1 /* 2 * Copyright (c) 2002-2004 Jeffrey Hsu. All rights reserved. 3 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 4 * The Regents of the University of California. All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 1. Redistributions of source code must retain the above copyright 10 * notice, this list of conditions and the following disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 3. All advertising materials mentioning features or use of this software 15 * must display the following acknowledgement: 16 * This product includes software developed by the University of 17 * California, Berkeley and its contributors. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 35 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $ 36 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.27 2004/05/20 04:28:52 hsu Exp $ 37 */ 38 39 #include "opt_ipfw.h" /* for ipfw_fwd */ 40 #include "opt_inet6.h" 41 #include "opt_ipsec.h" 42 #include "opt_tcpdebug.h" 43 #include "opt_tcp_input.h" 44 45 #include <sys/param.h> 46 #include <sys/systm.h> 47 #include <sys/kernel.h> 48 #include <sys/sysctl.h> 49 #include <sys/malloc.h> 50 #include <sys/mbuf.h> 51 #include <sys/proc.h> /* for proc0 declaration */ 52 #include <sys/protosw.h> 53 #include <sys/socket.h> 54 #include <sys/socketvar.h> 55 #include <sys/syslog.h> 56 #include <sys/in_cksum.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_systm.h> 65 #include <netinet/ip.h> 66 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 67 #include <netinet/in_var.h> 68 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 69 #include <netinet/in_pcb.h> 70 #include <netinet/ip_var.h> 71 #include <netinet/ip6.h> 72 #include <netinet/icmp6.h> 73 #include <netinet6/nd6.h> 74 #include <netinet6/ip6_var.h> 75 #include <netinet6/in6_pcb.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 86 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 87 struct tcphdr tcp_savetcp; 88 #endif /* TCPDEBUG */ 89 90 #ifdef FAST_IPSEC 91 #include <netipsec/ipsec.h> 92 #include <netipsec/ipsec6.h> 93 #endif 94 95 #ifdef IPSEC 96 #include <netinet6/ipsec.h> 97 #include <netinet6/ipsec6.h> 98 #include <netproto/key/key.h> 99 #endif /*IPSEC*/ 100 101 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 102 103 static const int tcprexmtthresh = 3; 104 tcp_cc tcp_ccgen; 105 static int log_in_vain = 0; 106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 107 &log_in_vain, 0, "Log all incoming TCP connections"); 108 109 static int blackhole = 0; 110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 111 &blackhole, 0, "Do not send RST when dropping refused connections"); 112 113 int tcp_delack_enabled = 1; 114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 115 &tcp_delack_enabled, 0, 116 "Delay ACK to try and piggyback it onto a data packet"); 117 118 #ifdef TCP_DROP_SYNFIN 119 static int drop_synfin = 0; 120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 121 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 122 #endif 123 124 static int tcp_do_limitedtransmit = 1; 125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW, 126 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)"); 127 128 static int tcp_do_early_retransmit = 0; 129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW, 130 &tcp_do_early_retransmit, 0, "Early retransmit"); 131 132 static int tcp_do_rfc3390 = 1; 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 static int tcp_do_eifel_detect = 1; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW, 139 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)"); 140 141 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 142 "TCP Segment Reassembly Queue"); 143 144 int tcp_reass_maxseg = 0; 145 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD, 146 &tcp_reass_maxseg, 0, 147 "Global maximum number of TCP Segments in Reassembly Queue"); 148 149 int tcp_reass_qsize = 0; 150 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 151 &tcp_reass_qsize, 0, 152 "Global number of TCP Segments currently in Reassembly Queue"); 153 154 static int tcp_reass_overflows = 0; 155 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 156 &tcp_reass_overflows, 0, 157 "Global number of TCP Segment Reassembly Queue Overflows"); 158 159 struct inpcbinfo tcbinfo[MAXCPU]; 160 161 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 162 static void tcp_pulloutofband(struct socket *, 163 struct tcphdr *, struct mbuf *, int); 164 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 165 struct mbuf *); 166 static void tcp_xmit_timer(struct tcpcb *, int); 167 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 168 169 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 170 #ifdef INET6 171 #define ND6_HINT(tp) \ 172 do { \ 173 if ((tp) && (tp)->t_inpcb && \ 174 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 175 (tp)->t_inpcb->in6p_route.ro_rt) \ 176 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 177 } while (0) 178 #else 179 #define ND6_HINT(tp) 180 #endif 181 182 /* 183 * Indicate whether this ack should be delayed. We can delay the ack if 184 * - delayed acks are enabled and 185 * - there is no delayed ack timer in progress and 186 * - our last ack wasn't a 0-sized window. We never want to delay 187 * the ack that opens up a 0-sized window. 188 */ 189 #define DELAY_ACK(tp) \ 190 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 191 (tp->t_flags & TF_RXWIN0SENT) == 0) 192 193 static int 194 tcp_reass(tp, th, tlenp, m) 195 struct tcpcb *tp; 196 struct tcphdr *th; 197 int *tlenp; 198 struct mbuf *m; 199 { 200 struct tseg_qent *q; 201 struct tseg_qent *p = NULL; 202 struct tseg_qent *nq; 203 struct tseg_qent *te; 204 struct socket *so = tp->t_inpcb->inp_socket; 205 int flags; 206 207 /* 208 * Call with th==0 after become established to 209 * force pre-ESTABLISHED data up to user socket. 210 */ 211 if (th == 0) 212 goto present; 213 214 /* 215 * Limit the number of segments in the reassembly queue to prevent 216 * holding on to too many segments (and thus running out of mbufs). 217 * Make sure to let the missing segment through which caused this 218 * queue. Always keep one global queue entry spare to be able to 219 * process the missing segment. 220 */ 221 if (th->th_seq != tp->rcv_nxt && 222 tcp_reass_qsize + 1 >= tcp_reass_maxseg) { 223 tcp_reass_overflows++; 224 tcpstat.tcps_rcvmemdrop++; 225 m_freem(m); 226 return (0); 227 } 228 229 /* Allocate a new queue entry. */ 230 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ, 231 M_INTWAIT | M_NULLOK); 232 if (te == NULL) { 233 tcpstat.tcps_rcvmemdrop++; 234 m_freem(m); 235 return (0); 236 } 237 tcp_reass_qsize++; 238 239 /* 240 * Find a segment which begins after this one does. 241 */ 242 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 243 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 244 break; 245 p = q; 246 } 247 248 /* 249 * If there is a preceding segment, it may provide some of 250 * our data already. If so, drop the data from the incoming 251 * segment. If it provides all of our data, drop us. 252 */ 253 if (p != NULL) { 254 int i; 255 /* conversion to int (in i) handles seq wraparound */ 256 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 257 if (i > 0) { 258 if (i >= *tlenp) { 259 tcpstat.tcps_rcvduppack++; 260 tcpstat.tcps_rcvdupbyte += *tlenp; 261 m_freem(m); 262 free(te, M_TSEGQ); 263 tcp_reass_qsize--; 264 /* 265 * Try to present any queued data 266 * at the left window edge to the user. 267 * This is needed after the 3-WHS 268 * completes. 269 */ 270 goto present; /* ??? */ 271 } 272 m_adj(m, i); 273 *tlenp -= i; 274 th->th_seq += i; 275 } 276 } 277 tcpstat.tcps_rcvoopack++; 278 tcpstat.tcps_rcvoobyte += *tlenp; 279 280 /* 281 * While we overlap succeeding segments trim them or, 282 * if they are completely covered, dequeue them. 283 */ 284 while (q) { 285 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 286 if (i <= 0) 287 break; 288 if (i < q->tqe_len) { 289 q->tqe_th->th_seq += i; 290 q->tqe_len -= i; 291 m_adj(q->tqe_m, i); 292 break; 293 } 294 295 nq = LIST_NEXT(q, tqe_q); 296 LIST_REMOVE(q, tqe_q); 297 m_freem(q->tqe_m); 298 free(q, M_TSEGQ); 299 tcp_reass_qsize--; 300 q = nq; 301 } 302 303 /* Insert the new segment queue entry into place. */ 304 te->tqe_m = m; 305 te->tqe_th = th; 306 te->tqe_len = *tlenp; 307 308 if (p == NULL) { 309 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 310 } else { 311 LIST_INSERT_AFTER(p, te, tqe_q); 312 } 313 314 present: 315 /* 316 * Present data to user, advancing rcv_nxt through 317 * completed sequence space. 318 */ 319 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 320 return (0); 321 q = LIST_FIRST(&tp->t_segq); 322 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 323 return (0); 324 do { 325 tp->rcv_nxt += q->tqe_len; 326 flags = q->tqe_th->th_flags & TH_FIN; 327 nq = LIST_NEXT(q, tqe_q); 328 LIST_REMOVE(q, tqe_q); 329 if (so->so_state & SS_CANTRCVMORE) 330 m_freem(q->tqe_m); 331 else 332 sbappend(&so->so_rcv, q->tqe_m); 333 free(q, M_TSEGQ); 334 tcp_reass_qsize--; 335 q = nq; 336 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 337 ND6_HINT(tp); 338 sorwakeup(so); 339 return (flags); 340 } 341 342 /* 343 * TCP input routine, follows pages 65-76 of the 344 * protocol specification dated September, 1981 very closely. 345 */ 346 #ifdef INET6 347 int 348 tcp6_input(mp, offp, proto) 349 struct mbuf **mp; 350 int *offp, proto; 351 { 352 struct mbuf *m = *mp; 353 struct in6_ifaddr *ia6; 354 355 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 356 357 /* 358 * draft-itojun-ipv6-tcp-to-anycast 359 * better place to put this in? 360 */ 361 ia6 = ip6_getdstifaddr(m); 362 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 363 struct ip6_hdr *ip6; 364 365 ip6 = mtod(m, struct ip6_hdr *); 366 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 367 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 368 return (IPPROTO_DONE); 369 } 370 371 tcp_input(m, *offp, proto); 372 return (IPPROTO_DONE); 373 } 374 #endif 375 376 void 377 tcp_input(m, off0, proto) 378 struct mbuf *m; 379 int off0, proto; 380 { 381 struct tcphdr *th; 382 struct ip *ip = NULL; 383 struct ipovly *ipov; 384 struct inpcb *inp = NULL; 385 u_char *optp = NULL; 386 int optlen = 0; 387 int len, tlen, off; 388 int drop_hdrlen; 389 struct tcpcb *tp = NULL; 390 int thflags; 391 struct socket *so = 0; 392 int todrop, acked, ourfinisacked, needoutput = 0; 393 u_long tiwin; 394 struct tcpopt to; /* options in this segment */ 395 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 396 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 397 struct sockaddr_in *next_hop = NULL; 398 int rstreason; /* For badport_bandlim accounting purposes */ 399 int cpu; 400 struct ip6_hdr *ip6 = NULL; 401 #ifdef INET6 402 boolean_t isipv6; 403 #else 404 const boolean_t isipv6 = FALSE; 405 #endif 406 #ifdef TCPDEBUG 407 short ostate = 0; 408 #endif 409 410 tcpstat.tcps_rcvtotal++; 411 412 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */ 413 while (m->m_type == MT_TAG) { 414 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 415 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 416 m = m->m_next; 417 } 418 419 #ifdef INET6 420 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE; 421 #endif 422 423 if (isipv6) { 424 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 425 ip6 = mtod(m, struct ip6_hdr *); 426 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 427 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 428 tcpstat.tcps_rcvbadsum++; 429 goto drop; 430 } 431 th = (struct tcphdr *)((caddr_t)ip6 + off0); 432 433 /* 434 * Be proactive about unspecified IPv6 address in source. 