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 struct sockaddr_in *next_hop = NULL; 433 int rstreason; /* For badport_bandlim accounting purposes */ 434 435 struct ip6_hdr *ip6 = NULL; 436 #ifdef INET6 437 int isipv6; 438 #else 439 const int isipv6 = 0; 440 #endif 441 442 #ifdef TCPDEBUG 443 /* 444 * The size of tcp_saveipgen must be the size of the max ip header, 445 * now IPv6. 446 */ 447 u_char tcp_saveipgen[40]; 448 struct tcphdr tcp_savetcp; 449 short ostate = 0; 450 #endif 451 452 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 453 next_hop = m_claim_next(m, PACKET_TAG_IPFORWARD); 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 /* IPFIREWALL_FORWARD section */ 615 if (next_hop != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 616 /* 617 * Transparently forwarded. Pretend to be the destination. 618 * already got one like this? 619 */ 620 inp = in_pcblookup_hash(&tcbinfo, ip->ip_src, th->th_sport, 621 ip->ip_dst, th->th_dport, 622 0, m->m_pkthdr.rcvif); 623 if (!inp) { 624 /* It's new. Try find the ambushing socket. */ 625 inp = in_pcblookup_hash(&tcbinfo, 626 ip->ip_src, th->th_sport, 627 next_hop->sin_addr, 628 next_hop->sin_port ? 629 ntohs(next_hop->sin_port) : 630 th->th_dport, 631 1, m->m_pkthdr.rcvif); 632 } 633 } else { 634 if (isipv6) { 635 #ifdef INET6 636 inp = in6_pcblookup_hash(&tcbinfo, 637 &ip6->ip6_src, th->th_sport, 638 &ip6->ip6_dst, th->th_dport, 639 1, m->m_pkthdr.rcvif); 640 #endif 641 } else 642 inp = in_pcblookup_hash(&tcbinfo, 643 ip->ip_src, th->th_sport, 644 ip->ip_dst, th->th_dport, 645 1, m->m_pkthdr.rcvif); 646 } 647 648 #if defined(IPSEC) || defined(FAST_IPSEC) 649 #ifdef INET6 650 if (isipv6) { 651 if (inp != NULL && ipsec6_in_reject(m, inp)) { 652 #ifdef IPSEC 653 ipsec6stat.in_polvio++; 654 #endif 655 goto drop; 656 } 657 } else 658 #endif /* INET6 */ 659 if (inp != NULL && ipsec4_in_reject(m, inp)) { 660 #ifdef IPSEC 661 ipsecstat.in_polvio++; 662 #endif 663 goto drop; 664 } 665 #endif /*IPSEC || FAST_IPSEC*/ 666 667 /* 668 * If the state is CLOSED (i.e., TCB does not exist) then 669 * all data in the incoming segment is discarded. 670 * If the TCB exists but is in CLOSED state, it is embryonic, 671 * but should either do a listen or a connect soon. 672 */ 673 if (inp == NULL) { 674 if (log_in_vain) { 675 #ifdef INET6 676 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 677 #else 678 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 679 #endif 680 681 if (isipv6) { 682 #ifdef INET6 683 strcpy(dbuf, "["); 684 strcpy(sbuf, "["); 685 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 686 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 687 strcat(dbuf, "]"); 688 strcat(sbuf, "]"); 689 #endif 690 } else { 691 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 692 strcpy(sbuf, inet_ntoa(ip->ip_src)); 693 } 694 switch (log_in_vain) { 695 case 1: 696 if ((thflags & TH_SYN) == 0) 697 break; 698 /* FALLTHROUGH */ 699 case 2: 700 log(LOG_INFO, 701 "Connection attempt to TCP %s:%d " 702 "from %s:%d flags:0x%02x\n", 703 dbuf, ntohs(th->th_dport), sbuf, 704 ntohs(th->th_sport), thflags); 705 break; 706 default: 707 break; 708 } 709 } 710 if (blackhole) { 711 switch (blackhole) { 712 case 1: 713 if (thflags & TH_SYN) 714 goto drop; 715 break; 716 case 2: 717 goto drop; 718 default: 719 goto drop; 720 } 721 } 722 rstreason = BANDLIM_RST_CLOSEDPORT; 723 goto dropwithreset; 724 } 725 INP_LOCK(inp); 726 if (inp->inp_vflag & INP_TIMEWAIT) { 727 /* 728 * The only option of relevance is TOF_CC, and only if 729 * present in a SYN segment. See tcp_timewait(). 730 */ 731 if (thflags & TH_SYN) 732 tcp_dooptions((struct tcpcb *)NULL, &to, optp, optlen, 1, th); 733 if (tcp_timewait((struct tcptw *)inp->inp_ppcb, 734 &to, th, m, tlen)) 735 goto findpcb; 736 /* 737 * tcp_timewait unlocks inp. 738 */ 739 INP_INFO_WUNLOCK(&tcbinfo); 740 return; 741 } 742 tp = intotcpcb(inp); 743 if (tp == 0) { 744 INP_UNLOCK(inp); 745 rstreason = BANDLIM_RST_CLOSEDPORT; 746 goto dropwithreset; 747 } 748 if (tp->t_state == TCPS_CLOSED) 749 goto drop; 750 751 /* Unscale the window into a 32-bit value. */ 752 if ((thflags & TH_SYN) == 0) 753 tiwin = th->th_win << tp->snd_scale; 754 else 755 tiwin = th->th_win; 756 757 #ifdef MAC 758 INP_LOCK_ASSERT(inp); 759 if (mac_check_inpcb_deliver(inp, m)) 760 goto drop; 761 #endif 762 so = inp->inp_socket; 763 #ifdef TCPDEBUG 764 if (so->so_options & SO_DEBUG) { 765 ostate = tp->t_state; 766 if (isipv6) 767 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 768 else 769 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 770 tcp_savetcp = *th; 771 } 772 #endif 773 if (so->so_options & SO_ACCEPTCONN) { 774 struct in_conninfo inc; 775 776 #ifdef INET6 777 inc.inc_isipv6 = isipv6; 778 #endif 779 if (isipv6) { 780 inc.inc6_faddr = ip6->ip6_src; 781 inc.inc6_laddr = ip6->ip6_dst; 782 } else { 783 inc.inc_faddr = ip->ip_src; 784 inc.inc_laddr = ip->ip_dst; 785 } 786 inc.inc_fport = th->th_sport; 787 inc.inc_lport = th->th_dport; 788 789 /* 790 * If the state is LISTEN then ignore segment if it contains 791 * a RST. If the segment contains an ACK then it is bad and 792 * send a RST. If it does not contain a SYN then it is not 793 * interesting; drop it. 794 * 795 * If the state is SYN_RECEIVED (syncache) and seg contains 796 * an ACK, but not for our SYN/ACK, send a RST. If the seg 797 * contains a RST, check the sequence number to see if it 798 * is a valid reset segment. 799 */ 800 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 801 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 802 if (!syncache_expand(&inc, th, &so, m)) { 803 /* 804 * No syncache entry, or ACK was not 805 * for our SYN/ACK. Send a RST. 806 */ 807 tcpstat.tcps_badsyn++; 808 rstreason = BANDLIM_RST_OPENPORT; 809 goto dropwithreset; 810 } 811 if (so == NULL) { 812 /* 813 * Could not complete 3-way handshake, 814 * connection is being closed down, and 815 * syncache will free mbuf. 816 */ 817 INP_UNLOCK(inp); 818 INP_INFO_WUNLOCK(&tcbinfo); 819 return; 820 } 821 /* 822 * Socket is created in state SYN_RECEIVED. 823 * Continue processing segment. 824 */ 825 INP_UNLOCK(inp); 826 inp = sotoinpcb(so); 827 INP_LOCK(inp); 828 tp = intotcpcb(inp); 829 /* 830 * This is what would have happened in 831 * tcp_output() when the SYN,ACK was sent. 832 */ 833 tp->snd_up = tp->snd_una; 834 tp->snd_max = tp->snd_nxt = tp->iss + 1; 835 tp->last_ack_sent = tp->rcv_nxt; 836 /* 837 * RFC1323: The window in SYN & SYN/ACK 838 * segments is never scaled. 839 */ 840 tp->snd_wnd = tiwin; /* unscaled */ 841 goto after_listen; 842 } 843 if (thflags & TH_RST) { 844 syncache_chkrst(&inc, th); 845 goto drop; 846 } 847 if (thflags & TH_ACK) { 848 syncache_badack(&inc); 849 tcpstat.tcps_badsyn++; 850 rstreason = BANDLIM_RST_OPENPORT; 851 goto dropwithreset; 852 } 853 goto drop; 854 } 855 856 /* 857 * Segment's flags are (SYN) or (SYN|FIN). 858 */ 859 #ifdef INET6 860 /* 861 * If deprecated address is forbidden, 862 * we do not accept SYN to deprecated interface 863 * address to prevent any new inbound connection from 864 * getting established. 865 * When we do not accept SYN, we send a TCP RST, 866 * with deprecated source address (instead of dropping 867 * it). We compromise it as it is much better for peer 868 * to send a RST, and RST will be the final packet 869 * for the exchange. 870 * 871 * If we do not forbid deprecated addresses, we accept 872 * the SYN packet. RFC2462 does not suggest dropping 873 * SYN in this case. 874 * If we decipher RFC2462 5.5.4, it says like this: 875 * 1. use of deprecated addr with existing 876 * communication is okay - "SHOULD continue to be 877 * used" 878 * 2. use of it with new communication: 879 * (2a) "SHOULD NOT be used if alternate address 880 * with sufficient scope is available" 881 * (2b) nothing mentioned otherwise. 882 * Here we fall into (2b) case as we have no choice in 883 * our source address selection - we must obey the peer. 884 * 885 * The wording in RFC2462 is confusing, and there are 886 * multiple description text for deprecated address 887 * handling - worse, they are not exactly the same. 888 * I believe 5.5.4 is the best one, so we follow 5.5.4. 889 */ 890 if (isipv6 && !ip6_use_deprecated) { 891 struct in6_ifaddr *ia6; 892 893 if ((ia6 = ip6_getdstifaddr(m)) && 894 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 895 INP_UNLOCK(inp); 896 tp = NULL; 897 rstreason = BANDLIM_RST_OPENPORT; 898 goto dropwithreset; 899 } 900 } 901 #endif 902 /* 903 * If it is from this socket, drop it, it must be forged. 904 * Don't bother responding if the destination was a broadcast. 905 */ 906 if (th->th_dport == th->th_sport) { 907 if (isipv6) { 908 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 909 &ip6->ip6_src)) 910 goto drop; 911 } else { 912 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 913 goto drop; 914 } 915 } 916 /* 917 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 918 * 919 * Note that it is quite possible to receive unicast 920 * link-layer packets with a broadcast IP address. Use 921 * in_broadcast() to find them. 922 */ 923 if (m->m_flags & (M_BCAST|M_MCAST)) 924 goto drop; 925 if (isipv6) { 926 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 927 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 928 goto drop; 929 } else { 930 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 931 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 932 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 933 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 934 goto drop; 935 } 936 /* 937 * SYN appears to be valid; create compressed TCP state 938 * for syncache, or perform t/tcp connection. 939 */ 940 if (so->so_qlen <= so->so_qlimit) { 941 #ifdef TCPDEBUG 942 if (so->so_options & SO_DEBUG) 943 tcp_trace(TA_INPUT, ostate, tp, 944 (void *)tcp_saveipgen, &tcp_savetcp, 0); 945 #endif 946 tcp_dooptions(tp, &to, optp, optlen, 1, th); 947 if (!syncache_add(&inc, &to, th, &so, m)) 948 goto drop; 949 if (so == NULL) { 950 /* 951 * Entry added to syncache, mbuf used to 952 * send SYN,ACK packet. 953 */ 954 KASSERT(headlocked, ("headlocked")); 955 INP_UNLOCK(inp); 956 INP_INFO_WUNLOCK(&tcbinfo); 957 return; 958 } 959 /* 960 * Segment passed TAO tests. 961 */ 962 INP_UNLOCK(inp); 963 inp = sotoinpcb(so); 964 INP_LOCK(inp); 965 tp = intotcpcb(inp); 966 tp->snd_wnd = tiwin; 967 tp->t_starttime = ticks; 968 tp->t_state = TCPS_ESTABLISHED; 969 970 /* 971 * T/TCP logic: 972 * If there is a FIN or if there is data, then 973 * delay SYN,ACK(SYN) in the hope of piggy-backing 974 * it on a response segment. Otherwise must send 975 * ACK now in case the other side is slow starting. 