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 IPSEC 88 #include <netinet6/ipsec.h> 89 #include <netinet6/ipsec6.h> 90 #include <netkey/key.h> 91 #endif /*IPSEC*/ 92 93 #include <machine/in_cksum.h> 94 95 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 96 97 static const int tcprexmtthresh = 3; 98 tcp_cc tcp_ccgen; 99 100 struct tcpstat tcpstat; 101 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 102 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 103 104 static int log_in_vain = 0; 105 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 106 &log_in_vain, 0, "Log all incoming TCP connections"); 107 108 static int blackhole = 0; 109 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 110 &blackhole, 0, "Do not send RST when dropping refused connections"); 111 112 int tcp_delack_enabled = 1; 113 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 114 &tcp_delack_enabled, 0, 115 "Delay ACK to try and piggyback it onto a data packet"); 116 117 #ifdef TCP_DROP_SYNFIN 118 static int drop_synfin = 0; 119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 120 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 121 #endif 122 123 struct inpcbhead tcb; 124 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 125 struct inpcbinfo tcbinfo; 126 struct mtx *tcbinfo_mtx; 127 128 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 129 static void tcp_pulloutofband(struct socket *, 130 struct tcphdr *, struct mbuf *, int); 131 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 132 struct mbuf *); 133 static void tcp_xmit_timer(struct tcpcb *, int); 134 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 135 136 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 137 #ifdef INET6 138 #define ND6_HINT(tp) \ 139 do { \ 140 if ((tp) && (tp)->t_inpcb && \ 141 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 142 (tp)->t_inpcb->in6p_route.ro_rt) \ 143 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 144 } while (0) 145 #else 146 #define ND6_HINT(tp) 147 #endif 148 149 /* 150 * Indicate whether this ack should be delayed. We can delay the ack if 151 * - delayed acks are enabled and 152 * - there is no delayed ack timer in progress and 153 * - our last ack wasn't a 0-sized window. We never want to delay 154 * the ack that opens up a 0-sized window. 155 */ 156 #define DELAY_ACK(tp) \ 157 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 158 (tp->t_flags & TF_RXWIN0SENT) == 0) 159 160 static int 161 tcp_reass(tp, th, tlenp, m) 162 register struct tcpcb *tp; 163 register struct tcphdr *th; 164 int *tlenp; 165 struct mbuf *m; 166 { 167 struct tseg_qent *q; 168 struct tseg_qent *p = NULL; 169 struct tseg_qent *nq; 170 struct tseg_qent *te; 171 struct socket *so = tp->t_inpcb->inp_socket; 172 int flags; 173 174 /* 175 * Call with th==0 after become established to 176 * force pre-ESTABLISHED data up to user socket. 177 */ 178 if (th == 0) 179 goto present; 180 181 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 182 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 183 M_NOWAIT); 184 if (te == NULL) { 185 tcpstat.tcps_rcvmemdrop++; 186 m_freem(m); 187 return (0); 188 } 189 190 /* 191 * Find a segment which begins after this one does. 192 */ 193 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 194 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 195 break; 196 p = q; 197 } 198 199 /* 200 * If there is a preceding segment, it may provide some of 201 * our data already. If so, drop the data from the incoming 202 * segment. If it provides all of our data, drop us. 203 */ 204 if (p != NULL) { 205 register int i; 206 /* conversion to int (in i) handles seq wraparound */ 207 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 208 if (i > 0) { 209 if (i >= *tlenp) { 210 tcpstat.tcps_rcvduppack++; 211 tcpstat.tcps_rcvdupbyte += *tlenp; 212 m_freem(m); 213 FREE(te, M_TSEGQ); 214 /* 215 * Try to present any queued data 216 * at the left window edge to the user. 217 * This is needed after the 3-WHS 218 * completes. 219 */ 220 goto present; /* ??? */ 221 } 222 m_adj(m, i); 223 *tlenp -= i; 224 th->th_seq += i; 225 } 226 } 227 tcpstat.tcps_rcvoopack++; 228 tcpstat.tcps_rcvoobyte += *tlenp; 229 230 /* 231 * While we overlap succeeding segments trim them or, 232 * if they are completely covered, dequeue them. 233 */ 234 while (q) { 235 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 236 if (i <= 0) 237 break; 238 if (i < q->tqe_len) { 239 q->tqe_th->th_seq += i; 240 q->tqe_len -= i; 241 m_adj(q->tqe_m, i); 242 break; 243 } 244 245 nq = LIST_NEXT(q, tqe_q); 246 LIST_REMOVE(q, tqe_q); 247 m_freem(q->tqe_m); 248 FREE(q, M_TSEGQ); 249 q = nq; 250 } 251 252 /* Insert the new segment queue entry into place. */ 253 te->tqe_m = m; 254 te->tqe_th = th; 255 te->tqe_len = *tlenp; 256 257 if (p == NULL) { 258 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 259 } else { 260 LIST_INSERT_AFTER(p, te, tqe_q); 261 } 262 263 present: 264 /* 265 * Present data to user, advancing rcv_nxt through 266 * completed sequence space. 267 */ 268 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 269 return (0); 270 q = LIST_FIRST(&tp->t_segq); 271 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 272 return (0); 273 do { 274 tp->rcv_nxt += q->tqe_len; 275 flags = q->tqe_th->th_flags & TH_FIN; 276 nq = LIST_NEXT(q, tqe_q); 277 LIST_REMOVE(q, tqe_q); 278 if (so->so_state & SS_CANTRCVMORE) 279 m_freem(q->tqe_m); 280 else 281 sbappend(&so->so_rcv, q->tqe_m); 282 FREE(q, M_TSEGQ); 283 q = nq; 284 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 285 ND6_HINT(tp); 286 sorwakeup(so); 287 return (flags); 288 } 289 290 /* 291 * TCP input routine, follows pages 65-76 of the 292 * protocol specification dated September, 1981 very closely. 293 */ 294 #ifdef INET6 295 int 296 tcp6_input(mp, offp, proto) 297 struct mbuf **mp; 298 int *offp, proto; 299 { 300 register struct mbuf *m = *mp; 301 struct in6_ifaddr *ia6; 302 303 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 304 305 /* 306 * draft-itojun-ipv6-tcp-to-anycast 307 * better place to put this in? 308 */ 309 ia6 = ip6_getdstifaddr(m); 310 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 311 struct ip6_hdr *ip6; 312 313 ip6 = mtod(m, struct ip6_hdr *); 314 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 315 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 316 return IPPROTO_DONE; 317 } 318 319 tcp_input(m, *offp); 320 return IPPROTO_DONE; 321 } 322 #endif 323 324 void 325 tcp_input(m, off0) 326 register struct mbuf *m; 327 int off0; 328 { 329 register struct tcphdr *th; 330 register struct ip *ip = NULL; 331 register struct ipovly *ipov; 332 register struct inpcb *inp = NULL; 333 u_char *optp = NULL; 334 int optlen = 0; 335 int len, tlen, off; 336 int drop_hdrlen; 337 register struct tcpcb *tp = 0; 338 register int thflags; 339 struct socket *so = 0; 340 int todrop, acked, ourfinisacked, needoutput = 0; 341 u_long tiwin; 342 struct tcpopt to; /* options in this segment */ 343 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 344 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 345 int headlocked = 0; 346 struct sockaddr_in *next_hop = NULL; 347 int rstreason; /* For badport_bandlim accounting purposes */ 348 349 struct ip6_hdr *ip6 = NULL; 350 #ifdef INET6 351 int isipv6; 352 #else 353 const int isipv6 = 0; 354 #endif 355 356 #ifdef TCPDEBUG 357 /* 358 * The size of tcp_saveipgen must be the size of the max ip header, 359 * now IPv6. 360 */ 361 u_char tcp_saveipgen[40]; 362 struct tcphdr tcp_savetcp; 363 short ostate = 0; 364 #endif 365 366 #ifdef MAC 367 int error; 368 #endif 369 370 /* Grab info from MT_TAG mbufs prepended to the chain. */ 371 for (;m && m->m_type == MT_TAG; m = m->m_next) { 372 if (m->m_tag_id == PACKET_TAG_IPFORWARD) 373 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 374 } 375 #ifdef INET6 376 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 377 #endif 378 bzero((char *)&to, sizeof(to)); 379 380 tcpstat.tcps_rcvtotal++; 381 382 if (isipv6) { 383 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 384 ip6 = mtod(m, struct ip6_hdr *); 385 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 386 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 387 tcpstat.tcps_rcvbadsum++; 388 goto drop; 389 } 390 th = (struct tcphdr *)((caddr_t)ip6 + off0); 391 392 /* 393 * Be proactive about unspecified IPv6 address in source. 394 * As we use all-zero to indicate unbounded/unconnected pcb, 395 * unspecified IPv6 address can be used to confuse us. 396 * 397 * Note that packets with unspecified IPv6 destination is 398 * already dropped in ip6_input. 399 */ 400 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 401 /* XXX stat */ 402 goto drop; 403 } 404 } else { 405 /* 406 * Get IP and TCP header together in first mbuf. 407 * Note: IP leaves IP header in first mbuf. 408 */ 409 if (off0 > sizeof (struct ip)) { 410 ip_stripoptions(m, (struct mbuf *)0); 411 off0 = sizeof(struct ip); 412 } 413 if (m->m_len < sizeof (struct tcpiphdr)) { 414 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 415 tcpstat.tcps_rcvshort++; 416 return; 417 } 418 } 419 ip = mtod(m, struct ip *); 420 ipov = (struct ipovly *)ip; 421 th = (struct tcphdr *)((caddr_t)ip + off0); 422 tlen = ip->ip_len; 423 424 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 425 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 426 th->th_sum = m->m_pkthdr.csum_data; 427 else 428 th->th_sum = in_pseudo(ip->ip_src.s_addr, 429 ip->ip_dst.s_addr, 430 htonl(m->m_pkthdr.csum_data + 431 ip->ip_len + 432 IPPROTO_TCP)); 433 th->th_sum ^= 0xffff; 434 } else { 435 /* 436 * Checksum extended TCP header and data. 437 */ 438 len = sizeof (struct ip) + tlen; 439 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 440 ipov->ih_len = (u_short)tlen; 441 ipov->ih_len = htons(ipov->ih_len); 442 th->th_sum = in_cksum(m, len); 443 } 444 if (th->th_sum) { 445 tcpstat.tcps_rcvbadsum++; 446 goto drop; 447 } 448 #ifdef INET6 449 /* Re-initialization for later version check */ 450 ip->ip_v = IPVERSION; 451 #endif 452 } 453 454 /* 455 * Check that TCP offset makes sense, 456 * pull out TCP options and adjust length. XXX 457 */ 458 off = th->th_off << 2; 459 if (off < sizeof (struct tcphdr) || off > tlen) { 460 tcpstat.tcps_rcvbadoff++; 461 goto drop; 462 } 463 tlen -= off; /* tlen is used instead of ti->ti_len */ 464 if (off > sizeof (struct tcphdr)) { 465 if (isipv6) { 466 IP6_EXTHDR_CHECK(m, off0, off, ); 467 ip6 = mtod(m, struct ip6_hdr *); 468 th = (struct tcphdr *)((caddr_t)ip6 + off0); 469 } else { 470 if (m->m_len < sizeof(struct ip) + off) { 471 if ((m = m_pullup(m, sizeof (struct ip) + off)) 472 == 0) { 473 tcpstat.tcps_rcvshort++; 474 return; 475 } 476 ip = mtod(m, struct ip *); 477 ipov = (struct ipovly *)ip; 478 th = (struct tcphdr *)((caddr_t)ip + off0); 479 } 480 } 481 optlen = off - sizeof (struct tcphdr); 482 optp = (u_char *)(th + 1); 483 } 484 thflags = th->th_flags; 485 486 #ifdef TCP_DROP_SYNFIN 487 /* 488 * If the drop_synfin option is enabled, drop all packets with 489 * both the SYN and FIN bits set. This prevents e.g. nmap from 490 * identifying the TCP/IP stack. 