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 #include <netinet/in_var.h> 65 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 66 #include <netinet/in_pcb.h> 67 #include <netinet/ip_var.h> 68 #ifdef INET6 69 #include <netinet/ip6.h> 70 #include <netinet/icmp6.h> 71 #include <netinet6/nd6.h> 72 #include <netinet6/ip6_var.h> 73 #include <netinet6/in6_pcb.h> 74 #endif 75 #include <netinet/tcp.h> 76 #include <netinet/tcp_fsm.h> 77 #include <netinet/tcp_seq.h> 78 #include <netinet/tcp_timer.h> 79 #include <netinet/tcp_var.h> 80 #ifdef INET6 81 #include <netinet6/tcp6_var.h> 82 #endif 83 #include <netinet/tcpip.h> 84 #ifdef TCPDEBUG 85 #include <netinet/tcp_debug.h> 86 87 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 88 struct tcphdr tcp_savetcp; 89 #endif /* TCPDEBUG */ 90 91 #ifdef IPSEC 92 #include <netinet6/ipsec.h> 93 #ifdef INET6 94 #include <netinet6/ipsec6.h> 95 #endif 96 #include <netkey/key.h> 97 #endif /*IPSEC*/ 98 99 #include <machine/in_cksum.h> 100 101 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 102 103 static int tcprexmtthresh = 3; 104 tcp_seq tcp_iss; 105 tcp_cc tcp_ccgen; 106 107 struct tcpstat tcpstat; 108 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RD, 109 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 110 111 static int log_in_vain = 0; 112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 113 &log_in_vain, 0, "Log all incoming TCP connections"); 114 115 static int blackhole = 0; 116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 117 &blackhole, 0, "Do not send RST when dropping refused connections"); 118 119 int tcp_delack_enabled = 1; 120 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 121 &tcp_delack_enabled, 0, 122 "Delay ACK to try and piggyback it onto a data packet"); 123 124 int tcp_lq_overflow = 1; 125 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_lq_overflow, CTLFLAG_RW, 126 &tcp_lq_overflow, 0, 127 "Listen Queue Overflow"); 128 129 #ifdef TCP_DROP_SYNFIN 130 static int drop_synfin = 0; 131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 132 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 133 #endif 134 135 struct inpcbhead tcb; 136 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 137 struct inpcbinfo tcbinfo; 138 139 static void tcp_dooptions __P((struct tcpcb *, 140 u_char *, int, struct tcphdr *, struct tcpopt *)); 141 static void tcp_pulloutofband __P((struct socket *, 142 struct tcphdr *, struct mbuf *, int)); 143 static int tcp_reass __P((struct tcpcb *, struct tcphdr *, int *, 144 struct mbuf *)); 145 static void tcp_xmit_timer __P((struct tcpcb *, int)); 146 static int tcp_newreno __P((struct tcpcb *, struct tcphdr *)); 147 148 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 149 #ifdef INET6 150 #define ND6_HINT(tp) \ 151 do { \ 152 if ((tp) && (tp)->t_inpcb && \ 153 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0 && \ 154 (tp)->t_inpcb->in6p_route.ro_rt) \ 155 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 156 } while (0) 157 #else 158 #define ND6_HINT(tp) 159 #endif 160 161 /* 162 * Indicate whether this ack should be delayed. 163 */ 164 #define DELAY_ACK(tp) \ 165 (tcp_delack_enabled && !callout_pending(tp->tt_delack)) 166 167 /* 168 * Insert segment which inludes th into reassembly queue of tcp with 169 * control block tp. Return TH_FIN if reassembly now includes 170 * a segment with FIN. The macro form does the common case inline 171 * (segment is the next to be received on an established connection, 172 * and the queue is empty), avoiding linkage into and removal 173 * from the queue and repetition of various conversions. 174 * Set DELACK for segments received in order, but ack immediately 175 * when segments are out of order (so fast retransmit can work). 176 */ 177 #define TCP_REASS(tp, th, tlenp, m, so, flags) { \ 178 if ((th)->th_seq == (tp)->rcv_nxt && \ 179 LIST_EMPTY(&(tp)->t_segq) && \ 180 (tp)->t_state == TCPS_ESTABLISHED) { \ 181 if (DELAY_ACK(tp)) \ 182 callout_reset(tp->tt_delack, tcp_delacktime, \ 183 tcp_timer_delack, tp); \ 184 else \ 185 tp->t_flags |= TF_ACKNOW; \ 186 (tp)->rcv_nxt += *(tlenp); \ 187 flags = (th)->th_flags & TH_FIN; \ 188 tcpstat.tcps_rcvpack++;\ 189 tcpstat.tcps_rcvbyte += *(tlenp);\ 190 ND6_HINT(tp); \ 191 sbappend(&(so)->so_rcv, (m)); \ 192 sorwakeup(so); \ 193 } else { \ 194 (flags) = tcp_reass((tp), (th), (tlenp), (m)); \ 195 tp->t_flags |= TF_ACKNOW; \ 196 } \ 197 } 198 199 static int 200 tcp_reass(tp, th, tlenp, m) 201 register struct tcpcb *tp; 202 register struct tcphdr *th; 203 int *tlenp; 204 struct mbuf *m; 205 { 206 struct tseg_qent *q; 207 struct tseg_qent *p = NULL; 208 struct tseg_qent *nq; 209 struct tseg_qent *te; 210 struct socket *so = tp->t_inpcb->inp_socket; 211 int flags; 212 213 /* 214 * Call with th==0 after become established to 215 * force pre-ESTABLISHED data up to user socket. 216 */ 217 if (th == 0) 218 goto present; 219 220 /* Allocate a new queue entry. If we can't, just drop the pkt. XXX */ 221 MALLOC(te, struct tseg_qent *, sizeof (struct tseg_qent), M_TSEGQ, 222 M_NOWAIT); 223 if (te == NULL) { 224 tcpstat.tcps_rcvmemdrop++; 225 m_freem(m); 226 return (0); 227 } 228 229 /* 230 * Find a segment which begins after this one does. 231 */ 232 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 233 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 234 break; 235 p = q; 236 } 237 238 /* 239 * If there is a preceding segment, it may provide some of 240 * our data already. If so, drop the data from the incoming 241 * segment. If it provides all of our data, drop us. 242 */ 243 if (p != NULL) { 244 register int i; 245 /* conversion to int (in i) handles seq wraparound */ 246 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 247 if (i > 0) { 248 if (i >= *tlenp) { 249 tcpstat.tcps_rcvduppack++; 250 tcpstat.tcps_rcvdupbyte += *tlenp; 251 m_freem(m); 252 FREE(te, M_TSEGQ); 253 /* 254 * Try to present any queued data 255 * at the left window edge to the user. 256 * This is needed after the 3-WHS 257 * completes. 258 */ 259 goto present; /* ??? */ 260 } 261 m_adj(m, i); 262 *tlenp -= i; 263 th->th_seq += i; 264 } 265 } 266 tcpstat.tcps_rcvoopack++; 267 tcpstat.tcps_rcvoobyte += *tlenp; 268 269 /* 270 * While we overlap succeeding segments trim them or, 271 * if they are completely covered, dequeue them. 272 */ 273 while (q) { 274 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 275 if (i <= 0) 276 break; 277 if (i < q->tqe_len) { 278 q->tqe_th->th_seq += i; 279 q->tqe_len -= i; 280 m_adj(q->tqe_m, i); 281 break; 282 } 283 284 nq = LIST_NEXT(q, tqe_q); 285 LIST_REMOVE(q, tqe_q); 286 m_freem(q->tqe_m); 287 FREE(q, M_TSEGQ); 288 q = nq; 289 } 290 291 /* Insert the new segment queue entry into place. */ 292 te->tqe_m = m; 293 te->tqe_th = th; 294 te->tqe_len = *tlenp; 295 296 if (p == NULL) { 297 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 298 } else { 299 LIST_INSERT_AFTER(p, te, tqe_q); 300 } 301 302 present: 303 /* 304 * Present data to user, advancing rcv_nxt through 305 * completed sequence space. 306 */ 307 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 308 return (0); 309 q = LIST_FIRST(&tp->t_segq); 310 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 311 return (0); 312 do { 313 tp->rcv_nxt += q->tqe_len; 314 flags = q->tqe_th->th_flags & TH_FIN; 315 nq = LIST_NEXT(q, tqe_q); 316 LIST_REMOVE(q, tqe_q); 317 if (so->so_state & SS_CANTRCVMORE) 318 m_freem(q->tqe_m); 319 else 320 sbappend(&so->so_rcv, q->tqe_m); 321 FREE(q, M_TSEGQ); 322 q = nq; 323 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 324 ND6_HINT(tp); 325 sorwakeup(so); 326 return (flags); 327 } 328 329 /* 330 * TCP input routine, follows pages 65-76 of the 331 * protocol specification dated September, 1981 very closely. 332 */ 333 #ifdef INET6 334 int 335 tcp6_input(mp, offp, proto) 336 struct mbuf **mp; 337 int *offp, proto; 338 { 339 register struct mbuf *m = *mp; 340 341 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 342 343 /* 344 * draft-itojun-ipv6-tcp-to-anycast 345 * better place to put this in? 346 */ 347 if (m->m_flags & M_ANYCAST6) { 348 struct ip6_hdr *ip6; 349 350 ip6 = mtod(m, struct ip6_hdr *); 351 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 352 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 353 return IPPROTO_DONE; 354 } 355 356 tcp_input(m, *offp, proto); 357 return IPPROTO_DONE; 358 } 359 #endif 360 361 void 362 tcp_input(m, off0, proto) 363 register struct mbuf *m; 364 int off0, proto; 365 { 366 register struct tcphdr *th; 367 register struct ip *ip = NULL; 368 register struct ipovly *ipov; 369 register struct inpcb *inp; 370 u_char *optp = NULL; 371 int optlen = 0; 372 int len, tlen, off; 373 int drop_hdrlen; 374 register struct tcpcb *tp = 0; 375 register int thflags; 376 struct socket *so = 0; 377 int todrop, acked, ourfinisacked, needoutput = 0; 378 struct in_addr laddr; 379 #ifdef INET6 380 struct in6_addr laddr6; 381 #endif 382 int dropsocket = 0; 383 int iss = 0; 384 u_long tiwin; 385 struct tcpopt to; /* options in this segment */ 386 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 387 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 388 #ifdef TCPDEBUG 389 short ostate = 0; 390 #endif 391 #ifdef INET6 392 struct ip6_hdr *ip6 = NULL; 393 int isipv6; 394 #endif /* INET6 */ 395 int rstreason; /* For badport_bandlim accounting purposes */ 396 397 #ifdef INET6 398 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 399 #endif 400 bzero((char *)&to, sizeof(to)); 401 402 tcpstat.tcps_rcvtotal++; 403 404 #ifdef INET6 405 if (isipv6) { 406 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 407 ip6 = mtod(m, struct ip6_hdr *); 408 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 409 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 410 tcpstat.tcps_rcvbadsum++; 411 goto drop; 412 } 413 th = (struct tcphdr *)((caddr_t)ip6 + off0); 414 } else 415 #endif /* INET6 */ 416 { 417 /* 418 * Get IP and TCP header together in first mbuf. 419 * Note: IP leaves IP header in first mbuf. 420 */ 421 if (off0 > sizeof (struct ip)) { 422 ip_stripoptions(m, (struct mbuf *)0); 423 off0 = sizeof(struct ip); 424 } 425 if (m->m_len < sizeof (struct tcpiphdr)) { 426 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 427 tcpstat.tcps_rcvshort++; 428 return; 429 } 430 } 431 ip = mtod(m, struct ip *); 432 ipov = (struct ipovly *)ip; 433 th = (struct tcphdr *)((caddr_t)ip + off0); 434 tlen = ip->ip_len; 435 436 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 437 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 438 th->th_sum = m->m_pkthdr.csum_data; 439 else 440 th->th_sum = in_pseudo(ip->ip_src.s_addr, 441 ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + 442 ip->ip_len + IPPROTO_TCP)); 443 th->th_sum ^= 0xffff; 444 } else { 445 /* 446 * Checksum extended TCP header and data. 447 */ 448 len = sizeof (struct ip) + tlen; 449 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 450 ipov->ih_len = (u_short)tlen; 451 HTONS(ipov->ih_len); 452 th->th_sum = in_cksum(m, len); 453 } 454 if (th->th_sum) { 455 tcpstat.tcps_rcvbadsum++; 456 goto drop; 457 } 458 #ifdef INET6 459 /* Re-initialization for later version check */ 460 ip->ip_v = IPVERSION; 461 #endif 462 } 463 464 /* 465 * Check that TCP offset makes sense, 466 * pull out TCP options and adjust length. XXX 467 */ 468 off = th->th_off << 2; 469 if (off < sizeof (struct tcphdr) || off > tlen) { 470 tcpstat.tcps_rcvbadoff++; 471 goto drop; 472 } 473 tlen -= off; /* tlen is used instead of ti->ti_len */ 474 if (off > sizeof (struct tcphdr)) { 475 #ifdef INET6 476 if (isipv6) { 477 IP6_EXTHDR_CHECK(m, off0, off, ); 478 ip6 = mtod(m, struct ip6_hdr *); 479 th = (struct tcphdr *)((caddr_t)ip6 + off0); 480 } else 481 #endif /* INET6 */ 482 { 483 if (m->m_len < sizeof(struct ip) + off) { 484 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) { 485 tcpstat.tcps_rcvshort++; 486 return; 487 } 488 ip = mtod(m, struct ip *); 489 ipov = (struct ipovly *)ip; 490 th = (struct tcphdr *)((caddr_t)ip + off0); 491 } 492 } 493 optlen = off - sizeof (struct tcphdr); 494 optp = (u_char *)(th + 1); 495 } 496 thflags = th->th_flags; 497 498 #ifdef TCP_DROP_SYNFIN 499 /* 500 * If the drop_synfin option is enabled, drop all packets with 501 * both the SYN and FIN bits set. This prevents e.g. nmap from 502 * identifying the TCP/IP stack. 503 * 504 * This is a violation of the TCP specification. 505 */ 506 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 507 goto drop; 508 #endif 509 510 /* 511 * Convert TCP protocol specific fields to host format. 512 */ 513 NTOHL(th->th_seq); 514 NTOHL(th->th_ack); 515 NTOHS(th->th_win); 516 NTOHS(th->th_urp); 517 518 /* 519 * Delay droping TCP, IP headers, IPv6 ext headers, and TCP options, 520 * until after ip6_savecontrol() is called and before other functions 521 * which don't want those proto headers. 522 * Because ip6_savecontrol() is going to parse the mbuf to 523 * search for data to be passed up to user-land, it wants mbuf 524 * parameters to be unchanged. 525 */ 526 drop_hdrlen = off0 + off; 527 528 /* 529 * Locate pcb for segment. 530 */ 531 findpcb: 532 #ifdef IPFIREWALL_FORWARD 533 if (ip_fw_fwd_addr != NULL 534 #ifdef INET6 535 && isipv6 == NULL /* IPv6 support is not yet */ 536 #endif /* INET6 */ 537 ) { 538 /* 539 * Diverted. Pretend to be the destination. 540 * already got one like this? 541 */ 542 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 543 ip->ip_dst, th->th_dport, 0, m->m_pkthdr.rcvif); 544 if (!inp) { 545 /* 546 * No, then it's new. Try find the ambushing socket 547 */ 548 if (!ip_fw_fwd_addr->sin_port) { 549 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, 550 th->th_sport, ip_fw_fwd_addr->sin_addr, 551 th->th_dport, 1, m->m_pkthdr.rcvif); 552 } else { 553 inp = in_pcblookup_hash(&tcbinfo, 554 ip->ip_src, th->th_sport, 555 ip_fw_fwd_addr->sin_addr, 556 ntohs(ip_fw_fwd_addr->sin_port), 1, 557 m->m_pkthdr.rcvif); 558 } 559 } 560 ip_fw_fwd_addr = NULL; 561 } else 562 #endif /* IPFIREWALL_FORWARD */ 563 { 564 #ifdef INET6 565 if (isipv6) 566 inp = in6_pcblookup_hash(&tcbinfo, &ip6->ip6_src, th->th_sport, 567 &ip6->ip6_dst, th->th_dport, 1, 568 m->m_pkthdr.rcvif); 569 else 570 #endif /* INET6 */ 571 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 572 ip->ip_dst, th->th_dport, 1, m->m_pkthdr.rcvif); 573 } 574 575 #ifdef IPSEC 576 #ifdef INET6 577 if (isipv6) { 578 if (inp != NULL && ipsec6_in_reject_so(m, inp->inp_socket)) { 579 ipsec6stat.in_polvio++; 580 goto drop; 581 } 582 } else 583 #endif /* INET6 */ 584 if (inp != NULL && ipsec4_in_reject_so(m, inp->inp_socket)) { 585 ipsecstat.in_polvio++; 586 goto drop; 587 } 588 #endif /*IPSEC*/ 589 590 /* 591 * If the state is CLOSED (i.e., TCB does not exist) then 592 * all data in the incoming segment is discarded. 593 * If the TCB exists but is in CLOSED state, it is embryonic, 594 * but should either do a listen or a connect soon. 595 */ 596 if (inp == NULL) { 597 if (log_in_vain) { 598 #ifdef INET6 599 char dbuf[INET6_ADDRSTRLEN], sbuf[INET6_ADDRSTRLEN]; 600 #else /* INET6 */ 601 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 602 #endif /* INET6 */ 603 604 #ifdef INET6 605 if (isipv6) { 606 strcpy(dbuf, ip6_sprintf(&ip6->ip6_dst)); 607 strcpy(sbuf, ip6_sprintf(&ip6->ip6_src)); 608 } else 609 #endif 610 { 611 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 612 strcpy(sbuf, inet_ntoa(ip->ip_src)); 613 } 614 switch (log_in_vain) { 615 case 1: 616 if(thflags & TH_SYN) 617 log(LOG_INFO, 618 "Connection attempt to TCP %s:%d from %s:%d\n", 619 dbuf, ntohs(th->th_dport), 620 sbuf, 621 ntohs(th->th_sport)); 622 break; 623 case 2: 624 log(LOG_INFO, 625 "Connection attempt to TCP %s:%d from %s:%d flags:0x%x\n", 626 dbuf, ntohs(th->th_dport), sbuf, 627 ntohs(th->th_sport), thflags); 628 break; 629 default: 630 break; 631 } 632 } 633 if (blackhole) { 634 switch (blackhole) { 635 case 1: 636 if (thflags & TH_SYN) 637 goto drop; 638 break; 639 case 2: 640 goto drop; 641 default: 642 goto drop; 643 } 644 } 645 rstreason = BANDLIM_RST_CLOSEDPORT; 646 goto dropwithreset; 647 } 648 tp = intotcpcb(inp); 649 if (tp == 0) { 650 rstreason = BANDLIM_RST_CLOSEDPORT; 651 goto dropwithreset; 652 } 653 if (tp->t_state == TCPS_CLOSED) 654 goto drop; 655 656 /* Unscale the window into a 32-bit value. */ 657 if ((thflags & TH_SYN) == 0) 658 tiwin = th->th_win << tp->snd_scale; 659 else 660 tiwin = th->th_win; 661 662 #ifdef INET6 663 /* save packet options if user wanted */ 664 if (isipv6 && inp->in6p_flags & INP_CONTROLOPTS) { 665 if (inp->in6p_options) { 666 m_freem(inp->in6p_options); 667 inp->in6p_options = 0; 668 } 669 ip6_savecontrol(inp, &inp->in6p_options, ip6, m); 670 } 671 /* else, should also do ip_srcroute() here? */ 672 #endif /* INET6 */ 673 674 so = inp->inp_socket; 675 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 676 #ifdef TCPDEBUG 677 if (so->so_options & SO_DEBUG) { 678 ostate = tp->t_state; 679 #ifdef INET6 680 if (isipv6) 681 bcopy((char *)ip6, (char *)tcp_saveipgen, 682 sizeof(*ip6)); 683 else 684 #endif /* INET6 */ 685 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 686 tcp_savetcp = *th; 687 } 688 #endif 689 if (so->so_options & SO_ACCEPTCONN) { 690 register struct tcpcb *tp0 = tp; 691 struct socket *so2; 692 #ifdef IPSEC 693 struct socket *oso; 694 #endif 695 #ifdef INET6 696 struct inpcb *oinp = sotoinpcb(so); 697 #endif /* INET6 */ 698 699 #ifndef IPSEC 700 /* 701 * Current IPsec implementation makes incorrect IPsec 702 * cache if this check is done here. 703 * So delay this until duplicated socket is created. 704 */ 705 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 706 /* 707 * Note: dropwithreset makes sure we don't 708 * send a RST in response to a RST. 709 */ 710 if (thflags & TH_ACK) { 711 tcpstat.tcps_badsyn++; 712 rstreason = BANDLIM_RST_OPENPORT; 713 goto dropwithreset; 714 } 715 goto drop; 716 } 717 #endif 718 so2 = sonewconn(so, 0); 719 if (so2 == 0) { 720 tcpstat.tcps_listendrop++; 721 so2 = sodropablereq(so); 722 if (so2) { 723 if (tcp_lq_overflow) 724 sototcpcb(so2)->t_flags |= 725 TF_LQ_OVERFLOW; 726 tcp_drop(sototcpcb(so2), ETIMEDOUT); 727 so2 = sonewconn(so, 0); 728 } 729 if (!so2) 730 goto drop; 731 } 732 #ifdef IPSEC 733 oso = so; 734 #endif 735 so = so2; 736 /* 737 * This is ugly, but .... 738 * 739 * Mark socket as temporary until we're 740 * committed to keeping it. The code at 741 * ``drop'' and ``dropwithreset'' check the 742 * flag dropsocket to see if the temporary 743 * socket created here should be discarded. 744 * We mark the socket as discardable until 745 * we're committed to it below in TCPS_LISTEN. 746 */ 747 dropsocket++; 748 inp = (struct inpcb *)so->so_pcb; 749 #ifdef INET6 750 if (isipv6) 751 inp->in6p_laddr = ip6->ip6_dst; 752 else { 753 if ((inp->inp_flags & IN6P_BINDV6ONLY) == 0) { 754 inp->inp_vflag &= ~INP_IPV6; 755 inp->inp_vflag |= INP_IPV4; 756 } 757 #endif /* INET6 */ 758 inp->inp_laddr = ip->ip_dst; 759 #ifdef INET6 760 } 761 #endif /* INET6 */ 762 inp->inp_lport = th->th_dport; 763 if (in_pcbinshash(inp) != 0) { 764 /* 765 * Undo the assignments above if we failed to 766 * put the PCB on the hash lists. 767 */ 768 #ifdef INET6 769 if (isipv6) 770 inp->in6p_laddr = in6addr_any; 771 else 772 #endif /* INET6 */ 773 inp->inp_laddr.s_addr = INADDR_ANY; 774 inp->inp_lport = 0; 775 goto drop; 776 } 777 #ifdef IPSEC 778 /* 779 * To avoid creating incorrectly cached IPsec 780 * association, this is need to be done here. 781 * 782 * Subject: (KAME-snap 748) 783 * From: Wayne Knowles <w.knowles@niwa.cri.nz> 784 * ftp://ftp.kame.net/pub/mail-list/snap-users/748 785 */ 786 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 787 /* 788 * Note: dropwithreset makes sure we don't 789 * send a RST in response to a RST. 790 */ 791 if (thflags & TH_ACK) { 792 tcpstat.tcps_badsyn++; 793 rstreason = BANDLIM_RST_OPENPORT; 794 goto dropwithreset; 795 } 796 goto drop; 797 } 798 #endif 799 #ifdef INET6 800 if (isipv6) { 801 /* 802 * inherit socket options from the listening 803 * socket. 804 */ 805 inp->inp_flags |= 806 oinp->inp_flags & INP_CONTROLOPTS; 807 if (inp->inp_flags & INP_CONTROLOPTS) { 808 if (inp->in6p_options) { 809 m_freem(inp->in6p_options); 810 inp->in6p_options = 0; 811 } 812 ip6_savecontrol(inp, 813 &inp->in6p_options, 814 ip6, m); 815 } 816 } else 817 #endif /* INET6 */ 818 inp->inp_options = ip_srcroute(); 819 #ifdef IPSEC 820 /* copy old policy into new socket's */ 821 if (ipsec_copy_policy(sotoinpcb(oso)->inp_sp, 822 inp->inp_sp)) 823 printf("tcp_input: could not copy policy\n"); 824 #endif 825 tp = intotcpcb(inp); 826 tp->t_state = TCPS_LISTEN; 827 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT); 828 829 /* Compute proper scaling value from buffer space */ 830 while (tp->request_r_scale < TCP_MAX_WINSHIFT && 831 TCP_MAXWIN << tp->request_r_scale < 832 so->so_rcv.sb_hiwat) 833 tp->request_r_scale++; 834 } 835 } 836 837 /* 838 * Segment received on connection. 839 * Reset idle time and keep-alive timer. 840 */ 841 tp->t_rcvtime = ticks; 842 if (TCPS_HAVEESTABLISHED(tp->t_state)) 843 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 844 845 /* 846 * Process options if not in LISTEN state, 847 * else do it below (after getting remote address). 