tcp_input.c revision 58698
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 * 3. All advertising materials mentioning features or use of this software 14 * must display the following acknowledgement: 15 * This product includes software developed by the University of 16 * California, Berkeley and its contributors. 17 * 4. Neither the name of the University nor the names of its contributors 18 * may be used to endorse or promote products derived from this software 19 * without specific prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 * 33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 34 * $FreeBSD: head/sys/netinet/tcp_input.c 58698 2000-03-27 19:14:27Z jlemon $ 35 */ 36 37#include "opt_ipfw.h" /* for ipfw_fwd */ 38#include "opt_inet6.h" 39#include "opt_ipsec.h" 40#include "opt_tcpdebug.h" 41#include "opt_tcp_input.h" 42 43#include <sys/param.h> 44#include <sys/systm.h> 45#include <sys/kernel.h> 46#include <sys/sysctl.h> 47#include <sys/malloc.h> 48#include <sys/mbuf.h> 49#include <sys/proc.h> /* for proc0 declaration */ 50#include <sys/protosw.h> 51#include <sys/socket.h> 52#include <sys/socketvar.h> 53#include <sys/syslog.h> 54 55#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57#include <net/if.h> 58#include <net/route.h> 59 60#include <netinet/in.h> 61#include <netinet/in_systm.h> 62#include <netinet/ip.h> 63#include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 64#ifdef INET6 65#include <netinet/ip6.h> 66#include <netinet/in_var.h> 67#include <netinet6/nd6.h> 68#include <netinet/icmp6.h> 69#endif 70#include <netinet/in_pcb.h> 71#ifdef INET6 72#include <netinet6/in6_pcb.h> 73#endif 74#include <netinet/ip_var.h> 75#ifdef INET6 76#include <netinet6/ip6_var.h> 77#endif 78#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 79#include <netinet/tcp.h> 80#include <netinet/tcp_fsm.h> 81#include <netinet/tcp_seq.h> 82#include <netinet/tcp_timer.h> 83#include <netinet/tcp_var.h> 84#ifdef INET6 85#include <netinet6/tcp6_var.h> 86#endif 87#include <netinet/tcpip.h> 88#ifdef TCPDEBUG 89#include <netinet/tcp_debug.h> 90 91u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 92struct tcphdr tcp_savetcp; 93#endif /* TCPDEBUG */ 94 95#ifdef IPSEC 96#include <netinet6/ipsec.h> 97#ifdef INET6 98#include <netinet6/ipsec6.h> 99#endif 100#include <netkey/key.h> 101#endif /*IPSEC*/ 102 103#include <machine/in_cksum.h> 104 105MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 106 107static int tcprexmtthresh = 3; 108tcp_seq tcp_iss; 109tcp_cc tcp_ccgen; 110 111struct tcpstat tcpstat; 112SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD, 113 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 114 115static int log_in_vain = 0; 116SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 117 &log_in_vain, 0, "Log all incoming TCP connections"); 118 119static int blackhole = 0; 120SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 121 &blackhole, 0, "Do not send RST when dropping refused connections"); 122 123int tcp_delack_enabled = 1; 124SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 125 &tcp_delack_enabled, 0, 126 "Delay ACK to try and piggyback it onto a data packet"); 127 128#ifdef TCP_DROP_SYNFIN 129static int drop_synfin = 0; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 131 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 132#endif 133 134#ifdef TCP_RESTRICT_RST 135static int restrict_rst = 0; 136SYSCTL_INT(_net_inet_tcp, OID_AUTO, restrict_rst, CTLFLAG_RW, 137 &restrict_rst, 0, "Restrict RST emission"); 138#endif 139 140struct inpcbhead tcb; 141#define tcb6 tcb /* for KAME src sync over BSD*'s */ 142struct inpcbinfo tcbinfo; 143 144static void tcp_dooptions __P((struct tcpcb *, 145 u_char *, int, struct tcphdr *, struct tcpopt *)); 146static void tcp_pulloutofband __P((struct socket *, 147 struct tcphdr *, struct mbuf *, int)); 148static int tcp_reass __P((struct tcpcb *, struct tcphdr *, int *, 149 struct mbuf *)); 150static void tcp_xmit_timer __P((struct tcpcb *, int)); 151 152/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 153#ifdef INET6 154#define ND6_HINT(tp) \ 155do { \ 156 if ((tp) && (tp)->t_inpcb && \ 157 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 158 (tp)->t_inpcb->in6p_route.ro_rt) \ 159 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL); \ 160} while (0) 161#else 162#define ND6_HINT(tp) 163#endif 164 165/* 166 * Insert segment which inludes th into reassembly queue of tcp with 167 * control block tp. Return TH_FIN if reassembly now includes 168 * a segment with FIN. The macro form does the common case inline 169 * (segment is the next to be received on an established connection, 170 * and the queue is empty), avoiding linkage into and removal 171 * from the queue and repetition of various conversions. 172 * Set DELACK for segments received in order, but ack immediately 173 * when segments are out of order (so fast retransmit can work). 174 */ 175#define TCP_REASS(tp, th, tlenp, m, so, flags) { \ 176 if ((th)->th_seq == (tp)->rcv_nxt && \ 177 LIST_EMPTY(&(tp)->t_segq) && \ 178 (tp)->t_state == TCPS_ESTABLISHED) { \ 179 if (tcp_delack_enabled) \ 180 callout_reset(tp->tt_delack, tcp_delacktime, \ 181 tcp_timer_delack, tp); \ 182 else \ 183 tp->t_flags |= TF_ACKNOW; \ 184 (tp)->rcv_nxt += *(tlenp); \ 185 flags = (th)->th_flags & TH_FIN; \ 186 tcpstat.tcps_rcvpack++;\ 187 tcpstat.tcps_rcvbyte += *(tlenp);\ 188 ND6_HINT(tp); \ 189 sbappend(&(so)->so_rcv, (m)); \ 190 sorwakeup(so); \ 191 } else { \ 192 (flags) = tcp_reass((tp), (th), (tlenp), (m)); \ 193 tp->t_flags |= TF_ACKNOW; \ 194 } \ 195} 196 197static int 198tcp_reass(tp, th, tlenp, m) 199 register struct tcpcb *tp; 200 register struct tcphdr *th; 201 int *tlenp; 202 struct mbuf *m; 203{ 204 struct tseg_qent *q; 205 struct tseg_qent *p = NULL; 206 struct tseg_qent *nq; 207 struct tseg_qent *te; 208 struct socket *so = tp->t_inpcb->inp_socket; 209 int flags; 210 211 /* 212 * Call with th==0 after become established to 213 * force pre-ESTABLISHED data up to user socket. 214 */ 215 if (th == 0) 216 goto present; 217 218 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 219 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 220 M_NOWAIT); 221 if (te == NULL) { 222 tcpstat.tcps_rcvmemdrop++; 223 m_freem(m); 224 return (0); 225 } 226 227 /* 228 * Find a segment which begins after this one does. 229 */ 230 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 231 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 232 break; 233 p = q; 234 } 235 236 /* 237 * If there is a preceding segment, it may provide some of 238 * our data already. If so, drop the data from the incoming 239 * segment. If it provides all of our data, drop us. 240 */ 241 if (p != NULL) { 242 register int i; 243 /* conversion to int (in i) handles seq wraparound */ 244 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 245 if (i > 0) { 246 if (i >= *tlenp) { 247 tcpstat.tcps_rcvduppack++; 248 tcpstat.tcps_rcvdupbyte += *tlenp; 249 m_freem(m); 250 FREE(te, M_TSEGQ); 251 /* 252 * Try to present any queued data 253 * at the left window edge to the user. 254 * This is needed after the 3-WHS 255 * completes. 256 */ 257 goto present; /* ??? */ 258 } 259 m_adj(m, i); 260 *tlenp -= i; 261 th->th_seq += i; 262 } 263 } 264 tcpstat.tcps_rcvoopack++; 265 tcpstat.tcps_rcvoobyte += *tlenp; 266 267 /* 268 * While we overlap succeeding segments trim them or, 269 * if they are completely covered, dequeue them. 270 */ 271 while (q) { 272 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 273 if (i <= 0) 274 break; 275 if (i < q->tqe_len) { 276 q->tqe_th->th_seq += i; 277 q->tqe_len -= i; 278 m_adj(q->tqe_m, i); 279 break; 280 } 281 282 nq = LIST_NEXT(q, tqe_q); 283 LIST_REMOVE(q, tqe_q); 284 m_freem(q->tqe_m); 285 FREE(q, M_TSEGQ); 286 q = nq; 287 } 288 289 /* Insert the new segment queue entry into place. */ 290 te->tqe_m = m; 291 te->tqe_th = th; 292 te->tqe_len = *tlenp; 293 294 if (p == NULL) { 295 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 296 } else { 297 LIST_INSERT_AFTER(p, te, tqe_q); 298 } 299 300present: 301 /* 302 * Present data to user, advancing rcv_nxt through 303 * completed sequence space. 304 */ 305 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 306 return (0); 307 q = LIST_FIRST(&tp->t_segq); 308 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 309 return (0); 310 do { 311 tp->rcv_nxt += q->tqe_len; 312 flags = q->tqe_th->th_flags & TH_FIN; 313 nq = LIST_NEXT(q, tqe_q); 314 LIST_REMOVE(q, tqe_q); 315 if (so->so_state & SS_CANTRCVMORE) 316 m_freem(q->tqe_m); 317 else 318 sbappend(&so->so_rcv, q->tqe_m); 319 FREE(q, M_TSEGQ); 320 q = nq; 321 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 322 ND6_HINT(tp); 323 sorwakeup(so); 324 return (flags); 325} 326 327/* 328 * TCP input routine, follows pages 65-76 of the 329 * protocol specification dated September, 1981 very closely. 