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#include "opt_tcp_sack.h" 41 42#include <sys/param.h> 43#include <sys/kernel.h> 44#include <sys/mac.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/proc.h> /* for proc0 declaration */ 48#include <sys/protosw.h> 49#include <sys/signalvar.h> 50#include <sys/socket.h> 51#include <sys/socketvar.h> 52#include <sys/sysctl.h> 53#include <sys/syslog.h> 54#include <sys/systm.h> 55 56#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 57 58#include <vm/uma.h> 59 60#include <net/if.h> 61#include <net/route.h> 62 63#include <netinet/in.h> 64#include <netinet/in_pcb.h> 65#include <netinet/in_systm.h> 66#include <netinet/in_var.h> 67#include <netinet/ip.h> 68#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 69#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 70#include <netinet/ip_var.h> 71#include <netinet/ip6.h> 72#include <netinet/icmp6.h> 73#include <netinet6/in6_pcb.h> 74#include <netinet6/ip6_var.h> 75#include <netinet6/nd6.h> 76#include <netinet/tcp.h> 77#include <netinet/tcp_fsm.h> 78#include <netinet/tcp_seq.h> 79#include <netinet/tcp_timer.h> 80#include <netinet/tcp_var.h> 81#include <netinet6/tcp6_var.h> 82#include <netinet/tcpip.h> 83#ifdef TCPDEBUG 84#include <netinet/tcp_debug.h> 85#endif /* TCPDEBUG */ 86 87#ifdef FAST_IPSEC 88#include <netipsec/ipsec.h> 89#include <netipsec/ipsec6.h> 90#endif /*FAST_IPSEC*/ 91 92#ifdef IPSEC 93#include <netinet6/ipsec.h> 94#include <netinet6/ipsec6.h> 95#include <netkey/key.h> 96#endif /*IPSEC*/ 97 98#include <machine/in_cksum.h> 99 100static const int tcprexmtthresh = 3; 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 134static int tcp_insecure_rst = 0; 135SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 136 &tcp_insecure_rst, 0, 137 "Follow the old (insecure) criteria for accepting RST packets."); 138 139SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 140 "TCP Segment Reassembly Queue"); 141 142static int tcp_reass_maxseg = 0; 143SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 144 &tcp_reass_maxseg, 0, 145 "Global maximum number of TCP Segments in Reassembly Queue"); 146 147int tcp_reass_qsize = 0; 148SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 149 &tcp_reass_qsize, 0, 150 "Global number of TCP Segments currently in Reassembly Queue"); 151 152static int tcp_reass_maxqlen = 48; 153SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 154 &tcp_reass_maxqlen, 0, 155 "Maximum number of TCP Segments per individual Reassembly Queue"); 156 157static int tcp_reass_overflows = 0; 158SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 159 &tcp_reass_overflows, 0, 160 "Global number of TCP Segment Reassembly Queue Overflows"); 161 162struct inpcbhead tcb; 163#define tcb6 tcb /* for KAME src sync over BSD*'s */ 164struct inpcbinfo tcbinfo; 165struct mtx *tcbinfo_mtx; 166 167static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 168 169static void tcp_pulloutofband(struct socket *, 170 struct tcphdr *, struct mbuf *, int); 171static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 172 struct mbuf *); 173static void tcp_xmit_timer(struct tcpcb *, int); 174static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 175static int tcp_timewait(struct tcptw *, struct tcpopt *, 176 struct tcphdr *, struct mbuf *, int); 177 178/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 179#ifdef INET6 180#define ND6_HINT(tp) \ 181do { \ 182 if ((tp) && (tp)->t_inpcb && \ 183 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 184 nd6_nud_hint(NULL, NULL, 0); \ 185} while (0) 186#else 187#define ND6_HINT(tp) 188#endif 189 190/* 191 * Indicate whether this ack should be delayed. We can delay the ack if 192 * - there is no delayed ack timer in progress and 193 * - our last ack wasn't a 0-sized window. We never want to delay 194 * the ack that opens up a 0-sized window and 195 * - delayed acks are enabled or 196 * - this is a half-synchronized T/TCP connection. 197 */ 198#define DELAY_ACK(tp) \ 199 ((!callout_active(tp->tt_delack) && \ 200 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 201 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 202 203/* Initialize TCP reassembly queue */ 204uma_zone_t tcp_reass_zone; 205void 206tcp_reass_init() 207{ 208 tcp_reass_maxseg = nmbclusters / 16; 209 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 210 &tcp_reass_maxseg); 211 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 212 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 213 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 214} 215 216static int 217tcp_reass(tp, th, tlenp, m) 218 register struct tcpcb *tp; 219 register struct tcphdr *th; 220 int *tlenp; 221 struct mbuf *m; 222{ 223 struct tseg_qent *q; 224 struct tseg_qent *p = NULL; 225 struct tseg_qent *nq; 226 struct tseg_qent *te = NULL; 227 struct socket *so = tp->t_inpcb->inp_socket; 228 int flags; 229 230 INP_LOCK_ASSERT(tp->t_inpcb); 231 232 /* 233 * XXX: tcp_reass() is rather inefficient with its data structures 234 * and should be rewritten (see NetBSD for optimizations). While 235 * doing that it should move to its own file tcp_reass.c. 236 */ 237 238 /* 239 * Call with th==NULL after become established to 240 * force pre-ESTABLISHED data up to user socket. 241 */ 242 if (th == NULL) 243 goto present; 244 245 /* 246 * Limit the number of segments in the reassembly queue to prevent 247 * holding on to too many segments (and thus running out of mbufs). 248 * Make sure to let the missing segment through which caused this 249 * queue. Always keep one global queue entry spare to be able to 250 * process the missing segment. 251 */ 252 if (th->th_seq != tp->rcv_nxt && 253 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 254 tp->t_segqlen >= tcp_reass_maxqlen)) { 255 tcp_reass_overflows++; 256 tcpstat.tcps_rcvmemdrop++; 257 m_freem(m); 258 *tlenp = 0; 259 return (0); 260 } 261 262 /* 263 * Allocate a new queue entry. If we can't, or hit the zone limit 264 * just drop the pkt. 265 */ 266 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 267 if (te == NULL) { 268 tcpstat.tcps_rcvmemdrop++; 269 m_freem(m); 270 *tlenp = 0; 271 return (0); 272 } 273 tp->t_segqlen++; 274 tcp_reass_qsize++; 275 276 /* 277 * Find a segment which begins after this one does. 278 */ 279 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 280 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 281 break; 282 p = q; 283 } 284 285 /* 286 * If there is a preceding segment, it may provide some of 287 * our data already. If so, drop the data from the incoming 288 * segment. If it provides all of our data, drop us. 289 */ 290 if (p != NULL) { 291 register int i; 292 /* conversion to int (in i) handles seq wraparound */ 293 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 294 if (i > 0) { 295 if (i >= *tlenp) { 296 tcpstat.tcps_rcvduppack++; 297 tcpstat.tcps_rcvdupbyte += *tlenp; 298 m_freem(m); 299 uma_zfree(tcp_reass_zone, te); 300 tp->t_segqlen--; 301 tcp_reass_qsize--; 302 /* 303 * Try to present any queued data 304 * at the left window edge to the user. 305 * This is needed after the 3-WHS 306 * completes. 307 */ 308 goto present; /* ??? */ 309 } 310 m_adj(m, i); 311 *tlenp -= i; 312 th->th_seq += i; 313 } 314 } 315 tcpstat.tcps_rcvoopack++; 316 tcpstat.tcps_rcvoobyte += *tlenp; 317 318 /* 319 * While we overlap succeeding segments trim them or, 320 * if they are completely covered, dequeue them. 321 */ 322 while (q) { 323 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 324 if (i <= 0) 325 break; 326 if (i < q->tqe_len) { 327 q->tqe_th->th_seq += i; 328 q->tqe_len -= i; 329 m_adj(q->tqe_m, i); 330 break; 331 } 332 333 nq = LIST_NEXT(q, tqe_q); 334 LIST_REMOVE(q, tqe_q); 335 m_freem(q->tqe_m); 336 uma_zfree(tcp_reass_zone, q); 337 tp->t_segqlen--; 338 tcp_reass_qsize--; 339 q = nq; 340 } 341 342 /* Insert the new segment queue entry into place. */ 343 te->tqe_m = m; 344 te->tqe_th = th; 345 te->tqe_len = *tlenp; 346 347 if (p == NULL) { 348 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 349 } else { 350 LIST_INSERT_AFTER(p, te, tqe_q); 351 } 352 353present: 354 /* 355 * Present data to user, advancing rcv_nxt through 356 * completed sequence space. 357 */ 358 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 359 return (0); 360 q = LIST_FIRST(&tp->t_segq); 361 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 362 return (0); 363 SOCKBUF_LOCK(&so->so_rcv); 364 do { 365 tp->rcv_nxt += q->tqe_len; 366 flags = q->tqe_th->th_flags & TH_FIN; 367 nq = LIST_NEXT(q, tqe_q); 368 LIST_REMOVE(q, tqe_q); 369 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 370 m_freem(q->tqe_m); 371 else 372 sbappendstream_locked(&so->so_rcv, q->tqe_m); 373 uma_zfree(tcp_reass_zone, q); 374 tp->t_segqlen--; 375 tcp_reass_qsize--; 376 q = nq; 377 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 378 ND6_HINT(tp); 379 sorwakeup_locked(so); 380 return (flags); 381} 382 383/* 384 * TCP input routine, follows pages 65-76 of the 385 * protocol specification dated September, 1981 very closely. 386 */ 387#ifdef INET6 388int 389tcp6_input(mp, offp, proto) 390 struct mbuf **mp; 391 int *offp, proto; 392{ 393 register struct mbuf *m = *mp; 394 struct in6_ifaddr *ia6; 395 396 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 397 398 /* 399 * draft-itojun-ipv6-tcp-to-anycast 400 * better place to put this in? 401 */ 402 ia6 = ip6_getdstifaddr(m); 403 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 404 struct ip6_hdr *ip6; 405 406 ip6 = mtod(m, struct ip6_hdr *); 407 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 408 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 409 return IPPROTO_DONE; 410 } 411 412 tcp_input(m, *offp); 413 return IPPROTO_DONE; 414} 415#endif 416 417void 418tcp_input(m, off0) 419 register struct mbuf *m; 420 int off0; 421{ 422 register struct tcphdr *th; 423 register struct ip *ip = NULL; 424 register struct ipovly *ipov; 425 register struct inpcb *inp = NULL; 426 u_char *optp = NULL; 427 int optlen = 0; 428 int len, tlen, off; 429 int drop_hdrlen; 430 register struct tcpcb *tp = 0; 431 register int thflags; 432 struct socket *so = 0; 433 int todrop, acked, ourfinisacked, needoutput = 0; 434 u_long tiwin; 435 struct tcpopt to; /* options in this segment */ 436 int headlocked = 0; 437#ifdef IPFIREWALL_FORWARD 438 struct m_tag *fwd_tag; 439#endif 440 int rstreason; /* For badport_bandlim accounting purposes */ 441 442 struct ip6_hdr *ip6 = NULL; 443#ifdef INET6 444 int isipv6; 445#else 446 const int isipv6 = 0; 447#endif 448 449#ifdef TCPDEBUG 450 /* 451 * The size of tcp_saveipgen must be the size of the max ip header, 452 * now IPv6. 453 */ 454 u_char tcp_saveipgen[40]; 455 struct tcphdr tcp_savetcp; 456 short ostate = 0; 457#endif 458 459#ifdef INET6 460 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 461#endif 462 bzero((char *)&to, sizeof(to)); 463 464 tcpstat.tcps_rcvtotal++; 465 466 if (isipv6) { 467#ifdef INET6 468 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 469 ip6 = mtod(m, struct ip6_hdr *); 470 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 471 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 472 tcpstat.tcps_rcvbadsum++; 473 goto drop; 474 } 475 th = (struct tcphdr *)((caddr_t)ip6 + off0); 476 477 /* 478 * Be proactive about unspecified IPv6 address in source. 479 * As we use all-zero to indicate unbounded/unconnected pcb, 480 * unspecified IPv6 address can be used to confuse us. 481 * 482 * Note that packets with unspecified IPv6 destination is 483 * already dropped in ip6_input. 484 */ 485 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 486 /* XXX stat */ 487 goto drop; 488 } 489#else 490 th = NULL; /* XXX: avoid compiler warning */ 491#endif 492 } else { 493 /* 494 * Get IP and TCP header together in first mbuf. 495 * Note: IP leaves IP header in first mbuf. 496 */ 497 if (off0 > sizeof (struct ip)) { 498 ip_stripoptions(m, (struct mbuf *)0); 499 off0 = sizeof(struct ip); 500 } 501 if (m->m_len < sizeof (struct tcpiphdr)) { 502 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 503 tcpstat.tcps_rcvshort++; 504 return; 505 } 506 } 507 ip = mtod(m, struct ip *); 508 ipov = (struct ipovly *)ip; 509 th = (struct tcphdr *)((caddr_t)ip + off0); 510 tlen = ip->ip_len; 511 512 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 513 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 514 th->th_sum = m->m_pkthdr.csum_data; 515 else 516 th->th_sum = in_pseudo(ip->ip_src.s_addr, 517 ip->ip_dst.s_addr, 518 htonl(m->m_pkthdr.csum_data + 519 ip->ip_len + 520 IPPROTO_TCP)); 521 th->th_sum ^= 0xffff; 522#ifdef TCPDEBUG 523 ipov->ih_len = (u_short)tlen; 524 ipov->ih_len = htons(ipov->ih_len); 525#endif 526 } else { 527 /* 528 * Checksum extended TCP header and data. 529 */ 530 len = sizeof (struct ip) + tlen; 531 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 532 ipov->ih_len = (u_short)tlen; 533 ipov->ih_len = htons(ipov->ih_len); 534 th->th_sum = in_cksum(m, len); 535 } 536 if (th->th_sum) { 537 tcpstat.tcps_rcvbadsum++; 538 goto drop; 539 } 540#ifdef INET6 541 /* Re-initialization for later version check */ 542 ip->ip_v = IPVERSION; 543#endif 544 } 545 546 /* 547 * Check that TCP offset makes sense, 548 * pull out TCP options and adjust length. XXX 549 */ 550 off = th->th_off << 2; 551 if (off < sizeof (struct tcphdr) || off > tlen) { 552 tcpstat.tcps_rcvbadoff++; 553 goto drop; 554 } 555 tlen -= off; /* tlen is used instead of ti->ti_len */ 556 if (off > sizeof (struct tcphdr)) { 557 if (isipv6) { 558#ifdef INET6 559 IP6_EXTHDR_CHECK(m, off0, off, ); 560 ip6 = mtod(m, struct ip6_hdr *); 561 th = (struct tcphdr *)((caddr_t)ip6 + off0); 562#endif 563 } else { 564 if (m->m_len < sizeof(struct ip) + off) { 565 if ((m = m_pullup(m, sizeof (struct ip) + off)) 566 == 0) { 567 tcpstat.tcps_rcvshort++; 568 return; 569 } 570 ip = mtod(m, struct ip *); 571 ipov = (struct ipovly *)ip; 572 th = (struct tcphdr *)((caddr_t)ip + off0); 573 } 574 } 575 optlen = off - sizeof (struct tcphdr); 576 optp = (u_char *)(th + 1); 577 } 578 thflags = th->th_flags; 579 580#ifdef TCP_DROP_SYNFIN 581 /* 582 * If the drop_synfin option is enabled, drop all packets with 583 * both the SYN and FIN bits set. This prevents e.g. nmap from 584 * identifying the TCP/IP stack. 585 * 586 * This is a violation of the TCP specification. 587 */ 588 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 589 goto drop; 590#endif 591 592 /* 593 * Convert TCP protocol specific fields to host format. 594 */ 595 th->th_seq = ntohl(th->th_seq); 596 th->th_ack = ntohl(th->th_ack); 597 th->th_win = ntohs(th->th_win); 598 th->th_urp = ntohs(th->th_urp); 599 600 /* 601 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 602 * until after ip6_savecontrol() is called and before other functions 603 * which don't want those proto headers. 604 * Because ip6_savecontrol() is going to parse the mbuf to 605 * search for data to be passed up to user-land, it wants mbuf 606 * parameters to be unchanged. 607 * XXX: the call of ip6_savecontrol() has been obsoleted based on 608 * latest version of the advanced API (20020110). 609 */ 610 drop_hdrlen = off0 + off; 611 612 /* 613 * Locate pcb for segment. 614 */ 615 INP_INFO_WLOCK(&tcbinfo); 616 headlocked = 1; 617findpcb: 618 KASSERT(headlocked, ("tcp_input: findpcb: head not locked")); 619#ifdef IPFIREWALL_FORWARD 620 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 621 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 622 623 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 624 struct sockaddr_in *next_hop; 625 626 next_hop = (struct sockaddr_in *)(fwd_tag+1); 627 /* 628 * Transparently forwarded. Pretend to be the destination. 629 * already got one like this? 630 */ 631 inp = in_pcblookup_hash(&tcbinfo, 632 ip->ip_src, th->th_sport, 633 ip->ip_dst, th->th_dport, 634 0, m->m_pkthdr.rcvif); 635 if (!inp) { 636 /* It's new. Try to find the ambushing socket. */ 637 inp = in_pcblookup_hash(&tcbinfo, 638 ip->ip_src, th->th_sport, 639 next_hop->sin_addr, 640 next_hop->sin_port ? 641 ntohs(next_hop->sin_port) : 642 th->th_dport, 643 1, m->m_pkthdr.rcvif); 644 } 645 /* Remove the tag from the packet. We don't need it anymore. */ 646 m_tag_delete(m, fwd_tag); 647 } else { 648#endif /* IPFIREWALL_FORWARD */ 649 if (isipv6) { 650#ifdef INET6 651 inp = in6_pcblookup_hash(&tcbinfo, 652 &ip6->ip6_src, th->th_sport, 653 &ip6->ip6_dst, th->th_dport, 654 1, m->m_pkthdr.rcvif); 655#endif 656 } else 657 inp = in_pcblookup_hash(&tcbinfo, 658 ip->ip_src, th->th_sport, 659 ip->ip_dst, th->th_dport, 660 1, m->m_pkthdr.rcvif); 661#ifdef IPFIREWALL_FORWARD 662 } 663#endif /* IPFIREWALL_FORWARD */ 664 665#if defined(IPSEC) || defined(FAST_IPSEC) 666#ifdef INET6 667 if (isipv6) { 668 if (inp != NULL && ipsec6_in_reject(m, inp)) { 669#ifdef IPSEC 670 ipsec6stat.in_polvio++; 671#endif 672 goto drop; 673 } 674 } else 675#endif /* INET6 */ 676 if (inp != NULL && ipsec4_in_reject(m, inp)) { 677#ifdef IPSEC 678 ipsecstat.in_polvio++; 679#endif 680 goto drop; 681 } 682#endif /*IPSEC || FAST_IPSEC*/ 683 684 /* 685 * If the state is CLOSED (i.e., TCB does not exist) then 686 * all data in the incoming segment is discarded. 687 * If the TCB exists but is in CLOSED state, it is embryonic, 688 * but should either do a listen or a connect soon. 689 */ 690 if (inp == NULL) { 691 if (log_in_vain) { 692#ifdef INET6 693 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 694#else 695 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 696#endif 697 698 if (isipv6) { 699#ifdef INET6 700 strcpy(dbuf, "["); 701 strcpy(sbuf, "["); 702 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 703 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 704 strcat(dbuf, "]"); 705 strcat(sbuf, "]"); 706#endif 707 } else { 708 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 709 strcpy(sbuf, inet_ntoa(ip->ip_src)); 710 } 711 switch (log_in_vain) { 712 case 1: 713 if ((thflags & TH_SYN) == 0) 714 break; 715 /* FALLTHROUGH */ 716 case 2: 717 log(LOG_INFO, 718 "Connection attempt to TCP %s:%d " 719 "from %s:%d flags:0x%02x\n", 720 dbuf, ntohs(th->th_dport), sbuf, 721 ntohs(th->th_sport), thflags); 722 break; 723 default: 724 break; 725 } 726 } 727 if (blackhole) { 728 switch (blackhole) { 729 case 1: 730 if (thflags & TH_SYN) 731 goto drop; 732 break; 733 case 2: 734 goto drop; 735 default: 736 goto drop; 737 } 738 } 739 rstreason = BANDLIM_RST_CLOSEDPORT; 740 goto dropwithreset; 741 } 742 INP_LOCK(inp); 743 if (inp->inp_vflag & INP_TIMEWAIT) { 744 /* 745 * The only option of relevance is TOF_CC, and only if 746 * present in a SYN segment. See tcp_timewait(). 747 */ 748 if (thflags & TH_SYN) 749 tcp_dooptions(&to, optp, optlen, 1); 750 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 751 &to, th, m, tlen)) 752 goto findpcb; 753 /* 754 * tcp_timewait unlocks inp. 755 */ 756 INP_INFO_WUNLOCK(&tcbinfo); 757 return; 758 } 759 tp = intotcpcb(inp); 760 if (tp == 0) { 761 INP_UNLOCK(inp); 762 rstreason = BANDLIM_RST_CLOSEDPORT; 763 goto dropwithreset; 764 } 765 if (tp->t_state == TCPS_CLOSED) 766 goto drop; 767 768 /* Unscale the window into a 32-bit value. */ 769 if ((thflags & TH_SYN) == 0) 770 tiwin = th->th_win << tp->snd_scale; 771 else 772 tiwin = th->th_win; 773 774#ifdef MAC 775 INP_LOCK_ASSERT(inp); 776 if (mac_check_inpcb_deliver(inp, m)) 777 goto drop; 778#endif 779 so = inp->inp_socket; 780#ifdef TCPDEBUG 781 if (so->so_options & SO_DEBUG) { 782 ostate = tp->t_state; 783 if (isipv6) 784 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 785 else 786 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 787 tcp_savetcp = *th; 788 } 789#endif 790 if (so->so_options & SO_ACCEPTCONN) { 791 struct in_conninfo inc; 792 793#ifdef INET6 794 inc.inc_isipv6 = isipv6; 795#endif 796 if (isipv6) { 797 inc.inc6_faddr = ip6->ip6_src; 798 inc.inc6_laddr = ip6->ip6_dst; 799 } else { 800 inc.inc_faddr = ip->ip_src; 801 inc.inc_laddr = ip->ip_dst; 802 } 803 inc.inc_fport = th->th_sport; 804 inc.inc_lport = th->th_dport; 805 806 /* 807 * If the state is LISTEN then ignore segment if it contains 808 * a RST. If the segment contains an ACK then it is bad and 809 * send a RST. If it does not contain a SYN then it is not 810 * interesting; drop it. 811 * 812 * If the state is SYN_RECEIVED (syncache) and seg contains 813 * an ACK, but not for our SYN/ACK, send a RST. If the seg 814 * contains a RST, check the sequence number to see if it 815 * is a valid reset segment. 816 */ 817 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 818 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 819 if (!syncache_expand(&inc, th, &so, m)) { 820 /* 821 * No syncache entry, or ACK was not 822 * for our SYN/ACK. Send a RST. 823 */ 824 tcpstat.tcps_badsyn++; 825 rstreason = BANDLIM_RST_OPENPORT; 826 goto dropwithreset; 827 } 828 if (so == NULL) { 829 /* 830 * Could not complete 3-way handshake, 831 * connection is being closed down, and 832 * syncache will free mbuf. 833 */ 834 INP_UNLOCK(inp); 835 INP_INFO_WUNLOCK(&tcbinfo); 836 return; 837 } 838 /* 839 * Socket is created in state SYN_RECEIVED. 840 * Continue processing segment. 841 */ 842 INP_UNLOCK(inp); 843 inp = sotoinpcb(so); 844 INP_LOCK(inp); 845 tp = intotcpcb(inp); 846 /* 847 * This is what would have happened in 848 * tcp_output() when the SYN,ACK was sent. 849 */ 850 tp->snd_up = tp->snd_una; 851 tp->snd_max = tp->snd_nxt = tp->iss + 1; 852 tp->last_ack_sent = tp->rcv_nxt; 853 /* 854 * RFC1323: The window in SYN & SYN/ACK 855 * segments is never scaled. 856 */ 857 tp->snd_wnd = tiwin; /* unscaled */ 858 goto after_listen; 859 } 860 if (thflags & TH_RST) { 861 syncache_chkrst(&inc, th); 862 goto drop; 863 } 864 if (thflags & TH_ACK) { 865 syncache_badack(&inc); 866 tcpstat.tcps_badsyn++; 867 rstreason = BANDLIM_RST_OPENPORT; 868 goto dropwithreset; 869 } 870 goto drop; 871 } 872 873 /* 874 * Segment's flags are (SYN) or (SYN|FIN). 875 */ 876#ifdef INET6 877 /* 878 * If deprecated address is forbidden, 879 * we do not accept SYN to deprecated interface 880 * address to prevent any new inbound connection from 881 * getting established. 882 * When we do not accept SYN, we send a TCP RST, 883 * with deprecated source address (instead of dropping 884 * it). We compromise it as it is much better for peer 885 * to send a RST, and RST will be the final packet 886 * for the exchange. 887 * 888 * If we do not forbid deprecated addresses, we accept 889 * the SYN packet. RFC2462 does not suggest dropping 890 * SYN in this case. 891 * If we decipher RFC2462 5.5.4, it says like this: 892 * 1. use of deprecated addr with existing 893 * communication is okay - "SHOULD continue to be 894 * used" 895 * 2. use of it with new communication: 896 * (2a) "SHOULD NOT be used if alternate address 897 * with sufficient scope is available" 898 * (2b) nothing mentioned otherwise. 899 * Here we fall into (2b) case as we have no choice in 900 * our source address selection - we must obey the peer. 901 * 902 * The wording in RFC2462 is confusing, and there are 903 * multiple description text for deprecated address 904 * handling - worse, they are not exactly the same. 905 * I believe 5.5.4 is the best one, so we follow 5.5.4. 906 */ 907 if (isipv6 && !ip6_use_deprecated) { 908 struct in6_ifaddr *ia6; 909 910 if ((ia6 = ip6_getdstifaddr(m)) && 911 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 912 INP_UNLOCK(inp); 913 tp = NULL; 914 rstreason = BANDLIM_RST_OPENPORT; 915 goto dropwithreset; 916 } 917 } 918#endif 919 /* 920 * If it is from this socket, drop it, it must be forged. 921 * Don't bother responding if the destination was a broadcast. 