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 #include <machine/in_cksum.h> 104 105 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 106 107 static int tcprexmtthresh = 3; 108 tcp_seq tcp_iss; 109 tcp_cc tcp_ccgen; 110 111 struct tcpstat tcpstat; 112 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD, 113 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 114 115 static int log_in_vain = 0; 116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 117 &log_in_vain, 0, "Log all incoming TCP connections"); 118 119 static int blackhole = 0; 120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 121 &blackhole, 0, "Do not send RST when dropping refused connections"); 122 123 int tcp_delack_enabled = 1; 124 SYSCTL_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 129 static int drop_synfin = 0; 130 SYSCTL_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 135 static int restrict_rst = 0; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, restrict_rst, CTLFLAG_RW, 137 &restrict_rst, 0, "Restrict RST emission"); 138 #endif 139 140 struct inpcbhead tcb; 141 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 142 struct inpcbinfo tcbinfo; 143 144 static void tcp_dooptions __P((struct tcpcb *, 145 u_char *, int, struct tcphdr *, struct tcpopt *)); 146 static void tcp_pulloutofband __P((struct socket *, 147 struct tcphdr *, struct mbuf *, int)); 148 static int tcp_reass __P((struct tcpcb *, struct tcphdr *, int *, 149 struct mbuf *)); 150 static 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) \ 155 do { \ 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 } 196 197 static int 198 tcp_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 } 299 300 present: 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 332 int 333 tcp6_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 358 359 void 360 tcp_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 */ 528 findpcb: 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 ((inp->inp_flags & IN6P_BINDV6ONLY) == 0) { 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 goto trimthenstep6; 1193 } 1194 1195 /* 1196 * If the state is SYN_RECEIVED: 1197 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1198 */ 1199 case TCPS_SYN_RECEIVED: 1200 if ((thflags & TH_ACK) && 1201 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1202 SEQ_GT(th->th_ack, tp->snd_max))) 1203 goto maybedropwithreset; 1204 break; 1205 1206 /* 1207 * If the state is SYN_SENT: 1208 * if seg contains an ACK, but not for our SYN, drop the input. 1209 * if seg contains a RST, then drop the connection. 1210 * if seg does not contain SYN, then drop it. 1211 * Otherwise this is an acceptable SYN segment 1212 * initialize tp->rcv_nxt and tp->irs 1213 * if seg contains ack then advance tp->snd_una 1214 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1215 * arrange for segment to be acked (eventually) 1216 * continue processing rest of data/controls, beginning with URG 1217 */ 1218 case TCPS_SYN_SENT: 1219 if ((taop = tcp_gettaocache(inp)) == NULL) { 1220 taop = &tao_noncached; 1221 bzero(taop, sizeof(*taop)); 1222 } 1223 1224 if ((thflags & TH_ACK) && 1225 (SEQ_LEQ(th->th_ack, tp->iss) || 1226 SEQ_GT(th->th_ack, tp->snd_max))) { 1227 /* 1228 * If we have a cached CCsent for the remote host, 1229 * hence we haven't just crashed and restarted, 1230 * do not send a RST. This may be a retransmission 1231 * from the other side after our earlier ACK was lost. 1232 * Our new SYN, when it arrives, will serve as the 1233 * needed ACK. 1234 */ 1235 if (taop->tao_ccsent != 0) 1236 goto drop; 1237 else 1238 goto dropwithreset; 1239 } 1240 if (thflags & TH_RST) { 1241 if (thflags & TH_ACK) 1242 tp = tcp_drop(tp, ECONNREFUSED); 1243 goto drop; 1244 } 1245 if ((thflags & TH_SYN) == 0) 1246 goto drop; 1247 tp->snd_wnd = th->th_win; /* initial send window */ 1248 tp->cc_recv = to.to_cc; /* foreign CC */ 1249 1250 tp->irs = th->th_seq; 1251 tcp_rcvseqinit(tp); 1252 if (thflags & TH_ACK) { 1253 /* 1254 * Our SYN was acked. If segment contains CC.ECHO 1255 * option, check it to make sure this segment really 1256 * matches our SYN. If not, just drop it as old 1257 * duplicate, but send an RST if we're still playing 1258 * by the old rules. If no CC.ECHO option, make sure 1259 * we don't get fooled into using T/TCP. 1260 */ 1261 if (to.to_flag & TOF_CCECHO) { 1262 if (tp->cc_send != to.to_ccecho) { 1263 if (taop->tao_ccsent != 0) 1264 goto drop; 1265 else 1266 goto dropwithreset; 1267 } 1268 } else 1269 tp->t_flags &= ~TF_RCVD_CC; 1270 tcpstat.tcps_connects++; 1271 soisconnected(so); 1272 /* Do window scaling on this connection? */ 1273 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1274 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1275 tp->snd_scale = tp->requested_s_scale; 1276 tp->rcv_scale = tp->request_r_scale; 1277 } 1278 /* Segment is acceptable, update cache if undefined. */ 1279 if (taop->tao_ccsent == 0) 1280 taop->tao_ccsent = to.to_ccecho; 1281 1282 tp->rcv_adv += tp->rcv_wnd; 1283 tp->snd_una++; /* SYN is acked */ 1284 /* 1285 * If there's data, delay ACK; if there's also a FIN 1286 * ACKNOW will be turned on later. 1287 */ 1288 if (tcp_delack_enabled && tlen != 0) 1289 callout_reset(tp->tt_delack, tcp_delacktime, 1290 tcp_timer_delack, tp); 1291 else 1292 tp->t_flags |= TF_ACKNOW; 1293 /* 1294 * Received <SYN,ACK> in SYN_SENT[*] state. 1295 * Transitions: 1296 * SYN_SENT --> ESTABLISHED 1297 * SYN_SENT* --> FIN_WAIT_1 1298 */ 1299 tp->t_starttime = ticks; 1300 if (tp->t_flags & TF_NEEDFIN) { 1301 tp->t_state = TCPS_FIN_WAIT_1; 1302 tp->t_flags &= ~TF_NEEDFIN; 1303 thflags &= ~TH_SYN; 1304 } else { 1305 tp->t_state = TCPS_ESTABLISHED; 1306 callout_reset(tp->tt_keep, tcp_keepidle, 1307 tcp_timer_keep, tp); 1308 } 1309 } else { 1310 /* 1311 * Received initial SYN in SYN-SENT[*] state => simul- 1312 * taneous open. If segment contains CC option and there is 1313 * a cached CC, apply TAO test; if it succeeds, connection is 1314 * half-synchronized. Otherwise, do 3-way handshake: 1315 * SYN-SENT -> SYN-RECEIVED 1316 * SYN-SENT* -> SYN-RECEIVED* 1317 * If there was no CC option, clear cached CC value. 