tcp_input.c revision 317375
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: stable/10/sys/netinet/tcp_input.c 317375 2017-04-24 16:31:28Z smh $"); 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_kdtrace.h" 58#include "opt_tcpdebug.h" 59 60#include <sys/param.h> 61#include <sys/kernel.h> 62#include <sys/hhook.h> 63#include <sys/malloc.h> 64#include <sys/mbuf.h> 65#include <sys/proc.h> /* for proc0 declaration */ 66#include <sys/protosw.h> 67#include <sys/sdt.h> 68#include <sys/signalvar.h> 69#include <sys/socket.h> 70#include <sys/socketvar.h> 71#include <sys/sysctl.h> 72#include <sys/syslog.h> 73#include <sys/systm.h> 74 75#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77#include <vm/uma.h> 78 79#include <net/if.h> 80#include <net/route.h> 81#include <net/vnet.h> 82 83#define TCPSTATES /* for logging */ 84 85#include <netinet/cc.h> 86#include <netinet/in.h> 87#include <netinet/in_kdtrace.h> 88#include <netinet/in_pcb.h> 89#include <netinet/in_systm.h> 90#include <netinet/in_var.h> 91#include <netinet/ip.h> 92#include <netinet/ip_icmp.h> /* required for icmp_var.h */ 93#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 94#include <netinet/ip_var.h> 95#include <netinet/ip_options.h> 96#include <netinet/ip6.h> 97#include <netinet/icmp6.h> 98#include <netinet6/in6_pcb.h> 99#include <netinet6/ip6_var.h> 100#include <netinet6/nd6.h> 101#ifdef TCP_RFC7413 102#include <netinet/tcp_fastopen.h> 103#endif 104#include <netinet/tcp_fsm.h> 105#include <netinet/tcp_seq.h> 106#include <netinet/tcp_timer.h> 107#include <netinet/tcp_var.h> 108#include <netinet6/tcp6_var.h> 109#include <netinet/tcpip.h> 110#include <netinet/tcp_syncache.h> 111#ifdef TCPDEBUG 112#include <netinet/tcp_debug.h> 113#endif /* TCPDEBUG */ 114#ifdef TCP_OFFLOAD 115#include <netinet/tcp_offload.h> 116#endif 117 118#ifdef IPSEC 119#include <netipsec/ipsec.h> 120#include <netipsec/ipsec6.h> 121#endif /*IPSEC*/ 122 123#include <machine/in_cksum.h> 124 125#include <security/mac/mac_framework.h> 126 127const int tcprexmtthresh = 3; 128 129int tcp_log_in_vain = 0; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 131 &tcp_log_in_vain, 0, 132 "Log all incoming TCP segments to closed ports"); 133 134VNET_DEFINE(int, blackhole) = 0; 135#define V_blackhole VNET(blackhole) 136SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 137 &VNET_NAME(blackhole), 0, 138 "Do not send RST on segments to closed ports"); 139 140VNET_DEFINE(int, tcp_delack_enabled) = 1; 141SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 142 &VNET_NAME(tcp_delack_enabled), 0, 143 "Delay ACK to try and piggyback it onto a data packet"); 144 145VNET_DEFINE(int, drop_synfin) = 0; 146#define V_drop_synfin VNET(drop_synfin) 147SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 148 &VNET_NAME(drop_synfin), 0, 149 "Drop TCP packets with SYN+FIN set"); 150 151VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 152SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_pipe, CTLFLAG_VNET | CTLFLAG_RW, 153 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 154 "Use calculated pipe/in-flight bytes per RFC 6675"); 155 156VNET_DEFINE(int, tcp_do_rfc3042) = 1; 157#define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 158SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 159 &VNET_NAME(tcp_do_rfc3042), 0, 160 "Enable RFC 3042 (Limited Transmit)"); 161 162VNET_DEFINE(int, tcp_do_rfc3390) = 1; 163SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 164 &VNET_NAME(tcp_do_rfc3390), 0, 165 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 166 167SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 168 "Experimental TCP extensions"); 169 170VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 171SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 172 &VNET_NAME(tcp_do_initcwnd10), 0, 173 "Enable RFC 6928 (Increasing initial CWND to 10)"); 174 175VNET_DEFINE(int, tcp_do_rfc3465) = 1; 176SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 177 &VNET_NAME(tcp_do_rfc3465), 0, 178 "Enable RFC 3465 (Appropriate Byte Counting)"); 179 180VNET_DEFINE(int, tcp_abc_l_var) = 2; 181SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 182 &VNET_NAME(tcp_abc_l_var), 2, 183 "Cap the max cwnd increment during slow-start to this number of segments"); 184 185static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 186 187VNET_DEFINE(int, tcp_do_ecn) = 2; 188SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 189 &VNET_NAME(tcp_do_ecn), 0, 190 "TCP ECN support"); 191 192VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 193SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 194 &VNET_NAME(tcp_ecn_maxretries), 0, 195 "Max retries before giving up on ECN"); 196 197VNET_DEFINE(int, tcp_insecure_rst) = 0; 198#define V_tcp_insecure_rst VNET(tcp_insecure_rst) 199SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 200 &VNET_NAME(tcp_insecure_rst), 0, 201 "Follow the old (insecure) criteria for accepting RST packets"); 202 203VNET_DEFINE(int, tcp_recvspace) = 1024*64; 204#define V_tcp_recvspace VNET(tcp_recvspace) 205SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 206 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 207 208VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 209#define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 210SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 211 &VNET_NAME(tcp_do_autorcvbuf), 0, 212 "Enable automatic receive buffer sizing"); 213 214VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 215#define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 216SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 217 &VNET_NAME(tcp_autorcvbuf_inc), 0, 218 "Incrementor step size of automatic receive buffer"); 219 220VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 221#define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 222SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 223 &VNET_NAME(tcp_autorcvbuf_max), 0, 224 "Max size of automatic receive buffer"); 225 226VNET_DEFINE(struct inpcbhead, tcb); 227#define tcb6 tcb /* for KAME src sync over BSD*'s */ 228VNET_DEFINE(struct inpcbinfo, tcbinfo); 229 230static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 231static void tcp_do_segment(struct mbuf *, struct tcphdr *, 232 struct socket *, struct tcpcb *, int, int, uint8_t, 233 int); 234static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 235 struct tcpcb *, int, int); 236static void tcp_pulloutofband(struct socket *, 237 struct tcphdr *, struct mbuf *, int); 238static void tcp_xmit_timer(struct tcpcb *, int); 239static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 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 * TCP statistics are stored in an "array" of counter(9)s. 249 */ 250VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 251VNET_PCPUSTAT_SYSINIT(tcpstat); 252SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 253 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 254 255#ifdef VIMAGE 256VNET_PCPUSTAT_SYSUNINIT(tcpstat); 257#endif /* VIMAGE */ 258/* 259 * Kernel module interface for updating tcpstat. The argument is an index 260 * into tcpstat treated as an array. 261 */ 262void 263kmod_tcpstat_inc(int statnum) 264{ 265 266 counter_u64_add(VNET(tcpstat)[statnum], 1); 267} 268 269/* 270 * Wrapper for the TCP established input helper hook. 271 */ 272static void inline 273hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 274{ 275 struct tcp_hhook_data hhook_data; 276 277 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 278 hhook_data.tp = tp; 279 hhook_data.th = th; 280 hhook_data.to = to; 281 282 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 283 tp->osd); 284 } 285} 286 287/* 288 * CC wrapper hook functions 289 */ 290static void inline 291cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 292{ 293 INP_WLOCK_ASSERT(tp->t_inpcb); 294 295 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 296 if (tp->snd_cwnd <= tp->snd_wnd) 297 tp->ccv->flags |= CCF_CWND_LIMITED; 298 else 299 tp->ccv->flags &= ~CCF_CWND_LIMITED; 300 301 if (type == CC_ACK) { 302 if (tp->snd_cwnd > tp->snd_ssthresh) { 303 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 304 V_tcp_abc_l_var * tp->t_maxseg); 305 if (tp->t_bytes_acked >= tp->snd_cwnd) { 306 tp->t_bytes_acked -= tp->snd_cwnd; 307 tp->ccv->flags |= CCF_ABC_SENTAWND; 308 } 309 } else { 310 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 311 tp->t_bytes_acked = 0; 312 } 313 } 314 315 if (CC_ALGO(tp)->ack_received != NULL) { 316 /* XXXLAS: Find a way to live without this */ 317 tp->ccv->curack = th->th_ack; 318 CC_ALGO(tp)->ack_received(tp->ccv, type); 319 } 320} 321 322static void inline 323cc_conn_init(struct tcpcb *tp) 324{ 325 struct hc_metrics_lite metrics; 326 struct inpcb *inp = tp->t_inpcb; 327 int rtt; 328 329 INP_WLOCK_ASSERT(tp->t_inpcb); 330 331 tcp_hc_get(&inp->inp_inc, &metrics); 332 333 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 334 tp->t_srtt = rtt; 335 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 336 TCPSTAT_INC(tcps_usedrtt); 337 if (metrics.rmx_rttvar) { 338 tp->t_rttvar = metrics.rmx_rttvar; 339 TCPSTAT_INC(tcps_usedrttvar); 340 } else { 341 /* default variation is +- 1 rtt */ 342 tp->t_rttvar = 343 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 344 } 345 TCPT_RANGESET(tp->t_rxtcur, 346 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 347 tp->t_rttmin, TCPTV_REXMTMAX); 348 } 349 if (metrics.rmx_ssthresh) { 350 /* 351 * There's some sort of gateway or interface 352 * buffer limit on the path. Use this to set 353 * the slow start threshhold, but set the 354 * threshold to no less than 2*mss. 355 */ 356 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 357 TCPSTAT_INC(tcps_usedssthresh); 358 } 359 360 /* 361 * Set the initial slow-start flight size. 362 * 363 * RFC5681 Section 3.1 specifies the default conservative values. 364 * RFC3390 specifies slightly more aggressive values. 365 * RFC6928 increases it to ten segments. 366 * 367 * If a SYN or SYN/ACK was lost and retransmitted, we have to 368 * reduce the initial CWND to one segment as congestion is likely 369 * requiring us to be cautious. 370 */ 371 if (tp->snd_cwnd == 1) 372 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 373 else if (V_tcp_do_initcwnd10) 374 tp->snd_cwnd = min(10 * tp->t_maxseg, 375 max(2 * tp->t_maxseg, 14600)); 376 else if (V_tcp_do_rfc3390) 377 tp->snd_cwnd = min(4 * tp->t_maxseg, 378 max(2 * tp->t_maxseg, 4380)); 379 else { 380 /* Per RFC5681 Section 3.1 */ 381 if (tp->t_maxseg > 2190) 382 tp->snd_cwnd = 2 * tp->t_maxseg; 383 else if (tp->t_maxseg > 1095) 384 tp->snd_cwnd = 3 * tp->t_maxseg; 385 else 386 tp->snd_cwnd = 4 * tp->t_maxseg; 387 } 388 389 if (CC_ALGO(tp)->conn_init != NULL) 390 CC_ALGO(tp)->conn_init(tp->ccv); 391} 392 393void inline 394cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 395{ 396 INP_WLOCK_ASSERT(tp->t_inpcb); 397 398 switch(type) { 399 case CC_NDUPACK: 400 if (!IN_FASTRECOVERY(tp->t_flags)) { 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_ECN: 407 if (!IN_CONGRECOVERY(tp->t_flags)) { 408 TCPSTAT_INC(tcps_ecn_rcwnd); 409 tp->snd_recover = tp->snd_max; 410 if (tp->t_flags & TF_ECN_PERMIT) 411 tp->t_flags |= TF_ECN_SND_CWR; 412 } 413 break; 414 case CC_RTO: 415 tp->t_dupacks = 0; 416 tp->t_bytes_acked = 0; 417 EXIT_RECOVERY(tp->t_flags); 418 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 419 tp->t_maxseg) * tp->t_maxseg; 420 tp->snd_cwnd = tp->t_maxseg; 421 break; 422 case CC_RTO_ERR: 423 TCPSTAT_INC(tcps_sndrexmitbad); 424 /* RTO was unnecessary, so reset everything. */ 425 tp->snd_cwnd = tp->snd_cwnd_prev; 426 tp->snd_ssthresh = tp->snd_ssthresh_prev; 427 tp->snd_recover = tp->snd_recover_prev; 428 if (tp->t_flags & TF_WASFRECOVERY) 429 ENTER_FASTRECOVERY(tp->t_flags); 430 if (tp->t_flags & TF_WASCRECOVERY) 431 ENTER_CONGRECOVERY(tp->t_flags); 432 tp->snd_nxt = tp->snd_max; 433 tp->t_flags &= ~TF_PREVVALID; 434 tp->t_badrxtwin = 0; 435 break; 436 } 437 438 if (CC_ALGO(tp)->cong_signal != NULL) { 439 if (th != NULL) 440 tp->ccv->curack = th->th_ack; 441 CC_ALGO(tp)->cong_signal(tp->ccv, type); 442 } 443} 444 445static void inline 446cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 447{ 448 INP_WLOCK_ASSERT(tp->t_inpcb); 449 450 /* XXXLAS: KASSERT that we're in recovery? */ 451 452 if (CC_ALGO(tp)->post_recovery != NULL) { 453 tp->ccv->curack = th->th_ack; 454 CC_ALGO(tp)->post_recovery(tp->ccv); 455 } 456 /* XXXLAS: EXIT_RECOVERY ? */ 457 tp->t_bytes_acked = 0; 458} 459 460#ifdef TCP_SIGNATURE 461static inline int 462tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 463 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 464{ 465 int ret; 466 467 tcp_fields_to_net(th); 468 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 469 tcp_fields_to_host(th); 470 return (ret); 471} 472#endif 473 474/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 475#ifdef INET6 476#define ND6_HINT(tp) \ 477do { \ 478 if ((tp) && (tp)->t_inpcb && \ 479 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 480 nd6_nud_hint(NULL, NULL, 0); \ 481} while (0) 482#else 483#define ND6_HINT(tp) 484#endif 485 486/* 487 * Indicate whether this ack should be delayed. We can delay the ack if 488 * - there is no delayed ack timer in progress and 489 * - our last ack wasn't a 0-sized window. We never want to delay 490 * the ack that opens up a 0-sized window and 491 * - delayed acks are enabled or 492 * - this is a half-synchronized T/TCP connection. 493 * - the segment size is not larger than the MSS and LRO wasn't used 494 * for this segment. 