1/*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52#include <sys/cdefs.h> 53__FBSDID("$FreeBSD$"); 54 55#include "opt_inet.h" 56#include "opt_inet6.h" 57#include "opt_ipsec.h" 58#include "opt_tcpdebug.h" 59 60#include <sys/param.h> 61#include <sys/arb.h> 62#include <sys/kernel.h> 63#ifdef TCP_HHOOK 64#include <sys/hhook.h> 65#endif 66#include <sys/malloc.h> 67#include <sys/mbuf.h> 68#include <sys/proc.h> /* for proc0 declaration */ 69#include <sys/protosw.h> 70#include <sys/qmath.h> 71#include <sys/sdt.h> 72#include <sys/signalvar.h> 73#include <sys/socket.h> 74#include <sys/socketvar.h> 75#include <sys/sysctl.h> 76#include <sys/syslog.h> 77#include <sys/systm.h> 78#include <sys/stats.h> 79 80#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 81 82#include <vm/uma.h> 83 84#include <net/if.h> 85#include <net/if_var.h> 86#include <net/route.h> 87#include <net/vnet.h> 88 89#define TCPSTATES /* for logging */ 90 91#include <netinet/in.h> 92#include <netinet/in_kdtrace.h> 93#include <netinet/in_pcb.h> 94#include <netinet/in_systm.h> 95#include <netinet/ip.h> 96#include <netinet/ip_icmp.h> /* required for icmp_var.h */ 97#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 98#include <netinet/ip_var.h> 99#include <netinet/ip_options.h> 100#include <netinet/ip6.h> 101#include <netinet/icmp6.h> 102#include <netinet6/in6_pcb.h> 103#include <netinet6/in6_var.h> 104#include <netinet6/ip6_var.h> 105#include <netinet6/nd6.h> 106#include <netinet/tcp.h> 107#include <netinet/tcp_fsm.h> 108#include <netinet/tcp_log_buf.h> 109#include <netinet/tcp_seq.h> 110#include <netinet/tcp_timer.h> 111#include <netinet/tcp_var.h> 112#include <netinet6/tcp6_var.h> 113#include <netinet/tcpip.h> 114#include <netinet/cc/cc.h> 115#include <netinet/tcp_fastopen.h> 116#ifdef TCPPCAP 117#include <netinet/tcp_pcap.h> 118#endif 119#include <netinet/tcp_syncache.h> 120#ifdef TCPDEBUG 121#include <netinet/tcp_debug.h> 122#endif /* TCPDEBUG */ 123#ifdef TCP_OFFLOAD 124#include <netinet/tcp_offload.h> 125#endif 126#include <netinet/udp.h> 127 128#include <netipsec/ipsec_support.h> 129 130#include <machine/in_cksum.h> 131 132#include <security/mac/mac_framework.h> 133 134const int tcprexmtthresh = 3; 135 136VNET_DEFINE(int, tcp_log_in_vain) = 0; 137SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 138 &VNET_NAME(tcp_log_in_vain), 0, 139 "Log all incoming TCP segments to closed ports"); 140 141VNET_DEFINE(int, blackhole) = 0; 142#define V_blackhole VNET(blackhole) 143SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 144 &VNET_NAME(blackhole), 0, 145 "Do not send RST on segments to closed ports"); 146 147VNET_DEFINE(int, tcp_delack_enabled) = 1; 148SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 149 &VNET_NAME(tcp_delack_enabled), 0, 150 "Delay ACK to try and piggyback it onto a data packet"); 151 152VNET_DEFINE(int, drop_synfin) = 0; 153SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 154 &VNET_NAME(drop_synfin), 0, 155 "Drop TCP packets with SYN+FIN set"); 156 157VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 158SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 159 &VNET_NAME(tcp_do_prr_conservative), 0, 160 "Do conservative Proportional Rate Reduction"); 161 162VNET_DEFINE(int, tcp_do_prr) = 1; 163SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_prr), 1, 165 "Enable Proportional Rate Reduction per RFC 6937"); 166 167VNET_DEFINE(int, tcp_do_newcwv) = 0; 168SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_newcwv), 0, 170 "Enable New Congestion Window Validation per RFC7661"); 171 172VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 173SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 175 "Use calculated pipe/in-flight bytes per RFC 6675"); 176 177VNET_DEFINE(int, tcp_do_rfc3042) = 1; 178SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 179 &VNET_NAME(tcp_do_rfc3042), 0, 180 "Enable RFC 3042 (Limited Transmit)"); 181 182VNET_DEFINE(int, tcp_do_rfc3390) = 1; 183SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 184 &VNET_NAME(tcp_do_rfc3390), 0, 185 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 186 187VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 188SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 189 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 190 "Slow-start flight size (initial congestion window) in number of segments"); 191 192VNET_DEFINE(int, tcp_do_rfc3465) = 1; 193SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 194 &VNET_NAME(tcp_do_rfc3465), 0, 195 "Enable RFC 3465 (Appropriate Byte Counting)"); 196 197VNET_DEFINE(int, tcp_abc_l_var) = 2; 198SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_abc_l_var), 2, 200 "Cap the max cwnd increment during slow-start to this number of segments"); 201 202static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 203 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 204 "TCP ECN"); 205 206VNET_DEFINE(int, tcp_do_ecn) = 2; 207SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_do_ecn), 0, 209 "TCP ECN support"); 210 211VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 212SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(tcp_ecn_maxretries), 0, 214 "Max retries before giving up on ECN"); 215 216VNET_DEFINE(int, tcp_insecure_syn) = 0; 217SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(tcp_insecure_syn), 0, 219 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 220 221VNET_DEFINE(int, tcp_insecure_rst) = 0; 222SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_insecure_rst), 0, 224 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 225 226VNET_DEFINE(int, tcp_recvspace) = 1024*64; 227#define V_tcp_recvspace VNET(tcp_recvspace) 228SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 229 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 230 231VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 232SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 233 &VNET_NAME(tcp_do_autorcvbuf), 0, 234 "Enable automatic receive buffer sizing"); 235 236VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 237SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 238 &VNET_NAME(tcp_autorcvbuf_max), 0, 239 "Max size of automatic receive buffer"); 240 241VNET_DEFINE(struct inpcbhead, tcb); 242#define tcb6 tcb /* for KAME src sync over BSD*'s */ 243VNET_DEFINE(struct inpcbinfo, tcbinfo); 244 245/* 246 * TCP statistics are stored in an array of counter(9)s, which size matches 247 * size of struct tcpstat. TCP running connection count is a regular array. 248 */ 249VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 250SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 253SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 254 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 255 "TCP connection counts by TCP state"); 256 257static void 258tcp_vnet_init(const void *unused) 259{ 260 261 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 262 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 263} 264VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 265 tcp_vnet_init, NULL); 266 267#ifdef VIMAGE 268static void 269tcp_vnet_uninit(const void *unused) 270{ 271 272 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 273 VNET_PCPUSTAT_FREE(tcpstat); 274} 275VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 276 tcp_vnet_uninit, NULL); 277#endif /* VIMAGE */ 278 279/* 280 * Kernel module interface for updating tcpstat. The first argument is an index 281 * into tcpstat treated as an array. 282 */ 283void 284kmod_tcpstat_add(int statnum, int val) 285{ 286 287 counter_u64_add(VNET(tcpstat)[statnum], val); 288} 289 290#ifdef TCP_HHOOK 291/* 292 * Wrapper for the TCP established input helper hook. 293 */ 294void 295hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 296{ 297 struct tcp_hhook_data hhook_data; 298 299 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 300 hhook_data.tp = tp; 301 hhook_data.th = th; 302 hhook_data.to = to; 303 304 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 305 tp->osd); 306 } 307} 308#endif 309 310/* 311 * CC wrapper hook functions 312 */ 313void 314cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 315 uint16_t type) 316{ 317#ifdef STATS 318 int32_t gput; 319#endif 320 321 INP_WLOCK_ASSERT(tp->t_inpcb); 322 323 tp->ccv->nsegs = nsegs; 324 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 325 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 326 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 327 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 328 tp->ccv->flags |= CCF_CWND_LIMITED; 329 else 330 tp->ccv->flags &= ~CCF_CWND_LIMITED; 331 332 if (type == CC_ACK) { 333#ifdef STATS 334 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 335 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 336 if (!IN_RECOVERY(tp->t_flags)) 337 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 338 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 339 if ((tp->t_flags & TF_GPUTINPROG) && 340 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 341 /* 342 * Compute goodput in bits per millisecond. 343 */ 344 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) / 345 max(1, tcp_ts_getticks() - tp->gput_ts); 346 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 347 gput); 348 /* 349 * XXXLAS: This is a temporary hack, and should be 350 * chained off VOI_TCP_GPUT when stats(9) grows an API 351 * to deal with chained VOIs. 352 */ 353 if (tp->t_stats_gput_prev > 0) 354 stats_voi_update_abs_s32(tp->t_stats, 355 VOI_TCP_GPUT_ND, 356 ((gput - tp->t_stats_gput_prev) * 100) / 357 tp->t_stats_gput_prev); 358 tp->t_flags &= ~TF_GPUTINPROG; 359 tp->t_stats_gput_prev = gput; 360 } 361#endif /* STATS */ 362 if (tp->snd_cwnd > tp->snd_ssthresh) { 363 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 364 if (tp->t_bytes_acked >= tp->snd_cwnd) { 365 tp->t_bytes_acked -= tp->snd_cwnd; 366 tp->ccv->flags |= CCF_ABC_SENTAWND; 367 } 368 } else { 369 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 370 tp->t_bytes_acked = 0; 371 } 372 } 373 374 if (CC_ALGO(tp)->ack_received != NULL) { 375 /* XXXLAS: Find a way to live without this */ 376 tp->ccv->curack = th->th_ack; 377 CC_ALGO(tp)->ack_received(tp->ccv, type); 378 } 379#ifdef STATS 380 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 381#endif 382} 383 384void 385cc_conn_init(struct tcpcb *tp) 386{ 387 struct hc_metrics_lite metrics; 388 struct inpcb *inp = tp->t_inpcb; 389 u_int maxseg; 390 int rtt; 391 392 INP_WLOCK_ASSERT(tp->t_inpcb); 393 394 tcp_hc_get(&inp->inp_inc, &metrics); 395 maxseg = tcp_maxseg(tp); 396 397 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 398 tp->t_srtt = rtt; 399 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 400 TCPSTAT_INC(tcps_usedrtt); 401 if (metrics.rmx_rttvar) { 402 tp->t_rttvar = metrics.rmx_rttvar; 403 TCPSTAT_INC(tcps_usedrttvar); 404 } else { 405 /* default variation is +- 1 rtt */ 406 tp->t_rttvar = 407 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 408 } 409 TCPT_RANGESET(tp->t_rxtcur, 410 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 411 tp->t_rttmin, TCPTV_REXMTMAX); 412 } 413 if (metrics.rmx_ssthresh) { 414 /* 415 * There's some sort of gateway or interface 416 * buffer limit on the path. Use this to set 417 * the slow start threshold, but set the 418 * threshold to no less than 2*mss. 419 */ 420 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 421 TCPSTAT_INC(tcps_usedssthresh); 422 } 423 424 /* 425 * Set the initial slow-start flight size. 426 * 427 * If a SYN or SYN/ACK was lost and retransmitted, we have to 428 * reduce the initial CWND to one segment as congestion is likely 429 * requiring us to be cautious. 430 */ 431 if (tp->snd_cwnd == 1) 432 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 433 else 434 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 435 436 if (CC_ALGO(tp)->conn_init != NULL) 437 CC_ALGO(tp)->conn_init(tp->ccv); 438} 439 440void inline 441cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 442{ 443 INP_WLOCK_ASSERT(tp->t_inpcb); 444 445#ifdef STATS 446 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 447#endif 448 449 switch(type) { 450 case CC_NDUPACK: 451 if (!IN_FASTRECOVERY(tp->t_flags)) { 452 tp->snd_recover = tp->snd_max; 453 if (tp->t_flags2 & TF2_ECN_PERMIT) 454 tp->t_flags2 |= TF2_ECN_SND_CWR; 455 } 456 break; 457 case CC_ECN: 458 if (!IN_CONGRECOVERY(tp->t_flags) || 459 /* 460 * Allow ECN reaction on ACK to CWR, if 461 * that data segment was also CE marked. 462 */ 463 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 464 EXIT_CONGRECOVERY(tp->t_flags); 465 TCPSTAT_INC(tcps_ecn_rcwnd); 466 tp->snd_recover = tp->snd_max + 1; 467 if (tp->t_flags2 & TF2_ECN_PERMIT) 468 tp->t_flags2 |= TF2_ECN_SND_CWR; 469 } 470 break; 471 case CC_RTO: 472 tp->t_dupacks = 0; 473 tp->t_bytes_acked = 0; 474 EXIT_RECOVERY(tp->t_flags); 475 if (tp->t_flags2 & TF2_ECN_PERMIT) 476 tp->t_flags2 |= TF2_ECN_SND_CWR; 477 break; 478 case CC_RTO_ERR: 479 TCPSTAT_INC(tcps_sndrexmitbad); 480 /* RTO was unnecessary, so reset everything. */ 481 tp->snd_cwnd = tp->snd_cwnd_prev; 482 tp->snd_ssthresh = tp->snd_ssthresh_prev; 483 tp->snd_recover = tp->snd_recover_prev; 484 if (tp->t_flags & TF_WASFRECOVERY) 485 ENTER_FASTRECOVERY(tp->t_flags); 486 if (tp->t_flags & TF_WASCRECOVERY) 487 ENTER_CONGRECOVERY(tp->t_flags); 488 tp->snd_nxt = tp->snd_max; 489 tp->t_flags &= ~TF_PREVVALID; 490 tp->t_badrxtwin = 0; 491 break; 492 } 493 494 if (CC_ALGO(tp)->cong_signal != NULL) { 495 if (th != NULL) 496 tp->ccv->curack = th->th_ack; 497 CC_ALGO(tp)->cong_signal(tp->ccv, type); 498 } 499} 500 501void inline 502cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 503{ 504 INP_WLOCK_ASSERT(tp->t_inpcb); 505 506 /* XXXLAS: KASSERT that we're in recovery? */ 507 508 if (CC_ALGO(tp)->post_recovery != NULL) { 509 tp->ccv->curack = th->th_ack; 510 CC_ALGO(tp)->post_recovery(tp->ccv); 511 } 512 /* XXXLAS: EXIT_RECOVERY ? */ 513 tp->t_bytes_acked = 0; 514 tp->sackhint.delivered_data = 0; 515 tp->sackhint.prr_out = 0; 516} 517 518/* 519 * Indicate whether this ack should be delayed. We can delay the ack if 520 * following conditions are met: 521 * - There is no delayed ack timer in progress. 522 * - Our last ack wasn't a 0-sized window. We never want to delay 523 * the ack that opens up a 0-sized window. 