tcp_input.c revision 309108
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 309108 2016-11-24 14:48:46Z jch $"); 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 1472static void 1473tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1474 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1475 int ti_locked) 1476{ 1477 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1478 int rstreason, todrop, win; 1479 u_long tiwin; 1480 char *s; 1481 struct in_conninfo *inc; 1482 struct mbuf *mfree; 1483 struct tcpopt to; 1484 int tfo_syn; 1485 1486#ifdef TCPDEBUG 1487 /* 1488 * The size of tcp_saveipgen must be the size of the max ip header, 1489 * now IPv6. 1490 */ 1491 u_char tcp_saveipgen[IP6_HDR_LEN]; 1492 struct tcphdr tcp_savetcp; 1493 short ostate = 0; 1494#endif 1495 thflags = th->th_flags; 1496 inc = &tp->t_inpcb->inp_inc; 1497 tp->sackhint.last_sack_ack = 0; 1498 sack_changed = 0; 1499 1500 /* 1501 * If this is either a state-changing packet or current state isn't 1502 * established, we require a write lock on tcbinfo. Otherwise, we 1503 * allow the tcbinfo to be in either alocked or unlocked, as the 1504 * caller may have unnecessarily acquired a write lock due to a race. 1505 */ 1506 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1507 tp->t_state != TCPS_ESTABLISHED) { 1508 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1509 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1510 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1511 } else { 1512#ifdef INVARIANTS 1513 if (ti_locked == TI_RLOCKED) 1514 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1515 else { 1516 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1517 "ti_locked: %d", __func__, ti_locked)); 1518 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1519 } 1520#endif 1521 } 1522 INP_WLOCK_ASSERT(tp->t_inpcb); 1523 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1524 __func__)); 1525 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1526 __func__)); 1527 1528 /* 1529 * Segment received on connection. 1530 * Reset idle time and keep-alive timer. 1531 * XXX: This should be done after segment 1532 * validation to ignore broken/spoofed segs. 1533 */ 1534 tp->t_rcvtime = ticks; 1535 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1536 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1537 1538 /* 1539 * Unscale the window into a 32-bit value. 1540 * For the SYN_SENT state the scale is zero. 1541 */ 1542 tiwin = th->th_win << tp->snd_scale; 1543 1544 /* 1545 * TCP ECN processing. 1546 */ 1547 if (tp->t_flags & TF_ECN_PERMIT) { 1548 if (thflags & TH_CWR) 1549 tp->t_flags &= ~TF_ECN_SND_ECE; 1550 switch (iptos & IPTOS_ECN_MASK) { 1551 case IPTOS_ECN_CE: 1552 tp->t_flags |= TF_ECN_SND_ECE; 1553 TCPSTAT_INC(tcps_ecn_ce); 1554 break; 1555 case IPTOS_ECN_ECT0: 1556 TCPSTAT_INC(tcps_ecn_ect0); 1557 break; 1558 case IPTOS_ECN_ECT1: 1559 TCPSTAT_INC(tcps_ecn_ect1); 1560 break; 1561 } 1562 /* Congestion experienced. */ 1563 if (thflags & TH_ECE) { 1564 cc_cong_signal(tp, th, CC_ECN); 1565 } 1566 } 1567 1568 /* 1569 * Parse options on any incoming segment. 1570 */ 1571 tcp_dooptions(&to, (u_char *)(th + 1), 1572 (th->th_off << 2) - sizeof(struct tcphdr), 1573 (thflags & TH_SYN) ? TO_SYN : 0); 1574 1575 /* 1576 * If echoed timestamp is later than the current time, 1577 * fall back to non RFC1323 RTT calculation. Normalize 1578 * timestamp if syncookies were used when this connection 1579 * was established. 1580 */ 1581 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1582 to.to_tsecr -= tp->ts_offset; 1583 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1584 to.to_tsecr = 0; 1585 } 1586 /* 1587 * If timestamps were negotiated during SYN/ACK they should 1588 * appear on every segment during this session and vice versa. 1589 */ 1590 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1591 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1592 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1593 "no action\n", s, __func__); 1594 free(s, M_TCPLOG); 1595 } 1596 } 1597 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1598 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1599 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1600 "no action\n", s, __func__); 1601 free(s, M_TCPLOG); 1602 } 1603 } 1604 1605 /* 1606 * Process options only when we get SYN/ACK back. The SYN case 1607 * for incoming connections is handled in tcp_syncache. 1608 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1609 * or <SYN,ACK>) segment itself is never scaled. 1610 * XXX this is traditional behavior, may need to be cleaned up. 1611 */ 1612 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1613 if ((to.to_flags & TOF_SCALE) && 1614 (tp->t_flags & TF_REQ_SCALE)) { 1615 tp->t_flags |= TF_RCVD_SCALE; 1616 tp->snd_scale = to.to_wscale; 1617 } 1618 /* 1619 * Initial send window. It will be updated with 1620 * the next incoming segment to the scaled value. 1621 */ 1622 tp->snd_wnd = th->th_win; 1623 if (to.to_flags & TOF_TS) { 1624 tp->t_flags |= TF_RCVD_TSTMP; 1625 tp->ts_recent = to.to_tsval; 1626 tp->ts_recent_age = tcp_ts_getticks(); 1627 } 1628 if (to.to_flags & TOF_MSS) 1629 tcp_mss(tp, to.to_mss); 1630 if ((tp->t_flags & TF_SACK_PERMIT) && 1631 (to.to_flags & TOF_SACKPERM) == 0) 1632 tp->t_flags &= ~TF_SACK_PERMIT; 1633 } 1634 1635 /* 1636 * Header prediction: check for the two common cases 1637 * of a uni-directional data xfer. If the packet has 1638 * no control flags, is in-sequence, the window didn't 1639 * change and we're not retransmitting, it's a 1640 * candidate. If the length is zero and the ack moved 1641 * forward, we're the sender side of the xfer. Just 1642 * free the data acked & wake any higher level process 1643 * that was blocked waiting for space. If the length 1644 * is non-zero and the ack didn't move, we're the 1645 * receiver side. If we're getting packets in-order 1646 * (the reassembly queue is empty), add the data to 1647 * the socket buffer and note that we need a delayed ack. 1648 * Make sure that the hidden state-flags are also off. 1649 * Since we check for TCPS_ESTABLISHED first, it can only 1650 * be TH_NEEDSYN. 1651 */ 1652 if (tp->t_state == TCPS_ESTABLISHED && 1653 th->th_seq == tp->rcv_nxt && 1654 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1655 tp->snd_nxt == tp->snd_max && 1656 tiwin && tiwin == tp->snd_wnd && 1657 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1658 LIST_EMPTY(&tp->t_segq) && 1659 ((to.to_flags & TOF_TS) == 0 || 1660 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1661 1662 /* 1663 * If last ACK falls within this segment's sequence numbers, 1664 * record the timestamp. 1665 * NOTE that the test is modified according to the latest 1666 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1667 */ 1668 if ((to.to_flags & TOF_TS) != 0 && 1669 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1670 tp->ts_recent_age = tcp_ts_getticks(); 1671 tp->ts_recent = to.to_tsval; 1672 } 1673 1674 if (tlen == 0) { 1675 if (SEQ_GT(th->th_ack, tp->snd_una) && 1676 SEQ_LEQ(th->th_ack, tp->snd_max) && 1677 !IN_RECOVERY(tp->t_flags) && 1678 (to.to_flags & TOF_SACK) == 0 && 1679 TAILQ_EMPTY(&tp->snd_holes)) { 1680 /* 1681 * This is a pure ack for outstanding data. 1682 */ 1683 if (ti_locked == TI_RLOCKED) 1684 INP_INFO_RUNLOCK(&V_tcbinfo); 1685 ti_locked = TI_UNLOCKED; 1686 1687 TCPSTAT_INC(tcps_predack); 1688 1689 /* 1690 * "bad retransmit" recovery. 1691 */ 1692 if (tp->t_rxtshift == 1 && 1693 tp->t_flags & TF_PREVVALID && 1694 (int)(ticks - tp->t_badrxtwin) < 0) { 1695 cc_cong_signal(tp, th, CC_RTO_ERR); 1696 } 1697 1698 /* 1699 * Recalculate the transmit timer / rtt. 1700 * 1701 * Some boxes send broken timestamp replies 1702 * during the SYN+ACK phase, ignore 1703 * timestamps of 0 or we could calculate a 1704 * huge RTT and blow up the retransmit timer. 1705 */ 1706 if ((to.to_flags & TOF_TS) != 0 && 1707 to.to_tsecr) { 1708 u_int t; 1709 1710 t = tcp_ts_getticks() - to.to_tsecr; 1711 if (!tp->t_rttlow || tp->t_rttlow > t) 1712 tp->t_rttlow = t; 1713 tcp_xmit_timer(tp, 1714 TCP_TS_TO_TICKS(t) + 1); 1715 } else if (tp->t_rtttime && 1716 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1717 if (!tp->t_rttlow || 1718 tp->t_rttlow > ticks - tp->t_rtttime) 1719 tp->t_rttlow = ticks - tp->t_rtttime; 1720 tcp_xmit_timer(tp, 1721 ticks - tp->t_rtttime); 1722 } 1723 acked = BYTES_THIS_ACK(tp, th); 1724 1725 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1726 hhook_run_tcp_est_in(tp, th, &to); 1727 1728 TCPSTAT_INC(tcps_rcvackpack); 1729 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1730 sbdrop(&so->so_snd, acked); 1731 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1732 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1733 tp->snd_recover = th->th_ack - 1; 1734 1735 /* 1736 * Let the congestion control algorithm update 1737 * congestion control related information. This 1738 * typically means increasing the congestion 1739 * window. 1740 */ 1741 cc_ack_received(tp, th, CC_ACK); 1742 1743 tp->snd_una = th->th_ack; 1744 /* 1745 * Pull snd_wl2 up to prevent seq wrap relative 1746 * to th_ack. 1747 */ 1748 tp->snd_wl2 = th->th_ack; 1749 tp->t_dupacks = 0; 1750 m_freem(m); 1751 ND6_HINT(tp); /* Some progress has been made. */ 1752 1753 /* 1754 * If all outstanding data are acked, stop 1755 * retransmit timer, otherwise restart timer 1756 * using current (possibly backed-off) value. 1757 * If process is waiting for space, 1758 * wakeup/selwakeup/signal. If data 1759 * are ready to send, let tcp_output 1760 * decide between more output or persist. 