495 */ 496#define DELAY_ACK(tp, tlen) \ 497 ((!tcp_timer_active(tp, TT_DELACK) && \ 498 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 499 (tlen <= tp->t_maxopd) && \ 500 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 501 502/* 503 * TCP input handling is split into multiple parts: 504 * tcp6_input is a thin wrapper around tcp_input for the extended 505 * ip6_protox[] call format in ip6_input 506 * tcp_input handles primary segment validation, inpcb lookup and 507 * SYN processing on listen sockets 508 * tcp_do_segment processes the ACK and text of the segment for 509 * establishing, established and closing connections 510 */ 511#ifdef INET6 512int 513tcp6_input(struct mbuf **mp, int *offp, int proto) 514{ 515 struct mbuf *m = *mp; 516 struct in6_ifaddr *ia6; 517 518 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 519 520 /* 521 * draft-itojun-ipv6-tcp-to-anycast 522 * better place to put this in? 523 */ 524 ia6 = ip6_getdstifaddr(m); 525 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 526 struct ip6_hdr *ip6; 527 528 ifa_free(&ia6->ia_ifa); 529 ip6 = mtod(m, struct ip6_hdr *); 530 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 531 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 532 return IPPROTO_DONE; 533 } 534 if (ia6) 535 ifa_free(&ia6->ia_ifa); 536 537 tcp_input(m, *offp); 538 return IPPROTO_DONE; 539} 540#endif /* INET6 */ 541 542void 543tcp_input(struct mbuf *m, int off0) 544{ 545 struct tcphdr *th = NULL; 546 struct ip *ip = NULL; 547 struct inpcb *inp = NULL; 548 struct tcpcb *tp = NULL; 549 struct socket *so = NULL; 550 u_char *optp = NULL; 551 int optlen = 0; 552#ifdef INET 553 int len; 554#endif 555 int tlen = 0, off; 556 int drop_hdrlen; 557 int thflags; 558 int rstreason = 0; /* For badport_bandlim accounting purposes */ 559#ifdef TCP_SIGNATURE 560 uint8_t sig_checked = 0; 561#endif 562 uint8_t iptos = 0; 563 struct m_tag *fwd_tag = NULL; 564#ifdef INET6 565 struct ip6_hdr *ip6 = NULL; 566 int isipv6; 567#else 568 const void *ip6 = NULL; 569#endif /* INET6 */ 570 struct tcpopt to; /* options in this segment */ 571 char *s = NULL; /* address and port logging */ 572 int ti_locked; 573#define TI_UNLOCKED 1 574#define TI_RLOCKED 2 575 576#ifdef TCPDEBUG 577 /* 578 * The size of tcp_saveipgen must be the size of the max ip header, 579 * now IPv6. 580 */ 581 u_char tcp_saveipgen[IP6_HDR_LEN]; 582 struct tcphdr tcp_savetcp; 583 short ostate = 0; 584#endif 585 586#ifdef INET6 587 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 588#endif 589 590 to.to_flags = 0; 591 TCPSTAT_INC(tcps_rcvtotal); 592 593#ifdef INET6 594 if (isipv6) { 595 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 596 597 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 598 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 599 if (m == NULL) { 600 TCPSTAT_INC(tcps_rcvshort); 601 return; 602 } 603 } 604 605 ip6 = mtod(m, struct ip6_hdr *); 606 th = (struct tcphdr *)((caddr_t)ip6 + off0); 607 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 608 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 609 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 610 th->th_sum = m->m_pkthdr.csum_data; 611 else 612 th->th_sum = in6_cksum_pseudo(ip6, tlen, 613 IPPROTO_TCP, m->m_pkthdr.csum_data); 614 th->th_sum ^= 0xffff; 615 } else 616 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 617 if (th->th_sum) { 618 TCPSTAT_INC(tcps_rcvbadsum); 619 goto drop; 620 } 621 622 /* 623 * Be proactive about unspecified IPv6 address in source. 624 * As we use all-zero to indicate unbounded/unconnected pcb, 625 * unspecified IPv6 address can be used to confuse us. 626 * 627 * Note that packets with unspecified IPv6 destination is 628 * already dropped in ip6_input. 629 */ 630 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 631 /* XXX stat */ 632 goto drop; 633 } 634 } 635#endif 636#if defined(INET) && defined(INET6) 637 else 638#endif 639#ifdef INET 640 { 641 /* 642 * Get IP and TCP header together in first mbuf. 643 * Note: IP leaves IP header in first mbuf. 644 */ 645 if (off0 > sizeof (struct ip)) { 646 ip_stripoptions(m); 647 off0 = sizeof(struct ip); 648 } 649 if (m->m_len < sizeof (struct tcpiphdr)) { 650 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 651 == NULL) { 652 TCPSTAT_INC(tcps_rcvshort); 653 return; 654 } 655 } 656 ip = mtod(m, struct ip *); 657 th = (struct tcphdr *)((caddr_t)ip + off0); 658 tlen = ntohs(ip->ip_len) - off0; 659 660 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 661 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 662 th->th_sum = m->m_pkthdr.csum_data; 663 else 664 th->th_sum = in_pseudo(ip->ip_src.s_addr, 665 ip->ip_dst.s_addr, 666 htonl(m->m_pkthdr.csum_data + tlen + 667 IPPROTO_TCP)); 668 th->th_sum ^= 0xffff; 669 } else { 670 struct ipovly *ipov = (struct ipovly *)ip; 671 672 /* 673 * Checksum extended TCP header and data. 674 */ 675 len = off0 + tlen; 676 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 677 ipov->ih_len = htons(tlen); 678 th->th_sum = in_cksum(m, len); 679 /* Reset length for SDT probes. */ 680 ip->ip_len = htons(tlen + off0); 681 } 682 683 if (th->th_sum) { 684 TCPSTAT_INC(tcps_rcvbadsum); 685 goto drop; 686 } 687 /* Re-initialization for later version check */ 688 ip->ip_v = IPVERSION; 689 } 690#endif /* INET */ 691 692#ifdef INET6 693 if (isipv6) 694 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 695#endif 696#if defined(INET) && defined(INET6) 697 else 698#endif 699#ifdef INET 700 iptos = ip->ip_tos; 701#endif 702 703 /* 704 * Check that TCP offset makes sense, 705 * pull out TCP options and adjust length. XXX 706 */ 707 off = th->th_off << 2; 708 if (off < sizeof (struct tcphdr) || off > tlen) { 709 TCPSTAT_INC(tcps_rcvbadoff); 710 goto drop; 711 } 712 tlen -= off; /* tlen is used instead of ti->ti_len */ 713 if (off > sizeof (struct tcphdr)) { 714#ifdef INET6 715 if (isipv6) { 716 IP6_EXTHDR_CHECK(m, off0, off, ); 717 ip6 = mtod(m, struct ip6_hdr *); 718 th = (struct tcphdr *)((caddr_t)ip6 + off0); 719 } 720#endif 721#if defined(INET) && defined(INET6) 722 else 723#endif 724#ifdef INET 725 { 726 if (m->m_len < sizeof(struct ip) + off) { 727 if ((m = m_pullup(m, sizeof (struct ip) + off)) 728 == NULL) { 729 TCPSTAT_INC(tcps_rcvshort); 730 return; 731 } 732 ip = mtod(m, struct ip *); 733 th = (struct tcphdr *)((caddr_t)ip + off0); 734 } 735 } 736#endif 737 optlen = off - sizeof (struct tcphdr); 738 optp = (u_char *)(th + 1); 739 } 740 thflags = th->th_flags; 741 742 /* 743 * Convert TCP protocol specific fields to host format. 744 */ 745 tcp_fields_to_host(th); 746 747 /* 748 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 749 */ 750 drop_hdrlen = off0 + off; 751 752 /* 753 * Locate pcb for segment; if we're likely to add or remove a 754 * connection then first acquire pcbinfo lock. There are three cases 755 * where we might discover later we need a write lock despite the 756 * flags: ACKs moving a connection out of the syncache, ACKs for a 757 * connection in TIMEWAIT and SYNs not targeting a listening socket. 758 */ 759 if ((thflags & (TH_FIN | TH_RST)) != 0) { 760 INP_INFO_RLOCK(&V_tcbinfo); 761 ti_locked = TI_RLOCKED; 762 } else 763 ti_locked = TI_UNLOCKED; 764 765 /* 766 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 767 */ 768 if ( 769#ifdef INET6 770 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 771#ifdef INET 772 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 773#endif 774#endif 775#if defined(INET) && !defined(INET6) 776 (m->m_flags & M_IP_NEXTHOP) 777#endif 778 ) 779 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 780 781findpcb: 782#ifdef INVARIANTS 783 if (ti_locked == TI_RLOCKED) { 784 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 785 } else { 786 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 787 } 788#endif 789#ifdef INET6 790 if (isipv6 && fwd_tag != NULL) { 791 struct sockaddr_in6 *next_hop6; 792 793 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 794 /* 795 * Transparently forwarded. Pretend to be the destination. 796 * Already got one like this? 797 */ 798 inp = in6_pcblookup_mbuf(&V_tcbinfo, 799 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 800 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 801 if (!inp) { 802 /* 803 * It's new. Try to find the ambushing socket. 804 * Because we've rewritten the destination address, 805 * any hardware-generated hash is ignored. 806 */ 807 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 808 th->th_sport, &next_hop6->sin6_addr, 809 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 810 th->th_dport, INPLOOKUP_WILDCARD | 811 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 812 } 813 } else if (isipv6) { 814 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 815 th->th_sport, &ip6->ip6_dst, th->th_dport, 816 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 817 m->m_pkthdr.rcvif, m); 818 } 819#endif /* INET6 */ 820#if defined(INET6) && defined(INET) 821 else 822#endif 823#ifdef INET 824 if (fwd_tag != NULL) { 825 struct sockaddr_in *next_hop; 826 827 next_hop = (struct sockaddr_in *)(fwd_tag+1); 828 /* 829 * Transparently forwarded. Pretend to be the destination. 830 * already got one like this? 831 */ 832 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 833 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 834 m->m_pkthdr.rcvif, m); 835 if (!inp) { 836 /* 837 * It's new. Try to find the ambushing socket. 838 * Because we've rewritten the destination address, 839 * any hardware-generated hash is ignored. 840 */ 841 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 842 th->th_sport, next_hop->sin_addr, 843 next_hop->sin_port ? ntohs(next_hop->sin_port) : 844 th->th_dport, INPLOOKUP_WILDCARD | 845 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 846 } 847 } else 848 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 849 th->th_sport, ip->ip_dst, th->th_dport, 850 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 851 m->m_pkthdr.rcvif, m); 852#endif /* INET */ 853 854 /* 855 * If the INPCB does not exist then all data in the incoming 856 * segment is discarded and an appropriate RST is sent back. 857 * XXX MRT Send RST using which routing table? 858 */ 859 if (inp == NULL) { 860 /* 861 * Log communication attempts to ports that are not 862 * in use. 863 */ 864 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 865 tcp_log_in_vain == 2) { 866 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 867 log(LOG_INFO, "%s; %s: Connection attempt " 868 "to closed port\n", s, __func__); 869 } 870 /* 871 * When blackholing do not respond with a RST but 872 * completely ignore the segment and drop it. 873 */ 874 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 875 V_blackhole == 2) 876 goto dropunlock; 877 878 rstreason = BANDLIM_RST_CLOSEDPORT; 879 goto dropwithreset; 880 } 881 INP_WLOCK_ASSERT(inp); 882 /* 883 * While waiting for inp lock during the lookup, another thread 884 * can have dropped the inpcb, in which case we need to loop back 885 * and try to find a new inpcb to deliver to. 886 */ 887 if (inp->inp_flags & INP_DROPPED) { 888 INP_WUNLOCK(inp); 889 inp = NULL; 890 goto findpcb; 891 } 892 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 893 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 894 ((inp->inp_socket == NULL) || 895 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 896 inp->inp_flowid = m->m_pkthdr.flowid; 897 inp->inp_flowtype = M_HASHTYPE_GET(m); 898 } 899#ifdef IPSEC 900#ifdef INET6 901 if (isipv6 && ipsec6_in_reject(m, inp)) { 902 IPSEC6STAT_INC(ips_in_polvio); 903 goto dropunlock; 904 } else 905#endif /* INET6 */ 906 if (ipsec4_in_reject(m, inp) != 0) { 907 IPSECSTAT_INC(ips_in_polvio); 908 goto dropunlock; 909 } 910#endif /* IPSEC */ 911 912 /* 913 * Check the minimum TTL for socket. 914 */ 915 if (inp->inp_ip_minttl != 0) { 916#ifdef INET6 917 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 918 goto dropunlock; 919 else 920#endif 921 if (inp->inp_ip_minttl > ip->ip_ttl) 922 goto dropunlock; 923 } 924 925 /* 926 * A previous connection in TIMEWAIT state is supposed to catch stray 927 * or duplicate segments arriving late. If this segment was a 928 * legitimate new connection attempt, the old INPCB gets removed and 929 * we can try again to find a listening socket. 930 * 931 * At this point, due to earlier optimism, we may hold only an inpcb 932 * lock, and not the inpcbinfo write lock. If so, we need to try to 933 * acquire it, or if that fails, acquire a reference on the inpcb, 934 * drop all locks, acquire a global write lock, and then re-acquire 935 * the inpcb lock. We may at that point discover that another thread 936 * has tried to free the inpcb, in which case we need to loop back 937 * and try to find a new inpcb to deliver to. 938 * 939 * XXXRW: It may be time to rethink timewait locking. 940 */ 941relocked: 942 if (inp->inp_flags & INP_TIMEWAIT) { 943 if (ti_locked == TI_UNLOCKED) { 944 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 945 in_pcbref(inp); 946 INP_WUNLOCK(inp); 947 INP_INFO_RLOCK(&V_tcbinfo); 948 ti_locked = TI_RLOCKED; 949 INP_WLOCK(inp); 950 if (in_pcbrele_wlocked(inp)) { 951 inp = NULL; 952 goto findpcb; 953 } else if (inp->inp_flags & INP_DROPPED) { 954 INP_WUNLOCK(inp); 955 inp = NULL; 956 goto findpcb; 957 } 958 } else 959 ti_locked = TI_RLOCKED; 960 } 961 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 962 963 if (thflags & TH_SYN) 964 tcp_dooptions(&to, optp, optlen, TO_SYN); 965 /* 966 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 967 */ 968 if (tcp_twcheck(inp, &to, th, m, tlen)) 969 goto findpcb; 970 INP_INFO_RUNLOCK(&V_tcbinfo); 971 return; 972 } 973 /* 974 * The TCPCB may no longer exist if the connection is winding 975 * down or it is in the CLOSED state. Either way we drop the 976 * segment and send an appropriate response. 977 */ 978 tp = intotcpcb(inp); 979 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 980 rstreason = BANDLIM_RST_CLOSEDPORT; 981 goto dropwithreset; 982 } 983 984#ifdef TCP_OFFLOAD 985 if (tp->t_flags & TF_TOE) { 986 tcp_offload_input(tp, m); 987 m = NULL; /* consumed by the TOE driver */ 988 goto dropunlock; 989 } 990#endif 991 992 /* 993 * We've identified a valid inpcb, but it could be that we need an 994 * inpcbinfo write lock but don't hold it. In this case, attempt to 995 * acquire using the same strategy as the TIMEWAIT case above. If we 996 * relock, we have to jump back to 'relocked' as the connection might 997 * now be in TIMEWAIT. 