976 */ 977 if (thflags & TH_FIN || tlen != 0) 978 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 979 else 980 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 981 tcpstat.tcps_connects++; 982 soisconnected(so); 983 goto trimthenstep6; 984 } 985 goto drop; 986 } 987 after_listen: 988 KASSERT(headlocked, ("tcp_input(): after_listen head is not locked")); 989 INP_LOCK_ASSERT(inp); 990 991 /* XXX temp debugging */ 992 /* should not happen - syncache should pick up these connections */ 993 if (tp->t_state == TCPS_LISTEN) 994 panic("tcp_input: TCPS_LISTEN"); 995 996 /* 997 * This is the second part of the MSS DoS prevention code (after 998 * minmss on the sending side) and it deals with too many too small 999 * tcp packets in a too short timeframe (1 second). 1000 * 1001 * For every full second we count the number of received packets 1002 * and bytes. If we get a lot of packets per second for this connection 1003 * (tcp_minmssoverload) we take a closer look at it and compute the 1004 * average packet size for the past second. If that is less than 1005 * tcp_minmss we get too many packets with very small payload which 1006 * is not good and burdens our system (and every packet generates 1007 * a wakeup to the process connected to our socket). We can reasonable 1008 * expect this to be small packet DoS attack to exhaust our CPU 1009 * cycles. 1010 * 1011 * Care has to be taken for the minimum packet overload value. This 1012 * value defines the minimum number of packets per second before we 1013 * start to worry. This must not be too low to avoid killing for 1014 * example interactive connections with many small packets like 1015 * telnet or SSH. 1016 * 1017 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1018 * this check. 1019 * 1020 * Account for packet if payload packet, skip over ACK, etc. 1021 */ 1022 if (tcp_minmss && tcp_minmssoverload && 1023 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1024 if (tp->rcv_second > ticks) { 1025 tp->rcv_pps++; 1026 tp->rcv_byps += tlen + off; 1027 if (tp->rcv_pps > tcp_minmssoverload) { 1028 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1029 printf("too many small tcp packets from " 1030 "%s:%u, av. %lubyte/packet, " 1031 "dropping connection\n", 1032 #ifdef INET6 1033 isipv6 ? 1034 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1035 #endif 1036 inet_ntoa(inp->inp_inc.inc_faddr), 1037 inp->inp_inc.inc_fport, 1038 tp->rcv_byps / tp->rcv_pps); 1039 tp = tcp_drop(tp, ECONNRESET); 1040 tcpstat.tcps_minmssdrops++; 1041 goto drop; 1042 } 1043 } 1044 } else { 1045 tp->rcv_second = ticks + hz; 1046 tp->rcv_pps = 1; 1047 tp->rcv_byps = tlen + off; 1048 } 1049 } 1050 1051 /* 1052 * Segment received on connection. 1053 * Reset idle time and keep-alive timer. 1054 */ 1055 tp->t_rcvtime = ticks; 1056 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1057 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1058 1059 /* 1060 * Process options only when we get SYN/ACK back. The SYN case 1061 * for incoming connections is handled in tcp_syncache. 1062 * XXX this is traditional behavior, may need to be cleaned up. 1063 */ 1064 tcp_dooptions(tp, &to, optp, optlen, thflags & TH_SYN, th); 1065 if (thflags & TH_SYN) { 1066 if (to.to_flags & TOF_SCALE) { 1067 tp->t_flags |= TF_RCVD_SCALE; 1068 tp->requested_s_scale = to.to_requested_s_scale; 1069 } 1070 if (to.to_flags & TOF_TS) { 1071 tp->t_flags |= TF_RCVD_TSTMP; 1072 tp->ts_recent = to.to_tsval; 1073 tp->ts_recent_age = ticks; 1074 } 1075 if (to.to_flags & (TOF_CC|TOF_CCNEW)) 1076 tp->t_flags |= TF_RCVD_CC; 1077 if (to.to_flags & TOF_MSS) 1078 tcp_mss(tp, to.to_mss); 1079 if (tp->sack_enable) { 1080 if (!(to.to_flags & TOF_SACK)) 1081 tp->sack_enable = 0; 1082 else 1083 tp->t_flags |= TF_SACK_PERMIT; 1084 } 1085 1086 } 1087 1088 if (tp->sack_enable) { 1089 /* Delete stale (cumulatively acked) SACK holes */ 1090 tcp_del_sackholes(tp, th); 1091 tp->rcv_laststart = th->th_seq; /* last recv'd segment*/ 1092 tp->rcv_lastend = th->th_seq + tlen; 1093 } 1094 1095 /* 1096 * Header prediction: check for the two common cases 1097 * of a uni-directional data xfer. If the packet has 1098 * no control flags, is in-sequence, the window didn't 1099 * change and we're not retransmitting, it's a 1100 * candidate. If the length is zero and the ack moved 1101 * forward, we're the sender side of the xfer. Just 1102 * free the data acked & wake any higher level process 1103 * that was blocked waiting for space. If the length 1104 * is non-zero and the ack didn't move, we're the 1105 * receiver side. If we're getting packets in-order 1106 * (the reassembly queue is empty), add the data to 1107 * the socket buffer and note that we need a delayed ack. 1108 * Make sure that the hidden state-flags are also off. 1109 * Since we check for TCPS_ESTABLISHED above, it can only 1110 * be TH_NEEDSYN. 1111 */ 1112 if (tp->t_state == TCPS_ESTABLISHED && 1113 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1114 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1115 ((to.to_flags & TOF_TS) == 0 || 1116 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1117 /* 1118 * Using the CC option is compulsory if once started: 1119 * the segment is OK if no T/TCP was negotiated or 1120 * if the segment has a CC option equal to CCrecv 1121 */ 1122 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) || 1123 ((to.to_flags & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) && 1124 th->th_seq == tp->rcv_nxt && 1125 tiwin && tiwin == tp->snd_wnd && 1126 tp->snd_nxt == tp->snd_max) { 1127 1128 /* 1129 * If last ACK falls within this segment's sequence numbers, 1130 * record the timestamp. 1131 * NOTE that the test is modified according to the latest 1132 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1133 */ 1134 if ((to.to_flags & TOF_TS) != 0 && 1135 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1136 tp->ts_recent_age = ticks; 1137 tp->ts_recent = to.to_tsval; 1138 } 1139 1140 if (tlen == 0) { 1141 if (SEQ_GT(th->th_ack, tp->snd_una) && 1142 SEQ_LEQ(th->th_ack, tp->snd_max) && 1143 tp->snd_cwnd >= tp->snd_wnd && 1144 ((!tcp_do_newreno && !tp->sack_enable && 1145 tp->t_dupacks < tcprexmtthresh) || 1146 ((tcp_do_newreno || tp->sack_enable) && 1147 !IN_FASTRECOVERY(tp)))) { 1148 KASSERT(headlocked, ("headlocked")); 1149 INP_INFO_WUNLOCK(&tcbinfo); 1150 /* 1151 * this is a pure ack for outstanding data. 1152 */ 1153 ++tcpstat.tcps_predack; 1154 /* 1155 * "bad retransmit" recovery 1156 */ 1157 if (tp->t_rxtshift == 1 && 1158 ticks < tp->t_badrxtwin) { 1159 ++tcpstat.tcps_sndrexmitbad; 1160 tp->snd_cwnd = tp->snd_cwnd_prev; 1161 tp->snd_ssthresh = 1162 tp->snd_ssthresh_prev; 1163 tp->snd_recover = tp->snd_recover_prev; 1164 if (tp->t_flags & TF_WASFRECOVERY) 1165 ENTER_FASTRECOVERY(tp); 1166 tp->snd_nxt = tp->snd_max; 1167 tp->t_badrxtwin = 0; 1168 } 1169 1170 /* 1171 * Recalculate the transmit timer / rtt. 1172 * 1173 * Some boxes send broken timestamp replies 1174 * during the SYN+ACK phase, ignore 1175 * timestamps of 0 or we could calculate a 1176 * huge RTT and blow up the retransmit timer. 1177 */ 1178 if ((to.to_flags & TOF_TS) != 0 && 1179 to.to_tsecr) { 1180 tcp_xmit_timer(tp, 1181 ticks - to.to_tsecr + 1); 1182 } else if (tp->t_rtttime && 1183 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1184 tcp_xmit_timer(tp, 1185 ticks - tp->t_rtttime); 1186 } 1187 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1188 acked = th->th_ack - tp->snd_una; 1189 tcpstat.tcps_rcvackpack++; 1190 tcpstat.tcps_rcvackbyte += acked; 1191 sbdrop(&so->so_snd, acked); 1192 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1193 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1194 tp->snd_recover = th->th_ack - 1; 1195 tp->snd_una = th->th_ack; 1196 /* 1197 * pull snd_wl2 up to prevent seq wrap relative 1198 * to th_ack. 1199 */ 1200 tp->snd_wl2 = th->th_ack; 1201 tp->t_dupacks = 0; 1202 m_freem(m); 1203 ND6_HINT(tp); /* some progress has been done */ 1204 1205 /* 1206 * If all outstanding data are acked, stop 1207 * retransmit timer, otherwise restart timer 1208 * using current (possibly backed-off) value. 1209 * If process is waiting for space, 1210 * wakeup/selwakeup/signal. If data 1211 * are ready to send, let tcp_output 1212 * decide between more output or persist. 1213 1214 #ifdef TCPDEBUG 1215 if (so->so_options & SO_DEBUG) 1216 tcp_trace(TA_INPUT, ostate, tp, 1217 (void *)tcp_saveipgen, 1218 &tcp_savetcp, 0); 1219 #endif 1220 */ 1221 if (tp->snd_una == tp->snd_max) 1222 callout_stop(tp->tt_rexmt); 1223 else if (!callout_active(tp->tt_persist)) 1224 callout_reset(tp->tt_rexmt, 1225 tp->t_rxtcur, 1226 tcp_timer_rexmt, tp); 1227 1228 sowwakeup(so); 1229 if (so->so_snd.sb_cc) 1230 (void) tcp_output(tp); 1231 goto check_delack; 1232 } 1233 } else if (th->th_ack == tp->snd_una && 1234 LIST_EMPTY(&tp->t_segq) && 1235 tlen <= sbspace(&so->so_rcv)) { 1236 KASSERT(headlocked, ("headlocked")); 1237 INP_INFO_WUNLOCK(&tcbinfo); 1238 /* 1239 * this is a pure, in-sequence data packet 1240 * with nothing on the reassembly queue and 1241 * we have enough buffer space to take it. 1242 */ 1243 /* Clean receiver SACK report if present */ 1244 if (tp->sack_enable && tp->rcv_numsacks) 1245 tcp_clean_sackreport(tp); 1246 ++tcpstat.tcps_preddat; 1247 tp->rcv_nxt += tlen; 1248 /* 1249 * Pull snd_wl1 up to prevent seq wrap relative to 1250 * th_seq. 1251 */ 1252 tp->snd_wl1 = th->th_seq; 1253 /* 1254 * Pull rcv_up up to prevent seq wrap relative to 1255 * rcv_nxt. 1256 */ 1257 tp->rcv_up = tp->rcv_nxt; 1258 tcpstat.tcps_rcvpack++; 1259 tcpstat.tcps_rcvbyte += tlen; 1260 ND6_HINT(tp); /* some progress has been done */ 1261 /* 1262 #ifdef TCPDEBUG 1263 if (so->so_options & SO_DEBUG) 1264 tcp_trace(TA_INPUT, ostate, tp, 1265 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1266 #endif 1267 * Add data to socket buffer. 1268 */ 1269 /* Unlocked read. */ 1270 SOCKBUF_LOCK(&so->so_rcv); 1271 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1272 m_freem(m); 1273 } else { 1274 m_adj(m, drop_hdrlen); /* delayed header drop */ 1275 sbappendstream_locked(&so->so_rcv, m); 1276 } 1277 sorwakeup_locked(so); 1278 if (DELAY_ACK(tp)) { 1279 tp->t_flags |= TF_DELACK; 1280 } else { 1281 tp->t_flags |= TF_ACKNOW; 1282 tcp_output(tp); 1283 } 1284 goto check_delack; 1285 } 1286 } 1287 1288 /* 1289 * Calculate amount of space in receive window, 1290 * and then do TCP input processing. 1291 * Receive window is amount of space in rcv queue, 1292 * but not less than advertised window. 