435 * As we use all-zero to indicate unbounded/unconnected pcb, 436 * unspecified IPv6 address can be used to confuse us. 437 * 438 * Note that packets with unspecified IPv6 destination is 439 * already dropped in ip6_input. 440 */ 441 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 442 /* XXX stat */ 443 goto drop; 444 } 445 } else { 446 /* 447 * Get IP and TCP header together in first mbuf. 448 * Note: IP leaves IP header in first mbuf. 449 */ 450 if (off0 > sizeof(struct ip)) { 451 ip_stripoptions(m); 452 off0 = sizeof(struct ip); 453 } 454 /* already checked and pulled up in ip_demux() */ 455 KASSERT(m->m_len >= sizeof(struct tcpiphdr), 456 ("TCP header not in one mbuf")); 457 ip = mtod(m, struct ip *); 458 ipov = (struct ipovly *)ip; 459 th = (struct tcphdr *)((caddr_t)ip + off0); 460 tlen = ip->ip_len; 461 462 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 463 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 464 th->th_sum = m->m_pkthdr.csum_data; 465 else 466 th->th_sum = in_pseudo(ip->ip_src.s_addr, 467 ip->ip_dst.s_addr, 468 htonl(m->m_pkthdr.csum_data + 469 ip->ip_len + 470 IPPROTO_TCP)); 471 th->th_sum ^= 0xffff; 472 } else { 473 /* 474 * Checksum extended TCP header and data. 475 */ 476 len = sizeof(struct ip) + tlen; 477 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 478 ipov->ih_len = (u_short)tlen; 479 ipov->ih_len = htons(ipov->ih_len); 480 th->th_sum = in_cksum(m, len); 481 } 482 if (th->th_sum) { 483 tcpstat.tcps_rcvbadsum++; 484 goto drop; 485 } 486 #ifdef INET6 487 /* Re-initialization for later version check */ 488 ip->ip_v = IPVERSION; 489 #endif 490 } 491 492 /* 493 * Check that TCP offset makes sense, 494 * pull out TCP options and adjust length. XXX 495 */ 496 off = th->th_off << 2; 497 /* already checked and pulled up in ip_demux() */ 498 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen, 499 ("bad TCP data offset")); 500 tlen -= off; /* tlen is used instead of ti->ti_len */ 501 if (off > sizeof(struct tcphdr)) { 502 if (isipv6) { 503 IP6_EXTHDR_CHECK(m, off0, off, ); 504 ip6 = mtod(m, struct ip6_hdr *); 505 th = (struct tcphdr *)((caddr_t)ip6 + off0); 506 } else { 507 /* already pulled up in ip_demux() */ 508 KASSERT(m->m_len >= sizeof(struct ip) + off, 509 ("TCP header and options not in one mbuf")); 510 } 511 optlen = off - sizeof(struct tcphdr); 512 optp = (u_char *)(th + 1); 513 } 514 thflags = th->th_flags; 515 516 #ifdef TCP_DROP_SYNFIN 517 /* 518 * If the drop_synfin option is enabled, drop all packets with 519 * both the SYN and FIN bits set. This prevents e.g. nmap from 520 * identifying the TCP/IP stack. 521 * 522 * This is a violation of the TCP specification. 523 */ 524 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 525 goto drop; 526 #endif 527 528 /* 529 * Convert TCP protocol specific fields to host format. 530 */ 531 th->th_seq = ntohl(th->th_seq); 532 th->th_ack = ntohl(th->th_ack); 533 th->th_win = ntohs(th->th_win); 534 th->th_urp = ntohs(th->th_urp); 535 536 /* 537 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 538 * until after ip6_savecontrol() is called and before other functions 539 * which don't want those proto headers. 540 * Because ip6_savecontrol() is going to parse the mbuf to 541 * search for data to be passed up to user-land, it wants mbuf 542 * parameters to be unchanged. 543 * XXX: the call of ip6_savecontrol() has been obsoleted based on 544 * latest version of the advanced API (20020110). 545 */ 546 drop_hdrlen = off0 + off; 547 548 /* 549 * Locate pcb for segment. 550 */ 551 findpcb: 552 /* IPFIREWALL_FORWARD section */ 553 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */ 554 /* 555 * Transparently forwarded. Pretend to be the destination. 556 * already got one like this? 557 */ 558 cpu = mycpu->gd_cpuid; 559 inp = in_pcblookup_hash(&tcbinfo[cpu], 560 ip->ip_src, th->th_sport, 561 ip->ip_dst, th->th_dport, 562 0, m->m_pkthdr.rcvif); 563 if (!inp) { 564 /* 565 * It's new. Try to find the ambushing socket. 566 */ 567 568 /* 569 * The rest of the ipfw code stores the port in 570 * host order. XXX 571 * (The IP address is still in network order.) 572 */ 573 in_port_t dport = next_hop->sin_port ? 574 htons(next_hop->sin_port) : 575 th->th_dport; 576 577 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport, 578 next_hop->sin_addr.s_addr, dport); 579 inp = in_pcblookup_hash(&tcbinfo[cpu], 580 ip->ip_src, th->th_sport, 581 next_hop->sin_addr, dport, 582 1, m->m_pkthdr.rcvif); 583 } 584 } else { 585 if (isipv6) { 586 inp = in6_pcblookup_hash(&tcbinfo[0], 587 &ip6->ip6_src, th->th_sport, 588 &ip6->ip6_dst, th->th_dport, 589 1, m->m_pkthdr.rcvif); 590 } else { 591 cpu = mycpu->gd_cpuid; 592 inp = in_pcblookup_hash(&tcbinfo[cpu], 593 ip->ip_src, th->th_sport, 594 ip->ip_dst, th->th_dport, 595 1, m->m_pkthdr.rcvif); 596 } 597 } 598 599 #ifdef IPSEC 600 if (isipv6) { 601 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 602 ipsec6stat.in_polvio++; 603 goto drop; 604 } 605 } else { 606 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 607 ipsecstat.in_polvio++; 608 goto drop; 609 } 610 } 611 #endif 612 #ifdef FAST_IPSEC 613 if (isipv6) { 614 if (inp != NULL && ipsec6_in_reject(m, inp)) { 615 goto drop; 616 } 617 } else { 618 if (inp != NULL && ipsec4_in_reject(m, inp)) { 619 goto drop; 620 } 621 } 622 #endif 623 624 /* 625 * If the state is CLOSED (i.e., TCB does not exist) then 626 * all data in the incoming segment is discarded. 627 * If the TCB exists but is in CLOSED state, it is embryonic, 628 * but should either do a listen or a connect soon. 629 */ 630 if (inp == NULL) { 631 if (log_in_vain) { 632 #ifdef INET6 633 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 634 #else 635 char dbuf[4 * sizeof "123"], sbuf[4 * sizeof "123"]; 636 #endif 637 if (isipv6) { 638 strcpy(dbuf, "["); 639 strcpy(sbuf, "["); 640 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 641 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 642 strcat(dbuf, "]"); 643 strcat(sbuf, "]"); 644 } else { 645 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 646 strcpy(sbuf, inet_ntoa(ip->ip_src)); 647 } 648 switch (log_in_vain) { 649 case 1: 650 if ((thflags & TH_SYN) == 0) 651 break; 652 case 2: 653 log(LOG_INFO, 654 "Connection attempt to TCP %s:%d " 655 "from %s:%d flags:0x%02x\n", 656 dbuf, ntohs(th->th_dport), sbuf, 657 ntohs(th->th_sport), thflags); 658 break; 659 default: 660 break; 661 } 662 } 663 if (blackhole) { 664 switch (blackhole) { 665 case 1: 666 if (thflags & TH_SYN) 667 goto drop; 668 break; 669 case 2: 670 goto drop; 671 default: 672 goto drop; 673 } 674 } 675 rstreason = BANDLIM_RST_CLOSEDPORT; 676 goto dropwithreset; 677 } 678 tp = intotcpcb(inp); 679 if (tp == NULL) { 680 rstreason = BANDLIM_RST_CLOSEDPORT; 681 goto dropwithreset; 682 } 683 if (tp->t_state == TCPS_CLOSED) 684 goto drop; 685 686 /* Unscale the window into a 32-bit value. */ 687 if ((thflags & TH_SYN) == 0) 688 tiwin = th->th_win << tp->snd_scale; 689 else 690 tiwin = th->th_win; 691 692 so = inp->inp_socket; 693 694 #ifdef TCPDEBUG 695 if (so->so_options & SO_DEBUG) { 696 ostate = tp->t_state; 697 if (isipv6) 698 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 699 else 700 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 701 tcp_savetcp = *th; 702 } 703 #endif 704 705 bzero((char *)&to, sizeof(to)); 706 707 if (so->so_options & SO_ACCEPTCONN) { 708 struct in_conninfo inc; 709 710 #ifdef INET6 711 inc.inc_isipv6 = (isipv6 == TRUE); 712 #endif 713 if (isipv6) { 714 inc.inc6_faddr = ip6->ip6_src; 715 inc.inc6_laddr = ip6->ip6_dst; 716 inc.inc6_route.ro_rt = NULL; /* XXX */ 717 } else { 718 inc.inc_faddr = ip->ip_src; 719 inc.inc_laddr = ip->ip_dst; 720 inc.inc_route.ro_rt = NULL; /* XXX */ 721 } 722 inc.inc_fport = th->th_sport; 723 inc.inc_lport = th->th_dport; 724 725 /* 726 * If the state is LISTEN then ignore segment if it contains 727 * a RST. If the segment contains an ACK then it is bad and 728 * send a RST. If it does not contain a SYN then it is not 729 * interesting; drop it. 730 * 731 * If the state is SYN_RECEIVED (syncache) and seg contains 732 * an ACK, but not for our SYN/ACK, send a RST. If the seg 733 * contains a RST, check the sequence number to see if it 734 * is a valid reset segment. 735 */ 736 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 737 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 738 if (!syncache_expand(&inc, th, &so, m)) { 739 /* 740 * No syncache entry, or ACK was not 741 * for our SYN/ACK. Send a RST. 742 */ 743 tcpstat.tcps_badsyn++; 744 rstreason = BANDLIM_RST_OPENPORT; 745 goto dropwithreset; 746 } 747 if (so == NULL) 748 /* 749 * Could not complete 3-way handshake, 750 * connection is being closed down, and 751 * syncache will free mbuf. 752 */ 753 return; 754 /* 755 * Socket is created in state SYN_RECEIVED. 756 * Continue processing segment. 757 */ 758 inp = sotoinpcb(so); 759 tp = intotcpcb(inp); 760 /* 761 * This is what would have happened in 762 * tcp_output() when the SYN,ACK was sent. 763 */ 764 tp->snd_up = tp->snd_una; 765 tp->snd_max = tp->snd_nxt = tp->iss + 1; 766 tp->last_ack_sent = tp->rcv_nxt; 767 /* 768 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 769 * until the _second_ ACK is received: 770 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 771 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 772 * move to ESTAB, set snd_wnd to tiwin. 773 */ 774 tp->snd_wnd = tiwin; /* unscaled */ 775 goto after_listen; 776 } 777 if (thflags & TH_RST) { 778 syncache_chkrst(&inc, th); 779 goto drop; 780 } 781 if (thflags & TH_ACK) { 782 syncache_badack(&inc); 783 tcpstat.tcps_badsyn++; 784 rstreason = BANDLIM_RST_OPENPORT; 785 goto dropwithreset; 786 } 787 goto drop; 788 } 789 790 /* 791 * Segment's flags are (SYN) or (SYN|FIN). 792 */ 793 #ifdef INET6 794 /* 795 * If deprecated address is forbidden, 796 * we do not accept SYN to deprecated interface 797 * address to prevent any new inbound connection from 798 * getting established. 