1 /* 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_ipfw.h" /* for ipfw_fwd */ 34 #include "opt_inet.h" 35 #include "opt_inet6.h" 36 #include "opt_ipsec.h" 37 #include "opt_mac.h" 38 #include "opt_tcpdebug.h" 39 #include "opt_tcp_input.h" 40 #include "opt_tcp_sack.h" 41 42 #include <sys/param.h> 43 #include <sys/kernel.h> 44 #include <sys/mac.h> 45 #include <sys/malloc.h> 46 #include <sys/mbuf.h> 47 #include <sys/proc.h> /* for proc0 declaration */ 48 #include <sys/protosw.h> 49 #include <sys/signalvar.h> 50 #include <sys/socket.h> 51 #include <sys/socketvar.h> 52 #include <sys/sysctl.h> 53 #include <sys/syslog.h> 54 #include <sys/systm.h> 55 56 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 57 58 #include <vm/uma.h> 59 60 #include <net/if.h> 61 #include <net/route.h> 62 63 #include <netinet/in.h> 64 #include <netinet/in_pcb.h> 65 #include <netinet/in_systm.h> 66 #include <netinet/in_var.h> 67 #include <netinet/ip.h> 68 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */ 69 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 70 #include <netinet/ip_var.h> 71 #include <netinet/ip6.h> 72 #include <netinet/icmp6.h> 73 #include <netinet6/in6_pcb.h> 74 #include <netinet6/ip6_var.h> 75 #include <netinet6/nd6.h> 76 #include <netinet/tcp.h> 77 #include <netinet/tcp_fsm.h> 78 #include <netinet/tcp_seq.h> 79 #include <netinet/tcp_timer.h> 80 #include <netinet/tcp_var.h> 81 #include <netinet6/tcp6_var.h> 82 #include <netinet/tcpip.h> 83 #ifdef TCPDEBUG 84 #include <netinet/tcp_debug.h> 85 #endif /* TCPDEBUG */ 86 87 #ifdef FAST_IPSEC 88 #include <netipsec/ipsec.h> 89 #include <netipsec/ipsec6.h> 90 #endif /*FAST_IPSEC*/ 91 92 #ifdef IPSEC 93 #include <netinet6/ipsec.h> 94 #include <netinet6/ipsec6.h> 95 #include <netkey/key.h> 96 #endif /*IPSEC*/ 97 98 #include <machine/in_cksum.h> 99 100 static const int tcprexmtthresh = 3; 101 tcp_cc tcp_ccgen; 102 103 struct tcpstat tcpstat; 104 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 105 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 106 107 static int log_in_vain = 0; 108 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 109 &log_in_vain, 0, "Log all incoming TCP connections"); 110 111 static int blackhole = 0; 112 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 113 &blackhole, 0, "Do not send RST when dropping refused connections"); 114 115 int tcp_delack_enabled = 1; 116 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 117 &tcp_delack_enabled, 0, 118 "Delay ACK to try and piggyback it onto a data packet"); 119 120 #ifdef TCP_DROP_SYNFIN 121 static int drop_synfin = 0; 122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 123 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 124 #endif 125 126 static int tcp_do_rfc3042 = 1; 127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 128 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 129 130 static int tcp_do_rfc3390 = 1; 131 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 132 &tcp_do_rfc3390, 0, 133 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 134 135 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 136 "TCP Segment Reassembly Queue"); 137 138 static int tcp_reass_maxseg = 0; 139 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 140 &tcp_reass_maxseg, 0, 141 "Global maximum number of TCP Segments in Reassembly Queue"); 142 143 int tcp_reass_qsize = 0; 144 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 145 &tcp_reass_qsize, 0, 146 "Global number of TCP Segments currently in Reassembly Queue"); 147 148 static int tcp_reass_maxqlen = 48; 149 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 150 &tcp_reass_maxqlen, 0, 151 "Maximum number of TCP Segments per individual Reassembly Queue"); 152 153 static int tcp_reass_overflows = 0; 154 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 155 &tcp_reass_overflows, 0, 156 "Global number of TCP Segment Reassembly Queue Overflows"); 157 158 struct inpcbhead tcb; 159 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 160 struct inpcbinfo tcbinfo; 161 struct mtx *tcbinfo_mtx; 162 163 static void tcp_dooptions(struct tcpcb *, struct tcpopt *, u_char *, 164 int, int, struct tcphdr *); 165 166 static void tcp_pulloutofband(struct socket *, 167 struct tcphdr *, struct mbuf *, int); 168 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 169 struct mbuf *); 170 static void tcp_xmit_timer(struct tcpcb *, int); 171 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 172 static int tcp_timewait(struct tcptw *, struct tcpopt *, 173 struct tcphdr *, struct mbuf *, int); 174 175 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 176 #ifdef INET6 177 #define ND6_HINT(tp) \ 178 do { \ 179 if ((tp) && (tp)->t_inpcb && \ 180 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 181 nd6_nud_hint(NULL, NULL, 0); \ 182 } while (0) 183 #else 184 #define ND6_HINT(tp) 185 #endif 186 187 /* 188 * Indicate whether this ack should be delayed. We can delay the ack if 189 * - there is no delayed ack timer in progress and 190 * - our last ack wasn't a 0-sized window. We never want to delay 191 * the ack that opens up a 0-sized window and 192 * - delayed acks are enabled or 193 * - this is a half-synchronized T/TCP connection. 194 */ 195 #define DELAY_ACK(tp) \ 196 ((!callout_active(tp->tt_delack) && \ 197 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 198 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 199 200 /* Initialize TCP reassembly queue */ 201 uma_zone_t tcp_reass_zone; 202 void 203 tcp_reass_init() 204 { 205 tcp_reass_maxseg = nmbclusters / 16; 206 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 207 &tcp_reass_maxseg); 208 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 209 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 210 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 211 } 212 213 static int 214 tcp_reass(tp, th, tlenp, m) 215 register struct tcpcb *tp; 216 register struct tcphdr *th; 217 int *tlenp; 218 struct mbuf *m; 219 { 220 struct tseg_qent *q; 221 struct tseg_qent *p = NULL; 222 struct tseg_qent *nq; 223 struct tseg_qent *te = NULL; 224 struct socket *so = tp->t_inpcb->inp_socket; 225 int flags; 226 227 /* 228 * XXX: tcp_reass() is rather inefficient with its data structures 229 * and should be rewritten (see NetBSD for optimizations). While 230 * doing that it should move to its own file tcp_reass.c. 231 */ 232 233 /* 234 * Call with th==0 after become established to 235 * force pre-ESTABLISHED data up to user socket. 236 */ 237 if (th == 0) 238 goto present; 239 240 /* 241 * Limit the number of segments in the reassembly queue to prevent 242 * holding on to too many segments (and thus running out of mbufs). 243 * Make sure to let the missing segment through which caused this 244 * queue. Always keep one global queue entry spare to be able to 245 * process the missing segment. 246 */ 247 if (th->th_seq != tp->rcv_nxt && 248 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 249 tp->t_segqlen >= tcp_reass_maxqlen)) { 250 tcp_reass_overflows++; 251 tcpstat.tcps_rcvmemdrop++; 252 m_freem(m); 253 return (0); 254 } 255 256 /* 257 * Allocate a new queue entry. If we can't, or hit the zone limit 258 * just drop the pkt. 259 */ 260 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 261 if (te == NULL) { 262 tcpstat.tcps_rcvmemdrop++; 263 m_freem(m); 264 return (0); 265 } 266 tp->t_segqlen++; 267 tcp_reass_qsize++; 268 269 /* 270 * Find a segment which begins after this one does. 271 */ 272 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 273 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 274 break; 275 p = q; 276 } 277 278 /* 279 * If there is a preceding segment, it may provide some of 280 * our data already. If so, drop the data from the incoming 281 * segment. If it provides all of our data, drop us. 282 */ 283 if (p != NULL) { 284 register int i; 285 /* conversion to int (in i) handles seq wraparound */ 286 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 287 if (i > 0) { 288 if (i >= *tlenp) { 289 tcpstat.tcps_rcvduppack++; 290 tcpstat.tcps_rcvdupbyte += *tlenp; 291 m_freem(m); 292 uma_zfree(tcp_reass_zone, te); 293 tp->t_segqlen--; 294 tcp_reass_qsize--; 295 /* 296 * Try to present any queued data 297 * at the left window edge to the user. 298 * This is needed after the 3-WHS 299 * completes. 300 */ 301 goto present; /* ??? */ 302 } 303 m_adj(m, i); 304 *tlenp -= i; 305 th->th_seq += i; 306 } 307 } 308 tcpstat.tcps_rcvoopack++; 309 tcpstat.tcps_rcvoobyte += *tlenp; 310 311 /* 312 * While we overlap succeeding segments trim them or, 313 * if they are completely covered, dequeue them. 314 */ 315 while (q) { 316 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 317 if (i <= 0) 318 break; 319 if (i < q->tqe_len) { 320 q->tqe_th->th_seq += i; 321 q->tqe_len -= i; 322 m_adj(q->tqe_m, i); 323 break; 324 } 325 326 nq = LIST_NEXT(q, tqe_q); 327 LIST_REMOVE(q, tqe_q); 328 m_freem(q->tqe_m); 329 uma_zfree(tcp_reass_zone, q); 330 tp->t_segqlen--; 331 tcp_reass_qsize--; 332 q = nq; 333 } 334 335 /* Insert the new segment queue entry into place. */ 336 te->tqe_m = m; 337 te->tqe_th = th; 338 te->tqe_len = *tlenp; 339 340 if (p == NULL) { 341 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 342 } else { 343 LIST_INSERT_AFTER(p, te, tqe_q); 344 } 345 346 present: 347 /* 348 * Present data to user, advancing rcv_nxt through 349 * completed sequence space. 350 */ 351 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 352 return (0); 353 q = LIST_FIRST(&tp->t_segq); 354 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 355 return (0); 356 SOCKBUF_LOCK(&so->so_rcv); 357 do { 358 tp->rcv_nxt += q->tqe_len; 359 flags = q->tqe_th->th_flags & TH_FIN; 360 nq = LIST_NEXT(q, tqe_q); 361 LIST_REMOVE(q, tqe_q); 362 /* Unlocked read. */ 363 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 364 m_freem(q->tqe_m); 365 else 366 sbappendstream_locked(&so->so_rcv, q->tqe_m); 367 uma_zfree(tcp_reass_zone, q); 368 tp->t_segqlen--; 369 tcp_reass_qsize--; 370 q = nq; 371 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 372 ND6_HINT(tp); 373 sorwakeup_locked(so); 374 return (flags); 375 } 376 377 /* 378 * TCP input routine, follows pages 65-76 of the 379 * protocol specification dated September, 1981 very closely. 380 */ 381 #ifdef INET6 382 int 383 tcp6_input(mp, offp, proto) 384 struct mbuf **mp; 385 int *offp, proto; 386 { 387 register struct mbuf *m = *mp; 388 struct in6_ifaddr *ia6; 389 390 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 391 392 /* 393 * draft-itojun-ipv6-tcp-to-anycast 394 * better place to put this in? 395 */ 396 ia6 = ip6_getdstifaddr(m); 397 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 398 struct ip6_hdr *ip6; 399 400 ip6 = mtod(m, struct ip6_hdr *); 401 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 402 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 403 return IPPROTO_DONE; 404 } 405 406 tcp_input(m, *offp); 407 return IPPROTO_DONE; 408 } 409 #endif 410 411 void 412 tcp_input(m, off0) 413 register struct mbuf *m; 414 int off0; 415 { 416 register struct tcphdr *th; 417 register struct ip *ip = NULL; 418 register struct ipovly *ipov; 419 register struct inpcb *inp = NULL; 420 u_char *optp = NULL; 421 int optlen = 0; 422 int len, tlen, off; 423 int drop_hdrlen; 424 register struct tcpcb *tp = 0; 425 register int thflags; 426 struct socket *so = 0; 427 int todrop, acked, ourfinisacked, needoutput = 0; 428 u_long tiwin; 429 struct tcpopt to; /* options in this segment */ 430 struct rmxp_tao tao; /* our TAO cache entry */ 431 int headlocked = 0; 432 #ifdef IPFIREWALL_FORWARD 433 struct m_tag *fwd_tag; 434 #endif 435 int rstreason; /* For badport_bandlim accounting purposes */ 436 437 struct ip6_hdr *ip6 = NULL; 438 #ifdef INET6 439 int isipv6; 440 #else 441 const int isipv6 = 0; 442 #endif 443 444 #ifdef TCPDEBUG 445 /* 446 * The size of tcp_saveipgen must be the size of the max ip header, 447 * now IPv6. 448 */ 449 u_char tcp_saveipgen[40]; 450 struct tcphdr tcp_savetcp; 451 short ostate = 0; 452 #endif 453 454 #ifdef INET6 455 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 456 #endif 457 bzero(&tao, sizeof(tao)); 458 bzero((char *)&to, sizeof(to)); 459 460 tcpstat.tcps_rcvtotal++; 461 462 if (isipv6) { 463 #ifdef INET6 464 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 465 ip6 = mtod(m, struct ip6_hdr *); 466 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 467 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 468 tcpstat.tcps_rcvbadsum++; 469 goto drop; 470 } 471 th = (struct tcphdr *)((caddr_t)ip6 + off0); 472 473 /* 474 * Be proactive about unspecified IPv6 address in source. 