330 */ 331#ifdef INET6 332int 333tcp6_input(mp, offp, proto) 334 struct mbuf **mp; 335 int *offp, proto; 336{ 337 register struct mbuf *m = *mp; 338 339 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 340 341 /* 342 * draft-itojun-ipv6-tcp-to-anycast 343 * better place to put this in? 344 */ 345 if (m->m_flags & M_ANYCAST6) { 346 struct ip6_hdr *ip6; 347 348 ip6 = mtod(m, struct ip6_hdr *); 349 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 350 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 351 return IPPROTO_DONE; 352 } 353 354 tcp_input(m, *offp, proto); 355 return IPPROTO_DONE; 356} 357#endif 358 359void 360tcp_input(m, off0, proto) 361 register struct mbuf *m; 362 int off0, proto; 363{ 364 register struct tcphdr *th; 365 register struct ip *ip = NULL; 366 register struct ipovly *ipov; 367 register struct inpcb *inp; 368 u_char *optp = NULL; 369 int optlen = 0; 370 int len, tlen, off; 371 int drop_hdrlen; 372 register struct tcpcb *tp = 0; 373 register int thflags; 374 struct socket *so = 0; 375 int todrop, acked, ourfinisacked, needoutput = 0; 376 struct in_addr laddr; 377#ifdef INET6 378 struct in6_addr laddr6; 379#endif 380 int dropsocket = 0; 381 int iss = 0; 382 u_long tiwin; 383 struct tcpopt to; /* options in this segment */ 384 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 385 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 386#ifdef TCPDEBUG 387 short ostate = 0; 388#endif 389#ifdef INET6 390 struct ip6_hdr *ip6 = NULL; 391 int isipv6; 392#endif /* INET6 */ 393 394#ifdef INET6 395 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 396#endif 397 bzero((char *)&to, sizeof(to)); 398 399 tcpstat.tcps_rcvtotal++; 400 401#ifdef INET6 402 if (isipv6) { 403 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 404 ip6 = mtod(m, struct ip6_hdr *); 405 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 406 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 407 tcpstat.tcps_rcvbadsum++; 408 goto drop; 409 } 410 th = (struct tcphdr *)((caddr_t)ip6 + off0); 411 } else 412#endif /* INET6 */ 413 { 414 /* 415 * Get IP and TCP header together in first mbuf. 416 * Note: IP leaves IP header in first mbuf. 417 */ 418 if (off0 > sizeof (struct ip)) { 419 ip_stripoptions(m, (struct mbuf *)0); 420 off0 = sizeof(struct ip); 421 } 422 if (m->m_len < sizeof (struct tcpiphdr)) { 423 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 424 tcpstat.tcps_rcvshort++; 425 return; 426 } 427 } 428 ip = mtod(m, struct ip *); 429 ipov = (struct ipovly *)ip; 430 th = (struct tcphdr *)((caddr_t)ip + off0); 431 tlen = ip->ip_len; 432 433 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 434 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 435 th->th_sum = m->m_pkthdr.csum_data; 436 else 437 th->th_sum = in_pseudo(ip->ip_src.s_addr, 438 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + 439 ip->ip_len + IPPROTO_TCP)); 440 th->th_sum ^= 0xffff; 441 } else { 442 /* 443 * Checksum extended TCP header and data. 444 */ 445 len = sizeof (struct ip) + tlen; 446 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 447 ipov->ih_len = (u_short)tlen; 448 HTONS(ipov->ih_len); 449 th->th_sum = in_cksum(m, len); 450 } 451 if (th->th_sum) { 452 tcpstat.tcps_rcvbadsum++; 453 goto drop; 454 } 455#ifdef INET6 456 /* Re-initialization for later version check */ 457 ip->ip_v = IPVERSION; 458#endif 459 } 460 461 /* 462 * Check that TCP offset makes sense, 463 * pull out TCP options and adjust length. XXX 464 */ 465 off = th->th_off << 2; 466 if (off < sizeof (struct tcphdr) || off > tlen) { 467 tcpstat.tcps_rcvbadoff++; 468 goto drop; 469 } 470 tlen -= off; /* tlen is used instead of ti->ti_len */ 471 if (off > sizeof (struct tcphdr)) { 472#ifdef INET6 473 if (isipv6) { 474 IP6_EXTHDR_CHECK(m, off0, off, ); 475 ip6 = mtod(m, struct ip6_hdr *); 476 th = (struct tcphdr *)((caddr_t)ip6 + off0); 477 } else 478#endif /* INET6 */ 479 { 480 if (m->m_len < sizeof(struct ip) + off) { 481 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 482 tcpstat.tcps_rcvshort++; 483 return; 484 } 485 ip = mtod(m, struct ip *); 486 ipov = (struct ipovly *)ip; 487 th = (struct tcphdr *)((caddr_t)ip + off0); 488 } 489 } 490 optlen = off - sizeof (struct tcphdr); 491 optp = (u_char *)(th + 1); 492 } 493 thflags = th->th_flags; 494 495#ifdef TCP_DROP_SYNFIN 496 /* 497 * If the drop_synfin option is enabled, drop all packets with 498 * both the SYN and FIN bits set. This prevents e.g. nmap from 499 * identifying the TCP/IP stack. 500 * 501 * This is incompatible with RFC1644 extensions (T/TCP). 502 */ 503 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 504 goto drop; 505#endif 506 507 /* 508 * Convert TCP protocol specific fields to host format. 509 */ 510 NTOHL(th->th_seq); 511 NTOHL(th->th_ack); 512 NTOHS(th->th_win); 513 NTOHS(th->th_urp); 514 515 /* 516 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 517 * until after ip6_savecontrol() is called and before other functions 518 * which don't want those proto headers. 519 * Because ip6_savecontrol() is going to parse the mbuf to 520 * search for data to be passed up to user-land, it wants mbuf 521 * parameters to be unchanged. 522 */ 523 drop_hdrlen = off0 + off; 524 525 /* 526 * Locate pcb for segment. 527 */ 528findpcb: 529#ifdef IPFIREWALL_FORWARD 530 if (ip_fw_fwd_addr != NULL 531#ifdef INET6 532 && isipv6 == NULL /* IPv6 support is not yet */ 533#endif /* INET6 */ 534 ) { 535 /* 536 * Diverted. Pretend to be the destination. 537 * already got one like this? 538 */ 539 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 540 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); 541 if (!inp) { 542 /* 543 * No, then it's new. Try find the ambushing socket 544 */ 545 if (!ip_fw_fwd_addr->sin_port) { 546 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, 547 th->th_sport, ip_fw_fwd_addr->sin_addr, 548 th->th_dport, 1, m->m_pkthdr.rcvif); 549 } else { 550 inp = in_pcblookup_hash(&tcbinfo, 551 ip->ip_src, th->th_sport, 552 ip_fw_fwd_addr->sin_addr, 553 ntohs(ip_fw_fwd_addr->sin_port), 1, 554 m->m_pkthdr.rcvif); 555 } 556 } 557 ip_fw_fwd_addr = NULL; 558 } else 559#endif /* IPFIREWALL_FORWARD */ 560 { 561#ifdef INET6 562 if (isipv6) 563 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, 564 &ip6->ip6_dst, th->th_dport, 1, 565 m->m_pkthdr.rcvif); 566 else 567#endif /* INET6 */ 568 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 569 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); 570 } 571 572#ifdef IPSEC 573#ifdef INET6 574 if (isipv6) { 575 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 576 ipsec6stat.in_polvio++; 577 goto drop; 578 } 579 } else 580#endif /* INET6 */ 581 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 582 ipsecstat.in_polvio++; 583 goto drop; 584 } 585#endif /*IPSEC*/ 586 587 /* 588 * If the state is CLOSED (i.e., TCB does not exist) then 589 * all data in the incoming segment is discarded. 590 * If the TCB exists but is in CLOSED state, it is embryonic, 591 * but should either do a listen or a connect soon. 592 */ 593 if (inp == NULL) { 594 if (log_in_vain) { 595#ifdef INET6 596 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN]; 597#else /* INET6 */ 598 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 599#endif /* INET6 */ 600 601#ifdef INET6 602 if (isipv6) { 603 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst)); 604 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src)); 605 } else 606#endif 607 { 608 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 609 strcpy(sbuf, inet_ntoa(ip->ip_src)); 610 } 611 switch (log_in_vain) { 612 case 1: 613 if(thflags & TH_SYN) 614 log(LOG_INFO, 615 "Connection attempt to TCP %s:%d from %s:%d\n", 616 dbuf, ntohs(th->th_dport), 617 sbuf, 618 ntohs(th->th_sport)); 619 break; 620 case 2: 621 log(LOG_INFO, 622 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 623 dbuf, ntohs(th->th_dport), sbuf, 624 ntohs(th->th_sport), thflags); 625 break; 626 default: 627 break; 628 } 629 } 630 if (blackhole) { 631 switch (blackhole) { 632 case 1: 633 if (thflags & TH_SYN) 634 goto drop; 635 break; 636 case 2: 637 goto drop; 638 default: 639 goto drop; 640 } 641 } 642 goto maybedropwithreset; 643 } 644 tp = intotcpcb(inp); 645 if (tp == 0) 646 goto maybedropwithreset; 647 if (tp->t_state == TCPS_CLOSED) 648 goto drop; 649 650 /* Unscale the window into a 32-bit value. */ 651 if ((thflags & TH_SYN) == 0) 652 tiwin = th->th_win << tp->snd_scale; 653 else 654 tiwin = th->th_win; 655 656#ifdef INET6 657 /* save packet options if user wanted */ 658 if (isipv6 && inp->in6p_flags & INP_CONTROLOPTS) { 659 if (inp->in6p_options) { 660 m_freem(inp->in6p_options); 661 inp->in6p_options = 0; 662 } 663 ip6_savecontrol(inp, &inp->in6p_options, ip6, m); 664 } 665 /* else, should also do ip_srcroute() here? */ 666#endif /* INET6 */ 667 668 so = inp->inp_socket; 669 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 670#ifdef TCPDEBUG 671 if (so->so_options & SO_DEBUG) { 672 ostate = tp->t_state; 673#ifdef INET6 674 if (isipv6) 675 bcopy((char *)ip6, (char *)tcp_saveipgen, 676 sizeof(*ip6)); 677 else 678#endif /* INET6 */ 679 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 680 tcp_savetcp = *th; 681 } 682#endif 683 if (so->so_options & SO_ACCEPTCONN) { 684 register struct tcpcb *tp0 = tp; 685 struct socket *so2; 686#ifdef IPSEC 687 struct socket *oso; 688#endif 689#ifdef INET6 690 struct inpcb *oinp = sotoinpcb(so); 691#endif /* INET6 */ 692 693#ifndef IPSEC 694 /* 695 * Current IPsec implementation makes incorrect IPsec 696 * cache if this check is done here. 697 * So delay this until duplicated socket is created. 698 */ 699 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 700 /* 701 * Note: dropwithreset makes sure we don't 702 * send a RST in response to a RST. 703 */ 704 if (thflags & TH_ACK) { 705 tcpstat.tcps_badsyn++; 706 goto maybedropwithreset; 707 } 708 goto drop; 709 } 710#endif 711 so2 = sonewconn(so, 0); 712 if (so2 == 0) { 713 tcpstat.tcps_listendrop++; 714 so2 = sodropablereq(so); 715 if (so2) { 716 tcp_drop(sototcpcb(so2), ETIMEDOUT); 717 so2 = sonewconn(so, 0); 718 } 719 if (!so2) 720 goto drop; 721 } 722#ifdef IPSEC 723 oso = so; 724#endif 725 so = so2; 726 /* 727 * This is ugly, but .... 728 * 729 * Mark socket as temporary until we're 730 * committed to keeping it. The code at 731 * ``drop'' and ``dropwithreset'' check the 732 * flag dropsocket to see if the temporary 733 * socket created here should be discarded. 734 * We mark the socket as discardable until 735 * we're committed to it below in TCPS_LISTEN. 736 */ 737 dropsocket++; 738 inp = (struct inpcb *)so->so_pcb; 739#ifdef INET6 740 if (isipv6) 741 inp->in6p_laddr = ip6->ip6_dst; 742 else { 743 if (ip6_mapped_addr_on) { 744 inp->inp_vflag &= ~INP_IPV6; 745 inp->inp_vflag |= INP_IPV4; 746 } 747#endif /* INET6 */ 748 inp->inp_laddr = ip->ip_dst; 749#ifdef INET6 750 } 751#endif /* INET6 */ 752 inp->inp_lport = th->th_dport; 753 if (in_pcbinshash(inp) != 0) { 754 /* 755 * Undo the assignments above if we failed to 756 * put the PCB on the hash lists. 757 */ 758#ifdef INET6 759 if (isipv6) 760 inp->in6p_laddr = in6addr_any; 761 else 762#endif /* INET6 */ 763 inp->inp_laddr.s_addr = INADDR_ANY; 764 inp->inp_lport = 0; 765 goto drop; 766 } 767#ifdef IPSEC 768 /* 769 * To avoid creating incorrectly cached IPsec 770 * association, this is need to be done here. 771 * 772 * Subject: (KAME-snap 748) 773 * From: Wayne Knowles <w.knowles@niwa.cri.nz> 774 * ftp://ftp.kame.net/pub/mail-list/snap-users/748 775 */ 776 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 777 /* 778 * Note: dropwithreset makes sure we don't 779 * send a RST in response to a RST. 780 */ 781 if (thflags & TH_ACK) { 782 tcpstat.tcps_badsyn++; 783 goto maybedropwithreset; 784 } 785 goto drop; 786 } 787#endif 788#ifdef INET6 789 if (isipv6) { 790 /* 791 * inherit socket options from the listening 792 * socket. 