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 126 127 #include <netipsec/ipsec_support.h> 128 129 #include <machine/in_cksum.h> 130 131 #include <security/mac/mac_framework.h> 132 133 const int tcprexmtthresh = 3; 134 135 VNET_DEFINE(int, tcp_log_in_vain) = 0; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 137 &VNET_NAME(tcp_log_in_vain), 0, 138 "Log all incoming TCP segments to closed ports"); 139 140 VNET_DEFINE(int, blackhole) = 0; 141 #define V_blackhole VNET(blackhole) 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 143 &VNET_NAME(blackhole), 0, 144 "Do not send RST on segments to closed ports"); 145 146 VNET_DEFINE(int, tcp_delack_enabled) = 1; 147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 148 &VNET_NAME(tcp_delack_enabled), 0, 149 "Delay ACK to try and piggyback it onto a data packet"); 150 151 VNET_DEFINE(int, drop_synfin) = 0; 152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 153 &VNET_NAME(drop_synfin), 0, 154 "Drop TCP packets with SYN+FIN set"); 155 156 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 158 &VNET_NAME(tcp_do_prr_conservative), 0, 159 "Do conservative Proportional Rate Reduction"); 160 161 VNET_DEFINE(int, tcp_do_prr) = 1; 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 163 &VNET_NAME(tcp_do_prr), 1, 164 "Enable Proportional Rate Reduction per RFC 6937"); 165 166 VNET_DEFINE(int, tcp_do_newcwv) = 0; 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 168 &VNET_NAME(tcp_do_newcwv), 0, 169 "Enable New Congestion Window Validation per RFC7661"); 170 171 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 173 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 174 "Use calculated pipe/in-flight bytes per RFC 6675"); 175 176 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 178 &VNET_NAME(tcp_do_rfc3042), 0, 179 "Enable RFC 3042 (Limited Transmit)"); 180 181 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 183 &VNET_NAME(tcp_do_rfc3390), 0, 184 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 185 186 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 188 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 189 "Slow-start flight size (initial congestion window) in number of segments"); 190 191 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_do_rfc3465), 0, 194 "Enable RFC 3465 (Appropriate Byte Counting)"); 195 196 VNET_DEFINE(int, tcp_abc_l_var) = 2; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_abc_l_var), 2, 199 "Cap the max cwnd increment during slow-start to this number of segments"); 200 201 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 202 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 203 "TCP ECN"); 204 205 VNET_DEFINE(int, tcp_do_ecn) = 2; 206 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(tcp_do_ecn), 0, 208 "TCP ECN support"); 209 210 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 211 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_ecn_maxretries), 0, 213 "Max retries before giving up on ECN"); 214 215 VNET_DEFINE(int, tcp_insecure_syn) = 0; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_insecure_syn), 0, 218 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 219 220 VNET_DEFINE(int, tcp_insecure_rst) = 0; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_insecure_rst), 0, 223 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 224 225 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 226 #define V_tcp_recvspace VNET(tcp_recvspace) 227 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 229 230 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 232 &VNET_NAME(tcp_do_autorcvbuf), 0, 233 "Enable automatic receive buffer sizing"); 234 235 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 237 &VNET_NAME(tcp_autorcvbuf_max), 0, 238 "Max size of automatic receive buffer"); 239 240 VNET_DEFINE(struct inpcbhead, tcb); 241 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 242 VNET_DEFINE(struct inpcbinfo, tcbinfo); 243 244 /* 245 * TCP statistics are stored in an array of counter(9)s, which size matches 246 * size of struct tcpstat. TCP running connection count is a regular array. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 250 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 251 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 252 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 253 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 254 "TCP connection counts by TCP state"); 255 256 static void 257 tcp_vnet_init(const void *unused) 258 { 259 260 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 261 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 262 } 263 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 264 tcp_vnet_init, NULL); 265 266 #ifdef VIMAGE 267 static void 268 tcp_vnet_uninit(const void *unused) 269 { 270 271 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 272 VNET_PCPUSTAT_FREE(tcpstat); 273 } 274 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 275 tcp_vnet_uninit, NULL); 276 #endif /* VIMAGE */ 277 278 /* 279 * Kernel module interface for updating tcpstat. The first argument is an index 280 * into tcpstat treated as an array. 281 */ 282 void 283 kmod_tcpstat_add(int statnum, int val) 284 { 285 286 counter_u64_add(VNET(tcpstat)[statnum], val); 287 } 288 289 #ifdef TCP_HHOOK 290 /* 291 * Wrapper for the TCP established input helper hook. 292 */ 293 void 294 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 295 { 296 struct tcp_hhook_data hhook_data; 297 298 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 299 hhook_data.tp = tp; 300 hhook_data.th = th; 301 hhook_data.to = to; 302 303 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 304 tp->osd); 305 } 306 } 307 #endif 308 309 /* 310 * CC wrapper hook functions 311 */ 312 void 313 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 314 uint16_t type) 315 { 316 #ifdef STATS 317 int32_t gput; 318 #endif 319 320 INP_WLOCK_ASSERT(tp->t_inpcb); 321 322 tp->ccv->nsegs = nsegs; 323 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 324 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 325 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 326 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 327 tp->ccv->flags |= CCF_CWND_LIMITED; 328 else 329 tp->ccv->flags &= ~CCF_CWND_LIMITED; 330 331 if (type == CC_ACK) { 332 #ifdef STATS 333 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 334 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 335 if (!IN_RECOVERY(tp->t_flags)) 336 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 337 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 338 if ((tp->t_flags & TF_GPUTINPROG) && 339 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 340 /* 341 * Compute goodput in bits per millisecond. 342 */ 343 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) / 344 max(1, tcp_ts_getticks() - tp->gput_ts); 345 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 346 gput); 347 /* 348 * XXXLAS: This is a temporary hack, and should be 349 * chained off VOI_TCP_GPUT when stats(9) grows an API 350 * to deal with chained VOIs. 351 */ 352 if (tp->t_stats_gput_prev > 0) 353 stats_voi_update_abs_s32(tp->t_stats, 354 VOI_TCP_GPUT_ND, 355 ((gput - tp->t_stats_gput_prev) * 100) / 356 tp->t_stats_gput_prev); 357 tp->t_flags &= ~TF_GPUTINPROG; 358 tp->t_stats_gput_prev = gput; 359 } 360 #endif /* STATS */ 361 if (tp->snd_cwnd > tp->snd_ssthresh) { 362 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 363 if (tp->t_bytes_acked >= tp->snd_cwnd) { 364 tp->t_bytes_acked -= tp->snd_cwnd; 365 tp->ccv->flags |= CCF_ABC_SENTAWND; 366 } 367 } else { 368 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 369 tp->t_bytes_acked = 0; 370 } 371 } 372 373 if (CC_ALGO(tp)->ack_received != NULL) { 374 /* XXXLAS: Find a way to live without this */ 375 tp->ccv->curack = th->th_ack; 376 CC_ALGO(tp)->ack_received(tp->ccv, type); 377 } 378 #ifdef STATS 379 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 380 #endif 381 } 382 383 void 384 cc_conn_init(struct tcpcb *tp) 385 { 386 struct hc_metrics_lite metrics; 387 struct inpcb *inp = tp->t_inpcb; 388 u_int maxseg; 389 int rtt; 390 391 INP_WLOCK_ASSERT(tp->t_inpcb); 392 393 tcp_hc_get(&inp->inp_inc, &metrics); 394 maxseg = tcp_maxseg(tp); 395 396 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 397 tp->t_srtt = rtt; 398 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 399 TCPSTAT_INC(tcps_usedrtt); 400 if (metrics.rmx_rttvar) { 401 tp->t_rttvar = metrics.rmx_rttvar; 402 TCPSTAT_INC(tcps_usedrttvar); 403 } else { 404 /* default variation is +- 1 rtt */ 405 tp->t_rttvar = 406 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 407 } 408 TCPT_RANGESET(tp->t_rxtcur, 409 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 410 tp->t_rttmin, TCPTV_REXMTMAX); 411 } 412 if (metrics.rmx_ssthresh) { 413 /* 414 * There's some sort of gateway or interface 415 * buffer limit on the path. Use this to set 416 * the slow start threshold, but set the 417 * threshold to no less than 2*mss. 418 */ 419 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 420 TCPSTAT_INC(tcps_usedssthresh); 421 } 422 423 /* 424 * Set the initial slow-start flight size. 425 * 426 * If a SYN or SYN/ACK was lost and retransmitted, we have to 427 * reduce the initial CWND to one segment as congestion is likely 428 * requiring us to be cautious. 429 */ 430 if (tp->snd_cwnd == 1) 431 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 432 else 433 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 434 435 if (CC_ALGO(tp)->conn_init != NULL) 436 CC_ALGO(tp)->conn_init(tp->ccv); 437 } 438 439 void inline 440 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 441 { 442 INP_WLOCK_ASSERT(tp->t_inpcb); 443 444 #ifdef STATS 445 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 446 #endif 447 448 switch(type) { 449 case CC_NDUPACK: 450 if (!IN_FASTRECOVERY(tp->t_flags)) { 451 tp->snd_recover = tp->snd_max; 452 if (tp->t_flags2 & TF2_ECN_PERMIT) 453 tp->t_flags2 |= TF2_ECN_SND_CWR; 454 } 455 break; 456 case CC_ECN: 457 if (!IN_CONGRECOVERY(tp->t_flags) || 458 /* 459 * Allow ECN reaction on ACK to CWR, if 460 * that data segment was also CE marked. 461 */ 462 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 463 EXIT_CONGRECOVERY(tp->t_flags); 464 TCPSTAT_INC(tcps_ecn_rcwnd); 465 tp->snd_recover = tp->snd_max + 1; 466 if (tp->t_flags2 & TF2_ECN_PERMIT) 467 tp->t_flags2 |= TF2_ECN_SND_CWR; 468 } 469 break; 470 case CC_RTO: 471 tp->t_dupacks = 0; 472 tp->t_bytes_acked = 0; 473 EXIT_RECOVERY(tp->t_flags); 474 if (tp->t_flags2 & TF2_ECN_PERMIT) 475 tp->t_flags2 |= TF2_ECN_SND_CWR; 476 break; 477 case CC_RTO_ERR: 478 TCPSTAT_INC(tcps_sndrexmitbad); 479 /* RTO was unnecessary, so reset everything. */ 480 tp->snd_cwnd = tp->snd_cwnd_prev; 481 tp->snd_ssthresh = tp->snd_ssthresh_prev; 482 tp->snd_recover = tp->snd_recover_prev; 483 if (tp->t_flags & TF_WASFRECOVERY) 484 ENTER_FASTRECOVERY(tp->t_flags); 485 if (tp->t_flags & TF_WASCRECOVERY) 486 ENTER_CONGRECOVERY(tp->t_flags); 487 tp->snd_nxt = tp->snd_max; 488 tp->t_flags &= ~TF_PREVVALID; 489 tp->t_badrxtwin = 0; 490 break; 491 } 492 493 if (CC_ALGO(tp)->cong_signal != NULL) { 494 if (th != NULL) 495 tp->ccv->curack = th->th_ack; 496 CC_ALGO(tp)->cong_signal(tp->ccv, type); 497 } 498 } 499 500 void inline 501 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 502 { 503 INP_WLOCK_ASSERT(tp->t_inpcb); 504 505 /* XXXLAS: KASSERT that we're in recovery? */ 506 507 if (CC_ALGO(tp)->post_recovery != NULL) { 508 tp->ccv->curack = th->th_ack; 509 CC_ALGO(tp)->post_recovery(tp->ccv); 510 } 511 /* XXXLAS: EXIT_RECOVERY ? */ 512 tp->t_bytes_acked = 0; 513 tp->sackhint.prr_out = 0; 514 } 515 516 /* 517 * Indicate whether this ack should be delayed. We can delay the ack if 518 * following conditions are met: 519 * - There is no delayed ack timer in progress. 520 * - Our last ack wasn't a 0-sized window. We never want to delay 521 * the ack that opens up a 0-sized window. 522 * - LRO wasn't used for this segment. We make sure by checking that the 523 * segment size is not larger than the MSS. 524 */ 525 #define DELAY_ACK(tp, tlen) \ 526 ((!tcp_timer_active(tp, TT_DELACK) && \ 527 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 528 (tlen <= tp->t_maxseg) && \ 529 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 530 531 void inline 532 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 533 { 534 INP_WLOCK_ASSERT(tp->t_inpcb); 535 536 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 537 switch (iptos & IPTOS_ECN_MASK) { 538 case IPTOS_ECN_CE: 539 tp->ccv->flags |= CCF_IPHDR_CE; 540 break; 541 case IPTOS_ECN_ECT0: 542 /* FALLTHROUGH */ 543 case IPTOS_ECN_ECT1: 544 /* FALLTHROUGH */ 545 case IPTOS_ECN_NOTECT: 546 tp->ccv->flags &= ~CCF_IPHDR_CE; 547 break; 548 } 549 550 if (th->th_flags & TH_CWR) 551 tp->ccv->flags |= CCF_TCPHDR_CWR; 552 else 553 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 554 555 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 556 557 if (tp->ccv->flags & CCF_ACKNOW) { 558 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 559 tp->t_flags |= TF_ACKNOW; 560 } 561 } 562 } 563 564 /* 565 * TCP input handling is split into multiple parts: 566 * tcp6_input is a thin wrapper around tcp_input for the extended 567 * ip6_protox[] call format in ip6_input 568 * tcp_input handles primary segment validation, inpcb lookup and 569 * SYN processing on listen sockets 570 * tcp_do_segment processes the ACK and text of the segment for 571 * establishing, established and closing connections 572 */ 573 #ifdef INET6 574 int 575 tcp6_input(struct mbuf **mp, int *offp, int proto) 576 { 577 struct mbuf *m; 578 struct in6_ifaddr *ia6; 579 struct ip6_hdr *ip6; 580 581 m = *mp; 582 if (m->m_len < *offp + sizeof(struct tcphdr)) { 583 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 584 if (m == NULL) { 585 *mp = m; 586 TCPSTAT_INC(tcps_rcvshort); 587 return (IPPROTO_DONE); 588 } 589 } 590 591 /* 592 * draft-itojun-ipv6-tcp-to-anycast 593 * better place to put this in? 594 */ 595 ip6 = mtod(m, struct ip6_hdr *); 596 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 597 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 598 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 599 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 600 *mp = NULL; 601 return (IPPROTO_DONE); 602 } 603 604 *mp = m; 605 return (tcp_input(mp, offp, proto)); 606 } 607 #endif /* INET6 */ 608 609 int 610 tcp_input(struct mbuf **mp, int *offp, int proto) 611 { 612 struct mbuf *m = *mp; 613 struct tcphdr *th = NULL; 614 struct ip *ip = NULL; 615 struct inpcb *inp = NULL; 616 struct tcpcb *tp = NULL; 617 struct socket *so = NULL; 618 u_char *optp = NULL; 619 int off0; 620 int optlen = 0; 621 #ifdef INET 622 int len; 623 uint8_t ipttl; 624 #endif 625 int tlen = 0, off; 626 int drop_hdrlen; 627 int thflags; 628 int rstreason = 0; /* For badport_bandlim accounting purposes */ 629 int lookupflag; 630 uint8_t iptos; 631 struct m_tag *fwd_tag = NULL; 632 #ifdef INET6 633 struct ip6_hdr *ip6 = NULL; 634 int isipv6; 635 #else 636 const void *ip6 = NULL; 637 #endif /* INET6 */ 638 struct tcpopt to; /* options in this segment */ 639 char *s = NULL; /* address and port logging */ 640 #ifdef TCPDEBUG 641 /* 642 * The size of tcp_saveipgen must be the size of the max ip header, 643 * now IPv6. 