998 */ 999#ifdef INVARIANTS 1000 if ((thflags & (TH_FIN | TH_RST)) != 0) 1001 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1002#endif 1003 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1004 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1005 !(tp->t_flags & TF_FASTOPEN)))) { 1006 if (ti_locked == TI_UNLOCKED) { 1007 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1008 in_pcbref(inp); 1009 INP_WUNLOCK(inp); 1010 INP_INFO_RLOCK(&V_tcbinfo); 1011 ti_locked = TI_RLOCKED; 1012 INP_WLOCK(inp); 1013 if (in_pcbrele_wlocked(inp)) { 1014 inp = NULL; 1015 goto findpcb; 1016 } else if (inp->inp_flags & INP_DROPPED) { 1017 INP_WUNLOCK(inp); 1018 inp = NULL; 1019 goto findpcb; 1020 } 1021 goto relocked; 1022 } else 1023 ti_locked = TI_RLOCKED; 1024 } 1025 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1026 } 1027 1028#ifdef MAC 1029 INP_WLOCK_ASSERT(inp); 1030 if (mac_inpcb_check_deliver(inp, m)) 1031 goto dropunlock; 1032#endif 1033 so = inp->inp_socket; 1034 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1035#ifdef TCPDEBUG 1036 if (so->so_options & SO_DEBUG) { 1037 ostate = tp->t_state; 1038#ifdef INET6 1039 if (isipv6) { 1040 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1041 } else 1042#endif 1043 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1044 tcp_savetcp = *th; 1045 } 1046#endif /* TCPDEBUG */ 1047 /* 1048 * When the socket is accepting connections (the INPCB is in LISTEN 1049 * state) we look into the SYN cache if this is a new connection 1050 * attempt or the completion of a previous one. 1051 */ 1052 if (so->so_options & SO_ACCEPTCONN) { 1053 struct in_conninfo inc; 1054 1055 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1056 "tp not listening", __func__)); 1057 bzero(&inc, sizeof(inc)); 1058#ifdef INET6 1059 if (isipv6) { 1060 inc.inc_flags |= INC_ISIPV6; 1061 inc.inc6_faddr = ip6->ip6_src; 1062 inc.inc6_laddr = ip6->ip6_dst; 1063 } else 1064#endif 1065 { 1066 inc.inc_faddr = ip->ip_src; 1067 inc.inc_laddr = ip->ip_dst; 1068 } 1069 inc.inc_fport = th->th_sport; 1070 inc.inc_lport = th->th_dport; 1071 inc.inc_fibnum = so->so_fibnum; 1072 1073 /* 1074 * Check for an existing connection attempt in syncache if 1075 * the flag is only ACK. A successful lookup creates a new 1076 * socket appended to the listen queue in SYN_RECEIVED state. 1077 */ 1078 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1079 1080 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1081 /* 1082 * Parse the TCP options here because 1083 * syncookies need access to the reflected 1084 * timestamp. 1085 */ 1086 tcp_dooptions(&to, optp, optlen, 0); 1087 /* 1088 * NB: syncache_expand() doesn't unlock 1089 * inp and tcpinfo locks. 1090 */ 1091 if (!syncache_expand(&inc, &to, th, &so, m)) { 1092 /* 1093 * No syncache entry or ACK was not 1094 * for our SYN/ACK. Send a RST. 1095 * NB: syncache did its own logging 1096 * of the failure cause. 1097 */ 1098 rstreason = BANDLIM_RST_OPENPORT; 1099 goto dropwithreset; 1100 } 1101#ifdef TCP_RFC7413 1102new_tfo_socket: 1103#endif 1104 if (so == NULL) { 1105 /* 1106 * We completed the 3-way handshake 1107 * but could not allocate a socket 1108 * either due to memory shortage, 1109 * listen queue length limits or 1110 * global socket limits. Send RST 1111 * or wait and have the remote end 1112 * retransmit the ACK for another 1113 * try. 1114 */ 1115 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1116 log(LOG_DEBUG, "%s; %s: Listen socket: " 1117 "Socket allocation failed due to " 1118 "limits or memory shortage, %s\n", 1119 s, __func__, 1120 V_tcp_sc_rst_sock_fail ? 1121 "sending RST" : "try again"); 1122 if (V_tcp_sc_rst_sock_fail) { 1123 rstreason = BANDLIM_UNLIMITED; 1124 goto dropwithreset; 1125 } else 1126 goto dropunlock; 1127 } 1128 /* 1129 * Socket is created in state SYN_RECEIVED. 1130 * Unlock the listen socket, lock the newly 1131 * created socket and update the tp variable. 1132 */ 1133 INP_WUNLOCK(inp); /* listen socket */ 1134 inp = sotoinpcb(so); 1135 /* 1136 * New connection inpcb is already locked by 1137 * syncache_expand(). 1138 */ 1139 INP_WLOCK_ASSERT(inp); 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#ifdef TCP_RFC7413 1366 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1367 goto new_tfo_socket; 1368#else 1369 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1370#endif 1371 /* 1372 * Entry added to syncache and mbuf consumed. 1373 * Only the listen socket is unlocked by syncache_add(). 1374 */ 1375 if (ti_locked == TI_RLOCKED) { 1376 INP_INFO_RUNLOCK(&V_tcbinfo); 1377 ti_locked = TI_UNLOCKED; 1378 } 1379 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1380 return; 1381 } else if (tp->t_state == TCPS_LISTEN) { 1382 /* 1383 * When a listen socket is torn down the SO_ACCEPTCONN 1384 * flag is removed first while connections are drained 1385 * from the accept queue in a unlock/lock cycle of the 1386 * ACCEPT_LOCK, opening a race condition allowing a SYN 1387 * attempt go through unhandled. 1388 */ 1389 goto dropunlock; 1390 } 1391 1392#ifdef TCP_SIGNATURE 1393 if (sig_checked == 0) { 1394 tcp_dooptions(&to, optp, optlen, 1395 (thflags & TH_SYN) ? TO_SYN : 0); 1396 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1397 th, tp->t_flags)) { 1398 1399 /* 1400 * In SYN_SENT state if it receives an RST, it is 1401 * allowed for further processing. 1402 */ 1403 if ((thflags & TH_RST) == 0 || 1404 (tp->t_state == TCPS_SYN_SENT) == 0) 1405 goto dropunlock; 1406 } 1407 sig_checked = 1; 1408 } 1409#endif 1410 1411 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1412 1413 /* 1414 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1415 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1416 * the inpcb, and unlocks pcbinfo. 1417 */ 1418 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1419 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1420 return; 1421 1422dropwithreset: 1423 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1424 1425 if (ti_locked == TI_RLOCKED) { 1426 INP_INFO_RUNLOCK(&V_tcbinfo); 1427 ti_locked = TI_UNLOCKED; 1428 } 1429#ifdef INVARIANTS 1430 else { 1431 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1432 "ti_locked: %d", __func__, ti_locked)); 1433 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1434 } 1435#endif 1436 1437 if (inp != NULL) { 1438 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1439 INP_WUNLOCK(inp); 1440 } else 1441 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1442 m = NULL; /* mbuf chain got consumed. */ 1443 goto drop; 1444 1445dropunlock: 1446 if (m != NULL) 1447 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1448 1449 if (ti_locked == TI_RLOCKED) { 1450 INP_INFO_RUNLOCK(&V_tcbinfo); 1451 ti_locked = TI_UNLOCKED; 1452 } 1453#ifdef INVARIANTS 1454 else { 1455 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1456 "ti_locked: %d", __func__, ti_locked)); 1457 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1458 } 1459#endif 1460 1461 if (inp != NULL) 1462 INP_WUNLOCK(inp); 1463 1464drop: 1465 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1466 if (s != NULL) 1467 free(s, M_TCPLOG); 1468 if (m != NULL) 1469 m_freem(m); 1470} 1471 1472/* 1473 * Automatic sizing of receive socket buffer. Often the send 1474 * buffer size is not optimally adjusted to the actual network 1475 * conditions at hand (delay bandwidth product). Setting the 1476 * buffer size too small limits throughput on links with high 1477 * bandwidth and high delay (eg. trans-continental/oceanic links). 1478 * 1479 * On the receive side the socket buffer memory is only rarely 1480 * used to any significant extent. This allows us to be much 1481 * more aggressive in scaling the receive socket buffer. For 1482 * the case that the buffer space is actually used to a large 1483 * extent and we run out of kernel memory we can simply drop 1484 * the new segments; TCP on the sender will just retransmit it 1485 * later. Setting the buffer size too big may only consume too 1486 * much kernel memory if the application doesn't read() from 1487 * the socket or packet loss or reordering makes use of the 1488 * reassembly queue. 1489 * 1490 * The criteria to step up the receive buffer one notch are: 1491 * 1. Application has not set receive buffer size with 1492 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1493 * 2. the number of bytes received during the time it takes 1494 * one timestamp to be reflected back to us (the RTT); 1495 * 3. received bytes per RTT is within seven eighth of the 1496 * current socket buffer size; 1497 * 4. receive buffer size has not hit maximal automatic size; 1498 * 1499 * This algorithm does one step per RTT at most and only if 1500 * we receive a bulk stream w/o packet losses or reorderings. 1501 * Shrinking the buffer during idle times is not necessary as 1502 * it doesn't consume any memory when idle. 1503 * 1504 * TODO: Only step up if the application is actually serving 1505 * the buffer to better manage the socket buffer resources. 1506 */ 1507int 1508tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1509 struct tcpcb *tp, int tlen) 1510{ 1511 int newsize = 0; 1512 1513 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1514 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1515 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1516 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1517 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1518 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1519 newsize = min(so->so_rcv.sb_hiwat + 1520 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1521 } 1522 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1523 1524 /* Start over with next RTT. */ 1525 tp->rfbuf_ts = 0; 1526 tp->rfbuf_cnt = 0; 1527 } else { 1528 tp->rfbuf_cnt += tlen; /* add up */ 1529 } 1530 1531 return (newsize); 1532} 1533 1534static void 1535tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1536 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1537 int ti_locked) 1538{ 1539 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1540 int rstreason, todrop, win; 1541 u_long tiwin; 1542 char *s; 1543 struct in_conninfo *inc; 1544 struct mbuf *mfree; 1545 struct tcpopt to; 1546 int tfo_syn; 1547 1548#ifdef TCPDEBUG 1549 /* 1550 * The size of tcp_saveipgen must be the size of the max ip header, 1551 * now IPv6. 1552 */ 1553 u_char tcp_saveipgen[IP6_HDR_LEN]; 1554 struct tcphdr tcp_savetcp; 1555 short ostate = 0; 1556#endif 1557 thflags = th->th_flags; 1558 inc = &tp->t_inpcb->inp_inc; 1559 tp->sackhint.last_sack_ack = 0; 1560 sack_changed = 0; 1561 1562 /* 1563 * If this is either a state-changing packet or current state isn't 1564 * established, we require a write lock on tcbinfo. Otherwise, we 1565 * allow the tcbinfo to be in either alocked or unlocked, as the 1566 * caller may have unnecessarily acquired a write lock due to a race. 1567 */ 1568 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1569 tp->t_state != TCPS_ESTABLISHED) { 1570 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1571 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1572 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1573 } else { 1574#ifdef INVARIANTS 1575 if (ti_locked == TI_RLOCKED) 1576 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1577 else { 1578 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1579 "ti_locked: %d", __func__, ti_locked)); 1580 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1581 } 1582#endif 1583 } 1584 INP_WLOCK_ASSERT(tp->t_inpcb); 1585 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1586 __func__)); 1587 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1588 __func__)); 1589 1590 /* 1591 * Segment received on connection. 1592 * Reset idle time and keep-alive timer. 1593 * XXX: This should be done after segment 1594 * validation to ignore broken/spoofed segs. 1595 */ 1596 tp->t_rcvtime = ticks; 1597 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1598 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1599 1600 /* 1601 * Unscale the window into a 32-bit value. 1602 * For the SYN_SENT state the scale is zero. 1603 */ 1604 tiwin = th->th_win << tp->snd_scale; 1605 1606 /* 1607 * TCP ECN processing. 1608 */ 1609 if (tp->t_flags & TF_ECN_PERMIT) { 1610 if (thflags & TH_CWR) 1611 tp->t_flags &= ~TF_ECN_SND_ECE; 1612 switch (iptos & IPTOS_ECN_MASK) { 1613 case IPTOS_ECN_CE: 1614 tp->t_flags |= TF_ECN_SND_ECE; 1615 TCPSTAT_INC(tcps_ecn_ce); 1616 break; 1617 case IPTOS_ECN_ECT0: 1618 TCPSTAT_INC(tcps_ecn_ect0); 1619 break; 1620 case IPTOS_ECN_ECT1: 1621 TCPSTAT_INC(tcps_ecn_ect1); 1622 break; 1623 } 1624 /* Congestion experienced. */ 1625 if (thflags & TH_ECE) { 1626 cc_cong_signal(tp, th, CC_ECN); 1627 } 1628 } 1629 1630 /* 1631 * Parse options on any incoming segment. 1632 */ 1633 tcp_dooptions(&to, (u_char *)(th + 1), 1634 (th->th_off << 2) - sizeof(struct tcphdr), 1635 (thflags & TH_SYN) ? TO_SYN : 0); 1636 1637 /* 1638 * If echoed timestamp is later than the current time, 1639 * fall back to non RFC1323 RTT calculation. Normalize 1640 * timestamp if syncookies were used when this connection 1641 * was established. 1642 */ 1643 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1644 to.to_tsecr -= tp->ts_offset; 1645 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1646 to.to_tsecr = 0; 1647 } 1648 /* 1649 * If timestamps were negotiated during SYN/ACK they should 1650 * appear on every segment during this session and vice versa. 1651 */ 1652 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1653 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1654 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1655 "no action\n", s, __func__); 1656 free(s, M_TCPLOG); 1657 } 1658 } 1659 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1660 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1661 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1662 "no action\n", s, __func__); 1663 free(s, M_TCPLOG); 1664 } 1665 } 1666 1667 /* 1668 * Process options only when we get SYN/ACK back. The SYN case 1669 * for incoming connections is handled in tcp_syncache. 