tcp_input.c revision 157376
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 157376 2006-04-01 16:36:36Z rwatson $ 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#ifdef MAC 782 INP_LOCK_ASSERT(inp); 783 if (mac_check_inpcb_deliver(inp, m)) 784 goto drop; 785#endif 786 so = inp->inp_socket; 787 KASSERT(so != NULL, ("tcp_input: so == NULL")); 788#ifdef TCPDEBUG 789 if (so->so_options & SO_DEBUG) { 790 ostate = tp->t_state; 791 if (isipv6) 792 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 793 else 794 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 795 tcp_savetcp = *th; 796 } 797#endif 798 if (so->so_options & SO_ACCEPTCONN) { 799 struct in_conninfo inc; 800 801#ifdef INET6 802 inc.inc_isipv6 = isipv6; 803#endif 804 if (isipv6) { 805 inc.inc6_faddr = ip6->ip6_src; 806 inc.inc6_laddr = ip6->ip6_dst; 807 } else { 808 inc.inc_faddr = ip->ip_src; 809 inc.inc_laddr = ip->ip_dst; 810 } 811 inc.inc_fport = th->th_sport; 812 inc.inc_lport = th->th_dport; 813 814 /* 815 * If the state is LISTEN then ignore segment if it contains 816 * a RST. If the segment contains an ACK then it is bad and 817 * send a RST. If it does not contain a SYN then it is not 818 * interesting; drop it. 819 * 820 * If the state is SYN_RECEIVED (syncache) and seg contains 821 * an ACK, but not for our SYN/ACK, send a RST. If the seg 822 * contains a RST, check the sequence number to see if it 823 * is a valid reset segment. 824 */ 825 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 826 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 827 if (!syncache_expand(&inc, th, &so, m)) { 828 /* 829 * No syncache entry, or ACK was not 830 * for our SYN/ACK. Send a RST. 831 */ 832 tcpstat.tcps_badsyn++; 833 rstreason = BANDLIM_RST_OPENPORT; 834 goto dropwithreset; 835 } 836 if (so == NULL) { 837 /* 838 * Could not complete 3-way handshake, 839 * connection is being closed down, and 840 * syncache has free'd mbuf. 841 */ 842 INP_UNLOCK(inp); 843 INP_INFO_WUNLOCK(&tcbinfo); 844 return; 845 } 846 /* 847 * Socket is created in state SYN_RECEIVED. 848 * Continue processing segment. 849 */ 850 INP_UNLOCK(inp); 851 inp = sotoinpcb(so); 852 INP_LOCK(inp); 853 tp = intotcpcb(inp); 854 /* 855 * This is what would have happened in 856 * tcp_output() when the SYN,ACK was sent. 857 */ 858 tp->snd_up = tp->snd_una; 859 tp->snd_max = tp->snd_nxt = tp->iss + 1; 860 tp->last_ack_sent = tp->rcv_nxt; 861 goto after_listen; 862 } 863 if (thflags & TH_RST) { 864 syncache_chkrst(&inc, th); 865 goto drop; 866 } 867 if (thflags & TH_ACK) { 868 syncache_badack(&inc); 869 tcpstat.tcps_badsyn++; 870 rstreason = BANDLIM_RST_OPENPORT; 871 goto dropwithreset; 872 } 873 goto drop; 874 } 875 876 /* 877 * Segment's flags are (SYN) or (SYN|FIN). 878 */ 879#ifdef INET6 880 /* 881 * If deprecated address is forbidden, 882 * we do not accept SYN to deprecated interface 883 * address to prevent any new inbound connection from 884 * getting established. 885 * When we do not accept SYN, we send a TCP RST, 886 * with deprecated source address (instead of dropping 887 * it). We compromise it as it is much better for peer 888 * to send a RST, and RST will be the final packet 889 * for the exchange. 890 * 891 * If we do not forbid deprecated addresses, we accept 892 * the SYN packet. RFC2462 does not suggest dropping 893 * SYN in this case. 894 * If we decipher RFC2462 5.5.4, it says like this: 895 * 1. use of deprecated addr with existing 896 * communication is okay - "SHOULD continue to be 897 * used" 898 * 2. use of it with new communication: 899 * (2a) "SHOULD NOT be used if alternate address 900 * with sufficient scope is available" 901 * (2b) nothing mentioned otherwise. 902 * Here we fall into (2b) case as we have no choice in 903 * our source address selection - we must obey the peer. 904 * 905 * The wording in RFC2462 is confusing, and there are 906 * multiple description text for deprecated address 907 * handling - worse, they are not exactly the same. 908 * I believe 5.5.4 is the best one, so we follow 5.5.4. 909 */ 910 if (isipv6 && !ip6_use_deprecated) { 911 struct in6_ifaddr *ia6; 912 913 if ((ia6 = ip6_getdstifaddr(m)) && 914 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 915 INP_UNLOCK(inp); 916 tp = NULL; 917 rstreason = BANDLIM_RST_OPENPORT; 918 goto dropwithreset; 919 } 920 } 921#endif 922 /* 923 * If it is from this socket, drop it, it must be forged. 924 * Don't bother responding if the destination was a broadcast. 925 */ 926 if (th->th_dport == th->th_sport) { 927 if (isipv6) { 928 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 929 &ip6->ip6_src)) 930 goto drop; 931 } else { 932 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 933 goto drop; 934 } 935 } 936 /* 937 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 938 * 939 * Note that it is quite possible to receive unicast 940 * link-layer packets with a broadcast IP address. Use 941 * in_broadcast() to find them. 942 */ 943 if (m->m_flags & (M_BCAST|M_MCAST)) 944 goto drop; 945 if (isipv6) { 946 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 947 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 948 goto drop; 949 } else { 950 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 951 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 952 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 953 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 954 goto drop; 955 } 956 /* 957 * SYN appears to be valid; create compressed TCP state 958 * for syncache, or perform t/tcp connection. 959 */ 960 if (so->so_qlen <= so->so_qlimit) { 961#ifdef TCPDEBUG 962 if (so->so_options & SO_DEBUG) 963 tcp_trace(TA_INPUT, ostate, tp, 964 (void *)tcp_saveipgen, &tcp_savetcp, 0); 965#endif 966 tcp_dooptions(&to, optp, optlen, 1); 967 if (!syncache_add(&inc, &to, th, &so, m)) 968 goto drop; 969 if (so == NULL) { 970 /* 971 * Entry added to syncache, mbuf used to 972 * send SYN,ACK packet. 973 */ 974 KASSERT(headlocked, ("headlocked")); 975 INP_UNLOCK(inp); 976 INP_INFO_WUNLOCK(&tcbinfo); 977 return; 978 } 979 /* 980 * Segment passed TAO tests. 981 * XXX: Can't happen at the moment. 982 */ 983 INP_UNLOCK(inp); 984 inp = sotoinpcb(so); 985 INP_LOCK(inp); 986 tp = intotcpcb(inp); 987 tp->t_starttime = ticks; 988 tp->t_state = TCPS_ESTABLISHED; 989 990 /* 991 * T/TCP logic: 992 * If there is a FIN or if there is data, then 993 * delay SYN,ACK(SYN) in the hope of piggy-backing 994 * it on a response segment. Otherwise must send 995 * ACK now in case the other side is slow starting. 996 */ 997 if (thflags & TH_FIN || tlen != 0) 998 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 999 else 1000 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1001 tiwin = th->th_win << tp->snd_scale; 1002 tcpstat.tcps_connects++; 1003 soisconnected(so); 1004 goto trimthenstep6; 1005 } 1006 goto drop; 1007 } 1008after_listen: 1009 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1010 INP_LOCK_ASSERT(inp); 1011 1012 /* Syncache takes care of sockets in the listen state. */ 1013 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN")); 1014 1015 /* 1016 * This is the second part of the MSS DoS prevention code (after 1017 * minmss on the sending side) and it deals with too many too small 1018 * tcp packets in a too short timeframe (1 second). 1019 * 1020 * For every full second we count the number of received packets 1021 * and bytes. If we get a lot of packets per second for this connection 1022 * (tcp_minmssoverload) we take a closer look at it and compute the 1023 * average packet size for the past second. If that is less than 1024 * tcp_minmss we get too many packets with very small payload which 1025 * is not good and burdens our system (and every packet generates 1026 * a wakeup to the process connected to our socket). We can reasonable 1027 * expect this to be small packet DoS attack to exhaust our CPU 1028 * cycles. 1029 * 1030 * Care has to be taken for the minimum packet overload value. This 1031 * value defines the minimum number of packets per second before we 1032 * start to worry. This must not be too low to avoid killing for 1033 * example interactive connections with many small packets like 1034 * telnet or SSH. 1035 * 1036 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1037 * this check. 1038 * 1039 * Account for packet if payload packet, skip over ACK, etc. 1040 */ 1041 if (tcp_minmss && tcp_minmssoverload && 1042 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1043 if ((unsigned int)(tp->rcv_second - ticks) < hz) { 1044 tp->rcv_pps++; 1045 tp->rcv_byps += tlen + off; 1046 if (tp->rcv_pps > tcp_minmssoverload) { 1047 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1048 printf("too many small tcp packets from " 1049 "%s:%u, av. %lubyte/packet, " 1050 "dropping connection\n", 1051#ifdef INET6 1052 isipv6 ? 1053 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1054#endif 1055 inet_ntoa(inp->inp_inc.inc_faddr), 1056 inp->inp_inc.inc_fport, 1057 tp->rcv_byps / tp->rcv_pps); 1058 KASSERT(headlocked, ("tcp_input: " 1059 "after_listen: tcp_drop: head " 1060 "not locked")); 1061 tp = tcp_drop(tp, ECONNRESET); 1062 tcpstat.tcps_minmssdrops++; 1063 goto drop; 1064 } 1065 } 1066 } else { 1067 tp->rcv_second = ticks + hz; 1068 tp->rcv_pps = 1; 1069 tp->rcv_byps = tlen + off; 1070 } 1071 } 1072 1073 /* 1074 * Segment received on connection. 1075 * Reset idle time and keep-alive timer. 1076 */ 1077 tp->t_rcvtime = ticks; 1078 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1079 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1080 1081 /* 1082 * Unscale the window into a 32-bit value. 1083 * This value is bogus for the TCPS_SYN_SENT state 1084 * and is overwritten later. 