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_tcpdebug.h" 41 #include "opt_tcp_input.h" 42 43 #include <sys/param.h> 44 #include <sys/kernel.h> 45 #include <sys/malloc.h> 46 #include <sys/mbuf.h> 47 #include <sys/proc.h> /* for proc0 declaration */ 48 #include <sys/protosw.h> 49 #include <sys/signalvar.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 #include <sys/sysctl.h> 53 #include <sys/syslog.h> 54 #include <sys/systm.h> 55 56 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 57 58 #include <net/if.h> 59 #include <net/route.h> 60 61 #include <netinet/in.h> 62 #include <netinet/in_pcb.h> 63 #include <netinet/in_systm.h> 64 #include <netinet/in_var.h> 65 #include <netinet/ip.h> 66 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 67 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 68 #include <netinet/ip_var.h> 69 #ifdef INET6 70 #include <netinet/ip6.h> 71 #include <netinet/icmp6.h> 72 #include <netinet6/in6_pcb.h> 73 #include <netinet6/ip6_var.h> 74 #include <netinet6/nd6.h> 75 #endif 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 #ifdef INET6 82 #include <netinet6/tcp6_var.h> 83 #endif 84 #include <netinet/tcpip.h> 85 #ifdef TCPDEBUG 86 #include <netinet/tcp_debug.h> 87 88 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 89 struct tcphdr tcp_savetcp; 90 #endif /* TCPDEBUG */ 91 92 #ifdef IPSEC 93 #include <netinet6/ipsec.h> 94 #ifdef INET6 95 #include <netinet6/ipsec6.h> 96 #endif 97 #include <netkey/key.h> 98 #endif /*IPSEC*/ 99 100 #include <machine/in_cksum.h> 101 102 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 103 104 static int tcprexmtthresh = 3; 105 tcp_cc tcp_ccgen; 106 107 struct tcpstat tcpstat; 108 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 109 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 110 111 static int log_in_vain = 0; 112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 113 &log_in_vain, 0, "Log all incoming TCP connections"); 114 115 static int blackhole = 0; 116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 117 &blackhole, 0, "Do not send RST when dropping refused connections"); 118 119 int tcp_delack_enabled = 1; 120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 121 &tcp_delack_enabled, 0, 122 "Delay ACK to try and piggyback it onto a data packet"); 123 124 #ifdef TCP_DROP_SYNFIN 125 static int drop_synfin = 0; 126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 127 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 128 #endif 129 130 struct inpcbhead tcb; 131 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 132 struct inpcbinfo tcbinfo; 133 struct mtx *tcbinfo_mtx; 134 135 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 136 static void tcp_pulloutofband(struct socket *, 137 struct tcphdr *, struct mbuf *, int); 138 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 139 struct mbuf *); 140 static void tcp_xmit_timer(struct tcpcb *, int); 141 static int tcp_newreno(struct tcpcb *, struct tcphdr *); 142 143 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 144 #ifdef INET6 145 #define ND6_HINT(tp) \ 146 do { \ 147 if ((tp) && (tp)->t_inpcb && \ 148 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 149 (tp)->t_inpcb->in6p_route.ro_rt) \ 150 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 151 } while (0) 152 #else 153 #define ND6_HINT(tp) 154 #endif 155 156 /* 157 * Indicate whether this ack should be delayed. We can delay the ack if 158 * - delayed acks are enabled and 159 * - there is no delayed ack timer in progress and 160 * - our last ack wasn't a 0-sized window. We never want to delay 161 * the ack that opens up a 0-sized window. 162 */ 163 #define DELAY_ACK(tp) \ 164 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 165 (tp->t_flags & TF_RXWIN0SENT) == 0) 166 167 static int 168 tcp_reass(tp, th, tlenp, m) 169 register struct tcpcb *tp; 170 register struct tcphdr *th; 171 int *tlenp; 172 struct mbuf *m; 173 { 174 struct tseg_qent *q; 175 struct tseg_qent *p = NULL; 176 struct tseg_qent *nq; 177 struct tseg_qent *te; 178 struct socket *so = tp->t_inpcb->inp_socket; 179 int flags; 180 181 /* 182 * Call with th==0 after become established to 183 * force pre-ESTABLISHED data up to user socket. 184 */ 185 if (th == 0) 186 goto present; 187 188 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 189 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 190 M_NOWAIT); 191 if (te == NULL) { 192 tcpstat.tcps_rcvmemdrop++; 193 m_freem(m); 194 return (0); 195 } 196 197 /* 198 * Find a segment which begins after this one does. 199 */ 200 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 201 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 202 break; 203 p = q; 204 } 205 206 /* 207 * If there is a preceding segment, it may provide some of 208 * our data already. If so, drop the data from the incoming 209 * segment. If it provides all of our data, drop us. 210 */ 211 if (p != NULL) { 212 register int i; 213 /* conversion to int (in i) handles seq wraparound */ 214 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 215 if (i > 0) { 216 if (i >= *tlenp) { 217 tcpstat.tcps_rcvduppack++; 218 tcpstat.tcps_rcvdupbyte += *tlenp; 219 m_freem(m); 220 FREE(te, M_TSEGQ); 221 /* 222 * Try to present any queued data 223 * at the left window edge to the user. 224 * This is needed after the 3-WHS 225 * completes. 226 */ 227 goto present; /* ??? */ 228 } 229 m_adj(m, i); 230 *tlenp -= i; 231 th->th_seq += i; 232 } 233 } 234 tcpstat.tcps_rcvoopack++; 235 tcpstat.tcps_rcvoobyte += *tlenp; 236 237 /* 238 * While we overlap succeeding segments trim them or, 239 * if they are completely covered, dequeue them. 240 */ 241 while (q) { 242 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 243 if (i <= 0) 244 break; 245 if (i < q->tqe_len) { 246 q->tqe_th->th_seq += i; 247 q->tqe_len -= i; 248 m_adj(q->tqe_m, i); 249 break; 250 } 251 252 nq = LIST_NEXT(q, tqe_q); 253 LIST_REMOVE(q, tqe_q); 254 m_freem(q->tqe_m); 255 FREE(q, M_TSEGQ); 256 q = nq; 257 } 258 259 /* Insert the new segment queue entry into place. */ 260 te->tqe_m = m; 261 te->tqe_th = th; 262 te->tqe_len = *tlenp; 263 264 if (p == NULL) { 265 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 266 } else { 267 LIST_INSERT_AFTER(p, te, tqe_q); 268 } 269 270 present: 271 /* 272 * Present data to user, advancing rcv_nxt through 273 * completed sequence space. 274 */ 275 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 276 return (0); 277 q = LIST_FIRST(&tp->t_segq); 278 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 279 return (0); 280 do { 281 tp->rcv_nxt += q->tqe_len; 282 flags = q->tqe_th->th_flags & TH_FIN; 283 nq = LIST_NEXT(q, tqe_q); 284 LIST_REMOVE(q, tqe_q); 285 if (so->so_state & SS_CANTRCVMORE) 286 m_freem(q->tqe_m); 287 else 288 sbappend(&so->so_rcv, q->tqe_m); 289 FREE(q, M_TSEGQ); 290 q = nq; 291 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 292 ND6_HINT(tp); 293 sorwakeup(so); 294 return (flags); 295 } 296 297 /* 298 * TCP input routine, follows pages 65-76 of the 299 * protocol specification dated September, 1981 very closely. 300 */ 301 #ifdef INET6 302 int 303 tcp6_input(mp, offp, proto) 304 struct mbuf **mp; 305 int *offp, proto; 306 { 307 register struct mbuf *m = *mp; 308 struct in6_ifaddr *ia6; 309 310 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 311 312 /* 313 * draft-itojun-ipv6-tcp-to-anycast 314 * better place to put this in? 315 */ 316 ia6 = ip6_getdstifaddr(m); 317 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 318 struct ip6_hdr *ip6; 319 320 ip6 = mtod(m, struct ip6_hdr *); 321 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 322 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 323 return IPPROTO_DONE; 324 } 325 326 tcp_input(m, *offp); 327 return IPPROTO_DONE; 328 } 329 #endif 330 331 void 332 tcp_input(m, off0) 333 register struct mbuf *m; 334 int off0; 335 { 336 register struct tcphdr *th; 337 register struct ip *ip = NULL; 338 register struct ipovly *ipov; 339 register struct inpcb *inp = NULL; 340 u_char *optp = NULL; 341 int optlen = 0; 342 int len, tlen, off; 343 int drop_hdrlen; 344 register struct tcpcb *tp = 0; 345 register int thflags; 346 struct socket *so = 0; 347 int todrop, acked, ourfinisacked, needoutput = 0; 348 u_long tiwin; 349 struct tcpopt to; /* options in this segment */ 350 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 351 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 352 int headlocked = 0; 353 354 #ifdef TCPDEBUG 355 short ostate = 0; 356 #endif 357 #ifdef INET6 358 struct ip6_hdr *ip6 = NULL; 359 int isipv6; 360 #endif /* INET6 */ 361 int rstreason; /* For badport_bandlim accounting purposes */ 362 363 #ifdef INET6 364 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 365 #endif 366 bzero((char *)&to, sizeof(to)); 367 368 tcpstat.tcps_rcvtotal++; 369 370 #ifdef INET6 371 if (isipv6) { 372 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 373 ip6 = mtod(m, struct ip6_hdr *); 374 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 375 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 376 tcpstat.tcps_rcvbadsum++; 377 goto drop; 378 } 379 th = (struct tcphdr *)((caddr_t)ip6 + off0); 380 381 /* 382 * Be proactive about unspecified IPv6 address in source. 383 * As we use all-zero to indicate unbounded/unconnected pcb, 384 * unspecified IPv6 address can be used to confuse us. 385 * 386 * Note that packets with unspecified IPv6 destination is 387 * already dropped in ip6_input. 