1 /* $OpenBSD: tcp_input.c,v 1.104 2002/01/24 22:42:48 provos Exp $ */ 2 /* $NetBSD: tcp_input.c,v 1.23 1996/02/13 23:43:44 christos Exp $ */ 3 4 /* 5 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994 6 * The Regents of the University of California. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 37 * 38 * NRL grants permission for redistribution and use in source and binary 39 * forms, with or without modification, of the software and documentation 40 * created at NRL provided that the following conditions are met: 41 * 42 * 1. Redistributions of source code must retain the above copyright 43 * notice, this list of conditions and the following disclaimer. 44 * 2. Redistributions in binary form must reproduce the above copyright 45 * notice, this list of conditions and the following disclaimer in the 46 * documentation and/or other materials provided with the distribution. 47 * 3. All advertising materials mentioning features or use of this software 48 * must display the following acknowledgements: 49 * This product includes software developed by the University of 50 * California, Berkeley and its contributors. 51 * This product includes software developed at the Information 52 * Technology Division, US Naval Research Laboratory. 53 * 4. Neither the name of the NRL nor the names of its contributors 54 * may be used to endorse or promote products derived from this software 55 * without specific prior written permission. 56 * 57 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 58 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 59 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 60 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 61 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 62 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 63 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 64 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 65 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 66 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 67 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 68 * 69 * The views and conclusions contained in the software and documentation 70 * are those of the authors and should not be interpreted as representing 71 * official policies, either expressed or implied, of the US Naval 72 * Research Laboratory (NRL). 73 */ 74 75 #ifndef TUBA_INCLUDE 76 #include <sys/param.h> 77 #include <sys/systm.h> 78 #include <sys/mbuf.h> 79 #include <sys/protosw.h> 80 #include <sys/socket.h> 81 #include <sys/socketvar.h> 82 83 #include <net/if.h> 84 #include <net/route.h> 85 86 #include <netinet/in.h> 87 #include <netinet/in_systm.h> 88 #include <netinet/ip.h> 89 #include <netinet/in_pcb.h> 90 #include <netinet/ip_var.h> 91 #include <netinet/tcp.h> 92 #include <netinet/tcp_fsm.h> 93 #include <netinet/tcp_seq.h> 94 #include <netinet/tcp_timer.h> 95 #include <netinet/tcp_var.h> 96 #include <netinet/tcpip.h> 97 #include <netinet/tcp_debug.h> 98 99 #ifdef INET6 100 #include <netinet6/in6_var.h> 101 #include <netinet6/nd6.h> 102 103 struct tcpiphdr tcp_saveti; 104 struct tcpipv6hdr tcp_saveti6; 105 106 /* for the packet header length in the mbuf */ 107 #define M_PH_LEN(m) (((struct mbuf *)(m))->m_pkthdr.len) 108 #define M_V6_LEN(m) (M_PH_LEN(m) - sizeof(struct ip6_hdr)) 109 #define M_V4_LEN(m) (M_PH_LEN(m) - sizeof(struct ip)) 110 #endif /* INET6 */ 111 112 int tcprexmtthresh = 3; 113 struct tcpiphdr tcp_saveti; 114 int tcptv_keep_init = TCPTV_KEEP_INIT; 115 116 extern u_long sb_max; 117 118 int tcp_rst_ppslim = 100; /* 100pps */ 119 int tcp_rst_ppslim_count = 0; 120 struct timeval tcp_rst_ppslim_last; 121 122 #endif /* TUBA_INCLUDE */ 123 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 124 125 /* for modulo comparisons of timestamps */ 126 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 127 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 128 129 /* 130 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 131 */ 132 #ifdef INET6 133 #define ND6_HINT(tp) \ 134 do { \ 135 if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \ 136 tp->t_inpcb->inp_route6.ro_rt) { \ 137 nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt, NULL, 0); \ 138 } \ 139 } while (0) 140 #else 141 #define ND6_HINT(tp) 142 #endif 143 144 /* 145 * Insert segment ti into reassembly queue of tcp with 146 * control block tp. Return TH_FIN if reassembly now includes 147 * a segment with FIN. The macro form does the common case inline 148 * (segment is the next to be received on an established connection, 149 * and the queue is empty), avoiding linkage into and removal 150 * from the queue and repetition of various conversions. 151 * Set DELACK for segments received in order, but ack immediately 152 * when segments are out of order (so fast retransmit can work). 153 */ 154 155 #ifndef TUBA_INCLUDE 156 157 int 158 tcp_reass(tp, th, m, tlen) 159 register struct tcpcb *tp; 160 register struct tcphdr *th; 161 struct mbuf *m; 162 int *tlen; 163 { 164 register struct ipqent *p, *q, *nq, *tiqe; 165 struct socket *so = tp->t_inpcb->inp_socket; 166 int flags; 167 168 /* 169 * Call with th==0 after become established to 170 * force pre-ESTABLISHED data up to user socket. 171 */ 172 if (th == 0) 173 goto present; 174 175 /* 176 * Allocate a new queue entry, before we throw away any data. 177 * If we can't, just drop the packet. XXX 178 */ 179 tiqe = pool_get(&ipqent_pool, PR_NOWAIT); 180 if (tiqe == NULL) { 181 tcpstat.tcps_rcvmemdrop++; 182 m_freem(m); 183 return (0); 184 } 185 186 /* 187 * Find a segment which begins after this one does. 188 */ 189 for (p = NULL, q = tp->segq.lh_first; q != NULL; 190 p = q, q = q->ipqe_q.le_next) 191 if (SEQ_GT(q->ipqe_tcp->th_seq, th->th_seq)) 192 break; 193 194 /* 195 * If there is a preceding segment, it may provide some of 196 * our data already. If so, drop the data from the incoming 197 * segment. If it provides all of our data, drop us. 198 */ 199 if (p != NULL) { 200 register struct tcphdr *phdr = p->ipqe_tcp; 201 register int i; 202 203 /* conversion to int (in i) handles seq wraparound */ 204 i = phdr->th_seq + phdr->th_reseqlen - th->th_seq; 205 if (i > 0) { 206 if (i >= *tlen) { 207 tcpstat.tcps_rcvduppack++; 208 tcpstat.tcps_rcvdupbyte += *tlen; 209 m_freem(m); 210 pool_put(&ipqent_pool, tiqe); 211 return (0); 212 } 213 m_adj(m, i); 214 *tlen -= i; 215 th->th_seq += i; 216 } 217 } 218 tcpstat.tcps_rcvoopack++; 219 tcpstat.tcps_rcvoobyte += *tlen; 220 221 /* 222 * While we overlap succeeding segments trim them or, 223 * if they are completely covered, dequeue them. 224 */ 225 for (; q != NULL; q = nq) { 226 register struct tcphdr *qhdr = q->ipqe_tcp; 227 register int i = (th->th_seq + *tlen) - qhdr->th_seq; 228 229 if (i <= 0) 230 break; 231 if (i < qhdr->th_reseqlen) { 232 qhdr->th_seq += i; 233 qhdr->th_reseqlen -= i; 234 m_adj(q->ipqe_m, i); 235 break; 236 } 237 nq = q->ipqe_q.le_next; 238 m_freem(q->ipqe_m); 239 LIST_REMOVE(q, ipqe_q); 240 pool_put(&ipqent_pool, q); 241 } 242 243 /* Insert the new fragment queue entry into place. */ 244 tiqe->ipqe_m = m; 245 th->th_reseqlen = *tlen; 246 tiqe->ipqe_tcp = th; 247 if (p == NULL) { 248 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 249 } else { 250 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 251 } 252 253 present: 254 /* 255 * Present data to user, advancing rcv_nxt through 256 * completed sequence space. 257 */ 258 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 259 return (0); 260 q = tp->segq.lh_first; 261 if (q == NULL || q->ipqe_tcp->th_seq != tp->rcv_nxt) 262 return (0); 263 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_tcp->th_reseqlen) 264 return (0); 265 do { 266 tp->rcv_nxt += q->ipqe_tcp->th_reseqlen; 267 flags = q->ipqe_tcp->th_flags & TH_FIN; 268 269 nq = q->ipqe_q.le_next; 270 LIST_REMOVE(q, ipqe_q); 271 ND6_HINT(tp); 272 if (so->so_state & SS_CANTRCVMORE) 273 m_freem(q->ipqe_m); 274 else 275 sbappend(&so->so_rcv, q->ipqe_m); 276 pool_put(&ipqent_pool, q); 277 q = nq; 278 } while (q != NULL && q->ipqe_tcp->th_seq == tp->rcv_nxt); 279 sorwakeup(so); 280 return (flags); 281 } 282 283 /* 284 * First check for a port-specific bomb. We do not want to drop half-opens 285 * for other ports if this is the only port being bombed. We only check 286 * the bottom 40 half open connections, to avoid wasting too much time. 287 * 288 * Or, otherwise it is more likely a generic syn bomb, so delete the oldest 289 * half-open connection. 290 */ 291 void 292 tcpdropoldhalfopen(avoidtp, port) 293 struct tcpcb *avoidtp; 294 u_int16_t port; 295 { 296 register struct inpcb *inp; 297 register struct tcpcb *tp; 298 int ncheck = 40; 299 int s; 300 301 s = splnet(); 302 inp = tcbtable.inpt_queue.cqh_first; 303 if (inp) /* XXX */ 304 for (; inp != (struct inpcb *)&tcbtable.inpt_queue && --ncheck; 305 inp = inp->inp_queue.cqe_prev) { 306 if ((tp = (struct tcpcb *)inp->inp_ppcb) && 307 tp != avoidtp && 308 tp->t_state == TCPS_SYN_RECEIVED && 309 port == inp->inp_lport) { 310 tcp_close(tp); 311 goto done; 312 } 313 } 314 315 inp = tcbtable.inpt_queue.cqh_first; 316 if (inp) /* XXX */ 317 for (; inp != (struct inpcb *)&tcbtable.inpt_queue; 318 inp = inp->inp_queue.cqe_prev) { 319 if ((tp = (struct tcpcb *)inp->inp_ppcb) && 320 tp != avoidtp && 321 tp->t_state == TCPS_SYN_RECEIVED) { 322 tcp_close(tp); 323 goto done; 324 } 325 } 326 done: 327 splx(s); 328 } 329 330 #ifdef INET6 331 int 332 tcp6_input(mp, offp, proto) 333 struct mbuf **mp; 334 int *offp, proto; 335 { 336 struct mbuf *m = *mp; 337 338 #if defined(NFAITH) && 0 < NFAITH 339 if (m->m_pkthdr.rcvif) { 340 if (m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 341 /* XXX send icmp6 host/port unreach? */ 342 m_freem(m); 343 return IPPROTO_DONE; 344 } 345 } 346 #endif 347 348 /* 349 * draft-itojun-ipv6-tcp-to-anycast 350 * better place to put this in? 351 */ 352 if (m->m_flags & M_ANYCAST6) { 353 if (m->m_len >= sizeof(struct ip6_hdr)) { 354 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 355 icmp6_error(m, ICMP6_DST_UNREACH, 356 ICMP6_DST_UNREACH_ADDR, 357 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 358 } else 359 m_freem(m); 360 return IPPROTO_DONE; 361 } 362 363 tcp_input(m, *offp, proto); 364 return IPPROTO_DONE; 365 } 366 #endif 367 368 /* 369 * TCP input routine, follows pages 65-76 of the 370 * protocol specification dated September, 1981 very closely. 371 */ 372 void 373 #if __STDC__ 374 tcp_input(struct mbuf *m, ...) 375 #else 376 tcp_input(m, va_alist) 377 register struct mbuf *m; 378 #endif 379 { 380 struct ip *ip; 381 register struct inpcb *inp; 382 caddr_t optp = NULL; 383 int optlen = 0; 384 int len, tlen, off; 385 register struct tcpcb *tp = 0; 386 register int tiflags; 387 struct socket *so = NULL; 388 int todrop, acked, ourfinisacked, needoutput = 0; 389 int hdroptlen = 0; 390 short ostate = 0; 391 struct in_addr laddr; 392 int dropsocket = 0; 393 int iss = 0; 394 u_long tiwin; 395 u_int32_t ts_val, ts_ecr; 396 int ts_present = 0; 397 int iphlen; 398 va_list ap; 399 register struct tcphdr *th; 400 #ifdef INET6 401 struct in6_addr laddr6; 402 struct ip6_hdr *ipv6 = NULL; 403 #endif /* INET6 */ 404 #ifdef IPSEC 405 struct m_tag *mtag; 406 struct tdb_ident *tdbi; 407 struct tdb *tdb; 408 int error, s; 409 #endif /* IPSEC */ 410 int af; 411 412 va_start(ap, m); 413 iphlen = va_arg(ap, int); 414 va_end(ap); 415 416 tcpstat.tcps_rcvtotal++; 417 418 /* 419 * Before we do ANYTHING, we have to figure out if it's TCP/IPv6 or 420 * TCP/IPv4. 