1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 34 * $FreeBSD$ 35 */ 36 37 #include "opt_ipfw.h" /* for ipfw_fwd */ 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_mac.h" 41 #include "opt_tcpdebug.h" 42 #include "opt_tcp_input.h" 43 44 #include <sys/param.h> 45 #include <sys/kernel.h> 46 #include <sys/mac.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #include <sys/proc.h> /* for proc0 declaration */ 50 #include <sys/protosw.h> 51 #include <sys/signalvar.h> 52 #include <sys/socket.h> 53 #include <sys/socketvar.h> 54 #include <sys/sysctl.h> 55 #include <sys/syslog.h> 56 #include <sys/systm.h> 57 58 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 59 60 #include <net/if.h> 61 #include <net/route.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_pcb.h> 65 #include <netinet/in_systm.h> 66 #include <netinet/in_var.h> 67 #include <netinet/ip.h> 68 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 69 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 70 #include <netinet/ip_var.h> 71 #include <netinet/ip6.h> 72 #include <netinet/icmp6.h> 73 #include <netinet6/in6_pcb.h> 74 #include <netinet6/ip6_var.h> 75 #include <netinet6/nd6.h> 76 #include <netinet/tcp.h> 77 #include <netinet/tcp_fsm.h> 78 #include <netinet/tcp_seq.h> 79 #include <netinet/tcp_timer.h> 80 #include <netinet/tcp_var.h> 81 #include <netinet6/tcp6_var.h> 82 #include <netinet/tcpip.h> 83 #ifdef TCPDEBUG 84 #include <netinet/tcp_debug.h> 85 #endif /* TCPDEBUG */ 86 87 #ifdef FAST_IPSEC 88 #include <netipsec/ipsec.h> 89 #include <netipsec/ipsec6.h> 90 #endif /*FAST_IPSEC*/ 91 92 #ifdef IPSEC 93 #include <netinet6/ipsec.h> 94 #include <netinet6/ipsec6.h> 95 #include <netkey/key.h> 96 #endif /*IPSEC*/ 97 98 #include <machine/in_cksum.h> 99 100 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 101 102 static const int tcprexmtthresh = 3; 103 tcp_cc tcp_ccgen; 104 105 struct tcpstat tcpstat; 106 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 107 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 108 109 static int log_in_vain = 0; 110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 111 &log_in_vain, 0, "Log all incoming TCP connections"); 112 113 static int blackhole = 0; 114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 115 &blackhole, 0, "Do not send RST when dropping refused connections"); 116 117 int tcp_delack_enabled = 1; 118 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 119 &tcp_delack_enabled, 0, 120 "Delay ACK to try and piggyback it onto a data packet"); 121 122 #ifdef TCP_DROP_SYNFIN 123 static int drop_synfin = 0; 124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 125 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 126 #endif 127 128 static int tcp_do_rfc3042 = 0; 129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 130 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 131 132 static int tcp_do_rfc3390 = 0; 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 134 &tcp_do_rfc3390, 0, 135 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 136 137 struct inpcbhead tcb; 138 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 139 struct inpcbinfo tcbinfo; 140 struct mtx *tcbinfo_mtx; 141 142 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 143 static void tcp_pulloutofband(struct socket *, 144 struct tcphdr *, struct mbuf *, int); 145 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 146 struct mbuf *); 147 static void tcp_xmit_timer(struct tcpcb *, int); 148 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 149 static int tcp_timewait(struct tcptw *, struct tcpopt *, 150 struct tcphdr *, struct mbuf *, int); 151 152 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 153 #ifdef INET6 154 #define ND6_HINT(tp) \ 155 do { \ 156 if ((tp) && (tp)->t_inpcb && \ 157 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 158 (tp)->t_inpcb->in6p_route.ro_rt) \ 159 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 160 } while (0) 161 #else 162 #define ND6_HINT(tp) 163 #endif 164 165 /* 166 * Indicate whether this ack should be delayed. We can delay the ack if 167 * - there is no delayed ack timer in progress and 168 * - our last ack wasn't a 0-sized window. We never want to delay 169 * the ack that opens up a 0-sized window and 170 * - delayed acks are enabled or 171 * - this is a half-synchronized T/TCP connection. 172 */ 173 #define DELAY_ACK(tp) \ 174 ((!callout_active(tp->tt_delack) && \ 175 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 176 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 177 178 static int 179 tcp_reass(tp, th, tlenp, m) 180 register struct tcpcb *tp; 181 register struct tcphdr *th; 182 int *tlenp; 183 struct mbuf *m; 184 { 185 struct tseg_qent *q; 186 struct tseg_qent *p = NULL; 187 struct tseg_qent *nq; 188 struct tseg_qent *te; 189 struct socket *so = tp->t_inpcb->inp_socket; 190 int flags; 191 192 /* 193 * Call with th==0 after become established to 194 * force pre-ESTABLISHED data up to user socket. 195 */ 196 if (th == 0) 197 goto present; 198 199 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 200 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 201 M_NOWAIT); 202 if (te == NULL) { 203 tcpstat.tcps_rcvmemdrop++; 204 m_freem(m); 205 return (0); 206 } 207 208 /* 209 * Find a segment which begins after this one does. 210 */ 211 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 212 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 213 break; 214 p = q; 215 } 216 217 /* 218 * If there is a preceding segment, it may provide some of 219 * our data already. If so, drop the data from the incoming 220 * segment. If it provides all of our data, drop us. 221 */ 222 if (p != NULL) { 223 register int i; 224 /* conversion to int (in i) handles seq wraparound */ 225 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 226 if (i > 0) { 227 if (i >= *tlenp) { 228 tcpstat.tcps_rcvduppack++; 229 tcpstat.tcps_rcvdupbyte += *tlenp; 230 m_freem(m); 231 FREE(te, M_TSEGQ); 232 /* 233 * Try to present any queued data 234 * at the left window edge to the user. 235 * This is needed after the 3-WHS 236 * completes. 237 */ 238 goto present; /* ??? */ 239 } 240 m_adj(m, i); 241 *tlenp -= i; 242 th->th_seq += i; 243 } 244 } 245 tcpstat.tcps_rcvoopack++; 246 tcpstat.tcps_rcvoobyte += *tlenp; 247 248 /* 249 * While we overlap succeeding segments trim them or, 250 * if they are completely covered, dequeue them. 251 */ 252 while (q) { 253 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 254 if (i <= 0) 255 break; 256 if (i < q->tqe_len) { 257 q->tqe_th->th_seq += i; 258 q->tqe_len -= i; 259 m_adj(q->tqe_m, i); 260 break; 261 } 262 263 nq = LIST_NEXT(q, tqe_q); 264 LIST_REMOVE(q, tqe_q); 265 m_freem(q->tqe_m); 266 FREE(q, M_TSEGQ); 267 q = nq; 268 } 269 270 /* Insert the new segment queue entry into place. */ 271 te->tqe_m = m; 272 te->tqe_th = th; 273 te->tqe_len = *tlenp; 274 275 if (p == NULL) { 276 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 277 } else { 278 LIST_INSERT_AFTER(p, te, tqe_q); 279 } 280 281 present: 282 /* 283 * Present data to user, advancing rcv_nxt through 284 * completed sequence space. 285 */ 286 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 287 return (0); 288 q = LIST_FIRST(&tp->t_segq); 289 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 290 return (0); 291 do { 292 tp->rcv_nxt += q->tqe_len; 293 flags = q->tqe_th->th_flags & TH_FIN; 294 nq = LIST_NEXT(q, tqe_q); 295 LIST_REMOVE(q, tqe_q); 296 if (so->so_state & SS_CANTRCVMORE) 297 m_freem(q->tqe_m); 298 else 299 sbappend(&so->so_rcv, q->tqe_m); 300 FREE(q, M_TSEGQ); 301 q = nq; 302 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 303 ND6_HINT(tp); 304 sorwakeup(so); 305 return (flags); 306 } 307 308 /* 309 * TCP input routine, follows pages 65-76 of the 310 * protocol specification dated September, 1981 very closely. 311 */ 312 #ifdef INET6 313 int 314 tcp6_input(mp, offp, proto) 315 struct mbuf **mp; 316 int *offp, proto; 317 { 318 register struct mbuf *m = *mp; 319 struct in6_ifaddr *ia6; 320 321 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 322 323 /* 324 * draft-itojun-ipv6-tcp-to-anycast 325 * better place to put this in? 326 */ 327 ia6 = ip6_getdstifaddr(m); 328 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 329 struct ip6_hdr *ip6; 330 331 ip6 = mtod(m, struct ip6_hdr *); 332 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 333 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 334 return IPPROTO_DONE; 335 } 336 337 tcp_input(m, *offp); 338 return IPPROTO_DONE; 339 } 340 #endif 341 342 void 343 tcp_input(m, off0) 344 register struct mbuf *m; 345 int off0; 346 { 347 register struct tcphdr *th; 348 register struct ip *ip = NULL; 349 register struct ipovly *ipov; 350 register struct inpcb *inp = NULL; 351 u_char *optp = NULL; 352 int optlen = 0; 353 int len, tlen, off; 354 int drop_hdrlen; 355 register struct tcpcb *tp = 0; 356 register int thflags; 357 struct socket *so = 0; 358 int todrop, acked, ourfinisacked, needoutput = 0; 359 u_long tiwin; 360 struct tcpopt to; /* options in this segment */ 361 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 362 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 363 int headlocked = 0; 364 struct sockaddr_in *next_hop = NULL; 365 int rstreason; /* For badport_bandlim accounting purposes */ 366 367 struct ip6_hdr *ip6 = NULL; 368 #ifdef INET6 369 int isipv6; 370 #else 371 const int isipv6 = 0; 372 #endif 373 374 #ifdef TCPDEBUG 375 /* 376 * The size of tcp_saveipgen must be the size of the max ip header, 377 * now IPv6. 378 */ 379 u_char tcp_saveipgen[40]; 380 struct tcphdr tcp_savetcp; 381 short ostate = 0; 382 #endif 383 384 /* Grab info from MT_TAG mbufs prepended to the chain. */ 385 for (;m && m->m_type == MT_TAG; m = m->m_next) { 386 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 387 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 388 } 389 #ifdef INET6 390 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 391 #endif 392 bzero((char *)&to, sizeof(to)); 393 394 tcpstat.tcps_rcvtotal++; 395 396 if (isipv6) { 397 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 398 ip6 = mtod(m, struct ip6_hdr *); 399 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 400 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 401 tcpstat.tcps_rcvbadsum++; 402 goto drop; 403 } 404 th = (struct tcphdr *)((caddr_t)ip6 + off0); 405 406 /* 407 * Be proactive about unspecified IPv6 address in source. 408 * As we use all-zero to indicate unbounded/unconnected pcb, 409 * unspecified IPv6 address can be used to confuse us. 410 * 411 * Note that packets with unspecified IPv6 destination is 412 * already dropped in ip6_input. 