1318 */ 1319 tp->t_flags |= TF_ACKNOW; 1320 callout_stop(tp->tt_rexmt); 1321 if (to.to_flag & TOF_CC) { 1322 if (taop->tao_cc != 0 && 1323 CC_GT(to.to_cc, taop->tao_cc)) { 1324 /* 1325 * update cache and make transition: 1326 * SYN-SENT -> ESTABLISHED* 1327 * SYN-SENT* -> FIN-WAIT-1* 1328 */ 1329 taop->tao_cc = to.to_cc; 1330 tp->t_starttime = ticks; 1331 if (tp->t_flags & TF_NEEDFIN) { 1332 tp->t_state = TCPS_FIN_WAIT_1; 1333 tp->t_flags &= ~TF_NEEDFIN; 1334 } else { 1335 tp->t_state = TCPS_ESTABLISHED; 1336 callout_reset(tp->tt_keep, 1337 tcp_keepidle, 1338 tcp_timer_keep, 1339 tp); 1340 } 1341 tp->t_flags |= TF_NEEDSYN; 1342 } else 1343 tp->t_state = TCPS_SYN_RECEIVED; 1344 } else { 1345 /* CC.NEW or no option => invalidate cache */ 1346 taop->tao_cc = 0; 1347 tp->t_state = TCPS_SYN_RECEIVED; 1348 } 1349 } 1350 1351 trimthenstep6: 1352 /* 1353 * Advance th->th_seq to correspond to first data byte. 1354 * If data, trim to stay within window, 1355 * dropping FIN if necessary. 1356 */ 1357 th->th_seq++; 1358 if (tlen > tp->rcv_wnd) { 1359 todrop = tlen - tp->rcv_wnd; 1360 m_adj(m, -todrop); 1361 tlen = tp->rcv_wnd; 1362 thflags &= ~TH_FIN; 1363 tcpstat.tcps_rcvpackafterwin++; 1364 tcpstat.tcps_rcvbyteafterwin += todrop; 1365 } 1366 tp->snd_wl1 = th->th_seq - 1; 1367 tp->rcv_up = th->th_seq; 1368 /* 1369 * Client side of transaction: already sent SYN and data. 1370 * If the remote host used T/TCP to validate the SYN, 1371 * our data will be ACK'd; if so, enter normal data segment 1372 * processing in the middle of step 5, ack processing. 1373 * Otherwise, goto step 6. 1374 */ 1375 if (thflags & TH_ACK) 1376 goto process_ACK; 1377 goto step6; 1378 /* 1379 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1380 * if segment contains a SYN and CC [not CC.NEW] option: 1381 * if state == TIME_WAIT and connection duration > MSL, 1382 * drop packet and send RST; 1383 * 1384 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1385 * ack the FIN (and data) in retransmission queue. 1386 * Complete close and delete TCPCB. Then reprocess 1387 * segment, hoping to find new TCPCB in LISTEN state; 1388 * 1389 * else must be old SYN; drop it. 1390 * else do normal processing. 1391 */ 1392 case TCPS_LAST_ACK: 1393 case TCPS_CLOSING: 1394 case TCPS_TIME_WAIT: 1395 if ((thflags & TH_SYN) && 1396 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 1397 if (tp->t_state == TCPS_TIME_WAIT && 1398 (ticks - tp->t_starttime) > tcp_msl) 1399 goto dropwithreset; 1400 if (CC_GT(to.to_cc, tp->cc_recv)) { 1401 tp = tcp_close(tp); 1402 goto findpcb; 1403 } 1404 else 1405 goto drop; 1406 } 1407 break; /* continue normal processing */ 1408 } 1409 1410 /* 1411 * States other than LISTEN or SYN_SENT. 1412 * First check the RST flag and sequence number since reset segments 1413 * are exempt from the timestamp and connection count tests. This 1414 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1415 * below which allowed reset segments in half the sequence space 1416 * to fall though and be processed (which gives forged reset 1417 * segments with a random sequence number a 50 percent chance of 1418 * killing a connection). 1419 * Then check timestamp, if present. 1420 * Then check the connection count, if present. 1421 * Then check that at least some bytes of segment are within 1422 * receive window. If segment begins before rcv_nxt, 1423 * drop leading data (and SYN); if nothing left, just ack. 1424 * 1425 * 1426 * If the RST bit is set, check the sequence number to see 1427 * if this is a valid reset segment. 1428 * RFC 793 page 37: 1429 * In all states except SYN-SENT, all reset (RST) segments 1430 * are validated by checking their SEQ-fields. A reset is 1431 * valid if its sequence number is in the window. 1432 * Note: this does not take into account delayed ACKs, so 1433 * we should test against last_ack_sent instead of rcv_nxt. 1434 * The sequence number in the reset segment is normally an 1435 * echo of our outgoing acknowlegement numbers, but some hosts 1436 * send a reset with the sequence number at the rightmost edge 1437 * of our receive window, and we have to handle this case. 1438 * If we have multiple segments in flight, the intial reset 1439 * segment sequence numbers will be to the left of last_ack_sent, 1440 * but they will eventually catch up. 1441 * In any case, it never made sense to trim reset segments to 1442 * fit the receive window since RFC 1122 says: 1443 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1444 * 1445 * A TCP SHOULD allow a received RST segment to include data. 1446 * 1447 * DISCUSSION 1448 * It has been suggested that a RST segment could contain 1449 * ASCII text that encoded and explained the cause of the 1450 * RST. No standard has yet been established for such 1451 * data. 1452 * 1453 * If the reset segment passes the sequence number test examine 1454 * the state: 1455 * SYN_RECEIVED STATE: 1456 * If passive open, return to LISTEN state. 1457 * If active open, inform user that connection was refused. 1458 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1459 * Inform user that connection was reset, and close tcb. 1460 * CLOSING, LAST_ACK STATES: 1461 * Close the tcb. 1462 * TIME_WAIT STATE: 1463 * Drop the segment - see Stevens, vol. 2, p. 964 and 1464 * RFC 1337. 1465 */ 1466 if (thflags & TH_RST) { 1467 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1468 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1469 switch (tp->t_state) { 1470 1471 case TCPS_SYN_RECEIVED: 1472 so->so_error = ECONNREFUSED; 1473 goto close; 1474 1475 case TCPS_ESTABLISHED: 1476 case TCPS_FIN_WAIT_1: 1477 case TCPS_FIN_WAIT_2: 1478 case TCPS_CLOSE_WAIT: 1479 so->so_error = ECONNRESET; 1480 close: 1481 tp->t_state = TCPS_CLOSED; 1482 tcpstat.tcps_drops++; 1483 tp = tcp_close(tp); 1484 break; 1485 1486 case TCPS_CLOSING: 1487 case TCPS_LAST_ACK: 1488 tp = tcp_close(tp); 1489 break; 1490 1491 case TCPS_TIME_WAIT: 1492 break; 1493 } 1494 } 1495 goto drop; 1496 } 1497 1498 /* 1499 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1500 * and it's less than ts_recent, drop it. 1501 */ 1502 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent && 1503 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1504 1505 /* Check to see if ts_recent is over 24 days old. */ 1506 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1507 /* 1508 * Invalidate ts_recent. If this segment updates 1509 * ts_recent, the age will be reset later and ts_recent 1510 * will get a valid value. If it does not, setting 1511 * ts_recent to zero will at least satisfy the 1512 * requirement that zero be placed in the timestamp 1513 * echo reply when ts_recent isn't valid. The 1514 * age isn't reset until we get a valid ts_recent 1515 * because we don't want out-of-order segments to be 1516 * dropped when ts_recent is old. 1517 */ 1518 tp->ts_recent = 0; 1519 } else { 1520 tcpstat.tcps_rcvduppack++; 1521 tcpstat.tcps_rcvdupbyte += tlen; 1522 tcpstat.tcps_pawsdrop++; 1523 goto dropafterack; 1524 } 1525 } 1526 1527 /* 1528 * T/TCP mechanism 1529 * If T/TCP was negotiated and the segment doesn't have CC, 1530 * or if its CC is wrong then drop the segment. 