tcp_input.c revision 166842
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 166842 2007-02-20 10:20:03Z rwatson $ 31 */ 32 33#include "opt_ipfw.h" /* for ipfw_fwd */ 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39#include "opt_tcp_input.h" 40#include "opt_tcp_sack.h" 41 42#include <sys/param.h> 43#include <sys/kernel.h> 44#include <sys/malloc.h> 45#include <sys/mbuf.h> 46#include <sys/proc.h> /* for proc0 declaration */ 47#include <sys/protosw.h> 48#include <sys/signalvar.h> 49#include <sys/socket.h> 50#include <sys/socketvar.h> 51#include <sys/sysctl.h> 52#include <sys/syslog.h> 53#include <sys/systm.h> 54 55#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57#include <vm/uma.h> 58 59#include <net/if.h> 60#include <net/route.h> 61 62#include <netinet/in.h> 63#include <netinet/in_pcb.h> 64#include <netinet/in_systm.h> 65#include <netinet/in_var.h> 66#include <netinet/ip.h> 67#include <netinet/ip_icmp.h> /* required for icmp_var.h */ 68#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 69#include <netinet/ip_var.h> 70#include <netinet/ip_options.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> 99 100#include <security/mac/mac_framework.h> 101 102static const int tcprexmtthresh = 3; 103 104struct tcpstat tcpstat; 105SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 106 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 107 108static int tcp_log_in_vain = 0; 109SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 110 &tcp_log_in_vain, 0, "Log all incoming TCP connections"); 111 112static int blackhole = 0; 113SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 114 &blackhole, 0, "Do not send RST when dropping refused connections"); 115 116int tcp_delack_enabled = 1; 117SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 118 &tcp_delack_enabled, 0, 119 "Delay ACK to try and piggyback it onto a data packet"); 120 121#ifdef TCP_DROP_SYNFIN 122static int drop_synfin = 0; 123SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 124 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 125#endif 126 127static int tcp_do_rfc3042 = 1; 128SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 129 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 130 131static int tcp_do_rfc3390 = 1; 132SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 133 &tcp_do_rfc3390, 0, 134 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 135 136static int tcp_insecure_rst = 0; 137SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 138 &tcp_insecure_rst, 0, 139 "Follow the old (insecure) criteria for accepting RST packets."); 140 141SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 142 "TCP Segment Reassembly Queue"); 143 144static int tcp_reass_maxseg = 0; 145SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 146 &tcp_reass_maxseg, 0, 147 "Global maximum number of TCP Segments in Reassembly Queue"); 148 149int tcp_reass_qsize = 0; 150SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 151 &tcp_reass_qsize, 0, 152 "Global number of TCP Segments currently in Reassembly Queue"); 153 154static int tcp_reass_maxqlen = 48; 155SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 156 &tcp_reass_maxqlen, 0, 157 "Maximum number of TCP Segments per individual Reassembly Queue"); 158 159static int tcp_reass_overflows = 0; 160SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 161 &tcp_reass_overflows, 0, 162 "Global number of TCP Segment Reassembly Queue Overflows"); 163 164int tcp_do_autorcvbuf = 1; 165SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 166 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing"); 167 168int tcp_autorcvbuf_inc = 16*1024; 169SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 170 &tcp_autorcvbuf_inc, 0, "Incrementor step size of automatic receive buffer"); 171 172int tcp_autorcvbuf_max = 256*1024; 173SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 174 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer"); 175 176struct inpcbhead tcb; 177#define tcb6 tcb /* for KAME src sync over BSD*'s */ 178struct inpcbinfo tcbinfo; 179struct mtx *tcbinfo_mtx; 180 181static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 182 183static void tcp_pulloutofband(struct socket *, 184 struct tcphdr *, struct mbuf *, int); 185static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 186 struct mbuf *); 187static void tcp_xmit_timer(struct tcpcb *, int); 188static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 189static int tcp_timewait(struct inpcb *, struct tcpopt *, 190 struct tcphdr *, struct mbuf *, int); 191 192/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 193#ifdef INET6 194#define ND6_HINT(tp) \ 195do { \ 196 if ((tp) && (tp)->t_inpcb && \ 197 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 198 nd6_nud_hint(NULL, NULL, 0); \ 199} while (0) 200#else 201#define ND6_HINT(tp) 202#endif 203 204/* 205 * Indicate whether this ack should be delayed. We can delay the ack if 206 * - there is no delayed ack timer in progress and 207 * - our last ack wasn't a 0-sized window. We never want to delay 208 * the ack that opens up a 0-sized window and 209 * - delayed acks are enabled or 210 * - this is a half-synchronized T/TCP connection. 211 */ 212#define DELAY_ACK(tp) \ 213 ((!callout_active(tp->tt_delack) && \ 214 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 215 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 216 217/* Initialize TCP reassembly queue */ 218static void 219tcp_reass_zone_change(void *tag) 220{ 221 222 tcp_reass_maxseg = nmbclusters / 16; 223 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 224} 225 226uma_zone_t tcp_reass_zone; 227void 228tcp_reass_init() 229{ 230 tcp_reass_maxseg = nmbclusters / 16; 231 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 232 &tcp_reass_maxseg); 233 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 234 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 235 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 236 EVENTHANDLER_REGISTER(nmbclusters_change, 237 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY); 238} 239 240static int 241tcp_reass(tp, th, tlenp, m) 242 register struct tcpcb *tp; 243 register struct tcphdr *th; 244 int *tlenp; 245 struct mbuf *m; 246{ 247 struct tseg_qent *q; 248 struct tseg_qent *p = NULL; 249 struct tseg_qent *nq; 250 struct tseg_qent *te = NULL; 251 struct socket *so = tp->t_inpcb->inp_socket; 252 int flags; 253 254 INP_LOCK_ASSERT(tp->t_inpcb); 255 256 /* 257 * XXX: tcp_reass() is rather inefficient with its data structures 258 * and should be rewritten (see NetBSD for optimizations). While 259 * doing that it should move to its own file tcp_reass.c. 260 */ 261 262 /* 263 * Call with th==NULL after become established to 264 * force pre-ESTABLISHED data up to user socket. 265 */ 266 if (th == NULL) 267 goto present; 268 269 /* 270 * Limit the number of segments in the reassembly queue to prevent 271 * holding on to too many segments (and thus running out of mbufs). 272 * Make sure to let the missing segment through which caused this 273 * queue. Always keep one global queue entry spare to be able to 274 * process the missing segment. 275 */ 276 if (th->th_seq != tp->rcv_nxt && 277 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 278 tp->t_segqlen >= tcp_reass_maxqlen)) { 279 tcp_reass_overflows++; 280 tcpstat.tcps_rcvmemdrop++; 281 m_freem(m); 282 *tlenp = 0; 283 return (0); 284 } 285 286 /* 287 * Allocate a new queue entry. If we can't, or hit the zone limit 288 * just drop the pkt. 289 */ 290 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 291 if (te == NULL) { 292 tcpstat.tcps_rcvmemdrop++; 293 m_freem(m); 294 *tlenp = 0; 295 return (0); 296 } 297 tp->t_segqlen++; 298 tcp_reass_qsize++; 299 300 /* 301 * Find a segment which begins after this one does. 302 */ 303 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 304 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 305 break; 306 p = q; 307 } 308 309 /* 310 * If there is a preceding segment, it may provide some of 311 * our data already. If so, drop the data from the incoming 312 * segment. If it provides all of our data, drop us. 313 */ 314 if (p != NULL) { 315 register int i; 316 /* conversion to int (in i) handles seq wraparound */ 317 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 318 if (i > 0) { 319 if (i >= *tlenp) { 320 tcpstat.tcps_rcvduppack++; 321 tcpstat.tcps_rcvdupbyte += *tlenp; 322 m_freem(m); 323 uma_zfree(tcp_reass_zone, te); 324 tp->t_segqlen--; 325 tcp_reass_qsize--; 326 /* 327 * Try to present any queued data 328 * at the left window edge to the user. 329 * This is needed after the 3-WHS 330 * completes. 331 */ 332 goto present; /* ??? */ 333 } 334 m_adj(m, i); 335 *tlenp -= i; 336 th->th_seq += i; 337 } 338 } 339 tcpstat.tcps_rcvoopack++; 340 tcpstat.tcps_rcvoobyte += *tlenp; 341 342 /* 343 * While we overlap succeeding segments trim them or, 344 * if they are completely covered, dequeue them. 345 */ 346 while (q) { 347 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 348 if (i <= 0) 349 break; 350 if (i < q->tqe_len) { 351 q->tqe_th->th_seq += i; 352 q->tqe_len -= i; 353 m_adj(q->tqe_m, i); 354 break; 355 } 356 357 nq = LIST_NEXT(q, tqe_q); 358 LIST_REMOVE(q, tqe_q); 359 m_freem(q->tqe_m); 360 uma_zfree(tcp_reass_zone, q); 361 tp->t_segqlen--; 362 tcp_reass_qsize--; 363 q = nq; 364 } 365 366 /* Insert the new segment queue entry into place. */ 367 te->tqe_m = m; 368 te->tqe_th = th; 369 te->tqe_len = *tlenp; 370 371 if (p == NULL) { 372 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 373 } else { 374 LIST_INSERT_AFTER(p, te, tqe_q); 375 } 376 377present: 378 /* 379 * Present data to user, advancing rcv_nxt through 380 * completed sequence space. 381 */ 382 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 383 return (0); 384 q = LIST_FIRST(&tp->t_segq); 385 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 386 return (0); 387 SOCKBUF_LOCK(&so->so_rcv); 388 do { 389 tp->rcv_nxt += q->tqe_len; 390 flags = q->tqe_th->th_flags & TH_FIN; 391 nq = LIST_NEXT(q, tqe_q); 392 LIST_REMOVE(q, tqe_q); 393 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 394 m_freem(q->tqe_m); 395 else 396 sbappendstream_locked(&so->so_rcv, q->tqe_m); 397 uma_zfree(tcp_reass_zone, q); 398 tp->t_segqlen--; 399 tcp_reass_qsize--; 400 q = nq; 401 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 402 ND6_HINT(tp); 403 sorwakeup_locked(so); 404 return (flags); 405} 406 407/* 408 * TCP input routine, follows pages 65-76 of the 409 * protocol specification dated September, 1981 very closely. 410 */ 411#ifdef INET6 412int 413tcp6_input(mp, offp, proto) 414 struct mbuf **mp; 415 int *offp, proto; 416{ 417 register struct mbuf *m = *mp; 418 struct in6_ifaddr *ia6; 419 420 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 421 422 /* 423 * draft-itojun-ipv6-tcp-to-anycast 424 * better place to put this in? 425 */ 426 ia6 = ip6_getdstifaddr(m); 427 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 428 struct ip6_hdr *ip6; 429 430 ip6 = mtod(m, struct ip6_hdr *); 431 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 432 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 433 return IPPROTO_DONE; 434 } 435 436 tcp_input(m, *offp); 437 return IPPROTO_DONE; 438} 439#endif 440 441void 442tcp_input(m, off0) 443 register struct mbuf *m; 444 int off0; 445{ 446 register struct tcphdr *th; 447 register struct ip *ip = NULL; 448 register struct ipovly *ipov; 449 register struct inpcb *inp = NULL; 450 u_char *optp = NULL; 451 int optlen = 0; 452 int len, tlen, off; 453 int drop_hdrlen; 454 register struct tcpcb *tp = 0; 455 register int thflags; 456 struct socket *so = 0; 457 int todrop, acked, ourfinisacked, needoutput = 0; 458 u_long tiwin; 459 struct tcpopt to; /* options in this segment */ 460 int headlocked = 0; 461#ifdef IPFIREWALL_FORWARD 462 struct m_tag *fwd_tag; 463#endif 464 int rstreason; /* For badport_bandlim accounting purposes */ 465 466 struct ip6_hdr *ip6 = NULL; 467#ifdef INET6 468 int isipv6; 469 char ip6buf[INET6_ADDRSTRLEN]; 470#else 471 const int isipv6 = 0; 472#endif 473 474#ifdef TCPDEBUG 475 /* 476 * The size of tcp_saveipgen must be the size of the max ip header, 477 * now IPv6. 478 */ 479 u_char tcp_saveipgen[40]; 480 struct tcphdr tcp_savetcp; 481 short ostate = 0; 482#endif 483 484#ifdef INET6 485 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 486#endif 487 bzero((char *)&to, sizeof(to)); 488 489 tcpstat.tcps_rcvtotal++; 490 491 if (isipv6) { 492#ifdef INET6 493 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 494 ip6 = mtod(m, struct ip6_hdr *); 495 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 496 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 497 tcpstat.tcps_rcvbadsum++; 498 goto drop; 499 } 500 th = (struct tcphdr *)((caddr_t)ip6 + off0); 501 502 /* 503 * Be proactive about unspecified IPv6 address in source. 