413 */ 414 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 415 /* XXX stat */ 416 goto drop; 417 } 418 } else { 419 /* 420 * Get IP and TCP header together in first mbuf. 421 * Note: IP leaves IP header in first mbuf. 422 */ 423 if (off0 > sizeof (struct ip)) { 424 ip_stripoptions(m, (struct mbuf *)0); 425 off0 = sizeof(struct ip); 426 } 427 if (m->m_len < sizeof (struct tcpiphdr)) { 428 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 429 tcpstat.tcps_rcvshort++; 430 return; 431 } 432 } 433 ip = mtod(m, struct ip *); 434 ipov = (struct ipovly *)ip; 435 th = (struct tcphdr *)((caddr_t)ip + off0); 436 tlen = ip->ip_len; 437 438 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 439 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 440 th->th_sum = m->m_pkthdr.csum_data; 441 else 442 th->th_sum = in_pseudo(ip->ip_src.s_addr, 443 ip->ip_dst.s_addr, 444 htonl(m->m_pkthdr.csum_data + 445 ip->ip_len + 446 IPPROTO_TCP)); 447 th->th_sum ^= 0xffff; 448 } else { 449 /* 450 * Checksum extended TCP header and data. 451 */ 452 len = sizeof (struct ip) + tlen; 453 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 454 ipov->ih_len = (u_short)tlen; 455 ipov->ih_len = htons(ipov->ih_len); 456 th->th_sum = in_cksum(m, len); 457 } 458 if (th->th_sum) { 459 tcpstat.tcps_rcvbadsum++; 460 goto drop; 461 } 462 #ifdef INET6 463 /* Re-initialization for later version check */ 464 ip->ip_v = IPVERSION; 465 #endif 466 } 467 468 /* 469 * Check that TCP offset makes sense, 470 * pull out TCP options and adjust length. XXX 471 */ 472 off = th->th_off << 2; 473 if (off < sizeof (struct tcphdr) || off > tlen) { 474 tcpstat.tcps_rcvbadoff++; 475 goto drop; 476 } 477 tlen -= off; /* tlen is used instead of ti->ti_len */ 478 if (off > sizeof (struct tcphdr)) { 479 if (isipv6) { 480 IP6_EXTHDR_CHECK(m, off0, off, ); 481 ip6 = mtod(m, struct ip6_hdr *); 482 th = (struct tcphdr *)((caddr_t)ip6 + off0); 483 } else { 484 if (m->m_len < sizeof(struct ip) + off) { 485 if ((m = m_pullup(m, sizeof (struct ip) + off)) 486 == 0) { 487 tcpstat.tcps_rcvshort++; 488 return; 489 } 490 ip = mtod(m, struct ip *); 491 ipov = (struct ipovly *)ip; 492 th = (struct tcphdr *)((caddr_t)ip + off0); 493 } 494 } 495 optlen = off - sizeof (struct tcphdr); 496 optp = (u_char *)(th + 1); 497 } 498 thflags = th->th_flags; 499 500 #ifdef TCP_DROP_SYNFIN 501 /* 502 * If the drop_synfin option is enabled, drop all packets with 503 * both the SYN and FIN bits set. This prevents e.g. nmap from 504 * identifying the TCP/IP stack. 505 * 506 * This is a violation of the TCP specification. 507 */ 508 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 509 goto drop; 510 #endif 511 512 /* 513 * Convert TCP protocol specific fields to host format. 514 */ 515 th->th_seq = ntohl(th->th_seq); 516 th->th_ack = ntohl(th->th_ack); 517 th->th_win = ntohs(th->th_win); 518 th->th_urp = ntohs(th->th_urp); 519 520 /* 521 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 522 * until after ip6_savecontrol() is called and before other functions 523 * which don't want those proto headers. 524 * Because ip6_savecontrol() is going to parse the mbuf to 525 * search for data to be passed up to user-land, it wants mbuf 526 * parameters to be unchanged. 527 * XXX: the call of ip6_savecontrol() has been obsoleted based on 528 * latest version of the advanced API (20020110). 529 */ 530 drop_hdrlen = off0 + off; 531 532 /* 533 * Locate pcb for segment. 534 */ 535 INP_INFO_WLOCK(&tcbinfo); 536 headlocked = 1; 537 findpcb: 538 /* IPFIREWALL_FORWARD section */ 539 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 540 /* 541 * Transparently forwarded. Pretend to be the destination. 542 * already got one like this? 543 */ 544 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 545 ip->ip_dst, th->th_dport, 546 0, m->m_pkthdr.rcvif); 547 if (!inp) { 548 /* It's new. Try find the ambushing socket. */ 549 inp = in_pcblookup_hash(&tcbinfo, 550 ip->ip_src, th->th_sport, 551 next_hop->sin_addr, 552 next_hop->sin_port ? 553 ntohs(next_hop->sin_port) : 554 th->th_dport, 555 1, m->m_pkthdr.rcvif); 556 } 557 } else { 558 if (isipv6) 559 inp = in6_pcblookup_hash(&tcbinfo, 560 &ip6->ip6_src, th->th_sport, 561 &ip6->ip6_dst, th->th_dport, 562 1, m->m_pkthdr.rcvif); 563 else 564 inp = in_pcblookup_hash(&tcbinfo, 565 ip->ip_src, th->th_sport, 566 ip->ip_dst, th->th_dport, 567 1, m->m_pkthdr.rcvif); 568 } 569 570 #ifdef IPSEC 571 if (isipv6) { 572 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 573 ipsec6stat.in_polvio++; 574 goto drop; 575 } 576 } else { 577 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 578 ipsecstat.in_polvio++; 579 goto drop; 580 } 581 } 582 #endif 583 #ifdef FAST_IPSEC 584 if (isipv6) { 585 if (inp != NULL && ipsec6_in_reject(m, inp)) { 586 goto drop; 587 } 588 } else 589 if (inp != NULL && ipsec4_in_reject(m, inp)) { 590 goto drop; 591 } 592 #endif /*FAST_IPSEC*/ 593 594 /* 595 * If the state is CLOSED (i.e., TCB does not exist) then 596 * all data in the incoming segment is discarded. 597 * If the TCB exists but is in CLOSED state, it is embryonic, 598 * but should either do a listen or a connect soon. 599 */ 600 if (inp == NULL) { 601 if (log_in_vain) { 602 #ifdef INET6 603 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 604 #else 605 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 606 #endif 607 608 if (isipv6) { 609 strcpy(dbuf, "["); 610 strcpy(sbuf, "["); 611 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 612 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 613 strcat(dbuf, "]"); 614 strcat(sbuf, "]"); 615 } else { 616 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 617 strcpy(sbuf, inet_ntoa(ip->ip_src)); 618 } 619 switch (log_in_vain) { 620 case 1: 621 if ((thflags & TH_SYN) == 0) 622 break; 623 /* FALLTHROUGH */ 624 case 2: 625 log(LOG_INFO, 626 "Connection attempt to TCP %s:%d " 627 "from %s:%d flags:0x%02x\n", 628 dbuf, ntohs(th->th_dport), sbuf, 629 ntohs(th->th_sport), thflags); 630 break; 631 default: 632 break; 633 } 634 } 635 if (blackhole) { 636 switch (blackhole) { 637 case 1: 638 if (thflags & TH_SYN) 639 goto drop; 640 break; 641 case 2: 642 goto drop; 643 default: 644 goto drop; 645 } 646 } 647 rstreason = BANDLIM_RST_CLOSEDPORT; 648 goto dropwithreset; 649 } 650 INP_LOCK(inp); 651 if (inp->inp_vflag & INP_TIMEWAIT) { 652 /* 653 * The only option of relevance is TOF_CC, and only if 654 * present in a SYN segment. See tcp_timewait(). 655 */ 656 if (thflags & TH_SYN) 657 tcp_dooptions(&to, optp, optlen, 1); 658 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 659 &to, th, m, tlen)) 660 goto findpcb; 661 /* 662 * tcp_timewait unlocks inp. 663 */ 664 INP_INFO_WUNLOCK(&tcbinfo); 665 return; 666 } 667 tp = intotcpcb(inp); 668 if (tp == 0) { 669 INP_UNLOCK(inp); 670 rstreason = BANDLIM_RST_CLOSEDPORT; 671 goto dropwithreset; 672 } 673 if (tp->t_state == TCPS_CLOSED) 674 goto drop; 675 676 /* Unscale the window into a 32-bit value. */ 677 if ((thflags & TH_SYN) == 0) 678 tiwin = th->th_win << tp->snd_scale; 679 else 680 tiwin = th->th_win; 681 682 so = inp->inp_socket; 683 #ifdef MAC 684 if (mac_check_socket_deliver(so, m)) 685 goto drop; 686 #endif 687 #ifdef TCPDEBUG 688 if (so->so_options & SO_DEBUG) { 689 ostate = tp->t_state; 690 if (isipv6) 691 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 692 else 693 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 694 tcp_savetcp = *th; 695 } 696 #endif 697 if (so->so_options & SO_ACCEPTCONN) { 698 struct in_conninfo inc; 699 700 #ifdef INET6 701 inc.inc_isipv6 = isipv6; 702 #endif 703 if (isipv6) { 704 inc.inc6_faddr = ip6->ip6_src; 705 inc.inc6_laddr = ip6->ip6_dst; 706 inc.inc6_route.ro_rt = NULL; /* XXX */ 707 } else { 708 inc.inc_faddr = ip->ip_src; 709 inc.inc_laddr = ip->ip_dst; 710 inc.inc_route.ro_rt = NULL; /* XXX */ 711 } 712 inc.inc_fport = th->th_sport; 713 inc.inc_lport = th->th_dport; 714 715 /* 716 * If the state is LISTEN then ignore segment if it contains 717 * a RST. If the segment contains an ACK then it is bad and 718 * send a RST. If it does not contain a SYN then it is not 719 * interesting; drop it. 720 * 721 * If the state is SYN_RECEIVED (syncache) and seg contains 722 * an ACK, but not for our SYN/ACK, send a RST. If the seg 723 * contains a RST, check the sequence number to see if it 724 * is a valid reset segment. 725 */ 726 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 727 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 728 if (!syncache_expand(&inc, th, &so, m)) { 729 /* 730 * No syncache entry, or ACK was not 731 * for our SYN/ACK. Send a RST. 732 */ 733 tcpstat.tcps_badsyn++; 734 rstreason = BANDLIM_RST_OPENPORT; 735 goto dropwithreset; 736 } 737 if (so == NULL) { 738 /* 739 * Could not complete 3-way handshake, 740 * connection is being closed down, and 741 * syncache will free mbuf. 742 */ 743 INP_UNLOCK(inp); 744 INP_INFO_WUNLOCK(&tcbinfo); 745 return; 746 } 747 /* 748 * Socket is created in state SYN_RECEIVED. 749 * Continue processing segment. 750 */ 751 INP_UNLOCK(inp); 752 inp = sotoinpcb(so); 753 INP_LOCK(inp); 754 tp = intotcpcb(inp); 755 /* 756 * This is what would have happened in 757 * tcp_output() when the SYN,ACK was sent. 758 */ 759 tp->snd_up = tp->snd_una; 760 tp->snd_max = tp->snd_nxt = tp->iss + 1; 761 tp->last_ack_sent = tp->rcv_nxt; 762 /* 763 * RFC1323: The window in SYN & SYN/ACK 764 * segments is never scaled. 765 */ 766 tp->snd_wnd = tiwin; /* unscaled */ 767 goto after_listen; 768 } 769 if (thflags & TH_RST) { 770 syncache_chkrst(&inc, th); 771 goto drop; 772 } 773 if (thflags & TH_ACK) { 774 syncache_badack(&inc); 775 tcpstat.tcps_badsyn++; 776 rstreason = BANDLIM_RST_OPENPORT; 777 goto dropwithreset; 778 } 779 goto drop; 780 } 781 782 /* 783 * Segment's flags are (SYN) or (SYN|FIN). 784 */ 785 #ifdef INET6 786 /* 787 * If deprecated address is forbidden, 788 * we do not accept SYN to deprecated interface 789 * address to prevent any new inbound connection from 790 * getting established. 791 * When we do not accept SYN, we send a TCP RST, 792 * with deprecated source address (instead of dropping 793 * it). We compromise it as it is much better for peer 794 * to send a RST, and RST will be the final packet 795 * for the exchange. 796 * 797 * If we do not forbid deprecated addresses, we accept 798 * the SYN packet. RFC2462 does not suggest dropping 799 * SYN in this case. 800 * If we decipher RFC2462 5.5.4, it says like this: 801 * 1. use of deprecated addr with existing 802 * communication is okay - "SHOULD continue to be 803 * used" 804 * 2. use of it with new communication: 805 * (2a) "SHOULD NOT be used if alternate address 806 * with sufficient scope is available" 807 * (2b) nothing mentioned otherwise. 808 * Here we fall into (2b) case as we have no choice in 809 * our source address selection - we must obey the peer. 810 * 811 * The wording in RFC2462 is confusing, and there are 812 * multiple description text for deprecated address 813 * handling - worse, they are not exactly the same. 814 * I believe 5.5.4 is the best one, so we follow 5.5.4. 815 */ 816 if (isipv6 && !ip6_use_deprecated) { 817 struct in6_ifaddr *ia6; 818 819 if ((ia6 = ip6_getdstifaddr(m)) && 820 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 821 INP_UNLOCK(inp); 822 tp = NULL; 823 rstreason = BANDLIM_RST_OPENPORT; 824 goto dropwithreset; 825 } 826 } 827 #endif 828 /* 829 * If it is from this socket, drop it, it must be forged. 830 * Don't bother responding if the destination was a broadcast. 