tcp_input.c revision 147605
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 147605 2005-06-25 22:24:45Z ups $ 31 */ 32 33#include "opt_ipfw.h" /* for ipfw_fwd */ 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39#include "opt_tcp_input.h" 40#include "opt_tcp_sack.h" 41 42#include <sys/param.h> 43#include <sys/kernel.h> 44#include <sys/mac.h> 45#include <sys/malloc.h> 46#include <sys/mbuf.h> 47#include <sys/proc.h> /* for proc0 declaration */ 48#include <sys/protosw.h> 49#include <sys/signalvar.h> 50#include <sys/socket.h> 51#include <sys/socketvar.h> 52#include <sys/sysctl.h> 53#include <sys/syslog.h> 54#include <sys/systm.h> 55 56#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 57 58#include <vm/uma.h> 59 60#include <net/if.h> 61#include <net/route.h> 62 63#include <netinet/in.h> 64#include <netinet/in_pcb.h> 65#include <netinet/in_systm.h> 66#include <netinet/in_var.h> 67#include <netinet/ip.h> 68#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 69#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 70#include <netinet/ip_var.h> 71#include <netinet/ip6.h> 72#include <netinet/icmp6.h> 73#include <netinet6/in6_pcb.h> 74#include <netinet6/ip6_var.h> 75#include <netinet6/nd6.h> 76#include <netinet/tcp.h> 77#include <netinet/tcp_fsm.h> 78#include <netinet/tcp_seq.h> 79#include <netinet/tcp_timer.h> 80#include <netinet/tcp_var.h> 81#include <netinet6/tcp6_var.h> 82#include <netinet/tcpip.h> 83#ifdef TCPDEBUG 84#include <netinet/tcp_debug.h> 85#endif /* TCPDEBUG */ 86 87#ifdef FAST_IPSEC 88#include <netipsec/ipsec.h> 89#include <netipsec/ipsec6.h> 90#endif /*FAST_IPSEC*/ 91 92#ifdef IPSEC 93#include <netinet6/ipsec.h> 94#include <netinet6/ipsec6.h> 95#include <netkey/key.h> 96#endif /*IPSEC*/ 97 98#include <machine/in_cksum.h> 99 100static const int tcprexmtthresh = 3; 101 102struct tcpstat tcpstat; 103SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 104 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 105 106static int log_in_vain = 0; 107SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 108 &log_in_vain, 0, "Log all incoming TCP connections"); 109 110static int blackhole = 0; 111SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 112 &blackhole, 0, "Do not send RST when dropping refused connections"); 113 114int tcp_delack_enabled = 1; 115SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 116 &tcp_delack_enabled, 0, 117 "Delay ACK to try and piggyback it onto a data packet"); 118 119#ifdef TCP_DROP_SYNFIN 120static int drop_synfin = 0; 121SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 122 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 123#endif 124 125static int tcp_do_rfc3042 = 1; 126SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 127 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 128 129static int tcp_do_rfc3390 = 1; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 131 &tcp_do_rfc3390, 0, 132 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 133 134static int tcp_insecure_rst = 0; 135SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 136 &tcp_insecure_rst, 0, 137 "Follow the old (insecure) criteria for accepting RST packets."); 138 139SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 140 "TCP Segment Reassembly Queue"); 141 142static int tcp_reass_maxseg = 0; 143SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 144 &tcp_reass_maxseg, 0, 145 "Global maximum number of TCP Segments in Reassembly Queue"); 146 147int tcp_reass_qsize = 0; 148SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 149 &tcp_reass_qsize, 0, 150 "Global number of TCP Segments currently in Reassembly Queue"); 151 152static int tcp_reass_maxqlen = 48; 153SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 154 &tcp_reass_maxqlen, 0, 155 "Maximum number of TCP Segments per individual Reassembly Queue"); 156 157static int tcp_reass_overflows = 0; 158SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 159 &tcp_reass_overflows, 0, 160 "Global number of TCP Segment Reassembly Queue Overflows"); 161 162struct inpcbhead tcb; 163#define tcb6 tcb /* for KAME src sync over BSD*'s */ 164struct inpcbinfo tcbinfo; 165struct mtx *tcbinfo_mtx; 166 167static void tcp_dooptions(struct tcpcb *, struct tcpopt *, u_char *, 168 int, int, struct tcphdr *); 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 if (inp->inp_vflag & INP_TIMEWAIT) { 745 /* 746 * The only option of relevance is TOF_CC, and only if 747 * present in a SYN segment. See tcp_timewait(). 748 */ 749 if (thflags & TH_SYN) 750 tcp_dooptions((struct tcpcb *)NULL, &to, optp, optlen, 1, th); 751 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 752 &to, th, m, tlen)) 753 goto findpcb; 754 /* 755 * tcp_timewait unlocks inp. 756 */ 757 INP_INFO_WUNLOCK(&tcbinfo); 758 return; 759 } 760 tp = intotcpcb(inp); 761 if (tp == 0) { 762 INP_UNLOCK(inp); 763 rstreason = BANDLIM_RST_CLOSEDPORT; 764 goto dropwithreset; 765 } 766 if (tp->t_state == TCPS_CLOSED) 767 goto drop; 768 769 /* Unscale the window into a 32-bit value. */ 770 if ((thflags & TH_SYN) == 0) 771 tiwin = th->th_win << tp->snd_scale; 772 else 773 tiwin = th->th_win; 774 775#ifdef MAC 776 INP_LOCK_ASSERT(inp); 777 if (mac_check_inpcb_deliver(inp, m)) 778 goto drop; 779#endif 780 so = inp->inp_socket; 781#ifdef TCPDEBUG 782 if (so->so_options & SO_DEBUG) { 783 ostate = tp->t_state; 784 if (isipv6) 785 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 786 else 787 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 788 tcp_savetcp = *th; 789 } 790#endif 791 if (so->so_options & SO_ACCEPTCONN) { 792 struct in_conninfo inc; 793 794#ifdef INET6 795 inc.inc_isipv6 = isipv6; 796#endif 797 if (isipv6) { 798 inc.inc6_faddr = ip6->ip6_src; 799 inc.inc6_laddr = ip6->ip6_dst; 800 } else { 801 inc.inc_faddr = ip->ip_src; 802 inc.inc_laddr = ip->ip_dst; 803 } 804 inc.inc_fport = th->th_sport; 805 inc.inc_lport = th->th_dport; 806 807 /* 808 * If the state is LISTEN then ignore segment if it contains 809 * a RST. If the segment contains an ACK then it is bad and 810 * send a RST. If it does not contain a SYN then it is not 811 * interesting; drop it. 812 * 813 * If the state is SYN_RECEIVED (syncache) and seg contains 814 * an ACK, but not for our SYN/ACK, send a RST. If the seg 815 * contains a RST, check the sequence number to see if it 816 * is a valid reset segment. 817 */ 818 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 819 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 820 if (!syncache_expand(&inc, th, &so, m)) { 821 /* 822 * No syncache entry, or ACK was not 823 * for our SYN/ACK. Send a RST. 824 */ 825 tcpstat.tcps_badsyn++; 826 rstreason = BANDLIM_RST_OPENPORT; 827 goto dropwithreset; 828 } 829 if (so == NULL) { 830 /* 831 * Could not complete 3-way handshake, 832 * connection is being closed down, and 833 * syncache will free mbuf. 834 */ 835 INP_UNLOCK(inp); 836 INP_INFO_WUNLOCK(&tcbinfo); 837 return; 838 } 839 /* 840 * Socket is created in state SYN_RECEIVED. 841 * Continue processing segment. 842 */ 843 INP_UNLOCK(inp); 844 inp = sotoinpcb(so); 845 INP_LOCK(inp); 846 tp = intotcpcb(inp); 847 /* 848 * This is what would have happened in 849 * tcp_output() when the SYN,ACK was sent. 850 */ 851 tp->snd_up = tp->snd_una; 852 tp->snd_max = tp->snd_nxt = tp->iss + 1; 853 tp->last_ack_sent = tp->rcv_nxt; 854 /* 855 * RFC1323: The window in SYN & SYN/ACK 856 * segments is never scaled. 857 */ 858 tp->snd_wnd = tiwin; /* unscaled */ 859 goto after_listen; 860 } 861 if (thflags & TH_RST) { 862 syncache_chkrst(&inc, th); 863 goto drop; 864 } 865 if (thflags & TH_ACK) { 866 syncache_badack(&inc); 867 tcpstat.tcps_badsyn++; 868 rstreason = BANDLIM_RST_OPENPORT; 869 goto dropwithreset; 870 } 871 goto drop; 872 } 873 874 /* 875 * Segment's flags are (SYN) or (SYN|FIN). 876 */ 877#ifdef INET6 878 /* 879 * If deprecated address is forbidden, 880 * we do not accept SYN to deprecated interface 881 * address to prevent any new inbound connection from 882 * getting established. 883 * When we do not accept SYN, we send a TCP RST, 884 * with deprecated source address (instead of dropping 885 * it). We compromise it as it is much better for peer 886 * to send a RST, and RST will be the final packet 887 * for the exchange. 888 * 889 * If we do not forbid deprecated addresses, we accept 890 * the SYN packet. RFC2462 does not suggest dropping 891 * SYN in this case. 892 * If we decipher RFC2462 5.5.4, it says like this: 893 * 1. use of deprecated addr with existing 894 * communication is okay - "SHOULD continue to be 895 * used" 896 * 2. use of it with new communication: 897 * (2a) "SHOULD NOT be used if alternate address 898 * with sufficient scope is available" 899 * (2b) nothing mentioned otherwise. 900 * Here we fall into (2b) case as we have no choice in 901 * our source address selection - we must obey the peer. 902 * 903 * The wording in RFC2462 is confusing, and there are 904 * multiple description text for deprecated address 905 * handling - worse, they are not exactly the same. 906 * I believe 5.5.4 is the best one, so we follow 5.5.4. 907 */ 908 if (isipv6 && !ip6_use_deprecated) { 909 struct in6_ifaddr *ia6; 910 911 if ((ia6 = ip6_getdstifaddr(m)) && 912 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 913 INP_UNLOCK(inp); 914 tp = NULL; 915 rstreason = BANDLIM_RST_OPENPORT; 916 goto dropwithreset; 917 } 918 } 919#endif 920 /* 921 * If it is from this socket, drop it, it must be forged. 922 * Don't bother responding if the destination was a broadcast. 923 */ 924 if (th->th_dport == th->th_sport) { 925 if (isipv6) { 926 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 927 &ip6->ip6_src)) 928 goto drop; 929 } else { 930 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 931 goto drop; 932 } 933 } 934 /* 935 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 936 * 937 * Note that it is quite possible to receive unicast 938 * link-layer packets with a broadcast IP address. Use 939 * in_broadcast() to find them. 940 */ 941 if (m->m_flags & (M_BCAST|M_MCAST)) 942 goto drop; 943 if (isipv6) { 944 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 945 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 946 goto drop; 947 } else { 948 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 949 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 950 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 951 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 952 goto drop; 953 } 954 /* 955 * SYN appears to be valid; create compressed TCP state 956 * for syncache, or perform t/tcp connection. 