491 * 492 * This is a violation of the TCP specification. 493 */ 494 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 495 goto drop; 496 #endif 497 498 /* 499 * Convert TCP protocol specific fields to host format. 500 */ 501 th->th_seq = ntohl(th->th_seq); 502 th->th_ack = ntohl(th->th_ack); 503 th->th_win = ntohs(th->th_win); 504 th->th_urp = ntohs(th->th_urp); 505 506 /* 507 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 508 * until after ip6_savecontrol() is called and before other functions 509 * which don't want those proto headers. 510 * Because ip6_savecontrol() is going to parse the mbuf to 511 * search for data to be passed up to user-land, it wants mbuf 512 * parameters to be unchanged. 513 * XXX: the call of ip6_savecontrol() has been obsoleted based on 514 * latest version of the advanced API (20020110). 515 */ 516 drop_hdrlen = off0 + off; 517 518 /* 519 * Locate pcb for segment. 520 */ 521 INP_INFO_WLOCK(&tcbinfo); 522 headlocked = 1; 523 findpcb: 524 /* IPFIREWALL_FORWARD section */ 525 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 526 /* 527 * Transparently forwarded. Pretend to be the destination. 528 * already got one like this? 529 */ 530 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 531 ip->ip_dst, th->th_dport, 532 0, m->m_pkthdr.rcvif); 533 if (!inp) { 534 /* It's new. Try find the ambushing socket. */ 535 inp = in_pcblookup_hash(&tcbinfo, 536 ip->ip_src, th->th_sport, 537 next_hop->sin_addr, 538 next_hop->sin_port ? 539 ntohs(next_hop->sin_port) : 540 th->th_dport, 541 1, m->m_pkthdr.rcvif); 542 } 543 } else { 544 if (isipv6) 545 inp = in6_pcblookup_hash(&tcbinfo, 546 &ip6->ip6_src, th->th_sport, 547 &ip6->ip6_dst, th->th_dport, 548 1, m->m_pkthdr.rcvif); 549 else 550 inp = in_pcblookup_hash(&tcbinfo, 551 ip->ip_src, th->th_sport, 552 ip->ip_dst, th->th_dport, 553 1, m->m_pkthdr.rcvif); 554 } 555 556 #ifdef IPSEC 557 if (isipv6) { 558 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 559 ipsec6stat.in_polvio++; 560 goto drop; 561 } 562 } else { 563 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 564 ipsecstat.in_polvio++; 565 goto drop; 566 } 567 } 568 #endif 569 570 /* 571 * If the state is CLOSED (i.e., TCB does not exist) then 572 * all data in the incoming segment is discarded. 573 * If the TCB exists but is in CLOSED state, it is embryonic, 574 * but should either do a listen or a connect soon. 575 */ 576 if (inp == NULL) { 577 if (log_in_vain) { 578 #ifdef INET6 579 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 580 #else 581 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 582 #endif 583 584 if (isipv6) { 585 strcpy(dbuf, "["); 586 strcpy(sbuf, "["); 587 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 588 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 589 strcat(dbuf, "]"); 590 strcat(sbuf, "]"); 591 } else { 592 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 593 strcpy(sbuf, inet_ntoa(ip->ip_src)); 594 } 595 switch (log_in_vain) { 596 case 1: 597 if (thflags & TH_SYN) 598 log(LOG_INFO, 599 "Connection attempt to TCP %s:%d " 600 "from %s:%d\n", 601 dbuf, ntohs(th->th_dport), sbuf, 602 ntohs(th->th_sport)); 603 break; 604 case 2: 605 log(LOG_INFO, 606 "Connection attempt to TCP %s:%d " 607 "from %s:%d flags:0x%x\n", 608 dbuf, ntohs(th->th_dport), sbuf, 609 ntohs(th->th_sport), thflags); 610 break; 611 default: 612 break; 613 } 614 } 615 if (blackhole) { 616 switch (blackhole) { 617 case 1: 618 if (thflags & TH_SYN) 619 goto drop; 620 break; 621 case 2: 622 goto drop; 623 default: 624 goto drop; 625 } 626 } 627 rstreason = BANDLIM_RST_CLOSEDPORT; 628 goto dropwithreset; 629 } 630 INP_LOCK(inp); 631 tp = intotcpcb(inp); 632 if (tp == 0) { 633 INP_UNLOCK(inp); 634 rstreason = BANDLIM_RST_CLOSEDPORT; 635 goto dropwithreset; 636 } 637 if (tp->t_state == TCPS_CLOSED) 638 goto drop; 639 640 /* Unscale the window into a 32-bit value. */ 641 if ((thflags & TH_SYN) == 0) 642 tiwin = th->th_win << tp->snd_scale; 643 else 644 tiwin = th->th_win; 645 646 so = inp->inp_socket; 647 #ifdef MAC 648 error = mac_check_socket_deliver(so, m); 649 if (error) 650 goto drop; 651 #endif 652 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 653 struct in_conninfo inc; 654 #ifdef TCPDEBUG 655 if (so->so_options & SO_DEBUG) { 656 ostate = tp->t_state; 657 if (isipv6) 658 bcopy((char *)ip6, (char *)tcp_saveipgen, 659 sizeof(*ip6)); 660 else 661 bcopy((char *)ip, (char *)tcp_saveipgen, 662 sizeof(*ip)); 663 tcp_savetcp = *th; 664 } 665 #endif 666 /* skip if this isn't a listen socket */ 667 if ((so->so_options & SO_ACCEPTCONN) == 0) 668 goto after_listen; 669 #ifdef INET6 670 inc.inc_isipv6 = isipv6; 671 #endif 672 if (isipv6) { 673 inc.inc6_faddr = ip6->ip6_src; 674 inc.inc6_laddr = ip6->ip6_dst; 675 inc.inc6_route.ro_rt = NULL; /* XXX */ 676 } else { 677 inc.inc_faddr = ip->ip_src; 678 inc.inc_laddr = ip->ip_dst; 679 inc.inc_route.ro_rt = NULL; /* XXX */ 680 } 681 inc.inc_fport = th->th_sport; 682 inc.inc_lport = th->th_dport; 683 684 /* 685 * If the state is LISTEN then ignore segment if it contains 686 * a RST. If the segment contains an ACK then it is bad and 687 * send a RST. If it does not contain a SYN then it is not 688 * interesting; drop it. 689 * 690 * If the state is SYN_RECEIVED (syncache) and seg contains 691 * an ACK, but not for our SYN/ACK, send a RST. If the seg 692 * contains a RST, check the sequence number to see if it 693 * is a valid reset segment. 694 */ 695 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 696 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 697 if (!syncache_expand(&inc, th, &so, m)) { 698 /* 699 * No syncache entry, or ACK was not 700 * for our SYN/ACK. Send a RST. 701 */ 702 tcpstat.tcps_badsyn++; 703 rstreason = BANDLIM_RST_OPENPORT; 704 goto dropwithreset; 705 } 706 if (so == NULL) { 707 /* 708 * Could not complete 3-way handshake, 709 * connection is being closed down, and 710 * syncache will free mbuf. 711 */ 712 INP_UNLOCK(inp); 713 INP_INFO_WUNLOCK(&tcbinfo); 714 return; 715 } 716 /* 717 * Socket is created in state SYN_RECEIVED. 718 * Continue processing segment. 719 */ 720 INP_UNLOCK(inp); 721 inp = sotoinpcb(so); 722 INP_LOCK(inp); 723 tp = intotcpcb(inp); 724 /* 725 * This is what would have happened in 726 * tcp_output() when the SYN,ACK was sent. 727 */ 728 tp->snd_up = tp->snd_una; 729 tp->snd_max = tp->snd_nxt = tp->iss + 1; 730 tp->last_ack_sent = tp->rcv_nxt; 731 /* 732 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 733 * until the _second_ ACK is received: 734 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 735 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 736 * move to ESTAB, set snd_wnd to tiwin. 737 */ 738 tp->snd_wnd = tiwin; /* unscaled */ 739 goto after_listen; 740 } 741 if (thflags & TH_RST) { 742 syncache_chkrst(&inc, th); 743 goto drop; 744 } 745 if (thflags & TH_ACK) { 746 syncache_badack(&inc); 747 tcpstat.tcps_badsyn++; 748 rstreason = BANDLIM_RST_OPENPORT; 749 goto dropwithreset; 750 } 751 goto drop; 752 } 753 754 /* 755 * Segment's flags are (SYN) or (SYN|FIN). 756 */ 757 #ifdef INET6 758 /* 759 * If deprecated address is forbidden, 760 * we do not accept SYN to deprecated interface 761 * address to prevent any new inbound connection from 762 * getting established. 763 * When we do not accept SYN, we send a TCP RST, 764 * with deprecated source address (instead of dropping 765 * it). We compromise it as it is much better for peer 766 * to send a RST, and RST will be the final packet 767 * for the exchange. 768 * 769 * If we do not forbid deprecated addresses, we accept 770 * the SYN packet. RFC2462 does not suggest dropping 771 * SYN in this case. 772 * If we decipher RFC2462 5.5.4, it says like this: 773 * 1. use of deprecated addr with existing 774 * communication is okay - "SHOULD continue to be 775 * used" 776 * 2. use of it with new communication: 777 * (2a) "SHOULD NOT be used if alternate address 778 * with sufficient scope is available" 779 * (2b) nothing mentioned otherwise. 780 * Here we fall into (2b) case as we have no choice in 781 * our source address selection - we must obey the peer. 782 * 783 * The wording in RFC2462 is confusing, and there are 784 * multiple description text for deprecated address 785 * handling - worse, they are not exactly the same. 786 * I believe 5.5.4 is the best one, so we follow 5.5.4. 787 */ 788 if (isipv6 && !ip6_use_deprecated) { 789 struct in6_ifaddr *ia6; 790 791 if ((ia6 = ip6_getdstifaddr(m)) && 792 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 793 INP_UNLOCK(inp); 794 tp = NULL; 795 rstreason = BANDLIM_RST_OPENPORT; 796 goto dropwithreset; 797 } 798 } 799 #endif 800 /* 801 * If it is from this socket, drop it, it must be forged. 802 * Don't bother responding if the destination was a broadcast. 