831 */ 832 if (th->th_dport == th->th_sport) { 833 if (isipv6) { 834 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 835 &ip6->ip6_src)) 836 goto drop; 837 } else { 838 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 839 goto drop; 840 } 841 } 842 /* 843 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 844 * 845 * Note that it is quite possible to receive unicast 846 * link-layer packets with a broadcast IP address. Use 847 * in_broadcast() to find them. 848 */ 849 if (m->m_flags & (M_BCAST|M_MCAST)) 850 goto drop; 851 if (isipv6) { 852 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 853 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 854 goto drop; 855 } else { 856 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 857 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 858 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 859 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 860 goto drop; 861 } 862 /* 863 * SYN appears to be valid; create compressed TCP state 864 * for syncache, or perform t/tcp connection. 865 */ 866 if (so->so_qlen <= so->so_qlimit) { 867 tcp_dooptions(&to, optp, optlen, 1); 868 if (!syncache_add(&inc, &to, th, &so, m)) 869 goto drop; 870 if (so == NULL) { 871 /* 872 * Entry added to syncache, mbuf used to 873 * send SYN,ACK packet. 874 */ 875 KASSERT(headlocked, ("headlocked")); 876 INP_UNLOCK(inp); 877 INP_INFO_WUNLOCK(&tcbinfo); 878 return; 879 } 880 /* 881 * Segment passed TAO tests. 882 */ 883 INP_UNLOCK(inp); 884 inp = sotoinpcb(so); 885 INP_LOCK(inp); 886 tp = intotcpcb(inp); 887 tp->snd_wnd = tiwin; 888 tp->t_starttime = ticks; 889 tp->t_state = TCPS_ESTABLISHED; 890 891 /* 892 * T/TCP logic: 893 * If there is a FIN or if there is data, then 894 * delay SYN,ACK(SYN) in the hope of piggy-backing 895 * it on a response segment. Otherwise must send 896 * ACK now in case the other side is slow starting. 897 */ 898 if (thflags & TH_FIN || tlen != 0) 899 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 900 else 901 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 902 tcpstat.tcps_connects++; 903 soisconnected(so); 904 goto trimthenstep6; 905 } 906 goto drop; 907 } 908 after_listen: 909 910 /* XXX temp debugging */ 911 /* should not happen - syncache should pick up these connections */ 912 if (tp->t_state == TCPS_LISTEN) 913 panic("tcp_input: TCPS_LISTEN"); 914 915 /* 916 * Segment received on connection. 917 * Reset idle time and keep-alive timer. 918 */ 919 tp->t_rcvtime = ticks; 920 if (TCPS_HAVEESTABLISHED(tp->t_state)) 921 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 922 923 /* 924 * Process options. 925 * XXX this is tradtitional behavior, may need to be cleaned up. 926 */ 927 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 928 if (thflags & TH_SYN) { 929 if (to.to_flags & TOF_SCALE) { 930 tp->t_flags |= TF_RCVD_SCALE; 931 tp->requested_s_scale = to.to_requested_s_scale; 932 } 933 if (to.to_flags & TOF_TS) { 934 tp->t_flags |= TF_RCVD_TSTMP; 935 tp->ts_recent = to.to_tsval; 936 tp->ts_recent_age = ticks; 937 } 938 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 939 tp->t_flags |= TF_RCVD_CC; 940 if (to.to_flags & TOF_MSS) 941 tcp_mss(tp, to.to_mss); 942 } 943 944 /* 945 * Header prediction: check for the two common cases 946 * of a uni-directional data xfer. If the packet has 947 * no control flags, is in-sequence, the window didn't 948 * change and we're not retransmitting, it's a 949 * candidate. If the length is zero and the ack moved 950 * forward, we're the sender side of the xfer. Just 951 * free the data acked & wake any higher level process 952 * that was blocked waiting for space. If the length 953 * is non-zero and the ack didn't move, we're the 954 * receiver side. If we're getting packets in-order 955 * (the reassembly queue is empty), add the data to 956 * the socket buffer and note that we need a delayed ack. 957 * Make sure that the hidden state-flags are also off. 958 * Since we check for TCPS_ESTABLISHED above, it can only 959 * be TH_NEEDSYN. 960 */ 961 if (tp->t_state == TCPS_ESTABLISHED && 962 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 963 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 964 ((to.to_flags & TOF_TS) == 0 || 965 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 966 /* 967 * Using the CC option is compulsory if once started: 968 * the segment is OK if no T/TCP was negotiated or 969 * if the segment has a CC option equal to CCrecv 970 */ 971 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 972 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 973 th->th_seq == tp->rcv_nxt && 974 tiwin && tiwin == tp->snd_wnd && 975 tp->snd_nxt == tp->snd_max) { 976 977 /* 978 * If last ACK falls within this segment's sequence numbers, 979 * record the timestamp. 980 * NOTE that the test is modified according to the latest 981 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 982 */ 983 if ((to.to_flags & TOF_TS) != 0 && 984 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 985 tp->ts_recent_age = ticks; 986 tp->ts_recent = to.to_tsval; 987 } 988 989 if (tlen == 0) { 990 if (SEQ_GT(th->th_ack, tp->snd_una) && 991 SEQ_LEQ(th->th_ack, tp->snd_max) && 992 tp->snd_cwnd >= tp->snd_wnd && 993 ((!tcp_do_newreno && 994 tp->t_dupacks < tcprexmtthresh) || 995 (tcp_do_newreno && 996 !SEQ_LT(tp->snd_una, tp->snd_recover)))) { 997 KASSERT(headlocked, ("headlocked")); 998 INP_INFO_WUNLOCK(&tcbinfo); 999 /* 1000 * this is a pure ack for outstanding data. 1001 */ 1002 ++tcpstat.tcps_predack; 1003 /* 1004 * "bad retransmit" recovery 1005 */ 1006 if (tp->t_rxtshift == 1 && 1007 ticks < tp->t_badrxtwin) { 1008 ++tcpstat.tcps_sndrexmitbad; 1009 tp->snd_cwnd = tp->snd_cwnd_prev; 1010 tp->snd_ssthresh = 1011 tp->snd_ssthresh_prev; 1012 tp->snd_high = tp->snd_high_prev; 1013 tp->snd_nxt = tp->snd_max; 1014 tp->t_badrxtwin = 0; 1015 } 1016 1017 /* 1018 * Recalculate the transmit timer / rtt. 1019 * 1020 * Some boxes send broken timestamp replies 1021 * during the SYN+ACK phase, ignore 1022 * timestamps of 0 or we could calculate a 1023 * huge RTT and blow up the retransmit timer. 1024 */ 1025 if ((to.to_flags & TOF_TS) != 0 && 1026 to.to_tsecr) { 1027 tcp_xmit_timer(tp, 1028 ticks - to.to_tsecr + 1); 1029 } else if (tp->t_rtttime && 1030 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1031 tcp_xmit_timer(tp, 1032 ticks - tp->t_rtttime); 1033 } 1034 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1035 acked = th->th_ack - tp->snd_una; 1036 tcpstat.tcps_rcvackpack++; 1037 tcpstat.tcps_rcvackbyte += acked; 1038 sbdrop(&so->so_snd, acked); 1039 if (SEQ_GT(tp->snd_una, tp->snd_high) && 1040 SEQ_LEQ(th->th_ack, tp->snd_high)) 1041 tp->snd_high = th->th_ack - 1; 1042 tp->snd_una = tp->snd_recover = th->th_ack; 1043 /* 1044 * pull snd_wl2 up to prevent seq wrap relative 1045 * to th_ack. 1046 */ 1047 tp->snd_wl2 = th->th_ack; 1048 tp->t_dupacks = 0; 1049 m_freem(m); 1050 ND6_HINT(tp); /* some progress has been done */ 1051 1052 /* 1053 * If all outstanding data are acked, stop 1054 * retransmit timer, otherwise restart timer 1055 * using current (possibly backed-off) value. 1056 * If process is waiting for space, 1057 * wakeup/selwakeup/signal. If data 1058 * are ready to send, let tcp_output 1059 * decide between more output or persist. 1060 */ 1061 if (tp->snd_una == tp->snd_max) 1062 callout_stop(tp->tt_rexmt); 1063 else if (!callout_active(tp->tt_persist)) 1064 callout_reset(tp->tt_rexmt, 1065 tp->t_rxtcur, 1066 tcp_timer_rexmt, tp); 1067 1068 sowwakeup(so); 1069 if (so->so_snd.sb_cc) 1070 (void) tcp_output(tp); 1071 goto check_delack; 1072 } 1073 } else if (th->th_ack == tp->snd_una && 1074 LIST_EMPTY(&tp->t_segq) && 1075 tlen <= sbspace(&so->so_rcv)) { 1076 KASSERT(headlocked, ("headlocked")); 1077 INP_INFO_WUNLOCK(&tcbinfo); 1078 /* 1079 * this is a pure, in-sequence data packet 1080 * with nothing on the reassembly queue and 1081 * we have enough buffer space to take it. 1082 */ 1083 ++tcpstat.tcps_preddat; 1084 tp->rcv_nxt += tlen; 1085 /* 1086 * Pull snd_wl1 up to prevent seq wrap relative to 1087 * th_seq. 1088 */ 1089 tp->snd_wl1 = th->th_seq; 1090 /* 1091 * Pull rcv_up up to prevent seq wrap relative to 1092 * rcv_nxt. 1093 */ 1094 tp->rcv_up = tp->rcv_nxt; 1095 tcpstat.tcps_rcvpack++; 1096 tcpstat.tcps_rcvbyte += tlen; 1097 ND6_HINT(tp); /* some progress has been done */ 1098 /* 1099 * Add data to socket buffer. 1100 */ 1101 if (so->so_state & SS_CANTRCVMORE) { 1102 m_freem(m); 1103 } else { 1104 m_adj(m, drop_hdrlen); /* delayed header drop */ 1105 sbappend(&so->so_rcv, m); 1106 } 1107 sorwakeup(so); 1108 if (DELAY_ACK(tp)) { 1109 tp->t_flags |= TF_DELACK; 1110 } else { 1111 tp->t_flags |= TF_ACKNOW; 1112 tcp_output(tp); 1113 } 1114 goto check_delack; 1115 } 1116 } 1117 1118 /* 1119 * Calculate amount of space in receive window, 1120 * and then do TCP input processing. 1121 * Receive window is amount of space in rcv queue, 1122 * but not less than advertised window. 1123 */ 1124 { int win; 1125 1126 win = sbspace(&so->so_rcv); 1127 if (win < 0) 1128 win = 0; 1129 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1130 } 1131 1132 switch (tp->t_state) { 1133 1134 /* 1135 * If the state is SYN_RECEIVED: 1136 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1137 */ 1138 case TCPS_SYN_RECEIVED: 1139 if ((thflags & TH_ACK) && 1140 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1141 SEQ_GT(th->th_ack, tp->snd_max))) { 1142 rstreason = BANDLIM_RST_OPENPORT; 1143 goto dropwithreset; 1144 } 1145 break; 1146 1147 /* 1148 * If the state is SYN_SENT: 1149 * if seg contains an ACK, but not for our SYN, drop the input. 1150 * if seg contains a RST, then drop the connection. 1151 * if seg does not contain SYN, then drop it. 1152 * Otherwise this is an acceptable SYN segment 1153 * initialize tp->rcv_nxt and tp->irs 1154 * if seg contains ack then advance tp->snd_una 1155 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1156 * arrange for segment to be acked (eventually) 1157 * continue processing rest of data/controls, beginning with URG 1158 */ 1159 case TCPS_SYN_SENT: 1160 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1161 taop = &tao_noncached; 1162 bzero(taop, sizeof(*taop)); 1163 } 1164 1165 if ((thflags & TH_ACK) && 1166 (SEQ_LEQ(th->th_ack, tp->iss) || 1167 SEQ_GT(th->th_ack, tp->snd_max))) { 1168 /* 1169 * If we have a cached CCsent for the remote host, 1170 * hence we haven't just crashed and restarted, 1171 * do not send a RST. This may be a retransmission 1172 * from the other side after our earlier ACK was lost. 1173 * Our new SYN, when it arrives, will serve as the 1174 * needed ACK. 1175 */ 1176 if (taop->tao_ccsent != 0) 1177 goto drop; 1178 else { 1179 rstreason = BANDLIM_UNLIMITED; 1180 goto dropwithreset; 1181 } 1182 } 1183 if (thflags & TH_RST) { 1184 if (thflags & TH_ACK) 1185 tp = tcp_drop(tp, ECONNREFUSED); 1186 goto drop; 1187 } 1188 if ((thflags & TH_SYN) == 0) 1189 goto drop; 1190 tp->snd_wnd = th->th_win; /* initial send window */ 1191 tp->cc_recv = to.to_cc; /* foreign CC */ 1192 1193 tp->irs = th->th_seq; 1194 tcp_rcvseqinit(tp); 1195 if (thflags & TH_ACK) { 1196 /* 1197 * Our SYN was acked. If segment contains CC.ECHO 1198 * option, check it to make sure this segment really 1199 * matches our SYN. If not, just drop it as old 1200 * duplicate, but send an RST if we're still playing 1201 * by the old rules. If no CC.ECHO option, make sure 1202 * we don't get fooled into using T/TCP. 1203 */ 1204 if (to.to_flags & TOF_CCECHO) { 1205 if (tp->cc_send != to.to_ccecho) { 1206 if (taop->tao_ccsent != 0) 1207 goto drop; 1208 else { 1209 rstreason = BANDLIM_UNLIMITED; 1210 goto dropwithreset; 1211 } 1212 } 1213 } else 1214 tp->t_flags &= ~TF_RCVD_CC; 1215 tcpstat.tcps_connects++; 1216 soisconnected(so); 1217 #ifdef MAC 1218 mac_set_socket_peer_from_mbuf(m, so); 1219 #endif 1220 /* Do window scaling on this connection? */ 1221 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1222 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1223 tp->snd_scale = tp->requested_s_scale; 1224 tp->rcv_scale = tp->request_r_scale; 1225 } 1226 /* Segment is acceptable, update cache if undefined. */ 1227 if (taop->tao_ccsent == 0) 1228 taop->tao_ccsent = to.to_ccecho; 1229 1230 tp->rcv_adv += tp->rcv_wnd; 1231 tp->snd_una++; /* SYN is acked */ 1232 /* 1233 * If there's data, delay ACK; if there's also a FIN 1234 * ACKNOW will be turned on later. 