388 */ 389 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 390 /* XXX stat */ 391 goto drop; 392 } 393 } else 394 #endif /* INET6 */ 395 { 396 /* 397 * Get IP and TCP header together in first mbuf. 398 * Note: IP leaves IP header in first mbuf. 399 */ 400 if (off0 > sizeof (struct ip)) { 401 ip_stripoptions(m, (struct mbuf *)0); 402 off0 = sizeof(struct ip); 403 } 404 if (m->m_len < sizeof (struct tcpiphdr)) { 405 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 406 tcpstat.tcps_rcvshort++; 407 return; 408 } 409 } 410 ip = mtod(m, struct ip *); 411 ipov = (struct ipovly *)ip; 412 th = (struct tcphdr *)((caddr_t)ip + off0); 413 tlen = ip->ip_len; 414 415 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 416 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 417 th->th_sum = m->m_pkthdr.csum_data; 418 else 419 th->th_sum = in_pseudo(ip->ip_src.s_addr, 420 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + 421 ip->ip_len + IPPROTO_TCP)); 422 th->th_sum ^= 0xffff; 423 } else { 424 /* 425 * Checksum extended TCP header and data. 426 */ 427 len = sizeof (struct ip) + tlen; 428 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 429 ipov->ih_len = (u_short)tlen; 430 ipov->ih_len = htons(ipov->ih_len); 431 th->th_sum = in_cksum(m, len); 432 } 433 if (th->th_sum) { 434 tcpstat.tcps_rcvbadsum++; 435 goto drop; 436 } 437 #ifdef INET6 438 /* Re-initialization for later version check */ 439 ip->ip_v = IPVERSION; 440 #endif 441 } 442 443 /* 444 * Check that TCP offset makes sense, 445 * pull out TCP options and adjust length. XXX 446 */ 447 off = th->th_off << 2; 448 if (off < sizeof (struct tcphdr) || off > tlen) { 449 tcpstat.tcps_rcvbadoff++; 450 goto drop; 451 } 452 tlen -= off; /* tlen is used instead of ti->ti_len */ 453 if (off > sizeof (struct tcphdr)) { 454 #ifdef INET6 455 if (isipv6) { 456 IP6_EXTHDR_CHECK(m, off0, off, ); 457 ip6 = mtod(m, struct ip6_hdr *); 458 th = (struct tcphdr *)((caddr_t)ip6 + off0); 459 } else 460 #endif /* INET6 */ 461 { 462 if (m->m_len < sizeof(struct ip) + off) { 463 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 464 tcpstat.tcps_rcvshort++; 465 return; 466 } 467 ip = mtod(m, struct ip *); 468 ipov = (struct ipovly *)ip; 469 th = (struct tcphdr *)((caddr_t)ip + off0); 470 } 471 } 472 optlen = off - sizeof (struct tcphdr); 473 optp = (u_char *)(th + 1); 474 } 475 thflags = th->th_flags; 476 477 #ifdef TCP_DROP_SYNFIN 478 /* 479 * If the drop_synfin option is enabled, drop all packets with 480 * both the SYN and FIN bits set. This prevents e.g. nmap from 481 * identifying the TCP/IP stack. 482 * 483 * This is a violation of the TCP specification. 484 */ 485 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 486 goto drop; 487 #endif 488 489 /* 490 * Convert TCP protocol specific fields to host format. 491 */ 492 th->th_seq = ntohl(th->th_seq); 493 th->th_ack = ntohl(th->th_ack); 494 th->th_win = ntohs(th->th_win); 495 th->th_urp = ntohs(th->th_urp); 496 497 /* 498 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 499 * until after ip6_savecontrol() is called and before other functions 500 * which don't want those proto headers. 501 * Because ip6_savecontrol() is going to parse the mbuf to 502 * search for data to be passed up to user-land, it wants mbuf 503 * parameters to be unchanged. 504 * XXX: the call of ip6_savecontrol() has been obsoleted based on 505 * latest version of the advanced API (20020110). 506 */ 507 drop_hdrlen = off0 + off; 508 509 /* 510 * Locate pcb for segment. 511 */ 512 INP_INFO_WLOCK(&tcbinfo); 513 headlocked = 1; 514 findpcb: 515 #ifdef IPFIREWALL_FORWARD 516 if (ip_fw_fwd_addr != NULL 517 #ifdef INET6 518 && isipv6 == NULL /* IPv6 support is not yet */ 519 #endif /* INET6 */ 520 ) { 521 /* 522 * Diverted. Pretend to be the destination. 523 * already got one like this? 524 */ 525 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 526 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); 527 if (!inp) { 528 /* 529 * No, then it's new. Try find the ambushing socket 530 */ 531 if (!ip_fw_fwd_addr->sin_port) { 532 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, 533 th->th_sport, ip_fw_fwd_addr->sin_addr, 534 th->th_dport, 1, m->m_pkthdr.rcvif); 535 } else { 536 inp = in_pcblookup_hash(&tcbinfo, 537 ip->ip_src, th->th_sport, 538 ip_fw_fwd_addr->sin_addr, 539 ntohs(ip_fw_fwd_addr->sin_port), 1, 540 m->m_pkthdr.rcvif); 541 } 542 } 543 ip_fw_fwd_addr = NULL; 544 } else 545 #endif /* IPFIREWALL_FORWARD */ 546 { 547 #ifdef INET6 548 if (isipv6) 549 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, 550 &ip6->ip6_dst, th->th_dport, 1, 551 m->m_pkthdr.rcvif); 552 else 553 #endif /* INET6 */ 554 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 555 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); 556 } 557 558 #ifdef IPSEC 559 #ifdef INET6 560 if (isipv6) { 561 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 562 ipsec6stat.in_polvio++; 563 goto drop; 564 } 565 } else 566 #endif /* INET6 */ 567 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 568 ipsecstat.in_polvio++; 569 goto drop; 570 } 571 #endif /*IPSEC*/ 572 573 /* 574 * If the state is CLOSED (i.e., TCB does not exist) then 575 * all data in the incoming segment is discarded. 576 * If the TCB exists but is in CLOSED state, it is embryonic, 577 * but should either do a listen or a connect soon. 578 */ 579 if (inp == NULL) { 580 if (log_in_vain) { 581 #ifdef INET6 582 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN]; 583 #else /* INET6 */ 584 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 585 #endif /* INET6 */ 586 587 #ifdef INET6 588 if (isipv6) { 589 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst)); 590 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src)); 591 } else 592 #endif 593 { 594 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 595 strcpy(sbuf, inet_ntoa(ip->ip_src)); 596 } 597 switch (log_in_vain) { 598 case 1: 599 if(thflags & TH_SYN) 600 log(LOG_INFO, 601 "Connection attempt to TCP %s:%d from %s:%d\n", 602 dbuf, ntohs(th->th_dport), 603 sbuf, 604 ntohs(th->th_sport)); 605 break; 606 case 2: 607 log(LOG_INFO, 608 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 609 dbuf, ntohs(th->th_dport), sbuf, 610 ntohs(th->th_sport), thflags); 611 break; 612 default: 613 break; 614 } 615 } 616 if (blackhole) { 617 switch (blackhole) { 618 case 1: 619 if (thflags & TH_SYN) 620 goto drop; 621 break; 622 case 2: 623 goto drop; 624 default: 625 goto drop; 626 } 627 } 628 rstreason = BANDLIM_RST_CLOSEDPORT; 629 goto dropwithreset; 630 } 631 INP_LOCK(inp); 632 tp = intotcpcb(inp); 633 if (tp == 0) { 634 INP_UNLOCK(inp); 635 rstreason = BANDLIM_RST_CLOSEDPORT; 636 goto dropwithreset; 637 } 638 if (tp->t_state == TCPS_CLOSED) 639 goto drop; 640 641 /* Unscale the window into a 32-bit value. */ 642 if ((thflags & TH_SYN) == 0) 643 tiwin = th->th_win << tp->snd_scale; 644 else 645 tiwin = th->th_win; 646 647 so = inp->inp_socket; 648 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 649 struct in_conninfo inc; 650 #ifdef TCPDEBUG 651 if (so->so_options & SO_DEBUG) { 652 ostate = tp->t_state; 653 #ifdef INET6 654 if (isipv6) 655 bcopy((char *)ip6, (char *)tcp_saveipgen, 656 sizeof(*ip6)); 657 else 658 #endif /* INET6 */ 659 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 660 tcp_savetcp = *th; 661 } 662 #endif 663 /* skip if this isn't a listen socket */ 664 if ((so->so_options & SO_ACCEPTCONN) == 0) 665 goto after_listen; 666 #ifdef INET6 667 inc.inc_isipv6 = isipv6; 668 if (isipv6) { 669 inc.inc6_faddr = ip6->ip6_src; 670 inc.inc6_laddr = ip6->ip6_dst; 671 inc.inc6_route.ro_rt = NULL; /* XXX */ 672 673 } else 674 #endif /* INET6 */ 675 { 676 inc.inc_faddr = ip->ip_src; 677 inc.inc_laddr = ip->ip_dst; 678 inc.inc_route.ro_rt = NULL; /* XXX */ 679 } 680 inc.inc_fport = th->th_sport; 681 inc.inc_lport = th->th_dport; 682 683 /* 684 * If the state is LISTEN then ignore segment if it contains 685 * a RST. If the segment contains an ACK then it is bad and 686 * send a RST. If it does not contain a SYN then it is not 687 * interesting; drop it. 688 * 689 * If the state is SYN_RECEIVED (syncache) and seg contains 690 * an ACK, but not for our SYN/ACK, send a RST. If the seg 691 * contains a RST, check the sequence number to see if it 692 * is a valid reset segment. 693 */ 694 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 695 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 696 if (!syncache_expand(&inc, th, &so, m)) { 697 /* 698 * No syncache entry, or ACK was not 699 * for our SYN/ACK. Send a RST. 700 */ 701 tcpstat.tcps_badsyn++; 702 rstreason = BANDLIM_RST_OPENPORT; 703 goto dropwithreset; 704 } 705 if (so == NULL) { 706 /* 707 * Could not complete 3-way handshake, 708 * connection is being closed down, and 709 * syncache will free mbuf. 710 */ 711 INP_UNLOCK(inp); 712 INP_INFO_WUNLOCK(&tcbinfo); 713 return; 714 } 715 /* 716 * Socket is created in state SYN_RECEIVED. 717 * Continue processing segment. 718 */ 719 INP_UNLOCK(inp); 720 inp = sotoinpcb(so); 721 INP_LOCK(inp); 722 tp = intotcpcb(inp); 723 /* 724 * This is what would have happened in 725 * tcp_output() when the SYN,ACK was sent. 726 */ 727 tp->snd_up = tp->snd_una; 728 tp->snd_max = tp->snd_nxt = tp->iss + 1; 729 tp->last_ack_sent = tp->rcv_nxt; 730 /* 731 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 732 * until the _second_ ACK is received: 733 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 734 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 735 * move to ESTAB, set snd_wnd to tiwin. 736 */ 737 tp->snd_wnd = tiwin; /* unscaled */ 738 goto after_listen; 739 } 740 if (thflags & TH_RST) { 741 syncache_chkrst(&inc, th); 742 goto drop; 743 } 744 if (thflags & TH_ACK) { 745 syncache_badack(&inc); 746 tcpstat.tcps_badsyn++; 747 rstreason = BANDLIM_RST_OPENPORT; 748 goto dropwithreset; 749 } 750 goto drop; 751 } 752 753 /* 754 * Segment's flags are (SYN) or (SYN|FIN). 755 */ 756 #ifdef INET6 757 /* 758 * If deprecated address is forbidden, 759 * we do not accept SYN to deprecated interface 760 * address to prevent any new inbound connection from 761 * getting established. 762 * When we do not accept SYN, we send a TCP RST, 763 * with deprecated source address (instead of dropping 764 * it). We compromise it as it is much better for peer 765 * to send a RST, and RST will be the final packet 766 * for the exchange. 767 * 768 * If we do not forbid deprecated addresses, we accept 769 * the SYN packet. RFC2462 does not suggest dropping 770 * SYN in this case. 771 * If we decipher RFC2462 5.5.4, it says like this: 772 * 1. use of deprecated addr with existing 773 * communication is okay - "SHOULD continue to be 774 * used" 775 * 2. use of it with new communication: 776 * (2a) "SHOULD NOT be used if alternate address 777 * with sufficient scope is available" 778 * (2b) nothing mentioned otherwise. 