1 /* $OpenBSD: tcp_input.c,v 1.168 2004/05/21 11:36:23 markus 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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * @(#)COPYRIGHT 1.1 (NRL) 17 January 1995 33 * 34 * NRL grants permission for redistribution and use in source and binary 35 * forms, with or without modification, of the software and documentation 36 * created at NRL provided that the following conditions are met: 37 * 38 * 1. Redistributions of source code must retain the above copyright 39 * notice, this list of conditions and the following disclaimer. 40 * 2. Redistributions in binary form must reproduce the above copyright 41 * notice, this list of conditions and the following disclaimer in the 42 * documentation and/or other materials provided with the distribution. 43 * 3. All advertising materials mentioning features or use of this software 44 * must display the following acknowledgements: 45 * This product includes software developed by the University of 46 * California, Berkeley and its contributors. 47 * This product includes software developed at the Information 48 * Technology Division, US Naval Research Laboratory. 49 * 4. Neither the name of the NRL nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS 54 * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 55 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A 56 * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR 57 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 58 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 59 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 60 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 61 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 62 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 63 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 64 * 65 * The views and conclusions contained in the software and documentation 66 * are those of the authors and should not be interpreted as representing 67 * official policies, either expressed or implied, of the US Naval 68 * Research Laboratory (NRL). 69 */ 70 71 #ifndef TUBA_INCLUDE 72 #include <sys/param.h> 73 #include <sys/systm.h> 74 #include <sys/mbuf.h> 75 #include <sys/protosw.h> 76 #include <sys/socket.h> 77 #include <sys/socketvar.h> 78 #include <sys/kernel.h> 79 80 #include <dev/rndvar.h> 81 82 #include <net/if.h> 83 #include <net/route.h> 84 85 #include <netinet/in.h> 86 #include <netinet/in_systm.h> 87 #include <netinet/ip.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/ip_var.h> 90 #include <netinet/tcp.h> 91 #include <netinet/tcp_fsm.h> 92 #include <netinet/tcp_seq.h> 93 #include <netinet/tcp_timer.h> 94 #include <netinet/tcp_var.h> 95 #include <netinet/tcpip.h> 96 #include <netinet/tcp_debug.h> 97 98 struct tcpiphdr tcp_saveti; 99 100 #ifdef INET6 101 #include <netinet6/in6_var.h> 102 #include <netinet6/nd6.h> 103 104 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 #ifdef TCP_SIGNATURE 113 #include <crypto/md5.h> 114 #endif 115 116 int tcprexmtthresh = 3; 117 int tcptv_keep_init = TCPTV_KEEP_INIT; 118 119 extern u_long sb_max; 120 121 int tcp_rst_ppslim = 100; /* 100pps */ 122 int tcp_rst_ppslim_count = 0; 123 struct timeval tcp_rst_ppslim_last; 124 125 int tcp_ackdrop_ppslim = 100; /* 100pps */ 126 int tcp_ackdrop_ppslim_count = 0; 127 struct timeval tcp_ackdrop_ppslim_last; 128 129 #endif /* TUBA_INCLUDE */ 130 #define TCP_PAWS_IDLE (24 * 24 * 60 * 60 * PR_SLOWHZ) 131 132 /* for modulo comparisons of timestamps */ 133 #define TSTMP_LT(a,b) ((int)((a)-(b)) < 0) 134 #define TSTMP_GEQ(a,b) ((int)((a)-(b)) >= 0) 135 136 /* 137 * Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. 138 */ 139 #ifdef INET6 140 #define ND6_HINT(tp) \ 141 do { \ 142 if (tp && tp->t_inpcb && (tp->t_inpcb->inp_flags & INP_IPV6) && \ 143 tp->t_inpcb->inp_route6.ro_rt) { \ 144 nd6_nud_hint(tp->t_inpcb->inp_route6.ro_rt, NULL, 0); \ 145 } \ 146 } while (0) 147 #else 148 #define ND6_HINT(tp) 149 #endif 150 151 #ifdef TCP_ECN 152 /* 153 * ECN (Explicit Congestion Notification) support based on RFC3168 154 * implementation note: 155 * snd_last is used to track a recovery phase. 156 * when cwnd is reduced, snd_last is set to snd_max. 157 * while snd_last > snd_una, the sender is in a recovery phase and 158 * its cwnd should not be reduced again. 159 * snd_last follows snd_una when not in a recovery phase. 160 */ 161 #endif 162 163 /* 164 * Macro to compute ACK transmission behavior. Delay the ACK unless 165 * we have already delayed an ACK (must send an ACK every two segments). 166 * We also ACK immediately if we received a PUSH and the ACK-on-PUSH 167 * option is enabled. 168 */ 169 #define TCP_SETUP_ACK(tp, tiflags) \ 170 do { \ 171 if ((tp)->t_flags & TF_DELACK || \ 172 (tcp_ack_on_push && (tiflags) & TH_PUSH)) \ 173 tp->t_flags |= TF_ACKNOW; \ 174 else \ 175 TCP_SET_DELACK(tp); \ 176 } while (0) 177 178 /* 179 * Insert segment ti into reassembly queue of tcp with 180 * control block tp. Return TH_FIN if reassembly now includes 181 * a segment with FIN. The macro form does the common case inline 182 * (segment is the next to be received on an established connection, 183 * and the queue is empty), avoiding linkage into and removal 184 * from the queue and repetition of various conversions. 185 * Set DELACK for segments received in order, but ack immediately 186 * when segments are out of order (so fast retransmit can work). 187 */ 188 189 #ifndef TUBA_INCLUDE 190 191 int 192 tcp_reass(tp, th, m, tlen) 193 struct tcpcb *tp; 194 struct tcphdr *th; 195 struct mbuf *m; 196 int *tlen; 197 { 198 struct ipqent *p, *q, *nq, *tiqe; 199 struct socket *so = tp->t_inpcb->inp_socket; 200 int flags; 201 202 /* 203 * Call with th==0 after become established to 204 * force pre-ESTABLISHED data up to user socket. 205 */ 206 if (th == 0) 207 goto present; 208 209 /* 210 * Allocate a new queue entry, before we throw away any data. 211 * If we can't, just drop the packet. XXX 212 */ 213 tiqe = pool_get(&tcpqe_pool, PR_NOWAIT); 214 if (tiqe == NULL) { 215 tiqe = LIST_FIRST(&tp->segq); 216 if (tiqe != NULL && th->th_seq == tp->rcv_nxt) { 217 /* Reuse last entry since new segment fills a hole */ 218 while ((p = LIST_NEXT(tiqe, ipqe_q)) != NULL) 219 tiqe = p; 220 m_freem(tiqe->ipqe_m); 221 LIST_REMOVE(tiqe, ipqe_q); 222 } 223 if (tiqe == NULL || th->th_seq != tp->rcv_nxt) { 224 /* Flush segment queue for this connection */ 225 tcp_freeq(tp); 226 tcpstat.tcps_rcvmemdrop++; 227 m_freem(m); 228 return (0); 229 } 230 } 231 232 /* 233 * Find a segment which begins after this one does. 234 */ 235 for (p = NULL, q = tp->segq.lh_first; q != NULL; 236 p = q, q = q->ipqe_q.le_next) 237 if (SEQ_GT(q->ipqe_tcp->th_seq, th->th_seq)) 238 break; 239 240 /* 241 * If there is a preceding segment, it may provide some of 242 * our data already. If so, drop the data from the incoming 243 * segment. If it provides all of our data, drop us. 244 */ 245 if (p != NULL) { 246 struct tcphdr *phdr = p->ipqe_tcp; 247 int i; 248 249 /* conversion to int (in i) handles seq wraparound */ 250 i = phdr->th_seq + phdr->th_reseqlen - th->th_seq; 251 if (i > 0) { 252 if (i >= *tlen) { 253 tcpstat.tcps_rcvduppack++; 254 tcpstat.tcps_rcvdupbyte += *tlen; 255 m_freem(m); 256 pool_put(&tcpqe_pool, tiqe); 257 return (0); 258 } 259 m_adj(m, i); 260 *tlen -= i; 261 th->th_seq += i; 262 } 263 } 264 tcpstat.tcps_rcvoopack++; 265 tcpstat.tcps_rcvoobyte += *tlen; 266 267 /* 268 * While we overlap succeeding segments trim them or, 269 * if they are completely covered, dequeue them. 270 */ 271 for (; q != NULL; q = nq) { 272 struct tcphdr *qhdr = q->ipqe_tcp; 273 int i = (th->th_seq + *tlen) - qhdr->th_seq; 274 275 if (i <= 0) 276 break; 277 if (i < qhdr->th_reseqlen) { 278 qhdr->th_seq += i; 279 qhdr->th_reseqlen -= i; 280 m_adj(q->ipqe_m, i); 281 break; 282 } 283 nq = q->ipqe_q.le_next; 284 m_freem(q->ipqe_m); 285 LIST_REMOVE(q, ipqe_q); 286 pool_put(&tcpqe_pool, q); 287 } 288 289 /* Insert the new segment queue entry into place. */ 290 tiqe->ipqe_m = m; 291 th->th_reseqlen = *tlen; 292 tiqe->ipqe_tcp = th; 293 if (p == NULL) { 294 LIST_INSERT_HEAD(&tp->segq, tiqe, ipqe_q); 295 } else { 296 LIST_INSERT_AFTER(p, tiqe, ipqe_q); 297 } 298 299 present: 300 /* 301 * Present data to user, advancing rcv_nxt through 302 * completed sequence space. 303 */ 304 if (TCPS_HAVEESTABLISHED(tp->t_state) == 0) 305 return (0); 306 q = tp->segq.lh_first; 307 if (q == NULL || q->ipqe_tcp->th_seq != tp->rcv_nxt) 308 return (0); 309 if (tp->t_state == TCPS_SYN_RECEIVED && q->ipqe_tcp->th_reseqlen) 310 return (0); 311 do { 312 tp->rcv_nxt += q->ipqe_tcp->th_reseqlen; 313 flags = q->ipqe_tcp->th_flags & TH_FIN; 314 315 nq = q->ipqe_q.le_next; 316 LIST_REMOVE(q, ipqe_q); 317 ND6_HINT(tp); 318 if (so->so_state & SS_CANTRCVMORE) 319 m_freem(q->ipqe_m); 320 else 321 sbappendstream(&so->so_rcv, q->ipqe_m); 322 pool_put(&tcpqe_pool, q); 323 q = nq; 324 } while (q != NULL && q->ipqe_tcp->th_seq == tp->rcv_nxt); 325 sorwakeup(so); 326 return (flags); 327 } 328 329 #ifdef INET6 330 int 331 tcp6_input(mp, offp, proto) 332 struct mbuf **mp; 333 int *offp, proto; 334 { 335 struct mbuf *m = *mp; 336 337 #if defined(NFAITH) && 0 < NFAITH 338 if (m->m_pkthdr.rcvif) { 339 if (m->m_pkthdr.rcvif->if_type == IFT_FAITH) { 340 /* XXX send icmp6 host/port unreach? */ 341 m_freem(m); 342 return IPPROTO_DONE; 343 } 344 } 345 #endif 346 347 /* 348 * draft-itojun-ipv6-tcp-to-anycast 349 * better place to put this in? 350 */ 351 if (m->m_flags & M_ANYCAST6) { 352 if (m->m_len >= sizeof(struct ip6_hdr)) { 353 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 354 icmp6_error(m, ICMP6_DST_UNREACH, 355 ICMP6_DST_UNREACH_ADDR, 356 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 357 } else 358 m_freem(m); 359 return IPPROTO_DONE; 360 } 361 362 tcp_input(m, *offp, proto); 363 return IPPROTO_DONE; 364 } 365 #endif 366 367 /* 368 * TCP input routine, follows pages 65-76 of the 369 * protocol specification dated September, 1981 very closely. 370 */ 371 void 372 tcp_input(struct mbuf *m, ...) 373 { 374 struct ip *ip; 375 struct inpcb *inp; 376 u_int8_t *optp = NULL; 377 int optlen = 0; 378 int len, tlen, off; 379 struct tcpcb *tp = 0; 380 int tiflags; 381 struct socket *so = NULL; 382 int todrop, acked, ourfinisacked, needoutput = 0; 383 int hdroptlen = 0; 384 short ostate = 0; 385 int iss = 0; 386 u_long tiwin; 387 struct tcp_opt_info opti; 388 int iphlen; 389 va_list ap; 390 struct tcphdr *th; 391 #ifdef INET6 392 struct ip6_hdr *ip6 = NULL; 393 #endif /* INET6 */ 394 #ifdef IPSEC 395 struct m_tag *mtag; 396 struct tdb_ident *tdbi; 397 struct tdb *tdb; 398 int error, s; 399 #endif /* IPSEC */ 400 int af; 401 #ifdef TCP_ECN 402 u_char iptos; 403 #endif 404 405 va_start(ap, m); 406 iphlen = va_arg(ap, int); 407 va_end(ap); 408 409 tcpstat.tcps_rcvtotal++; 410 411 opti.ts_present = 0; 412 opti.maxseg = 0; 413 414 /* 415 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 416 * See below for AF specific multicast. 417 */ 418 if (m->m_flags & (M_BCAST|M_MCAST)) 419 goto drop; 420 421 /* 422 * Before we do ANYTHING, we have to figure out if it's TCP/IPv6 or 423 * TCP/IPv4. 424 */ 425 switch (mtod(m, struct ip *)->ip_v) { 426 #ifdef INET6 427 case 6: 428 af = AF_INET6; 429 break; 430 #endif 431 case 4: 432 af = AF_INET; 433 break; 434 default: 435 m_freem(m); 436 return; /*EAFNOSUPPORT*/ 437 } 438 439 /* 440 * Get IP and TCP header together in first mbuf. 441 * Note: IP leaves IP header in first mbuf. 442 */ 443 switch (af) { 444 case AF_INET: 445 #ifdef DIAGNOSTIC 446 if (iphlen < sizeof(struct ip)) { 447 m_freem(m); 448 return; 449 } 450 #endif /* DIAGNOSTIC */ 451 break; 452 #ifdef INET6 453 case AF_INET6: 454 #ifdef DIAGNOSTIC 455 if (iphlen < sizeof(struct ip6_hdr)) { 456 m_freem(m); 457 return; 458 } 459 #endif /* DIAGNOSTIC */ 460 break; 461 #endif 462 default: 463 m_freem(m); 464 return; 465 } 466 467 IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, sizeof(*th)); 468 if (!th) { 469 tcpstat.tcps_rcvshort++; 470 return; 471 } 472 473 ip = NULL; 474 #ifdef INET6 475 ip6 = NULL; 476 #endif 477 switch (af) { 478 case AF_INET: 479 ip = mtod(m, struct ip *); 480 if (IN_MULTICAST(ip->ip_dst.s_addr) || 481 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 482 goto drop; 483 484 tlen = m->m_pkthdr.len - iphlen; 485 486 #ifdef TCP_ECN 487 /* save ip_tos before clearing it for checksum */ 488 iptos = ip->ip_tos; 489 #endif 490 /* 491 * Checksum extended TCP header and data. 492 */ 493 if ((m->m_pkthdr.csum & M_TCP_CSUM_IN_OK) == 0) { 494 if (m->m_pkthdr.csum & M_TCP_CSUM_IN_BAD) { 495 tcpstat.tcps_inhwcsum++; 496 tcpstat.tcps_rcvbadsum++; 497 goto drop; 498 } 499 len = m->m_pkthdr.len - iphlen; 500 if (in4_cksum(m, IPPROTO_TCP, iphlen, len) != 0) { 501 tcpstat.tcps_rcvbadsum++; 502 goto drop; 503 } 504 } else { 505 m->m_pkthdr.csum &= ~M_TCP_CSUM_IN_OK; 506 tcpstat.tcps_inhwcsum++; 507 } 508 break; 509 #ifdef INET6 510 case AF_INET6: 511 ip6 = mtod(m, struct ip6_hdr *); 512 tlen = m->m_pkthdr.len - iphlen; 513 #ifdef TCP_ECN 514 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 515 #endif 516 517 /* Be proactive about malicious use of IPv4 mapped address */ 518 if (IN6_IS_ADDR_V4MAPPED(&ip6->ip6_src) || 519 IN6_IS_ADDR_V4MAPPED(&ip6->ip6_dst)) { 520 /* XXX stat */ 521 goto drop; 522 } 523 524 /* 525 * Be proactive about unspecified IPv6 address in source. 526 * As we use all-zero to indicate unbounded/unconnected pcb, 527 * unspecified IPv6 address can be used to confuse us. 528 * 529 * Note that packets with unspecified IPv6 destination is 530 * already dropped in ip6_input. 531 */ 532 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 533 /* XXX stat */ 534 goto drop; 535 } 536 537 /* Discard packets to multicast */ 538 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst)) { 539 /* XXX stat */ 540 goto drop; 541 } 542 543 /* 544 * Checksum extended TCP header and data. 545 */ 546 if (in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr), tlen)) { 547 tcpstat.tcps_rcvbadsum++; 548 goto drop; 549 } 550 break; 551 #endif 552 } 553 #endif /* TUBA_INCLUDE */ 554 555 th = (struct tcphdr *)(mtod(m, caddr_t) + iphlen); 556 557 /* 558 * Check that TCP offset makes sense, 559 * pull out TCP options and adjust length. XXX 560 */ 561 off = th->th_off << 2; 562 if (off < sizeof(struct tcphdr) || off > tlen) { 563 tcpstat.tcps_rcvbadoff++; 564 goto drop; 565 } 566 tlen -= off; 567 if (off > sizeof(struct tcphdr)) { 568 IP6_EXTHDR_GET(th, struct tcphdr *, m, iphlen, off); 569 if (!th) { 570 tcpstat.tcps_rcvshort++; 571 return; 572 } 573 optlen = off - sizeof(struct tcphdr); 574 optp = mtod(m, u_int8_t *) + iphlen + sizeof(struct tcphdr); 575 /* 576 * Do quick retrieval of timestamp options ("options 577 * prediction?"). If timestamp is the only option and it's 578 * formatted as recommended in RFC 1323 appendix A, we 579 * quickly get the values now and not bother calling 580 * tcp_dooptions(), etc. 581 */ 582 if ((optlen == TCPOLEN_TSTAMP_APPA || 583 (optlen > TCPOLEN_TSTAMP_APPA && 584 optp[TCPOLEN_TSTAMP_APPA] == TCPOPT_EOL)) && 585 *(u_int32_t *)optp == htonl(TCPOPT_TSTAMP_HDR) && 586 (th->th_flags & TH_SYN) == 0) { 587 opti.ts_present = 1; 588 opti.ts_val = ntohl(*(u_int32_t *)(optp + 4)); 589 opti.ts_ecr = ntohl(*(u_int32_t *)(optp + 8)); 590 optp = NULL; /* we've parsed the options */ 591 } 592 } 593 tiflags = th->th_flags; 594 595 /* 596 * Convert TCP protocol specific fields to host format. 