421 */ 422 switch (mtod(m, struct ip *)->ip_v) { 423 #ifdef INET6 424 case 6: 425 af = AF_INET6; 426 break; 427 #endif 428 case 4: 429 af = AF_INET; 430 break; 431 default: 432 m_freem(m); 433 return; /*EAFNOSUPPORT*/ 434 } 435 436 /* 437 * Get IP and TCP header together in first mbuf. 438 * Note: IP leaves IP header in first mbuf. 439 */ 440 switch (af) { 441 case AF_INET: 442 #ifdef DIAGNOSTIC 443 if (iphlen < sizeof(struct ip)) { 444 m_freem(m); 445 return; 446 } 447 #endif /* DIAGNOSTIC */ 448 if (iphlen > sizeof(struct ip)) { 449 #if 0 /*XXX*/ 450 ip_stripoptions(m, (struct mbuf *)0); 451 iphlen = sizeof(struct ip); 452 #else 453 m_freem(m); 454 return; 455 #endif 456 } 457 break; 458 #ifdef INET6 459 case AF_INET6: 460 #ifdef DIAGNOSTIC 461 if (iphlen < sizeof(struct ip6_hdr)) { 462 m_freem(m); 463 return; 464 } 465 #endif /* DIAGNOSTIC */ 466 if (iphlen > sizeof(struct ip6_hdr)) { 467 #if 0 /*XXX*/ 468 ipv6_stripoptions(m, iphlen); 469 iphlen = sizeof(struct ip6_hdr); 470 #else 471 m_freem(m); 472 return; 473 #endif 474 } 475 break; 476 #endif 477 default: 478 m_freem(m); 479 return; 480 } 481 482 if (m->m_len < iphlen + sizeof(struct tcphdr)) { 483 m = m_pullup2(m, iphlen + sizeof(struct tcphdr)); 484 if (m == NULL) { 485 tcpstat.tcps_rcvshort++; 486 return; 487 } 488 } 489 490 ip = NULL; 491 #ifdef INET6 492 ipv6 = NULL; 493 #endif 494 switch (af) { 495 case AF_INET: 496 { 497 struct tcpiphdr *ti; 498 499 ip = mtod(m, struct ip *); 500 #if 1 501 tlen = m->m_pkthdr.len - iphlen; 502 #else 503 tlen = ((struct ip *)ti)->ip_len; 504 #endif 505 ti = mtod(m, struct tcpiphdr *); 506 507 /* 508 * Checksum extended TCP header and data. 509 */ 510 len = sizeof(struct ip) + tlen; 511 bzero(ti->ti_x1, sizeof ti->ti_x1); 512 ti->ti_len = (u_int16_t)tlen; 513 HTONS(ti->ti_len); 514 if ((m->m_pkthdr.csum & M_TCP_CSUM_IN_OK) == 0) { 515 if (m->m_pkthdr.csum & M_TCP_CSUM_IN_BAD) { 516 tcpstat.tcps_inhwcsum++; 517 tcpstat.tcps_rcvbadsum++; 518 goto drop; 519 } 520 if ((ti->ti_sum = in_cksum(m, len)) != 0) { 521 tcpstat.tcps_rcvbadsum++; 522 goto drop; 523 } 524 } else { 525 m->m_pkthdr.csum &= ~M_TCP_CSUM_IN_OK; 526 tcpstat.tcps_inhwcsum++; 527 } 528 break; 529 } 530 #ifdef INET6 531 case AF_INET6: 532 ipv6 = mtod(m, struct ip6_hdr *); 533 tlen = m->m_pkthdr.len - iphlen; 534 535 /* Be proactive about malicious use of IPv4 mapped address */ 536 if (IN6_IS_ADDR_V4MAPPED(&ipv6->ip6_src) || 537 IN6_IS_ADDR_V4MAPPED(&ipv6->ip6_dst)) { 538 /* XXX stat */ 539 goto drop; 540 } 541 542 /* 543 * Be proactive about unspecified IPv6 address in source. 544 * As we use all-zero to indicate unbounded/unconnected pcb, 545 * unspecified IPv6 address can be used to confuse us. 546 * 547 * Note that packets with unspecified IPv6 destination is 548 * already dropped in ip6_input. 549 */ 550 if (IN6_IS_ADDR_UNSPECIFIED(&ipv6->ip6_src)) { 551 /* XXX stat */ 552 goto drop; 553 } 554 555 /* 556 * Checksum extended TCP header and data. 557 */ 558 if (in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), tlen)) { 559 tcpstat.tcps_rcvbadsum++; 560 goto drop; 561 } 562 break; 563 #endif 564 } 565 #endif /* TUBA_INCLUDE */ 566 567 th = (struct tcphdr *)(mtod(m, caddr_t) + iphlen); 568 569 /* 570 * Check that TCP offset makes sense, 571 * pull out TCP options and adjust length. XXX 572 */ 573 off = th->th_off << 2; 574 if (off < sizeof(struct tcphdr) || off > tlen) { 575 tcpstat.tcps_rcvbadoff++; 576 goto drop; 577 } 578 tlen -= off; 579 if (off > sizeof(struct tcphdr)) { 580 if (m->m_len < iphlen + off) { 581 if ((m = m_pullup2(m, iphlen + off)) == NULL) { 582 tcpstat.tcps_rcvshort++; 583 return; 584 } 585 switch (af) { 586 case AF_INET: 587 ip = mtod(m, struct ip *); 588 break; 589 #ifdef INET6 590 case AF_INET6: 591 ipv6 = mtod(m, struct ip6_hdr *); 592 break; 593 #endif 594 } 595 th = (struct tcphdr *)(mtod(m, caddr_t) + iphlen); 596 } 597 optlen = off - sizeof(struct tcphdr); 598 optp = mtod(m, caddr_t) + iphlen + sizeof(struct tcphdr); 599 /* 600 * Do quick retrieval of timestamp options ("options 601 * prediction?"). If timestamp is the only option and it's 602 * formatted as recommended in RFC 1323 appendix A, we 603 * quickly get the values now and not bother calling 604 * tcp_dooptions(), etc. 605 */ 606 if ((optlen == TCPOLEN_TSTAMP_APPA || 607 (optlen > TCPOLEN_TSTAMP_APPA && 608 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 609 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 610 (th->th_flags & TH_SYN) == 0) { 611 ts_present = 1; 612 ts_val = ntohl(*(u_int32_t *)(optp + 4)); 613 ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 614 optp = NULL; /* we've parsed the options */ 615 } 616 } 617 tiflags = th->th_flags; 618 619 /* 620 * Convert TCP protocol specific fields to host format. 621 */ 622 NTOHL(th->th_seq); 623 NTOHL(th->th_ack); 624 NTOHS(th->th_win); 625 NTOHS(th->th_urp); 626 627 /* 628 * Locate pcb for segment. 629 */ 630 findpcb: 631 switch (af) { 632 #ifdef INET6 633 case AF_INET6: 634 inp = in6_pcbhashlookup(&tcbtable, &ipv6->ip6_src, th->th_sport, 635 &ipv6->ip6_dst, th->th_dport); 636 break; 637 #endif 638 case AF_INET: 639 inp = in_pcbhashlookup(&tcbtable, ip->ip_src, th->th_sport, 640 ip->ip_dst, th->th_dport); 641 break; 642 } 643 if (inp == 0) { 644 ++tcpstat.tcps_pcbhashmiss; 645 switch (af) { 646 #ifdef INET6 647 case AF_INET6: 648 inp = in_pcblookup(&tcbtable, &ipv6->ip6_src, 649 th->th_sport, &ipv6->ip6_dst, th->th_dport, 650 INPLOOKUP_WILDCARD | INPLOOKUP_IPV6); 651 break; 652 #endif /* INET6 */ 653 case AF_INET: 654 inp = in_pcblookup(&tcbtable, &ip->ip_src, th->th_sport, 655 &ip->ip_dst, th->th_dport, INPLOOKUP_WILDCARD); 656 break; 657 } 658 /* 659 * If the state is CLOSED (i.e., TCB does not exist) then 660 * all data in the incoming segment is discarded. 661 * If the TCB exists but is in CLOSED state, it is embryonic, 662 * but should either do a listen or a connect soon. 663 */ 664 if (inp == 0) { 665 ++tcpstat.tcps_noport; 666 goto dropwithreset_ratelim; 667 } 668 } 669 670 tp = intotcpcb(inp); 671 if (tp == 0) 672 goto dropwithreset_ratelim; 673 if (tp->t_state == TCPS_CLOSED) 674 goto drop; 675 676 /* Unscale the window into a 32-bit value. */ 677 if ((tiflags & TH_SYN) == 0) 678 tiwin = th->th_win << tp->snd_scale; 679 else 680 tiwin = th->th_win; 681 682 so = inp->inp_socket; 683 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 684 if (so->so_options & SO_DEBUG) { 685 ostate = tp->t_state; 686 switch (af) { 687 #ifdef INET6 688 case AF_INET6: 689 tcp_saveti6 = *(mtod(m, struct tcpipv6hdr *)); 690 break; 691 #endif 692 case AF_INET: 693 tcp_saveti = *(mtod(m, struct tcpiphdr *)); 694 break; 695 } 696 } 697 if (so->so_options & SO_ACCEPTCONN) { 698 struct socket *so1; 699 700 so1 = sonewconn(so, 0); 701 if (so1 == NULL) { 702 tcpdropoldhalfopen(tp, th->th_dport); 703 so1 = sonewconn(so, 0); 704 if (so1 == NULL) 705 goto drop; 706 } 707 so = so1; 708 /* 709 * This is ugly, but .... 710 * 711 * Mark socket as temporary until we're 712 * committed to keeping it. The code at 713 * ``drop'' and ``dropwithreset'' check the 714 * flag dropsocket to see if the temporary 715 * socket created here should be discarded. 716 * We mark the socket as discardable until 717 * we're committed to it below in TCPS_LISTEN. 718 */ 719 dropsocket++; 720 #ifdef IPSEC 721 /* 722 * We need to copy the required security levels 723 * from the old pcb. Ditto for any other 724 * IPsec-related information. 725 */ 726 { 727 struct inpcb *newinp = (struct inpcb *)so->so_pcb; 728 bcopy(inp->inp_seclevel, newinp->inp_seclevel, 729 sizeof(inp->inp_seclevel)); 730 newinp->inp_secrequire = inp->inp_secrequire; 731 if (inp->inp_ipsec_localid != NULL) { 732 newinp->inp_ipsec_localid = inp->inp_ipsec_localid; 733 inp->inp_ipsec_localid->ref_count++; 734 } 735 if (inp->inp_ipsec_remoteid != NULL) { 736 newinp->inp_ipsec_remoteid = inp->inp_ipsec_remoteid; 737 inp->inp_ipsec_remoteid->ref_count++; 738 } 739 if (inp->inp_ipsec_localcred != NULL) { 740 newinp->inp_ipsec_localcred = inp->inp_ipsec_localcred; 741 inp->inp_ipsec_localcred->ref_count++; 742 } 743 if (inp->inp_ipsec_remotecred != NULL) { 744 newinp->inp_ipsec_remotecred = inp->inp_ipsec_remotecred; 745 inp->inp_ipsec_remotecred->ref_count++; 746 } 747 if (inp->inp_ipsec_localauth != NULL) { 748 newinp->inp_ipsec_localauth 749 = inp->inp_ipsec_localauth; 750 inp->inp_ipsec_localauth->ref_count++; 751 } 752 if (inp->inp_ipsec_remoteauth != NULL) { 753 newinp->inp_ipsec_remoteauth 754 = inp->inp_ipsec_remoteauth; 755 inp->inp_ipsec_remoteauth->ref_count++; 756 } 757 } 758 #endif /* IPSEC */ 759 #ifdef INET6 760 /* 761 * inp still has the OLD in_pcb stuff, set the 762 * v6-related flags on the new guy, too. This is 763 * done particularly for the case where an AF_INET6 764 * socket is bound only to a port, and a v4 connection 765 * comes in on that port. 766 * we also copy the flowinfo from the original pcb 767 * to the new one. 768 */ 769 { 770 int flags = inp->inp_flags; 771 struct inpcb *oldinpcb = inp; 772 773 inp = (struct inpcb *)so->so_pcb; 774 inp->inp_flags |= (flags & INP_IPV6); 775 if ((inp->inp_flags & INP_IPV6) != 0) { 776 inp->inp_ipv6.ip6_hlim = 777 oldinpcb->inp_ipv6.ip6_hlim; 778 inp->inp_ipv6.ip6_flow = 779 oldinpcb->inp_ipv6.ip6_flow; 780 } 781 } 782 #else /* INET6 */ 783 inp = (struct inpcb *)so->so_pcb; 784 #endif /* INET6 */ 785 inp->inp_lport = th->th_dport; 786 switch (af) { 787 #ifdef INET6 788 case AF_INET6: 789 inp->inp_laddr6 = ipv6->ip6_dst; 790 791 /*inp->inp_options = ip6_srcroute();*/ /* soon. */ 792 /* 793 * still need to tweak outbound options 794 * processing to include this mbuf in 795 * the right place and put the correct 796 * NextHdr values in the right places. 797 * XXX rja 798 */ 799 break; 800 #endif /* INET6 */ 801 case AF_INET: 802 inp->inp_laddr = ip->ip_dst; 803 inp->inp_options = ip_srcroute(); 804 break; 805 } 806 in_pcbrehash(inp); 807 tp = intotcpcb(inp); 808 tp->t_state = TCPS_LISTEN; 809 810 /* Compute proper scaling value from buffer space */ 811 tcp_rscale(tp, so->so_rcv.sb_hiwat); 812 } 813 } 814 815 #ifdef IPSEC 816 /* Find most recent IPsec tag */ 817 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL); 818 s = splnet(); 819 if (mtag != NULL) { 820 tdbi = (struct tdb_ident *)(mtag + 1); 821 tdb = gettdb(tdbi->spi, &tdbi->dst, tdbi->proto); 822 } else 823 tdb = NULL; 824 ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN, 825 tdb, inp); 826 827 /* Latch SA */ 828 if (inp->inp_tdb_in != tdb) { 829 if (tdb) { 830 tdb_add_inp(tdb, inp, 1); 831 if (inp->inp_ipsec_remoteid == NULL && 832 tdb->tdb_srcid != NULL) { 833 inp->inp_ipsec_remoteid = tdb->tdb_srcid; 834 tdb->tdb_srcid->ref_count++; 835 } 836 if (inp->inp_ipsec_remotecred == NULL && 837 tdb->tdb_remote_cred != NULL) { 838 inp->inp_ipsec_remotecred = 839 tdb->tdb_remote_cred; 840 tdb->tdb_remote_cred->ref_count++; 841 } 842 if (inp->inp_ipsec_remoteauth == NULL && 843 tdb->tdb_remote_auth != NULL) { 844 inp->inp_ipsec_remoteauth = 845 tdb->tdb_remote_auth; 846 tdb->tdb_remote_auth->ref_count++; 847 } 848 } else { /* Just reset */ 849 TAILQ_REMOVE(&inp->inp_tdb_in->tdb_inp_in, inp, 850 inp_tdb_in_next); 851 inp->inp_tdb_in = NULL; 852 } 853 } 854 splx(s); 855 856 /* Error or otherwise drop-packet indication */ 857 if (error) 858 goto drop; 859 #endif /* IPSEC */ 860 861 /* 862 * Segment received on connection. 