793 */ 794 inp->inp_flags |= 795 oinp->inp_flags & INP_CONTROLOPTS; 796 if (inp->inp_flags & INP_CONTROLOPTS) { 797 if (inp->in6p_options) { 798 m_freem(inp->in6p_options); 799 inp->in6p_options = 0; 800 } 801 ip6_savecontrol(inp, 802 &inp->in6p_options, 803 ip6, m); 804 } 805 } else 806#endif /* INET6 */ 807 inp->inp_options = ip_srcroute(); 808#ifdef IPSEC 809 /* copy old policy into new socket's */ 810 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, 811 inp->inp_sp)) 812 printf("tcp_input: could not copy policy\n"); 813#endif 814 tp = intotcpcb(inp); 815 tp->t_state = TCPS_LISTEN; 816 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 817 818 /* Compute proper scaling value from buffer space */ 819 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 820 TCP_MAXWIN << tp->request_r_scale < 821 so->so_rcv.sb_hiwat) 822 tp->request_r_scale++; 823 } 824 } 825 826 /* 827 * Segment received on connection. 828 * Reset idle time and keep-alive timer. 829 */ 830 tp->t_rcvtime = ticks; 831 if (TCPS_HAVEESTABLISHED(tp->t_state)) 832 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 833 834 /* 835 * Process options if not in LISTEN state, 836 * else do it below (after getting remote address). 837 */ 838 if (tp->t_state != TCPS_LISTEN) 839 tcp_dooptions(tp, optp, optlen, th, &to); 840 841 /* 842 * Header prediction: check for the two common cases 843 * of a uni-directional data xfer. If the packet has 844 * no control flags, is in-sequence, the window didn't 845 * change and we're not retransmitting, it's a 846 * candidate. If the length is zero and the ack moved 847 * forward, we're the sender side of the xfer. Just 848 * free the data acked & wake any higher level process 849 * that was blocked waiting for space. If the length 850 * is non-zero and the ack didn't move, we're the 851 * receiver side. If we're getting packets in-order 852 * (the reassembly queue is empty), add the data to 853 * the socket buffer and note that we need a delayed ack. 854 * Make sure that the hidden state-flags are also off. 855 * Since we check for TCPS_ESTABLISHED above, it can only 856 * be TH_NEEDSYN. 857 */ 858 if (tp->t_state == TCPS_ESTABLISHED && 859 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 860 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 861 ((to.to_flag & TOF_TS) == 0 || 862 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 863 /* 864 * Using the CC option is compulsory if once started: 865 * the segment is OK if no T/TCP was negotiated or 866 * if the segment has a CC option equal to CCrecv 867 */ 868 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 869 ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 870 th->th_seq == tp->rcv_nxt && 871 tiwin && tiwin == tp->snd_wnd && 872 tp->snd_nxt == tp->snd_max) { 873 874 /* 875 * If last ACK falls within this segment's sequence numbers, 876 * record the timestamp. 877 * NOTE that the test is modified according to the latest 878 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 879 */ 880 if ((to.to_flag & TOF_TS) != 0 && 881 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 882 tp->ts_recent_age = ticks; 883 tp->ts_recent = to.to_tsval; 884 } 885 886 if (tlen == 0) { 887 if (SEQ_GT(th->th_ack, tp->snd_una) && 888 SEQ_LEQ(th->th_ack, tp->snd_max) && 889 tp->snd_cwnd >= tp->snd_wnd && 890 tp->t_dupacks < tcprexmtthresh) { 891 /* 892 * this is a pure ack for outstanding data. 893 */ 894 ++tcpstat.tcps_predack; 895 /* 896 * "bad retransmit" recovery 897 */ 898 if (tp->t_rxtshift == 1 && 899 ticks < tp->t_badrxtwin) { 900 tp->snd_cwnd = tp->snd_cwnd_prev; 901 tp->snd_ssthresh = 902 tp->snd_ssthresh_prev; 903 tp->snd_nxt = tp->snd_max; 904 tp->t_badrxtwin = 0; 905 } 906 if ((to.to_flag & TOF_TS) != 0) 907 tcp_xmit_timer(tp, 908 ticks - to.to_tsecr + 1); 909 else if (tp->t_rtttime && 910 SEQ_GT(th->th_ack, tp->t_rtseq)) 911 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 912 acked = th->th_ack - tp->snd_una; 913 tcpstat.tcps_rcvackpack++; 914 tcpstat.tcps_rcvackbyte += acked; 915 sbdrop(&so->so_snd, acked); 916 tp->snd_una = th->th_ack; 917 m_freem(m); 918 ND6_HINT(tp); /* some progress has been done */ 919 920 /* 921 * If all outstanding data are acked, stop 922 * retransmit timer, otherwise restart timer 923 * using current (possibly backed-off) value. 924 * If process is waiting for space, 925 * wakeup/selwakeup/signal. If data 926 * are ready to send, let tcp_output 927 * decide between more output or persist. 928 */ 929 if (tp->snd_una == tp->snd_max) 930 callout_stop(tp->tt_rexmt); 931 else if (!callout_active(tp->tt_persist)) 932 callout_reset(tp->tt_rexmt, 933 tp->t_rxtcur, 934 tcp_timer_rexmt, tp); 935 936 sowwakeup(so); 937 if (so->so_snd.sb_cc) 938 (void) tcp_output(tp); 939 return; 940 } 941 } else if (th->th_ack == tp->snd_una && 942 LIST_EMPTY(&tp->t_segq) && 943 tlen <= sbspace(&so->so_rcv)) { 944 /* 945 * this is a pure, in-sequence data packet 946 * with nothing on the reassembly queue and 947 * we have enough buffer space to take it. 948 */ 949 ++tcpstat.tcps_preddat; 950 tp->rcv_nxt += tlen; 951 tcpstat.tcps_rcvpack++; 952 tcpstat.tcps_rcvbyte += tlen; 953 ND6_HINT(tp); /* some progress has been done */ 954 /* 955 * Add data to socket buffer. 956 */ 957 m_adj(m, drop_hdrlen); /* delayed header drop */ 958 sbappend(&so->so_rcv, m); 959 sorwakeup(so); 960 if (tcp_delack_enabled) { 961 callout_reset(tp->tt_delack, tcp_delacktime, 962 tcp_timer_delack, tp); 963 } else { 964 tp->t_flags |= TF_ACKNOW; 965 tcp_output(tp); 966 } 967 return; 968 } 969 } 970 971 /* 972 * Calculate amount of space in receive window, 973 * and then do TCP input processing. 974 * Receive window is amount of space in rcv queue, 975 * but not less than advertised window. 976 */ 977 { int win; 978 979 win = sbspace(&so->so_rcv); 980 if (win < 0) 981 win = 0; 982 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 983 } 984 985 switch (tp->t_state) { 986 987 /* 988 * If the state is LISTEN then ignore segment if it contains an RST. 989 * If the segment contains an ACK then it is bad and send a RST. 990 * If it does not contain a SYN then it is not interesting; drop it. 991 * If it is from this socket, drop it, it must be forged. 992 * Don't bother responding if the destination was a broadcast. 993 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 994 * tp->iss, and send a segment: 995 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 996 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 997 * Fill in remote peer address fields if not previously specified. 998 * Enter SYN_RECEIVED state, and process any other fields of this 999 * segment in this state. 1000 */ 1001 case TCPS_LISTEN: { 1002 register struct sockaddr_in *sin; 1003#ifdef INET6 1004 register struct sockaddr_in6 *sin6; 1005#endif 1006 1007 if (thflags & TH_RST) 1008 goto drop; 1009 if (thflags & TH_ACK) 1010 goto maybedropwithreset; 1011 if ((thflags & TH_SYN) == 0) 1012 goto drop; 1013 if (th->th_dport == th->th_sport) { 1014#ifdef INET6 1015 if (isipv6) { 1016 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 1017 &ip6->ip6_src)) 1018 goto drop; 1019 } else 1020#endif /* INET6 */ 1021 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 1022 goto drop; 1023 } 1024 /* 1025 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1026 * in_broadcast() should never return true on a received 1027 * packet with M_BCAST not set. 1028 * 1029 * Packets with a multicast source address should also 1030 * be discarded. 1031 */ 1032 if (m->m_flags & (M_BCAST|M_MCAST)) 1033 goto drop; 1034#ifdef INET6 1035 if (isipv6) { 1036 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1037 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 1038 goto drop; 1039 } else 1040#endif 1041 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1042 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1043 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 1044 goto drop; 1045#ifdef INET6 1046 if (isipv6) { 1047 MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6, 1048 M_SONAME, M_NOWAIT); 1049 if (sin6 == NULL) 1050 goto drop; 1051 bzero(sin6, sizeof(*sin6)); 1052 sin6->sin6_family = AF_INET6; 1053 sin6->sin6_len = sizeof(*sin6); 1054 sin6->sin6_addr = ip6->ip6_src; 1055 sin6->sin6_port = th->th_sport; 1056 laddr6 = inp->in6p_laddr; 1057 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1058 inp->in6p_laddr = ip6->ip6_dst; 1059 if (in6_pcbconnect(inp, (struct sockaddr *)sin6, 1060 &proc0)) { 1061 inp->in6p_laddr = laddr6; 1062 FREE(sin6, M_SONAME); 1063 goto drop; 1064 } 1065 FREE(sin6, M_SONAME); 1066 } else 1067#endif 1068 { 1069 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 1070 M_NOWAIT); 1071 if (sin == NULL) 1072 goto drop; 1073 sin->sin_family = AF_INET; 1074 sin->sin_len = sizeof(*sin); 1075 sin->sin_addr = ip->ip_src; 1076 sin->sin_port = th->th_sport; 1077 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 1078 laddr = inp->inp_laddr; 1079 if (inp->inp_laddr.s_addr == INADDR_ANY) 1080 inp->inp_laddr = ip->ip_dst; 1081 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 1082 inp->inp_laddr = laddr; 1083 FREE(sin, M_SONAME); 1084 goto drop; 1085 } 1086 FREE(sin, M_SONAME); 1087 } 1088 tp->t_template = tcp_template(tp); 1089 if (tp->t_template == 0) { 1090 tp = tcp_drop(tp, ENOBUFS); 1091 dropsocket = 0; /* socket is already gone */ 1092 goto drop; 1093 } 1094 if ((taop = tcp_gettaocache(inp)) == NULL) { 1095 taop = &tao_noncached; 1096 bzero(taop, sizeof(*taop)); 1097 } 1098 tcp_dooptions(tp, optp, optlen, th, &to); 1099 if (iss) 1100 tp->iss = iss; 1101 else 1102 tp->iss = tcp_iss; 1103 tcp_iss += TCP_ISSINCR/4; 1104 tp->irs = th->th_seq; 1105 tcp_sendseqinit(tp); 1106 tcp_rcvseqinit(tp); 1107 /* 1108 * Initialization of the tcpcb for transaction; 1109 * set SND.WND = SEG.WND, 1110 * initialize CCsend and CCrecv. 1111 */ 1112 tp->snd_wnd = tiwin; /* initial send-window */ 1113 tp->cc_send = CC_INC(tcp_ccgen); 1114 tp->cc_recv = to.to_cc; 1115 /* 1116 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1117 * - compare SEG.CC against cached CC from the same host, 1118 * if any. 1119 * - if SEG.CC > chached value, SYN must be new and is accepted 1120 * immediately: save new CC in the cache, mark the socket 1121 * connected, enter ESTABLISHED state, turn on flag to 1122 * send a SYN in the next segment. 