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 && !IN_FASTRECOVERY(tp)))) { 996 KASSERT(headlocked, ("headlocked")); 997 INP_INFO_WUNLOCK(&tcbinfo); 998 /* 999 * this is a pure ack for outstanding data. 1000 */ 1001 ++tcpstat.tcps_predack; 1002 /* 1003 * "bad retransmit" recovery 1004 */ 1005 if (tp->t_rxtshift == 1 && 1006 ticks < tp->t_badrxtwin) { 1007 ++tcpstat.tcps_sndrexmitbad; 1008 tp->snd_cwnd = tp->snd_cwnd_prev; 1009 tp->snd_ssthresh = 1010 tp->snd_ssthresh_prev; 1011 tp->snd_recover = tp->snd_recover_prev; 1012 if (tp->t_flags & TF_WASFRECOVERY) 1013 ENTER_FASTRECOVERY(tp); 1014 tp->snd_nxt = tp->snd_max; 1015 tp->t_badrxtwin = 0; 1016 } 1017 1018 /* 1019 * Recalculate the transmit timer / rtt. 1020 * 1021 * Some boxes send broken timestamp replies 1022 * during the SYN+ACK phase, ignore 1023 * timestamps of 0 or we could calculate a 1024 * huge RTT and blow up the retransmit timer. 1025 */ 1026 if ((to.to_flags & TOF_TS) != 0 && 1027 to.to_tsecr) { 1028 tcp_xmit_timer(tp, 1029 ticks - to.to_tsecr + 1); 1030 } else if (tp->t_rtttime && 1031 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1032 tcp_xmit_timer(tp, 1033 ticks - tp->t_rtttime); 1034 } 1035 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1036 acked = th->th_ack - tp->snd_una; 1037 tcpstat.tcps_rcvackpack++; 1038 tcpstat.tcps_rcvackbyte += acked; 1039 sbdrop(&so->so_snd, acked); 1040 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1041 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1042 tp->snd_recover = th->th_ack - 1; 1043 tp->snd_una = th->th_ack; 1044 /* 1045 * pull snd_wl2 up to prevent seq wrap relative 1046 * to th_ack. 1047 */ 1048 tp->snd_wl2 = th->th_ack; 1049 tp->t_dupacks = 0; 1050 m_freem(m); 1051 ND6_HINT(tp); /* some progress has been done */ 1052 1053 /* 1054 * If all outstanding data are acked, stop 1055 * retransmit timer, otherwise restart timer 1056 * using current (possibly backed-off) value. 1057 * If process is waiting for space, 1058 * wakeup/selwakeup/signal. If data 1059 * are ready to send, let tcp_output 1060 * decide between more output or persist. 1061 */ 1062 if (tp->snd_una == tp->snd_max) 1063 callout_stop(tp->tt_rexmt); 1064 else if (!callout_active(tp->tt_persist)) 1065 callout_reset(tp->tt_rexmt, 1066 tp->t_rxtcur, 1067 tcp_timer_rexmt, tp); 1068 1069 sowwakeup(so); 1070 if (so->so_snd.sb_cc) 1071 (void) tcp_output(tp); 1072 goto check_delack; 1073 } 1074 } else if (th->th_ack == tp->snd_una && 1075 LIST_EMPTY(&tp->t_segq) && 1076 tlen <= sbspace(&so->so_rcv)) { 1077 KASSERT(headlocked, ("headlocked")); 1078 INP_INFO_WUNLOCK(&tcbinfo); 1079 /* 1080 * this is a pure, in-sequence data packet 1081 * with nothing on the reassembly queue and 1082 * we have enough buffer space to take it. 1083 */ 1084 ++tcpstat.tcps_preddat; 1085 tp->rcv_nxt += tlen; 1086 /* 1087 * Pull snd_wl1 up to prevent seq wrap relative to 1088 * th_seq. 1089 */ 1090 tp->snd_wl1 = th->th_seq; 1091 /* 1092 * Pull rcv_up up to prevent seq wrap relative to 1093 * rcv_nxt. 1094 */ 1095 tp->rcv_up = tp->rcv_nxt; 1096 tcpstat.tcps_rcvpack++; 1097 tcpstat.tcps_rcvbyte += tlen; 1098 ND6_HINT(tp); /* some progress has been done */ 1099 /* 1100 * Add data to socket buffer. 1101 */ 1102 if (so->so_state & SS_CANTRCVMORE) { 1103 m_freem(m); 1104 } else { 1105 m_adj(m, drop_hdrlen); /* delayed header drop */ 1106 sbappend(&so->so_rcv, m); 1107 } 1108 sorwakeup(so); 1109 if (DELAY_ACK(tp)) { 1110 tp->t_flags |= TF_DELACK; 1111 } else { 1112 tp->t_flags |= TF_ACKNOW; 1113 tcp_output(tp); 1114 } 1115 goto check_delack; 1116 } 1117 } 1118 1119 /* 1120 * Calculate amount of space in receive window, 1121 * and then do TCP input processing. 1122 * Receive window is amount of space in rcv queue, 1123 * but not less than advertised window. 1124 */ 1125 { int win; 1126 1127 win = sbspace(&so->so_rcv); 1128 if (win < 0) 1129 win = 0; 1130 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1131 } 1132 1133 switch (tp->t_state) { 1134 1135 /* 1136 * If the state is SYN_RECEIVED: 1137 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1138 */ 1139 case TCPS_SYN_RECEIVED: 1140 if ((thflags & TH_ACK) && 1141 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1142 SEQ_GT(th->th_ack, tp->snd_max))) { 1143 rstreason = BANDLIM_RST_OPENPORT; 1144 goto dropwithreset; 1145 } 1146 break; 1147 1148 /* 1149 * If the state is SYN_SENT: 1150 * if seg contains an ACK, but not for our SYN, drop the input. 1151 * if seg contains a RST, then drop the connection. 1152 * if seg does not contain SYN, then drop it. 1153 * Otherwise this is an acceptable SYN segment 1154 * initialize tp->rcv_nxt and tp->irs 1155 * if seg contains ack then advance tp->snd_una 1156 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1157 * arrange for segment to be acked (eventually) 1158 * continue processing rest of data/controls, beginning with URG 1159 */ 1160 case TCPS_SYN_SENT: 1161 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1162 taop = &tao_noncached; 1163 bzero(taop, sizeof(*taop)); 1164 } 1165 1166 if ((thflags & TH_ACK) && 1167 (SEQ_LEQ(th->th_ack, tp->iss) || 1168 SEQ_GT(th->th_ack, tp->snd_max))) { 1169 /* 1170 * If we have a cached CCsent for the remote host, 1171 * hence we haven't just crashed and restarted, 1172 * do not send a RST. This may be a retransmission 1173 * from the other side after our earlier ACK was lost. 1174 * Our new SYN, when it arrives, will serve as the 1175 * needed ACK. 1176 */ 1177 if (taop->tao_ccsent != 0) 1178 goto drop; 1179 else { 1180 rstreason = BANDLIM_UNLIMITED; 1181 goto dropwithreset; 1182 } 1183 } 1184 if (thflags & TH_RST) { 1185 if (thflags & TH_ACK) 1186 tp = tcp_drop(tp, ECONNREFUSED); 1187 goto drop; 1188 } 1189 if ((thflags & TH_SYN) == 0) 1190 goto drop; 1191 tp->snd_wnd = th->th_win; /* initial send window */ 1192 tp->cc_recv = to.to_cc; /* foreign CC */ 1193 1194 tp->irs = th->th_seq; 1195 tcp_rcvseqinit(tp); 1196 if (thflags & TH_ACK) { 1197 /* 1198 * Our SYN was acked. If segment contains CC.ECHO 1199 * option, check it to make sure this segment really 1200 * matches our SYN. If not, just drop it as old 1201 * duplicate, but send an RST if we're still playing 1202 * by the old rules. If no CC.ECHO option, make sure 1203 * we don't get fooled into using T/TCP. 1204 */ 1205 if (to.to_flags & TOF_CCECHO) { 1206 if (tp->cc_send != to.to_ccecho) { 1207 if (taop->tao_ccsent != 0) 1208 goto drop; 1209 else { 1210 rstreason = BANDLIM_UNLIMITED; 1211 goto dropwithreset; 1212 } 1213 } 1214 } else 1215 tp->t_flags &= ~TF_RCVD_CC; 1216 tcpstat.tcps_connects++; 1217 soisconnected(so); 1218 #ifdef MAC 1219 mac_set_socket_peer_from_mbuf(m, so); 1220 #endif 1221 /* Do window scaling on this connection? */ 1222 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1223 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1224 tp->snd_scale = tp->requested_s_scale; 1225 tp->rcv_scale = tp->request_r_scale; 1226 } 1227 /* Segment is acceptable, update cache if undefined. */ 1228 if (taop->tao_ccsent == 0) 1229 taop->tao_ccsent = to.to_ccecho; 1230 1231 tp->rcv_adv += tp->rcv_wnd; 1232 tp->snd_una++; /* SYN is acked */ 1233 /* 1234 * If there's data, delay ACK; if there's also a FIN 1235 * ACKNOW will be turned on later. 1236 */ 1237 if (DELAY_ACK(tp) && tlen != 0) 1238 callout_reset(tp->tt_delack, tcp_delacktime, 1239 tcp_timer_delack, tp); 1240 else 1241 tp->t_flags |= TF_ACKNOW; 1242 /* 1243 * Received <SYN,ACK> in SYN_SENT[*] state. 