848 */ 849 if (tp->t_state != TCPS_LISTEN) 850 tcp_dooptions(tp, optp, optlen, th, &to); 851 852 /* 853 * Header prediction: check for the two common cases 854 * of a uni-directional data xfer. If the packet has 855 * no control flags, is in-sequence, the window didn't 856 * change and we're not retransmitting, it's a 857 * candidate. If the length is zero and the ack moved 858 * forward, we're the sender side of the xfer. Just 859 * free the data acked & wake any higher level process 860 * that was blocked waiting for space. If the length 861 * is non-zero and the ack didn't move, we're the 862 * receiver side. If we're getting packets in-order 863 * (the reassembly queue is empty), add the data to 864 * the socket buffer and note that we need a delayed ack. 865 * Make sure that the hidden state-flags are also off. 866 * Since we check for TCPS_ESTABLISHED above, it can only 867 * be TH_NEEDSYN. 868 */ 869 if (tp->t_state == TCPS_ESTABLISHED && 870 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 871 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 872 ((to.to_flag & TOF_TS) == 0 || 873 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 874 /* 875 * Using the CC option is compulsory if once started: 876 * the segment is OK if no T/TCP was negotiated or 877 * if the segment has a CC option equal to CCrecv 878 */ 879 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 880 ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 881 th->th_seq == tp->rcv_nxt && 882 tiwin && tiwin == tp->snd_wnd && 883 tp->snd_nxt == tp->snd_max) { 884 885 /* 886 * If last ACK falls within this segment's sequence numbers, 887 * record the timestamp. 888 * NOTE that the test is modified according to the latest 889 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 890 */ 891 if ((to.to_flag & TOF_TS) != 0 && 892 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 893 tp->ts_recent_age = ticks; 894 tp->ts_recent = to.to_tsval; 895 } 896 897 if (tlen == 0) { 898 if (SEQ_GT(th->th_ack, tp->snd_una) && 899 SEQ_LEQ(th->th_ack, tp->snd_max) && 900 tp->snd_cwnd >= tp->snd_wnd && 901 tp->t_dupacks < tcprexmtthresh) { 902 /* 903 * this is a pure ack for outstanding data. 904 */ 905 ++tcpstat.tcps_predack; 906 /* 907 * "bad retransmit" recovery 908 */ 909 if (tp->t_rxtshift == 1 && 910 ticks < tp->t_badrxtwin) { 911 tp->snd_cwnd = tp->snd_cwnd_prev; 912 tp->snd_ssthresh = 913 tp->snd_ssthresh_prev; 914 tp->snd_nxt = tp->snd_max; 915 tp->t_badrxtwin = 0; 916 } 917 if ((to.to_flag & TOF_TS) != 0) 918 tcp_xmit_timer(tp, 919 ticks - to.to_tsecr + 1); 920 else if (tp->t_rtttime && 921 SEQ_GT(th->th_ack, tp->t_rtseq)) 922 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 923 acked = th->th_ack - tp->snd_una; 924 tcpstat.tcps_rcvackpack++; 925 tcpstat.tcps_rcvackbyte += acked; 926 sbdrop(&so->so_snd, acked); 927 tp->snd_una = th->th_ack; 928 m_freem(m); 929 ND6_HINT(tp); /* some progress has been done */ 930 931 /* 932 * If all outstanding data are acked, stop 933 * retransmit timer, otherwise restart timer 934 * using current (possibly backed-off) value. 935 * If process is waiting for space, 936 * wakeup/selwakeup/signal. If data 937 * are ready to send, let tcp_output 938 * decide between more output or persist. 939 */ 940 if (tp->snd_una == tp->snd_max) 941 callout_stop(tp->tt_rexmt); 942 else if (!callout_active(tp->tt_persist)) 943 callout_reset(tp->tt_rexmt, 944 tp->t_rxtcur, 945 tcp_timer_rexmt, tp); 946 947 sowwakeup(so); 948 if (so->so_snd.sb_cc) 949 (void) tcp_output(tp); 950 return; 951 } 952 } else if (th->th_ack == tp->snd_una && 953 LIST_EMPTY(&tp->t_segq) && 954 tlen <= sbspace(&so->so_rcv)) { 955 /* 956 * this is a pure, in-sequence data packet 957 * with nothing on the reassembly queue and 958 * we have enough buffer space to take it. 959 */ 960 ++tcpstat.tcps_preddat; 961 tp->rcv_nxt += tlen; 962 tcpstat.tcps_rcvpack++; 963 tcpstat.tcps_rcvbyte += tlen; 964 ND6_HINT(tp); /* some progress has been done */ 965 /* 966 * Add data to socket buffer. 967 */ 968 m_adj(m, drop_hdrlen); /* delayed header drop */ 969 sbappend(&so->so_rcv, m); 970 sorwakeup(so); 971 if (DELAY_ACK(tp)) { 972 callout_reset(tp->tt_delack, tcp_delacktime, 973 tcp_timer_delack, tp); 974 } else { 975 tp->t_flags |= TF_ACKNOW; 976 tcp_output(tp); 977 } 978 return; 979 } 980 } 981 982 /* 983 * Calculate amount of space in receive window, 984 * and then do TCP input processing. 985 * Receive window is amount of space in rcv queue, 986 * but not less than advertised window. 987 */ 988 { int win; 989 990 win = sbspace(&so->so_rcv); 991 if (win < 0) 992 win = 0; 993 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 994 } 995 996 switch (tp->t_state) { 997 998 /* 999 * If the state is LISTEN then ignore segment if it contains an RST. 1000 * If the segment contains an ACK then it is bad and send a RST. 1001 * If it does not contain a SYN then it is not interesting; drop it. 1002 * If it is from this socket, drop it, it must be forged. 1003 * Don't bother responding if the destination was a broadcast. 1004 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 1005 * tp->iss, and send a segment: 1006 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 1007 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 1008 * Fill in remote peer address fields if not previously specified. 1009 * Enter SYN_RECEIVED state, and process any other fields of this 1010 * segment in this state. 1011 */ 1012 case TCPS_LISTEN: { 1013 register struct sockaddr_in *sin; 1014 #ifdef INET6 1015 register struct sockaddr_in6 *sin6; 1016 #endif 1017 1018 if (thflags & TH_RST) 1019 goto drop; 1020 if (thflags & TH_ACK) { 1021 rstreason = BANDLIM_RST_OPENPORT; 1022 goto dropwithreset; 1023 } 1024 if ((thflags & TH_SYN) == 0) 1025 goto drop; 1026 if (th->th_dport == th->th_sport) { 1027 #ifdef INET6 1028 if (isipv6) { 1029 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 1030 &ip6->ip6_src)) 1031 goto drop; 1032 } else 1033 #endif /* INET6 */ 1034 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 1035 goto drop; 1036 } 1037 /* 1038 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1039 * in_broadcast() should never return true on a received 1040 * packet with M_BCAST not set. 1041 * 1042 * Packets with a multicast source address should also 1043 * be discarded. 1044 */ 1045 if (m->m_flags & (M_BCAST|M_MCAST)) 1046 goto drop; 1047 #ifdef INET6 1048 if (isipv6) { 1049 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1050 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 1051 goto drop; 1052 } else 1053 #endif 1054 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1055 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1056 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 1057 goto drop; 1058 #ifdef INET6 1059 if (isipv6) { 1060 MALLOC(sin6, struct sockaddr_in6 *, sizeof *sin6, 1061 M_SONAME, M_NOWAIT | M_ZERO); 1062 if (sin6 == NULL) 1063 goto drop; 1064 sin6->sin6_family = AF_INET6; 1065 sin6->sin6_len = sizeof(*sin6); 1066 sin6->sin6_addr = ip6->ip6_src; 1067 sin6->sin6_port = th->th_sport; 1068 laddr6 = inp->in6p_laddr; 1069 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) 1070 inp->in6p_laddr = ip6->ip6_dst; 1071 if (in6_pcbconnect(inp, (struct sockaddr *)sin6, 1072 &proc0)) { 1073 inp->in6p_laddr = laddr6; 1074 FREE(sin6, M_SONAME); 1075 goto drop; 1076 } 1077 FREE(sin6, M_SONAME); 1078 } else 1079 #endif 1080 { 1081 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME, 1082 M_NOWAIT); 1083 if (sin == NULL) 1084 goto drop; 1085 sin->sin_family = AF_INET; 1086 sin->sin_len = sizeof(*sin); 1087 sin->sin_addr = ip->ip_src; 1088 sin->sin_port = th->th_sport; 1089 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 1090 laddr = inp->inp_laddr; 1091 if (inp->inp_laddr.s_addr == INADDR_ANY) 1092 inp->inp_laddr = ip->ip_dst; 1093 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) { 1094 inp->inp_laddr = laddr; 1095 FREE(sin, M_SONAME); 1096 goto drop; 1097 } 1098 FREE(sin, M_SONAME); 1099 } 1100 tp->t_template = tcp_template(tp); 1101 if (tp->t_template == 0) { 1102 tp = tcp_drop(tp, ENOBUFS); 1103 dropsocket = 0; /* socket is already gone */ 1104 goto drop; 1105 } 1106 if ((taop = tcp_gettaocache(inp)) == NULL) { 1107 taop = &tao_noncached; 1108 bzero(taop, sizeof(*taop)); 1109 } 1110 tcp_dooptions(tp, optp, optlen, th, &to); 1111 if (iss) 1112 tp->iss = iss; 1113 else { 1114 tp->iss = tcp_rndiss_next(); 1115 } 1116 tp->irs = th->th_seq; 1117 tcp_sendseqinit(tp); 1118 tcp_rcvseqinit(tp); 1119 tp->snd_recover = tp->snd_una; 1120 /* 1121 * Initialization of the tcpcb for transaction; 1122 * set SND.WND = SEG.WND, 1123 * initialize CCsend and CCrecv. 1124 */ 1125 tp->snd_wnd = tiwin; /* initial send-window */ 1126 tp->cc_send = CC_INC(tcp_ccgen); 1127 tp->cc_recv = to.to_cc; 1128 /* 1129 * Perform TAO test on incoming CC (SEG.CC) option, if any. 1130 * - compare SEG.CC against cached CC from the same host, 1131 * if any. 1132 * - if SEG.CC > chached value, SYN must be new and is accepted 1133 * immediately: save new CC in the cache, mark the socket 1134 * connected, enter ESTABLISHED state, turn on flag to 1135 * send a SYN in the next segment. 1136 * A virtual advertised window is set in rcv_adv to 1137 * initialize SWS prevention. Then enter normal segment 1138 * processing: drop SYN, process data and FIN. 1139 * - otherwise do a normal 3-way handshake. 1140 */ 1141 if ((to.to_flag & TOF_CC) != 0) { 1142 if (((tp->t_flags & TF_NOPUSH) != 0) && 1143 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) { 1144 1145 taop->tao_cc = to.to_cc; 1146 tp->t_starttime = ticks; 1147 tp->t_state = TCPS_ESTABLISHED; 1148 1149 /* 1150 * If there is a FIN, or if there is data and the 1151 * connection is local, then delay SYN,ACK(SYN) in 1152 * the hope of piggy-backing it on a response 1153 * segment. Otherwise must send ACK now in case 1154 * the other side is slow starting. 1155 */ 1156 if (DELAY_ACK(tp) && ((thflags & TH_FIN) || 1157 (tlen != 0 && 1158 #ifdef INET6 1159 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) 1160 || 1161 (!isipv6 && 1162 #endif 1163 in_localaddr(inp->inp_faddr) 1164 #ifdef INET6 1165 )) 1166 #endif 1167 ))) { 1168 callout_reset(tp->tt_delack, tcp_delacktime, 1169 tcp_timer_delack, tp); 1170 tp->t_flags |= TF_NEEDSYN; 1171 } else 1172 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1173 1174 /* 1175 * Limit the `virtual advertised window' to TCP_MAXWIN 1176 * here. Even if we requested window scaling, it will 1177 * become effective only later when our SYN is acked. 