779 * Here we fall into (2b) case as we have no choice in 780 * our source address selection - we must obey the peer. 781 * 782 * The wording in RFC2462 is confusing, and there are 783 * multiple description text for deprecated address 784 * handling - worse, they are not exactly the same. 785 * I believe 5.5.4 is the best one, so we follow 5.5.4. 786 */ 787 if (isipv6 && !ip6_use_deprecated) { 788 struct in6_ifaddr *ia6; 789 790 if ((ia6 = ip6_getdstifaddr(m)) && 791 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 792 INP_UNLOCK(inp); 793 tp = NULL; 794 rstreason = BANDLIM_RST_OPENPORT; 795 goto dropwithreset; 796 } 797 } 798 #endif 799 /* 800 * If it is from this socket, drop it, it must be forged. 801 * Don't bother responding if the destination was a broadcast. 802 */ 803 if (th->th_dport == th->th_sport) { 804 #ifdef INET6 805 if (isipv6) { 806 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 807 &ip6->ip6_src)) 808 goto drop; 809 } else 810 #endif /* INET6 */ 811 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 812 goto drop; 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 #ifdef INET6 824 if (isipv6) { 825 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 826 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 827 goto drop; 828 } else 829 #endif 830 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 831 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 832 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 833 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 834 goto drop; 835 /* 836 * SYN appears to be valid; create compressed TCP state 837 * for syncache, or perform t/tcp connection. 838 */ 839 if (so->so_qlen <= so->so_qlimit) { 840 tcp_dooptions(&to, optp, optlen, 1); 841 if (!syncache_add(&inc, &to, th, &so, m)) 842 goto drop; 843 if (so == NULL) { 844 /* 845 * Entry added to syncache, mbuf used to 846 * send SYN,ACK packet. 847 */ 848 KASSERT(headlocked, ("headlocked")); 849 INP_UNLOCK(inp); 850 INP_INFO_WUNLOCK(&tcbinfo); 851 return; 852 } 853 /* 854 * Segment passed TAO tests. 855 */ 856 INP_UNLOCK(inp); 857 inp = sotoinpcb(so); 858 INP_LOCK(inp); 859 tp = intotcpcb(inp); 860 tp->snd_wnd = tiwin; 861 tp->t_starttime = ticks; 862 tp->t_state = TCPS_ESTABLISHED; 863 864 /* 865 * If there is a FIN, or if there is data and the 866 * connection is local, then delay SYN,ACK(SYN) in 867 * the hope of piggy-backing it on a response 868 * segment. Otherwise must send ACK now in case 869 * the other side is slow starting. 870 */ 871 if (DELAY_ACK(tp) && ((thflags & TH_FIN) || 872 (tlen != 0 && 873 #ifdef INET6 874 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) 875 || 876 (!isipv6 && 877 #endif 878 in_localaddr(inp->inp_faddr) 879 #ifdef INET6 880 )) 881 #endif 882 ))) { 883 callout_reset(tp->tt_delack, tcp_delacktime, 884 tcp_timer_delack, tp); 885 tp->t_flags |= TF_NEEDSYN; 886 } else 887 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 888 889 tcpstat.tcps_connects++; 890 soisconnected(so); 891 goto trimthenstep6; 892 } 893 goto drop; 894 } 895 after_listen: 896 897 /* XXX temp debugging */ 898 /* should not happen - syncache should pick up these connections */ 899 if (tp->t_state == TCPS_LISTEN) 900 panic("tcp_input: TCPS_LISTEN"); 901 902 /* 903 * Segment received on connection. 904 * Reset idle time and keep-alive timer. 905 */ 906 tp->t_rcvtime = ticks; 907 if (TCPS_HAVEESTABLISHED(tp->t_state)) 908 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 909 910 /* 911 * Process options. 912 * XXX this is tradtitional behavior, may need to be cleaned up. 913 */ 914 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 915 if (thflags & TH_SYN) { 916 if (to.to_flags & TOF_SCALE) { 917 tp->t_flags |= TF_RCVD_SCALE; 918 tp->requested_s_scale = to.to_requested_s_scale; 919 } 920 if (to.to_flags & TOF_TS) { 921 tp->t_flags |= TF_RCVD_TSTMP; 922 tp->ts_recent = to.to_tsval; 923 tp->ts_recent_age = ticks; 924 } 925 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 926 tp->t_flags |= TF_RCVD_CC; 927 if (to.to_flags & TOF_MSS) 928 tcp_mss(tp, to.to_mss); 929 } 930 931 /* 932 * Header prediction: check for the two common cases 933 * of a uni-directional data xfer. If the packet has 934 * no control flags, is in-sequence, the window didn't 935 * change and we're not retransmitting, it's a 936 * candidate. If the length is zero and the ack moved 937 * forward, we're the sender side of the xfer. Just 938 * free the data acked & wake any higher level process 939 * that was blocked waiting for space. If the length 940 * is non-zero and the ack didn't move, we're the 941 * receiver side. If we're getting packets in-order 942 * (the reassembly queue is empty), add the data to 943 * the socket buffer and note that we need a delayed ack. 944 * Make sure that the hidden state-flags are also off. 945 * Since we check for TCPS_ESTABLISHED above, it can only 946 * be TH_NEEDSYN. 947 */ 948 if (tp->t_state == TCPS_ESTABLISHED && 949 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 950 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 951 ((to.to_flags & TOF_TS) == 0 || 952 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 953 /* 954 * Using the CC option is compulsory if once started: 955 * the segment is OK if no T/TCP was negotiated or 956 * if the segment has a CC option equal to CCrecv 957 */ 958 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 959 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 960 th->th_seq == tp->rcv_nxt && 961 tiwin && tiwin == tp->snd_wnd && 962 tp->snd_nxt == tp->snd_max) { 963 964 /* 965 * If last ACK falls within this segment's sequence numbers, 966 * record the timestamp. 967 * NOTE that the test is modified according to the latest 968 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 969 */ 970 if ((to.to_flags & TOF_TS) != 0 && 971 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 972 tp->ts_recent_age = ticks; 973 tp->ts_recent = to.to_tsval; 974 } 975 976 if (tlen == 0) { 977 if (SEQ_GT(th->th_ack, tp->snd_una) && 978 SEQ_LEQ(th->th_ack, tp->snd_max) && 979 tp->snd_cwnd >= tp->snd_wnd && 980 tp->t_dupacks < tcprexmtthresh) { 981 KASSERT(headlocked, ("headlocked")); 982 INP_INFO_WUNLOCK(&tcbinfo); 983 headlocked = 0; 984 /* 985 * this is a pure ack for outstanding data. 986 */ 987 ++tcpstat.tcps_predack; 988 /* 989 * "bad retransmit" recovery 990 */ 991 if (tp->t_rxtshift == 1 && 992 ticks < tp->t_badrxtwin) { 993 tp->snd_cwnd = tp->snd_cwnd_prev; 994 tp->snd_ssthresh = 995 tp->snd_ssthresh_prev; 996 tp->snd_nxt = tp->snd_max; 997 tp->t_badrxtwin = 0; 998 } 999 if ((to.to_flags & TOF_TS) != 0) 1000 tcp_xmit_timer(tp, 1001 ticks - to.to_tsecr + 1); 1002 else if (tp->t_rtttime && 1003 SEQ_GT(th->th_ack, tp->t_rtseq)) 1004 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1005 acked = th->th_ack - tp->snd_una; 1006 tcpstat.tcps_rcvackpack++; 1007 tcpstat.tcps_rcvackbyte += acked; 1008 sbdrop(&so->so_snd, acked); 1009 tp->snd_una = th->th_ack; 1010 m_freem(m); 1011 ND6_HINT(tp); /* some progress has been done */ 1012 1013 /* 1014 * If all outstanding data are acked, stop 1015 * retransmit timer, otherwise restart timer 1016 * using current (possibly backed-off) value. 1017 * If process is waiting for space, 1018 * wakeup/selwakeup/signal. If data 1019 * are ready to send, let tcp_output 1020 * decide between more output or persist. 1021 */ 1022 if (tp->snd_una == tp->snd_max) 1023 callout_stop(tp->tt_rexmt); 1024 else if (!callout_active(tp->tt_persist)) 1025 callout_reset(tp->tt_rexmt, 1026 tp->t_rxtcur, 1027 tcp_timer_rexmt, tp); 1028 1029 sowwakeup(so); 1030 if (so->so_snd.sb_cc) 1031 (void) tcp_output(tp); 1032 INP_UNLOCK(inp); 1033 return; 1034 } 1035 } else if (th->th_ack == tp->snd_una && 1036 LIST_EMPTY(&tp->t_segq) && 1037 tlen <= sbspace(&so->so_rcv)) { 1038 KASSERT(headlocked, ("headlocked")); 1039 INP_INFO_WUNLOCK(&tcbinfo); 1040 headlocked = 0; 1041 /* 1042 * this is a pure, in-sequence data packet 1043 * with nothing on the reassembly queue and 1044 * we have enough buffer space to take it. 1045 */ 1046 ++tcpstat.tcps_preddat; 1047 tp->rcv_nxt += tlen; 1048 tcpstat.tcps_rcvpack++; 1049 tcpstat.tcps_rcvbyte += tlen; 1050 ND6_HINT(tp); /* some progress has been done */ 1051 /* 1052 * Add data to socket buffer. 1053 */ 1054 m_adj(m, drop_hdrlen); /* delayed header drop */ 1055 sbappend(&so->so_rcv, m); 1056 sorwakeup(so); 1057 if (DELAY_ACK(tp)) { 1058 callout_reset(tp->tt_delack, tcp_delacktime, 1059 tcp_timer_delack, tp); 1060 } else { 1061 tp->t_flags |= TF_ACKNOW; 1062 tcp_output(tp); 1063 } 1064 INP_UNLOCK(inp); 1065 return; 1066 } 1067 } 1068 1069 /* 1070 * Calculate amount of space in receive window, 1071 * and then do TCP input processing. 1072 * Receive window is amount of space in rcv queue, 1073 * but not less than advertised window. 1074 */ 1075 { int win; 1076 1077 win = sbspace(&so->so_rcv); 1078 if (win < 0) 1079 win = 0; 1080 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1081 } 1082 1083 switch (tp->t_state) { 1084 1085 /* 1086 * If the state is SYN_RECEIVED: 1087 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1088 */ 1089 case TCPS_SYN_RECEIVED: 1090 if ((thflags & TH_ACK) && 1091 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1092 SEQ_GT(th->th_ack, tp->snd_max))) { 1093 rstreason = BANDLIM_RST_OPENPORT; 1094 goto dropwithreset; 1095 } 1096 break; 1097 1098 /* 1099 * If the state is SYN_SENT: 1100 * if seg contains an ACK, but not for our SYN, drop the input. 1101 * if seg contains a RST, then drop the connection. 