803 */ 804 if (th->th_dport == th->th_sport) { 805 if (isipv6) { 806 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 807 &ip6->ip6_src)) 808 goto drop; 809 } else { 810 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 811 goto drop; 812 } 813 } 814 /* 815 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 816 * 817 * Note that it is quite possible to receive unicast 818 * link-layer packets with a broadcast IP address. Use 819 * in_broadcast() to find them. 820 */ 821 if (m->m_flags & (M_BCAST|M_MCAST)) 822 goto drop; 823 if (isipv6) { 824 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 825 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 826 goto drop; 827 } else { 828 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 829 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 830 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 831 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 832 goto drop; 833 } 834 /* 835 * SYN appears to be valid; create compressed TCP state 836 * for syncache, or perform t/tcp connection. 837 */ 838 if (so->so_qlen <= so->so_qlimit) { 839 tcp_dooptions(&to, optp, optlen, 1); 840 if (!syncache_add(&inc, &to, th, &so, m)) 841 goto drop; 842 if (so == NULL) { 843 /* 844 * Entry added to syncache, mbuf used to 845 * send SYN,ACK packet. 846 */ 847 KASSERT(headlocked, ("headlocked")); 848 INP_UNLOCK(inp); 849 INP_INFO_WUNLOCK(&tcbinfo); 850 return; 851 } 852 /* 853 * Segment passed TAO tests. 854 */ 855 INP_UNLOCK(inp); 856 inp = sotoinpcb(so); 857 INP_LOCK(inp); 858 tp = intotcpcb(inp); 859 tp->snd_wnd = tiwin; 860 tp->t_starttime = ticks; 861 tp->t_state = TCPS_ESTABLISHED; 862 863 /* 864 * If there is a FIN, or if there is data and the 865 * connection is local, then delay SYN,ACK(SYN) in 866 * the hope of piggy-backing it on a response 867 * segment. Otherwise must send ACK now in case 868 * the other side is slow starting. 869 */ 870 if (DELAY_ACK(tp) && 871 ((thflags & TH_FIN) || 872 (tlen != 0 && 873 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 874 (!isipv6 && in_localaddr(inp->inp_faddr)))))) { 875 callout_reset(tp->tt_delack, tcp_delacktime, 876 tcp_timer_delack, tp); 877 tp->t_flags |= TF_NEEDSYN; 878 } else 879 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 880 881 tcpstat.tcps_connects++; 882 soisconnected(so); 883 goto trimthenstep6; 884 } 885 goto drop; 886 } 887 after_listen: 888 889 /* XXX temp debugging */ 890 /* should not happen - syncache should pick up these connections */ 891 if (tp->t_state == TCPS_LISTEN) 892 panic("tcp_input: TCPS_LISTEN"); 893 894 /* 895 * Segment received on connection. 896 * Reset idle time and keep-alive timer. 897 */ 898 tp->t_rcvtime = ticks; 899 if (TCPS_HAVEESTABLISHED(tp->t_state)) 900 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 901 902 /* 903 * Process options. 904 * XXX this is tradtitional behavior, may need to be cleaned up. 905 */ 906 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 907 if (thflags & TH_SYN) { 908 if (to.to_flags & TOF_SCALE) { 909 tp->t_flags |= TF_RCVD_SCALE; 910 tp->requested_s_scale = to.to_requested_s_scale; 911 } 912 if (to.to_flags & TOF_TS) { 913 tp->t_flags |= TF_RCVD_TSTMP; 914 tp->ts_recent = to.to_tsval; 915 tp->ts_recent_age = ticks; 916 } 917 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 918 tp->t_flags |= TF_RCVD_CC; 919 if (to.to_flags & TOF_MSS) 920 tcp_mss(tp, to.to_mss); 921 } 922 923 /* 924 * Header prediction: check for the two common cases 925 * of a uni-directional data xfer. If the packet has 926 * no control flags, is in-sequence, the window didn't 927 * change and we're not retransmitting, it's a 928 * candidate. If the length is zero and the ack moved 929 * forward, we're the sender side of the xfer. Just 930 * free the data acked & wake any higher level process 931 * that was blocked waiting for space. If the length 932 * is non-zero and the ack didn't move, we're the 933 * receiver side. If we're getting packets in-order 934 * (the reassembly queue is empty), add the data to 935 * the socket buffer and note that we need a delayed ack. 936 * Make sure that the hidden state-flags are also off. 937 * Since we check for TCPS_ESTABLISHED above, it can only 938 * be TH_NEEDSYN. 939 */ 940 if (tp->t_state == TCPS_ESTABLISHED && 941 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 942 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 943 ((to.to_flags & TOF_TS) == 0 || 944 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 945 /* 946 * Using the CC option is compulsory if once started: 947 * the segment is OK if no T/TCP was negotiated or 948 * if the segment has a CC option equal to CCrecv 949 */ 950 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 951 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 952 th->th_seq == tp->rcv_nxt && 953 tiwin && tiwin == tp->snd_wnd && 954 tp->snd_nxt == tp->snd_max) { 955 956 /* 957 * If last ACK falls within this segment's sequence numbers, 958 * record the timestamp. 959 * NOTE that the test is modified according to the latest 960 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 961 */ 962 if ((to.to_flags & TOF_TS) != 0 && 963 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 964 tp->ts_recent_age = ticks; 965 tp->ts_recent = to.to_tsval; 966 } 967 968 if (tlen == 0) { 969 if (SEQ_GT(th->th_ack, tp->snd_una) && 970 SEQ_LEQ(th->th_ack, tp->snd_max) && 971 tp->snd_cwnd >= tp->snd_wnd && 972 tp->t_dupacks < tcprexmtthresh) { 973 KASSERT(headlocked, ("headlocked")); 974 INP_INFO_WUNLOCK(&tcbinfo); 975 headlocked = 0; 976 /* 977 * this is a pure ack for outstanding data. 978 */ 979 ++tcpstat.tcps_predack; 980 /* 981 * "bad retransmit" recovery 982 */ 983 if (tp->t_rxtshift == 1 && 984 ticks < tp->t_badrxtwin) { 985 tp->snd_cwnd = tp->snd_cwnd_prev; 986 tp->snd_ssthresh = 987 tp->snd_ssthresh_prev; 988 tp->snd_nxt = tp->snd_max; 989 tp->t_badrxtwin = 0; 990 } 991 if ((to.to_flags & TOF_TS) != 0) 992 tcp_xmit_timer(tp, 993 ticks - to.to_tsecr + 1); 994 else if (tp->t_rtttime && 995 SEQ_GT(th->th_ack, tp->t_rtseq)) 996 tcp_xmit_timer(tp, 997 ticks - tp->t_rtttime); 998 tcp_xmit_bandwidth_limit(tp, th->th_ack); 999 acked = th->th_ack - tp->snd_una; 1000 tcpstat.tcps_rcvackpack++; 1001 tcpstat.tcps_rcvackbyte += acked; 1002 sbdrop(&so->so_snd, acked); 1003 tp->snd_una = th->th_ack; 1004 tp->t_dupacks = 0; 1005 m_freem(m); 1006 ND6_HINT(tp); /* some progress has been done */ 1007 1008 /* 1009 * If all outstanding data are acked, stop 1010 * retransmit timer, otherwise restart timer 1011 * using current (possibly backed-off) value. 1012 * If process is waiting for space, 1013 * wakeup/selwakeup/signal. If data 1014 * are ready to send, let tcp_output 1015 * decide between more output or persist. 1016 */ 1017 if (tp->snd_una == tp->snd_max) 1018 callout_stop(tp->tt_rexmt); 1019 else if (!callout_active(tp->tt_persist)) 1020 callout_reset(tp->tt_rexmt, 1021 tp->t_rxtcur, 1022 tcp_timer_rexmt, tp); 1023 1024 sowwakeup(so); 1025 if (so->so_snd.sb_cc) 1026 (void) tcp_output(tp); 1027 INP_UNLOCK(inp); 1028 return; 1029 } 1030 } else if (th->th_ack == tp->snd_una && 1031 LIST_EMPTY(&tp->t_segq) && 1032 tlen <= sbspace(&so->so_rcv)) { 1033 KASSERT(headlocked, ("headlocked")); 1034 INP_INFO_WUNLOCK(&tcbinfo); 1035 headlocked = 0; 1036 /* 1037 * this is a pure, in-sequence data packet 1038 * with nothing on the reassembly queue and 1039 * we have enough buffer space to take it. 1040 */ 1041 ++tcpstat.tcps_preddat; 1042 tp->rcv_nxt += tlen; 1043 tcpstat.tcps_rcvpack++; 1044 tcpstat.tcps_rcvbyte += tlen; 1045 ND6_HINT(tp); /* some progress has been done */ 1046 /* 1047 * Add data to socket buffer. 1048 */ 1049 m_adj(m, drop_hdrlen); /* delayed header drop */ 1050 sbappend(&so->so_rcv, m); 1051 sorwakeup(so); 1052 if (DELAY_ACK(tp)) { 1053 callout_reset(tp->tt_delack, tcp_delacktime, 1054 tcp_timer_delack, tp); 1055 } else { 1056 tp->t_flags |= TF_ACKNOW; 1057 tcp_output(tp); 1058 } 1059 INP_UNLOCK(inp); 1060 return; 1061 } 1062 } 1063 1064 /* 1065 * Calculate amount of space in receive window, 1066 * and then do TCP input processing. 1067 * Receive window is amount of space in rcv queue, 1068 * but not less than advertised window. 1069 */ 1070 { int win; 1071 1072 win = sbspace(&so->so_rcv); 1073 if (win < 0) 1074 win = 0; 1075 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1076 } 1077 1078 switch (tp->t_state) { 1079 1080 /* 1081 * If the state is SYN_RECEIVED: 1082 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1083 */ 1084 case TCPS_SYN_RECEIVED: 1085 if ((thflags & TH_ACK) && 1086 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1087 SEQ_GT(th->th_ack, tp->snd_max))) { 1088 rstreason = BANDLIM_RST_OPENPORT; 1089 goto dropwithreset; 1090 } 1091 break; 1092 1093 /* 1094 * If the state is SYN_SENT: 1095 * if seg contains an ACK, but not for our SYN, drop the input. 1096 * if seg contains a RST, then drop the connection. 1097 * if seg does not contain SYN, then drop it. 1098 * Otherwise this is an acceptable SYN segment 1099 * initialize tp->rcv_nxt and tp->irs 1100 * if seg contains ack then advance tp->snd_una 1101 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1102 * arrange for segment to be acked (eventually) 1103 * continue processing rest of data/controls, beginning with URG 1104 */ 1105 case TCPS_SYN_SENT: 1106 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1107 taop = &tao_noncached; 1108 bzero(taop, sizeof(*taop)); 1109 } 1110 1111 if ((thflags & TH_ACK) && 1112 (SEQ_LEQ(th->th_ack, tp->iss) || 1113 SEQ_GT(th->th_ack, tp->snd_max))) { 1114 /* 1115 * If we have a cached CCsent for the remote host, 1116 * hence we haven't just crashed and restarted, 1117 * do not send a RST. This may be a retransmission 1118 * from the other side after our earlier ACK was lost. 1119 * Our new SYN, when it arrives, will serve as the 1120 * needed ACK. 1121 */ 1122 if (taop->tao_ccsent != 0) 1123 goto drop; 1124 else { 1125 rstreason = BANDLIM_UNLIMITED; 1126 goto dropwithreset; 1127 } 1128 } 1129 if (thflags & TH_RST) { 1130 if (thflags & TH_ACK) 1131 tp = tcp_drop(tp, ECONNREFUSED); 1132 goto drop; 1133 } 1134 if ((thflags & TH_SYN) == 0) 1135 goto drop; 1136 tp->snd_wnd = th->th_win; /* initial send window */ 1137 tp->cc_recv = to.to_cc; /* foreign CC */ 1138 1139 tp->irs = th->th_seq; 1140 tcp_rcvseqinit(tp); 1141 if (thflags & TH_ACK) { 1142 /* 1143 * Our SYN was acked. If segment contains CC.ECHO 1144 * option, check it to make sure this segment really 1145 * matches our SYN. If not, just drop it as old 1146 * duplicate, but send an RST if we're still playing 1147 * by the old rules. If no CC.ECHO option, make sure 1148 * we don't get fooled into using T/TCP. 1149 */ 1150 if (to.to_flags & TOF_CCECHO) { 1151 if (tp->cc_send != to.to_ccecho) { 1152 if (taop->tao_ccsent != 0) 1153 goto drop; 1154 else { 1155 rstreason = BANDLIM_UNLIMITED; 1156 goto dropwithreset; 1157 } 1158 } 1159 } else 1160 tp->t_flags &= ~TF_RCVD_CC; 1161 tcpstat.tcps_connects++; 1162 soisconnected(so); 1163 #ifdef MAC 1164 mac_set_socket_peer_from_mbuf(m, so); 1165 #endif 1166 /* Do window scaling on this connection? */ 1167 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1168 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1169 tp->snd_scale = tp->requested_s_scale; 1170 tp->rcv_scale = tp->request_r_scale; 1171 } 1172 /* Segment is acceptable, update cache if undefined. */ 1173 if (taop->tao_ccsent == 0) 1174 taop->tao_ccsent = to.to_ccecho; 1175 1176 tp->rcv_adv += tp->rcv_wnd; 1177 tp->snd_una++; /* SYN is acked */ 1178 /* 1179 * If there's data, delay ACK; if there's also a FIN 1180 * ACKNOW will be turned on later. 