504 * As we use all-zero to indicate unbounded/unconnected pcb, 505 * unspecified IPv6 address can be used to confuse us. 506 * 507 * Note that packets with unspecified IPv6 destination is 508 * already dropped in ip6_input. 509 */ 510 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 511 /* XXX stat */ 512 goto drop; 513 } 514#else 515 th = NULL; /* XXX: avoid compiler warning */ 516#endif 517 } else { 518 /* 519 * Get IP and TCP header together in first mbuf. 520 * Note: IP leaves IP header in first mbuf. 521 */ 522 if (off0 > sizeof (struct ip)) { 523 ip_stripoptions(m, (struct mbuf *)0); 524 off0 = sizeof(struct ip); 525 } 526 if (m->m_len < sizeof (struct tcpiphdr)) { 527 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 528 tcpstat.tcps_rcvshort++; 529 return; 530 } 531 } 532 ip = mtod(m, struct ip *); 533 ipov = (struct ipovly *)ip; 534 th = (struct tcphdr *)((caddr_t)ip + off0); 535 tlen = ip->ip_len; 536 537 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 538 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 539 th->th_sum = m->m_pkthdr.csum_data; 540 else 541 th->th_sum = in_pseudo(ip->ip_src.s_addr, 542 ip->ip_dst.s_addr, 543 htonl(m->m_pkthdr.csum_data + 544 ip->ip_len + 545 IPPROTO_TCP)); 546 th->th_sum ^= 0xffff; 547#ifdef TCPDEBUG 548 ipov->ih_len = (u_short)tlen; 549 ipov->ih_len = htons(ipov->ih_len); 550#endif 551 } else { 552 /* 553 * Checksum extended TCP header and data. 554 */ 555 len = sizeof (struct ip) + tlen; 556 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 557 ipov->ih_len = (u_short)tlen; 558 ipov->ih_len = htons(ipov->ih_len); 559 th->th_sum = in_cksum(m, len); 560 } 561 if (th->th_sum) { 562 tcpstat.tcps_rcvbadsum++; 563 goto drop; 564 } 565 /* Re-initialization for later version check */ 566 ip->ip_v = IPVERSION; 567 } 568 569 /* 570 * Check that TCP offset makes sense, 571 * pull out TCP options and adjust length. XXX 572 */ 573 off = th->th_off << 2; 574 if (off < sizeof (struct tcphdr) || off > tlen) { 575 tcpstat.tcps_rcvbadoff++; 576 goto drop; 577 } 578 tlen -= off; /* tlen is used instead of ti->ti_len */ 579 if (off > sizeof (struct tcphdr)) { 580 if (isipv6) { 581#ifdef INET6 582 IP6_EXTHDR_CHECK(m, off0, off, ); 583 ip6 = mtod(m, struct ip6_hdr *); 584 th = (struct tcphdr *)((caddr_t)ip6 + off0); 585#endif 586 } else { 587 if (m->m_len < sizeof(struct ip) + off) { 588 if ((m = m_pullup(m, sizeof (struct ip) + off)) 589 == 0) { 590 tcpstat.tcps_rcvshort++; 591 return; 592 } 593 ip = mtod(m, struct ip *); 594 ipov = (struct ipovly *)ip; 595 th = (struct tcphdr *)((caddr_t)ip + off0); 596 } 597 } 598 optlen = off - sizeof (struct tcphdr); 599 optp = (u_char *)(th + 1); 600 } 601 thflags = th->th_flags; 602 603#ifdef TCP_DROP_SYNFIN 604 /* 605 * If the drop_synfin option is enabled, drop all packets with 606 * both the SYN and FIN bits set. This prevents e.g. nmap from 607 * identifying the TCP/IP stack. 608 * 609 * This is a violation of the TCP specification. 610 */ 611 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 612 goto drop; 613#endif 614 615 /* 616 * Convert TCP protocol specific fields to host format. 617 */ 618 th->th_seq = ntohl(th->th_seq); 619 th->th_ack = ntohl(th->th_ack); 620 th->th_win = ntohs(th->th_win); 621 th->th_urp = ntohs(th->th_urp); 622 623 /* 624 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 625 * until after ip6_savecontrol() is called and before other functions 626 * which don't want those proto headers. 627 * Because ip6_savecontrol() is going to parse the mbuf to 628 * search for data to be passed up to user-land, it wants mbuf 629 * parameters to be unchanged. 630 * XXX: the call of ip6_savecontrol() has been obsoleted based on 631 * latest version of the advanced API (20020110). 632 */ 633 drop_hdrlen = off0 + off; 634 635 /* 636 * Locate pcb for segment. 637 */ 638 INP_INFO_WLOCK(&tcbinfo); 639 headlocked = 1; 640findpcb: 641 KASSERT(headlocked, ("tcp_input: findpcb: head not locked")); 642#ifdef IPFIREWALL_FORWARD 643 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 644 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 645 646 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 647 struct sockaddr_in *next_hop; 648 649 next_hop = (struct sockaddr_in *)(fwd_tag+1); 650 /* 651 * Transparently forwarded. Pretend to be the destination. 652 * already got one like this? 653 */ 654 inp = in_pcblookup_hash(&tcbinfo, 655 ip->ip_src, th->th_sport, 656 ip->ip_dst, th->th_dport, 657 0, m->m_pkthdr.rcvif); 658 if (!inp) { 659 /* It's new. Try to find the ambushing socket. */ 660 inp = in_pcblookup_hash(&tcbinfo, 661 ip->ip_src, th->th_sport, 662 next_hop->sin_addr, 663 next_hop->sin_port ? 664 ntohs(next_hop->sin_port) : 665 th->th_dport, 666 INPLOOKUP_WILDCARD, 667 m->m_pkthdr.rcvif); 668 } 669 /* Remove the tag from the packet. We don't need it anymore. */ 670 m_tag_delete(m, fwd_tag); 671 } else { 672#endif /* IPFIREWALL_FORWARD */ 673 if (isipv6) { 674#ifdef INET6 675 inp = in6_pcblookup_hash(&tcbinfo, 676 &ip6->ip6_src, th->th_sport, 677 &ip6->ip6_dst, th->th_dport, 678 INPLOOKUP_WILDCARD, 679 m->m_pkthdr.rcvif); 680#endif 681 } else 682 inp = in_pcblookup_hash(&tcbinfo, 683 ip->ip_src, th->th_sport, 684 ip->ip_dst, th->th_dport, 685 INPLOOKUP_WILDCARD, 686 m->m_pkthdr.rcvif); 687#ifdef IPFIREWALL_FORWARD 688 } 689#endif /* IPFIREWALL_FORWARD */ 690 691#if defined(IPSEC) || defined(FAST_IPSEC) 692#ifdef INET6 693 if (isipv6) { 694 if (inp != NULL && ipsec6_in_reject(m, inp)) { 695#ifdef IPSEC 696 ipsec6stat.in_polvio++; 697#endif 698 goto drop; 699 } 700 } else 701#endif /* INET6 */ 702 if (inp != NULL && ipsec4_in_reject(m, inp)) { 703#ifdef IPSEC 704 ipsecstat.in_polvio++; 705#endif 706 goto drop; 707 } 708#endif /*IPSEC || FAST_IPSEC*/ 709 710 /* 711 * If the state is CLOSED (i.e., TCB does not exist) then 712 * all data in the incoming segment is discarded. 713 * If the TCB exists but is in CLOSED state, it is embryonic, 714 * but should either do a listen or a connect soon. 715 */ 716 if (inp == NULL) { 717 if (tcp_log_in_vain) { 718#ifdef INET6 719 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 720#else 721 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 722#endif 723 724 if (isipv6) { 725#ifdef INET6 726 strcpy(dbuf, "["); 727 strcpy(sbuf, "["); 728 strcat(dbuf, 729 ip6_sprintf(ip6buf, &ip6->ip6_dst)); 730 strcat(sbuf, 731 ip6_sprintf(ip6buf, &ip6->ip6_src)); 732 strcat(dbuf, "]"); 733 strcat(sbuf, "]"); 734#endif 735 } else { 736 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 737 strcpy(sbuf, inet_ntoa(ip->ip_src)); 738 } 739 switch (tcp_log_in_vain) { 740 case 1: 741 if ((thflags & TH_SYN) == 0) 742 break; 743 /* FALLTHROUGH */ 744 case 2: 745 log(LOG_INFO, 746 "Connection attempt to TCP %s:%d " 747 "from %s:%d flags:0x%02x\n", 748 dbuf, ntohs(th->th_dport), sbuf, 749 ntohs(th->th_sport), thflags); 750 break; 751 default: 752 break; 753 } 754 } 755 if (blackhole) { 756 switch (blackhole) { 757 case 1: 758 if (thflags & TH_SYN) 759 goto drop; 760 break; 761 case 2: 762 goto drop; 763 default: 764 goto drop; 765 } 766 } 767 rstreason = BANDLIM_RST_CLOSEDPORT; 768 goto dropwithreset; 769 } 770 INP_LOCK(inp); 771 772 /* Check the minimum TTL for socket. */ 773 if (inp->inp_ip_minttl != 0) { 774#ifdef INET6 775 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 776 goto drop; 777 else 778#endif 779 if (inp->inp_ip_minttl > ip->ip_ttl) 780 goto drop; 781 } 782 783 if (inp->inp_vflag & INP_TIMEWAIT) { 784 /* 785 * The only option of relevance is TOF_CC, and only if 786 * present in a SYN segment. See tcp_timewait(). 787 */ 788 if (thflags & TH_SYN) 789 tcp_dooptions(&to, optp, optlen, TO_SYN); 790 if (tcp_timewait(inp, &to, th, m, tlen)) 791 goto findpcb; 792 /* 793 * tcp_timewait unlocks inp. 794 */ 795 INP_INFO_WUNLOCK(&tcbinfo); 796 return; 797 } 798 tp = intotcpcb(inp); 799 if (tp == 0) { 800 INP_UNLOCK(inp); 801 rstreason = BANDLIM_RST_CLOSEDPORT; 802 goto dropwithreset; 803 } 804 if (tp->t_state == TCPS_CLOSED) 805 goto drop; 806 807#ifdef MAC 808 INP_LOCK_ASSERT(inp); 809 if (mac_check_inpcb_deliver(inp, m)) 810 goto drop; 811#endif 812 so = inp->inp_socket; 813 KASSERT(so != NULL, ("tcp_input: so == NULL")); 814#ifdef TCPDEBUG 815 if (so->so_options & SO_DEBUG) { 816 ostate = tp->t_state; 817 if (isipv6) 818 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 819 else 820 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 821 tcp_savetcp = *th; 822 } 823#endif 824 if (so->so_options & SO_ACCEPTCONN) { 825 struct in_conninfo inc; 826 827 bzero(&inc, sizeof(inc)); 828#ifdef INET6 829 inc.inc_isipv6 = isipv6; 830#endif 831 if (isipv6) { 832 inc.inc6_faddr = ip6->ip6_src; 833 inc.inc6_laddr = ip6->ip6_dst; 834 } else { 835 inc.inc_faddr = ip->ip_src; 836 inc.inc_laddr = ip->ip_dst; 837 } 838 inc.inc_fport = th->th_sport; 839 inc.inc_lport = th->th_dport; 840 841 /* 842 * If the state is LISTEN then ignore segment if it contains 843 * a RST. If the segment contains an ACK then it is bad and 844 * send a RST. If it does not contain a SYN then it is not 845 * interesting; drop it. 846 * 847 * If the state is SYN_RECEIVED (syncache) and seg contains 848 * an ACK, but not for our SYN/ACK, send a RST. If the seg 849 * contains a RST, check the sequence number to see if it 850 * is a valid reset segment. 851 */ 852 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 853 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 854 /* 855 * Parse the TCP options here because 856 * syncookies need access to the reflected 857 * timestamp. 858 */ 859 tcp_dooptions(&to, optp, optlen, 0); 860 if (!syncache_expand(&inc, &to, th, &so, m)) { 861 /* 862 * No syncache entry, or ACK was not 863 * for our SYN/ACK. Send a RST. 864 */ 865 tcpstat.tcps_badsyn++; 866 rstreason = BANDLIM_RST_OPENPORT; 867 goto dropwithreset; 868 } 869 if (so == NULL) { 870 /* 871 * Could not complete 3-way handshake, 872 * connection is being closed down, and 873 * syncache has free'd mbuf. 874 */ 875 INP_UNLOCK(inp); 876 INP_INFO_WUNLOCK(&tcbinfo); 877 return; 878 } 879 /* 880 * Socket is created in state SYN_RECEIVED. 881 * Continue processing segment. 882 */ 883 INP_UNLOCK(inp); 884 inp = sotoinpcb(so); 885 INP_LOCK(inp); 886 tp = intotcpcb(inp); 887 /* 888 * This is what would have happened in 889 * tcp_output() when the SYN,ACK was sent. 890 */ 891 tp->snd_up = tp->snd_una; 892 tp->snd_max = tp->snd_nxt = tp->iss + 1; 893 tp->last_ack_sent = tp->rcv_nxt; 894 goto after_listen; 895 } 896 if (thflags & TH_RST) { 897 syncache_chkrst(&inc, th); 898 goto drop; 899 } 900 if (thflags & TH_ACK) { 901 syncache_badack(&inc); 902 tcpstat.tcps_badsyn++; 903 rstreason = BANDLIM_RST_OPENPORT; 904 goto dropwithreset; 905 } 906 goto drop; 907 } 908 909 /* 910 * Segment's flags are (SYN) or (SYN|FIN). 911 */ 912#ifdef INET6 913 /* 914 * If deprecated address is forbidden, 915 * we do not accept SYN to deprecated interface 916 * address to prevent any new inbound connection from 917 * getting established. 918 * When we do not accept SYN, we send a TCP RST, 919 * with deprecated source address (instead of dropping 920 * it). We compromise it as it is much better for peer 921 * to send a RST, and RST will be the final packet 922 * for the exchange. 923 * 924 * If we do not forbid deprecated addresses, we accept 925 * the SYN packet. RFC2462 does not suggest dropping 926 * SYN in this case. 927 * If we decipher RFC2462 5.5.4, it says like this: 928 * 1. use of deprecated addr with existing 929 * communication is okay - "SHOULD continue to be 930 * used" 931 * 2. use of it with new communication: 932 * (2a) "SHOULD NOT be used if alternate address 933 * with sufficient scope is available" 934 * (2b) nothing mentioned otherwise. 935 * Here we fall into (2b) case as we have no choice in 936 * our source address selection - we must obey the peer. 937 * 938 * The wording in RFC2462 is confusing, and there are 939 * multiple description text for deprecated address 940 * handling - worse, they are not exactly the same. 941 * I believe 5.5.4 is the best one, so we follow 5.5.4. 942 */ 943 if (isipv6 && !ip6_use_deprecated) { 944 struct in6_ifaddr *ia6; 945 946 if ((ia6 = ip6_getdstifaddr(m)) && 947 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 948 INP_UNLOCK(inp); 949 tp = NULL; 950 rstreason = BANDLIM_RST_OPENPORT; 951 goto dropwithreset; 952 } 953 } 954#endif 955 /* 956 * If it is from this socket, drop it, it must be forged. 957 * Don't bother responding if the destination was a broadcast. 