597 */ 598 NTOHL(th->th_seq); 599 NTOHL(th->th_ack); 600 NTOHS(th->th_win); 601 NTOHS(th->th_urp); 602 603 /* 604 * Locate pcb for segment. 605 */ 606 findpcb: 607 switch (af) { 608 #ifdef INET6 609 case AF_INET6: 610 inp = in6_pcbhashlookup(&tcbtable, &ip6->ip6_src, th->th_sport, 611 &ip6->ip6_dst, th->th_dport); 612 break; 613 #endif 614 case AF_INET: 615 inp = in_pcbhashlookup(&tcbtable, ip->ip_src, th->th_sport, 616 ip->ip_dst, th->th_dport); 617 break; 618 } 619 if (inp == 0) { 620 ++tcpstat.tcps_pcbhashmiss; 621 switch (af) { 622 #ifdef INET6 623 case AF_INET6: 624 inp = in6_pcblookup_listen(&tcbtable, 625 &ip6->ip6_dst, th->th_dport, m_tag_find(m, 626 PACKET_TAG_PF_TRANSLATE_LOCALHOST, NULL) != NULL); 627 break; 628 #endif /* INET6 */ 629 case AF_INET: 630 inp = in_pcblookup_listen(&tcbtable, 631 ip->ip_dst, th->th_dport, m_tag_find(m, 632 PACKET_TAG_PF_TRANSLATE_LOCALHOST, NULL) != NULL); 633 break; 634 } 635 /* 636 * If the state is CLOSED (i.e., TCB does not exist) then 637 * all data in the incoming segment is discarded. 638 * If the TCB exists but is in CLOSED state, it is embryonic, 639 * but should either do a listen or a connect soon. 640 */ 641 if (inp == 0) { 642 ++tcpstat.tcps_noport; 643 goto dropwithreset_ratelim; 644 } 645 } 646 647 tp = intotcpcb(inp); 648 if (tp == 0) 649 goto dropwithreset_ratelim; 650 if (tp->t_state == TCPS_CLOSED) 651 goto drop; 652 653 /* Unscale the window into a 32-bit value. */ 654 if ((tiflags & TH_SYN) == 0) 655 tiwin = th->th_win << tp->snd_scale; 656 else 657 tiwin = th->th_win; 658 659 so = inp->inp_socket; 660 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) { 661 union syn_cache_sa src; 662 union syn_cache_sa dst; 663 664 bzero(&src, sizeof(src)); 665 bzero(&dst, sizeof(dst)); 666 switch (af) { 667 #ifdef INET 668 case AF_INET: 669 src.sin.sin_len = sizeof(struct sockaddr_in); 670 src.sin.sin_family = AF_INET; 671 src.sin.sin_addr = ip->ip_src; 672 src.sin.sin_port = th->th_sport; 673 674 dst.sin.sin_len = sizeof(struct sockaddr_in); 675 dst.sin.sin_family = AF_INET; 676 dst.sin.sin_addr = ip->ip_dst; 677 dst.sin.sin_port = th->th_dport; 678 break; 679 #endif 680 #ifdef INET6 681 case AF_INET6: 682 src.sin6.sin6_len = sizeof(struct sockaddr_in6); 683 src.sin6.sin6_family = AF_INET6; 684 src.sin6.sin6_addr = ip6->ip6_src; 685 src.sin6.sin6_port = th->th_sport; 686 687 dst.sin6.sin6_len = sizeof(struct sockaddr_in6); 688 dst.sin6.sin6_family = AF_INET6; 689 dst.sin6.sin6_addr = ip6->ip6_dst; 690 dst.sin6.sin6_port = th->th_dport; 691 break; 692 #endif /* INET6 */ 693 default: 694 goto badsyn; /*sanity*/ 695 } 696 697 if (so->so_options & SO_DEBUG) { 698 ostate = tp->t_state; 699 switch (af) { 700 #ifdef INET6 701 case AF_INET6: 702 bcopy(ip6, &tcp_saveti6.ti6_i, sizeof(*ip6)); 703 bcopy(th, &tcp_saveti6.ti6_t, sizeof(*th)); 704 break; 705 #endif 706 case AF_INET: 707 bcopy(ip, &tcp_saveti.ti_i, sizeof(*ip)); 708 bcopy(th, &tcp_saveti.ti_t, sizeof(*th)); 709 break; 710 } 711 } 712 if (so->so_options & SO_ACCEPTCONN) { 713 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 714 if (tiflags & TH_RST) { 715 syn_cache_reset(&src.sa, &dst.sa, th); 716 } else if ((tiflags & (TH_ACK|TH_SYN)) == 717 (TH_ACK|TH_SYN)) { 718 /* 719 * Received a SYN,ACK. This should 720 * never happen while we are in 721 * LISTEN. Send an RST. 722 */ 723 goto badsyn; 724 } else if (tiflags & TH_ACK) { 725 so = syn_cache_get(&src.sa, &dst.sa, 726 th, iphlen, tlen, so, m); 727 if (so == NULL) { 728 /* 729 * We don't have a SYN for 730 * this ACK; send an RST. 731 */ 732 goto badsyn; 733 } else if (so == 734 (struct socket *)(-1)) { 735 /* 736 * We were unable to create 737 * the connection. If the 738 * 3-way handshake was 739 * completed, and RST has 740 * been sent to the peer. 741 * Since the mbuf might be 742 * in use for the reply, 743 * do not free it. 744 */ 745 m = NULL; 746 } else { 747 /* 748 * We have created a 749 * full-blown connection. 750 */ 751 tp = NULL; 752 inp = (struct inpcb *)so->so_pcb; 753 tp = intotcpcb(inp); 754 if (tp == NULL) 755 goto badsyn; /*XXX*/ 756 757 /* 758 * Compute proper scaling 759 * value from buffer space 760 */ 761 tcp_rscale(tp, so->so_rcv.sb_hiwat); 762 goto after_listen; 763 } 764 } else { 765 /* 766 * None of RST, SYN or ACK was set. 767 * This is an invalid packet for a 768 * TCB in LISTEN state. Send a RST. 769 */ 770 goto badsyn; 771 } 772 } else { 773 /* 774 * Received a SYN. 775 */ 776 #ifdef INET6 777 /* 778 * If deprecated address is forbidden, we do 779 * not accept SYN to deprecated interface 780 * address to prevent any new inbound 781 * connection from getting established. 782 * When we do not accept SYN, we send a TCP 783 * RST, with deprecated source address (instead 784 * of dropping it). We compromise it as it is 785 * much better for peer to send a RST, and 786 * RST will be the final packet for the 787 * exchange. 788 * 789 * If we do not forbid deprecated addresses, we 790 * accept the SYN packet. RFC2462 does not 791 * suggest dropping SYN in this case. 792 * If we decipher RFC2462 5.5.4, it says like 793 * this: 794 * 1. use of deprecated addr with existing 795 * communication is okay - "SHOULD continue 796 * to be used" 797 * 2. use of it with new communication: 798 * (2a) "SHOULD NOT be used if alternate 799 * address with sufficient scope is 800 * available" 801 * (2b) nothing mentioned otherwise. 802 * Here we fall into (2b) case as we have no 803 * choice in our source address selection - we 804 * must obey the peer. 805 * 806 * The wording in RFC2462 is confusing, and 807 * there are multiple description text for 808 * deprecated address handling - worse, they 809 * are not exactly the same. I believe 5.5.4 810 * is the best one, so we follow 5.5.4. 811 */ 812 if (ip6 && !ip6_use_deprecated) { 813 struct in6_ifaddr *ia6; 814 815 if ((ia6 = in6ifa_ifpwithaddr(m->m_pkthdr.rcvif, 816 &ip6->ip6_dst)) && 817 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 818 tp = NULL; 819 goto dropwithreset; 820 } 821 } 822 #endif 823 824 /* 825 * LISTEN socket received a SYN 826 * from itself? This can't possibly 827 * be valid; drop the packet. 828 */ 829 if (th->th_dport == th->th_sport) { 830 switch (af) { 831 #ifdef INET6 832 case AF_INET6: 833 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_src, 834 &ip6->ip6_dst)) { 835 tcpstat.tcps_badsyn++; 836 goto drop; 837 } 838 break; 839 #endif /* INET6 */ 840 case AF_INET: 841 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) { 842 tcpstat.tcps_badsyn++; 843 goto drop; 844 } 845 break; 846 } 847 } 848 849 /* 850 * SYN looks ok; create compressed TCP 851 * state for it. 852 */ 853 if (so->so_qlen <= so->so_qlimit && 854 syn_cache_add(&src.sa, &dst.sa, th, iphlen, 855 so, m, optp, optlen, &opti)) 856 m = NULL; 857 } 858 goto drop; 859 } 860 } 861 862 after_listen: 863 #ifdef DIAGNOSTIC 864 /* 865 * Should not happen now that all embryonic connections 866 * are handled with compressed state. 867 */ 868 if (tp->t_state == TCPS_LISTEN) 869 panic("tcp_input: TCPS_LISTEN"); 870 #endif 871 872 #ifdef IPSEC 873 /* Find most recent IPsec tag */ 874 mtag = m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL); 875 s = splnet(); 876 if (mtag != NULL) { 877 tdbi = (struct tdb_ident *)(mtag + 1); 878 tdb = gettdb(tdbi->spi, &tdbi->dst, tdbi->proto); 879 } else 880 tdb = NULL; 881 ipsp_spd_lookup(m, af, iphlen, &error, IPSP_DIRECTION_IN, 882 tdb, inp); 883 if (error) { 884 splx(s); 885 goto drop; 886 } 887 888 /* Latch SA */ 889 if (inp->inp_tdb_in != tdb) { 890 if (tdb) { 891 tdb_add_inp(tdb, inp, 1); 892 if (inp->inp_ipo == NULL) { 893 inp->inp_ipo = ipsec_add_policy(inp, af, 894 IPSP_DIRECTION_OUT); 895 if (inp->inp_ipo == NULL) { 896 splx(s); 897 goto drop; 898 } 899 } 900 if (inp->inp_ipo->ipo_dstid == NULL && 901 tdb->tdb_srcid != NULL) { 902 inp->inp_ipo->ipo_dstid = tdb->tdb_srcid; 903 tdb->tdb_srcid->ref_count++; 904 } 905 if (inp->inp_ipsec_remotecred == NULL && 906 tdb->tdb_remote_cred != NULL) { 907 inp->inp_ipsec_remotecred = 908 tdb->tdb_remote_cred; 909 tdb->tdb_remote_cred->ref_count++; 910 } 911 if (inp->inp_ipsec_remoteauth == NULL && 912 tdb->tdb_remote_auth != NULL) { 913 inp->inp_ipsec_remoteauth = 914 tdb->tdb_remote_auth; 915 tdb->tdb_remote_auth->ref_count++; 916 } 917 } else { /* Just reset */ 918 TAILQ_REMOVE(&inp->inp_tdb_in->tdb_inp_in, inp, 919 inp_tdb_in_next); 920 inp->inp_tdb_in = NULL; 921 } 922 } 923 splx(s); 924 #endif /* IPSEC */ 925 926 /* 927 * Segment received on connection. 928 * Reset idle time and keep-alive timer. 929 */ 930 tp->t_rcvtime = tcp_now; 931 if (TCPS_HAVEESTABLISHED(tp->t_state)) 932 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 933 934 #ifdef TCP_SACK 935 if (tp->sack_enable) 936 tcp_del_sackholes(tp, th); /* Delete stale SACK holes */ 937 #endif /* TCP_SACK */ 938 939 /* 940 * Process options. 941 */ 942 #ifdef TCP_SIGNATURE 943 if (optp || (tp->t_flags & TF_SIGNATURE)) 944 #else 945 if (optp) 946 #endif 947 if (tcp_dooptions(tp, optp, optlen, th, m, iphlen, &opti)) 948 goto drop; 949 950 #ifdef TCP_SACK 951 if (tp->sack_enable) { 952 tp->rcv_laststart = th->th_seq; /* last rec'vd segment*/ 953 tp->rcv_lastend = th->th_seq + tlen; 954 } 955 #endif /* TCP_SACK */ 956 #ifdef TCP_ECN 957 /* if congestion experienced, set ECE bit in subsequent packets. */ 958 if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) { 959 tp->t_flags |= TF_RCVD_CE; 960 tcpstat.tcps_ecn_rcvce++; 961 } 962 #endif 963 /* 964 * Header prediction: check for the two common cases 965 * of a uni-directional data xfer. If the packet has 966 * no control flags, is in-sequence, the window didn't 967 * change and we're not retransmitting, it's a 968 * candidate. If the length is zero and the ack moved 969 * forward, we're the sender side of the xfer. Just 970 * free the data acked & wake any higher level process 971 * that was blocked waiting for space. If the length 972 * is non-zero and the ack didn't move, we're the 973 * receiver side. If we're getting packets in-order 974 * (the reassembly queue is empty), add the data to 975 * the socket buffer and note that we need a delayed ack. 976 */ 977 if (tp->t_state == TCPS_ESTABLISHED && 978 #ifdef TCP_ECN 979 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ECE|TH_CWR|TH_ACK)) == TH_ACK && 980 #else 981 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 982 #endif 983 (!opti.ts_present || TSTMP_GEQ(opti.ts_val, tp->ts_recent)) && 984 th->th_seq == tp->rcv_nxt && 985 tiwin && tiwin == tp->snd_wnd && 986 tp->snd_nxt == tp->snd_max) { 987 988 /* 989 * If last ACK falls within this segment's sequence numbers, 990 * record the timestamp. 991 * Fix from Braden, see Stevens p. 870 992 */ 993 if (opti.ts_present && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 994 tp->ts_recent_age = tcp_now; 995 tp->ts_recent = opti.ts_val; 996 } 997 998 if (tlen == 0) { 999 if (SEQ_GT(th->th_ack, tp->snd_una) && 1000 SEQ_LEQ(th->th_ack, tp->snd_max) && 1001 tp->snd_cwnd >= tp->snd_wnd && 1002 tp->t_dupacks == 0) { 1003 /* 1004 * this is a pure ack for outstanding data. 1005 */ 1006 ++tcpstat.tcps_predack; 1007 if (opti.ts_present) 1008 tcp_xmit_timer(tp, tcp_now-opti.ts_ecr+1); 1009 else if (tp->t_rtttime && 1010 SEQ_GT(th->th_ack, tp->t_rtseq)) 1011 tcp_xmit_timer(tp, 1012 tcp_now - tp->t_rtttime); 1013 acked = th->th_ack - tp->snd_una; 1014 tcpstat.tcps_rcvackpack++; 1015 tcpstat.tcps_rcvackbyte += acked; 1016 ND6_HINT(tp); 1017 sbdrop(&so->so_snd, acked); 1018 tp->snd_una = th->th_ack; 1019 #if defined(TCP_SACK) || defined(TCP_ECN) 1020 /* 1021 * We want snd_last to track snd_una so 1022 * as to avoid sequence wraparound problems 1023 * for very large transfers. 1024 */ 1025 #ifdef TCP_ECN 1026 if (SEQ_GT(tp->snd_una, tp->snd_last)) 1027 #endif 1028 tp->snd_last = tp->snd_una; 1029 #endif /* TCP_SACK */ 1030 #if defined(TCP_SACK) && defined(TCP_FACK) 1031 tp->snd_fack = tp->snd_una; 1032 tp->retran_data = 0; 1033 #endif /* TCP_FACK */ 1034 m_freem(m); 1035 1036 /* 1037 * If all outstanding data are acked, stop 1038 * retransmit timer, otherwise restart timer 1039 * using current (possibly backed-off) value. 1040 * If process is waiting for space, 1041 * wakeup/selwakeup/signal. If data 1042 * are ready to send, let tcp_output 1043 * decide between more output or persist. 1044 */ 1045 if (tp->snd_una == tp->snd_max) 1046 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1047 else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1048 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1049 1050 if (sb_notify(&so->so_snd)) 1051 sowwakeup(so); 1052 if (so->so_snd.sb_cc) 1053 (void) tcp_output(tp); 1054 return; 1055 } 1056 } else if (th->th_ack == tp->snd_una && 1057 tp->segq.lh_first == NULL && 1058 tlen <= sbspace(&so->so_rcv)) { 1059 /* 1060 * This is a pure, in-sequence data packet 1061 * with nothing on the reassembly queue and 1062 * we have enough buffer space to take it. 1063 */ 1064 #ifdef TCP_SACK 1065 /* Clean receiver SACK report if present */ 1066 if (tp->sack_enable && tp->rcv_numsacks) 1067 tcp_clean_sackreport(tp); 1068 #endif /* TCP_SACK */ 1069 ++tcpstat.tcps_preddat; 1070 tp->rcv_nxt += tlen; 1071 tcpstat.tcps_rcvpack++; 1072 tcpstat.tcps_rcvbyte += tlen; 1073 ND6_HINT(tp); 1074 /* 1075 * Drop TCP, IP headers and TCP options then add data 1076 * to socket buffer. 1077 */ 1078 if (so->so_state & SS_CANTRCVMORE) 1079 m_freem(m); 1080 else { 1081 m_adj(m, iphlen + off); 1082 sbappendstream(&so->so_rcv, m); 1083 } 1084 sorwakeup(so); 1085 TCP_SETUP_ACK(tp, tiflags); 1086 if (tp->t_flags & TF_ACKNOW) 1087 (void) tcp_output(tp); 1088 return; 1089 } 1090 } 1091 1092 /* 1093 * Compute mbuf offset to TCP data segment. 1094 */ 1095 hdroptlen = iphlen + off; 1096 1097 /* 1098 * Calculate amount of space in receive window, 1099 * and then do TCP input processing. 1100 * Receive window is amount of space in rcv queue, 1101 * but not less than advertised window. 