tcp_input.c revision 154528
1/*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 30 * $FreeBSD: head/sys/netinet/tcp_input.c 154528 2006-01-18 18:59:30Z andre $ 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/ip_options.h> 72#include <netinet/ip6.h> 73#include <netinet/icmp6.h> 74#include <netinet6/in6_pcb.h> 75#include <netinet6/ip6_var.h> 76#include <netinet6/nd6.h> 77#include <netinet/tcp.h> 78#include <netinet/tcp_fsm.h> 79#include <netinet/tcp_seq.h> 80#include <netinet/tcp_timer.h> 81#include <netinet/tcp_var.h> 82#include <netinet6/tcp6_var.h> 83#include <netinet/tcpip.h> 84#ifdef TCPDEBUG 85#include <netinet/tcp_debug.h> 86#endif /* TCPDEBUG */ 87 88#ifdef FAST_IPSEC 89#include <netipsec/ipsec.h> 90#include <netipsec/ipsec6.h> 91#endif /*FAST_IPSEC*/ 92 93#ifdef IPSEC 94#include <netinet6/ipsec.h> 95#include <netinet6/ipsec6.h> 96#include <netkey/key.h> 97#endif /*IPSEC*/ 98 99#include <machine/in_cksum.h> 100 101static const int tcprexmtthresh = 3; 102 103struct tcpstat tcpstat; 104SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 105 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 106 107static int log_in_vain = 0; 108SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 109 &log_in_vain, 0, "Log all incoming TCP connections"); 110 111static int blackhole = 0; 112SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 113 &blackhole, 0, "Do not send RST when dropping refused connections"); 114 115int tcp_delack_enabled = 1; 116SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 117 &tcp_delack_enabled, 0, 118 "Delay ACK to try and piggyback it onto a data packet"); 119 120#ifdef TCP_DROP_SYNFIN 121static int drop_synfin = 0; 122SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 123 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 124#endif 125 126static int tcp_do_rfc3042 = 1; 127SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 128 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 129 130static int tcp_do_rfc3390 = 1; 131SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 132 &tcp_do_rfc3390, 0, 133 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 134 135static int tcp_insecure_rst = 0; 136SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 137 &tcp_insecure_rst, 0, 138 "Follow the old (insecure) criteria for accepting RST packets."); 139 140SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 141 "TCP Segment Reassembly Queue"); 142 143static int tcp_reass_maxseg = 0; 144SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 145 &tcp_reass_maxseg, 0, 146 "Global maximum number of TCP Segments in Reassembly Queue"); 147 148int tcp_reass_qsize = 0; 149SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 150 &tcp_reass_qsize, 0, 151 "Global number of TCP Segments currently in Reassembly Queue"); 152 153static int tcp_reass_maxqlen = 48; 154SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 155 &tcp_reass_maxqlen, 0, 156 "Maximum number of TCP Segments per individual Reassembly Queue"); 157 158static int tcp_reass_overflows = 0; 159SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 160 &tcp_reass_overflows, 0, 161 "Global number of TCP Segment Reassembly Queue Overflows"); 162 163struct inpcbhead tcb; 164#define tcb6 tcb /* for KAME src sync over BSD*'s */ 165struct inpcbinfo tcbinfo; 166struct mtx *tcbinfo_mtx; 167 168static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 169 170static void tcp_pulloutofband(struct socket *, 171 struct tcphdr *, struct mbuf *, int); 172static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 173 struct mbuf *); 174static void tcp_xmit_timer(struct tcpcb *, int); 175static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 176static int tcp_timewait(struct tcptw *, struct tcpopt *, 177 struct tcphdr *, struct mbuf *, int); 178 179/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 180#ifdef INET6 181#define ND6_HINT(tp) \ 182do { \ 183 if ((tp) && (tp)->t_inpcb && \ 184 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 185 nd6_nud_hint(NULL, NULL, 0); \ 186} while (0) 187#else 188#define ND6_HINT(tp) 189#endif 190 191/* 192 * Indicate whether this ack should be delayed. We can delay the ack if 193 * - there is no delayed ack timer in progress and 194 * - our last ack wasn't a 0-sized window. We never want to delay 195 * the ack that opens up a 0-sized window and 196 * - delayed acks are enabled or 197 * - this is a half-synchronized T/TCP connection. 198 */ 199#define DELAY_ACK(tp) \ 200 ((!callout_active(tp->tt_delack) && \ 201 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 202 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 203 204/* Initialize TCP reassembly queue */ 205uma_zone_t tcp_reass_zone; 206void 207tcp_reass_init() 208{ 209 tcp_reass_maxseg = nmbclusters / 16; 210 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 211 &tcp_reass_maxseg); 212 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 213 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 214 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 215} 216 217static int 218tcp_reass(tp, th, tlenp, m) 219 register struct tcpcb *tp; 220 register struct tcphdr *th; 221 int *tlenp; 222 struct mbuf *m; 223{ 224 struct tseg_qent *q; 225 struct tseg_qent *p = NULL; 226 struct tseg_qent *nq; 227 struct tseg_qent *te = NULL; 228 struct socket *so = tp->t_inpcb->inp_socket; 229 int flags; 230 231 INP_LOCK_ASSERT(tp->t_inpcb); 232 233 /* 234 * XXX: tcp_reass() is rather inefficient with its data structures 235 * and should be rewritten (see NetBSD for optimizations). While 236 * doing that it should move to its own file tcp_reass.c. 237 */ 238 239 /* 240 * Call with th==NULL after become established to 241 * force pre-ESTABLISHED data up to user socket. 242 */ 243 if (th == NULL) 244 goto present; 245 246 /* 247 * Limit the number of segments in the reassembly queue to prevent 248 * holding on to too many segments (and thus running out of mbufs). 249 * Make sure to let the missing segment through which caused this 250 * queue. Always keep one global queue entry spare to be able to 251 * process the missing segment. 252 */ 253 if (th->th_seq != tp->rcv_nxt && 254 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 255 tp->t_segqlen >= tcp_reass_maxqlen)) { 256 tcp_reass_overflows++; 257 tcpstat.tcps_rcvmemdrop++; 258 m_freem(m); 259 *tlenp = 0; 260 return (0); 261 } 262 263 /* 264 * Allocate a new queue entry. If we can't, or hit the zone limit 265 * just drop the pkt. 266 */ 267 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 268 if (te == NULL) { 269 tcpstat.tcps_rcvmemdrop++; 270 m_freem(m); 271 *tlenp = 0; 272 return (0); 273 } 274 tp->t_segqlen++; 275 tcp_reass_qsize++; 276 277 /* 278 * Find a segment which begins after this one does. 279 */ 280 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 281 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 282 break; 283 p = q; 284 } 285 286 /* 287 * If there is a preceding segment, it may provide some of 288 * our data already. If so, drop the data from the incoming 289 * segment. If it provides all of our data, drop us. 290 */ 291 if (p != NULL) { 292 register int i; 293 /* conversion to int (in i) handles seq wraparound */ 294 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 295 if (i > 0) { 296 if (i >= *tlenp) { 297 tcpstat.tcps_rcvduppack++; 298 tcpstat.tcps_rcvdupbyte += *tlenp; 299 m_freem(m); 300 uma_zfree(tcp_reass_zone, te); 301 tp->t_segqlen--; 302 tcp_reass_qsize--; 303 /* 304 * Try to present any queued data 305 * at the left window edge to the user. 306 * This is needed after the 3-WHS 307 * completes. 308 */ 309 goto present; /* ??? */ 310 } 311 m_adj(m, i); 312 *tlenp -= i; 313 th->th_seq += i; 314 } 315 } 316 tcpstat.tcps_rcvoopack++; 317 tcpstat.tcps_rcvoobyte += *tlenp; 318 319 /* 320 * While we overlap succeeding segments trim them or, 321 * if they are completely covered, dequeue them. 322 */ 323 while (q) { 324 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 325 if (i <= 0) 326 break; 327 if (i < q->tqe_len) { 328 q->tqe_th->th_seq += i; 329 q->tqe_len -= i; 330 m_adj(q->tqe_m, i); 331 break; 332 } 333 334 nq = LIST_NEXT(q, tqe_q); 335 LIST_REMOVE(q, tqe_q); 336 m_freem(q->tqe_m); 337 uma_zfree(tcp_reass_zone, q); 338 tp->t_segqlen--; 339 tcp_reass_qsize--; 340 q = nq; 341 } 342 343 /* Insert the new segment queue entry into place. */ 344 te->tqe_m = m; 345 te->tqe_th = th; 346 te->tqe_len = *tlenp; 347 348 if (p == NULL) { 349 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 350 } else { 351 LIST_INSERT_AFTER(p, te, tqe_q); 352 } 353 354present: 355 /* 356 * Present data to user, advancing rcv_nxt through 357 * completed sequence space. 358 */ 359 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 360 return (0); 361 q = LIST_FIRST(&tp->t_segq); 362 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 363 return (0); 364 SOCKBUF_LOCK(&so->so_rcv); 365 do { 366 tp->rcv_nxt += q->tqe_len; 367 flags = q->tqe_th->th_flags & TH_FIN; 368 nq = LIST_NEXT(q, tqe_q); 369 LIST_REMOVE(q, tqe_q); 370 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 371 m_freem(q->tqe_m); 372 else 373 sbappendstream_locked(&so->so_rcv, q->tqe_m); 374 uma_zfree(tcp_reass_zone, q); 375 tp->t_segqlen--; 376 tcp_reass_qsize--; 377 q = nq; 378 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 379 ND6_HINT(tp); 380 sorwakeup_locked(so); 381 return (flags); 382} 383 384/* 385 * TCP input routine, follows pages 65-76 of the 386 * protocol specification dated September, 1981 very closely. 387 */ 388#ifdef INET6 389int 390tcp6_input(mp, offp, proto) 391 struct mbuf **mp; 392 int *offp, proto; 393{ 394 register struct mbuf *m = *mp; 395 struct in6_ifaddr *ia6; 396 397 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 398 399 /* 400 * draft-itojun-ipv6-tcp-to-anycast 401 * better place to put this in? 402 */ 403 ia6 = ip6_getdstifaddr(m); 404 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 405 struct ip6_hdr *ip6; 406 407 ip6 = mtod(m, struct ip6_hdr *); 408 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 409 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 410 return IPPROTO_DONE; 411 } 412 413 tcp_input(m, *offp); 414 return IPPROTO_DONE; 415} 416#endif 417 418void 419tcp_input(m, off0) 420 register struct mbuf *m; 421 int off0; 422{ 423 register struct tcphdr *th; 424 register struct ip *ip = NULL; 425 register struct ipovly *ipov; 426 register struct inpcb *inp = NULL; 427 u_char *optp = NULL; 428 int optlen = 0; 429 int len, tlen, off; 430 int drop_hdrlen; 431 register struct tcpcb *tp = 0; 432 register int thflags; 433 struct socket *so = 0; 434 int todrop, acked, ourfinisacked, needoutput = 0; 435 u_long tiwin; 436 struct tcpopt to; /* options in this segment */ 437 int headlocked = 0; 438#ifdef IPFIREWALL_FORWARD 439 struct m_tag *fwd_tag; 440#endif 441 int rstreason; /* For badport_bandlim accounting purposes */ 442 443 struct ip6_hdr *ip6 = NULL; 444#ifdef INET6 445 int isipv6; 446#else 447 const int isipv6 = 0; 448#endif 449 450#ifdef TCPDEBUG 451 /* 452 * The size of tcp_saveipgen must be the size of the max ip header, 453 * now IPv6. 454 */ 455 u_char tcp_saveipgen[40]; 456 struct tcphdr tcp_savetcp; 457 short ostate = 0; 458#endif 459 460#ifdef INET6 461 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 462#endif 463 bzero((char *)&to, sizeof(to)); 464 465 tcpstat.tcps_rcvtotal++; 466 467 if (isipv6) { 468#ifdef INET6 469 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 470 ip6 = mtod(m, struct ip6_hdr *); 471 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 472 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 473 tcpstat.tcps_rcvbadsum++; 474 goto drop; 475 } 476 th = (struct tcphdr *)((caddr_t)ip6 + off0); 477 478 /* 479 * Be proactive about unspecified IPv6 address in source. 480 * As we use all-zero to indicate unbounded/unconnected pcb, 481 * unspecified IPv6 address can be used to confuse us. 482 * 483 * Note that packets with unspecified IPv6 destination is 484 * already dropped in ip6_input. 