524 * - LRO wasn't used for this segment. We make sure by checking that the 525 * segment size is not larger than the MSS. 526 */ 527#define DELAY_ACK(tp, tlen) \ 528 ((!tcp_timer_active(tp, TT_DELACK) && \ 529 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 530 (tlen <= tp->t_maxseg) && \ 531 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 532 533void inline 534cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 535{ 536 INP_WLOCK_ASSERT(tp->t_inpcb); 537 538 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 539 switch (iptos & IPTOS_ECN_MASK) { 540 case IPTOS_ECN_CE: 541 tp->ccv->flags |= CCF_IPHDR_CE; 542 break; 543 case IPTOS_ECN_ECT0: 544 /* FALLTHROUGH */ 545 case IPTOS_ECN_ECT1: 546 /* FALLTHROUGH */ 547 case IPTOS_ECN_NOTECT: 548 tp->ccv->flags &= ~CCF_IPHDR_CE; 549 break; 550 } 551 552 if (flags & TH_CWR) 553 tp->ccv->flags |= CCF_TCPHDR_CWR; 554 else 555 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 556 557 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 558 559 if (tp->ccv->flags & CCF_ACKNOW) { 560 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 561 tp->t_flags |= TF_ACKNOW; 562 } 563 } 564} 565 566void inline 567cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 568{ 569 cc_ecnpkt_handler_flags(tp, th->th_flags, iptos); 570} 571 572/* 573 * TCP input handling is split into multiple parts: 574 * tcp6_input is a thin wrapper around tcp_input for the extended 575 * ip6_protox[] call format in ip6_input 576 * tcp_input handles primary segment validation, inpcb lookup and 577 * SYN processing on listen sockets 578 * tcp_do_segment processes the ACK and text of the segment for 579 * establishing, established and closing connections 580 */ 581#ifdef INET6 582int 583tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 584{ 585 struct mbuf *m; 586 struct in6_ifaddr *ia6; 587 struct ip6_hdr *ip6; 588 589 m = *mp; 590 if (m->m_len < *offp + sizeof(struct tcphdr)) { 591 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 592 if (m == NULL) { 593 *mp = m; 594 TCPSTAT_INC(tcps_rcvshort); 595 return (IPPROTO_DONE); 596 } 597 } 598 599 /* 600 * draft-itojun-ipv6-tcp-to-anycast 601 * better place to put this in? 602 */ 603 ip6 = mtod(m, struct ip6_hdr *); 604 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 605 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 606 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 607 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 608 *mp = NULL; 609 return (IPPROTO_DONE); 610 } 611 612 *mp = m; 613 return (tcp_input_with_port(mp, offp, proto, port)); 614} 615 616int 617tcp6_input(struct mbuf **mp, int *offp, int proto) 618{ 619 620 return(tcp6_input_with_port(mp, offp, proto, 0)); 621} 622#endif /* INET6 */ 623 624int 625tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 626{ 627 struct mbuf *m = *mp; 628 struct tcphdr *th = NULL; 629 struct ip *ip = NULL; 630 struct inpcb *inp = NULL; 631 struct tcpcb *tp = NULL; 632 struct socket *so = NULL; 633 u_char *optp = NULL; 634 int off0; 635 int optlen = 0; 636#ifdef INET 637 int len; 638 uint8_t ipttl; 639#endif 640 int tlen = 0, off; 641 int drop_hdrlen; 642 int thflags; 643 int rstreason = 0; /* For badport_bandlim accounting purposes */ 644 uint8_t iptos; 645 struct m_tag *fwd_tag = NULL; 646#ifdef INET6 647 struct ip6_hdr *ip6 = NULL; 648 int isipv6; 649#else 650 const void *ip6 = NULL; 651#endif /* INET6 */ 652 struct tcpopt to; /* options in this segment */ 653 char *s = NULL; /* address and port logging */ 654#ifdef TCPDEBUG 655 /* 656 * The size of tcp_saveipgen must be the size of the max ip header, 657 * now IPv6. 658 */ 659 u_char tcp_saveipgen[IP6_HDR_LEN]; 660 struct tcphdr tcp_savetcp; 661 short ostate = 0; 662#endif 663 664 NET_EPOCH_ASSERT(); 665 666#ifdef INET6 667 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 668#endif 669 670 off0 = *offp; 671 m = *mp; 672 *mp = NULL; 673 to.to_flags = 0; 674 TCPSTAT_INC(tcps_rcvtotal); 675 676#ifdef INET6 677 if (isipv6) { 678 ip6 = mtod(m, struct ip6_hdr *); 679 th = (struct tcphdr *)((caddr_t)ip6 + off0); 680 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 681 if (port) 682 goto skip6_csum; 683 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 684 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 685 th->th_sum = m->m_pkthdr.csum_data; 686 else 687 th->th_sum = in6_cksum_pseudo(ip6, tlen, 688 IPPROTO_TCP, m->m_pkthdr.csum_data); 689 th->th_sum ^= 0xffff; 690 } else 691 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 692 if (th->th_sum) { 693 TCPSTAT_INC(tcps_rcvbadsum); 694 goto drop; 695 } 696 skip6_csum: 697 /* 698 * Be proactive about unspecified IPv6 address in source. 699 * As we use all-zero to indicate unbounded/unconnected pcb, 700 * unspecified IPv6 address can be used to confuse us. 701 * 702 * Note that packets with unspecified IPv6 destination is 703 * already dropped in ip6_input. 704 */ 705 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 706 /* XXX stat */ 707 goto drop; 708 } 709 iptos = IPV6_TRAFFIC_CLASS(ip6); 710 } 711#endif 712#if defined(INET) && defined(INET6) 713 else 714#endif 715#ifdef INET 716 { 717 /* 718 * Get IP and TCP header together in first mbuf. 719 * Note: IP leaves IP header in first mbuf. 720 */ 721 if (off0 > sizeof (struct ip)) { 722 ip_stripoptions(m); 723 off0 = sizeof(struct ip); 724 } 725 if (m->m_len < sizeof (struct tcpiphdr)) { 726 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 727 == NULL) { 728 TCPSTAT_INC(tcps_rcvshort); 729 return (IPPROTO_DONE); 730 } 731 } 732 ip = mtod(m, struct ip *); 733 th = (struct tcphdr *)((caddr_t)ip + off0); 734 tlen = ntohs(ip->ip_len) - off0; 735 736 iptos = ip->ip_tos; 737 if (port) 738 goto skip_csum; 739 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 740 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 741 th->th_sum = m->m_pkthdr.csum_data; 742 else 743 th->th_sum = in_pseudo(ip->ip_src.s_addr, 744 ip->ip_dst.s_addr, 745 htonl(m->m_pkthdr.csum_data + tlen + 746 IPPROTO_TCP)); 747 th->th_sum ^= 0xffff; 748 } else { 749 struct ipovly *ipov = (struct ipovly *)ip; 750 751 /* 752 * Checksum extended TCP header and data. 753 */ 754 len = off0 + tlen; 755 ipttl = ip->ip_ttl; 756 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 757 ipov->ih_len = htons(tlen); 758 th->th_sum = in_cksum(m, len); 759 /* Reset length for SDT probes. */ 760 ip->ip_len = htons(len); 761 /* Reset TOS bits */ 762 ip->ip_tos = iptos; 763 /* Re-initialization for later version check */ 764 ip->ip_ttl = ipttl; 765 ip->ip_v = IPVERSION; 766 ip->ip_hl = off0 >> 2; 767 } 768 skip_csum: 769 if (th->th_sum && (port == 0)) { 770 TCPSTAT_INC(tcps_rcvbadsum); 771 goto drop; 772 } 773 } 774#endif /* INET */ 775 776 /* 777 * Check that TCP offset makes sense, 778 * pull out TCP options and adjust length. XXX 779 */ 780 off = th->th_off << 2; 781 if (off < sizeof (struct tcphdr) || off > tlen) { 782 TCPSTAT_INC(tcps_rcvbadoff); 783 goto drop; 784 } 785 tlen -= off; /* tlen is used instead of ti->ti_len */ 786 if (off > sizeof (struct tcphdr)) { 787#ifdef INET6 788 if (isipv6) { 789 if (m->m_len < off0 + off) { 790 m = m_pullup(m, off0 + off); 791 if (m == NULL) { 792 TCPSTAT_INC(tcps_rcvshort); 793 return (IPPROTO_DONE); 794 } 795 } 796 ip6 = mtod(m, struct ip6_hdr *); 797 th = (struct tcphdr *)((caddr_t)ip6 + off0); 798 } 799#endif 800#if defined(INET) && defined(INET6) 801 else 802#endif 803#ifdef INET 804 { 805 if (m->m_len < sizeof(struct ip) + off) { 806 if ((m = m_pullup(m, sizeof (struct ip) + off)) 807 == NULL) { 808 TCPSTAT_INC(tcps_rcvshort); 809 return (IPPROTO_DONE); 810 } 811 ip = mtod(m, struct ip *); 812 th = (struct tcphdr *)((caddr_t)ip + off0); 813 } 814 } 815#endif 816 optlen = off - sizeof (struct tcphdr); 817 optp = (u_char *)(th + 1); 818 } 819 thflags = th->th_flags; 820 821 /* 822 * Convert TCP protocol specific fields to host format. 823 */ 824 tcp_fields_to_host(th); 825 826 /* 827 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 828 */ 829 drop_hdrlen = off0 + off; 830 831 /* 832 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 833 */ 834 if ( 835#ifdef INET6 836 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 837#ifdef INET 838 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 839#endif 840#endif 841#if defined(INET) && !defined(INET6) 842 (m->m_flags & M_IP_NEXTHOP) 843#endif 844 ) 845 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 846 847findpcb: 848#ifdef INET6 849 if (isipv6 && fwd_tag != NULL) { 850 struct sockaddr_in6 *next_hop6; 851 852 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 853 /* 854 * Transparently forwarded. Pretend to be the destination. 855 * Already got one like this? 856 */ 857 inp = in6_pcblookup_mbuf(&V_tcbinfo, 858 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 859 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 860 if (!inp) { 861 /* 862 * It's new. Try to find the ambushing socket. 863 * Because we've rewritten the destination address, 864 * any hardware-generated hash is ignored. 865 */ 866 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 867 th->th_sport, &next_hop6->sin6_addr, 868 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 869 th->th_dport, INPLOOKUP_WILDCARD | 870 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 871 } 872 } else if (isipv6) { 873 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 874 th->th_sport, &ip6->ip6_dst, th->th_dport, 875 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 876 m->m_pkthdr.rcvif, m); 877 } 878#endif /* INET6 */ 879#if defined(INET6) && defined(INET) 880 else 881#endif 882#ifdef INET 883 if (fwd_tag != NULL) { 884 struct sockaddr_in *next_hop; 885 886 next_hop = (struct sockaddr_in *)(fwd_tag+1); 887 /* 888 * Transparently forwarded. Pretend to be the destination. 889 * already got one like this? 890 */ 891 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 892 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 893 m->m_pkthdr.rcvif, m); 894 if (!inp) { 895 /* 896 * It's new. Try to find the ambushing socket. 897 * Because we've rewritten the destination address, 898 * any hardware-generated hash is ignored. 899 */ 900 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 901 th->th_sport, next_hop->sin_addr, 902 next_hop->sin_port ? ntohs(next_hop->sin_port) : 903 th->th_dport, INPLOOKUP_WILDCARD | 904 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 905 } 906 } else 907 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 908 th->th_sport, ip->ip_dst, th->th_dport, 909 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 910 m->m_pkthdr.rcvif, m); 911#endif /* INET */ 912 913 /* 914 * If the INPCB does not exist then all data in the incoming 915 * segment is discarded and an appropriate RST is sent back. 916 * XXX MRT Send RST using which routing table? 917 */ 918 if (inp == NULL) { 919 /* 920 * Log communication attempts to ports that are not 921 * in use. 922 */ 923 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 924 V_tcp_log_in_vain == 2) { 925 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 926 log(LOG_INFO, "%s; %s: Connection attempt " 927 "to closed port\n", s, __func__); 928 } 929 /* 930 * When blackholing do not respond with a RST but 931 * completely ignore the segment and drop it. 932 */ 933 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 934 V_blackhole == 2) 935 goto dropunlock; 936 937 rstreason = BANDLIM_RST_CLOSEDPORT; 938 goto dropwithreset; 939 } 940 INP_WLOCK_ASSERT(inp); 941 /* 942 * While waiting for inp lock during the lookup, another thread 943 * can have dropped the inpcb, in which case we need to loop back 944 * and try to find a new inpcb to deliver to. 945 */ 946 if (inp->inp_flags & INP_DROPPED) { 947 INP_WUNLOCK(inp); 948 inp = NULL; 949 goto findpcb; 950 } 951 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 952 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 953 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) { 954 inp->inp_flowid = m->m_pkthdr.flowid; 955 inp->inp_flowtype = M_HASHTYPE_GET(m); 956 } 957#if defined(IPSEC) || defined(IPSEC_SUPPORT) 958#ifdef INET6 959 if (isipv6 && IPSEC_ENABLED(ipv6) && 960 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 961 goto dropunlock; 962 } 963#ifdef INET 964 else 965#endif 966#endif /* INET6 */ 967#ifdef INET 968 if (IPSEC_ENABLED(ipv4) && 969 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 970 goto dropunlock; 971 } 972#endif /* INET */ 973#endif /* IPSEC */ 974 975 /* 976 * Check the minimum TTL for socket. 977 */ 978 if (inp->inp_ip_minttl != 0) { 979#ifdef INET6 980 if (isipv6) { 981 if (inp->inp_ip_minttl > ip6->ip6_hlim) 982 goto dropunlock; 983 } else 984#endif 985 if (inp->inp_ip_minttl > ip->ip_ttl) 986 goto dropunlock; 987 } 988 989 /* 990 * A previous connection in TIMEWAIT state is supposed to catch stray 991 * or duplicate segments arriving late. If this segment was a 992 * legitimate new connection attempt, the old INPCB gets removed and 993 * we can try again to find a listening socket. 994 * 995 * At this point, due to earlier optimism, we may hold only an inpcb 996 * lock, and not the inpcbinfo write lock. If so, we need to try to 997 * acquire it, or if that fails, acquire a reference on the inpcb, 998 * drop all locks, acquire a global write lock, and then re-acquire 999 * the inpcb lock. We may at that point discover that another thread 1000 * has tried to free the inpcb, in which case we need to loop back 1001 * and try to find a new inpcb to deliver to. 1002 * 1003 * XXXRW: It may be time to rethink timewait locking. 1004 */ 1005 if (inp->inp_flags & INP_TIMEWAIT) { 1006 tcp_dooptions(&to, optp, optlen, 1007 (thflags & TH_SYN) ? TO_SYN : 0); 1008 /* 1009 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1010 */ 1011 if (tcp_twcheck(inp, &to, th, m, tlen)) 1012 goto findpcb; 1013 return (IPPROTO_DONE); 1014 } 1015 /* 1016 * The TCPCB may no longer exist if the connection is winding 1017 * down or it is in the CLOSED state. Either way we drop the 1018 * segment and send an appropriate response. 1019 */ 1020 tp = intotcpcb(inp); 1021 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1022 rstreason = BANDLIM_RST_CLOSEDPORT; 1023 goto dropwithreset; 1024 } 1025 1026 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1027 rstreason = BANDLIM_RST_CLOSEDPORT; 1028 goto dropwithreset; 1029 } 1030 1031#ifdef TCP_OFFLOAD 1032 if (tp->t_flags & TF_TOE) { 1033 tcp_offload_input(tp, m); 1034 m = NULL; /* consumed by the TOE driver */ 1035 goto dropunlock; 1036 } 1037#endif 1038 1039#ifdef MAC 1040 INP_WLOCK_ASSERT(inp); 1041 if (mac_inpcb_check_deliver(inp, m)) 1042 goto dropunlock; 1043#endif 1044 so = inp->inp_socket; 1045 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1046#ifdef TCPDEBUG 1047 if (so->so_options & SO_DEBUG) { 1048 ostate = tp->t_state; 1049#ifdef INET6 1050 if (isipv6) { 1051 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1052 } else 1053#endif 1054 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1055 tcp_savetcp = *th; 1056 } 1057#endif /* TCPDEBUG */ 1058 /* 1059 * When the socket is accepting connections (the INPCB is in LISTEN 1060 * state) we look into the SYN cache if this is a new connection 1061 * attempt or the completion of a previous one. 1062 */ 1063 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1064 ("%s: so accepting but tp %p not listening", __func__, tp)); 1065 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1066 struct in_conninfo inc; 1067 1068 bzero(&inc, sizeof(inc)); 1069#ifdef INET6 1070 if (isipv6) { 1071 inc.inc_flags |= INC_ISIPV6; 1072 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1073 inc.inc_flags |= INC_IPV6MINMTU; 1074 inc.inc6_faddr = ip6->ip6_src; 1075 inc.inc6_laddr = ip6->ip6_dst; 1076 } else 1077#endif 1078 { 1079 inc.inc_faddr = ip->ip_src; 1080 inc.inc_laddr = ip->ip_dst; 1081 } 1082 inc.inc_fport = th->th_sport; 1083 inc.inc_lport = th->th_dport; 1084 inc.inc_fibnum = so->so_fibnum; 1085 1086 /* 1087 * Check for an existing connection attempt in syncache if 1088 * the flag is only ACK. A successful lookup creates a new 1089 * socket appended to the listen queue in SYN_RECEIVED state. 1090 */ 1091 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1092 /* 1093 * Parse the TCP options here because 1094 * syncookies need access to the reflected 1095 * timestamp. 1096 */ 1097 tcp_dooptions(&to, optp, optlen, 0); 1098 /* 1099 * NB: syncache_expand() doesn't unlock 1100 * inp and tcpinfo locks. 1101 */ 1102 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1103 if (rstreason < 0) { 1104 /* 1105 * A failing TCP MD5 signature comparison 1106 * must result in the segment being dropped 1107 * and must not produce any response back 1108 * to the sender. 1109 */ 1110 goto dropunlock; 1111 } else if (rstreason == 0) { 1112 /* 1113 * No syncache entry or ACK was not 1114 * for our SYN/ACK. Send a RST. 1115 * NB: syncache did its own logging 1116 * of the failure cause. 