1178 */ 1179 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN); 1180 tcpstat.tcps_connects++; 1181 soisconnected(so); 1182 callout_reset(tp->tt_keep, tcp_keepinit, 1183 tcp_timer_keep, tp); 1184 dropsocket = 0; /* committed to socket */ 1185 tcpstat.tcps_accepts++; 1186 goto trimthenstep6; 1187 } 1188 /* else do standard 3-way handshake */ 1189 } else { 1190 /* 1191 * No CC option, but maybe CC.NEW: 1192 * invalidate cached value. 1193 */ 1194 taop->tao_cc = 0; 1195 } 1196 /* 1197 * TAO test failed or there was no CC option, 1198 * do a standard 3-way handshake. 1199 */ 1200 tp->t_flags |= TF_ACKNOW; 1201 tp->t_state = TCPS_SYN_RECEIVED; 1202 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp); 1203 dropsocket = 0; /* committed to socket */ 1204 tcpstat.tcps_accepts++; 1205 goto trimthenstep6; 1206 } 1207 1208 /* 1209 * If the state is SYN_RECEIVED: 1210 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1211 */ 1212 case TCPS_SYN_RECEIVED: 1213 if ((thflags & TH_ACK) && 1214 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1215 SEQ_GT(th->th_ack, tp->snd_max))) { 1216 rstreason = BANDLIM_RST_OPENPORT; 1217 goto dropwithreset; 1218 } 1219 break; 1220 1221 /* 1222 * If the state is SYN_SENT: 1223 * if seg contains an ACK, but not for our SYN, drop the input. 1224 * if seg contains a RST, then drop the connection. 1225 * if seg does not contain SYN, then drop it. 1226 * Otherwise this is an acceptable SYN segment 1227 * initialize tp->rcv_nxt and tp->irs 1228 * if seg contains ack then advance tp->snd_una 1229 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1230 * arrange for segment to be acked (eventually) 1231 * continue processing rest of data/controls, beginning with URG 1232 */ 1233 case TCPS_SYN_SENT: 1234 if ((taop = tcp_gettaocache(inp)) == NULL) { 1235 taop = &tao_noncached; 1236 bzero(taop, sizeof(*taop)); 1237 } 1238 1239 if ((thflags & TH_ACK) && 1240 (SEQ_LEQ(th->th_ack, tp->iss) || 1241 SEQ_GT(th->th_ack, tp->snd_max))) { 1242 /* 1243 * If we have a cached CCsent for the remote host, 1244 * hence we haven't just crashed and restarted, 1245 * do not send a RST. This may be a retransmission 1246 * from the other side after our earlier ACK was lost. 1247 * Our new SYN, when it arrives, will serve as the 1248 * needed ACK. 1249 */ 1250 if (taop->tao_ccsent != 0) 1251 goto drop; 1252 else { 1253 rstreason = BANDLIM_UNLIMITED; 1254 goto dropwithreset; 1255 } 1256 } 1257 if (thflags & TH_RST) { 1258 if (thflags & TH_ACK) 1259 tp = tcp_drop(tp, ECONNREFUSED); 1260 goto drop; 1261 } 1262 if ((thflags & TH_SYN) == 0) 1263 goto drop; 1264 tp->snd_wnd = th->th_win; /* initial send window */ 1265 tp->cc_recv = to.to_cc; /* foreign CC */ 1266 1267 tp->irs = th->th_seq; 1268 tcp_rcvseqinit(tp); 1269 if (thflags & TH_ACK) { 1270 /* 1271 * Our SYN was acked. If segment contains CC.ECHO 1272 * option, check it to make sure this segment really 1273 * matches our SYN. If not, just drop it as old 1274 * duplicate, but send an RST if we're still playing 1275 * by the old rules. If no CC.ECHO option, make sure 1276 * we don't get fooled into using T/TCP. 1277 */ 1278 if (to.to_flag & TOF_CCECHO) { 1279 if (tp->cc_send != to.to_ccecho) { 1280 if (taop->tao_ccsent != 0) 1281 goto drop; 1282 else { 1283 rstreason = BANDLIM_UNLIMITED; 1284 goto dropwithreset; 1285 } 1286 } 1287 } else 1288 tp->t_flags &= ~TF_RCVD_CC; 1289 tcpstat.tcps_connects++; 1290 soisconnected(so); 1291 /* Do window scaling on this connection? */ 1292 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1293 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1294 tp->snd_scale = tp->requested_s_scale; 1295 tp->rcv_scale = tp->request_r_scale; 1296 } 1297 /* Segment is acceptable, update cache if undefined. */ 1298 if (taop->tao_ccsent == 0) 1299 taop->tao_ccsent = to.to_ccecho; 1300 1301 tp->rcv_adv += tp->rcv_wnd; 1302 tp->snd_una++; /* SYN is acked */ 1303 /* 1304 * If there's data, delay ACK; if there's also a FIN 1305 * ACKNOW will be turned on later. 1306 */ 1307 if (DELAY_ACK(tp) && tlen != 0) 1308 callout_reset(tp->tt_delack, tcp_delacktime, 1309 tcp_timer_delack, tp); 1310 else 1311 tp->t_flags |= TF_ACKNOW; 1312 /* 1313 * Received <SYN,ACK> in SYN_SENT[*] state. 1314 * Transitions: 1315 * SYN_SENT --> ESTABLISHED 1316 * SYN_SENT* --> FIN_WAIT_1 1317 */ 1318 tp->t_starttime = ticks; 1319 if (tp->t_flags & TF_NEEDFIN) { 1320 tp->t_state = TCPS_FIN_WAIT_1; 1321 tp->t_flags &= ~TF_NEEDFIN; 1322 thflags &= ~TH_SYN; 1323 } else { 1324 tp->t_state = TCPS_ESTABLISHED; 1325 callout_reset(tp->tt_keep, tcp_keepidle, 1326 tcp_timer_keep, tp); 1327 } 1328 } else { 1329 /* 1330 * Received initial SYN in SYN-SENT[*] state => simul- 1331 * taneous open. If segment contains CC option and there is 1332 * a cached CC, apply TAO test; if it succeeds, connection is 1333 * half-synchronized. Otherwise, do 3-way handshake: 1334 * SYN-SENT -> SYN-RECEIVED 1335 * SYN-SENT* -> SYN-RECEIVED* 1336 * If there was no CC option, clear cached CC value. 1337 */ 1338 tp->t_flags |= TF_ACKNOW; 1339 callout_stop(tp->tt_rexmt); 1340 if (to.to_flag & TOF_CC) { 1341 if (taop->tao_cc != 0 && 1342 CC_GT(to.to_cc, taop->tao_cc)) { 1343 /* 1344 * update cache and make transition: 1345 * SYN-SENT -> ESTABLISHED* 1346 * SYN-SENT* -> FIN-WAIT-1* 1347 */ 1348 taop->tao_cc = to.to_cc; 1349 tp->t_starttime = ticks; 1350 if (tp->t_flags & TF_NEEDFIN) { 1351 tp->t_state = TCPS_FIN_WAIT_1; 1352 tp->t_flags &= ~TF_NEEDFIN; 1353 } else { 1354 tp->t_state = TCPS_ESTABLISHED; 1355 callout_reset(tp->tt_keep, 1356 tcp_keepidle, 1357 tcp_timer_keep, 1358 tp); 1359 } 1360 tp->t_flags |= TF_NEEDSYN; 1361 } else 1362 tp->t_state = TCPS_SYN_RECEIVED; 1363 } else { 1364 /* CC.NEW or no option => invalidate cache */ 1365 taop->tao_cc = 0; 1366 tp->t_state = TCPS_SYN_RECEIVED; 1367 } 1368 } 1369 1370 trimthenstep6: 1371 /* 1372 * Advance th->th_seq to correspond to first data byte. 1373 * If data, trim to stay within window, 1374 * dropping FIN if necessary. 1375 */ 1376 th->th_seq++; 1377 if (tlen > tp->rcv_wnd) { 1378 todrop = tlen - tp->rcv_wnd; 1379 m_adj(m, -todrop); 1380 tlen = tp->rcv_wnd; 1381 thflags &= ~TH_FIN; 1382 tcpstat.tcps_rcvpackafterwin++; 1383 tcpstat.tcps_rcvbyteafterwin += todrop; 1384 } 1385 tp->snd_wl1 = th->th_seq - 1; 1386 tp->rcv_up = th->th_seq; 1387 /* 1388 * Client side of transaction: already sent SYN and data. 1389 * If the remote host used T/TCP to validate the SYN, 1390 * our data will be ACK'd; if so, enter normal data segment 1391 * processing in the middle of step 5, ack processing. 1392 * Otherwise, goto step 6. 1393 */ 1394 if (thflags & TH_ACK) 1395 goto process_ACK; 1396 goto step6; 1397 /* 1398 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1399 * if segment contains a SYN and CC [not CC.NEW] option: 1400 * if state == TIME_WAIT and connection duration > MSL, 1401 * drop packet and send RST; 1402 * 1403 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1404 * ack the FIN (and data) in retransmission queue. 1405 * Complete close and delete TCPCB. Then reprocess 1406 * segment, hoping to find new TCPCB in LISTEN state; 1407 * 1408 * else must be old SYN; drop it. 1409 * else do normal processing. 1410 */ 1411 case TCPS_LAST_ACK: 1412 case TCPS_CLOSING: 1413 case TCPS_TIME_WAIT: 1414 if ((thflags & TH_SYN) && 1415 (to.to_flag & TOF_CC) && tp->cc_recv != 0) { 1416 if (tp->t_state == TCPS_TIME_WAIT && 1417 (ticks - tp->t_starttime) > tcp_msl) { 1418 rstreason = BANDLIM_UNLIMITED; 1419 goto dropwithreset; 1420 } 1421 if (CC_GT(to.to_cc, tp->cc_recv)) { 1422 tp = tcp_close(tp); 1423 goto findpcb; 1424 } 1425 else 1426 goto drop; 1427 } 1428 break; /* continue normal processing */ 1429 } 1430 1431 /* 1432 * States other than LISTEN or SYN_SENT. 1433 * First check the RST flag and sequence number since reset segments 1434 * are exempt from the timestamp and connection count tests. This 1435 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1436 * below which allowed reset segments in half the sequence space 1437 * to fall though and be processed (which gives forged reset 1438 * segments with a random sequence number a 50 percent chance of 1439 * killing a connection). 1440 * Then check timestamp, if present. 1441 * Then check the connection count, if present. 1442 * Then check that at least some bytes of segment are within 1443 * receive window. If segment begins before rcv_nxt, 1444 * drop leading data (and SYN); if nothing left, just ack. 1445 * 1446 * 1447 * If the RST bit is set, check the sequence number to see 1448 * if this is a valid reset segment. 1449 * RFC 793 page 37: 1450 * In all states except SYN-SENT, all reset (RST) segments 1451 * are validated by checking their SEQ-fields. A reset is 1452 * valid if its sequence number is in the window. 1453 * Note: this does not take into account delayed ACKs, so 1454 * we should test against last_ack_sent instead of rcv_nxt. 1455 * The sequence number in the reset segment is normally an 1456 * echo of our outgoing acknowlegement numbers, but some hosts 1457 * send a reset with the sequence number at the rightmost edge 1458 * of our receive window, and we have to handle this case. 1459 * If we have multiple segments in flight, the intial reset 1460 * segment sequence numbers will be to the left of last_ack_sent, 1461 * but they will eventually catch up. 1462 * In any case, it never made sense to trim reset segments to 1463 * fit the receive window since RFC 1122 says: 1464 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1465 * 1466 * A TCP SHOULD allow a received RST segment to include data. 1467 * 1468 * DISCUSSION 1469 * It has been suggested that a RST segment could contain 1470 * ASCII text that encoded and explained the cause of the 1471 * RST. No standard has yet been established for such 1472 * data. 1473 * 1474 * If the reset segment passes the sequence number test examine 1475 * the state: 1476 * SYN_RECEIVED STATE: 1477 * If passive open, return to LISTEN state. 1478 * If active open, inform user that connection was refused. 1479 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1480 * Inform user that connection was reset, and close tcb. 1481 * CLOSING, LAST_ACK STATES: 1482 * Close the tcb. 1483 * TIME_WAIT STATE: 1484 * Drop the segment - see Stevens, vol. 2, p. 964 and 1485 * RFC 1337. 