475 * As we use all-zero to indicate unbounded/unconnected pcb, 476 * unspecified IPv6 address can be used to confuse us. 477 * 478 * Note that packets with unspecified IPv6 destination is 479 * already dropped in ip6_input. 480 */ 481 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 482 /* XXX stat */ 483 goto drop; 484 } 485 #else 486 th = NULL; /* XXX: avoid compiler warning */ 487 #endif 488 } else { 489 /* 490 * Get IP and TCP header together in first mbuf. 491 * Note: IP leaves IP header in first mbuf. 492 */ 493 if (off0 > sizeof (struct ip)) { 494 ip_stripoptions(m, (struct mbuf *)0); 495 off0 = sizeof(struct ip); 496 } 497 if (m->m_len < sizeof (struct tcpiphdr)) { 498 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 499 tcpstat.tcps_rcvshort++; 500 return; 501 } 502 } 503 ip = mtod(m, struct ip *); 504 ipov = (struct ipovly *)ip; 505 th = (struct tcphdr *)((caddr_t)ip + off0); 506 tlen = ip->ip_len; 507 508 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 509 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 510 th->th_sum = m->m_pkthdr.csum_data; 511 else 512 th->th_sum = in_pseudo(ip->ip_src.s_addr, 513 ip->ip_dst.s_addr, 514 htonl(m->m_pkthdr.csum_data + 515 ip->ip_len + 516 IPPROTO_TCP)); 517 th->th_sum ^= 0xffff; 518 #ifdef TCPDEBUG 519 ipov->ih_len = (u_short)tlen; 520 ipov->ih_len = htons(ipov->ih_len); 521 #endif 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 if (off < sizeof (struct tcphdr) || off > tlen) { 548 tcpstat.tcps_rcvbadoff++; 549 goto drop; 550 } 551 tlen -= off; /* tlen is used instead of ti->ti_len */ 552 if (off > sizeof (struct tcphdr)) { 553 if (isipv6) { 554 #ifdef INET6 555 IP6_EXTHDR_CHECK(m, off0, off, ); 556 ip6 = mtod(m, struct ip6_hdr *); 557 th = (struct tcphdr *)((caddr_t)ip6 + off0); 558 #endif 559 } else { 560 if (m->m_len < sizeof(struct ip) + off) { 561 if ((m = m_pullup(m, sizeof (struct ip) + off)) 562 == 0) { 563 tcpstat.tcps_rcvshort++; 564 return; 565 } 566 ip = mtod(m, struct ip *); 567 ipov = (struct ipovly *)ip; 568 th = (struct tcphdr *)((caddr_t)ip + off0); 569 } 570 } 571 optlen = off - sizeof (struct tcphdr); 572 optp = (u_char *)(th + 1); 573 } 574 thflags = th->th_flags; 575 576 #ifdef TCP_DROP_SYNFIN 577 /* 578 * If the drop_synfin option is enabled, drop all packets with 579 * both the SYN and FIN bits set. This prevents e.g. nmap from 580 * identifying the TCP/IP stack. 581 * 582 * This is a violation of the TCP specification. 583 */ 584 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 585 goto drop; 586 #endif 587 588 /* 589 * Convert TCP protocol specific fields to host format. 590 */ 591 th->th_seq = ntohl(th->th_seq); 592 th->th_ack = ntohl(th->th_ack); 593 th->th_win = ntohs(th->th_win); 594 th->th_urp = ntohs(th->th_urp); 595 596 /* 597 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 598 * until after ip6_savecontrol() is called and before other functions 599 * which don't want those proto headers. 600 * Because ip6_savecontrol() is going to parse the mbuf to 601 * search for data to be passed up to user-land, it wants mbuf 602 * parameters to be unchanged. 603 * XXX: the call of ip6_savecontrol() has been obsoleted based on 604 * latest version of the advanced API (20020110). 605 */ 606 drop_hdrlen = off0 + off; 607 608 /* 609 * Locate pcb for segment. 610 */ 611 INP_INFO_WLOCK(&tcbinfo); 612 headlocked = 1; 613 findpcb: 614 #ifdef IPFIREWALL_FORWARD 615 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 616 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 617 618 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 619 struct sockaddr_in *next_hop; 620 621 next_hop = (struct sockaddr_in *)(fwd_tag+1); 622 /* 623 * Transparently forwarded. Pretend to be the destination. 624 * already got one like this? 625 */ 626 inp = in_pcblookup_hash(&tcbinfo, 627 ip->ip_src, th->th_sport, 628 ip->ip_dst, th->th_dport, 629 0, m->m_pkthdr.rcvif); 630 if (!inp) { 631 /* It's new. Try to find the ambushing socket. */ 632 inp = in_pcblookup_hash(&tcbinfo, 633 ip->ip_src, th->th_sport, 634 next_hop->sin_addr, 635 next_hop->sin_port ? 636 ntohs(next_hop->sin_port) : 637 th->th_dport, 638 1, m->m_pkthdr.rcvif); 639 } 640 /* Remove the tag from the packet. We don't need it anymore. */ 641 m_tag_delete(m, fwd_tag); 642 } else { 643 #endif /* IPFIREWALL_FORWARD */ 644 if (isipv6) { 645 #ifdef INET6 646 inp = in6_pcblookup_hash(&tcbinfo, 647 &ip6->ip6_src, th->th_sport, 648 &ip6->ip6_dst, th->th_dport, 649 1, m->m_pkthdr.rcvif); 650 #endif 651 } else 652 inp = in_pcblookup_hash(&tcbinfo, 653 ip->ip_src, th->th_sport, 654 ip->ip_dst, th->th_dport, 655 1, m->m_pkthdr.rcvif); 656 #ifdef IPFIREWALL_FORWARD 657 } 658 #endif /* IPFIREWALL_FORWARD */ 659 660 #if defined(IPSEC) || defined(FAST_IPSEC) 661 #ifdef INET6 662 if (isipv6) { 663 if (inp != NULL && ipsec6_in_reject(m, inp)) { 664 #ifdef IPSEC 665 ipsec6stat.in_polvio++; 666 #endif 667 goto drop; 668 } 669 } else 670 #endif /* INET6 */ 671 if (inp != NULL && ipsec4_in_reject(m, inp)) { 672 #ifdef IPSEC 673 ipsecstat.in_polvio++; 674 #endif 675 goto drop; 676 } 677 #endif /*IPSEC || FAST_IPSEC*/ 678 679 /* 680 * If the state is CLOSED (i.e., TCB does not exist) then 681 * all data in the incoming segment is discarded. 682 * If the TCB exists but is in CLOSED state, it is embryonic, 683 * but should either do a listen or a connect soon. 684 */ 685 if (inp == NULL) { 686 if (log_in_vain) { 687 #ifdef INET6 688 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 689 #else 690 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 691 #endif 692 693 if (isipv6) { 694 #ifdef INET6 695 strcpy(dbuf, "["); 696 strcpy(sbuf, "["); 697 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 698 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 699 strcat(dbuf, "]"); 700 strcat(sbuf, "]"); 701 #endif 702 } else { 703 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 704 strcpy(sbuf, inet_ntoa(ip->ip_src)); 705 } 706 switch (log_in_vain) { 707 case 1: 708 if ((thflags & TH_SYN) == 0) 709 break; 710 /* FALLTHROUGH */ 711 case 2: 712 log(LOG_INFO, 713 "Connection attempt to TCP %s:%d " 714 "from %s:%d flags:0x%02x\n", 715 dbuf, ntohs(th->th_dport), sbuf, 716 ntohs(th->th_sport), thflags); 717 break; 718 default: 719 break; 720 } 721 } 722 if (blackhole) { 723 switch (blackhole) { 724 case 1: 725 if (thflags & TH_SYN) 726 goto drop; 727 break; 728 case 2: 729 goto drop; 730 default: 731 goto drop; 732 } 733 } 734 rstreason = BANDLIM_RST_CLOSEDPORT; 735 goto dropwithreset; 736 } 737 INP_LOCK(inp); 738 if (inp->inp_vflag & INP_TIMEWAIT) { 739 /* 740 * The only option of relevance is TOF_CC, and only if 741 * present in a SYN segment. See tcp_timewait(). 742 */ 743 if (thflags & TH_SYN) 744 tcp_dooptions((struct tcpcb *)NULL, &to, optp, optlen, 1, th); 745 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 746 &to, th, m, tlen)) 747 goto findpcb; 748 /* 749 * tcp_timewait unlocks inp. 750 */ 751 INP_INFO_WUNLOCK(&tcbinfo); 752 return; 753 } 754 tp = intotcpcb(inp); 755 if (tp == 0) { 756 INP_UNLOCK(inp); 757 rstreason = BANDLIM_RST_CLOSEDPORT; 758 goto dropwithreset; 759 } 760 if (tp->t_state == TCPS_CLOSED) 761 goto drop; 762 763 /* Unscale the window into a 32-bit value. */ 764 if ((thflags & TH_SYN) == 0) 765 tiwin = th->th_win << tp->snd_scale; 766 else 767 tiwin = th->th_win; 768 769 #ifdef MAC 770 INP_LOCK_ASSERT(inp); 771 if (mac_check_inpcb_deliver(inp, m)) 772 goto drop; 773 #endif 774 so = inp->inp_socket; 775 #ifdef TCPDEBUG 776 if (so->so_options & SO_DEBUG) { 777 ostate = tp->t_state; 778 if (isipv6) 779 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 780 else 781 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 782 tcp_savetcp = *th; 783 } 784 #endif 785 if (so->so_options & SO_ACCEPTCONN) { 786 struct in_conninfo inc; 787 788 #ifdef INET6 789 inc.inc_isipv6 = isipv6; 790 #endif 791 if (isipv6) { 792 inc.inc6_faddr = ip6->ip6_src; 793 inc.inc6_laddr = ip6->ip6_dst; 794 } else { 795 inc.inc_faddr = ip->ip_src; 796 inc.inc_laddr = ip->ip_dst; 797 } 798 inc.inc_fport = th->th_sport; 799 inc.inc_lport = th->th_dport; 800 801 /* 802 * If the state is LISTEN then ignore segment if it contains 803 * a RST. If the segment contains an ACK then it is bad and 804 * send a RST. If it does not contain a SYN then it is not 805 * interesting; drop it. 806 * 807 * If the state is SYN_RECEIVED (syncache) and seg contains 808 * an ACK, but not for our SYN/ACK, send a RST. If the seg 809 * contains a RST, check the sequence number to see if it 810 * is a valid reset segment. 811 */ 812 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 813 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 814 if (!syncache_expand(&inc, th, &so, m)) { 815 /* 816 * No syncache entry, or ACK was not 817 * for our SYN/ACK. Send a RST. 818 */ 819 tcpstat.tcps_badsyn++; 820 rstreason = BANDLIM_RST_OPENPORT; 821 goto dropwithreset; 822 } 823 if (so == NULL) { 824 /* 825 * Could not complete 3-way handshake, 826 * connection is being closed down, and 827 * syncache will free mbuf. 828 */ 829 INP_UNLOCK(inp); 830 INP_INFO_WUNLOCK(&tcbinfo); 831 return; 832 } 833 /* 834 * Socket is created in state SYN_RECEIVED. 835 * Continue processing segment. 836 */ 837 INP_UNLOCK(inp); 838 inp = sotoinpcb(so); 839 INP_LOCK(inp); 840 tp = intotcpcb(inp); 841 /* 842 * This is what would have happened in 843 * tcp_output() when the SYN,ACK was sent. 844 */ 845 tp->snd_up = tp->snd_una; 846 tp->snd_max = tp->snd_nxt = tp->iss + 1; 847 tp->last_ack_sent = tp->rcv_nxt; 848 /* 849 * RFC1323: The window in SYN & SYN/ACK 850 * segments is never scaled. 851 */ 852 tp->snd_wnd = tiwin; /* unscaled */ 853 goto after_listen; 854 } 855 if (thflags & TH_RST) { 856 syncache_chkrst(&inc, th); 857 goto drop; 858 } 859 if (thflags & TH_ACK) { 860 syncache_badack(&inc); 861 tcpstat.tcps_badsyn++; 862 rstreason = BANDLIM_RST_OPENPORT; 863 goto dropwithreset; 864 } 865 goto drop; 866 } 867 868 /* 869 * Segment's flags are (SYN) or (SYN|FIN). 870 */ 871 #ifdef INET6 872 /* 873 * If deprecated address is forbidden, 874 * we do not accept SYN to deprecated interface 875 * address to prevent any new inbound connection from 876 * getting established. 877 * When we do not accept SYN, we send a TCP RST, 878 * with deprecated source address (instead of dropping 879 * it). We compromise it as it is much better for peer 880 * to send a RST, and RST will be the final packet 881 * for the exchange. 882 * 883 * If we do not forbid deprecated addresses, we accept 884 * the SYN packet. RFC2462 does not suggest dropping 885 * SYN in this case. 886 * If we decipher RFC2462 5.5.4, it says like this: 887 * 1. use of deprecated addr with existing 888 * communication is okay - "SHOULD continue to be 889 * used" 890 * 2. use of it with new communication: 891 * (2a) "SHOULD NOT be used if alternate address 892 * with sufficient scope is available" 893 * (2b) nothing mentioned otherwise. 894 * Here we fall into (2b) case as we have no choice in 895 * our source address selection - we must obey the peer. 896 * 897 * The wording in RFC2462 is confusing, and there are 898 * multiple description text for deprecated address 899 * handling - worse, they are not exactly the same. 900 * I believe 5.5.4 is the best one, so we follow 5.5.4. 901 */ 902 if (isipv6 && !ip6_use_deprecated) { 903 struct in6_ifaddr *ia6; 904 905 if ((ia6 = ip6_getdstifaddr(m)) && 906 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 907 INP_UNLOCK(inp); 908 tp = NULL; 909 rstreason = BANDLIM_RST_OPENPORT; 910 goto dropwithreset; 911 } 912 } 913 #endif 914 /* 915 * If it is from this socket, drop it, it must be forged. 916 * Don't bother responding if the destination was a broadcast. 917 */ 918 if (th->th_dport == th->th_sport) { 919 if (isipv6) { 920 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 921 &ip6->ip6_src)) 922 goto drop; 923 } else { 924 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 925 goto drop; 926 } 927 } 928 /* 929 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 930 * 931 * Note that it is quite possible to receive unicast 932 * link-layer packets with a broadcast IP address. Use 933 * in_broadcast() to find them. 934 */ 935 if (m->m_flags & (M_BCAST|M_MCAST)) 936 goto drop; 937 if (isipv6) { 938 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 939 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 940 goto drop; 941 } else { 942 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 943 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 944 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 945 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 946 goto drop; 947 } 948 /* 949 * SYN appears to be valid; create compressed TCP state 950 * for syncache, or perform t/tcp connection. 