tcp_input.c revision 155819
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 155819 2006-02-18 17:05:00Z 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> /* required for icmp_var.h */ 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 if (!tp->t_rttlow || 1197 tp->t_rttlow > ticks - to.to_tsecr) 1198 tp->t_rttlow = ticks - to.to_tsecr; 1199 tcp_xmit_timer(tp, 1200 ticks - to.to_tsecr + 1); 1201 } else if (tp->t_rtttime && 1202 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1203 if (!tp->t_rttlow || 1204 tp->t_rttlow > ticks - tp->t_rtttime) 1205 tp->t_rttlow = ticks - tp->t_rtttime; 1206 tcp_xmit_timer(tp, 1207 ticks - tp->t_rtttime); 1208 } 1209 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1210 acked = th->th_ack - tp->snd_una; 1211 tcpstat.tcps_rcvackpack++; 1212 tcpstat.tcps_rcvackbyte += acked; 1213 sbdrop(&so->so_snd, acked); 1214 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1215 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1216 tp->snd_recover = th->th_ack - 1; 1217 tp->snd_una = th->th_ack; 1218 /* 1219 * pull snd_wl2 up to prevent seq wrap relative 1220 * to th_ack. 1221 */ 1222 tp->snd_wl2 = th->th_ack; 1223 tp->t_dupacks = 0; 1224 m_freem(m); 1225 ND6_HINT(tp); /* some progress has been done */ 1226 1227 /* 1228 * If all outstanding data are acked, stop 1229 * retransmit timer, otherwise restart timer 1230 * using current (possibly backed-off) value. 1231 * If process is waiting for space, 1232 * wakeup/selwakeup/signal. If data 1233 * are ready to send, let tcp_output 1234 * decide between more output or persist. 1235 1236#ifdef TCPDEBUG 1237 if (so->so_options & SO_DEBUG) 1238 tcp_trace(TA_INPUT, ostate, tp, 1239 (void *)tcp_saveipgen, 1240 &tcp_savetcp, 0); 1241#endif 1242 */ 1243 if (tp->snd_una == tp->snd_max) 1244 callout_stop(tp->tt_rexmt); 1245 else if (!callout_active(tp->tt_persist)) 1246 callout_reset(tp->tt_rexmt, 1247 tp->t_rxtcur, 1248 tcp_timer_rexmt, tp); 1249 1250 sowwakeup(so); 1251 if (so->so_snd.sb_cc) 1252 (void) tcp_output(tp); 1253 goto check_delack; 1254 } 1255 } else if (th->th_ack == tp->snd_una && 1256 LIST_EMPTY(&tp->t_segq) && 1257 tlen <= sbspace(&so->so_rcv)) { 1258 KASSERT(headlocked, ("headlocked")); 1259 INP_INFO_WUNLOCK(&tcbinfo); 1260 headlocked = 0; 1261 /* 1262 * this is a pure, in-sequence data packet 1263 * with nothing on the reassembly queue and 1264 * we have enough buffer space to take it. 1265 */ 1266 /* Clean receiver SACK report if present */ 1267 if (tp->sack_enable && tp->rcv_numsacks) 1268 tcp_clean_sackreport(tp); 1269 ++tcpstat.tcps_preddat; 1270 tp->rcv_nxt += tlen; 1271 /* 1272 * Pull snd_wl1 up to prevent seq wrap relative to 1273 * th_seq. 1274 */ 1275 tp->snd_wl1 = th->th_seq; 1276 /* 1277 * Pull rcv_up up to prevent seq wrap relative to 1278 * rcv_nxt. 1279 */ 1280 tp->rcv_up = tp->rcv_nxt; 1281 tcpstat.tcps_rcvpack++; 1282 tcpstat.tcps_rcvbyte += tlen; 1283 ND6_HINT(tp); /* some progress has been done */ 1284 /* 1285#ifdef TCPDEBUG 1286 if (so->so_options & SO_DEBUG) 1287 tcp_trace(TA_INPUT, ostate, tp, 1288 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1289#endif 1290 * Add data to socket buffer. 1291 */ 1292 SOCKBUF_LOCK(&so->so_rcv); 1293 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1294 m_freem(m); 1295 } else { 1296 m_adj(m, drop_hdrlen); /* delayed header drop */ 1297 sbappendstream_locked(&so->so_rcv, m); 1298 } 1299 sorwakeup_locked(so); 1300 if (DELAY_ACK(tp)) { 1301 tp->t_flags |= TF_DELACK; 1302 } else { 1303 tp->t_flags |= TF_ACKNOW; 1304 tcp_output(tp); 1305 } 1306 goto check_delack; 1307 } 1308 } 1309 1310 /* 1311 * Calculate amount of space in receive window, 1312 * and then do TCP input processing. 1313 * Receive window is amount of space in rcv queue, 1314 * but not less than advertised window. 1315 */ 1316 { int win; 1317 1318 win = sbspace(&so->so_rcv); 1319 if (win < 0) 1320 win = 0; 1321 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1322 } 1323 1324 switch (tp->t_state) { 1325 1326 /* 1327 * If the state is SYN_RECEIVED: 1328 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1329 */ 1330 case TCPS_SYN_RECEIVED: 1331 if ((thflags & TH_ACK) && 1332 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1333 SEQ_GT(th->th_ack, tp->snd_max))) { 1334 rstreason = BANDLIM_RST_OPENPORT; 1335 goto dropwithreset; 1336 } 1337 break; 1338 1339 /* 1340 * If the state is SYN_SENT: 1341 * if seg contains an ACK, but not for our SYN, drop the input. 1342 * if seg contains a RST, then drop the connection. 1343 * if seg does not contain SYN, then drop it. 1344 * Otherwise this is an acceptable SYN segment 1345 * initialize tp->rcv_nxt and tp->irs 1346 * if seg contains ack then advance tp->snd_una 1347 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1348 * arrange for segment to be acked (eventually) 1349 * continue processing rest of data/controls, beginning with URG 1350 */ 1351 case TCPS_SYN_SENT: 1352 if ((thflags & TH_ACK) && 1353 (SEQ_LEQ(th->th_ack, tp->iss) || 1354 SEQ_GT(th->th_ack, tp->snd_max))) { 1355 rstreason = BANDLIM_UNLIMITED; 1356 goto dropwithreset; 1357 } 1358 if (thflags & TH_RST) { 1359 if (thflags & TH_ACK) { 1360 KASSERT(headlocked, ("tcp_input: after_listen" 1361 ": tcp_drop.2: head not locked")); 1362 tp = tcp_drop(tp, ECONNREFUSED); 1363 } 1364 goto drop; 1365 } 1366 if ((thflags & TH_SYN) == 0) 1367 goto drop; 1368 tp->snd_wnd = th->th_win; /* initial send window */ 1369 1370 tp->irs = th->th_seq; 1371 tcp_rcvseqinit(tp); 1372 if (thflags & TH_ACK) { 1373 tcpstat.tcps_connects++; 1374 soisconnected(so); 1375#ifdef MAC 1376 SOCK_LOCK(so); 1377 mac_set_socket_peer_from_mbuf(m, so); 1378 SOCK_UNLOCK(so); 1379#endif 1380 /* Do window scaling on this connection? */ 1381 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1382 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1383 tp->snd_scale = tp->requested_s_scale; 1384 tp->rcv_scale = tp->request_r_scale; 1385 } 1386 tp->rcv_adv += tp->rcv_wnd; 1387 tp->snd_una++; /* SYN is acked */ 1388 /* 1389 * If there's data, delay ACK; if there's also a FIN 1390 * ACKNOW will be turned on later. 1391 */ 1392 if (DELAY_ACK(tp) && tlen != 0) 1393 callout_reset(tp->tt_delack, tcp_delacktime, 1394 tcp_timer_delack, tp); 1395 else 1396 tp->t_flags |= TF_ACKNOW; 1397 /* 1398 * Received <SYN,ACK> in SYN_SENT[*] state. 1399 * Transitions: 1400 * SYN_SENT --> ESTABLISHED 1401 * SYN_SENT* --> FIN_WAIT_1 1402 */ 1403 tp->t_starttime = ticks; 1404 if (tp->t_flags & TF_NEEDFIN) { 1405 tp->t_state = TCPS_FIN_WAIT_1; 1406 tp->t_flags &= ~TF_NEEDFIN; 1407 thflags &= ~TH_SYN; 1408 } else { 1409 tp->t_state = TCPS_ESTABLISHED; 1410 callout_reset(tp->tt_keep, tcp_keepidle, 1411 tcp_timer_keep, tp); 1412 } 1413 } else { 1414 /* 1415 * Received initial SYN in SYN-SENT[*] state => 1416 * simultaneous open. If segment contains CC option 1417 * and there is a cached CC, apply TAO test. 1418 * If it succeeds, connection is * half-synchronized. 1419 * Otherwise, do 3-way handshake: 1420 * SYN-SENT -> SYN-RECEIVED 1421 * SYN-SENT* -> SYN-RECEIVED* 1422 * If there was no CC option, clear cached CC value. 1423 */ 1424 tp->t_flags |= TF_ACKNOW; 1425 callout_stop(tp->tt_rexmt); 1426 tp->t_state = TCPS_SYN_RECEIVED; 1427 } 1428 1429trimthenstep6: 1430 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1431 "locked")); 1432 INP_LOCK_ASSERT(inp); 1433 1434 /* 1435 * Advance th->th_seq to correspond to first data byte. 1436 * If data, trim to stay within window, 1437 * dropping FIN if necessary. 1438 */ 1439 th->th_seq++; 1440 if (tlen > tp->rcv_wnd) { 1441 todrop = tlen - tp->rcv_wnd; 1442 m_adj(m, -todrop); 1443 tlen = tp->rcv_wnd; 1444 thflags &= ~TH_FIN; 1445 tcpstat.tcps_rcvpackafterwin++; 1446 tcpstat.tcps_rcvbyteafterwin += todrop; 1447 } 1448 tp->snd_wl1 = th->th_seq - 1; 1449 tp->rcv_up = th->th_seq; 1450 /* 1451 * Client side of transaction: already sent SYN and data. 1452 * If the remote host used T/TCP to validate the SYN, 1453 * our data will be ACK'd; if so, enter normal data segment 1454 * processing in the middle of step 5, ack processing. 1455 * Otherwise, goto step 6. 