863 * Reset idle time and keep-alive timer. 864 */ 865 tp->t_idle = 0; 866 if (tp->t_state != TCPS_SYN_RECEIVED) 867 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 868 869 #ifdef TCP_SACK 870 if (!tp->sack_disable) 871 tcp_del_sackholes(tp, th); /* Delete stale SACK holes */ 872 #endif /* TCP_SACK */ 873 874 /* 875 * Process options if not in LISTEN state, 876 * else do it below (after getting remote address). 877 */ 878 if (optp && tp->t_state != TCPS_LISTEN) 879 tcp_dooptions(tp, optp, optlen, th, 880 &ts_present, &ts_val, &ts_ecr); 881 882 #ifdef TCP_SACK 883 if (!tp->sack_disable) { 884 tp->rcv_laststart = th->th_seq; /* last rec'vd segment*/ 885 tp->rcv_lastend = th->th_seq + tlen; 886 } 887 #endif /* TCP_SACK */ 888 /* 889 * Header prediction: check for the two common cases 890 * of a uni-directional data xfer. If the packet has 891 * no control flags, is in-sequence, the window didn't 892 * change and we're not retransmitting, it's a 893 * candidate. If the length is zero and the ack moved 894 * forward, we're the sender side of the xfer. Just 895 * free the data acked & wake any higher level process 896 * that was blocked waiting for space. If the length 897 * is non-zero and the ack didn't move, we're the 898 * receiver side. If we're getting packets in-order 899 * (the reassembly queue is empty), add the data to 900 * the socket buffer and note that we need a delayed ack. 901 */ 902 if (tp->t_state == TCPS_ESTABLISHED && 903 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 904 (!ts_present || TSTMP_GEQ(ts_val, tp->ts_recent)) && 905 th->th_seq == tp->rcv_nxt && 906 tiwin && tiwin == tp->snd_wnd && 907 tp->snd_nxt == tp->snd_max) { 908 909 /* 910 * If last ACK falls within this segment's sequence numbers, 911 * record the timestamp. 912 * Fix from Braden, see Stevens p. 870 913 */ 914 if (ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 915 tp->ts_recent_age = tcp_now; 916 tp->ts_recent = ts_val; 917 } 918 919 if (tlen == 0) { 920 if (SEQ_GT(th->th_ack, tp->snd_una) && 921 SEQ_LEQ(th->th_ack, tp->snd_max) && 922 tp->snd_cwnd >= tp->snd_wnd && 923 tp->t_dupacks == 0) { 924 /* 925 * this is a pure ack for outstanding data. 926 */ 927 ++tcpstat.tcps_predack; 928 if (ts_present) 929 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 930 else if (tp->t_rtt && 931 SEQ_GT(th->th_ack, tp->t_rtseq)) 932 tcp_xmit_timer(tp, tp->t_rtt); 933 acked = th->th_ack - tp->snd_una; 934 tcpstat.tcps_rcvackpack++; 935 tcpstat.tcps_rcvackbyte += acked; 936 ND6_HINT(tp); 937 sbdrop(&so->so_snd, acked); 938 tp->snd_una = th->th_ack; 939 #if defined(TCP_SACK) 940 /* 941 * We want snd_last to track snd_una so 942 * as to avoid sequence wraparound problems 943 * for very large transfers. 944 */ 945 tp->snd_last = tp->snd_una; 946 #endif /* TCP_SACK */ 947 #if defined(TCP_SACK) && defined(TCP_FACK) 948 tp->snd_fack = tp->snd_una; 949 tp->retran_data = 0; 950 #endif /* TCP_FACK */ 951 m_freem(m); 952 953 /* 954 * If all outstanding data are acked, stop 955 * retransmit timer, otherwise restart timer 956 * using current (possibly backed-off) value. 957 * If process is waiting for space, 958 * wakeup/selwakeup/signal. If data 959 * are ready to send, let tcp_output 960 * decide between more output or persist. 961 */ 962 if (tp->snd_una == tp->snd_max) 963 TCP_TIMER_DISARM(tp, TCPT_REXMT); 964 else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 965 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 966 967 if (sb_notify(&so->so_snd)) 968 sowwakeup(so); 969 if (so->so_snd.sb_cc) 970 (void) tcp_output(tp); 971 return; 972 } 973 } else if (th->th_ack == tp->snd_una && 974 tp->segq.lh_first == NULL && 975 tlen <= sbspace(&so->so_rcv)) { 976 /* 977 * This is a pure, in-sequence data packet 978 * with nothing on the reassembly queue and 979 * we have enough buffer space to take it. 980 */ 981 #ifdef TCP_SACK 982 /* Clean receiver SACK report if present */ 983 if (!tp->sack_disable && tp->rcv_numsacks) 984 tcp_clean_sackreport(tp); 985 #endif /* TCP_SACK */ 986 ++tcpstat.tcps_preddat; 987 tp->rcv_nxt += tlen; 988 tcpstat.tcps_rcvpack++; 989 tcpstat.tcps_rcvbyte += tlen; 990 ND6_HINT(tp); 991 /* 992 * Drop TCP, IP headers and TCP options then add data 993 * to socket buffer. 994 */ 995 if (th->th_flags & TH_PUSH) 996 tp->t_flags |= TF_ACKNOW; 997 else 998 tp->t_flags |= TF_DELACK; 999 m_adj(m, iphlen + off); 1000 sbappend(&so->so_rcv, m); 1001 sorwakeup(so); 1002 return; 1003 } 1004 } 1005 1006 /* 1007 * Compute mbuf offset to TCP data segment. 1008 */ 1009 hdroptlen = iphlen + off; 1010 1011 /* 1012 * Calculate amount of space in receive window, 1013 * and then do TCP input processing. 1014 * Receive window is amount of space in rcv queue, 1015 * but not less than advertised window. 1016 */ 1017 { int win; 1018 1019 win = sbspace(&so->so_rcv); 1020 if (win < 0) 1021 win = 0; 1022 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1023 } 1024 1025 switch (tp->t_state) { 1026 1027 /* 1028 * If the state is LISTEN then ignore segment if it contains an RST. 1029 * If the segment contains an ACK then it is bad and send a RST. 1030 * If it does not contain a SYN then it is not interesting; drop it. 1031 * If it is from this socket, drop it, it must be forged. 1032 * Don't bother responding if the destination was a broadcast. 1033 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial 1034 * tp->iss, and send a segment: 1035 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 1036 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss. 1037 * Fill in remote peer address fields if not previously specified. 1038 * Enter SYN_RECEIVED state, and process any other fields of this 1039 * segment in this state. 1040 */ 1041 case TCPS_LISTEN: { 1042 struct mbuf *am; 1043 register struct sockaddr_in *sin; 1044 #ifdef INET6 1045 register struct sockaddr_in6 *sin6; 1046 #endif /* INET6 */ 1047 1048 if (tiflags & TH_RST) 1049 goto drop; 1050 if (tiflags & TH_ACK) 1051 goto dropwithreset; 1052 if ((tiflags & TH_SYN) == 0) 1053 goto drop; 1054 if (th->th_dport == th->th_sport) { 1055 switch (af) { 1056 #ifdef INET6 1057 case AF_INET6: 1058 if (IN6_ARE_ADDR_EQUAL(&ipv6->ip6_src, 1059 &ipv6->ip6_dst)) 1060 goto drop; 1061 break; 1062 #endif /* INET6 */ 1063 case AF_INET: 1064 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 1065 goto drop; 1066 break; 1067 } 1068 } 1069 1070 /* 1071 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 1072 * in_broadcast() should never return true on a received 1073 * packet with M_BCAST not set. 1074 */ 1075 if (m->m_flags & (M_BCAST|M_MCAST)) 1076 goto drop; 1077 switch (af) { 1078 #ifdef INET6 1079 case AF_INET6: 1080 /* XXX What about IPv6 Anycasting ?? :-( rja */ 1081 if (IN6_IS_ADDR_MULTICAST(&ipv6->ip6_dst)) 1082 goto drop; 1083 break; 1084 #endif /* INET6 */ 1085 case AF_INET: 1086 if (IN_MULTICAST(ip->ip_dst.s_addr)) 1087 goto drop; 1088 break; 1089 } 1090 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 1091 if (am == NULL) 1092 goto drop; 1093 switch (af) { 1094 #ifdef INET6 1095 case AF_INET6: 1096 /* 1097 * This is probably the place to set the tp->pf value. 1098 * (Don't forget to do it in the v4 code as well!) 1099 * 1100 * Also, remember to blank out things like flowlabel, or 1101 * set flowlabel for accepted sockets in v6. 1102 * 1103 * FURTHERMORE, this is PROBABLY the place where the 1104 * whole business of key munging is set up for passive 1105 * connections. 1106 */ 1107 am->m_len = sizeof(struct sockaddr_in6); 1108 sin6 = mtod(am, struct sockaddr_in6 *); 1109 sin6->sin6_family = AF_INET6; 1110 sin6->sin6_len = sizeof(struct sockaddr_in6); 1111 sin6->sin6_addr = ipv6->ip6_src; 1112 sin6->sin6_port = th->th_sport; 1113 sin6->sin6_flowinfo = htonl(0x0fffffff) & 1114 inp->inp_ipv6.ip6_flow; 1115 laddr6 = inp->inp_laddr6; 1116 if (IN6_IS_ADDR_UNSPECIFIED(&inp->inp_laddr6)) 1117 inp->inp_laddr6 = ipv6->ip6_dst; 1118 /* This is a good optimization. */ 1119 if (in6_pcbconnect(inp, am)) { 1120 inp->inp_laddr6 = laddr6; 1121 (void) m_free(am); 1122 goto drop; 1123 } 1124 break; 1125 #endif 1126 case AF_INET: 1127 /* drop IPv4 packet to AF_INET6 socket */ 1128 if (inp->inp_flags & INP_IPV6) { 1129 (void) m_free(am); 1130 goto drop; 1131 } 1132 am->m_len = sizeof(struct sockaddr_in); 1133 sin = mtod(am, struct sockaddr_in *); 1134 sin->sin_family = AF_INET; 1135 sin->sin_len = sizeof(*sin); 1136 sin->sin_addr = ip->ip_src; 1137 sin->sin_port = th->th_sport; 1138 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero)); 1139 laddr = inp->inp_laddr; 1140 if (inp->inp_laddr.s_addr == INADDR_ANY) 1141 inp->inp_laddr = ip->ip_dst; 1142 if (in_pcbconnect(inp, am)) { 1143 inp->inp_laddr = laddr; 1144 (void) m_free(am); 1145 goto drop; 1146 } 1147 (void) m_free(am); 1148 break; 1149 } 1150 tp->t_template = tcp_template(tp); 1151 if (tp->t_template == 0) { 1152 tp = tcp_drop(tp, ENOBUFS); 1153 dropsocket = 0; /* socket is already gone */ 1154 goto drop; 1155 } 1156 if (optp) 1157 tcp_dooptions(tp, optp, optlen, th, 1158 &ts_present, &ts_val, &ts_ecr); 1159 #ifdef TCP_SACK 1160 /* 1161 * If peer did not send a SACK_PERMITTED option (i.e., if 1162 * tcp_dooptions() did not set TF_SACK_PERMIT), set 1163 * sack_disable to 1 if it is currently 0. 1164 */ 1165 if (!tp->sack_disable) 1166 if ((tp->t_flags & TF_SACK_PERMIT) == 0) 1167 tp->sack_disable = 1; 1168 #endif 1169 1170 if (iss) 1171 tp->iss = iss; 1172 else { 1173 #ifdef TCP_COMPAT_42 1174 tcp_iss += TCP_ISSINCR/2; 1175 tp->iss = tcp_iss; 1176 #else /* TCP_COMPAT_42 */ 1177 tp->iss = tcp_rndiss_next(); 1178 #endif /* !TCP_COMPAT_42 */ 1179 } 1180 tp->irs = th->th_seq; 1181 tcp_sendseqinit(tp); 1182 #if defined (TCP_SACK) 1183 tp->snd_last = tp->snd_una; 1184 #endif /* TCP_SACK */ 1185 #if defined(TCP_SACK) && defined(TCP_FACK) 1186 tp->snd_fack = tp->snd_una; 1187 tp->retran_data = 0; 1188 tp->snd_awnd = 0; 1189 #endif /* TCP_FACK */ 1190 tcp_rcvseqinit(tp); 1191 tp->t_flags |= TF_ACKNOW; 1192 tp->t_state = TCPS_SYN_RECEIVED; 1193 TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init); 1194 dropsocket = 0; /* committed to socket */ 1195 tcpstat.tcps_accepts++; 1196 goto trimthenstep6; 1197 } 1198 1199 /* 1200 * If the state is SYN_RECEIVED: 1201 * if seg contains SYN/ACK, send an RST. 1202 * if seg contains an ACK, but not for our SYN/ACK, send an RST 1203 */ 1204 1205 case TCPS_SYN_RECEIVED: 1206 if (tiflags & TH_ACK) { 1207 if (tiflags & TH_SYN) { 1208 tcpstat.tcps_badsyn++; 1209 goto dropwithreset; 1210 } 1211 if (SEQ_LEQ(th->th_ack, tp->snd_una) || 1212 SEQ_GT(th->th_ack, tp->snd_max)) 1213 goto dropwithreset; 1214 } 1215 break; 1216 1217 /* 1218 * If the state is SYN_SENT: 1219 * if seg contains an ACK, but not for our SYN, drop the input. 1220 * if seg contains a RST, then drop the connection. 