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/malloc.h> 45 #include <sys/mbuf.h> 46 #include <sys/proc.h> /* for proc0 declaration */ 47 #include <sys/protosw.h> 48 #include <sys/signalvar.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/sysctl.h> 52 #include <sys/syslog.h> 53 #include <sys/systm.h> 54 55 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 56 57 #include <vm/uma.h> 58 59 #include <net/if.h> 60 #include <net/route.h> 61 62 #include <netinet/in.h> 63 #include <netinet/in_pcb.h> 64 #include <netinet/in_systm.h> 65 #include <netinet/in_var.h> 66 #include <netinet/ip.h> 67 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 68 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 69 #include <netinet/ip_var.h> 70 #include <netinet/ip_options.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 #include <security/mac/mac_framework.h> 101 102 static const int tcprexmtthresh = 3; 103 104 struct tcpstat tcpstat; 105 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 106 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 107 108 static int log_in_vain = 0; 109 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 110 &log_in_vain, 0, "Log all incoming TCP connections"); 111 112 static int blackhole = 0; 113 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 114 &blackhole, 0, "Do not send RST when dropping refused connections"); 115 116 int tcp_delack_enabled = 1; 117 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 118 &tcp_delack_enabled, 0, 119 "Delay ACK to try and piggyback it onto a data packet"); 120 121 #ifdef TCP_DROP_SYNFIN 122 static int drop_synfin = 0; 123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 124 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 125 #endif 126 127 static int tcp_do_rfc3042 = 1; 128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 129 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 130 131 static int tcp_do_rfc3390 = 1; 132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 133 &tcp_do_rfc3390, 0, 134 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 135 136 static int tcp_insecure_rst = 0; 137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 138 &tcp_insecure_rst, 0, 139 "Follow the old (insecure) criteria for accepting RST packets."); 140 141 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0, 142 "TCP Segment Reassembly Queue"); 143 144 static int tcp_reass_maxseg = 0; 145 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RDTUN, 146 &tcp_reass_maxseg, 0, 147 "Global maximum number of TCP Segments in Reassembly Queue"); 148 149 int tcp_reass_qsize = 0; 150 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD, 151 &tcp_reass_qsize, 0, 152 "Global number of TCP Segments currently in Reassembly Queue"); 153 154 static int tcp_reass_maxqlen = 48; 155 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxqlen, CTLFLAG_RW, 156 &tcp_reass_maxqlen, 0, 157 "Maximum number of TCP Segments per individual Reassembly Queue"); 158 159 static int tcp_reass_overflows = 0; 160 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD, 161 &tcp_reass_overflows, 0, 162 "Global number of TCP Segment Reassembly Queue Overflows"); 163 164 struct inpcbhead tcb; 165 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 166 struct inpcbinfo tcbinfo; 167 struct mtx *tcbinfo_mtx; 168 169 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 170 171 static void tcp_pulloutofband(struct socket *, 172 struct tcphdr *, struct mbuf *, int); 173 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *, 174 struct mbuf *); 175 static void tcp_xmit_timer(struct tcpcb *, int); 176 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 177 static int tcp_timewait(struct inpcb *, struct tcpopt *, 178 struct tcphdr *, struct mbuf *, int); 179 180 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 181 #ifdef INET6 182 #define ND6_HINT(tp) \ 183 do { \ 184 if ((tp) && (tp)->t_inpcb && \ 185 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 186 nd6_nud_hint(NULL, NULL, 0); \ 187 } while (0) 188 #else 189 #define ND6_HINT(tp) 190 #endif 191 192 /* 193 * Indicate whether this ack should be delayed. We can delay the ack if 194 * - there is no delayed ack timer in progress and 195 * - our last ack wasn't a 0-sized window. We never want to delay 196 * the ack that opens up a 0-sized window and 197 * - delayed acks are enabled or 198 * - this is a half-synchronized T/TCP connection. 199 */ 200 #define DELAY_ACK(tp) \ 201 ((!callout_active(tp->tt_delack) && \ 202 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 203 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 204 205 /* Initialize TCP reassembly queue */ 206 static void 207 tcp_reass_zone_change(void *tag) 208 { 209 210 tcp_reass_maxseg = nmbclusters / 16; 211 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 212 } 213 214 uma_zone_t tcp_reass_zone; 215 void 216 tcp_reass_init() 217 { 218 tcp_reass_maxseg = nmbclusters / 16; 219 TUNABLE_INT_FETCH("net.inet.tcp.reass.maxsegments", 220 &tcp_reass_maxseg); 221 tcp_reass_zone = uma_zcreate("tcpreass", sizeof (struct tseg_qent), 222 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 223 uma_zone_set_max(tcp_reass_zone, tcp_reass_maxseg); 224 EVENTHANDLER_REGISTER(nmbclusters_change, 225 tcp_reass_zone_change, NULL, EVENTHANDLER_PRI_ANY); 226 } 227 228 static int 229 tcp_reass(tp, th, tlenp, m) 230 register struct tcpcb *tp; 231 register struct tcphdr *th; 232 int *tlenp; 233 struct mbuf *m; 234 { 235 struct tseg_qent *q; 236 struct tseg_qent *p = NULL; 237 struct tseg_qent *nq; 238 struct tseg_qent *te = NULL; 239 struct socket *so = tp->t_inpcb->inp_socket; 240 int flags; 241 242 INP_LOCK_ASSERT(tp->t_inpcb); 243 244 /* 245 * XXX: tcp_reass() is rather inefficient with its data structures 246 * and should be rewritten (see NetBSD for optimizations). While 247 * doing that it should move to its own file tcp_reass.c. 248 */ 249 250 /* 251 * Call with th==NULL after become established to 252 * force pre-ESTABLISHED data up to user socket. 253 */ 254 if (th == NULL) 255 goto present; 256 257 /* 258 * Limit the number of segments in the reassembly queue to prevent 259 * holding on to too many segments (and thus running out of mbufs). 260 * Make sure to let the missing segment through which caused this 261 * queue. Always keep one global queue entry spare to be able to 262 * process the missing segment. 263 */ 264 if (th->th_seq != tp->rcv_nxt && 265 (tcp_reass_qsize + 1 >= tcp_reass_maxseg || 266 tp->t_segqlen >= tcp_reass_maxqlen)) { 267 tcp_reass_overflows++; 268 tcpstat.tcps_rcvmemdrop++; 269 m_freem(m); 270 *tlenp = 0; 271 return (0); 272 } 273 274 /* 275 * Allocate a new queue entry. If we can't, or hit the zone limit 276 * just drop the pkt. 277 */ 278 te = uma_zalloc(tcp_reass_zone, M_NOWAIT); 279 if (te == NULL) { 280 tcpstat.tcps_rcvmemdrop++; 281 m_freem(m); 282 *tlenp = 0; 283 return (0); 284 } 285 tp->t_segqlen++; 286 tcp_reass_qsize++; 287 288 /* 289 * Find a segment which begins after this one does. 290 */ 291 LIST_FOREACH(q, &tp->t_segq, tqe_q) { 292 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq)) 293 break; 294 p = q; 295 } 296 297 /* 298 * If there is a preceding segment, it may provide some of 299 * our data already. If so, drop the data from the incoming 300 * segment. If it provides all of our data, drop us. 301 */ 302 if (p != NULL) { 303 register int i; 304 /* conversion to int (in i) handles seq wraparound */ 305 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq; 306 if (i > 0) { 307 if (i >= *tlenp) { 308 tcpstat.tcps_rcvduppack++; 309 tcpstat.tcps_rcvdupbyte += *tlenp; 310 m_freem(m); 311 uma_zfree(tcp_reass_zone, te); 312 tp->t_segqlen--; 313 tcp_reass_qsize--; 314 /* 315 * Try to present any queued data 316 * at the left window edge to the user. 317 * This is needed after the 3-WHS 318 * completes. 319 */ 320 goto present; /* ??? */ 321 } 322 m_adj(m, i); 323 *tlenp -= i; 324 th->th_seq += i; 325 } 326 } 327 tcpstat.tcps_rcvoopack++; 328 tcpstat.tcps_rcvoobyte += *tlenp; 329 330 /* 331 * While we overlap succeeding segments trim them or, 332 * if they are completely covered, dequeue them. 333 */ 334 while (q) { 335 register int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq; 336 if (i <= 0) 337 break; 338 if (i < q->tqe_len) { 339 q->tqe_th->th_seq += i; 340 q->tqe_len -= i; 341 m_adj(q->tqe_m, i); 342 break; 343 } 344 345 nq = LIST_NEXT(q, tqe_q); 346 LIST_REMOVE(q, tqe_q); 347 m_freem(q->tqe_m); 348 uma_zfree(tcp_reass_zone, q); 349 tp->t_segqlen--; 350 tcp_reass_qsize--; 351 q = nq; 352 } 353 354 /* Insert the new segment queue entry into place. */ 355 te->tqe_m = m; 356 te->tqe_th = th; 357 te->tqe_len = *tlenp; 358 359 if (p == NULL) { 360 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q); 361 } else { 362 LIST_INSERT_AFTER(p, te, tqe_q); 363 } 364 365 present: 366 /* 367 * Present data to user, advancing rcv_nxt through 368 * completed sequence space. 369 */ 370 if (!TCPS_HAVEESTABLISHED(tp->t_state)) 371 return (0); 372 q = LIST_FIRST(&tp->t_segq); 373 if (!q || q->tqe_th->th_seq != tp->rcv_nxt) 374 return (0); 375 SOCKBUF_LOCK(&so->so_rcv); 376 do { 377 tp->rcv_nxt += q->tqe_len; 378 flags = q->tqe_th->th_flags & TH_FIN; 379 nq = LIST_NEXT(q, tqe_q); 380 LIST_REMOVE(q, tqe_q); 381 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 382 m_freem(q->tqe_m); 383 else 384 sbappendstream_locked(&so->so_rcv, q->tqe_m); 385 uma_zfree(tcp_reass_zone, q); 386 tp->t_segqlen--; 387 tcp_reass_qsize--; 388 q = nq; 389 } while (q && q->tqe_th->th_seq == tp->rcv_nxt); 390 ND6_HINT(tp); 391 sorwakeup_locked(so); 392 return (flags); 393 } 394 395 /* 396 * TCP input routine, follows pages 65-76 of the 397 * protocol specification dated September, 1981 very closely. 398 */ 399 #ifdef INET6 400 int 401 tcp6_input(mp, offp, proto) 402 struct mbuf **mp; 403 int *offp, proto; 404 { 405 register struct mbuf *m = *mp; 406 struct in6_ifaddr *ia6; 407 408 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 409 410 /* 411 * draft-itojun-ipv6-tcp-to-anycast 412 * better place to put this in? 413 */ 414 ia6 = ip6_getdstifaddr(m); 415 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 416 struct ip6_hdr *ip6; 417 418 ip6 = mtod(m, struct ip6_hdr *); 419 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 420 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 421 return IPPROTO_DONE; 422 } 423 424 tcp_input(m, *offp); 425 return IPPROTO_DONE; 426 } 427 #endif 428 429 void 430 tcp_input(m, off0) 431 register struct mbuf *m; 432 int off0; 433 { 434 register struct tcphdr *th; 435 register struct ip *ip = NULL; 436 register struct ipovly *ipov; 437 register struct inpcb *inp = NULL; 438 u_char *optp = NULL; 439 int optlen = 0; 440 int len, tlen, off; 441 int drop_hdrlen; 442 register struct tcpcb *tp = 0; 443 register int thflags; 444 struct socket *so = 0; 445 int todrop, acked, ourfinisacked, needoutput = 0; 446 u_long tiwin; 447 struct tcpopt to; /* options in this segment */ 448 int headlocked = 0; 449 #ifdef IPFIREWALL_FORWARD 450 struct m_tag *fwd_tag; 451 #endif 452 int rstreason; /* For badport_bandlim accounting purposes */ 453 454 struct ip6_hdr *ip6 = NULL; 455 #ifdef INET6 456 int isipv6; 457 #else 458 const int isipv6 = 0; 459 #endif 460 461 #ifdef TCPDEBUG 462 /* 463 * The size of tcp_saveipgen must be the size of the max ip header, 464 * now IPv6. 465 */ 466 u_char tcp_saveipgen[40]; 467 struct tcphdr tcp_savetcp; 468 short ostate = 0; 469 #endif 470 471 #ifdef INET6 472 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 473 #endif 474 bzero((char *)&to, sizeof(to)); 475 476 tcpstat.tcps_rcvtotal++; 477 478 if (isipv6) { 479 #ifdef INET6 480 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */ 481 ip6 = mtod(m, struct ip6_hdr *); 482 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 483 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 484 tcpstat.tcps_rcvbadsum++; 485 goto drop; 486 } 487 th = (struct tcphdr *)((caddr_t)ip6 + off0); 488 489 /* 490 * Be proactive about unspecified IPv6 address in source. 491 * As we use all-zero to indicate unbounded/unconnected pcb, 492 * unspecified IPv6 address can be used to confuse us. 493 * 494 * Note that packets with unspecified IPv6 destination is 495 * already dropped in ip6_input. 