1670 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1671 * or <SYN,ACK>) segment itself is never scaled. 1672 * XXX this is traditional behavior, may need to be cleaned up. 1673 */ 1674 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1675 if ((to.to_flags & TOF_SCALE) && 1676 (tp->t_flags & TF_REQ_SCALE)) { 1677 tp->t_flags |= TF_RCVD_SCALE; 1678 tp->snd_scale = to.to_wscale; 1679 } 1680 /* 1681 * Initial send window. It will be updated with 1682 * the next incoming segment to the scaled value. 1683 */ 1684 tp->snd_wnd = th->th_win; 1685 if (to.to_flags & TOF_TS) { 1686 tp->t_flags |= TF_RCVD_TSTMP; 1687 tp->ts_recent = to.to_tsval; 1688 tp->ts_recent_age = tcp_ts_getticks(); 1689 } 1690 if (to.to_flags & TOF_MSS) 1691 tcp_mss(tp, to.to_mss); 1692 if ((tp->t_flags & TF_SACK_PERMIT) && 1693 (to.to_flags & TOF_SACKPERM) == 0) 1694 tp->t_flags &= ~TF_SACK_PERMIT; 1695 } 1696 1697 /* 1698 * Header prediction: check for the two common cases 1699 * of a uni-directional data xfer. If the packet has 1700 * no control flags, is in-sequence, the window didn't 1701 * change and we're not retransmitting, it's a 1702 * candidate. If the length is zero and the ack moved 1703 * forward, we're the sender side of the xfer. Just 1704 * free the data acked & wake any higher level process 1705 * that was blocked waiting for space. If the length 1706 * is non-zero and the ack didn't move, we're the 1707 * receiver side. If we're getting packets in-order 1708 * (the reassembly queue is empty), add the data to 1709 * the socket buffer and note that we need a delayed ack. 1710 * Make sure that the hidden state-flags are also off. 1711 * Since we check for TCPS_ESTABLISHED first, it can only 1712 * be TH_NEEDSYN. 1713 */ 1714 if (tp->t_state == TCPS_ESTABLISHED && 1715 th->th_seq == tp->rcv_nxt && 1716 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1717 tp->snd_nxt == tp->snd_max && 1718 tiwin && tiwin == tp->snd_wnd && 1719 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1720 LIST_EMPTY(&tp->t_segq) && 1721 ((to.to_flags & TOF_TS) == 0 || 1722 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1723 1724 /* 1725 * If last ACK falls within this segment's sequence numbers, 1726 * record the timestamp. 1727 * NOTE that the test is modified according to the latest 1728 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1729 */ 1730 if ((to.to_flags & TOF_TS) != 0 && 1731 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1732 tp->ts_recent_age = tcp_ts_getticks(); 1733 tp->ts_recent = to.to_tsval; 1734 } 1735 1736 if (tlen == 0) { 1737 if (SEQ_GT(th->th_ack, tp->snd_una) && 1738 SEQ_LEQ(th->th_ack, tp->snd_max) && 1739 !IN_RECOVERY(tp->t_flags) && 1740 (to.to_flags & TOF_SACK) == 0 && 1741 TAILQ_EMPTY(&tp->snd_holes)) { 1742 /* 1743 * This is a pure ack for outstanding data. 1744 */ 1745 if (ti_locked == TI_RLOCKED) 1746 INP_INFO_RUNLOCK(&V_tcbinfo); 1747 ti_locked = TI_UNLOCKED; 1748 1749 TCPSTAT_INC(tcps_predack); 1750 1751 /* 1752 * "bad retransmit" recovery. 1753 */ 1754 if (tp->t_rxtshift == 1 && 1755 tp->t_flags & TF_PREVVALID && 1756 (int)(ticks - tp->t_badrxtwin) < 0) { 1757 cc_cong_signal(tp, th, CC_RTO_ERR); 1758 } 1759 1760 /* 1761 * Recalculate the transmit timer / rtt. 1762 * 1763 * Some boxes send broken timestamp replies 1764 * during the SYN+ACK phase, ignore 1765 * timestamps of 0 or we could calculate a 1766 * huge RTT and blow up the retransmit timer. 1767 */ 1768 if ((to.to_flags & TOF_TS) != 0 && 1769 to.to_tsecr) { 1770 u_int t; 1771 1772 t = tcp_ts_getticks() - to.to_tsecr; 1773 if (!tp->t_rttlow || tp->t_rttlow > t) 1774 tp->t_rttlow = t; 1775 tcp_xmit_timer(tp, 1776 TCP_TS_TO_TICKS(t) + 1); 1777 } else if (tp->t_rtttime && 1778 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1779 if (!tp->t_rttlow || 1780 tp->t_rttlow > ticks - tp->t_rtttime) 1781 tp->t_rttlow = ticks - tp->t_rtttime; 1782 tcp_xmit_timer(tp, 1783 ticks - tp->t_rtttime); 1784 } 1785 acked = BYTES_THIS_ACK(tp, th); 1786 1787 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1788 hhook_run_tcp_est_in(tp, th, &to); 1789 1790 TCPSTAT_INC(tcps_rcvackpack); 1791 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1792 sbdrop(&so->so_snd, acked); 1793 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1794 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1795 tp->snd_recover = th->th_ack - 1; 1796 1797 /* 1798 * Let the congestion control algorithm update 1799 * congestion control related information. This 1800 * typically means increasing the congestion 1801 * window. 1802 */ 1803 cc_ack_received(tp, th, CC_ACK); 1804 1805 tp->snd_una = th->th_ack; 1806 /* 1807 * Pull snd_wl2 up to prevent seq wrap relative 1808 * to th_ack. 1809 */ 1810 tp->snd_wl2 = th->th_ack; 1811 tp->t_dupacks = 0; 1812 m_freem(m); 1813 ND6_HINT(tp); /* Some progress has been made. */ 1814 1815 /* 1816 * If all outstanding data are acked, stop 1817 * retransmit timer, otherwise restart timer 1818 * using current (possibly backed-off) value. 1819 * If process is waiting for space, 1820 * wakeup/selwakeup/signal. If data 1821 * are ready to send, let tcp_output 1822 * decide between more output or persist. 1823 */ 1824#ifdef TCPDEBUG 1825 if (so->so_options & SO_DEBUG) 1826 tcp_trace(TA_INPUT, ostate, tp, 1827 (void *)tcp_saveipgen, 1828 &tcp_savetcp, 0); 1829#endif 1830 if (tp->snd_una == tp->snd_max) 1831 tcp_timer_activate(tp, TT_REXMT, 0); 1832 else if (!tcp_timer_active(tp, TT_PERSIST)) 1833 tcp_timer_activate(tp, TT_REXMT, 1834 tp->t_rxtcur); 1835 sowwakeup(so); 1836 if (so->so_snd.sb_cc) 1837 (void) tcp_output(tp); 1838 goto check_delack; 1839 } 1840 } else if (th->th_ack == tp->snd_una && 1841 tlen <= sbspace(&so->so_rcv)) { 1842 int newsize = 0; /* automatic sockbuf scaling */ 1843 1844 /* 1845 * This is a pure, in-sequence data packet with 1846 * nothing on the reassembly queue and we have enough 1847 * buffer space to take it. 1848 */ 1849 if (ti_locked == TI_RLOCKED) 1850 INP_INFO_RUNLOCK(&V_tcbinfo); 1851 ti_locked = TI_UNLOCKED; 1852 1853 /* Clean receiver SACK report if present */ 1854 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1855 tcp_clean_sackreport(tp); 1856 TCPSTAT_INC(tcps_preddat); 1857 tp->rcv_nxt += tlen; 1858 /* 1859 * Pull snd_wl1 up to prevent seq wrap relative to 1860 * th_seq. 1861 */ 1862 tp->snd_wl1 = th->th_seq; 1863 /* 1864 * Pull rcv_up up to prevent seq wrap relative to 1865 * rcv_nxt. 1866 */ 1867 tp->rcv_up = tp->rcv_nxt; 1868 TCPSTAT_INC(tcps_rcvpack); 1869 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1870 ND6_HINT(tp); /* Some progress has been made */ 1871#ifdef TCPDEBUG 1872 if (so->so_options & SO_DEBUG) 1873 tcp_trace(TA_INPUT, ostate, tp, 1874 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1875#endif 1876 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1877 1878 /* Add data to socket buffer. */ 1879 SOCKBUF_LOCK(&so->so_rcv); 1880 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1881 m_freem(m); 1882 } else { 1883 /* 1884 * Set new socket buffer size. 1885 * Give up when limit is reached. 1886 */ 1887 if (newsize) 1888 if (!sbreserve_locked(&so->so_rcv, 1889 newsize, so, NULL)) 1890 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1891 m_adj(m, drop_hdrlen); /* delayed header drop */ 1892 sbappendstream_locked(&so->so_rcv, m); 1893 } 1894 /* NB: sorwakeup_locked() does an implicit unlock. */ 1895 sorwakeup_locked(so); 1896 if (DELAY_ACK(tp, tlen)) { 1897 tp->t_flags |= TF_DELACK; 1898 } else { 1899 tp->t_flags |= TF_ACKNOW; 1900 tcp_output(tp); 1901 } 1902 goto check_delack; 1903 } 1904 } 1905 1906 /* 1907 * Calculate amount of space in receive window, 1908 * and then do TCP input processing. 1909 * Receive window is amount of space in rcv queue, 1910 * but not less than advertised window. 1911 */ 1912 win = sbspace(&so->so_rcv); 1913 if (win < 0) 1914 win = 0; 1915 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1916 1917 switch (tp->t_state) { 1918 1919 /* 1920 * If the state is SYN_RECEIVED: 1921 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1922 */ 1923 case TCPS_SYN_RECEIVED: 1924 if ((thflags & TH_ACK) && 1925 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1926 SEQ_GT(th->th_ack, tp->snd_max))) { 1927 rstreason = BANDLIM_RST_OPENPORT; 1928 goto dropwithreset; 1929 } 1930#ifdef TCP_RFC7413 1931 if (tp->t_flags & TF_FASTOPEN) { 1932 /* 1933 * When a TFO connection is in SYN_RECEIVED, the 1934 * only valid packets are the initial SYN, a 1935 * retransmit/copy of the initial SYN (possibly with 1936 * a subset of the original data), a valid ACK, a 1937 * FIN, or a RST. 1938 */ 1939 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1940 rstreason = BANDLIM_RST_OPENPORT; 1941 goto dropwithreset; 1942 } else if (thflags & TH_SYN) { 1943 /* non-initial SYN is ignored */ 1944 if ((tcp_timer_active(tp, TT_DELACK) || 1945 tcp_timer_active(tp, TT_REXMT))) 1946 goto drop; 1947 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1948 goto drop; 1949 } 1950 } 1951#endif 1952 break; 1953 1954 /* 1955 * If the state is SYN_SENT: 1956 * if seg contains an ACK, but not for our SYN, drop the input. 1957 * if seg contains a RST, then drop the connection. 1958 * if seg does not contain SYN, then drop it. 1959 * Otherwise this is an acceptable SYN segment 1960 * initialize tp->rcv_nxt and tp->irs 1961 * if seg contains ack then advance tp->snd_una 1962 * if seg contains an ECE and ECN support is enabled, the stream 1963 * is ECN capable. 1964 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1965 * arrange for segment to be acked (eventually) 1966 * continue processing rest of data/controls, beginning with URG 1967 */ 1968 case TCPS_SYN_SENT: 1969 if ((thflags & TH_ACK) && 1970 (SEQ_LEQ(th->th_ack, tp->iss) || 1971 SEQ_GT(th->th_ack, tp->snd_max))) { 1972 rstreason = BANDLIM_UNLIMITED; 1973 goto dropwithreset; 1974 } 1975 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1976 TCP_PROBE5(connect__refused, NULL, tp, 1977 mtod(m, const char *), tp, th); 1978 tp = tcp_drop(tp, ECONNREFUSED); 1979 } 1980 if (thflags & TH_RST) 1981 goto drop; 1982 if (!(thflags & TH_SYN)) 1983 goto drop; 1984 1985 tp->irs = th->th_seq; 1986 tcp_rcvseqinit(tp); 1987 if (thflags & TH_ACK) { 1988 TCPSTAT_INC(tcps_connects); 1989 soisconnected(so); 1990#ifdef MAC 1991 mac_socketpeer_set_from_mbuf(m, so); 1992#endif 1993 /* Do window scaling on this connection? */ 1994 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1995 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1996 tp->rcv_scale = tp->request_r_scale; 1997 } 1998 tp->rcv_adv += imin(tp->rcv_wnd, 1999 TCP_MAXWIN << tp->rcv_scale); 2000 tp->snd_una++; /* SYN is acked */ 2001 /* 2002 * If there's data, delay ACK; if there's also a FIN 2003 * ACKNOW will be turned on later. 2004 */ 2005 if (DELAY_ACK(tp, tlen) && tlen != 0) 2006 tcp_timer_activate(tp, TT_DELACK, 2007 tcp_delacktime); 2008 else 2009 tp->t_flags |= TF_ACKNOW; 2010 2011 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2012 tp->t_flags |= TF_ECN_PERMIT; 2013 TCPSTAT_INC(tcps_ecn_shs); 2014 } 2015 2016 /* 2017 * Received <SYN,ACK> in SYN_SENT[*] state. 2018 * Transitions: 2019 * SYN_SENT --> ESTABLISHED 2020 * SYN_SENT* --> FIN_WAIT_1 2021 */ 2022 tp->t_starttime = ticks; 2023 if (tp->t_flags & TF_NEEDFIN) { 2024 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2025 tp->t_flags &= ~TF_NEEDFIN; 2026 thflags &= ~TH_SYN; 2027 } else { 2028 tcp_state_change(tp, TCPS_ESTABLISHED); 2029 TCP_PROBE5(connect__established, NULL, tp, 2030 mtod(m, const char *), tp, th); 2031 cc_conn_init(tp); 2032 tcp_timer_activate(tp, TT_KEEP, 2033 TP_KEEPIDLE(tp)); 2034 } 2035 } else { 2036 /* 2037 * Received initial SYN in SYN-SENT[*] state => 2038 * simultaneous open. If segment contains CC option 2039 * and there is a cached CC, apply TAO test. 2040 * If it succeeds, connection is * half-synchronized. 2041 * Otherwise, do 3-way handshake: 2042 * SYN-SENT -> SYN-RECEIVED 2043 * SYN-SENT* -> SYN-RECEIVED* 2044 * If there was no CC option, clear cached CC value. 2045 */ 2046 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2047 tcp_timer_activate(tp, TT_REXMT, 0); 2048 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2049 } 2050 2051 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2052 "ti_locked %d", __func__, ti_locked)); 2053 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2054 INP_WLOCK_ASSERT(tp->t_inpcb); 2055 2056 /* 2057 * Advance th->th_seq to correspond to first data byte. 2058 * If data, trim to stay within window, 2059 * dropping FIN if necessary. 2060 */ 2061 th->th_seq++; 2062 if (tlen > tp->rcv_wnd) { 2063 todrop = tlen - tp->rcv_wnd; 2064 m_adj(m, -todrop); 2065 tlen = tp->rcv_wnd; 2066 thflags &= ~TH_FIN; 2067 TCPSTAT_INC(tcps_rcvpackafterwin); 2068 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2069 } 2070 tp->snd_wl1 = th->th_seq - 1; 2071 tp->rcv_up = th->th_seq; 2072 /* 2073 * Client side of transaction: already sent SYN and data. 2074 * If the remote host used T/TCP to validate the SYN, 2075 * our data will be ACK'd; if so, enter normal data segment 2076 * processing in the middle of step 5, ack processing. 2077 * Otherwise, goto step 6. 2078 */ 2079 if (thflags & TH_ACK) 2080 goto process_ACK; 2081 2082 goto step6; 2083 2084 /* 2085 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2086 * do normal processing. 2087 * 2088 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2089 */ 2090 case TCPS_LAST_ACK: 2091 case TCPS_CLOSING: 2092 break; /* continue normal processing */ 2093 } 2094 2095 /* 2096 * States other than LISTEN or SYN_SENT. 2097 * First check the RST flag and sequence number since reset segments 2098 * are exempt from the timestamp and connection count tests. This 2099 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2100 * below which allowed reset segments in half the sequence space 2101 * to fall though and be processed (which gives forged reset 2102 * segments with a random sequence number a 50 percent chance of 2103 * killing a connection). 