1221 * if seg does not contain SYN, then drop it. 1222 * Otherwise this is an acceptable SYN segment 1223 * initialize tp->rcv_nxt and tp->irs 1224 * if seg contains ack then advance tp->snd_una 1225 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1226 * arrange for segment to be acked (eventually) 1227 * continue processing rest of data/controls, beginning with URG 1228 */ 1229 case TCPS_SYN_SENT: 1230 if ((tiflags & TH_ACK) && 1231 (SEQ_LEQ(th->th_ack, tp->iss) || 1232 SEQ_GT(th->th_ack, tp->snd_max))) 1233 goto dropwithreset; 1234 if (tiflags & TH_RST) { 1235 if (tiflags & TH_ACK) 1236 tp = tcp_drop(tp, ECONNREFUSED); 1237 goto drop; 1238 } 1239 if ((tiflags & TH_SYN) == 0) 1240 goto drop; 1241 if (tiflags & TH_ACK) { 1242 tp->snd_una = th->th_ack; 1243 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1244 tp->snd_nxt = tp->snd_una; 1245 } 1246 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1247 tp->irs = th->th_seq; 1248 tcp_rcvseqinit(tp); 1249 tp->t_flags |= TF_ACKNOW; 1250 #ifdef TCP_SACK 1251 /* 1252 * If we've sent a SACK_PERMITTED option, and the peer 1253 * also replied with one, then TF_SACK_PERMIT should have 1254 * been set in tcp_dooptions(). If it was not, disable SACKs. 1255 */ 1256 if (!tp->sack_disable) 1257 if ((tp->t_flags & TF_SACK_PERMIT) == 0) 1258 tp->sack_disable = 1; 1259 #endif 1260 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 1261 tcpstat.tcps_connects++; 1262 soisconnected(so); 1263 tp->t_state = TCPS_ESTABLISHED; 1264 /* Do window scaling on this connection? */ 1265 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1266 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1267 tp->snd_scale = tp->requested_s_scale; 1268 tp->rcv_scale = tp->request_r_scale; 1269 } 1270 (void) tcp_reass(tp, (struct tcphdr *)0, 1271 (struct mbuf *)0, &tlen); 1272 /* 1273 * if we didn't have to retransmit the SYN, 1274 * use its rtt as our initial srtt & rtt var. 1275 */ 1276 if (tp->t_rtt) 1277 tcp_xmit_timer(tp, tp->t_rtt); 1278 /* 1279 * Since new data was acked (the SYN), open the 1280 * congestion window by one MSS. We do this 1281 * here, because we won't go through the normal 1282 * ACK processing below. And since this is the 1283 * start of the connection, we know we are in 1284 * the exponential phase of slow-start. 1285 */ 1286 tp->snd_cwnd += tp->t_maxseg; 1287 } else 1288 tp->t_state = TCPS_SYN_RECEIVED; 1289 1290 trimthenstep6: 1291 /* 1292 * Advance th->th_seq to correspond to first data byte. 1293 * If data, trim to stay within window, 1294 * dropping FIN if necessary. 1295 */ 1296 th->th_seq++; 1297 if (tlen > tp->rcv_wnd) { 1298 todrop = tlen - tp->rcv_wnd; 1299 m_adj(m, -todrop); 1300 tlen = tp->rcv_wnd; 1301 tiflags &= ~TH_FIN; 1302 tcpstat.tcps_rcvpackafterwin++; 1303 tcpstat.tcps_rcvbyteafterwin += todrop; 1304 } 1305 tp->snd_wl1 = th->th_seq - 1; 1306 tp->rcv_up = th->th_seq; 1307 goto step6; 1308 } 1309 1310 /* 1311 * States other than LISTEN or SYN_SENT. 1312 * First check timestamp, if present. 1313 * Then check that at least some bytes of segment are within 1314 * receive window. If segment begins before rcv_nxt, 1315 * drop leading data (and SYN); if nothing left, just ack. 1316 * 1317 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1318 * and it's less than ts_recent, drop it. 1319 */ 1320 if (ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1321 TSTMP_LT(ts_val, tp->ts_recent)) { 1322 1323 /* Check to see if ts_recent is over 24 days old. */ 1324 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1325 /* 1326 * Invalidate ts_recent. If this segment updates 1327 * ts_recent, the age will be reset later and ts_recent 1328 * will get a valid value. If it does not, setting 1329 * ts_recent to zero will at least satisfy the 1330 * requirement that zero be placed in the timestamp 1331 * echo reply when ts_recent isn't valid. The 1332 * age isn't reset until we get a valid ts_recent 1333 * because we don't want out-of-order segments to be 1334 * dropped when ts_recent is old. 1335 */ 1336 tp->ts_recent = 0; 1337 } else { 1338 tcpstat.tcps_rcvduppack++; 1339 tcpstat.tcps_rcvdupbyte += tlen; 1340 tcpstat.tcps_pawsdrop++; 1341 goto dropafterack; 1342 } 1343 } 1344 1345 todrop = tp->rcv_nxt - th->th_seq; 1346 if (todrop > 0) { 1347 if (tiflags & TH_SYN) { 1348 tiflags &= ~TH_SYN; 1349 th->th_seq++; 1350 if (th->th_urp > 1) 1351 th->th_urp--; 1352 else 1353 tiflags &= ~TH_URG; 1354 todrop--; 1355 } 1356 if (todrop >= tlen || 1357 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1358 /* 1359 * Any valid FIN must be to the left of the 1360 * window. At this point, FIN must be a 1361 * duplicate or out-of-sequence, so drop it. 1362 */ 1363 tiflags &= ~TH_FIN; 1364 /* 1365 * Send ACK to resynchronize, and drop any data, 1366 * but keep on processing for RST or ACK. 1367 */ 1368 tp->t_flags |= TF_ACKNOW; 1369 tcpstat.tcps_rcvdupbyte += todrop = tlen; 1370 tcpstat.tcps_rcvduppack++; 1371 } else { 1372 tcpstat.tcps_rcvpartduppack++; 1373 tcpstat.tcps_rcvpartdupbyte += todrop; 1374 } 1375 hdroptlen += todrop; /* drop from head afterwards */ 1376 th->th_seq += todrop; 1377 tlen -= todrop; 1378 if (th->th_urp > todrop) 1379 th->th_urp -= todrop; 1380 else { 1381 tiflags &= ~TH_URG; 1382 th->th_urp = 0; 1383 } 1384 } 1385 1386 /* 1387 * If new data are received on a connection after the 1388 * user processes are gone, then RST the other end. 1389 */ 1390 if ((so->so_state & SS_NOFDREF) && 1391 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1392 tp = tcp_close(tp); 1393 tcpstat.tcps_rcvafterclose++; 1394 goto dropwithreset; 1395 } 1396 1397 /* 1398 * If segment ends after window, drop trailing data 1399 * (and PUSH and FIN); if nothing left, just ACK. 1400 */ 1401 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1402 if (todrop > 0) { 1403 tcpstat.tcps_rcvpackafterwin++; 1404 if (todrop >= tlen) { 1405 tcpstat.tcps_rcvbyteafterwin += tlen; 1406 /* 1407 * If a new connection request is received 1408 * while in TIME_WAIT, drop the old connection 1409 * and start over if the sequence numbers 1410 * are above the previous ones. 1411 */ 1412 if (tiflags & TH_SYN && 1413 tp->t_state == TCPS_TIME_WAIT && 1414 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1415 iss = tp->snd_nxt + TCP_ISSINCR; 1416 tp = tcp_close(tp); 1417 goto findpcb; 1418 } 1419 /* 1420 * If window is closed can only take segments at 1421 * window edge, and have to drop data and PUSH from 1422 * incoming segments. Continue processing, but 1423 * remember to ack. Otherwise, drop segment 1424 * and ack. 1425 */ 1426 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1427 tp->t_flags |= TF_ACKNOW; 1428 tcpstat.tcps_rcvwinprobe++; 1429 } else 1430 goto dropafterack; 1431 } else 1432 tcpstat.tcps_rcvbyteafterwin += todrop; 1433 m_adj(m, -todrop); 1434 tlen -= todrop; 1435 tiflags &= ~(TH_PUSH|TH_FIN); 1436 } 1437 1438 /* 1439 * If last ACK falls within this segment's sequence numbers, 1440 * record its timestamp. 1441 * Fix from Braden, see Stevens p. 870 1442 */ 1443 if (ts_present && TSTMP_GEQ(ts_val, tp->ts_recent) && 1444 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1445 tp->ts_recent_age = tcp_now; 1446 tp->ts_recent = ts_val; 1447 } 1448 1449 /* 1450 * If the RST bit is set examine the state: 1451 * SYN_RECEIVED STATE: 1452 * If passive open, return to LISTEN state. 1453 * If active open, inform user that connection was refused. 1454 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1455 * Inform user that connection was reset, and close tcb. 1456 * CLOSING, LAST_ACK, TIME_WAIT STATES 1457 * Close the tcb. 1458 */ 1459 if (tiflags & TH_RST) { 1460 if (th->th_seq != tp->last_ack_sent) 1461 goto drop; 1462 1463 switch (tp->t_state) { 1464 case TCPS_SYN_RECEIVED: 1465 so->so_error = ECONNREFUSED; 1466 goto close; 1467 1468 case TCPS_ESTABLISHED: 1469 case TCPS_FIN_WAIT_1: 1470 case TCPS_FIN_WAIT_2: 1471 case TCPS_CLOSE_WAIT: 1472 so->so_error = ECONNRESET; 1473 close: 1474 tp->t_state = TCPS_CLOSED; 1475 tcpstat.tcps_drops++; 1476 tp = tcp_close(tp); 1477 goto drop; 1478 case TCPS_CLOSING: 1479 case TCPS_LAST_ACK: 1480 case TCPS_TIME_WAIT: 1481 tp = tcp_close(tp); 1482 goto drop; 1483 } 1484 } 1485 1486 /* 1487 * If a SYN is in the window, then this is an 1488 * error and we send an RST and drop the connection. 1489 */ 1490 if (tiflags & TH_SYN) { 1491 tp = tcp_drop(tp, ECONNRESET); 1492 goto dropwithreset; 1493 } 1494 1495 /* 1496 * If the ACK bit is off we drop the segment and return. 1497 */ 1498 if ((tiflags & TH_ACK) == 0) { 1499 if (tp->t_flags & TF_ACKNOW) 1500 goto dropafterack; 1501 else 1502 goto drop; 1503 } 1504 1505 /* 1506 * Ack processing. 1507 */ 1508 switch (tp->t_state) { 1509 1510 /* 1511 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1512 * ESTABLISHED state and continue processing. 1513 * The ACK was checked above. 1514 */ 1515 case TCPS_SYN_RECEIVED: 1516 tcpstat.tcps_connects++; 1517 soisconnected(so); 1518 tp->t_state = TCPS_ESTABLISHED; 1519 /* Do window scaling? */ 1520 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1521 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1522 tp->snd_scale = tp->requested_s_scale; 1523 tp->rcv_scale = tp->request_r_scale; 1524 } 1525 (void) tcp_reass(tp, (struct tcphdr *)0, (struct mbuf *)0, 1526 &tlen); 1527 tp->snd_wl1 = th->th_seq - 1; 1528 /* fall into ... */ 1529 1530 /* 1531 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1532 * ACKs. If the ack is in the range 1533 * tp->snd_una < th->th_ack <= tp->snd_max 1534 * then advance tp->snd_una to th->th_ack and drop 1535 * data from the retransmission queue. If this ACK reflects 1536 * more up to date window information we update our window information. 1537 */ 1538 case TCPS_ESTABLISHED: 1539 case TCPS_FIN_WAIT_1: 1540 case TCPS_FIN_WAIT_2: 1541 case TCPS_CLOSE_WAIT: 1542 case TCPS_CLOSING: 1543 case TCPS_LAST_ACK: 1544 case TCPS_TIME_WAIT: 1545 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1546 /* 1547 * Duplicate/old ACK processing. 1548 * Increments t_dupacks: 1549 * Pure duplicate (same seq/ack/window, no data) 1550 * Doesn't affect t_dupacks: 1551 * Data packets. 1552 * Normal window updates (window opens) 1553 * Resets t_dupacks: 1554 * New data ACKed. 1555 * Window shrinks 1556 * Old ACK 1557 */ 1558 if (tlen) 1559 break; 1560 /* 1561 * If we get an old ACK, there is probably packet 1562 * reordering going on. Be conservative and reset 1563 * t_dupacks so that we are less agressive in 1564 * doing a fast retransmit. 1565 */ 1566 if (th->th_ack != tp->snd_una) { 1567 tp->t_dupacks = 0; 1568 break; 1569 } 1570 if (tiwin == tp->snd_wnd) { 1571 tcpstat.tcps_rcvdupack++; 1572 /* 1573 * If we have outstanding data (other than 1574 * a window probe), this is a completely 1575 * duplicate ack (ie, window info didn't 1576 * change), the ack is the biggest we've 1577 * seen and we've seen exactly our rexmt 1578 * threshhold of them, assume a packet 1579 * has been dropped and retransmit it. 1580 * Kludge snd_nxt & the congestion 1581 * window so we send only this one 1582 * packet. 1583 * 1584 * We know we're losing at the current 1585 * window size so do congestion avoidance 1586 * (set ssthresh to half the current window 1587 * and pull our congestion window back to 1588 * the new ssthresh). 1589 * 1590 * Dup acks mean that packets have left the 1591 * network (they're now cached at the receiver) 1592 * so bump cwnd by the amount in the receiver 1593 * to keep a constant cwnd packets in the 1594 * network. 