496 */ 497 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 498 /* XXX stat */ 499 goto drop; 500 } 501 #else 502 th = NULL; /* XXX: avoid compiler warning */ 503 #endif 504 } else { 505 /* 506 * Get IP and TCP header together in first mbuf. 507 * Note: IP leaves IP header in first mbuf. 508 */ 509 if (off0 > sizeof (struct ip)) { 510 ip_stripoptions(m, (struct mbuf *)0); 511 off0 = sizeof(struct ip); 512 } 513 if (m->m_len < sizeof (struct tcpiphdr)) { 514 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) { 515 tcpstat.tcps_rcvshort++; 516 return; 517 } 518 } 519 ip = mtod(m, struct ip *); 520 ipov = (struct ipovly *)ip; 521 th = (struct tcphdr *)((caddr_t)ip + off0); 522 tlen = ip->ip_len; 523 524 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 525 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 526 th->th_sum = m->m_pkthdr.csum_data; 527 else 528 th->th_sum = in_pseudo(ip->ip_src.s_addr, 529 ip->ip_dst.s_addr, 530 htonl(m->m_pkthdr.csum_data + 531 ip->ip_len + 532 IPPROTO_TCP)); 533 th->th_sum ^= 0xffff; 534 #ifdef TCPDEBUG 535 ipov->ih_len = (u_short)tlen; 536 ipov->ih_len = htons(ipov->ih_len); 537 #endif 538 } else { 539 /* 540 * Checksum extended TCP header and data. 541 */ 542 len = sizeof (struct ip) + tlen; 543 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 544 ipov->ih_len = (u_short)tlen; 545 ipov->ih_len = htons(ipov->ih_len); 546 th->th_sum = in_cksum(m, len); 547 } 548 if (th->th_sum) { 549 tcpstat.tcps_rcvbadsum++; 550 goto drop; 551 } 552 /* Re-initialization for later version check */ 553 ip->ip_v = IPVERSION; 554 } 555 556 /* 557 * Check that TCP offset makes sense, 558 * pull out TCP options and adjust length. XXX 559 */ 560 off = th->th_off << 2; 561 if (off < sizeof (struct tcphdr) || off > tlen) { 562 tcpstat.tcps_rcvbadoff++; 563 goto drop; 564 } 565 tlen -= off; /* tlen is used instead of ti->ti_len */ 566 if (off > sizeof (struct tcphdr)) { 567 if (isipv6) { 568 #ifdef INET6 569 IP6_EXTHDR_CHECK(m, off0, off, ); 570 ip6 = mtod(m, struct ip6_hdr *); 571 th = (struct tcphdr *)((caddr_t)ip6 + off0); 572 #endif 573 } else { 574 if (m->m_len < sizeof(struct ip) + off) { 575 if ((m = m_pullup(m, sizeof (struct ip) + off)) 576 == 0) { 577 tcpstat.tcps_rcvshort++; 578 return; 579 } 580 ip = mtod(m, struct ip *); 581 ipov = (struct ipovly *)ip; 582 th = (struct tcphdr *)((caddr_t)ip + off0); 583 } 584 } 585 optlen = off - sizeof (struct tcphdr); 586 optp = (u_char *)(th + 1); 587 } 588 thflags = th->th_flags; 589 590 #ifdef TCP_DROP_SYNFIN 591 /* 592 * If the drop_synfin option is enabled, drop all packets with 593 * both the SYN and FIN bits set. This prevents e.g. nmap from 594 * identifying the TCP/IP stack. 595 * 596 * This is a violation of the TCP specification. 597 */ 598 if (drop_synfin && (thflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN)) 599 goto drop; 600 #endif 601 602 /* 603 * Convert TCP protocol specific fields to host format. 604 */ 605 th->th_seq = ntohl(th->th_seq); 606 th->th_ack = ntohl(th->th_ack); 607 th->th_win = ntohs(th->th_win); 608 th->th_urp = ntohs(th->th_urp); 609 610 /* 611 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options, 612 * until after ip6_savecontrol() is called and before other functions 613 * which don't want those proto headers. 614 * Because ip6_savecontrol() is going to parse the mbuf to 615 * search for data to be passed up to user-land, it wants mbuf 616 * parameters to be unchanged. 617 * XXX: the call of ip6_savecontrol() has been obsoleted based on 618 * latest version of the advanced API (20020110). 619 */ 620 drop_hdrlen = off0 + off; 621 622 /* 623 * Locate pcb for segment. 624 */ 625 INP_INFO_WLOCK(&tcbinfo); 626 headlocked = 1; 627 findpcb: 628 KASSERT(headlocked, ("tcp_input: findpcb: head not locked")); 629 #ifdef IPFIREWALL_FORWARD 630 /* Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ 631 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 632 633 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 634 struct sockaddr_in *next_hop; 635 636 next_hop = (struct sockaddr_in *)(fwd_tag+1); 637 /* 638 * Transparently forwarded. Pretend to be the destination. 639 * already got one like this? 640 */ 641 inp = in_pcblookup_hash(&tcbinfo, 642 ip->ip_src, th->th_sport, 643 ip->ip_dst, th->th_dport, 644 0, m->m_pkthdr.rcvif); 645 if (!inp) { 646 /* It's new. Try to find the ambushing socket. */ 647 inp = in_pcblookup_hash(&tcbinfo, 648 ip->ip_src, th->th_sport, 649 next_hop->sin_addr, 650 next_hop->sin_port ? 651 ntohs(next_hop->sin_port) : 652 th->th_dport, 653 INPLOOKUP_WILDCARD, 654 m->m_pkthdr.rcvif); 655 } 656 /* Remove the tag from the packet. We don't need it anymore. */ 657 m_tag_delete(m, fwd_tag); 658 } else { 659 #endif /* IPFIREWALL_FORWARD */ 660 if (isipv6) { 661 #ifdef INET6 662 inp = in6_pcblookup_hash(&tcbinfo, 663 &ip6->ip6_src, th->th_sport, 664 &ip6->ip6_dst, th->th_dport, 665 INPLOOKUP_WILDCARD, 666 m->m_pkthdr.rcvif); 667 #endif 668 } else 669 inp = in_pcblookup_hash(&tcbinfo, 670 ip->ip_src, th->th_sport, 671 ip->ip_dst, th->th_dport, 672 INPLOOKUP_WILDCARD, 673 m->m_pkthdr.rcvif); 674 #ifdef IPFIREWALL_FORWARD 675 } 676 #endif /* IPFIREWALL_FORWARD */ 677 678 #if defined(IPSEC) || defined(FAST_IPSEC) 679 #ifdef INET6 680 if (isipv6) { 681 if (inp != NULL && ipsec6_in_reject(m, inp)) { 682 #ifdef IPSEC 683 ipsec6stat.in_polvio++; 684 #endif 685 goto drop; 686 } 687 } else 688 #endif /* INET6 */ 689 if (inp != NULL && ipsec4_in_reject(m, inp)) { 690 #ifdef IPSEC 691 ipsecstat.in_polvio++; 692 #endif 693 goto drop; 694 } 695 #endif /*IPSEC || FAST_IPSEC*/ 696 697 /* 698 * If the state is CLOSED (i.e., TCB does not exist) then 699 * all data in the incoming segment is discarded. 700 * If the TCB exists but is in CLOSED state, it is embryonic, 701 * but should either do a listen or a connect soon. 702 */ 703 if (inp == NULL) { 704 if (log_in_vain) { 705 #ifdef INET6 706 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2]; 707 #else 708 char dbuf[4*sizeof "123"], sbuf[4*sizeof "123"]; 709 #endif 710 711 if (isipv6) { 712 #ifdef INET6 713 strcpy(dbuf, "["); 714 strcpy(sbuf, "["); 715 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst)); 716 strcat(sbuf, ip6_sprintf(&ip6->ip6_src)); 717 strcat(dbuf, "]"); 718 strcat(sbuf, "]"); 719 #endif 720 } else { 721 strcpy(dbuf, inet_ntoa(ip->ip_dst)); 722 strcpy(sbuf, inet_ntoa(ip->ip_src)); 723 } 724 switch (log_in_vain) { 725 case 1: 726 if ((thflags & TH_SYN) == 0) 727 break; 728 /* FALLTHROUGH */ 729 case 2: 730 log(LOG_INFO, 731 "Connection attempt to TCP %s:%d " 732 "from %s:%d flags:0x%02x\n", 733 dbuf, ntohs(th->th_dport), sbuf, 734 ntohs(th->th_sport), thflags); 735 break; 736 default: 737 break; 738 } 739 } 740 if (blackhole) { 741 switch (blackhole) { 742 case 1: 743 if (thflags & TH_SYN) 744 goto drop; 745 break; 746 case 2: 747 goto drop; 748 default: 749 goto drop; 750 } 751 } 752 rstreason = BANDLIM_RST_CLOSEDPORT; 753 goto dropwithreset; 754 } 755 INP_LOCK(inp); 756 757 /* Check the minimum TTL for socket. */ 758 if (inp->inp_ip_minttl != 0) { 759 #ifdef INET6 760 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 761 goto drop; 762 else 763 #endif 764 if (inp->inp_ip_minttl > ip->ip_ttl) 765 goto drop; 766 } 767 768 if (inp->inp_vflag & INP_TIMEWAIT) { 769 /* 770 * The only option of relevance is TOF_CC, and only if 771 * present in a SYN segment. See tcp_timewait(). 772 */ 773 if (thflags & TH_SYN) 774 tcp_dooptions(&to, optp, optlen, TO_SYN); 775 if (tcp_timewait(inp, &to, th, m, tlen)) 776 goto findpcb; 777 /* 778 * tcp_timewait unlocks inp. 779 */ 780 INP_INFO_WUNLOCK(&tcbinfo); 781 return; 782 } 783 tp = intotcpcb(inp); 784 if (tp == 0) { 785 INP_UNLOCK(inp); 786 rstreason = BANDLIM_RST_CLOSEDPORT; 787 goto dropwithreset; 788 } 789 if (tp->t_state == TCPS_CLOSED) 790 goto drop; 791 792 #ifdef MAC 793 INP_LOCK_ASSERT(inp); 794 if (mac_check_inpcb_deliver(inp, m)) 795 goto drop; 796 #endif 797 so = inp->inp_socket; 798 KASSERT(so != NULL, ("tcp_input: so == NULL")); 799 #ifdef TCPDEBUG 800 if (so->so_options & SO_DEBUG) { 801 ostate = tp->t_state; 802 if (isipv6) 803 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 804 else 805 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 806 tcp_savetcp = *th; 807 } 808 #endif 809 if (so->so_options & SO_ACCEPTCONN) { 810 struct in_conninfo inc; 811 812 bzero(&inc, sizeof(inc)); 813 #ifdef INET6 814 inc.inc_isipv6 = isipv6; 815 #endif 816 if (isipv6) { 817 inc.inc6_faddr = ip6->ip6_src; 818 inc.inc6_laddr = ip6->ip6_dst; 819 } else { 820 inc.inc_faddr = ip->ip_src; 821 inc.inc_laddr = ip->ip_dst; 822 } 823 inc.inc_fport = th->th_sport; 824 inc.inc_lport = th->th_dport; 825 826 /* 827 * If the state is LISTEN then ignore segment if it contains 828 * a RST. If the segment contains an ACK then it is bad and 829 * send a RST. If it does not contain a SYN then it is not 830 * interesting; drop it. 831 * 832 * If the state is SYN_RECEIVED (syncache) and seg contains 833 * an ACK, but not for our SYN/ACK, send a RST. If the seg 834 * contains a RST, check the sequence number to see if it 835 * is a valid reset segment. 836 */ 837 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) { 838 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 839 /* 840 * Parse the TCP options here because 841 * syncookies need access to the reflected 842 * timestamp. 843 */ 844 tcp_dooptions(&to, optp, optlen, 0); 845 if (!syncache_expand(&inc, &to, th, &so, m)) { 846 /* 847 * No syncache entry, or ACK was not 848 * for our SYN/ACK. Send a RST. 849 */ 850 tcpstat.tcps_badsyn++; 851 rstreason = BANDLIM_RST_OPENPORT; 852 goto dropwithreset; 853 } 854 if (so == NULL) { 855 /* 856 * Could not complete 3-way handshake, 857 * connection is being closed down, and 858 * syncache has free'd mbuf. 859 */ 860 INP_UNLOCK(inp); 861 INP_INFO_WUNLOCK(&tcbinfo); 862 return; 863 } 864 /* 865 * Socket is created in state SYN_RECEIVED. 866 * Continue processing segment. 867 */ 868 INP_UNLOCK(inp); 869 inp = sotoinpcb(so); 870 INP_LOCK(inp); 871 tp = intotcpcb(inp); 872 /* 873 * This is what would have happened in 874 * tcp_output() when the SYN,ACK was sent. 875 */ 876 tp->snd_up = tp->snd_una; 877 tp->snd_max = tp->snd_nxt = tp->iss + 1; 878 tp->last_ack_sent = tp->rcv_nxt; 879 goto after_listen; 880 } 881 if (thflags & TH_RST) { 882 syncache_chkrst(&inc, th); 883 goto drop; 884 } 885 if (thflags & TH_ACK) { 886 syncache_badack(&inc); 887 tcpstat.tcps_badsyn++; 888 rstreason = BANDLIM_RST_OPENPORT; 889 goto dropwithreset; 890 } 891 goto drop; 892 } 893 894 /* 895 * Segment's flags are (SYN) or (SYN|FIN). 896 */ 897 #ifdef INET6 898 /* 899 * If deprecated address is forbidden, 900 * we do not accept SYN to deprecated interface 901 * address to prevent any new inbound connection from 902 * getting established. 903 * When we do not accept SYN, we send a TCP RST, 904 * with deprecated source address (instead of dropping 905 * it). We compromise it as it is much better for peer 906 * to send a RST, and RST will be the final packet 907 * for the exchange. 908 * 909 * If we do not forbid deprecated addresses, we accept 910 * the SYN packet. RFC2462 does not suggest dropping 911 * SYN in this case. 912 * If we decipher RFC2462 5.5.4, it says like this: 913 * 1. use of deprecated addr with existing 914 * communication is okay - "SHOULD continue to be 915 * used" 916 * 2. use of it with new communication: 917 * (2a) "SHOULD NOT be used if alternate address 918 * with sufficient scope is available" 919 * (2b) nothing mentioned otherwise. 920 * Here we fall into (2b) case as we have no choice in 921 * our source address selection - we must obey the peer. 922 * 923 * The wording in RFC2462 is confusing, and there are 924 * multiple description text for deprecated address 925 * handling - worse, they are not exactly the same. 926 * I believe 5.5.4 is the best one, so we follow 5.5.4. 927 */ 928 if (isipv6 && !ip6_use_deprecated) { 929 struct in6_ifaddr *ia6; 930 931 if ((ia6 = ip6_getdstifaddr(m)) && 932 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 933 INP_UNLOCK(inp); 934 tp = NULL; 935 rstreason = BANDLIM_RST_OPENPORT; 936 goto dropwithreset; 937 } 938 } 939 #endif 940 /* 941 * If it is from this socket, drop it, it must be forged. 942 * Don't bother responding if the destination was a broadcast. 