tcp_input.c revision 128829
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 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 30 * $FreeBSD: head/sys/netinet/tcp_input.c 128829 2004-05-02 15:10:17Z darrenr $ 31 */ 32 33#include "opt_ipfw.h" /* for ipfw_fwd */ 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39#include "opt_tcp_input.h" 40 41#include <sys/param.h> 42#include <sys/kernel.h> 43#include <sys/mac.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/proc.h> /* for proc0 declaration */ 47#include <sys/protosw.h> 48#include <sys/signalvar.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/sysctl.h> 52#include <sys/syslog.h> 53#include <sys/systm.h> 54 55#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57#include <vm/uma.h> 58 59#include <net/if.h> 60#include <net/route.h> 61 62#include <netinet/in.h> 63#include <netinet/in_pcb.h> 64#include <netinet/in_systm.h> 65#include <netinet/in_var.h> 66#include <netinet/ip.h> 67#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 68#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 69#include <netinet/ip_var.h> 70#include <netinet/ip6.h> 71#include <netinet/icmp6.h> 72#include <netinet6/in6_pcb.h> 73#include <netinet6/ip6_var.h> 74#include <netinet6/nd6.h> 75#include <netinet/tcp.h> 76#include <netinet/tcp_fsm.h> 77#include <netinet/tcp_seq.h> 78#include <netinet/tcp_timer.h> 79#include <netinet/tcp_var.h> 80#include <netinet6/tcp6_var.h> 81#include <netinet/tcpip.h> 82#ifdef TCPDEBUG 83#include <netinet/tcp_debug.h> 84#endif /* TCPDEBUG */ 85 86#ifdef FAST_IPSEC 87#include <netipsec/ipsec.h> 88#include <netipsec/ipsec6.h> 89#endif /*FAST_IPSEC*/ 90 91#ifdef IPSEC 92#include <netinet6/ipsec.h> 93#include <netinet6/ipsec6.h> 94#include <netkey/key.h> 95#endif /*IPSEC*/ 96 97#include <machine/in_cksum.h> 98 99static const int tcprexmtthresh = 3; 100tcp_cc tcp_ccgen; 101 102struct tcpstat tcpstat; 103SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 105 106static int log_in_vain = 0; 107SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 108 &log_in_vain, 0, "Log all incoming TCP connections"); 109 110static int blackhole = 0; 111SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 112 &blackhole, 0, "Do not send RST when dropping refused connections"); 113 114int tcp_delack_enabled = 1; 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 116 &tcp_delack_enabled, 0, 117 "Delay ACK to try and piggyback it onto a data packet"); 118 119#ifdef TCP_DROP_SYNFIN 120static int drop_synfin = 0; 121SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 122 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 123#endif 124 125static int tcp_do_rfc3042 = 1; 126SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 127 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 128 129static int tcp_do_rfc3390 = 1; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 131 &tcp_do_rfc3390, 0, 132 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 133 134SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 135 "TCP Segment Reassembly Queue"); 136 137static int tcp_reass_maxseg = 0; 138SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 139 &tcp_reass_maxseg, 0, 140 "Global maximum number of TCP Segments in Reassembly Queue"); 141 142int tcp_reass_qsize = 0; 143SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 144 &tcp_reass_qsize, 0, 145 "Global number of TCP Segments currently in Reassembly Queue"); 146 147static int tcp_reass_maxqlen = 48; 148SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 149 &tcp_reass_maxqlen, 0, 150 "Maximum number of TCP Segments per individual Reassembly Queue"); 151 152static int tcp_reass_overflows = 0; 153SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 154 &tcp_reass_overflows, 0, 155 "Global number of TCP Segment Reassembly Queue Overflows"); 156 157struct inpcbhead tcb; 158#define tcb6 tcb /* for KAME src sync over BSD*'s */ 159struct inpcbinfo tcbinfo; 160struct mtx *tcbinfo_mtx; 161 162static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 163static void tcp_pulloutofband(struct socket *, 164 struct tcphdr *, struct mbuf *, int); 165static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 166 struct mbuf *); 167static void tcp_xmit_timer(struct tcpcb *, int); 168static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 169static int tcp_timewait(struct tcptw *, struct tcpopt *, 170 struct tcphdr *, struct mbuf *, int); 171 172/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 173#ifdef INET6 174#define ND6_HINT(tp) \ 175do { \ 176 if ((tp) && (tp)->t_inpcb && \ 177 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 178 nd6_nud_hint(NULL, NULL, 0); \ 179} while (0) 180#else 181#define ND6_HINT(tp) 182#endif 183 184/* 185 * Indicate whether this ack should be delayed. We can delay the ack if 186 * - there is no delayed ack timer in progress and 187 * - our last ack wasn't a 0-sized window. We never want to delay 188 * the ack that opens up a 0-sized window and 189 * - delayed acks are enabled or 190 * - this is a half-synchronized T/TCP connection. 191 */ 192#define DELAY_ACK(tp) \ 193 ((!callout_active(tp->tt_delack) && \ 194 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 195 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 196 197/* Initialize TCP reassembly queue */ 198uma_zone_t tcp_reass_zone; 199void 200tcp_reass_init() 201{ 202 tcp_reass_maxseg = nmbclusters / 16; 203 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 204 &tcp_reass_maxseg); 205 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 206 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 207 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 208} 209 210static int 211tcp_reass(tp, th, tlenp, m) 212 register struct tcpcb *tp; 213 register struct tcphdr *th; 214 int *tlenp; 215 struct mbuf *m; 216{ 217 struct tseg_qent *q; 218 struct tseg_qent *p = NULL; 219 struct tseg_qent *nq; 220 struct tseg_qent *te = NULL; 221 struct socket *so = tp->t_inpcb->inp_socket; 222 int flags; 223 224 /* 225 * XXX: tcp_reass() is rather inefficient with its data structures 226 * and should be rewritten (see NetBSD for optimizations). While 227 * doing that it should move to its own file tcp_reass.c. 228 */ 229 230 /* 231 * Call with th==0 after become established to 232 * force pre-ESTABLISHED data up to user socket. 233 */ 234 if (th == 0) 235 goto present; 236 237 /* 238 * Limit the number of segments in the reassembly queue to prevent 239 * holding on to too many segments (and thus running out of mbufs). 240 * Make sure to let the missing segment through which caused this 241 * queue. Always keep one global queue entry spare to be able to 242 * process the missing segment. 243 */ 244 if (th->th_seq != tp->rcv_nxt && 245 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 246 tp->t_segqlen >= tcp_reass_maxqlen)) { 247 tcp_reass_overflows++; 248 tcpstat.tcps_rcvmemdrop++; 249 m_freem(m); 250 return (0); 251 } 252 253 /* 254 * Allocate a new queue entry. If we can't, or hit the zone limit 255 * just drop the pkt. 256 */ 257 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 258 if (te == NULL) { 259 tcpstat.tcps_rcvmemdrop++; 260 m_freem(m); 261 return (0); 262 } 263 tp->t_segqlen++; 264 tcp_reass_qsize++; 265 266 /* 267 * Find a segment which begins after this one does. 268 */ 269 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 270 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 271 break; 272 p = q; 273 } 274 275 /* 276 * If there is a preceding segment, it may provide some of 277 * our data already. If so, drop the data from the incoming 278 * segment. If it provides all of our data, drop us. 279 */ 280 if (p != NULL) { 281 register int i; 282 /* conversion to int (in i) handles seq wraparound */ 283 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 284 if (i > 0) { 285 if (i >= *tlenp) { 286 tcpstat.tcps_rcvduppack++; 287 tcpstat.tcps_rcvdupbyte += *tlenp; 288 m_freem(m); 289 uma_zfree(tcp_reass_zone, te); 290 tp->t_segqlen--; 291 tcp_reass_qsize--; 292 /* 293 * Try to present any queued data 294 * at the left window edge to the user. 295 * This is needed after the 3-WHS 296 * completes. 297 */ 298 goto present; /* ??? */ 299 } 300 m_adj(m, i); 301 *tlenp -= i; 302 th->th_seq += i; 303 } 304 } 305 tcpstat.tcps_rcvoopack++; 306 tcpstat.tcps_rcvoobyte += *tlenp; 307 308 /* 309 * While we overlap succeeding segments trim them or, 310 * if they are completely covered, dequeue them. 311 */ 312 while (q) { 313 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 314 if (i <= 0) 315 break; 316 if (i < q->tqe_len) { 317 q->tqe_th->th_seq += i; 318 q->tqe_len -= i; 319 m_adj(q->tqe_m, i); 320 break; 321 } 322 323 nq = LIST_NEXT(q, tqe_q); 324 LIST_REMOVE(q, tqe_q); 325 m_freem(q->tqe_m); 326 uma_zfree(tcp_reass_zone, q); 327 tp->t_segqlen--; 328 tcp_reass_qsize--; 329 q = nq; 330 } 331 332 /* Insert the new segment queue entry into place. */ 333 te->tqe_m = m; 334 te->tqe_th = th; 335 te->tqe_len = *tlenp; 336 337 if (p == NULL) { 338 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 339 } else { 340 LIST_INSERT_AFTER(p, te, tqe_q); 341 } 342 343present: 344 /* 345 * Present data to user, advancing rcv_nxt through 346 * completed sequence space. 347 */ 348 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 349 return (0); 350 q = LIST_FIRST(&tp->t_segq); 351 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 352 return (0); 353 do { 354 tp->rcv_nxt += q->tqe_len; 355 flags = q->tqe_th->th_flags & TH_FIN; 356 nq = LIST_NEXT(q, tqe_q); 357 LIST_REMOVE(q, tqe_q); 358 if (so->so_state & SS_CANTRCVMORE) 359 m_freem(q->tqe_m); 360 else 361 sbappendstream(&so->so_rcv, q->tqe_m); 362 uma_zfree(tcp_reass_zone, q); 363 tp->t_segqlen--; 364 tcp_reass_qsize--; 365 q = nq; 366 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 367 ND6_HINT(tp); 368 sorwakeup(so); 369 return (flags); 370} 371 372/* 373 * TCP input routine, follows pages 65-76 of the 374 * protocol specification dated September, 1981 very closely. 375 */ 376#ifdef INET6 377int 378tcp6_input(mp, offp, proto) 379 struct mbuf **mp; 380 int *offp, proto; 381{ 382 register struct mbuf *m = *mp; 383 struct in6_ifaddr *ia6; 384 385 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 386 387 /* 388 * draft-itojun-ipv6-tcp-to-anycast 389 * better place to put this in? 390 */ 391 ia6 = ip6_getdstifaddr(m); 392 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 393 struct ip6_hdr *ip6; 394 395 ip6 = mtod(m, struct ip6_hdr *); 396 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 397 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 398 return IPPROTO_DONE; 399 } 400 401 tcp_input(m, *offp); 402 return IPPROTO_DONE; 403} 404#endif 405 406void 407tcp_input(m, off0) 408 register struct mbuf *m; 409 int off0; 410{ 411 register struct tcphdr *th; 412 register struct ip *ip = NULL; 413 register struct ipovly *ipov; 414 register struct inpcb *inp = NULL; 415 u_char *optp = NULL; 416 int optlen = 0; 417 int len, tlen, off; 418 int drop_hdrlen; 419 register struct tcpcb *tp = 0; 420 register int thflags; 421 struct socket *so = 0; 422 int todrop, acked, ourfinisacked, needoutput = 0; 423 u_long tiwin; 424 struct tcpopt to; /* options in this segment */ 425 struct rmxp_tao tao; /* our TAO cache entry */ 426 int headlocked = 0; 427 struct sockaddr_in *next_hop = NULL; 428 int rstreason; /* For badport_bandlim accounting purposes */ 429 430 struct ip6_hdr *ip6 = NULL; 431#ifdef INET6 432 int isipv6; 433#else 434 const int isipv6 = 0; 435#endif 436 437#ifdef TCPDEBUG 438 /* 439 * The size of tcp_saveipgen must be the size of the max ip header, 440 * now IPv6. 441 */ 442 u_char tcp_saveipgen[40]; 443 struct tcphdr tcp_savetcp; 444 short ostate = 0; 445#endif 446 447 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 448 next_hop = m_claim_next(m, PACKET_TAG_IPFORWARD); 449#ifdef INET6 450 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 451#endif 452 bzero(&tao, sizeof(tao)); 453 bzero((char *)&to, sizeof(to)); 454 455 tcpstat.tcps_rcvtotal++; 456 457 if (isipv6) { 458#ifdef INET6 459 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 460 ip6 = mtod(m, struct ip6_hdr *); 461 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 462 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 463 tcpstat.tcps_rcvbadsum++; 464 goto drop; 465 } 466 th = (struct tcphdr *)((caddr_t)ip6 + off0); 467 468 /* 469 * Be proactive about unspecified IPv6 address in source. 