tcp_input.c revision 256281
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 256186 2013-10-09 12:00:38Z glebius $"); 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#include <netinet/tcp_fsm.h> 102#include <netinet/tcp_seq.h> 103#include <netinet/tcp_timer.h> 104#include <netinet/tcp_var.h> 105#include <netinet6/tcp6_var.h> 106#include <netinet/tcpip.h> 107#include <netinet/tcp_syncache.h> 108#ifdef TCPDEBUG 109#include <netinet/tcp_debug.h> 110#endif /* TCPDEBUG */ 111#ifdef TCP_OFFLOAD 112#include <netinet/tcp_offload.h> 113#endif 114 115#ifdef IPSEC 116#include <netipsec/ipsec.h> 117#include <netipsec/ipsec6.h> 118#endif /*IPSEC*/ 119 120#include <machine/in_cksum.h> 121 122#include <security/mac/mac_framework.h> 123 124const int tcprexmtthresh = 3; 125 126int tcp_log_in_vain = 0; 127SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 128 &tcp_log_in_vain, 0, 129 "Log all incoming TCP segments to closed ports"); 130 131VNET_DEFINE(int, blackhole) = 0; 132#define V_blackhole VNET(blackhole) 133SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 134 &VNET_NAME(blackhole), 0, 135 "Do not send RST on segments to closed ports"); 136 137VNET_DEFINE(int, tcp_delack_enabled) = 1; 138SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 139 &VNET_NAME(tcp_delack_enabled), 0, 140 "Delay ACK to try and piggyback it onto a data packet"); 141 142VNET_DEFINE(int, drop_synfin) = 0; 143#define V_drop_synfin VNET(drop_synfin) 144SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 145 &VNET_NAME(drop_synfin), 0, 146 "Drop TCP packets with SYN+FIN set"); 147 148VNET_DEFINE(int, tcp_do_rfc3042) = 1; 149#define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 150SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 151 &VNET_NAME(tcp_do_rfc3042), 0, 152 "Enable RFC 3042 (Limited Transmit)"); 153 154VNET_DEFINE(int, tcp_do_rfc3390) = 1; 155SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 156 &VNET_NAME(tcp_do_rfc3390), 0, 157 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 158 159SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 160 "Experimental TCP extensions"); 161 162VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 163SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 164 &VNET_NAME(tcp_do_initcwnd10), 0, 165 "Enable draft-ietf-tcpm-initcwnd-05 (Increasing initial CWND to 10)"); 166 167VNET_DEFINE(int, tcp_do_rfc3465) = 1; 168SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 169 &VNET_NAME(tcp_do_rfc3465), 0, 170 "Enable RFC 3465 (Appropriate Byte Counting)"); 171 172VNET_DEFINE(int, tcp_abc_l_var) = 2; 173SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 174 &VNET_NAME(tcp_abc_l_var), 2, 175 "Cap the max cwnd increment during slow-start to this number of segments"); 176 177static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 178 179VNET_DEFINE(int, tcp_do_ecn) = 0; 180SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 181 &VNET_NAME(tcp_do_ecn), 0, 182 "TCP ECN support"); 183 184VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 185SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 186 &VNET_NAME(tcp_ecn_maxretries), 0, 187 "Max retries before giving up on ECN"); 188 189VNET_DEFINE(int, tcp_insecure_rst) = 0; 190#define V_tcp_insecure_rst VNET(tcp_insecure_rst) 191SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 192 &VNET_NAME(tcp_insecure_rst), 0, 193 "Follow the old (insecure) criteria for accepting RST packets"); 194 195VNET_DEFINE(int, tcp_recvspace) = 1024*64; 196#define V_tcp_recvspace VNET(tcp_recvspace) 197SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 198 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 199 200VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 201#define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 202SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 203 &VNET_NAME(tcp_do_autorcvbuf), 0, 204 "Enable automatic receive buffer sizing"); 205 206VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 207#define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 208SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 209 &VNET_NAME(tcp_autorcvbuf_inc), 0, 210 "Incrementor step size of automatic receive buffer"); 211 212VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 213#define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 214SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 215 &VNET_NAME(tcp_autorcvbuf_max), 0, 216 "Max size of automatic receive buffer"); 217 218VNET_DEFINE(struct inpcbhead, tcb); 219#define tcb6 tcb /* for KAME src sync over BSD*'s */ 220VNET_DEFINE(struct inpcbinfo, tcbinfo); 221 222static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 223static void tcp_do_segment(struct mbuf *, struct tcphdr *, 224 struct socket *, struct tcpcb *, int, int, uint8_t, 225 int); 226static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 227 struct tcpcb *, int, int); 228static void tcp_pulloutofband(struct socket *, 229 struct tcphdr *, struct mbuf *, int); 230static void tcp_xmit_timer(struct tcpcb *, int); 231static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 232static void inline tcp_fields_to_host(struct tcphdr *); 233#ifdef TCP_SIGNATURE 234static void inline tcp_fields_to_net(struct tcphdr *); 235static int inline tcp_signature_verify_input(struct mbuf *, int, int, 236 int, struct tcpopt *, struct tcphdr *, u_int); 237#endif 238static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 239 uint16_t type); 240static void inline cc_conn_init(struct tcpcb *tp); 241static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 242static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 243 struct tcphdr *th, struct tcpopt *to); 244 245/* 246 * TCP statistics are stored in an "array" of counter(9)s. 247 */ 248VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249VNET_PCPUSTAT_SYSINIT(tcpstat); 250SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252 253#ifdef VIMAGE 254VNET_PCPUSTAT_SYSUNINIT(tcpstat); 255#endif /* VIMAGE */ 256/* 257 * Kernel module interface for updating tcpstat. The argument is an index 258 * into tcpstat treated as an array. 259 */ 260void 261kmod_tcpstat_inc(int statnum) 262{ 263 264 counter_u64_add(VNET(tcpstat)[statnum], 1); 265} 266 267/* 268 * Wrapper for the TCP established input helper hook. 269 */ 270static void inline 271hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 272{ 273 struct tcp_hhook_data hhook_data; 274 275 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 276 hhook_data.tp = tp; 277 hhook_data.th = th; 278 hhook_data.to = to; 279 280 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 281 tp->osd); 282 } 283} 284 285/* 286 * CC wrapper hook functions 287 */ 288static void inline 289cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 290{ 291 INP_WLOCK_ASSERT(tp->t_inpcb); 292 293 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 294 if (tp->snd_cwnd <= tp->snd_wnd) 295 tp->ccv->flags |= CCF_CWND_LIMITED; 296 else 297 tp->ccv->flags &= ~CCF_CWND_LIMITED; 298 299 if (type == CC_ACK) { 300 if (tp->snd_cwnd > tp->snd_ssthresh) { 301 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 302 V_tcp_abc_l_var * tp->t_maxseg); 303 if (tp->t_bytes_acked >= tp->snd_cwnd) { 304 tp->t_bytes_acked -= tp->snd_cwnd; 305 tp->ccv->flags |= CCF_ABC_SENTAWND; 306 } 307 } else { 308 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 309 tp->t_bytes_acked = 0; 310 } 311 } 312 313 if (CC_ALGO(tp)->ack_received != NULL) { 314 /* XXXLAS: Find a way to live without this */ 315 tp->ccv->curack = th->th_ack; 316 CC_ALGO(tp)->ack_received(tp->ccv, type); 317 } 318} 319 320static void inline 321cc_conn_init(struct tcpcb *tp) 322{ 323 struct hc_metrics_lite metrics; 324 struct inpcb *inp = tp->t_inpcb; 325 int rtt; 326 327 INP_WLOCK_ASSERT(tp->t_inpcb); 328 329 tcp_hc_get(&inp->inp_inc, &metrics); 330 331 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 332 tp->t_srtt = rtt; 333 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 334 TCPSTAT_INC(tcps_usedrtt); 335 if (metrics.rmx_rttvar) { 336 tp->t_rttvar = metrics.rmx_rttvar; 337 TCPSTAT_INC(tcps_usedrttvar); 338 } else { 339 /* default variation is +- 1 rtt */ 340 tp->t_rttvar = 341 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 342 } 343 TCPT_RANGESET(tp->t_rxtcur, 344 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 345 tp->t_rttmin, TCPTV_REXMTMAX); 346 } 347 if (metrics.rmx_ssthresh) { 348 /* 349 * There's some sort of gateway or interface 350 * buffer limit on the path. Use this to set 351 * the slow start threshhold, but set the 352 * threshold to no less than 2*mss. 353 */ 354 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 355 TCPSTAT_INC(tcps_usedssthresh); 356 } 357 358 /* 359 * Set the initial slow-start flight size. 360 * 361 * RFC5681 Section 3.1 specifies the default conservative values. 362 * RFC3390 specifies slightly more aggressive values. 363 * Draft-ietf-tcpm-initcwnd-05 increases it to ten segments. 364 * 365 * If a SYN or SYN/ACK was lost and retransmitted, we have to 366 * reduce the initial CWND to one segment as congestion is likely 367 * requiring us to be cautious. 368 */ 369 if (tp->snd_cwnd == 1) 370 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 371 else if (V_tcp_do_initcwnd10) 372 tp->snd_cwnd = min(10 * tp->t_maxseg, 373 max(2 * tp->t_maxseg, 14600)); 374 else if (V_tcp_do_rfc3390) 375 tp->snd_cwnd = min(4 * tp->t_maxseg, 376 max(2 * tp->t_maxseg, 4380)); 377 else { 378 /* Per RFC5681 Section 3.1 */ 379 if (tp->t_maxseg > 2190) 380 tp->snd_cwnd = 2 * tp->t_maxseg; 381 else if (tp->t_maxseg > 1095) 382 tp->snd_cwnd = 3 * tp->t_maxseg; 383 else 384 tp->snd_cwnd = 4 * tp->t_maxseg; 385 } 386 387 if (CC_ALGO(tp)->conn_init != NULL) 388 CC_ALGO(tp)->conn_init(tp->ccv); 389} 390 391void inline 392cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 393{ 394 INP_WLOCK_ASSERT(tp->t_inpcb); 395 396 switch(type) { 397 case CC_NDUPACK: 398 if (!IN_FASTRECOVERY(tp->t_flags)) { 399 tp->snd_recover = tp->snd_max; 400 if (tp->t_flags & TF_ECN_PERMIT) 401 tp->t_flags |= TF_ECN_SND_CWR; 402 } 403 break; 404 case CC_ECN: 405 if (!IN_CONGRECOVERY(tp->t_flags)) { 406 TCPSTAT_INC(tcps_ecn_rcwnd); 407 tp->snd_recover = tp->snd_max; 408 if (tp->t_flags & TF_ECN_PERMIT) 409 tp->t_flags |= TF_ECN_SND_CWR; 410 } 411 break; 412 case CC_RTO: 413 tp->t_dupacks = 0; 414 tp->t_bytes_acked = 0; 415 EXIT_RECOVERY(tp->t_flags); 416 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 417 tp->t_maxseg) * tp->t_maxseg; 418 tp->snd_cwnd = tp->t_maxseg; 419 break; 420 case CC_RTO_ERR: 421 TCPSTAT_INC(tcps_sndrexmitbad); 422 /* RTO was unnecessary, so reset everything. */ 423 tp->snd_cwnd = tp->snd_cwnd_prev; 424 tp->snd_ssthresh = tp->snd_ssthresh_prev; 425 tp->snd_recover = tp->snd_recover_prev; 426 if (tp->t_flags & TF_WASFRECOVERY) 427 ENTER_FASTRECOVERY(tp->t_flags); 428 if (tp->t_flags & TF_WASCRECOVERY) 429 ENTER_CONGRECOVERY(tp->t_flags); 430 tp->snd_nxt = tp->snd_max; 431 tp->t_flags &= ~TF_PREVVALID; 432 tp->t_badrxtwin = 0; 433 break; 434 } 435 436 if (CC_ALGO(tp)->cong_signal != NULL) { 437 if (th != NULL) 438 tp->ccv->curack = th->th_ack; 439 CC_ALGO(tp)->cong_signal(tp->ccv, type); 440 } 441} 442 443static void inline 444cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 445{ 446 INP_WLOCK_ASSERT(tp->t_inpcb); 447 448 /* XXXLAS: KASSERT that we're in recovery? */ 449 450 if (CC_ALGO(tp)->post_recovery != NULL) { 451 tp->ccv->curack = th->th_ack; 452 CC_ALGO(tp)->post_recovery(tp->ccv); 453 } 454 /* XXXLAS: EXIT_RECOVERY ? */ 455 tp->t_bytes_acked = 0; 456} 457 458static inline void 459tcp_fields_to_host(struct tcphdr *th) 460{ 461 462 th->th_seq = ntohl(th->th_seq); 463 th->th_ack = ntohl(th->th_ack); 464 th->th_win = ntohs(th->th_win); 465 th->th_urp = ntohs(th->th_urp); 466} 467 468#ifdef TCP_SIGNATURE 469static inline void 470tcp_fields_to_net(struct tcphdr *th) 471{ 472 473 th->th_seq = htonl(th->th_seq); 474 th->th_ack = htonl(th->th_ack); 475 th->th_win = htons(th->th_win); 476 th->th_urp = htons(th->th_urp); 477} 478 479static inline int 480tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 481 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 482{ 483 int ret; 484 485 tcp_fields_to_net(th); 486 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 487 tcp_fields_to_host(th); 488 return (ret); 489} 490#endif 491 492/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 493#ifdef INET6 494#define ND6_HINT(tp) \ 495do { \ 496 if ((tp) && (tp)->t_inpcb && \ 497 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 498 nd6_nud_hint(NULL, NULL, 0); \ 499} while (0) 500#else 501#define ND6_HINT(tp) 502#endif 503 504/* 505 * Indicate whether this ack should be delayed. We can delay the ack if 506 * - there is no delayed ack timer in progress and 507 * - our last ack wasn't a 0-sized window. We never want to delay 508 * the ack that opens up a 0-sized window and 509 * - delayed acks are enabled or 510 * - this is a half-synchronized T/TCP connection. 511 */ 512#define DELAY_ACK(tp) \ 513 ((!tcp_timer_active(tp, TT_DELACK) && \ 514 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 515 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 516 517/* 518 * TCP input handling is split into multiple parts: 519 * tcp6_input is a thin wrapper around tcp_input for the extended 520 * ip6_protox[] call format in ip6_input 521 * tcp_input handles primary segment validation, inpcb lookup and 522 * SYN processing on listen sockets 523 * tcp_do_segment processes the ACK and text of the segment for 524 * establishing, established and closing connections 525 */ 526#ifdef INET6 527int 528tcp6_input(struct mbuf **mp, int *offp, int proto) 529{ 530 struct mbuf *m = *mp; 531 struct in6_ifaddr *ia6; 532 533 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 534 535 /* 536 * draft-itojun-ipv6-tcp-to-anycast 537 * better place to put this in? 538 */ 539 ia6 = ip6_getdstifaddr(m); 540 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 541 struct ip6_hdr *ip6; 542 543 ifa_free(&ia6->ia_ifa); 544 ip6 = mtod(m, struct ip6_hdr *); 545 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 546 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 547 return IPPROTO_DONE; 548 } 549 if (ia6) 550 ifa_free(&ia6->ia_ifa); 551 552 tcp_input(m, *offp); 553 return IPPROTO_DONE; 554} 555#endif /* INET6 */ 556 557void 558tcp_input(struct mbuf *m, int off0) 559{ 560 struct tcphdr *th = NULL; 561 struct ip *ip = NULL; 562 struct inpcb *inp = NULL; 563 struct tcpcb *tp = NULL; 564 struct socket *so = NULL; 565 u_char *optp = NULL; 566 int optlen = 0; 567#ifdef INET 568 int len; 569#endif 570 int tlen = 0, off; 571 int drop_hdrlen; 572 int thflags; 573 int rstreason = 0; /* For badport_bandlim accounting purposes */ 574#ifdef TCP_SIGNATURE 575 uint8_t sig_checked = 0; 576#endif 577 uint8_t iptos = 0; 578 struct m_tag *fwd_tag = NULL; 579#ifdef INET6 580 struct ip6_hdr *ip6 = NULL; 581 int isipv6; 582#else 583 const void *ip6 = NULL; 584#endif /* INET6 */ 585 struct tcpopt to; /* options in this segment */ 586 char *s = NULL; /* address and port logging */ 587 int ti_locked; 588#define TI_UNLOCKED 1 589#define TI_WLOCKED 2 590 591#ifdef TCPDEBUG 592 /* 593 * The size of tcp_saveipgen must be the size of the max ip header, 594 * now IPv6. 