1181 */ 1182 if (DELAY_ACK(tp) && tlen != 0) 1183 callout_reset(tp->tt_delack, tcp_delacktime, 1184 tcp_timer_delack, tp); 1185 else 1186 tp->t_flags |= TF_ACKNOW; 1187 /* 1188 * Received <SYN,ACK> in SYN_SENT[*] state. 1189 * Transitions: 1190 * SYN_SENT --> ESTABLISHED 1191 * SYN_SENT* --> FIN_WAIT_1 1192 */ 1193 tp->t_starttime = ticks; 1194 if (tp->t_flags & TF_NEEDFIN) { 1195 tp->t_state = TCPS_FIN_WAIT_1; 1196 tp->t_flags &= ~TF_NEEDFIN; 1197 thflags &= ~TH_SYN; 1198 } else { 1199 tp->t_state = TCPS_ESTABLISHED; 1200 callout_reset(tp->tt_keep, tcp_keepidle, 1201 tcp_timer_keep, tp); 1202 } 1203 } else { 1204 /* 1205 * Received initial SYN in SYN-SENT[*] state => 1206 * simultaneous open. If segment contains CC option 1207 * and there is a cached CC, apply TAO test. 1208 * If it succeeds, connection is * half-synchronized. 1209 * Otherwise, do 3-way handshake: 1210 * SYN-SENT -> SYN-RECEIVED 1211 * SYN-SENT* -> SYN-RECEIVED* 1212 * If there was no CC option, clear cached CC value. 1213 */ 1214 tp->t_flags |= TF_ACKNOW; 1215 callout_stop(tp->tt_rexmt); 1216 if (to.to_flags & TOF_CC) { 1217 if (taop->tao_cc != 0 && 1218 CC_GT(to.to_cc, taop->tao_cc)) { 1219 /* 1220 * update cache and make transition: 1221 * SYN-SENT -> ESTABLISHED* 1222 * SYN-SENT* -> FIN-WAIT-1* 1223 */ 1224 taop->tao_cc = to.to_cc; 1225 tp->t_starttime = ticks; 1226 if (tp->t_flags & TF_NEEDFIN) { 1227 tp->t_state = TCPS_FIN_WAIT_1; 1228 tp->t_flags &= ~TF_NEEDFIN; 1229 } else { 1230 tp->t_state = TCPS_ESTABLISHED; 1231 callout_reset(tp->tt_keep, 1232 tcp_keepidle, 1233 tcp_timer_keep, 1234 tp); 1235 } 1236 tp->t_flags |= TF_NEEDSYN; 1237 } else 1238 tp->t_state = TCPS_SYN_RECEIVED; 1239 } else { 1240 /* CC.NEW or no option => invalidate cache */ 1241 taop->tao_cc = 0; 1242 tp->t_state = TCPS_SYN_RECEIVED; 1243 } 1244 } 1245 1246 trimthenstep6: 1247 /* 1248 * Advance th->th_seq to correspond to first data byte. 1249 * If data, trim to stay within window, 1250 * dropping FIN if necessary. 1251 */ 1252 th->th_seq++; 1253 if (tlen > tp->rcv_wnd) { 1254 todrop = tlen - tp->rcv_wnd; 1255 m_adj(m, -todrop); 1256 tlen = tp->rcv_wnd; 1257 thflags &= ~TH_FIN; 1258 tcpstat.tcps_rcvpackafterwin++; 1259 tcpstat.tcps_rcvbyteafterwin += todrop; 1260 } 1261 tp->snd_wl1 = th->th_seq - 1; 1262 tp->rcv_up = th->th_seq; 1263 /* 1264 * Client side of transaction: already sent SYN and data. 1265 * If the remote host used T/TCP to validate the SYN, 1266 * our data will be ACK'd; if so, enter normal data segment 1267 * processing in the middle of step 5, ack processing. 1268 * Otherwise, goto step 6. 1269 */ 1270 if (thflags & TH_ACK) 1271 goto process_ACK; 1272 1273 goto step6; 1274 1275 /* 1276 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1277 * if segment contains a SYN and CC [not CC.NEW] option: 1278 * if state == TIME_WAIT and connection duration > MSL, 1279 * drop packet and send RST; 1280 * 1281 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1282 * ack the FIN (and data) in retransmission queue. 1283 * Complete close and delete TCPCB. Then reprocess 1284 * segment, hoping to find new TCPCB in LISTEN state; 1285 * 1286 * else must be old SYN; drop it. 1287 * else do normal processing. 1288 */ 1289 case TCPS_LAST_ACK: 1290 case TCPS_CLOSING: 1291 case TCPS_TIME_WAIT: 1292 if ((thflags & TH_SYN) && 1293 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1294 if (tp->t_state == TCPS_TIME_WAIT && 1295 (ticks - tp->t_starttime) > tcp_msl) { 1296 rstreason = BANDLIM_UNLIMITED; 1297 goto dropwithreset; 1298 } 1299 if (CC_GT(to.to_cc, tp->cc_recv)) { 1300 tp = tcp_close(tp); 1301 goto findpcb; 1302 } 1303 else 1304 goto drop; 1305 } 1306 break; /* continue normal processing */ 1307 } 1308 1309 /* 1310 * States other than LISTEN or SYN_SENT. 1311 * First check the RST flag and sequence number since reset segments 1312 * are exempt from the timestamp and connection count tests. This 1313 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1314 * below which allowed reset segments in half the sequence space 1315 * to fall though and be processed (which gives forged reset 1316 * segments with a random sequence number a 50 percent chance of 1317 * killing a connection). 1318 * Then check timestamp, if present. 1319 * Then check the connection count, if present. 1320 * Then check that at least some bytes of segment are within 1321 * receive window. If segment begins before rcv_nxt, 1322 * drop leading data (and SYN); if nothing left, just ack. 1323 * 1324 * 1325 * If the RST bit is set, check the sequence number to see 1326 * if this is a valid reset segment. 1327 * RFC 793 page 37: 1328 * In all states except SYN-SENT, all reset (RST) segments 1329 * are validated by checking their SEQ-fields. A reset is 1330 * valid if its sequence number is in the window. 1331 * Note: this does not take into account delayed ACKs, so 1332 * we should test against last_ack_sent instead of rcv_nxt. 1333 * The sequence number in the reset segment is normally an 1334 * echo of our outgoing acknowlegement numbers, but some hosts 1335 * send a reset with the sequence number at the rightmost edge 1336 * of our receive window, and we have to handle this case. 1337 * If we have multiple segments in flight, the intial reset 1338 * segment sequence numbers will be to the left of last_ack_sent, 1339 * but they will eventually catch up. 1340 * In any case, it never made sense to trim reset segments to 1341 * fit the receive window since RFC 1122 says: 1342 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1343 * 1344 * A TCP SHOULD allow a received RST segment to include data. 1345 * 1346 * DISCUSSION 1347 * It has been suggested that a RST segment could contain 1348 * ASCII text that encoded and explained the cause of the 1349 * RST. No standard has yet been established for such 1350 * data. 1351 * 1352 * If the reset segment passes the sequence number test examine 1353 * the state: 1354 * SYN_RECEIVED STATE: 1355 * If passive open, return to LISTEN state. 1356 * If active open, inform user that connection was refused. 1357 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1358 * Inform user that connection was reset, and close tcb. 1359 * CLOSING, LAST_ACK STATES: 1360 * Close the tcb. 1361 * TIME_WAIT STATE: 1362 * Drop the segment - see Stevens, vol. 2, p. 964 and 1363 * RFC 1337. 1364 */ 1365 if (thflags & TH_RST) { 1366 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1367 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1368 switch (tp->t_state) { 1369 1370 case TCPS_SYN_RECEIVED: 1371 so->so_error = ECONNREFUSED; 1372 goto close; 1373 1374 case TCPS_ESTABLISHED: 1375 case TCPS_FIN_WAIT_1: 1376 case TCPS_FIN_WAIT_2: 1377 case TCPS_CLOSE_WAIT: 1378 so->so_error = ECONNRESET; 1379 close: 1380 tp->t_state = TCPS_CLOSED; 1381 tcpstat.tcps_drops++; 1382 tp = tcp_close(tp); 1383 break; 1384 1385 case TCPS_CLOSING: 1386 case TCPS_LAST_ACK: 1387 tp = tcp_close(tp); 1388 break; 1389 1390 case TCPS_TIME_WAIT: 1391 break; 1392 } 1393 } 1394 goto drop; 1395 } 1396 1397 /* 1398 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1399 * and it's less than ts_recent, drop it. 1400 */ 1401 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1402 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1403 1404 /* Check to see if ts_recent is over 24 days old. */ 1405 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1406 /* 1407 * Invalidate ts_recent. If this segment updates 1408 * ts_recent, the age will be reset later and ts_recent 1409 * will get a valid value. If it does not, setting 1410 * ts_recent to zero will at least satisfy the 1411 * requirement that zero be placed in the timestamp 1412 * echo reply when ts_recent isn't valid. The 1413 * age isn't reset until we get a valid ts_recent 1414 * because we don't want out-of-order segments to be 1415 * dropped when ts_recent is old. 1416 */ 1417 tp->ts_recent = 0; 1418 } else { 1419 tcpstat.tcps_rcvduppack++; 1420 tcpstat.tcps_rcvdupbyte += tlen; 1421 tcpstat.tcps_pawsdrop++; 1422 goto dropafterack; 1423 } 1424 } 1425 1426 /* 1427 * T/TCP mechanism 1428 * If T/TCP was negotiated and the segment doesn't have CC, 1429 * or if its CC is wrong then drop the segment. 1430 * RST segments do not have to comply with this. 1431 */ 1432 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1433 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1434 goto dropafterack; 1435 1436 /* 1437 * In the SYN-RECEIVED state, validate that the packet belongs to 1438 * this connection before trimming the data to fit the receive 1439 * window. Check the sequence number versus IRS since we know 1440 * the sequence numbers haven't wrapped. This is a partial fix 1441 * for the "LAND" DoS attack. 1442 */ 1443 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1444 rstreason = BANDLIM_RST_OPENPORT; 1445 goto dropwithreset; 1446 } 1447 1448 todrop = tp->rcv_nxt - th->th_seq; 1449 if (todrop > 0) { 1450 if (thflags & TH_SYN) { 1451 thflags &= ~TH_SYN; 1452 th->th_seq++; 1453 if (th->th_urp > 1) 1454 th->th_urp--; 1455 else 1456 thflags &= ~TH_URG; 1457 todrop--; 1458 } 1459 /* 1460 * Following if statement from Stevens, vol. 2, p. 960. 