485 */ 486 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 487 /* XXX stat */ 488 goto drop; 489 } 490#else 491 th = NULL; /* XXX: avoid compiler warning */ 492#endif 493 } else { 494 /* 495 * Get IP and TCP header together in first mbuf. 496 * Note: IP leaves IP header in first mbuf. 497 */ 498 if (off0 > sizeof (struct ip)) { 499 ip_stripoptions(m, (struct mbuf *)0); 500 off0 = sizeof(struct ip); 501 } 502 if (m->m_len < sizeof (struct tcpiphdr)) { 503 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 504 tcpstat.tcps_rcvshort++; 505 return; 506 } 507 } 508 ip = mtod(m, struct ip *); 509 ipov = (struct ipovly *)ip; 510 th = (struct tcphdr *)((caddr_t)ip + off0); 511 tlen = ip->ip_len; 512 513 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 514 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 515 th->th_sum = m->m_pkthdr.csum_data; 516 else 517 th->th_sum = in_pseudo(ip->ip_src.s_addr, 518 ip->ip_dst.s_addr, 519 htonl(m->m_pkthdr.csum_data + 520 ip->ip_len + 521 IPPROTO_TCP)); 522 th->th_sum ^= 0xffff; 523#ifdef TCPDEBUG 524 ipov->ih_len = (u_short)tlen; 525 ipov->ih_len = htons(ipov->ih_len); 526#endif 527 } else { 528 /* 529 * Checksum extended TCP header and data. 530 */ 531 len = sizeof (struct ip) + tlen; 532 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 533 ipov->ih_len = (u_short)tlen; 534 ipov->ih_len = htons(ipov->ih_len); 535 th->th_sum = in_cksum(m, len); 536 } 537 if (th->th_sum) { 538 tcpstat.tcps_rcvbadsum++; 539 goto drop; 540 } 541#ifdef INET6 542 /* Re-initialization for later version check */ 543 ip->ip_v = IPVERSION; 544#endif 545 } 546 547 /* 548 * Check that TCP offset makes sense, 549 * pull out TCP options and adjust length. XXX 550 */ 551 off = th->th_off << 2; 552 if (off < sizeof (struct tcphdr) || off > tlen) { 553 tcpstat.tcps_rcvbadoff++; 554 goto drop; 555 } 556 tlen -= off; /* tlen is used instead of ti->ti_len */ 557 if (off > sizeof (struct tcphdr)) { 558 if (isipv6) { 559#ifdef INET6 560 IP6_EXTHDR_CHECK(m, off0, off, ); 561 ip6 = mtod(m, struct ip6_hdr *); 562 th = (struct tcphdr *)((caddr_t)ip6 + off0); 563#endif 564 } else { 565 if (m->m_len < sizeof(struct ip) + off) { 566 if ((m = m_pullup(m, sizeof (struct ip) + off)) 567 == 0) { 568 tcpstat.tcps_rcvshort++; 569 return; 570 } 571 ip = mtod(m, struct ip *); 572 ipov = (struct ipovly *)ip; 573 th = (struct tcphdr *)((caddr_t)ip + off0); 574 } 575 } 576 optlen = off - sizeof (struct tcphdr); 577 optp = (u_char *)(th + 1); 578 } 579 thflags = th->th_flags; 580 581#ifdef TCP_DROP_SYNFIN 582 /* 583 * If the drop_synfin option is enabled, drop all packets with 584 * both the SYN and FIN bits set. This prevents e.g. nmap from 585 * identifying the TCP/IP stack. 586 * 587 * This is a violation of the TCP specification. 588 */ 589 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 590 goto drop; 591#endif 592 593 /* 594 * Convert TCP protocol specific fields to host format. 595 */ 596 th->th_seq = ntohl(th->th_seq); 597 th->th_ack = ntohl(th->th_ack); 598 th->th_win = ntohs(th->th_win); 599 th->th_urp = ntohs(th->th_urp); 600 601 /* 602 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 603 * until after ip6_savecontrol() is called and before other functions 604 * which don't want those proto headers. 605 * Because ip6_savecontrol() is going to parse the mbuf to 606 * search for data to be passed up to user-land, it wants mbuf 607 * parameters to be unchanged. 608 * XXX: the call of ip6_savecontrol() has been obsoleted based on 609 * latest version of the advanced API (20020110). 610 */ 611 drop_hdrlen = off0 + off; 612 613 /* 614 * Locate pcb for segment. 615 */ 616 INP_INFO_WLOCK(&tcbinfo); 617 headlocked = 1; 618findpcb: 619 KASSERT(headlocked, ("tcp_input: findpcb: head not locked")); 620#ifdef IPFIREWALL_FORWARD 621 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 622 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 623 624 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 625 struct sockaddr_in *next_hop; 626 627 next_hop = (struct sockaddr_in *)(fwd_tag+1); 628 /* 629 * Transparently forwarded. Pretend to be the destination. 630 * already got one like this? 631 */ 632 inp = in_pcblookup_hash(&tcbinfo, 633 ip->ip_src, th->th_sport, 634 ip->ip_dst, th->th_dport, 635 0, m->m_pkthdr.rcvif); 636 if (!inp) { 637 /* It's new. Try to find the ambushing socket. */ 638 inp = in_pcblookup_hash(&tcbinfo, 639 ip->ip_src, th->th_sport, 640 next_hop->sin_addr, 641 next_hop->sin_port ? 642 ntohs(next_hop->sin_port) : 643 th->th_dport, 644 1, m->m_pkthdr.rcvif); 645 } 646 /* Remove the tag from the packet. We don't need it anymore. */ 647 m_tag_delete(m, fwd_tag); 648 } else { 649#endif /* IPFIREWALL_FORWARD */ 650 if (isipv6) { 651#ifdef INET6 652 inp = in6_pcblookup_hash(&tcbinfo, 653 &ip6->ip6_src, th->th_sport, 654 &ip6->ip6_dst, th->th_dport, 655 1, m->m_pkthdr.rcvif); 656#endif 657 } else 658 inp = in_pcblookup_hash(&tcbinfo, 659 ip->ip_src, th->th_sport, 660 ip->ip_dst, th->th_dport, 661 1, m->m_pkthdr.rcvif); 662#ifdef IPFIREWALL_FORWARD 663 } 664#endif /* IPFIREWALL_FORWARD */ 665 666#if defined(IPSEC) || defined(FAST_IPSEC) 667#ifdef INET6 668 if (isipv6) { 669 if (inp != NULL && ipsec6_in_reject(m, inp)) { 670#ifdef IPSEC 671 ipsec6stat.in_polvio++; 672#endif 673 goto drop; 674 } 675 } else 676#endif /* INET6 */ 677 if (inp != NULL && ipsec4_in_reject(m, inp)) { 678#ifdef IPSEC 679 ipsecstat.in_polvio++; 680#endif 681 goto drop; 682 } 683#endif /*IPSEC || FAST_IPSEC*/ 684 685 /* 686 * If the state is CLOSED (i.e., TCB does not exist) then 687 * all data in the incoming segment is discarded. 688 * If the TCB exists but is in CLOSED state, it is embryonic, 689 * but should either do a listen or a connect soon. 690 */ 691 if (inp == NULL) { 692 if (log_in_vain) { 693#ifdef INET6 694 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 695#else 696 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 697#endif 698 699 if (isipv6) { 700#ifdef INET6 701 strcpy(dbuf, "["); 702 strcpy(sbuf, "["); 703 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 704 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 705 strcat(dbuf, "]"); 706 strcat(sbuf, "]"); 707#endif 708 } else { 709 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 710 strcpy(sbuf, inet_ntoa(ip->ip_src)); 711 } 712 switch (log_in_vain) { 713 case 1: 714 if ((thflags & TH_SYN) == 0) 715 break; 716 /* FALLTHROUGH */ 717 case 2: 718 log(LOG_INFO, 719 "Connection attempt to TCP %s:%d " 720 "from %s:%d flags:0x%02x\n", 721 dbuf, ntohs(th->th_dport), sbuf, 722 ntohs(th->th_sport), thflags); 723 break; 724 default: 725 break; 726 } 727 } 728 if (blackhole) { 729 switch (blackhole) { 730 case 1: 731 if (thflags & TH_SYN) 732 goto drop; 733 break; 734 case 2: 735 goto drop; 736 default: 737 goto drop; 738 } 739 } 740 rstreason = BANDLIM_RST_CLOSEDPORT; 741 goto dropwithreset; 742 } 743 INP_LOCK(inp); 744 745 /* Check the minimum TTL for socket. */ 746 if (inp->inp_ip_minttl != 0) { 747#ifdef INET6 748 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 749 goto drop; 750 else 751#endif 752 if (inp->inp_ip_minttl > ip->ip_ttl) 753 goto drop; 754 } 755 756 if (inp->inp_vflag & INP_TIMEWAIT) { 757 /* 758 * The only option of relevance is TOF_CC, and only if 759 * present in a SYN segment. See tcp_timewait(). 760 */ 761 if (thflags & TH_SYN) 762 tcp_dooptions(&to, optp, optlen, 1); 763 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 764 &to, th, m, tlen)) 765 goto findpcb; 766 /* 767 * tcp_timewait unlocks inp. 768 */ 769 INP_INFO_WUNLOCK(&tcbinfo); 770 return; 771 } 772 tp = intotcpcb(inp); 773 if (tp == 0) { 774 INP_UNLOCK(inp); 775 rstreason = BANDLIM_RST_CLOSEDPORT; 776 goto dropwithreset; 777 } 778 if (tp->t_state == TCPS_CLOSED) 779 goto drop; 780 781 /* Unscale the window into a 32-bit value. */ 782 if ((thflags & TH_SYN) == 0) 783 tiwin = th->th_win << tp->snd_scale; 784 else 785 tiwin = th->th_win; 786 787#ifdef MAC 788 INP_LOCK_ASSERT(inp); 789 if (mac_check_inpcb_deliver(inp, m)) 790 goto drop; 791#endif 792 so = inp->inp_socket; 793#ifdef TCPDEBUG 794 if (so->so_options & SO_DEBUG) { 795 ostate = tp->t_state; 796 if (isipv6) 797 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 798 else 799 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 800 tcp_savetcp = *th; 801 } 802#endif 803 if (so->so_options & SO_ACCEPTCONN) { 804 struct in_conninfo inc; 805 806#ifdef INET6 807 inc.inc_isipv6 = isipv6; 808#endif 809 if (isipv6) { 810 inc.inc6_faddr = ip6->ip6_src; 811 inc.inc6_laddr = ip6->ip6_dst; 812 } else { 813 inc.inc_faddr = ip->ip_src; 814 inc.inc_laddr = ip->ip_dst; 815 } 816 inc.inc_fport = th->th_sport; 817 inc.inc_lport = th->th_dport; 818 819 /* 820 * If the state is LISTEN then ignore segment if it contains 821 * a RST. If the segment contains an ACK then it is bad and 822 * send a RST. If it does not contain a SYN then it is not 823 * interesting; drop it. 824 * 825 * If the state is SYN_RECEIVED (syncache) and seg contains 826 * an ACK, but not for our SYN/ACK, send a RST. If the seg 827 * contains a RST, check the sequence number to see if it 828 * is a valid reset segment. 829 */ 830 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 831 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 832 if (!syncache_expand(&inc, th, &so, m)) { 833 /* 834 * No syncache entry, or ACK was not 835 * for our SYN/ACK. Send a RST. 836 */ 837 tcpstat.tcps_badsyn++; 838 rstreason = BANDLIM_RST_OPENPORT; 839 goto dropwithreset; 840 } 841 if (so == NULL) { 842 /* 843 * Could not complete 3-way handshake, 844 * connection is being closed down, and 845 * syncache will free mbuf. 846 */ 847 INP_UNLOCK(inp); 848 INP_INFO_WUNLOCK(&tcbinfo); 849 return; 850 } 851 /* 852 * Socket is created in state SYN_RECEIVED. 853 * Continue processing segment. 854 */ 855 INP_UNLOCK(inp); 856 inp = sotoinpcb(so); 857 INP_LOCK(inp); 858 tp = intotcpcb(inp); 859 /* 860 * This is what would have happened in 861 * tcp_output() when the SYN,ACK was sent. 862 */ 863 tp->snd_up = tp->snd_una; 864 tp->snd_max = tp->snd_nxt = tp->iss + 1; 865 tp->last_ack_sent = tp->rcv_nxt; 866 /* 867 * RFC1323: The window in SYN & SYN/ACK 868 * segments is never scaled. 869 */ 870 tp->snd_wnd = tiwin; /* unscaled */ 871 goto after_listen; 872 } 873 if (thflags & TH_RST) { 874 syncache_chkrst(&inc, th); 875 goto drop; 876 } 877 if (thflags & TH_ACK) { 878 syncache_badack(&inc); 879 tcpstat.tcps_badsyn++; 880 rstreason = BANDLIM_RST_OPENPORT; 881 goto dropwithreset; 882 } 883 goto drop; 884 } 885 886 /* 887 * Segment's flags are (SYN) or (SYN|FIN). 888 */ 889#ifdef INET6 890 /* 891 * If deprecated address is forbidden, 892 * we do not accept SYN to deprecated interface 893 * address to prevent any new inbound connection from 894 * getting established. 895 * When we do not accept SYN, we send a TCP RST, 896 * with deprecated source address (instead of dropping 897 * it). We compromise it as it is much better for peer 898 * to send a RST, and RST will be the final packet 899 * for the exchange. 900 * 901 * If we do not forbid deprecated addresses, we accept 902 * the SYN packet. RFC2462 does not suggest dropping 903 * SYN in this case. 904 * If we decipher RFC2462 5.5.4, it says like this: 905 * 1. use of deprecated addr with existing 906 * communication is okay - "SHOULD continue to be 907 * used" 908 * 2. use of it with new communication: 909 * (2a) "SHOULD NOT be used if alternate address 910 * with sufficient scope is available" 911 * (2b) nothing mentioned otherwise. 912 * Here we fall into (2b) case as we have no choice in 913 * our source address selection - we must obey the peer. 914 * 915 * The wording in RFC2462 is confusing, and there are 916 * multiple description text for deprecated address 917 * handling - worse, they are not exactly the same. 918 * I believe 5.5.4 is the best one, so we follow 5.5.4. 919 */ 920 if (isipv6 && !ip6_use_deprecated) { 921 struct in6_ifaddr *ia6; 922 923 if ((ia6 = ip6_getdstifaddr(m)) && 924 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 925 INP_UNLOCK(inp); 926 tp = NULL; 927 rstreason = BANDLIM_RST_OPENPORT; 928 goto dropwithreset; 929 } 930 } 931#endif 932 /* 933 * If it is from this socket, drop it, it must be forged. 934 * Don't bother responding if the destination was a broadcast. 