922 */ 923 if (th->th_dport == th->th_sport) { 924 if (isipv6) { 925 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 926 &ip6->ip6_src)) 927 goto drop; 928 } else { 929 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 930 goto drop; 931 } 932 } 933 /* 934 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 935 * 936 * Note that it is quite possible to receive unicast 937 * link-layer packets with a broadcast IP address. Use 938 * in_broadcast() to find them. 939 */ 940 if (m->m_flags & (M_BCAST|M_MCAST)) 941 goto drop; 942 if (isipv6) { 943 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 944 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 945 goto drop; 946 } else { 947 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 948 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 949 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 950 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 951 goto drop; 952 } 953 /* 954 * SYN appears to be valid; create compressed TCP state 955 * for syncache, or perform t/tcp connection. 956 */ 957 if (so->so_qlen <= so->so_qlimit) { 958#ifdef TCPDEBUG 959 if (so->so_options & SO_DEBUG) 960 tcp_trace(TA_INPUT, ostate, tp, 961 (void *)tcp_saveipgen, &tcp_savetcp, 0); 962#endif 963 tcp_dooptions(&to, optp, optlen, 1); 964 if (!syncache_add(&inc, &to, th, &so, m)) 965 goto drop; 966 if (so == NULL) { 967 /* 968 * Entry added to syncache, mbuf used to 969 * send SYN,ACK packet. 970 */ 971 KASSERT(headlocked, ("headlocked")); 972 INP_UNLOCK(inp); 973 INP_INFO_WUNLOCK(&tcbinfo); 974 return; 975 } 976 /* 977 * Segment passed TAO tests. 978 */ 979 INP_UNLOCK(inp); 980 inp = sotoinpcb(so); 981 INP_LOCK(inp); 982 tp = intotcpcb(inp); 983 tp->snd_wnd = tiwin; 984 tp->t_starttime = ticks; 985 tp->t_state = TCPS_ESTABLISHED; 986 987 /* 988 * T/TCP logic: 989 * If there is a FIN or if there is data, then 990 * delay SYN,ACK(SYN) in the hope of piggy-backing 991 * it on a response segment. Otherwise must send 992 * ACK now in case the other side is slow starting. 993 */ 994 if (thflags & TH_FIN || tlen != 0) 995 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 996 else 997 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 998 tcpstat.tcps_connects++; 999 soisconnected(so); 1000 goto trimthenstep6; 1001 } 1002 goto drop; 1003 } 1004after_listen: 1005 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1006 INP_LOCK_ASSERT(inp); 1007 1008 /* XXX temp debugging */ 1009 /* should not happen - syncache should pick up these connections */ 1010 if (tp->t_state == TCPS_LISTEN) 1011 panic("tcp_input: TCPS_LISTEN"); 1012 1013 /* 1014 * This is the second part of the MSS DoS prevention code (after 1015 * minmss on the sending side) and it deals with too many too small 1016 * tcp packets in a too short timeframe (1 second). 1017 * 1018 * For every full second we count the number of received packets 1019 * and bytes. If we get a lot of packets per second for this connection 1020 * (tcp_minmssoverload) we take a closer look at it and compute the 1021 * average packet size for the past second. If that is less than 1022 * tcp_minmss we get too many packets with very small payload which 1023 * is not good and burdens our system (and every packet generates 1024 * a wakeup to the process connected to our socket). We can reasonable 1025 * expect this to be small packet DoS attack to exhaust our CPU 1026 * cycles. 1027 * 1028 * Care has to be taken for the minimum packet overload value. This 1029 * value defines the minimum number of packets per second before we 1030 * start to worry. This must not be too low to avoid killing for 1031 * example interactive connections with many small packets like 1032 * telnet or SSH. 1033 * 1034 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1035 * this check. 1036 * 1037 * Account for packet if payload packet, skip over ACK, etc. 1038 */ 1039 if (tcp_minmss && tcp_minmssoverload && 1040 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1041 if ((unsigned int)(tp->rcv_second - ticks) < hz) { 1042 tp->rcv_pps++; 1043 tp->rcv_byps += tlen + off; 1044 if (tp->rcv_pps > tcp_minmssoverload) { 1045 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1046 printf("too many small tcp packets from " 1047 "%s:%u, av. %lubyte/packet, " 1048 "dropping connection\n", 1049#ifdef INET6 1050 isipv6 ? 1051 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1052#endif 1053 inet_ntoa(inp->inp_inc.inc_faddr), 1054 inp->inp_inc.inc_fport, 1055 tp->rcv_byps / tp->rcv_pps); 1056 KASSERT(headlocked, ("tcp_input: " 1057 "after_listen: tcp_drop: head " 1058 "not locked")); 1059 tp = tcp_drop(tp, ECONNRESET); 1060 tcpstat.tcps_minmssdrops++; 1061 goto drop; 1062 } 1063 } 1064 } else { 1065 tp->rcv_second = ticks + hz; 1066 tp->rcv_pps = 1; 1067 tp->rcv_byps = tlen + off; 1068 } 1069 } 1070 1071 /* 1072 * Segment received on connection. 1073 * Reset idle time and keep-alive timer. 1074 */ 1075 tp->t_rcvtime = ticks; 1076 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1077 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1078 1079 /* 1080 * Process options only when we get SYN/ACK back. The SYN case 1081 * for incoming connections is handled in tcp_syncache. 1082 * XXX this is traditional behavior, may need to be cleaned up. 1083 */ 1084 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
| 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#include "opt_tcp_sack.h" 41 42#include <sys/param.h> 43#include <sys/kernel.h> 44#include <sys/mac.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/proc.h> /* for proc0 declaration */ 48#include <sys/protosw.h> 49#include <sys/signalvar.h> 50#include <sys/socket.h> 51#include <sys/socketvar.h> 52#include <sys/sysctl.h> 53#include <sys/syslog.h> 54#include <sys/systm.h> 55 56#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 57 58#include <vm/uma.h> 59 60#include <net/if.h> 61#include <net/route.h> 62 63#include <netinet/in.h> 64#include <netinet/in_pcb.h> 65#include <netinet/in_systm.h> 66#include <netinet/in_var.h> 67#include <netinet/ip.h> 68#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 69#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 70#include <netinet/ip_var.h> 71#include <netinet/ip6.h> 72#include <netinet/icmp6.h> 73#include <netinet6/in6_pcb.h> 74#include <netinet6/ip6_var.h> 75#include <netinet6/nd6.h> 76#include <netinet/tcp.h> 77#include <netinet/tcp_fsm.h> 78#include <netinet/tcp_seq.h> 79#include <netinet/tcp_timer.h> 80#include <netinet/tcp_var.h> 81#include <netinet6/tcp6_var.h> 82#include <netinet/tcpip.h> 83#ifdef TCPDEBUG 84#include <netinet/tcp_debug.h> 85#endif /* TCPDEBUG */ 86 87#ifdef FAST_IPSEC 88#include <netipsec/ipsec.h> 89#include <netipsec/ipsec6.h> 90#endif /*FAST_IPSEC*/ 91 92#ifdef IPSEC 93#include <netinet6/ipsec.h> 94#include <netinet6/ipsec6.h> 95#include <netkey/key.h> 96#endif /*IPSEC*/ 97 98#include <machine/in_cksum.h> 99 100static const int tcprexmtthresh = 3; 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 134static int tcp_insecure_rst = 0; 135SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 136 &tcp_insecure_rst, 0, 137 "Follow the old (insecure) criteria for accepting RST packets."); 138 139SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 140 "TCP Segment Reassembly Queue"); 141 142static int tcp_reass_maxseg = 0; 143SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 144 &tcp_reass_maxseg, 0, 145 "Global maximum number of TCP Segments in Reassembly Queue"); 146 147int tcp_reass_qsize = 0; 148SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 149 &tcp_reass_qsize, 0, 150 "Global number of TCP Segments currently in Reassembly Queue"); 151 152static int tcp_reass_maxqlen = 48; 153SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 154 &tcp_reass_maxqlen, 0, 155 "Maximum number of TCP Segments per individual Reassembly Queue"); 156 157static int tcp_reass_overflows = 0; 158SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 159 &tcp_reass_overflows, 0, 160 "Global number of TCP Segment Reassembly Queue Overflows"); 161 162struct inpcbhead tcb; 163#define tcb6 tcb /* for KAME src sync over BSD*'s */ 164struct inpcbinfo tcbinfo; 165struct mtx *tcbinfo_mtx; 166 167static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 168 169static void tcp_pulloutofband(struct socket *, 170 struct tcphdr *, struct mbuf *, int); 171static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 172 struct mbuf *); 173static void tcp_xmit_timer(struct tcpcb *, int); 174static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 175static int tcp_timewait(struct tcptw *, struct tcpopt *, 176 struct tcphdr *, struct mbuf *, int); 177 178/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 179#ifdef INET6 180#define ND6_HINT(tp) \ 181do { \ 182 if ((tp) && (tp)->t_inpcb && \ 183 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 184 nd6_nud_hint(NULL, NULL, 0); \ 185} while (0) 186#else 187#define ND6_HINT(tp) 188#endif 189 190/* 191 * Indicate whether this ack should be delayed. We can delay the ack if 192 * - there is no delayed ack timer in progress and 193 * - our last ack wasn't a 0-sized window. We never want to delay 194 * the ack that opens up a 0-sized window and 195 * - delayed acks are enabled or 196 * - this is a half-synchronized T/TCP connection. 197 */ 198#define DELAY_ACK(tp) \ 199 ((!callout_active(tp->tt_delack) && \ 200 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 201 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 202 203/* Initialize TCP reassembly queue */ 204uma_zone_t tcp_reass_zone; 205void 206tcp_reass_init() 207{ 208 tcp_reass_maxseg = nmbclusters / 16; 209 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 210 &tcp_reass_maxseg); 211 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 212 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 213 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 214} 215 216static int 217tcp_reass(tp, th, tlenp, m) 218 register struct tcpcb *tp; 219 register struct tcphdr *th; 220 int *tlenp; 221 struct mbuf *m; 222{ 223 struct tseg_qent *q; 224 struct tseg_qent *p = NULL; 225 struct tseg_qent *nq; 226 struct tseg_qent *te = NULL; 227 struct socket *so = tp->t_inpcb->inp_socket; 228 int flags; 229 230 INP_LOCK_ASSERT(tp->t_inpcb); 231 232 /* 233 * XXX: tcp_reass() is rather inefficient with its data structures 234 * and should be rewritten (see NetBSD for optimizations). While 235 * doing that it should move to its own file tcp_reass.c. 236 */ 237 238 /* 239 * Call with th==NULL after become established to 240 * force pre-ESTABLISHED data up to user socket. 241 */ 242 if (th == NULL) 243 goto present; 244 245 /* 246 * Limit the number of segments in the reassembly queue to prevent 247 * holding on to too many segments (and thus running out of mbufs). 248 * Make sure to let the missing segment through which caused this 249 * queue. Always keep one global queue entry spare to be able to 250 * process the missing segment. 251 */ 252 if (th->th_seq != tp->rcv_nxt && 253 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 254 tp->t_segqlen >= tcp_reass_maxqlen)) { 255 tcp_reass_overflows++; 256 tcpstat.tcps_rcvmemdrop++; 257 m_freem(m); 258 *tlenp = 0; 259 return (0); 260 } 261 262 /* 263 * Allocate a new queue entry. If we can't, or hit the zone limit 264 * just drop the pkt. 265 */ 266 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 267 if (te == NULL) { 268 tcpstat.tcps_rcvmemdrop++; 269 m_freem(m); 270 *tlenp = 0; 271 return (0); 272 } 273 tp->t_segqlen++; 274 tcp_reass_qsize++; 275 276 /* 277 * Find a segment which begins after this one does. 278 */ 279 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 280 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 281 break; 282 p = q; 283 } 284 285 /* 286 * If there is a preceding segment, it may provide some of 287 * our data already. If so, drop the data from the incoming 288 * segment. If it provides all of our data, drop us. 289 */ 290 if (p != NULL) { 291 register int i; 292 /* conversion to int (in i) handles seq wraparound */ 293 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 294 if (i > 0) { 295 if (i >= *tlenp) { 296 tcpstat.tcps_rcvduppack++; 297 tcpstat.tcps_rcvdupbyte += *tlenp; 298 m_freem(m); 299 uma_zfree(tcp_reass_zone, te); 300 tp->t_segqlen--; 301 tcp_reass_qsize--; 302 /* 303 * Try to present any queued data 304 * at the left window edge to the user. 305 * This is needed after the 3-WHS 306 * completes. 307 */ 308 goto present; /* ??? */ 309 } 310 m_adj(m, i); 311 *tlenp -= i; 312 th->th_seq += i; 313 } 314 } 315 tcpstat.tcps_rcvoopack++; 316 tcpstat.tcps_rcvoobyte += *tlenp; 317 318 /* 319 * While we overlap succeeding segments trim them or, 320 * if they are completely covered, dequeue them. 321 */ 322 while (q) { 323 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 324 if (i <= 0) 325 break; 326 if (i < q->tqe_len) { 327 q->tqe_th->th_seq += i; 328 q->tqe_len -= i; 329 m_adj(q->tqe_m, i); 330 break; 331 } 332 333 nq = LIST_NEXT(q, tqe_q); 334 LIST_REMOVE(q, tqe_q); 335 m_freem(q->tqe_m); 336 uma_zfree(tcp_reass_zone, q); 337 tp->t_segqlen--; 338 tcp_reass_qsize--; 339 q = nq; 340 } 341 342 /* Insert the new segment queue entry into place. */ 343 te->tqe_m = m; 344 te->tqe_th = th; 345 te->tqe_len = *tlenp; 346 347 if (p == NULL) { 348 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 349 } else { 350 LIST_INSERT_AFTER(p, te, tqe_q); 351 } 352 353present: 354 /* 355 * Present data to user, advancing rcv_nxt through 356 * completed sequence space. 357 */ 358 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 359 return (0); 360 q = LIST_FIRST(&tp->t_segq); 361 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 362 return (0); 363 SOCKBUF_LOCK(&so->so_rcv); 364 do { 365 tp->rcv_nxt += q->tqe_len; 366 flags = q->tqe_th->th_flags & TH_FIN; 367 nq = LIST_NEXT(q, tqe_q); 368 LIST_REMOVE(q, tqe_q); 369 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 370 m_freem(q->tqe_m); 371 else 372 sbappendstream_locked(&so->so_rcv, q->tqe_m); 373 uma_zfree(tcp_reass_zone, q); 374 tp->t_segqlen--; 375 tcp_reass_qsize--; 376 q = nq; 377 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 378 ND6_HINT(tp); 379 sorwakeup_locked(so); 380 return (flags); 381} 382 383/* 384 * TCP input routine, follows pages 65-76 of the 385 * protocol specification dated September, 1981 very closely. 386 */ 387#ifdef INET6 388int 389tcp6_input(mp, offp, proto) 390 struct mbuf **mp; 391 int *offp, proto; 392{ 393 register struct mbuf *m = *mp; 394 struct in6_ifaddr *ia6; 395 396 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 397 398 /* 399 * draft-itojun-ipv6-tcp-to-anycast 400 * better place to put this in? 401 */ 402 ia6 = ip6_getdstifaddr(m); 403 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 404 struct ip6_hdr *ip6; 405 406 ip6 = mtod(m, struct ip6_hdr *); 407 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 408 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 409 return IPPROTO_DONE; 410 } 411 412 tcp_input(m, *offp); 413 return IPPROTO_DONE; 414} 415#endif 416 417void 418tcp_input(m, off0) 419 register struct mbuf *m; 420 int off0; 421{ 422 register struct tcphdr *th; 423 register struct ip *ip = NULL; 424 register struct ipovly *ipov; 425 register struct inpcb *inp = NULL; 426 u_char *optp = NULL; 427 int optlen = 0; 428 int len, tlen, off; 429 int drop_hdrlen; 430 register struct tcpcb *tp = 0; 431 register int thflags; 432 struct socket *so = 0; 433 int todrop, acked, ourfinisacked, needoutput = 0; 434 u_long tiwin; 435 struct tcpopt to; /* options in this segment */ 436 int headlocked = 0; 437#ifdef IPFIREWALL_FORWARD 438 struct m_tag *fwd_tag; 439#endif 440 int rstreason; /* For badport_bandlim accounting purposes */ 441 442 struct ip6_hdr *ip6 = NULL; 443#ifdef INET6 444 int isipv6; 445#else 446 const int isipv6 = 0; 447#endif 448 449#ifdef TCPDEBUG 450 /* 451 * The size of tcp_saveipgen must be the size of the max ip header, 452 * now IPv6. 453 */ 454 u_char tcp_saveipgen[40]; 455 struct tcphdr tcp_savetcp; 456 short ostate = 0; 457#endif 458 459#ifdef INET6 460 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 461#endif 462 bzero((char *)&to, sizeof(to)); 463 464 tcpstat.tcps_rcvtotal++; 465 466 if (isipv6) { 467#ifdef INET6 468 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 469 ip6 = mtod(m, struct ip6_hdr *); 470 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 471 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 472 tcpstat.tcps_rcvbadsum++; 473 goto drop; 474 } 475 th = (struct tcphdr *)((caddr_t)ip6 + off0); 476 477 /* 478 * Be proactive about unspecified IPv6 address in source. 479 * As we use all-zero to indicate unbounded/unconnected pcb, 480 * unspecified IPv6 address can be used to confuse us. 481 * 482 * Note that packets with unspecified IPv6 destination is 483 * already dropped in ip6_input. 484 */ 485 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 486 /* XXX stat */ 487 goto drop; 488 } 489#else 490 th = NULL; /* XXX: avoid compiler warning */ 491#endif 492 } else { 493 /* 494 * Get IP and TCP header together in first mbuf. 495 * Note: IP leaves IP header in first mbuf. 496 */ 497 if (off0 > sizeof (struct ip)) { 498 ip_stripoptions(m, (struct mbuf *)0); 499 off0 = sizeof(struct ip); 500 } 501 if (m->m_len < sizeof (struct tcpiphdr)) { 502 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 503 tcpstat.tcps_rcvshort++; 504 return; 505 } 506 } 507 ip = mtod(m, struct ip *); 508 ipov = (struct ipovly *)ip; 509 th = (struct tcphdr *)((caddr_t)ip + off0); 510 tlen = ip->ip_len; 511 512 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 513 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 514 th->th_sum = m->m_pkthdr.csum_data; 515 else 516 th->th_sum = in_pseudo(ip->ip_src.s_addr, 517 ip->ip_dst.s_addr, 518 htonl(m->m_pkthdr.csum_data + 519 ip->ip_len + 520 IPPROTO_TCP)); 521 th->th_sum ^= 0xffff; 522#ifdef TCPDEBUG 523 ipov->ih_len = (u_short)tlen; 524 ipov->ih_len = htons(ipov->ih_len); 525#endif 526 } else { 527 /* 528 * Checksum extended TCP header and data. 529 */ 530 len = sizeof (struct ip) + tlen; 531 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 532 ipov->ih_len = (u_short)tlen; 533 ipov->ih_len = htons(ipov->ih_len); 534 th->th_sum = in_cksum(m, len); 535 } 536 if (th->th_sum) { 537 tcpstat.tcps_rcvbadsum++; 538 goto drop; 539 } 540#ifdef INET6 541 /* Re-initialization for later version check */ 542 ip->ip_v = IPVERSION; 543#endif 544 } 545 546 /* 547 * Check that TCP offset makes sense, 548 * pull out TCP options and adjust length. XXX 549 */ 550 off = th->th_off << 2; 551 if (off < sizeof (struct tcphdr) || off > tlen) { 552 tcpstat.tcps_rcvbadoff++; 553 goto drop; 554 } 555 tlen -= off; /* tlen is used instead of ti->ti_len */ 556 if (off > sizeof (struct tcphdr)) { 557 if (isipv6) { 558#ifdef INET6 559 IP6_EXTHDR_CHECK(m, off0, off, ); 560 ip6 = mtod(m, struct ip6_hdr *); 561 th = (struct tcphdr *)((caddr_t)ip6 + off0); 562#endif 563 } else { 564 if (m->m_len < sizeof(struct ip) + off) { 565 if ((m = m_pullup(m, sizeof (struct ip) + off)) 566 == 0) { 567 tcpstat.tcps_rcvshort++; 568 return; 569 } 570 ip = mtod(m, struct ip *); 571 ipov = (struct ipovly *)ip; 572 th = (struct tcphdr *)((caddr_t)ip + off0); 573 } 574 } 575 optlen = off - sizeof (struct tcphdr); 576 optp = (u_char *)(th + 1); 577 } 578 thflags = th->th_flags; 579 580#ifdef TCP_DROP_SYNFIN 581 /* 582 * If the drop_synfin option is enabled, drop all packets with 583 * both the SYN and FIN bits set. This prevents e.g. nmap from 584 * identifying the TCP/IP stack. 585 * 586 * This is a violation of the TCP specification. 587 */ 588 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 589 goto drop; 590#endif 591 592 /* 593 * Convert TCP protocol specific fields to host format. 594 */ 595 th->th_seq = ntohl(th->th_seq); 596 th->th_ack = ntohl(th->th_ack); 597 th->th_win = ntohs(th->th_win); 598 th->th_urp = ntohs(th->th_urp); 599 600 /* 601 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 602 * until after ip6_savecontrol() is called and before other functions 603 * which don't want those proto headers. 604 * Because ip6_savecontrol() is going to parse the mbuf to 605 * search for data to be passed up to user-land, it wants mbuf 606 * parameters to be unchanged. 607 * XXX: the call of ip6_savecontrol() has been obsoleted based on 608 * latest version of the advanced API (20020110). 609 */ 610 drop_hdrlen = off0 + off; 611 612 /* 613 * Locate pcb for segment. 614 */ 615 INP_INFO_WLOCK(&tcbinfo); 616 headlocked = 1; 617findpcb: 618 KASSERT(headlocked, ("tcp_input: findpcb: head not locked")); 619#ifdef IPFIREWALL_FORWARD 620 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 621 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 622 623 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 624 struct sockaddr_in *next_hop; 625 626 next_hop = (struct sockaddr_in *)(fwd_tag+1); 627 /* 628 * Transparently forwarded. Pretend to be the destination. 629 * already got one like this? 630 */ 631 inp = in_pcblookup_hash(&tcbinfo, 632 ip->ip_src, th->th_sport, 633 ip->ip_dst, th->th_dport, 634 0, m->m_pkthdr.rcvif); 635 if (!inp) { 636 /* It's new. Try to find the ambushing socket. */ 637 inp = in_pcblookup_hash(&tcbinfo, 638 ip->ip_src, th->th_sport, 639 next_hop->sin_addr, 640 next_hop->sin_port ? 641 ntohs(next_hop->sin_port) : 642 th->th_dport, 643 1, m->m_pkthdr.rcvif); 644 } 645 /* Remove the tag from the packet. We don't need it anymore. */ 646 m_tag_delete(m, fwd_tag); 647 } else { 648#endif /* IPFIREWALL_FORWARD */ 649 if (isipv6) { 650#ifdef INET6 651 inp = in6_pcblookup_hash(&tcbinfo, 652 &ip6->ip6_src, th->th_sport, 653 &ip6->ip6_dst, th->th_dport, 654 1, m->m_pkthdr.rcvif); 655#endif 656 } else 657 inp = in_pcblookup_hash(&tcbinfo, 658 ip->ip_src, th->th_sport, 659 ip->ip_dst, th->th_dport, 660 1, m->m_pkthdr.rcvif); 661#ifdef IPFIREWALL_FORWARD 662 } 663#endif /* IPFIREWALL_FORWARD */ 664 665#if defined(IPSEC) || defined(FAST_IPSEC) 666#ifdef INET6 667 if (isipv6) { 668 if (inp != NULL && ipsec6_in_reject(m, inp)) { 669#ifdef IPSEC 670 ipsec6stat.in_polvio++; 671#endif 672 goto drop; 673 } 674 } else 675#endif /* INET6 */ 676 if (inp != NULL && ipsec4_in_reject(m, inp)) { 677#ifdef IPSEC 678 ipsecstat.in_polvio++; 679#endif 680 goto drop; 681 } 682#endif /*IPSEC || FAST_IPSEC*/ 683 684 /* 685 * If the state is CLOSED (i.e., TCB does not exist) then 686 * all data in the incoming segment is discarded. 687 * If the TCB exists but is in CLOSED state, it is embryonic, 688 * but should either do a listen or a connect soon. 689 */ 690 if (inp == NULL) { 691 if (log_in_vain) { 692#ifdef INET6 693 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 694#else 695 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 696#endif 697 698 if (isipv6) { 699#ifdef INET6 700 strcpy(dbuf, "["); 701 strcpy(sbuf, "["); 702 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 703 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 704 strcat(dbuf, "]"); 705 strcat(sbuf, "]"); 706#endif 707 } else { 708 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 709 strcpy(sbuf, inet_ntoa(ip->ip_src)); 710 } 711 switch (log_in_vain) { 712 case 1: 713 if ((thflags & TH_SYN) == 0) 714 break; 715 /* FALLTHROUGH */ 716 case 2: 717 log(LOG_INFO, 718 "Connection attempt to TCP %s:%d " 719 "from %s:%d flags:0x%02x\n", 720 dbuf, ntohs(th->th_dport), sbuf, 721 ntohs(th->th_sport), thflags); 722 break; 723 default: 724 break; 725 } 726 } 727 if (blackhole) { 728 switch (blackhole) { 729 case 1: 730 if (thflags & TH_SYN) 731 goto drop; 732 break; 733 case 2: 734 goto drop; 735 default: 736 goto drop; 737 } 738 } 739 rstreason = BANDLIM_RST_CLOSEDPORT; 740 goto dropwithreset; 741 } 742 INP_LOCK(inp); 743 if (inp->inp_vflag & INP_TIMEWAIT) { 744 /* 745 * The only option of relevance is TOF_CC, and only if 746 * present in a SYN segment. See tcp_timewait(). 