tcp_input.c revision 107961
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: head/sys/netinet/tcp_input.c 107961 2002-12-17 00:24:48Z dillon $ 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 100MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 101 102static const int tcprexmtthresh = 3; 103tcp_cc tcp_ccgen; 104 105struct tcpstat tcpstat; 106SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 107 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 108 109static int log_in_vain = 0; 110SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 111 &log_in_vain, 0, "Log all incoming TCP connections"); 112 113static int blackhole = 0; 114SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 115 &blackhole, 0, "Do not send RST when dropping refused connections"); 116 117int tcp_delack_enabled = 1; 118SYSCTL_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 123static int drop_synfin = 0; 124SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 125 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 126#endif 127 128struct inpcbhead tcb; 129#define tcb6 tcb /* for KAME src sync over BSD*'s */ 130struct inpcbinfo tcbinfo; 131struct mtx *tcbinfo_mtx; 132 133static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 134static void tcp_pulloutofband(struct socket *, 135 struct tcphdr *, struct mbuf *, int); 136static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 137 struct mbuf *); 138static void tcp_xmit_timer(struct tcpcb *, int); 139static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 140 141/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 142#ifdef INET6 143#define ND6_HINT(tp) \ 144do { \ 145 if ((tp) && (tp)->t_inpcb && \ 146 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 147 (tp)->t_inpcb->in6p_route.ro_rt) \ 148 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 149} while (0) 150#else 151#define ND6_HINT(tp) 152#endif 153 154/* 155 * Indicate whether this ack should be delayed. We can delay the ack if 156 * - delayed acks are enabled and 157 * - there is no delayed ack timer in progress and 158 * - our last ack wasn't a 0-sized window. We never want to delay 159 * the ack that opens up a 0-sized window. 160 */ 161#define DELAY_ACK(tp) \ 162 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 163 (tp->t_flags & TF_RXWIN0SENT) == 0) 164 165static int 166tcp_reass(tp, th, tlenp, m) 167 register struct tcpcb *tp; 168 register struct tcphdr *th; 169 int *tlenp; 170 struct mbuf *m; 171{ 172 struct tseg_qent *q; 173 struct tseg_qent *p = NULL; 174 struct tseg_qent *nq; 175 struct tseg_qent *te; 176 struct socket *so = tp->t_inpcb->inp_socket; 177 int flags; 178 179 /* 180 * Call with th==0 after become established to 181 * force pre-ESTABLISHED data up to user socket. 182 */ 183 if (th == 0) 184 goto present; 185 186 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 187 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 188 M_NOWAIT); 189 if (te == NULL) { 190 tcpstat.tcps_rcvmemdrop++; 191 m_freem(m); 192 return (0); 193 } 194 195 /* 196 * Find a segment which begins after this one does. 197 */ 198 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 199 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 200 break; 201 p = q; 202 } 203 204 /* 205 * If there is a preceding segment, it may provide some of 206 * our data already. If so, drop the data from the incoming 207 * segment. If it provides all of our data, drop us. 208 */ 209 if (p != NULL) { 210 register int i; 211 /* conversion to int (in i) handles seq wraparound */ 212 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 213 if (i > 0) { 214 if (i >= *tlenp) { 215 tcpstat.tcps_rcvduppack++; 216 tcpstat.tcps_rcvdupbyte += *tlenp; 217 m_freem(m); 218 FREE(te, M_TSEGQ); 219 /* 220 * Try to present any queued data 221 * at the left window edge to the user. 222 * This is needed after the 3-WHS 223 * completes. 224 */ 225 goto present; /* ??? */ 226 } 227 m_adj(m, i); 228 *tlenp -= i; 229 th->th_seq += i; 230 } 231 } 232 tcpstat.tcps_rcvoopack++; 233 tcpstat.tcps_rcvoobyte += *tlenp; 234 235 /* 236 * While we overlap succeeding segments trim them or, 237 * if they are completely covered, dequeue them. 238 */ 239 while (q) { 240 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 241 if (i <= 0) 242 break; 243 if (i < q->tqe_len) { 244 q->tqe_th->th_seq += i; 245 q->tqe_len -= i; 246 m_adj(q->tqe_m, i); 247 break; 248 } 249 250 nq = LIST_NEXT(q, tqe_q); 251 LIST_REMOVE(q, tqe_q); 252 m_freem(q->tqe_m); 253 FREE(q, M_TSEGQ); 254 q = nq; 255 } 256 257 /* Insert the new segment queue entry into place. */ 258 te->tqe_m = m; 259 te->tqe_th = th; 260 te->tqe_len = *tlenp; 261 262 if (p == NULL) { 263 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 264 } else { 265 LIST_INSERT_AFTER(p, te, tqe_q); 266 } 267 268present: 269 /* 270 * Present data to user, advancing rcv_nxt through 271 * completed sequence space. 272 */ 273 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 274 return (0); 275 q = LIST_FIRST(&tp->t_segq); 276 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 277 return (0); 278 do { 279 tp->rcv_nxt += q->tqe_len; 280 flags = q->tqe_th->th_flags & TH_FIN; 281 nq = LIST_NEXT(q, tqe_q); 282 LIST_REMOVE(q, tqe_q); 283 if (so->so_state & SS_CANTRCVMORE) 284 m_freem(q->tqe_m); 285 else 286 sbappend(&so->so_rcv, q->tqe_m); 287 FREE(q, M_TSEGQ); 288 q = nq; 289 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 290 ND6_HINT(tp); 291 sorwakeup(so); 292 return (flags); 293} 294 295/* 296 * TCP input routine, follows pages 65-76 of the 297 * protocol specification dated September, 1981 very closely. 298 */ 299#ifdef INET6 300int 301tcp6_input(mp, offp, proto) 302 struct mbuf **mp; 303 int *offp, proto; 304{ 305 register struct mbuf *m = *mp; 306 struct in6_ifaddr *ia6; 307 308 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 309 310 /* 311 * draft-itojun-ipv6-tcp-to-anycast 312 * better place to put this in? 313 */ 314 ia6 = ip6_getdstifaddr(m); 315 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 316 struct ip6_hdr *ip6; 317 318 ip6 = mtod(m, struct ip6_hdr *); 319 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 320 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 321 return IPPROTO_DONE; 322 } 323 324 tcp_input(m, *offp); 325 return IPPROTO_DONE; 326} 327#endif 328 329void 330tcp_input(m, off0) 331 register struct mbuf *m; 332 int off0; 333{ 334 register struct tcphdr *th; 335 register struct ip *ip = NULL; 336 register struct ipovly *ipov; 337 register struct inpcb *inp = NULL; 338 u_char *optp = NULL; 339 int optlen = 0; 340 int len, tlen, off; 341 int drop_hdrlen; 342 register struct tcpcb *tp = 0; 343 register int thflags; 344 struct socket *so = 0; 345 int todrop, acked, ourfinisacked, needoutput = 0; 346 u_long tiwin; 347 struct tcpopt to; /* options in this segment */ 348 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 349 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 350 int headlocked = 0; 351 struct sockaddr_in *next_hop = NULL; 352 int rstreason; /* For badport_bandlim accounting purposes */ 353 354 struct ip6_hdr *ip6 = NULL; 355#ifdef INET6 356 int isipv6; 357#else 358 const int isipv6 = 0; 359#endif 360 361#ifdef TCPDEBUG 362 /* 363 * The size of tcp_saveipgen must be the size of the max ip header, 364 * now IPv6. 365 */ 366 u_char tcp_saveipgen[40]; 367 struct tcphdr tcp_savetcp; 368 short ostate = 0; 369#endif 370 371#ifdef MAC 372 int error; 373#endif 374 375 /* Grab info from MT_TAG mbufs prepended to the chain. */ 376 for (;m && m->m_type == MT_TAG; m = m->m_next) { 377 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 378 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 379 } 380#ifdef INET6 381 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 382#endif 383 bzero((char *)&to, sizeof(to)); 384 385 tcpstat.tcps_rcvtotal++; 386 387 if (isipv6) { 388 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 389 ip6 = mtod(m, struct ip6_hdr *); 390 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 391 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 392 tcpstat.tcps_rcvbadsum++; 393 goto drop; 394 } 395 th = (struct tcphdr *)((caddr_t)ip6 + off0); 396 397 /* 398 * Be proactive about unspecified IPv6 address in source. 399 * As we use all-zero to indicate unbounded/unconnected pcb, 400 * unspecified IPv6 address can be used to confuse us. 401 * 402 * Note that packets with unspecified IPv6 destination is 403 * already dropped in ip6_input. 404 */ 405 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 406 /* XXX stat */ 407 goto drop; 408 } 409 } else { 410 /* 411 * Get IP and TCP header together in first mbuf. 412 * Note: IP leaves IP header in first mbuf. 413 */ 414 if (off0 > sizeof (struct ip)) { 415 ip_stripoptions(m, (struct mbuf *)0); 416 off0 = sizeof(struct ip); 417 } 418 if (m->m_len < sizeof (struct tcpiphdr)) { 419 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 420 tcpstat.tcps_rcvshort++; 421 return; 422 } 423 } 424 ip = mtod(m, struct ip *); 425 ipov = (struct ipovly *)ip; 426 th = (struct tcphdr *)((caddr_t)ip + off0); 427 tlen = ip->ip_len; 428 429 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 430 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 431 th->th_sum = m->m_pkthdr.csum_data; 432 else 433 th->th_sum = in_pseudo(ip->ip_src.s_addr, 434 ip->ip_dst.s_addr, 435 htonl(m->m_pkthdr.csum_data + 436 ip->ip_len + 437 IPPROTO_TCP)); 438 th->th_sum ^= 0xffff; 439 } else { 440 /* 441 * Checksum extended TCP header and data. 442 */ 443 len = sizeof (struct ip) + tlen; 444 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 445 ipov->ih_len = (u_short)tlen; 446 ipov->ih_len = htons(ipov->ih_len); 447 th->th_sum = in_cksum(m, len); 448 } 449 if (th->th_sum) { 450 tcpstat.tcps_rcvbadsum++; 451 goto drop; 452 } 453#ifdef INET6 454 /* Re-initialization for later version check */ 455 ip->ip_v = IPVERSION; 456#endif 457 } 458 459 /* 460 * Check that TCP offset makes sense, 461 * pull out TCP options and adjust length. XXX 462 */ 463 off = th->th_off << 2; 464 if (off < sizeof (struct tcphdr) || off > tlen) { 465 tcpstat.tcps_rcvbadoff++; 466 goto drop; 467 } 468 tlen -= off; /* tlen is used instead of ti->ti_len */ 469 if (off > sizeof (struct tcphdr)) { 470 if (isipv6) { 471 IP6_EXTHDR_CHECK(m, off0, off, ); 472 ip6 = mtod(m, struct ip6_hdr *); 473 th = (struct tcphdr *)((caddr_t)ip6 + off0); 474 } else { 475 if (m->m_len < sizeof(struct ip) + off) { 476 if ((m = m_pullup(m, sizeof (struct ip) + off)) 477 == 0) { 478 tcpstat.tcps_rcvshort++; 479 return; 480 } 481 ip = mtod(m, struct ip *); 482 ipov = (struct ipovly *)ip; 483 th = (struct tcphdr *)((caddr_t)ip + off0); 484 } 485 } 486 optlen = off - sizeof (struct tcphdr); 487 optp = (u_char *)(th + 1); 488 } 489 thflags = th->th_flags; 490 491#ifdef TCP_DROP_SYNFIN 492 /* 493 * If the drop_synfin option is enabled, drop all packets with 494 * both the SYN and FIN bits set. This prevents e.g. nmap from 495 * identifying the TCP/IP stack. 