957 */ 958 if (so->so_qlen <= so->so_qlimit) { 959#ifdef TCPDEBUG 960 if (so->so_options & SO_DEBUG) 961 tcp_trace(TA_INPUT, ostate, tp, 962 (void *)tcp_saveipgen, &tcp_savetcp, 0); 963#endif 964 tcp_dooptions(tp, &to, optp, optlen, 1, th); 965 if (!syncache_add(&inc, &to, th, &so, m)) 966 goto drop; 967 if (so == NULL) { 968 /* 969 * Entry added to syncache, mbuf used to 970 * send SYN,ACK packet. 971 */ 972 KASSERT(headlocked, ("headlocked")); 973 INP_UNLOCK(inp); 974 INP_INFO_WUNLOCK(&tcbinfo); 975 return; 976 } 977 /* 978 * Segment passed TAO tests. 979 */ 980 INP_UNLOCK(inp); 981 inp = sotoinpcb(so); 982 INP_LOCK(inp); 983 tp = intotcpcb(inp); 984 tp->snd_wnd = tiwin; 985 tp->t_starttime = ticks; 986 tp->t_state = TCPS_ESTABLISHED; 987 988 /* 989 * T/TCP logic: 990 * If there is a FIN or if there is data, then 991 * delay SYN,ACK(SYN) in the hope of piggy-backing 992 * it on a response segment. Otherwise must send 993 * ACK now in case the other side is slow starting. 994 */ 995 if (thflags & TH_FIN || tlen != 0) 996 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 997 else 998 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 999 tcpstat.tcps_connects++; 1000 soisconnected(so); 1001 goto trimthenstep6; 1002 } 1003 goto drop; 1004 } 1005after_listen: 1006 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1007 INP_LOCK_ASSERT(inp); 1008 1009 /* XXX temp debugging */ 1010 /* should not happen - syncache should pick up these connections */ 1011 if (tp->t_state == TCPS_LISTEN) 1012 panic("tcp_input: TCPS_LISTEN"); 1013 1014 /* 1015 * This is the second part of the MSS DoS prevention code (after 1016 * minmss on the sending side) and it deals with too many too small 1017 * tcp packets in a too short timeframe (1 second). 1018 * 1019 * For every full second we count the number of received packets 1020 * and bytes. If we get a lot of packets per second for this connection 1021 * (tcp_minmssoverload) we take a closer look at it and compute the 1022 * average packet size for the past second. If that is less than 1023 * tcp_minmss we get too many packets with very small payload which 1024 * is not good and burdens our system (and every packet generates 1025 * a wakeup to the process connected to our socket). We can reasonable 1026 * expect this to be small packet DoS attack to exhaust our CPU 1027 * cycles. 1028 * 1029 * Care has to be taken for the minimum packet overload value. This 1030 * value defines the minimum number of packets per second before we 1031 * start to worry. This must not be too low to avoid killing for 1032 * example interactive connections with many small packets like 1033 * telnet or SSH. 1034 * 1035 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1036 * this check. 1037 * 1038 * Account for packet if payload packet, skip over ACK, etc. 1039 */ 1040 if (tcp_minmss && tcp_minmssoverload && 1041 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1042 if ((unsigned int)(tp->rcv_second - ticks) < hz) { 1043 tp->rcv_pps++; 1044 tp->rcv_byps += tlen + off; 1045 if (tp->rcv_pps > tcp_minmssoverload) { 1046 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1047 printf("too many small tcp packets from " 1048 "%s:%u, av. %lubyte/packet, " 1049 "dropping connection\n", 1050#ifdef INET6 1051 isipv6 ? 1052 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1053#endif 1054 inet_ntoa(inp->inp_inc.inc_faddr), 1055 inp->inp_inc.inc_fport, 1056 tp->rcv_byps / tp->rcv_pps); 1057 KASSERT(headlocked, ("tcp_input: " 1058 "after_listen: tcp_drop: head " 1059 "not locked")); 1060 tp = tcp_drop(tp, ECONNRESET); 1061 tcpstat.tcps_minmssdrops++; 1062 goto drop; 1063 } 1064 } 1065 } else { 1066 tp->rcv_second = ticks + hz; 1067 tp->rcv_pps = 1; 1068 tp->rcv_byps = tlen + off; 1069 } 1070 } 1071 1072 /* 1073 * Segment received on connection. 1074 * Reset idle time and keep-alive timer. 1075 */ 1076 tp->t_rcvtime = ticks; 1077 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1078 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1079 1080 /* 1081 * Process options only when we get SYN/ACK back. The SYN case 1082 * for incoming connections is handled in tcp_syncache. 1083 * XXX this is traditional behavior, may need to be cleaned up. 1084 */ 1085 tcp_dooptions(tp, &to, optp, optlen, thflags & TH_SYN, th); 1086 if (thflags & TH_SYN) { 1087 if (to.to_flags & TOF_SCALE) { 1088 tp->t_flags |= TF_RCVD_SCALE; 1089 tp->requested_s_scale = to.to_requested_s_scale; 1090 } 1091 if (to.to_flags & TOF_TS) { 1092 tp->t_flags |= TF_RCVD_TSTMP; 1093 tp->ts_recent = to.to_tsval; 1094 tp->ts_recent_age = ticks; 1095 } 1096 if (to.to_flags & TOF_MSS) 1097 tcp_mss(tp, to.to_mss); 1098 if (tp->sack_enable) { 1099 if (!(to.to_flags & TOF_SACK)) 1100 tp->sack_enable = 0; 1101 else 1102 tp->t_flags |= TF_SACK_PERMIT; 1103 } 1104 1105 } 1106 1107 if (tp->sack_enable) { 1108 /* Delete stale (cumulatively acked) SACK holes */ 1109 tcp_del_sackholes(tp, th); 1110 } 1111 1112 /* 1113 * Header prediction: check for the two common cases 1114 * of a uni-directional data xfer. If the packet has 1115 * no control flags, is in-sequence, the window didn't 1116 * change and we're not retransmitting, it's a 1117 * candidate. If the length is zero and the ack moved 1118 * forward, we're the sender side of the xfer. Just 1119 * free the data acked & wake any higher level process 1120 * that was blocked waiting for space. If the length 1121 * is non-zero and the ack didn't move, we're the 1122 * receiver side. If we're getting packets in-order 1123 * (the reassembly queue is empty), add the data to 1124 * the socket buffer and note that we need a delayed ack. 1125 * Make sure that the hidden state-flags are also off. 1126 * Since we check for TCPS_ESTABLISHED above, it can only 1127 * be TH_NEEDSYN. 1128 */ 1129 if (tp->t_state == TCPS_ESTABLISHED && 1130 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1131 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1132 ((to.to_flags & TOF_TS) == 0 || 1133 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1134 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1135 tp->snd_nxt == tp->snd_max) { 1136 1137 /* 1138 * If last ACK falls within this segment's sequence numbers, 1139 * record the timestamp. 1140 * NOTE that the test is modified according to the latest 1141 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1142 */ 1143 if ((to.to_flags & TOF_TS) != 0 && 1144 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1145 tp->ts_recent_age = ticks; 1146 tp->ts_recent = to.to_tsval; 1147 } 1148 1149 if (tlen == 0) { 1150 if (SEQ_GT(th->th_ack, tp->snd_una) && 1151 SEQ_LEQ(th->th_ack, tp->snd_max) && 1152 tp->snd_cwnd >= tp->snd_wnd && 1153 ((!tcp_do_newreno && !tp->sack_enable && 1154 tp->t_dupacks < tcprexmtthresh) || 1155 ((tcp_do_newreno || tp->sack_enable) && 1156 !IN_FASTRECOVERY(tp)))) { 1157 KASSERT(headlocked, ("headlocked")); 1158 INP_INFO_WUNLOCK(&tcbinfo); 1159 headlocked = 0; 1160 /* 1161 * this is a pure ack for outstanding data. 1162 */ 1163 ++tcpstat.tcps_predack; 1164 /* 1165 * "bad retransmit" recovery 1166 */ 1167 if (tp->t_rxtshift == 1 && 1168 ticks < tp->t_badrxtwin) { 1169 ++tcpstat.tcps_sndrexmitbad; 1170 tp->snd_cwnd = tp->snd_cwnd_prev; 1171 tp->snd_ssthresh = 1172 tp->snd_ssthresh_prev; 1173 tp->snd_recover = tp->snd_recover_prev; 1174 if (tp->t_flags & TF_WASFRECOVERY) 1175 ENTER_FASTRECOVERY(tp); 1176 tp->snd_nxt = tp->snd_max; 1177 tp->t_badrxtwin = 0; 1178 } 1179 1180 /* 1181 * Recalculate the transmit timer / rtt. 1182 * 1183 * Some boxes send broken timestamp replies 1184 * during the SYN+ACK phase, ignore 1185 * timestamps of 0 or we could calculate a 1186 * huge RTT and blow up the retransmit timer. 1187 */ 1188 if ((to.to_flags & TOF_TS) != 0 && 1189 to.to_tsecr) { 1190 tcp_xmit_timer(tp, 1191 ticks - to.to_tsecr + 1); 1192 } else if (tp->t_rtttime && 1193 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1194 tcp_xmit_timer(tp, 1195 ticks - tp->t_rtttime); 1196 } 1197 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1198 acked = th->th_ack - tp->snd_una; 1199 tcpstat.tcps_rcvackpack++; 1200 tcpstat.tcps_rcvackbyte += acked; 1201 sbdrop(&so->so_snd, acked); 1202 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1203 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1204 tp->snd_recover = th->th_ack - 1; 1205 tp->snd_una = th->th_ack; 1206 /* 1207 * pull snd_wl2 up to prevent seq wrap relative 1208 * to th_ack. 1209 */ 1210 tp->snd_wl2 = th->th_ack; 1211 tp->t_dupacks = 0; 1212 m_freem(m); 1213 ND6_HINT(tp); /* some progress has been done */ 1214 1215 /* 1216 * If all outstanding data are acked, stop 1217 * retransmit timer, otherwise restart timer 1218 * using current (possibly backed-off) value. 1219 * If process is waiting for space, 1220 * wakeup/selwakeup/signal. If data 1221 * are ready to send, let tcp_output 1222 * decide between more output or persist. 1223 1224#ifdef TCPDEBUG 1225 if (so->so_options & SO_DEBUG) 1226 tcp_trace(TA_INPUT, ostate, tp, 1227 (void *)tcp_saveipgen, 1228 &tcp_savetcp, 0); 1229#endif 1230 */ 1231 if (tp->snd_una == tp->snd_max) 1232 callout_stop(tp->tt_rexmt); 1233 else if (!callout_active(tp->tt_persist)) 1234 callout_reset(tp->tt_rexmt, 1235 tp->t_rxtcur, 1236 tcp_timer_rexmt, tp); 1237 1238 sowwakeup(so); 1239 if (so->so_snd.sb_cc) 1240 (void) tcp_output(tp); 1241 goto check_delack; 1242 } 1243 } else if (th->th_ack == tp->snd_una && 1244 LIST_EMPTY(&tp->t_segq) && 1245 tlen <= sbspace(&so->so_rcv)) { 1246 KASSERT(headlocked, ("headlocked")); 1247 INP_INFO_WUNLOCK(&tcbinfo); 1248 headlocked = 0; 1249 /* 1250 * this is a pure, in-sequence data packet 1251 * with nothing on the reassembly queue and 1252 * we have enough buffer space to take it. 1253 */ 1254 /* Clean receiver SACK report if present */ 1255 if (tp->sack_enable && tp->rcv_numsacks) 1256 tcp_clean_sackreport(tp); 1257 ++tcpstat.tcps_preddat; 1258 tp->rcv_nxt += tlen; 1259 /* 1260 * Pull snd_wl1 up to prevent seq wrap relative to 1261 * th_seq. 1262 */ 1263 tp->snd_wl1 = th->th_seq; 1264 /* 1265 * Pull rcv_up up to prevent seq wrap relative to 1266 * rcv_nxt. 1267 */ 1268 tp->rcv_up = tp->rcv_nxt; 1269 tcpstat.tcps_rcvpack++; 1270 tcpstat.tcps_rcvbyte += tlen; 1271 ND6_HINT(tp); /* some progress has been done */ 1272 /* 1273#ifdef TCPDEBUG 1274 if (so->so_options & SO_DEBUG) 1275 tcp_trace(TA_INPUT, ostate, tp, 1276 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1277#endif 1278 * Add data to socket buffer. 