951 */ 952 if (so->so_qlen <= so->so_qlimit) { 953 #ifdef TCPDEBUG 954 if (so->so_options & SO_DEBUG) 955 tcp_trace(TA_INPUT, ostate, tp, 956 (void *)tcp_saveipgen, &tcp_savetcp, 0); 957 #endif 958 tcp_dooptions(tp, &to, optp, optlen, 1, th); 959 if (!syncache_add(&inc, &to, th, &so, m)) 960 goto drop; 961 if (so == NULL) { 962 /* 963 * Entry added to syncache, mbuf used to 964 * send SYN,ACK packet. 965 */ 966 KASSERT(headlocked, ("headlocked")); 967 INP_UNLOCK(inp); 968 INP_INFO_WUNLOCK(&tcbinfo); 969 return; 970 } 971 /* 972 * Segment passed TAO tests. 973 */ 974 INP_UNLOCK(inp); 975 inp = sotoinpcb(so); 976 INP_LOCK(inp); 977 tp = intotcpcb(inp); 978 tp->snd_wnd = tiwin; 979 tp->t_starttime = ticks; 980 tp->t_state = TCPS_ESTABLISHED; 981 982 /* 983 * T/TCP logic: 984 * If there is a FIN or if there is data, then 985 * delay SYN,ACK(SYN) in the hope of piggy-backing 986 * it on a response segment. Otherwise must send 987 * ACK now in case the other side is slow starting. 988 */ 989 if (thflags & TH_FIN || tlen != 0) 990 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 991 else 992 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 993 tcpstat.tcps_connects++; 994 soisconnected(so); 995 goto trimthenstep6; 996 } 997 goto drop; 998 } 999 after_listen: 1000 KASSERT(headlocked, ("tcp_input(): after_listen head is not locked")); 1001 INP_LOCK_ASSERT(inp); 1002 1003 /* XXX temp debugging */ 1004 /* should not happen - syncache should pick up these connections */ 1005 if (tp->t_state == TCPS_LISTEN) 1006 panic("tcp_input: TCPS_LISTEN"); 1007 1008 /* 1009 * This is the second part of the MSS DoS prevention code (after 1010 * minmss on the sending side) and it deals with too many too small 1011 * tcp packets in a too short timeframe (1 second). 1012 * 1013 * For every full second we count the number of received packets 1014 * and bytes. If we get a lot of packets per second for this connection 1015 * (tcp_minmssoverload) we take a closer look at it and compute the 1016 * average packet size for the past second. If that is less than 1017 * tcp_minmss we get too many packets with very small payload which 1018 * is not good and burdens our system (and every packet generates 1019 * a wakeup to the process connected to our socket). We can reasonable 1020 * expect this to be small packet DoS attack to exhaust our CPU 1021 * cycles. 1022 * 1023 * Care has to be taken for the minimum packet overload value. This 1024 * value defines the minimum number of packets per second before we 1025 * start to worry. This must not be too low to avoid killing for 1026 * example interactive connections with many small packets like 1027 * telnet or SSH. 1028 * 1029 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1030 * this check. 1031 * 1032 * Account for packet if payload packet, skip over ACK, etc. 1033 */ 1034 if (tcp_minmss && tcp_minmssoverload && 1035 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1036 if (tp->rcv_second > ticks) { 1037 tp->rcv_pps++; 1038 tp->rcv_byps += tlen + off; 1039 if (tp->rcv_pps > tcp_minmssoverload) { 1040 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1041 printf("too many small tcp packets from " 1042 "%s:%u, av. %lubyte/packet, " 1043 "dropping connection\n", 1044 #ifdef INET6 1045 isipv6 ? 1046 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1047 #endif 1048 inet_ntoa(inp->inp_inc.inc_faddr), 1049 inp->inp_inc.inc_fport, 1050 tp->rcv_byps / tp->rcv_pps); 1051 tp = tcp_drop(tp, ECONNRESET); 1052 tcpstat.tcps_minmssdrops++; 1053 goto drop; 1054 } 1055 } 1056 } else { 1057 tp->rcv_second = ticks + hz; 1058 tp->rcv_pps = 1; 1059 tp->rcv_byps = tlen + off; 1060 } 1061 } 1062 1063 /* 1064 * Segment received on connection. 1065 * Reset idle time and keep-alive timer. 1066 */ 1067 tp->t_rcvtime = ticks; 1068 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1069 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1070 1071 /* 1072 * Process options only when we get SYN/ACK back. The SYN case 1073 * for incoming connections is handled in tcp_syncache. 1074 * XXX this is traditional behavior, may need to be cleaned up. 1075 */ 1076 tcp_dooptions(tp, &to, optp, optlen, thflags & TH_SYN, th); 1077 if (thflags & TH_SYN) { 1078 if (to.to_flags & TOF_SCALE) { 1079 tp->t_flags |= TF_RCVD_SCALE; 1080 tp->requested_s_scale = to.to_requested_s_scale; 1081 } 1082 if (to.to_flags & TOF_TS) { 1083 tp->t_flags |= TF_RCVD_TSTMP; 1084 tp->ts_recent = to.to_tsval; 1085 tp->ts_recent_age = ticks; 1086 } 1087 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 1088 tp->t_flags |= TF_RCVD_CC; 1089 if (to.to_flags & TOF_MSS) 1090 tcp_mss(tp, to.to_mss); 1091 if (tp->sack_enable) { 1092 if (!(to.to_flags & TOF_SACK)) 1093 tp->sack_enable = 0; 1094 else 1095 tp->t_flags |= TF_SACK_PERMIT; 1096 } 1097 1098 } 1099 1100 if (tp->sack_enable) { 1101 /* Delete stale (cumulatively acked) SACK holes */ 1102 tcp_del_sackholes(tp, th); 1103 tp->rcv_laststart = th->th_seq; /* last recv'd segment*/ 1104 tp->rcv_lastend = th->th_seq + tlen; 1105 } 1106 1107 /* 1108 * Header prediction: check for the two common cases 1109 * of a uni-directional data xfer. If the packet has 1110 * no control flags, is in-sequence, the window didn't 1111 * change and we're not retransmitting, it's a 1112 * candidate. If the length is zero and the ack moved 1113 * forward, we're the sender side of the xfer. Just 1114 * free the data acked & wake any higher level process 1115 * that was blocked waiting for space. If the length 1116 * is non-zero and the ack didn't move, we're the 1117 * receiver side. If we're getting packets in-order 1118 * (the reassembly queue is empty), add the data to 1119 * the socket buffer and note that we need a delayed ack. 1120 * Make sure that the hidden state-flags are also off. 1121 * Since we check for TCPS_ESTABLISHED above, it can only 1122 * be TH_NEEDSYN. 1123 */ 1124 if (tp->t_state == TCPS_ESTABLISHED && 1125 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1126 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1127 ((to.to_flags & TOF_TS) == 0 || 1128 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1129 /* 1130 * Using the CC option is compulsory if once started: 1131 * the segment is OK if no T/TCP was negotiated or 1132 * if the segment has a CC option equal to CCrecv 1133 */ 1134 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1135 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 1136 th->th_seq == tp->rcv_nxt && 1137 tiwin && tiwin == tp->snd_wnd && 1138 tp->snd_nxt == tp->snd_max) { 1139 1140 /* 1141 * If last ACK falls within this segment's sequence numbers, 1142 * record the timestamp. 1143 * NOTE that the test is modified according to the latest 1144 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1145 */ 1146 if ((to.to_flags & TOF_TS) != 0 && 1147 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1148 tp->ts_recent_age = ticks; 1149 tp->ts_recent = to.to_tsval; 1150 } 1151 1152 if (tlen == 0) { 1153 if (SEQ_GT(th->th_ack, tp->snd_una) && 1154 SEQ_LEQ(th->th_ack, tp->snd_max) && 1155 tp->snd_cwnd >= tp->snd_wnd && 1156 ((!tcp_do_newreno && !tp->sack_enable && 1157 tp->t_dupacks < tcprexmtthresh) || 1158 ((tcp_do_newreno || tp->sack_enable) && 1159 !IN_FASTRECOVERY(tp)))) { 1160 KASSERT(headlocked, ("headlocked")); 1161 INP_INFO_WUNLOCK(&tcbinfo); 1162 /* 1163 * this is a pure ack for outstanding data. 1164 */ 1165 ++tcpstat.tcps_predack; 1166 /* 1167 * "bad retransmit" recovery 1168 */ 1169 if (tp->t_rxtshift == 1 && 1170 ticks < tp->t_badrxtwin) { 1171 ++tcpstat.tcps_sndrexmitbad; 1172 tp->snd_cwnd = tp->snd_cwnd_prev; 1173 tp->snd_ssthresh = 1174 tp->snd_ssthresh_prev; 1175 tp->snd_recover = tp->snd_recover_prev; 1176 if (tp->t_flags & TF_WASFRECOVERY) 1177 ENTER_FASTRECOVERY(tp); 1178 tp->snd_nxt = tp->snd_max; 1179 tp->t_badrxtwin = 0; 1180 } 1181 1182 /* 1183 * Recalculate the transmit timer / rtt. 1184 * 1185 * Some boxes send broken timestamp replies 1186 * during the SYN+ACK phase, ignore 1187 * timestamps of 0 or we could calculate a 1188 * huge RTT and blow up the retransmit timer. 1189 */ 1190 if ((to.to_flags & TOF_TS) != 0 && 1191 to.to_tsecr) { 1192 tcp_xmit_timer(tp, 1193 ticks - to.to_tsecr + 1); 1194 } else if (tp->t_rtttime && 1195 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1196 tcp_xmit_timer(tp, 1197 ticks - tp->t_rtttime); 1198 } 1199 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1200 acked = th->th_ack - tp->snd_una; 1201 tcpstat.tcps_rcvackpack++; 1202 tcpstat.tcps_rcvackbyte += acked; 1203 sbdrop(&so->so_snd, acked); 1204 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1205 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1206 tp->snd_recover = th->th_ack - 1; 1207 tp->snd_una = th->th_ack; 1208 /* 1209 * pull snd_wl2 up to prevent seq wrap relative 1210 * to th_ack. 1211 */ 1212 tp->snd_wl2 = th->th_ack; 1213 tp->t_dupacks = 0; 1214 m_freem(m); 1215 ND6_HINT(tp); /* some progress has been done */ 1216 1217 /* 1218 * If all outstanding data are acked, stop 1219 * retransmit timer, otherwise restart timer 1220 * using current (possibly backed-off) value. 1221 * If process is waiting for space, 1222 * wakeup/selwakeup/signal. If data 1223 * are ready to send, let tcp_output 1224 * decide between more output or persist. 1225 1226 #ifdef TCPDEBUG 1227 if (so->so_options & SO_DEBUG) 1228 tcp_trace(TA_INPUT, ostate, tp, 1229 (void *)tcp_saveipgen, 1230 &tcp_savetcp, 0); 1231 #endif 1232 */ 1233 if (tp->snd_una == tp->snd_max) 1234 callout_stop(tp->tt_rexmt); 1235 else if (!callout_active(tp->tt_persist)) 1236 callout_reset(tp->tt_rexmt, 1237 tp->t_rxtcur, 1238 tcp_timer_rexmt, tp); 1239 1240 sowwakeup(so); 1241 if (so->so_snd.sb_cc) 1242 (void) tcp_output(tp); 1243 goto check_delack; 1244 } 1245 } else if (th->th_ack == tp->snd_una && 1246 LIST_EMPTY(&tp->t_segq) && 1247 tlen <= sbspace(&so->so_rcv)) { 1248 KASSERT(headlocked, ("headlocked")); 1249 INP_INFO_WUNLOCK(&tcbinfo); 1250 /* 1251 * this is a pure, in-sequence data packet 1252 * with nothing on the reassembly queue and 1253 * we have enough buffer space to take it. 1254 */ 1255 /* Clean receiver SACK report if present */ 1256 if (tp->sack_enable && tp->rcv_numsacks) 1257 tcp_clean_sackreport(tp); 1258 ++tcpstat.tcps_preddat; 1259 tp->rcv_nxt += tlen; 1260 /* 1261 * Pull snd_wl1 up to prevent seq wrap relative to 1262 * th_seq. 1263 */ 1264 tp->snd_wl1 = th->th_seq; 1265 /* 1266 * Pull rcv_up up to prevent seq wrap relative to 1267 * rcv_nxt. 1268 */ 1269 tp->rcv_up = tp->rcv_nxt; 1270 tcpstat.tcps_rcvpack++; 1271 tcpstat.tcps_rcvbyte += tlen; 1272 ND6_HINT(tp); /* some progress has been done */ 1273 /* 1274 #ifdef TCPDEBUG 1275 if (so->so_options & SO_DEBUG) 1276 tcp_trace(TA_INPUT, ostate, tp, 1277 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1278 #endif 1279 * Add data to socket buffer. 1280 */ 1281 /* Unlocked read. */ 1282 SOCKBUF_LOCK(&so->so_rcv); 1283 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1284 m_freem(m); 1285 } else { 1286 m_adj(m, drop_hdrlen); /* delayed header drop */ 1287 sbappendstream_locked(&so->so_rcv, m); 1288 } 1289 sorwakeup_locked(so); 1290 if (DELAY_ACK(tp)) { 1291 tp->t_flags |= TF_DELACK; 1292 } else { 1293 tp->t_flags |= TF_ACKNOW; 1294 tcp_output(tp); 1295 } 1296 goto check_delack; 1297 } 1298 } 1299 1300 /* 1301 * Calculate amount of space in receive window, 1302 * and then do TCP input processing. 1303 * Receive window is amount of space in rcv queue, 1304 * but not less than advertised window. 1305 */ 1306 { int win; 1307 1308 win = sbspace(&so->so_rcv); 1309 if (win < 0) 1310 win = 0; 1311 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1312 } 1313 1314 switch (tp->t_state) { 1315 1316 /* 1317 * If the state is SYN_RECEIVED: 1318 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1319 */ 1320 case TCPS_SYN_RECEIVED: 1321 if ((thflags & TH_ACK) && 1322 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1323 SEQ_GT(th->th_ack, tp->snd_max))) { 1324 rstreason = BANDLIM_RST_OPENPORT; 1325 goto dropwithreset; 1326 } 1327 break; 1328 1329 /* 1330 * If the state is SYN_SENT: 1331 * if seg contains an ACK, but not for our SYN, drop the input. 