943 */ 944 if (th->th_dport == th->th_sport) { 945 if (isipv6) { 946 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, 947 &ip6->ip6_src)) 948 goto drop; 949 } else { 950 if (ip->ip_dst.s_addr == ip->ip_src.s_addr) 951 goto drop; 952 } 953 } 954 /* 955 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN 956 * 957 * Note that it is quite possible to receive unicast 958 * link-layer packets with a broadcast IP address. Use 959 * in_broadcast() to find them. 960 */ 961 if (m->m_flags & (M_BCAST|M_MCAST)) 962 goto drop; 963 if (isipv6) { 964 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 965 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 966 goto drop; 967 } else { 968 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 969 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 970 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 971 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 972 goto drop; 973 } 974 /* 975 * SYN appears to be valid; create compressed TCP state 976 * for syncache, or perform t/tcp connection. 977 */ 978 if (so->so_qlen <= so->so_qlimit) { 979 #ifdef TCPDEBUG 980 if (so->so_options & SO_DEBUG) 981 tcp_trace(TA_INPUT, ostate, tp, 982 (void *)tcp_saveipgen, &tcp_savetcp, 0); 983 #endif 984 tcp_dooptions(&to, optp, optlen, TO_SYN); 985 if (!syncache_add(&inc, &to, th, inp, &so, m)) 986 goto drop; /* XXX: does not happen */ 987 if (so == NULL) { 988 /* 989 * Entry added to syncache, mbuf used to 990 * send SYN,ACK packet. Everything unlocked 991 * already. 992 */ 993 return; 994 } 995 panic("T/TCP not supported at the moment"); 996 #if 0 /* T/TCP */ 997 /* 998 * Segment passed TAO tests. 999 * XXX: Can't happen at the moment. 1000 */ 1001 INP_UNLOCK(inp); 1002 inp = sotoinpcb(so); 1003 INP_LOCK(inp); 1004 tp = intotcpcb(inp); 1005 tp->t_starttime = ticks; 1006 tp->t_state = TCPS_ESTABLISHED; 1007 1008 /* 1009 * T/TCP logic: 1010 * If there is a FIN or if there is data, then 1011 * delay SYN,ACK(SYN) in the hope of piggy-backing 1012 * it on a response segment. Otherwise must send 1013 * ACK now in case the other side is slow starting. 1014 */ 1015 if (thflags & TH_FIN || tlen != 0) 1016 tp->t_flags |= (TF_DELACK | TF_NEEDSYN); 1017 else 1018 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1019 tiwin = th->th_win << tp->snd_scale; 1020 tcpstat.tcps_connects++; 1021 soisconnected(so); 1022 goto trimthenstep6; 1023 #endif /* T/TCP */ 1024 } 1025 goto drop; 1026 } 1027 after_listen: 1028 KASSERT(headlocked, ("tcp_input: after_listen: head not locked")); 1029 INP_LOCK_ASSERT(inp); 1030 1031 /* Syncache takes care of sockets in the listen state. */ 1032 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN")); 1033 1034 /* 1035 * This is the second part of the MSS DoS prevention code (after 1036 * minmss on the sending side) and it deals with too many too small 1037 * tcp packets in a too short timeframe (1 second). 1038 * 1039 * For every full second we count the number of received packets 1040 * and bytes. If we get a lot of packets per second for this connection 1041 * (tcp_minmssoverload) we take a closer look at it and compute the 1042 * average packet size for the past second. If that is less than 1043 * tcp_minmss we get too many packets with very small payload which 1044 * is not good and burdens our system (and every packet generates 1045 * a wakeup to the process connected to our socket). We can reasonable 1046 * expect this to be small packet DoS attack to exhaust our CPU 1047 * cycles. 1048 * 1049 * Care has to be taken for the minimum packet overload value. This 1050 * value defines the minimum number of packets per second before we 1051 * start to worry. This must not be too low to avoid killing for 1052 * example interactive connections with many small packets like 1053 * telnet or SSH. 1054 * 1055 * Setting either tcp_minmssoverload or tcp_minmss to "0" disables 1056 * this check. 1057 * 1058 * Account for packet if payload packet, skip over ACK, etc. 1059 */ 1060 if (tcp_minmss && tcp_minmssoverload && 1061 tp->t_state == TCPS_ESTABLISHED && tlen > 0) { 1062 if ((unsigned int)(tp->rcv_second - ticks) < hz) { 1063 tp->rcv_pps++; 1064 tp->rcv_byps += tlen + off; 1065 if (tp->rcv_pps > tcp_minmssoverload) { 1066 if ((tp->rcv_byps / tp->rcv_pps) < tcp_minmss) { 1067 printf("too many small tcp packets from " 1068 "%s:%u, av. %lubyte/packet, " 1069 "dropping connection\n", 1070 #ifdef INET6 1071 isipv6 ? 1072 ip6_sprintf(&inp->inp_inc.inc6_faddr) : 1073 #endif 1074 inet_ntoa(inp->inp_inc.inc_faddr), 1075 inp->inp_inc.inc_fport, 1076 tp->rcv_byps / tp->rcv_pps); 1077 KASSERT(headlocked, ("tcp_input: " 1078 "after_listen: tcp_drop: head " 1079 "not locked")); 1080 tp = tcp_drop(tp, ECONNRESET); 1081 tcpstat.tcps_minmssdrops++; 1082 goto drop; 1083 } 1084 } 1085 } else { 1086 tp->rcv_second = ticks + hz; 1087 tp->rcv_pps = 1; 1088 tp->rcv_byps = tlen + off; 1089 } 1090 } 1091 1092 /* 1093 * Segment received on connection. 1094 * Reset idle time and keep-alive timer. 1095 */ 1096 tp->t_rcvtime = ticks; 1097 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1098 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp); 1099 1100 /* 1101 * Unscale the window into a 32-bit value. 1102 * This value is bogus for the TCPS_SYN_SENT state 1103 * and is overwritten later. 1104 */ 1105 tiwin = th->th_win << tp->snd_scale; 1106 1107 /* 1108 * Parse options on any incoming segment. 1109 */ 1110 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0); 1111 1112 /* 1113 * If echoed timestamp is later than the current time, 1114 * fall back to non RFC1323 RTT calculation. Normalize 1115 * timestamp if syncookies were used when this connection 1116 * was established. 1117 */ 1118 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1119 to.to_tsecr -= tp->ts_offset; 1120 if (TSTMP_GT(to.to_tsecr, ticks)) 1121 to.to_tsecr = 0; 1122 } 1123 1124 /* 1125 * Process options only when we get SYN/ACK back. The SYN case 1126 * for incoming connections is handled in tcp_syncache. 1127 * XXX this is traditional behavior, may need to be cleaned up. 1128 */ 1129 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1130 if ((to.to_flags & TOF_SCALE) && 1131 (tp->t_flags & TF_REQ_SCALE)) { 1132 tp->t_flags |= TF_RCVD_SCALE; 1133 tp->snd_scale = to.to_requested_s_scale; 1134 tp->snd_wnd = th->th_win << tp->snd_scale; 1135 tiwin = tp->snd_wnd; 1136 } 1137 if (to.to_flags & TOF_TS) { 1138 tp->t_flags |= TF_RCVD_TSTMP; 1139 tp->ts_recent = to.to_tsval; 1140 tp->ts_recent_age = ticks; 1141 } 1142 if (to.to_flags & TOF_MSS) 1143 tcp_mss(tp, to.to_mss); 1144 if (tp->sack_enable) { 1145 if (!(to.to_flags & TOF_SACK)) 1146 tp->sack_enable = 0; 1147 else 1148 tp->t_flags |= TF_SACK_PERMIT; 1149 } 1150 1151 } 1152 1153 /* 1154 * Header prediction: check for the two common cases 1155 * of a uni-directional data xfer. If the packet has 1156 * no control flags, is in-sequence, the window didn't 1157 * change and we're not retransmitting, it's a 1158 * candidate. If the length is zero and the ack moved 1159 * forward, we're the sender side of the xfer. Just 1160 * free the data acked & wake any higher level process 1161 * that was blocked waiting for space. If the length 1162 * is non-zero and the ack didn't move, we're the 1163 * receiver side. If we're getting packets in-order 1164 * (the reassembly queue is empty), add the data to 1165 * the socket buffer and note that we need a delayed ack. 1166 * Make sure that the hidden state-flags are also off. 1167 * Since we check for TCPS_ESTABLISHED above, it can only 1168 * be TH_NEEDSYN. 1169 */ 1170 if (tp->t_state == TCPS_ESTABLISHED && 1171 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1172 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1173 ((to.to_flags & TOF_TS) == 0 || 1174 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) && 1175 th->th_seq == tp->rcv_nxt && tiwin && tiwin == tp->snd_wnd && 1176 tp->snd_nxt == tp->snd_max) { 1177 1178 /* 1179 * If last ACK falls within this segment's sequence numbers, 1180 * record the timestamp. 1181 * NOTE that the test is modified according to the latest 1182 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1183 */ 1184 if ((to.to_flags & TOF_TS) != 0 && 1185 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1186 tp->ts_recent_age = ticks; 1187 tp->ts_recent = to.to_tsval; 1188 } 1189 1190 if (tlen == 0) { 1191 if (SEQ_GT(th->th_ack, tp->snd_una) && 1192 SEQ_LEQ(th->th_ack, tp->snd_max) && 1193 tp->snd_cwnd >= tp->snd_wnd && 1194 ((!tcp_do_newreno && !tp->sack_enable && 1195 tp->t_dupacks < tcprexmtthresh) || 1196 ((tcp_do_newreno || tp->sack_enable) && 1197 !IN_FASTRECOVERY(tp) && to.to_nsacks == 0 && 1198 TAILQ_EMPTY(&tp->snd_holes)))) { 1199 KASSERT(headlocked, ("headlocked")); 1200 INP_INFO_WUNLOCK(&tcbinfo); 1201 headlocked = 0; 1202 /* 1203 * this is a pure ack for outstanding data. 1204 */ 1205 ++tcpstat.tcps_predack; 1206 /* 1207 * "bad retransmit" recovery 1208 */ 1209 if (tp->t_rxtshift == 1 && 1210 ticks < tp->t_badrxtwin) { 1211 ++tcpstat.tcps_sndrexmitbad; 1212 tp->snd_cwnd = tp->snd_cwnd_prev; 1213 tp->snd_ssthresh = 1214 tp->snd_ssthresh_prev; 1215 tp->snd_recover = tp->snd_recover_prev; 1216 if (tp->t_flags & TF_WASFRECOVERY) 1217 ENTER_FASTRECOVERY(tp); 1218 tp->snd_nxt = tp->snd_max; 1219 tp->t_badrxtwin = 0; 1220 } 1221 1222 /* 1223 * Recalculate the transmit timer / rtt. 1224 * 1225 * Some boxes send broken timestamp replies 1226 * during the SYN+ACK phase, ignore 1227 * timestamps of 0 or we could calculate a 1228 * huge RTT and blow up the retransmit timer. 1229 */ 1230 if ((to.to_flags & TOF_TS) != 0 && 1231 to.to_tsecr) { 1232 if (!tp->t_rttlow || 1233 tp->t_rttlow > ticks - to.to_tsecr) 1234 tp->t_rttlow = ticks - to.to_tsecr; 1235 tcp_xmit_timer(tp, 1236 ticks - to.to_tsecr + 1); 1237 } else if (tp->t_rtttime && 1238 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1239 if (!tp->t_rttlow || 1240 tp->t_rttlow > ticks - tp->t_rtttime) 1241 tp->t_rttlow = ticks - tp->t_rtttime; 1242 tcp_xmit_timer(tp, 1243 ticks - tp->t_rtttime); 1244 } 1245 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1246 acked = th->th_ack - tp->snd_una; 1247 tcpstat.tcps_rcvackpack++; 1248 tcpstat.tcps_rcvackbyte += acked; 1249 sbdrop(&so->so_snd, acked); 1250 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1251 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1252 tp->snd_recover = th->th_ack - 1; 1253 tp->snd_una = th->th_ack; 1254 /* 1255 * pull snd_wl2 up to prevent seq wrap relative 1256 * to th_ack. 1257 */ 1258 tp->snd_wl2 = th->th_ack; 1259 tp->t_dupacks = 0; 1260 m_freem(m); 1261 ND6_HINT(tp); /* some progress has been done */ 1262 1263 /* 1264 * If all outstanding data are acked, stop 1265 * retransmit timer, otherwise restart timer 1266 * using current (possibly backed-off) value. 1267 * If process is waiting for space, 1268 * wakeup/selwakeup/signal. If data 1269 * are ready to send, let tcp_output 1270 * decide between more output or persist. 1271 1272 #ifdef TCPDEBUG 1273 if (so->so_options & SO_DEBUG) 1274 tcp_trace(TA_INPUT, ostate, tp, 1275 (void *)tcp_saveipgen, 1276 &tcp_savetcp, 0); 1277 #endif 1278 */ 1279 if (tp->snd_una == tp->snd_max) 1280 callout_stop(tp->tt_rexmt); 1281 else if (!callout_active(tp->tt_persist)) 1282 callout_reset(tp->tt_rexmt, 1283 tp->t_rxtcur, 1284 tcp_timer_rexmt, tp); 1285 1286 sowwakeup(so); 1287 if (so->so_snd.sb_cc) 1288 (void) tcp_output(tp); 1289 goto check_delack; 1290 } 1291 } else if (th->th_ack == tp->snd_una && 1292 LIST_EMPTY(&tp->t_segq) && 1293 tlen <= sbspace(&so->so_rcv)) { 1294 KASSERT(headlocked, ("headlocked")); 1295 INP_INFO_WUNLOCK(&tcbinfo); 1296 headlocked = 0; 1297 /* 1298 * this is a pure, in-sequence data packet 1299 * with nothing on the reassembly queue and 1300 * we have enough buffer space to take it. 1301 */ 1302 /* Clean receiver SACK report if present */ 1303 if (tp->sack_enable && tp->rcv_numsacks) 1304 tcp_clean_sackreport(tp); 1305 ++tcpstat.tcps_preddat; 1306 tp->rcv_nxt += tlen; 1307 /* 1308 * Pull snd_wl1 up to prevent seq wrap relative to 1309 * th_seq. 1310 */ 1311 tp->snd_wl1 = th->th_seq; 1312 /* 1313 * Pull rcv_up up to prevent seq wrap relative to 1314 * rcv_nxt. 1315 */ 1316 tp->rcv_up = tp->rcv_nxt; 1317 tcpstat.tcps_rcvpack++; 1318 tcpstat.tcps_rcvbyte += tlen; 1319 ND6_HINT(tp); /* some progress has been done */ 1320 /* 1321 #ifdef TCPDEBUG 1322 if (so->so_options & SO_DEBUG) 1323 tcp_trace(TA_INPUT, ostate, tp, 1324 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1325 #endif 1326 * Add data to socket buffer. 