1761 */ 1762#ifdef TCPDEBUG 1763 if (so->so_options & SO_DEBUG) 1764 tcp_trace(TA_INPUT, ostate, tp, 1765 (void *)tcp_saveipgen, 1766 &tcp_savetcp, 0); 1767#endif 1768 if (tp->snd_una == tp->snd_max) 1769 tcp_timer_activate(tp, TT_REXMT, 0); 1770 else if (!tcp_timer_active(tp, TT_PERSIST)) 1771 tcp_timer_activate(tp, TT_REXMT, 1772 tp->t_rxtcur); 1773 sowwakeup(so); 1774 if (so->so_snd.sb_cc) 1775 (void) tcp_output(tp); 1776 goto check_delack; 1777 } 1778 } else if (th->th_ack == tp->snd_una && 1779 tlen <= sbspace(&so->so_rcv)) { 1780 int newsize = 0; /* automatic sockbuf scaling */ 1781 1782 /* 1783 * This is a pure, in-sequence data packet with 1784 * nothing on the reassembly queue and we have enough 1785 * buffer space to take it. 1786 */ 1787 if (ti_locked == TI_RLOCKED) 1788 INP_INFO_RUNLOCK(&V_tcbinfo); 1789 ti_locked = TI_UNLOCKED; 1790 1791 /* Clean receiver SACK report if present */ 1792 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1793 tcp_clean_sackreport(tp); 1794 TCPSTAT_INC(tcps_preddat); 1795 tp->rcv_nxt += tlen; 1796 /* 1797 * Pull snd_wl1 up to prevent seq wrap relative to 1798 * th_seq. 1799 */ 1800 tp->snd_wl1 = th->th_seq; 1801 /* 1802 * Pull rcv_up up to prevent seq wrap relative to 1803 * rcv_nxt. 1804 */ 1805 tp->rcv_up = tp->rcv_nxt; 1806 TCPSTAT_INC(tcps_rcvpack); 1807 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1808 ND6_HINT(tp); /* Some progress has been made */ 1809#ifdef TCPDEBUG 1810 if (so->so_options & SO_DEBUG) 1811 tcp_trace(TA_INPUT, ostate, tp, 1812 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1813#endif 1814 /* 1815 * Automatic sizing of receive socket buffer. Often the send 1816 * buffer size is not optimally adjusted to the actual network 1817 * conditions at hand (delay bandwidth product). Setting the 1818 * buffer size too small limits throughput on links with high 1819 * bandwidth and high delay (eg. trans-continental/oceanic links). 1820 * 1821 * On the receive side the socket buffer memory is only rarely 1822 * used to any significant extent. This allows us to be much 1823 * more aggressive in scaling the receive socket buffer. For 1824 * the case that the buffer space is actually used to a large 1825 * extent and we run out of kernel memory we can simply drop 1826 * the new segments; TCP on the sender will just retransmit it 1827 * later. Setting the buffer size too big may only consume too 1828 * much kernel memory if the application doesn't read() from 1829 * the socket or packet loss or reordering makes use of the 1830 * reassembly queue. 1831 * 1832 * The criteria to step up the receive buffer one notch are: 1833 * 1. the number of bytes received during the time it takes 1834 * one timestamp to be reflected back to us (the RTT); 1835 * 2. received bytes per RTT is within seven eighth of the 1836 * current socket buffer size; 1837 * 3. receive buffer size has not hit maximal automatic size; 1838 * 1839 * This algorithm does one step per RTT at most and only if 1840 * we receive a bulk stream w/o packet losses or reorderings. 1841 * Shrinking the buffer during idle times is not necessary as 1842 * it doesn't consume any memory when idle. 1843 * 1844 * TODO: Only step up if the application is actually serving 1845 * the buffer to better manage the socket buffer resources. 1846 */ 1847 if (V_tcp_do_autorcvbuf && 1848 (to.to_flags & TOF_TS) && 1849 to.to_tsecr && 1850 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1851 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1852 to.to_tsecr - tp->rfbuf_ts < hz) { 1853 if (tp->rfbuf_cnt > 1854 (so->so_rcv.sb_hiwat / 8 * 7) && 1855 so->so_rcv.sb_hiwat < 1856 V_tcp_autorcvbuf_max) { 1857 newsize = 1858 min(so->so_rcv.sb_hiwat + 1859 V_tcp_autorcvbuf_inc, 1860 V_tcp_autorcvbuf_max); 1861 } 1862 /* Start over with next RTT. */ 1863 tp->rfbuf_ts = 0; 1864 tp->rfbuf_cnt = 0; 1865 } else 1866 tp->rfbuf_cnt += tlen; /* add up */ 1867 } 1868 1869 /* Add data to socket buffer. */ 1870 SOCKBUF_LOCK(&so->so_rcv); 1871 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1872 m_freem(m); 1873 } else { 1874 /* 1875 * Set new socket buffer size. 1876 * Give up when limit is reached. 1877 */ 1878 if (newsize) 1879 if (!sbreserve_locked(&so->so_rcv, 1880 newsize, so, NULL)) 1881 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1882 m_adj(m, drop_hdrlen); /* delayed header drop */ 1883 sbappendstream_locked(&so->so_rcv, m); 1884 } 1885 /* NB: sorwakeup_locked() does an implicit unlock. */ 1886 sorwakeup_locked(so); 1887 if (DELAY_ACK(tp, tlen)) { 1888 tp->t_flags |= TF_DELACK; 1889 } else { 1890 tp->t_flags |= TF_ACKNOW; 1891 tcp_output(tp); 1892 } 1893 goto check_delack; 1894 } 1895 } 1896 1897 /* 1898 * Calculate amount of space in receive window, 1899 * and then do TCP input processing. 1900 * Receive window is amount of space in rcv queue, 1901 * but not less than advertised window. 1902 */ 1903 win = sbspace(&so->so_rcv); 1904 if (win < 0) 1905 win = 0; 1906 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1907 1908 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1909 tp->rfbuf_ts = 0; 1910 tp->rfbuf_cnt = 0; 1911 1912 switch (tp->t_state) { 1913 1914 /* 1915 * If the state is SYN_RECEIVED: 1916 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1917 */ 1918 case TCPS_SYN_RECEIVED: 1919 if ((thflags & TH_ACK) && 1920 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1921 SEQ_GT(th->th_ack, tp->snd_max))) { 1922 rstreason = BANDLIM_RST_OPENPORT; 1923 goto dropwithreset; 1924 } 1925#ifdef TCP_RFC7413 1926 if (tp->t_flags & TF_FASTOPEN) { 1927 /* 1928 * When a TFO connection is in SYN_RECEIVED, the 1929 * only valid packets are the initial SYN, a 1930 * retransmit/copy of the initial SYN (possibly with 1931 * a subset of the original data), a valid ACK, a 1932 * FIN, or a RST. 1933 */ 1934 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1935 rstreason = BANDLIM_RST_OPENPORT; 1936 goto dropwithreset; 1937 } else if (thflags & TH_SYN) { 1938 /* non-initial SYN is ignored */ 1939 if ((tcp_timer_active(tp, TT_DELACK) || 1940 tcp_timer_active(tp, TT_REXMT))) 1941 goto drop; 1942 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1943 goto drop; 1944 } 1945 } 1946#endif 1947 break; 1948 1949 /* 1950 * If the state is SYN_SENT: 1951 * if seg contains an ACK, but not for our SYN, drop the input. 1952 * if seg contains a RST, then drop the connection. 1953 * if seg does not contain SYN, then drop it. 1954 * Otherwise this is an acceptable SYN segment 1955 * initialize tp->rcv_nxt and tp->irs 1956 * if seg contains ack then advance tp->snd_una 1957 * if seg contains an ECE and ECN support is enabled, the stream 1958 * is ECN capable. 1959 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1960 * arrange for segment to be acked (eventually) 1961 * continue processing rest of data/controls, beginning with URG 1962 */ 1963 case TCPS_SYN_SENT: 1964 if ((thflags & TH_ACK) && 1965 (SEQ_LEQ(th->th_ack, tp->iss) || 1966 SEQ_GT(th->th_ack, tp->snd_max))) { 1967 rstreason = BANDLIM_UNLIMITED; 1968 goto dropwithreset; 1969 } 1970 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1971 TCP_PROBE5(connect__refused, NULL, tp, 1972 mtod(m, const char *), tp, th); 1973 tp = tcp_drop(tp, ECONNREFUSED); 1974 } 1975 if (thflags & TH_RST) 1976 goto drop; 1977 if (!(thflags & TH_SYN)) 1978 goto drop; 1979 1980 tp->irs = th->th_seq; 1981 tcp_rcvseqinit(tp); 1982 if (thflags & TH_ACK) { 1983 TCPSTAT_INC(tcps_connects); 1984 soisconnected(so); 1985#ifdef MAC 1986 mac_socketpeer_set_from_mbuf(m, so); 1987#endif 1988 /* Do window scaling on this connection? */ 1989 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1990 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1991 tp->rcv_scale = tp->request_r_scale; 1992 } 1993 tp->rcv_adv += imin(tp->rcv_wnd, 1994 TCP_MAXWIN << tp->rcv_scale); 1995 tp->snd_una++; /* SYN is acked */ 1996 /* 1997 * If there's data, delay ACK; if there's also a FIN 1998 * ACKNOW will be turned on later. 1999 */ 2000 if (DELAY_ACK(tp, tlen) && tlen != 0) 2001 tcp_timer_activate(tp, TT_DELACK, 2002 tcp_delacktime); 2003 else 2004 tp->t_flags |= TF_ACKNOW; 2005 2006 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2007 tp->t_flags |= TF_ECN_PERMIT; 2008 TCPSTAT_INC(tcps_ecn_shs); 2009 } 2010 2011 /* 2012 * Received <SYN,ACK> in SYN_SENT[*] state. 2013 * Transitions: 2014 * SYN_SENT --> ESTABLISHED 2015 * SYN_SENT* --> FIN_WAIT_1 2016 */ 2017 tp->t_starttime = ticks; 2018 if (tp->t_flags & TF_NEEDFIN) { 2019 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2020 tp->t_flags &= ~TF_NEEDFIN; 2021 thflags &= ~TH_SYN; 2022 } else { 2023 tcp_state_change(tp, TCPS_ESTABLISHED); 2024 TCP_PROBE5(connect__established, NULL, tp, 2025 mtod(m, const char *), tp, th); 2026 cc_conn_init(tp); 2027 tcp_timer_activate(tp, TT_KEEP, 2028 TP_KEEPIDLE(tp)); 2029 } 2030 } else { 2031 /* 2032 * Received initial SYN in SYN-SENT[*] state => 2033 * simultaneous open. If segment contains CC option 2034 * and there is a cached CC, apply TAO test. 2035 * If it succeeds, connection is * half-synchronized. 2036 * Otherwise, do 3-way handshake: 2037 * SYN-SENT -> SYN-RECEIVED 2038 * SYN-SENT* -> SYN-RECEIVED* 2039 * If there was no CC option, clear cached CC value. 2040 */ 2041 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2042 tcp_timer_activate(tp, TT_REXMT, 0); 2043 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2044 } 2045 2046 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2047 "ti_locked %d", __func__, ti_locked)); 2048 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2049 INP_WLOCK_ASSERT(tp->t_inpcb); 2050 2051 /* 2052 * Advance th->th_seq to correspond to first data byte. 2053 * If data, trim to stay within window, 2054 * dropping FIN if necessary. 