496 * 497 * This is a violation of the TCP specification. 498 */ 499 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 500 goto drop; 501#endif 502 503 /* 504 * Convert TCP protocol specific fields to host format. 505 */ 506 th->th_seq = ntohl(th->th_seq); 507 th->th_ack = ntohl(th->th_ack); 508 th->th_win = ntohs(th->th_win); 509 th->th_urp = ntohs(th->th_urp); 510 511 /* 512 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 513 * until after ip6_savecontrol() is called and before other functions 514 * which don't want those proto headers. 515 * Because ip6_savecontrol() is going to parse the mbuf to 516 * search for data to be passed up to user-land, it wants mbuf 517 * parameters to be unchanged. 518 * XXX: the call of ip6_savecontrol() has been obsoleted based on 519 * latest version of the advanced API (20020110). 520 */ 521 drop_hdrlen = off0 + off; 522 523 /* 524 * Locate pcb for segment. 525 */ 526 INP_INFO_WLOCK(&tcbinfo); 527 headlocked = 1; 528findpcb: 529 /* IPFIREWALL_FORWARD section */ 530 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 531 /* 532 * Transparently forwarded. Pretend to be the destination. 533 * already got one like this? 534 */ 535 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 536 ip->ip_dst, th->th_dport, 537 0, m->m_pkthdr.rcvif); 538 if (!inp) { 539 /* It's new. Try find the ambushing socket. */ 540 inp = in_pcblookup_hash(&tcbinfo, 541 ip->ip_src, th->th_sport, 542 next_hop->sin_addr, 543 next_hop->sin_port ? 544 ntohs(next_hop->sin_port) : 545 th->th_dport, 546 1, m->m_pkthdr.rcvif); 547 } 548 } else { 549 if (isipv6) 550 inp = in6_pcblookup_hash(&tcbinfo, 551 &ip6->ip6_src, th->th_sport, 552 &ip6->ip6_dst, th->th_dport, 553 1, m->m_pkthdr.rcvif); 554 else 555 inp = in_pcblookup_hash(&tcbinfo, 556 ip->ip_src, th->th_sport, 557 ip->ip_dst, th->th_dport, 558 1, m->m_pkthdr.rcvif); 559 } 560 561#ifdef IPSEC 562 if (isipv6) { 563 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 564 ipsec6stat.in_polvio++; 565 goto drop; 566 } 567 } else { 568 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 569 ipsecstat.in_polvio++; 570 goto drop; 571 } 572 } 573#endif 574#ifdef FAST_IPSEC 575 if (isipv6) { 576 if (inp != NULL && ipsec6_in_reject(m, inp)) { 577 goto drop; 578 } 579 } else 580 if (inp != NULL && ipsec4_in_reject(m, inp)) { 581 goto drop; 582 } 583#endif /*FAST_IPSEC*/ 584 585 /* 586 * If the state is CLOSED (i.e., TCB does not exist) then 587 * all data in the incoming segment is discarded. 588 * If the TCB exists but is in CLOSED state, it is embryonic, 589 * but should either do a listen or a connect soon. 590 */ 591 if (inp == NULL) { 592 if (log_in_vain) { 593#ifdef INET6 594 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 595#else 596 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 597#endif 598 599 if (isipv6) { 600 strcpy(dbuf, "["); 601 strcpy(sbuf, "["); 602 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 603 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 604 strcat(dbuf, "]"); 605 strcat(sbuf, "]"); 606 } else { 607 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 608 strcpy(sbuf, inet_ntoa(ip->ip_src)); 609 } 610 switch (log_in_vain) { 611 case 1: 612 if (thflags & TH_SYN) 613 log(LOG_INFO, 614 "Connection attempt to TCP %s:%d " 615 "from %s:%d\n", 616 dbuf, ntohs(th->th_dport), sbuf, 617 ntohs(th->th_sport)); 618 break; 619 case 2: 620 log(LOG_INFO, 621 "Connection attempt to TCP %s:%d " 622 "from %s:%d flags:0x%x\n", 623 dbuf, ntohs(th->th_dport), sbuf, 624 ntohs(th->th_sport), thflags); 625 break; 626 default: 627 break; 628 } 629 } 630 if (blackhole) { 631 switch (blackhole) { 632 case 1: 633 if (thflags & TH_SYN) 634 goto drop; 635 break; 636 case 2: 637 goto drop; 638 default: 639 goto drop; 640 } 641 } 642 rstreason = BANDLIM_RST_CLOSEDPORT; 643 goto dropwithreset; 644 } 645 INP_LOCK(inp); 646 tp = intotcpcb(inp); 647 if (tp == 0) { 648 INP_UNLOCK(inp); 649 rstreason = BANDLIM_RST_CLOSEDPORT; 650 goto dropwithreset; 651 } 652 if (tp->t_state == TCPS_CLOSED) 653 goto drop; 654 655 /* Unscale the window into a 32-bit value. */ 656 if ((thflags & TH_SYN) == 0) 657 tiwin = th->th_win << tp->snd_scale; 658 else 659 tiwin = th->th_win; 660 661 so = inp->inp_socket; 662#ifdef MAC 663 error = mac_check_socket_deliver(so, m); 664 if (error) 665 goto drop; 666#endif 667 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 668 struct in_conninfo inc; 669#ifdef TCPDEBUG 670 if (so->so_options & SO_DEBUG) { 671 ostate = tp->t_state; 672 if (isipv6) 673 bcopy((char *)ip6, (char *)tcp_saveipgen, 674 sizeof(*ip6)); 675 else 676 bcopy((char *)ip, (char *)tcp_saveipgen, 677 sizeof(*ip)); 678 tcp_savetcp = *th; 679 } 680#endif 681 /* skip if this isn't a listen socket */ 682 if ((so->so_options & SO_ACCEPTCONN) == 0) 683 goto after_listen; 684#ifdef INET6 685 inc.inc_isipv6 = isipv6; 686#endif 687 if (isipv6) { 688 inc.inc6_faddr = ip6->ip6_src; 689 inc.inc6_laddr = ip6->ip6_dst; 690 inc.inc6_route.ro_rt = NULL; /* XXX */ 691 } else { 692 inc.inc_faddr = ip->ip_src; 693 inc.inc_laddr = ip->ip_dst; 694 inc.inc_route.ro_rt = NULL; /* XXX */ 695 } 696 inc.inc_fport = th->th_sport; 697 inc.inc_lport = th->th_dport; 698 699 /* 700 * If the state is LISTEN then ignore segment if it contains 701 * a RST. If the segment contains an ACK then it is bad and 702 * send a RST. If it does not contain a SYN then it is not 703 * interesting; drop it. 704 * 705 * If the state is SYN_RECEIVED (syncache) and seg contains 706 * an ACK, but not for our SYN/ACK, send a RST. If the seg 707 * contains a RST, check the sequence number to see if it 708 * is a valid reset segment. 709 */ 710 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 711 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 712 if (!syncache_expand(&inc, th, &so, m)) { 713 /* 714 * No syncache entry, or ACK was not 715 * for our SYN/ACK. Send a RST. 716 */ 717 tcpstat.tcps_badsyn++; 718 rstreason = BANDLIM_RST_OPENPORT; 719 goto dropwithreset; 720 } 721 if (so == NULL) { 722 /* 723 * Could not complete 3-way handshake, 724 * connection is being closed down, and 725 * syncache will free mbuf. 726 */ 727 INP_UNLOCK(inp); 728 INP_INFO_WUNLOCK(&tcbinfo); 729 return; 730 } 731 /* 732 * Socket is created in state SYN_RECEIVED. 733 * Continue processing segment. 734 */ 735 INP_UNLOCK(inp); 736 inp = sotoinpcb(so); 737 INP_LOCK(inp); 738 tp = intotcpcb(inp); 739 /* 740 * This is what would have happened in 741 * tcp_output() when the SYN,ACK was sent. 742 */ 743 tp->snd_up = tp->snd_una; 744 tp->snd_max = tp->snd_nxt = tp->iss + 1; 745 tp->last_ack_sent = tp->rcv_nxt; 746/* 747 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 748 * until the _second_ ACK is received: 749 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 750 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 751 * move to ESTAB, set snd_wnd to tiwin. 752 */ 753 tp->snd_wnd = tiwin; /* unscaled */ 754 goto after_listen; 755 } 756 if (thflags & TH_RST) { 757 syncache_chkrst(&inc, th); 758 goto drop; 759 } 760 if (thflags & TH_ACK) { 761 syncache_badack(&inc); 762 tcpstat.tcps_badsyn++; 763 rstreason = BANDLIM_RST_OPENPORT; 764 goto dropwithreset; 765 } 766 goto drop; 767 } 768 769 /* 770 * Segment's flags are (SYN) or (SYN|FIN). 771 */ 772#ifdef INET6 773 /* 774 * If deprecated address is forbidden, 775 * we do not accept SYN to deprecated interface 776 * address to prevent any new inbound connection from 777 * getting established. 778 * When we do not accept SYN, we send a TCP RST, 779 * with deprecated source address (instead of dropping 780 * it). We compromise it as it is much better for peer 781 * to send a RST, and RST will be the final packet 782 * for the exchange. 783 * 784 * If we do not forbid deprecated addresses, we accept 785 * the SYN packet. RFC2462 does not suggest dropping 786 * SYN in this case. 787 * If we decipher RFC2462 5.5.4, it says like this: 788 * 1. use of deprecated addr with existing 789 * communication is okay - "SHOULD continue to be 790 * used" 791 * 2. use of it with new communication: 792 * (2a) "SHOULD NOT be used if alternate address 793 * with sufficient scope is available" 794 * (2b) nothing mentioned otherwise. 795 * Here we fall into (2b) case as we have no choice in 796 * our source address selection - we must obey the peer. 797 * 798 * The wording in RFC2462 is confusing, and there are 799 * multiple description text for deprecated address 800 * handling - worse, they are not exactly the same. 801 * I believe 5.5.4 is the best one, so we follow 5.5.4. 802 */ 803 if (isipv6 && !ip6_use_deprecated) { 804 struct in6_ifaddr *ia6; 805 806 if ((ia6 = ip6_getdstifaddr(m)) && 807 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 808 INP_UNLOCK(inp); 809 tp = NULL; 810 rstreason = BANDLIM_RST_OPENPORT; 811 goto dropwithreset; 812 } 813 } 814#endif 815 /* 816 * If it is from this socket, drop it, it must be forged. 817 * Don't bother responding if the destination was a broadcast. 818 */ 819 if (th->th_dport == th->th_sport) { 820 if (isipv6) { 821 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 822 &ip6->ip6_src)) 823 goto drop; 824 } else { 825 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 826 goto drop; 827 } 828 } 829 /* 830 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 831 * 832 * Note that it is quite possible to receive unicast 833 * link-layer packets with a broadcast IP address. Use 834 * in_broadcast() to find them. 835 */ 836 if (m->m_flags & (M_BCAST|M_MCAST)) 837 goto drop; 838 if (isipv6) { 839 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 840 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 841 goto drop; 842 } else { 843 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 844 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 845 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 846 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 847 goto drop; 848 } 849 /* 850 * SYN appears to be valid; create compressed TCP state 851 * for syncache, or perform t/tcp connection. 