1456 */ 1457 if (thflags & TH_ACK) 1458 goto process_ACK; 1459 1460 goto step6; 1461 1462 /* 1463 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1464 * do normal processing. 1465 * 1466 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1467 */ 1468 case TCPS_LAST_ACK: 1469 case TCPS_CLOSING: 1470 case TCPS_TIME_WAIT: 1471 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1472 break; /* continue normal processing */ 1473 } 1474 1475 /* 1476 * States other than LISTEN or SYN_SENT. 1477 * First check the RST flag and sequence number since reset segments 1478 * are exempt from the timestamp and connection count tests. This 1479 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1480 * below which allowed reset segments in half the sequence space 1481 * to fall though and be processed (which gives forged reset 1482 * segments with a random sequence number a 50 percent chance of 1483 * killing a connection). 1484 * Then check timestamp, if present. 1485 * Then check the connection count, if present. 1486 * Then check that at least some bytes of segment are within 1487 * receive window. If segment begins before rcv_nxt, 1488 * drop leading data (and SYN); if nothing left, just ack. 1489 * 1490 * 1491 * If the RST bit is set, check the sequence number to see 1492 * if this is a valid reset segment. 1493 * RFC 793 page 37: 1494 * In all states except SYN-SENT, all reset (RST) segments 1495 * are validated by checking their SEQ-fields. A reset is 1496 * valid if its sequence number is in the window. 1497 * Note: this does not take into account delayed ACKs, so 1498 * we should test against last_ack_sent instead of rcv_nxt. 1499 * The sequence number in the reset segment is normally an 1500 * echo of our outgoing acknowlegement numbers, but some hosts 1501 * send a reset with the sequence number at the rightmost edge 1502 * of our receive window, and we have to handle this case. 1503 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1504 * that brute force RST attacks are possible. To combat this, 1505 * we use a much stricter check while in the ESTABLISHED state, 1506 * only accepting RSTs where the sequence number is equal to 1507 * last_ack_sent. In all other states (the states in which a 1508 * RST is more likely), the more permissive check is used. 1509 * If we have multiple segments in flight, the intial reset 1510 * segment sequence numbers will be to the left of last_ack_sent, 1511 * but they will eventually catch up. 1512 * In any case, it never made sense to trim reset segments to 1513 * fit the receive window since RFC 1122 says: 1514 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1515 * 1516 * A TCP SHOULD allow a received RST segment to include data. 1517 * 1518 * DISCUSSION 1519 * It has been suggested that a RST segment could contain 1520 * ASCII text that encoded and explained the cause of the 1521 * RST. No standard has yet been established for such 1522 * data. 1523 * 1524 * If the reset segment passes the sequence number test examine 1525 * the state: 1526 * SYN_RECEIVED STATE: 1527 * If passive open, return to LISTEN state. 1528 * If active open, inform user that connection was refused. 1529 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1530 * Inform user that connection was reset, and close tcb. 1531 * CLOSING, LAST_ACK STATES: 1532 * Close the tcb. 1533 * TIME_WAIT STATE: 1534 * Drop the segment - see Stevens, vol. 2, p. 964 and 1535 * RFC 1337. 1536 */ 1537 if (thflags & TH_RST) { 1538 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1539 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1540 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1541 switch (tp->t_state) { 1542 1543 case TCPS_SYN_RECEIVED: 1544 so->so_error = ECONNREFUSED; 1545 goto close; 1546 1547 case TCPS_ESTABLISHED: 1548 if (tp->last_ack_sent != th->th_seq && 1549 tcp_insecure_rst == 0) { 1550 tcpstat.tcps_badrst++; 1551 goto drop; 1552 } 1553 case TCPS_FIN_WAIT_1: 1554 case TCPS_FIN_WAIT_2: 1555 case TCPS_CLOSE_WAIT: 1556 so->so_error = ECONNRESET; 1557 close: 1558 tp->t_state = TCPS_CLOSED; 1559 tcpstat.tcps_drops++; 1560 KASSERT(headlocked, ("tcp_input: " 1561 "trimthenstep6: tcp_close: head not " 1562 "locked")); 1563 tp = tcp_close(tp); 1564 break; 1565 1566 case TCPS_CLOSING: 1567 case TCPS_LAST_ACK: 1568 KASSERT(headlocked, ("trimthenstep6: " 1569 "tcp_close.2: head not locked")); 1570 tp = tcp_close(tp); 1571 break; 1572 1573 case TCPS_TIME_WAIT: 1574 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1575 ("timewait")); 1576 break; 1577 } 1578 } 1579 goto drop; 1580 } 1581 1582 /* 1583 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1584 * and it's less than ts_recent, drop it. 1585 */ 1586 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1587 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1588 1589 /* Check to see if ts_recent is over 24 days old. */ 1590 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1591 /* 1592 * Invalidate ts_recent. If this segment updates 1593 * ts_recent, the age will be reset later and ts_recent 1594 * will get a valid value. If it does not, setting 1595 * ts_recent to zero will at least satisfy the 1596 * requirement that zero be placed in the timestamp 1597 * echo reply when ts_recent isn't valid. The 1598 * age isn't reset until we get a valid ts_recent 1599 * because we don't want out-of-order segments to be 1600 * dropped when ts_recent is old. 1601 */ 1602 tp->ts_recent = 0; 1603 } else { 1604 tcpstat.tcps_rcvduppack++; 1605 tcpstat.tcps_rcvdupbyte += tlen; 1606 tcpstat.tcps_pawsdrop++; 1607 if (tlen) 1608 goto dropafterack; 1609 goto drop; 1610 } 1611 } 1612 1613 /* 1614 * In the SYN-RECEIVED state, validate that the packet belongs to 1615 * this connection before trimming the data to fit the receive 1616 * window. Check the sequence number versus IRS since we know 1617 * the sequence numbers haven't wrapped. This is a partial fix 1618 * for the "LAND" DoS attack. 1619 */ 1620 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1621 rstreason = BANDLIM_RST_OPENPORT; 1622 goto dropwithreset; 1623 } 1624 1625 todrop = tp->rcv_nxt - th->th_seq; 1626 if (todrop > 0) { 1627 if (thflags & TH_SYN) { 1628 thflags &= ~TH_SYN; 1629 th->th_seq++; 1630 if (th->th_urp > 1) 1631 th->th_urp--; 1632 else 1633 thflags &= ~TH_URG; 1634 todrop--; 1635 } 1636 /* 1637 * Following if statement from Stevens, vol. 2, p. 960. 1638 */ 1639 if (todrop > tlen 1640 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1641 /* 1642 * Any valid FIN must be to the left of the window. 1643 * At this point the FIN must be a duplicate or out 1644 * of sequence; drop it. 1645 */ 1646 thflags &= ~TH_FIN; 1647 1648 /* 1649 * Send an ACK to resynchronize and drop any data. 1650 * But keep on processing for RST or ACK. 1651 */ 1652 tp->t_flags |= TF_ACKNOW; 1653 todrop = tlen; 1654 tcpstat.tcps_rcvduppack++; 1655 tcpstat.tcps_rcvdupbyte += todrop; 1656 } else { 1657 tcpstat.tcps_rcvpartduppack++; 1658 tcpstat.tcps_rcvpartdupbyte += todrop; 1659 } 1660 drop_hdrlen += todrop; /* drop from the top afterwards */ 1661 th->th_seq += todrop; 1662 tlen -= todrop; 1663 if (th->th_urp > todrop) 1664 th->th_urp -= todrop; 1665 else { 1666 thflags &= ~TH_URG; 1667 th->th_urp = 0; 1668 } 1669 } 1670 1671 /* 1672 * If new data are received on a connection after the 1673 * user processes are gone, then RST the other end. 1674 */ 1675 if ((so->so_state & SS_NOFDREF) && 1676 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1677 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not " 1678 "locked")); 1679 tp = tcp_close(tp); 1680 tcpstat.tcps_rcvafterclose++; 1681 rstreason = BANDLIM_UNLIMITED; 1682 goto dropwithreset; 1683 } 1684 1685 /* 1686 * If segment ends after window, drop trailing data 1687 * (and PUSH and FIN); if nothing left, just ACK. 1688 */ 1689 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1690 if (todrop > 0) { 1691 tcpstat.tcps_rcvpackafterwin++; 1692 if (todrop >= tlen) { 1693 tcpstat.tcps_rcvbyteafterwin += tlen; 1694 /* 1695 * If a new connection request is received 1696 * while in TIME_WAIT, drop the old connection 1697 * and start over if the sequence numbers 1698 * are above the previous ones. 1699 */ 1700 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1701 if (thflags & TH_SYN && 1702 tp->t_state == TCPS_TIME_WAIT && 1703 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1704 KASSERT(headlocked, ("trimthenstep6: " 1705 "tcp_close.4: head not locked")); 1706 tp = tcp_close(tp); 1707 goto findpcb; 1708 } 1709 /* 1710 * If window is closed can only take segments at 1711 * window edge, and have to drop data and PUSH from 1712 * incoming segments. Continue processing, but 1713 * remember to ack. Otherwise, drop segment 1714 * and ack. 1715 */ 1716 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1717 tp->t_flags |= TF_ACKNOW; 1718 tcpstat.tcps_rcvwinprobe++; 1719 } else 1720 goto dropafterack; 1721 } else 1722 tcpstat.tcps_rcvbyteafterwin += todrop; 1723 m_adj(m, -todrop); 1724 tlen -= todrop; 1725 thflags &= ~(TH_PUSH|TH_FIN); 1726 } 1727 1728 /* 1729 * If last ACK falls within this segment's sequence numbers, 1730 * record its timestamp. 