1327 */ 1328 SOCKBUF_LOCK(&so->so_rcv); 1329 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1330 m_freem(m); 1331 } else { 1332 m_adj(m, drop_hdrlen); /* delayed header drop */ 1333 sbappendstream_locked(&so->so_rcv, m); 1334 } 1335 sorwakeup_locked(so); 1336 if (DELAY_ACK(tp)) { 1337 tp->t_flags |= TF_DELACK; 1338 } else { 1339 tp->t_flags |= TF_ACKNOW; 1340 tcp_output(tp); 1341 } 1342 goto check_delack; 1343 } 1344 } 1345 1346 /* 1347 * Calculate amount of space in receive window, 1348 * and then do TCP input processing. 1349 * Receive window is amount of space in rcv queue, 1350 * but not less than advertised window. 1351 */ 1352 { int win; 1353 1354 win = sbspace(&so->so_rcv); 1355 if (win < 0) 1356 win = 0; 1357 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1358 } 1359 1360 switch (tp->t_state) { 1361 1362 /* 1363 * If the state is SYN_RECEIVED: 1364 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1365 */ 1366 case TCPS_SYN_RECEIVED: 1367 if ((thflags & TH_ACK) && 1368 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1369 SEQ_GT(th->th_ack, tp->snd_max))) { 1370 rstreason = BANDLIM_RST_OPENPORT; 1371 goto dropwithreset; 1372 } 1373 break; 1374 1375 /* 1376 * If the state is SYN_SENT: 1377 * if seg contains an ACK, but not for our SYN, drop the input. 1378 * if seg contains a RST, then drop the connection. 1379 * if seg does not contain SYN, then drop it. 1380 * Otherwise this is an acceptable SYN segment 1381 * initialize tp->rcv_nxt and tp->irs 1382 * if seg contains ack then advance tp->snd_una 1383 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1384 * arrange for segment to be acked (eventually) 1385 * continue processing rest of data/controls, beginning with URG 1386 */ 1387 case TCPS_SYN_SENT: 1388 if ((thflags & TH_ACK) && 1389 (SEQ_LEQ(th->th_ack, tp->iss) || 1390 SEQ_GT(th->th_ack, tp->snd_max))) { 1391 rstreason = BANDLIM_UNLIMITED; 1392 goto dropwithreset; 1393 } 1394 if (thflags & TH_RST) { 1395 if (thflags & TH_ACK) { 1396 KASSERT(headlocked, ("tcp_input: after_listen" 1397 ": tcp_drop.2: head not locked")); 1398 tp = tcp_drop(tp, ECONNREFUSED); 1399 } 1400 goto drop; 1401 } 1402 if ((thflags & TH_SYN) == 0) 1403 goto drop; 1404 1405 /* Initial send window, already scaled. */ 1406 tp->snd_wnd = th->th_win; 1407 1408 tp->irs = th->th_seq; 1409 tcp_rcvseqinit(tp); 1410 if (thflags & TH_ACK) { 1411 tcpstat.tcps_connects++; 1412 soisconnected(so); 1413 #ifdef MAC 1414 SOCK_LOCK(so); 1415 mac_set_socket_peer_from_mbuf(m, so); 1416 SOCK_UNLOCK(so); 1417 #endif 1418 /* Do window scaling on this connection? */ 1419 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1420 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1421 tp->rcv_scale = tp->request_r_scale; 1422 } 1423 tp->rcv_adv += tp->rcv_wnd; 1424 tp->snd_una++; /* SYN is acked */ 1425 /* 1426 * If there's data, delay ACK; if there's also a FIN 1427 * ACKNOW will be turned on later. 1428 */ 1429 if (DELAY_ACK(tp) && tlen != 0) 1430 callout_reset(tp->tt_delack, tcp_delacktime, 1431 tcp_timer_delack, tp); 1432 else 1433 tp->t_flags |= TF_ACKNOW; 1434 /* 1435 * Received <SYN,ACK> in SYN_SENT[*] state. 1436 * Transitions: 1437 * SYN_SENT --> ESTABLISHED 1438 * SYN_SENT* --> FIN_WAIT_1 1439 */ 1440 tp->t_starttime = ticks; 1441 if (tp->t_flags & TF_NEEDFIN) { 1442 tp->t_state = TCPS_FIN_WAIT_1; 1443 tp->t_flags &= ~TF_NEEDFIN; 1444 thflags &= ~TH_SYN; 1445 } else { 1446 tp->t_state = TCPS_ESTABLISHED; 1447 callout_reset(tp->tt_keep, tcp_keepidle, 1448 tcp_timer_keep, tp); 1449 } 1450 } else { 1451 /* 1452 * Received initial SYN in SYN-SENT[*] state => 1453 * simultaneous open. If segment contains CC option 1454 * and there is a cached CC, apply TAO test. 1455 * If it succeeds, connection is * half-synchronized. 1456 * Otherwise, do 3-way handshake: 1457 * SYN-SENT -> SYN-RECEIVED 1458 * SYN-SENT* -> SYN-RECEIVED* 1459 * If there was no CC option, clear cached CC value. 1460 */ 1461 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1462 callout_stop(tp->tt_rexmt); 1463 tp->t_state = TCPS_SYN_RECEIVED; 1464 } 1465 1466 #if 0 /* T/TCP */ 1467 trimthenstep6: 1468 #endif 1469 KASSERT(headlocked, ("tcp_input: trimthenstep6: head not " 1470 "locked")); 1471 INP_LOCK_ASSERT(inp); 1472 1473 /* 1474 * Advance th->th_seq to correspond to first data byte. 1475 * If data, trim to stay within window, 1476 * dropping FIN if necessary. 1477 */ 1478 th->th_seq++; 1479 if (tlen > tp->rcv_wnd) { 1480 todrop = tlen - tp->rcv_wnd; 1481 m_adj(m, -todrop); 1482 tlen = tp->rcv_wnd; 1483 thflags &= ~TH_FIN; 1484 tcpstat.tcps_rcvpackafterwin++; 1485 tcpstat.tcps_rcvbyteafterwin += todrop; 1486 } 1487 tp->snd_wl1 = th->th_seq - 1; 1488 tp->rcv_up = th->th_seq; 1489 /* 1490 * Client side of transaction: already sent SYN and data. 1491 * If the remote host used T/TCP to validate the SYN, 1492 * our data will be ACK'd; if so, enter normal data segment 1493 * processing in the middle of step 5, ack processing. 1494 * Otherwise, goto step 6. 1495 */ 1496 if (thflags & TH_ACK) 1497 goto process_ACK; 1498 1499 goto step6; 1500 1501 /* 1502 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1503 * do normal processing. 1504 * 1505 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1506 */ 1507 case TCPS_LAST_ACK: 1508 case TCPS_CLOSING: 1509 case TCPS_TIME_WAIT: 1510 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1511 break; /* continue normal processing */ 1512 } 1513 1514 /* 1515 * States other than LISTEN or SYN_SENT. 1516 * First check the RST flag and sequence number since reset segments 1517 * are exempt from the timestamp and connection count tests. This 1518 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1519 * below which allowed reset segments in half the sequence space 1520 * to fall though and be processed (which gives forged reset 1521 * segments with a random sequence number a 50 percent chance of 1522 * killing a connection). 1523 * Then check timestamp, if present. 1524 * Then check the connection count, if present. 1525 * Then check that at least some bytes of segment are within 1526 * receive window. If segment begins before rcv_nxt, 1527 * drop leading data (and SYN); if nothing left, just ack. 1528 * 1529 * 1530 * If the RST bit is set, check the sequence number to see 1531 * if this is a valid reset segment. 1532 * RFC 793 page 37: 1533 * In all states except SYN-SENT, all reset (RST) segments 1534 * are validated by checking their SEQ-fields. A reset is 1535 * valid if its sequence number is in the window. 1536 * Note: this does not take into account delayed ACKs, so 1537 * we should test against last_ack_sent instead of rcv_nxt. 1538 * The sequence number in the reset segment is normally an 1539 * echo of our outgoing acknowlegement numbers, but some hosts 1540 * send a reset with the sequence number at the rightmost edge 1541 * of our receive window, and we have to handle this case. 1542 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1543 * that brute force RST attacks are possible. To combat this, 1544 * we use a much stricter check while in the ESTABLISHED state, 1545 * only accepting RSTs where the sequence number is equal to 1546 * last_ack_sent. In all other states (the states in which a 1547 * RST is more likely), the more permissive check is used. 1548 * If we have multiple segments in flight, the intial reset 1549 * segment sequence numbers will be to the left of last_ack_sent, 1550 * but they will eventually catch up. 1551 * In any case, it never made sense to trim reset segments to 1552 * fit the receive window since RFC 1122 says: 1553 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1554 * 1555 * A TCP SHOULD allow a received RST segment to include data. 1556 * 1557 * DISCUSSION 1558 * It has been suggested that a RST segment could contain 1559 * ASCII text that encoded and explained the cause of the 1560 * RST. No standard has yet been established for such 1561 * data. 1562 * 1563 * If the reset segment passes the sequence number test examine 1564 * the state: 1565 * SYN_RECEIVED STATE: 1566 * If passive open, return to LISTEN state. 1567 * If active open, inform user that connection was refused. 1568 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1569 * Inform user that connection was reset, and close tcb. 1570 * CLOSING, LAST_ACK STATES: 1571 * Close the tcb. 1572 * TIME_WAIT STATE: 1573 * Drop the segment - see Stevens, vol. 2, p. 964 and 1574 * RFC 1337. 1575 */ 1576 if (thflags & TH_RST) { 1577 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 1578 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 1579 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 1580 switch (tp->t_state) { 1581 1582 case TCPS_SYN_RECEIVED: 1583 so->so_error = ECONNREFUSED; 1584 goto close; 1585 1586 case TCPS_ESTABLISHED: 1587 if (tp->last_ack_sent != th->th_seq && 1588 tcp_insecure_rst == 0) { 1589 tcpstat.tcps_badrst++; 1590 goto drop; 1591 } 1592 case TCPS_FIN_WAIT_1: 1593 case TCPS_FIN_WAIT_2: 1594 case TCPS_CLOSE_WAIT: 1595 so->so_error = ECONNRESET; 1596 close: 1597 tp->t_state = TCPS_CLOSED; 1598 tcpstat.tcps_drops++; 1599 KASSERT(headlocked, ("tcp_input: " 1600 "trimthenstep6: tcp_close: head not " 1601 "locked")); 1602 tp = tcp_close(tp); 1603 break; 1604 1605 case TCPS_CLOSING: 1606 case TCPS_LAST_ACK: 1607 KASSERT(headlocked, ("trimthenstep6: " 1608 "tcp_close.2: head not locked")); 1609 tp = tcp_close(tp); 1610 break; 1611 1612 case TCPS_TIME_WAIT: 1613 KASSERT(tp->t_state != TCPS_TIME_WAIT, 1614 ("timewait")); 1615 break; 1616 } 1617 } 1618 goto drop; 1619 } 1620 1621 /* 1622 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1623 * and it's less than ts_recent, drop it. 1624 */ 1625 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1626 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1627 1628 /* Check to see if ts_recent is over 24 days old. */ 1629 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1630 /* 1631 * Invalidate ts_recent. If this segment updates 1632 * ts_recent, the age will be reset later and ts_recent 1633 * will get a valid value. If it does not, setting 1634 * ts_recent to zero will at least satisfy the 1635 * requirement that zero be placed in the timestamp 1636 * echo reply when ts_recent isn't valid. The 1637 * age isn't reset until we get a valid ts_recent 1638 * because we don't want out-of-order segments to be 1639 * dropped when ts_recent is old. 1640 */ 1641 tp->ts_recent = 0; 1642 } else { 1643 tcpstat.tcps_rcvduppack++; 1644 tcpstat.tcps_rcvdupbyte += tlen; 1645 tcpstat.tcps_pawsdrop++; 1646 if (tlen) 1647 goto dropafterack; 1648 goto drop; 1649 } 1650 } 1651 1652 /* 1653 * In the SYN-RECEIVED state, validate that the packet belongs to 1654 * this connection before trimming the data to fit the receive 1655 * window. Check the sequence number versus IRS since we know 1656 * the sequence numbers haven't wrapped. This is a partial fix 1657 * for the "LAND" DoS attack. 1658 */ 1659 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1660 rstreason = BANDLIM_RST_OPENPORT; 1661 goto dropwithreset; 1662 } 1663 1664 todrop = tp->rcv_nxt - th->th_seq; 1665 if (todrop > 0) { 1666 if (thflags & TH_SYN) { 1667 thflags &= ~TH_SYN; 1668 th->th_seq++; 1669 if (th->th_urp > 1) 1670 th->th_urp--; 1671 else 1672 thflags &= ~TH_URG; 1673 todrop--; 1674 } 1675 /* 1676 * Following if statement from Stevens, vol. 2, p. 960. 1677 */ 1678 if (todrop > tlen 1679 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1680 /* 1681 * Any valid FIN must be to the left of the window. 