470 * As we use all-zero to indicate unbounded/unconnected pcb, 471 * unspecified IPv6 address can be used to confuse us. 472 * 473 * Note that packets with unspecified IPv6 destination is 474 * already dropped in ip6_input. 475 */ 476 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 477 /* XXX stat */ 478 goto drop; 479 } 480#else 481 th = NULL; /* XXX: avoid compiler warning */ 482#endif 483 } else { 484 /* 485 * Get IP and TCP header together in first mbuf. 486 * Note: IP leaves IP header in first mbuf. 487 */ 488 if (off0 > sizeof (struct ip)) { 489 ip_stripoptions(m, (struct mbuf *)0); 490 off0 = sizeof(struct ip); 491 } 492 if (m->m_len < sizeof (struct tcpiphdr)) { 493 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 494 tcpstat.tcps_rcvshort++; 495 return; 496 } 497 } 498 ip = mtod(m, struct ip *); 499 ipov = (struct ipovly *)ip; 500 th = (struct tcphdr *)((caddr_t)ip + off0); 501 tlen = ip->ip_len; 502 503 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 504 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 505 th->th_sum = m->m_pkthdr.csum_data; 506 else 507 th->th_sum = in_pseudo(ip->ip_src.s_addr, 508 ip->ip_dst.s_addr, 509 htonl(m->m_pkthdr.csum_data + 510 ip->ip_len + 511 IPPROTO_TCP)); 512 th->th_sum ^= 0xffff; 513#ifdef TCPDEBUG 514 ipov->ih_len = (u_short)tlen; 515 ipov->ih_len = htons(ipov->ih_len); 516#endif 517 } else { 518 /* 519 * Checksum extended TCP header and data. 520 */ 521 len = sizeof (struct ip) + tlen; 522 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 523 ipov->ih_len = (u_short)tlen; 524 ipov->ih_len = htons(ipov->ih_len); 525 th->th_sum = in_cksum(m, len); 526 } 527 if (th->th_sum) { 528 tcpstat.tcps_rcvbadsum++; 529 goto drop; 530 } 531#ifdef INET6 532 /* Re-initialization for later version check */ 533 ip->ip_v = IPVERSION; 534#endif 535 } 536 537 /* 538 * Check that TCP offset makes sense, 539 * pull out TCP options and adjust length. XXX 540 */ 541 off = th->th_off << 2; 542 if (off < sizeof (struct tcphdr) || off > tlen) { 543 tcpstat.tcps_rcvbadoff++; 544 goto drop; 545 } 546 tlen -= off; /* tlen is used instead of ti->ti_len */ 547 if (off > sizeof (struct tcphdr)) { 548 if (isipv6) { 549#ifdef INET6 550 IP6_EXTHDR_CHECK(m, off0, off, ); 551 ip6 = mtod(m, struct ip6_hdr *); 552 th = (struct tcphdr *)((caddr_t)ip6 + off0); 553#endif 554 } else { 555 if (m->m_len < sizeof(struct ip) + off) { 556 if ((m = m_pullup(m, sizeof (struct ip) + off)) 557 == 0) { 558 tcpstat.tcps_rcvshort++; 559 return; 560 } 561 ip = mtod(m, struct ip *); 562 ipov = (struct ipovly *)ip; 563 th = (struct tcphdr *)((caddr_t)ip + off0); 564 } 565 } 566 optlen = off - sizeof (struct tcphdr); 567 optp = (u_char *)(th + 1); 568 } 569 thflags = th->th_flags; 570 571#ifdef TCP_DROP_SYNFIN 572 /* 573 * If the drop_synfin option is enabled, drop all packets with 574 * both the SYN and FIN bits set. This prevents e.g. nmap from 575 * identifying the TCP/IP stack. 576 * 577 * This is a violation of the TCP specification. 578 */ 579 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 580 goto drop; 581#endif 582 583 /* 584 * Convert TCP protocol specific fields to host format. 585 */ 586 th->th_seq = ntohl(th->th_seq); 587 th->th_ack = ntohl(th->th_ack); 588 th->th_win = ntohs(th->th_win); 589 th->th_urp = ntohs(th->th_urp); 590 591 /* 592 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 593 * until after ip6_savecontrol() is called and before other functions 594 * which don't want those proto headers. 595 * Because ip6_savecontrol() is going to parse the mbuf to 596 * search for data to be passed up to user-land, it wants mbuf 597 * parameters to be unchanged. 598 * XXX: the call of ip6_savecontrol() has been obsoleted based on 599 * latest version of the advanced API (20020110). 600 */ 601 drop_hdrlen = off0 + off; 602 603 /* 604 * Locate pcb for segment. 605 */ 606 INP_INFO_WLOCK(&tcbinfo); 607 headlocked = 1; 608findpcb: 609 /* IPFIREWALL_FORWARD section */ 610 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 611 /* 612 * Transparently forwarded. Pretend to be the destination. 613 * already got one like this? 614 */ 615 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 616 ip->ip_dst, th->th_dport, 617 0, m->m_pkthdr.rcvif); 618 if (!inp) { 619 /* It's new. Try find the ambushing socket. */ 620 inp = in_pcblookup_hash(&tcbinfo, 621 ip->ip_src, th->th_sport, 622 next_hop->sin_addr, 623 next_hop->sin_port ? 624 ntohs(next_hop->sin_port) : 625 th->th_dport, 626 1, m->m_pkthdr.rcvif); 627 } 628 } else { 629 if (isipv6) { 630#ifdef INET6 631 inp = in6_pcblookup_hash(&tcbinfo, 632 &ip6->ip6_src, th->th_sport, 633 &ip6->ip6_dst, th->th_dport, 634 1, m->m_pkthdr.rcvif); 635#endif 636 } else 637 inp = in_pcblookup_hash(&tcbinfo, 638 ip->ip_src, th->th_sport, 639 ip->ip_dst, th->th_dport, 640 1, m->m_pkthdr.rcvif); 641 } 642 643#if defined(IPSEC) || defined(FAST_IPSEC) 644 if (isipv6) { 645 if (inp != NULL && ipsec6_in_reject(m, inp)) { 646#ifdef IPSEC 647 ipsec6stat.in_polvio++; 648#endif /*IPSEC*/ 649 goto drop; 650 } 651 } else if (inp != NULL && ipsec4_in_reject(m, inp)) { 652#ifdef IPSEC 653 ipsecstat.in_polvio++; 654#endif /*IPSEC*/ 655 goto drop; 656 } 657#endif /*IPSEC || FAST_IPSEC*/ 658 659 /* 660 * If the state is CLOSED (i.e., TCB does not exist) then 661 * all data in the incoming segment is discarded. 662 * If the TCB exists but is in CLOSED state, it is embryonic, 663 * but should either do a listen or a connect soon. 664 */ 665 if (inp == NULL) { 666 if (log_in_vain) { 667#ifdef INET6 668 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 669#else 670 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 671#endif 672 673 if (isipv6) { 674#ifdef INET6 675 strcpy(dbuf, "["); 676 strcpy(sbuf, "["); 677 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 678 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 679 strcat(dbuf, "]"); 680 strcat(sbuf, "]"); 681#endif 682 } else { 683 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 684 strcpy(sbuf, inet_ntoa(ip->ip_src)); 685 } 686 switch (log_in_vain) { 687 case 1: 688 if ((thflags & TH_SYN) == 0) 689 break; 690 /* FALLTHROUGH */ 691 case 2: 692 log(LOG_INFO, 693 "Connection attempt to TCP %s:%d " 694 "from %s:%d flags:0x%02x\n", 695 dbuf, ntohs(th->th_dport), sbuf, 696 ntohs(th->th_sport), thflags); 697 break; 698 default: 699 break; 700 } 701 } 702 if (blackhole) { 703 switch (blackhole) { 704 case 1: 705 if (thflags & TH_SYN) 706 goto drop; 707 break; 708 case 2: 709 goto drop; 710 default: 711 goto drop; 712 } 713 } 714 rstreason = BANDLIM_RST_CLOSEDPORT; 715 goto dropwithreset; 716 } 717 INP_LOCK(inp); 718 if (inp->inp_vflag & INP_TIMEWAIT) { 719 /* 720 * The only option of relevance is TOF_CC, and only if 721 * present in a SYN segment. See tcp_timewait(). 722 */ 723 if (thflags & TH_SYN) 724 tcp_dooptions(&to, optp, optlen, 1); 725 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 726 &to, th, m, tlen)) 727 goto findpcb; 728 /* 729 * tcp_timewait unlocks inp. 730 */ 731 INP_INFO_WUNLOCK(&tcbinfo); 732 return; 733 } 734 tp = intotcpcb(inp); 735 if (tp == 0) { 736 INP_UNLOCK(inp); 737 rstreason = BANDLIM_RST_CLOSEDPORT; 738 goto dropwithreset; 739 } 740 if (tp->t_state == TCPS_CLOSED) 741 goto drop; 742 743 /* Unscale the window into a 32-bit value. */ 744 if ((thflags & TH_SYN) == 0) 745 tiwin = th->th_win << tp->snd_scale; 746 else 747 tiwin = th->th_win; 748 749#ifdef MAC 750 if (mac_check_inpcb_deliver(inp, m)) 751 goto drop; 752#endif 753 so = inp->inp_socket; 754#ifdef TCPDEBUG 755 if (so->so_options & SO_DEBUG) { 756 ostate = tp->t_state; 757 if (isipv6) 758 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 759 else 760 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 761 tcp_savetcp = *th; 762 } 763#endif 764 if (so->so_options & SO_ACCEPTCONN) { 765 struct in_conninfo inc; 766 767#ifdef INET6 768 inc.inc_isipv6 = isipv6; 769#endif 770 if (isipv6) { 771 inc.inc6_faddr = ip6->ip6_src; 772 inc.inc6_laddr = ip6->ip6_dst; 773 } else { 774 inc.inc_faddr = ip->ip_src; 775 inc.inc_laddr = ip->ip_dst; 776 } 777 inc.inc_fport = th->th_sport; 778 inc.inc_lport = th->th_dport; 779 780 /* 781 * If the state is LISTEN then ignore segment if it contains 782 * a RST. If the segment contains an ACK then it is bad and 783 * send a RST. If it does not contain a SYN then it is not 784 * interesting; drop it. 785 * 786 * If the state is SYN_RECEIVED (syncache) and seg contains 787 * an ACK, but not for our SYN/ACK, send a RST. If the seg 788 * contains a RST, check the sequence number to see if it 789 * is a valid reset segment. 790 */ 791 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 792 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 793 if (!syncache_expand(&inc, th, &so, m)) { 794 /* 795 * No syncache entry, or ACK was not 796 * for our SYN/ACK. Send a RST. 797 */ 798 tcpstat.tcps_badsyn++; 799 rstreason = BANDLIM_RST_OPENPORT; 800 goto dropwithreset; 801 } 802 if (so == NULL) { 803 /* 804 * Could not complete 3-way handshake, 805 * connection is being closed down, and 806 * syncache will free mbuf. 807 */ 808 INP_UNLOCK(inp); 809 INP_INFO_WUNLOCK(&tcbinfo); 810 return; 811 } 812 /* 813 * Socket is created in state SYN_RECEIVED. 814 * Continue processing segment. 815 */ 816 INP_UNLOCK(inp); 817 inp = sotoinpcb(so); 818 INP_LOCK(inp); 819 tp = intotcpcb(inp); 820 /* 821 * This is what would have happened in 822 * tcp_output() when the SYN,ACK was sent. 823 */ 824 tp->snd_up = tp->snd_una; 825 tp->snd_max = tp->snd_nxt = tp->iss + 1; 826 tp->last_ack_sent = tp->rcv_nxt; 827 /* 828 * RFC1323: The window in SYN & SYN/ACK 829 * segments is never scaled. 830 */ 831 tp->snd_wnd = tiwin; /* unscaled */ 832 goto after_listen; 833 } 834 if (thflags & TH_RST) { 835 syncache_chkrst(&inc, th); 836 goto drop; 837 } 838 if (thflags & TH_ACK) { 839 syncache_badack(&inc); 840 tcpstat.tcps_badsyn++; 841 rstreason = BANDLIM_RST_OPENPORT; 842 goto dropwithreset; 843 } 844 goto drop; 845 } 846 847 /* 848 * Segment's flags are (SYN) or (SYN|FIN). 849 */ 850#ifdef INET6 851 /* 852 * If deprecated address is forbidden, 853 * we do not accept SYN to deprecated interface 854 * address to prevent any new inbound connection from 855 * getting established. 856 * When we do not accept SYN, we send a TCP RST, 857 * with deprecated source address (instead of dropping 858 * it). We compromise it as it is much better for peer 859 * to send a RST, and RST will be the final packet 860 * for the exchange. 861 * 862 * If we do not forbid deprecated addresses, we accept 863 * the SYN packet. RFC2462 does not suggest dropping 864 * SYN in this case. 865 * If we decipher RFC2462 5.5.4, it says like this: 866 * 1. use of deprecated addr with existing 867 * communication is okay - "SHOULD continue to be 868 * used" 869 * 2. use of it with new communication: 870 * (2a) "SHOULD NOT be used if alternate address 871 * with sufficient scope is available" 872 * (2b) nothing mentioned otherwise. 873 * Here we fall into (2b) case as we have no choice in 874 * our source address selection - we must obey the peer. 875 * 876 * The wording in RFC2462 is confusing, and there are 877 * multiple description text for deprecated address 878 * handling - worse, they are not exactly the same. 879 * I believe 5.5.4 is the best one, so we follow 5.5.4. 880 */ 881 if (isipv6 && !ip6_use_deprecated) { 882 struct in6_ifaddr *ia6; 883 884 if ((ia6 = ip6_getdstifaddr(m)) && 885 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 886 INP_UNLOCK(inp); 887 tp = NULL; 888 rstreason = BANDLIM_RST_OPENPORT; 889 goto dropwithreset; 890 } 891 } 892#endif 893 /* 894 * If it is from this socket, drop it, it must be forged. 895 * Don't bother responding if the destination was a broadcast. 