1102 * if seg does not contain SYN, then drop it. 1103 * Otherwise this is an acceptable SYN segment 1104 * initialize tp->rcv_nxt and tp->irs 1105 * if seg contains ack then advance tp->snd_una 1106 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1107 * arrange for segment to be acked (eventually) 1108 * continue processing rest of data/controls, beginning with URG 1109 */ 1110 case TCPS_SYN_SENT: 1111 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1112 taop = &tao_noncached; 1113 bzero(taop, sizeof(*taop)); 1114 } 1115 1116 if ((thflags & TH_ACK) && 1117 (SEQ_LEQ(th->th_ack, tp->iss) || 1118 SEQ_GT(th->th_ack, tp->snd_max))) { 1119 /* 1120 * If we have a cached CCsent for the remote host, 1121 * hence we haven't just crashed and restarted, 1122 * do not send a RST. This may be a retransmission 1123 * from the other side after our earlier ACK was lost. 1124 * Our new SYN, when it arrives, will serve as the 1125 * needed ACK. 1126 */ 1127 if (taop->tao_ccsent != 0) 1128 goto drop; 1129 else { 1130 rstreason = BANDLIM_UNLIMITED; 1131 goto dropwithreset; 1132 } 1133 } 1134 if (thflags & TH_RST) { 1135 if (thflags & TH_ACK) 1136 tp = tcp_drop(tp, ECONNREFUSED); 1137 goto drop; 1138 } 1139 if ((thflags & TH_SYN) == 0) 1140 goto drop; 1141 tp->snd_wnd = th->th_win; /* initial send window */ 1142 tp->cc_recv = to.to_cc; /* foreign CC */ 1143 1144 tp->irs = th->th_seq; 1145 tcp_rcvseqinit(tp); 1146 if (thflags & TH_ACK) { 1147 /* 1148 * Our SYN was acked. If segment contains CC.ECHO 1149 * option, check it to make sure this segment really 1150 * matches our SYN. If not, just drop it as old 1151 * duplicate, but send an RST if we're still playing 1152 * by the old rules. If no CC.ECHO option, make sure 1153 * we don't get fooled into using T/TCP. 1154 */ 1155 if (to.to_flags & TOF_CCECHO) { 1156 if (tp->cc_send != to.to_ccecho) { 1157 if (taop->tao_ccsent != 0) 1158 goto drop; 1159 else { 1160 rstreason = BANDLIM_UNLIMITED; 1161 goto dropwithreset; 1162 } 1163 } 1164 } else 1165 tp->t_flags &= ~TF_RCVD_CC; 1166 tcpstat.tcps_connects++; 1167 soisconnected(so); 1168 /* Do window scaling on this connection? */ 1169 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1170 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1171 tp->snd_scale = tp->requested_s_scale; 1172 tp->rcv_scale = tp->request_r_scale; 1173 } 1174 /* Segment is acceptable, update cache if undefined. */ 1175 if (taop->tao_ccsent == 0) 1176 taop->tao_ccsent = to.to_ccecho; 1177 1178 tp->rcv_adv += tp->rcv_wnd; 1179 tp->snd_una++; /* SYN is acked */ 1180 /* 1181 * If there's data, delay ACK; if there's also a FIN 1182 * ACKNOW will be turned on later. 1183 */ 1184 if (DELAY_ACK(tp) && tlen != 0) 1185 callout_reset(tp->tt_delack, tcp_delacktime, 1186 tcp_timer_delack, tp); 1187 else 1188 tp->t_flags |= TF_ACKNOW; 1189 /* 1190 * Received <SYN,ACK> in SYN_SENT[*] state. 1191 * Transitions: 1192 * SYN_SENT --> ESTABLISHED 1193 * SYN_SENT* --> FIN_WAIT_1 1194 */ 1195 tp->t_starttime = ticks; 1196 if (tp->t_flags & TF_NEEDFIN) { 1197 tp->t_state = TCPS_FIN_WAIT_1; 1198 tp->t_flags &= ~TF_NEEDFIN; 1199 thflags &= ~TH_SYN; 1200 } else { 1201 tp->t_state = TCPS_ESTABLISHED; 1202 callout_reset(tp->tt_keep, tcp_keepidle, 1203 tcp_timer_keep, tp); 1204 } 1205 } else { 1206 /* 1207 * Received initial SYN in SYN-SENT[*] state => simul- 1208 * taneous open. If segment contains CC option and there is 1209 * a cached CC, apply TAO test; if it succeeds, connection is 1210 * half-synchronized. Otherwise, do 3-way handshake: 1211 * SYN-SENT -> SYN-RECEIVED 1212 * SYN-SENT* -> SYN-RECEIVED* 1213 * If there was no CC option, clear cached CC value. 1214 */ 1215 tp->t_flags |= TF_ACKNOW; 1216 callout_stop(tp->tt_rexmt); 1217 if (to.to_flags & TOF_CC) { 1218 if (taop->tao_cc != 0 && 1219 CC_GT(to.to_cc, taop->tao_cc)) { 1220 /* 1221 * update cache and make transition: 1222 * SYN-SENT -> ESTABLISHED* 1223 * SYN-SENT* -> FIN-WAIT-1* 1224 */ 1225 taop->tao_cc = to.to_cc; 1226 tp->t_starttime = ticks; 1227 if (tp->t_flags & TF_NEEDFIN) { 1228 tp->t_state = TCPS_FIN_WAIT_1; 1229 tp->t_flags &= ~TF_NEEDFIN; 1230 } else { 1231 tp->t_state = TCPS_ESTABLISHED; 1232 callout_reset(tp->tt_keep, 1233 tcp_keepidle, 1234 tcp_timer_keep, 1235 tp); 1236 } 1237 tp->t_flags |= TF_NEEDSYN; 1238 } else 1239 tp->t_state = TCPS_SYN_RECEIVED; 1240 } else { 1241 /* CC.NEW or no option => invalidate cache */ 1242 taop->tao_cc = 0; 1243 tp->t_state = TCPS_SYN_RECEIVED; 1244 } 1245 } 1246 1247 trimthenstep6: 1248 /* 1249 * Advance th->th_seq to correspond to first data byte. 1250 * If data, trim to stay within window, 1251 * dropping FIN if necessary. 1252 */ 1253 th->th_seq++; 1254 if (tlen > tp->rcv_wnd) { 1255 todrop = tlen - tp->rcv_wnd; 1256 m_adj(m, -todrop); 1257 tlen = tp->rcv_wnd; 1258 thflags &= ~TH_FIN; 1259 tcpstat.tcps_rcvpackafterwin++; 1260 tcpstat.tcps_rcvbyteafterwin += todrop; 1261 } 1262 tp->snd_wl1 = th->th_seq - 1; 1263 tp->rcv_up = th->th_seq; 1264 /* 1265 * Client side of transaction: already sent SYN and data. 1266 * If the remote host used T/TCP to validate the SYN, 1267 * our data will be ACK'd; if so, enter normal data segment 1268 * processing in the middle of step 5, ack processing. 1269 * Otherwise, goto step 6. 1270 */ 1271 if (thflags & TH_ACK) 1272 goto process_ACK; 1273 goto step6; 1274 /* 1275 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1276 * if segment contains a SYN and CC [not CC.NEW] option: 1277 * if state == TIME_WAIT and connection duration > MSL, 1278 * drop packet and send RST; 1279 * 1280 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1281 * ack the FIN (and data) in retransmission queue. 1282 * Complete close and delete TCPCB. Then reprocess 1283 * segment, hoping to find new TCPCB in LISTEN state; 1284 * 1285 * else must be old SYN; drop it. 1286 * else do normal processing. 1287 */ 1288 case TCPS_LAST_ACK: 1289 case TCPS_CLOSING: 1290 case TCPS_TIME_WAIT: 1291 if ((thflags & TH_SYN) && 1292 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1293 if (tp->t_state == TCPS_TIME_WAIT && 1294 (ticks - tp->t_starttime) > tcp_msl) { 1295 rstreason = BANDLIM_UNLIMITED; 1296 goto dropwithreset; 1297 } 1298 if (CC_GT(to.to_cc, tp->cc_recv)) { 1299 tp = tcp_close(tp); 1300 goto findpcb; 1301 } 1302 else 1303 goto drop; 1304 } 1305 break; /* continue normal processing */ 1306 } 1307 1308 /* 1309 * States other than LISTEN or SYN_SENT. 1310 * First check the RST flag and sequence number since reset segments 1311 * are exempt from the timestamp and connection count tests. This 1312 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1313 * below which allowed reset segments in half the sequence space 1314 * to fall though and be processed (which gives forged reset 1315 * segments with a random sequence number a 50 percent chance of 1316 * killing a connection). 1317 * Then check timestamp, if present. 1318 * Then check the connection count, if present. 1319 * Then check that at least some bytes of segment are within 1320 * receive window. If segment begins before rcv_nxt, 1321 * drop leading data (and SYN); if nothing left, just ack. 1322 * 1323 * 1324 * If the RST bit is set, check the sequence number to see 1325 * if this is a valid reset segment. 1326 * RFC 793 page 37: 1327 * In all states except SYN-SENT, all reset (RST) segments 1328 * are validated by checking their SEQ-fields. A reset is 1329 * valid if its sequence number is in the window. 1330 * Note: this does not take into account delayed ACKs, so 1331 * we should test against last_ack_sent instead of rcv_nxt. 1332 * The sequence number in the reset segment is normally an 1333 * echo of our outgoing acknowlegement numbers, but some hosts 1334 * send a reset with the sequence number at the rightmost edge 1335 * of our receive window, and we have to handle this case. 1336 * If we have multiple segments in flight, the intial reset 1337 * segment sequence numbers will be to the left of last_ack_sent, 1338 * but they will eventually catch up. 1339 * In any case, it never made sense to trim reset segments to 1340 * fit the receive window since RFC 1122 says: 1341 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1342 * 1343 * A TCP SHOULD allow a received RST segment to include data. 1344 * 1345 * DISCUSSION 1346 * It has been suggested that a RST segment could contain 1347 * ASCII text that encoded and explained the cause of the 1348 * RST. No standard has yet been established for such 1349 * data. 1350 * 1351 * If the reset segment passes the sequence number test examine 1352 * the state: 1353 * SYN_RECEIVED STATE: 1354 * If passive open, return to LISTEN state. 1355 * If active open, inform user that connection was refused. 1356 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1357 * Inform user that connection was reset, and close tcb. 1358 * CLOSING, LAST_ACK STATES: 1359 * Close the tcb. 1360 * TIME_WAIT STATE: 1361 * Drop the segment - see Stevens, vol. 2, p. 964 and 1362 * RFC 1337. 1363 */ 1364 if (thflags & TH_RST) { 1365 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1366 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1367 switch (tp->t_state) { 1368 1369 case TCPS_SYN_RECEIVED: 1370 so->so_error = ECONNREFUSED; 1371 goto close; 1372 1373 case TCPS_ESTABLISHED: 1374 case TCPS_FIN_WAIT_1: 1375 case TCPS_FIN_WAIT_2: 1376 case TCPS_CLOSE_WAIT: 1377 so->so_error = ECONNRESET; 1378 close: 1379 tp->t_state = TCPS_CLOSED; 1380 tcpstat.tcps_drops++; 1381 tp = tcp_close(tp); 1382 break; 1383 1384 case TCPS_CLOSING: 1385 case TCPS_LAST_ACK: 1386 tp = tcp_close(tp); 1387 break; 1388 1389 case TCPS_TIME_WAIT: 1390 break; 1391 } 1392 } 1393 goto drop; 1394 } 1395 1396 /* 1397 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1398 * and it's less than ts_recent, drop it. 1399 */ 1400 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1401 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1402 1403 /* Check to see if ts_recent is over 24 days old. */ 1404 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1405 /* 1406 * Invalidate ts_recent. If this segment updates 1407 * ts_recent, the age will be reset later and ts_recent 1408 * will get a valid value. If it does not, setting 1409 * ts_recent to zero will at least satisfy the 1410 * requirement that zero be placed in the timestamp 1411 * echo reply when ts_recent isn't valid. The 1412 * age isn't reset until we get a valid ts_recent 1413 * because we don't want out-of-order segments to be 1414 * dropped when ts_recent is old. 1415 */ 1416 tp->ts_recent = 0; 1417 } else { 1418 tcpstat.tcps_rcvduppack++; 1419 tcpstat.tcps_rcvdupbyte += tlen; 1420 tcpstat.tcps_pawsdrop++; 1421 goto dropafterack; 1422 } 1423 } 1424 1425 /* 1426 * T/TCP mechanism 1427 * If T/TCP was negotiated and the segment doesn't have CC, 1428 * or if its CC is wrong then drop the segment. 