1595 */ 1596 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0) 1597 tp->t_dupacks = 0; 1598 #if defined(TCP_SACK) && defined(TCP_FACK) 1599 /* 1600 * In FACK, can enter fast rec. if the receiver 1601 * reports a reass. queue longer than 3 segs. 1602 */ 1603 else if (++tp->t_dupacks == tcprexmtthresh || 1604 ((SEQ_GT(tp->snd_fack, tcprexmtthresh * 1605 tp->t_maxseg + tp->snd_una)) && 1606 SEQ_GT(tp->snd_una, tp->snd_last))) { 1607 #else 1608 else if (++tp->t_dupacks == tcprexmtthresh) { 1609 #endif /* TCP_FACK */ 1610 tcp_seq onxt = tp->snd_nxt; 1611 u_long win = 1612 ulmin(tp->snd_wnd, tp->snd_cwnd) / 1613 2 / tp->t_maxseg; 1614 1615 #if defined(TCP_SACK) 1616 if (SEQ_LT(th->th_ack, tp->snd_last)){ 1617 /* 1618 * False fast retx after 1619 * timeout. Do not cut window. 1620 */ 1621 tp->t_dupacks = 0; 1622 goto drop; 1623 } 1624 #endif 1625 if (win < 2) 1626 win = 2; 1627 tp->snd_ssthresh = win * tp->t_maxseg; 1628 #if defined(TCP_SACK) 1629 tp->snd_last = tp->snd_max; 1630 #endif 1631 #ifdef TCP_SACK 1632 if (!tp->sack_disable) { 1633 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1634 tp->t_rtt = 0; 1635 tcpstat.tcps_sndrexmitfast++; 1636 #if defined(TCP_SACK) && defined(TCP_FACK) 1637 tp->t_dupacks = tcprexmtthresh; 1638 (void) tcp_output(tp); 1639 /* 1640 * During FR, snd_cwnd is held 1641 * constant for FACK. 1642 */ 1643 tp->snd_cwnd = tp->snd_ssthresh; 1644 #else 1645 /* 1646 * tcp_output() will send 1647 * oldest SACK-eligible rtx. 1648 */ 1649 (void) tcp_output(tp); 1650 tp->snd_cwnd = tp->snd_ssthresh+ 1651 tp->t_maxseg * tp->t_dupacks; 1652 #endif /* TCP_FACK */ 1653 goto drop; 1654 } 1655 #endif /* TCP_SACK */ 1656 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1657 tp->t_rtt = 0; 1658 tp->snd_nxt = th->th_ack; 1659 tp->snd_cwnd = tp->t_maxseg; 1660 tcpstat.tcps_sndrexmitfast++; 1661 (void) tcp_output(tp); 1662 1663 tp->snd_cwnd = tp->snd_ssthresh + 1664 tp->t_maxseg * tp->t_dupacks; 1665 if (SEQ_GT(onxt, tp->snd_nxt)) 1666 tp->snd_nxt = onxt; 1667 goto drop; 1668 } else if (tp->t_dupacks > tcprexmtthresh) { 1669 #if defined(TCP_SACK) && defined(TCP_FACK) 1670 /* 1671 * while (awnd < cwnd) 1672 * sendsomething(); 1673 */ 1674 if (!tp->sack_disable) { 1675 if (tp->snd_awnd < tp->snd_cwnd) 1676 tcp_output(tp); 1677 goto drop; 1678 } 1679 #endif /* TCP_FACK */ 1680 tp->snd_cwnd += tp->t_maxseg; 1681 (void) tcp_output(tp); 1682 goto drop; 1683 } 1684 } else if (tiwin < tp->snd_wnd) { 1685 /* 1686 * The window was retracted! Previous dup 1687 * ACKs may have been due to packets arriving 1688 * after the shrunken window, not a missing 1689 * packet, so play it safe and reset t_dupacks 1690 */ 1691 tp->t_dupacks = 0; 1692 } 1693 break; 1694 } 1695 /* 1696 * If the congestion window was inflated to account 1697 * for the other side's cached packets, retract it. 1698 */ 1699 #if defined(TCP_SACK) 1700 if (!tp->sack_disable) { 1701 if (tp->t_dupacks >= tcprexmtthresh) { 1702 /* Check for a partial ACK */ 1703 if (tcp_sack_partialack(tp, th)) { 1704 #if defined(TCP_SACK) && defined(TCP_FACK) 1705 /* Force call to tcp_output */ 1706 if (tp->snd_awnd < tp->snd_cwnd) 1707 needoutput = 1; 1708 #else 1709 tp->snd_cwnd += tp->t_maxseg; 1710 needoutput = 1; 1711 #endif /* TCP_FACK */ 1712 } else { 1713 /* Out of fast recovery */ 1714 tp->snd_cwnd = tp->snd_ssthresh; 1715 if (tcp_seq_subtract(tp->snd_max, 1716 th->th_ack) < tp->snd_ssthresh) 1717 tp->snd_cwnd = 1718 tcp_seq_subtract(tp->snd_max, 1719 th->th_ack); 1720 tp->t_dupacks = 0; 1721 #if defined(TCP_SACK) && defined(TCP_FACK) 1722 if (SEQ_GT(th->th_ack, tp->snd_fack)) 1723 tp->snd_fack = th->th_ack; 1724 #endif /* TCP_FACK */ 1725 } 1726 } 1727 } else { 1728 if (tp->t_dupacks >= tcprexmtthresh && 1729 !tcp_newreno(tp, th)) { 1730 /* Out of fast recovery */ 1731 tp->snd_cwnd = tp->snd_ssthresh; 1732 if (tcp_seq_subtract(tp->snd_max, th->th_ack) < 1733 tp->snd_ssthresh) 1734 tp->snd_cwnd = 1735 tcp_seq_subtract(tp->snd_max, 1736 th->th_ack); 1737 tp->t_dupacks = 0; 1738 } 1739 } 1740 if (tp->t_dupacks < tcprexmtthresh) 1741 tp->t_dupacks = 0; 1742 #else /* else no TCP_SACK */ 1743 if (tp->t_dupacks >= tcprexmtthresh && 1744 tp->snd_cwnd > tp->snd_ssthresh) 1745 tp->snd_cwnd = tp->snd_ssthresh; 1746 tp->t_dupacks = 0; 1747 #endif 1748 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1749 tcpstat.tcps_rcvacktoomuch++; 1750 goto dropafterack; 1751 } 1752 acked = th->th_ack - tp->snd_una; 1753 tcpstat.tcps_rcvackpack++; 1754 tcpstat.tcps_rcvackbyte += acked; 1755 1756 /* 1757 * If we have a timestamp reply, update smoothed 1758 * round trip time. If no timestamp is present but 1759 * transmit timer is running and timed sequence 1760 * number was acked, update smoothed round trip time. 1761 * Since we now have an rtt measurement, cancel the 1762 * timer backoff (cf., Phil Karn's retransmit alg.). 1763 * Recompute the initial retransmit timer. 1764 */ 1765 if (ts_present) 1766 tcp_xmit_timer(tp, tcp_now-ts_ecr+1); 1767 else if (tp->t_rtt && SEQ_GT(th->th_ack, tp->t_rtseq)) 1768 tcp_xmit_timer(tp,tp->t_rtt); 1769 1770 /* 1771 * If all outstanding data is acked, stop retransmit 1772 * timer and remember to restart (more output or persist). 1773 * If there is more data to be acked, restart retransmit 1774 * timer, using current (possibly backed-off) value. 1775 */ 1776 if (th->th_ack == tp->snd_max) { 1777 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1778 needoutput = 1; 1779 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1780 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1781 /* 1782 * When new data is acked, open the congestion window. 1783 * If the window gives us less than ssthresh packets 1784 * in flight, open exponentially (maxseg per packet). 1785 * Otherwise open linearly: maxseg per window 1786 * (maxseg^2 / cwnd per packet). 1787 */ 1788 { 1789 u_int cw = tp->snd_cwnd; 1790 u_int incr = tp->t_maxseg; 1791 1792 if (cw > tp->snd_ssthresh) 1793 incr = incr * incr / cw; 1794 #if defined (TCP_SACK) 1795 if (tp->t_dupacks < tcprexmtthresh) 1796 #endif 1797 tp->snd_cwnd = ulmin(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1798 } 1799 ND6_HINT(tp); 1800 if (acked > so->so_snd.sb_cc) { 1801 tp->snd_wnd -= so->so_snd.sb_cc; 1802 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1803 ourfinisacked = 1; 1804 } else { 1805 sbdrop(&so->so_snd, acked); 1806 tp->snd_wnd -= acked; 1807 ourfinisacked = 0; 1808 } 1809 if (sb_notify(&so->so_snd)) 1810 sowwakeup(so); 1811 tp->snd_una = th->th_ack; 1812 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1813 tp->snd_nxt = tp->snd_una; 1814 #if defined (TCP_SACK) && defined (TCP_FACK) 1815 if (SEQ_GT(tp->snd_una, tp->snd_fack)) { 1816 tp->snd_fack = tp->snd_una; 1817 /* Update snd_awnd for partial ACK 1818 * without any SACK blocks. 1819 */ 1820 tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, 1821 tp->snd_fack) + tp->retran_data; 1822 } 1823 #endif 1824 1825 switch (tp->t_state) { 1826 1827 /* 1828 * In FIN_WAIT_1 STATE in addition to the processing 1829 * for the ESTABLISHED state if our FIN is now acknowledged 1830 * then enter FIN_WAIT_2. 1831 */ 1832 case TCPS_FIN_WAIT_1: 1833 if (ourfinisacked) { 1834 /* 1835 * If we can't receive any more 1836 * data, then closing user can proceed. 1837 * Starting the timer is contrary to the 1838 * specification, but if we don't get a FIN 1839 * we'll hang forever. 1840 */ 1841 if (so->so_state & SS_CANTRCVMORE) { 1842 soisdisconnected(so); 1843 TCP_TIMER_ARM(tp, TCPT_2MSL, tcp_maxidle); 1844 } 1845 tp->t_state = TCPS_FIN_WAIT_2; 1846 } 1847 break; 1848 1849 /* 1850 * In CLOSING STATE in addition to the processing for 1851 * the ESTABLISHED state if the ACK acknowledges our FIN 1852 * then enter the TIME-WAIT state, otherwise ignore 1853 * the segment. 1854 */ 1855 case TCPS_CLOSING: 1856 if (ourfinisacked) { 1857 tp->t_state = TCPS_TIME_WAIT; 1858 tcp_canceltimers(tp); 1859 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1860 soisdisconnected(so); 1861 } 1862 break; 1863 1864 /* 1865 * In LAST_ACK, we may still be waiting for data to drain 1866 * and/or to be acked, as well as for the ack of our FIN. 1867 * If our FIN is now acknowledged, delete the TCB, 1868 * enter the closed state and return. 1869 */ 1870 case TCPS_LAST_ACK: 1871 if (ourfinisacked) { 1872 tp = tcp_close(tp); 1873 goto drop; 1874 } 1875 break; 1876 1877 /* 1878 * In TIME_WAIT state the only thing that should arrive 1879 * is a retransmission of the remote FIN. Acknowledge 1880 * it and restart the finack timer. 1881 */ 1882 case TCPS_TIME_WAIT: 1883 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 1884 goto dropafterack; 1885 } 1886 } 1887 1888 step6: 1889 /* 1890 * Update window information. 1891 * Don't look at window if no ACK: TAC's send garbage on first SYN. 1892 */ 1893 if ((tiflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || 1894 (tp->snd_wl1 == th->th_seq && SEQ_LT(tp->snd_wl2, th->th_ack)) || 1895 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))) { 1896 /* keep track of pure window updates */ 1897 if (tlen == 0 && 1898 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 1899 tcpstat.tcps_rcvwinupd++; 1900 tp->snd_wnd = tiwin; 1901 tp->snd_wl1 = th->th_seq; 1902 tp->snd_wl2 = th->th_ack; 1903 if (tp->snd_wnd > tp->max_sndwnd) 1904 tp->max_sndwnd = tp->snd_wnd; 1905 needoutput = 1; 1906 } 1907 1908 /* 1909 * Process segments with URG. 1910 */ 1911 if ((tiflags & TH_URG) && th->th_urp && 1912 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1913 /* 1914 * This is a kludge, but if we receive and accept 1915 * random urgent pointers, we'll crash in 1916 * soreceive. It's hard to imagine someone 1917 * actually wanting to send this much urgent data. 1918 */ 1919 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 1920 th->th_urp = 0; /* XXX */ 1921 tiflags &= ~TH_URG; /* XXX */ 1922 goto dodata; /* XXX */ 1923 } 1924 /* 1925 * If this segment advances the known urgent pointer, 1926 * then mark the data stream. This should not happen 1927 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 1928 * a FIN has been received from the remote side. 1929 * In these states we ignore the URG. 1930 * 1931 * According to RFC961 (Assigned Protocols), 1932 * the urgent pointer points to the last octet 1933 * of urgent data. We continue, however, 1934 * to consider it to indicate the first octet 1935 * of data past the urgent section as the original 1936 * spec states (in one of two places). 1937 */ 1938 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 1939 tp->rcv_up = th->th_seq + th->th_urp; 1940 so->so_oobmark = so->so_rcv.sb_cc + 1941 (tp->rcv_up - tp->rcv_nxt) - 1; 1942 if (so->so_oobmark == 0) 1943 so->so_state |= SS_RCVATMARK; 1944 sohasoutofband(so); 1945 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 1946 } 1947 /* 1948 * Remove out of band data so doesn't get presented to user. 1949 * This can happen independent of advancing the URG pointer, 1950 * but if two URG's are pending at once, some out-of-band 1951 * data may creep in... ick. 1952 */ 1953 if (th->th_urp <= (u_int16_t) tlen 1954 #ifdef SO_OOBINLINE 1955 && (so->so_options & SO_OOBINLINE) == 0 1956 #endif 1957 ) 1958 tcp_pulloutofband(so, th->th_urp, m, hdroptlen); 1959 } else 1960 /* 1961 * If no out of band data is expected, 1962 * pull receive urgent pointer along 1963 * with the receive window. 