958 */ 959 if (th->th_dport == th->th_sport) { 960 if (isipv6) { 961 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 962 &ip6->ip6_src)) 963 goto drop; 964 } else { 965 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 966 goto drop; 967 } 968 } 969 /* 970 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 971 * 972 * Note that it is quite possible to receive unicast 973 * link-layer packets with a broadcast IP address. Use 974 * in_broadcast() to find them. 975 */ 976 if (m->m_flags & (M_BCAST|M_MCAST)) 977 goto drop; 978 if (isipv6) { 979 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 980 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 981 goto drop; 982 } else { 983 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 984 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 985 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 986 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 987 goto drop; 988 } 989 /* 990 * SYN appears to be valid; create compressed TCP state 991 * for syncache, or perform t/tcp connection. 992 */ 993 if (so->so_qlen <= so->so_qlimit) { 994#ifdef TCPDEBUG 995 if (so->so_options & SO_DEBUG) 996 tcp_trace(TA_INPUT, ostate, tp, 997 (void *)tcp_saveipgen, &tcp_savetcp, 0); 998#endif 999 tcp_dooptions(&to, optp, optlen, TO_SYN); 1000 if (!syncache_add(&inc, &to, th, inp, &so, m)) 1001 goto drop; /* XXX: does not happen */ 1002 if (so == NULL) { 1003 /* 1004 * Entry added to syncache, mbuf used to 1005 * send SYN,ACK packet. Everything unlocked 1006 * already. 1007 */ 1008 return; 1009 } 1010 panic("T/TCP not supported at the moment"); 1011#if 0 /* T/TCP */ 1012 /* 1013 * Segment passed TAO tests. 1014 * XXX: Can't happen at the moment. 1015 */ 1016 INP_UNLOCK(inp); 1017 inp = sotoinpcb(so); 1018 INP_LOCK(inp); 1019 tp = intotcpcb(inp); 1020 tp->t_starttime = ticks; 1021 tp->t_state = TCPS_ESTABLISHED; 1022 1023 /* 1024 * T/TCP logic: 1025 * If there is a FIN or if there is data, then 1026 * delay SYN,ACK(SYN) in the hope of piggy-backing 1027 * it on a response segment. Otherwise must send 1028 * ACK now in case the other side is slow starting. 1029 */ 1030 if (thflags & TH_FIN || tlen != 0) 1031 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 1032 else 1033 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1034 tiwin = th->th_win << tp->snd_scale; 1035 tcpstat.tcps_connects++; 1036 soisconnected(so); 1037 goto trimthenstep6; 1038#endif /* T/TCP */ 1039 } 1040 goto drop; 1041 } 1042after_listen: 1043 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1044 INP_LOCK_ASSERT(inp); 1045 1046 /* Syncache takes care of sockets in the listen state. */ 1047 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN")); 1048 1049 /* 1050 * This is the second part of the MSS DoS prevention code (after 1051 * minmss on the sending side) and it deals with too many too small 1052 * tcp packets in a too short timeframe (1 second). 1053 * 1054 * For every full second we count the number of received packets 1055 * and bytes. If we get a lot of packets per second for this connection 1056 * (tcp_minmssoverload) we take a closer look at it and compute the 1057 * average packet size for the past second. If that is less than 1058 * tcp_minmss we get too many packets with very small payload which 1059 * is not good and burdens our system (and every packet generates 1060 * a wakeup to the process connected to our socket). We can reasonable 1061 * expect this to be small packet DoS attack to exhaust our CPU 1062 * cycles. 1063 * 1064 * Care has to be taken for the minimum packet overload value. This 1065 * value defines the minimum number of packets per second before we 1066 * start to worry. This must not be too low to avoid killing for 1067 * example interactive connections with many small packets like 1068 * telnet or SSH. 1069 * 1070 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1071 * this check. 1072 * 1073 * Account for packet if payload packet, skip over ACK, etc. 1074 */ 1075 if (tcp_minmss && tcp_minmssoverload && 1076 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1077 if ((unsigned int)(tp->rcv_second - ticks) < hz) { 1078 tp->rcv_pps++; 1079 tp->rcv_byps += tlen + off; 1080 if (tp->rcv_pps > tcp_minmssoverload) { 1081 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1082 printf("too many small tcp packets from " 1083 "%s:%u, av. %lubyte/packet, " 1084 "dropping connection\n", 1085#ifdef INET6 1086 isipv6 ? 1087 ip6_sprintf(ip6buf, 1088 &inp->inp_inc.inc6_faddr) : 1089#endif 1090 inet_ntoa(inp->inp_inc.inc_faddr), 1091 inp->inp_inc.inc_fport, 1092 tp->rcv_byps / tp->rcv_pps); 1093 KASSERT(headlocked, ("tcp_input: " 1094 "after_listen: tcp_drop: head " 1095 "not locked")); 1096 tp = tcp_drop(tp, ECONNRESET); 1097 tcpstat.tcps_minmssdrops++; 1098 goto drop; 1099 } 1100 } 1101 } else { 1102 tp->rcv_second = ticks + hz; 1103 tp->rcv_pps = 1; 1104 tp->rcv_byps = tlen + off; 1105 } 1106 } 1107 1108 /* 1109 * Segment received on connection. 1110 * Reset idle time and keep-alive timer. 1111 */ 1112 tp->t_rcvtime = ticks; 1113 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1114 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1115 1116 /* 1117 * Unscale the window into a 32-bit value. 1118 * This value is bogus for the TCPS_SYN_SENT state 1119 * and is overwritten later. 1120 */ 1121 tiwin = th->th_win << tp->snd_scale; 1122 1123 /* 1124 * Parse options on any incoming segment. 1125 */ 1126 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0); 1127 1128 /* 1129 * If echoed timestamp is later than the current time, 1130 * fall back to non RFC1323 RTT calculation. Normalize 1131 * timestamp if syncookies were used when this connection 1132 * was established. 1133 */ 1134 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1135 to.to_tsecr -= tp->ts_offset; 1136 if (TSTMP_GT(to.to_tsecr, ticks)) 1137 to.to_tsecr = 0; 1138 } 1139 1140 /* 1141 * Process options only when we get SYN/ACK back. The SYN case 1142 * for incoming connections is handled in tcp_syncache. 1143 * XXX this is traditional behavior, may need to be cleaned up. 1144 */ 1145 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1146 if ((to.to_flags & TOF_SCALE) && 1147 (tp->t_flags & TF_REQ_SCALE)) { 1148 tp->t_flags |= TF_RCVD_SCALE; 1149 tp->snd_scale = to.to_requested_s_scale; 1150 tp->snd_wnd = th->th_win << tp->snd_scale; 1151 tiwin = tp->snd_wnd; 1152 } 1153 if (to.to_flags & TOF_TS) { 1154 tp->t_flags |= TF_RCVD_TSTMP; 1155 tp->ts_recent = to.to_tsval; 1156 tp->ts_recent_age = ticks; 1157 } 1158 if (to.to_flags & TOF_MSS) 1159 tcp_mss(tp, to.to_mss); 1160 if (tp->sack_enable) { 1161 if (!(to.to_flags & TOF_SACK)) 1162 tp->sack_enable = 0; 1163 else 1164 tp->t_flags |= TF_SACK_PERMIT; 1165 } 1166 1167 } 1168 1169 /* 1170 * Header prediction: check for the two common cases 1171 * of a uni-directional data xfer. If the packet has 1172 * no control flags, is in-sequence, the window didn't 1173 * change and we're not retransmitting, it's a 1174 * candidate. If the length is zero and the ack moved 1175 * forward, we're the sender side of the xfer. Just 1176 * free the data acked & wake any higher level process 1177 * that was blocked waiting for space. If the length 1178 * is non-zero and the ack didn't move, we're the 1179 * receiver side. If we're getting packets in-order 1180 * (the reassembly queue is empty), add the data to 1181 * the socket buffer and note that we need a delayed ack. 1182 * Make sure that the hidden state-flags are also off. 1183 * Since we check for TCPS_ESTABLISHED above, it can only 1184 * be TH_NEEDSYN. 1185 */ 1186 if (tp->t_state == TCPS_ESTABLISHED && 1187 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1188 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1189 ((to.to_flags & TOF_TS) == 0 || 1190 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1191 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1192 tp->snd_nxt == tp->snd_max) { 1193 1194 /* 1195 * If last ACK falls within this segment's sequence numbers, 1196 * record the timestamp. 1197 * NOTE that the test is modified according to the latest 1198 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1199 */ 1200 if ((to.to_flags & TOF_TS) != 0 && 1201 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1202 tp->ts_recent_age = ticks; 1203 tp->ts_recent = to.to_tsval; 1204 } 1205 1206 if (tlen == 0) { 1207 if (SEQ_GT(th->th_ack, tp->snd_una) && 1208 SEQ_LEQ(th->th_ack, tp->snd_max) && 1209 tp->snd_cwnd >= tp->snd_wnd && 1210 ((!tcp_do_newreno && !tp->sack_enable && 1211 tp->t_dupacks < tcprexmtthresh) || 1212 ((tcp_do_newreno || tp->sack_enable) && 1213 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 && 1214 TAILQ_EMPTY(&tp->snd_holes)))) { 1215 KASSERT(headlocked, ("headlocked")); 1216 INP_INFO_WUNLOCK(&tcbinfo); 1217 headlocked = 0; 1218 /* 1219 * this is a pure ack for outstanding data. 1220 */ 1221 ++tcpstat.tcps_predack; 1222 /* 1223 * "bad retransmit" recovery 1224 */ 1225 if (tp->t_rxtshift == 1 && 1226 ticks < tp->t_badrxtwin) { 1227 ++tcpstat.tcps_sndrexmitbad; 1228 tp->snd_cwnd = tp->snd_cwnd_prev; 1229 tp->snd_ssthresh = 1230 tp->snd_ssthresh_prev; 1231 tp->snd_recover = tp->snd_recover_prev; 1232 if (tp->t_flags & TF_WASFRECOVERY) 1233 ENTER_FASTRECOVERY(tp); 1234 tp->snd_nxt = tp->snd_max; 1235 tp->t_badrxtwin = 0; 1236 } 1237 1238 /* 1239 * Recalculate the transmit timer / rtt. 1240 * 1241 * Some boxes send broken timestamp replies 1242 * during the SYN+ACK phase, ignore 1243 * timestamps of 0 or we could calculate a 1244 * huge RTT and blow up the retransmit timer. 1245 */ 1246 if ((to.to_flags & TOF_TS) != 0 && 1247 to.to_tsecr) { 1248 if (!tp->t_rttlow || 1249 tp->t_rttlow > ticks - to.to_tsecr) 1250 tp->t_rttlow = ticks - to.to_tsecr; 1251 tcp_xmit_timer(tp, 1252 ticks - to.to_tsecr + 1); 1253 } else if (tp->t_rtttime && 1254 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1255 if (!tp->t_rttlow || 1256 tp->t_rttlow > ticks - tp->t_rtttime) 1257 tp->t_rttlow = ticks - tp->t_rtttime; 1258 tcp_xmit_timer(tp, 1259 ticks - tp->t_rtttime); 1260 } 1261 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1262 acked = th->th_ack - tp->snd_una; 1263 tcpstat.tcps_rcvackpack++; 1264 tcpstat.tcps_rcvackbyte += acked; 1265 sbdrop(&so->so_snd, acked); 1266 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1267 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1268 tp->snd_recover = th->th_ack - 1; 1269 tp->snd_una = th->th_ack; 1270 /* 1271 * pull snd_wl2 up to prevent seq wrap relative 1272 * to th_ack. 1273 */ 1274 tp->snd_wl2 = th->th_ack; 1275 tp->t_dupacks = 0; 1276 m_freem(m); 1277 ND6_HINT(tp); /* some progress has been done */ 1278 1279 /* 1280 * If all outstanding data are acked, stop 1281 * retransmit timer, otherwise restart timer 1282 * using current (possibly backed-off) value. 1283 * If process is waiting for space, 1284 * wakeup/selwakeup/signal. If data 1285 * are ready to send, let tcp_output 1286 * decide between more output or persist. 1287 1288#ifdef TCPDEBUG 1289 if (so->so_options & SO_DEBUG) 1290 tcp_trace(TA_INPUT, ostate, tp, 1291 (void *)tcp_saveipgen, 1292 &tcp_savetcp, 0); 1293#endif 1294 */ 1295 if (tp->snd_una == tp->snd_max) 1296 callout_stop(tp->tt_rexmt); 1297 else if (!callout_active(tp->tt_persist)) 1298 callout_reset(tp->tt_rexmt, 1299 tp->t_rxtcur, 1300 tcp_timer_rexmt, tp); 1301 1302 sowwakeup(so); 1303 if (so->so_snd.sb_cc) 1304 (void) tcp_output(tp); 1305 goto check_delack; 1306 } 1307 } else if (th->th_ack == tp->snd_una && 1308 LIST_EMPTY(&tp->t_segq) && 1309 tlen <= sbspace(&so->so_rcv)) { 1310 int newsize = 0; /* automatic sockbuf scaling */ 1311 1312 KASSERT(headlocked, ("headlocked")); 1313 INP_INFO_WUNLOCK(&tcbinfo); 1314 headlocked = 0; 1315 /* 1316 * this is a pure, in-sequence data packet 1317 * with nothing on the reassembly queue and 1318 * we have enough buffer space to take it. 1319 */ 1320 /* Clean receiver SACK report if present */ 1321 if (tp->sack_enable && tp->rcv_numsacks) 1322 tcp_clean_sackreport(tp); 1323 ++tcpstat.tcps_preddat; 1324 tp->rcv_nxt += tlen; 1325 /* 1326 * Pull snd_wl1 up to prevent seq wrap relative to 1327 * th_seq. 1328 */ 1329 tp->snd_wl1 = th->th_seq; 1330 /* 1331 * Pull rcv_up up to prevent seq wrap relative to 1332 * rcv_nxt. 1333 */ 1334 tp->rcv_up = tp->rcv_nxt; 1335 tcpstat.tcps_rcvpack++; 1336 tcpstat.tcps_rcvbyte += tlen; 1337 ND6_HINT(tp); /* some progress has been done */ 1338#ifdef TCPDEBUG 1339 if (so->so_options & SO_DEBUG) 1340 tcp_trace(TA_INPUT, ostate, tp, 1341 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1342#endif 1343 /* 1344 * Automatic sizing of receive socket buffer. Often the send 1345 * buffer size is not optimally adjusted to the actual network 1346 * conditions at hand (delay bandwidth product). Setting the 1347 * buffer size too small limits throughput on links with high 1348 * bandwidth and high delay (eg. trans-continental/oceanic links). 