1 /* 2 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved. 3 * Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved. 4 * 5 * This code is derived from software contributed to The DragonFly Project 6 * by Jeffrey M. Hsu. 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 DragonFly Project nor the names of its 17 * contributors may be used to endorse or promote products derived 18 * from this software without specific, prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 31 * SUCH DAMAGE. 32 */ 33 34 /* 35 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved. 36 * 37 * License terms: all terms for the DragonFly license above plus the following: 38 * 39 * 4. All advertising materials mentioning features or use of this software 40 * must display the following acknowledgement: 41 * 42 * This product includes software developed by Jeffrey M. Hsu 43 * for the DragonFly Project. 44 * 45 * This requirement may be waived with permission from Jeffrey Hsu. 46 * This requirement will sunset and may be removed on July 8 2005, 47 * after which the standard DragonFly license (as shown above) will 48 * apply. 49 */ 50 51 /* 52 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 53 * The Regents of the University of California. All rights reserved. 54 * 55 * Redistribution and use in source and binary forms, with or without 56 * modification, are permitted provided that the following conditions 57 * are met: 58 * 1. Redistributions of source code must retain the above copyright 59 * notice, this list of conditions and the following disclaimer. 60 * 2. Redistributions in binary form must reproduce the above copyright 61 * notice, this list of conditions and the following disclaimer in the 62 * documentation and/or other materials provided with the distribution. 63 * 3. All advertising materials mentioning features or use of this software 64 * must display the following acknowledgement: 65 * This product includes software developed by the University of 66 * California, Berkeley and its contributors. 67 * 4. Neither the name of the University nor the names of its contributors 68 * may be used to endorse or promote products derived from this software 69 * without specific prior written permission. 70 * 71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 81 * SUCH DAMAGE. 82 * 83 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 84 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $ 85 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.36 2004/08/14 06:41:33 hsu Exp $ 86 */ 87 88 #include "opt_ipfw.h" /* for ipfw_fwd */ 89 #include "opt_inet6.h" 90 #include "opt_ipsec.h" 91 #include "opt_tcpdebug.h" 92 #include "opt_tcp_input.h" 93 94 #include <sys/param.h> 95 #include <sys/systm.h> 96 #include <sys/kernel.h> 97 #include <sys/sysctl.h> 98 #include <sys/malloc.h> 99 #include <sys/mbuf.h> 100 #include <sys/proc.h> /* for proc0 declaration */ 101 #include <sys/protosw.h> 102 #include <sys/socket.h> 103 #include <sys/socketvar.h> 104 #include <sys/syslog.h> 105 #include <sys/in_cksum.h> 106 107 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 108 #include <machine/stdarg.h> 109 110 #include <net/if.h> 111 #include <net/route.h> 112 113 #include <netinet/in.h> 114 #include <netinet/in_systm.h> 115 #include <netinet/ip.h> 116 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 117 #include <netinet/in_var.h> 118 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 119 #include <netinet/in_pcb.h> 120 #include <netinet/ip_var.h> 121 #include <netinet/ip6.h> 122 #include <netinet/icmp6.h> 123 #include <netinet6/nd6.h> 124 #include <netinet6/ip6_var.h> 125 #include <netinet6/in6_pcb.h> 126 #include <netinet/tcp.h> 127 #include <netinet/tcp_fsm.h> 128 #include <netinet/tcp_seq.h> 129 #include <netinet/tcp_timer.h> 130 #include <netinet/tcp_var.h> 131 #include <netinet6/tcp6_var.h> 132 #include <netinet/tcpip.h> 133 134 #ifdef TCPDEBUG 135 #include <netinet/tcp_debug.h> 136 137 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */ 138 struct tcphdr tcp_savetcp; 139 #endif 140 141 #ifdef FAST_IPSEC 142 #include <netipsec/ipsec.h> 143 #include <netipsec/ipsec6.h> 144 #endif 145 146 #ifdef IPSEC 147 #include <netinet6/ipsec.h> 148 #include <netinet6/ipsec6.h> 149 #include <netproto/key/key.h> 150 #endif 151 152 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry"); 153 154 static const int tcprexmtthresh = 3; 155 tcp_cc tcp_ccgen; 156 static int log_in_vain = 0; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 158 &log_in_vain, 0, "Log all incoming TCP connections"); 159 160 static int blackhole = 0; 161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 162 &blackhole, 0, "Do not send RST when dropping refused connections"); 163 164 int tcp_delack_enabled = 1; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 166 &tcp_delack_enabled, 0, 167 "Delay ACK to try and piggyback it onto a data packet"); 168 169 #ifdef TCP_DROP_SYNFIN 170 static int drop_synfin = 0; 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 172 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 173 #endif 174 175 static int tcp_do_limitedtransmit = 1; 176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW, 177 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)"); 178 179 static int tcp_do_early_retransmit = 0; 180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW, 181 &tcp_do_early_retransmit, 0, "Early retransmit"); 182 183 static int tcp_do_rfc3390 = 1; 184 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 185 &tcp_do_rfc3390, 0, 186 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 187 188 static int tcp_do_eifel_detect = 1; 189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW, 190 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)"); 191 192 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 193 "TCP Segment Reassembly Queue"); 194 195 int tcp_reass_maxseg = 0; 196 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD, 197 &tcp_reass_maxseg, 0, 198 "Global maximum number of TCP Segments in Reassembly Queue"); 199 200 int tcp_reass_qsize = 0; 201 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 202 &tcp_reass_qsize, 0, 203 "Global number of TCP Segments currently in Reassembly Queue"); 204 205 static int tcp_reass_overflows = 0; 206 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 207 &tcp_reass_overflows, 0, 208 "Global number of TCP Segment Reassembly Queue Overflows"); 209 210 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t); 211 static void tcp_pulloutofband(struct socket *, 212 struct tcphdr *, struct mbuf *, int); 213 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 214 struct mbuf *); 215 static void tcp_xmit_timer(struct tcpcb *, int); 216 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 217 218 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 219 #ifdef INET6 220 #define ND6_HINT(tp) \ 221 do { \ 222 if ((tp) && (tp)->t_inpcb && \ 223 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \ 224 (tp)->t_inpcb->in6p_route.ro_rt) \ 225 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \ 226 } while (0) 227 #else 228 #define ND6_HINT(tp) 229 #endif 230 231 /* 232 * Indicate whether this ack should be delayed. We can delay the ack if 233 * - delayed acks are enabled and 234 * - there is no delayed ack timer in progress and 235 * - our last ack wasn't a 0-sized window. We never want to delay 236 * the ack that opens up a 0-sized window. 237 */ 238 #define DELAY_ACK(tp) \ 239 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \ 240 !(tp->t_flags & TF_RXWIN0SENT)) 241 242 static int 243 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m) 244 { 245 struct tseg_qent *q; 246 struct tseg_qent *p = NULL; 247 struct tseg_qent *nq; 248 struct tseg_qent *te; 249 struct socket *so = tp->t_inpcb->inp_socket; 250 int flags; 251 252 /* 253 * Call with th == NULL after become established to 254 * force pre-ESTABLISHED data up to user socket. 255 */ 256 if (th == NULL) 257 goto present; 258 259 /* 260 * Limit the number of segments in the reassembly queue to prevent 261 * holding on to too many segments (and thus running out of mbufs). 262 * Make sure to let the missing segment through which caused this 263 * queue. Always keep one global queue entry spare to be able to 264 * process the missing segment. 265 */ 266 if (th->th_seq != tp->rcv_nxt && 267 tcp_reass_qsize + 1 >= tcp_reass_maxseg) { 268 tcp_reass_overflows++; 269 tcpstat.tcps_rcvmemdrop++; 270 m_freem(m); 271 return (0); 272 } 273 274 /* Allocate a new queue entry. */ 275 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ, 276 M_INTWAIT | M_NULLOK); 277 if (te == NULL) { 278 tcpstat.tcps_rcvmemdrop++; 279 m_freem(m); 280 return (0); 281 } 282 tcp_reass_qsize++; 283 284 /* 285 * Find a segment which begins after this one does. 286 */ 287 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 288 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 289 break; 290 p = q; 291 } 292 293 /* 294 * If there is a preceding segment, it may provide some of 295 * our data already. If so, drop the data from the incoming 296 * segment. If it provides all of our data, drop us. 297 */ 298 if (p != NULL) { 299 int i; 300 /* conversion to int (in i) handles seq wraparound */ 301 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 302 if (i > 0) { 303 if (i >= *tlenp) { 304 tcpstat.tcps_rcvduppack++; 305 tcpstat.tcps_rcvdupbyte += *tlenp; 306 m_freem(m); 307 free(te, M_TSEGQ); 308 tcp_reass_qsize--; 309 /* 310 * Try to present any queued data 311 * at the left window edge to the user. 312 * This is needed after the 3-WHS 313 * completes. 314 */ 315 goto present; /* ??? */ 316 } 317 m_adj(m, i); 318 *tlenp -= i; 319 th->th_seq += i; 320 } 321 } 322 tcpstat.tcps_rcvoopack++; 323 tcpstat.tcps_rcvoobyte += *tlenp; 324 325 /* 326 * While we overlap succeeding segments trim them or, 327 * if they are completely covered, dequeue them. 328 */ 329 while (q) { 330 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 331 if (i <= 0) 332 break; 333 if (i < q->tqe_len) { 334 q->tqe_th->th_seq += i; 335 q->tqe_len -= i; 336 m_adj(q->tqe_m, i); 337 break; 338 } 339 340 nq = LIST_NEXT(q, tqe_q); 341 LIST_REMOVE(q, tqe_q); 342 m_freem(q->tqe_m); 343 free(q, M_TSEGQ); 344 tcp_reass_qsize--; 345 q = nq; 346 } 347 348 /* Insert the new segment queue entry into place. */ 349 te->tqe_m = m; 350 te->tqe_th = th; 351 te->tqe_len = *tlenp; 352 353 if (p == NULL) { 354 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 355 } else { 356 LIST_INSERT_AFTER(p, te, tqe_q); 357 } 358 359 present: 360 /* 361 * Present data to user, advancing rcv_nxt through 362 * completed sequence space. 363 */ 364 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 365 return (0); 366 q = LIST_FIRST(&tp->t_segq); 367 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt) 368 return (0); 369 do { 370 tp->rcv_nxt += q->tqe_len; 371 flags = q->tqe_th->th_flags & TH_FIN; 372 nq = LIST_NEXT(q, tqe_q); 373 LIST_REMOVE(q, tqe_q); 374 if (so->so_state & SS_CANTRCVMORE) 375 m_freem(q->tqe_m); 376 else 377 sbappend(&so->so_rcv, q->tqe_m); 378 free(q, M_TSEGQ); 379 tcp_reass_qsize--; 380 q = nq; 381 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 382 ND6_HINT(tp); 383 sorwakeup(so); 384 return (flags); 385 } 386 387 /* 388 * TCP input routine, follows pages 65-76 of the 389 * protocol specification dated September, 1981 very closely. 