2055 */ 2056 th->th_seq++; 2057 if (tlen > tp->rcv_wnd) { 2058 todrop = tlen - tp->rcv_wnd; 2059 m_adj(m, -todrop); 2060 tlen = tp->rcv_wnd; 2061 thflags &= ~TH_FIN; 2062 TCPSTAT_INC(tcps_rcvpackafterwin); 2063 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2064 } 2065 tp->snd_wl1 = th->th_seq - 1; 2066 tp->rcv_up = th->th_seq; 2067 /* 2068 * Client side of transaction: already sent SYN and data. 2069 * If the remote host used T/TCP to validate the SYN, 2070 * our data will be ACK'd; if so, enter normal data segment 2071 * processing in the middle of step 5, ack processing. 2072 * Otherwise, goto step 6. 2073 */ 2074 if (thflags & TH_ACK) 2075 goto process_ACK; 2076 2077 goto step6; 2078 2079 /* 2080 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2081 * do normal processing. 2082 * 2083 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2084 */ 2085 case TCPS_LAST_ACK: 2086 case TCPS_CLOSING: 2087 break; /* continue normal processing */ 2088 } 2089 2090 /* 2091 * States other than LISTEN or SYN_SENT. 2092 * First check the RST flag and sequence number since reset segments 2093 * are exempt from the timestamp and connection count tests. This 2094 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2095 * below which allowed reset segments in half the sequence space 2096 * to fall though and be processed (which gives forged reset 2097 * segments with a random sequence number a 50 percent chance of 2098 * killing a connection). 2099 * Then check timestamp, if present. 2100 * Then check the connection count, if present. 2101 * Then check that at least some bytes of segment are within 2102 * receive window. If segment begins before rcv_nxt, 2103 * drop leading data (and SYN); if nothing left, just ack. 2104 * 2105 * 2106 * If the RST bit is set, check the sequence number to see 2107 * if this is a valid reset segment. 2108 * RFC 793 page 37: 2109 * In all states except SYN-SENT, all reset (RST) segments 2110 * are validated by checking their SEQ-fields. A reset is 2111 * valid if its sequence number is in the window. 2112 * Note: this does not take into account delayed ACKs, so 2113 * we should test against last_ack_sent instead of rcv_nxt. 2114 * The sequence number in the reset segment is normally an 2115 * echo of our outgoing acknowlegement numbers, but some hosts 2116 * send a reset with the sequence number at the rightmost edge 2117 * of our receive window, and we have to handle this case. 2118 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2119 * that brute force RST attacks are possible. To combat this, 2120 * we use a much stricter check while in the ESTABLISHED state, 2121 * only accepting RSTs where the sequence number is equal to 2122 * last_ack_sent. In all other states (the states in which a 2123 * RST is more likely), the more permissive check is used. 2124 * If we have multiple segments in flight, the initial reset 2125 * segment sequence numbers will be to the left of last_ack_sent, 2126 * but they will eventually catch up. 2127 * In any case, it never made sense to trim reset segments to 2128 * fit the receive window since RFC 1122 says: 2129 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2130 * 2131 * A TCP SHOULD allow a received RST segment to include data. 2132 * 2133 * DISCUSSION 2134 * It has been suggested that a RST segment could contain 2135 * ASCII text that encoded and explained the cause of the 2136 * RST. No standard has yet been established for such 2137 * data. 2138 * 2139 * If the reset segment passes the sequence number test examine 2140 * the state: 2141 * SYN_RECEIVED STATE: 2142 * If passive open, return to LISTEN state. 2143 * If active open, inform user that connection was refused. 2144 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2145 * Inform user that connection was reset, and close tcb. 2146 * CLOSING, LAST_ACK STATES: 2147 * Close the tcb. 2148 * TIME_WAIT STATE: 2149 * Drop the segment - see Stevens, vol. 2, p. 964 and 2150 * RFC 1337. 2151 */ 2152 if (thflags & TH_RST) { 2153 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2154 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2155 switch (tp->t_state) { 2156 2157 case TCPS_SYN_RECEIVED: 2158 so->so_error = ECONNREFUSED; 2159 goto close; 2160 2161 case TCPS_ESTABLISHED: 2162 if (V_tcp_insecure_rst == 0 && 2163 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2164 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2165 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2166 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2167 TCPSTAT_INC(tcps_badrst); 2168 goto drop; 2169 } 2170 /* FALLTHROUGH */ 2171 case TCPS_FIN_WAIT_1: 2172 case TCPS_FIN_WAIT_2: 2173 case TCPS_CLOSE_WAIT: 2174 so->so_error = ECONNRESET; 2175 close: 2176 KASSERT(ti_locked == TI_RLOCKED, 2177 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2178 ti_locked)); 2179 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2180 2181 tcp_state_change(tp, TCPS_CLOSED); 2182 TCPSTAT_INC(tcps_drops); 2183 tp = tcp_close(tp); 2184 break; 2185 2186 case TCPS_CLOSING: 2187 case TCPS_LAST_ACK: 2188 KASSERT(ti_locked == TI_RLOCKED, 2189 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2190 ti_locked)); 2191 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2192 2193 tp = tcp_close(tp); 2194 break; 2195 } 2196 } 2197 goto drop; 2198 } 2199 2200 /* 2201 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2202 * and it's less than ts_recent, drop it. 2203 */ 2204 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2205 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2206 2207 /* Check to see if ts_recent is over 24 days old. */ 2208 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2209 /* 2210 * Invalidate ts_recent. If this segment updates 2211 * ts_recent, the age will be reset later and ts_recent 2212 * will get a valid value. If it does not, setting 2213 * ts_recent to zero will at least satisfy the 2214 * requirement that zero be placed in the timestamp 2215 * echo reply when ts_recent isn't valid. The 2216 * age isn't reset until we get a valid ts_recent 2217 * because we don't want out-of-order segments to be 2218 * dropped when ts_recent is old. 2219 */ 2220 tp->ts_recent = 0; 2221 } else { 2222 TCPSTAT_INC(tcps_rcvduppack); 2223 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2224 TCPSTAT_INC(tcps_pawsdrop); 2225 if (tlen) 2226 goto dropafterack; 2227 goto drop; 2228 } 2229 } 2230 2231 /* 2232 * In the SYN-RECEIVED state, validate that the packet belongs to 2233 * this connection before trimming the data to fit the receive 2234 * window. Check the sequence number versus IRS since we know 2235 * the sequence numbers haven't wrapped. This is a partial fix 2236 * for the "LAND" DoS attack. 2237 */ 2238 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2239 rstreason = BANDLIM_RST_OPENPORT; 2240 goto dropwithreset; 2241 } 2242 2243 todrop = tp->rcv_nxt - th->th_seq; 2244 if (todrop > 0) { 2245 if (thflags & TH_SYN) { 2246 thflags &= ~TH_SYN; 2247 th->th_seq++; 2248 if (th->th_urp > 1) 2249 th->th_urp--; 2250 else 2251 thflags &= ~TH_URG; 2252 todrop--; 2253 } 2254 /* 2255 * Following if statement from Stevens, vol. 2, p. 960. 2256 */ 2257 if (todrop > tlen 2258 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2259 /* 2260 * Any valid FIN must be to the left of the window. 2261 * At this point the FIN must be a duplicate or out 2262 * of sequence; drop it. 2263 */ 2264 thflags &= ~TH_FIN; 2265 2266 /* 2267 * Send an ACK to resynchronize and drop any data. 2268 * But keep on processing for RST or ACK. 2269 */ 2270 tp->t_flags |= TF_ACKNOW; 2271 todrop = tlen; 2272 TCPSTAT_INC(tcps_rcvduppack); 2273 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2274 } else { 2275 TCPSTAT_INC(tcps_rcvpartduppack); 2276 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2277 } 2278 drop_hdrlen += todrop; /* drop from the top afterwards */ 2279 th->th_seq += todrop; 2280 tlen -= todrop; 2281 if (th->th_urp > todrop) 2282 th->th_urp -= todrop; 2283 else { 2284 thflags &= ~TH_URG; 2285 th->th_urp = 0; 2286 } 2287 } 2288 2289 /* 2290 * If new data are received on a connection after the 2291 * user processes are gone, then RST the other end. 2292 */ 2293 if ((so->so_state & SS_NOFDREF) && 2294 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2295 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2296 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2297 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2298 2299 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2300 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2301 "after socket was closed, " 2302 "sending RST and removing tcpcb\n", 2303 s, __func__, tcpstates[tp->t_state], tlen); 2304 free(s, M_TCPLOG); 2305 } 2306 tp = tcp_close(tp); 2307 TCPSTAT_INC(tcps_rcvafterclose); 2308 rstreason = BANDLIM_UNLIMITED; 2309 goto dropwithreset; 2310 } 2311 2312 /* 2313 * If segment ends after window, drop trailing data 2314 * (and PUSH and FIN); if nothing left, just ACK. 2315 */ 2316 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2317 if (todrop > 0) { 2318 TCPSTAT_INC(tcps_rcvpackafterwin); 2319 if (todrop >= tlen) { 2320 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2321 /* 2322 * If window is closed can only take segments at 2323 * window edge, and have to drop data and PUSH from 2324 * incoming segments. Continue processing, but 2325 * remember to ack. Otherwise, drop segment 2326 * and ack. 2327 */ 2328 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2329 tp->t_flags |= TF_ACKNOW; 2330 TCPSTAT_INC(tcps_rcvwinprobe); 2331 } else 2332 goto dropafterack; 2333 } else 2334 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2335 m_adj(m, -todrop); 2336 tlen -= todrop; 2337 thflags &= ~(TH_PUSH|TH_FIN); 2338 } 2339 2340 /* 2341 * If last ACK falls within this segment's sequence numbers, 2342 * record its timestamp. 