1102 */ 1103 { int win; 1104 1105 win = sbspace(&so->so_rcv); 1106 if (win < 0) 1107 win = 0; 1108 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1109 } 1110 1111 switch (tp->t_state) { 1112 1113 /* 1114 * If the state is SYN_RECEIVED: 1115 * if seg contains SYN/ACK, send an RST. 1116 * if seg contains an ACK, but not for our SYN/ACK, send an RST 1117 */ 1118 1119 case TCPS_SYN_RECEIVED: 1120 if (tiflags & TH_ACK) { 1121 if (tiflags & TH_SYN) { 1122 tcpstat.tcps_badsyn++; 1123 goto dropwithreset; 1124 } 1125 if (SEQ_LEQ(th->th_ack, tp->snd_una) || 1126 SEQ_GT(th->th_ack, tp->snd_max)) 1127 goto dropwithreset; 1128 } 1129 break; 1130 1131 /* 1132 * If the state is SYN_SENT: 1133 * if seg contains an ACK, but not for our SYN, drop the input. 1134 * if seg contains a RST, then drop the connection. 1135 * if seg does not contain SYN, then drop it. 1136 * Otherwise this is an acceptable SYN segment 1137 * initialize tp->rcv_nxt and tp->irs 1138 * if seg contains ack then advance tp->snd_una 1139 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1140 * arrange for segment to be acked (eventually) 1141 * continue processing rest of data/controls, beginning with URG 1142 */ 1143 case TCPS_SYN_SENT: 1144 if ((tiflags & TH_ACK) && 1145 (SEQ_LEQ(th->th_ack, tp->iss) || 1146 SEQ_GT(th->th_ack, tp->snd_max))) 1147 goto dropwithreset; 1148 if (tiflags & TH_RST) { 1149 #ifdef TCP_ECN 1150 /* if ECN is enabled, fall back to non-ecn at rexmit */ 1151 if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN)) 1152 goto drop; 1153 #endif 1154 if (tiflags & TH_ACK) 1155 tp = tcp_drop(tp, ECONNREFUSED); 1156 goto drop; 1157 } 1158 if ((tiflags & TH_SYN) == 0) 1159 goto drop; 1160 if (tiflags & TH_ACK) { 1161 tp->snd_una = th->th_ack; 1162 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 1163 tp->snd_nxt = tp->snd_una; 1164 } 1165 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1166 tp->irs = th->th_seq; 1167 tcp_mss(tp, opti.maxseg); 1168 /* Reset initial window to 1 segment for retransmit */ 1169 if (tp->t_rxtshift > 0) 1170 tp->snd_cwnd = tp->t_maxseg; 1171 tcp_rcvseqinit(tp); 1172 tp->t_flags |= TF_ACKNOW; 1173 #ifdef TCP_SACK 1174 /* 1175 * If we've sent a SACK_PERMITTED option, and the peer 1176 * also replied with one, then TF_SACK_PERMIT should have 1177 * been set in tcp_dooptions(). If it was not, disable SACKs. 1178 */ 1179 if (tp->sack_enable) 1180 tp->sack_enable = tp->t_flags & TF_SACK_PERMIT; 1181 #endif 1182 #ifdef TCP_ECN 1183 /* 1184 * if ECE is set but CWR is not set for SYN-ACK, or 1185 * both ECE and CWR are set for simultaneous open, 1186 * peer is ECN capable. 1187 */ 1188 if (tcp_do_ecn) { 1189 if ((tiflags & (TH_ACK|TH_ECE|TH_CWR)) 1190 == (TH_ACK|TH_ECE) || 1191 (tiflags & (TH_ACK|TH_ECE|TH_CWR)) 1192 == (TH_ECE|TH_CWR)) { 1193 tp->t_flags |= TF_ECN_PERMIT; 1194 tiflags &= ~(TH_ECE|TH_CWR); 1195 tcpstat.tcps_ecn_accepts++; 1196 } 1197 } 1198 #endif 1199 1200 if (tiflags & TH_ACK && SEQ_GT(tp->snd_una, tp->iss)) { 1201 tcpstat.tcps_connects++; 1202 soisconnected(so); 1203 tp->t_state = TCPS_ESTABLISHED; 1204 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1205 /* Do window scaling on this connection? */ 1206 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1207 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1208 tp->snd_scale = tp->requested_s_scale; 1209 tp->rcv_scale = tp->request_r_scale; 1210 } 1211 tcp_reass_lock(tp); 1212 (void) tcp_reass(tp, (struct tcphdr *)0, 1213 (struct mbuf *)0, &tlen); 1214 tcp_reass_unlock(tp); 1215 /* 1216 * if we didn't have to retransmit the SYN, 1217 * use its rtt as our initial srtt & rtt var. 1218 */ 1219 if (tp->t_rtttime) 1220 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1221 } else 1222 tp->t_state = TCPS_SYN_RECEIVED; 1223 1224 #if 0 1225 trimthenstep6: 1226 #endif 1227 /* 1228 * Advance th->th_seq to correspond to first data byte. 1229 * If data, trim to stay within window, 1230 * dropping FIN if necessary. 1231 */ 1232 th->th_seq++; 1233 if (tlen > tp->rcv_wnd) { 1234 todrop = tlen - tp->rcv_wnd; 1235 m_adj(m, -todrop); 1236 tlen = tp->rcv_wnd; 1237 tiflags &= ~TH_FIN; 1238 tcpstat.tcps_rcvpackafterwin++; 1239 tcpstat.tcps_rcvbyteafterwin += todrop; 1240 } 1241 tp->snd_wl1 = th->th_seq - 1; 1242 tp->rcv_up = th->th_seq; 1243 goto step6; 1244 } 1245 1246 /* 1247 * States other than LISTEN or SYN_SENT. 1248 * First check timestamp, if present. 1249 * Then check that at least some bytes of segment are within 1250 * receive window. If segment begins before rcv_nxt, 1251 * drop leading data (and SYN); if nothing left, just ack. 1252 * 1253 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1254 * and it's less than opti.ts_recent, drop it. 1255 */ 1256 if (opti.ts_present && (tiflags & TH_RST) == 0 && tp->ts_recent && 1257 TSTMP_LT(opti.ts_val, tp->ts_recent)) { 1258 1259 /* Check to see if ts_recent is over 24 days old. */ 1260 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1261 /* 1262 * Invalidate ts_recent. If this segment updates 1263 * ts_recent, the age will be reset later and ts_recent 1264 * will get a valid value. If it does not, setting 1265 * ts_recent to zero will at least satisfy the 1266 * requirement that zero be placed in the timestamp 1267 * echo reply when ts_recent isn't valid. The 1268 * age isn't reset until we get a valid ts_recent 1269 * because we don't want out-of-order segments to be 1270 * dropped when ts_recent is old. 1271 */ 1272 tp->ts_recent = 0; 1273 } else { 1274 tcpstat.tcps_rcvduppack++; 1275 tcpstat.tcps_rcvdupbyte += tlen; 1276 tcpstat.tcps_pawsdrop++; 1277 goto dropafterack; 1278 } 1279 } 1280 1281 todrop = tp->rcv_nxt - th->th_seq; 1282 if (todrop > 0) { 1283 if (tiflags & TH_SYN) { 1284 tiflags &= ~TH_SYN; 1285 th->th_seq++; 1286 if (th->th_urp > 1) 1287 th->th_urp--; 1288 else 1289 tiflags &= ~TH_URG; 1290 todrop--; 1291 } 1292 if (todrop > tlen || 1293 (todrop == tlen && (tiflags & TH_FIN) == 0)) { 1294 /* 1295 * Any valid FIN must be to the left of the 1296 * window. At this point, FIN must be a 1297 * duplicate or out-of-sequence, so drop it. 1298 */ 1299 tiflags &= ~TH_FIN; 1300 /* 1301 * Send ACK to resynchronize, and drop any data, 1302 * but keep on processing for RST or ACK. 1303 */ 1304 tp->t_flags |= TF_ACKNOW; 1305 tcpstat.tcps_rcvdupbyte += todrop = tlen; 1306 tcpstat.tcps_rcvduppack++; 1307 } else { 1308 tcpstat.tcps_rcvpartduppack++; 1309 tcpstat.tcps_rcvpartdupbyte += todrop; 1310 } 1311 hdroptlen += todrop; /* drop from head afterwards */ 1312 th->th_seq += todrop; 1313 tlen -= todrop; 1314 if (th->th_urp > todrop) 1315 th->th_urp -= todrop; 1316 else { 1317 tiflags &= ~TH_URG; 1318 th->th_urp = 0; 1319 } 1320 } 1321 1322 /* 1323 * If new data are received on a connection after the 1324 * user processes are gone, then RST the other end. 1325 */ 1326 if ((so->so_state & SS_NOFDREF) && 1327 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1328 tp = tcp_close(tp); 1329 tcpstat.tcps_rcvafterclose++; 1330 goto dropwithreset; 1331 } 1332 1333 /* 1334 * If segment ends after window, drop trailing data 1335 * (and PUSH and FIN); if nothing left, just ACK. 1336 */ 1337 todrop = (th->th_seq + tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1338 if (todrop > 0) { 1339 tcpstat.tcps_rcvpackafterwin++; 1340 if (todrop >= tlen) { 1341 tcpstat.tcps_rcvbyteafterwin += tlen; 1342 /* 1343 * If a new connection request is received 1344 * while in TIME_WAIT, drop the old connection 1345 * and start over if the sequence numbers 1346 * are above the previous ones. 1347 */ 1348 if (tiflags & TH_SYN && 1349 tp->t_state == TCPS_TIME_WAIT && 1350 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1351 iss = tp->snd_nxt + TCP_ISSINCR; 1352 tp = tcp_close(tp); 1353 goto findpcb; 1354 } 1355 /* 1356 * If window is closed can only take segments at 1357 * window edge, and have to drop data and PUSH from 1358 * incoming segments. Continue processing, but 1359 * remember to ack. Otherwise, drop segment 1360 * and ack. 1361 */ 1362 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1363 tp->t_flags |= TF_ACKNOW; 1364 tcpstat.tcps_rcvwinprobe++; 1365 } else 1366 goto dropafterack; 1367 } else 1368 tcpstat.tcps_rcvbyteafterwin += todrop; 1369 m_adj(m, -todrop); 1370 tlen -= todrop; 1371 tiflags &= ~(TH_PUSH|TH_FIN); 1372 } 1373 1374 /* 1375 * If last ACK falls within this segment's sequence numbers, 1376 * record its timestamp. 1377 * Fix from Braden, see Stevens p. 870 1378 */ 1379 if (opti.ts_present && TSTMP_GEQ(opti.ts_val, tp->ts_recent) && 1380 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1381 tp->ts_recent_age = tcp_now; 1382 tp->ts_recent = opti.ts_val; 1383 } 1384 1385 /* 1386 * If the RST bit is set examine the state: 1387 * SYN_RECEIVED STATE: 1388 * If passive open, return to LISTEN state. 1389 * If active open, inform user that connection was refused. 1390 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES: 1391 * Inform user that connection was reset, and close tcb. 1392 * CLOSING, LAST_ACK, TIME_WAIT STATES 1393 * Close the tcb. 1394 */ 1395 if (tiflags & TH_RST) { 1396 if (th->th_seq != tp->last_ack_sent) 1397 goto drop; 1398 1399 switch (tp->t_state) { 1400 case TCPS_SYN_RECEIVED: 1401 #ifdef TCP_ECN 1402 /* if ECN is enabled, fall back to non-ecn at rexmit */ 1403 if (tcp_do_ecn && !(tp->t_flags & TF_DISABLE_ECN)) 1404 goto drop; 1405 #endif 1406 so->so_error = ECONNREFUSED; 1407 goto close; 1408 1409 case TCPS_ESTABLISHED: 1410 case TCPS_FIN_WAIT_1: 1411 case TCPS_FIN_WAIT_2: 1412 case TCPS_CLOSE_WAIT: 1413 so->so_error = ECONNRESET; 1414 close: 1415 tp->t_state = TCPS_CLOSED; 1416 tcpstat.tcps_drops++; 1417 tp = tcp_close(tp); 1418 goto drop; 1419 case TCPS_CLOSING: 1420 case TCPS_LAST_ACK: 1421 case TCPS_TIME_WAIT: 1422 tp = tcp_close(tp); 1423 goto drop; 1424 } 1425 } 1426 1427 /* 1428 * If a SYN is in the window, then this is an 1429 * error and we ACK and drop the packet. 1430 */ 1431 if (tiflags & TH_SYN) 1432 goto dropafterack_ratelim; 1433 1434 /* 1435 * If the ACK bit is off we drop the segment and return. 1436 */ 1437 if ((tiflags & TH_ACK) == 0) { 1438 if (tp->t_flags & TF_ACKNOW) 1439 goto dropafterack; 1440 else 1441 goto drop; 1442 } 1443 1444 /* 1445 * Ack processing. 1446 */ 1447 switch (tp->t_state) { 1448 1449 /* 1450 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1451 * ESTABLISHED state and continue processing. 1452 * The ACK was checked above. 1453 */ 1454 case TCPS_SYN_RECEIVED: 1455 tcpstat.tcps_connects++; 1456 soisconnected(so); 1457 tp->t_state = TCPS_ESTABLISHED; 1458 TCP_TIMER_ARM(tp, TCPT_KEEP, tcp_keepidle); 1459 /* Do window scaling? */ 1460 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1461 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1462 tp->snd_scale = tp->requested_s_scale; 1463 tp->rcv_scale = tp->request_r_scale; 1464 } 1465 tcp_reass_lock(tp); 1466 (void) tcp_reass(tp, (struct tcphdr *)0, (struct mbuf *)0, 1467 &tlen); 1468 tcp_reass_unlock(tp); 1469 tp->snd_wl1 = th->th_seq - 1; 1470 /* fall into ... */ 1471 1472 /* 1473 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1474 * ACKs. If the ack is in the range 1475 * tp->snd_una < th->th_ack <= tp->snd_max 1476 * then advance tp->snd_una to th->th_ack and drop 1477 * data from the retransmission queue. If this ACK reflects 1478 * more up to date window information we update our window information. 1479 */ 1480 case TCPS_ESTABLISHED: 1481 case TCPS_FIN_WAIT_1: 1482 case TCPS_FIN_WAIT_2: 1483 case TCPS_CLOSE_WAIT: 1484 case TCPS_CLOSING: 1485 case TCPS_LAST_ACK: 1486 case TCPS_TIME_WAIT: 1487 #ifdef TCP_ECN 1488 /* 1489 * if we receive ECE and are not already in recovery phase, 1490 * reduce cwnd by half but don't slow-start. 1491 * advance snd_last to snd_max not to reduce cwnd again 1492 * until all outstanding packets are acked. 1493 */ 1494 if (tcp_do_ecn && (tiflags & TH_ECE)) { 1495 if ((tp->t_flags & TF_ECN_PERMIT) && 1496 SEQ_GEQ(tp->snd_una, tp->snd_last)) { 1497 u_int win; 1498 1499 win = min(tp->snd_wnd, tp->snd_cwnd) / tp->t_maxseg; 1500 if (win > 1) { 1501 tp->snd_ssthresh = win / 2 * tp->t_maxseg; 1502 tp->snd_cwnd = tp->snd_ssthresh; 1503 tp->snd_last = tp->snd_max; 1504 tp->t_flags |= TF_SEND_CWR; 1505 tcpstat.tcps_cwr_ecn++; 1506 } 1507 } 1508 tcpstat.tcps_ecn_rcvece++; 1509 } 1510 /* 1511 * if we receive CWR, we know that the peer has reduced 1512 * its congestion window. stop sending ecn-echo. 1513 */ 1514 if ((tiflags & TH_CWR)) { 1515 tp->t_flags &= ~TF_RCVD_CE; 1516 tcpstat.tcps_ecn_rcvcwr++; 1517 } 1518 #endif /* TCP_ECN */ 1519 1520 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1521 /* 1522 * Duplicate/old ACK processing. 1523 * Increments t_dupacks: 1524 * Pure duplicate (same seq/ack/window, no data) 1525 * Doesn't affect t_dupacks: 1526 * Data packets. 1527 * Normal window updates (window opens) 1528 * Resets t_dupacks: 1529 * New data ACKed. 1530 * Window shrinks 1531 * Old ACK 1532 */ 1533 if (tlen) { 1534 /* Drop very old ACKs unless th_seq matches */ 1535 if (th->th_seq != tp->rcv_nxt && 1536 SEQ_LT(th->th_ack, 1537 tp->snd_una - tp->max_sndwnd)) { 1538 tcpstat.tcps_rcvacktooold++; 1539 goto drop; 1540 } 1541 break; 1542 } 1543 /* 1544 * If we get an old ACK, there is probably packet 1545 * reordering going on. Be conservative and reset 1546 * t_dupacks so that we are less agressive in 1547 * doing a fast retransmit. 1548 */ 1549 if (th->th_ack != tp->snd_una) { 1550 tp->t_dupacks = 0; 1551 break; 1552 } 1553 if (tiwin == tp->snd_wnd) { 1554 tcpstat.tcps_rcvdupack++; 1555 /* 1556 * If we have outstanding data (other than 1557 * a window probe), this is a completely 1558 * duplicate ack (ie, window info didn't 1559 * change), the ack is the biggest we've 1560 * seen and we've seen exactly our rexmt 1561 * threshhold of them, assume a packet 1562 * has been dropped and retransmit it. 1563 * Kludge snd_nxt & the congestion 1564 * window so we send only this one 1565 * packet. 1566 * 1567 * We know we're losing at the current 1568 * window size so do congestion avoidance 1569 * (set ssthresh to half the current window 1570 * and pull our congestion window back to 1571 * the new ssthresh). 1572 * 1573 * Dup acks mean that packets have left the 1574 * network (they're now cached at the receiver) 1575 * so bump cwnd by the amount in the receiver 1576 * to keep a constant cwnd packets in the 1577 * network. 