1279 */ 1280 SOCKBUF_LOCK(&so->so_rcv); 1281 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1282 m_freem(m); 1283 } else { 1284 m_adj(m, drop_hdrlen); /* delayed header drop */ 1285 sbappendstream_locked(&so->so_rcv, m); 1286 } 1287 sorwakeup_locked(so); 1288 if (DELAY_ACK(tp)) { 1289 tp->t_flags |= TF_DELACK; 1290 } else { 1291 tp->t_flags |= TF_ACKNOW; 1292 tcp_output(tp); 1293 } 1294 goto check_delack; 1295 } 1296 } 1297 1298 /* 1299 * Calculate amount of space in receive window, 1300 * and then do TCP input processing. 1301 * Receive window is amount of space in rcv queue, 1302 * but not less than advertised window. 1303 */ 1304 { int win; 1305 1306 win = sbspace(&so->so_rcv); 1307 if (win < 0) 1308 win = 0; 1309 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1310 } 1311 1312 switch (tp->t_state) { 1313 1314 /* 1315 * If the state is SYN_RECEIVED: 1316 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1317 */ 1318 case TCPS_SYN_RECEIVED: 1319 if ((thflags & TH_ACK) && 1320 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1321 SEQ_GT(th->th_ack, tp->snd_max))) { 1322 rstreason = BANDLIM_RST_OPENPORT; 1323 goto dropwithreset; 1324 } 1325 break; 1326 1327 /* 1328 * If the state is SYN_SENT: 1329 * if seg contains an ACK, but not for our SYN, drop the input. 1330 * if seg contains a RST, then drop the connection. 1331 * if seg does not contain SYN, then drop it. 1332 * Otherwise this is an acceptable SYN segment 1333 * initialize tp->rcv_nxt and tp->irs 1334 * if seg contains ack then advance tp->snd_una 1335 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1336 * arrange for segment to be acked (eventually) 1337 * continue processing rest of data/controls, beginning with URG 1338 */ 1339 case TCPS_SYN_SENT: 1340 if ((thflags & TH_ACK) && 1341 (SEQ_LEQ(th->th_ack, tp->iss) || 1342 SEQ_GT(th->th_ack, tp->snd_max))) { 1343 rstreason = BANDLIM_UNLIMITED; 1344 goto dropwithreset; 1345 } 1346 if (thflags & TH_RST) { 1347 if (thflags & TH_ACK) { 1348 KASSERT(headlocked, ("tcp_input: after_listen" 1349 ": tcp_drop.2: head not locked")); 1350 tp = tcp_drop(tp, ECONNREFUSED); 1351 } 1352 goto drop; 1353 } 1354 if ((thflags & TH_SYN) == 0) 1355 goto drop; 1356 tp->snd_wnd = th->th_win; /* initial send window */ 1357 1358 tp->irs = th->th_seq; 1359 tcp_rcvseqinit(tp); 1360 if (thflags & TH_ACK) { 1361 tcpstat.tcps_connects++; 1362 soisconnected(so); 1363#ifdef MAC 1364 SOCK_LOCK(so); 1365 mac_set_socket_peer_from_mbuf(m, so); 1366 SOCK_UNLOCK(so); 1367#endif 1368 /* Do window scaling on this connection? */ 1369 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1370 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1371 tp->snd_scale = tp->requested_s_scale; 1372 tp->rcv_scale = tp->request_r_scale; 1373 } 1374 tp->rcv_adv += tp->rcv_wnd; 1375 tp->snd_una++; /* SYN is acked */ 1376 /* 1377 * If there's data, delay ACK; if there's also a FIN 1378 * ACKNOW will be turned on later. 1379 */ 1380 if (DELAY_ACK(tp) && tlen != 0) 1381 callout_reset(tp->tt_delack, tcp_delacktime, 1382 tcp_timer_delack, tp); 1383 else 1384 tp->t_flags |= TF_ACKNOW; 1385 /* 1386 * Received <SYN,ACK> in SYN_SENT[*] state. 1387 * Transitions: 1388 * SYN_SENT --> ESTABLISHED 1389 * SYN_SENT* --> FIN_WAIT_1 1390 */ 1391 tp->t_starttime = ticks; 1392 if (tp->t_flags & TF_NEEDFIN) { 1393 tp->t_state = TCPS_FIN_WAIT_1; 1394 tp->t_flags &= ~TF_NEEDFIN; 1395 thflags &= ~TH_SYN; 1396 } else { 1397 tp->t_state = TCPS_ESTABLISHED; 1398 callout_reset(tp->tt_keep, tcp_keepidle, 1399 tcp_timer_keep, tp); 1400 } 1401 } else { 1402 /* 1403 * Received initial SYN in SYN-SENT[*] state => 1404 * simultaneous open. If segment contains CC option 1405 * and there is a cached CC, apply TAO test. 1406 * If it succeeds, connection is * half-synchronized. 1407 * Otherwise, do 3-way handshake: 1408 * SYN-SENT -> SYN-RECEIVED 1409 * SYN-SENT* -> SYN-RECEIVED* 1410 * If there was no CC option, clear cached CC value. 1411 */ 1412 tp->t_flags |= TF_ACKNOW; 1413 callout_stop(tp->tt_rexmt); 1414 tp->t_state = TCPS_SYN_RECEIVED; 1415 } 1416 1417trimthenstep6: 1418 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1419 "locked")); 1420 INP_LOCK_ASSERT(inp); 1421 1422 /* 1423 * Advance th->th_seq to correspond to first data byte. 1424 * If data, trim to stay within window, 1425 * dropping FIN if necessary. 1426 */ 1427 th->th_seq++; 1428 if (tlen > tp->rcv_wnd) { 1429 todrop = tlen - tp->rcv_wnd; 1430 m_adj(m, -todrop); 1431 tlen = tp->rcv_wnd; 1432 thflags &= ~TH_FIN; 1433 tcpstat.tcps_rcvpackafterwin++; 1434 tcpstat.tcps_rcvbyteafterwin += todrop; 1435 } 1436 tp->snd_wl1 = th->th_seq - 1; 1437 tp->rcv_up = th->th_seq; 1438 /* 1439 * Client side of transaction: already sent SYN and data. 1440 * If the remote host used T/TCP to validate the SYN, 1441 * our data will be ACK'd; if so, enter normal data segment 1442 * processing in the middle of step 5, ack processing. 1443 * Otherwise, goto step 6. 1444 */ 1445 if (thflags & TH_ACK) 1446 goto process_ACK; 1447 1448 goto step6; 1449 1450 /* 1451 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1452 * do normal processing. 1453 * 1454 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1455 */ 1456 case TCPS_LAST_ACK: 1457 case TCPS_CLOSING: 1458 case TCPS_TIME_WAIT: 1459 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1460 break; /* continue normal processing */ 1461 } 1462 1463 /* 1464 * States other than LISTEN or SYN_SENT. 1465 * First check the RST flag and sequence number since reset segments 1466 * are exempt from the timestamp and connection count tests. This 1467 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1468 * below which allowed reset segments in half the sequence space 1469 * to fall though and be processed (which gives forged reset 1470 * segments with a random sequence number a 50 percent chance of 1471 * killing a connection). 1472 * Then check timestamp, if present. 1473 * Then check the connection count, if present. 1474 * Then check that at least some bytes of segment are within 1475 * receive window. If segment begins before rcv_nxt, 1476 * drop leading data (and SYN); if nothing left, just ack. 1477 * 1478 * 1479 * If the RST bit is set, check the sequence number to see 1480 * if this is a valid reset segment. 1481 * RFC 793 page 37: 1482 * In all states except SYN-SENT, all reset (RST) segments 1483 * are validated by checking their SEQ-fields. A reset is 1484 * valid if its sequence number is in the window. 1485 * Note: this does not take into account delayed ACKs, so 1486 * we should test against last_ack_sent instead of rcv_nxt. 1487 * The sequence number in the reset segment is normally an 1488 * echo of our outgoing acknowlegement numbers, but some hosts 1489 * send a reset with the sequence number at the rightmost edge 1490 * of our receive window, and we have to handle this case. 1491 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1492 * that brute force RST attacks are possible. To combat this, 1493 * we use a much stricter check while in the ESTABLISHED state, 1494 * only accepting RSTs where the sequence number is equal to 1495 * last_ack_sent. In all other states (the states in which a 1496 * RST is more likely), the more permissive check is used. 1497 * If we have multiple segments in flight, the intial reset 1498 * segment sequence numbers will be to the left of last_ack_sent, 1499 * but they will eventually catch up. 1500 * In any case, it never made sense to trim reset segments to 1501 * fit the receive window since RFC 1122 says: 1502 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1503 * 1504 * A TCP SHOULD allow a received RST segment to include data. 1505 * 1506 * DISCUSSION 1507 * It has been suggested that a RST segment could contain 1508 * ASCII text that encoded and explained the cause of the 1509 * RST. No standard has yet been established for such 1510 * data. 1511 * 1512 * If the reset segment passes the sequence number test examine 1513 * the state: 1514 * SYN_RECEIVED STATE: 1515 * If passive open, return to LISTEN state. 1516 * If active open, inform user that connection was refused. 1517 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1518 * Inform user that connection was reset, and close tcb. 1519 * CLOSING, LAST_ACK STATES: 1520 * Close the tcb. 1521 * TIME_WAIT STATE: 1522 * Drop the segment - see Stevens, vol. 2, p. 964 and 1523 * RFC 1337. 1524 */ 1525 if (thflags & TH_RST) { 1526 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1527 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1528 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1529 switch (tp->t_state) { 1530 1531 case TCPS_SYN_RECEIVED: 1532 so->so_error = ECONNREFUSED; 1533 goto close; 1534 1535 case TCPS_ESTABLISHED: 1536 if (tp->last_ack_sent != th->th_seq && 1537 tcp_insecure_rst == 0) { 1538 tcpstat.tcps_badrst++; 1539 goto drop; 1540 } 1541 case TCPS_FIN_WAIT_1: 1542 case TCPS_FIN_WAIT_2: 1543 case TCPS_CLOSE_WAIT: 1544 so->so_error = ECONNRESET; 1545 close: 1546 tp->t_state = TCPS_CLOSED; 1547 tcpstat.tcps_drops++; 1548 KASSERT(headlocked, ("tcp_input: " 1549 "trimthenstep6: tcp_close: head not " 1550 "locked")); 1551 tp = tcp_close(tp); 1552 break; 1553 1554 case TCPS_CLOSING: 1555 case TCPS_LAST_ACK: 1556 KASSERT(headlocked, ("trimthenstep6: " 1557 "tcp_close.2: head not locked")); 1558 tp = tcp_close(tp); 1559 break; 1560 1561 case TCPS_TIME_WAIT: 1562 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1563 ("timewait")); 1564 break; 1565 } 1566 } 1567 goto drop; 1568 } 1569 1570 /* 1571 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1572 * and it's less than ts_recent, drop it. 1573 */ 1574 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1575 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1576 1577 /* Check to see if ts_recent is over 24 days old. */ 1578 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1579 /* 1580 * Invalidate ts_recent. If this segment updates 1581 * ts_recent, the age will be reset later and ts_recent 1582 * will get a valid value. If it does not, setting 1583 * ts_recent to zero will at least satisfy the 1584 * requirement that zero be placed in the timestamp 1585 * echo reply when ts_recent isn't valid. The 1586 * age isn't reset until we get a valid ts_recent 1587 * because we don't want out-of-order segments to be 1588 * dropped when ts_recent is old. 1589 */ 1590 tp->ts_recent = 0; 1591 } else { 1592 tcpstat.tcps_rcvduppack++; 1593 tcpstat.tcps_rcvdupbyte += tlen; 1594 tcpstat.tcps_pawsdrop++; 1595 if (tlen) 1596 goto dropafterack; 1597 goto drop; 1598 } 1599 } 1600 1601 /* 1602 * In the SYN-RECEIVED state, validate that the packet belongs to 1603 * this connection before trimming the data to fit the receive 1604 * window. Check the sequence number versus IRS since we know 1605 * the sequence numbers haven't wrapped. This is a partial fix 1606 * for the "LAND" DoS attack. 