595 */ 596 u_char tcp_saveipgen[IP6_HDR_LEN]; 597 struct tcphdr tcp_savetcp; 598 short ostate = 0; 599#endif 600 601#ifdef INET6 602 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 603#endif 604 605 to.to_flags = 0; 606 TCPSTAT_INC(tcps_rcvtotal); 607 608#ifdef INET6 609 if (isipv6) { 610 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 611 612 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 613 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 614 if (m == NULL) { 615 TCPSTAT_INC(tcps_rcvshort); 616 return; 617 } 618 } 619 620 ip6 = mtod(m, struct ip6_hdr *); 621 th = (struct tcphdr *)((caddr_t)ip6 + off0); 622 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 623 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 624 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 625 th->th_sum = m->m_pkthdr.csum_data; 626 else 627 th->th_sum = in6_cksum_pseudo(ip6, tlen, 628 IPPROTO_TCP, m->m_pkthdr.csum_data); 629 th->th_sum ^= 0xffff; 630 } else 631 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 632 if (th->th_sum) { 633 TCPSTAT_INC(tcps_rcvbadsum); 634 goto drop; 635 } 636 637 /* 638 * Be proactive about unspecified IPv6 address in source. 639 * As we use all-zero to indicate unbounded/unconnected pcb, 640 * unspecified IPv6 address can be used to confuse us. 641 * 642 * Note that packets with unspecified IPv6 destination is 643 * already dropped in ip6_input. 644 */ 645 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 646 /* XXX stat */ 647 goto drop; 648 } 649 } 650#endif 651#if defined(INET) && defined(INET6) 652 else 653#endif 654#ifdef INET 655 { 656 /* 657 * Get IP and TCP header together in first mbuf. 658 * Note: IP leaves IP header in first mbuf. 659 */ 660 if (off0 > sizeof (struct ip)) { 661 ip_stripoptions(m); 662 off0 = sizeof(struct ip); 663 } 664 if (m->m_len < sizeof (struct tcpiphdr)) { 665 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 666 == NULL) { 667 TCPSTAT_INC(tcps_rcvshort); 668 return; 669 } 670 } 671 ip = mtod(m, struct ip *); 672 th = (struct tcphdr *)((caddr_t)ip + off0); 673 tlen = ntohs(ip->ip_len) - off0; 674 675 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 676 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 677 th->th_sum = m->m_pkthdr.csum_data; 678 else 679 th->th_sum = in_pseudo(ip->ip_src.s_addr, 680 ip->ip_dst.s_addr, 681 htonl(m->m_pkthdr.csum_data + tlen + 682 IPPROTO_TCP)); 683 th->th_sum ^= 0xffff; 684 } else { 685 struct ipovly *ipov = (struct ipovly *)ip; 686 687 /* 688 * Checksum extended TCP header and data. 689 */ 690 len = off0 + tlen; 691 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 692 ipov->ih_len = htons(tlen); 693 th->th_sum = in_cksum(m, len); 694 /* Reset length for SDT probes. */ 695 ip->ip_len = htons(tlen + off0); 696 } 697 698 if (th->th_sum) { 699 TCPSTAT_INC(tcps_rcvbadsum); 700 goto drop; 701 } 702 /* Re-initialization for later version check */ 703 ip->ip_v = IPVERSION; 704 } 705#endif /* INET */ 706 707#ifdef INET6 708 if (isipv6) 709 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 710#endif 711#if defined(INET) && defined(INET6) 712 else 713#endif 714#ifdef INET 715 iptos = ip->ip_tos; 716#endif 717 718 /* 719 * Check that TCP offset makes sense, 720 * pull out TCP options and adjust length. XXX 721 */ 722 off = th->th_off << 2; 723 if (off < sizeof (struct tcphdr) || off > tlen) { 724 TCPSTAT_INC(tcps_rcvbadoff); 725 goto drop; 726 } 727 tlen -= off; /* tlen is used instead of ti->ti_len */ 728 if (off > sizeof (struct tcphdr)) { 729#ifdef INET6 730 if (isipv6) { 731 IP6_EXTHDR_CHECK(m, off0, off, ); 732 ip6 = mtod(m, struct ip6_hdr *); 733 th = (struct tcphdr *)((caddr_t)ip6 + off0); 734 } 735#endif 736#if defined(INET) && defined(INET6) 737 else 738#endif 739#ifdef INET 740 { 741 if (m->m_len < sizeof(struct ip) + off) { 742 if ((m = m_pullup(m, sizeof (struct ip) + off)) 743 == NULL) { 744 TCPSTAT_INC(tcps_rcvshort); 745 return; 746 } 747 ip = mtod(m, struct ip *); 748 th = (struct tcphdr *)((caddr_t)ip + off0); 749 } 750 } 751#endif 752 optlen = off - sizeof (struct tcphdr); 753 optp = (u_char *)(th + 1); 754 } 755 thflags = th->th_flags; 756 757 /* 758 * Convert TCP protocol specific fields to host format. 759 */ 760 tcp_fields_to_host(th); 761 762 /* 763 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 764 */ 765 drop_hdrlen = off0 + off; 766 767 /* 768 * Locate pcb for segment; if we're likely to add or remove a 769 * connection then first acquire pcbinfo lock. There are two cases 770 * where we might discover later we need a write lock despite the 771 * flags: ACKs moving a connection out of the syncache, and ACKs for 772 * a connection in TIMEWAIT. 773 */ 774 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 775 INP_INFO_WLOCK(&V_tcbinfo); 776 ti_locked = TI_WLOCKED; 777 } else 778 ti_locked = TI_UNLOCKED; 779 780 /* 781 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 782 */ 783 if ( 784#ifdef INET6 785 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 786#ifdef INET 787 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 788#endif 789#endif 790#if defined(INET) && !defined(INET6) 791 (m->m_flags & M_IP_NEXTHOP) 792#endif 793 ) 794 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 795 796findpcb: 797#ifdef INVARIANTS 798 if (ti_locked == TI_WLOCKED) { 799 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 800 } else { 801 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 802 } 803#endif 804#ifdef INET6 805 if (isipv6 && fwd_tag != NULL) { 806 struct sockaddr_in6 *next_hop6; 807 808 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 809 /* 810 * Transparently forwarded. Pretend to be the destination. 811 * Already got one like this? 812 */ 813 inp = in6_pcblookup_mbuf(&V_tcbinfo, 814 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 815 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 816 if (!inp) { 817 /* 818 * It's new. Try to find the ambushing socket. 819 * Because we've rewritten the destination address, 820 * any hardware-generated hash is ignored. 821 */ 822 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 823 th->th_sport, &next_hop6->sin6_addr, 824 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 825 th->th_dport, INPLOOKUP_WILDCARD | 826 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 827 } 828 } else if (isipv6) { 829 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 830 th->th_sport, &ip6->ip6_dst, th->th_dport, 831 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 832 m->m_pkthdr.rcvif, m); 833 } 834#endif /* INET6 */ 835#if defined(INET6) && defined(INET) 836 else 837#endif 838#ifdef INET 839 if (fwd_tag != NULL) { 840 struct sockaddr_in *next_hop; 841 842 next_hop = (struct sockaddr_in *)(fwd_tag+1); 843 /* 844 * Transparently forwarded. Pretend to be the destination. 845 * already got one like this? 846 */ 847 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 848 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 849 m->m_pkthdr.rcvif, m); 850 if (!inp) { 851 /* 852 * It's new. Try to find the ambushing socket. 853 * Because we've rewritten the destination address, 854 * any hardware-generated hash is ignored. 855 */ 856 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 857 th->th_sport, next_hop->sin_addr, 858 next_hop->sin_port ? ntohs(next_hop->sin_port) : 859 th->th_dport, INPLOOKUP_WILDCARD | 860 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 861 } 862 } else 863 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 864 th->th_sport, ip->ip_dst, th->th_dport, 865 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 866 m->m_pkthdr.rcvif, m); 867#endif /* INET */ 868 869 /* 870 * If the INPCB does not exist then all data in the incoming 871 * segment is discarded and an appropriate RST is sent back. 872 * XXX MRT Send RST using which routing table? 873 */ 874 if (inp == NULL) { 875 /* 876 * Log communication attempts to ports that are not 877 * in use. 878 */ 879 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 880 tcp_log_in_vain == 2) { 881 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 882 log(LOG_INFO, "%s; %s: Connection attempt " 883 "to closed port\n", s, __func__); 884 } 885 /* 886 * When blackholing do not respond with a RST but 887 * completely ignore the segment and drop it. 888 */ 889 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 890 V_blackhole == 2) 891 goto dropunlock; 892 893 rstreason = BANDLIM_RST_CLOSEDPORT; 894 goto dropwithreset; 895 } 896 INP_WLOCK_ASSERT(inp); 897 if (!(inp->inp_flags & INP_HW_FLOWID) 898 && (m->m_flags & M_FLOWID) 899 && ((inp->inp_socket == NULL) 900 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 901 inp->inp_flags |= INP_HW_FLOWID; 902 inp->inp_flags &= ~INP_SW_FLOWID; 903 inp->inp_flowid = m->m_pkthdr.flowid; 904 } 905#ifdef IPSEC 906#ifdef INET6 907 if (isipv6 && ipsec6_in_reject(m, inp)) { 908 IPSEC6STAT_INC(ips_in_polvio); 909 goto dropunlock; 910 } else 911#endif /* INET6 */ 912 if (ipsec4_in_reject(m, inp) != 0) { 913 IPSECSTAT_INC(ips_in_polvio); 914 goto dropunlock; 915 } 916#endif /* IPSEC */ 917 918 /* 919 * Check the minimum TTL for socket. 920 */ 921 if (inp->inp_ip_minttl != 0) { 922#ifdef INET6 923 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 924 goto dropunlock; 925 else 926#endif 927 if (inp->inp_ip_minttl > ip->ip_ttl) 928 goto dropunlock; 929 } 930 931 /* 932 * A previous connection in TIMEWAIT state is supposed to catch stray 933 * or duplicate segments arriving late. If this segment was a 934 * legitimate new connection attempt, the old INPCB gets removed and 935 * we can try again to find a listening socket. 936 * 937 * At this point, due to earlier optimism, we may hold only an inpcb 938 * lock, and not the inpcbinfo write lock. If so, we need to try to 939 * acquire it, or if that fails, acquire a reference on the inpcb, 940 * drop all locks, acquire a global write lock, and then re-acquire 941 * the inpcb lock. We may at that point discover that another thread 942 * has tried to free the inpcb, in which case we need to loop back 943 * and try to find a new inpcb to deliver to. 944 * 945 * XXXRW: It may be time to rethink timewait locking. 946 */ 947relocked: 948 if (inp->inp_flags & INP_TIMEWAIT) { 949 if (ti_locked == TI_UNLOCKED) { 950 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 951 in_pcbref(inp); 952 INP_WUNLOCK(inp); 953 INP_INFO_WLOCK(&V_tcbinfo); 954 ti_locked = TI_WLOCKED; 955 INP_WLOCK(inp); 956 if (in_pcbrele_wlocked(inp)) { 957 inp = NULL; 958 goto findpcb; 959 } 960 } else 961 ti_locked = TI_WLOCKED; 962 } 963 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 964 965 if (thflags & TH_SYN) 966 tcp_dooptions(&to, optp, optlen, TO_SYN); 967 /* 968 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 969 */ 970 if (tcp_twcheck(inp, &to, th, m, tlen)) 971 goto findpcb; 972 INP_INFO_WUNLOCK(&V_tcbinfo); 973 return; 974 } 975 /* 976 * The TCPCB may no longer exist if the connection is winding 977 * down or it is in the CLOSED state. Either way we drop the 978 * segment and send an appropriate response. 979 */ 980 tp = intotcpcb(inp); 981 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 982 rstreason = BANDLIM_RST_CLOSEDPORT; 983 goto dropwithreset; 984 } 985 986#ifdef TCP_OFFLOAD 987 if (tp->t_flags & TF_TOE) { 988 tcp_offload_input(tp, m); 989 m = NULL; /* consumed by the TOE driver */ 990 goto dropunlock; 991 } 992#endif 993 994 /* 995 * We've identified a valid inpcb, but it could be that we need an 996 * inpcbinfo write lock but don't hold it. In this case, attempt to 997 * acquire using the same strategy as the TIMEWAIT case above. If we 998 * relock, we have to jump back to 'relocked' as the connection might 999 * now be in TIMEWAIT. 1000 */ 1001#ifdef INVARIANTS 1002 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 1003 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1004#endif 1005 if (tp->t_state != TCPS_ESTABLISHED) { 1006 if (ti_locked == TI_UNLOCKED) { 1007 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 1008 in_pcbref(inp); 1009 INP_WUNLOCK(inp); 1010 INP_INFO_WLOCK(&V_tcbinfo); 1011 ti_locked = TI_WLOCKED; 1012 INP_WLOCK(inp); 1013 if (in_pcbrele_wlocked(inp)) { 1014 inp = NULL; 1015 goto findpcb; 1016 } 1017 goto relocked; 1018 } else 1019 ti_locked = TI_WLOCKED; 1020 } 1021 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1022 } 1023 1024#ifdef MAC 1025 INP_WLOCK_ASSERT(inp); 1026 if (mac_inpcb_check_deliver(inp, m)) 1027 goto dropunlock; 1028#endif 1029 so = inp->inp_socket; 1030 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1031#ifdef TCPDEBUG 1032 if (so->so_options & SO_DEBUG) { 1033 ostate = tp->t_state; 1034#ifdef INET6 1035 if (isipv6) { 1036 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1037 } else 1038#endif 1039 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1040 tcp_savetcp = *th; 1041 } 1042#endif /* TCPDEBUG */ 1043 /* 1044 * When the socket is accepting connections (the INPCB is in LISTEN 1045 * state) we look into the SYN cache if this is a new connection 1046 * attempt or the completion of a previous one. Because listen 1047 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1048 * held in this case. 1049 */ 1050 if (so->so_options & SO_ACCEPTCONN) { 1051 struct in_conninfo inc; 1052 1053 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1054 "tp not listening", __func__)); 1055 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1056 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 * Parse the TCP options here because 1081 * syncookies need access to the reflected 1082 * timestamp. 