tcp_input.c revision 192912
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 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: head/sys/netinet/tcp_input.c 192912 2009-05-27 17:02:10Z zml $"); 34 35#include "opt_ipfw.h" /* for ipfw_fwd */ 36#include "opt_inet.h" 37#include "opt_inet6.h" 38#include "opt_ipsec.h" 39#include "opt_mac.h" 40#include "opt_tcpdebug.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#include <sys/vimage.h> 55 56#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 57 58#include <vm/uma.h> 59 60#include <net/if.h> 61#include <net/route.h> 62 63#define TCPSTATES /* for logging */ 64 65#include <netinet/in.h> 66#include <netinet/in_pcb.h> 67#include <netinet/in_systm.h> 68#include <netinet/in_var.h> 69#include <netinet/ip.h> 70#include <netinet/ip_icmp.h> /* required for icmp_var.h */ 71#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 72#include <netinet/ip_var.h> 73#include <netinet/ip_options.h> 74#include <netinet/ip6.h> 75#include <netinet/icmp6.h> 76#include <netinet6/in6_pcb.h> 77#include <netinet6/ip6_var.h> 78#include <netinet6/nd6.h> 79#include <netinet/tcp.h> 80#include <netinet/tcp_fsm.h> 81#include <netinet/tcp_seq.h> 82#include <netinet/tcp_timer.h> 83#include <netinet/tcp_var.h> 84#include <netinet6/tcp6_var.h> 85#include <netinet/tcpip.h> 86#include <netinet/tcp_syncache.h> 87#ifdef TCPDEBUG 88#include <netinet/tcp_debug.h> 89#endif /* TCPDEBUG */ 90#include <netinet/vinet.h> 91 92#ifdef INET6 93#include <netinet6/vinet6.h> 94#endif 95 96#ifdef IPSEC 97#include <netipsec/ipsec.h> 98#include <netipsec/ipsec6.h> 99#endif /*IPSEC*/ 100 101#include <machine/in_cksum.h> 102 103#include <security/mac/mac_framework.h> 104 105static const int tcprexmtthresh = 3; 106 107#ifdef VIMAGE_GLOBALS 108struct tcpstat tcpstat; 109int blackhole; 110int tcp_delack_enabled; 111int drop_synfin; 112int tcp_do_rfc3042; 113int tcp_do_rfc3390; 114int tcp_do_ecn; 115int tcp_ecn_maxretries; 116int tcp_insecure_rst; 117int tcp_do_autorcvbuf; 118int tcp_autorcvbuf_inc; 119int tcp_autorcvbuf_max; 120int tcp_do_rfc3465; 121int tcp_abc_l_var; 122#endif 123 124SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_STATS, stats, 125 CTLFLAG_RW, tcpstat , tcpstat, 126 "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 127 128int tcp_log_in_vain = 0; 129SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 130 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports"); 131 132SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 133 blackhole, 0, "Do not send RST on segments to closed ports"); 134 135SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, delayed_ack, 136 CTLFLAG_RW, tcp_delack_enabled, 0, 137 "Delay ACK to try and piggyback it onto a data packet"); 138 139SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, drop_synfin, 140 CTLFLAG_RW, drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 141 142SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 143 tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 144 145SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 146 tcp_do_rfc3390, 0, 147 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 148 149SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 150 tcp_do_rfc3465, 0, 151 "Enable RFC 3465 (Appropriate Byte Counting)"); 152SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 153 tcp_abc_l_var, 2, 154 "Cap the max cwnd increment during slow-start to this number of segments"); 155 156SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 157SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_ecn, OID_AUTO, enable, 158 CTLFLAG_RW, tcp_do_ecn, 0, "TCP ECN support"); 159SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_ecn, OID_AUTO, maxretries, 160 CTLFLAG_RW, tcp_ecn_maxretries, 0, "Max retries before giving up on ECN"); 161 162SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, insecure_rst, 163 CTLFLAG_RW, tcp_insecure_rst, 0, 164 "Follow the old (insecure) criteria for accepting RST packets"); 165 166SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, recvbuf_auto, 167 CTLFLAG_RW, tcp_do_autorcvbuf, 0, 168 "Enable automatic receive buffer sizing"); 169 170SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, recvbuf_inc, 171 CTLFLAG_RW, tcp_autorcvbuf_inc, 0, 172 "Incrementor step size of automatic receive buffer"); 173 174SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, recvbuf_max, 175 CTLFLAG_RW, tcp_autorcvbuf_max, 0, 176 "Max size of automatic receive buffer"); 177 178int tcp_read_locking = 1; 179SYSCTL_INT(_net_inet_tcp, OID_AUTO, read_locking, CTLFLAG_RW, 180 &tcp_read_locking, 0, "Enable read locking strategy"); 181 182int tcp_rlock_atfirst; 183SYSCTL_INT(_net_inet_tcp, OID_AUTO, rlock_atfirst, CTLFLAG_RD, 184 &tcp_rlock_atfirst, 0, ""); 185 186int tcp_wlock_atfirst; 187SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_wlock_atfirst, CTLFLAG_RD, 188 &tcp_wlock_atfirst, 0, ""); 189 190int tcp_wlock_upgraded; 191SYSCTL_INT(_net_inet_tcp, OID_AUTO, wlock_upgraded, CTLFLAG_RD, 192 &tcp_wlock_upgraded, 0, ""); 193 194int tcp_wlock_relocked; 195SYSCTL_INT(_net_inet_tcp, OID_AUTO, wlock_relocked, CTLFLAG_RD, 196 &tcp_wlock_relocked, 0, ""); 197 198int tcp_wlock_looped; 199SYSCTL_INT(_net_inet_tcp, OID_AUTO, wlock_looped, CTLFLAG_RD, 200 &tcp_wlock_looped, 0, ""); 201 202#ifdef VIMAGE_GLOBALS 203struct inpcbhead tcb; 204struct inpcbinfo tcbinfo; 205#endif 206#define tcb6 tcb /* for KAME src sync over BSD*'s */ 207 208static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 209static void tcp_do_segment(struct mbuf *, struct tcphdr *, 210 struct socket *, struct tcpcb *, int, int, uint8_t, 211 int); 212static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 213 struct tcpcb *, int, int); 214static void tcp_pulloutofband(struct socket *, 215 struct tcphdr *, struct mbuf *, int); 216static void tcp_xmit_timer(struct tcpcb *, int); 217static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 218static void inline 219 tcp_congestion_exp(struct tcpcb *); 220 221static void inline 222tcp_congestion_exp(struct tcpcb *tp) 223{ 224 u_int win; 225 226 win = min(tp->snd_wnd, tp->snd_cwnd) / 227 2 / tp->t_maxseg; 228 if (win < 2) 229 win = 2; 230 tp->snd_ssthresh = win * tp->t_maxseg; 231 ENTER_FASTRECOVERY(tp); 232 tp->snd_recover = tp->snd_max; 233 if (tp->t_flags & TF_ECN_PERMIT) 234 tp->t_flags |= TF_ECN_SND_CWR; 235} 236 237/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 238#ifdef INET6 239#define ND6_HINT(tp) \ 240do { \ 241 if ((tp) && (tp)->t_inpcb && \ 242 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 243 nd6_nud_hint(NULL, NULL, 0); \ 244} while (0) 245#else 246#define ND6_HINT(tp) 247#endif 248 249/* 250 * Indicate whether this ack should be delayed. We can delay the ack if 251 * - there is no delayed ack timer in progress and 252 * - our last ack wasn't a 0-sized window. We never want to delay 253 * the ack that opens up a 0-sized window and 254 * - delayed acks are enabled or 255 * - this is a half-synchronized T/TCP connection. 256 */ 257#define DELAY_ACK(tp) \ 258 ((!tcp_timer_active(tp, TT_DELACK) && \ 259 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 260 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 261 262/* 263 * TCP input handling is split into multiple parts: 264 * tcp6_input is a thin wrapper around tcp_input for the extended 265 * ip6_protox[] call format in ip6_input 266 * tcp_input handles primary segment validation, inpcb lookup and 267 * SYN processing on listen sockets 268 * tcp_do_segment processes the ACK and text of the segment for 269 * establishing, established and closing connections 270 */ 271#ifdef INET6 272int 273tcp6_input(struct mbuf **mp, int *offp, int proto) 274{ 275 INIT_VNET_INET6(curvnet); 276 struct mbuf *m = *mp; 277 struct in6_ifaddr *ia6; 278 279 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 280 281 /* 282 * draft-itojun-ipv6-tcp-to-anycast 283 * better place to put this in? 284 */ 285 ia6 = ip6_getdstifaddr(m); 286 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 287 struct ip6_hdr *ip6; 288 289 ip6 = mtod(m, struct ip6_hdr *); 290 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 291 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 292 return IPPROTO_DONE; 293 } 294 295 tcp_input(m, *offp); 296 return IPPROTO_DONE; 297} 298#endif 299 300void 301tcp_input(struct mbuf *m, int off0) 302{ 303 INIT_VNET_INET(curvnet); 304#ifdef INET6 305 INIT_VNET_INET6(curvnet); 306#endif 307#ifdef IPSEC 308 INIT_VNET_IPSEC(curvnet); 309#endif 310 struct tcphdr *th; 311 struct ip *ip = NULL; 312 struct ipovly *ipov; 313 struct inpcb *inp = NULL; 314 struct tcpcb *tp = NULL; 315 struct socket *so = NULL; 316 u_char *optp = NULL; 317 int optlen = 0; 318 int len, tlen, off; 319 int drop_hdrlen; 320 int thflags; 321 int rstreason = 0; /* For badport_bandlim accounting purposes */ 322 uint8_t iptos; 323#ifdef IPFIREWALL_FORWARD 324 struct m_tag *fwd_tag; 325#endif 326#ifdef INET6 327 struct ip6_hdr *ip6 = NULL; 328 int isipv6; 329#else 330 const void *ip6 = NULL; 331 const int isipv6 = 0; 332#endif 333 struct tcpopt to; /* options in this segment */ 334 char *s = NULL; /* address and port logging */ 335 int ti_locked; 336#define TI_UNLOCKED 1 337#define TI_RLOCKED 2 338#define TI_WLOCKED 3 339 340#ifdef TCPDEBUG 341 /* 342 * The size of tcp_saveipgen must be the size of the max ip header, 343 * now IPv6. 344 */ 345 u_char tcp_saveipgen[IP6_HDR_LEN]; 346 struct tcphdr tcp_savetcp; 347 short ostate = 0; 348#endif 349 350#ifdef INET6 351 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 352#endif 353 354 to.to_flags = 0; 355 TCPSTAT_INC(tcps_rcvtotal); 356 357 if (isipv6) { 358#ifdef INET6 359 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 360 ip6 = mtod(m, struct ip6_hdr *); 361 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 362 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 363 TCPSTAT_INC(tcps_rcvbadsum); 364 goto drop; 365 } 366 th = (struct tcphdr *)((caddr_t)ip6 + off0); 367 368 /* 369 * Be proactive about unspecified IPv6 address in source. 370 * As we use all-zero to indicate unbounded/unconnected pcb, 371 * unspecified IPv6 address can be used to confuse us. 372 * 373 * Note that packets with unspecified IPv6 destination is 374 * already dropped in ip6_input. 375 */ 376 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 377 /* XXX stat */ 378 goto drop; 379 } 380#else 381 th = NULL; /* XXX: Avoid compiler warning. */ 382#endif 383 } else { 384 /* 385 * Get IP and TCP header together in first mbuf. 386 * Note: IP leaves IP header in first mbuf. 387 */ 388 if (off0 > sizeof (struct ip)) { 389 ip_stripoptions(m, (struct mbuf *)0); 390 off0 = sizeof(struct ip); 391 } 392 if (m->m_len < sizeof (struct tcpiphdr)) { 393 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 394 == NULL) { 395 TCPSTAT_INC(tcps_rcvshort); 396 return; 397 } 398 } 399 ip = mtod(m, struct ip *); 400 ipov = (struct ipovly *)ip; 401 th = (struct tcphdr *)((caddr_t)ip + off0); 402 tlen = ip->ip_len; 403 404 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 405 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 406 th->th_sum = m->m_pkthdr.csum_data; 407 else 408 th->th_sum = in_pseudo(ip->ip_src.s_addr, 409 ip->ip_dst.s_addr, 410 htonl(m->m_pkthdr.csum_data + 411 ip->ip_len + 412 IPPROTO_TCP)); 413 th->th_sum ^= 0xffff; 414#ifdef TCPDEBUG 415 ipov->ih_len = (u_short)tlen; 416 ipov->ih_len = htons(ipov->ih_len); 417#endif 418 } else { 419 /* 420 * Checksum extended TCP header and data. 421 */ 422 len = sizeof (struct ip) + tlen; 423 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 424 ipov->ih_len = (u_short)tlen; 425 ipov->ih_len = htons(ipov->ih_len); 426 th->th_sum = in_cksum(m, len); 427 } 428 if (th->th_sum) { 429 TCPSTAT_INC(tcps_rcvbadsum); 430 goto drop; 431 } 432 /* Re-initialization for later version check */ 433 ip->ip_v = IPVERSION; 434 } 435 436#ifdef INET6 437 if (isipv6) 438 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 439 else 440#endif 441 iptos = ip->ip_tos; 442 443 /* 444 * Check that TCP offset makes sense, 445 * pull out TCP options and adjust length. XXX 446 */ 447 off = th->th_off << 2; 448 if (off < sizeof (struct tcphdr) || off > tlen) { 449 TCPSTAT_INC(tcps_rcvbadoff); 450 goto drop; 451 } 452 tlen -= off; /* tlen is used instead of ti->ti_len */ 453 if (off > sizeof (struct tcphdr)) { 454 if (isipv6) { 455#ifdef INET6 456 IP6_EXTHDR_CHECK(m, off0, off, ); 457 ip6 = mtod(m, struct ip6_hdr *); 458 th = (struct tcphdr *)((caddr_t)ip6 + off0); 459#endif 460 } else { 461 if (m->m_len < sizeof(struct ip) + off) { 462 if ((m = m_pullup(m, sizeof (struct ip) + off)) 463 == NULL) { 464 TCPSTAT_INC(tcps_rcvshort); 465 return; 466 } 467 ip = mtod(m, struct ip *); 468 ipov = (struct ipovly *)ip; 469 th = (struct tcphdr *)((caddr_t)ip + off0); 470 } 471 } 472 optlen = off - sizeof (struct tcphdr); 473 optp = (u_char *)(th + 1); 474 } 475 thflags = th->th_flags; 476 477 /* 478 * Convert TCP protocol specific fields to host format. 479 */ 480 th->th_seq = ntohl(th->th_seq); 481 th->th_ack = ntohl(th->th_ack); 482 th->th_win = ntohs(th->th_win); 483 th->th_urp = ntohs(th->th_urp); 484 485 /* 486 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 487 */ 488 drop_hdrlen = off0 + off; 489 490 /* 491 * Locate pcb for segment, which requires a lock on tcbinfo. 492 * Optimisticaly acquire a global read lock rather than a write lock 493 * unless header flags necessarily imply a state change. There are 494 * two cases where we might discover later we need a write lock 495 * despite the flags: ACKs moving a connection out of the syncache, 496 * and ACKs for a connection in TIMEWAIT. 