390 */ 391 #ifdef INET6 392 int 393 tcp6_input(struct mbuf **mp, int *offp, int proto) 394 { 395 struct mbuf *m = *mp; 396 struct in6_ifaddr *ia6; 397 398 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 399 400 /* 401 * draft-itojun-ipv6-tcp-to-anycast 402 * better place to put this in? 403 */ 404 ia6 = ip6_getdstifaddr(m); 405 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 406 struct ip6_hdr *ip6; 407 408 ip6 = mtod(m, struct ip6_hdr *); 409 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 410 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 411 return (IPPROTO_DONE); 412 } 413 414 tcp_input(m, *offp, proto); 415 return (IPPROTO_DONE); 416 } 417 #endif 418 419 void 420 tcp_input(struct mbuf *m, ...) 421 { 422 __va_list ap; 423 int off0, proto; 424 struct tcphdr *th; 425 struct ip *ip = NULL; 426 struct ipovly *ipov; 427 struct inpcb *inp = NULL; 428 u_char *optp = NULL; 429 int optlen = 0; 430 int len, tlen, off; 431 int drop_hdrlen; 432 struct tcpcb *tp = NULL; 433 int thflags; 434 struct socket *so = 0; 435 int todrop, acked; 436 boolean_t ourfinisacked, needoutput = FALSE; 437 u_long tiwin; 438 int recvwin; 439 struct tcpopt to; /* options in this segment */ 440 struct rmxp_tao *taop; /* pointer to our TAO cache entry */ 441 struct rmxp_tao tao_noncached; /* in case there's no cached entry */ 442 struct sockaddr_in *next_hop = NULL; 443 int rstreason; /* For badport_bandlim accounting purposes */ 444 int cpu; 445 struct ip6_hdr *ip6 = NULL; 446 #ifdef INET6 447 boolean_t isipv6; 448 #else 449 const boolean_t isipv6 = FALSE; 450 #endif 451 #ifdef TCPDEBUG 452 short ostate = 0; 453 #endif 454 455 __va_start(ap, m); 456 off0 = __va_arg(ap, int); 457 proto = __va_arg(ap, int); 458 __va_end(ap); 459 460 tcpstat.tcps_rcvtotal++; 461 462 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */ 463 while (m->m_type == MT_TAG) { 464 if (m->_m_tag_id == PACKET_TAG_IPFORWARD) 465 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data; 466 m = m->m_next; 467 } 468 469 #ifdef INET6 470 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE; 471 #endif 472 473 if (isipv6) { 474 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 475 ip6 = mtod(m, struct ip6_hdr *); 476 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 477 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 478 tcpstat.tcps_rcvbadsum++; 479 goto drop; 480 } 481 th = (struct tcphdr *)((caddr_t)ip6 + off0); 482 483 /* 484 * Be proactive about unspecified IPv6 address in source. 485 * As we use all-zero to indicate unbounded/unconnected pcb, 486 * unspecified IPv6 address can be used to confuse us. 487 * 488 * Note that packets with unspecified IPv6 destination is 489 * already dropped in ip6_input. 490 */ 491 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 492 /* XXX stat */ 493 goto drop; 494 } 495 } else { 496 /* 497 * Get IP and TCP header together in first mbuf. 498 * Note: IP leaves IP header in first mbuf. 499 */ 500 if (off0 > sizeof(struct ip)) { 501 ip_stripoptions(m); 502 off0 = sizeof(struct ip); 503 } 504 /* already checked and pulled up in ip_demux() */ 505 KASSERT(m->m_len >= sizeof(struct tcpiphdr), 506 ("TCP header not in one mbuf: m->m_len %d", m->m_len)); 507 ip = mtod(m, struct ip *); 508 ipov = (struct ipovly *)ip; 509 th = (struct tcphdr *)((caddr_t)ip + off0); 510 tlen = ip->ip_len; 511 512 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 513 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 514 th->th_sum = m->m_pkthdr.csum_data; 515 else 516 th->th_sum = in_pseudo(ip->ip_src.s_addr, 517 ip->ip_dst.s_addr, 518 htonl(m->m_pkthdr.csum_data + 519 ip->ip_len + 520 IPPROTO_TCP)); 521 th->th_sum ^= 0xffff; 522 } else { 523 /* 524 * Checksum extended TCP header and data. 525 */ 526 len = sizeof(struct ip) + tlen; 527 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 528 ipov->ih_len = (u_short)tlen; 529 ipov->ih_len = htons(ipov->ih_len); 530 th->th_sum = in_cksum(m, len); 531 } 532 if (th->th_sum) { 533 tcpstat.tcps_rcvbadsum++; 534 goto drop; 535 } 536 #ifdef INET6 537 /* Re-initialization for later version check */ 538 ip->ip_v = IPVERSION; 539 #endif 540 } 541 542 /* 543 * Check that TCP offset makes sense, 544 * pull out TCP options and adjust length. XXX 545 */ 546 off = th->th_off << 2; 547 /* already checked and pulled up in ip_demux() */ 548 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen, 549 ("bad TCP data offset %d (tlen %d)", off, tlen)); 550 tlen -= off; /* tlen is used instead of ti->ti_len */ 551 if (off > sizeof(struct tcphdr)) { 552 if (isipv6) { 553 IP6_EXTHDR_CHECK(m, off0, off, ); 554 ip6 = mtod(m, struct ip6_hdr *); 555 th = (struct tcphdr *)((caddr_t)ip6 + off0); 556 } else { 557 /* already pulled up in ip_demux() */ 558 KASSERT(m->m_len >= sizeof(struct ip) + off, 559 ("TCP header and options not in one mbuf: " 560 "m_len %d, off %d", m->m_len, off)); 561 } 562 optlen = off - sizeof(struct tcphdr); 563 optp = (u_char *)(th + 1); 564 } 565 thflags = th->th_flags; 566 567 #ifdef TCP_DROP_SYNFIN 568 /* 569 * If the drop_synfin option is enabled, drop all packets with 570 * both the SYN and FIN bits set. This prevents e.g. nmap from 571 * identifying the TCP/IP stack. 572 * 573 * This is a violation of the TCP specification. 574 */ 575 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN)) 576 goto drop; 577 #endif 578 579 /* 580 * Convert TCP protocol specific fields to host format. 581 */ 582 th->th_seq = ntohl(th->th_seq); 583 th->th_ack = ntohl(th->th_ack); 584 th->th_win = ntohs(th->th_win); 585 th->th_urp = ntohs(th->th_urp); 586 587 /* 588 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 589 * until after ip6_savecontrol() is called and before other functions 590 * which don't want those proto headers. 591 * Because ip6_savecontrol() is going to parse the mbuf to 592 * search for data to be passed up to user-land, it wants mbuf 593 * parameters to be unchanged. 594 * XXX: the call of ip6_savecontrol() has been obsoleted based on 595 * latest version of the advanced API (20020110). 596 */ 597 drop_hdrlen = off0 + off; 598 599 /* 600 * Locate pcb for segment. 601 */ 602 findpcb: 603 /* IPFIREWALL_FORWARD section */ 604 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */ 605 /* 606 * Transparently forwarded. Pretend to be the destination. 607 * already got one like this? 608 */ 609 cpu = mycpu->gd_cpuid; 610 inp = in_pcblookup_hash(&tcbinfo[cpu], 611 ip->ip_src, th->th_sport, 612 ip->ip_dst, th->th_dport, 613 0, m->m_pkthdr.rcvif); 614 if (!inp) { 615 /* 616 * It's new. Try to find the ambushing socket. 617 */ 618 619 /* 620 * The rest of the ipfw code stores the port in 621 * host order. XXX 622 * (The IP address is still in network order.) 623 */ 624 in_port_t dport = next_hop->sin_port ? 625 htons(next_hop->sin_port) : 626 th->th_dport; 627 628 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport, 629 next_hop->sin_addr.s_addr, dport); 630 inp = in_pcblookup_hash(&tcbinfo[cpu], 631 ip->ip_src, th->th_sport, 632 next_hop->sin_addr, dport, 633 1, m->m_pkthdr.rcvif); 634 } 635 } else { 636 if (isipv6) { 637 inp = in6_pcblookup_hash(&tcbinfo[0], 638 &ip6->ip6_src, th->th_sport, 639 &ip6->ip6_dst, th->th_dport, 640 1, m->m_pkthdr.rcvif); 641 } else { 642 cpu = mycpu->gd_cpuid; 643 inp = in_pcblookup_hash(&tcbinfo[cpu], 644 ip->ip_src, th->th_sport, 645 ip->ip_dst, th->th_dport, 646 1, m->m_pkthdr.rcvif); 647 } 648 } 649 650 /* 651 * If the state is CLOSED (i.e., TCB does not exist) then 652 * all data in the incoming segment is discarded. 653 * If the TCB exists but is in CLOSED state, it is embryonic, 654 * but should either do a listen or a connect soon. 655 */ 656 if (inp == NULL) { 657 if (log_in_vain) { 658 #ifdef INET6 659 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 660 #else 661 char dbuf[4 * sizeof "123"], sbuf[4 * sizeof "123"]; 662 #endif 663 if (isipv6) { 664 strcpy(dbuf, "["); 665 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 666 strcat(dbuf, "]"); 667 strcpy(sbuf, "["); 668 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 669 strcat(sbuf, "]"); 670 } else { 671 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 672 strcpy(sbuf, inet_ntoa(ip->ip_src)); 673 } 674 switch (log_in_vain) { 675 case 1: 676 if (!(thflags & TH_SYN)) 677 break; 678 case 2: 679 log(LOG_INFO, 680 "Connection attempt to TCP %s:%d " 681 "from %s:%d flags:0x%02x\n", 682 dbuf, ntohs(th->th_dport), sbuf, 683 ntohs(th->th_sport), thflags); 684 break; 685 default: 686 break; 687 } 688 } 689 if (blackhole) { 690 switch (blackhole) { 691 case 1: 692 if (thflags & TH_SYN) 693 goto drop; 694 break; 695 case 2: 696 goto drop; 697 default: 698 goto drop; 699 } 700 } 701 rstreason = BANDLIM_RST_CLOSEDPORT; 702 goto dropwithreset; 703 } 704 705 #ifdef IPSEC 706 if (isipv6) { 707 if (ipsec6_in_reject_so(m, inp->inp_socket)) { 708 ipsec6stat.in_polvio++; 709 goto drop; 710 } 711 } else { 712 if (ipsec4_in_reject_so(m, inp->inp_socket)) { 713 ipsecstat.in_polvio++; 714 goto drop; 715 } 716 } 717 #endif 718 #ifdef FAST_IPSEC 719 if (isipv6) { 720 if (ipsec6_in_reject(m, inp)) 721 goto drop; 722 } else { 723 if (ipsec4_in_reject(m, inp)) 724 goto drop; 725 } 726 #endif 727 728 tp = intotcpcb(inp); 729 if (tp == NULL) { 730 rstreason = BANDLIM_RST_CLOSEDPORT; 731 goto dropwithreset; 732 } 733 if (tp->t_state <= TCPS_CLOSED) 734 goto drop; 735 736 /* Unscale the window into a 32-bit value. */ 737 if (!(thflags & TH_SYN)) 738 tiwin = th->th_win << tp->snd_scale; 739 else 740 tiwin = th->th_win; 741 742 so = inp->inp_socket; 743 744 #ifdef TCPDEBUG 745 if (so->so_options & SO_DEBUG) { 746 ostate = tp->t_state; 747 if (isipv6) 748 bcopy(ip6, tcp_saveipgen, sizeof(*ip6)); 749 else 750 bcopy(ip, tcp_saveipgen, sizeof(*ip)); 751 tcp_savetcp = *th; 752 } 753 #endif 754 755 bzero(&to, sizeof(to)); 756 757 if (so->so_options & SO_ACCEPTCONN) { 758 struct in_conninfo inc; 759 760 #ifdef INET6 761 inc.inc_isipv6 = (isipv6 == TRUE); 762 #endif 763 if (isipv6) { 764 inc.inc6_faddr = ip6->ip6_src; 765 inc.inc6_laddr = ip6->ip6_dst; 766 inc.inc6_route.ro_rt = NULL; /* XXX */ 767 } else { 768 inc.inc_faddr = ip->ip_src; 769 inc.inc_laddr = ip->ip_dst; 770 inc.inc_route.ro_rt = NULL; /* XXX */ 771 } 772 inc.inc_fport = th->th_sport; 773 inc.inc_lport = th->th_dport; 774 775 /* 776 * If the state is LISTEN then ignore segment if it contains 777 * a RST. If the segment contains an ACK then it is bad and 778 * send a RST. If it does not contain a SYN then it is not 779 * interesting; drop it. 780 * 781 * If the state is SYN_RECEIVED (syncache) and seg contains 782 * an ACK, but not for our SYN/ACK, send a RST. If the seg 783 * contains a RST, check the sequence number to see if it 784 * is a valid reset segment. 