1682 * At this point the FIN must be a duplicate or out 1683 * of sequence; drop it. 1684 */ 1685 thflags &= ~TH_FIN; 1686 1687 /* 1688 * Send an ACK to resynchronize and drop any data. 1689 * But keep on processing for RST or ACK. 1690 */ 1691 tp->t_flags |= TF_ACKNOW; 1692 todrop = tlen; 1693 tcpstat.tcps_rcvduppack++; 1694 tcpstat.tcps_rcvdupbyte += todrop; 1695 } else { 1696 tcpstat.tcps_rcvpartduppack++; 1697 tcpstat.tcps_rcvpartdupbyte += todrop; 1698 } 1699 drop_hdrlen += todrop; /* drop from the top afterwards */ 1700 th->th_seq += todrop; 1701 tlen -= todrop; 1702 if (th->th_urp > todrop) 1703 th->th_urp -= todrop; 1704 else { 1705 thflags &= ~TH_URG; 1706 th->th_urp = 0; 1707 } 1708 } 1709 1710 /* 1711 * If new data are received on a connection after the 1712 * user processes are gone, then RST the other end. 1713 */ 1714 if ((so->so_state & SS_NOFDREF) && 1715 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1716 KASSERT(headlocked, ("trimthenstep6: tcp_close.3: head not " 1717 "locked")); 1718 tp = tcp_close(tp); 1719 tcpstat.tcps_rcvafterclose++; 1720 rstreason = BANDLIM_UNLIMITED; 1721 goto dropwithreset; 1722 } 1723 1724 /* 1725 * If segment ends after window, drop trailing data 1726 * (and PUSH and FIN); if nothing left, just ACK. 1727 */ 1728 todrop = (th->th_seq+tlen) - (tp->rcv_nxt+tp->rcv_wnd); 1729 if (todrop > 0) { 1730 tcpstat.tcps_rcvpackafterwin++; 1731 if (todrop >= tlen) { 1732 tcpstat.tcps_rcvbyteafterwin += tlen; 1733 /* 1734 * If a new connection request is received 1735 * while in TIME_WAIT, drop the old connection 1736 * and start over if the sequence numbers 1737 * are above the previous ones. 1738 */ 1739 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1740 if (thflags & TH_SYN && 1741 tp->t_state == TCPS_TIME_WAIT && 1742 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 1743 KASSERT(headlocked, ("trimthenstep6: " 1744 "tcp_close.4: head not locked")); 1745 tp = tcp_close(tp); 1746 goto findpcb; 1747 } 1748 /* 1749 * If window is closed can only take segments at 1750 * window edge, and have to drop data and PUSH from 1751 * incoming segments. Continue processing, but 1752 * remember to ack. Otherwise, drop segment 1753 * and ack. 1754 */ 1755 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1756 tp->t_flags |= TF_ACKNOW; 1757 tcpstat.tcps_rcvwinprobe++; 1758 } else 1759 goto dropafterack; 1760 } else 1761 tcpstat.tcps_rcvbyteafterwin += todrop; 1762 m_adj(m, -todrop); 1763 tlen -= todrop; 1764 thflags &= ~(TH_PUSH|TH_FIN); 1765 } 1766 1767 /* 1768 * If last ACK falls within this segment's sequence numbers, 1769 * record its timestamp. 1770 * NOTE: 1771 * 1) That the test incorporates suggestions from the latest 1772 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1773 * 2) That updating only on newer timestamps interferes with 1774 * our earlier PAWS tests, so this check should be solely 1775 * predicated on the sequence space of this segment. 1776 * 3) That we modify the segment boundary check to be 1777 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1778 * instead of RFC1323's 1779 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1780 * This modified check allows us to overcome RFC1323's 1781 * limitations as described in Stevens TCP/IP Illustrated 1782 * Vol. 2 p.869. In such cases, we can still calculate the 1783 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1784 */ 1785 if ((to.to_flags & TOF_TS) != 0 && 1786 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1787 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1788 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1789 tp->ts_recent_age = ticks; 1790 tp->ts_recent = to.to_tsval; 1791 } 1792 1793 /* 1794 * If a SYN is in the window, then this is an 1795 * error and we send an RST and drop the connection. 1796 */ 1797 if (thflags & TH_SYN) { 1798 KASSERT(headlocked, ("tcp_input: tcp_drop: trimthenstep6: " 1799 "head not locked")); 1800 tp = tcp_drop(tp, ECONNRESET); 1801 rstreason = BANDLIM_UNLIMITED; 1802 goto drop; 1803 } 1804 1805 /* 1806 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1807 * flag is on (half-synchronized state), then queue data for 1808 * later processing; else drop segment and return. 1809 */ 1810 if ((thflags & TH_ACK) == 0) { 1811 if (tp->t_state == TCPS_SYN_RECEIVED || 1812 (tp->t_flags & TF_NEEDSYN)) 1813 goto step6; 1814 else if (tp->t_flags & TF_ACKNOW) 1815 goto dropafterack; 1816 else 1817 goto drop; 1818 } 1819 1820 /* 1821 * Ack processing. 1822 */ 1823 switch (tp->t_state) { 1824 1825 /* 1826 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1827 * ESTABLISHED state and continue processing. 1828 * The ACK was checked above. 1829 */ 1830 case TCPS_SYN_RECEIVED: 1831 1832 tcpstat.tcps_connects++; 1833 soisconnected(so); 1834 /* Do window scaling? */ 1835 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1836 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1837 tp->rcv_scale = tp->request_r_scale; 1838 tp->snd_wnd = tiwin; 1839 } 1840 /* 1841 * Make transitions: 1842 * SYN-RECEIVED -> ESTABLISHED 1843 * SYN-RECEIVED* -> FIN-WAIT-1 1844 */ 1845 tp->t_starttime = ticks; 1846 if (tp->t_flags & TF_NEEDFIN) { 1847 tp->t_state = TCPS_FIN_WAIT_1; 1848 tp->t_flags &= ~TF_NEEDFIN; 1849 } else { 1850 tp->t_state = TCPS_ESTABLISHED; 1851 callout_reset(tp->tt_keep, tcp_keepidle, 1852 tcp_timer_keep, tp); 1853 } 1854 /* 1855 * If segment contains data or ACK, will call tcp_reass() 1856 * later; if not, do so now to pass queued data to user. 1857 */ 1858 if (tlen == 0 && (thflags & TH_FIN) == 0) 1859 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1860 (struct mbuf *)0); 1861 tp->snd_wl1 = th->th_seq - 1; 1862 /* FALLTHROUGH */ 1863 1864 /* 1865 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1866 * ACKs. If the ack is in the range 1867 * tp->snd_una < th->th_ack <= tp->snd_max 1868 * then advance tp->snd_una to th->th_ack and drop 1869 * data from the retransmission queue. If this ACK reflects 1870 * more up to date window information we update our window information. 1871 */ 1872 case TCPS_ESTABLISHED: 1873 case TCPS_FIN_WAIT_1: 1874 case TCPS_FIN_WAIT_2: 1875 case TCPS_CLOSE_WAIT: 1876 case TCPS_CLOSING: 1877 case TCPS_LAST_ACK: 1878 case TCPS_TIME_WAIT: 1879 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 1880 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1881 tcpstat.tcps_rcvacktoomuch++; 1882 goto dropafterack; 1883 } 1884 if (tp->sack_enable && 1885 (to.to_nsacks > 0 || !TAILQ_EMPTY(&tp->snd_holes))) 1886 tcp_sack_doack(tp, &to, th->th_ack); 1887 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1888 if (tlen == 0 && tiwin == tp->snd_wnd) { 1889 tcpstat.tcps_rcvdupack++; 1890 /* 1891 * If we have outstanding data (other than 1892 * a window probe), this is a completely 1893 * duplicate ack (ie, window info didn't 1894 * change), the ack is the biggest we've 1895 * seen and we've seen exactly our rexmt 1896 * threshhold of them, assume a packet 1897 * has been dropped and retransmit it. 1898 * Kludge snd_nxt & the congestion 1899 * window so we send only this one 1900 * packet. 1901 * 1902 * We know we're losing at the current 1903 * window size so do congestion avoidance 1904 * (set ssthresh to half the current window 1905 * and pull our congestion window back to 1906 * the new ssthresh). 1907 * 1908 * Dup acks mean that packets have left the 1909 * network (they're now cached at the receiver) 1910 * so bump cwnd by the amount in the receiver 1911 * to keep a constant cwnd packets in the 1912 * network. 1913 */ 1914 if (!callout_active(tp->tt_rexmt) || 1915 th->th_ack != tp->snd_una) 1916 tp->t_dupacks = 0; 1917 else if (++tp->t_dupacks > tcprexmtthresh || 1918 ((tcp_do_newreno || tp->sack_enable) && 1919 IN_FASTRECOVERY(tp))) { 1920 if (tp->sack_enable && IN_FASTRECOVERY(tp)) { 1921 int awnd; 1922 1923 /* 1924 * Compute the amount of data in flight first. 1925 * We can inject new data into the pipe iff 1926 * we have less than 1/2 the original window's 1927 * worth of data in flight. 1928 */ 1929 awnd = (tp->snd_nxt - tp->snd_fack) + 1930 tp->sackhint.sack_bytes_rexmit; 1931 if (awnd < tp->snd_ssthresh) { 1932 tp->snd_cwnd += tp->t_maxseg; 1933 if (tp->snd_cwnd > tp->snd_ssthresh) 1934 tp->snd_cwnd = tp->snd_ssthresh; 1935 } 1936 } else 1937 tp->snd_cwnd += tp->t_maxseg; 1938 (void) tcp_output(tp); 1939 goto drop; 1940 } else if (tp->t_dupacks == tcprexmtthresh) { 1941 tcp_seq onxt = tp->snd_nxt; 1942 u_int win; 1943 1944 /* 1945 * If we're doing sack, check to 1946 * see if we're already in sack 1947 * recovery. If we're not doing sack, 1948 * check to see if we're in newreno 1949 * recovery. 1950 */ 1951 if (tp->sack_enable) { 1952 if (IN_FASTRECOVERY(tp)) { 1953 tp->t_dupacks = 0; 1954 break; 1955 } 1956 } else if (tcp_do_newreno) { 1957 if (SEQ_LEQ(th->th_ack, 1958 tp->snd_recover)) { 1959 tp->t_dupacks = 0; 1960 break; 1961 } 1962 } 1963 win = min(tp->snd_wnd, tp->snd_cwnd) / 1964 2 / tp->t_maxseg; 1965 if (win < 2) 1966 win = 2; 1967 tp->snd_ssthresh = win * tp->t_maxseg; 1968 ENTER_FASTRECOVERY(tp); 1969 tp->snd_recover = tp->snd_max; 1970 callout_stop(tp->tt_rexmt); 1971 tp->t_rtttime = 0; 1972 if (tp->sack_enable) { 1973 tcpstat.tcps_sack_recovery_episode++; 1974 tp->sack_newdata = tp->snd_nxt; 1975 tp->snd_cwnd = tp->t_maxseg; 1976 (void) tcp_output(tp); 1977 goto drop; 1978 } 1979 tp->snd_nxt = th->th_ack; 1980 tp->snd_cwnd = tp->t_maxseg; 1981 (void) tcp_output(tp); 1982 KASSERT(tp->snd_limited <= 2, 1983 ("tp->snd_limited too big")); 1984 tp->snd_cwnd = tp->snd_ssthresh + 1985 tp->t_maxseg * 1986 (tp->t_dupacks - tp->snd_limited); 1987 if (SEQ_GT(onxt, tp->snd_nxt)) 1988 tp->snd_nxt = onxt; 1989 goto drop; 1990 } else if (tcp_do_rfc3042) { 1991 u_long oldcwnd = tp->snd_cwnd; 1992 tcp_seq oldsndmax = tp->snd_max; 1993 u_int sent; 1994 1995 KASSERT(tp->t_dupacks == 1 || 1996 tp->t_dupacks == 2, 1997 ("dupacks not 1 or 2")); 1998 if (tp->t_dupacks == 1) 1999 tp->snd_limited = 0; 2000 tp->snd_cwnd = 2001 (tp->snd_nxt - tp->snd_una) + 2002 (tp->t_dupacks - tp->snd_limited) * 2003 tp->t_maxseg; 2004 (void) tcp_output(tp); 2005 sent = tp->snd_max - oldsndmax; 2006 if (sent > tp->t_maxseg) { 2007 KASSERT((tp->t_dupacks == 2 && 2008 tp->snd_limited == 0) || 2009 (sent == tp->t_maxseg + 1 && 2010 tp->t_flags & TF_SENTFIN), 2011 ("sent too much")); 2012 tp->snd_limited = 2; 2013 } else if (sent > 0) 2014 ++tp->snd_limited; 2015 tp->snd_cwnd = oldcwnd; 2016 goto drop; 2017 } 2018 } else 2019 tp->t_dupacks = 0; 2020 break; 2021 } 2022 2023 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una")); 2024 2025 /* 2026 * If the congestion window was inflated to account 2027 * for the other side's cached packets, retract it. 2028 */ 2029 if (tcp_do_newreno || tp->sack_enable) { 2030 if (IN_FASTRECOVERY(tp)) { 2031 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2032 if (tp->sack_enable) 2033 tcp_sack_partialack(tp, th); 2034 else 2035 tcp_newreno_partial_ack(tp, th); 2036 } else { 2037 /* 2038 * Out of fast recovery. 2039 * Window inflation should have left us 2040 * with approximately snd_ssthresh 2041 * outstanding data. 2042 * But in case we would be inclined to 2043 * send a burst, better to do it via 2044 * the slow start mechanism. 2045 */ 2046 if (SEQ_GT(th->th_ack + 2047 tp->snd_ssthresh, 2048 tp->snd_max)) 2049 tp->snd_cwnd = tp->snd_max - 2050 th->th_ack + 2051 tp->t_maxseg; 2052 else 2053 tp->snd_cwnd = tp->snd_ssthresh; 2054 } 2055 } 2056 } else { 2057 if (tp->t_dupacks >= tcprexmtthresh && 2058 tp->snd_cwnd > tp->snd_ssthresh) 2059 tp->snd_cwnd = tp->snd_ssthresh; 2060 } 2061 tp->t_dupacks = 0; 2062 /* 2063 * If we reach this point, ACK is not a duplicate, 2064 * i.e., it ACKs something we sent. 2065 */ 2066 if (tp->t_flags & TF_NEEDSYN) { 2067 /* 2068 * T/TCP: Connection was half-synchronized, and our 2069 * SYN has been ACK'd (so connection is now fully 2070 * synchronized). Go to non-starred state, 2071 * increment snd_una for ACK of SYN, and check if 2072 * we can do window scaling. 2073 */ 2074 tp->t_flags &= ~TF_NEEDSYN; 2075 tp->snd_una++; 2076 /* Do window scaling? */ 2077 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2078 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2079 tp->rcv_scale = tp->request_r_scale; 2080 /* Send window already scaled. */ 2081 } 2082 } 2083 2084 process_ACK: 2085 KASSERT(headlocked, ("tcp_input: process_ACK: head not " 2086 "locked")); 2087 INP_LOCK_ASSERT(inp); 2088 2089 acked = th->th_ack - tp->snd_una; 2090 tcpstat.tcps_rcvackpack++; 2091 tcpstat.tcps_rcvackbyte += acked; 2092 2093 /* 2094 * If we just performed our first retransmit, and the ACK 2095 * arrives within our recovery window, then it was a mistake 2096 * to do the retransmit in the first place. Recover our 2097 * original cwnd and ssthresh, and proceed to transmit where 2098 * we left off. 2099 */ 2100 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2101 ++tcpstat.tcps_sndrexmitbad; 2102 tp->snd_cwnd = tp->snd_cwnd_prev; 2103 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2104 tp->snd_recover = tp->snd_recover_prev; 2105 if (tp->t_flags & TF_WASFRECOVERY) 2106 ENTER_FASTRECOVERY(tp); 2107 tp->snd_nxt = tp->snd_max; 2108 tp->t_badrxtwin = 0; /* XXX probably not required */ 2109 } 2110 2111 /* 2112 * If we have a timestamp reply, update smoothed 2113 * round trip time. If no timestamp is present but 2114 * transmit timer is running and timed sequence 2115 * number was acked, update smoothed round trip time. 2116 * Since we now have an rtt measurement, cancel the 2117 * timer backoff (cf., Phil Karn's retransmit alg.). 