1607 */ 1608 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1609 rstreason = BANDLIM_RST_OPENPORT; 1610 goto dropwithreset; 1611 } 1612 1613 todrop = tp->rcv_nxt - th->th_seq; 1614 if (todrop > 0) { 1615 if (thflags & TH_SYN) { 1616 thflags &= ~TH_SYN; 1617 th->th_seq++; 1618 if (th->th_urp > 1) 1619 th->th_urp--; 1620 else 1621 thflags &= ~TH_URG; 1622 todrop--; 1623 } 1624 /* 1625 * Following if statement from Stevens, vol. 2, p. 960. 1626 */ 1627 if (todrop > tlen 1628 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1629 /* 1630 * Any valid FIN must be to the left of the window. 1631 * At this point the FIN must be a duplicate or out 1632 * of sequence; drop it. 1633 */ 1634 thflags &= ~TH_FIN; 1635 1636 /* 1637 * Send an ACK to resynchronize and drop any data. 1638 * But keep on processing for RST or ACK. 1639 */ 1640 tp->t_flags |= TF_ACKNOW; 1641 todrop = tlen; 1642 tcpstat.tcps_rcvduppack++; 1643 tcpstat.tcps_rcvdupbyte += todrop; 1644 } else { 1645 tcpstat.tcps_rcvpartduppack++; 1646 tcpstat.tcps_rcvpartdupbyte += todrop; 1647 } 1648 drop_hdrlen += todrop; /* drop from the top afterwards */ 1649 th->th_seq += todrop; 1650 tlen -= todrop; 1651 if (th->th_urp > todrop) 1652 th->th_urp -= todrop; 1653 else { 1654 thflags &= ~TH_URG; 1655 th->th_urp = 0; 1656 } 1657 } 1658 1659 /* 1660 * If new data are received on a connection after the 1661 * user processes are gone, then RST the other end. 1662 */ 1663 if ((so->so_state & SS_NOFDREF) && 1664 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1665 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not " 1666 "locked")); 1667 tp = tcp_close(tp); 1668 tcpstat.tcps_rcvafterclose++; 1669 rstreason = BANDLIM_UNLIMITED; 1670 goto dropwithreset; 1671 } 1672 1673 /* 1674 * If segment ends after window, drop trailing data 1675 * (and PUSH and FIN); if nothing left, just ACK. 1676 */ 1677 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1678 if (todrop > 0) { 1679 tcpstat.tcps_rcvpackafterwin++; 1680 if (todrop >= tlen) { 1681 tcpstat.tcps_rcvbyteafterwin += tlen; 1682 /* 1683 * If a new connection request is received 1684 * while in TIME_WAIT, drop the old connection 1685 * and start over if the sequence numbers 1686 * are above the previous ones. 1687 */ 1688 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1689 if (thflags & TH_SYN && 1690 tp->t_state == TCPS_TIME_WAIT && 1691 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1692 KASSERT(headlocked, ("trimthenstep6: " 1693 "tcp_close.4: head not locked")); 1694 tp = tcp_close(tp); 1695 goto findpcb; 1696 } 1697 /* 1698 * If window is closed can only take segments at 1699 * window edge, and have to drop data and PUSH from 1700 * incoming segments. Continue processing, but 1701 * remember to ack. Otherwise, drop segment 1702 * and ack. 1703 */ 1704 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1705 tp->t_flags |= TF_ACKNOW; 1706 tcpstat.tcps_rcvwinprobe++; 1707 } else 1708 goto dropafterack; 1709 } else 1710 tcpstat.tcps_rcvbyteafterwin += todrop; 1711 m_adj(m, -todrop); 1712 tlen -= todrop; 1713 thflags &= ~(TH_PUSH|TH_FIN); 1714 } 1715 1716 /* 1717 * If last ACK falls within this segment's sequence numbers, 1718 * record its timestamp. 1719 * NOTE: 1720 * 1) That the test incorporates suggestions from the latest 1721 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1722 * 2) That updating only on newer timestamps interferes with 1723 * our earlier PAWS tests, so this check should be solely 1724 * predicated on the sequence space of this segment. 1725 * 3) That we modify the segment boundary check to be 1726 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1727 * instead of RFC1323's 1728 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1729 * This modified check allows us to overcome RFC1323's 1730 * limitations as described in Stevens TCP/IP Illustrated 1731 * Vol. 2 p.869. In such cases, we can still calculate the 1732 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1733 */ 1734 if ((to.to_flags & TOF_TS) != 0 && 1735 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1736 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1737 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1738 tp->ts_recent_age = ticks; 1739 tp->ts_recent = to.to_tsval; 1740 } 1741 1742 /* 1743 * If a SYN is in the window, then this is an 1744 * error and we send an RST and drop the connection. 1745 */ 1746 if (thflags & TH_SYN) { 1747 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: " 1748 "head not locked")); 1749 tp = tcp_drop(tp, ECONNRESET); 1750 rstreason = BANDLIM_UNLIMITED; 1751 goto drop; 1752 } 1753 1754 /* 1755 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1756 * flag is on (half-synchronized state), then queue data for 1757 * later processing; else drop segment and return. 1758 */ 1759 if ((thflags & TH_ACK) == 0) { 1760 if (tp->t_state == TCPS_SYN_RECEIVED || 1761 (tp->t_flags & TF_NEEDSYN)) 1762 goto step6; 1763 else 1764 goto drop; 1765 } 1766 1767 /* 1768 * Ack processing. 1769 */ 1770 switch (tp->t_state) { 1771 1772 /* 1773 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1774 * ESTABLISHED state and continue processing. 1775 * The ACK was checked above. 1776 */ 1777 case TCPS_SYN_RECEIVED: 1778 1779 tcpstat.tcps_connects++; 1780 soisconnected(so); 1781 /* Do window scaling? */ 1782 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1783 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1784 tp->snd_scale = tp->requested_s_scale; 1785 tp->rcv_scale = tp->request_r_scale; 1786 } 1787 /* 1788 * Make transitions: 1789 * SYN-RECEIVED -> ESTABLISHED 1790 * SYN-RECEIVED* -> FIN-WAIT-1 1791 */ 1792 tp->t_starttime = ticks; 1793 if (tp->t_flags & TF_NEEDFIN) { 1794 tp->t_state = TCPS_FIN_WAIT_1; 1795 tp->t_flags &= ~TF_NEEDFIN; 1796 } else { 1797 tp->t_state = TCPS_ESTABLISHED; 1798 callout_reset(tp->tt_keep, tcp_keepidle, 1799 tcp_timer_keep, tp); 1800 } 1801 /* 1802 * If segment contains data or ACK, will call tcp_reass() 1803 * later; if not, do so now to pass queued data to user. 1804 */ 1805 if (tlen == 0 && (thflags & TH_FIN) == 0) 1806 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1807 (struct mbuf *)0); 1808 tp->snd_wl1 = th->th_seq - 1; 1809 /* FALLTHROUGH */ 1810 1811 /* 1812 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1813 * ACKs. If the ack is in the range 1814 * tp->snd_una < th->th_ack <= tp->snd_max 1815 * then advance tp->snd_una to th->th_ack and drop 1816 * data from the retransmission queue. If this ACK reflects 1817 * more up to date window information we update our window information. 1818 */ 1819 case TCPS_ESTABLISHED: 1820 case TCPS_FIN_WAIT_1: 1821 case TCPS_FIN_WAIT_2: 1822 case TCPS_CLOSE_WAIT: 1823 case TCPS_CLOSING: 1824 case TCPS_LAST_ACK: 1825 case TCPS_TIME_WAIT: 1826 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1827 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1828 if (tlen == 0 && tiwin == tp->snd_wnd) { 1829 tcpstat.tcps_rcvdupack++; 1830 /* 1831 * If we have outstanding data (other than 1832 * a window probe), this is a completely 1833 * duplicate ack (ie, window info didn't 1834 * change), the ack is the biggest we've 1835 * seen and we've seen exactly our rexmt 1836 * threshhold of them, assume a packet 1837 * has been dropped and retransmit it. 1838 * Kludge snd_nxt & the congestion 1839 * window so we send only this one 1840 * packet. 1841 * 1842 * We know we're losing at the current 1843 * window size so do congestion avoidance 1844 * (set ssthresh to half the current window 1845 * and pull our congestion window back to 1846 * the new ssthresh). 1847 * 1848 * Dup acks mean that packets have left the 1849 * network (they're now cached at the receiver) 1850 * so bump cwnd by the amount in the receiver 1851 * to keep a constant cwnd packets in the 1852 * network. 1853 */ 1854 if (!callout_active(tp->tt_rexmt) || 1855 th->th_ack != tp->snd_una) 1856 tp->t_dupacks = 0; 1857 else if (++tp->t_dupacks > tcprexmtthresh || 1858 ((tcp_do_newreno || tp->sack_enable) && 1859 IN_FASTRECOVERY(tp))) { 1860 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1861 int awnd; 1862 1863 /* 1864 * Compute the amount of data in flight first. 1865 * We can inject new data into the pipe iff 1866 * we have less than 1/2 the original window's 1867 * worth of data in flight. 1868 */ 1869 awnd = (tp->snd_nxt - tp->snd_fack) + 1870 tp->sackhint.sack_bytes_rexmit; 1871 if (awnd < tp->snd_ssthresh) { 1872 tp->snd_cwnd += tp->t_maxseg; 1873 if (tp->snd_cwnd > tp->snd_ssthresh) 1874 tp->snd_cwnd = tp->snd_ssthresh; 1875 } 1876 } else 1877 tp->snd_cwnd += tp->t_maxseg; 1878 (void) tcp_output(tp); 1879 goto drop; 1880 } else if (tp->t_dupacks == tcprexmtthresh) { 1881 tcp_seq onxt = tp->snd_nxt; 1882 u_int win; 1883 1884 /* 1885 * If we're doing sack, check to 1886 * see if we're already in sack 1887 * recovery. If we're not doing sack, 1888 * check to see if we're in newreno 1889 * recovery. 