1578 */ 1579 if (TCP_TIMER_ISARMED(tp, TCPT_REXMT) == 0) 1580 tp->t_dupacks = 0; 1581 #if defined(TCP_SACK) && defined(TCP_FACK) 1582 /* 1583 * In FACK, can enter fast rec. if the receiver 1584 * reports a reass. queue longer than 3 segs. 1585 */ 1586 else if (++tp->t_dupacks == tcprexmtthresh || 1587 ((SEQ_GT(tp->snd_fack, tcprexmtthresh * 1588 tp->t_maxseg + tp->snd_una)) && 1589 SEQ_GT(tp->snd_una, tp->snd_last))) { 1590 #else 1591 else if (++tp->t_dupacks == tcprexmtthresh) { 1592 #endif /* TCP_FACK */ 1593 tcp_seq onxt = tp->snd_nxt; 1594 u_long win = 1595 ulmin(tp->snd_wnd, tp->snd_cwnd) / 1596 2 / tp->t_maxseg; 1597 1598 #if defined(TCP_SACK) || defined(TCP_ECN) 1599 if (SEQ_LT(th->th_ack, tp->snd_last)){ 1600 /* 1601 * False fast retx after 1602 * timeout. Do not cut window. 1603 */ 1604 tp->t_dupacks = 0; 1605 goto drop; 1606 } 1607 #endif 1608 if (win < 2) 1609 win = 2; 1610 tp->snd_ssthresh = win * tp->t_maxseg; 1611 #if defined(TCP_SACK) 1612 tp->snd_last = tp->snd_max; 1613 #endif 1614 #ifdef TCP_SACK 1615 if (tp->sack_enable) { 1616 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1617 tp->t_rtttime = 0; 1618 #ifdef TCP_ECN 1619 tp->t_flags |= TF_SEND_CWR; 1620 #endif 1621 #if 1 /* TCP_ECN */ 1622 tcpstat.tcps_cwr_frecovery++; 1623 #endif 1624 tcpstat.tcps_sndrexmitfast++; 1625 #if defined(TCP_SACK) && defined(TCP_FACK) 1626 tp->t_dupacks = tcprexmtthresh; 1627 (void) tcp_output(tp); 1628 /* 1629 * During FR, snd_cwnd is held 1630 * constant for FACK. 1631 */ 1632 tp->snd_cwnd = tp->snd_ssthresh; 1633 #else 1634 /* 1635 * tcp_output() will send 1636 * oldest SACK-eligible rtx. 1637 */ 1638 (void) tcp_output(tp); 1639 tp->snd_cwnd = tp->snd_ssthresh+ 1640 tp->t_maxseg * tp->t_dupacks; 1641 #endif /* TCP_FACK */ 1642 goto drop; 1643 } 1644 #endif /* TCP_SACK */ 1645 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1646 tp->t_rtttime = 0; 1647 tp->snd_nxt = th->th_ack; 1648 tp->snd_cwnd = tp->t_maxseg; 1649 #ifdef TCP_ECN 1650 tp->t_flags |= TF_SEND_CWR; 1651 #endif 1652 #if 1 /* TCP_ECN */ 1653 tcpstat.tcps_cwr_frecovery++; 1654 #endif 1655 tcpstat.tcps_sndrexmitfast++; 1656 (void) tcp_output(tp); 1657 1658 tp->snd_cwnd = tp->snd_ssthresh + 1659 tp->t_maxseg * tp->t_dupacks; 1660 if (SEQ_GT(onxt, tp->snd_nxt)) 1661 tp->snd_nxt = onxt; 1662 goto drop; 1663 } else if (tp->t_dupacks > tcprexmtthresh) { 1664 #if defined(TCP_SACK) && defined(TCP_FACK) 1665 /* 1666 * while (awnd < cwnd) 1667 * sendsomething(); 1668 */ 1669 if (tp->sack_enable) { 1670 if (tp->snd_awnd < tp->snd_cwnd) 1671 tcp_output(tp); 1672 goto drop; 1673 } 1674 #endif /* TCP_FACK */ 1675 tp->snd_cwnd += tp->t_maxseg; 1676 (void) tcp_output(tp); 1677 goto drop; 1678 } 1679 } else if (tiwin < tp->snd_wnd) { 1680 /* 1681 * The window was retracted! Previous dup 1682 * ACKs may have been due to packets arriving 1683 * after the shrunken window, not a missing 1684 * packet, so play it safe and reset t_dupacks 1685 */ 1686 tp->t_dupacks = 0; 1687 } 1688 break; 1689 } 1690 /* 1691 * If the congestion window was inflated to account 1692 * for the other side's cached packets, retract it. 1693 */ 1694 #if defined(TCP_SACK) 1695 if (tp->sack_enable) { 1696 if (tp->t_dupacks >= tcprexmtthresh) { 1697 /* Check for a partial ACK */ 1698 if (tcp_sack_partialack(tp, th)) { 1699 #if defined(TCP_SACK) && defined(TCP_FACK) 1700 /* Force call to tcp_output */ 1701 if (tp->snd_awnd < tp->snd_cwnd) 1702 needoutput = 1; 1703 #else 1704 tp->snd_cwnd += tp->t_maxseg; 1705 needoutput = 1; 1706 #endif /* TCP_FACK */ 1707 } else { 1708 /* Out of fast recovery */ 1709 tp->snd_cwnd = tp->snd_ssthresh; 1710 if (tcp_seq_subtract(tp->snd_max, 1711 th->th_ack) < tp->snd_ssthresh) 1712 tp->snd_cwnd = 1713 tcp_seq_subtract(tp->snd_max, 1714 th->th_ack); 1715 tp->t_dupacks = 0; 1716 #if defined(TCP_SACK) && defined(TCP_FACK) 1717 if (SEQ_GT(th->th_ack, tp->snd_fack)) 1718 tp->snd_fack = th->th_ack; 1719 #endif /* TCP_FACK */ 1720 } 1721 } 1722 } else { 1723 if (tp->t_dupacks >= tcprexmtthresh && 1724 !tcp_newreno(tp, th)) { 1725 /* Out of fast recovery */ 1726 tp->snd_cwnd = tp->snd_ssthresh; 1727 if (tcp_seq_subtract(tp->snd_max, th->th_ack) < 1728 tp->snd_ssthresh) 1729 tp->snd_cwnd = 1730 tcp_seq_subtract(tp->snd_max, 1731 th->th_ack); 1732 tp->t_dupacks = 0; 1733 } 1734 } 1735 if (tp->t_dupacks < tcprexmtthresh) 1736 tp->t_dupacks = 0; 1737 #else /* else no TCP_SACK */ 1738 if (tp->t_dupacks >= tcprexmtthresh && 1739 tp->snd_cwnd > tp->snd_ssthresh) 1740 tp->snd_cwnd = tp->snd_ssthresh; 1741 tp->t_dupacks = 0; 1742 #endif 1743 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1744 tcpstat.tcps_rcvacktoomuch++; 1745 goto dropafterack_ratelim; 1746 } 1747 acked = th->th_ack - tp->snd_una; 1748 tcpstat.tcps_rcvackpack++; 1749 tcpstat.tcps_rcvackbyte += acked; 1750 1751 /* 1752 * If we have a timestamp reply, update smoothed 1753 * round trip time. If no timestamp is present but 1754 * transmit timer is running and timed sequence 1755 * number was acked, update smoothed round trip time. 1756 * Since we now have an rtt measurement, cancel the 1757 * timer backoff (cf., Phil Karn's retransmit alg.). 1758 * Recompute the initial retransmit timer. 1759 */ 1760 if (opti.ts_present) 1761 tcp_xmit_timer(tp, tcp_now-opti.ts_ecr+1); 1762 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1763 tcp_xmit_timer(tp, tcp_now - tp->t_rtttime); 1764 1765 /* 1766 * If all outstanding data is acked, stop retransmit 1767 * timer and remember to restart (more output or persist). 1768 * If there is more data to be acked, restart retransmit 1769 * timer, using current (possibly backed-off) value. 1770 */ 1771 if (th->th_ack == tp->snd_max) { 1772 TCP_TIMER_DISARM(tp, TCPT_REXMT); 1773 needoutput = 1; 1774 } else if (TCP_TIMER_ISARMED(tp, TCPT_PERSIST) == 0) 1775 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 1776 /* 1777 * When new data is acked, open the congestion window. 1778 * If the window gives us less than ssthresh packets 1779 * in flight, open exponentially (maxseg per packet). 1780 * Otherwise open linearly: maxseg per window 1781 * (maxseg^2 / cwnd per packet). 1782 */ 1783 { 1784 u_int cw = tp->snd_cwnd; 1785 u_int incr = tp->t_maxseg; 1786 1787 if (cw > tp->snd_ssthresh) 1788 incr = incr * incr / cw; 1789 #if defined (TCP_SACK) 1790 if (tp->t_dupacks < tcprexmtthresh) 1791 #endif 1792 tp->snd_cwnd = ulmin(cw + incr, TCP_MAXWIN<<tp->snd_scale); 1793 } 1794 ND6_HINT(tp); 1795 if (acked > so->so_snd.sb_cc) { 1796 tp->snd_wnd -= so->so_snd.sb_cc; 1797 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 1798 ourfinisacked = 1; 1799 } else { 1800 sbdrop(&so->so_snd, acked); 1801 tp->snd_wnd -= acked; 1802 ourfinisacked = 0; 1803 } 1804 if (sb_notify(&so->so_snd)) 1805 sowwakeup(so); 1806 tp->snd_una = th->th_ack; 1807 #ifdef TCP_ECN 1808 /* sync snd_last with snd_una */ 1809 if (SEQ_GT(tp->snd_una, tp->snd_last)) 1810 tp->snd_last = tp->snd_una; 1811 #endif 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 tcp_reass_lock(tp); 1980 if (th->th_seq == tp->rcv_nxt && tp->segq.lh_first == NULL && 1981 tp->t_state == TCPS_ESTABLISHED) { 1982 tcp_reass_unlock(tp); 1983 TCP_SETUP_ACK(tp, tiflags); 1984 tp->rcv_nxt += tlen; 1985 tiflags = th->th_flags & TH_FIN; 1986 tcpstat.tcps_rcvpack++; 1987 tcpstat.tcps_rcvbyte += tlen; 1988 ND6_HINT(tp); 1989 if (so->so_state & SS_CANTRCVMORE) 1990 m_freem(m); 1991 else { 1992 m_adj(m, hdroptlen); 1993 sbappendstream(&so->so_rcv, m); 1994 } 1995 sorwakeup(so); 1996 } else { 1997 m_adj(m, hdroptlen); 1998 tiflags = tcp_reass(tp, th, m, &tlen); 1999 tcp_reass_unlock(tp); 2000 tp->t_flags |= TF_ACKNOW; 2001 } 2002 #ifdef TCP_SACK 2003 if (tp->sack_enable) 2004 tcp_update_sack_list(tp); 2005 #endif 2006 2007 /* 2008 * variable len never referenced again in modern BSD, 2009 * so why bother computing it ?? 2010 */ 2011 #if 0 2012 /* 2013 * Note the amount of data that peer has sent into 2014 * our window, in order to estimate the sender's 2015 * buffer size. 2016 */ 2017 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2018 #endif /* 0 */ 2019 } else { 2020 m_freem(m); 2021 tiflags &= ~TH_FIN; 2022 } 2023 2024 /* 2025 * If FIN is received ACK the FIN and let the user know 2026 * that the connection is closing. Ignore a FIN received before 2027 * the connection is fully established. 2028 */ 2029 if ((tiflags & TH_FIN) && TCPS_HAVEESTABLISHED(tp->t_state)) { 2030 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2031 socantrcvmore(so); 2032 tp->t_flags |= TF_ACKNOW; 2033 tp->rcv_nxt++; 2034 } 2035 switch (tp->t_state) { 2036 2037 /* 2038 * In ESTABLISHED STATE enter the CLOSE_WAIT state. 2039 */ 2040 case TCPS_ESTABLISHED: 2041 tp->t_state = TCPS_CLOSE_WAIT; 2042 break; 2043 2044 /* 2045 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2046 * enter the CLOSING state. 2047 */ 2048 case TCPS_FIN_WAIT_1: 2049 tp->t_state = TCPS_CLOSING; 2050 break; 2051 2052 /* 2053 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2054 * starting the time-wait timer, turning off the other 2055 * standard timers. 2056 */ 2057 case TCPS_FIN_WAIT_2: 2058 tp->t_state = TCPS_TIME_WAIT; 2059 tcp_canceltimers(tp); 2060 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2061 soisdisconnected(so); 2062 break; 2063 2064 /* 2065 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2066 */ 2067 case TCPS_TIME_WAIT: 2068 TCP_TIMER_ARM(tp, TCPT_2MSL, 2 * TCPTV_MSL); 2069 break; 2070 } 2071 } 2072 if (so->so_options & SO_DEBUG) { 2073 switch (tp->pf) { 2074 #ifdef INET6 2075 case PF_INET6: 2076 tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti6, 2077 0, tlen); 2078 break; 2079 #endif /* INET6 */ 2080 case PF_INET: 2081 tcp_trace(TA_INPUT, ostate, tp, (caddr_t) &tcp_saveti, 2082 0, tlen); 2083 break; 2084 } 2085 } 2086 2087 /* 2088 * Return any desired output. 2089 */ 2090 if (needoutput || (tp->t_flags & TF_ACKNOW)) { 2091 (void) tcp_output(tp); 2092 } 2093 return; 2094 2095 badsyn: 2096 /* 2097 * Received a bad SYN. Increment counters and dropwithreset. 2098 */ 2099 tcpstat.tcps_badsyn++; 2100 tp = NULL; 2101 goto dropwithreset; 2102 2103 dropafterack_ratelim: 2104 if (ppsratecheck(&tcp_ackdrop_ppslim_last, &tcp_ackdrop_ppslim_count, 2105 tcp_ackdrop_ppslim) == 0) { 2106 /* XXX stat */ 2107 goto drop; 2108 } 2109 /* ...fall into dropafterack... */ 2110 2111 dropafterack: 2112 /* 2113 * Generate an ACK dropping incoming segment if it occupies 2114 * sequence space, where the ACK reflects our state. 2115 */ 2116 if (tiflags & TH_RST) 2117 goto drop; 2118 m_freem(m); 2119 tp->t_flags |= TF_ACKNOW; 2120 (void) tcp_output(tp); 2121 return; 2122 2123 dropwithreset_ratelim: 2124 /* 2125 * We may want to rate-limit RSTs in certain situations, 2126 * particularly if we are sending an RST in response to 2127 * an attempt to connect to or otherwise communicate with 2128 * a port for which we have no socket. 2129 */ 2130 if (ppsratecheck(&tcp_rst_ppslim_last, &tcp_rst_ppslim_count, 2131 tcp_rst_ppslim) == 0) { 2132 /* XXX stat */ 2133 goto drop; 2134 } 2135 /* ...fall into dropwithreset... */ 2136 2137 dropwithreset: 2138 /* 2139 * Generate a RST, dropping incoming segment. 2140 * Make ACK acceptable to originator of segment. 2141 * Don't bother to respond to RST. 2142 */ 2143 if (tiflags & TH_RST) 2144 goto drop; 2145 if (tiflags & TH_ACK) { 2146 tcp_respond(tp, mtod(m, caddr_t), m, (tcp_seq)0, th->th_ack, 2147 TH_RST); 2148 } else { 2149 if (tiflags & TH_SYN) 2150 tlen++; 2151 tcp_respond(tp, mtod(m, caddr_t), m, th->th_seq + tlen, 2152 (tcp_seq)0, TH_RST|TH_ACK); 2153 } 2154 return; 2155 2156 drop: 2157 /* 2158 * Drop space held by incoming segment and return. 2159 */ 2160 if (tp && (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) { 2161 switch (tp->pf) { 2162 #ifdef INET6 2163 case PF_INET6: 2164 tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti6, 2165 0, tlen); 2166 break; 2167 #endif /* INET6 */ 2168 case PF_INET: 2169 tcp_trace(TA_DROP, ostate, tp, (caddr_t) &tcp_saveti, 2170 0, tlen); 2171 break; 2172 } 2173 } 2174 2175 m_freem(m); 2176 return; 2177 #ifndef TUBA_INCLUDE 2178 } 2179 2180 int 2181 tcp_dooptions(tp, cp, cnt, th, m, iphlen, oi) 2182 struct tcpcb *tp; 2183 u_char *cp; 2184 int cnt; 2185 struct tcphdr *th; 2186 struct mbuf *m; 2187 int iphlen; 2188 struct tcp_opt_info *oi; 2189 { 2190 u_int16_t mss = 0; 2191 int opt, optlen; 2192 #ifdef TCP_SIGNATURE 2193 caddr_t sigp = NULL; 2194 struct tdb *tdb = NULL; 2195 #endif /* TCP_SIGNATURE */ 2196 2197 #ifdef TCP_SIGNATURE 2198 if (cp) 2199 #endif /* TCP_SIGNATURE */ 2200 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2201 opt = cp[0]; 2202 if (opt == TCPOPT_EOL) 2203 break; 2204 if (opt == TCPOPT_NOP) 2205 optlen = 1; 2206 else { 2207 if (cnt < 2) 2208 break; 2209 optlen = cp[1]; 2210 if (optlen < 2 || optlen > cnt) 2211 break; 2212 } 2213 switch (opt) { 2214 2215 default: 2216 continue; 2217 2218 case TCPOPT_MAXSEG: 2219 if (optlen != TCPOLEN_MAXSEG) 2220 continue; 2221 if (!(th->th_flags & TH_SYN)) 2222 continue; 2223 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss)); 2224 NTOHS(mss); 2225 oi->maxseg = mss; 2226 break; 2227 2228 case TCPOPT_WINDOW: 2229 if (optlen != TCPOLEN_WINDOW) 2230 continue; 2231 if (!(th->th_flags & TH_SYN)) 2232 continue; 2233 tp->t_flags |= TF_RCVD_SCALE; 2234 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2235 break; 2236 2237 case TCPOPT_TIMESTAMP: 2238 if (optlen != TCPOLEN_TIMESTAMP) 2239 continue; 2240 oi->ts_present = 1; 2241 bcopy(cp + 2, &oi->ts_val, sizeof(oi->ts_val)); 2242 NTOHL(oi->ts_val); 2243 bcopy(cp + 6, &oi->ts_ecr, sizeof(oi->ts_ecr)); 2244 NTOHL(oi->ts_ecr); 2245 2246 /* 2247 * A timestamp received in a SYN makes 2248 * it ok to send timestamp requests and replies. 2249 */ 2250 if (th->th_flags & TH_SYN) { 2251 tp->t_flags |= TF_RCVD_TSTMP; 2252 tp->ts_recent = oi->ts_val; 2253 tp->ts_recent_age = tcp_now; 2254 } 2255 break; 2256 2257 #ifdef TCP_SACK 2258 case TCPOPT_SACK_PERMITTED: 2259 if (!tp->sack_enable || optlen!