1461 */ 1462 if (todrop > tlen 1463 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1464 /* 1465 * Any valid FIN must be to the left of the window. 1466 * At this point the FIN must be a duplicate or out 1467 * of sequence; drop it. 1468 */ 1469 thflags &= ~TH_FIN; 1470 1471 /* 1472 * Send an ACK to resynchronize and drop any data. 1473 * But keep on processing for RST or ACK. 1474 */ 1475 tp->t_flags |= TF_ACKNOW; 1476 todrop = tlen; 1477 tcpstat.tcps_rcvduppack++; 1478 tcpstat.tcps_rcvdupbyte += todrop; 1479 } else { 1480 tcpstat.tcps_rcvpartduppack++; 1481 tcpstat.tcps_rcvpartdupbyte += todrop; 1482 } 1483 drop_hdrlen += todrop; /* drop from the top afterwards */ 1484 th->th_seq += todrop; 1485 tlen -= todrop; 1486 if (th->th_urp > todrop) 1487 th->th_urp -= todrop; 1488 else { 1489 thflags &= ~TH_URG; 1490 th->th_urp = 0; 1491 } 1492 } 1493 1494 /* 1495 * If new data are received on a connection after the 1496 * user processes are gone, then RST the other end. 1497 */ 1498 if ((so->so_state & SS_NOFDREF) && 1499 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1500 tp = tcp_close(tp); 1501 tcpstat.tcps_rcvafterclose++; 1502 rstreason = BANDLIM_UNLIMITED; 1503 goto dropwithreset; 1504 } 1505 1506 /* 1507 * If segment ends after window, drop trailing data 1508 * (and PUSH and FIN); if nothing left, just ACK. 1509 */ 1510 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1511 if (todrop > 0) { 1512 tcpstat.tcps_rcvpackafterwin++; 1513 if (todrop >= tlen) { 1514 tcpstat.tcps_rcvbyteafterwin += tlen; 1515 /* 1516 * If a new connection request is received 1517 * while in TIME_WAIT, drop the old connection 1518 * and start over if the sequence numbers 1519 * are above the previous ones. 1520 */ 1521 if (thflags & TH_SYN && 1522 tp->t_state == TCPS_TIME_WAIT && 1523 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1524 tp = tcp_close(tp); 1525 goto findpcb; 1526 } 1527 /* 1528 * If window is closed can only take segments at 1529 * window edge, and have to drop data and PUSH from 1530 * incoming segments. Continue processing, but 1531 * remember to ack. Otherwise, drop segment 1532 * and ack. 1533 */ 1534 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1535 tp->t_flags |= TF_ACKNOW; 1536 tcpstat.tcps_rcvwinprobe++; 1537 } else 1538 goto dropafterack; 1539 } else 1540 tcpstat.tcps_rcvbyteafterwin += todrop; 1541 m_adj(m, -todrop); 1542 tlen -= todrop; 1543 thflags &= ~(TH_PUSH|TH_FIN); 1544 } 1545 1546 /* 1547 * If last ACK falls within this segment's sequence numbers, 1548 * record its timestamp. 1549 * NOTE that the test is modified according to the latest 1550 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1551 */ 1552 if ((to.to_flags & TOF_TS) != 0 && 1553 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1554 tp->ts_recent_age = ticks; 1555 tp->ts_recent = to.to_tsval; 1556 } 1557 1558 /* 1559 * If a SYN is in the window, then this is an 1560 * error and we send an RST and drop the connection. 1561 */ 1562 if (thflags & TH_SYN) { 1563 tp = tcp_drop(tp, ECONNRESET); 1564 rstreason = BANDLIM_UNLIMITED; 1565 goto dropwithreset; 1566 } 1567 1568 /* 1569 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1570 * flag is on (half-synchronized state), then queue data for 1571 * later processing; else drop segment and return. 1572 */ 1573 if ((thflags & TH_ACK) == 0) { 1574 if (tp->t_state == TCPS_SYN_RECEIVED || 1575 (tp->t_flags & TF_NEEDSYN)) 1576 goto step6; 1577 else 1578 goto drop; 1579 } 1580 1581 /* 1582 * Ack processing. 1583 */ 1584 switch (tp->t_state) { 1585 1586 /* 1587 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1588 * ESTABLISHED state and continue processing. 1589 * The ACK was checked above. 1590 */ 1591 case TCPS_SYN_RECEIVED: 1592 1593 tcpstat.tcps_connects++; 1594 soisconnected(so); 1595 /* Do window scaling? */ 1596 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1597 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1598 tp->snd_scale = tp->requested_s_scale; 1599 tp->rcv_scale = tp->request_r_scale; 1600 } 1601 /* 1602 * Upon successful completion of 3-way handshake, 1603 * update cache.CC if it was undefined, pass any queued 1604 * data to the user, and advance state appropriately. 1605 */ 1606 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1607 taop->tao_cc == 0) 1608 taop->tao_cc = tp->cc_recv; 1609 1610 /* 1611 * Make transitions: 1612 * SYN-RECEIVED -> ESTABLISHED 1613 * SYN-RECEIVED* -> FIN-WAIT-1 1614 */ 1615 tp->t_starttime = ticks; 1616 if (tp->t_flags & TF_NEEDFIN) { 1617 tp->t_state = TCPS_FIN_WAIT_1; 1618 tp->t_flags &= ~TF_NEEDFIN; 1619 } else { 1620 tp->t_state = TCPS_ESTABLISHED; 1621 callout_reset(tp->tt_keep, tcp_keepidle, 1622 tcp_timer_keep, tp); 1623 } 1624 /* 1625 * If segment contains data or ACK, will call tcp_reass() 1626 * later; if not, do so now to pass queued data to user. 1627 */ 1628 if (tlen == 0 && (thflags & TH_FIN) == 0) 1629 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1630 (struct mbuf *)0); 1631 tp->snd_wl1 = th->th_seq - 1; 1632 /* fall into ... */ 1633 1634 /* 1635 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1636 * ACKs. If the ack is in the range 1637 * tp->snd_una < th->th_ack <= tp->snd_max 1638 * then advance tp->snd_una to th->th_ack and drop 1639 * data from the retransmission queue. If this ACK reflects 1640 * more up to date window information we update our window information. 1641 */ 1642 case TCPS_ESTABLISHED: 1643 case TCPS_FIN_WAIT_1: 1644 case TCPS_FIN_WAIT_2: 1645 case TCPS_CLOSE_WAIT: 1646 case TCPS_CLOSING: 1647 case TCPS_LAST_ACK: 1648 case TCPS_TIME_WAIT: 1649 1650 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1651 if (tlen == 0 && tiwin == tp->snd_wnd) { 1652 tcpstat.tcps_rcvdupack++; 1653 /* 1654 * If we have outstanding data (other than 1655 * a window probe), this is a completely 1656 * duplicate ack (ie, window info didn't 1657 * change), the ack is the biggest we've 1658 * seen and we've seen exactly our rexmt 1659 * threshhold of them, assume a packet 1660 * has been dropped and retransmit it. 1661 * Kludge snd_nxt & the congestion 1662 * window so we send only this one 1663 * packet. 1664 * 1665 * We know we're losing at the current 1666 * window size so do congestion avoidance 1667 * (set ssthresh to half the current window 1668 * and pull our congestion window back to 1669 * the new ssthresh). 1670 * 1671 * Dup acks mean that packets have left the 1672 * network (they're now cached at the receiver) 1673 * so bump cwnd by the amount in the receiver 1674 * to keep a constant cwnd packets in the 1675 * network. 1676 */ 1677 if (!callout_active(tp->tt_rexmt) || 1678 th->th_ack != tp->snd_una) 1679 tp->t_dupacks = 0; 1680 else if (++tp->t_dupacks == tcprexmtthresh) { 1681 tcp_seq onxt = tp->snd_nxt; 1682 u_int win = 1683 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1684 tp->t_maxseg; 1685 if (tcp_do_newreno && 1686 SEQ_LT(th->th_ack, 1687 tp->snd_recover)) { 1688 /* False retransmit, should not 1689 * cut window 1690 */ 1691 tp->snd_cwnd += tp->t_maxseg; 1692 tp->t_dupacks = 0; 1693 (void) tcp_output(tp); 1694 goto drop; 1695 } 1696 if (win < 2) 1697 win = 2; 1698 tp->snd_ssthresh = win * tp->t_maxseg; 1699 tp->snd_recover = tp->snd_max; 1700 callout_stop(tp->tt_rexmt); 1701 tp->t_rtttime = 0; 1702 tp->snd_nxt = th->th_ack; 1703 tp->snd_cwnd = tp->t_maxseg; 1704 (void) tcp_output(tp); 1705 tp->snd_cwnd = tp->snd_ssthresh + 1706 tp->t_maxseg * tp->t_dupacks; 1707 if (SEQ_GT(onxt, tp->snd_nxt)) 1708 tp->snd_nxt = onxt; 1709 goto drop; 1710 } else if (tp->t_dupacks > tcprexmtthresh) { 1711 tp->snd_cwnd += tp->t_maxseg; 1712 (void) tcp_output(tp); 1713 goto drop; 1714 } 1715 } else 1716 tp->t_dupacks = 0; 1717 break; 1718 } 1719 /* 1720 * If the congestion window was inflated to account 1721 * for the other side's cached packets, retract it. 1722 */ 1723 if (tcp_do_newreno) { 1724 int is_partialack = SEQ_LT(th->th_ack, tp->snd_recover); 1725 if (tp->t_dupacks >= tcprexmtthresh) { 1726 if (is_partialack) { 1727 tcp_newreno_partial_ack(tp, th); 1728 } else { 1729 /* 1730 * Window inflation should have left us 1731 * with approximately snd_ssthresh 1732 * outstanding data. 1733 * But in case we would be inclined to 1734 * send a burst, better to do it via 1735 * the slow start mechanism. 1736 */ 1737 if (SEQ_GT(th->th_ack + 1738 tp->snd_ssthresh, 1739 tp->snd_max)) 1740 tp->snd_cwnd = tp->snd_max - 1741 th->th_ack + 1742 tp->t_maxseg; 1743 else 1744 tp->snd_cwnd = tp->snd_ssthresh; 1745 } 1746 } 1747 /* 1748 * Reset dupacks, except on partial acks in 1749 * fast recovery. 1750 */ 1751 if (!(tp->t_dupacks >= tcprexmtthresh && is_partialack)) 1752 tp->t_dupacks = 0; 1753 } else { 1754 if (tp->t_dupacks >= tcprexmtthresh && 1755 tp->snd_cwnd > tp->snd_ssthresh) 1756 tp->snd_cwnd = tp->snd_ssthresh; 1757 tp->t_dupacks = 0; 1758 } 1759 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1760 tcpstat.tcps_rcvacktoomuch++; 1761 goto dropafterack; 1762 } 1763 /* 1764 * If we reach this point, ACK is not a duplicate, 1765 * i.e., it ACKs something we sent. 1766 */ 1767 if (tp->t_flags & TF_NEEDSYN) { 1768 /* 1769 * T/TCP: Connection was half-synchronized, and our 1770 * SYN has been ACK'd (so connection is now fully 1771 * synchronized). Go to non-starred state, 1772 * increment snd_una for ACK of SYN, and check if 1773 * we can do window scaling. 1774 */ 1775 tp->t_flags &= ~TF_NEEDSYN; 1776 tp->snd_una++; 1777 /* Do window scaling? */ 1778 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1779 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1780 tp->snd_scale = tp->requested_s_scale; 1781 tp->rcv_scale = tp->request_r_scale; 1782 } 1783 } 1784 1785 process_ACK: 1786 acked = th->th_ack - tp->snd_una; 1787 tcpstat.tcps_rcvackpack++; 1788 tcpstat.tcps_rcvackbyte += acked; 1789 1790 /* 1791 * If we just performed our first retransmit, and the ACK 1792 * arrives within our recovery window, then it was a mistake 1793 * to do the retransmit in the first place. Recover our 1794 * original cwnd and ssthresh, and proceed to transmit where 1795 * we left off. 1796 */ 1797 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1798 ++tcpstat.tcps_sndrexmitbad; 1799 tp->snd_cwnd = tp->snd_cwnd_prev; 1800 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1801 tp->snd_nxt = tp->snd_max; 1802 tp->t_badrxtwin = 0; /* XXX probably not required */ 1803 } 1804 1805 /* 1806 * If we have a timestamp reply, update smoothed 1807 * round trip time. If no timestamp is present but 1808 * transmit timer is running and timed sequence 1809 * number was acked, update smoothed round trip time. 1810 * Since we now have an rtt measurement, cancel the 1811 * timer backoff (cf., Phil Karn's retransmit alg.). 