1293 */ 1294 { int win; 1295 1296 win = sbspace(&so->so_rcv); 1297 if (win < 0) 1298 win = 0; 1299 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1300 } 1301 1302 switch (tp->t_state) { 1303 1304 /* 1305 * If the state is SYN_RECEIVED: 1306 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1307 */ 1308 case TCPS_SYN_RECEIVED: 1309 if ((thflags & TH_ACK) && 1310 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1311 SEQ_GT(th->th_ack, tp->snd_max))) { 1312 rstreason = BANDLIM_RST_OPENPORT; 1313 goto dropwithreset; 1314 } 1315 break; 1316 1317 /* 1318 * If the state is SYN_SENT: 1319 * if seg contains an ACK, but not for our SYN, drop the input. 1320 * if seg contains a RST, then drop the connection. 1321 * if seg does not contain SYN, then drop it. 1322 * Otherwise this is an acceptable SYN segment 1323 * initialize tp->rcv_nxt and tp->irs 1324 * if seg contains ack then advance tp->snd_una 1325 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1326 * arrange for segment to be acked (eventually) 1327 * continue processing rest of data/controls, beginning with URG 1328 */ 1329 case TCPS_SYN_SENT: 1330 if (tcp_do_rfc1644) 1331 tcp_hc_gettao(&inp->inp_inc, &tao); 1332 1333 if ((thflags & TH_ACK) && 1334 (SEQ_LEQ(th->th_ack, tp->iss) || 1335 SEQ_GT(th->th_ack, tp->snd_max))) { 1336 /* 1337 * If we have a cached CCsent for the remote host, 1338 * hence we haven't just crashed and restarted, 1339 * do not send a RST. This may be a retransmission 1340 * from the other side after our earlier ACK was lost. 1341 * Our new SYN, when it arrives, will serve as the 1342 * needed ACK. 1343 */ 1344 if (tao.tao_ccsent != 0) 1345 goto drop; 1346 else { 1347 rstreason = BANDLIM_UNLIMITED; 1348 goto dropwithreset; 1349 } 1350 } 1351 if (thflags & TH_RST) { 1352 if (thflags & TH_ACK) 1353 tp = tcp_drop(tp, ECONNREFUSED); 1354 goto drop; 1355 } 1356 if ((thflags & TH_SYN) == 0) 1357 goto drop; 1358 tp->snd_wnd = th->th_win; /* initial send window */ 1359 tp->cc_recv = to.to_cc; /* foreign CC */ 1360 1361 tp->irs = th->th_seq; 1362 tcp_rcvseqinit(tp); 1363 if (thflags & TH_ACK) { 1364 /* 1365 * Our SYN was acked. If segment contains CC.ECHO 1366 * option, check it to make sure this segment really 1367 * matches our SYN. If not, just drop it as old 1368 * duplicate, but send an RST if we're still playing 1369 * by the old rules. If no CC.ECHO option, make sure 1370 * we don't get fooled into using T/TCP. 1371 */ 1372 if (to.to_flags & TOF_CCECHO) { 1373 if (tp->cc_send != to.to_ccecho) { 1374 if (tao.tao_ccsent != 0) 1375 goto drop; 1376 else { 1377 rstreason = BANDLIM_UNLIMITED; 1378 goto dropwithreset; 1379 } 1380 } 1381 } else 1382 tp->t_flags &= ~TF_RCVD_CC; 1383 tcpstat.tcps_connects++; 1384 soisconnected(so); 1385 #ifdef MAC 1386 SOCK_LOCK(so); 1387 mac_set_socket_peer_from_mbuf(m, so); 1388 SOCK_UNLOCK(so); 1389 #endif 1390 /* Do window scaling on this connection? */ 1391 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1392 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1393 tp->snd_scale = tp->requested_s_scale; 1394 tp->rcv_scale = tp->request_r_scale; 1395 } 1396 /* Segment is acceptable, update cache if undefined. */ 1397 if (tao.tao_ccsent == 0 && tcp_do_rfc1644) 1398 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CCSENT, to.to_ccecho, 0); 1399 1400 tp->rcv_adv += tp->rcv_wnd; 1401 tp->snd_una++; /* SYN is acked */ 1402 /* 1403 * If there's data, delay ACK; if there's also a FIN 1404 * ACKNOW will be turned on later. 1405 */ 1406 if (DELAY_ACK(tp) && tlen != 0) 1407 callout_reset(tp->tt_delack, tcp_delacktime, 1408 tcp_timer_delack, tp); 1409 else 1410 tp->t_flags |= TF_ACKNOW; 1411 /* 1412 * Received <SYN,ACK> in SYN_SENT[*] state. 1413 * Transitions: 1414 * SYN_SENT --> ESTABLISHED 1415 * SYN_SENT* --> FIN_WAIT_1 1416 */ 1417 tp->t_starttime = ticks; 1418 if (tp->t_flags & TF_NEEDFIN) { 1419 tp->t_state = TCPS_FIN_WAIT_1; 1420 tp->t_flags &= ~TF_NEEDFIN; 1421 thflags &= ~TH_SYN; 1422 } else { 1423 tp->t_state = TCPS_ESTABLISHED; 1424 callout_reset(tp->tt_keep, tcp_keepidle, 1425 tcp_timer_keep, tp); 1426 } 1427 } else { 1428 /* 1429 * Received initial SYN in SYN-SENT[*] state => 1430 * simultaneous open. If segment contains CC option 1431 * and there is a cached CC, apply TAO test. 1432 * If it succeeds, connection is * half-synchronized. 1433 * Otherwise, do 3-way handshake: 1434 * SYN-SENT -> SYN-RECEIVED 1435 * SYN-SENT* -> SYN-RECEIVED* 1436 * If there was no CC option, clear cached CC value. 1437 */ 1438 tp->t_flags |= TF_ACKNOW; 1439 callout_stop(tp->tt_rexmt); 1440 if (to.to_flags & TOF_CC) { 1441 if (tao.tao_cc != 0 && 1442 CC_GT(to.to_cc, tao.tao_cc)) { 1443 /* 1444 * update cache and make transition: 1445 * SYN-SENT -> ESTABLISHED* 1446 * SYN-SENT* -> FIN-WAIT-1* 1447 */ 1448 tao.tao_cc = to.to_cc; 1449 tcp_hc_updatetao(&inp->inp_inc, 1450 TCP_HC_TAO_CC, to.to_cc, 0); 1451 tp->t_starttime = ticks; 1452 if (tp->t_flags & TF_NEEDFIN) { 1453 tp->t_state = TCPS_FIN_WAIT_1; 1454 tp->t_flags &= ~TF_NEEDFIN; 1455 } else { 1456 tp->t_state = TCPS_ESTABLISHED; 1457 callout_reset(tp->tt_keep, 1458 tcp_keepidle, 1459 tcp_timer_keep, 1460 tp); 1461 } 1462 tp->t_flags |= TF_NEEDSYN; 1463 } else 1464 tp->t_state = TCPS_SYN_RECEIVED; 1465 } else { 1466 if (tcp_do_rfc1644) { 1467 /* CC.NEW or no option => invalidate cache */ 1468 tao.tao_cc = 0; 1469 tcp_hc_updatetao(&inp->inp_inc, 1470 TCP_HC_TAO_CC, to.to_cc, 0); 1471 } 1472 tp->t_state = TCPS_SYN_RECEIVED; 1473 } 1474 } 1475 1476 trimthenstep6: 1477 KASSERT(headlocked, 1478 ("tcp_input(): trimthenstep6 head is not locked")); 1479 INP_LOCK_ASSERT(inp); 1480 1481 /* 1482 * Advance th->th_seq to correspond to first data byte. 1483 * If data, trim to stay within window, 1484 * dropping FIN if necessary. 1485 */ 1486 th->th_seq++; 1487 if (tlen > tp->rcv_wnd) { 1488 todrop = tlen - tp->rcv_wnd; 1489 m_adj(m, -todrop); 1490 tlen = tp->rcv_wnd; 1491 thflags &= ~TH_FIN; 1492 tcpstat.tcps_rcvpackafterwin++; 1493 tcpstat.tcps_rcvbyteafterwin += todrop; 1494 } 1495 tp->snd_wl1 = th->th_seq - 1; 1496 tp->rcv_up = th->th_seq; 1497 /* 1498 * Client side of transaction: already sent SYN and data. 1499 * If the remote host used T/TCP to validate the SYN, 1500 * our data will be ACK'd; if so, enter normal data segment 1501 * processing in the middle of step 5, ack processing. 1502 * Otherwise, goto step 6. 1503 */ 1504 if (thflags & TH_ACK) 1505 goto process_ACK; 1506 1507 goto step6; 1508 1509 /* 1510 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1511 * if segment contains a SYN and CC [not CC.NEW] option: 1512 * if state == TIME_WAIT and connection duration > MSL, 1513 * drop packet and send RST; 1514 * 1515 * if SEG.CC > CCrecv then is new SYN, and can implicitly 1516 * ack the FIN (and data) in retransmission queue. 1517 * Complete close and delete TCPCB. Then reprocess 1518 * segment, hoping to find new TCPCB in LISTEN state; 1519 * 1520 * else must be old SYN; drop it. 1521 * else do normal processing. 1522 */ 1523 case TCPS_LAST_ACK: 1524 case TCPS_CLOSING: 1525 case TCPS_TIME_WAIT: 1526 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1527 if ((thflags & TH_SYN) && 1528 (to.to_flags & TOF_CC) && tp->cc_recv != 0) { 1529 if (tp->t_state == TCPS_TIME_WAIT && 1530 (ticks - tp->t_starttime) > tcp_msl) { 1531 rstreason = BANDLIM_UNLIMITED; 1532 goto dropwithreset; 1533 } 1534 if (CC_GT(to.to_cc, tp->cc_recv)) { 1535 tp = tcp_close(tp); 1536 goto findpcb; 1537 } 1538 else 1539 goto drop; 1540 } 1541 break; /* continue normal processing */ 1542 } 1543 1544 /* 1545 * States other than LISTEN or SYN_SENT. 1546 * First check the RST flag and sequence number since reset segments 1547 * are exempt from the timestamp and connection count tests. This 1548 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1549 * below which allowed reset segments in half the sequence space 1550 * to fall though and be processed (which gives forged reset 1551 * segments with a random sequence number a 50 percent chance of 1552 * killing a connection). 1553 * Then check timestamp, if present. 1554 * Then check the connection count, if present. 1555 * Then check that at least some bytes of segment are within 1556 * receive window. If segment begins before rcv_nxt, 1557 * drop leading data (and SYN); if nothing left, just ack. 1558 * 1559 * 1560 * If the RST bit is set, check the sequence number to see 1561 * if this is a valid reset segment. 1562 * RFC 793 page 37: 1563 * In all states except SYN-SENT, all reset (RST) segments 1564 * are validated by checking their SEQ-fields. A reset is 1565 * valid if its sequence number is in the window. 1566 * Note: this does not take into account delayed ACKs, so 1567 * we should test against last_ack_sent instead of rcv_nxt. 1568 * The sequence number in the reset segment is normally an 1569 * echo of our outgoing acknowlegement numbers, but some hosts 1570 * send a reset with the sequence number at the rightmost edge 1571 * of our receive window, and we have to handle this case. 1572 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1573 * that brute force RST attacks are possible. To combat this, 1574 * we use a much stricter check while in the ESTABLISHED state, 1575 * only accepting RSTs where the sequence number is equal to 1576 * last_ack_sent. In all other states (the states in which a 1577 * RST is more likely), the more permissive check is used. 1578 * If we have multiple segments in flight, the intial reset 1579 * segment sequence numbers will be to the left of last_ack_sent, 1580 * but they will eventually catch up. 1581 * In any case, it never made sense to trim reset segments to 1582 * fit the receive window since RFC 1122 says: 1583 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1584 * 1585 * A TCP SHOULD allow a received RST segment to include data. 1586 * 1587 * DISCUSSION 1588 * It has been suggested that a RST segment could contain 1589 * ASCII text that encoded and explained the cause of the 1590 * RST. No standard has yet been established for such 1591 * data. 1592 * 1593 * If the reset segment passes the sequence number test examine 1594 * the state: 1595 * SYN_RECEIVED STATE: 1596 * If passive open, return to LISTEN state. 1597 * If active open, inform user that connection was refused. 1598 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1599 * Inform user that connection was reset, and close tcb. 1600 * CLOSING, LAST_ACK STATES: 1601 * Close the tcb. 1602 * TIME_WAIT STATE: 1603 * Drop the segment - see Stevens, vol. 2, p. 964 and 1604 * RFC 1337. 