1244 * Transitions: 1245 * SYN_SENT --> ESTABLISHED 1246 * SYN_SENT* --> FIN_WAIT_1 1247 */ 1248 tp->t_starttime = ticks; 1249 if (tp->t_flags & TF_NEEDFIN) { 1250 tp->t_state = TCPS_FIN_WAIT_1; 1251 tp->t_flags &= ~TF_NEEDFIN; 1252 thflags &= ~TH_SYN; 1253 } else { 1254 tp->t_state = TCPS_ESTABLISHED; 1255 callout_reset(tp->tt_keep, tcp_keepidle, 1256 tcp_timer_keep, tp); 1257 } 1258 } else { 1259 /* 1260 * Received initial SYN in SYN-SENT[*] state => 1261 * simultaneous open. If segment contains CC option 1262 * and there is a cached CC, apply TAO test. 1263 * If it succeeds, connection is * half-synchronized. 1264 * Otherwise, do 3-way handshake: 1265 * SYN-SENT -> SYN-RECEIVED 1266 * SYN-SENT* -> SYN-RECEIVED* 1267 * If there was no CC option, clear cached CC value. 1268 */ 1269 tp->t_flags |= TF_ACKNOW; 1270 callout_stop(tp->tt_rexmt); 1271 if (to.to_flags & TOF_CC) { 1272 if (taop->tao_cc != 0 && 1273 CC_GT(to.to_cc, taop->tao_cc)) { 1274 /* 1275 * update cache and make transition: 1276 * SYN-SENT -> ESTABLISHED* 1277 * SYN-SENT* -> FIN-WAIT-1* 1278 */ 1279 taop->tao_cc = to.to_cc; 1280 tp->t_starttime = ticks; 1281 if (tp->t_flags & TF_NEEDFIN) { 1282 tp->t_state = TCPS_FIN_WAIT_1; 1283 tp->t_flags &= ~TF_NEEDFIN; 1284 } else { 1285 tp->t_state = TCPS_ESTABLISHED; 1286 callout_reset(tp->tt_keep, 1287 tcp_keepidle, 1288 tcp_timer_keep, 1289 tp); 1290 } 1291 tp->t_flags |= TF_NEEDSYN; 1292 } else 1293 tp->t_state = TCPS_SYN_RECEIVED; 1294 } else { 1295 /* CC.NEW or no option => invalidate cache */ 1296 taop->tao_cc = 0; 1297 tp->t_state = TCPS_SYN_RECEIVED; 1298 } 1299 } 1300 1301 trimthenstep6: 1302 /* 1303 * Advance th->th_seq to correspond to first data byte. 1304 * If data, trim to stay within window, 1305 * dropping FIN if necessary. 1306 */ 1307 th->th_seq++; 1308 if (tlen > tp->rcv_wnd) { 1309 todrop = tlen - tp->rcv_wnd; 1310 m_adj(m, -todrop); 1311 tlen = tp->rcv_wnd; 1312 thflags &= ~TH_FIN; 1313 tcpstat.tcps_rcvpackafterwin++; 1314 tcpstat.tcps_rcvbyteafterwin += todrop; 1315 } 1316 tp->snd_wl1 = th->th_seq - 1; 1317 tp->rcv_up = th->th_seq; 1318 /* 1319 * Client side of transaction: already sent SYN and data. 1320 * If the remote host used T/TCP to validate the SYN, 1321 * our data will be ACK'd; if so, enter normal data segment 1322 * processing in the middle of step 5, ack processing. 1323 * Otherwise, goto step 6. 1324 */ 1325 if (thflags & TH_ACK) 1326 goto process_ACK; 1327 1328 goto step6; 1329 1330 /* 1331 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1332 * if segment contains a SYN and CC [not CC.NEW] option: 1333 * if state == TIME_WAIT and connection duration > MSL, 1334 * drop packet and send RST; 1335 * 1336 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1337 * ack the FIN (and data) in retransmission queue. 1338 * Complete close and delete TCPCB. Then reprocess 1339 * segment, hoping to find new TCPCB in LISTEN state; 1340 * 1341 * else must be old SYN; drop it. 1342 * else do normal processing. 1343 */ 1344 case TCPS_LAST_ACK: 1345 case TCPS_CLOSING: 1346 case TCPS_TIME_WAIT: 1347 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1348 if ((thflags & TH_SYN) && 1349 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1350 if (tp->t_state == TCPS_TIME_WAIT && 1351 (ticks - tp->t_starttime) > tcp_msl) { 1352 rstreason = BANDLIM_UNLIMITED; 1353 goto dropwithreset; 1354 } 1355 if (CC_GT(to.to_cc, tp->cc_recv)) { 1356 tp = tcp_close(tp); 1357 goto findpcb; 1358 } 1359 else 1360 goto drop; 1361 } 1362 break; /* continue normal processing */ 1363 } 1364 1365 /* 1366 * States other than LISTEN or SYN_SENT. 1367 * First check the RST flag and sequence number since reset segments 1368 * are exempt from the timestamp and connection count tests. This 1369 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1370 * below which allowed reset segments in half the sequence space 1371 * to fall though and be processed (which gives forged reset 1372 * segments with a random sequence number a 50 percent chance of 1373 * killing a connection). 1374 * Then check timestamp, if present. 1375 * Then check the connection count, if present. 1376 * Then check that at least some bytes of segment are within 1377 * receive window. If segment begins before rcv_nxt, 1378 * drop leading data (and SYN); if nothing left, just ack. 1379 * 1380 * 1381 * If the RST bit is set, check the sequence number to see 1382 * if this is a valid reset segment. 1383 * RFC 793 page 37: 1384 * In all states except SYN-SENT, all reset (RST) segments 1385 * are validated by checking their SEQ-fields. A reset is 1386 * valid if its sequence number is in the window. 1387 * Note: this does not take into account delayed ACKs, so 1388 * we should test against last_ack_sent instead of rcv_nxt. 1389 * The sequence number in the reset segment is normally an 1390 * echo of our outgoing acknowlegement numbers, but some hosts 1391 * send a reset with the sequence number at the rightmost edge 1392 * of our receive window, and we have to handle this case. 1393 * If we have multiple segments in flight, the intial reset 1394 * segment sequence numbers will be to the left of last_ack_sent, 1395 * but they will eventually catch up. 1396 * In any case, it never made sense to trim reset segments to 1397 * fit the receive window since RFC 1122 says: 1398 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1399 * 1400 * A TCP SHOULD allow a received RST segment to include data. 1401 * 1402 * DISCUSSION 1403 * It has been suggested that a RST segment could contain 1404 * ASCII text that encoded and explained the cause of the 1405 * RST. No standard has yet been established for such 1406 * data. 1407 * 1408 * If the reset segment passes the sequence number test examine 1409 * the state: 1410 * SYN_RECEIVED STATE: 1411 * If passive open, return to LISTEN state. 1412 * If active open, inform user that connection was refused. 1413 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1414 * Inform user that connection was reset, and close tcb. 1415 * CLOSING, LAST_ACK STATES: 1416 * Close the tcb. 1417 * TIME_WAIT STATE: 1418 * Drop the segment - see Stevens, vol. 2, p. 964 and 1419 * RFC 1337. 1420 */ 1421 if (thflags & TH_RST) { 1422 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1423 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1424 switch (tp->t_state) { 1425 1426 case TCPS_SYN_RECEIVED: 1427 so->so_error = ECONNREFUSED; 1428 goto close; 1429 1430 case TCPS_ESTABLISHED: 1431 case TCPS_FIN_WAIT_1: 1432 case TCPS_FIN_WAIT_2: 1433 case TCPS_CLOSE_WAIT: 1434 so->so_error = ECONNRESET; 1435 close: 1436 tp->t_state = TCPS_CLOSED; 1437 tcpstat.tcps_drops++; 1438 tp = tcp_close(tp); 1439 break; 1440 1441 case TCPS_CLOSING: 1442 case TCPS_LAST_ACK: 1443 tp = tcp_close(tp); 1444 break; 1445 1446 case TCPS_TIME_WAIT: 1447 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1448 ("timewait")); 1449 break; 1450 } 1451 } 1452 goto drop; 1453 } 1454 1455 /* 1456 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1457 * and it's less than ts_recent, drop it. 1458 */ 1459 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1460 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1461 1462 /* Check to see if ts_recent is over 24 days old. */ 1463 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1464 /* 1465 * Invalidate ts_recent. If this segment updates 1466 * ts_recent, the age will be reset later and ts_recent 1467 * will get a valid value. If it does not, setting 1468 * ts_recent to zero will at least satisfy the 1469 * requirement that zero be placed in the timestamp 1470 * echo reply when ts_recent isn't valid. The 1471 * age isn't reset until we get a valid ts_recent 1472 * because we don't want out-of-order segments to be 1473 * dropped when ts_recent is old. 1474 */ 1475 tp->ts_recent = 0; 1476 } else { 1477 tcpstat.tcps_rcvduppack++; 1478 tcpstat.tcps_rcvdupbyte += tlen; 1479 tcpstat.tcps_pawsdrop++; 1480 if (tlen) 1481 goto dropafterack; 1482 goto drop; 1483 } 1484 } 1485 1486 /* 1487 * T/TCP mechanism 1488 * If T/TCP was negotiated and the segment doesn't have CC, 1489 * or if its CC is wrong then drop the segment. 