785 */ 786 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) { 787 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) { 788 if (!syncache_expand(&inc, th, &so, m)) { 789 /* 790 * No syncache entry, or ACK was not 791 * for our SYN/ACK. Send a RST. 792 */ 793 tcpstat.tcps_badsyn++; 794 rstreason = BANDLIM_RST_OPENPORT; 795 goto dropwithreset; 796 } 797 if (so == NULL) 798 /* 799 * Could not complete 3-way handshake, 800 * connection is being closed down, and 801 * syncache will free mbuf. 802 */ 803 return; 804 /* 805 * Socket is created in state SYN_RECEIVED. 806 * Continue processing segment. 807 */ 808 inp = sotoinpcb(so); 809 tp = intotcpcb(inp); 810 /* 811 * This is what would have happened in 812 * tcp_output() when the SYN,ACK was sent. 813 */ 814 tp->snd_up = tp->snd_una; 815 tp->snd_max = tp->snd_nxt = tp->iss + 1; 816 tp->last_ack_sent = tp->rcv_nxt; 817 /* 818 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled 819 * until the _second_ ACK is received: 820 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window. 821 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale, 822 * move to ESTAB, set snd_wnd to tiwin. 823 */ 824 tp->snd_wnd = tiwin; /* unscaled */ 825 goto after_listen; 826 } 827 if (thflags & TH_RST) { 828 syncache_chkrst(&inc, th); 829 goto drop; 830 } 831 if (thflags & TH_ACK) { 832 syncache_badack(&inc); 833 tcpstat.tcps_badsyn++; 834 rstreason = BANDLIM_RST_OPENPORT; 835 goto dropwithreset; 836 } 837 goto drop; 838 } 839 840 /* 841 * Segment's flags are (SYN) or (SYN | FIN). 842 */ 843 #ifdef INET6 844 /* 845 * If deprecated address is forbidden, 846 * we do not accept SYN to deprecated interface 847 * address to prevent any new inbound connection from 848 * getting established. 849 * When we do not accept SYN, we send a TCP RST, 850 * with deprecated source address (instead of dropping 851 * it). We compromise it as it is much better for peer 852 * to send a RST, and RST will be the final packet 853 * for the exchange. 854 * 855 * If we do not forbid deprecated addresses, we accept 856 * the SYN packet. RFC2462 does not suggest dropping 857 * SYN in this case. 858 * If we decipher RFC2462 5.5.4, it says like this: 859 * 1. use of deprecated addr with existing 860 * communication is okay - "SHOULD continue to be 861 * used" 862 * 2. use of it with new communication: 863 * (2a) "SHOULD NOT be used if alternate address 864 * with sufficient scope is available" 865 * (2b) nothing mentioned otherwise. 866 * Here we fall into (2b) case as we have no choice in 867 * our source address selection - we must obey the peer. 868 * 869 * The wording in RFC2462 is confusing, and there are 870 * multiple description text for deprecated address 871 * handling - worse, they are not exactly the same. 872 * I believe 5.5.4 is the best one, so we follow 5.5.4. 873 */ 874 if (isipv6 && !ip6_use_deprecated) { 875 struct in6_ifaddr *ia6; 876 877 if ((ia6 = ip6_getdstifaddr(m)) && 878 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 879 tp = NULL; 880 rstreason = BANDLIM_RST_OPENPORT; 881 goto dropwithreset; 882 } 883 } 884 #endif 885 /* 886 * If it is from this socket, drop it, it must be forged. 887 * Don't bother responding if the destination was a broadcast. 888 */ 889 if (th->th_dport == th->th_sport) { 890 if (isipv6) { 891 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 892 &ip6->ip6_src)) 893 goto drop; 894 } else { 895 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 896 goto drop; 897 } 898 } 899 /* 900 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 901 * 902 * Note that it is quite possible to receive unicast 903 * link-layer packets with a broadcast IP address. Use 904 * in_broadcast() to find them. 905 */ 906 if (m->m_flags & (M_BCAST | M_MCAST)) 907 goto drop; 908 if (isipv6) { 909 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 910 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 911 goto drop; 912 } else { 913 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 914 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 915 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 916 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 917 goto drop; 918 } 919 /* 920 * SYN appears to be valid; create compressed TCP state 921 * for syncache, or perform t/tcp connection. 922 */ 923 if (so->so_qlen <= so->so_qlimit) { 924 tcp_dooptions(&to, optp, optlen, TRUE); 925 if (!syncache_add(&inc, &to, th, &so, m)) 926 goto drop; 927 if (so == NULL) 928 /* 929 * Entry added to syncache, mbuf used to 930 * send SYN,ACK packet. 931 */ 932 return; 933 /* 934 * Segment passed TAO tests. 935 */ 936 inp = sotoinpcb(so); 937 tp = intotcpcb(inp); 938 tp->snd_wnd = tiwin; 939 tp->t_starttime = ticks; 940 tp->t_state = TCPS_ESTABLISHED; 941 942 /* 943 * If there is a FIN, or if there is data and the 944 * connection is local, then delay SYN,ACK(SYN) in 945 * the hope of piggy-backing it on a response 946 * segment. Otherwise must send ACK now in case 947 * the other side is slow starting. 948 */ 949 if (DELAY_ACK(tp) && 950 ((thflags & TH_FIN) || 951 (tlen != 0 && 952 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 953 (!isipv6 && in_localaddr(inp->inp_faddr)))))) { 954 callout_reset(tp->tt_delack, tcp_delacktime, 955 tcp_timer_delack, tp); 956 tp->t_flags |= TF_NEEDSYN; 957 } else 958 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 959 960 tcpstat.tcps_connects++; 961 soisconnected(so); 962 goto trimthenstep6; 963 } 964 goto drop; 965 } 966 after_listen: 967 968 /* should not happen - syncache should pick up these connections */ 969 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state")); 970 971 /* 972 * Segment received on connection. 973 * Reset idle time and keep-alive timer. 974 */ 975 tp->t_rcvtime = ticks; 976 if (TCPS_HAVEESTABLISHED(tp->t_state)) 977 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 978 979 /* 980 * Process options. 981 * XXX this is tradtitional behavior, may need to be cleaned up. 982 */ 983 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0); 984 if (thflags & TH_SYN) { 985 if (to.to_flags & TOF_SCALE) { 986 tp->t_flags |= TF_RCVD_SCALE; 987 tp->requested_s_scale = to.to_requested_s_scale; 988 } 989 if (to.to_flags & TOF_TS) { 990 tp->t_flags |= TF_RCVD_TSTMP; 991 tp->ts_recent = to.to_tsval; 992 tp->ts_recent_age = ticks; 993 } 994 if (to.to_flags & (TOF_CC | TOF_CCNEW)) 995 tp->t_flags |= TF_RCVD_CC; 996 if (to.to_flags & TOF_MSS) 997 tcp_mss(tp, to.to_mss); 998 } 999 1000 /* 1001 * Header prediction: check for the two common cases 1002 * of a uni-directional data xfer. If the packet has 1003 * no control flags, is in-sequence, the window didn't 1004 * change and we're not retransmitting, it's a 1005 * candidate. If the length is zero and the ack moved 1006 * forward, we're the sender side of the xfer. Just 1007 * free the data acked & wake any higher level process 1008 * that was blocked waiting for space. If the length 1009 * is non-zero and the ack didn't move, we're the 1010 * receiver side. If we're getting packets in-order 1011 * (the reassembly queue is empty), add the data to 1012 * the socket buffer and note that we need a delayed ack. 1013 * Make sure that the hidden state-flags are also off. 1014 * Since we check for TCPS_ESTABLISHED above, it can only 1015 * be TH_NEEDSYN. 1016 */ 1017 if (tp->t_state == TCPS_ESTABLISHED && 1018 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1019 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) && 1020 (!(to.to_flags & TOF_TS) || 1021 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1022 /* 1023 * Using the CC option is compulsory if once started: 1024 * the segment is OK if no T/TCP was negotiated or 1025 * if the segment has a CC option equal to CCrecv 1026 */ 1027 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1028 ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) && 1029 th->th_seq == tp->rcv_nxt && 1030 tiwin && tiwin == tp->snd_wnd && 1031 tp->snd_nxt == tp->snd_max) { 1032 1033 /* 1034 * If last ACK falls within this segment's sequence numbers, 1035 * record the timestamp. 1036 * NOTE that the test is modified according to the latest 1037 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1038 */ 1039 if ((to.to_flags & TOF_TS) && 1040 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1041 tp->ts_recent_age = ticks; 1042 tp->ts_recent = to.to_tsval; 1043 } 1044 1045 if (tlen == 0) { 1046 if (SEQ_GT(th->th_ack, tp->snd_una) && 1047 SEQ_LEQ(th->th_ack, tp->snd_max) && 1048 tp->snd_cwnd >= tp->snd_wnd && 1049 !IN_FASTRECOVERY(tp)) { 1050 /* 1051 * this is a pure ack for outstanding data. 1052 */ 1053 ++tcpstat.tcps_predack; 1054 /* 1055 * "bad retransmit" recovery 1056 * 1057 * If Eifel detection applies, then 1058 * it is deterministic, so use it 1059 * unconditionally over the old heuristic. 1060 * Otherwise, fall back to the old heuristic. 1061 */ 1062 if (tcp_do_eifel_detect && 1063 (to.to_flags & TOF_TS) && to.to_tsecr && 1064 (tp->t_flags & TF_FIRSTACCACK)) { 1065 /* Eifel detection applicable. */ 1066 if (to.to_tsecr < tp->t_rexmtTS) { 1067 tcp_revert_congestion_state(tp); 1068 ++tcpstat.tcps_eifeldetected; 1069 } 1070 } else if (tp->t_rxtshift == 1 && 1071 ticks < tp->t_badrxtwin) { 1072 tcp_revert_congestion_state(tp); 1073 ++tcpstat.tcps_rttdetected; 1074 } 1075 tp->t_flags &= ~(TF_FIRSTACCACK | 1076 TF_FASTREXMT | TF_EARLYREXMT); 1077 /* 1078 * Recalculate the retransmit timer / rtt. 1079 * 1080 * Some machines (certain windows boxes) 1081 * send broken timestamp replies during the 1082 * SYN+ACK phase, ignore timestamps of 0. 1083 */ 1084 if ((to.to_flags & TOF_TS) && to.to_tsecr) { 1085 tcp_xmit_timer(tp, 1086 ticks - to.to_tsecr + 1); 1087 } else if (tp->t_rtttime && 1088 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1089 tcp_xmit_timer(tp, 1090 ticks - tp->t_rtttime); 1091 } 1092 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1093 acked = th->th_ack - tp->snd_una; 1094 tcpstat.tcps_rcvackpack++; 1095 tcpstat.tcps_rcvackbyte += acked; 1096 sbdrop(&so->so_snd, acked); 1097 tp->snd_recover = th->th_ack - 1; 1098 tp->snd_una = th->th_ack; 1099 tp->t_dupacks = 0; 1100 m_freem(m); 1101 ND6_HINT(tp); /* some progress has been done */ 1102 1103 /* 1104 * If all outstanding data are acked, stop 1105 * retransmit timer, otherwise restart timer 1106 * using current (possibly backed-off) value. 1107 * If process is waiting for space, 1108 * wakeup/selwakeup/signal. If data 1109 * are ready to send, let tcp_output 1110 * decide between more output or persist. 1111 */ 1112 if (tp->snd_una == tp->snd_max) 1113 callout_stop(tp->tt_rexmt); 1114 else if (!callout_active(tp->tt_persist)) 1115 callout_reset(tp->tt_rexmt, 1116 tp->t_rxtcur, 1117 tcp_timer_rexmt, tp); 1118 1119 sowwakeup(so); 1120 if (so->so_snd.sb_cc) 1121 (void) tcp_output(tp); 1122 return; 1123 } 1124 } else if (th->th_ack == tp->snd_una && 1125 LIST_EMPTY(&tp->t_segq) && 1126 tlen <= sbspace(&so->so_rcv)) { 1127 /* 1128 * this is a pure, in-sequence data packet 1129 * with nothing on the reassembly queue and 1130 * we have enough buffer space to take it. 1131 */ 1132 ++tcpstat.tcps_preddat; 1133 tp->rcv_nxt += tlen; 1134 tcpstat.tcps_rcvpack++; 1135 tcpstat.tcps_rcvbyte += tlen; 1136 ND6_HINT(tp); /* some progress has been done */ 1137 /* 1138 * Add data to socket buffer. 