2118 * Recompute the initial retransmit timer. 2119 * 2120 * Some boxes send broken timestamp replies 2121 * during the SYN+ACK phase, ignore 2122 * timestamps of 0 or we could calculate a 2123 * huge RTT and blow up the retransmit timer. 2124 */ 2125 if ((to.to_flags & TOF_TS) != 0 && 2126 to.to_tsecr) { 2127 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2128 tp->t_rttlow = ticks - to.to_tsecr; 2129 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2130 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2131 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2132 tp->t_rttlow = ticks - tp->t_rtttime; 2133 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2134 } 2135 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2136 2137 /* 2138 * If all outstanding data is acked, stop retransmit 2139 * timer and remember to restart (more output or persist). 2140 * If there is more data to be acked, restart retransmit 2141 * timer, using current (possibly backed-off) value. 2142 */ 2143 if (th->th_ack == tp->snd_max) { 2144 callout_stop(tp->tt_rexmt); 2145 needoutput = 1; 2146 } else if (!callout_active(tp->tt_persist)) 2147 callout_reset(tp->tt_rexmt, tp->t_rxtcur, 2148 tcp_timer_rexmt, tp); 2149 2150 /* 2151 * If no data (only SYN) was ACK'd, 2152 * skip rest of ACK processing. 2153 */ 2154 if (acked == 0) 2155 goto step6; 2156 2157 /* 2158 * When new data is acked, open the congestion window. 2159 * If the window gives us less than ssthresh packets 2160 * in flight, open exponentially (maxseg per packet). 2161 * Otherwise open linearly: maxseg per window 2162 * (maxseg^2 / cwnd per packet). 2163 */ 2164 if ((!tcp_do_newreno && !tp->sack_enable) || 2165 !IN_FASTRECOVERY(tp)) { 2166 register u_int cw = tp->snd_cwnd; 2167 register u_int incr = tp->t_maxseg; 2168 if (cw > tp->snd_ssthresh) 2169 incr = incr * incr / cw; 2170 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2171 } 2172 SOCKBUF_LOCK(&so->so_snd); 2173 if (acked > so->so_snd.sb_cc) { 2174 tp->snd_wnd -= so->so_snd.sb_cc; 2175 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2176 ourfinisacked = 1; 2177 } else { 2178 sbdrop_locked(&so->so_snd, acked); 2179 tp->snd_wnd -= acked; 2180 ourfinisacked = 0; 2181 } 2182 sowwakeup_locked(so); 2183 /* detect una wraparound */ 2184 if ((tcp_do_newreno || tp->sack_enable) && 2185 !IN_FASTRECOVERY(tp) && 2186 SEQ_GT(tp->snd_una, tp->snd_recover) && 2187 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2188 tp->snd_recover = th->th_ack - 1; 2189 if ((tcp_do_newreno || tp->sack_enable) && 2190 IN_FASTRECOVERY(tp) && 2191 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2192 EXIT_FASTRECOVERY(tp); 2193 tp->snd_una = th->th_ack; 2194 if (tp->sack_enable) { 2195 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2196 tp->snd_recover = tp->snd_una; 2197 } 2198 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2199 tp->snd_nxt = tp->snd_una; 2200 2201 switch (tp->t_state) { 2202 2203 /* 2204 * In FIN_WAIT_1 STATE in addition to the processing 2205 * for the ESTABLISHED state if our FIN is now acknowledged 2206 * then enter FIN_WAIT_2. 2207 */ 2208 case TCPS_FIN_WAIT_1: 2209 if (ourfinisacked) { 2210 /* 2211 * If we can't receive any more 2212 * data, then closing user can proceed. 2213 * Starting the timer is contrary to the 2214 * specification, but if we don't get a FIN 2215 * we'll hang forever. 2216 */ 2217 /* XXXjl 2218 * we should release the tp also, and use a 2219 * compressed state. 2220 */ 2221 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2222 soisdisconnected(so); 2223 callout_reset(tp->tt_2msl, tcp_maxidle, 2224 tcp_timer_2msl, tp); 2225 } 2226 tp->t_state = TCPS_FIN_WAIT_2; 2227 } 2228 break; 2229 2230 /* 2231 * In CLOSING STATE in addition to the processing for 2232 * the ESTABLISHED state if the ACK acknowledges our FIN 2233 * then enter the TIME-WAIT state, otherwise ignore 2234 * the segment. 2235 */ 2236 case TCPS_CLOSING: 2237 if (ourfinisacked) { 2238 KASSERT(headlocked, ("tcp_input: process_ACK: " 2239 "head not locked")); 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 KASSERT(headlocked, ("tcp_input: process_ACK:" 2256 " tcp_close: head not locked")); 2257 tp = tcp_close(tp); 2258 goto drop; 2259 } 2260 break; 2261 2262 /* 2263 * In TIME_WAIT state the only thing that should arrive 2264 * is a retransmission of the remote FIN. Acknowledge 2265 * it and restart the finack timer. 2266 */ 2267 case TCPS_TIME_WAIT: 2268 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2269 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2270 tcp_timer_2msl, tp); 2271 goto dropafterack; 2272 } 2273 } 2274 2275 step6: 2276 KASSERT(headlocked, ("tcp_input: step6: head not locked")); 2277 INP_LOCK_ASSERT(inp); 2278 2279 /* 2280 * Update window information. 2281 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2282 */ 2283 if ((thflags & TH_ACK) && 2284 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2285 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2286 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2287 /* keep track of pure window updates */ 2288 if (tlen == 0 && 2289 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2290 tcpstat.tcps_rcvwinupd++; 2291 tp->snd_wnd = tiwin; 2292 tp->snd_wl1 = th->th_seq; 2293 tp->snd_wl2 = th->th_ack; 2294 if (tp->snd_wnd > tp->max_sndwnd) 2295 tp->max_sndwnd = tp->snd_wnd; 2296 needoutput = 1; 2297 } 2298 2299 /* 2300 * Process segments with URG. 2301 */ 2302 if ((thflags & TH_URG) && th->th_urp && 2303 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2304 /* 2305 * This is a kludge, but if we receive and accept 2306 * random urgent pointers, we'll crash in 2307 * soreceive. It's hard to imagine someone 2308 * actually wanting to send this much urgent data. 2309 */ 2310 SOCKBUF_LOCK(&so->so_rcv); 2311 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2312 th->th_urp = 0; /* XXX */ 2313 thflags &= ~TH_URG; /* XXX */ 2314 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2315 goto dodata; /* XXX */ 2316 } 2317 /* 2318 * If this segment advances the known urgent pointer, 2319 * then mark the data stream. This should not happen 2320 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2321 * a FIN has been received from the remote side. 2322 * In these states we ignore the URG. 2323 * 2324 * According to RFC961 (Assigned Protocols), 2325 * the urgent pointer points to the last octet 2326 * of urgent data. We continue, however, 2327 * to consider it to indicate the first octet 2328 * of data past the urgent section as the original 2329 * spec states (in one of two places). 2330 */ 2331 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2332 tp->rcv_up = th->th_seq + th->th_urp; 2333 so->so_oobmark = so->so_rcv.sb_cc + 2334 (tp->rcv_up - tp->rcv_nxt) - 1; 2335 if (so->so_oobmark == 0) 2336 so->so_rcv.sb_state |= SBS_RCVATMARK; 2337 sohasoutofband(so); 2338 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2339 } 2340 SOCKBUF_UNLOCK(&so->so_rcv); 2341 /* 2342 * Remove out of band data so doesn't get presented to user. 2343 * This can happen independent of advancing the URG pointer, 2344 * but if two URG's are pending at once, some out-of-band 2345 * data may creep in... ick. 2346 */ 2347 if (th->th_urp <= (u_long)tlen && 2348 !(so->so_options & SO_OOBINLINE)) { 2349 /* hdr drop is delayed */ 2350 tcp_pulloutofband(so, th, m, drop_hdrlen); 2351 } 2352 } else { 2353 /* 2354 * If no out of band data is expected, 2355 * pull receive urgent pointer along 2356 * with the receive window. 2357 */ 2358 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2359 tp->rcv_up = tp->rcv_nxt; 2360 } 2361 dodata: /* XXX */ 2362 KASSERT(headlocked, ("tcp_input: dodata: head not locked")); 2363 INP_LOCK_ASSERT(inp); 2364 2365 /* 2366 * Process the segment text, merging it into the TCP sequencing queue, 2367 * and arranging for acknowledgment of receipt if necessary. 2368 * This process logically involves adjusting tp->rcv_wnd as data 2369 * is presented to the user (this happens in tcp_usrreq.c, 2370 * case PRU_RCVD). If a FIN has already been received on this 2371 * connection then we just ignore the text. 2372 */ 2373 if ((tlen || (thflags & TH_FIN)) && 2374 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2375 tcp_seq save_start = th->th_seq; 2376 tcp_seq save_end = th->th_seq + tlen; 2377 m_adj(m, drop_hdrlen); /* delayed header drop */ 2378 /* 2379 * Insert segment which includes th into TCP reassembly queue 2380 * with control block tp. Set thflags to whether reassembly now 2381 * includes a segment with FIN. This handles the common case 2382 * inline (segment is the next to be received on an established 2383 * connection, and the queue is empty), avoiding linkage into 2384 * and removal from the queue and repetition of various 2385 * conversions. 2386 * Set DELACK for segments received in order, but ack 2387 * immediately when segments are out of order (so 2388 * fast retransmit can work). 2389 */ 2390 if (th->th_seq == tp->rcv_nxt && 2391 LIST_EMPTY(&tp->t_segq) && 2392 TCPS_HAVEESTABLISHED(tp->t_state)) { 2393 if (DELAY_ACK(tp)) 2394 tp->t_flags |= TF_DELACK; 2395 else 2396 tp->t_flags |= TF_ACKNOW; 2397 tp->rcv_nxt += tlen; 2398 thflags = th->th_flags & TH_FIN; 2399 tcpstat.tcps_rcvpack++; 2400 tcpstat.tcps_rcvbyte += tlen; 2401 ND6_HINT(tp); 2402 SOCKBUF_LOCK(&so->so_rcv); 2403 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2404 m_freem(m); 2405 else 2406 sbappendstream_locked(&so->so_rcv, m); 2407 sorwakeup_locked(so); 2408 } else { 2409 thflags = tcp_reass(tp, th, &tlen, m); 2410 tp->t_flags |= TF_ACKNOW; 2411 } 2412 if (tlen > 0 && tp->sack_enable) 2413 tcp_update_sack_list(tp, save_start, save_end); 2414 /* 2415 * Note the amount of data that peer has sent into 2416 * our window, in order to estimate the sender's 2417 * buffer size. 2418 */ 2419 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2420 } else { 2421 m_freem(m); 2422 thflags &= ~TH_FIN; 2423 } 2424 2425 /* 2426 * If FIN is received ACK the FIN and let the user know 2427 * that the connection is closing. 2428 */ 2429 if (thflags & TH_FIN) { 2430 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2431 socantrcvmore(so); 2432 /* 2433 * If connection is half-synchronized 2434 * (ie NEEDSYN flag on) then delay ACK, 2435 * so it may be piggybacked when SYN is sent. 2436 * Otherwise, since we received a FIN then no 2437 * more input can be expected, send ACK now. 2438 */ 2439 if (tp->t_flags & TF_NEEDSYN) 2440 tp->t_flags |= TF_DELACK; 2441 else 2442 tp->t_flags |= TF_ACKNOW; 2443 tp->rcv_nxt++; 2444 } 2445 switch (tp->t_state) { 2446 2447 /* 2448 * In SYN_RECEIVED and ESTABLISHED STATES 2449 * enter the CLOSE_WAIT state. 2450 */ 2451 case TCPS_SYN_RECEIVED: 2452 tp->t_starttime = ticks; 2453 /*FALLTHROUGH*/ 2454 case TCPS_ESTABLISHED: 2455 tp->t_state = TCPS_CLOSE_WAIT; 2456 break; 2457 2458 /* 2459 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2460 * enter the CLOSING state. 2461 */ 2462 case TCPS_FIN_WAIT_1: 2463 tp->t_state = TCPS_CLOSING; 2464 break; 2465 2466 /* 2467 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2468 * starting the time-wait timer, turning off the other 2469 * standard timers. 2470 */ 2471 case TCPS_FIN_WAIT_2: 2472 KASSERT(headlocked == 1, ("tcp_input: dodata: " 2473 "TCP_FIN_WAIT_2: head not locked")); 2474 tcp_twstart(tp); 2475 INP_INFO_WUNLOCK(&tcbinfo); 2476 return; 2477 2478 /* 2479 * In TIME_WAIT state restart the 2 MSL time_wait timer. 2480 */ 2481 case TCPS_TIME_WAIT: 2482 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("timewait")); 2483 callout_reset(tp->tt_2msl, 2 * tcp_msl, 2484 tcp_timer_2msl, tp); 2485 break; 2486 } 2487 } 2488 INP_INFO_WUNLOCK(&tcbinfo); 2489 headlocked = 0; 2490 #ifdef TCPDEBUG 2491 if (so->so_options & SO_DEBUG) 2492 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2493 &tcp_savetcp, 0); 2494 #endif 2495 2496 /* 2497 * Return any desired output. 