799 * When we do not accept SYN, we send a TCP RST, 800 * with deprecated source address (instead of dropping 801 * it). We compromise it as it is much better for peer 802 * to send a RST, and RST will be the final packet 803 * for the exchange. 804 * 805 * If we do not forbid deprecated addresses, we accept 806 * the SYN packet. RFC2462 does not suggest dropping 807 * SYN in this case. 808 * If we decipher RFC2462 5.5.4, it says like this: 809 * 1. use of deprecated addr with existing 810 * communication is okay - "SHOULD continue to be 811 * used" 812 * 2. use of it with new communication: 813 * (2a) "SHOULD NOT be used if alternate address 814 * with sufficient scope is available" 815 * (2b) nothing mentioned otherwise. 816 * Here we fall into (2b) case as we have no choice in 817 * our source address selection - we must obey the peer. 818 * 819 * The wording in RFC2462 is confusing, and there are 820 * multiple description text for deprecated address 821 * handling - worse, they are not exactly the same. 822 * I believe 5.5.4 is the best one, so we follow 5.5.4. 823 */ 824 if (isipv6 && !ip6_use_deprecated) { 825 struct in6_ifaddr *ia6; 826 827 if ((ia6 = ip6_getdstifaddr(m)) && 828 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 829 tp = NULL; 830 rstreason = BANDLIM_RST_OPENPORT; 831 goto dropwithreset; 832 } 833 } 834 #endif 835 /* 836 * If it is from this socket, drop it, it must be forged. 837 * Don't bother responding if the destination was a broadcast. 838 */ 839 if (th->th_dport == th->th_sport) { 840 if (isipv6) { 841 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 842 &ip6->ip6_src)) 843 goto drop; 844 } else { 845 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 846 goto drop; 847 } 848 } 849 /* 850 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 851 * 852 * Note that it is quite possible to receive unicast 853 * link-layer packets with a broadcast IP address. Use 854 * in_broadcast() to find them. 855 */ 856 if (m->m_flags & (M_BCAST|M_MCAST)) 857 goto drop; 858 if (isipv6) { 859 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 860 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 861 goto drop; 862 } else { 863 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 864 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 865 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 866 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 867 goto drop; 868 } 869 /* 870 * SYN appears to be valid; create compressed TCP state 871 * for syncache, or perform t/tcp connection. 872 */ 873 if (so->so_qlen <= so->so_qlimit) { 874 tcp_dooptions(&to, optp, optlen, 1); 875 if (!syncache_add(&inc, &to, th, &so, m)) 876 goto drop; 877 if (so == NULL) 878 /* 879 * Entry added to syncache, mbuf used to 880 * send SYN,ACK packet. 881 */ 882 return; 883 /* 884 * Segment passed TAO tests. 885 */ 886 inp = sotoinpcb(so); 887 tp = intotcpcb(inp); 888 tp->snd_wnd = tiwin; 889 tp->t_starttime = ticks; 890 tp->t_state = TCPS_ESTABLISHED; 891 892 /* 893 * If there is a FIN, or if there is data and the 894 * connection is local, then delay SYN,ACK(SYN) in 895 * the hope of piggy-backing it on a response 896 * segment. Otherwise must send ACK now in case 897 * the other side is slow starting. 898 */ 899 if (DELAY_ACK(tp) && 900 ((thflags & TH_FIN) || 901 (tlen != 0 && 902 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 903 (!isipv6 && in_localaddr(inp->inp_faddr)))))) { 904 callout_reset(tp->tt_delack, tcp_delacktime, 905 tcp_timer_delack, tp); 906 tp->t_flags |= TF_NEEDSYN; 907 } else 908 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 909 910 tcpstat.tcps_connects++; 911 soisconnected(so); 912 goto trimthenstep6; 913 } 914 goto drop; 915 } 916 after_listen: 917 918 /* XXX temp debugging */ 919 /* should not happen - syncache should pick up these connections */ 920 if (tp->t_state == TCPS_LISTEN) 921 panic("tcp_input: TCPS_LISTEN"); 922 923 /* 924 * Segment received on connection. 925 * Reset idle time and keep-alive timer. 926 */ 927 tp->t_rcvtime = ticks; 928 if (TCPS_HAVEESTABLISHED(tp->t_state)) 929 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 930 931 /* 932 * Process options. 933 * XXX this is tradtitional behavior, may need to be cleaned up. 934 */ 935 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 936 if (thflags & TH_SYN) { 937 if (to.to_flags & TOF_SCALE) { 938 tp->t_flags |= TF_RCVD_SCALE; 939 tp->requested_s_scale = to.to_requested_s_scale; 940 } 941 if (to.to_flags & TOF_TS) { 942 tp->t_flags |= TF_RCVD_TSTMP; 943 tp->ts_recent = to.to_tsval; 944 tp->ts_recent_age = ticks; 945 } 946 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 947 tp->t_flags |= TF_RCVD_CC; 948 if (to.to_flags & TOF_MSS) 949 tcp_mss(tp, to.to_mss); 950 } 951 952 /* 953 * Header prediction: check for the two common cases 954 * of a uni-directional data xfer. If the packet has 955 * no control flags, is in-sequence, the window didn't 956 * change and we're not retransmitting, it's a 957 * candidate. If the length is zero and the ack moved 958 * forward, we're the sender side of the xfer. Just 959 * free the data acked & wake any higher level process 960 * that was blocked waiting for space. If the length 961 * is non-zero and the ack didn't move, we're the 962 * receiver side. If we're getting packets in-order 963 * (the reassembly queue is empty), add the data to 964 * the socket buffer and note that we need a delayed ack. 965 * Make sure that the hidden state-flags are also off. 966 * Since we check for TCPS_ESTABLISHED above, it can only 967 * be TH_NEEDSYN. 968 */ 969 if (tp->t_state == TCPS_ESTABLISHED && 970 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 971 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 972 ((to.to_flags & TOF_TS) == 0 || 973 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 974 /* 975 * Using the CC option is compulsory if once started: 976 * the segment is OK if no T/TCP was negotiated or 977 * if the segment has a CC option equal to CCrecv 978 */ 979 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 980 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 981 th->th_seq == tp->rcv_nxt && 982 tiwin && tiwin == tp->snd_wnd && 983 tp->snd_nxt == tp->snd_max) { 984 985 /* 986 * If last ACK falls within this segment's sequence numbers, 987 * record the timestamp. 988 * NOTE that the test is modified according to the latest 989 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 990 */ 991 if ((to.to_flags & TOF_TS) != 0 && 992 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 993 tp->ts_recent_age = ticks; 994 tp->ts_recent = to.to_tsval; 995 } 996 997 if (tlen == 0) { 998 if (SEQ_GT(th->th_ack, tp->snd_una) && 999 SEQ_LEQ(th->th_ack, tp->snd_max) && 1000 tp->snd_cwnd >= tp->snd_wnd && 1001 ((!tcp_do_newreno && 1002 tp->t_dupacks < tcprexmtthresh) || 1003 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 1004 /* 1005 * this is a pure ack for outstanding data. 1006 */ 1007 ++tcpstat.tcps_predack; 1008 /* 1009 * "bad retransmit" recovery 1010 * 1011 * If Eifel detection applies, then 1012 * it is deterministic, so use it 1013 * unconditionally over the old heuristic. 1014 * Otherwise, fall back to the old heuristic. 1015 */ 1016 if (tcp_do_eifel_detect && 1017 (to.to_flags & TOF_TS) && to.to_tsecr && 1018 (tp->t_flags & TF_FIRSTACCACK)) { 1019 /* Eifel detection applicable. */ 1020 if (to.to_tsecr < tp->t_rexmtTS) { 1021 tcp_revert_congestion_state(tp); 1022 ++tcpstat.tcps_eifeldetected; 1023 } 1024 } else if (tp->t_rxtshift == 1 && 1025 ticks < tp->t_badrxtwin) { 1026 tcp_revert_congestion_state(tp); 1027 ++tcpstat.tcps_rttdetected; 1028 } 1029 tp->t_flags &= ~(TF_FIRSTACCACK | 1030 TF_FASTREXMT | TF_EARLYREXMT); 1031 /* 1032 * Recalculate the retransmit timer / rtt. 1033 * 1034 * Some machines (certain windows boxes) 1035 * send broken timestamp replies during the 1036 * SYN+ACK phase, ignore timestamps of 0. 1037 */ 1038 if ((to.to_flags & TOF_TS) != 0 && 1039 to.to_tsecr) { 1040 tcp_xmit_timer(tp, 1041 ticks - to.to_tsecr + 1); 1042 } else if (tp->t_rtttime && 1043 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1044 tcp_xmit_timer(tp, 1045 ticks - tp->t_rtttime); 1046 } 1047 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1048 acked = th->th_ack - tp->snd_una; 1049 tcpstat.tcps_rcvackpack++; 1050 tcpstat.tcps_rcvackbyte += acked; 1051 sbdrop(&so->so_snd, acked); 1052 tp->snd_recover = th->th_ack - 1; 1053 tp->snd_una = th->th_ack; 1054 tp->t_dupacks = 0; 1055 m_freem(m); 1056 ND6_HINT(tp); /* some progress has been done */ 1057 1058 /* 1059 * If all outstanding data are acked, stop 1060 * retransmit timer, otherwise restart timer 1061 * using current (possibly backed-off) value. 1062 * If process is waiting for space, 1063 * wakeup/selwakeup/signal. If data 1064 * are ready to send, let tcp_output 1065 * decide between more output or persist. 1066 */ 1067 if (tp->snd_una == tp->snd_max) 1068 callout_stop(tp->tt_rexmt); 1069 else if (!callout_active(tp->tt_persist)) 1070 callout_reset(tp->tt_rexmt, 1071 tp->t_rxtcur, 1072 tcp_timer_rexmt, tp); 1073 1074 sowwakeup(so); 1075 if (so->so_snd.sb_cc) 1076 (void) tcp_output(tp); 1077 return; 1078 } 1079 } else if (th->th_ack == tp->snd_una && 1080 LIST_EMPTY(&tp->t_segq) && 1081 tlen <= sbspace(&so->so_rcv)) { 1082 /* 1083 * this is a pure, in-sequence data packet 1084 * with nothing on the reassembly queue and 1085 * we have enough buffer space to take it. 1086 */ 1087 ++tcpstat.tcps_preddat; 1088 tp->rcv_nxt += tlen; 1089 tcpstat.tcps_rcvpack++; 1090 tcpstat.tcps_rcvbyte += tlen; 1091 ND6_HINT(tp); /* some progress has been done */ 1092 /* 1093 * Add data to socket buffer. 1094 */ 1095 if (so->so_state & SS_CANTRCVMORE) { 1096 m_freem(m); 1097 } else { 1098 m_adj(m, drop_hdrlen); /* delayed header drop */ 1099 sbappend(&so->so_rcv, m); 1100 } 1101 sorwakeup(so); 1102 if (DELAY_ACK(tp)) { 1103 callout_reset(tp->tt_delack, tcp_delacktime, 1104 tcp_timer_delack, tp); 1105 } else { 1106 tp->t_flags |= TF_ACKNOW; 1107 tcp_output(tp); 1108 } 1109 return; 1110 } 1111 } 1112 1113 /* 1114 * Calculate amount of space in receive window, 1115 * and then do TCP input processing. 1116 * Receive window is amount of space in rcv queue, 1117 * but not less than advertised window. 1118 */ 1119 { int win; 1120 1121 win = sbspace(&so->so_rcv); 1122 if (win < 0) 1123 win = 0; 1124 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1125 } 1126 1127 switch (tp->t_state) { 1128 1129 /* 1130 * If the state is SYN_RECEIVED: 1131 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1132 */ 1133 case TCPS_SYN_RECEIVED: 1134 if ((thflags & TH_ACK) && 1135 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1136 SEQ_GT(th->th_ack, tp->snd_max))) { 1137 rstreason = BANDLIM_RST_OPENPORT; 1138 goto dropwithreset; 1139 } 1140 break; 1141 1142 /* 1143 * If the state is SYN_SENT: 1144 * if seg contains an ACK, but not for our SYN, drop the input. 