1531 * RST segments do not have to comply with this. 1532 */ 1533 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1534 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1535 goto dropafterack; 1536 1537 /* 1538 * In the SYN-RECEIVED state, validate that the packet belongs to 1539 * this connection before trimming the data to fit the receive 1540 * window. Check the sequence number versus IRS since we know 1541 * the sequence numbers haven't wrapped. This is a partial fix 1542 * for the "LAND" DoS attack. 1543 */ 1544 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) 1545 goto maybedropwithreset; 1546 1547 todrop = tp->rcv_nxt - th->th_seq; 1548 if (todrop > 0) { 1549 if (thflags & TH_SYN) { 1550 thflags &= ~TH_SYN; 1551 th->th_seq++; 1552 if (th->th_urp > 1) 1553 th->th_urp--; 1554 else 1555 thflags &= ~TH_URG; 1556 todrop--; 1557 } 1558 /* 1559 * Following if statement from Stevens, vol. 2, p. 960. 1560 */ 1561 if (todrop > tlen 1562 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1563 /* 1564 * Any valid FIN must be to the left of the window. 1565 * At this point the FIN must be a duplicate or out 1566 * of sequence; drop it. 1567 */ 1568 thflags &= ~TH_FIN; 1569 1570 /* 1571 * Send an ACK to resynchronize and drop any data. 1572 * But keep on processing for RST or ACK. 1573 */ 1574 tp->t_flags |= TF_ACKNOW; 1575 todrop = tlen; 1576 tcpstat.tcps_rcvduppack++; 1577 tcpstat.tcps_rcvdupbyte += todrop; 1578 } else { 1579 tcpstat.tcps_rcvpartduppack++; 1580 tcpstat.tcps_rcvpartdupbyte += todrop; 1581 } 1582 drop_hdrlen += todrop; /* drop from the top afterwards */ 1583 th->th_seq += todrop; 1584 tlen -= todrop; 1585 if (th->th_urp > todrop) 1586 th->th_urp -= todrop; 1587 else { 1588 thflags &= ~TH_URG; 1589 th->th_urp = 0; 1590 } 1591 } 1592 1593 /* 1594 * If new data are received on a connection after the 1595 * user processes are gone, then RST the other end. 1596 */ 1597 if ((so->so_state & SS_NOFDREF) && 1598 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1599 tp = tcp_close(tp); 1600 tcpstat.tcps_rcvafterclose++; 1601 goto dropwithreset; 1602 } 1603 1604 /* 1605 * If segment ends after window, drop trailing data 1606 * (and PUSH and FIN); if nothing left, just ACK. 1607 */ 1608 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1609 if (todrop > 0) { 1610 tcpstat.tcps_rcvpackafterwin++; 1611 if (todrop >= tlen) { 1612 tcpstat.tcps_rcvbyteafterwin += tlen; 1613 /* 1614 * If a new connection request is received 1615 * while in TIME_WAIT, drop the old connection 1616 * and start over if the sequence numbers 1617 * are above the previous ones. 1618 */ 1619 if (thflags & TH_SYN && 1620 tp->t_state == TCPS_TIME_WAIT && 1621 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1622 iss = tp->snd_nxt + TCP_ISSINCR; 1623 tp = tcp_close(tp); 1624 goto findpcb; 1625 } 1626 /* 1627 * If window is closed can only take segments at 1628 * window edge, and have to drop data and PUSH from 1629 * incoming segments. Continue processing, but 1630 * remember to ack. Otherwise, drop segment 1631 * and ack. 1632 */ 1633 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1634 tp->t_flags |= TF_ACKNOW; 1635 tcpstat.tcps_rcvwinprobe++; 1636 } else 1637 goto dropafterack; 1638 } else 1639 tcpstat.tcps_rcvbyteafterwin += todrop; 1640 m_adj(m, -todrop); 1641 tlen -= todrop; 1642 thflags &= ~(TH_PUSH|TH_FIN); 1643 } 1644 1645 /* 1646 * If last ACK falls within this segment's sequence numbers, 1647 * record its timestamp. 1648 * NOTE that the test is modified according to the latest 1649 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1650 */ 1651 if ((to.to_flag & TOF_TS) != 0 && 1652 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1653 tp->ts_recent_age = ticks; 1654 tp->ts_recent = to.to_tsval; 1655 } 1656 1657 /* 1658 * If a SYN is in the window, then this is an 1659 * error and we send an RST and drop the connection. 1660 */ 1661 if (thflags & TH_SYN) { 1662 tp = tcp_drop(tp, ECONNRESET); 1663 goto dropwithreset; 1664 } 1665 1666 /* 1667 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1668 * flag is on (half-synchronized state), then queue data for 1669 * later processing; else drop segment and return. 1670 */ 1671 if ((thflags & TH_ACK) == 0) { 1672 if (tp->t_state == TCPS_SYN_RECEIVED || 1673 (tp->t_flags & TF_NEEDSYN)) 1674 goto step6; 1675 else 1676 goto drop; 1677 } 1678 1679 /* 1680 * Ack processing. 1681 */ 1682 switch (tp->t_state) { 1683 1684 /* 1685 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1686 * ESTABLISHED state and continue processing. 1687 * The ACK was checked above. 1688 */ 1689 case TCPS_SYN_RECEIVED: 1690 1691 tcpstat.tcps_connects++; 1692 soisconnected(so); 1693 /* Do window scaling? */ 1694 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1695 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1696 tp->snd_scale = tp->requested_s_scale; 1697 tp->rcv_scale = tp->request_r_scale; 1698 } 1699 /* 1700 * Upon successful completion of 3-way handshake, 1701 * update cache.CC if it was undefined, pass any queued 1702 * data to the user, and advance state appropriately. 1703 */ 1704 if ((taop = tcp_gettaocache(inp)) != NULL && 1705 taop->tao_cc == 0) 1706 taop->tao_cc = tp->cc_recv; 1707 1708 /* 1709 * Make transitions: 1710 * SYN-RECEIVED -> ESTABLISHED 1711 * SYN-RECEIVED* -> FIN-WAIT-1 1712 */ 1713 tp->t_starttime = ticks; 1714 if (tp->t_flags & TF_NEEDFIN) { 1715 tp->t_state = TCPS_FIN_WAIT_1; 1716 tp->t_flags &= ~TF_NEEDFIN; 1717 } else { 1718 tp->t_state = TCPS_ESTABLISHED; 1719 callout_reset(tp->tt_keep, tcp_keepidle, 1720 tcp_timer_keep, tp); 1721 } 1722 /* 1723 * If segment contains data or ACK, will call tcp_reass() 1724 * later; if not, do so now to pass queued data to user. 1725 */ 1726 if (tlen == 0 && (thflags & TH_FIN) == 0) 1727 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1728 (struct mbuf *)0); 1729 tp->snd_wl1 = th->th_seq - 1; 1730 /* fall into ... */ 1731 1732 /* 1733 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1734 * ACKs. If the ack is in the range 1735 * tp->snd_una < th->th_ack <= tp->snd_max 1736 * then advance tp->snd_una to th->th_ack and drop 1737 * data from the retransmission queue. If this ACK reflects 1738 * more up to date window information we update our window information. 