1349 * 1350 * On the receive side the socket buffer memory is only rarely 1351 * used to any significant extent. This allows us to be much 1352 * more aggressive in scaling the receive socket buffer. For 1353 * the case that the buffer space is actually used to a large 1354 * extent and we run out of kernel memory we can simply drop 1355 * the new segments; TCP on the sender will just retransmit it 1356 * later. Setting the buffer size too big may only consume too 1357 * much kernel memory if the application doesn't read() from 1358 * the socket or packet loss or reordering makes use of the 1359 * reassembly queue. 1360 * 1361 * The criteria to step up the receive buffer one notch are: 1362 * 1. the number of bytes received during the time it takes 1363 * one timestamp to be reflected back to us (the RTT); 1364 * 2. received bytes per RTT is within seven eighth of the 1365 * current socket buffer size; 1366 * 3. receive buffer size has not hit maximal automatic size; 1367 * 1368 * This algorithm does one step per RTT at most and only if 1369 * we receive a bulk stream w/o packet losses or reorderings. 1370 * Shrinking the buffer during idle times is not necessary as 1371 * it doesn't consume any memory when idle. 1372 * 1373 * TODO: Only step up if the application is actually serving 1374 * the buffer to better manage the socket buffer resources. 1375 */ 1376 if (tcp_do_autorcvbuf && 1377 to.to_tsecr && 1378 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1379 if (to.to_tsecr > tp->rfbuf_ts && 1380 to.to_tsecr - tp->rfbuf_ts < hz) { 1381 if (tp->rfbuf_cnt > 1382 (so->so_rcv.sb_hiwat / 8 * 7) && 1383 so->so_rcv.sb_hiwat < 1384 tcp_autorcvbuf_max) { 1385 newsize = 1386 min(so->so_rcv.sb_hiwat + 1387 tcp_autorcvbuf_inc, 1388 tcp_autorcvbuf_max); 1389 } 1390 /* Start over with next RTT. */ 1391 tp->rfbuf_ts = 0; 1392 tp->rfbuf_cnt = 0; 1393 } else 1394 tp->rfbuf_cnt += tlen; /* add up */ 1395 } 1396 1397 /* Add data to socket buffer. */ 1398 SOCKBUF_LOCK(&so->so_rcv); 1399 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1400 m_freem(m); 1401 } else { 1402 /* 1403 * Set new socket buffer size. 1404 * Give up when limit is reached. 1405 */ 1406 if (newsize) 1407 if (!sbreserve_locked(&so->so_rcv, 1408 newsize, so, curthread)) 1409 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1410 m_adj(m, drop_hdrlen); /* delayed header drop */ 1411 sbappendstream_locked(&so->so_rcv, m); 1412 } 1413 sorwakeup_locked(so); 1414 if (DELAY_ACK(tp)) { 1415 tp->t_flags |= TF_DELACK; 1416 } else { 1417 tp->t_flags |= TF_ACKNOW; 1418 tcp_output(tp); 1419 } 1420 goto check_delack; 1421 } 1422 } 1423 1424 /* 1425 * Calculate amount of space in receive window, 1426 * and then do TCP input processing. 1427 * Receive window is amount of space in rcv queue, 1428 * but not less than advertised window. 1429 */ 1430 { int win; 1431 1432 win = sbspace(&so->so_rcv); 1433 if (win < 0) 1434 win = 0; 1435 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1436 } 1437 1438 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1439 tp->rfbuf_ts = 0; 1440 tp->rfbuf_cnt = 0; 1441 1442 switch (tp->t_state) { 1443 1444 /* 1445 * If the state is SYN_RECEIVED: 1446 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1447 */ 1448 case TCPS_SYN_RECEIVED: 1449 if ((thflags & TH_ACK) && 1450 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1451 SEQ_GT(th->th_ack, tp->snd_max))) { 1452 rstreason = BANDLIM_RST_OPENPORT; 1453 goto dropwithreset; 1454 } 1455 break; 1456 1457 /* 1458 * If the state is SYN_SENT: 1459 * if seg contains an ACK, but not for our SYN, drop the input. 1460 * if seg contains a RST, then drop the connection. 1461 * if seg does not contain SYN, then drop it. 1462 * Otherwise this is an acceptable SYN segment 1463 * initialize tp->rcv_nxt and tp->irs 1464 * if seg contains ack then advance tp->snd_una 1465 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1466 * arrange for segment to be acked (eventually) 1467 * continue processing rest of data/controls, beginning with URG 1468 */ 1469 case TCPS_SYN_SENT: 1470 if ((thflags & TH_ACK) && 1471 (SEQ_LEQ(th->th_ack, tp->iss) || 1472 SEQ_GT(th->th_ack, tp->snd_max))) { 1473 rstreason = BANDLIM_UNLIMITED; 1474 goto dropwithreset; 1475 } 1476 if (thflags & TH_RST) { 1477 if (thflags & TH_ACK) { 1478 KASSERT(headlocked, ("tcp_input: after_listen" 1479 ": tcp_drop.2: head not locked")); 1480 tp = tcp_drop(tp, ECONNREFUSED); 1481 } 1482 goto drop; 1483 } 1484 if ((thflags & TH_SYN) == 0) 1485 goto drop; 1486 1487 /* Initial send window, already scaled. */ 1488 tp->snd_wnd = th->th_win; 1489 1490 tp->irs = th->th_seq; 1491 tcp_rcvseqinit(tp); 1492 if (thflags & TH_ACK) { 1493 tcpstat.tcps_connects++; 1494 soisconnected(so); 1495#ifdef MAC 1496 SOCK_LOCK(so); 1497 mac_set_socket_peer_from_mbuf(m, so); 1498 SOCK_UNLOCK(so); 1499#endif 1500 /* Do window scaling on this connection? */ 1501 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1502 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1503 tp->rcv_scale = tp->request_r_scale; 1504 } 1505 tp->rcv_adv += tp->rcv_wnd; 1506 tp->snd_una++; /* SYN is acked */ 1507 /* 1508 * If there's data, delay ACK; if there's also a FIN 1509 * ACKNOW will be turned on later. 1510 */ 1511 if (DELAY_ACK(tp) && tlen != 0) 1512 callout_reset(tp->tt_delack, tcp_delacktime, 1513 tcp_timer_delack, tp); 1514 else 1515 tp->t_flags |= TF_ACKNOW; 1516 /* 1517 * Received <SYN,ACK> in SYN_SENT[*] state. 1518 * Transitions: 1519 * SYN_SENT --> ESTABLISHED 1520 * SYN_SENT* --> FIN_WAIT_1 1521 */ 1522 tp->t_starttime = ticks; 1523 if (tp->t_flags & TF_NEEDFIN) { 1524 tp->t_state = TCPS_FIN_WAIT_1; 1525 tp->t_flags &= ~TF_NEEDFIN; 1526 thflags &= ~TH_SYN; 1527 } else { 1528 tp->t_state = TCPS_ESTABLISHED; 1529 callout_reset(tp->tt_keep, tcp_keepidle, 1530 tcp_timer_keep, tp); 1531 } 1532 } else { 1533 /* 1534 * Received initial SYN in SYN-SENT[*] state => 1535 * simultaneous open. If segment contains CC option 1536 * and there is a cached CC, apply TAO test. 1537 * If it succeeds, connection is * half-synchronized. 1538 * Otherwise, do 3-way handshake: 1539 * SYN-SENT -> SYN-RECEIVED 1540 * SYN-SENT* -> SYN-RECEIVED* 1541 * If there was no CC option, clear cached CC value. 1542 */ 1543 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1544 callout_stop(tp->tt_rexmt); 1545 tp->t_state = TCPS_SYN_RECEIVED; 1546 } 1547 1548#if 0 /* T/TCP */ 1549trimthenstep6: 1550#endif 1551 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1552 "locked")); 1553 INP_LOCK_ASSERT(inp); 1554 1555 /* 1556 * Advance th->th_seq to correspond to first data byte. 1557 * If data, trim to stay within window, 1558 * dropping FIN if necessary. 1559 */ 1560 th->th_seq++; 1561 if (tlen > tp->rcv_wnd) { 1562 todrop = tlen - tp->rcv_wnd; 1563 m_adj(m, -todrop); 1564 tlen = tp->rcv_wnd; 1565 thflags &= ~TH_FIN; 1566 tcpstat.tcps_rcvpackafterwin++; 1567 tcpstat.tcps_rcvbyteafterwin += todrop; 1568 } 1569 tp->snd_wl1 = th->th_seq - 1; 1570 tp->rcv_up = th->th_seq; 1571 /* 1572 * Client side of transaction: already sent SYN and data. 1573 * If the remote host used T/TCP to validate the SYN, 1574 * our data will be ACK'd; if so, enter normal data segment 1575 * processing in the middle of step 5, ack processing. 1576 * Otherwise, goto step 6. 1577 */ 1578 if (thflags & TH_ACK) 1579 goto process_ACK; 1580 1581 goto step6; 1582 1583 /* 1584 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1585 * do normal processing. 1586 * 1587 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1588 */ 1589 case TCPS_LAST_ACK: 1590 case TCPS_CLOSING: 1591 case TCPS_TIME_WAIT: 1592 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1593 break; /* continue normal processing */ 1594 } 1595 1596 /* 1597 * States other than LISTEN or SYN_SENT. 1598 * First check the RST flag and sequence number since reset segments 1599 * are exempt from the timestamp and connection count tests. This 1600 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1601 * below which allowed reset segments in half the sequence space 1602 * to fall though and be processed (which gives forged reset 1603 * segments with a random sequence number a 50 percent chance of 1604 * killing a connection). 1605 * Then check timestamp, if present. 1606 * Then check the connection count, if present. 1607 * Then check that at least some bytes of segment are within 1608 * receive window. If segment begins before rcv_nxt, 1609 * drop leading data (and SYN); if nothing left, just ack. 1610 * 1611 * 1612 * If the RST bit is set, check the sequence number to see 1613 * if this is a valid reset segment. 1614 * RFC 793 page 37: 1615 * In all states except SYN-SENT, all reset (RST) segments 1616 * are validated by checking their SEQ-fields. A reset is 1617 * valid if its sequence number is in the window. 1618 * Note: this does not take into account delayed ACKs, so 1619 * we should test against last_ack_sent instead of rcv_nxt. 1620 * The sequence number in the reset segment is normally an 1621 * echo of our outgoing acknowlegement numbers, but some hosts 1622 * send a reset with the sequence number at the rightmost edge 1623 * of our receive window, and we have to handle this case. 1624 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1625 * that brute force RST attacks are possible. To combat this, 1626 * we use a much stricter check while in the ESTABLISHED state, 1627 * only accepting RSTs where the sequence number is equal to 1628 * last_ack_sent. In all other states (the states in which a 1629 * RST is more likely), the more permissive check is used. 1630 * If we have multiple segments in flight, the intial reset 1631 * segment sequence numbers will be to the left of last_ack_sent, 1632 * but they will eventually catch up. 1633 * In any case, it never made sense to trim reset segments to 1634 * fit the receive window since RFC 1122 says: 1635 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1636 * 1637 * A TCP SHOULD allow a received RST segment to include data. 1638 * 1639 * DISCUSSION 1640 * It has been suggested that a RST segment could contain 1641 * ASCII text that encoded and explained the cause of the 1642 * RST. No standard has yet been established for such 1643 * data. 1644 * 1645 * If the reset segment passes the sequence number test examine 1646 * the state: 1647 * SYN_RECEIVED STATE: 1648 * If passive open, return to LISTEN state. 1649 * If active open, inform user that connection was refused. 1650 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1651 * Inform user that connection was reset, and close tcb. 1652 * CLOSING, LAST_ACK STATES: 1653 * Close the tcb. 1654 * TIME_WAIT STATE: 1655 * Drop the segment - see Stevens, vol. 2, p. 964 and 1656 * RFC 1337. 1657 */ 1658 if (thflags & TH_RST) { 1659 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1660 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1661 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1662 switch (tp->t_state) { 1663 1664 case TCPS_SYN_RECEIVED: 1665 so->so_error = ECONNREFUSED; 1666 goto close; 1667 1668 case TCPS_ESTABLISHED: 1669 if (tp->last_ack_sent != th->th_seq && 1670 tcp_insecure_rst == 0) { 1671 tcpstat.tcps_badrst++; 1672 goto drop; 1673 } 1674 case TCPS_FIN_WAIT_1: 1675 case TCPS_FIN_WAIT_2: 1676 case TCPS_CLOSE_WAIT: 1677 so->so_error = ECONNRESET; 1678 close: 1679 tp->t_state = TCPS_CLOSED; 1680 tcpstat.tcps_drops++; 1681 KASSERT(headlocked, ("tcp_input: " 1682 "trimthenstep6: tcp_close: head not " 1683 "locked")); 1684 tp = tcp_close(tp); 1685 break; 1686 1687 case TCPS_CLOSING: 1688 case TCPS_LAST_ACK: 1689 KASSERT(headlocked, ("trimthenstep6: " 1690 "tcp_close.2: head not locked")); 1691 tp = tcp_close(tp); 1692 break; 1693 1694 case TCPS_TIME_WAIT: 1695 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1696 ("timewait")); 1697 break; 1698 } 1699 } 1700 goto drop; 1701 } 1702 1703 /* 1704 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1705 * and it's less than ts_recent, drop it. 1706 */ 1707 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1708 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1709 1710 /* Check to see if ts_recent is over 24 days old. */ 1711 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1712 /* 1713 * Invalidate ts_recent. If this segment updates 1714 * ts_recent, the age will be reset later and ts_recent 1715 * will get a valid value. If it does not, setting 1716 * ts_recent to zero will at least satisfy the 1717 * requirement that zero be placed in the timestamp 1718 * echo reply when ts_recent isn't valid. The 1719 * age isn't reset until we get a valid ts_recent 1720 * because we don't want out-of-order segments to be 1721 * dropped when ts_recent is old. 1722 */ 1723 tp->ts_recent = 0; 1724 } else { 1725 tcpstat.tcps_rcvduppack++; 1726 tcpstat.tcps_rcvdupbyte += tlen; 1727 tcpstat.tcps_pawsdrop++; 1728 if (tlen) 1729 goto dropafterack; 1730 goto drop; 1731 } 1732 } 1733 1734 /* 1735 * In the SYN-RECEIVED state, validate that the packet belongs to 1736 * this connection before trimming the data to fit the receive 1737 * window. Check the sequence number versus IRS since we know 1738 * the sequence numbers haven't wrapped. This is a partial fix 1739 * for the "LAND" DoS attack. 