747 */ 748 if (thflags & TH_SYN) 749 tcp_dooptions(&to, optp, optlen, 1); 750 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 751 &to, th, m, tlen)) 752 goto findpcb; 753 /* 754 * tcp_timewait unlocks inp. 755 */ 756 INP_INFO_WUNLOCK(&tcbinfo); 757 return; 758 } 759 tp = intotcpcb(inp); 760 if (tp == 0) { 761 INP_UNLOCK(inp); 762 rstreason = BANDLIM_RST_CLOSEDPORT; 763 goto dropwithreset; 764 } 765 if (tp->t_state == TCPS_CLOSED) 766 goto drop; 767 768 /* Unscale the window into a 32-bit value. */ 769 if ((thflags & TH_SYN) == 0) 770 tiwin = th->th_win << tp->snd_scale; 771 else 772 tiwin = th->th_win; 773 774#ifdef MAC 775 INP_LOCK_ASSERT(inp); 776 if (mac_check_inpcb_deliver(inp, m)) 777 goto drop; 778#endif 779 so = inp->inp_socket; 780#ifdef TCPDEBUG 781 if (so->so_options & SO_DEBUG) { 782 ostate = tp->t_state; 783 if (isipv6) 784 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 785 else 786 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 787 tcp_savetcp = *th; 788 } 789#endif 790 if (so->so_options & SO_ACCEPTCONN) { 791 struct in_conninfo inc; 792 793#ifdef INET6 794 inc.inc_isipv6 = isipv6; 795#endif 796 if (isipv6) { 797 inc.inc6_faddr = ip6->ip6_src; 798 inc.inc6_laddr = ip6->ip6_dst; 799 } else { 800 inc.inc_faddr = ip->ip_src; 801 inc.inc_laddr = ip->ip_dst; 802 } 803 inc.inc_fport = th->th_sport; 804 inc.inc_lport = th->th_dport; 805 806 /* 807 * If the state is LISTEN then ignore segment if it contains 808 * a RST. If the segment contains an ACK then it is bad and 809 * send a RST. If it does not contain a SYN then it is not 810 * interesting; drop it. 811 * 812 * If the state is SYN_RECEIVED (syncache) and seg contains 813 * an ACK, but not for our SYN/ACK, send a RST. If the seg 814 * contains a RST, check the sequence number to see if it 815 * is a valid reset segment. 816 */ 817 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 818 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 819 if (!syncache_expand(&inc, th, &so, m)) { 820 /* 821 * No syncache entry, or ACK was not 822 * for our SYN/ACK. Send a RST. 823 */ 824 tcpstat.tcps_badsyn++; 825 rstreason = BANDLIM_RST_OPENPORT; 826 goto dropwithreset; 827 } 828 if (so == NULL) { 829 /* 830 * Could not complete 3-way handshake, 831 * connection is being closed down, and 832 * syncache will free mbuf. 833 */ 834 INP_UNLOCK(inp); 835 INP_INFO_WUNLOCK(&tcbinfo); 836 return; 837 } 838 /* 839 * Socket is created in state SYN_RECEIVED. 840 * Continue processing segment. 841 */ 842 INP_UNLOCK(inp); 843 inp = sotoinpcb(so); 844 INP_LOCK(inp); 845 tp = intotcpcb(inp); 846 /* 847 * This is what would have happened in 848 * tcp_output() when the SYN,ACK was sent. 849 */ 850 tp->snd_up = tp->snd_una; 851 tp->snd_max = tp->snd_nxt = tp->iss + 1; 852 tp->last_ack_sent = tp->rcv_nxt; 853 /* 854 * RFC1323: The window in SYN & SYN/ACK 855 * segments is never scaled. 856 */ 857 tp->snd_wnd = tiwin; /* unscaled */ 858 goto after_listen; 859 } 860 if (thflags & TH_RST) { 861 syncache_chkrst(&inc, th); 862 goto drop; 863 } 864 if (thflags & TH_ACK) { 865 syncache_badack(&inc); 866 tcpstat.tcps_badsyn++; 867 rstreason = BANDLIM_RST_OPENPORT; 868 goto dropwithreset; 869 } 870 goto drop; 871 } 872 873 /* 874 * Segment's flags are (SYN) or (SYN|FIN). 875 */ 876#ifdef INET6 877 /* 878 * If deprecated address is forbidden, 879 * we do not accept SYN to deprecated interface 880 * address to prevent any new inbound connection from 881 * getting established. 882 * When we do not accept SYN, we send a TCP RST, 883 * with deprecated source address (instead of dropping 884 * it). We compromise it as it is much better for peer 885 * to send a RST, and RST will be the final packet 886 * for the exchange. 887 * 888 * If we do not forbid deprecated addresses, we accept 889 * the SYN packet. RFC2462 does not suggest dropping 890 * SYN in this case. 891 * If we decipher RFC2462 5.5.4, it says like this: 892 * 1. use of deprecated addr with existing 893 * communication is okay - "SHOULD continue to be 894 * used" 895 * 2. use of it with new communication: 896 * (2a) "SHOULD NOT be used if alternate address 897 * with sufficient scope is available" 898 * (2b) nothing mentioned otherwise. 899 * Here we fall into (2b) case as we have no choice in 900 * our source address selection - we must obey the peer. 901 * 902 * The wording in RFC2462 is confusing, and there are 903 * multiple description text for deprecated address 904 * handling - worse, they are not exactly the same. 905 * I believe 5.5.4 is the best one, so we follow 5.5.4. 906 */ 907 if (isipv6 && !ip6_use_deprecated) { 908 struct in6_ifaddr *ia6; 909 910 if ((ia6 = ip6_getdstifaddr(m)) && 911 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 912 INP_UNLOCK(inp); 913 tp = NULL; 914 rstreason = BANDLIM_RST_OPENPORT; 915 goto dropwithreset; 916 } 917 } 918#endif 919 /* 920 * If it is from this socket, drop it, it must be forged. 921 * Don't bother responding if the destination was a broadcast. 922 */ 923 if (th->th_dport == th->th_sport) { 924 if (isipv6) { 925 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 926 &ip6->ip6_src)) 927 goto drop; 928 } else { 929 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 930 goto drop; 931 } 932 } 933 /* 934 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 935 * 936 * Note that it is quite possible to receive unicast 937 * link-layer packets with a broadcast IP address. Use 938 * in_broadcast() to find them. 939 */ 940 if (m->m_flags & (M_BCAST|M_MCAST)) 941 goto drop; 942 if (isipv6) { 943 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 944 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 945 goto drop; 946 } else { 947 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 948 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 949 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 950 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 951 goto drop; 952 } 953 /* 954 * SYN appears to be valid; create compressed TCP state 955 * for syncache, or perform t/tcp connection. 956 */ 957 if (so->so_qlen <= so->so_qlimit) { 958#ifdef TCPDEBUG 959 if (so->so_options & SO_DEBUG) 960 tcp_trace(TA_INPUT, ostate, tp, 961 (void *)tcp_saveipgen, &tcp_savetcp, 0); 962#endif 963 tcp_dooptions(&to, optp, optlen, 1); 964 if (!syncache_add(&inc, &to, th, &so, m)) 965 goto drop; 966 if (so == NULL) { 967 /* 968 * Entry added to syncache, mbuf used to 969 * send SYN,ACK packet. 970 */ 971 KASSERT(headlocked, ("headlocked")); 972 INP_UNLOCK(inp); 973 INP_INFO_WUNLOCK(&tcbinfo); 974 return; 975 } 976 /* 977 * Segment passed TAO tests. 978 */ 979 INP_UNLOCK(inp); 980 inp = sotoinpcb(so); 981 INP_LOCK(inp); 982 tp = intotcpcb(inp); 983 tp->snd_wnd = tiwin; 984 tp->t_starttime = ticks; 985 tp->t_state = TCPS_ESTABLISHED; 986 987 /* 988 * T/TCP logic: 989 * If there is a FIN or if there is data, then 990 * delay SYN,ACK(SYN) in the hope of piggy-backing 991 * it on a response segment. Otherwise must send 992 * ACK now in case the other side is slow starting. 993 */ 994 if (thflags & TH_FIN || tlen != 0) 995 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 996 else 997 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 998 tcpstat.tcps_connects++; 999 soisconnected(so); 1000 goto trimthenstep6; 1001 } 1002 goto drop; 1003 } 1004after_listen: 1005 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1006 INP_LOCK_ASSERT(inp); 1007 1008 /* XXX temp debugging */ 1009 /* should not happen - syncache should pick up these connections */ 1010 if (tp->t_state == TCPS_LISTEN) 1011 panic("tcp_input: TCPS_LISTEN"); 1012 1013 /* 1014 * This is the second part of the MSS DoS prevention code (after 1015 * minmss on the sending side) and it deals with too many too small 1016 * tcp packets in a too short timeframe (1 second). 1017 * 1018 * For every full second we count the number of received packets 1019 * and bytes. If we get a lot of packets per second for this connection 1020 * (tcp_minmssoverload) we take a closer look at it and compute the 1021 * average packet size for the past second. If that is less than 1022 * tcp_minmss we get too many packets with very small payload which 1023 * is not good and burdens our system (and every packet generates 1024 * a wakeup to the process connected to our socket). We can reasonable 1025 * expect this to be small packet DoS attack to exhaust our CPU 1026 * cycles. 1027 * 1028 * Care has to be taken for the minimum packet overload value. This 1029 * value defines the minimum number of packets per second before we 1030 * start to worry. This must not be too low to avoid killing for 1031 * example interactive connections with many small packets like 1032 * telnet or SSH. 1033 * 1034 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1035 * this check. 1036 * 1037 * Account for packet if payload packet, skip over ACK, etc. 1038 */ 1039 if (tcp_minmss && tcp_minmssoverload && 1040 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1041 if ((unsigned int)(tp->rcv_second - ticks) < hz) { 1042 tp->rcv_pps++; 1043 tp->rcv_byps += tlen + off; 1044 if (tp->rcv_pps > tcp_minmssoverload) { 1045 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1046 printf("too many small tcp packets from " 1047 "%s:%u, av. %lubyte/packet, " 1048 "dropping connection\n", 1049#ifdef INET6 1050 isipv6 ? 1051 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1052#endif 1053 inet_ntoa(inp->inp_inc.inc_faddr), 1054 inp->inp_inc.inc_fport, 1055 tp->rcv_byps / tp->rcv_pps); 1056 KASSERT(headlocked, ("tcp_input: " 1057 "after_listen: tcp_drop: head " 1058 "not locked")); 1059 tp = tcp_drop(tp, ECONNRESET); 1060 tcpstat.tcps_minmssdrops++; 1061 goto drop; 1062 } 1063 } 1064 } else { 1065 tp->rcv_second = ticks + hz; 1066 tp->rcv_pps = 1; 1067 tp->rcv_byps = tlen + off; 1068 } 1069 } 1070 1071 /* 1072 * Segment received on connection. 1073 * Reset idle time and keep-alive timer. 1074 */ 1075 tp->t_rcvtime = ticks; 1076 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1077 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1078 1079 /* 1080 * Process options only when we get SYN/ACK back. The SYN case 1081 * for incoming connections is handled in tcp_syncache. 1082 * XXX this is traditional behavior, may need to be cleaned up. 1083 */ 1084 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN);
|
1086 if (to.to_flags & TOF_SCALE) { 1087 tp->t_flags |= TF_RCVD_SCALE; 1088 tp->requested_s_scale = to.to_requested_s_scale; 1089 } 1090 if (to.to_flags & TOF_TS) { 1091 tp->t_flags |= TF_RCVD_TSTMP; 1092 tp->ts_recent = to.to_tsval; 1093 tp->ts_recent_age = ticks; 1094 } 1095 if (to.to_flags & TOF_MSS) 1096 tcp_mss(tp, to.to_mss); 1097 if (tp->sack_enable) { 1098 if (!(to.to_flags & TOF_SACK)) 1099 tp->sack_enable = 0; 1100 else 1101 tp->t_flags |= TF_SACK_PERMIT; 1102 } 1103 1104 } 1105 1106 /* 1107 * Header prediction: check for the two common cases 1108 * of a uni-directional data xfer. If the packet has 1109 * no control flags, is in-sequence, the window didn't 1110 * change and we're not retransmitting, it's a 1111 * candidate. If the length is zero and the ack moved 1112 * forward, we're the sender side of the xfer. Just 1113 * free the data acked & wake any higher level process 1114 * that was blocked waiting for space. If the length 1115 * is non-zero and the ack didn't move, we're the 1116 * receiver side. If we're getting packets in-order 1117 * (the reassembly queue is empty), add the data to 1118 * the socket buffer and note that we need a delayed ack. 1119 * Make sure that the hidden state-flags are also off. 1120 * Since we check for TCPS_ESTABLISHED above, it can only 1121 * be TH_NEEDSYN. 1122 */ 1123 if (tp->t_state == TCPS_ESTABLISHED && 1124 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1125 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1126 ((to.to_flags & TOF_TS) == 0 || 1127 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1128 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1129 tp->snd_nxt == tp->snd_max) { 1130 1131 /* 1132 * If last ACK falls within this segment's sequence numbers, 1133 * record the timestamp. 1134 * NOTE that the test is modified according to the latest 1135 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1136 */ 1137 if ((to.to_flags & TOF_TS) != 0 && 1138 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1139 tp->ts_recent_age = ticks; 1140 tp->ts_recent = to.to_tsval; 1141 } 1142 1143 if (tlen == 0) { 1144 if (SEQ_GT(th->th_ack, tp->snd_una) && 1145 SEQ_LEQ(th->th_ack, tp->snd_max) && 1146 tp->snd_cwnd >= tp->snd_wnd && 1147 ((!tcp_do_newreno && !tp->sack_enable && 1148 tp->t_dupacks < tcprexmtthresh) || 1149 ((tcp_do_newreno || tp->sack_enable) && 1150 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0))) { 1151 KASSERT(headlocked, ("headlocked")); 1152 INP_INFO_WUNLOCK(&tcbinfo); 1153 headlocked = 0; 1154 /* 1155 * this is a pure ack for outstanding data. 1156 */ 1157 ++tcpstat.tcps_predack; 1158 /* 1159 * "bad retransmit" recovery 1160 */ 1161 if (tp->t_rxtshift == 1 && 1162 ticks < tp->t_badrxtwin) { 1163 ++tcpstat.tcps_sndrexmitbad; 1164 tp->snd_cwnd = tp->snd_cwnd_prev; 1165 tp->snd_ssthresh = 1166 tp->snd_ssthresh_prev; 1167 tp->snd_recover = tp->snd_recover_prev; 1168 if (tp->t_flags & TF_WASFRECOVERY) 1169 ENTER_FASTRECOVERY(tp); 1170 tp->snd_nxt = tp->snd_max; 1171 tp->t_badrxtwin = 0; 1172 } 1173 1174 /* 1175 * Recalculate the transmit timer / rtt. 1176 * 1177 * Some boxes send broken timestamp replies 1178 * during the SYN+ACK phase, ignore 1179 * timestamps of 0 or we could calculate a 1180 * huge RTT and blow up the retransmit timer. 1181 */ 1182 if ((to.to_flags & TOF_TS) != 0 && 1183 to.to_tsecr) { 1184 tcp_xmit_timer(tp, 1185 ticks - to.to_tsecr + 1); 1186 } else if (tp->t_rtttime && 1187 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1188 tcp_xmit_timer(tp, 1189 ticks - tp->t_rtttime); 1190 } 1191 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1192 acked = th->th_ack - tp->snd_una; 1193 tcpstat.tcps_rcvackpack++; 1194 tcpstat.tcps_rcvackbyte += acked; 1195 sbdrop(&so->so_snd, acked); 1196 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1197 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1198 tp->snd_recover = th->th_ack - 1; 1199 tp->snd_una = th->th_ack; 1200 /* 1201 * pull snd_wl2 up to prevent seq wrap relative 1202 * to th_ack. 1203 */ 1204 tp->snd_wl2 = th->th_ack; 1205 tp->t_dupacks = 0; 1206 m_freem(m); 1207 ND6_HINT(tp); /* some progress has been done */ 1208 1209 /* 1210 * If all outstanding data are acked, stop 1211 * retransmit timer, otherwise restart timer 1212 * using current (possibly backed-off) value. 1213 * If process is waiting for space, 1214 * wakeup/selwakeup/signal. If data 1215 * are ready to send, let tcp_output 1216 * decide between more output or persist. 1217 1218#ifdef TCPDEBUG 1219 if (so->so_options & SO_DEBUG) 1220 tcp_trace(TA_INPUT, ostate, tp, 1221 (void *)tcp_saveipgen, 1222 &tcp_savetcp, 0); 1223#endif 1224 */ 1225 if (tp->snd_una == tp->snd_max) 1226 callout_stop(tp->tt_rexmt); 1227 else if (!callout_active(tp->tt_persist)) 1228 callout_reset(tp->tt_rexmt, 1229 tp->t_rxtcur, 1230 tcp_timer_rexmt, tp); 1231 1232 sowwakeup(so); 1233 if (so->so_snd.sb_cc) 1234 (void) tcp_output(tp); 1235 goto check_delack; 1236 } 1237 } else if (th->th_ack == tp->snd_una && 1238 LIST_EMPTY(&tp->t_segq) && 1239 tlen <= sbspace(&so->so_rcv)) { 1240 KASSERT(headlocked, ("headlocked")); 1241 INP_INFO_WUNLOCK(&tcbinfo); 1242 headlocked = 0; 1243 /* 1244 * this is a pure, in-sequence data packet 1245 * with nothing on the reassembly queue and 1246 * we have enough buffer space to take it. 1247 */ 1248 /* Clean receiver SACK report if present */ 1249 if (tp->sack_enable && tp->rcv_numsacks) 1250 tcp_clean_sackreport(tp); 1251 ++tcpstat.tcps_preddat; 1252 tp->rcv_nxt += tlen; 1253 /* 1254 * Pull snd_wl1 up to prevent seq wrap relative to 1255 * th_seq. 1256 */ 1257 tp->snd_wl1 = th->th_seq; 1258 /* 1259 * Pull rcv_up up to prevent seq wrap relative to 1260 * rcv_nxt. 1261 */ 1262 tp->rcv_up = tp->rcv_nxt; 1263 tcpstat.tcps_rcvpack++; 1264 tcpstat.tcps_rcvbyte += tlen; 1265 ND6_HINT(tp); /* some progress has been done */ 1266 /* 1267#ifdef TCPDEBUG 1268 if (so->so_options & SO_DEBUG) 1269 tcp_trace(TA_INPUT, ostate, tp, 1270 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1271#endif 1272 * Add data to socket buffer. 1273 */ 1274 SOCKBUF_LOCK(&so->so_rcv); 1275 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1276 m_freem(m); 1277 } else { 1278 m_adj(m, drop_hdrlen); /* delayed header drop */ 1279 sbappendstream_locked(&so->so_rcv, m); 1280 } 1281 sorwakeup_locked(so); 1282 if (DELAY_ACK(tp)) { 1283 tp->t_flags |= TF_DELACK; 1284 } else { 1285 tp->t_flags |= TF_ACKNOW; 1286 tcp_output(tp); 1287 } 1288 goto check_delack; 1289 } 1290 } 1291 1292 /* 1293 * Calculate amount of space in receive window, 1294 * and then do TCP input processing. 1295 * Receive window is amount of space in rcv queue, 1296 * but not less than advertised window. 1297 */ 1298 { int win; 1299 1300 win = sbspace(&so->so_rcv); 1301 if (win < 0) 1302 win = 0; 1303 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1304 } 1305 1306 switch (tp->t_state) { 1307 1308 /* 1309 * If the state is SYN_RECEIVED: 1310 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1311 */ 1312 case TCPS_SYN_RECEIVED: 1313 if ((thflags & TH_ACK) && 1314 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1315 SEQ_GT(th->th_ack, tp->snd_max))) { 1316 rstreason = BANDLIM_RST_OPENPORT; 1317 goto dropwithreset; 1318 } 1319 break; 1320 1321 /* 1322 * If the state is SYN_SENT: 1323 * if seg contains an ACK, but not for our SYN, drop the input. 1324 * if seg contains a RST, then drop the connection. 1325 * if seg does not contain SYN, then drop it. 1326 * Otherwise this is an acceptable SYN segment 1327 * initialize tp->rcv_nxt and tp->irs 1328 * if seg contains ack then advance tp->snd_una 1329 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1330 * arrange for segment to be acked (eventually) 1331 * continue processing rest of data/controls, beginning with URG 1332 */ 1333 case TCPS_SYN_SENT: 1334 if ((thflags & TH_ACK) && 1335 (SEQ_LEQ(th->th_ack, tp->iss) || 1336 SEQ_GT(th->th_ack, tp->snd_max))) { 1337 rstreason = BANDLIM_UNLIMITED; 1338 goto dropwithreset; 1339 } 1340 if (thflags & TH_RST) { 1341 if (thflags & TH_ACK) { 1342 KASSERT(headlocked, ("tcp_input: after_listen" 1343 ": tcp_drop.2: head not locked")); 1344 tp = tcp_drop(tp, ECONNREFUSED); 1345 } 1346 goto drop; 1347 } 1348 if ((thflags & TH_SYN) == 0) 1349 goto drop; 1350 tp->snd_wnd = th->th_win; /* initial send window */ 1351 1352 tp->irs = th->th_seq; 1353 tcp_rcvseqinit(tp); 1354 if (thflags & TH_ACK) { 1355 tcpstat.tcps_connects++; 1356 soisconnected(so); 1357#ifdef MAC 1358 SOCK_LOCK(so); 1359 mac_set_socket_peer_from_mbuf(m, so); 1360 SOCK_UNLOCK(so); 1361#endif 1362 /* Do window scaling on this connection? */ 1363 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1364 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1365 tp->snd_scale = tp->requested_s_scale; 1366 tp->rcv_scale = tp->request_r_scale; 1367 } 1368 tp->rcv_adv += tp->rcv_wnd; 1369 tp->snd_una++; /* SYN is acked */ 1370 /* 1371 * If there's data, delay ACK; if there's also a FIN 1372 * ACKNOW will be turned on later. 1373 */ 1374 if (DELAY_ACK(tp) && tlen != 0) 1375 callout_reset(tp->tt_delack, tcp_delacktime, 1376 tcp_timer_delack, tp); 1377 else 1378 tp->t_flags |= TF_ACKNOW; 1379 /* 1380 * Received <SYN,ACK> in SYN_SENT[*] state. 1381 * Transitions: 1382 * SYN_SENT --> ESTABLISHED 1383 * SYN_SENT* --> FIN_WAIT_1 1384 */ 1385 tp->t_starttime = ticks; 1386 if (tp->t_flags & TF_NEEDFIN) { 1387 tp->t_state = TCPS_FIN_WAIT_1; 1388 tp->t_flags &= ~TF_NEEDFIN; 1389 thflags &= ~TH_SYN; 1390 } else { 1391 tp->t_state = TCPS_ESTABLISHED; 1392 callout_reset(tp->tt_keep, tcp_keepidle, 1393 tcp_timer_keep, tp); 1394 } 1395 } else { 1396 /* 1397 * Received initial SYN in SYN-SENT[*] state => 1398 * simultaneous open. If segment contains CC option 1399 * and there is a cached CC, apply TAO test. 1400 * If it succeeds, connection is * half-synchronized. 1401 * Otherwise, do 3-way handshake: 1402 * SYN-SENT -> SYN-RECEIVED 1403 * SYN-SENT* -> SYN-RECEIVED* 1404 * If there was no CC option, clear cached CC value. 1405 */ 1406 tp->t_flags |= TF_ACKNOW; 1407 callout_stop(tp->tt_rexmt); 1408 tp->t_state = TCPS_SYN_RECEIVED; 1409 } 1410 1411trimthenstep6: 1412 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1413 "locked")); 1414 INP_LOCK_ASSERT(inp); 1415 1416 /* 1417 * Advance th->th_seq to correspond to first data byte. 1418 * If data, trim to stay within window, 1419 * dropping FIN if necessary. 1420 */ 1421 th->th_seq++; 1422 if (tlen > tp->rcv_wnd) { 1423 todrop = tlen - tp->rcv_wnd; 1424 m_adj(m, -todrop); 1425 tlen = tp->rcv_wnd; 1426 thflags &= ~TH_FIN; 1427 tcpstat.tcps_rcvpackafterwin++; 1428 tcpstat.tcps_rcvbyteafterwin += todrop; 1429 } 1430 tp->snd_wl1 = th->th_seq - 1; 1431 tp->rcv_up = th->th_seq; 1432 /* 1433 * Client side of transaction: already sent SYN and data. 1434 * If the remote host used T/TCP to validate the SYN, 1435 * our data will be ACK'd; if so, enter normal data segment 1436 * processing in the middle of step 5, ack processing. 1437 * Otherwise, goto step 6. 1438 */ 1439 if (thflags & TH_ACK) 1440 goto process_ACK; 1441 1442 goto step6; 1443 1444 /* 1445 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1446 * do normal processing. 1447 * 1448 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1449 */ 1450 case TCPS_LAST_ACK: 1451 case TCPS_CLOSING: 1452 case TCPS_TIME_WAIT: 1453 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1454 break; /* continue normal processing */ 1455 } 1456 1457 /* 1458 * States other than LISTEN or SYN_SENT. 1459 * First check the RST flag and sequence number since reset segments 1460 * are exempt from the timestamp and connection count tests. This 1461 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1462 * below which allowed reset segments in half the sequence space 1463 * to fall though and be processed (which gives forged reset 1464 * segments with a random sequence number a 50 percent chance of 1465 * killing a connection). 1466 * Then check timestamp, if present. 1467 * Then check the connection count, if present. 1468 * Then check that at least some bytes of segment are within 1469 * receive window. If segment begins before rcv_nxt, 1470 * drop leading data (and SYN); if nothing left, just ack. 1471 * 1472 * 1473 * If the RST bit is set, check the sequence number to see 1474 * if this is a valid reset segment. 1475 * RFC 793 page 37: 1476 * In all states except SYN-SENT, all reset (RST) segments 1477 * are validated by checking their SEQ-fields. A reset is 1478 * valid if its sequence number is in the window. 1479 * Note: this does not take into account delayed ACKs, so 1480 * we should test against last_ack_sent instead of rcv_nxt. 1481 * The sequence number in the reset segment is normally an 1482 * echo of our outgoing acknowlegement numbers, but some hosts 1483 * send a reset with the sequence number at the rightmost edge 1484 * of our receive window, and we have to handle this case. 1485 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1486 * that brute force RST attacks are possible. To combat this, 1487 * we use a much stricter check while in the ESTABLISHED state, 1488 * only accepting RSTs where the sequence number is equal to 1489 * last_ack_sent. In all other states (the states in which a 1490 * RST is more likely), the more permissive check is used. 1491 * If we have multiple segments in flight, the intial reset 1492 * segment sequence numbers will be to the left of last_ack_sent, 1493 * but they will eventually catch up. 1494 * In any case, it never made sense to trim reset segments to 1495 * fit the receive window since RFC 1122 says: 1496 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1497 * 1498 * A TCP SHOULD allow a received RST segment to include data. 1499 * 1500 * DISCUSSION 1501 * It has been suggested that a RST segment could contain 1502 * ASCII text that encoded and explained the cause of the 1503 * RST. No standard has yet been established for such 1504 * data. 1505 * 1506 * If the reset segment passes the sequence number test examine 1507 * the state: 1508 * SYN_RECEIVED STATE: 1509 * If passive open, return to LISTEN state. 1510 * If active open, inform user that connection was refused. 1511 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1512 * Inform user that connection was reset, and close tcb. 1513 * CLOSING, LAST_ACK STATES: 1514 * Close the tcb. 1515 * TIME_WAIT STATE: 1516 * Drop the segment - see Stevens, vol. 2, p. 964 and 1517 * RFC 1337. 