1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 __FBSDID("$FreeBSD$"); 54 55 #include "opt_inet.h" 56 #include "opt_inet6.h" 57 #include "opt_ipsec.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/arb.h> 62 #include <sys/kernel.h> 63 #ifdef TCP_HHOOK 64 #include <sys/hhook.h> 65 #endif 66 #include <sys/malloc.h> 67 #include <sys/mbuf.h> 68 #include <sys/proc.h> /* for proc0 declaration */ 69 #include <sys/protosw.h> 70 #include <sys/qmath.h> 71 #include <sys/sdt.h> 72 #include <sys/signalvar.h> 73 #include <sys/socket.h> 74 #include <sys/socketvar.h> 75 #include <sys/sysctl.h> 76 #include <sys/syslog.h> 77 #include <sys/systm.h> 78 #include <sys/stats.h> 79 80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 81 82 #include <vm/uma.h> 83 84 #include <net/if.h> 85 #include <net/if_var.h> 86 #include <net/route.h> 87 #include <net/vnet.h> 88 89 #define TCPSTATES /* for logging */ 90 91 #include <netinet/in.h> 92 #include <netinet/in_kdtrace.h> 93 #include <netinet/in_pcb.h> 94 #include <netinet/in_systm.h> 95 #include <netinet/ip.h> 96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 98 #include <netinet/ip_var.h> 99 #include <netinet/ip_options.h> 100 #include <netinet/ip6.h> 101 #include <netinet/icmp6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_var.h> 104 #include <netinet6/ip6_var.h> 105 #include <netinet6/nd6.h> 106 #include <netinet/tcp.h> 107 #include <netinet/tcp_fsm.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_seq.h> 110 #include <netinet/tcp_timer.h> 111 #include <netinet/tcp_var.h> 112 #include <netinet6/tcp6_var.h> 113 #include <netinet/tcpip.h> 114 #include <netinet/cc/cc.h> 115 #include <netinet/tcp_fastopen.h> 116 #ifdef TCPPCAP 117 #include <netinet/tcp_pcap.h> 118 #endif 119 #include <netinet/tcp_syncache.h> 120 #ifdef TCPDEBUG 121 #include <netinet/tcp_debug.h> 122 #endif /* TCPDEBUG */ 123 #ifdef TCP_OFFLOAD 124 #include <netinet/tcp_offload.h> 125 #endif 126 #include <netinet/tcp_ecn.h> 127 #include <netinet/udp.h> 128 129 #include <netipsec/ipsec_support.h> 130 131 #include <machine/in_cksum.h> 132 133 #include <security/mac/mac_framework.h> 134 135 const int tcprexmtthresh = 3; 136 137 VNET_DEFINE(int, tcp_log_in_vain) = 0; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(tcp_log_in_vain), 0, 140 "Log all incoming TCP segments to closed ports"); 141 142 VNET_DEFINE(int, blackhole) = 0; 143 #define V_blackhole VNET(blackhole) 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 145 &VNET_NAME(blackhole), 0, 146 "Do not send RST on segments to closed ports"); 147 148 VNET_DEFINE(bool, blackhole_local) = false; 149 #define V_blackhole_local VNET(blackhole_local) 150 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 151 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 152 "Enforce net.inet.tcp.blackhole for locally originated packets"); 153 154 VNET_DEFINE(int, tcp_delack_enabled) = 1; 155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 156 &VNET_NAME(tcp_delack_enabled), 0, 157 "Delay ACK to try and piggyback it onto a data packet"); 158 159 VNET_DEFINE(int, drop_synfin) = 0; 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 161 &VNET_NAME(drop_synfin), 0, 162 "Drop TCP packets with SYN+FIN set"); 163 164 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 166 &VNET_NAME(tcp_do_prr_conservative), 0, 167 "Do conservative Proportional Rate Reduction"); 168 169 VNET_DEFINE(int, tcp_do_prr) = 1; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 171 &VNET_NAME(tcp_do_prr), 1, 172 "Enable Proportional Rate Reduction per RFC 6937"); 173 174 VNET_DEFINE(int, tcp_do_lrd) = 0; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW, 176 &VNET_NAME(tcp_do_lrd), 1, 177 "Perform Lost Retransmission Detection"); 178 179 VNET_DEFINE(int, tcp_do_newcwv) = 0; 180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_do_newcwv), 0, 182 "Enable New Congestion Window Validation per RFC7661"); 183 184 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 186 &VNET_NAME(tcp_do_rfc3042), 0, 187 "Enable RFC 3042 (Limited Transmit)"); 188 189 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 191 &VNET_NAME(tcp_do_rfc3390), 0, 192 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 193 194 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 195 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 196 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 197 "Slow-start flight size (initial congestion window) in number of segments"); 198 199 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 201 &VNET_NAME(tcp_do_rfc3465), 0, 202 "Enable RFC 3465 (Appropriate Byte Counting)"); 203 204 VNET_DEFINE(int, tcp_abc_l_var) = 2; 205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 206 &VNET_NAME(tcp_abc_l_var), 2, 207 "Cap the max cwnd increment during slow-start to this number of segments"); 208 209 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 210 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 211 "TCP ECN"); 212 213 VNET_DEFINE(int, tcp_do_ecn) = 2; 214 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 215 &VNET_NAME(tcp_do_ecn), 0, 216 "TCP ECN support"); 217 218 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 219 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 220 &VNET_NAME(tcp_ecn_maxretries), 0, 221 "Max retries before giving up on ECN"); 222 223 VNET_DEFINE(int, tcp_insecure_syn) = 0; 224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 225 &VNET_NAME(tcp_insecure_syn), 0, 226 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 227 228 VNET_DEFINE(int, tcp_insecure_rst) = 0; 229 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 230 &VNET_NAME(tcp_insecure_rst), 0, 231 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 232 233 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 234 #define V_tcp_recvspace VNET(tcp_recvspace) 235 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 236 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 237 238 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 239 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 240 &VNET_NAME(tcp_do_autorcvbuf), 0, 241 "Enable automatic receive buffer sizing"); 242 243 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 244 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 245 &VNET_NAME(tcp_autorcvbuf_max), 0, 246 "Max size of automatic receive buffer"); 247 248 VNET_DEFINE(struct inpcbinfo, tcbinfo); 249 250 /* 251 * TCP statistics are stored in an array of counter(9)s, which size matches 252 * size of struct tcpstat. TCP running connection count is a regular array. 