1740 */ 1741 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1742 rstreason = BANDLIM_RST_OPENPORT; 1743 goto dropwithreset; 1744 } 1745 1746 todrop = tp->rcv_nxt - th->th_seq; 1747 if (todrop > 0) { 1748 if (thflags & TH_SYN) { 1749 thflags &= ~TH_SYN; 1750 th->th_seq++; 1751 if (th->th_urp > 1) 1752 th->th_urp--; 1753 else 1754 thflags &= ~TH_URG; 1755 todrop--; 1756 } 1757 /* 1758 * Following if statement from Stevens, vol. 2, p. 960. 1759 */ 1760 if (todrop > tlen 1761 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1762 /* 1763 * Any valid FIN must be to the left of the window. 1764 * At this point the FIN must be a duplicate or out 1765 * of sequence; drop it. 1766 */ 1767 thflags &= ~TH_FIN; 1768 1769 /* 1770 * Send an ACK to resynchronize and drop any data. 1771 * But keep on processing for RST or ACK. 1772 */ 1773 tp->t_flags |= TF_ACKNOW; 1774 todrop = tlen; 1775 tcpstat.tcps_rcvduppack++; 1776 tcpstat.tcps_rcvdupbyte += todrop; 1777 } else { 1778 tcpstat.tcps_rcvpartduppack++; 1779 tcpstat.tcps_rcvpartdupbyte += todrop; 1780 } 1781 drop_hdrlen += todrop; /* drop from the top afterwards */ 1782 th->th_seq += todrop; 1783 tlen -= todrop; 1784 if (th->th_urp > todrop) 1785 th->th_urp -= todrop; 1786 else { 1787 thflags &= ~TH_URG; 1788 th->th_urp = 0; 1789 } 1790 } 1791 1792 /* 1793 * If new data are received on a connection after the 1794 * user processes are gone, then RST the other end. 1795 */ 1796 if ((so->so_state & SS_NOFDREF) && 1797 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1798 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not " 1799 "locked")); 1800 tp = tcp_close(tp); 1801 tcpstat.tcps_rcvafterclose++; 1802 rstreason = BANDLIM_UNLIMITED; 1803 goto dropwithreset; 1804 } 1805 1806 /* 1807 * If segment ends after window, drop trailing data 1808 * (and PUSH and FIN); if nothing left, just ACK. 1809 */ 1810 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1811 if (todrop > 0) { 1812 tcpstat.tcps_rcvpackafterwin++; 1813 if (todrop >= tlen) { 1814 tcpstat.tcps_rcvbyteafterwin += tlen; 1815 /* 1816 * If a new connection request is received 1817 * while in TIME_WAIT, drop the old connection 1818 * and start over if the sequence numbers 1819 * are above the previous ones. 1820 */ 1821 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1822 if (thflags & TH_SYN && 1823 tp->t_state == TCPS_TIME_WAIT && 1824 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1825 KASSERT(headlocked, ("trimthenstep6: " 1826 "tcp_close.4: head not locked")); 1827 tp = tcp_close(tp); 1828 goto findpcb; 1829 } 1830 /* 1831 * If window is closed can only take segments at 1832 * window edge, and have to drop data and PUSH from 1833 * incoming segments. Continue processing, but 1834 * remember to ack. Otherwise, drop segment 1835 * and ack. 1836 */ 1837 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1838 tp->t_flags |= TF_ACKNOW; 1839 tcpstat.tcps_rcvwinprobe++; 1840 } else 1841 goto dropafterack; 1842 } else 1843 tcpstat.tcps_rcvbyteafterwin += todrop; 1844 m_adj(m, -todrop); 1845 tlen -= todrop; 1846 thflags &= ~(TH_PUSH|TH_FIN); 1847 } 1848 1849 /* 1850 * If last ACK falls within this segment's sequence numbers, 1851 * record its timestamp. 1852 * NOTE: 1853 * 1) That the test incorporates suggestions from the latest 1854 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1855 * 2) That updating only on newer timestamps interferes with 1856 * our earlier PAWS tests, so this check should be solely 1857 * predicated on the sequence space of this segment. 1858 * 3) That we modify the segment boundary check to be 1859 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1860 * instead of RFC1323's 1861 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1862 * This modified check allows us to overcome RFC1323's 1863 * limitations as described in Stevens TCP/IP Illustrated 1864 * Vol. 2 p.869. In such cases, we can still calculate the 1865 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1866 */ 1867 if ((to.to_flags & TOF_TS) != 0 && 1868 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1869 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1870 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1871 tp->ts_recent_age = ticks; 1872 tp->ts_recent = to.to_tsval; 1873 } 1874 1875 /* 1876 * If a SYN is in the window, then this is an 1877 * error and we send an RST and drop the connection. 1878 */ 1879 if (thflags & TH_SYN) { 1880 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: " 1881 "head not locked")); 1882 tp = tcp_drop(tp, ECONNRESET); 1883 rstreason = BANDLIM_UNLIMITED; 1884 goto drop; 1885 } 1886 1887 /* 1888 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1889 * flag is on (half-synchronized state), then queue data for 1890 * later processing; else drop segment and return. 1891 */ 1892 if ((thflags & TH_ACK) == 0) { 1893 if (tp->t_state == TCPS_SYN_RECEIVED || 1894 (tp->t_flags & TF_NEEDSYN)) 1895 goto step6; 1896 else if (tp->t_flags & TF_ACKNOW) 1897 goto dropafterack; 1898 else 1899 goto drop; 1900 } 1901 1902 /* 1903 * Ack processing. 1904 */ 1905 switch (tp->t_state) { 1906 1907 /* 1908 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1909 * ESTABLISHED state and continue processing. 1910 * The ACK was checked above. 1911 */ 1912 case TCPS_SYN_RECEIVED: 1913 1914 tcpstat.tcps_connects++; 1915 soisconnected(so); 1916 /* Do window scaling? */ 1917 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1918 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1919 tp->rcv_scale = tp->request_r_scale; 1920 tp->snd_wnd = tiwin; 1921 } 1922 /* 1923 * Make transitions: 1924 * SYN-RECEIVED -> ESTABLISHED 1925 * SYN-RECEIVED* -> FIN-WAIT-1 1926 */ 1927 tp->t_starttime = ticks; 1928 if (tp->t_flags & TF_NEEDFIN) { 1929 tp->t_state = TCPS_FIN_WAIT_1; 1930 tp->t_flags &= ~TF_NEEDFIN; 1931 } else { 1932 tp->t_state = TCPS_ESTABLISHED; 1933 callout_reset(tp->tt_keep, tcp_keepidle, 1934 tcp_timer_keep, tp); 1935 } 1936 /* 1937 * If segment contains data or ACK, will call tcp_reass() 1938 * later; if not, do so now to pass queued data to user. 1939 */ 1940 if (tlen == 0 && (thflags & TH_FIN) == 0) 1941 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1942 (struct mbuf *)0); 1943 tp->snd_wl1 = th->th_seq - 1; 1944 /* FALLTHROUGH */ 1945 1946 /* 1947 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1948 * ACKs. If the ack is in the range 1949 * tp->snd_una < th->th_ack <= tp->snd_max 1950 * then advance tp->snd_una to th->th_ack and drop 1951 * data from the retransmission queue. If this ACK reflects 1952 * more up to date window information we update our window information. 1953 */ 1954 case TCPS_ESTABLISHED: 1955 case TCPS_FIN_WAIT_1: 1956 case TCPS_FIN_WAIT_2: 1957 case TCPS_CLOSE_WAIT: 1958 case TCPS_CLOSING: 1959 case TCPS_LAST_ACK: 1960 case TCPS_TIME_WAIT: 1961 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1962 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1963 tcpstat.tcps_rcvacktoomuch++; 1964 goto dropafterack; 1965 } 1966 if (tp->sack_enable && 1967 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) 1968 tcp_sack_doack(tp, &to, th->th_ack); 1969 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1970 if (tlen == 0 && tiwin == tp->snd_wnd) { 1971 tcpstat.tcps_rcvdupack++; 1972 /* 1973 * If we have outstanding data (other than 1974 * a window probe), this is a completely 1975 * duplicate ack (ie, window info didn't 1976 * change), the ack is the biggest we've 1977 * seen and we've seen exactly our rexmt 1978 * threshhold of them, assume a packet 1979 * has been dropped and retransmit it. 1980 * Kludge snd_nxt & the congestion 1981 * window so we send only this one 1982 * packet. 1983 * 1984 * We know we're losing at the current 1985 * window size so do congestion avoidance 1986 * (set ssthresh to half the current window 1987 * and pull our congestion window back to 1988 * the new ssthresh). 1989 * 1990 * Dup acks mean that packets have left the 1991 * network (they're now cached at the receiver) 1992 * so bump cwnd by the amount in the receiver 1993 * to keep a constant cwnd packets in the 1994 * network. 1995 */ 1996 if (!callout_active(tp->tt_rexmt) || 1997 th->th_ack != tp->snd_una) 1998 tp->t_dupacks = 0; 1999 else if (++tp->t_dupacks > tcprexmtthresh || 2000 ((tcp_do_newreno || tp->sack_enable) && 2001 IN_FASTRECOVERY(tp))) { 2002 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 2003 int awnd; 2004 2005 /* 2006 * Compute the amount of data in flight first. 2007 * We can inject new data into the pipe iff 2008 * we have less than 1/2 the original window's 2009 * worth of data in flight. 2010 */ 2011 awnd = (tp->snd_nxt - tp->snd_fack) + 2012 tp->sackhint.sack_bytes_rexmit; 2013 if (awnd < tp->snd_ssthresh) { 2014 tp->snd_cwnd += tp->t_maxseg; 2015 if (tp->snd_cwnd > tp->snd_ssthresh) 2016 tp->snd_cwnd = tp->snd_ssthresh; 2017 } 2018 } else 2019 tp->snd_cwnd += tp->t_maxseg; 2020 (void) tcp_output(tp); 2021 goto drop; 2022 } else if (tp->t_dupacks == tcprexmtthresh) { 2023 tcp_seq onxt = tp->snd_nxt; 2024 u_int win; 2025 2026 /* 2027 * If we're doing sack, check to 2028 * see if we're already in sack 2029 * recovery. If we're not doing sack, 2030 * check to see if we're in newreno 2031 * recovery. 2032 */ 2033 if (tp->sack_enable) { 2034 if (IN_FASTRECOVERY(tp)) { 2035 tp->t_dupacks = 0; 2036 break; 2037 } 2038 } else if (tcp_do_newreno) { 2039 if (SEQ_LEQ(th->th_ack, 2040 tp->snd_recover)) { 2041 tp->t_dupacks = 0; 2042 break; 2043 } 2044 } 2045 win = min(tp->snd_wnd, tp->snd_cwnd) / 2046 2 / tp->t_maxseg; 2047 if (win < 2) 2048 win = 2; 2049 tp->snd_ssthresh = win * tp->t_maxseg; 2050 ENTER_FASTRECOVERY(tp); 2051 tp->snd_recover = tp->snd_max; 2052 callout_stop(tp->tt_rexmt); 2053 tp->t_rtttime = 0; 2054 if (tp->sack_enable) { 2055 tcpstat.tcps_sack_recovery_episode++; 2056 tp->sack_newdata = tp->snd_nxt; 2057 tp->snd_cwnd = tp->t_maxseg; 2058 (void) tcp_output(tp); 2059 goto drop; 2060 } 2061 tp->snd_nxt = th->th_ack; 2062 tp->snd_cwnd = tp->t_maxseg; 2063 (void) tcp_output(tp); 2064 KASSERT(tp->snd_limited <= 2, 2065 ("tp->snd_limited too big")); 2066 tp->snd_cwnd = tp->snd_ssthresh + 2067 tp->t_maxseg * 2068 (tp->t_dupacks - tp->snd_limited); 2069 if (SEQ_GT(onxt, tp->snd_nxt)) 2070 tp->snd_nxt = onxt; 2071 goto drop; 2072 } else if (tcp_do_rfc3042) { 2073 u_long oldcwnd = tp->snd_cwnd; 2074 tcp_seq oldsndmax = tp->snd_max; 2075 u_int sent; 2076 2077 KASSERT(tp->t_dupacks == 1 || 2078 tp->t_dupacks == 2, 2079 ("dupacks not 1 or 2")); 2080 if (tp->t_dupacks == 1) 2081 tp->snd_limited = 0; 2082 tp->snd_cwnd = 2083 (tp->snd_nxt - tp->snd_una) + 2084 (tp->t_dupacks - tp->snd_limited) * 2085 tp->t_maxseg; 2086 (void) tcp_output(tp); 2087 sent = tp->snd_max - oldsndmax; 2088 if (sent > tp->t_maxseg) { 2089 KASSERT((tp->t_dupacks == 2 && 2090 tp->snd_limited == 0) || 2091 (sent == tp->t_maxseg + 1 && 2092 tp->t_flags & TF_SENTFIN), 2093 ("sent too much")); 2094 tp->snd_limited = 2; 2095 } else if (sent > 0) 2096 ++tp->snd_limited; 2097 tp->snd_cwnd = oldcwnd; 2098 goto drop; 2099 } 2100 } else 2101 tp->t_dupacks = 0; 2102 break; 2103 } 2104 2105 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 2106 2107 /* 2108 * If the congestion window was inflated to account 2109 * for the other side's cached packets, retract it. 2110 */ 2111 if (tcp_do_newreno || tp->sack_enable) { 2112 if (IN_FASTRECOVERY(tp)) { 2113 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2114 if (tp->sack_enable) 2115 tcp_sack_partialack(tp, th); 2116 else 2117 tcp_newreno_partial_ack(tp, th); 2118 } else { 2119 /* 2120 * Out of fast recovery. 2121 * Window inflation should have left us 2122 * with approximately snd_ssthresh 2123 * outstanding data. 2124 * But in case we would be inclined to 2125 * send a burst, better to do it via 2126 * the slow start mechanism. 2127 */ 2128 if (SEQ_GT(th->th_ack + 2129 tp->snd_ssthresh, 2130 tp->snd_max)) 2131 tp->snd_cwnd = tp->snd_max - 2132 th->th_ack + 2133 tp->t_maxseg; 2134 else 2135 tp->snd_cwnd = tp->snd_ssthresh; 2136 } 2137 } 2138 } else { 2139 if (tp->t_dupacks >= tcprexmtthresh && 2140 tp->snd_cwnd > tp->snd_ssthresh) 2141 tp->snd_cwnd = tp->snd_ssthresh; 2142 } 2143 tp->t_dupacks = 0; 2144 /* 2145 * If we reach this point, ACK is not a duplicate, 2146 * i.e., it ACKs something we sent. 2147 */ 2148 if (tp->t_flags & TF_NEEDSYN) { 2149 /* 2150 * T/TCP: Connection was half-synchronized, and our 2151 * SYN has been ACK'd (so connection is now fully 2152 * synchronized). Go to non-starred state, 2153 * increment snd_una for ACK of SYN, and check if 2154 * we can do window scaling. 