852 */ 853 if (so->so_qlen <= so->so_qlimit) { 854 tcp_dooptions(&to, optp, optlen, 1); 855 if (!syncache_add(&inc, &to, th, &so, m)) 856 goto drop; 857 if (so == NULL) { 858 /* 859 * Entry added to syncache, mbuf used to 860 * send SYN,ACK packet. 861 */ 862 KASSERT(headlocked, ("headlocked")); 863 INP_UNLOCK(inp); 864 INP_INFO_WUNLOCK(&tcbinfo); 865 return; 866 } 867 /* 868 * Segment passed TAO tests. 869 */ 870 INP_UNLOCK(inp); 871 inp = sotoinpcb(so); 872 INP_LOCK(inp); 873 tp = intotcpcb(inp); 874 tp->snd_wnd = tiwin; 875 tp->t_starttime = ticks; 876 tp->t_state = TCPS_ESTABLISHED; 877 878 /* 879 * If there is a FIN, or if there is data and the 880 * connection is local, then delay SYN,ACK(SYN) in 881 * the hope of piggy-backing it on a response 882 * segment. Otherwise must send ACK now in case 883 * the other side is slow starting. 884 */ 885 if (DELAY_ACK(tp) && 886 ((thflags & TH_FIN) || 887 (tlen != 0 && 888 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 889 (!isipv6 && in_localaddr(inp->inp_faddr)))))) { 890 callout_reset(tp->tt_delack, tcp_delacktime, 891 tcp_timer_delack, tp); 892 tp->t_flags |= TF_NEEDSYN; 893 } else 894 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 895 896 tcpstat.tcps_connects++; 897 soisconnected(so); 898 goto trimthenstep6; 899 } 900 goto drop; 901 } 902after_listen: 903 904/* XXX temp debugging */ 905 /* should not happen - syncache should pick up these connections */ 906 if (tp->t_state == TCPS_LISTEN) 907 panic("tcp_input: TCPS_LISTEN"); 908 909 /* 910 * Segment received on connection. 911 * Reset idle time and keep-alive timer. 912 */ 913 tp->t_rcvtime = ticks; 914 if (TCPS_HAVEESTABLISHED(tp->t_state)) 915 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 916 917 /* 918 * Process options. 919 * XXX this is tradtitional behavior, may need to be cleaned up. 920 */ 921 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 922 if (thflags & TH_SYN) { 923 if (to.to_flags & TOF_SCALE) { 924 tp->t_flags |= TF_RCVD_SCALE; 925 tp->requested_s_scale = to.to_requested_s_scale; 926 } 927 if (to.to_flags & TOF_TS) { 928 tp->t_flags |= TF_RCVD_TSTMP; 929 tp->ts_recent = to.to_tsval; 930 tp->ts_recent_age = ticks; 931 } 932 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 933 tp->t_flags |= TF_RCVD_CC; 934 if (to.to_flags & TOF_MSS) 935 tcp_mss(tp, to.to_mss); 936 } 937 938 /* 939 * Header prediction: check for the two common cases 940 * of a uni-directional data xfer. If the packet has 941 * no control flags, is in-sequence, the window didn't 942 * change and we're not retransmitting, it's a 943 * candidate. If the length is zero and the ack moved 944 * forward, we're the sender side of the xfer. Just 945 * free the data acked & wake any higher level process 946 * that was blocked waiting for space. If the length 947 * is non-zero and the ack didn't move, we're the 948 * receiver side. If we're getting packets in-order 949 * (the reassembly queue is empty), add the data to 950 * the socket buffer and note that we need a delayed ack. 951 * Make sure that the hidden state-flags are also off. 952 * Since we check for TCPS_ESTABLISHED above, it can only 953 * be TH_NEEDSYN. 954 */ 955 if (tp->t_state == TCPS_ESTABLISHED && 956 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 957 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 958 ((to.to_flags & TOF_TS) == 0 || 959 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 960 /* 961 * Using the CC option is compulsory if once started: 962 * the segment is OK if no T/TCP was negotiated or 963 * if the segment has a CC option equal to CCrecv 964 */ 965 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 966 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 967 th->th_seq == tp->rcv_nxt && 968 tiwin && tiwin == tp->snd_wnd && 969 tp->snd_nxt == tp->snd_max) { 970 971 /* 972 * If last ACK falls within this segment's sequence numbers, 973 * record the timestamp. 974 * NOTE that the test is modified according to the latest 975 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 976 */ 977 if ((to.to_flags & TOF_TS) != 0 && 978 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 979 tp->ts_recent_age = ticks; 980 tp->ts_recent = to.to_tsval; 981 } 982 983 if (tlen == 0) { 984 if (SEQ_GT(th->th_ack, tp->snd_una) && 985 SEQ_LEQ(th->th_ack, tp->snd_max) && 986 tp->snd_cwnd >= tp->snd_wnd && 987 tp->t_dupacks < tcprexmtthresh) { 988 KASSERT(headlocked, ("headlocked")); 989 INP_INFO_WUNLOCK(&tcbinfo); 990 headlocked = 0; 991 /* 992 * this is a pure ack for outstanding data. 993 */ 994 ++tcpstat.tcps_predack; 995 /* 996 * "bad retransmit" recovery 997 */ 998 if (tp->t_rxtshift == 1 && 999 ticks < tp->t_badrxtwin) { 1000 tp->snd_cwnd = tp->snd_cwnd_prev; 1001 tp->snd_ssthresh = 1002 tp->snd_ssthresh_prev; 1003 tp->snd_nxt = tp->snd_max; 1004 tp->t_badrxtwin = 0; 1005 } 1006 1007 /* 1008 * Recalculate the transmit timer / rtt. 1009 * 1010 * Some boxes send broken timestamp replies 1011 * during the SYN+ACK phase, ignore 1012 * timestamps of 0 or we could calculate a 1013 * huge RTT and blow up the retransmit timer. 1014 */ 1015 if ((to.to_flags & TOF_TS) != 0 && 1016 to.to_tsecr) { 1017 tcp_xmit_timer(tp, 1018 ticks - to.to_tsecr + 1); 1019 } else if (tp->t_rtttime && 1020 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1021 tcp_xmit_timer(tp, 1022 ticks - tp->t_rtttime); 1023 } 1024 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1025 acked = th->th_ack - tp->snd_una; 1026 tcpstat.tcps_rcvackpack++; 1027 tcpstat.tcps_rcvackbyte += acked; 1028 sbdrop(&so->so_snd, acked); 1029 /* 1030 * pull snd_wl2 up to prevent seq wrap relative 1031 * to th_ack. 1032 */ 1033 tp->snd_wl2 = tp->snd_una = th->th_ack; 1034 tp->t_dupacks = 0; 1035 m_freem(m); 1036 ND6_HINT(tp); /* some progress has been done */ 1037 1038 /* 1039 * If all outstanding data are acked, stop 1040 * retransmit timer, otherwise restart timer 1041 * using current (possibly backed-off) value. 1042 * If process is waiting for space, 1043 * wakeup/selwakeup/signal. If data 1044 * are ready to send, let tcp_output 1045 * decide between more output or persist. 1046 */ 1047 if (tp->snd_una == tp->snd_max) 1048 callout_stop(tp->tt_rexmt); 1049 else if (!callout_active(tp->tt_persist)) 1050 callout_reset(tp->tt_rexmt, 1051 tp->t_rxtcur, 1052 tcp_timer_rexmt, tp); 1053 1054 sowwakeup(so); 1055 if (so->so_snd.sb_cc) 1056 (void) tcp_output(tp); 1057 INP_UNLOCK(inp); 1058 return; 1059 } 1060 } else if (th->th_ack == tp->snd_una && 1061 LIST_EMPTY(&tp->t_segq) && 1062 tlen <= sbspace(&so->so_rcv)) { 1063 KASSERT(headlocked, ("headlocked")); 1064 INP_INFO_WUNLOCK(&tcbinfo); 1065 headlocked = 0; 1066 /* 1067 * this is a pure, in-sequence data packet 1068 * with nothing on the reassembly queue and 1069 * we have enough buffer space to take it. 1070 */ 1071 ++tcpstat.tcps_preddat; 1072 tp->rcv_nxt += tlen; 1073 /* 1074 * Pull snd_wl1 up to prevent seq wrap relative to 1075 * th_seq. 1076 */ 1077 tp->snd_wl1 = th->th_seq; 1078 /* 1079 * Pull rcv_up up to prevent seq wrap relative to 1080 * rcv_nxt. 1081 */ 1082 tp->rcv_up = tp->rcv_nxt; 1083 tcpstat.tcps_rcvpack++; 1084 tcpstat.tcps_rcvbyte += tlen; 1085 ND6_HINT(tp); /* some progress has been done */ 1086 /* 1087 * Add data to socket buffer. 1088 */ 1089 if (so->so_state & SS_CANTRCVMORE) { 1090 m_freem(m); 1091 } else { 1092 m_adj(m, drop_hdrlen); /* delayed header drop */ 1093 sbappend(&so->so_rcv, m); 1094 } 1095 sorwakeup(so); 1096 if (DELAY_ACK(tp)) { 1097 callout_reset(tp->tt_delack, tcp_delacktime, 1098 tcp_timer_delack, tp); 1099 } else { 1100 tp->t_flags |= TF_ACKNOW; 1101 tcp_output(tp); 1102 } 1103 INP_UNLOCK(inp); 1104 return; 1105 } 1106 } 1107 1108 /* 1109 * Calculate amount of space in receive window, 1110 * and then do TCP input processing. 1111 * Receive window is amount of space in rcv queue, 1112 * but not less than advertised window. 1113 */ 1114 { int win; 1115 1116 win = sbspace(&so->so_rcv); 1117 if (win < 0) 1118 win = 0; 1119 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1120 } 1121 1122 switch (tp->t_state) { 1123 1124 /* 1125 * If the state is SYN_RECEIVED: 1126 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1127 */ 1128 case TCPS_SYN_RECEIVED: 1129 if ((thflags & TH_ACK) && 1130 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1131 SEQ_GT(th->th_ack, tp->snd_max))) { 1132 rstreason = BANDLIM_RST_OPENPORT; 1133 goto dropwithreset; 1134 } 1135 break; 1136 1137 /* 1138 * If the state is SYN_SENT: 1139 * if seg contains an ACK, but not for our SYN, drop the input. 1140 * if seg contains a RST, then drop the connection. 