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