1083 */ 1084 tcp_dooptions(&to, optp, optlen, 0); 1085 /* 1086 * NB: syncache_expand() doesn't unlock 1087 * inp and tcpinfo locks. 1088 */ 1089 if (!syncache_expand(&inc, &to, th, &so, m)) { 1090 /* 1091 * No syncache entry or ACK was not 1092 * for our SYN/ACK. Send a RST. 1093 * NB: syncache did its own logging 1094 * of the failure cause. 1095 */ 1096 rstreason = BANDLIM_RST_OPENPORT; 1097 goto dropwithreset; 1098 } 1099 if (so == NULL) { 1100 /* 1101 * We completed the 3-way handshake 1102 * but could not allocate a socket 1103 * either due to memory shortage, 1104 * listen queue length limits or 1105 * global socket limits. Send RST 1106 * or wait and have the remote end 1107 * retransmit the ACK for another 1108 * try. 1109 */ 1110 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1111 log(LOG_DEBUG, "%s; %s: Listen socket: " 1112 "Socket allocation failed due to " 1113 "limits or memory shortage, %s\n", 1114 s, __func__, 1115 V_tcp_sc_rst_sock_fail ? 1116 "sending RST" : "try again"); 1117 if (V_tcp_sc_rst_sock_fail) { 1118 rstreason = BANDLIM_UNLIMITED; 1119 goto dropwithreset; 1120 } else 1121 goto dropunlock; 1122 } 1123 /* 1124 * Socket is created in state SYN_RECEIVED. 1125 * Unlock the listen socket, lock the newly 1126 * created socket and update the tp variable. 1127 */ 1128 INP_WUNLOCK(inp); /* listen socket */ 1129 inp = sotoinpcb(so); 1130 INP_WLOCK(inp); /* new connection */ 1131 tp = intotcpcb(inp); 1132 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1133 ("%s: ", __func__)); 1134#ifdef TCP_SIGNATURE 1135 if (sig_checked == 0) { 1136 tcp_dooptions(&to, optp, optlen, 1137 (thflags & TH_SYN) ? TO_SYN : 0); 1138 if (!tcp_signature_verify_input(m, off0, tlen, 1139 optlen, &to, th, tp->t_flags)) { 1140 1141 /* 1142 * In SYN_SENT state if it receives an 1143 * RST, it is allowed for further 1144 * processing. 1145 */ 1146 if ((thflags & TH_RST) == 0 || 1147 (tp->t_state == TCPS_SYN_SENT) == 0) 1148 goto dropunlock; 1149 } 1150 sig_checked = 1; 1151 } 1152#endif 1153 1154 /* 1155 * Process the segment and the data it 1156 * contains. tcp_do_segment() consumes 1157 * the mbuf chain and unlocks the inpcb. 1158 */ 1159 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1160 iptos, ti_locked); 1161 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1162 return; 1163 } 1164 /* 1165 * Segment flag validation for new connection attempts: 1166 * 1167 * Our (SYN|ACK) response was rejected. 1168 * Check with syncache and remove entry to prevent 1169 * retransmits. 1170 * 1171 * NB: syncache_chkrst does its own logging of failure 1172 * causes. 1173 */ 1174 if (thflags & TH_RST) { 1175 syncache_chkrst(&inc, th); 1176 goto dropunlock; 1177 } 1178 /* 1179 * We can't do anything without SYN. 1180 */ 1181 if ((thflags & TH_SYN) == 0) { 1182 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1183 log(LOG_DEBUG, "%s; %s: Listen socket: " 1184 "SYN is missing, segment ignored\n", 1185 s, __func__); 1186 TCPSTAT_INC(tcps_badsyn); 1187 goto dropunlock; 1188 } 1189 /* 1190 * (SYN|ACK) is bogus on a listen socket. 1191 */ 1192 if (thflags & TH_ACK) { 1193 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1194 log(LOG_DEBUG, "%s; %s: Listen socket: " 1195 "SYN|ACK invalid, segment rejected\n", 1196 s, __func__); 1197 syncache_badack(&inc); /* XXX: Not needed! */ 1198 TCPSTAT_INC(tcps_badsyn); 1199 rstreason = BANDLIM_RST_OPENPORT; 1200 goto dropwithreset; 1201 } 1202 /* 1203 * If the drop_synfin option is enabled, drop all 1204 * segments with both the SYN and FIN bits set. 1205 * This prevents e.g. nmap from identifying the 1206 * TCP/IP stack. 1207 * XXX: Poor reasoning. nmap has other methods 1208 * and is constantly refining its stack detection 1209 * strategies. 1210 * XXX: This is a violation of the TCP specification 1211 * and was used by RFC1644. 1212 */ 1213 if ((thflags & TH_FIN) && V_drop_synfin) { 1214 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1215 log(LOG_DEBUG, "%s; %s: Listen socket: " 1216 "SYN|FIN segment ignored (based on " 1217 "sysctl setting)\n", s, __func__); 1218 TCPSTAT_INC(tcps_badsyn); 1219 goto dropunlock; 1220 } 1221 /* 1222 * Segment's flags are (SYN) or (SYN|FIN). 1223 * 1224 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1225 * as they do not affect the state of the TCP FSM. 1226 * The data pointed to by TH_URG and th_urp is ignored. 1227 */ 1228 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1229 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1230 KASSERT(thflags & (TH_SYN), 1231 ("%s: Listen socket: TH_SYN not set", __func__)); 1232#ifdef INET6 1233 /* 1234 * If deprecated address is forbidden, 1235 * we do not accept SYN to deprecated interface 1236 * address to prevent any new inbound connection from 1237 * getting established. 1238 * When we do not accept SYN, we send a TCP RST, 1239 * with deprecated source address (instead of dropping 1240 * it). We compromise it as it is much better for peer 1241 * to send a RST, and RST will be the final packet 1242 * for the exchange. 1243 * 1244 * If we do not forbid deprecated addresses, we accept 1245 * the SYN packet. RFC2462 does not suggest dropping 1246 * SYN in this case. 1247 * If we decipher RFC2462 5.5.4, it says like this: 1248 * 1. use of deprecated addr with existing 1249 * communication is okay - "SHOULD continue to be 1250 * used" 1251 * 2. use of it with new communication: 1252 * (2a) "SHOULD NOT be used if alternate address 1253 * with sufficient scope is available" 1254 * (2b) nothing mentioned otherwise. 1255 * Here we fall into (2b) case as we have no choice in 1256 * our source address selection - we must obey the peer. 1257 * 1258 * The wording in RFC2462 is confusing, and there are 1259 * multiple description text for deprecated address 1260 * handling - worse, they are not exactly the same. 1261 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1262 */ 1263 if (isipv6 && !V_ip6_use_deprecated) { 1264 struct in6_ifaddr *ia6; 1265 1266 ia6 = ip6_getdstifaddr(m); 1267 if (ia6 != NULL && 1268 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1269 ifa_free(&ia6->ia_ifa); 1270 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1271 log(LOG_DEBUG, "%s; %s: Listen socket: " 1272 "Connection attempt to deprecated " 1273 "IPv6 address rejected\n", 1274 s, __func__); 1275 rstreason = BANDLIM_RST_OPENPORT; 1276 goto dropwithreset; 1277 } 1278 if (ia6) 1279 ifa_free(&ia6->ia_ifa); 1280 } 1281#endif /* INET6 */ 1282 /* 1283 * Basic sanity checks on incoming SYN requests: 1284 * Don't respond if the destination is a link layer 1285 * broadcast according to RFC1122 4.2.3.10, p. 104. 1286 * If it is from this socket it must be forged. 1287 * Don't respond if the source or destination is a 1288 * global or subnet broad- or multicast address. 1289 * Note that it is quite possible to receive unicast 1290 * link-layer packets with a broadcast IP address. Use 1291 * in_broadcast() to find them. 1292 */ 1293 if (m->m_flags & (M_BCAST|M_MCAST)) { 1294 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1295 log(LOG_DEBUG, "%s; %s: Listen socket: " 1296 "Connection attempt from broad- or multicast " 1297 "link layer address ignored\n", s, __func__); 1298 goto dropunlock; 1299 } 1300#ifdef INET6 1301 if (isipv6) { 1302 if (th->th_dport == th->th_sport && 1303 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1304 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1305 log(LOG_DEBUG, "%s; %s: Listen socket: " 1306 "Connection attempt to/from self " 1307 "ignored\n", s, __func__); 1308 goto dropunlock; 1309 } 1310 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1311 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1312 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1313 log(LOG_DEBUG, "%s; %s: Listen socket: " 1314 "Connection attempt from/to multicast " 1315 "address ignored\n", s, __func__); 1316 goto dropunlock; 1317 } 1318 } 1319#endif 1320#if defined(INET) && defined(INET6) 1321 else 1322#endif 1323#ifdef INET 1324 { 1325 if (th->th_dport == th->th_sport && 1326 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1327 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1328 log(LOG_DEBUG, "%s; %s: Listen socket: " 1329 "Connection attempt from/to self " 1330 "ignored\n", s, __func__); 1331 goto dropunlock; 1332 } 1333 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1334 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1335 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1336 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1337 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1338 log(LOG_DEBUG, "%s; %s: Listen socket: " 1339 "Connection attempt from/to broad- " 1340 "or multicast address ignored\n", 1341 s, __func__); 1342 goto dropunlock; 1343 } 1344 } 1345#endif 1346 /* 1347 * SYN appears to be valid. Create compressed TCP state 1348 * for syncache. 1349 */ 1350#ifdef TCPDEBUG 1351 if (so->so_options & SO_DEBUG) 1352 tcp_trace(TA_INPUT, ostate, tp, 1353 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1354#endif 1355 tcp_dooptions(&to, optp, optlen, TO_SYN); 1356 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1357 /* 1358 * Entry added to syncache and mbuf consumed. 1359 * Everything already unlocked by syncache_add(). 1360 */ 1361 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1362 return; 1363 } else if (tp->t_state == TCPS_LISTEN) { 1364 /* 1365 * When a listen socket is torn down the SO_ACCEPTCONN 1366 * flag is removed first while connections are drained 1367 * from the accept queue in a unlock/lock cycle of the 1368 * ACCEPT_LOCK, opening a race condition allowing a SYN 1369 * attempt go through unhandled. 1370 */ 1371 goto dropunlock; 1372 } 1373 1374#ifdef TCP_SIGNATURE 1375 if (sig_checked == 0) { 1376 tcp_dooptions(&to, optp, optlen, 1377 (thflags & TH_SYN) ? TO_SYN : 0); 1378 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1379 th, tp->t_flags)) { 1380 1381 /* 1382 * In SYN_SENT state if it receives an RST, it is 1383 * allowed for further processing. 1384 */ 1385 if ((thflags & TH_RST) == 0 || 1386 (tp->t_state == TCPS_SYN_SENT) == 0) 1387 goto dropunlock; 1388 } 1389 sig_checked = 1; 1390 } 1391#endif 1392 1393 TCP_PROBE5(receive, NULL, tp, m->m_data, tp, th); 1394 1395 /* 1396 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1397 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1398 * the inpcb, and unlocks pcbinfo. 1399 */ 1400 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1401 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1402 return; 1403 1404dropwithreset: 1405 TCP_PROBE5(receive, NULL, tp, m->m_data, tp, th); 1406 1407 if (ti_locked == TI_WLOCKED) { 1408 INP_INFO_WUNLOCK(&V_tcbinfo); 1409 ti_locked = TI_UNLOCKED; 1410 } 1411#ifdef INVARIANTS 1412 else { 1413 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1414 "ti_locked: %d", __func__, ti_locked)); 1415 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1416 } 1417#endif 1418 1419 if (inp != NULL) { 1420 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1421 INP_WUNLOCK(inp); 1422 } else 1423 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1424 m = NULL; /* mbuf chain got consumed. */ 1425 goto drop; 1426 1427dropunlock: 1428 if (m != NULL) 1429 TCP_PROBE5(receive, NULL, tp, m->m_data, tp, th); 1430 1431 if (ti_locked == TI_WLOCKED) { 1432 INP_INFO_WUNLOCK(&V_tcbinfo); 1433 ti_locked = TI_UNLOCKED; 1434 } 1435#ifdef INVARIANTS 1436 else { 1437 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1438 "ti_locked: %d", __func__, ti_locked)); 1439 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1440 } 1441#endif 1442 1443 if (inp != NULL) 1444 INP_WUNLOCK(inp); 1445 1446drop: 1447 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1448 if (s != NULL) 1449 free(s, M_TCPLOG); 1450 if (m != NULL) 1451 m_freem(m); 1452} 1453 1454static void 1455tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1456 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1457 int ti_locked) 1458{ 1459 int thflags, acked, ourfinisacked, needoutput = 0; 1460 int rstreason, todrop, win; 1461 u_long tiwin; 1462 char *s; 1463 struct in_conninfo *inc; 1464 struct mbuf *mfree; 1465 struct tcpopt to; 1466 1467#ifdef TCPDEBUG 1468 /* 1469 * The size of tcp_saveipgen must be the size of the max ip header, 1470 * now IPv6. 1471 */ 1472 u_char tcp_saveipgen[IP6_HDR_LEN]; 1473 struct tcphdr tcp_savetcp; 1474 short ostate = 0; 1475#endif 1476 thflags = th->th_flags; 1477 inc = &tp->t_inpcb->inp_inc; 1478 tp->sackhint.last_sack_ack = 0; 1479 1480 /* 1481 * If this is either a state-changing packet or current state isn't 1482 * established, we require a write lock on tcbinfo. Otherwise, we 1483 * allow the tcbinfo to be in either alocked or unlocked, as the 1484 * caller may have unnecessarily acquired a write lock due to a race. 1485 */ 1486 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1487 tp->t_state != TCPS_ESTABLISHED) { 1488 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1489 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1490 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1491 } else { 1492#ifdef INVARIANTS 1493 if (ti_locked == TI_WLOCKED) 1494 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1495 else { 1496 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1497 "ti_locked: %d", __func__, ti_locked)); 1498 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1499 } 1500#endif 1501 } 1502 INP_WLOCK_ASSERT(tp->t_inpcb); 1503 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1504 __func__)); 1505 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1506 __func__)); 1507 1508 /* 1509 * Segment received on connection. 