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