1605 */ 1606 if (thflags & TH_RST) { 1607 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1608 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1609 switch (tp->t_state) { 1610 1611 case TCPS_SYN_RECEIVED: 1612 so->so_error = ECONNREFUSED; 1613 goto close; 1614 1615 case TCPS_ESTABLISHED: 1616 if (tp->last_ack_sent != th->th_seq) { 1617 tcpstat.tcps_badrst++; 1618 goto drop; 1619 } 1620 case TCPS_FIN_WAIT_1: 1621 case TCPS_FIN_WAIT_2: 1622 case TCPS_CLOSE_WAIT: 1623 so->so_error = ECONNRESET; 1624 close: 1625 tp->t_state = TCPS_CLOSED; 1626 tcpstat.tcps_drops++; 1627 tp = tcp_close(tp); 1628 break; 1629 1630 case TCPS_CLOSING: 1631 case TCPS_LAST_ACK: 1632 tp = tcp_close(tp); 1633 break; 1634 1635 case TCPS_TIME_WAIT: 1636 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1637 ("timewait")); 1638 break; 1639 } 1640 } 1641 goto drop; 1642 } 1643 1644 /* 1645 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1646 * and it's less than ts_recent, drop it. 1647 */ 1648 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1649 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1650 1651 /* Check to see if ts_recent is over 24 days old. */ 1652 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1653 /* 1654 * Invalidate ts_recent. If this segment updates 1655 * ts_recent, the age will be reset later and ts_recent 1656 * will get a valid value. If it does not, setting 1657 * ts_recent to zero will at least satisfy the 1658 * requirement that zero be placed in the timestamp 1659 * echo reply when ts_recent isn't valid. The 1660 * age isn't reset until we get a valid ts_recent 1661 * because we don't want out-of-order segments to be 1662 * dropped when ts_recent is old. 1663 */ 1664 tp->ts_recent = 0; 1665 } else { 1666 tcpstat.tcps_rcvduppack++; 1667 tcpstat.tcps_rcvdupbyte += tlen; 1668 tcpstat.tcps_pawsdrop++; 1669 if (tlen) 1670 goto dropafterack; 1671 goto drop; 1672 } 1673 } 1674 1675 /* 1676 * T/TCP mechanism 1677 * If T/TCP was negotiated and the segment doesn't have CC, 1678 * or if its CC is wrong then drop the segment. 1679 * RST segments do not have to comply with this. 1680 */ 1681 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) && 1682 ((to.to_flags & TOF_CC) == 0 || tp->cc_recv != to.to_cc)) 1683 goto dropafterack; 1684 1685 /* 1686 * In the SYN-RECEIVED state, validate that the packet belongs to 1687 * this connection before trimming the data to fit the receive 1688 * window. Check the sequence number versus IRS since we know 1689 * the sequence numbers haven't wrapped. This is a partial fix 1690 * for the "LAND" DoS attack. 1691 */ 1692 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1693 rstreason = BANDLIM_RST_OPENPORT; 1694 goto dropwithreset; 1695 } 1696 1697 todrop = tp->rcv_nxt - th->th_seq; 1698 if (todrop > 0) { 1699 if (thflags & TH_SYN) { 1700 thflags &= ~TH_SYN; 1701 th->th_seq++; 1702 if (th->th_urp > 1) 1703 th->th_urp--; 1704 else 1705 thflags &= ~TH_URG; 1706 todrop--; 1707 } 1708 /* 1709 * Following if statement from Stevens, vol. 2, p. 960. 1710 */ 1711 if (todrop > tlen 1712 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1713 /* 1714 * Any valid FIN must be to the left of the window. 1715 * At this point the FIN must be a duplicate or out 1716 * of sequence; drop it. 1717 */ 1718 thflags &= ~TH_FIN; 1719 1720 /* 1721 * Send an ACK to resynchronize and drop any data. 1722 * But keep on processing for RST or ACK. 1723 */ 1724 tp->t_flags |= TF_ACKNOW; 1725 todrop = tlen; 1726 tcpstat.tcps_rcvduppack++; 1727 tcpstat.tcps_rcvdupbyte += todrop; 1728 } else { 1729 tcpstat.tcps_rcvpartduppack++; 1730 tcpstat.tcps_rcvpartdupbyte += todrop; 1731 } 1732 drop_hdrlen += todrop; /* drop from the top afterwards */ 1733 th->th_seq += todrop; 1734 tlen -= todrop; 1735 if (th->th_urp > todrop) 1736 th->th_urp -= todrop; 1737 else { 1738 thflags &= ~TH_URG; 1739 th->th_urp = 0; 1740 } 1741 } 1742 1743 /* 1744 * If new data are received on a connection after the 1745 * user processes are gone, then RST the other end. 1746 */ 1747 if ((so->so_state & SS_NOFDREF) && 1748 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1749 tp = tcp_close(tp); 1750 tcpstat.tcps_rcvafterclose++; 1751 rstreason = BANDLIM_UNLIMITED; 1752 goto dropwithreset; 1753 } 1754 1755 /* 1756 * If segment ends after window, drop trailing data 1757 * (and PUSH and FIN); if nothing left, just ACK. 1758 */ 1759 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1760 if (todrop > 0) { 1761 tcpstat.tcps_rcvpackafterwin++; 1762 if (todrop >= tlen) { 1763 tcpstat.tcps_rcvbyteafterwin += tlen; 1764 /* 1765 * If a new connection request is received 1766 * while in TIME_WAIT, drop the old connection 1767 * and start over if the sequence numbers 1768 * are above the previous ones. 1769 */ 1770 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1771 if (thflags & TH_SYN && 1772 tp->t_state == TCPS_TIME_WAIT && 1773 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1774 tp = tcp_close(tp); 1775 goto findpcb; 1776 } 1777 /* 1778 * If window is closed can only take segments at 1779 * window edge, and have to drop data and PUSH from 1780 * incoming segments. Continue processing, but 1781 * remember to ack. Otherwise, drop segment 1782 * and ack. 1783 */ 1784 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1785 tp->t_flags |= TF_ACKNOW; 1786 tcpstat.tcps_rcvwinprobe++; 1787 } else 1788 goto dropafterack; 1789 } else 1790 tcpstat.tcps_rcvbyteafterwin += todrop; 1791 m_adj(m, -todrop); 1792 tlen -= todrop; 1793 thflags &= ~(TH_PUSH|TH_FIN); 1794 } 1795 1796 /* 1797 * If last ACK falls within this segment's sequence numbers, 1798 * record its timestamp. 1799 * NOTE that the test is modified according to the latest 1800 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1801 */ 1802 if ((to.to_flags & TOF_TS) != 0 && 1803 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1804 tp->ts_recent_age = ticks; 1805 tp->ts_recent = to.to_tsval; 1806 } 1807 1808 /* 1809 * If a SYN is in the window, then this is an 1810 * error and we send an RST and drop the connection. 1811 */ 1812 if (thflags & TH_SYN) { 1813 tp = tcp_drop(tp, ECONNRESET); 1814 rstreason = BANDLIM_UNLIMITED; 1815 goto drop; 1816 } 1817 1818 /* 1819 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1820 * flag is on (half-synchronized state), then queue data for 1821 * later processing; else drop segment and return. 1822 */ 1823 if ((thflags & TH_ACK) == 0) { 1824 if (tp->t_state == TCPS_SYN_RECEIVED || 1825 (tp->t_flags & TF_NEEDSYN)) 1826 goto step6; 1827 else 1828 goto drop; 1829 } 1830 1831 /* 1832 * Ack processing. 1833 */ 1834 switch (tp->t_state) { 1835 1836 /* 1837 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1838 * ESTABLISHED state and continue processing. 1839 * The ACK was checked above. 1840 */ 1841 case TCPS_SYN_RECEIVED: 1842 1843 tcpstat.tcps_connects++; 1844 soisconnected(so); 1845 /* Do window scaling? */ 1846 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1847 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1848 tp->snd_scale = tp->requested_s_scale; 1849 tp->rcv_scale = tp->request_r_scale; 1850 } 1851 /* 1852 * Upon successful completion of 3-way handshake, 1853 * update cache.CC, pass any queued data to the user, 1854 * and advance state appropriately. 1855 */ 1856 if (tcp_do_rfc1644) { 1857 tao.tao_cc = tp->cc_recv; 1858 tcp_hc_updatetao(&inp->inp_inc, TCP_HC_TAO_CC, 1859 tp->cc_recv, 0); 1860 } 1861 /* 1862 * Make transitions: 1863 * SYN-RECEIVED -> ESTABLISHED 1864 * SYN-RECEIVED* -> FIN-WAIT-1 1865 */ 1866 tp->t_starttime = ticks; 1867 if (tp->t_flags & TF_NEEDFIN) { 1868 tp->t_state = TCPS_FIN_WAIT_1; 1869 tp->t_flags &= ~TF_NEEDFIN; 1870 } else { 1871 tp->t_state = TCPS_ESTABLISHED; 1872 callout_reset(tp->tt_keep, tcp_keepidle, 1873 tcp_timer_keep, tp); 1874 } 1875 /* 1876 * If segment contains data or ACK, will call tcp_reass() 1877 * later; if not, do so now to pass queued data to user. 1878 */ 1879 if (tlen == 0 && (thflags & TH_FIN) == 0) 1880 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1881 (struct mbuf *)0); 1882 tp->snd_wl1 = th->th_seq - 1; 1883 /* FALLTHROUGH */ 1884 1885 /* 1886 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1887 * ACKs. If the ack is in the range 1888 * tp->snd_una < th->th_ack <= tp->snd_max 1889 * then advance tp->snd_una to th->th_ack and drop 1890 * data from the retransmission queue. If this ACK reflects 1891 * more up to date window information we update our window information. 1892 */ 1893 case TCPS_ESTABLISHED: 1894 case TCPS_FIN_WAIT_1: 1895 case TCPS_FIN_WAIT_2: 1896 case TCPS_CLOSE_WAIT: 1897 case TCPS_CLOSING: 1898 case TCPS_LAST_ACK: 1899 case TCPS_TIME_WAIT: 1900 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1901 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1902 if (tlen == 0 && tiwin == tp->snd_wnd) { 1903 tcpstat.tcps_rcvdupack++; 1904 /* 1905 * If we have outstanding data (other than 1906 * a window probe), this is a completely 1907 * duplicate ack (ie, window info didn't 1908 * change), the ack is the biggest we've 1909 * seen and we've seen exactly our rexmt 1910 * threshhold of them, assume a packet 1911 * has been dropped and retransmit it. 1912 * Kludge snd_nxt & the congestion 1913 * window so we send only this one 1914 * packet. 1915 * 1916 * We know we're losing at the current 1917 * window size so do congestion avoidance 1918 * (set ssthresh to half the current window 1919 * and pull our congestion window back to 1920 * the new ssthresh). 1921 * 1922 * Dup acks mean that packets have left the 1923 * network (they're now cached at the receiver) 1924 * so bump cwnd by the amount in the receiver 1925 * to keep a constant cwnd packets in the 1926 * network. 1927 */ 1928 if (!callout_active(tp->tt_rexmt) || 1929 th->th_ack != tp->snd_una) 1930 tp->t_dupacks = 0; 1931 else if (++tp->t_dupacks > tcprexmtthresh || 1932 ((tcp_do_newreno || tp->sack_enable) && 1933 IN_FASTRECOVERY(tp))) { 1934 tp->snd_cwnd += tp->t_maxseg; 1935 (void) tcp_output(tp); 1936 goto drop; 1937 } else if (tp->t_dupacks == tcprexmtthresh) { 1938 tcp_seq onxt = tp->snd_nxt; 1939 u_int win; 1940 1941 /* 1942 * If we're doing sack, check to 1943 * see if we're already in sack 1944 * recovery. If we're not doing sack, 1945 * check to see if we're in newreno 1946 * recovery. 