1490 * RST segments do not have to comply with this. 1491 */ 1492 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1493 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1494 goto dropafterack; 1495 1496 /* 1497 * In the SYN-RECEIVED state, validate that the packet belongs to 1498 * this connection before trimming the data to fit the receive 1499 * window. Check the sequence number versus IRS since we know 1500 * the sequence numbers haven't wrapped. This is a partial fix 1501 * for the "LAND" DoS attack. 1502 */ 1503 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1504 rstreason = BANDLIM_RST_OPENPORT; 1505 goto dropwithreset; 1506 } 1507 1508 todrop = tp->rcv_nxt - th->th_seq; 1509 if (todrop > 0) { 1510 if (thflags & TH_SYN) { 1511 thflags &= ~TH_SYN; 1512 th->th_seq++; 1513 if (th->th_urp > 1) 1514 th->th_urp--; 1515 else 1516 thflags &= ~TH_URG; 1517 todrop--; 1518 } 1519 /* 1520 * Following if statement from Stevens, vol. 2, p. 960. 1521 */ 1522 if (todrop > tlen 1523 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1524 /* 1525 * Any valid FIN must be to the left of the window. 1526 * At this point the FIN must be a duplicate or out 1527 * of sequence; drop it. 1528 */ 1529 thflags &= ~TH_FIN; 1530 1531 /* 1532 * Send an ACK to resynchronize and drop any data. 1533 * But keep on processing for RST or ACK. 1534 */ 1535 tp->t_flags |= TF_ACKNOW; 1536 todrop = tlen; 1537 tcpstat.tcps_rcvduppack++; 1538 tcpstat.tcps_rcvdupbyte += todrop; 1539 } else { 1540 tcpstat.tcps_rcvpartduppack++; 1541 tcpstat.tcps_rcvpartdupbyte += todrop; 1542 } 1543 drop_hdrlen += todrop; /* drop from the top afterwards */ 1544 th->th_seq += todrop; 1545 tlen -= todrop; 1546 if (th->th_urp > todrop) 1547 th->th_urp -= todrop; 1548 else { 1549 thflags &= ~TH_URG; 1550 th->th_urp = 0; 1551 } 1552 } 1553 1554 /* 1555 * If new data are received on a connection after the 1556 * user processes are gone, then RST the other end. 1557 */ 1558 if ((so->so_state & SS_NOFDREF) && 1559 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1560 tp = tcp_close(tp); 1561 tcpstat.tcps_rcvafterclose++; 1562 rstreason = BANDLIM_UNLIMITED; 1563 goto dropwithreset; 1564 } 1565 1566 /* 1567 * If segment ends after window, drop trailing data 1568 * (and PUSH and FIN); if nothing left, just ACK. 1569 */ 1570 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1571 if (todrop > 0) { 1572 tcpstat.tcps_rcvpackafterwin++; 1573 if (todrop >= tlen) { 1574 tcpstat.tcps_rcvbyteafterwin += tlen; 1575 /* 1576 * If a new connection request is received 1577 * while in TIME_WAIT, drop the old connection 1578 * and start over if the sequence numbers 1579 * are above the previous ones. 1580 */ 1581 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1582 if (thflags & TH_SYN && 1583 tp->t_state == TCPS_TIME_WAIT && 1584 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1585 tp = tcp_close(tp); 1586 goto findpcb; 1587 } 1588 /* 1589 * If window is closed can only take segments at 1590 * window edge, and have to drop data and PUSH from 1591 * incoming segments. Continue processing, but 1592 * remember to ack. Otherwise, drop segment 1593 * and ack. 1594 */ 1595 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1596 tp->t_flags |= TF_ACKNOW; 1597 tcpstat.tcps_rcvwinprobe++; 1598 } else 1599 goto dropafterack; 1600 } else 1601 tcpstat.tcps_rcvbyteafterwin += todrop; 1602 m_adj(m, -todrop); 1603 tlen -= todrop; 1604 thflags &= ~(TH_PUSH|TH_FIN); 1605 } 1606 1607 /* 1608 * If last ACK falls within this segment's sequence numbers, 1609 * record its timestamp. 1610 * NOTE that the test is modified according to the latest 1611 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1612 */ 1613 if ((to.to_flags & TOF_TS) != 0 && 1614 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1615 tp->ts_recent_age = ticks; 1616 tp->ts_recent = to.to_tsval; 1617 } 1618 1619 /* 1620 * If a SYN is in the window, then this is an 1621 * error and we send an RST and drop the connection. 1622 */ 1623 if (thflags & TH_SYN) { 1624 tp = tcp_drop(tp, ECONNRESET); 1625 rstreason = BANDLIM_UNLIMITED; 1626 goto dropwithreset; 1627 } 1628 1629 /* 1630 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1631 * flag is on (half-synchronized state), then queue data for 1632 * later processing; else drop segment and return. 1633 */ 1634 if ((thflags & TH_ACK) == 0) { 1635 if (tp->t_state == TCPS_SYN_RECEIVED || 1636 (tp->t_flags & TF_NEEDSYN)) 1637 goto step6; 1638 else 1639 goto drop; 1640 } 1641 1642 /* 1643 * Ack processing. 1644 */ 1645 switch (tp->t_state) { 1646 1647 /* 1648 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1649 * ESTABLISHED state and continue processing. 1650 * The ACK was checked above. 1651 */ 1652 case TCPS_SYN_RECEIVED: 1653 1654 tcpstat.tcps_connects++; 1655 soisconnected(so); 1656 /* Do window scaling? */ 1657 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1658 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1659 tp->snd_scale = tp->requested_s_scale; 1660 tp->rcv_scale = tp->request_r_scale; 1661 } 1662 /* 1663 * Upon successful completion of 3-way handshake, 1664 * update cache.CC if it was undefined, pass any queued 1665 * data to the user, and advance state appropriately. 1666 */ 1667 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1668 taop->tao_cc == 0) 1669 taop->tao_cc = tp->cc_recv; 1670 1671 /* 1672 * Make transitions: 1673 * SYN-RECEIVED -> ESTABLISHED 1674 * SYN-RECEIVED* -> FIN-WAIT-1 1675 */ 1676 tp->t_starttime = ticks; 1677 if (tp->t_flags & TF_NEEDFIN) { 1678 tp->t_state = TCPS_FIN_WAIT_1; 1679 tp->t_flags &= ~TF_NEEDFIN; 1680 } else { 1681 tp->t_state = TCPS_ESTABLISHED; 1682 callout_reset(tp->tt_keep, tcp_keepidle, 1683 tcp_timer_keep, tp); 1684 } 1685 /* 1686 * If segment contains data or ACK, will call tcp_reass() 1687 * later; if not, do so now to pass queued data to user. 1688 */ 1689 if (tlen == 0 && (thflags & TH_FIN) == 0) 1690 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1691 (struct mbuf *)0); 1692 tp->snd_wl1 = th->th_seq - 1; 1693 /* FALLTHROUGH */ 1694 1695 /* 1696 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1697 * ACKs. If the ack is in the range 1698 * tp->snd_una < th->th_ack <= tp->snd_max 1699 * then advance tp->snd_una to th->th_ack and drop 1700 * data from the retransmission queue. If this ACK reflects 1701 * more up to date window information we update our window information. 1702 */ 1703 case TCPS_ESTABLISHED: 1704 case TCPS_FIN_WAIT_1: 1705 case TCPS_FIN_WAIT_2: 1706 case TCPS_CLOSE_WAIT: 1707 case TCPS_CLOSING: 1708 case TCPS_LAST_ACK: 1709 case TCPS_TIME_WAIT: 1710 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1711 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1712 if (tlen == 0 && tiwin == tp->snd_wnd) { 1713 tcpstat.tcps_rcvdupack++; 1714 /* 1715 * If we have outstanding data (other than 1716 * a window probe), this is a completely 1717 * duplicate ack (ie, window info didn't 1718 * change), the ack is the biggest we've 1719 * seen and we've seen exactly our rexmt 1720 * threshhold of them, assume a packet 1721 * has been dropped and retransmit it. 1722 * Kludge snd_nxt & the congestion 1723 * window so we send only this one 1724 * packet. 1725 * 1726 * We know we're losing at the current 1727 * window size so do congestion avoidance 1728 * (set ssthresh to half the current window 1729 * and pull our congestion window back to 1730 * the new ssthresh). 1731 * 1732 * Dup acks mean that packets have left the 1733 * network (they're now cached at the receiver) 1734 * so bump cwnd by the amount in the receiver 1735 * to keep a constant cwnd packets in the 1736 * network. 