935 */ 936 if (th->th_dport == th->th_sport) { 937 if (isipv6) { 938 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 939 &ip6->ip6_src)) 940 goto drop; 941 } else { 942 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 943 goto drop; 944 } 945 } 946 /* 947 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 948 * 949 * Note that it is quite possible to receive unicast 950 * link-layer packets with a broadcast IP address. Use 951 * in_broadcast() to find them. 952 */ 953 if (m->m_flags & (M_BCAST|M_MCAST)) 954 goto drop; 955 if (isipv6) { 956 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 957 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 958 goto drop; 959 } else { 960 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 961 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 962 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 963 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 964 goto drop; 965 } 966 /* 967 * SYN appears to be valid; create compressed TCP state 968 * for syncache, or perform t/tcp connection. 969 */ 970 if (so->so_qlen <= so->so_qlimit) { 971#ifdef TCPDEBUG 972 if (so->so_options & SO_DEBUG) 973 tcp_trace(TA_INPUT, ostate, tp, 974 (void *)tcp_saveipgen, &tcp_savetcp, 0); 975#endif 976 tcp_dooptions(&to, optp, optlen, 1); 977 if (!syncache_add(&inc, &to, th, &so, m)) 978 goto drop; 979 if (so == NULL) { 980 /* 981 * Entry added to syncache, mbuf used to 982 * send SYN,ACK packet. 983 */ 984 KASSERT(headlocked, ("headlocked")); 985 INP_UNLOCK(inp); 986 INP_INFO_WUNLOCK(&tcbinfo); 987 return; 988 } 989 /* 990 * Segment passed TAO tests. 991 */ 992 INP_UNLOCK(inp); 993 inp = sotoinpcb(so); 994 INP_LOCK(inp); 995 tp = intotcpcb(inp); 996 tp->snd_wnd = tiwin; 997 tp->t_starttime = ticks; 998 tp->t_state = TCPS_ESTABLISHED; 999 1000 /* 1001 * T/TCP logic: 1002 * If there is a FIN or if there is data, then 1003 * delay SYN,ACK(SYN) in the hope of piggy-backing 1004 * it on a response segment. Otherwise must send 1005 * ACK now in case the other side is slow starting. 1006 */ 1007 if (thflags & TH_FIN || tlen != 0) 1008 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 1009 else 1010 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1011 tcpstat.tcps_connects++; 1012 soisconnected(so); 1013 goto trimthenstep6; 1014 } 1015 goto drop; 1016 } 1017after_listen: 1018 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1019 INP_LOCK_ASSERT(inp); 1020 1021 /* Syncache takes care of sockets in the listen state. */ 1022 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN")); 1023 1024 /* 1025 * This is the second part of the MSS DoS prevention code (after 1026 * minmss on the sending side) and it deals with too many too small 1027 * tcp packets in a too short timeframe (1 second). 1028 * 1029 * For every full second we count the number of received packets 1030 * and bytes. If we get a lot of packets per second for this connection 1031 * (tcp_minmssoverload) we take a closer look at it and compute the 1032 * average packet size for the past second. If that is less than 1033 * tcp_minmss we get too many packets with very small payload which 1034 * is not good and burdens our system (and every packet generates 1035 * a wakeup to the process connected to our socket). We can reasonable 1036 * expect this to be small packet DoS attack to exhaust our CPU 1037 * cycles. 1038 * 1039 * Care has to be taken for the minimum packet overload value. This 1040 * value defines the minimum number of packets per second before we 1041 * start to worry. This must not be too low to avoid killing for 1042 * example interactive connections with many small packets like 1043 * telnet or SSH. 1044 * 1045 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1046 * this check. 1047 * 1048 * Account for packet if payload packet, skip over ACK, etc. 1049 */ 1050 if (tcp_minmss && tcp_minmssoverload && 1051 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1052 if ((unsigned int)(tp->rcv_second - ticks) < hz) { 1053 tp->rcv_pps++; 1054 tp->rcv_byps += tlen + off; 1055 if (tp->rcv_pps > tcp_minmssoverload) { 1056 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1057 printf("too many small tcp packets from " 1058 "%s:%u, av. %lubyte/packet, " 1059 "dropping connection\n", 1060#ifdef INET6 1061 isipv6 ? 1062 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1063#endif 1064 inet_ntoa(inp->inp_inc.inc_faddr), 1065 inp->inp_inc.inc_fport, 1066 tp->rcv_byps / tp->rcv_pps); 1067 KASSERT(headlocked, ("tcp_input: " 1068 "after_listen: tcp_drop: head " 1069 "not locked")); 1070 tp = tcp_drop(tp, ECONNRESET); 1071 tcpstat.tcps_minmssdrops++; 1072 goto drop; 1073 } 1074 } 1075 } else { 1076 tp->rcv_second = ticks + hz; 1077 tp->rcv_pps = 1; 1078 tp->rcv_byps = tlen + off; 1079 } 1080 } 1081 1082 /* 1083 * Segment received on connection. 1084 * Reset idle time and keep-alive timer. 1085 */ 1086 tp->t_rcvtime = ticks; 1087 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1088 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1089 1090 /* 1091 * Process options only when we get SYN/ACK back. The SYN case 1092 * for incoming connections is handled in tcp_syncache. 1093 * XXX this is traditional behavior, may need to be cleaned up. 1094 */ 1095 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 1096 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1097 if (to.to_flags & TOF_SCALE) { 1098 tp->t_flags |= TF_RCVD_SCALE; 1099 tp->requested_s_scale = to.to_requested_s_scale; 1100 } 1101 if (to.to_flags & TOF_TS) { 1102 tp->t_flags |= TF_RCVD_TSTMP; 1103 tp->ts_recent = to.to_tsval; 1104 tp->ts_recent_age = ticks; 1105 } 1106 if (to.to_flags & TOF_MSS) 1107 tcp_mss(tp, to.to_mss); 1108 if (tp->sack_enable) { 1109 if (!(to.to_flags & TOF_SACK)) 1110 tp->sack_enable = 0; 1111 else 1112 tp->t_flags |= TF_SACK_PERMIT; 1113 } 1114 1115 } 1116 1117 /* 1118 * Header prediction: check for the two common cases 1119 * of a uni-directional data xfer. If the packet has 1120 * no control flags, is in-sequence, the window didn't 1121 * change and we're not retransmitting, it's a 1122 * candidate. If the length is zero and the ack moved 1123 * forward, we're the sender side of the xfer. Just 1124 * free the data acked & wake any higher level process 1125 * that was blocked waiting for space. If the length 1126 * is non-zero and the ack didn't move, we're the 1127 * receiver side. If we're getting packets in-order 1128 * (the reassembly queue is empty), add the data to 1129 * the socket buffer and note that we need a delayed ack. 1130 * Make sure that the hidden state-flags are also off. 1131 * Since we check for TCPS_ESTABLISHED above, it can only 1132 * be TH_NEEDSYN. 1133 */ 1134 if (tp->t_state == TCPS_ESTABLISHED && 1135 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1136 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1137 ((to.to_flags & TOF_TS) == 0 || 1138 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1139 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1140 tp->snd_nxt == tp->snd_max) { 1141 1142 /* 1143 * If last ACK falls within this segment's sequence numbers, 1144 * record the timestamp. 1145 * NOTE that the test is modified according to the latest 1146 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1147 */ 1148 if ((to.to_flags & TOF_TS) != 0 && 1149 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1150 tp->ts_recent_age = ticks; 1151 tp->ts_recent = to.to_tsval; 1152 } 1153 1154 if (tlen == 0) { 1155 if (SEQ_GT(th->th_ack, tp->snd_una) && 1156 SEQ_LEQ(th->th_ack, tp->snd_max) && 1157 tp->snd_cwnd >= tp->snd_wnd && 1158 ((!tcp_do_newreno && !tp->sack_enable && 1159 tp->t_dupacks < tcprexmtthresh) || 1160 ((tcp_do_newreno || tp->sack_enable) && 1161 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 && 1162 TAILQ_EMPTY(&tp->snd_holes)))) { 1163 KASSERT(headlocked, ("headlocked")); 1164 INP_INFO_WUNLOCK(&tcbinfo); 1165 headlocked = 0; 1166 /* 1167 * this is a pure ack for outstanding data. 1168 */ 1169 ++tcpstat.tcps_predack; 1170 /* 1171 * "bad retransmit" recovery 1172 */ 1173 if (tp->t_rxtshift == 1 && 1174 ticks < tp->t_badrxtwin) { 1175 ++tcpstat.tcps_sndrexmitbad; 1176 tp->snd_cwnd = tp->snd_cwnd_prev; 1177 tp->snd_ssthresh = 1178 tp->snd_ssthresh_prev; 1179 tp->snd_recover = tp->snd_recover_prev; 1180 if (tp->t_flags & TF_WASFRECOVERY) 1181 ENTER_FASTRECOVERY(tp); 1182 tp->snd_nxt = tp->snd_max; 1183 tp->t_badrxtwin = 0; 1184 } 1185 1186 /* 1187 * Recalculate the transmit timer / rtt. 1188 * 1189 * Some boxes send broken timestamp replies 1190 * during the SYN+ACK phase, ignore 1191 * timestamps of 0 or we could calculate a 1192 * huge RTT and blow up the retransmit timer. 1193 */ 1194 if ((to.to_flags & TOF_TS) != 0 && 1195 to.to_tsecr) { 1196 tcp_xmit_timer(tp, 1197 ticks - to.to_tsecr + 1); 1198 } else if (tp->t_rtttime && 1199 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1200 tcp_xmit_timer(tp, 1201 ticks - tp->t_rtttime); 1202 } 1203 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1204 acked = th->th_ack - tp->snd_una; 1205 tcpstat.tcps_rcvackpack++; 1206 tcpstat.tcps_rcvackbyte += acked; 1207 sbdrop(&so->so_snd, acked); 1208 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1209 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1210 tp->snd_recover = th->th_ack - 1; 1211 tp->snd_una = th->th_ack; 1212 /* 1213 * pull snd_wl2 up to prevent seq wrap relative 1214 * to th_ack. 1215 */ 1216 tp->snd_wl2 = th->th_ack; 1217 tp->t_dupacks = 0; 1218 m_freem(m); 1219 ND6_HINT(tp); /* some progress has been done */ 1220 1221 /* 1222 * If all outstanding data are acked, stop 1223 * retransmit timer, otherwise restart timer 1224 * using current (possibly backed-off) value. 1225 * If process is waiting for space, 1226 * wakeup/selwakeup/signal. If data 1227 * are ready to send, let tcp_output 1228 * decide between more output or persist. 1229 1230#ifdef TCPDEBUG 1231 if (so->so_options & SO_DEBUG) 1232 tcp_trace(TA_INPUT, ostate, tp, 1233 (void *)tcp_saveipgen, 1234 &tcp_savetcp, 0); 1235#endif 1236 */ 1237 if (tp->snd_una == tp->snd_max) 1238 callout_stop(tp->tt_rexmt); 1239 else if (!callout_active(tp->tt_persist)) 1240 callout_reset(tp->tt_rexmt, 1241 tp->t_rxtcur, 1242 tcp_timer_rexmt, tp); 1243 1244 sowwakeup(so); 1245 if (so->so_snd.sb_cc) 1246 (void) tcp_output(tp); 1247 goto check_delack; 1248 } 1249 } else if (th->th_ack == tp->snd_una && 1250 LIST_EMPTY(&tp->t_segq) && 1251 tlen <= sbspace(&so->so_rcv)) { 1252 KASSERT(headlocked, ("headlocked")); 1253 INP_INFO_WUNLOCK(&tcbinfo); 1254 headlocked = 0; 1255 /* 1256 * this is a pure, in-sequence data packet 1257 * with nothing on the reassembly queue and 1258 * we have enough buffer space to take it. 1259 */ 1260 /* Clean receiver SACK report if present */ 1261 if (tp->sack_enable && tp->rcv_numsacks) 1262 tcp_clean_sackreport(tp); 1263 ++tcpstat.tcps_preddat; 1264 tp->rcv_nxt += tlen; 1265 /* 1266 * Pull snd_wl1 up to prevent seq wrap relative to 1267 * th_seq. 1268 */ 1269 tp->snd_wl1 = th->th_seq; 1270 /* 1271 * Pull rcv_up up to prevent seq wrap relative to 1272 * rcv_nxt. 1273 */ 1274 tp->rcv_up = tp->rcv_nxt; 1275 tcpstat.tcps_rcvpack++; 1276 tcpstat.tcps_rcvbyte += tlen; 1277 ND6_HINT(tp); /* some progress has been done */ 1278 /* 1279#ifdef TCPDEBUG 1280 if (so->so_options & SO_DEBUG) 1281 tcp_trace(TA_INPUT, ostate, tp, 1282 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1283#endif 1284 * Add data to socket buffer. 1285 */ 1286 SOCKBUF_LOCK(&so->so_rcv); 1287 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1288 m_freem(m); 1289 } else { 1290 m_adj(m, drop_hdrlen); /* delayed header drop */ 1291 sbappendstream_locked(&so->so_rcv, m); 1292 } 1293 sorwakeup_locked(so); 1294 if (DELAY_ACK(tp)) { 1295 tp->t_flags |= TF_DELACK; 1296 } else { 1297 tp->t_flags |= TF_ACKNOW; 1298 tcp_output(tp); 1299 } 1300 goto check_delack; 1301 } 1302 } 1303 1304 /* 1305 * Calculate amount of space in receive window, 1306 * and then do TCP input processing. 1307 * Receive window is amount of space in rcv queue, 1308 * but not less than advertised window. 