1947 */ 1948 if (tp->sack_enable) { 1949 if (IN_FASTRECOVERY(tp)) { 1950 tp->t_dupacks = 0; 1951 break; 1952 } 1953 } else if (tcp_do_newreno) { 1954 if (SEQ_LEQ(th->th_ack, 1955 tp->snd_recover)) { 1956 tp->t_dupacks = 0; 1957 break; 1958 } 1959 } 1960 win = min(tp->snd_wnd, tp->snd_cwnd) / 1961 2 / tp->t_maxseg; 1962 if (win < 2) 1963 win = 2; 1964 tp->snd_ssthresh = win * tp->t_maxseg; 1965 ENTER_FASTRECOVERY(tp); 1966 tp->snd_recover = tp->snd_max; 1967 callout_stop(tp->tt_rexmt); 1968 tp->t_rtttime = 0; 1969 if (tp->sack_enable) { 1970 tcpstat.tcps_sack_recovery_episode++; 1971 tp->snd_cwnd = 1972 tp->t_maxseg * 1973 tp->t_dupacks; 1974 (void) tcp_output(tp); 1975 tp->snd_cwnd += 1976 tp->snd_ssthresh; 1977 goto drop; 1978 } 1979 1980 tp->snd_nxt = th->th_ack; 1981 tp->snd_cwnd = tp->t_maxseg; 1982 (void) tcp_output(tp); 1983 KASSERT(tp->snd_limited <= 2, 1984 ("tp->snd_limited too big")); 1985 tp->snd_cwnd = tp->snd_ssthresh + 1986 tp->t_maxseg * 1987 (tp->t_dupacks - tp->snd_limited); 1988 if (SEQ_GT(onxt, tp->snd_nxt)) 1989 tp->snd_nxt = onxt; 1990 goto drop; 1991 } else if (tcp_do_rfc3042) { 1992 u_long oldcwnd = tp->snd_cwnd; 1993 tcp_seq oldsndmax = tp->snd_max; 1994 u_int sent; 1995 1996 KASSERT(tp->t_dupacks == 1 || 1997 tp->t_dupacks == 2, 1998 ("dupacks not 1 or 2")); 1999 if (tp->t_dupacks == 1) 2000 tp->snd_limited = 0; 2001 tp->snd_cwnd = 2002 (tp->snd_nxt - tp->snd_una) + 2003 (tp->t_dupacks - tp->snd_limited) * 2004 tp->t_maxseg; 2005 (void) tcp_output(tp); 2006 sent = tp->snd_max - oldsndmax; 2007 if (sent > tp->t_maxseg) { 2008 KASSERT((tp->t_dupacks == 2 && 2009 tp->snd_limited == 0) || 2010 (sent == tp->t_maxseg + 1 && 2011 tp->t_flags & TF_SENTFIN), 2012 ("sent too much")); 2013 tp->snd_limited = 2; 2014 } else if (sent > 0) 2015 ++tp->snd_limited; 2016 tp->snd_cwnd = oldcwnd; 2017 goto drop; 2018 } 2019 } else 2020 tp->t_dupacks = 0; 2021 break; 2022 } 2023 2024 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 2025 2026 /* 2027 * If the congestion window was inflated to account 2028 * for the other side's cached packets, retract it. 2029 */ 2030 if (tcp_do_newreno || tp->sack_enable) { 2031 if (IN_FASTRECOVERY(tp)) { 2032 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2033 if (tp->sack_enable) 2034 tcp_sack_partialack(tp, th); 2035 else 2036 tcp_newreno_partial_ack(tp, th); 2037 } else { 2038 /* 2039 * Out of fast recovery. 2040 * Window inflation should have left us 2041 * with approximately snd_ssthresh 2042 * outstanding data. 2043 * But in case we would be inclined to 2044 * send a burst, better to do it via 2045 * the slow start mechanism. 2046 */ 2047 if (SEQ_GT(th->th_ack + 2048 tp->snd_ssthresh, 2049 tp->snd_max)) 2050 tp->snd_cwnd = tp->snd_max - 2051 th->th_ack + 2052 tp->t_maxseg; 2053 else 2054 tp->snd_cwnd = tp->snd_ssthresh; 2055 } 2056 } 2057 } else { 2058 if (tp->t_dupacks >= tcprexmtthresh && 2059 tp->snd_cwnd > tp->snd_ssthresh) 2060 tp->snd_cwnd = tp->snd_ssthresh; 2061 } 2062 tp->t_dupacks = 0; 2063 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2064 tcpstat.tcps_rcvacktoomuch++; 2065 goto dropafterack; 2066 } 2067 /* 2068 * If we reach this point, ACK is not a duplicate, 2069 * i.e., it ACKs something we sent. 2070 */ 2071 if (tp->t_flags & TF_NEEDSYN) { 2072 /* 2073 * T/TCP: Connection was half-synchronized, and our 2074 * SYN has been ACK'd (so connection is now fully 2075 * synchronized). Go to non-starred state, 2076 * increment snd_una for ACK of SYN, and check if 2077 * we can do window scaling. 2078 */ 2079 tp->t_flags &= ~TF_NEEDSYN; 2080 tp->snd_una++; 2081 /* Do window scaling? */ 2082 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2083 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2084 tp->snd_scale = tp->requested_s_scale; 2085 tp->rcv_scale = tp->request_r_scale; 2086 } 2087 } 2088 2089 process_ACK: 2090 KASSERT(headlocked, 2091 ("tcp_input(): process_ACK head is not locked")); 2092 INP_LOCK_ASSERT(inp); 2093 2094 acked = th->th_ack - tp->snd_una; 2095 tcpstat.tcps_rcvackpack++; 2096 tcpstat.tcps_rcvackbyte += acked; 2097 2098 /* 2099 * If we just performed our first retransmit, and the ACK 2100 * arrives within our recovery window, then it was a mistake 2101 * to do the retransmit in the first place. Recover our 2102 * original cwnd and ssthresh, and proceed to transmit where 2103 * we left off. 2104 */ 2105 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2106 ++tcpstat.tcps_sndrexmitbad; 2107 tp->snd_cwnd = tp->snd_cwnd_prev; 2108 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2109 tp->snd_recover = tp->snd_recover_prev; 2110 if (tp->t_flags & TF_WASFRECOVERY) 2111 ENTER_FASTRECOVERY(tp); 2112 tp->snd_nxt = tp->snd_max; 2113 tp->t_badrxtwin = 0; /* XXX probably not required */ 2114 } 2115 2116 /* 2117 * If we have a timestamp reply, update smoothed 2118 * round trip time. If no timestamp is present but 2119 * transmit timer is running and timed sequence 2120 * number was acked, update smoothed round trip time. 2121 * Since we now have an rtt measurement, cancel the 2122 * timer backoff (cf., Phil Karn's retransmit alg.). 2123 * Recompute the initial retransmit timer. 2124 * 2125 * Some boxes send broken timestamp replies 2126 * during the SYN+ACK phase, ignore 2127 * timestamps of 0 or we could calculate a 2128 * huge RTT and blow up the retransmit timer. 2129 */ 2130 if ((to.to_flags & TOF_TS) != 0 && 2131 to.to_tsecr) { 2132 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2133 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2134 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2135 } 2136 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2137 2138 /* 2139 * If all outstanding data is acked, stop retransmit 2140 * timer and remember to restart (more output or persist). 2141 * If there is more data to be acked, restart retransmit 2142 * timer, using current (possibly backed-off) value. 2143 */ 2144 if (th->th_ack == tp->snd_max) { 2145 callout_stop(tp->tt_rexmt); 2146 needoutput = 1; 2147 } else if (!callout_active(tp->tt_persist)) 2148 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2149 tcp_timer_rexmt, tp); 2150 2151 /* 2152 * If no data (only SYN) was ACK'd, 2153 * skip rest of ACK processing. 2154 */ 2155 if (acked == 0) 2156 goto step6; 2157 2158 /* 2159 * When new data is acked, open the congestion window. 2160 * If the window gives us less than ssthresh packets 2161 * in flight, open exponentially (maxseg per packet). 2162 * Otherwise open linearly: maxseg per window 2163 * (maxseg^2 / cwnd per packet). 2164 */ 2165 if ((!tcp_do_newreno && !tp->sack_enable) || 2166 !IN_FASTRECOVERY(tp)) { 2167 register u_int cw = tp->snd_cwnd; 2168 register u_int incr = tp->t_maxseg; 2169 if (cw > tp->snd_ssthresh) 2170 incr = incr * incr / cw; 2171 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2172 } 2173 SOCKBUF_LOCK(&so->so_snd); 2174 if (acked > so->so_snd.sb_cc) { 2175 tp->snd_wnd -= so->so_snd.sb_cc; 2176 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2177 ourfinisacked = 1; 2178 } else { 2179 sbdrop_locked(&so->so_snd, acked); 2180 tp->snd_wnd -= acked; 2181 ourfinisacked = 0; 2182 } 2183 sowwakeup_locked(so); 2184 /* detect una wraparound */ 2185 if ((tcp_do_newreno || tp->sack_enable) && 2186 !IN_FASTRECOVERY(tp) && 2187 SEQ_GT(tp->snd_una, tp->snd_recover) && 2188 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2189 tp->snd_recover = th->th_ack - 1; 2190 if ((tcp_do_newreno || tp->sack_enable) && 2191 IN_FASTRECOVERY(tp) && 2192 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2193 EXIT_FASTRECOVERY(tp); 2194 tp->snd_una = th->th_ack; 2195 if (tp->sack_enable) { 2196 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2197 tp->snd_recover = tp->snd_una; 2198 } 2199 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2200 tp->snd_nxt = tp->snd_una; 2201 2202 switch (tp->t_state) { 2203 2204 /* 2205 * In FIN_WAIT_1 STATE in addition to the processing 2206 * for the ESTABLISHED state if our FIN is now acknowledged 2207 * then enter FIN_WAIT_2. 2208 */ 2209 case TCPS_FIN_WAIT_1: 2210 if (ourfinisacked) { 2211 /* 2212 * If we can't receive any more 2213 * data, then closing user can proceed. 2214 * Starting the timer is contrary to the 2215 * specification, but if we don't get a FIN 2216 * we'll hang forever. 2217 */ 2218 /* XXXjl 2219 * we should release the tp also, and use a 2220 * compressed state. 2221 */ 2222 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2223 soisdisconnected(so); 2224 callout_reset(tp->tt_2msl, tcp_maxidle, 2225 tcp_timer_2msl, tp); 2226 } 2227 tp->t_state = TCPS_FIN_WAIT_2; 2228 } 2229 break; 2230 2231 /* 2232 * In CLOSING STATE in addition to the processing for 2233 * the ESTABLISHED state if the ACK acknowledges our FIN 2234 * then enter the TIME-WAIT state, otherwise ignore 2235 * the segment. 2236 */ 2237 case TCPS_CLOSING: 2238 if (ourfinisacked) { 2239 KASSERT(headlocked, ("headlocked")); 2240 tcp_twstart(tp); 2241 INP_INFO_WUNLOCK(&tcbinfo); 2242 m_freem(m); 2243 return; 2244 } 2245 break; 2246 2247 /* 2248 * In LAST_ACK, we may still be waiting for data to drain 2249 * and/or to be acked, as well as for the ack of our FIN. 2250 * If our FIN is now acknowledged, delete the TCB, 2251 * enter the closed state and return. 2252 */ 2253 case TCPS_LAST_ACK: 2254 if (ourfinisacked) { 2255 tp = tcp_close(tp); 2256 goto drop; 2257 } 2258 break; 2259 2260 /* 2261 * In TIME_WAIT state the only thing that should arrive 2262 * is a retransmission of the remote FIN. Acknowledge 2263 * it and restart the finack timer. 2264 */ 2265 case TCPS_TIME_WAIT: 2266 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2267 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2268 tcp_timer_2msl, tp); 2269 goto dropafterack; 2270 } 2271 } 2272 2273 step6: 2274 KASSERT(headlocked, ("tcp_input(): step6 head is not locked")); 2275 INP_LOCK_ASSERT(inp); 2276 2277 /* 2278 * Update window information. 2279 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2280 */ 2281 if ((thflags & TH_ACK) && 2282 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2283 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2284 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2285 /* keep track of pure window updates */ 2286 if (tlen == 0 && 2287 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2288 tcpstat.tcps_rcvwinupd++; 2289 tp->snd_wnd = tiwin; 2290 tp->snd_wl1 = th->th_seq; 2291 tp->snd_wl2 = th->th_ack; 2292 if (tp->snd_wnd > tp->max_sndwnd) 2293 tp->max_sndwnd = tp->snd_wnd; 2294 needoutput = 1; 2295 } 2296 2297 /* 2298 * Process segments with URG. 2299 */ 2300 if ((thflags & TH_URG) && th->th_urp && 2301 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2302 /* 2303 * This is a kludge, but if we receive and accept 2304 * random urgent pointers, we'll crash in 2305 * soreceive. It's hard to imagine someone 2306 * actually wanting to send this much urgent data. 2307 */ 2308 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2309 th->th_urp = 0; /* XXX */ 2310 thflags &= ~TH_URG; /* XXX */ 2311 goto dodata; /* XXX */ 2312 } 2313 /* 2314 * If this segment advances the known urgent pointer, 2315 * then mark the data stream. This should not happen 2316 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2317 * a FIN has been received from the remote side. 2318 * In these states we ignore the URG. 2319 * 2320 * According to RFC961 (Assigned Protocols), 2321 * the urgent pointer points to the last octet 2322 * of urgent data. We continue, however, 2323 * to consider it to indicate the first octet 2324 * of data past the urgent section as the original 2325 * spec states (in one of two places). 