=TCPOLEN_SACK_PERMITTED) 2260 continue; 2261 if (th->th_flags & TH_SYN) 2262 /* MUST only be set on SYN */ 2263 tp->t_flags |= TF_SACK_PERMIT; 2264 break; 2265 case TCPOPT_SACK: 2266 if (tcp_sack_option(tp, th, cp, optlen)) 2267 continue; 2268 break; 2269 #endif 2270 #ifdef TCP_SIGNATURE 2271 case TCPOPT_SIGNATURE: 2272 if (optlen != TCPOLEN_SIGNATURE) 2273 continue; 2274 2275 if (sigp && bcmp(sigp, cp + 2, 16)) 2276 return (-1); 2277 2278 sigp = cp + 2; 2279 break; 2280 #endif /* TCP_SIGNATURE */ 2281 } 2282 } 2283 2284 #ifdef TCP_SIGNATURE 2285 if (tp->t_flags & TF_SIGNATURE) { 2286 union sockaddr_union src, dst; 2287 2288 memset(&src, 0, sizeof(union sockaddr_union)); 2289 memset(&dst, 0, sizeof(union sockaddr_union)); 2290 2291 switch (tp->pf) { 2292 case 0: 2293 #ifdef INET 2294 case AF_INET: 2295 src.sa.sa_len = sizeof(struct sockaddr_in); 2296 src.sa.sa_family = AF_INET; 2297 src.sin.sin_addr = mtod(m, struct ip *)->ip_src; 2298 dst.sa.sa_len = sizeof(struct sockaddr_in); 2299 dst.sa.sa_family = AF_INET; 2300 dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst; 2301 break; 2302 #endif 2303 #ifdef INET6 2304 case AF_INET6: 2305 src.sa.sa_len = sizeof(struct sockaddr_in6); 2306 src.sa.sa_family = AF_INET6; 2307 src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src; 2308 dst.sa.sa_len = sizeof(struct sockaddr_in6); 2309 dst.sa.sa_family = AF_INET6; 2310 dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst; 2311 break; 2312 #endif /* INET6 */ 2313 } 2314 2315 tdb = gettdbbysrcdst(0, &src, &dst, IPPROTO_TCP); 2316 2317 /* 2318 * We don't have an SA for this peer, so we turn off 2319 * TF_SIGNATURE on the listen socket 2320 */ 2321 if (tdb == NULL && tp->t_state == TCPS_LISTEN) 2322 tp->t_flags &= ~TF_SIGNATURE; 2323 2324 } 2325 2326 if ((sigp ? TF_SIGNATURE : 0) ^ (tp->t_flags & TF_SIGNATURE)) { 2327 tcpstat.tcps_rcvbadsig++; 2328 return (-1); 2329 } 2330 2331 if (sigp) { 2332 MD5_CTX ctx; 2333 char sig[16]; 2334 2335 if (tdb == NULL) { 2336 tcpstat.tcps_rcvbadsig++; 2337 return (-1); 2338 } 2339 2340 MD5Init(&ctx); 2341 2342 switch(tp->pf) { 2343 case 0: 2344 #ifdef INET 2345 case AF_INET: 2346 { 2347 struct ippseudo ippseudo; 2348 2349 ippseudo.ippseudo_src = 2350 mtod(m, struct ip *)->ip_src; 2351 ippseudo.ippseudo_dst = 2352 mtod(m, struct ip *)->ip_dst; 2353 ippseudo.ippseudo_pad = 0; 2354 ippseudo.ippseudo_p = IPPROTO_TCP; 2355 ippseudo.ippseudo_len = htons( 2356 m->m_pkthdr.len - iphlen); 2357 2358 MD5Update(&ctx, (char *)&ippseudo, 2359 sizeof(struct ippseudo)); 2360 } 2361 break; 2362 #endif /* INET */ 2363 #ifdef INET6 2364 case AF_INET6: 2365 { 2366 struct ip6_hdr_pseudo ip6pseudo; 2367 2368 bzero(&ip6pseudo, sizeof(ip6pseudo)); 2369 ip6pseudo.ip6ph_src = 2370 mtod(m, struct ip6_hdr *)->ip6_src; 2371 ip6pseudo.ip6ph_dst = 2372 mtod(m, struct ip6_hdr *)->ip6_dst; 2373 in6_clearscope(&ip6pseudo.ip6ph_src); 2374 in6_clearscope(&ip6pseudo.ip6ph_dst); 2375 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 2376 ip6pseudo.ip6ph_len = htonl(m->m_pkthdr.len - 2377 iphlen); 2378 2379 MD5Update(&ctx, (char *)&ip6pseudo, 2380 sizeof(ip6pseudo)); 2381 } 2382 break; 2383 #endif /* INET6 */ 2384 } 2385 2386 { 2387 struct tcphdr tcphdr; 2388 2389 tcphdr.th_sport = th->th_sport; 2390 tcphdr.th_dport = th->th_dport; 2391 tcphdr.th_seq = htonl(th->th_seq); 2392 tcphdr.th_ack = htonl(th->th_ack); 2393 tcphdr.th_off = th->th_off; 2394 tcphdr.th_x2 = th->th_x2; 2395 tcphdr.th_flags = th->th_flags; 2396 tcphdr.th_win = htons(th->th_win); 2397 tcphdr.th_sum = 0; 2398 tcphdr.th_urp = htons(th->th_urp); 2399 2400 MD5Update(&ctx, (char *)&tcphdr, 2401 sizeof(struct tcphdr)); 2402 } 2403 2404 if (m_apply(m, iphlen + th->th_off * sizeof(uint32_t), 2405 m->m_pkthdr.len - (iphlen + th->th_off * sizeof(uint32_t)), 2406 tcp_signature_apply, (caddr_t)&ctx)) 2407 return (-1); 2408 2409 MD5Update(&ctx, tdb->tdb_amxkey, tdb->tdb_amxkeylen); 2410 MD5Final(sig, &ctx); 2411 2412 if (bcmp(sig, sigp, 16)) { 2413 tcpstat.tcps_rcvbadsig++; 2414 return (-1); 2415 } 2416 2417 tcpstat.tcps_rcvgoodsig++; 2418 } 2419 #endif /* TCP_SIGNATURE */ 2420 2421 return (0); 2422 } 2423 2424 #if defined(TCP_SACK) 2425 u_long 2426 tcp_seq_subtract(a, b) 2427 u_long a, b; 2428 { 2429 return ((long)(a - b)); 2430 } 2431 #endif 2432 2433 2434 #ifdef TCP_SACK 2435 /* 2436 * This function is called upon receipt of new valid data (while not in header 2437 * prediction mode), and it updates the ordered list of sacks. 2438 */ 2439 void 2440 tcp_update_sack_list(tp) 2441 struct tcpcb *tp; 2442 { 2443 /* 2444 * First reported block MUST be the most recent one. Subsequent 2445 * blocks SHOULD be in the order in which they arrived at the 2446 * receiver. These two conditions make the implementation fully 2447 * compliant with RFC 2018. 2448 */ 2449 int i, j = 0, count = 0, lastpos = -1; 2450 struct sackblk sack, firstsack, temp[MAX_SACK_BLKS]; 2451 2452 /* First clean up current list of sacks */ 2453 for (i = 0; i < tp->rcv_numsacks; i++) { 2454 sack = tp->sackblks[i]; 2455 if (sack.start == 0 && sack.end == 0) { 2456 count++; /* count = number of blocks to be discarded */ 2457 continue; 2458 } 2459 if (SEQ_LEQ(sack.end, tp->rcv_nxt)) { 2460 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2461 count++; 2462 } else { 2463 temp[j].start = tp->sackblks[i].start; 2464 temp[j++].end = tp->sackblks[i].end; 2465 } 2466 } 2467 tp->rcv_numsacks -= count; 2468 if (tp->rcv_numsacks == 0) { /* no sack blocks currently (fast path) */ 2469 tcp_clean_sackreport(tp); 2470 if (SEQ_LT(tp->rcv_nxt, tp->rcv_laststart)) { 2471 /* ==> need first sack block */ 2472 tp->sackblks[0].start = tp->rcv_laststart; 2473 tp->sackblks[0].end = tp->rcv_lastend; 2474 tp->rcv_numsacks = 1; 2475 } 2476 return; 2477 } 2478 /* Otherwise, sack blocks are already present. */ 2479 for (i = 0; i < tp->rcv_numsacks; i++) 2480 tp->sackblks[i] = temp[i]; /* first copy back sack list */ 2481 if (SEQ_GEQ(tp->rcv_nxt, tp->rcv_lastend)) 2482 return; /* sack list remains unchanged */ 2483 /* 2484 * From here, segment just received should be (part of) the 1st sack. 2485 * Go through list, possibly coalescing sack block entries. 2486 */ 2487 firstsack.start = tp->rcv_laststart; 2488 firstsack.end = tp->rcv_lastend; 2489 for (i = 0; i < tp->rcv_numsacks; i++) { 2490 sack = tp->sackblks[i]; 2491 if (SEQ_LT(sack.end, firstsack.start) || 2492 SEQ_GT(sack.start, firstsack.end)) 2493 continue; /* no overlap */ 2494 if (sack.start == firstsack.start && sack.end == firstsack.end){ 2495 /* 2496 * identical block; delete it here since we will 2497 * move it to the front of the list. 2498 */ 2499 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2500 lastpos = i; /* last posn with a zero entry */ 2501 continue; 2502 } 2503 if (SEQ_LEQ(sack.start, firstsack.start)) 2504 firstsack.start = sack.start; /* merge blocks */ 2505 if (SEQ_GEQ(sack.end, firstsack.end)) 2506 firstsack.end = sack.end; /* merge blocks */ 2507 tp->sackblks[i].start = tp->sackblks[i].end = 0; 2508 lastpos = i; /* last posn with a zero entry */ 2509 } 2510 if (lastpos != -1) { /* at least one merge */ 2511 for (i = 0, j = 1; i < tp->rcv_numsacks; i++) { 2512 sack = tp->sackblks[i]; 2513 if (sack.start == 0 && sack.end == 0) 2514 continue; 2515 temp[j++] = sack; 2516 } 2517 tp->rcv_numsacks = j; /* including first blk (added later) */ 2518 for (i = 1; i < tp->rcv_numsacks; i++) /* now copy back */ 2519 tp->sackblks[i] = temp[i]; 2520 } else { /* no merges -- shift sacks by 1 */ 2521 if (tp->rcv_numsacks < MAX_SACK_BLKS) 2522 tp->rcv_numsacks++; 2523 for (i = tp->rcv_numsacks-1; i > 0; i--) 2524 tp->sackblks[i] = tp->sackblks[i-1]; 2525 } 2526 tp->sackblks[0] = firstsack; 2527 return; 2528 } 2529 2530 /* 2531 * Process the TCP SACK option. Returns 1 if tcp_dooptions() should continue, 2532 * and 0 otherwise, if the option was fine. tp->snd_holes is an ordered list 2533 * of holes (oldest to newest, in terms of the sequence space). 2534 */ 2535 int 2536 tcp_sack_option(struct tcpcb *tp, struct tcphdr *th, u_char *cp, int optlen) 2537 { 2538 int tmp_olen; 2539 u_char *tmp_cp; 2540 struct sackhole *cur, *p, *temp; 2541 2542 if (!tp->sack_enable) 2543 return (1); 2544 2545 /* Note: TCPOLEN_SACK must be 2*sizeof(tcp_seq) */ 2546 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2547 return (1); 2548 tmp_cp = cp + 2; 2549 tmp_olen = optlen - 2; 2550 if (tp->snd_numholes < 0) 2551 tp->snd_numholes = 0; 2552 if (tp->t_maxseg == 0) 2553 panic("tcp_sack_option"); /* Should never happen */ 2554 while (tmp_olen > 0) { 2555 struct sackblk sack; 2556 2557 bcopy(tmp_cp, (char *) &(sack.start), sizeof(tcp_seq)); 2558 NTOHL(sack.start); 2559 bcopy(tmp_cp + sizeof(tcp_seq), 2560 (char *) &(sack.end), sizeof(tcp_seq)); 2561 NTOHL(sack.end); 2562 tmp_olen -= TCPOLEN_SACK; 2563 tmp_cp += TCPOLEN_SACK; 2564 if (SEQ_LEQ(sack.end, sack.start)) 2565 continue; /* bad SACK fields */ 2566 if (SEQ_LEQ(sack.end, tp->snd_una)) 2567 continue; /* old block */ 2568 #if defined(TCP_SACK) && defined(TCP_FACK) 2569 /* Updates snd_fack. */ 2570 if (SEQ_GT(sack.end, tp->snd_fack)) 2571 tp->snd_fack = sack.end; 2572 #endif /* TCP_FACK */ 2573 if (SEQ_GT(th->th_ack, tp->snd_una)) { 2574 if (SEQ_LT(sack.start, th->th_ack)) 2575 continue; 2576 } 2577 if (SEQ_GT(sack.end, tp->snd_max)) 2578 continue; 2579 if (tp->snd_holes == NULL) { /* first hole */ 2580 tp->snd_holes = (struct sackhole *) 2581 pool_get(&sackhl_pool, PR_NOWAIT); 2582 if (tp->snd_holes == NULL) { 2583 /* ENOBUFS, so ignore SACKed block for now*/ 2584 continue; 2585 } 2586 cur = tp->snd_holes; 2587 cur->start = th->th_ack; 2588 cur->end = sack.start; 2589 cur->rxmit = cur->start; 2590 cur->next = NULL; 2591 tp->snd_numholes = 1; 2592 tp->rcv_lastsack = sack.end; 2593 /* 2594 * dups is at least one. If more data has been 2595 * SACKed, it can be greater than one. 2596 */ 2597 cur->dups = min(tcprexmtthresh, 2598 ((sack.end - cur->end)/tp->t_maxseg)); 2599 if (cur->dups < 1) 2600 cur->dups = 1; 2601 continue; /* with next sack block */ 2602 } 2603 /* Go thru list of holes: p = previous, cur = current */ 2604 p = cur = tp->snd_holes; 2605 while (cur) { 2606 if (SEQ_LEQ(sack.end, cur->start)) 2607 /* SACKs data before the current hole */ 2608 break; /* no use going through more holes */ 2609 if (SEQ_GEQ(sack.start, cur->end)) { 2610 /* SACKs data beyond the current hole */ 2611 cur->dups++; 2612 if (((sack.end - cur->end)/tp->t_maxseg) >= 2613 tcprexmtthresh) 2614 cur->dups = tcprexmtthresh; 2615 p = cur; 2616 cur = cur->next; 2617 continue; 2618 } 2619 if (SEQ_LEQ(sack.start, cur->start)) { 2620 /* Data acks at least the beginning of hole */ 2621 #if defined(TCP_SACK) && defined(TCP_FACK) 2622 if (SEQ_GT(sack.end, cur->rxmit)) 2623 tp->retran_data -= 2624 tcp_seq_subtract(cur->rxmit, 2625 cur->start); 2626 else 2627 tp->retran_data -= 2628 tcp_seq_subtract(sack.end, 2629 cur->start); 2630 #endif /* TCP_FACK */ 2631 if (SEQ_GEQ(sack.end, cur->end)) { 2632 /* Acks entire hole, so delete hole */ 2633 if (p != cur) { 2634 p->next = cur->next; 2635 pool_put(&sackhl_pool, cur); 2636 cur = p->next; 2637 } else { 2638 cur = cur->next; 2639 pool_put(&sackhl_pool, p); 2640 p = cur; 2641 tp->snd_holes = p; 2642 } 2643 tp->snd_numholes--; 2644 continue; 2645 } 2646 /* otherwise, move start of hole forward */ 2647 cur->start = sack.end; 2648 cur->rxmit = max (cur->rxmit, cur->start); 2649 p = cur; 2650 cur = cur->next; 2651 continue; 2652 } 2653 /* move end of hole backward */ 2654 if (SEQ_GEQ(sack.end, cur->end)) { 2655 #if defined(TCP_SACK) && defined(TCP_FACK) 2656 if (SEQ_GT(cur->rxmit, sack.start)) 2657 tp->retran_data -= 2658 tcp_seq_subtract(cur->rxmit, 2659 sack.start); 2660 #endif /* TCP_FACK */ 2661 cur->end = sack.start; 2662 cur->rxmit = min(cur->rxmit, cur->end); 2663 cur->dups++; 2664 if (((sack.end - cur->end)/tp->t_maxseg) >= 2665 tcprexmtthresh) 2666 cur->dups = tcprexmtthresh; 2667 p = cur; 2668 cur = cur->next; 2669 continue; 2670 } 2671 if (SEQ_LT(cur->start, sack.start) && 2672 SEQ_GT(cur->end, sack.end)) { 2673 /* 2674 * ACKs some data in middle of a hole; need to 2675 * split current hole 2676 */ 2677 temp = (struct sackhole *) 2678 pool_get(&sackhl_pool, PR_NOWAIT); 2679 if (temp == NULL) 2680 continue; /* ENOBUFS */ 2681 #if defined(TCP_SACK) && defined(TCP_FACK) 2682 if (SEQ_GT(cur->rxmit, sack.end)) 2683 tp->retran_data -= 2684 tcp_seq_subtract(sack.end, 2685 sack.start); 2686 else if (SEQ_GT(cur->rxmit, sack.start)) 2687 tp->retran_data -= 2688 tcp_seq_subtract(cur->rxmit, 2689 sack.start); 2690 #endif /* TCP_FACK */ 2691 temp->next = cur->next; 2692 temp->start = sack.end; 2693 temp->end = cur->end; 2694 temp->dups = cur->dups; 2695 temp->rxmit = max(cur->rxmit, temp->start); 2696 cur->end = sack.start; 2697 cur->rxmit = min(cur->rxmit, cur->end); 2698 cur->dups++; 2699 if (((sack.end - cur->end)/tp->t_maxseg) >= 2700 tcprexmtthresh) 2701 cur->dups = tcprexmtthresh; 2702 cur->next = temp; 2703 p = temp; 2704 cur = p->next; 2705 tp->snd_numholes++; 2706 } 2707 } 2708 /* At this point, p points to the last hole on the list */ 2709 if (SEQ_LT(tp->rcv_lastsack, sack.start)) { 2710 /* 2711 * Need to append new hole at end. 2712 * Last hole is p (and it's not NULL). 2713 */ 2714 temp = (struct sackhole *) 2715 pool_get(&sackhl_pool, PR_NOWAIT); 2716 if (temp == NULL) 2717 continue; /* ENOBUFS */ 2718 temp->start = tp->rcv_lastsack; 2719 temp->end = sack.start; 2720 temp->dups = min(tcprexmtthresh, 2721 ((sack.end - sack.start)/tp->t_maxseg)); 2722 if (temp->dups < 1) 2723 temp->dups = 1; 2724 temp->rxmit = temp->start; 2725 temp->next = 0; 2726 p->next = temp; 2727 tp->rcv_lastsack = sack.end; 2728 tp->snd_numholes++; 2729 } 2730 } 2731 #if defined(TCP_SACK) && defined(TCP_FACK) 2732 /* 2733 * Update retran_data and snd_awnd. Go through the list of 2734 * holes. Increment retran_data by (hole->rxmit - hole->start). 2735 */ 2736 tp->retran_data = 0; 2737 cur = tp->snd_holes; 2738 while (cur) { 2739 tp->retran_data += cur->rxmit - cur->start; 2740 cur = cur->next; 2741 } 2742 tp->snd_awnd = tcp_seq_subtract(tp->snd_nxt, tp->snd_fack) + 2743 tp->retran_data; 2744 #endif /* TCP_FACK */ 2745 2746 return (0); 2747 } 2748 2749 /* 2750 * Delete stale (i.e, cumulatively ack'd) holes. Hole is deleted only if 2751 * it is completely acked; otherwise, tcp_sack_option(), called from 2752 * tcp_dooptions(), will fix up the hole. 2753 */ 2754 void 2755 tcp_del_sackholes(tp, th) 2756 struct tcpcb *tp; 2757 struct tcphdr *th; 2758 { 2759 if (tp->sack_enable && tp->t_state != TCPS_LISTEN) { 2760 /* max because this could be an older ack just arrived */ 2761 tcp_seq lastack = SEQ_GT(th->th_ack, tp->snd_una) ? 2762 th->th_ack : tp->snd_una; 2763 struct sackhole *cur = tp->snd_holes; 2764 struct sackhole *prev; 2765 while (cur) 2766 if (SEQ_LEQ(cur->end, lastack)) { 2767 prev = cur; 2768 cur = cur->next; 2769 pool_put(&sackhl_pool, prev); 2770 tp->snd_numholes--; 2771 } else if (SEQ_LT(cur->start, lastack)) { 2772 cur->start = lastack; 2773 if (SEQ_LT(cur->rxmit, cur->start)) 2774 cur->rxmit = cur->start; 2775 break; 2776 } else 2777 break; 2778 tp->snd_holes = cur; 2779 } 2780 } 2781 2782 /* 2783 * Delete all receiver-side SACK information. 2784 */ 2785 void 2786 tcp_clean_sackreport(tp) 2787 struct tcpcb *tp; 2788 { 2789 int i; 2790 2791 tp->rcv_numsacks = 0; 2792 for (i = 0; i < MAX_SACK_BLKS; i++) 2793 tp->sackblks[i].start = tp->sackblks[i].end=0; 2794 2795 } 2796 2797 /* 2798 * Checks for partial ack. If partial ack arrives, turn off retransmission 2799 * timer, deflate the window, do not clear tp->t_dupacks, and return 1. 2800 * If the ack advances at least to tp->snd_last, return 0. 