2104 * Then check timestamp, if present. 2105 * Then check the connection count, if present. 2106 * Then check that at least some bytes of segment are within 2107 * receive window. If segment begins before rcv_nxt, 2108 * drop leading data (and SYN); if nothing left, just ack. 2109 * 2110 * 2111 * If the RST bit is set, check the sequence number to see 2112 * if this is a valid reset segment. 2113 * RFC 793 page 37: 2114 * In all states except SYN-SENT, all reset (RST) segments 2115 * are validated by checking their SEQ-fields. A reset is 2116 * valid if its sequence number is in the window. 2117 * Note: this does not take into account delayed ACKs, so 2118 * we should test against last_ack_sent instead of rcv_nxt. 2119 * The sequence number in the reset segment is normally an 2120 * echo of our outgoing acknowlegement numbers, but some hosts 2121 * send a reset with the sequence number at the rightmost edge 2122 * of our receive window, and we have to handle this case. 2123 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2124 * that brute force RST attacks are possible. To combat this, 2125 * we use a much stricter check while in the ESTABLISHED state, 2126 * only accepting RSTs where the sequence number is equal to 2127 * last_ack_sent. In all other states (the states in which a 2128 * RST is more likely), the more permissive check is used. 2129 * If we have multiple segments in flight, the initial reset 2130 * segment sequence numbers will be to the left of last_ack_sent, 2131 * but they will eventually catch up. 2132 * In any case, it never made sense to trim reset segments to 2133 * fit the receive window since RFC 1122 says: 2134 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2135 * 2136 * A TCP SHOULD allow a received RST segment to include data. 2137 * 2138 * DISCUSSION 2139 * It has been suggested that a RST segment could contain 2140 * ASCII text that encoded and explained the cause of the 2141 * RST. No standard has yet been established for such 2142 * data. 2143 * 2144 * If the reset segment passes the sequence number test examine 2145 * the state: 2146 * SYN_RECEIVED STATE: 2147 * If passive open, return to LISTEN state. 2148 * If active open, inform user that connection was refused. 2149 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2150 * Inform user that connection was reset, and close tcb. 2151 * CLOSING, LAST_ACK STATES: 2152 * Close the tcb. 2153 * TIME_WAIT STATE: 2154 * Drop the segment - see Stevens, vol. 2, p. 964 and 2155 * RFC 1337. 2156 */ 2157 if (thflags & TH_RST) { 2158 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2159 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2160 switch (tp->t_state) { 2161 2162 case TCPS_SYN_RECEIVED: 2163 so->so_error = ECONNREFUSED; 2164 goto close; 2165 2166 case TCPS_ESTABLISHED: 2167 if (V_tcp_insecure_rst == 0 && 2168 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2169 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2170 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2171 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2172 TCPSTAT_INC(tcps_badrst); 2173 goto drop; 2174 } 2175 /* FALLTHROUGH */ 2176 case TCPS_FIN_WAIT_1: 2177 case TCPS_FIN_WAIT_2: 2178 case TCPS_CLOSE_WAIT: 2179 so->so_error = ECONNRESET; 2180 close: 2181 KASSERT(ti_locked == TI_RLOCKED, 2182 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2183 ti_locked)); 2184 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2185 2186 tcp_state_change(tp, TCPS_CLOSED); 2187 TCPSTAT_INC(tcps_drops); 2188 tp = tcp_close(tp); 2189 break; 2190 2191 case TCPS_CLOSING: 2192 case TCPS_LAST_ACK: 2193 KASSERT(ti_locked == TI_RLOCKED, 2194 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2195 ti_locked)); 2196 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2197 2198 tp = tcp_close(tp); 2199 break; 2200 } 2201 } 2202 goto drop; 2203 } 2204 2205 /* 2206 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2207 * and it's less than ts_recent, drop it. 2208 */ 2209 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2210 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2211 2212 /* Check to see if ts_recent is over 24 days old. */ 2213 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2214 /* 2215 * Invalidate ts_recent. If this segment updates 2216 * ts_recent, the age will be reset later and ts_recent 2217 * will get a valid value. If it does not, setting 2218 * ts_recent to zero will at least satisfy the 2219 * requirement that zero be placed in the timestamp 2220 * echo reply when ts_recent isn't valid. The 2221 * age isn't reset until we get a valid ts_recent 2222 * because we don't want out-of-order segments to be 2223 * dropped when ts_recent is old. 2224 */ 2225 tp->ts_recent = 0; 2226 } else { 2227 TCPSTAT_INC(tcps_rcvduppack); 2228 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2229 TCPSTAT_INC(tcps_pawsdrop); 2230 if (tlen) 2231 goto dropafterack; 2232 goto drop; 2233 } 2234 } 2235 2236 /* 2237 * In the SYN-RECEIVED state, validate that the packet belongs to 2238 * this connection before trimming the data to fit the receive 2239 * window. Check the sequence number versus IRS since we know 2240 * the sequence numbers haven't wrapped. This is a partial fix 2241 * for the "LAND" DoS attack. 2242 */ 2243 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2244 rstreason = BANDLIM_RST_OPENPORT; 2245 goto dropwithreset; 2246 } 2247 2248 todrop = tp->rcv_nxt - th->th_seq; 2249 if (todrop > 0) { 2250 if (thflags & TH_SYN) { 2251 thflags &= ~TH_SYN; 2252 th->th_seq++; 2253 if (th->th_urp > 1) 2254 th->th_urp--; 2255 else 2256 thflags &= ~TH_URG; 2257 todrop--; 2258 } 2259 /* 2260 * Following if statement from Stevens, vol. 2, p. 960. 2261 */ 2262 if (todrop > tlen 2263 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2264 /* 2265 * Any valid FIN must be to the left of the window. 2266 * At this point the FIN must be a duplicate or out 2267 * of sequence; drop it. 2268 */ 2269 thflags &= ~TH_FIN; 2270 2271 /* 2272 * Send an ACK to resynchronize and drop any data. 2273 * But keep on processing for RST or ACK. 2274 */ 2275 tp->t_flags |= TF_ACKNOW; 2276 todrop = tlen; 2277 TCPSTAT_INC(tcps_rcvduppack); 2278 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2279 } else { 2280 TCPSTAT_INC(tcps_rcvpartduppack); 2281 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2282 } 2283 drop_hdrlen += todrop; /* drop from the top afterwards */ 2284 th->th_seq += todrop; 2285 tlen -= todrop; 2286 if (th->th_urp > todrop) 2287 th->th_urp -= todrop; 2288 else { 2289 thflags &= ~TH_URG; 2290 th->th_urp = 0; 2291 } 2292 } 2293 2294 /* 2295 * If new data are received on a connection after the 2296 * user processes are gone, then RST the other end. 2297 */ 2298 if ((so->so_state & SS_NOFDREF) && 2299 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2300 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2301 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2302 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2303 2304 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2305 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2306 "after socket was closed, " 2307 "sending RST and removing tcpcb\n", 2308 s, __func__, tcpstates[tp->t_state], tlen); 2309 free(s, M_TCPLOG); 2310 } 2311 tp = tcp_close(tp); 2312 TCPSTAT_INC(tcps_rcvafterclose); 2313 rstreason = BANDLIM_UNLIMITED; 2314 goto dropwithreset; 2315 } 2316 2317 /* 2318 * If segment ends after window, drop trailing data 2319 * (and PUSH and FIN); if nothing left, just ACK. 2320 */ 2321 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2322 if (todrop > 0) { 2323 TCPSTAT_INC(tcps_rcvpackafterwin); 2324 if (todrop >= tlen) { 2325 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2326 /* 2327 * If window is closed can only take segments at 2328 * window edge, and have to drop data and PUSH from 2329 * incoming segments. Continue processing, but 2330 * remember to ack. Otherwise, drop segment 2331 * and ack. 2332 */ 2333 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2334 tp->t_flags |= TF_ACKNOW; 2335 TCPSTAT_INC(tcps_rcvwinprobe); 2336 } else 2337 goto dropafterack; 2338 } else 2339 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2340 m_adj(m, -todrop); 2341 tlen -= todrop; 2342 thflags &= ~(TH_PUSH|TH_FIN); 2343 } 2344 2345 /* 2346 * If last ACK falls within this segment's sequence numbers, 2347 * record its timestamp. 2348 * NOTE: 2349 * 1) That the test incorporates suggestions from the latest 2350 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2351 * 2) That updating only on newer timestamps interferes with 2352 * our earlier PAWS tests, so this check should be solely 2353 * predicated on the sequence space of this segment. 2354 * 3) That we modify the segment boundary check to be 2355 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2356 * instead of RFC1323's 2357 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2358 * This modified check allows us to overcome RFC1323's 2359 * limitations as described in Stevens TCP/IP Illustrated 2360 * Vol. 2 p.869. In such cases, we can still calculate the 2361 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2362 */ 2363 if ((to.to_flags & TOF_TS) != 0 && 2364 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2365 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2366 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2367 tp->ts_recent_age = tcp_ts_getticks(); 2368 tp->ts_recent = to.to_tsval; 2369 } 2370 2371 /* 2372 * If a SYN is in the window, then this is an 2373 * error and we send an RST and drop the connection. 2374 */ 2375 if (thflags & TH_SYN) { 2376 KASSERT(ti_locked == TI_RLOCKED, 2377 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2378 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2379 2380 tp = tcp_drop(tp, ECONNRESET); 2381 rstreason = BANDLIM_UNLIMITED; 2382 goto drop; 2383 } 2384 2385 /* 2386 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2387 * flag is on (half-synchronized state), then queue data for 2388 * later processing; else drop segment and return. 2389 */ 2390 if ((thflags & TH_ACK) == 0) { 2391 if (tp->t_state == TCPS_SYN_RECEIVED || 2392 (tp->t_flags & TF_NEEDSYN)) { 2393#ifdef TCP_RFC7413 2394 if (tp->t_state == TCPS_SYN_RECEIVED && 2395 tp->t_flags & TF_FASTOPEN) { 2396 tp->snd_wnd = tiwin; 2397 cc_conn_init(tp); 2398 } 2399#endif 2400 goto step6; 2401 } else if (tp->t_flags & TF_ACKNOW) 2402 goto dropafterack; 2403 else 2404 goto drop; 2405 } 2406 2407 /* 2408 * Ack processing. 2409 */ 2410 switch (tp->t_state) { 2411 2412 /* 2413 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2414 * ESTABLISHED state and continue processing. 2415 * The ACK was checked above. 2416 */ 2417 case TCPS_SYN_RECEIVED: 2418 2419 TCPSTAT_INC(tcps_connects); 2420 soisconnected(so); 2421 /* Do window scaling? */ 2422 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2423 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2424 tp->rcv_scale = tp->request_r_scale; 2425 tp->snd_wnd = tiwin; 2426 } 2427 /* 2428 * Make transitions: 2429 * SYN-RECEIVED -> ESTABLISHED 2430 * SYN-RECEIVED* -> FIN-WAIT-1 2431 */ 2432 tp->t_starttime = ticks; 2433 if (tp->t_flags & TF_NEEDFIN) { 2434 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2435 tp->t_flags &= ~TF_NEEDFIN; 2436 } else { 2437 tcp_state_change(tp, TCPS_ESTABLISHED); 2438 TCP_PROBE5(accept__established, NULL, tp, 2439 mtod(m, const char *), tp, th); 2440#ifdef TCP_RFC7413 2441 if (tp->t_tfo_pending) { 2442 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2443 tp->t_tfo_pending = NULL; 2444 2445 /* 2446 * Account for the ACK of our SYN prior to 2447 * regular ACK processing below. 2448 */ 2449 tp->snd_una++; 2450 } 2451 /* 2452 * TFO connections call cc_conn_init() during SYN 2453 * processing. Calling it again here for such 2454 * connections is not harmless as it would undo the 2455 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2456 * is retransmitted. 2457 */ 2458 if (!(tp->t_flags & TF_FASTOPEN)) 2459#endif 2460 cc_conn_init(tp); 2461 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2462 } 2463 /* 2464 * If segment contains data or ACK, will call tcp_reass() 2465 * later; if not, do so now to pass queued data to user. 2466 */ 2467 if (tlen == 0 && (thflags & TH_FIN) == 0) 2468 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2469 (struct mbuf *)0); 2470 tp->snd_wl1 = th->th_seq - 1; 2471 /* FALLTHROUGH */ 2472 2473 /* 2474 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2475 * ACKs. If the ack is in the range 2476 * tp->snd_una < th->th_ack <= tp->snd_max 2477 * then advance tp->snd_una to th->th_ack and drop 2478 * data from the retransmission queue. If this ACK reflects 2479 * more up to date window information we update our window information. 2480 */ 2481 case TCPS_ESTABLISHED: 2482 case TCPS_FIN_WAIT_1: 2483 case TCPS_FIN_WAIT_2: 2484 case TCPS_CLOSE_WAIT: 2485 case TCPS_CLOSING: 2486 case TCPS_LAST_ACK: 2487 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2488 TCPSTAT_INC(tcps_rcvacktoomuch); 2489 goto dropafterack; 2490 } 2491 if ((tp->t_flags & TF_SACK_PERMIT) && 2492 ((to.to_flags & TOF_SACK) || 2493 !TAILQ_EMPTY(&tp->snd_holes))) 2494 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2495 else 2496 /* 2497 * Reset the value so that previous (valid) value 2498 * from the last ack with SACK doesn't get used. 2499 */ 2500 tp->sackhint.sacked_bytes = 0; 2501 2502 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2503 hhook_run_tcp_est_in(tp, th, &to); 2504 2505 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2506 if (tlen == 0 && 2507 (tiwin == tp->snd_wnd || 2508 (tp->t_flags & TF_SACK_PERMIT))) { 2509 TCPSTAT_INC(tcps_rcvdupack); 2510 /* 2511 * If we have outstanding data (other than 2512 * a window probe), this is a completely 2513 * duplicate ack (ie, window info didn't 2514 * change), the ack is the biggest we've 2515 * seen and we've seen exactly our rexmt 2516 * threshhold of them, assume a packet 2517 * has been dropped and retransmit it. 2518 * Kludge snd_nxt & the congestion 2519 * window so we send only this one 2520 * packet. 