1117 */ 1118 rstreason = BANDLIM_RST_OPENPORT; 1119 goto dropwithreset; 1120 } 1121tfo_socket_result: 1122 if (so == NULL) { 1123 /* 1124 * We completed the 3-way handshake 1125 * but could not allocate a socket 1126 * either due to memory shortage, 1127 * listen queue length limits or 1128 * global socket limits. Send RST 1129 * or wait and have the remote end 1130 * retransmit the ACK for another 1131 * try. 1132 */ 1133 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1134 log(LOG_DEBUG, "%s; %s: Listen socket: " 1135 "Socket allocation failed due to " 1136 "limits or memory shortage, %s\n", 1137 s, __func__, 1138 V_tcp_sc_rst_sock_fail ? 1139 "sending RST" : "try again"); 1140 if (V_tcp_sc_rst_sock_fail) { 1141 rstreason = BANDLIM_UNLIMITED; 1142 goto dropwithreset; 1143 } else 1144 goto dropunlock; 1145 } 1146 /* 1147 * Socket is created in state SYN_RECEIVED. 1148 * Unlock the listen socket, lock the newly 1149 * created socket and update the tp variable. 1150 */ 1151 INP_WUNLOCK(inp); /* listen socket */ 1152 inp = sotoinpcb(so); 1153 /* 1154 * New connection inpcb is already locked by 1155 * syncache_expand(). 1156 */ 1157 INP_WLOCK_ASSERT(inp); 1158 tp = intotcpcb(inp); 1159 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1160 ("%s: ", __func__)); 1161 /* 1162 * Process the segment and the data it 1163 * contains. tcp_do_segment() consumes 1164 * the mbuf chain and unlocks the inpcb. 1165 */ 1166 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1167 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1168 iptos); 1169 return (IPPROTO_DONE); 1170 } 1171 /* 1172 * Segment flag validation for new connection attempts: 1173 * 1174 * Our (SYN|ACK) response was rejected. 1175 * Check with syncache and remove entry to prevent 1176 * retransmits. 1177 * 1178 * NB: syncache_chkrst does its own logging of failure 1179 * causes. 1180 */ 1181 if (thflags & TH_RST) { 1182 syncache_chkrst(&inc, th, m, port); 1183 goto dropunlock; 1184 } 1185 /* 1186 * We can't do anything without SYN. 1187 */ 1188 if ((thflags & TH_SYN) == 0) { 1189 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1190 log(LOG_DEBUG, "%s; %s: Listen socket: " 1191 "SYN is missing, segment ignored\n", 1192 s, __func__); 1193 TCPSTAT_INC(tcps_badsyn); 1194 goto dropunlock; 1195 } 1196 /* 1197 * (SYN|ACK) is bogus on a listen socket. 1198 */ 1199 if (thflags & TH_ACK) { 1200 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1201 log(LOG_DEBUG, "%s; %s: Listen socket: " 1202 "SYN|ACK invalid, segment rejected\n", 1203 s, __func__); 1204 syncache_badack(&inc, port); /* XXX: Not needed! */ 1205 TCPSTAT_INC(tcps_badsyn); 1206 rstreason = BANDLIM_RST_OPENPORT; 1207 goto dropwithreset; 1208 } 1209 /* 1210 * If the drop_synfin option is enabled, drop all 1211 * segments with both the SYN and FIN bits set. 1212 * This prevents e.g. nmap from identifying the 1213 * TCP/IP stack. 1214 * XXX: Poor reasoning. nmap has other methods 1215 * and is constantly refining its stack detection 1216 * strategies. 1217 * XXX: This is a violation of the TCP specification 1218 * and was used by RFC1644. 1219 */ 1220 if ((thflags & TH_FIN) && V_drop_synfin) { 1221 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1222 log(LOG_DEBUG, "%s; %s: Listen socket: " 1223 "SYN|FIN segment ignored (based on " 1224 "sysctl setting)\n", s, __func__); 1225 TCPSTAT_INC(tcps_badsyn); 1226 goto dropunlock; 1227 } 1228 /* 1229 * Segment's flags are (SYN) or (SYN|FIN). 1230 * 1231 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1232 * as they do not affect the state of the TCP FSM. 1233 * The data pointed to by TH_URG and th_urp is ignored. 1234 */ 1235 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1236 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1237 KASSERT(thflags & (TH_SYN), 1238 ("%s: Listen socket: TH_SYN not set", __func__)); 1239#ifdef INET6 1240 /* 1241 * If deprecated address is forbidden, 1242 * we do not accept SYN to deprecated interface 1243 * address to prevent any new inbound connection from 1244 * getting established. 1245 * When we do not accept SYN, we send a TCP RST, 1246 * with deprecated source address (instead of dropping 1247 * it). We compromise it as it is much better for peer 1248 * to send a RST, and RST will be the final packet 1249 * for the exchange. 1250 * 1251 * If we do not forbid deprecated addresses, we accept 1252 * the SYN packet. RFC2462 does not suggest dropping 1253 * SYN in this case. 1254 * If we decipher RFC2462 5.5.4, it says like this: 1255 * 1. use of deprecated addr with existing 1256 * communication is okay - "SHOULD continue to be 1257 * used" 1258 * 2. use of it with new communication: 1259 * (2a) "SHOULD NOT be used if alternate address 1260 * with sufficient scope is available" 1261 * (2b) nothing mentioned otherwise. 1262 * Here we fall into (2b) case as we have no choice in 1263 * our source address selection - we must obey the peer. 1264 * 1265 * The wording in RFC2462 is confusing, and there are 1266 * multiple description text for deprecated address 1267 * handling - worse, they are not exactly the same. 1268 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1269 */ 1270 if (isipv6 && !V_ip6_use_deprecated) { 1271 struct in6_ifaddr *ia6; 1272 1273 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1274 if (ia6 != NULL && 1275 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1276 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1277 log(LOG_DEBUG, "%s; %s: Listen socket: " 1278 "Connection attempt to deprecated " 1279 "IPv6 address rejected\n", 1280 s, __func__); 1281 rstreason = BANDLIM_RST_OPENPORT; 1282 goto dropwithreset; 1283 } 1284 } 1285#endif /* INET6 */ 1286 /* 1287 * Basic sanity checks on incoming SYN requests: 1288 * Don't respond if the destination is a link layer 1289 * broadcast according to RFC1122 4.2.3.10, p. 104. 1290 * If it is from this socket it must be forged. 1291 * Don't respond if the source or destination is a 1292 * global or subnet broad- or multicast address. 1293 * Note that it is quite possible to receive unicast 1294 * link-layer packets with a broadcast IP address. Use 1295 * in_broadcast() to find them. 1296 */ 1297 if (m->m_flags & (M_BCAST|M_MCAST)) { 1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1299 log(LOG_DEBUG, "%s; %s: Listen socket: " 1300 "Connection attempt from broad- or multicast " 1301 "link layer address ignored\n", s, __func__); 1302 goto dropunlock; 1303 } 1304#ifdef INET6 1305 if (isipv6) { 1306 if (th->th_dport == th->th_sport && 1307 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1309 log(LOG_DEBUG, "%s; %s: Listen socket: " 1310 "Connection attempt to/from self " 1311 "ignored\n", s, __func__); 1312 goto dropunlock; 1313 } 1314 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1315 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1316 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1317 log(LOG_DEBUG, "%s; %s: Listen socket: " 1318 "Connection attempt from/to multicast " 1319 "address ignored\n", s, __func__); 1320 goto dropunlock; 1321 } 1322 } 1323#endif 1324#if defined(INET) && defined(INET6) 1325 else 1326#endif 1327#ifdef INET 1328 { 1329 if (th->th_dport == th->th_sport && 1330 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1331 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1332 log(LOG_DEBUG, "%s; %s: Listen socket: " 1333 "Connection attempt from/to self " 1334 "ignored\n", s, __func__); 1335 goto dropunlock; 1336 } 1337 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1338 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1339 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1340 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1341 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1342 log(LOG_DEBUG, "%s; %s: Listen socket: " 1343 "Connection attempt from/to broad- " 1344 "or multicast address ignored\n", 1345 s, __func__); 1346 goto dropunlock; 1347 } 1348 } 1349#endif 1350 /* 1351 * SYN appears to be valid. Create compressed TCP state 1352 * for syncache. 1353 */ 1354#ifdef TCPDEBUG 1355 if (so->so_options & SO_DEBUG) 1356 tcp_trace(TA_INPUT, ostate, tp, 1357 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1358#endif 1359 TCP_PROBE3(debug__input, tp, th, m); 1360 tcp_dooptions(&to, optp, optlen, TO_SYN); 1361 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos, 1362 port)) 1363 goto tfo_socket_result; 1364 1365 /* 1366 * Entry added to syncache and mbuf consumed. 1367 * Only the listen socket is unlocked by syncache_add(). 1368 */ 1369 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1370 return (IPPROTO_DONE); 1371 } else if (tp->t_state == TCPS_LISTEN) { 1372 /* 1373 * When a listen socket is torn down the SO_ACCEPTCONN 1374 * flag is removed first while connections are drained 1375 * from the accept queue in a unlock/lock cycle of the 1376 * ACCEPT_LOCK, opening a race condition allowing a SYN 1377 * attempt go through unhandled. 1378 */ 1379 goto dropunlock; 1380 } 1381#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1382 if (tp->t_flags & TF_SIGNATURE) { 1383 tcp_dooptions(&to, optp, optlen, thflags); 1384 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1385 TCPSTAT_INC(tcps_sig_err_nosigopt); 1386 goto dropunlock; 1387 } 1388 if (!TCPMD5_ENABLED() || 1389 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1390 goto dropunlock; 1391 } 1392#endif 1393 TCP_PROBE5(receive, NULL, tp, m, 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 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1401 return (IPPROTO_DONE); 1402 1403dropwithreset: 1404 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1405 1406 if (inp != NULL) { 1407 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1408 INP_WUNLOCK(inp); 1409 } else 1410 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1411 m = NULL; /* mbuf chain got consumed. */ 1412 goto drop; 1413 1414dropunlock: 1415 if (m != NULL) 1416 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1417 1418 if (inp != NULL) 1419 INP_WUNLOCK(inp); 1420 1421drop: 1422 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1423 if (s != NULL) 1424 free(s, M_TCPLOG); 1425 if (m != NULL) 1426 m_freem(m); 1427 return (IPPROTO_DONE); 1428} 1429 1430/* 1431 * Automatic sizing of receive socket buffer. Often the send 1432 * buffer size is not optimally adjusted to the actual network 1433 * conditions at hand (delay bandwidth product). Setting the 1434 * buffer size too small limits throughput on links with high 1435 * bandwidth and high delay (eg. trans-continental/oceanic links). 1436 * 1437 * On the receive side the socket buffer memory is only rarely 1438 * used to any significant extent. This allows us to be much 1439 * more aggressive in scaling the receive socket buffer. For 1440 * the case that the buffer space is actually used to a large 1441 * extent and we run out of kernel memory we can simply drop 1442 * the new segments; TCP on the sender will just retransmit it 1443 * later. Setting the buffer size too big may only consume too 1444 * much kernel memory if the application doesn't read() from 1445 * the socket or packet loss or reordering makes use of the 1446 * reassembly queue. 1447 * 1448 * The criteria to step up the receive buffer one notch are: 1449 * 1. Application has not set receive buffer size with 1450 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1451 * 2. the number of bytes received during 1/2 of an sRTT 1452 * is at least 3/8 of the current socket buffer size. 1453 * 3. receive buffer size has not hit maximal automatic size; 1454 * 1455 * If all of the criteria are met we increaset the socket buffer 1456 * by a 1/2 (bounded by the max). This allows us to keep ahead 1457 * of slow-start but also makes it so our peer never gets limited 1458 * by our rwnd which we then open up causing a burst. 1459 * 1460 * This algorithm does two steps per RTT at most and only if 1461 * we receive a bulk stream w/o packet losses or reorderings. 1462 * Shrinking the buffer during idle times is not necessary as 1463 * it doesn't consume any memory when idle. 1464 * 1465 * TODO: Only step up if the application is actually serving 1466 * the buffer to better manage the socket buffer resources. 1467 */ 1468int 1469tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1470 struct tcpcb *tp, int tlen) 1471{ 1472 int newsize = 0; 1473 1474 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1475 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1476 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1477 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1478 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1479 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1480 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1481 } 1482 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1483 1484 /* Start over with next RTT. */ 1485 tp->rfbuf_ts = 0; 1486 tp->rfbuf_cnt = 0; 1487 } else { 1488 tp->rfbuf_cnt += tlen; /* add up */ 1489 } 1490 return (newsize); 1491} 1492 1493int 1494tcp_input(struct mbuf **mp, int *offp, int proto) 1495{ 1496 return(tcp_input_with_port(mp, offp, proto, 0)); 1497} 1498 1499void 1500tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1501{ 1502 /* 1503 * Since tp might be gone if the session entered 1504 * the TIME_WAIT state before coming here, we need 1505 * to check if the socket is still connected. 1506 */ 1507 if (tp == NULL) { 1508 return; 1509 } 1510 if (so == NULL) { 1511 return; 1512 } 1513 INP_LOCK_ASSERT(tp->t_inpcb); 1514 if (tp->t_flags & TF_WAKESOR) { 1515 tp->t_flags &= ~TF_WAKESOR; 1516 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1517 sorwakeup_locked(so); 1518 } 1519} 1520 1521void 1522tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1523 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1524{ 1525 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1526 int rstreason, todrop, win, incforsyn = 0; 1527 uint32_t tiwin; 1528 uint16_t nsegs; 1529 char *s; 1530 struct in_conninfo *inc; 1531 struct mbuf *mfree; 1532 struct tcpopt to; 1533 int tfo_syn; 1534 u_int maxseg; 1535 1536#ifdef TCPDEBUG 1537 /* 1538 * The size of tcp_saveipgen must be the size of the max ip header, 1539 * now IPv6. 1540 */ 1541 u_char tcp_saveipgen[IP6_HDR_LEN]; 1542 struct tcphdr tcp_savetcp; 1543 short ostate = 0; 1544#endif 1545 thflags = th->th_flags; 1546 inc = &tp->t_inpcb->inp_inc; 1547 tp->sackhint.last_sack_ack = 0; 1548 sack_changed = 0; 1549 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1550 1551 NET_EPOCH_ASSERT(); 1552 INP_WLOCK_ASSERT(tp->t_inpcb); 1553 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1554 __func__)); 1555 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1556 __func__)); 1557 1558#ifdef TCPPCAP 1559 /* Save segment, if requested. */ 1560 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1561#endif 1562 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1563 tlen, NULL, true); 1564 1565 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1566 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1567 log(LOG_DEBUG, "%s; %s: " 1568 "SYN|FIN segment ignored (based on " 1569 "sysctl setting)\n", s, __func__); 1570 free(s, M_TCPLOG); 1571 } 1572 goto drop; 1573 } 1574 1575 /* 1576 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1577 * check SEQ.ACK first. 1578 */ 1579 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1580 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1581 rstreason = BANDLIM_UNLIMITED; 1582 goto dropwithreset; 1583 } 1584 1585 /* 1586 * Segment received on connection. 