1429 * RST segments do not have to comply with this. 1430 */ 1431 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1432 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1433 goto dropafterack; 1434 1435 /* 1436 * In the SYN-RECEIVED state, validate that the packet belongs to 1437 * this connection before trimming the data to fit the receive 1438 * window. Check the sequence number versus IRS since we know 1439 * the sequence numbers haven't wrapped. This is a partial fix 1440 * for the "LAND" DoS attack. 1441 */ 1442 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1443 rstreason = BANDLIM_RST_OPENPORT; 1444 goto dropwithreset; 1445 } 1446 1447 todrop = tp->rcv_nxt - th->th_seq; 1448 if (todrop > 0) { 1449 if (thflags & TH_SYN) { 1450 thflags &= ~TH_SYN; 1451 th->th_seq++; 1452 if (th->th_urp > 1) 1453 th->th_urp--; 1454 else 1455 thflags &= ~TH_URG; 1456 todrop--; 1457 } 1458 /* 1459 * Following if statement from Stevens, vol. 2, p. 960. 1460 */ 1461 if (todrop > tlen 1462 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1463 /* 1464 * Any valid FIN must be to the left of the window. 1465 * At this point the FIN must be a duplicate or out 1466 * of sequence; drop it. 1467 */ 1468 thflags &= ~TH_FIN; 1469 1470 /* 1471 * Send an ACK to resynchronize and drop any data. 1472 * But keep on processing for RST or ACK. 1473 */ 1474 tp->t_flags |= TF_ACKNOW; 1475 todrop = tlen; 1476 tcpstat.tcps_rcvduppack++; 1477 tcpstat.tcps_rcvdupbyte += todrop; 1478 } else { 1479 tcpstat.tcps_rcvpartduppack++; 1480 tcpstat.tcps_rcvpartdupbyte += todrop; 1481 } 1482 drop_hdrlen += todrop; /* drop from the top afterwards */ 1483 th->th_seq += todrop; 1484 tlen -= todrop; 1485 if (th->th_urp > todrop) 1486 th->th_urp -= todrop; 1487 else { 1488 thflags &= ~TH_URG; 1489 th->th_urp = 0; 1490 } 1491 } 1492 1493 /* 1494 * If new data are received on a connection after the 1495 * user processes are gone, then RST the other end. 1496 */ 1497 if ((so->so_state & SS_NOFDREF) && 1498 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1499 tp = tcp_close(tp); 1500 tcpstat.tcps_rcvafterclose++; 1501 rstreason = BANDLIM_UNLIMITED; 1502 goto dropwithreset; 1503 } 1504 1505 /* 1506 * If segment ends after window, drop trailing data 1507 * (and PUSH and FIN); if nothing left, just ACK. 1508 */ 1509 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1510 if (todrop > 0) { 1511 tcpstat.tcps_rcvpackafterwin++; 1512 if (todrop >= tlen) { 1513 tcpstat.tcps_rcvbyteafterwin += tlen; 1514 /* 1515 * If a new connection request is received 1516 * while in TIME_WAIT, drop the old connection 1517 * and start over if the sequence numbers 1518 * are above the previous ones. 1519 */ 1520 if (thflags & TH_SYN && 1521 tp->t_state == TCPS_TIME_WAIT && 1522 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1523 tp = tcp_close(tp); 1524 goto findpcb; 1525 } 1526 /* 1527 * If window is closed can only take segments at 1528 * window edge, and have to drop data and PUSH from 1529 * incoming segments. Continue processing, but 1530 * remember to ack. Otherwise, drop segment 1531 * and ack. 1532 */ 1533 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1534 tp->t_flags |= TF_ACKNOW; 1535 tcpstat.tcps_rcvwinprobe++; 1536 } else 1537 goto dropafterack; 1538 } else 1539 tcpstat.tcps_rcvbyteafterwin += todrop; 1540 m_adj(m, -todrop); 1541 tlen -= todrop; 1542 thflags &= ~(TH_PUSH|TH_FIN); 1543 } 1544 1545 /* 1546 * If last ACK falls within this segment's sequence numbers, 1547 * record its timestamp. 1548 * NOTE that the test is modified according to the latest 1549 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1550 */ 1551 if ((to.to_flags & TOF_TS) != 0 && 1552 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1553 tp->ts_recent_age = ticks; 1554 tp->ts_recent = to.to_tsval; 1555 } 1556 1557 /* 1558 * If a SYN is in the window, then this is an 1559 * error and we send an RST and drop the connection. 1560 */ 1561 if (thflags & TH_SYN) { 1562 tp = tcp_drop(tp, ECONNRESET); 1563 rstreason = BANDLIM_UNLIMITED; 1564 goto dropwithreset; 1565 } 1566 1567 /* 1568 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1569 * flag is on (half-synchronized state), then queue data for 1570 * later processing; else drop segment and return. 1571 */ 1572 if ((thflags & TH_ACK) == 0) { 1573 if (tp->t_state == TCPS_SYN_RECEIVED || 1574 (tp->t_flags & TF_NEEDSYN)) 1575 goto step6; 1576 else 1577 goto drop; 1578 } 1579 1580 /* 1581 * Ack processing. 1582 */ 1583 switch (tp->t_state) { 1584 1585 /* 1586 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1587 * ESTABLISHED state and continue processing. 1588 * The ACK was checked above. 1589 */ 1590 case TCPS_SYN_RECEIVED: 1591 1592 tcpstat.tcps_connects++; 1593 soisconnected(so); 1594 /* Do window scaling? */ 1595 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1596 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1597 tp->snd_scale = tp->requested_s_scale; 1598 tp->rcv_scale = tp->request_r_scale; 1599 } 1600 /* 1601 * Upon successful completion of 3-way handshake, 1602 * update cache.CC if it was undefined, pass any queued 1603 * data to the user, and advance state appropriately. 1604 */ 1605 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1606 taop->tao_cc == 0) 1607 taop->tao_cc = tp->cc_recv; 1608 1609 /* 1610 * Make transitions: 1611 * SYN-RECEIVED -> ESTABLISHED 1612 * SYN-RECEIVED* -> FIN-WAIT-1 1613 */ 1614 tp->t_starttime = ticks; 1615 if (tp->t_flags & TF_NEEDFIN) { 1616 tp->t_state = TCPS_FIN_WAIT_1; 1617 tp->t_flags &= ~TF_NEEDFIN; 1618 } else { 1619 tp->t_state = TCPS_ESTABLISHED; 1620 callout_reset(tp->tt_keep, tcp_keepidle, 1621 tcp_timer_keep, tp); 1622 } 1623 /* 1624 * If segment contains data or ACK, will call tcp_reass() 1625 * later; if not, do so now to pass queued data to user. 1626 */ 1627 if (tlen == 0 && (thflags & TH_FIN) == 0) 1628 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1629 (struct mbuf *)0); 1630 tp->snd_wl1 = th->th_seq - 1; 1631 /* fall into ... */ 1632 1633 /* 1634 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1635 * ACKs. If the ack is in the range 1636 * tp->snd_una < th->th_ack <= tp->snd_max 1637 * then advance tp->snd_una to th->th_ack and drop 1638 * data from the retransmission queue. If this ACK reflects 1639 * more up to date window information we update our window information. 1640 */ 1641 case TCPS_ESTABLISHED: 1642 case TCPS_FIN_WAIT_1: 1643 case TCPS_FIN_WAIT_2: 1644 case TCPS_CLOSE_WAIT: 1645 case TCPS_CLOSING: 1646 case TCPS_LAST_ACK: 1647 case TCPS_TIME_WAIT: 1648 1649 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1650 if (tlen == 0 && tiwin == tp->snd_wnd) { 1651 tcpstat.tcps_rcvdupack++; 1652 /* 1653 * If we have outstanding data (other than 1654 * a window probe), this is a completely 1655 * duplicate ack (ie, window info didn't 1656 * change), the ack is the biggest we've 1657 * seen and we've seen exactly our rexmt 1658 * threshhold of them, assume a packet 1659 * has been dropped and retransmit it. 1660 * Kludge snd_nxt & the congestion 1661 * window so we send only this one 1662 * packet. 1663 * 1664 * We know we're losing at the current 1665 * window size so do congestion avoidance 1666 * (set ssthresh to half the current window 1667 * and pull our congestion window back to 1668 * the new ssthresh). 1669 * 1670 * Dup acks mean that packets have left the 1671 * network (they're now cached at the receiver) 1672 * so bump cwnd by the amount in the receiver 1673 * to keep a constant cwnd packets in the 1674 * network. 1675 */ 1676 if (!callout_active(tp->tt_rexmt) || 1677 th->th_ack != tp->snd_una) 1678 tp->t_dupacks = 0; 1679 else if (++tp->t_dupacks == tcprexmtthresh) { 1680 tcp_seq onxt = tp->snd_nxt; 1681 u_int win = 1682 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1683 tp->t_maxseg; 1684 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1685 tp->snd_recover)) { 1686 /* False retransmit, should not 1687 * cut window 1688 */ 1689 tp->snd_cwnd += tp->t_maxseg; 1690 tp->t_dupacks = 0; 1691 (void) tcp_output(tp); 1692 goto drop; 1693 } 1694 if (win < 2) 1695 win = 2; 1696 tp->snd_ssthresh = win * tp->t_maxseg; 1697 tp->snd_recover = tp->snd_max; 1698 callout_stop(tp->tt_rexmt); 1699 tp->t_rtttime = 0; 1700 tp->snd_nxt = th->th_ack; 1701 tp->snd_cwnd = tp->t_maxseg; 1702 (void) tcp_output(tp); 1703 tp->snd_cwnd = tp->snd_ssthresh + 1704 tp->t_maxseg * tp->t_dupacks; 1705 if (SEQ_GT(onxt, tp->snd_nxt)) 1706 tp->snd_nxt = onxt; 1707 goto drop; 1708 } else if (tp->t_dupacks > tcprexmtthresh) { 1709 tp->snd_cwnd += tp->t_maxseg; 1710 (void) tcp_output(tp); 1711 goto drop; 1712 } 1713 } else 1714 tp->t_dupacks = 0; 1715 break; 1716 } 1717 /* 1718 * If the congestion window was inflated to account 1719 * for the other side's cached packets, retract it. 1720 */ 1721 if (tcp_do_newreno == 0) { 1722 if (tp->t_dupacks >= tcprexmtthresh && 1723 tp->snd_cwnd > tp->snd_ssthresh) 1724 tp->snd_cwnd = tp->snd_ssthresh; 1725 tp->t_dupacks = 0; 1726 } else if (tp->t_dupacks >= tcprexmtthresh && 1727 !tcp_newreno(tp, th)) { 1728 /* 1729 * Window inflation should have left us with approx. 1730 * snd_ssthresh outstanding data. But in case we 1731 * would be inclined to send a burst, better to do 1732 * it via the slow start mechanism. 1733 */ 1734 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max)) 1735 tp->snd_cwnd = 1736 tp->snd_max - th->th_ack + tp->t_maxseg; 1737 else 1738 tp->snd_cwnd = tp->snd_ssthresh; 1739 tp->t_dupacks = 0; 1740 } 1741 if (tp->t_dupacks < tcprexmtthresh) 1742 tp->t_dupacks = 0; 1743 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1744 tcpstat.tcps_rcvacktoomuch++; 1745 goto dropafterack; 1746 } 1747 /* 1748 * If we reach this point, ACK is not a duplicate, 1749 * i.e., it ACKs something we sent. 1750 */ 1751 if (tp->t_flags & TF_NEEDSYN) { 1752 /* 1753 * T/TCP: Connection was half-synchronized, and our 1754 * SYN has been ACK'd (so connection is now fully 1755 * synchronized). Go to non-starred state, 1756 * increment snd_una for ACK of SYN, and check if 1757 * we can do window scaling. 