1739 */ 1740 case TCPS_ESTABLISHED: 1741 case TCPS_FIN_WAIT_1: 1742 case TCPS_FIN_WAIT_2: 1743 case TCPS_CLOSE_WAIT: 1744 case TCPS_CLOSING: 1745 case TCPS_LAST_ACK: 1746 case TCPS_TIME_WAIT: 1747 1748 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1749 if (tlen == 0 && tiwin == tp->snd_wnd) { 1750 tcpstat.tcps_rcvdupack++; 1751 /* 1752 * If we have outstanding data (other than 1753 * a window probe), this is a completely 1754 * duplicate ack (ie, window info didn't 1755 * change), the ack is the biggest we've 1756 * seen and we've seen exactly our rexmt 1757 * threshhold of them, assume a packet 1758 * has been dropped and retransmit it. 1759 * Kludge snd_nxt & the congestion 1760 * window so we send only this one 1761 * packet. 1762 * 1763 * We know we're losing at the current 1764 * window size so do congestion avoidance 1765 * (set ssthresh to half the current window 1766 * and pull our congestion window back to 1767 * the new ssthresh). 1768 * 1769 * Dup acks mean that packets have left the 1770 * network (they're now cached at the receiver) 1771 * so bump cwnd by the amount in the receiver 1772 * to keep a constant cwnd packets in the 1773 * network. 1774 */ 1775 if (!callout_active(tp->tt_rexmt) || 1776 th->th_ack != tp->snd_una) 1777 tp->t_dupacks = 0; 1778 else if (++tp->t_dupacks == tcprexmtthresh) { 1779 tcp_seq onxt = tp->snd_nxt; 1780 u_int win = 1781 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1782 tp->t_maxseg; 1783 1784 if (win < 2) 1785 win = 2; 1786 tp->snd_ssthresh = win * tp->t_maxseg; 1787 callout_stop(tp->tt_rexmt); 1788 tp->t_rtttime = 0; 1789 tp->snd_nxt = th->th_ack; 1790 tp->snd_cwnd = tp->t_maxseg; 1791 (void) tcp_output(tp); 1792 tp->snd_cwnd = tp->snd_ssthresh + 1793 tp->t_maxseg * tp->t_dupacks; 1794 if (SEQ_GT(onxt, tp->snd_nxt)) 1795 tp->snd_nxt = onxt; 1796 goto drop; 1797 } else if (tp->t_dupacks > tcprexmtthresh) { 1798 tp->snd_cwnd += tp->t_maxseg; 1799 (void) tcp_output(tp); 1800 goto drop; 1801 } 1802 } else 1803 tp->t_dupacks = 0; 1804 break; 1805 } 1806 /* 1807 * If the congestion window was inflated to account 1808 * for the other side's cached packets, retract it. 1809 */ 1810 if (tp->t_dupacks >= tcprexmtthresh && 1811 tp->snd_cwnd > tp->snd_ssthresh) 1812 tp->snd_cwnd = tp->snd_ssthresh; 1813 tp->t_dupacks = 0; 1814 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1815 tcpstat.tcps_rcvacktoomuch++; 1816 goto dropafterack; 1817 } 1818 /* 1819 * If we reach this point, ACK is not a duplicate, 1820 * i.e., it ACKs something we sent. 1821 */ 1822 if (tp->t_flags & TF_NEEDSYN) { 1823 /* 1824 * T/TCP: Connection was half-synchronized, and our 1825 * SYN has been ACK'd (so connection is now fully 1826 * synchronized). Go to non-starred state, 1827 * increment snd_una for ACK of SYN, and check if 1828 * we can do window scaling. 1829 */ 1830 tp->t_flags &= ~TF_NEEDSYN; 1831 tp->snd_una++; 1832 /* Do window scaling? */ 1833 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1834 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1835 tp->snd_scale = tp->requested_s_scale; 1836 tp->rcv_scale = tp->request_r_scale; 1837 } 1838 } 1839 1840 process_ACK: 1841 acked = th->th_ack - tp->snd_una; 1842 tcpstat.tcps_rcvackpack++; 1843 tcpstat.tcps_rcvackbyte += acked; 1844 1845 /* 1846 * If we just performed our first retransmit, and the ACK 1847 * arrives within our recovery window, then it was a mistake 1848 * to do the retransmit in the first place. Recover our 1849 * original cwnd and ssthresh, and proceed to transmit where 1850 * we left off. 1851 */ 1852 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1853 tp->snd_cwnd = tp->snd_cwnd_prev; 1854 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1855 tp->snd_nxt = tp->snd_max; 1856 tp->t_badrxtwin = 0; /* XXX probably not required */ 1857 } 1858 1859 /* 1860 * If we have a timestamp reply, update smoothed 1861 * round trip time. If no timestamp is present but 1862 * transmit timer is running and timed sequence 1863 * number was acked, update smoothed round trip time. 1864 * Since we now have an rtt measurement, cancel the 1865 * timer backoff (cf., Phil Karn's retransmit alg.). 1866 * Recompute the initial retransmit timer. 1867 */ 1868 if (to.to_flag & TOF_TS) 1869 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1870 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1871 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1872 1873 /* 1874 * If all outstanding data is acked, stop retransmit 1875 * timer and remember to restart (more output or persist). 1876 * If there is more data to be acked, restart retransmit 1877 * timer, using current (possibly backed-off) value. 1878 */ 1879 if (th->th_ack == tp->snd_max) { 1880 callout_stop(tp->tt_rexmt); 1881 needoutput = 1; 1882 } else if (!callout_active(tp->tt_persist)) 1883 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1884 tcp_timer_rexmt, tp); 1885 1886 /* 1887 * If no data (only SYN) was ACK'd, 1888 * skip rest of ACK processing. 1889 */ 1890 if (acked == 0) 1891 goto step6; 1892 1893 /* 1894 * When new data is acked, open the congestion window. 1895 * If the window gives us less than ssthresh packets 1896 * in flight, open exponentially (maxseg per packet). 1897 * Otherwise open linearly: maxseg per window 1898 * (maxseg^2 / cwnd per packet). 1899 */ 1900 { 1901 register u_int cw = tp->snd_cwnd; 1902 register u_int incr = tp->t_maxseg; 1903 1904 if (cw > tp->snd_ssthresh) 1905 incr = incr * incr / cw; 1906 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN << tp->snd_scale); 1907 } 1908 if (acked > so->so_snd.sb_cc) { 1909 tp->snd_wnd -= so->so_snd.sb_cc; 1910 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1911 ourfinisacked = 1; 1912 } else { 1913 sbdrop(&so->so_snd, acked); 1914 tp->snd_wnd -= acked; 1915 ourfinisacked = 0; 1916 } 1917 sowwakeup(so); 1918 tp->snd_una = th->th_ack; 1919 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1920 tp->snd_nxt = tp->snd_una; 1921 1922 switch (tp->t_state) { 1923 1924 /* 1925 * In FIN_WAIT_1 STATE in addition to the processing 1926 * for the ESTABLISHED state if our FIN is now acknowledged 1927 * then enter FIN_WAIT_2. 1928 */ 1929 case TCPS_FIN_WAIT_1: 1930 if (ourfinisacked) { 1931 /* 1932 * If we can't receive any more 1933 * data, then closing user can proceed. 