1587 * Reset idle time and keep-alive timer. 1588 * XXX: This should be done after segment 1589 * validation to ignore broken/spoofed segs. 1590 */ 1591 tp->t_rcvtime = ticks; 1592 1593 /* 1594 * Scale up the window into a 32-bit value. 1595 * For the SYN_SENT state the scale is zero. 1596 */ 1597 tiwin = th->th_win << tp->snd_scale; 1598#ifdef STATS 1599 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1600#endif 1601 1602 /* 1603 * TCP ECN processing. 1604 */ 1605 if (tp->t_flags2 & TF2_ECN_PERMIT) { 1606 if (thflags & TH_CWR) { 1607 tp->t_flags2 &= ~TF2_ECN_SND_ECE; 1608 tp->t_flags |= TF_ACKNOW; 1609 } 1610 switch (iptos & IPTOS_ECN_MASK) { 1611 case IPTOS_ECN_CE: 1612 tp->t_flags2 |= TF2_ECN_SND_ECE; 1613 TCPSTAT_INC(tcps_ecn_ce); 1614 break; 1615 case IPTOS_ECN_ECT0: 1616 TCPSTAT_INC(tcps_ecn_ect0); 1617 break; 1618 case IPTOS_ECN_ECT1: 1619 TCPSTAT_INC(tcps_ecn_ect1); 1620 break; 1621 } 1622 1623 /* Process a packet differently from RFC3168. */ 1624 cc_ecnpkt_handler(tp, th, iptos); 1625 1626 /* Congestion experienced. */ 1627 if (thflags & TH_ECE) { 1628 cc_cong_signal(tp, th, CC_ECN); 1629 } 1630 } 1631 1632 /* 1633 * Parse options on any incoming segment. 1634 */ 1635 tcp_dooptions(&to, (u_char *)(th + 1), 1636 (th->th_off << 2) - sizeof(struct tcphdr), 1637 (thflags & TH_SYN) ? TO_SYN : 0); 1638 1639#if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1640 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1641 (to.to_flags & TOF_SIGNATURE) == 0) { 1642 TCPSTAT_INC(tcps_sig_err_sigopt); 1643 /* XXX: should drop? */ 1644 } 1645#endif 1646 /* 1647 * If echoed timestamp is later than the current time, 1648 * fall back to non RFC1323 RTT calculation. Normalize 1649 * timestamp if syncookies were used when this connection 1650 * was established. 1651 */ 1652 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1653 to.to_tsecr -= tp->ts_offset; 1654 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1655 to.to_tsecr = 0; 1656 else if (tp->t_flags & TF_PREVVALID && 1657 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1658 cc_cong_signal(tp, th, CC_RTO_ERR); 1659 } 1660 /* 1661 * Process options only when we get SYN/ACK back. The SYN case 1662 * for incoming connections is handled in tcp_syncache. 1663 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1664 * or <SYN,ACK>) segment itself is never scaled. 1665 * XXX this is traditional behavior, may need to be cleaned up. 1666 */ 1667 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1668 /* Handle parallel SYN for ECN */ 1669 if (!(thflags & TH_ACK) && 1670 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) && 1671 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) { 1672 tp->t_flags2 |= TF2_ECN_PERMIT; 1673 tp->t_flags2 |= TF2_ECN_SND_ECE; 1674 TCPSTAT_INC(tcps_ecn_shs); 1675 } 1676 if ((to.to_flags & TOF_SCALE) && 1677 (tp->t_flags & TF_REQ_SCALE) && 1678 !(tp->t_flags & TF_NOOPT)) { 1679 tp->t_flags |= TF_RCVD_SCALE; 1680 tp->snd_scale = to.to_wscale; 1681 } else 1682 tp->t_flags &= ~TF_REQ_SCALE; 1683 /* 1684 * Initial send window. It will be updated with 1685 * the next incoming segment to the scaled value. 1686 */ 1687 tp->snd_wnd = th->th_win; 1688 if ((to.to_flags & TOF_TS) && 1689 (tp->t_flags & TF_REQ_TSTMP) && 1690 !(tp->t_flags & TF_NOOPT)) { 1691 tp->t_flags |= TF_RCVD_TSTMP; 1692 tp->ts_recent = to.to_tsval; 1693 tp->ts_recent_age = tcp_ts_getticks(); 1694 } else 1695 tp->t_flags &= ~TF_REQ_TSTMP; 1696 if (to.to_flags & TOF_MSS) 1697 tcp_mss(tp, to.to_mss); 1698 if ((tp->t_flags & TF_SACK_PERMIT) && 1699 (!(to.to_flags & TOF_SACKPERM) || 1700 (tp->t_flags & TF_NOOPT))) 1701 tp->t_flags &= ~TF_SACK_PERMIT; 1702 if (IS_FASTOPEN(tp->t_flags)) { 1703 if ((to.to_flags & TOF_FASTOPEN) && 1704 !(tp->t_flags & TF_NOOPT)) { 1705 uint16_t mss; 1706 1707 if (to.to_flags & TOF_MSS) 1708 mss = to.to_mss; 1709 else 1710 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1711 mss = TCP6_MSS; 1712 else 1713 mss = TCP_MSS; 1714 tcp_fastopen_update_cache(tp, mss, 1715 to.to_tfo_len, to.to_tfo_cookie); 1716 } else 1717 tcp_fastopen_disable_path(tp); 1718 } 1719 } 1720 1721 /* 1722 * If timestamps were negotiated during SYN/ACK and a 1723 * segment without a timestamp is received, silently drop 1724 * the segment, unless it is a RST segment or missing timestamps are 1725 * tolerated. 1726 * See section 3.2 of RFC 7323. 1727 */ 1728 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1729 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1730 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1731 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1732 "segment processed normally\n", 1733 s, __func__); 1734 free(s, M_TCPLOG); 1735 } 1736 } else { 1737 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1738 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1739 "segment silently dropped\n", s, __func__); 1740 free(s, M_TCPLOG); 1741 } 1742 goto drop; 1743 } 1744 } 1745 /* 1746 * If timestamps were not negotiated during SYN/ACK and a 1747 * segment with a timestamp is received, ignore the 1748 * timestamp and process the packet normally. 1749 * See section 3.2 of RFC 7323. 1750 */ 1751 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1752 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1753 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1754 "segment processed normally\n", s, __func__); 1755 free(s, M_TCPLOG); 1756 } 1757 } 1758 1759 /* 1760 * Header prediction: check for the two common cases 1761 * of a uni-directional data xfer. If the packet has 1762 * no control flags, is in-sequence, the window didn't 1763 * change and we're not retransmitting, it's a 1764 * candidate. If the length is zero and the ack moved 1765 * forward, we're the sender side of the xfer. Just 1766 * free the data acked & wake any higher level process 1767 * that was blocked waiting for space. If the length 1768 * is non-zero and the ack didn't move, we're the 1769 * receiver side. If we're getting packets in-order 1770 * (the reassembly queue is empty), add the data to 1771 * the socket buffer and note that we need a delayed ack. 1772 * Make sure that the hidden state-flags are also off. 1773 * Since we check for TCPS_ESTABLISHED first, it can only 1774 * be TH_NEEDSYN. 1775 */ 1776 if (tp->t_state == TCPS_ESTABLISHED && 1777 th->th_seq == tp->rcv_nxt && 1778 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1779 tp->snd_nxt == tp->snd_max && 1780 tiwin && tiwin == tp->snd_wnd && 1781 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1782 SEGQ_EMPTY(tp) && 1783 ((to.to_flags & TOF_TS) == 0 || 1784 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1785 /* 1786 * If last ACK falls within this segment's sequence numbers, 1787 * record the timestamp. 1788 * NOTE that the test is modified according to the latest 1789 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1790 */ 1791 if ((to.to_flags & TOF_TS) != 0 && 1792 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1793 tp->ts_recent_age = tcp_ts_getticks(); 1794 tp->ts_recent = to.to_tsval; 1795 } 1796 1797 if (tlen == 0) { 1798 if (SEQ_GT(th->th_ack, tp->snd_una) && 1799 SEQ_LEQ(th->th_ack, tp->snd_max) && 1800 !IN_RECOVERY(tp->t_flags) && 1801 (to.to_flags & TOF_SACK) == 0 && 1802 TAILQ_EMPTY(&tp->snd_holes)) { 1803 /* 1804 * This is a pure ack for outstanding data. 1805 */ 1806 TCPSTAT_INC(tcps_predack); 1807 1808 /* 1809 * "bad retransmit" recovery without timestamps. 1810 */ 1811 if ((to.to_flags & TOF_TS) == 0 && 1812 tp->t_rxtshift == 1 && 1813 tp->t_flags & TF_PREVVALID && 1814 (int)(ticks - tp->t_badrxtwin) < 0) { 1815 cc_cong_signal(tp, th, CC_RTO_ERR); 1816 } 1817 1818 /* 1819 * Recalculate the transmit timer / rtt. 1820 * 1821 * Some boxes send broken timestamp replies 1822 * during the SYN+ACK phase, ignore 1823 * timestamps of 0 or we could calculate a 1824 * huge RTT and blow up the retransmit timer. 1825 */ 1826 if ((to.to_flags & TOF_TS) != 0 && 1827 to.to_tsecr) { 1828 uint32_t t; 1829 1830 t = tcp_ts_getticks() - to.to_tsecr; 1831 if (!tp->t_rttlow || tp->t_rttlow > t) 1832 tp->t_rttlow = t; 1833 tcp_xmit_timer(tp, 1834 TCP_TS_TO_TICKS(t) + 1); 1835 } else if (tp->t_rtttime && 1836 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1837 if (!tp->t_rttlow || 1838 tp->t_rttlow > ticks - tp->t_rtttime) 1839 tp->t_rttlow = ticks - tp->t_rtttime; 1840 tcp_xmit_timer(tp, 1841 ticks - tp->t_rtttime); 1842 } 1843 acked = BYTES_THIS_ACK(tp, th); 1844 1845#ifdef TCP_HHOOK 1846 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1847 hhook_run_tcp_est_in(tp, th, &to); 1848#endif 1849 1850 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1851 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1852 sbdrop(&so->so_snd, acked); 1853 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1854 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1855 tp->snd_recover = th->th_ack - 1; 1856 1857 /* 1858 * Let the congestion control algorithm update 1859 * congestion control related information. This 1860 * typically means increasing the congestion 1861 * window. 1862 */ 1863 cc_ack_received(tp, th, nsegs, CC_ACK); 1864 1865 tp->snd_una = th->th_ack; 1866 /* 1867 * Pull snd_wl2 up to prevent seq wrap relative 1868 * to th_ack. 1869 */ 1870 tp->snd_wl2 = th->th_ack; 1871 tp->t_dupacks = 0; 1872 m_freem(m); 1873 1874 /* 1875 * If all outstanding data are acked, stop 1876 * retransmit timer, otherwise restart timer 1877 * using current (possibly backed-off) value. 1878 * If process is waiting for space, 1879 * wakeup/selwakeup/signal. If data 1880 * are ready to send, let tcp_output 1881 * decide between more output or persist. 1882 */ 1883#ifdef TCPDEBUG 1884 if (so->so_options & SO_DEBUG) 1885 tcp_trace(TA_INPUT, ostate, tp, 1886 (void *)tcp_saveipgen, 1887 &tcp_savetcp, 0); 1888#endif 1889 TCP_PROBE3(debug__input, tp, th, m); 1890 if (tp->snd_una == tp->snd_max) 1891 tcp_timer_activate(tp, TT_REXMT, 0); 1892 else if (!tcp_timer_active(tp, TT_PERSIST)) 1893 tcp_timer_activate(tp, TT_REXMT, 1894 tp->t_rxtcur); 1895 sowwakeup(so); 1896 if (sbavail(&so->so_snd)) 1897 (void) tp->t_fb->tfb_tcp_output(tp); 1898 goto check_delack; 1899 } 1900 } else if (th->th_ack == tp->snd_una && 1901 tlen <= sbspace(&so->so_rcv)) { 1902 int newsize = 0; /* automatic sockbuf scaling */ 1903 1904 /* 1905 * This is a pure, in-sequence data packet with 1906 * nothing on the reassembly queue and we have enough 1907 * buffer space to take it. 1908 */ 1909 /* Clean receiver SACK report if present */ 1910 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1911 tcp_clean_sackreport(tp); 1912 TCPSTAT_INC(tcps_preddat); 1913 tp->rcv_nxt += tlen; 1914 if (tlen && 1915 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1916 (tp->t_fbyte_in == 0)) { 1917 tp->t_fbyte_in = ticks; 1918 if (tp->t_fbyte_in == 0) 1919 tp->t_fbyte_in = 1; 1920 if (tp->t_fbyte_out && tp->t_fbyte_in) 1921 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1922 } 1923 /* 1924 * Pull snd_wl1 up to prevent seq wrap relative to 1925 * th_seq. 1926 */ 1927 tp->snd_wl1 = th->th_seq; 1928 /* 1929 * Pull rcv_up up to prevent seq wrap relative to 1930 * rcv_nxt. 1931 */ 1932 tp->rcv_up = tp->rcv_nxt; 1933 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1934 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1935#ifdef TCPDEBUG 1936 if (so->so_options & SO_DEBUG) 1937 tcp_trace(TA_INPUT, ostate, tp, 1938 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1939#endif 1940 TCP_PROBE3(debug__input, tp, th, m); 1941 1942 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1943 1944 /* Add data to socket buffer. */ 1945 SOCKBUF_LOCK(&so->so_rcv); 1946 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1947 m_freem(m); 1948 } else { 1949 /* 1950 * Set new socket buffer size. 1951 * Give up when limit is reached. 1952 */ 1953 if (newsize) 1954 if (!sbreserve_locked(&so->so_rcv, 1955 newsize, so, NULL)) 1956 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1957 m_adj(m, drop_hdrlen); /* delayed header drop */ 1958 sbappendstream_locked(&so->so_rcv, m, 0); 1959 } 1960 /* NB: sorwakeup_locked() does an implicit unlock. */ 1961 sorwakeup_locked(so); 1962 if (DELAY_ACK(tp, tlen)) { 1963 tp->t_flags |= TF_DELACK; 1964 } else { 1965 tp->t_flags |= TF_ACKNOW; 1966 tp->t_fb->tfb_tcp_output(tp); 1967 } 1968 goto check_delack; 1969 } 1970 } 1971 1972 /* 1973 * Calculate amount of space in receive window, 1974 * and then do TCP input processing. 1975 * Receive window is amount of space in rcv queue, 1976 * but not less than advertised window. 1977 */ 1978 win = sbspace(&so->so_rcv); 1979 if (win < 0) 1980 win = 0; 1981 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1982 1983 switch (tp->t_state) { 1984 /* 1985 * If the state is SYN_RECEIVED: 1986 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1987 */ 1988 case TCPS_SYN_RECEIVED: 1989 if ((thflags & TH_ACK) && 1990 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1991 SEQ_GT(th->th_ack, tp->snd_max))) { 1992 rstreason = BANDLIM_RST_OPENPORT; 1993 goto dropwithreset; 1994 } 1995 if (IS_FASTOPEN(tp->t_flags)) { 1996 /* 1997 * When a TFO connection is in SYN_RECEIVED, the 1998 * only valid packets are the initial SYN, a 1999 * retransmit/copy of the initial SYN (possibly with 2000 * a subset of the original data), a valid ACK, a 2001 * FIN, or a RST. 2002 */ 2003 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 2004 rstreason = BANDLIM_RST_OPENPORT; 2005 goto dropwithreset; 2006 } else if (thflags & TH_SYN) { 2007 /* non-initial SYN is ignored */ 2008 if ((tcp_timer_active(tp, TT_DELACK) || 2009 tcp_timer_active(tp, TT_REXMT))) 2010 goto drop; 2011 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 2012 goto drop; 2013 } 2014 } 2015 break; 2016 2017 /* 2018 * If the state is SYN_SENT: 2019 * if seg contains a RST with valid ACK (SEQ.ACK has already 2020 * been verified), then drop the connection. 2021 * if seg contains a RST without an ACK, drop the seg. 2022 * if seg does not contain SYN, then drop the seg. 2023 * Otherwise this is an acceptable SYN segment 2024 * initialize tp->rcv_nxt and tp->irs 2025 * if seg contains ack then advance tp->snd_una 2026 * if seg contains an ECE and ECN support is enabled, the stream 2027 * is ECN capable. 2028 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2029 * arrange for segment to be acked (eventually) 2030 * continue processing rest of data/controls, beginning with URG 2031 */ 2032 case TCPS_SYN_SENT: 2033 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2034 TCP_PROBE5(connect__refused, NULL, tp, 2035 m, tp, th); 2036 tp = tcp_drop(tp, ECONNREFUSED); 2037 } 2038 if (thflags & TH_RST) 2039 goto drop; 2040 if (!(thflags & TH_SYN)) 2041 goto drop; 2042 2043 tp->irs = th->th_seq; 2044 tcp_rcvseqinit(tp); 2045 if (thflags & TH_ACK) { 2046 int tfo_partial_ack = 0; 2047 2048 TCPSTAT_INC(tcps_connects); 2049 soisconnected(so); 2050#ifdef MAC 2051 mac_socketpeer_set_from_mbuf(m, so); 2052#endif 2053 /* Do window scaling on this connection? */ 2054 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2055 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2056 tp->rcv_scale = tp->request_r_scale; 2057 } 2058 tp->rcv_adv += min(tp->rcv_wnd, 2059 TCP_MAXWIN << tp->rcv_scale); 2060 tp->snd_una++; /* SYN is acked */ 2061 /* 2062 * If not all the data that was sent in the TFO SYN 2063 * has been acked, resend the remainder right away. 2064 */ 2065 if (IS_FASTOPEN(tp->t_flags) && 2066 (tp->snd_una != tp->snd_max)) { 2067 tp->snd_nxt = th->th_ack; 2068 tfo_partial_ack = 1; 2069 } 2070 /* 2071 * If there's data, delay ACK; if there's also a FIN 2072 * ACKNOW will be turned on later. 