1737 */ 1738 if (!callout_active(tp->tt_rexmt) || 1739 th->th_ack != tp->snd_una) 1740 tp->t_dupacks = 0; 1741 else if (++tp->t_dupacks > tcprexmtthresh || 1742 (tcp_do_newreno && 1743 IN_FASTRECOVERY(tp))) { 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, 1752 tp->snd_recover)) { 1753 tp->t_dupacks = 0; 1754 break; 1755 } 1756 win = min(tp->snd_wnd, tp->snd_cwnd) / 1757 2 / tp->t_maxseg; 1758 if (win < 2) 1759 win = 2; 1760 tp->snd_ssthresh = win * tp->t_maxseg; 1761 ENTER_FASTRECOVERY(tp); 1762 tp->snd_recover = tp->snd_max; 1763 callout_stop(tp->tt_rexmt); 1764 tp->t_rtttime = 0; 1765 tp->snd_nxt = th->th_ack; 1766 tp->snd_cwnd = tp->t_maxseg; 1767 (void) tcp_output(tp); 1768 KASSERT(tp->snd_limited <= 2, 1769 ("tp->snd_limited too big")); 1770 tp->snd_cwnd = tp->snd_ssthresh + 1771 tp->t_maxseg * 1772 (tp->t_dupacks - tp->snd_limited); 1773 if (SEQ_GT(onxt, tp->snd_nxt)) 1774 tp->snd_nxt = onxt; 1775 goto drop; 1776 } else if (tcp_do_rfc3042) { 1777 u_long oldcwnd = tp->snd_cwnd; 1778 tcp_seq oldsndmax = tp->snd_max; 1779 u_int sent; 1780 KASSERT(tp->t_dupacks == 1 || 1781 tp->t_dupacks == 2, 1782 ("dupacks not 1 or 2")); 1783 if (tp->t_dupacks == 1) { 1784 tp->snd_limited = 0; 1785 tp->snd_cwnd += tp->t_maxseg; 1786 } else { 1787 tp->snd_cwnd += 1788 tp->t_maxseg * 2; 1789 } 1790 (void) tcp_output(tp); 1791 sent = tp->snd_max - oldsndmax; 1792 if (sent > tp->t_maxseg) { 1793 KASSERT(tp->snd_limited == 0 && 1794 tp->t_dupacks == 2, 1795 ("sent too much")); 1796 tp->snd_limited = 2; 1797 } else if (sent > 0) 1798 ++tp->snd_limited; 1799 tp->snd_cwnd = oldcwnd; 1800 goto drop; 1801 } 1802 } else 1803 tp->t_dupacks = 0; 1804 break; 1805 } 1806 1807 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1808 1809 /* 1810 * If the congestion window was inflated to account 1811 * for the other side's cached packets, retract it. 1812 */ 1813 if (tcp_do_newreno) { 1814 if (IN_FASTRECOVERY(tp)) { 1815 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1816 tcp_newreno_partial_ack(tp, th); 1817 } else { 1818 /* 1819 * Window inflation should have left us 1820 * with approximately snd_ssthresh 1821 * outstanding data. 1822 * But in case we would be inclined to 1823 * send a burst, better to do it via 1824 * the slow start mechanism. 1825 */ 1826 if (SEQ_GT(th->th_ack + 1827 tp->snd_ssthresh, 1828 tp->snd_max)) 1829 tp->snd_cwnd = tp->snd_max - 1830 th->th_ack + 1831 tp->t_maxseg; 1832 else 1833 tp->snd_cwnd = tp->snd_ssthresh; 1834 } 1835 } 1836 } else { 1837 if (tp->t_dupacks >= tcprexmtthresh && 1838 tp->snd_cwnd > tp->snd_ssthresh) 1839 tp->snd_cwnd = tp->snd_ssthresh; 1840 } 1841 tp->t_dupacks = 0; 1842 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1843 tcpstat.tcps_rcvacktoomuch++; 1844 goto dropafterack; 1845 } 1846 /* 1847 * If we reach this point, ACK is not a duplicate, 1848 * i.e., it ACKs something we sent. 1849 */ 1850 if (tp->t_flags & TF_NEEDSYN) { 1851 /* 1852 * T/TCP: Connection was half-synchronized, and our 1853 * SYN has been ACK'd (so connection is now fully 1854 * synchronized). Go to non-starred state, 1855 * increment snd_una for ACK of SYN, and check if 1856 * we can do window scaling. 1857 */ 1858 tp->t_flags &= ~TF_NEEDSYN; 1859 tp->snd_una++; 1860 /* Do window scaling? */ 1861 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1862 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1863 tp->snd_scale = tp->requested_s_scale; 1864 tp->rcv_scale = tp->request_r_scale; 1865 } 1866 } 1867 1868 process_ACK: 1869 acked = th->th_ack - tp->snd_una; 1870 tcpstat.tcps_rcvackpack++; 1871 tcpstat.tcps_rcvackbyte += acked; 1872 1873 /* 1874 * If we just performed our first retransmit, and the ACK 1875 * arrives within our recovery window, then it was a mistake 1876 * to do the retransmit in the first place. Recover our 1877 * original cwnd and ssthresh, and proceed to transmit where 1878 * we left off. 1879 */ 1880 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1881 ++tcpstat.tcps_sndrexmitbad; 1882 tp->snd_cwnd = tp->snd_cwnd_prev; 1883 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1884 tp->snd_recover = tp->snd_recover_prev; 1885 if (tp->t_flags & TF_WASFRECOVERY) 1886 ENTER_FASTRECOVERY(tp); 1887 tp->snd_nxt = tp->snd_max; 1888 tp->t_badrxtwin = 0; /* XXX probably not required */ 1889 } 1890 1891 /* 1892 * If we have a timestamp reply, update smoothed 1893 * round trip time. If no timestamp is present but 1894 * transmit timer is running and timed sequence 1895 * number was acked, update smoothed round trip time. 1896 * Since we now have an rtt measurement, cancel the 1897 * timer backoff (cf., Phil Karn's retransmit alg.). 1898 * Recompute the initial retransmit timer. 1899 * 1900 * Some boxes send broken timestamp replies 1901 * during the SYN+ACK phase, ignore 1902 * timestamps of 0 or we could calculate a 1903 * huge RTT and blow up the retransmit timer. 1904 */ 1905 if ((to.to_flags & TOF_TS) != 0 && 1906 to.to_tsecr) { 1907 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1908 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1909 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1910 } 1911 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1912 1913 /* 1914 * If all outstanding data is acked, stop retransmit 1915 * timer and remember to restart (more output or persist). 1916 * If there is more data to be acked, restart retransmit 1917 * timer, using current (possibly backed-off) value. 1918 */ 1919 if (th->th_ack == tp->snd_max) { 1920 callout_stop(tp->tt_rexmt); 1921 needoutput = 1; 1922 } else if (!callout_active(tp->tt_persist)) 1923 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1924 tcp_timer_rexmt, tp); 1925 1926 /* 1927 * If no data (only SYN) was ACK'd, 1928 * skip rest of ACK processing. 1929 */ 1930 if (acked == 0) 1931 goto step6; 1932 1933 /* 1934 * When new data is acked, open the congestion window. 1935 * If the window gives us less than ssthresh packets 1936 * in flight, open exponentially (maxseg per packet). 1937 * Otherwise open linearly: maxseg per window 1938 * (maxseg^2 / cwnd per packet). 1939 */ 1940 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 1941 register u_int cw = tp->snd_cwnd; 1942 register u_int incr = tp->t_maxseg; 1943 if (cw > tp->snd_ssthresh) 1944 incr = incr * incr / cw; 1945 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 1946 } 1947 if (acked > so->so_snd.sb_cc) { 1948 tp->snd_wnd -= so->so_snd.sb_cc; 1949 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1950 ourfinisacked = 1; 1951 } else { 1952 sbdrop(&so->so_snd, acked); 1953 tp->snd_wnd -= acked; 1954 ourfinisacked = 0; 1955 } 1956 sowwakeup(so); 1957 /* detect una wraparound */ 1958 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 1959 SEQ_GT(tp->snd_una, tp->snd_recover) && 1960 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1961 tp->snd_recover = th->th_ack - 1; 1962 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 1963 SEQ_GEQ(th->th_ack, tp->snd_recover)) 1964 EXIT_FASTRECOVERY(tp); 1965 tp->snd_una = th->th_ack; 1966 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1967 tp->snd_nxt = tp->snd_una; 1968 1969 switch (tp->t_state) { 1970 1971 /* 1972 * In FIN_WAIT_1 STATE in addition to the processing 1973 * for the ESTABLISHED state if our FIN is now acknowledged 1974 * then enter FIN_WAIT_2. 1975 */ 1976 case TCPS_FIN_WAIT_1: 1977 if (ourfinisacked) { 1978 /* 1979 * If we can't receive any more 1980 * data, then closing user can proceed. 1981 * Starting the timer is contrary to the 1982 * specification, but if we don't get a FIN 1983 * we'll hang forever. 1984 */ 1985 /* XXXjl 1986 * we should release the tp also, and use a 1987 * compressed state. 