497 */ 498 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 499 tcp_read_locking == 0) { 500 INP_INFO_WLOCK(&V_tcbinfo); 501 ti_locked = TI_WLOCKED; 502 tcp_wlock_atfirst++; 503 } else { 504 INP_INFO_RLOCK(&V_tcbinfo); 505 ti_locked = TI_RLOCKED; 506 tcp_rlock_atfirst++; 507 } 508 509findpcb: 510#ifdef INVARIANTS 511 if (ti_locked == TI_RLOCKED) 512 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 513 else if (ti_locked == TI_WLOCKED) 514 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 515 else 516 panic("%s: findpcb ti_locked %d\n", __func__, ti_locked); 517#endif 518 519#ifdef IPFIREWALL_FORWARD 520 /* 521 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 522 */ 523 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 524 525 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 526 struct sockaddr_in *next_hop; 527 528 next_hop = (struct sockaddr_in *)(fwd_tag+1); 529 /* 530 * Transparently forwarded. Pretend to be the destination. 531 * already got one like this? 532 */ 533 inp = in_pcblookup_hash(&V_tcbinfo, 534 ip->ip_src, th->th_sport, 535 ip->ip_dst, th->th_dport, 536 0, m->m_pkthdr.rcvif); 537 if (!inp) { 538 /* It's new. Try to find the ambushing socket. */ 539 inp = in_pcblookup_hash(&V_tcbinfo, 540 ip->ip_src, th->th_sport, 541 next_hop->sin_addr, 542 next_hop->sin_port ? 543 ntohs(next_hop->sin_port) : 544 th->th_dport, 545 INPLOOKUP_WILDCARD, 546 m->m_pkthdr.rcvif); 547 } 548 /* Remove the tag from the packet. We don't need it anymore. */ 549 m_tag_delete(m, fwd_tag); 550 } else 551#endif /* IPFIREWALL_FORWARD */ 552 { 553 if (isipv6) { 554#ifdef INET6 555 inp = in6_pcblookup_hash(&V_tcbinfo, 556 &ip6->ip6_src, th->th_sport, 557 &ip6->ip6_dst, th->th_dport, 558 INPLOOKUP_WILDCARD, 559 m->m_pkthdr.rcvif); 560#endif 561 } else 562 inp = in_pcblookup_hash(&V_tcbinfo, 563 ip->ip_src, th->th_sport, 564 ip->ip_dst, th->th_dport, 565 INPLOOKUP_WILDCARD, 566 m->m_pkthdr.rcvif); 567 } 568 569 /* 570 * If the INPCB does not exist then all data in the incoming 571 * segment is discarded and an appropriate RST is sent back. 572 * XXX MRT Send RST using which routing table? 573 */ 574 if (inp == NULL) { 575 /* 576 * Log communication attempts to ports that are not 577 * in use. 578 */ 579 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 580 tcp_log_in_vain == 2) { 581 if ((s = tcp_log_addrs(NULL, th, (void *)ip, ip6))) 582 log(LOG_INFO, "%s; %s: Connection attempt " 583 "to closed port\n", s, __func__); 584 } 585 /* 586 * When blackholing do not respond with a RST but 587 * completely ignore the segment and drop it. 588 */ 589 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 590 V_blackhole == 2) 591 goto dropunlock; 592 593 rstreason = BANDLIM_RST_CLOSEDPORT; 594 goto dropwithreset; 595 } 596 INP_WLOCK(inp); 597 if (!(inp->inp_flags & INP_HW_FLOWID) 598 && (m->m_flags & M_FLOWID) 599 && ((inp->inp_socket == NULL) 600 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 601 inp->inp_flags |= INP_HW_FLOWID; 602 inp->inp_flags &= ~INP_SW_FLOWID; 603 inp->inp_flowid = m->m_pkthdr.flowid; 604 } 605#ifdef IPSEC 606#ifdef INET6 607 if (isipv6 && ipsec6_in_reject(m, inp)) { 608 V_ipsec6stat.in_polvio++; 609 goto dropunlock; 610 } else 611#endif /* INET6 */ 612 if (ipsec4_in_reject(m, inp) != 0) { 613 V_ipsec4stat.in_polvio++; 614 goto dropunlock; 615 } 616#endif /* IPSEC */ 617 618 /* 619 * Check the minimum TTL for socket. 620 */ 621 if (inp->inp_ip_minttl != 0) { 622#ifdef INET6 623 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 624 goto dropunlock; 625 else 626#endif 627 if (inp->inp_ip_minttl > ip->ip_ttl) 628 goto dropunlock; 629 } 630 631 /* 632 * A previous connection in TIMEWAIT state is supposed to catch stray 633 * or duplicate segments arriving late. If this segment was a 634 * legitimate new connection attempt the old INPCB gets removed and 635 * we can try again to find a listening socket. 636 * 637 * At this point, due to earlier optimism, we may hold a read lock on 638 * the inpcbinfo, rather than a write lock. If so, we need to 639 * upgrade, or if that fails, acquire a reference on the inpcb, drop 640 * all locks, acquire a global write lock, and then re-acquire the 641 * inpcb lock. We may at that point discover that another thread has 642 * tried to free the inpcb, in which case we need to loop back and 643 * try to find a new inpcb to deliver to. 644 */ 645 if (inp->inp_flags & INP_TIMEWAIT) { 646 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 647 ("%s: INP_TIMEWAIT ti_locked %d", __func__, ti_locked)); 648 649 if (ti_locked == TI_RLOCKED) { 650 if (rw_try_upgrade(&V_tcbinfo.ipi_lock) == 0) { 651 in_pcbref(inp); 652 INP_WUNLOCK(inp); 653 INP_INFO_RUNLOCK(&V_tcbinfo); 654 INP_INFO_WLOCK(&V_tcbinfo); 655 ti_locked = TI_WLOCKED; 656 INP_WLOCK(inp); 657 if (in_pcbrele(inp)) { 658 tcp_wlock_looped++; 659 inp = NULL; 660 goto findpcb; 661 } 662 tcp_wlock_relocked++; 663 } else { 664 ti_locked = TI_WLOCKED; 665 tcp_wlock_upgraded++; 666 } 667 } 668 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 669 670 if (thflags & TH_SYN) 671 tcp_dooptions(&to, optp, optlen, TO_SYN); 672 /* 673 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 674 */ 675 if (tcp_twcheck(inp, &to, th, m, tlen)) 676 goto findpcb; 677 INP_INFO_WUNLOCK(&V_tcbinfo); 678 return; 679 } 680 /* 681 * The TCPCB may no longer exist if the connection is winding 682 * down or it is in the CLOSED state. Either way we drop the 683 * segment and send an appropriate response. 684 */ 685 tp = intotcpcb(inp); 686 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 687 rstreason = BANDLIM_RST_CLOSEDPORT; 688 goto dropwithreset; 689 } 690 691 /* 692 * We've identified a valid inpcb, but it could be that we need an 693 * inpcbinfo write lock and have only a read lock. In this case, 694 * attempt to upgrade/relock using the same strategy as the TIMEWAIT 695 * case above. 696 */ 697 if (tp->t_state != TCPS_ESTABLISHED || 698 (thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 699 tcp_read_locking == 0) { 700 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 701 ("%s: upgrade check ti_locked %d", __func__, ti_locked)); 702 703 if (ti_locked == TI_RLOCKED) { 704 if (rw_try_upgrade(&V_tcbinfo.ipi_lock) == 0) { 705 in_pcbref(inp); 706 INP_WUNLOCK(inp); 707 INP_INFO_RUNLOCK(&V_tcbinfo); 708 INP_INFO_WLOCK(&V_tcbinfo); 709 ti_locked = TI_WLOCKED; 710 INP_WLOCK(inp); 711 if (in_pcbrele(inp)) { 712 tcp_wlock_looped++; 713 inp = NULL; 714 goto findpcb; 715 } 716 tcp_wlock_relocked++; 717 } else { 718 ti_locked = TI_WLOCKED; 719 tcp_wlock_upgraded++; 720 } 721 } 722 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 723 } 724 725#ifdef MAC 726 INP_WLOCK_ASSERT(inp); 727 if (mac_inpcb_check_deliver(inp, m)) 728 goto dropunlock; 729#endif 730 so = inp->inp_socket; 731 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 732#ifdef TCPDEBUG 733 if (so->so_options & SO_DEBUG) { 734 ostate = tp->t_state; 735 if (isipv6) { 736#ifdef INET6 737 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 738#endif 739 } else 740 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 741 tcp_savetcp = *th; 742 } 743#endif 744 /* 745 * When the socket is accepting connections (the INPCB is in LISTEN 746 * state) we look into the SYN cache if this is a new connection 747 * attempt or the completion of a previous one. 748 */ 749 if (so->so_options & SO_ACCEPTCONN) { 750 struct in_conninfo inc; 751 752 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 753 "tp not listening", __func__)); 754 755 bzero(&inc, sizeof(inc)); 756#ifdef INET6 757 if (isipv6) { 758 inc.inc_flags |= INC_ISIPV6; 759 inc.inc6_faddr = ip6->ip6_src; 760 inc.inc6_laddr = ip6->ip6_dst; 761 } else 762#endif 763 { 764 inc.inc_faddr = ip->ip_src; 765 inc.inc_laddr = ip->ip_dst; 766 } 767 inc.inc_fport = th->th_sport; 768 inc.inc_lport = th->th_dport; 769 770 /* 771 * Check for an existing connection attempt in syncache if 772 * the flag is only ACK. A successful lookup creates a new 773 * socket appended to the listen queue in SYN_RECEIVED state. 774 */ 775 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 776 /* 777 * Parse the TCP options here because 778 * syncookies need access to the reflected 779 * timestamp. 780 */ 781 tcp_dooptions(&to, optp, optlen, 0); 782 /* 783 * NB: syncache_expand() doesn't unlock 784 * inp and tcpinfo locks. 785 */ 786 if (!syncache_expand(&inc, &to, th, &so, m)) { 787 /* 788 * No syncache entry or ACK was not 789 * for our SYN/ACK. Send a RST. 790 * NB: syncache did its own logging 791 * of the failure cause. 792 */ 793 rstreason = BANDLIM_RST_OPENPORT; 794 goto dropwithreset; 795 } 796 if (so == NULL) { 797 /* 798 * We completed the 3-way handshake 799 * but could not allocate a socket 800 * either due to memory shortage, 801 * listen queue length limits or 802 * global socket limits. Send RST 803 * or wait and have the remote end 804 * retransmit the ACK for another 805 * try. 806 */ 807 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 808 log(LOG_DEBUG, "%s; %s: Listen socket: " 809 "Socket allocation failed due to " 810 "limits or memory shortage, %s\n", 811 s, __func__, 812 V_tcp_sc_rst_sock_fail ? 813 "sending RST" : "try again"); 814 if (V_tcp_sc_rst_sock_fail) { 815 rstreason = BANDLIM_UNLIMITED; 816 goto dropwithreset; 817 } else 818 goto dropunlock; 819 } 820 /* 821 * Socket is created in state SYN_RECEIVED. 822 * Unlock the listen socket, lock the newly 823 * created socket and update the tp variable. 824 */ 825 INP_WUNLOCK(inp); /* listen socket */ 826 inp = sotoinpcb(so); 827 INP_WLOCK(inp); /* new connection */ 828 tp = intotcpcb(inp); 829 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 830 ("%s: ", __func__)); 831 /* 832 * Process the segment and the data it 833 * contains. tcp_do_segment() consumes 834 * the mbuf chain and unlocks the inpcb. 835 */ 836 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 837 iptos, ti_locked); 838 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 839 return; 840 } 841 /* 842 * Segment flag validation for new connection attempts: 843 * 844 * Our (SYN|ACK) response was rejected. 845 * Check with syncache and remove entry to prevent 846 * retransmits. 847 * 848 * NB: syncache_chkrst does its own logging of failure 849 * causes. 850 */ 851 if (thflags & TH_RST) { 852 syncache_chkrst(&inc, th); 853 goto dropunlock; 854 } 855 /* 856 * We can't do anything without SYN. 857 */ 858 if ((thflags & TH_SYN) == 0) { 859 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 860 log(LOG_DEBUG, "%s; %s: Listen socket: " 861 "SYN is missing, segment ignored\n", 862 s, __func__); 863 TCPSTAT_INC(tcps_badsyn); 864 goto dropunlock; 865 } 866 /* 867 * (SYN|ACK) is bogus on a listen socket. 868 */ 869 if (thflags & TH_ACK) { 870 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 871 log(LOG_DEBUG, "%s; %s: Listen socket: " 872 "SYN|ACK invalid, segment rejected\n", 873 s, __func__); 874 syncache_badack(&inc); /* XXX: Not needed! */ 875 TCPSTAT_INC(tcps_badsyn); 876 rstreason = BANDLIM_RST_OPENPORT; 877 goto dropwithreset; 878 } 879 /* 880 * If the drop_synfin option is enabled, drop all 881 * segments with both the SYN and FIN bits set. 882 * This prevents e.g. nmap from identifying the 883 * TCP/IP stack. 884 * XXX: Poor reasoning. nmap has other methods 885 * and is constantly refining its stack detection 886 * strategies. 887 * XXX: This is a violation of the TCP specification 888 * and was used by RFC1644. 889 */ 890 if ((thflags & TH_FIN) && V_drop_synfin) { 891 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 892 log(LOG_DEBUG, "%s; %s: Listen socket: " 893 "SYN|FIN segment ignored (based on " 894 "sysctl setting)\n", s, __func__); 895 TCPSTAT_INC(tcps_badsyn); 896 goto dropunlock; 897 } 898 /* 899 * Segment's flags are (SYN) or (SYN|FIN). 900 * 901 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 902 * as they do not affect the state of the TCP FSM. 903 * The data pointed to by TH_URG and th_urp is ignored. 904 */ 905 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 906 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 907 KASSERT(thflags & (TH_SYN), 908 ("%s: Listen socket: TH_SYN not set", __func__)); 909#ifdef INET6 910 /* 911 * If deprecated address is forbidden, 912 * we do not accept SYN to deprecated interface 913 * address to prevent any new inbound connection from 914 * getting established. 915 * When we do not accept SYN, we send a TCP RST, 916 * with deprecated source address (instead of dropping 917 * it). We compromise it as it is much better for peer 918 * to send a RST, and RST will be the final packet 919 * for the exchange. 920 * 921 * If we do not forbid deprecated addresses, we accept 922 * the SYN packet. RFC2462 does not suggest dropping 923 * SYN in this case. 924 * If we decipher RFC2462 5.5.4, it says like this: 925 * 1. use of deprecated addr with existing 926 * communication is okay - "SHOULD continue to be 927 * used" 928 * 2. use of it with new communication: 929 * (2a) "SHOULD NOT be used if alternate address 930 * with sufficient scope is available" 931 * (2b) nothing mentioned otherwise. 932 * Here we fall into (2b) case as we have no choice in 933 * our source address selection - we must obey the peer. 934 * 935 * The wording in RFC2462 is confusing, and there are 936 * multiple description text for deprecated address 937 * handling - worse, they are not exactly the same. 938 * I believe 5.5.4 is the best one, so we follow 5.5.4. 