253 */ 254 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 255 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 256 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 257 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 258 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 259 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 260 "TCP connection counts by TCP state"); 261 262 /* 263 * Kernel module interface for updating tcpstat. The first argument is an index 264 * into tcpstat treated as an array. 265 */ 266 void 267 kmod_tcpstat_add(int statnum, int val) 268 { 269 270 counter_u64_add(VNET(tcpstat)[statnum], val); 271 } 272 273 #ifdef TCP_HHOOK 274 /* 275 * Wrapper for the TCP established input helper hook. 276 */ 277 void 278 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 279 { 280 struct tcp_hhook_data hhook_data; 281 282 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 283 hhook_data.tp = tp; 284 hhook_data.th = th; 285 hhook_data.to = to; 286 287 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 288 tp->osd); 289 } 290 } 291 #endif 292 293 /* 294 * CC wrapper hook functions 295 */ 296 void 297 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 298 uint16_t type) 299 { 300 #ifdef STATS 301 int32_t gput; 302 #endif 303 304 INP_WLOCK_ASSERT(tp->t_inpcb); 305 306 tp->ccv->nsegs = nsegs; 307 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 308 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 309 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 310 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 311 tp->ccv->flags |= CCF_CWND_LIMITED; 312 else 313 tp->ccv->flags &= ~CCF_CWND_LIMITED; 314 315 if (type == CC_ACK) { 316 #ifdef STATS 317 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 318 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 319 if (!IN_RECOVERY(tp->t_flags)) 320 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 321 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 322 if ((tp->t_flags & TF_GPUTINPROG) && 323 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 324 /* 325 * Compute goodput in bits per millisecond. 326 */ 327 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 328 max(1, tcp_ts_getticks() - tp->gput_ts); 329 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 330 gput); 331 /* 332 * XXXLAS: This is a temporary hack, and should be 333 * chained off VOI_TCP_GPUT when stats(9) grows an API 334 * to deal with chained VOIs. 335 */ 336 if (tp->t_stats_gput_prev > 0) 337 stats_voi_update_abs_s32(tp->t_stats, 338 VOI_TCP_GPUT_ND, 339 ((gput - tp->t_stats_gput_prev) * 100) / 340 tp->t_stats_gput_prev); 341 tp->t_flags &= ~TF_GPUTINPROG; 342 tp->t_stats_gput_prev = gput; 343 } 344 #endif /* STATS */ 345 if (tp->snd_cwnd > tp->snd_ssthresh) { 346 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 347 if (tp->t_bytes_acked >= tp->snd_cwnd) { 348 tp->t_bytes_acked -= tp->snd_cwnd; 349 tp->ccv->flags |= CCF_ABC_SENTAWND; 350 } 351 } else { 352 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 353 tp->t_bytes_acked = 0; 354 } 355 } 356 357 if (CC_ALGO(tp)->ack_received != NULL) { 358 /* XXXLAS: Find a way to live without this */ 359 tp->ccv->curack = th->th_ack; 360 CC_ALGO(tp)->ack_received(tp->ccv, type); 361 } 362 #ifdef STATS 363 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 364 #endif 365 } 366 367 void 368 cc_conn_init(struct tcpcb *tp) 369 { 370 struct hc_metrics_lite metrics; 371 struct inpcb *inp = tp->t_inpcb; 372 u_int maxseg; 373 int rtt; 374 375 INP_WLOCK_ASSERT(tp->t_inpcb); 376 377 tcp_hc_get(&inp->inp_inc, &metrics); 378 maxseg = tcp_maxseg(tp); 379 380 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 381 tp->t_srtt = rtt; 382 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 383 TCPSTAT_INC(tcps_usedrtt); 384 if (metrics.rmx_rttvar) { 385 tp->t_rttvar = metrics.rmx_rttvar; 386 TCPSTAT_INC(tcps_usedrttvar); 387 } else { 388 /* default variation is +- 1 rtt */ 389 tp->t_rttvar = 390 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 391 } 392 TCPT_RANGESET(tp->t_rxtcur, 393 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 394 tp->t_rttmin, TCPTV_REXMTMAX); 395 } 396 if (metrics.rmx_ssthresh) { 397 /* 398 * There's some sort of gateway or interface 399 * buffer limit on the path. Use this to set 400 * the slow start threshold, but set the 401 * threshold to no less than 2*mss. 402 */ 403 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 404 TCPSTAT_INC(tcps_usedssthresh); 405 } 406 407 /* 408 * Set the initial slow-start flight size. 409 * 410 * If a SYN or SYN/ACK was lost and retransmitted, we have to 411 * reduce the initial CWND to one segment as congestion is likely 412 * requiring us to be cautious. 413 */ 414 if (tp->snd_cwnd == 1) 415 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 416 else 417 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 418 419 if (CC_ALGO(tp)->conn_init != NULL) 420 CC_ALGO(tp)->conn_init(tp->ccv); 421 } 422 423 void inline 424 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 425 { 426 INP_WLOCK_ASSERT(tp->t_inpcb); 427 428 #ifdef STATS 429 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 430 #endif 431 432 switch(type) { 433 case CC_NDUPACK: 434 if (!IN_FASTRECOVERY(tp->t_flags)) { 435 tp->snd_recover = tp->snd_max; 436 if (tp->t_flags2 & TF2_ECN_PERMIT) 437 tp->t_flags2 |= TF2_ECN_SND_CWR; 438 } 439 break; 440 case CC_ECN: 441 if (!IN_CONGRECOVERY(tp->t_flags) || 442 /* 443 * Allow ECN reaction on ACK to CWR, if 444 * that data segment was also CE marked. 