2155 */ 2156 tp->t_flags &= ~TF_NEEDSYN; 2157 tp->snd_una++; 2158 /* Do window scaling? */ 2159 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2160 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2161 tp->rcv_scale = tp->request_r_scale; 2162 /* Send window already scaled. */ 2163 } 2164 } 2165 2166process_ACK: 2167 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 2168 "locked")); 2169 INP_LOCK_ASSERT(inp); 2170 2171 acked = th->th_ack - tp->snd_una; 2172 tcpstat.tcps_rcvackpack++; 2173 tcpstat.tcps_rcvackbyte += acked; 2174 2175 /* 2176 * If we just performed our first retransmit, and the ACK 2177 * arrives within our recovery window, then it was a mistake 2178 * to do the retransmit in the first place. Recover our 2179 * original cwnd and ssthresh, and proceed to transmit where 2180 * we left off. 2181 */ 2182 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2183 ++tcpstat.tcps_sndrexmitbad; 2184 tp->snd_cwnd = tp->snd_cwnd_prev; 2185 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2186 tp->snd_recover = tp->snd_recover_prev; 2187 if (tp->t_flags & TF_WASFRECOVERY) 2188 ENTER_FASTRECOVERY(tp); 2189 tp->snd_nxt = tp->snd_max; 2190 tp->t_badrxtwin = 0; /* XXX probably not required */ 2191 } 2192 2193 /* 2194 * If we have a timestamp reply, update smoothed 2195 * round trip time. If no timestamp is present but 2196 * transmit timer is running and timed sequence 2197 * number was acked, update smoothed round trip time. 2198 * Since we now have an rtt measurement, cancel the 2199 * timer backoff (cf., Phil Karn's retransmit alg.). 2200 * Recompute the initial retransmit timer. 2201 * 2202 * Some boxes send broken timestamp replies 2203 * during the SYN+ACK phase, ignore 2204 * timestamps of 0 or we could calculate a 2205 * huge RTT and blow up the retransmit timer. 2206 */ 2207 if ((to.to_flags & TOF_TS) != 0 && 2208 to.to_tsecr) { 2209 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2210 tp->t_rttlow = ticks - to.to_tsecr; 2211 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2212 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2213 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2214 tp->t_rttlow = ticks - tp->t_rtttime; 2215 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2216 } 2217 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2218 2219 /* 2220 * If all outstanding data is acked, stop retransmit 2221 * timer and remember to restart (more output or persist). 2222 * If there is more data to be acked, restart retransmit 2223 * timer, using current (possibly backed-off) value. 2224 */ 2225 if (th->th_ack == tp->snd_max) { 2226 callout_stop(tp->tt_rexmt); 2227 needoutput = 1; 2228 } else if (!callout_active(tp->tt_persist)) 2229 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2230 tcp_timer_rexmt, tp); 2231 2232 /* 2233 * If no data (only SYN) was ACK'd, 2234 * skip rest of ACK processing. 2235 */ 2236 if (acked == 0) 2237 goto step6; 2238 2239 /* 2240 * When new data is acked, open the congestion window. 2241 * If the window gives us less than ssthresh packets 2242 * in flight, open exponentially (maxseg per packet). 2243 * Otherwise open linearly: maxseg per window 2244 * (maxseg^2 / cwnd per packet). 2245 */ 2246 if ((!tcp_do_newreno && !tp->sack_enable) || 2247 !IN_FASTRECOVERY(tp)) { 2248 register u_int cw = tp->snd_cwnd; 2249 register u_int incr = tp->t_maxseg; 2250 if (cw > tp->snd_ssthresh) 2251 incr = incr * incr / cw; 2252 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2253 } 2254 SOCKBUF_LOCK(&so->so_snd); 2255 if (acked > so->so_snd.sb_cc) { 2256 tp->snd_wnd -= so->so_snd.sb_cc; 2257 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2258 ourfinisacked = 1; 2259 } else { 2260 sbdrop_locked(&so->so_snd, acked); 2261 tp->snd_wnd -= acked; 2262 ourfinisacked = 0; 2263 } 2264 sowwakeup_locked(so); 2265 /* detect una wraparound */ 2266 if ((tcp_do_newreno || tp->sack_enable) && 2267 !IN_FASTRECOVERY(tp) && 2268 SEQ_GT(tp->snd_una, tp->snd_recover) && 2269 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2270 tp->snd_recover = th->th_ack - 1; 2271 if ((tcp_do_newreno || tp->sack_enable) && 2272 IN_FASTRECOVERY(tp) && 2273 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2274 EXIT_FASTRECOVERY(tp); 2275 tp->snd_una = th->th_ack; 2276 if (tp->sack_enable) { 2277 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2278 tp->snd_recover = tp->snd_una; 2279 } 2280 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2281 tp->snd_nxt = tp->snd_una; 2282 2283 switch (tp->t_state) { 2284 2285 /* 2286 * In FIN_WAIT_1 STATE in addition to the processing 2287 * for the ESTABLISHED state if our FIN is now acknowledged 2288 * then enter FIN_WAIT_2. 2289 */ 2290 case TCPS_FIN_WAIT_1: 2291 if (ourfinisacked) { 2292 /* 2293 * If we can't receive any more 2294 * data, then closing user can proceed. 2295 * Starting the timer is contrary to the 2296 * specification, but if we don't get a FIN 2297 * we'll hang forever. 2298 */ 2299 /* XXXjl 2300 * we should release the tp also, and use a 2301 * compressed state. 2302 */ 2303 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2304 soisdisconnected(so); 2305 callout_reset(tp->tt_2msl, tcp_maxidle, 2306 tcp_timer_2msl, tp); 2307 } 2308 tp->t_state = TCPS_FIN_WAIT_2; 2309 } 2310 break; 2311 2312 /* 2313 * In CLOSING STATE in addition to the processing for 2314 * the ESTABLISHED state if the ACK acknowledges our FIN 2315 * then enter the TIME-WAIT state, otherwise ignore 2316 * the segment. 2317 */ 2318 case TCPS_CLOSING: 2319 if (ourfinisacked) { 2320 KASSERT(headlocked, ("tcp_input: process_ACK: " 2321 "head not locked")); 2322 tcp_twstart(tp); 2323 INP_INFO_WUNLOCK(&tcbinfo); 2324 m_freem(m); 2325 return; 2326 } 2327 break; 2328 2329 /* 2330 * In LAST_ACK, we may still be waiting for data to drain 2331 * and/or to be acked, as well as for the ack of our FIN. 2332 * If our FIN is now acknowledged, delete the TCB, 2333 * enter the closed state and return. 2334 */ 2335 case TCPS_LAST_ACK: 2336 if (ourfinisacked) { 2337 KASSERT(headlocked, ("tcp_input: process_ACK:" 2338 " tcp_close: head not locked")); 2339 tp = tcp_close(tp); 2340 goto drop; 2341 } 2342 break; 2343 2344 /* 2345 * In TIME_WAIT state the only thing that should arrive 2346 * is a retransmission of the remote FIN. Acknowledge 2347 * it and restart the finack timer. 2348 */ 2349 case TCPS_TIME_WAIT: 2350 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2351 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2352 tcp_timer_2msl, tp); 2353 goto dropafterack; 2354 } 2355 } 2356 2357step6: 2358 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2359 INP_LOCK_ASSERT(inp); 2360 2361 /* 2362 * Update window information. 2363 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2364 */ 2365 if ((thflags & TH_ACK) && 2366 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2367 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2368 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2369 /* keep track of pure window updates */ 2370 if (tlen == 0 && 2371 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2372 tcpstat.tcps_rcvwinupd++; 2373 tp->snd_wnd = tiwin; 2374 tp->snd_wl1 = th->th_seq; 2375 tp->snd_wl2 = th->th_ack; 2376 if (tp->snd_wnd > tp->max_sndwnd) 2377 tp->max_sndwnd = tp->snd_wnd; 2378 needoutput = 1; 2379 } 2380 2381 /* 2382 * Process segments with URG. 2383 */ 2384 if ((thflags & TH_URG) && th->th_urp && 2385 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2386 /* 2387 * This is a kludge, but if we receive and accept 2388 * random urgent pointers, we'll crash in 2389 * soreceive. It's hard to imagine someone 2390 * actually wanting to send this much urgent data. 2391 */ 2392 SOCKBUF_LOCK(&so->so_rcv); 2393 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2394 th->th_urp = 0; /* XXX */ 2395 thflags &= ~TH_URG; /* XXX */ 2396 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2397 goto dodata; /* XXX */ 2398 } 2399 /* 2400 * If this segment advances the known urgent pointer, 2401 * then mark the data stream. This should not happen 2402 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2403 * a FIN has been received from the remote side. 2404 * In these states we ignore the URG. 2405 * 2406 * According to RFC961 (Assigned Protocols), 2407 * the urgent pointer points to the last octet 2408 * of urgent data. We continue, however, 2409 * to consider it to indicate the first octet 2410 * of data past the urgent section as the original 2411 * spec states (in one of two places). 2412 */ 2413 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2414 tp->rcv_up = th->th_seq + th->th_urp; 2415 so->so_oobmark = so->so_rcv.sb_cc + 2416 (tp->rcv_up - tp->rcv_nxt) - 1; 2417 if (so->so_oobmark == 0) 2418 so->so_rcv.sb_state |= SBS_RCVATMARK; 2419 sohasoutofband(so); 2420 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2421 } 2422 SOCKBUF_UNLOCK(&so->so_rcv); 2423 /* 2424 * Remove out of band data so doesn't get presented to user. 2425 * This can happen independent of advancing the URG pointer, 2426 * but if two URG's are pending at once, some out-of-band 2427 * data may creep in... ick. 2428 */ 2429 if (th->th_urp <= (u_long)tlen && 2430 !(so->so_options & SO_OOBINLINE)) { 2431 /* hdr drop is delayed */ 2432 tcp_pulloutofband(so, th, m, drop_hdrlen); 2433 } 2434 } else { 2435 /* 2436 * If no out of band data is expected, 2437 * pull receive urgent pointer along 2438 * with the receive window. 2439 */ 2440 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2441 tp->rcv_up = tp->rcv_nxt; 2442 } 2443dodata: /* XXX */ 2444 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2445 INP_LOCK_ASSERT(inp); 2446 2447 /* 2448 * Process the segment text, merging it into the TCP sequencing queue, 2449 * and arranging for acknowledgment of receipt if necessary. 2450 * This process logically involves adjusting tp->rcv_wnd as data 2451 * is presented to the user (this happens in tcp_usrreq.c, 2452 * case PRU_RCVD). If a FIN has already been received on this 2453 * connection then we just ignore the text. 2454 */ 2455 if ((tlen || (thflags & TH_FIN)) && 2456 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2457 tcp_seq save_start = th->th_seq; 2458 tcp_seq save_end = th->th_seq + tlen; 2459 m_adj(m, drop_hdrlen); /* delayed header drop */ 2460 /* 2461 * Insert segment which includes th into TCP reassembly queue 2462 * with control block tp. Set thflags to whether reassembly now 2463 * includes a segment with FIN. This handles the common case 2464 * inline (segment is the next to be received on an established 2465 * connection, and the queue is empty), avoiding linkage into 2466 * and removal from the queue and repetition of various 2467 * conversions. 2468 * Set DELACK for segments received in order, but ack 2469 * immediately when segments are out of order (so 2470 * fast retransmit can work). 2471 */ 2472 if (th->th_seq == tp->rcv_nxt && 2473 LIST_EMPTY(&tp->t_segq) && 2474 TCPS_HAVEESTABLISHED(tp->t_state)) { 2475 if (DELAY_ACK(tp)) 2476 tp->t_flags |= TF_DELACK; 2477 else 2478 tp->t_flags |= TF_ACKNOW; 2479 tp->rcv_nxt += tlen; 2480 thflags = th->th_flags & TH_FIN; 2481 tcpstat.tcps_rcvpack++; 2482 tcpstat.tcps_rcvbyte += tlen; 2483 ND6_HINT(tp); 2484 SOCKBUF_LOCK(&so->so_rcv); 2485 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2486 m_freem(m); 2487 else 2488 sbappendstream_locked(&so->so_rcv, m); 2489 sorwakeup_locked(so); 2490 } else { 2491 thflags = tcp_reass(tp, th, &tlen, m); 2492 tp->t_flags |= TF_ACKNOW; 2493 } 2494 if (tlen > 0 && tp->sack_enable) 2495 tcp_update_sack_list(tp, save_start, save_end); 2496 /* 2497 * Note the amount of data that peer has sent into 2498 * our window, in order to estimate the sender's 2499 * buffer size. 2500 */ 2501 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2502 } else { 2503 m_freem(m); 2504 thflags &= ~TH_FIN; 2505 } 2506 2507 /* 2508 * If FIN is received ACK the FIN and let the user know 2509 * that the connection is closing. 2510 */ 2511 if (thflags & TH_FIN) { 2512 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2513 socantrcvmore(so); 2514 /* 2515 * If connection is half-synchronized 2516 * (ie NEEDSYN flag on) then delay ACK, 2517 * so it may be piggybacked when SYN is sent. 2518 * Otherwise, since we received a FIN then no 2519 * more input can be expected, send ACK now. 2520 */ 2521 if (tp->t_flags & TF_NEEDSYN) 2522 tp->t_flags |= TF_DELACK; 2523 else 2524 tp->t_flags |= TF_ACKNOW; 2525 tp->rcv_nxt++; 2526 } 2527 switch (tp->t_state) { 2528 2529 /* 2530 * In SYN_RECEIVED and ESTABLISHED STATES 2531 * enter the CLOSE_WAIT state. 2532 */ 2533 case TCPS_SYN_RECEIVED: 2534 tp->t_starttime = ticks; 2535 /*FALLTHROUGH*/ 2536 case TCPS_ESTABLISHED: 2537 tp->t_state = TCPS_CLOSE_WAIT; 2538 break; 2539 2540 /* 2541 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2542 * enter the CLOSING state. 