1145 * if seg contains a RST, then drop the connection. 1146 * if seg does not contain SYN, then drop it. 1147 * Otherwise this is an acceptable SYN segment 1148 * initialize tp->rcv_nxt and tp->irs 1149 * if seg contains ack then advance tp->snd_una 1150 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1151 * arrange for segment to be acked (eventually) 1152 * continue processing rest of data/controls, beginning with URG 1153 */ 1154 case TCPS_SYN_SENT: 1155 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1156 taop = &tao_noncached; 1157 bzero(taop, sizeof(*taop)); 1158 } 1159 1160 if ((thflags & TH_ACK) && 1161 (SEQ_LEQ(th->th_ack, tp->iss) || 1162 SEQ_GT(th->th_ack, tp->snd_max))) { 1163 /* 1164 * If we have a cached CCsent for the remote host, 1165 * hence we haven't just crashed and restarted, 1166 * do not send a RST. This may be a retransmission 1167 * from the other side after our earlier ACK was lost. 1168 * Our new SYN, when it arrives, will serve as the 1169 * needed ACK. 1170 */ 1171 if (taop->tao_ccsent != 0) 1172 goto drop; 1173 else { 1174 rstreason = BANDLIM_UNLIMITED; 1175 goto dropwithreset; 1176 } 1177 } 1178 if (thflags & TH_RST) { 1179 if (thflags & TH_ACK) 1180 tp = tcp_drop(tp, ECONNREFUSED); 1181 goto drop; 1182 } 1183 if ((thflags & TH_SYN) == 0) 1184 goto drop; 1185 tp->snd_wnd = th->th_win; /* initial send window */ 1186 tp->cc_recv = to.to_cc; /* foreign CC */ 1187 1188 tp->irs = th->th_seq; 1189 tcp_rcvseqinit(tp); 1190 if (thflags & TH_ACK) { 1191 /* 1192 * Our SYN was acked. If segment contains CC.ECHO 1193 * option, check it to make sure this segment really 1194 * matches our SYN. If not, just drop it as old 1195 * duplicate, but send an RST if we're still playing 1196 * by the old rules. If no CC.ECHO option, make sure 1197 * we don't get fooled into using T/TCP. 1198 */ 1199 if (to.to_flags & TOF_CCECHO) { 1200 if (tp->cc_send != to.to_ccecho) { 1201 if (taop->tao_ccsent != 0) 1202 goto drop; 1203 else { 1204 rstreason = BANDLIM_UNLIMITED; 1205 goto dropwithreset; 1206 } 1207 } 1208 } else 1209 tp->t_flags &= ~TF_RCVD_CC; 1210 tcpstat.tcps_connects++; 1211 soisconnected(so); 1212 /* Do window scaling on this connection? */ 1213 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1214 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1215 tp->snd_scale = tp->requested_s_scale; 1216 tp->rcv_scale = tp->request_r_scale; 1217 } 1218 /* Segment is acceptable, update cache if undefined. */ 1219 if (taop->tao_ccsent == 0) 1220 taop->tao_ccsent = to.to_ccecho; 1221 1222 tp->rcv_adv += tp->rcv_wnd; 1223 tp->snd_una++; /* SYN is acked */ 1224 /* 1225 * If there's data, delay ACK; if there's also a FIN 1226 * ACKNOW will be turned on later. 1227 */ 1228 if (DELAY_ACK(tp) && tlen != 0) 1229 callout_reset(tp->tt_delack, tcp_delacktime, 1230 tcp_timer_delack, tp); 1231 else 1232 tp->t_flags |= TF_ACKNOW; 1233 /* 1234 * Received <SYN,ACK> in SYN_SENT[*] state. 1235 * Transitions: 1236 * SYN_SENT --> ESTABLISHED 1237 * SYN_SENT* --> FIN_WAIT_1 1238 */ 1239 tp->t_starttime = ticks; 1240 if (tp->t_flags & TF_NEEDFIN) { 1241 tp->t_state = TCPS_FIN_WAIT_1; 1242 tp->t_flags &= ~TF_NEEDFIN; 1243 thflags &= ~TH_SYN; 1244 } else { 1245 tp->t_state = TCPS_ESTABLISHED; 1246 callout_reset(tp->tt_keep, tcp_keepidle, 1247 tcp_timer_keep, tp); 1248 } 1249 } else { 1250 /* 1251 * Received initial SYN in SYN-SENT[*] state => 1252 * simultaneous open. If segment contains CC option 1253 * and there is a cached CC, apply TAO test. 1254 * If it succeeds, connection is * half-synchronized. 1255 * Otherwise, do 3-way handshake: 1256 * SYN-SENT -> SYN-RECEIVED 1257 * SYN-SENT* -> SYN-RECEIVED* 1258 * If there was no CC option, clear cached CC value. 1259 */ 1260 tp->t_flags |= TF_ACKNOW; 1261 callout_stop(tp->tt_rexmt); 1262 if (to.to_flags & TOF_CC) { 1263 if (taop->tao_cc != 0 && 1264 CC_GT(to.to_cc, taop->tao_cc)) { 1265 /* 1266 * update cache and make transition: 1267 * SYN-SENT -> ESTABLISHED* 1268 * SYN-SENT* -> FIN-WAIT-1* 1269 */ 1270 taop->tao_cc = to.to_cc; 1271 tp->t_starttime = ticks; 1272 if (tp->t_flags & TF_NEEDFIN) { 1273 tp->t_state = TCPS_FIN_WAIT_1; 1274 tp->t_flags &= ~TF_NEEDFIN; 1275 } else { 1276 tp->t_state = TCPS_ESTABLISHED; 1277 callout_reset(tp->tt_keep, 1278 tcp_keepidle, 1279 tcp_timer_keep, 1280 tp); 1281 } 1282 tp->t_flags |= TF_NEEDSYN; 1283 } else 1284 tp->t_state = TCPS_SYN_RECEIVED; 1285 } else { 1286 /* CC.NEW or no option => invalidate cache */ 1287 taop->tao_cc = 0; 1288 tp->t_state = TCPS_SYN_RECEIVED; 1289 } 1290 } 1291 1292 trimthenstep6: 1293 /* 1294 * Advance th->th_seq to correspond to first data byte. 1295 * If data, trim to stay within window, 1296 * dropping FIN if necessary. 1297 */ 1298 th->th_seq++; 1299 if (tlen > tp->rcv_wnd) { 1300 todrop = tlen - tp->rcv_wnd; 1301 m_adj(m, -todrop); 1302 tlen = tp->rcv_wnd; 1303 thflags &= ~TH_FIN; 1304 tcpstat.tcps_rcvpackafterwin++; 1305 tcpstat.tcps_rcvbyteafterwin += todrop; 1306 } 1307 tp->snd_wl1 = th->th_seq - 1; 1308 tp->rcv_up = th->th_seq; 1309 /* 1310 * Client side of transaction: already sent SYN and data. 1311 * If the remote host used T/TCP to validate the SYN, 1312 * our data will be ACK'd; if so, enter normal data segment 1313 * processing in the middle of step 5, ack processing. 1314 * Otherwise, goto step 6. 1315 */ 1316 if (thflags & TH_ACK) 1317 goto process_ACK; 1318 1319 goto step6; 1320 1321 /* 1322 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1323 * if segment contains a SYN and CC [not CC.NEW] option: 1324 * if state == TIME_WAIT and connection duration > MSL, 1325 * drop packet and send RST; 1326 * 1327 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1328 * ack the FIN (and data) in retransmission queue. 1329 * Complete close and delete TCPCB. Then reprocess 1330 * segment, hoping to find new TCPCB in LISTEN state; 1331 * 1332 * else must be old SYN; drop it. 1333 * else do normal processing. 1334 */ 1335 case TCPS_LAST_ACK: 1336 case TCPS_CLOSING: 1337 case TCPS_TIME_WAIT: 1338 if ((thflags & TH_SYN) && 1339 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1340 if (tp->t_state == TCPS_TIME_WAIT && 1341 (ticks - tp->t_starttime) > tcp_msl) { 1342 rstreason = BANDLIM_UNLIMITED; 1343 goto dropwithreset; 1344 } 1345 if (CC_GT(to.to_cc, tp->cc_recv)) { 1346 tp = tcp_close(tp); 1347 goto findpcb; 1348 } 1349 else 1350 goto drop; 1351 } 1352 break; /* continue normal processing */ 1353 } 1354 1355 /* 1356 * States other than LISTEN or SYN_SENT. 1357 * First check the RST flag and sequence number since reset segments 1358 * are exempt from the timestamp and connection count tests. This 1359 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1360 * below which allowed reset segments in half the sequence space 1361 * to fall though and be processed (which gives forged reset 1362 * segments with a random sequence number a 50 percent chance of 1363 * killing a connection). 1364 * Then check timestamp, if present. 1365 * Then check the connection count, if present. 1366 * Then check that at least some bytes of segment are within 1367 * receive window. If segment begins before rcv_nxt, 1368 * drop leading data (and SYN); if nothing left, just ack. 1369 * 1370 * 1371 * If the RST bit is set, check the sequence number to see 1372 * if this is a valid reset segment. 1373 * RFC 793 page 37: 1374 * In all states except SYN-SENT, all reset (RST) segments 1375 * are validated by checking their SEQ-fields. A reset is 1376 * valid if its sequence number is in the window. 1377 * Note: this does not take into account delayed ACKs, so 1378 * we should test against last_ack_sent instead of rcv_nxt. 1379 * The sequence number in the reset segment is normally an 1380 * echo of our outgoing acknowlegement numbers, but some hosts 1381 * send a reset with the sequence number at the rightmost edge 1382 * of our receive window, and we have to handle this case. 1383 * If we have multiple segments in flight, the intial reset 1384 * segment sequence numbers will be to the left of last_ack_sent, 1385 * but they will eventually catch up. 1386 * In any case, it never made sense to trim reset segments to 1387 * fit the receive window since RFC 1122 says: 1388 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1389 * 1390 * A TCP SHOULD allow a received RST segment to include data. 1391 * 1392 * DISCUSSION 1393 * It has been suggested that a RST segment could contain 1394 * ASCII text that encoded and explained the cause of the 1395 * RST. No standard has yet been established for such 1396 * data. 1397 * 1398 * If the reset segment passes the sequence number test examine 1399 * the state: 1400 * SYN_RECEIVED STATE: 1401 * If passive open, return to LISTEN state. 1402 * If active open, inform user that connection was refused. 1403 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1404 * Inform user that connection was reset, and close tcb. 1405 * CLOSING, LAST_ACK STATES: 1406 * Close the tcb. 1407 * TIME_WAIT STATE: 1408 * Drop the segment - see Stevens, vol. 2, p. 964 and 1409 * RFC 1337. 1410 */ 1411 if (thflags & TH_RST) { 1412 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1413 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1414 switch (tp->t_state) { 1415 1416 case TCPS_SYN_RECEIVED: 1417 so->so_error = ECONNREFUSED; 1418 goto close; 1419 1420 case TCPS_ESTABLISHED: 1421 case TCPS_FIN_WAIT_1: 1422 case TCPS_FIN_WAIT_2: 1423 case TCPS_CLOSE_WAIT: 1424 so->so_error = ECONNRESET; 1425 close: 1426 tp->t_state = TCPS_CLOSED; 1427 tcpstat.tcps_drops++; 1428 tp = tcp_close(tp); 1429 break; 1430 1431 case TCPS_CLOSING: 1432 case TCPS_LAST_ACK: 1433 tp = tcp_close(tp); 1434 break; 1435 1436 case TCPS_TIME_WAIT: 1437 break; 1438 } 1439 } 1440 goto drop; 1441 } 1442 1443 /* 1444 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1445 * and it's less than ts_recent, drop it. 1446 */ 1447 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1448 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1449 1450 /* Check to see if ts_recent is over 24 days old. */ 1451 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1452 /* 1453 * Invalidate ts_recent. If this segment updates 1454 * ts_recent, the age will be reset later and ts_recent 1455 * will get a valid value. If it does not, setting 1456 * ts_recent to zero will at least satisfy the 1457 * requirement that zero be placed in the timestamp 1458 * echo reply when ts_recent isn't valid. The 1459 * age isn't reset until we get a valid ts_recent 1460 * because we don't want out-of-order segments to be 1461 * dropped when ts_recent is old. 1462 */ 1463 tp->ts_recent = 0; 1464 } else { 1465 tcpstat.tcps_rcvduppack++; 1466 tcpstat.tcps_rcvdupbyte += tlen; 1467 tcpstat.tcps_pawsdrop++; 1468 if (tlen) 1469 goto dropafterack; 1470 goto drop; 1471 } 1472 } 1473 1474 /* 1475 * T/TCP mechanism 1476 * If T/TCP was negotiated and the segment doesn't have CC, 1477 * or if its CC is wrong then drop the segment. 