2521 * 2522 * We know we're losing at the current 2523 * window size so do congestion avoidance 2524 * (set ssthresh to half the current window 2525 * and pull our congestion window back to 2526 * the new ssthresh). 2527 * 2528 * Dup acks mean that packets have left the 2529 * network (they're now cached at the receiver) 2530 * so bump cwnd by the amount in the receiver 2531 * to keep a constant cwnd packets in the 2532 * network. 2533 * 2534 * When using TCP ECN, notify the peer that 2535 * we reduced the cwnd. 2536 */ 2537 /* 2538 * Following 2 kinds of acks should not affect 2539 * dupack counting: 2540 * 1) Old acks 2541 * 2) Acks with SACK but without any new SACK 2542 * information in them. These could result from 2543 * any anomaly in the network like a switch 2544 * duplicating packets or a possible DoS attack. 2545 */ 2546 if (th->th_ack != tp->snd_una || 2547 ((tp->t_flags & TF_SACK_PERMIT) && 2548 !sack_changed)) 2549 break; 2550 else if (!tcp_timer_active(tp, TT_REXMT)) 2551 tp->t_dupacks = 0; 2552 else if (++tp->t_dupacks > tcprexmtthresh || 2553 IN_FASTRECOVERY(tp->t_flags)) { 2554 cc_ack_received(tp, th, CC_DUPACK); 2555 if ((tp->t_flags & TF_SACK_PERMIT) && 2556 IN_FASTRECOVERY(tp->t_flags)) { 2557 int awnd; 2558 2559 /* 2560 * Compute the amount of data in flight first. 2561 * We can inject new data into the pipe iff 2562 * we have less than 1/2 the original window's 2563 * worth of data in flight. 2564 */ 2565 if (V_tcp_do_rfc6675_pipe) 2566 awnd = tcp_compute_pipe(tp); 2567 else 2568 awnd = (tp->snd_nxt - tp->snd_fack) + 2569 tp->sackhint.sack_bytes_rexmit; 2570 2571 if (awnd < tp->snd_ssthresh) { 2572 tp->snd_cwnd += tp->t_maxseg; 2573 if (tp->snd_cwnd > tp->snd_ssthresh) 2574 tp->snd_cwnd = tp->snd_ssthresh; 2575 } 2576 } else 2577 tp->snd_cwnd += tp->t_maxseg; 2578 if ((thflags & TH_FIN) && 2579 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2580 /* 2581 * If its a fin we need to process 2582 * it to avoid a race where both 2583 * sides enter FIN-WAIT and send FIN|ACK 2584 * at the same time. 2585 */ 2586 break; 2587 } 2588 (void) tcp_output(tp); 2589 goto drop; 2590 } else if (tp->t_dupacks == tcprexmtthresh) { 2591 tcp_seq onxt = tp->snd_nxt; 2592 2593 /* 2594 * If we're doing sack, check to 2595 * see if we're already in sack 2596 * recovery. If we're not doing sack, 2597 * check to see if we're in newreno 2598 * recovery. 2599 */ 2600 if (tp->t_flags & TF_SACK_PERMIT) { 2601 if (IN_FASTRECOVERY(tp->t_flags)) { 2602 tp->t_dupacks = 0; 2603 break; 2604 } 2605 } else { 2606 if (SEQ_LEQ(th->th_ack, 2607 tp->snd_recover)) { 2608 tp->t_dupacks = 0; 2609 break; 2610 } 2611 } 2612 /* Congestion signal before ack. */ 2613 cc_cong_signal(tp, th, CC_NDUPACK); 2614 cc_ack_received(tp, th, CC_DUPACK); 2615 tcp_timer_activate(tp, TT_REXMT, 0); 2616 tp->t_rtttime = 0; 2617 if (tp->t_flags & TF_SACK_PERMIT) { 2618 TCPSTAT_INC( 2619 tcps_sack_recovery_episode); 2620 tp->sack_newdata = tp->snd_nxt; 2621 tp->snd_cwnd = tp->t_maxseg; 2622 (void) tcp_output(tp); 2623 goto drop; 2624 } 2625 tp->snd_nxt = th->th_ack; 2626 tp->snd_cwnd = tp->t_maxseg; 2627 if ((thflags & TH_FIN) && 2628 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2629 /* 2630 * If its a fin we need to process 2631 * it to avoid a race where both 2632 * sides enter FIN-WAIT and send FIN|ACK 2633 * at the same time. 2634 */ 2635 break; 2636 } 2637 (void) tcp_output(tp); 2638 KASSERT(tp->snd_limited <= 2, 2639 ("%s: tp->snd_limited too big", 2640 __func__)); 2641 tp->snd_cwnd = tp->snd_ssthresh + 2642 tp->t_maxseg * 2643 (tp->t_dupacks - tp->snd_limited); 2644 if (SEQ_GT(onxt, tp->snd_nxt)) 2645 tp->snd_nxt = onxt; 2646 goto drop; 2647 } else if (V_tcp_do_rfc3042) { 2648 /* 2649 * Process first and second duplicate 2650 * ACKs. Each indicates a segment 2651 * leaving the network, creating room 2652 * for more. Make sure we can send a 2653 * packet on reception of each duplicate 2654 * ACK by increasing snd_cwnd by one 2655 * segment. Restore the original 2656 * snd_cwnd after packet transmission. 2657 */ 2658 cc_ack_received(tp, th, CC_DUPACK); 2659 u_long oldcwnd = tp->snd_cwnd; 2660 tcp_seq oldsndmax = tp->snd_max; 2661 u_int sent; 2662 int avail; 2663 2664 KASSERT(tp->t_dupacks == 1 || 2665 tp->t_dupacks == 2, 2666 ("%s: dupacks not 1 or 2", 2667 __func__)); 2668 if (tp->t_dupacks == 1) 2669 tp->snd_limited = 0; 2670 tp->snd_cwnd = 2671 (tp->snd_nxt - tp->snd_una) + 2672 (tp->t_dupacks - tp->snd_limited) * 2673 tp->t_maxseg; 2674 if ((thflags & TH_FIN) && 2675 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2676 /* 2677 * If its a fin we need to process 2678 * it to avoid a race where both 2679 * sides enter FIN-WAIT and send FIN|ACK 2680 * at the same time. 2681 */ 2682 break; 2683 } 2684 /* 2685 * Only call tcp_output when there 2686 * is new data available to be sent. 2687 * Otherwise we would send pure ACKs. 2688 */ 2689 SOCKBUF_LOCK(&so->so_snd); 2690 avail = so->so_snd.sb_cc - 2691 (tp->snd_nxt - tp->snd_una); 2692 SOCKBUF_UNLOCK(&so->so_snd); 2693 if (avail > 0) 2694 (void) tcp_output(tp); 2695 sent = tp->snd_max - oldsndmax; 2696 if (sent > tp->t_maxseg) { 2697 KASSERT((tp->t_dupacks == 2 && 2698 tp->snd_limited == 0) || 2699 (sent == tp->t_maxseg + 1 && 2700 tp->t_flags & TF_SENTFIN), 2701 ("%s: sent too much", 2702 __func__)); 2703 tp->snd_limited = 2; 2704 } else if (sent > 0) 2705 ++tp->snd_limited; 2706 tp->snd_cwnd = oldcwnd; 2707 goto drop; 2708 } 2709 } 2710 break; 2711 } else { 2712 /* 2713 * This ack is advancing the left edge, reset the 2714 * counter. 2715 */ 2716 tp->t_dupacks = 0; 2717 /* 2718 * If this ack also has new SACK info, increment the 2719 * counter as per rfc6675. 2720 */ 2721 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2722 tp->t_dupacks++; 2723 } 2724 2725 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2726 ("%s: th_ack <= snd_una", __func__)); 2727 2728 /* 2729 * If the congestion window was inflated to account 2730 * for the other side's cached packets, retract it. 2731 */ 2732 if (IN_FASTRECOVERY(tp->t_flags)) { 2733 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2734 if (tp->t_flags & TF_SACK_PERMIT) 2735 tcp_sack_partialack(tp, th); 2736 else 2737 tcp_newreno_partial_ack(tp, th); 2738 } else 2739 cc_post_recovery(tp, th); 2740 } 2741 /* 2742 * If we reach this point, ACK is not a duplicate, 2743 * i.e., it ACKs something we sent. 2744 */ 2745 if (tp->t_flags & TF_NEEDSYN) { 2746 /* 2747 * T/TCP: Connection was half-synchronized, and our 2748 * SYN has been ACK'd (so connection is now fully 2749 * synchronized). Go to non-starred state, 2750 * increment snd_una for ACK of SYN, and check if 2751 * we can do window scaling. 2752 */ 2753 tp->t_flags &= ~TF_NEEDSYN; 2754 tp->snd_una++; 2755 /* Do window scaling? */ 2756 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2757 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2758 tp->rcv_scale = tp->request_r_scale; 2759 /* Send window already scaled. */ 2760 } 2761 } 2762 2763process_ACK: 2764 INP_WLOCK_ASSERT(tp->t_inpcb); 2765 2766 acked = BYTES_THIS_ACK(tp, th); 2767 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2768 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2769 tp->snd_una, th->th_ack, tp, m)); 2770 TCPSTAT_INC(tcps_rcvackpack); 2771 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2772 2773 /* 2774 * If we just performed our first retransmit, and the ACK 2775 * arrives within our recovery window, then it was a mistake 2776 * to do the retransmit in the first place. Recover our 2777 * original cwnd and ssthresh, and proceed to transmit where 2778 * we left off. 2779 */ 2780 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2781 (int)(ticks - tp->t_badrxtwin) < 0) 2782 cc_cong_signal(tp, th, CC_RTO_ERR); 2783 2784 /* 2785 * If we have a timestamp reply, update smoothed 2786 * round trip time. If no timestamp is present but 2787 * transmit timer is running and timed sequence 2788 * number was acked, update smoothed round trip time. 2789 * Since we now have an rtt measurement, cancel the 2790 * timer backoff (cf., Phil Karn's retransmit alg.). 2791 * Recompute the initial retransmit timer. 2792 * 2793 * Some boxes send broken timestamp replies 2794 * during the SYN+ACK phase, ignore 2795 * timestamps of 0 or we could calculate a 2796 * huge RTT and blow up the retransmit timer. 2797 */ 2798 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2799 u_int t; 2800 2801 t = tcp_ts_getticks() - to.to_tsecr; 2802 if (!tp->t_rttlow || tp->t_rttlow > t) 2803 tp->t_rttlow = t; 2804 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2805 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2806 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2807 tp->t_rttlow = ticks - tp->t_rtttime; 2808 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2809 } 2810 2811 /* 2812 * If all outstanding data is acked, stop retransmit 2813 * timer and remember to restart (more output or persist). 2814 * If there is more data to be acked, restart retransmit 2815 * timer, using current (possibly backed-off) value. 2816 */ 2817 if (th->th_ack == tp->snd_max) { 2818 tcp_timer_activate(tp, TT_REXMT, 0); 2819 needoutput = 1; 2820 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2821 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2822 2823 /* 2824 * If no data (only SYN) was ACK'd, 2825 * skip rest of ACK processing. 2826 */ 2827 if (acked == 0) 2828 goto step6; 2829 2830 /* 2831 * Let the congestion control algorithm update congestion 2832 * control related information. This typically means increasing 2833 * the congestion window. 2834 */ 2835 cc_ack_received(tp, th, CC_ACK); 2836 2837 SOCKBUF_LOCK(&so->so_snd); 2838 if (acked > so->so_snd.sb_cc) { 2839 if (tp->snd_wnd >= so->so_snd.sb_cc) 2840 tp->snd_wnd -= so->so_snd.sb_cc; 2841 else 2842 tp->snd_wnd = 0; 2843 mfree = sbcut_locked(&so->so_snd, 2844 (int)so->so_snd.sb_cc); 2845 ourfinisacked = 1; 2846 } else { 2847 mfree = sbcut_locked(&so->so_snd, acked); 2848 if (tp->snd_wnd >= (u_long) acked) 2849 tp->snd_wnd -= acked; 2850 else 2851 tp->snd_wnd = 0; 2852 ourfinisacked = 0; 2853 } 2854 /* NB: sowwakeup_locked() does an implicit unlock. */ 2855 sowwakeup_locked(so); 2856 m_freem(mfree); 2857 /* Detect una wraparound. */ 2858 if (!IN_RECOVERY(tp->t_flags) && 2859 SEQ_GT(tp->snd_una, tp->snd_recover) && 2860 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2861 tp->snd_recover = th->th_ack - 1; 2862 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2863 if (IN_RECOVERY(tp->t_flags) && 2864 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2865 EXIT_RECOVERY(tp->t_flags); 2866 } 2867 tp->snd_una = th->th_ack; 2868 if (tp->t_flags & TF_SACK_PERMIT) { 2869 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2870 tp->snd_recover = tp->snd_una; 2871 } 2872 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2873 tp->snd_nxt = tp->snd_una; 2874 2875 switch (tp->t_state) { 2876 2877 /* 2878 * In FIN_WAIT_1 STATE in addition to the processing 2879 * for the ESTABLISHED state if our FIN is now acknowledged 2880 * then enter FIN_WAIT_2. 2881 */ 2882 case TCPS_FIN_WAIT_1: 2883 if (ourfinisacked) { 2884 /* 2885 * If we can't receive any more 2886 * data, then closing user can proceed. 2887 * Starting the timer is contrary to the 2888 * specification, but if we don't get a FIN 2889 * we'll hang forever. 2890 * 2891 * XXXjl: 2892 * we should release the tp also, and use a 2893 * compressed state. 2894 */ 2895 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2896 soisdisconnected(so); 2897 tcp_timer_activate(tp, TT_2MSL, 2898 (tcp_fast_finwait2_recycle ? 2899 tcp_finwait2_timeout : 2900 TP_MAXIDLE(tp))); 2901 } 2902 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2903 } 2904 break; 2905 2906 /* 2907 * In CLOSING STATE in addition to the processing for 2908 * the ESTABLISHED state if the ACK acknowledges our FIN 2909 * then enter the TIME-WAIT state, otherwise ignore 2910 * the segment. 2911 */ 2912 case TCPS_CLOSING: 2913 if (ourfinisacked) { 2914 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2915 tcp_twstart(tp); 2916 INP_INFO_RUNLOCK(&V_tcbinfo); 2917 m_freem(m); 2918 return; 2919 } 2920 break; 2921 2922 /* 2923 * In LAST_ACK, we may still be waiting for data to drain 2924 * and/or to be acked, as well as for the ack of our FIN. 2925 * If our FIN is now acknowledged, delete the TCB, 2926 * enter the closed state and return. 2927 */ 2928 case TCPS_LAST_ACK: 2929 if (ourfinisacked) { 2930 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2931 tp = tcp_close(tp); 2932 goto drop; 2933 } 2934 break; 2935 } 2936 } 2937 2938step6: 2939 INP_WLOCK_ASSERT(tp->t_inpcb); 2940 2941 /* 2942 * Update window information. 2943 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2944 */ 2945 if ((thflags & TH_ACK) && 2946 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2947 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2948 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2949 /* keep track of pure window updates */ 2950 if (tlen == 0 && 2951 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2952 TCPSTAT_INC(tcps_rcvwinupd); 2953 tp->snd_wnd = tiwin; 2954 tp->snd_wl1 = th->th_seq; 2955 tp->snd_wl2 = th->th_ack; 2956 if (tp->snd_wnd > tp->max_sndwnd) 2957 tp->max_sndwnd = tp->snd_wnd; 2958 needoutput = 1; 2959 } 2960 2961 /* 2962 * Process segments with URG. 2963 */ 2964 if ((thflags & TH_URG) && th->th_urp && 2965 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2966 /* 2967 * This is a kludge, but if we receive and accept 2968 * random urgent pointers, we'll crash in 2969 * soreceive. It's hard to imagine someone 2970 * actually wanting to send this much urgent data. 2971 */ 2972 SOCKBUF_LOCK(&so->so_rcv); 2973 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2974 th->th_urp = 0; /* XXX */ 2975 thflags &= ~TH_URG; /* XXX */ 2976 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2977 goto dodata; /* XXX */ 2978 } 2979 /* 2980 * If this segment advances the known urgent pointer, 2981 * then mark the data stream. This should not happen 2982 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2983 * a FIN has been received from the remote side. 2984 * In these states we ignore the URG. 2985 * 2986 * According to RFC961 (Assigned Protocols), 2987 * the urgent pointer points to the last octet 2988 * of urgent data. We continue, however, 2989 * to consider it to indicate the first octet 2990 * of data past the urgent section as the original 2991 * spec states (in one of two places). 