1934 * Starting the timer is contrary to the 1935 * specification, but if we don't get a FIN 1936 * we'll hang forever. 1937 */ 1938 if (so->so_state & SS_CANTRCVMORE) { 1939 soisdisconnected(so); 1940 callout_reset(tp->tt_2msl, tcp_maxidle, 1941 tcp_timer_2msl, tp); 1942 } 1943 tp->t_state = TCPS_FIN_WAIT_2; 1944 } 1945 break; 1946 1947 /* 1948 * In CLOSING STATE in addition to the processing for 1949 * the ESTABLISHED state if the ACK acknowledges our FIN 1950 * then enter the TIME-WAIT state, otherwise ignore 1951 * the segment. 1952 */ 1953 case TCPS_CLOSING: 1954 if (ourfinisacked) { 1955 tp->t_state = TCPS_TIME_WAIT; 1956 tcp_canceltimers(tp); 1957 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 1958 if (tp->cc_recv != 0 && 1959 (ticks - tp->t_starttime) < tcp_msl) 1960 callout_reset(tp->tt_2msl, 1961 tp->t_rxtcur * 1962 TCPTV_TWTRUNC, 1963 tcp_timer_2msl, tp); 1964 else 1965 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1966 tcp_timer_2msl, tp); 1967 soisdisconnected(so); 1968 } 1969 break; 1970 1971 /* 1972 * In LAST_ACK, we may still be waiting for data to drain 1973 * and/or to be acked, as well as for the ack of our FIN. 1974 * If our FIN is now acknowledged, delete the TCB, 1975 * enter the closed state and return. 1976 */ 1977 case TCPS_LAST_ACK: 1978 if (ourfinisacked) { 1979 tp = tcp_close(tp); 1980 goto drop; 1981 } 1982 break; 1983 1984 /* 1985 * In TIME_WAIT state the only thing that should arrive 1986 * is a retransmission of the remote FIN. Acknowledge 1987 * it and restart the finack timer. 1988 */ 1989 case TCPS_TIME_WAIT: 1990 callout_reset(tp->tt_2msl, 2 * tcp_msl, 1991 tcp_timer_2msl, tp); 1992 goto dropafterack; 1993 } 1994 } 1995 1996 step6: 1997 /* 1998 * Update window information. 1999 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2000 */ 2001 if ((thflags & TH_ACK) && 2002 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2003 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2004 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2005 /* keep track of pure window updates */ 2006 if (tlen == 0 && 2007 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2008 tcpstat.tcps_rcvwinupd++; 2009 tp->snd_wnd = tiwin; 2010 tp->snd_wl1 = th->th_seq; 2011 tp->snd_wl2 = th->th_ack; 2012 if (tp->snd_wnd > tp->max_sndwnd) 2013 tp->max_sndwnd = tp->snd_wnd; 2014 needoutput = 1; 2015 } 2016 2017 /* 2018 * Process segments with URG. 2019 */ 2020 if ((thflags & TH_URG) && th->th_urp && 2021 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2022 /* 2023 * This is a kludge, but if we receive and accept 2024 * random urgent pointers, we'll crash in 2025 * soreceive. It's hard to imagine someone 2026 * actually wanting to send this much urgent data. 2027 */ 2028 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2029 th->th_urp = 0; /* XXX */ 2030 thflags &= ~TH_URG; /* XXX */ 2031 goto dodata; /* XXX */ 2032 } 2033 /* 2034 * If this segment advances the known urgent pointer, 2035 * then mark the data stream. This should not happen 2036 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2037 * a FIN has been received from the remote side. 2038 * In these states we ignore the URG. 2039 * 2040 * According to RFC961 (Assigned Protocols), 2041 * the urgent pointer points to the last octet 2042 * of urgent data. We continue, however, 2043 * to consider it to indicate the first octet 2044 * of data past the urgent section as the original 2045 * spec states (in one of two places). 2046 */ 2047 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2048 tp->rcv_up = th->th_seq + th->th_urp; 2049 so->so_oobmark = so->so_rcv.sb_cc + 2050 (tp->rcv_up - tp->rcv_nxt) - 1; 2051 if (so->so_oobmark == 0) 2052 so->so_state |= SS_RCVATMARK; 2053 sohasoutofband(so); 2054 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2055 } 2056 /* 2057 * Remove out of band data so doesn't get presented to user. 2058 * This can happen independent of advancing the URG pointer, 2059 * but if two URG's are pending at once, some out-of-band 2060 * data may creep in... ick. 2061 */ 2062 if (th->th_urp <= (u_long)tlen 2063 #ifdef SO_OOBINLINE 2064 && (so->so_options & SO_OOBINLINE) == 0 2065 #endif 2066 ) 2067 tcp_pulloutofband(so, th, m, 2068 drop_hdrlen); /* hdr drop is delayed */ 2069 } else 2070 /* 2071 * If no out of band data is expected, 2072 * pull receive urgent pointer along 2073 * with the receive window. 2074 */ 2075 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2076 tp->rcv_up = tp->rcv_nxt; 2077 dodata: /* XXX */ 2078 2079 /* 2080 * Process the segment text, merging it into the TCP sequencing queue, 2081 * and arranging for acknowledgment of receipt if necessary. 2082 * This process logically involves adjusting tp->rcv_wnd as data 2083 * is presented to the user (this happens in tcp_usrreq.c, 2084 * case PRU_RCVD). If a FIN has already been received on this 2085 * connection then we just ignore the text. 2086 */ 2087 if ((tlen || (thflags&TH_FIN)) && 2088 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2089 m_adj(m, drop_hdrlen); /* delayed header drop */ 2090 TCP_REASS(tp, th, &tlen, m, so, thflags); 2091 /* 2092 * Note the amount of data that peer has sent into 2093 * our window, in order to estimate the sender's 2094 * buffer size. 2095 */ 2096 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2097 } else { 2098 m_freem(m); 2099 thflags &= ~TH_FIN; 2100 } 2101 2102 /* 2103 * If FIN is received ACK the FIN and let the user know 2104 * that the connection is closing. 2105 */ 2106 if (thflags & TH_FIN) { 2107 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2108 socantrcvmore(so); 2109 /* 2110 * If connection is half-synchronized 2111 * (ie NEEDSYN flag on) then delay ACK, 2112 * so it may be piggybacked when SYN is sent. 2113 * Otherwise, since we received a FIN then no 2114 * more input can be expected, send ACK now. 2115 */ 2116 if (tcp_delack_enabled && (tp->t_flags & TF_NEEDSYN)) 2117 callout_reset(tp->tt_delack, tcp_delacktime, 2118 tcp_timer_delack, tp); 2119 else 2120 tp->t_flags |= TF_ACKNOW; 2121 tp->rcv_nxt++; 2122 } 2123 switch (tp->t_state) { 2124 2125 /* 2126 * In SYN_RECEIVED and ESTABLISHED STATES 2127 * enter the CLOSE_WAIT state. 2128 */ 2129 case TCPS_SYN_RECEIVED: 2130 tp->t_starttime = ticks; 2131 /*FALLTHROUGH*/ 2132 case TCPS_ESTABLISHED: 2133 tp->t_state = TCPS_CLOSE_WAIT; 2134 break; 2135 2136 /* 2137 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2138 * enter the CLOSING state. 