1235 */ 1236 if (DELAY_ACK(tp) && tlen != 0) 1237 callout_reset(tp->tt_delack, tcp_delacktime, 1238 tcp_timer_delack, tp); 1239 else 1240 tp->t_flags |= TF_ACKNOW; 1241 /* 1242 * Received <SYN,ACK> in SYN_SENT[*] state. 1243 * Transitions: 1244 * SYN_SENT --> ESTABLISHED 1245 * SYN_SENT* --> FIN_WAIT_1 1246 */ 1247 tp->t_starttime = ticks; 1248 if (tp->t_flags & TF_NEEDFIN) { 1249 tp->t_state = TCPS_FIN_WAIT_1; 1250 tp->t_flags &= ~TF_NEEDFIN; 1251 thflags &= ~TH_SYN; 1252 } else { 1253 tp->t_state = TCPS_ESTABLISHED; 1254 callout_reset(tp->tt_keep, tcp_keepidle, 1255 tcp_timer_keep, tp); 1256 } 1257 } else { 1258 /* 1259 * Received initial SYN in SYN-SENT[*] state => 1260 * simultaneous open. If segment contains CC option 1261 * and there is a cached CC, apply TAO test. 1262 * If it succeeds, connection is * half-synchronized. 1263 * Otherwise, do 3-way handshake: 1264 * SYN-SENT -> SYN-RECEIVED 1265 * SYN-SENT* -> SYN-RECEIVED* 1266 * If there was no CC option, clear cached CC value. 1267 */ 1268 tp->t_flags |= TF_ACKNOW; 1269 callout_stop(tp->tt_rexmt); 1270 if (to.to_flags & TOF_CC) { 1271 if (taop->tao_cc != 0 && 1272 CC_GT(to.to_cc, taop->tao_cc)) { 1273 /* 1274 * update cache and make transition: 1275 * SYN-SENT -> ESTABLISHED* 1276 * SYN-SENT* -> FIN-WAIT-1* 1277 */ 1278 taop->tao_cc = to.to_cc; 1279 tp->t_starttime = ticks; 1280 if (tp->t_flags & TF_NEEDFIN) { 1281 tp->t_state = TCPS_FIN_WAIT_1; 1282 tp->t_flags &= ~TF_NEEDFIN; 1283 } else { 1284 tp->t_state = TCPS_ESTABLISHED; 1285 callout_reset(tp->tt_keep, 1286 tcp_keepidle, 1287 tcp_timer_keep, 1288 tp); 1289 } 1290 tp->t_flags |= TF_NEEDSYN; 1291 } else 1292 tp->t_state = TCPS_SYN_RECEIVED; 1293 } else { 1294 /* CC.NEW or no option => invalidate cache */ 1295 taop->tao_cc = 0; 1296 tp->t_state = TCPS_SYN_RECEIVED; 1297 } 1298 } 1299 1300 trimthenstep6: 1301 /* 1302 * Advance th->th_seq to correspond to first data byte. 1303 * If data, trim to stay within window, 1304 * dropping FIN if necessary. 1305 */ 1306 th->th_seq++; 1307 if (tlen > tp->rcv_wnd) { 1308 todrop = tlen - tp->rcv_wnd; 1309 m_adj(m, -todrop); 1310 tlen = tp->rcv_wnd; 1311 thflags &= ~TH_FIN; 1312 tcpstat.tcps_rcvpackafterwin++; 1313 tcpstat.tcps_rcvbyteafterwin += todrop; 1314 } 1315 tp->snd_wl1 = th->th_seq - 1; 1316 tp->rcv_up = th->th_seq; 1317 /* 1318 * Client side of transaction: already sent SYN and data. 1319 * If the remote host used T/TCP to validate the SYN, 1320 * our data will be ACK'd; if so, enter normal data segment 1321 * processing in the middle of step 5, ack processing. 1322 * Otherwise, goto step 6. 1323 */ 1324 if (thflags & TH_ACK) 1325 goto process_ACK; 1326 1327 goto step6; 1328 1329 /* 1330 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1331 * if segment contains a SYN and CC [not CC.NEW] option: 1332 * if state == TIME_WAIT and connection duration > MSL, 1333 * drop packet and send RST; 1334 * 1335 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1336 * ack the FIN (and data) in retransmission queue. 1337 * Complete close and delete TCPCB. Then reprocess 1338 * segment, hoping to find new TCPCB in LISTEN state; 1339 * 1340 * else must be old SYN; drop it. 1341 * else do normal processing. 1342 */ 1343 case TCPS_LAST_ACK: 1344 case TCPS_CLOSING: 1345 case TCPS_TIME_WAIT: 1346 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1347 if ((thflags & TH_SYN) && 1348 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1349 if (tp->t_state == TCPS_TIME_WAIT && 1350 (ticks - tp->t_starttime) > tcp_msl) { 1351 rstreason = BANDLIM_UNLIMITED; 1352 goto dropwithreset; 1353 } 1354 if (CC_GT(to.to_cc, tp->cc_recv)) { 1355 tp = tcp_close(tp); 1356 goto findpcb; 1357 } 1358 else 1359 goto drop; 1360 } 1361 break; /* continue normal processing */ 1362 } 1363 1364 /* 1365 * States other than LISTEN or SYN_SENT. 1366 * First check the RST flag and sequence number since reset segments 1367 * are exempt from the timestamp and connection count tests. This 1368 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1369 * below which allowed reset segments in half the sequence space 1370 * to fall though and be processed (which gives forged reset 1371 * segments with a random sequence number a 50 percent chance of 1372 * killing a connection). 1373 * Then check timestamp, if present. 1374 * Then check the connection count, if present. 1375 * Then check that at least some bytes of segment are within 1376 * receive window. If segment begins before rcv_nxt, 1377 * drop leading data (and SYN); if nothing left, just ack. 1378 * 1379 * 1380 * If the RST bit is set, check the sequence number to see 1381 * if this is a valid reset segment. 1382 * RFC 793 page 37: 1383 * In all states except SYN-SENT, all reset (RST) segments 1384 * are validated by checking their SEQ-fields. A reset is 1385 * valid if its sequence number is in the window. 1386 * Note: this does not take into account delayed ACKs, so 1387 * we should test against last_ack_sent instead of rcv_nxt. 1388 * The sequence number in the reset segment is normally an 1389 * echo of our outgoing acknowlegement numbers, but some hosts 1390 * send a reset with the sequence number at the rightmost edge 1391 * of our receive window, and we have to handle this case. 1392 * If we have multiple segments in flight, the intial reset 1393 * segment sequence numbers will be to the left of last_ack_sent, 1394 * but they will eventually catch up. 1395 * In any case, it never made sense to trim reset segments to 1396 * fit the receive window since RFC 1122 says: 1397 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1398 * 1399 * A TCP SHOULD allow a received RST segment to include data. 1400 * 1401 * DISCUSSION 1402 * It has been suggested that a RST segment could contain 1403 * ASCII text that encoded and explained the cause of the 1404 * RST. No standard has yet been established for such 1405 * data. 1406 * 1407 * If the reset segment passes the sequence number test examine 1408 * the state: 1409 * SYN_RECEIVED STATE: 1410 * If passive open, return to LISTEN state. 1411 * If active open, inform user that connection was refused. 1412 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1413 * Inform user that connection was reset, and close tcb. 1414 * CLOSING, LAST_ACK STATES: 1415 * Close the tcb. 1416 * TIME_WAIT STATE: 1417 * Drop the segment - see Stevens, vol. 2, p. 964 and 1418 * RFC 1337. 1419 */ 1420 if (thflags & TH_RST) { 1421 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1422 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1423 switch (tp->t_state) { 1424 1425 case TCPS_SYN_RECEIVED: 1426 so->so_error = ECONNREFUSED; 1427 goto close; 1428 1429 case TCPS_ESTABLISHED: 1430 case TCPS_FIN_WAIT_1: 1431 case TCPS_FIN_WAIT_2: 1432 case TCPS_CLOSE_WAIT: 1433 so->so_error = ECONNRESET; 1434 close: 1435 tp->t_state = TCPS_CLOSED; 1436 tcpstat.tcps_drops++; 1437 tp = tcp_close(tp); 1438 break; 1439 1440 case TCPS_CLOSING: 1441 case TCPS_LAST_ACK: 1442 tp = tcp_close(tp); 1443 break; 1444 1445 case TCPS_TIME_WAIT: 1446 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1447 ("timewait")); 1448 break; 1449 } 1450 } 1451 goto drop; 1452 } 1453 1454 /* 1455 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1456 * and it's less than ts_recent, drop it. 1457 */ 1458 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1459 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1460 1461 /* Check to see if ts_recent is over 24 days old. */ 1462 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1463 /* 1464 * Invalidate ts_recent. If this segment updates 1465 * ts_recent, the age will be reset later and ts_recent 1466 * will get a valid value. If it does not, setting 1467 * ts_recent to zero will at least satisfy the 1468 * requirement that zero be placed in the timestamp 1469 * echo reply when ts_recent isn't valid. The 1470 * age isn't reset until we get a valid ts_recent 1471 * because we don't want out-of-order segments to be 1472 * dropped when ts_recent is old. 1473 */ 1474 tp->ts_recent = 0; 1475 } else { 1476 tcpstat.tcps_rcvduppack++; 1477 tcpstat.tcps_rcvdupbyte += tlen; 1478 tcpstat.tcps_pawsdrop++; 1479 if (tlen) 1480 goto dropafterack; 1481 goto drop; 1482 } 1483 } 1484 1485 /* 1486 * T/TCP mechanism 1487 * If T/TCP was negotiated and the segment doesn't have CC, 1488 * or if its CC is wrong then drop the segment. 1489 * RST segments do not have to comply with this. 1490 */ 1491 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1492 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1493 goto dropafterack; 1494 1495 /* 1496 * In the SYN-RECEIVED state, validate that the packet belongs to 1497 * this connection before trimming the data to fit the receive 1498 * window. Check the sequence number versus IRS since we know 1499 * the sequence numbers haven't wrapped. This is a partial fix 1500 * for the "LAND" DoS attack. 1501 */ 1502 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1503 rstreason = BANDLIM_RST_OPENPORT; 1504 goto dropwithreset; 1505 } 1506 1507 todrop = tp->rcv_nxt - th->th_seq; 1508 if (todrop > 0) { 1509 if (thflags & TH_SYN) { 1510 thflags &= ~TH_SYN; 1511 th->th_seq++; 1512 if (th->th_urp > 1) 1513 th->th_urp--; 1514 else 1515 thflags &= ~TH_URG; 1516 todrop--; 1517 } 1518 /* 1519 * Following if statement from Stevens, vol. 2, p. 960. 1520 */ 1521 if (todrop > tlen 1522 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1523 /* 1524 * Any valid FIN must be to the left of the window. 