445 */ 446 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 447 EXIT_CONGRECOVERY(tp->t_flags); 448 TCPSTAT_INC(tcps_ecn_rcwnd); 449 tp->snd_recover = tp->snd_max + 1; 450 if (tp->t_flags2 & TF2_ECN_PERMIT) 451 tp->t_flags2 |= TF2_ECN_SND_CWR; 452 } 453 break; 454 case CC_RTO: 455 tp->t_dupacks = 0; 456 tp->t_bytes_acked = 0; 457 EXIT_RECOVERY(tp->t_flags); 458 if (tp->t_flags2 & TF2_ECN_PERMIT) 459 tp->t_flags2 |= TF2_ECN_SND_CWR; 460 break; 461 case CC_RTO_ERR: 462 TCPSTAT_INC(tcps_sndrexmitbad); 463 /* RTO was unnecessary, so reset everything. */ 464 tp->snd_cwnd = tp->snd_cwnd_prev; 465 tp->snd_ssthresh = tp->snd_ssthresh_prev; 466 tp->snd_recover = tp->snd_recover_prev; 467 if (tp->t_flags & TF_WASFRECOVERY) 468 ENTER_FASTRECOVERY(tp->t_flags); 469 if (tp->t_flags & TF_WASCRECOVERY) 470 ENTER_CONGRECOVERY(tp->t_flags); 471 tp->snd_nxt = tp->snd_max; 472 tp->t_flags &= ~TF_PREVVALID; 473 tp->t_badrxtwin = 0; 474 break; 475 } 476 477 if (CC_ALGO(tp)->cong_signal != NULL) { 478 if (th != NULL) 479 tp->ccv->curack = th->th_ack; 480 CC_ALGO(tp)->cong_signal(tp->ccv, type); 481 } 482 } 483 484 void inline 485 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 486 { 487 INP_WLOCK_ASSERT(tp->t_inpcb); 488 489 /* XXXLAS: KASSERT that we're in recovery? */ 490 491 if (CC_ALGO(tp)->post_recovery != NULL) { 492 tp->ccv->curack = th->th_ack; 493 CC_ALGO(tp)->post_recovery(tp->ccv); 494 } 495 /* XXXLAS: EXIT_RECOVERY ? */ 496 tp->t_bytes_acked = 0; 497 tp->sackhint.delivered_data = 0; 498 tp->sackhint.prr_out = 0; 499 } 500 501 /* 502 * Indicate whether this ack should be delayed. We can delay the ack if 503 * following conditions are met: 504 * - There is no delayed ack timer in progress. 505 * - Our last ack wasn't a 0-sized window. We never want to delay 506 * the ack that opens up a 0-sized window. 507 * - LRO wasn't used for this segment. We make sure by checking that the 508 * segment size is not larger than the MSS. 509 */ 510 #define DELAY_ACK(tp, tlen) \ 511 ((!tcp_timer_active(tp, TT_DELACK) && \ 512 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 513 (tlen <= tp->t_maxseg) && \ 514 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 515 516 void inline 517 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 518 { 519 INP_WLOCK_ASSERT(tp->t_inpcb); 520 521 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 522 switch (iptos & IPTOS_ECN_MASK) { 523 case IPTOS_ECN_CE: 524 tp->ccv->flags |= CCF_IPHDR_CE; 525 break; 526 case IPTOS_ECN_ECT0: 527 /* FALLTHROUGH */ 528 case IPTOS_ECN_ECT1: 529 /* FALLTHROUGH */ 530 case IPTOS_ECN_NOTECT: 531 tp->ccv->flags &= ~CCF_IPHDR_CE; 532 break; 533 } 534 535 if (flags & TH_CWR) 536 tp->ccv->flags |= CCF_TCPHDR_CWR; 537 else 538 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 539 540 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 541 542 if (tp->ccv->flags & CCF_ACKNOW) { 543 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 544 tp->t_flags |= TF_ACKNOW; 545 } 546 } 547 } 548 549 void inline 550 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 551 { 552 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 553 } 554 555 /* 556 * TCP input handling is split into multiple parts: 557 * tcp6_input is a thin wrapper around tcp_input for the extended 558 * ip6_protox[] call format in ip6_input 559 * tcp_input handles primary segment validation, inpcb lookup and 560 * SYN processing on listen sockets 561 * tcp_do_segment processes the ACK and text of the segment for 562 * establishing, established and closing connections 563 */ 564 #ifdef INET6 565 int 566 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 567 { 568 struct mbuf *m; 569 struct in6_ifaddr *ia6; 570 struct ip6_hdr *ip6; 571 572 m = *mp; 573 if (m->m_len < *offp + sizeof(struct tcphdr)) { 574 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 575 if (m == NULL) { 576 *mp = m; 577 TCPSTAT_INC(tcps_rcvshort); 578 return (IPPROTO_DONE); 579 } 580 } 581 582 /* 583 * draft-itojun-ipv6-tcp-to-anycast 584 * better place to put this in? 585 */ 586 ip6 = mtod(m, struct ip6_hdr *); 587 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 588 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 589 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 590 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 591 *mp = NULL; 592 return (IPPROTO_DONE); 593 } 594 595 *mp = m; 596 return (tcp_input_with_port(mp, offp, proto, port)); 597 } 598 599 int 600 tcp6_input(struct mbuf **mp, int *offp, int proto) 601 { 602 603 return(tcp6_input_with_port(mp, offp, proto, 0)); 604 } 605 #endif /* INET6 */ 606 607 int 608 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 609 { 610 struct mbuf *m = *mp; 611 struct tcphdr *th = NULL; 612 struct ip *ip = NULL; 613 struct inpcb *inp = NULL; 614 struct tcpcb *tp = NULL; 615 struct socket *so = NULL; 616 u_char *optp = NULL; 617 int off0; 618 int optlen = 0; 619 #ifdef INET 620 int len; 621 uint8_t ipttl; 622 #endif 623 int tlen = 0, off; 624 int drop_hdrlen; 625 int thflags; 626 int rstreason = 0; /* For badport_bandlim accounting purposes */ 627 int lookupflag; 628 uint8_t iptos; 629 struct m_tag *fwd_tag = NULL; 630 #ifdef INET6 631 struct ip6_hdr *ip6 = NULL; 632 int isipv6; 633 #else 634 const void *ip6 = NULL; 635 #endif /* INET6 */ 636 struct tcpopt to; /* options in this segment */ 637 char *s = NULL; /* address and port logging */ 638 #ifdef TCPDEBUG 639 /* 640 * The size of tcp_saveipgen must be the size of the max ip header, 641 * now IPv6. 