1758 */ 1759 tp->t_flags &= ~TF_NEEDSYN; 1760 tp->snd_una++; 1761 /* Do window scaling? */ 1762 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1763 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1764 tp->snd_scale = tp->requested_s_scale; 1765 tp->rcv_scale = tp->request_r_scale; 1766 } 1767 } 1768 1769 process_ACK: 1770 acked = th->th_ack - tp->snd_una; 1771 tcpstat.tcps_rcvackpack++; 1772 tcpstat.tcps_rcvackbyte += acked; 1773 1774 /* 1775 * If we just performed our first retransmit, and the ACK 1776 * arrives within our recovery window, then it was a mistake 1777 * to do the retransmit in the first place. Recover our 1778 * original cwnd and ssthresh, and proceed to transmit where 1779 * we left off. 1780 */ 1781 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1782 tp->snd_cwnd = tp->snd_cwnd_prev; 1783 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1784 tp->snd_nxt = tp->snd_max; 1785 tp->t_badrxtwin = 0; /* XXX probably not required */ 1786 } 1787 1788 /* 1789 * If we have a timestamp reply, update smoothed 1790 * round trip time. If no timestamp is present but 1791 * transmit timer is running and timed sequence 1792 * number was acked, update smoothed round trip time. 1793 * Since we now have an rtt measurement, cancel the 1794 * timer backoff (cf., Phil Karn's retransmit alg.). 1795 * Recompute the initial retransmit timer. 1796 */ 1797 if (to.to_flags & TOF_TS) 1798 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1799 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1800 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1801 1802 /* 1803 * If all outstanding data is acked, stop retransmit 1804 * timer and remember to restart (more output or persist). 1805 * If there is more data to be acked, restart retransmit 1806 * timer, using current (possibly backed-off) value. 1807 */ 1808 if (th->th_ack == tp->snd_max) { 1809 callout_stop(tp->tt_rexmt); 1810 needoutput = 1; 1811 } else if (!callout_active(tp->tt_persist)) 1812 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1813 tcp_timer_rexmt, tp); 1814 1815 /* 1816 * If no data (only SYN) was ACK'd, 1817 * skip rest of ACK processing. 1818 */ 1819 if (acked == 0) 1820 goto step6; 1821 1822 /* 1823 * When new data is acked, open the congestion window. 1824 * If the window gives us less than ssthresh packets 1825 * in flight, open exponentially (maxseg per packet). 1826 * Otherwise open linearly: maxseg per window 1827 * (maxseg^2 / cwnd per packet). 1828 */ 1829 { 1830 register u_int cw = tp->snd_cwnd; 1831 register u_int incr = tp->t_maxseg; 1832 1833 if (cw > tp->snd_ssthresh) 1834 incr = incr * incr / cw; 1835 /* 1836 * If t_dupacks != 0 here, it indicates that we are still 1837 * in NewReno fast recovery mode, so we leave the congestion 1838 * window alone. 1839 */ 1840 if (tcp_do_newreno == 0 || tp->t_dupacks == 0) 1841 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1842 } 1843 if (acked > so->so_snd.sb_cc) { 1844 tp->snd_wnd -= so->so_snd.sb_cc; 1845 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1846 ourfinisacked = 1; 1847 } else { 1848 sbdrop(&so->so_snd, acked); 1849 tp->snd_wnd -= acked; 1850 ourfinisacked = 0; 1851 } 1852 sowwakeup(so); 1853 tp->snd_una = th->th_ack; 1854 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1855 tp->snd_nxt = tp->snd_una; 1856 1857 switch (tp->t_state) { 1858 1859 /* 1860 * In FIN_WAIT_1 STATE in addition to the processing 1861 * for the ESTABLISHED state if our FIN is now acknowledged 1862 * then enter FIN_WAIT_2. 1863 */ 1864 case TCPS_FIN_WAIT_1: 1865 if (ourfinisacked) { 1866 /* 1867 * If we can't receive any more 1868 * data, then closing user can proceed. 1869 * Starting the timer is contrary to the 1870 * specification, but if we don't get a FIN 1871 * we'll hang forever. 1872 */ 1873 if (so->so_state & SS_CANTRCVMORE) { 1874 soisdisconnected(so); 1875 callout_reset(tp->tt_2msl, tcp_maxidle, 1876 tcp_timer_2msl, tp); 1877 } 1878 tp->t_state = TCPS_FIN_WAIT_2; 1879 } 1880 break; 1881 1882 /* 1883 * In CLOSING STATE in addition to the processing for 1884 * the ESTABLISHED state if the ACK acknowledges our FIN 1885 * then enter the TIME-WAIT state, otherwise ignore 1886 * the segment. 1887 */ 1888 case TCPS_CLOSING: 1889 if (ourfinisacked) { 1890 tp->t_state = TCPS_TIME_WAIT; 1891 tcp_canceltimers(tp); 1892 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1893 if (tp->cc_recv != 0 && 1894 (ticks - tp->t_starttime) < tcp_msl) 1895 callout_reset(tp->tt_2msl, 1896 tp->t_rxtcur * 1897 TCPTV_TWTRUNC, 1898 tcp_timer_2msl, tp); 1899 else 1900 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1901 tcp_timer_2msl, tp); 1902 soisdisconnected(so); 1903 } 1904 break; 1905 1906 /* 1907 * In LAST_ACK, we may still be waiting for data to drain 1908 * and/or to be acked, as well as for the ack of our FIN. 1909 * If our FIN is now acknowledged, delete the TCB, 1910 * enter the closed state and return. 1911 */ 1912 case TCPS_LAST_ACK: 1913 if (ourfinisacked) { 1914 tp = tcp_close(tp); 1915 goto drop; 1916 } 1917 break; 1918 1919 /* 1920 * In TIME_WAIT state the only thing that should arrive 1921 * is a retransmission of the remote FIN. Acknowledge 1922 * it and restart the finack timer. 1923 */ 1924 case TCPS_TIME_WAIT: 1925 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1926 tcp_timer_2msl, tp); 1927 goto dropafterack; 1928 } 1929 } 1930 1931 step6: 1932 /* 1933 * Update window information. 1934 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1935 */ 1936 if ((thflags & TH_ACK) && 1937 (SEQ_LT(tp->snd_wl1, th->th_seq) || 1938 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 1939 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 1940 /* keep track of pure window updates */ 1941 if (tlen == 0 && 1942 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1943 tcpstat.tcps_rcvwinupd++; 1944 tp->snd_wnd = tiwin; 1945 tp->snd_wl1 = th->th_seq; 1946 tp->snd_wl2 = th->th_ack; 1947 if (tp->snd_wnd > tp->max_sndwnd) 1948 tp->max_sndwnd = tp->snd_wnd; 1949 needoutput = 1; 1950 } 1951 1952 /* 1953 * Process segments with URG. 1954 */ 1955 if ((thflags & TH_URG) && th->th_urp && 1956 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1957 /* 1958 * This is a kludge, but if we receive and accept 1959 * random urgent pointers, we'll crash in 1960 * soreceive. It's hard to imagine someone 1961 * actually wanting to send this much urgent data. 1962 */ 1963 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1964 th->th_urp = 0; /* XXX */ 1965 thflags &= ~TH_URG; /* XXX */ 1966 goto dodata; /* XXX */ 1967 } 1968 /* 1969 * If this segment advances the known urgent pointer, 1970 * then mark the data stream. This should not happen 1971 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1972 * a FIN has been received from the remote side. 1973 * In these states we ignore the URG. 1974 * 1975 * According to RFC961 (Assigned Protocols), 1976 * the urgent pointer points to the last octet 1977 * of urgent data. We continue, however, 1978 * to consider it to indicate the first octet 1979 * of data past the urgent section as the original 1980 * spec states (in one of two places). 1981 */ 1982 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1983 tp->rcv_up = th->th_seq + th->th_urp; 1984 so->so_oobmark = so->so_rcv.sb_cc + 1985 (tp->rcv_up - tp->rcv_nxt) - 1; 1986 if (so->so_oobmark == 0) 1987 so->so_state |= SS_RCVATMARK; 1988 sohasoutofband(so); 1989 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1990 } 1991 /* 1992 * Remove out of band data so doesn't get presented to user. 1993 * This can happen independent of advancing the URG pointer, 1994 * but if two URG's are pending at once, some out-of-band 1995 * data may creep in... ick. 1996 */ 1997 if (th->th_urp <= (u_long)tlen 1998 #ifdef SO_OOBINLINE 1999 && (so->so_options & SO_OOBINLINE) == 0 2000 #endif 2001 ) 2002 tcp_pulloutofband(so, th, m, 2003 drop_hdrlen); /* hdr drop is delayed */ 2004 } else 2005 /* 2006 * If no out of band data is expected, 2007 * pull receive urgent pointer along 2008 * with the receive window. 2009 */ 2010 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2011 tp->rcv_up = tp->rcv_nxt; 2012 dodata: /* XXX */ 2013 KASSERT(headlocked, ("headlocked")); 2014 INP_INFO_WUNLOCK(&tcbinfo); 2015 headlocked = 0; 2016 /* 2017 * Process the segment text, merging it into the TCP sequencing queue, 2018 * and arranging for acknowledgment of receipt if necessary. 2019 * This process logically involves adjusting tp->rcv_wnd as data 2020 * is presented to the user (this happens in tcp_usrreq.c, 2021 * case PRU_RCVD). If a FIN has already been received on this 2022 * connection then we just ignore the text. 2023 */ 2024 if ((tlen || (thflags&TH_FIN)) && 2025 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2026 m_adj(m, drop_hdrlen); /* delayed header drop */ 2027 /* 2028 * Insert segment which inludes th into reassembly queue of tcp with 2029 * control block tp. Return TH_FIN if reassembly now includes 2030 * a segment with FIN. This handle the common case inline (segment 2031 * is the next to be received on an established connection, and the 2032 * queue is empty), avoiding linkage into and removal from the queue 2033 * and repetition of various conversions. 2034 * Set DELACK for segments received in order, but ack immediately 2035 * when segments are out of order (so fast retransmit can work). 2036 */ 2037 if (th->th_seq == tp->rcv_nxt && 2038 LIST_EMPTY(&tp->t_segq) && 2039 TCPS_HAVEESTABLISHED(tp->t_state)) { 2040 if (DELAY_ACK(tp)) 2041 callout_reset(tp->tt_delack, tcp_delacktime, 2042 tcp_timer_delack, tp); 2043 else 2044 tp->t_flags |= TF_ACKNOW; 2045 tp->rcv_nxt += tlen; 2046 thflags = th->th_flags & TH_FIN; 2047 tcpstat.tcps_rcvpack++; 2048 tcpstat.tcps_rcvbyte += tlen; 2049 ND6_HINT(tp); 2050 sbappend(&so->so_rcv, m); 2051 sorwakeup(so); 2052 } else { 2053 thflags = tcp_reass(tp, th, &tlen, m); 2054 tp->t_flags |= TF_ACKNOW; 2055 } 2056 2057 /* 2058 * Note the amount of data that peer has sent into 2059 * our window, in order to estimate the sender's 2060 * buffer size. 2061 */ 2062 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2063 } else { 2064 m_freem(m); 2065 thflags &= ~TH_FIN; 2066 } 2067 2068 /* 2069 * If FIN is received ACK the FIN and let the user know 2070 * that the connection is closing. 