1890 */ 1891 if (tp->sack_enable) { 1892 if (IN_FASTRECOVERY(tp)) { 1893 tp->t_dupacks = 0; 1894 break; 1895 } 1896 } else if (tcp_do_newreno) { 1897 if (SEQ_LEQ(th->th_ack, 1898 tp->snd_recover)) { 1899 tp->t_dupacks = 0; 1900 break; 1901 } 1902 } 1903 win = min(tp->snd_wnd, tp->snd_cwnd) / 1904 2 / tp->t_maxseg; 1905 if (win < 2) 1906 win = 2; 1907 tp->snd_ssthresh = win * tp->t_maxseg; 1908 ENTER_FASTRECOVERY(tp); 1909 tp->snd_recover = tp->snd_max; 1910 callout_stop(tp->tt_rexmt); 1911 tp->t_rtttime = 0; 1912 if (tp->sack_enable) { 1913 KASSERT(tp->sackhint. 1914 sack_bytes_rexmit == 0, 1915 ("sackhint rexmit == 0")); 1916 tcpstat.tcps_sack_recovery_episode++; 1917 tp->sack_newdata = tp->snd_nxt; 1918 tp->snd_cwnd = tp->t_maxseg; 1919 (void) tcp_output(tp); 1920 goto drop; 1921 } 1922 tp->snd_nxt = th->th_ack; 1923 tp->snd_cwnd = tp->t_maxseg; 1924 (void) tcp_output(tp); 1925 KASSERT(tp->snd_limited <= 2, 1926 ("tp->snd_limited too big")); 1927 tp->snd_cwnd = tp->snd_ssthresh + 1928 tp->t_maxseg * 1929 (tp->t_dupacks - tp->snd_limited); 1930 if (SEQ_GT(onxt, tp->snd_nxt)) 1931 tp->snd_nxt = onxt; 1932 goto drop; 1933 } else if (tcp_do_rfc3042) { 1934 u_long oldcwnd = tp->snd_cwnd; 1935 tcp_seq oldsndmax = tp->snd_max; 1936 u_int sent; 1937 1938 KASSERT(tp->t_dupacks == 1 || 1939 tp->t_dupacks == 2, 1940 ("dupacks not 1 or 2")); 1941 if (tp->t_dupacks == 1) 1942 tp->snd_limited = 0; 1943 tp->snd_cwnd = 1944 (tp->snd_nxt - tp->snd_una) + 1945 (tp->t_dupacks - tp->snd_limited) * 1946 tp->t_maxseg; 1947 (void) tcp_output(tp); 1948 sent = tp->snd_max - oldsndmax; 1949 if (sent > tp->t_maxseg) { 1950 KASSERT((tp->t_dupacks == 2 && 1951 tp->snd_limited == 0) || 1952 (sent == tp->t_maxseg + 1 && 1953 tp->t_flags & TF_SENTFIN), 1954 ("sent too much")); 1955 tp->snd_limited = 2; 1956 } else if (sent > 0) 1957 ++tp->snd_limited; 1958 tp->snd_cwnd = oldcwnd; 1959 goto drop; 1960 } 1961 } else 1962 tp->t_dupacks = 0; 1963 break; 1964 } 1965 1966 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1967 1968 /* 1969 * If the congestion window was inflated to account 1970 * for the other side's cached packets, retract it. 1971 */ 1972 if (tcp_do_newreno || tp->sack_enable) { 1973 if (IN_FASTRECOVERY(tp)) { 1974 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1975 if (tp->sack_enable) 1976 tcp_sack_partialack(tp, th); 1977 else 1978 tcp_newreno_partial_ack(tp, th); 1979 } else { 1980 /* 1981 * Out of fast recovery. 1982 * Window inflation should have left us 1983 * with approximately snd_ssthresh 1984 * outstanding data. 1985 * But in case we would be inclined to 1986 * send a burst, better to do it via 1987 * the slow start mechanism. 1988 */ 1989 if (SEQ_GT(th->th_ack + 1990 tp->snd_ssthresh, 1991 tp->snd_max)) 1992 tp->snd_cwnd = tp->snd_max - 1993 th->th_ack + 1994 tp->t_maxseg; 1995 else 1996 tp->snd_cwnd = tp->snd_ssthresh; 1997 } 1998 } 1999 } else { 2000 if (tp->t_dupacks >= tcprexmtthresh && 2001 tp->snd_cwnd > tp->snd_ssthresh) 2002 tp->snd_cwnd = tp->snd_ssthresh; 2003 } 2004 tp->t_dupacks = 0; 2005 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2006 tcpstat.tcps_rcvacktoomuch++; 2007 goto dropafterack; 2008 } 2009 /* 2010 * If we reach this point, ACK is not a duplicate, 2011 * i.e., it ACKs something we sent. 2012 */ 2013 if (tp->t_flags & TF_NEEDSYN) { 2014 /* 2015 * T/TCP: Connection was half-synchronized, and our 2016 * SYN has been ACK'd (so connection is now fully 2017 * synchronized). Go to non-starred state, 2018 * increment snd_una for ACK of SYN, and check if 2019 * we can do window scaling. 2020 */ 2021 tp->t_flags &= ~TF_NEEDSYN; 2022 tp->snd_una++; 2023 /* Do window scaling? */ 2024 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2025 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2026 tp->snd_scale = tp->requested_s_scale; 2027 tp->rcv_scale = tp->request_r_scale; 2028 } 2029 } 2030 2031process_ACK: 2032 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 2033 "locked")); 2034 INP_LOCK_ASSERT(inp); 2035 2036 acked = th->th_ack - tp->snd_una; 2037 tcpstat.tcps_rcvackpack++; 2038 tcpstat.tcps_rcvackbyte += acked; 2039 2040 /* 2041 * If we just performed our first retransmit, and the ACK 2042 * arrives within our recovery window, then it was a mistake 2043 * to do the retransmit in the first place. Recover our 2044 * original cwnd and ssthresh, and proceed to transmit where 2045 * we left off. 2046 */ 2047 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2048 ++tcpstat.tcps_sndrexmitbad; 2049 tp->snd_cwnd = tp->snd_cwnd_prev; 2050 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2051 tp->snd_recover = tp->snd_recover_prev; 2052 if (tp->t_flags & TF_WASFRECOVERY) 2053 ENTER_FASTRECOVERY(tp); 2054 tp->snd_nxt = tp->snd_max; 2055 tp->t_badrxtwin = 0; /* XXX probably not required */ 2056 } 2057 2058 /* 2059 * If we have a timestamp reply, update smoothed 2060 * round trip time. If no timestamp is present but 2061 * transmit timer is running and timed sequence 2062 * number was acked, update smoothed round trip time. 2063 * Since we now have an rtt measurement, cancel the 2064 * timer backoff (cf., Phil Karn's retransmit alg.). 2065 * Recompute the initial retransmit timer. 2066 * 2067 * Some boxes send broken timestamp replies 2068 * during the SYN+ACK phase, ignore 2069 * timestamps of 0 or we could calculate a 2070 * huge RTT and blow up the retransmit timer. 2071 */ 2072 if ((to.to_flags & TOF_TS) != 0 && 2073 to.to_tsecr) { 2074 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2075 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2076 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2077 } 2078 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2079 2080 /* 2081 * If all outstanding data is acked, stop retransmit 2082 * timer and remember to restart (more output or persist). 2083 * If there is more data to be acked, restart retransmit 2084 * timer, using current (possibly backed-off) value. 2085 */ 2086 if (th->th_ack == tp->snd_max) { 2087 callout_stop(tp->tt_rexmt); 2088 needoutput = 1; 2089 } else if (!callout_active(tp->tt_persist)) 2090 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2091 tcp_timer_rexmt, tp); 2092 2093 /* 2094 * If no data (only SYN) was ACK'd, 2095 * skip rest of ACK processing. 2096 */ 2097 if (acked == 0) 2098 goto step6; 2099 2100 /* 2101 * When new data is acked, open the congestion window. 2102 * If the window gives us less than ssthresh packets 2103 * in flight, open exponentially (maxseg per packet). 2104 * Otherwise open linearly: maxseg per window 2105 * (maxseg^2 / cwnd per packet). 2106 */ 2107 if ((!tcp_do_newreno && !tp->sack_enable) || 2108 !IN_FASTRECOVERY(tp)) { 2109 register u_int cw = tp->snd_cwnd; 2110 register u_int incr = tp->t_maxseg; 2111 if (cw > tp->snd_ssthresh) 2112 incr = incr * incr / cw; 2113 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2114 } 2115 SOCKBUF_LOCK(&so->so_snd); 2116 if (acked > so->so_snd.sb_cc) { 2117 tp->snd_wnd -= so->so_snd.sb_cc; 2118 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2119 ourfinisacked = 1; 2120 } else { 2121 sbdrop_locked(&so->so_snd, acked); 2122 tp->snd_wnd -= acked; 2123 ourfinisacked = 0; 2124 } 2125 sowwakeup_locked(so); 2126 /* detect una wraparound */ 2127 if ((tcp_do_newreno || tp->sack_enable) && 2128 !IN_FASTRECOVERY(tp) && 2129 SEQ_GT(tp->snd_una, tp->snd_recover) && 2130 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2131 tp->snd_recover = th->th_ack - 1; 2132 if ((tcp_do_newreno || tp->sack_enable) && 2133 IN_FASTRECOVERY(tp) && 2134 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2135 EXIT_FASTRECOVERY(tp); 2136 tp->snd_una = th->th_ack; 2137 if (tp->sack_enable) { 2138 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2139 tp->snd_recover = tp->snd_una; 2140 } 2141 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2142 tp->snd_nxt = tp->snd_una; 2143 2144 switch (tp->t_state) { 2145 2146 /* 2147 * In FIN_WAIT_1 STATE in addition to the processing 2148 * for the ESTABLISHED state if our FIN is now acknowledged 2149 * then enter FIN_WAIT_2. 2150 */ 2151 case TCPS_FIN_WAIT_1: 2152 if (ourfinisacked) { 2153 /* 2154 * If we can't receive any more 2155 * data, then closing user can proceed. 2156 * Starting the timer is contrary to the 2157 * specification, but if we don't get a FIN 2158 * we'll hang forever. 2159 */ 2160 /* XXXjl 2161 * we should release the tp also, and use a 2162 * compressed state. 2163 */ 2164 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2165 soisdisconnected(so); 2166 callout_reset(tp->tt_2msl, tcp_maxidle, 2167 tcp_timer_2msl, tp); 2168 } 2169 tp->t_state = TCPS_FIN_WAIT_2; 2170 } 2171 break; 2172 2173 /* 2174 * In CLOSING STATE in addition to the processing for 2175 * the ESTABLISHED state if the ACK acknowledges our FIN 2176 * then enter the TIME-WAIT state, otherwise ignore 2177 * the segment. 2178 */ 2179 case TCPS_CLOSING: 2180 if (ourfinisacked) { 2181 KASSERT(headlocked, ("tcp_input: process_ACK: " 2182 "head not locked")); 2183 tcp_twstart(tp); 2184 INP_INFO_WUNLOCK(&tcbinfo); 2185 m_freem(m); 2186 return; 2187 } 2188 break; 2189 2190 /* 2191 * In LAST_ACK, we may still be waiting for data to drain 2192 * and/or to be acked, as well as for the ack of our FIN. 2193 * If our FIN is now acknowledged, delete the TCB, 2194 * enter the closed state and return. 2195 */ 2196 case TCPS_LAST_ACK: 2197 if (ourfinisacked) { 2198 KASSERT(headlocked, ("tcp_input: process_ACK:" 2199 " tcp_close: head not locked")); 2200 tp = tcp_close(tp); 2201 goto drop; 2202 } 2203 break; 2204 2205 /* 2206 * In TIME_WAIT state the only thing that should arrive 2207 * is a retransmission of the remote FIN. Acknowledge 2208 * it and restart the finack timer. 2209 */ 2210 case TCPS_TIME_WAIT: 2211 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2212 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2213 tcp_timer_2msl, tp); 2214 goto dropafterack; 2215 } 2216 } 2217 2218step6: 2219 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2220 INP_LOCK_ASSERT(inp); 2221 2222 /* 2223 * Update window information. 