1478 * RST segments do not have to comply with this. 1479 */ 1480 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1481 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1482 goto dropafterack; 1483 1484 /* 1485 * In the SYN-RECEIVED state, validate that the packet belongs to 1486 * this connection before trimming the data to fit the receive 1487 * window. Check the sequence number versus IRS since we know 1488 * the sequence numbers haven't wrapped. This is a partial fix 1489 * for the "LAND" DoS attack. 1490 */ 1491 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1492 rstreason = BANDLIM_RST_OPENPORT; 1493 goto dropwithreset; 1494 } 1495 1496 todrop = tp->rcv_nxt - th->th_seq; 1497 if (todrop > 0) { 1498 if (thflags & TH_SYN) { 1499 thflags &= ~TH_SYN; 1500 th->th_seq++; 1501 if (th->th_urp > 1) 1502 th->th_urp--; 1503 else 1504 thflags &= ~TH_URG; 1505 todrop--; 1506 } 1507 /* 1508 * Following if statement from Stevens, vol. 2, p. 960. 1509 */ 1510 if (todrop > tlen 1511 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1512 /* 1513 * Any valid FIN must be to the left of the window. 1514 * At this point the FIN must be a duplicate or out 1515 * of sequence; drop it. 1516 */ 1517 thflags &= ~TH_FIN; 1518 1519 /* 1520 * Send an ACK to resynchronize and drop any data. 1521 * But keep on processing for RST or ACK. 1522 */ 1523 tp->t_flags |= TF_ACKNOW; 1524 todrop = tlen; 1525 tcpstat.tcps_rcvduppack++; 1526 tcpstat.tcps_rcvdupbyte += todrop; 1527 } else { 1528 tcpstat.tcps_rcvpartduppack++; 1529 tcpstat.tcps_rcvpartdupbyte += todrop; 1530 } 1531 drop_hdrlen += todrop; /* drop from the top afterwards */ 1532 th->th_seq += todrop; 1533 tlen -= todrop; 1534 if (th->th_urp > todrop) 1535 th->th_urp -= todrop; 1536 else { 1537 thflags &= ~TH_URG; 1538 th->th_urp = 0; 1539 } 1540 } 1541 1542 /* 1543 * If new data are received on a connection after the 1544 * user processes are gone, then RST the other end. 1545 */ 1546 if ((so->so_state & SS_NOFDREF) && 1547 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1548 tp = tcp_close(tp); 1549 tcpstat.tcps_rcvafterclose++; 1550 rstreason = BANDLIM_UNLIMITED; 1551 goto dropwithreset; 1552 } 1553 1554 /* 1555 * If segment ends after window, drop trailing data 1556 * (and PUSH and FIN); if nothing left, just ACK. 1557 */ 1558 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1559 if (todrop > 0) { 1560 tcpstat.tcps_rcvpackafterwin++; 1561 if (todrop >= tlen) { 1562 tcpstat.tcps_rcvbyteafterwin += tlen; 1563 /* 1564 * If a new connection request is received 1565 * while in TIME_WAIT, drop the old connection 1566 * and start over if the sequence numbers 1567 * are above the previous ones. 1568 */ 1569 if (thflags & TH_SYN && 1570 tp->t_state == TCPS_TIME_WAIT && 1571 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1572 tp = tcp_close(tp); 1573 goto findpcb; 1574 } 1575 /* 1576 * If window is closed can only take segments at 1577 * window edge, and have to drop data and PUSH from 1578 * incoming segments. Continue processing, but 1579 * remember to ack. Otherwise, drop segment 1580 * and ack. 1581 */ 1582 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1583 tp->t_flags |= TF_ACKNOW; 1584 tcpstat.tcps_rcvwinprobe++; 1585 } else 1586 goto dropafterack; 1587 } else 1588 tcpstat.tcps_rcvbyteafterwin += todrop; 1589 m_adj(m, -todrop); 1590 tlen -= todrop; 1591 thflags &= ~(TH_PUSH|TH_FIN); 1592 } 1593 1594 /* 1595 * If last ACK falls within this segment's sequence numbers, 1596 * record its timestamp. 1597 * NOTE that the test is modified according to the latest 1598 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1599 */ 1600 if ((to.to_flags & TOF_TS) != 0 && 1601 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1602 tp->ts_recent_age = ticks; 1603 tp->ts_recent = to.to_tsval; 1604 } 1605 1606 /* 1607 * If a SYN is in the window, then this is an 1608 * error and we send an RST and drop the connection. 1609 */ 1610 if (thflags & TH_SYN) { 1611 tp = tcp_drop(tp, ECONNRESET); 1612 rstreason = BANDLIM_UNLIMITED; 1613 goto dropwithreset; 1614 } 1615 1616 /* 1617 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1618 * flag is on (half-synchronized state), then queue data for 1619 * later processing; else drop segment and return. 1620 */ 1621 if ((thflags & TH_ACK) == 0) { 1622 if (tp->t_state == TCPS_SYN_RECEIVED || 1623 (tp->t_flags & TF_NEEDSYN)) 1624 goto step6; 1625 else 1626 goto drop; 1627 } 1628 1629 /* 1630 * Ack processing. 1631 */ 1632 switch (tp->t_state) { 1633 1634 /* 1635 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1636 * ESTABLISHED state and continue processing. 1637 * The ACK was checked above. 1638 */ 1639 case TCPS_SYN_RECEIVED: 1640 1641 tcpstat.tcps_connects++; 1642 soisconnected(so); 1643 /* Do window scaling? */ 1644 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1645 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1646 tp->snd_scale = tp->requested_s_scale; 1647 tp->rcv_scale = tp->request_r_scale; 1648 } 1649 /* 1650 * Upon successful completion of 3-way handshake, 1651 * update cache.CC if it was undefined, pass any queued 1652 * data to the user, and advance state appropriately. 1653 */ 1654 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1655 taop->tao_cc == 0) 1656 taop->tao_cc = tp->cc_recv; 1657 1658 /* 1659 * Make transitions: 1660 * SYN-RECEIVED -> ESTABLISHED 1661 * SYN-RECEIVED* -> FIN-WAIT-1 1662 */ 1663 tp->t_starttime = ticks; 1664 if (tp->t_flags & TF_NEEDFIN) { 1665 tp->t_state = TCPS_FIN_WAIT_1; 1666 tp->t_flags &= ~TF_NEEDFIN; 1667 } else { 1668 tp->t_state = TCPS_ESTABLISHED; 1669 callout_reset(tp->tt_keep, tcp_keepidle, 1670 tcp_timer_keep, tp); 1671 } 1672 /* 1673 * If segment contains data or ACK, will call tcp_reass() 1674 * later; if not, do so now to pass queued data to user. 1675 */ 1676 if (tlen == 0 && (thflags & TH_FIN) == 0) 1677 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1678 (struct mbuf *)0); 1679 tp->snd_wl1 = th->th_seq - 1; 1680 /* fall into ... */ 1681 1682 /* 1683 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1684 * ACKs. If the ack is in the range 1685 * tp->snd_una < th->th_ack <= tp->snd_max 1686 * then advance tp->snd_una to th->th_ack and drop 1687 * data from the retransmission queue. If this ACK reflects 1688 * more up to date window information we update our window information. 1689 */ 1690 case TCPS_ESTABLISHED: 1691 case TCPS_FIN_WAIT_1: 1692 case TCPS_FIN_WAIT_2: 1693 case TCPS_CLOSE_WAIT: 1694 case TCPS_CLOSING: 1695 case TCPS_LAST_ACK: 1696 case TCPS_TIME_WAIT: 1697 1698 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1699 if (tlen == 0 && tiwin == tp->snd_wnd) { 1700 tcpstat.tcps_rcvdupack++; 1701 /* 1702 * If we have outstanding data (other than 1703 * a window probe), this is a completely 1704 * duplicate ack (ie, window info didn't 1705 * change), the ack is the biggest we've 1706 * seen and we've seen exactly our rexmt 1707 * threshhold of them, assume a packet 1708 * has been dropped and retransmit it. 1709 * Kludge snd_nxt & the congestion 1710 * window so we send only this one 1711 * packet. 1712 * 1713 * We know we're losing at the current 1714 * window size so do congestion avoidance 1715 * (set ssthresh to half the current window 1716 * and pull our congestion window back to 1717 * the new ssthresh). 1718 * 1719 * Dup acks mean that packets have left the 1720 * network (they're now cached at the receiver) 1721 * so bump cwnd by the amount in the receiver 1722 * to keep a constant cwnd packets in the 1723 * network. 1724 */ 1725 if (!callout_active(tp->tt_rexmt) || 1726 th->th_ack != tp->snd_una) 1727 tp->t_dupacks = 0; 1728 else if (++tp->t_dupacks > tcprexmtthresh || 1729 (tcp_do_newreno && 1730 IN_FASTRECOVERY(tp))) { 1731 tp->snd_cwnd += tp->t_maxseg; 1732 (void) tcp_output(tp); 1733 goto drop; 1734 } else if (tp->t_dupacks == tcprexmtthresh) { 1735 tcp_seq onxt; 1736 u_int win; 1737 1738 if (tcp_do_newreno && 1739 SEQ_LEQ(th->th_ack, 1740 tp->snd_recover)) { 1741 tp->t_dupacks = 0; 1742 break; 1743 } 1744 fastretransmit: 1745 if (tcp_do_eifel_detect && 1746 (tp->t_flags & TF_RCVD_TSTMP)) { 1747 tcp_save_congestion_state(tp); 1748 tp->t_flags |= TF_FASTREXMT; 1749 } 1750 win = min(tp->snd_wnd, tp->snd_cwnd) / 1751 2 / tp->t_maxseg; 1752 if (win < 2) 1753 win = 2; 1754 tp->snd_ssthresh = win * tp->t_maxseg; 1755 ENTER_FASTRECOVERY(tp); 1756 tp->snd_recover = tp->snd_max; 1757 callout_stop(tp->tt_rexmt); 1758 tp->t_rtttime = 0; 1759 onxt = tp->snd_nxt; 1760 tp->snd_nxt = th->th_ack; 1761 tp->snd_cwnd = tp->t_maxseg; 1762 (void) tcp_output(tp); 1763 ++tcpstat.tcps_sndfastrexmit; 1764 KASSERT(tp->snd_limited <= 2, 1765 ("tp->snd_limited too big")); 1766 tp->snd_cwnd = tp->snd_ssthresh + 1767 (tp->t_maxseg * 1768 (tp->t_dupacks - tp->snd_limited)); 1769 if (SEQ_GT(onxt, tp->snd_nxt)) 1770 tp->snd_nxt = onxt; 1771 goto drop; 1772 } else if (tcp_do_limitedtransmit) { 1773 u_long oldcwnd = tp->snd_cwnd; 1774 tcp_seq oldsndmax = tp->snd_max; 1775 /* outstanding data */ 1776 uint32_t ownd = 1777 tp->snd_max - tp->snd_una; 1778 u_int sent; 1779 1780 #define iceildiv(n, d) (((n)+(d)-1) / (d)) 1781 1782 KASSERT(tp->t_dupacks == 1 || 1783 tp->t_dupacks == 2, 1784 ("dupacks not 1 or 2")); 1785 if (tp->t_dupacks == 1) 1786 tp->snd_limited = 0; 1787 tp->snd_cwnd = ownd + 1788 (tp->t_dupacks - tp->snd_limited) * 1789 tp->t_maxseg; 1790 (void) tcp_output(tp); 1791 tp->snd_cwnd = oldcwnd; 1792 sent = tp->snd_max - oldsndmax; 1793 if (sent > tp->t_maxseg) { 1794 KASSERT((tp->t_dupacks == 2 && 1795 tp->snd_limited == 0) || 1796 (sent == tp->t_maxseg + 1 && 1797 tp->t_flags & TF_SENTFIN), 1798 ("sent too much")); 1799 KASSERT(sent <= 1800 tp->t_maxseg * 2, 1801 ("sent too many segments")); 1802 tp->snd_limited = 2; 1803 tcpstat.tcps_sndlimited += 2; 1804 } else if (sent > 0) { 1805 ++tp->snd_limited; 1806 ++tcpstat.tcps_sndlimited; 1807 } else if (tcp_do_early_retransmit && 1808 (tcp_do_eifel_detect && 1809 (tp->t_flags & TF_RCVD_TSTMP)) && 1810 tcp_do_newreno && 1811 tp->t_dupacks + 1 >= 1812 iceildiv(ownd, tp->t_maxseg)) { 1813 ++tcpstat.tcps_sndearlyrexmit; 1814 tp->t_flags |= TF_EARLYREXMT; 1815 goto fastretransmit; 1816 } 1817 goto drop; 1818 } 1819 } else 1820 tp->t_dupacks = 0; 1821 break; 1822 } 1823 1824 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1825 1826 /* 1827 * If the congestion window was inflated to account 1828 * for the other side's cached packets, retract it. 1829 */ 1830 if (tcp_do_newreno) { 1831 if (IN_FASTRECOVERY(tp)) { 1832 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1833 tcp_newreno_partial_ack(tp, th); 1834 } else { 1835 /* 1836 * Window inflation should have left us 1837 * with approximately snd_ssthresh 1838 * outstanding data. 1839 * But in case we would be inclined to 1840 * send a burst, better to do it via 1841 * the slow start mechanism. 