1085 */ 1086 tiwin = th->th_win << tp->snd_scale; 1087 1088 /* 1089 * Process options only when we get SYN/ACK back. The SYN case 1090 * for incoming connections is handled in tcp_syncache. 1091 * XXX this is traditional behavior, may need to be cleaned up. 1092 */ 1093 tcp_dooptions(&to, optp, optlen, thflags & TH_SYN); 1094 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1095 if ((to.to_flags & TOF_SCALE) && 1096 (tp->t_flags & TF_REQ_SCALE)) { 1097 tp->t_flags |= TF_RCVD_SCALE; 1098 tp->snd_scale = to.to_requested_s_scale; 1099 tp->snd_wnd = th->th_win << tp->snd_scale; 1100 tiwin = tp->snd_wnd; 1101 } 1102 if (to.to_flags & TOF_TS) { 1103 tp->t_flags |= TF_RCVD_TSTMP; 1104 tp->ts_recent = to.to_tsval; 1105 tp->ts_recent_age = ticks; 1106 } 1107 if (to.to_flags & TOF_MSS) 1108 tcp_mss(tp, to.to_mss); 1109 if (tp->sack_enable) { 1110 if (!(to.to_flags & TOF_SACK)) 1111 tp->sack_enable = 0; 1112 else 1113 tp->t_flags |= TF_SACK_PERMIT; 1114 } 1115 1116 } 1117 1118 /* 1119 * Header prediction: check for the two common cases 1120 * of a uni-directional data xfer. If the packet has 1121 * no control flags, is in-sequence, the window didn't 1122 * change and we're not retransmitting, it's a 1123 * candidate. If the length is zero and the ack moved 1124 * forward, we're the sender side of the xfer. Just 1125 * free the data acked & wake any higher level process 1126 * that was blocked waiting for space. If the length 1127 * is non-zero and the ack didn't move, we're the 1128 * receiver side. If we're getting packets in-order 1129 * (the reassembly queue is empty), add the data to 1130 * the socket buffer and note that we need a delayed ack. 1131 * Make sure that the hidden state-flags are also off. 1132 * Since we check for TCPS_ESTABLISHED above, it can only 1133 * be TH_NEEDSYN. 1134 */ 1135 if (tp->t_state == TCPS_ESTABLISHED && 1136 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1137 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1138 ((to.to_flags & TOF_TS) == 0 || 1139 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1140 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1141 tp->snd_nxt == tp->snd_max) { 1142 1143 /* 1144 * If last ACK falls within this segment's sequence numbers, 1145 * record the timestamp. 1146 * NOTE that the test is modified according to the latest 1147 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1148 */ 1149 if ((to.to_flags & TOF_TS) != 0 && 1150 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1151 tp->ts_recent_age = ticks; 1152 tp->ts_recent = to.to_tsval; 1153 } 1154 1155 if (tlen == 0) { 1156 if (SEQ_GT(th->th_ack, tp->snd_una) && 1157 SEQ_LEQ(th->th_ack, tp->snd_max) && 1158 tp->snd_cwnd >= tp->snd_wnd && 1159 ((!tcp_do_newreno && !tp->sack_enable && 1160 tp->t_dupacks < tcprexmtthresh) || 1161 ((tcp_do_newreno || tp->sack_enable) && 1162 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 && 1163 TAILQ_EMPTY(&tp->snd_holes)))) { 1164 KASSERT(headlocked, ("headlocked")); 1165 INP_INFO_WUNLOCK(&tcbinfo); 1166 headlocked = 0; 1167 /* 1168 * this is a pure ack for outstanding data. 1169 */ 1170 ++tcpstat.tcps_predack; 1171 /* 1172 * "bad retransmit" recovery 1173 */ 1174 if (tp->t_rxtshift == 1 && 1175 ticks < tp->t_badrxtwin) { 1176 ++tcpstat.tcps_sndrexmitbad; 1177 tp->snd_cwnd = tp->snd_cwnd_prev; 1178 tp->snd_ssthresh = 1179 tp->snd_ssthresh_prev; 1180 tp->snd_recover = tp->snd_recover_prev; 1181 if (tp->t_flags & TF_WASFRECOVERY) 1182 ENTER_FASTRECOVERY(tp); 1183 tp->snd_nxt = tp->snd_max; 1184 tp->t_badrxtwin = 0; 1185 } 1186 1187 /* 1188 * Recalculate the transmit timer / rtt. 1189 * 1190 * Some boxes send broken timestamp replies 1191 * during the SYN+ACK phase, ignore 1192 * timestamps of 0 or we could calculate a 1193 * huge RTT and blow up the retransmit timer. 1194 */ 1195 if ((to.to_flags & TOF_TS) != 0 && 1196 to.to_tsecr) { 1197 if (!tp->t_rttlow || 1198 tp->t_rttlow > ticks - to.to_tsecr) 1199 tp->t_rttlow = ticks - to.to_tsecr; 1200 tcp_xmit_timer(tp, 1201 ticks - to.to_tsecr + 1); 1202 } else if (tp->t_rtttime && 1203 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1204 if (!tp->t_rttlow || 1205 tp->t_rttlow > ticks - tp->t_rtttime) 1206 tp->t_rttlow = ticks - tp->t_rtttime; 1207 tcp_xmit_timer(tp, 1208 ticks - tp->t_rtttime); 1209 } 1210 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1211 acked = th->th_ack - tp->snd_una; 1212 tcpstat.tcps_rcvackpack++; 1213 tcpstat.tcps_rcvackbyte += acked; 1214 sbdrop(&so->so_snd, acked); 1215 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1216 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1217 tp->snd_recover = th->th_ack - 1; 1218 tp->snd_una = th->th_ack; 1219 /* 1220 * pull snd_wl2 up to prevent seq wrap relative 1221 * to th_ack. 1222 */ 1223 tp->snd_wl2 = th->th_ack; 1224 tp->t_dupacks = 0; 1225 m_freem(m); 1226 ND6_HINT(tp); /* some progress has been done */ 1227 1228 /* 1229 * If all outstanding data are acked, stop 1230 * retransmit timer, otherwise restart timer 1231 * using current (possibly backed-off) value. 1232 * If process is waiting for space, 1233 * wakeup/selwakeup/signal. If data 1234 * are ready to send, let tcp_output 1235 * decide between more output or persist. 1236 1237#ifdef TCPDEBUG 1238 if (so->so_options & SO_DEBUG) 1239 tcp_trace(TA_INPUT, ostate, tp, 1240 (void *)tcp_saveipgen, 1241 &tcp_savetcp, 0); 1242#endif 1243 */ 1244 if (tp->snd_una == tp->snd_max) 1245 callout_stop(tp->tt_rexmt); 1246 else if (!callout_active(tp->tt_persist)) 1247 callout_reset(tp->tt_rexmt, 1248 tp->t_rxtcur, 1249 tcp_timer_rexmt, tp); 1250 1251 sowwakeup(so); 1252 if (so->so_snd.sb_cc) 1253 (void) tcp_output(tp); 1254 goto check_delack; 1255 } 1256 } else if (th->th_ack == tp->snd_una && 1257 LIST_EMPTY(&tp->t_segq) && 1258 tlen <= sbspace(&so->so_rcv)) { 1259 KASSERT(headlocked, ("headlocked")); 1260 INP_INFO_WUNLOCK(&tcbinfo); 1261 headlocked = 0; 1262 /* 1263 * this is a pure, in-sequence data packet 1264 * with nothing on the reassembly queue and 1265 * we have enough buffer space to take it. 1266 */ 1267 /* Clean receiver SACK report if present */ 1268 if (tp->sack_enable && tp->rcv_numsacks) 1269 tcp_clean_sackreport(tp); 1270 ++tcpstat.tcps_preddat; 1271 tp->rcv_nxt += tlen; 1272 /* 1273 * Pull snd_wl1 up to prevent seq wrap relative to 1274 * th_seq. 1275 */ 1276 tp->snd_wl1 = th->th_seq; 1277 /* 1278 * Pull rcv_up up to prevent seq wrap relative to 1279 * rcv_nxt. 1280 */ 1281 tp->rcv_up = tp->rcv_nxt; 1282 tcpstat.tcps_rcvpack++; 1283 tcpstat.tcps_rcvbyte += tlen; 1284 ND6_HINT(tp); /* some progress has been done */ 1285 /* 1286#ifdef TCPDEBUG 1287 if (so->so_options & SO_DEBUG) 1288 tcp_trace(TA_INPUT, ostate, tp, 1289 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1290#endif 1291 * Add data to socket buffer. 1292 */ 1293 SOCKBUF_LOCK(&so->so_rcv); 1294 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1295 m_freem(m); 1296 } else { 1297 m_adj(m, drop_hdrlen); /* delayed header drop */ 1298 sbappendstream_locked(&so->so_rcv, m); 1299 } 1300 sorwakeup_locked(so); 1301 if (DELAY_ACK(tp)) { 1302 tp->t_flags |= TF_DELACK; 1303 } else { 1304 tp->t_flags |= TF_ACKNOW; 1305 tcp_output(tp); 1306 } 1307 goto check_delack; 1308 } 1309 } 1310 1311 /* 1312 * Calculate amount of space in receive window, 1313 * and then do TCP input processing. 1314 * Receive window is amount of space in rcv queue, 1315 * but not less than advertised window. 1316 */ 1317 { int win; 1318 1319 win = sbspace(&so->so_rcv); 1320 if (win < 0) 1321 win = 0; 1322 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1323 } 1324 1325 switch (tp->t_state) { 1326 1327 /* 1328 * If the state is SYN_RECEIVED: 1329 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1330 */ 1331 case TCPS_SYN_RECEIVED: 1332 if ((thflags & TH_ACK) && 1333 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1334 SEQ_GT(th->th_ack, tp->snd_max))) { 1335 rstreason = BANDLIM_RST_OPENPORT; 1336 goto dropwithreset; 1337 } 1338 break; 1339 1340 /* 1341 * If the state is SYN_SENT: 1342 * if seg contains an ACK, but not for our SYN, drop the input. 1343 * if seg contains a RST, then drop the connection. 1344 * if seg does not contain SYN, then drop it. 1345 * Otherwise this is an acceptable SYN segment 1346 * initialize tp->rcv_nxt and tp->irs 1347 * if seg contains ack then advance tp->snd_una 1348 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1349 * arrange for segment to be acked (eventually) 1350 * continue processing rest of data/controls, beginning with URG 1351 */ 1352 case TCPS_SYN_SENT: 1353 if ((thflags & TH_ACK) && 1354 (SEQ_LEQ(th->th_ack, tp->iss) || 1355 SEQ_GT(th->th_ack, tp->snd_max))) { 1356 rstreason = BANDLIM_UNLIMITED; 1357 goto dropwithreset; 1358 } 1359 if (thflags & TH_RST) { 1360 if (thflags & TH_ACK) { 1361 KASSERT(headlocked, ("tcp_input: after_listen" 1362 ": tcp_drop.2: head not locked")); 1363 tp = tcp_drop(tp, ECONNREFUSED); 1364 } 1365 goto drop; 1366 } 1367 if ((thflags & TH_SYN) == 0) 1368 goto drop; 1369 1370 /* Initial send window, already scaled. */ 1371 tp->snd_wnd = th->th_win; 1372 1373 tp->irs = th->th_seq; 1374 tcp_rcvseqinit(tp); 1375 if (thflags & TH_ACK) { 1376 tcpstat.tcps_connects++; 1377 soisconnected(so); 1378#ifdef MAC 1379 SOCK_LOCK(so); 1380 mac_set_socket_peer_from_mbuf(m, so); 1381 SOCK_UNLOCK(so); 1382#endif 1383 /* Do window scaling on this connection? */ 1384 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1385 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1386 tp->rcv_scale = tp->request_r_scale; 1387 } 1388 tp->rcv_adv += tp->rcv_wnd; 1389 tp->snd_una++; /* SYN is acked */ 1390 /* 1391 * If there's data, delay ACK; if there's also a FIN 1392 * ACKNOW will be turned on later. 1393 */ 1394 if (DELAY_ACK(tp) && tlen != 0) 1395 callout_reset(tp->tt_delack, tcp_delacktime, 1396 tcp_timer_delack, tp); 1397 else 1398 tp->t_flags |= TF_ACKNOW; 1399 /* 1400 * Received <SYN,ACK> in SYN_SENT[*] state. 