896 */ 897 if (th->th_dport == th->th_sport) { 898 if (isipv6) { 899 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 900 &ip6->ip6_src)) 901 goto drop; 902 } else { 903 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 904 goto drop; 905 } 906 } 907 /* 908 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 909 * 910 * Note that it is quite possible to receive unicast 911 * link-layer packets with a broadcast IP address. Use 912 * in_broadcast() to find them. 913 */ 914 if (m->m_flags & (M_BCAST|M_MCAST)) 915 goto drop; 916 if (isipv6) { 917 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 918 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 919 goto drop; 920 } else { 921 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 922 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 923 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 924 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 925 goto drop; 926 } 927 /* 928 * SYN appears to be valid; create compressed TCP state 929 * for syncache, or perform t/tcp connection. 930 */ 931 if (so->so_qlen <= so->so_qlimit) { 932#ifdef TCPDEBUG 933 if (so->so_options & SO_DEBUG) 934 tcp_trace(TA_INPUT, ostate, tp, 935 (void *)tcp_saveipgen, &tcp_savetcp, 0); 936#endif 937 tcp_dooptions(&to, optp, optlen, 1); 938 if (!syncache_add(&inc, &to, th, &so, m)) 939 goto drop; 940 if (so == NULL) { 941 /* 942 * Entry added to syncache, mbuf used to 943 * send SYN,ACK packet. 944 */ 945 KASSERT(headlocked, ("headlocked")); 946 INP_UNLOCK(inp); 947 INP_INFO_WUNLOCK(&tcbinfo); 948 return; 949 } 950 /* 951 * Segment passed TAO tests. 952 */ 953 INP_UNLOCK(inp); 954 inp = sotoinpcb(so); 955 INP_LOCK(inp); 956 tp = intotcpcb(inp); 957 tp->snd_wnd = tiwin; 958 tp->t_starttime = ticks; 959 tp->t_state = TCPS_ESTABLISHED; 960 961 /* 962 * T/TCP logic: 963 * If there is a FIN or if there is data, then 964 * delay SYN,ACK(SYN) in the hope of piggy-backing 965 * it on a response segment. Otherwise must send 966 * ACK now in case the other side is slow starting. 967 */ 968 if (thflags & TH_FIN || tlen != 0) 969 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 970 else 971 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 972 tcpstat.tcps_connects++; 973 soisconnected(so); 974 goto trimthenstep6; 975 } 976 goto drop; 977 } 978after_listen: 979 980 /* XXX temp debugging */ 981 /* should not happen - syncache should pick up these connections */ 982 if (tp->t_state == TCPS_LISTEN) 983 panic("tcp_input: TCPS_LISTEN"); 984 985 /* 986 * This is the second part of the MSS DoS prevention code (after 987 * minmss on the sending side) and it deals with too many too small 988 * tcp packets in a too short timeframe (1 second). 989 * 990 * For every full second we count the number of received packets 991 * and bytes. If we get a lot of packets per second for this connection 992 * (tcp_minmssoverload) we take a closer look at it and compute the 993 * average packet size for the past second. If that is less than 994 * tcp_minmss we get too many packets with very small payload which 995 * is not good and burdens our system (and every packet generates 996 * a wakeup to the process connected to our socket). We can reasonable 997 * expect this to be small packet DoS attack to exhaust our CPU 998 * cycles. 999 * 1000 * Care has to be taken for the minimum packet overload value. This 1001 * value defines the minimum number of packets per second before we 1002 * start to worry. This must not be too low to avoid killing for 1003 * example interactive connections with many small packets like 1004 * telnet or SSH. 1005 * 1006 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1007 * this check. 1008 * 1009 * Account for packet if payload packet, skip over ACK, etc. 1010 */ 1011 if (tcp_minmss && tcp_minmssoverload && 1012 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1013 if (tp->rcv_second > ticks) { 1014 tp->rcv_pps++; 1015 tp->rcv_byps += tlen + off; 1016 if (tp->rcv_pps > tcp_minmssoverload) { 1017 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1018 printf("too many small tcp packets from " 1019 "%s:%u, av. %lubyte/packet, " 1020 "dropping connection\n", 1021#ifdef INET6 1022 isipv6 ? 1023 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1024#endif 1025 inet_ntoa(inp->inp_inc.inc_faddr), 1026 inp->inp_inc.inc_fport, 1027 tp->rcv_byps / tp->rcv_pps); 1028 tp = tcp_drop(tp, ECONNRESET); 1029 tcpstat.tcps_minmssdrops++; 1030 goto drop; 1031 } 1032 } 1033 } else { 1034 tp->rcv_second = ticks + hz; 1035 tp->rcv_pps = 1; 1036 tp->rcv_byps = tlen + off; 1037 } 1038 } 1039 1040 /* 1041 * Segment received on connection. 1042 * Reset idle time and keep-alive timer. 1043 */ 1044 tp->t_rcvtime = ticks; 1045 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1046 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1047 1048 /* 1049 * Process options only when we get SYN/ACK back. The SYN case 1050 * for incoming connections is handled in tcp_syncache. 1051 * XXX this is traditional behavior, may need to be cleaned up. 1052 */ 1053 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 1054 if (thflags & TH_SYN) { 1055 if (to.to_flags & TOF_SCALE) { 1056 tp->t_flags |= TF_RCVD_SCALE; 1057 tp->requested_s_scale = to.to_requested_s_scale; 1058 } 1059 if (to.to_flags & TOF_TS) { 1060 tp->t_flags |= TF_RCVD_TSTMP; 1061 tp->ts_recent = to.to_tsval; 1062 tp->ts_recent_age = ticks; 1063 } 1064 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 1065 tp->t_flags |= TF_RCVD_CC; 1066 if (to.to_flags & TOF_MSS) 1067 tcp_mss(tp, to.to_mss); 1068 } 1069 1070 /* 1071 * Header prediction: check for the two common cases 1072 * of a uni-directional data xfer. If the packet has 1073 * no control flags, is in-sequence, the window didn't 1074 * change and we're not retransmitting, it's a 1075 * candidate. If the length is zero and the ack moved 1076 * forward, we're the sender side of the xfer. Just 1077 * free the data acked & wake any higher level process 1078 * that was blocked waiting for space. If the length 1079 * is non-zero and the ack didn't move, we're the 1080 * receiver side. If we're getting packets in-order 1081 * (the reassembly queue is empty), add the data to 1082 * the socket buffer and note that we need a delayed ack. 1083 * Make sure that the hidden state-flags are also off. 1084 * Since we check for TCPS_ESTABLISHED above, it can only 1085 * be TH_NEEDSYN. 1086 */ 1087 if (tp->t_state == TCPS_ESTABLISHED && 1088 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1089 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1090 ((to.to_flags & TOF_TS) == 0 || 1091 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1092 /* 1093 * Using the CC option is compulsory if once started: 1094 * the segment is OK if no T/TCP was negotiated or 1095 * if the segment has a CC option equal to CCrecv 1096 */ 1097 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1098 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 1099 th->th_seq == tp->rcv_nxt && 1100 tiwin && tiwin == tp->snd_wnd && 1101 tp->snd_nxt == tp->snd_max) { 1102 1103 /* 1104 * If last ACK falls within this segment's sequence numbers, 1105 * record the timestamp. 1106 * NOTE that the test is modified according to the latest 1107 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1108 */ 1109 if ((to.to_flags & TOF_TS) != 0 && 1110 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1111 tp->ts_recent_age = ticks; 1112 tp->ts_recent = to.to_tsval; 1113 } 1114 1115 if (tlen == 0) { 1116 if (SEQ_GT(th->th_ack, tp->snd_una) && 1117 SEQ_LEQ(th->th_ack, tp->snd_max) && 1118 tp->snd_cwnd >= tp->snd_wnd && 1119 ((!tcp_do_newreno && 1120 tp->t_dupacks < tcprexmtthresh) || 1121 (tcp_do_newreno && !IN_FASTRECOVERY(tp)))) { 1122 KASSERT(headlocked, ("headlocked")); 1123 INP_INFO_WUNLOCK(&tcbinfo); 1124 /* 1125 * this is a pure ack for outstanding data. 1126 */ 1127 ++tcpstat.tcps_predack; 1128 /* 1129 * "bad retransmit" recovery 1130 */ 1131 if (tp->t_rxtshift == 1 && 1132 ticks < tp->t_badrxtwin) { 1133 ++tcpstat.tcps_sndrexmitbad; 1134 tp->snd_cwnd = tp->snd_cwnd_prev; 1135 tp->snd_ssthresh = 1136 tp->snd_ssthresh_prev; 1137 tp->snd_recover = tp->snd_recover_prev; 1138 if (tp->t_flags & TF_WASFRECOVERY) 1139 ENTER_FASTRECOVERY(tp); 1140 tp->snd_nxt = tp->snd_max; 1141 tp->t_badrxtwin = 0; 1142 } 1143 1144 /* 1145 * Recalculate the transmit timer / rtt. 1146 * 1147 * Some boxes send broken timestamp replies 1148 * during the SYN+ACK phase, ignore 1149 * timestamps of 0 or we could calculate a 1150 * huge RTT and blow up the retransmit timer. 1151 */ 1152 if ((to.to_flags & TOF_TS) != 0 && 1153 to.to_tsecr) { 1154 tcp_xmit_timer(tp, 1155 ticks - to.to_tsecr + 1); 1156 } else if (tp->t_rtttime && 1157 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1158 tcp_xmit_timer(tp, 1159 ticks - tp->t_rtttime); 1160 } 1161 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1162 acked = th->th_ack - tp->snd_una; 1163 tcpstat.tcps_rcvackpack++; 1164 tcpstat.tcps_rcvackbyte += acked; 1165 sbdrop(&so->so_snd, acked); 1166 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1167 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1168 tp->snd_recover = th->th_ack - 1; 1169 tp->snd_una = th->th_ack; 1170 /* 1171 * pull snd_wl2 up to prevent seq wrap relative 1172 * to th_ack. 1173 */ 1174 tp->snd_wl2 = th->th_ack; 1175 tp->t_dupacks = 0; 1176 m_freem(m); 1177 ND6_HINT(tp); /* some progress has been done */ 1178 1179 /* 1180 * If all outstanding data are acked, stop 1181 * retransmit timer, otherwise restart timer 1182 * using current (possibly backed-off) value. 1183 * If process is waiting for space, 1184 * wakeup/selwakeup/signal. If data 1185 * are ready to send, let tcp_output 1186 * decide between more output or persist. 1187 1188#ifdef TCPDEBUG 1189 if (so->so_options & SO_DEBUG) 1190 tcp_trace(TA_INPUT, ostate, tp, 1191 (void *)tcp_saveipgen, 1192 &tcp_savetcp, 0); 1193#endif 1194 */ 1195 if (tp->snd_una == tp->snd_max) 1196 callout_stop(tp->tt_rexmt); 1197 else if (!callout_active(tp->tt_persist)) 1198 callout_reset(tp->tt_rexmt, 1199 tp->t_rxtcur, 1200 tcp_timer_rexmt, tp); 1201 1202 sowwakeup(so); 1203 if (so->so_snd.sb_cc) 1204 (void) tcp_output(tp); 1205 goto check_delack; 1206 } 1207 } else if (th->th_ack == tp->snd_una && 1208 LIST_EMPTY(&tp->t_segq) && 1209 tlen <= sbspace(&so->so_rcv)) { 1210 KASSERT(headlocked, ("headlocked")); 1211 INP_INFO_WUNLOCK(&tcbinfo); 1212 /* 1213 * this is a pure, in-sequence data packet 1214 * with nothing on the reassembly queue and 1215 * we have enough buffer space to take it. 1216 */ 1217 ++tcpstat.tcps_preddat; 1218 tp->rcv_nxt += tlen; 1219 /* 1220 * Pull snd_wl1 up to prevent seq wrap relative to 1221 * th_seq. 1222 */ 1223 tp->snd_wl1 = th->th_seq; 1224 /* 1225 * Pull rcv_up up to prevent seq wrap relative to 1226 * rcv_nxt. 1227 */ 1228 tp->rcv_up = tp->rcv_nxt; 1229 tcpstat.tcps_rcvpack++; 1230 tcpstat.tcps_rcvbyte += tlen; 1231 ND6_HINT(tp); /* some progress has been done */ 1232 /* 1233#ifdef TCPDEBUG 1234 if (so->so_options & SO_DEBUG) 1235 tcp_trace(TA_INPUT, ostate, tp, 1236 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1237#endif 1238 * Add data to socket buffer. 1239 */ 1240 if (so->so_state & SS_CANTRCVMORE) { 1241 m_freem(m); 1242 } else { 1243 m_adj(m, drop_hdrlen); /* delayed header drop */ 1244 sbappendstream(&so->so_rcv, m); 1245 } 1246 sorwakeup(so); 1247 if (DELAY_ACK(tp)) { 1248 tp->t_flags |= TF_DELACK; 1249 } else { 1250 tp->t_flags |= TF_ACKNOW; 1251 tcp_output(tp); 1252 } 1253 goto check_delack; 1254 } 1255 } 1256 1257 /* 1258 * Calculate amount of space in receive window, 1259 * and then do TCP input processing. 1260 * Receive window is amount of space in rcv queue, 1261 * but not less than advertised window. 1262 */ 1263 { int win; 1264 1265 win = sbspace(&so->so_rcv); 1266 if (win < 0) 1267 win = 0; 1268 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1269 } 1270 1271 switch (tp->t_state) { 1272 1273 /* 1274 * If the state is SYN_RECEIVED: 1275 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1276 */ 1277 case TCPS_SYN_RECEIVED: 1278 if ((thflags & TH_ACK) && 1279 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1280 SEQ_GT(th->th_ack, tp->snd_max))) { 1281 rstreason = BANDLIM_RST_OPENPORT; 1282 goto dropwithreset; 1283 } 1284 break; 1285 1286 /* 1287 * If the state is SYN_SENT: 1288 * if seg contains an ACK, but not for our SYN, drop the input. 