1/*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 4. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50#include <sys/cdefs.h> 51__FBSDID("$FreeBSD$"); 52 53#include "opt_ipfw.h" /* for ipfw_fwd */ 54#include "opt_inet.h" 55#include "opt_inet6.h" 56#include "opt_ipsec.h" 57#include "opt_tcpdebug.h" 58 59#include <sys/param.h> 60#include <sys/kernel.h> 61#include <sys/hhook.h> 62#include <sys/malloc.h> 63#include <sys/mbuf.h> 64#include <sys/proc.h> /* for proc0 declaration */ 65#include <sys/protosw.h> 66#include <sys/signalvar.h> 67#include <sys/socket.h> 68#include <sys/socketvar.h> 69#include <sys/sysctl.h> 70#include <sys/syslog.h> 71#include <sys/systm.h> 72 73#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 74 75#include <vm/uma.h> 76 77#include <net/if.h> 78#include <net/route.h> 79#include <net/vnet.h> 80 81#define TCPSTATES /* for logging */ 82 83#include <netinet/cc.h> 84#include <netinet/in.h> 85#include <netinet/in_pcb.h> 86#include <netinet/in_systm.h> 87#include <netinet/in_var.h> 88#include <netinet/ip.h> 89#include <netinet/ip_icmp.h> /* required for icmp_var.h */ 90#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 91#include <netinet/ip_var.h> 92#include <netinet/ip_options.h> 93#include <netinet/ip6.h> 94#include <netinet/icmp6.h> 95#include <netinet6/in6_pcb.h> 96#include <netinet6/ip6_var.h> 97#include <netinet6/nd6.h> 98#include <netinet/tcp_fsm.h> 99#include <netinet/tcp_seq.h> 100#include <netinet/tcp_timer.h> 101#include <netinet/tcp_var.h> 102#include <netinet6/tcp6_var.h> 103#include <netinet/tcpip.h> 104#include <netinet/tcp_syncache.h> 105#ifdef TCPDEBUG 106#include <netinet/tcp_debug.h> 107#endif /* TCPDEBUG */ 108#ifdef TCP_OFFLOAD 109#include <netinet/tcp_offload.h> 110#endif 111 112#ifdef IPSEC 113#include <netipsec/ipsec.h> 114#include <netipsec/ipsec6.h> 115#endif /*IPSEC*/ 116 117#include <machine/in_cksum.h> 118 119#include <security/mac/mac_framework.h> 120 121const int tcprexmtthresh = 3; 122 123VNET_DEFINE(struct tcpstat, tcpstat); 124SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 125 &VNET_NAME(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, 131 "Log all incoming TCP segments to closed ports"); 132 133VNET_DEFINE(int, blackhole) = 0; 134#define V_blackhole VNET(blackhole) 135SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 136 &VNET_NAME(blackhole), 0, 137 "Do not send RST on segments to closed ports"); 138 139VNET_DEFINE(int, tcp_delack_enabled) = 1; 140SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 141 &VNET_NAME(tcp_delack_enabled), 0, 142 "Delay ACK to try and piggyback it onto a data packet"); 143 144VNET_DEFINE(int, drop_synfin) = 0; 145#define V_drop_synfin VNET(drop_synfin) 146SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 147 &VNET_NAME(drop_synfin), 0, 148 "Drop TCP packets with SYN+FIN set"); 149 150VNET_DEFINE(int, tcp_do_rfc3042) = 1; 151#define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 152SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 153 &VNET_NAME(tcp_do_rfc3042), 0, 154 "Enable RFC 3042 (Limited Transmit)"); 155 156VNET_DEFINE(int, tcp_do_rfc3390) = 1; 157SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 158 &VNET_NAME(tcp_do_rfc3390), 0, 159 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 160 161SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 162 "Experimental TCP extensions"); 163 164VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 165SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 166 &VNET_NAME(tcp_do_initcwnd10), 0, 167 "Enable RFC 6928 (Increasing initial CWND to 10)"); 168 169VNET_DEFINE(int, tcp_do_rfc3465) = 1; 170SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 171 &VNET_NAME(tcp_do_rfc3465), 0, 172 "Enable RFC 3465 (Appropriate Byte Counting)"); 173 174VNET_DEFINE(int, tcp_abc_l_var) = 2; 175SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 176 &VNET_NAME(tcp_abc_l_var), 2, 177 "Cap the max cwnd increment during slow-start to this number of segments"); 178 179static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 180 181VNET_DEFINE(int, tcp_do_ecn) = 0; 182SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 183 &VNET_NAME(tcp_do_ecn), 0, 184 "TCP ECN support"); 185 186VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 187SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 188 &VNET_NAME(tcp_ecn_maxretries), 0, 189 "Max retries before giving up on ECN"); 190 191VNET_DEFINE(int, tcp_insecure_rst) = 0; 192#define V_tcp_insecure_rst VNET(tcp_insecure_rst) 193SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 194 &VNET_NAME(tcp_insecure_rst), 0, 195 "Follow the old (insecure) criteria for accepting RST packets"); 196 197VNET_DEFINE(int, tcp_recvspace) = 1024*64; 198#define V_tcp_recvspace VNET(tcp_recvspace) 199SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 200 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 201 202VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 203#define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 204SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 205 &VNET_NAME(tcp_do_autorcvbuf), 0, 206 "Enable automatic receive buffer sizing"); 207 208VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 209#define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 210SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 211 &VNET_NAME(tcp_autorcvbuf_inc), 0, 212 "Incrementor step size of automatic receive buffer"); 213 214VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 215#define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 216SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 217 &VNET_NAME(tcp_autorcvbuf_max), 0, 218 "Max size of automatic receive buffer"); 219 220VNET_DEFINE(struct inpcbhead, tcb); 221#define tcb6 tcb /* for KAME src sync over BSD*'s */ 222VNET_DEFINE(struct inpcbinfo, tcbinfo); 223 224static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 225static void tcp_do_segment(struct mbuf *, struct tcphdr *, 226 struct socket *, struct tcpcb *, int, int, uint8_t, 227 int); 228static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 229 struct tcpcb *, int, int); 230static void tcp_pulloutofband(struct socket *, 231 struct tcphdr *, struct mbuf *, int); 232static void tcp_xmit_timer(struct tcpcb *, int); 233static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 234static void inline tcp_fields_to_host(struct tcphdr *); 235#ifdef TCP_SIGNATURE 236static void inline tcp_fields_to_net(struct tcphdr *); 237static int inline tcp_signature_verify_input(struct mbuf *, int, int, 238 int, struct tcpopt *, struct tcphdr *, u_int); 239#endif 240static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 241 uint16_t type); 242static void inline cc_conn_init(struct tcpcb *tp); 243static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 244static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 245 struct tcphdr *th, struct tcpopt *to); 246 247/* 248 * Kernel module interface for updating tcpstat. The argument is an index 249 * into tcpstat treated as an array of u_long. While this encodes the 250 * general layout of tcpstat into the caller, it doesn't encode its location, 251 * so that future changes to add, for example, per-CPU stats support won't 252 * cause binary compatibility problems for kernel modules. 253 */ 254void 255kmod_tcpstat_inc(int statnum) 256{ 257 258 (*((u_long *)&V_tcpstat + statnum))++; 259} 260 261/* 262 * Wrapper for the TCP established input helper hook. 263 */ 264static void inline 265hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 266{ 267 struct tcp_hhook_data hhook_data; 268 269 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 270 hhook_data.tp = tp; 271 hhook_data.th = th; 272 hhook_data.to = to; 273 274 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 275 tp->osd); 276 } 277} 278 279/* 280 * CC wrapper hook functions 281 */ 282static void inline 283cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 284{ 285 INP_WLOCK_ASSERT(tp->t_inpcb); 286 287 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 288 if (tp->snd_cwnd <= tp->snd_wnd) 289 tp->ccv->flags |= CCF_CWND_LIMITED; 290 else 291 tp->ccv->flags &= ~CCF_CWND_LIMITED; 292 293 if (type == CC_ACK) { 294 if (tp->snd_cwnd > tp->snd_ssthresh) { 295 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 296 V_tcp_abc_l_var * tp->t_maxseg); 297 if (tp->t_bytes_acked >= tp->snd_cwnd) { 298 tp->t_bytes_acked -= tp->snd_cwnd; 299 tp->ccv->flags |= CCF_ABC_SENTAWND; 300 } 301 } else { 302 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 303 tp->t_bytes_acked = 0; 304 } 305 } 306 307 if (CC_ALGO(tp)->ack_received != NULL) { 308 /* XXXLAS: Find a way to live without this */ 309 tp->ccv->curack = th->th_ack; 310 CC_ALGO(tp)->ack_received(tp->ccv, type); 311 } 312} 313 314static void inline 315cc_conn_init(struct tcpcb *tp) 316{ 317 struct hc_metrics_lite metrics; 318 struct inpcb *inp = tp->t_inpcb; 319 int rtt; 320 321 INP_WLOCK_ASSERT(tp->t_inpcb); 322 323 tcp_hc_get(&inp->inp_inc, &metrics); 324 325 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 326 tp->t_srtt = rtt; 327 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 328 TCPSTAT_INC(tcps_usedrtt); 329 if (metrics.rmx_rttvar) { 330 tp->t_rttvar = metrics.rmx_rttvar; 331 TCPSTAT_INC(tcps_usedrttvar); 332 } else { 333 /* default variation is +- 1 rtt */ 334 tp->t_rttvar = 335 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 336 } 337 TCPT_RANGESET(tp->t_rxtcur, 338 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 339 tp->t_rttmin, TCPTV_REXMTMAX); 340 } 341 if (metrics.rmx_ssthresh) { 342 /* 343 * There's some sort of gateway or interface 344 * buffer limit on the path. Use this to set 345 * the slow start threshhold, but set the 346 * threshold to no less than 2*mss. 347 */ 348 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 349 TCPSTAT_INC(tcps_usedssthresh); 350 } 351 352 /* 353 * Set the initial slow-start flight size. 354 * 355 * RFC5681 Section 3.1 specifies the default conservative values. 356 * RFC3390 specifies slightly more aggressive values. 357 * RFC6928 increases it to ten segments. 358 * 359 * If a SYN or SYN/ACK was lost and retransmitted, we have to 360 * reduce the initial CWND to one segment as congestion is likely 361 * requiring us to be cautious. 362 */ 363 if (tp->snd_cwnd == 1) 364 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 365 else if (V_tcp_do_initcwnd10) 366 tp->snd_cwnd = min(10 * tp->t_maxseg, 367 max(2 * tp->t_maxseg, 14600)); 368 else if (V_tcp_do_rfc3390) 369 tp->snd_cwnd = min(4 * tp->t_maxseg, 370 max(2 * tp->t_maxseg, 4380)); 371 else { 372 /* Per RFC5681 Section 3.1 */ 373 if (tp->t_maxseg > 2190) 374 tp->snd_cwnd = 2 * tp->t_maxseg; 375 else if (tp->t_maxseg > 1095) 376 tp->snd_cwnd = 3 * tp->t_maxseg; 377 else 378 tp->snd_cwnd = 4 * tp->t_maxseg; 379 } 380 381 if (CC_ALGO(tp)->conn_init != NULL) 382 CC_ALGO(tp)->conn_init(tp->ccv); 383} 384 385void inline 386cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 387{ 388 INP_WLOCK_ASSERT(tp->t_inpcb); 389 390 switch(type) { 391 case CC_NDUPACK: 392 if (!IN_FASTRECOVERY(tp->t_flags)) { 393 tp->snd_recover = tp->snd_max; 394 if (tp->t_flags & TF_ECN_PERMIT) 395 tp->t_flags |= TF_ECN_SND_CWR; 396 } 397 break; 398 case CC_ECN: 399 if (!IN_CONGRECOVERY(tp->t_flags)) { 400 TCPSTAT_INC(tcps_ecn_rcwnd); 401 tp->snd_recover = tp->snd_max; 402 if (tp->t_flags & TF_ECN_PERMIT) 403 tp->t_flags |= TF_ECN_SND_CWR; 404 } 405 break; 406 case CC_RTO: 407 tp->t_dupacks = 0; 408 tp->t_bytes_acked = 0; 409 EXIT_RECOVERY(tp->t_flags); 410 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 411 tp->t_maxseg) * tp->t_maxseg; 412 tp->snd_cwnd = tp->t_maxseg; 413 break; 414 case CC_RTO_ERR: 415 TCPSTAT_INC(tcps_sndrexmitbad); 416 /* RTO was unnecessary, so reset everything. */ 417 tp->snd_cwnd = tp->snd_cwnd_prev; 418 tp->snd_ssthresh = tp->snd_ssthresh_prev; 419 tp->snd_recover = tp->snd_recover_prev; 420 if (tp->t_flags & TF_WASFRECOVERY) 421 ENTER_FASTRECOVERY(tp->t_flags); 422 if (tp->t_flags & TF_WASCRECOVERY) 423 ENTER_CONGRECOVERY(tp->t_flags); 424 tp->snd_nxt = tp->snd_max; 425 tp->t_flags &= ~TF_PREVVALID; 426 tp->t_badrxtwin = 0; 427 break; 428 } 429 430 if (CC_ALGO(tp)->cong_signal != NULL) { 431 if (th != NULL) 432 tp->ccv->curack = th->th_ack; 433 CC_ALGO(tp)->cong_signal(tp->ccv, type); 434 } 435} 436 437static void inline 438cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 439{ 440 INP_WLOCK_ASSERT(tp->t_inpcb); 441 442 /* XXXLAS: KASSERT that we're in recovery? */ 443 444 if (CC_ALGO(tp)->post_recovery != NULL) { 445 tp->ccv->curack = th->th_ack; 446 CC_ALGO(tp)->post_recovery(tp->ccv); 447 } 448 /* XXXLAS: EXIT_RECOVERY ? */ 449 tp->t_bytes_acked = 0; 450} 451 452static inline void 453tcp_fields_to_host(struct tcphdr *th) 454{ 455 456 th->th_seq = ntohl(th->th_seq); 457 th->th_ack = ntohl(th->th_ack); 458 th->th_win = ntohs(th->th_win); 459 th->th_urp = ntohs(th->th_urp); 460} 461 462#ifdef TCP_SIGNATURE 463static inline void 464tcp_fields_to_net(struct tcphdr *th) 465{ 466 467 th->th_seq = htonl(th->th_seq); 468 th->th_ack = htonl(th->th_ack); 469 th->th_win = htons(th->th_win); 470 th->th_urp = htons(th->th_urp); 471} 472 473static inline int 474tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 475 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 476{ 477 int ret; 478 479 tcp_fields_to_net(th); 480 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 481 tcp_fields_to_host(th); 482 return (ret); 483} 484#endif 485 486/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 487#ifdef INET6 488#define ND6_HINT(tp) \ 489do { \ 490 if ((tp) && (tp)->t_inpcb && \ 491 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 492 nd6_nud_hint(NULL, NULL, 0); \ 493} while (0) 494#else 495#define ND6_HINT(tp) 496#endif 497 498/* 499 * Indicate whether this ack should be delayed. We can delay the ack if 500 * - there is no delayed ack timer in progress and 501 * - our last ack wasn't a 0-sized window. We never want to delay 502 * the ack that opens up a 0-sized window and 503 * - delayed acks are enabled or 504 * - this is a half-synchronized T/TCP connection. 