644 */ 645 u_char tcp_saveipgen[IP6_HDR_LEN]; 646 struct tcphdr tcp_savetcp; 647 short ostate = 0; 648 #endif 649 650 NET_EPOCH_ASSERT(); 651 652 #ifdef INET6 653 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 654 #endif 655 656 off0 = *offp; 657 m = *mp; 658 *mp = NULL; 659 to.to_flags = 0; 660 TCPSTAT_INC(tcps_rcvtotal); 661 662 #ifdef INET6 663 if (isipv6) { 664 ip6 = mtod(m, struct ip6_hdr *); 665 th = (struct tcphdr *)((caddr_t)ip6 + off0); 666 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 667 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 668 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 669 th->th_sum = m->m_pkthdr.csum_data; 670 else 671 th->th_sum = in6_cksum_pseudo(ip6, tlen, 672 IPPROTO_TCP, m->m_pkthdr.csum_data); 673 th->th_sum ^= 0xffff; 674 } else 675 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 676 if (th->th_sum) { 677 TCPSTAT_INC(tcps_rcvbadsum); 678 goto drop; 679 } 680 681 /* 682 * Be proactive about unspecified IPv6 address in source. 683 * As we use all-zero to indicate unbounded/unconnected pcb, 684 * unspecified IPv6 address can be used to confuse us. 685 * 686 * Note that packets with unspecified IPv6 destination is 687 * already dropped in ip6_input. 688 */ 689 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 690 /* XXX stat */ 691 goto drop; 692 } 693 iptos = IPV6_TRAFFIC_CLASS(ip6); 694 } 695 #endif 696 #if defined(INET) && defined(INET6) 697 else 698 #endif 699 #ifdef INET 700 { 701 /* 702 * Get IP and TCP header together in first mbuf. 703 * Note: IP leaves IP header in first mbuf. 704 */ 705 if (off0 > sizeof (struct ip)) { 706 ip_stripoptions(m); 707 off0 = sizeof(struct ip); 708 } 709 if (m->m_len < sizeof (struct tcpiphdr)) { 710 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 711 == NULL) { 712 TCPSTAT_INC(tcps_rcvshort); 713 return (IPPROTO_DONE); 714 } 715 } 716 ip = mtod(m, struct ip *); 717 th = (struct tcphdr *)((caddr_t)ip + off0); 718 tlen = ntohs(ip->ip_len) - off0; 719 720 iptos = ip->ip_tos; 721 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 722 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 723 th->th_sum = m->m_pkthdr.csum_data; 724 else 725 th->th_sum = in_pseudo(ip->ip_src.s_addr, 726 ip->ip_dst.s_addr, 727 htonl(m->m_pkthdr.csum_data + tlen + 728 IPPROTO_TCP)); 729 th->th_sum ^= 0xffff; 730 } else { 731 struct ipovly *ipov = (struct ipovly *)ip; 732 733 /* 734 * Checksum extended TCP header and data. 735 */ 736 len = off0 + tlen; 737 ipttl = ip->ip_ttl; 738 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 739 ipov->ih_len = htons(tlen); 740 th->th_sum = in_cksum(m, len); 741 /* Reset length for SDT probes. */ 742 ip->ip_len = htons(len); 743 /* Reset TOS bits */ 744 ip->ip_tos = iptos; 745 /* Re-initialization for later version check */ 746 ip->ip_ttl = ipttl; 747 ip->ip_v = IPVERSION; 748 ip->ip_hl = off0 >> 2; 749 } 750 751 if (th->th_sum) { 752 TCPSTAT_INC(tcps_rcvbadsum); 753 goto drop; 754 } 755 } 756 #endif /* INET */ 757 758 /* 759 * Check that TCP offset makes sense, 760 * pull out TCP options and adjust length. XXX 761 */ 762 off = th->th_off << 2; 763 if (off < sizeof (struct tcphdr) || off > tlen) { 764 TCPSTAT_INC(tcps_rcvbadoff); 765 goto drop; 766 } 767 tlen -= off; /* tlen is used instead of ti->ti_len */ 768 if (off > sizeof (struct tcphdr)) { 769 #ifdef INET6 770 if (isipv6) { 771 if (m->m_len < off0 + off) { 772 m = m_pullup(m, off0 + off); 773 if (m == NULL) { 774 TCPSTAT_INC(tcps_rcvshort); 775 return (IPPROTO_DONE); 776 } 777 } 778 ip6 = mtod(m, struct ip6_hdr *); 779 th = (struct tcphdr *)((caddr_t)ip6 + off0); 780 } 781 #endif 782 #if defined(INET) && defined(INET6) 783 else 784 #endif 785 #ifdef INET 786 { 787 if (m->m_len < sizeof(struct ip) + off) { 788 if ((m = m_pullup(m, sizeof (struct ip) + off)) 789 == NULL) { 790 TCPSTAT_INC(tcps_rcvshort); 791 return (IPPROTO_DONE); 792 } 793 ip = mtod(m, struct ip *); 794 th = (struct tcphdr *)((caddr_t)ip + off0); 795 } 796 } 797 #endif 798 optlen = off - sizeof (struct tcphdr); 799 optp = (u_char *)(th + 1); 800 } 801 thflags = th->th_flags; 802 803 /* 804 * Convert TCP protocol specific fields to host format. 805 */ 806 tcp_fields_to_host(th); 807 808 /* 809 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 810 */ 811 drop_hdrlen = off0 + off; 812 813 /* 814 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 815 */ 816 if ( 817 #ifdef INET6 818 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 819 #ifdef INET 820 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 821 #endif 822 #endif 823 #if defined(INET) && !defined(INET6) 824 (m->m_flags & M_IP_NEXTHOP) 825 #endif 826 ) 827 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 828 829 /* 830 * For initial SYN packets arriving on listening socket, 831 * we don't need write lock. 832 */ 833 lookupflag = (thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 834 INPLOOKUP_RLOCKLISTEN : INPLOOKUP_WLOCKPCB; 835 findpcb: 836 #ifdef INET6 837 if (isipv6 && fwd_tag != NULL) { 838 struct sockaddr_in6 *next_hop6; 839 840 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 841 /* 842 * Transparently forwarded. Pretend to be the destination. 843 * Already got one like this? 844 */ 845 inp = in6_pcblookup_mbuf(&V_tcbinfo, 846 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 847 lookupflag, m->m_pkthdr.rcvif, m); 848 if (!inp) { 849 /* 850 * It's new. Try to find the ambushing socket. 851 * Because we've rewritten the destination address, 852 * any hardware-generated hash is ignored. 853 */ 854 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 855 th->th_sport, &next_hop6->sin6_addr, 856 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 857 th->th_dport, INPLOOKUP_WILDCARD | lookupflag, 858 m->m_pkthdr.rcvif); 859 } 860 } else if (isipv6) { 861 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 862 th->th_sport, &ip6->ip6_dst, th->th_dport, 863 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m); 864 } 865 #endif /* INET6 */ 866 #if defined(INET6) && defined(INET) 867 else 868 #endif 869 #ifdef INET 870 if (fwd_tag != NULL) { 871 struct sockaddr_in *next_hop; 872 873 next_hop = (struct sockaddr_in *)(fwd_tag+1); 874 /* 875 * Transparently forwarded. Pretend to be the destination. 876 * already got one like this? 877 */ 878 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 879 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); 880 if (!inp) { 881 /* 882 * It's new. Try to find the ambushing socket. 883 * Because we've rewritten the destination address, 884 * any hardware-generated hash is ignored. 885 */ 886 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 887 th->th_sport, next_hop->sin_addr, 888 next_hop->sin_port ? ntohs(next_hop->sin_port) : 889 th->th_dport, INPLOOKUP_WILDCARD | lookupflag, 890 m->m_pkthdr.rcvif); 891 } 892 } else 893 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 894 th->th_sport, ip->ip_dst, th->th_dport, 895 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m); 896 #endif /* INET */ 897 898 /* 899 * If the INPCB does not exist then all data in the incoming 900 * segment is discarded and an appropriate RST is sent back. 901 * XXX MRT Send RST using which routing table? 902 */ 903 if (inp == NULL) { 904 /* 905 * Log communication attempts to ports that are not 906 * in use. 907 */ 908 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 909 V_tcp_log_in_vain == 2) { 910 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 911 log(LOG_INFO, "%s; %s: Connection attempt " 912 "to closed port\n", s, __func__); 913 } 914 /* 915 * When blackholing do not respond with a RST but 916 * completely ignore the segment and drop it. 917 */ 918 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 919 V_blackhole == 2) 920 goto dropunlock; 921 922 rstreason = BANDLIM_RST_CLOSEDPORT; 923 goto dropwithreset; 924 } 925 INP_LOCK_ASSERT(inp); 926 /* 927 * While waiting for inp lock during the lookup, another thread 928 * can have dropped the inpcb, in which case we need to loop back 929 * and try to find a new inpcb to deliver to. 930 */ 931 if (inp->inp_flags & INP_DROPPED) { 932 INP_UNLOCK(inp); 933 inp = NULL; 934 goto findpcb; 935 } 936 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 937 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 938 ((inp->inp_socket == NULL) || 939 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 940 inp->inp_flowid = m->m_pkthdr.flowid; 941 inp->inp_flowtype = M_HASHTYPE_GET(m); 942 } 943 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 944 #ifdef INET6 945 if (isipv6 && IPSEC_ENABLED(ipv6) && 946 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 947 goto dropunlock; 948 } 949 #ifdef INET 950 else 951 #endif 952 #endif /* INET6 */ 953 #ifdef INET 954 if (IPSEC_ENABLED(ipv4) && 955 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 956 goto dropunlock; 957 } 958 #endif /* INET */ 959 #endif /* IPSEC */ 960 961 /* 962 * Check the minimum TTL for socket. 963 */ 964 if (inp->inp_ip_minttl != 0) { 965 #ifdef INET6 966 if (isipv6) { 967 if (inp->inp_ip_minttl > ip6->ip6_hlim) 968 goto dropunlock; 969 } else 970 #endif 971 if (inp->inp_ip_minttl > ip->ip_ttl) 972 goto dropunlock; 973 } 974 975 /* 976 * A previous connection in TIMEWAIT state is supposed to catch stray 977 * or duplicate segments arriving late. If this segment was a 978 * legitimate new connection attempt, the old INPCB gets removed and 979 * we can try again to find a listening socket. 980 * 981 * At this point, due to earlier optimism, we may hold only an inpcb 982 * lock, and not the inpcbinfo write lock. If so, we need to try to 983 * acquire it, or if that fails, acquire a reference on the inpcb, 984 * drop all locks, acquire a global write lock, and then re-acquire 985 * the inpcb lock. We may at that point discover that another thread 986 * has tried to free the inpcb, in which case we need to loop back 987 * and try to find a new inpcb to deliver to. 988 * 989 * XXXRW: It may be time to rethink timewait locking. 990 */ 991 if (inp->inp_flags & INP_TIMEWAIT) { 992 tcp_dooptions(&to, optp, optlen, 993 (thflags & TH_SYN) ? TO_SYN : 0); 994 /* 995 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 996 */ 997 if (tcp_twcheck(inp, &to, th, m, tlen)) 998 goto findpcb; 999 return (IPPROTO_DONE); 1000 } 1001 /* 1002 * The TCPCB may no longer exist if the connection is winding 1003 * down or it is in the CLOSED state. Either way we drop the 1004 * segment and send an appropriate response. 1005 */ 1006 tp = intotcpcb(inp); 1007 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1008 rstreason = BANDLIM_RST_CLOSEDPORT; 1009 goto dropwithreset; 1010 } 1011 1012 #ifdef TCP_OFFLOAD 1013 if (tp->t_flags & TF_TOE) { 1014 tcp_offload_input(tp, m); 1015 m = NULL; /* consumed by the TOE driver */ 1016 goto dropunlock; 1017 } 1018 #endif 1019 1020 #ifdef MAC 1021 if (mac_inpcb_check_deliver(inp, m)) 1022 goto dropunlock; 1023 #endif 1024 so = inp->inp_socket; 1025 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1026 #ifdef TCPDEBUG 1027 if (so->so_options & SO_DEBUG) { 1028 ostate = tp->t_state; 1029 #ifdef INET6 1030 if (isipv6) { 1031 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1032 } else 1033 #endif 1034 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1035 tcp_savetcp = *th; 1036 } 1037 #endif /* TCPDEBUG */ 1038 /* 1039 * When the socket is accepting connections (the INPCB is in LISTEN 1040 * state) we look into the SYN cache if this is a new connection 1041 * attempt or the completion of a previous one. 1042 */ 1043 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1044 ("%s: so accepting but tp %p not listening", __func__, tp)); 1045 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1046 struct in_conninfo inc; 1047 1048 bzero(&inc, sizeof(inc)); 1049 #ifdef INET6 1050 if (isipv6) { 1051 inc.inc_flags |= INC_ISIPV6; 1052 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1053 inc.inc_flags |= INC_IPV6MINMTU; 1054 inc.inc6_faddr = ip6->ip6_src; 1055 inc.inc6_laddr = ip6->ip6_dst; 1056 } else 1057 #endif 1058 { 1059 inc.inc_faddr = ip->ip_src; 1060 inc.inc_laddr = ip->ip_dst; 1061 } 1062 inc.inc_fport = th->th_sport; 1063 inc.inc_lport = th->th_dport; 1064 inc.inc_fibnum = so->so_fibnum; 1065 1066 /* 1067 * Check for an existing connection attempt in syncache if 1068 * the flag is only ACK. A successful lookup creates a new 1069 * socket appended to the listen queue in SYN_RECEIVED state. 1070 */ 1071 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1072 /* 1073 * Parse the TCP options here because 1074 * syncookies need access to the reflected 1075 * timestamp. 1076 */ 1077 tcp_dooptions(&to, optp, optlen, 0); 1078 /* 1079 * NB: syncache_expand() doesn't unlock 1080 * inp and tcpinfo locks. 1081 */ 1082 rstreason = syncache_expand(&inc, &to, th, &so, m); 1083 if (rstreason < 0) { 1084 /* 1085 * A failing TCP MD5 signature comparison 1086 * must result in the segment being dropped 1087 * and must not produce any response back 1088 * to the sender. 1089 */ 1090 goto dropunlock; 1091 } else if (rstreason == 0) { 1092 /* 1093 * No syncache entry or ACK was not 1094 * for our SYN/ACK. Send a RST. 1095 * NB: syncache did its own logging 1096 * of the failure cause. 1097 */ 1098 rstreason = BANDLIM_RST_OPENPORT; 1099 goto dropwithreset; 1100 } 1101 tfo_socket_result: 1102 if (so == NULL) { 1103 /* 1104 * We completed the 3-way handshake 1105 * but could not allocate a socket 1106 * either due to memory shortage, 1107 * listen queue length limits or 1108 * global socket limits. Send RST 1109 * or wait and have the remote end 1110 * retransmit the ACK for another 1111 * try. 1112 */ 1113 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1114 log(LOG_DEBUG, "%s; %s: Listen socket: " 1115 "Socket allocation failed due to " 1116 "limits or memory shortage, %s\n", 1117 s, __func__, 1118 V_tcp_sc_rst_sock_fail ? 1119 "sending RST" : "try again"); 1120 if (V_tcp_sc_rst_sock_fail) { 1121 rstreason = BANDLIM_UNLIMITED; 1122 goto dropwithreset; 1123 } else 1124 goto dropunlock; 1125 } 1126 /* 1127 * Socket is created in state SYN_RECEIVED. 1128 * Unlock the listen socket, lock the newly 1129 * created socket and update the tp variable. 1130 * If we came here via jump to tfo_socket_result, 1131 * then listening socket is read-locked. 1132 */ 1133 INP_UNLOCK(inp); /* listen socket */ 1134 inp = sotoinpcb(so); 1135 /* 1136 * New connection inpcb is already locked by 1137 * syncache_expand(). 1138 */ 1139 INP_WLOCK_ASSERT(inp); 1140 tp = intotcpcb(inp); 1141 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1142 ("%s: ", __func__)); 1143 /* 1144 * Process the segment and the data it 1145 * contains. tcp_do_segment() consumes 1146 * the mbuf chain and unlocks the inpcb. 1147 */ 1148 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1149 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1150 iptos); 1151 return (IPPROTO_DONE); 1152 } 1153 /* 1154 * Segment flag validation for new connection attempts: 1155 * 1156 * Our (SYN|ACK) response was rejected. 1157 * Check with syncache and remove entry to prevent 1158 * retransmits. 1159 * 1160 * NB: syncache_chkrst does its own logging of failure 1161 * causes. 