1141 * if seg does not contain SYN, then drop it. 1142 * Otherwise this is an acceptable SYN segment 1143 * initialize tp->rcv_nxt and tp->irs 1144 * if seg contains ack then advance tp->snd_una 1145 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1146 * arrange for segment to be acked (eventually) 1147 * continue processing rest of data/controls, beginning with URG 1148 */ 1149 case TCPS_SYN_SENT: 1150 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1151 taop = &tao_noncached; 1152 bzero(taop, sizeof(*taop)); 1153 } 1154 1155 if ((thflags & TH_ACK) && 1156 (SEQ_LEQ(th->th_ack, tp->iss) || 1157 SEQ_GT(th->th_ack, tp->snd_max))) { 1158 /* 1159 * If we have a cached CCsent for the remote host, 1160 * hence we haven't just crashed and restarted, 1161 * do not send a RST. This may be a retransmission 1162 * from the other side after our earlier ACK was lost. 1163 * Our new SYN, when it arrives, will serve as the 1164 * needed ACK. 1165 */ 1166 if (taop->tao_ccsent != 0) 1167 goto drop; 1168 else { 1169 rstreason = BANDLIM_UNLIMITED; 1170 goto dropwithreset; 1171 } 1172 } 1173 if (thflags & TH_RST) { 1174 if (thflags & TH_ACK) 1175 tp = tcp_drop(tp, ECONNREFUSED); 1176 goto drop; 1177 } 1178 if ((thflags & TH_SYN) == 0) 1179 goto drop; 1180 tp->snd_wnd = th->th_win; /* initial send window */ 1181 tp->cc_recv = to.to_cc; /* foreign CC */ 1182 1183 tp->irs = th->th_seq; 1184 tcp_rcvseqinit(tp); 1185 if (thflags & TH_ACK) { 1186 /* 1187 * Our SYN was acked. If segment contains CC.ECHO 1188 * option, check it to make sure this segment really 1189 * matches our SYN. If not, just drop it as old 1190 * duplicate, but send an RST if we're still playing 1191 * by the old rules. If no CC.ECHO option, make sure 1192 * we don't get fooled into using T/TCP. 1193 */ 1194 if (to.to_flags & TOF_CCECHO) { 1195 if (tp->cc_send != to.to_ccecho) { 1196 if (taop->tao_ccsent != 0) 1197 goto drop; 1198 else { 1199 rstreason = BANDLIM_UNLIMITED; 1200 goto dropwithreset; 1201 } 1202 } 1203 } else 1204 tp->t_flags &= ~TF_RCVD_CC; 1205 tcpstat.tcps_connects++; 1206 soisconnected(so); 1207#ifdef MAC 1208 mac_set_socket_peer_from_mbuf(m, so); 1209#endif 1210 /* Do window scaling on this connection? */ 1211 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1212 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1213 tp->snd_scale = tp->requested_s_scale; 1214 tp->rcv_scale = tp->request_r_scale; 1215 } 1216 /* Segment is acceptable, update cache if undefined. */ 1217 if (taop->tao_ccsent == 0) 1218 taop->tao_ccsent = to.to_ccecho; 1219 1220 tp->rcv_adv += tp->rcv_wnd; 1221 tp->snd_una++; /* SYN is acked */ 1222 /* 1223 * If there's data, delay ACK; if there's also a FIN 1224 * ACKNOW will be turned on later. 1225 */ 1226 if (DELAY_ACK(tp) && tlen != 0) 1227 callout_reset(tp->tt_delack, tcp_delacktime, 1228 tcp_timer_delack, tp); 1229 else 1230 tp->t_flags |= TF_ACKNOW; 1231 /* 1232 * Received <SYN,ACK> in SYN_SENT[*] state. 1233 * Transitions: 1234 * SYN_SENT --> ESTABLISHED 1235 * SYN_SENT* --> FIN_WAIT_1 1236 */ 1237 tp->t_starttime = ticks; 1238 if (tp->t_flags & TF_NEEDFIN) { 1239 tp->t_state = TCPS_FIN_WAIT_1; 1240 tp->t_flags &= ~TF_NEEDFIN; 1241 thflags &= ~TH_SYN; 1242 } else { 1243 tp->t_state = TCPS_ESTABLISHED; 1244 callout_reset(tp->tt_keep, tcp_keepidle, 1245 tcp_timer_keep, tp); 1246 } 1247 } else { 1248 /* 1249 * Received initial SYN in SYN-SENT[*] state => 1250 * simultaneous open. If segment contains CC option 1251 * and there is a cached CC, apply TAO test. 1252 * If it succeeds, connection is * half-synchronized. 1253 * Otherwise, do 3-way handshake: 1254 * SYN-SENT -> SYN-RECEIVED 1255 * SYN-SENT* -> SYN-RECEIVED* 1256 * If there was no CC option, clear cached CC value. 1257 */ 1258 tp->t_flags |= TF_ACKNOW; 1259 callout_stop(tp->tt_rexmt); 1260 if (to.to_flags & TOF_CC) { 1261 if (taop->tao_cc != 0 && 1262 CC_GT(to.to_cc, taop->tao_cc)) { 1263 /* 1264 * update cache and make transition: 1265 * SYN-SENT -> ESTABLISHED* 1266 * SYN-SENT* -> FIN-WAIT-1* 1267 */ 1268 taop->tao_cc = to.to_cc; 1269 tp->t_starttime = ticks; 1270 if (tp->t_flags & TF_NEEDFIN) { 1271 tp->t_state = TCPS_FIN_WAIT_1; 1272 tp->t_flags &= ~TF_NEEDFIN; 1273 } else { 1274 tp->t_state = TCPS_ESTABLISHED; 1275 callout_reset(tp->tt_keep, 1276 tcp_keepidle, 1277 tcp_timer_keep, 1278 tp); 1279 } 1280 tp->t_flags |= TF_NEEDSYN; 1281 } else 1282 tp->t_state = TCPS_SYN_RECEIVED; 1283 } else { 1284 /* CC.NEW or no option => invalidate cache */ 1285 taop->tao_cc = 0; 1286 tp->t_state = TCPS_SYN_RECEIVED; 1287 } 1288 } 1289 1290trimthenstep6: 1291 /* 1292 * Advance th->th_seq to correspond to first data byte. 1293 * If data, trim to stay within window, 1294 * dropping FIN if necessary. 1295 */ 1296 th->th_seq++; 1297 if (tlen > tp->rcv_wnd) { 1298 todrop = tlen - tp->rcv_wnd; 1299 m_adj(m, -todrop); 1300 tlen = tp->rcv_wnd; 1301 thflags &= ~TH_FIN; 1302 tcpstat.tcps_rcvpackafterwin++; 1303 tcpstat.tcps_rcvbyteafterwin += todrop; 1304 } 1305 tp->snd_wl1 = th->th_seq - 1; 1306 tp->rcv_up = th->th_seq; 1307 /* 1308 * Client side of transaction: already sent SYN and data. 1309 * If the remote host used T/TCP to validate the SYN, 1310 * our data will be ACK'd; if so, enter normal data segment 1311 * processing in the middle of step 5, ack processing. 1312 * Otherwise, goto step 6. 1313 */ 1314 if (thflags & TH_ACK) 1315 goto process_ACK; 1316 1317 goto step6; 1318 1319 /* 1320 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1321 * if segment contains a SYN and CC [not CC.NEW] option: 1322 * if state == TIME_WAIT and connection duration > MSL, 1323 * drop packet and send RST; 1324 * 1325 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1326 * ack the FIN (and data) in retransmission queue. 1327 * Complete close and delete TCPCB. Then reprocess 1328 * segment, hoping to find new TCPCB in LISTEN state; 1329 * 1330 * else must be old SYN; drop it. 1331 * else do normal processing. 1332 */ 1333 case TCPS_LAST_ACK: 1334 case TCPS_CLOSING: 1335 case TCPS_TIME_WAIT: 1336 if ((thflags & TH_SYN) && 1337 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1338 if (tp->t_state == TCPS_TIME_WAIT && 1339 (ticks - tp->t_starttime) > tcp_msl) { 1340 rstreason = BANDLIM_UNLIMITED; 1341 goto dropwithreset; 1342 } 1343 if (CC_GT(to.to_cc, tp->cc_recv)) { 1344 tp = tcp_close(tp); 1345 goto findpcb; 1346 } 1347 else 1348 goto drop; 1349 } 1350 break; /* continue normal processing */ 1351 } 1352 1353 /* 1354 * States other than LISTEN or SYN_SENT. 1355 * First check the RST flag and sequence number since reset segments 1356 * are exempt from the timestamp and connection count tests. This 1357 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1358 * below which allowed reset segments in half the sequence space 1359 * to fall though and be processed (which gives forged reset 1360 * segments with a random sequence number a 50 percent chance of 1361 * killing a connection). 1362 * Then check timestamp, if present. 1363 * Then check the connection count, if present. 1364 * Then check that at least some bytes of segment are within 1365 * receive window. If segment begins before rcv_nxt, 1366 * drop leading data (and SYN); if nothing left, just ack. 1367 * 1368 * 1369 * If the RST bit is set, check the sequence number to see 1370 * if this is a valid reset segment. 1371 * RFC 793 page 37: 1372 * In all states except SYN-SENT, all reset (RST) segments 1373 * are validated by checking their SEQ-fields. A reset is 1374 * valid if its sequence number is in the window. 1375 * Note: this does not take into account delayed ACKs, so 1376 * we should test against last_ack_sent instead of rcv_nxt. 1377 * The sequence number in the reset segment is normally an 1378 * echo of our outgoing acknowlegement numbers, but some hosts 1379 * send a reset with the sequence number at the rightmost edge 1380 * of our receive window, and we have to handle this case. 1381 * If we have multiple segments in flight, the intial reset 1382 * segment sequence numbers will be to the left of last_ack_sent, 1383 * but they will eventually catch up. 1384 * In any case, it never made sense to trim reset segments to 1385 * fit the receive window since RFC 1122 says: 1386 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1387 * 1388 * A TCP SHOULD allow a received RST segment to include data. 1389 * 1390 * DISCUSSION 1391 * It has been suggested that a RST segment could contain 1392 * ASCII text that encoded and explained the cause of the 1393 * RST. No standard has yet been established for such 1394 * data. 1395 * 1396 * If the reset segment passes the sequence number test examine 1397 * the state: 1398 * SYN_RECEIVED STATE: 1399 * If passive open, return to LISTEN state. 1400 * If active open, inform user that connection was refused. 1401 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1402 * Inform user that connection was reset, and close tcb. 1403 * CLOSING, LAST_ACK STATES: 1404 * Close the tcb. 1405 * TIME_WAIT STATE: 1406 * Drop the segment - see Stevens, vol. 2, p. 964 and 1407 * RFC 1337. 1408 */ 1409 if (thflags & TH_RST) { 1410 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1411 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1412 switch (tp->t_state) { 1413 1414 case TCPS_SYN_RECEIVED: 1415 so->so_error = ECONNREFUSED; 1416 goto close; 1417 1418 case TCPS_ESTABLISHED: 1419 case TCPS_FIN_WAIT_1: 1420 case TCPS_FIN_WAIT_2: 1421 case TCPS_CLOSE_WAIT: 1422 so->so_error = ECONNRESET; 1423 close: 1424 tp->t_state = TCPS_CLOSED; 1425 tcpstat.tcps_drops++; 1426 tp = tcp_close(tp); 1427 break; 1428 1429 case TCPS_CLOSING: 1430 case TCPS_LAST_ACK: 1431 tp = tcp_close(tp); 1432 break; 1433 1434 case TCPS_TIME_WAIT: 1435 break; 1436 } 1437 } 1438 goto drop; 1439 } 1440 1441 /* 1442 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1443 * and it's less than ts_recent, drop it. 1444 */ 1445 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1446 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1447 1448 /* Check to see if ts_recent is over 24 days old. */ 1449 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1450 /* 1451 * Invalidate ts_recent. If this segment updates 1452 * ts_recent, the age will be reset later and ts_recent 1453 * will get a valid value. If it does not, setting 1454 * ts_recent to zero will at least satisfy the 1455 * requirement that zero be placed in the timestamp 1456 * echo reply when ts_recent isn't valid. The 1457 * age isn't reset until we get a valid ts_recent 1458 * because we don't want out-of-order segments to be 1459 * dropped when ts_recent is old. 1460 */ 1461 tp->ts_recent = 0; 1462 } else { 1463 tcpstat.tcps_rcvduppack++; 1464 tcpstat.tcps_rcvdupbyte += tlen; 1465 tcpstat.tcps_pawsdrop++; 1466 if (tlen) { 1467 goto dropafterack; 1468 } else { 1469 printf("PAWS ack-on-ack loop avoided\n"); 1470 goto drop; 1471 } 1472 } 1473 } 1474 1475 /* 1476 * T/TCP mechanism 1477 * If T/TCP was negotiated and the segment doesn't have CC, 1478 * or if its CC is wrong then drop the segment. 1479 * RST segments do not have to comply with this. 1480 */ 1481 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1482 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1483 goto dropafterack; 1484 1485 /* 1486 * In the SYN-RECEIVED state, validate that the packet belongs to 1487 * this connection before trimming the data to fit the receive 1488 * window. Check the sequence number versus IRS since we know 1489 * the sequence numbers haven't wrapped. This is a partial fix 1490 * for the "LAND" DoS attack. 