1289 * if seg contains a RST, then drop the connection. 1290 * if seg does not contain SYN, then drop it. 1291 * Otherwise this is an acceptable SYN segment 1292 * initialize tp->rcv_nxt and tp->irs 1293 * if seg contains ack then advance tp->snd_una 1294 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1295 * arrange for segment to be acked (eventually) 1296 * continue processing rest of data/controls, beginning with URG 1297 */ 1298 case TCPS_SYN_SENT: 1299 if (tcp_do_rfc1644) 1300 tcp_hc_gettao(&inp->inp_inc, &tao); 1301 1302 if ((thflags & TH_ACK) && 1303 (SEQ_LEQ(th->th_ack, tp->iss) || 1304 SEQ_GT(th->th_ack, tp->snd_max))) { 1305 /* 1306 * If we have a cached CCsent for the remote host, 1307 * hence we haven't just crashed and restarted, 1308 * do not send a RST. This may be a retransmission 1309 * from the other side after our earlier ACK was lost. 1310 * Our new SYN, when it arrives, will serve as the 1311 * needed ACK. 1312 */ 1313 if (tao.tao_ccsent != 0) 1314 goto drop; 1315 else { 1316 rstreason = BANDLIM_UNLIMITED; 1317 goto dropwithreset; 1318 } 1319 } 1320 if (thflags & TH_RST) { 1321 if (thflags & TH_ACK) 1322 tp = tcp_drop(tp, ECONNREFUSED); 1323 goto drop; 1324 } 1325 if ((thflags & TH_SYN) == 0) 1326 goto drop; 1327 tp->snd_wnd = th->th_win; /* initial send window */ 1328 tp->cc_recv = to.to_cc; /* foreign CC */ 1329 1330 tp->irs = th->th_seq; 1331 tcp_rcvseqinit(tp); 1332 if (thflags & TH_ACK) { 1333 /* 1334 * Our SYN was acked. If segment contains CC.ECHO 1335 * option, check it to make sure this segment really 1336 * matches our SYN. If not, just drop it as old 1337 * duplicate, but send an RST if we're still playing 1338 * by the old rules. If no CC.ECHO option, make sure 1339 * we don't get fooled into using T/TCP. 1340 */ 1341 if (to.to_flags & TOF_CCECHO) { 1342 if (tp->cc_send != to.to_ccecho) { 1343 if (tao.tao_ccsent != 0) 1344 goto drop; 1345 else { 1346 rstreason = BANDLIM_UNLIMITED; 1347 goto dropwithreset; 1348 } 1349 } 1350 } else 1351 tp->t_flags &= ~TF_RCVD_CC; 1352 tcpstat.tcps_connects++; 1353 soisconnected(so); 1354#ifdef MAC 1355 mac_set_socket_peer_from_mbuf(m, so); 1356#endif 1357 /* Do window scaling on this connection? */ 1358 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1359 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1360 tp->snd_scale = tp->requested_s_scale; 1361 tp->rcv_scale = tp->request_r_scale; 1362 } 1363 /* Segment is acceptable, update cache if undefined. */ 1364 if (tao.tao_ccsent == 0 && tcp_do_rfc1644) 1365 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0); 1366 1367 tp->rcv_adv += tp->rcv_wnd; 1368 tp->snd_una++; /* SYN is acked */ 1369 /* 1370 * If there's data, delay ACK; if there's also a FIN 1371 * ACKNOW will be turned on later. 1372 */ 1373 if (DELAY_ACK(tp) && tlen != 0) 1374 callout_reset(tp->tt_delack, tcp_delacktime, 1375 tcp_timer_delack, tp); 1376 else 1377 tp->t_flags |= TF_ACKNOW; 1378 /* 1379 * Received <SYN,ACK> in SYN_SENT[*] state. 1380 * Transitions: 1381 * SYN_SENT --> ESTABLISHED 1382 * SYN_SENT* --> FIN_WAIT_1 1383 */ 1384 tp->t_starttime = ticks; 1385 if (tp->t_flags & TF_NEEDFIN) { 1386 tp->t_state = TCPS_FIN_WAIT_1; 1387 tp->t_flags &= ~TF_NEEDFIN; 1388 thflags &= ~TH_SYN; 1389 } else { 1390 tp->t_state = TCPS_ESTABLISHED; 1391 callout_reset(tp->tt_keep, tcp_keepidle, 1392 tcp_timer_keep, tp); 1393 } 1394 } else { 1395 /* 1396 * Received initial SYN in SYN-SENT[*] state => 1397 * simultaneous open. If segment contains CC option 1398 * and there is a cached CC, apply TAO test. 1399 * If it succeeds, connection is * half-synchronized. 1400 * Otherwise, do 3-way handshake: 1401 * SYN-SENT -> SYN-RECEIVED 1402 * SYN-SENT* -> SYN-RECEIVED* 1403 * If there was no CC option, clear cached CC value. 1404 */ 1405 tp->t_flags |= TF_ACKNOW; 1406 callout_stop(tp->tt_rexmt); 1407 if (to.to_flags & TOF_CC) { 1408 if (tao.tao_cc != 0 && 1409 CC_GT(to.to_cc, tao.tao_cc)) { 1410 /* 1411 * update cache and make transition: 1412 * SYN-SENT -> ESTABLISHED* 1413 * SYN-SENT* -> FIN-WAIT-1* 1414 */ 1415 tao.tao_cc = to.to_cc; 1416 tcp_hc_updatetao(&inp->inp_inc, 1417 TCP_HC_TAO_CC, to.to_cc, 0); 1418 tp->t_starttime = ticks; 1419 if (tp->t_flags & TF_NEEDFIN) { 1420 tp->t_state = TCPS_FIN_WAIT_1; 1421 tp->t_flags &= ~TF_NEEDFIN; 1422 } else { 1423 tp->t_state = TCPS_ESTABLISHED; 1424 callout_reset(tp->tt_keep, 1425 tcp_keepidle, 1426 tcp_timer_keep, 1427 tp); 1428 } 1429 tp->t_flags |= TF_NEEDSYN; 1430 } else 1431 tp->t_state = TCPS_SYN_RECEIVED; 1432 } else { 1433 if (tcp_do_rfc1644) { 1434 /* CC.NEW or no option => invalidate cache */ 1435 tao.tao_cc = 0; 1436 tcp_hc_updatetao(&inp->inp_inc, 1437 TCP_HC_TAO_CC, to.to_cc, 0); 1438 } 1439 tp->t_state = TCPS_SYN_RECEIVED; 1440 } 1441 } 1442 1443trimthenstep6: 1444 /* 1445 * Advance th->th_seq to correspond to first data byte. 1446 * If data, trim to stay within window, 1447 * dropping FIN if necessary. 1448 */ 1449 th->th_seq++; 1450 if (tlen > tp->rcv_wnd) { 1451 todrop = tlen - tp->rcv_wnd; 1452 m_adj(m, -todrop); 1453 tlen = tp->rcv_wnd; 1454 thflags &= ~TH_FIN; 1455 tcpstat.tcps_rcvpackafterwin++; 1456 tcpstat.tcps_rcvbyteafterwin += todrop; 1457 } 1458 tp->snd_wl1 = th->th_seq - 1; 1459 tp->rcv_up = th->th_seq; 1460 /* 1461 * Client side of transaction: already sent SYN and data. 1462 * If the remote host used T/TCP to validate the SYN, 1463 * our data will be ACK'd; if so, enter normal data segment 1464 * processing in the middle of step 5, ack processing. 1465 * Otherwise, goto step 6. 1466 */ 1467 if (thflags & TH_ACK) 1468 goto process_ACK; 1469 1470 goto step6; 1471 1472 /* 1473 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1474 * if segment contains a SYN and CC [not CC.NEW] option: 1475 * if state == TIME_WAIT and connection duration > MSL, 1476 * drop packet and send RST; 1477 * 1478 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1479 * ack the FIN (and data) in retransmission queue. 1480 * Complete close and delete TCPCB. Then reprocess 1481 * segment, hoping to find new TCPCB in LISTEN state; 1482 * 1483 * else must be old SYN; drop it. 1484 * else do normal processing. 1485 */ 1486 case TCPS_LAST_ACK: 1487 case TCPS_CLOSING: 1488 case TCPS_TIME_WAIT: 1489 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1490 if ((thflags & TH_SYN) && 1491 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1492 if (tp->t_state == TCPS_TIME_WAIT && 1493 (ticks - tp->t_starttime) > tcp_msl) { 1494 rstreason = BANDLIM_UNLIMITED; 1495 goto dropwithreset; 1496 } 1497 if (CC_GT(to.to_cc, tp->cc_recv)) { 1498 tp = tcp_close(tp); 1499 goto findpcb; 1500 } 1501 else 1502 goto drop; 1503 } 1504 break; /* continue normal processing */ 1505 } 1506 1507 /* 1508 * States other than LISTEN or SYN_SENT. 1509 * First check the RST flag and sequence number since reset segments 1510 * are exempt from the timestamp and connection count tests. This 1511 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1512 * below which allowed reset segments in half the sequence space 1513 * to fall though and be processed (which gives forged reset 1514 * segments with a random sequence number a 50 percent chance of 1515 * killing a connection). 1516 * Then check timestamp, if present. 1517 * Then check the connection count, if present. 1518 * Then check that at least some bytes of segment are within 1519 * receive window. If segment begins before rcv_nxt, 1520 * drop leading data (and SYN); if nothing left, just ack. 1521 * 1522 * 1523 * If the RST bit is set, check the sequence number to see 1524 * if this is a valid reset segment. 1525 * RFC 793 page 37: 1526 * In all states except SYN-SENT, all reset (RST) segments 1527 * are validated by checking their SEQ-fields. A reset is 1528 * valid if its sequence number is in the window. 1529 * Note: this does not take into account delayed ACKs, so 1530 * we should test against last_ack_sent instead of rcv_nxt. 1531 * The sequence number in the reset segment is normally an 1532 * echo of our outgoing acknowlegement numbers, but some hosts 1533 * send a reset with the sequence number at the rightmost edge 1534 * of our receive window, and we have to handle this case. 1535 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1536 * that brute force RST attacks are possible. To combat this, 1537 * we use a much stricter check while in the ESTABLISHED state, 1538 * only accepting RSTs where the sequence number is equal to 1539 * last_ack_sent. In all other states (the states in which a 1540 * RST is more likely), the more permissive check is used. 1541 * If we have multiple segments in flight, the intial reset 1542 * segment sequence numbers will be to the left of last_ack_sent, 1543 * but they will eventually catch up. 1544 * In any case, it never made sense to trim reset segments to 1545 * fit the receive window since RFC 1122 says: 1546 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1547 * 1548 * A TCP SHOULD allow a received RST segment to include data. 1549 * 1550 * DISCUSSION 1551 * It has been suggested that a RST segment could contain 1552 * ASCII text that encoded and explained the cause of the 1553 * RST. No standard has yet been established for such 1554 * data. 1555 * 1556 * If the reset segment passes the sequence number test examine 1557 * the state: 1558 * SYN_RECEIVED STATE: 1559 * If passive open, return to LISTEN state. 1560 * If active open, inform user that connection was refused. 1561 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1562 * Inform user that connection was reset, and close tcb. 1563 * CLOSING, LAST_ACK STATES: 1564 * Close the tcb. 1565 * TIME_WAIT STATE: 1566 * Drop the segment - see Stevens, vol. 2, p. 964 and 1567 * RFC 1337. 1568 */ 1569 if (thflags & TH_RST) { 1570 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1571 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1572 switch (tp->t_state) { 1573 1574 case TCPS_SYN_RECEIVED: 1575 so->so_error = ECONNREFUSED; 1576 goto close; 1577 1578 case TCPS_ESTABLISHED: 1579 if (tp->last_ack_sent != th->th_seq) { 1580 tcpstat.tcps_badrst++; 1581 goto drop; 1582 } 1583 case TCPS_FIN_WAIT_1: 1584 case TCPS_FIN_WAIT_2: 1585 case TCPS_CLOSE_WAIT: 1586 so->so_error = ECONNRESET; 1587 close: 1588 tp->t_state = TCPS_CLOSED; 1589 tcpstat.tcps_drops++; 1590 tp = tcp_close(tp); 1591 break; 1592 1593 case TCPS_CLOSING: 1594 case TCPS_LAST_ACK: 1595 tp = tcp_close(tp); 1596 break; 1597 1598 case TCPS_TIME_WAIT: 1599 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1600 ("timewait")); 1601 break; 1602 } 1603 } 1604 goto drop; 1605 } 1606 1607 /* 1608 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1609 * and it's less than ts_recent, drop it. 1610 */ 1611 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1612 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1613 1614 /* Check to see if ts_recent is over 24 days old. */ 1615 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1616 /* 1617 * Invalidate ts_recent. If this segment updates 1618 * ts_recent, the age will be reset later and ts_recent 1619 * will get a valid value. If it does not, setting 1620 * ts_recent to zero will at least satisfy the 1621 * requirement that zero be placed in the timestamp 1622 * echo reply when ts_recent isn't valid. The 1623 * age isn't reset until we get a valid ts_recent 1624 * because we don't want out-of-order segments to be 1625 * dropped when ts_recent is old. 1626 */ 1627 tp->ts_recent = 0; 1628 } else { 1629 tcpstat.tcps_rcvduppack++; 1630 tcpstat.tcps_rcvdupbyte += tlen; 1631 tcpstat.tcps_pawsdrop++; 1632 if (tlen) 1633 goto dropafterack; 1634 goto drop; 1635 } 1636 } 1637 1638 /* 1639 * T/TCP mechanism 1640 * If T/TCP was negotiated and the segment doesn't have CC, 1641 * or if its CC is wrong then drop the segment. 