1486 */ 1487 if (thflags & TH_RST) { 1488 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1489 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1490 switch (tp->t_state) { 1491 1492 case TCPS_SYN_RECEIVED: 1493 so->so_error = ECONNREFUSED; 1494 goto close; 1495 1496 case TCPS_ESTABLISHED: 1497 case TCPS_FIN_WAIT_1: 1498 case TCPS_FIN_WAIT_2: 1499 case TCPS_CLOSE_WAIT: 1500 so->so_error = ECONNRESET; 1501 close: 1502 tp->t_state = TCPS_CLOSED; 1503 tcpstat.tcps_drops++; 1504 tp = tcp_close(tp); 1505 break; 1506 1507 case TCPS_CLOSING: 1508 case TCPS_LAST_ACK: 1509 tp = tcp_close(tp); 1510 break; 1511 1512 case TCPS_TIME_WAIT: 1513 break; 1514 } 1515 } 1516 goto drop; 1517 } 1518 1519 /* 1520 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1521 * and it's less than ts_recent, drop it. 1522 */ 1523 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent && 1524 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1525 1526 /* Check to see if ts_recent is over 24 days old. */ 1527 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1528 /* 1529 * Invalidate ts_recent. If this segment updates 1530 * ts_recent, the age will be reset later and ts_recent 1531 * will get a valid value. If it does not, setting 1532 * ts_recent to zero will at least satisfy the 1533 * requirement that zero be placed in the timestamp 1534 * echo reply when ts_recent isn't valid. The 1535 * age isn't reset until we get a valid ts_recent 1536 * because we don't want out-of-order segments to be 1537 * dropped when ts_recent is old. 1538 */ 1539 tp->ts_recent = 0; 1540 } else { 1541 tcpstat.tcps_rcvduppack++; 1542 tcpstat.tcps_rcvdupbyte += tlen; 1543 tcpstat.tcps_pawsdrop++; 1544 goto dropafterack; 1545 } 1546 } 1547 1548 /* 1549 * T/TCP mechanism 1550 * If T/TCP was negotiated and the segment doesn't have CC, 1551 * or if its CC is wrong then drop the segment. 1552 * RST segments do not have to comply with this. 1553 */ 1554 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1555 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1556 goto dropafterack; 1557 1558 /* 1559 * In the SYN-RECEIVED state, validate that the packet belongs to 1560 * this connection before trimming the data to fit the receive 1561 * window. Check the sequence number versus IRS since we know 1562 * the sequence numbers haven't wrapped. This is a partial fix 1563 * for the "LAND" DoS attack. 1564 */ 1565 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1566 rstreason = BANDLIM_RST_OPENPORT; 1567 goto dropwithreset; 1568 } 1569 1570 todrop = tp->rcv_nxt - th->th_seq; 1571 if (todrop > 0) { 1572 if (thflags & TH_SYN) { 1573 thflags &= ~TH_SYN; 1574 th->th_seq++; 1575 if (th->th_urp > 1) 1576 th->th_urp--; 1577 else 1578 thflags &= ~TH_URG; 1579 todrop--; 1580 } 1581 /* 1582 * Following if statement from Stevens, vol. 2, p. 960. 1583 */ 1584 if (todrop > tlen 1585 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1586 /* 1587 * Any valid FIN must be to the left of the window. 1588 * At this point the FIN must be a duplicate or out 1589 * of sequence; drop it. 1590 */ 1591 thflags &= ~TH_FIN; 1592 1593 /* 1594 * Send an ACK to resynchronize and drop any data. 1595 * But keep on processing for RST or ACK. 1596 */ 1597 tp->t_flags |= TF_ACKNOW; 1598 todrop = tlen; 1599 tcpstat.tcps_rcvduppack++; 1600 tcpstat.tcps_rcvdupbyte += todrop; 1601 } else { 1602 tcpstat.tcps_rcvpartduppack++; 1603 tcpstat.tcps_rcvpartdupbyte += todrop; 1604 } 1605 drop_hdrlen += todrop; /* drop from the top afterwards */ 1606 th->th_seq += todrop; 1607 tlen -= todrop; 1608 if (th->th_urp > todrop) 1609 th->th_urp -= todrop; 1610 else { 1611 thflags &= ~TH_URG; 1612 th->th_urp = 0; 1613 } 1614 } 1615 1616 /* 1617 * If new data are received on a connection after the 1618 * user processes are gone, then RST the other end. 1619 */ 1620 if ((so->so_state & SS_NOFDREF) && 1621 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1622 tp = tcp_close(tp); 1623 tcpstat.tcps_rcvafterclose++; 1624 rstreason = BANDLIM_UNLIMITED; 1625 goto dropwithreset; 1626 } 1627 1628 /* 1629 * If segment ends after window, drop trailing data 1630 * (and PUSH and FIN); if nothing left, just ACK. 1631 */ 1632 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1633 if (todrop > 0) { 1634 tcpstat.tcps_rcvpackafterwin++; 1635 if (todrop >= tlen) { 1636 tcpstat.tcps_rcvbyteafterwin += tlen; 1637 /* 1638 * If a new connection request is received 1639 * while in TIME_WAIT, drop the old connection 1640 * and start over if the sequence numbers 1641 * are above the previous ones. 1642 */ 1643 if (thflags & TH_SYN && 1644 tp->t_state == TCPS_TIME_WAIT && 1645 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1646 iss = tcp_rndiss_next(); 1647 tp = tcp_close(tp); 1648 goto findpcb; 1649 } 1650 /* 1651 * If window is closed can only take segments at 1652 * window edge, and have to drop data and PUSH from 1653 * incoming segments. Continue processing, but 1654 * remember to ack. Otherwise, drop segment 1655 * and ack. 1656 */ 1657 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1658 tp->t_flags |= TF_ACKNOW; 1659 tcpstat.tcps_rcvwinprobe++; 1660 } else 1661 goto dropafterack; 1662 } else 1663 tcpstat.tcps_rcvbyteafterwin += todrop; 1664 m_adj(m, -todrop); 1665 tlen -= todrop; 1666 thflags &= ~(TH_PUSH|TH_FIN); 1667 } 1668 1669 /* 1670 * If last ACK falls within this segment's sequence numbers, 1671 * record its timestamp. 1672 * NOTE that the test is modified according to the latest 1673 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1674 */ 1675 if ((to.to_flag & TOF_TS) != 0 && 1676 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1677 tp->ts_recent_age = ticks; 1678 tp->ts_recent = to.to_tsval; 1679 } 1680 1681 /* 1682 * If a SYN is in the window, then this is an 1683 * error and we send an RST and drop the connection. 1684 */ 1685 if (thflags & TH_SYN) { 1686 tp = tcp_drop(tp, ECONNRESET); 1687 rstreason = BANDLIM_UNLIMITED; 1688 goto dropwithreset; 1689 } 1690 1691 /* 1692 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1693 * flag is on (half-synchronized state), then queue data for 1694 * later processing; else drop segment and return. 1695 */ 1696 if ((thflags & TH_ACK) == 0) { 1697 if (tp->t_state == TCPS_SYN_RECEIVED || 1698 (tp->t_flags & TF_NEEDSYN)) 1699 goto step6; 1700 else 1701 goto drop; 1702 } 1703 1704 /* 1705 * Ack processing. 1706 */ 1707 switch (tp->t_state) { 1708 1709 /* 1710 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1711 * ESTABLISHED state and continue processing. 1712 * The ACK was checked above. 1713 */ 1714 case TCPS_SYN_RECEIVED: 1715 1716 tcpstat.tcps_connects++; 1717 soisconnected(so); 1718 /* Do window scaling? */ 1719 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1720 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1721 tp->snd_scale = tp->requested_s_scale; 1722 tp->rcv_scale = tp->request_r_scale; 1723 } 1724 /* 1725 * Upon successful completion of 3-way handshake, 1726 * update cache.CC if it was undefined, pass any queued 1727 * data to the user, and advance state appropriately. 1728 */ 1729 if ((taop = tcp_gettaocache(inp)) != NULL && 1730 taop->tao_cc == 0) 1731 taop->tao_cc = tp->cc_recv; 1732 1733 /* 1734 * Make transitions: 1735 * SYN-RECEIVED -> ESTABLISHED 1736 * SYN-RECEIVED* -> FIN-WAIT-1 1737 */ 1738 tp->t_starttime = ticks; 1739 if (tp->t_flags & TF_NEEDFIN) { 1740 tp->t_state = TCPS_FIN_WAIT_1; 1741 tp->t_flags &= ~TF_NEEDFIN; 1742 } else { 1743 tp->t_state = TCPS_ESTABLISHED; 1744 callout_reset(tp->tt_keep, tcp_keepidle, 1745 tcp_timer_keep, tp); 1746 } 1747 /* 1748 * If segment contains data or ACK, will call tcp_reass() 1749 * later; if not, do so now to pass queued data to user. 1750 */ 1751 if (tlen == 0 && (thflags & TH_FIN) == 0) 1752 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1753 (struct mbuf *)0); 1754 tp->snd_wl1 = th->th_seq - 1; 1755 /* fall into ... */ 1756 1757 /* 1758 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1759 * ACKs. If the ack is in the range 1760 * tp->snd_una < th->th_ack <= tp->snd_max 1761 * then advance tp->snd_una to th->th_ack and drop 1762 * data from the retransmission queue. If this ACK reflects 1763 * more up to date window information we update our window information. 1764 */ 1765 case TCPS_ESTABLISHED: 1766 case TCPS_FIN_WAIT_1: 1767 case TCPS_FIN_WAIT_2: 1768 case TCPS_CLOSE_WAIT: 1769 case TCPS_CLOSING: 1770 case TCPS_LAST_ACK: 1771 case TCPS_TIME_WAIT: 1772 1773 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1774 if (tlen == 0 && tiwin == tp->snd_wnd) { 1775 tcpstat.tcps_rcvdupack++; 1776 /* 1777 * If we have outstanding data (other than 1778 * a window probe), this is a completely 1779 * duplicate ack (ie, window info didn't 1780 * change), the ack is the biggest we've 1781 * seen and we've seen exactly our rexmt 1782 * threshhold of them, assume a packet 1783 * has been dropped and retransmit it. 1784 * Kludge snd_nxt & the congestion 1785 * window so we send only this one 1786 * packet. 1787 * 1788 * We know we're losing at the current 1789 * window size so do congestion avoidance 1790 * (set ssthresh to half the current window 1791 * and pull our congestion window back to 1792 * the new ssthresh). 1793 * 1794 * Dup acks mean that packets have left the 1795 * network (they're now cached at the receiver) 1796 * so bump cwnd by the amount in the receiver 1797 * to keep a constant cwnd packets in the 1798 * network. 1799 */ 1800 if (!callout_active(tp->tt_rexmt) || 1801 th->th_ack != tp->snd_una) 1802 tp->t_dupacks = 0; 1803 else if (++tp->t_dupacks == tcprexmtthresh) { 1804 tcp_seq onxt = tp->snd_nxt; 1805 u_int win = 1806 min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1807 tp->t_maxseg; 1808 if (tcp_do_newreno && SEQ_LT(th->th_ack, 1809 tp->snd_recover)) { 1810 /* False retransmit, should not 1811 * cut window 1812 */ 1813 tp->snd_cwnd += tp->t_maxseg; 1814 tp->t_dupacks = 0; 1815 (void) tcp_output(tp); 1816 goto drop; 1817 } 1818 if (win < 2) 1819 win = 2; 1820 tp->snd_ssthresh = win * tp->t_maxseg; 1821 tp->snd_recover = tp->snd_max; 1822 callout_stop(tp->tt_rexmt); 1823 tp->t_rtttime = 0; 1824 tp->snd_nxt = th->th_ack; 1825 tp->snd_cwnd = tp->t_maxseg; 1826 (void) tcp_output(tp); 1827 tp->snd_cwnd = tp->snd_ssthresh + 1828 tp->t_maxseg * tp->t_dupacks; 1829 if (SEQ_GT(onxt, tp->snd_nxt)) 1830 tp->snd_nxt = onxt; 1831 goto drop; 1832 } else if (tp->t_dupacks > tcprexmtthresh) { 1833 tp->snd_cwnd += tp->t_maxseg; 1834 (void) tcp_output(tp); 1835 goto drop; 1836 } 1837 } else 1838 tp->t_dupacks = 0; 1839 break; 1840 } 1841 /* 1842 * If the congestion window was inflated to account 1843 * for the other side's cached packets, retract it. 1844 */ 1845 if (tcp_do_newreno == 0) { 1846 if (tp->t_dupacks >= tcprexmtthresh && 1847 tp->snd_cwnd > tp->snd_ssthresh) 1848 tp->snd_cwnd = tp->snd_ssthresh; 1849 tp->t_dupacks = 0; 1850 } else if (tp->t_dupacks >= tcprexmtthresh && 1851 !tcp_newreno(tp, th)) { 1852 /* 1853 * Window inflation should have left us with approx. 