1491 */ 1492 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1493 rstreason = BANDLIM_RST_OPENPORT; 1494 goto dropwithreset; 1495 } 1496 1497 todrop = tp->rcv_nxt - th->th_seq; 1498 if (todrop > 0) { 1499 if (thflags & TH_SYN) { 1500 thflags &= ~TH_SYN; 1501 th->th_seq++; 1502 if (th->th_urp > 1) 1503 th->th_urp--; 1504 else 1505 thflags &= ~TH_URG; 1506 todrop--; 1507 } 1508 /* 1509 * Following if statement from Stevens, vol. 2, p. 960. 1510 */ 1511 if (todrop > tlen 1512 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1513 /* 1514 * Any valid FIN must be to the left of the window. 1515 * At this point the FIN must be a duplicate or out 1516 * of sequence; drop it. 1517 */ 1518 thflags &= ~TH_FIN; 1519 1520 /* 1521 * Send an ACK to resynchronize and drop any data. 1522 * But keep on processing for RST or ACK. 1523 */ 1524 tp->t_flags |= TF_ACKNOW; 1525 todrop = tlen; 1526 tcpstat.tcps_rcvduppack++; 1527 tcpstat.tcps_rcvdupbyte += todrop; 1528 } else { 1529 tcpstat.tcps_rcvpartduppack++; 1530 tcpstat.tcps_rcvpartdupbyte += todrop; 1531 } 1532 drop_hdrlen += todrop; /* drop from the top afterwards */ 1533 th->th_seq += todrop; 1534 tlen -= todrop; 1535 if (th->th_urp > todrop) 1536 th->th_urp -= todrop; 1537 else { 1538 thflags &= ~TH_URG; 1539 th->th_urp = 0; 1540 } 1541 } 1542 1543 /* 1544 * If new data are received on a connection after the 1545 * user processes are gone, then RST the other end. 1546 */ 1547 if ((so->so_state & SS_NOFDREF) && 1548 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1549 tp = tcp_close(tp); 1550 tcpstat.tcps_rcvafterclose++; 1551 rstreason = BANDLIM_UNLIMITED; 1552 goto dropwithreset; 1553 } 1554 1555 /* 1556 * If segment ends after window, drop trailing data 1557 * (and PUSH and FIN); if nothing left, just ACK. 1558 */ 1559 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1560 if (todrop > 0) { 1561 tcpstat.tcps_rcvpackafterwin++; 1562 if (todrop >= tlen) { 1563 tcpstat.tcps_rcvbyteafterwin += tlen; 1564 /* 1565 * If a new connection request is received 1566 * while in TIME_WAIT, drop the old connection 1567 * and start over if the sequence numbers 1568 * are above the previous ones. 1569 */ 1570 if (thflags & TH_SYN && 1571 tp->t_state == TCPS_TIME_WAIT && 1572 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1573 tp = tcp_close(tp); 1574 goto findpcb; 1575 } 1576 /* 1577 * If window is closed can only take segments at 1578 * window edge, and have to drop data and PUSH from 1579 * incoming segments. Continue processing, but 1580 * remember to ack. Otherwise, drop segment 1581 * and ack. 1582 */ 1583 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1584 tp->t_flags |= TF_ACKNOW; 1585 tcpstat.tcps_rcvwinprobe++; 1586 } else 1587 goto dropafterack; 1588 } else 1589 tcpstat.tcps_rcvbyteafterwin += todrop; 1590 m_adj(m, -todrop); 1591 tlen -= todrop; 1592 thflags &= ~(TH_PUSH|TH_FIN); 1593 } 1594 1595 /* 1596 * If last ACK falls within this segment's sequence numbers, 1597 * record its timestamp. 1598 * NOTE that the test is modified according to the latest 1599 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1600 */ 1601 if ((to.to_flags & TOF_TS) != 0 && 1602 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1603 tp->ts_recent_age = ticks; 1604 tp->ts_recent = to.to_tsval; 1605 } 1606 1607 /* 1608 * If a SYN is in the window, then this is an 1609 * error and we send an RST and drop the connection. 1610 */ 1611 if (thflags & TH_SYN) { 1612 tp = tcp_drop(tp, ECONNRESET); 1613 rstreason = BANDLIM_UNLIMITED; 1614 goto dropwithreset; 1615 } 1616 1617 /* 1618 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1619 * flag is on (half-synchronized state), then queue data for 1620 * later processing; else drop segment and return. 1621 */ 1622 if ((thflags & TH_ACK) == 0) { 1623 if (tp->t_state == TCPS_SYN_RECEIVED || 1624 (tp->t_flags & TF_NEEDSYN)) 1625 goto step6; 1626 else 1627 goto drop; 1628 } 1629 1630 /* 1631 * Ack processing. 1632 */ 1633 switch (tp->t_state) { 1634 1635 /* 1636 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1637 * ESTABLISHED state and continue processing. 1638 * The ACK was checked above. 1639 */ 1640 case TCPS_SYN_RECEIVED: 1641 1642 tcpstat.tcps_connects++; 1643 soisconnected(so); 1644 /* Do window scaling? */ 1645 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1646 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1647 tp->snd_scale = tp->requested_s_scale; 1648 tp->rcv_scale = tp->request_r_scale; 1649 } 1650 /* 1651 * Upon successful completion of 3-way handshake, 1652 * update cache.CC if it was undefined, pass any queued 1653 * data to the user, and advance state appropriately. 1654 */ 1655 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1656 taop->tao_cc == 0) 1657 taop->tao_cc = tp->cc_recv; 1658 1659 /* 1660 * Make transitions: 1661 * SYN-RECEIVED -> ESTABLISHED 1662 * SYN-RECEIVED* -> FIN-WAIT-1 1663 */ 1664 tp->t_starttime = ticks; 1665 if (tp->t_flags & TF_NEEDFIN) { 1666 tp->t_state = TCPS_FIN_WAIT_1; 1667 tp->t_flags &= ~TF_NEEDFIN; 1668 } else { 1669 tp->t_state = TCPS_ESTABLISHED; 1670 callout_reset(tp->tt_keep, tcp_keepidle, 1671 tcp_timer_keep, tp); 1672 } 1673 /* 1674 * If segment contains data or ACK, will call tcp_reass() 1675 * later; if not, do so now to pass queued data to user. 1676 */ 1677 if (tlen == 0 && (thflags & TH_FIN) == 0) 1678 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1679 (struct mbuf *)0); 1680 tp->snd_wl1 = th->th_seq - 1; 1681 /* FALLTHROUGH */ 1682 1683 /* 1684 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1685 * ACKs. If the ack is in the range 1686 * tp->snd_una < th->th_ack <= tp->snd_max 1687 * then advance tp->snd_una to th->th_ack and drop 1688 * data from the retransmission queue. If this ACK reflects 1689 * more up to date window information we update our window information. 1690 */ 1691 case TCPS_ESTABLISHED: 1692 case TCPS_FIN_WAIT_1: 1693 case TCPS_FIN_WAIT_2: 1694 case TCPS_CLOSE_WAIT: 1695 case TCPS_CLOSING: 1696 case TCPS_LAST_ACK: 1697 case TCPS_TIME_WAIT: 1698 1699 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1700 if (tlen == 0 && tiwin == tp->snd_wnd) { 1701 tcpstat.tcps_rcvdupack++; 1702 /* 1703 * If we have outstanding data (other than 1704 * a window probe), this is a completely 1705 * duplicate ack (ie, window info didn't 1706 * change), the ack is the biggest we've 1707 * seen and we've seen exactly our rexmt 1708 * threshhold of them, assume a packet 1709 * has been dropped and retransmit it. 1710 * Kludge snd_nxt & the congestion 1711 * window so we send only this one 1712 * packet. 1713 * 1714 * We know we're losing at the current 1715 * window size so do congestion avoidance 1716 * (set ssthresh to half the current window 1717 * and pull our congestion window back to 1718 * the new ssthresh). 1719 * 1720 * Dup acks mean that packets have left the 1721 * network (they're now cached at the receiver) 1722 * so bump cwnd by the amount in the receiver 1723 * to keep a constant cwnd packets in the 1724 * network. 1725 */ 1726 if (!callout_active(tp->tt_rexmt) || 1727 th->th_ack != tp->snd_una) 1728 tp->t_dupacks = 0; 1729 else if (++tp->t_dupacks == tcprexmtthresh) { 1730 tcp_seq onxt = tp->snd_nxt; 1731 u_int win = 1732 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1733 tp->t_maxseg; 1734 if (tcp_do_newreno && 1735 SEQ_LT(th->th_ack, 1736 tp->snd_recover)) { 1737 /* False retransmit, should not 1738 * cut window 1739 */ 1740 tp->snd_cwnd += tp->t_maxseg; 1741 tp->t_dupacks = 0; 1742 (void) tcp_output(tp); 1743 goto drop; 1744 } 1745 if (win < 2) 1746 win = 2; 1747 tp->snd_ssthresh = win * tp->t_maxseg; 1748 tp->snd_recover = tp->snd_max; 1749 callout_stop(tp->tt_rexmt); 1750 tp->t_rtttime = 0; 1751 tp->snd_nxt = th->th_ack; 1752 tp->snd_cwnd = tp->t_maxseg; 1753 (void) tcp_output(tp); 1754 tp->snd_cwnd = tp->snd_ssthresh + 1755 tp->t_maxseg * tp->t_dupacks; 1756 if (SEQ_GT(onxt, tp->snd_nxt)) 1757 tp->snd_nxt = onxt; 1758 goto drop; 1759 } else if (tp->t_dupacks > tcprexmtthresh) { 1760 tp->snd_cwnd += tp->t_maxseg; 1761 (void) tcp_output(tp); 1762 goto drop; 1763 } 1764 } else 1765 tp->t_dupacks = 0; 1766 break; 1767 } 1768 /* 1769 * If the congestion window was inflated to account 1770 * for the other side's cached packets, retract it. 1771 */ 1772 if (tcp_do_newreno) { 1773 int is_partialack = SEQ_LT(th->th_ack, tp->snd_recover); 1774 if (tp->t_dupacks >= tcprexmtthresh) { 1775 if (is_partialack) { 1776 tcp_newreno_partial_ack(tp, th); 1777 } else { 1778 /* 1779 * Window inflation should have left us 1780 * with approximately snd_ssthresh 1781 * outstanding data. 1782 * But in case we would be inclined to 1783 * send a burst, better to do it via 1784 * the slow start mechanism. 1785 */ 1786 if (SEQ_GT(th->th_ack + 1787 tp->snd_ssthresh, 1788 tp->snd_max)) 1789 tp->snd_cwnd = tp->snd_max - 1790 th->th_ack + 1791 tp->t_maxseg; 1792 else 1793 tp->snd_cwnd = tp->snd_ssthresh; 1794 } 1795 } 1796 /* 1797 * Reset dupacks, except on partial acks in 1798 * fast recovery. 1799 */ 1800 if (!(tp->t_dupacks >= tcprexmtthresh && is_partialack)) 1801 tp->t_dupacks = 0; 1802 } else { 1803 if (tp->t_dupacks >= tcprexmtthresh && 1804 tp->snd_cwnd > tp->snd_ssthresh) 1805 tp->snd_cwnd = tp->snd_ssthresh; 1806 tp->t_dupacks = 0; 1807 } 1808 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1809 tcpstat.tcps_rcvacktoomuch++; 1810 goto dropafterack; 1811 } 1812 /* 1813 * If we reach this point, ACK is not a duplicate, 1814 * i.e., it ACKs something we sent. 1815 */ 1816 if (tp->t_flags & TF_NEEDSYN) { 1817 /* 1818 * T/TCP: Connection was half-synchronized, and our 1819 * SYN has been ACK'd (so connection is now fully 1820 * synchronized). Go to non-starred state, 1821 * increment snd_una for ACK of SYN, and check if 1822 * we can do window scaling. 