2073 */ 2074 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2075 tcp_timer_activate(tp, TT_DELACK, 2076 tcp_delacktime); 2077 else 2078 tp->t_flags |= TF_ACKNOW; 2079 2080 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2081 (V_tcp_do_ecn == 1)) { 2082 tp->t_flags2 |= TF2_ECN_PERMIT; 2083 TCPSTAT_INC(tcps_ecn_shs); 2084 } 2085 2086 /* 2087 * Received <SYN,ACK> in SYN_SENT[*] state. 2088 * Transitions: 2089 * SYN_SENT --> ESTABLISHED 2090 * SYN_SENT* --> FIN_WAIT_1 2091 */ 2092 tp->t_starttime = ticks; 2093 if (tp->t_flags & TF_NEEDFIN) { 2094 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2095 tp->t_flags &= ~TF_NEEDFIN; 2096 thflags &= ~TH_SYN; 2097 } else { 2098 tcp_state_change(tp, TCPS_ESTABLISHED); 2099 TCP_PROBE5(connect__established, NULL, tp, 2100 m, tp, th); 2101 cc_conn_init(tp); 2102 tcp_timer_activate(tp, TT_KEEP, 2103 TP_KEEPIDLE(tp)); 2104 } 2105 } else { 2106 /* 2107 * Received initial SYN in SYN-SENT[*] state => 2108 * simultaneous open. 2109 * If it succeeds, connection is * half-synchronized. 2110 * Otherwise, do 3-way handshake: 2111 * SYN-SENT -> SYN-RECEIVED 2112 * SYN-SENT* -> SYN-RECEIVED* 2113 */ 2114 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2115 tcp_timer_activate(tp, TT_REXMT, 0); 2116 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2117 } 2118 2119 INP_WLOCK_ASSERT(tp->t_inpcb); 2120 2121 /* 2122 * Advance th->th_seq to correspond to first data byte. 2123 * If data, trim to stay within window, 2124 * dropping FIN if necessary. 2125 */ 2126 th->th_seq++; 2127 if (tlen > tp->rcv_wnd) { 2128 todrop = tlen - tp->rcv_wnd; 2129 m_adj(m, -todrop); 2130 tlen = tp->rcv_wnd; 2131 thflags &= ~TH_FIN; 2132 TCPSTAT_INC(tcps_rcvpackafterwin); 2133 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2134 } 2135 tp->snd_wl1 = th->th_seq - 1; 2136 tp->rcv_up = th->th_seq; 2137 /* 2138 * Client side of transaction: already sent SYN and data. 2139 * If the remote host used T/TCP to validate the SYN, 2140 * our data will be ACK'd; if so, enter normal data segment 2141 * processing in the middle of step 5, ack processing. 2142 * Otherwise, goto step 6. 2143 */ 2144 if (thflags & TH_ACK) 2145 goto process_ACK; 2146 2147 goto step6; 2148 2149 /* 2150 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2151 * do normal processing. 2152 * 2153 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2154 */ 2155 case TCPS_LAST_ACK: 2156 case TCPS_CLOSING: 2157 break; /* continue normal processing */ 2158 } 2159 2160 /* 2161 * States other than LISTEN or SYN_SENT. 2162 * First check the RST flag and sequence number since reset segments 2163 * are exempt from the timestamp and connection count tests. This 2164 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2165 * below which allowed reset segments in half the sequence space 2166 * to fall though and be processed (which gives forged reset 2167 * segments with a random sequence number a 50 percent chance of 2168 * killing a connection). 2169 * Then check timestamp, if present. 2170 * Then check the connection count, if present. 2171 * Then check that at least some bytes of segment are within 2172 * receive window. If segment begins before rcv_nxt, 2173 * drop leading data (and SYN); if nothing left, just ack. 2174 */ 2175 if (thflags & TH_RST) { 2176 /* 2177 * RFC5961 Section 3.2 2178 * 2179 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2180 * - If RST is in window, we send challenge ACK. 2181 * 2182 * Note: to take into account delayed ACKs, we should 2183 * test against last_ack_sent instead of rcv_nxt. 2184 * Note 2: we handle special case of closed window, not 2185 * covered by the RFC. 2186 */ 2187 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2188 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2189 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2190 KASSERT(tp->t_state != TCPS_SYN_SENT, 2191 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2192 __func__, th, tp)); 2193 2194 if (V_tcp_insecure_rst || 2195 tp->last_ack_sent == th->th_seq) { 2196 TCPSTAT_INC(tcps_drops); 2197 /* Drop the connection. */ 2198 switch (tp->t_state) { 2199 case TCPS_SYN_RECEIVED: 2200 so->so_error = ECONNREFUSED; 2201 goto close; 2202 case TCPS_ESTABLISHED: 2203 case TCPS_FIN_WAIT_1: 2204 case TCPS_FIN_WAIT_2: 2205 case TCPS_CLOSE_WAIT: 2206 case TCPS_CLOSING: 2207 case TCPS_LAST_ACK: 2208 so->so_error = ECONNRESET; 2209 close: 2210 /* FALLTHROUGH */ 2211 default: 2212 tp = tcp_close(tp); 2213 } 2214 } else { 2215 TCPSTAT_INC(tcps_badrst); 2216 /* Send challenge ACK. */ 2217 tcp_respond(tp, mtod(m, void *), th, m, 2218 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2219 tp->last_ack_sent = tp->rcv_nxt; 2220 m = NULL; 2221 } 2222 } 2223 goto drop; 2224 } 2225 2226 /* 2227 * RFC5961 Section 4.2 2228 * Send challenge ACK for any SYN in synchronized state. 2229 */ 2230 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2231 tp->t_state != TCPS_SYN_RECEIVED) { 2232 TCPSTAT_INC(tcps_badsyn); 2233 if (V_tcp_insecure_syn && 2234 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2235 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2236 tp = tcp_drop(tp, ECONNRESET); 2237 rstreason = BANDLIM_UNLIMITED; 2238 } else { 2239 /* Send challenge ACK. */ 2240 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2241 tp->snd_nxt, TH_ACK); 2242 tp->last_ack_sent = tp->rcv_nxt; 2243 m = NULL; 2244 } 2245 goto drop; 2246 } 2247 2248 /* 2249 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2250 * and it's less than ts_recent, drop it. 2251 */ 2252 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2253 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2254 /* Check to see if ts_recent is over 24 days old. */ 2255 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2256 /* 2257 * Invalidate ts_recent. If this segment updates 2258 * ts_recent, the age will be reset later and ts_recent 2259 * will get a valid value. If it does not, setting 2260 * ts_recent to zero will at least satisfy the 2261 * requirement that zero be placed in the timestamp 2262 * echo reply when ts_recent isn't valid. The 2263 * age isn't reset until we get a valid ts_recent 2264 * because we don't want out-of-order segments to be 2265 * dropped when ts_recent is old. 2266 */ 2267 tp->ts_recent = 0; 2268 } else { 2269 TCPSTAT_INC(tcps_rcvduppack); 2270 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2271 TCPSTAT_INC(tcps_pawsdrop); 2272 if (tlen) 2273 goto dropafterack; 2274 goto drop; 2275 } 2276 } 2277 2278 /* 2279 * In the SYN-RECEIVED state, validate that the packet belongs to 2280 * this connection before trimming the data to fit the receive 2281 * window. Check the sequence number versus IRS since we know 2282 * the sequence numbers haven't wrapped. This is a partial fix 2283 * for the "LAND" DoS attack. 2284 */ 2285 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2286 rstreason = BANDLIM_RST_OPENPORT; 2287 goto dropwithreset; 2288 } 2289 2290 todrop = tp->rcv_nxt - th->th_seq; 2291 if (todrop > 0) { 2292 if (thflags & TH_SYN) { 2293 thflags &= ~TH_SYN; 2294 th->th_seq++; 2295 if (th->th_urp > 1) 2296 th->th_urp--; 2297 else 2298 thflags &= ~TH_URG; 2299 todrop--; 2300 } 2301 /* 2302 * Following if statement from Stevens, vol. 2, p. 960. 2303 */ 2304 if (todrop > tlen 2305 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2306 /* 2307 * Any valid FIN must be to the left of the window. 2308 * At this point the FIN must be a duplicate or out 2309 * of sequence; drop it. 2310 */ 2311 thflags &= ~TH_FIN; 2312 2313 /* 2314 * Send an ACK to resynchronize and drop any data. 2315 * But keep on processing for RST or ACK. 2316 */ 2317 tp->t_flags |= TF_ACKNOW; 2318 todrop = tlen; 2319 TCPSTAT_INC(tcps_rcvduppack); 2320 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2321 } else { 2322 TCPSTAT_INC(tcps_rcvpartduppack); 2323 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2324 } 2325 /* 2326 * DSACK - add SACK block for dropped range 2327 */ 2328 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2329 tcp_update_sack_list(tp, th->th_seq, 2330 th->th_seq + todrop); 2331 /* 2332 * ACK now, as the next in-sequence segment 2333 * will clear the DSACK block again 2334 */ 2335 tp->t_flags |= TF_ACKNOW; 2336 } 2337 drop_hdrlen += todrop; /* drop from the top afterwards */ 2338 th->th_seq += todrop; 2339 tlen -= todrop; 2340 if (th->th_urp > todrop) 2341 th->th_urp -= todrop; 2342 else { 2343 thflags &= ~TH_URG; 2344 th->th_urp = 0; 2345 } 2346 } 2347 2348 /* 2349 * If new data are received on a connection after the 2350 * user processes are gone, then RST the other end. 2351 */ 2352 if ((so->so_state & SS_NOFDREF) && 2353 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2354 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2355 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2356 "after socket was closed, " 2357 "sending RST and removing tcpcb\n", 2358 s, __func__, tcpstates[tp->t_state], tlen); 2359 free(s, M_TCPLOG); 2360 } 2361 tp = tcp_close(tp); 2362 TCPSTAT_INC(tcps_rcvafterclose); 2363 rstreason = BANDLIM_UNLIMITED; 2364 goto dropwithreset; 2365 } 2366 2367 /* 2368 * If segment ends after window, drop trailing data 2369 * (and PUSH and FIN); if nothing left, just ACK. 2370 */ 2371 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2372 if (todrop > 0) { 2373 TCPSTAT_INC(tcps_rcvpackafterwin); 2374 if (todrop >= tlen) { 2375 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2376 /* 2377 * If window is closed can only take segments at 2378 * window edge, and have to drop data and PUSH from 2379 * incoming segments. Continue processing, but 2380 * remember to ack. Otherwise, drop segment 2381 * and ack. 2382 */ 2383 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2384 tp->t_flags |= TF_ACKNOW; 2385 TCPSTAT_INC(tcps_rcvwinprobe); 2386 } else 2387 goto dropafterack; 2388 } else 2389 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2390 m_adj(m, -todrop); 2391 tlen -= todrop; 2392 thflags &= ~(TH_PUSH|TH_FIN); 2393 } 2394 2395 /* 2396 * If last ACK falls within this segment's sequence numbers, 2397 * record its timestamp. 2398 * NOTE: 2399 * 1) That the test incorporates suggestions from the latest 2400 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2401 * 2) That updating only on newer timestamps interferes with 2402 * our earlier PAWS tests, so this check should be solely 2403 * predicated on the sequence space of this segment. 2404 * 3) That we modify the segment boundary check to be 2405 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2406 * instead of RFC1323's 2407 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2408 * This modified check allows us to overcome RFC1323's 2409 * limitations as described in Stevens TCP/IP Illustrated 2410 * Vol. 2 p.869. In such cases, we can still calculate the 2411 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2412 */ 2413 if ((to.to_flags & TOF_TS) != 0 && 2414 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2415 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2416 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2417 tp->ts_recent_age = tcp_ts_getticks(); 2418 tp->ts_recent = to.to_tsval; 2419 } 2420 2421 /* 2422 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2423 * flag is on (half-synchronized state), then queue data for 2424 * later processing; else drop segment and return. 2425 */ 2426 if ((thflags & TH_ACK) == 0) { 2427 if (tp->t_state == TCPS_SYN_RECEIVED || 2428 (tp->t_flags & TF_NEEDSYN)) { 2429 if (tp->t_state == TCPS_SYN_RECEIVED && 2430 IS_FASTOPEN(tp->t_flags)) { 2431 tp->snd_wnd = tiwin; 2432 cc_conn_init(tp); 2433 } 2434 goto step6; 2435 } else if (tp->t_flags & TF_ACKNOW) 2436 goto dropafterack; 2437 else 2438 goto drop; 2439 } 2440 2441 /* 2442 * Ack processing. 2443 */ 2444 switch (tp->t_state) { 2445 /* 2446 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2447 * ESTABLISHED state and continue processing. 2448 * The ACK was checked above. 2449 */ 2450 case TCPS_SYN_RECEIVED: 2451 2452 TCPSTAT_INC(tcps_connects); 2453 soisconnected(so); 2454 /* Do window scaling? */ 2455 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2456 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2457 tp->rcv_scale = tp->request_r_scale; 2458 } 2459 tp->snd_wnd = tiwin; 2460 /* 2461 * Make transitions: 2462 * SYN-RECEIVED -> ESTABLISHED 2463 * SYN-RECEIVED* -> FIN-WAIT-1 2464 */ 2465 tp->t_starttime = ticks; 2466 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2467 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2468 tp->t_tfo_pending = NULL; 2469 } 2470 if (tp->t_flags & TF_NEEDFIN) { 2471 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2472 tp->t_flags &= ~TF_NEEDFIN; 2473 } else { 2474 tcp_state_change(tp, TCPS_ESTABLISHED); 2475 TCP_PROBE5(accept__established, NULL, tp, 2476 m, tp, th); 2477 /* 2478 * TFO connections call cc_conn_init() during SYN 2479 * processing. Calling it again here for such 2480 * connections is not harmless as it would undo the 2481 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2482 * is retransmitted. 2483 */ 2484 if (!IS_FASTOPEN(tp->t_flags)) 2485 cc_conn_init(tp); 2486 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2487 } 2488 /* 2489 * Account for the ACK of our SYN prior to 2490 * regular ACK processing below, except for 2491 * simultaneous SYN, which is handled later. 2492 */ 2493 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2494 incforsyn = 1; 2495 /* 2496 * If segment contains data or ACK, will call tcp_reass() 2497 * later; if not, do so now to pass queued data to user. 2498 */ 2499 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2500 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2501 (struct mbuf *)0); 2502 tcp_handle_wakeup(tp, so); 2503 } 2504 tp->snd_wl1 = th->th_seq - 1; 2505 /* FALLTHROUGH */ 2506 2507 /* 2508 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2509 * ACKs. If the ack is in the range 2510 * tp->snd_una < th->th_ack <= tp->snd_max 2511 * then advance tp->snd_una to th->th_ack and drop 2512 * data from the retransmission queue. If this ACK reflects 2513 * more up to date window information we update our window information. 2514 */ 2515 case TCPS_ESTABLISHED: 2516 case TCPS_FIN_WAIT_1: 2517 case TCPS_FIN_WAIT_2: 2518 case TCPS_CLOSE_WAIT: 2519 case TCPS_CLOSING: 2520 case TCPS_LAST_ACK: 2521 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2522 TCPSTAT_INC(tcps_rcvacktoomuch); 2523 goto dropafterack; 2524 } 2525 if ((tp->t_flags & TF_SACK_PERMIT) && 2526 ((to.to_flags & TOF_SACK) || 2527 !TAILQ_EMPTY(&tp->snd_holes))) 2528 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2529 else 2530 /* 2531 * Reset the value so that previous (valid) value 2532 * from the last ack with SACK doesn't get used. 2533 */ 2534 tp->sackhint.sacked_bytes = 0; 2535 2536#ifdef TCP_HHOOK 2537 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2538 hhook_run_tcp_est_in(tp, th, &to); 2539#endif 2540 2541 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2542 maxseg = tcp_maxseg(tp); 2543 if (tlen == 0 && 2544 (tiwin == tp->snd_wnd || 2545 (tp->t_flags & TF_SACK_PERMIT))) { 2546 /* 2547 * If this is the first time we've seen a 2548 * FIN from the remote, this is not a 2549 * duplicate and it needs to be processed 2550 * normally. This happens during a 2551 * simultaneous close. 