1525 * At this point the FIN must be a duplicate or out 1526 * of sequence; drop it. 1527 */ 1528 thflags &= ~TH_FIN; 1529 1530 /* 1531 * Send an ACK to resynchronize and drop any data. 1532 * But keep on processing for RST or ACK. 1533 */ 1534 tp->t_flags |= TF_ACKNOW; 1535 todrop = tlen; 1536 tcpstat.tcps_rcvduppack++; 1537 tcpstat.tcps_rcvdupbyte += todrop; 1538 } else { 1539 tcpstat.tcps_rcvpartduppack++; 1540 tcpstat.tcps_rcvpartdupbyte += todrop; 1541 } 1542 drop_hdrlen += todrop; /* drop from the top afterwards */ 1543 th->th_seq += todrop; 1544 tlen -= todrop; 1545 if (th->th_urp > todrop) 1546 th->th_urp -= todrop; 1547 else { 1548 thflags &= ~TH_URG; 1549 th->th_urp = 0; 1550 } 1551 } 1552 1553 /* 1554 * If new data are received on a connection after the 1555 * user processes are gone, then RST the other end. 1556 */ 1557 if ((so->so_state & SS_NOFDREF) && 1558 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1559 tp = tcp_close(tp); 1560 tcpstat.tcps_rcvafterclose++; 1561 rstreason = BANDLIM_UNLIMITED; 1562 goto dropwithreset; 1563 } 1564 1565 /* 1566 * If segment ends after window, drop trailing data 1567 * (and PUSH and FIN); if nothing left, just ACK. 1568 */ 1569 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1570 if (todrop > 0) { 1571 tcpstat.tcps_rcvpackafterwin++; 1572 if (todrop >= tlen) { 1573 tcpstat.tcps_rcvbyteafterwin += tlen; 1574 /* 1575 * If a new connection request is received 1576 * while in TIME_WAIT, drop the old connection 1577 * and start over if the sequence numbers 1578 * are above the previous ones. 1579 */ 1580 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1581 if (thflags & TH_SYN && 1582 tp->t_state == TCPS_TIME_WAIT && 1583 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1584 tp = tcp_close(tp); 1585 goto findpcb; 1586 } 1587 /* 1588 * If window is closed can only take segments at 1589 * window edge, and have to drop data and PUSH from 1590 * incoming segments. Continue processing, but 1591 * remember to ack. Otherwise, drop segment 1592 * and ack. 1593 */ 1594 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1595 tp->t_flags |= TF_ACKNOW; 1596 tcpstat.tcps_rcvwinprobe++; 1597 } else 1598 goto dropafterack; 1599 } else 1600 tcpstat.tcps_rcvbyteafterwin += todrop; 1601 m_adj(m, -todrop); 1602 tlen -= todrop; 1603 thflags &= ~(TH_PUSH|TH_FIN); 1604 } 1605 1606 /* 1607 * If last ACK falls within this segment's sequence numbers, 1608 * record its timestamp. 1609 * NOTE that the test is modified according to the latest 1610 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1611 */ 1612 if ((to.to_flags & TOF_TS) != 0 && 1613 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1614 tp->ts_recent_age = ticks; 1615 tp->ts_recent = to.to_tsval; 1616 } 1617 1618 /* 1619 * If a SYN is in the window, then this is an 1620 * error and we send an RST and drop the connection. 1621 */ 1622 if (thflags & TH_SYN) { 1623 tp = tcp_drop(tp, ECONNRESET); 1624 rstreason = BANDLIM_UNLIMITED; 1625 goto dropwithreset; 1626 } 1627 1628 /* 1629 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1630 * flag is on (half-synchronized state), then queue data for 1631 * later processing; else drop segment and return. 1632 */ 1633 if ((thflags & TH_ACK) == 0) { 1634 if (tp->t_state == TCPS_SYN_RECEIVED || 1635 (tp->t_flags & TF_NEEDSYN)) 1636 goto step6; 1637 else 1638 goto drop; 1639 } 1640 1641 /* 1642 * Ack processing. 1643 */ 1644 switch (tp->t_state) { 1645 1646 /* 1647 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1648 * ESTABLISHED state and continue processing. 1649 * The ACK was checked above. 1650 */ 1651 case TCPS_SYN_RECEIVED: 1652 1653 tcpstat.tcps_connects++; 1654 soisconnected(so); 1655 /* Do window scaling? */ 1656 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1657 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1658 tp->snd_scale = tp->requested_s_scale; 1659 tp->rcv_scale = tp->request_r_scale; 1660 } 1661 /* 1662 * Upon successful completion of 3-way handshake, 1663 * update cache.CC if it was undefined, pass any queued 1664 * data to the user, and advance state appropriately. 1665 */ 1666 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1667 taop->tao_cc == 0) 1668 taop->tao_cc = tp->cc_recv; 1669 1670 /* 1671 * Make transitions: 1672 * SYN-RECEIVED -> ESTABLISHED 1673 * SYN-RECEIVED* -> FIN-WAIT-1 1674 */ 1675 tp->t_starttime = ticks; 1676 if (tp->t_flags & TF_NEEDFIN) { 1677 tp->t_state = TCPS_FIN_WAIT_1; 1678 tp->t_flags &= ~TF_NEEDFIN; 1679 } else { 1680 tp->t_state = TCPS_ESTABLISHED; 1681 callout_reset(tp->tt_keep, tcp_keepidle, 1682 tcp_timer_keep, tp); 1683 } 1684 /* 1685 * If segment contains data or ACK, will call tcp_reass() 1686 * later; if not, do so now to pass queued data to user. 1687 */ 1688 if (tlen == 0 && (thflags & TH_FIN) == 0) 1689 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1690 (struct mbuf *)0); 1691 tp->snd_wl1 = th->th_seq - 1; 1692 /* FALLTHROUGH */ 1693 1694 /* 1695 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1696 * ACKs. If the ack is in the range 1697 * tp->snd_una < th->th_ack <= tp->snd_max 1698 * then advance tp->snd_una to th->th_ack and drop 1699 * data from the retransmission queue. If this ACK reflects 1700 * more up to date window information we update our window information. 1701 */ 1702 case TCPS_ESTABLISHED: 1703 case TCPS_FIN_WAIT_1: 1704 case TCPS_FIN_WAIT_2: 1705 case TCPS_CLOSE_WAIT: 1706 case TCPS_CLOSING: 1707 case TCPS_LAST_ACK: 1708 case TCPS_TIME_WAIT: 1709 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1710 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1711 if (tlen == 0 && tiwin == tp->snd_wnd) { 1712 tcpstat.tcps_rcvdupack++; 1713 /* 1714 * If we have outstanding data (other than 1715 * a window probe), this is a completely 1716 * duplicate ack (ie, window info didn't 1717 * change), the ack is the biggest we've 1718 * seen and we've seen exactly our rexmt 1719 * threshhold of them, assume a packet 1720 * has been dropped and retransmit it. 1721 * Kludge snd_nxt & the congestion 1722 * window so we send only this one 1723 * packet. 1724 * 1725 * We know we're losing at the current 1726 * window size so do congestion avoidance 1727 * (set ssthresh to half the current window 1728 * and pull our congestion window back to 1729 * the new ssthresh). 1730 * 1731 * Dup acks mean that packets have left the 1732 * network (they're now cached at the receiver) 1733 * so bump cwnd by the amount in the receiver 1734 * to keep a constant cwnd packets in the 1735 * network. 1736 */ 1737 if (!callout_active(tp->tt_rexmt) || 1738 th->th_ack != tp->snd_una) 1739 tp->t_dupacks = 0; 1740 else if (++tp->t_dupacks > tcprexmtthresh || 1741 (tcp_do_newreno && 1742 SEQ_LT(tp->snd_una, 1743 tp->snd_recover))) { 1744 tp->snd_cwnd += tp->t_maxseg; 1745 (void) tcp_output(tp); 1746 goto drop; 1747 } else if (tp->t_dupacks == tcprexmtthresh) { 1748 tcp_seq onxt = tp->snd_nxt; 1749 u_int win; 1750 if (tcp_do_newreno && 1751 SEQ_LEQ(th->th_ack, tp->snd_high)) { 1752 tp->t_dupacks = 0; 1753 break; 1754 } 1755 win = min(tp->snd_wnd, tp->snd_cwnd) / 1756 2 / tp->t_maxseg; 1757 if (win < 2) 1758 win = 2; 1759 tp->snd_ssthresh = win * tp->t_maxseg; 1760 tp->snd_recover = tp->snd_max; 1761 callout_stop(tp->tt_rexmt); 1762 tp->t_rtttime = 0; 1763 tp->snd_nxt = th->th_ack; 1764 tp->snd_cwnd = tp->t_maxseg; 1765 (void) tcp_output(tp); 1766 KASSERT(tp->snd_limited <= 2, 1767 ("tp->snd_limited too big")); 1768 tp->snd_cwnd = tp->snd_ssthresh + 1769 tp->t_maxseg * 1770 (tp->t_dupacks - tp->snd_limited); 1771 if (SEQ_GT(onxt, tp->snd_nxt)) 1772 tp->snd_nxt = onxt; 1773 goto drop; 1774 } else if (tcp_do_rfc3042) { 1775 u_long oldcwnd = tp->snd_cwnd; 1776 tcp_seq oldsndmax = tp->snd_max; 1777 u_int sent; 1778 KASSERT(tp->t_dupacks == 1 || 1779 tp->t_dupacks == 2, 1780 ("dupacks not 1 or 2")); 1781 if (tp->t_dupacks == 1) { 1782 tp->snd_limited = 0; 1783 tp->snd_cwnd += tp->t_maxseg; 1784 } else { 1785 tp->snd_cwnd += 1786 tp->t_maxseg * 2; 1787 } 1788 (void) tcp_output(tp); 1789 sent = tp->snd_max - oldsndmax; 1790 if (sent > tp->t_maxseg) { 1791 KASSERT(tp->snd_limited == 0 && 1792 tp->t_dupacks == 2, 1793 ("sent too much")); 1794 tp->snd_limited = 2; 1795 } else if (sent > 0) 1796 ++tp->snd_limited; 1797 tp->snd_cwnd = oldcwnd; 1798 goto drop; 1799 } 1800 } else 1801 tp->t_dupacks = 0; 1802 break; 1803 } 1804 1805 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1806 1807 /* 1808 * If the congestion window was inflated to account 1809 * for the other side's cached packets, retract it. 1810 */ 1811 if (tcp_do_newreno) { 1812 if (SEQ_LT(tp->snd_una, tp->snd_recover)) { 1813 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1814 tcp_newreno_partial_ack(tp, th); 1815 } else { 1816 /* 1817 * Window inflation should have left us 1818 * with approximately snd_ssthresh 1819 * outstanding data. 1820 * But in case we would be inclined to 1821 * send a burst, better to do it via 1822 * the slow start mechanism. 1823 */ 1824 if (SEQ_GT(th->th_ack + 1825 tp->snd_ssthresh, 1826 tp->snd_max)) 1827 tp->snd_cwnd = tp->snd_max - 1828 th->th_ack + 1829 tp->t_maxseg; 1830 else 1831 tp->snd_cwnd = tp->snd_ssthresh; 1832 } 1833 } 1834 } else { 1835 if (tp->t_dupacks >= tcprexmtthresh && 1836 tp->snd_cwnd > tp->snd_ssthresh) 1837 tp->snd_cwnd = tp->snd_ssthresh; 1838 } 1839 tp->t_dupacks = 0; 1840 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1841 tcpstat.tcps_rcvacktoomuch++; 1842 goto dropafterack; 1843 } 1844 /* 1845 * If we reach this point, ACK is not a duplicate, 1846 * i.e., it ACKs something we sent. 1847 */ 1848 if (tp->t_flags & TF_NEEDSYN) { 1849 /* 1850 * T/TCP: Connection was half-synchronized, and our 1851 * SYN has been ACK'd (so connection is now fully 1852 * synchronized). Go to non-starred state, 1853 * increment snd_una for ACK of SYN, and check if 1854 * we can do window scaling. 1855 */ 1856 tp->t_flags &= ~TF_NEEDSYN; 1857 tp->snd_una++; 1858 /* Do window scaling? */ 1859 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1860 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1861 tp->snd_scale = tp->requested_s_scale; 1862 tp->rcv_scale = tp->request_r_scale; 1863 } 1864 } 1865 1866 process_ACK: 1867 acked = th->th_ack - tp->snd_una; 1868 tcpstat.tcps_rcvackpack++; 1869 tcpstat.tcps_rcvackbyte += acked; 1870 1871 /* 1872 * If we just performed our first retransmit, and the ACK 1873 * arrives within our recovery window, then it was a mistake 1874 * to do the retransmit in the first place. Recover our 1875 * original cwnd and ssthresh, and proceed to transmit where 1876 * we left off. 1877 */ 1878 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1879 ++tcpstat.tcps_sndrexmitbad; 1880 tp->snd_cwnd = tp->snd_cwnd_prev; 1881 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1882 tp->snd_high = tp->snd_high_prev; 1883 tp->snd_nxt = tp->snd_max; 1884 tp->t_badrxtwin = 0; /* XXX probably not required */ 1885 } 1886 1887 /* 1888 * If we have a timestamp reply, update smoothed 1889 * round trip time. If no timestamp is present but 1890 * transmit timer is running and timed sequence 1891 * number was acked, update smoothed round trip time. 1892 * Since we now have an rtt measurement, cancel the 1893 * timer backoff (cf., Phil Karn's retransmit alg.). 1894 * Recompute the initial retransmit timer. 1895 * 1896 * Some boxes send broken timestamp replies 1897 * during the SYN+ACK phase, ignore 1898 * timestamps of 0 or we could calculate a 1899 * huge RTT and blow up the retransmit timer. 1900 */ 1901 if ((to.to_flags & TOF_TS) != 0 && 1902 to.to_tsecr) { 1903 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1904 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1905 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1906 } 1907 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1908 1909 /* 1910 * If all outstanding data is acked, stop retransmit 1911 * timer and remember to restart (more output or persist). 