1332 * if seg contains a RST, then drop the connection. 1333 * if seg does not contain SYN, then drop it. 1334 * Otherwise this is an acceptable SYN segment 1335 * initialize tp->rcv_nxt and tp->irs 1336 * if seg contains ack then advance tp->snd_una 1337 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1338 * arrange for segment to be acked (eventually) 1339 * continue processing rest of data/controls, beginning with URG 1340 */ 1341 case TCPS_SYN_SENT: 1342 if (tcp_do_rfc1644) 1343 tcp_hc_gettao(&inp->inp_inc, &tao); 1344 1345 if ((thflags & TH_ACK) && 1346 (SEQ_LEQ(th->th_ack, tp->iss) || 1347 SEQ_GT(th->th_ack, tp->snd_max))) { 1348 /* 1349 * If we have a cached CCsent for the remote host, 1350 * hence we haven't just crashed and restarted, 1351 * do not send a RST. This may be a retransmission 1352 * from the other side after our earlier ACK was lost. 1353 * Our new SYN, when it arrives, will serve as the 1354 * needed ACK. 1355 */ 1356 if (tao.tao_ccsent != 0) 1357 goto drop; 1358 else { 1359 rstreason = BANDLIM_UNLIMITED; 1360 goto dropwithreset; 1361 } 1362 } 1363 if (thflags & TH_RST) { 1364 if (thflags & TH_ACK) 1365 tp = tcp_drop(tp, ECONNREFUSED); 1366 goto drop; 1367 } 1368 if ((thflags & TH_SYN) == 0) 1369 goto drop; 1370 tp->snd_wnd = th->th_win; /* initial send window */ 1371 tp->cc_recv = to.to_cc; /* foreign CC */ 1372 1373 tp->irs = th->th_seq; 1374 tcp_rcvseqinit(tp); 1375 if (thflags & TH_ACK) { 1376 /* 1377 * Our SYN was acked. If segment contains CC.ECHO 1378 * option, check it to make sure this segment really 1379 * matches our SYN. If not, just drop it as old 1380 * duplicate, but send an RST if we're still playing 1381 * by the old rules. If no CC.ECHO option, make sure 1382 * we don't get fooled into using T/TCP. 1383 */ 1384 if (to.to_flags & TOF_CCECHO) { 1385 if (tp->cc_send != to.to_ccecho) { 1386 if (tao.tao_ccsent != 0) 1387 goto drop; 1388 else { 1389 rstreason = BANDLIM_UNLIMITED; 1390 goto dropwithreset; 1391 } 1392 } 1393 } else 1394 tp->t_flags &= ~TF_RCVD_CC; 1395 tcpstat.tcps_connects++; 1396 soisconnected(so); 1397 #ifdef MAC 1398 SOCK_LOCK(so); 1399 mac_set_socket_peer_from_mbuf(m, so); 1400 SOCK_UNLOCK(so); 1401 #endif 1402 /* Do window scaling on this connection? */ 1403 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1404 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1405 tp->snd_scale = tp->requested_s_scale; 1406 tp->rcv_scale = tp->request_r_scale; 1407 } 1408 /* Segment is acceptable, update cache if undefined. */ 1409 if (tao.tao_ccsent == 0 && tcp_do_rfc1644) 1410 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0); 1411 1412 tp->rcv_adv += tp->rcv_wnd; 1413 tp->snd_una++; /* SYN is acked */ 1414 /* 1415 * If there's data, delay ACK; if there's also a FIN 1416 * ACKNOW will be turned on later. 1417 */ 1418 if (DELAY_ACK(tp) && tlen != 0) 1419 callout_reset(tp->tt_delack, tcp_delacktime, 1420 tcp_timer_delack, tp); 1421 else 1422 tp->t_flags |= TF_ACKNOW; 1423 /* 1424 * Received <SYN,ACK> in SYN_SENT[*] state. 1425 * Transitions: 1426 * SYN_SENT --> ESTABLISHED 1427 * SYN_SENT* --> FIN_WAIT_1 1428 */ 1429 tp->t_starttime = ticks; 1430 if (tp->t_flags & TF_NEEDFIN) { 1431 tp->t_state = TCPS_FIN_WAIT_1; 1432 tp->t_flags &= ~TF_NEEDFIN; 1433 thflags &= ~TH_SYN; 1434 } else { 1435 tp->t_state = TCPS_ESTABLISHED; 1436 callout_reset(tp->tt_keep, tcp_keepidle, 1437 tcp_timer_keep, tp); 1438 } 1439 } else { 1440 /* 1441 * Received initial SYN in SYN-SENT[*] state => 1442 * simultaneous open. If segment contains CC option 1443 * and there is a cached CC, apply TAO test. 1444 * If it succeeds, connection is * half-synchronized. 1445 * Otherwise, do 3-way handshake: 1446 * SYN-SENT -> SYN-RECEIVED 1447 * SYN-SENT* -> SYN-RECEIVED* 1448 * If there was no CC option, clear cached CC value. 1449 */ 1450 tp->t_flags |= TF_ACKNOW; 1451 callout_stop(tp->tt_rexmt); 1452 if (to.to_flags & TOF_CC) { 1453 if (tao.tao_cc != 0 && 1454 CC_GT(to.to_cc, tao.tao_cc)) { 1455 /* 1456 * update cache and make transition: 1457 * SYN-SENT -> ESTABLISHED* 1458 * SYN-SENT* -> FIN-WAIT-1* 1459 */ 1460 tao.tao_cc = to.to_cc; 1461 tcp_hc_updatetao(&inp->inp_inc, 1462 TCP_HC_TAO_CC, to.to_cc, 0); 1463 tp->t_starttime = ticks; 1464 if (tp->t_flags & TF_NEEDFIN) { 1465 tp->t_state = TCPS_FIN_WAIT_1; 1466 tp->t_flags &= ~TF_NEEDFIN; 1467 } else { 1468 tp->t_state = TCPS_ESTABLISHED; 1469 callout_reset(tp->tt_keep, 1470 tcp_keepidle, 1471 tcp_timer_keep, 1472 tp); 1473 } 1474 tp->t_flags |= TF_NEEDSYN; 1475 } else 1476 tp->t_state = TCPS_SYN_RECEIVED; 1477 } else { 1478 if (tcp_do_rfc1644) { 1479 /* CC.NEW or no option => invalidate cache */ 1480 tao.tao_cc = 0; 1481 tcp_hc_updatetao(&inp->inp_inc, 1482 TCP_HC_TAO_CC, to.to_cc, 0); 1483 } 1484 tp->t_state = TCPS_SYN_RECEIVED; 1485 } 1486 } 1487 1488 trimthenstep6: 1489 KASSERT(headlocked, 1490 ("tcp_input(): trimthenstep6 head is not locked")); 1491 INP_LOCK_ASSERT(inp); 1492 1493 /* 1494 * Advance th->th_seq to correspond to first data byte. 1495 * If data, trim to stay within window, 1496 * dropping FIN if necessary. 1497 */ 1498 th->th_seq++; 1499 if (tlen > tp->rcv_wnd) { 1500 todrop = tlen - tp->rcv_wnd; 1501 m_adj(m, -todrop); 1502 tlen = tp->rcv_wnd; 1503 thflags &= ~TH_FIN; 1504 tcpstat.tcps_rcvpackafterwin++; 1505 tcpstat.tcps_rcvbyteafterwin += todrop; 1506 } 1507 tp->snd_wl1 = th->th_seq - 1; 1508 tp->rcv_up = th->th_seq; 1509 /* 1510 * Client side of transaction: already sent SYN and data. 1511 * If the remote host used T/TCP to validate the SYN, 1512 * our data will be ACK'd; if so, enter normal data segment 1513 * processing in the middle of step 5, ack processing. 1514 * Otherwise, goto step 6. 1515 */ 1516 if (thflags & TH_ACK) 1517 goto process_ACK; 1518 1519 goto step6; 1520 1521 /* 1522 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1523 * if segment contains a SYN and CC [not CC.NEW] option: 1524 * if state == TIME_WAIT and connection duration > MSL, 1525 * drop packet and send RST; 1526 * 1527 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1528 * ack the FIN (and data) in retransmission queue. 1529 * Complete close and delete TCPCB. Then reprocess 1530 * segment, hoping to find new TCPCB in LISTEN state; 1531 * 1532 * else must be old SYN; drop it. 1533 * else do normal processing. 1534 */ 1535 case TCPS_LAST_ACK: 1536 case TCPS_CLOSING: 1537 case TCPS_TIME_WAIT: 1538 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1539 if ((thflags & TH_SYN) && 1540 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1541 if (tp->t_state == TCPS_TIME_WAIT && 1542 (ticks - tp->t_starttime) > tcp_msl) { 1543 rstreason = BANDLIM_UNLIMITED; 1544 goto dropwithreset; 1545 } 1546 if (CC_GT(to.to_cc, tp->cc_recv)) { 1547 tp = tcp_close(tp); 1548 goto findpcb; 1549 } 1550 else 1551 goto drop; 1552 } 1553 break; /* continue normal processing */ 1554 } 1555 1556 /* 1557 * States other than LISTEN or SYN_SENT. 1558 * First check the RST flag and sequence number since reset segments 1559 * are exempt from the timestamp and connection count tests. This 1560 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1561 * below which allowed reset segments in half the sequence space 1562 * to fall though and be processed (which gives forged reset 1563 * segments with a random sequence number a 50 percent chance of 1564 * killing a connection). 1565 * Then check timestamp, if present. 1566 * Then check the connection count, if present. 1567 * Then check that at least some bytes of segment are within 1568 * receive window. If segment begins before rcv_nxt, 1569 * drop leading data (and SYN); if nothing left, just ack. 1570 * 1571 * 1572 * If the RST bit is set, check the sequence number to see 1573 * if this is a valid reset segment. 1574 * RFC 793 page 37: 1575 * In all states except SYN-SENT, all reset (RST) segments 1576 * are validated by checking their SEQ-fields. A reset is 1577 * valid if its sequence number is in the window. 1578 * Note: this does not take into account delayed ACKs, so 1579 * we should test against last_ack_sent instead of rcv_nxt. 1580 * The sequence number in the reset segment is normally an 1581 * echo of our outgoing acknowlegement numbers, but some hosts 1582 * send a reset with the sequence number at the rightmost edge 1583 * of our receive window, and we have to handle this case. 1584 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1585 * that brute force RST attacks are possible. To combat this, 1586 * we use a much stricter check while in the ESTABLISHED state, 1587 * only accepting RSTs where the sequence number is equal to 1588 * last_ack_sent. In all other states (the states in which a 1589 * RST is more likely), the more permissive check is used. 1590 * If we have multiple segments in flight, the intial reset 1591 * segment sequence numbers will be to the left of last_ack_sent, 1592 * but they will eventually catch up. 1593 * In any case, it never made sense to trim reset segments to 1594 * fit the receive window since RFC 1122 says: 1595 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1596 * 1597 * A TCP SHOULD allow a received RST segment to include data. 1598 * 1599 * DISCUSSION 1600 * It has been suggested that a RST segment could contain 1601 * ASCII text that encoded and explained the cause of the 1602 * RST. No standard has yet been established for such 1603 * data. 1604 * 1605 * If the reset segment passes the sequence number test examine 1606 * the state: 1607 * SYN_RECEIVED STATE: 1608 * If passive open, return to LISTEN state. 1609 * If active open, inform user that connection was refused. 1610 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1611 * Inform user that connection was reset, and close tcb. 1612 * CLOSING, LAST_ACK STATES: 1613 * Close the tcb. 1614 * TIME_WAIT STATE: 1615 * Drop the segment - see Stevens, vol. 2, p. 964 and 1616 * RFC 1337. 1617 */ 1618 if (thflags & TH_RST) { 1619 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1620 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1621 switch (tp->t_state) { 1622 1623 case TCPS_SYN_RECEIVED: 1624 so->so_error = ECONNREFUSED; 1625 goto close; 1626 1627 case TCPS_ESTABLISHED: 1628 if (tp->last_ack_sent != th->th_seq) { 1629 tcpstat.tcps_badrst++; 1630 goto drop; 1631 } 1632 case TCPS_FIN_WAIT_1: 1633 case TCPS_FIN_WAIT_2: 1634 case TCPS_CLOSE_WAIT: 1635 so->so_error = ECONNRESET; 1636 close: 1637 tp->t_state = TCPS_CLOSED; 1638 tcpstat.tcps_drops++; 1639 tp = tcp_close(tp); 1640 break; 1641 1642 case TCPS_CLOSING: 1643 case TCPS_LAST_ACK: 1644 tp = tcp_close(tp); 1645 break; 1646 1647 case TCPS_TIME_WAIT: 1648 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1649 ("timewait")); 1650 break; 1651 } 1652 } 1653 goto drop; 1654 } 1655 1656 /* 1657 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1658 * and it's less than ts_recent, drop it. 1659 */ 1660 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1661 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1662 1663 /* Check to see if ts_recent is over 24 days old. */ 1664 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1665 /* 1666 * Invalidate ts_recent. If this segment updates 1667 * ts_recent, the age will be reset later and ts_recent 1668 * will get a valid value. If it does not, setting 1669 * ts_recent to zero will at least satisfy the 1670 * requirement that zero be placed in the timestamp 1671 * echo reply when ts_recent isn't valid. The 1672 * age isn't reset until we get a valid ts_recent 1673 * because we don't want out-of-order segments to be 1674 * dropped when ts_recent is old. 1675 */ 1676 tp->ts_recent = 0; 1677 } else { 1678 tcpstat.tcps_rcvduppack++; 1679 tcpstat.tcps_rcvdupbyte += tlen; 1680 tcpstat.tcps_pawsdrop++; 1681 if (tlen) 1682 goto dropafterack; 1683 goto drop; 1684 } 1685 } 1686 1687 /* 1688 * T/TCP mechanism 1689 * If T/TCP was negotiated and the segment doesn't have CC, 1690 * or if its CC is wrong then drop the segment. 1691 * RST segments do not have to comply with this. 1692 */ 1693 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1694 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1695 goto dropafterack; 1696 1697 /* 1698 * In the SYN-RECEIVED state, validate that the packet belongs to 1699 * this connection before trimming the data to fit the receive 1700 * window. Check the sequence number versus IRS since we know 1701 * the sequence numbers haven't wrapped. This is a partial fix 1702 * for the "LAND" DoS attack. 