2552 */ 2553 if ((thflags & TH_FIN) && 2554 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2555 tp->t_dupacks = 0; 2556 break; 2557 } 2558 TCPSTAT_INC(tcps_rcvdupack); 2559 /* 2560 * If we have outstanding data (other than 2561 * a window probe), this is a completely 2562 * duplicate ack (ie, window info didn't 2563 * change and FIN isn't set), 2564 * the ack is the biggest we've 2565 * seen and we've seen exactly our rexmt 2566 * threshold of them, assume a packet 2567 * has been dropped and retransmit it. 2568 * Kludge snd_nxt & the congestion 2569 * window so we send only this one 2570 * packet. 2571 * 2572 * We know we're losing at the current 2573 * window size so do congestion avoidance 2574 * (set ssthresh to half the current window 2575 * and pull our congestion window back to 2576 * the new ssthresh). 2577 * 2578 * Dup acks mean that packets have left the 2579 * network (they're now cached at the receiver) 2580 * so bump cwnd by the amount in the receiver 2581 * to keep a constant cwnd packets in the 2582 * network. 2583 * 2584 * When using TCP ECN, notify the peer that 2585 * we reduced the cwnd. 2586 */ 2587 /* 2588 * Following 2 kinds of acks should not affect 2589 * dupack counting: 2590 * 1) Old acks 2591 * 2) Acks with SACK but without any new SACK 2592 * information in them. These could result from 2593 * any anomaly in the network like a switch 2594 * duplicating packets or a possible DoS attack. 2595 */ 2596 if (th->th_ack != tp->snd_una || 2597 ((tp->t_flags & TF_SACK_PERMIT) && 2598 (to.to_flags & TOF_SACK) && 2599 !sack_changed)) 2600 break; 2601 else if (!tcp_timer_active(tp, TT_REXMT)) 2602 tp->t_dupacks = 0; 2603 else if (++tp->t_dupacks > tcprexmtthresh || 2604 IN_FASTRECOVERY(tp->t_flags)) { 2605 cc_ack_received(tp, th, nsegs, 2606 CC_DUPACK); 2607 if (V_tcp_do_prr && 2608 IN_FASTRECOVERY(tp->t_flags) && 2609 (tp->t_flags & TF_SACK_PERMIT)) { 2610 tcp_do_prr_ack(tp, th, &to); 2611 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2612 (to.to_flags & TOF_SACK) && 2613 IN_FASTRECOVERY(tp->t_flags)) { 2614 int awnd; 2615 2616 /* 2617 * Compute the amount of data in flight first. 2618 * We can inject new data into the pipe iff 2619 * we have less than 1/2 the original window's 2620 * worth of data in flight. 2621 */ 2622 if (V_tcp_do_rfc6675_pipe) 2623 awnd = tcp_compute_pipe(tp); 2624 else 2625 awnd = (tp->snd_nxt - tp->snd_fack) + 2626 tp->sackhint.sack_bytes_rexmit; 2627 2628 if (awnd < tp->snd_ssthresh) { 2629 tp->snd_cwnd += maxseg; 2630 if (tp->snd_cwnd > tp->snd_ssthresh) 2631 tp->snd_cwnd = tp->snd_ssthresh; 2632 } 2633 } else 2634 tp->snd_cwnd += maxseg; 2635 (void) tp->t_fb->tfb_tcp_output(tp); 2636 goto drop; 2637 } else if (tp->t_dupacks == tcprexmtthresh || 2638 (tp->t_flags & TF_SACK_PERMIT && 2639 V_tcp_do_rfc6675_pipe && 2640 tp->sackhint.sacked_bytes > 2641 (tcprexmtthresh - 1) * maxseg)) { 2642enter_recovery: 2643 /* 2644 * Above is the RFC6675 trigger condition of 2645 * more than (dupthresh-1)*maxseg sacked data. 2646 * If the count of holes in the 2647 * scoreboard is >= dupthresh, we could 2648 * also enter loss recovery, but don't 2649 * have that value readily available. 2650 */ 2651 tp->t_dupacks = tcprexmtthresh; 2652 tcp_seq onxt = tp->snd_nxt; 2653 2654 /* 2655 * If we're doing sack, or prr, check 2656 * to see if we're already in sack 2657 * recovery. If we're not doing sack, 2658 * check to see if we're in newreno 2659 * recovery. 2660 */ 2661 if (V_tcp_do_prr || 2662 (tp->t_flags & TF_SACK_PERMIT)) { 2663 if (IN_FASTRECOVERY(tp->t_flags)) { 2664 tp->t_dupacks = 0; 2665 break; 2666 } 2667 } else { 2668 if (SEQ_LEQ(th->th_ack, 2669 tp->snd_recover)) { 2670 tp->t_dupacks = 0; 2671 break; 2672 } 2673 } 2674 /* Congestion signal before ack. */ 2675 cc_cong_signal(tp, th, CC_NDUPACK); 2676 cc_ack_received(tp, th, nsegs, 2677 CC_DUPACK); 2678 tcp_timer_activate(tp, TT_REXMT, 0); 2679 tp->t_rtttime = 0; 2680 if (V_tcp_do_prr) { 2681 /* 2682 * snd_ssthresh is already updated by 2683 * cc_cong_signal. 2684 */ 2685 tp->sackhint.prr_delivered = 2686 tp->sackhint.sacked_bytes; 2687 tp->sackhint.recover_fs = max(1, 2688 tp->snd_nxt - tp->snd_una); 2689 } 2690 if ((tp->t_flags & TF_SACK_PERMIT) && 2691 (to.to_flags & TOF_SACK)) { 2692 TCPSTAT_INC( 2693 tcps_sack_recovery_episode); 2694 tp->snd_recover = tp->snd_nxt; 2695 tp->snd_cwnd = maxseg; 2696 (void) tp->t_fb->tfb_tcp_output(tp); 2697 if (SEQ_GT(th->th_ack, tp->snd_una)) 2698 goto resume_partialack; 2699 goto drop; 2700 } 2701 tp->snd_nxt = th->th_ack; 2702 tp->snd_cwnd = maxseg; 2703 (void) tp->t_fb->tfb_tcp_output(tp); 2704 KASSERT(tp->snd_limited <= 2, 2705 ("%s: tp->snd_limited too big", 2706 __func__)); 2707 tp->snd_cwnd = tp->snd_ssthresh + 2708 maxseg * 2709 (tp->t_dupacks - tp->snd_limited); 2710 if (SEQ_GT(onxt, tp->snd_nxt)) 2711 tp->snd_nxt = onxt; 2712 goto drop; 2713 } else if (V_tcp_do_rfc3042) { 2714 /* 2715 * Process first and second duplicate 2716 * ACKs. Each indicates a segment 2717 * leaving the network, creating room 2718 * for more. Make sure we can send a 2719 * packet on reception of each duplicate 2720 * ACK by increasing snd_cwnd by one 2721 * segment. Restore the original 2722 * snd_cwnd after packet transmission. 2723 */ 2724 cc_ack_received(tp, th, nsegs, 2725 CC_DUPACK); 2726 uint32_t oldcwnd = tp->snd_cwnd; 2727 tcp_seq oldsndmax = tp->snd_max; 2728 u_int sent; 2729 int avail; 2730 2731 KASSERT(tp->t_dupacks == 1 || 2732 tp->t_dupacks == 2, 2733 ("%s: dupacks not 1 or 2", 2734 __func__)); 2735 if (tp->t_dupacks == 1) 2736 tp->snd_limited = 0; 2737 tp->snd_cwnd = 2738 (tp->snd_nxt - tp->snd_una) + 2739 (tp->t_dupacks - tp->snd_limited) * 2740 maxseg; 2741 /* 2742 * Only call tcp_output when there 2743 * is new data available to be sent. 2744 * Otherwise we would send pure ACKs. 2745 */ 2746 SOCKBUF_LOCK(&so->so_snd); 2747 avail = sbavail(&so->so_snd) - 2748 (tp->snd_nxt - tp->snd_una); 2749 SOCKBUF_UNLOCK(&so->so_snd); 2750 if (avail > 0) 2751 (void) tp->t_fb->tfb_tcp_output(tp); 2752 sent = tp->snd_max - oldsndmax; 2753 if (sent > maxseg) { 2754 KASSERT((tp->t_dupacks == 2 && 2755 tp->snd_limited == 0) || 2756 (sent == maxseg + 1 && 2757 tp->t_flags & TF_SENTFIN), 2758 ("%s: sent too much", 2759 __func__)); 2760 tp->snd_limited = 2; 2761 } else if (sent > 0) 2762 ++tp->snd_limited; 2763 tp->snd_cwnd = oldcwnd; 2764 goto drop; 2765 } 2766 } 2767 break; 2768 } else { 2769 /* 2770 * This ack is advancing the left edge, reset the 2771 * counter. 2772 */ 2773 tp->t_dupacks = 0; 2774 /* 2775 * If this ack also has new SACK info, increment the 2776 * counter as per rfc6675. The variable 2777 * sack_changed tracks all changes to the SACK 2778 * scoreboard, including when partial ACKs without 2779 * SACK options are received, and clear the scoreboard 2780 * from the left side. Such partial ACKs should not be 2781 * counted as dupacks here. 2782 */ 2783 if ((tp->t_flags & TF_SACK_PERMIT) && 2784 (to.to_flags & TOF_SACK) && 2785 sack_changed) { 2786 tp->t_dupacks++; 2787 /* limit overhead by setting maxseg last */ 2788 if (!IN_FASTRECOVERY(tp->t_flags) && 2789 (tp->sackhint.sacked_bytes > 2790 ((tcprexmtthresh - 1) * 2791 (maxseg = tcp_maxseg(tp))))) { 2792 goto enter_recovery; 2793 } 2794 } 2795 } 2796 2797resume_partialack: 2798 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2799 ("%s: th_ack <= snd_una", __func__)); 2800 2801 /* 2802 * If the congestion window was inflated to account 2803 * for the other side's cached packets, retract it. 2804 */ 2805 if (IN_FASTRECOVERY(tp->t_flags)) { 2806 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2807 if (tp->t_flags & TF_SACK_PERMIT) 2808 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2809 tcp_timer_activate(tp, TT_REXMT, 0); 2810 tp->t_rtttime = 0; 2811 tcp_do_prr_ack(tp, th, &to); 2812 tp->t_flags |= TF_ACKNOW; 2813 (void) tcp_output(tp); 2814 } else 2815 tcp_sack_partialack(tp, th); 2816 else 2817 tcp_newreno_partial_ack(tp, th); 2818 } else 2819 cc_post_recovery(tp, th); 2820 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2821 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2822 if (V_tcp_do_prr) { 2823 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2824 tp->snd_fack = th->th_ack; 2825 tcp_do_prr_ack(tp, th, &to); 2826 (void) tcp_output(tp); 2827 } 2828 } else 2829 cc_post_recovery(tp, th); 2830 } 2831 /* 2832 * If we reach this point, ACK is not a duplicate, 2833 * i.e., it ACKs something we sent. 2834 */ 2835 if (tp->t_flags & TF_NEEDSYN) { 2836 /* 2837 * T/TCP: Connection was half-synchronized, and our 2838 * SYN has been ACK'd (so connection is now fully 2839 * synchronized). Go to non-starred state, 2840 * increment snd_una for ACK of SYN, and check if 2841 * we can do window scaling. 2842 */ 2843 tp->t_flags &= ~TF_NEEDSYN; 2844 tp->snd_una++; 2845 /* Do window scaling? */ 2846 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2847 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2848 tp->rcv_scale = tp->request_r_scale; 2849 /* Send window already scaled. */ 2850 } 2851 } 2852 2853process_ACK: 2854 INP_WLOCK_ASSERT(tp->t_inpcb); 2855 2856 /* 2857 * Adjust for the SYN bit in sequence space, 2858 * but don't account for it in cwnd calculations. 2859 * This is for the SYN_RECEIVED, non-simultaneous 2860 * SYN case. SYN_SENT and simultaneous SYN are 2861 * treated elsewhere. 2862 */ 2863 if (incforsyn) 2864 tp->snd_una++; 2865 acked = BYTES_THIS_ACK(tp, th); 2866 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2867 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2868 tp->snd_una, th->th_ack, tp, m)); 2869 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2870 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2871 2872 /* 2873 * If we just performed our first retransmit, and the ACK 2874 * arrives within our recovery window, then it was a mistake 2875 * to do the retransmit in the first place. Recover our 2876 * original cwnd and ssthresh, and proceed to transmit where 2877 * we left off. 2878 */ 2879 if (tp->t_rxtshift == 1 && 2880 tp->t_flags & TF_PREVVALID && 2881 tp->t_badrxtwin && 2882 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2883 cc_cong_signal(tp, th, CC_RTO_ERR); 2884 2885 /* 2886 * If we have a timestamp reply, update smoothed 2887 * round trip time. If no timestamp is present but 2888 * transmit timer is running and timed sequence 2889 * number was acked, update smoothed round trip time. 2890 * Since we now have an rtt measurement, cancel the 2891 * timer backoff (cf., Phil Karn's retransmit alg.). 2892 * Recompute the initial retransmit timer. 2893 * 2894 * Some boxes send broken timestamp replies 2895 * during the SYN+ACK phase, ignore 2896 * timestamps of 0 or we could calculate a 2897 * huge RTT and blow up the retransmit timer. 2898 */ 2899 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2900 uint32_t t; 2901 2902 t = tcp_ts_getticks() - to.to_tsecr; 2903 if (!tp->t_rttlow || tp->t_rttlow > t) 2904 tp->t_rttlow = t; 2905 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2906 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2907 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2908 tp->t_rttlow = ticks - tp->t_rtttime; 2909 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2910 } 2911 2912 /* 2913 * If all outstanding data is acked, stop retransmit 2914 * timer and remember to restart (more output or persist). 2915 * If there is more data to be acked, restart retransmit 2916 * timer, using current (possibly backed-off) value. 2917 */ 2918 if (th->th_ack == tp->snd_max) { 2919 tcp_timer_activate(tp, TT_REXMT, 0); 2920 needoutput = 1; 2921 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2922 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2923 2924 /* 2925 * If no data (only SYN) was ACK'd, 2926 * skip rest of ACK processing. 2927 */ 2928 if (acked == 0) 2929 goto step6; 2930 2931 /* 2932 * Let the congestion control algorithm update congestion 2933 * control related information. This typically means increasing 2934 * the congestion window. 2935 */ 2936 cc_ack_received(tp, th, nsegs, CC_ACK); 2937 2938 SOCKBUF_LOCK(&so->so_snd); 2939 if (acked > sbavail(&so->so_snd)) { 2940 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2941 tp->snd_wnd -= sbavail(&so->so_snd); 2942 else 2943 tp->snd_wnd = 0; 2944 mfree = sbcut_locked(&so->so_snd, 2945 (int)sbavail(&so->so_snd)); 2946 ourfinisacked = 1; 2947 } else { 2948 mfree = sbcut_locked(&so->so_snd, acked); 2949 if (tp->snd_wnd >= (uint32_t) acked) 2950 tp->snd_wnd -= acked; 2951 else 2952 tp->snd_wnd = 0; 2953 ourfinisacked = 0; 2954 } 2955 /* NB: sowwakeup_locked() does an implicit unlock. */ 2956 sowwakeup_locked(so); 2957 m_freem(mfree); 2958 /* Detect una wraparound. */ 2959 if (!IN_RECOVERY(tp->t_flags) && 2960 SEQ_GT(tp->snd_una, tp->snd_recover) && 2961 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2962 tp->snd_recover = th->th_ack - 1; 2963 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2964 if (IN_RECOVERY(tp->t_flags) && 2965 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2966 EXIT_RECOVERY(tp->t_flags); 2967 } 2968 tp->snd_una = th->th_ack; 2969 if (tp->t_flags & TF_SACK_PERMIT) { 2970 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2971 tp->snd_recover = tp->snd_una; 2972 } 2973 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2974 tp->snd_nxt = tp->snd_una; 2975 2976 switch (tp->t_state) { 2977 /* 2978 * In FIN_WAIT_1 STATE in addition to the processing 2979 * for the ESTABLISHED state if our FIN is now acknowledged 2980 * then enter FIN_WAIT_2. 2981 */ 2982 case TCPS_FIN_WAIT_1: 2983 if (ourfinisacked) { 2984 /* 2985 * If we can't receive any more 2986 * data, then closing user can proceed. 2987 * Starting the timer is contrary to the 2988 * specification, but if we don't get a FIN 2989 * we'll hang forever. 2990 * 2991 * XXXjl: 2992 * we should release the tp also, and use a 2993 * compressed state. 2994 */ 2995 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2996 soisdisconnected(so); 2997 tcp_timer_activate(tp, TT_2MSL, 2998 (tcp_fast_finwait2_recycle ? 2999 tcp_finwait2_timeout : 3000 TP_MAXIDLE(tp))); 3001 } 3002 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3003 } 3004 break; 3005 3006 /* 3007 * In CLOSING STATE in addition to the processing for 3008 * the ESTABLISHED state if the ACK acknowledges our FIN 3009 * then enter the TIME-WAIT state, otherwise ignore 3010 * the segment. 3011 */ 3012 case TCPS_CLOSING: 3013 if (ourfinisacked) { 3014 tcp_twstart(tp); 3015 m_freem(m); 3016 return; 3017 } 3018 break; 3019 3020 /* 3021 * In LAST_ACK, we may still be waiting for data to drain 3022 * and/or to be acked, as well as for the ack of our FIN. 3023 * If our FIN is now acknowledged, delete the TCB, 3024 * enter the closed state and return. 3025 */ 3026 case TCPS_LAST_ACK: 3027 if (ourfinisacked) { 3028 tp = tcp_close(tp); 3029 goto drop; 3030 } 3031 break; 3032 } 3033 } 3034 3035step6: 3036 INP_WLOCK_ASSERT(tp->t_inpcb); 3037 3038 /* 3039 * Update window information. 