2343 * NOTE: 2344 * 1) That the test incorporates suggestions from the latest 2345 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2346 * 2) That updating only on newer timestamps interferes with 2347 * our earlier PAWS tests, so this check should be solely 2348 * predicated on the sequence space of this segment. 2349 * 3) That we modify the segment boundary check to be 2350 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2351 * instead of RFC1323's 2352 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2353 * This modified check allows us to overcome RFC1323's 2354 * limitations as described in Stevens TCP/IP Illustrated 2355 * Vol. 2 p.869. In such cases, we can still calculate the 2356 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2357 */ 2358 if ((to.to_flags & TOF_TS) != 0 && 2359 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2360 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2361 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2362 tp->ts_recent_age = tcp_ts_getticks(); 2363 tp->ts_recent = to.to_tsval; 2364 } 2365 2366 /* 2367 * If a SYN is in the window, then this is an 2368 * error and we send an RST and drop the connection. 2369 */ 2370 if (thflags & TH_SYN) { 2371 KASSERT(ti_locked == TI_RLOCKED, 2372 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2373 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2374 2375 tp = tcp_drop(tp, ECONNRESET); 2376 rstreason = BANDLIM_UNLIMITED; 2377 goto drop; 2378 } 2379 2380 /* 2381 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2382 * flag is on (half-synchronized state), then queue data for 2383 * later processing; else drop segment and return. 2384 */ 2385 if ((thflags & TH_ACK) == 0) { 2386 if (tp->t_state == TCPS_SYN_RECEIVED || 2387 (tp->t_flags & TF_NEEDSYN)) { 2388#ifdef TCP_RFC7413 2389 if (tp->t_state == TCPS_SYN_RECEIVED && 2390 tp->t_flags & TF_FASTOPEN) { 2391 tp->snd_wnd = tiwin; 2392 cc_conn_init(tp); 2393 } 2394#endif 2395 goto step6; 2396 } else if (tp->t_flags & TF_ACKNOW) 2397 goto dropafterack; 2398 else 2399 goto drop; 2400 } 2401 2402 /* 2403 * Ack processing. 2404 */ 2405 switch (tp->t_state) { 2406 2407 /* 2408 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2409 * ESTABLISHED state and continue processing. 2410 * The ACK was checked above. 2411 */ 2412 case TCPS_SYN_RECEIVED: 2413 2414 TCPSTAT_INC(tcps_connects); 2415 soisconnected(so); 2416 /* Do window scaling? */ 2417 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2418 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2419 tp->rcv_scale = tp->request_r_scale; 2420 tp->snd_wnd = tiwin; 2421 } 2422 /* 2423 * Make transitions: 2424 * SYN-RECEIVED -> ESTABLISHED 2425 * SYN-RECEIVED* -> FIN-WAIT-1 2426 */ 2427 tp->t_starttime = ticks; 2428 if (tp->t_flags & TF_NEEDFIN) { 2429 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2430 tp->t_flags &= ~TF_NEEDFIN; 2431 } else { 2432 tcp_state_change(tp, TCPS_ESTABLISHED); 2433 TCP_PROBE5(accept__established, NULL, tp, 2434 mtod(m, const char *), tp, th); 2435#ifdef TCP_RFC7413 2436 if (tp->t_tfo_pending) { 2437 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2438 tp->t_tfo_pending = NULL; 2439 2440 /* 2441 * Account for the ACK of our SYN prior to 2442 * regular ACK processing below. 2443 */ 2444 tp->snd_una++; 2445 } 2446 /* 2447 * TFO connections call cc_conn_init() during SYN 2448 * processing. Calling it again here for such 2449 * connections is not harmless as it would undo the 2450 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2451 * is retransmitted. 2452 */ 2453 if (!(tp->t_flags & TF_FASTOPEN)) 2454#endif 2455 cc_conn_init(tp); 2456 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2457 } 2458 /* 2459 * If segment contains data or ACK, will call tcp_reass() 2460 * later; if not, do so now to pass queued data to user. 2461 */ 2462 if (tlen == 0 && (thflags & TH_FIN) == 0) 2463 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2464 (struct mbuf *)0); 2465 tp->snd_wl1 = th->th_seq - 1; 2466 /* FALLTHROUGH */ 2467 2468 /* 2469 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2470 * ACKs. If the ack is in the range 2471 * tp->snd_una < th->th_ack <= tp->snd_max 2472 * then advance tp->snd_una to th->th_ack and drop 2473 * data from the retransmission queue. If this ACK reflects 2474 * more up to date window information we update our window information. 2475 */ 2476 case TCPS_ESTABLISHED: 2477 case TCPS_FIN_WAIT_1: 2478 case TCPS_FIN_WAIT_2: 2479 case TCPS_CLOSE_WAIT: 2480 case TCPS_CLOSING: 2481 case TCPS_LAST_ACK: 2482 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2483 TCPSTAT_INC(tcps_rcvacktoomuch); 2484 goto dropafterack; 2485 } 2486 if ((tp->t_flags & TF_SACK_PERMIT) && 2487 ((to.to_flags & TOF_SACK) || 2488 !TAILQ_EMPTY(&tp->snd_holes))) 2489 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2490 else 2491 /* 2492 * Reset the value so that previous (valid) value 2493 * from the last ack with SACK doesn't get used. 2494 */ 2495 tp->sackhint.sacked_bytes = 0; 2496 2497 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2498 hhook_run_tcp_est_in(tp, th, &to); 2499 2500 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2501 if (tlen == 0 && 2502 (tiwin == tp->snd_wnd || 2503 (tp->t_flags & TF_SACK_PERMIT))) { 2504 TCPSTAT_INC(tcps_rcvdupack); 2505 /* 2506 * If we have outstanding data (other than 2507 * a window probe), this is a completely 2508 * duplicate ack (ie, window info didn't 2509 * change), the ack is the biggest we've 2510 * seen and we've seen exactly our rexmt 2511 * threshhold of them, assume a packet 2512 * has been dropped and retransmit it. 2513 * Kludge snd_nxt & the congestion 2514 * window so we send only this one 2515 * packet. 2516 * 2517 * We know we're losing at the current 2518 * window size so do congestion avoidance 2519 * (set ssthresh to half the current window 2520 * and pull our congestion window back to 2521 * the new ssthresh). 2522 * 2523 * Dup acks mean that packets have left the 2524 * network (they're now cached at the receiver) 2525 * so bump cwnd by the amount in the receiver 2526 * to keep a constant cwnd packets in the 2527 * network. 2528 * 2529 * When using TCP ECN, notify the peer that 2530 * we reduced the cwnd. 2531 */ 2532 /* 2533 * Following 2 kinds of acks should not affect 2534 * dupack counting: 2535 * 1) Old acks 2536 * 2) Acks with SACK but without any new SACK 2537 * information in them. These could result from 2538 * any anomaly in the network like a switch 2539 * duplicating packets or a possible DoS attack. 2540 */ 2541 if (th->th_ack != tp->snd_una || 2542 ((tp->t_flags & TF_SACK_PERMIT) && 2543 !sack_changed)) 2544 break; 2545 else if (!tcp_timer_active(tp, TT_REXMT)) 2546 tp->t_dupacks = 0; 2547 else if (++tp->t_dupacks > tcprexmtthresh || 2548 IN_FASTRECOVERY(tp->t_flags)) { 2549 cc_ack_received(tp, th, CC_DUPACK); 2550 if ((tp->t_flags & TF_SACK_PERMIT) && 2551 IN_FASTRECOVERY(tp->t_flags)) { 2552 int awnd; 2553 2554 /* 2555 * Compute the amount of data in flight first. 2556 * We can inject new data into the pipe iff 2557 * we have less than 1/2 the original window's 2558 * worth of data in flight. 2559 */ 2560 if (V_tcp_do_rfc6675_pipe) 2561 awnd = tcp_compute_pipe(tp); 2562 else 2563 awnd = (tp->snd_nxt - tp->snd_fack) + 2564 tp->sackhint.sack_bytes_rexmit; 2565 2566 if (awnd < tp->snd_ssthresh) { 2567 tp->snd_cwnd += tp->t_maxseg; 2568 if (tp->snd_cwnd > tp->snd_ssthresh) 2569 tp->snd_cwnd = tp->snd_ssthresh; 2570 } 2571 } else 2572 tp->snd_cwnd += tp->t_maxseg; 2573 if ((thflags & TH_FIN) && 2574 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2575 /* 2576 * If its a fin we need to process 2577 * it to avoid a race where both 2578 * sides enter FIN-WAIT and send FIN|ACK 2579 * at the same time. 2580 */ 2581 break; 2582 } 2583 (void) tcp_output(tp); 2584 goto drop; 2585 } else if (tp->t_dupacks == tcprexmtthresh) { 2586 tcp_seq onxt = tp->snd_nxt; 2587 2588 /* 2589 * If we're doing sack, check to 2590 * see if we're already in sack 2591 * recovery. If we're not doing sack, 2592 * check to see if we're in newreno 2593 * recovery. 2594 */ 2595 if (tp->t_flags & TF_SACK_PERMIT) { 2596 if (IN_FASTRECOVERY(tp->t_flags)) { 2597 tp->t_dupacks = 0; 2598 break; 2599 } 2600 } else { 2601 if (SEQ_LEQ(th->th_ack, 2602 tp->snd_recover)) { 2603 tp->t_dupacks = 0; 2604 break; 2605 } 2606 } 2607 /* Congestion signal before ack. */ 2608 cc_cong_signal(tp, th, CC_NDUPACK); 2609 cc_ack_received(tp, th, CC_DUPACK); 2610 tcp_timer_activate(tp, TT_REXMT, 0); 2611 tp->t_rtttime = 0; 2612 if (tp->t_flags & TF_SACK_PERMIT) { 2613 TCPSTAT_INC( 2614 tcps_sack_recovery_episode); 2615 tp->sack_newdata = tp->snd_nxt; 2616 tp->snd_cwnd = tp->t_maxseg; 2617 (void) tcp_output(tp); 2618 goto drop; 2619 } 2620 tp->snd_nxt = th->th_ack; 2621 tp->snd_cwnd = tp->t_maxseg; 2622 if ((thflags & TH_FIN) && 2623 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2624 /* 2625 * If its a fin we need to process 2626 * it to avoid a race where both 2627 * sides enter FIN-WAIT and send FIN|ACK 2628 * at the same time. 2629 */ 2630 break; 2631 } 2632 (void) tcp_output(tp); 2633 KASSERT(tp->snd_limited <= 2, 2634 ("%s: tp->snd_limited too big", 2635 __func__)); 2636 tp->snd_cwnd = tp->snd_ssthresh + 2637 tp->t_maxseg * 2638 (tp->t_dupacks - tp->snd_limited); 2639 if (SEQ_GT(onxt, tp->snd_nxt)) 2640 tp->snd_nxt = onxt; 2641 goto drop; 2642 } else if (V_tcp_do_rfc3042) { 2643 /* 2644 * Process first and second duplicate 2645 * ACKs. Each indicates a segment 2646 * leaving the network, creating room 2647 * for more. Make sure we can send a 2648 * packet on reception of each duplicate 2649 * ACK by increasing snd_cwnd by one 2650 * segment. Restore the original 2651 * snd_cwnd after packet transmission. 