1309 */ 1310 { int win; 1311 1312 win = sbspace(&so->so_rcv); 1313 if (win < 0) 1314 win = 0; 1315 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1316 } 1317 1318 switch (tp->t_state) { 1319 1320 /* 1321 * If the state is SYN_RECEIVED: 1322 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1323 */ 1324 case TCPS_SYN_RECEIVED: 1325 if ((thflags & TH_ACK) && 1326 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1327 SEQ_GT(th->th_ack, tp->snd_max))) { 1328 rstreason = BANDLIM_RST_OPENPORT; 1329 goto dropwithreset; 1330 } 1331 break; 1332 1333 /* 1334 * If the state is SYN_SENT: 1335 * if seg contains an ACK, but not for our SYN, drop the input. 1336 * if seg contains a RST, then drop the connection. 1337 * if seg does not contain SYN, then drop it. 1338 * Otherwise this is an acceptable SYN segment 1339 * initialize tp->rcv_nxt and tp->irs 1340 * if seg contains ack then advance tp->snd_una 1341 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1342 * arrange for segment to be acked (eventually) 1343 * continue processing rest of data/controls, beginning with URG 1344 */ 1345 case TCPS_SYN_SENT: 1346 if ((thflags & TH_ACK) && 1347 (SEQ_LEQ(th->th_ack, tp->iss) || 1348 SEQ_GT(th->th_ack, tp->snd_max))) { 1349 rstreason = BANDLIM_UNLIMITED; 1350 goto dropwithreset; 1351 } 1352 if (thflags & TH_RST) { 1353 if (thflags & TH_ACK) { 1354 KASSERT(headlocked, ("tcp_input: after_listen" 1355 ": tcp_drop.2: head not locked")); 1356 tp = tcp_drop(tp, ECONNREFUSED); 1357 } 1358 goto drop; 1359 } 1360 if ((thflags & TH_SYN) == 0) 1361 goto drop; 1362 tp->snd_wnd = th->th_win; /* initial send window */ 1363 1364 tp->irs = th->th_seq; 1365 tcp_rcvseqinit(tp); 1366 if (thflags & TH_ACK) { 1367 tcpstat.tcps_connects++; 1368 soisconnected(so); 1369#ifdef MAC 1370 SOCK_LOCK(so); 1371 mac_set_socket_peer_from_mbuf(m, so); 1372 SOCK_UNLOCK(so); 1373#endif 1374 /* Do window scaling on this connection? */ 1375 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1376 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1377 tp->snd_scale = tp->requested_s_scale; 1378 tp->rcv_scale = tp->request_r_scale; 1379 } 1380 tp->rcv_adv += tp->rcv_wnd; 1381 tp->snd_una++; /* SYN is acked */ 1382 /* 1383 * If there's data, delay ACK; if there's also a FIN 1384 * ACKNOW will be turned on later. 1385 */ 1386 if (DELAY_ACK(tp) && tlen != 0) 1387 callout_reset(tp->tt_delack, tcp_delacktime, 1388 tcp_timer_delack, tp); 1389 else 1390 tp->t_flags |= TF_ACKNOW; 1391 /* 1392 * Received <SYN,ACK> in SYN_SENT[*] state. 1393 * Transitions: 1394 * SYN_SENT --> ESTABLISHED 1395 * SYN_SENT* --> FIN_WAIT_1 1396 */ 1397 tp->t_starttime = ticks; 1398 if (tp->t_flags & TF_NEEDFIN) { 1399 tp->t_state = TCPS_FIN_WAIT_1; 1400 tp->t_flags &= ~TF_NEEDFIN; 1401 thflags &= ~TH_SYN; 1402 } else { 1403 tp->t_state = TCPS_ESTABLISHED; 1404 callout_reset(tp->tt_keep, tcp_keepidle, 1405 tcp_timer_keep, tp); 1406 } 1407 } else { 1408 /* 1409 * Received initial SYN in SYN-SENT[*] state => 1410 * simultaneous open. If segment contains CC option 1411 * and there is a cached CC, apply TAO test. 1412 * If it succeeds, connection is * half-synchronized. 1413 * Otherwise, do 3-way handshake: 1414 * SYN-SENT -> SYN-RECEIVED 1415 * SYN-SENT* -> SYN-RECEIVED* 1416 * If there was no CC option, clear cached CC value. 1417 */ 1418 tp->t_flags |= TF_ACKNOW; 1419 callout_stop(tp->tt_rexmt); 1420 tp->t_state = TCPS_SYN_RECEIVED; 1421 } 1422 1423trimthenstep6: 1424 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1425 "locked")); 1426 INP_LOCK_ASSERT(inp); 1427 1428 /* 1429 * Advance th->th_seq to correspond to first data byte. 1430 * If data, trim to stay within window, 1431 * dropping FIN if necessary. 1432 */ 1433 th->th_seq++; 1434 if (tlen > tp->rcv_wnd) { 1435 todrop = tlen - tp->rcv_wnd; 1436 m_adj(m, -todrop); 1437 tlen = tp->rcv_wnd; 1438 thflags &= ~TH_FIN; 1439 tcpstat.tcps_rcvpackafterwin++; 1440 tcpstat.tcps_rcvbyteafterwin += todrop; 1441 } 1442 tp->snd_wl1 = th->th_seq - 1; 1443 tp->rcv_up = th->th_seq; 1444 /* 1445 * Client side of transaction: already sent SYN and data. 1446 * If the remote host used T/TCP to validate the SYN, 1447 * our data will be ACK'd; if so, enter normal data segment 1448 * processing in the middle of step 5, ack processing. 1449 * Otherwise, goto step 6. 1450 */ 1451 if (thflags & TH_ACK) 1452 goto process_ACK; 1453 1454 goto step6; 1455 1456 /* 1457 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1458 * do normal processing. 1459 * 1460 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1461 */ 1462 case TCPS_LAST_ACK: 1463 case TCPS_CLOSING: 1464 case TCPS_TIME_WAIT: 1465 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1466 break; /* continue normal processing */ 1467 } 1468 1469 /* 1470 * States other than LISTEN or SYN_SENT. 1471 * First check the RST flag and sequence number since reset segments 1472 * are exempt from the timestamp and connection count tests. This 1473 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1474 * below which allowed reset segments in half the sequence space 1475 * to fall though and be processed (which gives forged reset 1476 * segments with a random sequence number a 50 percent chance of 1477 * killing a connection). 1478 * Then check timestamp, if present. 1479 * Then check the connection count, if present. 1480 * Then check that at least some bytes of segment are within 1481 * receive window. If segment begins before rcv_nxt, 1482 * drop leading data (and SYN); if nothing left, just ack. 1483 * 1484 * 1485 * If the RST bit is set, check the sequence number to see 1486 * if this is a valid reset segment. 1487 * RFC 793 page 37: 1488 * In all states except SYN-SENT, all reset (RST) segments 1489 * are validated by checking their SEQ-fields. A reset is 1490 * valid if its sequence number is in the window. 1491 * Note: this does not take into account delayed ACKs, so 1492 * we should test against last_ack_sent instead of rcv_nxt. 1493 * The sequence number in the reset segment is normally an 1494 * echo of our outgoing acknowlegement numbers, but some hosts 1495 * send a reset with the sequence number at the rightmost edge 1496 * of our receive window, and we have to handle this case. 1497 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1498 * that brute force RST attacks are possible. To combat this, 1499 * we use a much stricter check while in the ESTABLISHED state, 1500 * only accepting RSTs where the sequence number is equal to 1501 * last_ack_sent. In all other states (the states in which a 1502 * RST is more likely), the more permissive check is used. 1503 * If we have multiple segments in flight, the intial reset 1504 * segment sequence numbers will be to the left of last_ack_sent, 1505 * but they will eventually catch up. 1506 * In any case, it never made sense to trim reset segments to 1507 * fit the receive window since RFC 1122 says: 1508 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1509 * 1510 * A TCP SHOULD allow a received RST segment to include data. 1511 * 1512 * DISCUSSION 1513 * It has been suggested that a RST segment could contain 1514 * ASCII text that encoded and explained the cause of the 1515 * RST. No standard has yet been established for such 1516 * data. 1517 * 1518 * If the reset segment passes the sequence number test examine 1519 * the state: 1520 * SYN_RECEIVED STATE: 1521 * If passive open, return to LISTEN state. 1522 * If active open, inform user that connection was refused. 1523 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1524 * Inform user that connection was reset, and close tcb. 1525 * CLOSING, LAST_ACK STATES: 1526 * Close the tcb. 1527 * TIME_WAIT STATE: 1528 * Drop the segment - see Stevens, vol. 2, p. 964 and 1529 * RFC 1337. 1530 */ 1531 if (thflags & TH_RST) { 1532 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1533 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1534 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1535 switch (tp->t_state) { 1536 1537 case TCPS_SYN_RECEIVED: 1538 so->so_error = ECONNREFUSED; 1539 goto close; 1540 1541 case TCPS_ESTABLISHED: 1542 if (tp->last_ack_sent != th->th_seq && 1543 tcp_insecure_rst == 0) { 1544 tcpstat.tcps_badrst++; 1545 goto drop; 1546 } 1547 case TCPS_FIN_WAIT_1: 1548 case TCPS_FIN_WAIT_2: 1549 case TCPS_CLOSE_WAIT: 1550 so->so_error = ECONNRESET; 1551 close: 1552 tp->t_state = TCPS_CLOSED; 1553 tcpstat.tcps_drops++; 1554 KASSERT(headlocked, ("tcp_input: " 1555 "trimthenstep6: tcp_close: head not " 1556 "locked")); 1557 tp = tcp_close(tp); 1558 break; 1559 1560 case TCPS_CLOSING: 1561 case TCPS_LAST_ACK: 1562 KASSERT(headlocked, ("trimthenstep6: " 1563 "tcp_close.2: head not locked")); 1564 tp = tcp_close(tp); 1565 break; 1566 1567 case TCPS_TIME_WAIT: 1568 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1569 ("timewait")); 1570 break; 1571 } 1572 } 1573 goto drop; 1574 } 1575 1576 /* 1577 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1578 * and it's less than ts_recent, drop it. 1579 */ 1580 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1581 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1582 1583 /* Check to see if ts_recent is over 24 days old. */ 1584 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1585 /* 1586 * Invalidate ts_recent. If this segment updates 1587 * ts_recent, the age will be reset later and ts_recent 1588 * will get a valid value. If it does not, setting 1589 * ts_recent to zero will at least satisfy the 1590 * requirement that zero be placed in the timestamp 1591 * echo reply when ts_recent isn't valid. The 1592 * age isn't reset until we get a valid ts_recent 1593 * because we don't want out-of-order segments to be 1594 * dropped when ts_recent is old. 1595 */ 1596 tp->ts_recent = 0; 1597 } else { 1598 tcpstat.tcps_rcvduppack++; 1599 tcpstat.tcps_rcvdupbyte += tlen; 1600 tcpstat.tcps_pawsdrop++; 1601 if (tlen) 1602 goto dropafterack; 1603 goto drop; 1604 } 1605 } 1606 1607 /* 1608 * In the SYN-RECEIVED state, validate that the packet belongs to 1609 * this connection before trimming the data to fit the receive 1610 * window. Check the sequence number versus IRS since we know 1611 * the sequence numbers haven't wrapped. This is a partial fix 1612 * for the "LAND" DoS attack. 1613 */ 1614 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1615 rstreason = BANDLIM_RST_OPENPORT; 1616 goto dropwithreset; 1617 } 1618 1619 todrop = tp->rcv_nxt - th->th_seq; 1620 if (todrop > 0) { 1621 if (thflags & TH_SYN) { 1622 thflags &= ~TH_SYN; 1623 th->th_seq++; 1624 if (th->th_urp > 1) 1625 th->th_urp--; 1626 else 1627 thflags &= ~TH_URG; 1628 todrop--; 1629 } 1630 /* 1631 * Following if statement from Stevens, vol. 2, p. 960. 1632 */ 1633 if (todrop > tlen 1634 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1635 /* 1636 * Any valid FIN must be to the left of the window. 1637 * At this point the FIN must be a duplicate or out 1638 * of sequence; drop it. 1639 */ 1640 thflags &= ~TH_FIN; 1641 1642 /* 1643 * Send an ACK to resynchronize and drop any data. 1644 * But keep on processing for RST or ACK. 1645 */ 1646 tp->t_flags |= TF_ACKNOW; 1647 todrop = tlen; 1648 tcpstat.tcps_rcvduppack++; 1649 tcpstat.tcps_rcvdupbyte += todrop; 1650 } else { 1651 tcpstat.tcps_rcvpartduppack++; 1652 tcpstat.tcps_rcvpartdupbyte += todrop; 1653 } 1654 drop_hdrlen += todrop; /* drop from the top afterwards */ 1655 th->th_seq += todrop; 1656 tlen -= todrop; 1657 if (th->th_urp > todrop) 1658 th->th_urp -= todrop; 1659 else { 1660 thflags &= ~TH_URG; 1661 th->th_urp = 0; 1662 } 1663 } 1664 1665 /* 1666 * If new data are received on a connection after the 1667 * user processes are gone, then RST the other end. 1668 */ 1669 if ((so->so_state & SS_NOFDREF) && 1670 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1671 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not " 1672 "locked")); 1673 tp = tcp_close(tp); 1674 tcpstat.tcps_rcvafterclose++; 1675 rstreason = BANDLIM_UNLIMITED; 1676 goto dropwithreset; 1677 } 1678 1679 /* 1680 * If segment ends after window, drop trailing data 1681 * (and PUSH and FIN); if nothing left, just ACK. 1682 */ 1683 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1684 if (todrop > 0) { 1685 tcpstat.tcps_rcvpackafterwin++; 1686 if (todrop >= tlen) { 1687 tcpstat.tcps_rcvbyteafterwin += tlen; 1688 /* 1689 * If a new connection request is received 1690 * while in TIME_WAIT, drop the old connection 1691 * and start over if the sequence numbers 1692 * are above the previous ones. 1693 */ 1694 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1695 if (thflags & TH_SYN && 1696 tp->t_state == TCPS_TIME_WAIT && 1697 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1698 KASSERT(headlocked, ("trimthenstep6: " 1699 "tcp_close.4: head not locked")); 1700 tp = tcp_close(tp); 1701 goto findpcb; 1702 } 1703 /* 1704 * If window is closed can only take segments at 1705 * window edge, and have to drop data and PUSH from 1706 * incoming segments. Continue processing, but 1707 * remember to ack. Otherwise, drop segment 1708 * and ack. 1709 */ 1710 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1711 tp->t_flags |= TF_ACKNOW; 1712 tcpstat.tcps_rcvwinprobe++; 1713 } else 1714 goto dropafterack; 1715 } else 1716 tcpstat.tcps_rcvbyteafterwin += todrop; 1717 m_adj(m, -todrop); 1718 tlen -= todrop; 1719 thflags &= ~(TH_PUSH|TH_FIN); 1720 } 1721 1722 /* 1723 * If last ACK falls within this segment's sequence numbers, 1724 * record its timestamp. 