642 */ 643 u_char tcp_saveipgen[IP6_HDR_LEN]; 644 struct tcphdr tcp_savetcp; 645 short ostate = 0; 646 #endif 647 648 NET_EPOCH_ASSERT(); 649 650 #ifdef INET6 651 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 652 #endif 653 654 off0 = *offp; 655 m = *mp; 656 *mp = NULL; 657 to.to_flags = 0; 658 TCPSTAT_INC(tcps_rcvtotal); 659 660 #ifdef INET6 661 if (isipv6) { 662 ip6 = mtod(m, struct ip6_hdr *); 663 th = (struct tcphdr *)((caddr_t)ip6 + off0); 664 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 665 if (port) 666 goto skip6_csum; 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 skip6_csum: 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 (port) 722 goto skip_csum; 723 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 724 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 725 th->th_sum = m->m_pkthdr.csum_data; 726 else 727 th->th_sum = in_pseudo(ip->ip_src.s_addr, 728 ip->ip_dst.s_addr, 729 htonl(m->m_pkthdr.csum_data + tlen + 730 IPPROTO_TCP)); 731 th->th_sum ^= 0xffff; 732 } else { 733 struct ipovly *ipov = (struct ipovly *)ip; 734 735 /* 736 * Checksum extended TCP header and data. 737 */ 738 len = off0 + tlen; 739 ipttl = ip->ip_ttl; 740 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 741 ipov->ih_len = htons(tlen); 742 th->th_sum = in_cksum(m, len); 743 /* Reset length for SDT probes. */ 744 ip->ip_len = htons(len); 745 /* Reset TOS bits */ 746 ip->ip_tos = iptos; 747 /* Re-initialization for later version check */ 748 ip->ip_ttl = ipttl; 749 ip->ip_v = IPVERSION; 750 ip->ip_hl = off0 >> 2; 751 } 752 skip_csum: 753 if (th->th_sum && (port == 0)) { 754 TCPSTAT_INC(tcps_rcvbadsum); 755 goto drop; 756 } 757 } 758 #endif /* INET */ 759 760 /* 761 * Check that TCP offset makes sense, 762 * pull out TCP options and adjust length. XXX 763 */ 764 off = th->th_off << 2; 765 if (off < sizeof (struct tcphdr) || off > tlen) { 766 TCPSTAT_INC(tcps_rcvbadoff); 767 goto drop; 768 } 769 tlen -= off; /* tlen is used instead of ti->ti_len */ 770 if (off > sizeof (struct tcphdr)) { 771 #ifdef INET6 772 if (isipv6) { 773 if (m->m_len < off0 + off) { 774 m = m_pullup(m, off0 + off); 775 if (m == NULL) { 776 TCPSTAT_INC(tcps_rcvshort); 777 return (IPPROTO_DONE); 778 } 779 } 780 ip6 = mtod(m, struct ip6_hdr *); 781 th = (struct tcphdr *)((caddr_t)ip6 + off0); 782 } 783 #endif 784 #if defined(INET) && defined(INET6) 785 else 786 #endif 787 #ifdef INET 788 { 789 if (m->m_len < sizeof(struct ip) + off) { 790 if ((m = m_pullup(m, sizeof (struct ip) + off)) 791 == NULL) { 792 TCPSTAT_INC(tcps_rcvshort); 793 return (IPPROTO_DONE); 794 } 795 ip = mtod(m, struct ip *); 796 th = (struct tcphdr *)((caddr_t)ip + off0); 797 } 798 } 799 #endif 800 optlen = off - sizeof (struct tcphdr); 801 optp = (u_char *)(th + 1); 802 } 803 thflags = tcp_get_flags(th); 804 805 /* 806 * Convert TCP protocol specific fields to host format. 807 */ 808 tcp_fields_to_host(th); 809 810 /* 811 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 812 */ 813 drop_hdrlen = off0 + off; 814 815 /* 816 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 817 */ 818 if ( 819 #ifdef INET6 820 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 821 #ifdef INET 822 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 823 #endif 824 #endif 825 #if defined(INET) && !defined(INET6) 826 (m->m_flags & M_IP_NEXTHOP) 827 #endif 828 ) 829 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 830 831 /* 832 * For initial SYN packets we don't need write lock on matching 833 * PCB, be it a listening one or a synchronized one. The packet 834 * shall not modify its state. 835 */ 836 lookupflag = (thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 837 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB; 838 findpcb: 839 #ifdef INET6 840 if (isipv6 && fwd_tag != NULL) { 841 struct sockaddr_in6 *next_hop6; 842 843 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 844 /* 845 * Transparently forwarded. Pretend to be the destination. 846 * Already got one like this? 847 */ 848 inp = in6_pcblookup_mbuf(&V_tcbinfo, 849 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 850 lookupflag, m->m_pkthdr.rcvif, m); 851 if (!inp) { 852 /* 853 * It's new. Try to find the ambushing socket. 854 * Because we've rewritten the destination address, 855 * any hardware-generated hash is ignored. 856 */ 857 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 858 th->th_sport, &next_hop6->sin6_addr, 859 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 860 th->th_dport, INPLOOKUP_WILDCARD | lookupflag, 861 m->m_pkthdr.rcvif); 862 } 863 } else if (isipv6) { 864 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 865 th->th_sport, &ip6->ip6_dst, th->th_dport, 866 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m); 867 } 868 #endif /* INET6 */ 869 #if defined(INET6) && defined(INET) 870 else 871 #endif 872 #ifdef INET 873 if (fwd_tag != NULL) { 874 struct sockaddr_in *next_hop; 875 876 next_hop = (struct sockaddr_in *)(fwd_tag+1); 877 /* 878 * Transparently forwarded. Pretend to be the destination. 879 * already got one like this? 880 */ 881 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 882 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); 883 if (!inp) { 884 /* 885 * It's new. Try to find the ambushing socket. 886 * Because we've rewritten the destination address, 887 * any hardware-generated hash is ignored. 888 */ 889 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 890 th->th_sport, next_hop->sin_addr, 891 next_hop->sin_port ? ntohs(next_hop->sin_port) : 892 th->th_dport, INPLOOKUP_WILDCARD | lookupflag, 893 m->m_pkthdr.rcvif); 894 } 895 } else 896 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 897 th->th_sport, ip->ip_dst, th->th_dport, 898 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m); 899 #endif /* INET */ 900 901 /* 902 * If the INPCB does not exist then all data in the incoming 903 * segment is discarded and an appropriate RST is sent back. 904 * XXX MRT Send RST using which routing table? 