1731 * NOTE: 1732 * 1) That the test incorporates suggestions from the latest 1733 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1734 * 2) That updating only on newer timestamps interferes with 1735 * our earlier PAWS tests, so this check should be solely 1736 * predicated on the sequence space of this segment. 1737 * 3) That we modify the segment boundary check to be 1738 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1739 * instead of RFC1323's 1740 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1741 * This modified check allows us to overcome RFC1323's 1742 * limitations as described in Stevens TCP/IP Illustrated 1743 * Vol. 2 p.869. In such cases, we can still calculate the 1744 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1745 */ 1746 if ((to.to_flags & TOF_TS) != 0 && 1747 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1748 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1749 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1750 tp->ts_recent_age = ticks; 1751 tp->ts_recent = to.to_tsval; 1752 } 1753 1754 /* 1755 * If a SYN is in the window, then this is an 1756 * error and we send an RST and drop the connection. 1757 */ 1758 if (thflags & TH_SYN) { 1759 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: " 1760 "head not locked")); 1761 tp = tcp_drop(tp, ECONNRESET); 1762 rstreason = BANDLIM_UNLIMITED; 1763 goto drop; 1764 } 1765 1766 /* 1767 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1768 * flag is on (half-synchronized state), then queue data for 1769 * later processing; else drop segment and return. 1770 */ 1771 if ((thflags & TH_ACK) == 0) { 1772 if (tp->t_state == TCPS_SYN_RECEIVED || 1773 (tp->t_flags & TF_NEEDSYN)) 1774 goto step6; 1775 else 1776 goto drop; 1777 } 1778 1779 /* 1780 * Ack processing. 1781 */ 1782 switch (tp->t_state) { 1783 1784 /* 1785 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1786 * ESTABLISHED state and continue processing. 1787 * The ACK was checked above. 1788 */ 1789 case TCPS_SYN_RECEIVED: 1790 1791 tcpstat.tcps_connects++; 1792 soisconnected(so); 1793 /* Do window scaling? */ 1794 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1795 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1796 tp->snd_scale = tp->requested_s_scale; 1797 tp->rcv_scale = tp->request_r_scale; 1798 } 1799 /* 1800 * Make transitions: 1801 * SYN-RECEIVED -> ESTABLISHED 1802 * SYN-RECEIVED* -> FIN-WAIT-1 1803 */ 1804 tp->t_starttime = ticks; 1805 if (tp->t_flags & TF_NEEDFIN) { 1806 tp->t_state = TCPS_FIN_WAIT_1; 1807 tp->t_flags &= ~TF_NEEDFIN; 1808 } else { 1809 tp->t_state = TCPS_ESTABLISHED; 1810 callout_reset(tp->tt_keep, tcp_keepidle, 1811 tcp_timer_keep, tp); 1812 } 1813 /* 1814 * If segment contains data or ACK, will call tcp_reass() 1815 * later; if not, do so now to pass queued data to user. 1816 */ 1817 if (tlen == 0 && (thflags & TH_FIN) == 0) 1818 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1819 (struct mbuf *)0); 1820 tp->snd_wl1 = th->th_seq - 1; 1821 /* FALLTHROUGH */ 1822 1823 /* 1824 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1825 * ACKs. If the ack is in the range 1826 * tp->snd_una < th->th_ack <= tp->snd_max 1827 * then advance tp->snd_una to th->th_ack and drop 1828 * data from the retransmission queue. If this ACK reflects 1829 * more up to date window information we update our window information. 1830 */ 1831 case TCPS_ESTABLISHED: 1832 case TCPS_FIN_WAIT_1: 1833 case TCPS_FIN_WAIT_2: 1834 case TCPS_CLOSE_WAIT: 1835 case TCPS_CLOSING: 1836 case TCPS_LAST_ACK: 1837 case TCPS_TIME_WAIT: 1838 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1839 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1840 tcpstat.tcps_rcvacktoomuch++; 1841 goto dropafterack; 1842 } 1843 if (tp->sack_enable && 1844 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) 1845 tcp_sack_doack(tp, &to, th->th_ack); 1846 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1847 if (tlen == 0 && tiwin == tp->snd_wnd) { 1848 tcpstat.tcps_rcvdupack++; 1849 /* 1850 * If we have outstanding data (other than 1851 * a window probe), this is a completely 1852 * duplicate ack (ie, window info didn't 1853 * change), the ack is the biggest we've 1854 * seen and we've seen exactly our rexmt 1855 * threshhold of them, assume a packet 1856 * has been dropped and retransmit it. 1857 * Kludge snd_nxt & the congestion 1858 * window so we send only this one 1859 * packet. 1860 * 1861 * We know we're losing at the current 1862 * window size so do congestion avoidance 1863 * (set ssthresh to half the current window 1864 * and pull our congestion window back to 1865 * the new ssthresh). 1866 * 1867 * Dup acks mean that packets have left the 1868 * network (they're now cached at the receiver) 1869 * so bump cwnd by the amount in the receiver 1870 * to keep a constant cwnd packets in the 1871 * network. 1872 */ 1873 if (!callout_active(tp->tt_rexmt) || 1874 th->th_ack != tp->snd_una) 1875 tp->t_dupacks = 0; 1876 else if (++tp->t_dupacks > tcprexmtthresh || 1877 ((tcp_do_newreno || tp->sack_enable) && 1878 IN_FASTRECOVERY(tp))) { 1879 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1880 int awnd; 1881 1882 /* 1883 * Compute the amount of data in flight first. 1884 * We can inject new data into the pipe iff 1885 * we have less than 1/2 the original window's 1886 * worth of data in flight. 1887 */ 1888 awnd = (tp->snd_nxt - tp->snd_fack) + 1889 tp->sackhint.sack_bytes_rexmit; 1890 if (awnd < tp->snd_ssthresh) { 1891 tp->snd_cwnd += tp->t_maxseg; 1892 if (tp->snd_cwnd > tp->snd_ssthresh) 1893 tp->snd_cwnd = tp->snd_ssthresh; 1894 } 1895 } else 1896 tp->snd_cwnd += tp->t_maxseg; 1897 (void) tcp_output(tp); 1898 goto drop; 1899 } else if (tp->t_dupacks == tcprexmtthresh) { 1900 tcp_seq onxt = tp->snd_nxt; 1901 u_int win; 1902 1903 /* 1904 * If we're doing sack, check to 1905 * see if we're already in sack 1906 * recovery. If we're not doing sack, 1907 * check to see if we're in newreno 1908 * recovery. 