2224 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2225 */ 2226 if ((thflags & TH_ACK) && 2227 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2228 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2229 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2230 /* keep track of pure window updates */ 2231 if (tlen == 0 && 2232 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2233 tcpstat.tcps_rcvwinupd++; 2234 tp->snd_wnd = tiwin; 2235 tp->snd_wl1 = th->th_seq; 2236 tp->snd_wl2 = th->th_ack; 2237 if (tp->snd_wnd > tp->max_sndwnd) 2238 tp->max_sndwnd = tp->snd_wnd; 2239 needoutput = 1; 2240 } 2241 2242 /* 2243 * Process segments with URG. 2244 */ 2245 if ((thflags & TH_URG) && th->th_urp && 2246 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2247 /* 2248 * This is a kludge, but if we receive and accept 2249 * random urgent pointers, we'll crash in 2250 * soreceive. It's hard to imagine someone 2251 * actually wanting to send this much urgent data. 2252 */ 2253 SOCKBUF_LOCK(&so->so_rcv); 2254 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2255 th->th_urp = 0; /* XXX */ 2256 thflags &= ~TH_URG; /* XXX */ 2257 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2258 goto dodata; /* XXX */ 2259 } 2260 /* 2261 * If this segment advances the known urgent pointer, 2262 * then mark the data stream. This should not happen 2263 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2264 * a FIN has been received from the remote side. 2265 * In these states we ignore the URG. 2266 * 2267 * According to RFC961 (Assigned Protocols), 2268 * the urgent pointer points to the last octet 2269 * of urgent data. We continue, however, 2270 * to consider it to indicate the first octet 2271 * of data past the urgent section as the original 2272 * spec states (in one of two places). 2273 */ 2274 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2275 tp->rcv_up = th->th_seq + th->th_urp; 2276 so->so_oobmark = so->so_rcv.sb_cc + 2277 (tp->rcv_up - tp->rcv_nxt) - 1; 2278 if (so->so_oobmark == 0) 2279 so->so_rcv.sb_state |= SBS_RCVATMARK; 2280 sohasoutofband(so); 2281 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2282 } 2283 SOCKBUF_UNLOCK(&so->so_rcv); 2284 /* 2285 * Remove out of band data so doesn't get presented to user. 2286 * This can happen independent of advancing the URG pointer, 2287 * but if two URG's are pending at once, some out-of-band 2288 * data may creep in... ick. 2289 */ 2290 if (th->th_urp <= (u_long)tlen && 2291 !(so->so_options & SO_OOBINLINE)) { 2292 /* hdr drop is delayed */ 2293 tcp_pulloutofband(so, th, m, drop_hdrlen); 2294 } 2295 } else { 2296 /* 2297 * If no out of band data is expected, 2298 * pull receive urgent pointer along 2299 * with the receive window. 2300 */ 2301 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2302 tp->rcv_up = tp->rcv_nxt; 2303 } 2304dodata: /* XXX */ 2305 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2306 INP_LOCK_ASSERT(inp); 2307 2308 /* 2309 * Process the segment text, merging it into the TCP sequencing queue, 2310 * and arranging for acknowledgment of receipt if necessary. 2311 * This process logically involves adjusting tp->rcv_wnd as data 2312 * is presented to the user (this happens in tcp_usrreq.c, 2313 * case PRU_RCVD). If a FIN has already been received on this 2314 * connection then we just ignore the text. 2315 */ 2316 if ((tlen || (thflags & TH_FIN)) && 2317 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2318 m_adj(m, drop_hdrlen); /* delayed header drop */ 2319 /* 2320 * Insert segment which includes th into TCP reassembly queue 2321 * with control block tp. Set thflags to whether reassembly now 2322 * includes a segment with FIN. This handles the common case 2323 * inline (segment is the next to be received on an established 2324 * connection, and the queue is empty), avoiding linkage into 2325 * and removal from the queue and repetition of various 2326 * conversions. 2327 * Set DELACK for segments received in order, but ack 2328 * immediately when segments are out of order (so 2329 * fast retransmit can work). 2330 */ 2331 if (th->th_seq == tp->rcv_nxt && 2332 LIST_EMPTY(&tp->t_segq) && 2333 TCPS_HAVEESTABLISHED(tp->t_state)) { 2334 if (DELAY_ACK(tp)) 2335 tp->t_flags |= TF_DELACK; 2336 else 2337 tp->t_flags |= TF_ACKNOW; 2338 tp->rcv_nxt += tlen; 2339 thflags = th->th_flags & TH_FIN; 2340 tcpstat.tcps_rcvpack++; 2341 tcpstat.tcps_rcvbyte += tlen; 2342 ND6_HINT(tp); 2343 SOCKBUF_LOCK(&so->so_rcv); 2344 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2345 m_freem(m); 2346 else 2347 sbappendstream_locked(&so->so_rcv, m); 2348 sorwakeup_locked(so); 2349 } else { 2350 thflags = tcp_reass(tp, th, &tlen, m); 2351 tp->t_flags |= TF_ACKNOW; 2352 } 2353 if (tlen > 0 && tp->sack_enable) 2354 tcp_update_sack_list(tp, th->th_seq, th->th_seq + tlen); 2355 /* 2356 * Note the amount of data that peer has sent into 2357 * our window, in order to estimate the sender's 2358 * buffer size. 2359 */ 2360 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2361 } else { 2362 m_freem(m); 2363 thflags &= ~TH_FIN; 2364 } 2365 2366 /* 2367 * If FIN is received ACK the FIN and let the user know 2368 * that the connection is closing. 2369 */ 2370 if (thflags & TH_FIN) { 2371 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2372 socantrcvmore(so); 2373 /* 2374 * If connection is half-synchronized 2375 * (ie NEEDSYN flag on) then delay ACK, 2376 * so it may be piggybacked when SYN is sent. 2377 * Otherwise, since we received a FIN then no 2378 * more input can be expected, send ACK now. 2379 */ 2380 if (tp->t_flags & TF_NEEDSYN) 2381 tp->t_flags |= TF_DELACK; 2382 else 2383 tp->t_flags |= TF_ACKNOW; 2384 tp->rcv_nxt++; 2385 } 2386 switch (tp->t_state) { 2387 2388 /* 2389 * In SYN_RECEIVED and ESTABLISHED STATES 2390 * enter the CLOSE_WAIT state. 2391 */ 2392 case TCPS_SYN_RECEIVED: 2393 tp->t_starttime = ticks; 2394 /*FALLTHROUGH*/ 2395 case TCPS_ESTABLISHED: 2396 tp->t_state = TCPS_CLOSE_WAIT; 2397 break; 2398 2399 /* 2400 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2401 * enter the CLOSING state. 2402 */ 2403 case TCPS_FIN_WAIT_1: 2404 tp->t_state = TCPS_CLOSING; 2405 break; 2406 2407 /* 2408 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2409 * starting the time-wait timer, turning off the other 2410 * standard timers. 2411 */ 2412 case TCPS_FIN_WAIT_2: 2413 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2414 "TCP_FIN_WAIT_2: head not locked")); 2415 tcp_twstart(tp); 2416 INP_INFO_WUNLOCK(&tcbinfo); 2417 return; 2418 2419 /* 2420 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2421 */ 2422 case TCPS_TIME_WAIT: 2423 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2424 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2425 tcp_timer_2msl, tp); 2426 break; 2427 } 2428 } 2429 INP_INFO_WUNLOCK(&tcbinfo); 2430 headlocked = 0; 2431#ifdef TCPDEBUG 2432 if (so->so_options & SO_DEBUG) 2433 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2434 &tcp_savetcp, 0); 2435#endif 2436 2437 /* 2438 * Return any desired output. 2439 */ 2440 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2441 (void) tcp_output(tp); 2442 2443check_delack: 2444 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2445 INP_LOCK_ASSERT(inp); 2446 if (tp->t_flags & TF_DELACK) { 2447 tp->t_flags &= ~TF_DELACK; 2448 callout_reset(tp->tt_delack, tcp_delacktime, 2449 tcp_timer_delack, tp); 2450 } 2451 INP_UNLOCK(inp); 2452 return; 2453 2454dropafterack: 2455 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2456 /* 2457 * Generate an ACK dropping incoming segment if it occupies 2458 * sequence space, where the ACK reflects our state. 2459 * 2460 * We can now skip the test for the RST flag since all 2461 * paths to this code happen after packets containing 2462 * RST have been dropped. 2463 * 2464 * In the SYN-RECEIVED state, don't send an ACK unless the 2465 * segment we received passes the SYN-RECEIVED ACK test. 2466 * If it fails send a RST. This breaks the loop in the 2467 * "LAND" DoS attack, and also prevents an ACK storm 2468 * between two listening ports that have been sent forged 2469 * SYN segments, each with the source address of the other. 2470 */ 2471 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2472 (SEQ_GT(tp->snd_una, th->th_ack) || 2473 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2474 rstreason = BANDLIM_RST_OPENPORT; 2475 goto dropwithreset; 2476 } 2477#ifdef TCPDEBUG 2478 if (so->so_options & SO_DEBUG) 2479 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2480 &tcp_savetcp, 0); 2481#endif 2482 KASSERT(headlocked, ("headlocked should be 1")); 2483 INP_INFO_WUNLOCK(&tcbinfo); 2484 tp->t_flags |= TF_ACKNOW; 2485 (void) tcp_output(tp); 2486 INP_UNLOCK(inp); 2487 m_freem(m); 2488 return; 2489 2490dropwithreset: 2491 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2492 /* 2493 * Generate a RST, dropping incoming segment. 2494 * Make ACK acceptable to originator of segment. 2495 * Don't bother to respond if destination was broadcast/multicast. 2496 */ 2497 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2498 goto drop; 2499 if (isipv6) { 2500 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2501 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2502 goto drop; 2503 } else { 2504 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2505 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2506 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2507 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2508 goto drop; 2509 } 2510 /* IPv6 anycast check is done at tcp6_input() */ 2511 2512 /* 2513 * Perform bandwidth limiting. 2514 */ 2515 if (badport_bandlim(rstreason) < 0) 2516 goto drop; 2517 2518#ifdef TCPDEBUG 2519 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2520 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2521 &tcp_savetcp, 0); 2522#endif 2523 2524 if (thflags & TH_ACK) 2525 /* mtod() below is safe as long as hdr dropping is delayed */ 2526 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2527 TH_RST); 2528 else { 2529 if (thflags & TH_SYN) 2530 tlen++; 2531 /* mtod() below is safe as long as hdr dropping is delayed */ 2532 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2533 (tcp_seq)0, TH_RST|TH_ACK); 2534 } 2535 2536 if (tp) 2537 INP_UNLOCK(inp); 2538 if (headlocked) 2539 INP_INFO_WUNLOCK(&tcbinfo); 2540 return; 2541 2542drop: 2543 /* 2544 * Drop space held by incoming segment and return. 2545 */ 2546#ifdef TCPDEBUG 2547 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2548 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2549 &tcp_savetcp, 0); 2550#endif 2551 if (tp) 2552 INP_UNLOCK(inp); 2553 if (headlocked) 2554 INP_INFO_WUNLOCK(&tcbinfo); 2555 m_freem(m); 2556 return; 2557} 2558 2559/* 2560 * Parse TCP options and place in tcpopt. 