2139 */ 2140 case TCPS_FIN_WAIT_1: 2141 tp->t_state = TCPS_CLOSING; 2142 break; 2143 2144 /* 2145 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2146 * starting the time-wait timer, turning off the other 2147 * standard timers. 2148 */ 2149 case TCPS_FIN_WAIT_2: 2150 tp->t_state = TCPS_TIME_WAIT; 2151 tcp_canceltimers(tp); 2152 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2153 if (tp->cc_recv != 0 && 2154 (ticks - tp->t_starttime) < tcp_msl) { 2155 callout_reset(tp->tt_2msl, 2156 tp->t_rxtcur * TCPTV_TWTRUNC, 2157 tcp_timer_2msl, tp); 2158 /* For transaction client, force ACK now. */ 2159 tp->t_flags |= TF_ACKNOW; 2160 } 2161 else 2162 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2163 tcp_timer_2msl, tp); 2164 soisdisconnected(so); 2165 break; 2166 2167 /* 2168 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2169 */ 2170 case TCPS_TIME_WAIT: 2171 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2172 tcp_timer_2msl, tp); 2173 break; 2174 } 2175 } 2176 #ifdef TCPDEBUG 2177 if (so->so_options & SO_DEBUG) 2178 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2179 &tcp_savetcp, 0); 2180 #endif 2181 2182 /* 2183 * Return any desired output. 2184 */ 2185 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2186 (void) tcp_output(tp); 2187 return; 2188 2189 dropafterack: 2190 /* 2191 * Generate an ACK dropping incoming segment if it occupies 2192 * sequence space, where the ACK reflects our state. 2193 * 2194 * We can now skip the test for the RST flag since all 2195 * paths to this code happen after packets containing 2196 * RST have been dropped. 2197 * 2198 * In the SYN-RECEIVED state, don't send an ACK unless the 2199 * segment we received passes the SYN-RECEIVED ACK test. 2200 * If it fails send a RST. This breaks the loop in the 2201 * "LAND" DoS attack, and also prevents an ACK storm 2202 * between two listening ports that have been sent forged 2203 * SYN segments, each with the source address of the other. 2204 */ 2205 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2206 (SEQ_GT(tp->snd_una, th->th_ack) || 2207 SEQ_GT(th->th_ack, tp->snd_max)) ) 2208 goto maybedropwithreset; 2209 #ifdef TCPDEBUG 2210 if (so->so_options & SO_DEBUG) 2211 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2212 &tcp_savetcp, 0); 2213 #endif 2214 m_freem(m); 2215 tp->t_flags |= TF_ACKNOW; 2216 (void) tcp_output(tp); 2217 return; 2218 2219 2220 /* 2221 * Conditionally drop with reset or just drop depending on whether 2222 * we think we are under attack or not. 2223 */ 2224 maybedropwithreset: 2225 #ifdef ICMP_BANDLIM 2226 if (badport_bandlim(1) < 0) 2227 goto drop; 2228 #endif 2229 /* fall through */ 2230 dropwithreset: 2231 #ifdef TCP_RESTRICT_RST 2232 if (restrict_rst) 2233 goto drop; 2234 #endif 2235 /* 2236 * Generate a RST, dropping incoming segment. 2237 * Make ACK acceptable to originator of segment. 2238 * Don't bother to respond if destination was broadcast/multicast. 2239 */ 2240 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2241 goto drop; 2242 #ifdef INET6 2243 if (isipv6) { 2244 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2245 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2246 goto drop; 2247 } else 2248 #endif /* INET6 */ 2249 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2250 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2251 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 2252 goto drop; 2253 /* IPv6 anycast check is done at tcp6_input() */ 2254 #ifdef TCPDEBUG 2255 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2256 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2257 &tcp_savetcp, 0); 2258 #endif 2259 if (thflags & TH_ACK) 2260 /* mtod() below is safe as long as hdr dropping is delayed */ 2261 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2262 TH_RST); 2263 else { 2264 if (thflags & TH_SYN) 2265 tlen++; 2266 /* mtod() below is safe as long as hdr dropping is delayed */ 2267 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2268 (tcp_seq)0, TH_RST|TH_ACK); 2269 } 2270 /* destroy temporarily created socket */ 2271 if (dropsocket) 2272 (void) soabort(so); 2273 return; 2274 2275 drop: 2276 /* 2277 * Drop space held by incoming segment and return. 2278 */ 2279 #ifdef TCPDEBUG 2280 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2281 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2282 &tcp_savetcp, 0); 2283 #endif 2284 m_freem(m); 2285 /* destroy temporarily created socket */ 2286 if (dropsocket) 2287 (void) soabort(so); 2288 return; 2289 } 2290 2291 static void 2292 tcp_dooptions(tp, cp, cnt, th, to) 2293 struct tcpcb *tp; 2294 u_char *cp; 2295 int cnt; 2296 struct tcphdr *th; 2297 struct tcpopt *to; 2298 { 2299 u_short mss = 0; 2300 int opt, optlen; 2301 2302 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2303 opt = cp[0]; 2304 if (opt == TCPOPT_EOL) 2305 break; 2306 if (opt == TCPOPT_NOP) 2307 optlen = 1; 2308 else { 2309 optlen = cp[1]; 2310 if (optlen <= 0) 2311 break; 2312 } 2313 switch (opt) { 2314 2315 default: 2316 continue; 2317 2318 case TCPOPT_MAXSEG: 2319 if (optlen != TCPOLEN_MAXSEG) 2320 continue; 2321 if (!(th->th_flags & TH_SYN)) 2322 continue; 2323 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 2324 NTOHS(mss); 2325 break; 2326 2327 case TCPOPT_WINDOW: 2328 if (optlen != TCPOLEN_WINDOW) 2329 continue; 2330 if (!(th->th_flags & TH_SYN)) 2331 continue; 2332 tp->t_flags |= TF_RCVD_SCALE; 2333 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2334 break; 2335 2336 case TCPOPT_TIMESTAMP: 2337 if (optlen != TCPOLEN_TIMESTAMP) 2338 continue; 2339 to->to_flag |= TOF_TS; 2340 bcopy((char *)cp + 2, 2341 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2342 NTOHL(to->to_tsval); 2343 bcopy((char *)cp + 6, 2344 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2345 NTOHL(to->to_tsecr); 2346 2347 /* 2348 * A timestamp received in a SYN makes 2349 * it ok to send timestamp requests and replies. 2350 */ 2351 if (th->th_flags & TH_SYN) { 2352 tp->t_flags |= TF_RCVD_TSTMP; 2353 tp->ts_recent = to->to_tsval; 2354 tp->ts_recent_age = ticks; 2355 } 2356 break; 2357 case TCPOPT_CC: 2358 if (optlen != TCPOLEN_CC) 2359 continue; 2360 to->to_flag |= TOF_CC; 2361 bcopy((char *)cp + 2, 2362 (char *)&to->to_cc, sizeof(to->to_cc)); 2363 NTOHL(to->to_cc); 2364 /* 2365 * A CC or CC.new option received in a SYN makes 2366 * it ok to send CC in subsequent segments. 