1725 * NOTE: 1726 * 1) That the test incorporates suggestions from the latest 1727 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1728 * 2) That updating only on newer timestamps interferes with 1729 * our earlier PAWS tests, so this check should be solely 1730 * predicated on the sequence space of this segment. 1731 * 3) That we modify the segment boundary check to be 1732 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1733 * instead of RFC1323's 1734 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1735 * This modified check allows us to overcome RFC1323's 1736 * limitations as described in Stevens TCP/IP Illustrated 1737 * Vol. 2 p.869. In such cases, we can still calculate the 1738 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1739 */ 1740 if ((to.to_flags & TOF_TS) != 0 && 1741 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1742 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1743 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1744 tp->ts_recent_age = ticks; 1745 tp->ts_recent = to.to_tsval; 1746 } 1747 1748 /* 1749 * If a SYN is in the window, then this is an 1750 * error and we send an RST and drop the connection. 1751 */ 1752 if (thflags & TH_SYN) { 1753 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: " 1754 "head not locked")); 1755 tp = tcp_drop(tp, ECONNRESET); 1756 rstreason = BANDLIM_UNLIMITED; 1757 goto drop; 1758 } 1759 1760 /* 1761 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1762 * flag is on (half-synchronized state), then queue data for 1763 * later processing; else drop segment and return. 1764 */ 1765 if ((thflags & TH_ACK) == 0) { 1766 if (tp->t_state == TCPS_SYN_RECEIVED || 1767 (tp->t_flags & TF_NEEDSYN)) 1768 goto step6; 1769 else 1770 goto drop; 1771 } 1772 1773 /* 1774 * Ack processing. 1775 */ 1776 switch (tp->t_state) { 1777 1778 /* 1779 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1780 * ESTABLISHED state and continue processing. 1781 * The ACK was checked above. 1782 */ 1783 case TCPS_SYN_RECEIVED: 1784 1785 tcpstat.tcps_connects++; 1786 soisconnected(so); 1787 /* Do window scaling? */ 1788 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1789 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1790 tp->snd_scale = tp->requested_s_scale; 1791 tp->rcv_scale = tp->request_r_scale; 1792 } 1793 /* 1794 * Make transitions: 1795 * SYN-RECEIVED -> ESTABLISHED 1796 * SYN-RECEIVED* -> FIN-WAIT-1 1797 */ 1798 tp->t_starttime = ticks; 1799 if (tp->t_flags & TF_NEEDFIN) { 1800 tp->t_state = TCPS_FIN_WAIT_1; 1801 tp->t_flags &= ~TF_NEEDFIN; 1802 } else { 1803 tp->t_state = TCPS_ESTABLISHED; 1804 callout_reset(tp->tt_keep, tcp_keepidle, 1805 tcp_timer_keep, tp); 1806 } 1807 /* 1808 * If segment contains data or ACK, will call tcp_reass() 1809 * later; if not, do so now to pass queued data to user. 1810 */ 1811 if (tlen == 0 && (thflags & TH_FIN) == 0) 1812 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1813 (struct mbuf *)0); 1814 tp->snd_wl1 = th->th_seq - 1; 1815 /* FALLTHROUGH */ 1816 1817 /* 1818 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1819 * ACKs. If the ack is in the range 1820 * tp->snd_una < th->th_ack <= tp->snd_max 1821 * then advance tp->snd_una to th->th_ack and drop 1822 * data from the retransmission queue. If this ACK reflects 1823 * more up to date window information we update our window information. 1824 */ 1825 case TCPS_ESTABLISHED: 1826 case TCPS_FIN_WAIT_1: 1827 case TCPS_FIN_WAIT_2: 1828 case TCPS_CLOSE_WAIT: 1829 case TCPS_CLOSING: 1830 case TCPS_LAST_ACK: 1831 case TCPS_TIME_WAIT: 1832 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1833 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1834 tcpstat.tcps_rcvacktoomuch++; 1835 goto dropafterack; 1836 } 1837 if (tp->sack_enable && 1838 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) 1839 tcp_sack_doack(tp, &to, th->th_ack); 1840 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1841 if (tlen == 0 && tiwin == tp->snd_wnd) { 1842 tcpstat.tcps_rcvdupack++; 1843 /* 1844 * If we have outstanding data (other than 1845 * a window probe), this is a completely 1846 * duplicate ack (ie, window info didn't 1847 * change), the ack is the biggest we've 1848 * seen and we've seen exactly our rexmt 1849 * threshhold of them, assume a packet 1850 * has been dropped and retransmit it. 1851 * Kludge snd_nxt & the congestion 1852 * window so we send only this one 1853 * packet. 1854 * 1855 * We know we're losing at the current 1856 * window size so do congestion avoidance 1857 * (set ssthresh to half the current window 1858 * and pull our congestion window back to 1859 * the new ssthresh). 1860 * 1861 * Dup acks mean that packets have left the 1862 * network (they're now cached at the receiver) 1863 * so bump cwnd by the amount in the receiver 1864 * to keep a constant cwnd packets in the 1865 * network. 1866 */ 1867 if (!callout_active(tp->tt_rexmt) || 1868 th->th_ack != tp->snd_una) 1869 tp->t_dupacks = 0; 1870 else if (++tp->t_dupacks > tcprexmtthresh || 1871 ((tcp_do_newreno || tp->sack_enable) && 1872 IN_FASTRECOVERY(tp))) { 1873 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1874 int awnd; 1875 1876 /* 1877 * Compute the amount of data in flight first. 1878 * We can inject new data into the pipe iff 1879 * we have less than 1/2 the original window's 1880 * worth of data in flight. 1881 */ 1882 awnd = (tp->snd_nxt - tp->snd_fack) + 1883 tp->sackhint.sack_bytes_rexmit; 1884 if (awnd < tp->snd_ssthresh) { 1885 tp->snd_cwnd += tp->t_maxseg; 1886 if (tp->snd_cwnd > tp->snd_ssthresh) 1887 tp->snd_cwnd = tp->snd_ssthresh; 1888 } 1889 } else 1890 tp->snd_cwnd += tp->t_maxseg; 1891 (void) tcp_output(tp); 1892 goto drop; 1893 } else if (tp->t_dupacks == tcprexmtthresh) { 1894 tcp_seq onxt = tp->snd_nxt; 1895 u_int win; 1896 1897 /* 1898 * If we're doing sack, check to 1899 * see if we're already in sack 1900 * recovery. If we're not doing sack, 1901 * check to see if we're in newreno 1902 * recovery. 1903 */ 1904 if (tp->sack_enable) { 1905 if (IN_FASTRECOVERY(tp)) { 1906 tp->t_dupacks = 0; 1907 break; 1908 } 1909 } else if (tcp_do_newreno) { 1910 if (SEQ_LEQ(th->th_ack, 1911 tp->snd_recover)) { 1912 tp->t_dupacks = 0; 1913 break; 1914 } 1915 } 1916 win = min(tp->snd_wnd, tp->snd_cwnd) / 1917 2 / tp->t_maxseg; 1918 if (win < 2) 1919 win = 2; 1920 tp->snd_ssthresh = win * tp->t_maxseg; 1921 ENTER_FASTRECOVERY(tp); 1922 tp->snd_recover = tp->snd_max; 1923 callout_stop(tp->tt_rexmt); 1924 tp->t_rtttime = 0; 1925 if (tp->sack_enable) { 1926 tcpstat.tcps_sack_recovery_episode++; 1927 tp->sack_newdata = tp->snd_nxt; 1928 tp->snd_cwnd = tp->t_maxseg; 1929 (void) tcp_output(tp); 1930 goto drop; 1931 } 1932 tp->snd_nxt = th->th_ack; 1933 tp->snd_cwnd = tp->t_maxseg; 1934 (void) tcp_output(tp); 1935 KASSERT(tp->snd_limited <= 2, 1936 ("tp->snd_limited too big")); 1937 tp->snd_cwnd = tp->snd_ssthresh + 1938 tp->t_maxseg * 1939 (tp->t_dupacks - tp->snd_limited); 1940 if (SEQ_GT(onxt, tp->snd_nxt)) 1941 tp->snd_nxt = onxt; 1942 goto drop; 1943 } else if (tcp_do_rfc3042) { 1944 u_long oldcwnd = tp->snd_cwnd; 1945 tcp_seq oldsndmax = tp->snd_max; 1946 u_int sent; 1947 1948 KASSERT(tp->t_dupacks == 1 || 1949 tp->t_dupacks == 2, 1950 ("dupacks not 1 or 2")); 1951 if (tp->t_dupacks == 1) 1952 tp->snd_limited = 0; 1953 tp->snd_cwnd = 1954 (tp->snd_nxt - tp->snd_una) + 1955 (tp->t_dupacks - tp->snd_limited) * 1956 tp->t_maxseg; 1957 (void) tcp_output(tp); 1958 sent = tp->snd_max - oldsndmax; 1959 if (sent > tp->t_maxseg) { 1960 KASSERT((tp->t_dupacks == 2 && 1961 tp->snd_limited == 0) || 1962 (sent == tp->t_maxseg + 1 && 1963 tp->t_flags & TF_SENTFIN), 1964 ("sent too much")); 1965 tp->snd_limited = 2; 1966 } else if (sent > 0) 1967 ++tp->snd_limited; 1968 tp->snd_cwnd = oldcwnd; 1969 goto drop; 1970 } 1971 } else 1972 tp->t_dupacks = 0; 1973 break; 1974 } 1975 1976 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1977 1978 /* 1979 * If the congestion window was inflated to account 1980 * for the other side's cached packets, retract it. 1981 */ 1982 if (tcp_do_newreno || tp->sack_enable) { 1983 if (IN_FASTRECOVERY(tp)) { 1984 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1985 if (tp->sack_enable) 1986 tcp_sack_partialack(tp, th); 1987 else 1988 tcp_newreno_partial_ack(tp, th); 1989 } else { 1990 /* 1991 * Out of fast recovery. 1992 * Window inflation should have left us 1993 * with approximately snd_ssthresh 1994 * outstanding data. 1995 * But in case we would be inclined to 1996 * send a burst, better to do it via 1997 * the slow start mechanism. 1998 */ 1999 if (SEQ_GT(th->th_ack + 2000 tp->snd_ssthresh, 2001 tp->snd_max)) 2002 tp->snd_cwnd = tp->snd_max - 2003 th->th_ack + 2004 tp->t_maxseg; 2005 else 2006 tp->snd_cwnd = tp->snd_ssthresh; 2007 } 2008 } 2009 } else { 2010 if (tp->t_dupacks >= tcprexmtthresh && 2011 tp->snd_cwnd > tp->snd_ssthresh) 2012 tp->snd_cwnd = tp->snd_ssthresh; 2013 } 2014 tp->t_dupacks = 0; 2015 /* 2016 * If we reach this point, ACK is not a duplicate, 2017 * i.e., it ACKs something we sent. 2018 */ 2019 if (tp->t_flags & TF_NEEDSYN) { 2020 /* 2021 * T/TCP: Connection was half-synchronized, and our 2022 * SYN has been ACK'd (so connection is now fully 2023 * synchronized). Go to non-starred state, 2024 * increment snd_una for ACK of SYN, and check if 2025 * we can do window scaling. 2026 */ 2027 tp->t_flags &= ~TF_NEEDSYN; 2028 tp->snd_una++; 2029 /* Do window scaling? */ 2030 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2031 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2032 tp->snd_scale = tp->requested_s_scale; 2033 tp->rcv_scale = tp->request_r_scale; 2034 } 2035 } 2036 2037process_ACK: 2038 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 2039 "locked")); 2040 INP_LOCK_ASSERT(inp); 2041 2042 acked = th->th_ack - tp->snd_una; 2043 tcpstat.tcps_rcvackpack++; 2044 tcpstat.tcps_rcvackbyte += acked; 2045 2046 /* 2047 * If we just performed our first retransmit, and the ACK 2048 * arrives within our recovery window, then it was a mistake 2049 * to do the retransmit in the first place. Recover our 2050 * original cwnd and ssthresh, and proceed to transmit where 2051 * we left off. 2052 */ 2053 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2054 ++tcpstat.tcps_sndrexmitbad; 2055 tp->snd_cwnd = tp->snd_cwnd_prev; 2056 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2057 tp->snd_recover = tp->snd_recover_prev; 2058 if (tp->t_flags & TF_WASFRECOVERY) 2059 ENTER_FASTRECOVERY(tp); 2060 tp->snd_nxt = tp->snd_max; 2061 tp->t_badrxtwin = 0; /* XXX probably not required */ 2062 } 2063 2064 /* 2065 * If we have a timestamp reply, update smoothed 2066 * round trip time. If no timestamp is present but 2067 * transmit timer is running and timed sequence 2068 * number was acked, update smoothed round trip time. 2069 * Since we now have an rtt measurement, cancel the 2070 * timer backoff (cf., Phil Karn's retransmit alg.). 