505 */ 506#define DELAY_ACK(tp) \ 507 ((!tcp_timer_active(tp, TT_DELACK) && \ 508 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 509 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 510 511/* 512 * TCP input handling is split into multiple parts: 513 * tcp6_input is a thin wrapper around tcp_input for the extended 514 * ip6_protox[] call format in ip6_input 515 * tcp_input handles primary segment validation, inpcb lookup and 516 * SYN processing on listen sockets 517 * tcp_do_segment processes the ACK and text of the segment for 518 * establishing, established and closing connections 519 */ 520#ifdef INET6 521int 522tcp6_input(struct mbuf **mp, int *offp, int proto) 523{ 524 struct mbuf *m = *mp; 525 struct in6_ifaddr *ia6; 526 527 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 528 529 /* 530 * draft-itojun-ipv6-tcp-to-anycast 531 * better place to put this in? 532 */ 533 ia6 = ip6_getdstifaddr(m); 534 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 535 struct ip6_hdr *ip6; 536 537 ifa_free(&ia6->ia_ifa); 538 ip6 = mtod(m, struct ip6_hdr *); 539 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 540 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 541 return IPPROTO_DONE; 542 } 543 if (ia6) 544 ifa_free(&ia6->ia_ifa); 545 546 tcp_input(m, *offp); 547 return IPPROTO_DONE; 548} 549#endif /* INET6 */ 550 551void 552tcp_input(struct mbuf *m, int off0) 553{ 554 struct tcphdr *th = NULL; 555 struct ip *ip = NULL; 556#ifdef INET 557 struct ipovly *ipov; 558#endif 559 struct inpcb *inp = NULL; 560 struct tcpcb *tp = NULL; 561 struct socket *so = NULL; 562 u_char *optp = NULL; 563 int optlen = 0; 564#ifdef INET 565 int len; 566#endif 567 int tlen = 0, off; 568 int drop_hdrlen; 569 int thflags; 570 int rstreason = 0; /* For badport_bandlim accounting purposes */ 571#ifdef TCP_SIGNATURE 572 uint8_t sig_checked = 0; 573#endif 574 uint8_t iptos = 0; 575 struct m_tag *fwd_tag = NULL; 576#ifdef INET6 577 struct ip6_hdr *ip6 = NULL; 578 int isipv6; 579#else 580 const void *ip6 = NULL; 581#endif /* INET6 */ 582 struct tcpopt to; /* options in this segment */ 583 char *s = NULL; /* address and port logging */ 584 int ti_locked; 585#define TI_UNLOCKED 1 586#define TI_WLOCKED 2 587 588#ifdef TCPDEBUG 589 /* 590 * The size of tcp_saveipgen must be the size of the max ip header, 591 * now IPv6. 592 */ 593 u_char tcp_saveipgen[IP6_HDR_LEN]; 594 struct tcphdr tcp_savetcp; 595 short ostate = 0; 596#endif 597 598#ifdef INET6 599 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 600#endif 601 602 to.to_flags = 0; 603 TCPSTAT_INC(tcps_rcvtotal); 604 605#ifdef INET6 606 if (isipv6) { 607 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 608 609 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 610 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 611 if (m == NULL) { 612 TCPSTAT_INC(tcps_rcvshort); 613 return; 614 } 615 } 616 617 ip6 = mtod(m, struct ip6_hdr *); 618 th = (struct tcphdr *)((caddr_t)ip6 + off0); 619 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 620 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 621 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 622 th->th_sum = m->m_pkthdr.csum_data; 623 else 624 th->th_sum = in6_cksum_pseudo(ip6, tlen, 625 IPPROTO_TCP, m->m_pkthdr.csum_data); 626 th->th_sum ^= 0xffff; 627 } else 628 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 629 if (th->th_sum) { 630 TCPSTAT_INC(tcps_rcvbadsum); 631 goto drop; 632 } 633 634 /* 635 * Be proactive about unspecified IPv6 address in source. 636 * As we use all-zero to indicate unbounded/unconnected pcb, 637 * unspecified IPv6 address can be used to confuse us. 638 * 639 * Note that packets with unspecified IPv6 destination is 640 * already dropped in ip6_input. 641 */ 642 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 643 /* XXX stat */ 644 goto drop; 645 } 646 } 647#endif 648#if defined(INET) && defined(INET6) 649 else 650#endif 651#ifdef INET 652 { 653 /* 654 * Get IP and TCP header together in first mbuf. 655 * Note: IP leaves IP header in first mbuf. 656 */ 657 if (off0 > sizeof (struct ip)) { 658 ip_stripoptions(m, (struct mbuf *)0); 659 off0 = sizeof(struct ip); 660 } 661 if (m->m_len < sizeof (struct tcpiphdr)) { 662 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 663 == NULL) { 664 TCPSTAT_INC(tcps_rcvshort); 665 return; 666 } 667 } 668 ip = mtod(m, struct ip *); 669 ipov = (struct ipovly *)ip; 670 th = (struct tcphdr *)((caddr_t)ip + off0); 671 tlen = ip->ip_len; 672 673 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 674 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 675 th->th_sum = m->m_pkthdr.csum_data; 676 else 677 th->th_sum = in_pseudo(ip->ip_src.s_addr, 678 ip->ip_dst.s_addr, 679 htonl(m->m_pkthdr.csum_data + 680 ip->ip_len + 681 IPPROTO_TCP)); 682 th->th_sum ^= 0xffff; 683#ifdef TCPDEBUG 684 ipov->ih_len = (u_short)tlen; 685 ipov->ih_len = htons(ipov->ih_len); 686#endif 687 } else { 688 /* 689 * Checksum extended TCP header and data. 690 */ 691 len = sizeof (struct ip) + tlen; 692 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 693 ipov->ih_len = (u_short)tlen; 694 ipov->ih_len = htons(ipov->ih_len); 695 th->th_sum = in_cksum(m, len); 696 } 697 if (th->th_sum) { 698 TCPSTAT_INC(tcps_rcvbadsum); 699 goto drop; 700 } 701 /* Re-initialization for later version check */ 702 ip->ip_v = IPVERSION; 703 } 704#endif /* INET */ 705 706#ifdef INET6 707 if (isipv6) 708 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 709#endif 710#if defined(INET) && defined(INET6) 711 else 712#endif 713#ifdef INET 714 iptos = ip->ip_tos; 715#endif 716 717 /* 718 * Check that TCP offset makes sense, 719 * pull out TCP options and adjust length. XXX 720 */ 721 off = th->th_off << 2; 722 if (off < sizeof (struct tcphdr) || off > tlen) { 723 TCPSTAT_INC(tcps_rcvbadoff); 724 goto drop; 725 } 726 tlen -= off; /* tlen is used instead of ti->ti_len */ 727 if (off > sizeof (struct tcphdr)) { 728#ifdef INET6 729 if (isipv6) { 730 IP6_EXTHDR_CHECK(m, off0, off, ); 731 ip6 = mtod(m, struct ip6_hdr *); 732 th = (struct tcphdr *)((caddr_t)ip6 + off0); 733 } 734#endif 735#if defined(INET) && defined(INET6) 736 else 737#endif 738#ifdef INET 739 { 740 if (m->m_len < sizeof(struct ip) + off) { 741 if ((m = m_pullup(m, sizeof (struct ip) + off)) 742 == NULL) { 743 TCPSTAT_INC(tcps_rcvshort); 744 return; 745 } 746 ip = mtod(m, struct ip *); 747 ipov = (struct ipovly *)ip; 748 th = (struct tcphdr *)((caddr_t)ip + off0); 749 } 750 } 751#endif 752 optlen = off - sizeof (struct tcphdr); 753 optp = (u_char *)(th + 1); 754 } 755 thflags = th->th_flags; 756 757 /* 758 * Convert TCP protocol specific fields to host format. 759 */ 760 tcp_fields_to_host(th); 761 762 /* 763 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 764 */ 765 drop_hdrlen = off0 + off; 766 767 /* 768 * Locate pcb for segment; if we're likely to add or remove a 769 * connection then first acquire pcbinfo lock. There are two cases 770 * where we might discover later we need a write lock despite the 771 * flags: ACKs moving a connection out of the syncache, and ACKs for 772 * a connection in TIMEWAIT. 773 */ 774 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 775 INP_INFO_WLOCK(&V_tcbinfo); 776 ti_locked = TI_WLOCKED; 777 } else 778 ti_locked = TI_UNLOCKED; 779 780findpcb: 781#ifdef INVARIANTS 782 if (ti_locked == TI_WLOCKED) { 783 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 784 } else { 785 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 786 } 787#endif 788 789 /* 790 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 791 */ 792 if ( 793#ifdef INET6 794 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 795#ifdef INET 796 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 797#endif 798#endif 799#if defined(INET) && !defined(INET6) 800 (m->m_flags & M_IP_NEXTHOP) 801#endif 802 ) 803 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 804 805#ifdef INET6 806 if (isipv6 && fwd_tag != NULL) { 807 struct sockaddr_in6 *next_hop6; 808 809 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 810 /* 811 * Transparently forwarded. Pretend to be the destination. 812 * Already got one like this? 813 */ 814 inp = in6_pcblookup_mbuf(&V_tcbinfo, 815 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 816 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 817 if (!inp) { 818 /* 819 * It's new. Try to find the ambushing socket. 820 * Because we've rewritten the destination address, 821 * any hardware-generated hash is ignored. 822 */ 823 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 824 th->th_sport, &next_hop6->sin6_addr, 825 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 826 th->th_dport, INPLOOKUP_WILDCARD | 827 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 828 } 829 /* Remove the tag from the packet. We don't need it anymore. */ 830 m_tag_delete(m, fwd_tag); 831 m->m_flags &= ~M_IP6_NEXTHOP; 832 fwd_tag = NULL; 833 } else if (isipv6) { 834 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 835 th->th_sport, &ip6->ip6_dst, th->th_dport, 836 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 837 m->m_pkthdr.rcvif, m); 838 } 839#endif /* INET6 */ 840#if defined(INET6) && defined(INET) 841 else 842#endif 843#ifdef INET 844 if (fwd_tag != NULL) { 845 struct sockaddr_in *next_hop; 846 847 next_hop = (struct sockaddr_in *)(fwd_tag+1); 848 /* 849 * Transparently forwarded. Pretend to be the destination. 850 * already got one like this? 851 */ 852 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 853 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 854 m->m_pkthdr.rcvif, m); 855 if (!inp) { 856 /* 857 * It's new. Try to find the ambushing socket. 858 * Because we've rewritten the destination address, 859 * any hardware-generated hash is ignored. 860 */ 861 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 862 th->th_sport, next_hop->sin_addr, 863 next_hop->sin_port ? ntohs(next_hop->sin_port) : 864 th->th_dport, INPLOOKUP_WILDCARD | 865 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 866 } 867 /* Remove the tag from the packet. We don't need it anymore. */ 868 m_tag_delete(m, fwd_tag); 869 m->m_flags &= ~M_IP_NEXTHOP; 870 fwd_tag = NULL; 871 } else 872 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 873 th->th_sport, ip->ip_dst, th->th_dport, 874 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 875 m->m_pkthdr.rcvif, m); 876#endif /* INET */ 877 878 /* 879 * If the INPCB does not exist then all data in the incoming 880 * segment is discarded and an appropriate RST is sent back. 881 * XXX MRT Send RST using which routing table? 882 */ 883 if (inp == NULL) { 884 /* 885 * Log communication attempts to ports that are not 886 * in use. 887 */ 888 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 889 tcp_log_in_vain == 2) { 890 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 891 log(LOG_INFO, "%s; %s: Connection attempt " 892 "to closed port\n", s, __func__); 893 } 894 /* 895 * When blackholing do not respond with a RST but 896 * completely ignore the segment and drop it. 897 */ 898 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 899 V_blackhole == 2) 900 goto dropunlock; 901 902 rstreason = BANDLIM_RST_CLOSEDPORT; 903 goto dropwithreset; 904 } 905 INP_WLOCK_ASSERT(inp); 906 if (!(inp->inp_flags & INP_HW_FLOWID) 907 && (m->m_flags & M_FLOWID) 908 && ((inp->inp_socket == NULL) 909 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 910 inp->inp_flags |= INP_HW_FLOWID; 911 inp->inp_flags &= ~INP_SW_FLOWID; 912 inp->inp_flowid = m->m_pkthdr.flowid; 913 } 914#ifdef IPSEC 915#ifdef INET6 916 if (isipv6 && ipsec6_in_reject(m, inp)) { 917 IPSEC6STAT_INC(in_polvio); 918 goto dropunlock; 919 } else 920#endif /* INET6 */ 921 if (ipsec4_in_reject(m, inp) != 0) { 922 IPSECSTAT_INC(in_polvio); 923 goto dropunlock; 924 } 925#endif /* IPSEC */ 926 927 /* 928 * Check the minimum TTL for socket. 929 */ 930 if (inp->inp_ip_minttl != 0) { 931#ifdef INET6 932 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 933 goto dropunlock; 934 else 935#endif 936 if (inp->inp_ip_minttl > ip->ip_ttl) 937 goto dropunlock; 938 } 939 940 /* 941 * A previous connection in TIMEWAIT state is supposed to catch stray 942 * or duplicate segments arriving late. If this segment was a 943 * legitimate new connection attempt the old INPCB gets removed and 944 * we can try again to find a listening socket. 945 * 946 * At this point, due to earlier optimism, we may hold only an inpcb 947 * lock, and not the inpcbinfo write lock. If so, we need to try to 948 * acquire it, or if that fails, acquire a reference on the inpcb, 949 * drop all locks, acquire a global write lock, and then re-acquire 950 * the inpcb lock. We may at that point discover that another thread 951 * has tried to free the inpcb, in which case we need to loop back 952 * and try to find a new inpcb to deliver to. 953 * 954 * XXXRW: It may be time to rethink timewait locking. 955 */ 956relocked: 957 if (inp->inp_flags & INP_TIMEWAIT) { 958 if (ti_locked == TI_UNLOCKED) { 959 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 960 in_pcbref(inp); 961 INP_WUNLOCK(inp); 962 INP_INFO_WLOCK(&V_tcbinfo); 963 ti_locked = TI_WLOCKED; 964 INP_WLOCK(inp); 965 if (in_pcbrele_wlocked(inp)) { 966 inp = NULL; 967 goto findpcb; 968 } 969 } else 970 ti_locked = TI_WLOCKED; 971 } 972 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 973 974 if (thflags & TH_SYN) 975 tcp_dooptions(&to, optp, optlen, TO_SYN); 976 /* 977 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 978 */ 979 if (tcp_twcheck(inp, &to, th, m, tlen)) 980 goto findpcb; 981 INP_INFO_WUNLOCK(&V_tcbinfo); 982 return; 983 } 984 /* 985 * The TCPCB may no longer exist if the connection is winding 986 * down or it is in the CLOSED state. Either way we drop the 987 * segment and send an appropriate response. 988 */ 989 tp = intotcpcb(inp); 990 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 991 rstreason = BANDLIM_RST_CLOSEDPORT; 992 goto dropwithreset; 993 } 994 995#ifdef TCP_OFFLOAD 996 if (tp->t_flags & TF_TOE) { 997 tcp_offload_input(tp, m); 998 m = NULL; /* consumed by the TOE driver */ 999 goto dropunlock; 1000 } 1001#endif 1002 1003 /* 1004 * We've identified a valid inpcb, but it could be that we need an 1005 * inpcbinfo write lock but don't hold it. In this case, attempt to 1006 * acquire using the same strategy as the TIMEWAIT case above. If we 1007 * relock, we have to jump back to 'relocked' as the connection might 1008 * now be in TIMEWAIT. 