1964 */ 1965 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 1966 tp->rcv_up = tp->rcv_nxt; 1967 dodata: /* XXX */ 1968 1969 /* 1970 * Process the segment text, merging it into the TCP sequencing queue, 1971 * and arranging for acknowledgment of receipt if necessary. 1972 * This process logically involves adjusting tp->rcv_wnd as data 1973 * is presented to the user (this happens in tcp_usrreq.c, 1974 * case PRU_RCVD). If a FIN has already been received on this 1975 * connection then we just ignore the text. 1976 */ 1977 if ((tlen || (tiflags & TH_FIN)) && 1978 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 1979 if (th->th_seq == tp->rcv_nxt && tp->segq.lh_first == NULL && 1980 tp->t_state == TCPS_ESTABLISHED) { 1981 if (th->th_flags & TH_PUSH) 1982 tp->t_flags |= TF_ACKNOW; 1983 else 1984 tp->t_flags |= TF_DELACK; 1985 tp->rcv_nxt += tlen; 1986 tiflags = th->th_flags & TH_FIN; 1987 tcpstat.tcps_rcvpack++; 1988 tcpstat.tcps_rcvbyte += tlen; 1989 ND6_HINT(tp); 1990 m_adj(m, hdroptlen); 1991 sbappend(&so->so_rcv, m); 1992 sorwakeup(so); 1993 } else { 1994 m_adj(m, hdroptlen); 1995 tiflags = tcp_reass(tp, th, m, &tlen); 1996 tp->t_flags |= TF_ACKNOW; 1997 } 1998 #ifdef TCP_SACK 1999 if (!tp->sack_disable) 2000 tcp_update_sack_list(tp); 2001 #endif 2002 2003 /* 2004 * variable len never referenced again in modern BSD, 2005 * so why bother computing it ?? 2006 */ 2007 #if 0 2008 /* 2009 * Note the amount of data that peer has sent into 2010 * our window, in order to estimate the sender's 2011 * buffer size. 2012 */ 2013 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2014 #endif /* 0 */ 2015 } else { 2016 m_freem(m); 2017 tiflags &= ~TH_FIN; 2018 } 2019 2020 /* 2021 * If FIN is received ACK the FIN and let the user know 2022 * that the connection is closing. Ignore a FIN received before 2023 * the connection is fully established. 2024 */ 2025 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2026 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2027 socantrcvmore(so); 2028 tp->t_flags |= TF_ACKNOW; 2029 tp->rcv_nxt++; 2030 } 2031 switch (tp->t_state) { 2032 2033 /* 2034 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2035 */ 2036 case TCPS_ESTABLISHED: 2037 tp->t_state = TCPS_CLOSE_WAIT; 2038 break; 2039 2040 /* 2041 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2042 * enter the CLOSING state. 2043 */ 2044 case TCPS_FIN_WAIT_1: 2045 tp->t_state = TCPS_CLOSING; 2046 break; 2047 2048 /* 2049 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2050 * starting the time-wait timer, turning off the other 2051 * standard timers. 2052 */ 2053 case TCPS_FIN_WAIT_2: 2054 tp->t_state = TCPS_TIME_WAIT; 2055 tcp_canceltimers(tp); 2056 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2057 soisdisconnected(so); 2058 break; 2059 2060 /* 2061 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2062 */ 2063 case TCPS_TIME_WAIT: 2064 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2065 break; 2066 } 2067 } 2068 if (so->so_options & SO_DEBUG) { 2069 switch (tp->pf == PF_INET6) { 2070 #ifdef INET6 2071 case PF_INET6: 2072 tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti6, 2073 0, tlen); 2074 break; 2075 #endif /* INET6 */ 2076 case PF_INET: 2077 tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti, 2078 0, tlen); 2079 break; 2080 } 2081 } 2082 2083 /* 2084 * Return any desired output. 2085 */ 2086 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2087 (void) tcp_output(tp); 2088 } 2089 return; 2090 2091 dropafterack: 2092 /* 2093 * Generate an ACK dropping incoming segment if it occupies 2094 * sequence space, where the ACK reflects our state. 2095 */ 2096 if (tiflags & TH_RST) 2097 goto drop; 2098 m_freem(m); 2099 tp->t_flags |= TF_ACKNOW; 2100 (void) tcp_output(tp); 2101 return; 2102 2103 dropwithreset_ratelim: 2104 /* 2105 * We may want to rate-limit RSTs in certain situations, 2106 * particularly if we are sending an RST in response to 2107 * an attempt to connect to or otherwise communicate with 2108 * a port for which we have no socket. 2109 */ 2110 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2111 tcp_rst_ppslim) == 0) { 2112 /* XXX stat */ 2113 goto drop; 2114 } 2115 /* ...fall into dropwithreset... */ 2116 2117 dropwithreset: 2118 /* 2119 * Generate a RST, dropping incoming segment. 2120 * Make ACK acceptable to originator of segment. 2121 * Don't bother to respond if destination was broadcast/multicast. 2122 */ 2123 if ((tiflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2124 goto drop; 2125 switch (af) { 2126 #ifdef INET6 2127 case AF_INET6: 2128 /* For following calls to tcp_respond */ 2129 if (IN6_IS_ADDR_MULTICAST(&ipv6->ip6_dst)) 2130 goto drop; 2131 break; 2132 #endif /* INET6 */ 2133 case AF_INET: 2134 if (IN_MULTICAST(ip->ip_dst.s_addr)) 2135 goto drop; 2136 } 2137 if (tiflags & TH_ACK) { 2138 tcp_respond(tp, mtod(m, caddr_t), m, (tcp_seq)0, th->th_ack, 2139 TH_RST); 2140 } else { 2141 if (tiflags & TH_SYN) 2142 tlen++; 2143 tcp_respond(tp, mtod(m, caddr_t), m, th->th_seq + tlen, 2144 (tcp_seq)0, TH_RST|TH_ACK); 2145 } 2146 /* destroy temporarily created socket */ 2147 if (dropsocket) 2148 (void) soabort(so); 2149 return; 2150 2151 drop: 2152 /* 2153 * Drop space held by incoming segment and return. 2154 */ 2155 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) { 2156 switch (tp->pf) { 2157 #ifdef INET6 2158 case PF_INET6: 2159 tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti6, 2160 0, tlen); 2161 break; 2162 #endif /* INET6 */ 2163 case PF_INET: 2164 tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti, 2165 0, tlen); 2166 break; 2167 } 2168 } 2169 2170 m_freem(m); 2171 /* destroy temporarily created socket */ 2172 if (dropsocket) 2173 (void) soabort(so); 2174 return; 2175 #ifndef TUBA_INCLUDE 2176 } 2177 2178 void 2179 tcp_dooptions(tp, cp, cnt, th, ts_present, ts_val, ts_ecr) 2180 struct tcpcb *tp; 2181 u_char *cp; 2182 int cnt; 2183 struct tcphdr *th; 2184 int *ts_present; 2185 u_int32_t *ts_val, *ts_ecr; 2186 { 2187 u_int16_t mss = 0; 2188 int opt, optlen; 2189 2190 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2191 opt = cp[0]; 2192 if (opt == TCPOPT_EOL) 2193 break; 2194 if (opt == TCPOPT_NOP) 2195 optlen = 1; 2196 else { 2197 if (cnt < 2) 2198 break; 2199 optlen = cp[1]; 2200 if (optlen < 2 || optlen > cnt) 2201 break; 2202 } 2203 switch (opt) { 2204 2205 default: 2206 continue; 2207 2208 case TCPOPT_MAXSEG: 2209 if (optlen != TCPOLEN_MAXSEG) 2210 continue; 2211 if (!(th->th_flags & TH_SYN)) 2212 continue; 2213 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 2214 NTOHS(mss); 2215 break; 2216 2217 case TCPOPT_WINDOW: 2218 if (optlen != TCPOLEN_WINDOW) 2219 continue; 2220 if (!(th->th_flags & TH_SYN)) 2221 continue; 2222 tp->t_flags |= TF_RCVD_SCALE; 2223 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2224 break; 2225 2226 case TCPOPT_TIMESTAMP: 2227 if (optlen != TCPOLEN_TIMESTAMP) 2228 continue; 2229 *ts_present = 1; 2230 bcopy((char *)cp + 2, (char *) ts_val, sizeof(*ts_val)); 2231 NTOHL(*ts_val); 2232 bcopy((char *)cp + 6, (char *) ts_ecr, sizeof(*ts_ecr)); 2233 NTOHL(*ts_ecr); 2234 2235 /* 2236 * A timestamp received in a SYN makes 2237 * it ok to send timestamp requests and replies. 2238 */ 2239 if (th->th_flags & TH_SYN) { 2240 tp->t_flags |= TF_RCVD_TSTMP; 2241 tp->ts_recent = *ts_val; 2242 tp->ts_recent_age = tcp_now; 2243 } 2244 break; 2245 2246 #ifdef TCP_SACK 2247 case TCPOPT_SACK_PERMITTED: 2248 if (tp->sack_disable || optlen!=TCPOLEN_SACK_PERMITTED) 2249 continue; 2250 if (th->th_flags & TH_SYN) 2251 /* MUST only be set on SYN */ 2252 tp->t_flags |= TF_SACK_PERMIT; 2253 break; 2254 case TCPOPT_SACK: 2255 if (tcp_sack_option(tp, th, cp, optlen)) 2256 continue; 2257 break; 2258 #endif 2259 } 2260 } 2261 /* Update t_maxopd and t_maxseg after all options are processed */ 2262 if (th->th_flags & TH_SYN) { 2263 (void) tcp_mss(tp, mss); /* sets t_maxseg */ 2264 2265 if (mss) 2266 tcp_mss_update(tp); 2267 } 2268 } 2269 2270 #if defined(TCP_SACK) 2271 u_long 2272 tcp_seq_subtract(a, b) 2273 u_long a, b; 2274 { 2275 return ((long)(a - b)); 2276 } 2277 #endif 2278 2279 2280 #ifdef TCP_SACK 2281 /* 2282 * This function is called upon receipt of new valid data (while not in header 2283 * prediction mode), and it updates the ordered list of sacks. 2284 */ 2285 void 2286 tcp_update_sack_list(tp) 2287 struct tcpcb *tp; 2288 { 2289 /* 2290 * First reported block MUST be the most recent one. Subsequent 2291 * blocks SHOULD be in the order in which they arrived at the 2292 * receiver. These two conditions make the implementation fully 2293 * compliant with RFC 2018. 2294 */ 2295 int i, j = 0, count = 0, lastpos = -1; 2296 struct sackblk sack, firstsack, temp[MAX_SACK_BLKS]; 2297 2298 /* First clean up current list of sacks */ 2299 for (i = 0; i < tp->rcv_numsacks; i++) { 2300 sack = tp->sackblks[i]; 2301 if (sack.start == 0 && sack.end == 0) { 2302 count++; /* count = number of blocks to be discarded */ 2303 continue; 2304 } 2305 if (SEQ_LEQ(sack.end, tp->rcv_nxt)) { 2306 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2307 count++; 2308 } else { 2309 temp[j].start = tp->sackblks[i].start; 2310 temp[j++].end = tp->sackblks[i].end; 2311 } 2312 } 2313 tp->rcv_numsacks -= count; 2314 if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */ 2315 tcp_clean_sackreport(tp); 2316 if (SEQ_LT(tp->rcv_nxt, tp->rcv_laststart)) { 2317 /* ==> need first sack block */ 2318 tp->sackblks[0].start = tp->rcv_laststart; 2319 tp->sackblks[0].end = tp->rcv_lastend; 2320 tp->rcv_numsacks = 1; 2321 } 2322 return; 2323 } 2324 /* Otherwise, sack blocks are already present. */ 2325 for (i = 0; i < tp->rcv_numsacks; i++) 2326 tp->sackblks[i] = temp[i]; /* first copy back sack list */ 2327 if (SEQ_GEQ(tp->rcv_nxt, tp->rcv_lastend)) 2328 return; /* sack list remains unchanged */ 2329 /* 2330 * From here, segment just received should be (part of) the 1st sack. 2331 * Go through list, possibly coalescing sack block entries. 2332 */ 2333 firstsack.start = tp->rcv_laststart; 2334 firstsack.end = tp->rcv_lastend; 2335 for (i = 0; i < tp->rcv_numsacks; i++) { 2336 sack = tp->sackblks[i]; 2337 if (SEQ_LT(sack.end, firstsack.start) || 2338 SEQ_GT(sack.start, firstsack.end)) 2339 continue; /* no overlap */ 2340 if (sack.start == firstsack.start && sack.end == firstsack.end){ 2341 /* 2342 * identical block; delete it here since we will 2343 * move it to the front of the list. 2344 */ 2345 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2346 lastpos = i; /* last posn with a zero entry */ 2347 continue; 2348 } 2349 if (SEQ_LEQ(sack.start, firstsack.start)) 2350 firstsack.start = sack.start; /* merge blocks */ 2351 if (SEQ_GEQ(sack.end, firstsack.end)) 2352 firstsack.end = sack.end; /* merge blocks */ 2353 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2354 lastpos = i; /* last posn with a zero entry */ 2355 } 2356 if (lastpos != -1) { /* at least one merge */ 2357 for (i = 0, j = 1; i < tp->rcv_numsacks; i++) { 2358 sack = tp->sackblks[i]; 2359 if (sack.start == 0 && sack.end == 0) 2360 continue; 2361 temp[j++] = sack; 2362 } 2363 tp->rcv_numsacks = j; /* including first blk (added later) */ 2364 for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */ 2365 tp->sackblks[i] = temp[i]; 2366 } else { /* no merges -- shift sacks by 1 */ 2367 if (tp->rcv_numsacks < MAX_SACK_BLKS) 2368 tp->rcv_numsacks++; 2369 for (i = tp->rcv_numsacks-1; i > 0; i--) 2370 tp->sackblks[i] = tp->sackblks[i-1]; 2371 } 2372 tp->sackblks[0] = firstsack; 2373 return; 2374 } 2375 2376 /* 2377 * Process the TCP SACK option. Returns 1 if tcp_dooptions() should continue, 2378 * and 0 otherwise, if the option was fine. tp->snd_holes is an ordered list 2379 * of holes (oldest to newest, in terms of the sequence space). 