905 */ 906 if (inp == NULL) { 907 /* 908 * Log communication attempts to ports that are not 909 * in use. 910 */ 911 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 912 V_tcp_log_in_vain == 2) { 913 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 914 log(LOG_INFO, "%s; %s: Connection attempt " 915 "to closed port\n", s, __func__); 916 } 917 /* 918 * When blackholing do not respond with a RST but 919 * completely ignore the segment and drop it. 920 */ 921 if (((V_blackhole == 1 && (thflags & TH_SYN)) || 922 V_blackhole == 2) && (V_blackhole_local || ( 923 #ifdef INET6 924 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 925 #endif 926 #ifdef INET 927 !in_localip(ip->ip_src) 928 #else 929 true 930 #endif 931 ))) 932 goto dropunlock; 933 934 rstreason = BANDLIM_RST_CLOSEDPORT; 935 goto dropwithreset; 936 } 937 INP_LOCK_ASSERT(inp); 938 /* 939 * While waiting for inp lock during the lookup, another thread 940 * can have dropped the inpcb, in which case we need to loop back 941 * and try to find a new inpcb to deliver to. 942 */ 943 if (inp->inp_flags & INP_DROPPED) { 944 INP_UNLOCK(inp); 945 inp = NULL; 946 goto findpcb; 947 } 948 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 949 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 950 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) { 951 inp->inp_flowid = m->m_pkthdr.flowid; 952 inp->inp_flowtype = M_HASHTYPE_GET(m); 953 } 954 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 955 #ifdef INET6 956 if (isipv6 && IPSEC_ENABLED(ipv6) && 957 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 958 goto dropunlock; 959 } 960 #ifdef INET 961 else 962 #endif 963 #endif /* INET6 */ 964 #ifdef INET 965 if (IPSEC_ENABLED(ipv4) && 966 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 967 goto dropunlock; 968 } 969 #endif /* INET */ 970 #endif /* IPSEC */ 971 972 /* 973 * Check the minimum TTL for socket. 974 */ 975 if (inp->inp_ip_minttl != 0) { 976 #ifdef INET6 977 if (isipv6) { 978 if (inp->inp_ip_minttl > ip6->ip6_hlim) 979 goto dropunlock; 980 } else 981 #endif 982 if (inp->inp_ip_minttl > ip->ip_ttl) 983 goto dropunlock; 984 } 985 986 /* 987 * A previous connection in TIMEWAIT state is supposed to catch stray 988 * or duplicate segments arriving late. If this segment was a 989 * legitimate new connection attempt, the old INPCB gets removed and 990 * we can try again to find a listening socket. 991 */ 992 if (inp->inp_flags & INP_TIMEWAIT) { 993 tcp_dooptions(&to, optp, optlen, 994 (thflags & TH_SYN) ? TO_SYN : 0); 995 /* 996 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 997 */ 998 if (tcp_twcheck(inp, &to, th, m, tlen)) 999 goto findpcb; 1000 return (IPPROTO_DONE); 1001 } 1002 /* 1003 * The TCPCB may no longer exist if the connection is winding 1004 * down or it is in the CLOSED state. Either way we drop the 1005 * segment and send an appropriate response. 1006 */ 1007 tp = intotcpcb(inp); 1008 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1009 rstreason = BANDLIM_RST_CLOSEDPORT; 1010 goto dropwithreset; 1011 } 1012 1013 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1014 rstreason = BANDLIM_RST_CLOSEDPORT; 1015 goto dropwithreset; 1016 } 1017 1018 #ifdef TCP_OFFLOAD 1019 if (tp->t_flags & TF_TOE) { 1020 tcp_offload_input(tp, m); 1021 m = NULL; /* consumed by the TOE driver */ 1022 goto dropunlock; 1023 } 1024 #endif 1025 1026 #ifdef MAC 1027 if (mac_inpcb_check_deliver(inp, m)) 1028 goto dropunlock; 1029 #endif 1030 so = inp->inp_socket; 1031 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1032 #ifdef TCPDEBUG 1033 if (so->so_options & SO_DEBUG) { 1034 ostate = tp->t_state; 1035 #ifdef INET6 1036 if (isipv6) { 1037 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1038 } else 1039 #endif 1040 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1041 tcp_savetcp = *th; 1042 } 1043 #endif /* TCPDEBUG */ 1044 /* 1045 * When the socket is accepting connections (the INPCB is in LISTEN 1046 * state) we look into the SYN cache if this is a new connection 1047 * attempt or the completion of a previous one. 1048 */ 1049 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1050 ("%s: so accepting but tp %p not listening", __func__, tp)); 1051 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1052 struct in_conninfo inc; 1053 1054 bzero(&inc, sizeof(inc)); 1055 #ifdef INET6 1056 if (isipv6) { 1057 inc.inc_flags |= INC_ISIPV6; 1058 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1059 inc.inc_flags |= INC_IPV6MINMTU; 1060 inc.inc6_faddr = ip6->ip6_src; 1061 inc.inc6_laddr = ip6->ip6_dst; 1062 } else 1063 #endif 1064 { 1065 inc.inc_faddr = ip->ip_src; 1066 inc.inc_laddr = ip->ip_dst; 1067 } 1068 inc.inc_fport = th->th_sport; 1069 inc.inc_lport = th->th_dport; 1070 inc.inc_fibnum = so->so_fibnum; 1071 1072 /* 1073 * Check for an existing connection attempt in syncache if 1074 * the flag is only ACK. A successful lookup creates a new 1075 * socket appended to the listen queue in SYN_RECEIVED state. 1076 */ 1077 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1078 /* 1079 * Parse the TCP options here because 1080 * syncookies need access to the reflected 1081 * timestamp. 1082 */ 1083 tcp_dooptions(&to, optp, optlen, 0); 1084 /* 1085 * NB: syncache_expand() doesn't unlock 1086 * inp and tcpinfo locks. 1087 */ 1088 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1089 if (rstreason < 0) { 1090 /* 1091 * A failing TCP MD5 signature comparison 1092 * must result in the segment being dropped 1093 * and must not produce any response back 1094 * to the sender. 1095 */ 1096 goto dropunlock; 1097 } else if (rstreason == 0) { 1098 /* 1099 * No syncache entry or ACK was not 1100 * for our SYN/ACK. Send a RST. 1101 * NB: syncache did its own logging 1102 * of the failure cause. 