1642 * RST segments do not have to comply with this. 1643 */ 1644 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1645 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1646 goto dropafterack; 1647 1648 /* 1649 * In the SYN-RECEIVED state, validate that the packet belongs to 1650 * this connection before trimming the data to fit the receive 1651 * window. Check the sequence number versus IRS since we know 1652 * the sequence numbers haven't wrapped. This is a partial fix 1653 * for the "LAND" DoS attack. 1654 */ 1655 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1656 rstreason = BANDLIM_RST_OPENPORT; 1657 goto dropwithreset; 1658 } 1659 1660 todrop = tp->rcv_nxt - th->th_seq; 1661 if (todrop > 0) { 1662 if (thflags & TH_SYN) { 1663 thflags &= ~TH_SYN; 1664 th->th_seq++; 1665 if (th->th_urp > 1) 1666 th->th_urp--; 1667 else 1668 thflags &= ~TH_URG; 1669 todrop--; 1670 } 1671 /* 1672 * Following if statement from Stevens, vol. 2, p. 960. 1673 */ 1674 if (todrop > tlen 1675 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1676 /* 1677 * Any valid FIN must be to the left of the window. 1678 * At this point the FIN must be a duplicate or out 1679 * of sequence; drop it. 1680 */ 1681 thflags &= ~TH_FIN; 1682 1683 /* 1684 * Send an ACK to resynchronize and drop any data. 1685 * But keep on processing for RST or ACK. 1686 */ 1687 tp->t_flags |= TF_ACKNOW; 1688 todrop = tlen; 1689 tcpstat.tcps_rcvduppack++; 1690 tcpstat.tcps_rcvdupbyte += todrop; 1691 } else { 1692 tcpstat.tcps_rcvpartduppack++; 1693 tcpstat.tcps_rcvpartdupbyte += todrop; 1694 } 1695 drop_hdrlen += todrop; /* drop from the top afterwards */ 1696 th->th_seq += todrop; 1697 tlen -= todrop; 1698 if (th->th_urp > todrop) 1699 th->th_urp -= todrop; 1700 else { 1701 thflags &= ~TH_URG; 1702 th->th_urp = 0; 1703 } 1704 } 1705 1706 /* 1707 * If new data are received on a connection after the 1708 * user processes are gone, then RST the other end. 1709 */ 1710 if ((so->so_state & SS_NOFDREF) && 1711 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1712 tp = tcp_close(tp); 1713 tcpstat.tcps_rcvafterclose++; 1714 rstreason = BANDLIM_UNLIMITED; 1715 goto dropwithreset; 1716 } 1717 1718 /* 1719 * If segment ends after window, drop trailing data 1720 * (and PUSH and FIN); if nothing left, just ACK. 1721 */ 1722 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1723 if (todrop > 0) { 1724 tcpstat.tcps_rcvpackafterwin++; 1725 if (todrop >= tlen) { 1726 tcpstat.tcps_rcvbyteafterwin += tlen; 1727 /* 1728 * If a new connection request is received 1729 * while in TIME_WAIT, drop the old connection 1730 * and start over if the sequence numbers 1731 * are above the previous ones. 1732 */ 1733 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1734 if (thflags & TH_SYN && 1735 tp->t_state == TCPS_TIME_WAIT && 1736 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1737 tp = tcp_close(tp); 1738 goto findpcb; 1739 } 1740 /* 1741 * If window is closed can only take segments at 1742 * window edge, and have to drop data and PUSH from 1743 * incoming segments. Continue processing, but 1744 * remember to ack. Otherwise, drop segment 1745 * and ack. 1746 */ 1747 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1748 tp->t_flags |= TF_ACKNOW; 1749 tcpstat.tcps_rcvwinprobe++; 1750 } else 1751 goto dropafterack; 1752 } else 1753 tcpstat.tcps_rcvbyteafterwin += todrop; 1754 m_adj(m, -todrop); 1755 tlen -= todrop; 1756 thflags &= ~(TH_PUSH|TH_FIN); 1757 } 1758 1759 /* 1760 * If last ACK falls within this segment's sequence numbers, 1761 * record its timestamp. 1762 * NOTE that the test is modified according to the latest 1763 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1764 */ 1765 if ((to.to_flags & TOF_TS) != 0 && 1766 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1767 tp->ts_recent_age = ticks; 1768 tp->ts_recent = to.to_tsval; 1769 } 1770 1771 /* 1772 * If a SYN is in the window, then this is an 1773 * error and we send an RST and drop the connection. 1774 */ 1775 if (thflags & TH_SYN) { 1776 tp = tcp_drop(tp, ECONNRESET); 1777 rstreason = BANDLIM_UNLIMITED; 1778 goto drop; 1779 } 1780 1781 /* 1782 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1783 * flag is on (half-synchronized state), then queue data for 1784 * later processing; else drop segment and return. 1785 */ 1786 if ((thflags & TH_ACK) == 0) { 1787 if (tp->t_state == TCPS_SYN_RECEIVED || 1788 (tp->t_flags & TF_NEEDSYN)) 1789 goto step6; 1790 else 1791 goto drop; 1792 } 1793 1794 /* 1795 * Ack processing. 1796 */ 1797 switch (tp->t_state) { 1798 1799 /* 1800 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1801 * ESTABLISHED state and continue processing. 1802 * The ACK was checked above. 1803 */ 1804 case TCPS_SYN_RECEIVED: 1805 1806 tcpstat.tcps_connects++; 1807 soisconnected(so); 1808 /* Do window scaling? */ 1809 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1810 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1811 tp->snd_scale = tp->requested_s_scale; 1812 tp->rcv_scale = tp->request_r_scale; 1813 } 1814 /* 1815 * Upon successful completion of 3-way handshake, 1816 * update cache.CC, pass any queued data to the user, 1817 * and advance state appropriately. 1818 */ 1819 if (tcp_do_rfc1644) { 1820 tao.tao_cc = tp->cc_recv; 1821 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC, 1822 tp->cc_recv, 0); 1823 } 1824 /* 1825 * Make transitions: 1826 * SYN-RECEIVED -> ESTABLISHED 1827 * SYN-RECEIVED* -> FIN-WAIT-1 1828 */ 1829 tp->t_starttime = ticks; 1830 if (tp->t_flags & TF_NEEDFIN) { 1831 tp->t_state = TCPS_FIN_WAIT_1; 1832 tp->t_flags &= ~TF_NEEDFIN; 1833 } else { 1834 tp->t_state = TCPS_ESTABLISHED; 1835 callout_reset(tp->tt_keep, tcp_keepidle, 1836 tcp_timer_keep, tp); 1837 } 1838 /* 1839 * If segment contains data or ACK, will call tcp_reass() 1840 * later; if not, do so now to pass queued data to user. 1841 */ 1842 if (tlen == 0 && (thflags & TH_FIN) == 0) 1843 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1844 (struct mbuf *)0); 1845 tp->snd_wl1 = th->th_seq - 1; 1846 /* FALLTHROUGH */ 1847 1848 /* 1849 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1850 * ACKs. If the ack is in the range 1851 * tp->snd_una < th->th_ack <= tp->snd_max 1852 * then advance tp->snd_una to th->th_ack and drop 1853 * data from the retransmission queue. If this ACK reflects 1854 * more up to date window information we update our window information. 1855 */ 1856 case TCPS_ESTABLISHED: 1857 case TCPS_FIN_WAIT_1: 1858 case TCPS_FIN_WAIT_2: 1859 case TCPS_CLOSE_WAIT: 1860 case TCPS_CLOSING: 1861 case TCPS_LAST_ACK: 1862 case TCPS_TIME_WAIT: 1863 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1864 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1865 if (tlen == 0 && tiwin == tp->snd_wnd) { 1866 tcpstat.tcps_rcvdupack++; 1867 /* 1868 * If we have outstanding data (other than 1869 * a window probe), this is a completely 1870 * duplicate ack (ie, window info didn't 1871 * change), the ack is the biggest we've 1872 * seen and we've seen exactly our rexmt 1873 * threshhold of them, assume a packet 1874 * has been dropped and retransmit it. 1875 * Kludge snd_nxt & the congestion 1876 * window so we send only this one 1877 * packet. 1878 * 1879 * We know we're losing at the current 1880 * window size so do congestion avoidance 1881 * (set ssthresh to half the current window 1882 * and pull our congestion window back to 1883 * the new ssthresh). 1884 * 1885 * Dup acks mean that packets have left the 1886 * network (they're now cached at the receiver) 1887 * so bump cwnd by the amount in the receiver 1888 * to keep a constant cwnd packets in the 1889 * network. 1890 */ 1891 if (!callout_active(tp->tt_rexmt) || 1892 th->th_ack != tp->snd_una) 1893 tp->t_dupacks = 0; 1894 else if (++tp->t_dupacks > tcprexmtthresh || 1895 (tcp_do_newreno && 1896 IN_FASTRECOVERY(tp))) { 1897 tp->snd_cwnd += tp->t_maxseg; 1898 (void) tcp_output(tp); 1899 goto drop; 1900 } else if (tp->t_dupacks == tcprexmtthresh) { 1901 tcp_seq onxt = tp->snd_nxt; 1902 u_int win; 1903 if (tcp_do_newreno && 1904 SEQ_LEQ(th->th_ack, 1905 tp->snd_recover)) { 1906 tp->t_dupacks = 0; 1907 break; 1908 } 1909 win = min(tp->snd_wnd, tp->snd_cwnd) / 1910 2 / tp->t_maxseg; 1911 if (win < 2) 1912 win = 2; 1913 tp->snd_ssthresh = win * tp->t_maxseg; 1914 ENTER_FASTRECOVERY(tp); 1915 tp->snd_recover = tp->snd_max; 1916 callout_stop(tp->tt_rexmt); 1917 tp->t_rtttime = 0; 1918 tp->snd_nxt = th->th_ack; 1919 tp->snd_cwnd = tp->t_maxseg; 1920 (void) tcp_output(tp); 1921 KASSERT(tp->snd_limited <= 2, 1922 ("tp->snd_limited too big")); 1923 tp->snd_cwnd = tp->snd_ssthresh + 1924 tp->t_maxseg * 1925 (tp->t_dupacks - tp->snd_limited); 1926 if (SEQ_GT(onxt, tp->snd_nxt)) 1927 tp->snd_nxt = onxt; 1928 goto drop; 1929 } else if (tcp_do_rfc3042) { 1930 u_long oldcwnd = tp->snd_cwnd; 1931 tcp_seq oldsndmax = tp->snd_max; 1932 u_int sent; 1933 1934 KASSERT(tp->t_dupacks == 1 || 1935 tp->t_dupacks == 2, 1936 ("dupacks not 1 or 2")); 1937 if (tp->t_dupacks == 1) 1938 tp->snd_limited = 0; 1939 tp->snd_cwnd = 1940 (tp->snd_nxt - tp->snd_una) + 1941 (tp->t_dupacks - tp->snd_limited) * 1942 tp->t_maxseg; 1943 (void) tcp_output(tp); 1944 sent = tp->snd_max - oldsndmax; 1945 if (sent > tp->t_maxseg) { 1946 KASSERT((tp->t_dupacks == 2 && 1947 tp->snd_limited == 0) || 1948 (sent == tp->t_maxseg + 1 && 1949 tp->t_flags & TF_SENTFIN), 1950 ("sent too much")); 1951 tp->snd_limited = 2; 1952 } else if (sent > 0) 1953 ++tp->snd_limited; 1954 tp->snd_cwnd = oldcwnd; 1955 goto drop; 1956 } 1957 } else 1958 tp->t_dupacks = 0; 1959 break; 1960 } 1961 1962 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1963 1964 /* 1965 * If the congestion window was inflated to account 1966 * for the other side's cached packets, retract it. 1967 */ 1968 if (tcp_do_newreno) { 1969 if (IN_FASTRECOVERY(tp)) { 1970 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1971 tcp_newreno_partial_ack(tp, th); 1972 } else { 1973 /* 1974 * Window inflation should have left us 1975 * with approximately snd_ssthresh 1976 * outstanding data. 1977 * But in case we would be inclined to 1978 * send a burst, better to do it via 1979 * the slow start mechanism. 1980 */ 1981 if (SEQ_GT(th->th_ack + 1982 tp->snd_ssthresh, 1983 tp->snd_max)) 1984 tp->snd_cwnd = tp->snd_max - 1985 th->th_ack + 1986 tp->t_maxseg; 1987 else 1988 tp->snd_cwnd = tp->snd_ssthresh; 1989 } 1990 } 1991 } else { 1992 if (tp->t_dupacks >= tcprexmtthresh && 1993 tp->snd_cwnd > tp->snd_ssthresh) 1994 tp->snd_cwnd = tp->snd_ssthresh; 1995 } 1996 tp->t_dupacks = 0; 1997 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1998 tcpstat.tcps_rcvacktoomuch++; 1999 goto dropafterack; 2000 } 2001 /* 2002 * If we reach this point, ACK is not a duplicate, 2003 * i.e., it ACKs something we sent. 2004 */ 2005 if (tp->t_flags & TF_NEEDSYN) { 2006 /* 2007 * T/TCP: Connection was half-synchronized, and our 2008 * SYN has been ACK'd (so connection is now fully 2009 * synchronized). Go to non-starred state, 2010 * increment snd_una for ACK of SYN, and check if 2011 * we can do window scaling. 2012 */ 2013 tp->t_flags &= ~TF_NEEDSYN; 2014 tp->snd_una++; 2015 /* Do window scaling? */ 2016 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2017 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2018 tp->snd_scale = tp->requested_s_scale; 2019 tp->rcv_scale = tp->request_r_scale; 2020 } 2021 } 2022 2023process_ACK: 2024 acked = th->th_ack - tp->snd_una; 2025 tcpstat.tcps_rcvackpack++; 2026 tcpstat.tcps_rcvackbyte += acked; 2027 2028 /* 2029 * If we just performed our first retransmit, and the ACK 2030 * arrives within our recovery window, then it was a mistake 2031 * to do the retransmit in the first place. Recover our 2032 * original cwnd and ssthresh, and proceed to transmit where 2033 * we left off. 2034 */ 2035 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2036 ++tcpstat.tcps_sndrexmitbad; 2037 tp->snd_cwnd = tp->snd_cwnd_prev; 2038 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2039 tp->snd_recover = tp->snd_recover_prev; 2040 if (tp->t_flags & TF_WASFRECOVERY) 2041 ENTER_FASTRECOVERY(tp); 2042 tp->snd_nxt = tp->snd_max; 2043 tp->t_badrxtwin = 0; /* XXX probably not required */ 2044 } 2045 2046 /* 2047 * If we have a timestamp reply, update smoothed 2048 * round trip time. If no timestamp is present but 2049 * transmit timer is running and timed sequence 2050 * number was acked, update smoothed round trip time. 2051 * Since we now have an rtt measurement, cancel the 2052 * timer backoff (cf., Phil Karn's retransmit alg.). 