1988 */ 1989 if (so->so_state & SS_CANTRCVMORE) { 1990 soisdisconnected(so); 1991 callout_reset(tp->tt_2msl, tcp_maxidle, 1992 tcp_timer_2msl, tp); 1993 } 1994 tp->t_state = TCPS_FIN_WAIT_2; 1995 } 1996 break; 1997 1998 /* 1999 * In CLOSING STATE in addition to the processing for 2000 * the ESTABLISHED state if the ACK acknowledges our FIN 2001 * then enter the TIME-WAIT state, otherwise ignore 2002 * the segment. 2003 */ 2004 case TCPS_CLOSING: 2005 if (ourfinisacked) { 2006 KASSERT(headlocked, ("headlocked")); 2007 tcp_twstart(tp); 2008 INP_INFO_WUNLOCK(&tcbinfo); 2009 m_freem(m); 2010 return; 2011 } 2012 break; 2013 2014 /* 2015 * In LAST_ACK, we may still be waiting for data to drain 2016 * and/or to be acked, as well as for the ack of our FIN. 2017 * If our FIN is now acknowledged, delete the TCB, 2018 * enter the closed state and return. 2019 */ 2020 case TCPS_LAST_ACK: 2021 if (ourfinisacked) { 2022 tp = tcp_close(tp); 2023 goto drop; 2024 } 2025 break; 2026 2027 /* 2028 * In TIME_WAIT state the only thing that should arrive 2029 * is a retransmission of the remote FIN. Acknowledge 2030 * it and restart the finack timer. 2031 */ 2032 case TCPS_TIME_WAIT: 2033 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2034 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2035 tcp_timer_2msl, tp); 2036 goto dropafterack; 2037 } 2038 } 2039 2040 step6: 2041 /* 2042 * Update window information. 2043 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2044 */ 2045 if ((thflags & TH_ACK) && 2046 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2047 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2048 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2049 /* keep track of pure window updates */ 2050 if (tlen == 0 && 2051 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2052 tcpstat.tcps_rcvwinupd++; 2053 tp->snd_wnd = tiwin; 2054 tp->snd_wl1 = th->th_seq; 2055 tp->snd_wl2 = th->th_ack; 2056 if (tp->snd_wnd > tp->max_sndwnd) 2057 tp->max_sndwnd = tp->snd_wnd; 2058 needoutput = 1; 2059 } 2060 2061 /* 2062 * Process segments with URG. 2063 */ 2064 if ((thflags & TH_URG) && th->th_urp && 2065 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2066 /* 2067 * This is a kludge, but if we receive and accept 2068 * random urgent pointers, we'll crash in 2069 * soreceive. It's hard to imagine someone 2070 * actually wanting to send this much urgent data. 2071 */ 2072 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2073 th->th_urp = 0; /* XXX */ 2074 thflags &= ~TH_URG; /* XXX */ 2075 goto dodata; /* XXX */ 2076 } 2077 /* 2078 * If this segment advances the known urgent pointer, 2079 * then mark the data stream. This should not happen 2080 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2081 * a FIN has been received from the remote side. 2082 * In these states we ignore the URG. 2083 * 2084 * According to RFC961 (Assigned Protocols), 2085 * the urgent pointer points to the last octet 2086 * of urgent data. We continue, however, 2087 * to consider it to indicate the first octet 2088 * of data past the urgent section as the original 2089 * spec states (in one of two places). 2090 */ 2091 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2092 tp->rcv_up = th->th_seq + th->th_urp; 2093 so->so_oobmark = so->so_rcv.sb_cc + 2094 (tp->rcv_up - tp->rcv_nxt) - 1; 2095 if (so->so_oobmark == 0) 2096 so->so_state |= SS_RCVATMARK; 2097 sohasoutofband(so); 2098 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2099 } 2100 /* 2101 * Remove out of band data so doesn't get presented to user. 2102 * This can happen independent of advancing the URG pointer, 2103 * but if two URG's are pending at once, some out-of-band 2104 * data may creep in... ick. 2105 */ 2106 if (th->th_urp <= (u_long)tlen && 2107 !(so->so_options & SO_OOBINLINE)) { 2108 /* hdr drop is delayed */ 2109 tcp_pulloutofband(so, th, m, drop_hdrlen); 2110 } 2111 } else { 2112 /* 2113 * If no out of band data is expected, 2114 * pull receive urgent pointer along 2115 * with the receive window. 2116 */ 2117 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2118 tp->rcv_up = tp->rcv_nxt; 2119 } 2120 dodata: /* XXX */ 2121 KASSERT(headlocked, ("headlocked")); 2122 /* 2123 * Process the segment text, merging it into the TCP sequencing queue, 2124 * and arranging for acknowledgment of receipt if necessary. 2125 * This process logically involves adjusting tp->rcv_wnd as data 2126 * is presented to the user (this happens in tcp_usrreq.c, 2127 * case PRU_RCVD). If a FIN has already been received on this 2128 * connection then we just ignore the text. 2129 */ 2130 if ((tlen || (thflags & TH_FIN)) && 2131 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2132 m_adj(m, drop_hdrlen); /* delayed header drop */ 2133 /* 2134 * Insert segment which includes th into TCP reassembly queue 2135 * with control block tp. Set thflags to whether reassembly now 2136 * includes a segment with FIN. This handles the common case 2137 * inline (segment is the next to be received on an established 2138 * connection, and the queue is empty), avoiding linkage into 2139 * and removal from the queue and repetition of various 2140 * conversions. 2141 * Set DELACK for segments received in order, but ack 2142 * immediately when segments are out of order (so 2143 * fast retransmit can work). 2144 */ 2145 if (th->th_seq == tp->rcv_nxt && 2146 LIST_EMPTY(&tp->t_segq) && 2147 TCPS_HAVEESTABLISHED(tp->t_state)) { 2148 if (DELAY_ACK(tp)) 2149 tp->t_flags |= TF_DELACK; 2150 else 2151 tp->t_flags |= TF_ACKNOW; 2152 tp->rcv_nxt += tlen; 2153 thflags = th->th_flags & TH_FIN; 2154 tcpstat.tcps_rcvpack++; 2155 tcpstat.tcps_rcvbyte += tlen; 2156 ND6_HINT(tp); 2157 if (so->so_state & SS_CANTRCVMORE) 2158 m_freem(m); 2159 else 2160 sbappend(&so->so_rcv, m); 2161 sorwakeup(so); 2162 } else { 2163 thflags = tcp_reass(tp, th, &tlen, m); 2164 tp->t_flags |= TF_ACKNOW; 2165 } 2166 2167 /* 2168 * Note the amount of data that peer has sent into 2169 * our window, in order to estimate the sender's 2170 * buffer size. 2171 */ 2172 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2173 } else { 2174 m_freem(m); 2175 thflags &= ~TH_FIN; 2176 } 2177 2178 /* 2179 * If FIN is received ACK the FIN and let the user know 2180 * that the connection is closing. 2181 */ 2182 if (thflags & TH_FIN) { 2183 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2184 socantrcvmore(so); 2185 /* 2186 * If connection is half-synchronized 2187 * (ie NEEDSYN flag on) then delay ACK, 2188 * so it may be piggybacked when SYN is sent. 2189 * Otherwise, since we received a FIN then no 2190 * more input can be expected, send ACK now. 2191 */ 2192 if (tp->t_flags & TF_NEEDSYN) 2193 tp->t_flags |= TF_DELACK; 2194 else 2195 tp->t_flags |= TF_ACKNOW; 2196 tp->rcv_nxt++; 2197 } 2198 switch (tp->t_state) { 2199 2200 /* 2201 * In SYN_RECEIVED and ESTABLISHED STATES 2202 * enter the CLOSE_WAIT state. 2203 */ 2204 case TCPS_SYN_RECEIVED: 2205 tp->t_starttime = ticks; 2206 /*FALLTHROUGH*/ 2207 case TCPS_ESTABLISHED: 2208 tp->t_state = TCPS_CLOSE_WAIT; 2209 break; 2210 2211 /* 2212 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2213 * enter the CLOSING state. 2214 */ 2215 case TCPS_FIN_WAIT_1: 2216 tp->t_state = TCPS_CLOSING; 2217 break; 2218 2219 /* 2220 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2221 * starting the time-wait timer, turning off the other 2222 * standard timers. 2223 */ 2224 case TCPS_FIN_WAIT_2: 2225 KASSERT(headlocked == 1, ("headlocked should be 1")); 2226 tcp_twstart(tp); 2227 INP_INFO_WUNLOCK(&tcbinfo); 2228 return; 2229 2230 /* 2231 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2232 */ 2233 case TCPS_TIME_WAIT: 2234 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2235 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2236 tcp_timer_2msl, tp); 2237 break; 2238 } 2239 } 2240 INP_INFO_WUNLOCK(&tcbinfo); 2241 #ifdef TCPDEBUG 2242 if (so->so_options & SO_DEBUG) 2243 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2244 &tcp_savetcp, 0); 2245 #endif 2246 2247 /* 2248 * Return any desired output. 