2992 */ 2993 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2994 tp->rcv_up = th->th_seq + th->th_urp; 2995 so->so_oobmark = so->so_rcv.sb_cc + 2996 (tp->rcv_up - tp->rcv_nxt) - 1; 2997 if (so->so_oobmark == 0) 2998 so->so_rcv.sb_state |= SBS_RCVATMARK; 2999 sohasoutofband(so); 3000 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3001 } 3002 SOCKBUF_UNLOCK(&so->so_rcv); 3003 /* 3004 * Remove out of band data so doesn't get presented to user. 3005 * This can happen independent of advancing the URG pointer, 3006 * but if two URG's are pending at once, some out-of-band 3007 * data may creep in... ick. 3008 */ 3009 if (th->th_urp <= (u_long)tlen && 3010 !(so->so_options & SO_OOBINLINE)) { 3011 /* hdr drop is delayed */ 3012 tcp_pulloutofband(so, th, m, drop_hdrlen); 3013 } 3014 } else { 3015 /* 3016 * If no out of band data is expected, 3017 * pull receive urgent pointer along 3018 * with the receive window. 3019 */ 3020 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3021 tp->rcv_up = tp->rcv_nxt; 3022 } 3023dodata: /* XXX */ 3024 INP_WLOCK_ASSERT(tp->t_inpcb); 3025 3026 /* 3027 * Process the segment text, merging it into the TCP sequencing queue, 3028 * and arranging for acknowledgment of receipt if necessary. 3029 * This process logically involves adjusting tp->rcv_wnd as data 3030 * is presented to the user (this happens in tcp_usrreq.c, 3031 * case PRU_RCVD). If a FIN has already been received on this 3032 * connection then we just ignore the text. 3033 */ 3034 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3035 (tp->t_flags & TF_FASTOPEN)); 3036 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3037 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3038 tcp_seq save_start = th->th_seq; 3039 3040 m_adj(m, drop_hdrlen); /* delayed header drop */ 3041 /* 3042 * Insert segment which includes th into TCP reassembly queue 3043 * with control block tp. Set thflags to whether reassembly now 3044 * includes a segment with FIN. This handles the common case 3045 * inline (segment is the next to be received on an established 3046 * connection, and the queue is empty), avoiding linkage into 3047 * and removal from the queue and repetition of various 3048 * conversions. 3049 * Set DELACK for segments received in order, but ack 3050 * immediately when segments are out of order (so 3051 * fast retransmit can work). 3052 */ 3053 if (th->th_seq == tp->rcv_nxt && 3054 LIST_EMPTY(&tp->t_segq) && 3055 (TCPS_HAVEESTABLISHED(tp->t_state) || 3056 tfo_syn)) { 3057 if (DELAY_ACK(tp, tlen) || tfo_syn) 3058 tp->t_flags |= TF_DELACK; 3059 else 3060 tp->t_flags |= TF_ACKNOW; 3061 tp->rcv_nxt += tlen; 3062 thflags = th->th_flags & TH_FIN; 3063 TCPSTAT_INC(tcps_rcvpack); 3064 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3065 ND6_HINT(tp); 3066 SOCKBUF_LOCK(&so->so_rcv); 3067 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3068 m_freem(m); 3069 else 3070 sbappendstream_locked(&so->so_rcv, m); 3071 /* NB: sorwakeup_locked() does an implicit unlock. */ 3072 sorwakeup_locked(so); 3073 } else { 3074 /* 3075 * XXX: Due to the header drop above "th" is 3076 * theoretically invalid by now. Fortunately 3077 * m_adj() doesn't actually frees any mbufs 3078 * when trimming from the head. 3079 */ 3080 thflags = tcp_reass(tp, th, &tlen, m); 3081 tp->t_flags |= TF_ACKNOW; 3082 } 3083 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3084 tcp_update_sack_list(tp, save_start, save_start + tlen); 3085#if 0 3086 /* 3087 * Note the amount of data that peer has sent into 3088 * our window, in order to estimate the sender's 3089 * buffer size. 3090 * XXX: Unused. 3091 */ 3092 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3093 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3094 else 3095 len = so->so_rcv.sb_hiwat; 3096#endif 3097 } else { 3098 m_freem(m); 3099 thflags &= ~TH_FIN; 3100 } 3101 3102 /* 3103 * If FIN is received ACK the FIN and let the user know 3104 * that the connection is closing. 3105 */ 3106 if (thflags & TH_FIN) { 3107 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3108 socantrcvmore(so); 3109 /* 3110 * If connection is half-synchronized 3111 * (ie NEEDSYN flag on) then delay ACK, 3112 * so it may be piggybacked when SYN is sent. 3113 * Otherwise, since we received a FIN then no 3114 * more input can be expected, send ACK now. 3115 */ 3116 if (tp->t_flags & TF_NEEDSYN) 3117 tp->t_flags |= TF_DELACK; 3118 else 3119 tp->t_flags |= TF_ACKNOW; 3120 tp->rcv_nxt++; 3121 } 3122 switch (tp->t_state) { 3123 3124 /* 3125 * In SYN_RECEIVED and ESTABLISHED STATES 3126 * enter the CLOSE_WAIT state. 3127 */ 3128 case TCPS_SYN_RECEIVED: 3129 tp->t_starttime = ticks; 3130 /* FALLTHROUGH */ 3131 case TCPS_ESTABLISHED: 3132 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3133 break; 3134 3135 /* 3136 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3137 * enter the CLOSING state. 3138 */ 3139 case TCPS_FIN_WAIT_1: 3140 tcp_state_change(tp, TCPS_CLOSING); 3141 break; 3142 3143 /* 3144 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3145 * starting the time-wait timer, turning off the other 3146 * standard timers. 3147 */ 3148 case TCPS_FIN_WAIT_2: 3149 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3150 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3151 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3152 ti_locked)); 3153 3154 tcp_twstart(tp); 3155 INP_INFO_RUNLOCK(&V_tcbinfo); 3156 return; 3157 } 3158 } 3159 if (ti_locked == TI_RLOCKED) 3160 INP_INFO_RUNLOCK(&V_tcbinfo); 3161 ti_locked = TI_UNLOCKED; 3162 3163#ifdef TCPDEBUG 3164 if (so->so_options & SO_DEBUG) 3165 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3166 &tcp_savetcp, 0); 3167#endif 3168 3169 /* 3170 * Return any desired output. 3171 */ 3172 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3173 (void) tcp_output(tp); 3174 3175check_delack: 3176 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3177 __func__, ti_locked)); 3178 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3179 INP_WLOCK_ASSERT(tp->t_inpcb); 3180 3181 if (tp->t_flags & TF_DELACK) { 3182 tp->t_flags &= ~TF_DELACK; 3183 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3184 } 3185 INP_WUNLOCK(tp->t_inpcb); 3186 return; 3187 3188dropafterack: 3189 /* 3190 * Generate an ACK dropping incoming segment if it occupies 3191 * sequence space, where the ACK reflects our state. 3192 * 3193 * We can now skip the test for the RST flag since all 3194 * paths to this code happen after packets containing 3195 * RST have been dropped. 3196 * 3197 * In the SYN-RECEIVED state, don't send an ACK unless the 3198 * segment we received passes the SYN-RECEIVED ACK test. 3199 * If it fails send a RST. This breaks the loop in the 3200 * "LAND" DoS attack, and also prevents an ACK storm 3201 * between two listening ports that have been sent forged 3202 * SYN segments, each with the source address of the other. 3203 */ 3204 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3205 (SEQ_GT(tp->snd_una, th->th_ack) || 3206 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3207 rstreason = BANDLIM_RST_OPENPORT; 3208 goto dropwithreset; 3209 } 3210#ifdef TCPDEBUG 3211 if (so->so_options & SO_DEBUG) 3212 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3213 &tcp_savetcp, 0); 3214#endif 3215 if (ti_locked == TI_RLOCKED) 3216 INP_INFO_RUNLOCK(&V_tcbinfo); 3217 ti_locked = TI_UNLOCKED; 3218 3219 tp->t_flags |= TF_ACKNOW; 3220 (void) tcp_output(tp); 3221 INP_WUNLOCK(tp->t_inpcb); 3222 m_freem(m); 3223 return; 3224 3225dropwithreset: 3226 if (ti_locked == TI_RLOCKED) 3227 INP_INFO_RUNLOCK(&V_tcbinfo); 3228 ti_locked = TI_UNLOCKED; 3229 3230 if (tp != NULL) { 3231 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3232 INP_WUNLOCK(tp->t_inpcb); 3233 } else 3234 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3235 return; 3236 3237drop: 3238 if (ti_locked == TI_RLOCKED) { 3239 INP_INFO_RUNLOCK(&V_tcbinfo); 3240 ti_locked = TI_UNLOCKED; 3241 } 3242#ifdef INVARIANTS 3243 else 3244 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3245#endif 3246 3247 /* 3248 * Drop space held by incoming segment and return. 3249 */ 3250#ifdef TCPDEBUG 3251 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3252 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3253 &tcp_savetcp, 0); 3254#endif 3255 if (tp != NULL) 3256 INP_WUNLOCK(tp->t_inpcb); 3257 m_freem(m); 3258} 3259 3260/* 3261 * Issue RST and make ACK acceptable to originator of segment. 3262 * The mbuf must still include the original packet header. 3263 * tp may be NULL. 3264 */ 3265static void 3266tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3267 int tlen, int rstreason) 3268{ 3269#ifdef INET 3270 struct ip *ip; 3271#endif 3272#ifdef INET6 3273 struct ip6_hdr *ip6; 3274#endif 3275 3276 if (tp != NULL) { 3277 INP_WLOCK_ASSERT(tp->t_inpcb); 3278 } 3279 3280 /* Don't bother if destination was broadcast/multicast. */ 3281 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3282 goto drop; 3283#ifdef INET6 3284 if (mtod(m, struct ip *)->ip_v == 6) { 3285 ip6 = mtod(m, struct ip6_hdr *); 3286 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3287 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3288 goto drop; 3289 /* IPv6 anycast check is done at tcp6_input() */ 3290 } 3291#endif 3292#if defined(INET) && defined(INET6) 3293 else 3294#endif 3295#ifdef INET 3296 { 3297 ip = mtod(m, struct ip *); 3298 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3299 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3300 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3301 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3302 goto drop; 3303 } 3304#endif 3305 3306 /* Perform bandwidth limiting. */ 3307 if (badport_bandlim(rstreason) < 0) 3308 goto drop; 3309 3310 /* tcp_respond consumes the mbuf chain. */ 3311 if (th->th_flags & TH_ACK) { 3312 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3313 th->th_ack, TH_RST); 3314 } else { 3315 if (th->th_flags & TH_SYN) 3316 tlen++; 3317 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3318 (tcp_seq)0, TH_RST|TH_ACK); 3319 } 3320 return; 3321drop: 3322 m_freem(m); 3323} 3324 3325/* 3326 * Parse TCP options and place in tcpopt. 3327 */ 3328static void 3329tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3330{ 3331 int opt, optlen; 3332 3333 to->to_flags = 0; 3334 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3335 opt = cp[0]; 3336 if (opt == TCPOPT_EOL) 3337 break; 3338 if (opt == TCPOPT_NOP) 3339 optlen = 1; 3340 else { 3341 if (cnt < 2) 3342 break; 3343 optlen = cp[1]; 3344 if (optlen < 2 || optlen > cnt) 3345 break; 3346 } 3347 switch (opt) { 3348 case TCPOPT_MAXSEG: 3349 if (optlen != TCPOLEN_MAXSEG) 3350 continue; 3351 if (!(flags & TO_SYN)) 3352 continue; 3353 to->to_flags |= TOF_MSS; 3354 bcopy((char *)cp + 2, 3355 (char *)&to->to_mss, sizeof(to->to_mss)); 3356 to->to_mss = ntohs(to->to_mss); 3357 break; 3358 case TCPOPT_WINDOW: 3359 if (optlen != TCPOLEN_WINDOW) 3360 continue; 3361 if (!(flags & TO_SYN)) 3362 continue; 3363 to->to_flags |= TOF_SCALE; 3364 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3365 break; 3366 case TCPOPT_TIMESTAMP: 3367 if (optlen != TCPOLEN_TIMESTAMP) 3368 continue; 3369 to->to_flags |= TOF_TS; 3370 bcopy((char *)cp + 2, 3371 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3372 to->to_tsval = ntohl(to->to_tsval); 3373 bcopy((char *)cp + 6, 3374 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3375 to->to_tsecr = ntohl(to->to_tsecr); 3376 break; 3377#ifdef TCP_SIGNATURE 3378 /* 3379 * XXX In order to reply to a host which has set the 3380 * TCP_SIGNATURE option in its initial SYN, we have to 3381 * record the fact that the option was observed here 3382 * for the syncache code to perform the correct response. 3383 */ 3384 case TCPOPT_SIGNATURE: 3385 if (optlen != TCPOLEN_SIGNATURE) 3386 continue; 3387 to->to_flags |= TOF_SIGNATURE; 3388 to->to_signature = cp + 2; 3389 break; 3390#endif 3391 case TCPOPT_SACK_PERMITTED: 3392 if (optlen != TCPOLEN_SACK_PERMITTED) 3393 continue; 3394 if (!(flags & TO_SYN)) 3395 continue; 3396 if (!V_tcp_do_sack) 3397 continue; 3398 to->to_flags |= TOF_SACKPERM; 3399 break; 3400 case TCPOPT_SACK: 3401 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3402 continue; 3403 if (flags & TO_SYN) 3404 continue; 3405 to->to_flags |= TOF_SACK; 3406 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3407 to->to_sacks = cp + 2; 3408 TCPSTAT_INC(tcps_sack_rcv_blocks); 3409 break; 3410#ifdef TCP_RFC7413 3411 case TCPOPT_FAST_OPEN: 3412 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3413 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3414 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3415 continue; 3416 if (!(flags & TO_SYN)) 3417 continue; 3418 if (!V_tcp_fastopen_enabled) 3419 continue; 3420 to->to_flags |= TOF_FASTOPEN; 3421 to->to_tfo_len = optlen - 2; 3422 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3423 break; 3424#endif 3425 default: 3426 continue; 3427 } 3428 } 3429} 3430 3431/* 3432 * Pull out of band byte out of a segment so 3433 * it doesn't appear in the user's data queue. 3434 * It is still reflected in the segment length for 3435 * sequencing purposes. 