1909 */ 1910 if (tp->sack_enable) { 1911 if (IN_FASTRECOVERY(tp)) { 1912 tp->t_dupacks = 0; 1913 break; 1914 } 1915 } else if (tcp_do_newreno) { 1916 if (SEQ_LEQ(th->th_ack, 1917 tp->snd_recover)) { 1918 tp->t_dupacks = 0; 1919 break; 1920 } 1921 } 1922 win = min(tp->snd_wnd, tp->snd_cwnd) / 1923 2 / tp->t_maxseg; 1924 if (win < 2) 1925 win = 2; 1926 tp->snd_ssthresh = win * tp->t_maxseg; 1927 ENTER_FASTRECOVERY(tp); 1928 tp->snd_recover = tp->snd_max; 1929 callout_stop(tp->tt_rexmt); 1930 tp->t_rtttime = 0; 1931 if (tp->sack_enable) { 1932 tcpstat.tcps_sack_recovery_episode++; 1933 tp->sack_newdata = tp->snd_nxt; 1934 tp->snd_cwnd = tp->t_maxseg; 1935 (void) tcp_output(tp); 1936 goto drop; 1937 } 1938 tp->snd_nxt = th->th_ack; 1939 tp->snd_cwnd = tp->t_maxseg; 1940 (void) tcp_output(tp); 1941 KASSERT(tp->snd_limited <= 2, 1942 ("tp->snd_limited too big")); 1943 tp->snd_cwnd = tp->snd_ssthresh + 1944 tp->t_maxseg * 1945 (tp->t_dupacks - tp->snd_limited); 1946 if (SEQ_GT(onxt, tp->snd_nxt)) 1947 tp->snd_nxt = onxt; 1948 goto drop; 1949 } else if (tcp_do_rfc3042) { 1950 u_long oldcwnd = tp->snd_cwnd; 1951 tcp_seq oldsndmax = tp->snd_max; 1952 u_int sent; 1953 1954 KASSERT(tp->t_dupacks == 1 || 1955 tp->t_dupacks == 2, 1956 ("dupacks not 1 or 2")); 1957 if (tp->t_dupacks == 1) 1958 tp->snd_limited = 0; 1959 tp->snd_cwnd = 1960 (tp->snd_nxt - tp->snd_una) + 1961 (tp->t_dupacks - tp->snd_limited) * 1962 tp->t_maxseg; 1963 (void) tcp_output(tp); 1964 sent = tp->snd_max - oldsndmax; 1965 if (sent > tp->t_maxseg) { 1966 KASSERT((tp->t_dupacks == 2 && 1967 tp->snd_limited == 0) || 1968 (sent == tp->t_maxseg + 1 && 1969 tp->t_flags & TF_SENTFIN), 1970 ("sent too much")); 1971 tp->snd_limited = 2; 1972 } else if (sent > 0) 1973 ++tp->snd_limited; 1974 tp->snd_cwnd = oldcwnd; 1975 goto drop; 1976 } 1977 } else 1978 tp->t_dupacks = 0; 1979 break; 1980 } 1981 1982 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1983 1984 /* 1985 * If the congestion window was inflated to account 1986 * for the other side's cached packets, retract it. 1987 */ 1988 if (tcp_do_newreno || tp->sack_enable) { 1989 if (IN_FASTRECOVERY(tp)) { 1990 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1991 if (tp->sack_enable) 1992 tcp_sack_partialack(tp, th); 1993 else 1994 tcp_newreno_partial_ack(tp, th); 1995 } else { 1996 /* 1997 * Out of fast recovery. 1998 * Window inflation should have left us 1999 * with approximately snd_ssthresh 2000 * outstanding data. 2001 * But in case we would be inclined to 2002 * send a burst, better to do it via 2003 * the slow start mechanism. 2004 */ 2005 if (SEQ_GT(th->th_ack + 2006 tp->snd_ssthresh, 2007 tp->snd_max)) 2008 tp->snd_cwnd = tp->snd_max - 2009 th->th_ack + 2010 tp->t_maxseg; 2011 else 2012 tp->snd_cwnd = tp->snd_ssthresh; 2013 } 2014 } 2015 } else { 2016 if (tp->t_dupacks >= tcprexmtthresh && 2017 tp->snd_cwnd > tp->snd_ssthresh) 2018 tp->snd_cwnd = tp->snd_ssthresh; 2019 } 2020 tp->t_dupacks = 0; 2021 /* 2022 * If we reach this point, ACK is not a duplicate, 2023 * i.e., it ACKs something we sent. 2024 */ 2025 if (tp->t_flags & TF_NEEDSYN) { 2026 /* 2027 * T/TCP: Connection was half-synchronized, and our 2028 * SYN has been ACK'd (so connection is now fully 2029 * synchronized). Go to non-starred state, 2030 * increment snd_una for ACK of SYN, and check if 2031 * we can do window scaling. 2032 */ 2033 tp->t_flags &= ~TF_NEEDSYN; 2034 tp->snd_una++; 2035 /* Do window scaling? */ 2036 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2037 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2038 tp->snd_scale = tp->requested_s_scale; 2039 tp->rcv_scale = tp->request_r_scale; 2040 } 2041 } 2042 2043process_ACK: 2044 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 2045 "locked")); 2046 INP_LOCK_ASSERT(inp); 2047 2048 acked = th->th_ack - tp->snd_una; 2049 tcpstat.tcps_rcvackpack++; 2050 tcpstat.tcps_rcvackbyte += acked; 2051 2052 /* 2053 * If we just performed our first retransmit, and the ACK 2054 * arrives within our recovery window, then it was a mistake 2055 * to do the retransmit in the first place. Recover our 2056 * original cwnd and ssthresh, and proceed to transmit where 2057 * we left off. 2058 */ 2059 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2060 ++tcpstat.tcps_sndrexmitbad; 2061 tp->snd_cwnd = tp->snd_cwnd_prev; 2062 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2063 tp->snd_recover = tp->snd_recover_prev; 2064 if (tp->t_flags & TF_WASFRECOVERY) 2065 ENTER_FASTRECOVERY(tp); 2066 tp->snd_nxt = tp->snd_max; 2067 tp->t_badrxtwin = 0; /* XXX probably not required */ 2068 } 2069 2070 /* 2071 * If we have a timestamp reply, update smoothed 2072 * round trip time. If no timestamp is present but 2073 * transmit timer is running and timed sequence 2074 * number was acked, update smoothed round trip time. 2075 * Since we now have an rtt measurement, cancel the 2076 * timer backoff (cf., Phil Karn's retransmit alg.). 2077 * Recompute the initial retransmit timer. 2078 * 2079 * Some boxes send broken timestamp replies 2080 * during the SYN+ACK phase, ignore 2081 * timestamps of 0 or we could calculate a 2082 * huge RTT and blow up the retransmit timer. 2083 */ 2084 if ((to.to_flags & TOF_TS) != 0 && 2085 to.to_tsecr) { 2086 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2087 tp->t_rttlow = ticks - to.to_tsecr; 2088 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2089 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2090 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2091 tp->t_rttlow = ticks - tp->t_rtttime; 2092 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2093 } 2094 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2095 2096 /* 2097 * If all outstanding data is acked, stop retransmit 2098 * timer and remember to restart (more output or persist). 2099 * If there is more data to be acked, restart retransmit 2100 * timer, using current (possibly backed-off) value. 2101 */ 2102 if (th->th_ack == tp->snd_max) { 2103 callout_stop(tp->tt_rexmt); 2104 needoutput = 1; 2105 } else if (!callout_active(tp->tt_persist)) 2106 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2107 tcp_timer_rexmt, tp); 2108 2109 /* 2110 * If no data (only SYN) was ACK'd, 2111 * skip rest of ACK processing. 2112 */ 2113 if (acked == 0) 2114 goto step6; 2115 2116 /* 2117 * When new data is acked, open the congestion window. 2118 * If the window gives us less than ssthresh packets 2119 * in flight, open exponentially (maxseg per packet). 2120 * Otherwise open linearly: maxseg per window 2121 * (maxseg^2 / cwnd per packet). 2122 */ 2123 if ((!tcp_do_newreno && !tp->sack_enable) || 2124 !IN_FASTRECOVERY(tp)) { 2125 register u_int cw = tp->snd_cwnd; 2126 register u_int incr = tp->t_maxseg; 2127 if (cw > tp->snd_ssthresh) 2128 incr = incr * incr / cw; 2129 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2130 } 2131 SOCKBUF_LOCK(&so->so_snd); 2132 if (acked > so->so_snd.sb_cc) { 2133 tp->snd_wnd -= so->so_snd.sb_cc; 2134 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2135 ourfinisacked = 1; 2136 } else { 2137 sbdrop_locked(&so->so_snd, acked); 2138 tp->snd_wnd -= acked; 2139 ourfinisacked = 0; 2140 } 2141 sowwakeup_locked(so); 2142 /* detect una wraparound */ 2143 if ((tcp_do_newreno || tp->sack_enable) && 2144 !IN_FASTRECOVERY(tp) && 2145 SEQ_GT(tp->snd_una, tp->snd_recover) && 2146 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2147 tp->snd_recover = th->th_ack - 1; 2148 if ((tcp_do_newreno || tp->sack_enable) && 2149 IN_FASTRECOVERY(tp) && 2150 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2151 EXIT_FASTRECOVERY(tp); 2152 tp->snd_una = th->th_ack; 2153 if (tp->sack_enable) { 2154 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2155 tp->snd_recover = tp->snd_una; 2156 } 2157 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2158 tp->snd_nxt = tp->snd_una; 2159 2160 switch (tp->t_state) { 2161 2162 /* 2163 * In FIN_WAIT_1 STATE in addition to the processing 2164 * for the ESTABLISHED state if our FIN is now acknowledged 2165 * then enter FIN_WAIT_2. 2166 */ 2167 case TCPS_FIN_WAIT_1: 2168 if (ourfinisacked) { 2169 /* 2170 * If we can't receive any more 2171 * data, then closing user can proceed. 2172 * Starting the timer is contrary to the 2173 * specification, but if we don't get a FIN 2174 * we'll hang forever. 2175 */ 2176 /* XXXjl 2177 * we should release the tp also, and use a 2178 * compressed state. 2179 */ 2180 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2181 soisdisconnected(so); 2182 callout_reset(tp->tt_2msl, tcp_maxidle, 2183 tcp_timer_2msl, tp); 2184 } 2185 tp->t_state = TCPS_FIN_WAIT_2; 2186 } 2187 break; 2188 2189 /* 2190 * In CLOSING STATE in addition to the processing for 2191 * the ESTABLISHED state if the ACK acknowledges our FIN 2192 * then enter the TIME-WAIT state, otherwise ignore 2193 * the segment. 2194 */ 2195 case TCPS_CLOSING: 2196 if (ourfinisacked) { 2197 KASSERT(headlocked, ("tcp_input: process_ACK: " 2198 "head not locked")); 2199 tcp_twstart(tp); 2200 INP_INFO_WUNLOCK(&tcbinfo); 2201 m_freem(m); 2202 return; 2203 } 2204 break; 2205 2206 /* 2207 * In LAST_ACK, we may still be waiting for data to drain 2208 * and/or to be acked, as well as for the ack of our FIN. 2209 * If our FIN is now acknowledged, delete the TCB, 2210 * enter the closed state and return. 