1510 * Reset idle time and keep-alive timer. 1511 * XXX: This should be done after segment 1512 * validation to ignore broken/spoofed segs. 1513 */ 1514 tp->t_rcvtime = ticks; 1515 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1516 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1517 1518 /* 1519 * Unscale the window into a 32-bit value. 1520 * For the SYN_SENT state the scale is zero. 1521 */ 1522 tiwin = th->th_win << tp->snd_scale; 1523 1524 /* 1525 * TCP ECN processing. 1526 */ 1527 if (tp->t_flags & TF_ECN_PERMIT) { 1528 if (thflags & TH_CWR) 1529 tp->t_flags &= ~TF_ECN_SND_ECE; 1530 switch (iptos & IPTOS_ECN_MASK) { 1531 case IPTOS_ECN_CE: 1532 tp->t_flags |= TF_ECN_SND_ECE; 1533 TCPSTAT_INC(tcps_ecn_ce); 1534 break; 1535 case IPTOS_ECN_ECT0: 1536 TCPSTAT_INC(tcps_ecn_ect0); 1537 break; 1538 case IPTOS_ECN_ECT1: 1539 TCPSTAT_INC(tcps_ecn_ect1); 1540 break; 1541 } 1542 /* Congestion experienced. */ 1543 if (thflags & TH_ECE) { 1544 cc_cong_signal(tp, th, CC_ECN); 1545 } 1546 } 1547 1548 /* 1549 * Parse options on any incoming segment. 1550 */ 1551 tcp_dooptions(&to, (u_char *)(th + 1), 1552 (th->th_off << 2) - sizeof(struct tcphdr), 1553 (thflags & TH_SYN) ? TO_SYN : 0); 1554 1555 /* 1556 * If echoed timestamp is later than the current time, 1557 * fall back to non RFC1323 RTT calculation. Normalize 1558 * timestamp if syncookies were used when this connection 1559 * was established. 1560 */ 1561 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1562 to.to_tsecr -= tp->ts_offset; 1563 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1564 to.to_tsecr = 0; 1565 } 1566 /* 1567 * If timestamps were negotiated during SYN/ACK they should 1568 * appear on every segment during this session and vice versa. 1569 */ 1570 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1571 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1572 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1573 "no action\n", s, __func__); 1574 free(s, M_TCPLOG); 1575 } 1576 } 1577 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1578 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1579 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1580 "no action\n", s, __func__); 1581 free(s, M_TCPLOG); 1582 } 1583 } 1584 1585 /* 1586 * Process options only when we get SYN/ACK back. The SYN case 1587 * for incoming connections is handled in tcp_syncache. 1588 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1589 * or <SYN,ACK>) segment itself is never scaled. 1590 * XXX this is traditional behavior, may need to be cleaned up. 1591 */ 1592 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1593 if ((to.to_flags & TOF_SCALE) && 1594 (tp->t_flags & TF_REQ_SCALE)) { 1595 tp->t_flags |= TF_RCVD_SCALE; 1596 tp->snd_scale = to.to_wscale; 1597 } 1598 /* 1599 * Initial send window. It will be updated with 1600 * the next incoming segment to the scaled value. 1601 */ 1602 tp->snd_wnd = th->th_win; 1603 if (to.to_flags & TOF_TS) { 1604 tp->t_flags |= TF_RCVD_TSTMP; 1605 tp->ts_recent = to.to_tsval; 1606 tp->ts_recent_age = tcp_ts_getticks(); 1607 } 1608 if (to.to_flags & TOF_MSS) 1609 tcp_mss(tp, to.to_mss); 1610 if ((tp->t_flags & TF_SACK_PERMIT) && 1611 (to.to_flags & TOF_SACKPERM) == 0) 1612 tp->t_flags &= ~TF_SACK_PERMIT; 1613 } 1614 1615 /* 1616 * Header prediction: check for the two common cases 1617 * of a uni-directional data xfer. If the packet has 1618 * no control flags, is in-sequence, the window didn't 1619 * change and we're not retransmitting, it's a 1620 * candidate. If the length is zero and the ack moved 1621 * forward, we're the sender side of the xfer. Just 1622 * free the data acked & wake any higher level process 1623 * that was blocked waiting for space. If the length 1624 * is non-zero and the ack didn't move, we're the 1625 * receiver side. If we're getting packets in-order 1626 * (the reassembly queue is empty), add the data to 1627 * the socket buffer and note that we need a delayed ack. 1628 * Make sure that the hidden state-flags are also off. 1629 * Since we check for TCPS_ESTABLISHED first, it can only 1630 * be TH_NEEDSYN. 1631 */ 1632 if (tp->t_state == TCPS_ESTABLISHED && 1633 th->th_seq == tp->rcv_nxt && 1634 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1635 tp->snd_nxt == tp->snd_max && 1636 tiwin && tiwin == tp->snd_wnd && 1637 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1638 LIST_EMPTY(&tp->t_segq) && 1639 ((to.to_flags & TOF_TS) == 0 || 1640 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1641 1642 /* 1643 * If last ACK falls within this segment's sequence numbers, 1644 * record the timestamp. 1645 * NOTE that the test is modified according to the latest 1646 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1647 */ 1648 if ((to.to_flags & TOF_TS) != 0 && 1649 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1650 tp->ts_recent_age = tcp_ts_getticks(); 1651 tp->ts_recent = to.to_tsval; 1652 } 1653 1654 if (tlen == 0) { 1655 if (SEQ_GT(th->th_ack, tp->snd_una) && 1656 SEQ_LEQ(th->th_ack, tp->snd_max) && 1657 !IN_RECOVERY(tp->t_flags) && 1658 (to.to_flags & TOF_SACK) == 0 && 1659 TAILQ_EMPTY(&tp->snd_holes)) { 1660 /* 1661 * This is a pure ack for outstanding data. 1662 */ 1663 if (ti_locked == TI_WLOCKED) 1664 INP_INFO_WUNLOCK(&V_tcbinfo); 1665 ti_locked = TI_UNLOCKED; 1666 1667 TCPSTAT_INC(tcps_predack); 1668 1669 /* 1670 * "bad retransmit" recovery. 1671 */ 1672 if (tp->t_rxtshift == 1 && 1673 tp->t_flags & TF_PREVVALID && 1674 (int)(ticks - tp->t_badrxtwin) < 0) { 1675 cc_cong_signal(tp, th, CC_RTO_ERR); 1676 } 1677 1678 /* 1679 * Recalculate the transmit timer / rtt. 1680 * 1681 * Some boxes send broken timestamp replies 1682 * during the SYN+ACK phase, ignore 1683 * timestamps of 0 or we could calculate a 1684 * huge RTT and blow up the retransmit timer. 1685 */ 1686 if ((to.to_flags & TOF_TS) != 0 && 1687 to.to_tsecr) { 1688 u_int t; 1689 1690 t = tcp_ts_getticks() - to.to_tsecr; 1691 if (!tp->t_rttlow || tp->t_rttlow > t) 1692 tp->t_rttlow = t; 1693 tcp_xmit_timer(tp, 1694 TCP_TS_TO_TICKS(t) + 1); 1695 } else if (tp->t_rtttime && 1696 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1697 if (!tp->t_rttlow || 1698 tp->t_rttlow > ticks - tp->t_rtttime) 1699 tp->t_rttlow = ticks - tp->t_rtttime; 1700 tcp_xmit_timer(tp, 1701 ticks - tp->t_rtttime); 1702 } 1703 acked = BYTES_THIS_ACK(tp, th); 1704 1705 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1706 hhook_run_tcp_est_in(tp, th, &to); 1707 1708 TCPSTAT_INC(tcps_rcvackpack); 1709 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1710 sbdrop(&so->so_snd, acked); 1711 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1712 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1713 tp->snd_recover = th->th_ack - 1; 1714 1715 /* 1716 * Let the congestion control algorithm update 1717 * congestion control related information. This 1718 * typically means increasing the congestion 1719 * window. 1720 */ 1721 cc_ack_received(tp, th, CC_ACK); 1722 1723 tp->snd_una = th->th_ack; 1724 /* 1725 * Pull snd_wl2 up to prevent seq wrap relative 1726 * to th_ack. 1727 */ 1728 tp->snd_wl2 = th->th_ack; 1729 tp->t_dupacks = 0; 1730 m_freem(m); 1731 ND6_HINT(tp); /* Some progress has been made. */ 1732 1733 /* 1734 * If all outstanding data are acked, stop 1735 * retransmit timer, otherwise restart timer 1736 * using current (possibly backed-off) value. 1737 * If process is waiting for space, 1738 * wakeup/selwakeup/signal. If data 1739 * are ready to send, let tcp_output 1740 * decide between more output or persist. 1741 */ 1742#ifdef TCPDEBUG 1743 if (so->so_options & SO_DEBUG) 1744 tcp_trace(TA_INPUT, ostate, tp, 1745 (void *)tcp_saveipgen, 1746 &tcp_savetcp, 0); 1747#endif 1748 if (tp->snd_una == tp->snd_max) 1749 tcp_timer_activate(tp, TT_REXMT, 0); 1750 else if (!tcp_timer_active(tp, TT_PERSIST)) 1751 tcp_timer_activate(tp, TT_REXMT, 1752 tp->t_rxtcur); 1753 sowwakeup(so); 1754 if (so->so_snd.sb_cc) 1755 (void) tcp_output(tp); 1756 goto check_delack; 1757 } 1758 } else if (th->th_ack == tp->snd_una && 1759 tlen <= sbspace(&so->so_rcv)) { 1760 int newsize = 0; /* automatic sockbuf scaling */ 1761 1762 /* 1763 * This is a pure, in-sequence data packet with 1764 * nothing on the reassembly queue and we have enough 1765 * buffer space to take it. 1766 */ 1767 if (ti_locked == TI_WLOCKED) 1768 INP_INFO_WUNLOCK(&V_tcbinfo); 1769 ti_locked = TI_UNLOCKED; 1770 1771 /* Clean receiver SACK report if present */ 1772 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1773 tcp_clean_sackreport(tp); 1774 TCPSTAT_INC(tcps_preddat); 1775 tp->rcv_nxt += tlen; 1776 /* 1777 * Pull snd_wl1 up to prevent seq wrap relative to 1778 * th_seq. 1779 */ 1780 tp->snd_wl1 = th->th_seq; 1781 /* 1782 * Pull rcv_up up to prevent seq wrap relative to 1783 * rcv_nxt. 1784 */ 1785 tp->rcv_up = tp->rcv_nxt; 1786 TCPSTAT_INC(tcps_rcvpack); 1787 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1788 ND6_HINT(tp); /* Some progress has been made */ 1789#ifdef TCPDEBUG 1790 if (so->so_options & SO_DEBUG) 1791 tcp_trace(TA_INPUT, ostate, tp, 1792 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1793#endif 1794 /* 1795 * Automatic sizing of receive socket buffer. Often the send 1796 * buffer size is not optimally adjusted to the actual network 1797 * conditions at hand (delay bandwidth product). Setting the 1798 * buffer size too small limits throughput on links with high 1799 * bandwidth and high delay (eg. trans-continental/oceanic links). 1800 * 1801 * On the receive side the socket buffer memory is only rarely 1802 * used to any significant extent. This allows us to be much 1803 * more aggressive in scaling the receive socket buffer. For 1804 * the case that the buffer space is actually used to a large 1805 * extent and we run out of kernel memory we can simply drop 1806 * the new segments; TCP on the sender will just retransmit it 1807 * later. Setting the buffer size too big may only consume too 1808 * much kernel memory if the application doesn't read() from 1809 * the socket or packet loss or reordering makes use of the 1810 * reassembly queue. 1811 * 1812 * The criteria to step up the receive buffer one notch are: 1813 * 1. the number of bytes received during the time it takes 1814 * one timestamp to be reflected back to us (the RTT); 1815 * 2. received bytes per RTT is within seven eighth of the 1816 * current socket buffer size; 1817 * 3. receive buffer size has not hit maximal automatic size; 1818 * 1819 * This algorithm does one step per RTT at most and only if 1820 * we receive a bulk stream w/o packet losses or reorderings. 1821 * Shrinking the buffer during idle times is not necessary as 1822 * it doesn't consume any memory when idle. 1823 * 1824 * TODO: Only step up if the application is actually serving 1825 * the buffer to better manage the socket buffer resources. 1826 */ 1827 if (V_tcp_do_autorcvbuf && 1828 to.to_tsecr && 1829 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1830 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1831 to.to_tsecr - tp->rfbuf_ts < hz) { 1832 if (tp->rfbuf_cnt > 1833 (so->so_rcv.sb_hiwat / 8 * 7) && 1834 so->so_rcv.sb_hiwat < 1835 V_tcp_autorcvbuf_max) { 1836 newsize = 1837 min(so->so_rcv.sb_hiwat + 1838 V_tcp_autorcvbuf_inc, 1839 V_tcp_autorcvbuf_max); 1840 } 1841 /* Start over with next RTT. */ 1842 tp->rfbuf_ts = 0; 1843 tp->rfbuf_cnt = 0; 1844 } else 1845 tp->rfbuf_cnt += tlen; /* add up */ 1846 } 1847 1848 /* Add data to socket buffer. */ 1849 SOCKBUF_LOCK(&so->so_rcv); 1850 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1851 m_freem(m); 1852 } else { 1853 /* 1854 * Set new socket buffer size. 1855 * Give up when limit is reached. 1856 */ 1857 if (newsize) 1858 if (!sbreserve_locked(&so->so_rcv, 1859 newsize, so, NULL)) 1860 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1861 m_adj(m, drop_hdrlen); /* delayed header drop */ 1862 sbappendstream_locked(&so->so_rcv, m); 1863 } 1864 /* NB: sorwakeup_locked() does an implicit unlock. */ 1865 sorwakeup_locked(so); 1866 if (DELAY_ACK(tp)) { 1867 tp->t_flags |= TF_DELACK; 1868 } else { 1869 tp->t_flags |= TF_ACKNOW; 1870 tcp_output(tp); 1871 } 1872 goto check_delack; 1873 } 1874 } 1875 1876 /* 1877 * Calculate amount of space in receive window, 1878 * and then do TCP input processing. 1879 * Receive window is amount of space in rcv queue, 1880 * but not less than advertised window. 1881 */ 1882 win = sbspace(&so->so_rcv); 1883 if (win < 0) 1884 win = 0; 1885 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1886 1887 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1888 tp->rfbuf_ts = 0; 1889 tp->rfbuf_cnt = 0; 1890 1891 switch (tp->t_state) { 1892 1893 /* 1894 * If the state is SYN_RECEIVED: 1895 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1896 */ 1897 case TCPS_SYN_RECEIVED: 1898 if ((thflags & TH_ACK) && 1899 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1900 SEQ_GT(th->th_ack, tp->snd_max))) { 1901 rstreason = BANDLIM_RST_OPENPORT; 1902 goto dropwithreset; 1903 } 1904 break; 1905 1906 /* 1907 * If the state is SYN_SENT: 1908 * if seg contains an ACK, but not for our SYN, drop the input. 1909 * if seg contains a RST, then drop the connection. 1910 * if seg does not contain SYN, then drop it. 1911 * Otherwise this is an acceptable SYN segment 1912 * initialize tp->rcv_nxt and tp->irs 1913 * if seg contains ack then advance tp->snd_una 1914 * if seg contains an ECE and ECN support is enabled, the stream 1915 * is ECN capable. 1916 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1917 * arrange for segment to be acked (eventually) 1918 * continue processing rest of data/controls, beginning with URG 1919 */ 1920 case TCPS_SYN_SENT: 1921 if ((thflags & TH_ACK) && 1922 (SEQ_LEQ(th->th_ack, tp->iss) || 1923 SEQ_GT(th->th_ack, tp->snd_max))) { 1924 rstreason = BANDLIM_UNLIMITED; 1925 goto dropwithreset; 1926 } 1927 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1928 TCP_PROBE5(connect_refused, NULL, tp, m->m_data, tp, 1929 th); 1930 tp = tcp_drop(tp, ECONNREFUSED); 1931 } 1932 if (thflags & TH_RST) 1933 goto drop; 1934 if (!