3436 */ 3437static void 3438tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3439 int off) 3440{ 3441 int cnt = off + th->th_urp - 1; 3442 3443 while (cnt >= 0) { 3444 if (m->m_len > cnt) { 3445 char *cp = mtod(m, caddr_t) + cnt; 3446 struct tcpcb *tp = sototcpcb(so); 3447 3448 INP_WLOCK_ASSERT(tp->t_inpcb); 3449 3450 tp->t_iobc = *cp; 3451 tp->t_oobflags |= TCPOOB_HAVEDATA; 3452 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3453 m->m_len--; 3454 if (m->m_flags & M_PKTHDR) 3455 m->m_pkthdr.len--; 3456 return; 3457 } 3458 cnt -= m->m_len; 3459 m = m->m_next; 3460 if (m == NULL) 3461 break; 3462 } 3463 panic("tcp_pulloutofband"); 3464} 3465 3466/* 3467 * Collect new round-trip time estimate 3468 * and update averages and current timeout. 3469 */ 3470static void 3471tcp_xmit_timer(struct tcpcb *tp, int rtt) 3472{ 3473 int delta; 3474 3475 INP_WLOCK_ASSERT(tp->t_inpcb); 3476 3477 TCPSTAT_INC(tcps_rttupdated); 3478 tp->t_rttupdated++; 3479 if (tp->t_srtt != 0) { 3480 /* 3481 * srtt is stored as fixed point with 5 bits after the 3482 * binary point (i.e., scaled by 8). The following magic 3483 * is equivalent to the smoothing algorithm in rfc793 with 3484 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3485 * point). Adjust rtt to origin 0. 3486 */ 3487 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3488 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3489 3490 if ((tp->t_srtt += delta) <= 0) 3491 tp->t_srtt = 1; 3492 3493 /* 3494 * We accumulate a smoothed rtt variance (actually, a 3495 * smoothed mean difference), then set the retransmit 3496 * timer to smoothed rtt + 4 times the smoothed variance. 3497 * rttvar is stored as fixed point with 4 bits after the 3498 * binary point (scaled by 16). The following is 3499 * equivalent to rfc793 smoothing with an alpha of .75 3500 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3501 * rfc793's wired-in beta. 3502 */ 3503 if (delta < 0) 3504 delta = -delta; 3505 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3506 if ((tp->t_rttvar += delta) <= 0) 3507 tp->t_rttvar = 1; 3508 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3509 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3510 } else { 3511 /* 3512 * No rtt measurement yet - use the unsmoothed rtt. 3513 * Set the variance to half the rtt (so our first 3514 * retransmit happens at 3*rtt). 3515 */ 3516 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3517 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3518 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3519 } 3520 tp->t_rtttime = 0; 3521 tp->t_rxtshift = 0; 3522 3523 /* 3524 * the retransmit should happen at rtt + 4 * rttvar. 3525 * Because of the way we do the smoothing, srtt and rttvar 3526 * will each average +1/2 tick of bias. When we compute 3527 * the retransmit timer, we want 1/2 tick of rounding and 3528 * 1 extra tick because of +-1/2 tick uncertainty in the 3529 * firing of the timer. The bias will give us exactly the 3530 * 1.5 tick we need. But, because the bias is 3531 * statistical, we have to test that we don't drop below 3532 * the minimum feasible timer (which is 2 ticks). 3533 */ 3534 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3535 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3536 3537 /* 3538 * We received an ack for a packet that wasn't retransmitted; 3539 * it is probably safe to discard any error indications we've 3540 * received recently. This isn't quite right, but close enough 3541 * for now (a route might have failed after we sent a segment, 3542 * and the return path might not be symmetrical). 3543 */ 3544 tp->t_softerror = 0; 3545} 3546 3547/* 3548 * Determine a reasonable value for maxseg size. 3549 * If the route is known, check route for mtu. 3550 * If none, use an mss that can be handled on the outgoing interface 3551 * without forcing IP to fragment. If no route is found, route has no mtu, 3552 * or the destination isn't local, use a default, hopefully conservative 3553 * size (usually 512 or the default IP max size, but no more than the mtu 3554 * of the interface), as we can't discover anything about intervening 3555 * gateways or networks. We also initialize the congestion/slow start 3556 * window to be a single segment if the destination isn't local. 3557 * While looking at the routing entry, we also initialize other path-dependent 3558 * parameters from pre-set or cached values in the routing entry. 3559 * 3560 * Also take into account the space needed for options that we 3561 * send regularly. Make maxseg shorter by that amount to assure 3562 * that we can send maxseg amount of data even when the options 3563 * are present. Store the upper limit of the length of options plus 3564 * data in maxopd. 3565 * 3566 * NOTE that this routine is only called when we process an incoming 3567 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3568 * settings are handled in tcp_mssopt(). 3569 */ 3570void 3571tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3572 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3573{ 3574 int mss = 0; 3575 u_long maxmtu = 0; 3576 struct inpcb *inp = tp->t_inpcb; 3577 struct hc_metrics_lite metrics; 3578 int origoffer; 3579#ifdef INET6 3580 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3581 size_t min_protoh = isipv6 ? 3582 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3583 sizeof (struct tcpiphdr); 3584#else 3585 const size_t min_protoh = sizeof(struct tcpiphdr); 3586#endif 3587 3588 INP_WLOCK_ASSERT(tp->t_inpcb); 3589 3590 if (mtuoffer != -1) { 3591 KASSERT(offer == -1, ("%s: conflict", __func__)); 3592 offer = mtuoffer - min_protoh; 3593 } 3594 origoffer = offer; 3595 3596 /* Initialize. */ 3597#ifdef INET6 3598 if (isipv6) { 3599 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3600 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3601 } 3602#endif 3603#if defined(INET) && defined(INET6) 3604 else 3605#endif 3606#ifdef INET 3607 { 3608 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3609 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3610 } 3611#endif 3612 3613 /* 3614 * No route to sender, stay with default mss and return. 3615 */ 3616 if (maxmtu == 0) { 3617 /* 3618 * In case we return early we need to initialize metrics 3619 * to a defined state as tcp_hc_get() would do for us 3620 * if there was no cache hit. 3621 */ 3622 if (metricptr != NULL) 3623 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3624 return; 3625 } 3626 3627 /* What have we got? */ 3628 switch (offer) { 3629 case 0: 3630 /* 3631 * Offer == 0 means that there was no MSS on the SYN 3632 * segment, in this case we use tcp_mssdflt as 3633 * already assigned to t_maxopd above. 3634 */ 3635 offer = tp->t_maxopd; 3636 break; 3637 3638 case -1: 3639 /* 3640 * Offer == -1 means that we didn't receive SYN yet. 3641 */ 3642 /* FALLTHROUGH */ 3643 3644 default: 3645 /* 3646 * Prevent DoS attack with too small MSS. Round up 3647 * to at least minmss. 3648 */ 3649 offer = max(offer, V_tcp_minmss); 3650 } 3651 3652 /* 3653 * rmx information is now retrieved from tcp_hostcache. 3654 */ 3655 tcp_hc_get(&inp->inp_inc, &metrics); 3656 if (metricptr != NULL) 3657 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3658 3659 /* 3660 * If there's a discovered mtu int tcp hostcache, use it 3661 * else, use the link mtu. 3662 */ 3663 if (metrics.rmx_mtu) 3664 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3665 else { 3666#ifdef INET6 3667 if (isipv6) { 3668 mss = maxmtu - min_protoh; 3669 if (!V_path_mtu_discovery && 3670 !in6_localaddr(&inp->in6p_faddr)) 3671 mss = min(mss, V_tcp_v6mssdflt); 3672 } 3673#endif 3674#if defined(INET) && defined(INET6) 3675 else 3676#endif 3677#ifdef INET 3678 { 3679 mss = maxmtu - min_protoh; 3680 if (!V_path_mtu_discovery && 3681 !in_localaddr(inp->inp_faddr)) 3682 mss = min(mss, V_tcp_mssdflt); 3683 } 3684#endif 3685 /* 3686 * XXX - The above conditional (mss = maxmtu - min_protoh) 3687 * probably violates the TCP spec. 3688 * The problem is that, since we don't know the 3689 * other end's MSS, we are supposed to use a conservative 3690 * default. But, if we do that, then MTU discovery will 3691 * never actually take place, because the conservative 3692 * default is much less than the MTUs typically seen 3693 * on the Internet today. For the moment, we'll sweep 3694 * this under the carpet. 3695 * 3696 * The conservative default might not actually be a problem 3697 * if the only case this occurs is when sending an initial 3698 * SYN with options and data to a host we've never talked 3699 * to before. Then, they will reply with an MSS value which 3700 * will get recorded and the new parameters should get 3701 * recomputed. For Further Study. 3702 */ 3703 } 3704 mss = min(mss, offer); 3705 3706 /* 3707 * Sanity check: make sure that maxopd will be large 3708 * enough to allow some data on segments even if the 3709 * all the option space is used (40bytes). Otherwise 3710 * funny things may happen in tcp_output. 3711 */ 3712 mss = max(mss, 64); 3713 3714 /* 3715 * maxopd stores the maximum length of data AND options 3716 * in a segment; maxseg is the amount of data in a normal 3717 * segment. We need to store this value (maxopd) apart 3718 * from maxseg, because now every segment carries options 3719 * and thus we normally have somewhat less data in segments. 3720 */ 3721 tp->t_maxopd = mss; 3722 3723 /* 3724 * origoffer==-1 indicates that no segments were received yet. 3725 * In this case we just guess. 3726 */ 3727 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3728 (origoffer == -1 || 3729 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3730 mss -= TCPOLEN_TSTAMP_APPA; 3731 3732 tp->t_maxseg = mss; 3733} 3734 3735void 3736tcp_mss(struct tcpcb *tp, int offer) 3737{ 3738 int mss; 3739 u_long bufsize; 3740 struct inpcb *inp; 3741 struct socket *so; 3742 struct hc_metrics_lite metrics; 3743 struct tcp_ifcap cap; 3744 3745 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3746 3747 bzero(&cap, sizeof(cap)); 3748 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3749 3750 mss = tp->t_maxseg; 3751 inp = tp->t_inpcb; 3752 3753 /* 3754 * If there's a pipesize, change the socket buffer to that size, 3755 * don't change if sb_hiwat is different than default (then it 3756 * has been changed on purpose with setsockopt). 3757 * Make the socket buffers an integral number of mss units; 3758 * if the mss is larger than the socket buffer, decrease the mss. 3759 */ 3760 so = inp->inp_socket; 3761 SOCKBUF_LOCK(&so->so_snd); 3762 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3763 bufsize = metrics.rmx_sendpipe; 3764 else 3765 bufsize = so->so_snd.sb_hiwat; 3766 if (bufsize < mss) 3767 mss = bufsize; 3768 else { 3769 bufsize = roundup(bufsize, mss); 3770 if (bufsize > sb_max) 3771 bufsize = sb_max; 3772 if (bufsize > so->so_snd.sb_hiwat) 3773 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3774 } 3775 SOCKBUF_UNLOCK(&so->so_snd); 3776 tp->t_maxseg = mss; 3777 3778 SOCKBUF_LOCK(&so->so_rcv); 3779 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3780 bufsize = metrics.rmx_recvpipe; 3781 else 3782 bufsize = so->so_rcv.sb_hiwat; 3783 if (bufsize > mss) { 3784 bufsize = roundup(bufsize, mss); 3785 if (bufsize > sb_max) 3786 bufsize = sb_max; 3787 if (bufsize > so->so_rcv.sb_hiwat) 3788 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3789 } 3790 SOCKBUF_UNLOCK(&so->so_rcv); 3791 3792 /* Check the interface for TSO capabilities. */ 3793 if (cap.ifcap & CSUM_TSO) { 3794 tp->t_flags |= TF_TSO; 3795 tp->t_tsomax = cap.tsomax; 3796 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3797 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3798 } 3799} 3800 3801/* 3802 * Determine the MSS option to send on an outgoing SYN. 3803 */ 3804int 3805tcp_mssopt(struct in_conninfo *inc) 3806{ 3807 int mss = 0; 3808 u_long maxmtu = 0; 3809 u_long thcmtu = 0; 3810 size_t min_protoh; 3811 3812 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3813 3814#ifdef INET6 3815 if (inc->inc_flags & INC_ISIPV6) { 3816 mss = V_tcp_v6mssdflt; 3817 maxmtu = tcp_maxmtu6(inc, NULL); 3818 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3819 } 3820#endif 3821#if defined(INET) && defined(INET6) 3822 else 3823#endif 3824#ifdef INET 3825 { 3826 mss = V_tcp_mssdflt; 3827 maxmtu = tcp_maxmtu(inc, NULL); 3828 min_protoh = sizeof(struct tcpiphdr); 3829 } 3830#endif 3831#if defined(INET6) || defined(INET) 3832 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3833#endif 3834 3835 if (maxmtu && thcmtu) 3836 mss = min(maxmtu, thcmtu) - min_protoh; 3837 else if (maxmtu || thcmtu) 3838 mss = max(maxmtu, thcmtu) - min_protoh; 3839 3840 return (mss); 3841} 3842 3843 3844/* 3845 * On a partial ack arrives, force the retransmission of the 3846 * next unacknowledged segment. Do not clear tp->t_dupacks. 3847 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3848 * be started again. 3849 */ 3850static void 3851tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3852{ 3853 tcp_seq onxt = tp->snd_nxt; 3854 u_long ocwnd = tp->snd_cwnd; 3855 3856 INP_WLOCK_ASSERT(tp->t_inpcb); 3857 3858 tcp_timer_activate(tp, TT_REXMT, 0); 3859 tp->t_rtttime = 0; 3860 tp->snd_nxt = th->th_ack; 3861 /* 3862 * Set snd_cwnd to one segment beyond acknowledged offset. 3863 * (tp->snd_una has not yet been updated when this function is called.) 3864 */ 3865 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3866 tp->t_flags |= TF_ACKNOW; 3867 (void) tcp_output(tp); 3868 tp->snd_cwnd = ocwnd; 3869 if (SEQ_GT(onxt, tp->snd_nxt)) 3870 tp->snd_nxt = onxt; 3871 /* 3872 * Partial window deflation. Relies on fact that tp->snd_una 3873 * not updated yet. 3874 */ 3875 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3876 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3877 else 3878 tp->snd_cwnd = 0; 3879 tp->snd_cwnd += tp->t_maxseg; 3880} 3881 3882int 3883tcp_compute_pipe(struct tcpcb *tp) 3884{ 3885 return (tp->snd_max - tp->snd_una + 3886 tp->sackhint.sack_bytes_rexmit - 3887 tp->sackhint.sacked_bytes); 3888} 3889