939 */ 940 if (isipv6 && !V_ip6_use_deprecated) { 941 struct in6_ifaddr *ia6; 942 943 if ((ia6 = ip6_getdstifaddr(m)) && 944 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 945 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 946 log(LOG_DEBUG, "%s; %s: Listen socket: " 947 "Connection attempt to deprecated " 948 "IPv6 address rejected\n", 949 s, __func__); 950 rstreason = BANDLIM_RST_OPENPORT; 951 goto dropwithreset; 952 } 953 } 954#endif 955 /* 956 * Basic sanity checks on incoming SYN requests: 957 * Don't respond if the destination is a link layer 958 * broadcast according to RFC1122 4.2.3.10, p. 104. 959 * If it is from this socket it must be forged. 960 * Don't respond if the source or destination is a 961 * global or subnet broad- or multicast address. 962 * Note that it is quite possible to receive unicast 963 * link-layer packets with a broadcast IP address. Use 964 * in_broadcast() to find them. 965 */ 966 if (m->m_flags & (M_BCAST|M_MCAST)) { 967 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 968 log(LOG_DEBUG, "%s; %s: Listen socket: " 969 "Connection attempt from broad- or multicast " 970 "link layer address ignored\n", s, __func__); 971 goto dropunlock; 972 } 973 if (isipv6) { 974#ifdef INET6 975 if (th->th_dport == th->th_sport && 976 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 977 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 978 log(LOG_DEBUG, "%s; %s: Listen socket: " 979 "Connection attempt to/from self " 980 "ignored\n", s, __func__); 981 goto dropunlock; 982 } 983 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 984 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 985 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 986 log(LOG_DEBUG, "%s; %s: Listen socket: " 987 "Connection attempt from/to multicast " 988 "address ignored\n", s, __func__); 989 goto dropunlock; 990 } 991#endif 992 } else { 993 if (th->th_dport == th->th_sport && 994 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 995 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 996 log(LOG_DEBUG, "%s; %s: Listen socket: " 997 "Connection attempt from/to self " 998 "ignored\n", s, __func__); 999 goto dropunlock; 1000 } 1001 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1002 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1003 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1004 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1005 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1006 log(LOG_DEBUG, "%s; %s: Listen socket: " 1007 "Connection attempt from/to broad- " 1008 "or multicast address ignored\n", 1009 s, __func__); 1010 goto dropunlock; 1011 } 1012 } 1013 /* 1014 * SYN appears to be valid. Create compressed TCP state 1015 * for syncache. 1016 */ 1017#ifdef TCPDEBUG 1018 if (so->so_options & SO_DEBUG) 1019 tcp_trace(TA_INPUT, ostate, tp, 1020 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1021#endif 1022 tcp_dooptions(&to, optp, optlen, TO_SYN); 1023 syncache_add(&inc, &to, th, inp, &so, m); 1024 /* 1025 * Entry added to syncache and mbuf consumed. 1026 * Everything already unlocked by syncache_add(). 1027 */ 1028 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1029 return; 1030 } 1031 1032 /* 1033 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1034 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1035 * the inpcb, and unlocks pcbinfo. 1036 */ 1037 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1038 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1039 return; 1040 1041dropwithreset: 1042 if (ti_locked == TI_RLOCKED) 1043 INP_INFO_RUNLOCK(&V_tcbinfo); 1044 else if (ti_locked == TI_WLOCKED) 1045 INP_INFO_WUNLOCK(&V_tcbinfo); 1046 else 1047 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked); 1048 ti_locked = TI_UNLOCKED; 1049 1050 if (inp != NULL) { 1051 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1052 INP_WUNLOCK(inp); 1053 } else 1054 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1055 m = NULL; /* mbuf chain got consumed. */ 1056 goto drop; 1057 1058dropunlock: 1059 if (ti_locked == TI_RLOCKED) 1060 INP_INFO_RUNLOCK(&V_tcbinfo); 1061 else if (ti_locked == TI_WLOCKED) 1062 INP_INFO_WUNLOCK(&V_tcbinfo); 1063 else 1064 panic("%s: dropunlock ti_locked %d", __func__, ti_locked); 1065 ti_locked = TI_UNLOCKED; 1066 1067 if (inp != NULL) 1068 INP_WUNLOCK(inp); 1069 1070drop: 1071 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1072 if (s != NULL) 1073 free(s, M_TCPLOG); 1074 if (m != NULL) 1075 m_freem(m); 1076} 1077 1078static void 1079tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1080 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1081 int ti_locked) 1082{ 1083 INIT_VNET_INET(tp->t_vnet); 1084 int thflags, acked, ourfinisacked, needoutput = 0; 1085 int rstreason, todrop, win; 1086 u_long tiwin; 1087 struct tcpopt to; 1088 1089#ifdef TCPDEBUG 1090 /* 1091 * The size of tcp_saveipgen must be the size of the max ip header, 1092 * now IPv6. 1093 */ 1094 u_char tcp_saveipgen[IP6_HDR_LEN]; 1095 struct tcphdr tcp_savetcp; 1096 short ostate = 0; 1097#endif 1098 thflags = th->th_flags; 1099 1100 /* 1101 * If this is either a state-changing packet or current state isn't 1102 * established, we require a write lock on tcbinfo. Otherwise, we 1103 * allow either a read lock or a write lock, as we may have acquired 1104 * a write lock due to a race. 1105 * 1106 * Require a global write lock for SYN/FIN/RST segments or 1107 * non-established connections; otherwise accept either a read or 1108 * write lock, as we may have conservatively acquired a write lock in 1109 * certain cases in tcp_input() (is this still true?). Currently we 1110 * will never enter with no lock, so we try to drop it quickly in the 1111 * common pure ack/pure data cases. 1112 */ 1113 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1114 tp->t_state != TCPS_ESTABLISHED) { 1115 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1116 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1117 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1118 } else { 1119#ifdef INVARIANTS 1120 if (ti_locked == TI_RLOCKED) 1121 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1122 else if (ti_locked == TI_WLOCKED) 1123 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1124 else 1125 panic("%s: ti_locked %d for EST", __func__, 1126 ti_locked); 1127#endif 1128 } 1129 INP_WLOCK_ASSERT(tp->t_inpcb); 1130 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1131 __func__)); 1132 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1133 __func__)); 1134 1135 /* 1136 * Segment received on connection. 1137 * Reset idle time and keep-alive timer. 1138 * XXX: This should be done after segment 1139 * validation to ignore broken/spoofed segs. 1140 */ 1141 tp->t_rcvtime = ticks; 1142 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1143 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1144 1145 /* 1146 * Unscale the window into a 32-bit value. 1147 * For the SYN_SENT state the scale is zero. 1148 */ 1149 tiwin = th->th_win << tp->snd_scale; 1150 1151 /* 1152 * TCP ECN processing. 1153 */ 1154 if (tp->t_flags & TF_ECN_PERMIT) { 1155 switch (iptos & IPTOS_ECN_MASK) { 1156 case IPTOS_ECN_CE: 1157 tp->t_flags |= TF_ECN_SND_ECE; 1158 TCPSTAT_INC(tcps_ecn_ce); 1159 break; 1160 case IPTOS_ECN_ECT0: 1161 TCPSTAT_INC(tcps_ecn_ect0); 1162 break; 1163 case IPTOS_ECN_ECT1: 1164 TCPSTAT_INC(tcps_ecn_ect1); 1165 break; 1166 } 1167 1168 if (thflags & TH_CWR) 1169 tp->t_flags &= ~TF_ECN_SND_ECE; 1170 1171 /* 1172 * Congestion experienced. 1173 * Ignore if we are already trying to recover. 1174 */ 1175 if ((thflags & TH_ECE) && 1176 SEQ_LEQ(th->th_ack, tp->snd_recover)) { 1177 TCPSTAT_INC(tcps_ecn_rcwnd); 1178 tcp_congestion_exp(tp); 1179 } 1180 } 1181 1182 /* 1183 * Parse options on any incoming segment. 1184 */ 1185 tcp_dooptions(&to, (u_char *)(th + 1), 1186 (th->th_off << 2) - sizeof(struct tcphdr), 1187 (thflags & TH_SYN) ? TO_SYN : 0); 1188 1189 /* 1190 * If echoed timestamp is later than the current time, 1191 * fall back to non RFC1323 RTT calculation. Normalize 1192 * timestamp if syncookies were used when this connection 1193 * was established. 1194 */ 1195 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1196 to.to_tsecr -= tp->ts_offset; 1197 if (TSTMP_GT(to.to_tsecr, ticks)) 1198 to.to_tsecr = 0; 1199 } 1200 1201 /* 1202 * Process options only when we get SYN/ACK back. The SYN case 1203 * for incoming connections is handled in tcp_syncache. 1204 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1205 * or <SYN,ACK>) segment itself is never scaled. 1206 * XXX this is traditional behavior, may need to be cleaned up. 1207 */ 1208 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1209 if ((to.to_flags & TOF_SCALE) && 1210 (tp->t_flags & TF_REQ_SCALE)) { 1211 tp->t_flags |= TF_RCVD_SCALE; 1212 tp->snd_scale = to.to_wscale; 1213 } 1214 /* 1215 * Initial send window. It will be updated with 1216 * the next incoming segment to the scaled value. 1217 */ 1218 tp->snd_wnd = th->th_win; 1219 if (to.to_flags & TOF_TS) { 1220 tp->t_flags |= TF_RCVD_TSTMP; 1221 tp->ts_recent = to.to_tsval; 1222 tp->ts_recent_age = ticks; 1223 } 1224 if (to.to_flags & TOF_MSS) 1225 tcp_mss(tp, to.to_mss); 1226 if ((tp->t_flags & TF_SACK_PERMIT) && 1227 (to.to_flags & TOF_SACKPERM) == 0) 1228 tp->t_flags &= ~TF_SACK_PERMIT; 1229 } 1230 1231 /* 1232 * Header prediction: check for the two common cases 1233 * of a uni-directional data xfer. If the packet has 1234 * no control flags, is in-sequence, the window didn't 1235 * change and we're not retransmitting, it's a 1236 * candidate. If the length is zero and the ack moved 1237 * forward, we're the sender side of the xfer. Just 1238 * free the data acked & wake any higher level process 1239 * that was blocked waiting for space. If the length 1240 * is non-zero and the ack didn't move, we're the 1241 * receiver side. If we're getting packets in-order 1242 * (the reassembly queue is empty), add the data to 1243 * the socket buffer and note that we need a delayed ack. 1244 * Make sure that the hidden state-flags are also off. 1245 * Since we check for TCPS_ESTABLISHED first, it can only 1246 * be TH_NEEDSYN. 1247 */ 1248 if (tp->t_state == TCPS_ESTABLISHED && 1249 th->th_seq == tp->rcv_nxt && 1250 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1251 tp->snd_nxt == tp->snd_max && 1252 tiwin && tiwin == tp->snd_wnd && 1253 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1254 LIST_EMPTY(&tp->t_segq) && 1255 ((to.to_flags & TOF_TS) == 0 || 1256 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1257 1258 /* 1259 * If last ACK falls within this segment's sequence numbers, 1260 * record the timestamp. 1261 * NOTE that the test is modified according to the latest 1262 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1263 */ 1264 if ((to.to_flags & TOF_TS) != 0 && 1265 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1266 tp->ts_recent_age = ticks; 1267 tp->ts_recent = to.to_tsval; 1268 } 1269 1270 if (tlen == 0) { 1271 if (SEQ_GT(th->th_ack, tp->snd_una) && 1272 SEQ_LEQ(th->th_ack, tp->snd_max) && 1273 tp->snd_cwnd >= tp->snd_wnd && 1274 ((!V_tcp_do_newreno && 1275 !(tp->t_flags & TF_SACK_PERMIT) && 1276 tp->t_dupacks < tcprexmtthresh) || 1277 ((V_tcp_do_newreno || 1278 (tp->t_flags & TF_SACK_PERMIT)) && 1279 !IN_FASTRECOVERY(tp) && 1280 (to.to_flags & TOF_SACK) == 0 && 1281 TAILQ_EMPTY(&tp->snd_holes)))) { 1282 /* 1283 * This is a pure ack for outstanding data. 1284 */ 1285 if (ti_locked == TI_RLOCKED) 1286 INP_INFO_RUNLOCK(&V_tcbinfo); 1287 else if (ti_locked == TI_WLOCKED) 1288 INP_INFO_WUNLOCK(&V_tcbinfo); 1289 else 1290 panic("%s: ti_locked %d on pure ACK", 1291 __func__, ti_locked); 1292 ti_locked = TI_UNLOCKED; 1293 1294 TCPSTAT_INC(tcps_predack); 1295 1296 /* 1297 * "bad retransmit" recovery. 1298 */ 1299 if (tp->t_rxtshift == 1 && 1300 ticks < tp->t_badrxtwin) { 1301 TCPSTAT_INC(tcps_sndrexmitbad); 1302 tp->snd_cwnd = tp->snd_cwnd_prev; 1303 tp->snd_ssthresh = 1304 tp->snd_ssthresh_prev; 1305 tp->snd_recover = tp->snd_recover_prev; 1306 if (tp->t_flags & TF_WASFRECOVERY) 1307 ENTER_FASTRECOVERY(tp); 1308 tp->snd_nxt = tp->snd_max; 1309 tp->t_badrxtwin = 0; 1310 } 1311 1312 /* 1313 * Recalculate the transmit timer / rtt. 1314 * 1315 * Some boxes send broken timestamp replies 1316 * during the SYN+ACK phase, ignore 1317 * timestamps of 0 or we could calculate a 1318 * huge RTT and blow up the retransmit timer. 1319 */ 1320 if ((to.to_flags & TOF_TS) != 0 && 1321 to.to_tsecr) { 1322 if (!tp->t_rttlow || 1323 tp->t_rttlow > ticks - to.to_tsecr) 1324 tp->t_rttlow = ticks - to.to_tsecr; 1325 tcp_xmit_timer(tp, 1326 ticks - to.to_tsecr + 1); 1327 } else if (tp->t_rtttime && 1328 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1329 if (!tp->t_rttlow || 1330 tp->t_rttlow > ticks - tp->t_rtttime) 1331 tp->t_rttlow = ticks - tp->t_rtttime; 1332 tcp_xmit_timer(tp, 1333 ticks - tp->t_rtttime); 1334 } 1335 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1336 acked = th->th_ack - tp->snd_una; 1337 TCPSTAT_INC(tcps_rcvackpack); 1338 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1339 sbdrop(&so->so_snd, acked); 1340 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1341 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1342 tp->snd_recover = th->th_ack - 1; 1343 tp->snd_una = th->th_ack; 1344 /* 1345 * Pull snd_wl2 up to prevent seq wrap relative 1346 * to th_ack. 1347 */ 1348 tp->snd_wl2 = th->th_ack; 1349 tp->t_dupacks = 0; 1350 m_freem(m); 1351 ND6_HINT(tp); /* Some progress has been made. */ 1352 1353 /* 1354 * If all outstanding data are acked, stop 1355 * retransmit timer, otherwise restart timer 1356 * using current (possibly backed-off) value. 