1162 */ 1163 if (thflags & TH_RST) { 1164 syncache_chkrst(&inc, th, m); 1165 goto dropunlock; 1166 } 1167 /* 1168 * We can't do anything without SYN. 1169 */ 1170 if ((thflags & TH_SYN) == 0) { 1171 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1172 log(LOG_DEBUG, "%s; %s: Listen socket: " 1173 "SYN is missing, segment ignored\n", 1174 s, __func__); 1175 TCPSTAT_INC(tcps_badsyn); 1176 goto dropunlock; 1177 } 1178 /* 1179 * (SYN|ACK) is bogus on a listen socket. 1180 */ 1181 if (thflags & TH_ACK) { 1182 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1183 log(LOG_DEBUG, "%s; %s: Listen socket: " 1184 "SYN|ACK invalid, segment rejected\n", 1185 s, __func__); 1186 syncache_badack(&inc); /* XXX: Not needed! */ 1187 TCPSTAT_INC(tcps_badsyn); 1188 rstreason = BANDLIM_RST_OPENPORT; 1189 goto dropwithreset; 1190 } 1191 /* 1192 * If the drop_synfin option is enabled, drop all 1193 * segments with both the SYN and FIN bits set. 1194 * This prevents e.g. nmap from identifying the 1195 * TCP/IP stack. 1196 * XXX: Poor reasoning. nmap has other methods 1197 * and is constantly refining its stack detection 1198 * strategies. 1199 * XXX: This is a violation of the TCP specification 1200 * and was used by RFC1644. 1201 */ 1202 if ((thflags & TH_FIN) && V_drop_synfin) { 1203 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1204 log(LOG_DEBUG, "%s; %s: Listen socket: " 1205 "SYN|FIN segment ignored (based on " 1206 "sysctl setting)\n", s, __func__); 1207 TCPSTAT_INC(tcps_badsyn); 1208 goto dropunlock; 1209 } 1210 /* 1211 * Segment's flags are (SYN) or (SYN|FIN). 1212 * 1213 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1214 * as they do not affect the state of the TCP FSM. 1215 * The data pointed to by TH_URG and th_urp is ignored. 1216 */ 1217 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1218 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1219 KASSERT(thflags & (TH_SYN), 1220 ("%s: Listen socket: TH_SYN not set", __func__)); 1221 INP_RLOCK_ASSERT(inp); 1222 #ifdef INET6 1223 /* 1224 * If deprecated address is forbidden, 1225 * we do not accept SYN to deprecated interface 1226 * address to prevent any new inbound connection from 1227 * getting established. 1228 * When we do not accept SYN, we send a TCP RST, 1229 * with deprecated source address (instead of dropping 1230 * it). We compromise it as it is much better for peer 1231 * to send a RST, and RST will be the final packet 1232 * for the exchange. 1233 * 1234 * If we do not forbid deprecated addresses, we accept 1235 * the SYN packet. RFC2462 does not suggest dropping 1236 * SYN in this case. 1237 * If we decipher RFC2462 5.5.4, it says like this: 1238 * 1. use of deprecated addr with existing 1239 * communication is okay - "SHOULD continue to be 1240 * used" 1241 * 2. use of it with new communication: 1242 * (2a) "SHOULD NOT be used if alternate address 1243 * with sufficient scope is available" 1244 * (2b) nothing mentioned otherwise. 1245 * Here we fall into (2b) case as we have no choice in 1246 * our source address selection - we must obey the peer. 1247 * 1248 * The wording in RFC2462 is confusing, and there are 1249 * multiple description text for deprecated address 1250 * handling - worse, they are not exactly the same. 1251 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1252 */ 1253 if (isipv6 && !V_ip6_use_deprecated) { 1254 struct in6_ifaddr *ia6; 1255 1256 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1257 if (ia6 != NULL && 1258 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1259 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1260 log(LOG_DEBUG, "%s; %s: Listen socket: " 1261 "Connection attempt to deprecated " 1262 "IPv6 address rejected\n", 1263 s, __func__); 1264 rstreason = BANDLIM_RST_OPENPORT; 1265 goto dropwithreset; 1266 } 1267 } 1268 #endif /* INET6 */ 1269 /* 1270 * Basic sanity checks on incoming SYN requests: 1271 * Don't respond if the destination is a link layer 1272 * broadcast according to RFC1122 4.2.3.10, p. 104. 1273 * If it is from this socket it must be forged. 1274 * Don't respond if the source or destination is a 1275 * global or subnet broad- or multicast address. 1276 * Note that it is quite possible to receive unicast 1277 * link-layer packets with a broadcast IP address. Use 1278 * in_broadcast() to find them. 1279 */ 1280 if (m->m_flags & (M_BCAST|M_MCAST)) { 1281 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1282 log(LOG_DEBUG, "%s; %s: Listen socket: " 1283 "Connection attempt from broad- or multicast " 1284 "link layer address ignored\n", s, __func__); 1285 goto dropunlock; 1286 } 1287 #ifdef INET6 1288 if (isipv6) { 1289 if (th->th_dport == th->th_sport && 1290 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1291 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1292 log(LOG_DEBUG, "%s; %s: Listen socket: " 1293 "Connection attempt to/from self " 1294 "ignored\n", s, __func__); 1295 goto dropunlock; 1296 } 1297 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1298 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1299 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1300 log(LOG_DEBUG, "%s; %s: Listen socket: " 1301 "Connection attempt from/to multicast " 1302 "address ignored\n", s, __func__); 1303 goto dropunlock; 1304 } 1305 } 1306 #endif 1307 #if defined(INET) && defined(INET6) 1308 else 1309 #endif 1310 #ifdef INET 1311 { 1312 if (th->th_dport == th->th_sport && 1313 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1314 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1315 log(LOG_DEBUG, "%s; %s: Listen socket: " 1316 "Connection attempt from/to self " 1317 "ignored\n", s, __func__); 1318 goto dropunlock; 1319 } 1320 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1321 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1322 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1323 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1324 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1325 log(LOG_DEBUG, "%s; %s: Listen socket: " 1326 "Connection attempt from/to broad- " 1327 "or multicast address ignored\n", 1328 s, __func__); 1329 goto dropunlock; 1330 } 1331 } 1332 #endif 1333 /* 1334 * SYN appears to be valid. Create compressed TCP state 1335 * for syncache. 1336 */ 1337 #ifdef TCPDEBUG 1338 if (so->so_options & SO_DEBUG) 1339 tcp_trace(TA_INPUT, ostate, tp, 1340 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1341 #endif 1342 TCP_PROBE3(debug__input, tp, th, m); 1343 tcp_dooptions(&to, optp, optlen, TO_SYN); 1344 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1345 iptos)) != NULL) 1346 goto tfo_socket_result; 1347 1348 /* 1349 * Entry added to syncache and mbuf consumed. 1350 * Only the listen socket is unlocked by syncache_add(). 1351 */ 1352 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1353 return (IPPROTO_DONE); 1354 } else if (tp->t_state == TCPS_LISTEN) { 1355 /* 1356 * When a listen socket is torn down the SO_ACCEPTCONN 1357 * flag is removed first while connections are drained 1358 * from the accept queue in a unlock/lock cycle of the 1359 * ACCEPT_LOCK, opening a race condition allowing a SYN 1360 * attempt go through unhandled. 1361 */ 1362 goto dropunlock; 1363 } 1364 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1365 if (tp->t_flags & TF_SIGNATURE) { 1366 tcp_dooptions(&to, optp, optlen, thflags); 1367 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1368 TCPSTAT_INC(tcps_sig_err_nosigopt); 1369 goto dropunlock; 1370 } 1371 if (!TCPMD5_ENABLED() || 1372 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1373 goto dropunlock; 1374 } 1375 #endif 1376 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1377 1378 /* 1379 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1380 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1381 * the inpcb, and unlocks pcbinfo. 1382 */ 1383 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1384 return (IPPROTO_DONE); 1385 1386 dropwithreset: 1387 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1388 1389 if (inp != NULL) { 1390 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1391 INP_UNLOCK(inp); 1392 } else 1393 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1394 m = NULL; /* mbuf chain got consumed. */ 1395 goto drop; 1396 1397 dropunlock: 1398 if (m != NULL) 1399 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1400 1401 if (inp != NULL) 1402 INP_UNLOCK(inp); 1403 1404 drop: 1405 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1406 if (s != NULL) 1407 free(s, M_TCPLOG); 1408 if (m != NULL) 1409 m_freem(m); 1410 return (IPPROTO_DONE); 1411 } 1412 1413 /* 1414 * Automatic sizing of receive socket buffer. Often the send 1415 * buffer size is not optimally adjusted to the actual network 1416 * conditions at hand (delay bandwidth product). Setting the 1417 * buffer size too small limits throughput on links with high 1418 * bandwidth and high delay (eg. trans-continental/oceanic links). 1419 * 1420 * On the receive side the socket buffer memory is only rarely 1421 * used to any significant extent. This allows us to be much 1422 * more aggressive in scaling the receive socket buffer. For 1423 * the case that the buffer space is actually used to a large 1424 * extent and we run out of kernel memory we can simply drop 1425 * the new segments; TCP on the sender will just retransmit it 1426 * later. Setting the buffer size too big may only consume too 1427 * much kernel memory if the application doesn't read() from 1428 * the socket or packet loss or reordering makes use of the 1429 * reassembly queue. 1430 * 1431 * The criteria to step up the receive buffer one notch are: 1432 * 1. Application has not set receive buffer size with 1433 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1434 * 2. the number of bytes received during 1/2 of an sRTT 1435 * is at least 3/8 of the current socket buffer size. 1436 * 3. receive buffer size has not hit maximal automatic size; 1437 * 1438 * If all of the criteria are met we increaset the socket buffer 1439 * by a 1/2 (bounded by the max). This allows us to keep ahead 1440 * of slow-start but also makes it so our peer never gets limited 1441 * by our rwnd which we then open up causing a burst. 1442 * 1443 * This algorithm does two steps per RTT at most and only if 1444 * we receive a bulk stream w/o packet losses or reorderings. 1445 * Shrinking the buffer during idle times is not necessary as 1446 * it doesn't consume any memory when idle. 1447 * 1448 * TODO: Only step up if the application is actually serving 1449 * the buffer to better manage the socket buffer resources. 1450 */ 1451 int 1452 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1453 struct tcpcb *tp, int tlen) 1454 { 1455 int newsize = 0; 1456 1457 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1458 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1459 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1460 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1461 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1462 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1463 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1464 } 1465 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1466 1467 /* Start over with next RTT. */ 1468 tp->rfbuf_ts = 0; 1469 tp->rfbuf_cnt = 0; 1470 } else { 1471 tp->rfbuf_cnt += tlen; /* add up */ 1472 } 1473 return (newsize); 1474 } 1475 1476 void 1477 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1478 { 1479 /* 1480 * Since tp might be gone if the session entered 1481 * the TIME_WAIT state before coming here, we need 1482 * to check if the socket is still connected. 1483 */ 1484 if ((so->so_state & SS_ISCONNECTED) == 0) 1485 return; 1486 INP_LOCK_ASSERT(tp->t_inpcb); 1487 if (tp->t_flags & TF_WAKESOR) { 1488 tp->t_flags &= ~TF_WAKESOR; 1489 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 1490 sorwakeup(so); 1491 } 1492 if (tp->t_flags & TF_WAKESOW) { 1493 tp->t_flags &= ~TF_WAKESOW; 1494 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 1495 sowwakeup(so); 1496 } 1497 } 1498 1499 void 1500 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1501 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1502 { 1503 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1504 int rstreason, todrop, win, incforsyn = 0; 1505 uint32_t tiwin; 1506 uint16_t nsegs; 1507 char *s; 1508 struct in_conninfo *inc; 1509 struct mbuf *mfree; 1510 struct tcpopt to; 1511 int tfo_syn; 1512 u_int maxseg; 1513 1514 #ifdef TCPDEBUG 1515 /* 1516 * The size of tcp_saveipgen must be the size of the max ip header, 1517 * now IPv6. 1518 */ 1519 u_char tcp_saveipgen[IP6_HDR_LEN]; 1520 struct tcphdr tcp_savetcp; 1521 short ostate = 0; 1522 #endif 1523 thflags = th->th_flags; 1524 inc = &tp->t_inpcb->inp_inc; 1525 tp->sackhint.last_sack_ack = 0; 1526 sack_changed = 0; 1527 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1528 1529 NET_EPOCH_ASSERT(); 1530 INP_WLOCK_ASSERT(tp->t_inpcb); 1531 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1532 __func__)); 1533 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1534 __func__)); 1535 1536 #ifdef TCPPCAP 1537 /* Save segment, if requested. */ 1538 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1539 #endif 1540 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1541 tlen, NULL, true); 1542 1543 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1544 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1545 log(LOG_DEBUG, "%s; %s: " 1546 "SYN|FIN segment ignored (based on " 1547 "sysctl setting)\n", s, __func__); 1548 free(s, M_TCPLOG); 1549 } 1550 goto drop; 1551 } 1552 1553 /* 1554 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1555 * check SEQ.ACK first. 1556 */ 1557 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1558 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1559 rstreason = BANDLIM_UNLIMITED; 1560 goto dropwithreset; 1561 } 1562 1563 /* 1564 * Segment received on connection. 1565 * Reset idle time and keep-alive timer. 1566 * XXX: This should be done after segment 1567 * validation to ignore broken/spoofed segs. 1568 */ 1569 tp->t_rcvtime = ticks; 1570 1571 /* 1572 * Scale up the window into a 32-bit value. 1573 * For the SYN_SENT state the scale is zero. 1574 */ 1575 tiwin = th->th_win << tp->snd_scale; 1576 #ifdef STATS 1577 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1578 #endif 1579 1580 /* 1581 * TCP ECN processing. 1582 */ 1583 if (tp->t_flags2 & TF2_ECN_PERMIT) { 1584 if (thflags & TH_CWR) { 1585 tp->t_flags2 &= ~TF2_ECN_SND_ECE; 1586 tp->t_flags |= TF_ACKNOW; 1587 } 1588 switch (iptos & IPTOS_ECN_MASK) { 1589 case IPTOS_ECN_CE: 1590 tp->t_flags2 |= TF2_ECN_SND_ECE; 1591 TCPSTAT_INC(tcps_ecn_ce); 1592 break; 1593 case IPTOS_ECN_ECT0: 1594 TCPSTAT_INC(tcps_ecn_ect0); 1595 break; 1596 case IPTOS_ECN_ECT1: 1597 TCPSTAT_INC(tcps_ecn_ect1); 1598 break; 1599 } 1600 1601 /* Process a packet differently from RFC3168. */ 1602 cc_ecnpkt_handler(tp, th, iptos); 1603 1604 /* Congestion experienced. */ 1605 if (thflags & TH_ECE) { 1606 cc_cong_signal(tp, th, CC_ECN); 1607 } 1608 } 1609 1610 /* 1611 * Parse options on any incoming segment. 