1854 * snd_ssthresh outstanding data. But in case we 1855 * would be inclined to send a burst, better to do 1856 * it via the slow start mechanism. 1857 */ 1858 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, tp->snd_max)) 1859 tp->snd_cwnd = 1860 tp->snd_max - th->th_ack + tp->t_maxseg; 1861 else 1862 tp->snd_cwnd = tp->snd_ssthresh; 1863 tp->t_dupacks = 0; 1864 } 1865 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1866 tcpstat.tcps_rcvacktoomuch++; 1867 goto dropafterack; 1868 } 1869 /* 1870 * If we reach this point, ACK is not a duplicate, 1871 * i.e., it ACKs something we sent. 1872 */ 1873 if (tp->t_flags & TF_NEEDSYN) { 1874 /* 1875 * T/TCP: Connection was half-synchronized, and our 1876 * SYN has been ACK'd (so connection is now fully 1877 * synchronized). Go to non-starred state, 1878 * increment snd_una for ACK of SYN, and check if 1879 * we can do window scaling. 1880 */ 1881 tp->t_flags &= ~TF_NEEDSYN; 1882 tp->snd_una++; 1883 /* Do window scaling? */ 1884 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1885 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1886 tp->snd_scale = tp->requested_s_scale; 1887 tp->rcv_scale = tp->request_r_scale; 1888 } 1889 } 1890 1891 process_ACK: 1892 acked = th->th_ack - tp->snd_una; 1893 tcpstat.tcps_rcvackpack++; 1894 tcpstat.tcps_rcvackbyte += acked; 1895 1896 /* 1897 * If we just performed our first retransmit, and the ACK 1898 * arrives within our recovery window, then it was a mistake 1899 * to do the retransmit in the first place. Recover our 1900 * original cwnd and ssthresh, and proceed to transmit where 1901 * we left off. 1902 */ 1903 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1904 tp->snd_cwnd = tp->snd_cwnd_prev; 1905 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1906 tp->snd_nxt = tp->snd_max; 1907 tp->t_badrxtwin = 0; /* XXX probably not required */ 1908 } 1909 1910 /* 1911 * If we have a timestamp reply, update smoothed 1912 * round trip time. If no timestamp is present but 1913 * transmit timer is running and timed sequence 1914 * number was acked, update smoothed round trip time. 1915 * Since we now have an rtt measurement, cancel the 1916 * timer backoff (cf., Phil Karn's retransmit alg.). 1917 * Recompute the initial retransmit timer. 1918 */ 1919 if (to.to_flag & TOF_TS) 1920 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1921 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1922 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1923 1924 /* 1925 * If all outstanding data is acked, stop retransmit 1926 * timer and remember to restart (more output or persist). 1927 * If there is more data to be acked, restart retransmit 1928 * timer, using current (possibly backed-off) value. 1929 */ 1930 if (th->th_ack == tp->snd_max) { 1931 callout_stop(tp->tt_rexmt); 1932 needoutput = 1; 1933 } else if (!callout_active(tp->tt_persist)) 1934 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1935 tcp_timer_rexmt, tp); 1936 1937 /* 1938 * If no data (only SYN) was ACK'd, 1939 * skip rest of ACK processing. 1940 */ 1941 if (acked == 0) 1942 goto step6; 1943 1944 /* 1945 * When new data is acked, open the congestion window. 1946 * If the window gives us less than ssthresh packets 1947 * in flight, open exponentially (maxseg per packet). 1948 * Otherwise open linearly: maxseg per window 1949 * (maxseg^2 / cwnd per packet). 1950 */ 1951 { 1952 register u_int cw = tp->snd_cwnd; 1953 register u_int incr = tp->t_maxseg; 1954 1955 if (cw > tp->snd_ssthresh) 1956 incr = incr * incr / cw; 1957 /* 1958 * If t_dupacks != 0 here, it indicates that we are still 1959 * in NewReno fast recovery mode, so we leave the congestion 1960 * window alone. 1961 */ 1962 if (tcp_do_newreno == 0 || tp->t_dupacks == 0) 1963 tp->snd_cwnd = min(cw + incr,TCP_MAXWIN<<tp->snd_scale); 1964 } 1965 if (acked > so->so_snd.sb_cc) { 1966 tp->snd_wnd -= so->so_snd.sb_cc; 1967 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1968 ourfinisacked = 1; 1969 } else { 1970 sbdrop(&so->so_snd, acked); 1971 tp->snd_wnd -= acked; 1972 ourfinisacked = 0; 1973 } 1974 sowwakeup(so); 1975 tp->snd_una = th->th_ack; 1976 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1977 tp->snd_nxt = tp->snd_una; 1978 1979 switch (tp->t_state) { 1980 1981 /* 1982 * In FIN_WAIT_1 STATE in addition to the processing 1983 * for the ESTABLISHED state if our FIN is now acknowledged 1984 * then enter FIN_WAIT_2. 1985 */ 1986 case TCPS_FIN_WAIT_1: 1987 if (ourfinisacked) { 1988 /* 1989 * If we can't receive any more 1990 * data, then closing user can proceed. 1991 * Starting the timer is contrary to the 1992 * specification, but if we don't get a FIN 1993 * we'll hang forever. 1994 */ 1995 if (so->so_state & SS_CANTRCVMORE) { 1996 soisdisconnected(so); 1997 callout_reset(tp->tt_2msl, tcp_maxidle, 1998 tcp_timer_2msl, tp); 1999 } 2000 tp->t_state = TCPS_FIN_WAIT_2; 2001 } 2002 break; 2003 2004 /* 2005 * In CLOSING STATE in addition to the processing for 2006 * the ESTABLISHED state if the ACK acknowledges our FIN 2007 * then enter the TIME-WAIT state, otherwise ignore 2008 * the segment. 2009 */ 2010 case TCPS_CLOSING: 2011 if (ourfinisacked) { 2012 tp->t_state = TCPS_TIME_WAIT; 2013 tcp_canceltimers(tp); 2014 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2015 if (tp->cc_recv != 0 && 2016 (ticks - tp->t_starttime) < tcp_msl) 2017 callout_reset(tp->tt_2msl, 2018 tp->t_rxtcur * 2019 TCPTV_TWTRUNC, 2020 tcp_timer_2msl, tp); 2021 else 2022 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2023 tcp_timer_2msl, tp); 2024 soisdisconnected(so); 2025 } 2026 break; 2027 2028 /* 2029 * In LAST_ACK, we may still be waiting for data to drain 2030 * and/or to be acked, as well as for the ack of our FIN. 2031 * If our FIN is now acknowledged, delete the TCB, 2032 * enter the closed state and return. 2033 */ 2034 case TCPS_LAST_ACK: 2035 if (ourfinisacked) { 2036 tp = tcp_close(tp); 2037 goto drop; 2038 } 2039 break; 2040 2041 /* 2042 * In TIME_WAIT state the only thing that should arrive 2043 * is a retransmission of the remote FIN. Acknowledge 2044 * it and restart the finack timer. 2045 */ 2046 case TCPS_TIME_WAIT: 2047 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2048 tcp_timer_2msl, tp); 2049 goto dropafterack; 2050 } 2051 } 2052 2053 step6: 2054 /* 2055 * Update window information. 2056 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2057 */ 2058 if ((thflags & TH_ACK) && 2059 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2060 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2061 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2062 /* keep track of pure window updates */ 2063 if (tlen == 0 && 2064 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2065 tcpstat.tcps_rcvwinupd++; 2066 tp->snd_wnd = tiwin; 2067 tp->snd_wl1 = th->th_seq; 2068 tp->snd_wl2 = th->th_ack; 2069 if (tp->snd_wnd > tp->max_sndwnd) 2070 tp->max_sndwnd = tp->snd_wnd; 2071 needoutput = 1; 2072 } 2073 2074 /* 2075 * Process segments with URG. 2076 */ 2077 if ((thflags & TH_URG) && th->th_urp && 2078 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2079 /* 2080 * This is a kludge, but if we receive and accept 2081 * random urgent pointers, we'll crash in 2082 * soreceive. It's hard to imagine someone 2083 * actually wanting to send this much urgent data. 2084 */ 2085 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2086 th->th_urp = 0; /* XXX */ 2087 thflags &= ~TH_URG; /* XXX */ 2088 goto dodata; /* XXX */ 2089 } 2090 /* 2091 * If this segment advances the known urgent pointer, 2092 * then mark the data stream. This should not happen 2093 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2094 * a FIN has been received from the remote side. 2095 * In these states we ignore the URG. 2096 * 2097 * According to RFC961 (Assigned Protocols), 2098 * the urgent pointer points to the last octet 2099 * of urgent data. We continue, however, 2100 * to consider it to indicate the first octet 2101 * of data past the urgent section as the original 2102 * spec states (in one of two places). 2103 */ 2104 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2105 tp->rcv_up = th->th_seq + th->th_urp; 2106 so->so_oobmark = so->so_rcv.sb_cc + 2107 (tp->rcv_up - tp->rcv_nxt) - 1; 2108 if (so->so_oobmark == 0) 2109 so->so_state |= SS_RCVATMARK; 2110 sohasoutofband(so); 2111 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2112 } 2113 /* 2114 * Remove out of band data so doesn't get presented to user. 2115 * This can happen independent of advancing the URG pointer, 2116 * but if two URG's are pending at once, some out-of-band 2117 * data may creep in... ick. 2118 */ 2119 if (th->th_urp <= (u_long)tlen 2120 #ifdef SO_OOBINLINE 2121 && (so->so_options & SO_OOBINLINE) == 0 2122 #endif 2123 ) 2124 tcp_pulloutofband(so, th, m, 2125 drop_hdrlen); /* hdr drop is delayed */ 2126 } else 2127 /* 2128 * If no out of band data is expected, 2129 * pull receive urgent pointer along 2130 * with the receive window. 2131 */ 2132 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2133 tp->rcv_up = tp->rcv_nxt; 2134 dodata: /* XXX */ 2135 2136 /* 2137 * Process the segment text, merging it into the TCP sequencing queue, 2138 * and arranging for acknowledgment of receipt if necessary. 2139 * This process logically involves adjusting tp->rcv_wnd as data 2140 * is presented to the user (this happens in tcp_usrreq.c, 2141 * case PRU_RCVD). If a FIN has already been received on this 2142 * connection then we just ignore the text. 2143 */ 2144 if ((tlen || (thflags&TH_FIN)) && 2145 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2146 m_adj(m, drop_hdrlen); /* delayed header drop */ 2147 TCP_REASS(tp, th, &tlen, m, so, thflags); 2148 /* 2149 * Note the amount of data that peer has sent into 2150 * our window, in order to estimate the sender's 2151 * buffer size. 2152 */ 2153 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2154 } else { 2155 m_freem(m); 2156 thflags &= ~TH_FIN; 2157 } 2158 2159 /* 2160 * If FIN is received ACK the FIN and let the user know 2161 * that the connection is closing. 