1139 */ 1140 if (so->so_state & SS_CANTRCVMORE) { 1141 m_freem(m); 1142 } else { 1143 m_adj(m, drop_hdrlen); /* delayed header drop */ 1144 sbappend(&so->so_rcv, m); 1145 } 1146 sorwakeup(so); 1147 1148 /* 1149 * This code is responsible for most of the ACKs 1150 * the TCP stack sends back after receiving a data 1151 * packet. Note that the DELAY_ACK check fails if 1152 * the delack timer is already running, which results 1153 * in an ack being sent every other packet (which is 1154 * what we want). 1155 */ 1156 if (DELAY_ACK(tp)) { 1157 callout_reset(tp->tt_delack, tcp_delacktime, 1158 tcp_timer_delack, tp); 1159 } else { 1160 tp->t_flags |= TF_ACKNOW; 1161 if (!(tp->t_flags & TF_ONOUTPUTQ)) { 1162 tp->t_flags |= TF_ONOUTPUTQ; 1163 tp->tt_cpu = mycpu->gd_cpuid; 1164 TAILQ_INSERT_TAIL( 1165 &tcpcbackq[tp->tt_cpu], 1166 tp, t_outputq); 1167 } 1168 } 1169 return; 1170 } 1171 } 1172 1173 /* 1174 * Calculate amount of space in receive window, 1175 * and then do TCP input processing. 1176 * Receive window is amount of space in rcv queue, 1177 * but not less than advertised window. 1178 */ 1179 recvwin = sbspace(&so->so_rcv); 1180 if (recvwin < 0) 1181 recvwin = 0; 1182 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt)); 1183 1184 switch (tp->t_state) { 1185 /* 1186 * If the state is SYN_RECEIVED: 1187 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1188 */ 1189 case TCPS_SYN_RECEIVED: 1190 if ((thflags & TH_ACK) && 1191 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1192 SEQ_GT(th->th_ack, tp->snd_max))) { 1193 rstreason = BANDLIM_RST_OPENPORT; 1194 goto dropwithreset; 1195 } 1196 break; 1197 1198 /* 1199 * If the state is SYN_SENT: 1200 * if seg contains an ACK, but not for our SYN, drop the input. 1201 * if seg contains a RST, then drop the connection. 1202 * if seg does not contain SYN, then drop it. 1203 * Otherwise this is an acceptable SYN segment 1204 * initialize tp->rcv_nxt and tp->irs 1205 * if seg contains ack then advance tp->snd_una 1206 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1207 * arrange for segment to be acked (eventually) 1208 * continue processing rest of data/controls, beginning with URG 1209 */ 1210 case TCPS_SYN_SENT: 1211 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) { 1212 taop = &tao_noncached; 1213 bzero(taop, sizeof(*taop)); 1214 } 1215 1216 if ((thflags & TH_ACK) && 1217 (SEQ_LEQ(th->th_ack, tp->iss) || 1218 SEQ_GT(th->th_ack, tp->snd_max))) { 1219 /* 1220 * If we have a cached CCsent for the remote host, 1221 * hence we haven't just crashed and restarted, 1222 * do not send a RST. This may be a retransmission 1223 * from the other side after our earlier ACK was lost. 1224 * Our new SYN, when it arrives, will serve as the 1225 * needed ACK. 1226 */ 1227 if (taop->tao_ccsent != 0) 1228 goto drop; 1229 else { 1230 rstreason = BANDLIM_UNLIMITED; 1231 goto dropwithreset; 1232 } 1233 } 1234 if (thflags & TH_RST) { 1235 if (thflags & TH_ACK) 1236 tp = tcp_drop(tp, ECONNREFUSED); 1237 goto drop; 1238 } 1239 if (!(thflags & TH_SYN)) 1240 goto drop; 1241 tp->snd_wnd = th->th_win; /* initial send window */ 1242 tp->cc_recv = to.to_cc; /* foreign CC */ 1243 1244 tp->irs = th->th_seq; 1245 tcp_rcvseqinit(tp); 1246 if (thflags & TH_ACK) { 1247 /* 1248 * Our SYN was acked. If segment contains CC.ECHO 1249 * option, check it to make sure this segment really 1250 * matches our SYN. If not, just drop it as old 1251 * duplicate, but send an RST if we're still playing 1252 * by the old rules. If no CC.ECHO option, make sure 1253 * we don't get fooled into using T/TCP. 1254 */ 1255 if (to.to_flags & TOF_CCECHO) { 1256 if (tp->cc_send != to.to_ccecho) { 1257 if (taop->tao_ccsent != 0) 1258 goto drop; 1259 else { 1260 rstreason = BANDLIM_UNLIMITED; 1261 goto dropwithreset; 1262 } 1263 } 1264 } else 1265 tp->t_flags &= ~TF_RCVD_CC; 1266 tcpstat.tcps_connects++; 1267 soisconnected(so); 1268 /* Do window scaling on this connection? */ 1269 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1270 (TF_RCVD_SCALE | TF_REQ_SCALE)) { 1271 tp->snd_scale = tp->requested_s_scale; 1272 tp->rcv_scale = tp->request_r_scale; 1273 } 1274 /* Segment is acceptable, update cache if undefined. */ 1275 if (taop->tao_ccsent == 0) 1276 taop->tao_ccsent = to.to_ccecho; 1277 1278 tp->rcv_adv += tp->rcv_wnd; 1279 tp->snd_una++; /* SYN is acked */ 1280 /* 1281 * If there's data, delay ACK; if there's also a FIN 1282 * ACKNOW will be turned on later. 1283 */ 1284 if (DELAY_ACK(tp) && tlen != 0) 1285 callout_reset(tp->tt_delack, tcp_delacktime, 1286 tcp_timer_delack, tp); 1287 else 1288 tp->t_flags |= TF_ACKNOW; 1289 /* 1290 * Received <SYN,ACK> in SYN_SENT[*] state. 1291 * Transitions: 1292 * SYN_SENT --> ESTABLISHED 1293 * SYN_SENT* --> FIN_WAIT_1 1294 */ 1295 tp->t_starttime = ticks; 1296 if (tp->t_flags & TF_NEEDFIN) { 1297 tp->t_state = TCPS_FIN_WAIT_1; 1298 tp->t_flags &= ~TF_NEEDFIN; 1299 thflags &= ~TH_SYN; 1300 } else { 1301 tp->t_state = TCPS_ESTABLISHED; 1302 callout_reset(tp->tt_keep, tcp_keepidle, 1303 tcp_timer_keep, tp); 1304 } 1305 } else { 1306 /* 1307 * Received initial SYN in SYN-SENT[*] state => 1308 * simultaneous open. If segment contains CC option 1309 * and there is a cached CC, apply TAO test. 1310 * If it succeeds, connection is * half-synchronized. 1311 * Otherwise, do 3-way handshake: 1312 * SYN-SENT -> SYN-RECEIVED 1313 * SYN-SENT* -> SYN-RECEIVED* 1314 * If there was no CC option, clear cached CC value. 1315 */ 1316 tp->t_flags |= TF_ACKNOW; 1317 callout_stop(tp->tt_rexmt); 1318 if (to.to_flags & TOF_CC) { 1319 if (taop->tao_cc != 0 && 1320 CC_GT(to.to_cc, taop->tao_cc)) { 1321 /* 1322 * update cache and make transition: 1323 * SYN-SENT -> ESTABLISHED* 1324 * SYN-SENT* -> FIN-WAIT-1* 1325 */ 1326 taop->tao_cc = to.to_cc; 1327 tp->t_starttime = ticks; 1328 if (tp->t_flags & TF_NEEDFIN) { 1329 tp->t_state = TCPS_FIN_WAIT_1; 1330 tp->t_flags &= ~TF_NEEDFIN; 1331 } else { 1332 tp->t_state = TCPS_ESTABLISHED; 1333 callout_reset(tp->tt_keep, 1334 tcp_keepidle, 1335 tcp_timer_keep, 1336 tp); 1337 } 1338 tp->t_flags |= TF_NEEDSYN; 1339 } else 1340 tp->t_state = TCPS_SYN_RECEIVED; 1341 } else { 1342 /* CC.NEW or no option => invalidate cache */ 1343 taop->tao_cc = 0; 1344 tp->t_state = TCPS_SYN_RECEIVED; 1345 } 1346 } 1347 1348 trimthenstep6: 1349 /* 1350 * Advance th->th_seq to correspond to first data byte. 1351 * If data, trim to stay within window, 1352 * dropping FIN if necessary. 1353 */ 1354 th->th_seq++; 1355 if (tlen > tp->rcv_wnd) { 1356 todrop = tlen - tp->rcv_wnd; 1357 m_adj(m, -todrop); 1358 tlen = tp->rcv_wnd; 1359 thflags &= ~TH_FIN; 1360 tcpstat.tcps_rcvpackafterwin++; 1361 tcpstat.tcps_rcvbyteafterwin += todrop; 1362 } 1363 tp->snd_wl1 = th->th_seq - 1; 1364 tp->rcv_up = th->th_seq; 1365 /* 1366 * Client side of transaction: already sent SYN and data. 1367 * If the remote host used T/TCP to validate the SYN, 1368 * our data will be ACK'd; if so, enter normal data segment 1369 * processing in the middle of step 5, ack processing. 1370 * Otherwise, goto step 6. 1371 */ 1372 if (thflags & TH_ACK) 1373 goto process_ACK; 1374 1375 goto step6; 1376 1377 /* 1378 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1379 * if segment contains a SYN and CC [not CC.NEW] option: 1380 * if state == TIME_WAIT and connection duration > MSL, 1381 * drop packet and send RST; 1382 * 1383 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1384 * ack the FIN (and data) in retransmission queue. 1385 * Complete close and delete TCPCB. Then reprocess 1386 * segment, hoping to find new TCPCB in LISTEN state; 1387 * 1388 * else must be old SYN; drop it. 1389 * else do normal processing. 1390 */ 1391 case TCPS_LAST_ACK: 1392 case TCPS_CLOSING: 1393 case TCPS_TIME_WAIT: 1394 if ((thflags & TH_SYN) && 1395 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1396 if (tp->t_state == TCPS_TIME_WAIT && 1397 (ticks - tp->t_starttime) > tcp_msl) { 1398 rstreason = BANDLIM_UNLIMITED; 1399 goto dropwithreset; 1400 } 1401 if (CC_GT(to.to_cc, tp->cc_recv)) { 1402 tp = tcp_close(tp); 1403 goto findpcb; 1404 } 1405 else 1406 goto drop; 1407 } 1408 break; /* continue normal processing */ 1409 } 1410 1411 /* 1412 * States other than LISTEN or SYN_SENT. 1413 * First check the RST flag and sequence number since reset segments 1414 * are exempt from the timestamp and connection count tests. This 1415 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1416 * below which allowed reset segments in half the sequence space 1417 * to fall though and be processed (which gives forged reset 1418 * segments with a random sequence number a 50 percent chance of 1419 * killing a connection). 1420 * Then check timestamp, if present. 1421 * Then check the connection count, if present. 1422 * Then check that at least some bytes of segment are within 1423 * receive window. If segment begins before rcv_nxt, 1424 * drop leading data (and SYN); if nothing left, just ack. 1425 * 1426 * 1427 * If the RST bit is set, check the sequence number to see 1428 * if this is a valid reset segment. 1429 * RFC 793 page 37: 1430 * In all states except SYN-SENT, all reset (RST) segments 1431 * are validated by checking their SEQ-fields. A reset is 1432 * valid if its sequence number is in the window. 1433 * Note: this does not take into account delayed ACKs, so 1434 * we should test against last_ack_sent instead of rcv_nxt. 1435 * The sequence number in the reset segment is normally an 1436 * echo of our outgoing acknowlegement numbers, but some hosts 1437 * send a reset with the sequence number at the rightmost edge 1438 * of our receive window, and we have to handle this case. 1439 * If we have multiple segments in flight, the intial reset 1440 * segment sequence numbers will be to the left of last_ack_sent, 1441 * but they will eventually catch up. 1442 * In any case, it never made sense to trim reset segments to 1443 * fit the receive window since RFC 1122 says: 1444 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1445 * 1446 * A TCP SHOULD allow a received RST segment to include data. 1447 * 1448 * DISCUSSION 1449 * It has been suggested that a RST segment could contain 1450 * ASCII text that encoded and explained the cause of the 1451 * RST. No standard has yet been established for such 1452 * data. 1453 * 1454 * If the reset segment passes the sequence number test examine 1455 * the state: 1456 * SYN_RECEIVED STATE: 1457 * If passive open, return to LISTEN state. 1458 * If active open, inform user that connection was refused. 1459 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1460 * Inform user that connection was reset, and close tcb. 1461 * CLOSING, LAST_ACK STATES: 1462 * Close the tcb. 1463 * TIME_WAIT STATE: 1464 * Drop the segment - see Stevens, vol. 2, p. 964 and 1465 * RFC 1337. 