1612 */ 1613 tcp_dooptions(&to, (u_char *)(th + 1), 1614 (th->th_off << 2) - sizeof(struct tcphdr), 1615 (thflags & TH_SYN) ? TO_SYN : 0); 1616 1617 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1618 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1619 (to.to_flags & TOF_SIGNATURE) == 0) { 1620 TCPSTAT_INC(tcps_sig_err_sigopt); 1621 /* XXX: should drop? */ 1622 } 1623 #endif 1624 /* 1625 * If echoed timestamp is later than the current time, 1626 * fall back to non RFC1323 RTT calculation. Normalize 1627 * timestamp if syncookies were used when this connection 1628 * was established. 1629 */ 1630 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1631 to.to_tsecr -= tp->ts_offset; 1632 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1633 to.to_tsecr = 0; 1634 else if (tp->t_flags & TF_PREVVALID && 1635 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1636 cc_cong_signal(tp, th, CC_RTO_ERR); 1637 } 1638 /* 1639 * Process options only when we get SYN/ACK back. The SYN case 1640 * for incoming connections is handled in tcp_syncache. 1641 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1642 * or <SYN,ACK>) segment itself is never scaled. 1643 * XXX this is traditional behavior, may need to be cleaned up. 1644 */ 1645 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1646 /* Handle parallel SYN for ECN */ 1647 if (!(thflags & TH_ACK) && 1648 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) && 1649 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) { 1650 tp->t_flags2 |= TF2_ECN_PERMIT; 1651 tp->t_flags2 |= TF2_ECN_SND_ECE; 1652 TCPSTAT_INC(tcps_ecn_shs); 1653 } 1654 if ((to.to_flags & TOF_SCALE) && 1655 (tp->t_flags & TF_REQ_SCALE) && 1656 !(tp->t_flags & TF_NOOPT)) { 1657 tp->t_flags |= TF_RCVD_SCALE; 1658 tp->snd_scale = to.to_wscale; 1659 } else 1660 tp->t_flags &= ~TF_REQ_SCALE; 1661 /* 1662 * Initial send window. It will be updated with 1663 * the next incoming segment to the scaled value. 1664 */ 1665 tp->snd_wnd = th->th_win; 1666 if ((to.to_flags & TOF_TS) && 1667 (tp->t_flags & TF_REQ_TSTMP) && 1668 !(tp->t_flags & TF_NOOPT)) { 1669 tp->t_flags |= TF_RCVD_TSTMP; 1670 tp->ts_recent = to.to_tsval; 1671 tp->ts_recent_age = tcp_ts_getticks(); 1672 } else 1673 tp->t_flags &= ~TF_REQ_TSTMP; 1674 if (to.to_flags & TOF_MSS) 1675 tcp_mss(tp, to.to_mss); 1676 if ((tp->t_flags & TF_SACK_PERMIT) && 1677 (!(to.to_flags & TOF_SACKPERM) || 1678 (tp->t_flags & TF_NOOPT))) 1679 tp->t_flags &= ~TF_SACK_PERMIT; 1680 if (IS_FASTOPEN(tp->t_flags)) { 1681 if ((to.to_flags & TOF_FASTOPEN) && 1682 !(tp->t_flags & TF_NOOPT)) { 1683 uint16_t mss; 1684 1685 if (to.to_flags & TOF_MSS) 1686 mss = to.to_mss; 1687 else 1688 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1689 mss = TCP6_MSS; 1690 else 1691 mss = TCP_MSS; 1692 tcp_fastopen_update_cache(tp, mss, 1693 to.to_tfo_len, to.to_tfo_cookie); 1694 } else 1695 tcp_fastopen_disable_path(tp); 1696 } 1697 } 1698 1699 /* 1700 * If timestamps were negotiated during SYN/ACK and a 1701 * segment without a timestamp is received, silently drop 1702 * the segment, unless it is a RST segment or missing timestamps are 1703 * tolerated. 1704 * See section 3.2 of RFC 7323. 1705 */ 1706 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1707 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1708 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1709 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1710 "segment processed normally\n", 1711 s, __func__); 1712 free(s, M_TCPLOG); 1713 } 1714 } else { 1715 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1716 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1717 "segment silently dropped\n", s, __func__); 1718 free(s, M_TCPLOG); 1719 } 1720 goto drop; 1721 } 1722 } 1723 /* 1724 * If timestamps were not negotiated during SYN/ACK and a 1725 * segment with a timestamp is received, ignore the 1726 * timestamp and process the packet normally. 1727 * See section 3.2 of RFC 7323. 1728 */ 1729 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1730 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1731 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1732 "segment processed normally\n", s, __func__); 1733 free(s, M_TCPLOG); 1734 } 1735 } 1736 1737 /* 1738 * Header prediction: check for the two common cases 1739 * of a uni-directional data xfer. If the packet has 1740 * no control flags, is in-sequence, the window didn't 1741 * change and we're not retransmitting, it's a 1742 * candidate. If the length is zero and the ack moved 1743 * forward, we're the sender side of the xfer. Just 1744 * free the data acked & wake any higher level process 1745 * that was blocked waiting for space. If the length 1746 * is non-zero and the ack didn't move, we're the 1747 * receiver side. If we're getting packets in-order 1748 * (the reassembly queue is empty), add the data to 1749 * the socket buffer and note that we need a delayed ack. 1750 * Make sure that the hidden state-flags are also off. 1751 * Since we check for TCPS_ESTABLISHED first, it can only 1752 * be TH_NEEDSYN. 1753 */ 1754 if (tp->t_state == TCPS_ESTABLISHED && 1755 th->th_seq == tp->rcv_nxt && 1756 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1757 tp->snd_nxt == tp->snd_max && 1758 tiwin && tiwin == tp->snd_wnd && 1759 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1760 SEGQ_EMPTY(tp) && 1761 ((to.to_flags & TOF_TS) == 0 || 1762 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1763 /* 1764 * If last ACK falls within this segment's sequence numbers, 1765 * record the timestamp. 1766 * NOTE that the test is modified according to the latest 1767 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1768 */ 1769 if ((to.to_flags & TOF_TS) != 0 && 1770 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1771 tp->ts_recent_age = tcp_ts_getticks(); 1772 tp->ts_recent = to.to_tsval; 1773 } 1774 1775 if (tlen == 0) { 1776 if (SEQ_GT(th->th_ack, tp->snd_una) && 1777 SEQ_LEQ(th->th_ack, tp->snd_max) && 1778 !IN_RECOVERY(tp->t_flags) && 1779 (to.to_flags & TOF_SACK) == 0 && 1780 TAILQ_EMPTY(&tp->snd_holes)) { 1781 /* 1782 * This is a pure ack for outstanding data. 1783 */ 1784 TCPSTAT_INC(tcps_predack); 1785 1786 /* 1787 * "bad retransmit" recovery without timestamps. 1788 */ 1789 if ((to.to_flags & TOF_TS) == 0 && 1790 tp->t_rxtshift == 1 && 1791 tp->t_flags & TF_PREVVALID && 1792 (int)(ticks - tp->t_badrxtwin) < 0) { 1793 cc_cong_signal(tp, th, CC_RTO_ERR); 1794 } 1795 1796 /* 1797 * Recalculate the transmit timer / rtt. 1798 * 1799 * Some boxes send broken timestamp replies 1800 * during the SYN+ACK phase, ignore 1801 * timestamps of 0 or we could calculate a 1802 * huge RTT and blow up the retransmit timer. 1803 */ 1804 if ((to.to_flags & TOF_TS) != 0 && 1805 to.to_tsecr) { 1806 uint32_t t; 1807 1808 t = tcp_ts_getticks() - to.to_tsecr; 1809 if (!tp->t_rttlow || tp->t_rttlow > t) 1810 tp->t_rttlow = t; 1811 tcp_xmit_timer(tp, 1812 TCP_TS_TO_TICKS(t) + 1); 1813 } else if (tp->t_rtttime && 1814 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1815 if (!tp->t_rttlow || 1816 tp->t_rttlow > ticks - tp->t_rtttime) 1817 tp->t_rttlow = ticks - tp->t_rtttime; 1818 tcp_xmit_timer(tp, 1819 ticks - tp->t_rtttime); 1820 } 1821 acked = BYTES_THIS_ACK(tp, th); 1822 1823 #ifdef TCP_HHOOK 1824 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1825 hhook_run_tcp_est_in(tp, th, &to); 1826 #endif 1827 1828 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1829 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1830 sbdrop(&so->so_snd, acked); 1831 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1832 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1833 tp->snd_recover = th->th_ack - 1; 1834 1835 /* 1836 * Let the congestion control algorithm update 1837 * congestion control related information. This 1838 * typically means increasing the congestion 1839 * window. 1840 */ 1841 cc_ack_received(tp, th, nsegs, CC_ACK); 1842 1843 tp->snd_una = th->th_ack; 1844 /* 1845 * Pull snd_wl2 up to prevent seq wrap relative 1846 * to th_ack. 1847 */ 1848 tp->snd_wl2 = th->th_ack; 1849 tp->t_dupacks = 0; 1850 m_freem(m); 1851 1852 /* 1853 * If all outstanding data are acked, stop 1854 * retransmit timer, otherwise restart timer 1855 * using current (possibly backed-off) value. 1856 * If process is waiting for space, 1857 * wakeup/selwakeup/signal. If data 1858 * are ready to send, let tcp_output 1859 * decide between more output or persist. 1860 */ 1861 #ifdef TCPDEBUG 1862 if (so->so_options & SO_DEBUG) 1863 tcp_trace(TA_INPUT, ostate, tp, 1864 (void *)tcp_saveipgen, 1865 &tcp_savetcp, 0); 1866 #endif 1867 TCP_PROBE3(debug__input, tp, th, m); 1868 if (tp->snd_una == tp->snd_max) 1869 tcp_timer_activate(tp, TT_REXMT, 0); 1870 else if (!tcp_timer_active(tp, TT_PERSIST)) 1871 tcp_timer_activate(tp, TT_REXMT, 1872 tp->t_rxtcur); 1873 tp->t_flags |= TF_WAKESOW; 1874 if (sbavail(&so->so_snd)) 1875 (void) tp->t_fb->tfb_tcp_output(tp); 1876 goto check_delack; 1877 } 1878 } else if (th->th_ack == tp->snd_una && 1879 tlen <= sbspace(&so->so_rcv)) { 1880 int newsize = 0; /* automatic sockbuf scaling */ 1881 1882 /* 1883 * This is a pure, in-sequence data packet with 1884 * nothing on the reassembly queue and we have enough 1885 * buffer space to take it. 1886 */ 1887 /* Clean receiver SACK report if present */ 1888 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1889 tcp_clean_sackreport(tp); 1890 TCPSTAT_INC(tcps_preddat); 1891 tp->rcv_nxt += tlen; 1892 if (tlen && 1893 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1894 (tp->t_fbyte_in == 0)) { 1895 tp->t_fbyte_in = ticks; 1896 if (tp->t_fbyte_in == 0) 1897 tp->t_fbyte_in = 1; 1898 if (tp->t_fbyte_out && tp->t_fbyte_in) 1899 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1900 } 1901 /* 1902 * Pull snd_wl1 up to prevent seq wrap relative to 1903 * th_seq. 1904 */ 1905 tp->snd_wl1 = th->th_seq; 1906 /* 1907 * Pull rcv_up up to prevent seq wrap relative to 1908 * rcv_nxt. 1909 */ 1910 tp->rcv_up = tp->rcv_nxt; 1911 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1912 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1913 #ifdef TCPDEBUG 1914 if (so->so_options & SO_DEBUG) 1915 tcp_trace(TA_INPUT, ostate, tp, 1916 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1917 #endif 1918 TCP_PROBE3(debug__input, tp, th, m); 1919 1920 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1921 1922 /* Add data to socket buffer. */ 1923 SOCKBUF_LOCK(&so->so_rcv); 1924 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1925 m_freem(m); 1926 } else { 1927 /* 1928 * Set new socket buffer size. 1929 * Give up when limit is reached. 1930 */ 1931 if (newsize) 1932 if (!sbreserve_locked(&so->so_rcv, 1933 newsize, so, NULL)) 1934 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1935 m_adj(m, drop_hdrlen); /* delayed header drop */ 1936 sbappendstream_locked(&so->so_rcv, m, 0); 1937 } 1938 SOCKBUF_UNLOCK(&so->so_rcv); 1939 tp->t_flags |= TF_WAKESOR; 1940 if (DELAY_ACK(tp, tlen)) { 1941 tp->t_flags |= TF_DELACK; 1942 } else { 1943 tp->t_flags |= TF_ACKNOW; 1944 tp->t_fb->tfb_tcp_output(tp); 1945 } 1946 goto check_delack; 1947 } 1948 } 1949 1950 /* 1951 * Calculate amount of space in receive window, 1952 * and then do TCP input processing. 1953 * Receive window is amount of space in rcv queue, 1954 * but not less than advertised window. 1955 */ 1956 win = sbspace(&so->so_rcv); 1957 if (win < 0) 1958 win = 0; 1959 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1960 1961 switch (tp->t_state) { 1962 /* 1963 * If the state is SYN_RECEIVED: 1964 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1965 */ 1966 case TCPS_SYN_RECEIVED: 1967 if ((thflags & TH_ACK) && 1968 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1969 SEQ_GT(th->th_ack, tp->snd_max))) { 1970 rstreason = BANDLIM_RST_OPENPORT; 1971 goto dropwithreset; 1972 } 1973 if (IS_FASTOPEN(tp->t_flags)) { 1974 /* 1975 * When a TFO connection is in SYN_RECEIVED, the 1976 * only valid packets are the initial SYN, a 1977 * retransmit/copy of the initial SYN (possibly with 1978 * a subset of the original data), a valid ACK, a 1979 * FIN, or a RST. 1980 */ 1981 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1982 rstreason = BANDLIM_RST_OPENPORT; 1983 goto dropwithreset; 1984 } else if (thflags & TH_SYN) { 1985 /* non-initial SYN is ignored */ 1986 if ((tcp_timer_active(tp, TT_DELACK) || 1987 tcp_timer_active(tp, TT_REXMT))) 1988 goto drop; 1989 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1990 goto drop; 1991 } 1992 } 1993 break; 1994 1995 /* 1996 * If the state is SYN_SENT: 1997 * if seg contains a RST with valid ACK (SEQ.ACK has already 1998 * been verified), then drop the connection. 1999 * if seg contains a RST without an ACK, drop the seg. 2000 * if seg does not contain SYN, then drop the seg. 2001 * Otherwise this is an acceptable SYN segment 2002 * initialize tp->rcv_nxt and tp->irs 2003 * if seg contains ack then advance tp->snd_una 2004 * if seg contains an ECE and ECN support is enabled, the stream 2005 * is ECN capable. 2006 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2007 * arrange for segment to be acked (eventually) 2008 * continue processing rest of data/controls, beginning with URG 2009 */ 2010 case TCPS_SYN_SENT: 2011 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2012 TCP_PROBE5(connect__refused, NULL, tp, 2013 m, tp, th); 2014 tp = tcp_drop(tp, ECONNREFUSED); 2015 } 2016 if (thflags & TH_RST) 2017 goto drop; 2018 if (!(thflags & TH_SYN)) 2019 goto drop; 2020 2021 tp->irs = th->th_seq; 2022 tcp_rcvseqinit(tp); 2023 if (thflags & TH_ACK) { 2024 int tfo_partial_ack = 0; 2025 2026 TCPSTAT_INC(tcps_connects); 2027 soisconnected(so); 2028 #ifdef MAC 2029 mac_socketpeer_set_from_mbuf(m, so); 2030 #endif 2031 /* Do window scaling on this connection? */ 2032 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2033 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2034 tp->rcv_scale = tp->request_r_scale; 2035 } 2036 tp->rcv_adv += min(tp->rcv_wnd, 2037 TCP_MAXWIN << tp->rcv_scale); 2038 tp->snd_una++; /* SYN is acked */ 2039 /* 2040 * If not all the data that was sent in the TFO SYN 2041 * has been acked, resend the remainder right away. 2042 */ 2043 if (IS_FASTOPEN(tp->t_flags) && 2044 (tp->snd_una != tp->snd_max)) { 2045 tp->snd_nxt = th->th_ack; 2046 tfo_partial_ack = 1; 2047 } 2048 /* 2049 * If there's data, delay ACK; if there's also a FIN 2050 * ACKNOW will be turned on later. 2051 */ 2052 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2053 tcp_timer_activate(tp, TT_DELACK, 2054 tcp_delacktime); 2055 else 2056 tp->t_flags |= TF_ACKNOW; 2057 2058 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2059 (V_tcp_do_ecn == 1)) { 2060 tp->t_flags2 |= TF2_ECN_PERMIT; 2061 TCPSTAT_INC(tcps_ecn_shs); 2062 } 2063 2064 /* 2065 * Received <SYN,ACK> in SYN_SENT[*] state. 2066 * Transitions: 2067 * SYN_SENT --> ESTABLISHED 2068 * SYN_SENT* --> FIN_WAIT_1 2069 */ 2070 tp->t_starttime = ticks; 2071 if (tp->t_flags & TF_NEEDFIN) { 2072 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2073 tp->t_flags &= ~TF_NEEDFIN; 2074 thflags &= ~TH_SYN; 2075 } else { 2076 tcp_state_change(tp, TCPS_ESTABLISHED); 2077 TCP_PROBE5(connect__established, NULL, tp, 2078 m, tp, th); 2079 cc_conn_init(tp); 2080 tcp_timer_activate(tp, TT_KEEP, 2081 TP_KEEPIDLE(tp)); 2082 } 2083 } else { 2084 /* 2085 * Received initial SYN in SYN-SENT[*] state => 2086 * simultaneous open. 2087 * If it succeeds, connection is * half-synchronized. 