3040 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3041 */ 3042 if ((thflags & TH_ACK) && 3043 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3044 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3045 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3046 /* keep track of pure window updates */ 3047 if (tlen == 0 && 3048 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3049 TCPSTAT_INC(tcps_rcvwinupd); 3050 tp->snd_wnd = tiwin; 3051 tp->snd_wl1 = th->th_seq; 3052 tp->snd_wl2 = th->th_ack; 3053 if (tp->snd_wnd > tp->max_sndwnd) 3054 tp->max_sndwnd = tp->snd_wnd; 3055 needoutput = 1; 3056 } 3057 3058 /* 3059 * Process segments with URG. 3060 */ 3061 if ((thflags & TH_URG) && th->th_urp && 3062 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3063 /* 3064 * This is a kludge, but if we receive and accept 3065 * random urgent pointers, we'll crash in 3066 * soreceive. It's hard to imagine someone 3067 * actually wanting to send this much urgent data. 3068 */ 3069 SOCKBUF_LOCK(&so->so_rcv); 3070 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3071 th->th_urp = 0; /* XXX */ 3072 thflags &= ~TH_URG; /* XXX */ 3073 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3074 goto dodata; /* XXX */ 3075 } 3076 /* 3077 * If this segment advances the known urgent pointer, 3078 * then mark the data stream. This should not happen 3079 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3080 * a FIN has been received from the remote side. 3081 * In these states we ignore the URG. 3082 * 3083 * According to RFC961 (Assigned Protocols), 3084 * the urgent pointer points to the last octet 3085 * of urgent data. We continue, however, 3086 * to consider it to indicate the first octet 3087 * of data past the urgent section as the original 3088 * spec states (in one of two places). 3089 */ 3090 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3091 tp->rcv_up = th->th_seq + th->th_urp; 3092 so->so_oobmark = sbavail(&so->so_rcv) + 3093 (tp->rcv_up - tp->rcv_nxt) - 1; 3094 if (so->so_oobmark == 0) 3095 so->so_rcv.sb_state |= SBS_RCVATMARK; 3096 sohasoutofband(so); 3097 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3098 } 3099 SOCKBUF_UNLOCK(&so->so_rcv); 3100 /* 3101 * Remove out of band data so doesn't get presented to user. 3102 * This can happen independent of advancing the URG pointer, 3103 * but if two URG's are pending at once, some out-of-band 3104 * data may creep in... ick. 3105 */ 3106 if (th->th_urp <= (uint32_t)tlen && 3107 !(so->so_options & SO_OOBINLINE)) { 3108 /* hdr drop is delayed */ 3109 tcp_pulloutofband(so, th, m, drop_hdrlen); 3110 } 3111 } else { 3112 /* 3113 * If no out of band data is expected, 3114 * pull receive urgent pointer along 3115 * with the receive window. 3116 */ 3117 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3118 tp->rcv_up = tp->rcv_nxt; 3119 } 3120dodata: /* XXX */ 3121 INP_WLOCK_ASSERT(tp->t_inpcb); 3122 3123 /* 3124 * Process the segment text, merging it into the TCP sequencing queue, 3125 * and arranging for acknowledgment of receipt if necessary. 3126 * This process logically involves adjusting tp->rcv_wnd as data 3127 * is presented to the user (this happens in tcp_usrreq.c, 3128 * case PRU_RCVD). If a FIN has already been received on this 3129 * connection then we just ignore the text. 3130 */ 3131 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3132 IS_FASTOPEN(tp->t_flags)); 3133 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3134 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3135 tcp_seq save_start = th->th_seq; 3136 tcp_seq save_rnxt = tp->rcv_nxt; 3137 int save_tlen = tlen; 3138 m_adj(m, drop_hdrlen); /* delayed header drop */ 3139 /* 3140 * Insert segment which includes th into TCP reassembly queue 3141 * with control block tp. Set thflags to whether reassembly now 3142 * includes a segment with FIN. This handles the common case 3143 * inline (segment is the next to be received on an established 3144 * connection, and the queue is empty), avoiding linkage into 3145 * and removal from the queue and repetition of various 3146 * conversions. 3147 * Set DELACK for segments received in order, but ack 3148 * immediately when segments are out of order (so 3149 * fast retransmit can work). 3150 */ 3151 if (th->th_seq == tp->rcv_nxt && 3152 SEGQ_EMPTY(tp) && 3153 (TCPS_HAVEESTABLISHED(tp->t_state) || 3154 tfo_syn)) { 3155 if (DELAY_ACK(tp, tlen) || tfo_syn) 3156 tp->t_flags |= TF_DELACK; 3157 else 3158 tp->t_flags |= TF_ACKNOW; 3159 tp->rcv_nxt += tlen; 3160 if (tlen && 3161 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3162 (tp->t_fbyte_in == 0)) { 3163 tp->t_fbyte_in = ticks; 3164 if (tp->t_fbyte_in == 0) 3165 tp->t_fbyte_in = 1; 3166 if (tp->t_fbyte_out && tp->t_fbyte_in) 3167 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3168 } 3169 thflags = th->th_flags & TH_FIN; 3170 TCPSTAT_INC(tcps_rcvpack); 3171 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3172 SOCKBUF_LOCK(&so->so_rcv); 3173 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3174 m_freem(m); 3175 else 3176 sbappendstream_locked(&so->so_rcv, m, 0); 3177 tp->t_flags |= TF_WAKESOR; 3178 } else { 3179 /* 3180 * XXX: Due to the header drop above "th" is 3181 * theoretically invalid by now. Fortunately 3182 * m_adj() doesn't actually frees any mbufs 3183 * when trimming from the head. 3184 */ 3185 tcp_seq temp = save_start; 3186 3187 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3188 tp->t_flags |= TF_ACKNOW; 3189 } 3190 if ((tp->t_flags & TF_SACK_PERMIT) && 3191 (save_tlen > 0) && 3192 TCPS_HAVEESTABLISHED(tp->t_state)) { 3193 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3194 /* 3195 * DSACK actually handled in the fastpath 3196 * above. 3197 */ 3198 tcp_update_sack_list(tp, save_start, 3199 save_start + save_tlen); 3200 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3201 if ((tp->rcv_numsacks >= 1) && 3202 (tp->sackblks[0].end == save_start)) { 3203 /* 3204 * Partial overlap, recorded at todrop 3205 * above. 3206 */ 3207 tcp_update_sack_list(tp, 3208 tp->sackblks[0].start, 3209 tp->sackblks[0].end); 3210 } else { 3211 tcp_update_dsack_list(tp, save_start, 3212 save_start + save_tlen); 3213 } 3214 } else if (tlen >= save_tlen) { 3215 /* Update of sackblks. */ 3216 tcp_update_dsack_list(tp, save_start, 3217 save_start + save_tlen); 3218 } else if (tlen > 0) { 3219 tcp_update_dsack_list(tp, save_start, 3220 save_start + tlen); 3221 } 3222 } 3223 tcp_handle_wakeup(tp, so); 3224#if 0 3225 /* 3226 * Note the amount of data that peer has sent into 3227 * our window, in order to estimate the sender's 3228 * buffer size. 3229 * XXX: Unused. 3230 */ 3231 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3232 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3233 else 3234 len = so->so_rcv.sb_hiwat; 3235#endif 3236 } else { 3237 m_freem(m); 3238 thflags &= ~TH_FIN; 3239 } 3240 3241 /* 3242 * If FIN is received ACK the FIN and let the user know 3243 * that the connection is closing. 3244 */ 3245 if (thflags & TH_FIN) { 3246 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3247 /* The socket upcall is handled by socantrcvmore. */ 3248 socantrcvmore(so); 3249 /* 3250 * If connection is half-synchronized 3251 * (ie NEEDSYN flag on) then delay ACK, 3252 * so it may be piggybacked when SYN is sent. 3253 * Otherwise, since we received a FIN then no 3254 * more input can be expected, send ACK now. 3255 */ 3256 if (tp->t_flags & TF_NEEDSYN) 3257 tp->t_flags |= TF_DELACK; 3258 else 3259 tp->t_flags |= TF_ACKNOW; 3260 tp->rcv_nxt++; 3261 } 3262 switch (tp->t_state) { 3263 /* 3264 * In SYN_RECEIVED and ESTABLISHED STATES 3265 * enter the CLOSE_WAIT state. 3266 */ 3267 case TCPS_SYN_RECEIVED: 3268 tp->t_starttime = ticks; 3269 /* FALLTHROUGH */ 3270 case TCPS_ESTABLISHED: 3271 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3272 break; 3273 3274 /* 3275 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3276 * enter the CLOSING state. 3277 */ 3278 case TCPS_FIN_WAIT_1: 3279 tcp_state_change(tp, TCPS_CLOSING); 3280 break; 3281 3282 /* 3283 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3284 * starting the time-wait timer, turning off the other 3285 * standard timers. 3286 */ 3287 case TCPS_FIN_WAIT_2: 3288 tcp_twstart(tp); 3289 return; 3290 } 3291 } 3292#ifdef TCPDEBUG 3293 if (so->so_options & SO_DEBUG) 3294 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3295 &tcp_savetcp, 0); 3296#endif 3297 TCP_PROBE3(debug__input, tp, th, m); 3298 3299 /* 3300 * Return any desired output. 3301 */ 3302 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3303 (void) tp->t_fb->tfb_tcp_output(tp); 3304 3305check_delack: 3306 INP_WLOCK_ASSERT(tp->t_inpcb); 3307 3308 if (tp->t_flags & TF_DELACK) { 3309 tp->t_flags &= ~TF_DELACK; 3310 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3311 } 3312 INP_WUNLOCK(tp->t_inpcb); 3313 return; 3314 3315dropafterack: 3316 /* 3317 * Generate an ACK dropping incoming segment if it occupies 3318 * sequence space, where the ACK reflects our state. 3319 * 3320 * We can now skip the test for the RST flag since all 3321 * paths to this code happen after packets containing 3322 * RST have been dropped. 3323 * 3324 * In the SYN-RECEIVED state, don't send an ACK unless the 3325 * segment we received passes the SYN-RECEIVED ACK test. 3326 * If it fails send a RST. This breaks the loop in the 3327 * "LAND" DoS attack, and also prevents an ACK storm 3328 * between two listening ports that have been sent forged 3329 * SYN segments, each with the source address of the other. 3330 */ 3331 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3332 (SEQ_GT(tp->snd_una, th->th_ack) || 3333 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3334 rstreason = BANDLIM_RST_OPENPORT; 3335 goto dropwithreset; 3336 } 3337#ifdef TCPDEBUG 3338 if (so->so_options & SO_DEBUG) 3339 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3340 &tcp_savetcp, 0); 3341#endif 3342 TCP_PROBE3(debug__input, tp, th, m); 3343 tp->t_flags |= TF_ACKNOW; 3344 (void) tp->t_fb->tfb_tcp_output(tp); 3345 INP_WUNLOCK(tp->t_inpcb); 3346 m_freem(m); 3347 return; 3348 3349dropwithreset: 3350 if (tp != NULL) { 3351 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3352 INP_WUNLOCK(tp->t_inpcb); 3353 } else 3354 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3355 return; 3356 3357drop: 3358 /* 3359 * Drop space held by incoming segment and return. 3360 */ 3361#ifdef TCPDEBUG 3362 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3363 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3364 &tcp_savetcp, 0); 3365#endif 3366 TCP_PROBE3(debug__input, tp, th, m); 3367 if (tp != NULL) { 3368 INP_WUNLOCK(tp->t_inpcb); 3369 } 3370 m_freem(m); 3371} 3372 3373/* 3374 * Issue RST and make ACK acceptable to originator of segment. 3375 * The mbuf must still include the original packet header. 3376 * tp may be NULL. 3377 */ 3378void 3379tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3380 int tlen, int rstreason) 3381{ 3382#ifdef INET 3383 struct ip *ip; 3384#endif 3385#ifdef INET6 3386 struct ip6_hdr *ip6; 3387#endif 3388 3389 if (tp != NULL) { 3390 INP_WLOCK_ASSERT(tp->t_inpcb); 3391 } 3392 3393 /* Don't bother if destination was broadcast/multicast. */ 3394 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3395 goto drop; 3396#ifdef INET6 3397 if (mtod(m, struct ip *)->ip_v == 6) { 3398 ip6 = mtod(m, struct ip6_hdr *); 3399 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3400 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3401 goto drop; 3402 /* IPv6 anycast check is done at tcp6_input() */ 3403 } 3404#endif 3405#if defined(INET) && defined(INET6) 3406 else 3407#endif 3408#ifdef INET 3409 { 3410 ip = mtod(m, struct ip *); 3411 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3412 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3413 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3414 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3415 goto drop; 3416 } 3417#endif 3418 3419 /* Perform bandwidth limiting. */ 3420 if (badport_bandlim(rstreason) < 0) 3421 goto drop; 3422 3423 /* tcp_respond consumes the mbuf chain. */ 3424 if (th->th_flags & TH_ACK) { 3425 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3426 th->th_ack, TH_RST); 3427 } else { 3428 if (th->th_flags & TH_SYN) 3429 tlen++; 3430 if (th->th_flags & TH_FIN) 3431 tlen++; 3432 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3433 (tcp_seq)0, TH_RST|TH_ACK); 3434 } 3435 return; 3436drop: 3437 m_freem(m); 3438} 3439 3440/* 3441 * Parse TCP options and place in tcpopt. 3442 */ 3443void 3444tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3445{ 3446 int opt, optlen; 3447 3448 to->to_flags = 0; 3449 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3450 opt = cp[0]; 3451 if (opt == TCPOPT_EOL) 3452 break; 3453 if (opt == TCPOPT_NOP) 3454 optlen = 1; 3455 else { 3456 if (cnt < 2) 3457 break; 3458 optlen = cp[1]; 3459 if (optlen < 2 || optlen > cnt) 3460 break; 3461 } 3462 switch (opt) { 3463 case TCPOPT_MAXSEG: 3464 if (optlen != TCPOLEN_MAXSEG) 3465 continue; 3466 if (!(flags & TO_SYN)) 3467 continue; 3468 to->to_flags |= TOF_MSS; 3469 bcopy((char *)cp + 2, 3470 (char *)&to->to_mss, sizeof(to->to_mss)); 3471 to->to_mss = ntohs(to->to_mss); 3472 break; 3473 case TCPOPT_WINDOW: 3474 if (optlen != TCPOLEN_WINDOW) 3475 continue; 3476 if (!(flags & TO_SYN)) 3477 continue; 3478 to->to_flags |= TOF_SCALE; 3479 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3480 break; 3481 case TCPOPT_TIMESTAMP: 3482 if (optlen != TCPOLEN_TIMESTAMP) 3483 continue; 3484 to->to_flags |= TOF_TS; 3485 bcopy((char *)cp + 2, 3486 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3487 to->to_tsval = ntohl(to->to_tsval); 3488 bcopy((char *)cp + 6, 3489 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3490 to->to_tsecr = ntohl(to->to_tsecr); 3491 break; 3492 case TCPOPT_SIGNATURE: 3493 /* 3494 * In order to reply to a host which has set the 3495 * TCP_SIGNATURE option in its initial SYN, we have 3496 * to record the fact that the option was observed 3497 * here for the syncache code to perform the correct 3498 * response. 3499 */ 3500 if (optlen != TCPOLEN_SIGNATURE) 3501 continue; 3502 to->to_flags |= TOF_SIGNATURE; 3503 to->to_signature = cp + 2; 3504 break; 3505 case TCPOPT_SACK_PERMITTED: 3506 if (optlen != TCPOLEN_SACK_PERMITTED) 3507 continue; 3508 if (!(flags & TO_SYN)) 3509 continue; 3510 if (!V_tcp_do_sack) 3511 continue; 3512 to->to_flags |= TOF_SACKPERM; 3513 break; 3514 case TCPOPT_SACK: 3515 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3516 continue; 3517 if (flags & TO_SYN) 3518 continue; 3519 to->to_flags |= TOF_SACK; 3520 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3521 to->to_sacks = cp + 2; 3522 TCPSTAT_INC(tcps_sack_rcv_blocks); 3523 break; 3524 case TCPOPT_FAST_OPEN: 3525 /* 3526 * Cookie length validation is performed by the 3527 * server side cookie checking code or the client 3528 * side cookie cache update code. 3529 */ 3530 if (!(flags & TO_SYN)) 3531 continue; 3532 if (!V_tcp_fastopen_client_enable && 3533 !V_tcp_fastopen_server_enable) 3534 continue; 3535 to->to_flags |= TOF_FASTOPEN; 3536 to->to_tfo_len = optlen - 2; 3537 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3538 break; 3539 default: 3540 continue; 3541 } 3542 } 3543} 3544 3545/* 3546 * Pull out of band byte out of a segment so 3547 * it doesn't appear in the user's data queue. 3548 * It is still reflected in the segment length for 3549 * sequencing purposes. 3550 */ 3551void 3552tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3553 int off) 3554{ 3555 int cnt = off + th->th_urp - 1; 3556 3557 while (cnt >= 0) { 3558 if (m->m_len > cnt) { 3559 char *cp = mtod(m, caddr_t) + cnt; 3560 struct tcpcb *tp = sototcpcb(so); 3561 3562 INP_WLOCK_ASSERT(tp->t_inpcb); 3563 3564 tp->t_iobc = *cp; 3565 tp->t_oobflags |= TCPOOB_HAVEDATA; 3566 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3567 m->m_len--; 3568 if (m->m_flags & M_PKTHDR) 3569 m->m_pkthdr.len--; 3570 return; 3571 } 3572 cnt -= m->m_len; 3573 m = m->m_next; 3574 if (m == NULL) 3575 break; 3576 } 3577 panic("tcp_pulloutofband"); 3578} 3579 3580/* 3581 * Collect new round-trip time estimate 3582 * and update averages and current timeout. 