1009 */ 1010#ifdef INVARIANTS 1011 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 1012 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1013#endif 1014 if (tp->t_state != TCPS_ESTABLISHED) { 1015 if (ti_locked == TI_UNLOCKED) { 1016 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 1017 in_pcbref(inp); 1018 INP_WUNLOCK(inp); 1019 INP_INFO_WLOCK(&V_tcbinfo); 1020 ti_locked = TI_WLOCKED; 1021 INP_WLOCK(inp); 1022 if (in_pcbrele_wlocked(inp)) { 1023 inp = NULL; 1024 goto findpcb; 1025 } 1026 goto relocked; 1027 } else 1028 ti_locked = TI_WLOCKED; 1029 } 1030 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1031 } 1032 1033#ifdef MAC 1034 INP_WLOCK_ASSERT(inp); 1035 if (mac_inpcb_check_deliver(inp, m)) 1036 goto dropunlock; 1037#endif 1038 so = inp->inp_socket; 1039 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1040#ifdef TCPDEBUG 1041 if (so->so_options & SO_DEBUG) { 1042 ostate = tp->t_state; 1043#ifdef INET6 1044 if (isipv6) { 1045 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1046 } else 1047#endif 1048 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1049 tcp_savetcp = *th; 1050 } 1051#endif /* TCPDEBUG */ 1052 /* 1053 * When the socket is accepting connections (the INPCB is in LISTEN 1054 * state) we look into the SYN cache if this is a new connection 1055 * attempt or the completion of a previous one. Because listen 1056 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1057 * held in this case. 1058 */ 1059 if (so->so_options & SO_ACCEPTCONN) { 1060 struct in_conninfo inc; 1061 1062 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1063 "tp not listening", __func__)); 1064 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1065 1066 bzero(&inc, sizeof(inc)); 1067#ifdef INET6 1068 if (isipv6) { 1069 inc.inc_flags |= INC_ISIPV6; 1070 inc.inc6_faddr = ip6->ip6_src; 1071 inc.inc6_laddr = ip6->ip6_dst; 1072 } else 1073#endif 1074 { 1075 inc.inc_faddr = ip->ip_src; 1076 inc.inc_laddr = ip->ip_dst; 1077 } 1078 inc.inc_fport = th->th_sport; 1079 inc.inc_lport = th->th_dport; 1080 inc.inc_fibnum = so->so_fibnum; 1081 1082 /* 1083 * Check for an existing connection attempt in syncache if 1084 * the flag is only ACK. A successful lookup creates a new 1085 * socket appended to the listen queue in SYN_RECEIVED state. 1086 */ 1087 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1088 /* 1089 * Parse the TCP options here because 1090 * syncookies need access to the reflected 1091 * timestamp. 1092 */ 1093 tcp_dooptions(&to, optp, optlen, 0); 1094 /* 1095 * NB: syncache_expand() doesn't unlock 1096 * inp and tcpinfo locks. 1097 */ 1098 if (!syncache_expand(&inc, &to, th, &so, m)) { 1099 /* 1100 * No syncache entry or ACK was not 1101 * for our SYN/ACK. Send a RST. 1102 * NB: syncache did its own logging 1103 * of the failure cause. 1104 */ 1105 rstreason = BANDLIM_RST_OPENPORT; 1106 goto dropwithreset; 1107 } 1108 if (so == NULL) { 1109 /* 1110 * We completed the 3-way handshake 1111 * but could not allocate a socket 1112 * either due to memory shortage, 1113 * listen queue length limits or 1114 * global socket limits. Send RST 1115 * or wait and have the remote end 1116 * retransmit the ACK for another 1117 * try. 1118 */ 1119 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1120 log(LOG_DEBUG, "%s; %s: Listen socket: " 1121 "Socket allocation failed due to " 1122 "limits or memory shortage, %s\n", 1123 s, __func__, 1124 V_tcp_sc_rst_sock_fail ? 1125 "sending RST" : "try again"); 1126 if (V_tcp_sc_rst_sock_fail) { 1127 rstreason = BANDLIM_UNLIMITED; 1128 goto dropwithreset; 1129 } else 1130 goto dropunlock; 1131 } 1132 /* 1133 * Socket is created in state SYN_RECEIVED. 1134 * Unlock the listen socket, lock the newly 1135 * created socket and update the tp variable. 1136 */ 1137 INP_WUNLOCK(inp); /* listen socket */ 1138 inp = sotoinpcb(so); 1139 INP_WLOCK(inp); /* new connection */ 1140 tp = intotcpcb(inp); 1141 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1142 ("%s: ", __func__)); 1143#ifdef TCP_SIGNATURE 1144 if (sig_checked == 0) { 1145 tcp_dooptions(&to, optp, optlen, 1146 (thflags & TH_SYN) ? TO_SYN : 0); 1147 if (!tcp_signature_verify_input(m, off0, tlen, 1148 optlen, &to, th, tp->t_flags)) { 1149 1150 /* 1151 * In SYN_SENT state if it receives an 1152 * RST, it is allowed for further 1153 * processing. 1154 */ 1155 if ((thflags & TH_RST) == 0 || 1156 (tp->t_state == TCPS_SYN_SENT) == 0) 1157 goto dropunlock; 1158 } 1159 sig_checked = 1; 1160 } 1161#endif 1162 1163 /* 1164 * Process the segment and the data it 1165 * contains. tcp_do_segment() consumes 1166 * the mbuf chain and unlocks the inpcb. 1167 */ 1168 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1169 iptos, ti_locked); 1170 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1171 return; 1172 } 1173 /* 1174 * Segment flag validation for new connection attempts: 1175 * 1176 * Our (SYN|ACK) response was rejected. 1177 * Check with syncache and remove entry to prevent 1178 * retransmits. 1179 * 1180 * NB: syncache_chkrst does its own logging of failure 1181 * causes. 1182 */ 1183 if (thflags & TH_RST) { 1184 syncache_chkrst(&inc, th); 1185 goto dropunlock; 1186 } 1187 /* 1188 * We can't do anything without SYN. 1189 */ 1190 if ((thflags & TH_SYN) == 0) { 1191 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1192 log(LOG_DEBUG, "%s; %s: Listen socket: " 1193 "SYN is missing, segment ignored\n", 1194 s, __func__); 1195 TCPSTAT_INC(tcps_badsyn); 1196 goto dropunlock; 1197 } 1198 /* 1199 * (SYN|ACK) is bogus on a listen socket. 1200 */ 1201 if (thflags & TH_ACK) { 1202 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1203 log(LOG_DEBUG, "%s; %s: Listen socket: " 1204 "SYN|ACK invalid, segment rejected\n", 1205 s, __func__); 1206 syncache_badack(&inc); /* XXX: Not needed! */ 1207 TCPSTAT_INC(tcps_badsyn); 1208 rstreason = BANDLIM_RST_OPENPORT; 1209 goto dropwithreset; 1210 } 1211 /* 1212 * If the drop_synfin option is enabled, drop all 1213 * segments with both the SYN and FIN bits set. 1214 * This prevents e.g. nmap from identifying the 1215 * TCP/IP stack. 1216 * XXX: Poor reasoning. nmap has other methods 1217 * and is constantly refining its stack detection 1218 * strategies. 1219 * XXX: This is a violation of the TCP specification 1220 * and was used by RFC1644. 1221 */ 1222 if ((thflags & TH_FIN) && V_drop_synfin) { 1223 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1224 log(LOG_DEBUG, "%s; %s: Listen socket: " 1225 "SYN|FIN segment ignored (based on " 1226 "sysctl setting)\n", s, __func__); 1227 TCPSTAT_INC(tcps_badsyn); 1228 goto dropunlock; 1229 } 1230 /* 1231 * Segment's flags are (SYN) or (SYN|FIN). 1232 * 1233 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1234 * as they do not affect the state of the TCP FSM. 1235 * The data pointed to by TH_URG and th_urp is ignored. 1236 */ 1237 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1238 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1239 KASSERT(thflags & (TH_SYN), 1240 ("%s: Listen socket: TH_SYN not set", __func__)); 1241#ifdef INET6 1242 /* 1243 * If deprecated address is forbidden, 1244 * we do not accept SYN to deprecated interface 1245 * address to prevent any new inbound connection from 1246 * getting established. 1247 * When we do not accept SYN, we send a TCP RST, 1248 * with deprecated source address (instead of dropping 1249 * it). We compromise it as it is much better for peer 1250 * to send a RST, and RST will be the final packet 1251 * for the exchange. 1252 * 1253 * If we do not forbid deprecated addresses, we accept 1254 * the SYN packet. RFC2462 does not suggest dropping 1255 * SYN in this case. 1256 * If we decipher RFC2462 5.5.4, it says like this: 1257 * 1. use of deprecated addr with existing 1258 * communication is okay - "SHOULD continue to be 1259 * used" 1260 * 2. use of it with new communication: 1261 * (2a) "SHOULD NOT be used if alternate address 1262 * with sufficient scope is available" 1263 * (2b) nothing mentioned otherwise. 1264 * Here we fall into (2b) case as we have no choice in 1265 * our source address selection - we must obey the peer. 1266 * 1267 * The wording in RFC2462 is confusing, and there are 1268 * multiple description text for deprecated address 1269 * handling - worse, they are not exactly the same. 1270 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1271 */ 1272 if (isipv6 && !V_ip6_use_deprecated) { 1273 struct in6_ifaddr *ia6; 1274 1275 ia6 = ip6_getdstifaddr(m); 1276 if (ia6 != NULL && 1277 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1278 ifa_free(&ia6->ia_ifa); 1279 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1280 log(LOG_DEBUG, "%s; %s: Listen socket: " 1281 "Connection attempt to deprecated " 1282 "IPv6 address rejected\n", 1283 s, __func__); 1284 rstreason = BANDLIM_RST_OPENPORT; 1285 goto dropwithreset; 1286 } 1287 if (ia6) 1288 ifa_free(&ia6->ia_ifa); 1289 } 1290#endif /* INET6 */ 1291 /* 1292 * Basic sanity checks on incoming SYN requests: 1293 * Don't respond if the destination is a link layer 1294 * broadcast according to RFC1122 4.2.3.10, p. 104. 1295 * If it is from this socket it must be forged. 1296 * Don't respond if the source or destination is a 1297 * global or subnet broad- or multicast address. 1298 * Note that it is quite possible to receive unicast 1299 * link-layer packets with a broadcast IP address. Use 1300 * in_broadcast() to find them. 1301 */ 1302 if (m->m_flags & (M_BCAST|M_MCAST)) { 1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1304 log(LOG_DEBUG, "%s; %s: Listen socket: " 1305 "Connection attempt from broad- or multicast " 1306 "link layer address ignored\n", s, __func__); 1307 goto dropunlock; 1308 } 1309#ifdef INET6 1310 if (isipv6) { 1311 if (th->th_dport == th->th_sport && 1312 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1313 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1314 log(LOG_DEBUG, "%s; %s: Listen socket: " 1315 "Connection attempt to/from self " 1316 "ignored\n", s, __func__); 1317 goto dropunlock; 1318 } 1319 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1320 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1321 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1322 log(LOG_DEBUG, "%s; %s: Listen socket: " 1323 "Connection attempt from/to multicast " 1324 "address ignored\n", s, __func__); 1325 goto dropunlock; 1326 } 1327 } 1328#endif 1329#if defined(INET) && defined(INET6) 1330 else 1331#endif 1332#ifdef INET 1333 { 1334 if (th->th_dport == th->th_sport && 1335 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1336 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1337 log(LOG_DEBUG, "%s; %s: Listen socket: " 1338 "Connection attempt from/to self " 1339 "ignored\n", s, __func__); 1340 goto dropunlock; 1341 } 1342 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1343 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1344 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1345 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1346 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1347 log(LOG_DEBUG, "%s; %s: Listen socket: " 1348 "Connection attempt from/to broad- " 1349 "or multicast address ignored\n", 1350 s, __func__); 1351 goto dropunlock; 1352 } 1353 } 1354#endif 1355 /* 1356 * SYN appears to be valid. Create compressed TCP state 1357 * for syncache. 1358 */ 1359#ifdef TCPDEBUG 1360 if (so->so_options & SO_DEBUG) 1361 tcp_trace(TA_INPUT, ostate, tp, 1362 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1363#endif 1364 tcp_dooptions(&to, optp, optlen, TO_SYN); 1365 syncache_add(&inc, &to, th, inp, &so, m); 1366 /* 1367 * Entry added to syncache and mbuf consumed. 1368 * Everything already unlocked by syncache_add(). 1369 */ 1370 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1371 return; 1372 } else if (tp->t_state == TCPS_LISTEN) { 1373 /* 1374 * When a listen socket is torn down the SO_ACCEPTCONN 1375 * flag is removed first while connections are drained 1376 * from the accept queue in a unlock/lock cycle of the 1377 * ACCEPT_LOCK, opening a race condition allowing a SYN 1378 * attempt go through unhandled. 1379 */ 1380 goto dropunlock; 1381 } 1382 1383#ifdef TCP_SIGNATURE 1384 if (sig_checked == 0) { 1385 tcp_dooptions(&to, optp, optlen, 1386 (thflags & TH_SYN) ? TO_SYN : 0); 1387 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1388 th, tp->t_flags)) { 1389 1390 /* 1391 * In SYN_SENT state if it receives an RST, it is 1392 * allowed for further processing. 1393 */ 1394 if ((thflags & TH_RST) == 0 || 1395 (tp->t_state == TCPS_SYN_SENT) == 0) 1396 goto dropunlock; 1397 } 1398 sig_checked = 1; 1399 } 1400#endif 1401 1402 /* 1403 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1404 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1405 * the inpcb, and unlocks pcbinfo. 1406 */ 1407 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1408 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1409 return; 1410 1411dropwithreset: 1412 if (ti_locked == TI_WLOCKED) { 1413 INP_INFO_WUNLOCK(&V_tcbinfo); 1414 ti_locked = TI_UNLOCKED; 1415 } 1416#ifdef INVARIANTS 1417 else { 1418 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1419 "ti_locked: %d", __func__, ti_locked)); 1420 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1421 } 1422#endif 1423 1424 if (inp != NULL) { 1425 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1426 INP_WUNLOCK(inp); 1427 } else 1428 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1429 m = NULL; /* mbuf chain got consumed. */ 1430 goto drop; 1431 1432dropunlock: 1433 if (ti_locked == TI_WLOCKED) { 1434 INP_INFO_WUNLOCK(&V_tcbinfo); 1435 ti_locked = TI_UNLOCKED; 1436 } 1437#ifdef INVARIANTS 1438 else { 1439 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1440 "ti_locked: %d", __func__, ti_locked)); 1441 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1442 } 1443#endif 1444 1445 if (inp != NULL) 1446 INP_WUNLOCK(inp); 1447 1448drop: 1449 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1450 if (s != NULL) 1451 free(s, M_TCPLOG); 1452 if (m != NULL) 1453 m_freem(m); 1454} 1455 1456static void 1457tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1458 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1459 int ti_locked) 1460{ 1461 int thflags, acked, ourfinisacked, needoutput = 0; 1462 int rstreason, todrop, win; 1463 u_long tiwin; 1464 struct tcpopt to; 1465 1466#ifdef TCPDEBUG 1467 /* 1468 * The size of tcp_saveipgen must be the size of the max ip header, 1469 * now IPv6. 