2498 */ 2499 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2500 (void) tcp_output(tp); 2501 2502 check_delack: 2503 KASSERT(headlocked == 0, ("tcp_input: check_delack: head locked")); 2504 INP_LOCK_ASSERT(inp); 2505 if (tp->t_flags & TF_DELACK) { 2506 tp->t_flags &= ~TF_DELACK; 2507 callout_reset(tp->tt_delack, tcp_delacktime, 2508 tcp_timer_delack, tp); 2509 } 2510 INP_UNLOCK(inp); 2511 return; 2512 2513 dropafterack: 2514 KASSERT(headlocked, ("tcp_input: dropafterack: head not locked")); 2515 /* 2516 * Generate an ACK dropping incoming segment if it occupies 2517 * sequence space, where the ACK reflects our state. 2518 * 2519 * We can now skip the test for the RST flag since all 2520 * paths to this code happen after packets containing 2521 * RST have been dropped. 2522 * 2523 * In the SYN-RECEIVED state, don't send an ACK unless the 2524 * segment we received passes the SYN-RECEIVED ACK test. 2525 * If it fails send a RST. This breaks the loop in the 2526 * "LAND" DoS attack, and also prevents an ACK storm 2527 * between two listening ports that have been sent forged 2528 * SYN segments, each with the source address of the other. 2529 */ 2530 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2531 (SEQ_GT(tp->snd_una, th->th_ack) || 2532 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2533 rstreason = BANDLIM_RST_OPENPORT; 2534 goto dropwithreset; 2535 } 2536 #ifdef TCPDEBUG 2537 if (so->so_options & SO_DEBUG) 2538 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2539 &tcp_savetcp, 0); 2540 #endif 2541 KASSERT(headlocked, ("headlocked should be 1")); 2542 INP_INFO_WUNLOCK(&tcbinfo); 2543 tp->t_flags |= TF_ACKNOW; 2544 (void) tcp_output(tp); 2545 INP_UNLOCK(inp); 2546 m_freem(m); 2547 return; 2548 2549 dropwithreset: 2550 KASSERT(headlocked, ("tcp_input: dropwithreset: head not locked")); 2551 /* 2552 * Generate a RST, dropping incoming segment. 2553 * Make ACK acceptable to originator of segment. 2554 * Don't bother to respond if destination was broadcast/multicast. 2555 */ 2556 if ((thflags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2557 goto drop; 2558 if (isipv6) { 2559 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2560 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2561 goto drop; 2562 } else { 2563 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2564 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2565 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2566 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2567 goto drop; 2568 } 2569 /* IPv6 anycast check is done at tcp6_input() */ 2570 2571 /* 2572 * Perform bandwidth limiting. 2573 */ 2574 if (badport_bandlim(rstreason) < 0) 2575 goto drop; 2576 2577 #ifdef TCPDEBUG 2578 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2579 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2580 &tcp_savetcp, 0); 2581 #endif 2582 2583 if (thflags & TH_ACK) 2584 /* mtod() below is safe as long as hdr dropping is delayed */ 2585 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, 2586 TH_RST); 2587 else { 2588 if (thflags & TH_SYN) 2589 tlen++; 2590 /* mtod() below is safe as long as hdr dropping is delayed */ 2591 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2592 (tcp_seq)0, TH_RST|TH_ACK); 2593 } 2594 2595 if (tp != NULL) 2596 INP_UNLOCK(inp); 2597 if (headlocked) 2598 INP_INFO_WUNLOCK(&tcbinfo); 2599 return; 2600 2601 drop: 2602 /* 2603 * Drop space held by incoming segment and return. 2604 */ 2605 #ifdef TCPDEBUG 2606 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2607 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2608 &tcp_savetcp, 0); 2609 #endif 2610 if (tp != NULL) 2611 INP_UNLOCK(inp); 2612 if (headlocked) 2613 INP_INFO_WUNLOCK(&tcbinfo); 2614 m_freem(m); 2615 return; 2616 } 2617 2618 /* 2619 * Parse TCP options and place in tcpopt. 2620 */ 2621 static void 2622 tcp_dooptions(to, cp, cnt, flags) 2623 struct tcpopt *to; 2624 u_char *cp; 2625 int cnt; 2626 int flags; 2627 { 2628 int opt, optlen; 2629 2630 to->to_flags = 0; 2631 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2632 opt = cp[0]; 2633 if (opt == TCPOPT_EOL) 2634 break; 2635 if (opt == TCPOPT_NOP) 2636 optlen = 1; 2637 else { 2638 if (cnt < 2) 2639 break; 2640 optlen = cp[1]; 2641 if (optlen < 2 || optlen > cnt) 2642 break; 2643 } 2644 switch (opt) { 2645 case TCPOPT_MAXSEG: 2646 if (optlen != TCPOLEN_MAXSEG) 2647 continue; 2648 if (!(flags & TO_SYN)) 2649 continue; 2650 to->to_flags |= TOF_MSS; 2651 bcopy((char *)cp + 2, 2652 (char *)&to->to_mss, sizeof(to->to_mss)); 2653 to->to_mss = ntohs(to->to_mss); 2654 break; 2655 case TCPOPT_WINDOW: 2656 if (optlen != TCPOLEN_WINDOW) 2657 continue; 2658 if (!(flags & TO_SYN)) 2659 continue; 2660 to->to_flags |= TOF_SCALE; 2661 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT); 2662 break; 2663 case TCPOPT_TIMESTAMP: 2664 if (optlen != TCPOLEN_TIMESTAMP) 2665 continue; 2666 to->to_flags |= TOF_TS; 2667 bcopy((char *)cp + 2, 2668 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2669 to->to_tsval = ntohl(to->to_tsval); 2670 bcopy((char *)cp + 6, 2671 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2672 to->to_tsecr = ntohl(to->to_tsecr); 2673 break; 2674 #ifdef TCP_SIGNATURE 2675 /* 2676 * XXX In order to reply to a host which has set the 2677 * TCP_SIGNATURE option in its initial SYN, we have to 2678 * record the fact that the option was observed here 2679 * for the syncache code to perform the correct response. 2680 */ 2681 case TCPOPT_SIGNATURE: 2682 if (optlen != TCPOLEN_SIGNATURE) 2683 continue; 2684 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN); 2685 break; 2686 #endif 2687 case TCPOPT_SACK_PERMITTED: 2688 if (optlen != TCPOLEN_SACK_PERMITTED) 2689 continue; 2690 if (!(flags & TO_SYN)) 2691 continue; 2692 if (!tcp_do_sack) 2693 continue; 2694 to->to_flags |= TOF_SACK; 2695 break; 2696 case TCPOPT_SACK: 2697 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2698 continue; 2699 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2700 to->to_sacks = cp + 2; 2701 tcpstat.tcps_sack_rcv_blocks++; 2702 break; 2703 default: 2704 continue; 2705 } 2706 } 2707 } 2708 2709 /* 2710 * Pull out of band byte out of a segment so 2711 * it doesn't appear in the user's data queue. 2712 * It is still reflected in the segment length for 2713 * sequencing purposes. 2714 */ 2715 static void 2716 tcp_pulloutofband(so, th, m, off) 2717 struct socket *so; 2718 struct tcphdr *th; 2719 register struct mbuf *m; 2720 int off; /* delayed to be droped hdrlen */ 2721 { 2722 int cnt = off + th->th_urp - 1; 2723 2724 while (cnt >= 0) { 2725 if (m->m_len > cnt) { 2726 char *cp = mtod(m, caddr_t) + cnt; 2727 struct tcpcb *tp = sototcpcb(so); 2728 2729 tp->t_iobc = *cp; 2730 tp->t_oobflags |= TCPOOB_HAVEDATA; 2731 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2732 m->m_len--; 2733 if (m->m_flags & M_PKTHDR) 2734 m->m_pkthdr.len--; 2735 return; 2736 } 2737 cnt -= m->m_len; 2738 m = m->m_next; 2739 if (m == 0) 2740 break; 2741 } 2742 panic("tcp_pulloutofband"); 2743 } 2744 2745 /* 2746 * Collect new round-trip time estimate 2747 * and update averages and current timeout. 2748 */ 2749 static void 2750 tcp_xmit_timer(tp, rtt) 2751 register struct tcpcb *tp; 2752 int rtt; 2753 { 2754 register int delta; 2755 2756 INP_LOCK_ASSERT(tp->t_inpcb); 2757 2758 tcpstat.tcps_rttupdated++; 2759 tp->t_rttupdated++; 2760 if (tp->t_srtt != 0) { 2761 /* 2762 * srtt is stored as fixed point with 5 bits after the 2763 * binary point (i.e., scaled by 8). The following magic 2764 * is equivalent to the smoothing algorithm in rfc793 with 2765 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2766 * point). Adjust rtt to origin 0. 2767 */ 2768 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2769 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2770 2771 if ((tp->t_srtt += delta) <= 0) 2772 tp->t_srtt = 1; 2773 2774 /* 2775 * We accumulate a smoothed rtt variance (actually, a 2776 * smoothed mean difference), then set the retransmit 2777 * timer to smoothed rtt + 4 times the smoothed variance. 2778 * rttvar is stored as fixed point with 4 bits after the 2779 * binary point (scaled by 16). The following is 2780 * equivalent to rfc793 smoothing with an alpha of .75 2781 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2782 * rfc793's wired-in beta. 2783 */ 2784 if (delta < 0) 2785 delta = -delta; 2786 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2787 if ((tp->t_rttvar += delta) <= 0) 2788 tp->t_rttvar = 1; 2789 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2790 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2791 } else { 2792 /* 2793 * No rtt measurement yet - use the unsmoothed rtt. 2794 * Set the variance to half the rtt (so our first 2795 * retransmit happens at 3*rtt). 2796 */ 2797 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2798 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2799 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2800 } 2801 tp->t_rtttime = 0; 2802 tp->t_rxtshift = 0; 2803 2804 /* 2805 * the retransmit should happen at rtt + 4 * rttvar. 2806 * Because of the way we do the smoothing, srtt and rttvar 2807 * will each average +1/2 tick of bias. When we compute 2808 * the retransmit timer, we want 1/2 tick of rounding and 2809 * 1 extra tick because of +-1/2 tick uncertainty in the 2810 * firing of the timer. The bias will give us exactly the 2811 * 1.5 tick we need. But, because the bias is 2812 * statistical, we have to test that we don't drop below 2813 * the minimum feasible timer (which is 2 ticks). 2814 */ 2815 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2816 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2817 2818 /* 2819 * We received an ack for a packet that wasn't retransmitted; 2820 * it is probably safe to discard any error indications we've 2821 * received recently. This isn't quite right, but close enough 2822 * for now (a route might have failed after we sent a segment, 2823 * and the return path might not be symmetrical). 2824 */ 2825 tp->t_softerror = 0; 2826 } 2827 2828 /* 2829 * Determine a reasonable value for maxseg size. 2830 * If the route is known, check route for mtu. 2831 * If none, use an mss that can be handled on the outgoing 2832 * interface without forcing IP to fragment; if bigger than 2833 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2834 * to utilize large mbufs. If no route is found, route has no mtu, 2835 * or the destination isn't local, use a default, hopefully conservative 2836 * size (usually 512 or the default IP max size, but no more than the mtu 2837 * of the interface), as we can't discover anything about intervening 2838 * gateways or networks. We also initialize the congestion/slow start 2839 * window to be a single segment if the destination isn't local. 2840 * While looking at the routing entry, we also initialize other path-dependent 2841 * parameters from pre-set or cached values in the routing entry. 2842 * 2843 * Also take into account the space needed for options that we 2844 * send regularly. Make maxseg shorter by that amount to assure 2845 * that we can send maxseg amount of data even when the options 2846 * are present. Store the upper limit of the length of options plus 2847 * data in maxopd. 2848 * 2849 * 2850 * In case of T/TCP, we call this routine during implicit connection 2851 * setup as well (offer = -1), to initialize maxseg from the cached 2852 * MSS of our peer. 2853 * 2854 * NOTE that this routine is only called when we process an incoming 2855 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2856 */ 2857 void 2858 tcp_mss(tp, offer) 2859 struct tcpcb *tp; 2860 int offer; 2861 { 2862 int rtt, mss; 2863 u_long bufsize; 2864 u_long maxmtu; 2865 struct inpcb *inp = tp->t_inpcb; 2866 struct socket *so; 2867 struct hc_metrics_lite metrics; 2868 int origoffer = offer; 2869 int mtuflags = 0; 2870 #ifdef INET6 2871 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2872 size_t min_protoh = isipv6 ? 2873 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2874 sizeof (struct tcpiphdr); 2875 #else 2876 const size_t min_protoh = sizeof(struct tcpiphdr); 2877 #endif 2878 2879 /* initialize */ 2880 #ifdef INET6 2881 if (isipv6) { 2882 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags); 2883 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2884 } else 2885 #endif 2886 { 2887 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags); 2888 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2889 } 2890 so = inp->inp_socket; 2891 2892 /* 2893 * no route to sender, stay with default mss and return 2894 */ 2895 if (maxmtu == 0) 2896 return; 2897 2898 /* what have we got? */ 2899 switch (offer) { 2900 case 0: 2901 /* 2902 * Offer == 0 means that there was no MSS on the SYN 2903 * segment, in this case we use tcp_mssdflt. 