1912 * If there is more data to be acked, restart retransmit 1913 * timer, using current (possibly backed-off) value. 1914 */ 1915 if (th->th_ack == tp->snd_max) { 1916 callout_stop(tp->tt_rexmt); 1917 needoutput = 1; 1918 } else if (!callout_active(tp->tt_persist)) 1919 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1920 tcp_timer_rexmt, tp); 1921 1922 /* 1923 * If no data (only SYN) was ACK'd, 1924 * skip rest of ACK processing. 1925 */ 1926 if (acked == 0) 1927 goto step6; 1928 1929 /* 1930 * When new data is acked, open the congestion window. 1931 * If the window gives us less than ssthresh packets 1932 * in flight, open exponentially (maxseg per packet). 1933 * Otherwise open linearly: maxseg per window 1934 * (maxseg^2 / cwnd per packet). 1935 */ 1936 if (!tcp_do_newreno || SEQ_GEQ(tp->snd_una, tp->snd_recover)) { 1937 register u_int cw = tp->snd_cwnd; 1938 register u_int incr = tp->t_maxseg; 1939 if (cw > tp->snd_ssthresh) 1940 incr = incr * incr / cw; 1941 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 1942 } 1943 if (acked > so->so_snd.sb_cc) { 1944 tp->snd_wnd -= so->so_snd.sb_cc; 1945 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1946 ourfinisacked = 1; 1947 } else { 1948 sbdrop(&so->so_snd, acked); 1949 tp->snd_wnd -= acked; 1950 ourfinisacked = 0; 1951 } 1952 sowwakeup(so); 1953 /* detect una wraparound */ 1954 if (SEQ_GEQ(tp->snd_una, tp->snd_recover) && 1955 SEQ_LT(th->th_ack, tp->snd_recover)) 1956 tp->snd_recover = th->th_ack; 1957 if (SEQ_GT(tp->snd_una, tp->snd_high) && 1958 SEQ_LEQ(th->th_ack, tp->snd_high)) 1959 tp->snd_high = th->th_ack - 1; 1960 tp->snd_una = th->th_ack; 1961 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1962 tp->snd_nxt = tp->snd_una; 1963 1964 switch (tp->t_state) { 1965 1966 /* 1967 * In FIN_WAIT_1 STATE in addition to the processing 1968 * for the ESTABLISHED state if our FIN is now acknowledged 1969 * then enter FIN_WAIT_2. 1970 */ 1971 case TCPS_FIN_WAIT_1: 1972 if (ourfinisacked) { 1973 /* 1974 * If we can't receive any more 1975 * data, then closing user can proceed. 1976 * Starting the timer is contrary to the 1977 * specification, but if we don't get a FIN 1978 * we'll hang forever. 1979 */ 1980 /* XXXjl 1981 * we should release the tp also, and use a 1982 * compressed state. 1983 */ 1984 if (so->so_state & SS_CANTRCVMORE) { 1985 soisdisconnected(so); 1986 callout_reset(tp->tt_2msl, tcp_maxidle, 1987 tcp_timer_2msl, tp); 1988 } 1989 tp->t_state = TCPS_FIN_WAIT_2; 1990 } 1991 break; 1992 1993 /* 1994 * In CLOSING STATE in addition to the processing for 1995 * the ESTABLISHED state if the ACK acknowledges our FIN 1996 * then enter the TIME-WAIT state, otherwise ignore 1997 * the segment. 1998 */ 1999 case TCPS_CLOSING: 2000 if (ourfinisacked) { 2001 KASSERT(headlocked, ("headlocked")); 2002 tcp_twstart(tp); 2003 INP_INFO_WUNLOCK(&tcbinfo); 2004 m_freem(m); 2005 return; 2006 } 2007 break; 2008 2009 /* 2010 * In LAST_ACK, we may still be waiting for data to drain 2011 * and/or to be acked, as well as for the ack of our FIN. 2012 * If our FIN is now acknowledged, delete the TCB, 2013 * enter the closed state and return. 2014 */ 2015 case TCPS_LAST_ACK: 2016 if (ourfinisacked) { 2017 tp = tcp_close(tp); 2018 goto drop; 2019 } 2020 break; 2021 2022 /* 2023 * In TIME_WAIT state the only thing that should arrive 2024 * is a retransmission of the remote FIN. Acknowledge 2025 * it and restart the finack timer. 2026 */ 2027 case TCPS_TIME_WAIT: 2028 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2029 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2030 tcp_timer_2msl, tp); 2031 goto dropafterack; 2032 } 2033 } 2034 2035 step6: 2036 /* 2037 * Update window information. 2038 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2039 */ 2040 if ((thflags & TH_ACK) && 2041 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2042 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2043 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2044 /* keep track of pure window updates */ 2045 if (tlen == 0 && 2046 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2047 tcpstat.tcps_rcvwinupd++; 2048 tp->snd_wnd = tiwin; 2049 tp->snd_wl1 = th->th_seq; 2050 tp->snd_wl2 = th->th_ack; 2051 if (tp->snd_wnd > tp->max_sndwnd) 2052 tp->max_sndwnd = tp->snd_wnd; 2053 needoutput = 1; 2054 } 2055 2056 /* 2057 * Process segments with URG. 2058 */ 2059 if ((thflags & TH_URG) && th->th_urp && 2060 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2061 /* 2062 * This is a kludge, but if we receive and accept 2063 * random urgent pointers, we'll crash in 2064 * soreceive. It's hard to imagine someone 2065 * actually wanting to send this much urgent data. 2066 */ 2067 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2068 th->th_urp = 0; /* XXX */ 2069 thflags &= ~TH_URG; /* XXX */ 2070 goto dodata; /* XXX */ 2071 } 2072 /* 2073 * If this segment advances the known urgent pointer, 2074 * then mark the data stream. This should not happen 2075 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2076 * a FIN has been received from the remote side. 2077 * In these states we ignore the URG. 2078 * 2079 * According to RFC961 (Assigned Protocols), 2080 * the urgent pointer points to the last octet 2081 * of urgent data. We continue, however, 2082 * to consider it to indicate the first octet 2083 * of data past the urgent section as the original 2084 * spec states (in one of two places). 2085 */ 2086 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2087 tp->rcv_up = th->th_seq + th->th_urp; 2088 so->so_oobmark = so->so_rcv.sb_cc + 2089 (tp->rcv_up - tp->rcv_nxt) - 1; 2090 if (so->so_oobmark == 0) 2091 so->so_state |= SS_RCVATMARK; 2092 sohasoutofband(so); 2093 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2094 } 2095 /* 2096 * Remove out of band data so doesn't get presented to user. 2097 * This can happen independent of advancing the URG pointer, 2098 * but if two URG's are pending at once, some out-of-band 2099 * data may creep in... ick. 2100 */ 2101 if (th->th_urp <= (u_long)tlen && 2102 !(so->so_options & SO_OOBINLINE)) { 2103 /* hdr drop is delayed */ 2104 tcp_pulloutofband(so, th, m, drop_hdrlen); 2105 } 2106 } else { 2107 /* 2108 * If no out of band data is expected, 2109 * pull receive urgent pointer along 2110 * with the receive window. 2111 */ 2112 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2113 tp->rcv_up = tp->rcv_nxt; 2114 } 2115 dodata: /* XXX */ 2116 KASSERT(headlocked, ("headlocked")); 2117 /* 2118 * Process the segment text, merging it into the TCP sequencing queue, 2119 * and arranging for acknowledgment of receipt if necessary. 2120 * This process logically involves adjusting tp->rcv_wnd as data 2121 * is presented to the user (this happens in tcp_usrreq.c, 2122 * case PRU_RCVD). If a FIN has already been received on this 2123 * connection then we just ignore the text. 2124 */ 2125 if ((tlen || (thflags & TH_FIN)) && 2126 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2127 m_adj(m, drop_hdrlen); /* delayed header drop */ 2128 /* 2129 * Insert segment which includes th into TCP reassembly queue 2130 * with control block tp. Set thflags to whether reassembly now 2131 * includes a segment with FIN. This handles the common case 2132 * inline (segment is the next to be received on an established 2133 * connection, and the queue is empty), avoiding linkage into 2134 * and removal from the queue and repetition of various 2135 * conversions. 2136 * Set DELACK for segments received in order, but ack 2137 * immediately when segments are out of order (so 2138 * fast retransmit can work). 2139 */ 2140 if (th->th_seq == tp->rcv_nxt && 2141 LIST_EMPTY(&tp->t_segq) && 2142 TCPS_HAVEESTABLISHED(tp->t_state)) { 2143 if (DELAY_ACK(tp)) 2144 tp->t_flags |= TF_DELACK; 2145 else 2146 tp->t_flags |= TF_ACKNOW; 2147 tp->rcv_nxt += tlen; 2148 thflags = th->th_flags & TH_FIN; 2149 tcpstat.tcps_rcvpack++; 2150 tcpstat.tcps_rcvbyte += tlen; 2151 ND6_HINT(tp); 2152 if (so->so_state & SS_CANTRCVMORE) 2153 m_freem(m); 2154 else 2155 sbappend(&so->so_rcv, m); 2156 sorwakeup(so); 2157 } else { 2158 thflags = tcp_reass(tp, th, &tlen, m); 2159 tp->t_flags |= TF_ACKNOW; 2160 } 2161 2162 /* 2163 * Note the amount of data that peer has sent into 2164 * our window, in order to estimate the sender's 2165 * buffer size. 2166 */ 2167 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2168 } else { 2169 m_freem(m); 2170 thflags &= ~TH_FIN; 2171 } 2172 2173 /* 2174 * If FIN is received ACK the FIN and let the user know 2175 * that the connection is closing. 2176 */ 2177 if (thflags & TH_FIN) { 2178 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2179 socantrcvmore(so); 2180 /* 2181 * If connection is half-synchronized 2182 * (ie NEEDSYN flag on) then delay ACK, 2183 * so it may be piggybacked when SYN is sent. 2184 * Otherwise, since we received a FIN then no 2185 * more input can be expected, send ACK now. 2186 */ 2187 if (tp->t_flags & TF_NEEDSYN) 2188 tp->t_flags |= TF_DELACK; 2189 else 2190 tp->t_flags |= TF_ACKNOW; 2191 tp->rcv_nxt++; 2192 } 2193 switch (tp->t_state) { 2194 2195 /* 2196 * In SYN_RECEIVED and ESTABLISHED STATES 2197 * enter the CLOSE_WAIT state. 2198 */ 2199 case TCPS_SYN_RECEIVED: 2200 tp->t_starttime = ticks; 2201 /*FALLTHROUGH*/ 2202 case TCPS_ESTABLISHED: 2203 tp->t_state = TCPS_CLOSE_WAIT; 2204 break; 2205 2206 /* 2207 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2208 * enter the CLOSING state. 2209 */ 2210 case TCPS_FIN_WAIT_1: 2211 tp->t_state = TCPS_CLOSING; 2212 break; 2213 2214 /* 2215 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2216 * starting the time-wait timer, turning off the other 2217 * standard timers. 2218 */ 2219 case TCPS_FIN_WAIT_2: 2220 KASSERT(headlocked == 1, ("headlocked should be 1")); 2221 tcp_twstart(tp); 2222 INP_INFO_WUNLOCK(&tcbinfo); 2223 return; 2224 2225 /* 2226 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2227 */ 2228 case TCPS_TIME_WAIT: 2229 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2230 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2231 tcp_timer_2msl, tp); 2232 break; 2233 } 2234 } 2235 INP_INFO_WUNLOCK(&tcbinfo); 2236 #ifdef TCPDEBUG 2237 if (so->so_options & SO_DEBUG) 2238 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2239 &tcp_savetcp, 0); 2240 #endif 2241 2242 /* 2243 * Return any desired output. 2244 */ 2245 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2246 (void) tcp_output(tp); 2247 2248 check_delack: 2249 if (tp->t_flags & TF_DELACK) { 2250 tp->t_flags &= ~TF_DELACK; 2251 callout_reset(tp->tt_delack, tcp_delacktime, 2252 tcp_timer_delack, tp); 2253 } 2254 INP_UNLOCK(inp); 2255 return; 2256 2257 dropafterack: 2258 /* 2259 * Generate an ACK dropping incoming segment if it occupies 2260 * sequence space, where the ACK reflects our state. 2261 * 2262 * We can now skip the test for the RST flag since all 2263 * paths to this code happen after packets containing 2264 * RST have been dropped. 