2162 */ 2163 if (thflags & TH_FIN) { 2164 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2165 socantrcvmore(so); 2166 /* 2167 * If connection is half-synchronized 2168 * (ie NEEDSYN flag on) then delay ACK, 2169 * so it may be piggybacked when SYN is sent. 2170 * Otherwise, since we received a FIN then no 2171 * more input can be expected, send ACK now. 2172 */ 2173 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2174 callout_reset(tp->tt_delack, tcp_delacktime, 2175 tcp_timer_delack, tp); 2176 else 2177 tp->t_flags |= TF_ACKNOW; 2178 tp->rcv_nxt++; 2179 } 2180 switch (tp->t_state) { 2181 2182 /* 2183 * In SYN_RECEIVED and ESTABLISHED STATES 2184 * enter the CLOSE_WAIT state. 2185 */ 2186 case TCPS_SYN_RECEIVED: 2187 tp->t_starttime = ticks; 2188 /*FALLTHROUGH*/ 2189 case TCPS_ESTABLISHED: 2190 tp->t_state = TCPS_CLOSE_WAIT; 2191 break; 2192 2193 /* 2194 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2195 * enter the CLOSING state. 2196 */ 2197 case TCPS_FIN_WAIT_1: 2198 tp->t_state = TCPS_CLOSING; 2199 break; 2200 2201 /* 2202 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2203 * starting the time-wait timer, turning off the other 2204 * standard timers. 2205 */ 2206 case TCPS_FIN_WAIT_2: 2207 tp->t_state = TCPS_TIME_WAIT; 2208 tcp_canceltimers(tp); 2209 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2210 if (tp->cc_recv != 0 && 2211 (ticks - tp->t_starttime) < tcp_msl) { 2212 callout_reset(tp->tt_2msl, 2213 tp->t_rxtcur * TCPTV_TWTRUNC, 2214 tcp_timer_2msl, tp); 2215 /* For transaction client, force ACK now. */ 2216 tp->t_flags |= TF_ACKNOW; 2217 } 2218 else 2219 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2220 tcp_timer_2msl, tp); 2221 soisdisconnected(so); 2222 break; 2223 2224 /* 2225 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2226 */ 2227 case TCPS_TIME_WAIT: 2228 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2229 tcp_timer_2msl, tp); 2230 break; 2231 } 2232 } 2233 #ifdef TCPDEBUG 2234 if (so->so_options & SO_DEBUG) 2235 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2236 &tcp_savetcp, 0); 2237 #endif 2238 2239 /* 2240 * Return any desired output. 2241 */ 2242 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2243 (void) tcp_output(tp); 2244 return; 2245 2246 dropafterack: 2247 /* 2248 * Generate an ACK dropping incoming segment if it occupies 2249 * sequence space, where the ACK reflects our state. 2250 * 2251 * We can now skip the test for the RST flag since all 2252 * paths to this code happen after packets containing 2253 * RST have been dropped. 2254 * 2255 * In the SYN-RECEIVED state, don't send an ACK unless the 2256 * segment we received passes the SYN-RECEIVED ACK test. 2257 * If it fails send a RST. This breaks the loop in the 2258 * "LAND" DoS attack, and also prevents an ACK storm 2259 * between two listening ports that have been sent forged 2260 * SYN segments, each with the source address of the other. 2261 */ 2262 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2263 (SEQ_GT(tp->snd_una, th->th_ack) || 2264 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2265 rstreason = BANDLIM_RST_OPENPORT; 2266 goto dropwithreset; 2267 } 2268 #ifdef TCPDEBUG 2269 if (so->so_options & SO_DEBUG) 2270 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2271 &tcp_savetcp, 0); 2272 #endif 2273 m_freem(m); 2274 tp->t_flags |= TF_ACKNOW; 2275 (void) tcp_output(tp); 2276 return; 2277 2278 dropwithreset: 2279 /* 2280 * Generate a RST, dropping incoming segment. 2281 * Make ACK acceptable to originator of segment. 2282 * Don't bother to respond if destination was broadcast/multicast. 2283 */ 2284 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2285 goto drop; 2286 #ifdef INET6 2287 if (isipv6) { 2288 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2289 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2290 goto drop; 2291 } else 2292 #endif /* INET6 */ 2293 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2294 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2295 ip->ip_src.s_addr == htonl(INADDR_BROADCAST)) 2296 goto drop; 2297 /* IPv6 anycast check is done at tcp6_input() */ 2298 2299 /* 2300 * Perform bandwidth limiting. 2301 */ 2302 if (badport_bandlim(rstreason) < 0) 2303 goto drop; 2304 2305 #ifdef TCPDEBUG 2306 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2307 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2308 &tcp_savetcp, 0); 2309 #endif 2310 if (thflags & TH_ACK) 2311 /* mtod() below is safe as long as hdr dropping is delayed */ 2312 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2313 TH_RST); 2314 else { 2315 if (thflags & TH_SYN) 2316 tlen++; 2317 /* mtod() below is safe as long as hdr dropping is delayed */ 2318 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2319 (tcp_seq)0, TH_RST|TH_ACK); 2320 } 2321 /* destroy temporarily created socket */ 2322 if (dropsocket) 2323 (void) soabort(so); 2324 return; 2325 2326 drop: 2327 /* 2328 * Drop space held by incoming segment and return. 2329 */ 2330 #ifdef TCPDEBUG 2331 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2332 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2333 &tcp_savetcp, 0); 2334 #endif 2335 m_freem(m); 2336 /* destroy temporarily created socket */ 2337 if (dropsocket) 2338 (void) soabort(so); 2339 return; 2340 } 2341 2342 static void 2343 tcp_dooptions(tp, cp, cnt, th, to) 2344 struct tcpcb *tp; 2345 u_char *cp; 2346 int cnt; 2347 struct tcphdr *th; 2348 struct tcpopt *to; 2349 { 2350 u_short mss = 0; 2351 int opt, optlen; 2352 2353 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2354 opt = cp[0]; 2355 if (opt == TCPOPT_EOL) 2356 break; 2357 if (opt == TCPOPT_NOP) 2358 optlen = 1; 2359 else { 2360 if (cnt < 2) 2361 break; 2362 optlen = cp[1]; 2363 if (optlen < 2 || optlen > cnt) 2364 break; 2365 } 2366 switch (opt) { 2367 2368 default: 2369 continue; 2370 2371 case TCPOPT_MAXSEG: 2372 if (optlen != TCPOLEN_MAXSEG) 2373 continue; 2374 if (!(th->th_flags & TH_SYN)) 2375 continue; 2376 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 2377 NTOHS(mss); 2378 break; 2379 2380 case TCPOPT_WINDOW: 2381 if (optlen != TCPOLEN_WINDOW) 2382 continue; 2383 if (!(th->th_flags & TH_SYN)) 2384 continue; 2385 tp->t_flags |= TF_RCVD_SCALE; 2386 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2387 break; 2388 2389 case TCPOPT_TIMESTAMP: 2390 if (optlen != TCPOLEN_TIMESTAMP) 2391 continue; 2392 to->to_flag |= TOF_TS; 2393 bcopy((char *)cp + 2, 2394 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2395 NTOHL(to->to_tsval); 2396 bcopy((char *)cp + 6, 2397 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2398 NTOHL(to->to_tsecr); 2399 2400 /* 2401 * A timestamp received in a SYN makes 2402 * it ok to send timestamp requests and replies. 2403 */ 2404 if (th->th_flags & TH_SYN) { 2405 tp->t_flags |= TF_RCVD_TSTMP; 2406 tp->ts_recent = to->to_tsval; 2407 tp->ts_recent_age = ticks; 2408 } 2409 break; 2410 case TCPOPT_CC: 2411 if (optlen != TCPOLEN_CC) 2412 continue; 2413 to->to_flag |= TOF_CC; 2414 bcopy((char *)cp + 2, 2415 (char *)&to->to_cc, sizeof(to->to_cc)); 2416 NTOHL(to->to_cc); 2417 /* 2418 * A CC or CC.new option received in a SYN makes 2419 * it ok to send CC in subsequent segments. 2420 */ 2421 if (th->th_flags & TH_SYN) 2422 tp->t_flags |= TF_RCVD_CC; 2423 break; 2424 case TCPOPT_CCNEW: 2425 if (optlen != TCPOLEN_CC) 2426 continue; 2427 if (!(th->th_flags & TH_SYN)) 2428 continue; 2429 to->to_flag |= TOF_CCNEW; 2430 bcopy((char *)cp + 2, 2431 (char *)&to->to_cc, sizeof(to->to_cc)); 2432 NTOHL(to->to_cc); 2433 /* 2434 * A CC or CC.new option received in a SYN makes 2435 * it ok to send CC in subsequent segments. 2436 */ 2437 tp->t_flags |= TF_RCVD_CC; 2438 break; 2439 case TCPOPT_CCECHO: 2440 if (optlen != TCPOLEN_CC) 2441 continue; 2442 if (!(th->th_flags & TH_SYN)) 2443 continue; 2444 to->to_flag |= TOF_CCECHO; 2445 bcopy((char *)cp + 2, 2446 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2447 NTOHL(to->to_ccecho); 2448 break; 2449 } 2450 } 2451 if (th->th_flags & TH_SYN) 2452 tcp_mss(tp, mss); /* sets t_maxseg */ 2453 } 2454 2455 /* 2456 * Pull out of band byte out of a segment so 2457 * it doesn't appear in the user's data queue. 2458 * It is still reflected in the segment length for 2459 * sequencing purposes. 2460 */ 2461 static void 2462 tcp_pulloutofband(so, th, m, off) 2463 struct socket *so; 2464 struct tcphdr *th; 2465 register struct mbuf *m; 2466 int off; /* delayed to be droped hdrlen */ 2467 { 2468 int cnt = off + th->th_urp - 1; 2469 2470 while (cnt >= 0) { 2471 if (m->m_len > cnt) { 2472 char *cp = mtod(m, caddr_t) + cnt; 2473 struct tcpcb *tp = sototcpcb(so); 2474 2475 tp->t_iobc = *cp; 2476 tp->t_oobflags |= TCPOOB_HAVEDATA; 2477 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2478 m->m_len--; 2479 if (m->m_flags & M_PKTHDR) 2480 m->m_pkthdr.len--; 2481 return; 2482 } 2483 cnt -= m->m_len; 2484 m = m->m_next; 2485 if (m == 0) 2486 break; 2487 } 2488 panic("tcp_pulloutofband"); 2489 } 2490 2491 /* 2492 * Collect new round-trip time estimate 2493 * and update averages and current timeout. 2494 */ 2495 static void 2496 tcp_xmit_timer(tp, rtt) 2497 register struct tcpcb *tp; 2498 int rtt; 2499 { 2500 register int delta; 2501 2502 tcpstat.tcps_rttupdated++; 2503 tp->t_rttupdated++; 2504 if (tp->t_srtt != 0) { 2505 /* 2506 * srtt is stored as fixed point with 5 bits after the 2507 * binary point (i.e., scaled by 8). The following magic 2508 * is equivalent to the smoothing algorithm in rfc793 with 2509 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2510 * point). Adjust rtt to origin 0. 2511 */ 2512 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2513 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2514 2515 if ((tp->t_srtt += delta) <= 0) 2516 tp->t_srtt = 1; 2517 2518 /* 2519 * We accumulate a smoothed rtt variance (actually, a 2520 * smoothed mean difference), then set the retransmit 2521 * timer to smoothed rtt + 4 times the smoothed variance. 2522 * rttvar is stored as fixed point with 4 bits after the 2523 * binary point (scaled by 16). The following is 2524 * equivalent to rfc793 smoothing with an alpha of .75 2525 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2526 * rfc793's wired-in beta. 2527 */ 2528 if (delta < 0) 2529 delta = -delta; 2530 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2531 if ((tp->t_rttvar += delta) <= 0) 2532 tp->t_rttvar = 1; 2533 } else { 2534 /* 2535 * No rtt measurement yet - use the unsmoothed rtt. 2536 * Set the variance to half the rtt (so our first 2537 * retransmit happens at 3*rtt). 2538 */ 2539 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2540 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2541 } 2542 tp->t_rtttime = 0; 2543 tp->t_rxtshift = 0; 2544 2545 /* 2546 * the retransmit should happen at rtt + 4 * rttvar. 