2071 * Recompute the initial retransmit timer. 2072 * 2073 * Some boxes send broken timestamp replies 2074 * during the SYN+ACK phase, ignore 2075 * timestamps of 0 or we could calculate a 2076 * huge RTT and blow up the retransmit timer. 2077 */ 2078 if ((to.to_flags & TOF_TS) != 0 && 2079 to.to_tsecr) { 2080 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2081 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2082 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2083 } 2084 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2085 2086 /* 2087 * If all outstanding data is acked, stop retransmit 2088 * timer and remember to restart (more output or persist). 2089 * If there is more data to be acked, restart retransmit 2090 * timer, using current (possibly backed-off) value. 2091 */ 2092 if (th->th_ack == tp->snd_max) { 2093 callout_stop(tp->tt_rexmt); 2094 needoutput = 1; 2095 } else if (!callout_active(tp->tt_persist)) 2096 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2097 tcp_timer_rexmt, tp); 2098 2099 /* 2100 * If no data (only SYN) was ACK'd, 2101 * skip rest of ACK processing. 2102 */ 2103 if (acked == 0) 2104 goto step6; 2105 2106 /* 2107 * When new data is acked, open the congestion window. 2108 * If the window gives us less than ssthresh packets 2109 * in flight, open exponentially (maxseg per packet). 2110 * Otherwise open linearly: maxseg per window 2111 * (maxseg^2 / cwnd per packet). 2112 */ 2113 if ((!tcp_do_newreno && !tp->sack_enable) || 2114 !IN_FASTRECOVERY(tp)) { 2115 register u_int cw = tp->snd_cwnd; 2116 register u_int incr = tp->t_maxseg; 2117 if (cw > tp->snd_ssthresh) 2118 incr = incr * incr / cw; 2119 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2120 } 2121 SOCKBUF_LOCK(&so->so_snd); 2122 if (acked > so->so_snd.sb_cc) { 2123 tp->snd_wnd -= so->so_snd.sb_cc; 2124 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2125 ourfinisacked = 1; 2126 } else { 2127 sbdrop_locked(&so->so_snd, acked); 2128 tp->snd_wnd -= acked; 2129 ourfinisacked = 0; 2130 } 2131 sowwakeup_locked(so); 2132 /* detect una wraparound */ 2133 if ((tcp_do_newreno || tp->sack_enable) && 2134 !IN_FASTRECOVERY(tp) && 2135 SEQ_GT(tp->snd_una, tp->snd_recover) && 2136 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2137 tp->snd_recover = th->th_ack - 1; 2138 if ((tcp_do_newreno || tp->sack_enable) && 2139 IN_FASTRECOVERY(tp) && 2140 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2141 EXIT_FASTRECOVERY(tp); 2142 tp->snd_una = th->th_ack; 2143 if (tp->sack_enable) { 2144 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2145 tp->snd_recover = tp->snd_una; 2146 } 2147 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2148 tp->snd_nxt = tp->snd_una; 2149 2150 switch (tp->t_state) { 2151 2152 /* 2153 * In FIN_WAIT_1 STATE in addition to the processing 2154 * for the ESTABLISHED state if our FIN is now acknowledged 2155 * then enter FIN_WAIT_2. 2156 */ 2157 case TCPS_FIN_WAIT_1: 2158 if (ourfinisacked) { 2159 /* 2160 * If we can't receive any more 2161 * data, then closing user can proceed. 2162 * Starting the timer is contrary to the 2163 * specification, but if we don't get a FIN 2164 * we'll hang forever. 2165 */ 2166 /* XXXjl 2167 * we should release the tp also, and use a 2168 * compressed state. 2169 */ 2170 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2171 soisdisconnected(so); 2172 callout_reset(tp->tt_2msl, tcp_maxidle, 2173 tcp_timer_2msl, tp); 2174 } 2175 tp->t_state = TCPS_FIN_WAIT_2; 2176 } 2177 break; 2178 2179 /* 2180 * In CLOSING STATE in addition to the processing for 2181 * the ESTABLISHED state if the ACK acknowledges our FIN 2182 * then enter the TIME-WAIT state, otherwise ignore 2183 * the segment. 2184 */ 2185 case TCPS_CLOSING: 2186 if (ourfinisacked) { 2187 KASSERT(headlocked, ("tcp_input: process_ACK: " 2188 "head not locked")); 2189 tcp_twstart(tp); 2190 INP_INFO_WUNLOCK(&tcbinfo); 2191 m_freem(m); 2192 return; 2193 } 2194 break; 2195 2196 /* 2197 * In LAST_ACK, we may still be waiting for data to drain 2198 * and/or to be acked, as well as for the ack of our FIN. 2199 * If our FIN is now acknowledged, delete the TCB, 2200 * enter the closed state and return. 2201 */ 2202 case TCPS_LAST_ACK: 2203 if (ourfinisacked) { 2204 KASSERT(headlocked, ("tcp_input: process_ACK:" 2205 " tcp_close: head not locked")); 2206 tp = tcp_close(tp); 2207 goto drop; 2208 } 2209 break; 2210 2211 /* 2212 * In TIME_WAIT state the only thing that should arrive 2213 * is a retransmission of the remote FIN. Acknowledge 2214 * it and restart the finack timer. 2215 */ 2216 case TCPS_TIME_WAIT: 2217 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2218 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2219 tcp_timer_2msl, tp); 2220 goto dropafterack; 2221 } 2222 } 2223 2224step6: 2225 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2226 INP_LOCK_ASSERT(inp); 2227 2228 /* 2229 * Update window information. 2230 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2231 */ 2232 if ((thflags & TH_ACK) && 2233 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2234 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2235 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2236 /* keep track of pure window updates */ 2237 if (tlen == 0 && 2238 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2239 tcpstat.tcps_rcvwinupd++; 2240 tp->snd_wnd = tiwin; 2241 tp->snd_wl1 = th->th_seq; 2242 tp->snd_wl2 = th->th_ack; 2243 if (tp->snd_wnd > tp->max_sndwnd) 2244 tp->max_sndwnd = tp->snd_wnd; 2245 needoutput = 1; 2246 } 2247 2248 /* 2249 * Process segments with URG. 2250 */ 2251 if ((thflags & TH_URG) && th->th_urp && 2252 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2253 /* 2254 * This is a kludge, but if we receive and accept 2255 * random urgent pointers, we'll crash in 2256 * soreceive. It's hard to imagine someone 2257 * actually wanting to send this much urgent data. 2258 */ 2259 SOCKBUF_LOCK(&so->so_rcv); 2260 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2261 th->th_urp = 0; /* XXX */ 2262 thflags &= ~TH_URG; /* XXX */ 2263 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2264 goto dodata; /* XXX */ 2265 } 2266 /* 2267 * If this segment advances the known urgent pointer, 2268 * then mark the data stream. This should not happen 2269 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2270 * a FIN has been received from the remote side. 2271 * In these states we ignore the URG. 2272 * 2273 * According to RFC961 (Assigned Protocols), 2274 * the urgent pointer points to the last octet 2275 * of urgent data. We continue, however, 2276 * to consider it to indicate the first octet 2277 * of data past the urgent section as the original 2278 * spec states (in one of two places). 2279 */ 2280 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2281 tp->rcv_up = th->th_seq + th->th_urp; 2282 so->so_oobmark = so->so_rcv.sb_cc + 2283 (tp->rcv_up - tp->rcv_nxt) - 1; 2284 if (so->so_oobmark == 0) 2285 so->so_rcv.sb_state |= SBS_RCVATMARK; 2286 sohasoutofband(so); 2287 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2288 } 2289 SOCKBUF_UNLOCK(&so->so_rcv); 2290 /* 2291 * Remove out of band data so doesn't get presented to user. 2292 * This can happen independent of advancing the URG pointer, 2293 * but if two URG's are pending at once, some out-of-band 2294 * data may creep in... ick. 2295 */ 2296 if (th->th_urp <= (u_long)tlen && 2297 !(so->so_options & SO_OOBINLINE)) { 2298 /* hdr drop is delayed */ 2299 tcp_pulloutofband(so, th, m, drop_hdrlen); 2300 } 2301 } else { 2302 /* 2303 * If no out of band data is expected, 2304 * pull receive urgent pointer along 2305 * with the receive window. 2306 */ 2307 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2308 tp->rcv_up = tp->rcv_nxt; 2309 } 2310dodata: /* XXX */ 2311 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2312 INP_LOCK_ASSERT(inp); 2313 2314 /* 2315 * Process the segment text, merging it into the TCP sequencing queue, 2316 * and arranging for acknowledgment of receipt if necessary. 2317 * This process logically involves adjusting tp->rcv_wnd as data 2318 * is presented to the user (this happens in tcp_usrreq.c, 2319 * case PRU_RCVD). If a FIN has already been received on this 2320 * connection then we just ignore the text. 2321 */ 2322 if ((tlen || (thflags & TH_FIN)) && 2323 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2324 tcp_seq save_start = th->th_seq; 2325 tcp_seq save_end = th->th_seq + tlen; 2326 m_adj(m, drop_hdrlen); /* delayed header drop */ 2327 /* 2328 * Insert segment which includes th into TCP reassembly queue 2329 * with control block tp. Set thflags to whether reassembly now 2330 * includes a segment with FIN. This handles the common case 2331 * inline (segment is the next to be received on an established 2332 * connection, and the queue is empty), avoiding linkage into 2333 * and removal from the queue and repetition of various 2334 * conversions. 2335 * Set DELACK for segments received in order, but ack 2336 * immediately when segments are out of order (so 2337 * fast retransmit can work). 2338 */ 2339 if (th->th_seq == tp->rcv_nxt && 2340 LIST_EMPTY(&tp->t_segq) && 2341 TCPS_HAVEESTABLISHED(tp->t_state)) { 2342 if (DELAY_ACK(tp)) 2343 tp->t_flags |= TF_DELACK; 2344 else 2345 tp->t_flags |= TF_ACKNOW; 2346 tp->rcv_nxt += tlen; 2347 thflags = th->th_flags & TH_FIN; 2348 tcpstat.tcps_rcvpack++; 2349 tcpstat.tcps_rcvbyte += tlen; 2350 ND6_HINT(tp); 2351 SOCKBUF_LOCK(&so->so_rcv); 2352 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2353 m_freem(m); 2354 else 2355 sbappendstream_locked(&so->so_rcv, m); 2356 sorwakeup_locked(so); 2357 } else { 2358 thflags = tcp_reass(tp, th, &tlen, m); 2359 tp->t_flags |= TF_ACKNOW; 2360 } 2361 if (tlen > 0 && tp->sack_enable) 2362 tcp_update_sack_list(tp, save_start, save_end); 2363 /* 2364 * Note the amount of data that peer has sent into 2365 * our window, in order to estimate the sender's 2366 * buffer size. 2367 */ 2368 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2369 } else { 2370 m_freem(m); 2371 thflags &= ~TH_FIN; 2372 } 2373 2374 /* 2375 * If FIN is received ACK the FIN and let the user know 2376 * that the connection is closing. 2377 */ 2378 if (thflags & TH_FIN) { 2379 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2380 socantrcvmore(so); 2381 /* 2382 * If connection is half-synchronized 2383 * (ie NEEDSYN flag on) then delay ACK, 2384 * so it may be piggybacked when SYN is sent. 2385 * Otherwise, since we received a FIN then no 2386 * more input can be expected, send ACK now. 2387 */ 2388 if (tp->t_flags & TF_NEEDSYN) 2389 tp->t_flags |= TF_DELACK; 2390 else 2391 tp->t_flags |= TF_ACKNOW; 2392 tp->rcv_nxt++; 2393 } 2394 switch (tp->t_state) { 2395 2396 /* 2397 * In SYN_RECEIVED and ESTABLISHED STATES 2398 * enter the CLOSE_WAIT state. 2399 */ 2400 case TCPS_SYN_RECEIVED: 2401 tp->t_starttime = ticks; 2402 /*FALLTHROUGH*/ 2403 case TCPS_ESTABLISHED: 2404 tp->t_state = TCPS_CLOSE_WAIT; 2405 break; 2406 2407 /* 2408 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2409 * enter the CLOSING state. 2410 */ 2411 case TCPS_FIN_WAIT_1: 2412 tp->t_state = TCPS_CLOSING; 2413 break; 2414 2415 /* 2416 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2417 * starting the time-wait timer, turning off the other 2418 * standard timers. 2419 */ 2420 case TCPS_FIN_WAIT_2: 2421 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2422 "TCP_FIN_WAIT_2: head not locked")); 2423 tcp_twstart(tp); 2424 INP_INFO_WUNLOCK(&tcbinfo); 2425 return; 2426 2427 /* 2428 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2429 */ 2430 case TCPS_TIME_WAIT: 2431 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2432 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2433 tcp_timer_2msl, tp); 2434 break; 2435 } 2436 } 2437 INP_INFO_WUNLOCK(&tcbinfo); 2438 headlocked = 0; 2439#ifdef TCPDEBUG 2440 if (so->so_options & SO_DEBUG) 2441 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2442 &tcp_savetcp, 0); 2443#endif 2444 2445 /* 2446 * Return any desired output. 