2071 */ 2072 if (thflags & TH_FIN) { 2073 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2074 socantrcvmore(so); 2075 /* 2076 * If connection is half-synchronized 2077 * (ie NEEDSYN flag on) then delay ACK, 2078 * so it may be piggybacked when SYN is sent. 2079 * Otherwise, since we received a FIN then no 2080 * more input can be expected, send ACK now. 2081 */ 2082 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2083 callout_reset(tp->tt_delack, tcp_delacktime, 2084 tcp_timer_delack, tp); 2085 else 2086 tp->t_flags |= TF_ACKNOW; 2087 tp->rcv_nxt++; 2088 } 2089 switch (tp->t_state) { 2090 2091 /* 2092 * In SYN_RECEIVED and ESTABLISHED STATES 2093 * enter the CLOSE_WAIT state. 2094 */ 2095 case TCPS_SYN_RECEIVED: 2096 tp->t_starttime = ticks; 2097 /*FALLTHROUGH*/ 2098 case TCPS_ESTABLISHED: 2099 tp->t_state = TCPS_CLOSE_WAIT; 2100 break; 2101 2102 /* 2103 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2104 * enter the CLOSING state. 2105 */ 2106 case TCPS_FIN_WAIT_1: 2107 tp->t_state = TCPS_CLOSING; 2108 break; 2109 2110 /* 2111 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2112 * starting the time-wait timer, turning off the other 2113 * standard timers. 2114 */ 2115 case TCPS_FIN_WAIT_2: 2116 tp->t_state = TCPS_TIME_WAIT; 2117 tcp_canceltimers(tp); 2118 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2119 if (tp->cc_recv != 0 && 2120 (ticks - tp->t_starttime) < tcp_msl) { 2121 callout_reset(tp->tt_2msl, 2122 tp->t_rxtcur * TCPTV_TWTRUNC, 2123 tcp_timer_2msl, tp); 2124 /* For transaction client, force ACK now. */ 2125 tp->t_flags |= TF_ACKNOW; 2126 } 2127 else 2128 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2129 tcp_timer_2msl, tp); 2130 soisdisconnected(so); 2131 break; 2132 2133 /* 2134 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2135 */ 2136 case TCPS_TIME_WAIT: 2137 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2138 tcp_timer_2msl, tp); 2139 break; 2140 } 2141 } 2142 #ifdef TCPDEBUG 2143 if (so->so_options & SO_DEBUG) 2144 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2145 &tcp_savetcp, 0); 2146 #endif 2147 2148 /* 2149 * Return any desired output. 2150 */ 2151 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2152 (void) tcp_output(tp); 2153 INP_UNLOCK(inp); 2154 return; 2155 2156 dropafterack: 2157 /* 2158 * Generate an ACK dropping incoming segment if it occupies 2159 * sequence space, where the ACK reflects our state. 2160 * 2161 * We can now skip the test for the RST flag since all 2162 * paths to this code happen after packets containing 2163 * RST have been dropped. 2164 * 2165 * In the SYN-RECEIVED state, don't send an ACK unless the 2166 * segment we received passes the SYN-RECEIVED ACK test. 2167 * If it fails send a RST. This breaks the loop in the 2168 * "LAND" DoS attack, and also prevents an ACK storm 2169 * between two listening ports that have been sent forged 2170 * SYN segments, each with the source address of the other. 2171 */ 2172 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2173 (SEQ_GT(tp->snd_una, th->th_ack) || 2174 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2175 rstreason = BANDLIM_RST_OPENPORT; 2176 goto dropwithreset; 2177 } 2178 #ifdef TCPDEBUG 2179 if (so->so_options & SO_DEBUG) 2180 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2181 &tcp_savetcp, 0); 2182 #endif 2183 if (headlocked) 2184 INP_INFO_WUNLOCK(&tcbinfo); 2185 m_freem(m); 2186 tp->t_flags |= TF_ACKNOW; 2187 (void) tcp_output(tp); 2188 INP_UNLOCK(inp); 2189 return; 2190 2191 dropwithreset: 2192 /* 2193 * Generate a RST, dropping incoming segment. 2194 * Make ACK acceptable to originator of segment. 2195 * Don't bother to respond if destination was broadcast/multicast. 2196 */ 2197 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2198 goto drop; 2199 #ifdef INET6 2200 if (isipv6) { 2201 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2202 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2203 goto drop; 2204 } else 2205 #endif /* INET6 */ 2206 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2207 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2208 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2209 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2210 goto drop; 2211 /* IPv6 anycast check is done at tcp6_input() */ 2212 2213 if (tp) 2214 INP_UNLOCK(inp); 2215 /* 2216 * Perform bandwidth limiting. 2217 */ 2218 if (badport_bandlim(rstreason) < 0) 2219 goto drop; 2220 2221 #ifdef TCPDEBUG 2222 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2223 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2224 &tcp_savetcp, 0); 2225 #endif 2226 if (thflags & TH_ACK) 2227 /* mtod() below is safe as long as hdr dropping is delayed */ 2228 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2229 TH_RST); 2230 else { 2231 if (thflags & TH_SYN) 2232 tlen++; 2233 /* mtod() below is safe as long as hdr dropping is delayed */ 2234 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2235 (tcp_seq)0, TH_RST|TH_ACK); 2236 } 2237 if (headlocked) 2238 INP_INFO_WUNLOCK(&tcbinfo); 2239 return; 2240 2241 drop: 2242 /* 2243 * Drop space held by incoming segment and return. 2244 */ 2245 #ifdef TCPDEBUG 2246 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2247 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2248 &tcp_savetcp, 0); 2249 #endif 2250 if (tp) 2251 INP_UNLOCK(inp); 2252 m_freem(m); 2253 if (headlocked) 2254 INP_INFO_WUNLOCK(&tcbinfo); 2255 return; 2256 } 2257 2258 /* 2259 * Parse TCP options and place in tcpopt. 2260 */ 2261 static void 2262 tcp_dooptions(to, cp, cnt, is_syn) 2263 struct tcpopt *to; 2264 u_char *cp; 2265 int cnt; 2266 { 2267 int opt, optlen; 2268 2269 to->to_flags = 0; 2270 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2271 opt = cp[0]; 2272 if (opt == TCPOPT_EOL) 2273 break; 2274 if (opt == TCPOPT_NOP) 2275 optlen = 1; 2276 else { 2277 if (cnt < 2) 2278 break; 2279 optlen = cp[1]; 2280 if (optlen < 2 || optlen > cnt) 2281 break; 2282 } 2283 switch (opt) { 2284 case TCPOPT_MAXSEG: 2285 if (optlen != TCPOLEN_MAXSEG) 2286 continue; 2287 if (!is_syn) 2288 continue; 2289 to->to_flags |= TOF_MSS; 2290 bcopy((char *)cp + 2, 2291 (char *)&to->to_mss, sizeof(to->to_mss)); 2292 to->to_mss = ntohs(to->to_mss); 2293 break; 2294 case TCPOPT_WINDOW: 2295 if (optlen != TCPOLEN_WINDOW) 2296 continue; 2297 if (! is_syn) 2298 continue; 2299 to->to_flags |= TOF_SCALE; 2300 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2301 break; 2302 case TCPOPT_TIMESTAMP: 2303 if (optlen != TCPOLEN_TIMESTAMP) 2304 continue; 2305 to->to_flags |= TOF_TS; 2306 bcopy((char *)cp + 2, 2307 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2308 to->to_tsval = ntohl(to->to_tsval); 2309 bcopy((char *)cp + 6, 2310 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2311 to->to_tsecr = ntohl(to->to_tsecr); 2312 break; 2313 case TCPOPT_CC: 2314 if (optlen != TCPOLEN_CC) 2315 continue; 2316 to->to_flags |= TOF_CC; 2317 bcopy((char *)cp + 2, 2318 (char *)&to->to_cc, sizeof(to->to_cc)); 2319 to->to_cc = ntohl(to->to_cc); 2320 break; 2321 case TCPOPT_CCNEW: 2322 if (optlen != TCPOLEN_CC) 2323 continue; 2324 if (!is_syn) 2325 continue; 2326 to->to_flags |= TOF_CCNEW; 2327 bcopy((char *)cp + 2, 2328 (char *)&to->to_cc, sizeof(to->to_cc)); 2329 to->to_cc = ntohl(to->to_cc); 2330 break; 2331 case TCPOPT_CCECHO: 2332 if (optlen != TCPOLEN_CC) 2333 continue; 2334 if (!is_syn) 2335 continue; 2336 to->to_flags |= TOF_CCECHO; 2337 bcopy((char *)cp + 2, 2338 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2339 to->to_ccecho = ntohl(to->to_ccecho); 2340 break; 2341 default: 2342 continue; 2343 } 2344 } 2345 } 2346 2347 /* 2348 * Pull out of band byte out of a segment so 2349 * it doesn't appear in the user's data queue. 2350 * It is still reflected in the segment length for 2351 * sequencing purposes. 2352 */ 2353 static void 2354 tcp_pulloutofband(so, th, m, off) 2355 struct socket *so; 2356 struct tcphdr *th; 2357 register struct mbuf *m; 2358 int off; /* delayed to be droped hdrlen */ 2359 { 2360 int cnt = off + th->th_urp - 1; 2361 2362 while (cnt >= 0) { 2363 if (m->m_len > cnt) { 2364 char *cp = mtod(m, caddr_t) + cnt; 2365 struct tcpcb *tp = sototcpcb(so); 2366 2367 tp->t_iobc = *cp; 2368 tp->t_oobflags |= TCPOOB_HAVEDATA; 2369 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2370 m->m_len--; 2371 if (m->m_flags & M_PKTHDR) 2372 m->m_pkthdr.len--; 2373 return; 2374 } 2375 cnt -= m->m_len; 2376 m = m->m_next; 2377 if (m == 0) 2378 break; 2379 } 2380 panic("tcp_pulloutofband"); 2381 } 2382 2383 /* 2384 * Collect new round-trip time estimate 2385 * and update averages and current timeout. 2386 */ 2387 static void 2388 tcp_xmit_timer(tp, rtt) 2389 register struct tcpcb *tp; 2390 int rtt; 2391 { 2392 register int delta; 2393 2394 tcpstat.tcps_rttupdated++; 2395 tp->t_rttupdated++; 2396 if (tp->t_srtt != 0) { 2397 /* 2398 * srtt is stored as fixed point with 5 bits after the 2399 * binary point (i.e., scaled by 8). The following magic 2400 * is equivalent to the smoothing algorithm in rfc793 with 2401 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2402 * point). Adjust rtt to origin 0. 2403 */ 2404 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2405 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2406 2407 if ((tp->t_srtt += delta) <= 0) 2408 tp->t_srtt = 1; 2409 2410 /* 2411 * We accumulate a smoothed rtt variance (actually, a 2412 * smoothed mean difference), then set the retransmit 2413 * timer to smoothed rtt + 4 times the smoothed variance. 2414 * rttvar is stored as fixed point with 4 bits after the 2415 * binary point (scaled by 16). The following is 2416 * equivalent to rfc793 smoothing with an alpha of .75 2417 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2418 * rfc793's wired-in beta. 2419 */ 2420 if (delta < 0) 2421 delta = -delta; 2422 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2423 if ((tp->t_rttvar += delta) <= 0) 2424 tp->t_rttvar = 1; 2425 } else { 2426 /* 2427 * No rtt measurement yet - use the unsmoothed rtt. 2428 * Set the variance to half the rtt (so our first 2429 * retransmit happens at 3*rtt). 