1470 */ 1471 u_char tcp_saveipgen[IP6_HDR_LEN]; 1472 struct tcphdr tcp_savetcp; 1473 short ostate = 0; 1474#endif 1475 thflags = th->th_flags; 1476 tp->sackhint.last_sack_ack = 0; 1477 1478 /* 1479 * If this is either a state-changing packet or current state isn't 1480 * established, we require a write lock on tcbinfo. Otherwise, we 1481 * allow either a read lock or a write lock, as we may have acquired 1482 * a write lock due to a race. 1483 * 1484 * Require a global write lock for SYN/FIN/RST segments or 1485 * non-established connections; otherwise accept either a read or 1486 * write lock, as we may have conservatively acquired a write lock in 1487 * certain cases in tcp_input() (is this still true?). Currently we 1488 * will never enter with no lock, so we try to drop it quickly in the 1489 * common pure ack/pure data cases. 1490 */ 1491 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1492 tp->t_state != TCPS_ESTABLISHED) { 1493 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1494 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1495 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1496 } else { 1497#ifdef INVARIANTS 1498 if (ti_locked == TI_WLOCKED) 1499 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1500 else { 1501 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1502 "ti_locked: %d", __func__, ti_locked)); 1503 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1504 } 1505#endif 1506 } 1507 INP_WLOCK_ASSERT(tp->t_inpcb); 1508 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1509 __func__)); 1510 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1511 __func__)); 1512 1513 /* 1514 * Segment received on connection. 1515 * Reset idle time and keep-alive timer. 1516 * XXX: This should be done after segment 1517 * validation to ignore broken/spoofed segs. 1518 */ 1519 tp->t_rcvtime = ticks; 1520 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1521 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1522 1523 /* 1524 * Unscale the window into a 32-bit value. 1525 * For the SYN_SENT state the scale is zero. 1526 */ 1527 tiwin = th->th_win << tp->snd_scale; 1528 1529 /* 1530 * TCP ECN processing. 1531 */ 1532 if (tp->t_flags & TF_ECN_PERMIT) { 1533 if (thflags & TH_CWR) 1534 tp->t_flags &= ~TF_ECN_SND_ECE; 1535 switch (iptos & IPTOS_ECN_MASK) { 1536 case IPTOS_ECN_CE: 1537 tp->t_flags |= TF_ECN_SND_ECE; 1538 TCPSTAT_INC(tcps_ecn_ce); 1539 break; 1540 case IPTOS_ECN_ECT0: 1541 TCPSTAT_INC(tcps_ecn_ect0); 1542 break; 1543 case IPTOS_ECN_ECT1: 1544 TCPSTAT_INC(tcps_ecn_ect1); 1545 break; 1546 } 1547 /* Congestion experienced. */ 1548 if (thflags & TH_ECE) { 1549 cc_cong_signal(tp, th, CC_ECN); 1550 } 1551 } 1552 1553 /* 1554 * Parse options on any incoming segment. 1555 */ 1556 tcp_dooptions(&to, (u_char *)(th + 1), 1557 (th->th_off << 2) - sizeof(struct tcphdr), 1558 (thflags & TH_SYN) ? TO_SYN : 0); 1559 1560 /* 1561 * If echoed timestamp is later than the current time, 1562 * fall back to non RFC1323 RTT calculation. Normalize 1563 * timestamp if syncookies were used when this connection 1564 * was established. 1565 */ 1566 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1567 to.to_tsecr -= tp->ts_offset; 1568 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1569 to.to_tsecr = 0; 1570 } 1571 1572 /* 1573 * Process options only when we get SYN/ACK back. The SYN case 1574 * for incoming connections is handled in tcp_syncache. 1575 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1576 * or <SYN,ACK>) segment itself is never scaled. 1577 * XXX this is traditional behavior, may need to be cleaned up. 1578 */ 1579 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1580 if ((to.to_flags & TOF_SCALE) && 1581 (tp->t_flags & TF_REQ_SCALE)) { 1582 tp->t_flags |= TF_RCVD_SCALE; 1583 tp->snd_scale = to.to_wscale; 1584 } 1585 /* 1586 * Initial send window. It will be updated with 1587 * the next incoming segment to the scaled value. 1588 */ 1589 tp->snd_wnd = th->th_win; 1590 if (to.to_flags & TOF_TS) { 1591 tp->t_flags |= TF_RCVD_TSTMP; 1592 tp->ts_recent = to.to_tsval; 1593 tp->ts_recent_age = tcp_ts_getticks(); 1594 } 1595 if (to.to_flags & TOF_MSS) 1596 tcp_mss(tp, to.to_mss); 1597 if ((tp->t_flags & TF_SACK_PERMIT) && 1598 (to.to_flags & TOF_SACKPERM) == 0) 1599 tp->t_flags &= ~TF_SACK_PERMIT; 1600 } 1601 1602 /* 1603 * Header prediction: check for the two common cases 1604 * of a uni-directional data xfer. If the packet has 1605 * no control flags, is in-sequence, the window didn't 1606 * change and we're not retransmitting, it's a 1607 * candidate. If the length is zero and the ack moved 1608 * forward, we're the sender side of the xfer. Just 1609 * free the data acked & wake any higher level process 1610 * that was blocked waiting for space. If the length 1611 * is non-zero and the ack didn't move, we're the 1612 * receiver side. If we're getting packets in-order 1613 * (the reassembly queue is empty), add the data to 1614 * the socket buffer and note that we need a delayed ack. 1615 * Make sure that the hidden state-flags are also off. 1616 * Since we check for TCPS_ESTABLISHED first, it can only 1617 * be TH_NEEDSYN. 1618 */ 1619 if (tp->t_state == TCPS_ESTABLISHED && 1620 th->th_seq == tp->rcv_nxt && 1621 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1622 tp->snd_nxt == tp->snd_max && 1623 tiwin && tiwin == tp->snd_wnd && 1624 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1625 LIST_EMPTY(&tp->t_segq) && 1626 ((to.to_flags & TOF_TS) == 0 || 1627 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1628 1629 /* 1630 * If last ACK falls within this segment's sequence numbers, 1631 * record the timestamp. 1632 * NOTE that the test is modified according to the latest 1633 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1634 */ 1635 if ((to.to_flags & TOF_TS) != 0 && 1636 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1637 tp->ts_recent_age = tcp_ts_getticks(); 1638 tp->ts_recent = to.to_tsval; 1639 } 1640 1641 if (tlen == 0) { 1642 if (SEQ_GT(th->th_ack, tp->snd_una) && 1643 SEQ_LEQ(th->th_ack, tp->snd_max) && 1644 !IN_RECOVERY(tp->t_flags) && 1645 (to.to_flags & TOF_SACK) == 0 && 1646 TAILQ_EMPTY(&tp->snd_holes)) { 1647 /* 1648 * This is a pure ack for outstanding data. 1649 */ 1650 if (ti_locked == TI_WLOCKED) 1651 INP_INFO_WUNLOCK(&V_tcbinfo); 1652 ti_locked = TI_UNLOCKED; 1653 1654 TCPSTAT_INC(tcps_predack); 1655 1656 /* 1657 * "bad retransmit" recovery. 1658 */ 1659 if (tp->t_rxtshift == 1 && 1660 tp->t_flags & TF_PREVVALID && 1661 (int)(ticks - tp->t_badrxtwin) < 0) { 1662 cc_cong_signal(tp, th, CC_RTO_ERR); 1663 } 1664 1665 /* 1666 * Recalculate the transmit timer / rtt. 1667 * 1668 * Some boxes send broken timestamp replies 1669 * during the SYN+ACK phase, ignore 1670 * timestamps of 0 or we could calculate a 1671 * huge RTT and blow up the retransmit timer. 1672 */ 1673 if ((to.to_flags & TOF_TS) != 0 && 1674 to.to_tsecr) { 1675 u_int t; 1676 1677 t = tcp_ts_getticks() - to.to_tsecr; 1678 if (!tp->t_rttlow || tp->t_rttlow > t) 1679 tp->t_rttlow = t; 1680 tcp_xmit_timer(tp, 1681 TCP_TS_TO_TICKS(t) + 1); 1682 } else if (tp->t_rtttime && 1683 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1684 if (!tp->t_rttlow || 1685 tp->t_rttlow > ticks - tp->t_rtttime) 1686 tp->t_rttlow = ticks - tp->t_rtttime; 1687 tcp_xmit_timer(tp, 1688 ticks - tp->t_rtttime); 1689 } 1690 acked = BYTES_THIS_ACK(tp, th); 1691 1692 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1693 hhook_run_tcp_est_in(tp, th, &to); 1694 1695 TCPSTAT_INC(tcps_rcvackpack); 1696 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1697 sbdrop(&so->so_snd, acked); 1698 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1699 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1700 tp->snd_recover = th->th_ack - 1; 1701 1702 /* 1703 * Let the congestion control algorithm update 1704 * congestion control related information. This 1705 * typically means increasing the congestion 1706 * window. 1707 */ 1708 cc_ack_received(tp, th, CC_ACK); 1709 1710 tp->snd_una = th->th_ack; 1711 /* 1712 * Pull snd_wl2 up to prevent seq wrap relative 1713 * to th_ack. 1714 */ 1715 tp->snd_wl2 = th->th_ack; 1716 tp->t_dupacks = 0; 1717 m_freem(m); 1718 ND6_HINT(tp); /* Some progress has been made. */ 1719 1720 /* 1721 * If all outstanding data are acked, stop 1722 * retransmit timer, otherwise restart timer 1723 * using current (possibly backed-off) value. 1724 * If process is waiting for space, 1725 * wakeup/selwakeup/signal. If data 1726 * are ready to send, let tcp_output 1727 * decide between more output or persist. 1728 */ 1729#ifdef TCPDEBUG 1730 if (so->so_options & SO_DEBUG) 1731 tcp_trace(TA_INPUT, ostate, tp, 1732 (void *)tcp_saveipgen, 1733 &tcp_savetcp, 0); 1734#endif 1735 if (tp->snd_una == tp->snd_max) 1736 tcp_timer_activate(tp, TT_REXMT, 0); 1737 else if (!tcp_timer_active(tp, TT_PERSIST)) 1738 tcp_timer_activate(tp, TT_REXMT, 1739 tp->t_rxtcur); 1740 sowwakeup(so); 1741 if (so->so_snd.sb_cc) 1742 (void) tcp_output(tp); 1743 goto check_delack; 1744 } 1745 } else if (th->th_ack == tp->snd_una && 1746 tlen <= sbspace(&so->so_rcv)) { 1747 int newsize = 0; /* automatic sockbuf scaling */ 1748 1749 /* 1750 * This is a pure, in-sequence data packet with 1751 * nothing on the reassembly queue and we have enough 1752 * buffer space to take it. 1753 */ 1754 if (ti_locked == TI_WLOCKED) 1755 INP_INFO_WUNLOCK(&V_tcbinfo); 1756 ti_locked = TI_UNLOCKED; 1757 1758 /* Clean receiver SACK report if present */ 1759 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1760 tcp_clean_sackreport(tp); 1761 TCPSTAT_INC(tcps_preddat); 1762 tp->rcv_nxt += tlen; 1763 /* 1764 * Pull snd_wl1 up to prevent seq wrap relative to 1765 * th_seq. 1766 */ 1767 tp->snd_wl1 = th->th_seq; 1768 /* 1769 * Pull rcv_up up to prevent seq wrap relative to 1770 * rcv_nxt. 1771 */ 1772 tp->rcv_up = tp->rcv_nxt; 1773 TCPSTAT_INC(tcps_rcvpack); 1774 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1775 ND6_HINT(tp); /* Some progress has been made */ 1776#ifdef TCPDEBUG 1777 if (so->so_options & SO_DEBUG) 1778 tcp_trace(TA_INPUT, ostate, tp, 1779 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1780#endif 1781 /* 1782 * Automatic sizing of receive socket buffer. Often the send 1783 * buffer size is not optimally adjusted to the actual network 1784 * conditions at hand (delay bandwidth product). Setting the 1785 * buffer size too small limits throughput on links with high 1786 * bandwidth and high delay (eg. trans-continental/oceanic links). 1787 * 1788 * On the receive side the socket buffer memory is only rarely 1789 * used to any significant extent. This allows us to be much 1790 * more aggressive in scaling the receive socket buffer. For 1791 * the case that the buffer space is actually used to a large 1792 * extent and we run out of kernel memory we can simply drop 1793 * the new segments; TCP on the sender will just retransmit it 1794 * later. Setting the buffer size too big may only consume too 1795 * much kernel memory if the application doesn't read() from 1796 * the socket or packet loss or reordering makes use of the 1797 * reassembly queue. 1798 * 1799 * The criteria to step up the receive buffer one notch are: 1800 * 1. the number of bytes received during the time it takes 1801 * one timestamp to be reflected back to us (the RTT); 1802 * 2. received bytes per RTT is within seven eighth of the 1803 * current socket buffer size; 1804 * 3. receive buffer size has not hit maximal automatic size; 1805 * 1806 * This algorithm does one step per RTT at most and only if 1807 * we receive a bulk stream w/o packet losses or reorderings. 1808 * Shrinking the buffer during idle times is not necessary as 1809 * it doesn't consume any memory when idle. 1810 * 1811 * TODO: Only step up if the application is actually serving 1812 * the buffer to better manage the socket buffer resources. 1813 */ 1814 if (V_tcp_do_autorcvbuf && 1815 to.to_tsecr && 1816 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1817 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1818 to.to_tsecr - tp->rfbuf_ts < hz) { 1819 if (tp->rfbuf_cnt > 1820 (so->so_rcv.sb_hiwat / 8 * 7) && 1821 so->so_rcv.sb_hiwat < 1822 V_tcp_autorcvbuf_max) { 1823 newsize = 1824 min(so->so_rcv.sb_hiwat + 1825 V_tcp_autorcvbuf_inc, 1826 V_tcp_autorcvbuf_max); 1827 } 1828 /* Start over with next RTT. */ 1829 tp->rfbuf_ts = 0; 1830 tp->rfbuf_cnt = 0; 1831 } else 1832 tp->rfbuf_cnt += tlen; /* add up */ 1833 } 1834 1835 /* Add data to socket buffer. */ 1836 SOCKBUF_LOCK(&so->so_rcv); 1837 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1838 m_freem(m); 1839 } else { 1840 /* 1841 * Set new socket buffer size. 1842 * Give up when limit is reached. 1843 */ 1844 if (newsize) 1845 if (!sbreserve_locked(&so->so_rcv, 1846 newsize, so, NULL)) 1847 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1848 m_adj(m, drop_hdrlen); /* delayed header drop */ 1849 sbappendstream_locked(&so->so_rcv, m); 1850 } 1851 /* NB: sorwakeup_locked() does an implicit unlock. */ 1852 sorwakeup_locked(so); 1853 if (DELAY_ACK(tp)) { 1854 tp->t_flags |= TF_DELACK; 1855 } else { 1856 tp->t_flags |= TF_ACKNOW; 1857 tcp_output(tp); 1858 } 1859 goto check_delack; 1860 } 1861 } 1862 1863 /* 1864 * Calculate amount of space in receive window, 1865 * and then do TCP input processing. 1866 * Receive window is amount of space in rcv queue, 1867 * but not less than advertised window. 1868 */ 1869 win = sbspace(&so->so_rcv); 1870 if (win < 0) 1871 win = 0; 1872 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1873 1874 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1875 tp->rfbuf_ts = 0; 1876 tp->rfbuf_cnt = 0; 1877 1878 switch (tp->t_state) { 1879 1880 /* 1881 * If the state is SYN_RECEIVED: 1882 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1883 */ 1884 case TCPS_SYN_RECEIVED: 1885 if ((thflags & TH_ACK) && 1886 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1887 SEQ_GT(th->th_ack, tp->snd_max))) { 1888 rstreason = BANDLIM_RST_OPENPORT; 1889 goto dropwithreset; 1890 } 1891 break; 1892 1893 /* 1894 * If the state is SYN_SENT: 1895 * if seg contains an ACK, but not for our SYN, drop the input. 