2380 */ 2381 int 2382 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen) 2383 { 2384 int tmp_olen; 2385 u_char *tmp_cp; 2386 struct sackhole *cur, *p, *temp; 2387 2388 if (tp->sack_disable) 2389 return (1); 2390 2391 /* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */ 2392 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2393 return (1); 2394 tmp_cp = cp + 2; 2395 tmp_olen = optlen - 2; 2396 if (tp->snd_numholes < 0) 2397 tp->snd_numholes = 0; 2398 if (tp->t_maxseg == 0) 2399 panic("tcp_sack_option"); /* Should never happen */ 2400 while (tmp_olen > 0) { 2401 struct sackblk sack; 2402 2403 bcopy(tmp_cp, (char *) &(sack.start), sizeof(tcp_seq)); 2404 NTOHL(sack.start); 2405 bcopy(tmp_cp + sizeof(tcp_seq), 2406 (char *) &(sack.end), sizeof(tcp_seq)); 2407 NTOHL(sack.end); 2408 tmp_olen -= TCPOLEN_SACK; 2409 tmp_cp += TCPOLEN_SACK; 2410 if (SEQ_LEQ(sack.end, sack.start)) 2411 continue; /* bad SACK fields */ 2412 if (SEQ_LEQ(sack.end, tp->snd_una)) 2413 continue; /* old block */ 2414 #if defined(TCP_SACK) && defined(TCP_FACK) 2415 /* Updates snd_fack. */ 2416 if (SEQ_GT(sack.end, tp->snd_fack)) 2417 tp->snd_fack = sack.end; 2418 #endif /* TCP_FACK */ 2419 if (SEQ_GT(th->th_ack, tp->snd_una)) { 2420 if (SEQ_LT(sack.start, th->th_ack)) 2421 continue; 2422 } 2423 if (SEQ_GT(sack.end, tp->snd_max)) 2424 continue; 2425 if (tp->snd_holes == NULL) { /* first hole */ 2426 tp->snd_holes = (struct sackhole *) 2427 pool_get(&sackhl_pool, PR_NOWAIT); 2428 if (tp->snd_holes == NULL) { 2429 /* ENOBUFS, so ignore SACKed block for now*/ 2430 continue; 2431 } 2432 cur = tp->snd_holes; 2433 cur->start = th->th_ack; 2434 cur->end = sack.start; 2435 cur->rxmit = cur->start; 2436 cur->next = NULL; 2437 tp->snd_numholes = 1; 2438 tp->rcv_lastsack = sack.end; 2439 /* 2440 * dups is at least one. If more data has been 2441 * SACKed, it can be greater than one. 2442 */ 2443 cur->dups = min(tcprexmtthresh, 2444 ((sack.end - cur->end)/tp->t_maxseg)); 2445 if (cur->dups < 1) 2446 cur->dups = 1; 2447 continue; /* with next sack block */ 2448 } 2449 /* Go thru list of holes: p = previous, cur = current */ 2450 p = cur = tp->snd_holes; 2451 while (cur) { 2452 if (SEQ_LEQ(sack.end, cur->start)) 2453 /* SACKs data before the current hole */ 2454 break; /* no use going through more holes */ 2455 if (SEQ_GEQ(sack.start, cur->end)) { 2456 /* SACKs data beyond the current hole */ 2457 cur->dups++; 2458 if (((sack.end - cur->end)/tp->t_maxseg) >= 2459 tcprexmtthresh) 2460 cur->dups = tcprexmtthresh; 2461 p = cur; 2462 cur = cur->next; 2463 continue; 2464 } 2465 if (SEQ_LEQ(sack.start, cur->start)) { 2466 /* Data acks at least the beginning of hole */ 2467 #if defined(TCP_SACK) && defined(TCP_FACK) 2468 if (SEQ_GT(sack.end, cur->rxmit)) 2469 tp->retran_data -= 2470 tcp_seq_subtract(cur->rxmit, 2471 cur->start); 2472 else 2473 tp->retran_data -= 2474 tcp_seq_subtract(sack.end, 2475 cur->start); 2476 #endif /* TCP_FACK */ 2477 if (SEQ_GEQ(sack.end, cur->end)) { 2478 /* Acks entire hole, so delete hole */ 2479 if (p != cur) { 2480 p->next = cur->next; 2481 pool_put(&sackhl_pool, cur); 2482 cur = p->next; 2483 } else { 2484 cur = cur->next; 2485 pool_put(&sackhl_pool, p); 2486 p = cur; 2487 tp->snd_holes = p; 2488 } 2489 tp->snd_numholes--; 2490 continue; 2491 } 2492 /* otherwise, move start of hole forward */ 2493 cur->start = sack.end; 2494 cur->rxmit = max (cur->rxmit, cur->start); 2495 p = cur; 2496 cur = cur->next; 2497 continue; 2498 } 2499 /* move end of hole backward */ 2500 if (SEQ_GEQ(sack.end, cur->end)) { 2501 #if defined(TCP_SACK) && defined(TCP_FACK) 2502 if (SEQ_GT(cur->rxmit, sack.start)) 2503 tp->retran_data -= 2504 tcp_seq_subtract(cur->rxmit, 2505 sack.start); 2506 #endif /* TCP_FACK */ 2507 cur->end = sack.start; 2508 cur->rxmit = min(cur->rxmit, cur->end); 2509 cur->dups++; 2510 if (((sack.end - cur->end)/tp->t_maxseg) >= 2511 tcprexmtthresh) 2512 cur->dups = tcprexmtthresh; 2513 p = cur; 2514 cur = cur->next; 2515 continue; 2516 } 2517 if (SEQ_LT(cur->start, sack.start) && 2518 SEQ_GT(cur->end, sack.end)) { 2519 /* 2520 * ACKs some data in middle of a hole; need to 2521 * split current hole 2522 */ 2523 temp = (struct sackhole *) 2524 pool_get(&sackhl_pool, PR_NOWAIT); 2525 if (temp == NULL) 2526 continue; /* ENOBUFS */ 2527 #if defined(TCP_SACK) && defined(TCP_FACK) 2528 if (SEQ_GT(cur->rxmit, sack.end)) 2529 tp->retran_data -= 2530 tcp_seq_subtract(sack.end, 2531 sack.start); 2532 else if (SEQ_GT(cur->rxmit, sack.start)) 2533 tp->retran_data -= 2534 tcp_seq_subtract(cur->rxmit, 2535 sack.start); 2536 #endif /* TCP_FACK */ 2537 temp->next = cur->next; 2538 temp->start = sack.end; 2539 temp->end = cur->end; 2540 temp->dups = cur->dups; 2541 temp->rxmit = max(cur->rxmit, temp->start); 2542 cur->end = sack.start; 2543 cur->rxmit = min(cur->rxmit, cur->end); 2544 cur->dups++; 2545 if (((sack.end - cur->end)/tp->t_maxseg) >= 2546 tcprexmtthresh) 2547 cur->dups = tcprexmtthresh; 2548 cur->next = temp; 2549 p = temp; 2550 cur = p->next; 2551 tp->snd_numholes++; 2552 } 2553 } 2554 /* At this point, p points to the last hole on the list */ 2555 if (SEQ_LT(tp->rcv_lastsack, sack.start)) { 2556 /* 2557 * Need to append new hole at end. 2558 * Last hole is p (and it's not NULL). 2559 */ 2560 temp = (struct sackhole *) 2561 pool_get(&sackhl_pool, PR_NOWAIT); 2562 if (temp == NULL) 2563 continue; /* ENOBUFS */ 2564 temp->start = tp->rcv_lastsack; 2565 temp->end = sack.start; 2566 temp->dups = min(tcprexmtthresh, 2567 ((sack.end - sack.start)/tp->t_maxseg)); 2568 if (temp->dups < 1) 2569 temp->dups = 1; 2570 temp->rxmit = temp->start; 2571 temp->next = 0; 2572 p->next = temp; 2573 tp->rcv_lastsack = sack.end; 2574 tp->snd_numholes++; 2575 } 2576 } 2577 #if defined(TCP_SACK) && defined(TCP_FACK) 2578 /* 2579 * Update retran_data and snd_awnd. Go through the list of 2580 * holes. Increment retran_data by (hole->rxmit - hole->start). 2581 */ 2582 tp->retran_data = 0; 2583 cur = tp->snd_holes; 2584 while (cur) { 2585 tp->retran_data += cur->rxmit - cur->start; 2586 cur = cur->next; 2587 } 2588 tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, tp->snd_fack) + 2589 tp->retran_data; 2590 #endif /* TCP_FACK */ 2591 2592 return (0); 2593 } 2594 2595 /* 2596 * Delete stale (i.e, cumulatively ack'd) holes. Hole is deleted only if 2597 * it is completely acked; otherwise, tcp_sack_option(), called from 2598 * tcp_dooptions(), will fix up the hole. 2599 */ 2600 void 2601 tcp_del_sackholes(tp, th) 2602 struct tcpcb *tp; 2603 struct tcphdr *th; 2604 { 2605 if (!tp->sack_disable && tp->t_state != TCPS_LISTEN) { 2606 /* max because this could be an older ack just arrived */ 2607 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 2608 th->th_ack : tp->snd_una; 2609 struct sackhole *cur = tp->snd_holes; 2610 struct sackhole *prev; 2611 while (cur) 2612 if (SEQ_LEQ(cur->end, lastack)) { 2613 prev = cur; 2614 cur = cur->next; 2615 pool_put(&sackhl_pool, prev); 2616 tp->snd_numholes--; 2617 } else if (SEQ_LT(cur->start, lastack)) { 2618 cur->start = lastack; 2619 if (SEQ_LT(cur->rxmit, cur->start)) 2620 cur->rxmit = cur->start; 2621 break; 2622 } else 2623 break; 2624 tp->snd_holes = cur; 2625 } 2626 } 2627 2628 /* 2629 * Delete all receiver-side SACK information. 2630 */ 2631 void 2632 tcp_clean_sackreport(tp) 2633 struct tcpcb *tp; 2634 { 2635 int i; 2636 2637 tp->rcv_numsacks = 0; 2638 for (i = 0; i < MAX_SACK_BLKS; i++) 2639 tp->sackblks[i].start = tp->sackblks[i].end=0; 2640 2641 } 2642 2643 /* 2644 * Checks for partial ack. If partial ack arrives, turn off retransmission 2645 * timer, deflate the window, do not clear tp->t_dupacks, and return 1. 2646 * If the ack advances at least to tp->snd_last, return 0. 2647 */ 2648 int 2649 tcp_sack_partialack(tp, th) 2650 struct tcpcb *tp; 2651 struct tcphdr *th; 2652 { 2653 if (SEQ_LT(th->th_ack, tp->snd_last)) { 2654 /* Turn off retx. timer (will start again next segment) */ 2655 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2656 tp->t_rtt = 0; 2657 #ifndef TCP_FACK 2658 /* 2659 * Partial window deflation. This statement relies on the 2660 * fact that tp->snd_una has not been updated yet. In FACK 2661 * hold snd_cwnd constant during fast recovery. 2662 */ 2663 if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) { 2664 tp->snd_cwnd -= th->th_ack - tp->snd_una; 2665 tp->snd_cwnd += tp->t_maxseg; 2666 } else 2667 tp->snd_cwnd = tp->t_maxseg; 2668 #endif 2669 return (1); 2670 } 2671 return (0); 2672 } 2673 #endif /* TCP_SACK */ 2674 2675 /* 2676 * Pull out of band byte out of a segment so 2677 * it doesn't appear in the user's data queue. 2678 * It is still reflected in the segment length for 2679 * sequencing purposes. 2680 */ 2681 void 2682 tcp_pulloutofband(so, urgent, m, off) 2683 struct socket *so; 2684 u_int urgent; 2685 register struct mbuf *m; 2686 int off; 2687 { 2688 int cnt = off + urgent - 1; 2689 2690 while (cnt >= 0) { 2691 if (m->m_len > cnt) { 2692 char *cp = mtod(m, caddr_t) + cnt; 2693 struct tcpcb *tp = sototcpcb(so); 2694 2695 tp->t_iobc = *cp; 2696 tp->t_oobflags |= TCPOOB_HAVEDATA; 2697 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2698 m->m_len--; 2699 return; 2700 } 2701 cnt -= m->m_len; 2702 m = m->m_next; 2703 if (m == 0) 2704 break; 2705 } 2706 panic("tcp_pulloutofband"); 2707 } 2708 2709 /* 2710 * Collect new round-trip time estimate 2711 * and update averages and current timeout. 2712 */ 2713 void 2714 tcp_xmit_timer(tp, rtt) 2715 register struct tcpcb *tp; 2716 short rtt; 2717 { 2718 register short delta; 2719 short rttmin; 2720 2721 tcpstat.tcps_rttupdated++; 2722 --rtt; 2723 if (tp->t_srtt != 0) { 2724 /* 2725 * srtt is stored as fixed point with 3 bits after the 2726 * binary point (i.e., scaled by 8). The following magic 2727 * is equivalent to the smoothing algorithm in rfc793 with 2728 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2729 * point). Adjust rtt to origin 0. 2730 */ 2731 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2732 if ((tp->t_srtt += delta) <= 0) 2733 tp->t_srtt = 1; 2734 /* 2735 * We accumulate a smoothed rtt variance (actually, a 2736 * smoothed mean difference), then set the retransmit 2737 * timer to smoothed rtt + 4 times the smoothed variance. 2738 * rttvar is stored as fixed point with 2 bits after the 2739 * binary point (scaled by 4). The following is 2740 * equivalent to rfc793 smoothing with an alpha of .75 2741 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2742 * rfc793's wired-in beta. 2743 */ 2744 if (delta < 0) 2745 delta = -delta; 2746 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2747 if ((tp->t_rttvar += delta) <= 0) 2748 tp->t_rttvar = 1; 2749 } else { 2750 /* 2751 * No rtt measurement yet - use the unsmoothed rtt. 2752 * Set the variance to half the rtt (so our first 2753 * retransmit happens at 3*rtt). 