2053 * Recompute the initial retransmit timer. 2054 * 2055 * Some boxes send broken timestamp replies 2056 * during the SYN+ACK phase, ignore 2057 * timestamps of 0 or we could calculate a 2058 * huge RTT and blow up the retransmit timer. 2059 */ 2060 if ((to.to_flags & TOF_TS) != 0 && 2061 to.to_tsecr) { 2062 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2063 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2064 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2065 } 2066 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2067 2068 /* 2069 * If all outstanding data is acked, stop retransmit 2070 * timer and remember to restart (more output or persist). 2071 * If there is more data to be acked, restart retransmit 2072 * timer, using current (possibly backed-off) value. 2073 */ 2074 if (th->th_ack == tp->snd_max) { 2075 callout_stop(tp->tt_rexmt); 2076 needoutput = 1; 2077 } else if (!callout_active(tp->tt_persist)) 2078 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2079 tcp_timer_rexmt, tp); 2080 2081 /* 2082 * If no data (only SYN) was ACK'd, 2083 * skip rest of ACK processing. 2084 */ 2085 if (acked == 0) 2086 goto step6; 2087 2088 /* 2089 * When new data is acked, open the congestion window. 2090 * If the window gives us less than ssthresh packets 2091 * in flight, open exponentially (maxseg per packet). 2092 * Otherwise open linearly: maxseg per window 2093 * (maxseg^2 / cwnd per packet). 2094 */ 2095 if (!tcp_do_newreno || !IN_FASTRECOVERY(tp)) { 2096 register u_int cw = tp->snd_cwnd; 2097 register u_int incr = tp->t_maxseg; 2098 if (cw > tp->snd_ssthresh) 2099 incr = incr * incr / cw; 2100 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2101 } 2102 if (acked > so->so_snd.sb_cc) { 2103 tp->snd_wnd -= so->so_snd.sb_cc; 2104 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2105 ourfinisacked = 1; 2106 } else { 2107 sbdrop(&so->so_snd, acked); 2108 tp->snd_wnd -= acked; 2109 ourfinisacked = 0; 2110 } 2111 sowwakeup(so); 2112 /* detect una wraparound */ 2113 if (tcp_do_newreno && !IN_FASTRECOVERY(tp) && 2114 SEQ_GT(tp->snd_una, tp->snd_recover) && 2115 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2116 tp->snd_recover = th->th_ack - 1; 2117 if (tcp_do_newreno && IN_FASTRECOVERY(tp) && 2118 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2119 EXIT_FASTRECOVERY(tp); 2120 tp->snd_una = th->th_ack; 2121 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2122 tp->snd_nxt = tp->snd_una; 2123 2124 switch (tp->t_state) { 2125 2126 /* 2127 * In FIN_WAIT_1 STATE in addition to the processing 2128 * for the ESTABLISHED state if our FIN is now acknowledged 2129 * then enter FIN_WAIT_2. 2130 */ 2131 case TCPS_FIN_WAIT_1: 2132 if (ourfinisacked) { 2133 /* 2134 * If we can't receive any more 2135 * data, then closing user can proceed. 2136 * Starting the timer is contrary to the 2137 * specification, but if we don't get a FIN 2138 * we'll hang forever. 2139 */ 2140 /* XXXjl 2141 * we should release the tp also, and use a 2142 * compressed state. 2143 */ 2144 if (so->so_state & SS_CANTRCVMORE) { 2145 soisdisconnected(so); 2146 callout_reset(tp->tt_2msl, tcp_maxidle, 2147 tcp_timer_2msl, tp); 2148 } 2149 tp->t_state = TCPS_FIN_WAIT_2; 2150 } 2151 break; 2152 2153 /* 2154 * In CLOSING STATE in addition to the processing for 2155 * the ESTABLISHED state if the ACK acknowledges our FIN 2156 * then enter the TIME-WAIT state, otherwise ignore 2157 * the segment. 2158 */ 2159 case TCPS_CLOSING: 2160 if (ourfinisacked) { 2161 KASSERT(headlocked, ("headlocked")); 2162 tcp_twstart(tp); 2163 INP_INFO_WUNLOCK(&tcbinfo); 2164 m_freem(m); 2165 return; 2166 } 2167 break; 2168 2169 /* 2170 * In LAST_ACK, we may still be waiting for data to drain 2171 * and/or to be acked, as well as for the ack of our FIN. 2172 * If our FIN is now acknowledged, delete the TCB, 2173 * enter the closed state and return. 2174 */ 2175 case TCPS_LAST_ACK: 2176 if (ourfinisacked) { 2177 tp = tcp_close(tp); 2178 goto drop; 2179 } 2180 break; 2181 2182 /* 2183 * In TIME_WAIT state the only thing that should arrive 2184 * is a retransmission of the remote FIN. Acknowledge 2185 * it and restart the finack timer. 2186 */ 2187 case TCPS_TIME_WAIT: 2188 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2189 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2190 tcp_timer_2msl, tp); 2191 goto dropafterack; 2192 } 2193 } 2194 2195step6: 2196 /* 2197 * Update window information. 2198 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2199 */ 2200 if ((thflags & TH_ACK) && 2201 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2202 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2203 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2204 /* keep track of pure window updates */ 2205 if (tlen == 0 && 2206 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2207 tcpstat.tcps_rcvwinupd++; 2208 tp->snd_wnd = tiwin; 2209 tp->snd_wl1 = th->th_seq; 2210 tp->snd_wl2 = th->th_ack; 2211 if (tp->snd_wnd > tp->max_sndwnd) 2212 tp->max_sndwnd = tp->snd_wnd; 2213 needoutput = 1; 2214 } 2215 2216 /* 2217 * Process segments with URG. 2218 */ 2219 if ((thflags & TH_URG) && th->th_urp && 2220 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2221 /* 2222 * This is a kludge, but if we receive and accept 2223 * random urgent pointers, we'll crash in 2224 * soreceive. It's hard to imagine someone 2225 * actually wanting to send this much urgent data. 2226 */ 2227 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2228 th->th_urp = 0; /* XXX */ 2229 thflags &= ~TH_URG; /* XXX */ 2230 goto dodata; /* XXX */ 2231 } 2232 /* 2233 * If this segment advances the known urgent pointer, 2234 * then mark the data stream. This should not happen 2235 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2236 * a FIN has been received from the remote side. 2237 * In these states we ignore the URG. 2238 * 2239 * According to RFC961 (Assigned Protocols), 2240 * the urgent pointer points to the last octet 2241 * of urgent data. We continue, however, 2242 * to consider it to indicate the first octet 2243 * of data past the urgent section as the original 2244 * spec states (in one of two places). 2245 */ 2246 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2247 tp->rcv_up = th->th_seq + th->th_urp; 2248 so->so_oobmark = so->so_rcv.sb_cc + 2249 (tp->rcv_up - tp->rcv_nxt) - 1; 2250 if (so->so_oobmark == 0) 2251 so->so_state |= SS_RCVATMARK; 2252 sohasoutofband(so); 2253 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2254 } 2255 /* 2256 * Remove out of band data so doesn't get presented to user. 2257 * This can happen independent of advancing the URG pointer, 2258 * but if two URG's are pending at once, some out-of-band 2259 * data may creep in... ick. 2260 */ 2261 if (th->th_urp <= (u_long)tlen && 2262 !(so->so_options & SO_OOBINLINE)) { 2263 /* hdr drop is delayed */ 2264 tcp_pulloutofband(so, th, m, drop_hdrlen); 2265 } 2266 } else { 2267 /* 2268 * If no out of band data is expected, 2269 * pull receive urgent pointer along 2270 * with the receive window. 2271 */ 2272 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2273 tp->rcv_up = tp->rcv_nxt; 2274 } 2275dodata: /* XXX */ 2276 KASSERT(headlocked, ("headlocked")); 2277 /* 2278 * Process the segment text, merging it into the TCP sequencing queue, 2279 * and arranging for acknowledgment of receipt if necessary. 2280 * This process logically involves adjusting tp->rcv_wnd as data 2281 * is presented to the user (this happens in tcp_usrreq.c, 2282 * case PRU_RCVD). If a FIN has already been received on this 2283 * connection then we just ignore the text. 2284 */ 2285 if ((tlen || (thflags & TH_FIN)) && 2286 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2287 m_adj(m, drop_hdrlen); /* delayed header drop */ 2288 /* 2289 * Insert segment which includes th into TCP reassembly queue 2290 * with control block tp. Set thflags to whether reassembly now 2291 * includes a segment with FIN. This handles the common case 2292 * inline (segment is the next to be received on an established 2293 * connection, and the queue is empty), avoiding linkage into 2294 * and removal from the queue and repetition of various 2295 * conversions. 2296 * Set DELACK for segments received in order, but ack 2297 * immediately when segments are out of order (so 2298 * fast retransmit can work). 2299 */ 2300 if (th->th_seq == tp->rcv_nxt && 2301 LIST_EMPTY(&tp->t_segq) && 2302 TCPS_HAVEESTABLISHED(tp->t_state)) { 2303 if (DELAY_ACK(tp)) 2304 tp->t_flags |= TF_DELACK; 2305 else 2306 tp->t_flags |= TF_ACKNOW; 2307 tp->rcv_nxt += tlen; 2308 thflags = th->th_flags & TH_FIN; 2309 tcpstat.tcps_rcvpack++; 2310 tcpstat.tcps_rcvbyte += tlen; 2311 ND6_HINT(tp); 2312 if (so->so_state & SS_CANTRCVMORE) 2313 m_freem(m); 2314 else 2315 sbappendstream(&so->so_rcv, m); 2316 sorwakeup(so); 2317 } else { 2318 thflags = tcp_reass(tp, th, &tlen, m); 2319 tp->t_flags |= TF_ACKNOW; 2320 } 2321 2322 /* 2323 * Note the amount of data that peer has sent into 2324 * our window, in order to estimate the sender's 2325 * buffer size. 2326 */ 2327 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2328 } else { 2329 m_freem(m); 2330 thflags &= ~TH_FIN; 2331 } 2332 2333 /* 2334 * If FIN is received ACK the FIN and let the user know 2335 * that the connection is closing. 2336 */ 2337 if (thflags & TH_FIN) { 2338 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2339 socantrcvmore(so); 2340 /* 2341 * If connection is half-synchronized 2342 * (ie NEEDSYN flag on) then delay ACK, 2343 * so it may be piggybacked when SYN is sent. 2344 * Otherwise, since we received a FIN then no 2345 * more input can be expected, send ACK now. 2346 */ 2347 if (tp->t_flags & TF_NEEDSYN) 2348 tp->t_flags |= TF_DELACK; 2349 else 2350 tp->t_flags |= TF_ACKNOW; 2351 tp->rcv_nxt++; 2352 } 2353 switch (tp->t_state) { 2354 2355 /* 2356 * In SYN_RECEIVED and ESTABLISHED STATES 2357 * enter the CLOSE_WAIT state. 2358 */ 2359 case TCPS_SYN_RECEIVED: 2360 tp->t_starttime = ticks; 2361 /*FALLTHROUGH*/ 2362 case TCPS_ESTABLISHED: 2363 tp->t_state = TCPS_CLOSE_WAIT; 2364 break; 2365 2366 /* 2367 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2368 * enter the CLOSING state. 2369 */ 2370 case TCPS_FIN_WAIT_1: 2371 tp->t_state = TCPS_CLOSING; 2372 break; 2373 2374 /* 2375 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2376 * starting the time-wait timer, turning off the other 2377 * standard timers. 2378 */ 2379 case TCPS_FIN_WAIT_2: 2380 KASSERT(headlocked == 1, ("headlocked should be 1")); 2381 tcp_twstart(tp); 2382 INP_INFO_WUNLOCK(&tcbinfo); 2383 return; 2384 2385 /* 2386 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2387 */ 2388 case TCPS_TIME_WAIT: 2389 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2390 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2391 tcp_timer_2msl, tp); 2392 break; 2393 } 2394 } 2395 INP_INFO_WUNLOCK(&tcbinfo); 2396#ifdef TCPDEBUG 2397 if (so->so_options & SO_DEBUG) 2398 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2399 &tcp_savetcp, 0); 2400#endif 2401 2402 /* 2403 * Return any desired output. 2404 */ 2405 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2406 (void) tcp_output(tp); 2407 2408check_delack: 2409 if (tp->t_flags & TF_DELACK) { 2410 tp->t_flags &= ~TF_DELACK; 2411 callout_reset(tp->tt_delack, tcp_delacktime, 2412 tcp_timer_delack, tp); 2413 } 2414 INP_UNLOCK(inp); 2415 return; 2416 2417dropafterack: 2418 /* 2419 * Generate an ACK dropping incoming segment if it occupies 2420 * sequence space, where the ACK reflects our state. 2421 * 2422 * We can now skip the test for the RST flag since all 2423 * paths to this code happen after packets containing 2424 * RST have been dropped. 