(thflags & TH_SYN)) 1935 goto drop; 1936 1937 tp->irs = th->th_seq; 1938 tcp_rcvseqinit(tp); 1939 if (thflags & TH_ACK) { 1940 TCPSTAT_INC(tcps_connects); 1941 soisconnected(so); 1942#ifdef MAC 1943 mac_socketpeer_set_from_mbuf(m, so); 1944#endif 1945 /* Do window scaling on this connection? */ 1946 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1947 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1948 tp->rcv_scale = tp->request_r_scale; 1949 } 1950 tp->rcv_adv += imin(tp->rcv_wnd, 1951 TCP_MAXWIN << tp->rcv_scale); 1952 tp->snd_una++; /* SYN is acked */ 1953 /* 1954 * If there's data, delay ACK; if there's also a FIN 1955 * ACKNOW will be turned on later. 1956 */ 1957 if (DELAY_ACK(tp) && tlen != 0) 1958 tcp_timer_activate(tp, TT_DELACK, 1959 tcp_delacktime); 1960 else 1961 tp->t_flags |= TF_ACKNOW; 1962 1963 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1964 tp->t_flags |= TF_ECN_PERMIT; 1965 TCPSTAT_INC(tcps_ecn_shs); 1966 } 1967 1968 /* 1969 * Received <SYN,ACK> in SYN_SENT[*] state. 1970 * Transitions: 1971 * SYN_SENT --> ESTABLISHED 1972 * SYN_SENT* --> FIN_WAIT_1 1973 */ 1974 tp->t_starttime = ticks; 1975 if (tp->t_flags & TF_NEEDFIN) { 1976 tcp_state_change(tp, TCPS_FIN_WAIT_1); 1977 tp->t_flags &= ~TF_NEEDFIN; 1978 thflags &= ~TH_SYN; 1979 } else { 1980 tcp_state_change(tp, TCPS_ESTABLISHED); 1981 TCP_PROBE5(connect_established, NULL, tp, 1982 m->m_data, tp, th); 1983 cc_conn_init(tp); 1984 tcp_timer_activate(tp, TT_KEEP, 1985 TP_KEEPIDLE(tp)); 1986 } 1987 } else { 1988 /* 1989 * Received initial SYN in SYN-SENT[*] state => 1990 * simultaneous open. If segment contains CC option 1991 * and there is a cached CC, apply TAO test. 1992 * If it succeeds, connection is * half-synchronized. 1993 * Otherwise, do 3-way handshake: 1994 * SYN-SENT -> SYN-RECEIVED 1995 * SYN-SENT* -> SYN-RECEIVED* 1996 * If there was no CC option, clear cached CC value. 1997 */ 1998 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1999 tcp_timer_activate(tp, TT_REXMT, 0); 2000 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2001 } 2002 2003 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 2004 "ti_locked %d", __func__, ti_locked)); 2005 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2006 INP_WLOCK_ASSERT(tp->t_inpcb); 2007 2008 /* 2009 * Advance th->th_seq to correspond to first data byte. 2010 * If data, trim to stay within window, 2011 * dropping FIN if necessary. 2012 */ 2013 th->th_seq++; 2014 if (tlen > tp->rcv_wnd) { 2015 todrop = tlen - tp->rcv_wnd; 2016 m_adj(m, -todrop); 2017 tlen = tp->rcv_wnd; 2018 thflags &= ~TH_FIN; 2019 TCPSTAT_INC(tcps_rcvpackafterwin); 2020 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2021 } 2022 tp->snd_wl1 = th->th_seq - 1; 2023 tp->rcv_up = th->th_seq; 2024 /* 2025 * Client side of transaction: already sent SYN and data. 2026 * If the remote host used T/TCP to validate the SYN, 2027 * our data will be ACK'd; if so, enter normal data segment 2028 * processing in the middle of step 5, ack processing. 2029 * Otherwise, goto step 6. 2030 */ 2031 if (thflags & TH_ACK) 2032 goto process_ACK; 2033 2034 goto step6; 2035 2036 /* 2037 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2038 * do normal processing. 2039 * 2040 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2041 */ 2042 case TCPS_LAST_ACK: 2043 case TCPS_CLOSING: 2044 break; /* continue normal processing */ 2045 } 2046 2047 /* 2048 * States other than LISTEN or SYN_SENT. 2049 * First check the RST flag and sequence number since reset segments 2050 * are exempt from the timestamp and connection count tests. This 2051 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2052 * below which allowed reset segments in half the sequence space 2053 * to fall though and be processed (which gives forged reset 2054 * segments with a random sequence number a 50 percent chance of 2055 * killing a connection). 2056 * Then check timestamp, if present. 2057 * Then check the connection count, if present. 2058 * Then check that at least some bytes of segment are within 2059 * receive window. If segment begins before rcv_nxt, 2060 * drop leading data (and SYN); if nothing left, just ack. 2061 * 2062 * 2063 * If the RST bit is set, check the sequence number to see 2064 * if this is a valid reset segment. 2065 * RFC 793 page 37: 2066 * In all states except SYN-SENT, all reset (RST) segments 2067 * are validated by checking their SEQ-fields. A reset is 2068 * valid if its sequence number is in the window. 2069 * Note: this does not take into account delayed ACKs, so 2070 * we should test against last_ack_sent instead of rcv_nxt. 2071 * The sequence number in the reset segment is normally an 2072 * echo of our outgoing acknowlegement numbers, but some hosts 2073 * send a reset with the sequence number at the rightmost edge 2074 * of our receive window, and we have to handle this case. 2075 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2076 * that brute force RST attacks are possible. To combat this, 2077 * we use a much stricter check while in the ESTABLISHED state, 2078 * only accepting RSTs where the sequence number is equal to 2079 * last_ack_sent. In all other states (the states in which a 2080 * RST is more likely), the more permissive check is used. 2081 * If we have multiple segments in flight, the initial reset 2082 * segment sequence numbers will be to the left of last_ack_sent, 2083 * but they will eventually catch up. 2084 * In any case, it never made sense to trim reset segments to 2085 * fit the receive window since RFC 1122 says: 2086 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2087 * 2088 * A TCP SHOULD allow a received RST segment to include data. 2089 * 2090 * DISCUSSION 2091 * It has been suggested that a RST segment could contain 2092 * ASCII text that encoded and explained the cause of the 2093 * RST. No standard has yet been established for such 2094 * data. 2095 * 2096 * If the reset segment passes the sequence number test examine 2097 * the state: 2098 * SYN_RECEIVED STATE: 2099 * If passive open, return to LISTEN state. 2100 * If active open, inform user that connection was refused. 2101 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2102 * Inform user that connection was reset, and close tcb. 2103 * CLOSING, LAST_ACK STATES: 2104 * Close the tcb. 2105 * TIME_WAIT STATE: 2106 * Drop the segment - see Stevens, vol. 2, p. 964 and 2107 * RFC 1337. 2108 */ 2109 if (thflags & TH_RST) { 2110 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2111 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2112 switch (tp->t_state) { 2113 2114 case TCPS_SYN_RECEIVED: 2115 so->so_error = ECONNREFUSED; 2116 goto close; 2117 2118 case TCPS_ESTABLISHED: 2119 if (V_tcp_insecure_rst == 0 && 2120 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2121 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2122 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2123 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2124 TCPSTAT_INC(tcps_badrst); 2125 goto drop; 2126 } 2127 /* FALLTHROUGH */ 2128 case TCPS_FIN_WAIT_1: 2129 case TCPS_FIN_WAIT_2: 2130 case TCPS_CLOSE_WAIT: 2131 so->so_error = ECONNRESET; 2132 close: 2133 KASSERT(ti_locked == TI_WLOCKED, 2134 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2135 ti_locked)); 2136 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2137 2138 tcp_state_change(tp, TCPS_CLOSED); 2139 TCPSTAT_INC(tcps_drops); 2140 tp = tcp_close(tp); 2141 break; 2142 2143 case TCPS_CLOSING: 2144 case TCPS_LAST_ACK: 2145 KASSERT(ti_locked == TI_WLOCKED, 2146 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2147 ti_locked)); 2148 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2149 2150 tp = tcp_close(tp); 2151 break; 2152 } 2153 } 2154 goto drop; 2155 } 2156 2157 /* 2158 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2159 * and it's less than ts_recent, drop it. 2160 */ 2161 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2162 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2163 2164 /* Check to see if ts_recent is over 24 days old. */ 2165 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2166 /* 2167 * Invalidate ts_recent. If this segment updates 2168 * ts_recent, the age will be reset later and ts_recent 2169 * will get a valid value. If it does not, setting 2170 * ts_recent to zero will at least satisfy the 2171 * requirement that zero be placed in the timestamp 2172 * echo reply when ts_recent isn't valid. The 2173 * age isn't reset until we get a valid ts_recent 2174 * because we don't want out-of-order segments to be 2175 * dropped when ts_recent is old. 2176 */ 2177 tp->ts_recent = 0; 2178 } else { 2179 TCPSTAT_INC(tcps_rcvduppack); 2180 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2181 TCPSTAT_INC(tcps_pawsdrop); 2182 if (tlen) 2183 goto dropafterack; 2184 goto drop; 2185 } 2186 } 2187 2188 /* 2189 * In the SYN-RECEIVED state, validate that the packet belongs to 2190 * this connection before trimming the data to fit the receive 2191 * window. Check the sequence number versus IRS since we know 2192 * the sequence numbers haven't wrapped. This is a partial fix 2193 * for the "LAND" DoS attack. 2194 */ 2195 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2196 rstreason = BANDLIM_RST_OPENPORT; 2197 goto dropwithreset; 2198 } 2199 2200 todrop = tp->rcv_nxt - th->th_seq; 2201 if (todrop > 0) { 2202 /* 2203 * If this is a duplicate SYN for our current connection, 2204 * advance over it and pretend and it's not a SYN. 2205 */ 2206 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2207 thflags &= ~TH_SYN; 2208 th->th_seq++; 2209 if (th->th_urp > 1) 2210 th->th_urp--; 2211 else 2212 thflags &= ~TH_URG; 2213 todrop--; 2214 } 2215 /* 2216 * Following if statement from Stevens, vol. 2, p. 960. 2217 */ 2218 if (todrop > tlen 2219 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2220 /* 2221 * Any valid FIN must be to the left of the window. 2222 * At this point the FIN must be a duplicate or out 2223 * of sequence; drop it. 2224 */ 2225 thflags &= ~TH_FIN; 2226 2227 /* 2228 * Send an ACK to resynchronize and drop any data. 2229 * But keep on processing for RST or ACK. 2230 */ 2231 tp->t_flags |= TF_ACKNOW; 2232 todrop = tlen; 2233 TCPSTAT_INC(tcps_rcvduppack); 2234 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2235 } else { 2236 TCPSTAT_INC(tcps_rcvpartduppack); 2237 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2238 } 2239 drop_hdrlen += todrop; /* drop from the top afterwards */ 2240 th->th_seq += todrop; 2241 tlen -= todrop; 2242 if (th->th_urp > todrop) 2243 th->th_urp -= todrop; 2244 else { 2245 thflags &= ~TH_URG; 2246 th->th_urp = 0; 2247 } 2248 } 2249 2250 /* 2251 * If new data are received on a connection after the 2252 * user processes are gone, then RST the other end. 2253 */ 2254 if ((so->so_state & SS_NOFDREF) && 2255 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2256 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2257 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2258 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2259 2260 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2261 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2262 "after socket was closed, " 2263 "sending RST and removing tcpcb\n", 2264 s, __func__, tcpstates[tp->t_state], tlen); 2265 free(s, M_TCPLOG); 2266 } 2267 tp = tcp_close(tp); 2268 TCPSTAT_INC(tcps_rcvafterclose); 2269 rstreason = BANDLIM_UNLIMITED; 2270 goto dropwithreset; 2271 } 2272 2273 /* 2274 * If segment ends after window, drop trailing data 2275 * (and PUSH and FIN); if nothing left, just ACK. 2276 */ 2277 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2278 if (todrop > 0) { 2279 TCPSTAT_INC(tcps_rcvpackafterwin); 2280 if (todrop >= tlen) { 2281 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2282 /* 2283 * If window is closed can only take segments at 2284 * window edge, and have to drop data and PUSH from 2285 * incoming segments. Continue processing, but 2286 * remember to ack. Otherwise, drop segment 2287 * and ack. 2288 */ 2289 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2290 tp->t_flags |= TF_ACKNOW; 2291 TCPSTAT_INC(tcps_rcvwinprobe); 2292 } else 2293 goto dropafterack; 2294 } else 2295 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2296 m_adj(m, -todrop); 2297 tlen -= todrop; 2298 thflags &= ~(TH_PUSH|TH_FIN); 2299 } 2300 2301 /* 2302 * If last ACK falls within this segment's sequence numbers, 2303 * record its timestamp. 2304 * NOTE: 2305 * 1) That the test incorporates suggestions from the latest 2306 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2307 * 2) That updating only on newer timestamps interferes with 2308 * our earlier PAWS tests, so this check should be solely 2309 * predicated on the sequence space of this segment. 2310 * 3) That we modify the segment boundary check to be 2311 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2312 * instead of RFC1323's 2313 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2314 * This modified check allows us to overcome RFC1323's 2315 * limitations as described in Stevens TCP/IP Illustrated 2316 * Vol. 2 p.869. In such cases, we can still calculate the 2317 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2318 */ 2319 if ((to.to_flags & TOF_TS) != 0 && 2320 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2321 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2322 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2323 tp->ts_recent_age = tcp_ts_getticks(); 2324 tp->ts_recent = to.to_tsval; 2325 } 2326 2327 /* 2328 * If a SYN is in the window, then this is an 2329 * error and we send an RST and drop the connection. 2330 */ 2331 if (thflags & TH_SYN) { 2332 KASSERT(ti_locked == TI_WLOCKED, 2333 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2334 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2335 2336 tp = tcp_drop(tp, ECONNRESET); 2337 rstreason = BANDLIM_UNLIMITED; 2338 goto drop; 2339 } 2340 2341 /* 2342 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2343 * flag is on (half-synchronized state), then queue data for 2344 * later processing; else drop segment and return. 