2447 */ 2448 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2449 (void) tcp_output(tp); 2450 2451check_delack: 2452 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2453 INP_LOCK_ASSERT(inp); 2454 if (tp->t_flags & TF_DELACK) { 2455 tp->t_flags &= ~TF_DELACK; 2456 callout_reset(tp->tt_delack, tcp_delacktime, 2457 tcp_timer_delack, tp); 2458 } 2459 INP_UNLOCK(inp); 2460 return; 2461 2462dropafterack: 2463 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2464 /* 2465 * Generate an ACK dropping incoming segment if it occupies 2466 * sequence space, where the ACK reflects our state. 2467 * 2468 * We can now skip the test for the RST flag since all 2469 * paths to this code happen after packets containing 2470 * RST have been dropped. 2471 * 2472 * In the SYN-RECEIVED state, don't send an ACK unless the 2473 * segment we received passes the SYN-RECEIVED ACK test. 2474 * If it fails send a RST. This breaks the loop in the 2475 * "LAND" DoS attack, and also prevents an ACK storm 2476 * between two listening ports that have been sent forged 2477 * SYN segments, each with the source address of the other. 2478 */ 2479 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2480 (SEQ_GT(tp->snd_una, th->th_ack) || 2481 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2482 rstreason = BANDLIM_RST_OPENPORT; 2483 goto dropwithreset; 2484 } 2485#ifdef TCPDEBUG 2486 if (so->so_options & SO_DEBUG) 2487 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2488 &tcp_savetcp, 0); 2489#endif 2490 KASSERT(headlocked, ("headlocked should be 1")); 2491 INP_INFO_WUNLOCK(&tcbinfo); 2492 tp->t_flags |= TF_ACKNOW; 2493 (void) tcp_output(tp); 2494 INP_UNLOCK(inp); 2495 m_freem(m); 2496 return; 2497 2498dropwithreset: 2499 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2500 /* 2501 * Generate a RST, dropping incoming segment. 2502 * Make ACK acceptable to originator of segment. 2503 * Don't bother to respond if destination was broadcast/multicast. 2504 */ 2505 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2506 goto drop; 2507 if (isipv6) { 2508 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2509 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2510 goto drop; 2511 } else { 2512 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2513 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2514 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2515 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2516 goto drop; 2517 } 2518 /* IPv6 anycast check is done at tcp6_input() */ 2519 2520 /* 2521 * Perform bandwidth limiting. 2522 */ 2523 if (badport_bandlim(rstreason) < 0) 2524 goto drop; 2525 2526#ifdef TCPDEBUG 2527 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2528 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2529 &tcp_savetcp, 0); 2530#endif 2531 2532 if (thflags & TH_ACK) 2533 /* mtod() below is safe as long as hdr dropping is delayed */ 2534 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2535 TH_RST); 2536 else { 2537 if (thflags & TH_SYN) 2538 tlen++; 2539 /* mtod() below is safe as long as hdr dropping is delayed */ 2540 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2541 (tcp_seq)0, TH_RST|TH_ACK); 2542 } 2543 2544 if (tp) 2545 INP_UNLOCK(inp); 2546 if (headlocked) 2547 INP_INFO_WUNLOCK(&tcbinfo); 2548 return; 2549 2550drop: 2551 /* 2552 * Drop space held by incoming segment and return. 2553 */ 2554#ifdef TCPDEBUG 2555 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2556 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2557 &tcp_savetcp, 0); 2558#endif 2559 if (tp) 2560 INP_UNLOCK(inp); 2561 if (headlocked) 2562 INP_INFO_WUNLOCK(&tcbinfo); 2563 m_freem(m); 2564 return; 2565} 2566 2567/* 2568 * Parse TCP options and place in tcpopt. 2569 */ 2570static void 2571tcp_dooptions(to, cp, cnt, is_syn) 2572 struct tcpopt *to; 2573 u_char *cp; 2574 int cnt; 2575 int is_syn; 2576{ 2577 int opt, optlen; 2578 2579 to->to_flags = 0; 2580 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2581 opt = cp[0]; 2582 if (opt == TCPOPT_EOL) 2583 break; 2584 if (opt == TCPOPT_NOP) 2585 optlen = 1; 2586 else { 2587 if (cnt < 2) 2588 break; 2589 optlen = cp[1]; 2590 if (optlen < 2 || optlen > cnt) 2591 break; 2592 } 2593 switch (opt) { 2594 case TCPOPT_MAXSEG: 2595 if (optlen != TCPOLEN_MAXSEG) 2596 continue; 2597 if (!is_syn) 2598 continue; 2599 to->to_flags |= TOF_MSS; 2600 bcopy((char *)cp + 2, 2601 (char *)&to->to_mss, sizeof(to->to_mss)); 2602 to->to_mss = ntohs(to->to_mss); 2603 break; 2604 case TCPOPT_WINDOW: 2605 if (optlen != TCPOLEN_WINDOW) 2606 continue; 2607 if (! is_syn) 2608 continue; 2609 to->to_flags |= TOF_SCALE; 2610 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2611 break; 2612 case TCPOPT_TIMESTAMP: 2613 if (optlen != TCPOLEN_TIMESTAMP) 2614 continue; 2615 to->to_flags |= TOF_TS; 2616 bcopy((char *)cp + 2, 2617 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2618 to->to_tsval = ntohl(to->to_tsval); 2619 bcopy((char *)cp + 6, 2620 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2621 to->to_tsecr = ntohl(to->to_tsecr); 2622 /* 2623 * If echoed timestamp is later than the current time, 2624 * fall back to non RFC1323 RTT calculation. 2625 */ 2626 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks)) 2627 to->to_tsecr = 0; 2628 break; 2629#ifdef TCP_SIGNATURE 2630 /* 2631 * XXX In order to reply to a host which has set the 2632 * TCP_SIGNATURE option in its initial SYN, we have to 2633 * record the fact that the option was observed here 2634 * for the syncache code to perform the correct response. 2635 */ 2636 case TCPOPT_SIGNATURE: 2637 if (optlen != TCPOLEN_SIGNATURE) 2638 continue; 2639 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2640 break; 2641#endif 2642 case TCPOPT_SACK_PERMITTED: 2643 if (!tcp_do_sack || 2644 optlen != TCPOLEN_SACK_PERMITTED) 2645 continue; 2646 if (is_syn) { 2647 /* MUST only be set on SYN */ 2648 to->to_flags |= TOF_SACK; 2649 } 2650 break; 2651 case TCPOPT_SACK: 2652 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2653 continue; 2654 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2655 to->to_sacks = cp + 2; 2656 tcpstat.tcps_sack_rcv_blocks++; 2657 break; 2658 default: 2659 continue; 2660 } 2661 } 2662} 2663 2664/* 2665 * Pull out of band byte out of a segment so 2666 * it doesn't appear in the user's data queue. 2667 * It is still reflected in the segment length for 2668 * sequencing purposes. 2669 */ 2670static void 2671tcp_pulloutofband(so, th, m, off) 2672 struct socket *so; 2673 struct tcphdr *th; 2674 register struct mbuf *m; 2675 int off; /* delayed to be droped hdrlen */ 2676{ 2677 int cnt = off + th->th_urp - 1; 2678 2679 while (cnt >= 0) { 2680 if (m->m_len > cnt) { 2681 char *cp = mtod(m, caddr_t) + cnt; 2682 struct tcpcb *tp = sototcpcb(so); 2683 2684 tp->t_iobc = *cp; 2685 tp->t_oobflags |= TCPOOB_HAVEDATA; 2686 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2687 m->m_len--; 2688 if (m->m_flags & M_PKTHDR) 2689 m->m_pkthdr.len--; 2690 return; 2691 } 2692 cnt -= m->m_len; 2693 m = m->m_next; 2694 if (m == 0) 2695 break; 2696 } 2697 panic("tcp_pulloutofband"); 2698} 2699 2700/* 2701 * Collect new round-trip time estimate 2702 * and update averages and current timeout. 2703 */ 2704static void 2705tcp_xmit_timer(tp, rtt) 2706 register struct tcpcb *tp; 2707 int rtt; 2708{ 2709 register int delta; 2710 2711 INP_LOCK_ASSERT(tp->t_inpcb); 2712 2713 tcpstat.tcps_rttupdated++; 2714 tp->t_rttupdated++; 2715 if (tp->t_srtt != 0) { 2716 /* 2717 * srtt is stored as fixed point with 5 bits after the 2718 * binary point (i.e., scaled by 8). The following magic 2719 * is equivalent to the smoothing algorithm in rfc793 with 2720 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2721 * point). Adjust rtt to origin 0. 2722 */ 2723 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2724 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2725 2726 if ((tp->t_srtt += delta) <= 0) 2727 tp->t_srtt = 1; 2728 2729 /* 2730 * We accumulate a smoothed rtt variance (actually, a 2731 * smoothed mean difference), then set the retransmit 2732 * timer to smoothed rtt + 4 times the smoothed variance. 2733 * rttvar is stored as fixed point with 4 bits after the 2734 * binary point (scaled by 16). The following is 2735 * equivalent to rfc793 smoothing with an alpha of .75 2736 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2737 * rfc793's wired-in beta. 2738 */ 2739 if (delta < 0) 2740 delta = -delta; 2741 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2742 if ((tp->t_rttvar += delta) <= 0) 2743 tp->t_rttvar = 1; 2744 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2745 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2746 } else { 2747 /* 2748 * No rtt measurement yet - use the unsmoothed rtt. 2749 * Set the variance to half the rtt (so our first 2750 * retransmit happens at 3*rtt). 2751 */ 2752 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2753 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2754 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2755 } 2756 tp->t_rtttime = 0; 2757 tp->t_rxtshift = 0; 2758 2759 /* 2760 * the retransmit should happen at rtt + 4 * rttvar. 2761 * Because of the way we do the smoothing, srtt and rttvar 2762 * will each average +1/2 tick of bias. When we compute 2763 * the retransmit timer, we want 1/2 tick of rounding and 2764 * 1 extra tick because of +-1/2 tick uncertainty in the 2765 * firing of the timer. The bias will give us exactly the 2766 * 1.5 tick we need. But, because the bias is 2767 * statistical, we have to test that we don't drop below 2768 * the minimum feasible timer (which is 2 ticks). 2769 */ 2770 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2771 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2772 2773 /* 2774 * We received an ack for a packet that wasn't retransmitted; 2775 * it is probably safe to discard any error indications we've 2776 * received recently. This isn't quite right, but close enough 2777 * for now (a route might have failed after we sent a segment, 2778 * and the return path might not be symmetrical). 2779 */ 2780 tp->t_softerror = 0; 2781} 2782 2783/* 2784 * Determine a reasonable value for maxseg size. 2785 * If the route is known, check route for mtu. 2786 * If none, use an mss that can be handled on the outgoing 2787 * interface without forcing IP to fragment; if bigger than 2788 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2789 * to utilize large mbufs. If no route is found, route has no mtu, 2790 * or the destination isn't local, use a default, hopefully conservative 2791 * size (usually 512 or the default IP max size, but no more than the mtu 2792 * of the interface), as we can't discover anything about intervening 2793 * gateways or networks. We also initialize the congestion/slow start 2794 * window to be a single segment if the destination isn't local. 2795 * While looking at the routing entry, we also initialize other path-dependent 2796 * parameters from pre-set or cached values in the routing entry. 2797 * 2798 * Also take into account the space needed for options that we 2799 * send regularly. Make maxseg shorter by that amount to assure 2800 * that we can send maxseg amount of data even when the options 2801 * are present. Store the upper limit of the length of options plus 2802 * data in maxopd. 2803 * 2804 * 2805 * In case of T/TCP, we call this routine during implicit connection 2806 * setup as well (offer = -1), to initialize maxseg from the cached 2807 * MSS of our peer. 2808 * 2809 * NOTE that this routine is only called when we process an incoming 2810 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2811 */ 2812void 2813tcp_mss(tp, offer) 2814 struct tcpcb *tp; 2815 int offer; 2816{ 2817 int rtt, mss; 2818 u_long bufsize; 2819 u_long maxmtu; 2820 struct inpcb *inp = tp->t_inpcb; 2821 struct socket *so; 2822 struct hc_metrics_lite metrics; 2823 int origoffer = offer; 2824#ifdef INET6 2825 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2826 size_t min_protoh = isipv6 ? 