1123 * A virtual advertised window is set in rcv_adv to 1124 * initialize SWS prevention. Then enter normal segment 1125 * processing: drop SYN, process data and FIN. 1126 * - otherwise do a normal 3-way handshake. 1127 */ 1128 if ((to.to_flag & TOF_CC) != 0) { 1129 if (((tp->t_flags & TF_NOPUSH) != 0) && 1130 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 1131 1132 taop->tao_cc = to.to_cc; 1133 tp->t_starttime = ticks; 1134 tp->t_state = TCPS_ESTABLISHED; 1135 1136 /* 1137 * If there is a FIN, or if there is data and the 1138 * connection is local, then delay SYN,ACK(SYN) in 1139 * the hope of piggy-backing it on a response 1140 * segment. Otherwise must send ACK now in case 1141 * the other side is slow starting. 1142 */ 1143 if (tcp_delack_enabled && ((thflags & TH_FIN) || 1144 (tlen != 0 && 1145#ifdef INET6 1146 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) 1147 || 1148 (!isipv6 && 1149#endif 1150 in_localaddr(inp->inp_faddr) 1151#ifdef INET6 1152 )) 1153#endif 1154 ))) { 1155 callout_reset(tp->tt_delack, tcp_delacktime, 1156 tcp_timer_delack, tp); 1157 tp->t_flags |= TF_NEEDSYN; 1158 } else 1159 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1160 1161 /* 1162 * Limit the `virtual advertised window' to TCP_MAXWIN 1163 * here. Even if we requested window scaling, it will 1164 * become effective only later when our SYN is acked. 1165 */ 1166 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 1167 tcpstat.tcps_connects++; 1168 soisconnected(so); 1169 callout_reset(tp->tt_keep, tcp_keepinit, 1170 tcp_timer_keep, tp); 1171 dropsocket = 0; /* committed to socket */ 1172 tcpstat.tcps_accepts++; 1173 goto trimthenstep6; 1174 } 1175 /* else do standard 3-way handshake */ 1176 } else { 1177 /* 1178 * No CC option, but maybe CC.NEW: 1179 * invalidate cached value. 1180 */ 1181 taop->tao_cc = 0; 1182 } 1183 /* 1184 * TAO test failed or there was no CC option, 1185 * do a standard 3-way handshake. 1186 */ 1187 tp->t_flags |= TF_ACKNOW; 1188 tp->t_state = TCPS_SYN_RECEIVED; 1189 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 1190 dropsocket = 0; /* committed to socket */ 1191 tcpstat.tcps_accepts++; 1192 ND6_HINT((struct tcpcb *)inp->inp_ppcb); 1193 goto trimthenstep6; 1194 } 1195 1196 /* 1197 * If the state is SYN_RECEIVED: 1198 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1199 */ 1200 case TCPS_SYN_RECEIVED: 1201 if ((thflags & TH_ACK) && 1202 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1203 SEQ_GT(th->th_ack, tp->snd_max))) 1204 goto maybedropwithreset; 1205 break; 1206 1207 /* 1208 * If the state is SYN_SENT: 1209 * if seg contains an ACK, but not for our SYN, drop the input. 1210 * if seg contains a RST, then drop the connection. 1211 * if seg does not contain SYN, then drop it. 1212 * Otherwise this is an acceptable SYN segment 1213 * initialize tp->rcv_nxt and tp->irs 1214 * if seg contains ack then advance tp->snd_una 1215 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1216 * arrange for segment to be acked (eventually) 1217 * continue processing rest of data/controls, beginning with URG 1218 */ 1219 case TCPS_SYN_SENT: 1220 if ((taop = tcp_gettaocache(inp)) == NULL) { 1221 taop = &tao_noncached; 1222 bzero(taop, sizeof(*taop)); 1223 } 1224 1225 if ((thflags & TH_ACK) && 1226 (SEQ_LEQ(th->th_ack, tp->iss) || 1227 SEQ_GT(th->th_ack, tp->snd_max))) { 1228 /* 1229 * If we have a cached CCsent for the remote host, 1230 * hence we haven't just crashed and restarted, 1231 * do not send a RST. This may be a retransmission 1232 * from the other side after our earlier ACK was lost. 1233 * Our new SYN, when it arrives, will serve as the 1234 * needed ACK. 1235 */ 1236 if (taop->tao_ccsent != 0) 1237 goto drop; 1238 else 1239 goto dropwithreset; 1240 } 1241 if (thflags & TH_RST) { 1242 if (thflags & TH_ACK) 1243 tp = tcp_drop(tp, ECONNREFUSED); 1244 goto drop; 1245 } 1246 if ((thflags & TH_SYN) == 0) 1247 goto drop; 1248 tp->snd_wnd = th->th_win; /* initial send window */ 1249 tp->cc_recv = to.to_cc; /* foreign CC */ 1250 1251 tp->irs = th->th_seq; 1252 tcp_rcvseqinit(tp); 1253 if (thflags & TH_ACK) { 1254 /* 1255 * Our SYN was acked. If segment contains CC.ECHO 1256 * option, check it to make sure this segment really 1257 * matches our SYN. If not, just drop it as old 1258 * duplicate, but send an RST if we're still playing 1259 * by the old rules. If no CC.ECHO option, make sure 1260 * we don't get fooled into using T/TCP. 1261 */ 1262 if (to.to_flag & TOF_CCECHO) { 1263 if (tp->cc_send != to.to_ccecho) { 1264 if (taop->tao_ccsent != 0) 1265 goto drop; 1266 else 1267 goto dropwithreset; 1268 } 1269 } else 1270 tp->t_flags &= ~TF_RCVD_CC; 1271 tcpstat.tcps_connects++; 1272 soisconnected(so); 1273 /* Do window scaling on this connection? */ 1274 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1275 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1276 tp->snd_scale = tp->requested_s_scale; 1277 tp->rcv_scale = tp->request_r_scale; 1278 } 1279 /* Segment is acceptable, update cache if undefined. */ 1280 if (taop->tao_ccsent == 0) 1281 taop->tao_ccsent = to.to_ccecho; 1282 1283 tp->rcv_adv += tp->rcv_wnd; 1284 tp->snd_una++; /* SYN is acked */ 1285 /* 1286 * If there's data, delay ACK; if there's also a FIN 1287 * ACKNOW will be turned on later. 1288 */ 1289 if (tcp_delack_enabled && tlen != 0) 1290 callout_reset(tp->tt_delack, tcp_delacktime, 1291 tcp_timer_delack, tp); 1292 else 1293 tp->t_flags |= TF_ACKNOW; 1294 /* 1295 * Received <SYN,ACK> in SYN_SENT[*] state. 1296 * Transitions: 1297 * SYN_SENT --> ESTABLISHED 1298 * SYN_SENT* --> FIN_WAIT_1 1299 */ 1300 tp->t_starttime = ticks; 1301 if (tp->t_flags & TF_NEEDFIN) { 1302 tp->t_state = TCPS_FIN_WAIT_1; 1303 tp->t_flags &= ~TF_NEEDFIN; 1304 thflags &= ~TH_SYN; 1305 } else { 1306 tp->t_state = TCPS_ESTABLISHED; 1307 callout_reset(tp->tt_keep, tcp_keepidle, 1308 tcp_timer_keep, tp); 1309 } 1310 } else { 1311 /* 1312 * Received initial SYN in SYN-SENT[*] state => simul- 1313 * taneous open. If segment contains CC option and there is 1314 * a cached CC, apply TAO test; if it succeeds, connection is 1315 * half-synchronized. Otherwise, do 3-way handshake: 1316 * SYN-SENT -> SYN-RECEIVED 1317 * SYN-SENT* -> SYN-RECEIVED* 1318 * If there was no CC option, clear cached CC value. 1319 */ 1320 tp->t_flags |= TF_ACKNOW; 1321 callout_stop(tp->tt_rexmt); 1322 if (to.to_flag & TOF_CC) { 1323 if (taop->tao_cc != 0 && 1324 CC_GT(to.to_cc, taop->tao_cc)) { 1325 /* 1326 * update cache and make transition: 1327 * SYN-SENT -> ESTABLISHED* 1328 * SYN-SENT* -> FIN-WAIT-1* 1329 */ 1330 taop->tao_cc = to.to_cc; 1331 tp->t_starttime = ticks; 1332 if (tp->t_flags & TF_NEEDFIN) { 1333 tp->t_state = TCPS_FIN_WAIT_1; 1334 tp->t_flags &= ~TF_NEEDFIN; 1335 } else { 1336 tp->t_state = TCPS_ESTABLISHED; 1337 callout_reset(tp->tt_keep, 1338 tcp_keepidle, 1339 tcp_timer_keep, 1340 tp); 1341 } 1342 tp->t_flags |= TF_NEEDSYN; 1343 } else 1344 tp->t_state = TCPS_SYN_RECEIVED; 1345 } else { 1346 /* CC.NEW or no option => invalidate cache */ 1347 taop->tao_cc = 0; 1348 tp->t_state = TCPS_SYN_RECEIVED; 1349 } 1350 } 1351 1352trimthenstep6: 1353 /* 1354 * Advance th->th_seq to correspond to first data byte. 1355 * If data, trim to stay within window, 1356 * dropping FIN if necessary. 1357 */ 1358 th->th_seq++; 1359 if (tlen > tp->rcv_wnd) { 1360 todrop = tlen - tp->rcv_wnd; 1361 m_adj(m, -todrop); 1362 tlen = tp->rcv_wnd; 1363 thflags &= ~TH_FIN; 1364 tcpstat.tcps_rcvpackafterwin++; 1365 tcpstat.tcps_rcvbyteafterwin += todrop; 1366 } 1367 tp->snd_wl1 = th->th_seq - 1; 1368 tp->rcv_up = th->th_seq; 1369 /* 1370 * Client side of transaction: already sent SYN and data. 1371 * If the remote host used T/TCP to validate the SYN, 1372 * our data will be ACK'd; if so, enter normal data segment 1373 * processing in the middle of step 5, ack processing. 1374 * Otherwise, goto step 6. 1375 */ 1376 if (thflags & TH_ACK) 1377 goto process_ACK; 1378 goto step6; 1379 /* 1380 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1381 * if segment contains a SYN and CC [not CC.NEW] option: 1382 * if state == TIME_WAIT and connection duration > MSL, 1383 * drop packet and send RST; 1384 * 1385 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1386 * ack the FIN (and data) in retransmission queue. 1387 * Complete close and delete TCPCB. Then reprocess 1388 * segment, hoping to find new TCPCB in LISTEN state; 1389 * 1390 * else must be old SYN; drop it. 1391 * else do normal processing. 1392 */ 1393 case TCPS_LAST_ACK: 1394 case TCPS_CLOSING: 1395 case TCPS_TIME_WAIT: 1396 if ((thflags & TH_SYN) && 1397 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 1398 if (tp->t_state == TCPS_TIME_WAIT && 1399 (ticks - tp->t_starttime) > tcp_msl) 1400 goto dropwithreset; 1401 if (CC_GT(to.to_cc, tp->cc_recv)) { 1402 tp = tcp_close(tp); 1403 goto findpcb; 1404 } 1405 else 1406 goto drop; 1407 } 1408 break; /* continue normal processing */ 1409 } 1410 1411 /* 1412 * States other than LISTEN or SYN_SENT. 1413 * First check the RST flag and sequence number since reset segments 1414 * are exempt from the timestamp and connection count tests. This 1415 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1416 * below which allowed reset segments in half the sequence space 1417 * to fall though and be processed (which gives forged reset 1418 * segments with a random sequence number a 50 percent chance of 1419 * killing a connection). 1420 * Then check timestamp, if present. 1421 * Then check the connection count, if present. 1422 * Then check that at least some bytes of segment are within 1423 * receive window. If segment begins before rcv_nxt, 1424 * drop leading data (and SYN); if nothing left, just ack. 1425 * 1426 * 1427 * If the RST bit is set, check the sequence number to see 1428 * if this is a valid reset segment. 1429 * RFC 793 page 37: 1430 * In all states except SYN-SENT, all reset (RST) segments 1431 * are validated by checking their SEQ-fields. A reset is 1432 * valid if its sequence number is in the window. 