2367 */ 2368 if (th->th_flags & TH_SYN) 2369 tp->t_flags |= TF_RCVD_CC; 2370 break; 2371 case TCPOPT_CCNEW: 2372 if (optlen != TCPOLEN_CC) 2373 continue; 2374 if (!(th->th_flags & TH_SYN)) 2375 continue; 2376 to->to_flag |= TOF_CCNEW; 2377 bcopy((char *)cp + 2, 2378 (char *)&to->to_cc, sizeof(to->to_cc)); 2379 NTOHL(to->to_cc); 2380 /* 2381 * A CC or CC.new option received in a SYN makes 2382 * it ok to send CC in subsequent segments. 2383 */ 2384 tp->t_flags |= TF_RCVD_CC; 2385 break; 2386 case TCPOPT_CCECHO: 2387 if (optlen != TCPOLEN_CC) 2388 continue; 2389 if (!(th->th_flags & TH_SYN)) 2390 continue; 2391 to->to_flag |= TOF_CCECHO; 2392 bcopy((char *)cp + 2, 2393 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2394 NTOHL(to->to_ccecho); 2395 break; 2396 } 2397 } 2398 if (th->th_flags & TH_SYN) 2399 tcp_mss(tp, mss); /* sets t_maxseg */ 2400 } 2401 2402 /* 2403 * Pull out of band byte out of a segment so 2404 * it doesn't appear in the user's data queue. 2405 * It is still reflected in the segment length for 2406 * sequencing purposes. 2407 */ 2408 static void 2409 tcp_pulloutofband(so, th, m, off) 2410 struct socket *so; 2411 struct tcphdr *th; 2412 register struct mbuf *m; 2413 int off; /* delayed to be droped hdrlen */ 2414 { 2415 int cnt = off + th->th_urp - 1; 2416 2417 while (cnt >= 0) { 2418 if (m->m_len > cnt) { 2419 char *cp = mtod(m, caddr_t) + cnt; 2420 struct tcpcb *tp = sototcpcb(so); 2421 2422 tp->t_iobc = *cp; 2423 tp->t_oobflags |= TCPOOB_HAVEDATA; 2424 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2425 m->m_len--; 2426 if (m->m_flags & M_PKTHDR) 2427 m->m_pkthdr.len--; 2428 return; 2429 } 2430 cnt -= m->m_len; 2431 m = m->m_next; 2432 if (m == 0) 2433 break; 2434 } 2435 panic("tcp_pulloutofband"); 2436 } 2437 2438 /* 2439 * Collect new round-trip time estimate 2440 * and update averages and current timeout. 2441 */ 2442 static void 2443 tcp_xmit_timer(tp, rtt) 2444 register struct tcpcb *tp; 2445 int rtt; 2446 { 2447 register int delta; 2448 2449 tcpstat.tcps_rttupdated++; 2450 tp->t_rttupdated++; 2451 if (tp->t_srtt != 0) { 2452 /* 2453 * srtt is stored as fixed point with 5 bits after the 2454 * binary point (i.e., scaled by 8). The following magic 2455 * is equivalent to the smoothing algorithm in rfc793 with 2456 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2457 * point). Adjust rtt to origin 0. 2458 */ 2459 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2460 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2461 2462 if ((tp->t_srtt += delta) <= 0) 2463 tp->t_srtt = 1; 2464 2465 /* 2466 * We accumulate a smoothed rtt variance (actually, a 2467 * smoothed mean difference), then set the retransmit 2468 * timer to smoothed rtt + 4 times the smoothed variance. 2469 * rttvar is stored as fixed point with 4 bits after the 2470 * binary point (scaled by 16). The following is 2471 * equivalent to rfc793 smoothing with an alpha of .75 2472 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2473 * rfc793's wired-in beta. 2474 */ 2475 if (delta < 0) 2476 delta = -delta; 2477 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2478 if ((tp->t_rttvar += delta) <= 0) 2479 tp->t_rttvar = 1; 2480 } else { 2481 /* 2482 * No rtt measurement yet - use the unsmoothed rtt. 2483 * Set the variance to half the rtt (so our first 2484 * retransmit happens at 3*rtt). 2485 */ 2486 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2487 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2488 } 2489 tp->t_rtttime = 0; 2490 tp->t_rxtshift = 0; 2491 2492 /* 2493 * the retransmit should happen at rtt + 4 * rttvar. 2494 * Because of the way we do the smoothing, srtt and rttvar 2495 * will each average +1/2 tick of bias. When we compute 2496 * the retransmit timer, we want 1/2 tick of rounding and 2497 * 1 extra tick because of +-1/2 tick uncertainty in the 2498 * firing of the timer. The bias will give us exactly the 2499 * 1.5 tick we need. But, because the bias is 2500 * statistical, we have to test that we don't drop below 2501 * the minimum feasible timer (which is 2 ticks). 2502 */ 2503 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2504 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2505 2506 /* 2507 * We received an ack for a packet that wasn't retransmitted; 2508 * it is probably safe to discard any error indications we've 2509 * received recently. This isn't quite right, but close enough 2510 * for now (a route might have failed after we sent a segment, 2511 * and the return path might not be symmetrical). 2512 */ 2513 tp->t_softerror = 0; 2514 } 2515 2516 /* 2517 * Determine a reasonable value for maxseg size. 2518 * If the route is known, check route for mtu. 2519 * If none, use an mss that can be handled on the outgoing 2520 * interface without forcing IP to fragment; if bigger than 2521 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2522 * to utilize large mbufs. If no route is found, route has no mtu, 2523 * or the destination isn't local, use a default, hopefully conservative 2524 * size (usually 512 or the default IP max size, but no more than the mtu 2525 * of the interface), as we can't discover anything about intervening 2526 * gateways or networks. We also initialize the congestion/slow start 2527 * window to be a single segment if the destination isn't local. 2528 * While looking at the routing entry, we also initialize other path-dependent 2529 * parameters from pre-set or cached values in the routing entry. 2530 * 2531 * Also take into account the space needed for options that we 2532 * send regularly. Make maxseg shorter by that amount to assure 2533 * that we can send maxseg amount of data even when the options 2534 * are present. Store the upper limit of the length of options plus 2535 * data in maxopd. 2536 * 2537 * NOTE that this routine is only called when we process an incoming 2538 * segment, for outgoing segments only tcp_mssopt is called. 