1842 */ 1843 if (SEQ_GT(th->th_ack + 1844 tp->snd_ssthresh, 1845 tp->snd_max)) 1846 tp->snd_cwnd = tp->snd_max - 1847 th->th_ack + 1848 tp->t_maxseg; 1849 else 1850 tp->snd_cwnd = tp->snd_ssthresh; 1851 } 1852 } 1853 } else { 1854 if (tp->t_dupacks >= tcprexmtthresh && 1855 tp->snd_cwnd > tp->snd_ssthresh) 1856 tp->snd_cwnd = tp->snd_ssthresh; 1857 } 1858 tp->t_dupacks = 0; 1859 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1860 tcpstat.tcps_rcvacktoomuch++; 1861 goto dropafterack; 1862 } 1863 /* 1864 * If we reach this point, ACK is not a duplicate, 1865 * i.e., it ACKs something we sent. 1866 */ 1867 if (tp->t_flags & TF_NEEDSYN) { 1868 /* 1869 * T/TCP: Connection was half-synchronized, and our 1870 * SYN has been ACK'd (so connection is now fully 1871 * synchronized). Go to non-starred state, 1872 * increment snd_una for ACK of SYN, and check if 1873 * we can do window scaling. 1874 */ 1875 tp->t_flags &= ~TF_NEEDSYN; 1876 tp->snd_una++; 1877 /* Do window scaling? */ 1878 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1879 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1880 tp->snd_scale = tp->requested_s_scale; 1881 tp->rcv_scale = tp->request_r_scale; 1882 } 1883 } 1884 1885 process_ACK: 1886 acked = th->th_ack - tp->snd_una; 1887 tcpstat.tcps_rcvackpack++; 1888 tcpstat.tcps_rcvackbyte += acked; 1889 1890 /* 1891 * If we just performed our first retransmit, and the ACK 1892 * arrives within our recovery window, then it was a mistake 1893 * to do the retransmit in the first place. Recover our 1894 * original cwnd and ssthresh, and proceed to transmit where 1895 * we left off. 1896 */ 1897 if (tcp_do_eifel_detect && acked && 1898 (to.to_flags & TOF_TS) && to.to_tsecr && 1899 (tp->t_flags & TF_FIRSTACCACK)) { 1900 /* Eifel detection applicable. */ 1901 if (to.to_tsecr < tp->t_rexmtTS) { 1902 ++tcpstat.tcps_eifeldetected; 1903 tcp_revert_congestion_state(tp); 1904 if (tp->t_rxtshift == 1 && 1905 ticks >= tp->t_badrxtwin) 1906 ++tcpstat.tcps_rttcantdetect; 1907 } 1908 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1909 tcp_revert_congestion_state(tp); 1910 ++tcpstat.tcps_rttdetected; 1911 } 1912 1913 /* 1914 * If we have a timestamp reply, update smoothed 1915 * round trip time. If no timestamp is present but 1916 * transmit timer is running and timed sequence 1917 * number was acked, update smoothed round trip time. 1918 * Since we now have an rtt measurement, cancel the 1919 * timer backoff (cf., Phil Karn's retransmit alg.). 1920 * Recompute the initial retransmit timer. 1921 * 1922 * Some machines (certain windows boxes) send broken 1923 * timestamp replies during the SYN+ACK phase, ignore 1924 * timestamps of 0. 1925 */ 1926 if ((to.to_flags & TOF_TS) != 0 && 1927 to.to_tsecr) { 1928 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1929 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1930 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1931 } 1932 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1933 1934 /* 1935 * If all outstanding data is acked, stop retransmit 1936 * timer and remember to restart (more output or persist). 1937 * If there is more data to be acked, restart retransmit 1938 * timer, using current (possibly backed-off) value. 1939 */ 1940 if (th->th_ack == tp->snd_max) { 1941 callout_stop(tp->tt_rexmt); 1942 needoutput = 1; 1943 } else if (!callout_active(tp->tt_persist)) 1944 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1945 tcp_timer_rexmt, tp); 1946 1947 /* 1948 * If no data (only SYN) was ACK'd, 1949 * skip rest of ACK processing. 1950 */ 1951 if (acked == 0) 1952 goto step6; 1953 1954 /* Stop looking for an acceptable ACK since one was received. */ 1955 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT); 1956 1957 /* 1958 * When new data is acked, open the congestion window. 1959 * If the window gives us less than ssthresh packets 1960 * in flight, open exponentially (maxseg per packet). 1961 * Otherwise open linearly: maxseg per window 1962 * (maxseg^2 / cwnd per packet). 1963 */ 1964 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 1965 u_int cw = tp->snd_cwnd; 1966 u_int incr = tp->t_maxseg; 1967 if (cw > tp->snd_ssthresh) 1968 incr = incr * incr / cw; 1969 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 1970 } 1971 if (acked > so->so_snd.sb_cc) { 1972 tp->snd_wnd -= so->so_snd.sb_cc; 1973 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1974 ourfinisacked = 1; 1975 } else { 1976 sbdrop(&so->so_snd, acked); 1977 tp->snd_wnd -= acked; 1978 ourfinisacked = 0; 1979 } 1980 sowwakeup(so); 1981 if (tcp_do_newreno) { 1982 if (IN_FASTRECOVERY(tp)) { 1983 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) 1984 EXIT_FASTRECOVERY(tp); 1985 } else { 1986 tp->snd_recover = th->th_ack - 1; 1987 } 1988 } 1989 tp->snd_una = th->th_ack; 1990 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1991 tp->snd_nxt = tp->snd_una; 1992 1993 switch (tp->t_state) { 1994 1995 /* 1996 * In FIN_WAIT_1 STATE in addition to the processing 1997 * for the ESTABLISHED state if our FIN is now acknowledged 1998 * then enter FIN_WAIT_2. 1999 */ 2000 case TCPS_FIN_WAIT_1: 2001 if (ourfinisacked) { 2002 /* 2003 * If we can't receive any more 2004 * data, then closing user can proceed. 2005 * Starting the timer is contrary to the 2006 * specification, but if we don't get a FIN 2007 * we'll hang forever. 2008 */ 2009 if (so->so_state & SS_CANTRCVMORE) { 2010 soisdisconnected(so); 2011 callout_reset(tp->tt_2msl, tcp_maxidle, 2012 tcp_timer_2msl, tp); 2013 } 2014 tp->t_state = TCPS_FIN_WAIT_2; 2015 } 2016 break; 2017 2018 /* 2019 * In CLOSING STATE in addition to the processing for 2020 * the ESTABLISHED state if the ACK acknowledges our FIN 2021 * then enter the TIME-WAIT state, otherwise ignore 2022 * the segment. 2023 */ 2024 case TCPS_CLOSING: 2025 if (ourfinisacked) { 2026 tp->t_state = TCPS_TIME_WAIT; 2027 tcp_canceltimers(tp); 2028 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2029 if (tp->cc_recv != 0 && 2030 (ticks - tp->t_starttime) < tcp_msl) 2031 callout_reset(tp->tt_2msl, 2032 tp->t_rxtcur * 2033 TCPTV_TWTRUNC, 2034 tcp_timer_2msl, tp); 2035 else 2036 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2037 tcp_timer_2msl, tp); 2038 soisdisconnected(so); 2039 } 2040 break; 2041 2042 /* 2043 * In LAST_ACK, we may still be waiting for data to drain 2044 * and/or to be acked, as well as for the ack of our FIN. 2045 * If our FIN is now acknowledged, delete the TCB, 2046 * enter the closed state and return. 2047 */ 2048 case TCPS_LAST_ACK: 2049 if (ourfinisacked) { 2050 tp = tcp_close(tp); 2051 goto drop; 2052 } 2053 break; 2054 2055 /* 2056 * In TIME_WAIT state the only thing that should arrive 2057 * is a retransmission of the remote FIN. Acknowledge 2058 * it and restart the finack timer. 2059 */ 2060 case TCPS_TIME_WAIT: 2061 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2062 tcp_timer_2msl, tp); 2063 goto dropafterack; 2064 } 2065 } 2066 2067 step6: 2068 /* 2069 * Update window information. 2070 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2071 */ 2072 if ((thflags & TH_ACK) && 2073 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2074 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2075 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2076 /* keep track of pure window updates */ 2077 if (tlen == 0 && 2078 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2079 tcpstat.tcps_rcvwinupd++; 2080 tp->snd_wnd = tiwin; 2081 tp->snd_wl1 = th->th_seq; 2082 tp->snd_wl2 = th->th_ack; 2083 if (tp->snd_wnd > tp->max_sndwnd) 2084 tp->max_sndwnd = tp->snd_wnd; 2085 needoutput = 1; 2086 } 2087 2088 /* 2089 * Process segments with URG. 2090 */ 2091 if ((thflags & TH_URG) && th->th_urp && 2092 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2093 /* 2094 * This is a kludge, but if we receive and accept 2095 * random urgent pointers, we'll crash in 2096 * soreceive. It's hard to imagine someone 2097 * actually wanting to send this much urgent data. 2098 */ 2099 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2100 th->th_urp = 0; /* XXX */ 2101 thflags &= ~TH_URG; /* XXX */ 2102 goto dodata; /* XXX */ 2103 } 2104 /* 2105 * If this segment advances the known urgent pointer, 2106 * then mark the data stream. This should not happen 2107 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2108 * a FIN has been received from the remote side. 2109 * In these states we ignore the URG. 2110 * 2111 * According to RFC961 (Assigned Protocols), 2112 * the urgent pointer points to the last octet 2113 * of urgent data. We continue, however, 2114 * to consider it to indicate the first octet 2115 * of data past the urgent section as the original 2116 * spec states (in one of two places). 2117 */ 2118 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2119 tp->rcv_up = th->th_seq + th->th_urp; 2120 so->so_oobmark = so->so_rcv.sb_cc + 2121 (tp->rcv_up - tp->rcv_nxt) - 1; 2122 if (so->so_oobmark == 0) 2123 so->so_state |= SS_RCVATMARK; 2124 sohasoutofband(so); 2125 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2126 } 2127 /* 2128 * Remove out of band data so doesn't get presented to user. 2129 * This can happen independent of advancing the URG pointer, 2130 * but if two URG's are pending at once, some out-of-band 2131 * data may creep in... ick. 2132 */ 2133 if (th->th_urp <= (u_long)tlen 2134 #ifdef SO_OOBINLINE 2135 && (so->so_options & SO_OOBINLINE) == 0 2136 #endif 2137 ) 2138 tcp_pulloutofband(so, th, m, 2139 drop_hdrlen); /* hdr drop is delayed */ 2140 } else { 2141 /* 2142 * If no out of band data is expected, 2143 * pull receive urgent pointer along 2144 * with the receive window. 2145 */ 2146 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2147 tp->rcv_up = tp->rcv_nxt; 2148 } 2149 dodata: /* XXX */ 2150 2151 /* 2152 * Process the segment text, merging it into the TCP sequencing queue, 2153 * and arranging for acknowledgment of receipt if necessary. 2154 * This process logically involves adjusting tp->rcv_wnd as data 2155 * is presented to the user (this happens in tcp_usrreq.c, 2156 * case PRU_RCVD). If a FIN has already been received on this 2157 * connection then we just ignore the text. 2158 */ 2159 if ((tlen || (thflags & TH_FIN)) && 2160 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2161 m_adj(m, drop_hdrlen); /* delayed header drop */ 2162 /* 2163 * Insert segment which includes th into TCP reassembly queue 2164 * with control block tp. Set thflags to whether reassembly now 2165 * includes a segment with FIN. This handles the common case 2166 * inline (segment is the next to be received on an established 2167 * connection, and the queue is empty), avoiding linkage into 2168 * and removal from the queue and repetition of various 2169 * conversions. 