2425 * 2426 * In the SYN-RECEIVED state, don't send an ACK unless the 2427 * segment we received passes the SYN-RECEIVED ACK test. 2428 * If it fails send a RST. This breaks the loop in the 2429 * "LAND" DoS attack, and also prevents an ACK storm 2430 * between two listening ports that have been sent forged 2431 * SYN segments, each with the source address of the other. 2432 */ 2433 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2434 (SEQ_GT(tp->snd_una, th->th_ack) || 2435 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2436 rstreason = BANDLIM_RST_OPENPORT; 2437 goto dropwithreset; 2438 } 2439#ifdef TCPDEBUG 2440 if (so->so_options & SO_DEBUG) 2441 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2442 &tcp_savetcp, 0); 2443#endif 2444 KASSERT(headlocked, ("headlocked should be 1")); 2445 INP_INFO_WUNLOCK(&tcbinfo); 2446 m_freem(m); 2447 tp->t_flags |= TF_ACKNOW; 2448 (void) tcp_output(tp); 2449 INP_UNLOCK(inp); 2450 return; 2451 2452dropwithreset: 2453 /* 2454 * Generate a RST, dropping incoming segment. 2455 * Make ACK acceptable to originator of segment. 2456 * Don't bother to respond if destination was broadcast/multicast. 2457 */ 2458 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2459 goto drop; 2460 if (isipv6) { 2461 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2462 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2463 goto drop; 2464 } else { 2465 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2466 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2467 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2468 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2469 goto drop; 2470 } 2471 /* IPv6 anycast check is done at tcp6_input() */ 2472 2473 /* 2474 * Perform bandwidth limiting. 2475 */ 2476 if (badport_bandlim(rstreason) < 0) 2477 goto drop; 2478 2479#ifdef TCPDEBUG 2480 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2481 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2482 &tcp_savetcp, 0); 2483#endif 2484 2485 if (thflags & TH_ACK) 2486 /* mtod() below is safe as long as hdr dropping is delayed */ 2487 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2488 TH_RST); 2489 else { 2490 if (thflags & TH_SYN) 2491 tlen++; 2492 /* mtod() below is safe as long as hdr dropping is delayed */ 2493 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2494 (tcp_seq)0, TH_RST|TH_ACK); 2495 } 2496 2497 if (tp) 2498 INP_UNLOCK(inp); 2499 if (headlocked) 2500 INP_INFO_WUNLOCK(&tcbinfo); 2501 return; 2502 2503drop: 2504 /* 2505 * Drop space held by incoming segment and return. 2506 */ 2507#ifdef TCPDEBUG 2508 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2509 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2510 &tcp_savetcp, 0); 2511#endif 2512 if (tp) 2513 INP_UNLOCK(inp); 2514 m_freem(m); 2515 if (headlocked) 2516 INP_INFO_WUNLOCK(&tcbinfo); 2517 return; 2518} 2519 2520/* 2521 * Parse TCP options and place in tcpopt. 2522 */ 2523static void 2524tcp_dooptions(to, cp, cnt, is_syn) 2525 struct tcpopt *to; 2526 u_char *cp; 2527 int cnt; 2528 int is_syn; 2529{ 2530 int opt, optlen; 2531 2532 to->to_flags = 0; 2533 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2534 opt = cp[0]; 2535 if (opt == TCPOPT_EOL) 2536 break; 2537 if (opt == TCPOPT_NOP) 2538 optlen = 1; 2539 else { 2540 if (cnt < 2) 2541 break; 2542 optlen = cp[1]; 2543 if (optlen < 2 || optlen > cnt) 2544 break; 2545 } 2546 switch (opt) { 2547 case TCPOPT_MAXSEG: 2548 if (optlen != TCPOLEN_MAXSEG) 2549 continue; 2550 if (!is_syn) 2551 continue; 2552 to->to_flags |= TOF_MSS; 2553 bcopy((char *)cp + 2, 2554 (char *)&to->to_mss, sizeof(to->to_mss)); 2555 to->to_mss = ntohs(to->to_mss); 2556 break; 2557 case TCPOPT_WINDOW: 2558 if (optlen != TCPOLEN_WINDOW) 2559 continue; 2560 if (! is_syn) 2561 continue; 2562 to->to_flags |= TOF_SCALE; 2563 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2564 break; 2565 case TCPOPT_TIMESTAMP: 2566 if (optlen != TCPOLEN_TIMESTAMP) 2567 continue; 2568 to->to_flags |= TOF_TS; 2569 bcopy((char *)cp + 2, 2570 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2571 to->to_tsval = ntohl(to->to_tsval); 2572 bcopy((char *)cp + 6, 2573 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2574 to->to_tsecr = ntohl(to->to_tsecr); 2575 break; 2576 case TCPOPT_CC: 2577 if (optlen != TCPOLEN_CC) 2578 continue; 2579 to->to_flags |= TOF_CC; 2580 bcopy((char *)cp + 2, 2581 (char *)&to->to_cc, sizeof(to->to_cc)); 2582 to->to_cc = ntohl(to->to_cc); 2583 break; 2584 case TCPOPT_CCNEW: 2585 if (optlen != TCPOLEN_CC) 2586 continue; 2587 if (!is_syn) 2588 continue; 2589 to->to_flags |= TOF_CCNEW; 2590 bcopy((char *)cp + 2, 2591 (char *)&to->to_cc, sizeof(to->to_cc)); 2592 to->to_cc = ntohl(to->to_cc); 2593 break; 2594 case TCPOPT_CCECHO: 2595 if (optlen != TCPOLEN_CC) 2596 continue; 2597 if (!is_syn) 2598 continue; 2599 to->to_flags |= TOF_CCECHO; 2600 bcopy((char *)cp + 2, 2601 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2602 to->to_ccecho = ntohl(to->to_ccecho); 2603 break; 2604#ifdef TCP_SIGNATURE 2605 /* 2606 * XXX In order to reply to a host which has set the 2607 * TCP_SIGNATURE option in its initial SYN, we have to 2608 * record the fact that the option was observed here 2609 * for the syncache code to perform the correct response. 2610 */ 2611 case TCPOPT_SIGNATURE: 2612 if (optlen != TCPOLEN_SIGNATURE) 2613 continue; 2614 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2615 break; 2616#endif 2617 default: 2618 continue; 2619 } 2620 } 2621} 2622 2623/* 2624 * Pull out of band byte out of a segment so 2625 * it doesn't appear in the user's data queue. 2626 * It is still reflected in the segment length for 2627 * sequencing purposes. 2628 */ 2629static void 2630tcp_pulloutofband(so, th, m, off) 2631 struct socket *so; 2632 struct tcphdr *th; 2633 register struct mbuf *m; 2634 int off; /* delayed to be droped hdrlen */ 2635{ 2636 int cnt = off + th->th_urp - 1; 2637 2638 while (cnt >= 0) { 2639 if (m->m_len > cnt) { 2640 char *cp = mtod(m, caddr_t) + cnt; 2641 struct tcpcb *tp = sototcpcb(so); 2642 2643 tp->t_iobc = *cp; 2644 tp->t_oobflags |= TCPOOB_HAVEDATA; 2645 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2646 m->m_len--; 2647 if (m->m_flags & M_PKTHDR) 2648 m->m_pkthdr.len--; 2649 return; 2650 } 2651 cnt -= m->m_len; 2652 m = m->m_next; 2653 if (m == 0) 2654 break; 2655 } 2656 panic("tcp_pulloutofband"); 2657} 2658 2659/* 2660 * Collect new round-trip time estimate 2661 * and update averages and current timeout. 2662 */ 2663static void 2664tcp_xmit_timer(tp, rtt) 2665 register struct tcpcb *tp; 2666 int rtt; 2667{ 2668 register int delta; 2669 2670 tcpstat.tcps_rttupdated++; 2671 tp->t_rttupdated++; 2672 if (tp->t_srtt != 0) { 2673 /* 2674 * srtt is stored as fixed point with 5 bits after the 2675 * binary point (i.e., scaled by 8). The following magic 2676 * is equivalent to the smoothing algorithm in rfc793 with 2677 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2678 * point). Adjust rtt to origin 0. 2679 */ 2680 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2681 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2682 2683 if ((tp->t_srtt += delta) <= 0) 2684 tp->t_srtt = 1; 2685 2686 /* 2687 * We accumulate a smoothed rtt variance (actually, a 2688 * smoothed mean difference), then set the retransmit 2689 * timer to smoothed rtt + 4 times the smoothed variance. 2690 * rttvar is stored as fixed point with 4 bits after the 2691 * binary point (scaled by 16). The following is 2692 * equivalent to rfc793 smoothing with an alpha of .75 2693 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2694 * rfc793's wired-in beta. 2695 */ 2696 if (delta < 0) 2697 delta = -delta; 2698 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2699 if ((tp->t_rttvar += delta) <= 0) 2700 tp->t_rttvar = 1; 2701 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2702 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2703 } else { 2704 /* 2705 * No rtt measurement yet - use the unsmoothed rtt. 2706 * Set the variance to half the rtt (so our first 2707 * retransmit happens at 3*rtt). 2708 */ 2709 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2710 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2711 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2712 } 2713 tp->t_rtttime = 0; 2714 tp->t_rxtshift = 0; 2715 2716 /* 2717 * the retransmit should happen at rtt + 4 * rttvar. 2718 * Because of the way we do the smoothing, srtt and rttvar 2719 * will each average +1/2 tick of bias. When we compute 2720 * the retransmit timer, we want 1/2 tick of rounding and 2721 * 1 extra tick because of +-1/2 tick uncertainty in the 2722 * firing of the timer. The bias will give us exactly the 2723 * 1.5 tick we need. But, because the bias is 2724 * statistical, we have to test that we don't drop below 2725 * the minimum feasible timer (which is 2 ticks). 2726 */ 2727 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2728 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2729 2730 /* 2731 * We received an ack for a packet that wasn't retransmitted; 2732 * it is probably safe to discard any error indications we've 2733 * received recently. This isn't quite right, but close enough 2734 * for now (a route might have failed after we sent a segment, 2735 * and the return path might not be symmetrical). 2736 */ 2737 tp->t_softerror = 0; 2738} 2739 2740/* 2741 * Determine a reasonable value for maxseg size. 2742 * If the route is known, check route for mtu. 2743 * If none, use an mss that can be handled on the outgoing 2744 * interface without forcing IP to fragment; if bigger than 2745 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2746 * to utilize large mbufs. If no route is found, route has no mtu, 2747 * or the destination isn't local, use a default, hopefully conservative 2748 * size (usually 512 or the default IP max size, but no more than the mtu 2749 * of the interface), as we can't discover anything about intervening 2750 * gateways or networks. We also initialize the congestion/slow start 2751 * window to be a single segment if the destination isn't local. 2752 * While looking at the routing entry, we also initialize other path-dependent 2753 * parameters from pre-set or cached values in the routing entry. 2754 * 2755 * Also take into account the space needed for options that we 2756 * send regularly. Make maxseg shorter by that amount to assure 2757 * that we can send maxseg amount of data even when the options 2758 * are present. Store the upper limit of the length of options plus 2759 * data in maxopd. 2760 * 2761 * 2762 * In case of T/TCP, we call this routine during implicit connection 2763 * setup as well (offer = -1), to initialize maxseg from the cached 2764 * MSS of our peer. 2765 * 2766 * NOTE that this routine is only called when we process an incoming 2767 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2768 */ 2769void 2770tcp_mss(tp, offer) 2771 struct tcpcb *tp; 2772 int offer; 2773{ 2774 int rtt, mss; 2775 u_long bufsize; 2776 u_long maxmtu; 2777 struct inpcb *inp = tp->t_inpcb; 2778 struct socket *so; 2779 struct hc_metrics_lite metrics; 2780 struct rmxp_tao tao; 2781 int origoffer = offer; 2782#ifdef INET6 2783 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2784 size_t min_protoh = isipv6 ? 2785 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2786 sizeof (struct tcpiphdr); 2787#else 2788 const size_t min_protoh = sizeof(struct tcpiphdr); 2789#endif 2790 bzero(&tao, sizeof(tao)); 2791 2792 /* initialize */ 2793#ifdef INET6 2794 if (isipv6) { 2795 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2796 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2797 } else 2798#endif 2799 { 2800 maxmtu = tcp_maxmtu(&inp->inp_inc); 2801 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2802 } 2803 so = inp->inp_socket; 2804 2805 /* 2806 * no route to sender, stay with default mss and return 2807 */ 2808 if (maxmtu == 0) 2809 return; 2810 2811 /* what have we got? */ 2812 switch (offer) { 2813 case 0: 2814 /* 2815 * Offer == 0 means that there was no MSS on the SYN 2816 * segment, in this case we use tcp_mssdflt. 