1812 * Recompute the initial retransmit timer. 1813 */ 1814 if (to.to_flags & TOF_TS) 1815 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1816 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1817 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1818 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1819 1820 /* 1821 * If all outstanding data is acked, stop retransmit 1822 * timer and remember to restart (more output or persist). 1823 * If there is more data to be acked, restart retransmit 1824 * timer, using current (possibly backed-off) value. 1825 */ 1826 if (th->th_ack == tp->snd_max) { 1827 callout_stop(tp->tt_rexmt); 1828 needoutput = 1; 1829 } else if (!callout_active(tp->tt_persist)) 1830 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1831 tcp_timer_rexmt, tp); 1832 1833 /* 1834 * If no data (only SYN) was ACK'd, 1835 * skip rest of ACK processing. 1836 */ 1837 if (acked == 0) 1838 goto step6; 1839 1840 /* 1841 * When new data is acked, open the congestion window. 1842 * If the window gives us less than ssthresh packets 1843 * in flight, open exponentially (maxseg per packet). 1844 * Otherwise open linearly: maxseg per window 1845 * (maxseg^2 / cwnd per packet). 1846 */ 1847 { 1848 register u_int cw = tp->snd_cwnd; 1849 register u_int incr = tp->t_maxseg; 1850 1851 if (cw > tp->snd_ssthresh) 1852 incr = incr * incr / cw; 1853 /* 1854 * If t_dupacks != 0 here, it indicates that we are still 1855 * in NewReno fast recovery mode, so we leave the congestion 1856 * window alone. 1857 */ 1858 if (!tcp_do_newreno || tp->t_dupacks == 0) 1859 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1860 } 1861 if (acked > so->so_snd.sb_cc) { 1862 tp->snd_wnd -= so->so_snd.sb_cc; 1863 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1864 ourfinisacked = 1; 1865 } else { 1866 sbdrop(&so->so_snd, acked); 1867 tp->snd_wnd -= acked; 1868 ourfinisacked = 0; 1869 } 1870 sowwakeup(so); 1871 tp->snd_una = th->th_ack; 1872 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1873 tp->snd_nxt = tp->snd_una; 1874 1875 switch (tp->t_state) { 1876 1877 /* 1878 * In FIN_WAIT_1 STATE in addition to the processing 1879 * for the ESTABLISHED state if our FIN is now acknowledged 1880 * then enter FIN_WAIT_2. 1881 */ 1882 case TCPS_FIN_WAIT_1: 1883 if (ourfinisacked) { 1884 /* 1885 * If we can't receive any more 1886 * data, then closing user can proceed. 1887 * Starting the timer is contrary to the 1888 * specification, but if we don't get a FIN 1889 * we'll hang forever. 1890 */ 1891 if (so->so_state & SS_CANTRCVMORE) { 1892 soisdisconnected(so); 1893 callout_reset(tp->tt_2msl, tcp_maxidle, 1894 tcp_timer_2msl, tp); 1895 } 1896 tp->t_state = TCPS_FIN_WAIT_2; 1897 } 1898 break; 1899 1900 /* 1901 * In CLOSING STATE in addition to the processing for 1902 * the ESTABLISHED state if the ACK acknowledges our FIN 1903 * then enter the TIME-WAIT state, otherwise ignore 1904 * the segment. 1905 */ 1906 case TCPS_CLOSING: 1907 if (ourfinisacked) { 1908 tp->t_state = TCPS_TIME_WAIT; 1909 tcp_canceltimers(tp); 1910 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1911 if (tp->cc_recv != 0 && 1912 (ticks - tp->t_starttime) < tcp_msl) 1913 callout_reset(tp->tt_2msl, 1914 tp->t_rxtcur * 1915 TCPTV_TWTRUNC, 1916 tcp_timer_2msl, tp); 1917 else 1918 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1919 tcp_timer_2msl, tp); 1920 soisdisconnected(so); 1921 } 1922 break; 1923 1924 /* 1925 * In LAST_ACK, we may still be waiting for data to drain 1926 * and/or to be acked, as well as for the ack of our FIN. 1927 * If our FIN is now acknowledged, delete the TCB, 1928 * enter the closed state and return. 1929 */ 1930 case TCPS_LAST_ACK: 1931 if (ourfinisacked) { 1932 tp = tcp_close(tp); 1933 goto drop; 1934 } 1935 break; 1936 1937 /* 1938 * In TIME_WAIT state the only thing that should arrive 1939 * is a retransmission of the remote FIN. Acknowledge 1940 * it and restart the finack timer. 1941 */ 1942 case TCPS_TIME_WAIT: 1943 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1944 tcp_timer_2msl, tp); 1945 goto dropafterack; 1946 } 1947 } 1948 1949 step6: 1950 /* 1951 * Update window information. 1952 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1953 */ 1954 if ((thflags & TH_ACK) && 1955 (SEQ_LT(tp->snd_wl1, th->th_seq) || 1956 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 1957 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1958 /* keep track of pure window updates */ 1959 if (tlen == 0 && 1960 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1961 tcpstat.tcps_rcvwinupd++; 1962 tp->snd_wnd = tiwin; 1963 tp->snd_wl1 = th->th_seq; 1964 tp->snd_wl2 = th->th_ack; 1965 if (tp->snd_wnd > tp->max_sndwnd) 1966 tp->max_sndwnd = tp->snd_wnd; 1967 needoutput = 1; 1968 } 1969 1970 /* 1971 * Process segments with URG. 1972 */ 1973 if ((thflags & TH_URG) && th->th_urp && 1974 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1975 /* 1976 * This is a kludge, but if we receive and accept 1977 * random urgent pointers, we'll crash in 1978 * soreceive. It's hard to imagine someone 1979 * actually wanting to send this much urgent data. 1980 */ 1981 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1982 th->th_urp = 0; /* XXX */ 1983 thflags &= ~TH_URG; /* XXX */ 1984 goto dodata; /* XXX */ 1985 } 1986 /* 1987 * If this segment advances the known urgent pointer, 1988 * then mark the data stream. This should not happen 1989 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1990 * a FIN has been received from the remote side. 1991 * In these states we ignore the URG. 1992 * 1993 * According to RFC961 (Assigned Protocols), 1994 * the urgent pointer points to the last octet 1995 * of urgent data. We continue, however, 1996 * to consider it to indicate the first octet 1997 * of data past the urgent section as the original 1998 * spec states (in one of two places). 1999 */ 2000 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2001 tp->rcv_up = th->th_seq + th->th_urp; 2002 so->so_oobmark = so->so_rcv.sb_cc + 2003 (tp->rcv_up - tp->rcv_nxt) - 1; 2004 if (so->so_oobmark == 0) 2005 so->so_state |= SS_RCVATMARK; 2006 sohasoutofband(so); 2007 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2008 } 2009 /* 2010 * Remove out of band data so doesn't get presented to user. 2011 * This can happen independent of advancing the URG pointer, 2012 * but if two URG's are pending at once, some out-of-band 2013 * data may creep in... ick. 2014 */ 2015 if (th->th_urp <= (u_long)tlen 2016 #ifdef SO_OOBINLINE 2017 && (so->so_options & SO_OOBINLINE) == 0 2018 #endif 2019 ) 2020 tcp_pulloutofband(so, th, m, 2021 drop_hdrlen); /* hdr drop is delayed */ 2022 } else { 2023 /* 2024 * If no out of band data is expected, 2025 * pull receive urgent pointer along 2026 * with the receive window. 2027 */ 2028 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2029 tp->rcv_up = tp->rcv_nxt; 2030 } 2031 dodata: /* XXX */ 2032 KASSERT(headlocked, ("headlocked")); 2033 INP_INFO_WUNLOCK(&tcbinfo); 2034 headlocked = 0; 2035 /* 2036 * Process the segment text, merging it into the TCP sequencing queue, 2037 * and arranging for acknowledgment of receipt if necessary. 2038 * This process logically involves adjusting tp->rcv_wnd as data 2039 * is presented to the user (this happens in tcp_usrreq.c, 2040 * case PRU_RCVD). If a FIN has already been received on this 2041 * connection then we just ignore the text. 2042 */ 2043 if ((tlen || (thflags & TH_FIN)) && 2044 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2045 m_adj(m, drop_hdrlen); /* delayed header drop */ 2046 /* 2047 * Insert segment which includes th into TCP reassembly queue 2048 * with control block tp. Set thflags to whether reassembly now 2049 * includes a segment with FIN. This handles the common case 2050 * inline (segment is the next to be received on an established 2051 * connection, and the queue is empty), avoiding linkage into 2052 * and removal from the queue and repetition of various 2053 * conversions. 2054 * Set DELACK for segments received in order, but ack 2055 * immediately when segments are out of order (so 2056 * fast retransmit can work). 2057 */ 2058 if (th->th_seq == tp->rcv_nxt && 2059 LIST_EMPTY(&tp->t_segq) && 2060 TCPS_HAVEESTABLISHED(tp->t_state)) { 2061 if (DELAY_ACK(tp)) 2062 callout_reset(tp->tt_delack, tcp_delacktime, 2063 tcp_timer_delack, tp); 2064 else 2065 tp->t_flags |= TF_ACKNOW; 2066 tp->rcv_nxt += tlen; 2067 thflags = th->th_flags & TH_FIN; 2068 tcpstat.tcps_rcvpack++; 2069 tcpstat.tcps_rcvbyte += tlen; 2070 ND6_HINT(tp); 2071 sbappend(&so->so_rcv, m); 2072 sorwakeup(so); 2073 } else { 2074 thflags = tcp_reass(tp, th, &tlen, m); 2075 tp->t_flags |= TF_ACKNOW; 2076 } 2077 2078 /* 2079 * Note the amount of data that peer has sent into 2080 * our window, in order to estimate the sender's 2081 * buffer size. 2082 */ 2083 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2084 } else { 2085 m_freem(m); 2086 thflags &= ~TH_FIN; 2087 } 2088 2089 /* 2090 * If FIN is received ACK the FIN and let the user know 2091 * that the connection is closing. 2092 */ 2093 if (thflags & TH_FIN) { 2094 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2095 socantrcvmore(so); 2096 /* 2097 * If connection is half-synchronized 2098 * (ie NEEDSYN flag on) then delay ACK, 2099 * so it may be piggybacked when SYN is sent. 2100 * Otherwise, since we received a FIN then no 2101 * more input can be expected, send ACK now. 2102 */ 2103 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2104 callout_reset(tp->tt_delack, tcp_delacktime, 2105 tcp_timer_delack, tp); 2106 else 2107 tp->t_flags |= TF_ACKNOW; 2108 tp->rcv_nxt++; 2109 } 2110 switch (tp->t_state) { 2111 2112 /* 2113 * In SYN_RECEIVED and ESTABLISHED STATES 2114 * enter the CLOSE_WAIT state. 