2249 */ 2250 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2251 (void) tcp_output(tp); 2252 2253 check_delack: 2254 if (tp->t_flags & TF_DELACK) { 2255 tp->t_flags &= ~TF_DELACK; 2256 callout_reset(tp->tt_delack, tcp_delacktime, 2257 tcp_timer_delack, tp); 2258 } 2259 INP_UNLOCK(inp); 2260 return; 2261 2262 dropafterack: 2263 /* 2264 * Generate an ACK dropping incoming segment if it occupies 2265 * sequence space, where the ACK reflects our state. 2266 * 2267 * We can now skip the test for the RST flag since all 2268 * paths to this code happen after packets containing 2269 * RST have been dropped. 2270 * 2271 * In the SYN-RECEIVED state, don't send an ACK unless the 2272 * segment we received passes the SYN-RECEIVED ACK test. 2273 * If it fails send a RST. This breaks the loop in the 2274 * "LAND" DoS attack, and also prevents an ACK storm 2275 * between two listening ports that have been sent forged 2276 * SYN segments, each with the source address of the other. 2277 */ 2278 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2279 (SEQ_GT(tp->snd_una, th->th_ack) || 2280 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2281 rstreason = BANDLIM_RST_OPENPORT; 2282 goto dropwithreset; 2283 } 2284 #ifdef TCPDEBUG 2285 if (so->so_options & SO_DEBUG) 2286 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2287 &tcp_savetcp, 0); 2288 #endif 2289 KASSERT(headlocked, ("headlocked should be 1")); 2290 INP_INFO_WUNLOCK(&tcbinfo); 2291 m_freem(m); 2292 tp->t_flags |= TF_ACKNOW; 2293 (void) tcp_output(tp); 2294 INP_UNLOCK(inp); 2295 return; 2296 2297 dropwithreset: 2298 /* 2299 * Generate a RST, dropping incoming segment. 2300 * Make ACK acceptable to originator of segment. 2301 * Don't bother to respond if destination was broadcast/multicast. 2302 */ 2303 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2304 goto drop; 2305 if (isipv6) { 2306 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2307 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2308 goto drop; 2309 } else { 2310 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2311 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2312 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2313 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2314 goto drop; 2315 } 2316 /* IPv6 anycast check is done at tcp6_input() */ 2317 2318 /* 2319 * Perform bandwidth limiting. 2320 */ 2321 if (badport_bandlim(rstreason) < 0) 2322 goto drop; 2323 2324 #ifdef TCPDEBUG 2325 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2326 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2327 &tcp_savetcp, 0); 2328 #endif 2329 2330 if (tp) 2331 INP_UNLOCK(inp); 2332 2333 if (thflags & TH_ACK) 2334 /* mtod() below is safe as long as hdr dropping is delayed */ 2335 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2336 TH_RST); 2337 else { 2338 if (thflags & TH_SYN) 2339 tlen++; 2340 /* mtod() below is safe as long as hdr dropping is delayed */ 2341 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2342 (tcp_seq)0, TH_RST|TH_ACK); 2343 } 2344 if (headlocked) 2345 INP_INFO_WUNLOCK(&tcbinfo); 2346 return; 2347 2348 drop: 2349 /* 2350 * Drop space held by incoming segment and return. 2351 */ 2352 #ifdef TCPDEBUG 2353 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2354 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2355 &tcp_savetcp, 0); 2356 #endif 2357 if (tp) 2358 INP_UNLOCK(inp); 2359 m_freem(m); 2360 if (headlocked) 2361 INP_INFO_WUNLOCK(&tcbinfo); 2362 return; 2363 } 2364 2365 /* 2366 * Parse TCP options and place in tcpopt. 2367 */ 2368 static void 2369 tcp_dooptions(to, cp, cnt, is_syn) 2370 struct tcpopt *to; 2371 u_char *cp; 2372 int cnt; 2373 int is_syn; 2374 { 2375 int opt, optlen; 2376 2377 to->to_flags = 0; 2378 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2379 opt = cp[0]; 2380 if (opt == TCPOPT_EOL) 2381 break; 2382 if (opt == TCPOPT_NOP) 2383 optlen = 1; 2384 else { 2385 if (cnt < 2) 2386 break; 2387 optlen = cp[1]; 2388 if (optlen < 2 || optlen > cnt) 2389 break; 2390 } 2391 switch (opt) { 2392 case TCPOPT_MAXSEG: 2393 if (optlen != TCPOLEN_MAXSEG) 2394 continue; 2395 if (!is_syn) 2396 continue; 2397 to->to_flags |= TOF_MSS; 2398 bcopy((char *)cp + 2, 2399 (char *)&to->to_mss, sizeof(to->to_mss)); 2400 to->to_mss = ntohs(to->to_mss); 2401 break; 2402 case TCPOPT_WINDOW: 2403 if (optlen != TCPOLEN_WINDOW) 2404 continue; 2405 if (! is_syn) 2406 continue; 2407 to->to_flags |= TOF_SCALE; 2408 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2409 break; 2410 case TCPOPT_TIMESTAMP: 2411 if (optlen != TCPOLEN_TIMESTAMP) 2412 continue; 2413 to->to_flags |= TOF_TS; 2414 bcopy((char *)cp + 2, 2415 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2416 to->to_tsval = ntohl(to->to_tsval); 2417 bcopy((char *)cp + 6, 2418 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2419 to->to_tsecr = ntohl(to->to_tsecr); 2420 break; 2421 case TCPOPT_CC: 2422 if (optlen != TCPOLEN_CC) 2423 continue; 2424 to->to_flags |= TOF_CC; 2425 bcopy((char *)cp + 2, 2426 (char *)&to->to_cc, sizeof(to->to_cc)); 2427 to->to_cc = ntohl(to->to_cc); 2428 break; 2429 case TCPOPT_CCNEW: 2430 if (optlen != TCPOLEN_CC) 2431 continue; 2432 if (!is_syn) 2433 continue; 2434 to->to_flags |= TOF_CCNEW; 2435 bcopy((char *)cp + 2, 2436 (char *)&to->to_cc, sizeof(to->to_cc)); 2437 to->to_cc = ntohl(to->to_cc); 2438 break; 2439 case TCPOPT_CCECHO: 2440 if (optlen != TCPOLEN_CC) 2441 continue; 2442 if (!is_syn) 2443 continue; 2444 to->to_flags |= TOF_CCECHO; 2445 bcopy((char *)cp + 2, 2446 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2447 to->to_ccecho = ntohl(to->to_ccecho); 2448 break; 2449 default: 2450 continue; 2451 } 2452 } 2453 } 2454 2455 /* 2456 * Pull out of band byte out of a segment so 2457 * it doesn't appear in the user's data queue. 2458 * It is still reflected in the segment length for 2459 * sequencing purposes. 2460 */ 2461 static void 2462 tcp_pulloutofband(so, th, m, off) 2463 struct socket *so; 2464 struct tcphdr *th; 2465 register struct mbuf *m; 2466 int off; /* delayed to be droped hdrlen */ 2467 { 2468 int cnt = off + th->th_urp - 1; 2469 2470 while (cnt >= 0) { 2471 if (m->m_len > cnt) { 2472 char *cp = mtod(m, caddr_t) + cnt; 2473 struct tcpcb *tp = sototcpcb(so); 2474 2475 tp->t_iobc = *cp; 2476 tp->t_oobflags |= TCPOOB_HAVEDATA; 2477 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2478 m->m_len--; 2479 if (m->m_flags & M_PKTHDR) 2480 m->m_pkthdr.len--; 2481 return; 2482 } 2483 cnt -= m->m_len; 2484 m = m->m_next; 2485 if (m == 0) 2486 break; 2487 } 2488 panic("tcp_pulloutofband"); 2489 } 2490 2491 /* 2492 * Collect new round-trip time estimate 2493 * and update averages and current timeout. 2494 */ 2495 static void 2496 tcp_xmit_timer(tp, rtt) 2497 register struct tcpcb *tp; 2498 int rtt; 2499 { 2500 register int delta; 2501 2502 tcpstat.tcps_rttupdated++; 2503 tp->t_rttupdated++; 2504 if (tp->t_srtt != 0) { 2505 /* 2506 * srtt is stored as fixed point with 5 bits after the 2507 * binary point (i.e., scaled by 8). The following magic 2508 * is equivalent to the smoothing algorithm in rfc793 with 2509 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2510 * point). Adjust rtt to origin 0. 2511 */ 2512 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2513 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2514 2515 if ((tp->t_srtt += delta) <= 0) 2516 tp->t_srtt = 1; 2517 2518 /* 2519 * We accumulate a smoothed rtt variance (actually, a 2520 * smoothed mean difference), then set the retransmit 2521 * timer to smoothed rtt + 4 times the smoothed variance. 