2211 */ 2212 case TCPS_LAST_ACK: 2213 if (ourfinisacked) { 2214 KASSERT(headlocked, ("tcp_input: process_ACK:" 2215 " tcp_close: head not locked")); 2216 tp = tcp_close(tp); 2217 goto drop; 2218 } 2219 break; 2220 2221 /* 2222 * In TIME_WAIT state the only thing that should arrive 2223 * is a retransmission of the remote FIN. Acknowledge 2224 * it and restart the finack timer. 2225 */ 2226 case TCPS_TIME_WAIT: 2227 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2228 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2229 tcp_timer_2msl, tp); 2230 goto dropafterack; 2231 } 2232 } 2233 2234step6: 2235 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2236 INP_LOCK_ASSERT(inp); 2237 2238 /* 2239 * Update window information. 2240 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2241 */ 2242 if ((thflags & TH_ACK) && 2243 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2244 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2245 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2246 /* keep track of pure window updates */ 2247 if (tlen == 0 && 2248 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2249 tcpstat.tcps_rcvwinupd++; 2250 tp->snd_wnd = tiwin; 2251 tp->snd_wl1 = th->th_seq; 2252 tp->snd_wl2 = th->th_ack; 2253 if (tp->snd_wnd > tp->max_sndwnd) 2254 tp->max_sndwnd = tp->snd_wnd; 2255 needoutput = 1; 2256 } 2257 2258 /* 2259 * Process segments with URG. 2260 */ 2261 if ((thflags & TH_URG) && th->th_urp && 2262 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2263 /* 2264 * This is a kludge, but if we receive and accept 2265 * random urgent pointers, we'll crash in 2266 * soreceive. It's hard to imagine someone 2267 * actually wanting to send this much urgent data. 2268 */ 2269 SOCKBUF_LOCK(&so->so_rcv); 2270 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2271 th->th_urp = 0; /* XXX */ 2272 thflags &= ~TH_URG; /* XXX */ 2273 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2274 goto dodata; /* XXX */ 2275 } 2276 /* 2277 * If this segment advances the known urgent pointer, 2278 * then mark the data stream. This should not happen 2279 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2280 * a FIN has been received from the remote side. 2281 * In these states we ignore the URG. 2282 * 2283 * According to RFC961 (Assigned Protocols), 2284 * the urgent pointer points to the last octet 2285 * of urgent data. We continue, however, 2286 * to consider it to indicate the first octet 2287 * of data past the urgent section as the original 2288 * spec states (in one of two places). 2289 */ 2290 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2291 tp->rcv_up = th->th_seq + th->th_urp; 2292 so->so_oobmark = so->so_rcv.sb_cc + 2293 (tp->rcv_up - tp->rcv_nxt) - 1; 2294 if (so->so_oobmark == 0) 2295 so->so_rcv.sb_state |= SBS_RCVATMARK; 2296 sohasoutofband(so); 2297 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2298 } 2299 SOCKBUF_UNLOCK(&so->so_rcv); 2300 /* 2301 * Remove out of band data so doesn't get presented to user. 2302 * This can happen independent of advancing the URG pointer, 2303 * but if two URG's are pending at once, some out-of-band 2304 * data may creep in... ick. 2305 */ 2306 if (th->th_urp <= (u_long)tlen && 2307 !(so->so_options & SO_OOBINLINE)) { 2308 /* hdr drop is delayed */ 2309 tcp_pulloutofband(so, th, m, drop_hdrlen); 2310 } 2311 } else { 2312 /* 2313 * If no out of band data is expected, 2314 * pull receive urgent pointer along 2315 * with the receive window. 2316 */ 2317 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2318 tp->rcv_up = tp->rcv_nxt; 2319 } 2320dodata: /* XXX */ 2321 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2322 INP_LOCK_ASSERT(inp); 2323 2324 /* 2325 * Process the segment text, merging it into the TCP sequencing queue, 2326 * and arranging for acknowledgment of receipt if necessary. 2327 * This process logically involves adjusting tp->rcv_wnd as data 2328 * is presented to the user (this happens in tcp_usrreq.c, 2329 * case PRU_RCVD). If a FIN has already been received on this 2330 * connection then we just ignore the text. 2331 */ 2332 if ((tlen || (thflags & TH_FIN)) && 2333 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2334 tcp_seq save_start = th->th_seq; 2335 tcp_seq save_end = th->th_seq + tlen; 2336 m_adj(m, drop_hdrlen); /* delayed header drop */ 2337 /* 2338 * Insert segment which includes th into TCP reassembly queue 2339 * with control block tp. Set thflags to whether reassembly now 2340 * includes a segment with FIN. This handles the common case 2341 * inline (segment is the next to be received on an established 2342 * connection, and the queue is empty), avoiding linkage into 2343 * and removal from the queue and repetition of various 2344 * conversions. 2345 * Set DELACK for segments received in order, but ack 2346 * immediately when segments are out of order (so 2347 * fast retransmit can work). 2348 */ 2349 if (th->th_seq == tp->rcv_nxt && 2350 LIST_EMPTY(&tp->t_segq) && 2351 TCPS_HAVEESTABLISHED(tp->t_state)) { 2352 if (DELAY_ACK(tp)) 2353 tp->t_flags |= TF_DELACK; 2354 else 2355 tp->t_flags |= TF_ACKNOW; 2356 tp->rcv_nxt += tlen; 2357 thflags = th->th_flags & TH_FIN; 2358 tcpstat.tcps_rcvpack++; 2359 tcpstat.tcps_rcvbyte += tlen; 2360 ND6_HINT(tp); 2361 SOCKBUF_LOCK(&so->so_rcv); 2362 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2363 m_freem(m); 2364 else 2365 sbappendstream_locked(&so->so_rcv, m); 2366 sorwakeup_locked(so); 2367 } else { 2368 thflags = tcp_reass(tp, th, &tlen, m); 2369 tp->t_flags |= TF_ACKNOW; 2370 } 2371 if (tlen > 0 && tp->sack_enable) 2372 tcp_update_sack_list(tp, save_start, save_end); 2373 /* 2374 * Note the amount of data that peer has sent into 2375 * our window, in order to estimate the sender's 2376 * buffer size. 2377 */ 2378 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2379 } else { 2380 m_freem(m); 2381 thflags &= ~TH_FIN; 2382 } 2383 2384 /* 2385 * If FIN is received ACK the FIN and let the user know 2386 * that the connection is closing. 2387 */ 2388 if (thflags & TH_FIN) { 2389 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2390 socantrcvmore(so); 2391 /* 2392 * If connection is half-synchronized 2393 * (ie NEEDSYN flag on) then delay ACK, 2394 * so it may be piggybacked when SYN is sent. 2395 * Otherwise, since we received a FIN then no 2396 * more input can be expected, send ACK now. 2397 */ 2398 if (tp->t_flags & TF_NEEDSYN) 2399 tp->t_flags |= TF_DELACK; 2400 else 2401 tp->t_flags |= TF_ACKNOW; 2402 tp->rcv_nxt++; 2403 } 2404 switch (tp->t_state) { 2405 2406 /* 2407 * In SYN_RECEIVED and ESTABLISHED STATES 2408 * enter the CLOSE_WAIT state. 2409 */ 2410 case TCPS_SYN_RECEIVED: 2411 tp->t_starttime = ticks; 2412 /*FALLTHROUGH*/ 2413 case TCPS_ESTABLISHED: 2414 tp->t_state = TCPS_CLOSE_WAIT; 2415 break; 2416 2417 /* 2418 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2419 * enter the CLOSING state. 2420 */ 2421 case TCPS_FIN_WAIT_1: 2422 tp->t_state = TCPS_CLOSING; 2423 break; 2424 2425 /* 2426 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2427 * starting the time-wait timer, turning off the other 2428 * standard timers. 2429 */ 2430 case TCPS_FIN_WAIT_2: 2431 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2432 "TCP_FIN_WAIT_2: head not locked")); 2433 tcp_twstart(tp); 2434 INP_INFO_WUNLOCK(&tcbinfo); 2435 return; 2436 2437 /* 2438 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2439 */ 2440 case TCPS_TIME_WAIT: 2441 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2442 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2443 tcp_timer_2msl, tp); 2444 break; 2445 } 2446 } 2447 INP_INFO_WUNLOCK(&tcbinfo); 2448 headlocked = 0; 2449#ifdef TCPDEBUG 2450 if (so->so_options & SO_DEBUG) 2451 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2452 &tcp_savetcp, 0); 2453#endif 2454 2455 /* 2456 * Return any desired output. 