2561 */ 2562static void 2563tcp_dooptions(tp, to, cp, cnt, is_syn, th) 2564 struct tcpcb *tp; 2565 struct tcpopt *to; 2566 u_char *cp; 2567 int cnt; 2568 int is_syn; 2569 struct tcphdr *th; 2570{ 2571 int opt, optlen; 2572 2573 to->to_flags = 0; 2574 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2575 opt = cp[0]; 2576 if (opt == TCPOPT_EOL) 2577 break; 2578 if (opt == TCPOPT_NOP) 2579 optlen = 1; 2580 else { 2581 if (cnt < 2) 2582 break; 2583 optlen = cp[1]; 2584 if (optlen < 2 || optlen > cnt) 2585 break; 2586 } 2587 switch (opt) { 2588 case TCPOPT_MAXSEG: 2589 if (optlen != TCPOLEN_MAXSEG) 2590 continue; 2591 if (!is_syn) 2592 continue; 2593 to->to_flags |= TOF_MSS; 2594 bcopy((char *)cp + 2, 2595 (char *)&to->to_mss, sizeof(to->to_mss)); 2596 to->to_mss = ntohs(to->to_mss); 2597 break; 2598 case TCPOPT_WINDOW: 2599 if (optlen != TCPOLEN_WINDOW) 2600 continue; 2601 if (! is_syn) 2602 continue; 2603 to->to_flags |= TOF_SCALE; 2604 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2605 break; 2606 case TCPOPT_TIMESTAMP: 2607 if (optlen != TCPOLEN_TIMESTAMP) 2608 continue; 2609 to->to_flags |= TOF_TS; 2610 bcopy((char *)cp + 2, 2611 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2612 to->to_tsval = ntohl(to->to_tsval); 2613 bcopy((char *)cp + 6, 2614 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2615 to->to_tsecr = ntohl(to->to_tsecr); 2616 /* 2617 * If echoed timestamp is later than the current time, 2618 * fall back to non RFC1323 RTT calculation. 2619 */ 2620 if ((to->to_tsecr != 0) && TSTMP_GT(to->to_tsecr, ticks)) 2621 to->to_tsecr = 0; 2622 break; 2623#ifdef TCP_SIGNATURE 2624 /* 2625 * XXX In order to reply to a host which has set the 2626 * TCP_SIGNATURE option in its initial SYN, we have to 2627 * record the fact that the option was observed here 2628 * for the syncache code to perform the correct response. 2629 */ 2630 case TCPOPT_SIGNATURE: 2631 if (optlen != TCPOLEN_SIGNATURE) 2632 continue; 2633 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2634 break; 2635#endif 2636 case TCPOPT_SACK_PERMITTED: 2637 if (!tcp_do_sack || 2638 optlen != TCPOLEN_SACK_PERMITTED) 2639 continue; 2640 if (is_syn) { 2641 /* MUST only be set on SYN */ 2642 to->to_flags |= TOF_SACK; 2643 } 2644 break; 2645 2646 case TCPOPT_SACK: 2647 if (!tp || tcp_sack_option(tp, th, cp, optlen)) 2648 continue; 2649 break; 2650 default: 2651 continue; 2652 } 2653 } 2654} 2655 2656/* 2657 * Pull out of band byte out of a segment so 2658 * it doesn't appear in the user's data queue. 2659 * It is still reflected in the segment length for 2660 * sequencing purposes. 2661 */ 2662static void 2663tcp_pulloutofband(so, th, m, off) 2664 struct socket *so; 2665 struct tcphdr *th; 2666 register struct mbuf *m; 2667 int off; /* delayed to be droped hdrlen */ 2668{ 2669 int cnt = off + th->th_urp - 1; 2670 2671 while (cnt >= 0) { 2672 if (m->m_len > cnt) { 2673 char *cp = mtod(m, caddr_t) + cnt; 2674 struct tcpcb *tp = sototcpcb(so); 2675 2676 tp->t_iobc = *cp; 2677 tp->t_oobflags |= TCPOOB_HAVEDATA; 2678 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2679 m->m_len--; 2680 if (m->m_flags & M_PKTHDR) 2681 m->m_pkthdr.len--; 2682 return; 2683 } 2684 cnt -= m->m_len; 2685 m = m->m_next; 2686 if (m == 0) 2687 break; 2688 } 2689 panic("tcp_pulloutofband"); 2690} 2691 2692/* 2693 * Collect new round-trip time estimate 2694 * and update averages and current timeout. 2695 */ 2696static void 2697tcp_xmit_timer(tp, rtt) 2698 register struct tcpcb *tp; 2699 int rtt; 2700{ 2701 register int delta; 2702 2703 INP_LOCK_ASSERT(tp->t_inpcb); 2704 2705 tcpstat.tcps_rttupdated++; 2706 tp->t_rttupdated++; 2707 if (tp->t_srtt != 0) { 2708 /* 2709 * srtt is stored as fixed point with 5 bits after the 2710 * binary point (i.e., scaled by 8). The following magic 2711 * is equivalent to the smoothing algorithm in rfc793 with 2712 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2713 * point). Adjust rtt to origin 0. 2714 */ 2715 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2716 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2717 2718 if ((tp->t_srtt += delta) <= 0) 2719 tp->t_srtt = 1; 2720 2721 /* 2722 * We accumulate a smoothed rtt variance (actually, a 2723 * smoothed mean difference), then set the retransmit 2724 * timer to smoothed rtt + 4 times the smoothed variance. 2725 * rttvar is stored as fixed point with 4 bits after the 2726 * binary point (scaled by 16). The following is 2727 * equivalent to rfc793 smoothing with an alpha of .75 2728 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2729 * rfc793's wired-in beta. 2730 */ 2731 if (delta < 0) 2732 delta = -delta; 2733 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2734 if ((tp->t_rttvar += delta) <= 0) 2735 tp->t_rttvar = 1; 2736 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2737 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2738 } else { 2739 /* 2740 * No rtt measurement yet - use the unsmoothed rtt. 2741 * Set the variance to half the rtt (so our first 2742 * retransmit happens at 3*rtt). 2743 */ 2744 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2745 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2746 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2747 } 2748 tp->t_rtttime = 0; 2749 tp->t_rxtshift = 0; 2750 2751 /* 2752 * the retransmit should happen at rtt + 4 * rttvar. 2753 * Because of the way we do the smoothing, srtt and rttvar 2754 * will each average +1/2 tick of bias. When we compute 2755 * the retransmit timer, we want 1/2 tick of rounding and 2756 * 1 extra tick because of +-1/2 tick uncertainty in the 2757 * firing of the timer. The bias will give us exactly the 2758 * 1.5 tick we need. But, because the bias is 2759 * statistical, we have to test that we don't drop below 2760 * the minimum feasible timer (which is 2 ticks). 2761 */ 2762 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2763 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2764 2765 /* 2766 * We received an ack for a packet that wasn't retransmitted; 2767 * it is probably safe to discard any error indications we've 2768 * received recently. This isn't quite right, but close enough 2769 * for now (a route might have failed after we sent a segment, 2770 * and the return path might not be symmetrical). 2771 */ 2772 tp->t_softerror = 0; 2773} 2774 2775/* 2776 * Determine a reasonable value for maxseg size. 2777 * If the route is known, check route for mtu. 2778 * If none, use an mss that can be handled on the outgoing 2779 * interface without forcing IP to fragment; if bigger than 2780 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2781 * to utilize large mbufs. If no route is found, route has no mtu, 2782 * or the destination isn't local, use a default, hopefully conservative 2783 * size (usually 512 or the default IP max size, but no more than the mtu 2784 * of the interface), as we can't discover anything about intervening 2785 * gateways or networks. We also initialize the congestion/slow start 2786 * window to be a single segment if the destination isn't local. 2787 * While looking at the routing entry, we also initialize other path-dependent 2788 * parameters from pre-set or cached values in the routing entry. 2789 * 2790 * Also take into account the space needed for options that we 2791 * send regularly. Make maxseg shorter by that amount to assure 2792 * that we can send maxseg amount of data even when the options 2793 * are present. Store the upper limit of the length of options plus 2794 * data in maxopd. 2795 * 2796 * 2797 * In case of T/TCP, we call this routine during implicit connection 2798 * setup as well (offer = -1), to initialize maxseg from the cached 2799 * MSS of our peer. 2800 * 2801 * NOTE that this routine is only called when we process an incoming 2802 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2803 */ 2804void 2805tcp_mss(tp, offer) 2806 struct tcpcb *tp; 2807 int offer; 2808{ 2809 int rtt, mss; 2810 u_long bufsize; 2811 u_long maxmtu; 2812 struct inpcb *inp = tp->t_inpcb; 2813 struct socket *so; 2814 struct hc_metrics_lite metrics; 2815 int origoffer = offer; 2816#ifdef INET6 2817 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2818 size_t min_protoh = isipv6 ? 2819 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2820 sizeof (struct tcpiphdr); 2821#else 2822 const size_t min_protoh = sizeof(struct tcpiphdr); 2823#endif 2824 2825 /* initialize */ 2826#ifdef INET6 2827 if (isipv6) { 2828 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2829 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2830 } else 2831#endif 2832 { 2833 maxmtu = tcp_maxmtu(&inp->inp_inc); 2834 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2835 } 2836 so = inp->inp_socket; 2837 2838 /* 2839 * no route to sender, stay with default mss and return 2840 */ 2841 if (maxmtu == 0) 2842 return; 2843 2844 /* what have we got? */ 2845 switch (offer) { 2846 case 0: 2847 /* 2848 * Offer == 0 means that there was no MSS on the SYN 2849 * segment, in this case we use tcp_mssdflt. 2850 */ 2851 offer = 2852#ifdef INET6 2853 isipv6 ? tcp_v6mssdflt : 2854#endif 2855 tcp_mssdflt; 2856 break; 2857 2858 case -1: 2859 /* 2860 * Offer == -1 means that we didn't receive SYN yet. 2861 */ 2862 /* FALLTHROUGH */ 2863 2864 default: 2865 /* 2866 * Prevent DoS attack with too small MSS. Round up 2867 * to at least minmss. 2868 */ 2869 offer = max(offer, tcp_minmss); 2870 /* 2871 * Sanity check: make sure that maxopd will be large 2872 * enough to allow some data on segments even if the 2873 * all the option space is used (40bytes). Otherwise 2874 * funny things may happen in tcp_output. 2875 */ 2876 offer = max(offer, 64); 2877 } 2878 2879 /* 2880 * rmx information is now retrieved from tcp_hostcache 2881 */ 2882 tcp_hc_get(&inp->inp_inc, &metrics); 2883 2884 /* 2885 * if there's a discovered mtu int tcp hostcache, use it 2886 * else, use the link mtu. 2887 */ 2888 if (metrics.rmx_mtu) 2889 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2890 else { 2891#ifdef INET6 2892 if (isipv6) { 2893 mss = maxmtu - min_protoh; 2894 if (!path_mtu_discovery && 2895 !in6_localaddr(&inp->in6p_faddr)) 2896 mss = min(mss, tcp_v6mssdflt); 2897 } else 2898#endif 2899 { 2900 mss = maxmtu - min_protoh; 2901 if (!path_mtu_discovery && 2902 !in_localaddr(inp->inp_faddr)) 2903 mss = min(mss, tcp_mssdflt); 2904 } 2905 } 2906 mss = min(mss, offer); 2907 2908 /* 2909 * maxopd stores the maximum length of data AND options 2910 * in a segment; maxseg is the amount of data in a normal 2911 * segment. We need to store this value (maxopd) apart 2912 * from maxseg, because now every segment carries options 2913 * and thus we normally have somewhat less data in segments. 