1357 * If process is waiting for space, 1358 * wakeup/selwakeup/signal. If data 1359 * are ready to send, let tcp_output 1360 * decide between more output or persist. 1361 */ 1362#ifdef TCPDEBUG 1363 if (so->so_options & SO_DEBUG) 1364 tcp_trace(TA_INPUT, ostate, tp, 1365 (void *)tcp_saveipgen, 1366 &tcp_savetcp, 0); 1367#endif 1368 if (tp->snd_una == tp->snd_max) 1369 tcp_timer_activate(tp, TT_REXMT, 0); 1370 else if (!tcp_timer_active(tp, TT_PERSIST)) 1371 tcp_timer_activate(tp, TT_REXMT, 1372 tp->t_rxtcur); 1373 sowwakeup(so); 1374 if (so->so_snd.sb_cc) 1375 (void) tcp_output(tp); 1376 goto check_delack; 1377 } 1378 } else if (th->th_ack == tp->snd_una && 1379 tlen <= sbspace(&so->so_rcv)) { 1380 int newsize = 0; /* automatic sockbuf scaling */ 1381 1382 /* 1383 * This is a pure, in-sequence data packet with 1384 * nothing on the reassembly queue and we have enough 1385 * buffer space to take it. 1386 */ 1387 if (ti_locked == TI_RLOCKED) 1388 INP_INFO_RUNLOCK(&V_tcbinfo); 1389 else if (ti_locked == TI_WLOCKED) 1390 INP_INFO_WUNLOCK(&V_tcbinfo); 1391 else 1392 panic("%s: ti_locked %d on pure data " 1393 "segment", __func__, ti_locked); 1394 ti_locked = TI_UNLOCKED; 1395 1396 /* Clean receiver SACK report if present */ 1397 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1398 tcp_clean_sackreport(tp); 1399 TCPSTAT_INC(tcps_preddat); 1400 tp->rcv_nxt += tlen; 1401 /* 1402 * Pull snd_wl1 up to prevent seq wrap relative to 1403 * th_seq. 1404 */ 1405 tp->snd_wl1 = th->th_seq; 1406 /* 1407 * Pull rcv_up up to prevent seq wrap relative to 1408 * rcv_nxt. 1409 */ 1410 tp->rcv_up = tp->rcv_nxt; 1411 TCPSTAT_INC(tcps_rcvpack); 1412 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1413 ND6_HINT(tp); /* Some progress has been made */ 1414#ifdef TCPDEBUG 1415 if (so->so_options & SO_DEBUG) 1416 tcp_trace(TA_INPUT, ostate, tp, 1417 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1418#endif 1419 /* 1420 * Automatic sizing of receive socket buffer. Often the send 1421 * buffer size is not optimally adjusted to the actual network 1422 * conditions at hand (delay bandwidth product). Setting the 1423 * buffer size too small limits throughput on links with high 1424 * bandwidth and high delay (eg. trans-continental/oceanic links). 1425 * 1426 * On the receive side the socket buffer memory is only rarely 1427 * used to any significant extent. This allows us to be much 1428 * more aggressive in scaling the receive socket buffer. For 1429 * the case that the buffer space is actually used to a large 1430 * extent and we run out of kernel memory we can simply drop 1431 * the new segments; TCP on the sender will just retransmit it 1432 * later. Setting the buffer size too big may only consume too 1433 * much kernel memory if the application doesn't read() from 1434 * the socket or packet loss or reordering makes use of the 1435 * reassembly queue. 1436 * 1437 * The criteria to step up the receive buffer one notch are: 1438 * 1. the number of bytes received during the time it takes 1439 * one timestamp to be reflected back to us (the RTT); 1440 * 2. received bytes per RTT is within seven eighth of the 1441 * current socket buffer size; 1442 * 3. receive buffer size has not hit maximal automatic size; 1443 * 1444 * This algorithm does one step per RTT at most and only if 1445 * we receive a bulk stream w/o packet losses or reorderings. 1446 * Shrinking the buffer during idle times is not necessary as 1447 * it doesn't consume any memory when idle. 1448 * 1449 * TODO: Only step up if the application is actually serving 1450 * the buffer to better manage the socket buffer resources. 1451 */ 1452 if (V_tcp_do_autorcvbuf && 1453 to.to_tsecr && 1454 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1455 if (to.to_tsecr > tp->rfbuf_ts && 1456 to.to_tsecr - tp->rfbuf_ts < hz) { 1457 if (tp->rfbuf_cnt > 1458 (so->so_rcv.sb_hiwat / 8 * 7) && 1459 so->so_rcv.sb_hiwat < 1460 V_tcp_autorcvbuf_max) { 1461 newsize = 1462 min(so->so_rcv.sb_hiwat + 1463 V_tcp_autorcvbuf_inc, 1464 V_tcp_autorcvbuf_max); 1465 } 1466 /* Start over with next RTT. */ 1467 tp->rfbuf_ts = 0; 1468 tp->rfbuf_cnt = 0; 1469 } else 1470 tp->rfbuf_cnt += tlen; /* add up */ 1471 } 1472 1473 /* Add data to socket buffer. */ 1474 SOCKBUF_LOCK(&so->so_rcv); 1475 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1476 m_freem(m); 1477 } else { 1478 /* 1479 * Set new socket buffer size. 1480 * Give up when limit is reached. 1481 */ 1482 if (newsize) 1483 if (!sbreserve_locked(&so->so_rcv, 1484 newsize, so, NULL)) 1485 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1486 m_adj(m, drop_hdrlen); /* delayed header drop */ 1487 sbappendstream_locked(&so->so_rcv, m); 1488 } 1489 /* NB: sorwakeup_locked() does an implicit unlock. */ 1490 sorwakeup_locked(so); 1491 if (DELAY_ACK(tp)) { 1492 tp->t_flags |= TF_DELACK; 1493 } else { 1494 tp->t_flags |= TF_ACKNOW; 1495 tcp_output(tp); 1496 } 1497 goto check_delack; 1498 } 1499 } 1500 1501 /* 1502 * Calculate amount of space in receive window, 1503 * and then do TCP input processing. 1504 * Receive window is amount of space in rcv queue, 1505 * but not less than advertised window. 1506 */ 1507 win = sbspace(&so->so_rcv); 1508 if (win < 0) 1509 win = 0; 1510 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1511 1512 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1513 tp->rfbuf_ts = 0; 1514 tp->rfbuf_cnt = 0; 1515 1516 switch (tp->t_state) { 1517 1518 /* 1519 * If the state is SYN_RECEIVED: 1520 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1521 */ 1522 case TCPS_SYN_RECEIVED: 1523 if ((thflags & TH_ACK) && 1524 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1525 SEQ_GT(th->th_ack, tp->snd_max))) { 1526 rstreason = BANDLIM_RST_OPENPORT; 1527 goto dropwithreset; 1528 } 1529 break; 1530 1531 /* 1532 * If the state is SYN_SENT: 1533 * if seg contains an ACK, but not for our SYN, drop the input. 1534 * if seg contains a RST, then drop the connection. 1535 * if seg does not contain SYN, then drop it. 1536 * Otherwise this is an acceptable SYN segment 1537 * initialize tp->rcv_nxt and tp->irs 1538 * if seg contains ack then advance tp->snd_una 1539 * if seg contains an ECE and ECN support is enabled, the stream 1540 * is ECN capable. 1541 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1542 * arrange for segment to be acked (eventually) 1543 * continue processing rest of data/controls, beginning with URG 1544 */ 1545 case TCPS_SYN_SENT: 1546 if ((thflags & TH_ACK) && 1547 (SEQ_LEQ(th->th_ack, tp->iss) || 1548 SEQ_GT(th->th_ack, tp->snd_max))) { 1549 rstreason = BANDLIM_UNLIMITED; 1550 goto dropwithreset; 1551 } 1552 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1553 tp = tcp_drop(tp, ECONNREFUSED); 1554 if (thflags & TH_RST) 1555 goto drop; 1556 if (!(thflags & TH_SYN)) 1557 goto drop; 1558 1559 tp->irs = th->th_seq; 1560 tcp_rcvseqinit(tp); 1561 if (thflags & TH_ACK) { 1562 TCPSTAT_INC(tcps_connects); 1563 soisconnected(so); 1564#ifdef MAC 1565 SOCK_LOCK(so); 1566 mac_socketpeer_set_from_mbuf(m, so); 1567 SOCK_UNLOCK(so); 1568#endif 1569 /* Do window scaling on this connection? */ 1570 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1571 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1572 tp->rcv_scale = tp->request_r_scale; 1573 } 1574 tp->rcv_adv += tp->rcv_wnd; 1575 tp->snd_una++; /* SYN is acked */ 1576 /* 1577 * If there's data, delay ACK; if there's also a FIN 1578 * ACKNOW will be turned on later. 1579 */ 1580 if (DELAY_ACK(tp) && tlen != 0) 1581 tcp_timer_activate(tp, TT_DELACK, 1582 tcp_delacktime); 1583 else 1584 tp->t_flags |= TF_ACKNOW; 1585 1586 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1587 tp->t_flags |= TF_ECN_PERMIT; 1588 TCPSTAT_INC(tcps_ecn_shs); 1589 } 1590 1591 /* 1592 * Received <SYN,ACK> in SYN_SENT[*] state. 1593 * Transitions: 1594 * SYN_SENT --> ESTABLISHED 1595 * SYN_SENT* --> FIN_WAIT_1 1596 */ 1597 tp->t_starttime = ticks; 1598 if (tp->t_flags & TF_NEEDFIN) { 1599 tp->t_state = TCPS_FIN_WAIT_1; 1600 tp->t_flags &= ~TF_NEEDFIN; 1601 thflags &= ~TH_SYN; 1602 } else { 1603 tp->t_state = TCPS_ESTABLISHED; 1604 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1605 } 1606 } else { 1607 /* 1608 * Received initial SYN in SYN-SENT[*] state => 1609 * simultaneous open. If segment contains CC option 1610 * and there is a cached CC, apply TAO test. 1611 * If it succeeds, connection is * half-synchronized. 1612 * Otherwise, do 3-way handshake: 1613 * SYN-SENT -> SYN-RECEIVED 1614 * SYN-SENT* -> SYN-RECEIVED* 1615 * If there was no CC option, clear cached CC value. 1616 */ 1617 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1618 tcp_timer_activate(tp, TT_REXMT, 0); 1619 tp->t_state = TCPS_SYN_RECEIVED; 1620 } 1621 1622 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1623 "ti_locked %d", __func__, ti_locked)); 1624 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1625 INP_WLOCK_ASSERT(tp->t_inpcb); 1626 1627 /* 1628 * Advance th->th_seq to correspond to first data byte. 1629 * If data, trim to stay within window, 1630 * dropping FIN if necessary. 1631 */ 1632 th->th_seq++; 1633 if (tlen > tp->rcv_wnd) { 1634 todrop = tlen - tp->rcv_wnd; 1635 m_adj(m, -todrop); 1636 tlen = tp->rcv_wnd; 1637 thflags &= ~TH_FIN; 1638 TCPSTAT_INC(tcps_rcvpackafterwin); 1639 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1640 } 1641 tp->snd_wl1 = th->th_seq - 1; 1642 tp->rcv_up = th->th_seq; 1643 /* 1644 * Client side of transaction: already sent SYN and data. 1645 * If the remote host used T/TCP to validate the SYN, 1646 * our data will be ACK'd; if so, enter normal data segment 1647 * processing in the middle of step 5, ack processing. 1648 * Otherwise, goto step 6. 1649 */ 1650 if (thflags & TH_ACK) 1651 goto process_ACK; 1652 1653 goto step6; 1654 1655 /* 1656 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1657 * do normal processing. 1658 * 1659 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1660 */ 1661 case TCPS_LAST_ACK: 1662 case TCPS_CLOSING: 1663 break; /* continue normal processing */ 1664 } 1665 1666 /* 1667 * States other than LISTEN or SYN_SENT. 1668 * First check the RST flag and sequence number since reset segments 1669 * are exempt from the timestamp and connection count tests. This 1670 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1671 * below which allowed reset segments in half the sequence space 1672 * to fall though and be processed (which gives forged reset 1673 * segments with a random sequence number a 50 percent chance of 1674 * killing a connection). 1675 * Then check timestamp, if present. 1676 * Then check the connection count, if present. 1677 * Then check that at least some bytes of segment are within 1678 * receive window. If segment begins before rcv_nxt, 1679 * drop leading data (and SYN); if nothing left, just ack. 1680 * 1681 * 1682 * If the RST bit is set, check the sequence number to see 1683 * if this is a valid reset segment. 1684 * RFC 793 page 37: 1685 * In all states except SYN-SENT, all reset (RST) segments 1686 * are validated by checking their SEQ-fields. A reset is 1687 * valid if its sequence number is in the window. 1688 * Note: this does not take into account delayed ACKs, so 1689 * we should test against last_ack_sent instead of rcv_nxt. 1690 * The sequence number in the reset segment is normally an 1691 * echo of our outgoing acknowlegement numbers, but some hosts 1692 * send a reset with the sequence number at the rightmost edge 1693 * of our receive window, and we have to handle this case. 1694 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1695 * that brute force RST attacks are possible. To combat this, 1696 * we use a much stricter check while in the ESTABLISHED state, 1697 * only accepting RSTs where the sequence number is equal to 1698 * last_ack_sent. In all other states (the states in which a 1699 * RST is more likely), the more permissive check is used. 1700 * If we have multiple segments in flight, the initial reset 1701 * segment sequence numbers will be to the left of last_ack_sent, 1702 * but they will eventually catch up. 1703 * In any case, it never made sense to trim reset segments to 1704 * fit the receive window since RFC 1122 says: 1705 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1706 * 1707 * A TCP SHOULD allow a received RST segment to include data. 1708 * 1709 * DISCUSSION 1710 * It has been suggested that a RST segment could contain 1711 * ASCII text that encoded and explained the cause of the 1712 * RST. No standard has yet been established for such 1713 * data. 1714 * 1715 * If the reset segment passes the sequence number test examine 1716 * the state: 1717 * SYN_RECEIVED STATE: 1718 * If passive open, return to LISTEN state. 1719 * If active open, inform user that connection was refused. 1720 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1721 * Inform user that connection was reset, and close tcb. 1722 * CLOSING, LAST_ACK STATES: 1723 * Close the tcb. 1724 * TIME_WAIT STATE: 1725 * Drop the segment - see Stevens, vol. 2, p. 964 and 1726 * RFC 1337. 