2457 */ 2458 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2459 (void) tcp_output(tp); 2460 2461check_delack: 2462 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2463 INP_LOCK_ASSERT(inp); 2464 if (tp->t_flags & TF_DELACK) { 2465 tp->t_flags &= ~TF_DELACK; 2466 callout_reset(tp->tt_delack, tcp_delacktime, 2467 tcp_timer_delack, tp); 2468 } 2469 INP_UNLOCK(inp); 2470 return; 2471 2472dropafterack: 2473 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2474 /* 2475 * Generate an ACK dropping incoming segment if it occupies 2476 * sequence space, where the ACK reflects our state. 2477 * 2478 * We can now skip the test for the RST flag since all 2479 * paths to this code happen after packets containing 2480 * RST have been dropped. 2481 * 2482 * In the SYN-RECEIVED state, don't send an ACK unless the 2483 * segment we received passes the SYN-RECEIVED ACK test. 2484 * If it fails send a RST. This breaks the loop in the 2485 * "LAND" DoS attack, and also prevents an ACK storm 2486 * between two listening ports that have been sent forged 2487 * SYN segments, each with the source address of the other. 2488 */ 2489 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2490 (SEQ_GT(tp->snd_una, th->th_ack) || 2491 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2492 rstreason = BANDLIM_RST_OPENPORT; 2493 goto dropwithreset; 2494 } 2495#ifdef TCPDEBUG 2496 if (so->so_options & SO_DEBUG) 2497 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2498 &tcp_savetcp, 0); 2499#endif 2500 KASSERT(headlocked, ("headlocked should be 1")); 2501 INP_INFO_WUNLOCK(&tcbinfo); 2502 tp->t_flags |= TF_ACKNOW; 2503 (void) tcp_output(tp); 2504 INP_UNLOCK(inp); 2505 m_freem(m); 2506 return; 2507 2508dropwithreset: 2509 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2510 /* 2511 * Generate a RST, dropping incoming segment. 2512 * Make ACK acceptable to originator of segment. 2513 * Don't bother to respond if destination was broadcast/multicast. 2514 */ 2515 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2516 goto drop; 2517 if (isipv6) { 2518 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2519 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2520 goto drop; 2521 } else { 2522 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2523 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2524 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2525 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2526 goto drop; 2527 } 2528 /* IPv6 anycast check is done at tcp6_input() */ 2529 2530 /* 2531 * Perform bandwidth limiting. 2532 */ 2533 if (badport_bandlim(rstreason) < 0) 2534 goto drop; 2535 2536#ifdef TCPDEBUG 2537 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2538 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2539 &tcp_savetcp, 0); 2540#endif 2541 2542 if (thflags & TH_ACK) 2543 /* mtod() below is safe as long as hdr dropping is delayed */ 2544 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2545 TH_RST); 2546 else { 2547 if (thflags & TH_SYN) 2548 tlen++; 2549 /* mtod() below is safe as long as hdr dropping is delayed */ 2550 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2551 (tcp_seq)0, TH_RST|TH_ACK); 2552 } 2553 2554 if (tp) 2555 INP_UNLOCK(inp); 2556 if (headlocked) 2557 INP_INFO_WUNLOCK(&tcbinfo); 2558 return; 2559 2560drop: 2561 /* 2562 * Drop space held by incoming segment and return. 2563 */ 2564#ifdef TCPDEBUG 2565 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2566 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2567 &tcp_savetcp, 0); 2568#endif 2569 if (tp) 2570 INP_UNLOCK(inp); 2571 if (headlocked) 2572 INP_INFO_WUNLOCK(&tcbinfo); 2573 m_freem(m); 2574 return; 2575} 2576 2577/* 2578 * Parse TCP options and place in tcpopt. 2579 */ 2580static void 2581tcp_dooptions(to, cp, cnt, is_syn) 2582 struct tcpopt *to; 2583 u_char *cp; 2584 int cnt; 2585 int is_syn; 2586{ 2587 int opt, optlen; 2588 2589 to->to_flags = 0; 2590 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2591 opt = cp[0]; 2592 if (opt == TCPOPT_EOL) 2593 break; 2594 if (opt == TCPOPT_NOP) 2595 optlen = 1; 2596 else { 2597 if (cnt < 2) 2598 break; 2599 optlen = cp[1]; 2600 if (optlen < 2 || optlen > cnt) 2601 break; 2602 } 2603 switch (opt) { 2604 case TCPOPT_MAXSEG: 2605 if (optlen != TCPOLEN_MAXSEG) 2606 continue; 2607 if (!is_syn) 2608 continue; 2609 to->to_flags |= TOF_MSS; 2610 bcopy((char *)cp + 2, 2611 (char *)&to->to_mss, sizeof(to->to_mss)); 2612 to->to_mss = ntohs(to->to_mss); 2613 break; 2614 case TCPOPT_WINDOW: 2615 if (optlen != TCPOLEN_WINDOW) 2616 continue; 2617 if (! is_syn) 2618 continue; 2619 to->to_flags |= TOF_SCALE; 2620 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2621 break; 2622 case TCPOPT_TIMESTAMP: 2623 if (optlen != TCPOLEN_TIMESTAMP) 2624 continue; 2625 to->to_flags |= TOF_TS; 2626 bcopy((char *)cp + 2, 2627 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2628 to->to_tsval = ntohl(to->to_tsval); 2629 bcopy((char *)cp + 6, 2630 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2631 to->to_tsecr = ntohl(to->to_tsecr); 2632 /* 2633 * If echoed timestamp is later than the current time, 2634 * fall back to non RFC1323 RTT calculation. 2635 */ 2636 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks)) 2637 to->to_tsecr = 0; 2638 break; 2639#ifdef TCP_SIGNATURE 2640 /* 2641 * XXX In order to reply to a host which has set the 2642 * TCP_SIGNATURE option in its initial SYN, we have to 2643 * record the fact that the option was observed here 2644 * for the syncache code to perform the correct response. 2645 */ 2646 case TCPOPT_SIGNATURE: 2647 if (optlen != TCPOLEN_SIGNATURE) 2648 continue; 2649 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2650 break; 2651#endif 2652 case TCPOPT_SACK_PERMITTED: 2653 if (!tcp_do_sack || 2654 optlen != TCPOLEN_SACK_PERMITTED) 2655 continue; 2656 if (is_syn) { 2657 /* MUST only be set on SYN */ 2658 to->to_flags |= TOF_SACK; 2659 } 2660 break; 2661 case TCPOPT_SACK: 2662 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2663 continue; 2664 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2665 to->to_sacks = cp + 2; 2666 tcpstat.tcps_sack_rcv_blocks++; 2667 break; 2668 default: 2669 continue; 2670 } 2671 } 2672} 2673 2674/* 2675 * Pull out of band byte out of a segment so 2676 * it doesn't appear in the user's data queue. 2677 * It is still reflected in the segment length for 2678 * sequencing purposes. 2679 */ 2680static void 2681tcp_pulloutofband(so, th, m, off) 2682 struct socket *so; 2683 struct tcphdr *th; 2684 register struct mbuf *m; 2685 int off; /* delayed to be droped hdrlen */ 2686{ 2687 int cnt = off + th->th_urp - 1; 2688 2689 while (cnt >= 0) { 2690 if (m->m_len > cnt) { 2691 char *cp = mtod(m, caddr_t) + cnt; 2692 struct tcpcb *tp = sototcpcb(so); 2693 2694 tp->t_iobc = *cp; 2695 tp->t_oobflags |= TCPOOB_HAVEDATA; 2696 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2697 m->m_len--; 2698 if (m->m_flags & M_PKTHDR) 2699 m->m_pkthdr.len--; 2700 return; 2701 } 2702 cnt -= m->m_len; 2703 m = m->m_next; 2704 if (m == 0) 2705 break; 2706 } 2707 panic("tcp_pulloutofband"); 2708} 2709 2710/* 2711 * Collect new round-trip time estimate 2712 * and update averages and current timeout. 2713 */ 2714static void 2715tcp_xmit_timer(tp, rtt) 2716 register struct tcpcb *tp; 2717 int rtt; 2718{ 2719 register int delta; 2720 2721 INP_LOCK_ASSERT(tp->t_inpcb); 2722 2723 tcpstat.tcps_rttupdated++; 2724 tp->t_rttupdated++; 2725 if (tp->t_srtt != 0) { 2726 /* 2727 * srtt is stored as fixed point with 5 bits after the 2728 * binary point (i.e., scaled by 8). The following magic 2729 * is equivalent to the smoothing algorithm in rfc793 with 2730 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2731 * point). Adjust rtt to origin 0. 2732 */ 2733 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2734 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2735 2736 if ((tp->t_srtt += delta) <= 0) 2737 tp->t_srtt = 1; 2738 2739 /* 2740 * We accumulate a smoothed rtt variance (actually, a 2741 * smoothed mean difference), then set the retransmit 2742 * timer to smoothed rtt + 4 times the smoothed variance. 