2326 */ 2327 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2328 tp->rcv_up = th->th_seq + th->th_urp; 2329 SOCKBUF_LOCK(&so->so_rcv); 2330 so->so_oobmark = so->so_rcv.sb_cc + 2331 (tp->rcv_up - tp->rcv_nxt) - 1; 2332 if (so->so_oobmark == 0) 2333 so->so_rcv.sb_state |= SBS_RCVATMARK; 2334 SOCKBUF_UNLOCK(&so->so_rcv); 2335 sohasoutofband(so); 2336 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2337 } 2338 /* 2339 * Remove out of band data so doesn't get presented to user. 2340 * This can happen independent of advancing the URG pointer, 2341 * but if two URG's are pending at once, some out-of-band 2342 * data may creep in... ick. 2343 */ 2344 if (th->th_urp <= (u_long)tlen && 2345 !(so->so_options & SO_OOBINLINE)) { 2346 /* hdr drop is delayed */ 2347 tcp_pulloutofband(so, th, m, drop_hdrlen); 2348 } 2349 } else { 2350 /* 2351 * If no out of band data is expected, 2352 * pull receive urgent pointer along 2353 * with the receive window. 2354 */ 2355 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2356 tp->rcv_up = tp->rcv_nxt; 2357 } 2358 dodata: /* XXX */ 2359 KASSERT(headlocked, ("tcp_input(): dodata head is not locked")); 2360 INP_LOCK_ASSERT(inp); 2361 2362 /* 2363 * Process the segment text, merging it into the TCP sequencing queue, 2364 * and arranging for acknowledgment of receipt if necessary. 2365 * This process logically involves adjusting tp->rcv_wnd as data 2366 * is presented to the user (this happens in tcp_usrreq.c, 2367 * case PRU_RCVD). If a FIN has already been received on this 2368 * connection then we just ignore the text. 2369 */ 2370 if ((tlen || (thflags & TH_FIN)) && 2371 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2372 m_adj(m, drop_hdrlen); /* delayed header drop */ 2373 /* 2374 * Insert segment which includes th into TCP reassembly queue 2375 * with control block tp. Set thflags to whether reassembly now 2376 * includes a segment with FIN. This handles the common case 2377 * inline (segment is the next to be received on an established 2378 * connection, and the queue is empty), avoiding linkage into 2379 * and removal from the queue and repetition of various 2380 * conversions. 2381 * Set DELACK for segments received in order, but ack 2382 * immediately when segments are out of order (so 2383 * fast retransmit can work). 2384 */ 2385 if (th->th_seq == tp->rcv_nxt && 2386 LIST_EMPTY(&tp->t_segq) && 2387 TCPS_HAVEESTABLISHED(tp->t_state)) { 2388 if (DELAY_ACK(tp)) 2389 tp->t_flags |= TF_DELACK; 2390 else 2391 tp->t_flags |= TF_ACKNOW; 2392 tp->rcv_nxt += tlen; 2393 thflags = th->th_flags & TH_FIN; 2394 tcpstat.tcps_rcvpack++; 2395 tcpstat.tcps_rcvbyte += tlen; 2396 ND6_HINT(tp); 2397 /* Unlocked read. */ 2398 SOCKBUF_LOCK(&so->so_rcv); 2399 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2400 m_freem(m); 2401 else 2402 sbappendstream_locked(&so->so_rcv, m); 2403 sorwakeup_locked(so); 2404 } else { 2405 thflags = tcp_reass(tp, th, &tlen, m); 2406 tp->t_flags |= TF_ACKNOW; 2407 } 2408 if (tp->sack_enable) 2409 tcp_update_sack_list(tp); 2410 /* 2411 * Note the amount of data that peer has sent into 2412 * our window, in order to estimate the sender's 2413 * buffer size. 2414 */ 2415 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2416 } else { 2417 m_freem(m); 2418 thflags &= ~TH_FIN; 2419 } 2420 2421 /* 2422 * If FIN is received ACK the FIN and let the user know 2423 * that the connection is closing. 2424 */ 2425 if (thflags & TH_FIN) { 2426 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2427 socantrcvmore(so); 2428 /* 2429 * If connection is half-synchronized 2430 * (ie NEEDSYN flag on) then delay ACK, 2431 * so it may be piggybacked when SYN is sent. 2432 * Otherwise, since we received a FIN then no 2433 * more input can be expected, send ACK now. 2434 */ 2435 if (tp->t_flags & TF_NEEDSYN) 2436 tp->t_flags |= TF_DELACK; 2437 else 2438 tp->t_flags |= TF_ACKNOW; 2439 tp->rcv_nxt++; 2440 } 2441 switch (tp->t_state) { 2442 2443 /* 2444 * In SYN_RECEIVED and ESTABLISHED STATES 2445 * enter the CLOSE_WAIT state. 2446 */ 2447 case TCPS_SYN_RECEIVED: 2448 tp->t_starttime = ticks; 2449 /*FALLTHROUGH*/ 2450 case TCPS_ESTABLISHED: 2451 tp->t_state = TCPS_CLOSE_WAIT; 2452 break; 2453 2454 /* 2455 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2456 * enter the CLOSING state. 2457 */ 2458 case TCPS_FIN_WAIT_1: 2459 tp->t_state = TCPS_CLOSING; 2460 break; 2461 2462 /* 2463 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2464 * starting the time-wait timer, turning off the other 2465 * standard timers. 2466 */ 2467 case TCPS_FIN_WAIT_2: 2468 KASSERT(headlocked == 1, ("headlocked should be 1")); 2469 tcp_twstart(tp); 2470 INP_INFO_WUNLOCK(&tcbinfo); 2471 return; 2472 2473 /* 2474 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2475 */ 2476 case TCPS_TIME_WAIT: 2477 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2478 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2479 tcp_timer_2msl, tp); 2480 break; 2481 } 2482 } 2483 INP_INFO_WUNLOCK(&tcbinfo); 2484 #ifdef TCPDEBUG 2485 if (so->so_options & SO_DEBUG) 2486 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2487 &tcp_savetcp, 0); 2488 #endif 2489 2490 /* 2491 * Return any desired output. 2492 */ 2493 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2494 (void) tcp_output(tp); 2495 2496 check_delack: 2497 INP_LOCK_ASSERT(inp); 2498 if (tp->t_flags & TF_DELACK) { 2499 tp->t_flags &= ~TF_DELACK; 2500 callout_reset(tp->tt_delack, tcp_delacktime, 2501 tcp_timer_delack, tp); 2502 } 2503 INP_UNLOCK(inp); 2504 return; 2505 2506 dropafterack: 2507 /* 2508 * Generate an ACK dropping incoming segment if it occupies 2509 * sequence space, where the ACK reflects our state. 2510 * 2511 * We can now skip the test for the RST flag since all 2512 * paths to this code happen after packets containing 2513 * RST have been dropped. 2514 * 2515 * In the SYN-RECEIVED state, don't send an ACK unless the 2516 * segment we received passes the SYN-RECEIVED ACK test. 2517 * If it fails send a RST. This breaks the loop in the 2518 * "LAND" DoS attack, and also prevents an ACK storm 2519 * between two listening ports that have been sent forged 2520 * SYN segments, each with the source address of the other. 2521 */ 2522 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2523 (SEQ_GT(tp->snd_una, th->th_ack) || 2524 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2525 rstreason = BANDLIM_RST_OPENPORT; 2526 goto dropwithreset; 2527 } 2528 #ifdef TCPDEBUG 2529 if (so->so_options & SO_DEBUG) 2530 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2531 &tcp_savetcp, 0); 2532 #endif 2533 KASSERT(headlocked, ("headlocked should be 1")); 2534 INP_INFO_WUNLOCK(&tcbinfo); 2535 m_freem(m); 2536 tp->t_flags |= TF_ACKNOW; 2537 (void) tcp_output(tp); 2538 INP_UNLOCK(inp); 2539 return; 2540 2541 dropwithreset: 2542 /* 2543 * Generate a RST, dropping incoming segment. 2544 * Make ACK acceptable to originator of segment. 2545 * Don't bother to respond if destination was broadcast/multicast. 2546 */ 2547 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2548 goto drop; 2549 if (isipv6) { 2550 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2551 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2552 goto drop; 2553 } else { 2554 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2555 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2556 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2557 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2558 goto drop; 2559 } 2560 /* IPv6 anycast check is done at tcp6_input() */ 2561 2562 /* 2563 * Perform bandwidth limiting. 2564 */ 2565 if (badport_bandlim(rstreason) < 0) 2566 goto drop; 2567 2568 #ifdef TCPDEBUG 2569 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2570 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2571 &tcp_savetcp, 0); 2572 #endif 2573 2574 if (thflags & TH_ACK) 2575 /* mtod() below is safe as long as hdr dropping is delayed */ 2576 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2577 TH_RST); 2578 else { 2579 if (thflags & TH_SYN) 2580 tlen++; 2581 /* mtod() below is safe as long as hdr dropping is delayed */ 2582 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2583 (tcp_seq)0, TH_RST|TH_ACK); 2584 } 2585 2586 if (tp) 2587 INP_UNLOCK(inp); 2588 if (headlocked) 2589 INP_INFO_WUNLOCK(&tcbinfo); 2590 return; 2591 2592 drop: 2593 /* 2594 * Drop space held by incoming segment and return. 2595 */ 2596 #ifdef TCPDEBUG 2597 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2598 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2599 &tcp_savetcp, 0); 2600 #endif 2601 if (tp) 2602 INP_UNLOCK(inp); 2603 m_freem(m); 2604 if (headlocked) 2605 INP_INFO_WUNLOCK(&tcbinfo); 2606 return; 2607 } 2608 2609 /* 2610 * Parse TCP options and place in tcpopt. 