1518 */ 1519 if (thflags & TH_RST) { 1520 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1521 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1522 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1523 switch (tp->t_state) { 1524 1525 case TCPS_SYN_RECEIVED: 1526 so->so_error = ECONNREFUSED; 1527 goto close; 1528 1529 case TCPS_ESTABLISHED: 1530 if (tp->last_ack_sent != th->th_seq && 1531 tcp_insecure_rst == 0) { 1532 tcpstat.tcps_badrst++; 1533 goto drop; 1534 } 1535 case TCPS_FIN_WAIT_1: 1536 case TCPS_FIN_WAIT_2: 1537 case TCPS_CLOSE_WAIT: 1538 so->so_error = ECONNRESET; 1539 close: 1540 tp->t_state = TCPS_CLOSED; 1541 tcpstat.tcps_drops++; 1542 KASSERT(headlocked, ("tcp_input: " 1543 "trimthenstep6: tcp_close: head not " 1544 "locked")); 1545 tp = tcp_close(tp); 1546 break; 1547 1548 case TCPS_CLOSING: 1549 case TCPS_LAST_ACK: 1550 KASSERT(headlocked, ("trimthenstep6: " 1551 "tcp_close.2: head not locked")); 1552 tp = tcp_close(tp); 1553 break; 1554 1555 case TCPS_TIME_WAIT: 1556 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1557 ("timewait")); 1558 break; 1559 } 1560 } 1561 goto drop; 1562 } 1563 1564 /* 1565 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1566 * and it's less than ts_recent, drop it. 1567 */ 1568 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1569 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1570 1571 /* Check to see if ts_recent is over 24 days old. */ 1572 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1573 /* 1574 * Invalidate ts_recent. If this segment updates 1575 * ts_recent, the age will be reset later and ts_recent 1576 * will get a valid value. If it does not, setting 1577 * ts_recent to zero will at least satisfy the 1578 * requirement that zero be placed in the timestamp 1579 * echo reply when ts_recent isn't valid. The 1580 * age isn't reset until we get a valid ts_recent 1581 * because we don't want out-of-order segments to be 1582 * dropped when ts_recent is old. 1583 */ 1584 tp->ts_recent = 0; 1585 } else { 1586 tcpstat.tcps_rcvduppack++; 1587 tcpstat.tcps_rcvdupbyte += tlen; 1588 tcpstat.tcps_pawsdrop++; 1589 if (tlen) 1590 goto dropafterack; 1591 goto drop; 1592 } 1593 } 1594 1595 /* 1596 * In the SYN-RECEIVED state, validate that the packet belongs to 1597 * this connection before trimming the data to fit the receive 1598 * window. Check the sequence number versus IRS since we know 1599 * the sequence numbers haven't wrapped. This is a partial fix 1600 * for the "LAND" DoS attack. 1601 */ 1602 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1603 rstreason = BANDLIM_RST_OPENPORT; 1604 goto dropwithreset; 1605 } 1606 1607 todrop = tp->rcv_nxt - th->th_seq; 1608 if (todrop > 0) { 1609 if (thflags & TH_SYN) { 1610 thflags &= ~TH_SYN; 1611 th->th_seq++; 1612 if (th->th_urp > 1) 1613 th->th_urp--; 1614 else 1615 thflags &= ~TH_URG; 1616 todrop--; 1617 } 1618 /* 1619 * Following if statement from Stevens, vol. 2, p. 960. 1620 */ 1621 if (todrop > tlen 1622 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1623 /* 1624 * Any valid FIN must be to the left of the window. 1625 * At this point the FIN must be a duplicate or out 1626 * of sequence; drop it. 1627 */ 1628 thflags &= ~TH_FIN; 1629 1630 /* 1631 * Send an ACK to resynchronize and drop any data. 1632 * But keep on processing for RST or ACK. 1633 */ 1634 tp->t_flags |= TF_ACKNOW; 1635 todrop = tlen; 1636 tcpstat.tcps_rcvduppack++; 1637 tcpstat.tcps_rcvdupbyte += todrop; 1638 } else { 1639 tcpstat.tcps_rcvpartduppack++; 1640 tcpstat.tcps_rcvpartdupbyte += todrop; 1641 } 1642 drop_hdrlen += todrop; /* drop from the top afterwards */ 1643 th->th_seq += todrop; 1644 tlen -= todrop; 1645 if (th->th_urp > todrop) 1646 th->th_urp -= todrop; 1647 else { 1648 thflags &= ~TH_URG; 1649 th->th_urp = 0; 1650 } 1651 } 1652 1653 /* 1654 * If new data are received on a connection after the 1655 * user processes are gone, then RST the other end. 1656 */ 1657 if ((so->so_state & SS_NOFDREF) && 1658 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1659 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not " 1660 "locked")); 1661 tp = tcp_close(tp); 1662 tcpstat.tcps_rcvafterclose++; 1663 rstreason = BANDLIM_UNLIMITED; 1664 goto dropwithreset; 1665 } 1666 1667 /* 1668 * If segment ends after window, drop trailing data 1669 * (and PUSH and FIN); if nothing left, just ACK. 1670 */ 1671 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1672 if (todrop > 0) { 1673 tcpstat.tcps_rcvpackafterwin++; 1674 if (todrop >= tlen) { 1675 tcpstat.tcps_rcvbyteafterwin += tlen; 1676 /* 1677 * If a new connection request is received 1678 * while in TIME_WAIT, drop the old connection 1679 * and start over if the sequence numbers 1680 * are above the previous ones. 1681 */ 1682 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1683 if (thflags & TH_SYN && 1684 tp->t_state == TCPS_TIME_WAIT && 1685 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1686 KASSERT(headlocked, ("trimthenstep6: " 1687 "tcp_close.4: head not locked")); 1688 tp = tcp_close(tp); 1689 goto findpcb; 1690 } 1691 /* 1692 * If window is closed can only take segments at 1693 * window edge, and have to drop data and PUSH from 1694 * incoming segments. Continue processing, but 1695 * remember to ack. Otherwise, drop segment 1696 * and ack. 1697 */ 1698 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1699 tp->t_flags |= TF_ACKNOW; 1700 tcpstat.tcps_rcvwinprobe++; 1701 } else 1702 goto dropafterack; 1703 } else 1704 tcpstat.tcps_rcvbyteafterwin += todrop; 1705 m_adj(m, -todrop); 1706 tlen -= todrop; 1707 thflags &= ~(TH_PUSH|TH_FIN); 1708 } 1709 1710 /* 1711 * If last ACK falls within this segment's sequence numbers, 1712 * record its timestamp. 1713 * NOTE: 1714 * 1) That the test incorporates suggestions from the latest 1715 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1716 * 2) That updating only on newer timestamps interferes with 1717 * our earlier PAWS tests, so this check should be solely 1718 * predicated on the sequence space of this segment. 1719 * 3) That we modify the segment boundary check to be 1720 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1721 * instead of RFC1323's 1722 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1723 * This modified check allows us to overcome RFC1323's 1724 * limitations as described in Stevens TCP/IP Illustrated 1725 * Vol. 2 p.869. In such cases, we can still calculate the 1726 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1727 */ 1728 if ((to.to_flags & TOF_TS) != 0 && 1729 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1730 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1731 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1732 tp->ts_recent_age = ticks; 1733 tp->ts_recent = to.to_tsval; 1734 } 1735 1736 /* 1737 * If a SYN is in the window, then this is an 1738 * error and we send an RST and drop the connection. 1739 */ 1740 if (thflags & TH_SYN) { 1741 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: " 1742 "head not locked")); 1743 tp = tcp_drop(tp, ECONNRESET); 1744 rstreason = BANDLIM_UNLIMITED; 1745 goto drop; 1746 } 1747 1748 /* 1749 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1750 * flag is on (half-synchronized state), then queue data for 1751 * later processing; else drop segment and return. 1752 */ 1753 if ((thflags & TH_ACK) == 0) { 1754 if (tp->t_state == TCPS_SYN_RECEIVED || 1755 (tp->t_flags & TF_NEEDSYN)) 1756 goto step6; 1757 else 1758 goto drop; 1759 } 1760 1761 /* 1762 * Ack processing. 1763 */ 1764 switch (tp->t_state) { 1765 1766 /* 1767 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1768 * ESTABLISHED state and continue processing. 1769 * The ACK was checked above. 1770 */ 1771 case TCPS_SYN_RECEIVED: 1772 1773 tcpstat.tcps_connects++; 1774 soisconnected(so); 1775 /* Do window scaling? */ 1776 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1777 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1778 tp->snd_scale = tp->requested_s_scale; 1779 tp->rcv_scale = tp->request_r_scale; 1780 } 1781 /* 1782 * Make transitions: 1783 * SYN-RECEIVED -> ESTABLISHED 1784 * SYN-RECEIVED* -> FIN-WAIT-1 1785 */ 1786 tp->t_starttime = ticks; 1787 if (tp->t_flags & TF_NEEDFIN) { 1788 tp->t_state = TCPS_FIN_WAIT_1; 1789 tp->t_flags &= ~TF_NEEDFIN; 1790 } else { 1791 tp->t_state = TCPS_ESTABLISHED; 1792 callout_reset(tp->tt_keep, tcp_keepidle, 1793 tcp_timer_keep, tp); 1794 } 1795 /* 1796 * If segment contains data or ACK, will call tcp_reass() 1797 * later; if not, do so now to pass queued data to user. 1798 */ 1799 if (tlen == 0 && (thflags & TH_FIN) == 0) 1800 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1801 (struct mbuf *)0); 1802 tp->snd_wl1 = th->th_seq - 1; 1803 /* FALLTHROUGH */ 1804 1805 /* 1806 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1807 * ACKs. If the ack is in the range 1808 * tp->snd_una < th->th_ack <= tp->snd_max 1809 * then advance tp->snd_una to th->th_ack and drop 1810 * data from the retransmission queue. If this ACK reflects 1811 * more up to date window information we update our window information. 1812 */ 1813 case TCPS_ESTABLISHED: 1814 case TCPS_FIN_WAIT_1: 1815 case TCPS_FIN_WAIT_2: 1816 case TCPS_CLOSE_WAIT: 1817 case TCPS_CLOSING: 1818 case TCPS_LAST_ACK: 1819 case TCPS_TIME_WAIT: 1820 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1821 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1822 tcpstat.tcps_rcvacktoomuch++; 1823 goto dropafterack; 1824 } 1825 if (tp->sack_enable) 1826 tcp_sack_doack(tp, &to, th->th_ack); 1827 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1828 if (tlen == 0 && tiwin == tp->snd_wnd) { 1829 tcpstat.tcps_rcvdupack++; 1830 /* 1831 * If we have outstanding data (other than 1832 * a window probe), this is a completely 1833 * duplicate ack (ie, window info didn't 1834 * change), the ack is the biggest we've 1835 * seen and we've seen exactly our rexmt 1836 * threshhold of them, assume a packet 1837 * has been dropped and retransmit it. 1838 * Kludge snd_nxt & the congestion 1839 * window so we send only this one 1840 * packet. 1841 * 1842 * We know we're losing at the current 1843 * window size so do congestion avoidance 1844 * (set ssthresh to half the current window 1845 * and pull our congestion window back to 1846 * the new ssthresh). 1847 * 1848 * Dup acks mean that packets have left the 1849 * network (they're now cached at the receiver) 1850 * so bump cwnd by the amount in the receiver 1851 * to keep a constant cwnd packets in the 1852 * network. 1853 */ 1854 if (!callout_active(tp->tt_rexmt) || 1855 th->th_ack != tp->snd_una) 1856 tp->t_dupacks = 0; 1857 else if (++tp->t_dupacks > tcprexmtthresh || 1858 ((tcp_do_newreno || tp->sack_enable) && 1859 IN_FASTRECOVERY(tp))) { 1860 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1861 int awnd; 1862 1863 /* 1864 * Compute the amount of data in flight first. 1865 * We can inject new data into the pipe iff 1866 * we have less than 1/2 the original window's 1867 * worth of data in flight. 1868 */ 1869 awnd = (tp->snd_nxt - tp->snd_fack) + 1870 tp->sackhint.sack_bytes_rexmit; 1871 if (awnd < tp->snd_ssthresh) { 1872 tp->snd_cwnd += tp->t_maxseg; 1873 if (tp->snd_cwnd > tp->snd_ssthresh) 1874 tp->snd_cwnd = tp->snd_ssthresh; 1875 } 1876 } else 1877 tp->snd_cwnd += tp->t_maxseg; 1878 (void) tcp_output(tp); 1879 goto drop; 1880 } else if (tp->t_dupacks == tcprexmtthresh) { 1881 tcp_seq onxt = tp->snd_nxt; 1882 u_int win; 1883 1884 /* 1885 * If we're doing sack, check to 1886 * see if we're already in sack 1887 * recovery. If we're not doing sack, 1888 * check to see if we're in newreno 1889 * recovery. 1890 */ 1891 if (tp->sack_enable) { 1892 if (IN_FASTRECOVERY(tp)) { 1893 tp->t_dupacks = 0; 1894 break; 1895 } 1896 } else if (tcp_do_newreno) { 1897 if (SEQ_LEQ(th->th_ack, 1898 tp->snd_recover)) { 1899 tp->t_dupacks = 0; 1900 break; 1901 } 1902 } 1903 win = min(tp->snd_wnd, tp->snd_cwnd) / 1904 2 / tp->t_maxseg; 1905 if (win < 2) 1906 win = 2; 1907 tp->snd_ssthresh = win * tp->t_maxseg; 1908 ENTER_FASTRECOVERY(tp); 1909 tp->snd_recover = tp->snd_max; 1910 callout_stop(tp->tt_rexmt); 1911 tp->t_rtttime = 0; 1912 if (tp->sack_enable) { 1913 KASSERT(tp->sackhint. 1914 sack_bytes_rexmit == 0, 1915 ("sackhint rexmit == 0")); 1916 tcpstat.tcps_sack_recovery_episode++; 1917 tp->sack_newdata = tp->snd_nxt; 1918 tp->snd_cwnd = tp->t_maxseg; 1919 (void) tcp_output(tp); 1920 goto drop; 1921 } 1922 tp->snd_nxt = th->th_ack; 1923 tp->snd_cwnd = tp->t_maxseg; 1924 (void) tcp_output(tp); 1925 KASSERT(tp->snd_limited <= 2, 1926 ("tp->snd_limited too big")); 1927 tp->snd_cwnd = tp->snd_ssthresh + 1928 tp->t_maxseg * 1929 (tp->t_dupacks - tp->snd_limited); 1930 if (SEQ_GT(onxt, tp->snd_nxt)) 1931 tp->snd_nxt = onxt; 1932 goto drop; 1933 } else if (tcp_do_rfc3042) { 1934 u_long oldcwnd = tp->snd_cwnd; 1935 tcp_seq oldsndmax = tp->snd_max; 1936 u_int sent; 1937 1938 KASSERT(tp->t_dupacks == 1 || 1939 tp->t_dupacks == 2, 1940 ("dupacks not 1 or 2")); 1941 if (tp->t_dupacks == 1) 1942 tp->snd_limited = 0; 1943 tp->snd_cwnd = 1944 (tp->snd_nxt - tp->snd_una) + 1945 (tp->t_dupacks - tp->snd_limited) * 1946 tp->t_maxseg; 1947 (void) tcp_output(tp); 1948 sent = tp->snd_max - oldsndmax; 1949 if (sent > tp->t_maxseg) { 1950 KASSERT((tp->t_dupacks == 2 && 1951 tp->snd_limited == 0) || 1952 (sent == tp->t_maxseg + 1 && 1953 tp->t_flags & TF_SENTFIN), 1954 ("sent too much")); 1955 tp->snd_limited = 2; 1956 } else if (sent > 0) 1957 ++tp->snd_limited; 1958 tp->snd_cwnd = oldcwnd; 1959 goto drop; 1960 } 1961 } else 1962 tp->t_dupacks = 0; 1963 break; 1964 } 1965 1966 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1967 1968 /* 1969 * If the congestion window was inflated to account 1970 * for the other side's cached packets, retract it. 1971 */ 1972 if (tcp_do_newreno || tp->sack_enable) { 1973 if (IN_FASTRECOVERY(tp)) { 1974 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1975 if (tp->sack_enable) 1976 tcp_sack_partialack(tp, th); 1977 else 1978 tcp_newreno_partial_ack(tp, th); 1979 } else { 1980 /* 1981 * Out of fast recovery. 1982 * Window inflation should have left us 1983 * with approximately snd_ssthresh 1984 * outstanding data. 1985 * But in case we would be inclined to 1986 * send a burst, better to do it via 1987 * the slow start mechanism. 1988 */ 1989 if (SEQ_GT(th->th_ack + 1990 tp->snd_ssthresh, 1991 tp->snd_max)) 1992 tp->snd_cwnd = tp->snd_max - 1993 th->th_ack + 1994 tp->t_maxseg; 1995 else 1996 tp->snd_cwnd = tp->snd_ssthresh; 1997 } 1998 } 1999 } else { 2000 if (tp->t_dupacks >= tcprexmtthresh && 2001 tp->snd_cwnd > tp->snd_ssthresh) 2002 tp->snd_cwnd = tp->snd_ssthresh; 2003 } 2004 tp->t_dupacks = 0; 2005 /* 2006 * If we reach this point, ACK is not a duplicate, 2007 * i.e., it ACKs something we sent. 2008 */ 2009 if (tp->t_flags & TF_NEEDSYN) { 2010 /* 2011 * T/TCP: Connection was half-synchronized, and our 2012 * SYN has been ACK'd (so connection is now fully 2013 * synchronized). Go to non-starred state, 2014 * increment snd_una for ACK of SYN, and check if 2015 * we can do window scaling. 2016 */ 2017 tp->t_flags &= ~TF_NEEDSYN; 2018 tp->snd_una++; 2019 /* Do window scaling? */ 2020 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2021 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2022 tp->snd_scale = tp->requested_s_scale; 2023 tp->rcv_scale = tp->request_r_scale; 2024 } 2025 } 2026 2027process_ACK: 2028 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 2029 "locked")); 2030 INP_LOCK_ASSERT(inp); 2031 2032 acked = th->th_ack - tp->snd_una; 2033 tcpstat.tcps_rcvackpack++; 2034 tcpstat.tcps_rcvackbyte += acked; 2035 2036 /* 2037 * If we just performed our first retransmit, and the ACK 2038 * arrives within our recovery window, then it was a mistake 2039 * to do the retransmit in the first place. Recover our 2040 * original cwnd and ssthresh, and proceed to transmit where 2041 * we left off. 2042 */ 2043 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2044 ++tcpstat.tcps_sndrexmitbad; 2045 tp->snd_cwnd = tp->snd_cwnd_prev; 2046 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2047 tp->snd_recover = tp->snd_recover_prev; 2048 if (tp->t_flags & TF_WASFRECOVERY) 2049 ENTER_FASTRECOVERY(tp); 2050 tp->snd_nxt = tp->snd_max; 2051 tp->t_badrxtwin = 0; /* XXX probably not required */ 2052 } 2053 2054 /* 2055 * If we have a timestamp reply, update smoothed 2056 * round trip time. If no timestamp is present but 2057 * transmit timer is running and timed sequence 2058 * number was acked, update smoothed round trip time. 2059 * Since we now have an rtt measurement, cancel the 2060 * timer backoff (cf., Phil Karn's retransmit alg.). 2061 * Recompute the initial retransmit timer. 2062 * 2063 * Some boxes send broken timestamp replies 2064 * during the SYN+ACK phase, ignore 2065 * timestamps of 0 or we could calculate a 2066 * huge RTT and blow up the retransmit timer. 2067 */ 2068 if ((to.to_flags & TOF_TS) != 0 && 2069 to.to_tsecr) { 2070 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2071 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2072 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2073 } 2074 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2075 2076 /* 2077 * If all outstanding data is acked, stop retransmit 2078 * timer and remember to restart (more output or persist). 2079 * If there is more data to be acked, restart retransmit 2080 * timer, using current (possibly backed-off) value. 2081 */ 2082 if (th->th_ack == tp->snd_max) { 2083 callout_stop(tp->tt_rexmt); 2084 needoutput = 1; 2085 } else if (!callout_active(tp->tt_persist)) 2086 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2087 tcp_timer_rexmt, tp); 2088 2089 /* 2090 * If no data (only SYN) was ACK'd, 2091 * skip rest of ACK processing. 2092 */ 2093 if (acked == 0) 2094 goto step6; 2095 2096 /* 2097 * When new data is acked, open the congestion window. 2098 * If the window gives us less than ssthresh packets 2099 * in flight, open exponentially (maxseg per packet). 2100 * Otherwise open linearly: maxseg per window 2101 * (maxseg^2 / cwnd per packet). 2102 */ 2103 if ((!tcp_do_newreno && !tp->sack_enable) || 2104 !IN_FASTRECOVERY(tp)) { 2105 register u_int cw = tp->snd_cwnd; 2106 register u_int incr = tp->t_maxseg; 2107 if (cw > tp->snd_ssthresh) 2108 incr = incr * incr / cw; 2109 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2110 } 2111 SOCKBUF_LOCK(&so->so_snd); 2112 if (acked > so->so_snd.sb_cc) { 2113 tp->snd_wnd -= so->so_snd.sb_cc; 2114 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2115 ourfinisacked = 1; 2116 } else { 2117 sbdrop_locked(&so->so_snd, acked); 2118 tp->snd_wnd -= acked; 2119 ourfinisacked = 0; 2120 } 2121 sowwakeup_locked(so); 2122 /* detect una wraparound */ 2123 if ((tcp_do_newreno || tp->sack_enable) && 2124 !IN_FASTRECOVERY(tp) && 2125 SEQ_GT(tp->snd_una, tp->snd_recover) && 2126 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2127 tp->snd_recover = th->th_ack - 1; 2128 if ((tcp_do_newreno || tp->sack_enable) && 2129 IN_FASTRECOVERY(tp) && 2130 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2131 EXIT_FASTRECOVERY(tp); 2132 tp->snd_una = th->th_ack; 2133 if (tp->sack_enable) { 2134 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2135 tp->snd_recover = tp->snd_una; 2136 } 2137 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2138 tp->snd_nxt = tp->snd_una; 2139 2140 switch (tp->t_state) { 2141 2142 /* 2143 * In FIN_WAIT_1 STATE in addition to the processing 2144 * for the ESTABLISHED state if our FIN is now acknowledged 2145 * then enter FIN_WAIT_2. 2146 */ 2147 case TCPS_FIN_WAIT_1: 2148 if (ourfinisacked) { 2149 /* 2150 * If we can't receive any more 2151 * data, then closing user can proceed. 2152 * Starting the timer is contrary to the 2153 * specification, but if we don't get a FIN 2154 * we'll hang forever. 2155 */ 2156 /* XXXjl 2157 * we should release the tp also, and use a 2158 * compressed state. 2159 */ 2160 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2161 soisdisconnected(so); 2162 callout_reset(tp->tt_2msl, tcp_maxidle, 2163 tcp_timer_2msl, tp); 2164 } 2165 tp->t_state = TCPS_FIN_WAIT_2; 2166 } 2167 break; 2168 2169 /* 2170 * In CLOSING STATE in addition to the processing for 2171 * the ESTABLISHED state if the ACK acknowledges our FIN 2172 * then enter the TIME-WAIT state, otherwise ignore 2173 * the segment. 2174 */ 2175 case TCPS_CLOSING: 2176 if (ourfinisacked) { 2177 KASSERT(headlocked, ("tcp_input: process_ACK: " 2178 "head not locked")); 2179 tcp_twstart(tp); 2180 INP_INFO_WUNLOCK(&tcbinfo); 2181 m_freem(m); 2182 return; 2183 } 2184 break; 2185 2186 /* 2187 * In LAST_ACK, we may still be waiting for data to drain 2188 * and/or to be acked, as well as for the ack of our FIN. 2189 * If our FIN is now acknowledged, delete the TCB, 2190 * enter the closed state and return. 2191 */ 2192 case TCPS_LAST_ACK: 2193 if (ourfinisacked) { 2194 KASSERT(headlocked, ("tcp_input: process_ACK:" 2195 " tcp_close: head not locked")); 2196 tp = tcp_close(tp); 2197 goto drop; 2198 } 2199 break; 2200 2201 /* 2202 * In TIME_WAIT state the only thing that should arrive 2203 * is a retransmission of the remote FIN. Acknowledge 2204 * it and restart the finack timer. 2205 */ 2206 case TCPS_TIME_WAIT: 2207 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2208 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2209 tcp_timer_2msl, tp); 2210 goto dropafterack; 2211 } 2212 } 2213 2214step6: 2215 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2216 INP_LOCK_ASSERT(inp); 2217 2218 /* 2219 * Update window information. 2220 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2221 */ 2222 if ((thflags & TH_ACK) && 2223 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2224 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2225 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2226 /* keep track of pure window updates */ 2227 if (tlen == 0 && 2228 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2229 tcpstat.tcps_rcvwinupd++; 2230 tp->snd_wnd = tiwin; 2231 tp->snd_wl1 = th->th_seq; 2232 tp->snd_wl2 = th->th_ack; 2233 if (tp->snd_wnd > tp->max_sndwnd) 2234 tp->max_sndwnd = tp->snd_wnd; 2235 needoutput = 1; 2236 } 2237 2238 /* 2239 * Process segments with URG. 2240 */ 2241 if ((thflags & TH_URG) && th->th_urp && 2242 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2243 /* 2244 * This is a kludge, but if we receive and accept 2245 * random urgent pointers, we'll crash in 2246 * soreceive. It's hard to imagine someone 2247 * actually wanting to send this much urgent data. 2248 */ 2249 SOCKBUF_LOCK(&so->so_rcv); 2250 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2251 th->th_urp = 0; /* XXX */ 2252 thflags &= ~TH_URG; /* XXX */ 2253 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2254 goto dodata; /* XXX */ 2255 } 2256 /* 2257 * If this segment advances the known urgent pointer, 2258 * then mark the data stream. This should not happen 2259 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2260 * a FIN has been received from the remote side. 2261 * In these states we ignore the URG. 2262 * 2263 * According to RFC961 (Assigned Protocols), 2264 * the urgent pointer points to the last octet 2265 * of urgent data. We continue, however, 2266 * to consider it to indicate the first octet 2267 * of data past the urgent section as the original 2268 * spec states (in one of two places). 2269 */ 2270 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2271 tp->rcv_up = th->th_seq + th->th_urp; 2272 so->so_oobmark = so->so_rcv.sb_cc + 2273 (tp->rcv_up - tp->rcv_nxt) - 1; 2274 if (so->so_oobmark == 0) 2275 so->so_rcv.sb_state |= SBS_RCVATMARK; 2276 sohasoutofband(so); 2277 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2278 } 2279 SOCKBUF_UNLOCK(&so->so_rcv); 2280 /* 2281 * Remove out of band data so doesn't get presented to user. 2282 * This can happen independent of advancing the URG pointer, 2283 * but if two URG's are pending at once, some out-of-band 2284 * data may creep in... ick. 2285 */ 2286 if (th->th_urp <= (u_long)tlen && 2287 !