1401 * Transitions: 1402 * SYN_SENT --> ESTABLISHED 1403 * SYN_SENT* --> FIN_WAIT_1 1404 */ 1405 tp->t_starttime = ticks; 1406 if (tp->t_flags & TF_NEEDFIN) { 1407 tp->t_state = TCPS_FIN_WAIT_1; 1408 tp->t_flags &= ~TF_NEEDFIN; 1409 thflags &= ~TH_SYN; 1410 } else { 1411 tp->t_state = TCPS_ESTABLISHED; 1412 callout_reset(tp->tt_keep, tcp_keepidle, 1413 tcp_timer_keep, tp); 1414 } 1415 } else { 1416 /* 1417 * Received initial SYN in SYN-SENT[*] state => 1418 * simultaneous open. If segment contains CC option 1419 * and there is a cached CC, apply TAO test. 1420 * If it succeeds, connection is * half-synchronized. 1421 * Otherwise, do 3-way handshake: 1422 * SYN-SENT -> SYN-RECEIVED 1423 * SYN-SENT* -> SYN-RECEIVED* 1424 * If there was no CC option, clear cached CC value. 1425 */ 1426 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1427 callout_stop(tp->tt_rexmt); 1428 tp->t_state = TCPS_SYN_RECEIVED; 1429 } 1430 1431trimthenstep6: 1432 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1433 "locked")); 1434 INP_LOCK_ASSERT(inp); 1435 1436 /* 1437 * Advance th->th_seq to correspond to first data byte. 1438 * If data, trim to stay within window, 1439 * dropping FIN if necessary. 1440 */ 1441 th->th_seq++; 1442 if (tlen > tp->rcv_wnd) { 1443 todrop = tlen - tp->rcv_wnd; 1444 m_adj(m, -todrop); 1445 tlen = tp->rcv_wnd; 1446 thflags &= ~TH_FIN; 1447 tcpstat.tcps_rcvpackafterwin++; 1448 tcpstat.tcps_rcvbyteafterwin += todrop; 1449 } 1450 tp->snd_wl1 = th->th_seq - 1; 1451 tp->rcv_up = th->th_seq; 1452 /* 1453 * Client side of transaction: already sent SYN and data. 1454 * If the remote host used T/TCP to validate the SYN, 1455 * our data will be ACK'd; if so, enter normal data segment 1456 * processing in the middle of step 5, ack processing. 1457 * Otherwise, goto step 6. 1458 */ 1459 if (thflags & TH_ACK) 1460 goto process_ACK; 1461 1462 goto step6; 1463 1464 /* 1465 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1466 * do normal processing. 1467 * 1468 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1469 */ 1470 case TCPS_LAST_ACK: 1471 case TCPS_CLOSING: 1472 case TCPS_TIME_WAIT: 1473 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1474 break; /* continue normal processing */ 1475 } 1476 1477 /* 1478 * States other than LISTEN or SYN_SENT. 1479 * First check the RST flag and sequence number since reset segments 1480 * are exempt from the timestamp and connection count tests. This 1481 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1482 * below which allowed reset segments in half the sequence space 1483 * to fall though and be processed (which gives forged reset 1484 * segments with a random sequence number a 50 percent chance of 1485 * killing a connection). 1486 * Then check timestamp, if present. 1487 * Then check the connection count, if present. 1488 * Then check that at least some bytes of segment are within 1489 * receive window. If segment begins before rcv_nxt, 1490 * drop leading data (and SYN); if nothing left, just ack. 1491 * 1492 * 1493 * If the RST bit is set, check the sequence number to see 1494 * if this is a valid reset segment. 1495 * RFC 793 page 37: 1496 * In all states except SYN-SENT, all reset (RST) segments 1497 * are validated by checking their SEQ-fields. A reset is 1498 * valid if its sequence number is in the window. 1499 * Note: this does not take into account delayed ACKs, so 1500 * we should test against last_ack_sent instead of rcv_nxt. 1501 * The sequence number in the reset segment is normally an 1502 * echo of our outgoing acknowlegement numbers, but some hosts 1503 * send a reset with the sequence number at the rightmost edge 1504 * of our receive window, and we have to handle this case. 1505 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1506 * that brute force RST attacks are possible. To combat this, 1507 * we use a much stricter check while in the ESTABLISHED state, 1508 * only accepting RSTs where the sequence number is equal to 1509 * last_ack_sent. In all other states (the states in which a 1510 * RST is more likely), the more permissive check is used. 1511 * If we have multiple segments in flight, the intial reset 1512 * segment sequence numbers will be to the left of last_ack_sent, 1513 * but they will eventually catch up. 1514 * In any case, it never made sense to trim reset segments to 1515 * fit the receive window since RFC 1122 says: 1516 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1517 * 1518 * A TCP SHOULD allow a received RST segment to include data. 1519 * 1520 * DISCUSSION 1521 * It has been suggested that a RST segment could contain 1522 * ASCII text that encoded and explained the cause of the 1523 * RST. No standard has yet been established for such 1524 * data. 1525 * 1526 * If the reset segment passes the sequence number test examine 1527 * the state: 1528 * SYN_RECEIVED STATE: 1529 * If passive open, return to LISTEN state. 1530 * If active open, inform user that connection was refused. 1531 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1532 * Inform user that connection was reset, and close tcb. 1533 * CLOSING, LAST_ACK STATES: 1534 * Close the tcb. 1535 * TIME_WAIT STATE: 1536 * Drop the segment - see Stevens, vol. 2, p. 964 and 1537 * RFC 1337. 1538 */ 1539 if (thflags & TH_RST) { 1540 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1541 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1542 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1543 switch (tp->t_state) { 1544 1545 case TCPS_SYN_RECEIVED: 1546 so->so_error = ECONNREFUSED; 1547 goto close; 1548 1549 case TCPS_ESTABLISHED: 1550 if (tp->last_ack_sent != th->th_seq && 1551 tcp_insecure_rst == 0) { 1552 tcpstat.tcps_badrst++; 1553 goto drop; 1554 } 1555 case TCPS_FIN_WAIT_1: 1556 case TCPS_FIN_WAIT_2: 1557 case TCPS_CLOSE_WAIT: 1558 so->so_error = ECONNRESET; 1559 close: 1560 tp->t_state = TCPS_CLOSED; 1561 tcpstat.tcps_drops++; 1562 KASSERT(headlocked, ("tcp_input: " 1563 "trimthenstep6: tcp_close: head not " 1564 "locked")); 1565 tp = tcp_close(tp); 1566 break; 1567 1568 case TCPS_CLOSING: 1569 case TCPS_LAST_ACK: 1570 KASSERT(headlocked, ("trimthenstep6: " 1571 "tcp_close.2: head not locked")); 1572 tp = tcp_close(tp); 1573 break; 1574 1575 case TCPS_TIME_WAIT: 1576 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1577 ("timewait")); 1578 break; 1579 } 1580 } 1581 goto drop; 1582 } 1583 1584 /* 1585 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1586 * and it's less than ts_recent, drop it. 1587 */ 1588 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1589 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1590 1591 /* Check to see if ts_recent is over 24 days old. */ 1592 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1593 /* 1594 * Invalidate ts_recent. If this segment updates 1595 * ts_recent, the age will be reset later and ts_recent 1596 * will get a valid value. If it does not, setting 1597 * ts_recent to zero will at least satisfy the 1598 * requirement that zero be placed in the timestamp 1599 * echo reply when ts_recent isn't valid. The 1600 * age isn't reset until we get a valid ts_recent 1601 * because we don't want out-of-order segments to be 1602 * dropped when ts_recent is old. 1603 */ 1604 tp->ts_recent = 0; 1605 } else { 1606 tcpstat.tcps_rcvduppack++; 1607 tcpstat.tcps_rcvdupbyte += tlen; 1608 tcpstat.tcps_pawsdrop++; 1609 if (tlen) 1610 goto dropafterack; 1611 goto drop; 1612 } 1613 } 1614 1615 /* 1616 * In the SYN-RECEIVED state, validate that the packet belongs to 1617 * this connection before trimming the data to fit the receive 1618 * window. Check the sequence number versus IRS since we know 1619 * the sequence numbers haven't wrapped. This is a partial fix 1620 * for the "LAND" DoS attack. 1621 */ 1622 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1623 rstreason = BANDLIM_RST_OPENPORT; 1624 goto dropwithreset; 1625 } 1626 1627 todrop = tp->rcv_nxt - th->th_seq; 1628 if (todrop > 0) { 1629 if (thflags & TH_SYN) { 1630 thflags &= ~TH_SYN; 1631 th->th_seq++; 1632 if (th->th_urp > 1) 1633 th->th_urp--; 1634 else 1635 thflags &= ~TH_URG; 1636 todrop--; 1637 } 1638 /* 1639 * Following if statement from Stevens, vol. 2, p. 960. 1640 */ 1641 if (todrop > tlen 1642 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1643 /* 1644 * Any valid FIN must be to the left of the window. 1645 * At this point the FIN must be a duplicate or out 1646 * of sequence; drop it. 1647 */ 1648 thflags &= ~TH_FIN; 1649 1650 /* 1651 * Send an ACK to resynchronize and drop any data. 1652 * But keep on processing for RST or ACK. 1653 */ 1654 tp->t_flags |= TF_ACKNOW; 1655 todrop = tlen; 1656 tcpstat.tcps_rcvduppack++; 1657 tcpstat.tcps_rcvdupbyte += todrop; 1658 } else { 1659 tcpstat.tcps_rcvpartduppack++; 1660 tcpstat.tcps_rcvpartdupbyte += todrop; 1661 } 1662 drop_hdrlen += todrop; /* drop from the top afterwards */ 1663 th->th_seq += todrop; 1664 tlen -= todrop; 1665 if (th->th_urp > todrop) 1666 th->th_urp -= todrop; 1667 else { 1668 thflags &= ~TH_URG; 1669 th->th_urp = 0; 1670 } 1671 } 1672 1673 /* 1674 * If new data are received on a connection after the 1675 * user processes are gone, then RST the other end. 1676 */ 1677 if ((so->so_state & SS_NOFDREF) && 1678 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1679 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not " 1680 "locked")); 1681 tp = tcp_close(tp); 1682 tcpstat.tcps_rcvafterclose++; 1683 rstreason = BANDLIM_UNLIMITED; 1684 goto dropwithreset; 1685 } 1686 1687 /* 1688 * If segment ends after window, drop trailing data 1689 * (and PUSH and FIN); if nothing left, just ACK. 1690 */ 1691 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1692 if (todrop > 0) { 1693 tcpstat.tcps_rcvpackafterwin++; 1694 if (todrop >= tlen) { 1695 tcpstat.tcps_rcvbyteafterwin += tlen; 1696 /* 1697 * If a new connection request is received 1698 * while in TIME_WAIT, drop the old connection 1699 * and start over if the sequence numbers 1700 * are above the previous ones. 1701 */ 1702 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1703 if (thflags & TH_SYN && 1704 tp->t_state == TCPS_TIME_WAIT && 1705 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1706 KASSERT(headlocked, ("trimthenstep6: " 1707 "tcp_close.4: head not locked")); 1708 tp = tcp_close(tp); 1709 goto findpcb; 1710 } 1711 /* 1712 * If window is closed can only take segments at 1713 * window edge, and have to drop data and PUSH from 1714 * incoming segments. Continue processing, but 1715 * remember to ack. Otherwise, drop segment 1716 * and ack. 1717 */ 1718 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1719 tp->t_flags |= TF_ACKNOW; 1720 tcpstat.tcps_rcvwinprobe++; 1721 } else 1722 goto dropafterack; 1723 } else 1724 tcpstat.tcps_rcvbyteafterwin += todrop; 1725 m_adj(m, -todrop); 1726 tlen -= todrop; 1727 thflags &= ~(TH_PUSH|TH_FIN); 1728 } 1729 1730 /* 1731 * If last ACK falls within this segment's sequence numbers, 1732 * record its timestamp. 