2547 * Because of the way we do the smoothing, srtt and rttvar 2548 * will each average +1/2 tick of bias. When we compute 2549 * the retransmit timer, we want 1/2 tick of rounding and 2550 * 1 extra tick because of +-1/2 tick uncertainty in the 2551 * firing of the timer. The bias will give us exactly the 2552 * 1.5 tick we need. But, because the bias is 2553 * statistical, we have to test that we don't drop below 2554 * the minimum feasible timer (which is 2 ticks). 2555 */ 2556 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2557 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2558 2559 /* 2560 * We received an ack for a packet that wasn't retransmitted; 2561 * it is probably safe to discard any error indications we've 2562 * received recently. This isn't quite right, but close enough 2563 * for now (a route might have failed after we sent a segment, 2564 * and the return path might not be symmetrical). 2565 */ 2566 tp->t_softerror = 0; 2567 } 2568 2569 /* 2570 * Determine a reasonable value for maxseg size. 2571 * If the route is known, check route for mtu. 2572 * If none, use an mss that can be handled on the outgoing 2573 * interface without forcing IP to fragment; if bigger than 2574 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2575 * to utilize large mbufs. If no route is found, route has no mtu, 2576 * or the destination isn't local, use a default, hopefully conservative 2577 * size (usually 512 or the default IP max size, but no more than the mtu 2578 * of the interface), as we can't discover anything about intervening 2579 * gateways or networks. We also initialize the congestion/slow start 2580 * window to be a single segment if the destination isn't local. 2581 * While looking at the routing entry, we also initialize other path-dependent 2582 * parameters from pre-set or cached values in the routing entry. 2583 * 2584 * Also take into account the space needed for options that we 2585 * send regularly. Make maxseg shorter by that amount to assure 2586 * that we can send maxseg amount of data even when the options 2587 * are present. Store the upper limit of the length of options plus 2588 * data in maxopd. 2589 * 2590 * NOTE that this routine is only called when we process an incoming 2591 * segment, for outgoing segments only tcp_mssopt is called. 2592 * 2593 * In case of T/TCP, we call this routine during implicit connection 2594 * setup as well (offer = -1), to initialize maxseg from the cached 2595 * MSS of our peer. 2596 */ 2597 void 2598 tcp_mss(tp, offer) 2599 struct tcpcb *tp; 2600 int offer; 2601 { 2602 register struct rtentry *rt; 2603 struct ifnet *ifp; 2604 register int rtt, mss; 2605 u_long bufsize; 2606 struct inpcb *inp; 2607 struct socket *so; 2608 struct rmxp_tao *taop; 2609 int origoffer = offer; 2610 #ifdef INET6 2611 int isipv6; 2612 int min_protoh; 2613 #endif 2614 2615 inp = tp->t_inpcb; 2616 #ifdef INET6 2617 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2618 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2619 : sizeof (struct tcpiphdr); 2620 #else 2621 #define min_protoh (sizeof (struct tcpiphdr)) 2622 #endif 2623 #ifdef INET6 2624 if (isipv6) 2625 rt = tcp_rtlookup6(inp); 2626 else 2627 #endif 2628 rt = tcp_rtlookup(inp); 2629 if (rt == NULL) { 2630 tp->t_maxopd = tp->t_maxseg = 2631 #ifdef INET6 2632 isipv6 ? tcp_v6mssdflt : 2633 #endif /* INET6 */ 2634 tcp_mssdflt; 2635 return; 2636 } 2637 ifp = rt->rt_ifp; 2638 so = inp->inp_socket; 2639 2640 taop = rmx_taop(rt->rt_rmx); 2641 /* 2642 * Offer == -1 means that we didn't receive SYN yet, 2643 * use cached value in that case; 2644 */ 2645 if (offer == -1) 2646 offer = taop->tao_mssopt; 2647 /* 2648 * Offer == 0 means that there was no MSS on the SYN segment, 2649 * in this case we use tcp_mssdflt. 2650 */ 2651 if (offer == 0) 2652 offer = 2653 #ifdef INET6 2654 isipv6 ? tcp_v6mssdflt : 2655 #endif /* INET6 */ 2656 tcp_mssdflt; 2657 else 2658 /* 2659 * Sanity check: make sure that maxopd will be large 2660 * enough to allow some data on segments even is the 2661 * all the option space is used (40bytes). Otherwise 2662 * funny things may happen in tcp_output. 2663 */ 2664 offer = max(offer, 64); 2665 taop->tao_mssopt = offer; 2666 2667 /* 2668 * While we're here, check if there's an initial rtt 2669 * or rttvar. Convert from the route-table units 2670 * to scaled multiples of the slow timeout timer. 2671 */ 2672 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2673 /* 2674 * XXX the lock bit for RTT indicates that the value 2675 * is also a minimum value; this is subject to time. 2676 */ 2677 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2678 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2679 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2680 tcpstat.tcps_usedrtt++; 2681 if (rt->rt_rmx.rmx_rttvar) { 2682 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2683 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2684 tcpstat.tcps_usedrttvar++; 2685 } else { 2686 /* default variation is +- 1 rtt */ 2687 tp->t_rttvar = 2688 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2689 } 2690 TCPT_RANGESET(tp->t_rxtcur, 2691 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2692 tp->t_rttmin, TCPTV_REXMTMAX); 2693 } 2694 /* 2695 * if there's an mtu associated with the route, use it 2696 * else, use the link mtu. 2697 */ 2698 if (rt->rt_rmx.rmx_mtu) 2699 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2700 else 2701 { 2702 mss = 2703 #ifdef INET6 2704 (isipv6 ? nd_ifinfo[rt->rt_ifp->if_index].linkmtu : 2705 #endif 2706 ifp->if_mtu 2707 #ifdef INET6 2708 ) 2709 #endif 2710 - min_protoh; 2711 #ifdef INET6 2712 if (isipv6) { 2713 if (!in6_localaddr(&inp->in6p_faddr)) 2714 mss = min(mss, tcp_v6mssdflt); 2715 } else 2716 #endif 2717 if (!in_localaddr(inp->inp_faddr)) 2718 mss = min(mss, tcp_mssdflt); 2719 } 2720 mss = min(mss, offer); 2721 /* 2722 * maxopd stores the maximum length of data AND options 2723 * in a segment; maxseg is the amount of data in a normal 2724 * segment. We need to store this value (maxopd) apart 2725 * from maxseg, because now every segment carries options 2726 * and thus we normally have somewhat less data in segments. 2727 */ 2728 tp->t_maxopd = mss; 2729 2730 /* 2731 * In case of T/TCP, origoffer==-1 indicates, that no segments 2732 * were received yet. In this case we just guess, otherwise 2733 * we do the same as before T/TCP. 2734 */ 2735 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2736 (origoffer == -1 || 2737 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2738 mss -= TCPOLEN_TSTAMP_APPA; 2739 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 2740 (origoffer == -1 || 2741 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2742 mss -= TCPOLEN_CC_APPA; 2743 2744 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2745 if (mss > MCLBYTES) 2746 mss &= ~(MCLBYTES-1); 2747 #else 2748 if (mss > MCLBYTES) 2749 mss = mss / MCLBYTES * MCLBYTES; 2750 #endif 2751 /* 2752 * If there's a pipesize, change the socket buffer 2753 * to that size. Make the socket buffers an integral 2754 * number of mss units; if the mss is larger than 2755 * the socket buffer, decrease the mss. 2756 */ 2757 #ifdef RTV_SPIPE 2758 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2759 #endif 2760 bufsize = so->so_snd.sb_hiwat; 2761 if (bufsize < mss) 2762 mss = bufsize; 2763 else { 2764 bufsize = roundup(bufsize, mss); 2765 if (bufsize > sb_max) 2766 bufsize = sb_max; 2767 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2768 } 2769 tp->t_maxseg = mss; 2770 2771 #ifdef RTV_RPIPE 2772 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2773 #endif 2774 bufsize = so->so_rcv.sb_hiwat; 2775 if (bufsize > mss) { 2776 bufsize = roundup(bufsize, mss); 2777 if (bufsize > sb_max) 2778 bufsize = sb_max; 2779 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2780 } 2781 2782 /* 2783 * Set the slow-start flight size depending on whether this 2784 * is a local network or not. 2785 */ 2786 if ( 2787 #ifdef INET6 2788 (isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2789 (!isipv6 && 2790 #endif 2791 in_localaddr(inp->inp_faddr) 2792 #ifdef INET6 2793 ) 2794 #endif 2795 ) 2796 tp->snd_cwnd = mss * ss_fltsz_local; 2797 else 2798 tp->snd_cwnd = mss * ss_fltsz; 2799 2800 if (rt->rt_rmx.rmx_ssthresh) { 2801 /* 2802 * There's some sort of gateway or interface 2803 * buffer limit on the path. Use this to set 2804 * the slow start threshhold, but set the 2805 * threshold to no less than 2*mss. 2806 */ 2807 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2808 tcpstat.tcps_usedssthresh++; 2809 } 2810 } 2811 2812 /* 2813 * Determine the MSS option to send on an outgoing SYN. 2814 */ 2815 int 2816 tcp_mssopt(tp) 2817 struct tcpcb *tp; 2818 { 2819 struct rtentry *rt; 2820 #ifdef INET6 2821 int isipv6; 2822 int min_protoh; 2823 #endif 2824 2825 #ifdef INET6 2826 isipv6 = ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2827 min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) 2828 : sizeof (struct tcpiphdr); 2829 #else 2830 #define min_protoh (sizeof (struct tcpiphdr)) 2831 #endif 2832 #ifdef INET6 2833 if (isipv6) 2834 rt = tcp_rtlookup6(tp->t_inpcb); 2835 else 2836 #endif /* INET6 */ 2837 rt = tcp_rtlookup(tp->t_inpcb); 2838 if (rt == NULL) 2839 return 2840 #ifdef INET6 2841 isipv6 ? tcp_v6mssdflt : 2842 #endif /* INET6 */ 2843 tcp_mssdflt; 2844 2845 return rt->rt_ifp->if_mtu - min_protoh; 2846 } 2847 2848 2849 /* 2850 * Checks for partial ack. If partial ack arrives, force the retransmission 2851 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 2852 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 2853 * be started again. If the ack advances at least to tp->snd_recover, return 0. 2854 */ 2855 static int 2856 tcp_newreno(tp, th) 2857 struct tcpcb *tp; 2858 struct tcphdr *th; 2859 { 2860 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2861 tcp_seq onxt = tp->snd_nxt; 2862 u_long ocwnd = tp->snd_cwnd; 2863 2864 callout_stop(tp->tt_rexmt); 2865 tp->t_rtttime = 0; 2866 tp->snd_nxt = th->th_ack; 2867 /* 2868 * Set snd_cwnd to one segment beyond acknowledged offset 2869 * (tp->snd_una has not yet been updated when this function 2870 * is called) 2871 */ 2872 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2873 (void) tcp_output(tp); 2874 tp->snd_cwnd = ocwnd; 2875 if (SEQ_GT(onxt, tp->snd_nxt)) 2876 tp->snd_nxt = onxt; 2877 /* 2878 * Partial window deflation. Relies on fact that tp->snd_una 2879 * not updated yet. 2880 */ 2881 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2882 return (1); 2883 } 2884 return (0); 2885 } 2886