2115 */ 2116 case TCPS_SYN_RECEIVED: 2117 tp->t_starttime = ticks; 2118 /*FALLTHROUGH*/ 2119 case TCPS_ESTABLISHED: 2120 tp->t_state = TCPS_CLOSE_WAIT; 2121 break; 2122 2123 /* 2124 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2125 * enter the CLOSING state. 2126 */ 2127 case TCPS_FIN_WAIT_1: 2128 tp->t_state = TCPS_CLOSING; 2129 break; 2130 2131 /* 2132 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2133 * starting the time-wait timer, turning off the other 2134 * standard timers. 2135 */ 2136 case TCPS_FIN_WAIT_2: 2137 tp->t_state = TCPS_TIME_WAIT; 2138 tcp_canceltimers(tp); 2139 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2140 if (tp->cc_recv != 0 && 2141 (ticks - tp->t_starttime) < tcp_msl) { 2142 callout_reset(tp->tt_2msl, 2143 tp->t_rxtcur * TCPTV_TWTRUNC, 2144 tcp_timer_2msl, tp); 2145 /* For transaction client, force ACK now. */ 2146 tp->t_flags |= TF_ACKNOW; 2147 } 2148 else 2149 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2150 tcp_timer_2msl, tp); 2151 soisdisconnected(so); 2152 break; 2153 2154 /* 2155 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2156 */ 2157 case TCPS_TIME_WAIT: 2158 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2159 tcp_timer_2msl, tp); 2160 break; 2161 } 2162 } 2163 #ifdef TCPDEBUG 2164 if (so->so_options & SO_DEBUG) 2165 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2166 &tcp_savetcp, 0); 2167 #endif 2168 2169 /* 2170 * Return any desired output. 2171 */ 2172 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2173 (void) tcp_output(tp); 2174 INP_UNLOCK(inp); 2175 return; 2176 2177 dropafterack: 2178 /* 2179 * Generate an ACK dropping incoming segment if it occupies 2180 * sequence space, where the ACK reflects our state. 2181 * 2182 * We can now skip the test for the RST flag since all 2183 * paths to this code happen after packets containing 2184 * RST have been dropped. 2185 * 2186 * In the SYN-RECEIVED state, don't send an ACK unless the 2187 * segment we received passes the SYN-RECEIVED ACK test. 2188 * If it fails send a RST. This breaks the loop in the 2189 * "LAND" DoS attack, and also prevents an ACK storm 2190 * between two listening ports that have been sent forged 2191 * SYN segments, each with the source address of the other. 2192 */ 2193 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2194 (SEQ_GT(tp->snd_una, th->th_ack) || 2195 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2196 rstreason = BANDLIM_RST_OPENPORT; 2197 goto dropwithreset; 2198 } 2199 #ifdef TCPDEBUG 2200 if (so->so_options & SO_DEBUG) 2201 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2202 &tcp_savetcp, 0); 2203 #endif 2204 if (headlocked) 2205 INP_INFO_WUNLOCK(&tcbinfo); 2206 m_freem(m); 2207 tp->t_flags |= TF_ACKNOW; 2208 (void) tcp_output(tp); 2209 INP_UNLOCK(inp); 2210 return; 2211 2212 dropwithreset: 2213 /* 2214 * Generate a RST, dropping incoming segment. 2215 * Make ACK acceptable to originator of segment. 2216 * Don't bother to respond if destination was broadcast/multicast. 2217 */ 2218 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2219 goto drop; 2220 if (isipv6) { 2221 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2222 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2223 goto drop; 2224 } else { 2225 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2226 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2227 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2228 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2229 goto drop; 2230 } 2231 /* IPv6 anycast check is done at tcp6_input() */ 2232 2233 /* 2234 * Perform bandwidth limiting. 2235 */ 2236 if (badport_bandlim(rstreason) < 0) 2237 goto drop; 2238 2239 #ifdef TCPDEBUG 2240 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2241 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2242 &tcp_savetcp, 0); 2243 #endif 2244 2245 if (tp) 2246 INP_UNLOCK(inp); 2247 2248 if (thflags & TH_ACK) 2249 /* mtod() below is safe as long as hdr dropping is delayed */ 2250 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2251 TH_RST); 2252 else { 2253 if (thflags & TH_SYN) 2254 tlen++; 2255 /* mtod() below is safe as long as hdr dropping is delayed */ 2256 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2257 (tcp_seq)0, TH_RST|TH_ACK); 2258 } 2259 if (headlocked) 2260 INP_INFO_WUNLOCK(&tcbinfo); 2261 return; 2262 2263 drop: 2264 /* 2265 * Drop space held by incoming segment and return. 2266 */ 2267 #ifdef TCPDEBUG 2268 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2269 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2270 &tcp_savetcp, 0); 2271 #endif 2272 if (tp) 2273 INP_UNLOCK(inp); 2274 m_freem(m); 2275 if (headlocked) 2276 INP_INFO_WUNLOCK(&tcbinfo); 2277 return; 2278 } 2279 2280 /* 2281 * Parse TCP options and place in tcpopt. 2282 */ 2283 static void 2284 tcp_dooptions(to, cp, cnt, is_syn) 2285 struct tcpopt *to; 2286 u_char *cp; 2287 int cnt; 2288 { 2289 int opt, optlen; 2290 2291 to->to_flags = 0; 2292 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2293 opt = cp[0]; 2294 if (opt == TCPOPT_EOL) 2295 break; 2296 if (opt == TCPOPT_NOP) 2297 optlen = 1; 2298 else { 2299 if (cnt < 2) 2300 break; 2301 optlen = cp[1]; 2302 if (optlen < 2 || optlen > cnt) 2303 break; 2304 } 2305 switch (opt) { 2306 case TCPOPT_MAXSEG: 2307 if (optlen != TCPOLEN_MAXSEG) 2308 continue; 2309 if (!is_syn) 2310 continue; 2311 to->to_flags |= TOF_MSS; 2312 bcopy((char *)cp + 2, 2313 (char *)&to->to_mss, sizeof(to->to_mss)); 2314 to->to_mss = ntohs(to->to_mss); 2315 break; 2316 case TCPOPT_WINDOW: 2317 if (optlen != TCPOLEN_WINDOW) 2318 continue; 2319 if (! is_syn) 2320 continue; 2321 to->to_flags |= TOF_SCALE; 2322 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2323 break; 2324 case TCPOPT_TIMESTAMP: 2325 if (optlen != TCPOLEN_TIMESTAMP) 2326 continue; 2327 to->to_flags |= TOF_TS; 2328 bcopy((char *)cp + 2, 2329 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2330 to->to_tsval = ntohl(to->to_tsval); 2331 bcopy((char *)cp + 6, 2332 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2333 to->to_tsecr = ntohl(to->to_tsecr); 2334 break; 2335 case TCPOPT_CC: 2336 if (optlen != TCPOLEN_CC) 2337 continue; 2338 to->to_flags |= TOF_CC; 2339 bcopy((char *)cp + 2, 2340 (char *)&to->to_cc, sizeof(to->to_cc)); 2341 to->to_cc = ntohl(to->to_cc); 2342 break; 2343 case TCPOPT_CCNEW: 2344 if (optlen != TCPOLEN_CC) 2345 continue; 2346 if (!is_syn) 2347 continue; 2348 to->to_flags |= TOF_CCNEW; 2349 bcopy((char *)cp + 2, 2350 (char *)&to->to_cc, sizeof(to->to_cc)); 2351 to->to_cc = ntohl(to->to_cc); 2352 break; 2353 case TCPOPT_CCECHO: 2354 if (optlen != TCPOLEN_CC) 2355 continue; 2356 if (!is_syn) 2357 continue; 2358 to->to_flags |= TOF_CCECHO; 2359 bcopy((char *)cp + 2, 2360 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2361 to->to_ccecho = ntohl(to->to_ccecho); 2362 break; 2363 default: 2364 continue; 2365 } 2366 } 2367 } 2368 2369 /* 2370 * Pull out of band byte out of a segment so 2371 * it doesn't appear in the user's data queue. 2372 * It is still reflected in the segment length for 2373 * sequencing purposes. 2374 */ 2375 static void 2376 tcp_pulloutofband(so, th, m, off) 2377 struct socket *so; 2378 struct tcphdr *th; 2379 register struct mbuf *m; 2380 int off; /* delayed to be droped hdrlen */ 2381 { 2382 int cnt = off + th->th_urp - 1; 2383 2384 while (cnt >= 0) { 2385 if (m->m_len > cnt) { 2386 char *cp = mtod(m, caddr_t) + cnt; 2387 struct tcpcb *tp = sototcpcb(so); 2388 2389 tp->t_iobc = *cp; 2390 tp->t_oobflags |= TCPOOB_HAVEDATA; 2391 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2392 m->m_len--; 2393 if (m->m_flags & M_PKTHDR) 2394 m->m_pkthdr.len--; 2395 return; 2396 } 2397 cnt -= m->m_len; 2398 m = m->m_next; 2399 if (m == 0) 2400 break; 2401 } 2402 panic("tcp_pulloutofband"); 2403 } 2404 2405 /* 2406 * Collect new round-trip time estimate 2407 * and update averages and current timeout. 2408 */ 2409 static void 2410 tcp_xmit_timer(tp, rtt) 2411 register struct tcpcb *tp; 2412 int rtt; 2413 { 2414 register int delta; 2415 2416 tcpstat.tcps_rttupdated++; 2417 tp->t_rttupdated++; 2418 if (tp->t_srtt != 0) { 2419 /* 2420 * srtt is stored as fixed point with 5 bits after the 2421 * binary point (i.e., scaled by 8). The following magic 2422 * is equivalent to the smoothing algorithm in rfc793 with 2423 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2424 * point). Adjust rtt to origin 0. 2425 */ 2426 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2427 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2428 2429 if ((tp->t_srtt += delta) <= 0) 2430 tp->t_srtt = 1; 2431 2432 /* 2433 * We accumulate a smoothed rtt variance (actually, a 2434 * smoothed mean difference), then set the retransmit 2435 * timer to smoothed rtt + 4 times the smoothed variance. 2436 * rttvar is stored as fixed point with 4 bits after the 2437 * binary point (scaled by 16). The following is 2438 * equivalent to rfc793 smoothing with an alpha of .75 2439 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2440 * rfc793's wired-in beta. 2441 */ 2442 if (delta < 0) 2443 delta = -delta; 2444 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2445 if ((tp->t_rttvar += delta) <= 0) 2446 tp->t_rttvar = 1; 2447 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2448 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2449 } else { 2450 /* 2451 * No rtt measurement yet - use the unsmoothed rtt. 2452 * Set the variance to half the rtt (so our first 2453 * retransmit happens at 3*rtt). 2454 */ 2455 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2456 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2457 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2458 } 2459 tp->t_rtttime = 0; 2460 tp->t_rxtshift = 0; 2461 2462 /* 2463 * the retransmit should happen at rtt + 4 * rttvar. 