2904 */ 2905 offer = 2906 #ifdef INET6 2907 isipv6 ? tcp_v6mssdflt : 2908 #endif 2909 tcp_mssdflt; 2910 break; 2911 2912 case -1: 2913 /* 2914 * Offer == -1 means that we didn't receive SYN yet. 2915 */ 2916 /* FALLTHROUGH */ 2917 2918 default: 2919 /* 2920 * Prevent DoS attack with too small MSS. Round up 2921 * to at least minmss. 2922 */ 2923 offer = max(offer, tcp_minmss); 2924 /* 2925 * Sanity check: make sure that maxopd will be large 2926 * enough to allow some data on segments even if the 2927 * all the option space is used (40bytes). Otherwise 2928 * funny things may happen in tcp_output. 2929 */ 2930 offer = max(offer, 64); 2931 } 2932 2933 /* 2934 * rmx information is now retrieved from tcp_hostcache 2935 */ 2936 tcp_hc_get(&inp->inp_inc, &metrics); 2937 2938 /* 2939 * if there's a discovered mtu int tcp hostcache, use it 2940 * else, use the link mtu. 2941 */ 2942 if (metrics.rmx_mtu) 2943 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2944 else { 2945 #ifdef INET6 2946 if (isipv6) { 2947 mss = maxmtu - min_protoh; 2948 if (!path_mtu_discovery && 2949 !in6_localaddr(&inp->in6p_faddr)) 2950 mss = min(mss, tcp_v6mssdflt); 2951 } else 2952 #endif 2953 { 2954 mss = maxmtu - min_protoh; 2955 if (!path_mtu_discovery && 2956 !in_localaddr(inp->inp_faddr)) 2957 mss = min(mss, tcp_mssdflt); 2958 } 2959 } 2960 mss = min(mss, offer); 2961 2962 /* 2963 * maxopd stores the maximum length of data AND options 2964 * in a segment; maxseg is the amount of data in a normal 2965 * segment. We need to store this value (maxopd) apart 2966 * from maxseg, because now every segment carries options 2967 * and thus we normally have somewhat less data in segments. 2968 */ 2969 tp->t_maxopd = mss; 2970 2971 /* 2972 * origoffer==-1 indicates, that no segments were received yet. 2973 * In this case we just guess. 2974 */ 2975 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2976 (origoffer == -1 || 2977 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2978 mss -= TCPOLEN_TSTAMP_APPA; 2979 tp->t_maxseg = mss; 2980 2981 #if (MCLBYTES & (MCLBYTES - 1)) == 0 2982 if (mss > MCLBYTES) 2983 mss &= ~(MCLBYTES-1); 2984 #else 2985 if (mss > MCLBYTES) 2986 mss = mss / MCLBYTES * MCLBYTES; 2987 #endif 2988 tp->t_maxseg = mss; 2989 2990 /* 2991 * If there's a pipesize, change the socket buffer to that size, 2992 * don't change if sb_hiwat is different than default (then it 2993 * has been changed on purpose with setsockopt). 2994 * Make the socket buffers an integral number of mss units; 2995 * if the mss is larger than the socket buffer, decrease the mss. 2996 */ 2997 SOCKBUF_LOCK(&so->so_snd); 2998 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2999 bufsize = metrics.rmx_sendpipe; 3000 else 3001 bufsize = so->so_snd.sb_hiwat; 3002 if (bufsize < mss) 3003 mss = bufsize; 3004 else { 3005 bufsize = roundup(bufsize, mss); 3006 if (bufsize > sb_max) 3007 bufsize = sb_max; 3008 if (bufsize > so->so_snd.sb_hiwat) 3009 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3010 } 3011 SOCKBUF_UNLOCK(&so->so_snd); 3012 tp->t_maxseg = mss; 3013 3014 SOCKBUF_LOCK(&so->so_rcv); 3015 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3016 bufsize = metrics.rmx_recvpipe; 3017 else 3018 bufsize = so->so_rcv.sb_hiwat; 3019 if (bufsize > mss) { 3020 bufsize = roundup(bufsize, mss); 3021 if (bufsize > sb_max) 3022 bufsize = sb_max; 3023 if (bufsize > so->so_rcv.sb_hiwat) 3024 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3025 } 3026 SOCKBUF_UNLOCK(&so->so_rcv); 3027 /* 3028 * While we're here, check the others too 3029 */ 3030 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3031 tp->t_srtt = rtt; 3032 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3033 tcpstat.tcps_usedrtt++; 3034 if (metrics.rmx_rttvar) { 3035 tp->t_rttvar = metrics.rmx_rttvar; 3036 tcpstat.tcps_usedrttvar++; 3037 } else { 3038 /* default variation is +- 1 rtt */ 3039 tp->t_rttvar = 3040 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3041 } 3042 TCPT_RANGESET(tp->t_rxtcur, 3043 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3044 tp->t_rttmin, TCPTV_REXMTMAX); 3045 } 3046 if (metrics.rmx_ssthresh) { 3047 /* 3048 * There's some sort of gateway or interface 3049 * buffer limit on the path. Use this to set 3050 * the slow start threshhold, but set the 3051 * threshold to no less than 2*mss. 3052 */ 3053 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3054 tcpstat.tcps_usedssthresh++; 3055 } 3056 if (metrics.rmx_bandwidth) 3057 tp->snd_bandwidth = metrics.rmx_bandwidth; 3058 3059 /* 3060 * Set the slow-start flight size depending on whether this 3061 * is a local network or not. 3062 * 3063 * Extend this so we cache the cwnd too and retrieve it here. 3064 * Make cwnd even bigger than RFC3390 suggests but only if we 3065 * have previous experience with the remote host. Be careful 3066 * not make cwnd bigger than remote receive window or our own 3067 * send socket buffer. Maybe put some additional upper bound 3068 * on the retrieved cwnd. Should do incremental updates to 3069 * hostcache when cwnd collapses so next connection doesn't 3070 * overloads the path again. 3071 * 3072 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3073 * We currently check only in syncache_socket for that. 3074 */ 3075 #define TCP_METRICS_CWND 3076 #ifdef TCP_METRICS_CWND 3077 if (metrics.rmx_cwnd) 3078 tp->snd_cwnd = max(mss, 3079 min(metrics.rmx_cwnd / 2, 3080 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3081 else 3082 #endif 3083 if (tcp_do_rfc3390) 3084 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3085 #ifdef INET6 3086 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3087 (!isipv6 && in_localaddr(inp->inp_faddr))) 3088 #else 3089 else if (in_localaddr(inp->inp_faddr)) 3090 #endif 3091 tp->snd_cwnd = mss * ss_fltsz_local; 3092 else 3093 tp->snd_cwnd = mss * ss_fltsz; 3094 3095 /* Check the interface for TSO capabilities. */ 3096 if (mtuflags & CSUM_TSO) 3097 tp->t_flags |= TF_TSO; 3098 } 3099 3100 /* 3101 * Determine the MSS option to send on an outgoing SYN. 3102 */ 3103 int 3104 tcp_mssopt(inc) 3105 struct in_conninfo *inc; 3106 { 3107 int mss = 0; 3108 u_long maxmtu = 0; 3109 u_long thcmtu = 0; 3110 size_t min_protoh; 3111 #ifdef INET6 3112 int isipv6 = inc->inc_isipv6 ? 1 : 0; 3113 #endif 3114 3115 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3116 3117 #ifdef INET6 3118 if (isipv6) { 3119 mss = tcp_v6mssdflt; 3120 maxmtu = tcp_maxmtu6(inc, NULL); 3121 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3122 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3123 } else 3124 #endif 3125 { 3126 mss = tcp_mssdflt; 3127 maxmtu = tcp_maxmtu(inc, NULL); 3128 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3129 min_protoh = sizeof(struct tcpiphdr); 3130 } 3131 if (maxmtu && thcmtu) 3132 mss = min(maxmtu, thcmtu) - min_protoh; 3133 else if (maxmtu || thcmtu) 3134 mss = max(maxmtu, thcmtu) - min_protoh; 3135 3136 return (mss); 3137 } 3138 3139 3140 /* 3141 * On a partial ack arrives, force the retransmission of the 3142 * next unacknowledged segment. Do not clear tp->t_dupacks. 3143 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3144 * be started again. 3145 */ 3146 static void 3147 tcp_newreno_partial_ack(tp, th) 3148 struct tcpcb *tp; 3149 struct tcphdr *th; 3150 { 3151 tcp_seq onxt = tp->snd_nxt; 3152 u_long ocwnd = tp->snd_cwnd; 3153 3154 callout_stop(tp->tt_rexmt); 3155 tp->t_rtttime = 0; 3156 tp->snd_nxt = th->th_ack; 3157 /* 3158 * Set snd_cwnd to one segment beyond acknowledged offset. 3159 * (tp->snd_una has not yet been updated when this function is called.) 3160 */ 3161 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3162 tp->t_flags |= TF_ACKNOW; 3163 (void) tcp_output(tp); 3164 tp->snd_cwnd = ocwnd; 3165 if (SEQ_GT(onxt, tp->snd_nxt)) 3166 tp->snd_nxt = onxt; 3167 /* 3168 * Partial window deflation. Relies on fact that tp->snd_una 3169 * not updated yet. 3170 */ 3171 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3172 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3173 else 3174 tp->snd_cwnd = 0; 3175 tp->snd_cwnd += tp->t_maxseg; 3176 } 3177 3178 /* 3179 * Returns 1 if the TIME_WAIT state was killed and we should start over, 3180 * looking for a pcb in the listen state. Returns 0 otherwise. 3181 */ 3182 static int 3183 tcp_timewait(inp, to, th, m, tlen) 3184 struct inpcb *inp; 3185 struct tcpopt *to; 3186 struct tcphdr *th; 3187 struct mbuf *m; 3188 int tlen; 3189 { 3190 struct tcptw *tw; 3191 int thflags; 3192 tcp_seq seq; 3193 #ifdef INET6 3194 int isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 3195 #else 3196 const int isipv6 = 0; 3197 #endif 3198 3199 /* tcbinfo lock required for tcp_twclose(), tcp_timer_2msl_reset(). */ 3200 INP_INFO_WLOCK_ASSERT(&tcbinfo); 3201 INP_LOCK_ASSERT(inp); 3202 3203 /* 3204 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is 3205 * still present. This is undesirable, but temporarily necessary 3206 * until we work out how to handle inpcb's who's timewait state has 3207 * been removed. 3208 */ 3209 tw = intotw(inp); 3210 if (tw == NULL) 3211 goto drop; 3212 3213 thflags = th->th_flags; 3214 3215 /* 3216 * NOTE: for FIN_WAIT_2 (to be added later), 3217 * must validate sequence number before accepting RST 3218 */ 3219 3220 /* 3221 * If the segment contains RST: 3222 * Drop the segment - see Stevens, vol. 2, p. 964 and 3223 * RFC 1337. 3224 */ 3225 if (thflags & TH_RST) 3226 goto drop; 3227 3228 #if 0 3229 /* PAWS not needed at the moment */ 3230 /* 3231 * RFC 1323 PAWS: If we have a timestamp reply on this segment 3232 * and it's less than ts_recent, drop it. 3233 */ 3234 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 3235 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 3236 if ((thflags & TH_ACK) == 0) 3237 goto drop; 3238 goto ack; 3239 } 3240 /* 3241 * ts_recent is never updated because we never accept new segments. 3242 */ 3243 #endif 3244 3245 /* 3246 * If a new connection request is received 3247 * while in TIME_WAIT, drop the old connection 3248 * and start over if the sequence numbers 3249 * are above the previous ones. 3250 */ 3251 if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) { 3252 tcp_twclose(tw, 0); 3253 return (1); 3254 } 3255 3256 /* 3257 * Drop the the segment if it does not contain an ACK. 3258 */ 3259 if ((thflags & TH_ACK) == 0) 3260 goto drop; 3261 3262 /* 3263 * Reset the 2MSL timer if this is a duplicate FIN. 3264 */ 3265 if (thflags & TH_FIN) { 3266 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 3267 if (seq + 1 == tw->rcv_nxt) 3268 tcp_timer_2msl_reset(tw, 1); 3269 } 3270 3271 /* 3272 * Acknowledge the segment if it has data or is not a duplicate ACK. 3273 */ 3274 if (thflags != TH_ACK || tlen != 0 || 3275 th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt) 3276 tcp_twrespond(tw, TH_ACK); 3277 goto drop; 3278 3279 /* 3280 * Generate a RST, dropping incoming segment. 3281 * Make ACK acceptable to originator of segment. 3282 * Don't bother to respond if destination was broadcast/multicast. 3283 */ 3284 if (m->m_flags & (M_BCAST|M_MCAST)) 3285 goto drop; 3286 if (isipv6) { 3287 struct ip6_hdr *ip6; 3288 3289 /* IPv6 anycast check is done at tcp6_input() */ 3290 ip6 = mtod(m, struct ip6_hdr *); 3291 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3292 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3293 goto drop; 3294 } else { 3295 struct ip *ip; 3296 3297 ip = mtod(m, struct ip *); 3298 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3299 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3300 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3301 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3302 goto drop; 3303 } 3304 if (thflags & TH_ACK) { 3305 tcp_respond(NULL, 3306 mtod(m, void *), th, m, 0, th->th_ack, TH_RST); 3307 } else { 3308 seq = th->th_seq + (thflags & TH_SYN ? 1 : 0); 3309 tcp_respond(NULL, 3310 mtod(m, void *), th, m, seq, 0, TH_RST|TH_ACK); 3311 } 3312 INP_UNLOCK(inp); 3313 return (0); 3314 3315 drop: 3316 INP_UNLOCK(inp); 3317 m_freem(m); 3318 return (0); 3319 } 3320