2088 * Otherwise, do 3-way handshake: 2089 * SYN-SENT -> SYN-RECEIVED 2090 * SYN-SENT* -> SYN-RECEIVED* 2091 */ 2092 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2093 tcp_timer_activate(tp, TT_REXMT, 0); 2094 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2095 } 2096 2097 INP_WLOCK_ASSERT(tp->t_inpcb); 2098 2099 /* 2100 * Advance th->th_seq to correspond to first data byte. 2101 * If data, trim to stay within window, 2102 * dropping FIN if necessary. 2103 */ 2104 th->th_seq++; 2105 if (tlen > tp->rcv_wnd) { 2106 todrop = tlen - tp->rcv_wnd; 2107 m_adj(m, -todrop); 2108 tlen = tp->rcv_wnd; 2109 thflags &= ~TH_FIN; 2110 TCPSTAT_INC(tcps_rcvpackafterwin); 2111 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2112 } 2113 tp->snd_wl1 = th->th_seq - 1; 2114 tp->rcv_up = th->th_seq; 2115 /* 2116 * Client side of transaction: already sent SYN and data. 2117 * If the remote host used T/TCP to validate the SYN, 2118 * our data will be ACK'd; if so, enter normal data segment 2119 * processing in the middle of step 5, ack processing. 2120 * Otherwise, goto step 6. 2121 */ 2122 if (thflags & TH_ACK) 2123 goto process_ACK; 2124 2125 goto step6; 2126 2127 /* 2128 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2129 * do normal processing. 2130 * 2131 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2132 */ 2133 case TCPS_LAST_ACK: 2134 case TCPS_CLOSING: 2135 break; /* continue normal processing */ 2136 } 2137 2138 /* 2139 * States other than LISTEN or SYN_SENT. 2140 * First check the RST flag and sequence number since reset segments 2141 * are exempt from the timestamp and connection count tests. This 2142 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2143 * below which allowed reset segments in half the sequence space 2144 * to fall though and be processed (which gives forged reset 2145 * segments with a random sequence number a 50 percent chance of 2146 * killing a connection). 2147 * Then check timestamp, if present. 2148 * Then check the connection count, if present. 2149 * Then check that at least some bytes of segment are within 2150 * receive window. If segment begins before rcv_nxt, 2151 * drop leading data (and SYN); if nothing left, just ack. 2152 */ 2153 if (thflags & TH_RST) { 2154 /* 2155 * RFC5961 Section 3.2 2156 * 2157 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2158 * - If RST is in window, we send challenge ACK. 2159 * 2160 * Note: to take into account delayed ACKs, we should 2161 * test against last_ack_sent instead of rcv_nxt. 2162 * Note 2: we handle special case of closed window, not 2163 * covered by the RFC. 2164 */ 2165 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2166 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2167 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2168 KASSERT(tp->t_state != TCPS_SYN_SENT, 2169 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2170 __func__, th, tp)); 2171 2172 if (V_tcp_insecure_rst || 2173 tp->last_ack_sent == th->th_seq) { 2174 TCPSTAT_INC(tcps_drops); 2175 /* Drop the connection. */ 2176 switch (tp->t_state) { 2177 case TCPS_SYN_RECEIVED: 2178 so->so_error = ECONNREFUSED; 2179 goto close; 2180 case TCPS_ESTABLISHED: 2181 case TCPS_FIN_WAIT_1: 2182 case TCPS_FIN_WAIT_2: 2183 case TCPS_CLOSE_WAIT: 2184 case TCPS_CLOSING: 2185 case TCPS_LAST_ACK: 2186 so->so_error = ECONNRESET; 2187 close: 2188 /* FALLTHROUGH */ 2189 default: 2190 tp = tcp_close(tp); 2191 } 2192 } else { 2193 TCPSTAT_INC(tcps_badrst); 2194 /* Send challenge ACK. */ 2195 tcp_respond(tp, mtod(m, void *), th, m, 2196 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2197 tp->last_ack_sent = tp->rcv_nxt; 2198 m = NULL; 2199 } 2200 } 2201 goto drop; 2202 } 2203 2204 /* 2205 * RFC5961 Section 4.2 2206 * Send challenge ACK for any SYN in synchronized state. 2207 */ 2208 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2209 tp->t_state != TCPS_SYN_RECEIVED) { 2210 TCPSTAT_INC(tcps_badsyn); 2211 if (V_tcp_insecure_syn && 2212 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2213 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2214 tp = tcp_drop(tp, ECONNRESET); 2215 rstreason = BANDLIM_UNLIMITED; 2216 } else { 2217 /* Send challenge ACK. */ 2218 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2219 tp->snd_nxt, TH_ACK); 2220 tp->last_ack_sent = tp->rcv_nxt; 2221 m = NULL; 2222 } 2223 goto drop; 2224 } 2225 2226 /* 2227 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2228 * and it's less than ts_recent, drop it. 2229 */ 2230 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2231 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2232 /* Check to see if ts_recent is over 24 days old. */ 2233 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2234 /* 2235 * Invalidate ts_recent. If this segment updates 2236 * ts_recent, the age will be reset later and ts_recent 2237 * will get a valid value. If it does not, setting 2238 * ts_recent to zero will at least satisfy the 2239 * requirement that zero be placed in the timestamp 2240 * echo reply when ts_recent isn't valid. The 2241 * age isn't reset until we get a valid ts_recent 2242 * because we don't want out-of-order segments to be 2243 * dropped when ts_recent is old. 2244 */ 2245 tp->ts_recent = 0; 2246 } else { 2247 TCPSTAT_INC(tcps_rcvduppack); 2248 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2249 TCPSTAT_INC(tcps_pawsdrop); 2250 if (tlen) 2251 goto dropafterack; 2252 goto drop; 2253 } 2254 } 2255 2256 /* 2257 * In the SYN-RECEIVED state, validate that the packet belongs to 2258 * this connection before trimming the data to fit the receive 2259 * window. Check the sequence number versus IRS since we know 2260 * the sequence numbers haven't wrapped. This is a partial fix 2261 * for the "LAND" DoS attack. 2262 */ 2263 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2264 rstreason = BANDLIM_RST_OPENPORT; 2265 goto dropwithreset; 2266 } 2267 2268 todrop = tp->rcv_nxt - th->th_seq; 2269 if (todrop > 0) { 2270 if (thflags & TH_SYN) { 2271 thflags &= ~TH_SYN; 2272 th->th_seq++; 2273 if (th->th_urp > 1) 2274 th->th_urp--; 2275 else 2276 thflags &= ~TH_URG; 2277 todrop--; 2278 } 2279 /* 2280 * Following if statement from Stevens, vol. 2, p. 960. 2281 */ 2282 if (todrop > tlen 2283 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2284 /* 2285 * Any valid FIN must be to the left of the window. 2286 * At this point the FIN must be a duplicate or out 2287 * of sequence; drop it. 2288 */ 2289 thflags &= ~TH_FIN; 2290 2291 /* 2292 * Send an ACK to resynchronize and drop any data. 2293 * But keep on processing for RST or ACK. 2294 */ 2295 tp->t_flags |= TF_ACKNOW; 2296 todrop = tlen; 2297 TCPSTAT_INC(tcps_rcvduppack); 2298 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2299 } else { 2300 TCPSTAT_INC(tcps_rcvpartduppack); 2301 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2302 } 2303 /* 2304 * DSACK - add SACK block for dropped range 2305 */ 2306 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2307 tcp_update_sack_list(tp, th->th_seq, 2308 th->th_seq + todrop); 2309 /* 2310 * ACK now, as the next in-sequence segment 2311 * will clear the DSACK block again 2312 */ 2313 tp->t_flags |= TF_ACKNOW; 2314 } 2315 drop_hdrlen += todrop; /* drop from the top afterwards */ 2316 th->th_seq += todrop; 2317 tlen -= todrop; 2318 if (th->th_urp > todrop) 2319 th->th_urp -= todrop; 2320 else { 2321 thflags &= ~TH_URG; 2322 th->th_urp = 0; 2323 } 2324 } 2325 2326 /* 2327 * If new data are received on a connection after the 2328 * user processes are gone, then RST the other end. 2329 */ 2330 if ((so->so_state & SS_NOFDREF) && 2331 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2332 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2333 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2334 "after socket was closed, " 2335 "sending RST and removing tcpcb\n", 2336 s, __func__, tcpstates[tp->t_state], tlen); 2337 free(s, M_TCPLOG); 2338 } 2339 tp = tcp_close(tp); 2340 TCPSTAT_INC(tcps_rcvafterclose); 2341 rstreason = BANDLIM_UNLIMITED; 2342 goto dropwithreset; 2343 } 2344 2345 /* 2346 * If segment ends after window, drop trailing data 2347 * (and PUSH and FIN); if nothing left, just ACK. 2348 */ 2349 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2350 if (todrop > 0) { 2351 TCPSTAT_INC(tcps_rcvpackafterwin); 2352 if (todrop >= tlen) { 2353 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2354 /* 2355 * If window is closed can only take segments at 2356 * window edge, and have to drop data and PUSH from 2357 * incoming segments. Continue processing, but 2358 * remember to ack. Otherwise, drop segment 2359 * and ack. 2360 */ 2361 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2362 tp->t_flags |= TF_ACKNOW; 2363 TCPSTAT_INC(tcps_rcvwinprobe); 2364 } else 2365 goto dropafterack; 2366 } else 2367 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2368 m_adj(m, -todrop); 2369 tlen -= todrop; 2370 thflags &= ~(TH_PUSH|TH_FIN); 2371 } 2372 2373 /* 2374 * If last ACK falls within this segment's sequence numbers, 2375 * record its timestamp. 2376 * NOTE: 2377 * 1) That the test incorporates suggestions from the latest 2378 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2379 * 2) That updating only on newer timestamps interferes with 2380 * our earlier PAWS tests, so this check should be solely 2381 * predicated on the sequence space of this segment. 2382 * 3) That we modify the segment boundary check to be 2383 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2384 * instead of RFC1323's 2385 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2386 * This modified check allows us to overcome RFC1323's 2387 * limitations as described in Stevens TCP/IP Illustrated 2388 * Vol. 2 p.869. In such cases, we can still calculate the 2389 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2390 */ 2391 if ((to.to_flags & TOF_TS) != 0 && 2392 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2393 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2394 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2395 tp->ts_recent_age = tcp_ts_getticks(); 2396 tp->ts_recent = to.to_tsval; 2397 } 2398 2399 /* 2400 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2401 * flag is on (half-synchronized state), then queue data for 2402 * later processing; else drop segment and return. 2403 */ 2404 if ((thflags & TH_ACK) == 0) { 2405 if (tp->t_state == TCPS_SYN_RECEIVED || 2406 (tp->t_flags & TF_NEEDSYN)) { 2407 if (tp->t_state == TCPS_SYN_RECEIVED && 2408 IS_FASTOPEN(tp->t_flags)) { 2409 tp->snd_wnd = tiwin; 2410 cc_conn_init(tp); 2411 } 2412 goto step6; 2413 } else if (tp->t_flags & TF_ACKNOW) 2414 goto dropafterack; 2415 else 2416 goto drop; 2417 } 2418 2419 /* 2420 * Ack processing. 2421 */ 2422 switch (tp->t_state) { 2423 /* 2424 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2425 * ESTABLISHED state and continue processing. 2426 * The ACK was checked above. 2427 */ 2428 case TCPS_SYN_RECEIVED: 2429 2430 TCPSTAT_INC(tcps_connects); 2431 soisconnected(so); 2432 /* Do window scaling? */ 2433 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2434 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2435 tp->rcv_scale = tp->request_r_scale; 2436 } 2437 tp->snd_wnd = tiwin; 2438 /* 2439 * Make transitions: 2440 * SYN-RECEIVED -> ESTABLISHED 2441 * SYN-RECEIVED* -> FIN-WAIT-1 2442 */ 2443 tp->t_starttime = ticks; 2444 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2445 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2446 tp->t_tfo_pending = NULL; 2447 } 2448 if (tp->t_flags & TF_NEEDFIN) { 2449 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2450 tp->t_flags &= ~TF_NEEDFIN; 2451 } else { 2452 tcp_state_change(tp, TCPS_ESTABLISHED); 2453 TCP_PROBE5(accept__established, NULL, tp, 2454 m, tp, th); 2455 /* 2456 * TFO connections call cc_conn_init() during SYN 2457 * processing. Calling it again here for such 2458 * connections is not harmless as it would undo the 2459 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2460 * is retransmitted. 2461 */ 2462 if (!IS_FASTOPEN(tp->t_flags)) 2463 cc_conn_init(tp); 2464 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2465 } 2466 /* 2467 * Account for the ACK of our SYN prior to 2468 * regular ACK processing below, except for 2469 * simultaneous SYN, which is handled later. 2470 */ 2471 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2472 incforsyn = 1; 2473 /* 2474 * If segment contains data or ACK, will call tcp_reass() 2475 * later; if not, do so now to pass queued data to user. 2476 */ 2477 if (tlen == 0 && (thflags & TH_FIN) == 0) 2478 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2479 (struct mbuf *)0); 2480 tp->snd_wl1 = th->th_seq - 1; 2481 /* FALLTHROUGH */ 2482 2483 /* 2484 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2485 * ACKs. If the ack is in the range 2486 * tp->snd_una < th->th_ack <= tp->snd_max 2487 * then advance tp->snd_una to th->th_ack and drop 2488 * data from the retransmission queue. If this ACK reflects 2489 * more up to date window information we update our window information. 2490 */ 2491 case TCPS_ESTABLISHED: 2492 case TCPS_FIN_WAIT_1: 2493 case TCPS_FIN_WAIT_2: 2494 case TCPS_CLOSE_WAIT: 2495 case TCPS_CLOSING: 2496 case TCPS_LAST_ACK: 2497 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2498 TCPSTAT_INC(tcps_rcvacktoomuch); 2499 goto dropafterack; 2500 } 2501 if ((tp->t_flags & TF_SACK_PERMIT) && 2502 ((to.to_flags & TOF_SACK) || 2503 !TAILQ_EMPTY(&tp->snd_holes))) 2504 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2505 else 2506 /* 2507 * Reset the value so that previous (valid) value 2508 * from the last ack with SACK doesn't get used. 2509 */ 2510 tp->sackhint.sacked_bytes = 0; 2511 2512 #ifdef TCP_HHOOK 2513 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2514 hhook_run_tcp_est_in(tp, th, &to); 2515 #endif 2516 2517 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2518 maxseg = tcp_maxseg(tp); 2519 if (tlen == 0 && 2520 (tiwin == tp->snd_wnd || 2521 (tp->t_flags & TF_SACK_PERMIT))) { 2522 /* 2523 * If this is the first time we've seen a 2524 * FIN from the remote, this is not a 2525 * duplicate and it needs to be processed 2526 * normally. This happens during a 2527 * simultaneous close. 2528 */ 2529 if ((thflags & TH_FIN) && 2530 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2531 tp->t_dupacks = 0; 2532 break; 2533 } 2534 TCPSTAT_INC(tcps_rcvdupack); 2535 /* 2536 * If we have outstanding data (other than 2537 * a window probe), this is a completely 2538 * duplicate ack (ie, window info didn't 2539 * change and FIN isn't set), 2540 * the ack is the biggest we've 2541 * seen and we've seen exactly our rexmt 2542 * threshold of them, assume a packet 2543 * has been dropped and retransmit it. 2544 * Kludge snd_nxt & the congestion 2545 * window so we send only this one 2546 * packet. 2547 * 2548 * We know we're losing at the current 2549 * window size so do congestion avoidance 2550 * (set ssthresh to half the current window 2551 * and pull our congestion window back to 2552 * the new ssthresh). 2553 * 2554 * Dup acks mean that packets have left the 2555 * network (they're now cached at the receiver) 2556 * so bump cwnd by the amount in the receiver 2557 * to keep a constant cwnd packets in the 2558 * network. 