3583 */ 3584void 3585tcp_xmit_timer(struct tcpcb *tp, int rtt) 3586{ 3587 int delta; 3588 3589 INP_WLOCK_ASSERT(tp->t_inpcb); 3590 3591 TCPSTAT_INC(tcps_rttupdated); 3592 tp->t_rttupdated++; 3593#ifdef STATS 3594 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3595 imax(0, rtt * 1000 / hz)); 3596#endif 3597 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3598 /* 3599 * srtt is stored as fixed point with 5 bits after the 3600 * binary point (i.e., scaled by 8). The following magic 3601 * is equivalent to the smoothing algorithm in rfc793 with 3602 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3603 * point). Adjust rtt to origin 0. 3604 */ 3605 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3606 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3607 3608 if ((tp->t_srtt += delta) <= 0) 3609 tp->t_srtt = 1; 3610 3611 /* 3612 * We accumulate a smoothed rtt variance (actually, a 3613 * smoothed mean difference), then set the retransmit 3614 * timer to smoothed rtt + 4 times the smoothed variance. 3615 * rttvar is stored as fixed point with 4 bits after the 3616 * binary point (scaled by 16). The following is 3617 * equivalent to rfc793 smoothing with an alpha of .75 3618 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3619 * rfc793's wired-in beta. 3620 */ 3621 if (delta < 0) 3622 delta = -delta; 3623 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3624 if ((tp->t_rttvar += delta) <= 0) 3625 tp->t_rttvar = 1; 3626 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3627 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3628 } else { 3629 /* 3630 * No rtt measurement yet - use the unsmoothed rtt. 3631 * Set the variance to half the rtt (so our first 3632 * retransmit happens at 3*rtt). 3633 */ 3634 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3635 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3636 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3637 } 3638 tp->t_rtttime = 0; 3639 tp->t_rxtshift = 0; 3640 3641 /* 3642 * the retransmit should happen at rtt + 4 * rttvar. 3643 * Because of the way we do the smoothing, srtt and rttvar 3644 * will each average +1/2 tick of bias. When we compute 3645 * the retransmit timer, we want 1/2 tick of rounding and 3646 * 1 extra tick because of +-1/2 tick uncertainty in the 3647 * firing of the timer. The bias will give us exactly the 3648 * 1.5 tick we need. But, because the bias is 3649 * statistical, we have to test that we don't drop below 3650 * the minimum feasible timer (which is 2 ticks). 3651 */ 3652 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3653 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3654 3655 /* 3656 * We received an ack for a packet that wasn't retransmitted; 3657 * it is probably safe to discard any error indications we've 3658 * received recently. This isn't quite right, but close enough 3659 * for now (a route might have failed after we sent a segment, 3660 * and the return path might not be symmetrical). 3661 */ 3662 tp->t_softerror = 0; 3663} 3664 3665/* 3666 * Determine a reasonable value for maxseg size. 3667 * If the route is known, check route for mtu. 3668 * If none, use an mss that can be handled on the outgoing interface 3669 * without forcing IP to fragment. If no route is found, route has no mtu, 3670 * or the destination isn't local, use a default, hopefully conservative 3671 * size (usually 512 or the default IP max size, but no more than the mtu 3672 * of the interface), as we can't discover anything about intervening 3673 * gateways or networks. We also initialize the congestion/slow start 3674 * window to be a single segment if the destination isn't local. 3675 * While looking at the routing entry, we also initialize other path-dependent 3676 * parameters from pre-set or cached values in the routing entry. 3677 * 3678 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3679 * IP options, e.g. IPSEC data, since length of this data may vary, and 3680 * thus it is calculated for every segment separately in tcp_output(). 3681 * 3682 * NOTE that this routine is only called when we process an incoming 3683 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3684 * settings are handled in tcp_mssopt(). 3685 */ 3686void 3687tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3688 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3689{ 3690 int mss = 0; 3691 uint32_t maxmtu = 0; 3692 struct inpcb *inp = tp->t_inpcb; 3693 struct hc_metrics_lite metrics; 3694#ifdef INET6 3695 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3696 size_t min_protoh = isipv6 ? 3697 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3698 sizeof (struct tcpiphdr); 3699#else 3700 size_t min_protoh = sizeof(struct tcpiphdr); 3701#endif 3702 3703 INP_WLOCK_ASSERT(tp->t_inpcb); 3704 3705 if (tp->t_port) 3706 min_protoh += V_tcp_udp_tunneling_overhead; 3707 if (mtuoffer != -1) { 3708 KASSERT(offer == -1, ("%s: conflict", __func__)); 3709 offer = mtuoffer - min_protoh; 3710 } 3711 3712 /* Initialize. */ 3713#ifdef INET6 3714 if (isipv6) { 3715 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3716 tp->t_maxseg = V_tcp_v6mssdflt; 3717 } 3718#endif 3719#if defined(INET) && defined(INET6) 3720 else 3721#endif 3722#ifdef INET 3723 { 3724 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3725 tp->t_maxseg = V_tcp_mssdflt; 3726 } 3727#endif 3728 3729 /* 3730 * No route to sender, stay with default mss and return. 3731 */ 3732 if (maxmtu == 0) { 3733 /* 3734 * In case we return early we need to initialize metrics 3735 * to a defined state as tcp_hc_get() would do for us 3736 * if there was no cache hit. 3737 */ 3738 if (metricptr != NULL) 3739 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3740 return; 3741 } 3742 3743 /* What have we got? */ 3744 switch (offer) { 3745 case 0: 3746 /* 3747 * Offer == 0 means that there was no MSS on the SYN 3748 * segment, in this case we use tcp_mssdflt as 3749 * already assigned to t_maxseg above. 3750 */ 3751 offer = tp->t_maxseg; 3752 break; 3753 3754 case -1: 3755 /* 3756 * Offer == -1 means that we didn't receive SYN yet. 3757 */ 3758 /* FALLTHROUGH */ 3759 3760 default: 3761 /* 3762 * Prevent DoS attack with too small MSS. Round up 3763 * to at least minmss. 3764 */ 3765 offer = max(offer, V_tcp_minmss); 3766 } 3767 3768 /* 3769 * rmx information is now retrieved from tcp_hostcache. 3770 */ 3771 tcp_hc_get(&inp->inp_inc, &metrics); 3772 if (metricptr != NULL) 3773 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3774 3775 /* 3776 * If there's a discovered mtu in tcp hostcache, use it. 3777 * Else, use the link mtu. 3778 */ 3779 if (metrics.rmx_mtu) 3780 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3781 else { 3782#ifdef INET6 3783 if (isipv6) { 3784 mss = maxmtu - min_protoh; 3785 if (!V_path_mtu_discovery && 3786 !in6_localaddr(&inp->in6p_faddr)) 3787 mss = min(mss, V_tcp_v6mssdflt); 3788 } 3789#endif 3790#if defined(INET) && defined(INET6) 3791 else 3792#endif 3793#ifdef INET 3794 { 3795 mss = maxmtu - min_protoh; 3796 if (!V_path_mtu_discovery && 3797 !in_localaddr(inp->inp_faddr)) 3798 mss = min(mss, V_tcp_mssdflt); 3799 } 3800#endif 3801 /* 3802 * XXX - The above conditional (mss = maxmtu - min_protoh) 3803 * probably violates the TCP spec. 3804 * The problem is that, since we don't know the 3805 * other end's MSS, we are supposed to use a conservative 3806 * default. But, if we do that, then MTU discovery will 3807 * never actually take place, because the conservative 3808 * default is much less than the MTUs typically seen 3809 * on the Internet today. For the moment, we'll sweep 3810 * this under the carpet. 3811 * 3812 * The conservative default might not actually be a problem 3813 * if the only case this occurs is when sending an initial 3814 * SYN with options and data to a host we've never talked 3815 * to before. Then, they will reply with an MSS value which 3816 * will get recorded and the new parameters should get 3817 * recomputed. For Further Study. 3818 */ 3819 } 3820 mss = min(mss, offer); 3821 3822 /* 3823 * Sanity check: make sure that maxseg will be large 3824 * enough to allow some data on segments even if the 3825 * all the option space is used (40bytes). Otherwise 3826 * funny things may happen in tcp_output. 3827 * 3828 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3829 */ 3830 mss = max(mss, 64); 3831 3832 tp->t_maxseg = mss; 3833} 3834 3835void 3836tcp_mss(struct tcpcb *tp, int offer) 3837{ 3838 int mss; 3839 uint32_t bufsize; 3840 struct inpcb *inp; 3841 struct socket *so; 3842 struct hc_metrics_lite metrics; 3843 struct tcp_ifcap cap; 3844 3845 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3846 3847 bzero(&cap, sizeof(cap)); 3848 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3849 3850 mss = tp->t_maxseg; 3851 inp = tp->t_inpcb; 3852 3853 /* 3854 * If there's a pipesize, change the socket buffer to that size, 3855 * don't change if sb_hiwat is different than default (then it 3856 * has been changed on purpose with setsockopt). 3857 * Make the socket buffers an integral number of mss units; 3858 * if the mss is larger than the socket buffer, decrease the mss. 3859 */ 3860 so = inp->inp_socket; 3861 SOCKBUF_LOCK(&so->so_snd); 3862 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3863 bufsize = metrics.rmx_sendpipe; 3864 else 3865 bufsize = so->so_snd.sb_hiwat; 3866 if (bufsize < mss) 3867 mss = bufsize; 3868 else { 3869 bufsize = roundup(bufsize, mss); 3870 if (bufsize > sb_max) 3871 bufsize = sb_max; 3872 if (bufsize > so->so_snd.sb_hiwat) 3873 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3874 } 3875 SOCKBUF_UNLOCK(&so->so_snd); 3876 /* 3877 * Sanity check: make sure that maxseg will be large 3878 * enough to allow some data on segments even if the 3879 * all the option space is used (40bytes). Otherwise 3880 * funny things may happen in tcp_output. 3881 * 3882 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3883 */ 3884 tp->t_maxseg = max(mss, 64); 3885 3886 SOCKBUF_LOCK(&so->so_rcv); 3887 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3888 bufsize = metrics.rmx_recvpipe; 3889 else 3890 bufsize = so->so_rcv.sb_hiwat; 3891 if (bufsize > mss) { 3892 bufsize = roundup(bufsize, mss); 3893 if (bufsize > sb_max) 3894 bufsize = sb_max; 3895 if (bufsize > so->so_rcv.sb_hiwat) 3896 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3897 } 3898 SOCKBUF_UNLOCK(&so->so_rcv); 3899 3900 /* Check the interface for TSO capabilities. */ 3901 if (cap.ifcap & CSUM_TSO) { 3902 tp->t_flags |= TF_TSO; 3903 tp->t_tsomax = cap.tsomax; 3904 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3905 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3906 } 3907} 3908 3909/* 3910 * Determine the MSS option to send on an outgoing SYN. 3911 */ 3912int 3913tcp_mssopt(struct in_conninfo *inc) 3914{ 3915 int mss = 0; 3916 uint32_t thcmtu = 0; 3917 uint32_t maxmtu = 0; 3918 size_t min_protoh; 3919 3920 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3921 3922#ifdef INET6 3923 if (inc->inc_flags & INC_ISIPV6) { 3924 mss = V_tcp_v6mssdflt; 3925 maxmtu = tcp_maxmtu6(inc, NULL); 3926 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3927 } 3928#endif 3929#if defined(INET) && defined(INET6) 3930 else 3931#endif 3932#ifdef INET 3933 { 3934 mss = V_tcp_mssdflt; 3935 maxmtu = tcp_maxmtu(inc, NULL); 3936 min_protoh = sizeof(struct tcpiphdr); 3937 } 3938#endif 3939#if defined(INET6) || defined(INET) 3940 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3941#endif 3942 3943 if (maxmtu && thcmtu) 3944 mss = min(maxmtu, thcmtu) - min_protoh; 3945 else if (maxmtu || thcmtu) 3946 mss = max(maxmtu, thcmtu) - min_protoh; 3947 3948 return (mss); 3949} 3950 3951void 3952tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3953{ 3954 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3955 int maxseg = tcp_maxseg(tp); 3956 3957 INP_WLOCK_ASSERT(tp->t_inpcb); 3958 3959 /* 3960 * Compute the amount of data that this ACK is indicating 3961 * (del_data) and an estimate of how many bytes are in the 3962 * network. 3963 */ 3964 del_data = tp->sackhint.delivered_data; 3965 if (V_tcp_do_rfc6675_pipe) 3966 pipe = tcp_compute_pipe(tp); 3967 else 3968 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; 3969 tp->sackhint.prr_delivered += del_data; 3970 /* 3971 * Proportional Rate Reduction 3972 */ 3973 if (pipe >= tp->snd_ssthresh) { 3974 if (tp->sackhint.recover_fs == 0) 3975 tp->sackhint.recover_fs = 3976 imax(1, tp->snd_nxt - tp->snd_una); 3977 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3978 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3979 tp->sackhint.prr_out; 3980 } else { 3981 if (V_tcp_do_prr_conservative) 3982 limit = tp->sackhint.prr_delivered - 3983 tp->sackhint.prr_out; 3984 else 3985 limit = imax(tp->sackhint.prr_delivered - 3986 tp->sackhint.prr_out, del_data) + 3987 maxseg; 3988 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 3989 } 3990 snd_cnt = imax(snd_cnt, 0) / maxseg; 3991 /* 3992 * Send snd_cnt new data into the network in response to this ack. 3993 * If there is going to be a SACK retransmission, adjust snd_cwnd 3994 * accordingly. 3995 */ 3996 if (IN_FASTRECOVERY(tp->t_flags)) { 3997 tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover + 3998 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg)); 3999 } else if (IN_CONGRECOVERY(tp->t_flags)) 4000 tp->snd_cwnd = imax(maxseg, pipe - del_data + 4001 (snd_cnt * maxseg)); 4002} 4003 4004/* 4005 * On a partial ack arrives, force the retransmission of the 4006 * next unacknowledged segment. Do not clear tp->t_dupacks. 4007 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4008 * be started again. 4009 */ 4010void 4011tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4012{ 4013 tcp_seq onxt = tp->snd_nxt; 4014 uint32_t ocwnd = tp->snd_cwnd; 4015 u_int maxseg = tcp_maxseg(tp); 4016 4017 INP_WLOCK_ASSERT(tp->t_inpcb); 4018 4019 tcp_timer_activate(tp, TT_REXMT, 0); 4020 tp->t_rtttime = 0; 4021 tp->snd_nxt = th->th_ack; 4022 /* 4023 * Set snd_cwnd to one segment beyond acknowledged offset. 4024 * (tp->snd_una has not yet been updated when this function is called.) 4025 */ 4026 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4027 tp->t_flags |= TF_ACKNOW; 4028 (void) tp->t_fb->tfb_tcp_output(tp); 4029 tp->snd_cwnd = ocwnd; 4030 if (SEQ_GT(onxt, tp->snd_nxt)) 4031 tp->snd_nxt = onxt; 4032 /* 4033 * Partial window deflation. Relies on fact that tp->snd_una 4034 * not updated yet. 4035 */ 4036 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4037 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4038 else 4039 tp->snd_cwnd = 0; 4040 tp->snd_cwnd += maxseg; 4041} 4042 4043int 4044tcp_compute_pipe(struct tcpcb *tp) 4045{ 4046 return (tp->snd_max - tp->snd_una + 4047 tp->sackhint.sack_bytes_rexmit - 4048 tp->sackhint.sacked_bytes); 4049} 4050 4051uint32_t 4052tcp_compute_initwnd(uint32_t maxseg) 4053{ 4054 /* 4055 * Calculate the Initial Window, also used as Restart Window 4056 * 4057 * RFC5681 Section 3.1 specifies the default conservative values. 4058 * RFC3390 specifies slightly more aggressive values. 4059 * RFC6928 increases it to ten segments. 4060 * Support for user specified value for initial flight size. 4061 */ 4062 if (V_tcp_initcwnd_segments) 4063 return min(V_tcp_initcwnd_segments * maxseg, 4064 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4065 else if (V_tcp_do_rfc3390) 4066 return min(4 * maxseg, max(2 * maxseg, 4380)); 4067 else { 4068 /* Per RFC5681 Section 3.1 */ 4069 if (maxseg > 2190) 4070 return (2 * maxseg); 4071 else if (maxseg > 1095) 4072 return (3 * maxseg); 4073 else 4074 return (4 * maxseg); 4075 } 4076} 4077