1733 * NOTE: 1734 * 1) That the test incorporates suggestions from the latest 1735 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1736 * 2) That updating only on newer timestamps interferes with 1737 * our earlier PAWS tests, so this check should be solely 1738 * predicated on the sequence space of this segment. 1739 * 3) That we modify the segment boundary check to be 1740 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1741 * instead of RFC1323's 1742 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1743 * This modified check allows us to overcome RFC1323's 1744 * limitations as described in Stevens TCP/IP Illustrated 1745 * Vol. 2 p.869. In such cases, we can still calculate the 1746 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1747 */ 1748 if ((to.to_flags & TOF_TS) != 0 && 1749 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1750 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1751 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1752 tp->ts_recent_age = ticks; 1753 tp->ts_recent = to.to_tsval; 1754 } 1755 1756 /* 1757 * If a SYN is in the window, then this is an 1758 * error and we send an RST and drop the connection. 1759 */ 1760 if (thflags & TH_SYN) { 1761 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: " 1762 "head not locked")); 1763 tp = tcp_drop(tp, ECONNRESET); 1764 rstreason = BANDLIM_UNLIMITED; 1765 goto drop; 1766 } 1767 1768 /* 1769 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1770 * flag is on (half-synchronized state), then queue data for 1771 * later processing; else drop segment and return. 1772 */ 1773 if ((thflags & TH_ACK) == 0) { 1774 if (tp->t_state == TCPS_SYN_RECEIVED || 1775 (tp->t_flags & TF_NEEDSYN)) 1776 goto step6; 1777 else 1778 goto drop; 1779 } 1780 1781 /* 1782 * Ack processing. 1783 */ 1784 switch (tp->t_state) { 1785 1786 /* 1787 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1788 * ESTABLISHED state and continue processing. 1789 * The ACK was checked above. 1790 */ 1791 case TCPS_SYN_RECEIVED: 1792 1793 tcpstat.tcps_connects++; 1794 soisconnected(so); 1795 /* Do window scaling? */ 1796 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1797 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1798 tp->rcv_scale = tp->request_r_scale; 1799 tp->snd_wnd = tiwin; 1800 } 1801 /* 1802 * Make transitions: 1803 * SYN-RECEIVED -> ESTABLISHED 1804 * SYN-RECEIVED* -> FIN-WAIT-1 1805 */ 1806 tp->t_starttime = ticks; 1807 if (tp->t_flags & TF_NEEDFIN) { 1808 tp->t_state = TCPS_FIN_WAIT_1; 1809 tp->t_flags &= ~TF_NEEDFIN; 1810 } else { 1811 tp->t_state = TCPS_ESTABLISHED; 1812 callout_reset(tp->tt_keep, tcp_keepidle, 1813 tcp_timer_keep, tp); 1814 } 1815 /* 1816 * If segment contains data or ACK, will call tcp_reass() 1817 * later; if not, do so now to pass queued data to user. 1818 */ 1819 if (tlen == 0 && (thflags & TH_FIN) == 0) 1820 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1821 (struct mbuf *)0); 1822 tp->snd_wl1 = th->th_seq - 1; 1823 /* FALLTHROUGH */ 1824 1825 /* 1826 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1827 * ACKs. If the ack is in the range 1828 * tp->snd_una < th->th_ack <= tp->snd_max 1829 * then advance tp->snd_una to th->th_ack and drop 1830 * data from the retransmission queue. If this ACK reflects 1831 * more up to date window information we update our window information. 1832 */ 1833 case TCPS_ESTABLISHED: 1834 case TCPS_FIN_WAIT_1: 1835 case TCPS_FIN_WAIT_2: 1836 case TCPS_CLOSE_WAIT: 1837 case TCPS_CLOSING: 1838 case TCPS_LAST_ACK: 1839 case TCPS_TIME_WAIT: 1840 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1841 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1842 tcpstat.tcps_rcvacktoomuch++; 1843 goto dropafterack; 1844 } 1845 if (tp->sack_enable && 1846 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) 1847 tcp_sack_doack(tp, &to, th->th_ack); 1848 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1849 if (tlen == 0 && tiwin == tp->snd_wnd) { 1850 tcpstat.tcps_rcvdupack++; 1851 /* 1852 * If we have outstanding data (other than 1853 * a window probe), this is a completely 1854 * duplicate ack (ie, window info didn't 1855 * change), the ack is the biggest we've 1856 * seen and we've seen exactly our rexmt 1857 * threshhold of them, assume a packet 1858 * has been dropped and retransmit it. 1859 * Kludge snd_nxt & the congestion 1860 * window so we send only this one 1861 * packet. 1862 * 1863 * We know we're losing at the current 1864 * window size so do congestion avoidance 1865 * (set ssthresh to half the current window 1866 * and pull our congestion window back to 1867 * the new ssthresh). 1868 * 1869 * Dup acks mean that packets have left the 1870 * network (they're now cached at the receiver) 1871 * so bump cwnd by the amount in the receiver 1872 * to keep a constant cwnd packets in the 1873 * network. 1874 */ 1875 if (!callout_active(tp->tt_rexmt) || 1876 th->th_ack != tp->snd_una) 1877 tp->t_dupacks = 0; 1878 else if (++tp->t_dupacks > tcprexmtthresh || 1879 ((tcp_do_newreno || tp->sack_enable) && 1880 IN_FASTRECOVERY(tp))) { 1881 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1882 int awnd; 1883 1884 /* 1885 * Compute the amount of data in flight first. 1886 * We can inject new data into the pipe iff 1887 * we have less than 1/2 the original window's 1888 * worth of data in flight. 1889 */ 1890 awnd = (tp->snd_nxt - tp->snd_fack) + 1891 tp->sackhint.sack_bytes_rexmit; 1892 if (awnd < tp->snd_ssthresh) { 1893 tp->snd_cwnd += tp->t_maxseg; 1894 if (tp->snd_cwnd > tp->snd_ssthresh) 1895 tp->snd_cwnd = tp->snd_ssthresh; 1896 } 1897 } else 1898 tp->snd_cwnd += tp->t_maxseg; 1899 (void) tcp_output(tp); 1900 goto drop; 1901 } else if (tp->t_dupacks == tcprexmtthresh) { 1902 tcp_seq onxt = tp->snd_nxt; 1903 u_int win; 1904 1905 /* 1906 * If we're doing sack, check to 1907 * see if we're already in sack 1908 * recovery. If we're not doing sack, 1909 * check to see if we're in newreno 1910 * recovery. 1911 */ 1912 if (tp->sack_enable) { 1913 if (IN_FASTRECOVERY(tp)) { 1914 tp->t_dupacks = 0; 1915 break; 1916 } 1917 } else if (tcp_do_newreno) { 1918 if (SEQ_LEQ(th->th_ack, 1919 tp->snd_recover)) { 1920 tp->t_dupacks = 0; 1921 break; 1922 } 1923 } 1924 win = min(tp->snd_wnd, tp->snd_cwnd) / 1925 2 / tp->t_maxseg; 1926 if (win < 2) 1927 win = 2; 1928 tp->snd_ssthresh = win * tp->t_maxseg; 1929 ENTER_FASTRECOVERY(tp); 1930 tp->snd_recover = tp->snd_max; 1931 callout_stop(tp->tt_rexmt); 1932 tp->t_rtttime = 0; 1933 if (tp->sack_enable) { 1934 tcpstat.tcps_sack_recovery_episode++; 1935 tp->sack_newdata = tp->snd_nxt; 1936 tp->snd_cwnd = tp->t_maxseg; 1937 (void) tcp_output(tp); 1938 goto drop; 1939 } 1940 tp->snd_nxt = th->th_ack; 1941 tp->snd_cwnd = tp->t_maxseg; 1942 (void) tcp_output(tp); 1943 KASSERT(tp->snd_limited <= 2, 1944 ("tp->snd_limited too big")); 1945 tp->snd_cwnd = tp->snd_ssthresh + 1946 tp->t_maxseg * 1947 (tp->t_dupacks - tp->snd_limited); 1948 if (SEQ_GT(onxt, tp->snd_nxt)) 1949 tp->snd_nxt = onxt; 1950 goto drop; 1951 } else if (tcp_do_rfc3042) { 1952 u_long oldcwnd = tp->snd_cwnd; 1953 tcp_seq oldsndmax = tp->snd_max; 1954 u_int sent; 1955 1956 KASSERT(tp->t_dupacks == 1 || 1957 tp->t_dupacks == 2, 1958 ("dupacks not 1 or 2")); 1959 if (tp->t_dupacks == 1) 1960 tp->snd_limited = 0; 1961 tp->snd_cwnd = 1962 (tp->snd_nxt - tp->snd_una) + 1963 (tp->t_dupacks - tp->snd_limited) * 1964 tp->t_maxseg; 1965 (void) tcp_output(tp); 1966 sent = tp->snd_max - oldsndmax; 1967 if (sent > tp->t_maxseg) { 1968 KASSERT((tp->t_dupacks == 2 && 1969 tp->snd_limited == 0) || 1970 (sent == tp->t_maxseg + 1 && 1971 tp->t_flags & TF_SENTFIN), 1972 ("sent too much")); 1973 tp->snd_limited = 2; 1974 } else if (sent > 0) 1975 ++tp->snd_limited; 1976 tp->snd_cwnd = oldcwnd; 1977 goto drop; 1978 } 1979 } else 1980 tp->t_dupacks = 0; 1981 break; 1982 } 1983 1984 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1985 1986 /* 1987 * If the congestion window was inflated to account 1988 * for the other side's cached packets, retract it. 1989 */ 1990 if (tcp_do_newreno || tp->sack_enable) { 1991 if (IN_FASTRECOVERY(tp)) { 1992 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1993 if (tp->sack_enable) 1994 tcp_sack_partialack(tp, th); 1995 else 1996 tcp_newreno_partial_ack(tp, th); 1997 } else { 1998 /* 1999 * Out of fast recovery. 2000 * Window inflation should have left us 2001 * with approximately snd_ssthresh 2002 * outstanding data. 2003 * But in case we would be inclined to 2004 * send a burst, better to do it via 2005 * the slow start mechanism. 