2754 */ 2755 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2756 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2757 } 2758 tp->t_rtt = 0; 2759 tp->t_rxtshift = 0; 2760 2761 /* 2762 * the retransmit should happen at rtt + 4 * rttvar. 2763 * Because of the way we do the smoothing, srtt and rttvar 2764 * will each average +1/2 tick of bias. When we compute 2765 * the retransmit timer, we want 1/2 tick of rounding and 2766 * 1 extra tick because of +-1/2 tick uncertainty in the 2767 * firing of the timer. The bias will give us exactly the 2768 * 1.5 tick we need. But, because the bias is 2769 * statistical, we have to test that we don't drop below 2770 * the minimum feasible timer (which is 2 ticks). 2771 */ 2772 if (tp->t_rttmin > rtt + 2) 2773 rttmin = tp->t_rttmin; 2774 else 2775 rttmin = rtt + 2; 2776 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2777 2778 /* 2779 * We received an ack for a packet that wasn't retransmitted; 2780 * it is probably safe to discard any error indications we've 2781 * received recently. This isn't quite right, but close enough 2782 * for now (a route might have failed after we sent a segment, 2783 * and the return path might not be symmetrical). 2784 */ 2785 tp->t_softerror = 0; 2786 } 2787 2788 /* 2789 * Determine a reasonable value for maxseg size. 2790 * If the route is known, check route for mtu. 2791 * If none, use an mss that can be handled on the outgoing 2792 * interface without forcing IP to fragment; if bigger than 2793 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2794 * to utilize large mbufs. If no route is found, route has no mtu, 2795 * or the destination isn't local, use a default, hopefully conservative 2796 * size (usually 512 or the default IP max size, but no more than the mtu 2797 * of the interface), as we can't discover anything about intervening 2798 * gateways or networks. We also initialize the congestion/slow start 2799 * window to be a single segment if the destination isn't local. 2800 * While looking at the routing entry, we also initialize other path-dependent 2801 * parameters from pre-set or cached values in the routing entry. 2802 * 2803 * Also take into account the space needed for options that we 2804 * send regularly. Make maxseg shorter by that amount to assure 2805 * that we can send maxseg amount of data even when the options 2806 * are present. Store the upper limit of the length of options plus 2807 * data in maxopd. 2808 * 2809 * NOTE: offer == -1 indicates that the maxseg size changed due to 2810 * Path MTU discovery. 2811 */ 2812 int 2813 tcp_mss(tp, offer) 2814 register struct tcpcb *tp; 2815 int offer; 2816 { 2817 struct rtentry *rt; 2818 struct ifnet *ifp; 2819 int mss, mssopt; 2820 int iphlen; 2821 int is_ipv6 = 0; 2822 struct inpcb *inp; 2823 2824 inp = tp->t_inpcb; 2825 2826 mssopt = mss = tcp_mssdflt; 2827 2828 rt = in_pcbrtentry(inp); 2829 2830 if (rt == NULL) 2831 goto out; 2832 2833 ifp = rt->rt_ifp; 2834 2835 switch (tp->pf) { 2836 #ifdef INET6 2837 case AF_INET6: 2838 iphlen = sizeof(struct ip6_hdr); 2839 is_ipv6 = 1; 2840 break; 2841 #endif 2842 case AF_INET: 2843 iphlen = sizeof(struct ip); 2844 break; 2845 default: 2846 /* the family does not support path MTU discovery */ 2847 goto out; 2848 } 2849 2850 #ifdef RTV_MTU 2851 /* 2852 * if there's an mtu associated with the route and we support 2853 * path MTU discovery for the underlying protocol family, use it. 2854 */ 2855 if (rt->rt_rmx.rmx_mtu) { 2856 /* 2857 * One may wish to lower MSS to take into account options, 2858 * especially security-related options. 2859 */ 2860 mss = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 2861 } else 2862 #endif /* RTV_MTU */ 2863 if (!ifp) 2864 /* 2865 * ifp may be null and rmx_mtu may be zero in certain 2866 * v6 cases (e.g., if ND wasn't able to resolve the 2867 * destination host. 2868 */ 2869 goto out; 2870 else if (ifp->if_flags & IFF_LOOPBACK) 2871 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2872 else if (!is_ipv6) { 2873 if (ip_mtudisc) 2874 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2875 else if (inp && in_localaddr(inp->inp_faddr)) 2876 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2877 } 2878 #ifdef INET6 2879 else if (is_ipv6) { 2880 if (inp && IN6_IS_ADDR_V4MAPPED(&inp->inp_faddr6)) { 2881 /* mapped addr case */ 2882 struct in_addr d; 2883 bcopy(&inp->inp_faddr6.s6_addr32[3], &d, sizeof(d)); 2884 if (ip_mtudisc || in_localaddr(d)) 2885 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2886 } else { 2887 /* 2888 * for IPv6, path MTU discovery is always turned on, 2889 * or the node must use packet size <= 1280. 2890 */ 2891 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2892 } 2893 } 2894 #endif /* INET6 */ 2895 2896 /* Calculate the value that we offer in TCPOPT_MAXSEG */ 2897 if (offer != -1) { 2898 mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 2899 mssopt = max(tcp_mssdflt, mssopt); 2900 } 2901 2902 out: 2903 /* 2904 * The current mss, t_maxseg, is initialized to the default value. 2905 * If we compute a smaller value, reduce the current mss. 2906 * If we compute a larger value, return it for use in sending 2907 * a max seg size option, but don't store it for use 2908 * unless we received an offer at least that large from peer. 2909 * However, do not accept offers under 64 bytes. 2910 */ 2911 if (offer > 0) 2912 tp->t_peermss = offer; 2913 if (tp->t_peermss) 2914 mss = min(mss, tp->t_peermss); 2915 mss = max(mss, 64); /* sanity - at least max opt. space */ 2916 2917 /* 2918 * maxopd stores the maximum length of data AND options 2919 * in a segment; maxseg is the amount of data in a normal 2920 * segment. We need to store this value (maxopd) apart 2921 * from maxseg, because now every segment carries options 2922 * and thus we normally have somewhat less data in segments. 2923 */ 2924 tp->t_maxopd = mss; 2925 2926 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2927 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 2928 mss -= TCPOLEN_TSTAMP_APPA; 2929 2930 if (offer == -1) { 2931 /* mss changed due to Path MTU discovery */ 2932 if (mss < tp->t_maxseg) { 2933 /* 2934 * Follow suggestion in RFC 2414 to reduce the 2935 * congestion window by the ratio of the old 2936 * segment size to the new segment size. 2937 */ 2938 tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) * 2939 mss, mss); 2940 } 2941 } else 2942 tp->snd_cwnd = mss; 2943 2944 tp->t_maxseg = mss; 2945 2946 return (offer != -1 ? mssopt : mss); 2947 } 2948 2949 /* 2950 * Set connection variables based on the effective MSS. 2951 * We are passed the TCPCB for the actual connection. If we 2952 * are the server, we are called by the compressed state engine 2953 * when the 3-way handshake is complete. If we are the client, 2954 * we are called when we receive the SYN,ACK from the server. 2955 * 2956 * NOTE: The t_maxseg value must be initialized in the TCPCB 2957 * before this routine is called! 2958 */ 2959 void 2960 tcp_mss_update(tp) 2961 struct tcpcb *tp; 2962 { 2963 int mss, rtt; 2964 u_long bufsize; 2965 struct rtentry *rt; 2966 struct socket *so; 2967 2968 so = tp->t_inpcb->inp_socket; 2969 mss = tp->t_maxseg; 2970 2971 rt = in_pcbrtentry(tp->t_inpcb); 2972 2973 if (rt == NULL) 2974 return; 2975 2976 #ifdef RTV_MTU /* if route characteristics exist ... */ 2977 /* 2978 * While we're here, check if there's an initial rtt 2979 * or rttvar. Convert from the route-table units 2980 * to scaled multiples of the slow timeout timer. 2981 */ 2982 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2983 /* 2984 * XXX the lock bit for MTU indicates that the value 2985 * is also a minimum value; this is subject to time. 2986 */ 2987 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2988 TCPT_RANGESET(tp->t_rttmin, 2989 rtt / (RTM_RTTUNIT / PR_SLOWHZ), 2990 TCPTV_MIN, TCPTV_REXMTMAX); 2991 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE)); 2992 if (rt->rt_rmx.rmx_rttvar) 2993 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2994 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE)); 2995 else 2996 /* default variation is +- 1 rtt */ 2997 tp->t_rttvar = 2998 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2999 TCPT_RANGESET((long) tp->t_rxtcur, 3000 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3001 tp->t_rttmin, TCPTV_REXMTMAX); 3002 } 3003 #endif 3004 3005 /* 3006 * If there's a pipesize, change the socket buffer 3007 * to that size. Make the socket buffers an integral 3008 * number of mss units; if the mss is larger than 3009 * the socket buffer, decrease the mss. 3010 */ 3011 #ifdef RTV_SPIPE 3012 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 3013 #endif 3014 bufsize = so->so_snd.sb_hiwat; 3015 if (bufsize < mss) { 3016 mss = bufsize; 3017 /* Update t_maxseg and t_maxopd */ 3018 tcp_mss(tp, mss); 3019 } else { 3020 bufsize = roundup(bufsize, mss); 3021 if (bufsize > sb_max) 3022 bufsize = sb_max; 3023 (void)sbreserve(&so->so_snd, bufsize); 3024 } 3025 3026 #ifdef RTV_RPIPE 3027 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 3028 #endif 3029 bufsize = so->so_rcv.sb_hiwat; 3030 if (bufsize > mss) { 3031 bufsize = roundup(bufsize, mss); 3032 if (bufsize > sb_max) 3033 bufsize = sb_max; 3034 (void)sbreserve(&so->so_rcv, bufsize); 3035 #ifdef RTV_RPIPE 3036 if (rt->rt_rmx.rmx_recvpipe > 0) 3037 tcp_rscale(tp, so->so_rcv.sb_hiwat); 3038 #endif 3039 } 3040 3041 #ifdef RTV_SSTHRESH 3042 if (rt->rt_rmx.rmx_ssthresh) { 3043 /* 3044 * There's some sort of gateway or interface 3045 * buffer limit on the path. Use this to set 3046 * the slow start threshhold, but set the 3047 * threshold to no less than 2*mss. 3048 */ 3049 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 3050 } 3051 #endif /* RTV_MTU */ 3052 } 3053 #endif /* TUBA_INCLUDE */ 3054 3055 #if defined (TCP_SACK) 3056 /* 3057 * Checks for partial ack. If partial ack arrives, force the retransmission 3058 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 3059 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 3060 * be started again. If the ack advances at least to tp->snd_last, return 0. 3061 */ 3062 int 3063 tcp_newreno(tp, th) 3064 struct tcpcb *tp; 3065 struct tcphdr *th; 3066 { 3067 if (SEQ_LT(th->th_ack, tp->snd_last)) { 3068 /* 3069 * snd_una has not been updated and the socket send buffer 3070 * not yet drained of the acked data, so we have to leave 3071 * snd_una as it was to get the correct data offset in 3072 * tcp_output(). 3073 */ 3074 tcp_seq onxt = tp->snd_nxt; 3075 u_long ocwnd = tp->snd_cwnd; 3076 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3077 tp->t_rtt = 0; 3078 tp->snd_nxt = th->th_ack; 3079 /* 3080 * Set snd_cwnd to one segment beyond acknowledged offset 3081 * (tp->snd_una not yet updated when this function is called) 3082 */ 3083 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3084 (void) tcp_output(tp); 3085 tp->snd_cwnd = ocwnd; 3086 if (SEQ_GT(onxt, tp->snd_nxt)) 3087 tp->snd_nxt = onxt; 3088 /* 3089 * Partial window deflation. Relies on fact that tp->snd_una 3090 * not updated yet. 3091 */ 3092 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3093 return 1; 3094 } 3095 return 0; 3096 } 3097 #endif /* TCP_SACK */ 3098