2543 */ 2544 case TCPS_FIN_WAIT_1: 2545 tp->t_state = TCPS_CLOSING; 2546 break; 2547 2548 /* 2549 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2550 * starting the time-wait timer, turning off the other 2551 * standard timers. 2552 */ 2553 case TCPS_FIN_WAIT_2: 2554 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2555 "TCP_FIN_WAIT_2: head not locked")); 2556 tcp_twstart(tp); 2557 INP_INFO_WUNLOCK(&tcbinfo); 2558 return; 2559 2560 /* 2561 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2562 */ 2563 case TCPS_TIME_WAIT: 2564 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2565 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2566 tcp_timer_2msl, tp); 2567 break; 2568 } 2569 } 2570 INP_INFO_WUNLOCK(&tcbinfo); 2571 headlocked = 0; 2572#ifdef TCPDEBUG 2573 if (so->so_options & SO_DEBUG) 2574 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2575 &tcp_savetcp, 0); 2576#endif 2577 2578 /* 2579 * Return any desired output. 2580 */ 2581 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2582 (void) tcp_output(tp); 2583 2584check_delack: 2585 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2586 INP_LOCK_ASSERT(inp); 2587 if (tp->t_flags & TF_DELACK) { 2588 tp->t_flags &= ~TF_DELACK; 2589 callout_reset(tp->tt_delack, tcp_delacktime, 2590 tcp_timer_delack, tp); 2591 } 2592 INP_UNLOCK(inp); 2593 return; 2594 2595dropafterack: 2596 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2597 /* 2598 * Generate an ACK dropping incoming segment if it occupies 2599 * sequence space, where the ACK reflects our state. 2600 * 2601 * We can now skip the test for the RST flag since all 2602 * paths to this code happen after packets containing 2603 * RST have been dropped. 2604 * 2605 * In the SYN-RECEIVED state, don't send an ACK unless the 2606 * segment we received passes the SYN-RECEIVED ACK test. 2607 * If it fails send a RST. This breaks the loop in the 2608 * "LAND" DoS attack, and also prevents an ACK storm 2609 * between two listening ports that have been sent forged 2610 * SYN segments, each with the source address of the other. 2611 */ 2612 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2613 (SEQ_GT(tp->snd_una, th->th_ack) || 2614 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2615 rstreason = BANDLIM_RST_OPENPORT; 2616 goto dropwithreset; 2617 } 2618#ifdef TCPDEBUG 2619 if (so->so_options & SO_DEBUG) 2620 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2621 &tcp_savetcp, 0); 2622#endif 2623 KASSERT(headlocked, ("headlocked should be 1")); 2624 INP_INFO_WUNLOCK(&tcbinfo); 2625 tp->t_flags |= TF_ACKNOW; 2626 (void) tcp_output(tp); 2627 INP_UNLOCK(inp); 2628 m_freem(m); 2629 return; 2630 2631dropwithreset: 2632 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2633 /* 2634 * Generate a RST, dropping incoming segment. 2635 * Make ACK acceptable to originator of segment. 2636 * Don't bother to respond if destination was broadcast/multicast. 2637 */ 2638 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2639 goto drop; 2640 if (isipv6) { 2641 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2642 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2643 goto drop; 2644 } else { 2645 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2646 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2647 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2648 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2649 goto drop; 2650 } 2651 /* IPv6 anycast check is done at tcp6_input() */ 2652 2653 /* 2654 * Perform bandwidth limiting. 2655 */ 2656 if (badport_bandlim(rstreason) < 0) 2657 goto drop; 2658 2659#ifdef TCPDEBUG 2660 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2661 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2662 &tcp_savetcp, 0); 2663#endif 2664 2665 if (thflags & TH_ACK) 2666 /* mtod() below is safe as long as hdr dropping is delayed */ 2667 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2668 TH_RST); 2669 else { 2670 if (thflags & TH_SYN) 2671 tlen++; 2672 /* mtod() below is safe as long as hdr dropping is delayed */ 2673 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2674 (tcp_seq)0, TH_RST|TH_ACK); 2675 } 2676 2677 if (tp != NULL) 2678 INP_UNLOCK(inp); 2679 if (headlocked) 2680 INP_INFO_WUNLOCK(&tcbinfo); 2681 return; 2682 2683drop: 2684 /* 2685 * Drop space held by incoming segment and return. 2686 */ 2687#ifdef TCPDEBUG 2688 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2689 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2690 &tcp_savetcp, 0); 2691#endif 2692 if (tp != NULL) 2693 INP_UNLOCK(inp); 2694 if (headlocked) 2695 INP_INFO_WUNLOCK(&tcbinfo); 2696 m_freem(m); 2697 return; 2698} 2699 2700/* 2701 * Parse TCP options and place in tcpopt. 2702 */ 2703static void 2704tcp_dooptions(to, cp, cnt, flags) 2705 struct tcpopt *to; 2706 u_char *cp; 2707 int cnt; 2708 int flags; 2709{ 2710 int opt, optlen; 2711 2712 to->to_flags = 0; 2713 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2714 opt = cp[0]; 2715 if (opt == TCPOPT_EOL) 2716 break; 2717 if (opt == TCPOPT_NOP) 2718 optlen = 1; 2719 else { 2720 if (cnt < 2) 2721 break; 2722 optlen = cp[1]; 2723 if (optlen < 2 || optlen > cnt) 2724 break; 2725 } 2726 switch (opt) { 2727 case TCPOPT_MAXSEG: 2728 if (optlen != TCPOLEN_MAXSEG) 2729 continue; 2730 if (!(flags & TO_SYN)) 2731 continue; 2732 to->to_flags |= TOF_MSS; 2733 bcopy((char *)cp + 2, 2734 (char *)&to->to_mss, sizeof(to->to_mss)); 2735 to->to_mss = ntohs(to->to_mss); 2736 break; 2737 case TCPOPT_WINDOW: 2738 if (optlen != TCPOLEN_WINDOW) 2739 continue; 2740 if (!(flags & TO_SYN)) 2741 continue; 2742 to->to_flags |= TOF_SCALE; 2743 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2744 break; 2745 case TCPOPT_TIMESTAMP: 2746 if (optlen != TCPOLEN_TIMESTAMP) 2747 continue; 2748 to->to_flags |= TOF_TS; 2749 bcopy((char *)cp + 2, 2750 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2751 to->to_tsval = ntohl(to->to_tsval); 2752 bcopy((char *)cp + 6, 2753 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2754 to->to_tsecr = ntohl(to->to_tsecr); 2755 break; 2756#ifdef TCP_SIGNATURE 2757 /* 2758 * XXX In order to reply to a host which has set the 2759 * TCP_SIGNATURE option in its initial SYN, we have to 2760 * record the fact that the option was observed here 2761 * for the syncache code to perform the correct response. 2762 */ 2763 case TCPOPT_SIGNATURE: 2764 if (optlen != TCPOLEN_SIGNATURE) 2765 continue; 2766 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2767 break; 2768#endif 2769 case TCPOPT_SACK_PERMITTED: 2770 if (optlen != TCPOLEN_SACK_PERMITTED) 2771 continue; 2772 if (!(flags & TO_SYN)) 2773 continue; 2774 if (!tcp_do_sack) 2775 continue; 2776 to->to_flags |= TOF_SACK; 2777 break; 2778 case TCPOPT_SACK: 2779 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2780 continue; 2781 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2782 to->to_sacks = cp + 2; 2783 tcpstat.tcps_sack_rcv_blocks++; 2784 break; 2785 default: 2786 continue; 2787 } 2788 } 2789} 2790 2791/* 2792 * Pull out of band byte out of a segment so 2793 * it doesn't appear in the user's data queue. 2794 * It is still reflected in the segment length for 2795 * sequencing purposes. 2796 */ 2797static void 2798tcp_pulloutofband(so, th, m, off) 2799 struct socket *so; 2800 struct tcphdr *th; 2801 register struct mbuf *m; 2802 int off; /* delayed to be droped hdrlen */ 2803{ 2804 int cnt = off + th->th_urp - 1; 2805 2806 while (cnt >= 0) { 2807 if (m->m_len > cnt) { 2808 char *cp = mtod(m, caddr_t) + cnt; 2809 struct tcpcb *tp = sototcpcb(so); 2810 2811 tp->t_iobc = *cp; 2812 tp->t_oobflags |= TCPOOB_HAVEDATA; 2813 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2814 m->m_len--; 2815 if (m->m_flags & M_PKTHDR) 2816 m->m_pkthdr.len--; 2817 return; 2818 } 2819 cnt -= m->m_len; 2820 m = m->m_next; 2821 if (m == 0) 2822 break; 2823 } 2824 panic("tcp_pulloutofband"); 2825} 2826 2827/* 2828 * Collect new round-trip time estimate 2829 * and update averages and current timeout. 2830 */ 2831static void 2832tcp_xmit_timer(tp, rtt) 2833 register struct tcpcb *tp; 2834 int rtt; 2835{ 2836 register int delta; 2837 2838 INP_LOCK_ASSERT(tp->t_inpcb); 2839 2840 tcpstat.tcps_rttupdated++; 2841 tp->t_rttupdated++; 2842 if (tp->t_srtt != 0) { 2843 /* 2844 * srtt is stored as fixed point with 5 bits after the 2845 * binary point (i.e., scaled by 8). The following magic 2846 * is equivalent to the smoothing algorithm in rfc793 with 2847 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2848 * point). Adjust rtt to origin 0. 2849 */ 2850 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2851 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2852 2853 if ((tp->t_srtt += delta) <= 0) 2854 tp->t_srtt = 1; 2855 2856 /* 2857 * We accumulate a smoothed rtt variance (actually, a 2858 * smoothed mean difference), then set the retransmit 2859 * timer to smoothed rtt + 4 times the smoothed variance. 2860 * rttvar is stored as fixed point with 4 bits after the 2861 * binary point (scaled by 16). The following is 2862 * equivalent to rfc793 smoothing with an alpha of .75 2863 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2864 * rfc793's wired-in beta. 2865 */ 2866 if (delta < 0) 2867 delta = -delta; 2868 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2869 if ((tp->t_rttvar += delta) <= 0) 2870 tp->t_rttvar = 1; 2871 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2872 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2873 } else { 2874 /* 2875 * No rtt measurement yet - use the unsmoothed rtt. 2876 * Set the variance to half the rtt (so our first 2877 * retransmit happens at 3*rtt). 2878 */ 2879 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2880 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2881 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2882 } 2883 tp->t_rtttime = 0; 2884 tp->t_rxtshift = 0; 2885 2886 /* 2887 * the retransmit should happen at rtt + 4 * rttvar. 2888 * Because of the way we do the smoothing, srtt and rttvar 2889 * will each average +1/2 tick of bias. When we compute 2890 * the retransmit timer, we want 1/2 tick of rounding and 2891 * 1 extra tick because of +-1/2 tick uncertainty in the 2892 * firing of the timer. The bias will give us exactly the 2893 * 1.5 tick we need. But, because the bias is 2894 * statistical, we have to test that we don't drop below 2895 * the minimum feasible timer (which is 2 ticks). 2896 */ 2897 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2898 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2899 2900 /* 2901 * We received an ack for a packet that wasn't retransmitted; 2902 * it is probably safe to discard any error indications we've 2903 * received recently. This isn't quite right, but close enough 2904 * for now (a route might have failed after we sent a segment, 2905 * and the return path might not be symmetrical). 2906 */ 2907 tp->t_softerror = 0; 2908} 2909 2910/* 2911 * Determine a reasonable value for maxseg size. 2912 * If the route is known, check route for mtu. 2913 * If none, use an mss that can be handled on the outgoing 2914 * interface without forcing IP to fragment; if bigger than 2915 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2916 * to utilize large mbufs. If no route is found, route has no mtu, 2917 * or the destination isn't local, use a default, hopefully conservative 2918 * size (usually 512 or the default IP max size, but no more than the mtu 2919 * of the interface), as we can't discover anything about intervening 2920 * gateways or networks. We also initialize the congestion/slow start 2921 * window to be a single segment if the destination isn't local. 2922 * While looking at the routing entry, we also initialize other path-dependent 2923 * parameters from pre-set or cached values in the routing entry. 2924 * 2925 * Also take into account the space needed for options that we 2926 * send regularly. Make maxseg shorter by that amount to assure 2927 * that we can send maxseg amount of data even when the options 2928 * are present. Store the upper limit of the length of options plus 2929 * data in maxopd. 2930 * 2931 * 2932 * In case of T/TCP, we call this routine during implicit connection 2933 * setup as well (offer = -1), to initialize maxseg from the cached 2934 * MSS of our peer. 2935 * 2936 * NOTE that this routine is only called when we process an incoming 2937 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2938 */ 2939void 2940tcp_mss(tp, offer) 2941 struct tcpcb *tp; 2942 int offer; 2943{ 2944 int rtt, mss; 2945 u_long bufsize; 2946 u_long maxmtu; 2947 struct inpcb *inp = tp->t_inpcb; 2948 struct socket *so; 2949 struct hc_metrics_lite metrics; 2950 int origoffer = offer; 2951 int mtuflags = 0; 2952#ifdef INET6 2953 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2954 size_t min_protoh = isipv6 ? 