2559 * 2560 * When using TCP ECN, notify the peer that 2561 * we reduced the cwnd. 2562 */ 2563 /* 2564 * Following 2 kinds of acks should not affect 2565 * dupack counting: 2566 * 1) Old acks 2567 * 2) Acks with SACK but without any new SACK 2568 * information in them. These could result from 2569 * any anomaly in the network like a switch 2570 * duplicating packets or a possible DoS attack. 2571 */ 2572 if (th->th_ack != tp->snd_una || 2573 ((tp->t_flags & TF_SACK_PERMIT) && 2574 (to.to_flags & TOF_SACK) && 2575 !sack_changed)) 2576 break; 2577 else if (!tcp_timer_active(tp, TT_REXMT)) 2578 tp->t_dupacks = 0; 2579 else if (++tp->t_dupacks > tcprexmtthresh || 2580 IN_FASTRECOVERY(tp->t_flags)) { 2581 cc_ack_received(tp, th, nsegs, 2582 CC_DUPACK); 2583 if (V_tcp_do_prr && 2584 IN_FASTRECOVERY(tp->t_flags) && 2585 (tp->t_flags & TF_SACK_PERMIT)) { 2586 tcp_do_prr_ack(tp, th, &to); 2587 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2588 (to.to_flags & TOF_SACK) && 2589 IN_FASTRECOVERY(tp->t_flags)) { 2590 int awnd; 2591 2592 /* 2593 * Compute the amount of data in flight first. 2594 * We can inject new data into the pipe iff 2595 * we have less than 1/2 the original window's 2596 * worth of data in flight. 2597 */ 2598 if (V_tcp_do_rfc6675_pipe) 2599 awnd = tcp_compute_pipe(tp); 2600 else 2601 awnd = (tp->snd_nxt - tp->snd_fack) + 2602 tp->sackhint.sack_bytes_rexmit; 2603 2604 if (awnd < tp->snd_ssthresh) { 2605 tp->snd_cwnd += maxseg; 2606 if (tp->snd_cwnd > tp->snd_ssthresh) 2607 tp->snd_cwnd = tp->snd_ssthresh; 2608 } 2609 } else 2610 tp->snd_cwnd += maxseg; 2611 (void) tp->t_fb->tfb_tcp_output(tp); 2612 goto drop; 2613 } else if (tp->t_dupacks == tcprexmtthresh || 2614 (tp->t_flags & TF_SACK_PERMIT && 2615 V_tcp_do_rfc6675_pipe && 2616 tp->sackhint.sacked_bytes > 2617 (tcprexmtthresh - 1) * maxseg)) { 2618 enter_recovery: 2619 /* 2620 * Above is the RFC6675 trigger condition of 2621 * more than (dupthresh-1)*maxseg sacked data. 2622 * If the count of holes in the 2623 * scoreboard is >= dupthresh, we could 2624 * also enter loss recovery, but don't 2625 * have that value readily available. 2626 */ 2627 tp->t_dupacks = tcprexmtthresh; 2628 tcp_seq onxt = tp->snd_nxt; 2629 2630 /* 2631 * If we're doing sack, or prr, check 2632 * to see if we're already in sack 2633 * recovery. If we're not doing sack, 2634 * check to see if we're in newreno 2635 * recovery. 2636 */ 2637 if (V_tcp_do_prr || 2638 (tp->t_flags & TF_SACK_PERMIT)) { 2639 if (IN_FASTRECOVERY(tp->t_flags)) { 2640 tp->t_dupacks = 0; 2641 break; 2642 } 2643 } else { 2644 if (SEQ_LEQ(th->th_ack, 2645 tp->snd_recover)) { 2646 tp->t_dupacks = 0; 2647 break; 2648 } 2649 } 2650 /* Congestion signal before ack. */ 2651 cc_cong_signal(tp, th, CC_NDUPACK); 2652 cc_ack_received(tp, th, nsegs, 2653 CC_DUPACK); 2654 tcp_timer_activate(tp, TT_REXMT, 0); 2655 tp->t_rtttime = 0; 2656 if (V_tcp_do_prr) { 2657 /* 2658 * snd_ssthresh is already updated by 2659 * cc_cong_signal. 2660 */ 2661 tp->sackhint.prr_delivered = 2662 tp->sackhint.sacked_bytes; 2663 tp->sackhint.recover_fs = max(1, 2664 tp->snd_nxt - tp->snd_una); 2665 } 2666 if ((tp->t_flags & TF_SACK_PERMIT) && 2667 (to.to_flags & TOF_SACK)) { 2668 TCPSTAT_INC( 2669 tcps_sack_recovery_episode); 2670 tp->snd_recover = tp->snd_nxt; 2671 tp->snd_cwnd = maxseg; 2672 (void) tp->t_fb->tfb_tcp_output(tp); 2673 if (SEQ_GT(th->th_ack, tp->snd_una)) 2674 goto resume_partialack; 2675 goto drop; 2676 } 2677 tp->snd_nxt = th->th_ack; 2678 tp->snd_cwnd = maxseg; 2679 (void) tp->t_fb->tfb_tcp_output(tp); 2680 KASSERT(tp->snd_limited <= 2, 2681 ("%s: tp->snd_limited too big", 2682 __func__)); 2683 tp->snd_cwnd = tp->snd_ssthresh + 2684 maxseg * 2685 (tp->t_dupacks - tp->snd_limited); 2686 if (SEQ_GT(onxt, tp->snd_nxt)) 2687 tp->snd_nxt = onxt; 2688 goto drop; 2689 } else if (V_tcp_do_rfc3042) { 2690 /* 2691 * Process first and second duplicate 2692 * ACKs. Each indicates a segment 2693 * leaving the network, creating room 2694 * for more. Make sure we can send a 2695 * packet on reception of each duplicate 2696 * ACK by increasing snd_cwnd by one 2697 * segment. Restore the original 2698 * snd_cwnd after packet transmission. 2699 */ 2700 cc_ack_received(tp, th, nsegs, 2701 CC_DUPACK); 2702 uint32_t oldcwnd = tp->snd_cwnd; 2703 tcp_seq oldsndmax = tp->snd_max; 2704 u_int sent; 2705 int avail; 2706 2707 KASSERT(tp->t_dupacks == 1 || 2708 tp->t_dupacks == 2, 2709 ("%s: dupacks not 1 or 2", 2710 __func__)); 2711 if (tp->t_dupacks == 1) 2712 tp->snd_limited = 0; 2713 tp->snd_cwnd = 2714 (tp->snd_nxt - tp->snd_una) + 2715 (tp->t_dupacks - tp->snd_limited) * 2716 maxseg; 2717 /* 2718 * Only call tcp_output when there 2719 * is new data available to be sent. 2720 * Otherwise we would send pure ACKs. 2721 */ 2722 SOCKBUF_LOCK(&so->so_snd); 2723 avail = sbavail(&so->so_snd) - 2724 (tp->snd_nxt - tp->snd_una); 2725 SOCKBUF_UNLOCK(&so->so_snd); 2726 if (avail > 0) 2727 (void) tp->t_fb->tfb_tcp_output(tp); 2728 sent = tp->snd_max - oldsndmax; 2729 if (sent > maxseg) { 2730 KASSERT((tp->t_dupacks == 2 && 2731 tp->snd_limited == 0) || 2732 (sent == maxseg + 1 && 2733 tp->t_flags & TF_SENTFIN), 2734 ("%s: sent too much", 2735 __func__)); 2736 tp->snd_limited = 2; 2737 } else if (sent > 0) 2738 ++tp->snd_limited; 2739 tp->snd_cwnd = oldcwnd; 2740 goto drop; 2741 } 2742 } 2743 break; 2744 } else { 2745 /* 2746 * This ack is advancing the left edge, reset the 2747 * counter. 2748 */ 2749 tp->t_dupacks = 0; 2750 /* 2751 * If this ack also has new SACK info, increment the 2752 * counter as per rfc6675. The variable 2753 * sack_changed tracks all changes to the SACK 2754 * scoreboard, including when partial ACKs without 2755 * SACK options are received, and clear the scoreboard 2756 * from the left side. Such partial ACKs should not be 2757 * counted as dupacks here. 2758 */ 2759 if ((tp->t_flags & TF_SACK_PERMIT) && 2760 (to.to_flags & TOF_SACK) && 2761 sack_changed) { 2762 tp->t_dupacks++; 2763 /* limit overhead by setting maxseg last */ 2764 if (!IN_FASTRECOVERY(tp->t_flags) && 2765 (tp->sackhint.sacked_bytes > 2766 ((tcprexmtthresh - 1) * 2767 (maxseg = tcp_maxseg(tp))))) { 2768 goto enter_recovery; 2769 } 2770 } 2771 } 2772 2773 resume_partialack: 2774 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2775 ("%s: th_ack <= snd_una", __func__)); 2776 2777 /* 2778 * If the congestion window was inflated to account 2779 * for the other side's cached packets, retract it. 2780 */ 2781 if (IN_FASTRECOVERY(tp->t_flags)) { 2782 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2783 if (tp->t_flags & TF_SACK_PERMIT) 2784 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2785 tcp_timer_activate(tp, TT_REXMT, 0); 2786 tp->t_rtttime = 0; 2787 tcp_do_prr_ack(tp, th, &to); 2788 tp->t_flags |= TF_ACKNOW; 2789 (void) tcp_output(tp); 2790 } else 2791 tcp_sack_partialack(tp, th); 2792 else 2793 tcp_newreno_partial_ack(tp, th); 2794 } else 2795 cc_post_recovery(tp, th); 2796 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2797 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2798 if (V_tcp_do_prr) { 2799 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2800 tp->snd_fack = th->th_ack; 2801 tcp_do_prr_ack(tp, th, &to); 2802 (void) tcp_output(tp); 2803 } 2804 } else 2805 cc_post_recovery(tp, th); 2806 } 2807 /* 2808 * If we reach this point, ACK is not a duplicate, 2809 * i.e., it ACKs something we sent. 2810 */ 2811 if (tp->t_flags & TF_NEEDSYN) { 2812 /* 2813 * T/TCP: Connection was half-synchronized, and our 2814 * SYN has been ACK'd (so connection is now fully 2815 * synchronized). Go to non-starred state, 2816 * increment snd_una for ACK of SYN, and check if 2817 * we can do window scaling. 2818 */ 2819 tp->t_flags &= ~TF_NEEDSYN; 2820 tp->snd_una++; 2821 /* Do window scaling? */ 2822 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2823 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2824 tp->rcv_scale = tp->request_r_scale; 2825 /* Send window already scaled. */ 2826 } 2827 } 2828 2829 process_ACK: 2830 INP_WLOCK_ASSERT(tp->t_inpcb); 2831 2832 /* 2833 * Adjust for the SYN bit in sequence space, 2834 * but don't account for it in cwnd calculations. 2835 * This is for the SYN_RECEIVED, non-simultaneous 2836 * SYN case. SYN_SENT and simultaneous SYN are 2837 * treated elsewhere. 2838 */ 2839 if (incforsyn) 2840 tp->snd_una++; 2841 acked = BYTES_THIS_ACK(tp, th); 2842 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2843 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2844 tp->snd_una, th->th_ack, tp, m)); 2845 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2846 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2847 2848 /* 2849 * If we just performed our first retransmit, and the ACK 2850 * arrives within our recovery window, then it was a mistake 2851 * to do the retransmit in the first place. Recover our 2852 * original cwnd and ssthresh, and proceed to transmit where 2853 * we left off. 2854 */ 2855 if (tp->t_rxtshift == 1 && 2856 tp->t_flags & TF_PREVVALID && 2857 tp->t_badrxtwin && 2858 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2859 cc_cong_signal(tp, th, CC_RTO_ERR); 2860 2861 /* 2862 * If we have a timestamp reply, update smoothed 2863 * round trip time. If no timestamp is present but 2864 * transmit timer is running and timed sequence 2865 * number was acked, update smoothed round trip time. 2866 * Since we now have an rtt measurement, cancel the 2867 * timer backoff (cf., Phil Karn's retransmit alg.). 2868 * Recompute the initial retransmit timer. 2869 * 2870 * Some boxes send broken timestamp replies 2871 * during the SYN+ACK phase, ignore 2872 * timestamps of 0 or we could calculate a 2873 * huge RTT and blow up the retransmit timer. 2874 */ 2875 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2876 uint32_t t; 2877 2878 t = tcp_ts_getticks() - to.to_tsecr; 2879 if (!tp->t_rttlow || tp->t_rttlow > t) 2880 tp->t_rttlow = t; 2881 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2882 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2883 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2884 tp->t_rttlow = ticks - tp->t_rtttime; 2885 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2886 } 2887 2888 /* 2889 * If all outstanding data is acked, stop retransmit 2890 * timer and remember to restart (more output or persist). 2891 * If there is more data to be acked, restart retransmit 2892 * timer, using current (possibly backed-off) value. 2893 */ 2894 if (th->th_ack == tp->snd_max) { 2895 tcp_timer_activate(tp, TT_REXMT, 0); 2896 needoutput = 1; 2897 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2898 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2899 2900 /* 2901 * If no data (only SYN) was ACK'd, 2902 * skip rest of ACK processing. 2903 */ 2904 if (acked == 0) 2905 goto step6; 2906 2907 /* 2908 * Let the congestion control algorithm update congestion 2909 * control related information. This typically means increasing 2910 * the congestion window. 2911 */ 2912 cc_ack_received(tp, th, nsegs, CC_ACK); 2913 2914 SOCKBUF_LOCK(&so->so_snd); 2915 if (acked > sbavail(&so->so_snd)) { 2916 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2917 tp->snd_wnd -= sbavail(&so->so_snd); 2918 else 2919 tp->snd_wnd = 0; 2920 mfree = sbcut_locked(&so->so_snd, 2921 (int)sbavail(&so->so_snd)); 2922 ourfinisacked = 1; 2923 } else { 2924 mfree = sbcut_locked(&so->so_snd, acked); 2925 if (tp->snd_wnd >= (uint32_t) acked) 2926 tp->snd_wnd -= acked; 2927 else 2928 tp->snd_wnd = 0; 2929 ourfinisacked = 0; 2930 } 2931 SOCKBUF_UNLOCK(&so->so_snd); 2932 tp->t_flags |= TF_WAKESOW; 2933 m_freem(mfree); 2934 /* Detect una wraparound. */ 2935 if (!IN_RECOVERY(tp->t_flags) && 2936 SEQ_GT(tp->snd_una, tp->snd_recover) && 2937 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2938 tp->snd_recover = th->th_ack - 1; 2939 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2940 if (IN_RECOVERY(tp->t_flags) && 2941 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2942 EXIT_RECOVERY(tp->t_flags); 2943 } 2944 tp->snd_una = th->th_ack; 2945 if (tp->t_flags & TF_SACK_PERMIT) { 2946 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2947 tp->snd_recover = tp->snd_una; 2948 } 2949 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2950 tp->snd_nxt = tp->snd_una; 2951 2952 switch (tp->t_state) { 2953 /* 2954 * In FIN_WAIT_1 STATE in addition to the processing 2955 * for the ESTABLISHED state if our FIN is now acknowledged 2956 * then enter FIN_WAIT_2. 2957 */ 2958 case TCPS_FIN_WAIT_1: 2959 if (ourfinisacked) { 2960 /* 2961 * If we can't receive any more 2962 * data, then closing user can proceed. 2963 * Starting the timer is contrary to the 2964 * specification, but if we don't get a FIN 2965 * we'll hang forever. 2966 * 2967 * XXXjl: 2968 * we should release the tp also, and use a 2969 * compressed state. 2970 */ 2971 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2972 soisdisconnected(so); 2973 tcp_timer_activate(tp, TT_2MSL, 2974 (tcp_fast_finwait2_recycle ? 2975 tcp_finwait2_timeout : 2976 TP_MAXIDLE(tp))); 2977 } 2978 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2979 } 2980 break; 2981 2982 /* 2983 * In CLOSING STATE in addition to the processing for 2984 * the ESTABLISHED state if the ACK acknowledges our FIN 2985 * then enter the TIME-WAIT state, otherwise ignore 2986 * the segment. 2987 */ 2988 case TCPS_CLOSING: 2989 if (ourfinisacked) { 2990 tcp_twstart(tp); 2991 m_freem(m); 2992 return; 2993 } 2994 break; 2995 2996 /* 2997 * In LAST_ACK, we may still be waiting for data to drain 2998 * and/or to be acked, as well as for the ack of our FIN. 2999 * If our FIN is now acknowledged, delete the TCB, 3000 * enter the closed state and return. 3001 */ 3002 case TCPS_LAST_ACK: 3003 if (ourfinisacked) { 3004 tp = tcp_close(tp); 3005 goto drop; 3006 } 3007 break; 3008 } 3009 } 3010 3011 step6: 3012 INP_WLOCK_ASSERT(tp->t_inpcb); 3013 3014 /* 3015 * Update window information. 3016 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3017 */ 3018 if ((thflags & TH_ACK) && 3019 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3020 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3021 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3022 /* keep track of pure window updates */ 3023 if (tlen == 0 && 3024 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3025 TCPSTAT_INC(tcps_rcvwinupd); 3026 tp->snd_wnd = tiwin; 3027 tp->snd_wl1 = th->th_seq; 3028 tp->snd_wl2 = th->th_ack; 3029 if (tp->snd_wnd > tp->max_sndwnd) 3030 tp->max_sndwnd = tp->snd_wnd; 3031 needoutput = 1; 3032 } 3033 3034 /* 3035 * Process segments with URG. 3036 */ 3037 if ((thflags & TH_URG) && th->th_urp && 3038 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3039 /* 3040 * This is a kludge, but if we receive and accept 3041 * random urgent pointers, we'll crash in 3042 * soreceive. It's hard to imagine someone 3043 * actually wanting to send this much urgent data. 3044 */ 3045 SOCKBUF_LOCK(&so->so_rcv); 3046 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3047 th->th_urp = 0; /* XXX */ 3048 thflags &= ~TH_URG; /* XXX */ 3049 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3050 goto dodata; /* XXX */ 3051 } 3052 /* 3053 * If this segment advances the known urgent pointer, 3054 * then mark the data stream. This should not happen 3055 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3056 * a FIN has been received from the remote side. 3057 * In these states we ignore the URG. 3058 * 3059 * According to RFC961 (Assigned Protocols), 3060 * the urgent pointer points to the last octet 3061 * of urgent data. We continue, however, 3062 * to consider it to indicate the first octet 3063 * of data past the urgent section as the original 3064 * spec states (in one of two places). 