2539 * 2540 * In case of T/TCP, we call this routine during implicit connection 2541 * setup as well (offer = -1), to initialize maxseg from the cached 2542 * MSS of our peer. 2543 */ 2544 void 2545 tcp_mss(tp, offer) 2546 struct tcpcb *tp; 2547 int offer; 2548 { 2549 register struct rtentry *rt; 2550 struct ifnet *ifp; 2551 register int rtt, mss; 2552 u_long bufsize; 2553 struct inpcb *inp; 2554 struct socket *so; 2555 struct rmxp_tao *taop; 2556 int origoffer = offer; 2557 #ifdef INET6 2558 int isipv6; 2559 int min_protoh; 2560 #endif 2561 2562 inp = tp->t_inpcb; 2563 #ifdef INET6 2564 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2565 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2566 : sizeof (struct tcpiphdr); 2567 #else 2568 #define min_protoh (sizeof (struct tcpiphdr)) 2569 #endif 2570 #ifdef INET6 2571 if (isipv6) 2572 rt = tcp_rtlookup6(inp); 2573 else 2574 #endif 2575 rt = tcp_rtlookup(inp); 2576 if (rt == NULL) { 2577 tp->t_maxopd = tp->t_maxseg = 2578 #ifdef INET6 2579 isipv6 ? tcp_v6mssdflt : 2580 #endif /* INET6 */ 2581 tcp_mssdflt; 2582 return; 2583 } 2584 ifp = rt->rt_ifp; 2585 so = inp->inp_socket; 2586 2587 taop = rmx_taop(rt->rt_rmx); 2588 /* 2589 * Offer == -1 means that we didn't receive SYN yet, 2590 * use cached value in that case; 2591 */ 2592 if (offer == -1) 2593 offer = taop->tao_mssopt; 2594 /* 2595 * Offer == 0 means that there was no MSS on the SYN segment, 2596 * in this case we use tcp_mssdflt. 2597 */ 2598 if (offer == 0) 2599 offer = 2600 #ifdef INET6 2601 isipv6 ? tcp_v6mssdflt : 2602 #endif /* INET6 */ 2603 tcp_mssdflt; 2604 else 2605 /* 2606 * Sanity check: make sure that maxopd will be large 2607 * enough to allow some data on segments even is the 2608 * all the option space is used (40bytes). Otherwise 2609 * funny things may happen in tcp_output. 2610 */ 2611 offer = max(offer, 64); 2612 taop->tao_mssopt = offer; 2613 2614 /* 2615 * While we're here, check if there's an initial rtt 2616 * or rttvar. Convert from the route-table units 2617 * to scaled multiples of the slow timeout timer. 2618 */ 2619 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2620 /* 2621 * XXX the lock bit for RTT indicates that the value 2622 * is also a minimum value; this is subject to time. 2623 */ 2624 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2625 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2626 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2627 tcpstat.tcps_usedrtt++; 2628 if (rt->rt_rmx.rmx_rttvar) { 2629 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2630 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2631 tcpstat.tcps_usedrttvar++; 2632 } else { 2633 /* default variation is +- 1 rtt */ 2634 tp->t_rttvar = 2635 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2636 } 2637 TCPT_RANGESET(tp->t_rxtcur, 2638 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2639 tp->t_rttmin, TCPTV_REXMTMAX); 2640 } 2641 /* 2642 * if there's an mtu associated with the route, use it 2643 * else, use the link mtu. 2644 */ 2645 if (rt->rt_rmx.rmx_mtu) 2646 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2647 else 2648 { 2649 mss = 2650 #ifdef INET6 2651 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2652 #endif 2653 ifp->if_mtu 2654 #ifdef INET6 2655 ) 2656 #endif 2657 - min_protoh; 2658 #ifdef INET6 2659 if (isipv6) { 2660 if (!in6_localaddr(&inp->in6p_faddr)) 2661 mss = min(mss, tcp_v6mssdflt); 2662 } else 2663 #endif 2664 if (!in_localaddr(inp->inp_faddr)) 2665 mss = min(mss, tcp_mssdflt); 2666 } 2667 mss = min(mss, offer); 2668 /* 2669 * maxopd stores the maximum length of data AND options 2670 * in a segment; maxseg is the amount of data in a normal 2671 * segment. We need to store this value (maxopd) apart 2672 * from maxseg, because now every segment carries options 2673 * and thus we normally have somewhat less data in segments. 2674 */ 2675 tp->t_maxopd = mss; 2676 2677 /* 2678 * In case of T/TCP, origoffer==-1 indicates, that no segments 2679 * were received yet. In this case we just guess, otherwise 2680 * we do the same as before T/TCP. 2681 */ 2682 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2683 (origoffer == -1 || 2684 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2685 mss -= TCPOLEN_TSTAMP_APPA; 2686 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2687 (origoffer == -1 || 2688 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2689 mss -= TCPOLEN_CC_APPA; 2690 2691 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2692 if (mss > MCLBYTES) 2693 mss &= ~(MCLBYTES-1); 2694 #else 2695 if (mss > MCLBYTES) 2696 mss = mss / MCLBYTES * MCLBYTES; 2697 #endif 2698 /* 2699 * If there's a pipesize, change the socket buffer 2700 * to that size. Make the socket buffers an integral 2701 * number of mss units; if the mss is larger than 2702 * the socket buffer, decrease the mss. 2703 */ 2704 #ifdef RTV_SPIPE 2705 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2706 #endif 2707 bufsize = so->so_snd.sb_hiwat; 2708 if (bufsize < mss) 2709 mss = bufsize; 2710 else { 2711 bufsize = roundup(bufsize, mss); 2712 if (bufsize > sb_max) 2713 bufsize = sb_max; 2714 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2715 } 2716 tp->t_maxseg = mss; 2717 2718 #ifdef RTV_RPIPE 2719 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2720 #endif 2721 bufsize = so->so_rcv.sb_hiwat; 2722 if (bufsize > mss) { 2723 bufsize = roundup(bufsize, mss); 2724 if (bufsize > sb_max) 2725 bufsize = sb_max; 2726 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2727 } 2728 2729 /* 2730 * Set the slow-start flight size depending on whether this 2731 * is a local network or not. 2732 */ 2733 if ( 2734 #ifdef INET6 2735 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2736 (!isipv6 && 2737 #endif 2738 in_localaddr(inp->inp_faddr) 2739 #ifdef INET6 2740 ) 2741 #endif 2742 ) 2743 tp->snd_cwnd = mss * ss_fltsz_local; 2744 else 2745 tp->snd_cwnd = mss * ss_fltsz; 2746 2747 if (rt->rt_rmx.rmx_ssthresh) { 2748 /* 2749 * There's some sort of gateway or interface 2750 * buffer limit on the path. Use this to set 2751 * the slow start threshhold, but set the 2752 * threshold to no less than 2*mss. 2753 */ 2754 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2755 tcpstat.tcps_usedssthresh++; 2756 } 2757 } 2758 2759 /* 2760 * Determine the MSS option to send on an outgoing SYN. 2761 */ 2762 int 2763 tcp_mssopt(tp) 2764 struct tcpcb *tp; 2765 { 2766 struct rtentry *rt; 2767 #ifdef INET6 2768 int isipv6; 2769 int min_protoh; 2770 #endif 2771 2772 #ifdef INET6 2773 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2774 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2775 : sizeof (struct tcpiphdr); 2776 #else 2777 #define min_protoh (sizeof (struct tcpiphdr)) 2778 #endif 2779 #ifdef INET6 2780 if (isipv6) 2781 rt = tcp_rtlookup6(tp->t_inpcb); 2782 else 2783 #endif /* INET6 */ 2784 rt = tcp_rtlookup(tp->t_inpcb); 2785 if (rt == NULL) 2786 return 2787 #ifdef INET6 2788 isipv6 ? tcp_v6mssdflt : 2789 #endif /* INET6 */ 2790 tcp_mssdflt; 2791 2792 return rt->rt_ifp->if_mtu - min_protoh; 2793 } 2794