1727 */ 1728 if (thflags & TH_RST) { 1729 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1730 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1731 switch (tp->t_state) { 1732 1733 case TCPS_SYN_RECEIVED: 1734 so->so_error = ECONNREFUSED; 1735 goto close; 1736 1737 case TCPS_ESTABLISHED: 1738 if (V_tcp_insecure_rst == 0 && 1739 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 1740 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 1741 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1742 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 1743 TCPSTAT_INC(tcps_badrst); 1744 goto drop; 1745 } 1746 /* FALLTHROUGH */ 1747 case TCPS_FIN_WAIT_1: 1748 case TCPS_FIN_WAIT_2: 1749 case TCPS_CLOSE_WAIT: 1750 so->so_error = ECONNRESET; 1751 close: 1752 KASSERT(ti_locked == TI_WLOCKED, 1753 ("tcp_do_segment: TH_RST 1 ti_locked %d", 1754 ti_locked)); 1755 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1756 1757 tp->t_state = TCPS_CLOSED; 1758 TCPSTAT_INC(tcps_drops); 1759 tp = tcp_close(tp); 1760 break; 1761 1762 case TCPS_CLOSING: 1763 case TCPS_LAST_ACK: 1764 KASSERT(ti_locked == TI_WLOCKED, 1765 ("tcp_do_segment: TH_RST 2 ti_locked %d", 1766 ti_locked)); 1767 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1768 1769 tp = tcp_close(tp); 1770 break; 1771 } 1772 } 1773 goto drop; 1774 } 1775 1776 /* 1777 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1778 * and it's less than ts_recent, drop it. 1779 */ 1780 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1781 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1782 1783 /* Check to see if ts_recent is over 24 days old. */ 1784 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1785 /* 1786 * Invalidate ts_recent. If this segment updates 1787 * ts_recent, the age will be reset later and ts_recent 1788 * will get a valid value. If it does not, setting 1789 * ts_recent to zero will at least satisfy the 1790 * requirement that zero be placed in the timestamp 1791 * echo reply when ts_recent isn't valid. The 1792 * age isn't reset until we get a valid ts_recent 1793 * because we don't want out-of-order segments to be 1794 * dropped when ts_recent is old. 1795 */ 1796 tp->ts_recent = 0; 1797 } else { 1798 TCPSTAT_INC(tcps_rcvduppack); 1799 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 1800 TCPSTAT_INC(tcps_pawsdrop); 1801 if (tlen) 1802 goto dropafterack; 1803 goto drop; 1804 } 1805 } 1806 1807 /* 1808 * In the SYN-RECEIVED state, validate that the packet belongs to 1809 * this connection before trimming the data to fit the receive 1810 * window. Check the sequence number versus IRS since we know 1811 * the sequence numbers haven't wrapped. This is a partial fix 1812 * for the "LAND" DoS attack. 1813 */ 1814 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1815 rstreason = BANDLIM_RST_OPENPORT; 1816 goto dropwithreset; 1817 } 1818 1819 todrop = tp->rcv_nxt - th->th_seq; 1820 if (todrop > 0) { 1821 /* 1822 * If this is a duplicate SYN for our current connection, 1823 * advance over it and pretend and it's not a SYN. 1824 */ 1825 if (thflags & TH_SYN && th->th_seq == tp->irs) { 1826 thflags &= ~TH_SYN; 1827 th->th_seq++; 1828 if (th->th_urp > 1) 1829 th->th_urp--; 1830 else 1831 thflags &= ~TH_URG; 1832 todrop--; 1833 } 1834 /* 1835 * Following if statement from Stevens, vol. 2, p. 960. 1836 */ 1837 if (todrop > tlen 1838 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1839 /* 1840 * Any valid FIN must be to the left of the window. 1841 * At this point the FIN must be a duplicate or out 1842 * of sequence; drop it. 1843 */ 1844 thflags &= ~TH_FIN; 1845 1846 /* 1847 * Send an ACK to resynchronize and drop any data. 1848 * But keep on processing for RST or ACK. 1849 */ 1850 tp->t_flags |= TF_ACKNOW; 1851 todrop = tlen; 1852 TCPSTAT_INC(tcps_rcvduppack); 1853 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 1854 } else { 1855 TCPSTAT_INC(tcps_rcvpartduppack); 1856 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 1857 } 1858 drop_hdrlen += todrop; /* drop from the top afterwards */ 1859 th->th_seq += todrop; 1860 tlen -= todrop; 1861 if (th->th_urp > todrop) 1862 th->th_urp -= todrop; 1863 else { 1864 thflags &= ~TH_URG; 1865 th->th_urp = 0; 1866 } 1867 } 1868 1869 /* 1870 * If new data are received on a connection after the 1871 * user processes are gone, then RST the other end. 1872 */ 1873 if ((so->so_state & SS_NOFDREF) && 1874 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1875 char *s; 1876 1877 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 1878 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 1879 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1880 1881 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 1882 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 1883 "was closed, sending RST and removing tcpcb\n", 1884 s, __func__, tcpstates[tp->t_state], tlen); 1885 free(s, M_TCPLOG); 1886 } 1887 tp = tcp_close(tp); 1888 TCPSTAT_INC(tcps_rcvafterclose); 1889 rstreason = BANDLIM_UNLIMITED; 1890 goto dropwithreset; 1891 } 1892 1893 /* 1894 * If segment ends after window, drop trailing data 1895 * (and PUSH and FIN); if nothing left, just ACK. 1896 */ 1897 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1898 if (todrop > 0) { 1899 TCPSTAT_INC(tcps_rcvpackafterwin); 1900 if (todrop >= tlen) { 1901 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 1902 /* 1903 * If window is closed can only take segments at 1904 * window edge, and have to drop data and PUSH from 1905 * incoming segments. Continue processing, but 1906 * remember to ack. Otherwise, drop segment 1907 * and ack. 1908 */ 1909 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1910 tp->t_flags |= TF_ACKNOW; 1911 TCPSTAT_INC(tcps_rcvwinprobe); 1912 } else 1913 goto dropafterack; 1914 } else 1915 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1916 m_adj(m, -todrop); 1917 tlen -= todrop; 1918 thflags &= ~(TH_PUSH|TH_FIN); 1919 } 1920 1921 /* 1922 * If last ACK falls within this segment's sequence numbers, 1923 * record its timestamp. 1924 * NOTE: 1925 * 1) That the test incorporates suggestions from the latest 1926 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1927 * 2) That updating only on newer timestamps interferes with 1928 * our earlier PAWS tests, so this check should be solely 1929 * predicated on the sequence space of this segment. 1930 * 3) That we modify the segment boundary check to be 1931 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1932 * instead of RFC1323's 1933 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1934 * This modified check allows us to overcome RFC1323's 1935 * limitations as described in Stevens TCP/IP Illustrated 1936 * Vol. 2 p.869. In such cases, we can still calculate the 1937 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1938 */ 1939 if ((to.to_flags & TOF_TS) != 0 && 1940 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1941 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1942 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1943 tp->ts_recent_age = ticks; 1944 tp->ts_recent = to.to_tsval; 1945 } 1946 1947 /* 1948 * If a SYN is in the window, then this is an 1949 * error and we send an RST and drop the connection. 1950 */ 1951 if (thflags & TH_SYN) { 1952 KASSERT(ti_locked == TI_WLOCKED, 1953 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 1954 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1955 1956 tp = tcp_drop(tp, ECONNRESET); 1957 rstreason = BANDLIM_UNLIMITED; 1958 goto drop; 1959 } 1960 1961 /* 1962 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1963 * flag is on (half-synchronized state), then queue data for 1964 * later processing; else drop segment and return. 1965 */ 1966 if ((thflags & TH_ACK) == 0) { 1967 if (tp->t_state == TCPS_SYN_RECEIVED || 1968 (tp->t_flags & TF_NEEDSYN)) 1969 goto step6; 1970 else if (tp->t_flags & TF_ACKNOW) 1971 goto dropafterack; 1972 else 1973 goto drop; 1974 } 1975 1976 /* 1977 * Ack processing. 1978 */ 1979 switch (tp->t_state) { 1980 1981 /* 1982 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1983 * ESTABLISHED state and continue processing. 1984 * The ACK was checked above. 1985 */ 1986 case TCPS_SYN_RECEIVED: 1987 1988 TCPSTAT_INC(tcps_connects); 1989 soisconnected(so); 1990 /* Do window scaling? */ 1991 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1992 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1993 tp->rcv_scale = tp->request_r_scale; 1994 tp->snd_wnd = tiwin; 1995 } 1996 /* 1997 * Make transitions: 1998 * SYN-RECEIVED -> ESTABLISHED 1999 * SYN-RECEIVED* -> FIN-WAIT-1 2000 */ 2001 tp->t_starttime = ticks; 2002 if (tp->t_flags & TF_NEEDFIN) { 2003 tp->t_state = TCPS_FIN_WAIT_1; 2004 tp->t_flags &= ~TF_NEEDFIN; 2005 } else { 2006 tp->t_state = TCPS_ESTABLISHED; 2007 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 2008 } 2009 /* 2010 * If segment contains data or ACK, will call tcp_reass() 2011 * later; if not, do so now to pass queued data to user. 2012 */ 2013 if (tlen == 0 && (thflags & TH_FIN) == 0) 2014 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2015 (struct mbuf *)0); 2016 tp->snd_wl1 = th->th_seq - 1; 2017 /* FALLTHROUGH */ 2018 2019 /* 2020 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2021 * ACKs. If the ack is in the range 2022 * tp->snd_una < th->th_ack <= tp->snd_max 2023 * then advance tp->snd_una to th->th_ack and drop 2024 * data from the retransmission queue. If this ACK reflects 2025 * more up to date window information we update our window information. 2026 */ 2027 case TCPS_ESTABLISHED: 2028 case TCPS_FIN_WAIT_1: 2029 case TCPS_FIN_WAIT_2: 2030 case TCPS_CLOSE_WAIT: 2031 case TCPS_CLOSING: 2032 case TCPS_LAST_ACK: 2033 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2034 TCPSTAT_INC(tcps_rcvacktoomuch); 2035 goto dropafterack; 2036 } 2037 if ((tp->t_flags & TF_SACK_PERMIT) && 2038 ((to.to_flags & TOF_SACK) || 2039 !TAILQ_EMPTY(&tp->snd_holes))) 2040 tcp_sack_doack(tp, &to, th->th_ack); 2041 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2042 if (tlen == 0 && tiwin == tp->snd_wnd) { 2043 TCPSTAT_INC(tcps_rcvdupack); 2044 /* 2045 * If we have outstanding data (other than 2046 * a window probe), this is a completely 2047 * duplicate ack (ie, window info didn't 2048 * change), the ack is the biggest we've 2049 * seen and we've seen exactly our rexmt 2050 * threshhold of them, assume a packet 2051 * has been dropped and retransmit it. 2052 * Kludge snd_nxt & the congestion 2053 * window so we send only this one 2054 * packet. 2055 * 2056 * We know we're losing at the current 2057 * window size so do congestion avoidance 2058 * (set ssthresh to half the current window 2059 * and pull our congestion window back to 2060 * the new ssthresh). 2061 * 2062 * Dup acks mean that packets have left the 2063 * network (they're now cached at the receiver) 2064 * so bump cwnd by the amount in the receiver 2065 * to keep a constant cwnd packets in the 2066 * network. 2067 * 2068 * When using TCP ECN, notify the peer that 2069 * we reduced the cwnd. 2070 */ 2071 if (!tcp_timer_active(tp, TT_REXMT) || 2072 th->th_ack != tp->snd_una) 2073 tp->t_dupacks = 0; 2074 else if (++tp->t_dupacks > tcprexmtthresh || 2075 ((V_tcp_do_newreno || 2076 (tp->t_flags & TF_SACK_PERMIT)) && 2077 IN_FASTRECOVERY(tp))) { 2078 if ((tp->t_flags & TF_SACK_PERMIT) && 2079 IN_FASTRECOVERY(tp)) { 2080 int awnd; 2081 2082 /* 2083 * Compute the amount of data in flight first. 2084 * We can inject new data into the pipe iff 2085 * we have less than 1/2 the original window's 2086 * worth of data in flight. 2087 */ 2088 awnd = (tp->snd_nxt - tp->snd_fack) + 2089 tp->sackhint.sack_bytes_rexmit; 2090 if (awnd < tp->snd_ssthresh) { 2091 tp->snd_cwnd += tp->t_maxseg; 2092 if (tp->snd_cwnd > tp->snd_ssthresh) 2093 tp->snd_cwnd = tp->snd_ssthresh; 2094 } 2095 } else 2096 tp->snd_cwnd += tp->t_maxseg; 2097 (void) tcp_output(tp); 2098 goto drop; 2099 } else if (tp->t_dupacks == tcprexmtthresh) { 2100 tcp_seq onxt = tp->snd_nxt; 2101 2102 /* 2103 * If we're doing sack, check to 2104 * see if we're already in sack 2105 * recovery. If we're not doing sack, 2106 * check to see if we're in newreno 2107 * recovery. 2108 */ 2109 if (tp->t_flags & TF_SACK_PERMIT) { 2110 if (IN_FASTRECOVERY(tp)) { 2111 tp->t_dupacks = 0; 2112 break; 2113 } 2114 } else if (V_tcp_do_newreno || 2115 V_tcp_do_ecn) { 2116 if (SEQ_LEQ(th->th_ack, 2117 tp->snd_recover)) { 2118 tp->t_dupacks = 0; 2119 break; 2120 } 2121 } 2122 tcp_congestion_exp(tp); 2123 tcp_timer_activate(tp, TT_REXMT, 0); 2124 tp->t_rtttime = 0; 2125 if (tp->t_flags & TF_SACK_PERMIT) { 2126 TCPSTAT_INC( 2127 tcps_sack_recovery_episode); 2128 tp->sack_newdata = tp->snd_nxt; 2129 tp->snd_cwnd = tp->t_maxseg; 2130 (void) tcp_output(tp); 2131 goto drop; 2132 } 2133 tp->snd_nxt = th->th_ack; 2134 tp->snd_cwnd = tp->t_maxseg; 2135 (void) tcp_output(tp); 2136 KASSERT(tp->snd_limited <= 2, 2137 ("%s: tp->snd_limited too big", 2138 __func__)); 2139 tp->snd_cwnd = tp->snd_ssthresh + 2140 tp->t_maxseg * 2141 (tp->t_dupacks - tp->snd_limited); 2142 if (SEQ_GT(onxt, tp->snd_nxt)) 2143 tp->snd_nxt = onxt; 2144 goto drop; 2145 } else if (V_tcp_do_rfc3042) { 2146 u_long oldcwnd = tp->snd_cwnd; 2147 tcp_seq oldsndmax = tp->snd_max; 2148 u_int sent; 2149 2150 KASSERT(tp->t_dupacks == 1 || 2151 tp->t_dupacks == 2, 2152 ("%s: dupacks not 1 or 2", 2153 __func__)); 2154 if (tp->t_dupacks == 1) 2155 tp->snd_limited = 0; 2156 tp->snd_cwnd = 2157 (tp->snd_nxt - tp->snd_una) + 2158 (tp->t_dupacks - tp->snd_limited) * 2159 tp->t_maxseg; 2160 (void) tcp_output(tp); 2161 sent = tp->snd_max - oldsndmax; 2162 if (sent > tp->t_maxseg) { 2163 KASSERT((tp->t_dupacks == 2 && 2164 tp->snd_limited == 0) || 2165 (sent == tp->t_maxseg + 1 && 2166 tp->t_flags & TF_SENTFIN), 2167 ("%s: sent too much", 2168 __func__)); 2169 tp->snd_limited = 2; 2170 } else if (sent > 0) 2171 ++tp->snd_limited; 2172 tp->snd_cwnd = oldcwnd; 2173 goto drop; 2174 } 2175 } else 2176 tp->t_dupacks = 0; 2177 break; 2178 } 2179 2180 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2181 ("%s: th_ack <= snd_una", __func__)); 2182 2183 /* 2184 * If the congestion window was inflated to account 2185 * for the other side's cached packets, retract it. 2186 */ 2187 if (V_tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) { 2188 if (IN_FASTRECOVERY(tp)) { 2189 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2190 if (tp->t_flags & TF_SACK_PERMIT) 2191 tcp_sack_partialack(tp, th); 2192 else 2193 tcp_newreno_partial_ack(tp, th); 2194 } else { 2195 /* 2196 * Out of fast recovery. 