1896 * if seg contains a RST, then drop the connection. 1897 * if seg does not contain SYN, then drop it. 1898 * Otherwise this is an acceptable SYN segment 1899 * initialize tp->rcv_nxt and tp->irs 1900 * if seg contains ack then advance tp->snd_una 1901 * if seg contains an ECE and ECN support is enabled, the stream 1902 * is ECN capable. 1903 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1904 * arrange for segment to be acked (eventually) 1905 * continue processing rest of data/controls, beginning with URG 1906 */ 1907 case TCPS_SYN_SENT: 1908 if ((thflags & TH_ACK) && 1909 (SEQ_LEQ(th->th_ack, tp->iss) || 1910 SEQ_GT(th->th_ack, tp->snd_max))) { 1911 rstreason = BANDLIM_UNLIMITED; 1912 goto dropwithreset; 1913 } 1914 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1915 tp = tcp_drop(tp, ECONNREFUSED); 1916 if (thflags & TH_RST) 1917 goto drop; 1918 if (!(thflags & TH_SYN)) 1919 goto drop; 1920 1921 tp->irs = th->th_seq; 1922 tcp_rcvseqinit(tp); 1923 if (thflags & TH_ACK) { 1924 TCPSTAT_INC(tcps_connects); 1925 soisconnected(so); 1926#ifdef MAC 1927 mac_socketpeer_set_from_mbuf(m, so); 1928#endif 1929 /* Do window scaling on this connection? */ 1930 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1931 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1932 tp->rcv_scale = tp->request_r_scale; 1933 } 1934 tp->rcv_adv += imin(tp->rcv_wnd, 1935 TCP_MAXWIN << tp->rcv_scale); 1936 tp->snd_una++; /* SYN is acked */ 1937 /* 1938 * If there's data, delay ACK; if there's also a FIN 1939 * ACKNOW will be turned on later. 1940 */ 1941 if (DELAY_ACK(tp) && tlen != 0) 1942 tcp_timer_activate(tp, TT_DELACK, 1943 tcp_delacktime); 1944 else 1945 tp->t_flags |= TF_ACKNOW; 1946 1947 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1948 tp->t_flags |= TF_ECN_PERMIT; 1949 TCPSTAT_INC(tcps_ecn_shs); 1950 } 1951 1952 /* 1953 * Received <SYN,ACK> in SYN_SENT[*] state. 1954 * Transitions: 1955 * SYN_SENT --> ESTABLISHED 1956 * SYN_SENT* --> FIN_WAIT_1 1957 */ 1958 tp->t_starttime = ticks; 1959 if (tp->t_flags & TF_NEEDFIN) { 1960 tp->t_state = TCPS_FIN_WAIT_1; 1961 tp->t_flags &= ~TF_NEEDFIN; 1962 thflags &= ~TH_SYN; 1963 } else { 1964 tp->t_state = TCPS_ESTABLISHED; 1965 cc_conn_init(tp); 1966 tcp_timer_activate(tp, TT_KEEP, 1967 TP_KEEPIDLE(tp)); 1968 } 1969 } else { 1970 /* 1971 * Received initial SYN in SYN-SENT[*] state => 1972 * simultaneous open. If segment contains CC option 1973 * and there is a cached CC, apply TAO test. 1974 * If it succeeds, connection is * half-synchronized. 1975 * Otherwise, do 3-way handshake: 1976 * SYN-SENT -> SYN-RECEIVED 1977 * SYN-SENT* -> SYN-RECEIVED* 1978 * If there was no CC option, clear cached CC value. 1979 */ 1980 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1981 tcp_timer_activate(tp, TT_REXMT, 0); 1982 tp->t_state = TCPS_SYN_RECEIVED; 1983 } 1984 1985 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1986 "ti_locked %d", __func__, ti_locked)); 1987 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1988 INP_WLOCK_ASSERT(tp->t_inpcb); 1989 1990 /* 1991 * Advance th->th_seq to correspond to first data byte. 1992 * If data, trim to stay within window, 1993 * dropping FIN if necessary. 1994 */ 1995 th->th_seq++; 1996 if (tlen > tp->rcv_wnd) { 1997 todrop = tlen - tp->rcv_wnd; 1998 m_adj(m, -todrop); 1999 tlen = tp->rcv_wnd; 2000 thflags &= ~TH_FIN; 2001 TCPSTAT_INC(tcps_rcvpackafterwin); 2002 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2003 } 2004 tp->snd_wl1 = th->th_seq - 1; 2005 tp->rcv_up = th->th_seq; 2006 /* 2007 * Client side of transaction: already sent SYN and data. 2008 * If the remote host used T/TCP to validate the SYN, 2009 * our data will be ACK'd; if so, enter normal data segment 2010 * processing in the middle of step 5, ack processing. 2011 * Otherwise, goto step 6. 2012 */ 2013 if (thflags & TH_ACK) 2014 goto process_ACK; 2015 2016 goto step6; 2017 2018 /* 2019 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2020 * do normal processing. 2021 * 2022 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2023 */ 2024 case TCPS_LAST_ACK: 2025 case TCPS_CLOSING: 2026 break; /* continue normal processing */ 2027 } 2028 2029 /* 2030 * States other than LISTEN or SYN_SENT. 2031 * First check the RST flag and sequence number since reset segments 2032 * are exempt from the timestamp and connection count tests. This 2033 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2034 * below which allowed reset segments in half the sequence space 2035 * to fall though and be processed (which gives forged reset 2036 * segments with a random sequence number a 50 percent chance of 2037 * killing a connection). 2038 * Then check timestamp, if present. 2039 * Then check the connection count, if present. 2040 * Then check that at least some bytes of segment are within 2041 * receive window. If segment begins before rcv_nxt, 2042 * drop leading data (and SYN); if nothing left, just ack. 2043 * 2044 * 2045 * If the RST bit is set, check the sequence number to see 2046 * if this is a valid reset segment. 2047 * RFC 793 page 37: 2048 * In all states except SYN-SENT, all reset (RST) segments 2049 * are validated by checking their SEQ-fields. A reset is 2050 * valid if its sequence number is in the window. 2051 * Note: this does not take into account delayed ACKs, so 2052 * we should test against last_ack_sent instead of rcv_nxt. 2053 * The sequence number in the reset segment is normally an 2054 * echo of our outgoing acknowlegement numbers, but some hosts 2055 * send a reset with the sequence number at the rightmost edge 2056 * of our receive window, and we have to handle this case. 2057 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2058 * that brute force RST attacks are possible. To combat this, 2059 * we use a much stricter check while in the ESTABLISHED state, 2060 * only accepting RSTs where the sequence number is equal to 2061 * last_ack_sent. In all other states (the states in which a 2062 * RST is more likely), the more permissive check is used. 2063 * If we have multiple segments in flight, the initial reset 2064 * segment sequence numbers will be to the left of last_ack_sent, 2065 * but they will eventually catch up. 2066 * In any case, it never made sense to trim reset segments to 2067 * fit the receive window since RFC 1122 says: 2068 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2069 * 2070 * A TCP SHOULD allow a received RST segment to include data. 2071 * 2072 * DISCUSSION 2073 * It has been suggested that a RST segment could contain 2074 * ASCII text that encoded and explained the cause of the 2075 * RST. No standard has yet been established for such 2076 * data. 2077 * 2078 * If the reset segment passes the sequence number test examine 2079 * the state: 2080 * SYN_RECEIVED STATE: 2081 * If passive open, return to LISTEN state. 2082 * If active open, inform user that connection was refused. 2083 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2084 * Inform user that connection was reset, and close tcb. 2085 * CLOSING, LAST_ACK STATES: 2086 * Close the tcb. 2087 * TIME_WAIT STATE: 2088 * Drop the segment - see Stevens, vol. 2, p. 964 and 2089 * RFC 1337. 2090 */ 2091 if (thflags & TH_RST) { 2092 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2093 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2094 switch (tp->t_state) { 2095 2096 case TCPS_SYN_RECEIVED: 2097 so->so_error = ECONNREFUSED; 2098 goto close; 2099 2100 case TCPS_ESTABLISHED: 2101 if (V_tcp_insecure_rst == 0 && 2102 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2103 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2104 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2105 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2106 TCPSTAT_INC(tcps_badrst); 2107 goto drop; 2108 } 2109 /* FALLTHROUGH */ 2110 case TCPS_FIN_WAIT_1: 2111 case TCPS_FIN_WAIT_2: 2112 case TCPS_CLOSE_WAIT: 2113 so->so_error = ECONNRESET; 2114 close: 2115 KASSERT(ti_locked == TI_WLOCKED, 2116 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2117 ti_locked)); 2118 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2119 2120 tp->t_state = TCPS_CLOSED; 2121 TCPSTAT_INC(tcps_drops); 2122 tp = tcp_close(tp); 2123 break; 2124 2125 case TCPS_CLOSING: 2126 case TCPS_LAST_ACK: 2127 KASSERT(ti_locked == TI_WLOCKED, 2128 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2129 ti_locked)); 2130 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2131 2132 tp = tcp_close(tp); 2133 break; 2134 } 2135 } 2136 goto drop; 2137 } 2138 2139 /* 2140 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2141 * and it's less than ts_recent, drop it. 2142 */ 2143 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2144 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2145 2146 /* Check to see if ts_recent is over 24 days old. */ 2147 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2148 /* 2149 * Invalidate ts_recent. If this segment updates 2150 * ts_recent, the age will be reset later and ts_recent 2151 * will get a valid value. If it does not, setting 2152 * ts_recent to zero will at least satisfy the 2153 * requirement that zero be placed in the timestamp 2154 * echo reply when ts_recent isn't valid. The 2155 * age isn't reset until we get a valid ts_recent 2156 * because we don't want out-of-order segments to be 2157 * dropped when ts_recent is old. 2158 */ 2159 tp->ts_recent = 0; 2160 } else { 2161 TCPSTAT_INC(tcps_rcvduppack); 2162 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2163 TCPSTAT_INC(tcps_pawsdrop); 2164 if (tlen) 2165 goto dropafterack; 2166 goto drop; 2167 } 2168 } 2169 2170 /* 2171 * In the SYN-RECEIVED state, validate that the packet belongs to 2172 * this connection before trimming the data to fit the receive 2173 * window. Check the sequence number versus IRS since we know 2174 * the sequence numbers haven't wrapped. This is a partial fix 2175 * for the "LAND" DoS attack. 2176 */ 2177 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2178 rstreason = BANDLIM_RST_OPENPORT; 2179 goto dropwithreset; 2180 } 2181 2182 todrop = tp->rcv_nxt - th->th_seq; 2183 if (todrop > 0) { 2184 if (thflags & TH_SYN) { 2185 thflags &= ~TH_SYN; 2186 th->th_seq++; 2187 if (th->th_urp > 1) 2188 th->th_urp--; 2189 else 2190 thflags &= ~TH_URG; 2191 todrop--; 2192 } 2193 /* 2194 * Following if statement from Stevens, vol. 2, p. 960. 2195 */ 2196 if (todrop > tlen 2197 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2198 /* 2199 * Any valid FIN must be to the left of the window. 2200 * At this point the FIN must be a duplicate or out 2201 * of sequence; drop it. 2202 */ 2203 thflags &= ~TH_FIN; 2204 2205 /* 2206 * Send an ACK to resynchronize and drop any data. 2207 * But keep on processing for RST or ACK. 2208 */ 2209 tp->t_flags |= TF_ACKNOW; 2210 todrop = tlen; 2211 TCPSTAT_INC(tcps_rcvduppack); 2212 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2213 } else { 2214 TCPSTAT_INC(tcps_rcvpartduppack); 2215 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2216 } 2217 drop_hdrlen += todrop; /* drop from the top afterwards */ 2218 th->th_seq += todrop; 2219 tlen -= todrop; 2220 if (th->th_urp > todrop) 2221 th->th_urp -= todrop; 2222 else { 2223 thflags &= ~TH_URG; 2224 th->th_urp = 0; 2225 } 2226 } 2227 2228 /* 2229 * If new data are received on a connection after the 2230 * user processes are gone, then RST the other end. 2231 */ 2232 if ((so->so_state & SS_NOFDREF) && 2233 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2234 char *s; 2235 2236 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2237 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2238 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2239 2240 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 2241 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 2242 "was closed, sending RST and removing tcpcb\n", 2243 s, __func__, tcpstates[tp->t_state], tlen); 2244 free(s, M_TCPLOG); 2245 } 2246 tp = tcp_close(tp); 2247 TCPSTAT_INC(tcps_rcvafterclose); 2248 rstreason = BANDLIM_UNLIMITED; 2249 goto dropwithreset; 2250 } 2251 2252 /* 2253 * If segment ends after window, drop trailing data 2254 * (and PUSH and FIN); if nothing left, just ACK. 2255 */ 2256 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2257 if (todrop > 0) { 2258 TCPSTAT_INC(tcps_rcvpackafterwin); 2259 if (todrop >= tlen) { 2260 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2261 /* 2262 * If window is closed can only take segments at 2263 * window edge, and have to drop data and PUSH from 2264 * incoming segments. Continue processing, but 2265 * remember to ack. Otherwise, drop segment 2266 * and ack. 2267 */ 2268 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2269 tp->t_flags |= TF_ACKNOW; 2270 TCPSTAT_INC(tcps_rcvwinprobe); 2271 } else 2272 goto dropafterack; 2273 } else 2274 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2275 m_adj(m, -todrop); 2276 tlen -= todrop; 2277 thflags &= ~(TH_PUSH|TH_FIN); 2278 } 2279 2280 /* 2281 * If last ACK falls within this segment's sequence numbers, 2282 * record its timestamp. 2283 * NOTE: 2284 * 1) That the test incorporates suggestions from the latest 2285 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2286 * 2) That updating only on newer timestamps interferes with 2287 * our earlier PAWS tests, so this check should be solely 2288 * predicated on the sequence space of this segment. 2289 * 3) That we modify the segment boundary check to be 2290 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2291 * instead of RFC1323's 2292 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2293 * This modified check allows us to overcome RFC1323's 2294 * limitations as described in Stevens TCP/IP Illustrated 2295 * Vol. 2 p.869. In such cases, we can still calculate the 2296 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2297 */ 2298 if ((to.to_flags & TOF_TS) != 0 && 2299 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2300 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2301 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2302 tp->ts_recent_age = tcp_ts_getticks(); 2303 tp->ts_recent = to.to_tsval; 2304 } 2305 2306 /* 2307 * If a SYN is in the window, then this is an 2308 * error and we send an RST and drop the connection. 