2801 */ 2802 int 2803 tcp_sack_partialack(tp, th) 2804 struct tcpcb *tp; 2805 struct tcphdr *th; 2806 { 2807 if (SEQ_LT(th->th_ack, tp->snd_last)) { 2808 /* Turn off retx. timer (will start again next segment) */ 2809 TCP_TIMER_DISARM(tp, TCPT_REXMT); 2810 tp->t_rtttime = 0; 2811 #ifndef TCP_FACK 2812 /* 2813 * Partial window deflation. This statement relies on the 2814 * fact that tp->snd_una has not been updated yet. In FACK 2815 * hold snd_cwnd constant during fast recovery. 2816 */ 2817 if (tp->snd_cwnd > (th->th_ack - tp->snd_una)) { 2818 tp->snd_cwnd -= th->th_ack - tp->snd_una; 2819 tp->snd_cwnd += tp->t_maxseg; 2820 } else 2821 tp->snd_cwnd = tp->t_maxseg; 2822 #endif 2823 return (1); 2824 } 2825 return (0); 2826 } 2827 #endif /* TCP_SACK */ 2828 2829 /* 2830 * Pull out of band byte out of a segment so 2831 * it doesn't appear in the user's data queue. 2832 * It is still reflected in the segment length for 2833 * sequencing purposes. 2834 */ 2835 void 2836 tcp_pulloutofband(so, urgent, m, off) 2837 struct socket *so; 2838 u_int urgent; 2839 struct mbuf *m; 2840 int off; 2841 { 2842 int cnt = off + urgent - 1; 2843 2844 while (cnt >= 0) { 2845 if (m->m_len > cnt) { 2846 char *cp = mtod(m, caddr_t) + cnt; 2847 struct tcpcb *tp = sototcpcb(so); 2848 2849 tp->t_iobc = *cp; 2850 tp->t_oobflags |= TCPOOB_HAVEDATA; 2851 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2852 m->m_len--; 2853 return; 2854 } 2855 cnt -= m->m_len; 2856 m = m->m_next; 2857 if (m == 0) 2858 break; 2859 } 2860 panic("tcp_pulloutofband"); 2861 } 2862 2863 /* 2864 * Collect new round-trip time estimate 2865 * and update averages and current timeout. 2866 */ 2867 void 2868 tcp_xmit_timer(tp, rtt) 2869 struct tcpcb *tp; 2870 short rtt; 2871 { 2872 short delta; 2873 short rttmin; 2874 2875 tcpstat.tcps_rttupdated++; 2876 --rtt; 2877 if (tp->t_srtt != 0) { 2878 /* 2879 * srtt is stored as fixed point with 3 bits after the 2880 * binary point (i.e., scaled by 8). The following magic 2881 * is equivalent to the smoothing algorithm in rfc793 with 2882 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2883 * point). Adjust rtt to origin 0. 2884 */ 2885 delta = (rtt << 2) - (tp->t_srtt >> TCP_RTT_SHIFT); 2886 if ((tp->t_srtt += delta) <= 0) 2887 tp->t_srtt = 1; 2888 /* 2889 * We accumulate a smoothed rtt variance (actually, a 2890 * smoothed mean difference), then set the retransmit 2891 * timer to smoothed rtt + 4 times the smoothed variance. 2892 * rttvar is stored as fixed point with 2 bits after the 2893 * binary point (scaled by 4). The following is 2894 * equivalent to rfc793 smoothing with an alpha of .75 2895 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2896 * rfc793's wired-in beta. 2897 */ 2898 if (delta < 0) 2899 delta = -delta; 2900 delta -= (tp->t_rttvar >> TCP_RTTVAR_SHIFT); 2901 if ((tp->t_rttvar += delta) <= 0) 2902 tp->t_rttvar = 1; 2903 } else { 2904 /* 2905 * No rtt measurement yet - use the unsmoothed rtt. 2906 * Set the variance to half the rtt (so our first 2907 * retransmit happens at 3*rtt). 2908 */ 2909 tp->t_srtt = rtt << (TCP_RTT_SHIFT + 2); 2910 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT + 2 - 1); 2911 } 2912 tp->t_rtttime = 0; 2913 tp->t_rxtshift = 0; 2914 2915 /* 2916 * the retransmit should happen at rtt + 4 * rttvar. 2917 * Because of the way we do the smoothing, srtt and rttvar 2918 * will each average +1/2 tick of bias. When we compute 2919 * the retransmit timer, we want 1/2 tick of rounding and 2920 * 1 extra tick because of +-1/2 tick uncertainty in the 2921 * firing of the timer. The bias will give us exactly the 2922 * 1.5 tick we need. But, because the bias is 2923 * statistical, we have to test that we don't drop below 2924 * the minimum feasible timer (which is 2 ticks). 2925 */ 2926 if (tp->t_rttmin > rtt + 2) 2927 rttmin = tp->t_rttmin; 2928 else 2929 rttmin = rtt + 2; 2930 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), rttmin, TCPTV_REXMTMAX); 2931 2932 /* 2933 * We received an ack for a packet that wasn't retransmitted; 2934 * it is probably safe to discard any error indications we've 2935 * received recently. This isn't quite right, but close enough 2936 * for now (a route might have failed after we sent a segment, 2937 * and the return path might not be symmetrical). 2938 */ 2939 tp->t_softerror = 0; 2940 } 2941 2942 /* 2943 * Determine a reasonable value for maxseg size. 2944 * If the route is known, check route for mtu. 2945 * If none, use an mss that can be handled on the outgoing 2946 * interface without forcing IP to fragment; if bigger than 2947 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2948 * to utilize large mbufs. If no route is found, route has no mtu, 2949 * or the destination isn't local, use a default, hopefully conservative 2950 * size (usually 512 or the default IP max size, but no more than the mtu 2951 * of the interface), as we can't discover anything about intervening 2952 * gateways or networks. We also initialize the congestion/slow start 2953 * window to be a single segment if the destination isn't local. 2954 * While looking at the routing entry, we also initialize other path-dependent 2955 * parameters from pre-set or cached values in the routing entry. 2956 * 2957 * Also take into account the space needed for options that we 2958 * send regularly. Make maxseg shorter by that amount to assure 2959 * that we can send maxseg amount of data even when the options 2960 * are present. Store the upper limit of the length of options plus 2961 * data in maxopd. 2962 * 2963 * NOTE: offer == -1 indicates that the maxseg size changed due to 2964 * Path MTU discovery. 2965 */ 2966 int 2967 tcp_mss(tp, offer) 2968 struct tcpcb *tp; 2969 int offer; 2970 { 2971 struct rtentry *rt; 2972 struct ifnet *ifp; 2973 int mss, mssopt; 2974 int iphlen; 2975 struct inpcb *inp; 2976 2977 inp = tp->t_inpcb; 2978 2979 mssopt = mss = tcp_mssdflt; 2980 2981 rt = in_pcbrtentry(inp); 2982 2983 if (rt == NULL) 2984 goto out; 2985 2986 ifp = rt->rt_ifp; 2987 2988 switch (tp->pf) { 2989 #ifdef INET6 2990 case AF_INET6: 2991 iphlen = sizeof(struct ip6_hdr); 2992 break; 2993 #endif 2994 case AF_INET: 2995 iphlen = sizeof(struct ip); 2996 break; 2997 default: 2998 /* the family does not support path MTU discovery */ 2999 goto out; 3000 } 3001 3002 #ifdef RTV_MTU 3003 /* 3004 * if there's an mtu associated with the route and we support 3005 * path MTU discovery for the underlying protocol family, use it. 3006 */ 3007 if (rt->rt_rmx.rmx_mtu) { 3008 /* 3009 * One may wish to lower MSS to take into account options, 3010 * especially security-related options. 3011 */ 3012 if (tp->pf == AF_INET6 && rt->rt_rmx.rmx_mtu < IPV6_MMTU) { 3013 /* 3014 * RFC2460 section 5, last paragraph: if path MTU is 3015 * smaller than 1280, use 1280 as packet size and 3016 * attach fragment header. 3017 */ 3018 mss = IPV6_MMTU - iphlen - sizeof(struct ip6_frag) - 3019 sizeof(struct tcphdr); 3020 } else 3021 mss = rt->rt_rmx.rmx_mtu - iphlen - sizeof(struct tcphdr); 3022 } else 3023 #endif /* RTV_MTU */ 3024 if (!ifp) 3025 /* 3026 * ifp may be null and rmx_mtu may be zero in certain 3027 * v6 cases (e.g., if ND wasn't able to resolve the 3028 * destination host. 3029 */ 3030 goto out; 3031 else if (ifp->if_flags & IFF_LOOPBACK) 3032 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 3033 else if (tp->pf == AF_INET) { 3034 if (ip_mtudisc) 3035 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 3036 else if (inp && in_localaddr(inp->inp_faddr)) 3037 mss = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 3038 } 3039 #ifdef INET6 3040 else if (tp->pf == AF_INET6) { 3041 /* 3042 * for IPv6, path MTU discovery is always turned on, 3043 * or the node must use packet size <= 1280. 3044 */ 3045 mss = IN6_LINKMTU(ifp) - iphlen - sizeof(struct tcphdr); 3046 } 3047 #endif /* INET6 */ 3048 3049 /* Calculate the value that we offer in TCPOPT_MAXSEG */ 3050 if (offer != -1) { 3051 #ifndef INET6 3052 mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 3053 #else 3054 if (tp->pf == AF_INET6) 3055 mssopt = IN6_LINKMTU(ifp) - iphlen - 3056 sizeof(struct tcphdr); 3057 else 3058 mssopt = ifp->if_mtu - iphlen - sizeof(struct tcphdr); 3059 #endif 3060 3061 mssopt = max(tcp_mssdflt, mssopt); 3062 } 3063 3064 out: 3065 /* 3066 * The current mss, t_maxseg, is initialized to the default value. 3067 * If we compute a smaller value, reduce the current mss. 3068 * If we compute a larger value, return it for use in sending 3069 * a max seg size option, but don't store it for use 3070 * unless we received an offer at least that large from peer. 3071 * 3072 * However, do not accept offers lower than the minimum of 3073 * the interface MTU and 216. 3074 */ 3075 if (offer > 0) 3076 tp->t_peermss = offer; 3077 if (tp->t_peermss) 3078 mss = min(mss, max(tp->t_peermss, 216)); 3079 3080 /* sanity - at least max opt. space */ 3081 mss = max(mss, 64); 3082 3083 /* 3084 * maxopd stores the maximum length of data AND options 3085 * in a segment; maxseg is the amount of data in a normal 3086 * segment. We need to store this value (maxopd) apart 3087 * from maxseg, because now every segment carries options 3088 * and thus we normally have somewhat less data in segments. 3089 */ 3090 tp->t_maxopd = mss; 3091 3092 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3093 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP) 3094 mss -= TCPOLEN_TSTAMP_APPA; 3095 #ifdef TCP_SIGNATURE 3096 if (tp->t_flags & TF_SIGNATURE) 3097 mss -= TCPOLEN_SIGLEN; 3098 #endif 3099 3100 if (offer == -1) { 3101 /* mss changed due to Path MTU discovery */ 3102 if (mss < tp->t_maxseg) { 3103 /* 3104 * Follow suggestion in RFC 2414 to reduce the 3105 * congestion window by the ratio of the old 3106 * segment size to the new segment size. 3107 */ 3108 tp->snd_cwnd = ulmax((tp->snd_cwnd / tp->t_maxseg) * 3109 mss, mss); 3110 } 3111 } else if (tcp_do_rfc3390) { 3112 /* increase initial window */ 3113 tp->snd_cwnd = ulmin(4 * mss, ulmax(2 * mss, 4380)); 3114 } else 3115 tp->snd_cwnd = mss; 3116 3117 tp->t_maxseg = mss; 3118 3119 return (offer != -1 ? mssopt : mss); 3120 } 3121 3122 /* 3123 * Set connection variables based on the effective MSS. 3124 * We are passed the TCPCB for the actual connection. If we 3125 * are the server, we are called by the compressed state engine 3126 * when the 3-way handshake is complete. If we are the client, 3127 * we are called when we receive the SYN,ACK from the server. 3128 * 3129 * NOTE: The t_maxseg value must be initialized in the TCPCB 3130 * before this routine is called! 3131 */ 3132 void 3133 tcp_mss_update(tp) 3134 struct tcpcb *tp; 3135 { 3136 int mss; 3137 u_long bufsize; 3138 struct rtentry *rt; 3139 struct socket *so; 3140 3141 so = tp->t_inpcb->inp_socket; 3142 mss = tp->t_maxseg; 3143 3144 rt = in_pcbrtentry(tp->t_inpcb); 3145 3146 if (rt == NULL) 3147 return; 3148 3149 bufsize = so->so_snd.sb_hiwat; 3150 if (bufsize < mss) { 3151 mss = bufsize; 3152 /* Update t_maxseg and t_maxopd */ 3153 tcp_mss(tp, mss); 3154 } else { 3155 bufsize = roundup(bufsize, mss); 3156 if (bufsize > sb_max) 3157 bufsize = sb_max; 3158 (void)sbreserve(&so->so_snd, bufsize); 3159 } 3160 3161 bufsize = so->so_rcv.sb_hiwat; 3162 if (bufsize > mss) { 3163 bufsize = roundup(bufsize, mss); 3164 if (bufsize > sb_max) 3165 bufsize = sb_max; 3166 (void)sbreserve(&so->so_rcv, bufsize); 3167 } 3168 3169 } 3170 #endif /* TUBA_INCLUDE */ 3171 3172 #if defined (TCP_SACK) 3173 /* 3174 * Checks for partial ack. If partial ack arrives, force the retransmission 3175 * of the next unacknowledged segment, do not clear tp->t_dupacks, and return 3176 * 1. By setting snd_nxt to ti_ack, this forces retransmission timer to 3177 * be started again. If the ack advances at least to tp->snd_last, return 0. 3178 */ 3179 int 3180 tcp_newreno(tp, th) 3181 struct tcpcb *tp; 3182 struct tcphdr *th; 3183 { 3184 if (SEQ_LT(th->th_ack, tp->snd_last)) { 3185 /* 3186 * snd_una has not been updated and the socket send buffer 3187 * not yet drained of the acked data, so we have to leave 3188 * snd_una as it was to get the correct data offset in 3189 * tcp_output(). 3190 */ 3191 tcp_seq onxt = tp->snd_nxt; 3192 u_long ocwnd = tp->snd_cwnd; 3193 TCP_TIMER_DISARM(tp, TCPT_REXMT); 3194 tp->t_rtttime = 0; 3195 tp->snd_nxt = th->th_ack; 3196 /* 3197 * Set snd_cwnd to one segment beyond acknowledged offset 3198 * (tp->snd_una not yet updated when this function is called) 3199 */ 3200 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3201 (void) tcp_output(tp); 3202 tp->snd_cwnd = ocwnd; 3203 if (SEQ_GT(onxt, tp->snd_nxt)) 3204 tp->snd_nxt = onxt; 3205 /* 3206 * Partial window deflation. Relies on fact that tp->snd_una 3207 * not updated yet. 3208 */ 3209 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3210 return 1; 3211 } 3212 return 0; 3213 } 3214 #endif /* TCP_SACK */ 3215 3216 static int 3217 tcp_mss_adv(struct ifnet *ifp, int af) 3218 { 3219 int mss = 0; 3220 int iphlen; 3221 3222 switch (af) { 3223 case AF_INET: 3224 if (ifp != NULL) 3225 mss = ifp->if_mtu; 3226 iphlen = sizeof(struct ip); 3227 break; 3228 #ifdef INET6 3229 case AF_INET6: 3230 if (ifp != NULL) 3231 mss = IN6_LINKMTU(ifp); 3232 iphlen = sizeof(struct ip6_hdr); 3233 break; 3234 #endif 3235 } 3236 mss = mss - iphlen - sizeof(struct tcphdr); 3237 return (max(mss, tcp_mssdflt)); 3238 } 3239 3240 /* 3241 * TCP compressed state engine. Currently used to hold compressed 3242 * state for SYN_RECEIVED. 3243 */ 3244 3245 u_long syn_cache_count; 3246 u_int32_t syn_hash1, syn_hash2; 3247 3248 #define SYN_HASH(sa, sp, dp) \ 3249 ((((sa)->s_addr^syn_hash1)*(((((u_int32_t)(dp))<<16) + \ 3250 ((u_int32_t)(sp)))^syn_hash2))) 3251 #ifndef INET6 3252 #define SYN_HASHALL(hash, src, dst) \ 3253 do { \ 3254 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3255 ((struct sockaddr_in *)(src))->sin_port, \ 3256 ((struct sockaddr_in *)(dst))->sin_port); \ 3257 } while (/*CONSTCOND*/ 0) 3258 #else 3259 #define SYN_HASH6(sa, sp, dp) \ 3260 ((((sa)->s6_addr32[0] ^ (sa)->s6_addr32[3] ^ syn_hash1) * \ 3261 (((((u_int32_t)(dp))<<16) + ((u_int32_t)(sp)))^syn_hash2)) \ 3262 & 0x7fffffff) 3263 3264 #define SYN_HASHALL(hash, src, dst) \ 3265 do { \ 3266 switch ((src)->sa_family) { \ 3267 case AF_INET: \ 3268 hash = SYN_HASH(&((struct sockaddr_in *)(src))->sin_addr, \ 3269 ((struct sockaddr_in *)(src))->sin_port, \ 3270 ((struct sockaddr_in *)(dst))->sin_port); \ 3271 break; \ 3272 case AF_INET6: \ 3273 hash = SYN_HASH6(&((struct sockaddr_in6 *)(src))->sin6_addr, \ 3274 ((struct sockaddr_in6 *)(src))->sin6_port, \ 3275 ((struct sockaddr_in6 *)(dst))->sin6_port); \ 3276 break; \ 3277 default: \ 3278 hash = 0; \ 3279 } \ 3280 } while (/*CONSTCOND*/0) 3281 #endif /* INET6 */ 3282 3283 #define SYN_CACHE_RM(sc) \ 3284 do { \ 3285 TAILQ_REMOVE(&tcp_syn_cache[(sc)->sc_bucketidx].sch_bucket, \ 3286 (sc), sc_bucketq); \ 3287 (sc)->sc_tp = NULL; \ 3288 LIST_REMOVE((sc), sc_tpq); \ 3289 tcp_syn_cache[(sc)->sc_bucketidx].sch_length--; \ 3290 timeout_del(&(sc)->sc_timer); \ 3291 syn_cache_count--; \ 3292 } while (/*CONSTCOND*/0) 3293 3294 #define SYN_CACHE_PUT(sc) \ 3295 do { \ 3296 if ((sc)->sc_ipopts) \ 3297 (void) m_free((sc)->sc_ipopts); \ 3298 if ((sc)->sc_route4.ro_rt != NULL) \ 3299 RTFREE((sc)->sc_route4.ro_rt); \ 3300 pool_put(&syn_cache_pool, (sc)); \ 3301 } while (/*CONSTCOND*/0) 3302 3303 struct pool syn_cache_pool; 3304 3305 /* 3306 * We don't estimate RTT with SYNs, so each packet starts with the default 3307 * RTT and each timer step has a fixed timeout value. 