2197 * Window inflation should have left us 2198 * with approximately snd_ssthresh 2199 * outstanding data. 2200 * But in case we would be inclined to 2201 * send a burst, better to do it via 2202 * the slow start mechanism. 2203 */ 2204 if (SEQ_GT(th->th_ack + 2205 tp->snd_ssthresh, 2206 tp->snd_max)) 2207 tp->snd_cwnd = tp->snd_max - 2208 th->th_ack + 2209 tp->t_maxseg; 2210 else 2211 tp->snd_cwnd = tp->snd_ssthresh; 2212 } 2213 } 2214 } else { 2215 if (tp->t_dupacks >= tcprexmtthresh && 2216 tp->snd_cwnd > tp->snd_ssthresh) 2217 tp->snd_cwnd = tp->snd_ssthresh; 2218 } 2219 tp->t_dupacks = 0; 2220 /* 2221 * If we reach this point, ACK is not a duplicate, 2222 * i.e., it ACKs something we sent. 2223 */ 2224 if (tp->t_flags & TF_NEEDSYN) { 2225 /* 2226 * T/TCP: Connection was half-synchronized, and our 2227 * SYN has been ACK'd (so connection is now fully 2228 * synchronized). Go to non-starred state, 2229 * increment snd_una for ACK of SYN, and check if 2230 * we can do window scaling. 2231 */ 2232 tp->t_flags &= ~TF_NEEDSYN; 2233 tp->snd_una++; 2234 /* Do window scaling? */ 2235 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2236 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2237 tp->rcv_scale = tp->request_r_scale; 2238 /* Send window already scaled. */ 2239 } 2240 } 2241 2242process_ACK: 2243 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2244 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2245 ("tcp_input: process_ACK ti_locked %d", ti_locked)); 2246 INP_WLOCK_ASSERT(tp->t_inpcb); 2247 2248 acked = th->th_ack - tp->snd_una; 2249 TCPSTAT_INC(tcps_rcvackpack); 2250 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2251 2252 /* 2253 * If we just performed our first retransmit, and the ACK 2254 * arrives within our recovery window, then it was a mistake 2255 * to do the retransmit in the first place. Recover our 2256 * original cwnd and ssthresh, and proceed to transmit where 2257 * we left off. 2258 */ 2259 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2260 TCPSTAT_INC(tcps_sndrexmitbad); 2261 tp->snd_cwnd = tp->snd_cwnd_prev; 2262 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2263 tp->snd_recover = tp->snd_recover_prev; 2264 if (tp->t_flags & TF_WASFRECOVERY) 2265 ENTER_FASTRECOVERY(tp); 2266 tp->snd_nxt = tp->snd_max; 2267 tp->t_badrxtwin = 0; /* XXX probably not required */ 2268 } 2269 2270 /* 2271 * If we have a timestamp reply, update smoothed 2272 * round trip time. If no timestamp is present but 2273 * transmit timer is running and timed sequence 2274 * number was acked, update smoothed round trip time. 2275 * Since we now have an rtt measurement, cancel the 2276 * timer backoff (cf., Phil Karn's retransmit alg.). 2277 * Recompute the initial retransmit timer. 2278 * 2279 * Some boxes send broken timestamp replies 2280 * during the SYN+ACK phase, ignore 2281 * timestamps of 0 or we could calculate a 2282 * huge RTT and blow up the retransmit timer. 2283 */ 2284 if ((to.to_flags & TOF_TS) != 0 && 2285 to.to_tsecr) { 2286 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2287 tp->t_rttlow = ticks - to.to_tsecr; 2288 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2289 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2290 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2291 tp->t_rttlow = ticks - tp->t_rtttime; 2292 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2293 } 2294 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2295 2296 /* 2297 * If all outstanding data is acked, stop retransmit 2298 * timer and remember to restart (more output or persist). 2299 * If there is more data to be acked, restart retransmit 2300 * timer, using current (possibly backed-off) value. 2301 */ 2302 if (th->th_ack == tp->snd_max) { 2303 tcp_timer_activate(tp, TT_REXMT, 0); 2304 needoutput = 1; 2305 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2306 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2307 2308 /* 2309 * If no data (only SYN) was ACK'd, 2310 * skip rest of ACK processing. 2311 */ 2312 if (acked == 0) 2313 goto step6; 2314 2315 /* 2316 * When new data is acked, open the congestion window. 2317 * Method depends on which congestion control state we're 2318 * in (slow start or cong avoid) and if ABC (RFC 3465) is 2319 * enabled. 2320 * 2321 * slow start: cwnd <= ssthresh 2322 * cong avoid: cwnd > ssthresh 2323 * 2324 * slow start and ABC (RFC 3465): 2325 * Grow cwnd exponentially by the amount of data 2326 * ACKed capping the max increment per ACK to 2327 * (abc_l_var * maxseg) bytes. 2328 * 2329 * slow start without ABC (RFC 2581): 2330 * Grow cwnd exponentially by maxseg per ACK. 2331 * 2332 * cong avoid and ABC (RFC 3465): 2333 * Grow cwnd linearly by maxseg per RTT for each 2334 * cwnd worth of ACKed data. 2335 * 2336 * cong avoid without ABC (RFC 2581): 2337 * Grow cwnd linearly by approximately maxseg per RTT using 2338 * maxseg^2 / cwnd per ACK as the increment. 2339 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to 2340 * avoid capping cwnd. 2341 */ 2342 if ((!V_tcp_do_newreno && !(tp->t_flags & TF_SACK_PERMIT)) || 2343 !IN_FASTRECOVERY(tp)) { 2344 u_int cw = tp->snd_cwnd; 2345 u_int incr = tp->t_maxseg; 2346 /* In congestion avoidance? */ 2347 if (cw > tp->snd_ssthresh) { 2348 if (V_tcp_do_rfc3465) { 2349 tp->t_bytes_acked += acked; 2350 if (tp->t_bytes_acked >= tp->snd_cwnd) 2351 tp->t_bytes_acked -= cw; 2352 else 2353 incr = 0; 2354 } 2355 else 2356 incr = max((incr * incr / cw), 1); 2357 /* 2358 * In slow-start with ABC enabled and no RTO in sight? 2359 * (Must not use abc_l_var > 1 if slow starting after an 2360 * RTO. On RTO, snd_nxt = snd_una, so the snd_nxt == 2361 * snd_max check is sufficient to handle this). 2362 */ 2363 } else if (V_tcp_do_rfc3465 && 2364 tp->snd_nxt == tp->snd_max) 2365 incr = min(acked, 2366 V_tcp_abc_l_var * tp->t_maxseg); 2367 /* ABC is on by default, so (incr == 0) frequently. */ 2368 if (incr > 0) 2369 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2370 } 2371 SOCKBUF_LOCK(&so->so_snd); 2372 if (acked > so->so_snd.sb_cc) { 2373 tp->snd_wnd -= so->so_snd.sb_cc; 2374 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2375 ourfinisacked = 1; 2376 } else { 2377 sbdrop_locked(&so->so_snd, acked); 2378 tp->snd_wnd -= acked; 2379 ourfinisacked = 0; 2380 } 2381 /* NB: sowwakeup_locked() does an implicit unlock. */ 2382 sowwakeup_locked(so); 2383 /* Detect una wraparound. */ 2384 if ((V_tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2385 !IN_FASTRECOVERY(tp) && 2386 SEQ_GT(tp->snd_una, tp->snd_recover) && 2387 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2388 tp->snd_recover = th->th_ack - 1; 2389 if ((V_tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2390 IN_FASTRECOVERY(tp) && 2391 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2392 EXIT_FASTRECOVERY(tp); 2393 tp->t_bytes_acked = 0; 2394 } 2395 tp->snd_una = th->th_ack; 2396 if (tp->t_flags & TF_SACK_PERMIT) { 2397 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2398 tp->snd_recover = tp->snd_una; 2399 } 2400 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2401 tp->snd_nxt = tp->snd_una; 2402 2403 switch (tp->t_state) { 2404 2405 /* 2406 * In FIN_WAIT_1 STATE in addition to the processing 2407 * for the ESTABLISHED state if our FIN is now acknowledged 2408 * then enter FIN_WAIT_2. 2409 */ 2410 case TCPS_FIN_WAIT_1: 2411 if (ourfinisacked) { 2412 /* 2413 * If we can't receive any more 2414 * data, then closing user can proceed. 2415 * Starting the timer is contrary to the 2416 * specification, but if we don't get a FIN 2417 * we'll hang forever. 2418 * 2419 * XXXjl: 2420 * we should release the tp also, and use a 2421 * compressed state. 2422 */ 2423 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2424 int timeout; 2425 2426 soisdisconnected(so); 2427 timeout = (tcp_fast_finwait2_recycle) ? 2428 tcp_finwait2_timeout : tcp_maxidle; 2429 tcp_timer_activate(tp, TT_2MSL, timeout); 2430 } 2431 tp->t_state = TCPS_FIN_WAIT_2; 2432 } 2433 break; 2434 2435 /* 2436 * In CLOSING STATE in addition to the processing for 2437 * the ESTABLISHED state if the ACK acknowledges our FIN 2438 * then enter the TIME-WAIT state, otherwise ignore 2439 * the segment. 2440 */ 2441 case TCPS_CLOSING: 2442 if (ourfinisacked) { 2443 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2444 tcp_twstart(tp); 2445 INP_INFO_WUNLOCK(&V_tcbinfo); 2446 m_freem(m); 2447 return; 2448 } 2449 break; 2450 2451 /* 2452 * In LAST_ACK, we may still be waiting for data to drain 2453 * and/or to be acked, as well as for the ack of our FIN. 2454 * If our FIN is now acknowledged, delete the TCB, 2455 * enter the closed state and return. 2456 */ 2457 case TCPS_LAST_ACK: 2458 if (ourfinisacked) { 2459 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2460 tp = tcp_close(tp); 2461 goto drop; 2462 } 2463 break; 2464 } 2465 } 2466 2467step6: 2468 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2469 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2470 ("tcp_do_segment: step6 ti_locked %d", ti_locked)); 2471 INP_WLOCK_ASSERT(tp->t_inpcb); 2472 2473 /* 2474 * Update window information. 2475 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2476 */ 2477 if ((thflags & TH_ACK) && 2478 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2479 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2480 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2481 /* keep track of pure window updates */ 2482 if (tlen == 0 && 2483 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2484 TCPSTAT_INC(tcps_rcvwinupd); 2485 tp->snd_wnd = tiwin; 2486 tp->snd_wl1 = th->th_seq; 2487 tp->snd_wl2 = th->th_ack; 2488 if (tp->snd_wnd > tp->max_sndwnd) 2489 tp->max_sndwnd = tp->snd_wnd; 2490 needoutput = 1; 2491 } 2492 2493 /* 2494 * Process segments with URG. 2495 */ 2496 if ((thflags & TH_URG) && th->th_urp && 2497 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2498 /* 2499 * This is a kludge, but if we receive and accept 2500 * random urgent pointers, we'll crash in 2501 * soreceive. It's hard to imagine someone 2502 * actually wanting to send this much urgent data. 2503 */ 2504 SOCKBUF_LOCK(&so->so_rcv); 2505 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2506 th->th_urp = 0; /* XXX */ 2507 thflags &= ~TH_URG; /* XXX */ 2508 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2509 goto dodata; /* XXX */ 2510 } 2511 /* 2512 * If this segment advances the known urgent pointer, 2513 * then mark the data stream. This should not happen 2514 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2515 * a FIN has been received from the remote side. 2516 * In these states we ignore the URG. 2517 * 2518 * According to RFC961 (Assigned Protocols), 2519 * the urgent pointer points to the last octet 2520 * of urgent data. We continue, however, 2521 * to consider it to indicate the first octet 2522 * of data past the urgent section as the original 2523 * spec states (in one of two places). 2524 */ 2525 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2526 tp->rcv_up = th->th_seq + th->th_urp; 2527 so->so_oobmark = so->so_rcv.sb_cc + 2528 (tp->rcv_up - tp->rcv_nxt) - 1; 2529 if (so->so_oobmark == 0) 2530 so->so_rcv.sb_state |= SBS_RCVATMARK; 2531 sohasoutofband(so); 2532 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2533 } 2534 SOCKBUF_UNLOCK(&so->so_rcv); 2535 /* 2536 * Remove out of band data so doesn't get presented to user. 2537 * This can happen independent of advancing the URG pointer, 2538 * but if two URG's are pending at once, some out-of-band 2539 * data may creep in... ick. 2540 */ 2541 if (th->th_urp <= (u_long)tlen && 2542 !(so->so_options & SO_OOBINLINE)) { 2543 /* hdr drop is delayed */ 2544 tcp_pulloutofband(so, th, m, drop_hdrlen); 2545 } 2546 } else { 2547 /* 2548 * If no out of band data is expected, 2549 * pull receive urgent pointer along 2550 * with the receive window. 2551 */ 2552 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2553 tp->rcv_up = tp->rcv_nxt; 2554 } 2555dodata: /* XXX */ 2556 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2557 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2558 ("tcp_do_segment: dodata ti_locked %d", ti_locked)); 2559 INP_WLOCK_ASSERT(tp->t_inpcb); 2560 2561 /* 2562 * Process the segment text, merging it into the TCP sequencing queue, 2563 * and arranging for acknowledgment of receipt if necessary. 2564 * This process logically involves adjusting tp->rcv_wnd as data 2565 * is presented to the user (this happens in tcp_usrreq.c, 2566 * case PRU_RCVD). If a FIN has already been received on this 2567 * connection then we just ignore the text. 2568 */ 2569 if ((tlen || (thflags & TH_FIN)) && 2570 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2571 tcp_seq save_start = th->th_seq; 2572 m_adj(m, drop_hdrlen); /* delayed header drop */ 2573 /* 2574 * Insert segment which includes th into TCP reassembly queue 2575 * with control block tp. Set thflags to whether reassembly now 2576 * includes a segment with FIN. This handles the common case 2577 * inline (segment is the next to be received on an established 2578 * connection, and the queue is empty), avoiding linkage into 2579 * and removal from the queue and repetition of various 2580 * conversions. 