2170 * Set DELACK for segments received in order, but ack 2171 * immediately when segments are out of order (so 2172 * fast retransmit can work). 2173 */ 2174 if (th->th_seq == tp->rcv_nxt && 2175 LIST_EMPTY(&tp->t_segq) && 2176 TCPS_HAVEESTABLISHED(tp->t_state)) { 2177 if (DELAY_ACK(tp)) 2178 callout_reset(tp->tt_delack, tcp_delacktime, 2179 tcp_timer_delack, tp); 2180 else 2181 tp->t_flags |= TF_ACKNOW; 2182 tp->rcv_nxt += tlen; 2183 thflags = th->th_flags & TH_FIN; 2184 tcpstat.tcps_rcvpack++; 2185 tcpstat.tcps_rcvbyte += tlen; 2186 ND6_HINT(tp); 2187 if (so->so_state & SS_CANTRCVMORE) 2188 m_freem(m); 2189 else 2190 sbappend(&so->so_rcv, m); 2191 sorwakeup(so); 2192 } else { 2193 thflags = tcp_reass(tp, th, &tlen, m); 2194 tp->t_flags |= TF_ACKNOW; 2195 } 2196 2197 /* 2198 * Note the amount of data that peer has sent into 2199 * our window, in order to estimate the sender's 2200 * buffer size. 2201 */ 2202 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2203 } else { 2204 m_freem(m); 2205 thflags &= ~TH_FIN; 2206 } 2207 2208 /* 2209 * If FIN is received ACK the FIN and let the user know 2210 * that the connection is closing. 2211 */ 2212 if (thflags & TH_FIN) { 2213 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2214 socantrcvmore(so); 2215 /* 2216 * If connection is half-synchronized 2217 * (ie NEEDSYN flag on) then delay ACK, 2218 * so it may be piggybacked when SYN is sent. 2219 * Otherwise, since we received a FIN then no 2220 * more input can be expected, send ACK now. 2221 */ 2222 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2223 callout_reset(tp->tt_delack, tcp_delacktime, 2224 tcp_timer_delack, tp); 2225 else 2226 tp->t_flags |= TF_ACKNOW; 2227 tp->rcv_nxt++; 2228 } 2229 switch (tp->t_state) { 2230 2231 /* 2232 * In SYN_RECEIVED and ESTABLISHED STATES 2233 * enter the CLOSE_WAIT state. 2234 */ 2235 case TCPS_SYN_RECEIVED: 2236 tp->t_starttime = ticks; 2237 /*FALLTHROUGH*/ 2238 case TCPS_ESTABLISHED: 2239 tp->t_state = TCPS_CLOSE_WAIT; 2240 break; 2241 2242 /* 2243 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2244 * enter the CLOSING state. 2245 */ 2246 case TCPS_FIN_WAIT_1: 2247 tp->t_state = TCPS_CLOSING; 2248 break; 2249 2250 /* 2251 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2252 * starting the time-wait timer, turning off the other 2253 * standard timers. 2254 */ 2255 case TCPS_FIN_WAIT_2: 2256 tp->t_state = TCPS_TIME_WAIT; 2257 tcp_canceltimers(tp); 2258 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2259 if (tp->cc_recv != 0 && 2260 (ticks - tp->t_starttime) < tcp_msl) { 2261 callout_reset(tp->tt_2msl, 2262 tp->t_rxtcur * TCPTV_TWTRUNC, 2263 tcp_timer_2msl, tp); 2264 /* For transaction client, force ACK now. */ 2265 tp->t_flags |= TF_ACKNOW; 2266 } 2267 else 2268 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2269 tcp_timer_2msl, tp); 2270 soisdisconnected(so); 2271 break; 2272 2273 /* 2274 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2275 */ 2276 case TCPS_TIME_WAIT: 2277 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2278 tcp_timer_2msl, tp); 2279 break; 2280 } 2281 } 2282 #ifdef TCPDEBUG 2283 if (so->so_options & SO_DEBUG) 2284 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2285 &tcp_savetcp, 0); 2286 #endif 2287 2288 /* 2289 * Return any desired output. 2290 */ 2291 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2292 (void) tcp_output(tp); 2293 return; 2294 2295 dropafterack: 2296 /* 2297 * Generate an ACK dropping incoming segment if it occupies 2298 * sequence space, where the ACK reflects our state. 2299 * 2300 * We can now skip the test for the RST flag since all 2301 * paths to this code happen after packets containing 2302 * RST have been dropped. 2303 * 2304 * In the SYN-RECEIVED state, don't send an ACK unless the 2305 * segment we received passes the SYN-RECEIVED ACK test. 2306 * If it fails send a RST. This breaks the loop in the 2307 * "LAND" DoS attack, and also prevents an ACK storm 2308 * between two listening ports that have been sent forged 2309 * SYN segments, each with the source address of the other. 2310 */ 2311 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2312 (SEQ_GT(tp->snd_una, th->th_ack) || 2313 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2314 rstreason = BANDLIM_RST_OPENPORT; 2315 goto dropwithreset; 2316 } 2317 #ifdef TCPDEBUG 2318 if (so->so_options & SO_DEBUG) 2319 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2320 &tcp_savetcp, 0); 2321 #endif 2322 m_freem(m); 2323 tp->t_flags |= TF_ACKNOW; 2324 (void) tcp_output(tp); 2325 return; 2326 2327 dropwithreset: 2328 /* 2329 * Generate a RST, dropping incoming segment. 2330 * Make ACK acceptable to originator of segment. 2331 * Don't bother to respond if destination was broadcast/multicast. 2332 */ 2333 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2334 goto drop; 2335 if (isipv6) { 2336 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2337 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2338 goto drop; 2339 } else { 2340 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2341 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2342 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2343 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2344 goto drop; 2345 } 2346 /* IPv6 anycast check is done at tcp6_input() */ 2347 2348 /* 2349 * Perform bandwidth limiting. 2350 */ 2351 #ifdef ICMP_BANDLIM 2352 if (badport_bandlim(rstreason) < 0) 2353 goto drop; 2354 #endif 2355 2356 #ifdef TCPDEBUG 2357 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2358 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2359 &tcp_savetcp, 0); 2360 #endif 2361 if (thflags & TH_ACK) 2362 /* mtod() below is safe as long as hdr dropping is delayed */ 2363 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2364 TH_RST); 2365 else { 2366 if (thflags & TH_SYN) 2367 tlen++; 2368 /* mtod() below is safe as long as hdr dropping is delayed */ 2369 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2370 (tcp_seq)0, TH_RST|TH_ACK); 2371 } 2372 return; 2373 2374 drop: 2375 /* 2376 * Drop space held by incoming segment and return. 2377 */ 2378 #ifdef TCPDEBUG 2379 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2380 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2381 &tcp_savetcp, 0); 2382 #endif 2383 m_freem(m); 2384 return; 2385 } 2386 2387 /* 2388 * Parse TCP options and place in tcpopt. 2389 */ 2390 static void 2391 tcp_dooptions(to, cp, cnt, is_syn) 2392 struct tcpopt *to; 2393 u_char *cp; 2394 int cnt; 2395 { 2396 int opt, optlen; 2397 2398 to->to_flags = 0; 2399 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2400 opt = cp[0]; 2401 if (opt == TCPOPT_EOL) 2402 break; 2403 if (opt == TCPOPT_NOP) 2404 optlen = 1; 2405 else { 2406 if (cnt < 2) 2407 break; 2408 optlen = cp[1]; 2409 if (optlen < 2 || optlen > cnt) 2410 break; 2411 } 2412 switch (opt) { 2413 case TCPOPT_MAXSEG: 2414 if (optlen != TCPOLEN_MAXSEG) 2415 continue; 2416 if (!is_syn) 2417 continue; 2418 to->to_flags |= TOF_MSS; 2419 bcopy((char *)cp + 2, 2420 (char *)&to->to_mss, sizeof(to->to_mss)); 2421 to->to_mss = ntohs(to->to_mss); 2422 break; 2423 case TCPOPT_WINDOW: 2424 if (optlen != TCPOLEN_WINDOW) 2425 continue; 2426 if (! is_syn) 2427 continue; 2428 to->to_flags |= TOF_SCALE; 2429 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2430 break; 2431 case TCPOPT_TIMESTAMP: 2432 if (optlen != TCPOLEN_TIMESTAMP) 2433 continue; 2434 to->to_flags |= TOF_TS; 2435 bcopy((char *)cp + 2, 2436 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2437 to->to_tsval = ntohl(to->to_tsval); 2438 bcopy((char *)cp + 6, 2439 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2440 to->to_tsecr = ntohl(to->to_tsecr); 2441 break; 2442 case TCPOPT_CC: 2443 if (optlen != TCPOLEN_CC) 2444 continue; 2445 to->to_flags |= TOF_CC; 2446 bcopy((char *)cp + 2, 2447 (char *)&to->to_cc, sizeof(to->to_cc)); 2448 to->to_cc = ntohl(to->to_cc); 2449 break; 2450 case TCPOPT_CCNEW: 2451 if (optlen != TCPOLEN_CC) 2452 continue; 2453 if (!is_syn) 2454 continue; 2455 to->to_flags |= TOF_CCNEW; 2456 bcopy((char *)cp + 2, 2457 (char *)&to->to_cc, sizeof(to->to_cc)); 2458 to->to_cc = ntohl(to->to_cc); 2459 break; 2460 case TCPOPT_CCECHO: 2461 if (optlen != TCPOLEN_CC) 2462 continue; 2463 if (!is_syn) 2464 continue; 2465 to->to_flags |= TOF_CCECHO; 2466 bcopy((char *)cp + 2, 2467 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2468 to->to_ccecho = ntohl(to->to_ccecho); 2469 break; 2470 default: 2471 continue; 2472 } 2473 } 2474 } 2475 2476 /* 2477 * Pull out of band byte out of a segment so 2478 * it doesn't appear in the user's data queue. 2479 * It is still reflected in the segment length for 2480 * sequencing purposes. 2481 */ 2482 static void 2483 tcp_pulloutofband(so, th, m, off) 2484 struct socket *so; 2485 struct tcphdr *th; 2486 struct mbuf *m; 2487 int off; /* delayed to be droped hdrlen */ 2488 { 2489 int cnt = off + th->th_urp - 1; 2490 2491 while (cnt >= 0) { 2492 if (m->m_len > cnt) { 2493 char *cp = mtod(m, caddr_t) + cnt; 2494 struct tcpcb *tp = sototcpcb(so); 2495 2496 tp->t_iobc = *cp; 2497 tp->t_oobflags |= TCPOOB_HAVEDATA; 2498 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2499 m->m_len--; 2500 if (m->m_flags & M_PKTHDR) 2501 m->m_pkthdr.len--; 2502 return; 2503 } 2504 cnt -= m->m_len; 2505 m = m->m_next; 2506 if (m == 0) 2507 break; 2508 } 2509 panic("tcp_pulloutofband"); 2510 } 2511 2512 /* 2513 * Collect new round-trip time estimate 2514 * and update averages and current timeout. 2515 */ 2516 static void 2517 tcp_xmit_timer(tp, rtt) 2518 struct tcpcb *tp; 2519 int rtt; 2520 { 2521 int delta; 2522 2523 tcpstat.tcps_rttupdated++; 2524 tp->t_rttupdated++; 2525 if (tp->t_srtt != 0) { 2526 /* 2527 * srtt is stored as fixed point with 5 bits after the 2528 * binary point (i.e., scaled by 8). The following magic 2529 * is equivalent to the smoothing algorithm in rfc793 with 2530 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2531 * point). Adjust rtt to origin 0. 2532 */ 2533 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2534 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2535 2536 if ((tp->t_srtt += delta) <= 0) 2537 tp->t_srtt = 1; 2538 2539 /* 2540 * We accumulate a smoothed rtt variance (actually, a 2541 * smoothed mean difference), then set the retransmit 2542 * timer to smoothed rtt + 4 times the smoothed variance. 2543 * rttvar is stored as fixed point with 4 bits after the 2544 * binary point (scaled by 16). The following is 2545 * equivalent to rfc793 smoothing with an alpha of .75 2546 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2547 * rfc793's wired-in beta. 2548 */ 2549 if (delta < 0) 2550 delta = -delta; 2551 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2552 if ((tp->t_rttvar += delta) <= 0) 2553 tp->t_rttvar = 1; 2554 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2555 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2556 } else { 2557 /* 2558 * No rtt measurement yet - use the unsmoothed rtt. 2559 * Set the variance to half the rtt (so our first 2560 * retransmit happens at 3*rtt). 