1466 */ 1467 if (thflags & TH_RST) { 1468 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1469 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1470 switch (tp->t_state) { 1471 1472 case TCPS_SYN_RECEIVED: 1473 so->so_error = ECONNREFUSED; 1474 goto close; 1475 1476 case TCPS_ESTABLISHED: 1477 case TCPS_FIN_WAIT_1: 1478 case TCPS_FIN_WAIT_2: 1479 case TCPS_CLOSE_WAIT: 1480 so->so_error = ECONNRESET; 1481 close: 1482 tp->t_state = TCPS_CLOSED; 1483 tcpstat.tcps_drops++; 1484 tp = tcp_close(tp); 1485 break; 1486 1487 case TCPS_CLOSING: 1488 case TCPS_LAST_ACK: 1489 tp = tcp_close(tp); 1490 break; 1491 1492 case TCPS_TIME_WAIT: 1493 break; 1494 } 1495 } 1496 goto drop; 1497 } 1498 1499 /* 1500 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1501 * and it's less than ts_recent, drop it. 1502 */ 1503 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 && 1504 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1505 1506 /* Check to see if ts_recent is over 24 days old. */ 1507 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1508 /* 1509 * Invalidate ts_recent. If this segment updates 1510 * ts_recent, the age will be reset later and ts_recent 1511 * will get a valid value. If it does not, setting 1512 * ts_recent to zero will at least satisfy the 1513 * requirement that zero be placed in the timestamp 1514 * echo reply when ts_recent isn't valid. The 1515 * age isn't reset until we get a valid ts_recent 1516 * because we don't want out-of-order segments to be 1517 * dropped when ts_recent is old. 1518 */ 1519 tp->ts_recent = 0; 1520 } else { 1521 tcpstat.tcps_rcvduppack++; 1522 tcpstat.tcps_rcvdupbyte += tlen; 1523 tcpstat.tcps_pawsdrop++; 1524 if (tlen) 1525 goto dropafterack; 1526 goto drop; 1527 } 1528 } 1529 1530 /* 1531 * T/TCP mechanism 1532 * If T/TCP was negotiated and the segment doesn't have CC, 1533 * or if its CC is wrong then drop the segment. 1534 * RST segments do not have to comply with this. 1535 */ 1536 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1537 (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc)) 1538 goto dropafterack; 1539 1540 /* 1541 * In the SYN-RECEIVED state, validate that the packet belongs to 1542 * this connection before trimming the data to fit the receive 1543 * window. Check the sequence number versus IRS since we know 1544 * the sequence numbers haven't wrapped. This is a partial fix 1545 * for the "LAND" DoS attack. 1546 */ 1547 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1548 rstreason = BANDLIM_RST_OPENPORT; 1549 goto dropwithreset; 1550 } 1551 1552 todrop = tp->rcv_nxt - th->th_seq; 1553 if (todrop > 0) { 1554 if (thflags & TH_SYN) { 1555 thflags &= ~TH_SYN; 1556 th->th_seq++; 1557 if (th->th_urp > 1) 1558 th->th_urp--; 1559 else 1560 thflags &= ~TH_URG; 1561 todrop--; 1562 } 1563 /* 1564 * Following if statement from Stevens, vol. 2, p. 960. 1565 */ 1566 if (todrop > tlen || 1567 (todrop == tlen && !(thflags & TH_FIN))) { 1568 /* 1569 * Any valid FIN must be to the left of the window. 1570 * At this point the FIN must be a duplicate or out 1571 * of sequence; drop it. 1572 */ 1573 thflags &= ~TH_FIN; 1574 1575 /* 1576 * Send an ACK to resynchronize and drop any data. 1577 * But keep on processing for RST or ACK. 1578 */ 1579 tp->t_flags |= TF_ACKNOW; 1580 todrop = tlen; 1581 tcpstat.tcps_rcvduppack++; 1582 tcpstat.tcps_rcvdupbyte += todrop; 1583 } else { 1584 tcpstat.tcps_rcvpartduppack++; 1585 tcpstat.tcps_rcvpartdupbyte += todrop; 1586 } 1587 drop_hdrlen += todrop; /* drop from the top afterwards */ 1588 th->th_seq += todrop; 1589 tlen -= todrop; 1590 if (th->th_urp > todrop) 1591 th->th_urp -= todrop; 1592 else { 1593 thflags &= ~TH_URG; 1594 th->th_urp = 0; 1595 } 1596 } 1597 1598 /* 1599 * If new data are received on a connection after the 1600 * user processes are gone, then RST the other end. 1601 */ 1602 if ((so->so_state & SS_NOFDREF) && 1603 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1604 tp = tcp_close(tp); 1605 tcpstat.tcps_rcvafterclose++; 1606 rstreason = BANDLIM_UNLIMITED; 1607 goto dropwithreset; 1608 } 1609 1610 /* 1611 * If segment ends after window, drop trailing data 1612 * (and PUSH and FIN); if nothing left, just ACK. 1613 */ 1614 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1615 if (todrop > 0) { 1616 tcpstat.tcps_rcvpackafterwin++; 1617 if (todrop >= tlen) { 1618 tcpstat.tcps_rcvbyteafterwin += tlen; 1619 /* 1620 * If a new connection request is received 1621 * while in TIME_WAIT, drop the old connection 1622 * and start over if the sequence numbers 1623 * are above the previous ones. 1624 */ 1625 if (thflags & TH_SYN && 1626 tp->t_state == TCPS_TIME_WAIT && 1627 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1628 tp = tcp_close(tp); 1629 goto findpcb; 1630 } 1631 /* 1632 * If window is closed can only take segments at 1633 * window edge, and have to drop data and PUSH from 1634 * incoming segments. Continue processing, but 1635 * remember to ack. Otherwise, drop segment 1636 * and ack. 1637 */ 1638 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1639 tp->t_flags |= TF_ACKNOW; 1640 tcpstat.tcps_rcvwinprobe++; 1641 } else 1642 goto dropafterack; 1643 } else 1644 tcpstat.tcps_rcvbyteafterwin += todrop; 1645 m_adj(m, -todrop); 1646 tlen -= todrop; 1647 thflags &= ~(TH_PUSH | TH_FIN); 1648 } 1649 1650 /* 1651 * If last ACK falls within this segment's sequence numbers, 1652 * record its timestamp. 1653 * NOTE that the test is modified according to the latest 1654 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1655 */ 1656 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1657 tp->ts_recent_age = ticks; 1658 tp->ts_recent = to.to_tsval; 1659 } 1660 1661 /* 1662 * If a SYN is in the window, then this is an 1663 * error and we send an RST and drop the connection. 1664 */ 1665 if (thflags & TH_SYN) { 1666 tp = tcp_drop(tp, ECONNRESET); 1667 rstreason = BANDLIM_UNLIMITED; 1668 goto dropwithreset; 1669 } 1670 1671 /* 1672 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1673 * flag is on (half-synchronized state), then queue data for 1674 * later processing; else drop segment and return. 1675 */ 1676 if (!(thflags & TH_ACK)) { 1677 if (tp->t_state == TCPS_SYN_RECEIVED || 1678 (tp->t_flags & TF_NEEDSYN)) 1679 goto step6; 1680 else 1681 goto drop; 1682 } 1683 1684 /* 1685 * Ack processing. 1686 */ 1687 switch (tp->t_state) { 1688 /* 1689 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1690 * ESTABLISHED state and continue processing. 1691 * The ACK was checked above. 1692 */ 1693 case TCPS_SYN_RECEIVED: 1694 1695 tcpstat.tcps_connects++; 1696 soisconnected(so); 1697 /* Do window scaling? */ 1698 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1699 (TF_RCVD_SCALE | TF_REQ_SCALE)) { 1700 tp->snd_scale = tp->requested_s_scale; 1701 tp->rcv_scale = tp->request_r_scale; 1702 } 1703 /* 1704 * Upon successful completion of 3-way handshake, 1705 * update cache.CC if it was undefined, pass any queued 1706 * data to the user, and advance state appropriately. 1707 */ 1708 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL && 1709 taop->tao_cc == 0) 1710 taop->tao_cc = tp->cc_recv; 1711 1712 /* 1713 * Make transitions: 1714 * SYN-RECEIVED -> ESTABLISHED 1715 * SYN-RECEIVED* -> FIN-WAIT-1 1716 */ 1717 tp->t_starttime = ticks; 1718 if (tp->t_flags & TF_NEEDFIN) { 1719 tp->t_state = TCPS_FIN_WAIT_1; 1720 tp->t_flags &= ~TF_NEEDFIN; 1721 } else { 1722 tp->t_state = TCPS_ESTABLISHED; 1723 callout_reset(tp->tt_keep, tcp_keepidle, 1724 tcp_timer_keep, tp); 1725 } 1726 /* 1727 * If segment contains data or ACK, will call tcp_reass() 1728 * later; if not, do so now to pass queued data to user. 1729 */ 1730 if (tlen == 0 && !(thflags & TH_FIN)) 1731 (void) tcp_reass(tp, NULL, NULL, NULL); 1732 /* fall into ... */ 1733 1734 /* 1735 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1736 * ACKs. If the ack is in the range 1737 * tp->snd_una < th->th_ack <= tp->snd_max 1738 * then advance tp->snd_una to th->th_ack and drop 1739 * data from the retransmission queue. If this ACK reflects 1740 * more up to date window information we update our window information. 1741 */ 1742 case TCPS_ESTABLISHED: 1743 case TCPS_FIN_WAIT_1: 1744 case TCPS_FIN_WAIT_2: 1745 case TCPS_CLOSE_WAIT: 1746 case TCPS_CLOSING: 1747 case TCPS_LAST_ACK: 1748 case TCPS_TIME_WAIT: 1749 1750 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1751 if (tlen != 0 || tiwin != tp->snd_wnd) { 1752 tp->t_dupacks = 0; 1753 break; 1754 } 1755 tcpstat.tcps_rcvdupack++; 1756 /* 1757 * If we have outstanding data (other than 1758 * a window probe), this is a completely 1759 * duplicate ack (ie, window info didn't 1760 * change), the ack is the biggest we've 1761 * seen and we've seen exactly our rexmt 1762 * threshhold of them, assume a packet 1763 * has been dropped and retransmit it. 1764 * Kludge snd_nxt & the congestion 1765 * window so we send only this one 1766 * packet. 1767 * 1768 * We know we're losing at the current 1769 * window size so do congestion avoidance 1770 * (set ssthresh to half the current window 1771 * and pull our congestion window back to 1772 * the new ssthresh). 1773 * 1774 * Dup acks mean that packets have left the 1775 * network (they're now cached at the receiver) 1776 * so bump cwnd by the amount in the receiver 1777 * to keep a constant cwnd packets in the 1778 * network. 1779 */ 1780 if (!callout_active(tp->tt_rexmt) || 1781 th->th_ack != tp->snd_una) { 1782 tp->t_dupacks = 0; 1783 break; 1784 } 1785 if (IN_FASTRECOVERY(tp)) { 1786 tp->snd_cwnd += tp->t_maxseg; 1787 (void) tcp_output(tp); 1788 } else if (++tp->t_dupacks == tcprexmtthresh) { 1789 tcp_seq onxt; 1790 u_int win; 1791 1792 if (SEQ_LEQ(th->th_ack, tp->snd_recover)) { 1793 tp->t_dupacks = 0; 1794 break; 1795 } 1796 fastretransmit: 1797 if (tcp_do_eifel_detect && 1798 (tp->t_flags & TF_RCVD_TSTMP)) { 1799 tcp_save_congestion_state(tp); 1800 tp->t_flags |= TF_FASTREXMT; 1801 } 1802 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / 1803 tp->t_maxseg; 1804 if (win < 2) 1805 win = 2; 1806 tp->snd_ssthresh = win * tp->t_maxseg; 1807 ENTER_FASTRECOVERY(tp); 1808 tp->snd_recover = tp->snd_max; 1809 callout_stop(tp->tt_rexmt); 1810 tp->t_rtttime = 0; 1811 onxt = tp->snd_nxt; 1812 tp->snd_nxt = th->th_ack; 1813 tp->snd_cwnd = tp->t_maxseg; 1814 (void) tcp_output(tp); 1815 ++tcpstat.tcps_sndfastrexmit; 1816 KASSERT(tp->snd_limited <= 2, 1817 ("tp->snd_limited too big")); 1818 tp->snd_cwnd = tp->snd_ssthresh + 1819 (tp->t_maxseg * 1820 (tp->t_dupacks - tp->snd_limited)); 1821 if (SEQ_GT(onxt, tp->snd_nxt)) 1822 tp->snd_nxt = onxt; 1823 } else if (tcp_do_limitedtransmit) { 1824 u_long oldcwnd = tp->snd_cwnd; 1825 tcp_seq oldsndmax = tp->snd_max; 1826 /* outstanding data */ 1827 uint32_t ownd = tp->snd_max - tp->snd_una; 1828 u_int sent; 1829 1830 #define iceildiv(n, d) (((n)+(d)-1) / (d)) 1831 1832 KASSERT(tp->t_dupacks == 1 || 1833 tp->t_dupacks == 2, 1834 ("dupacks not 1 or 2")); 1835 if (tp->t_dupacks == 1) 1836 tp->snd_limited = 0; 1837 tp->snd_cwnd = ownd + 1838 (tp->t_dupacks - tp->snd_limited) * 1839 tp->t_maxseg; 1840 (void) tcp_output(tp); 1841 tp->snd_cwnd = oldcwnd; 1842 sent = tp->snd_max - oldsndmax; 1843 if (sent > tp->t_maxseg) { 1844 KASSERT((tp->t_dupacks == 2 && 1845 tp->snd_limited == 0) || 1846 (sent == tp->t_maxseg + 1 && 1847 tp->t_flags & TF_SENTFIN), 1848 ("sent too much")); 1849 KASSERT(sent <= tp->t_maxseg * 2, 1850 ("sent too many segments")); 1851 tp->snd_limited = 2; 1852 tcpstat.tcps_sndlimited += 2; 1853 } else if (sent > 0) { 1854 ++tp->snd_limited; 1855 ++tcpstat.tcps_sndlimited; 1856 } else if (tcp_do_early_retransmit && 1857 (tcp_do_eifel_detect && 1858 (tp->t_flags & TF_RCVD_TSTMP)) && 1859 tp->t_dupacks + 1 >= 1860 iceildiv(ownd, tp->t_maxseg)) { 1861 ++tcpstat.tcps_sndearlyrexmit; 1862 tp->t_flags |= TF_EARLYREXMT; 1863 goto fastretransmit; 1864 } 1865 } 1866 goto drop; 1867 } 1868 1869 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 1870 1871 /* 1872 * If the congestion window was inflated to account 1873 * for the other side's cached packets, retract it. 1874 */ 1875 if (IN_FASTRECOVERY(tp)) { 1876 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1877 tcp_newreno_partial_ack(tp, th); 1878 } else { 1879 /* 1880 * Window inflation should have left us 1881 * with approximately snd_ssthresh 1882 * outstanding data. 1883 * But in case we would be inclined to 1884 * send a burst, better to do it via 1885 * the slow start mechanism. 