2464 * Because of the way we do the smoothing, srtt and rttvar 2465 * will each average +1/2 tick of bias. When we compute 2466 * the retransmit timer, we want 1/2 tick of rounding and 2467 * 1 extra tick because of +-1/2 tick uncertainty in the 2468 * firing of the timer. The bias will give us exactly the 2469 * 1.5 tick we need. But, because the bias is 2470 * statistical, we have to test that we don't drop below 2471 * the minimum feasible timer (which is 2 ticks). 2472 */ 2473 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2474 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2475 2476 /* 2477 * We received an ack for a packet that wasn't retransmitted; 2478 * it is probably safe to discard any error indications we've 2479 * received recently. This isn't quite right, but close enough 2480 * for now (a route might have failed after we sent a segment, 2481 * and the return path might not be symmetrical). 2482 */ 2483 tp->t_softerror = 0; 2484 } 2485 2486 /* 2487 * Determine a reasonable value for maxseg size. 2488 * If the route is known, check route for mtu. 2489 * If none, use an mss that can be handled on the outgoing 2490 * interface without forcing IP to fragment; if bigger than 2491 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2492 * to utilize large mbufs. If no route is found, route has no mtu, 2493 * or the destination isn't local, use a default, hopefully conservative 2494 * size (usually 512 or the default IP max size, but no more than the mtu 2495 * of the interface), as we can't discover anything about intervening 2496 * gateways or networks. We also initialize the congestion/slow start 2497 * window to be a single segment if the destination isn't local. 2498 * While looking at the routing entry, we also initialize other path-dependent 2499 * parameters from pre-set or cached values in the routing entry. 2500 * 2501 * Also take into account the space needed for options that we 2502 * send regularly. Make maxseg shorter by that amount to assure 2503 * that we can send maxseg amount of data even when the options 2504 * are present. Store the upper limit of the length of options plus 2505 * data in maxopd. 2506 * 2507 * NOTE that this routine is only called when we process an incoming 2508 * segment, for outgoing segments only tcp_mssopt is called. 2509 * 2510 * In case of T/TCP, we call this routine during implicit connection 2511 * setup as well (offer = -1), to initialize maxseg from the cached 2512 * MSS of our peer. 2513 */ 2514 void 2515 tcp_mss(tp, offer) 2516 struct tcpcb *tp; 2517 int offer; 2518 { 2519 register struct rtentry *rt; 2520 struct ifnet *ifp; 2521 register int rtt, mss; 2522 u_long bufsize; 2523 struct inpcb *inp = tp->t_inpcb; 2524 struct socket *so; 2525 struct rmxp_tao *taop; 2526 int origoffer = offer; 2527 #ifdef INET6 2528 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2529 size_t min_protoh = isipv6 ? 2530 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2531 sizeof (struct tcpiphdr); 2532 #else 2533 const int isipv6 = 0; 2534 const size_t min_protoh = sizeof (struct tcpiphdr); 2535 #endif 2536 2537 if (isipv6) 2538 rt = tcp_rtlookup6(&inp->inp_inc); 2539 else 2540 rt = tcp_rtlookup(&inp->inp_inc); 2541 if (rt == NULL) { 2542 tp->t_maxopd = tp->t_maxseg = 2543 isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2544 return; 2545 } 2546 ifp = rt->rt_ifp; 2547 so = inp->inp_socket; 2548 2549 taop = rmx_taop(rt->rt_rmx); 2550 /* 2551 * Offer == -1 means that we didn't receive SYN yet, 2552 * use cached value in that case; 2553 */ 2554 if (offer == -1) 2555 offer = taop->tao_mssopt; 2556 /* 2557 * Offer == 0 means that there was no MSS on the SYN segment, 2558 * in this case we use tcp_mssdflt. 2559 */ 2560 if (offer == 0) 2561 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2562 else 2563 /* 2564 * Sanity check: make sure that maxopd will be large 2565 * enough to allow some data on segments even is the 2566 * all the option space is used (40bytes). Otherwise 2567 * funny things may happen in tcp_output. 2568 */ 2569 offer = max(offer, 64); 2570 taop->tao_mssopt = offer; 2571 2572 /* 2573 * While we're here, check if there's an initial rtt 2574 * or rttvar. Convert from the route-table units 2575 * to scaled multiples of the slow timeout timer. 2576 */ 2577 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2578 /* 2579 * XXX the lock bit for RTT indicates that the value 2580 * is also a minimum value; this is subject to time. 2581 */ 2582 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2583 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2584 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2585 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2586 tcpstat.tcps_usedrtt++; 2587 if (rt->rt_rmx.rmx_rttvar) { 2588 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2589 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2590 tcpstat.tcps_usedrttvar++; 2591 } else { 2592 /* default variation is +- 1 rtt */ 2593 tp->t_rttvar = 2594 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2595 } 2596 TCPT_RANGESET(tp->t_rxtcur, 2597 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2598 tp->t_rttmin, TCPTV_REXMTMAX); 2599 } 2600 /* 2601 * if there's an mtu associated with the route, use it 2602 * else, use the link mtu. 2603 */ 2604 if (rt->rt_rmx.rmx_mtu) 2605 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2606 else { 2607 if (isipv6) { 2608 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2609 min_protoh; 2610 if (!in6_localaddr(&inp->in6p_faddr)) 2611 mss = min(mss, tcp_v6mssdflt); 2612 } else { 2613 mss = ifp->if_mtu - min_protoh; 2614 if (!in_localaddr(inp->inp_faddr)) 2615 mss = min(mss, tcp_mssdflt); 2616 } 2617 } 2618 mss = min(mss, offer); 2619 /* 2620 * maxopd stores the maximum length of data AND options 2621 * in a segment; maxseg is the amount of data in a normal 2622 * segment. We need to store this value (maxopd) apart 2623 * from maxseg, because now every segment carries options 2624 * and thus we normally have somewhat less data in segments. 2625 */ 2626 tp->t_maxopd = mss; 2627 2628 /* 2629 * In case of T/TCP, origoffer==-1 indicates, that no segments 2630 * were received yet. In this case we just guess, otherwise 2631 * we do the same as before T/TCP. 2632 */ 2633 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2634 (origoffer == -1 || 2635 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2636 mss -= TCPOLEN_TSTAMP_APPA; 2637 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2638 (origoffer == -1 || 2639 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2640 mss -= TCPOLEN_CC_APPA; 2641 2642 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2643 if (mss > MCLBYTES) 2644 mss &= ~(MCLBYTES-1); 2645 #else 2646 if (mss > MCLBYTES) 2647 mss = mss / MCLBYTES * MCLBYTES; 2648 #endif 2649 /* 2650 * If there's a pipesize, change the socket buffer 2651 * to that size. Make the socket buffers an integral 2652 * number of mss units; if the mss is larger than 2653 * the socket buffer, decrease the mss. 2654 */ 2655 #ifdef RTV_SPIPE 2656 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2657 #endif 2658 bufsize = so->so_snd.sb_hiwat; 2659 if (bufsize < mss) 2660 mss = bufsize; 2661 else { 2662 bufsize = roundup(bufsize, mss); 2663 if (bufsize > sb_max) 2664 bufsize = sb_max; 2665 if (bufsize > so->so_snd.sb_hiwat) 2666 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2667 } 2668 tp->t_maxseg = mss; 2669 2670 #ifdef RTV_RPIPE 2671 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2672 #endif 2673 bufsize = so->so_rcv.sb_hiwat; 2674 if (bufsize > mss) { 2675 bufsize = roundup(bufsize, mss); 2676 if (bufsize > sb_max) 2677 bufsize = sb_max; 2678 if (bufsize > so->so_rcv.sb_hiwat) 2679 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2680 } 2681 2682 /* 2683 * Set the slow-start flight size depending on whether this 2684 * is a local network or not. 2685 */ 2686 if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2687 (!isipv6 && in_localaddr(inp->inp_faddr))) 2688 tp->snd_cwnd = mss * ss_fltsz_local; 2689 else 2690 tp->snd_cwnd = mss * ss_fltsz; 2691 2692 if (rt->rt_rmx.rmx_ssthresh) { 2693 /* 2694 * There's some sort of gateway or interface 2695 * buffer limit on the path. Use this to set 2696 * the slow start threshhold, but set the 2697 * threshold to no less than 2*mss. 2698 */ 2699 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2700 tcpstat.tcps_usedssthresh++; 2701 } 2702 } 2703 2704 /* 2705 * Determine the MSS option to send on an outgoing SYN. 2706 */ 2707 int 2708 tcp_mssopt(tp) 2709 struct tcpcb *tp; 2710 { 2711 struct rtentry *rt; 2712 #ifdef INET6 2713 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2714 size_t min_protoh = isipv6 ? 2715 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2716 sizeof (struct tcpiphdr); 2717 #else 2718 const int isipv6 = 0; 2719 const size_t min_protoh = sizeof (struct tcpiphdr); 2720 #endif 2721 2722 if (isipv6) 2723 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2724 else 2725 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2726 if (rt == NULL) 2727 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2728 2729 return (rt->rt_ifp->if_mtu - min_protoh); 2730 } 2731 2732 2733 /* 2734 * On a partial ack arrives, force the retransmission of the 2735 * next unacknowledged segment. Do not clear tp->t_dupacks. 2736 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2737 * be started again. 2738 */ 2739 static void 2740 tcp_newreno_partial_ack(tp, th) 2741 struct tcpcb *tp; 2742 struct tcphdr *th; 2743 { 2744 tcp_seq onxt = tp->snd_nxt; 2745 u_long ocwnd = tp->snd_cwnd; 2746 2747 callout_stop(tp->tt_rexmt); 2748 tp->t_rtttime = 0; 2749 tp->snd_nxt = th->th_ack; 2750 /* 2751 * Set snd_cwnd to one segment beyond acknowledged offset. 2752 * (tp->snd_una has not yet been updated when this function is called.) 2753 */ 2754 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2755 (void) tcp_output(tp); 2756 tp->snd_cwnd = ocwnd; 2757 if (SEQ_GT(onxt, tp->snd_nxt)) 2758 tp->snd_nxt = onxt; 2759 /* 2760 * Partial window deflation. Relies on fact that tp->snd_una 2761 * not updated yet. 2762 */ 2763 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2764 } 2765