2743 * rttvar is stored as fixed point with 4 bits after the 2744 * binary point (scaled by 16). The following is 2745 * equivalent to rfc793 smoothing with an alpha of .75 2746 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2747 * rfc793's wired-in beta. 2748 */ 2749 if (delta < 0) 2750 delta = -delta; 2751 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2752 if ((tp->t_rttvar += delta) <= 0) 2753 tp->t_rttvar = 1; 2754 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2755 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2756 } else { 2757 /* 2758 * No rtt measurement yet - use the unsmoothed rtt. 2759 * Set the variance to half the rtt (so our first 2760 * retransmit happens at 3*rtt). 2761 */ 2762 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2763 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2764 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2765 } 2766 tp->t_rtttime = 0; 2767 tp->t_rxtshift = 0; 2768 2769 /* 2770 * the retransmit should happen at rtt + 4 * rttvar. 2771 * Because of the way we do the smoothing, srtt and rttvar 2772 * will each average +1/2 tick of bias. When we compute 2773 * the retransmit timer, we want 1/2 tick of rounding and 2774 * 1 extra tick because of +-1/2 tick uncertainty in the 2775 * firing of the timer. The bias will give us exactly the 2776 * 1.5 tick we need. But, because the bias is 2777 * statistical, we have to test that we don't drop below 2778 * the minimum feasible timer (which is 2 ticks). 2779 */ 2780 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2781 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2782 2783 /* 2784 * We received an ack for a packet that wasn't retransmitted; 2785 * it is probably safe to discard any error indications we've 2786 * received recently. This isn't quite right, but close enough 2787 * for now (a route might have failed after we sent a segment, 2788 * and the return path might not be symmetrical). 2789 */ 2790 tp->t_softerror = 0; 2791} 2792 2793/* 2794 * Determine a reasonable value for maxseg size. 2795 * If the route is known, check route for mtu. 2796 * If none, use an mss that can be handled on the outgoing 2797 * interface without forcing IP to fragment; if bigger than 2798 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2799 * to utilize large mbufs. If no route is found, route has no mtu, 2800 * or the destination isn't local, use a default, hopefully conservative 2801 * size (usually 512 or the default IP max size, but no more than the mtu 2802 * of the interface), as we can't discover anything about intervening 2803 * gateways or networks. We also initialize the congestion/slow start 2804 * window to be a single segment if the destination isn't local. 2805 * While looking at the routing entry, we also initialize other path-dependent 2806 * parameters from pre-set or cached values in the routing entry. 2807 * 2808 * Also take into account the space needed for options that we 2809 * send regularly. Make maxseg shorter by that amount to assure 2810 * that we can send maxseg amount of data even when the options 2811 * are present. Store the upper limit of the length of options plus 2812 * data in maxopd. 2813 * 2814 * 2815 * In case of T/TCP, we call this routine during implicit connection 2816 * setup as well (offer = -1), to initialize maxseg from the cached 2817 * MSS of our peer. 2818 * 2819 * NOTE that this routine is only called when we process an incoming 2820 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2821 */ 2822void 2823tcp_mss(tp, offer) 2824 struct tcpcb *tp; 2825 int offer; 2826{ 2827 int rtt, mss; 2828 u_long bufsize; 2829 u_long maxmtu; 2830 struct inpcb *inp = tp->t_inpcb; 2831 struct socket *so; 2832 struct hc_metrics_lite metrics; 2833 int origoffer = offer; 2834#ifdef INET6 2835 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2836 size_t min_protoh = isipv6 ? 2837 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2838 sizeof (struct tcpiphdr); 2839#else 2840 const size_t min_protoh = sizeof(struct tcpiphdr); 2841#endif 2842 2843 /* initialize */ 2844#ifdef INET6 2845 if (isipv6) { 2846 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2847 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2848 } else 2849#endif 2850 { 2851 maxmtu = tcp_maxmtu(&inp->inp_inc); 2852 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2853 } 2854 so = inp->inp_socket; 2855 2856 /* 2857 * no route to sender, stay with default mss and return 2858 */ 2859 if (maxmtu == 0) 2860 return; 2861 2862 /* what have we got? */ 2863 switch (offer) { 2864 case 0: 2865 /* 2866 * Offer == 0 means that there was no MSS on the SYN 2867 * segment, in this case we use tcp_mssdflt. 2868 */ 2869 offer = 2870#ifdef INET6 2871 isipv6 ? tcp_v6mssdflt : 2872#endif 2873 tcp_mssdflt; 2874 break; 2875 2876 case -1: 2877 /* 2878 * Offer == -1 means that we didn't receive SYN yet. 2879 */ 2880 /* FALLTHROUGH */ 2881 2882 default: 2883 /* 2884 * Prevent DoS attack with too small MSS. Round up 2885 * to at least minmss. 2886 */ 2887 offer = max(offer, tcp_minmss); 2888 /* 2889 * Sanity check: make sure that maxopd will be large 2890 * enough to allow some data on segments even if the 2891 * all the option space is used (40bytes). Otherwise 2892 * funny things may happen in tcp_output. 2893 */ 2894 offer = max(offer, 64); 2895 } 2896 2897 /* 2898 * rmx information is now retrieved from tcp_hostcache 2899 */ 2900 tcp_hc_get(&inp->inp_inc, &metrics); 2901 2902 /* 2903 * if there's a discovered mtu int tcp hostcache, use it 2904 * else, use the link mtu. 2905 */ 2906 if (metrics.rmx_mtu) 2907 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2908 else { 2909#ifdef INET6 2910 if (isipv6) { 2911 mss = maxmtu - min_protoh; 2912 if (!path_mtu_discovery && 2913 !in6_localaddr(&inp->in6p_faddr)) 2914 mss = min(mss, tcp_v6mssdflt); 2915 } else 2916#endif 2917 { 2918 mss = maxmtu - min_protoh; 2919 if (!path_mtu_discovery && 2920 !in_localaddr(inp->inp_faddr)) 2921 mss = min(mss, tcp_mssdflt); 2922 } 2923 } 2924 mss = min(mss, offer); 2925 2926 /* 2927 * maxopd stores the maximum length of data AND options 2928 * in a segment; maxseg is the amount of data in a normal 2929 * segment. We need to store this value (maxopd) apart 2930 * from maxseg, because now every segment carries options 2931 * and thus we normally have somewhat less data in segments. 2932 */ 2933 tp->t_maxopd = mss; 2934 2935 /* 2936 * origoffer==-1 indicates, that no segments were received yet. 2937 * In this case we just guess. 2938 */ 2939 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2940 (origoffer == -1 || 2941 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2942 mss -= TCPOLEN_TSTAMP_APPA; 2943 tp->t_maxseg = mss; 2944 2945#if (MCLBYTES & (MCLBYTES - 1)) == 0 2946 if (mss > MCLBYTES) 2947 mss &= ~(MCLBYTES-1); 2948#else 2949 if (mss > MCLBYTES) 2950 mss = mss / MCLBYTES * MCLBYTES; 2951#endif 2952 tp->t_maxseg = mss; 2953 2954 /* 2955 * If there's a pipesize, change the socket buffer to that size, 2956 * don't change if sb_hiwat is different than default (then it 2957 * has been changed on purpose with setsockopt). 