2522 * rttvar is stored as fixed point with 4 bits after the 2523 * binary point (scaled by 16). The following is 2524 * equivalent to rfc793 smoothing with an alpha of .75 2525 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2526 * rfc793's wired-in beta. 2527 */ 2528 if (delta < 0) 2529 delta = -delta; 2530 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2531 if ((tp->t_rttvar += delta) <= 0) 2532 tp->t_rttvar = 1; 2533 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2534 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2535 } else { 2536 /* 2537 * No rtt measurement yet - use the unsmoothed rtt. 2538 * Set the variance to half the rtt (so our first 2539 * retransmit happens at 3*rtt). 2540 */ 2541 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2542 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2543 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2544 } 2545 tp->t_rtttime = 0; 2546 tp->t_rxtshift = 0; 2547 2548 /* 2549 * the retransmit should happen at rtt + 4 * rttvar. 2550 * Because of the way we do the smoothing, srtt and rttvar 2551 * will each average +1/2 tick of bias. When we compute 2552 * the retransmit timer, we want 1/2 tick of rounding and 2553 * 1 extra tick because of +-1/2 tick uncertainty in the 2554 * firing of the timer. The bias will give us exactly the 2555 * 1.5 tick we need. But, because the bias is 2556 * statistical, we have to test that we don't drop below 2557 * the minimum feasible timer (which is 2 ticks). 2558 */ 2559 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2560 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2561 2562 /* 2563 * We received an ack for a packet that wasn't retransmitted; 2564 * it is probably safe to discard any error indications we've 2565 * received recently. This isn't quite right, but close enough 2566 * for now (a route might have failed after we sent a segment, 2567 * and the return path might not be symmetrical). 2568 */ 2569 tp->t_softerror = 0; 2570 } 2571 2572 /* 2573 * Determine a reasonable value for maxseg size. 2574 * If the route is known, check route for mtu. 2575 * If none, use an mss that can be handled on the outgoing 2576 * interface without forcing IP to fragment; if bigger than 2577 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2578 * to utilize large mbufs. If no route is found, route has no mtu, 2579 * or the destination isn't local, use a default, hopefully conservative 2580 * size (usually 512 or the default IP max size, but no more than the mtu 2581 * of the interface), as we can't discover anything about intervening 2582 * gateways or networks. We also initialize the congestion/slow start 2583 * window to be a single segment if the destination isn't local. 2584 * While looking at the routing entry, we also initialize other path-dependent 2585 * parameters from pre-set or cached values in the routing entry. 2586 * 2587 * Also take into account the space needed for options that we 2588 * send regularly. Make maxseg shorter by that amount to assure 2589 * that we can send maxseg amount of data even when the options 2590 * are present. Store the upper limit of the length of options plus 2591 * data in maxopd. 2592 * 2593 * NOTE that this routine is only called when we process an incoming 2594 * segment, for outgoing segments only tcp_mssopt is called. 2595 * 2596 * In case of T/TCP, we call this routine during implicit connection 2597 * setup as well (offer = -1), to initialize maxseg from the cached 2598 * MSS of our peer. 2599 */ 2600 void 2601 tcp_mss(tp, offer) 2602 struct tcpcb *tp; 2603 int offer; 2604 { 2605 register struct rtentry *rt; 2606 struct ifnet *ifp; 2607 register int rtt, mss; 2608 u_long bufsize; 2609 struct inpcb *inp = tp->t_inpcb; 2610 struct socket *so; 2611 struct rmxp_tao *taop; 2612 int origoffer = offer; 2613 #ifdef INET6 2614 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2615 size_t min_protoh = isipv6 ? 2616 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2617 sizeof (struct tcpiphdr); 2618 #else 2619 const int isipv6 = 0; 2620 const size_t min_protoh = sizeof (struct tcpiphdr); 2621 #endif 2622 2623 if (isipv6) 2624 rt = tcp_rtlookup6(&inp->inp_inc); 2625 else 2626 rt = tcp_rtlookup(&inp->inp_inc); 2627 if (rt == NULL) { 2628 tp->t_maxopd = tp->t_maxseg = 2629 isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2630 return; 2631 } 2632 ifp = rt->rt_ifp; 2633 so = inp->inp_socket; 2634 2635 taop = rmx_taop(rt->rt_rmx); 2636 /* 2637 * Offer == -1 means that we didn't receive SYN yet, 2638 * use cached value in that case; 2639 */ 2640 if (offer == -1) 2641 offer = taop->tao_mssopt; 2642 /* 2643 * Offer == 0 means that there was no MSS on the SYN segment, 2644 * in this case we use tcp_mssdflt. 2645 */ 2646 if (offer == 0) 2647 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2648 else 2649 /* 2650 * Sanity check: make sure that maxopd will be large 2651 * enough to allow some data on segments even is the 2652 * all the option space is used (40bytes). Otherwise 2653 * funny things may happen in tcp_output. 2654 */ 2655 offer = max(offer, 64); 2656 taop->tao_mssopt = offer; 2657 2658 /* 2659 * While we're here, check if there's an initial rtt 2660 * or rttvar. Convert from the route-table units 2661 * to scaled multiples of the slow timeout timer. 2662 */ 2663 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2664 /* 2665 * XXX the lock bit for RTT indicates that the value 2666 * is also a minimum value; this is subject to time. 2667 */ 2668 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2669 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2670 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2671 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2672 tcpstat.tcps_usedrtt++; 2673 if (rt->rt_rmx.rmx_rttvar) { 2674 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2675 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2676 tcpstat.tcps_usedrttvar++; 2677 } else { 2678 /* default variation is +- 1 rtt */ 2679 tp->t_rttvar = 2680 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2681 } 2682 TCPT_RANGESET(tp->t_rxtcur, 2683 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2684 tp->t_rttmin, TCPTV_REXMTMAX); 2685 } 2686 /* 2687 * if there's an mtu associated with the route, use it 2688 * else, use the link mtu. 2689 */ 2690 if (rt->rt_rmx.rmx_mtu) 2691 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2692 else { 2693 if (isipv6) { 2694 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2695 min_protoh; 2696 if (!in6_localaddr(&inp->in6p_faddr)) 2697 mss = min(mss, tcp_v6mssdflt); 2698 } else { 2699 mss = ifp->if_mtu - min_protoh; 2700 if (!in_localaddr(inp->inp_faddr)) 2701 mss = min(mss, tcp_mssdflt); 2702 } 2703 } 2704 mss = min(mss, offer); 2705 /* 2706 * maxopd stores the maximum length of data AND options 2707 * in a segment; maxseg is the amount of data in a normal 2708 * segment. We need to store this value (maxopd) apart 2709 * from maxseg, because now every segment carries options 2710 * and thus we normally have somewhat less data in segments. 