1103 */ 1104 rstreason = BANDLIM_RST_OPENPORT; 1105 goto dropwithreset; 1106 } 1107 tfo_socket_result: 1108 if (so == NULL) { 1109 /* 1110 * We completed the 3-way handshake 1111 * but could not allocate a socket 1112 * either due to memory shortage, 1113 * listen queue length limits or 1114 * global socket limits. Send RST 1115 * or wait and have the remote end 1116 * retransmit the ACK for another 1117 * try. 1118 */ 1119 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1120 log(LOG_DEBUG, "%s; %s: Listen socket: " 1121 "Socket allocation failed due to " 1122 "limits or memory shortage, %s\n", 1123 s, __func__, 1124 V_tcp_sc_rst_sock_fail ? 1125 "sending RST" : "try again"); 1126 if (V_tcp_sc_rst_sock_fail) { 1127 rstreason = BANDLIM_UNLIMITED; 1128 goto dropwithreset; 1129 } else 1130 goto dropunlock; 1131 } 1132 /* 1133 * Socket is created in state SYN_RECEIVED. 1134 * Unlock the listen socket, lock the newly 1135 * created socket and update the tp variable. 1136 * If we came here via jump to tfo_socket_result, 1137 * then listening socket is read-locked. 1138 */ 1139 INP_UNLOCK(inp); /* listen socket */ 1140 inp = sotoinpcb(so); 1141 /* 1142 * New connection inpcb is already locked by 1143 * syncache_expand(). 1144 */ 1145 INP_WLOCK_ASSERT(inp); 1146 tp = intotcpcb(inp); 1147 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1148 ("%s: ", __func__)); 1149 /* 1150 * Process the segment and the data it 1151 * contains. tcp_do_segment() consumes 1152 * the mbuf chain and unlocks the inpcb. 1153 */ 1154 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1155 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1156 iptos); 1157 return (IPPROTO_DONE); 1158 } 1159 /* 1160 * Segment flag validation for new connection attempts: 1161 * 1162 * Our (SYN|ACK) response was rejected. 1163 * Check with syncache and remove entry to prevent 1164 * retransmits. 1165 * 1166 * NB: syncache_chkrst does its own logging of failure 1167 * causes. 1168 */ 1169 if (thflags & TH_RST) { 1170 syncache_chkrst(&inc, th, m, port); 1171 goto dropunlock; 1172 } 1173 /* 1174 * We can't do anything without SYN. 1175 */ 1176 if ((thflags & TH_SYN) == 0) { 1177 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1178 log(LOG_DEBUG, "%s; %s: Listen socket: " 1179 "SYN is missing, segment ignored\n", 1180 s, __func__); 1181 TCPSTAT_INC(tcps_badsyn); 1182 goto dropunlock; 1183 } 1184 /* 1185 * (SYN|ACK) is bogus on a listen socket. 1186 */ 1187 if (thflags & TH_ACK) { 1188 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1189 log(LOG_DEBUG, "%s; %s: Listen socket: " 1190 "SYN|ACK invalid, segment rejected\n", 1191 s, __func__); 1192 syncache_badack(&inc, port); /* XXX: Not needed! */ 1193 TCPSTAT_INC(tcps_badsyn); 1194 rstreason = BANDLIM_RST_OPENPORT; 1195 goto dropwithreset; 1196 } 1197 /* 1198 * If the drop_synfin option is enabled, drop all 1199 * segments with both the SYN and FIN bits set. 1200 * This prevents e.g. nmap from identifying the 1201 * TCP/IP stack. 1202 * XXX: Poor reasoning. nmap has other methods 1203 * and is constantly refining its stack detection 1204 * strategies. 1205 * XXX: This is a violation of the TCP specification 1206 * and was used by RFC1644. 1207 */ 1208 if ((thflags & TH_FIN) && V_drop_synfin) { 1209 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1210 log(LOG_DEBUG, "%s; %s: Listen socket: " 1211 "SYN|FIN segment ignored (based on " 1212 "sysctl setting)\n", s, __func__); 1213 TCPSTAT_INC(tcps_badsyn); 1214 goto dropunlock; 1215 } 1216 /* 1217 * Segment's flags are (SYN) or (SYN|FIN). 1218 * 1219 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1220 * as they do not affect the state of the TCP FSM. 1221 * The data pointed to by TH_URG and th_urp is ignored. 1222 */ 1223 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1224 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1225 KASSERT(thflags & (TH_SYN), 1226 ("%s: Listen socket: TH_SYN not set", __func__)); 1227 INP_RLOCK_ASSERT(inp); 1228 #ifdef INET6 1229 /* 1230 * If deprecated address is forbidden, 1231 * we do not accept SYN to deprecated interface 1232 * address to prevent any new inbound connection from 1233 * getting established. 1234 * When we do not accept SYN, we send a TCP RST, 1235 * with deprecated source address (instead of dropping 1236 * it). We compromise it as it is much better for peer 1237 * to send a RST, and RST will be the final packet 1238 * for the exchange. 1239 * 1240 * If we do not forbid deprecated addresses, we accept 1241 * the SYN packet. RFC2462 does not suggest dropping 1242 * SYN in this case. 1243 * If we decipher RFC2462 5.5.4, it says like this: 1244 * 1. use of deprecated addr with existing 1245 * communication is okay - "SHOULD continue to be 1246 * used" 1247 * 2. use of it with new communication: 1248 * (2a) "SHOULD NOT be used if alternate address 1249 * with sufficient scope is available" 1250 * (2b) nothing mentioned otherwise. 1251 * Here we fall into (2b) case as we have no choice in 1252 * our source address selection - we must obey the peer. 1253 * 1254 * The wording in RFC2462 is confusing, and there are 1255 * multiple description text for deprecated address 1256 * handling - worse, they are not exactly the same. 1257 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1258 */ 1259 if (isipv6 && !V_ip6_use_deprecated) { 1260 struct in6_ifaddr *ia6; 1261 1262 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1263 if (ia6 != NULL && 1264 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1265 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1266 log(LOG_DEBUG, "%s; %s: Listen socket: " 1267 "Connection attempt to deprecated " 1268 "IPv6 address rejected\n", 1269 s, __func__); 1270 rstreason = BANDLIM_RST_OPENPORT; 1271 goto dropwithreset; 1272 } 1273 } 1274 #endif /* INET6 */ 1275 /* 1276 * Basic sanity checks on incoming SYN requests: 1277 * Don't respond if the destination is a link layer 1278 * broadcast according to RFC1122 4.2.3.10, p. 104. 