2581 * Set DELACK for segments received in order, but ack 2582 * immediately when segments are out of order (so 2583 * fast retransmit can work). 2584 */ 2585 if (th->th_seq == tp->rcv_nxt && 2586 LIST_EMPTY(&tp->t_segq) && 2587 TCPS_HAVEESTABLISHED(tp->t_state)) { 2588 if (DELAY_ACK(tp)) 2589 tp->t_flags |= TF_DELACK; 2590 else 2591 tp->t_flags |= TF_ACKNOW; 2592 tp->rcv_nxt += tlen; 2593 thflags = th->th_flags & TH_FIN; 2594 TCPSTAT_INC(tcps_rcvpack); 2595 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2596 ND6_HINT(tp); 2597 SOCKBUF_LOCK(&so->so_rcv); 2598 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2599 m_freem(m); 2600 else 2601 sbappendstream_locked(&so->so_rcv, m); 2602 /* NB: sorwakeup_locked() does an implicit unlock. */ 2603 sorwakeup_locked(so); 2604 } else { 2605 /* 2606 * XXX: Due to the header drop above "th" is 2607 * theoretically invalid by now. Fortunately 2608 * m_adj() doesn't actually frees any mbufs 2609 * when trimming from the head. 2610 */ 2611 thflags = tcp_reass(tp, th, &tlen, m); 2612 tp->t_flags |= TF_ACKNOW; 2613 } 2614 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2615 tcp_update_sack_list(tp, save_start, save_start + tlen); 2616#if 0 2617 /* 2618 * Note the amount of data that peer has sent into 2619 * our window, in order to estimate the sender's 2620 * buffer size. 2621 * XXX: Unused. 2622 */ 2623 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2624#endif 2625 } else { 2626 m_freem(m); 2627 thflags &= ~TH_FIN; 2628 } 2629 2630 /* 2631 * If FIN is received ACK the FIN and let the user know 2632 * that the connection is closing. 2633 */ 2634 if (thflags & TH_FIN) { 2635 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2636 socantrcvmore(so); 2637 /* 2638 * If connection is half-synchronized 2639 * (ie NEEDSYN flag on) then delay ACK, 2640 * so it may be piggybacked when SYN is sent. 2641 * Otherwise, since we received a FIN then no 2642 * more input can be expected, send ACK now. 2643 */ 2644 if (tp->t_flags & TF_NEEDSYN) 2645 tp->t_flags |= TF_DELACK; 2646 else 2647 tp->t_flags |= TF_ACKNOW; 2648 tp->rcv_nxt++; 2649 } 2650 switch (tp->t_state) { 2651 2652 /* 2653 * In SYN_RECEIVED and ESTABLISHED STATES 2654 * enter the CLOSE_WAIT state. 2655 */ 2656 case TCPS_SYN_RECEIVED: 2657 tp->t_starttime = ticks; 2658 /* FALLTHROUGH */ 2659 case TCPS_ESTABLISHED: 2660 tp->t_state = TCPS_CLOSE_WAIT; 2661 break; 2662 2663 /* 2664 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2665 * enter the CLOSING state. 2666 */ 2667 case TCPS_FIN_WAIT_1: 2668 tp->t_state = TCPS_CLOSING; 2669 break; 2670 2671 /* 2672 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2673 * starting the time-wait timer, turning off the other 2674 * standard timers. 2675 */ 2676 case TCPS_FIN_WAIT_2: 2677 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2678 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2679 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2680 ti_locked)); 2681 2682 tcp_twstart(tp); 2683 INP_INFO_WUNLOCK(&V_tcbinfo); 2684 return; 2685 } 2686 } 2687 if (ti_locked == TI_RLOCKED) 2688 INP_INFO_RUNLOCK(&V_tcbinfo); 2689 else if (ti_locked == TI_WLOCKED) 2690 INP_INFO_WUNLOCK(&V_tcbinfo); 2691 else 2692 panic("%s: dodata epilogue ti_locked %d", __func__, 2693 ti_locked); 2694 ti_locked = TI_UNLOCKED; 2695 2696#ifdef TCPDEBUG 2697 if (so->so_options & SO_DEBUG) 2698 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2699 &tcp_savetcp, 0); 2700#endif 2701 2702 /* 2703 * Return any desired output. 2704 */ 2705 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2706 (void) tcp_output(tp); 2707 2708check_delack: 2709 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2710 __func__, ti_locked)); 2711 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2712 INP_WLOCK_ASSERT(tp->t_inpcb); 2713 2714 if (tp->t_flags & TF_DELACK) { 2715 tp->t_flags &= ~TF_DELACK; 2716 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2717 } 2718 INP_WUNLOCK(tp->t_inpcb); 2719 return; 2720 2721dropafterack: 2722 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2723 ("tcp_do_segment: dropafterack ti_locked %d", ti_locked)); 2724 2725 /* 2726 * Generate an ACK dropping incoming segment if it occupies 2727 * sequence space, where the ACK reflects our state. 2728 * 2729 * We can now skip the test for the RST flag since all 2730 * paths to this code happen after packets containing 2731 * RST have been dropped. 2732 * 2733 * In the SYN-RECEIVED state, don't send an ACK unless the 2734 * segment we received passes the SYN-RECEIVED ACK test. 2735 * If it fails send a RST. This breaks the loop in the 2736 * "LAND" DoS attack, and also prevents an ACK storm 2737 * between two listening ports that have been sent forged 2738 * SYN segments, each with the source address of the other. 2739 */ 2740 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2741 (SEQ_GT(tp->snd_una, th->th_ack) || 2742 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2743 rstreason = BANDLIM_RST_OPENPORT; 2744 goto dropwithreset; 2745 } 2746#ifdef TCPDEBUG 2747 if (so->so_options & SO_DEBUG) 2748 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2749 &tcp_savetcp, 0); 2750#endif 2751 if (ti_locked == TI_RLOCKED) 2752 INP_INFO_RUNLOCK(&V_tcbinfo); 2753 else if (ti_locked == TI_WLOCKED) 2754 INP_INFO_WUNLOCK(&V_tcbinfo); 2755 else 2756 panic("%s: dropafterack epilogue ti_locked %d", __func__, 2757 ti_locked); 2758 ti_locked = TI_UNLOCKED; 2759 2760 tp->t_flags |= TF_ACKNOW; 2761 (void) tcp_output(tp); 2762 INP_WUNLOCK(tp->t_inpcb); 2763 m_freem(m); 2764 return; 2765 2766dropwithreset: 2767 if (ti_locked == TI_RLOCKED) 2768 INP_INFO_RUNLOCK(&V_tcbinfo); 2769 else if (ti_locked == TI_WLOCKED) 2770 INP_INFO_WUNLOCK(&V_tcbinfo); 2771 else 2772 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked); 2773 ti_locked = TI_UNLOCKED; 2774 2775 if (tp != NULL) { 2776 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2777 INP_WUNLOCK(tp->t_inpcb); 2778 } else 2779 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 2780 return; 2781 2782drop: 2783 if (ti_locked == TI_RLOCKED) 2784 INP_INFO_RUNLOCK(&V_tcbinfo); 2785 else if (ti_locked == TI_WLOCKED) 2786 INP_INFO_WUNLOCK(&V_tcbinfo); 2787#ifdef INVARIANTS 2788 else 2789 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2790#endif 2791 ti_locked = TI_UNLOCKED; 2792 2793 /* 2794 * Drop space held by incoming segment and return. 2795 */ 2796#ifdef TCPDEBUG 2797 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2798 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2799 &tcp_savetcp, 0); 2800#endif 2801 if (tp != NULL) 2802 INP_WUNLOCK(tp->t_inpcb); 2803 m_freem(m); 2804} 2805 2806/* 2807 * Issue RST and make ACK acceptable to originator of segment. 2808 * The mbuf must still include the original packet header. 2809 * tp may be NULL. 2810 */ 2811static void 2812tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 2813 int tlen, int rstreason) 2814{ 2815 struct ip *ip; 2816#ifdef INET6 2817 struct ip6_hdr *ip6; 2818#endif 2819 2820 if (tp != NULL) { 2821 INP_WLOCK_ASSERT(tp->t_inpcb); 2822 } 2823 2824 /* Don't bother if destination was broadcast/multicast. */ 2825 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2826 goto drop; 2827#ifdef INET6 2828 if (mtod(m, struct ip *)->ip_v == 6) { 2829 ip6 = mtod(m, struct ip6_hdr *); 2830 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2831 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2832 goto drop; 2833 /* IPv6 anycast check is done at tcp6_input() */ 2834 } else 2835#endif 2836 { 2837 ip = mtod(m, struct ip *); 2838 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2839 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2840 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2841 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2842 goto drop; 2843 } 2844 2845 /* Perform bandwidth limiting. */ 2846 if (badport_bandlim(rstreason) < 0) 2847 goto drop; 2848 2849 /* tcp_respond consumes the mbuf chain. */ 2850 if (th->th_flags & TH_ACK) { 2851 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 2852 th->th_ack, TH_RST); 2853 } else { 2854 if (th->th_flags & TH_SYN) 2855 tlen++; 2856 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2857 (tcp_seq)0, TH_RST|TH_ACK); 2858 } 2859 return; 2860drop: 2861 m_freem(m); 2862} 2863 2864/* 2865 * Parse TCP options and place in tcpopt. 2866 */ 2867static void 2868tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 2869{ 2870 INIT_VNET_INET(curvnet); 2871 int opt, optlen; 2872 2873 to->to_flags = 0; 2874 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2875 opt = cp[0]; 2876 if (opt == TCPOPT_EOL) 2877 break; 2878 if (opt == TCPOPT_NOP) 2879 optlen = 1; 2880 else { 2881 if (cnt < 2) 2882 break; 2883 optlen = cp[1]; 2884 if (optlen < 2 || optlen > cnt) 2885 break; 2886 } 2887 switch (opt) { 2888 case TCPOPT_MAXSEG: 2889 if (optlen != TCPOLEN_MAXSEG) 2890 continue; 2891 if (!(flags & TO_SYN)) 2892 continue; 2893 to->to_flags |= TOF_MSS; 2894 bcopy((char *)cp + 2, 2895 (char *)&to->to_mss, sizeof(to->to_mss)); 2896 to->to_mss = ntohs(to->to_mss); 2897 break; 2898 case TCPOPT_WINDOW: 2899 if (optlen != TCPOLEN_WINDOW) 2900 continue; 2901 if (!(flags & TO_SYN)) 2902 continue; 2903 to->to_flags |= TOF_SCALE; 2904 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 2905 break; 2906 case TCPOPT_TIMESTAMP: 2907 if (optlen != TCPOLEN_TIMESTAMP) 2908 continue; 2909 to->to_flags |= TOF_TS; 2910 bcopy((char *)cp + 2, 2911 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2912 to->to_tsval = ntohl(to->to_tsval); 2913 bcopy((char *)cp + 6, 2914 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2915 to->to_tsecr = ntohl(to->to_tsecr); 2916 break; 2917#ifdef TCP_SIGNATURE 2918 /* 2919 * XXX In order to reply to a host which has set the 2920 * TCP_SIGNATURE option in its initial SYN, we have to 2921 * record the fact that the option was observed here 2922 * for the syncache code to perform the correct response. 2923 */ 2924 case TCPOPT_SIGNATURE: 2925 if (optlen != TCPOLEN_SIGNATURE) 2926 continue; 2927 to->to_flags |= TOF_SIGNATURE; 2928 to->to_signature = cp + 2; 2929 break; 2930#endif 2931 case TCPOPT_SACK_PERMITTED: 2932 if (optlen != TCPOLEN_SACK_PERMITTED) 2933 continue; 2934 if (!(flags & TO_SYN)) 2935 continue; 2936 if (!V_tcp_do_sack) 2937 continue; 2938 to->to_flags |= TOF_SACKPERM; 2939 break; 2940 case TCPOPT_SACK: 2941 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2942 continue; 2943 if (flags & TO_SYN) 2944 continue; 2945 to->to_flags |= TOF_SACK; 2946 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2947 to->to_sacks = cp + 2; 2948 TCPSTAT_INC(tcps_sack_rcv_blocks); 2949 break; 2950 default: 2951 continue; 2952 } 2953 } 2954} 2955 2956/* 2957 * Pull out of band byte out of a segment so 2958 * it doesn't appear in the user's data queue. 2959 * It is still reflected in the segment length for 2960 * sequencing purposes. 2961 */ 2962static void 2963tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 2964 int off) 2965{ 2966 int cnt = off + th->th_urp - 1; 2967 2968 while (cnt >= 0) { 2969 if (m->m_len > cnt) { 2970 char *cp = mtod(m, caddr_t) + cnt; 2971 struct tcpcb *tp = sototcpcb(so); 2972 2973 INP_WLOCK_ASSERT(tp->t_inpcb); 2974 2975 tp->t_iobc = *cp; 2976 tp->t_oobflags |= TCPOOB_HAVEDATA; 2977 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2978 m->m_len--; 2979 if (m->m_flags & M_PKTHDR) 2980 m->m_pkthdr.len--; 2981 return; 2982 } 2983 cnt -= m->m_len; 2984 m = m->m_next; 2985 if (m == NULL) 2986 break; 2987 } 2988 panic("tcp_pulloutofband"); 2989} 2990 2991/* 2992 * Collect new round-trip time estimate 2993 * and update averages and current timeout. 2994 */ 2995static void 2996tcp_xmit_timer(struct tcpcb *tp, int rtt) 2997{ 2998 INIT_VNET_INET(tp->t_inpcb->inp_vnet); 2999 int delta; 3000 3001 INP_WLOCK_ASSERT(tp->t_inpcb); 3002 3003 TCPSTAT_INC(tcps_rttupdated); 3004 tp->t_rttupdated++; 3005 if (tp->t_srtt != 0) { 3006 /* 3007 * srtt is stored as fixed point with 5 bits after the 3008 * binary point (i.e., scaled by 8). The following magic 3009 * is equivalent to the smoothing algorithm in rfc793 with 3010 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3011 * point). Adjust rtt to origin 0. 3012 */ 3013 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3014 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3015 3016 if ((tp->t_srtt += delta) <= 0) 3017 tp->t_srtt = 1; 3018 3019 /* 3020 * We accumulate a smoothed rtt variance (actually, a 3021 * smoothed mean difference), then set the retransmit 3022 * timer to smoothed rtt + 4 times the smoothed variance. 3023 * rttvar is stored as fixed point with 4 bits after the 3024 * binary point (scaled by 16). The following is 3025 * equivalent to rfc793 smoothing with an alpha of .75 3026 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3027 * rfc793's wired-in beta. 3028 */ 3029 if (delta < 0) 3030 delta = -delta; 3031 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3032 if ((tp->t_rttvar += delta) <= 0) 3033 tp->t_rttvar = 1; 3034 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3035 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3036 } else { 3037 /* 3038 * No rtt measurement yet - use the unsmoothed rtt. 3039 * Set the variance to half the rtt (so our first 3040 * retransmit happens at 3*rtt). 3041 */ 3042 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3043 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3044 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3045 } 3046 tp->t_rtttime = 0; 3047 tp->t_rxtshift = 0; 3048 3049 /* 3050 * the retransmit should happen at rtt + 4 * rttvar. 