2652 */ 2653 cc_ack_received(tp, th, CC_DUPACK); 2654 u_long oldcwnd = tp->snd_cwnd; 2655 tcp_seq oldsndmax = tp->snd_max; 2656 u_int sent; 2657 int avail; 2658 2659 KASSERT(tp->t_dupacks == 1 || 2660 tp->t_dupacks == 2, 2661 ("%s: dupacks not 1 or 2", 2662 __func__)); 2663 if (tp->t_dupacks == 1) 2664 tp->snd_limited = 0; 2665 tp->snd_cwnd = 2666 (tp->snd_nxt - tp->snd_una) + 2667 (tp->t_dupacks - tp->snd_limited) * 2668 tp->t_maxseg; 2669 if ((thflags & TH_FIN) && 2670 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2671 /* 2672 * If its a fin we need to process 2673 * it to avoid a race where both 2674 * sides enter FIN-WAIT and send FIN|ACK 2675 * at the same time. 2676 */ 2677 break; 2678 } 2679 /* 2680 * Only call tcp_output when there 2681 * is new data available to be sent. 2682 * Otherwise we would send pure ACKs. 2683 */ 2684 SOCKBUF_LOCK(&so->so_snd); 2685 avail = so->so_snd.sb_cc - 2686 (tp->snd_nxt - tp->snd_una); 2687 SOCKBUF_UNLOCK(&so->so_snd); 2688 if (avail > 0) 2689 (void) tcp_output(tp); 2690 sent = tp->snd_max - oldsndmax; 2691 if (sent > tp->t_maxseg) { 2692 KASSERT((tp->t_dupacks == 2 && 2693 tp->snd_limited == 0) || 2694 (sent == tp->t_maxseg + 1 && 2695 tp->t_flags & TF_SENTFIN), 2696 ("%s: sent too much", 2697 __func__)); 2698 tp->snd_limited = 2; 2699 } else if (sent > 0) 2700 ++tp->snd_limited; 2701 tp->snd_cwnd = oldcwnd; 2702 goto drop; 2703 } 2704 } 2705 break; 2706 } else { 2707 /* 2708 * This ack is advancing the left edge, reset the 2709 * counter. 2710 */ 2711 tp->t_dupacks = 0; 2712 /* 2713 * If this ack also has new SACK info, increment the 2714 * counter as per rfc6675. 2715 */ 2716 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2717 tp->t_dupacks++; 2718 } 2719 2720 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2721 ("%s: th_ack <= snd_una", __func__)); 2722 2723 /* 2724 * If the congestion window was inflated to account 2725 * for the other side's cached packets, retract it. 2726 */ 2727 if (IN_FASTRECOVERY(tp->t_flags)) { 2728 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2729 if (tp->t_flags & TF_SACK_PERMIT) 2730 tcp_sack_partialack(tp, th); 2731 else 2732 tcp_newreno_partial_ack(tp, th); 2733 } else 2734 cc_post_recovery(tp, th); 2735 } 2736 /* 2737 * If we reach this point, ACK is not a duplicate, 2738 * i.e., it ACKs something we sent. 2739 */ 2740 if (tp->t_flags & TF_NEEDSYN) { 2741 /* 2742 * T/TCP: Connection was half-synchronized, and our 2743 * SYN has been ACK'd (so connection is now fully 2744 * synchronized). Go to non-starred state, 2745 * increment snd_una for ACK of SYN, and check if 2746 * we can do window scaling. 2747 */ 2748 tp->t_flags &= ~TF_NEEDSYN; 2749 tp->snd_una++; 2750 /* Do window scaling? */ 2751 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2752 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2753 tp->rcv_scale = tp->request_r_scale; 2754 /* Send window already scaled. */ 2755 } 2756 } 2757 2758process_ACK: 2759 INP_WLOCK_ASSERT(tp->t_inpcb); 2760 2761 acked = BYTES_THIS_ACK(tp, th); 2762 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2763 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2764 tp->snd_una, th->th_ack, tp, m)); 2765 TCPSTAT_INC(tcps_rcvackpack); 2766 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2767 2768 /* 2769 * If we just performed our first retransmit, and the ACK 2770 * arrives within our recovery window, then it was a mistake 2771 * to do the retransmit in the first place. Recover our 2772 * original cwnd and ssthresh, and proceed to transmit where 2773 * we left off. 2774 */ 2775 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2776 (int)(ticks - tp->t_badrxtwin) < 0) 2777 cc_cong_signal(tp, th, CC_RTO_ERR); 2778 2779 /* 2780 * If we have a timestamp reply, update smoothed 2781 * round trip time. If no timestamp is present but 2782 * transmit timer is running and timed sequence 2783 * number was acked, update smoothed round trip time. 2784 * Since we now have an rtt measurement, cancel the 2785 * timer backoff (cf., Phil Karn's retransmit alg.). 2786 * Recompute the initial retransmit timer. 2787 * 2788 * Some boxes send broken timestamp replies 2789 * during the SYN+ACK phase, ignore 2790 * timestamps of 0 or we could calculate a 2791 * huge RTT and blow up the retransmit timer. 2792 */ 2793 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2794 u_int t; 2795 2796 t = tcp_ts_getticks() - to.to_tsecr; 2797 if (!tp->t_rttlow || tp->t_rttlow > t) 2798 tp->t_rttlow = t; 2799 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2800 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2801 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2802 tp->t_rttlow = ticks - tp->t_rtttime; 2803 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2804 } 2805 2806 /* 2807 * If all outstanding data is acked, stop retransmit 2808 * timer and remember to restart (more output or persist). 2809 * If there is more data to be acked, restart retransmit 2810 * timer, using current (possibly backed-off) value. 2811 */ 2812 if (th->th_ack == tp->snd_max) { 2813 tcp_timer_activate(tp, TT_REXMT, 0); 2814 needoutput = 1; 2815 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2816 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2817 2818 /* 2819 * If no data (only SYN) was ACK'd, 2820 * skip rest of ACK processing. 2821 */ 2822 if (acked == 0) 2823 goto step6; 2824 2825 /* 2826 * Let the congestion control algorithm update congestion 2827 * control related information. This typically means increasing 2828 * the congestion window. 2829 */ 2830 cc_ack_received(tp, th, CC_ACK); 2831 2832 SOCKBUF_LOCK(&so->so_snd); 2833 if (acked > so->so_snd.sb_cc) { 2834 if (tp->snd_wnd >= so->so_snd.sb_cc) 2835 tp->snd_wnd -= so->so_snd.sb_cc; 2836 else 2837 tp->snd_wnd = 0; 2838 mfree = sbcut_locked(&so->so_snd, 2839 (int)so->so_snd.sb_cc); 2840 ourfinisacked = 1; 2841 } else { 2842 mfree = sbcut_locked(&so->so_snd, acked); 2843 if (tp->snd_wnd >= (u_long) acked) 2844 tp->snd_wnd -= acked; 2845 else 2846 tp->snd_wnd = 0; 2847 ourfinisacked = 0; 2848 } 2849 /* NB: sowwakeup_locked() does an implicit unlock. */ 2850 sowwakeup_locked(so); 2851 m_freem(mfree); 2852 /* Detect una wraparound. */ 2853 if (!IN_RECOVERY(tp->t_flags) && 2854 SEQ_GT(tp->snd_una, tp->snd_recover) && 2855 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2856 tp->snd_recover = th->th_ack - 1; 2857 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2858 if (IN_RECOVERY(tp->t_flags) && 2859 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2860 EXIT_RECOVERY(tp->t_flags); 2861 } 2862 tp->snd_una = th->th_ack; 2863 if (tp->t_flags & TF_SACK_PERMIT) { 2864 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2865 tp->snd_recover = tp->snd_una; 2866 } 2867 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2868 tp->snd_nxt = tp->snd_una; 2869 2870 switch (tp->t_state) { 2871 2872 /* 2873 * In FIN_WAIT_1 STATE in addition to the processing 2874 * for the ESTABLISHED state if our FIN is now acknowledged 2875 * then enter FIN_WAIT_2. 2876 */ 2877 case TCPS_FIN_WAIT_1: 2878 if (ourfinisacked) { 2879 /* 2880 * If we can't receive any more 2881 * data, then closing user can proceed. 2882 * Starting the timer is contrary to the 2883 * specification, but if we don't get a FIN 2884 * we'll hang forever. 2885 * 2886 * XXXjl: 2887 * we should release the tp also, and use a 2888 * compressed state. 2889 */ 2890 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2891 soisdisconnected(so); 2892 tcp_timer_activate(tp, TT_2MSL, 2893 (tcp_fast_finwait2_recycle ? 2894 tcp_finwait2_timeout : 2895 TP_MAXIDLE(tp))); 2896 } 2897 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2898 } 2899 break; 2900 2901 /* 2902 * In CLOSING STATE in addition to the processing for 2903 * the ESTABLISHED state if the ACK acknowledges our FIN 2904 * then enter the TIME-WAIT state, otherwise ignore 2905 * the segment. 2906 */ 2907 case TCPS_CLOSING: 2908 if (ourfinisacked) { 2909 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2910 tcp_twstart(tp); 2911 INP_INFO_RUNLOCK(&V_tcbinfo); 2912 m_freem(m); 2913 return; 2914 } 2915 break; 2916 2917 /* 2918 * In LAST_ACK, we may still be waiting for data to drain 2919 * and/or to be acked, as well as for the ack of our FIN. 2920 * If our FIN is now acknowledged, delete the TCB, 2921 * enter the closed state and return. 2922 */ 2923 case TCPS_LAST_ACK: 2924 if (ourfinisacked) { 2925 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2926 tp = tcp_close(tp); 2927 goto drop; 2928 } 2929 break; 2930 } 2931 } 2932 2933step6: 2934 INP_WLOCK_ASSERT(tp->t_inpcb); 2935 2936 /* 2937 * Update window information. 