1886 */ 1887 if (SEQ_GT(th->th_ack + tp->snd_ssthresh, 1888 tp->snd_max)) 1889 tp->snd_cwnd = tp->snd_max - 1890 th->th_ack + tp->t_maxseg; 1891 else 1892 tp->snd_cwnd = tp->snd_ssthresh; 1893 } 1894 } 1895 tp->t_dupacks = 0; 1896 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1897 /* 1898 * Detected optimistic ACK attack. 1899 * Force slow-start to de-synchronize attack. 1900 */ 1901 tp->snd_cwnd = tp->t_maxseg; 1902 1903 tcpstat.tcps_rcvacktoomuch++; 1904 goto dropafterack; 1905 } 1906 /* 1907 * If we reach this point, ACK is not a duplicate, 1908 * i.e., it ACKs something we sent. 1909 */ 1910 if (tp->t_flags & TF_NEEDSYN) { 1911 /* 1912 * T/TCP: Connection was half-synchronized, and our 1913 * SYN has been ACK'd (so connection is now fully 1914 * synchronized). Go to non-starred state, 1915 * increment snd_una for ACK of SYN, and check if 1916 * we can do window scaling. 1917 */ 1918 tp->t_flags &= ~TF_NEEDSYN; 1919 tp->snd_una++; 1920 /* Do window scaling? */ 1921 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) == 1922 (TF_RCVD_SCALE | TF_REQ_SCALE)) { 1923 tp->snd_scale = tp->requested_s_scale; 1924 tp->rcv_scale = tp->request_r_scale; 1925 } 1926 } 1927 1928 process_ACK: 1929 acked = th->th_ack - tp->snd_una; 1930 tcpstat.tcps_rcvackpack++; 1931 tcpstat.tcps_rcvackbyte += acked; 1932 1933 if (tcp_do_eifel_detect && acked > 0 && 1934 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) && 1935 (tp->t_flags & TF_FIRSTACCACK)) { 1936 /* Eifel detection applicable. */ 1937 if (to.to_tsecr < tp->t_rexmtTS) { 1938 ++tcpstat.tcps_eifeldetected; 1939 tcp_revert_congestion_state(tp); 1940 if (tp->t_rxtshift == 1 && 1941 ticks >= tp->t_badrxtwin) 1942 ++tcpstat.tcps_rttcantdetect; 1943 } 1944 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1945 /* 1946 * If we just performed our first retransmit, 1947 * and the ACK arrives within our recovery window, 1948 * then it was a mistake to do the retransmit 1949 * in the first place. Recover our original cwnd 1950 * and ssthresh, and proceed to transmit where we 1951 * left off. 1952 */ 1953 tcp_revert_congestion_state(tp); 1954 ++tcpstat.tcps_rttdetected; 1955 } 1956 1957 /* 1958 * If we have a timestamp reply, update smoothed 1959 * round trip time. If no timestamp is present but 1960 * transmit timer is running and timed sequence 1961 * number was acked, update smoothed round trip time. 1962 * Since we now have an rtt measurement, cancel the 1963 * timer backoff (cf., Phil Karn's retransmit alg.). 1964 * Recompute the initial retransmit timer. 1965 * 1966 * Some machines (certain windows boxes) send broken 1967 * timestamp replies during the SYN+ACK phase, ignore 1968 * timestamps of 0. 1969 */ 1970 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) 1971 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1972 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) 1973 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1974 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1975 1976 /* 1977 * If all outstanding data is acked, stop retransmit 1978 * timer and remember to restart (more output or persist). 1979 * If there is more data to be acked, restart retransmit 1980 * timer, using current (possibly backed-off) value. 1981 */ 1982 if (th->th_ack == tp->snd_max) { 1983 callout_stop(tp->tt_rexmt); 1984 needoutput = TRUE; 1985 } else if (!callout_active(tp->tt_persist)) 1986 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 1987 tcp_timer_rexmt, tp); 1988 1989 /* 1990 * If no data (only SYN) was ACK'd, 1991 * skip rest of ACK processing. 1992 */ 1993 if (acked == 0) 1994 goto step6; 1995 1996 /* Stop looking for an acceptable ACK since one was received. */ 1997 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT); 1998 1999 /* 2000 * When new data is acked, open the congestion window. 2001 * If the window gives us less than ssthresh packets 2002 * in flight, open exponentially (maxseg per packet). 2003 * Otherwise open linearly: maxseg per window 2004 * (maxseg^2 / cwnd per packet). 2005 */ 2006 if (!IN_FASTRECOVERY(tp)) { 2007 u_int cw = tp->snd_cwnd; 2008 u_int incr = tp->t_maxseg; 2009 2010 if (cw > tp->snd_ssthresh) 2011 incr = incr * incr / cw; 2012 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2013 } 2014 if (acked > so->so_snd.sb_cc) { 2015 tp->snd_wnd -= so->so_snd.sb_cc; 2016 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc); 2017 ourfinisacked = TRUE; 2018 } else { 2019 sbdrop(&so->so_snd, acked); 2020 tp->snd_wnd -= acked; 2021 ourfinisacked = FALSE; 2022 } 2023 sowwakeup(so); 2024 if (IN_FASTRECOVERY(tp)) { 2025 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) 2026 EXIT_FASTRECOVERY(tp); 2027 } else { 2028 tp->snd_recover = th->th_ack - 1; 2029 } 2030 tp->snd_una = th->th_ack; 2031 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2032 tp->snd_nxt = tp->snd_una; 2033 2034 switch (tp->t_state) { 2035 /* 2036 * In FIN_WAIT_1 STATE in addition to the processing 2037 * for the ESTABLISHED state if our FIN is now acknowledged 2038 * then enter FIN_WAIT_2. 2039 */ 2040 case TCPS_FIN_WAIT_1: 2041 if (ourfinisacked) { 2042 /* 2043 * If we can't receive any more 2044 * data, then closing user can proceed. 2045 * Starting the timer is contrary to the 2046 * specification, but if we don't get a FIN 2047 * we'll hang forever. 2048 */ 2049 if (so->so_state & SS_CANTRCVMORE) { 2050 soisdisconnected(so); 2051 callout_reset(tp->tt_2msl, tcp_maxidle, 2052 tcp_timer_2msl, tp); 2053 } 2054 tp->t_state = TCPS_FIN_WAIT_2; 2055 } 2056 break; 2057 2058 /* 2059 * In CLOSING STATE in addition to the processing for 2060 * the ESTABLISHED state if the ACK acknowledges our FIN 2061 * then enter the TIME-WAIT state, otherwise ignore 2062 * the segment. 2063 */ 2064 case TCPS_CLOSING: 2065 if (ourfinisacked) { 2066 tp->t_state = TCPS_TIME_WAIT; 2067 tcp_canceltimers(tp); 2068 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2069 if (tp->cc_recv != 0 && 2070 (ticks - tp->t_starttime) < tcp_msl) 2071 callout_reset(tp->tt_2msl, 2072 tp->t_rxtcur * TCPTV_TWTRUNC, 2073 tcp_timer_2msl, tp); 2074 else 2075 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2076 tcp_timer_2msl, tp); 2077 soisdisconnected(so); 2078 } 2079 break; 2080 2081 /* 2082 * In LAST_ACK, we may still be waiting for data to drain 2083 * and/or to be acked, as well as for the ack of our FIN. 2084 * If our FIN is now acknowledged, delete the TCB, 2085 * enter the closed state and return. 2086 */ 2087 case TCPS_LAST_ACK: 2088 if (ourfinisacked) { 2089 tp = tcp_close(tp); 2090 goto drop; 2091 } 2092 break; 2093 2094 /* 2095 * In TIME_WAIT state the only thing that should arrive 2096 * is a retransmission of the remote FIN. Acknowledge 2097 * it and restart the finack timer. 2098 */ 2099 case TCPS_TIME_WAIT: 2100 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2101 tcp_timer_2msl, tp); 2102 goto dropafterack; 2103 } 2104 } 2105 2106 step6: 2107 /* 2108 * Update window information. 2109 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2110 */ 2111 if ((thflags & TH_ACK) && 2112 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2113 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2114 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2115 /* keep track of pure window updates */ 2116 if (tlen == 0 && tp->snd_wl2 == th->th_ack && 2117 tiwin > tp->snd_wnd) 2118 tcpstat.tcps_rcvwinupd++; 2119 tp->snd_wnd = tiwin; 2120 tp->snd_wl1 = th->th_seq; 2121 tp->snd_wl2 = th->th_ack; 2122 if (tp->snd_wnd > tp->max_sndwnd) 2123 tp->max_sndwnd = tp->snd_wnd; 2124 needoutput = TRUE; 2125 } 2126 2127 /* 2128 * Process segments with URG. 2129 */ 2130 if ((thflags & TH_URG) && th->th_urp && 2131 !TCPS_HAVERCVDFIN(tp->t_state)) { 2132 /* 2133 * This is a kludge, but if we receive and accept 2134 * random urgent pointers, we'll crash in 2135 * soreceive. It's hard to imagine someone 2136 * actually wanting to send this much urgent data. 2137 */ 2138 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2139 th->th_urp = 0; /* XXX */ 2140 thflags &= ~TH_URG; /* XXX */ 2141 goto dodata; /* XXX */ 2142 } 2143 /* 2144 * If this segment advances the known urgent pointer, 2145 * then mark the data stream. This should not happen 2146 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2147 * a FIN has been received from the remote side. 2148 * In these states we ignore the URG. 2149 * 2150 * According to RFC961 (Assigned Protocols), 2151 * the urgent pointer points to the last octet 2152 * of urgent data. We continue, however, 2153 * to consider it to indicate the first octet 2154 * of data past the urgent section as the original 2155 * spec states (in one of two places). 2156 */ 2157 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) { 2158 tp->rcv_up = th->th_seq + th->th_urp; 2159 so->so_oobmark = so->so_rcv.sb_cc + 2160 (tp->rcv_up - tp->rcv_nxt) - 1; 2161 if (so->so_oobmark == 0) 2162 so->so_state |= SS_RCVATMARK; 2163 sohasoutofband(so); 2164 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2165 } 2166 /* 2167 * Remove out of band data so doesn't get presented to user. 2168 * This can happen independent of advancing the URG pointer, 2169 * but if two URG's are pending at once, some out-of-band 2170 * data may creep in... ick. 2171 */ 2172 if (th->th_urp <= (u_long)tlen && 2173 !(so->so_options & SO_OOBINLINE)) { 2174 /* hdr drop is delayed */ 2175 tcp_pulloutofband(so, th, m, drop_hdrlen); 2176 } 2177 } else { 2178 /* 2179 * If no out of band data is expected, 2180 * pull receive urgent pointer along 2181 * with the receive window. 2182 */ 2183 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2184 tp->rcv_up = tp->rcv_nxt; 2185 } 2186 2187 dodata: /* XXX */ 2188 /* 2189 * Process the segment text, merging it into the TCP sequencing queue, 2190 * and arranging for acknowledgment of receipt if necessary. 