2265 * 2266 * In the SYN-RECEIVED state, don't send an ACK unless the 2267 * segment we received passes the SYN-RECEIVED ACK test. 2268 * If it fails send a RST. This breaks the loop in the 2269 * "LAND" DoS attack, and also prevents an ACK storm 2270 * between two listening ports that have been sent forged 2271 * SYN segments, each with the source address of the other. 2272 */ 2273 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2274 (SEQ_GT(tp->snd_una, th->th_ack) || 2275 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2276 rstreason = BANDLIM_RST_OPENPORT; 2277 goto dropwithreset; 2278 } 2279 #ifdef TCPDEBUG 2280 if (so->so_options & SO_DEBUG) 2281 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2282 &tcp_savetcp, 0); 2283 #endif 2284 KASSERT(headlocked, ("headlocked should be 1")); 2285 INP_INFO_WUNLOCK(&tcbinfo); 2286 m_freem(m); 2287 tp->t_flags |= TF_ACKNOW; 2288 (void) tcp_output(tp); 2289 INP_UNLOCK(inp); 2290 return; 2291 2292 dropwithreset: 2293 /* 2294 * Generate a RST, dropping incoming segment. 2295 * Make ACK acceptable to originator of segment. 2296 * Don't bother to respond if destination was broadcast/multicast. 2297 */ 2298 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2299 goto drop; 2300 if (isipv6) { 2301 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2302 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2303 goto drop; 2304 } else { 2305 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2306 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2307 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2308 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2309 goto drop; 2310 } 2311 /* IPv6 anycast check is done at tcp6_input() */ 2312 2313 /* 2314 * Perform bandwidth limiting. 2315 */ 2316 if (badport_bandlim(rstreason) < 0) 2317 goto drop; 2318 2319 #ifdef TCPDEBUG 2320 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2321 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2322 &tcp_savetcp, 0); 2323 #endif 2324 2325 if (tp) 2326 INP_UNLOCK(inp); 2327 2328 if (thflags & TH_ACK) 2329 /* mtod() below is safe as long as hdr dropping is delayed */ 2330 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2331 TH_RST); 2332 else { 2333 if (thflags & TH_SYN) 2334 tlen++; 2335 /* mtod() below is safe as long as hdr dropping is delayed */ 2336 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2337 (tcp_seq)0, TH_RST|TH_ACK); 2338 } 2339 if (headlocked) 2340 INP_INFO_WUNLOCK(&tcbinfo); 2341 return; 2342 2343 drop: 2344 /* 2345 * Drop space held by incoming segment and return. 2346 */ 2347 #ifdef TCPDEBUG 2348 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2349 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2350 &tcp_savetcp, 0); 2351 #endif 2352 if (tp) 2353 INP_UNLOCK(inp); 2354 m_freem(m); 2355 if (headlocked) 2356 INP_INFO_WUNLOCK(&tcbinfo); 2357 return; 2358 } 2359 2360 /* 2361 * Parse TCP options and place in tcpopt. 2362 */ 2363 static void 2364 tcp_dooptions(to, cp, cnt, is_syn) 2365 struct tcpopt *to; 2366 u_char *cp; 2367 int cnt; 2368 int is_syn; 2369 { 2370 int opt, optlen; 2371 2372 to->to_flags = 0; 2373 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2374 opt = cp[0]; 2375 if (opt == TCPOPT_EOL) 2376 break; 2377 if (opt == TCPOPT_NOP) 2378 optlen = 1; 2379 else { 2380 if (cnt < 2) 2381 break; 2382 optlen = cp[1]; 2383 if (optlen < 2 || optlen > cnt) 2384 break; 2385 } 2386 switch (opt) { 2387 case TCPOPT_MAXSEG: 2388 if (optlen != TCPOLEN_MAXSEG) 2389 continue; 2390 if (!is_syn) 2391 continue; 2392 to->to_flags |= TOF_MSS; 2393 bcopy((char *)cp + 2, 2394 (char *)&to->to_mss, sizeof(to->to_mss)); 2395 to->to_mss = ntohs(to->to_mss); 2396 break; 2397 case TCPOPT_WINDOW: 2398 if (optlen != TCPOLEN_WINDOW) 2399 continue; 2400 if (! is_syn) 2401 continue; 2402 to->to_flags |= TOF_SCALE; 2403 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2404 break; 2405 case TCPOPT_TIMESTAMP: 2406 if (optlen != TCPOLEN_TIMESTAMP) 2407 continue; 2408 to->to_flags |= TOF_TS; 2409 bcopy((char *)cp + 2, 2410 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2411 to->to_tsval = ntohl(to->to_tsval); 2412 bcopy((char *)cp + 6, 2413 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2414 to->to_tsecr = ntohl(to->to_tsecr); 2415 break; 2416 case TCPOPT_CC: 2417 if (optlen != TCPOLEN_CC) 2418 continue; 2419 to->to_flags |= TOF_CC; 2420 bcopy((char *)cp + 2, 2421 (char *)&to->to_cc, sizeof(to->to_cc)); 2422 to->to_cc = ntohl(to->to_cc); 2423 break; 2424 case TCPOPT_CCNEW: 2425 if (optlen != TCPOLEN_CC) 2426 continue; 2427 if (!is_syn) 2428 continue; 2429 to->to_flags |= TOF_CCNEW; 2430 bcopy((char *)cp + 2, 2431 (char *)&to->to_cc, sizeof(to->to_cc)); 2432 to->to_cc = ntohl(to->to_cc); 2433 break; 2434 case TCPOPT_CCECHO: 2435 if (optlen != TCPOLEN_CC) 2436 continue; 2437 if (!is_syn) 2438 continue; 2439 to->to_flags |= TOF_CCECHO; 2440 bcopy((char *)cp + 2, 2441 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2442 to->to_ccecho = ntohl(to->to_ccecho); 2443 break; 2444 default: 2445 continue; 2446 } 2447 } 2448 } 2449 2450 /* 2451 * Pull out of band byte out of a segment so 2452 * it doesn't appear in the user's data queue. 2453 * It is still reflected in the segment length for 2454 * sequencing purposes. 2455 */ 2456 static void 2457 tcp_pulloutofband(so, th, m, off) 2458 struct socket *so; 2459 struct tcphdr *th; 2460 register struct mbuf *m; 2461 int off; /* delayed to be droped hdrlen */ 2462 { 2463 int cnt = off + th->th_urp - 1; 2464 2465 while (cnt >= 0) { 2466 if (m->m_len > cnt) { 2467 char *cp = mtod(m, caddr_t) + cnt; 2468 struct tcpcb *tp = sototcpcb(so); 2469 2470 tp->t_iobc = *cp; 2471 tp->t_oobflags |= TCPOOB_HAVEDATA; 2472 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2473 m->m_len--; 2474 if (m->m_flags & M_PKTHDR) 2475 m->m_pkthdr.len--; 2476 return; 2477 } 2478 cnt -= m->m_len; 2479 m = m->m_next; 2480 if (m == 0) 2481 break; 2482 } 2483 panic("tcp_pulloutofband"); 2484 } 2485 2486 /* 2487 * Collect new round-trip time estimate 2488 * and update averages and current timeout. 2489 */ 2490 static void 2491 tcp_xmit_timer(tp, rtt) 2492 register struct tcpcb *tp; 2493 int rtt; 2494 { 2495 register int delta; 2496 2497 tcpstat.tcps_rttupdated++; 2498 tp->t_rttupdated++; 2499 if (tp->t_srtt != 0) { 2500 /* 2501 * srtt is stored as fixed point with 5 bits after the 2502 * binary point (i.e., scaled by 8). The following magic 2503 * is equivalent to the smoothing algorithm in rfc793 with 2504 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2505 * point). Adjust rtt to origin 0. 2506 */ 2507 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2508 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2509 2510 if ((tp->t_srtt += delta) <= 0) 2511 tp->t_srtt = 1; 2512 2513 /* 2514 * We accumulate a smoothed rtt variance (actually, a 2515 * smoothed mean difference), then set the retransmit 2516 * timer to smoothed rtt + 4 times the smoothed variance. 2517 * rttvar is stored as fixed point with 4 bits after the 2518 * binary point (scaled by 16). The following is 2519 * equivalent to rfc793 smoothing with an alpha of .75 2520 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2521 * rfc793's wired-in beta. 2522 */ 2523 if (delta < 0) 2524 delta = -delta; 2525 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2526 if ((tp->t_rttvar += delta) <= 0) 2527 tp->t_rttvar = 1; 2528 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2529 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2530 } else { 2531 /* 2532 * No rtt measurement yet - use the unsmoothed rtt. 2533 * Set the variance to half the rtt (so our first 2534 * retransmit happens at 3*rtt). 2535 */ 2536 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2537 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2538 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2539 } 2540 tp->t_rtttime = 0; 2541 tp->t_rxtshift = 0; 2542 2543 /* 2544 * the retransmit should happen at rtt + 4 * rttvar. 2545 * Because of the way we do the smoothing, srtt and rttvar 2546 * will each average +1/2 tick of bias. When we compute 2547 * the retransmit timer, we want 1/2 tick of rounding and 2548 * 1 extra tick because of +-1/2 tick uncertainty in the 2549 * firing of the timer. The bias will give us exactly the 2550 * 1.5 tick we need. But, because the bias is 2551 * statistical, we have to test that we don't drop below 2552 * the minimum feasible timer (which is 2 ticks). 2553 */ 2554 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2555 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2556 2557 /* 2558 * We received an ack for a packet that wasn't retransmitted; 2559 * it is probably safe to discard any error indications we've 2560 * received recently. This isn't quite right, but close enough 2561 * for now (a route might have failed after we sent a segment, 2562 * and the return path might not be symmetrical). 2563 */ 2564 tp->t_softerror = 0; 2565 } 2566 2567 /* 2568 * Determine a reasonable value for maxseg size. 2569 * If the route is known, check route for mtu. 2570 * If none, use an mss that can be handled on the outgoing 2571 * interface without forcing IP to fragment; if bigger than 2572 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2573 * to utilize large mbufs. If no route is found, route has no mtu, 2574 * or the destination isn't local, use a default, hopefully conservative 2575 * size (usually 512 or the default IP max size, but no more than the mtu 2576 * of the interface), as we can't discover anything about intervening 2577 * gateways or networks. We also initialize the congestion/slow start 2578 * window to be a single segment if the destination isn't local. 2579 * While looking at the routing entry, we also initialize other path-dependent 2580 * parameters from pre-set or cached values in the routing entry. 2581 * 2582 * Also take into account the space needed for options that we 2583 * send regularly. Make maxseg shorter by that amount to assure 2584 * that we can send maxseg amount of data even when the options 2585 * are present. Store the upper limit of the length of options plus 2586 * data in maxopd. 2587 * 2588 * NOTE that this routine is only called when we process an incoming 2589 * segment, for outgoing segments only tcp_mssopt is called. 2590 * 2591 * In case of T/TCP, we call this routine during implicit connection 2592 * setup as well (offer = -1), to initialize maxseg from the cached 2593 * MSS of our peer. 2594 */ 2595 void 2596 tcp_mss(tp, offer) 2597 struct tcpcb *tp; 2598 int offer; 2599 { 2600 register struct rtentry *rt; 2601 struct ifnet *ifp; 2602 register int rtt, mss; 2603 u_long bufsize; 2604 struct inpcb *inp = tp->t_inpcb; 2605 struct socket *so; 2606 struct rmxp_tao *taop; 2607 int origoffer = offer; 2608 #ifdef INET6 2609 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2610 size_t min_protoh = isipv6 ? 