2006 */ 2007 if (SEQ_GT(th->th_ack + 2008 tp->snd_ssthresh, 2009 tp->snd_max)) 2010 tp->snd_cwnd = tp->snd_max - 2011 th->th_ack + 2012 tp->t_maxseg; 2013 else 2014 tp->snd_cwnd = tp->snd_ssthresh; 2015 } 2016 } 2017 } else { 2018 if (tp->t_dupacks >= tcprexmtthresh && 2019 tp->snd_cwnd > tp->snd_ssthresh) 2020 tp->snd_cwnd = tp->snd_ssthresh; 2021 } 2022 tp->t_dupacks = 0; 2023 /* 2024 * If we reach this point, ACK is not a duplicate, 2025 * i.e., it ACKs something we sent. 2026 */ 2027 if (tp->t_flags & TF_NEEDSYN) { 2028 /* 2029 * T/TCP: Connection was half-synchronized, and our 2030 * SYN has been ACK'd (so connection is now fully 2031 * synchronized). Go to non-starred state, 2032 * increment snd_una for ACK of SYN, and check if 2033 * we can do window scaling. 2034 */ 2035 tp->t_flags &= ~TF_NEEDSYN; 2036 tp->snd_una++; 2037 /* Do window scaling? */ 2038 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2039 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2040 tp->rcv_scale = tp->request_r_scale; 2041 /* Send window already scaled. */ 2042 } 2043 } 2044 2045process_ACK: 2046 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 2047 "locked")); 2048 INP_LOCK_ASSERT(inp); 2049 2050 acked = th->th_ack - tp->snd_una; 2051 tcpstat.tcps_rcvackpack++; 2052 tcpstat.tcps_rcvackbyte += acked; 2053 2054 /* 2055 * If we just performed our first retransmit, and the ACK 2056 * arrives within our recovery window, then it was a mistake 2057 * to do the retransmit in the first place. Recover our 2058 * original cwnd and ssthresh, and proceed to transmit where 2059 * we left off. 2060 */ 2061 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2062 ++tcpstat.tcps_sndrexmitbad; 2063 tp->snd_cwnd = tp->snd_cwnd_prev; 2064 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2065 tp->snd_recover = tp->snd_recover_prev; 2066 if (tp->t_flags & TF_WASFRECOVERY) 2067 ENTER_FASTRECOVERY(tp); 2068 tp->snd_nxt = tp->snd_max; 2069 tp->t_badrxtwin = 0; /* XXX probably not required */ 2070 } 2071 2072 /* 2073 * If we have a timestamp reply, update smoothed 2074 * round trip time. If no timestamp is present but 2075 * transmit timer is running and timed sequence 2076 * number was acked, update smoothed round trip time. 2077 * Since we now have an rtt measurement, cancel the 2078 * timer backoff (cf., Phil Karn's retransmit alg.). 2079 * Recompute the initial retransmit timer. 2080 * 2081 * Some boxes send broken timestamp replies 2082 * during the SYN+ACK phase, ignore 2083 * timestamps of 0 or we could calculate a 2084 * huge RTT and blow up the retransmit timer. 2085 */ 2086 if ((to.to_flags & TOF_TS) != 0 && 2087 to.to_tsecr) { 2088 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2089 tp->t_rttlow = ticks - to.to_tsecr; 2090 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2091 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2092 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2093 tp->t_rttlow = ticks - tp->t_rtttime; 2094 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2095 } 2096 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2097 2098 /* 2099 * If all outstanding data is acked, stop retransmit 2100 * timer and remember to restart (more output or persist). 2101 * If there is more data to be acked, restart retransmit 2102 * timer, using current (possibly backed-off) value. 2103 */ 2104 if (th->th_ack == tp->snd_max) { 2105 callout_stop(tp->tt_rexmt); 2106 needoutput = 1; 2107 } else if (!callout_active(tp->tt_persist)) 2108 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2109 tcp_timer_rexmt, tp); 2110 2111 /* 2112 * If no data (only SYN) was ACK'd, 2113 * skip rest of ACK processing. 2114 */ 2115 if (acked == 0) 2116 goto step6; 2117 2118 /* 2119 * When new data is acked, open the congestion window. 2120 * If the window gives us less than ssthresh packets 2121 * in flight, open exponentially (maxseg per packet). 2122 * Otherwise open linearly: maxseg per window 2123 * (maxseg^2 / cwnd per packet). 2124 */ 2125 if ((!tcp_do_newreno && !tp->sack_enable) || 2126 !IN_FASTRECOVERY(tp)) { 2127 register u_int cw = tp->snd_cwnd; 2128 register u_int incr = tp->t_maxseg; 2129 if (cw > tp->snd_ssthresh) 2130 incr = incr * incr / cw; 2131 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2132 } 2133 SOCKBUF_LOCK(&so->so_snd); 2134 if (acked > so->so_snd.sb_cc) { 2135 tp->snd_wnd -= so->so_snd.sb_cc; 2136 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2137 ourfinisacked = 1; 2138 } else { 2139 sbdrop_locked(&so->so_snd, acked); 2140 tp->snd_wnd -= acked; 2141 ourfinisacked = 0; 2142 } 2143 sowwakeup_locked(so); 2144 /* detect una wraparound */ 2145 if ((tcp_do_newreno || tp->sack_enable) && 2146 !IN_FASTRECOVERY(tp) && 2147 SEQ_GT(tp->snd_una, tp->snd_recover) && 2148 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2149 tp->snd_recover = th->th_ack - 1; 2150 if ((tcp_do_newreno || tp->sack_enable) && 2151 IN_FASTRECOVERY(tp) && 2152 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2153 EXIT_FASTRECOVERY(tp); 2154 tp->snd_una = th->th_ack; 2155 if (tp->sack_enable) { 2156 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2157 tp->snd_recover = tp->snd_una; 2158 } 2159 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2160 tp->snd_nxt = tp->snd_una; 2161 2162 switch (tp->t_state) { 2163 2164 /* 2165 * In FIN_WAIT_1 STATE in addition to the processing 2166 * for the ESTABLISHED state if our FIN is now acknowledged 2167 * then enter FIN_WAIT_2. 2168 */ 2169 case TCPS_FIN_WAIT_1: 2170 if (ourfinisacked) { 2171 /* 2172 * If we can't receive any more 2173 * data, then closing user can proceed. 2174 * Starting the timer is contrary to the 2175 * specification, but if we don't get a FIN 2176 * we'll hang forever. 2177 */ 2178 /* XXXjl 2179 * we should release the tp also, and use a 2180 * compressed state. 2181 */ 2182 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2183 soisdisconnected(so); 2184 callout_reset(tp->tt_2msl, tcp_maxidle, 2185 tcp_timer_2msl, tp); 2186 } 2187 tp->t_state = TCPS_FIN_WAIT_2; 2188 } 2189 break; 2190 2191 /* 2192 * In CLOSING STATE in addition to the processing for 2193 * the ESTABLISHED state if the ACK acknowledges our FIN 2194 * then enter the TIME-WAIT state, otherwise ignore 2195 * the segment. 2196 */ 2197 case TCPS_CLOSING: 2198 if (ourfinisacked) { 2199 KASSERT(headlocked, ("tcp_input: process_ACK: " 2200 "head not locked")); 2201 tcp_twstart(tp); 2202 INP_INFO_WUNLOCK(&tcbinfo); 2203 m_freem(m); 2204 return; 2205 } 2206 break; 2207 2208 /* 2209 * In LAST_ACK, we may still be waiting for data to drain 2210 * and/or to be acked, as well as for the ack of our FIN. 2211 * If our FIN is now acknowledged, delete the TCB, 2212 * enter the closed state and return. 2213 */ 2214 case TCPS_LAST_ACK: 2215 if (ourfinisacked) { 2216 KASSERT(headlocked, ("tcp_input: process_ACK:" 2217 " tcp_close: head not locked")); 2218 tp = tcp_close(tp); 2219 goto drop; 2220 } 2221 break; 2222 2223 /* 2224 * In TIME_WAIT state the only thing that should arrive 2225 * is a retransmission of the remote FIN. Acknowledge 2226 * it and restart the finack timer. 2227 */ 2228 case TCPS_TIME_WAIT: 2229 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2230 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2231 tcp_timer_2msl, tp); 2232 goto dropafterack; 2233 } 2234 } 2235 2236step6: 2237 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2238 INP_LOCK_ASSERT(inp); 2239 2240 /* 2241 * Update window information. 2242 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2243 */ 2244 if ((thflags & TH_ACK) && 2245 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2246 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2247 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2248 /* keep track of pure window updates */ 2249 if (tlen == 0 && 2250 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2251 tcpstat.tcps_rcvwinupd++; 2252 tp->snd_wnd = tiwin; 2253 tp->snd_wl1 = th->th_seq; 2254 tp->snd_wl2 = th->th_ack; 2255 if (tp->snd_wnd > tp->max_sndwnd) 2256 tp->max_sndwnd = tp->snd_wnd; 2257 needoutput = 1; 2258 } 2259 2260 /* 2261 * Process segments with URG. 2262 */ 2263 if ((thflags & TH_URG) && th->th_urp && 2264 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2265 /* 2266 * This is a kludge, but if we receive and accept 2267 * random urgent pointers, we'll crash in 2268 * soreceive. It's hard to imagine someone 2269 * actually wanting to send this much urgent data. 2270 */ 2271 SOCKBUF_LOCK(&so->so_rcv); 2272 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2273 th->th_urp = 0; /* XXX */ 2274 thflags &= ~TH_URG; /* XXX */ 2275 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2276 goto dodata; /* XXX */ 2277 } 2278 /* 2279 * If this segment advances the known urgent pointer, 2280 * then mark the data stream. This should not happen 2281 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2282 * a FIN has been received from the remote side. 2283 * In these states we ignore the URG. 2284 * 2285 * According to RFC961 (Assigned Protocols), 2286 * the urgent pointer points to the last octet 2287 * of urgent data. We continue, however, 2288 * to consider it to indicate the first octet 2289 * of data past the urgent section as the original 2290 * spec states (in one of two places). 2291 */ 2292 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2293 tp->rcv_up = th->th_seq + th->th_urp; 2294 so->so_oobmark = so->so_rcv.sb_cc + 2295 (tp->rcv_up - tp->rcv_nxt) - 1; 2296 if (so->so_oobmark == 0) 2297 so->so_rcv.sb_state |= SBS_RCVATMARK; 2298 sohasoutofband(so); 2299 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2300 } 2301 SOCKBUF_UNLOCK(&so->so_rcv); 2302 /* 2303 * Remove out of band data so doesn't get presented to user. 2304 * This can happen independent of advancing the URG pointer, 2305 * but if two URG's are pending at once, some out-of-band 2306 * data may creep in... ick. 2307 */ 2308 if (th->th_urp <= (u_long)tlen && 2309 !