3308 */ 3309 #define SYN_CACHE_TIMER_ARM(sc) \ 3310 do { \ 3311 TCPT_RANGESET((sc)->sc_rxtcur, \ 3312 TCPTV_SRTTDFLT * tcp_backoff[(sc)->sc_rxtshift], TCPTV_MIN, \ 3313 TCPTV_REXMTMAX); \ 3314 if (!timeout_initialized(&(sc)->sc_timer)) \ 3315 timeout_set(&(sc)->sc_timer, syn_cache_timer, (sc)); \ 3316 timeout_add(&(sc)->sc_timer, (sc)->sc_rxtcur * (hz / PR_SLOWHZ)); \ 3317 } while (/*CONSTCOND*/0) 3318 3319 #define SYN_CACHE_TIMESTAMP(sc) tcp_now 3320 3321 void 3322 syn_cache_init() 3323 { 3324 int i; 3325 3326 /* Initialize the hash buckets. */ 3327 for (i = 0; i < tcp_syn_cache_size; i++) 3328 TAILQ_INIT(&tcp_syn_cache[i].sch_bucket); 3329 3330 /* Initialize the syn cache pool. */ 3331 pool_init(&syn_cache_pool, sizeof(struct syn_cache), 0, 0, 0, 3332 "synpl", NULL); 3333 } 3334 3335 void 3336 syn_cache_insert(sc, tp) 3337 struct syn_cache *sc; 3338 struct tcpcb *tp; 3339 { 3340 struct syn_cache_head *scp; 3341 struct syn_cache *sc2; 3342 int s; 3343 3344 /* 3345 * If there are no entries in the hash table, reinitialize 3346 * the hash secrets. 3347 */ 3348 if (syn_cache_count == 0) { 3349 syn_hash1 = arc4random(); 3350 syn_hash2 = arc4random(); 3351 } 3352 3353 SYN_HASHALL(sc->sc_hash, &sc->sc_src.sa, &sc->sc_dst.sa); 3354 sc->sc_bucketidx = sc->sc_hash % tcp_syn_cache_size; 3355 scp = &tcp_syn_cache[sc->sc_bucketidx]; 3356 3357 /* 3358 * Make sure that we don't overflow the per-bucket 3359 * limit or the total cache size limit. 3360 */ 3361 s = splsoftnet(); 3362 if (scp->sch_length >= tcp_syn_bucket_limit) { 3363 tcpstat.tcps_sc_bucketoverflow++; 3364 /* 3365 * The bucket is full. Toss the oldest element in the 3366 * bucket. This will be the first entry in the bucket. 3367 */ 3368 sc2 = TAILQ_FIRST(&scp->sch_bucket); 3369 #ifdef DIAGNOSTIC 3370 /* 3371 * This should never happen; we should always find an 3372 * entry in our bucket. 3373 */ 3374 if (sc2 == NULL) 3375 panic("syn_cache_insert: bucketoverflow: impossible"); 3376 #endif 3377 SYN_CACHE_RM(sc2); 3378 SYN_CACHE_PUT(sc2); 3379 } else if (syn_cache_count >= tcp_syn_cache_limit) { 3380 struct syn_cache_head *scp2, *sce; 3381 3382 tcpstat.tcps_sc_overflowed++; 3383 /* 3384 * The cache is full. Toss the oldest entry in the 3385 * first non-empty bucket we can find. 3386 * 3387 * XXX We would really like to toss the oldest 3388 * entry in the cache, but we hope that this 3389 * condition doesn't happen very often. 3390 */ 3391 scp2 = scp; 3392 if (TAILQ_EMPTY(&scp2->sch_bucket)) { 3393 sce = &tcp_syn_cache[tcp_syn_cache_size]; 3394 for (++scp2; scp2 != scp; scp2++) { 3395 if (scp2 >= sce) 3396 scp2 = &tcp_syn_cache[0]; 3397 if (! TAILQ_EMPTY(&scp2->sch_bucket)) 3398 break; 3399 } 3400 #ifdef DIAGNOSTIC 3401 /* 3402 * This should never happen; we should always find a 3403 * non-empty bucket. 3404 */ 3405 if (scp2 == scp) 3406 panic("syn_cache_insert: cacheoverflow: " 3407 "impossible"); 3408 #endif 3409 } 3410 sc2 = TAILQ_FIRST(&scp2->sch_bucket); 3411 SYN_CACHE_RM(sc2); 3412 SYN_CACHE_PUT(sc2); 3413 } 3414 3415 /* 3416 * Initialize the entry's timer. 3417 */ 3418 sc->sc_rxttot = 0; 3419 sc->sc_rxtshift = 0; 3420 SYN_CACHE_TIMER_ARM(sc); 3421 3422 /* Link it from tcpcb entry */ 3423 LIST_INSERT_HEAD(&tp->t_sc, sc, sc_tpq); 3424 3425 /* Put it into the bucket. */ 3426 TAILQ_INSERT_TAIL(&scp->sch_bucket, sc, sc_bucketq); 3427 scp->sch_length++; 3428 syn_cache_count++; 3429 3430 tcpstat.tcps_sc_added++; 3431 splx(s); 3432 } 3433 3434 /* 3435 * Walk the timer queues, looking for SYN,ACKs that need to be retransmitted. 3436 * If we have retransmitted an entry the maximum number of times, expire 3437 * that entry. 3438 */ 3439 void 3440 syn_cache_timer(void *arg) 3441 { 3442 struct syn_cache *sc = arg; 3443 int s; 3444 3445 s = splsoftnet(); 3446 3447 if (__predict_false(sc->sc_rxtshift == TCP_MAXRXTSHIFT)) { 3448 /* Drop it -- too many retransmissions. */ 3449 goto dropit; 3450 } 3451 3452 /* 3453 * Compute the total amount of time this entry has 3454 * been on a queue. If this entry has been on longer 3455 * than the keep alive timer would allow, expire it. 3456 */ 3457 sc->sc_rxttot += sc->sc_rxtcur; 3458 if (sc->sc_rxttot >= tcptv_keep_init) 3459 goto dropit; 3460 3461 tcpstat.tcps_sc_retransmitted++; 3462 (void) syn_cache_respond(sc, NULL); 3463 3464 /* Advance the timer back-off. */ 3465 sc->sc_rxtshift++; 3466 SYN_CACHE_TIMER_ARM(sc); 3467 3468 splx(s); 3469 return; 3470 3471 dropit: 3472 tcpstat.tcps_sc_timed_out++; 3473 SYN_CACHE_RM(sc); 3474 SYN_CACHE_PUT(sc); 3475 splx(s); 3476 } 3477 3478 /* 3479 * Remove syn cache created by the specified tcb entry, 3480 * because this does not make sense to keep them 3481 * (if there's no tcb entry, syn cache entry will never be used) 3482 */ 3483 void 3484 syn_cache_cleanup(tp) 3485 struct tcpcb *tp; 3486 { 3487 struct syn_cache *sc, *nsc; 3488 int s; 3489 3490 s = splsoftnet(); 3491 3492 for (sc = LIST_FIRST(&tp->t_sc); sc != NULL; sc = nsc) { 3493 nsc = LIST_NEXT(sc, sc_tpq); 3494 3495 #ifdef DIAGNOSTIC 3496 if (sc->sc_tp != tp) 3497 panic("invalid sc_tp in syn_cache_cleanup"); 3498 #endif 3499 SYN_CACHE_RM(sc); 3500 SYN_CACHE_PUT(sc); 3501 } 3502 /* just for safety */ 3503 LIST_INIT(&tp->t_sc); 3504 3505 splx(s); 3506 } 3507 3508 /* 3509 * Find an entry in the syn cache. 3510 */ 3511 struct syn_cache * 3512 syn_cache_lookup(src, dst, headp) 3513 struct sockaddr *src; 3514 struct sockaddr *dst; 3515 struct syn_cache_head **headp; 3516 { 3517 struct syn_cache *sc; 3518 struct syn_cache_head *scp; 3519 u_int32_t hash; 3520 int s; 3521 3522 SYN_HASHALL(hash, src, dst); 3523 3524 scp = &tcp_syn_cache[hash % tcp_syn_cache_size]; 3525 *headp = scp; 3526 s = splsoftnet(); 3527 for (sc = TAILQ_FIRST(&scp->sch_bucket); sc != NULL; 3528 sc = TAILQ_NEXT(sc, sc_bucketq)) { 3529 if (sc->sc_hash != hash) 3530 continue; 3531 if (!bcmp(&sc->sc_src, src, src->sa_len) && 3532 !bcmp(&sc->sc_dst, dst, dst->sa_len)) { 3533 splx(s); 3534 return (sc); 3535 } 3536 } 3537 splx(s); 3538 return (NULL); 3539 } 3540 3541 /* 3542 * This function gets called when we receive an ACK for a 3543 * socket in the LISTEN state. We look up the connection 3544 * in the syn cache, and if its there, we pull it out of 3545 * the cache and turn it into a full-blown connection in 3546 * the SYN-RECEIVED state. 3547 * 3548 * The return values may not be immediately obvious, and their effects 3549 * can be subtle, so here they are: 3550 * 3551 * NULL SYN was not found in cache; caller should drop the 3552 * packet and send an RST. 3553 * 3554 * -1 We were unable to create the new connection, and are 3555 * aborting it. An ACK,RST is being sent to the peer 3556 * (unless we got screwey sequence numbners; see below), 3557 * because the 3-way handshake has been completed. Caller 3558 * should not free the mbuf, since we may be using it. If 3559 * we are not, we will free it. 3560 * 3561 * Otherwise, the return value is a pointer to the new socket 3562 * associated with the connection. 3563 */ 3564 struct socket * 3565 syn_cache_get(src, dst, th, hlen, tlen, so, m) 3566 struct sockaddr *src; 3567 struct sockaddr *dst; 3568 struct tcphdr *th; 3569 unsigned int hlen, tlen; 3570 struct socket *so; 3571 struct mbuf *m; 3572 { 3573 struct syn_cache *sc; 3574 struct syn_cache_head *scp; 3575 struct inpcb *inp = NULL; 3576 struct tcpcb *tp = 0; 3577 struct mbuf *am; 3578 int s; 3579 struct socket *oso; 3580 3581 s = splsoftnet(); 3582 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3583 splx(s); 3584 return (NULL); 3585 } 3586 3587 /* 3588 * Verify the sequence and ack numbers. Try getting the correct 3589 * response again. 3590 */ 3591 if ((th->th_ack != sc->sc_iss + 1) || 3592 SEQ_LEQ(th->th_seq, sc->sc_irs) || 3593 SEQ_GT(th->th_seq, sc->sc_irs + 1 + sc->sc_win)) { 3594 (void) syn_cache_respond(sc, m); 3595 splx(s); 3596 return ((struct socket *)(-1)); 3597 } 3598 3599 /* Remove this cache entry */ 3600 SYN_CACHE_RM(sc); 3601 splx(s); 3602 3603 /* 3604 * Ok, create the full blown connection, and set things up 3605 * as they would have been set up if we had created the 3606 * connection when the SYN arrived. If we can't create 3607 * the connection, abort it. 3608 */ 3609 oso = so; 3610 so = sonewconn(so, SS_ISCONNECTED); 3611 if (so == NULL) 3612 goto resetandabort; 3613 3614 inp = sotoinpcb(oso); 3615 #ifdef IPSEC 3616 /* 3617 * We need to copy the required security levels 3618 * from the old pcb. Ditto for any other 3619 * IPsec-related information. 3620 */ 3621 { 3622 struct inpcb *newinp = (struct inpcb *)so->so_pcb; 3623 bcopy(inp->inp_seclevel, newinp->inp_seclevel, 3624 sizeof(inp->inp_seclevel)); 3625 newinp->inp_secrequire = inp->inp_secrequire; 3626 if (inp->inp_ipo != NULL) { 3627 newinp->inp_ipo = inp->inp_ipo; 3628 inp->inp_ipo->ipo_ref_count++; 3629 } 3630 if (inp->inp_ipsec_remotecred != NULL) { 3631 newinp->inp_ipsec_remotecred = inp->inp_ipsec_remotecred; 3632 inp->inp_ipsec_remotecred->ref_count++; 3633 } 3634 if (inp->inp_ipsec_remoteauth != NULL) { 3635 newinp->inp_ipsec_remoteauth 3636 = inp->inp_ipsec_remoteauth; 3637 inp->inp_ipsec_remoteauth->ref_count++; 3638 } 3639 } 3640 #endif /* IPSEC */ 3641 #ifdef INET6 3642 /* 3643 * inp still has the OLD in_pcb stuff, set the 3644 * v6-related flags on the new guy, too. 3645 */ 3646 { 3647 int flags = inp->inp_flags; 3648 struct inpcb *oldinpcb = inp; 3649 3650 inp = (struct inpcb *)so->so_pcb; 3651 inp->inp_flags |= (flags & INP_IPV6); 3652 if ((inp->inp_flags & INP_IPV6) != 0) { 3653 inp->inp_ipv6.ip6_hlim = 3654 oldinpcb->inp_ipv6.ip6_hlim; 3655 } 3656 } 3657 #else /* INET6 */ 3658 inp = (struct inpcb *)so->so_pcb; 3659 #endif /* INET6 */ 3660 3661 inp->inp_lport = th->th_dport; 3662 switch (src->sa_family) { 3663 #ifdef INET6 3664 case AF_INET6: 3665 inp->inp_laddr6 = ((struct sockaddr_in6 *)dst)->sin6_addr; 3666 break; 3667 #endif /* INET6 */ 3668 case AF_INET: 3669 3670 inp->inp_laddr = ((struct sockaddr_in *)dst)->sin_addr; 3671 inp->inp_options = ip_srcroute(); 3672 if (inp->inp_options == NULL) { 3673 inp->inp_options = sc->sc_ipopts; 3674 sc->sc_ipopts = NULL; 3675 } 3676 break; 3677 } 3678 in_pcbrehash(inp); 3679 3680 /* 3681 * Give the new socket our cached route reference. 3682 */ 3683 if (inp) 3684 inp->inp_route = sc->sc_route4; /* struct assignment */ 3685 #ifdef INET6 3686 else 3687 inp->inp_route6 = sc->sc_route6; 3688 #endif 3689 sc->sc_route4.ro_rt = NULL; 3690 3691 am = m_get(M_DONTWAIT, MT_SONAME); /* XXX */ 3692 if (am == NULL) 3693 goto resetandabort; 3694 am->m_len = src->sa_len; 3695 bcopy(src, mtod(am, caddr_t), src->sa_len); 3696 3697 switch (src->sa_family) { 3698 case AF_INET: 3699 /* drop IPv4 packet to AF_INET6 socket */ 3700 if (inp->inp_flags & INP_IPV6) { 3701 (void) m_free(am); 3702 goto resetandabort; 3703 } 3704 if (in_pcbconnect(inp, am)) { 3705 (void) m_free(am); 3706 goto resetandabort; 3707 } 3708 break; 3709 #ifdef INET6 3710 case AF_INET6: 3711 if (in6_pcbconnect(inp, am)) { 3712 (void) m_free(am); 3713 goto resetandabort; 3714 } 3715 break; 3716 #endif 3717 } 3718 (void) m_free(am); 3719 3720 tp = intotcpcb(inp); 3721 tp->t_flags = sototcpcb(oso)->t_flags & TF_NODELAY; 3722 if (sc->sc_request_r_scale != 15) { 3723 tp->requested_s_scale = sc->sc_requested_s_scale; 3724 tp->request_r_scale = sc->sc_request_r_scale; 3725 tp->snd_scale = sc->sc_requested_s_scale; 3726 tp->rcv_scale = sc->sc_request_r_scale; 3727 tp->t_flags |= TF_REQ_SCALE|TF_RCVD_SCALE; 3728 } 3729 if (sc->sc_flags & SCF_TIMESTAMP) 3730 tp->t_flags |= TF_REQ_TSTMP|TF_RCVD_TSTMP; 3731 3732 tp->t_template = tcp_template(tp); 3733 if (tp->t_template == 0) { 3734 tp = tcp_drop(tp, ENOBUFS); /* destroys socket */ 3735 so = NULL; 3736 m_freem(m); 3737 goto abort; 3738 } 3739 #ifdef TCP_SACK 3740 tp->sack_enable = sc->sc_flags & SCF_SACK_PERMIT; 3741 #endif 3742 3743 tp->iss = sc->sc_iss; 3744 tp->irs = sc->sc_irs; 3745 tcp_sendseqinit(tp); 3746 #if defined (TCP_SACK) || defined(TCP_ECN) 3747 tp->snd_last = tp->snd_una; 3748 #endif /* TCP_SACK */ 3749 #if defined(TCP_SACK) && defined(TCP_FACK) 3750 tp->snd_fack = tp->snd_una; 3751 tp->retran_data = 0; 3752 tp->snd_awnd = 0; 3753 #endif /* TCP_FACK */ 3754 #ifdef TCP_ECN 3755 if (sc->sc_flags & SCF_ECN_PERMIT) { 3756 tp->t_flags |= TF_ECN_PERMIT; 3757 tcpstat.tcps_ecn_accepts++; 3758 } 3759 #endif 3760 #ifdef TCP_SACK 3761 if (sc->sc_flags & SCF_SACK_PERMIT) 3762 tp->t_flags |= TF_SACK_PERMIT; 3763 #endif 3764 #ifdef TCP_SIGNATURE 3765 if (sc->sc_flags & SCF_SIGNATURE) 3766 tp->t_flags |= TF_SIGNATURE; 3767 #endif 3768 tcp_rcvseqinit(tp); 3769 tp->t_state = TCPS_SYN_RECEIVED; 3770 tp->t_rcvtime = tcp_now; 3771 TCP_TIMER_ARM(tp, TCPT_KEEP, tcptv_keep_init); 3772 tcpstat.tcps_accepts++; 3773 3774 tcp_mss(tp, sc->sc_peermaxseg); /* sets t_maxseg */ 3775 if (sc->sc_peermaxseg) 3776 tcp_mss_update(tp); 3777 /* Reset initial window to 1 segment for retransmit */ 3778 if (sc->sc_rxtshift > 0) 3779 tp->snd_cwnd = tp->t_maxseg; 3780 tp->snd_wl1 = sc->sc_irs; 3781 tp->rcv_up = sc->sc_irs + 1; 3782 3783 /* 3784 * This is what whould have happened in tcp_output() when 3785 * the SYN,ACK was sent. 3786 */ 3787 tp->snd_up = tp->snd_una; 3788 tp->snd_max = tp->snd_nxt = tp->iss+1; 3789 TCP_TIMER_ARM(tp, TCPT_REXMT, tp->t_rxtcur); 3790 if (sc->sc_win > 0 && SEQ_GT(tp->rcv_nxt + sc->sc_win, tp->rcv_adv)) 3791 tp->rcv_adv = tp->rcv_nxt + sc->sc_win; 3792 tp->last_ack_sent = tp->rcv_nxt; 3793 3794 tcpstat.tcps_sc_completed++; 3795 SYN_CACHE_PUT(sc); 3796 return (so); 3797 3798 resetandabort: 3799 tcp_respond(NULL, mtod(m, caddr_t), m, (tcp_seq)0, th->th_ack, TH_RST); 3800 abort: 3801 if (so != NULL) 3802 (void) soabort(so); 3803 SYN_CACHE_PUT(sc); 3804 tcpstat.tcps_sc_aborted++; 3805 return ((struct socket *)(-1)); 3806 } 3807 3808 /* 3809 * This function is called when we get a RST for a 3810 * non-existent connection, so that we can see if the 3811 * connection is in the syn cache. If it is, zap it. 3812 */ 3813 3814 void 3815 syn_cache_reset(src, dst, th) 3816 struct sockaddr *src; 3817 struct sockaddr *dst; 3818 struct tcphdr *th; 3819 { 3820 struct syn_cache *sc; 3821 struct syn_cache_head *scp; 3822 int s = splsoftnet(); 3823 3824 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3825 splx(s); 3826 return; 3827 } 3828 if (SEQ_LT(th->th_seq, sc->sc_irs) || 3829 SEQ_GT(th->th_seq, sc->sc_irs+1)) { 3830 splx(s); 3831 return; 3832 } 3833 SYN_CACHE_RM(sc); 3834 splx(s); 3835 tcpstat.tcps_sc_reset++; 3836 SYN_CACHE_PUT(sc); 3837 } 3838 3839 void 3840 syn_cache_unreach(src, dst, th) 3841 struct sockaddr *src; 3842 struct sockaddr *dst; 3843 struct tcphdr *th; 3844 { 3845 struct syn_cache *sc; 3846 struct syn_cache_head *scp; 3847 int s; 3848 3849 s = splsoftnet(); 3850 if ((sc = syn_cache_lookup(src, dst, &scp)) == NULL) { 3851 splx(s); 3852 return; 3853 } 3854 /* If the sequence number != sc_iss, then it's a bogus ICMP msg */ 3855 if (ntohl (th->th_seq) != sc->sc_iss) { 3856 splx(s); 3857 return; 3858 } 3859 3860 /* 3861 * If we've retransmitted 3 times and this is our second error, 3862 * we remove the entry. Otherwise, we allow it to continue on. 3863 * This prevents us from incorrectly nuking an entry during a 3864 * spurious network outage. 