2561 */ 2562 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2563 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2564 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2565 } 2566 tp->t_rtttime = 0; 2567 tp->t_rxtshift = 0; 2568 2569 /* 2570 * the retransmit should happen at rtt + 4 * rttvar. 2571 * Because of the way we do the smoothing, srtt and rttvar 2572 * will each average +1/2 tick of bias. When we compute 2573 * the retransmit timer, we want 1/2 tick of rounding and 2574 * 1 extra tick because of +-1/2 tick uncertainty in the 2575 * firing of the timer. The bias will give us exactly the 2576 * 1.5 tick we need. But, because the bias is 2577 * statistical, we have to test that we don't drop below 2578 * the minimum feasible timer (which is 2 ticks). 2579 */ 2580 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2581 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2582 2583 /* 2584 * We received an ack for a packet that wasn't retransmitted; 2585 * it is probably safe to discard any error indications we've 2586 * received recently. This isn't quite right, but close enough 2587 * for now (a route might have failed after we sent a segment, 2588 * and the return path might not be symmetrical). 2589 */ 2590 tp->t_softerror = 0; 2591 } 2592 2593 /* 2594 * Determine a reasonable value for maxseg size. 2595 * If the route is known, check route for mtu. 2596 * If none, use an mss that can be handled on the outgoing 2597 * interface without forcing IP to fragment; if bigger than 2598 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2599 * to utilize large mbufs. If no route is found, route has no mtu, 2600 * or the destination isn't local, use a default, hopefully conservative 2601 * size (usually 512 or the default IP max size, but no more than the mtu 2602 * of the interface), as we can't discover anything about intervening 2603 * gateways or networks. We also initialize the congestion/slow start 2604 * window to be a single segment if the destination isn't local. 2605 * While looking at the routing entry, we also initialize other path-dependent 2606 * parameters from pre-set or cached values in the routing entry. 2607 * 2608 * Also take into account the space needed for options that we 2609 * send regularly. Make maxseg shorter by that amount to assure 2610 * that we can send maxseg amount of data even when the options 2611 * are present. Store the upper limit of the length of options plus 2612 * data in maxopd. 2613 * 2614 * NOTE that this routine is only called when we process an incoming 2615 * segment, for outgoing segments only tcp_mssopt is called. 2616 * 2617 * In case of T/TCP, we call this routine during implicit connection 2618 * setup as well (offer = -1), to initialize maxseg from the cached 2619 * MSS of our peer. 2620 */ 2621 void 2622 tcp_mss(tp, offer) 2623 struct tcpcb *tp; 2624 int offer; 2625 { 2626 struct rtentry *rt; 2627 struct ifnet *ifp; 2628 int rtt, mss; 2629 u_long bufsize; 2630 struct inpcb *inp = tp->t_inpcb; 2631 struct socket *so; 2632 struct rmxp_tao *taop; 2633 int origoffer = offer; 2634 #ifdef INET6 2635 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE); 2636 size_t min_protoh = isipv6 ? 2637 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2638 sizeof(struct tcpiphdr); 2639 #else 2640 const boolean_t isipv6 = FALSE; 2641 const size_t min_protoh = sizeof(struct tcpiphdr); 2642 #endif 2643 2644 if (isipv6) 2645 rt = tcp_rtlookup6(&inp->inp_inc); 2646 else 2647 rt = tcp_rtlookup(&inp->inp_inc); 2648 if (rt == NULL) { 2649 tp->t_maxopd = tp->t_maxseg = 2650 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2651 return; 2652 } 2653 ifp = rt->rt_ifp; 2654 so = inp->inp_socket; 2655 2656 taop = rmx_taop(rt->rt_rmx); 2657 /* 2658 * Offer == -1 means that we didn't receive SYN yet, 2659 * use cached value in that case; 2660 */ 2661 if (offer == -1) 2662 offer = taop->tao_mssopt; 2663 /* 2664 * Offer == 0 means that there was no MSS on the SYN segment, 2665 * in this case we use tcp_mssdflt. 2666 */ 2667 if (offer == 0) 2668 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2669 else 2670 /* 2671 * Sanity check: make sure that maxopd will be large 2672 * enough to allow some data on segments even is the 2673 * all the option space is used (40bytes). Otherwise 2674 * funny things may happen in tcp_output. 2675 */ 2676 offer = max(offer, 64); 2677 taop->tao_mssopt = offer; 2678 2679 /* 2680 * While we're here, check if there's an initial rtt 2681 * or rttvar. Convert from the route-table units 2682 * to scaled multiples of the slow timeout timer. 2683 */ 2684 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2685 /* 2686 * XXX the lock bit for RTT indicates that the value 2687 * is also a minimum value; this is subject to time. 2688 */ 2689 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2690 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2691 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2692 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2693 tcpstat.tcps_usedrtt++; 2694 if (rt->rt_rmx.rmx_rttvar) { 2695 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2696 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2697 tcpstat.tcps_usedrttvar++; 2698 } else { 2699 /* default variation is +- 1 rtt */ 2700 tp->t_rttvar = 2701 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2702 } 2703 TCPT_RANGESET(tp->t_rxtcur, 2704 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2705 tp->t_rttmin, TCPTV_REXMTMAX); 2706 } 2707 /* 2708 * if there's an mtu associated with the route, use it 2709 * else, use the link mtu. 2710 */ 2711 if (rt->rt_rmx.rmx_mtu) 2712 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2713 else { 2714 if (isipv6) { 2715 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2716 min_protoh; 2717 if (!in6_localaddr(&inp->in6p_faddr)) 2718 mss = min(mss, tcp_v6mssdflt); 2719 } else { 2720 mss = ifp->if_mtu - min_protoh; 2721 if (!in_localaddr(inp->inp_faddr)) 2722 mss = min(mss, tcp_mssdflt); 2723 } 2724 } 2725 mss = min(mss, offer); 2726 /* 2727 * maxopd stores the maximum length of data AND options 2728 * in a segment; maxseg is the amount of data in a normal 2729 * segment. We need to store this value (maxopd) apart 2730 * from maxseg, because now every segment carries options 2731 * and thus we normally have somewhat less data in segments. 2732 */ 2733 tp->t_maxopd = mss; 2734 2735 /* 2736 * In case of T/TCP, origoffer==-1 indicates, that no segments 2737 * were received yet. In this case we just guess, otherwise 2738 * we do the same as before T/TCP. 2739 */ 2740 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2741 (origoffer == -1 || 2742 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2743 mss -= TCPOLEN_TSTAMP_APPA; 2744 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2745 (origoffer == -1 || 2746 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2747 mss -= TCPOLEN_CC_APPA; 2748 2749 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2750 if (mss > MCLBYTES) 2751 mss &= ~(MCLBYTES-1); 2752 #else 2753 if (mss > MCLBYTES) 2754 mss = mss / MCLBYTES * MCLBYTES; 2755 #endif 2756 /* 2757 * If there's a pipesize, change the socket buffer 2758 * to that size. Make the socket buffers an integral 2759 * number of mss units; if the mss is larger than 2760 * the socket buffer, decrease the mss. 2761 */ 2762 #ifdef RTV_SPIPE 2763 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2764 #endif 2765 bufsize = so->so_snd.sb_hiwat; 2766 if (bufsize < mss) 2767 mss = bufsize; 2768 else { 2769 bufsize = roundup(bufsize, mss); 2770 if (bufsize > sb_max) 2771 bufsize = sb_max; 2772 if (bufsize > so->so_snd.sb_hiwat) 2773 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2774 } 2775 tp->t_maxseg = mss; 2776 2777 #ifdef RTV_RPIPE 2778 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2779 #endif 2780 bufsize = so->so_rcv.sb_hiwat; 2781 if (bufsize > mss) { 2782 bufsize = roundup(bufsize, mss); 2783 if (bufsize > sb_max) 2784 bufsize = sb_max; 2785 if (bufsize > so->so_rcv.sb_hiwat) 2786 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2787 } 2788 2789 /* 2790 * Set the slow-start flight size depending on whether this 2791 * is a local network or not. 2792 */ 2793 if (tcp_do_rfc3390) 2794 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2795 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2796 (!isipv6 && in_localaddr(inp->inp_faddr))) 2797 tp->snd_cwnd = mss * ss_fltsz_local; 2798 else 2799 tp->snd_cwnd = mss * ss_fltsz; 2800 2801 if (rt->rt_rmx.rmx_ssthresh) { 2802 /* 2803 * There's some sort of gateway or interface 2804 * buffer limit on the path. Use this to set 2805 * the slow start threshhold, but set the 2806 * threshold to no less than 2*mss. 2807 */ 2808 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2809 tcpstat.tcps_usedssthresh++; 2810 } 2811 } 2812 2813 /* 2814 * Determine the MSS option to send on an outgoing SYN. 2815 */ 2816 int 2817 tcp_mssopt(tp) 2818 struct tcpcb *tp; 2819 { 2820 struct rtentry *rt; 2821 #ifdef INET6 2822 boolean_t isipv6 = 2823 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE); 2824 int min_protoh = isipv6 ? 2825 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2826 sizeof(struct tcpiphdr); 2827 #else 2828 const boolean_t isipv6 = FALSE; 2829 const size_t min_protoh = sizeof(struct tcpiphdr); 2830 #endif 2831 2832 if (isipv6) 2833 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2834 else 2835 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2836 if (rt == NULL) 2837 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2838 2839 return (rt->rt_ifp->if_mtu - min_protoh); 2840 } 2841 2842 2843 /* 2844 * When a partial ack arrives, force the retransmission of the 2845 * next unacknowledged segment. Do not clear tp->t_dupacks. 2846 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2847 * be started again. 2848 */ 2849 static void 2850 tcp_newreno_partial_ack(tp, th) 2851 struct tcpcb *tp; 2852 struct tcphdr *th; 2853 { 2854 tcp_seq onxt = tp->snd_nxt; 2855 u_long ocwnd = tp->snd_cwnd; 2856 2857 callout_stop(tp->tt_rexmt); 2858 tp->t_rtttime = 0; 2859 tp->snd_nxt = th->th_ack; 2860 /* 2861 * Set snd_cwnd to one segment beyond acknowledged offset 2862 * (tp->snd_una has not yet been updated when this function is called.) 2863 */ 2864 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2865 tp->t_flags |= TF_ACKNOW; 2866 (void) tcp_output(tp); 2867 tp->snd_cwnd = ocwnd; 2868 if (SEQ_GT(onxt, tp->snd_nxt)) 2869 tp->snd_nxt = onxt; 2870 /* 2871 * Partial window deflation. Relies on fact that tp->snd_una 2872 * not updated yet. 2873 */ 2874 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2875 } 2876