3065 */ 3066 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3067 tp->rcv_up = th->th_seq + th->th_urp; 3068 so->so_oobmark = sbavail(&so->so_rcv) + 3069 (tp->rcv_up - tp->rcv_nxt) - 1; 3070 if (so->so_oobmark == 0) 3071 so->so_rcv.sb_state |= SBS_RCVATMARK; 3072 sohasoutofband(so); 3073 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3074 } 3075 SOCKBUF_UNLOCK(&so->so_rcv); 3076 /* 3077 * Remove out of band data so doesn't get presented to user. 3078 * This can happen independent of advancing the URG pointer, 3079 * but if two URG's are pending at once, some out-of-band 3080 * data may creep in... ick. 3081 */ 3082 if (th->th_urp <= (uint32_t)tlen && 3083 !(so->so_options & SO_OOBINLINE)) { 3084 /* hdr drop is delayed */ 3085 tcp_pulloutofband(so, th, m, drop_hdrlen); 3086 } 3087 } else { 3088 /* 3089 * If no out of band data is expected, 3090 * pull receive urgent pointer along 3091 * with the receive window. 3092 */ 3093 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3094 tp->rcv_up = tp->rcv_nxt; 3095 } 3096 dodata: /* XXX */ 3097 INP_WLOCK_ASSERT(tp->t_inpcb); 3098 3099 /* 3100 * Process the segment text, merging it into the TCP sequencing queue, 3101 * and arranging for acknowledgment of receipt if necessary. 3102 * This process logically involves adjusting tp->rcv_wnd as data 3103 * is presented to the user (this happens in tcp_usrreq.c, 3104 * case PRU_RCVD). If a FIN has already been received on this 3105 * connection then we just ignore the text. 3106 */ 3107 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3108 IS_FASTOPEN(tp->t_flags)); 3109 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3110 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3111 tcp_seq save_start = th->th_seq; 3112 tcp_seq save_rnxt = tp->rcv_nxt; 3113 int save_tlen = tlen; 3114 m_adj(m, drop_hdrlen); /* delayed header drop */ 3115 /* 3116 * Insert segment which includes th into TCP reassembly queue 3117 * with control block tp. Set thflags to whether reassembly now 3118 * includes a segment with FIN. This handles the common case 3119 * inline (segment is the next to be received on an established 3120 * connection, and the queue is empty), avoiding linkage into 3121 * and removal from the queue and repetition of various 3122 * conversions. 3123 * Set DELACK for segments received in order, but ack 3124 * immediately when segments are out of order (so 3125 * fast retransmit can work). 3126 */ 3127 if (th->th_seq == tp->rcv_nxt && 3128 SEGQ_EMPTY(tp) && 3129 (TCPS_HAVEESTABLISHED(tp->t_state) || 3130 tfo_syn)) { 3131 if (DELAY_ACK(tp, tlen) || tfo_syn) 3132 tp->t_flags |= TF_DELACK; 3133 else 3134 tp->t_flags |= TF_ACKNOW; 3135 tp->rcv_nxt += tlen; 3136 if (tlen && 3137 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3138 (tp->t_fbyte_in == 0)) { 3139 tp->t_fbyte_in = ticks; 3140 if (tp->t_fbyte_in == 0) 3141 tp->t_fbyte_in = 1; 3142 if (tp->t_fbyte_out && tp->t_fbyte_in) 3143 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3144 } 3145 thflags = th->th_flags & TH_FIN; 3146 TCPSTAT_INC(tcps_rcvpack); 3147 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3148 SOCKBUF_LOCK(&so->so_rcv); 3149 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3150 m_freem(m); 3151 else 3152 sbappendstream_locked(&so->so_rcv, m, 0); 3153 SOCKBUF_UNLOCK(&so->so_rcv); 3154 tp->t_flags |= TF_WAKESOR; 3155 } else { 3156 /* 3157 * XXX: Due to the header drop above "th" is 3158 * theoretically invalid by now. Fortunately 3159 * m_adj() doesn't actually frees any mbufs 3160 * when trimming from the head. 3161 */ 3162 tcp_seq temp = save_start; 3163 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3164 tp->t_flags |= TF_ACKNOW; 3165 } 3166 if ((tp->t_flags & TF_SACK_PERMIT) && 3167 (save_tlen > 0) && 3168 TCPS_HAVEESTABLISHED(tp->t_state)) { 3169 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3170 /* 3171 * DSACK actually handled in the fastpath 3172 * above. 3173 */ 3174 tcp_update_sack_list(tp, save_start, 3175 save_start + save_tlen); 3176 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3177 if ((tp->rcv_numsacks >= 1) && 3178 (tp->sackblks[0].end == save_start)) { 3179 /* 3180 * Partial overlap, recorded at todrop 3181 * above. 3182 */ 3183 tcp_update_sack_list(tp, 3184 tp->sackblks[0].start, 3185 tp->sackblks[0].end); 3186 } else { 3187 tcp_update_dsack_list(tp, save_start, 3188 save_start + save_tlen); 3189 } 3190 } else if (tlen >= save_tlen) { 3191 /* Update of sackblks. */ 3192 tcp_update_dsack_list(tp, save_start, 3193 save_start + save_tlen); 3194 } else if (tlen > 0) { 3195 tcp_update_dsack_list(tp, save_start, 3196 save_start + tlen); 3197 } 3198 } 3199 #if 0 3200 /* 3201 * Note the amount of data that peer has sent into 3202 * our window, in order to estimate the sender's 3203 * buffer size. 3204 * XXX: Unused. 3205 */ 3206 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3207 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3208 else 3209 len = so->so_rcv.sb_hiwat; 3210 #endif 3211 } else { 3212 m_freem(m); 3213 thflags &= ~TH_FIN; 3214 } 3215 3216 /* 3217 * If FIN is received ACK the FIN and let the user know 3218 * that the connection is closing. 3219 */ 3220 if (thflags & TH_FIN) { 3221 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3222 socantrcvmore(so); 3223 /* The socket upcall is handled by socantrcvmore. */ 3224 tp->t_flags &= ~TF_WAKESOR; 3225 /* 3226 * If connection is half-synchronized 3227 * (ie NEEDSYN flag on) then delay ACK, 3228 * so it may be piggybacked when SYN is sent. 3229 * Otherwise, since we received a FIN then no 3230 * more input can be expected, send ACK now. 3231 */ 3232 if (tp->t_flags & TF_NEEDSYN) 3233 tp->t_flags |= TF_DELACK; 3234 else 3235 tp->t_flags |= TF_ACKNOW; 3236 tp->rcv_nxt++; 3237 } 3238 switch (tp->t_state) { 3239 /* 3240 * In SYN_RECEIVED and ESTABLISHED STATES 3241 * enter the CLOSE_WAIT state. 3242 */ 3243 case TCPS_SYN_RECEIVED: 3244 tp->t_starttime = ticks; 3245 /* FALLTHROUGH */ 3246 case TCPS_ESTABLISHED: 3247 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3248 break; 3249 3250 /* 3251 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3252 * enter the CLOSING state. 3253 */ 3254 case TCPS_FIN_WAIT_1: 3255 tcp_state_change(tp, TCPS_CLOSING); 3256 break; 3257 3258 /* 3259 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3260 * starting the time-wait timer, turning off the other 3261 * standard timers. 3262 */ 3263 case TCPS_FIN_WAIT_2: 3264 tcp_twstart(tp); 3265 return; 3266 } 3267 } 3268 #ifdef TCPDEBUG 3269 if (so->so_options & SO_DEBUG) 3270 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3271 &tcp_savetcp, 0); 3272 #endif 3273 TCP_PROBE3(debug__input, tp, th, m); 3274 3275 /* 3276 * Return any desired output. 3277 */ 3278 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3279 (void) tp->t_fb->tfb_tcp_output(tp); 3280 3281 check_delack: 3282 INP_WLOCK_ASSERT(tp->t_inpcb); 3283 3284 if (tp->t_flags & TF_DELACK) { 3285 tp->t_flags &= ~TF_DELACK; 3286 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3287 } 3288 tcp_handle_wakeup(tp, so); 3289 INP_WUNLOCK(tp->t_inpcb); 3290 return; 3291 3292 dropafterack: 3293 /* 3294 * Generate an ACK dropping incoming segment if it occupies 3295 * sequence space, where the ACK reflects our state. 3296 * 3297 * We can now skip the test for the RST flag since all 3298 * paths to this code happen after packets containing 3299 * RST have been dropped. 3300 * 3301 * In the SYN-RECEIVED state, don't send an ACK unless the 3302 * segment we received passes the SYN-RECEIVED ACK test. 3303 * If it fails send a RST. This breaks the loop in the 3304 * "LAND" DoS attack, and also prevents an ACK storm 3305 * between two listening ports that have been sent forged 3306 * SYN segments, each with the source address of the other. 3307 */ 3308 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3309 (SEQ_GT(tp->snd_una, th->th_ack) || 3310 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3311 rstreason = BANDLIM_RST_OPENPORT; 3312 goto dropwithreset; 3313 } 3314 #ifdef TCPDEBUG 3315 if (so->so_options & SO_DEBUG) 3316 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3317 &tcp_savetcp, 0); 3318 #endif 3319 TCP_PROBE3(debug__input, tp, th, m); 3320 tp->t_flags |= TF_ACKNOW; 3321 (void) tp->t_fb->tfb_tcp_output(tp); 3322 tcp_handle_wakeup(tp, so); 3323 INP_WUNLOCK(tp->t_inpcb); 3324 m_freem(m); 3325 return; 3326 3327 dropwithreset: 3328 if (tp != NULL) { 3329 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3330 tcp_handle_wakeup(tp, so); 3331 INP_WUNLOCK(tp->t_inpcb); 3332 } else 3333 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3334 return; 3335 3336 drop: 3337 /* 3338 * Drop space held by incoming segment and return. 3339 */ 3340 #ifdef TCPDEBUG 3341 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3342 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3343 &tcp_savetcp, 0); 3344 #endif 3345 TCP_PROBE3(debug__input, tp, th, m); 3346 if (tp != NULL) { 3347 tcp_handle_wakeup(tp, so); 3348 INP_WUNLOCK(tp->t_inpcb); 3349 } 3350 m_freem(m); 3351 } 3352 3353 /* 3354 * Issue RST and make ACK acceptable to originator of segment. 3355 * The mbuf must still include the original packet header. 3356 * tp may be NULL. 3357 */ 3358 void 3359 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3360 int tlen, int rstreason) 3361 { 3362 #ifdef INET 3363 struct ip *ip; 3364 #endif 3365 #ifdef INET6 3366 struct ip6_hdr *ip6; 3367 #endif 3368 3369 if (tp != NULL) { 3370 INP_LOCK_ASSERT(tp->t_inpcb); 3371 } 3372 3373 /* Don't bother if destination was broadcast/multicast. */ 3374 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3375 goto drop; 3376 #ifdef INET6 3377 if (mtod(m, struct ip *)->ip_v == 6) { 3378 ip6 = mtod(m, struct ip6_hdr *); 3379 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3380 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3381 goto drop; 3382 /* IPv6 anycast check is done at tcp6_input() */ 3383 } 3384 #endif 3385 #if defined(INET) && defined(INET6) 3386 else 3387 #endif 3388 #ifdef INET 3389 { 3390 ip = mtod(m, struct ip *); 3391 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3392 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3393 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3394 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3395 goto drop; 3396 } 3397 #endif 3398 3399 /* Perform bandwidth limiting. */ 3400 if (badport_bandlim(rstreason) < 0) 3401 goto drop; 3402 3403 /* tcp_respond consumes the mbuf chain. */ 3404 if (th->th_flags & TH_ACK) { 3405 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3406 th->th_ack, TH_RST); 3407 } else { 3408 if (th->th_flags & TH_SYN) 3409 tlen++; 3410 if (th->th_flags & TH_FIN) 3411 tlen++; 3412 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3413 (tcp_seq)0, TH_RST|TH_ACK); 3414 } 3415 return; 3416 drop: 3417 m_freem(m); 3418 } 3419 3420 /* 3421 * Parse TCP options and place in tcpopt. 3422 */ 3423 void 3424 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3425 { 3426 int opt, optlen; 3427 3428 to->to_flags = 0; 3429 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3430 opt = cp[0]; 3431 if (opt == TCPOPT_EOL) 3432 break; 3433 if (opt == TCPOPT_NOP) 3434 optlen = 1; 3435 else { 3436 if (cnt < 2) 3437 break; 3438 optlen = cp[1]; 3439 if (optlen < 2 || optlen > cnt) 3440 break; 3441 } 3442 switch (opt) { 3443 case TCPOPT_MAXSEG: 3444 if (optlen != TCPOLEN_MAXSEG) 3445 continue; 3446 if (!(flags & TO_SYN)) 3447 continue; 3448 to->to_flags |= TOF_MSS; 3449 bcopy((char *)cp + 2, 3450 (char *)&to->to_mss, sizeof(to->to_mss)); 3451 to->to_mss = ntohs(to->to_mss); 3452 break; 3453 case TCPOPT_WINDOW: 3454 if (optlen != TCPOLEN_WINDOW) 3455 continue; 3456 if (!(flags & TO_SYN)) 3457 continue; 3458 to->to_flags |= TOF_SCALE; 3459 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3460 break; 3461 case TCPOPT_TIMESTAMP: 3462 if (optlen != TCPOLEN_TIMESTAMP) 3463 continue; 3464 to->to_flags |= TOF_TS; 3465 bcopy((char *)cp + 2, 3466 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3467 to->to_tsval = ntohl(to->to_tsval); 3468 bcopy((char *)cp + 6, 3469 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3470 to->to_tsecr = ntohl(to->to_tsecr); 3471 break; 3472 case TCPOPT_SIGNATURE: 3473 /* 3474 * In order to reply to a host which has set the 3475 * TCP_SIGNATURE option in its initial SYN, we have 3476 * to record the fact that the option was observed 3477 * here for the syncache code to perform the correct 3478 * response. 3479 */ 3480 if (optlen != TCPOLEN_SIGNATURE) 3481 continue; 3482 to->to_flags |= TOF_SIGNATURE; 3483 to->to_signature = cp + 2; 3484 break; 3485 case TCPOPT_SACK_PERMITTED: 3486 if (optlen != TCPOLEN_SACK_PERMITTED) 3487 continue; 3488 if (!(flags & TO_SYN)) 3489 continue; 3490 if (!V_tcp_do_sack) 3491 continue; 3492 to->to_flags |= TOF_SACKPERM; 3493 break; 3494 case TCPOPT_SACK: 3495 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3496 continue; 3497 if (flags & TO_SYN) 3498 continue; 3499 to->to_flags |= TOF_SACK; 3500 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3501 to->to_sacks = cp + 2; 3502 TCPSTAT_INC(tcps_sack_rcv_blocks); 3503 break; 3504 case TCPOPT_FAST_OPEN: 3505 /* 3506 * Cookie length validation is performed by the 3507 * server side cookie checking code or the client 3508 * side cookie cache update code. 3509 */ 3510 if (!(flags & TO_SYN)) 3511 continue; 3512 if (!V_tcp_fastopen_client_enable && 3513 !V_tcp_fastopen_server_enable) 3514 continue; 3515 to->to_flags |= TOF_FASTOPEN; 3516 to->to_tfo_len = optlen - 2; 3517 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3518 break; 3519 default: 3520 continue; 3521 } 3522 } 3523 } 3524 3525 /* 3526 * Pull out of band byte out of a segment so 3527 * it doesn't appear in the user's data queue. 3528 * It is still reflected in the segment length for 3529 * sequencing purposes. 3530 */ 3531 void 3532 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3533 int off) 3534 { 3535 int cnt = off + th->th_urp - 1; 3536 3537 while (cnt >= 0) { 3538 if (m->m_len > cnt) { 3539 char *cp = mtod(m, caddr_t) + cnt; 3540 struct tcpcb *tp = sototcpcb(so); 3541 3542 INP_WLOCK_ASSERT(tp->t_inpcb); 3543 3544 tp->t_iobc = *cp; 3545 tp->t_oobflags |= TCPOOB_HAVEDATA; 3546 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3547 m->m_len--; 3548 if (m->m_flags & M_PKTHDR) 3549 m->m_pkthdr.len--; 3550 return; 3551 } 3552 cnt -= m->m_len; 3553 m = m->m_next; 3554 if (m == NULL) 3555 break; 3556 } 3557 panic("tcp_pulloutofband"); 3558 } 3559 3560 /* 3561 * Collect new round-trip time estimate 3562 * and update averages and current timeout. 