2938 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2939 */ 2940 if ((thflags & TH_ACK) && 2941 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2942 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2943 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2944 /* keep track of pure window updates */ 2945 if (tlen == 0 && 2946 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2947 TCPSTAT_INC(tcps_rcvwinupd); 2948 tp->snd_wnd = tiwin; 2949 tp->snd_wl1 = th->th_seq; 2950 tp->snd_wl2 = th->th_ack; 2951 if (tp->snd_wnd > tp->max_sndwnd) 2952 tp->max_sndwnd = tp->snd_wnd; 2953 needoutput = 1; 2954 } 2955 2956 /* 2957 * Process segments with URG. 2958 */ 2959 if ((thflags & TH_URG) && th->th_urp && 2960 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2961 /* 2962 * This is a kludge, but if we receive and accept 2963 * random urgent pointers, we'll crash in 2964 * soreceive. It's hard to imagine someone 2965 * actually wanting to send this much urgent data. 2966 */ 2967 SOCKBUF_LOCK(&so->so_rcv); 2968 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2969 th->th_urp = 0; /* XXX */ 2970 thflags &= ~TH_URG; /* XXX */ 2971 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2972 goto dodata; /* XXX */ 2973 } 2974 /* 2975 * If this segment advances the known urgent pointer, 2976 * then mark the data stream. This should not happen 2977 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2978 * a FIN has been received from the remote side. 2979 * In these states we ignore the URG. 2980 * 2981 * According to RFC961 (Assigned Protocols), 2982 * the urgent pointer points to the last octet 2983 * of urgent data. We continue, however, 2984 * to consider it to indicate the first octet 2985 * of data past the urgent section as the original 2986 * spec states (in one of two places). 2987 */ 2988 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2989 tp->rcv_up = th->th_seq + th->th_urp; 2990 so->so_oobmark = so->so_rcv.sb_cc + 2991 (tp->rcv_up - tp->rcv_nxt) - 1; 2992 if (so->so_oobmark == 0) 2993 so->so_rcv.sb_state |= SBS_RCVATMARK; 2994 sohasoutofband(so); 2995 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2996 } 2997 SOCKBUF_UNLOCK(&so->so_rcv); 2998 /* 2999 * Remove out of band data so doesn't get presented to user. 3000 * This can happen independent of advancing the URG pointer, 3001 * but if two URG's are pending at once, some out-of-band 3002 * data may creep in... ick. 3003 */ 3004 if (th->th_urp <= (u_long)tlen && 3005 !(so->so_options & SO_OOBINLINE)) { 3006 /* hdr drop is delayed */ 3007 tcp_pulloutofband(so, th, m, drop_hdrlen); 3008 } 3009 } else { 3010 /* 3011 * If no out of band data is expected, 3012 * pull receive urgent pointer along 3013 * with the receive window. 3014 */ 3015 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3016 tp->rcv_up = tp->rcv_nxt; 3017 } 3018dodata: /* XXX */ 3019 INP_WLOCK_ASSERT(tp->t_inpcb); 3020 3021 /* 3022 * Process the segment text, merging it into the TCP sequencing queue, 3023 * and arranging for acknowledgment of receipt if necessary. 3024 * This process logically involves adjusting tp->rcv_wnd as data 3025 * is presented to the user (this happens in tcp_usrreq.c, 3026 * case PRU_RCVD). If a FIN has already been received on this 3027 * connection then we just ignore the text. 3028 */ 3029 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3030 (tp->t_flags & TF_FASTOPEN)); 3031 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3032 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3033 tcp_seq save_start = th->th_seq; 3034 3035 m_adj(m, drop_hdrlen); /* delayed header drop */ 3036 /* 3037 * Insert segment which includes th into TCP reassembly queue 3038 * with control block tp. Set thflags to whether reassembly now 3039 * includes a segment with FIN. This handles the common case 3040 * inline (segment is the next to be received on an established 3041 * connection, and the queue is empty), avoiding linkage into 3042 * and removal from the queue and repetition of various 3043 * conversions. 3044 * Set DELACK for segments received in order, but ack 3045 * immediately when segments are out of order (so 3046 * fast retransmit can work). 3047 */ 3048 if (th->th_seq == tp->rcv_nxt && 3049 LIST_EMPTY(&tp->t_segq) && 3050 (TCPS_HAVEESTABLISHED(tp->t_state) || 3051 tfo_syn)) { 3052 if (DELAY_ACK(tp, tlen) || tfo_syn) 3053 tp->t_flags |= TF_DELACK; 3054 else 3055 tp->t_flags |= TF_ACKNOW; 3056 tp->rcv_nxt += tlen; 3057 thflags = th->th_flags & TH_FIN; 3058 TCPSTAT_INC(tcps_rcvpack); 3059 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3060 ND6_HINT(tp); 3061 SOCKBUF_LOCK(&so->so_rcv); 3062 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3063 m_freem(m); 3064 else 3065 sbappendstream_locked(&so->so_rcv, m); 3066 /* NB: sorwakeup_locked() does an implicit unlock. */ 3067 sorwakeup_locked(so); 3068 } else { 3069 /* 3070 * XXX: Due to the header drop above "th" is 3071 * theoretically invalid by now. Fortunately 3072 * m_adj() doesn't actually frees any mbufs 3073 * when trimming from the head. 3074 */ 3075 thflags = tcp_reass(tp, th, &tlen, m); 3076 tp->t_flags |= TF_ACKNOW; 3077 } 3078 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3079 tcp_update_sack_list(tp, save_start, save_start + tlen); 3080#if 0 3081 /* 3082 * Note the amount of data that peer has sent into 3083 * our window, in order to estimate the sender's 3084 * buffer size. 3085 * XXX: Unused. 3086 */ 3087 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3088 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3089 else 3090 len = so->so_rcv.sb_hiwat; 3091#endif 3092 } else { 3093 m_freem(m); 3094 thflags &= ~TH_FIN; 3095 } 3096 3097 /* 3098 * If FIN is received ACK the FIN and let the user know 3099 * that the connection is closing. 3100 */ 3101 if (thflags & TH_FIN) { 3102 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3103 socantrcvmore(so); 3104 /* 3105 * If connection is half-synchronized 3106 * (ie NEEDSYN flag on) then delay ACK, 3107 * so it may be piggybacked when SYN is sent. 3108 * Otherwise, since we received a FIN then no 3109 * more input can be expected, send ACK now. 3110 */ 3111 if (tp->t_flags & TF_NEEDSYN) 3112 tp->t_flags |= TF_DELACK; 3113 else 3114 tp->t_flags |= TF_ACKNOW; 3115 tp->rcv_nxt++; 3116 } 3117 switch (tp->t_state) { 3118 3119 /* 3120 * In SYN_RECEIVED and ESTABLISHED STATES 3121 * enter the CLOSE_WAIT state. 3122 */ 3123 case TCPS_SYN_RECEIVED: 3124 tp->t_starttime = ticks; 3125 /* FALLTHROUGH */ 3126 case TCPS_ESTABLISHED: 3127 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3128 break; 3129 3130 /* 3131 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3132 * enter the CLOSING state. 3133 */ 3134 case TCPS_FIN_WAIT_1: 3135 tcp_state_change(tp, TCPS_CLOSING); 3136 break; 3137 3138 /* 3139 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3140 * starting the time-wait timer, turning off the other 3141 * standard timers. 3142 */ 3143 case TCPS_FIN_WAIT_2: 3144 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3145 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3146 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3147 ti_locked)); 3148 3149 tcp_twstart(tp); 3150 INP_INFO_RUNLOCK(&V_tcbinfo); 3151 return; 3152 } 3153 } 3154 if (ti_locked == TI_RLOCKED) 3155 INP_INFO_RUNLOCK(&V_tcbinfo); 3156 ti_locked = TI_UNLOCKED; 3157 3158#ifdef TCPDEBUG 3159 if (so->so_options & SO_DEBUG) 3160 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3161 &tcp_savetcp, 0); 3162#endif 3163 3164 /* 3165 * Return any desired output. 3166 */ 3167 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3168 (void) tcp_output(tp); 3169 3170check_delack: 3171 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3172 __func__, ti_locked)); 3173 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3174 INP_WLOCK_ASSERT(tp->t_inpcb); 3175 3176 if (tp->t_flags & TF_DELACK) { 3177 tp->t_flags &= ~TF_DELACK; 3178 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3179 } 3180 INP_WUNLOCK(tp->t_inpcb); 3181 return; 3182 3183dropafterack: 3184 /* 3185 * Generate an ACK dropping incoming segment if it occupies 3186 * sequence space, where the ACK reflects our state. 3187 * 3188 * We can now skip the test for the RST flag since all 3189 * paths to this code happen after packets containing 3190 * RST have been dropped. 3191 * 3192 * In the SYN-RECEIVED state, don't send an ACK unless the 3193 * segment we received passes the SYN-RECEIVED ACK test. 3194 * If it fails send a RST. This breaks the loop in the 3195 * "LAND" DoS attack, and also prevents an ACK storm 3196 * between two listening ports that have been sent forged 3197 * SYN segments, each with the source address of the other. 3198 */ 3199 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3200 (SEQ_GT(tp->snd_una, th->th_ack) || 3201 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3202 rstreason = BANDLIM_RST_OPENPORT; 3203 goto dropwithreset; 3204 } 3205#ifdef TCPDEBUG 3206 if (so->so_options & SO_DEBUG) 3207 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3208 &tcp_savetcp, 0); 3209#endif 3210 if (ti_locked == TI_RLOCKED) 3211 INP_INFO_RUNLOCK(&V_tcbinfo); 3212 ti_locked = TI_UNLOCKED; 3213 3214 tp->t_flags |= TF_ACKNOW; 3215 (void) tcp_output(tp); 3216 INP_WUNLOCK(tp->t_inpcb); 3217 m_freem(m); 3218 return; 3219 3220dropwithreset: 3221 if (ti_locked == TI_RLOCKED) 3222 INP_INFO_RUNLOCK(&V_tcbinfo); 3223 ti_locked = TI_UNLOCKED; 3224 3225 if (tp != NULL) { 3226 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3227 INP_WUNLOCK(tp->t_inpcb); 3228 } else 3229 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3230 return; 3231 3232drop: 3233 if (ti_locked == TI_RLOCKED) { 3234 INP_INFO_RUNLOCK(&V_tcbinfo); 3235 ti_locked = TI_UNLOCKED; 3236 } 3237#ifdef INVARIANTS 3238 else 3239 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3240#endif 3241 3242 /* 3243 * Drop space held by incoming segment and return. 