1703 */ 1704 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1705 rstreason = BANDLIM_RST_OPENPORT; 1706 goto dropwithreset; 1707 } 1708 1709 todrop = tp->rcv_nxt - th->th_seq; 1710 if (todrop > 0) { 1711 if (thflags & TH_SYN) { 1712 thflags &= ~TH_SYN; 1713 th->th_seq++; 1714 if (th->th_urp > 1) 1715 th->th_urp--; 1716 else 1717 thflags &= ~TH_URG; 1718 todrop--; 1719 } 1720 /* 1721 * Following if statement from Stevens, vol. 2, p. 960. 1722 */ 1723 if (todrop > tlen 1724 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1725 /* 1726 * Any valid FIN must be to the left of the window. 1727 * At this point the FIN must be a duplicate or out 1728 * of sequence; drop it. 1729 */ 1730 thflags &= ~TH_FIN; 1731 1732 /* 1733 * Send an ACK to resynchronize and drop any data. 1734 * But keep on processing for RST or ACK. 1735 */ 1736 tp->t_flags |= TF_ACKNOW; 1737 todrop = tlen; 1738 tcpstat.tcps_rcvduppack++; 1739 tcpstat.tcps_rcvdupbyte += todrop; 1740 } else { 1741 tcpstat.tcps_rcvpartduppack++; 1742 tcpstat.tcps_rcvpartdupbyte += todrop; 1743 } 1744 drop_hdrlen += todrop; /* drop from the top afterwards */ 1745 th->th_seq += todrop; 1746 tlen -= todrop; 1747 if (th->th_urp > todrop) 1748 th->th_urp -= todrop; 1749 else { 1750 thflags &= ~TH_URG; 1751 th->th_urp = 0; 1752 } 1753 } 1754 1755 /* 1756 * If new data are received on a connection after the 1757 * user processes are gone, then RST the other end. 1758 */ 1759 if ((so->so_state & SS_NOFDREF) && 1760 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1761 tp = tcp_close(tp); 1762 tcpstat.tcps_rcvafterclose++; 1763 rstreason = BANDLIM_UNLIMITED; 1764 goto dropwithreset; 1765 } 1766 1767 /* 1768 * If segment ends after window, drop trailing data 1769 * (and PUSH and FIN); if nothing left, just ACK. 1770 */ 1771 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1772 if (todrop > 0) { 1773 tcpstat.tcps_rcvpackafterwin++; 1774 if (todrop >= tlen) { 1775 tcpstat.tcps_rcvbyteafterwin += tlen; 1776 /* 1777 * If a new connection request is received 1778 * while in TIME_WAIT, drop the old connection 1779 * and start over if the sequence numbers 1780 * are above the previous ones. 1781 */ 1782 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1783 if (thflags & TH_SYN && 1784 tp->t_state == TCPS_TIME_WAIT && 1785 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1786 tp = tcp_close(tp); 1787 goto findpcb; 1788 } 1789 /* 1790 * If window is closed can only take segments at 1791 * window edge, and have to drop data and PUSH from 1792 * incoming segments. Continue processing, but 1793 * remember to ack. Otherwise, drop segment 1794 * and ack. 1795 */ 1796 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1797 tp->t_flags |= TF_ACKNOW; 1798 tcpstat.tcps_rcvwinprobe++; 1799 } else 1800 goto dropafterack; 1801 } else 1802 tcpstat.tcps_rcvbyteafterwin += todrop; 1803 m_adj(m, -todrop); 1804 tlen -= todrop; 1805 thflags &= ~(TH_PUSH|TH_FIN); 1806 } 1807 1808 /* 1809 * If last ACK falls within this segment's sequence numbers, 1810 * record its timestamp. 1811 * NOTE that the test is modified according to the latest 1812 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1813 */ 1814 if ((to.to_flags & TOF_TS) != 0 && 1815 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1816 tp->ts_recent_age = ticks; 1817 tp->ts_recent = to.to_tsval; 1818 } 1819 1820 /* 1821 * If a SYN is in the window, then this is an 1822 * error and we send an RST and drop the connection. 1823 */ 1824 if (thflags & TH_SYN) { 1825 tp = tcp_drop(tp, ECONNRESET); 1826 rstreason = BANDLIM_UNLIMITED; 1827 goto drop; 1828 } 1829 1830 /* 1831 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1832 * flag is on (half-synchronized state), then queue data for 1833 * later processing; else drop segment and return. 1834 */ 1835 if ((thflags & TH_ACK) == 0) { 1836 if (tp->t_state == TCPS_SYN_RECEIVED || 1837 (tp->t_flags & TF_NEEDSYN)) 1838 goto step6; 1839 else 1840 goto drop; 1841 } 1842 1843 /* 1844 * Ack processing. 1845 */ 1846 switch (tp->t_state) { 1847 1848 /* 1849 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1850 * ESTABLISHED state and continue processing. 1851 * The ACK was checked above. 1852 */ 1853 case TCPS_SYN_RECEIVED: 1854 1855 tcpstat.tcps_connects++; 1856 soisconnected(so); 1857 /* Do window scaling? */ 1858 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1859 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1860 tp->snd_scale = tp->requested_s_scale; 1861 tp->rcv_scale = tp->request_r_scale; 1862 } 1863 /* 1864 * Upon successful completion of 3-way handshake, 1865 * update cache.CC, pass any queued data to the user, 1866 * and advance state appropriately. 1867 */ 1868 if (tcp_do_rfc1644) { 1869 tao.tao_cc = tp->cc_recv; 1870 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC, 1871 tp->cc_recv, 0); 1872 } 1873 /* 1874 * Make transitions: 1875 * SYN-RECEIVED -> ESTABLISHED 1876 * SYN-RECEIVED* -> FIN-WAIT-1 1877 */ 1878 tp->t_starttime = ticks; 1879 if (tp->t_flags & TF_NEEDFIN) { 1880 tp->t_state = TCPS_FIN_WAIT_1; 1881 tp->t_flags &= ~TF_NEEDFIN; 1882 } else { 1883 tp->t_state = TCPS_ESTABLISHED; 1884 callout_reset(tp->tt_keep, tcp_keepidle, 1885 tcp_timer_keep, tp); 1886 } 1887 /* 1888 * If segment contains data or ACK, will call tcp_reass() 1889 * later; if not, do so now to pass queued data to user. 1890 */ 1891 if (tlen == 0 && (thflags & TH_FIN) == 0) 1892 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1893 (struct mbuf *)0); 1894 tp->snd_wl1 = th->th_seq - 1; 1895 /* FALLTHROUGH */ 1896 1897 /* 1898 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1899 * ACKs. If the ack is in the range 1900 * tp->snd_una < th->th_ack <= tp->snd_max 1901 * then advance tp->snd_una to th->th_ack and drop 1902 * data from the retransmission queue. If this ACK reflects 1903 * more up to date window information we update our window information. 1904 */ 1905 case TCPS_ESTABLISHED: 1906 case TCPS_FIN_WAIT_1: 1907 case TCPS_FIN_WAIT_2: 1908 case TCPS_CLOSE_WAIT: 1909 case TCPS_CLOSING: 1910 case TCPS_LAST_ACK: 1911 case TCPS_TIME_WAIT: 1912 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1913 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1914 if (tlen == 0 && tiwin == tp->snd_wnd) { 1915 tcpstat.tcps_rcvdupack++; 1916 /* 1917 * If we have outstanding data (other than 1918 * a window probe), this is a completely 1919 * duplicate ack (ie, window info didn't 1920 * change), the ack is the biggest we've 1921 * seen and we've seen exactly our rexmt 1922 * threshhold of them, assume a packet 1923 * has been dropped and retransmit it. 1924 * Kludge snd_nxt & the congestion 1925 * window so we send only this one 1926 * packet. 1927 * 1928 * We know we're losing at the current 1929 * window size so do congestion avoidance 1930 * (set ssthresh to half the current window 1931 * and pull our congestion window back to 1932 * the new ssthresh). 1933 * 1934 * Dup acks mean that packets have left the 1935 * network (they're now cached at the receiver) 1936 * so bump cwnd by the amount in the receiver 1937 * to keep a constant cwnd packets in the 1938 * network. 1939 */ 1940 if (!callout_active(tp->tt_rexmt) || 1941 th->th_ack != tp->snd_una) 1942 tp->t_dupacks = 0; 1943 else if (++tp->t_dupacks > tcprexmtthresh || 1944 ((tcp_do_newreno || tp->sack_enable) && 1945 IN_FASTRECOVERY(tp))) { 1946 tp->snd_cwnd += tp->t_maxseg; 1947 (void) tcp_output(tp); 1948 goto drop; 1949 } else if (tp->t_dupacks == tcprexmtthresh) { 1950 tcp_seq onxt = tp->snd_nxt; 1951 u_int win; 1952 1953 /* 1954 * If we're doing sack, check to 1955 * see if we're already in sack 1956 * recovery. If we're not doing sack, 1957 * check to see if we're in newreno 1958 * recovery. 1959 */ 1960 if (tp->sack_enable) { 1961 if (IN_FASTRECOVERY(tp)) { 1962 tp->t_dupacks = 0; 1963 break; 1964 } 1965 } else if (tcp_do_newreno) { 1966 if (SEQ_LEQ(th->th_ack, 1967 tp->snd_recover)) { 1968 tp->t_dupacks = 0; 1969 break; 1970 } 1971 } 1972 win = min(tp->snd_wnd, tp->snd_cwnd) / 1973 2 / tp->t_maxseg; 1974 if (win < 2) 1975 win = 2; 1976 tp->snd_ssthresh = win * tp->t_maxseg; 1977 ENTER_FASTRECOVERY(tp); 1978 tp->snd_recover = tp->snd_max; 1979 callout_stop(tp->tt_rexmt); 1980 tp->t_rtttime = 0; 1981 if (tp->sack_enable) { 1982 tcpstat.tcps_sack_recovery_episode++; 1983 tp->snd_cwnd = 1984 tp->t_maxseg * 1985 tp->t_dupacks; 1986 (void) tcp_output(tp); 1987 tp->snd_cwnd += 1988 tp->snd_ssthresh; 1989 goto drop; 1990 } 1991 1992 tp->snd_nxt = th->th_ack; 1993 tp->snd_cwnd = tp->t_maxseg; 1994 (void) tcp_output(tp); 1995 KASSERT(tp->snd_limited <= 2, 1996 ("tp->snd_limited too big")); 1997 tp->snd_cwnd = tp->snd_ssthresh + 1998 tp->t_maxseg * 1999 (tp->t_dupacks - tp->snd_limited); 2000 if (SEQ_GT(onxt, tp->snd_nxt)) 2001 tp->snd_nxt = onxt; 2002 goto drop; 2003 } else if (tcp_do_rfc3042) { 2004 u_long oldcwnd = tp->snd_cwnd; 2005 tcp_seq oldsndmax = tp->snd_max; 2006 u_int sent; 2007 2008 KASSERT(tp->t_dupacks == 1 || 2009 tp->t_dupacks == 2, 2010 ("dupacks not 1 or 2")); 2011 if (tp->t_dupacks == 1) 2012 tp->snd_limited = 0; 2013 tp->snd_cwnd = 2014 (tp->snd_nxt - tp->snd_una) + 2015 (tp->t_dupacks - tp->snd_limited) * 2016 tp->t_maxseg; 2017 (void) tcp_output(tp); 2018 sent = tp->snd_max - oldsndmax; 2019 if (sent > tp->t_maxseg) { 2020 KASSERT((tp->t_dupacks == 2 && 2021 tp->snd_limited == 0) || 2022 (sent == tp->t_maxseg + 1 && 2023 tp->t_flags & TF_SENTFIN), 2024 ("sent too much")); 2025 tp->snd_limited = 2; 2026 } else if (sent > 0) 2027 ++tp->snd_limited; 2028 tp->snd_cwnd = oldcwnd; 2029 goto drop; 2030 } 2031 } else 2032 tp->t_dupacks = 0; 2033 break; 2034 } 2035 2036 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 2037 2038 /* 2039 * If the congestion window was inflated to account 2040 * for the other side's cached packets, retract it. 2041 */ 2042 if (tcp_do_newreno || tp->sack_enable) { 2043 if (IN_FASTRECOVERY(tp)) { 2044 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2045 if (tp->sack_enable) 2046 tcp_sack_partialack(tp, th); 2047 else 2048 tcp_newreno_partial_ack(tp, th); 2049 } else { 2050 /* 2051 * Out of fast recovery. 2052 * Window inflation should have left us 2053 * with approximately snd_ssthresh 2054 * outstanding data. 2055 * But in case we would be inclined to 2056 * send a burst, better to do it via 2057 * the slow start mechanism. 2058 */ 2059 if (SEQ_GT(th->th_ack + 2060 tp->snd_ssthresh, 2061 tp->snd_max)) 2062 tp->snd_cwnd = tp->snd_max - 2063 th->th_ack + 2064 tp->t_maxseg; 2065 else 2066 tp->snd_cwnd = tp->snd_ssthresh; 2067 } 2068 } 2069 } else { 2070 if (tp->t_dupacks >= tcprexmtthresh && 2071 tp->snd_cwnd > tp->snd_ssthresh) 2072 tp->snd_cwnd = tp->snd_ssthresh; 2073 } 2074 tp->t_dupacks = 0; 2075 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2076 tcpstat.tcps_rcvacktoomuch++; 2077 goto dropafterack; 2078 } 2079 /* 2080 * If we reach this point, ACK is not a duplicate, 2081 * i.e., it ACKs something we sent. 2082 */ 2083 if (tp->t_flags & TF_NEEDSYN) { 2084 /* 2085 * T/TCP: Connection was half-synchronized, and our 2086 * SYN has been ACK'd (so connection is now fully 2087 * synchronized). Go to non-starred state, 2088 * increment snd_una for ACK of SYN, and check if 2089 * we can do window scaling. 2090 */ 2091 tp->t_flags &= ~TF_NEEDSYN; 2092 tp->snd_una++; 2093 /* Do window scaling? */ 2094 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2095 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2096 tp->snd_scale = tp->requested_s_scale; 2097 tp->rcv_scale = tp->request_r_scale; 2098 } 2099 } 2100 2101 process_ACK: 2102 KASSERT(headlocked, 2103 ("tcp_input(): process_ACK head is not locked")); 2104 INP_LOCK_ASSERT(inp); 2105 2106 acked = th->th_ack - tp->snd_una; 2107 tcpstat.tcps_rcvackpack++; 2108 tcpstat.tcps_rcvackbyte += acked; 2109 2110 /* 2111 * If we just performed our first retransmit, and the ACK 2112 * arrives within our recovery window, then it was a mistake 2113 * to do the retransmit in the first place. Recover our 2114 * original cwnd and ssthresh, and proceed to transmit where 2115 * we left off. 2116 */ 2117 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2118 ++tcpstat.tcps_sndrexmitbad; 2119 tp->snd_cwnd = tp->snd_cwnd_prev; 2120 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2121 tp->snd_recover = tp->snd_recover_prev; 2122 if (tp->t_flags & TF_WASFRECOVERY) 2123 ENTER_FASTRECOVERY(tp); 2124 tp->snd_nxt = tp->snd_max; 2125 tp->t_badrxtwin = 0; /* XXX probably not required */ 2126 } 2127 2128 /* 2129 * If we have a timestamp reply, update smoothed 2130 * round trip time. If no timestamp is present but 2131 * transmit timer is running and timed sequence 2132 * number was acked, update smoothed round trip time. 2133 * Since we now have an rtt measurement, cancel the 2134 * timer backoff (cf., Phil Karn's retransmit alg.). 