2309 */ 2310 if (thflags & TH_SYN) { 2311 KASSERT(ti_locked == TI_WLOCKED, 2312 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2313 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2314 2315 tp = tcp_drop(tp, ECONNRESET); 2316 rstreason = BANDLIM_UNLIMITED; 2317 goto drop; 2318 } 2319 2320 /* 2321 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2322 * flag is on (half-synchronized state), then queue data for 2323 * later processing; else drop segment and return. 2324 */ 2325 if ((thflags & TH_ACK) == 0) { 2326 if (tp->t_state == TCPS_SYN_RECEIVED || 2327 (tp->t_flags & TF_NEEDSYN)) 2328 goto step6; 2329 else if (tp->t_flags & TF_ACKNOW) 2330 goto dropafterack; 2331 else 2332 goto drop; 2333 } 2334 2335 /* 2336 * Ack processing. 2337 */ 2338 switch (tp->t_state) { 2339 2340 /* 2341 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2342 * ESTABLISHED state and continue processing. 2343 * The ACK was checked above. 2344 */ 2345 case TCPS_SYN_RECEIVED: 2346 2347 TCPSTAT_INC(tcps_connects); 2348 soisconnected(so); 2349 /* Do window scaling? */ 2350 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2351 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2352 tp->rcv_scale = tp->request_r_scale; 2353 tp->snd_wnd = tiwin; 2354 } 2355 /* 2356 * Make transitions: 2357 * SYN-RECEIVED -> ESTABLISHED 2358 * SYN-RECEIVED* -> FIN-WAIT-1 2359 */ 2360 tp->t_starttime = ticks; 2361 if (tp->t_flags & TF_NEEDFIN) { 2362 tp->t_state = TCPS_FIN_WAIT_1; 2363 tp->t_flags &= ~TF_NEEDFIN; 2364 } else { 2365 tp->t_state = TCPS_ESTABLISHED; 2366 cc_conn_init(tp); 2367 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2368 } 2369 /* 2370 * If segment contains data or ACK, will call tcp_reass() 2371 * later; if not, do so now to pass queued data to user. 2372 */ 2373 if (tlen == 0 && (thflags & TH_FIN) == 0) 2374 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2375 (struct mbuf *)0); 2376 tp->snd_wl1 = th->th_seq - 1; 2377 /* FALLTHROUGH */ 2378 2379 /* 2380 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2381 * ACKs. If the ack is in the range 2382 * tp->snd_una < th->th_ack <= tp->snd_max 2383 * then advance tp->snd_una to th->th_ack and drop 2384 * data from the retransmission queue. If this ACK reflects 2385 * more up to date window information we update our window information. 2386 */ 2387 case TCPS_ESTABLISHED: 2388 case TCPS_FIN_WAIT_1: 2389 case TCPS_FIN_WAIT_2: 2390 case TCPS_CLOSE_WAIT: 2391 case TCPS_CLOSING: 2392 case TCPS_LAST_ACK: 2393 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2394 TCPSTAT_INC(tcps_rcvacktoomuch); 2395 goto dropafterack; 2396 } 2397 if ((tp->t_flags & TF_SACK_PERMIT) && 2398 ((to.to_flags & TOF_SACK) || 2399 !TAILQ_EMPTY(&tp->snd_holes))) 2400 tcp_sack_doack(tp, &to, th->th_ack); 2401 2402 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2403 hhook_run_tcp_est_in(tp, th, &to); 2404 2405 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2406 if (tlen == 0 && tiwin == tp->snd_wnd) { 2407 TCPSTAT_INC(tcps_rcvdupack); 2408 /* 2409 * If we have outstanding data (other than 2410 * a window probe), this is a completely 2411 * duplicate ack (ie, window info didn't 2412 * change), the ack is the biggest we've 2413 * seen and we've seen exactly our rexmt 2414 * threshhold of them, assume a packet 2415 * has been dropped and retransmit it. 2416 * Kludge snd_nxt & the congestion 2417 * window so we send only this one 2418 * packet. 2419 * 2420 * We know we're losing at the current 2421 * window size so do congestion avoidance 2422 * (set ssthresh to half the current window 2423 * and pull our congestion window back to 2424 * the new ssthresh). 2425 * 2426 * Dup acks mean that packets have left the 2427 * network (they're now cached at the receiver) 2428 * so bump cwnd by the amount in the receiver 2429 * to keep a constant cwnd packets in the 2430 * network. 2431 * 2432 * When using TCP ECN, notify the peer that 2433 * we reduced the cwnd. 2434 */ 2435 if (!tcp_timer_active(tp, TT_REXMT) || 2436 th->th_ack != tp->snd_una) 2437 tp->t_dupacks = 0; 2438 else if (++tp->t_dupacks > tcprexmtthresh || 2439 IN_FASTRECOVERY(tp->t_flags)) { 2440 cc_ack_received(tp, th, CC_DUPACK); 2441 if ((tp->t_flags & TF_SACK_PERMIT) && 2442 IN_FASTRECOVERY(tp->t_flags)) { 2443 int awnd; 2444 2445 /* 2446 * Compute the amount of data in flight first. 2447 * We can inject new data into the pipe iff 2448 * we have less than 1/2 the original window's 2449 * worth of data in flight. 2450 */ 2451 awnd = (tp->snd_nxt - tp->snd_fack) + 2452 tp->sackhint.sack_bytes_rexmit; 2453 if (awnd < tp->snd_ssthresh) { 2454 tp->snd_cwnd += tp->t_maxseg; 2455 if (tp->snd_cwnd > tp->snd_ssthresh) 2456 tp->snd_cwnd = tp->snd_ssthresh; 2457 } 2458 } else 2459 tp->snd_cwnd += tp->t_maxseg; 2460 if ((thflags & TH_FIN) && 2461 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2462 /* 2463 * If its a fin we need to process 2464 * it to avoid a race where both 2465 * sides enter FIN-WAIT and send FIN|ACK 2466 * at the same time. 2467 */ 2468 break; 2469 } 2470 (void) tcp_output(tp); 2471 goto drop; 2472 } else if (tp->t_dupacks == tcprexmtthresh) { 2473 tcp_seq onxt = tp->snd_nxt; 2474 2475 /* 2476 * If we're doing sack, check to 2477 * see if we're already in sack 2478 * recovery. If we're not doing sack, 2479 * check to see if we're in newreno 2480 * recovery. 2481 */ 2482 if (tp->t_flags & TF_SACK_PERMIT) { 2483 if (IN_FASTRECOVERY(tp->t_flags)) { 2484 tp->t_dupacks = 0; 2485 break; 2486 } 2487 } else { 2488 if (SEQ_LEQ(th->th_ack, 2489 tp->snd_recover)) { 2490 tp->t_dupacks = 0; 2491 break; 2492 } 2493 } 2494 /* Congestion signal before ack. */ 2495 cc_cong_signal(tp, th, CC_NDUPACK); 2496 cc_ack_received(tp, th, CC_DUPACK); 2497 tcp_timer_activate(tp, TT_REXMT, 0); 2498 tp->t_rtttime = 0; 2499 if (tp->t_flags & TF_SACK_PERMIT) { 2500 TCPSTAT_INC( 2501 tcps_sack_recovery_episode); 2502 tp->sack_newdata = tp->snd_nxt; 2503 tp->snd_cwnd = tp->t_maxseg; 2504 (void) tcp_output(tp); 2505 goto drop; 2506 } 2507 tp->snd_nxt = th->th_ack; 2508 tp->snd_cwnd = tp->t_maxseg; 2509 if ((thflags & TH_FIN) && 2510 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2511 /* 2512 * If its a fin we need to process 2513 * it to avoid a race where both 2514 * sides enter FIN-WAIT and send FIN|ACK 2515 * at the same time. 2516 */ 2517 break; 2518 } 2519 (void) tcp_output(tp); 2520 KASSERT(tp->snd_limited <= 2, 2521 ("%s: tp->snd_limited too big", 2522 __func__)); 2523 tp->snd_cwnd = tp->snd_ssthresh + 2524 tp->t_maxseg * 2525 (tp->t_dupacks - tp->snd_limited); 2526 if (SEQ_GT(onxt, tp->snd_nxt)) 2527 tp->snd_nxt = onxt; 2528 goto drop; 2529 } else if (V_tcp_do_rfc3042) { 2530 cc_ack_received(tp, th, CC_DUPACK); 2531 u_long oldcwnd = tp->snd_cwnd; 2532 tcp_seq oldsndmax = tp->snd_max; 2533 u_int sent; 2534 int avail; 2535 2536 KASSERT(tp->t_dupacks == 1 || 2537 tp->t_dupacks == 2, 2538 ("%s: dupacks not 1 or 2", 2539 __func__)); 2540 if (tp->t_dupacks == 1) 2541 tp->snd_limited = 0; 2542 tp->snd_cwnd = 2543 (tp->snd_nxt - tp->snd_una) + 2544 (tp->t_dupacks - tp->snd_limited) * 2545 tp->t_maxseg; 2546 if ((thflags & TH_FIN) && 2547 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2548 /* 2549 * If its a fin we need to process 2550 * it to avoid a race where both 2551 * sides enter FIN-WAIT and send FIN|ACK 2552 * at the same time. 2553 */ 2554 break; 2555 } 2556 /* 2557 * Only call tcp_output when there 2558 * is new data available to be sent. 2559 * Otherwise we would send pure ACKs. 2560 */ 2561 SOCKBUF_LOCK(&so->so_snd); 2562 avail = so->so_snd.sb_cc - 2563 (tp->snd_nxt - tp->snd_una); 2564 SOCKBUF_UNLOCK(&so->so_snd); 2565 if (avail > 0) 2566 (void) tcp_output(tp); 2567 sent = tp->snd_max - oldsndmax; 2568 if (sent > tp->t_maxseg) { 2569 KASSERT((tp->t_dupacks == 2 && 2570 tp->snd_limited == 0) || 2571 (sent == tp->t_maxseg + 1 && 2572 tp->t_flags & TF_SENTFIN), 2573 ("%s: sent too much", 2574 __func__)); 2575 tp->snd_limited = 2; 2576 } else if (sent > 0) 2577 ++tp->snd_limited; 2578 tp->snd_cwnd = oldcwnd; 2579 goto drop; 2580 } 2581 } else 2582 tp->t_dupacks = 0; 2583 break; 2584 } 2585 2586 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2587 ("%s: th_ack <= snd_una", __func__)); 2588 2589 /* 2590 * If the congestion window was inflated to account 2591 * for the other side's cached packets, retract it. 2592 */ 2593 if (IN_FASTRECOVERY(tp->t_flags)) { 2594 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2595 if (tp->t_flags & TF_SACK_PERMIT) 2596 tcp_sack_partialack(tp, th); 2597 else 2598 tcp_newreno_partial_ack(tp, th); 2599 } else 2600 cc_post_recovery(tp, th); 2601 } 2602 tp->t_dupacks = 0; 2603 /* 2604 * If we reach this point, ACK is not a duplicate, 2605 * i.e., it ACKs something we sent. 2606 */ 2607 if (tp->t_flags & TF_NEEDSYN) { 2608 /* 2609 * T/TCP: Connection was half-synchronized, and our 2610 * SYN has been ACK'd (so connection is now fully 2611 * synchronized). Go to non-starred state, 2612 * increment snd_una for ACK of SYN, and check if 2613 * we can do window scaling. 2614 */ 2615 tp->t_flags &= ~TF_NEEDSYN; 2616 tp->snd_una++; 2617 /* Do window scaling? */ 2618 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2619 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2620 tp->rcv_scale = tp->request_r_scale; 2621 /* Send window already scaled. */ 2622 } 2623 } 2624 2625process_ACK: 2626 INP_WLOCK_ASSERT(tp->t_inpcb); 2627 2628 acked = BYTES_THIS_ACK(tp, th); 2629 TCPSTAT_INC(tcps_rcvackpack); 2630 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2631 2632 /* 2633 * If we just performed our first retransmit, and the ACK 2634 * arrives within our recovery window, then it was a mistake 2635 * to do the retransmit in the first place. Recover our 2636 * original cwnd and ssthresh, and proceed to transmit where 2637 * we left off. 2638 */ 2639 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2640 (int)(ticks - tp->t_badrxtwin) < 0) 2641 cc_cong_signal(tp, th, CC_RTO_ERR); 2642 2643 /* 2644 * If we have a timestamp reply, update smoothed 2645 * round trip time. If no timestamp is present but 2646 * transmit timer is running and timed sequence 2647 * number was acked, update smoothed round trip time. 2648 * Since we now have an rtt measurement, cancel the 2649 * timer backoff (cf., Phil Karn's retransmit alg.). 2650 * Recompute the initial retransmit timer. 2651 * 2652 * Some boxes send broken timestamp replies 2653 * during the SYN+ACK phase, ignore 2654 * timestamps of 0 or we could calculate a 2655 * huge RTT and blow up the retransmit timer. 2656 */ 2657 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2658 u_int t; 2659 2660 t = tcp_ts_getticks() - to.to_tsecr; 2661 if (!tp->t_rttlow || tp->t_rttlow > t) 2662 tp->t_rttlow = t; 2663 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2664 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2665 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2666 tp->t_rttlow = ticks - tp->t_rtttime; 2667 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2668 } 2669 2670 /* 2671 * If all outstanding data is acked, stop retransmit 2672 * timer and remember to restart (more output or persist). 2673 * If there is more data to be acked, restart retransmit 2674 * timer, using current (possibly backed-off) value. 2675 */ 2676 if (th->th_ack == tp->snd_max) { 2677 tcp_timer_activate(tp, TT_REXMT, 0); 2678 needoutput = 1; 2679 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2680 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2681 2682 /* 2683 * If no data (only SYN) was ACK'd, 2684 * skip rest of ACK processing. 2685 */ 2686 if (acked == 0) 2687 goto step6; 2688 2689 /* 2690 * Let the congestion control algorithm update congestion 2691 * control related information. This typically means increasing 2692 * the congestion window. 2693 */ 2694 cc_ack_received(tp, th, CC_ACK); 2695 2696 SOCKBUF_LOCK(&so->so_snd); 2697 if (acked > so->so_snd.sb_cc) { 2698 tp->snd_wnd -= so->so_snd.sb_cc; 2699 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2700 ourfinisacked = 1; 2701 } else { 2702 sbdrop_locked(&so->so_snd, acked); 2703 tp->snd_wnd -= acked; 2704 ourfinisacked = 0; 2705 } 2706 /* NB: sowwakeup_locked() does an implicit unlock. */ 2707 sowwakeup_locked(so); 2708 /* Detect una wraparound. */ 2709 if (!IN_RECOVERY(tp->t_flags) && 2710 SEQ_GT(tp->snd_una, tp->snd_recover) && 2711 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2712 tp->snd_recover = th->th_ack - 1; 2713 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2714 if (IN_RECOVERY(tp->t_flags) && 2715 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2716 EXIT_RECOVERY(tp->t_flags); 2717 } 2718 tp->snd_una = th->th_ack; 2719 if (tp->t_flags & TF_SACK_PERMIT) { 2720 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2721 tp->snd_recover = tp->snd_una; 2722 } 2723 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2724 tp->snd_nxt = tp->snd_una; 2725 2726 switch (tp->t_state) { 2727 2728 /* 2729 * In FIN_WAIT_1 STATE in addition to the processing 2730 * for the ESTABLISHED state if our FIN is now acknowledged 2731 * then enter FIN_WAIT_2. 2732 */ 2733 case TCPS_FIN_WAIT_1: 2734 if (ourfinisacked) { 2735 /* 2736 * If we can't receive any more 2737 * data, then closing user can proceed. 2738 * Starting the timer is contrary to the 2739 * specification, but if we don't get a FIN 2740 * we'll hang forever. 2741 * 2742 * XXXjl: 2743 * we should release the tp also, and use a 2744 * compressed state. 2745 */ 2746 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2747 soisdisconnected(so); 2748 tcp_timer_activate(tp, TT_2MSL, 2749 (tcp_fast_finwait2_recycle ? 2750 tcp_finwait2_timeout : 2751 TP_MAXIDLE(tp))); 2752 } 2753 tp->t_state = TCPS_FIN_WAIT_2; 2754 } 2755 break; 2756 2757 /* 2758 * In CLOSING STATE in addition to the processing for 2759 * the ESTABLISHED state if the ACK acknowledges our FIN 2760 * then enter the TIME-WAIT state, otherwise ignore 2761 * the segment. 2762 */ 2763 case TCPS_CLOSING: 2764 if (ourfinisacked) { 2765 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2766 tcp_twstart(tp); 2767 INP_INFO_WUNLOCK(&V_tcbinfo); 2768 m_freem(m); 2769 return; 2770 } 2771 break; 2772 2773 /* 2774 * In LAST_ACK, we may still be waiting for data to drain 2775 * and/or to be acked, as well as for the ack of our FIN. 2776 * If our FIN is now acknowledged, delete the TCB, 2777 * enter the closed state and return. 