2191 * This process logically involves adjusting tp->rcv_wnd as data 2192 * is presented to the user (this happens in tcp_usrreq.c, 2193 * case PRU_RCVD). If a FIN has already been received on this 2194 * connection then we just ignore the text. 2195 */ 2196 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) { 2197 m_adj(m, drop_hdrlen); /* delayed header drop */ 2198 /* 2199 * Insert segment which includes th into TCP reassembly queue 2200 * with control block tp. Set thflags to whether reassembly now 2201 * includes a segment with FIN. This handles the common case 2202 * inline (segment is the next to be received on an established 2203 * connection, and the queue is empty), avoiding linkage into 2204 * and removal from the queue and repetition of various 2205 * conversions. 2206 * Set DELACK for segments received in order, but ack 2207 * immediately when segments are out of order (so 2208 * fast retransmit can work). 2209 */ 2210 if (th->th_seq == tp->rcv_nxt && 2211 LIST_EMPTY(&tp->t_segq) && 2212 TCPS_HAVEESTABLISHED(tp->t_state)) { 2213 if (DELAY_ACK(tp)) 2214 callout_reset(tp->tt_delack, tcp_delacktime, 2215 tcp_timer_delack, tp); 2216 else 2217 tp->t_flags |= TF_ACKNOW; 2218 tp->rcv_nxt += tlen; 2219 thflags = th->th_flags & TH_FIN; 2220 tcpstat.tcps_rcvpack++; 2221 tcpstat.tcps_rcvbyte += tlen; 2222 ND6_HINT(tp); 2223 if (so->so_state & SS_CANTRCVMORE) 2224 m_freem(m); 2225 else 2226 sbappend(&so->so_rcv, m); 2227 sorwakeup(so); 2228 } else { 2229 thflags = tcp_reass(tp, th, &tlen, m); 2230 tp->t_flags |= TF_ACKNOW; 2231 } 2232 2233 /* 2234 * Note the amount of data that peer has sent into 2235 * our window, in order to estimate the sender's 2236 * buffer size. 2237 */ 2238 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2239 } else { 2240 m_freem(m); 2241 thflags &= ~TH_FIN; 2242 } 2243 2244 /* 2245 * If FIN is received ACK the FIN and let the user know 2246 * that the connection is closing. 2247 */ 2248 if (thflags & TH_FIN) { 2249 if (!TCPS_HAVERCVDFIN(tp->t_state)) { 2250 socantrcvmore(so); 2251 /* 2252 * If connection is half-synchronized 2253 * (ie NEEDSYN flag on) then delay ACK, 2254 * so it may be piggybacked when SYN is sent. 2255 * Otherwise, since we received a FIN then no 2256 * more input can be expected, send ACK now. 2257 */ 2258 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) 2259 callout_reset(tp->tt_delack, tcp_delacktime, 2260 tcp_timer_delack, tp); 2261 else 2262 tp->t_flags |= TF_ACKNOW; 2263 tp->rcv_nxt++; 2264 } 2265 2266 switch (tp->t_state) { 2267 /* 2268 * In SYN_RECEIVED and ESTABLISHED STATES 2269 * enter the CLOSE_WAIT state. 2270 */ 2271 case TCPS_SYN_RECEIVED: 2272 tp->t_starttime = ticks; 2273 /*FALLTHROUGH*/ 2274 case TCPS_ESTABLISHED: 2275 tp->t_state = TCPS_CLOSE_WAIT; 2276 break; 2277 2278 /* 2279 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2280 * enter the CLOSING state. 2281 */ 2282 case TCPS_FIN_WAIT_1: 2283 tp->t_state = TCPS_CLOSING; 2284 break; 2285 2286 /* 2287 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2288 * starting the time-wait timer, turning off the other 2289 * standard timers. 2290 */ 2291 case TCPS_FIN_WAIT_2: 2292 tp->t_state = TCPS_TIME_WAIT; 2293 tcp_canceltimers(tp); 2294 /* Shorten TIME_WAIT [RFC-1644, p.28] */ 2295 if (tp->cc_recv != 0 && 2296 (ticks - tp->t_starttime) < tcp_msl) { 2297 callout_reset(tp->tt_2msl, 2298 tp->t_rxtcur * TCPTV_TWTRUNC, 2299 tcp_timer_2msl, tp); 2300 /* For transaction client, force ACK now. */ 2301 tp->t_flags |= TF_ACKNOW; 2302 } 2303 else 2304 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2305 tcp_timer_2msl, tp); 2306 soisdisconnected(so); 2307 break; 2308 2309 /* 2310 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2311 */ 2312 case TCPS_TIME_WAIT: 2313 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2314 tcp_timer_2msl, tp); 2315 break; 2316 } 2317 } 2318 2319 #ifdef TCPDEBUG 2320 if (so->so_options & SO_DEBUG) 2321 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2322 &tcp_savetcp, 0); 2323 #endif 2324 2325 /* 2326 * Return any desired output. 2327 */ 2328 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2329 (void) tcp_output(tp); 2330 return; 2331 2332 dropafterack: 2333 /* 2334 * Generate an ACK dropping incoming segment if it occupies 2335 * sequence space, where the ACK reflects our state. 2336 * 2337 * We can now skip the test for the RST flag since all 2338 * paths to this code happen after packets containing 2339 * RST have been dropped. 2340 * 2341 * In the SYN-RECEIVED state, don't send an ACK unless the 2342 * segment we received passes the SYN-RECEIVED ACK test. 2343 * If it fails send a RST. This breaks the loop in the 2344 * "LAND" DoS attack, and also prevents an ACK storm 2345 * between two listening ports that have been sent forged 2346 * SYN segments, each with the source address of the other. 2347 */ 2348 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2349 (SEQ_GT(tp->snd_una, th->th_ack) || 2350 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2351 rstreason = BANDLIM_RST_OPENPORT; 2352 goto dropwithreset; 2353 } 2354 #ifdef TCPDEBUG 2355 if (so->so_options & SO_DEBUG) 2356 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2357 &tcp_savetcp, 0); 2358 #endif 2359 m_freem(m); 2360 tp->t_flags |= TF_ACKNOW; 2361 (void) tcp_output(tp); 2362 return; 2363 2364 dropwithreset: 2365 /* 2366 * Generate a RST, dropping incoming segment. 2367 * Make ACK acceptable to originator of segment. 2368 * Don't bother to respond if destination was broadcast/multicast. 2369 */ 2370 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST)) 2371 goto drop; 2372 if (isipv6) { 2373 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2374 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2375 goto drop; 2376 } else { 2377 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2378 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2379 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2380 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2381 goto drop; 2382 } 2383 /* IPv6 anycast check is done at tcp6_input() */ 2384 2385 /* 2386 * Perform bandwidth limiting. 2387 */ 2388 #ifdef ICMP_BANDLIM 2389 if (badport_bandlim(rstreason) < 0) 2390 goto drop; 2391 #endif 2392 2393 #ifdef TCPDEBUG 2394 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2395 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2396 &tcp_savetcp, 0); 2397 #endif 2398 if (thflags & TH_ACK) 2399 /* mtod() below is safe as long as hdr dropping is delayed */ 2400 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2401 TH_RST); 2402 else { 2403 if (thflags & TH_SYN) 2404 tlen++; 2405 /* mtod() below is safe as long as hdr dropping is delayed */ 2406 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen, 2407 (tcp_seq)0, TH_RST | TH_ACK); 2408 } 2409 return; 2410 2411 drop: 2412 /* 2413 * Drop space held by incoming segment and return. 2414 */ 2415 #ifdef TCPDEBUG 2416 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2417 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2418 &tcp_savetcp, 0); 2419 #endif 2420 m_freem(m); 2421 return; 2422 } 2423 2424 /* 2425 * Parse TCP options and place in tcpopt. 2426 */ 2427 static void 2428 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn) 2429 { 2430 int opt, optlen; 2431 2432 to->to_flags = 0; 2433 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2434 opt = cp[0]; 2435 if (opt == TCPOPT_EOL) 2436 break; 2437 if (opt == TCPOPT_NOP) 2438 optlen = 1; 2439 else { 2440 if (cnt < 2) 2441 break; 2442 optlen = cp[1]; 2443 if (optlen < 2 || optlen > cnt) 2444 break; 2445 } 2446 switch (opt) { 2447 case TCPOPT_MAXSEG: 2448 if (optlen != TCPOLEN_MAXSEG) 2449 continue; 2450 if (!is_syn) 2451 continue; 2452 to->to_flags |= TOF_MSS; 2453 bcopy(cp + 2, &to->to_mss, sizeof(to->to_mss)); 2454 to->to_mss = ntohs(to->to_mss); 2455 break; 2456 case TCPOPT_WINDOW: 2457 if (optlen != TCPOLEN_WINDOW) 2458 continue; 2459 if (!is_syn) 2460 continue; 2461 to->to_flags |= TOF_SCALE; 2462 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2463 break; 2464 case TCPOPT_TIMESTAMP: 2465 if (optlen != TCPOLEN_TIMESTAMP) 2466 continue; 2467 to->to_flags |= TOF_TS; 2468 bcopy(cp + 2, &to->to_tsval, sizeof(to->to_tsval)); 2469 to->to_tsval = ntohl(to->to_tsval); 2470 bcopy(cp + 6, &to->to_tsecr, sizeof(to->to_tsecr)); 2471 to->to_tsecr = ntohl(to->to_tsecr); 2472 break; 2473 case TCPOPT_CC: 2474 if (optlen != TCPOLEN_CC) 2475 continue; 2476 to->to_flags |= TOF_CC; 2477 bcopy(cp + 2, &to->to_cc, sizeof(to->to_cc)); 2478 to->to_cc = ntohl(to->to_cc); 2479 break; 2480 case TCPOPT_CCNEW: 2481 if (optlen != TCPOLEN_CC) 2482 continue; 2483 if (!is_syn) 2484 continue; 2485 to->to_flags |= TOF_CCNEW; 2486 bcopy(cp + 2, &to->to_cc, sizeof(to->to_cc)); 2487 to->to_cc = ntohl(to->to_cc); 2488 break; 2489 case TCPOPT_CCECHO: 2490 if (optlen != TCPOLEN_CC) 2491 continue; 2492 if (!is_syn) 2493 continue; 2494 to->to_flags |= TOF_CCECHO; 2495 bcopy(cp + 2, &to->to_ccecho, sizeof(to->to_ccecho)); 2496 to->to_ccecho = ntohl(to->to_ccecho); 2497 break; 2498 default: 2499 continue; 2500 } 2501 } 2502 } 2503 2504 /* 2505 * Pull out of band byte out of a segment so 2506 * it doesn't appear in the user's data queue. 2507 * It is still reflected in the segment length for 2508 * sequencing purposes. 2509 * "off" is the delayed to be dropped hdrlen. 2510 */ 2511 static void 2512 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) 2513 { 2514 int cnt = off + th->th_urp - 1; 2515 2516 while (cnt >= 0) { 2517 if (m->m_len > cnt) { 2518 char *cp = mtod(m, caddr_t) + cnt; 2519 struct tcpcb *tp = sototcpcb(so); 2520 2521 tp->t_iobc = *cp; 2522 tp->t_oobflags |= TCPOOB_HAVEDATA; 2523 bcopy(cp + 1, cp, m->m_len - cnt - 1); 2524 m->m_len--; 2525 if (m->m_flags & M_PKTHDR) 2526 m->m_pkthdr.len--; 2527 return; 2528 } 2529 cnt -= m->m_len; 2530 m = m->m_next; 2531 if (m == 0) 2532 break; 2533 } 2534 panic("tcp_pulloutofband"); 2535 } 2536 2537 /* 2538 * Collect new round-trip time estimate 2539 * and update averages and current timeout. 2540 */ 2541 static void 2542 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2543 { 2544 int delta; 2545 2546 tcpstat.tcps_rttupdated++; 2547 tp->t_rttupdated++; 2548 if (tp->t_srtt != 0) { 2549 /* 2550 * srtt is stored as fixed point with 5 bits after the 2551 * binary point (i.e., scaled by 8). The following magic 2552 * is equivalent to the smoothing algorithm in rfc793 with 2553 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2554 * point). Adjust rtt to origin 0. 2555 */ 2556 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2557 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2558 2559 if ((tp->t_srtt += delta) <= 0) 2560 tp->t_srtt = 1; 2561 2562 /* 2563 * We accumulate a smoothed rtt variance (actually, a 2564 * smoothed mean difference), then set the retransmit 2565 * timer to smoothed rtt + 4 times the smoothed variance. 