2135 * Recompute the initial retransmit timer. 2136 * 2137 * Some boxes send broken timestamp replies 2138 * during the SYN+ACK phase, ignore 2139 * timestamps of 0 or we could calculate a 2140 * huge RTT and blow up the retransmit timer. 2141 */ 2142 if ((to.to_flags & TOF_TS) != 0 && 2143 to.to_tsecr) { 2144 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2145 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2146 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2147 } 2148 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2149 2150 /* 2151 * If all outstanding data is acked, stop retransmit 2152 * timer and remember to restart (more output or persist). 2153 * If there is more data to be acked, restart retransmit 2154 * timer, using current (possibly backed-off) value. 2155 */ 2156 if (th->th_ack == tp->snd_max) { 2157 callout_stop(tp->tt_rexmt); 2158 needoutput = 1; 2159 } else if (!callout_active(tp->tt_persist)) 2160 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2161 tcp_timer_rexmt, tp); 2162 2163 /* 2164 * If no data (only SYN) was ACK'd, 2165 * skip rest of ACK processing. 2166 */ 2167 if (acked == 0) 2168 goto step6; 2169 2170 /* 2171 * When new data is acked, open the congestion window. 2172 * If the window gives us less than ssthresh packets 2173 * in flight, open exponentially (maxseg per packet). 2174 * Otherwise open linearly: maxseg per window 2175 * (maxseg^2 / cwnd per packet). 2176 */ 2177 if ((!tcp_do_newreno && !tp->sack_enable) || 2178 !IN_FASTRECOVERY(tp)) { 2179 register u_int cw = tp->snd_cwnd; 2180 register u_int incr = tp->t_maxseg; 2181 if (cw > tp->snd_ssthresh) 2182 incr = incr * incr / cw; 2183 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2184 } 2185 SOCKBUF_LOCK(&so->so_snd); 2186 if (acked > so->so_snd.sb_cc) { 2187 tp->snd_wnd -= so->so_snd.sb_cc; 2188 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2189 ourfinisacked = 1; 2190 } else { 2191 sbdrop_locked(&so->so_snd, acked); 2192 tp->snd_wnd -= acked; 2193 ourfinisacked = 0; 2194 } 2195 sowwakeup_locked(so); 2196 /* detect una wraparound */ 2197 if ((tcp_do_newreno || tp->sack_enable) && 2198 !IN_FASTRECOVERY(tp) && 2199 SEQ_GT(tp->snd_una, tp->snd_recover) && 2200 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2201 tp->snd_recover = th->th_ack - 1; 2202 if ((tcp_do_newreno || tp->sack_enable) && 2203 IN_FASTRECOVERY(tp) && 2204 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2205 EXIT_FASTRECOVERY(tp); 2206 tp->snd_una = th->th_ack; 2207 if (tp->sack_enable) { 2208 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2209 tp->snd_recover = tp->snd_una; 2210 } 2211 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2212 tp->snd_nxt = tp->snd_una; 2213 2214 switch (tp->t_state) { 2215 2216 /* 2217 * In FIN_WAIT_1 STATE in addition to the processing 2218 * for the ESTABLISHED state if our FIN is now acknowledged 2219 * then enter FIN_WAIT_2. 2220 */ 2221 case TCPS_FIN_WAIT_1: 2222 if (ourfinisacked) { 2223 /* 2224 * If we can't receive any more 2225 * data, then closing user can proceed. 2226 * Starting the timer is contrary to the 2227 * specification, but if we don't get a FIN 2228 * we'll hang forever. 2229 */ 2230 /* XXXjl 2231 * we should release the tp also, and use a 2232 * compressed state. 2233 */ 2234 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2235 soisdisconnected(so); 2236 callout_reset(tp->tt_2msl, tcp_maxidle, 2237 tcp_timer_2msl, tp); 2238 } 2239 tp->t_state = TCPS_FIN_WAIT_2; 2240 } 2241 break; 2242 2243 /* 2244 * In CLOSING STATE in addition to the processing for 2245 * the ESTABLISHED state if the ACK acknowledges our FIN 2246 * then enter the TIME-WAIT state, otherwise ignore 2247 * the segment. 2248 */ 2249 case TCPS_CLOSING: 2250 if (ourfinisacked) { 2251 KASSERT(headlocked, ("headlocked")); 2252 tcp_twstart(tp); 2253 INP_INFO_WUNLOCK(&tcbinfo); 2254 m_freem(m); 2255 return; 2256 } 2257 break; 2258 2259 /* 2260 * In LAST_ACK, we may still be waiting for data to drain 2261 * and/or to be acked, as well as for the ack of our FIN. 2262 * If our FIN is now acknowledged, delete the TCB, 2263 * enter the closed state and return. 2264 */ 2265 case TCPS_LAST_ACK: 2266 if (ourfinisacked) { 2267 tp = tcp_close(tp); 2268 goto drop; 2269 } 2270 break; 2271 2272 /* 2273 * In TIME_WAIT state the only thing that should arrive 2274 * is a retransmission of the remote FIN. Acknowledge 2275 * it and restart the finack timer. 2276 */ 2277 case TCPS_TIME_WAIT: 2278 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2279 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2280 tcp_timer_2msl, tp); 2281 goto dropafterack; 2282 } 2283 } 2284 2285 step6: 2286 KASSERT(headlocked, ("tcp_input(): step6 head is not locked")); 2287 INP_LOCK_ASSERT(inp); 2288 2289 /* 2290 * Update window information. 2291 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2292 */ 2293 if ((thflags & TH_ACK) && 2294 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2295 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2296 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2297 /* keep track of pure window updates */ 2298 if (tlen == 0 && 2299 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2300 tcpstat.tcps_rcvwinupd++; 2301 tp->snd_wnd = tiwin; 2302 tp->snd_wl1 = th->th_seq; 2303 tp->snd_wl2 = th->th_ack; 2304 if (tp->snd_wnd > tp->max_sndwnd) 2305 tp->max_sndwnd = tp->snd_wnd; 2306 needoutput = 1; 2307 } 2308 2309 /* 2310 * Process segments with URG. 2311 */ 2312 if ((thflags & TH_URG) && th->th_urp && 2313 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2314 /* 2315 * This is a kludge, but if we receive and accept 2316 * random urgent pointers, we'll crash in 2317 * soreceive. It's hard to imagine someone 2318 * actually wanting to send this much urgent data. 2319 */ 2320 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2321 th->th_urp = 0; /* XXX */ 2322 thflags &= ~TH_URG; /* XXX */ 2323 goto dodata; /* XXX */ 2324 } 2325 /* 2326 * If this segment advances the known urgent pointer, 2327 * then mark the data stream. This should not happen 2328 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2329 * a FIN has been received from the remote side. 2330 * In these states we ignore the URG. 2331 * 2332 * According to RFC961 (Assigned Protocols), 2333 * the urgent pointer points to the last octet 2334 * of urgent data. We continue, however, 2335 * to consider it to indicate the first octet 2336 * of data past the urgent section as the original 2337 * spec states (in one of two places). 2338 */ 2339 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2340 tp->rcv_up = th->th_seq + th->th_urp; 2341 SOCKBUF_LOCK(&so->so_rcv); 2342 so->so_oobmark = so->so_rcv.sb_cc + 2343 (tp->rcv_up - tp->rcv_nxt) - 1; 2344 if (so->so_oobmark == 0) 2345 so->so_rcv.sb_state |= SBS_RCVATMARK; 2346 SOCKBUF_UNLOCK(&so->so_rcv); 2347 sohasoutofband(so); 2348 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2349 } 2350 /* 2351 * Remove out of band data so doesn't get presented to user. 2352 * This can happen independent of advancing the URG pointer, 2353 * but if two URG's are pending at once, some out-of-band 2354 * data may creep in... ick. 2355 */ 2356 if (th->th_urp <= (u_long)tlen && 2357 !(so->so_options & SO_OOBINLINE)) { 2358 /* hdr drop is delayed */ 2359 tcp_pulloutofband(so, th, m, drop_hdrlen); 2360 } 2361 } else { 2362 /* 2363 * If no out of band data is expected, 2364 * pull receive urgent pointer along 2365 * with the receive window. 2366 */ 2367 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2368 tp->rcv_up = tp->rcv_nxt; 2369 } 2370 dodata: /* XXX */ 2371 KASSERT(headlocked, ("tcp_input(): dodata head is not locked")); 2372 INP_LOCK_ASSERT(inp); 2373 2374 /* 2375 * Process the segment text, merging it into the TCP sequencing queue, 2376 * and arranging for acknowledgment of receipt if necessary. 2377 * This process logically involves adjusting tp->rcv_wnd as data 2378 * is presented to the user (this happens in tcp_usrreq.c, 2379 * case PRU_RCVD). If a FIN has already been received on this 2380 * connection then we just ignore the text. 2381 */ 2382 if ((tlen || (thflags & TH_FIN)) && 2383 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2384 m_adj(m, drop_hdrlen); /* delayed header drop */ 2385 /* 2386 * Insert segment which includes th into TCP reassembly queue 2387 * with control block tp. Set thflags to whether reassembly now 2388 * includes a segment with FIN. This handles the common case 2389 * inline (segment is the next to be received on an established 2390 * connection, and the queue is empty), avoiding linkage into 2391 * and removal from the queue and repetition of various 2392 * conversions. 2393 * Set DELACK for segments received in order, but ack 2394 * immediately when segments are out of order (so 2395 * fast retransmit can work). 2396 */ 2397 if (th->th_seq == tp->rcv_nxt && 2398 LIST_EMPTY(&tp->t_segq) && 2399 TCPS_HAVEESTABLISHED(tp->t_state)) { 2400 if (DELAY_ACK(tp)) 2401 tp->t_flags |= TF_DELACK; 2402 else 2403 tp->t_flags |= TF_ACKNOW; 2404 tp->rcv_nxt += tlen; 2405 thflags = th->th_flags & TH_FIN; 2406 tcpstat.tcps_rcvpack++; 2407 tcpstat.tcps_rcvbyte += tlen; 2408 ND6_HINT(tp); 2409 /* Unlocked read. */ 2410 SOCKBUF_LOCK(&so->so_rcv); 2411 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2412 m_freem(m); 2413 else 2414 sbappendstream_locked(&so->so_rcv, m); 2415 sorwakeup_locked(so); 2416 } else { 2417 thflags = tcp_reass(tp, th, &tlen, m); 2418 tp->t_flags |= TF_ACKNOW; 2419 } 2420 if (tp->sack_enable) 2421 tcp_update_sack_list(tp); 2422 /* 2423 * Note the amount of data that peer has sent into 2424 * our window, in order to estimate the sender's 2425 * buffer size. 2426 */ 2427 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2428 } else { 2429 m_freem(m); 2430 thflags &= ~TH_FIN; 2431 } 2432 2433 /* 2434 * If FIN is received ACK the FIN and let the user know 2435 * that the connection is closing. 2436 */ 2437 if (thflags & TH_FIN) { 2438 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2439 socantrcvmore(so); 2440 /* 2441 * If connection is half-synchronized 2442 * (ie NEEDSYN flag on) then delay ACK, 2443 * so it may be piggybacked when SYN is sent. 2444 * Otherwise, since we received a FIN then no 2445 * more input can be expected, send ACK now. 2446 */ 2447 if (tp->t_flags & TF_NEEDSYN) 2448 tp->t_flags |= TF_DELACK; 2449 else 2450 tp->t_flags |= TF_ACKNOW; 2451 tp->rcv_nxt++; 2452 } 2453 switch (tp->t_state) { 2454 2455 /* 2456 * In SYN_RECEIVED and ESTABLISHED STATES 2457 * enter the CLOSE_WAIT state. 2458 */ 2459 case TCPS_SYN_RECEIVED: 2460 tp->t_starttime = ticks; 2461 /*FALLTHROUGH*/ 2462 case TCPS_ESTABLISHED: 2463 tp->t_state = TCPS_CLOSE_WAIT; 2464 break; 2465 2466 /* 2467 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2468 * enter the CLOSING state. 2469 */ 2470 case TCPS_FIN_WAIT_1: 2471 tp->t_state = TCPS_CLOSING; 2472 break; 2473 2474 /* 2475 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2476 * starting the time-wait timer, turning off the other 2477 * standard timers. 2478 */ 2479 case TCPS_FIN_WAIT_2: 2480 KASSERT(headlocked == 1, ("headlocked should be 1")); 2481 tcp_twstart(tp); 2482 INP_INFO_WUNLOCK(&tcbinfo); 2483 return; 2484 2485 /* 2486 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2487 */ 2488 case TCPS_TIME_WAIT: 2489 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2490 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2491 tcp_timer_2msl, tp); 2492 break; 2493 } 2494 } 2495 INP_INFO_WUNLOCK(&tcbinfo); 2496 #ifdef TCPDEBUG 2497 if (so->so_options & SO_DEBUG) 2498 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2499 &tcp_savetcp, 0); 2500 #endif 2501 2502 /* 2503 * Return any desired output. 2504 */ 2505 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2506 (void) tcp_output(tp); 2507 2508 check_delack: 2509 INP_LOCK_ASSERT(inp); 2510 if (tp->t_flags & TF_DELACK) { 2511 tp->t_flags &= ~TF_DELACK; 2512 callout_reset(tp->tt_delack, tcp_delacktime, 2513 tcp_timer_delack, tp); 2514 } 2515 INP_UNLOCK(inp); 2516 return; 2517 2518 dropafterack: 2519 /* 2520 * Generate an ACK dropping incoming segment if it occupies 2521 * sequence space, where the ACK reflects our state. 