2345 */ 2346 if ((thflags & TH_ACK) == 0) { 2347 if (tp->t_state == TCPS_SYN_RECEIVED || 2348 (tp->t_flags & TF_NEEDSYN)) 2349 goto step6; 2350 else if (tp->t_flags & TF_ACKNOW) 2351 goto dropafterack; 2352 else 2353 goto drop; 2354 } 2355 2356 /* 2357 * Ack processing. 2358 */ 2359 switch (tp->t_state) { 2360 2361 /* 2362 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2363 * ESTABLISHED state and continue processing. 2364 * The ACK was checked above. 2365 */ 2366 case TCPS_SYN_RECEIVED: 2367 2368 TCPSTAT_INC(tcps_connects); 2369 soisconnected(so); 2370 /* Do window scaling? */ 2371 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2372 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2373 tp->rcv_scale = tp->request_r_scale; 2374 tp->snd_wnd = tiwin; 2375 } 2376 /* 2377 * Make transitions: 2378 * SYN-RECEIVED -> ESTABLISHED 2379 * SYN-RECEIVED* -> FIN-WAIT-1 2380 */ 2381 tp->t_starttime = ticks; 2382 if (tp->t_flags & TF_NEEDFIN) { 2383 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2384 tp->t_flags &= ~TF_NEEDFIN; 2385 } else { 2386 tcp_state_change(tp, TCPS_ESTABLISHED); 2387 TCP_PROBE5(accept_established, NULL, tp, m->m_data, tp, 2388 th); 2389 cc_conn_init(tp); 2390 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2391 } 2392 /* 2393 * If segment contains data or ACK, will call tcp_reass() 2394 * later; if not, do so now to pass queued data to user. 2395 */ 2396 if (tlen == 0 && (thflags & TH_FIN) == 0) 2397 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2398 (struct mbuf *)0); 2399 tp->snd_wl1 = th->th_seq - 1; 2400 /* FALLTHROUGH */ 2401 2402 /* 2403 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2404 * ACKs. If the ack is in the range 2405 * tp->snd_una < th->th_ack <= tp->snd_max 2406 * then advance tp->snd_una to th->th_ack and drop 2407 * data from the retransmission queue. If this ACK reflects 2408 * more up to date window information we update our window information. 2409 */ 2410 case TCPS_ESTABLISHED: 2411 case TCPS_FIN_WAIT_1: 2412 case TCPS_FIN_WAIT_2: 2413 case TCPS_CLOSE_WAIT: 2414 case TCPS_CLOSING: 2415 case TCPS_LAST_ACK: 2416 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2417 TCPSTAT_INC(tcps_rcvacktoomuch); 2418 goto dropafterack; 2419 } 2420 if ((tp->t_flags & TF_SACK_PERMIT) && 2421 ((to.to_flags & TOF_SACK) || 2422 !TAILQ_EMPTY(&tp->snd_holes))) 2423 tcp_sack_doack(tp, &to, th->th_ack); 2424 2425 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2426 hhook_run_tcp_est_in(tp, th, &to); 2427 2428 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2429 if (tlen == 0 && tiwin == tp->snd_wnd) { 2430 TCPSTAT_INC(tcps_rcvdupack); 2431 /* 2432 * If we have outstanding data (other than 2433 * a window probe), this is a completely 2434 * duplicate ack (ie, window info didn't 2435 * change), the ack is the biggest we've 2436 * seen and we've seen exactly our rexmt 2437 * threshhold of them, assume a packet 2438 * has been dropped and retransmit it. 2439 * Kludge snd_nxt & the congestion 2440 * window so we send only this one 2441 * packet. 2442 * 2443 * We know we're losing at the current 2444 * window size so do congestion avoidance 2445 * (set ssthresh to half the current window 2446 * and pull our congestion window back to 2447 * the new ssthresh). 2448 * 2449 * Dup acks mean that packets have left the 2450 * network (they're now cached at the receiver) 2451 * so bump cwnd by the amount in the receiver 2452 * to keep a constant cwnd packets in the 2453 * network. 2454 * 2455 * When using TCP ECN, notify the peer that 2456 * we reduced the cwnd. 2457 */ 2458 if (!tcp_timer_active(tp, TT_REXMT) || 2459 th->th_ack != tp->snd_una) 2460 tp->t_dupacks = 0; 2461 else if (++tp->t_dupacks > tcprexmtthresh || 2462 IN_FASTRECOVERY(tp->t_flags)) { 2463 cc_ack_received(tp, th, CC_DUPACK); 2464 if ((tp->t_flags & TF_SACK_PERMIT) && 2465 IN_FASTRECOVERY(tp->t_flags)) { 2466 int awnd; 2467 2468 /* 2469 * Compute the amount of data in flight first. 2470 * We can inject new data into the pipe iff 2471 * we have less than 1/2 the original window's 2472 * worth of data in flight. 2473 */ 2474 awnd = (tp->snd_nxt - tp->snd_fack) + 2475 tp->sackhint.sack_bytes_rexmit; 2476 if (awnd < tp->snd_ssthresh) { 2477 tp->snd_cwnd += tp->t_maxseg; 2478 if (tp->snd_cwnd > tp->snd_ssthresh) 2479 tp->snd_cwnd = tp->snd_ssthresh; 2480 } 2481 } else 2482 tp->snd_cwnd += tp->t_maxseg; 2483 if ((thflags & TH_FIN) && 2484 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2485 /* 2486 * If its a fin we need to process 2487 * it to avoid a race where both 2488 * sides enter FIN-WAIT and send FIN|ACK 2489 * at the same time. 2490 */ 2491 break; 2492 } 2493 (void) tcp_output(tp); 2494 goto drop; 2495 } else if (tp->t_dupacks == tcprexmtthresh) { 2496 tcp_seq onxt = tp->snd_nxt; 2497 2498 /* 2499 * If we're doing sack, check to 2500 * see if we're already in sack 2501 * recovery. If we're not doing sack, 2502 * check to see if we're in newreno 2503 * recovery. 2504 */ 2505 if (tp->t_flags & TF_SACK_PERMIT) { 2506 if (IN_FASTRECOVERY(tp->t_flags)) { 2507 tp->t_dupacks = 0; 2508 break; 2509 } 2510 } else { 2511 if (SEQ_LEQ(th->th_ack, 2512 tp->snd_recover)) { 2513 tp->t_dupacks = 0; 2514 break; 2515 } 2516 } 2517 /* Congestion signal before ack. */ 2518 cc_cong_signal(tp, th, CC_NDUPACK); 2519 cc_ack_received(tp, th, CC_DUPACK); 2520 tcp_timer_activate(tp, TT_REXMT, 0); 2521 tp->t_rtttime = 0; 2522 if (tp->t_flags & TF_SACK_PERMIT) { 2523 TCPSTAT_INC( 2524 tcps_sack_recovery_episode); 2525 tp->sack_newdata = tp->snd_nxt; 2526 tp->snd_cwnd = tp->t_maxseg; 2527 (void) tcp_output(tp); 2528 goto drop; 2529 } 2530 tp->snd_nxt = th->th_ack; 2531 tp->snd_cwnd = tp->t_maxseg; 2532 if ((thflags & TH_FIN) && 2533 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2534 /* 2535 * If its a fin we need to process 2536 * it to avoid a race where both 2537 * sides enter FIN-WAIT and send FIN|ACK 2538 * at the same time. 2539 */ 2540 break; 2541 } 2542 (void) tcp_output(tp); 2543 KASSERT(tp->snd_limited <= 2, 2544 ("%s: tp->snd_limited too big", 2545 __func__)); 2546 tp->snd_cwnd = tp->snd_ssthresh + 2547 tp->t_maxseg * 2548 (tp->t_dupacks - tp->snd_limited); 2549 if (SEQ_GT(onxt, tp->snd_nxt)) 2550 tp->snd_nxt = onxt; 2551 goto drop; 2552 } else if (V_tcp_do_rfc3042) { 2553 cc_ack_received(tp, th, CC_DUPACK); 2554 u_long oldcwnd = tp->snd_cwnd; 2555 tcp_seq oldsndmax = tp->snd_max; 2556 u_int sent; 2557 int avail; 2558 2559 KASSERT(tp->t_dupacks == 1 || 2560 tp->t_dupacks == 2, 2561 ("%s: dupacks not 1 or 2", 2562 __func__)); 2563 if (tp->t_dupacks == 1) 2564 tp->snd_limited = 0; 2565 tp->snd_cwnd = 2566 (tp->snd_nxt - tp->snd_una) + 2567 (tp->t_dupacks - tp->snd_limited) * 2568 tp->t_maxseg; 2569 if ((thflags & TH_FIN) && 2570 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2571 /* 2572 * If its a fin we need to process 2573 * it to avoid a race where both 2574 * sides enter FIN-WAIT and send FIN|ACK 2575 * at the same time. 2576 */ 2577 break; 2578 } 2579 /* 2580 * Only call tcp_output when there 2581 * is new data available to be sent. 2582 * Otherwise we would send pure ACKs. 2583 */ 2584 SOCKBUF_LOCK(&so->so_snd); 2585 avail = so->so_snd.sb_cc - 2586 (tp->snd_nxt - tp->snd_una); 2587 SOCKBUF_UNLOCK(&so->so_snd); 2588 if (avail > 0) 2589 (void) tcp_output(tp); 2590 sent = tp->snd_max - oldsndmax; 2591 if (sent > tp->t_maxseg) { 2592 KASSERT((tp->t_dupacks == 2 && 2593 tp->snd_limited == 0) || 2594 (sent == tp->t_maxseg + 1 && 2595 tp->t_flags & TF_SENTFIN), 2596 ("%s: sent too much", 2597 __func__)); 2598 tp->snd_limited = 2; 2599 } else if (sent > 0) 2600 ++tp->snd_limited; 2601 tp->snd_cwnd = oldcwnd; 2602 goto drop; 2603 } 2604 } else 2605 tp->t_dupacks = 0; 2606 break; 2607 } 2608 2609 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2610 ("%s: th_ack <= snd_una", __func__)); 2611 2612 /* 2613 * If the congestion window was inflated to account 2614 * for the other side's cached packets, retract it. 2615 */ 2616 if (IN_FASTRECOVERY(tp->t_flags)) { 2617 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2618 if (tp->t_flags & TF_SACK_PERMIT) 2619 tcp_sack_partialack(tp, th); 2620 else 2621 tcp_newreno_partial_ack(tp, th); 2622 } else 2623 cc_post_recovery(tp, th); 2624 } 2625 tp->t_dupacks = 0; 2626 /* 2627 * If we reach this point, ACK is not a duplicate, 2628 * i.e., it ACKs something we sent. 2629 */ 2630 if (tp->t_flags & TF_NEEDSYN) { 2631 /* 2632 * T/TCP: Connection was half-synchronized, and our 2633 * SYN has been ACK'd (so connection is now fully 2634 * synchronized). Go to non-starred state, 2635 * increment snd_una for ACK of SYN, and check if 2636 * we can do window scaling. 2637 */ 2638 tp->t_flags &= ~TF_NEEDSYN; 2639 tp->snd_una++; 2640 /* Do window scaling? */ 2641 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2642 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2643 tp->rcv_scale = tp->request_r_scale; 2644 /* Send window already scaled. */ 2645 } 2646 } 2647 2648process_ACK: 2649 INP_WLOCK_ASSERT(tp->t_inpcb); 2650 2651 acked = BYTES_THIS_ACK(tp, th); 2652 TCPSTAT_INC(tcps_rcvackpack); 2653 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2654 2655 /* 2656 * If we just performed our first retransmit, and the ACK 2657 * arrives within our recovery window, then it was a mistake 2658 * to do the retransmit in the first place. Recover our 2659 * original cwnd and ssthresh, and proceed to transmit where 2660 * we left off. 2661 */ 2662 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2663 (int)(ticks - tp->t_badrxtwin) < 0) 2664 cc_cong_signal(tp, th, CC_RTO_ERR); 2665 2666 /* 2667 * If we have a timestamp reply, update smoothed 2668 * round trip time. If no timestamp is present but 2669 * transmit timer is running and timed sequence 2670 * number was acked, update smoothed round trip time. 2671 * Since we now have an rtt measurement, cancel the 2672 * timer backoff (cf., Phil Karn's retransmit alg.). 2673 * Recompute the initial retransmit timer. 2674 * 2675 * Some boxes send broken timestamp replies 2676 * during the SYN+ACK phase, ignore 2677 * timestamps of 0 or we could calculate a 2678 * huge RTT and blow up the retransmit timer. 2679 */ 2680 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2681 u_int t; 2682 2683 t = tcp_ts_getticks() - to.to_tsecr; 2684 if (!tp->t_rttlow || tp->t_rttlow > t) 2685 tp->t_rttlow = t; 2686 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2687 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2688 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2689 tp->t_rttlow = ticks - tp->t_rtttime; 2690 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2691 } 2692 2693 /* 2694 * If all outstanding data is acked, stop retransmit 2695 * timer and remember to restart (more output or persist). 2696 * If there is more data to be acked, restart retransmit 2697 * timer, using current (possibly backed-off) value. 2698 */ 2699 if (th->th_ack == tp->snd_max) { 2700 tcp_timer_activate(tp, TT_REXMT, 0); 2701 needoutput = 1; 2702 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2703 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2704 2705 /* 2706 * If no data (only SYN) was ACK'd, 2707 * skip rest of ACK processing. 2708 */ 2709 if (acked == 0) 2710 goto step6; 2711 2712 /* 2713 * Let the congestion control algorithm update congestion 2714 * control related information. This typically means increasing 2715 * the congestion window. 2716 */ 2717 cc_ack_received(tp, th, CC_ACK); 2718 2719 SOCKBUF_LOCK(&so->so_snd); 2720 if (acked > so->so_snd.sb_cc) { 2721 tp->snd_wnd -= so->so_snd.sb_cc; 2722 mfree = sbcut_locked(&so->so_snd, 2723 (int)so->so_snd.sb_cc); 2724 ourfinisacked = 1; 2725 } else { 2726 mfree = sbcut_locked(&so->so_snd, acked); 2727 tp->snd_wnd -= acked; 2728 ourfinisacked = 0; 2729 } 2730 /* NB: sowwakeup_locked() does an implicit unlock. */ 2731 sowwakeup_locked(so); 2732 m_freem(mfree); 2733 /* Detect una wraparound. */ 2734 if (!IN_RECOVERY(tp->t_flags) && 2735 SEQ_GT(tp->snd_una, tp->snd_recover) && 2736 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2737 tp->snd_recover = th->th_ack - 1; 2738 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2739 if (IN_RECOVERY(tp->t_flags) && 2740 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2741 EXIT_RECOVERY(tp->t_flags); 2742 } 2743 tp->snd_una = th->th_ack; 2744 if (tp->t_flags & TF_SACK_PERMIT) { 2745 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2746 tp->snd_recover = tp->snd_una; 2747 } 2748 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2749 tp->snd_nxt = tp->snd_una; 2750 2751 switch (tp->t_state) { 2752 2753 /* 2754 * In FIN_WAIT_1 STATE in addition to the processing 2755 * for the ESTABLISHED state if our FIN is now acknowledged 2756 * then enter FIN_WAIT_2. 2757 */ 2758 case TCPS_FIN_WAIT_1: 2759 if (ourfinisacked) { 2760 /* 2761 * If we can't receive any more 2762 * data, then closing user can proceed. 2763 * Starting the timer is contrary to the 2764 * specification, but if we don't get a FIN 2765 * we'll hang forever. 2766 * 2767 * XXXjl: 2768 * we should release the tp also, and use a 2769 * compressed state. 2770 */ 2771 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2772 soisdisconnected(so); 2773 tcp_timer_activate(tp, TT_2MSL, 2774 (tcp_fast_finwait2_recycle ? 2775 tcp_finwait2_timeout : 2776 TP_MAXIDLE(tp))); 2777 } 2778 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2779 } 2780 break; 2781 2782 /* 2783 * In CLOSING STATE in addition to the processing for 2784 * the ESTABLISHED state if the ACK acknowledges our FIN 2785 * then enter the TIME-WAIT state, otherwise ignore 2786 * the segment. 2787 */ 2788 case TCPS_CLOSING: 2789 if (ourfinisacked) { 2790 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2791 tcp_twstart(tp); 2792 INP_INFO_WUNLOCK(&V_tcbinfo); 2793 m_freem(m); 2794 return; 2795 } 2796 break; 2797 2798 /* 2799 * In LAST_ACK, we may still be waiting for data to drain 2800 * and/or to be acked, as well as for the ack of our FIN. 2801 * If our FIN is now acknowledged, delete the TCB, 2802 * enter the closed state and return. 