2778 */ 2779 case TCPS_LAST_ACK: 2780 if (ourfinisacked) { 2781 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2782 tp = tcp_close(tp); 2783 goto drop; 2784 } 2785 break; 2786 } 2787 } 2788 2789step6: 2790 INP_WLOCK_ASSERT(tp->t_inpcb); 2791 2792 /* 2793 * Update window information. 2794 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2795 */ 2796 if ((thflags & TH_ACK) && 2797 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2798 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2799 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2800 /* keep track of pure window updates */ 2801 if (tlen == 0 && 2802 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2803 TCPSTAT_INC(tcps_rcvwinupd); 2804 tp->snd_wnd = tiwin; 2805 tp->snd_wl1 = th->th_seq; 2806 tp->snd_wl2 = th->th_ack; 2807 if (tp->snd_wnd > tp->max_sndwnd) 2808 tp->max_sndwnd = tp->snd_wnd; 2809 needoutput = 1; 2810 } 2811 2812 /* 2813 * Process segments with URG. 2814 */ 2815 if ((thflags & TH_URG) && th->th_urp && 2816 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2817 /* 2818 * This is a kludge, but if we receive and accept 2819 * random urgent pointers, we'll crash in 2820 * soreceive. It's hard to imagine someone 2821 * actually wanting to send this much urgent data. 2822 */ 2823 SOCKBUF_LOCK(&so->so_rcv); 2824 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2825 th->th_urp = 0; /* XXX */ 2826 thflags &= ~TH_URG; /* XXX */ 2827 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2828 goto dodata; /* XXX */ 2829 } 2830 /* 2831 * If this segment advances the known urgent pointer, 2832 * then mark the data stream. This should not happen 2833 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2834 * a FIN has been received from the remote side. 2835 * In these states we ignore the URG. 2836 * 2837 * According to RFC961 (Assigned Protocols), 2838 * the urgent pointer points to the last octet 2839 * of urgent data. We continue, however, 2840 * to consider it to indicate the first octet 2841 * of data past the urgent section as the original 2842 * spec states (in one of two places). 2843 */ 2844 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2845 tp->rcv_up = th->th_seq + th->th_urp; 2846 so->so_oobmark = so->so_rcv.sb_cc + 2847 (tp->rcv_up - tp->rcv_nxt) - 1; 2848 if (so->so_oobmark == 0) 2849 so->so_rcv.sb_state |= SBS_RCVATMARK; 2850 sohasoutofband(so); 2851 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2852 } 2853 SOCKBUF_UNLOCK(&so->so_rcv); 2854 /* 2855 * Remove out of band data so doesn't get presented to user. 2856 * This can happen independent of advancing the URG pointer, 2857 * but if two URG's are pending at once, some out-of-band 2858 * data may creep in... ick. 2859 */ 2860 if (th->th_urp <= (u_long)tlen && 2861 !(so->so_options & SO_OOBINLINE)) { 2862 /* hdr drop is delayed */ 2863 tcp_pulloutofband(so, th, m, drop_hdrlen); 2864 } 2865 } else { 2866 /* 2867 * If no out of band data is expected, 2868 * pull receive urgent pointer along 2869 * with the receive window. 2870 */ 2871 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2872 tp->rcv_up = tp->rcv_nxt; 2873 } 2874dodata: /* XXX */ 2875 INP_WLOCK_ASSERT(tp->t_inpcb); 2876 2877 /* 2878 * Process the segment text, merging it into the TCP sequencing queue, 2879 * and arranging for acknowledgment of receipt if necessary. 2880 * This process logically involves adjusting tp->rcv_wnd as data 2881 * is presented to the user (this happens in tcp_usrreq.c, 2882 * case PRU_RCVD). If a FIN has already been received on this 2883 * connection then we just ignore the text. 2884 */ 2885 if ((tlen || (thflags & TH_FIN)) && 2886 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2887 tcp_seq save_start = th->th_seq; 2888 m_adj(m, drop_hdrlen); /* delayed header drop */ 2889 /* 2890 * Insert segment which includes th into TCP reassembly queue 2891 * with control block tp. Set thflags to whether reassembly now 2892 * includes a segment with FIN. This handles the common case 2893 * inline (segment is the next to be received on an established 2894 * connection, and the queue is empty), avoiding linkage into 2895 * and removal from the queue and repetition of various 2896 * conversions. 2897 * Set DELACK for segments received in order, but ack 2898 * immediately when segments are out of order (so 2899 * fast retransmit can work). 2900 */ 2901 if (th->th_seq == tp->rcv_nxt && 2902 LIST_EMPTY(&tp->t_segq) && 2903 TCPS_HAVEESTABLISHED(tp->t_state)) { 2904 if (DELAY_ACK(tp)) 2905 tp->t_flags |= TF_DELACK; 2906 else 2907 tp->t_flags |= TF_ACKNOW; 2908 tp->rcv_nxt += tlen; 2909 thflags = th->th_flags & TH_FIN; 2910 TCPSTAT_INC(tcps_rcvpack); 2911 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2912 ND6_HINT(tp); 2913 SOCKBUF_LOCK(&so->so_rcv); 2914 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2915 m_freem(m); 2916 else 2917 sbappendstream_locked(&so->so_rcv, m); 2918 /* NB: sorwakeup_locked() does an implicit unlock. */ 2919 sorwakeup_locked(so); 2920 } else { 2921 /* 2922 * XXX: Due to the header drop above "th" is 2923 * theoretically invalid by now. Fortunately 2924 * m_adj() doesn't actually frees any mbufs 2925 * when trimming from the head. 2926 */ 2927 thflags = tcp_reass(tp, th, &tlen, m); 2928 tp->t_flags |= TF_ACKNOW; 2929 } 2930 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2931 tcp_update_sack_list(tp, save_start, save_start + tlen); 2932#if 0 2933 /* 2934 * Note the amount of data that peer has sent into 2935 * our window, in order to estimate the sender's 2936 * buffer size. 2937 * XXX: Unused. 2938 */ 2939 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2940 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2941 else 2942 len = so->so_rcv.sb_hiwat; 2943#endif 2944 } else { 2945 m_freem(m); 2946 thflags &= ~TH_FIN; 2947 } 2948 2949 /* 2950 * If FIN is received ACK the FIN and let the user know 2951 * that the connection is closing. 2952 */ 2953 if (thflags & TH_FIN) { 2954 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2955 socantrcvmore(so); 2956 /* 2957 * If connection is half-synchronized 2958 * (ie NEEDSYN flag on) then delay ACK, 2959 * so it may be piggybacked when SYN is sent. 2960 * Otherwise, since we received a FIN then no 2961 * more input can be expected, send ACK now. 2962 */ 2963 if (tp->t_flags & TF_NEEDSYN) 2964 tp->t_flags |= TF_DELACK; 2965 else 2966 tp->t_flags |= TF_ACKNOW; 2967 tp->rcv_nxt++; 2968 } 2969 switch (tp->t_state) { 2970 2971 /* 2972 * In SYN_RECEIVED and ESTABLISHED STATES 2973 * enter the CLOSE_WAIT state. 2974 */ 2975 case TCPS_SYN_RECEIVED: 2976 tp->t_starttime = ticks; 2977 /* FALLTHROUGH */ 2978 case TCPS_ESTABLISHED: 2979 tp->t_state = TCPS_CLOSE_WAIT; 2980 break; 2981 2982 /* 2983 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2984 * enter the CLOSING state. 2985 */ 2986 case TCPS_FIN_WAIT_1: 2987 tp->t_state = TCPS_CLOSING; 2988 break; 2989 2990 /* 2991 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2992 * starting the time-wait timer, turning off the other 2993 * standard timers. 2994 */ 2995 case TCPS_FIN_WAIT_2: 2996 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2997 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2998 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2999 ti_locked)); 3000 3001 tcp_twstart(tp); 3002 INP_INFO_WUNLOCK(&V_tcbinfo); 3003 return; 3004 } 3005 } 3006 if (ti_locked == TI_WLOCKED) 3007 INP_INFO_WUNLOCK(&V_tcbinfo); 3008 ti_locked = TI_UNLOCKED; 3009 3010#ifdef TCPDEBUG 3011 if (so->so_options & SO_DEBUG) 3012 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3013 &tcp_savetcp, 0); 3014#endif 3015 3016 /* 3017 * Return any desired output. 3018 */ 3019 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3020 (void) tcp_output(tp); 3021 3022check_delack: 3023 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3024 __func__, ti_locked)); 3025 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3026 INP_WLOCK_ASSERT(tp->t_inpcb); 3027 3028 if (tp->t_flags & TF_DELACK) { 3029 tp->t_flags &= ~TF_DELACK; 3030 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3031 } 3032 INP_WUNLOCK(tp->t_inpcb); 3033 return; 3034 3035dropafterack: 3036 /* 3037 * Generate an ACK dropping incoming segment if it occupies 3038 * sequence space, where the ACK reflects our state. 3039 * 3040 * We can now skip the test for the RST flag since all 3041 * paths to this code happen after packets containing 3042 * RST have been dropped. 3043 * 3044 * In the SYN-RECEIVED state, don't send an ACK unless the 3045 * segment we received passes the SYN-RECEIVED ACK test. 3046 * If it fails send a RST. This breaks the loop in the 3047 * "LAND" DoS attack, and also prevents an ACK storm 3048 * between two listening ports that have been sent forged 3049 * SYN segments, each with the source address of the other. 3050 */ 3051 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3052 (SEQ_GT(tp->snd_una, th->th_ack) || 3053 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3054 rstreason = BANDLIM_RST_OPENPORT; 3055 goto dropwithreset; 3056 } 3057#ifdef TCPDEBUG 3058 if (so->so_options & SO_DEBUG) 3059 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3060 &tcp_savetcp, 0); 3061#endif 3062 if (ti_locked == TI_WLOCKED) 3063 INP_INFO_WUNLOCK(&V_tcbinfo); 3064 ti_locked = TI_UNLOCKED; 3065 3066 tp->t_flags |= TF_ACKNOW; 3067 (void) tcp_output(tp); 3068 INP_WUNLOCK(tp->t_inpcb); 3069 m_freem(m); 3070 return; 3071 3072dropwithreset: 3073 if (ti_locked == TI_WLOCKED) 3074 INP_INFO_WUNLOCK(&V_tcbinfo); 3075 ti_locked = TI_UNLOCKED; 3076 3077 if (tp != NULL) { 3078 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3079 INP_WUNLOCK(tp->t_inpcb); 3080 } else 3081 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3082 return; 3083 3084drop: 3085 if (ti_locked == TI_WLOCKED) { 3086 INP_INFO_WUNLOCK(&V_tcbinfo); 3087 ti_locked = TI_UNLOCKED; 3088 } 3089#ifdef INVARIANTS 3090 else 3091 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3092#endif 3093 3094 /* 3095 * Drop space held by incoming segment and return. 3096 */ 3097#ifdef TCPDEBUG 3098 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3099 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3100 &tcp_savetcp, 0); 3101#endif 3102 if (tp != NULL) 3103 INP_WUNLOCK(tp->t_inpcb); 3104 m_freem(m); 3105} 3106 3107/* 3108 * Issue RST and make ACK acceptable to originator of segment. 3109 * The mbuf must still include the original packet header. 3110 * tp may be NULL. 3111 */ 3112static void 3113tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3114 int tlen, int rstreason) 3115{ 3116#ifdef INET 3117 struct ip *ip; 3118#endif 3119#ifdef INET6 3120 struct ip6_hdr *ip6; 3121#endif 3122 3123 if (tp != NULL) { 3124 INP_WLOCK_ASSERT(tp->t_inpcb); 3125 } 3126 3127 /* Don't bother if destination was broadcast/multicast. */ 3128 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3129 goto drop; 3130#ifdef INET6 3131 if (mtod(m, struct ip *)->ip_v == 6) { 3132 ip6 = mtod(m, struct ip6_hdr *); 3133 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3134 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3135 goto drop; 3136 /* IPv6 anycast check is done at tcp6_input() */ 3137 } 3138#endif 3139#if defined(INET) && defined(INET6) 3140 else 3141#endif 3142#ifdef INET 3143 { 3144 ip = mtod(m, struct ip *); 3145 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3146 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3147 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3148 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3149 goto drop; 3150 } 3151#endif 3152 3153 /* Perform bandwidth limiting. */ 3154 if (badport_bandlim(rstreason) < 0) 3155 goto drop; 3156 3157 /* tcp_respond consumes the mbuf chain. */ 3158 if (th->th_flags & TH_ACK) { 3159 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3160 th->th_ack, TH_RST); 3161 } else { 3162 if (th->th_flags & TH_SYN) 3163 tlen++; 3164 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3165 (tcp_seq)0, TH_RST|TH_ACK); 3166 } 3167 return; 3168drop: 3169 m_freem(m); 3170} 3171 3172/* 3173 * Parse TCP options and place in tcpopt. 3174 */ 3175static void 3176tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3177{ 3178 int opt, optlen; 3179 3180 to->to_flags = 0; 3181 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3182 opt = cp[0]; 3183 if (opt == TCPOPT_EOL) 3184 break; 3185 if (opt == TCPOPT_NOP) 3186 optlen = 1; 3187 else { 3188 if (cnt < 2) 3189 break; 3190 optlen = cp[1]; 3191 if (optlen < 2 || optlen > cnt) 3192 break; 3193 } 3194 switch (opt) { 3195 case TCPOPT_MAXSEG: 3196 if (optlen != TCPOLEN_MAXSEG) 3197 continue; 3198 if (!(flags & TO_SYN)) 3199 continue; 3200 to->to_flags |= TOF_MSS; 3201 bcopy((char *)cp + 2, 3202 (char *)&to->to_mss, sizeof(to->to_mss)); 3203 to->to_mss = ntohs(to->to_mss); 3204 break; 3205 case TCPOPT_WINDOW: 3206 if (optlen != TCPOLEN_WINDOW) 3207 continue; 3208 if (!(flags & TO_SYN)) 3209 continue; 3210 to->to_flags |= TOF_SCALE; 3211 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3212 break; 3213 case TCPOPT_TIMESTAMP: 3214 if (optlen != TCPOLEN_TIMESTAMP) 3215 continue; 3216 to->to_flags |= TOF_TS; 3217 bcopy((char *)cp + 2, 3218 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3219 to->to_tsval = ntohl(to->to_tsval); 3220 bcopy((char *)cp + 6, 3221 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3222 to->to_tsecr = ntohl(to->to_tsecr); 3223 break; 3224#ifdef TCP_SIGNATURE 3225 /* 3226 * XXX In order to reply to a host which has set the 3227 * TCP_SIGNATURE option in its initial SYN, we have to 3228 * record the fact that the option was observed here 3229 * for the syncache code to perform the correct response. 3230 */ 3231 case TCPOPT_SIGNATURE: 3232 if (optlen != TCPOLEN_SIGNATURE) 3233 continue; 3234 to->to_flags |= TOF_SIGNATURE; 3235 to->to_signature = cp + 2; 3236 break; 3237#endif 3238 case TCPOPT_SACK_PERMITTED: 3239 if (optlen != TCPOLEN_SACK_PERMITTED) 3240 continue; 3241 if (!