1433 * Note: this does not take into account delayed ACKs, so 1434 * we should test against last_ack_sent instead of rcv_nxt. 1435 * The sequence number in the reset segment is normally an 1436 * echo of our outgoing acknowlegement numbers, but some hosts 1437 * send a reset with the sequence number at the rightmost edge 1438 * of our receive window, and we have to handle this case. 1439 * If we have multiple segments in flight, the intial reset 1440 * segment sequence numbers will be to the left of last_ack_sent, 1441 * but they will eventually catch up. 1442 * In any case, it never made sense to trim reset segments to 1443 * fit the receive window since RFC 1122 says: 1444 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1445 * 1446 * A TCP SHOULD allow a received RST segment to include data. 1447 * 1448 * DISCUSSION 1449 * It has been suggested that a RST segment could contain 1450 * ASCII text that encoded and explained the cause of the 1451 * RST. No standard has yet been established for such 1452 * data. 1453 * 1454 * If the reset segment passes the sequence number test examine 1455 * the state: 1456 * SYN_RECEIVED STATE: 1457 * If passive open, return to LISTEN state. 1458 * If active open, inform user that connection was refused. 1459 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1460 * Inform user that connection was reset, and close tcb. 1461 * CLOSING, LAST_ACK STATES: 1462 * Close the tcb. 1463 * TIME_WAIT STATE: 1464 * Drop the segment - see Stevens, vol. 2, p. 964 and 1465 * RFC 1337. 1466 */ 1467 if (thflags & TH_RST) { 1468 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1469 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1470 switch (tp->t_state) { 1471 1472 case TCPS_SYN_RECEIVED: 1473 so->so_error = ECONNREFUSED; 1474 goto close; 1475 1476 case TCPS_ESTABLISHED: 1477 case TCPS_FIN_WAIT_1: 1478 case TCPS_FIN_WAIT_2: 1479 case TCPS_CLOSE_WAIT: 1480 so->so_error = ECONNRESET; 1481 close: 1482 tp->t_state = TCPS_CLOSED; 1483 tcpstat.tcps_drops++; 1484 tp = tcp_close(tp); 1485 break; 1486 1487 case TCPS_CLOSING: 1488 case TCPS_LAST_ACK: 1489 tp = tcp_close(tp); 1490 break; 1491 1492 case TCPS_TIME_WAIT: 1493 break; 1494 } 1495 } 1496 goto drop; 1497 } 1498 1499 /* 1500 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1501 * and it's less than ts_recent, drop it. 1502 */ 1503 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent && 1504 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1505 1506 /* Check to see if ts_recent is over 24 days old. */ 1507 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1508 /* 1509 * Invalidate ts_recent. If this segment updates 1510 * ts_recent, the age will be reset later and ts_recent 1511 * will get a valid value. If it does not, setting 1512 * ts_recent to zero will at least satisfy the 1513 * requirement that zero be placed in the timestamp 1514 * echo reply when ts_recent isn't valid. The 1515 * age isn't reset until we get a valid ts_recent 1516 * because we don't want out-of-order segments to be 1517 * dropped when ts_recent is old. 1518 */ 1519 tp->ts_recent = 0; 1520 } else { 1521 tcpstat.tcps_rcvduppack++; 1522 tcpstat.tcps_rcvdupbyte += tlen; 1523 tcpstat.tcps_pawsdrop++; 1524 goto dropafterack; 1525 } 1526 } 1527 1528 /* 1529 * T/TCP mechanism 1530 * If T/TCP was negotiated and the segment doesn't have CC, 1531 * or if its CC is wrong then drop the segment. 1532 * RST segments do not have to comply with this. 1533 */ 1534 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1535 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1536 goto dropafterack; 1537 1538 /* 1539 * In the SYN-RECEIVED state, validate that the packet belongs to 1540 * this connection before trimming the data to fit the receive 1541 * window. Check the sequence number versus IRS since we know 1542 * the sequence numbers haven't wrapped. This is a partial fix 1543 * for the "LAND" DoS attack. 1544 */ 1545 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) 1546 goto maybedropwithreset; 1547 1548 todrop = tp->rcv_nxt - th->th_seq; 1549 if (todrop > 0) { 1550 if (thflags & TH_SYN) { 1551 thflags &= ~TH_SYN; 1552 th->th_seq++; 1553 if (th->th_urp > 1) 1554 th->th_urp--; 1555 else 1556 thflags &= ~TH_URG; 1557 todrop--; 1558 } 1559 /* 1560 * Following if statement from Stevens, vol. 2, p. 960. 1561 */ 1562 if (todrop > tlen 1563 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1564 /* 1565 * Any valid FIN must be to the left of the window. 1566 * At this point the FIN must be a duplicate or out 1567 * of sequence; drop it. 1568 */ 1569 thflags &= ~TH_FIN; 1570 1571 /* 1572 * Send an ACK to resynchronize and drop any data. 1573 * But keep on processing for RST or ACK. 1574 */ 1575 tp->t_flags |= TF_ACKNOW; 1576 todrop = tlen; 1577 tcpstat.tcps_rcvduppack++; 1578 tcpstat.tcps_rcvdupbyte += todrop; 1579 } else { 1580 tcpstat.tcps_rcvpartduppack++; 1581 tcpstat.tcps_rcvpartdupbyte += todrop; 1582 } 1583 drop_hdrlen += todrop; /* drop from the top afterwards */ 1584 th->th_seq += todrop; 1585 tlen -= todrop; 1586 if (th->th_urp > todrop) 1587 th->th_urp -= todrop; 1588 else { 1589 thflags &= ~TH_URG; 1590 th->th_urp = 0; 1591 } 1592 } 1593 1594 /* 1595 * If new data are received on a connection after the 1596 * user processes are gone, then RST the other end. 1597 */ 1598 if ((so->so_state & SS_NOFDREF) && 1599 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1600 tp = tcp_close(tp); 1601 tcpstat.tcps_rcvafterclose++; 1602 goto dropwithreset; 1603 } 1604 1605 /* 1606 * If segment ends after window, drop trailing data 1607 * (and PUSH and FIN); if nothing left, just ACK. 1608 */ 1609 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1610 if (todrop > 0) { 1611 tcpstat.tcps_rcvpackafterwin++; 1612 if (todrop >= tlen) { 1613 tcpstat.tcps_rcvbyteafterwin += tlen; 1614 /* 1615 * If a new connection request is received 1616 * while in TIME_WAIT, drop the old connection 1617 * and start over if the sequence numbers 1618 * are above the previous ones. 1619 */ 1620 if (thflags & TH_SYN && 1621 tp->t_state == TCPS_TIME_WAIT && 1622 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1623 iss = tp->snd_nxt + TCP_ISSINCR; 1624 tp = tcp_close(tp); 1625 goto findpcb; 1626 } 1627 /* 1628 * If window is closed can only take segments at 1629 * window edge, and have to drop data and PUSH from 1630 * incoming segments. Continue processing, but 1631 * remember to ack. Otherwise, drop segment 1632 * and ack. 1633 */ 1634 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1635 tp->t_flags |= TF_ACKNOW; 1636 tcpstat.tcps_rcvwinprobe++; 1637 } else 1638 goto dropafterack; 1639 } else 1640 tcpstat.tcps_rcvbyteafterwin += todrop; 1641 m_adj(m, -todrop); 1642 tlen -= todrop; 1643 thflags &= ~(TH_PUSH|TH_FIN); 1644 } 1645 1646 /* 1647 * If last ACK falls within this segment's sequence numbers, 1648 * record its timestamp. 1649 * NOTE that the test is modified according to the latest 1650 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1651 */ 1652 if ((to.to_flag & TOF_TS) != 0 && 1653 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1654 tp->ts_recent_age = ticks; 1655 tp->ts_recent = to.to_tsval; 1656 } 1657 1658 /* 1659 * If a SYN is in the window, then this is an 1660 * error and we send an RST and drop the connection. 1661 */ 1662 if (thflags & TH_SYN) { 1663 tp = tcp_drop(tp, ECONNRESET); 1664 goto dropwithreset; 1665 } 1666 1667 /* 1668 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1669 * flag is on (half-synchronized state), then queue data for 1670 * later processing; else drop segment and return. 1671 */ 1672 if ((thflags & TH_ACK) == 0) { 1673 if (tp->t_state == TCPS_SYN_RECEIVED || 1674 (tp->t_flags & TF_NEEDSYN)) 1675 goto step6; 1676 else 1677 goto drop; 1678 } 1679 1680 /* 1681 * Ack processing. 1682 */ 1683 switch (tp->t_state) { 1684 1685 /* 1686 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1687 * ESTABLISHED state and continue processing. 1688 * The ACK was checked above. 1689 */ 1690 case TCPS_SYN_RECEIVED: 1691 1692 tcpstat.tcps_connects++; 1693 soisconnected(so); 1694 /* Do window scaling? */ 1695 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1696 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1697 tp->snd_scale = tp->requested_s_scale; 1698 tp->rcv_scale = tp->request_r_scale; 1699 } 1700 /* 1701 * Upon successful completion of 3-way handshake, 1702 * update cache.CC if it was undefined, pass any queued 1703 * data to the user, and advance state appropriately. 1704 */ 1705 if ((taop = tcp_gettaocache(inp)) != NULL && 1706 taop->tao_cc == 0) 1707 taop->tao_cc = tp->cc_recv; 1708 1709 /* 1710 * Make transitions: 1711 * SYN-RECEIVED -> ESTABLISHED 1712 * SYN-RECEIVED* -> FIN-WAIT-1 1713 */ 1714 tp->t_starttime = ticks; 1715 if (tp->t_flags & TF_NEEDFIN) { 1716 tp->t_state = TCPS_FIN_WAIT_1; 1717 tp->t_flags &= ~TF_NEEDFIN; 1718 } else { 1719 tp->t_state = TCPS_ESTABLISHED; 1720 callout_reset(tp->tt_keep, tcp_keepidle, 1721 tcp_timer_keep, tp); 1722 } 1723 /* 1724 * If segment contains data or ACK, will call tcp_reass() 1725 * later; if not, do so now to pass queued data to user. 1726 */ 1727 if (tlen == 0 && (thflags & TH_FIN) == 0) 1728 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1729 (struct mbuf *)0); 1730 tp->snd_wl1 = th->th_seq - 1; 1731 /* fall into ... */ 1732 1733 /* 1734 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1735 * ACKs. If the ack is in the range 1736 * tp->snd_una < th->th_ack <= tp->snd_max 1737 * then advance tp->snd_una to th->th_ack and drop 1738 * data from the retransmission queue. If this ACK reflects 1739 * more up to date window information we update our window information. 1740 */ 1741 case TCPS_ESTABLISHED: 1742 case TCPS_FIN_WAIT_1: 1743 case TCPS_FIN_WAIT_2: 1744 case TCPS_CLOSE_WAIT: 1745 case TCPS_CLOSING: 1746 case TCPS_LAST_ACK: 1747 case TCPS_TIME_WAIT: 1748 1749 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1750 if (tlen == 0 && tiwin == tp->snd_wnd) { 1751 tcpstat.tcps_rcvdupack++; 1752 /* 1753 * If we have outstanding data (other than 1754 * a window probe), this is a completely 1755 * duplicate ack (ie, window info didn't 1756 * change), the ack is the biggest we've 1757 * seen and we've seen exactly our rexmt 1758 * threshhold of them, assume a packet 1759 * has been dropped and retransmit it. 1760 * Kludge snd_nxt & the congestion 1761 * window so we send only this one 1762 * packet. 1763 * 1764 * We know we're losing at the current 1765 * window size so do congestion avoidance 1766 * (set ssthresh to half the current window 1767 * and pull our congestion window back to 1768 * the new ssthresh). 