2522 * 2523 * We can now skip the test for the RST flag since all 2524 * paths to this code happen after packets containing 2525 * RST have been dropped. 2526 * 2527 * In the SYN-RECEIVED state, don't send an ACK unless the 2528 * segment we received passes the SYN-RECEIVED ACK test. 2529 * If it fails send a RST. This breaks the loop in the 2530 * "LAND" DoS attack, and also prevents an ACK storm 2531 * between two listening ports that have been sent forged 2532 * SYN segments, each with the source address of the other. 2533 */ 2534 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2535 (SEQ_GT(tp->snd_una, th->th_ack) || 2536 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2537 rstreason = BANDLIM_RST_OPENPORT; 2538 goto dropwithreset; 2539 } 2540 #ifdef TCPDEBUG 2541 if (so->so_options & SO_DEBUG) 2542 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2543 &tcp_savetcp, 0); 2544 #endif 2545 KASSERT(headlocked, ("headlocked should be 1")); 2546 INP_INFO_WUNLOCK(&tcbinfo); 2547 m_freem(m); 2548 tp->t_flags |= TF_ACKNOW; 2549 (void) tcp_output(tp); 2550 INP_UNLOCK(inp); 2551 return; 2552 2553 dropwithreset: 2554 /* 2555 * Generate a RST, dropping incoming segment. 2556 * Make ACK acceptable to originator of segment. 2557 * Don't bother to respond if destination was broadcast/multicast. 2558 */ 2559 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2560 goto drop; 2561 if (isipv6) { 2562 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2563 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2564 goto drop; 2565 } else { 2566 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2567 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2568 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2569 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2570 goto drop; 2571 } 2572 /* IPv6 anycast check is done at tcp6_input() */ 2573 2574 /* 2575 * Perform bandwidth limiting. 2576 */ 2577 if (badport_bandlim(rstreason) < 0) 2578 goto drop; 2579 2580 #ifdef TCPDEBUG 2581 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2582 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2583 &tcp_savetcp, 0); 2584 #endif 2585 2586 if (thflags & TH_ACK) 2587 /* mtod() below is safe as long as hdr dropping is delayed */ 2588 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2589 TH_RST); 2590 else { 2591 if (thflags & TH_SYN) 2592 tlen++; 2593 /* mtod() below is safe as long as hdr dropping is delayed */ 2594 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2595 (tcp_seq)0, TH_RST|TH_ACK); 2596 } 2597 2598 if (tp) 2599 INP_UNLOCK(inp); 2600 if (headlocked) 2601 INP_INFO_WUNLOCK(&tcbinfo); 2602 return; 2603 2604 drop: 2605 /* 2606 * Drop space held by incoming segment and return. 2607 */ 2608 #ifdef TCPDEBUG 2609 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2610 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2611 &tcp_savetcp, 0); 2612 #endif 2613 if (tp) 2614 INP_UNLOCK(inp); 2615 m_freem(m); 2616 if (headlocked) 2617 INP_INFO_WUNLOCK(&tcbinfo); 2618 return; 2619 } 2620 2621 /* 2622 * Parse TCP options and place in tcpopt. 2623 */ 2624 static void 2625 tcp_dooptions(tp, to, cp, cnt, is_syn, th) 2626 struct tcpcb *tp; 2627 struct tcpopt *to; 2628 u_char *cp; 2629 int cnt; 2630 int is_syn; 2631 struct tcphdr *th; 2632 { 2633 int opt, optlen; 2634 2635 to->to_flags = 0; 2636 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2637 opt = cp[0]; 2638 if (opt == TCPOPT_EOL) 2639 break; 2640 if (opt == TCPOPT_NOP) 2641 optlen = 1; 2642 else { 2643 if (cnt < 2) 2644 break; 2645 optlen = cp[1]; 2646 if (optlen < 2 || optlen > cnt) 2647 break; 2648 } 2649 switch (opt) { 2650 case TCPOPT_MAXSEG: 2651 if (optlen != TCPOLEN_MAXSEG) 2652 continue; 2653 if (!is_syn) 2654 continue; 2655 to->to_flags |= TOF_MSS; 2656 bcopy((char *)cp + 2, 2657 (char *)&to->to_mss, sizeof(to->to_mss)); 2658 to->to_mss = ntohs(to->to_mss); 2659 break; 2660 case TCPOPT_WINDOW: 2661 if (optlen != TCPOLEN_WINDOW) 2662 continue; 2663 if (! is_syn) 2664 continue; 2665 to->to_flags |= TOF_SCALE; 2666 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2667 break; 2668 case TCPOPT_TIMESTAMP: 2669 if (optlen != TCPOLEN_TIMESTAMP) 2670 continue; 2671 to->to_flags |= TOF_TS; 2672 bcopy((char *)cp + 2, 2673 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2674 to->to_tsval = ntohl(to->to_tsval); 2675 bcopy((char *)cp + 6, 2676 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2677 to->to_tsecr = ntohl(to->to_tsecr); 2678 break; 2679 case TCPOPT_CC: 2680 if (optlen != TCPOLEN_CC) 2681 continue; 2682 to->to_flags |= TOF_CC; 2683 bcopy((char *)cp + 2, 2684 (char *)&to->to_cc, sizeof(to->to_cc)); 2685 to->to_cc = ntohl(to->to_cc); 2686 break; 2687 case TCPOPT_CCNEW: 2688 if (optlen != TCPOLEN_CC) 2689 continue; 2690 if (!is_syn) 2691 continue; 2692 to->to_flags |= TOF_CCNEW; 2693 bcopy((char *)cp + 2, 2694 (char *)&to->to_cc, sizeof(to->to_cc)); 2695 to->to_cc = ntohl(to->to_cc); 2696 break; 2697 case TCPOPT_CCECHO: 2698 if (optlen != TCPOLEN_CC) 2699 continue; 2700 if (!is_syn) 2701 continue; 2702 to->to_flags |= TOF_CCECHO; 2703 bcopy((char *)cp + 2, 2704 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2705 to->to_ccecho = ntohl(to->to_ccecho); 2706 break; 2707 #ifdef TCP_SIGNATURE 2708 /* 2709 * XXX In order to reply to a host which has set the 2710 * TCP_SIGNATURE option in its initial SYN, we have to 2711 * record the fact that the option was observed here 2712 * for the syncache code to perform the correct response. 2713 */ 2714 case TCPOPT_SIGNATURE: 2715 if (optlen != TCPOLEN_SIGNATURE) 2716 continue; 2717 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2718 break; 2719 #endif 2720 case TCPOPT_SACK_PERMITTED: 2721 if (!tcp_do_sack || 2722 optlen != TCPOLEN_SACK_PERMITTED) 2723 continue; 2724 if (is_syn) { 2725 /* MUST only be set on SYN */ 2726 to->to_flags |= TOF_SACK; 2727 } 2728 break; 2729 2730 case TCPOPT_SACK: 2731 if (!tp || tcp_sack_option(tp, th, cp, optlen)) 2732 continue; 2733 break; 2734 default: 2735 continue; 2736 } 2737 } 2738 } 2739 2740 /* 2741 * Pull out of band byte out of a segment so 2742 * it doesn't appear in the user's data queue. 2743 * It is still reflected in the segment length for 2744 * sequencing purposes. 2745 */ 2746 static void 2747 tcp_pulloutofband(so, th, m, off) 2748 struct socket *so; 2749 struct tcphdr *th; 2750 register struct mbuf *m; 2751 int off; /* delayed to be droped hdrlen */ 2752 { 2753 int cnt = off + th->th_urp - 1; 2754 2755 while (cnt >= 0) { 2756 if (m->m_len > cnt) { 2757 char *cp = mtod(m, caddr_t) + cnt; 2758 struct tcpcb *tp = sototcpcb(so); 2759 2760 tp->t_iobc = *cp; 2761 tp->t_oobflags |= TCPOOB_HAVEDATA; 2762 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2763 m->m_len--; 2764 if (m->m_flags & M_PKTHDR) 2765 m->m_pkthdr.len--; 2766 return; 2767 } 2768 cnt -= m->m_len; 2769 m = m->m_next; 2770 if (m == 0) 2771 break; 2772 } 2773 panic("tcp_pulloutofband"); 2774 } 2775 2776 /* 2777 * Collect new round-trip time estimate 2778 * and update averages and current timeout. 2779 */ 2780 static void 2781 tcp_xmit_timer(tp, rtt) 2782 register struct tcpcb *tp; 2783 int rtt; 2784 { 2785 register int delta; 2786 2787 tcpstat.tcps_rttupdated++; 2788 tp->t_rttupdated++; 2789 if (tp->t_srtt != 0) { 2790 /* 2791 * srtt is stored as fixed point with 5 bits after the 2792 * binary point (i.e., scaled by 8). The following magic 2793 * is equivalent to the smoothing algorithm in rfc793 with 2794 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2795 * point). Adjust rtt to origin 0. 2796 */ 2797 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2798 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2799 2800 if ((tp->t_srtt += delta) <= 0) 2801 tp->t_srtt = 1; 2802 2803 /* 2804 * We accumulate a smoothed rtt variance (actually, a 2805 * smoothed mean difference), then set the retransmit 2806 * timer to smoothed rtt + 4 times the smoothed variance. 2807 * rttvar is stored as fixed point with 4 bits after the 2808 * binary point (scaled by 16). The following is 2809 * equivalent to rfc793 smoothing with an alpha of .75 2810 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2811 * rfc793's wired-in beta. 2812 */ 2813 if (delta < 0) 2814 delta = -delta; 2815 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2816 if ((tp->t_rttvar += delta) <= 0) 2817 tp->t_rttvar = 1; 2818 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2819 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2820 } else { 2821 /* 2822 * No rtt measurement yet - use the unsmoothed rtt. 2823 * Set the variance to half the rtt (so our first 2824 * retransmit happens at 3*rtt). 2825 */ 2826 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2827 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2828 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2829 } 2830 tp->t_rtttime = 0; 2831 tp->t_rxtshift = 0; 2832 2833 /* 2834 * the retransmit should happen at rtt + 4 * rttvar. 2835 * Because of the way we do the smoothing, srtt and rttvar 2836 * will each average +1/2 tick of bias. When we compute 2837 * the retransmit timer, we want 1/2 tick of rounding and 2838 * 1 extra tick because of +-1/2 tick uncertainty in the 2839 * firing of the timer. The bias will give us exactly the 2840 * 1.5 tick we need. But, because the bias is 2841 * statistical, we have to test that we don't drop below 2842 * the minimum feasible timer (which is 2 ticks). 2843 */ 2844 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2845 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2846 2847 /* 2848 * We received an ack for a packet that wasn't retransmitted; 2849 * it is probably safe to discard any error indications we've 2850 * received recently. This isn't quite right, but close enough 2851 * for now (a route might have failed after we sent a segment, 2852 * and the return path might not be symmetrical). 2853 */ 2854 tp->t_softerror = 0; 2855 } 2856 2857 /* 2858 * Determine a reasonable value for maxseg size. 2859 * If the route is known, check route for mtu. 2860 * If none, use an mss that can be handled on the outgoing 2861 * interface without forcing IP to fragment; if bigger than 2862 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2863 * to utilize large mbufs. If no route is found, route has no mtu, 2864 * or the destination isn't local, use a default, hopefully conservative 2865 * size (usually 512 or the default IP max size, but no more than the mtu 2866 * of the interface), as we can't discover anything about intervening 2867 * gateways or networks. We also initialize the congestion/slow start 2868 * window to be a single segment if the destination isn't local. 2869 * While looking at the routing entry, we also initialize other path-dependent 2870 * parameters from pre-set or cached values in the routing entry. 2871 * 2872 * Also take into account the space needed for options that we 2873 * send regularly. Make maxseg shorter by that amount to assure 2874 * that we can send maxseg amount of data even when the options 2875 * are present. Store the upper limit of the length of options plus 2876 * data in maxopd. 2877 * 2878 * 2879 * In case of T/TCP, we call this routine during implicit connection 2880 * setup as well (offer = -1), to initialize maxseg from the cached 2881 * MSS of our peer. 2882 * 2883 * NOTE that this routine is only called when we process an incoming 2884 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2885 */ 2886 void 2887 tcp_mss(tp, offer) 2888 struct tcpcb *tp; 2889 int offer; 2890 { 2891 int rtt, mss; 2892 u_long bufsize; 2893 u_long maxmtu; 2894 struct inpcb *inp = tp->t_inpcb; 2895 struct socket *so; 2896 struct hc_metrics_lite metrics; 2897 struct rmxp_tao tao; 2898 int origoffer = offer; 2899 #ifdef INET6 2900 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2901 size_t min_protoh = isipv6 ? 2902 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2903 sizeof (struct tcpiphdr); 2904 #else 2905 const size_t min_protoh = sizeof(struct tcpiphdr); 2906 #endif 2907 bzero(&tao, sizeof(tao)); 2908 2909 /* initialize */ 2910 #ifdef INET6 2911 if (isipv6) { 2912 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2913 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2914 } else 2915 #endif 2916 { 2917 maxmtu = tcp_maxmtu(&inp->inp_inc); 2918 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2919 } 2920 so = inp->inp_socket; 2921 2922 /* 2923 * no route to sender, stay with default mss and return 2924 */ 2925 if (maxmtu == 0) 2926 return; 2927 2928 /* what have we got? */ 2929 switch (offer) { 2930 case 0: 2931 /* 2932 * Offer == 0 means that there was no MSS on the SYN 2933 * segment, in this case we use tcp_mssdflt. 