(flags & TO_SYN)) 3242 continue; 3243 if (!V_tcp_do_sack) 3244 continue; 3245 to->to_flags |= TOF_SACKPERM; 3246 break; 3247 case TCPOPT_SACK: 3248 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3249 continue; 3250 if (flags & TO_SYN) 3251 continue; 3252 to->to_flags |= TOF_SACK; 3253 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3254 to->to_sacks = cp + 2; 3255 TCPSTAT_INC(tcps_sack_rcv_blocks); 3256 break; 3257 default: 3258 continue; 3259 } 3260 } 3261} 3262 3263/* 3264 * Pull out of band byte out of a segment so 3265 * it doesn't appear in the user's data queue. 3266 * It is still reflected in the segment length for 3267 * sequencing purposes. 3268 */ 3269static void 3270tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3271 int off) 3272{ 3273 int cnt = off + th->th_urp - 1; 3274 3275 while (cnt >= 0) { 3276 if (m->m_len > cnt) { 3277 char *cp = mtod(m, caddr_t) + cnt; 3278 struct tcpcb *tp = sototcpcb(so); 3279 3280 INP_WLOCK_ASSERT(tp->t_inpcb); 3281 3282 tp->t_iobc = *cp; 3283 tp->t_oobflags |= TCPOOB_HAVEDATA; 3284 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3285 m->m_len--; 3286 if (m->m_flags & M_PKTHDR) 3287 m->m_pkthdr.len--; 3288 return; 3289 } 3290 cnt -= m->m_len; 3291 m = m->m_next; 3292 if (m == NULL) 3293 break; 3294 } 3295 panic("tcp_pulloutofband"); 3296} 3297 3298/* 3299 * Collect new round-trip time estimate 3300 * and update averages and current timeout. 3301 */ 3302static void 3303tcp_xmit_timer(struct tcpcb *tp, int rtt) 3304{ 3305 int delta; 3306 3307 INP_WLOCK_ASSERT(tp->t_inpcb); 3308 3309 TCPSTAT_INC(tcps_rttupdated); 3310 tp->t_rttupdated++; 3311 if (tp->t_srtt != 0) { 3312 /* 3313 * srtt is stored as fixed point with 5 bits after the 3314 * binary point (i.e., scaled by 8). The following magic 3315 * is equivalent to the smoothing algorithm in rfc793 with 3316 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3317 * point). Adjust rtt to origin 0. 3318 */ 3319 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3320 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3321 3322 if ((tp->t_srtt += delta) <= 0) 3323 tp->t_srtt = 1; 3324 3325 /* 3326 * We accumulate a smoothed rtt variance (actually, a 3327 * smoothed mean difference), then set the retransmit 3328 * timer to smoothed rtt + 4 times the smoothed variance. 3329 * rttvar is stored as fixed point with 4 bits after the 3330 * binary point (scaled by 16). The following is 3331 * equivalent to rfc793 smoothing with an alpha of .75 3332 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3333 * rfc793's wired-in beta. 3334 */ 3335 if (delta < 0) 3336 delta = -delta; 3337 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3338 if ((tp->t_rttvar += delta) <= 0) 3339 tp->t_rttvar = 1; 3340 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3341 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3342 } else { 3343 /* 3344 * No rtt measurement yet - use the unsmoothed rtt. 3345 * Set the variance to half the rtt (so our first 3346 * retransmit happens at 3*rtt). 3347 */ 3348 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3349 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3350 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3351 } 3352 tp->t_rtttime = 0; 3353 tp->t_rxtshift = 0; 3354 3355 /* 3356 * the retransmit should happen at rtt + 4 * rttvar. 3357 * Because of the way we do the smoothing, srtt and rttvar 3358 * will each average +1/2 tick of bias. When we compute 3359 * the retransmit timer, we want 1/2 tick of rounding and 3360 * 1 extra tick because of +-1/2 tick uncertainty in the 3361 * firing of the timer. The bias will give us exactly the 3362 * 1.5 tick we need. But, because the bias is 3363 * statistical, we have to test that we don't drop below 3364 * the minimum feasible timer (which is 2 ticks). 3365 */ 3366 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3367 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3368 3369 /* 3370 * We received an ack for a packet that wasn't retransmitted; 3371 * it is probably safe to discard any error indications we've 3372 * received recently. This isn't quite right, but close enough 3373 * for now (a route might have failed after we sent a segment, 3374 * and the return path might not be symmetrical). 3375 */ 3376 tp->t_softerror = 0; 3377} 3378 3379/* 3380 * Determine a reasonable value for maxseg size. 3381 * If the route is known, check route for mtu. 3382 * If none, use an mss that can be handled on the outgoing interface 3383 * without forcing IP to fragment. If no route is found, route has no mtu, 3384 * or the destination isn't local, use a default, hopefully conservative 3385 * size (usually 512 or the default IP max size, but no more than the mtu 3386 * of the interface), as we can't discover anything about intervening 3387 * gateways or networks. We also initialize the congestion/slow start 3388 * window to be a single segment if the destination isn't local. 3389 * While looking at the routing entry, we also initialize other path-dependent 3390 * parameters from pre-set or cached values in the routing entry. 3391 * 3392 * Also take into account the space needed for options that we 3393 * send regularly. Make maxseg shorter by that amount to assure 3394 * that we can send maxseg amount of data even when the options 3395 * are present. Store the upper limit of the length of options plus 3396 * data in maxopd. 3397 * 3398 * NOTE that this routine is only called when we process an incoming 3399 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3400 * settings are handled in tcp_mssopt(). 3401 */ 3402void 3403tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3404 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3405{ 3406 int mss = 0; 3407 u_long maxmtu = 0; 3408 struct inpcb *inp = tp->t_inpcb; 3409 struct hc_metrics_lite metrics; 3410 int origoffer; 3411#ifdef INET6 3412 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3413 size_t min_protoh = isipv6 ? 3414 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3415 sizeof (struct tcpiphdr); 3416#else 3417 const size_t min_protoh = sizeof(struct tcpiphdr); 3418#endif 3419 3420 INP_WLOCK_ASSERT(tp->t_inpcb); 3421 3422 if (mtuoffer != -1) { 3423 KASSERT(offer == -1, ("%s: conflict", __func__)); 3424 offer = mtuoffer - min_protoh; 3425 } 3426 origoffer = offer; 3427 3428 /* Initialize. */ 3429#ifdef INET6 3430 if (isipv6) { 3431 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3432 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3433 } 3434#endif 3435#if defined(INET) && defined(INET6) 3436 else 3437#endif 3438#ifdef INET 3439 { 3440 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3441 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3442 } 3443#endif 3444 3445 /* 3446 * No route to sender, stay with default mss and return. 3447 */ 3448 if (maxmtu == 0) { 3449 /* 3450 * In case we return early we need to initialize metrics 3451 * to a defined state as tcp_hc_get() would do for us 3452 * if there was no cache hit. 3453 */ 3454 if (metricptr != NULL) 3455 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3456 return; 3457 } 3458 3459 /* What have we got? */ 3460 switch (offer) { 3461 case 0: 3462 /* 3463 * Offer == 0 means that there was no MSS on the SYN 3464 * segment, in this case we use tcp_mssdflt as 3465 * already assigned to t_maxopd above. 3466 */ 3467 offer = tp->t_maxopd; 3468 break; 3469 3470 case -1: 3471 /* 3472 * Offer == -1 means that we didn't receive SYN yet. 3473 */ 3474 /* FALLTHROUGH */ 3475 3476 default: 3477 /* 3478 * Prevent DoS attack with too small MSS. Round up 3479 * to at least minmss. 3480 */ 3481 offer = max(offer, V_tcp_minmss); 3482 } 3483 3484 /* 3485 * rmx information is now retrieved from tcp_hostcache. 3486 */ 3487 tcp_hc_get(&inp->inp_inc, &metrics); 3488 if (metricptr != NULL) 3489 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3490 3491 /* 3492 * If there's a discovered mtu int tcp hostcache, use it 3493 * else, use the link mtu. 3494 */ 3495 if (metrics.rmx_mtu) 3496 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3497 else { 3498#ifdef INET6 3499 if (isipv6) { 3500 mss = maxmtu - min_protoh; 3501 if (!V_path_mtu_discovery && 3502 !in6_localaddr(&inp->in6p_faddr)) 3503 mss = min(mss, V_tcp_v6mssdflt); 3504 } 3505#endif 3506#if defined(INET) && defined(INET6) 3507 else 3508#endif 3509#ifdef INET 3510 { 3511 mss = maxmtu - min_protoh; 3512 if (!V_path_mtu_discovery && 3513 !in_localaddr(inp->inp_faddr)) 3514 mss = min(mss, V_tcp_mssdflt); 3515 } 3516#endif 3517 /* 3518 * XXX - The above conditional (mss = maxmtu - min_protoh) 3519 * probably violates the TCP spec. 3520 * The problem is that, since we don't know the 3521 * other end's MSS, we are supposed to use a conservative 3522 * default. But, if we do that, then MTU discovery will 3523 * never actually take place, because the conservative 3524 * default is much less than the MTUs typically seen 3525 * on the Internet today. For the moment, we'll sweep 3526 * this under the carpet. 3527 * 3528 * The conservative default might not actually be a problem 3529 * if the only case this occurs is when sending an initial 3530 * SYN with options and data to a host we've never talked 3531 * to before. Then, they will reply with an MSS value which 3532 * will get recorded and the new parameters should get 3533 * recomputed. For Further Study. 3534 */ 3535 } 3536 mss = min(mss, offer); 3537 3538 /* 3539 * Sanity check: make sure that maxopd will be large 3540 * enough to allow some data on segments even if the 3541 * all the option space is used (40bytes). Otherwise 3542 * funny things may happen in tcp_output. 3543 */ 3544 mss = max(mss, 64); 3545 3546 /* 3547 * maxopd stores the maximum length of data AND options 3548 * in a segment; maxseg is the amount of data in a normal 3549 * segment. We need to store this value (maxopd) apart 3550 * from maxseg, because now every segment carries options 3551 * and thus we normally have somewhat less data in segments. 3552 */ 3553 tp->t_maxopd = mss; 3554 3555 /* 3556 * origoffer==-1 indicates that no segments were received yet. 3557 * In this case we just guess. 3558 */ 3559 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3560 (origoffer == -1 || 3561 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3562 mss -= TCPOLEN_TSTAMP_APPA; 3563 3564 tp->t_maxseg = mss; 3565} 3566 3567void 3568tcp_mss(struct tcpcb *tp, int offer) 3569{ 3570 int mss; 3571 u_long bufsize; 3572 struct inpcb *inp; 3573 struct socket *so; 3574 struct hc_metrics_lite metrics; 3575 struct tcp_ifcap cap; 3576 3577 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3578 3579 bzero(&cap, sizeof(cap)); 3580 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3581 3582 mss = tp->t_maxseg; 3583 inp = tp->t_inpcb; 3584 3585 /* 3586 * If there's a pipesize, change the socket buffer to that size, 3587 * don't change if sb_hiwat is different than default (then it 3588 * has been changed on purpose with setsockopt). 3589 * Make the socket buffers an integral number of mss units; 3590 * if the mss is larger than the socket buffer, decrease the mss. 3591 */ 3592 so = inp->inp_socket; 3593 SOCKBUF_LOCK(&so->so_snd); 3594 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3595 bufsize = metrics.rmx_sendpipe; 3596 else 3597 bufsize = so->so_snd.sb_hiwat; 3598 if (bufsize < mss) 3599 mss = bufsize; 3600 else { 3601 bufsize = roundup(bufsize, mss); 3602 if (bufsize > sb_max) 3603 bufsize = sb_max; 3604 if (bufsize > so->so_snd.sb_hiwat) 3605 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3606 } 3607 SOCKBUF_UNLOCK(&so->so_snd); 3608 tp->t_maxseg = mss; 3609 3610 SOCKBUF_LOCK(&so->so_rcv); 3611 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3612 bufsize = metrics.rmx_recvpipe; 3613 else 3614 bufsize = so->so_rcv.sb_hiwat; 3615 if (bufsize > mss) { 3616 bufsize = roundup(bufsize, mss); 3617 if (bufsize > sb_max) 3618 bufsize = sb_max; 3619 if (bufsize > so->so_rcv.sb_hiwat) 3620 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3621 } 3622 SOCKBUF_UNLOCK(&so->so_rcv); 3623 3624 /* Check the interface for TSO capabilities. */ 3625 if (cap.ifcap & CSUM_TSO) { 3626 tp->t_flags |= TF_TSO; 3627 tp->t_tsomax = cap.tsomax; 3628 } 3629} 3630 3631/* 3632 * Determine the MSS option to send on an outgoing SYN. 3633 */ 3634int 3635tcp_mssopt(struct in_conninfo *inc) 3636{ 3637 int mss = 0; 3638 u_long maxmtu = 0; 3639 u_long thcmtu = 0; 3640 size_t min_protoh; 3641 3642 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3643 3644#ifdef INET6 3645 if (inc->inc_flags & INC_ISIPV6) { 3646 mss = V_tcp_v6mssdflt; 3647 maxmtu = tcp_maxmtu6(inc, NULL); 3648 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3649 } 3650#endif 3651#if defined(INET) && defined(INET6) 3652 else 3653#endif 3654#ifdef INET 3655 { 3656 mss = V_tcp_mssdflt; 3657 maxmtu = tcp_maxmtu(inc, NULL); 3658 min_protoh = sizeof(struct tcpiphdr); 3659 } 3660#endif 3661#if defined(INET6) || defined(INET) 3662 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3663#endif 3664 3665 if (maxmtu && thcmtu) 3666 mss = min(maxmtu, thcmtu) - min_protoh; 3667 else if (maxmtu || thcmtu) 3668 mss = max(maxmtu, thcmtu) - min_protoh; 3669 3670 return (mss); 3671} 3672 3673 3674/* 3675 * On a partial ack arrives, force the retransmission of the 3676 * next unacknowledged segment. Do not clear tp->t_dupacks. 3677 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3678 * be started again. 3679 */ 3680static void 3681tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3682{ 3683 tcp_seq onxt = tp->snd_nxt; 3684 u_long ocwnd = tp->snd_cwnd; 3685 3686 INP_WLOCK_ASSERT(tp->t_inpcb); 3687 3688 tcp_timer_activate(tp, TT_REXMT, 0); 3689 tp->t_rtttime = 0; 3690 tp->snd_nxt = th->th_ack; 3691 /* 3692 * Set snd_cwnd to one segment beyond acknowledged offset. 3693 * (tp->snd_una has not yet been updated when this function is called.) 3694 */ 3695 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3696 tp->t_flags |= TF_ACKNOW; 3697 (void) tcp_output(tp); 3698 tp->snd_cwnd = ocwnd; 3699 if (SEQ_GT(onxt, tp->snd_nxt)) 3700 tp->snd_nxt = onxt; 3701 /* 3702 * Partial window deflation. Relies on fact that tp->snd_una 3703 * not updated yet. 3704 */ 3705 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3706 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3707 else 3708 tp->snd_cwnd = 0; 3709 tp->snd_cwnd += tp->t_maxseg; 3710} 3711