3865 * 3866 * See tcp_notify(). 3867 */ 3868 if ((sc->sc_flags & SCF_UNREACH) == 0 || sc->sc_rxtshift < 3) { 3869 sc->sc_flags |= SCF_UNREACH; 3870 splx(s); 3871 return; 3872 } 3873 3874 SYN_CACHE_RM(sc); 3875 splx(s); 3876 tcpstat.tcps_sc_unreach++; 3877 SYN_CACHE_PUT(sc); 3878 } 3879 3880 /* 3881 * Given a LISTEN socket and an inbound SYN request, add 3882 * this to the syn cache, and send back a segment: 3883 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK> 3884 * to the source. 3885 * 3886 * IMPORTANT NOTE: We do _NOT_ ACK data that might accompany the SYN. 3887 * Doing so would require that we hold onto the data and deliver it 3888 * to the application. However, if we are the target of a SYN-flood 3889 * DoS attack, an attacker could send data which would eventually 3890 * consume all available buffer space if it were ACKed. By not ACKing 3891 * the data, we avoid this DoS scenario. 3892 */ 3893 3894 int 3895 syn_cache_add(src, dst, th, iphlen, so, m, optp, optlen, oi) 3896 struct sockaddr *src; 3897 struct sockaddr *dst; 3898 struct tcphdr *th; 3899 unsigned int iphlen; 3900 struct socket *so; 3901 struct mbuf *m; 3902 u_char *optp; 3903 int optlen; 3904 struct tcp_opt_info *oi; 3905 { 3906 struct tcpcb tb, *tp; 3907 long win; 3908 struct syn_cache *sc; 3909 struct syn_cache_head *scp; 3910 struct mbuf *ipopts; 3911 3912 tp = sototcpcb(so); 3913 3914 /* 3915 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 3916 * 3917 * Note this check is performed in tcp_input() very early on. 3918 */ 3919 3920 /* 3921 * Initialize some local state. 3922 */ 3923 win = sbspace(&so->so_rcv); 3924 if (win > TCP_MAXWIN) 3925 win = TCP_MAXWIN; 3926 3927 #ifdef TCP_SIGNATURE 3928 if (optp || (tp->t_flags & TF_SIGNATURE)) { 3929 #else 3930 if (optp) { 3931 #endif 3932 tb.pf = tp->pf; 3933 #ifdef TCP_SACK 3934 tb.sack_enable = tcp_do_sack; 3935 #endif 3936 tb.t_flags = tcp_do_rfc1323 ? (TF_REQ_SCALE|TF_REQ_TSTMP) : 0; 3937 #ifdef TCP_SIGNATURE 3938 tb.t_state = TCPS_LISTEN; 3939 if (tp->t_flags & TF_SIGNATURE) 3940 tb.t_flags |= TF_SIGNATURE; 3941 #endif 3942 if (tcp_dooptions(&tb, optp, optlen, th, m, iphlen, oi)) 3943 return (0); 3944 } else 3945 tb.t_flags = 0; 3946 3947 switch (src->sa_family) { 3948 #ifdef INET 3949 case AF_INET: 3950 /* 3951 * Remember the IP options, if any. 3952 */ 3953 ipopts = ip_srcroute(); 3954 break; 3955 #endif 3956 default: 3957 ipopts = NULL; 3958 } 3959 3960 /* 3961 * See if we already have an entry for this connection. 3962 * If we do, resend the SYN,ACK. We do not count this 3963 * as a retransmission (XXX though maybe we should). 3964 */ 3965 if ((sc = syn_cache_lookup(src, dst, &scp)) != NULL) { 3966 tcpstat.tcps_sc_dupesyn++; 3967 if (ipopts) { 3968 /* 3969 * If we were remembering a previous source route, 3970 * forget it and use the new one we've been given. 3971 */ 3972 if (sc->sc_ipopts) 3973 (void) m_free(sc->sc_ipopts); 3974 sc->sc_ipopts = ipopts; 3975 } 3976 sc->sc_timestamp = tb.ts_recent; 3977 if (syn_cache_respond(sc, m) == 0) { 3978 tcpstat.tcps_sndacks++; 3979 tcpstat.tcps_sndtotal++; 3980 } 3981 return (1); 3982 } 3983 3984 sc = pool_get(&syn_cache_pool, PR_NOWAIT); 3985 if (sc == NULL) { 3986 if (ipopts) 3987 (void) m_free(ipopts); 3988 return (0); 3989 } 3990 3991 /* 3992 * Fill in the cache, and put the necessary IP and TCP 3993 * options into the reply. 3994 */ 3995 bzero(sc, sizeof(struct syn_cache)); 3996 bzero(&sc->sc_timer, sizeof(sc->sc_timer)); 3997 bcopy(src, &sc->sc_src, src->sa_len); 3998 bcopy(dst, &sc->sc_dst, dst->sa_len); 3999 sc->sc_flags = 0; 4000 sc->sc_ipopts = ipopts; 4001 sc->sc_irs = th->th_seq; 4002 4003 #ifdef TCP_COMPAT_42 4004 tcp_iss += TCP_ISSINCR/2; 4005 sc->sc_iss = tcp_iss; 4006 #else 4007 sc->sc_iss = tcp_rndiss_next(); 4008 #endif 4009 sc->sc_peermaxseg = oi->maxseg; 4010 sc->sc_ourmaxseg = tcp_mss_adv(m->m_flags & M_PKTHDR ? 4011 m->m_pkthdr.rcvif : NULL, sc->sc_src.sa.sa_family); 4012 sc->sc_win = win; 4013 sc->sc_timestamp = tb.ts_recent; 4014 if ((tb.t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP)) == 4015 (TF_REQ_TSTMP|TF_RCVD_TSTMP)) 4016 sc->sc_flags |= SCF_TIMESTAMP; 4017 if ((tb.t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 4018 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 4019 sc->sc_requested_s_scale = tb.requested_s_scale; 4020 sc->sc_request_r_scale = 0; 4021 while (sc->sc_request_r_scale < TCP_MAX_WINSHIFT && 4022 TCP_MAXWIN << sc->sc_request_r_scale < 4023 so->so_rcv.sb_hiwat) 4024 sc->sc_request_r_scale++; 4025 } else { 4026 sc->sc_requested_s_scale = 15; 4027 sc->sc_request_r_scale = 15; 4028 } 4029 #ifdef TCP_ECN 4030 /* 4031 * if both ECE and CWR flag bits are set, peer is ECN capable. 4032 */ 4033 if (tcp_do_ecn && 4034 (th->th_flags & (TH_ECE|TH_CWR)) == (TH_ECE|TH_CWR)) 4035 sc->sc_flags |= SCF_ECN_PERMIT; 4036 #endif 4037 #ifdef TCP_SACK 4038 /* 4039 * Set SCF_SACK_PERMIT if peer did send a SACK_PERMITTED option 4040 * (i.e., if tcp_dooptions() did set TF_SACK_PERMIT). 4041 */ 4042 if (tb.sack_enable && (tb.t_flags & TF_SACK_PERMIT)) 4043 sc->sc_flags |= SCF_SACK_PERMIT; 4044 #endif 4045 #ifdef TCP_SIGNATURE 4046 if (tb.t_flags & TF_SIGNATURE) 4047 sc->sc_flags |= SCF_SIGNATURE; 4048 #endif 4049 sc->sc_tp = tp; 4050 if (syn_cache_respond(sc, m) == 0) { 4051 syn_cache_insert(sc, tp); 4052 tcpstat.tcps_sndacks++; 4053 tcpstat.tcps_sndtotal++; 4054 } else { 4055 SYN_CACHE_PUT(sc); 4056 tcpstat.tcps_sc_dropped++; 4057 } 4058 return (1); 4059 } 4060 4061 int 4062 syn_cache_respond(sc, m) 4063 struct syn_cache *sc; 4064 struct mbuf *m; 4065 { 4066 struct route *ro; 4067 u_int8_t *optp; 4068 int optlen, error; 4069 u_int16_t tlen; 4070 struct ip *ip = NULL; 4071 #ifdef INET6 4072 struct ip6_hdr *ip6 = NULL; 4073 #endif 4074 struct tcphdr *th; 4075 u_int hlen; 4076 struct inpcb *inp; 4077 4078 switch (sc->sc_src.sa.sa_family) { 4079 case AF_INET: 4080 hlen = sizeof(struct ip); 4081 ro = &sc->sc_route4; 4082 break; 4083 #ifdef INET6 4084 case AF_INET6: 4085 hlen = sizeof(struct ip6_hdr); 4086 ro = (struct route *)&sc->sc_route6; 4087 break; 4088 #endif 4089 default: 4090 if (m) 4091 m_freem(m); 4092 return (EAFNOSUPPORT); 4093 } 4094 4095 /* Compute the size of the TCP options. */ 4096 optlen = 4 + (sc->sc_request_r_scale != 15 ? 4 : 0) + 4097 #ifdef TCP_SACK 4098 ((sc->sc_flags & SCF_SACK_PERMIT) ? 4 : 0) + 4099 #endif 4100 #ifdef TCP_SIGNATURE 4101 ((sc->sc_flags & SCF_SIGNATURE) ? TCPOLEN_SIGLEN : 0) + 4102 #endif 4103 ((sc->sc_flags & SCF_TIMESTAMP) ? TCPOLEN_TSTAMP_APPA : 0); 4104 4105 tlen = hlen + sizeof(struct tcphdr) + optlen; 4106 4107 /* 4108 * Create the IP+TCP header from scratch. 4109 */ 4110 if (m) 4111 m_freem(m); 4112 #ifdef DIAGNOSTIC 4113 if (max_linkhdr + tlen > MCLBYTES) 4114 return (ENOBUFS); 4115 #endif 4116 MGETHDR(m, M_DONTWAIT, MT_DATA); 4117 if (m && tlen > MHLEN) { 4118 MCLGET(m, M_DONTWAIT); 4119 if ((m->m_flags & M_EXT) == 0) { 4120 m_freem(m); 4121 m = NULL; 4122 } 4123 } 4124 if (m == NULL) 4125 return (ENOBUFS); 4126 4127 /* Fixup the mbuf. */ 4128 m->m_data += max_linkhdr; 4129 m->m_len = m->m_pkthdr.len = tlen; 4130 m->m_pkthdr.rcvif = NULL; 4131 memset(mtod(m, u_char *), 0, tlen); 4132 4133 switch (sc->sc_src.sa.sa_family) { 4134 case AF_INET: 4135 ip = mtod(m, struct ip *); 4136 ip->ip_dst = sc->sc_src.sin.sin_addr; 4137 ip->ip_src = sc->sc_dst.sin.sin_addr; 4138 ip->ip_p = IPPROTO_TCP; 4139 th = (struct tcphdr *)(ip + 1); 4140 th->th_dport = sc->sc_src.sin.sin_port; 4141 th->th_sport = sc->sc_dst.sin.sin_port; 4142 break; 4143 #ifdef INET6 4144 case AF_INET6: 4145 ip6 = mtod(m, struct ip6_hdr *); 4146 ip6->ip6_dst = sc->sc_src.sin6.sin6_addr; 4147 ip6->ip6_src = sc->sc_dst.sin6.sin6_addr; 4148 ip6->ip6_nxt = IPPROTO_TCP; 4149 /* ip6_plen will be updated in ip6_output() */ 4150 th = (struct tcphdr *)(ip6 + 1); 4151 th->th_dport = sc->sc_src.sin6.sin6_port; 4152 th->th_sport = sc->sc_dst.sin6.sin6_port; 4153 break; 4154 #endif 4155 default: 4156 th = NULL; 4157 } 4158 4159 th->th_seq = htonl(sc->sc_iss); 4160 th->th_ack = htonl(sc->sc_irs + 1); 4161 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2; 4162 th->th_flags = TH_SYN|TH_ACK; 4163 #ifdef TCP_ECN 4164 /* Set ECE for SYN-ACK if peer supports ECN. */ 4165 if (tcp_do_ecn && (sc->sc_flags & SCF_ECN_PERMIT)) 4166 th->th_flags |= TH_ECE; 4167 #endif 4168 th->th_win = htons(sc->sc_win); 4169 /* th_sum already 0 */ 4170 /* th_urp already 0 */ 4171 4172 /* Tack on the TCP options. */ 4173 optp = (u_int8_t *)(th + 1); 4174 *optp++ = TCPOPT_MAXSEG; 4175 *optp++ = 4; 4176 *optp++ = (sc->sc_ourmaxseg >> 8) & 0xff; 4177 *optp++ = sc->sc_ourmaxseg & 0xff; 4178 4179 #ifdef TCP_SACK 4180 /* Include SACK_PERMIT_HDR option if peer has already done so. */ 4181 if (sc->sc_flags & SCF_SACK_PERMIT) { 4182 *((u_int32_t *)optp) = htonl(TCPOPT_SACK_PERMIT_HDR); 4183 optp += 4; 4184 } 4185 #endif 4186 4187 if (sc->sc_request_r_scale != 15) { 4188 *((u_int32_t *)optp) = htonl(TCPOPT_NOP << 24 | 4189 TCPOPT_WINDOW << 16 | TCPOLEN_WINDOW << 8 | 4190 sc->sc_request_r_scale); 4191 optp += 4; 4192 } 4193 4194 if (sc->sc_flags & SCF_TIMESTAMP) { 4195 u_int32_t *lp = (u_int32_t *)(optp); 4196 /* Form timestamp option as shown in appendix A of RFC 1323. */ 4197 *lp++ = htonl(TCPOPT_TSTAMP_HDR); 4198 *lp++ = htonl(SYN_CACHE_TIMESTAMP(sc)); 4199 *lp = htonl(sc->sc_timestamp); 4200 optp += TCPOLEN_TSTAMP_APPA; 4201 } 4202 4203 #ifdef TCP_SIGNATURE 4204 if (sc->sc_flags & SCF_SIGNATURE) { 4205 MD5_CTX ctx; 4206 union sockaddr_union src, dst; 4207 struct tdb *tdb; 4208 4209 bzero(&src, sizeof(union sockaddr_union)); 4210 bzero(&dst, sizeof(union sockaddr_union)); 4211 src.sa.sa_len = sc->sc_src.sa.sa_len; 4212 src.sa.sa_family = sc->sc_src.sa.sa_family; 4213 dst.sa.sa_len = sc->sc_dst.sa.sa_len; 4214 dst.sa.sa_family = sc->sc_dst.sa.sa_family; 4215 4216 switch (sc->sc_src.sa.sa_family) { 4217 case 0: /*default to PF_INET*/ 4218 #ifdef INET 4219 case AF_INET: 4220 src.sin.sin_addr = mtod(m, struct ip *)->ip_src; 4221 dst.sin.sin_addr = mtod(m, struct ip *)->ip_dst; 4222 break; 4223 #endif /* INET */ 4224 #ifdef INET6 4225 case AF_INET6: 4226 src.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_src; 4227 dst.sin6.sin6_addr = mtod(m, struct ip6_hdr *)->ip6_dst; 4228 break; 4229 #endif /* INET6 */ 4230 } 4231 4232 tdb = gettdbbysrcdst(0, &src, &dst, IPPROTO_TCP); 4233 if (tdb == NULL) { 4234 if (m) 4235 m_freem(m); 4236 return (EPERM); 4237 } 4238 4239 MD5Init(&ctx); 4240 4241 switch (sc->sc_src.sa.sa_family) { 4242 case 0: /*default to PF_INET*/ 4243 #ifdef INET 4244 case AF_INET: 4245 { 4246 struct ippseudo ippseudo; 4247 4248 ippseudo.ippseudo_src = ip->ip_src; 4249 ippseudo.ippseudo_dst = ip->ip_dst; 4250 ippseudo.ippseudo_pad = 0; 4251 ippseudo.ippseudo_p = IPPROTO_TCP; 4252 ippseudo.ippseudo_len = htons(tlen - hlen); 4253 4254 MD5Update(&ctx, (char *)&ippseudo, 4255 sizeof(struct ippseudo)); 4256 4257 } 4258 break; 4259 #endif /* INET */ 4260 #ifdef INET6 4261 case AF_INET6: 4262 { 4263 struct ip6_hdr_pseudo ip6pseudo; 4264 4265 bzero(&ip6pseudo, sizeof(ip6pseudo)); 4266 ip6pseudo.ip6ph_src = ip6->ip6_src; 4267 ip6pseudo.ip6ph_dst = ip6->ip6_dst; 4268 in6_clearscope(&ip6pseudo.ip6ph_src); 4269 in6_clearscope(&ip6pseudo.ip6ph_dst); 4270 ip6pseudo.ip6ph_nxt = IPPROTO_TCP; 4271 ip6pseudo.ip6ph_len = htonl(tlen - hlen); 4272 4273 MD5Update(&ctx, (char *)&ip6pseudo, 4274 sizeof(ip6pseudo)); 4275 } 4276 break; 4277 #endif /* INET6 */ 4278 } 4279 4280 th->th_sum = 0; 4281 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr)); 4282 MD5Update(&ctx, tdb->tdb_amxkey, tdb->tdb_amxkeylen); 4283 4284 /* Send signature option */ 4285 *(optp++) = TCPOPT_SIGNATURE; 4286 *(optp++) = TCPOLEN_SIGNATURE; 4287 4288 MD5Final(optp, &ctx); 4289 optp += 16; 4290 4291 /* Pad options list to the next 32 bit boundary and 4292 * terminate it. 4293 */ 4294 *optp++ = TCPOPT_NOP; 4295 *optp++ = TCPOPT_EOL; 4296 } 4297 #endif /* TCP_SIGNATURE */ 4298 4299 /* Compute the packet's checksum. */ 4300 switch (sc->sc_src.sa.sa_family) { 4301 case AF_INET: 4302 ip->ip_len = htons(tlen - hlen); 4303 th->th_sum = 0; 4304 th->th_sum = in_cksum(m, tlen); 4305 break; 4306 #ifdef INET6 4307 case AF_INET6: 4308 ip6->ip6_plen = htons(tlen - hlen); 4309 th->th_sum = 0; 4310 th->th_sum = in6_cksum(m, IPPROTO_TCP, hlen, tlen - hlen); 4311 break; 4312 #endif 4313 } 4314 4315 /* 4316 * Fill in some straggling IP bits. Note the stack expects 4317 * ip_len to be in host order, for convenience. 4318 */ 4319 switch (sc->sc_src.sa.sa_family) { 4320 #ifdef INET 4321 case AF_INET: 4322 ip->ip_len = htons(tlen); 4323 ip->ip_ttl = ip_defttl; 4324 /* XXX tos? */ 4325 break; 4326 #endif 4327 #ifdef INET6 4328 case AF_INET6: 4329 ip6->ip6_vfc &= ~IPV6_VERSION_MASK; 4330 ip6->ip6_vfc |= IPV6_VERSION; 4331 ip6->ip6_plen = htons(tlen - hlen); 4332 /* ip6_hlim will be initialized afterwards */ 4333 /* leave flowlabel = 0, it is legal and require no state mgmt */ 4334 break; 4335 #endif 4336 } 4337 4338 /* use IPsec policy from listening socket, on SYN ACK */ 4339 inp = sc->sc_tp ? sc->sc_tp->t_inpcb : NULL; 4340 4341 switch (sc->sc_src.sa.sa_family) { 4342 #ifdef INET 4343 case AF_INET: 4344 error = ip_output(m, sc->sc_ipopts, ro, 4345 (ip_mtudisc ? IP_MTUDISC : 0), 4346 (struct ip_moptions *)NULL, inp); 4347 break; 4348 #endif 4349 #ifdef INET6 4350 case AF_INET6: 4351 ip6->ip6_hlim = in6_selecthlim(NULL, 4352 ro->ro_rt ? ro->ro_rt->rt_ifp : NULL); 4353 4354 error = ip6_output(m, NULL /*XXX*/, (struct route_in6 *)ro, 0, 4355 (struct ip6_moptions *)0, NULL); 4356 break; 4357 #endif 4358 default: 4359 error = EAFNOSUPPORT; 4360 break; 4361 } 4362 return (error); 4363 } 4364