3051 * Because of the way we do the smoothing, srtt and rttvar 3052 * will each average +1/2 tick of bias. When we compute 3053 * the retransmit timer, we want 1/2 tick of rounding and 3054 * 1 extra tick because of +-1/2 tick uncertainty in the 3055 * firing of the timer. The bias will give us exactly the 3056 * 1.5 tick we need. But, because the bias is 3057 * statistical, we have to test that we don't drop below 3058 * the minimum feasible timer (which is 2 ticks). 3059 */ 3060 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3061 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3062 3063 /* 3064 * We received an ack for a packet that wasn't retransmitted; 3065 * it is probably safe to discard any error indications we've 3066 * received recently. This isn't quite right, but close enough 3067 * for now (a route might have failed after we sent a segment, 3068 * and the return path might not be symmetrical). 3069 */ 3070 tp->t_softerror = 0; 3071} 3072 3073/* 3074 * Determine a reasonable value for maxseg size. 3075 * If the route is known, check route for mtu. 3076 * If none, use an mss that can be handled on the outgoing 3077 * interface without forcing IP to fragment; if bigger than 3078 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 3079 * to utilize large mbufs. If no route is found, route has no mtu, 3080 * or the destination isn't local, use a default, hopefully conservative 3081 * size (usually 512 or the default IP max size, but no more than the mtu 3082 * of the interface), as we can't discover anything about intervening 3083 * gateways or networks. We also initialize the congestion/slow start 3084 * window to be a single segment if the destination isn't local. 3085 * While looking at the routing entry, we also initialize other path-dependent 3086 * parameters from pre-set or cached values in the routing entry. 3087 * 3088 * Also take into account the space needed for options that we 3089 * send regularly. Make maxseg shorter by that amount to assure 3090 * that we can send maxseg amount of data even when the options 3091 * are present. Store the upper limit of the length of options plus 3092 * data in maxopd. 3093 * 3094 * In case of T/TCP, we call this routine during implicit connection 3095 * setup as well (offer = -1), to initialize maxseg from the cached 3096 * MSS of our peer. 3097 * 3098 * NOTE that this routine is only called when we process an incoming 3099 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 3100 */ 3101void 3102tcp_mss_update(struct tcpcb *tp, int offer, 3103 struct hc_metrics_lite *metricptr, int *mtuflags) 3104{ 3105 INIT_VNET_INET(tp->t_inpcb->inp_vnet); 3106 int mss; 3107 u_long maxmtu; 3108 struct inpcb *inp = tp->t_inpcb; 3109 struct hc_metrics_lite metrics; 3110 int origoffer = offer; 3111#ifdef INET6 3112 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3113 size_t min_protoh = isipv6 ? 3114 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3115 sizeof (struct tcpiphdr); 3116#else 3117 const size_t min_protoh = sizeof(struct tcpiphdr); 3118#endif 3119 3120 INP_WLOCK_ASSERT(tp->t_inpcb); 3121 3122 /* Initialize. */ 3123#ifdef INET6 3124 if (isipv6) { 3125 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags); 3126 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3127 } else 3128#endif 3129 { 3130 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags); 3131 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3132 } 3133 3134 /* 3135 * No route to sender, stay with default mss and return. 3136 */ 3137 if (maxmtu == 0) { 3138 /* 3139 * In case we return early we need to initialize metrics 3140 * to a defined state as tcp_hc_get() would do for us 3141 * if there was no cache hit. 3142 */ 3143 if (metricptr != NULL) 3144 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3145 return; 3146 } 3147 3148 /* What have we got? */ 3149 switch (offer) { 3150 case 0: 3151 /* 3152 * Offer == 0 means that there was no MSS on the SYN 3153 * segment, in this case we use tcp_mssdflt as 3154 * already assigned to t_maxopd above. 3155 */ 3156 offer = tp->t_maxopd; 3157 break; 3158 3159 case -1: 3160 /* 3161 * Offer == -1 means that we didn't receive SYN yet. 3162 */ 3163 /* FALLTHROUGH */ 3164 3165 default: 3166 /* 3167 * Prevent DoS attack with too small MSS. Round up 3168 * to at least minmss. 3169 */ 3170 offer = max(offer, V_tcp_minmss); 3171 } 3172 3173 /* 3174 * rmx information is now retrieved from tcp_hostcache. 3175 */ 3176 tcp_hc_get(&inp->inp_inc, &metrics); 3177 if (metricptr != NULL) 3178 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3179 3180 /* 3181 * If there's a discovered mtu int tcp hostcache, use it 3182 * else, use the link mtu. 3183 */ 3184 if (metrics.rmx_mtu) 3185 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3186 else { 3187#ifdef INET6 3188 if (isipv6) { 3189 mss = maxmtu - min_protoh; 3190 if (!V_path_mtu_discovery && 3191 !in6_localaddr(&inp->in6p_faddr)) 3192 mss = min(mss, V_tcp_v6mssdflt); 3193 } else 3194#endif 3195 { 3196 mss = maxmtu - min_protoh; 3197 if (!V_path_mtu_discovery && 3198 !in_localaddr(inp->inp_faddr)) 3199 mss = min(mss, V_tcp_mssdflt); 3200 } 3201 /* 3202 * XXX - The above conditional (mss = maxmtu - min_protoh) 3203 * probably violates the TCP spec. 3204 * The problem is that, since we don't know the 3205 * other end's MSS, we are supposed to use a conservative 3206 * default. But, if we do that, then MTU discovery will 3207 * never actually take place, because the conservative 3208 * default is much less than the MTUs typically seen 3209 * on the Internet today. For the moment, we'll sweep 3210 * this under the carpet. 3211 * 3212 * The conservative default might not actually be a problem 3213 * if the only case this occurs is when sending an initial 3214 * SYN with options and data to a host we've never talked 3215 * to before. Then, they will reply with an MSS value which 3216 * will get recorded and the new parameters should get 3217 * recomputed. For Further Study. 3218 */ 3219 } 3220 mss = min(mss, offer); 3221 3222 /* 3223 * Sanity check: make sure that maxopd will be large 3224 * enough to allow some data on segments even if the 3225 * all the option space is used (40bytes). Otherwise 3226 * funny things may happen in tcp_output. 3227 */ 3228 mss = max(mss, 64); 3229 3230 /* 3231 * maxopd stores the maximum length of data AND options 3232 * in a segment; maxseg is the amount of data in a normal 3233 * segment. We need to store this value (maxopd) apart 3234 * from maxseg, because now every segment carries options 3235 * and thus we normally have somewhat less data in segments. 3236 */ 3237 tp->t_maxopd = mss; 3238 3239 /* 3240 * origoffer==-1 indicates that no segments were received yet. 3241 * In this case we just guess. 3242 */ 3243 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3244 (origoffer == -1 || 3245 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3246 mss -= TCPOLEN_TSTAMP_APPA; 3247 3248#if (MCLBYTES & (MCLBYTES - 1)) == 0 3249 if (mss > MCLBYTES) 3250 mss &= ~(MCLBYTES-1); 3251#else 3252 if (mss > MCLBYTES) 3253 mss = mss / MCLBYTES * MCLBYTES; 3254#endif 3255 tp->t_maxseg = mss; 3256} 3257 3258void 3259tcp_mss(struct tcpcb *tp, int offer) 3260{ 3261 int rtt, mss; 3262 u_long bufsize; 3263 struct inpcb *inp; 3264 struct socket *so; 3265 struct hc_metrics_lite metrics; 3266 int mtuflags = 0; 3267#ifdef INET6 3268 int isipv6; 3269#endif 3270 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3271 INIT_VNET_INET(tp->t_vnet); 3272 3273 tcp_mss_update(tp, offer, &metrics, &mtuflags); 3274 3275 mss = tp->t_maxseg; 3276 inp = tp->t_inpcb; 3277#ifdef INET6 3278 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3279#endif 3280 3281 /* 3282 * If there's a pipesize, change the socket buffer to that size, 3283 * don't change if sb_hiwat is different than default (then it 3284 * has been changed on purpose with setsockopt). 3285 * Make the socket buffers an integral number of mss units; 3286 * if the mss is larger than the socket buffer, decrease the mss. 3287 */ 3288 so = inp->inp_socket; 3289 SOCKBUF_LOCK(&so->so_snd); 3290 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3291 bufsize = metrics.rmx_sendpipe; 3292 else 3293 bufsize = so->so_snd.sb_hiwat; 3294 if (bufsize < mss) 3295 mss = bufsize; 3296 else { 3297 bufsize = roundup(bufsize, mss); 3298 if (bufsize > sb_max) 3299 bufsize = sb_max; 3300 if (bufsize > so->so_snd.sb_hiwat) 3301 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3302 } 3303 SOCKBUF_UNLOCK(&so->so_snd); 3304 tp->t_maxseg = mss; 3305 3306 SOCKBUF_LOCK(&so->so_rcv); 3307 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3308 bufsize = metrics.rmx_recvpipe; 3309 else 3310 bufsize = so->so_rcv.sb_hiwat; 3311 if (bufsize > mss) { 3312 bufsize = roundup(bufsize, mss); 3313 if (bufsize > sb_max) 3314 bufsize = sb_max; 3315 if (bufsize > so->so_rcv.sb_hiwat) 3316 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3317 } 3318 SOCKBUF_UNLOCK(&so->so_rcv); 3319 /* 3320 * While we're here, check the others too. 3321 */ 3322 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3323 tp->t_srtt = rtt; 3324 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3325 TCPSTAT_INC(tcps_usedrtt); 3326 if (metrics.rmx_rttvar) { 3327 tp->t_rttvar = metrics.rmx_rttvar; 3328 TCPSTAT_INC(tcps_usedrttvar); 3329 } else { 3330 /* default variation is +- 1 rtt */ 3331 tp->t_rttvar = 3332 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3333 } 3334 TCPT_RANGESET(tp->t_rxtcur, 3335 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3336 tp->t_rttmin, TCPTV_REXMTMAX); 3337 } 3338 if (metrics.rmx_ssthresh) { 3339 /* 3340 * There's some sort of gateway or interface 3341 * buffer limit on the path. Use this to set 3342 * the slow start threshhold, but set the 3343 * threshold to no less than 2*mss. 3344 */ 3345 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3346 TCPSTAT_INC(tcps_usedssthresh); 3347 } 3348 if (metrics.rmx_bandwidth) 3349 tp->snd_bandwidth = metrics.rmx_bandwidth; 3350 3351 /* 3352 * Set the slow-start flight size depending on whether this 3353 * is a local network or not. 3354 * 3355 * Extend this so we cache the cwnd too and retrieve it here. 3356 * Make cwnd even bigger than RFC3390 suggests but only if we 3357 * have previous experience with the remote host. Be careful 3358 * not make cwnd bigger than remote receive window or our own 3359 * send socket buffer. Maybe put some additional upper bound 3360 * on the retrieved cwnd. Should do incremental updates to 3361 * hostcache when cwnd collapses so next connection doesn't 3362 * overloads the path again. 3363 * 3364 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3365 * We currently check only in syncache_socket for that. 3366 */ 3367#define TCP_METRICS_CWND 3368#ifdef TCP_METRICS_CWND 3369 if (metrics.rmx_cwnd) 3370 tp->snd_cwnd = max(mss, 3371 min(metrics.rmx_cwnd / 2, 3372 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3373 else 3374#endif 3375 if (V_tcp_do_rfc3390) 3376 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3377#ifdef INET6 3378 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3379 (!isipv6 && in_localaddr(inp->inp_faddr))) 3380#else 3381 else if (in_localaddr(inp->inp_faddr)) 3382#endif 3383 tp->snd_cwnd = mss * V_ss_fltsz_local; 3384 else 3385 tp->snd_cwnd = mss * V_ss_fltsz; 3386 3387 /* Check the interface for TSO capabilities. */ 3388 if (mtuflags & CSUM_TSO) 3389 tp->t_flags |= TF_TSO; 3390} 3391 3392/* 3393 * Determine the MSS option to send on an outgoing SYN. 3394 */ 3395int 3396tcp_mssopt(struct in_conninfo *inc) 3397{ 3398 INIT_VNET_INET(curvnet); 3399 int mss = 0; 3400 u_long maxmtu = 0; 3401 u_long thcmtu = 0; 3402 size_t min_protoh; 3403 3404 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3405 3406#ifdef INET6 3407 if (inc->inc_flags & INC_ISIPV6) { 3408 mss = V_tcp_v6mssdflt; 3409 maxmtu = tcp_maxmtu6(inc, NULL); 3410 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3411 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3412 } else 3413#endif 3414 { 3415 mss = V_tcp_mssdflt; 3416 maxmtu = tcp_maxmtu(inc, NULL); 3417 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3418 min_protoh = sizeof(struct tcpiphdr); 3419 } 3420 if (maxmtu && thcmtu) 3421 mss = min(maxmtu, thcmtu) - min_protoh; 3422 else if (maxmtu || thcmtu) 3423 mss = max(maxmtu, thcmtu) - min_protoh; 3424 3425 return (mss); 3426} 3427 3428 3429/* 3430 * On a partial ack arrives, force the retransmission of the 3431 * next unacknowledged segment. Do not clear tp->t_dupacks. 3432 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3433 * be started again. 3434 */ 3435static void 3436tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3437{ 3438 tcp_seq onxt = tp->snd_nxt; 3439 u_long ocwnd = tp->snd_cwnd; 3440 3441 INP_WLOCK_ASSERT(tp->t_inpcb); 3442 3443 tcp_timer_activate(tp, TT_REXMT, 0); 3444 tp->t_rtttime = 0; 3445 tp->snd_nxt = th->th_ack; 3446 /* 3447 * Set snd_cwnd to one segment beyond acknowledged offset. 3448 * (tp->snd_una has not yet been updated when this function is called.) 3449 */ 3450 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3451 tp->t_flags |= TF_ACKNOW; 3452 (void) tcp_output(tp); 3453 tp->snd_cwnd = ocwnd; 3454 if (SEQ_GT(onxt, tp->snd_nxt)) 3455 tp->snd_nxt = onxt; 3456 /* 3457 * Partial window deflation. Relies on fact that tp->snd_una 3458 * not updated yet. 3459 */ 3460 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3461 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3462 else 3463 tp->snd_cwnd = 0; 3464 tp->snd_cwnd += tp->t_maxseg; 3465} 3466