1279 * If it is from this socket it must be forged. 1280 * Don't respond if the source or destination is a 1281 * global or subnet broad- or multicast address. 1282 * Note that it is quite possible to receive unicast 1283 * link-layer packets with a broadcast IP address. Use 1284 * in_broadcast() to find them. 1285 */ 1286 if (m->m_flags & (M_BCAST|M_MCAST)) { 1287 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1288 log(LOG_DEBUG, "%s; %s: Listen socket: " 1289 "Connection attempt from broad- or multicast " 1290 "link layer address ignored\n", s, __func__); 1291 goto dropunlock; 1292 } 1293 #ifdef INET6 1294 if (isipv6) { 1295 if (th->th_dport == th->th_sport && 1296 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1297 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1298 log(LOG_DEBUG, "%s; %s: Listen socket: " 1299 "Connection attempt to/from self " 1300 "ignored\n", s, __func__); 1301 goto dropunlock; 1302 } 1303 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1304 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1305 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1306 log(LOG_DEBUG, "%s; %s: Listen socket: " 1307 "Connection attempt from/to multicast " 1308 "address ignored\n", s, __func__); 1309 goto dropunlock; 1310 } 1311 } 1312 #endif 1313 #if defined(INET) && defined(INET6) 1314 else 1315 #endif 1316 #ifdef INET 1317 { 1318 if (th->th_dport == th->th_sport && 1319 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1320 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1321 log(LOG_DEBUG, "%s; %s: Listen socket: " 1322 "Connection attempt from/to self " 1323 "ignored\n", s, __func__); 1324 goto dropunlock; 1325 } 1326 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1327 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1328 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1329 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1330 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1331 log(LOG_DEBUG, "%s; %s: Listen socket: " 1332 "Connection attempt from/to broad- " 1333 "or multicast address ignored\n", 1334 s, __func__); 1335 goto dropunlock; 1336 } 1337 } 1338 #endif 1339 /* 1340 * SYN appears to be valid. Create compressed TCP state 1341 * for syncache. 1342 */ 1343 #ifdef TCPDEBUG 1344 if (so->so_options & SO_DEBUG) 1345 tcp_trace(TA_INPUT, ostate, tp, 1346 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1347 #endif 1348 TCP_PROBE3(debug__input, tp, th, m); 1349 tcp_dooptions(&to, optp, optlen, TO_SYN); 1350 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1351 iptos, port)) != NULL) 1352 goto tfo_socket_result; 1353 1354 /* 1355 * Entry added to syncache and mbuf consumed. 1356 * Only the listen socket is unlocked by syncache_add(). 1357 */ 1358 return (IPPROTO_DONE); 1359 } else if (tp->t_state == TCPS_LISTEN) { 1360 /* 1361 * When a listen socket is torn down the SO_ACCEPTCONN 1362 * flag is removed first while connections are drained 1363 * from the accept queue in a unlock/lock cycle of the 1364 * ACCEPT_LOCK, opening a race condition allowing a SYN 1365 * attempt go through unhandled. 1366 */ 1367 goto dropunlock; 1368 } 1369 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1370 if (tp->t_flags & TF_SIGNATURE) { 1371 tcp_dooptions(&to, optp, optlen, thflags); 1372 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1373 TCPSTAT_INC(tcps_sig_err_nosigopt); 1374 goto dropunlock; 1375 } 1376 if (!TCPMD5_ENABLED() || 1377 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1378 goto dropunlock; 1379 } 1380 #endif 1381 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1382 1383 /* 1384 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1385 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1386 * the inpcb, and unlocks pcbinfo. 1387 * 1388 * XXXGL: in case of a pure SYN arriving on existing connection 1389 * TCP stacks won't need to modify the PCB, they would either drop 1390 * the segment silently, or send a challenge ACK. However, we try 1391 * to upgrade the lock, because calling convention for stacks is 1392 * write-lock on PCB. If upgrade fails, drop the SYN. 1393 */ 1394 if (lookupflag == INPLOOKUP_RLOCKPCB && INP_TRY_UPGRADE(inp) == 0) 1395 goto dropunlock; 1396 1397 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1398 return (IPPROTO_DONE); 1399 1400 dropwithreset: 1401 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1402 1403 if (inp != NULL) { 1404 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1405 INP_UNLOCK(inp); 1406 } else 1407 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1408 m = NULL; /* mbuf chain got consumed. */ 1409 goto drop; 1410 1411 dropunlock: 1412 if (m != NULL) 1413 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1414 1415 if (inp != NULL) 1416 INP_UNLOCK(inp); 1417 1418 drop: 1419 if (s != NULL) 1420 free(s, M_TCPLOG); 1421 if (m != NULL) 1422 m_freem(m); 1423 return (IPPROTO_DONE); 1424 } 1425 1426 /* 1427 * Automatic sizing of receive socket buffer. Often the send 1428 * buffer size is not optimally adjusted to the actual network 1429 * conditions at hand (delay bandwidth product). Setting the 1430 * buffer size too small limits throughput on links with high 1431 * bandwidth and high delay (eg. trans-continental/oceanic links). 1432 * 1433 * On the receive side the socket buffer memory is only rarely 1434 * used to any significant extent. This allows us to be much 1435 * more aggressive in scaling the receive socket buffer. For 1436 * the case that the buffer space is actually used to a large 1437 * extent and we run out of kernel memory we can simply drop 1438 * the new segments; TCP on the sender will just retransmit it 1439 * later. Setting the buffer size too big may only consume too 1440 * much kernel memory if the application doesn't read() from 1441 * the socket or packet loss or reordering makes use of the 1442 * reassembly queue. 