1823 */ 1824 tp->t_flags &= ~TF_NEEDSYN; 1825 tp->snd_una++; 1826 /* Do window scaling? */ 1827 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1828 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1829 tp->snd_scale = tp->requested_s_scale; 1830 tp->rcv_scale = tp->request_r_scale; 1831 } 1832 } 1833 1834process_ACK: 1835 acked = th->th_ack - tp->snd_una; 1836 tcpstat.tcps_rcvackpack++; 1837 tcpstat.tcps_rcvackbyte += acked; 1838 1839 /* 1840 * If we just performed our first retransmit, and the ACK 1841 * arrives within our recovery window, then it was a mistake 1842 * to do the retransmit in the first place. Recover our 1843 * original cwnd and ssthresh, and proceed to transmit where 1844 * we left off. 1845 */ 1846 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1847 ++tcpstat.tcps_sndrexmitbad; 1848 tp->snd_cwnd = tp->snd_cwnd_prev; 1849 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1850 tp->snd_nxt = tp->snd_max; 1851 tp->t_badrxtwin = 0; /* XXX probably not required */ 1852 } 1853 1854 /* 1855 * If we have a timestamp reply, update smoothed 1856 * round trip time. If no timestamp is present but 1857 * transmit timer is running and timed sequence 1858 * number was acked, update smoothed round trip time. 1859 * Since we now have an rtt measurement, cancel the 1860 * timer backoff (cf., Phil Karn's retransmit alg.). 1861 * Recompute the initial retransmit timer. 1862 * 1863 * Some boxes send broken timestamp replies 1864 * during the SYN+ACK phase, ignore 1865 * timestamps of 0 or we could calculate a 1866 * huge RTT and blow up the retransmit timer. 1867 */ 1868 if ((to.to_flags & TOF_TS) != 0 && 1869 to.to_tsecr) { 1870 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1871 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1872 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1873 } 1874 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1875 1876 /* 1877 * If all outstanding data is acked, stop retransmit 1878 * timer and remember to restart (more output or persist). 1879 * If there is more data to be acked, restart retransmit 1880 * timer, using current (possibly backed-off) value. 1881 */ 1882 if (th->th_ack == tp->snd_max) { 1883 callout_stop(tp->tt_rexmt); 1884 needoutput = 1; 1885 } else if (!callout_active(tp->tt_persist)) 1886 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1887 tcp_timer_rexmt, tp); 1888 1889 /* 1890 * If no data (only SYN) was ACK'd, 1891 * skip rest of ACK processing. 1892 */ 1893 if (acked == 0) 1894 goto step6; 1895 1896 /* 1897 * When new data is acked, open the congestion window. 1898 * If the window gives us less than ssthresh packets 1899 * in flight, open exponentially (maxseg per packet). 1900 * Otherwise open linearly: maxseg per window 1901 * (maxseg^2 / cwnd per packet). 1902 */ 1903 { 1904 register u_int cw = tp->snd_cwnd; 1905 register u_int incr = tp->t_maxseg; 1906 1907 if (cw > tp->snd_ssthresh) 1908 incr = incr * incr / cw; 1909 /* 1910 * If t_dupacks != 0 here, it indicates that we are still 1911 * in NewReno fast recovery mode, so we leave the congestion 1912 * window alone. 1913 */ 1914 if (!tcp_do_newreno || tp->t_dupacks == 0) 1915 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1916 } 1917 if (acked > so->so_snd.sb_cc) { 1918 tp->snd_wnd -= so->so_snd.sb_cc; 1919 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1920 ourfinisacked = 1; 1921 } else { 1922 sbdrop(&so->so_snd, acked); 1923 tp->snd_wnd -= acked; 1924 ourfinisacked = 0; 1925 } 1926 sowwakeup(so); 1927 tp->snd_una = th->th_ack; 1928 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1929 tp->snd_nxt = tp->snd_una; 1930 1931 switch (tp->t_state) { 1932 1933 /* 1934 * In FIN_WAIT_1 STATE in addition to the processing 1935 * for the ESTABLISHED state if our FIN is now acknowledged 1936 * then enter FIN_WAIT_2. 1937 */ 1938 case TCPS_FIN_WAIT_1: 1939 if (ourfinisacked) { 1940 /* 1941 * If we can't receive any more 1942 * data, then closing user can proceed. 1943 * Starting the timer is contrary to the 1944 * specification, but if we don't get a FIN 1945 * we'll hang forever. 1946 */ 1947 if (so->so_state & SS_CANTRCVMORE) { 1948 soisdisconnected(so); 1949 callout_reset(tp->tt_2msl, tcp_maxidle, 1950 tcp_timer_2msl, tp); 1951 } 1952 tp->t_state = TCPS_FIN_WAIT_2; 1953 } 1954 break; 1955 1956 /* 1957 * In CLOSING STATE in addition to the processing for 1958 * the ESTABLISHED state if the ACK acknowledges our FIN 1959 * then enter the TIME-WAIT state, otherwise ignore 1960 * the segment. 1961 */ 1962 case TCPS_CLOSING: 1963 if (ourfinisacked) { 1964 tp->t_state = TCPS_TIME_WAIT; 1965 tcp_canceltimers(tp); 1966 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1967 if (tp->cc_recv != 0 && 1968 (ticks - tp->t_starttime) < tcp_msl) 1969 callout_reset(tp->tt_2msl, 1970 tp->t_rxtcur * 1971 TCPTV_TWTRUNC, 1972 tcp_timer_2msl, tp); 1973 else 1974 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1975 tcp_timer_2msl, tp); 1976 soisdisconnected(so); 1977 } 1978 break; 1979 1980 /* 1981 * In LAST_ACK, we may still be waiting for data to drain 1982 * and/or to be acked, as well as for the ack of our FIN. 1983 * If our FIN is now acknowledged, delete the TCB, 1984 * enter the closed state and return. 1985 */ 1986 case TCPS_LAST_ACK: 1987 if (ourfinisacked) { 1988 tp = tcp_close(tp); 1989 goto drop; 1990 } 1991 break; 1992 1993 /* 1994 * In TIME_WAIT state the only thing that should arrive 1995 * is a retransmission of the remote FIN. Acknowledge 1996 * it and restart the finack timer. 1997 */ 1998 case TCPS_TIME_WAIT: 1999 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2000 tcp_timer_2msl, tp); 2001 goto dropafterack; 2002 } 2003 } 2004 2005step6: 2006 /* 2007 * Update window information. 2008 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2009 */ 2010 if ((thflags & TH_ACK) && 2011 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2012 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2013 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2014 /* keep track of pure window updates */ 2015 if (tlen == 0 && 2016 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2017 tcpstat.tcps_rcvwinupd++; 2018 tp->snd_wnd = tiwin; 2019 tp->snd_wl1 = th->th_seq; 2020 tp->snd_wl2 = th->th_ack; 2021 if (tp->snd_wnd > tp->max_sndwnd) 2022 tp->max_sndwnd = tp->snd_wnd; 2023 needoutput = 1; 2024 } 2025 2026 /* 2027 * Process segments with URG. 2028 */ 2029 if ((thflags & TH_URG) && th->th_urp && 2030 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2031 /* 2032 * This is a kludge, but if we receive and accept 2033 * random urgent pointers, we'll crash in 2034 * soreceive. It's hard to imagine someone 2035 * actually wanting to send this much urgent data. 2036 */ 2037 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2038 th->th_urp = 0; /* XXX */ 2039 thflags &= ~TH_URG; /* XXX */ 2040 goto dodata; /* XXX */ 2041 } 2042 /* 2043 * If this segment advances the known urgent pointer, 2044 * then mark the data stream. This should not happen 2045 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2046 * a FIN has been received from the remote side. 2047 * In these states we ignore the URG. 2048 * 2049 * According to RFC961 (Assigned Protocols), 2050 * the urgent pointer points to the last octet 2051 * of urgent data. We continue, however, 2052 * to consider it to indicate the first octet 2053 * of data past the urgent section as the original 2054 * spec states (in one of two places). 2055 */ 2056 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2057 tp->rcv_up = th->th_seq + th->th_urp; 2058 so->so_oobmark = so->so_rcv.sb_cc + 2059 (tp->rcv_up - tp->rcv_nxt) - 1; 2060 if (so->so_oobmark == 0) 2061 so->so_state |= SS_RCVATMARK; 2062 sohasoutofband(so); 2063 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2064 } 2065 /* 2066 * Remove out of band data so doesn't get presented to user. 2067 * This can happen independent of advancing the URG pointer, 2068 * but if two URG's are pending at once, some out-of-band 2069 * data may creep in... ick. 2070 */ 2071 if (th->th_urp <= (u_long)tlen && 2072 !(so->so_options & SO_OOBINLINE)) { 2073 /* hdr drop is delayed */ 2074 tcp_pulloutofband(so, th, m, drop_hdrlen); 2075 } 2076 } else { 2077 /* 2078 * If no out of band data is expected, 2079 * pull receive urgent pointer along 2080 * with the receive window. 2081 */ 2082 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2083 tp->rcv_up = tp->rcv_nxt; 2084 } 2085dodata: /* XXX */ 2086 KASSERT(headlocked, ("headlocked")); 2087 INP_INFO_WUNLOCK(&tcbinfo); 2088 headlocked = 0; 2089 /* 2090 * Process the segment text, merging it into the TCP sequencing queue, 2091 * and arranging for acknowledgment of receipt if necessary. 2092 * This process logically involves adjusting tp->rcv_wnd as data 2093 * is presented to the user (this happens in tcp_usrreq.c, 2094 * case PRU_RCVD). If a FIN has already been received on this 2095 * connection then we just ignore the text. 2096 */ 2097 if ((tlen || (thflags & TH_FIN)) && 2098 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2099 m_adj(m, drop_hdrlen); /* delayed header drop */ 2100 /* 2101 * Insert segment which includes th into TCP reassembly queue 2102 * with control block tp. Set thflags to whether reassembly now 2103 * includes a segment with FIN. This handles the common case 2104 * inline (segment is the next to be received on an established 2105 * connection, and the queue is empty), avoiding linkage into 2106 * and removal from the queue and repetition of various 2107 * conversions. 2108 * Set DELACK for segments received in order, but ack 2109 * immediately when segments are out of order (so 2110 * fast retransmit can work). 2111 */ 2112 if (th->th_seq == tp->rcv_nxt && 2113 LIST_EMPTY(&tp->t_segq) && 2114 TCPS_HAVEESTABLISHED(tp->t_state)) { 2115 if (DELAY_ACK(tp)) 2116 callout_reset(tp->tt_delack, tcp_delacktime, 2117 tcp_timer_delack, tp); 2118 else 2119 tp->t_flags |= TF_ACKNOW; 2120 tp->rcv_nxt += tlen; 2121 thflags = th->th_flags & TH_FIN; 2122 tcpstat.tcps_rcvpack++; 2123 tcpstat.tcps_rcvbyte += tlen; 2124 ND6_HINT(tp); 2125 if (so->so_state & SS_CANTRCVMORE) 2126 m_freem(m); 2127 else 2128 sbappend(&so->so_rcv, m); 2129 sorwakeup(so); 2130 } else { 2131 thflags = tcp_reass(tp, th, &tlen, m); 2132 tp->t_flags |= TF_ACKNOW; 2133 } 2134 2135 /* 2136 * Note the amount of data that peer has sent into 2137 * our window, in order to estimate the sender's 2138 * buffer size. 2139 */ 2140 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2141 } else { 2142 m_freem(m); 2143 thflags &= ~TH_FIN; 2144 } 2145 2146 /* 2147 * If FIN is received ACK the FIN and let the user know 2148 * that the connection is closing. 