(so->so_options & SO_OOBINLINE)) { 2310 /* hdr drop is delayed */ 2311 tcp_pulloutofband(so, th, m, drop_hdrlen); 2312 } 2313 } else { 2314 /* 2315 * If no out of band data is expected, 2316 * pull receive urgent pointer along 2317 * with the receive window. 2318 */ 2319 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2320 tp->rcv_up = tp->rcv_nxt; 2321 } 2322dodata: /* XXX */ 2323 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2324 INP_LOCK_ASSERT(inp); 2325 2326 /* 2327 * Process the segment text, merging it into the TCP sequencing queue, 2328 * and arranging for acknowledgment of receipt if necessary. 2329 * This process logically involves adjusting tp->rcv_wnd as data 2330 * is presented to the user (this happens in tcp_usrreq.c, 2331 * case PRU_RCVD). If a FIN has already been received on this 2332 * connection then we just ignore the text. 2333 */ 2334 if ((tlen || (thflags & TH_FIN)) && 2335 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2336 tcp_seq save_start = th->th_seq; 2337 tcp_seq save_end = th->th_seq + tlen; 2338 m_adj(m, drop_hdrlen); /* delayed header drop */ 2339 /* 2340 * Insert segment which includes th into TCP reassembly queue 2341 * with control block tp. Set thflags to whether reassembly now 2342 * includes a segment with FIN. This handles the common case 2343 * inline (segment is the next to be received on an established 2344 * connection, and the queue is empty), avoiding linkage into 2345 * and removal from the queue and repetition of various 2346 * conversions. 2347 * Set DELACK for segments received in order, but ack 2348 * immediately when segments are out of order (so 2349 * fast retransmit can work). 2350 */ 2351 if (th->th_seq == tp->rcv_nxt && 2352 LIST_EMPTY(&tp->t_segq) && 2353 TCPS_HAVEESTABLISHED(tp->t_state)) { 2354 if (DELAY_ACK(tp)) 2355 tp->t_flags |= TF_DELACK; 2356 else 2357 tp->t_flags |= TF_ACKNOW; 2358 tp->rcv_nxt += tlen; 2359 thflags = th->th_flags & TH_FIN; 2360 tcpstat.tcps_rcvpack++; 2361 tcpstat.tcps_rcvbyte += tlen; 2362 ND6_HINT(tp); 2363 SOCKBUF_LOCK(&so->so_rcv); 2364 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2365 m_freem(m); 2366 else 2367 sbappendstream_locked(&so->so_rcv, m); 2368 sorwakeup_locked(so); 2369 } else { 2370 thflags = tcp_reass(tp, th, &tlen, m); 2371 tp->t_flags |= TF_ACKNOW; 2372 } 2373 if (tlen > 0 && tp->sack_enable) 2374 tcp_update_sack_list(tp, save_start, save_end); 2375 /* 2376 * Note the amount of data that peer has sent into 2377 * our window, in order to estimate the sender's 2378 * buffer size. 2379 */ 2380 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2381 } else { 2382 m_freem(m); 2383 thflags &= ~TH_FIN; 2384 } 2385 2386 /* 2387 * If FIN is received ACK the FIN and let the user know 2388 * that the connection is closing. 2389 */ 2390 if (thflags & TH_FIN) { 2391 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2392 socantrcvmore(so); 2393 /* 2394 * If connection is half-synchronized 2395 * (ie NEEDSYN flag on) then delay ACK, 2396 * so it may be piggybacked when SYN is sent. 2397 * Otherwise, since we received a FIN then no 2398 * more input can be expected, send ACK now. 2399 */ 2400 if (tp->t_flags & TF_NEEDSYN) 2401 tp->t_flags |= TF_DELACK; 2402 else 2403 tp->t_flags |= TF_ACKNOW; 2404 tp->rcv_nxt++; 2405 } 2406 switch (tp->t_state) { 2407 2408 /* 2409 * In SYN_RECEIVED and ESTABLISHED STATES 2410 * enter the CLOSE_WAIT state. 2411 */ 2412 case TCPS_SYN_RECEIVED: 2413 tp->t_starttime = ticks; 2414 /*FALLTHROUGH*/ 2415 case TCPS_ESTABLISHED: 2416 tp->t_state = TCPS_CLOSE_WAIT; 2417 break; 2418 2419 /* 2420 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2421 * enter the CLOSING state. 2422 */ 2423 case TCPS_FIN_WAIT_1: 2424 tp->t_state = TCPS_CLOSING; 2425 break; 2426 2427 /* 2428 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2429 * starting the time-wait timer, turning off the other 2430 * standard timers. 2431 */ 2432 case TCPS_FIN_WAIT_2: 2433 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2434 "TCP_FIN_WAIT_2: head not locked")); 2435 tcp_twstart(tp); 2436 INP_INFO_WUNLOCK(&tcbinfo); 2437 return; 2438 2439 /* 2440 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2441 */ 2442 case TCPS_TIME_WAIT: 2443 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2444 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2445 tcp_timer_2msl, tp); 2446 break; 2447 } 2448 } 2449 INP_INFO_WUNLOCK(&tcbinfo); 2450 headlocked = 0; 2451#ifdef TCPDEBUG 2452 if (so->so_options & SO_DEBUG) 2453 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2454 &tcp_savetcp, 0); 2455#endif 2456 2457 /* 2458 * Return any desired output. 2459 */ 2460 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2461 (void) tcp_output(tp); 2462 2463check_delack: 2464 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2465 INP_LOCK_ASSERT(inp); 2466 if (tp->t_flags & TF_DELACK) { 2467 tp->t_flags &= ~TF_DELACK; 2468 callout_reset(tp->tt_delack, tcp_delacktime, 2469 tcp_timer_delack, tp); 2470 } 2471 INP_UNLOCK(inp); 2472 return; 2473 2474dropafterack: 2475 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2476 /* 2477 * Generate an ACK dropping incoming segment if it occupies 2478 * sequence space, where the ACK reflects our state. 2479 * 2480 * We can now skip the test for the RST flag since all 2481 * paths to this code happen after packets containing 2482 * RST have been dropped. 2483 * 2484 * In the SYN-RECEIVED state, don't send an ACK unless the 2485 * segment we received passes the SYN-RECEIVED ACK test. 2486 * If it fails send a RST. This breaks the loop in the 2487 * "LAND" DoS attack, and also prevents an ACK storm 2488 * between two listening ports that have been sent forged 2489 * SYN segments, each with the source address of the other. 2490 */ 2491 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2492 (SEQ_GT(tp->snd_una, th->th_ack) || 2493 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2494 rstreason = BANDLIM_RST_OPENPORT; 2495 goto dropwithreset; 2496 } 2497#ifdef TCPDEBUG 2498 if (so->so_options & SO_DEBUG) 2499 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2500 &tcp_savetcp, 0); 2501#endif 2502 KASSERT(headlocked, ("headlocked should be 1")); 2503 INP_INFO_WUNLOCK(&tcbinfo); 2504 tp->t_flags |= TF_ACKNOW; 2505 (void) tcp_output(tp); 2506 INP_UNLOCK(inp); 2507 m_freem(m); 2508 return; 2509 2510dropwithreset: 2511 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2512 /* 2513 * Generate a RST, dropping incoming segment. 2514 * Make ACK acceptable to originator of segment. 2515 * Don't bother to respond if destination was broadcast/multicast. 2516 */ 2517 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2518 goto drop; 2519 if (isipv6) { 2520 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2521 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2522 goto drop; 2523 } else { 2524 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2525 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2526 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2527 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2528 goto drop; 2529 } 2530 /* IPv6 anycast check is done at tcp6_input() */ 2531 2532 /* 2533 * Perform bandwidth limiting. 2534 */ 2535 if (badport_bandlim(rstreason) < 0) 2536 goto drop; 2537 2538#ifdef TCPDEBUG 2539 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2540 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2541 &tcp_savetcp, 0); 2542#endif 2543 2544 if (thflags & TH_ACK) 2545 /* mtod() below is safe as long as hdr dropping is delayed */ 2546 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2547 TH_RST); 2548 else { 2549 if (thflags & TH_SYN) 2550 tlen++; 2551 /* mtod() below is safe as long as hdr dropping is delayed */ 2552 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2553 (tcp_seq)0, TH_RST|TH_ACK); 2554 } 2555 2556 if (tp != NULL) 2557 INP_UNLOCK(inp); 2558 if (headlocked) 2559 INP_INFO_WUNLOCK(&tcbinfo); 2560 return; 2561 2562drop: 2563 /* 2564 * Drop space held by incoming segment and return. 2565 */ 2566#ifdef TCPDEBUG 2567 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2568 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2569 &tcp_savetcp, 0); 2570#endif 2571 if (tp != NULL) 2572 INP_UNLOCK(inp); 2573 if (headlocked) 2574 INP_INFO_WUNLOCK(&tcbinfo); 2575 m_freem(m); 2576 return; 2577} 2578 2579/* 2580 * Parse TCP options and place in tcpopt. 2581 */ 2582static void 2583tcp_dooptions(to, cp, cnt, is_syn) 2584 struct tcpopt *to; 2585 u_char *cp; 2586 int cnt; 2587 int is_syn; 2588{ 2589 int opt, optlen; 2590 2591 to->to_flags = 0; 2592 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2593 opt = cp[0]; 2594 if (opt == TCPOPT_EOL) 2595 break; 2596 if (opt == TCPOPT_NOP) 2597 optlen = 1; 2598 else { 2599 if (cnt < 2) 2600 break; 2601 optlen = cp[1]; 2602 if (optlen < 2 || optlen > cnt) 2603 break; 2604 } 2605 switch (opt) { 2606 case TCPOPT_MAXSEG: 2607 if (optlen != TCPOLEN_MAXSEG) 2608 continue; 2609 if (!is_syn) 2610 continue; 2611 to->to_flags |= TOF_MSS; 2612 bcopy((char *)cp + 2, 2613 (char *)&to->to_mss, sizeof(to->to_mss)); 2614 to->to_mss = ntohs(to->to_mss); 2615 break; 2616 case TCPOPT_WINDOW: 2617 if (optlen != TCPOLEN_WINDOW) 2618 continue; 2619 if (! is_syn) 2620 continue; 2621 to->to_flags |= TOF_SCALE; 2622 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2623 break; 2624 case TCPOPT_TIMESTAMP: 2625 if (optlen != TCPOLEN_TIMESTAMP) 2626 continue; 2627 to->to_flags |= TOF_TS; 2628 bcopy((char *)cp + 2, 2629 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2630 to->to_tsval = ntohl(to->to_tsval); 2631 bcopy((char *)cp + 6, 2632 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2633 to->to_tsecr = ntohl(to->to_tsecr); 2634 /* 2635 * If echoed timestamp is later than the current time, 2636 * fall back to non RFC1323 RTT calculation. 