2430 */ 2431 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2432 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2433 } 2434 tp->t_rtttime = 0; 2435 tp->t_rxtshift = 0; 2436 2437 /* 2438 * the retransmit should happen at rtt + 4 * rttvar. 2439 * Because of the way we do the smoothing, srtt and rttvar 2440 * will each average +1/2 tick of bias. When we compute 2441 * the retransmit timer, we want 1/2 tick of rounding and 2442 * 1 extra tick because of +-1/2 tick uncertainty in the 2443 * firing of the timer. The bias will give us exactly the 2444 * 1.5 tick we need. But, because the bias is 2445 * statistical, we have to test that we don't drop below 2446 * the minimum feasible timer (which is 2 ticks). 2447 */ 2448 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2449 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2450 2451 /* 2452 * We received an ack for a packet that wasn't retransmitted; 2453 * it is probably safe to discard any error indications we've 2454 * received recently. This isn't quite right, but close enough 2455 * for now (a route might have failed after we sent a segment, 2456 * and the return path might not be symmetrical). 2457 */ 2458 tp->t_softerror = 0; 2459 } 2460 2461 /* 2462 * Determine a reasonable value for maxseg size. 2463 * If the route is known, check route for mtu. 2464 * If none, use an mss that can be handled on the outgoing 2465 * interface without forcing IP to fragment; if bigger than 2466 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2467 * to utilize large mbufs. If no route is found, route has no mtu, 2468 * or the destination isn't local, use a default, hopefully conservative 2469 * size (usually 512 or the default IP max size, but no more than the mtu 2470 * of the interface), as we can't discover anything about intervening 2471 * gateways or networks. We also initialize the congestion/slow start 2472 * window to be a single segment if the destination isn't local. 2473 * While looking at the routing entry, we also initialize other path-dependent 2474 * parameters from pre-set or cached values in the routing entry. 2475 * 2476 * Also take into account the space needed for options that we 2477 * send regularly. Make maxseg shorter by that amount to assure 2478 * that we can send maxseg amount of data even when the options 2479 * are present. Store the upper limit of the length of options plus 2480 * data in maxopd. 2481 * 2482 * NOTE that this routine is only called when we process an incoming 2483 * segment, for outgoing segments only tcp_mssopt is called. 2484 * 2485 * In case of T/TCP, we call this routine during implicit connection 2486 * setup as well (offer = -1), to initialize maxseg from the cached 2487 * MSS of our peer. 2488 */ 2489 void 2490 tcp_mss(tp, offer) 2491 struct tcpcb *tp; 2492 int offer; 2493 { 2494 register struct rtentry *rt; 2495 struct ifnet *ifp; 2496 register int rtt, mss; 2497 u_long bufsize; 2498 struct inpcb *inp; 2499 struct socket *so; 2500 struct rmxp_tao *taop; 2501 int origoffer = offer; 2502 #ifdef INET6 2503 int isipv6; 2504 int min_protoh; 2505 #endif 2506 2507 inp = tp->t_inpcb; 2508 #ifdef INET6 2509 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2510 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2511 : sizeof (struct tcpiphdr); 2512 #else 2513 #define min_protoh (sizeof (struct tcpiphdr)) 2514 #endif 2515 #ifdef INET6 2516 if (isipv6) 2517 rt = tcp_rtlookup6(&inp->inp_inc); 2518 else 2519 #endif 2520 rt = tcp_rtlookup(&inp->inp_inc); 2521 if (rt == NULL) { 2522 tp->t_maxopd = tp->t_maxseg = 2523 #ifdef INET6 2524 isipv6 ? tcp_v6mssdflt : 2525 #endif /* INET6 */ 2526 tcp_mssdflt; 2527 return; 2528 } 2529 ifp = rt->rt_ifp; 2530 so = inp->inp_socket; 2531 2532 taop = rmx_taop(rt->rt_rmx); 2533 /* 2534 * Offer == -1 means that we didn't receive SYN yet, 2535 * use cached value in that case; 2536 */ 2537 if (offer == -1) 2538 offer = taop->tao_mssopt; 2539 /* 2540 * Offer == 0 means that there was no MSS on the SYN segment, 2541 * in this case we use tcp_mssdflt. 2542 */ 2543 if (offer == 0) 2544 offer = 2545 #ifdef INET6 2546 isipv6 ? tcp_v6mssdflt : 2547 #endif /* INET6 */ 2548 tcp_mssdflt; 2549 else 2550 /* 2551 * Sanity check: make sure that maxopd will be large 2552 * enough to allow some data on segments even is the 2553 * all the option space is used (40bytes). Otherwise 2554 * funny things may happen in tcp_output. 2555 */ 2556 offer = max(offer, 64); 2557 taop->tao_mssopt = offer; 2558 2559 /* 2560 * While we're here, check if there's an initial rtt 2561 * or rttvar. Convert from the route-table units 2562 * to scaled multiples of the slow timeout timer. 2563 */ 2564 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2565 /* 2566 * XXX the lock bit for RTT indicates that the value 2567 * is also a minimum value; this is subject to time. 2568 */ 2569 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2570 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2571 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2572 tcpstat.tcps_usedrtt++; 2573 if (rt->rt_rmx.rmx_rttvar) { 2574 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2575 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2576 tcpstat.tcps_usedrttvar++; 2577 } else { 2578 /* default variation is +- 1 rtt */ 2579 tp->t_rttvar = 2580 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2581 } 2582 TCPT_RANGESET(tp->t_rxtcur, 2583 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2584 tp->t_rttmin, TCPTV_REXMTMAX); 2585 } 2586 /* 2587 * if there's an mtu associated with the route, use it 2588 * else, use the link mtu. 2589 */ 2590 if (rt->rt_rmx.rmx_mtu) 2591 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2592 else 2593 { 2594 mss = 2595 #ifdef INET6 2596 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2597 #endif 2598 ifp->if_mtu 2599 #ifdef INET6 2600 ) 2601 #endif 2602 - min_protoh; 2603 #ifdef INET6 2604 if (isipv6) { 2605 if (!in6_localaddr(&inp->in6p_faddr)) 2606 mss = min(mss, tcp_v6mssdflt); 2607 } else 2608 #endif 2609 if (!in_localaddr(inp->inp_faddr)) 2610 mss = min(mss, tcp_mssdflt); 2611 } 2612 mss = min(mss, offer); 2613 /* 2614 * maxopd stores the maximum length of data AND options 2615 * in a segment; maxseg is the amount of data in a normal 2616 * segment. We need to store this value (maxopd) apart 2617 * from maxseg, because now every segment carries options 2618 * and thus we normally have somewhat less data in segments. 2619 */ 2620 tp->t_maxopd = mss; 2621 2622 /* 2623 * In case of T/TCP, origoffer==-1 indicates, that no segments 2624 * were received yet. In this case we just guess, otherwise 2625 * we do the same as before T/TCP. 2626 */ 2627 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2628 (origoffer == -1 || 2629 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2630 mss -= TCPOLEN_TSTAMP_APPA; 2631 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2632 (origoffer == -1 || 2633 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2634 mss -= TCPOLEN_CC_APPA; 2635 2636 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2637 if (mss > MCLBYTES) 2638 mss &= ~(MCLBYTES-1); 2639 #else 2640 if (mss > MCLBYTES) 2641 mss = mss / MCLBYTES * MCLBYTES; 2642 #endif 2643 /* 2644 * If there's a pipesize, change the socket buffer 2645 * to that size. Make the socket buffers an integral 2646 * number of mss units; if the mss is larger than 2647 * the socket buffer, decrease the mss. 2648 */ 2649 #ifdef RTV_SPIPE 2650 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2651 #endif 2652 bufsize = so->so_snd.sb_hiwat; 2653 if (bufsize < mss) 2654 mss = bufsize; 2655 else { 2656 bufsize = roundup(bufsize, mss); 2657 if (bufsize > sb_max) 2658 bufsize = sb_max; 2659 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2660 } 2661 tp->t_maxseg = mss; 2662 2663 #ifdef RTV_RPIPE 2664 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2665 #endif 2666 bufsize = so->so_rcv.sb_hiwat; 2667 if (bufsize > mss) { 2668 bufsize = roundup(bufsize, mss); 2669 if (bufsize > sb_max) 2670 bufsize = sb_max; 2671 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2672 } 2673 2674 /* 2675 * Set the slow-start flight size depending on whether this 2676 * is a local network or not. 2677 */ 2678 if ( 2679 #ifdef INET6 2680 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2681 (!isipv6 && 2682 #endif 2683 in_localaddr(inp->inp_faddr) 2684 #ifdef INET6 2685 ) 2686 #endif 2687 ) 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; 2714 int min_protoh; 2715 #endif 2716 2717 #ifdef INET6 2718 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2719 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2720 : sizeof (struct tcpiphdr); 2721 #else 2722 #define min_protoh (sizeof (struct tcpiphdr)) 2723 #endif 2724 #ifdef INET6 2725 if (isipv6) 2726 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2727 else 2728 #endif /* INET6 */ 2729 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2730 if (rt == NULL) 2731 return 2732 #ifdef INET6 2733 isipv6 ? tcp_v6mssdflt : 2734 #endif /* INET6 */ 2735 tcp_mssdflt; 2736 2737 return rt->rt_ifp->if_mtu - min_protoh; 2738 } 2739 2740 2741 /* 2742 * Checks for partial ack. If partial ack arrives, force the retransmission 2743 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2744 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 2745 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2746 */ 2747 static int 2748 tcp_newreno(tp, th) 2749 struct tcpcb *tp; 2750 struct tcphdr *th; 2751 { 2752 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2753 tcp_seq onxt = tp->snd_nxt; 2754 u_long ocwnd = tp->snd_cwnd; 2755 2756 callout_stop(tp->tt_rexmt); 2757 tp->t_rtttime = 0; 2758 tp->snd_nxt = th->th_ack; 2759 /* 2760 * Set snd_cwnd to one segment beyond acknowledged offset 2761 * (tp->snd_una has not yet been updated when this function 2762 * is called) 2763 */ 2764 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2765 (void) tcp_output(tp); 2766 tp->snd_cwnd = ocwnd; 2767 if (SEQ_GT(onxt, tp->snd_nxt)) 2768 tp->snd_nxt = onxt; 2769 /* 2770 * Partial window deflation. Relies on fact that tp->snd_una 2771 * not updated yet. 2772 */ 2773 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2774 return (1); 2775 } 2776 return (0); 2777 } 2778