(so->so_options & SO_OOBINLINE)) { 2288 /* hdr drop is delayed */ 2289 tcp_pulloutofband(so, th, m, drop_hdrlen); 2290 } 2291 } else { 2292 /* 2293 * If no out of band data is expected, 2294 * pull receive urgent pointer along 2295 * with the receive window. 2296 */ 2297 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2298 tp->rcv_up = tp->rcv_nxt; 2299 } 2300dodata: /* XXX */ 2301 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2302 INP_LOCK_ASSERT(inp); 2303 2304 /* 2305 * Process the segment text, merging it into the TCP sequencing queue, 2306 * and arranging for acknowledgment of receipt if necessary. 2307 * This process logically involves adjusting tp->rcv_wnd as data 2308 * is presented to the user (this happens in tcp_usrreq.c, 2309 * case PRU_RCVD). If a FIN has already been received on this 2310 * connection then we just ignore the text. 2311 */ 2312 if ((tlen || (thflags & TH_FIN)) && 2313 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2314 m_adj(m, drop_hdrlen); /* delayed header drop */ 2315 /* 2316 * Insert segment which includes th into TCP reassembly queue 2317 * with control block tp. Set thflags to whether reassembly now 2318 * includes a segment with FIN. This handles the common case 2319 * inline (segment is the next to be received on an established 2320 * connection, and the queue is empty), avoiding linkage into 2321 * and removal from the queue and repetition of various 2322 * conversions. 2323 * Set DELACK for segments received in order, but ack 2324 * immediately when segments are out of order (so 2325 * fast retransmit can work). 2326 */ 2327 if (th->th_seq == tp->rcv_nxt && 2328 LIST_EMPTY(&tp->t_segq) && 2329 TCPS_HAVEESTABLISHED(tp->t_state)) { 2330 if (DELAY_ACK(tp)) 2331 tp->t_flags |= TF_DELACK; 2332 else 2333 tp->t_flags |= TF_ACKNOW; 2334 tp->rcv_nxt += tlen; 2335 thflags = th->th_flags & TH_FIN; 2336 tcpstat.tcps_rcvpack++; 2337 tcpstat.tcps_rcvbyte += tlen; 2338 ND6_HINT(tp); 2339 SOCKBUF_LOCK(&so->so_rcv); 2340 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2341 m_freem(m); 2342 else 2343 sbappendstream_locked(&so->so_rcv, m); 2344 sorwakeup_locked(so); 2345 } else { 2346 thflags = tcp_reass(tp, th, &tlen, m); 2347 tp->t_flags |= TF_ACKNOW; 2348 } 2349 if (tlen > 0 && tp->sack_enable) 2350 tcp_update_sack_list(tp, th->th_seq, th->th_seq + tlen); 2351 /* 2352 * Note the amount of data that peer has sent into 2353 * our window, in order to estimate the sender's 2354 * buffer size. 2355 */ 2356 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2357 } else { 2358 m_freem(m); 2359 thflags &= ~TH_FIN; 2360 } 2361 2362 /* 2363 * If FIN is received ACK the FIN and let the user know 2364 * that the connection is closing. 2365 */ 2366 if (thflags & TH_FIN) { 2367 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2368 socantrcvmore(so); 2369 /* 2370 * If connection is half-synchronized 2371 * (ie NEEDSYN flag on) then delay ACK, 2372 * so it may be piggybacked when SYN is sent. 2373 * Otherwise, since we received a FIN then no 2374 * more input can be expected, send ACK now. 2375 */ 2376 if (tp->t_flags & TF_NEEDSYN) 2377 tp->t_flags |= TF_DELACK; 2378 else 2379 tp->t_flags |= TF_ACKNOW; 2380 tp->rcv_nxt++; 2381 } 2382 switch (tp->t_state) { 2383 2384 /* 2385 * In SYN_RECEIVED and ESTABLISHED STATES 2386 * enter the CLOSE_WAIT state. 2387 */ 2388 case TCPS_SYN_RECEIVED: 2389 tp->t_starttime = ticks; 2390 /*FALLTHROUGH*/ 2391 case TCPS_ESTABLISHED: 2392 tp->t_state = TCPS_CLOSE_WAIT; 2393 break; 2394 2395 /* 2396 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2397 * enter the CLOSING state. 2398 */ 2399 case TCPS_FIN_WAIT_1: 2400 tp->t_state = TCPS_CLOSING; 2401 break; 2402 2403 /* 2404 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2405 * starting the time-wait timer, turning off the other 2406 * standard timers. 2407 */ 2408 case TCPS_FIN_WAIT_2: 2409 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2410 "TCP_FIN_WAIT_2: head not locked")); 2411 tcp_twstart(tp); 2412 INP_INFO_WUNLOCK(&tcbinfo); 2413 return; 2414 2415 /* 2416 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2417 */ 2418 case TCPS_TIME_WAIT: 2419 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2420 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2421 tcp_timer_2msl, tp); 2422 break; 2423 } 2424 } 2425 INP_INFO_WUNLOCK(&tcbinfo); 2426 headlocked = 0; 2427#ifdef TCPDEBUG 2428 if (so->so_options & SO_DEBUG) 2429 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2430 &tcp_savetcp, 0); 2431#endif 2432 2433 /* 2434 * Return any desired output. 2435 */ 2436 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2437 (void) tcp_output(tp); 2438 2439check_delack: 2440 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2441 INP_LOCK_ASSERT(inp); 2442 if (tp->t_flags & TF_DELACK) { 2443 tp->t_flags &= ~TF_DELACK; 2444 callout_reset(tp->tt_delack, tcp_delacktime, 2445 tcp_timer_delack, tp); 2446 } 2447 INP_UNLOCK(inp); 2448 return; 2449 2450dropafterack: 2451 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2452 /* 2453 * Generate an ACK dropping incoming segment if it occupies 2454 * sequence space, where the ACK reflects our state. 2455 * 2456 * We can now skip the test for the RST flag since all 2457 * paths to this code happen after packets containing 2458 * RST have been dropped. 2459 * 2460 * In the SYN-RECEIVED state, don't send an ACK unless the 2461 * segment we received passes the SYN-RECEIVED ACK test. 2462 * If it fails send a RST. This breaks the loop in the 2463 * "LAND" DoS attack, and also prevents an ACK storm 2464 * between two listening ports that have been sent forged 2465 * SYN segments, each with the source address of the other. 2466 */ 2467 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2468 (SEQ_GT(tp->snd_una, th->th_ack) || 2469 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2470 rstreason = BANDLIM_RST_OPENPORT; 2471 goto dropwithreset; 2472 } 2473#ifdef TCPDEBUG 2474 if (so->so_options & SO_DEBUG) 2475 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2476 &tcp_savetcp, 0); 2477#endif 2478 KASSERT(headlocked, ("headlocked should be 1")); 2479 INP_INFO_WUNLOCK(&tcbinfo); 2480 tp->t_flags |= TF_ACKNOW; 2481 (void) tcp_output(tp); 2482 INP_UNLOCK(inp); 2483 m_freem(m); 2484 return; 2485 2486dropwithreset: 2487 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2488 /* 2489 * Generate a RST, dropping incoming segment. 2490 * Make ACK acceptable to originator of segment. 2491 * Don't bother to respond if destination was broadcast/multicast. 2492 */ 2493 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2494 goto drop; 2495 if (isipv6) { 2496 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2497 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2498 goto drop; 2499 } else { 2500 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2501 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2502 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2503 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2504 goto drop; 2505 } 2506 /* IPv6 anycast check is done at tcp6_input() */ 2507 2508 /* 2509 * Perform bandwidth limiting. 2510 */ 2511 if (badport_bandlim(rstreason) < 0) 2512 goto drop; 2513 2514#ifdef TCPDEBUG 2515 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2516 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2517 &tcp_savetcp, 0); 2518#endif 2519 2520 if (thflags & TH_ACK) 2521 /* mtod() below is safe as long as hdr dropping is delayed */ 2522 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2523 TH_RST); 2524 else { 2525 if (thflags & TH_SYN) 2526 tlen++; 2527 /* mtod() below is safe as long as hdr dropping is delayed */ 2528 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2529 (tcp_seq)0, TH_RST|TH_ACK); 2530 } 2531 2532 if (tp) 2533 INP_UNLOCK(inp); 2534 if (headlocked) 2535 INP_INFO_WUNLOCK(&tcbinfo); 2536 return; 2537 2538drop: 2539 /* 2540 * Drop space held by incoming segment and return. 2541 */ 2542#ifdef TCPDEBUG 2543 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2544 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2545 &tcp_savetcp, 0); 2546#endif 2547 if (tp) 2548 INP_UNLOCK(inp); 2549 if (headlocked) 2550 INP_INFO_WUNLOCK(&tcbinfo); 2551 m_freem(m); 2552 return; 2553} 2554 2555/* 2556 * Parse TCP options and place in tcpopt. 2557 */ 2558static void 2559tcp_dooptions(to, cp, cnt, is_syn) 2560 struct tcpopt *to; 2561 u_char *cp; 2562 int cnt; 2563 int is_syn; 2564{ 2565 int opt, optlen; 2566 2567 to->to_flags = 0; 2568 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2569 opt = cp[0]; 2570 if (opt == TCPOPT_EOL) 2571 break; 2572 if (opt == TCPOPT_NOP) 2573 optlen = 1; 2574 else { 2575 if (cnt < 2) 2576 break; 2577 optlen = cp[1]; 2578 if (optlen < 2 || optlen > cnt) 2579 break; 2580 } 2581 switch (opt) { 2582 case TCPOPT_MAXSEG: 2583 if (optlen != TCPOLEN_MAXSEG) 2584 continue; 2585 if (!is_syn) 2586 continue; 2587 to->to_flags |= TOF_MSS; 2588 bcopy((char *)cp + 2, 2589 (char *)&to->to_mss, sizeof(to->to_mss)); 2590 to->to_mss = ntohs(to->to_mss); 2591 break; 2592 case TCPOPT_WINDOW: 2593 if (optlen != TCPOLEN_WINDOW) 2594 continue; 2595 if (! is_syn) 2596 continue; 2597 to->to_flags |= TOF_SCALE; 2598 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2599 break; 2600 case TCPOPT_TIMESTAMP: 2601 if (optlen != TCPOLEN_TIMESTAMP) 2602 continue; 2603 to->to_flags |= TOF_TS; 2604 bcopy((char *)cp + 2, 2605 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2606 to->to_tsval = ntohl(to->to_tsval); 2607 bcopy((char *)cp + 6, 2608 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2609 to->to_tsecr = ntohl(to->to_tsecr); 2610 /* 2611 * If echoed timestamp is later than the current time, 2612 * fall back to non RFC1323 RTT calculation. 2613 */ 2614 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks)) 2615 to->to_tsecr = 0; 2616 break; 2617#ifdef TCP_SIGNATURE 2618 /* 2619 * XXX In order to reply to a host which has set the 2620 * TCP_SIGNATURE option in its initial SYN, we have to 2621 * record the fact that the option was observed here 2622 * for the syncache code to perform the correct response. 2623 */ 2624 case TCPOPT_SIGNATURE: 2625 if (optlen != TCPOLEN_SIGNATURE) 2626 continue; 2627 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2628 break; 2629#endif 2630 case TCPOPT_SACK_PERMITTED: 2631 if (!tcp_do_sack || 2632 optlen != TCPOLEN_SACK_PERMITTED) 2633 continue; 2634 if (is_syn) { 2635 /* MUST only be set on SYN */ 2636 to->to_flags |= TOF_SACK; 2637 } 2638 break; 2639 case TCPOPT_SACK: 2640 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2641 continue; 2642 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2643 to->to_sacks = cp + 2; 2644 tcpstat.tcps_sack_rcv_blocks++; 2645 break; 2646 default: 2647 continue; 2648 } 2649 } 2650} 2651 2652/* 2653 * Pull out of band byte out of a segment so 2654 * it doesn't appear in the user's data queue. 2655 * It is still reflected in the segment length for 2656 * sequencing purposes. 2657 */ 2658static void 2659tcp_pulloutofband(so, th, m, off) 2660 struct socket *so; 2661 struct tcphdr *th; 2662 register struct mbuf *m; 2663 int off; /* delayed to be droped hdrlen */ 2664{ 2665 int cnt = off + th->th_urp - 1; 2666 2667 while (cnt >= 0) { 2668 if (m->m_len > cnt) { 2669 char *cp = mtod(m, caddr_t) + cnt; 2670 struct tcpcb *tp = sototcpcb(so); 2671 2672 tp->t_iobc = *cp; 2673 tp->t_oobflags |= TCPOOB_HAVEDATA; 2674 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2675 m->m_len--; 2676 if (m->m_flags & M_PKTHDR) 2677 m->m_pkthdr.len--; 2678 return; 2679 } 2680 cnt -= m->m_len; 2681 m = m->m_next; 2682 if (m == 0) 2683 break; 2684 } 2685 panic("tcp_pulloutofband"); 2686} 2687 2688/* 2689 * Collect new round-trip time estimate 2690 * and update averages and current timeout. 2691 */ 2692static void 2693tcp_xmit_timer(tp, rtt) 2694 register struct tcpcb *tp; 2695 int rtt; 2696{ 2697 register int delta; 2698 2699 INP_LOCK_ASSERT(tp->t_inpcb); 2700 2701 tcpstat.tcps_rttupdated++; 2702 tp->t_rttupdated++; 2703 if (tp->t_srtt != 0) { 2704 /* 2705 * srtt is stored as fixed point with 5 bits after the 2706 * binary point (i.e., scaled by 8). The following magic 2707 * is equivalent to the smoothing algorithm in rfc793 with 2708 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2709 * point). Adjust rtt to origin 0. 2710 */ 2711 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2712 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2713 2714 if ((tp->t_srtt += delta) <= 0) 2715 tp->t_srtt = 1; 2716 2717 /* 2718 * We accumulate a smoothed rtt variance (actually, a 2719 * smoothed mean difference), then set the retransmit 2720 * timer to smoothed rtt + 4 times the smoothed variance. 2721 * rttvar is stored as fixed point with 4 bits after the 2722 * binary point (scaled by 16). The following is 2723 * equivalent to rfc793 smoothing with an alpha of .75 2724 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2725 * rfc793's wired-in beta. 2726 */ 2727 if (delta < 0) 2728 delta = -delta; 2729 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2730 if ((tp->t_rttvar += delta) <= 0) 2731 tp->t_rttvar = 1; 2732 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2733 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2734 } else { 2735 /* 2736 * No rtt measurement yet - use the unsmoothed rtt. 2737 * Set the variance to half the rtt (so our first 2738 * retransmit happens at 3*rtt). 2739 */ 2740 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2741 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2742 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2743 } 2744 tp->t_rtttime = 0; 2745 tp->t_rxtshift = 0; 2746 2747 /* 2748 * the retransmit should happen at rtt + 4 * rttvar. 2749 * Because of the way we do the smoothing, srtt and rttvar 2750 * will each average +1/2 tick of bias. When we compute 2751 * the retransmit timer, we want 1/2 tick of rounding and 2752 * 1 extra tick because of +-1/2 tick uncertainty in the 2753 * firing of the timer. The bias will give us exactly the 2754 * 1.5 tick we need. But, because the bias is 2755 * statistical, we have to test that we don't drop below 2756 * the minimum feasible timer (which is 2 ticks). 2757 */ 2758 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2759 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2760 2761 /* 2762 * We received an ack for a packet that wasn't retransmitted; 2763 * it is probably safe to discard any error indications we've 2764 * received recently. This isn't quite right, but close enough 2765 * for now (a route might have failed after we sent a segment, 2766 * and the return path might not be symmetrical). 2767 */ 2768 tp->t_softerror = 0; 2769} 2770 2771/* 2772 * Determine a reasonable value for maxseg size. 2773 * If the route is known, check route for mtu. 2774 * If none, use an mss that can be handled on the outgoing 2775 * interface without forcing IP to fragment; if bigger than 2776 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2777 * to utilize large mbufs. If no route is found, route has no mtu, 2778 * or the destination isn't local, use a default, hopefully conservative 2779 * size (usually 512 or the default IP max size, but no more than the mtu 2780 * of the interface), as we can't discover anything about intervening 2781 * gateways or networks. We also initialize the congestion/slow start 2782 * window to be a single segment if the destination isn't local. 2783 * While looking at the routing entry, we also initialize other path-dependent 2784 * parameters from pre-set or cached values in the routing entry. 2785 * 2786 * Also take into account the space needed for options that we 2787 * send regularly. Make maxseg shorter by that amount to assure 2788 * that we can send maxseg amount of data even when the options 2789 * are present. Store the upper limit of the length of options plus 2790 * data in maxopd. 2791 * 2792 * 2793 * In case of T/TCP, we call this routine during implicit connection 2794 * setup as well (offer = -1), to initialize maxseg from the cached 2795 * MSS of our peer. 2796 * 2797 * NOTE that this routine is only called when we process an incoming 2798 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2799 */ 2800void 2801tcp_mss(tp, offer) 2802 struct tcpcb *tp; 2803 int offer; 2804{ 2805 int rtt, mss; 2806 u_long bufsize; 2807 u_long maxmtu; 2808 struct inpcb *inp = tp->t_inpcb; 2809 struct socket *so; 2810 struct hc_metrics_lite metrics; 2811 int origoffer = offer; 2812#ifdef INET6 2813 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2814 size_t min_protoh = isipv6 ? 2815 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2816 sizeof (struct tcpiphdr); 2817#else 2818 const size_t min_protoh = sizeof(struct tcpiphdr); 2819#endif 2820 2821 /* initialize */ 2822#ifdef INET6 2823 if (isipv6) { 2824 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2825 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2826 } else 2827#endif 2828 { 2829 maxmtu = tcp_maxmtu(&inp->inp_inc); 2830 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2831 } 2832 so = inp->inp_socket; 2833 2834 /* 2835 * no route to sender, stay with default mss and return 2836 */ 2837 if (maxmtu == 0) 2838 return; 2839 2840 /* what have we got? */ 2841 switch (offer) { 2842 case 0: 2843 /* 2844 * Offer == 0 means that there was no MSS on the SYN 2845 * segment, in this case we use tcp_mssdflt. 2846 */ 2847 offer = 2848#ifdef INET6 2849 isipv6 ? tcp_v6mssdflt : 2850#endif 2851 tcp_mssdflt; 2852 break; 2853 2854 case -1: 2855 /* 2856 * Offer == -1 means that we didn't receive SYN yet. 2857 */ 2858 /* FALLTHROUGH */ 2859 2860 default: 2861 /* 2862 * Prevent DoS attack with too small MSS. Round up 2863 * to at least minmss. 2864 */ 2865 offer = max(offer, tcp_minmss); 2866 /* 2867 * Sanity check: make sure that maxopd will be large 2868 * enough to allow some data on segments even if the 2869 * all the option space is used (40bytes). Otherwise 2870 * funny things may happen in tcp_output. 2871 */ 2872 offer = max(offer, 64); 2873 } 2874 2875 /* 2876 * rmx information is now retrieved from tcp_hostcache 2877 */ 2878 tcp_hc_get(&inp->inp_inc, &metrics); 2879 2880 /* 2881 * if there's a discovered mtu int tcp hostcache, use it 2882 * else, use the link mtu. 2883 */ 2884 if (metrics.rmx_mtu) 2885 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2886 else { 2887#ifdef INET6 2888 if (isipv6) { 2889 mss = maxmtu - min_protoh; 2890 if (!path_mtu_discovery && 2891 !in6_localaddr(&inp->in6p_faddr)) 2892 mss = min(mss, tcp_v6mssdflt); 2893 } else 2894#endif 2895 { 2896 mss = maxmtu - min_protoh; 2897 if (!path_mtu_discovery && 2898 !in_localaddr(inp->inp_faddr)) 2899 mss = min(mss, tcp_mssdflt); 2900 } 2901 } 2902 mss = min(mss, offer); 2903 2904 /* 2905 * maxopd stores the maximum length of data AND options 2906 * in a segment; maxseg is the amount of data in a normal 2907 * segment. We need to store this value (maxopd) apart 2908 * from maxseg, because now every segment carries options 2909 * and thus we normally have somewhat less data in segments. 2910 */ 2911 tp->t_maxopd = mss; 2912 2913 /* 2914 * origoffer==-1 indicates, that no segments were received yet. 2915 * In this case we just guess. 2916 */ 2917 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2918 (origoffer == -1 || 2919 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2920 mss -= TCPOLEN_TSTAMP_APPA; 2921 tp->t_maxseg = mss; 2922 2923#if (MCLBYTES & (MCLBYTES - 1)) == 0 2924 if (mss > MCLBYTES) 2925 mss &= ~(MCLBYTES-1); 2926#else 2927 if (mss > MCLBYTES) 2928 mss = mss / MCLBYTES * MCLBYTES; 2929#endif 2930 tp->t_maxseg = mss; 2931 2932 /* 2933 * If there's a pipesize, change the socket buffer to that size, 2934 * don't change if sb_hiwat is different than default (then it 2935 * has been changed on purpose with setsockopt). 2936 * Make the socket buffers an integral number of mss units; 2937 * if the mss is larger than the socket buffer, decrease the mss. 2938 */ 2939 SOCKBUF_LOCK(&so->so_snd); 2940 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2941 bufsize = metrics.rmx_sendpipe; 2942 else 2943 bufsize = so->so_snd.sb_hiwat; 2944 if (bufsize < mss) 2945 mss = bufsize; 2946 else { 2947 bufsize = roundup(bufsize, mss); 2948 if (bufsize > sb_max) 2949 bufsize = sb_max; 2950 if (bufsize > so->so_snd.sb_hiwat) 2951 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2952 } 2953 SOCKBUF_UNLOCK(&so->so_snd); 2954 tp->t_maxseg = mss; 2955 2956 SOCKBUF_LOCK(&so->so_rcv); 2957 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2958 bufsize = metrics.rmx_recvpipe; 2959 else 2960 bufsize = so->so_rcv.sb_hiwat; 2961 if (bufsize > mss) { 2962 bufsize = roundup(bufsize, mss); 2963 if (bufsize > sb_max) 2964 bufsize = sb_max; 2965 if (bufsize > so->so_rcv.sb_hiwat) 2966 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2967 } 2968 SOCKBUF_UNLOCK(&so->so_rcv); 2969 /* 2970 * While we're here, check the others too 2971 */ 2972 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2973 tp->t_srtt = rtt; 2974 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2975 tcpstat.tcps_usedrtt++; 2976 if (metrics.rmx_rttvar) { 2977 tp->t_rttvar = metrics.rmx_rttvar; 2978 tcpstat.tcps_usedrttvar++; 2979 } else { 2980 /* default variation is +- 1 rtt */ 2981 tp->t_rttvar = 2982 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2983 } 2984 TCPT_RANGESET(tp->t_rxtcur, 2985 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2986 tp->t_rttmin, TCPTV_REXMTMAX); 2987 } 2988 if (metrics.rmx_ssthresh) { 2989 /* 2990 * There's some sort of gateway or interface 2991 * buffer limit on the path. Use this to set 2992 * the slow start threshhold, but set the 2993 * threshold to no less than 2*mss. 2994 */ 2995 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2996 tcpstat.tcps_usedssthresh++; 2997 } 2998 if (metrics.rmx_bandwidth) 2999 tp->snd_bandwidth = metrics.rmx_bandwidth; 3000 3001 /* 3002 * Set the slow-start flight size depending on whether this 3003 * is a local network or not. 3004 * 3005 * Extend this so we cache the cwnd too and retrieve it here. 3006 * Make cwnd even bigger than RFC3390 suggests but only if we 3007 * have previous experience with the remote host. Be careful 3008 * not make cwnd bigger than remote receive window or our own 3009 * send socket buffer. Maybe put some additional upper bound 3010 * on the retrieved cwnd. Should do incremental updates to 3011 * hostcache when cwnd collapses so next connection doesn't 3012 * overloads the path again. 3013 * 3014 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3015 * We currently check only in syncache_socket for that. 3016 */ 3017#define TCP_METRICS_CWND 3018#ifdef TCP_METRICS_CWND 3019 if (metrics.rmx_cwnd) 3020 tp->snd_cwnd = max(mss, 3021 min(metrics.rmx_cwnd / 2, 3022 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3023 else 3024#endif 3025 if (tcp_do_rfc3390) 3026 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3027#ifdef INET6 3028 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3029 (!isipv6 && in_localaddr(inp->inp_faddr))) 3030#else 3031 else if (in_localaddr(inp->inp_faddr)) 3032#endif 3033 tp->snd_cwnd = mss * ss_fltsz_local; 3034 else 3035 tp->snd_cwnd = mss * ss_fltsz; 3036} 3037 3038/* 3039 * Determine the MSS option to send on an outgoing SYN. 3040 */ 3041int 3042tcp_mssopt(inc) 3043 struct in_conninfo *inc; 3044{ 3045 int mss = 0; 3046 u_long maxmtu = 0; 3047 u_long thcmtu = 0; 3048 size_t min_protoh; 3049#ifdef INET6 3050 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3051#endif 3052 3053 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3054 3055#ifdef INET6 3056 if (isipv6) { 3057 mss = tcp_v6mssdflt; 3058 maxmtu = tcp_maxmtu6(inc); 3059 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3060 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3061 } else 3062#endif 3063 { 3064 mss = tcp_mssdflt; 3065 maxmtu = tcp_maxmtu(inc); 3066 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3067 min_protoh = sizeof(struct tcpiphdr); 3068 } 3069 if (maxmtu && thcmtu) 3070 mss = min(maxmtu, thcmtu) - min_protoh; 3071 else if (maxmtu || thcmtu) 3072 mss = max(maxmtu, thcmtu) - min_protoh; 3073 3074 return (mss); 3075} 3076 3077 3078/* 3079 * On a partial ack arrives, force the retransmission of the 3080 * next unacknowledged segment. Do not clear tp->t_dupacks. 