2611 */ 2612 static void 2613 tcp_dooptions(tp, to, cp, cnt, is_syn, th) 2614 struct tcpcb *tp; 2615 struct tcpopt *to; 2616 u_char *cp; 2617 int cnt; 2618 int is_syn; 2619 struct tcphdr *th; 2620 { 2621 int opt, optlen; 2622 2623 to->to_flags = 0; 2624 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2625 opt = cp[0]; 2626 if (opt == TCPOPT_EOL) 2627 break; 2628 if (opt == TCPOPT_NOP) 2629 optlen = 1; 2630 else { 2631 if (cnt < 2) 2632 break; 2633 optlen = cp[1]; 2634 if (optlen < 2 || optlen > cnt) 2635 break; 2636 } 2637 switch (opt) { 2638 case TCPOPT_MAXSEG: 2639 if (optlen != TCPOLEN_MAXSEG) 2640 continue; 2641 if (!is_syn) 2642 continue; 2643 to->to_flags |= TOF_MSS; 2644 bcopy((char *)cp + 2, 2645 (char *)&to->to_mss, sizeof(to->to_mss)); 2646 to->to_mss = ntohs(to->to_mss); 2647 break; 2648 case TCPOPT_WINDOW: 2649 if (optlen != TCPOLEN_WINDOW) 2650 continue; 2651 if (! is_syn) 2652 continue; 2653 to->to_flags |= TOF_SCALE; 2654 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2655 break; 2656 case TCPOPT_TIMESTAMP: 2657 if (optlen != TCPOLEN_TIMESTAMP) 2658 continue; 2659 to->to_flags |= TOF_TS; 2660 bcopy((char *)cp + 2, 2661 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2662 to->to_tsval = ntohl(to->to_tsval); 2663 bcopy((char *)cp + 6, 2664 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2665 to->to_tsecr = ntohl(to->to_tsecr); 2666 break; 2667 case TCPOPT_CC: 2668 if (optlen != TCPOLEN_CC) 2669 continue; 2670 to->to_flags |= TOF_CC; 2671 bcopy((char *)cp + 2, 2672 (char *)&to->to_cc, sizeof(to->to_cc)); 2673 to->to_cc = ntohl(to->to_cc); 2674 break; 2675 case TCPOPT_CCNEW: 2676 if (optlen != TCPOLEN_CC) 2677 continue; 2678 if (!is_syn) 2679 continue; 2680 to->to_flags |= TOF_CCNEW; 2681 bcopy((char *)cp + 2, 2682 (char *)&to->to_cc, sizeof(to->to_cc)); 2683 to->to_cc = ntohl(to->to_cc); 2684 break; 2685 case TCPOPT_CCECHO: 2686 if (optlen != TCPOLEN_CC) 2687 continue; 2688 if (!is_syn) 2689 continue; 2690 to->to_flags |= TOF_CCECHO; 2691 bcopy((char *)cp + 2, 2692 (char *)&to->to_ccecho, sizeof(to->to_ccecho)); 2693 to->to_ccecho = ntohl(to->to_ccecho); 2694 break; 2695 #ifdef TCP_SIGNATURE 2696 /* 2697 * XXX In order to reply to a host which has set the 2698 * TCP_SIGNATURE option in its initial SYN, we have to 2699 * record the fact that the option was observed here 2700 * for the syncache code to perform the correct response. 2701 */ 2702 case TCPOPT_SIGNATURE: 2703 if (optlen != TCPOLEN_SIGNATURE) 2704 continue; 2705 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2706 break; 2707 #endif 2708 case TCPOPT_SACK_PERMITTED: 2709 if (!tcp_do_sack || 2710 optlen != TCPOLEN_SACK_PERMITTED) 2711 continue; 2712 if (is_syn) { 2713 /* MUST only be set on SYN */ 2714 to->to_flags |= TOF_SACK; 2715 } 2716 break; 2717 2718 case TCPOPT_SACK: 2719 if (!tp || tcp_sack_option(tp, th, cp, optlen)) 2720 continue; 2721 break; 2722 default: 2723 continue; 2724 } 2725 } 2726 } 2727 2728 /* 2729 * Pull out of band byte out of a segment so 2730 * it doesn't appear in the user's data queue. 2731 * It is still reflected in the segment length for 2732 * sequencing purposes. 2733 */ 2734 static void 2735 tcp_pulloutofband(so, th, m, off) 2736 struct socket *so; 2737 struct tcphdr *th; 2738 register struct mbuf *m; 2739 int off; /* delayed to be droped hdrlen */ 2740 { 2741 int cnt = off + th->th_urp - 1; 2742 2743 while (cnt >= 0) { 2744 if (m->m_len > cnt) { 2745 char *cp = mtod(m, caddr_t) + cnt; 2746 struct tcpcb *tp = sototcpcb(so); 2747 2748 tp->t_iobc = *cp; 2749 tp->t_oobflags |= TCPOOB_HAVEDATA; 2750 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2751 m->m_len--; 2752 if (m->m_flags & M_PKTHDR) 2753 m->m_pkthdr.len--; 2754 return; 2755 } 2756 cnt -= m->m_len; 2757 m = m->m_next; 2758 if (m == 0) 2759 break; 2760 } 2761 panic("tcp_pulloutofband"); 2762 } 2763 2764 /* 2765 * Collect new round-trip time estimate 2766 * and update averages and current timeout. 2767 */ 2768 static void 2769 tcp_xmit_timer(tp, rtt) 2770 register struct tcpcb *tp; 2771 int rtt; 2772 { 2773 register int delta; 2774 2775 tcpstat.tcps_rttupdated++; 2776 tp->t_rttupdated++; 2777 if (tp->t_srtt != 0) { 2778 /* 2779 * srtt is stored as fixed point with 5 bits after the 2780 * binary point (i.e., scaled by 8). The following magic 2781 * is equivalent to the smoothing algorithm in rfc793 with 2782 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2783 * point). Adjust rtt to origin 0. 2784 */ 2785 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2786 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2787 2788 if ((tp->t_srtt += delta) <= 0) 2789 tp->t_srtt = 1; 2790 2791 /* 2792 * We accumulate a smoothed rtt variance (actually, a 2793 * smoothed mean difference), then set the retransmit 2794 * timer to smoothed rtt + 4 times the smoothed variance. 2795 * rttvar is stored as fixed point with 4 bits after the 2796 * binary point (scaled by 16). The following is 2797 * equivalent to rfc793 smoothing with an alpha of .75 2798 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2799 * rfc793's wired-in beta. 2800 */ 2801 if (delta < 0) 2802 delta = -delta; 2803 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2804 if ((tp->t_rttvar += delta) <= 0) 2805 tp->t_rttvar = 1; 2806 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2807 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2808 } else { 2809 /* 2810 * No rtt measurement yet - use the unsmoothed rtt. 2811 * Set the variance to half the rtt (so our first 2812 * retransmit happens at 3*rtt). 2813 */ 2814 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2815 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2816 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2817 } 2818 tp->t_rtttime = 0; 2819 tp->t_rxtshift = 0; 2820 2821 /* 2822 * the retransmit should happen at rtt + 4 * rttvar. 2823 * Because of the way we do the smoothing, srtt and rttvar 2824 * will each average +1/2 tick of bias. When we compute 2825 * the retransmit timer, we want 1/2 tick of rounding and 2826 * 1 extra tick because of +-1/2 tick uncertainty in the 2827 * firing of the timer. The bias will give us exactly the 2828 * 1.5 tick we need. But, because the bias is 2829 * statistical, we have to test that we don't drop below 2830 * the minimum feasible timer (which is 2 ticks). 2831 */ 2832 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2833 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2834 2835 /* 2836 * We received an ack for a packet that wasn't retransmitted; 2837 * it is probably safe to discard any error indications we've 2838 * received recently. This isn't quite right, but close enough 2839 * for now (a route might have failed after we sent a segment, 2840 * and the return path might not be symmetrical). 2841 */ 2842 tp->t_softerror = 0; 2843 } 2844 2845 /* 2846 * Determine a reasonable value for maxseg size. 2847 * If the route is known, check route for mtu. 2848 * If none, use an mss that can be handled on the outgoing 2849 * interface without forcing IP to fragment; if bigger than 2850 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2851 * to utilize large mbufs. If no route is found, route has no mtu, 2852 * or the destination isn't local, use a default, hopefully conservative 2853 * size (usually 512 or the default IP max size, but no more than the mtu 2854 * of the interface), as we can't discover anything about intervening 2855 * gateways or networks. We also initialize the congestion/slow start 2856 * window to be a single segment if the destination isn't local. 2857 * While looking at the routing entry, we also initialize other path-dependent 2858 * parameters from pre-set or cached values in the routing entry. 2859 * 2860 * Also take into account the space needed for options that we 2861 * send regularly. Make maxseg shorter by that amount to assure 2862 * that we can send maxseg amount of data even when the options 2863 * are present. Store the upper limit of the length of options plus 2864 * data in maxopd. 2865 * 2866 * 2867 * In case of T/TCP, we call this routine during implicit connection 2868 * setup as well (offer = -1), to initialize maxseg from the cached 2869 * MSS of our peer. 2870 * 2871 * NOTE that this routine is only called when we process an incoming 2872 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2873 */ 2874 void 2875 tcp_mss(tp, offer) 2876 struct tcpcb *tp; 2877 int offer; 2878 { 2879 int rtt, mss; 2880 u_long bufsize; 2881 u_long maxmtu; 2882 struct inpcb *inp = tp->t_inpcb; 2883 struct socket *so; 2884 struct hc_metrics_lite metrics; 2885 struct rmxp_tao tao; 2886 int origoffer = offer; 2887 #ifdef INET6 2888 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2889 size_t min_protoh = isipv6 ? 2890 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2891 sizeof (struct tcpiphdr); 2892 #else 2893 const size_t min_protoh = sizeof(struct tcpiphdr); 2894 #endif 2895 bzero(&tao, sizeof(tao)); 2896 2897 /* initialize */ 2898 #ifdef INET6 2899 if (isipv6) { 2900 maxmtu = tcp_maxmtu6(&inp->inp_inc); 2901 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2902 } else 2903 #endif 2904 { 2905 maxmtu = tcp_maxmtu(&inp->inp_inc); 2906 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2907 } 2908 so = inp->inp_socket; 2909 2910 /* 2911 * no route to sender, stay with default mss and return 2912 */ 2913 if (maxmtu == 0) 2914 return; 2915 2916 /* what have we got? */ 2917 switch (offer) { 2918 case 0: 2919 /* 2920 * Offer == 0 means that there was no MSS on the SYN 2921 * segment, in this case we use tcp_mssdflt. 2922 */ 2923 offer = 2924 #ifdef INET6 2925 isipv6 ? tcp_v6mssdflt : 2926 #endif 2927 tcp_mssdflt; 2928 break; 2929 2930 case -1: 2931 /* 2932 * Offer == -1 means that we didn't receive SYN yet, 2933 * use cached value in that case; 2934 */ 2935 if (tcp_do_rfc1644) 2936 tcp_hc_gettao(&inp->inp_inc, &tao); 2937 if (tao.tao_mssopt != 0) 2938 offer = tao.tao_mssopt; 2939 /* FALLTHROUGH */ 2940 2941 default: 2942 /* 2943 * Prevent DoS attack with too small MSS. Round up 2944 * to at least minmss. 2945 */ 2946 offer = max(offer, tcp_minmss); 2947 /* 2948 * Sanity check: make sure that maxopd will be large 2949 * enough to allow some data on segments even if the 2950 * all the option space is used (40bytes). Otherwise 2951 * funny things may happen in tcp_output. 2952 */ 2953 offer = max(offer, 64); 2954 if (tcp_do_rfc1644) 2955 tcp_hc_updatetao(&inp->inp_inc, 2956 TCP_HC_TAO_MSSOPT, 0, offer); 2957 } 2958 2959 /* 2960 * rmx information is now retrieved from tcp_hostcache 2961 */ 2962 tcp_hc_get(&inp->inp_inc, &metrics); 2963 2964 /* 2965 * if there's a discovered mtu int tcp hostcache, use it 2966 * else, use the link mtu. 2967 */ 2968 if (metrics.rmx_mtu) 2969 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2970 else { 2971 #ifdef INET6 2972 if (isipv6) { 2973 mss = maxmtu - min_protoh; 2974 if (!path_mtu_discovery && 2975 !in6_localaddr(&inp->in6p_faddr)) 2976 mss = min(mss, tcp_v6mssdflt); 2977 } else 2978 #endif 2979 { 2980 mss = maxmtu - min_protoh; 2981 if (!path_mtu_discovery && 2982 !in_localaddr(inp->inp_faddr)) 2983 mss = min(mss, tcp_mssdflt); 2984 } 2985 } 2986 mss = min(mss, offer); 2987 2988 /* 2989 * maxopd stores the maximum length of data AND options 2990 * in a segment; maxseg is the amount of data in a normal 2991 * segment. We need to store this value (maxopd) apart 2992 * from maxseg, because now every segment carries options 2993 * and thus we normally have somewhat less data in segments. 