2934 */ 2935 offer = 2936 #ifdef INET6 2937 isipv6 ? tcp_v6mssdflt : 2938 #endif 2939 tcp_mssdflt; 2940 break; 2941 2942 case -1: 2943 /* 2944 * Offer == -1 means that we didn't receive SYN yet, 2945 * use cached value in that case; 2946 */ 2947 if (tcp_do_rfc1644) 2948 tcp_hc_gettao(&inp->inp_inc, &tao); 2949 if (tao.tao_mssopt != 0) 2950 offer = tao.tao_mssopt; 2951 /* FALLTHROUGH */ 2952 2953 default: 2954 /* 2955 * Prevent DoS attack with too small MSS. Round up 2956 * to at least minmss. 2957 */ 2958 offer = max(offer, tcp_minmss); 2959 /* 2960 * Sanity check: make sure that maxopd will be large 2961 * enough to allow some data on segments even if the 2962 * all the option space is used (40bytes). Otherwise 2963 * funny things may happen in tcp_output. 2964 */ 2965 offer = max(offer, 64); 2966 if (tcp_do_rfc1644) 2967 tcp_hc_updatetao(&inp->inp_inc, 2968 TCP_HC_TAO_MSSOPT, 0, offer); 2969 } 2970 2971 /* 2972 * rmx information is now retrieved from tcp_hostcache 2973 */ 2974 tcp_hc_get(&inp->inp_inc, &metrics); 2975 2976 /* 2977 * if there's a discovered mtu int tcp hostcache, use it 2978 * else, use the link mtu. 2979 */ 2980 if (metrics.rmx_mtu) 2981 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2982 else { 2983 #ifdef INET6 2984 if (isipv6) { 2985 mss = maxmtu - min_protoh; 2986 if (!path_mtu_discovery && 2987 !in6_localaddr(&inp->in6p_faddr)) 2988 mss = min(mss, tcp_v6mssdflt); 2989 } else 2990 #endif 2991 { 2992 mss = maxmtu - min_protoh; 2993 if (!path_mtu_discovery && 2994 !in_localaddr(inp->inp_faddr)) 2995 mss = min(mss, tcp_mssdflt); 2996 } 2997 } 2998 mss = min(mss, offer); 2999 3000 /* 3001 * maxopd stores the maximum length of data AND options 3002 * in a segment; maxseg is the amount of data in a normal 3003 * segment. We need to store this value (maxopd) apart 3004 * from maxseg, because now every segment carries options 3005 * and thus we normally have somewhat less data in segments. 3006 */ 3007 tp->t_maxopd = mss; 3008 3009 /* 3010 * In case of T/TCP, origoffer==-1 indicates, that no segments 3011 * were received yet. In this case we just guess, otherwise 3012 * we do the same as before T/TCP. 3013 */ 3014 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3015 (origoffer == -1 || 3016 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3017 mss -= TCPOLEN_TSTAMP_APPA; 3018 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 3019 (origoffer == -1 || 3020 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 3021 mss -= TCPOLEN_CC_APPA; 3022 tp->t_maxseg = mss; 3023 3024 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3025 if (mss > MCLBYTES) 3026 mss &= ~(MCLBYTES-1); 3027 #else 3028 if (mss > MCLBYTES) 3029 mss = mss / MCLBYTES * MCLBYTES; 3030 #endif 3031 tp->t_maxseg = mss; 3032 3033 /* 3034 * If there's a pipesize, change the socket buffer to that size, 3035 * don't change if sb_hiwat is different than default (then it 3036 * has been changed on purpose with setsockopt). 3037 * Make the socket buffers an integral number of mss units; 3038 * if the mss is larger than the socket buffer, decrease the mss. 3039 */ 3040 SOCKBUF_LOCK(&so->so_snd); 3041 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3042 bufsize = metrics.rmx_sendpipe; 3043 else 3044 bufsize = so->so_snd.sb_hiwat; 3045 if (bufsize < mss) 3046 mss = bufsize; 3047 else { 3048 bufsize = roundup(bufsize, mss); 3049 if (bufsize > sb_max) 3050 bufsize = sb_max; 3051 if (bufsize > so->so_snd.sb_hiwat) 3052 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3053 } 3054 SOCKBUF_UNLOCK(&so->so_snd); 3055 tp->t_maxseg = mss; 3056 3057 SOCKBUF_LOCK(&so->so_rcv); 3058 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3059 bufsize = metrics.rmx_recvpipe; 3060 else 3061 bufsize = so->so_rcv.sb_hiwat; 3062 if (bufsize > mss) { 3063 bufsize = roundup(bufsize, mss); 3064 if (bufsize > sb_max) 3065 bufsize = sb_max; 3066 if (bufsize > so->so_rcv.sb_hiwat) 3067 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3068 } 3069 SOCKBUF_UNLOCK(&so->so_rcv); 3070 /* 3071 * While we're here, check the others too 3072 */ 3073 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3074 tp->t_srtt = rtt; 3075 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3076 tcpstat.tcps_usedrtt++; 3077 if (metrics.rmx_rttvar) { 3078 tp->t_rttvar = metrics.rmx_rttvar; 3079 tcpstat.tcps_usedrttvar++; 3080 } else { 3081 /* default variation is +- 1 rtt */ 3082 tp->t_rttvar = 3083 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3084 } 3085 TCPT_RANGESET(tp->t_rxtcur, 3086 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3087 tp->t_rttmin, TCPTV_REXMTMAX); 3088 } 3089 if (metrics.rmx_ssthresh) { 3090 /* 3091 * There's some sort of gateway or interface 3092 * buffer limit on the path. Use this to set 3093 * the slow start threshhold, but set the 3094 * threshold to no less than 2*mss. 3095 */ 3096 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3097 tcpstat.tcps_usedssthresh++; 3098 } 3099 if (metrics.rmx_bandwidth) 3100 tp->snd_bandwidth = metrics.rmx_bandwidth; 3101 3102 /* 3103 * Set the slow-start flight size depending on whether this 3104 * is a local network or not. 3105 * 3106 * Extend this so we cache the cwnd too and retrieve it here. 3107 * Make cwnd even bigger than RFC3390 suggests but only if we 3108 * have previous experience with the remote host. Be careful 3109 * not make cwnd bigger than remote receive window or our own 3110 * send socket buffer. Maybe put some additional upper bound 3111 * on the retrieved cwnd. Should do incremental updates to 3112 * hostcache when cwnd collapses so next connection doesn't 3113 * overloads the path again. 3114 * 3115 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3116 * We currently check only in syncache_socket for that. 3117 */ 3118 #define TCP_METRICS_CWND 3119 #ifdef TCP_METRICS_CWND 3120 if (metrics.rmx_cwnd) 3121 tp->snd_cwnd = max(mss, 3122 min(metrics.rmx_cwnd / 2, 3123 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3124 else 3125 #endif 3126 if (tcp_do_rfc3390) 3127 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3128 #ifdef INET6 3129 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3130 (!isipv6 && in_localaddr(inp->inp_faddr))) 3131 #else 3132 else if (in_localaddr(inp->inp_faddr)) 3133 #endif 3134 tp->snd_cwnd = mss * ss_fltsz_local; 3135 else 3136 tp->snd_cwnd = mss * ss_fltsz; 3137 } 3138 3139 /* 3140 * Determine the MSS option to send on an outgoing SYN. 3141 */ 3142 int 3143 tcp_mssopt(inc) 3144 struct in_conninfo *inc; 3145 { 3146 int mss = 0; 3147 u_long maxmtu = 0; 3148 u_long thcmtu = 0; 3149 size_t min_protoh; 3150 #ifdef INET6 3151 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3152 #endif 3153 3154 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3155 3156 #ifdef INET6 3157 if (isipv6) { 3158 mss = tcp_v6mssdflt; 3159 maxmtu = tcp_maxmtu6(inc); 3160 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3161 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3162 } else 3163 #endif 3164 { 3165 mss = tcp_mssdflt; 3166 maxmtu = tcp_maxmtu(inc); 3167 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3168 min_protoh = sizeof(struct tcpiphdr); 3169 } 3170 if (maxmtu && thcmtu) 3171 mss = min(maxmtu, thcmtu) - min_protoh; 3172 else if (maxmtu || thcmtu) 3173 mss = max(maxmtu, thcmtu) - min_protoh; 3174 3175 return (mss); 3176 } 3177 3178 3179 /* 3180 * On a partial ack arrives, force the retransmission of the 3181 * next unacknowledged segment. Do not clear tp->t_dupacks. 3182 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3183 * be started again. 3184 */ 3185 static void 3186 tcp_newreno_partial_ack(tp, th) 3187 struct tcpcb *tp; 3188 struct tcphdr *th; 3189 { 3190 tcp_seq onxt = tp->snd_nxt; 3191 u_long ocwnd = tp->snd_cwnd; 3192 3193 callout_stop(tp->tt_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 has not yet been updated when this function is called.) 3199 */ 3200 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3201 tp->t_flags |= TF_ACKNOW; 3202 (void) tcp_output(tp); 3203 tp->snd_cwnd = ocwnd; 3204 if (SEQ_GT(onxt, tp->snd_nxt)) 3205 tp->snd_nxt = onxt; 3206 /* 3207 * Partial window deflation. Relies on fact that tp->snd_una 3208 * not updated yet. 3209 */ 3210 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3211 } 3212 3213 /* 3214 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3215 * looking for a pcb in the listen state. Returns 0 otherwise. 3216 */ 3217 static int 3218 tcp_timewait(tw, to, th, m, tlen) 3219 struct tcptw *tw; 3220 struct tcpopt *to; 3221 struct tcphdr *th; 3222 struct mbuf *m; 3223 int tlen; 3224 { 3225 int thflags; 3226 tcp_seq seq; 3227 #ifdef INET6 3228 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3229 #else 3230 const int isipv6 = 0; 3231 #endif 3232 3233 thflags = th->th_flags; 3234 3235 /* 3236 * NOTE: for FIN_WAIT_2 (to be added later), 3237 * must validate sequence number before accepting RST 3238 */ 3239 3240 /* 3241 * If the segment contains RST: 3242 * Drop the segment - see Stevens, vol. 2, p. 964 and 3243 * RFC 1337. 3244 */ 3245 if (thflags & TH_RST) 3246 goto drop; 3247 3248 /* 3249 * If segment contains a SYN and CC [not CC.NEW] option: 3250 * if connection duration > MSL, drop packet and send RST; 3251 * 3252 * if SEG.CC > CCrecv then is new SYN. 3253 * Complete close and delete TCPCB. Then reprocess 3254 * segment, hoping to find new TCPCB in LISTEN state; 3255 * 3256 * else must be old SYN; drop it. 3257 * else do normal processing. 3258 */ 3259 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 3260 if ((ticks - tw->t_starttime) > tcp_msl) 3261 goto reset; 3262 if (CC_GT(to->to_cc, tw->cc_recv)) { 3263 (void) tcp_twclose(tw, 0); 3264 return (1); 3265 } 3266 goto drop; 3267 } 3268 3269 #if 0 3270 /* PAWS not needed at the moment */ 3271 /* 3272 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3273 * and it's less than ts_recent, drop it. 3274 */ 3275 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3276 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3277 if ((thflags & TH_ACK) == 0) 3278 goto drop; 3279 goto ack; 3280 } 3281 /* 3282 * ts_recent is never updated because we never accept new segments. 3283 */ 3284 #endif 3285 3286 /* 3287 * If a new connection request is received 3288 * while in TIME_WAIT, drop the old connection 3289 * and start over if the sequence numbers 3290 * are above the previous ones. 3291 */ 3292 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3293 (void) tcp_twclose(tw, 0); 3294 return (1); 3295 } 3296 3297 /* 3298 * Drop the the segment if it does not contain an ACK. 3299 */ 3300 if ((thflags & TH_ACK) == 0) 3301 goto drop; 3302 3303 /* 3304 * Reset the 2MSL timer if this is a duplicate FIN. 3305 */ 3306 if (thflags & TH_FIN) { 3307 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3308 if (seq + 1 == tw->rcv_nxt) 3309 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3310 } 3311 3312 /* 3313 * Acknowledge the segment if it has data or is not a duplicate ACK. 3314 */ 3315 if (thflags != TH_ACK || tlen != 0 || 3316 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3317 tcp_twrespond(tw, TH_ACK); 3318 goto drop; 3319 3320 reset: 3321 /* 3322 * Generate a RST, dropping incoming segment. 3323 * Make ACK acceptable to originator of segment. 3324 * Don't bother to respond if destination was broadcast/multicast. 3325 */ 3326 if (m->m_flags & (M_BCAST|M_MCAST)) 3327 goto drop; 3328 if (isipv6) { 3329 struct ip6_hdr *ip6; 3330 3331 /* IPv6 anycast check is done at tcp6_input() */ 3332 ip6 = mtod(m, struct ip6_hdr *); 3333 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3334 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3335 goto drop; 3336 } else { 3337 struct ip *ip; 3338 3339 ip = mtod(m, struct ip *); 3340 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3341 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3342 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3343 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3344 goto drop; 3345 } 3346 if (thflags & TH_ACK) { 3347 tcp_respond(NULL, 3348 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3349 } else { 3350 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3351 tcp_respond(NULL, 3352 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3353 } 3354 INP_UNLOCK(tw->tw_inpcb); 3355 return (0); 3356 3357 drop: 3358 INP_UNLOCK(tw->tw_inpcb); 3359 m_freem(m); 3360 return (0); 3361 } 3362