2566 * rttvar is stored as fixed point with 4 bits after the 2567 * binary point (scaled by 16). The following is 2568 * equivalent to rfc793 smoothing with an alpha of .75 2569 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2570 * rfc793's wired-in beta. 2571 */ 2572 if (delta < 0) 2573 delta = -delta; 2574 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2575 if ((tp->t_rttvar += delta) <= 0) 2576 tp->t_rttvar = 1; 2577 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2578 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2579 } else { 2580 /* 2581 * No rtt measurement yet - use the unsmoothed rtt. 2582 * Set the variance to half the rtt (so our first 2583 * retransmit happens at 3*rtt). 2584 */ 2585 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2586 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2587 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2588 } 2589 tp->t_rtttime = 0; 2590 tp->t_rxtshift = 0; 2591 2592 /* 2593 * the retransmit should happen at rtt + 4 * rttvar. 2594 * Because of the way we do the smoothing, srtt and rttvar 2595 * will each average +1/2 tick of bias. When we compute 2596 * the retransmit timer, we want 1/2 tick of rounding and 2597 * 1 extra tick because of +-1/2 tick uncertainty in the 2598 * firing of the timer. The bias will give us exactly the 2599 * 1.5 tick we need. But, because the bias is 2600 * statistical, we have to test that we don't drop below 2601 * the minimum feasible timer (which is 2 ticks). 2602 */ 2603 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2604 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2605 2606 /* 2607 * We received an ack for a packet that wasn't retransmitted; 2608 * it is probably safe to discard any error indications we've 2609 * received recently. This isn't quite right, but close enough 2610 * for now (a route might have failed after we sent a segment, 2611 * and the return path might not be symmetrical). 2612 */ 2613 tp->t_softerror = 0; 2614 } 2615 2616 /* 2617 * Determine a reasonable value for maxseg size. 2618 * If the route is known, check route for mtu. 2619 * If none, use an mss that can be handled on the outgoing 2620 * interface without forcing IP to fragment; if bigger than 2621 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2622 * to utilize large mbufs. If no route is found, route has no mtu, 2623 * or the destination isn't local, use a default, hopefully conservative 2624 * size (usually 512 or the default IP max size, but no more than the mtu 2625 * of the interface), as we can't discover anything about intervening 2626 * gateways or networks. We also initialize the congestion/slow start 2627 * window to be a single segment if the destination isn't local. 2628 * While looking at the routing entry, we also initialize other path-dependent 2629 * parameters from pre-set or cached values in the routing entry. 2630 * 2631 * Also take into account the space needed for options that we 2632 * send regularly. Make maxseg shorter by that amount to assure 2633 * that we can send maxseg amount of data even when the options 2634 * are present. Store the upper limit of the length of options plus 2635 * data in maxopd. 2636 * 2637 * NOTE that this routine is only called when we process an incoming 2638 * segment, for outgoing segments only tcp_mssopt is called. 2639 * 2640 * In case of T/TCP, we call this routine during implicit connection 2641 * setup as well (offer = -1), to initialize maxseg from the cached 2642 * MSS of our peer. 2643 */ 2644 void 2645 tcp_mss(struct tcpcb *tp, int offer) 2646 { 2647 struct rtentry *rt; 2648 struct ifnet *ifp; 2649 int rtt, mss; 2650 u_long bufsize; 2651 struct inpcb *inp = tp->t_inpcb; 2652 struct socket *so; 2653 struct rmxp_tao *taop; 2654 int origoffer = offer; 2655 #ifdef INET6 2656 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE); 2657 size_t min_protoh = isipv6 ? 2658 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2659 sizeof(struct tcpiphdr); 2660 #else 2661 const boolean_t isipv6 = FALSE; 2662 const size_t min_protoh = sizeof(struct tcpiphdr); 2663 #endif 2664 2665 if (isipv6) 2666 rt = tcp_rtlookup6(&inp->inp_inc); 2667 else 2668 rt = tcp_rtlookup(&inp->inp_inc); 2669 if (rt == NULL) { 2670 tp->t_maxopd = tp->t_maxseg = 2671 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2672 return; 2673 } 2674 ifp = rt->rt_ifp; 2675 so = inp->inp_socket; 2676 2677 taop = rmx_taop(rt->rt_rmx); 2678 /* 2679 * Offer == -1 means that we didn't receive SYN yet, 2680 * use cached value in that case; 2681 */ 2682 if (offer == -1) 2683 offer = taop->tao_mssopt; 2684 /* 2685 * Offer == 0 means that there was no MSS on the SYN segment, 2686 * in this case we use tcp_mssdflt. 2687 */ 2688 if (offer == 0) 2689 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2690 else 2691 /* 2692 * Sanity check: make sure that maxopd will be large 2693 * enough to allow some data on segments even is the 2694 * all the option space is used (40bytes). Otherwise 2695 * funny things may happen in tcp_output. 2696 */ 2697 offer = max(offer, 64); 2698 taop->tao_mssopt = offer; 2699 2700 /* 2701 * While we're here, check if there's an initial rtt 2702 * or rttvar. Convert from the route-table units 2703 * to scaled multiples of the slow timeout timer. 2704 */ 2705 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) { 2706 /* 2707 * XXX the lock bit for RTT indicates that the value 2708 * is also a minimum value; this is subject to time. 2709 */ 2710 if (rt->rt_rmx.rmx_locks & RTV_RTT) 2711 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz); 2712 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE)); 2713 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2714 tcpstat.tcps_usedrtt++; 2715 if (rt->rt_rmx.rmx_rttvar) { 2716 tp->t_rttvar = rt->rt_rmx.rmx_rttvar / 2717 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE)); 2718 tcpstat.tcps_usedrttvar++; 2719 } else { 2720 /* default variation is +- 1 rtt */ 2721 tp->t_rttvar = 2722 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2723 } 2724 TCPT_RANGESET(tp->t_rxtcur, 2725 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2726 tp->t_rttmin, TCPTV_REXMTMAX); 2727 } 2728 /* 2729 * if there's an mtu associated with the route, use it 2730 * else, use the link mtu. 2731 */ 2732 if (rt->rt_rmx.rmx_mtu) 2733 mss = rt->rt_rmx.rmx_mtu - min_protoh; 2734 else { 2735 if (isipv6) { 2736 mss = nd_ifinfo[rt->rt_ifp->if_index].linkmtu - 2737 min_protoh; 2738 if (!in6_localaddr(&inp->in6p_faddr)) 2739 mss = min(mss, tcp_v6mssdflt); 2740 } else { 2741 mss = ifp->if_mtu - min_protoh; 2742 if (!in_localaddr(inp->inp_faddr)) 2743 mss = min(mss, tcp_mssdflt); 2744 } 2745 } 2746 mss = min(mss, offer); 2747 /* 2748 * maxopd stores the maximum length of data AND options 2749 * in a segment; maxseg is the amount of data in a normal 2750 * segment. We need to store this value (maxopd) apart 2751 * from maxseg, because now every segment carries options 2752 * and thus we normally have somewhat less data in segments. 2753 */ 2754 tp->t_maxopd = mss; 2755 2756 /* 2757 * In case of T/TCP, origoffer==-1 indicates, that no segments 2758 * were received yet. In this case we just guess, otherwise 2759 * we do the same as before T/TCP. 2760 */ 2761 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP && 2762 (origoffer == -1 || 2763 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2764 mss -= TCPOLEN_TSTAMP_APPA; 2765 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC && 2766 (origoffer == -1 || 2767 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 2768 mss -= TCPOLEN_CC_APPA; 2769 2770 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2771 if (mss > MCLBYTES) 2772 mss &= ~(MCLBYTES-1); 2773 #else 2774 if (mss > MCLBYTES) 2775 mss = mss / MCLBYTES * MCLBYTES; 2776 #endif 2777 /* 2778 * If there's a pipesize, change the socket buffer 2779 * to that size. Make the socket buffers an integral 2780 * number of mss units; if the mss is larger than 2781 * the socket buffer, decrease the mss. 2782 */ 2783 #ifdef RTV_SPIPE 2784 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0) 2785 #endif 2786 bufsize = so->so_snd.sb_hiwat; 2787 if (bufsize < mss) 2788 mss = bufsize; 2789 else { 2790 bufsize = roundup(bufsize, mss); 2791 if (bufsize > sb_max) 2792 bufsize = sb_max; 2793 if (bufsize > so->so_snd.sb_hiwat) 2794 (void)sbreserve(&so->so_snd, bufsize, so, NULL); 2795 } 2796 tp->t_maxseg = mss; 2797 2798 #ifdef RTV_RPIPE 2799 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0) 2800 #endif 2801 bufsize = so->so_rcv.sb_hiwat; 2802 if (bufsize > mss) { 2803 bufsize = roundup(bufsize, mss); 2804 if (bufsize > sb_max) 2805 bufsize = sb_max; 2806 if (bufsize > so->so_rcv.sb_hiwat) 2807 (void)sbreserve(&so->so_rcv, bufsize, so, NULL); 2808 } 2809 2810 /* 2811 * Set the slow-start flight size depending on whether this 2812 * is a local network or not. 2813 */ 2814 if (tcp_do_rfc3390) 2815 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2816 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2817 (!isipv6 && in_localaddr(inp->inp_faddr))) 2818 tp->snd_cwnd = mss * ss_fltsz_local; 2819 else 2820 tp->snd_cwnd = mss * ss_fltsz; 2821 2822 if (rt->rt_rmx.rmx_ssthresh) { 2823 /* 2824 * There's some sort of gateway or interface 2825 * buffer limit on the path. Use this to set 2826 * the slow start threshhold, but set the 2827 * threshold to no less than 2*mss. 2828 */ 2829 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh); 2830 tcpstat.tcps_usedssthresh++; 2831 } 2832 } 2833 2834 /* 2835 * Determine the MSS option to send on an outgoing SYN. 2836 */ 2837 int 2838 tcp_mssopt(struct tcpcb *tp) 2839 { 2840 struct rtentry *rt; 2841 #ifdef INET6 2842 boolean_t isipv6 = 2843 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE); 2844 int min_protoh = isipv6 ? 2845 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) : 2846 sizeof(struct tcpiphdr); 2847 #else 2848 const boolean_t isipv6 = FALSE; 2849 const size_t min_protoh = sizeof(struct tcpiphdr); 2850 #endif 2851 2852 if (isipv6) 2853 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc); 2854 else 2855 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc); 2856 if (rt == NULL) 2857 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt); 2858 2859 return (rt->rt_ifp->if_mtu - min_protoh); 2860 } 2861 2862 /* 2863 * When a partial ack arrives, force the retransmission of the 2864 * next unacknowledged segment. Do not clear tp->t_dupacks. 2865 * By setting snd_nxt to th_ack, this forces retransmission timer to 2866 * be started again. 2867 */ 2868 static void 2869 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 2870 { 2871 tcp_seq onxt = tp->snd_nxt; 2872 u_long ocwnd = tp->snd_cwnd; 2873 2874 callout_stop(tp->tt_rexmt); 2875 tp->t_rtttime = 0; 2876 tp->snd_nxt = th->th_ack; 2877 /* 2878 * Set snd_cwnd to one segment beyond acknowledged offset 2879 * (tp->snd_una has not yet been updated when this function is called.) 2880 */ 2881 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 2882 tp->t_flags |= TF_ACKNOW; 2883 (void) tcp_output(tp); 2884 tp->snd_cwnd = ocwnd; 2885 if (SEQ_GT(onxt, tp->snd_nxt)) 2886 tp->snd_nxt = onxt; 2887 /* 2888 * Partial window deflation. Relies on fact that tp->snd_una 2889 * not updated yet. 2890 */ 2891 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 2892 } 2893