2803 */ 2804 case TCPS_LAST_ACK: 2805 if (ourfinisacked) { 2806 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2807 tp = tcp_close(tp); 2808 goto drop; 2809 } 2810 break; 2811 } 2812 } 2813 2814step6: 2815 INP_WLOCK_ASSERT(tp->t_inpcb); 2816 2817 /* 2818 * Update window information. 2819 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2820 */ 2821 if ((thflags & TH_ACK) && 2822 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2823 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2824 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2825 /* keep track of pure window updates */ 2826 if (tlen == 0 && 2827 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2828 TCPSTAT_INC(tcps_rcvwinupd); 2829 tp->snd_wnd = tiwin; 2830 tp->snd_wl1 = th->th_seq; 2831 tp->snd_wl2 = th->th_ack; 2832 if (tp->snd_wnd > tp->max_sndwnd) 2833 tp->max_sndwnd = tp->snd_wnd; 2834 needoutput = 1; 2835 } 2836 2837 /* 2838 * Process segments with URG. 2839 */ 2840 if ((thflags & TH_URG) && th->th_urp && 2841 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2842 /* 2843 * This is a kludge, but if we receive and accept 2844 * random urgent pointers, we'll crash in 2845 * soreceive. It's hard to imagine someone 2846 * actually wanting to send this much urgent data. 2847 */ 2848 SOCKBUF_LOCK(&so->so_rcv); 2849 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2850 th->th_urp = 0; /* XXX */ 2851 thflags &= ~TH_URG; /* XXX */ 2852 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2853 goto dodata; /* XXX */ 2854 } 2855 /* 2856 * If this segment advances the known urgent pointer, 2857 * then mark the data stream. This should not happen 2858 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2859 * a FIN has been received from the remote side. 2860 * In these states we ignore the URG. 2861 * 2862 * According to RFC961 (Assigned Protocols), 2863 * the urgent pointer points to the last octet 2864 * of urgent data. We continue, however, 2865 * to consider it to indicate the first octet 2866 * of data past the urgent section as the original 2867 * spec states (in one of two places). 2868 */ 2869 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2870 tp->rcv_up = th->th_seq + th->th_urp; 2871 so->so_oobmark = so->so_rcv.sb_cc + 2872 (tp->rcv_up - tp->rcv_nxt) - 1; 2873 if (so->so_oobmark == 0) 2874 so->so_rcv.sb_state |= SBS_RCVATMARK; 2875 sohasoutofband(so); 2876 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2877 } 2878 SOCKBUF_UNLOCK(&so->so_rcv); 2879 /* 2880 * Remove out of band data so doesn't get presented to user. 2881 * This can happen independent of advancing the URG pointer, 2882 * but if two URG's are pending at once, some out-of-band 2883 * data may creep in... ick. 2884 */ 2885 if (th->th_urp <= (u_long)tlen && 2886 !(so->so_options & SO_OOBINLINE)) { 2887 /* hdr drop is delayed */ 2888 tcp_pulloutofband(so, th, m, drop_hdrlen); 2889 } 2890 } else { 2891 /* 2892 * If no out of band data is expected, 2893 * pull receive urgent pointer along 2894 * with the receive window. 2895 */ 2896 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2897 tp->rcv_up = tp->rcv_nxt; 2898 } 2899dodata: /* XXX */ 2900 INP_WLOCK_ASSERT(tp->t_inpcb); 2901 2902 /* 2903 * Process the segment text, merging it into the TCP sequencing queue, 2904 * and arranging for acknowledgment of receipt if necessary. 2905 * This process logically involves adjusting tp->rcv_wnd as data 2906 * is presented to the user (this happens in tcp_usrreq.c, 2907 * case PRU_RCVD). If a FIN has already been received on this 2908 * connection then we just ignore the text. 2909 */ 2910 if ((tlen || (thflags & TH_FIN)) && 2911 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2912 tcp_seq save_start = th->th_seq; 2913 m_adj(m, drop_hdrlen); /* delayed header drop */ 2914 /* 2915 * Insert segment which includes th into TCP reassembly queue 2916 * with control block tp. Set thflags to whether reassembly now 2917 * includes a segment with FIN. This handles the common case 2918 * inline (segment is the next to be received on an established 2919 * connection, and the queue is empty), avoiding linkage into 2920 * and removal from the queue and repetition of various 2921 * conversions. 2922 * Set DELACK for segments received in order, but ack 2923 * immediately when segments are out of order (so 2924 * fast retransmit can work). 2925 */ 2926 if (th->th_seq == tp->rcv_nxt && 2927 LIST_EMPTY(&tp->t_segq) && 2928 TCPS_HAVEESTABLISHED(tp->t_state)) { 2929 if (DELAY_ACK(tp)) 2930 tp->t_flags |= TF_DELACK; 2931 else 2932 tp->t_flags |= TF_ACKNOW; 2933 tp->rcv_nxt += tlen; 2934 thflags = th->th_flags & TH_FIN; 2935 TCPSTAT_INC(tcps_rcvpack); 2936 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2937 ND6_HINT(tp); 2938 SOCKBUF_LOCK(&so->so_rcv); 2939 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2940 m_freem(m); 2941 else 2942 sbappendstream_locked(&so->so_rcv, m); 2943 /* NB: sorwakeup_locked() does an implicit unlock. */ 2944 sorwakeup_locked(so); 2945 } else { 2946 /* 2947 * XXX: Due to the header drop above "th" is 2948 * theoretically invalid by now. Fortunately 2949 * m_adj() doesn't actually frees any mbufs 2950 * when trimming from the head. 2951 */ 2952 thflags = tcp_reass(tp, th, &tlen, m); 2953 tp->t_flags |= TF_ACKNOW; 2954 } 2955 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2956 tcp_update_sack_list(tp, save_start, save_start + tlen); 2957#if 0 2958 /* 2959 * Note the amount of data that peer has sent into 2960 * our window, in order to estimate the sender's 2961 * buffer size. 2962 * XXX: Unused. 2963 */ 2964 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2965 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2966 else 2967 len = so->so_rcv.sb_hiwat; 2968#endif 2969 } else { 2970 m_freem(m); 2971 thflags &= ~TH_FIN; 2972 } 2973 2974 /* 2975 * If FIN is received ACK the FIN and let the user know 2976 * that the connection is closing. 2977 */ 2978 if (thflags & TH_FIN) { 2979 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2980 socantrcvmore(so); 2981 /* 2982 * If connection is half-synchronized 2983 * (ie NEEDSYN flag on) then delay ACK, 2984 * so it may be piggybacked when SYN is sent. 2985 * Otherwise, since we received a FIN then no 2986 * more input can be expected, send ACK now. 2987 */ 2988 if (tp->t_flags & TF_NEEDSYN) 2989 tp->t_flags |= TF_DELACK; 2990 else 2991 tp->t_flags |= TF_ACKNOW; 2992 tp->rcv_nxt++; 2993 } 2994 switch (tp->t_state) { 2995 2996 /* 2997 * In SYN_RECEIVED and ESTABLISHED STATES 2998 * enter the CLOSE_WAIT state. 2999 */ 3000 case TCPS_SYN_RECEIVED: 3001 tp->t_starttime = ticks; 3002 /* FALLTHROUGH */ 3003 case TCPS_ESTABLISHED: 3004 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3005 break; 3006 3007 /* 3008 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3009 * enter the CLOSING state. 3010 */ 3011 case TCPS_FIN_WAIT_1: 3012 tcp_state_change(tp, TCPS_CLOSING); 3013 break; 3014 3015 /* 3016 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3017 * starting the time-wait timer, turning off the other 3018 * standard timers. 3019 */ 3020 case TCPS_FIN_WAIT_2: 3021 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 3022 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 3023 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3024 ti_locked)); 3025 3026 tcp_twstart(tp); 3027 INP_INFO_WUNLOCK(&V_tcbinfo); 3028 return; 3029 } 3030 } 3031 if (ti_locked == TI_WLOCKED) 3032 INP_INFO_WUNLOCK(&V_tcbinfo); 3033 ti_locked = TI_UNLOCKED; 3034 3035#ifdef TCPDEBUG 3036 if (so->so_options & SO_DEBUG) 3037 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3038 &tcp_savetcp, 0); 3039#endif 3040 3041 /* 3042 * Return any desired output. 3043 */ 3044 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3045 (void) tcp_output(tp); 3046 3047check_delack: 3048 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3049 __func__, ti_locked)); 3050 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3051 INP_WLOCK_ASSERT(tp->t_inpcb); 3052 3053 if (tp->t_flags & TF_DELACK) { 3054 tp->t_flags &= ~TF_DELACK; 3055 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3056 } 3057 INP_WUNLOCK(tp->t_inpcb); 3058 return; 3059 3060dropafterack: 3061 /* 3062 * Generate an ACK dropping incoming segment if it occupies 3063 * sequence space, where the ACK reflects our state. 3064 * 3065 * We can now skip the test for the RST flag since all 3066 * paths to this code happen after packets containing 3067 * RST have been dropped. 3068 * 3069 * In the SYN-RECEIVED state, don't send an ACK unless the 3070 * segment we received passes the SYN-RECEIVED ACK test. 3071 * If it fails send a RST. This breaks the loop in the 3072 * "LAND" DoS attack, and also prevents an ACK storm 3073 * between two listening ports that have been sent forged 3074 * SYN segments, each with the source address of the other. 3075 */ 3076 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3077 (SEQ_GT(tp->snd_una, th->th_ack) || 3078 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3079 rstreason = BANDLIM_RST_OPENPORT; 3080 goto dropwithreset; 3081 } 3082#ifdef TCPDEBUG 3083 if (so->so_options & SO_DEBUG) 3084 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3085 &tcp_savetcp, 0); 3086#endif 3087 if (ti_locked == TI_WLOCKED) 3088 INP_INFO_WUNLOCK(&V_tcbinfo); 3089 ti_locked = TI_UNLOCKED; 3090 3091 tp->t_flags |= TF_ACKNOW; 3092 (void) tcp_output(tp); 3093 INP_WUNLOCK(tp->t_inpcb); 3094 m_freem(m); 3095 return; 3096 3097dropwithreset: 3098 if (ti_locked == TI_WLOCKED) 3099 INP_INFO_WUNLOCK(&V_tcbinfo); 3100 ti_locked = TI_UNLOCKED; 3101 3102 if (tp != NULL) { 3103 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3104 INP_WUNLOCK(tp->t_inpcb); 3105 } else 3106 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3107 return; 3108 3109drop: 3110 if (ti_locked == TI_WLOCKED) { 3111 INP_INFO_WUNLOCK(&V_tcbinfo); 3112 ti_locked = TI_UNLOCKED; 3113 } 3114#ifdef INVARIANTS 3115 else 3116 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3117#endif 3118 3119 /* 3120 * Drop space held by incoming segment and return. 3121 */ 3122#ifdef TCPDEBUG 3123 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3124 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3125 &tcp_savetcp, 0); 3126#endif 3127 if (tp != NULL) 3128 INP_WUNLOCK(tp->t_inpcb); 3129 m_freem(m); 3130} 3131 3132/* 3133 * Issue RST and make ACK acceptable to originator of segment. 3134 * The mbuf must still include the original packet header. 3135 * tp may be NULL. 3136 */ 3137static void 3138tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3139 int tlen, int rstreason) 3140{ 3141#ifdef INET 3142 struct ip *ip; 3143#endif 3144#ifdef INET6 3145 struct ip6_hdr *ip6; 3146#endif 3147 3148 if (tp != NULL) { 3149 INP_WLOCK_ASSERT(tp->t_inpcb); 3150 } 3151 3152 /* Don't bother if destination was broadcast/multicast. */ 3153 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3154 goto drop; 3155#ifdef INET6 3156 if (mtod(m, struct ip *)->ip_v == 6) { 3157 ip6 = mtod(m, struct ip6_hdr *); 3158 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3159 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3160 goto drop; 3161 /* IPv6 anycast check is done at tcp6_input() */ 3162 } 3163#endif 3164#if defined(INET) && defined(INET6) 3165 else 3166#endif 3167#ifdef INET 3168 { 3169 ip = mtod(m, struct ip *); 3170 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3171 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3172 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3173 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3174 goto drop; 3175 } 3176#endif 3177 3178 /* Perform bandwidth limiting. */ 3179 if (badport_bandlim(rstreason) < 0) 3180 goto drop; 3181 3182 /* tcp_respond consumes the mbuf chain. */ 3183 if (th->th_flags & TH_ACK) { 3184 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3185 th->th_ack, TH_RST); 3186 } else { 3187 if (th->th_flags & TH_SYN) 3188 tlen++; 3189 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3190 (tcp_seq)0, TH_RST|TH_ACK); 3191 } 3192 return; 3193drop: 3194 m_freem(m); 3195} 3196 3197/* 3198 * Parse TCP options and place in tcpopt. 3199 */ 3200static void 3201tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3202{ 3203 int opt, optlen; 3204 3205 to->to_flags = 0; 3206 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3207 opt = cp[0]; 3208 if (opt == TCPOPT_EOL) 3209 break; 3210 if (opt == TCPOPT_NOP) 3211 optlen = 1; 3212 else { 3213 if (cnt < 2) 3214 break; 3215 optlen = cp[1]; 3216 if (optlen < 2 || optlen > cnt) 3217 break; 3218 } 3219 switch (opt) { 3220 case TCPOPT_MAXSEG: 3221 if (optlen != TCPOLEN_MAXSEG) 3222 continue; 3223 if (!(flags & TO_SYN)) 3224 continue; 3225 to->to_flags |= TOF_MSS; 3226 bcopy((char *)cp + 2, 3227 (char *)&to->to_mss, sizeof(to->to_mss)); 3228 to->to_mss = ntohs(to->to_mss); 3229 break; 3230 case TCPOPT_WINDOW: 3231 if (optlen != TCPOLEN_WINDOW) 3232 continue; 3233 if (!(flags & TO_SYN)) 3234 continue; 3235 to->to_flags |= TOF_SCALE; 3236 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3237 break; 3238 case TCPOPT_TIMESTAMP: 3239 if (optlen != TCPOLEN_TIMESTAMP) 3240 continue; 3241 to->to_flags |= TOF_TS; 3242 bcopy((char *)cp + 2, 3243 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3244 to->to_tsval = ntohl(to->to_tsval); 3245 bcopy((char *)cp + 6, 3246 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3247 to->to_tsecr = ntohl(to->to_tsecr); 3248 break; 3249#ifdef TCP_SIGNATURE 3250 /* 3251 * XXX In order to reply to a host which has set the 3252 * TCP_SIGNATURE option in its initial SYN, we have to 3253 * record the fact that the option was observed here 3254 * for the syncache code to perform the correct response. 3255 */ 3256 case TCPOPT_SIGNATURE: 3257 if (optlen != TCPOLEN_SIGNATURE) 3258 continue; 3259 to->to_flags |= TOF_SIGNATURE; 3260 to->to_signature = cp + 2; 3261 break; 3262#endif 3263 case TCPOPT_SACK_PERMITTED: 3264 if (optlen != TCPOLEN_SACK_PERMITTED) 3265 continue; 3266 if (!