1769 * 1770 * Dup acks mean that packets have left the 1771 * network (they're now cached at the receiver) 1772 * so bump cwnd by the amount in the receiver 1773 * to keep a constant cwnd packets in the 1774 * network. 1775 */ 1776 if (!callout_active(tp->tt_rexmt) || 1777 th->th_ack != tp->snd_una) 1778 tp->t_dupacks = 0; 1779 else if (++tp->t_dupacks == tcprexmtthresh) { 1780 tcp_seq onxt = tp->snd_nxt; 1781 u_int win = 1782 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1783 tp->t_maxseg; 1784 1785 if (win < 2) 1786 win = 2; 1787 tp->snd_ssthresh = win * tp->t_maxseg; 1788 callout_stop(tp->tt_rexmt); 1789 tp->t_rtttime = 0; 1790 tp->snd_nxt = th->th_ack; 1791 tp->snd_cwnd = tp->t_maxseg; 1792 (void) tcp_output(tp); 1793 tp->snd_cwnd = tp->snd_ssthresh + 1794 tp->t_maxseg * tp->t_dupacks; 1795 if (SEQ_GT(onxt, tp->snd_nxt)) 1796 tp->snd_nxt = onxt; 1797 goto drop; 1798 } else if (tp->t_dupacks > tcprexmtthresh) { 1799 tp->snd_cwnd += tp->t_maxseg; 1800 (void) tcp_output(tp); 1801 goto drop; 1802 } 1803 } else 1804 tp->t_dupacks = 0; 1805 break; 1806 } 1807 /* 1808 * If the congestion window was inflated to account 1809 * for the other side's cached packets, retract it. 1810 */ 1811 if (tp->t_dupacks >= tcprexmtthresh && 1812 tp->snd_cwnd > tp->snd_ssthresh) 1813 tp->snd_cwnd = tp->snd_ssthresh; 1814 tp->t_dupacks = 0; 1815 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1816 tcpstat.tcps_rcvacktoomuch++; 1817 goto dropafterack; 1818 } 1819 /* 1820 * If we reach this point, ACK is not a duplicate, 1821 * i.e., it ACKs something we sent. 1822 */ 1823 if (tp->t_flags & TF_NEEDSYN) { 1824 /* 1825 * T/TCP: Connection was half-synchronized, and our 1826 * SYN has been ACK'd (so connection is now fully 1827 * synchronized). Go to non-starred state, 1828 * increment snd_una for ACK of SYN, and check if 1829 * we can do window scaling. 1830 */ 1831 tp->t_flags &= ~TF_NEEDSYN; 1832 tp->snd_una++; 1833 /* Do window scaling? */ 1834 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1835 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1836 tp->snd_scale = tp->requested_s_scale; 1837 tp->rcv_scale = tp->request_r_scale; 1838 } 1839 } 1840 1841process_ACK: 1842 acked = th->th_ack - tp->snd_una; 1843 tcpstat.tcps_rcvackpack++; 1844 tcpstat.tcps_rcvackbyte += acked; 1845 1846 /* 1847 * If we just performed our first retransmit, and the ACK 1848 * arrives within our recovery window, then it was a mistake 1849 * to do the retransmit in the first place. Recover our 1850 * original cwnd and ssthresh, and proceed to transmit where 1851 * we left off. 1852 */ 1853 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1854 tp->snd_cwnd = tp->snd_cwnd_prev; 1855 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1856 tp->snd_nxt = tp->snd_max; 1857 tp->t_badrxtwin = 0; /* XXX probably not required */ 1858 } 1859 1860 /* 1861 * If we have a timestamp reply, update smoothed 1862 * round trip time. If no timestamp is present but 1863 * transmit timer is running and timed sequence 1864 * number was acked, update smoothed round trip time. 1865 * Since we now have an rtt measurement, cancel the 1866 * timer backoff (cf., Phil Karn's retransmit alg.). 1867 * Recompute the initial retransmit timer. 1868 */ 1869 if (to.to_flag & TOF_TS) 1870 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1871 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1872 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1873 1874 /* 1875 * If all outstanding data is acked, stop retransmit 1876 * timer and remember to restart (more output or persist). 1877 * If there is more data to be acked, restart retransmit 1878 * timer, using current (possibly backed-off) value. 1879 */ 1880 if (th->th_ack == tp->snd_max) { 1881 callout_stop(tp->tt_rexmt); 1882 needoutput = 1; 1883 } else if (!callout_active(tp->tt_persist)) 1884 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1885 tcp_timer_rexmt, tp); 1886 1887 /* 1888 * If no data (only SYN) was ACK'd, 1889 * skip rest of ACK processing. 1890 */ 1891 if (acked == 0) 1892 goto step6; 1893 1894 /* 1895 * When new data is acked, open the congestion window. 1896 * If the window gives us less than ssthresh packets 1897 * in flight, open exponentially (maxseg per packet). 1898 * Otherwise open linearly: maxseg per window 1899 * (maxseg^2 / cwnd per packet). 1900 */ 1901 { 1902 register u_int cw = tp->snd_cwnd; 1903 register u_int incr = tp->t_maxseg; 1904 1905 if (cw > tp->snd_ssthresh) 1906 incr = incr * incr / cw; 1907 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale); 1908 } 1909 if (acked > so->so_snd.sb_cc) { 1910 tp->snd_wnd -= so->so_snd.sb_cc; 1911 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1912 ourfinisacked = 1; 1913 } else { 1914 sbdrop(&so->so_snd, acked); 1915 tp->snd_wnd -= acked; 1916 ourfinisacked = 0; 1917 } 1918 sowwakeup(so); 1919 tp->snd_una = th->th_ack; 1920 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1921 tp->snd_nxt = tp->snd_una; 1922 1923 switch (tp->t_state) { 1924 1925 /* 1926 * In FIN_WAIT_1 STATE in addition to the processing 1927 * for the ESTABLISHED state if our FIN is now acknowledged 1928 * then enter FIN_WAIT_2. 1929 */ 1930 case TCPS_FIN_WAIT_1: 1931 if (ourfinisacked) { 1932 /* 1933 * If we can't receive any more 1934 * data, then closing user can proceed. 1935 * Starting the timer is contrary to the 1936 * specification, but if we don't get a FIN 1937 * we'll hang forever. 1938 */ 1939 if (so->so_state & SS_CANTRCVMORE) { 1940 soisdisconnected(so); 1941 callout_reset(tp->tt_2msl, tcp_maxidle, 1942 tcp_timer_2msl, tp); 1943 } 1944 tp->t_state = TCPS_FIN_WAIT_2; 1945 } 1946 break; 1947 1948 /* 1949 * In CLOSING STATE in addition to the processing for 1950 * the ESTABLISHED state if the ACK acknowledges our FIN 1951 * then enter the TIME-WAIT state, otherwise ignore 1952 * the segment. 1953 */ 1954 case TCPS_CLOSING: 1955 if (ourfinisacked) { 1956 tp->t_state = TCPS_TIME_WAIT; 1957 tcp_canceltimers(tp); 1958 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1959 if (tp->cc_recv != 0 && 1960 (ticks - tp->t_starttime) < tcp_msl) 1961 callout_reset(tp->tt_2msl, 1962 tp->t_rxtcur * 1963 TCPTV_TWTRUNC, 1964 tcp_timer_2msl, tp); 1965 else 1966 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1967 tcp_timer_2msl, tp); 1968 soisdisconnected(so); 1969 } 1970 break; 1971 1972 /* 1973 * In LAST_ACK, we may still be waiting for data to drain 1974 * and/or to be acked, as well as for the ack of our FIN. 1975 * If our FIN is now acknowledged, delete the TCB, 1976 * enter the closed state and return. 1977 */ 1978 case TCPS_LAST_ACK: 1979 if (ourfinisacked) { 1980 tp = tcp_close(tp); 1981 goto drop; 1982 } 1983 break; 1984 1985 /* 1986 * In TIME_WAIT state the only thing that should arrive 1987 * is a retransmission of the remote FIN. Acknowledge 1988 * it and restart the finack timer. 1989 */ 1990 case TCPS_TIME_WAIT: 1991 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1992 tcp_timer_2msl, tp); 1993 goto dropafterack; 1994 } 1995 } 1996 1997step6: 1998 /* 1999 * Update window information. 2000 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2001 */ 2002 if ((thflags & TH_ACK) && 2003 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2004 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2005 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2006 /* keep track of pure window updates */ 2007 if (tlen == 0 && 2008 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2009 tcpstat.tcps_rcvwinupd++; 2010 tp->snd_wnd = tiwin; 2011 tp->snd_wl1 = th->th_seq; 2012 tp->snd_wl2 = th->th_ack; 2013 if (tp->snd_wnd > tp->max_sndwnd) 2014 tp->max_sndwnd = tp->snd_wnd; 2015 needoutput = 1; 2016 } 2017 2018 /* 2019 * Process segments with URG. 2020 */ 2021 if ((thflags & TH_URG) && th->th_urp && 2022 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2023 /* 2024 * This is a kludge, but if we receive and accept 2025 * random urgent pointers, we'll crash in 2026 * soreceive. It's hard to imagine someone 2027 * actually wanting to send this much urgent data. 2028 */ 2029 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2030 th->th_urp = 0; /* XXX */ 2031 thflags &= ~TH_URG; /* XXX */ 2032 goto dodata; /* XXX */ 2033 } 2034 /* 2035 * If this segment advances the known urgent pointer, 2036 * then mark the data stream. This should not happen 2037 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2038 * a FIN has been received from the remote side. 2039 * In these states we ignore the URG. 2040 * 2041 * According to RFC961 (Assigned Protocols), 2042 * the urgent pointer points to the last octet 2043 * of urgent data. We continue, however, 2044 * to consider it to indicate the first octet 2045 * of data past the urgent section as the original 2046 * spec states (in one of two places). 2047 */ 2048 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2049 tp->rcv_up = th->th_seq + th->th_urp; 2050 so->so_oobmark = so->so_rcv.sb_cc + 2051 (tp->rcv_up - tp->rcv_nxt) - 1; 2052 if (so->so_oobmark == 0) 2053 so->so_state |= SS_RCVATMARK; 2054 sohasoutofband(so); 2055 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2056 } 2057 /* 2058 * Remove out of band data so doesn't get presented to user. 2059 * This can happen independent of advancing the URG pointer, 2060 * but if two URG's are pending at once, some out-of-band 2061 * data may creep in... ick. 2062 */ 2063 if (th->th_urp <= (u_long)tlen 2064#ifdef SO_OOBINLINE 2065 && (so->so_options & SO_OOBINLINE) == 0 2066#endif 2067 ) 2068 tcp_pulloutofband(so, th, m, 2069 drop_hdrlen); /* hdr drop is delayed */ 2070 } else 2071 /* 2072 * If no out of band data is expected, 2073 * pull receive urgent pointer along 2074 * with the receive window. 2075 */ 2076 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2077 tp->rcv_up = tp->rcv_nxt; 2078dodata: /* XXX */ 2079 2080 /* 2081 * Process the segment text, merging it into the TCP sequencing queue, 2082 * and arranging for acknowledgment of receipt if necessary. 