2958 * Make the socket buffers an integral number of mss units; 2959 * if the mss is larger than the socket buffer, decrease the mss. 2960 */ 2961 SOCKBUF_LOCK(&so->so_snd); 2962 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2963 bufsize = metrics.rmx_sendpipe; 2964 else 2965 bufsize = so->so_snd.sb_hiwat; 2966 if (bufsize < mss) 2967 mss = bufsize; 2968 else { 2969 bufsize = roundup(bufsize, mss); 2970 if (bufsize > sb_max) 2971 bufsize = sb_max; 2972 if (bufsize > so->so_snd.sb_hiwat) 2973 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2974 } 2975 SOCKBUF_UNLOCK(&so->so_snd); 2976 tp->t_maxseg = mss; 2977 2978 SOCKBUF_LOCK(&so->so_rcv); 2979 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2980 bufsize = metrics.rmx_recvpipe; 2981 else 2982 bufsize = so->so_rcv.sb_hiwat; 2983 if (bufsize > mss) { 2984 bufsize = roundup(bufsize, mss); 2985 if (bufsize > sb_max) 2986 bufsize = sb_max; 2987 if (bufsize > so->so_rcv.sb_hiwat) 2988 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2989 } 2990 SOCKBUF_UNLOCK(&so->so_rcv); 2991 /* 2992 * While we're here, check the others too 2993 */ 2994 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2995 tp->t_srtt = rtt; 2996 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2997 tcpstat.tcps_usedrtt++; 2998 if (metrics.rmx_rttvar) { 2999 tp->t_rttvar = metrics.rmx_rttvar; 3000 tcpstat.tcps_usedrttvar++; 3001 } else { 3002 /* default variation is +- 1 rtt */ 3003 tp->t_rttvar = 3004 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3005 } 3006 TCPT_RANGESET(tp->t_rxtcur, 3007 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3008 tp->t_rttmin, TCPTV_REXMTMAX); 3009 } 3010 if (metrics.rmx_ssthresh) { 3011 /* 3012 * There's some sort of gateway or interface 3013 * buffer limit on the path. Use this to set 3014 * the slow start threshhold, but set the 3015 * threshold to no less than 2*mss. 3016 */ 3017 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3018 tcpstat.tcps_usedssthresh++; 3019 } 3020 if (metrics.rmx_bandwidth) 3021 tp->snd_bandwidth = metrics.rmx_bandwidth; 3022 3023 /* 3024 * Set the slow-start flight size depending on whether this 3025 * is a local network or not. 3026 * 3027 * Extend this so we cache the cwnd too and retrieve it here. 3028 * Make cwnd even bigger than RFC3390 suggests but only if we 3029 * have previous experience with the remote host. Be careful 3030 * not make cwnd bigger than remote receive window or our own 3031 * send socket buffer. Maybe put some additional upper bound 3032 * on the retrieved cwnd. Should do incremental updates to 3033 * hostcache when cwnd collapses so next connection doesn't 3034 * overloads the path again. 3035 * 3036 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3037 * We currently check only in syncache_socket for that. 3038 */ 3039#define TCP_METRICS_CWND 3040#ifdef TCP_METRICS_CWND 3041 if (metrics.rmx_cwnd) 3042 tp->snd_cwnd = max(mss, 3043 min(metrics.rmx_cwnd / 2, 3044 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3045 else 3046#endif 3047 if (tcp_do_rfc3390) 3048 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3049#ifdef INET6 3050 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3051 (!isipv6 && in_localaddr(inp->inp_faddr))) 3052#else 3053 else if (in_localaddr(inp->inp_faddr)) 3054#endif 3055 tp->snd_cwnd = mss * ss_fltsz_local; 3056 else 3057 tp->snd_cwnd = mss * ss_fltsz; 3058} 3059 3060/* 3061 * Determine the MSS option to send on an outgoing SYN. 3062 */ 3063int 3064tcp_mssopt(inc) 3065 struct in_conninfo *inc; 3066{ 3067 int mss = 0; 3068 u_long maxmtu = 0; 3069 u_long thcmtu = 0; 3070 size_t min_protoh; 3071#ifdef INET6 3072 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3073#endif 3074 3075 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3076 3077#ifdef INET6 3078 if (isipv6) { 3079 mss = tcp_v6mssdflt; 3080 maxmtu = tcp_maxmtu6(inc); 3081 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3082 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3083 } else 3084#endif 3085 { 3086 mss = tcp_mssdflt; 3087 maxmtu = tcp_maxmtu(inc); 3088 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3089 min_protoh = sizeof(struct tcpiphdr); 3090 } 3091 if (maxmtu && thcmtu) 3092 mss = min(maxmtu, thcmtu) - min_protoh; 3093 else if (maxmtu || thcmtu) 3094 mss = max(maxmtu, thcmtu) - min_protoh; 3095 3096 return (mss); 3097} 3098 3099 3100/* 3101 * On a partial ack arrives, force the retransmission of the 3102 * next unacknowledged segment. Do not clear tp->t_dupacks. 3103 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3104 * be started again. 3105 */ 3106static void 3107tcp_newreno_partial_ack(tp, th) 3108 struct tcpcb *tp; 3109 struct tcphdr *th; 3110{ 3111 tcp_seq onxt = tp->snd_nxt; 3112 u_long ocwnd = tp->snd_cwnd; 3113 3114 callout_stop(tp->tt_rexmt); 3115 tp->t_rtttime = 0; 3116 tp->snd_nxt = th->th_ack; 3117 /* 3118 * Set snd_cwnd to one segment beyond acknowledged offset. 3119 * (tp->snd_una has not yet been updated when this function is called.) 3120 */ 3121 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3122 tp->t_flags |= TF_ACKNOW; 3123 (void) tcp_output(tp); 3124 tp->snd_cwnd = ocwnd; 3125 if (SEQ_GT(onxt, tp->snd_nxt)) 3126 tp->snd_nxt = onxt; 3127 /* 3128 * Partial window deflation. Relies on fact that tp->snd_una 3129 * not updated yet. 3130 */ 3131 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3132 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3133 else 3134 tp->snd_cwnd = 0; 3135 tp->snd_cwnd += tp->t_maxseg; 3136} 3137 3138/* 3139 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3140 * looking for a pcb in the listen state. Returns 0 otherwise. 3141 */ 3142static int 3143tcp_timewait(tw, to, th, m, tlen) 3144 struct tcptw *tw; 3145 struct tcpopt *to; 3146 struct tcphdr *th; 3147 struct mbuf *m; 3148 int tlen; 3149{ 3150 int thflags; 3151 tcp_seq seq; 3152#ifdef INET6 3153 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3154#else 3155 const int isipv6 = 0; 3156#endif 3157 3158 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */ 3159 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3160 INP_LOCK_ASSERT(tw->tw_inpcb); 3161 3162 thflags = th->th_flags; 3163 3164 /* 3165 * NOTE: for FIN_WAIT_2 (to be added later), 3166 * must validate sequence number before accepting RST 3167 */ 3168 3169 /* 3170 * If the segment contains RST: 3171 * Drop the segment - see Stevens, vol. 2, p. 964 and 3172 * RFC 1337. 3173 */ 3174 if (thflags & TH_RST) 3175 goto drop; 3176 3177#if 0 3178/* PAWS not needed at the moment */ 3179 /* 3180 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3181 * and it's less than ts_recent, drop it. 3182 */ 3183 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3184 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3185 if ((thflags & TH_ACK) == 0) 3186 goto drop; 3187 goto ack; 3188 } 3189 /* 3190 * ts_recent is never updated because we never accept new segments. 3191 */ 3192#endif 3193 3194 /* 3195 * If a new connection request is received 3196 * while in TIME_WAIT, drop the old connection 3197 * and start over if the sequence numbers 3198 * are above the previous ones. 3199 */ 3200 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3201 (void) tcp_twclose(tw, 0); 3202 return (1); 3203 } 3204 3205 /* 3206 * Drop the the segment if it does not contain an ACK. 3207 */ 3208 if ((thflags & TH_ACK) == 0) 3209 goto drop; 3210 3211 /* 3212 * Reset the 2MSL timer if this is a duplicate FIN. 3213 */ 3214 if (thflags & TH_FIN) { 3215 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3216 if (seq + 1 == tw->rcv_nxt) 3217 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3218 } 3219 3220 /* 3221 * Acknowledge the segment if it has data or is not a duplicate ACK. 3222 */ 3223 if (thflags != TH_ACK || tlen != 0 || 3224 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3225 tcp_twrespond(tw, TH_ACK); 3226 goto drop; 3227 3228 /* 3229 * Generate a RST, dropping incoming segment. 3230 * Make ACK acceptable to originator of segment. 3231 * Don't bother to respond if destination was broadcast/multicast. 3232 */ 3233 if (m->m_flags & (M_BCAST|M_MCAST)) 3234 goto drop; 3235 if (isipv6) { 3236 struct ip6_hdr *ip6; 3237 3238 /* IPv6 anycast check is done at tcp6_input() */ 3239 ip6 = mtod(m, struct ip6_hdr *); 3240 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3241 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3242 goto drop; 3243 } else { 3244 struct ip *ip; 3245 3246 ip = mtod(m, struct ip *); 3247 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3248 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3249 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3250 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3251 goto drop; 3252 } 3253 if (thflags & TH_ACK) { 3254 tcp_respond(NULL, 3255 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3256 } else { 3257 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3258 tcp_respond(NULL, 3259 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3260 } 3261 INP_UNLOCK(tw->tw_inpcb); 3262 return (0); 3263 3264drop: 3265 INP_UNLOCK(tw->tw_inpcb); 3266 m_freem(m); 3267 return (0); 3268} 3269