2827 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2828 sizeof (struct tcpiphdr); 2829#else 2830 const size_t min_protoh = sizeof(struct tcpiphdr); 2831#endif 2832 2833 /* initialize */ 2834#ifdef INET6 2835 if (isipv6) { 2836 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2837 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2838 } else 2839#endif 2840 { 2841 maxmtu = tcp_maxmtu(&inp->inp_inc); 2842 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2843 } 2844 so = inp->inp_socket; 2845 2846 /* 2847 * no route to sender, stay with default mss and return 2848 */ 2849 if (maxmtu == 0) 2850 return; 2851 2852 /* what have we got? */ 2853 switch (offer) { 2854 case 0: 2855 /* 2856 * Offer == 0 means that there was no MSS on the SYN 2857 * segment, in this case we use tcp_mssdflt. 2858 */ 2859 offer = 2860#ifdef INET6 2861 isipv6 ? tcp_v6mssdflt : 2862#endif 2863 tcp_mssdflt; 2864 break; 2865 2866 case -1: 2867 /* 2868 * Offer == -1 means that we didn't receive SYN yet. 2869 */ 2870 /* FALLTHROUGH */ 2871 2872 default: 2873 /* 2874 * Prevent DoS attack with too small MSS. Round up 2875 * to at least minmss. 2876 */ 2877 offer = max(offer, tcp_minmss); 2878 /* 2879 * Sanity check: make sure that maxopd will be large 2880 * enough to allow some data on segments even if the 2881 * all the option space is used (40bytes). Otherwise 2882 * funny things may happen in tcp_output. 2883 */ 2884 offer = max(offer, 64); 2885 } 2886 2887 /* 2888 * rmx information is now retrieved from tcp_hostcache 2889 */ 2890 tcp_hc_get(&inp->inp_inc, &metrics); 2891 2892 /* 2893 * if there's a discovered mtu int tcp hostcache, use it 2894 * else, use the link mtu. 2895 */ 2896 if (metrics.rmx_mtu) 2897 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2898 else { 2899#ifdef INET6 2900 if (isipv6) { 2901 mss = maxmtu - min_protoh; 2902 if (!path_mtu_discovery && 2903 !in6_localaddr(&inp->in6p_faddr)) 2904 mss = min(mss, tcp_v6mssdflt); 2905 } else 2906#endif 2907 { 2908 mss = maxmtu - min_protoh; 2909 if (!path_mtu_discovery && 2910 !in_localaddr(inp->inp_faddr)) 2911 mss = min(mss, tcp_mssdflt); 2912 } 2913 } 2914 mss = min(mss, offer); 2915 2916 /* 2917 * maxopd stores the maximum length of data AND options 2918 * in a segment; maxseg is the amount of data in a normal 2919 * segment. We need to store this value (maxopd) apart 2920 * from maxseg, because now every segment carries options 2921 * and thus we normally have somewhat less data in segments. 2922 */ 2923 tp->t_maxopd = mss; 2924 2925 /* 2926 * origoffer==-1 indicates, that no segments were received yet. 2927 * In this case we just guess. 2928 */ 2929 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2930 (origoffer == -1 || 2931 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2932 mss -= TCPOLEN_TSTAMP_APPA; 2933 tp->t_maxseg = mss; 2934 2935#if (MCLBYTES & (MCLBYTES - 1)) == 0 2936 if (mss > MCLBYTES) 2937 mss &= ~(MCLBYTES-1); 2938#else 2939 if (mss > MCLBYTES) 2940 mss = mss / MCLBYTES * MCLBYTES; 2941#endif 2942 tp->t_maxseg = mss; 2943 2944 /* 2945 * If there's a pipesize, change the socket buffer to that size, 2946 * don't change if sb_hiwat is different than default (then it 2947 * has been changed on purpose with setsockopt). 2948 * Make the socket buffers an integral number of mss units; 2949 * if the mss is larger than the socket buffer, decrease the mss. 2950 */ 2951 SOCKBUF_LOCK(&so->so_snd); 2952 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2953 bufsize = metrics.rmx_sendpipe; 2954 else 2955 bufsize = so->so_snd.sb_hiwat; 2956 if (bufsize < mss) 2957 mss = bufsize; 2958 else { 2959 bufsize = roundup(bufsize, mss); 2960 if (bufsize > sb_max) 2961 bufsize = sb_max; 2962 if (bufsize > so->so_snd.sb_hiwat) 2963 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2964 } 2965 SOCKBUF_UNLOCK(&so->so_snd); 2966 tp->t_maxseg = mss; 2967 2968 SOCKBUF_LOCK(&so->so_rcv); 2969 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2970 bufsize = metrics.rmx_recvpipe; 2971 else 2972 bufsize = so->so_rcv.sb_hiwat; 2973 if (bufsize > mss) { 2974 bufsize = roundup(bufsize, mss); 2975 if (bufsize > sb_max) 2976 bufsize = sb_max; 2977 if (bufsize > so->so_rcv.sb_hiwat) 2978 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2979 } 2980 SOCKBUF_UNLOCK(&so->so_rcv); 2981 /* 2982 * While we're here, check the others too 2983 */ 2984 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2985 tp->t_srtt = rtt; 2986 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2987 tcpstat.tcps_usedrtt++; 2988 if (metrics.rmx_rttvar) { 2989 tp->t_rttvar = metrics.rmx_rttvar; 2990 tcpstat.tcps_usedrttvar++; 2991 } else { 2992 /* default variation is +- 1 rtt */ 2993 tp->t_rttvar = 2994 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2995 } 2996 TCPT_RANGESET(tp->t_rxtcur, 2997 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2998 tp->t_rttmin, TCPTV_REXMTMAX); 2999 } 3000 if (metrics.rmx_ssthresh) { 3001 /* 3002 * There's some sort of gateway or interface 3003 * buffer limit on the path. Use this to set 3004 * the slow start threshhold, but set the 3005 * threshold to no less than 2*mss. 3006 */ 3007 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3008 tcpstat.tcps_usedssthresh++; 3009 } 3010 if (metrics.rmx_bandwidth) 3011 tp->snd_bandwidth = metrics.rmx_bandwidth; 3012 3013 /* 3014 * Set the slow-start flight size depending on whether this 3015 * is a local network or not. 3016 * 3017 * Extend this so we cache the cwnd too and retrieve it here. 3018 * Make cwnd even bigger than RFC3390 suggests but only if we 3019 * have previous experience with the remote host. Be careful 3020 * not make cwnd bigger than remote receive window or our own 3021 * send socket buffer. Maybe put some additional upper bound 3022 * on the retrieved cwnd. Should do incremental updates to 3023 * hostcache when cwnd collapses so next connection doesn't 3024 * overloads the path again. 3025 * 3026 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3027 * We currently check only in syncache_socket for that. 3028 */ 3029#define TCP_METRICS_CWND 3030#ifdef TCP_METRICS_CWND 3031 if (metrics.rmx_cwnd) 3032 tp->snd_cwnd = max(mss, 3033 min(metrics.rmx_cwnd / 2, 3034 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3035 else 3036#endif 3037 if (tcp_do_rfc3390) 3038 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3039#ifdef INET6 3040 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3041 (!isipv6 && in_localaddr(inp->inp_faddr))) 3042#else 3043 else if (in_localaddr(inp->inp_faddr)) 3044#endif 3045 tp->snd_cwnd = mss * ss_fltsz_local; 3046 else 3047 tp->snd_cwnd = mss * ss_fltsz; 3048} 3049 3050/* 3051 * Determine the MSS option to send on an outgoing SYN. 3052 */ 3053int 3054tcp_mssopt(inc) 3055 struct in_conninfo *inc; 3056{ 3057 int mss = 0; 3058 u_long maxmtu = 0; 3059 u_long thcmtu = 0; 3060 size_t min_protoh; 3061#ifdef INET6 3062 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3063#endif 3064 3065 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3066 3067#ifdef INET6 3068 if (isipv6) { 3069 mss = tcp_v6mssdflt; 3070 maxmtu = tcp_maxmtu6(inc); 3071 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3072 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3073 } else 3074#endif 3075 { 3076 mss = tcp_mssdflt; 3077 maxmtu = tcp_maxmtu(inc); 3078 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3079 min_protoh = sizeof(struct tcpiphdr); 3080 } 3081 if (maxmtu && thcmtu) 3082 mss = min(maxmtu, thcmtu) - min_protoh; 3083 else if (maxmtu || thcmtu) 3084 mss = max(maxmtu, thcmtu) - min_protoh; 3085 3086 return (mss); 3087} 3088 3089 3090/* 3091 * On a partial ack arrives, force the retransmission of the 3092 * next unacknowledged segment. Do not clear tp->t_dupacks. 3093 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3094 * be started again. 3095 */ 3096static void 3097tcp_newreno_partial_ack(tp, th) 3098 struct tcpcb *tp; 3099 struct tcphdr *th; 3100{ 3101 tcp_seq onxt = tp->snd_nxt; 3102 u_long ocwnd = tp->snd_cwnd; 3103 3104 callout_stop(tp->tt_rexmt); 3105 tp->t_rtttime = 0; 3106 tp->snd_nxt = th->th_ack; 3107 /* 3108 * Set snd_cwnd to one segment beyond acknowledged offset. 3109 * (tp->snd_una has not yet been updated when this function is called.) 3110 */ 3111 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3112 tp->t_flags |= TF_ACKNOW; 3113 (void) tcp_output(tp); 3114 tp->snd_cwnd = ocwnd; 3115 if (SEQ_GT(onxt, tp->snd_nxt)) 3116 tp->snd_nxt = onxt; 3117 /* 3118 * Partial window deflation. Relies on fact that tp->snd_una 3119 * not updated yet. 3120 */ 3121 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3122 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3123 else 3124 tp->snd_cwnd = 0; 3125 tp->snd_cwnd += tp->t_maxseg; 3126} 3127 3128/* 3129 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3130 * looking for a pcb in the listen state. Returns 0 otherwise. 3131 */ 3132static int 3133tcp_timewait(tw, to, th, m, tlen) 3134 struct tcptw *tw; 3135 struct tcpopt *to; 3136 struct tcphdr *th; 3137 struct mbuf *m; 3138 int tlen; 3139{ 3140 int thflags; 3141 tcp_seq seq; 3142#ifdef INET6 3143 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3144#else 3145 const int isipv6 = 0; 3146#endif 3147 3148 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */ 3149 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3150 INP_LOCK_ASSERT(tw->tw_inpcb); 3151 3152 thflags = th->th_flags; 3153 3154 /* 3155 * NOTE: for FIN_WAIT_2 (to be added later), 3156 * must validate sequence number before accepting RST 3157 */ 3158 3159 /* 3160 * If the segment contains RST: 3161 * Drop the segment - see Stevens, vol. 2, p. 964 and 3162 * RFC 1337. 3163 */ 3164 if (thflags & TH_RST) 3165 goto drop; 3166 3167#if 0 3168/* PAWS not needed at the moment */ 3169 /* 3170 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3171 * and it's less than ts_recent, drop it. 3172 */ 3173 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3174 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3175 if ((thflags & TH_ACK) == 0) 3176 goto drop; 3177 goto ack; 3178 } 3179 /* 3180 * ts_recent is never updated because we never accept new segments. 3181 */ 3182#endif 3183 3184 /* 3185 * If a new connection request is received 3186 * while in TIME_WAIT, drop the old connection 3187 * and start over if the sequence numbers 3188 * are above the previous ones. 3189 */ 3190 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3191 (void) tcp_twclose(tw, 0); 3192 return (1); 3193 } 3194 3195 /* 3196 * Drop the the segment if it does not contain an ACK. 3197 */ 3198 if ((thflags & TH_ACK) == 0) 3199 goto drop; 3200 3201 /* 3202 * Reset the 2MSL timer if this is a duplicate FIN. 3203 */ 3204 if (thflags & TH_FIN) { 3205 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3206 if (seq + 1 == tw->rcv_nxt) 3207 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3208 } 3209 3210 /* 3211 * Acknowledge the segment if it has data or is not a duplicate ACK. 3212 */ 3213 if (thflags != TH_ACK || tlen != 0 || 3214 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3215 tcp_twrespond(tw, TH_ACK); 3216 goto drop; 3217 3218 /* 3219 * Generate a RST, dropping incoming segment. 3220 * Make ACK acceptable to originator of segment. 3221 * Don't bother to respond if destination was broadcast/multicast. 3222 */ 3223 if (m->m_flags & (M_BCAST|M_MCAST)) 3224 goto drop; 3225 if (isipv6) { 3226 struct ip6_hdr *ip6; 3227 3228 /* IPv6 anycast check is done at tcp6_input() */ 3229 ip6 = mtod(m, struct ip6_hdr *); 3230 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3231 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3232 goto drop; 3233 } else { 3234 struct ip *ip; 3235 3236 ip = mtod(m, struct ip *); 3237 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3238 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3239 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3240 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3241 goto drop; 3242 } 3243 if (thflags & TH_ACK) { 3244 tcp_respond(NULL, 3245 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3246 } else { 3247 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3248 tcp_respond(NULL, 3249 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3250 } 3251 INP_UNLOCK(tw->tw_inpcb); 3252 return (0); 3253 3254drop: 3255 INP_UNLOCK(tw->tw_inpcb); 3256 m_freem(m); 3257 return (0); 3258} 3259