2955 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2956 sizeof (struct tcpiphdr); 2957#else 2958 const size_t min_protoh = sizeof(struct tcpiphdr); 2959#endif 2960 2961 /* initialize */ 2962#ifdef INET6 2963 if (isipv6) { 2964 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags); 2965 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2966 } else 2967#endif 2968 { 2969 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags); 2970 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2971 } 2972 so = inp->inp_socket; 2973 2974 /* 2975 * no route to sender, stay with default mss and return 2976 */ 2977 if (maxmtu == 0) 2978 return; 2979 2980 /* what have we got? */ 2981 switch (offer) { 2982 case 0: 2983 /* 2984 * Offer == 0 means that there was no MSS on the SYN 2985 * segment, in this case we use tcp_mssdflt. 2986 */ 2987 offer = 2988#ifdef INET6 2989 isipv6 ? tcp_v6mssdflt : 2990#endif 2991 tcp_mssdflt; 2992 break; 2993 2994 case -1: 2995 /* 2996 * Offer == -1 means that we didn't receive SYN yet. 2997 */ 2998 /* FALLTHROUGH */ 2999 3000 default: 3001 /* 3002 * Prevent DoS attack with too small MSS. Round up 3003 * to at least minmss. 3004 */ 3005 offer = max(offer, tcp_minmss); 3006 /* 3007 * Sanity check: make sure that maxopd will be large 3008 * enough to allow some data on segments even if the 3009 * all the option space is used (40bytes). Otherwise 3010 * funny things may happen in tcp_output. 3011 */ 3012 offer = max(offer, 64); 3013 } 3014 3015 /* 3016 * rmx information is now retrieved from tcp_hostcache 3017 */ 3018 tcp_hc_get(&inp->inp_inc, &metrics); 3019 3020 /* 3021 * if there's a discovered mtu int tcp hostcache, use it 3022 * else, use the link mtu. 3023 */ 3024 if (metrics.rmx_mtu) 3025 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3026 else { 3027#ifdef INET6 3028 if (isipv6) { 3029 mss = maxmtu - min_protoh; 3030 if (!path_mtu_discovery && 3031 !in6_localaddr(&inp->in6p_faddr)) 3032 mss = min(mss, tcp_v6mssdflt); 3033 } else 3034#endif 3035 { 3036 mss = maxmtu - min_protoh; 3037 if (!path_mtu_discovery && 3038 !in_localaddr(inp->inp_faddr)) 3039 mss = min(mss, tcp_mssdflt); 3040 } 3041 } 3042 mss = min(mss, offer); 3043 3044 /* 3045 * maxopd stores the maximum length of data AND options 3046 * in a segment; maxseg is the amount of data in a normal 3047 * segment. We need to store this value (maxopd) apart 3048 * from maxseg, because now every segment carries options 3049 * and thus we normally have somewhat less data in segments. 3050 */ 3051 tp->t_maxopd = mss; 3052 3053 /* 3054 * origoffer==-1 indicates, that no segments were received yet. 3055 * In this case we just guess. 3056 */ 3057 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3058 (origoffer == -1 || 3059 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3060 mss -= TCPOLEN_TSTAMP_APPA; 3061 tp->t_maxseg = mss; 3062 3063#if (MCLBYTES & (MCLBYTES - 1)) == 0 3064 if (mss > MCLBYTES) 3065 mss &= ~(MCLBYTES-1); 3066#else 3067 if (mss > MCLBYTES) 3068 mss = mss / MCLBYTES * MCLBYTES; 3069#endif 3070 tp->t_maxseg = mss; 3071 3072 /* 3073 * If there's a pipesize, change the socket buffer to that size, 3074 * don't change if sb_hiwat is different than default (then it 3075 * has been changed on purpose with setsockopt). 3076 * Make the socket buffers an integral number of mss units; 3077 * if the mss is larger than the socket buffer, decrease the mss. 3078 */ 3079 SOCKBUF_LOCK(&so->so_snd); 3080 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3081 bufsize = metrics.rmx_sendpipe; 3082 else 3083 bufsize = so->so_snd.sb_hiwat; 3084 if (bufsize < mss) 3085 mss = bufsize; 3086 else { 3087 bufsize = roundup(bufsize, mss); 3088 if (bufsize > sb_max) 3089 bufsize = sb_max; 3090 if (bufsize > so->so_snd.sb_hiwat) 3091 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3092 } 3093 SOCKBUF_UNLOCK(&so->so_snd); 3094 tp->t_maxseg = mss; 3095 3096 SOCKBUF_LOCK(&so->so_rcv); 3097 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3098 bufsize = metrics.rmx_recvpipe; 3099 else 3100 bufsize = so->so_rcv.sb_hiwat; 3101 if (bufsize > mss) { 3102 bufsize = roundup(bufsize, mss); 3103 if (bufsize > sb_max) 3104 bufsize = sb_max; 3105 if (bufsize > so->so_rcv.sb_hiwat) 3106 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3107 } 3108 SOCKBUF_UNLOCK(&so->so_rcv); 3109 /* 3110 * While we're here, check the others too 3111 */ 3112 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3113 tp->t_srtt = rtt; 3114 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3115 tcpstat.tcps_usedrtt++; 3116 if (metrics.rmx_rttvar) { 3117 tp->t_rttvar = metrics.rmx_rttvar; 3118 tcpstat.tcps_usedrttvar++; 3119 } else { 3120 /* default variation is +- 1 rtt */ 3121 tp->t_rttvar = 3122 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3123 } 3124 TCPT_RANGESET(tp->t_rxtcur, 3125 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3126 tp->t_rttmin, TCPTV_REXMTMAX); 3127 } 3128 if (metrics.rmx_ssthresh) { 3129 /* 3130 * There's some sort of gateway or interface 3131 * buffer limit on the path. Use this to set 3132 * the slow start threshhold, but set the 3133 * threshold to no less than 2*mss. 3134 */ 3135 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3136 tcpstat.tcps_usedssthresh++; 3137 } 3138 if (metrics.rmx_bandwidth) 3139 tp->snd_bandwidth = metrics.rmx_bandwidth; 3140 3141 /* 3142 * Set the slow-start flight size depending on whether this 3143 * is a local network or not. 3144 * 3145 * Extend this so we cache the cwnd too and retrieve it here. 3146 * Make cwnd even bigger than RFC3390 suggests but only if we 3147 * have previous experience with the remote host. Be careful 3148 * not make cwnd bigger than remote receive window or our own 3149 * send socket buffer. Maybe put some additional upper bound 3150 * on the retrieved cwnd. Should do incremental updates to 3151 * hostcache when cwnd collapses so next connection doesn't 3152 * overloads the path again. 3153 * 3154 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3155 * We currently check only in syncache_socket for that. 3156 */ 3157#define TCP_METRICS_CWND 3158#ifdef TCP_METRICS_CWND 3159 if (metrics.rmx_cwnd) 3160 tp->snd_cwnd = max(mss, 3161 min(metrics.rmx_cwnd / 2, 3162 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3163 else 3164#endif 3165 if (tcp_do_rfc3390) 3166 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3167#ifdef INET6 3168 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3169 (!isipv6 && in_localaddr(inp->inp_faddr))) 3170#else 3171 else if (in_localaddr(inp->inp_faddr)) 3172#endif 3173 tp->snd_cwnd = mss * ss_fltsz_local; 3174 else 3175 tp->snd_cwnd = mss * ss_fltsz; 3176 3177 /* Check the interface for TSO capabilities. */ 3178 if (mtuflags & CSUM_TSO) 3179 tp->t_flags |= TF_TSO; 3180} 3181 3182/* 3183 * Determine the MSS option to send on an outgoing SYN. 3184 */ 3185int 3186tcp_mssopt(inc) 3187 struct in_conninfo *inc; 3188{ 3189 int mss = 0; 3190 u_long maxmtu = 0; 3191 u_long thcmtu = 0; 3192 size_t min_protoh; 3193#ifdef INET6 3194 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3195#endif 3196 3197 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3198 3199#ifdef INET6 3200 if (isipv6) { 3201 mss = tcp_v6mssdflt; 3202 maxmtu = tcp_maxmtu6(inc, NULL); 3203 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3204 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3205 } else 3206#endif 3207 { 3208 mss = tcp_mssdflt; 3209 maxmtu = tcp_maxmtu(inc, NULL); 3210 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3211 min_protoh = sizeof(struct tcpiphdr); 3212 } 3213 if (maxmtu && thcmtu) 3214 mss = min(maxmtu, thcmtu) - min_protoh; 3215 else if (maxmtu || thcmtu) 3216 mss = max(maxmtu, thcmtu) - min_protoh; 3217 3218 return (mss); 3219} 3220 3221 3222/* 3223 * On a partial ack arrives, force the retransmission of the 3224 * next unacknowledged segment. Do not clear tp->t_dupacks. 3225 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3226 * be started again. 3227 */ 3228static void 3229tcp_newreno_partial_ack(tp, th) 3230 struct tcpcb *tp; 3231 struct tcphdr *th; 3232{ 3233 tcp_seq onxt = tp->snd_nxt; 3234 u_long ocwnd = tp->snd_cwnd; 3235 3236 callout_stop(tp->tt_rexmt); 3237 tp->t_rtttime = 0; 3238 tp->snd_nxt = th->th_ack; 3239 /* 3240 * Set snd_cwnd to one segment beyond acknowledged offset. 3241 * (tp->snd_una has not yet been updated when this function is called.) 3242 */ 3243 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3244 tp->t_flags |= TF_ACKNOW; 3245 (void) tcp_output(tp); 3246 tp->snd_cwnd = ocwnd; 3247 if (SEQ_GT(onxt, tp->snd_nxt)) 3248 tp->snd_nxt = onxt; 3249 /* 3250 * Partial window deflation. Relies on fact that tp->snd_una 3251 * not updated yet. 3252 */ 3253 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3254 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3255 else 3256 tp->snd_cwnd = 0; 3257 tp->snd_cwnd += tp->t_maxseg; 3258} 3259 3260/* 3261 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3262 * looking for a pcb in the listen state. Returns 0 otherwise. 3263 */ 3264static int 3265tcp_timewait(inp, to, th, m, tlen) 3266 struct inpcb *inp; 3267 struct tcpopt *to; 3268 struct tcphdr *th; 3269 struct mbuf *m; 3270 int tlen; 3271{ 3272 struct tcptw *tw; 3273 int thflags; 3274 tcp_seq seq; 3275#ifdef INET6 3276 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3277#else 3278 const int isipv6 = 0; 3279#endif 3280 3281 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */ 3282 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3283 INP_LOCK_ASSERT(inp); 3284 3285 /* 3286 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 3287 * still present. This is undesirable, but temporarily necessary 3288 * until we work out how to handle inpcb's who's timewait state has 3289 * been removed. 3290 */ 3291 tw = intotw(inp); 3292 if (tw == NULL) 3293 goto drop; 3294 3295 thflags = th->th_flags; 3296 3297 /* 3298 * NOTE: for FIN_WAIT_2 (to be added later), 3299 * must validate sequence number before accepting RST 3300 */ 3301 3302 /* 3303 * If the segment contains RST: 3304 * Drop the segment - see Stevens, vol. 2, p. 964 and 3305 * RFC 1337. 3306 */ 3307 if (thflags & TH_RST) 3308 goto drop; 3309 3310#if 0 3311/* PAWS not needed at the moment */ 3312 /* 3313 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3314 * and it's less than ts_recent, drop it. 3315 */ 3316 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3317 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3318 if ((thflags & TH_ACK) == 0) 3319 goto drop; 3320 goto ack; 3321 } 3322 /* 3323 * ts_recent is never updated because we never accept new segments. 3324 */ 3325#endif 3326 3327 /* 3328 * If a new connection request is received 3329 * while in TIME_WAIT, drop the old connection 3330 * and start over if the sequence numbers 3331 * are above the previous ones. 3332 */ 3333 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3334 tcp_twclose(tw, 0); 3335 return (1); 3336 } 3337 3338 /* 3339 * Drop the the segment if it does not contain an ACK. 3340 */ 3341 if ((thflags & TH_ACK) == 0) 3342 goto drop; 3343 3344 /* 3345 * Reset the 2MSL timer if this is a duplicate FIN. 3346 */ 3347 if (thflags & TH_FIN) { 3348 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3349 if (seq + 1 == tw->rcv_nxt) 3350 tcp_timer_2msl_reset(tw, 1); 3351 } 3352 3353 /* 3354 * Acknowledge the segment if it has data or is not a duplicate ACK. 3355 */ 3356 if (thflags != TH_ACK || tlen != 0 || 3357 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3358 tcp_twrespond(tw, TH_ACK); 3359 goto drop; 3360 3361 /* 3362 * Generate a RST, dropping incoming segment. 3363 * Make ACK acceptable to originator of segment. 3364 * Don't bother to respond if destination was broadcast/multicast. 3365 */ 3366 if (m->m_flags & (M_BCAST|M_MCAST)) 3367 goto drop; 3368 if (isipv6) { 3369 struct ip6_hdr *ip6; 3370 3371 /* IPv6 anycast check is done at tcp6_input() */ 3372 ip6 = mtod(m, struct ip6_hdr *); 3373 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3374 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3375 goto drop; 3376 } else { 3377 struct ip *ip; 3378 3379 ip = mtod(m, struct ip *); 3380 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3381 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3382 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3383 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3384 goto drop; 3385 } 3386 if (thflags & TH_ACK) { 3387 tcp_respond(NULL, 3388 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3389 } else { 3390 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3391 tcp_respond(NULL, 3392 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3393 } 3394 INP_UNLOCK(inp); 3395 return (0); 3396 3397drop: 3398 INP_UNLOCK(inp); 3399 m_freem(m); 3400 return (0); 3401} 3402