2083 * This process logically involves adjusting tp->rcv_wnd as data 2084 * is presented to the user (this happens in tcp_usrreq.c, 2085 * case PRU_RCVD). If a FIN has already been received on this 2086 * connection then we just ignore the text. 2087 */ 2088 if ((tlen || (thflags&TH_FIN)) && 2089 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2090 m_adj(m, drop_hdrlen); /* delayed header drop */ 2091 TCP_REASS(tp, th, &tlen, m, so, thflags); 2092 /* 2093 * Note the amount of data that peer has sent into 2094 * our window, in order to estimate the sender's 2095 * buffer size. 2096 */ 2097 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2098 } else { 2099 m_freem(m); 2100 thflags &= ~TH_FIN; 2101 } 2102 2103 /* 2104 * If FIN is received ACK the FIN and let the user know 2105 * that the connection is closing. 2106 */ 2107 if (thflags & TH_FIN) { 2108 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2109 socantrcvmore(so); 2110 /* 2111 * If connection is half-synchronized 2112 * (ie NEEDSYN flag on) then delay ACK, 2113 * so it may be piggybacked when SYN is sent. 2114 * Otherwise, since we received a FIN then no 2115 * more input can be expected, send ACK now. 2116 */ 2117 if (tcp_delack_enabled && (tp->t_flags & TF_NEEDSYN)) 2118 callout_reset(tp->tt_delack, tcp_delacktime, 2119 tcp_timer_delack, tp); 2120 else 2121 tp->t_flags |= TF_ACKNOW; 2122 tp->rcv_nxt++; 2123 } 2124 switch (tp->t_state) { 2125 2126 /* 2127 * In SYN_RECEIVED and ESTABLISHED STATES 2128 * enter the CLOSE_WAIT state. 2129 */ 2130 case TCPS_SYN_RECEIVED: 2131 tp->t_starttime = ticks; 2132 /*FALLTHROUGH*/ 2133 case TCPS_ESTABLISHED: 2134 tp->t_state = TCPS_CLOSE_WAIT; 2135 break; 2136 2137 /* 2138 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2139 * enter the CLOSING state. 2140 */ 2141 case TCPS_FIN_WAIT_1: 2142 tp->t_state = TCPS_CLOSING; 2143 break; 2144 2145 /* 2146 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2147 * starting the time-wait timer, turning off the other 2148 * standard timers. 2149 */ 2150 case TCPS_FIN_WAIT_2: 2151 tp->t_state = TCPS_TIME_WAIT; 2152 tcp_canceltimers(tp); 2153 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2154 if (tp->cc_recv != 0 && 2155 (ticks - tp->t_starttime) < tcp_msl) { 2156 callout_reset(tp->tt_2msl, 2157 tp->t_rxtcur * TCPTV_TWTRUNC, 2158 tcp_timer_2msl, tp); 2159 /* For transaction client, force ACK now. */ 2160 tp->t_flags |= TF_ACKNOW; 2161 } 2162 else 2163 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2164 tcp_timer_2msl, tp); 2165 soisdisconnected(so); 2166 break; 2167 2168 /* 2169 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2170 */ 2171 case TCPS_TIME_WAIT: 2172 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2173 tcp_timer_2msl, tp); 2174 break; 2175 } 2176 } 2177#ifdef TCPDEBUG 2178 if (so->so_options & SO_DEBUG) 2179 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2180 &tcp_savetcp, 0); 2181#endif 2182 2183 /* 2184 * Return any desired output. 2185 */ 2186 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2187 (void) tcp_output(tp); 2188 return; 2189 2190dropafterack: 2191 /* 2192 * Generate an ACK dropping incoming segment if it occupies 2193 * sequence space, where the ACK reflects our state. 2194 * 2195 * We can now skip the test for the RST flag since all 2196 * paths to this code happen after packets containing 2197 * RST have been dropped. 2198 * 2199 * In the SYN-RECEIVED state, don't send an ACK unless the 2200 * segment we received passes the SYN-RECEIVED ACK test. 2201 * If it fails send a RST. This breaks the loop in the 2202 * "LAND" DoS attack, and also prevents an ACK storm 2203 * between two listening ports that have been sent forged 2204 * SYN segments, each with the source address of the other. 2205 */ 2206 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2207 (SEQ_GT(tp->snd_una, th->th_ack) || 2208 SEQ_GT(th->th_ack, tp->snd_max)) ) 2209 goto maybedropwithreset; 2210#ifdef TCPDEBUG 2211 if (so->so_options & SO_DEBUG) 2212 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2213 &tcp_savetcp, 0); 2214#endif 2215 m_freem(m); 2216 tp->t_flags |= TF_ACKNOW; 2217 (void) tcp_output(tp); 2218 return; 2219 2220 2221 /* 2222 * Conditionally drop with reset or just drop depending on whether 2223 * we think we are under attack or not. 2224 */ 2225maybedropwithreset: 2226#ifdef ICMP_BANDLIM 2227 if (badport_bandlim(1) < 0) 2228 goto drop; 2229#endif 2230 /* fall through */ 2231dropwithreset: 2232#ifdef TCP_RESTRICT_RST 2233 if (restrict_rst) 2234 goto drop; 2235#endif 2236 /* 2237 * Generate a RST, dropping incoming segment. 2238 * Make ACK acceptable to originator of segment. 2239 * Don't bother to respond if destination was broadcast/multicast. 2240 */ 2241 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2242 goto drop; 2243#ifdef INET6 2244 if (isipv6) { 2245 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2246 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2247 goto drop; 2248 } else 2249#endif /* INET6 */ 2250 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2251 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2252 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 2253 goto drop; 2254 /* IPv6 anycast check is done at tcp6_input() */ 2255#ifdef TCPDEBUG 2256 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2257 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2258 &tcp_savetcp, 0); 2259#endif 2260 if (thflags & TH_ACK) 2261 /* mtod() below is safe as long as hdr dropping is delayed */ 2262 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2263 TH_RST); 2264 else { 2265 if (thflags & TH_SYN) 2266 tlen++; 2267 /* mtod() below is safe as long as hdr dropping is delayed */ 2268 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2269 (tcp_seq)0, TH_RST|TH_ACK); 2270 } 2271 /* destroy temporarily created socket */ 2272 if (dropsocket) 2273 (void) soabort(so); 2274 return; 2275 2276drop: 2277 /* 2278 * Drop space held by incoming segment and return. 2279 */ 2280#ifdef TCPDEBUG 2281 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2282 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2283 &tcp_savetcp, 0); 2284#endif 2285 m_freem(m); 2286 /* destroy temporarily created socket */ 2287 if (dropsocket) 2288 (void) soabort(so); 2289 return; 2290} 2291 2292static void 2293tcp_dooptions(tp, cp, cnt, th, to) 2294 struct tcpcb *tp; 2295 u_char *cp; 2296 int cnt; 2297 struct tcphdr *th; 2298 struct tcpopt *to; 2299{ 2300 u_short mss = 0; 2301 int opt, optlen; 2302 2303 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2304 opt = cp[0]; 2305 if (opt == TCPOPT_EOL) 2306 break; 2307 if (opt == TCPOPT_NOP) 2308 optlen = 1; 2309 else { 2310 optlen = cp[1]; 2311 if (optlen <= 0) 2312 break; 2313 } 2314 switch (opt) { 2315 2316 default: 2317 continue; 2318 2319 case TCPOPT_MAXSEG: 2320 if (optlen != TCPOLEN_MAXSEG) 2321 continue; 2322 if (!(th->th_flags & TH_SYN)) 2323 continue; 2324 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 2325 NTOHS(mss); 2326 break; 2327 2328 case TCPOPT_WINDOW: 2329 if (optlen != TCPOLEN_WINDOW) 2330 continue; 2331 if (!(th->th_flags & TH_SYN)) 2332 continue; 2333 tp->t_flags |= TF_RCVD_SCALE; 2334 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2335 break; 2336 2337 case TCPOPT_TIMESTAMP: 2338 if (optlen != TCPOLEN_TIMESTAMP) 2339 continue; 2340 to->to_flag |= TOF_TS; 2341 bcopy((char *)cp + 2, 2342 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2343 NTOHL(to->to_tsval); 2344 bcopy((char *)cp + 6, 2345 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2346 NTOHL(to->to_tsecr); 2347 2348 /* 2349 * A timestamp received in a SYN makes 2350 * it ok to send timestamp requests and replies. 2351 */ 2352 if (th->th_flags & TH_SYN) { 2353 tp->t_flags |= TF_RCVD_TSTMP; 2354 tp->ts_recent = to->to_tsval; 2355 tp->ts_recent_age = ticks; 2356 } 2357 break; 2358 case TCPOPT_CC: 2359 if (optlen != TCPOLEN_CC) 2360 continue; 2361 to->to_flag |= TOF_CC; 2362 bcopy((char *)cp + 2, 2363 (char *)&to->to_cc, sizeof(to->to_cc)); 2364 NTOHL(to->to_cc); 2365 /* 2366 * A CC or CC.new option received in a SYN makes 2367 * it ok to send CC in subsequent segments. 2368 */ 2369 if (th->th_flags & TH_SYN) 2370 tp->t_flags |= TF_RCVD_CC; 2371 break; 2372 case TCPOPT_CCNEW: 2373 if (optlen != TCPOLEN_CC) 2374 continue; 2375 if (!(th->th_flags & TH_SYN)) 2376 continue; 2377 to->to_flag |= TOF_CCNEW; 2378 bcopy((char *)cp + 2, 2379 (char *)&to->to_cc, sizeof(to->to_cc)); 2380 NTOHL(to->to_cc); 2381 /* 2382 * A CC or CC.new option received in a SYN makes 2383 * it ok to send CC in subsequent segments. 2384 */ 2385 tp->t_flags |= TF_RCVD_CC; 2386 break; 2387 case TCPOPT_CCECHO: 2388 if (optlen != TCPOLEN_CC) 2389 continue; 2390 if (!(th->th_flags & TH_SYN)) 2391 continue; 2392 to->to_flag |= TOF_CCECHO; 2393 bcopy((char *)cp + 2, 2394 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2395 NTOHL(to->to_ccecho); 2396 break; 2397 } 2398 } 2399 if (th->th_flags & TH_SYN) 2400 tcp_mss(tp, mss); /* sets t_maxseg */ 2401} 2402 2403/* 2404 * Pull out of band byte out of a segment so 2405 * it doesn't appear in the user's data queue. 2406 * It is still reflected in the segment length for 2407 * sequencing purposes. 2408 */ 2409static void 2410tcp_pulloutofband(so, th, m, off) 2411 struct socket *so; 2412 struct tcphdr *th; 2413 register struct mbuf *m; 2414 int off; /* delayed to be droped hdrlen */ 2415{ 2416 int cnt = off + th->th_urp - 1; 2417 2418 while (cnt >= 0) { 2419 if (m->m_len > cnt) { 2420 char *cp = mtod(m, caddr_t) + cnt; 2421 struct tcpcb *tp = sototcpcb(so); 2422 2423 tp->t_iobc = *cp; 2424 tp->t_oobflags |= TCPOOB_HAVEDATA; 2425 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2426 m->m_len--; 2427 if (m->m_flags & M_PKTHDR) 2428 m->m_pkthdr.len--; 2429 return; 2430 } 2431 cnt -= m->m_len; 2432 m = m->m_next; 2433 if (m == 0) 2434 break; 2435 } 2436 panic("tcp_pulloutofband"); 2437} 2438 2439/* 2440 * Collect new round-trip time estimate 2441 * and update averages and current timeout. 2442 */ 2443static void 2444tcp_xmit_timer(tp, rtt) 2445 register struct tcpcb *tp; 2446 int rtt; 2447{ 2448 register int delta; 2449 2450 tcpstat.tcps_rttupdated++; 2451 tp->t_rttupdated++; 2452 if (tp->t_srtt != 0) { 2453 /* 2454 * srtt is stored as fixed point with 5 bits after the 2455 * binary point (i.e., scaled by 8). The following magic 2456 * is equivalent to the smoothing algorithm in rfc793 with 2457 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2458 * point). Adjust rtt to origin 0. 2459 */ 2460 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2461 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2462 2463 if ((tp->t_srtt += delta) <= 0) 2464 tp->t_srtt = 1; 2465 2466 /* 2467 * We accumulate a smoothed rtt variance (actually, a 2468 * smoothed mean difference), then set the retransmit 2469 * timer to smoothed rtt + 4 times the smoothed variance. 