2711 */ 2712 tp->t_maxopd = mss; 2713 2714 /* 2715 * In case of T/TCP, origoffer==-1 indicates, that no segments 2716 * were received yet. In this case we just guess, otherwise 2717 * we do the same as before T/TCP. 2718 */ 2719 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2720 (origoffer == -1 || 2721 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2722 mss -= TCPOLEN_TSTAMP_APPA; 2723 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2724 (origoffer == -1 || 2725 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2726 mss -= TCPOLEN_CC_APPA; 2727 2728 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2729 if (mss > MCLBYTES) 2730 mss &= ~(MCLBYTES-1); 2731 #else 2732 if (mss > MCLBYTES) 2733 mss = mss / MCLBYTES * MCLBYTES; 2734 #endif 2735 /* 2736 * If there's a pipesize, change the socket buffer 2737 * to that size. Make the socket buffers an integral 2738 * number of mss units; if the mss is larger than 2739 * the socket buffer, decrease the mss. 2740 */ 2741 #ifdef RTV_SPIPE 2742 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2743 #endif 2744 bufsize = so->so_snd.sb_hiwat; 2745 if (bufsize < mss) 2746 mss = bufsize; 2747 else { 2748 bufsize = roundup(bufsize, mss); 2749 if (bufsize > sb_max) 2750 bufsize = sb_max; 2751 if (bufsize > so->so_snd.sb_hiwat) 2752 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2753 } 2754 tp->t_maxseg = mss; 2755 2756 #ifdef RTV_RPIPE 2757 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2758 #endif 2759 bufsize = so->so_rcv.sb_hiwat; 2760 if (bufsize > mss) { 2761 bufsize = roundup(bufsize, mss); 2762 if (bufsize > sb_max) 2763 bufsize = sb_max; 2764 if (bufsize > so->so_rcv.sb_hiwat) 2765 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2766 } 2767 2768 /* 2769 * Set the slow-start flight size depending on whether this 2770 * is a local network or not. 2771 */ 2772 if (tcp_do_rfc3390) 2773 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2774 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2775 (!isipv6 && in_localaddr(inp->inp_faddr))) 2776 tp->snd_cwnd = mss * ss_fltsz_local; 2777 else 2778 tp->snd_cwnd = mss * ss_fltsz; 2779 2780 if (rt->rt_rmx.rmx_ssthresh) { 2781 /* 2782 * There's some sort of gateway or interface 2783 * buffer limit on the path. Use this to set 2784 * the slow start threshhold, but set the 2785 * threshold to no less than 2*mss. 2786 */ 2787 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2788 tcpstat.tcps_usedssthresh++; 2789 } 2790 } 2791 2792 /* 2793 * Determine the MSS option to send on an outgoing SYN. 2794 */ 2795 int 2796 tcp_mssopt(tp) 2797 struct tcpcb *tp; 2798 { 2799 struct rtentry *rt; 2800 #ifdef INET6 2801 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2802 size_t min_protoh = isipv6 ? 2803 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2804 sizeof (struct tcpiphdr); 2805 #else 2806 const int isipv6 = 0; 2807 const size_t min_protoh = sizeof (struct tcpiphdr); 2808 #endif 2809 2810 if (isipv6) 2811 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2812 else 2813 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2814 if (rt == NULL) 2815 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2816 2817 return (rt->rt_ifp->if_mtu - min_protoh); 2818 } 2819 2820 2821 /* 2822 * On a partial ack arrives, force the retransmission of the 2823 * next unacknowledged segment. Do not clear tp->t_dupacks. 2824 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2825 * be started again. 2826 */ 2827 static void 2828 tcp_newreno_partial_ack(tp, th) 2829 struct tcpcb *tp; 2830 struct tcphdr *th; 2831 { 2832 tcp_seq onxt = tp->snd_nxt; 2833 u_long ocwnd = tp->snd_cwnd; 2834 2835 callout_stop(tp->tt_rexmt); 2836 tp->t_rtttime = 0; 2837 tp->snd_nxt = th->th_ack; 2838 /* 2839 * Set snd_cwnd to one segment beyond acknowledged offset. 2840 * (tp->snd_una has not yet been updated when this function is called.) 2841 */ 2842 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2843 tp->t_flags |= TF_ACKNOW; 2844 (void) tcp_output(tp); 2845 tp->snd_cwnd = ocwnd; 2846 if (SEQ_GT(onxt, tp->snd_nxt)) 2847 tp->snd_nxt = onxt; 2848 /* 2849 * Partial window deflation. Relies on fact that tp->snd_una 2850 * not updated yet. 2851 */ 2852 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2853 } 2854 2855 /* 2856 * Returns 1 if the TIME_WAIT state was killed and we should start over, 2857 * looking for a pcb in the listen state. Returns 0 otherwise. 2858 */ 2859 static int 2860 tcp_timewait(tw, to, th, m, tlen) 2861 struct tcptw *tw; 2862 struct tcpopt *to; 2863 struct tcphdr *th; 2864 struct mbuf *m; 2865 int tlen; 2866 { 2867 int thflags; 2868 tcp_seq seq; 2869 #ifdef INET6 2870 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 2871 #else 2872 const int isipv6 = 0; 2873 #endif 2874 2875 thflags = th->th_flags; 2876 2877 /* 2878 * NOTE: for FIN_WAIT_2 (to be added later), 2879 * must validate sequence number before accepting RST 2880 */ 2881 2882 /* 2883 * If the segment contains RST: 2884 * Drop the segment - see Stevens, vol. 2, p. 964 and 2885 * RFC 1337. 2886 */ 2887 if (thflags & TH_RST) 2888 goto drop; 2889 2890 /* 2891 * If segment contains a SYN and CC [not CC.NEW] option: 2892 * if connection duration > MSL, drop packet and send RST; 2893 * 2894 * if SEG.CC > CCrecv then is new SYN. 2895 * Complete close and delete TCPCB. Then reprocess 2896 * segment, hoping to find new TCPCB in LISTEN state; 2897 * 2898 * else must be old SYN; drop it. 2899 * else do normal processing. 2900 */ 2901 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 2902 if ((ticks - tw->t_starttime) > tcp_msl) 2903 goto reset; 2904 if (CC_GT(to->to_cc, tw->cc_recv)) { 2905 (void) tcp_twclose(tw, 0); 2906 return (1); 2907 } 2908 goto drop; 2909 } 2910 2911 #if 0 2912 /* PAWS not needed at the moment */ 2913 /* 2914 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2915 * and it's less than ts_recent, drop it. 2916 */ 2917 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2918 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2919 if ((thflags & TH_ACK) == 0) 2920 goto drop; 2921 goto ack; 2922 } 2923 /* 2924 * ts_recent is never updated because we never accept new segments. 2925 */ 2926 #endif 2927 2928 /* 2929 * If a new connection request is received 2930 * while in TIME_WAIT, drop the old connection 2931 * and start over if the sequence numbers 2932 * are above the previous ones. 2933 */ 2934 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 2935 (void) tcp_twclose(tw, 0); 2936 return (1); 2937 } 2938 2939 /* 2940 * Drop the the segment if it does not contain an ACK. 2941 */ 2942 if ((thflags & TH_ACK) == 0) 2943 goto drop; 2944 2945 /* 2946 * Reset the 2MSL timer if this is a duplicate FIN. 2947 */ 2948 if (thflags & TH_FIN) { 2949 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 2950 if (seq + 1 == tw->rcv_nxt) 2951 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 2952 } 2953 2954 /* 2955 * Acknowledge the segment if it has data or is not a duplicate ACK. 2956 */ 2957 if (thflags != TH_ACK || tlen != 0 || 2958 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 2959 tcp_twrespond(tw, NULL, m, TH_ACK); 2960 goto drop; 2961 2962 reset: 2963 /* 2964 * Generate a RST, dropping incoming segment. 2965 * Make ACK acceptable to originator of segment. 2966 * Don't bother to respond if destination was broadcast/multicast. 2967 */ 2968 if (m->m_flags & (M_BCAST|M_MCAST)) 2969 goto drop; 2970 if (isipv6) { 2971 struct ip6_hdr *ip6; 2972 2973 /* IPv6 anycast check is done at tcp6_input() */ 2974 ip6 = mtod(m, struct ip6_hdr *); 2975 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2976 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2977 goto drop; 2978 } else { 2979 struct ip *ip; 2980 2981 ip = mtod(m, struct ip *); 2982 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2983 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2984 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2985 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2986 goto drop; 2987 } 2988 if (thflags & TH_ACK) { 2989 tcp_respond(NULL, 2990 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 2991 } else { 2992 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 2993 tcp_respond(NULL, 2994 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 2995 } 2996 INP_UNLOCK(tw->tw_inpcb); 2997 return (0); 2998 2999 drop: 3000 INP_UNLOCK(tw->tw_inpcb); 3001 m_freem(m); 3002 return (0); 3003 } 3004