3081 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3082 * be started again. 3083 */ 3084static void 3085tcp_newreno_partial_ack(tp, th) 3086 struct tcpcb *tp; 3087 struct tcphdr *th; 3088{ 3089 tcp_seq onxt = tp->snd_nxt; 3090 u_long ocwnd = tp->snd_cwnd; 3091 3092 callout_stop(tp->tt_rexmt); 3093 tp->t_rtttime = 0; 3094 tp->snd_nxt = th->th_ack; 3095 /* 3096 * Set snd_cwnd to one segment beyond acknowledged offset. 3097 * (tp->snd_una has not yet been updated when this function is called.) 3098 */ 3099 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3100 tp->t_flags |= TF_ACKNOW; 3101 (void) tcp_output(tp); 3102 tp->snd_cwnd = ocwnd; 3103 if (SEQ_GT(onxt, tp->snd_nxt)) 3104 tp->snd_nxt = onxt; 3105 /* 3106 * Partial window deflation. Relies on fact that tp->snd_una 3107 * not updated yet. 3108 */ 3109 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3110} 3111 3112/* 3113 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3114 * looking for a pcb in the listen state. Returns 0 otherwise. 3115 */ 3116static int 3117tcp_timewait(tw, to, th, m, tlen) 3118 struct tcptw *tw; 3119 struct tcpopt *to; 3120 struct tcphdr *th; 3121 struct mbuf *m; 3122 int tlen; 3123{ 3124 int thflags; 3125 tcp_seq seq; 3126#ifdef INET6 3127 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3128#else 3129 const int isipv6 = 0; 3130#endif 3131 3132 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */ 3133 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3134 INP_LOCK_ASSERT(tw->tw_inpcb); 3135 3136 thflags = th->th_flags; 3137 3138 /* 3139 * NOTE: for FIN_WAIT_2 (to be added later), 3140 * must validate sequence number before accepting RST 3141 */ 3142 3143 /* 3144 * If the segment contains RST: 3145 * Drop the segment - see Stevens, vol. 2, p. 964 and 3146 * RFC 1337. 3147 */ 3148 if (thflags & TH_RST) 3149 goto drop; 3150 3151#if 0 3152/* PAWS not needed at the moment */ 3153 /* 3154 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3155 * and it's less than ts_recent, drop it. 3156 */ 3157 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3158 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3159 if ((thflags & TH_ACK) == 0) 3160 goto drop; 3161 goto ack; 3162 } 3163 /* 3164 * ts_recent is never updated because we never accept new segments. 3165 */ 3166#endif 3167 3168 /* 3169 * If a new connection request is received 3170 * while in TIME_WAIT, drop the old connection 3171 * and start over if the sequence numbers 3172 * are above the previous ones. 3173 */ 3174 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3175 (void) tcp_twclose(tw, 0); 3176 return (1); 3177 } 3178 3179 /* 3180 * Drop the the segment if it does not contain an ACK. 3181 */ 3182 if ((thflags & TH_ACK) == 0) 3183 goto drop; 3184 3185 /* 3186 * Reset the 2MSL timer if this is a duplicate FIN. 3187 */ 3188 if (thflags & TH_FIN) { 3189 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3190 if (seq + 1 == tw->rcv_nxt) 3191 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3192 } 3193 3194 /* 3195 * Acknowledge the segment if it has data or is not a duplicate ACK. 3196 */ 3197 if (thflags != TH_ACK || tlen != 0 || 3198 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3199 tcp_twrespond(tw, TH_ACK); 3200 goto drop; 3201 3202 /* 3203 * Generate a RST, dropping incoming segment. 3204 * Make ACK acceptable to originator of segment. 3205 * Don't bother to respond if destination was broadcast/multicast. 3206 */ 3207 if (m->m_flags & (M_BCAST|M_MCAST)) 3208 goto drop; 3209 if (isipv6) { 3210 struct ip6_hdr *ip6; 3211 3212 /* IPv6 anycast check is done at tcp6_input() */ 3213 ip6 = mtod(m, struct ip6_hdr *); 3214 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3215 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3216 goto drop; 3217 } else { 3218 struct ip *ip; 3219 3220 ip = mtod(m, struct ip *); 3221 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3222 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3223 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3224 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3225 goto drop; 3226 } 3227 if (thflags & TH_ACK) { 3228 tcp_respond(NULL, 3229 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3230 } else { 3231 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3232 tcp_respond(NULL, 3233 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3234 } 3235 INP_UNLOCK(tw->tw_inpcb); 3236 return (0); 3237 3238drop: 3239 INP_UNLOCK(tw->tw_inpcb); 3240 m_freem(m); 3241 return (0); 3242}
| 1086 if (to.to_flags & TOF_SCALE) { 1087 tp->t_flags |= TF_RCVD_SCALE; 1088 tp->requested_s_scale = to.to_requested_s_scale; 1089 } 1090 if (to.to_flags & TOF_TS) { 1091 tp->t_flags |= TF_RCVD_TSTMP; 1092 tp->ts_recent = to.to_tsval; 1093 tp->ts_recent_age = ticks; 1094 } 1095 if (to.to_flags & TOF_MSS) 1096 tcp_mss(tp, to.to_mss); 1097 if (tp->sack_enable) { 1098 if (!(to.to_flags & TOF_SACK)) 1099 tp->sack_enable = 0; 1100 else 1101 tp->t_flags |= TF_SACK_PERMIT; 1102 } 1103 1104 } 1105 1106 /* 1107 * Header prediction: check for the two common cases 1108 * of a uni-directional data xfer. If the packet has 1109 * no control flags, is in-sequence, the window didn't 1110 * change and we're not retransmitting, it's a 1111 * candidate. If the length is zero and the ack moved 1112 * forward, we're the sender side of the xfer. Just 1113 * free the data acked & wake any higher level process 1114 * that was blocked waiting for space. If the length 1115 * is non-zero and the ack didn't move, we're the 1116 * receiver side. If we're getting packets in-order 1117 * (the reassembly queue is empty), add the data to 1118 * the socket buffer and note that we need a delayed ack. 1119 * Make sure that the hidden state-flags are also off. 1120 * Since we check for TCPS_ESTABLISHED above, it can only 1121 * be TH_NEEDSYN. 1122 */ 1123 if (tp->t_state == TCPS_ESTABLISHED && 1124 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1125 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1126 ((to.to_flags & TOF_TS) == 0 || 1127 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1128 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1129 tp->snd_nxt == tp->snd_max) { 1130 1131 /* 1132 * If last ACK falls within this segment's sequence numbers, 1133 * record the timestamp. 1134 * NOTE that the test is modified according to the latest 1135 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1136 */ 1137 if ((to.to_flags & TOF_TS) != 0 && 1138 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1139 tp->ts_recent_age = ticks; 1140 tp->ts_recent = to.to_tsval; 1141 } 1142 1143 if (tlen == 0) { 1144 if (SEQ_GT(th->th_ack, tp->snd_una) && 1145 SEQ_LEQ(th->th_ack, tp->snd_max) && 1146 tp->snd_cwnd >= tp->snd_wnd && 1147 ((!tcp_do_newreno && !tp->sack_enable && 1148 tp->t_dupacks < tcprexmtthresh) || 1149 ((tcp_do_newreno || tp->sack_enable) && 1150 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0))) { 1151 KASSERT(headlocked, ("headlocked")); 1152 INP_INFO_WUNLOCK(&tcbinfo); 1153 headlocked = 0; 1154 /* 1155 * this is a pure ack for outstanding data. 1156 */ 1157 ++tcpstat.tcps_predack; 1158 /* 1159 * "bad retransmit" recovery 1160 */ 1161 if (tp->t_rxtshift == 1 && 1162 ticks < tp->t_badrxtwin) { 1163 ++tcpstat.tcps_sndrexmitbad; 1164 tp->snd_cwnd = tp->snd_cwnd_prev; 1165 tp->snd_ssthresh = 1166 tp->snd_ssthresh_prev; 1167 tp->snd_recover = tp->snd_recover_prev; 1168 if (tp->t_flags & TF_WASFRECOVERY) 1169 ENTER_FASTRECOVERY(tp); 1170 tp->snd_nxt = tp->snd_max; 1171 tp->t_badrxtwin = 0; 1172 } 1173 1174 /* 1175 * Recalculate the transmit timer / rtt. 1176 * 1177 * Some boxes send broken timestamp replies 1178 * during the SYN+ACK phase, ignore 1179 * timestamps of 0 or we could calculate a 1180 * huge RTT and blow up the retransmit timer. 1181 */ 1182 if ((to.to_flags & TOF_TS) != 0 && 1183 to.to_tsecr) { 1184 tcp_xmit_timer(tp, 1185 ticks - to.to_tsecr + 1); 1186 } else if (tp->t_rtttime && 1187 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1188 tcp_xmit_timer(tp, 1189 ticks - tp->t_rtttime); 1190 } 1191 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1192 acked = th->th_ack - tp->snd_una; 1193 tcpstat.tcps_rcvackpack++; 1194 tcpstat.tcps_rcvackbyte += acked; 1195 sbdrop(&so->so_snd, acked); 1196 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1197 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1198 tp->snd_recover = th->th_ack - 1; 1199 tp->snd_una = th->th_ack; 1200 /* 1201 * pull snd_wl2 up to prevent seq wrap relative 1202 * to th_ack. 1203 */ 1204 tp->snd_wl2 = th->th_ack; 1205 tp->t_dupacks = 0; 1206 m_freem(m); 1207 ND6_HINT(tp); /* some progress has been done */ 1208 1209 /* 1210 * If all outstanding data are acked, stop 1211 * retransmit timer, otherwise restart timer 1212 * using current (possibly backed-off) value. 1213 * If process is waiting for space, 1214 * wakeup/selwakeup/signal. If data 1215 * are ready to send, let tcp_output 1216 * decide between more output or persist. 1217 1218#ifdef TCPDEBUG 1219 if (so->so_options & SO_DEBUG) 1220 tcp_trace(TA_INPUT, ostate, tp, 1221 (void *)tcp_saveipgen, 1222 &tcp_savetcp, 0); 1223#endif 1224 */ 1225 if (tp->snd_una == tp->snd_max) 1226 callout_stop(tp->tt_rexmt); 1227 else if (!callout_active(tp->tt_persist)) 1228 callout_reset(tp->tt_rexmt, 1229 tp->t_rxtcur, 1230 tcp_timer_rexmt, tp); 1231 1232 sowwakeup(so); 1233 if (so->so_snd.sb_cc) 1234 (void) tcp_output(tp); 1235 goto check_delack; 1236 } 1237 } else if (th->th_ack == tp->snd_una && 1238 LIST_EMPTY(&tp->t_segq) && 1239 tlen <= sbspace(&so->so_rcv)) { 1240 KASSERT(headlocked, ("headlocked")); 1241 INP_INFO_WUNLOCK(&tcbinfo); 1242 headlocked = 0; 1243 /* 1244 * this is a pure, in-sequence data packet 1245 * with nothing on the reassembly queue and 1246 * we have enough buffer space to take it. 1247 */ 1248 /* Clean receiver SACK report if present */ 1249 if (tp->sack_enable && tp->rcv_numsacks) 1250 tcp_clean_sackreport(tp); 1251 ++tcpstat.tcps_preddat; 1252 tp->rcv_nxt += tlen; 1253 /* 1254 * Pull snd_wl1 up to prevent seq wrap relative to 1255 * th_seq. 1256 */ 1257 tp->snd_wl1 = th->th_seq; 1258 /* 1259 * Pull rcv_up up to prevent seq wrap relative to 1260 * rcv_nxt. 1261 */ 1262 tp->rcv_up = tp->rcv_nxt; 1263 tcpstat.tcps_rcvpack++; 1264 tcpstat.tcps_rcvbyte += tlen; 1265 ND6_HINT(tp); /* some progress has been done */ 1266 /* 1267#ifdef TCPDEBUG 1268 if (so->so_options & SO_DEBUG) 1269 tcp_trace(TA_INPUT, ostate, tp, 1270 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1271#endif 1272 * Add data to socket buffer. 1273 */ 1274 SOCKBUF_LOCK(&so->so_rcv); 1275 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1276 m_freem(m); 1277 } else { 1278 m_adj(m, drop_hdrlen); /* delayed header drop */ 1279 sbappendstream_locked(&so->so_rcv, m); 1280 } 1281 sorwakeup_locked(so); 1282 if (DELAY_ACK(tp)) { 1283 tp->t_flags |= TF_DELACK; 1284 } else { 1285 tp->t_flags |= TF_ACKNOW; 1286 tcp_output(tp); 1287 } 1288 goto check_delack; 1289 } 1290 } 1291 1292 /* 1293 * Calculate amount of space in receive window, 1294 * and then do TCP input processing. 1295 * Receive window is amount of space in rcv queue, 1296 * but not less than advertised window. 1297 */ 1298 { int win; 1299 1300 win = sbspace(&so->so_rcv); 1301 if (win < 0) 1302 win = 0; 1303 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1304 } 1305 1306 switch (tp->t_state) { 1307 1308 /* 1309 * If the state is SYN_RECEIVED: 1310 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1311 */ 1312 case TCPS_SYN_RECEIVED: 1313 if ((thflags & TH_ACK) && 1314 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1315 SEQ_GT(th->th_ack, tp->snd_max))) { 1316 rstreason = BANDLIM_RST_OPENPORT; 1317 goto dropwithreset; 1318 } 1319 break; 1320 1321 /* 1322 * If the state is SYN_SENT: 1323 * if seg contains an ACK, but not for our SYN, drop the input. 1324 * if seg contains a RST, then drop the connection. 1325 * if seg does not contain SYN, then drop it. 1326 * Otherwise this is an acceptable SYN segment 1327 * initialize tp->rcv_nxt and tp->irs 1328 * if seg contains ack then advance tp->snd_una 1329 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1330 * arrange for segment to be acked (eventually) 1331 * continue processing rest of data/controls, beginning with URG 1332 */ 1333 case TCPS_SYN_SENT: 1334 if ((thflags & TH_ACK) && 1335 (SEQ_LEQ(th->th_ack, tp->iss) || 1336 SEQ_GT(th->th_ack, tp->snd_max))) { 1337 rstreason = BANDLIM_UNLIMITED; 1338 goto dropwithreset; 1339 } 1340 if (thflags & TH_RST) { 1341 if (thflags & TH_ACK) { 1342 KASSERT(headlocked, ("tcp_input: after_listen" 1343 ": tcp_drop.2: head not locked")); 1344 tp = tcp_drop(tp, ECONNREFUSED); 1345 } 1346 goto drop; 1347 } 1348 if ((thflags & TH_SYN) == 0) 1349 goto drop; 1350 tp->snd_wnd = th->th_win; /* initial send window */ 1351 1352 tp->irs = th->th_seq; 1353 tcp_rcvseqinit(tp); 1354 if (thflags & TH_ACK) { 1355 tcpstat.tcps_connects++; 1356 soisconnected(so); 1357#ifdef MAC 1358 SOCK_LOCK(so); 1359 mac_set_socket_peer_from_mbuf(m, so); 1360 SOCK_UNLOCK(so); 1361#endif 1362 /* Do window scaling on this connection? */ 1363 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1364 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1365 tp->snd_scale = tp->requested_s_scale; 1366 tp->rcv_scale = tp->request_r_scale; 1367 } 1368 tp->rcv_adv += tp->rcv_wnd; 1369 tp->snd_una++; /* SYN is acked */ 1370 /* 1371 * If there's data, delay ACK; if there's also a FIN 1372 * ACKNOW will be turned on later. 1373 */ 1374 if (DELAY_ACK(tp) && tlen != 0) 1375 callout_reset(tp->tt_delack, tcp_delacktime, 1376 tcp_timer_delack, tp); 1377 else 1378 tp->t_flags |= TF_ACKNOW; 1379 /* 1380 * Received <SYN,ACK> in SYN_SENT[*] state. 1381 * Transitions: 1382 * SYN_SENT --> ESTABLISHED 1383 * SYN_SENT* --> FIN_WAIT_1 1384 */ 1385 tp->t_starttime = ticks; 1386 if (tp->t_flags & TF_NEEDFIN) { 1387 tp->t_state = TCPS_FIN_WAIT_1; 1388 tp->t_flags &= ~TF_NEEDFIN; 1389 thflags &= ~TH_SYN; 1390 } else { 1391 tp->t_state = TCPS_ESTABLISHED; 1392 callout_reset(tp->tt_keep, tcp_keepidle, 1393 tcp_timer_keep, tp); 1394 } 1395 } else { 1396 /* 1397 * Received initial SYN in SYN-SENT[*] state => 1398 * simultaneous open. If segment contains CC option 1399 * and there is a cached CC, apply TAO test. 1400 * If it succeeds, connection is * half-synchronized. 1401 * Otherwise, do 3-way handshake: 1402 * SYN-SENT -> SYN-RECEIVED 1403 * SYN-SENT* -> SYN-RECEIVED* 1404 * If there was no CC option, clear cached CC value. 1405 */ 1406 tp->t_flags |= TF_ACKNOW; 1407 callout_stop(tp->tt_rexmt); 1408 tp->t_state = TCPS_SYN_RECEIVED; 1409 } 1410 1411trimthenstep6: 1412 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1413 "locked")); 1414 INP_LOCK_ASSERT(inp); 1415 1416 /* 1417 * Advance th->th_seq to correspond to first data byte. 1418 * If data, trim to stay within window, 1419 * dropping FIN if necessary. 1420 */ 1421 th->th_seq++; 1422 if (tlen > tp->rcv_wnd) { 1423 todrop = tlen - tp->rcv_wnd; 1424 m_adj(m, -todrop); 1425 tlen = tp->rcv_wnd; 1426 thflags &= ~TH_FIN; 1427 tcpstat.tcps_rcvpackafterwin++; 1428 tcpstat.tcps_rcvbyteafterwin += todrop; 1429 } 1430 tp->snd_wl1 = th->th_seq - 1; 1431 tp->rcv_up = th->th_seq; 1432 /* 1433 * Client side of transaction: already sent SYN and data. 1434 * If the remote host used T/TCP to validate the SYN, 1435 * our data will be ACK'd; if so, enter normal data segment 1436 * processing in the middle of step 5, ack processing. 1437 * Otherwise, goto step 6. 1438 */ 1439 if (thflags & TH_ACK) 1440 goto process_ACK; 1441 1442 goto step6; 1443 1444 /* 1445 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1446 * do normal processing. 1447 * 1448 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1449 */ 1450 case TCPS_LAST_ACK: 1451 case TCPS_CLOSING: 1452 case TCPS_TIME_WAIT: 1453 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1454 break; /* continue normal processing */ 1455 } 1456 1457 /* 1458 * States other than LISTEN or SYN_SENT. 1459 * First check the RST flag and sequence number since reset segments 1460 * are exempt from the timestamp and connection count tests. This 1461 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1462 * below which allowed reset segments in half the sequence space 1463 * to fall though and be processed (which gives forged reset 1464 * segments with a random sequence number a 50 percent chance of 1465 * killing a connection). 1466 * Then check timestamp, if present. 1467 * Then check the connection count, if present. 1468 * Then check that at least some bytes of segment are within 1469 * receive window. If segment begins before rcv_nxt, 1470 * drop leading data (and SYN); if nothing left, just ack. 1471 * 1472 * 1473 * If the RST bit is set, check the sequence number to see 1474 * if this is a valid reset segment. 1475 * RFC 793 page 37: 1476 * In all states except SYN-SENT, all reset (RST) segments 1477 * are validated by checking their SEQ-fields. A reset is 1478 * valid if its sequence number is in the window. 1479 * Note: this does not take into account delayed ACKs, so 1480 * we should test against last_ack_sent instead of rcv_nxt. 1481 * The sequence number in the reset segment is normally an 1482 * echo of our outgoing acknowlegement numbers, but some hosts 1483 * send a reset with the sequence number at the rightmost edge 1484 * of our receive window, and we have to handle this case. 1485 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1486 * that brute force RST attacks are possible. To combat this, 1487 * we use a much stricter check while in the ESTABLISHED state, 1488 * only accepting RSTs where the sequence number is equal to 1489 * last_ack_sent. In all other states (the states in which a 1490 * RST is more likely), the more permissive check is used. 1491 * If we have multiple segments in flight, the intial reset 1492 * segment sequence numbers will be to the left of last_ack_sent, 1493 * but they will eventually catch up. 1494 * In any case, it never made sense to trim reset segments to 1495 * fit the receive window since RFC 1122 says: 1496 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1497 * 1498 * A TCP SHOULD allow a received RST segment to include data. 1499 * 1500 * DISCUSSION 1501 * It has been suggested that a RST segment could contain 1502 * ASCII text that encoded and explained the cause of the 1503 * RST. No standard has yet been established for such 1504 * data. 1505 * 1506 * If the reset segment passes the sequence number test examine 1507 * the state: 1508 * SYN_RECEIVED STATE: 1509 * If passive open, return to LISTEN state. 1510 * If active open, inform user that connection was refused. 1511 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1512 * Inform user that connection was reset, and close tcb. 1513 * CLOSING, LAST_ACK STATES: 1514 * Close the tcb. 1515 * TIME_WAIT STATE: 1516 * Drop the segment - see Stevens, vol. 2, p. 964 and 1517 * RFC 1337. 1518 */ 1519 if (thflags & TH_RST) { 1520 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1521 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1522 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1523 switch (tp->t_state) { 1524 1525 case TCPS_SYN_RECEIVED: 1526 so->so_error = ECONNREFUSED; 1527 goto close; 1528 1529 case TCPS_ESTABLISHED: 1530 if (tp->last_ack_sent != th->th_seq && 1531 tcp_insecure_rst == 0) { 1532 tcpstat.tcps_badrst++; 1533 goto drop; 1534 } 1535 case TCPS_FIN_WAIT_1: 1536 case TCPS_FIN_WAIT_2: 1537 case TCPS_CLOSE_WAIT: 1538 so->so_error = ECONNRESET; 1539 close: 1540 tp->t_state = TCPS_CLOSED; 1541 tcpstat.tcps_drops++; 1542 KASSERT(headlocked, ("tcp_input: " 1543 "trimthenstep6: tcp_close: head not " 1544 "locked")); 1545 tp = tcp_close(tp); 1546 break; 1547 1548 case TCPS_CLOSING: 1549 case TCPS_LAST_ACK: 1550 KASSERT(headlocked, ("trimthenstep6: " 1551 "tcp_close.2: head not locked")); 1552 tp = tcp_close(tp); 1553 break; 1554 1555 case TCPS_TIME_WAIT: 1556 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1557 ("timewait")); 1558 break; 1559 } 1560 } 1561 goto drop; 1562 } 1563 1564 /* 1565 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1566 * and it's less than ts_recent, drop it. 1567 */ 1568 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1569 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1570 1571 /* Check to see if ts_recent is over 24 days old. */ 1572 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1573 /* 1574 * Invalidate ts_recent. If this segment updates 1575 * ts_recent, the age will be reset later and ts_recent 1576 * will get a valid value. If it does not, setting 1577 * ts_recent to zero will at least satisfy the 1578 * requirement that zero be placed in the timestamp 1579 * echo reply when ts_recent isn't valid. The 1580 * age isn't reset until we get a valid ts_recent 1581 * because we don't want out-of-order segments to be 1582 * dropped when ts_recent is old. 1583 */ 1584 tp->ts_recent = 0; 1585 } else { 1586 tcpstat.tcps_rcvduppack++; 1587 tcpstat.tcps_rcvdupbyte += tlen; 1588 tcpstat.tcps_pawsdrop++; 1589 if (tlen) 1590 goto dropafterack; 1591 goto drop; 1592 } 1593 } 1594 1595 /* 1596 * In the SYN-RECEIVED state, validate that the packet belongs to 1597 * this connection before trimming the data to fit the receive 1598 * window. Check the sequence number versus IRS since we know 1599 * the sequence numbers haven't wrapped. This is a partial fix 1600 * for the "LAND" DoS attack. 1601 */ 1602 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1603 rstreason = BANDLIM_RST_OPENPORT; 1604 goto dropwithreset; 1605 } 1606 1607 todrop = tp->rcv_nxt - th->th_seq; 1608 if (todrop > 0) { 1609 if (thflags & TH_SYN) { 1610 thflags &= ~TH_SYN; 1611 th->th_seq++; 1612 if (th->th_urp > 1) 1613 th->th_urp--; 1614 else 1615 thflags &= ~TH_URG; 1616 todrop--; 1617 } 1618 /* 1619 * Following if statement from Stevens, vol. 2, p. 960. 1620 */ 1621 if (todrop > tlen 1622 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1623 /* 1624 * Any valid FIN must be to the left of the window. 1625 * At this point the FIN must be a duplicate or out 1626 * of sequence; drop it. 1627 */ 1628 thflags &= ~TH_FIN; 1629 1630 /* 1631 * Send an ACK to resynchronize and drop any data. 1632 * But keep on processing for RST or ACK. 1633 */ 1634 tp->t_flags |= TF_ACKNOW; 1635 todrop = tlen; 1636 tcpstat.tcps_rcvduppack++; 1637 tcpstat.tcps_rcvdupbyte += todrop; 1638 } else { 1639 tcpstat.tcps_rcvpartduppack++; 1640 tcpstat.tcps_rcvpartdupbyte += todrop; 1641 } 1642 drop_hdrlen += todrop; /* drop from the top afterwards */ 1643 th->th_seq += todrop; 1644 tlen -= todrop; 1645 if (th->th_urp > todrop) 1646 th->th_urp -= todrop; 1647 else { 1648 thflags &= ~TH_URG; 1649 th->th_urp = 0; 1650 } 1651 } 1652 1653 /* 1654 * If new data are received on a connection after the 1655 * user processes are gone, then RST the other end. 1656 */ 1657 if ((so->so_state & SS_NOFDREF) && 1658 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1659 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not " 1660 "locked")); 1661 tp = tcp_close(tp); 1662 tcpstat.tcps_rcvafterclose++; 1663 rstreason = BANDLIM_UNLIMITED; 1664 goto dropwithreset; 1665 } 1666 1667 /* 1668 * If segment ends after window, drop trailing data 1669 * (and PUSH and FIN); if nothing left, just ACK. 1670 */ 1671 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1672 if (todrop > 0) { 1673 tcpstat.tcps_rcvpackafterwin++; 1674 if (todrop >= tlen) { 1675 tcpstat.tcps_rcvbyteafterwin += tlen; 1676 /* 1677 * If a new connection request is received 1678 * while in TIME_WAIT, drop the old connection 1679 * and start over if the sequence numbers 1680 * are above the previous ones. 1681 */ 1682 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1683 if (thflags & TH_SYN && 1684 tp->t_state == TCPS_TIME_WAIT && 1685 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1686 KASSERT(headlocked, ("trimthenstep6: " 1687 "tcp_close.4: head not locked")); 1688 tp = tcp_close(tp); 1689 goto findpcb; 1690 } 1691 /* 1692 * If window is closed can only take segments at 1693 * window edge, and have to drop data and PUSH from 1694 * incoming segments. Continue processing, but 1695 * remember to ack. Otherwise, drop segment 1696 * and ack. 1697 */ 1698 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1699 tp->t_flags |= TF_ACKNOW; 1700 tcpstat.tcps_rcvwinprobe++; 1701 } else 1702 goto dropafterack; 1703 } else 1704 tcpstat.tcps_rcvbyteafterwin += todrop; 1705 m_adj(m, -todrop); 1706 tlen -= todrop; 1707 thflags &= ~(TH_PUSH|TH_FIN); 1708 } 1709 1710 /* 1711 * If last ACK falls within this segment's sequence numbers, 1712 * record its timestamp. 