2637 */ 2638 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks)) 2639 to->to_tsecr = 0; 2640 break; 2641#ifdef TCP_SIGNATURE 2642 /* 2643 * XXX In order to reply to a host which has set the 2644 * TCP_SIGNATURE option in its initial SYN, we have to 2645 * record the fact that the option was observed here 2646 * for the syncache code to perform the correct response. 2647 */ 2648 case TCPOPT_SIGNATURE: 2649 if (optlen != TCPOLEN_SIGNATURE) 2650 continue; 2651 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2652 break; 2653#endif 2654 case TCPOPT_SACK_PERMITTED: 2655 if (!tcp_do_sack || 2656 optlen != TCPOLEN_SACK_PERMITTED) 2657 continue; 2658 if (is_syn) { 2659 /* MUST only be set on SYN */ 2660 to->to_flags |= TOF_SACK; 2661 } 2662 break; 2663 case TCPOPT_SACK: 2664 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2665 continue; 2666 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2667 to->to_sacks = cp + 2; 2668 tcpstat.tcps_sack_rcv_blocks++; 2669 break; 2670 default: 2671 continue; 2672 } 2673 } 2674} 2675 2676/* 2677 * Pull out of band byte out of a segment so 2678 * it doesn't appear in the user's data queue. 2679 * It is still reflected in the segment length for 2680 * sequencing purposes. 2681 */ 2682static void 2683tcp_pulloutofband(so, th, m, off) 2684 struct socket *so; 2685 struct tcphdr *th; 2686 register struct mbuf *m; 2687 int off; /* delayed to be droped hdrlen */ 2688{ 2689 int cnt = off + th->th_urp - 1; 2690 2691 while (cnt >= 0) { 2692 if (m->m_len > cnt) { 2693 char *cp = mtod(m, caddr_t) + cnt; 2694 struct tcpcb *tp = sototcpcb(so); 2695 2696 tp->t_iobc = *cp; 2697 tp->t_oobflags |= TCPOOB_HAVEDATA; 2698 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2699 m->m_len--; 2700 if (m->m_flags & M_PKTHDR) 2701 m->m_pkthdr.len--; 2702 return; 2703 } 2704 cnt -= m->m_len; 2705 m = m->m_next; 2706 if (m == 0) 2707 break; 2708 } 2709 panic("tcp_pulloutofband"); 2710} 2711 2712/* 2713 * Collect new round-trip time estimate 2714 * and update averages and current timeout. 2715 */ 2716static void 2717tcp_xmit_timer(tp, rtt) 2718 register struct tcpcb *tp; 2719 int rtt; 2720{ 2721 register int delta; 2722 2723 INP_LOCK_ASSERT(tp->t_inpcb); 2724 2725 tcpstat.tcps_rttupdated++; 2726 tp->t_rttupdated++; 2727 if (tp->t_srtt != 0) { 2728 /* 2729 * srtt is stored as fixed point with 5 bits after the 2730 * binary point (i.e., scaled by 8). The following magic 2731 * is equivalent to the smoothing algorithm in rfc793 with 2732 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2733 * point). Adjust rtt to origin 0. 2734 */ 2735 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2736 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2737 2738 if ((tp->t_srtt += delta) <= 0) 2739 tp->t_srtt = 1; 2740 2741 /* 2742 * We accumulate a smoothed rtt variance (actually, a 2743 * smoothed mean difference), then set the retransmit 2744 * timer to smoothed rtt + 4 times the smoothed variance. 2745 * rttvar is stored as fixed point with 4 bits after the 2746 * binary point (scaled by 16). The following is 2747 * equivalent to rfc793 smoothing with an alpha of .75 2748 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2749 * rfc793's wired-in beta. 2750 */ 2751 if (delta < 0) 2752 delta = -delta; 2753 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2754 if ((tp->t_rttvar += delta) <= 0) 2755 tp->t_rttvar = 1; 2756 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2757 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2758 } else { 2759 /* 2760 * No rtt measurement yet - use the unsmoothed rtt. 2761 * Set the variance to half the rtt (so our first 2762 * retransmit happens at 3*rtt). 2763 */ 2764 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2765 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2766 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2767 } 2768 tp->t_rtttime = 0; 2769 tp->t_rxtshift = 0; 2770 2771 /* 2772 * the retransmit should happen at rtt + 4 * rttvar. 2773 * Because of the way we do the smoothing, srtt and rttvar 2774 * will each average +1/2 tick of bias. When we compute 2775 * the retransmit timer, we want 1/2 tick of rounding and 2776 * 1 extra tick because of +-1/2 tick uncertainty in the 2777 * firing of the timer. The bias will give us exactly the 2778 * 1.5 tick we need. But, because the bias is 2779 * statistical, we have to test that we don't drop below 2780 * the minimum feasible timer (which is 2 ticks). 2781 */ 2782 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2783 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2784 2785 /* 2786 * We received an ack for a packet that wasn't retransmitted; 2787 * it is probably safe to discard any error indications we've 2788 * received recently. This isn't quite right, but close enough 2789 * for now (a route might have failed after we sent a segment, 2790 * and the return path might not be symmetrical). 2791 */ 2792 tp->t_softerror = 0; 2793} 2794 2795/* 2796 * Determine a reasonable value for maxseg size. 2797 * If the route is known, check route for mtu. 2798 * If none, use an mss that can be handled on the outgoing 2799 * interface without forcing IP to fragment; if bigger than 2800 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2801 * to utilize large mbufs. If no route is found, route has no mtu, 2802 * or the destination isn't local, use a default, hopefully conservative 2803 * size (usually 512 or the default IP max size, but no more than the mtu 2804 * of the interface), as we can't discover anything about intervening 2805 * gateways or networks. We also initialize the congestion/slow start 2806 * window to be a single segment if the destination isn't local. 2807 * While looking at the routing entry, we also initialize other path-dependent 2808 * parameters from pre-set or cached values in the routing entry. 2809 * 2810 * Also take into account the space needed for options that we 2811 * send regularly. Make maxseg shorter by that amount to assure 2812 * that we can send maxseg amount of data even when the options 2813 * are present. Store the upper limit of the length of options plus 2814 * data in maxopd. 2815 * 2816 * 2817 * In case of T/TCP, we call this routine during implicit connection 2818 * setup as well (offer = -1), to initialize maxseg from the cached 2819 * MSS of our peer. 2820 * 2821 * NOTE that this routine is only called when we process an incoming 2822 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2823 */ 2824void 2825tcp_mss(tp, offer) 2826 struct tcpcb *tp; 2827 int offer; 2828{ 2829 int rtt, mss; 2830 u_long bufsize; 2831 u_long maxmtu; 2832 struct inpcb *inp = tp->t_inpcb; 2833 struct socket *so; 2834 struct hc_metrics_lite metrics; 2835 int origoffer = offer; 2836#ifdef INET6 2837 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2838 size_t min_protoh = isipv6 ? 2839 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2840 sizeof (struct tcpiphdr); 2841#else 2842 const size_t min_protoh = sizeof(struct tcpiphdr); 2843#endif 2844 2845 /* initialize */ 2846#ifdef INET6 2847 if (isipv6) { 2848 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2849 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2850 } else 2851#endif 2852 { 2853 maxmtu = tcp_maxmtu(&inp->inp_inc); 2854 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2855 } 2856 so = inp->inp_socket; 2857 2858 /* 2859 * no route to sender, stay with default mss and return 2860 */ 2861 if (maxmtu == 0) 2862 return; 2863 2864 /* what have we got? */ 2865 switch (offer) { 2866 case 0: 2867 /* 2868 * Offer == 0 means that there was no MSS on the SYN 2869 * segment, in this case we use tcp_mssdflt. 2870 */ 2871 offer = 2872#ifdef INET6 2873 isipv6 ? tcp_v6mssdflt : 2874#endif 2875 tcp_mssdflt; 2876 break; 2877 2878 case -1: 2879 /* 2880 * Offer == -1 means that we didn't receive SYN yet. 2881 */ 2882 /* FALLTHROUGH */ 2883 2884 default: 2885 /* 2886 * Prevent DoS attack with too small MSS. Round up 2887 * to at least minmss. 2888 */ 2889 offer = max(offer, tcp_minmss); 2890 /* 2891 * Sanity check: make sure that maxopd will be large 2892 * enough to allow some data on segments even if the 2893 * all the option space is used (40bytes). Otherwise 2894 * funny things may happen in tcp_output. 2895 */ 2896 offer = max(offer, 64); 2897 } 2898 2899 /* 2900 * rmx information is now retrieved from tcp_hostcache 2901 */ 2902 tcp_hc_get(&inp->inp_inc, &metrics); 2903 2904 /* 2905 * if there's a discovered mtu int tcp hostcache, use it 2906 * else, use the link mtu. 2907 */ 2908 if (metrics.rmx_mtu) 2909 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2910 else { 2911#ifdef INET6 2912 if (isipv6) { 2913 mss = maxmtu - min_protoh; 2914 if (!path_mtu_discovery && 2915 !in6_localaddr(&inp->in6p_faddr)) 2916 mss = min(mss, tcp_v6mssdflt); 2917 } else 2918#endif 2919 { 2920 mss = maxmtu - min_protoh; 2921 if (!path_mtu_discovery && 2922 !in_localaddr(inp->inp_faddr)) 2923 mss = min(mss, tcp_mssdflt); 2924 } 2925 } 2926 mss = min(mss, offer); 2927 2928 /* 2929 * maxopd stores the maximum length of data AND options 2930 * in a segment; maxseg is the amount of data in a normal 2931 * segment. We need to store this value (maxopd) apart 2932 * from maxseg, because now every segment carries options 2933 * and thus we normally have somewhat less data in segments. 2934 */ 2935 tp->t_maxopd = mss; 2936 2937 /* 2938 * origoffer==-1 indicates, that no segments were received yet. 2939 * In this case we just guess. 