2149 */ 2150 if (thflags & TH_FIN) { 2151 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2152 socantrcvmore(so); 2153 /* 2154 * If connection is half-synchronized 2155 * (ie NEEDSYN flag on) then delay ACK, 2156 * so it may be piggybacked when SYN is sent. 2157 * Otherwise, since we received a FIN then no 2158 * more input can be expected, send ACK now. 2159 */ 2160 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2161 callout_reset(tp->tt_delack, tcp_delacktime, 2162 tcp_timer_delack, tp); 2163 else 2164 tp->t_flags |= TF_ACKNOW; 2165 tp->rcv_nxt++; 2166 } 2167 switch (tp->t_state) { 2168 2169 /* 2170 * In SYN_RECEIVED and ESTABLISHED STATES 2171 * enter the CLOSE_WAIT state. 2172 */ 2173 case TCPS_SYN_RECEIVED: 2174 tp->t_starttime = ticks; 2175 /*FALLTHROUGH*/ 2176 case TCPS_ESTABLISHED: 2177 tp->t_state = TCPS_CLOSE_WAIT; 2178 break; 2179 2180 /* 2181 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2182 * enter the CLOSING state. 2183 */ 2184 case TCPS_FIN_WAIT_1: 2185 tp->t_state = TCPS_CLOSING; 2186 break; 2187 2188 /* 2189 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2190 * starting the time-wait timer, turning off the other 2191 * standard timers. 2192 */ 2193 case TCPS_FIN_WAIT_2: 2194 tp->t_state = TCPS_TIME_WAIT; 2195 tcp_canceltimers(tp); 2196 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2197 if (tp->cc_recv != 0 && 2198 (ticks - tp->t_starttime) < tcp_msl) { 2199 callout_reset(tp->tt_2msl, 2200 tp->t_rxtcur * TCPTV_TWTRUNC, 2201 tcp_timer_2msl, tp); 2202 /* For transaction client, force ACK now. */ 2203 tp->t_flags |= TF_ACKNOW; 2204 } 2205 else 2206 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2207 tcp_timer_2msl, tp); 2208 soisdisconnected(so); 2209 break; 2210 2211 /* 2212 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2213 */ 2214 case TCPS_TIME_WAIT: 2215 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2216 tcp_timer_2msl, tp); 2217 break; 2218 } 2219 } 2220#ifdef TCPDEBUG 2221 if (so->so_options & SO_DEBUG) 2222 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2223 &tcp_savetcp, 0); 2224#endif 2225 2226 /* 2227 * Return any desired output. 2228 */ 2229 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2230 (void) tcp_output(tp); 2231 INP_UNLOCK(inp); 2232 return; 2233 2234dropafterack: 2235 /* 2236 * Generate an ACK dropping incoming segment if it occupies 2237 * sequence space, where the ACK reflects our state. 2238 * 2239 * We can now skip the test for the RST flag since all 2240 * paths to this code happen after packets containing 2241 * RST have been dropped. 2242 * 2243 * In the SYN-RECEIVED state, don't send an ACK unless the 2244 * segment we received passes the SYN-RECEIVED ACK test. 2245 * If it fails send a RST. This breaks the loop in the 2246 * "LAND" DoS attack, and also prevents an ACK storm 2247 * between two listening ports that have been sent forged 2248 * SYN segments, each with the source address of the other. 2249 */ 2250 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2251 (SEQ_GT(tp->snd_una, th->th_ack) || 2252 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2253 rstreason = BANDLIM_RST_OPENPORT; 2254 goto dropwithreset; 2255 } 2256#ifdef TCPDEBUG 2257 if (so->so_options & SO_DEBUG) 2258 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2259 &tcp_savetcp, 0); 2260#endif 2261 if (headlocked) 2262 INP_INFO_WUNLOCK(&tcbinfo); 2263 m_freem(m); 2264 tp->t_flags |= TF_ACKNOW; 2265 (void) tcp_output(tp); 2266 INP_UNLOCK(inp); 2267 return; 2268 2269dropwithreset: 2270 /* 2271 * Generate a RST, dropping incoming segment. 2272 * Make ACK acceptable to originator of segment. 2273 * Don't bother to respond if destination was broadcast/multicast. 2274 */ 2275 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2276 goto drop; 2277 if (isipv6) { 2278 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2279 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2280 goto drop; 2281 } else { 2282 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2283 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2284 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2285 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2286 goto drop; 2287 } 2288 /* IPv6 anycast check is done at tcp6_input() */ 2289 2290 /* 2291 * Perform bandwidth limiting. 2292 */ 2293 if (badport_bandlim(rstreason) < 0) 2294 goto drop; 2295 2296#ifdef TCPDEBUG 2297 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2298 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2299 &tcp_savetcp, 0); 2300#endif 2301 2302 if (tp) 2303 INP_UNLOCK(inp); 2304 2305 if (thflags & TH_ACK) 2306 /* mtod() below is safe as long as hdr dropping is delayed */ 2307 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2308 TH_RST); 2309 else { 2310 if (thflags & TH_SYN) 2311 tlen++; 2312 /* mtod() below is safe as long as hdr dropping is delayed */ 2313 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2314 (tcp_seq)0, TH_RST|TH_ACK); 2315 } 2316 if (headlocked) 2317 INP_INFO_WUNLOCK(&tcbinfo); 2318 return; 2319 2320drop: 2321 /* 2322 * Drop space held by incoming segment and return. 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 if (tp) 2330 INP_UNLOCK(inp); 2331 m_freem(m); 2332 if (headlocked) 2333 INP_INFO_WUNLOCK(&tcbinfo); 2334 return; 2335} 2336 2337/* 2338 * Parse TCP options and place in tcpopt. 2339 */ 2340static void 2341tcp_dooptions(to, cp, cnt, is_syn) 2342 struct tcpopt *to; 2343 u_char *cp; 2344 int cnt; 2345{ 2346 int opt, optlen; 2347 2348 to->to_flags = 0; 2349 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2350 opt = cp[0]; 2351 if (opt == TCPOPT_EOL) 2352 break; 2353 if (opt == TCPOPT_NOP) 2354 optlen = 1; 2355 else { 2356 if (cnt < 2) 2357 break; 2358 optlen = cp[1]; 2359 if (optlen < 2 || optlen > cnt) 2360 break; 2361 } 2362 switch (opt) { 2363 case TCPOPT_MAXSEG: 2364 if (optlen != TCPOLEN_MAXSEG) 2365 continue; 2366 if (!is_syn) 2367 continue; 2368 to->to_flags |= TOF_MSS; 2369 bcopy((char *)cp + 2, 2370 (char *)&to->to_mss, sizeof(to->to_mss)); 2371 to->to_mss = ntohs(to->to_mss); 2372 break; 2373 case TCPOPT_WINDOW: 2374 if (optlen != TCPOLEN_WINDOW) 2375 continue; 2376 if (! is_syn) 2377 continue; 2378 to->to_flags |= TOF_SCALE; 2379 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2380 break; 2381 case TCPOPT_TIMESTAMP: 2382 if (optlen != TCPOLEN_TIMESTAMP) 2383 continue; 2384 to->to_flags |= TOF_TS; 2385 bcopy((char *)cp + 2, 2386 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2387 to->to_tsval = ntohl(to->to_tsval); 2388 bcopy((char *)cp + 6, 2389 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2390 to->to_tsecr = ntohl(to->to_tsecr); 2391 break; 2392 case TCPOPT_CC: 2393 if (optlen != TCPOLEN_CC) 2394 continue; 2395 to->to_flags |= TOF_CC; 2396 bcopy((char *)cp + 2, 2397 (char *)&to->to_cc, sizeof(to->to_cc)); 2398 to->to_cc = ntohl(to->to_cc); 2399 break; 2400 case TCPOPT_CCNEW: 2401 if (optlen != TCPOLEN_CC) 2402 continue; 2403 if (!is_syn) 2404 continue; 2405 to->to_flags |= TOF_CCNEW; 2406 bcopy((char *)cp + 2, 2407 (char *)&to->to_cc, sizeof(to->to_cc)); 2408 to->to_cc = ntohl(to->to_cc); 2409 break; 2410 case TCPOPT_CCECHO: 2411 if (optlen != TCPOLEN_CC) 2412 continue; 2413 if (!is_syn) 2414 continue; 2415 to->to_flags |= TOF_CCECHO; 2416 bcopy((char *)cp + 2, 2417 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2418 to->to_ccecho = ntohl(to->to_ccecho); 2419 break; 2420 default: 2421 continue; 2422 } 2423 } 2424} 2425 2426/* 2427 * Pull out of band byte out of a segment so 2428 * it doesn't appear in the user's data queue. 2429 * It is still reflected in the segment length for 2430 * sequencing purposes. 2431 */ 2432static void 2433tcp_pulloutofband(so, th, m, off) 2434 struct socket *so; 2435 struct tcphdr *th; 2436 register struct mbuf *m; 2437 int off; /* delayed to be droped hdrlen */ 2438{ 2439 int cnt = off + th->th_urp - 1; 2440 2441 while (cnt >= 0) { 2442 if (m->m_len > cnt) { 2443 char *cp = mtod(m, caddr_t) + cnt; 2444 struct tcpcb *tp = sototcpcb(so); 2445 2446 tp->t_iobc = *cp; 2447 tp->t_oobflags |= TCPOOB_HAVEDATA; 2448 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2449 m->m_len--; 2450 if (m->m_flags & M_PKTHDR) 2451 m->m_pkthdr.len--; 2452 return; 2453 } 2454 cnt -= m->m_len; 2455 m = m->m_next; 2456 if (m == 0) 2457 break; 2458 } 2459 panic("tcp_pulloutofband"); 2460} 2461 2462/* 2463 * Collect new round-trip time estimate 2464 * and update averages and current timeout. 2465 */ 2466static void 2467tcp_xmit_timer(tp, rtt) 2468 register struct tcpcb *tp; 2469 int rtt; 2470{ 2471 register int delta; 2472 2473 tcpstat.tcps_rttupdated++; 2474 tp->t_rttupdated++; 2475 if (tp->t_srtt != 0) { 2476 /* 2477 * srtt is stored as fixed point with 5 bits after the 2478 * binary point (i.e., scaled by 8). The following magic 2479 * is equivalent to the smoothing algorithm in rfc793 with 2480 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2481 * point). Adjust rtt to origin 0. 2482 */ 2483 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2484 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2485 2486 if ((tp->t_srtt += delta) <= 0) 2487 tp->t_srtt = 1; 2488 2489 /* 2490 * We accumulate a smoothed rtt variance (actually, a 2491 * smoothed mean difference), then set the retransmit 2492 * timer to smoothed rtt + 4 times the smoothed variance. 2493 * rttvar is stored as fixed point with 4 bits after the 2494 * binary point (scaled by 16). The following is 2495 * equivalent to rfc793 smoothing with an alpha of .75 2496 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2497 * rfc793's wired-in beta. 2498 */ 2499 if (delta < 0) 2500 delta = -delta; 2501 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2502 if ((tp->t_rttvar += delta) <= 0) 2503 tp->t_rttvar = 1; 2504 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2505 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2506 } else { 2507 /* 2508 * No rtt measurement yet - use the unsmoothed rtt. 2509 * Set the variance to half the rtt (so our first 2510 * retransmit happens at 3*rtt). 2511 */ 2512 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2513 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2514 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2515 } 2516 tp->t_rtttime = 0; 2517 tp->t_rxtshift = 0; 2518 2519 /* 2520 * the retransmit should happen at rtt + 4 * rttvar. 