3244 */ 3245#ifdef TCPDEBUG 3246 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3247 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3248 &tcp_savetcp, 0); 3249#endif 3250 if (tp != NULL) 3251 INP_WUNLOCK(tp->t_inpcb); 3252 m_freem(m); 3253} 3254 3255/* 3256 * Issue RST and make ACK acceptable to originator of segment. 3257 * The mbuf must still include the original packet header. 3258 * tp may be NULL. 3259 */ 3260static void 3261tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3262 int tlen, int rstreason) 3263{ 3264#ifdef INET 3265 struct ip *ip; 3266#endif 3267#ifdef INET6 3268 struct ip6_hdr *ip6; 3269#endif 3270 3271 if (tp != NULL) { 3272 INP_WLOCK_ASSERT(tp->t_inpcb); 3273 } 3274 3275 /* Don't bother if destination was broadcast/multicast. */ 3276 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3277 goto drop; 3278#ifdef INET6 3279 if (mtod(m, struct ip *)->ip_v == 6) { 3280 ip6 = mtod(m, struct ip6_hdr *); 3281 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3282 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3283 goto drop; 3284 /* IPv6 anycast check is done at tcp6_input() */ 3285 } 3286#endif 3287#if defined(INET) && defined(INET6) 3288 else 3289#endif 3290#ifdef INET 3291 { 3292 ip = mtod(m, struct ip *); 3293 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3294 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3295 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3296 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3297 goto drop; 3298 } 3299#endif 3300 3301 /* Perform bandwidth limiting. */ 3302 if (badport_bandlim(rstreason) < 0) 3303 goto drop; 3304 3305 /* tcp_respond consumes the mbuf chain. */ 3306 if (th->th_flags & TH_ACK) { 3307 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3308 th->th_ack, TH_RST); 3309 } else { 3310 if (th->th_flags & TH_SYN) 3311 tlen++; 3312 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3313 (tcp_seq)0, TH_RST|TH_ACK); 3314 } 3315 return; 3316drop: 3317 m_freem(m); 3318} 3319 3320/* 3321 * Parse TCP options and place in tcpopt. 3322 */ 3323static void 3324tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3325{ 3326 int opt, optlen; 3327 3328 to->to_flags = 0; 3329 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3330 opt = cp[0]; 3331 if (opt == TCPOPT_EOL) 3332 break; 3333 if (opt == TCPOPT_NOP) 3334 optlen = 1; 3335 else { 3336 if (cnt < 2) 3337 break; 3338 optlen = cp[1]; 3339 if (optlen < 2 || optlen > cnt) 3340 break; 3341 } 3342 switch (opt) { 3343 case TCPOPT_MAXSEG: 3344 if (optlen != TCPOLEN_MAXSEG) 3345 continue; 3346 if (!(flags & TO_SYN)) 3347 continue; 3348 to->to_flags |= TOF_MSS; 3349 bcopy((char *)cp + 2, 3350 (char *)&to->to_mss, sizeof(to->to_mss)); 3351 to->to_mss = ntohs(to->to_mss); 3352 break; 3353 case TCPOPT_WINDOW: 3354 if (optlen != TCPOLEN_WINDOW) 3355 continue; 3356 if (!(flags & TO_SYN)) 3357 continue; 3358 to->to_flags |= TOF_SCALE; 3359 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3360 break; 3361 case TCPOPT_TIMESTAMP: 3362 if (optlen != TCPOLEN_TIMESTAMP) 3363 continue; 3364 to->to_flags |= TOF_TS; 3365 bcopy((char *)cp + 2, 3366 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3367 to->to_tsval = ntohl(to->to_tsval); 3368 bcopy((char *)cp + 6, 3369 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3370 to->to_tsecr = ntohl(to->to_tsecr); 3371 break; 3372#ifdef TCP_SIGNATURE 3373 /* 3374 * XXX In order to reply to a host which has set the 3375 * TCP_SIGNATURE option in its initial SYN, we have to 3376 * record the fact that the option was observed here 3377 * for the syncache code to perform the correct response. 3378 */ 3379 case TCPOPT_SIGNATURE: 3380 if (optlen != TCPOLEN_SIGNATURE) 3381 continue; 3382 to->to_flags |= TOF_SIGNATURE; 3383 to->to_signature = cp + 2; 3384 break; 3385#endif 3386 case TCPOPT_SACK_PERMITTED: 3387 if (optlen != TCPOLEN_SACK_PERMITTED) 3388 continue; 3389 if (!(flags & TO_SYN)) 3390 continue; 3391 if (!V_tcp_do_sack) 3392 continue; 3393 to->to_flags |= TOF_SACKPERM; 3394 break; 3395 case TCPOPT_SACK: 3396 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3397 continue; 3398 if (flags & TO_SYN) 3399 continue; 3400 to->to_flags |= TOF_SACK; 3401 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3402 to->to_sacks = cp + 2; 3403 TCPSTAT_INC(tcps_sack_rcv_blocks); 3404 break; 3405#ifdef TCP_RFC7413 3406 case TCPOPT_FAST_OPEN: 3407 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3408 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3409 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3410 continue; 3411 if (!(flags & TO_SYN)) 3412 continue; 3413 if (!V_tcp_fastopen_enabled) 3414 continue; 3415 to->to_flags |= TOF_FASTOPEN; 3416 to->to_tfo_len = optlen - 2; 3417 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3418 break; 3419#endif 3420 default: 3421 continue; 3422 } 3423 } 3424} 3425 3426/* 3427 * Pull out of band byte out of a segment so 3428 * it doesn't appear in the user's data queue. 3429 * It is still reflected in the segment length for 3430 * sequencing purposes. 3431 */ 3432static void 3433tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3434 int off) 3435{ 3436 int cnt = off + th->th_urp - 1; 3437 3438 while (cnt >= 0) { 3439 if (m->m_len > cnt) { 3440 char *cp = mtod(m, caddr_t) + cnt; 3441 struct tcpcb *tp = sototcpcb(so); 3442 3443 INP_WLOCK_ASSERT(tp->t_inpcb); 3444 3445 tp->t_iobc = *cp; 3446 tp->t_oobflags |= TCPOOB_HAVEDATA; 3447 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3448 m->m_len--; 3449 if (m->m_flags & M_PKTHDR) 3450 m->m_pkthdr.len--; 3451 return; 3452 } 3453 cnt -= m->m_len; 3454 m = m->m_next; 3455 if (m == NULL) 3456 break; 3457 } 3458 panic("tcp_pulloutofband"); 3459} 3460 3461/* 3462 * Collect new round-trip time estimate 3463 * and update averages and current timeout. 3464 */ 3465static void 3466tcp_xmit_timer(struct tcpcb *tp, int rtt) 3467{ 3468 int delta; 3469 3470 INP_WLOCK_ASSERT(tp->t_inpcb); 3471 3472 TCPSTAT_INC(tcps_rttupdated); 3473 tp->t_rttupdated++; 3474 if (tp->t_srtt != 0) { 3475 /* 3476 * srtt is stored as fixed point with 5 bits after the 3477 * binary point (i.e., scaled by 8). The following magic 3478 * is equivalent to the smoothing algorithm in rfc793 with 3479 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3480 * point). Adjust rtt to origin 0. 3481 */ 3482 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3483 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3484 3485 if ((tp->t_srtt += delta) <= 0) 3486 tp->t_srtt = 1; 3487 3488 /* 3489 * We accumulate a smoothed rtt variance (actually, a 3490 * smoothed mean difference), then set the retransmit 3491 * timer to smoothed rtt + 4 times the smoothed variance. 3492 * rttvar is stored as fixed point with 4 bits after the 3493 * binary point (scaled by 16). The following is 3494 * equivalent to rfc793 smoothing with an alpha of .75 3495 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3496 * rfc793's wired-in beta. 3497 */ 3498 if (delta < 0) 3499 delta = -delta; 3500 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3501 if ((tp->t_rttvar += delta) <= 0) 3502 tp->t_rttvar = 1; 3503 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3504 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3505 } else { 3506 /* 3507 * No rtt measurement yet - use the unsmoothed rtt. 3508 * Set the variance to half the rtt (so our first 3509 * retransmit happens at 3*rtt). 3510 */ 3511 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3512 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3513 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3514 } 3515 tp->t_rtttime = 0; 3516 tp->t_rxtshift = 0; 3517 3518 /* 3519 * the retransmit should happen at rtt + 4 * rttvar. 3520 * Because of the way we do the smoothing, srtt and rttvar 3521 * will each average +1/2 tick of bias. When we compute 3522 * the retransmit timer, we want 1/2 tick of rounding and 3523 * 1 extra tick because of +-1/2 tick uncertainty in the 3524 * firing of the timer. The bias will give us exactly the 3525 * 1.5 tick we need. But, because the bias is 3526 * statistical, we have to test that we don't drop below 3527 * the minimum feasible timer (which is 2 ticks). 3528 */ 3529 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3530 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3531 3532 /* 3533 * We received an ack for a packet that wasn't retransmitted; 3534 * it is probably safe to discard any error indications we've 3535 * received recently. This isn't quite right, but close enough 3536 * for now (a route might have failed after we sent a segment, 3537 * and the return path might not be symmetrical). 3538 */ 3539 tp->t_softerror = 0; 3540} 3541 3542/* 3543 * Determine a reasonable value for maxseg size. 3544 * If the route is known, check route for mtu. 3545 * If none, use an mss that can be handled on the outgoing interface 3546 * without forcing IP to fragment. If no route is found, route has no mtu, 3547 * or the destination isn't local, use a default, hopefully conservative 3548 * size (usually 512 or the default IP max size, but no more than the mtu 3549 * of the interface), as we can't discover anything about intervening 3550 * gateways or networks. We also initialize the congestion/slow start 3551 * window to be a single segment if the destination isn't local. 3552 * While looking at the routing entry, we also initialize other path-dependent 3553 * parameters from pre-set or cached values in the routing entry. 