2940 */ 2941 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2942 (origoffer == -1 || 2943 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2944 mss -= TCPOLEN_TSTAMP_APPA; 2945 tp->t_maxseg = mss; 2946 2947#if (MCLBYTES & (MCLBYTES - 1)) == 0 2948 if (mss > MCLBYTES) 2949 mss &= ~(MCLBYTES-1); 2950#else 2951 if (mss > MCLBYTES) 2952 mss = mss / MCLBYTES * MCLBYTES; 2953#endif 2954 tp->t_maxseg = mss; 2955 2956 /* 2957 * If there's a pipesize, change the socket buffer to that size, 2958 * don't change if sb_hiwat is different than default (then it 2959 * has been changed on purpose with setsockopt). 2960 * Make the socket buffers an integral number of mss units; 2961 * if the mss is larger than the socket buffer, decrease the mss. 2962 */ 2963 SOCKBUF_LOCK(&so->so_snd); 2964 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2965 bufsize = metrics.rmx_sendpipe; 2966 else 2967 bufsize = so->so_snd.sb_hiwat; 2968 if (bufsize < mss) 2969 mss = bufsize; 2970 else { 2971 bufsize = roundup(bufsize, mss); 2972 if (bufsize > sb_max) 2973 bufsize = sb_max; 2974 if (bufsize > so->so_snd.sb_hiwat) 2975 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2976 } 2977 SOCKBUF_UNLOCK(&so->so_snd); 2978 tp->t_maxseg = mss; 2979 2980 SOCKBUF_LOCK(&so->so_rcv); 2981 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2982 bufsize = metrics.rmx_recvpipe; 2983 else 2984 bufsize = so->so_rcv.sb_hiwat; 2985 if (bufsize > mss) { 2986 bufsize = roundup(bufsize, mss); 2987 if (bufsize > sb_max) 2988 bufsize = sb_max; 2989 if (bufsize > so->so_rcv.sb_hiwat) 2990 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2991 } 2992 SOCKBUF_UNLOCK(&so->so_rcv); 2993 /* 2994 * While we're here, check the others too 2995 */ 2996 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2997 tp->t_srtt = rtt; 2998 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2999 tcpstat.tcps_usedrtt++; 3000 if (metrics.rmx_rttvar) { 3001 tp->t_rttvar = metrics.rmx_rttvar; 3002 tcpstat.tcps_usedrttvar++; 3003 } else { 3004 /* default variation is +- 1 rtt */ 3005 tp->t_rttvar = 3006 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3007 } 3008 TCPT_RANGESET(tp->t_rxtcur, 3009 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3010 tp->t_rttmin, TCPTV_REXMTMAX); 3011 } 3012 if (metrics.rmx_ssthresh) { 3013 /* 3014 * There's some sort of gateway or interface 3015 * buffer limit on the path. Use this to set 3016 * the slow start threshhold, but set the 3017 * threshold to no less than 2*mss. 3018 */ 3019 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3020 tcpstat.tcps_usedssthresh++; 3021 } 3022 if (metrics.rmx_bandwidth) 3023 tp->snd_bandwidth = metrics.rmx_bandwidth; 3024 3025 /* 3026 * Set the slow-start flight size depending on whether this 3027 * is a local network or not. 3028 * 3029 * Extend this so we cache the cwnd too and retrieve it here. 3030 * Make cwnd even bigger than RFC3390 suggests but only if we 3031 * have previous experience with the remote host. Be careful 3032 * not make cwnd bigger than remote receive window or our own 3033 * send socket buffer. Maybe put some additional upper bound 3034 * on the retrieved cwnd. Should do incremental updates to 3035 * hostcache when cwnd collapses so next connection doesn't 3036 * overloads the path again. 3037 * 3038 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3039 * We currently check only in syncache_socket for that. 3040 */ 3041#define TCP_METRICS_CWND 3042#ifdef TCP_METRICS_CWND 3043 if (metrics.rmx_cwnd) 3044 tp->snd_cwnd = max(mss, 3045 min(metrics.rmx_cwnd / 2, 3046 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3047 else 3048#endif 3049 if (tcp_do_rfc3390) 3050 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3051#ifdef INET6 3052 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3053 (!isipv6 && in_localaddr(inp->inp_faddr))) 3054#else 3055 else if (in_localaddr(inp->inp_faddr)) 3056#endif 3057 tp->snd_cwnd = mss * ss_fltsz_local; 3058 else 3059 tp->snd_cwnd = mss * ss_fltsz; 3060} 3061 3062/* 3063 * Determine the MSS option to send on an outgoing SYN. 3064 */ 3065int 3066tcp_mssopt(inc) 3067 struct in_conninfo *inc; 3068{ 3069 int mss = 0; 3070 u_long maxmtu = 0; 3071 u_long thcmtu = 0; 3072 size_t min_protoh; 3073#ifdef INET6 3074 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3075#endif 3076 3077 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3078 3079#ifdef INET6 3080 if (isipv6) { 3081 mss = tcp_v6mssdflt; 3082 maxmtu = tcp_maxmtu6(inc); 3083 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3084 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3085 } else 3086#endif 3087 { 3088 mss = tcp_mssdflt; 3089 maxmtu = tcp_maxmtu(inc); 3090 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3091 min_protoh = sizeof(struct tcpiphdr); 3092 } 3093 if (maxmtu && thcmtu) 3094 mss = min(maxmtu, thcmtu) - min_protoh; 3095 else if (maxmtu || thcmtu) 3096 mss = max(maxmtu, thcmtu) - min_protoh; 3097 3098 return (mss); 3099} 3100 3101 3102/* 3103 * On a partial ack arrives, force the retransmission of the 3104 * next unacknowledged segment. Do not clear tp->t_dupacks. 3105 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3106 * be started again. 3107 */ 3108static void 3109tcp_newreno_partial_ack(tp, th) 3110 struct tcpcb *tp; 3111 struct tcphdr *th; 3112{ 3113 tcp_seq onxt = tp->snd_nxt; 3114 u_long ocwnd = tp->snd_cwnd; 3115 3116 callout_stop(tp->tt_rexmt); 3117 tp->t_rtttime = 0; 3118 tp->snd_nxt = th->th_ack; 3119 /* 3120 * Set snd_cwnd to one segment beyond acknowledged offset. 3121 * (tp->snd_una has not yet been updated when this function is called.) 3122 */ 3123 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3124 tp->t_flags |= TF_ACKNOW; 3125 (void) tcp_output(tp); 3126 tp->snd_cwnd = ocwnd; 3127 if (SEQ_GT(onxt, tp->snd_nxt)) 3128 tp->snd_nxt = onxt; 3129 /* 3130 * Partial window deflation. Relies on fact that tp->snd_una 3131 * not updated yet. 3132 */ 3133 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3134 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3135 else 3136 tp->snd_cwnd = 0; 3137 tp->snd_cwnd += tp->t_maxseg; 3138} 3139 3140/* 3141 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3142 * looking for a pcb in the listen state. Returns 0 otherwise. 3143 */ 3144static int 3145tcp_timewait(tw, to, th, m, tlen) 3146 struct tcptw *tw; 3147 struct tcpopt *to; 3148 struct tcphdr *th; 3149 struct mbuf *m; 3150 int tlen; 3151{ 3152 int thflags; 3153 tcp_seq seq; 3154#ifdef INET6 3155 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3156#else 3157 const int isipv6 = 0; 3158#endif 3159 3160 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */ 3161 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3162 INP_LOCK_ASSERT(tw->tw_inpcb); 3163 3164 thflags = th->th_flags; 3165 3166 /* 3167 * NOTE: for FIN_WAIT_2 (to be added later), 3168 * must validate sequence number before accepting RST 3169 */ 3170 3171 /* 3172 * If the segment contains RST: 3173 * Drop the segment - see Stevens, vol. 2, p. 964 and 3174 * RFC 1337. 3175 */ 3176 if (thflags & TH_RST) 3177 goto drop; 3178 3179#if 0 3180/* PAWS not needed at the moment */ 3181 /* 3182 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3183 * and it's less than ts_recent, drop it. 3184 */ 3185 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3186 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3187 if ((thflags & TH_ACK) == 0) 3188 goto drop; 3189 goto ack; 3190 } 3191 /* 3192 * ts_recent is never updated because we never accept new segments. 3193 */ 3194#endif 3195 3196 /* 3197 * If a new connection request is received 3198 * while in TIME_WAIT, drop the old connection 3199 * and start over if the sequence numbers 3200 * are above the previous ones. 3201 */ 3202 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3203 tcp_twclose(tw, 0); 3204 return (1); 3205 } 3206 3207 /* 3208 * Drop the the segment if it does not contain an ACK. 3209 */ 3210 if ((thflags & TH_ACK) == 0) 3211 goto drop; 3212 3213 /* 3214 * Reset the 2MSL timer if this is a duplicate FIN. 3215 */ 3216 if (thflags & TH_FIN) { 3217 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3218 if (seq + 1 == tw->rcv_nxt) 3219 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3220 } 3221 3222 /* 3223 * Acknowledge the segment if it has data or is not a duplicate ACK. 3224 */ 3225 if (thflags != TH_ACK || tlen != 0 || 3226 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3227 tcp_twrespond(tw, TH_ACK); 3228 goto drop; 3229 3230 /* 3231 * Generate a RST, dropping incoming segment. 3232 * Make ACK acceptable to originator of segment. 3233 * Don't bother to respond if destination was broadcast/multicast. 3234 */ 3235 if (m->m_flags & (M_BCAST|M_MCAST)) 3236 goto drop; 3237 if (isipv6) { 3238 struct ip6_hdr *ip6; 3239 3240 /* IPv6 anycast check is done at tcp6_input() */ 3241 ip6 = mtod(m, struct ip6_hdr *); 3242 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3243 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3244 goto drop; 3245 } else { 3246 struct ip *ip; 3247 3248 ip = mtod(m, struct ip *); 3249 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3250 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3251 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3252 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3253 goto drop; 3254 } 3255 if (thflags & TH_ACK) { 3256 tcp_respond(NULL, 3257 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3258 } else { 3259 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3260 tcp_respond(NULL, 3261 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3262 } 3263 INP_UNLOCK(tw->tw_inpcb); 3264 return (0); 3265 3266drop: 3267 INP_UNLOCK(tw->tw_inpcb); 3268 m_freem(m); 3269 return (0); 3270} 3271