2611 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2612 sizeof (struct tcpiphdr); 2613 #else 2614 const int isipv6 = 0; 2615 const size_t min_protoh = sizeof (struct tcpiphdr); 2616 #endif 2617 2618 if (isipv6) 2619 rt = tcp_rtlookup6(&inp->inp_inc); 2620 else 2621 rt = tcp_rtlookup(&inp->inp_inc); 2622 if (rt == NULL) { 2623 tp->t_maxopd = tp->t_maxseg = 2624 isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2625 return; 2626 } 2627 ifp = rt->rt_ifp; 2628 so = inp->inp_socket; 2629 2630 taop = rmx_taop(rt->rt_rmx); 2631 /* 2632 * Offer == -1 means that we didn't receive SYN yet, 2633 * use cached value in that case; 2634 */ 2635 if (offer == -1) 2636 offer = taop->tao_mssopt; 2637 /* 2638 * Offer == 0 means that there was no MSS on the SYN segment, 2639 * in this case we use tcp_mssdflt. 2640 */ 2641 if (offer == 0) 2642 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2643 else 2644 /* 2645 * Sanity check: make sure that maxopd will be large 2646 * enough to allow some data on segments even is the 2647 * all the option space is used (40bytes). Otherwise 2648 * funny things may happen in tcp_output. 2649 */ 2650 offer = max(offer, 64); 2651 taop->tao_mssopt = offer; 2652 2653 /* 2654 * While we're here, check if there's an initial rtt 2655 * or rttvar. Convert from the route-table units 2656 * to scaled multiples of the slow timeout timer. 2657 */ 2658 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2659 /* 2660 * XXX the lock bit for RTT indicates that the value 2661 * is also a minimum value; this is subject to time. 2662 */ 2663 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2664 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2665 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2666 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2667 tcpstat.tcps_usedrtt++; 2668 if (rt->rt_rmx.rmx_rttvar) { 2669 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2670 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2671 tcpstat.tcps_usedrttvar++; 2672 } else { 2673 /* default variation is +- 1 rtt */ 2674 tp->t_rttvar = 2675 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2676 } 2677 TCPT_RANGESET(tp->t_rxtcur, 2678 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2679 tp->t_rttmin, TCPTV_REXMTMAX); 2680 } 2681 /* 2682 * if there's an mtu associated with the route, use it 2683 * else, use the link mtu. 2684 */ 2685 if (rt->rt_rmx.rmx_mtu) 2686 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2687 else { 2688 if (isipv6) { 2689 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2690 min_protoh; 2691 if (!in6_localaddr(&inp->in6p_faddr)) 2692 mss = min(mss, tcp_v6mssdflt); 2693 } else { 2694 mss = ifp->if_mtu - min_protoh; 2695 if (!in_localaddr(inp->inp_faddr)) 2696 mss = min(mss, tcp_mssdflt); 2697 } 2698 } 2699 mss = min(mss, offer); 2700 /* 2701 * maxopd stores the maximum length of data AND options 2702 * in a segment; maxseg is the amount of data in a normal 2703 * segment. We need to store this value (maxopd) apart 2704 * from maxseg, because now every segment carries options 2705 * and thus we normally have somewhat less data in segments. 2706 */ 2707 tp->t_maxopd = mss; 2708 2709 /* 2710 * In case of T/TCP, origoffer==-1 indicates, that no segments 2711 * were received yet. In this case we just guess, otherwise 2712 * we do the same as before T/TCP. 2713 */ 2714 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2715 (origoffer == -1 || 2716 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2717 mss -= TCPOLEN_TSTAMP_APPA; 2718 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2719 (origoffer == -1 || 2720 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2721 mss -= TCPOLEN_CC_APPA; 2722 2723 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2724 if (mss > MCLBYTES) 2725 mss &= ~(MCLBYTES-1); 2726 #else 2727 if (mss > MCLBYTES) 2728 mss = mss / MCLBYTES * MCLBYTES; 2729 #endif 2730 /* 2731 * If there's a pipesize, change the socket buffer 2732 * to that size. Make the socket buffers an integral 2733 * number of mss units; if the mss is larger than 2734 * the socket buffer, decrease the mss. 2735 */ 2736 #ifdef RTV_SPIPE 2737 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2738 #endif 2739 bufsize = so->so_snd.sb_hiwat; 2740 if (bufsize < mss) 2741 mss = bufsize; 2742 else { 2743 bufsize = roundup(bufsize, mss); 2744 if (bufsize > sb_max) 2745 bufsize = sb_max; 2746 if (bufsize > so->so_snd.sb_hiwat) 2747 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2748 } 2749 tp->t_maxseg = mss; 2750 2751 #ifdef RTV_RPIPE 2752 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2753 #endif 2754 bufsize = so->so_rcv.sb_hiwat; 2755 if (bufsize > mss) { 2756 bufsize = roundup(bufsize, mss); 2757 if (bufsize > sb_max) 2758 bufsize = sb_max; 2759 if (bufsize > so->so_rcv.sb_hiwat) 2760 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2761 } 2762 2763 /* 2764 * Set the slow-start flight size depending on whether this 2765 * is a local network or not. 2766 */ 2767 if (tcp_do_rfc3390) 2768 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2769 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2770 (!isipv6 && in_localaddr(inp->inp_faddr))) 2771 tp->snd_cwnd = mss * ss_fltsz_local; 2772 else 2773 tp->snd_cwnd = mss * ss_fltsz; 2774 2775 if (rt->rt_rmx.rmx_ssthresh) { 2776 /* 2777 * There's some sort of gateway or interface 2778 * buffer limit on the path. Use this to set 2779 * the slow start threshhold, but set the 2780 * threshold to no less than 2*mss. 2781 */ 2782 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2783 tcpstat.tcps_usedssthresh++; 2784 } 2785 } 2786 2787 /* 2788 * Determine the MSS option to send on an outgoing SYN. 2789 */ 2790 int 2791 tcp_mssopt(tp) 2792 struct tcpcb *tp; 2793 { 2794 struct rtentry *rt; 2795 #ifdef INET6 2796 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2797 size_t min_protoh = isipv6 ? 2798 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2799 sizeof (struct tcpiphdr); 2800 #else 2801 const int isipv6 = 0; 2802 const size_t min_protoh = sizeof (struct tcpiphdr); 2803 #endif 2804 2805 if (isipv6) 2806 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2807 else 2808 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2809 if (rt == NULL) 2810 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2811 2812 return (rt->rt_ifp->if_mtu - min_protoh); 2813 } 2814 2815 2816 /* 2817 * On a partial ack arrives, force the retransmission of the 2818 * next unacknowledged segment. Do not clear tp->t_dupacks. 2819 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2820 * be started again. 2821 */ 2822 static void 2823 tcp_newreno_partial_ack(tp, th) 2824 struct tcpcb *tp; 2825 struct tcphdr *th; 2826 { 2827 tcp_seq onxt = tp->snd_nxt; 2828 u_long ocwnd = tp->snd_cwnd; 2829 2830 callout_stop(tp->tt_rexmt); 2831 tp->t_rtttime = 0; 2832 tp->snd_nxt = th->th_ack; 2833 /* 2834 * Set snd_cwnd to one segment beyond acknowledged offset. 2835 * (tp->snd_una has not yet been updated when this function is called.) 2836 */ 2837 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2838 tp->t_flags |= TF_ACKNOW; 2839 (void) tcp_output(tp); 2840 tp->snd_cwnd = ocwnd; 2841 if (SEQ_GT(onxt, tp->snd_nxt)) 2842 tp->snd_nxt = onxt; 2843 /* 2844 * Partial window deflation. Relies on fact that tp->snd_una 2845 * not updated yet. 2846 */ 2847 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2848 } 2849 2850 /* 2851 * Returns 1 if the TIME_WAIT state was killed and we should start over, 2852 * looking for a pcb in the listen state. Returns 0 otherwise. 2853 */ 2854 static int 2855 tcp_timewait(tw, to, th, m, tlen) 2856 struct tcptw *tw; 2857 struct tcpopt *to; 2858 struct tcphdr *th; 2859 struct mbuf *m; 2860 int tlen; 2861 { 2862 int thflags; 2863 tcp_seq seq; 2864 #ifdef INET6 2865 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 2866 #else 2867 const int isipv6 = 0; 2868 #endif 2869 2870 thflags = th->th_flags; 2871 2872 /* 2873 * NOTE: for FIN_WAIT_2 (to be added later), 2874 * must validate sequence number before accepting RST 2875 */ 2876 2877 /* 2878 * If the segment contains RST: 2879 * Drop the segment - see Stevens, vol. 2, p. 964 and 2880 * RFC 1337. 2881 */ 2882 if (thflags & TH_RST) 2883 goto drop; 2884 2885 /* 2886 * If segment contains a SYN and CC [not CC.NEW] option: 2887 * if connection duration > MSL, drop packet and send RST; 2888 * 2889 * if SEG.CC > CCrecv then is new SYN. 2890 * Complete close and delete TCPCB. Then reprocess 2891 * segment, hoping to find new TCPCB in LISTEN state; 2892 * 2893 * else must be old SYN; drop it. 2894 * else do normal processing. 2895 */ 2896 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 2897 if ((ticks - tw->t_starttime) > tcp_msl) 2898 goto reset; 2899 if (CC_GT(to->to_cc, tw->cc_recv)) { 2900 (void) tcp_twclose(tw, 0); 2901 return (1); 2902 } 2903 goto drop; 2904 } 2905 2906 #if 0 2907 /* PAWS not needed at the moment */ 2908 /* 2909 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2910 * and it's less than ts_recent, drop it. 2911 */ 2912 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2913 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2914 if ((thflags & TH_ACK) == 0) 2915 goto drop; 2916 goto ack; 2917 } 2918 /* 2919 * ts_recent is never updated because we never accept new segments. 2920 */ 2921 #endif 2922 2923 /* 2924 * If a new connection request is received 2925 * while in TIME_WAIT, drop the old connection 2926 * and start over if the sequence numbers 2927 * are above the previous ones. 2928 */ 2929 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 2930 (void) tcp_twclose(tw, 0); 2931 return (1); 2932 } 2933 2934 /* 2935 * Drop the the segment if it does not contain an ACK. 2936 */ 2937 if ((thflags & TH_ACK) == 0) 2938 goto drop; 2939 2940 /* 2941 * Reset the 2MSL timer if this is a duplicate FIN. 2942 */ 2943 if (thflags & TH_FIN) { 2944 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 2945 if (seq + 1 == tw->rcv_nxt) 2946 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 2947 } 2948 2949 /* 2950 * Acknowledge the segment if it has data or is not a duplicate ACK. 2951 */ 2952 if (thflags != TH_ACK || tlen != 0 || 2953 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 2954 tcp_twrespond(tw, NULL, m, TH_ACK); 2955 goto drop; 2956 2957 reset: 2958 /* 2959 * Generate a RST, dropping incoming segment. 2960 * Make ACK acceptable to originator of segment. 2961 * Don't bother to respond if destination was broadcast/multicast. 2962 */ 2963 if (m->m_flags & (M_BCAST|M_MCAST)) 2964 goto drop; 2965 if (isipv6) { 2966 struct ip6_hdr *ip6; 2967 2968 /* IPv6 anycast check is done at tcp6_input() */ 2969 ip6 = mtod(m, struct ip6_hdr *); 2970 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2971 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2972 goto drop; 2973 } else { 2974 struct ip *ip; 2975 2976 ip = mtod(m, struct ip *); 2977 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2978 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2979 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2980 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2981 goto drop; 2982 } 2983 if (thflags & TH_ACK) { 2984 tcp_respond(NULL, 2985 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 2986 } else { 2987 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 2988 tcp_respond(NULL, 2989 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 2990 } 2991 INP_UNLOCK(tw->tw_inpcb); 2992 return (0); 2993 2994 drop: 2995 INP_UNLOCK(tw->tw_inpcb); 2996 m_freem(m); 2997 return (0); 2998 } 2999