3554 * 3555 * Also take into account the space needed for options that we 3556 * send regularly. Make maxseg shorter by that amount to assure 3557 * that we can send maxseg amount of data even when the options 3558 * are present. Store the upper limit of the length of options plus 3559 * data in maxopd. 3560 * 3561 * NOTE that this routine is only called when we process an incoming 3562 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3563 * settings are handled in tcp_mssopt(). 3564 */ 3565void 3566tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3567 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3568{ 3569 int mss = 0; 3570 u_long maxmtu = 0; 3571 struct inpcb *inp = tp->t_inpcb; 3572 struct hc_metrics_lite metrics; 3573 int origoffer; 3574#ifdef INET6 3575 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3576 size_t min_protoh = isipv6 ? 3577 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3578 sizeof (struct tcpiphdr); 3579#else 3580 const size_t min_protoh = sizeof(struct tcpiphdr); 3581#endif 3582 3583 INP_WLOCK_ASSERT(tp->t_inpcb); 3584 3585 if (mtuoffer != -1) { 3586 KASSERT(offer == -1, ("%s: conflict", __func__)); 3587 offer = mtuoffer - min_protoh; 3588 } 3589 origoffer = offer; 3590 3591 /* Initialize. */ 3592#ifdef INET6 3593 if (isipv6) { 3594 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3595 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3596 } 3597#endif 3598#if defined(INET) && defined(INET6) 3599 else 3600#endif 3601#ifdef INET 3602 { 3603 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3604 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3605 } 3606#endif 3607 3608 /* 3609 * No route to sender, stay with default mss and return. 3610 */ 3611 if (maxmtu == 0) { 3612 /* 3613 * In case we return early we need to initialize metrics 3614 * to a defined state as tcp_hc_get() would do for us 3615 * if there was no cache hit. 3616 */ 3617 if (metricptr != NULL) 3618 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3619 return; 3620 } 3621 3622 /* What have we got? */ 3623 switch (offer) { 3624 case 0: 3625 /* 3626 * Offer == 0 means that there was no MSS on the SYN 3627 * segment, in this case we use tcp_mssdflt as 3628 * already assigned to t_maxopd above. 3629 */ 3630 offer = tp->t_maxopd; 3631 break; 3632 3633 case -1: 3634 /* 3635 * Offer == -1 means that we didn't receive SYN yet. 3636 */ 3637 /* FALLTHROUGH */ 3638 3639 default: 3640 /* 3641 * Prevent DoS attack with too small MSS. Round up 3642 * to at least minmss. 3643 */ 3644 offer = max(offer, V_tcp_minmss); 3645 } 3646 3647 /* 3648 * rmx information is now retrieved from tcp_hostcache. 3649 */ 3650 tcp_hc_get(&inp->inp_inc, &metrics); 3651 if (metricptr != NULL) 3652 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3653 3654 /* 3655 * If there's a discovered mtu int tcp hostcache, use it 3656 * else, use the link mtu. 3657 */ 3658 if (metrics.rmx_mtu) 3659 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3660 else { 3661#ifdef INET6 3662 if (isipv6) { 3663 mss = maxmtu - min_protoh; 3664 if (!V_path_mtu_discovery && 3665 !in6_localaddr(&inp->in6p_faddr)) 3666 mss = min(mss, V_tcp_v6mssdflt); 3667 } 3668#endif 3669#if defined(INET) && defined(INET6) 3670 else 3671#endif 3672#ifdef INET 3673 { 3674 mss = maxmtu - min_protoh; 3675 if (!V_path_mtu_discovery && 3676 !in_localaddr(inp->inp_faddr)) 3677 mss = min(mss, V_tcp_mssdflt); 3678 } 3679#endif 3680 /* 3681 * XXX - The above conditional (mss = maxmtu - min_protoh) 3682 * probably violates the TCP spec. 3683 * The problem is that, since we don't know the 3684 * other end's MSS, we are supposed to use a conservative 3685 * default. But, if we do that, then MTU discovery will 3686 * never actually take place, because the conservative 3687 * default is much less than the MTUs typically seen 3688 * on the Internet today. For the moment, we'll sweep 3689 * this under the carpet. 3690 * 3691 * The conservative default might not actually be a problem 3692 * if the only case this occurs is when sending an initial 3693 * SYN with options and data to a host we've never talked 3694 * to before. Then, they will reply with an MSS value which 3695 * will get recorded and the new parameters should get 3696 * recomputed. For Further Study. 3697 */ 3698 } 3699 mss = min(mss, offer); 3700 3701 /* 3702 * Sanity check: make sure that maxopd will be large 3703 * enough to allow some data on segments even if the 3704 * all the option space is used (40bytes). Otherwise 3705 * funny things may happen in tcp_output. 3706 */ 3707 mss = max(mss, 64); 3708 3709 /* 3710 * maxopd stores the maximum length of data AND options 3711 * in a segment; maxseg is the amount of data in a normal 3712 * segment. We need to store this value (maxopd) apart 3713 * from maxseg, because now every segment carries options 3714 * and thus we normally have somewhat less data in segments. 3715 */ 3716 tp->t_maxopd = mss; 3717 3718 /* 3719 * origoffer==-1 indicates that no segments were received yet. 3720 * In this case we just guess. 3721 */ 3722 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3723 (origoffer == -1 || 3724 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3725 mss -= TCPOLEN_TSTAMP_APPA; 3726 3727 tp->t_maxseg = mss; 3728} 3729 3730void 3731tcp_mss(struct tcpcb *tp, int offer) 3732{ 3733 int mss; 3734 u_long bufsize; 3735 struct inpcb *inp; 3736 struct socket *so; 3737 struct hc_metrics_lite metrics; 3738 struct tcp_ifcap cap; 3739 3740 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3741 3742 bzero(&cap, sizeof(cap)); 3743 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3744 3745 mss = tp->t_maxseg; 3746 inp = tp->t_inpcb; 3747 3748 /* 3749 * If there's a pipesize, change the socket buffer to that size, 3750 * don't change if sb_hiwat is different than default (then it 3751 * has been changed on purpose with setsockopt). 3752 * Make the socket buffers an integral number of mss units; 3753 * if the mss is larger than the socket buffer, decrease the mss. 3754 */ 3755 so = inp->inp_socket; 3756 SOCKBUF_LOCK(&so->so_snd); 3757 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3758 bufsize = metrics.rmx_sendpipe; 3759 else 3760 bufsize = so->so_snd.sb_hiwat; 3761 if (bufsize < mss) 3762 mss = bufsize; 3763 else { 3764 bufsize = roundup(bufsize, mss); 3765 if (bufsize > sb_max) 3766 bufsize = sb_max; 3767 if (bufsize > so->so_snd.sb_hiwat) 3768 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3769 } 3770 SOCKBUF_UNLOCK(&so->so_snd); 3771 tp->t_maxseg = mss; 3772 3773 SOCKBUF_LOCK(&so->so_rcv); 3774 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3775 bufsize = metrics.rmx_recvpipe; 3776 else 3777 bufsize = so->so_rcv.sb_hiwat; 3778 if (bufsize > mss) { 3779 bufsize = roundup(bufsize, mss); 3780 if (bufsize > sb_max) 3781 bufsize = sb_max; 3782 if (bufsize > so->so_rcv.sb_hiwat) 3783 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3784 } 3785 SOCKBUF_UNLOCK(&so->so_rcv); 3786 3787 /* Check the interface for TSO capabilities. */ 3788 if (cap.ifcap & CSUM_TSO) { 3789 tp->t_flags |= TF_TSO; 3790 tp->t_tsomax = cap.tsomax; 3791 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3792 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3793 } 3794} 3795 3796/* 3797 * Determine the MSS option to send on an outgoing SYN. 3798 */ 3799int 3800tcp_mssopt(struct in_conninfo *inc) 3801{ 3802 int mss = 0; 3803 u_long maxmtu = 0; 3804 u_long thcmtu = 0; 3805 size_t min_protoh; 3806 3807 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3808 3809#ifdef INET6 3810 if (inc->inc_flags & INC_ISIPV6) { 3811 mss = V_tcp_v6mssdflt; 3812 maxmtu = tcp_maxmtu6(inc, NULL); 3813 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3814 } 3815#endif 3816#if defined(INET) && defined(INET6) 3817 else 3818#endif 3819#ifdef INET 3820 { 3821 mss = V_tcp_mssdflt; 3822 maxmtu = tcp_maxmtu(inc, NULL); 3823 min_protoh = sizeof(struct tcpiphdr); 3824 } 3825#endif 3826#if defined(INET6) || defined(INET) 3827 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3828#endif 3829 3830 if (maxmtu && thcmtu) 3831 mss = min(maxmtu, thcmtu) - min_protoh; 3832 else if (maxmtu || thcmtu) 3833 mss = max(maxmtu, thcmtu) - min_protoh; 3834 3835 return (mss); 3836} 3837 3838 3839/* 3840 * On a partial ack arrives, force the retransmission of the 3841 * next unacknowledged segment. Do not clear tp->t_dupacks. 3842 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3843 * be started again. 3844 */ 3845static void 3846tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3847{ 3848 tcp_seq onxt = tp->snd_nxt; 3849 u_long ocwnd = tp->snd_cwnd; 3850 3851 INP_WLOCK_ASSERT(tp->t_inpcb); 3852 3853 tcp_timer_activate(tp, TT_REXMT, 0); 3854 tp->t_rtttime = 0; 3855 tp->snd_nxt = th->th_ack; 3856 /* 3857 * Set snd_cwnd to one segment beyond acknowledged offset. 3858 * (tp->snd_una has not yet been updated when this function is called.) 3859 */ 3860 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3861 tp->t_flags |= TF_ACKNOW; 3862 (void) tcp_output(tp); 3863 tp->snd_cwnd = ocwnd; 3864 if (SEQ_GT(onxt, tp->snd_nxt)) 3865 tp->snd_nxt = onxt; 3866 /* 3867 * Partial window deflation. Relies on fact that tp->snd_una 3868 * not updated yet. 3869 */ 3870 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3871 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3872 else 3873 tp->snd_cwnd = 0; 3874 tp->snd_cwnd += tp->t_maxseg; 3875} 3876 3877int 3878tcp_compute_pipe(struct tcpcb *tp) 3879{ 3880 return (tp->snd_max - tp->snd_una + 3881 tp->sackhint.sack_bytes_rexmit - 3882 tp->sackhint.sacked_bytes); 3883} 3884