1713 * NOTE: 1714 * 1) That the test incorporates suggestions from the latest 1715 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1716 * 2) That updating only on newer timestamps interferes with 1717 * our earlier PAWS tests, so this check should be solely 1718 * predicated on the sequence space of this segment. 1719 * 3) That we modify the segment boundary check to be 1720 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1721 * instead of RFC1323's 1722 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1723 * This modified check allows us to overcome RFC1323's 1724 * limitations as described in Stevens TCP/IP Illustrated 1725 * Vol. 2 p.869. In such cases, we can still calculate the 1726 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1727 */ 1728 if ((to.to_flags & TOF_TS) != 0 && 1729 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1730 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1731 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1732 tp->ts_recent_age = ticks; 1733 tp->ts_recent = to.to_tsval; 1734 } 1735 1736 /* 1737 * If a SYN is in the window, then this is an 1738 * error and we send an RST and drop the connection. 1739 */ 1740 if (thflags & TH_SYN) { 1741 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: " 1742 "head not locked")); 1743 tp = tcp_drop(tp, ECONNRESET); 1744 rstreason = BANDLIM_UNLIMITED; 1745 goto drop; 1746 } 1747 1748 /* 1749 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1750 * flag is on (half-synchronized state), then queue data for 1751 * later processing; else drop segment and return. 1752 */ 1753 if ((thflags & TH_ACK) == 0) { 1754 if (tp->t_state == TCPS_SYN_RECEIVED || 1755 (tp->t_flags & TF_NEEDSYN)) 1756 goto step6; 1757 else 1758 goto drop; 1759 } 1760 1761 /* 1762 * Ack processing. 1763 */ 1764 switch (tp->t_state) { 1765 1766 /* 1767 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1768 * ESTABLISHED state and continue processing. 1769 * The ACK was checked above. 1770 */ 1771 case TCPS_SYN_RECEIVED: 1772 1773 tcpstat.tcps_connects++; 1774 soisconnected(so); 1775 /* Do window scaling? */ 1776 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1777 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1778 tp->snd_scale = tp->requested_s_scale; 1779 tp->rcv_scale = tp->request_r_scale; 1780 } 1781 /* 1782 * Make transitions: 1783 * SYN-RECEIVED -> ESTABLISHED 1784 * SYN-RECEIVED* -> FIN-WAIT-1 1785 */ 1786 tp->t_starttime = ticks; 1787 if (tp->t_flags & TF_NEEDFIN) { 1788 tp->t_state = TCPS_FIN_WAIT_1; 1789 tp->t_flags &= ~TF_NEEDFIN; 1790 } else { 1791 tp->t_state = TCPS_ESTABLISHED; 1792 callout_reset(tp->tt_keep, tcp_keepidle, 1793 tcp_timer_keep, tp); 1794 } 1795 /* 1796 * If segment contains data or ACK, will call tcp_reass() 1797 * later; if not, do so now to pass queued data to user. 1798 */ 1799 if (tlen == 0 && (thflags & TH_FIN) == 0) 1800 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1801 (struct mbuf *)0); 1802 tp->snd_wl1 = th->th_seq - 1; 1803 /* FALLTHROUGH */ 1804 1805 /* 1806 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1807 * ACKs. If the ack is in the range 1808 * tp->snd_una < th->th_ack <= tp->snd_max 1809 * then advance tp->snd_una to th->th_ack and drop 1810 * data from the retransmission queue. If this ACK reflects 1811 * more up to date window information we update our window information. 1812 */ 1813 case TCPS_ESTABLISHED: 1814 case TCPS_FIN_WAIT_1: 1815 case TCPS_FIN_WAIT_2: 1816 case TCPS_CLOSE_WAIT: 1817 case TCPS_CLOSING: 1818 case TCPS_LAST_ACK: 1819 case TCPS_TIME_WAIT: 1820 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1821 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1822 tcpstat.tcps_rcvacktoomuch++; 1823 goto dropafterack; 1824 } 1825 if (tp->sack_enable) 1826 tcp_sack_doack(tp, &to, th->th_ack); 1827 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1828 if (tlen == 0 && tiwin == tp->snd_wnd) { 1829 tcpstat.tcps_rcvdupack++; 1830 /* 1831 * If we have outstanding data (other than 1832 * a window probe), this is a completely 1833 * duplicate ack (ie, window info didn't 1834 * change), the ack is the biggest we've 1835 * seen and we've seen exactly our rexmt 1836 * threshhold of them, assume a packet 1837 * has been dropped and retransmit it. 1838 * Kludge snd_nxt & the congestion 1839 * window so we send only this one 1840 * packet. 1841 * 1842 * We know we're losing at the current 1843 * window size so do congestion avoidance 1844 * (set ssthresh to half the current window 1845 * and pull our congestion window back to 1846 * the new ssthresh). 1847 * 1848 * Dup acks mean that packets have left the 1849 * network (they're now cached at the receiver) 1850 * so bump cwnd by the amount in the receiver 1851 * to keep a constant cwnd packets in the 1852 * network. 1853 */ 1854 if (!callout_active(tp->tt_rexmt) || 1855 th->th_ack != tp->snd_una) 1856 tp->t_dupacks = 0; 1857 else if (++tp->t_dupacks > tcprexmtthresh || 1858 ((tcp_do_newreno || tp->sack_enable) && 1859 IN_FASTRECOVERY(tp))) { 1860 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1861 int awnd; 1862 1863 /* 1864 * Compute the amount of data in flight first. 1865 * We can inject new data into the pipe iff 1866 * we have less than 1/2 the original window's 1867 * worth of data in flight. 1868 */ 1869 awnd = (tp->snd_nxt - tp->snd_fack) + 1870 tp->sackhint.sack_bytes_rexmit; 1871 if (awnd < tp->snd_ssthresh) { 1872 tp->snd_cwnd += tp->t_maxseg; 1873 if (tp->snd_cwnd > tp->snd_ssthresh) 1874 tp->snd_cwnd = tp->snd_ssthresh; 1875 } 1876 } else 1877 tp->snd_cwnd += tp->t_maxseg; 1878 (void) tcp_output(tp); 1879 goto drop; 1880 } else if (tp->t_dupacks == tcprexmtthresh) { 1881 tcp_seq onxt = tp->snd_nxt; 1882 u_int win; 1883 1884 /* 1885 * If we're doing sack, check to 1886 * see if we're already in sack 1887 * recovery. If we're not doing sack, 1888 * check to see if we're in newreno 1889 * recovery. 1890 */ 1891 if (tp->sack_enable) { 1892 if (IN_FASTRECOVERY(tp)) { 1893 tp->t_dupacks = 0; 1894 break; 1895 } 1896 } else if (tcp_do_newreno) { 1897 if (SEQ_LEQ(th->th_ack, 1898 tp->snd_recover)) { 1899 tp->t_dupacks = 0; 1900 break; 1901 } 1902 } 1903 win = min(tp->snd_wnd, tp->snd_cwnd) / 1904 2 / tp->t_maxseg; 1905 if (win < 2) 1906 win = 2; 1907 tp->snd_ssthresh = win * tp->t_maxseg; 1908 ENTER_FASTRECOVERY(tp); 1909 tp->snd_recover = tp->snd_max; 1910 callout_stop(tp->tt_rexmt); 1911 tp->t_rtttime = 0; 1912 if (tp->sack_enable) { 1913 KASSERT(tp->sackhint. 1914 sack_bytes_rexmit == 0, 1915 ("sackhint rexmit == 0")); 1916 tcpstat.tcps_sack_recovery_episode++; 1917 tp->sack_newdata = tp->snd_nxt; 1918 tp->snd_cwnd = tp->t_maxseg; 1919 (void) tcp_output(tp); 1920 goto drop; 1921 } 1922 tp->snd_nxt = th->th_ack; 1923 tp->snd_cwnd = tp->t_maxseg; 1924 (void) tcp_output(tp); 1925 KASSERT(tp->snd_limited <= 2, 1926 ("tp->snd_limited too big")); 1927 tp->snd_cwnd = tp->snd_ssthresh + 1928 tp->t_maxseg * 1929 (tp->t_dupacks - tp->snd_limited); 1930 if (SEQ_GT(onxt, tp->snd_nxt)) 1931 tp->snd_nxt = onxt; 1932 goto drop; 1933 } else if (tcp_do_rfc3042) { 1934 u_long oldcwnd = tp->snd_cwnd; 1935 tcp_seq oldsndmax = tp->snd_max; 1936 u_int sent; 1937 1938 KASSERT(tp->t_dupacks == 1 || 1939 tp->t_dupacks == 2, 1940 ("dupacks not 1 or 2")); 1941 if (tp->t_dupacks == 1) 1942 tp->snd_limited = 0; 1943 tp->snd_cwnd = 1944 (tp->snd_nxt - tp->snd_una) + 1945 (tp->t_dupacks - tp->snd_limited) * 1946 tp->t_maxseg; 1947 (void) tcp_output(tp); 1948 sent = tp->snd_max - oldsndmax; 1949 if (sent > tp->t_maxseg) { 1950 KASSERT((tp->t_dupacks == 2 && 1951 tp->snd_limited == 0) || 1952 (sent == tp->t_maxseg + 1 && 1953 tp->t_flags & TF_SENTFIN), 1954 ("sent too much")); 1955 tp->snd_limited = 2; 1956 } else if (sent > 0) 1957 ++tp->snd_limited; 1958 tp->snd_cwnd = oldcwnd; 1959 goto drop; 1960 } 1961 } else 1962 tp->t_dupacks = 0; 1963 break; 1964 } 1965 1966 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1967 1968 /* 1969 * If the congestion window was inflated to account 1970 * for the other side's cached packets, retract it. 1971 */ 1972 if (tcp_do_newreno || tp->sack_enable) { 1973 if (IN_FASTRECOVERY(tp)) { 1974 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1975 if (tp->sack_enable) 1976 tcp_sack_partialack(tp, th); 1977 else 1978 tcp_newreno_partial_ack(tp, th); 1979 } else { 1980 /* 1981 * Out of fast recovery. 1982 * Window inflation should have left us 1983 * with approximately snd_ssthresh 1984 * outstanding data. 1985 * But in case we would be inclined to 1986 * send a burst, better to do it via 1987 * the slow start mechanism. 1988 */ 1989 if (SEQ_GT(th->th_ack + 1990 tp->snd_ssthresh, 1991 tp->snd_max)) 1992 tp->snd_cwnd = tp->snd_max - 1993 th->th_ack + 1994 tp->t_maxseg; 1995 else 1996 tp->snd_cwnd = tp->snd_ssthresh; 1997 } 1998 } 1999 } else { 2000 if (tp->t_dupacks >= tcprexmtthresh && 2001 tp->snd_cwnd > tp->snd_ssthresh) 2002 tp->snd_cwnd = tp->snd_ssthresh; 2003 } 2004 tp->t_dupacks = 0; 2005 /* 2006 * If we reach this point, ACK is not a duplicate, 2007 * i.e., it ACKs something we sent. 2008 */ 2009 if (tp->t_flags & TF_NEEDSYN) { 2010 /* 2011 * T/TCP: Connection was half-synchronized, and our 2012 * SYN has been ACK'd (so connection is now fully 2013 * synchronized). Go to non-starred state, 2014 * increment snd_una for ACK of SYN, and check if 2015 * we can do window scaling. 2016 */ 2017 tp->t_flags &= ~TF_NEEDSYN; 2018 tp->snd_una++; 2019 /* Do window scaling? */ 2020 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2021 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2022 tp->snd_scale = tp->requested_s_scale; 2023 tp->rcv_scale = tp->request_r_scale; 2024 } 2025 } 2026 2027process_ACK: 2028 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 2029 "locked")); 2030 INP_LOCK_ASSERT(inp); 2031 2032 acked = th->th_ack - tp->snd_una; 2033 tcpstat.tcps_rcvackpack++; 2034 tcpstat.tcps_rcvackbyte += acked; 2035 2036 /* 2037 * If we just performed our first retransmit, and the ACK 2038 * arrives within our recovery window, then it was a mistake 2039 * to do the retransmit in the first place. Recover our 2040 * original cwnd and ssthresh, and proceed to transmit where 2041 * we left off. 2042 */ 2043 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2044 ++tcpstat.tcps_sndrexmitbad; 2045 tp->snd_cwnd = tp->snd_cwnd_prev; 2046 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2047 tp->snd_recover = tp->snd_recover_prev; 2048 if (tp->t_flags & TF_WASFRECOVERY) 2049 ENTER_FASTRECOVERY(tp); 2050 tp->snd_nxt = tp->snd_max; 2051 tp->t_badrxtwin = 0; /* XXX probably not required */ 2052 } 2053 2054 /* 2055 * If we have a timestamp reply, update smoothed 2056 * round trip time. If no timestamp is present but 2057 * transmit timer is running and timed sequence 2058 * number was acked, update smoothed round trip time. 2059 * Since we now have an rtt measurement, cancel the 2060 * timer backoff (cf., Phil Karn's retransmit alg.). 2061 * Recompute the initial retransmit timer. 2062 * 2063 * Some boxes send broken timestamp replies 2064 * during the SYN+ACK phase, ignore 2065 * timestamps of 0 or we could calculate a 2066 * huge RTT and blow up the retransmit timer. 2067 */ 2068 if ((to.to_flags & TOF_TS) != 0 && 2069 to.to_tsecr) { 2070 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2071 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2072 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2073 } 2074 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2075 2076 /* 2077 * If all outstanding data is acked, stop retransmit 2078 * timer and remember to restart (more output or persist). 2079 * If there is more data to be acked, restart retransmit 2080 * timer, using current (possibly backed-off) value. 2081 */ 2082 if (th->th_ack == tp->snd_max) { 2083 callout_stop(tp->tt_rexmt); 2084 needoutput = 1; 2085 } else if (!callout_active(tp->tt_persist)) 2086 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2087 tcp_timer_rexmt, tp); 2088 2089 /* 2090 * If no data (only SYN) was ACK'd, 2091 * skip rest of ACK processing. 2092 */ 2093 if (acked == 0) 2094 goto step6; 2095 2096 /* 2097 * When new data is acked, open the congestion window. 2098 * If the window gives us less than ssthresh packets 2099 * in flight, open exponentially (maxseg per packet). 2100 * Otherwise open linearly: maxseg per window 2101 * (maxseg^2 / cwnd per packet). 2102 */ 2103 if ((!tcp_do_newreno && !tp->sack_enable) || 2104 !IN_FASTRECOVERY(tp)) { 2105 register u_int cw = tp->snd_cwnd; 2106 register u_int incr = tp->t_maxseg; 2107 if (cw > tp->snd_ssthresh) 2108 incr = incr * incr / cw; 2109 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2110 } 2111 SOCKBUF_LOCK(&so->so_snd); 2112 if (acked > so->so_snd.sb_cc) { 2113 tp->snd_wnd -= so->so_snd.sb_cc; 2114 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2115 ourfinisacked = 1; 2116 } else { 2117 sbdrop_locked(&so->so_snd, acked); 2118 tp->snd_wnd -= acked; 2119 ourfinisacked = 0; 2120 } 2121 sowwakeup_locked(so); 2122 /* detect una wraparound */ 2123 if ((tcp_do_newreno || tp->sack_enable) && 2124 !IN_FASTRECOVERY(tp) && 2125 SEQ_GT(tp->snd_una, tp->snd_recover) && 2126 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2127 tp->snd_recover = th->th_ack - 1; 2128 if ((tcp_do_newreno || tp->sack_enable) && 2129 IN_FASTRECOVERY(tp) && 2130 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2131 EXIT_FASTRECOVERY(tp); 2132 tp->snd_una = th->th_ack; 2133 if (tp->sack_enable) { 2134 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2135 tp->snd_recover = tp->snd_una; 2136 } 2137 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2138 tp->snd_nxt = tp->snd_una; 2139 2140 switch (tp->t_state) { 2141 2142 /* 2143 * In FIN_WAIT_1 STATE in addition to the processing 2144 * for the ESTABLISHED state if our FIN is now acknowledged 2145 * then enter FIN_WAIT_2. 2146 */ 2147 case TCPS_FIN_WAIT_1: 2148 if (ourfinisacked) { 2149 /* 2150 * If we can't receive any more 2151 * data, then closing user can proceed. 2152 * Starting the timer is contrary to the 2153 * specification, but if we don't get a FIN 2154 * we'll hang forever. 2155 */ 2156 /* XXXjl 2157 * we should release the tp also, and use a 2158 * compressed state. 2159 */ 2160 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2161 soisdisconnected(so); 2162 callout_reset(tp->tt_2msl, tcp_maxidle, 2163 tcp_timer_2msl, tp); 2164 } 2165 tp->t_state = TCPS_FIN_WAIT_2; 2166 } 2167 break; 2168 2169 /* 2170 * In CLOSING STATE in addition to the processing for 2171 * the ESTABLISHED state if the ACK acknowledges our FIN 2172 * then enter the TIME-WAIT state, otherwise ignore 2173 * the segment. 2174 */ 2175 case TCPS_CLOSING: 2176 if (ourfinisacked) { 2177 KASSERT(headlocked, ("tcp_input: process_ACK: " 2178 "head not locked")); 2179 tcp_twstart(tp); 2180 INP_INFO_WUNLOCK(&tcbinfo); 2181 m_freem(m); 2182 return; 2183 } 2184 break; 2185 2186 /* 2187 * In LAST_ACK, we may still be waiting for data to drain 2188 * and/or to be acked, as well as for the ack of our FIN. 2189 * If our FIN is now acknowledged, delete the TCB, 2190 * enter the closed state and return. 2191 */ 2192 case TCPS_LAST_ACK: 2193 if (ourfinisacked) { 2194 KASSERT(headlocked, ("tcp_input: process_ACK:" 2195 " tcp_close: head not locked")); 2196 tp = tcp_close(tp); 2197 goto drop; 2198 } 2199 break; 2200 2201 /* 2202 * In TIME_WAIT state the only thing that should arrive 2203 * is a retransmission of the remote FIN. Acknowledge 2204 * it and restart the finack timer. 2205 */ 2206 case TCPS_TIME_WAIT: 2207 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2208 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2209 tcp_timer_2msl, tp); 2210 goto dropafterack; 2211 } 2212 } 2213 2214step6: 2215 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2216 INP_LOCK_ASSERT(inp); 2217 2218 /* 2219 * Update window information. 2220 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2221 */ 2222 if ((thflags & TH_ACK) && 2223 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2224 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2225 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2226 /* keep track of pure window updates */ 2227 if (tlen == 0 && 2228 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2229 tcpstat.tcps_rcvwinupd++; 2230 tp->snd_wnd = tiwin; 2231 tp->snd_wl1 = th->th_seq; 2232 tp->snd_wl2 = th->th_ack; 2233 if (tp->snd_wnd > tp->max_sndwnd) 2234 tp->max_sndwnd = tp->snd_wnd; 2235 needoutput = 1; 2236 } 2237 2238 /* 2239 * Process segments with URG. 2240 */ 2241 if ((thflags & TH_URG) && th->th_urp && 2242 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2243 /* 2244 * This is a kludge, but if we receive and accept 2245 * random urgent pointers, we'll crash in 2246 * soreceive. It's hard to imagine someone 2247 * actually wanting to send this much urgent data. 2248 */ 2249 SOCKBUF_LOCK(&so->so_rcv); 2250 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2251 th->th_urp = 0; /* XXX */ 2252 thflags &= ~TH_URG; /* XXX */ 2253 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2254 goto dodata; /* XXX */ 2255 } 2256 /* 2257 * If this segment advances the known urgent pointer, 2258 * then mark the data stream. This should not happen 2259 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2260 * a FIN has been received from the remote side. 2261 * In these states we ignore the URG. 2262 * 2263 * According to RFC961 (Assigned Protocols), 2264 * the urgent pointer points to the last octet 2265 * of urgent data. We continue, however, 2266 * to consider it to indicate the first octet 2267 * of data past the urgent section as the original 2268 * spec states (in one of two places). 2269 */ 2270 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2271 tp->rcv_up = th->th_seq + th->th_urp; 2272 so->so_oobmark = so->so_rcv.sb_cc + 2273 (tp->rcv_up - tp->rcv_nxt) - 1; 2274 if (so->so_oobmark == 0) 2275 so->so_rcv.sb_state |= SBS_RCVATMARK; 2276 sohasoutofband(so); 2277 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2278 } 2279 SOCKBUF_UNLOCK(&so->so_rcv); 2280 /* 2281 * Remove out of band data so doesn't get presented to user. 2282 * This can happen independent of advancing the URG pointer, 2283 * but if two URG's are pending at once, some out-of-band 2284 * data may creep in... ick. 2285 */ 2286 if (th->th_urp <= (u_long)tlen && 2287 !(so->so_options & SO_OOBINLINE)) { 2288 /* hdr drop is delayed */ 2289 tcp_pulloutofband(so, th, m, drop_hdrlen); 2290 } 2291 } else { 2292 /* 2293 * If no out of band data is expected, 2294 * pull receive urgent pointer along 2295 * with the receive window. 2296 */ 2297 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2298 tp->rcv_up = tp->rcv_nxt; 2299 } 2300dodata: /* XXX */ 2301 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2302 INP_LOCK_ASSERT(inp); 2303 2304 /* 2305 * Process the segment text, merging it into the TCP sequencing queue, 2306 * and arranging for acknowledgment of receipt if necessary. 2307 * This process logically involves adjusting tp->rcv_wnd as data 2308 * is presented to the user (this happens in tcp_usrreq.c, 2309 * case PRU_RCVD). If a FIN has already been received on this 2310 * connection then we just ignore the text. 2311 */ 2312 if ((tlen || (thflags & TH_FIN)) && 2313 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2314 m_adj(m, drop_hdrlen); /* delayed header drop */ 2315 /* 2316 * Insert segment which includes th into TCP reassembly queue 2317 * with control block tp. Set thflags to whether reassembly now 2318 * includes a segment with FIN. This handles the common case 2319 * inline (segment is the next to be received on an established 2320 * connection, and the queue is empty), avoiding linkage into 2321 * and removal from the queue and repetition of various 2322 * conversions. 2323 * Set DELACK for segments received in order, but ack 2324 * immediately when segments are out of order (so 2325 * fast retransmit can work). 2326 */ 2327 if (th->th_seq == tp->rcv_nxt && 2328 LIST_EMPTY(&tp->t_segq) && 2329 TCPS_HAVEESTABLISHED(tp->t_state)) { 2330 if (DELAY_ACK(tp)) 2331 tp->t_flags |= TF_DELACK; 2332 else 2333 tp->t_flags |= TF_ACKNOW; 2334 tp->rcv_nxt += tlen; 2335 thflags = th->th_flags & TH_FIN; 2336 tcpstat.tcps_rcvpack++; 2337 tcpstat.tcps_rcvbyte += tlen; 2338 ND6_HINT(tp); 2339 SOCKBUF_LOCK(&so->so_rcv); 2340 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2341 m_freem(m); 2342 else 2343 sbappendstream_locked(&so->so_rcv, m); 2344 sorwakeup_locked(so); 2345 } else { 2346 thflags = tcp_reass(tp, th, &tlen, m); 2347 tp->t_flags |= TF_ACKNOW; 2348 } 2349 if (tlen > 0 && tp->sack_enable) 2350 tcp_update_sack_list(tp, th->th_seq, th->th_seq + tlen); 2351 /* 2352 * Note the amount of data that peer has sent into 2353 * our window, in order to estimate the sender's 2354 * buffer size. 2355 */ 2356 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2357 } else { 2358 m_freem(m); 2359 thflags &= ~TH_FIN; 2360 } 2361 2362 /* 2363 * If FIN is received ACK the FIN and let the user know 2364 * that the connection is closing. 2365 */ 2366 if (thflags & TH_FIN) { 2367 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2368 socantrcvmore(so); 2369 /* 2370 * If connection is half-synchronized 2371 * (ie NEEDSYN flag on) then delay ACK, 2372 * so it may be piggybacked when SYN is sent. 2373 * Otherwise, since we received a FIN then no 2374 * more input can be expected, send ACK now. 2375 */ 2376 if (tp->t_flags & TF_NEEDSYN) 2377 tp->t_flags |= TF_DELACK; 2378 else 2379 tp->t_flags |= TF_ACKNOW; 2380 tp->rcv_nxt++; 2381 } 2382 switch (tp->t_state) { 2383 2384 /* 2385 * In SYN_RECEIVED and ESTABLISHED STATES 2386 * enter the CLOSE_WAIT state. 2387 */ 2388 case TCPS_SYN_RECEIVED: 2389 tp->t_starttime = ticks; 2390 /*FALLTHROUGH*/ 2391 case TCPS_ESTABLISHED: 2392 tp->t_state = TCPS_CLOSE_WAIT; 2393 break; 2394 2395 /* 2396 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2397 * enter the CLOSING state. 2398 */ 2399 case TCPS_FIN_WAIT_1: 2400 tp->t_state = TCPS_CLOSING; 2401 break; 2402 2403 /* 2404 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2405 * starting the time-wait timer, turning off the other 2406 * standard timers. 2407 */ 2408 case TCPS_FIN_WAIT_2: 2409 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2410 "TCP_FIN_WAIT_2: head not locked")); 2411 tcp_twstart(tp); 2412 INP_INFO_WUNLOCK(&tcbinfo); 2413 return; 2414 2415 /* 2416 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2417 */ 2418 case TCPS_TIME_WAIT: 2419 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2420 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2421 tcp_timer_2msl, tp); 2422 break; 2423 } 2424 } 2425 INP_INFO_WUNLOCK(&tcbinfo); 2426 headlocked = 0; 2427#ifdef TCPDEBUG 2428 if (so->so_options & SO_DEBUG) 2429 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2430 &tcp_savetcp, 0); 2431#endif 2432 2433 /* 2434 * Return any desired output. 