2994 */ 2995 tp->t_maxopd = mss; 2996 2997 /* 2998 * In case of T/TCP, origoffer==-1 indicates, that no segments 2999 * were received yet. In this case we just guess, otherwise 3000 * we do the same as before T/TCP. 3001 */ 3002 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3003 (origoffer == -1 || 3004 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3005 mss -= TCPOLEN_TSTAMP_APPA; 3006 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC && 3007 (origoffer == -1 || 3008 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)) 3009 mss -= TCPOLEN_CC_APPA; 3010 tp->t_maxseg = mss; 3011 3012 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3013 if (mss > MCLBYTES) 3014 mss &= ~(MCLBYTES-1); 3015 #else 3016 if (mss > MCLBYTES) 3017 mss = mss / MCLBYTES * MCLBYTES; 3018 #endif 3019 tp->t_maxseg = mss; 3020 3021 /* 3022 * If there's a pipesize, change the socket buffer to that size, 3023 * don't change if sb_hiwat is different than default (then it 3024 * has been changed on purpose with setsockopt). 3025 * Make the socket buffers an integral number of mss units; 3026 * if the mss is larger than the socket buffer, decrease the mss. 3027 */ 3028 SOCKBUF_LOCK(&so->so_snd); 3029 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3030 bufsize = metrics.rmx_sendpipe; 3031 else 3032 bufsize = so->so_snd.sb_hiwat; 3033 if (bufsize < mss) 3034 mss = bufsize; 3035 else { 3036 bufsize = roundup(bufsize, mss); 3037 if (bufsize > sb_max) 3038 bufsize = sb_max; 3039 if (bufsize > so->so_snd.sb_hiwat) 3040 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3041 } 3042 SOCKBUF_UNLOCK(&so->so_snd); 3043 tp->t_maxseg = mss; 3044 3045 SOCKBUF_LOCK(&so->so_rcv); 3046 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3047 bufsize = metrics.rmx_recvpipe; 3048 else 3049 bufsize = so->so_rcv.sb_hiwat; 3050 if (bufsize > mss) { 3051 bufsize = roundup(bufsize, mss); 3052 if (bufsize > sb_max) 3053 bufsize = sb_max; 3054 if (bufsize > so->so_rcv.sb_hiwat) 3055 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3056 } 3057 SOCKBUF_UNLOCK(&so->so_rcv); 3058 /* 3059 * While we're here, check the others too 3060 */ 3061 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3062 tp->t_srtt = rtt; 3063 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3064 tcpstat.tcps_usedrtt++; 3065 if (metrics.rmx_rttvar) { 3066 tp->t_rttvar = metrics.rmx_rttvar; 3067 tcpstat.tcps_usedrttvar++; 3068 } else { 3069 /* default variation is +- 1 rtt */ 3070 tp->t_rttvar = 3071 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3072 } 3073 TCPT_RANGESET(tp->t_rxtcur, 3074 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3075 tp->t_rttmin, TCPTV_REXMTMAX); 3076 } 3077 if (metrics.rmx_ssthresh) { 3078 /* 3079 * There's some sort of gateway or interface 3080 * buffer limit on the path. Use this to set 3081 * the slow start threshhold, but set the 3082 * threshold to no less than 2*mss. 3083 */ 3084 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3085 tcpstat.tcps_usedssthresh++; 3086 } 3087 if (metrics.rmx_bandwidth) 3088 tp->snd_bandwidth = metrics.rmx_bandwidth; 3089 3090 /* 3091 * Set the slow-start flight size depending on whether this 3092 * is a local network or not. 3093 * 3094 * Extend this so we cache the cwnd too and retrieve it here. 3095 * Make cwnd even bigger than RFC3390 suggests but only if we 3096 * have previous experience with the remote host. Be careful 3097 * not make cwnd bigger than remote receive window or our own 3098 * send socket buffer. Maybe put some additional upper bound 3099 * on the retrieved cwnd. Should do incremental updates to 3100 * hostcache when cwnd collapses so next connection doesn't 3101 * overloads the path again. 3102 * 3103 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3104 * We currently check only in syncache_socket for that. 3105 */ 3106 #define TCP_METRICS_CWND 3107 #ifdef TCP_METRICS_CWND 3108 if (metrics.rmx_cwnd) 3109 tp->snd_cwnd = max(mss, 3110 min(metrics.rmx_cwnd / 2, 3111 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3112 else 3113 #endif 3114 if (tcp_do_rfc3390) 3115 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3116 #ifdef INET6 3117 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3118 (!isipv6 && in_localaddr(inp->inp_faddr))) 3119 #else 3120 else if (in_localaddr(inp->inp_faddr)) 3121 #endif 3122 tp->snd_cwnd = mss * ss_fltsz_local; 3123 else 3124 tp->snd_cwnd = mss * ss_fltsz; 3125 } 3126 3127 /* 3128 * Determine the MSS option to send on an outgoing SYN. 3129 */ 3130 int 3131 tcp_mssopt(inc) 3132 struct in_conninfo *inc; 3133 { 3134 int mss = 0; 3135 u_long maxmtu = 0; 3136 u_long thcmtu = 0; 3137 size_t min_protoh; 3138 #ifdef INET6 3139 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3140 #endif 3141 3142 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3143 3144 #ifdef INET6 3145 if (isipv6) { 3146 mss = tcp_v6mssdflt; 3147 maxmtu = tcp_maxmtu6(inc); 3148 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3149 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3150 } else 3151 #endif 3152 { 3153 mss = tcp_mssdflt; 3154 maxmtu = tcp_maxmtu(inc); 3155 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3156 min_protoh = sizeof(struct tcpiphdr); 3157 } 3158 if (maxmtu && thcmtu) 3159 mss = min(maxmtu, thcmtu) - min_protoh; 3160 else if (maxmtu || thcmtu) 3161 mss = max(maxmtu, thcmtu) - min_protoh; 3162 3163 return (mss); 3164 } 3165 3166 3167 /* 3168 * On a partial ack arrives, force the retransmission of the 3169 * next unacknowledged segment. Do not clear tp->t_dupacks. 3170 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3171 * be started again. 3172 */ 3173 static void 3174 tcp_newreno_partial_ack(tp, th) 3175 struct tcpcb *tp; 3176 struct tcphdr *th; 3177 { 3178 tcp_seq onxt = tp->snd_nxt; 3179 u_long ocwnd = tp->snd_cwnd; 3180 3181 callout_stop(tp->tt_rexmt); 3182 tp->t_rtttime = 0; 3183 tp->snd_nxt = th->th_ack; 3184 /* 3185 * Set snd_cwnd to one segment beyond acknowledged offset. 3186 * (tp->snd_una has not yet been updated when this function is called.) 3187 */ 3188 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3189 tp->t_flags |= TF_ACKNOW; 3190 (void) tcp_output(tp); 3191 tp->snd_cwnd = ocwnd; 3192 if (SEQ_GT(onxt, tp->snd_nxt)) 3193 tp->snd_nxt = onxt; 3194 /* 3195 * Partial window deflation. Relies on fact that tp->snd_una 3196 * not updated yet. 3197 */ 3198 tp->snd_cwnd -= (th->th_ack - tp->snd_una - tp->t_maxseg); 3199 } 3200 3201 /* 3202 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3203 * looking for a pcb in the listen state. Returns 0 otherwise. 3204 */ 3205 static int 3206 tcp_timewait(tw, to, th, m, tlen) 3207 struct tcptw *tw; 3208 struct tcpopt *to; 3209 struct tcphdr *th; 3210 struct mbuf *m; 3211 int tlen; 3212 { 3213 int thflags; 3214 tcp_seq seq; 3215 #ifdef INET6 3216 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3217 #else 3218 const int isipv6 = 0; 3219 #endif 3220 3221 thflags = th->th_flags; 3222 3223 /* 3224 * NOTE: for FIN_WAIT_2 (to be added later), 3225 * must validate sequence number before accepting RST 3226 */ 3227 3228 /* 3229 * If the segment contains RST: 3230 * Drop the segment - see Stevens, vol. 2, p. 964 and 3231 * RFC 1337. 3232 */ 3233 if (thflags & TH_RST) 3234 goto drop; 3235 3236 /* 3237 * If segment contains a SYN and CC [not CC.NEW] option: 3238 * if connection duration > MSL, drop packet and send RST; 3239 * 3240 * if SEG.CC > CCrecv then is new SYN. 3241 * Complete close and delete TCPCB. Then reprocess 3242 * segment, hoping to find new TCPCB in LISTEN state; 3243 * 3244 * else must be old SYN; drop it. 3245 * else do normal processing. 3246 */ 3247 if ((thflags & TH_SYN) && (to->to_flags & TOF_CC) && tw->cc_recv != 0) { 3248 if ((ticks - tw->t_starttime) > tcp_msl) 3249 goto reset; 3250 if (CC_GT(to->to_cc, tw->cc_recv)) { 3251 (void) tcp_twclose(tw, 0); 3252 return (1); 3253 } 3254 goto drop; 3255 } 3256 3257 #if 0 3258 /* PAWS not needed at the moment */ 3259 /* 3260 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3261 * and it's less than ts_recent, drop it. 3262 */ 3263 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3264 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3265 if ((thflags & TH_ACK) == 0) 3266 goto drop; 3267 goto ack; 3268 } 3269 /* 3270 * ts_recent is never updated because we never accept new segments. 3271 */ 3272 #endif 3273 3274 /* 3275 * If a new connection request is received 3276 * while in TIME_WAIT, drop the old connection 3277 * and start over if the sequence numbers 3278 * are above the previous ones. 3279 */ 3280 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3281 (void) tcp_twclose(tw, 0); 3282 return (1); 3283 } 3284 3285 /* 3286 * Drop the the segment if it does not contain an ACK. 3287 */ 3288 if ((thflags & TH_ACK) == 0) 3289 goto drop; 3290 3291 /* 3292 * Reset the 2MSL timer if this is a duplicate FIN. 3293 */ 3294 if (thflags & TH_FIN) { 3295 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3296 if (seq + 1 == tw->rcv_nxt) 3297 tcp_timer_2msl_reset(tw, 2 * tcp_msl); 3298 } 3299 3300 /* 3301 * Acknowledge the segment if it has data or is not a duplicate ACK. 3302 */ 3303 if (thflags != TH_ACK || tlen != 0 || 3304 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3305 tcp_twrespond(tw, TH_ACK); 3306 goto drop; 3307 3308 reset: 3309 /* 3310 * Generate a RST, dropping incoming segment. 3311 * Make ACK acceptable to originator of segment. 3312 * Don't bother to respond if destination was broadcast/multicast. 3313 */ 3314 if (m->m_flags & (M_BCAST|M_MCAST)) 3315 goto drop; 3316 if (isipv6) { 3317 struct ip6_hdr *ip6; 3318 3319 /* IPv6 anycast check is done at tcp6_input() */ 3320 ip6 = mtod(m, struct ip6_hdr *); 3321 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3322 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3323 goto drop; 3324 } else { 3325 struct ip *ip; 3326 3327 ip = mtod(m, struct ip *); 3328 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3329 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3330 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3331 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3332 goto drop; 3333 } 3334 if (thflags & TH_ACK) { 3335 tcp_respond(NULL, 3336 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3337 } else { 3338 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3339 tcp_respond(NULL, 3340 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3341 } 3342 INP_UNLOCK(tw->tw_inpcb); 3343 return (0); 3344 3345 drop: 3346 INP_UNLOCK(tw->tw_inpcb); 3347 m_freem(m); 3348 return (0); 3349 } 3350