2470 * rttvar is stored as fixed point with 4 bits after the 2471 * binary point (scaled by 16). The following is 2472 * equivalent to rfc793 smoothing with an alpha of .75 2473 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2474 * rfc793's wired-in beta. 2475 */ 2476 if (delta < 0) 2477 delta = -delta; 2478 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2479 if ((tp->t_rttvar += delta) <= 0) 2480 tp->t_rttvar = 1; 2481 } else { 2482 /* 2483 * No rtt measurement yet - use the unsmoothed rtt. 2484 * Set the variance to half the rtt (so our first 2485 * retransmit happens at 3*rtt). 2486 */ 2487 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2488 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2489 } 2490 tp->t_rtttime = 0; 2491 tp->t_rxtshift = 0; 2492 2493 /* 2494 * the retransmit should happen at rtt + 4 * rttvar. 2495 * Because of the way we do the smoothing, srtt and rttvar 2496 * will each average +1/2 tick of bias. When we compute 2497 * the retransmit timer, we want 1/2 tick of rounding and 2498 * 1 extra tick because of +-1/2 tick uncertainty in the 2499 * firing of the timer. The bias will give us exactly the 2500 * 1.5 tick we need. But, because the bias is 2501 * statistical, we have to test that we don't drop below 2502 * the minimum feasible timer (which is 2 ticks). 2503 */ 2504 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2505 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2506 2507 /* 2508 * We received an ack for a packet that wasn't retransmitted; 2509 * it is probably safe to discard any error indications we've 2510 * received recently. This isn't quite right, but close enough 2511 * for now (a route might have failed after we sent a segment, 2512 * and the return path might not be symmetrical). 2513 */ 2514 tp->t_softerror = 0; 2515} 2516 2517/* 2518 * Determine a reasonable value for maxseg size. 2519 * If the route is known, check route for mtu. 2520 * If none, use an mss that can be handled on the outgoing 2521 * interface without forcing IP to fragment; if bigger than 2522 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2523 * to utilize large mbufs. If no route is found, route has no mtu, 2524 * or the destination isn't local, use a default, hopefully conservative 2525 * size (usually 512 or the default IP max size, but no more than the mtu 2526 * of the interface), as we can't discover anything about intervening 2527 * gateways or networks. We also initialize the congestion/slow start 2528 * window to be a single segment if the destination isn't local. 2529 * While looking at the routing entry, we also initialize other path-dependent 2530 * parameters from pre-set or cached values in the routing entry. 2531 * 2532 * Also take into account the space needed for options that we 2533 * send regularly. Make maxseg shorter by that amount to assure 2534 * that we can send maxseg amount of data even when the options 2535 * are present. Store the upper limit of the length of options plus 2536 * data in maxopd. 2537 * 2538 * NOTE that this routine is only called when we process an incoming 2539 * segment, for outgoing segments only tcp_mssopt is called. 2540 * 2541 * In case of T/TCP, we call this routine during implicit connection 2542 * setup as well (offer = -1), to initialize maxseg from the cached 2543 * MSS of our peer. 2544 */ 2545void 2546tcp_mss(tp, offer) 2547 struct tcpcb *tp; 2548 int offer; 2549{ 2550 register struct rtentry *rt; 2551 struct ifnet *ifp; 2552 register int rtt, mss; 2553 u_long bufsize; 2554 struct inpcb *inp; 2555 struct socket *so; 2556 struct rmxp_tao *taop; 2557 int origoffer = offer; 2558#ifdef INET6 2559 int isipv6; 2560 int min_protoh; 2561#endif 2562 2563 inp = tp->t_inpcb; 2564#ifdef INET6 2565 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2566 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2567 : sizeof (struct tcpiphdr); 2568#else 2569#define min_protoh (sizeof (struct tcpiphdr)) 2570#endif 2571#ifdef INET6 2572 if (isipv6) 2573 rt = tcp_rtlookup6(inp); 2574 else 2575#endif 2576 rt = tcp_rtlookup(inp); 2577 if (rt == NULL) { 2578 tp->t_maxopd = tp->t_maxseg = 2579#ifdef INET6 2580 isipv6 ? tcp_v6mssdflt : 2581#endif /* INET6 */ 2582 tcp_mssdflt; 2583 return; 2584 } 2585 ifp = rt->rt_ifp; 2586 so = inp->inp_socket; 2587 2588 taop = rmx_taop(rt->rt_rmx); 2589 /* 2590 * Offer == -1 means that we didn't receive SYN yet, 2591 * use cached value in that case; 2592 */ 2593 if (offer == -1) 2594 offer = taop->tao_mssopt; 2595 /* 2596 * Offer == 0 means that there was no MSS on the SYN segment, 2597 * in this case we use tcp_mssdflt. 2598 */ 2599 if (offer == 0) 2600 offer = 2601#ifdef INET6 2602 isipv6 ? tcp_v6mssdflt : 2603#endif /* INET6 */ 2604 tcp_mssdflt; 2605 else 2606 /* 2607 * Sanity check: make sure that maxopd will be large 2608 * enough to allow some data on segments even is the 2609 * all the option space is used (40bytes). Otherwise 2610 * funny things may happen in tcp_output. 2611 */ 2612 offer = max(offer, 64); 2613 taop->tao_mssopt = offer; 2614 2615 /* 2616 * While we're here, check if there's an initial rtt 2617 * or rttvar. Convert from the route-table units 2618 * to scaled multiples of the slow timeout timer. 2619 */ 2620 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2621 /* 2622 * XXX the lock bit for RTT indicates that the value 2623 * is also a minimum value; this is subject to time. 2624 */ 2625 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2626 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2627 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2628 tcpstat.tcps_usedrtt++; 2629 if (rt->rt_rmx.rmx_rttvar) { 2630 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2631 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2632 tcpstat.tcps_usedrttvar++; 2633 } else { 2634 /* default variation is +- 1 rtt */ 2635 tp->t_rttvar = 2636 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2637 } 2638 TCPT_RANGESET(tp->t_rxtcur, 2639 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2640 tp->t_rttmin, TCPTV_REXMTMAX); 2641 } 2642 /* 2643 * if there's an mtu associated with the route, use it 2644 * else, use the link mtu. 2645 */ 2646 if (rt->rt_rmx.rmx_mtu) 2647 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2648 else 2649 { 2650 mss = 2651#ifdef INET6 2652 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2653#endif 2654 ifp->if_mtu 2655#ifdef INET6 2656 ) 2657#endif 2658 - min_protoh; 2659#ifdef INET6 2660 if (isipv6) { 2661 if (!in6_localaddr(&inp->in6p_faddr)) 2662 mss = min(mss, tcp_v6mssdflt); 2663 } else 2664#endif 2665 if (!in_localaddr(inp->inp_faddr)) 2666 mss = min(mss, tcp_mssdflt); 2667 } 2668 mss = min(mss, offer); 2669 /* 2670 * maxopd stores the maximum length of data AND options 2671 * in a segment; maxseg is the amount of data in a normal 2672 * segment. We need to store this value (maxopd) apart 2673 * from maxseg, because now every segment carries options 2674 * and thus we normally have somewhat less data in segments. 2675 */ 2676 tp->t_maxopd = mss; 2677 2678 /* 2679 * In case of T/TCP, origoffer==-1 indicates, that no segments 2680 * were received yet. In this case we just guess, otherwise 2681 * we do the same as before T/TCP. 2682 */ 2683 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2684 (origoffer == -1 || 2685 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2686 mss -= TCPOLEN_TSTAMP_APPA; 2687 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2688 (origoffer == -1 || 2689 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2690 mss -= TCPOLEN_CC_APPA; 2691 2692#if (MCLBYTES & (MCLBYTES - 1)) == 0 2693 if (mss > MCLBYTES) 2694 mss &= ~(MCLBYTES-1); 2695#else 2696 if (mss > MCLBYTES) 2697 mss = mss / MCLBYTES * MCLBYTES; 2698#endif 2699 /* 2700 * If there's a pipesize, change the socket buffer 2701 * to that size. Make the socket buffers an integral 2702 * number of mss units; if the mss is larger than 2703 * the socket buffer, decrease the mss. 2704 */ 2705#ifdef RTV_SPIPE 2706 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2707#endif 2708 bufsize = so->so_snd.sb_hiwat; 2709 if (bufsize < mss) 2710 mss = bufsize; 2711 else { 2712 bufsize = roundup(bufsize, mss); 2713 if (bufsize > sb_max) 2714 bufsize = sb_max; 2715 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2716 } 2717 tp->t_maxseg = mss; 2718 2719#ifdef RTV_RPIPE 2720 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2721#endif 2722 bufsize = so->so_rcv.sb_hiwat; 2723 if (bufsize > mss) { 2724 bufsize = roundup(bufsize, mss); 2725 if (bufsize > sb_max) 2726 bufsize = sb_max; 2727 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2728 } 2729 2730 /* 2731 * Set the slow-start flight size depending on whether this 2732 * is a local network or not. 2733 */ 2734 if ( 2735#ifdef INET6 2736 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2737 (!isipv6 && 2738#endif 2739 in_localaddr(inp->inp_faddr) 2740#ifdef INET6 2741 ) 2742#endif 2743 ) 2744 tp->snd_cwnd = mss * ss_fltsz_local; 2745 else 2746 tp->snd_cwnd = mss * ss_fltsz; 2747 2748 if (rt->rt_rmx.rmx_ssthresh) { 2749 /* 2750 * There's some sort of gateway or interface 2751 * buffer limit on the path. Use this to set 2752 * the slow start threshhold, but set the 2753 * threshold to no less than 2*mss. 2754 */ 2755 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2756 tcpstat.tcps_usedssthresh++; 2757 } 2758} 2759 2760/* 2761 * Determine the MSS option to send on an outgoing SYN. 2762 */ 2763int 2764tcp_mssopt(tp) 2765 struct tcpcb *tp; 2766{ 2767 struct rtentry *rt; 2768#ifdef INET6 2769 int isipv6; 2770 int min_protoh; 2771#endif 2772 2773#ifdef INET6 2774 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2775 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2776 : sizeof (struct tcpiphdr); 2777#else 2778#define min_protoh (sizeof (struct tcpiphdr)) 2779#endif 2780#ifdef INET6 2781 if (isipv6) 2782 rt = tcp_rtlookup6(tp->t_inpcb); 2783 else 2784#endif /* INET6 */ 2785 rt = tcp_rtlookup(tp->t_inpcb); 2786 if (rt == NULL) 2787 return 2788#ifdef INET6 2789 isipv6 ? tcp_v6mssdflt : 2790#endif /* INET6 */ 2791 tcp_mssdflt; 2792 2793 return rt->rt_ifp->if_mtu - min_protoh; 2794} 2795