2521 * Because of the way we do the smoothing, srtt and rttvar 2522 * will each average +1/2 tick of bias. When we compute 2523 * the retransmit timer, we want 1/2 tick of rounding and 2524 * 1 extra tick because of +-1/2 tick uncertainty in the 2525 * firing of the timer. The bias will give us exactly the 2526 * 1.5 tick we need. But, because the bias is 2527 * statistical, we have to test that we don't drop below 2528 * the minimum feasible timer (which is 2 ticks). 2529 */ 2530 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2531 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2532 2533 /* 2534 * We received an ack for a packet that wasn't retransmitted; 2535 * it is probably safe to discard any error indications we've 2536 * received recently. This isn't quite right, but close enough 2537 * for now (a route might have failed after we sent a segment, 2538 * and the return path might not be symmetrical). 2539 */ 2540 tp->t_softerror = 0; 2541} 2542 2543/* 2544 * Determine a reasonable value for maxseg size. 2545 * If the route is known, check route for mtu. 2546 * If none, use an mss that can be handled on the outgoing 2547 * interface without forcing IP to fragment; if bigger than 2548 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2549 * to utilize large mbufs. If no route is found, route has no mtu, 2550 * or the destination isn't local, use a default, hopefully conservative 2551 * size (usually 512 or the default IP max size, but no more than the mtu 2552 * of the interface), as we can't discover anything about intervening 2553 * gateways or networks. We also initialize the congestion/slow start 2554 * window to be a single segment if the destination isn't local. 2555 * While looking at the routing entry, we also initialize other path-dependent 2556 * parameters from pre-set or cached values in the routing entry. 2557 * 2558 * Also take into account the space needed for options that we 2559 * send regularly. Make maxseg shorter by that amount to assure 2560 * that we can send maxseg amount of data even when the options 2561 * are present. Store the upper limit of the length of options plus 2562 * data in maxopd. 2563 * 2564 * NOTE that this routine is only called when we process an incoming 2565 * segment, for outgoing segments only tcp_mssopt is called. 2566 * 2567 * In case of T/TCP, we call this routine during implicit connection 2568 * setup as well (offer = -1), to initialize maxseg from the cached 2569 * MSS of our peer. 2570 */ 2571void 2572tcp_mss(tp, offer) 2573 struct tcpcb *tp; 2574 int offer; 2575{ 2576 register struct rtentry *rt; 2577 struct ifnet *ifp; 2578 register int rtt, mss; 2579 u_long bufsize; 2580 struct inpcb *inp = tp->t_inpcb; 2581 struct socket *so; 2582 struct rmxp_tao *taop; 2583 int origoffer = offer; 2584#ifdef INET6 2585 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2586 size_t min_protoh = isipv6 ? 2587 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2588 sizeof (struct tcpiphdr); 2589#else 2590 const int isipv6 = 0; 2591 const size_t min_protoh = sizeof (struct tcpiphdr); 2592#endif 2593 2594 if (isipv6) 2595 rt = tcp_rtlookup6(&inp->inp_inc); 2596 else 2597 rt = tcp_rtlookup(&inp->inp_inc); 2598 if (rt == NULL) { 2599 tp->t_maxopd = tp->t_maxseg = 2600 isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2601 return; 2602 } 2603 ifp = rt->rt_ifp; 2604 so = inp->inp_socket; 2605 2606 taop = rmx_taop(rt->rt_rmx); 2607 /* 2608 * Offer == -1 means that we didn't receive SYN yet, 2609 * use cached value in that case; 2610 */ 2611 if (offer == -1) 2612 offer = taop->tao_mssopt; 2613 /* 2614 * Offer == 0 means that there was no MSS on the SYN segment, 2615 * in this case we use tcp_mssdflt. 2616 */ 2617 if (offer == 0) 2618 offer = isipv6 ? tcp_v6mssdflt : tcp_mssdflt; 2619 else 2620 /* 2621 * Sanity check: make sure that maxopd will be large 2622 * enough to allow some data on segments even is the 2623 * all the option space is used (40bytes). Otherwise 2624 * funny things may happen in tcp_output. 2625 */ 2626 offer = max(offer, 64); 2627 taop->tao_mssopt = offer; 2628 2629 /* 2630 * While we're here, check if there's an initial rtt 2631 * or rttvar. Convert from the route-table units 2632 * to scaled multiples of the slow timeout timer. 2633 */ 2634 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2635 /* 2636 * XXX the lock bit for RTT indicates that the value 2637 * is also a minimum value; this is subject to time. 2638 */ 2639 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2640 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2641 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2642 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2643 tcpstat.tcps_usedrtt++; 2644 if (rt->rt_rmx.rmx_rttvar) { 2645 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2646 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2647 tcpstat.tcps_usedrttvar++; 2648 } else { 2649 /* default variation is +- 1 rtt */ 2650 tp->t_rttvar = 2651 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2652 } 2653 TCPT_RANGESET(tp->t_rxtcur, 2654 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2655 tp->t_rttmin, TCPTV_REXMTMAX); 2656 } 2657 /* 2658 * if there's an mtu associated with the route, use it 2659 * else, use the link mtu. 2660 */ 2661 if (rt->rt_rmx.rmx_mtu) 2662 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2663 else { 2664 if (isipv6) { 2665 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2666 min_protoh; 2667 if (!in6_localaddr(&inp->in6p_faddr)) 2668 mss = min(mss, tcp_v6mssdflt); 2669 } else { 2670 mss = ifp->if_mtu - min_protoh; 2671 if (!in_localaddr(inp->inp_faddr)) 2672 mss = min(mss, tcp_mssdflt); 2673 } 2674 } 2675 mss = min(mss, offer); 2676 /* 2677 * maxopd stores the maximum length of data AND options 2678 * in a segment; maxseg is the amount of data in a normal 2679 * segment. We need to store this value (maxopd) apart 2680 * from maxseg, because now every segment carries options 2681 * and thus we normally have somewhat less data in segments. 2682 */ 2683 tp->t_maxopd = mss; 2684 2685 /* 2686 * In case of T/TCP, origoffer==-1 indicates, that no segments 2687 * were received yet. In this case we just guess, otherwise 2688 * we do the same as before T/TCP. 2689 */ 2690 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2691 (origoffer == -1 || 2692 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2693 mss -= TCPOLEN_TSTAMP_APPA; 2694 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2695 (origoffer == -1 || 2696 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2697 mss -= TCPOLEN_CC_APPA; 2698 2699#if (MCLBYTES & (MCLBYTES - 1)) == 0 2700 if (mss > MCLBYTES) 2701 mss &= ~(MCLBYTES-1); 2702#else 2703 if (mss > MCLBYTES) 2704 mss = mss / MCLBYTES * MCLBYTES; 2705#endif 2706 /* 2707 * If there's a pipesize, change the socket buffer 2708 * to that size. Make the socket buffers an integral 2709 * number of mss units; if the mss is larger than 2710 * the socket buffer, decrease the mss. 2711 */ 2712#ifdef RTV_SPIPE 2713 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2714#endif 2715 bufsize = so->so_snd.sb_hiwat; 2716 if (bufsize < mss) 2717 mss = bufsize; 2718 else { 2719 bufsize = roundup(bufsize, mss); 2720 if (bufsize > sb_max) 2721 bufsize = sb_max; 2722 if (bufsize > so->so_snd.sb_hiwat) 2723 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2724 } 2725 tp->t_maxseg = mss; 2726 2727#ifdef RTV_RPIPE 2728 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2729#endif 2730 bufsize = so->so_rcv.sb_hiwat; 2731 if (bufsize > mss) { 2732 bufsize = roundup(bufsize, mss); 2733 if (bufsize > sb_max) 2734 bufsize = sb_max; 2735 if (bufsize > so->so_rcv.sb_hiwat) 2736 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2737 } 2738 2739 /* 2740 * Set the slow-start flight size depending on whether this 2741 * is a local network or not. 2742 */ 2743 if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2744 (!isipv6 && in_localaddr(inp->inp_faddr))) 2745 tp->snd_cwnd = mss * ss_fltsz_local; 2746 else 2747 tp->snd_cwnd = mss * ss_fltsz; 2748 2749 if (rt->rt_rmx.rmx_ssthresh) { 2750 /* 2751 * There's some sort of gateway or interface 2752 * buffer limit on the path. Use this to set 2753 * the slow start threshhold, but set the 2754 * threshold to no less than 2*mss. 2755 */ 2756 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2757 tcpstat.tcps_usedssthresh++; 2758 } 2759} 2760 2761/* 2762 * Determine the MSS option to send on an outgoing SYN. 2763 */ 2764int 2765tcp_mssopt(tp) 2766 struct tcpcb *tp; 2767{ 2768 struct rtentry *rt; 2769#ifdef INET6 2770 int isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2771 size_t min_protoh = isipv6 ? 2772 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2773 sizeof (struct tcpiphdr); 2774#else 2775 const int isipv6 = 0; 2776 const size_t min_protoh = sizeof (struct tcpiphdr); 2777#endif 2778 2779 if (isipv6) 2780 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2781 else 2782 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2783 if (rt == NULL) 2784 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2785 2786 return (rt->rt_ifp->if_mtu - min_protoh); 2787} 2788 2789 2790/* 2791 * On a partial ack arrives, force the retransmission of the 2792 * next unacknowledged segment. Do not clear tp->t_dupacks. 2793 * By setting snd_nxt to ti_ack, this forces retransmission timer to 2794 * be started again. 2795 */ 2796static void 2797tcp_newreno_partial_ack(tp, th) 2798 struct tcpcb *tp; 2799 struct tcphdr *th; 2800{ 2801 tcp_seq onxt = tp->snd_nxt; 2802 u_long ocwnd = tp->snd_cwnd; 2803 2804 callout_stop(tp->tt_rexmt); 2805 tp->t_rtttime = 0; 2806 tp->snd_nxt = th->th_ack; 2807 /* 2808 * Set snd_cwnd to one segment beyond acknowledged offset. 2809 * (tp->snd_una has not yet been updated when this function is called.) 2810 */ 2811 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2812 tp->t_flags |= TF_ACKNOW; 2813 (void) tcp_output(tp); 2814 tp->snd_cwnd = ocwnd; 2815 if (SEQ_GT(onxt, tp->snd_nxt)) 2816 tp->snd_nxt = onxt; 2817 /* 2818 * Partial window deflation. Relies on fact that tp->snd_una 2819 * not updated yet. 2820 */ 2821 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2822} 2823