2914 */ 2915 tp->t_maxopd = mss; 2916 2917 /* 2918 * origoffer==-1 indicates, that no segments were received yet. 2919 * In this case we just guess. 2920 */ 2921 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2922 (origoffer == -1 || 2923 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2924 mss -= TCPOLEN_TSTAMP_APPA; 2925 tp->t_maxseg = mss; 2926 2927#if (MCLBYTES & (MCLBYTES - 1)) == 0 2928 if (mss > MCLBYTES) 2929 mss &= ~(MCLBYTES-1); 2930#else 2931 if (mss > MCLBYTES) 2932 mss = mss / MCLBYTES * MCLBYTES; 2933#endif 2934 tp->t_maxseg = mss; 2935 2936 /* 2937 * If there's a pipesize, change the socket buffer to that size, 2938 * don't change if sb_hiwat is different than default (then it 2939 * has been changed on purpose with setsockopt). 2940 * Make the socket buffers an integral number of mss units; 2941 * if the mss is larger than the socket buffer, decrease the mss. 2942 */ 2943 SOCKBUF_LOCK(&so->so_snd); 2944 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2945 bufsize = metrics.rmx_sendpipe; 2946 else 2947 bufsize = so->so_snd.sb_hiwat; 2948 if (bufsize < mss) 2949 mss = bufsize; 2950 else { 2951 bufsize = roundup(bufsize, mss); 2952 if (bufsize > sb_max) 2953 bufsize = sb_max; 2954 if (bufsize > so->so_snd.sb_hiwat) 2955 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2956 } 2957 SOCKBUF_UNLOCK(&so->so_snd); 2958 tp->t_maxseg = mss; 2959 2960 SOCKBUF_LOCK(&so->so_rcv); 2961 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2962 bufsize = metrics.rmx_recvpipe; 2963 else 2964 bufsize = so->so_rcv.sb_hiwat; 2965 if (bufsize > mss) { 2966 bufsize = roundup(bufsize, mss); 2967 if (bufsize > sb_max) 2968 bufsize = sb_max; 2969 if (bufsize > so->so_rcv.sb_hiwat) 2970 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2971 } 2972 SOCKBUF_UNLOCK(&so->so_rcv); 2973 /* 2974 * While we're here, check the others too 2975 */ 2976 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2977 tp->t_srtt = rtt; 2978 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2979 tcpstat.tcps_usedrtt++; 2980 if (metrics.rmx_rttvar) { 2981 tp->t_rttvar = metrics.rmx_rttvar; 2982 tcpstat.tcps_usedrttvar++; 2983 } else { 2984 /* default variation is +- 1 rtt */ 2985 tp->t_rttvar = 2986 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2987 } 2988 TCPT_RANGESET(tp->t_rxtcur, 2989 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2990 tp->t_rttmin, TCPTV_REXMTMAX); 2991 } 2992 if (metrics.rmx_ssthresh) { 2993 /* 2994 * There's some sort of gateway or interface 2995 * buffer limit on the path. Use this to set 2996 * the slow start threshhold, but set the 2997 * threshold to no less than 2*mss. 2998 */ 2999 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3000 tcpstat.tcps_usedssthresh++; 3001 } 3002 if (metrics.rmx_bandwidth) 3003 tp->snd_bandwidth = metrics.rmx_bandwidth; 3004 3005 /* 3006 * Set the slow-start flight size depending on whether this 3007 * is a local network or not. 3008 * 3009 * Extend this so we cache the cwnd too and retrieve it here. 3010 * Make cwnd even bigger than RFC3390 suggests but only if we 3011 * have previous experience with the remote host. Be careful 3012 * not make cwnd bigger than remote receive window or our own 3013 * send socket buffer. Maybe put some additional upper bound 3014 * on the retrieved cwnd. Should do incremental updates to 3015 * hostcache when cwnd collapses so next connection doesn't 3016 * overloads the path again. 3017 * 3018 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3019 * We currently check only in syncache_socket for that. 3020 */ 3021#define TCP_METRICS_CWND 3022#ifdef TCP_METRICS_CWND 3023 if (metrics.rmx_cwnd) 3024 tp->snd_cwnd = max(mss, 3025 min(metrics.rmx_cwnd / 2, 3026 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3027 else 3028#endif 3029 if (tcp_do_rfc3390) 3030 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3031#ifdef INET6 3032 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3033 (!isipv6 && in_localaddr(inp->inp_faddr))) 3034#else 3035 else if (in_localaddr(inp->inp_faddr)) 3036#endif 3037 tp->snd_cwnd = mss * ss_fltsz_local; 3038 else 3039 tp->snd_cwnd = mss * ss_fltsz; 3040} 3041 3042/* 3043 * Determine the MSS option to send on an outgoing SYN. 3044 */ 3045int 3046tcp_mssopt(inc) 3047 struct in_conninfo *inc; 3048{ 3049 int mss = 0; 3050 u_long maxmtu = 0; 3051 u_long thcmtu = 0; 3052 size_t min_protoh; 3053#ifdef INET6 3054 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3055#endif 3056 3057 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3058 3059#ifdef INET6 3060 if (isipv6) { 3061 mss = tcp_v6mssdflt; 3062 maxmtu = tcp_maxmtu6(inc); 3063 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3064 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3065 } else 3066#endif 3067 { 3068 mss = tcp_mssdflt; 3069 maxmtu = tcp_maxmtu(inc); 3070 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3071 min_protoh = sizeof(struct tcpiphdr); 3072 } 3073 if (maxmtu && thcmtu) 3074 mss = min(maxmtu, thcmtu) - min_protoh; 3075 else if (maxmtu || thcmtu) 3076 mss = max(maxmtu, thcmtu) - min_protoh; 3077 3078 return (mss); 3079} 3080 3081 3082/* 3083 * On a partial ack arrives, force the retransmission of the 3084 * next unacknowledged segment. Do not clear tp->t_dupacks. 3085 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3086 * be started again. 3087 */ 3088static void 3089tcp_newreno_partial_ack(tp, th) 3090 struct tcpcb *tp; 3091 struct tcphdr *th; 3092{ 3093 tcp_seq onxt = tp->snd_nxt; 3094 u_long ocwnd = tp->snd_cwnd; 3095 3096 callout_stop(tp->tt_rexmt); 3097 tp->t_rtttime = 0; 3098 tp->snd_nxt = th->th_ack; 3099 /* 3100 * Set snd_cwnd to one segment beyond acknowledged offset. 3101 * (tp->snd_una has not yet been updated when this function is called.) 3102 */ 3103 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3104 tp->t_flags |= TF_ACKNOW; 3105 (void) tcp_output(tp); 3106 tp->snd_cwnd = ocwnd; 3107 if (SEQ_GT(onxt, tp->snd_nxt)) 3108 tp->snd_nxt = onxt; 3109 /* 3110 * Partial window deflation. Relies on fact that tp->snd_una 3111 * not updated yet. 3112 */ 3113 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3114} 3115 3116/* 3117 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3118 * looking for a pcb in the listen state. Returns 0 otherwise. 3119 */ 3120static int 3121tcp_timewait(tw, to, th, m, tlen) 3122 struct tcptw *tw; 3123 struct tcpopt *to; 3124 struct tcphdr *th; 3125 struct mbuf *m; 3126 int tlen; 3127{ 3128 int thflags; 3129 tcp_seq seq; 3130#ifdef INET6 3131 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3132#else 3133 const int isipv6 = 0; 3134#endif 3135 3136 /* tcbinfo lock required for tcp_twclose(), tcp_2msl_reset. */ 3137 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3138 INP_LOCK_ASSERT(tw->tw_inpcb); 3139 3140 thflags = th->th_flags; 3141 3142 /* 3143 * NOTE: for FIN_WAIT_2 (to be added later), 3144 * must validate sequence number before accepting RST 3145 */ 3146 3147 /* 3148 * If the segment contains RST: 3149 * Drop the segment - see Stevens, vol. 2, p. 964 and 3150 * RFC 1337. 3151 */ 3152 if (thflags & TH_RST) 3153 goto drop; 3154 3155#if 0 3156/* PAWS not needed at the moment */ 3157 /* 3158 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3159 * and it's less than ts_recent, drop it. 3160 */ 3161 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3162 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3163 if ((thflags & TH_ACK) == 0) 3164 goto drop; 3165 goto ack; 3166 } 3167 /* 3168 * ts_recent is never updated because we never accept new segments. 3169 */ 3170#endif 3171 3172 /* 3173 * If a new connection request is received 3174 * while in TIME_WAIT, drop the old connection 3175 * and start over if the sequence numbers 3176 * are above the previous ones. 3177 */ 3178 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3179 (void) tcp_twclose(tw, 0); 3180 return (1); 3181 } 3182 3183 /* 3184 * Drop the the segment if it does not contain an ACK. 3185 */ 3186 if ((thflags & TH_ACK) == 0) 3187 goto drop; 3188 3189 /* 3190 * Reset the 2MSL timer if this is a duplicate FIN. 3191 */ 3192 if (thflags & TH_FIN) { 3193 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3194 if (seq + 1 == tw->rcv_nxt) 3195 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3196 } 3197 3198 /* 3199 * Acknowledge the segment if it has data or is not a duplicate ACK. 3200 */ 3201 if (thflags != TH_ACK || tlen != 0 || 3202 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3203 tcp_twrespond(tw, TH_ACK); 3204 goto drop; 3205 3206 /* 3207 * Generate a RST, dropping incoming segment. 3208 * Make ACK acceptable to originator of segment. 3209 * Don't bother to respond if destination was broadcast/multicast. 3210 */ 3211 if (m->m_flags & (M_BCAST|M_MCAST)) 3212 goto drop; 3213 if (isipv6) { 3214 struct ip6_hdr *ip6; 3215 3216 /* IPv6 anycast check is done at tcp6_input() */ 3217 ip6 = mtod(m, struct ip6_hdr *); 3218 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3219 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3220 goto drop; 3221 } else { 3222 struct ip *ip; 3223 3224 ip = mtod(m, struct ip *); 3225 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3226 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3227 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3228 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3229 goto drop; 3230 } 3231 if (thflags & TH_ACK) { 3232 tcp_respond(NULL, 3233 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3234 } else { 3235 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3236 tcp_respond(NULL, 3237 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3238 } 3239 INP_UNLOCK(tw->tw_inpcb); 3240 return (0); 3241 3242drop: 3243 INP_UNLOCK(tw->tw_inpcb); 3244 m_freem(m); 3245 return (0); 3246} 3247