2817 */ 2818 offer = 2819#ifdef INET6 2820 isipv6 ? tcp_v6mssdflt : 2821#endif 2822 tcp_mssdflt; 2823 break; 2824 2825 case -1: 2826 /* 2827 * Offer == -1 means that we didn't receive SYN yet, 2828 * use cached value in that case; 2829 */ 2830 if (tcp_do_rfc1644) 2831 tcp_hc_gettao(&inp->inp_inc, &tao); 2832 if (tao.tao_mssopt != 0) 2833 offer = tao.tao_mssopt; 2834 /* FALLTHROUGH */ 2835 2836 default: 2837 /* 2838 * Prevent DoS attack with too small MSS. Round up 2839 * to at least minmss. 2840 */ 2841 offer = max(offer, tcp_minmss); 2842 /* 2843 * Sanity check: make sure that maxopd will be large 2844 * enough to allow some data on segments even if the 2845 * all the option space is used (40bytes). Otherwise 2846 * funny things may happen in tcp_output. 2847 */ 2848 offer = max(offer, 64); 2849 if (tcp_do_rfc1644) 2850 tcp_hc_updatetao(&inp->inp_inc, 2851 TCP_HC_TAO_MSSOPT, 0, offer); 2852 } 2853 2854 /* 2855 * rmx information is now retrieved from tcp_hostcache 2856 */ 2857 tcp_hc_get(&inp->inp_inc, &metrics); 2858 2859 /* 2860 * if there's a discovered mtu int tcp hostcache, use it 2861 * else, use the link mtu. 2862 */ 2863 if (metrics.rmx_mtu) 2864 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2865 else { 2866#ifdef INET6 2867 if (isipv6) { 2868 mss = maxmtu - min_protoh; 2869 if (!path_mtu_discovery && 2870 !in6_localaddr(&inp->in6p_faddr)) 2871 mss = min(mss, tcp_v6mssdflt); 2872 } else 2873#endif 2874 { 2875 mss = maxmtu - min_protoh; 2876 if (!path_mtu_discovery && 2877 !in_localaddr(inp->inp_faddr)) 2878 mss = min(mss, tcp_mssdflt); 2879 } 2880 } 2881 mss = min(mss, offer); 2882 2883 /* 2884 * maxopd stores the maximum length of data AND options 2885 * in a segment; maxseg is the amount of data in a normal 2886 * segment. We need to store this value (maxopd) apart 2887 * from maxseg, because now every segment carries options 2888 * and thus we normally have somewhat less data in segments. 2889 */ 2890 tp->t_maxopd = mss; 2891 2892 /* 2893 * In case of T/TCP, origoffer==-1 indicates, that no segments 2894 * were received yet. In this case we just guess, otherwise 2895 * we do the same as before T/TCP. 2896 */ 2897 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2898 (origoffer == -1 || 2899 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2900 mss -= TCPOLEN_TSTAMP_APPA; 2901 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2902 (origoffer == -1 || 2903 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2904 mss -= TCPOLEN_CC_APPA; 2905 tp->t_maxseg = mss; 2906 2907#if (MCLBYTES & (MCLBYTES - 1)) == 0 2908 if (mss > MCLBYTES) 2909 mss &= ~(MCLBYTES-1); 2910#else 2911 if (mss > MCLBYTES) 2912 mss = mss / MCLBYTES * MCLBYTES; 2913#endif 2914 tp->t_maxseg = mss; 2915 2916 /* 2917 * If there's a pipesize, change the socket buffer to that size, 2918 * don't change if sb_hiwat is different than default (then it 2919 * has been changed on purpose with setsockopt). 2920 * Make the socket buffers an integral number of mss units; 2921 * if the mss is larger than the socket buffer, decrease the mss. 2922 */ 2923 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2924 bufsize = metrics.rmx_sendpipe; 2925 else 2926 bufsize = so->so_snd.sb_hiwat; 2927 if (bufsize < mss) 2928 mss = bufsize; 2929 else { 2930 bufsize = roundup(bufsize, mss); 2931 if (bufsize > sb_max) 2932 bufsize = sb_max; 2933 if (bufsize > so->so_snd.sb_hiwat) 2934 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2935 } 2936 tp->t_maxseg = mss; 2937 2938 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2939 bufsize = metrics.rmx_recvpipe; 2940 else 2941 bufsize = so->so_rcv.sb_hiwat; 2942 if (bufsize > mss) { 2943 bufsize = roundup(bufsize, mss); 2944 if (bufsize > sb_max) 2945 bufsize = sb_max; 2946 if (bufsize > so->so_rcv.sb_hiwat) 2947 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2948 } 2949 /* 2950 * While we're here, check the others too 2951 */ 2952 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2953 tp->t_srtt = rtt; 2954 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2955 tcpstat.tcps_usedrtt++; 2956 if (metrics.rmx_rttvar) { 2957 tp->t_rttvar = metrics.rmx_rttvar; 2958 tcpstat.tcps_usedrttvar++; 2959 } else { 2960 /* default variation is +- 1 rtt */ 2961 tp->t_rttvar = 2962 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2963 } 2964 TCPT_RANGESET(tp->t_rxtcur, 2965 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2966 tp->t_rttmin, TCPTV_REXMTMAX); 2967 } 2968 if (metrics.rmx_ssthresh) { 2969 /* 2970 * There's some sort of gateway or interface 2971 * buffer limit on the path. Use this to set 2972 * the slow start threshhold, but set the 2973 * threshold to no less than 2*mss. 2974 */ 2975 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2976 tcpstat.tcps_usedssthresh++; 2977 } 2978 if (metrics.rmx_bandwidth) 2979 tp->snd_bandwidth = metrics.rmx_bandwidth; 2980 2981 /* 2982 * Set the slow-start flight size depending on whether this 2983 * is a local network or not. 2984 * 2985 * Extend this so we cache the cwnd too and retrieve it here. 2986 * Make cwnd even bigger than RFC3390 suggests but only if we 2987 * have previous experience with the remote host. Be careful 2988 * not make cwnd bigger than remote receive window or our own 2989 * send socket buffer. Maybe put some additional upper bound 2990 * on the retrieved cwnd. Should do incremental updates to 2991 * hostcache when cwnd collapses so next connection doesn't 2992 * overloads the path again. 2993 * 2994 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2995 * We currently check only in syncache_socket for that. 2996 */ 2997#define TCP_METRICS_CWND 2998#ifdef TCP_METRICS_CWND 2999 if (metrics.rmx_cwnd) 3000 tp->snd_cwnd = max(mss, 3001 min(metrics.rmx_cwnd / 2, 3002 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3003 else 3004#endif 3005 if (tcp_do_rfc3390) 3006 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3007#ifdef INET6 3008 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3009 (!isipv6 && in_localaddr(inp->inp_faddr))) 3010#else 3011 else if (in_localaddr(inp->inp_faddr)) 3012#endif 3013 tp->snd_cwnd = mss * ss_fltsz_local; 3014 else 3015 tp->snd_cwnd = mss * ss_fltsz; 3016} 3017 3018/* 3019 * Determine the MSS option to send on an outgoing SYN. 3020 */ 3021int 3022tcp_mssopt(inc) 3023 struct in_conninfo *inc; 3024{ 3025 int mss = 0; 3026 u_long maxmtu = 0; 3027 u_long thcmtu = 0; 3028 size_t min_protoh; 3029#ifdef INET6 3030 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3031#endif 3032 3033 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3034 3035#ifdef INET6 3036 if (isipv6) { 3037 mss = tcp_v6mssdflt; 3038 maxmtu = tcp_maxmtu6(inc); 3039 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3040 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3041 } else 3042#endif 3043 { 3044 mss = tcp_mssdflt; 3045 maxmtu = tcp_maxmtu(inc); 3046 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3047 min_protoh = sizeof(struct tcpiphdr); 3048 } 3049 if (maxmtu && thcmtu) 3050 mss = min(maxmtu, thcmtu) - min_protoh; 3051 else if (maxmtu || thcmtu) 3052 mss = max(maxmtu, thcmtu) - min_protoh; 3053 3054 return (mss); 3055} 3056 3057 3058/* 3059 * On a partial ack arrives, force the retransmission of the 3060 * next unacknowledged segment. Do not clear tp->t_dupacks. 3061 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3062 * be started again. 3063 */ 3064static void 3065tcp_newreno_partial_ack(tp, th) 3066 struct tcpcb *tp; 3067 struct tcphdr *th; 3068{ 3069 tcp_seq onxt = tp->snd_nxt; 3070 u_long ocwnd = tp->snd_cwnd; 3071 3072 callout_stop(tp->tt_rexmt); 3073 tp->t_rtttime = 0; 3074 tp->snd_nxt = th->th_ack; 3075 /* 3076 * Set snd_cwnd to one segment beyond acknowledged offset. 3077 * (tp->snd_una has not yet been updated when this function is called.) 3078 */ 3079 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3080 tp->t_flags |= TF_ACKNOW; 3081 (void) tcp_output(tp); 3082 tp->snd_cwnd = ocwnd; 3083 if (SEQ_GT(onxt, tp->snd_nxt)) 3084 tp->snd_nxt = onxt; 3085 /* 3086 * Partial window deflation. Relies on fact that tp->snd_una 3087 * not updated yet. 3088 */ 3089 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3090} 3091 3092/* 3093 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3094 * looking for a pcb in the listen state. Returns 0 otherwise. 3095 */ 3096static int 3097tcp_timewait(tw, to, th, m, tlen) 3098 struct tcptw *tw; 3099 struct tcpopt *to; 3100 struct tcphdr *th; 3101 struct mbuf *m; 3102 int tlen; 3103{ 3104 int thflags; 3105 tcp_seq seq; 3106#ifdef INET6 3107 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3108#else 3109 const int isipv6 = 0; 3110#endif 3111 3112 thflags = th->th_flags; 3113 3114 /* 3115 * NOTE: for FIN_WAIT_2 (to be added later), 3116 * must validate sequence number before accepting RST 3117 */ 3118 3119 /* 3120 * If the segment contains RST: 3121 * Drop the segment - see Stevens, vol. 2, p. 964 and 3122 * RFC 1337. 3123 */ 3124 if (thflags & TH_RST) 3125 goto drop; 3126 3127 /* 3128 * If segment contains a SYN and CC [not CC.NEW] option: 3129 * if connection duration > MSL, drop packet and send RST; 3130 * 3131 * if SEG.CC > CCrecv then is new SYN. 3132 * Complete close and delete TCPCB. Then reprocess 3133 * segment, hoping to find new TCPCB in LISTEN state; 3134 * 3135 * else must be old SYN; drop it. 3136 * else do normal processing. 3137 */ 3138 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 3139 if ((ticks - tw->t_starttime) > tcp_msl) 3140 goto reset; 3141 if (CC_GT(to->to_cc, tw->cc_recv)) { 3142 (void) tcp_twclose(tw, 0); 3143 return (1); 3144 } 3145 goto drop; 3146 } 3147 3148#if 0 3149/* PAWS not needed at the moment */ 3150 /* 3151 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3152 * and it's less than ts_recent, drop it. 3153 */ 3154 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3155 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3156 if ((thflags & TH_ACK) == 0) 3157 goto drop; 3158 goto ack; 3159 } 3160 /* 3161 * ts_recent is never updated because we never accept new segments. 3162 */ 3163#endif 3164 3165 /* 3166 * If a new connection request is received 3167 * while in TIME_WAIT, drop the old connection 3168 * and start over if the sequence numbers 3169 * are above the previous ones. 3170 */ 3171 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3172 (void) tcp_twclose(tw, 0); 3173 return (1); 3174 } 3175 3176 /* 3177 * Drop the the segment if it does not contain an ACK. 3178 */ 3179 if ((thflags & TH_ACK) == 0) 3180 goto drop; 3181 3182 /* 3183 * Reset the 2MSL timer if this is a duplicate FIN. 3184 */ 3185 if (thflags & TH_FIN) { 3186 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3187 if (seq + 1 == tw->rcv_nxt) 3188 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3189 } 3190 3191 /* 3192 * Acknowledge the segment if it has data or is not a duplicate ACK. 3193 */ 3194 if (thflags != TH_ACK || tlen != 0 || 3195 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3196 tcp_twrespond(tw, TH_ACK); 3197 goto drop; 3198 3199reset: 3200 /* 3201 * Generate a RST, dropping incoming segment. 3202 * Make ACK acceptable to originator of segment. 3203 * Don't bother to respond if destination was broadcast/multicast. 3204 */ 3205 if (m->m_flags & (M_BCAST|M_MCAST)) 3206 goto drop; 3207 if (isipv6) { 3208 struct ip6_hdr *ip6; 3209 3210 /* IPv6 anycast check is done at tcp6_input() */ 3211 ip6 = mtod(m, struct ip6_hdr *); 3212 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3213 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3214 goto drop; 3215 } else { 3216 struct ip *ip; 3217 3218 ip = mtod(m, struct ip *); 3219 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3220 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3221 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3222 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3223 goto drop; 3224 } 3225 if (thflags & TH_ACK) { 3226 tcp_respond(NULL, 3227 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3228 } else { 3229 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3230 tcp_respond(NULL, 3231 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3232 } 3233 INP_UNLOCK(tw->tw_inpcb); 3234 return (0); 3235 3236drop: 3237 INP_UNLOCK(tw->tw_inpcb); 3238 m_freem(m); 3239 return (0); 3240} 3241