(flags & TO_SYN)) 3267 continue; 3268 if (!V_tcp_do_sack) 3269 continue; 3270 to->to_flags |= TOF_SACKPERM; 3271 break; 3272 case TCPOPT_SACK: 3273 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3274 continue; 3275 if (flags & TO_SYN) 3276 continue; 3277 to->to_flags |= TOF_SACK; 3278 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3279 to->to_sacks = cp + 2; 3280 TCPSTAT_INC(tcps_sack_rcv_blocks); 3281 break; 3282 default: 3283 continue; 3284 } 3285 } 3286} 3287 3288/* 3289 * Pull out of band byte out of a segment so 3290 * it doesn't appear in the user's data queue. 3291 * It is still reflected in the segment length for 3292 * sequencing purposes. 3293 */ 3294static void 3295tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3296 int off) 3297{ 3298 int cnt = off + th->th_urp - 1; 3299 3300 while (cnt >= 0) { 3301 if (m->m_len > cnt) { 3302 char *cp = mtod(m, caddr_t) + cnt; 3303 struct tcpcb *tp = sototcpcb(so); 3304 3305 INP_WLOCK_ASSERT(tp->t_inpcb); 3306 3307 tp->t_iobc = *cp; 3308 tp->t_oobflags |= TCPOOB_HAVEDATA; 3309 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3310 m->m_len--; 3311 if (m->m_flags & M_PKTHDR) 3312 m->m_pkthdr.len--; 3313 return; 3314 } 3315 cnt -= m->m_len; 3316 m = m->m_next; 3317 if (m == NULL) 3318 break; 3319 } 3320 panic("tcp_pulloutofband"); 3321} 3322 3323/* 3324 * Collect new round-trip time estimate 3325 * and update averages and current timeout. 3326 */ 3327static void 3328tcp_xmit_timer(struct tcpcb *tp, int rtt) 3329{ 3330 int delta; 3331 3332 INP_WLOCK_ASSERT(tp->t_inpcb); 3333 3334 TCPSTAT_INC(tcps_rttupdated); 3335 tp->t_rttupdated++; 3336 if (tp->t_srtt != 0) { 3337 /* 3338 * srtt is stored as fixed point with 5 bits after the 3339 * binary point (i.e., scaled by 8). The following magic 3340 * is equivalent to the smoothing algorithm in rfc793 with 3341 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3342 * point). Adjust rtt to origin 0. 3343 */ 3344 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3345 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3346 3347 if ((tp->t_srtt += delta) <= 0) 3348 tp->t_srtt = 1; 3349 3350 /* 3351 * We accumulate a smoothed rtt variance (actually, a 3352 * smoothed mean difference), then set the retransmit 3353 * timer to smoothed rtt + 4 times the smoothed variance. 3354 * rttvar is stored as fixed point with 4 bits after the 3355 * binary point (scaled by 16). The following is 3356 * equivalent to rfc793 smoothing with an alpha of .75 3357 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3358 * rfc793's wired-in beta. 3359 */ 3360 if (delta < 0) 3361 delta = -delta; 3362 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3363 if ((tp->t_rttvar += delta) <= 0) 3364 tp->t_rttvar = 1; 3365 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3366 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3367 } else { 3368 /* 3369 * No rtt measurement yet - use the unsmoothed rtt. 3370 * Set the variance to half the rtt (so our first 3371 * retransmit happens at 3*rtt). 3372 */ 3373 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3374 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3375 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3376 } 3377 tp->t_rtttime = 0; 3378 tp->t_rxtshift = 0; 3379 3380 /* 3381 * the retransmit should happen at rtt + 4 * rttvar. 3382 * Because of the way we do the smoothing, srtt and rttvar 3383 * will each average +1/2 tick of bias. When we compute 3384 * the retransmit timer, we want 1/2 tick of rounding and 3385 * 1 extra tick because of +-1/2 tick uncertainty in the 3386 * firing of the timer. The bias will give us exactly the 3387 * 1.5 tick we need. But, because the bias is 3388 * statistical, we have to test that we don't drop below 3389 * the minimum feasible timer (which is 2 ticks). 3390 */ 3391 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3392 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3393 3394 /* 3395 * We received an ack for a packet that wasn't retransmitted; 3396 * it is probably safe to discard any error indications we've 3397 * received recently. This isn't quite right, but close enough 3398 * for now (a route might have failed after we sent a segment, 3399 * and the return path might not be symmetrical). 3400 */ 3401 tp->t_softerror = 0; 3402} 3403 3404/* 3405 * Determine a reasonable value for maxseg size. 3406 * If the route is known, check route for mtu. 3407 * If none, use an mss that can be handled on the outgoing interface 3408 * without forcing IP to fragment. If no route is found, route has no mtu, 3409 * or the destination isn't local, use a default, hopefully conservative 3410 * size (usually 512 or the default IP max size, but no more than the mtu 3411 * of the interface), as we can't discover anything about intervening 3412 * gateways or networks. We also initialize the congestion/slow start 3413 * window to be a single segment if the destination isn't local. 3414 * While looking at the routing entry, we also initialize other path-dependent 3415 * parameters from pre-set or cached values in the routing entry. 3416 * 3417 * Also take into account the space needed for options that we 3418 * send regularly. Make maxseg shorter by that amount to assure 3419 * that we can send maxseg amount of data even when the options 3420 * are present. Store the upper limit of the length of options plus 3421 * data in maxopd. 3422 * 3423 * NOTE that this routine is only called when we process an incoming 3424 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3425 * settings are handled in tcp_mssopt(). 3426 */ 3427void 3428tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3429 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3430{ 3431 int mss = 0; 3432 u_long maxmtu = 0; 3433 struct inpcb *inp = tp->t_inpcb; 3434 struct hc_metrics_lite metrics; 3435 int origoffer; 3436#ifdef INET6 3437 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3438 size_t min_protoh = isipv6 ? 3439 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3440 sizeof (struct tcpiphdr); 3441#else 3442 const size_t min_protoh = sizeof(struct tcpiphdr); 3443#endif 3444 3445 INP_WLOCK_ASSERT(tp->t_inpcb); 3446 3447 if (mtuoffer != -1) { 3448 KASSERT(offer == -1, ("%s: conflict", __func__)); 3449 offer = mtuoffer - min_protoh; 3450 } 3451 origoffer = offer; 3452 3453 /* Initialize. */ 3454#ifdef INET6 3455 if (isipv6) { 3456 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3457 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3458 } 3459#endif 3460#if defined(INET) && defined(INET6) 3461 else 3462#endif 3463#ifdef INET 3464 { 3465 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3466 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3467 } 3468#endif 3469 3470 /* 3471 * No route to sender, stay with default mss and return. 3472 */ 3473 if (maxmtu == 0) { 3474 /* 3475 * In case we return early we need to initialize metrics 3476 * to a defined state as tcp_hc_get() would do for us 3477 * if there was no cache hit. 3478 */ 3479 if (metricptr != NULL) 3480 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3481 return; 3482 } 3483 3484 /* What have we got? */ 3485 switch (offer) { 3486 case 0: 3487 /* 3488 * Offer == 0 means that there was no MSS on the SYN 3489 * segment, in this case we use tcp_mssdflt as 3490 * already assigned to t_maxopd above. 3491 */ 3492 offer = tp->t_maxopd; 3493 break; 3494 3495 case -1: 3496 /* 3497 * Offer == -1 means that we didn't receive SYN yet. 3498 */ 3499 /* FALLTHROUGH */ 3500 3501 default: 3502 /* 3503 * Prevent DoS attack with too small MSS. Round up 3504 * to at least minmss. 3505 */ 3506 offer = max(offer, V_tcp_minmss); 3507 } 3508 3509 /* 3510 * rmx information is now retrieved from tcp_hostcache. 3511 */ 3512 tcp_hc_get(&inp->inp_inc, &metrics); 3513 if (metricptr != NULL) 3514 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3515 3516 /* 3517 * If there's a discovered mtu int tcp hostcache, use it 3518 * else, use the link mtu. 3519 */ 3520 if (metrics.rmx_mtu) 3521 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3522 else { 3523#ifdef INET6 3524 if (isipv6) { 3525 mss = maxmtu - min_protoh; 3526 if (!V_path_mtu_discovery && 3527 !in6_localaddr(&inp->in6p_faddr)) 3528 mss = min(mss, V_tcp_v6mssdflt); 3529 } 3530#endif 3531#if defined(INET) && defined(INET6) 3532 else 3533#endif 3534#ifdef INET 3535 { 3536 mss = maxmtu - min_protoh; 3537 if (!V_path_mtu_discovery && 3538 !in_localaddr(inp->inp_faddr)) 3539 mss = min(mss, V_tcp_mssdflt); 3540 } 3541#endif 3542 /* 3543 * XXX - The above conditional (mss = maxmtu - min_protoh) 3544 * probably violates the TCP spec. 3545 * The problem is that, since we don't know the 3546 * other end's MSS, we are supposed to use a conservative 3547 * default. But, if we do that, then MTU discovery will 3548 * never actually take place, because the conservative 3549 * default is much less than the MTUs typically seen 3550 * on the Internet today. For the moment, we'll sweep 3551 * this under the carpet. 3552 * 3553 * The conservative default might not actually be a problem 3554 * if the only case this occurs is when sending an initial 3555 * SYN with options and data to a host we've never talked 3556 * to before. Then, they will reply with an MSS value which 3557 * will get recorded and the new parameters should get 3558 * recomputed. For Further Study. 3559 */ 3560 } 3561 mss = min(mss, offer); 3562 3563 /* 3564 * Sanity check: make sure that maxopd will be large 3565 * enough to allow some data on segments even if the 3566 * all the option space is used (40bytes). Otherwise 3567 * funny things may happen in tcp_output. 3568 */ 3569 mss = max(mss, 64); 3570 3571 /* 3572 * maxopd stores the maximum length of data AND options 3573 * in a segment; maxseg is the amount of data in a normal 3574 * segment. We need to store this value (maxopd) apart 3575 * from maxseg, because now every segment carries options 3576 * and thus we normally have somewhat less data in segments. 3577 */ 3578 tp->t_maxopd = mss; 3579 3580 /* 3581 * origoffer==-1 indicates that no segments were received yet. 3582 * In this case we just guess. 3583 */ 3584 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3585 (origoffer == -1 || 3586 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3587 mss -= TCPOLEN_TSTAMP_APPA; 3588 3589 tp->t_maxseg = mss; 3590} 3591 3592void 3593tcp_mss(struct tcpcb *tp, int offer) 3594{ 3595 int mss; 3596 u_long bufsize; 3597 struct inpcb *inp; 3598 struct socket *so; 3599 struct hc_metrics_lite metrics; 3600 struct tcp_ifcap cap; 3601 3602 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3603 3604 bzero(&cap, sizeof(cap)); 3605 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3606 3607 mss = tp->t_maxseg; 3608 inp = tp->t_inpcb; 3609 3610 /* 3611 * If there's a pipesize, change the socket buffer to that size, 3612 * don't change if sb_hiwat is different than default (then it 3613 * has been changed on purpose with setsockopt). 3614 * Make the socket buffers an integral number of mss units; 3615 * if the mss is larger than the socket buffer, decrease the mss. 3616 */ 3617 so = inp->inp_socket; 3618 SOCKBUF_LOCK(&so->so_snd); 3619 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3620 bufsize = metrics.rmx_sendpipe; 3621 else 3622 bufsize = so->so_snd.sb_hiwat; 3623 if (bufsize < mss) 3624 mss = bufsize; 3625 else { 3626 bufsize = roundup(bufsize, mss); 3627 if (bufsize > sb_max) 3628 bufsize = sb_max; 3629 if (bufsize > so->so_snd.sb_hiwat) 3630 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3631 } 3632 SOCKBUF_UNLOCK(&so->so_snd); 3633 tp->t_maxseg = mss; 3634 3635 SOCKBUF_LOCK(&so->so_rcv); 3636 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3637 bufsize = metrics.rmx_recvpipe; 3638 else 3639 bufsize = so->so_rcv.sb_hiwat; 3640 if (bufsize > mss) { 3641 bufsize = roundup(bufsize, mss); 3642 if (bufsize > sb_max) 3643 bufsize = sb_max; 3644 if (bufsize > so->so_rcv.sb_hiwat) 3645 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3646 } 3647 SOCKBUF_UNLOCK(&so->so_rcv); 3648 3649 /* Check the interface for TSO capabilities. */ 3650 if (cap.ifcap & CSUM_TSO) { 3651 tp->t_flags |= TF_TSO; 3652 tp->t_tsomax = cap.tsomax; 3653 } 3654} 3655 3656/* 3657 * Determine the MSS option to send on an outgoing SYN. 3658 */ 3659int 3660tcp_mssopt(struct in_conninfo *inc) 3661{ 3662 int mss = 0; 3663 u_long maxmtu = 0; 3664 u_long thcmtu = 0; 3665 size_t min_protoh; 3666 3667 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3668 3669#ifdef INET6 3670 if (inc->inc_flags & INC_ISIPV6) { 3671 mss = V_tcp_v6mssdflt; 3672 maxmtu = tcp_maxmtu6(inc, NULL); 3673 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3674 } 3675#endif 3676#if defined(INET) && defined(INET6) 3677 else 3678#endif 3679#ifdef INET 3680 { 3681 mss = V_tcp_mssdflt; 3682 maxmtu = tcp_maxmtu(inc, NULL); 3683 min_protoh = sizeof(struct tcpiphdr); 3684 } 3685#endif 3686#if defined(INET6) || defined(INET) 3687 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3688#endif 3689 3690 if (maxmtu && thcmtu) 3691 mss = min(maxmtu, thcmtu) - min_protoh; 3692 else if (maxmtu || thcmtu) 3693 mss = max(maxmtu, thcmtu) - min_protoh; 3694 3695 return (mss); 3696} 3697 3698 3699/* 3700 * On a partial ack arrives, force the retransmission of the 3701 * next unacknowledged segment. Do not clear tp->t_dupacks. 3702 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3703 * be started again. 3704 */ 3705static void 3706tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3707{ 3708 tcp_seq onxt = tp->snd_nxt; 3709 u_long ocwnd = tp->snd_cwnd; 3710 3711 INP_WLOCK_ASSERT(tp->t_inpcb); 3712 3713 tcp_timer_activate(tp, TT_REXMT, 0); 3714 tp->t_rtttime = 0; 3715 tp->snd_nxt = th->th_ack; 3716 /* 3717 * Set snd_cwnd to one segment beyond acknowledged offset. 3718 * (tp->snd_una has not yet been updated when this function is called.) 3719 */ 3720 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3721 tp->t_flags |= TF_ACKNOW; 3722 (void) tcp_output(tp); 3723 tp->snd_cwnd = ocwnd; 3724 if (SEQ_GT(onxt, tp->snd_nxt)) 3725 tp->snd_nxt = onxt; 3726 /* 3727 * Partial window deflation. Relies on fact that tp->snd_una 3728 * not updated yet. 3729 */ 3730 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3731 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3732 else 3733 tp->snd_cwnd = 0; 3734 tp->snd_cwnd += tp->t_maxseg; 3735} 3736