2435 */ 2436 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2437 (void) tcp_output(tp); 2438 2439check_delack: 2440 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2441 INP_LOCK_ASSERT(inp); 2442 if (tp->t_flags & TF_DELACK) { 2443 tp->t_flags &= ~TF_DELACK; 2444 callout_reset(tp->tt_delack, tcp_delacktime, 2445 tcp_timer_delack, tp); 2446 } 2447 INP_UNLOCK(inp); 2448 return; 2449 2450dropafterack: 2451 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2452 /* 2453 * Generate an ACK dropping incoming segment if it occupies 2454 * sequence space, where the ACK reflects our state. 2455 * 2456 * We can now skip the test for the RST flag since all 2457 * paths to this code happen after packets containing 2458 * RST have been dropped. 2459 * 2460 * In the SYN-RECEIVED state, don't send an ACK unless the 2461 * segment we received passes the SYN-RECEIVED ACK test. 2462 * If it fails send a RST. This breaks the loop in the 2463 * "LAND" DoS attack, and also prevents an ACK storm 2464 * between two listening ports that have been sent forged 2465 * SYN segments, each with the source address of the other. 2466 */ 2467 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2468 (SEQ_GT(tp->snd_una, th->th_ack) || 2469 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2470 rstreason = BANDLIM_RST_OPENPORT; 2471 goto dropwithreset; 2472 } 2473#ifdef TCPDEBUG 2474 if (so->so_options & SO_DEBUG) 2475 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2476 &tcp_savetcp, 0); 2477#endif 2478 KASSERT(headlocked, ("headlocked should be 1")); 2479 INP_INFO_WUNLOCK(&tcbinfo); 2480 tp->t_flags |= TF_ACKNOW; 2481 (void) tcp_output(tp); 2482 INP_UNLOCK(inp); 2483 m_freem(m); 2484 return; 2485 2486dropwithreset: 2487 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2488 /* 2489 * Generate a RST, dropping incoming segment. 2490 * Make ACK acceptable to originator of segment. 2491 * Don't bother to respond if destination was broadcast/multicast. 2492 */ 2493 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2494 goto drop; 2495 if (isipv6) { 2496 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2497 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2498 goto drop; 2499 } else { 2500 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2501 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2502 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2503 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2504 goto drop; 2505 } 2506 /* IPv6 anycast check is done at tcp6_input() */ 2507 2508 /* 2509 * Perform bandwidth limiting. 2510 */ 2511 if (badport_bandlim(rstreason) < 0) 2512 goto drop; 2513 2514#ifdef TCPDEBUG 2515 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2516 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2517 &tcp_savetcp, 0); 2518#endif 2519 2520 if (thflags & TH_ACK) 2521 /* mtod() below is safe as long as hdr dropping is delayed */ 2522 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2523 TH_RST); 2524 else { 2525 if (thflags & TH_SYN) 2526 tlen++; 2527 /* mtod() below is safe as long as hdr dropping is delayed */ 2528 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2529 (tcp_seq)0, TH_RST|TH_ACK); 2530 } 2531 2532 if (tp) 2533 INP_UNLOCK(inp); 2534 if (headlocked) 2535 INP_INFO_WUNLOCK(&tcbinfo); 2536 return; 2537 2538drop: 2539 /* 2540 * Drop space held by incoming segment and return. 2541 */ 2542#ifdef TCPDEBUG 2543 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2544 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2545 &tcp_savetcp, 0); 2546#endif 2547 if (tp) 2548 INP_UNLOCK(inp); 2549 if (headlocked) 2550 INP_INFO_WUNLOCK(&tcbinfo); 2551 m_freem(m); 2552 return; 2553} 2554 2555/* 2556 * Parse TCP options and place in tcpopt. 2557 */ 2558static void 2559tcp_dooptions(to, cp, cnt, is_syn) 2560 struct tcpopt *to; 2561 u_char *cp; 2562 int cnt; 2563 int is_syn; 2564{ 2565 int opt, optlen; 2566 2567 to->to_flags = 0; 2568 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2569 opt = cp[0]; 2570 if (opt == TCPOPT_EOL) 2571 break; 2572 if (opt == TCPOPT_NOP) 2573 optlen = 1; 2574 else { 2575 if (cnt < 2) 2576 break; 2577 optlen = cp[1]; 2578 if (optlen < 2 || optlen > cnt) 2579 break; 2580 } 2581 switch (opt) { 2582 case TCPOPT_MAXSEG: 2583 if (optlen != TCPOLEN_MAXSEG) 2584 continue; 2585 if (!is_syn) 2586 continue; 2587 to->to_flags |= TOF_MSS; 2588 bcopy((char *)cp + 2, 2589 (char *)&to->to_mss, sizeof(to->to_mss)); 2590 to->to_mss = ntohs(to->to_mss); 2591 break; 2592 case TCPOPT_WINDOW: 2593 if (optlen != TCPOLEN_WINDOW) 2594 continue; 2595 if (! is_syn) 2596 continue; 2597 to->to_flags |= TOF_SCALE; 2598 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2599 break; 2600 case TCPOPT_TIMESTAMP: 2601 if (optlen != TCPOLEN_TIMESTAMP) 2602 continue; 2603 to->to_flags |= TOF_TS; 2604 bcopy((char *)cp + 2, 2605 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2606 to->to_tsval = ntohl(to->to_tsval); 2607 bcopy((char *)cp + 6, 2608 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2609 to->to_tsecr = ntohl(to->to_tsecr); 2610 /* 2611 * If echoed timestamp is later than the current time, 2612 * fall back to non RFC1323 RTT calculation. 2613 */ 2614 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks)) 2615 to->to_tsecr = 0; 2616 break; 2617#ifdef TCP_SIGNATURE 2618 /* 2619 * XXX In order to reply to a host which has set the 2620 * TCP_SIGNATURE option in its initial SYN, we have to 2621 * record the fact that the option was observed here 2622 * for the syncache code to perform the correct response. 2623 */ 2624 case TCPOPT_SIGNATURE: 2625 if (optlen != TCPOLEN_SIGNATURE) 2626 continue; 2627 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2628 break; 2629#endif 2630 case TCPOPT_SACK_PERMITTED: 2631 if (!tcp_do_sack || 2632 optlen != TCPOLEN_SACK_PERMITTED) 2633 continue; 2634 if (is_syn) { 2635 /* MUST only be set on SYN */ 2636 to->to_flags |= TOF_SACK; 2637 } 2638 break; 2639 case TCPOPT_SACK: 2640 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2641 continue; 2642 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2643 to->to_sacks = cp + 2; 2644 tcpstat.tcps_sack_rcv_blocks++; 2645 break; 2646 default: 2647 continue; 2648 } 2649 } 2650} 2651 2652/* 2653 * Pull out of band byte out of a segment so 2654 * it doesn't appear in the user's data queue. 2655 * It is still reflected in the segment length for 2656 * sequencing purposes. 2657 */ 2658static void 2659tcp_pulloutofband(so, th, m, off) 2660 struct socket *so; 2661 struct tcphdr *th; 2662 register struct mbuf *m; 2663 int off; /* delayed to be droped hdrlen */ 2664{ 2665 int cnt = off + th->th_urp - 1; 2666 2667 while (cnt >= 0) { 2668 if (m->m_len > cnt) { 2669 char *cp = mtod(m, caddr_t) + cnt; 2670 struct tcpcb *tp = sototcpcb(so); 2671 2672 tp->t_iobc = *cp; 2673 tp->t_oobflags |= TCPOOB_HAVEDATA; 2674 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2675 m->m_len--; 2676 if (m->m_flags & M_PKTHDR) 2677 m->m_pkthdr.len--; 2678 return; 2679 } 2680 cnt -= m->m_len; 2681 m = m->m_next; 2682 if (m == 0) 2683 break; 2684 } 2685 panic("tcp_pulloutofband"); 2686} 2687 2688/* 2689 * Collect new round-trip time estimate 2690 * and update averages and current timeout. 2691 */ 2692static void 2693tcp_xmit_timer(tp, rtt) 2694 register struct tcpcb *tp; 2695 int rtt; 2696{ 2697 register int delta; 2698 2699 INP_LOCK_ASSERT(tp->t_inpcb); 2700 2701 tcpstat.tcps_rttupdated++; 2702 tp->t_rttupdated++; 2703 if (tp->t_srtt != 0) { 2704 /* 2705 * srtt is stored as fixed point with 5 bits after the 2706 * binary point (i.e., scaled by 8). The following magic 2707 * is equivalent to the smoothing algorithm in rfc793 with 2708 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2709 * point). Adjust rtt to origin 0. 2710 */ 2711 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2712 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2713 2714 if ((tp->t_srtt += delta) <= 0) 2715 tp->t_srtt = 1; 2716 2717 /* 2718 * We accumulate a smoothed rtt variance (actually, a 2719 * smoothed mean difference), then set the retransmit 2720 * timer to smoothed rtt + 4 times the smoothed variance. 2721 * rttvar is stored as fixed point with 4 bits after the 2722 * binary point (scaled by 16). The following is 2723 * equivalent to rfc793 smoothing with an alpha of .75 2724 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2725 * rfc793's wired-in beta. 2726 */ 2727 if (delta < 0) 2728 delta = -delta; 2729 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2730 if ((tp->t_rttvar += delta) <= 0) 2731 tp->t_rttvar = 1; 2732 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2733 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2734 } else { 2735 /* 2736 * No rtt measurement yet - use the unsmoothed rtt. 2737 * Set the variance to half the rtt (so our first 2738 * retransmit happens at 3*rtt). 2739 */ 2740 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2741 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2742 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2743 } 2744 tp->t_rtttime = 0; 2745 tp->t_rxtshift = 0; 2746 2747 /* 2748 * the retransmit should happen at rtt + 4 * rttvar. 2749 * Because of the way we do the smoothing, srtt and rttvar 2750 * will each average +1/2 tick of bias. When we compute 2751 * the retransmit timer, we want 1/2 tick of rounding and 2752 * 1 extra tick because of +-1/2 tick uncertainty in the 2753 * firing of the timer. The bias will give us exactly the 2754 * 1.5 tick we need. But, because the bias is 2755 * statistical, we have to test that we don't drop below 2756 * the minimum feasible timer (which is 2 ticks). 2757 */ 2758 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2759 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2760 2761 /* 2762 * We received an ack for a packet that wasn't retransmitted; 2763 * it is probably safe to discard any error indications we've 2764 * received recently. This isn't quite right, but close enough 2765 * for now (a route might have failed after we sent a segment, 2766 * and the return path might not be symmetrical). 2767 */ 2768 tp->t_softerror = 0; 2769} 2770 2771/* 2772 * Determine a reasonable value for maxseg size. 2773 * If the route is known, check route for mtu. 2774 * If none, use an mss that can be handled on the outgoing 2775 * interface without forcing IP to fragment; if bigger than 2776 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2777 * to utilize large mbufs. If no route is found, route has no mtu, 2778 * or the destination isn't local, use a default, hopefully conservative 2779 * size (usually 512 or the default IP max size, but no more than the mtu 2780 * of the interface), as we can't discover anything about intervening 2781 * gateways or networks. We also initialize the congestion/slow start 2782 * window to be a single segment if the destination isn't local. 2783 * While looking at the routing entry, we also initialize other path-dependent 2784 * parameters from pre-set or cached values in the routing entry. 2785 * 2786 * Also take into account the space needed for options that we 2787 * send regularly. Make maxseg shorter by that amount to assure 2788 * that we can send maxseg amount of data even when the options 2789 * are present. Store the upper limit of the length of options plus 2790 * data in maxopd. 2791 * 2792 * 2793 * In case of T/TCP, we call this routine during implicit connection 2794 * setup as well (offer = -1), to initialize maxseg from the cached 2795 * MSS of our peer. 2796 * 2797 * NOTE that this routine is only called when we process an incoming 2798 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2799 */ 2800void 2801tcp_mss(tp, offer) 2802 struct tcpcb *tp; 2803 int offer; 2804{ 2805 int rtt, mss; 2806 u_long bufsize; 2807 u_long maxmtu; 2808 struct inpcb *inp = tp->t_inpcb; 2809 struct socket *so; 2810 struct hc_metrics_lite metrics; 2811 int origoffer = offer; 2812#ifdef INET6 2813 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2814 size_t min_protoh = isipv6 ? 2815 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2816 sizeof (struct tcpiphdr); 2817#else 2818 const size_t min_protoh = sizeof(struct tcpiphdr); 2819#endif 2820 2821 /* initialize */ 2822#ifdef INET6 2823 if (isipv6) { 2824 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2825 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2826 } else 2827#endif 2828 { 2829 maxmtu = tcp_maxmtu(&inp->inp_inc); 2830 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2831 } 2832 so = inp->inp_socket; 2833 2834 /* 2835 * no route to sender, stay with default mss and return 2836 */ 2837 if (maxmtu == 0) 2838 return; 2839 2840 /* what have we got? */ 2841 switch (offer) { 2842 case 0: 2843 /* 2844 * Offer == 0 means that there was no MSS on the SYN 2845 * segment, in this case we use tcp_mssdflt. 2846 */ 2847 offer = 2848#ifdef INET6 2849 isipv6 ? tcp_v6mssdflt : 2850#endif 2851 tcp_mssdflt; 2852 break; 2853 2854 case -1: 2855 /* 2856 * Offer == -1 means that we didn't receive SYN yet. 2857 */ 2858 /* FALLTHROUGH */ 2859 2860 default: 2861 /* 2862 * Prevent DoS attack with too small MSS. Round up 2863 * to at least minmss. 2864 */ 2865 offer = max(offer, tcp_minmss); 2866 /* 2867 * Sanity check: make sure that maxopd will be large 2868 * enough to allow some data on segments even if the 2869 * all the option space is used (40bytes). Otherwise 2870 * funny things may happen in tcp_output. 2871 */ 2872 offer = max(offer, 64); 2873 } 2874 2875 /* 2876 * rmx information is now retrieved from tcp_hostcache 2877 */ 2878 tcp_hc_get(&inp->inp_inc, &metrics); 2879 2880 /* 2881 * if there's a discovered mtu int tcp hostcache, use it 2882 * else, use the link mtu. 2883 */ 2884 if (metrics.rmx_mtu) 2885 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2886 else { 2887#ifdef INET6 2888 if (isipv6) { 2889 mss = maxmtu - min_protoh; 2890 if (!path_mtu_discovery && 2891 !in6_localaddr(&inp->in6p_faddr)) 2892 mss = min(mss, tcp_v6mssdflt); 2893 } else 2894#endif 2895 { 2896 mss = maxmtu - min_protoh; 2897 if (!path_mtu_discovery && 2898 !in_localaddr(inp->inp_faddr)) 2899 mss = min(mss, tcp_mssdflt); 2900 } 2901 } 2902 mss = min(mss, offer); 2903 2904 /* 2905 * maxopd stores the maximum length of data AND options 2906 * in a segment; maxseg is the amount of data in a normal 2907 * segment. We need to store this value (maxopd) apart 2908 * from maxseg, because now every segment carries options 2909 * and thus we normally have somewhat less data in segments. 2910 */ 2911 tp->t_maxopd = mss; 2912 2913 /* 2914 * origoffer==-1 indicates, that no segments were received yet. 2915 * In this case we just guess. 2916 */ 2917 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2918 (origoffer == -1 || 2919 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2920 mss -= TCPOLEN_TSTAMP_APPA; 2921 tp->t_maxseg = mss; 2922 2923#if (MCLBYTES & (MCLBYTES - 1)) == 0 2924 if (mss > MCLBYTES) 2925 mss &= ~(MCLBYTES-1); 2926#else 2927 if (mss > MCLBYTES) 2928 mss = mss / MCLBYTES * MCLBYTES; 2929#endif 2930 tp->t_maxseg = mss; 2931 2932 /* 2933 * If there's a pipesize, change the socket buffer to that size, 2934 * don't change if sb_hiwat is different than default (then it 2935 * has been changed on purpose with setsockopt). 2936 * Make the socket buffers an integral number of mss units; 2937 * if the mss is larger than the socket buffer, decrease the mss. 2938 */ 2939 SOCKBUF_LOCK(&so->so_snd); 2940 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2941 bufsize = metrics.rmx_sendpipe; 2942 else 2943 bufsize = so->so_snd.sb_hiwat; 2944 if (bufsize < mss) 2945 mss = bufsize; 2946 else { 2947 bufsize = roundup(bufsize, mss); 2948 if (bufsize > sb_max) 2949 bufsize = sb_max; 2950 if (bufsize > so->so_snd.sb_hiwat) 2951 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2952 } 2953 SOCKBUF_UNLOCK(&so->so_snd); 2954 tp->t_maxseg = mss; 2955 2956 SOCKBUF_LOCK(&so->so_rcv); 2957 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2958 bufsize = metrics.rmx_recvpipe; 2959 else 2960 bufsize = so->so_rcv.sb_hiwat; 2961 if (bufsize > mss) { 2962 bufsize = roundup(bufsize, mss); 2963 if (bufsize > sb_max) 2964 bufsize = sb_max; 2965 if (bufsize > so->so_rcv.sb_hiwat) 2966 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2967 } 2968 SOCKBUF_UNLOCK(&so->so_rcv); 2969 /* 2970 * While we're here, check the others too 2971 */ 2972 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2973 tp->t_srtt = rtt; 2974 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2975 tcpstat.tcps_usedrtt++; 2976 if (metrics.rmx_rttvar) { 2977 tp->t_rttvar = metrics.rmx_rttvar; 2978 tcpstat.tcps_usedrttvar++; 2979 } else { 2980 /* default variation is +- 1 rtt */ 2981 tp->t_rttvar = 2982 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2983 } 2984 TCPT_RANGESET(tp->t_rxtcur, 2985 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2986 tp->t_rttmin, TCPTV_REXMTMAX); 2987 } 2988 if (metrics.rmx_ssthresh) { 2989 /* 2990 * There's some sort of gateway or interface 2991 * buffer limit on the path. Use this to set 2992 * the slow start threshhold, but set the 2993 * threshold to no less than 2*mss. 2994 */ 2995 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2996 tcpstat.tcps_usedssthresh++; 2997 } 2998 if (metrics.rmx_bandwidth) 2999 tp->snd_bandwidth = metrics.rmx_bandwidth; 3000 3001 /* 3002 * Set the slow-start flight size depending on whether this 3003 * is a local network or not. 3004 * 3005 * Extend this so we cache the cwnd too and retrieve it here. 3006 * Make cwnd even bigger than RFC3390 suggests but only if we 3007 * have previous experience with the remote host. Be careful 3008 * not make cwnd bigger than remote receive window or our own 3009 * send socket buffer. Maybe put some additional upper bound 3010 * on the retrieved cwnd. Should do incremental updates to 3011 * hostcache when cwnd collapses so next connection doesn't 3012 * overloads the path again. 3013 * 3014 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3015 * We currently check only in syncache_socket for that. 3016 */ 3017#define TCP_METRICS_CWND 3018#ifdef TCP_METRICS_CWND 3019 if (metrics.rmx_cwnd) 3020 tp->snd_cwnd = max(mss, 3021 min(metrics.rmx_cwnd / 2, 3022 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3023 else 3024#endif 3025 if (tcp_do_rfc3390) 3026 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3027#ifdef INET6 3028 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3029 (!isipv6 && in_localaddr(inp->inp_faddr))) 3030#else 3031 else if (in_localaddr(inp->inp_faddr)) 3032#endif 3033 tp->snd_cwnd = mss * ss_fltsz_local; 3034 else 3035 tp->snd_cwnd = mss * ss_fltsz; 3036} 3037 3038/* 3039 * Determine the MSS option to send on an outgoing SYN. 3040 */ 3041int 3042tcp_mssopt(inc) 3043 struct in_conninfo *inc; 3044{ 3045 int mss = 0; 3046 u_long maxmtu = 0; 3047 u_long thcmtu = 0; 3048 size_t min_protoh; 3049#ifdef INET6 3050 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3051#endif 3052 3053 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3054 3055#ifdef INET6 3056 if (isipv6) { 3057 mss = tcp_v6mssdflt; 3058 maxmtu = tcp_maxmtu6(inc); 3059 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3060 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3061 } else 3062#endif 3063 { 3064 mss = tcp_mssdflt; 3065 maxmtu = tcp_maxmtu(inc); 3066 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3067 min_protoh = sizeof(struct tcpiphdr); 3068 } 3069 if (maxmtu && thcmtu) 3070 mss = min(maxmtu, thcmtu) - min_protoh; 3071 else if (maxmtu || thcmtu) 3072 mss = max(maxmtu, thcmtu) - min_protoh; 3073 3074 return (mss); 3075} 3076 3077 3078/* 3079 * On a partial ack arrives, force the retransmission of the 3080 * next unacknowledged segment. Do not clear tp->t_dupacks. 3081 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3082 * be started again. 3083 */ 3084static void 3085tcp_newreno_partial_ack(tp, th) 3086 struct tcpcb *tp; 3087 struct tcphdr *th; 3088{ 3089 tcp_seq onxt = tp->snd_nxt; 3090 u_long ocwnd = tp->snd_cwnd; 3091 3092 callout_stop(tp->tt_rexmt); 3093 tp->t_rtttime = 0; 3094 tp->snd_nxt = th->th_ack; 3095 /* 3096 * Set snd_cwnd to one segment beyond acknowledged offset. 3097 * (tp->snd_una has not yet been updated when this function is called.) 3098 */ 3099 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3100 tp->t_flags |= TF_ACKNOW; 3101 (void) tcp_output(tp); 3102 tp->snd_cwnd = ocwnd; 3103 if (SEQ_GT(onxt, tp->snd_nxt)) 3104 tp->snd_nxt = onxt; 3105 /* 3106 * Partial window deflation. Relies on fact that tp->snd_una 3107 * not updated yet. 3108 */ 3109 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3110} 3111 3112/* 3113 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3114 * looking for a pcb in the listen state. Returns 0 otherwise. 3115 */ 3116static int 3117tcp_timewait(tw, to, th, m, tlen) 3118 struct tcptw *tw; 3119 struct tcpopt *to; 3120 struct tcphdr *th; 3121 struct mbuf *m; 3122 int tlen; 3123{ 3124 int thflags; 3125 tcp_seq seq; 3126#ifdef INET6 3127 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3128#else 3129 const int isipv6 = 0; 3130#endif 3131 3132 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */ 3133 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3134 INP_LOCK_ASSERT(tw->tw_inpcb); 3135 3136 thflags = th->th_flags; 3137 3138 /* 3139 * NOTE: for FIN_WAIT_2 (to be added later), 3140 * must validate sequence number before accepting RST 3141 */ 3142 3143 /* 3144 * If the segment contains RST: 3145 * Drop the segment - see Stevens, vol. 2, p. 964 and 3146 * RFC 1337. 3147 */ 3148 if (thflags & TH_RST) 3149 goto drop; 3150 3151#if 0 3152/* PAWS not needed at the moment */ 3153 /* 3154 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3155 * and it's less than ts_recent, drop it. 3156 */ 3157 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3158 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3159 if ((thflags & TH_ACK) == 0) 3160 goto drop; 3161 goto ack; 3162 } 3163 /* 3164 * ts_recent is never updated because we never accept new segments. 3165 */ 3166#endif 3167 3168 /* 3169 * If a new connection request is received 3170 * while in TIME_WAIT, drop the old connection 3171 * and start over if the sequence numbers 3172 * are above the previous ones. 3173 */ 3174 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3175 (void) tcp_twclose(tw, 0); 3176 return (1); 3177 } 3178 3179 /* 3180 * Drop the the segment if it does not contain an ACK. 3181 */ 3182 if ((thflags & TH_ACK) == 0) 3183 goto drop; 3184 3185 /* 3186 * Reset the 2MSL timer if this is a duplicate FIN. 3187 */ 3188 if (thflags & TH_FIN) { 3189 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3190 if (seq + 1 == tw->rcv_nxt) 3191 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3192 } 3193 3194 /* 3195 * Acknowledge the segment if it has data or is not a duplicate ACK. 3196 */ 3197 if (thflags != TH_ACK || tlen != 0 || 3198 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3199 tcp_twrespond(tw, TH_ACK); 3200 goto drop; 3201 3202 /* 3203 * Generate a RST, dropping incoming segment. 3204 * Make ACK acceptable to originator of segment. 3205 * Don't bother to respond if destination was broadcast/multicast. 3206 */ 3207 if (m->m_flags & (M_BCAST|M_MCAST)) 3208 goto drop; 3209 if (isipv6) { 3210 struct ip6_hdr *ip6; 3211 3212 /* IPv6 anycast check is done at tcp6_input() */ 3213 ip6 = mtod(m, struct ip6_hdr *); 3214 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3215 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3216 goto drop; 3217 } else { 3218 struct ip *ip; 3219 3220 ip = mtod(m, struct ip *); 3221 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3222 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3223 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3224 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3225 goto drop; 3226 } 3227 if (thflags & TH_ACK) { 3228 tcp_respond(NULL, 3229 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3230 } else { 3231 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3232 tcp_respond(NULL, 3233 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3234 } 3235 INP_UNLOCK(tw->tw_inpcb); 3236 return (0); 3237 3238drop: 3239 INP_UNLOCK(tw->tw_inpcb); 3240 m_freem(m); 3241 return (0); 3242}
|