3563 */ 3564 void 3565 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3566 { 3567 int delta; 3568 3569 INP_WLOCK_ASSERT(tp->t_inpcb); 3570 3571 TCPSTAT_INC(tcps_rttupdated); 3572 tp->t_rttupdated++; 3573 #ifdef STATS 3574 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3575 imax(0, rtt * 1000 / hz)); 3576 #endif 3577 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3578 /* 3579 * srtt is stored as fixed point with 5 bits after the 3580 * binary point (i.e., scaled by 8). The following magic 3581 * is equivalent to the smoothing algorithm in rfc793 with 3582 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3583 * point). Adjust rtt to origin 0. 3584 */ 3585 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3586 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3587 3588 if ((tp->t_srtt += delta) <= 0) 3589 tp->t_srtt = 1; 3590 3591 /* 3592 * We accumulate a smoothed rtt variance (actually, a 3593 * smoothed mean difference), then set the retransmit 3594 * timer to smoothed rtt + 4 times the smoothed variance. 3595 * rttvar is stored as fixed point with 4 bits after the 3596 * binary point (scaled by 16). The following is 3597 * equivalent to rfc793 smoothing with an alpha of .75 3598 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3599 * rfc793's wired-in beta. 3600 */ 3601 if (delta < 0) 3602 delta = -delta; 3603 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3604 if ((tp->t_rttvar += delta) <= 0) 3605 tp->t_rttvar = 1; 3606 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3607 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3608 } else { 3609 /* 3610 * No rtt measurement yet - use the unsmoothed rtt. 3611 * Set the variance to half the rtt (so our first 3612 * retransmit happens at 3*rtt). 3613 */ 3614 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3615 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3616 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3617 } 3618 tp->t_rtttime = 0; 3619 tp->t_rxtshift = 0; 3620 3621 /* 3622 * the retransmit should happen at rtt + 4 * rttvar. 3623 * Because of the way we do the smoothing, srtt and rttvar 3624 * will each average +1/2 tick of bias. When we compute 3625 * the retransmit timer, we want 1/2 tick of rounding and 3626 * 1 extra tick because of +-1/2 tick uncertainty in the 3627 * firing of the timer. The bias will give us exactly the 3628 * 1.5 tick we need. But, because the bias is 3629 * statistical, we have to test that we don't drop below 3630 * the minimum feasible timer (which is 2 ticks). 3631 */ 3632 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3633 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3634 3635 /* 3636 * We received an ack for a packet that wasn't retransmitted; 3637 * it is probably safe to discard any error indications we've 3638 * received recently. This isn't quite right, but close enough 3639 * for now (a route might have failed after we sent a segment, 3640 * and the return path might not be symmetrical). 3641 */ 3642 tp->t_softerror = 0; 3643 } 3644 3645 /* 3646 * Determine a reasonable value for maxseg size. 3647 * If the route is known, check route for mtu. 3648 * If none, use an mss that can be handled on the outgoing interface 3649 * without forcing IP to fragment. If no route is found, route has no mtu, 3650 * or the destination isn't local, use a default, hopefully conservative 3651 * size (usually 512 or the default IP max size, but no more than the mtu 3652 * of the interface), as we can't discover anything about intervening 3653 * gateways or networks. We also initialize the congestion/slow start 3654 * window to be a single segment if the destination isn't local. 3655 * While looking at the routing entry, we also initialize other path-dependent 3656 * parameters from pre-set or cached values in the routing entry. 3657 * 3658 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3659 * IP options, e.g. IPSEC data, since length of this data may vary, and 3660 * thus it is calculated for every segment separately in tcp_output(). 3661 * 3662 * NOTE that this routine is only called when we process an incoming 3663 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3664 * settings are handled in tcp_mssopt(). 3665 */ 3666 void 3667 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3668 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3669 { 3670 int mss = 0; 3671 uint32_t maxmtu = 0; 3672 struct inpcb *inp = tp->t_inpcb; 3673 struct hc_metrics_lite metrics; 3674 #ifdef INET6 3675 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3676 size_t min_protoh = isipv6 ? 3677 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3678 sizeof (struct tcpiphdr); 3679 #else 3680 const size_t min_protoh = sizeof(struct tcpiphdr); 3681 #endif 3682 3683 INP_WLOCK_ASSERT(tp->t_inpcb); 3684 3685 if (mtuoffer != -1) { 3686 KASSERT(offer == -1, ("%s: conflict", __func__)); 3687 offer = mtuoffer - min_protoh; 3688 } 3689 3690 /* Initialize. */ 3691 #ifdef INET6 3692 if (isipv6) { 3693 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3694 tp->t_maxseg = V_tcp_v6mssdflt; 3695 } 3696 #endif 3697 #if defined(INET) && defined(INET6) 3698 else 3699 #endif 3700 #ifdef INET 3701 { 3702 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3703 tp->t_maxseg = V_tcp_mssdflt; 3704 } 3705 #endif 3706 3707 /* 3708 * No route to sender, stay with default mss and return. 3709 */ 3710 if (maxmtu == 0) { 3711 /* 3712 * In case we return early we need to initialize metrics 3713 * to a defined state as tcp_hc_get() would do for us 3714 * if there was no cache hit. 3715 */ 3716 if (metricptr != NULL) 3717 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3718 return; 3719 } 3720 3721 /* What have we got? */ 3722 switch (offer) { 3723 case 0: 3724 /* 3725 * Offer == 0 means that there was no MSS on the SYN 3726 * segment, in this case we use tcp_mssdflt as 3727 * already assigned to t_maxseg above. 3728 */ 3729 offer = tp->t_maxseg; 3730 break; 3731 3732 case -1: 3733 /* 3734 * Offer == -1 means that we didn't receive SYN yet. 3735 */ 3736 /* FALLTHROUGH */ 3737 3738 default: 3739 /* 3740 * Prevent DoS attack with too small MSS. Round up 3741 * to at least minmss. 3742 */ 3743 offer = max(offer, V_tcp_minmss); 3744 } 3745 3746 /* 3747 * rmx information is now retrieved from tcp_hostcache. 3748 */ 3749 tcp_hc_get(&inp->inp_inc, &metrics); 3750 if (metricptr != NULL) 3751 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3752 3753 /* 3754 * If there's a discovered mtu in tcp hostcache, use it. 3755 * Else, use the link mtu. 3756 */ 3757 if (metrics.rmx_mtu) 3758 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3759 else { 3760 #ifdef INET6 3761 if (isipv6) { 3762 mss = maxmtu - min_protoh; 3763 if (!V_path_mtu_discovery && 3764 !in6_localaddr(&inp->in6p_faddr)) 3765 mss = min(mss, V_tcp_v6mssdflt); 3766 } 3767 #endif 3768 #if defined(INET) && defined(INET6) 3769 else 3770 #endif 3771 #ifdef INET 3772 { 3773 mss = maxmtu - min_protoh; 3774 if (!V_path_mtu_discovery && 3775 !in_localaddr(inp->inp_faddr)) 3776 mss = min(mss, V_tcp_mssdflt); 3777 } 3778 #endif 3779 /* 3780 * XXX - The above conditional (mss = maxmtu - min_protoh) 3781 * probably violates the TCP spec. 3782 * The problem is that, since we don't know the 3783 * other end's MSS, we are supposed to use a conservative 3784 * default. But, if we do that, then MTU discovery will 3785 * never actually take place, because the conservative 3786 * default is much less than the MTUs typically seen 3787 * on the Internet today. For the moment, we'll sweep 3788 * this under the carpet. 3789 * 3790 * The conservative default might not actually be a problem 3791 * if the only case this occurs is when sending an initial 3792 * SYN with options and data to a host we've never talked 3793 * to before. Then, they will reply with an MSS value which 3794 * will get recorded and the new parameters should get 3795 * recomputed. For Further Study. 3796 */ 3797 } 3798 mss = min(mss, offer); 3799 3800 /* 3801 * Sanity check: make sure that maxseg will be large 3802 * enough to allow some data on segments even if the 3803 * all the option space is used (40bytes). Otherwise 3804 * funny things may happen in tcp_output. 3805 * 3806 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3807 */ 3808 mss = max(mss, 64); 3809 3810 tp->t_maxseg = mss; 3811 } 3812 3813 void 3814 tcp_mss(struct tcpcb *tp, int offer) 3815 { 3816 int mss; 3817 uint32_t bufsize; 3818 struct inpcb *inp; 3819 struct socket *so; 3820 struct hc_metrics_lite metrics; 3821 struct tcp_ifcap cap; 3822 3823 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3824 3825 bzero(&cap, sizeof(cap)); 3826 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3827 3828 mss = tp->t_maxseg; 3829 inp = tp->t_inpcb; 3830 3831 /* 3832 * If there's a pipesize, change the socket buffer to that size, 3833 * don't change if sb_hiwat is different than default (then it 3834 * has been changed on purpose with setsockopt). 3835 * Make the socket buffers an integral number of mss units; 3836 * if the mss is larger than the socket buffer, decrease the mss. 3837 */ 3838 so = inp->inp_socket; 3839 SOCKBUF_LOCK(&so->so_snd); 3840 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3841 bufsize = metrics.rmx_sendpipe; 3842 else 3843 bufsize = so->so_snd.sb_hiwat; 3844 if (bufsize < mss) 3845 mss = bufsize; 3846 else { 3847 bufsize = roundup(bufsize, mss); 3848 if (bufsize > sb_max) 3849 bufsize = sb_max; 3850 if (bufsize > so->so_snd.sb_hiwat) 3851 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3852 } 3853 SOCKBUF_UNLOCK(&so->so_snd); 3854 /* 3855 * Sanity check: make sure that maxseg will be large 3856 * enough to allow some data on segments even if the 3857 * all the option space is used (40bytes). Otherwise 3858 * funny things may happen in tcp_output. 3859 * 3860 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3861 */ 3862 tp->t_maxseg = max(mss, 64); 3863 3864 SOCKBUF_LOCK(&so->so_rcv); 3865 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3866 bufsize = metrics.rmx_recvpipe; 3867 else 3868 bufsize = so->so_rcv.sb_hiwat; 3869 if (bufsize > mss) { 3870 bufsize = roundup(bufsize, mss); 3871 if (bufsize > sb_max) 3872 bufsize = sb_max; 3873 if (bufsize > so->so_rcv.sb_hiwat) 3874 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3875 } 3876 SOCKBUF_UNLOCK(&so->so_rcv); 3877 3878 /* Check the interface for TSO capabilities. */ 3879 if (cap.ifcap & CSUM_TSO) { 3880 tp->t_flags |= TF_TSO; 3881 tp->t_tsomax = cap.tsomax; 3882 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3883 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3884 } 3885 } 3886 3887 /* 3888 * Determine the MSS option to send on an outgoing SYN. 3889 */ 3890 int 3891 tcp_mssopt(struct in_conninfo *inc) 3892 { 3893 int mss = 0; 3894 uint32_t thcmtu = 0; 3895 uint32_t maxmtu = 0; 3896 size_t min_protoh; 3897 3898 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3899 3900 #ifdef INET6 3901 if (inc->inc_flags & INC_ISIPV6) { 3902 mss = V_tcp_v6mssdflt; 3903 maxmtu = tcp_maxmtu6(inc, NULL); 3904 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3905 } 3906 #endif 3907 #if defined(INET) && defined(INET6) 3908 else 3909 #endif 3910 #ifdef INET 3911 { 3912 mss = V_tcp_mssdflt; 3913 maxmtu = tcp_maxmtu(inc, NULL); 3914 min_protoh = sizeof(struct tcpiphdr); 3915 } 3916 #endif 3917 #if defined(INET6) || defined(INET) 3918 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3919 #endif 3920 3921 if (maxmtu && thcmtu) 3922 mss = min(maxmtu, thcmtu) - min_protoh; 3923 else if (maxmtu || thcmtu) 3924 mss = max(maxmtu, thcmtu) - min_protoh; 3925 3926 return (mss); 3927 } 3928 3929 void 3930 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3931 { 3932 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3933 int maxseg = tcp_maxseg(tp); 3934 3935 INP_WLOCK_ASSERT(tp->t_inpcb); 3936 3937 /* 3938 * Compute the amount of data that this ACK is indicating 3939 * (del_data) and an estimate of how many bytes are in the 3940 * network. 3941 */ 3942 del_data = tp->sackhint.delivered_data; 3943 if (V_tcp_do_rfc6675_pipe) 3944 pipe = tcp_compute_pipe(tp); 3945 else 3946 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; 3947 tp->sackhint.prr_delivered += del_data; 3948 /* 3949 * Proportional Rate Reduction 3950 */ 3951 if (pipe >= tp->snd_ssthresh) { 3952 if (tp->sackhint.recover_fs == 0) 3953 tp->sackhint.recover_fs = 3954 imax(1, tp->snd_nxt - tp->snd_una); 3955 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3956 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3957 tp->sackhint.prr_out; 3958 } else { 3959 if (V_tcp_do_prr_conservative) 3960 limit = tp->sackhint.prr_delivered - 3961 tp->sackhint.prr_out; 3962 else 3963 limit = imax(tp->sackhint.prr_delivered - 3964 tp->sackhint.prr_out, del_data) + 3965 maxseg; 3966 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 3967 } 3968 snd_cnt = imax(snd_cnt, 0) / maxseg; 3969 /* 3970 * Send snd_cnt new data into the network in response to this ack. 3971 * If there is going to be a SACK retransmission, adjust snd_cwnd 3972 * accordingly. 3973 */ 3974 if (IN_FASTRECOVERY(tp->t_flags)) { 3975 tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover + 3976 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg)); 3977 } else if (IN_CONGRECOVERY(tp->t_flags)) 3978 tp->snd_cwnd = imax(maxseg, pipe - del_data + 3979 (snd_cnt * maxseg)); 3980 } 3981 3982 /* 3983 * On a partial ack arrives, force the retransmission of the 3984 * next unacknowledged segment. Do not clear tp->t_dupacks. 3985 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3986 * be started again. 3987 */ 3988 void 3989 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3990 { 3991 tcp_seq onxt = tp->snd_nxt; 3992 uint32_t ocwnd = tp->snd_cwnd; 3993 u_int maxseg = tcp_maxseg(tp); 3994 3995 INP_WLOCK_ASSERT(tp->t_inpcb); 3996 3997 tcp_timer_activate(tp, TT_REXMT, 0); 3998 tp->t_rtttime = 0; 3999 tp->snd_nxt = th->th_ack; 4000 /* 4001 * Set snd_cwnd to one segment beyond acknowledged offset. 4002 * (tp->snd_una has not yet been updated when this function is called.) 4003 */ 4004 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4005 tp->t_flags |= TF_ACKNOW; 4006 (void) tp->t_fb->tfb_tcp_output(tp); 4007 tp->snd_cwnd = ocwnd; 4008 if (SEQ_GT(onxt, tp->snd_nxt)) 4009 tp->snd_nxt = onxt; 4010 /* 4011 * Partial window deflation. Relies on fact that tp->snd_una 4012 * not updated yet. 4013 */ 4014 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4015 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4016 else 4017 tp->snd_cwnd = 0; 4018 tp->snd_cwnd += maxseg; 4019 } 4020 4021 int 4022 tcp_compute_pipe(struct tcpcb *tp) 4023 { 4024 return (tp->snd_max - tp->snd_una + 4025 tp->sackhint.sack_bytes_rexmit - 4026 tp->sackhint.sacked_bytes); 4027 } 4028 4029 uint32_t 4030 tcp_compute_initwnd(uint32_t maxseg) 4031 { 4032 /* 4033 * Calculate the Initial Window, also used as Restart Window 4034 * 4035 * RFC5681 Section 3.1 specifies the default conservative values. 4036 * RFC3390 specifies slightly more aggressive values. 4037 * RFC6928 increases it to ten segments. 4038 * Support for user specified value for initial flight size. 4039 */ 4040 if (V_tcp_initcwnd_segments) 4041 return min(V_tcp_initcwnd_segments * maxseg, 4042 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4043 else if (V_tcp_do_rfc3390) 4044 return min(4 * maxseg, max(2 * maxseg, 4380)); 4045 else { 4046 /* Per RFC5681 Section 3.1 */ 4047 if (maxseg > 2190) 4048 return (2 * maxseg); 4049 else if (maxseg > 1095) 4050 return (3 * maxseg); 4051 else 4052 return (4 * maxseg); 4053 } 4054 } 4055