1443 * 1444 * The criteria to step up the receive buffer one notch are: 1445 * 1. Application has not set receive buffer size with 1446 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1447 * 2. the number of bytes received during 1/2 of an sRTT 1448 * is at least 3/8 of the current socket buffer size. 1449 * 3. receive buffer size has not hit maximal automatic size; 1450 * 1451 * If all of the criteria are met we increaset the socket buffer 1452 * by a 1/2 (bounded by the max). This allows us to keep ahead 1453 * of slow-start but also makes it so our peer never gets limited 1454 * by our rwnd which we then open up causing a burst. 1455 * 1456 * This algorithm does two steps per RTT at most and only if 1457 * we receive a bulk stream w/o packet losses or reorderings. 1458 * Shrinking the buffer during idle times is not necessary as 1459 * it doesn't consume any memory when idle. 1460 * 1461 * TODO: Only step up if the application is actually serving 1462 * the buffer to better manage the socket buffer resources. 1463 */ 1464 int 1465 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1466 struct tcpcb *tp, int tlen) 1467 { 1468 int newsize = 0; 1469 1470 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1471 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1472 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1473 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1474 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1475 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1476 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1477 } 1478 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1479 1480 /* Start over with next RTT. */ 1481 tp->rfbuf_ts = 0; 1482 tp->rfbuf_cnt = 0; 1483 } else { 1484 tp->rfbuf_cnt += tlen; /* add up */ 1485 } 1486 return (newsize); 1487 } 1488 1489 int 1490 tcp_input(struct mbuf **mp, int *offp, int proto) 1491 { 1492 return(tcp_input_with_port(mp, offp, proto, 0)); 1493 } 1494 1495 void 1496 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1497 { 1498 /* 1499 * Since tp might be gone if the session entered 1500 * the TIME_WAIT state before coming here, we need 1501 * to check if the socket is still connected. 1502 */ 1503 if (tp == NULL) { 1504 return; 1505 } 1506 if (so == NULL) { 1507 return; 1508 } 1509 INP_LOCK_ASSERT(tp->t_inpcb); 1510 if (tp->t_flags & TF_WAKESOR) { 1511 tp->t_flags &= ~TF_WAKESOR; 1512 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1513 sorwakeup_locked(so); 1514 } 1515 } 1516 1517 void 1518 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1519 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1520 { 1521 uint16_t thflags; 1522 int acked, ourfinisacked, needoutput = 0, sack_changed; 1523 int rstreason, todrop, win, incforsyn = 0; 1524 uint32_t tiwin; 1525 uint16_t nsegs; 1526 char *s; 1527 struct in_conninfo *inc; 1528 struct mbuf *mfree; 1529 struct tcpopt to; 1530 int tfo_syn; 1531 u_int maxseg; 1532 1533 #ifdef TCPDEBUG 1534 /* 1535 * The size of tcp_saveipgen must be the size of the max ip header, 1536 * now IPv6. 1537 */ 1538 u_char tcp_saveipgen[IP6_HDR_LEN]; 1539 struct tcphdr tcp_savetcp; 1540 short ostate = 0; 1541 #endif 1542 thflags = tcp_get_flags(th); 1543 inc = &tp->t_inpcb->inp_inc; 1544 tp->sackhint.last_sack_ack = 0; 1545 sack_changed = 0; 1546 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1547 1548 NET_EPOCH_ASSERT(); 1549 INP_WLOCK_ASSERT(tp->t_inpcb); 1550 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1551 __func__)); 1552 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1553 __func__)); 1554 1555 #ifdef TCPPCAP 1556 /* Save segment, if requested. */ 1557 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1558 #endif 1559 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1560 tlen, NULL, true); 1561 1562 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1563 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1564 log(LOG_DEBUG, "%s; %s: " 1565 "SYN|FIN segment ignored (based on " 1566 "sysctl setting)\n", s, __func__); 1567 free(s, M_TCPLOG); 1568 } 1569 goto drop; 1570 } 1571 1572 /* 1573 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1574 * check SEQ.ACK first. 1575 */ 1576 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1577 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1578 rstreason = BANDLIM_UNLIMITED; 1579 goto dropwithreset; 1580 } 1581 1582 /* 1583 * Segment received on connection. 1584 * Reset idle time and keep-alive timer. 1585 * XXX: This should be done after segment 1586 * validation to ignore broken/spoofed segs. 1587 */ 1588 tp->t_rcvtime = ticks; 1589 1590 /* 1591 * Scale up the window into a 32-bit value. 1592 * For the SYN_SENT state the scale is zero. 1593 */ 1594 tiwin = th->th_win << tp->snd_scale; 1595 #ifdef STATS 1596 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1597 #endif 1598 1599 /* 1600 * TCP ECN processing. 1601 */ 1602 if (tcp_ecn_input_segment(tp, thflags, iptos)) 1603 cc_cong_signal(tp, th, CC_ECN); 1604 1605 /* 1606 * Parse options on any incoming segment. 1607 */ 1608 tcp_dooptions(&to, (u_char *)(th + 1), 1609 (th->th_off << 2) - sizeof(struct tcphdr), 1610 (thflags & TH_SYN) ? TO_SYN : 0); 1611 1612 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1613 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1614 (to.to_flags & TOF_SIGNATURE) == 0) { 1615 TCPSTAT_INC(tcps_sig_err_sigopt); 1616 /* XXX: should drop? */ 1617 } 1618 #endif 1619 /* 1620 * If echoed timestamp is later than the current time, 1621 * fall back to non RFC1323 RTT calculation. Normalize 1622 * timestamp if syncookies were used when this connection 1623 * was established. 1624 */ 1625 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1626 to.to_tsecr -= tp->ts_offset; 1627 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1628 to.to_tsecr = 0; 1629 else if (tp->t_rxtshift == 1 && 1630 tp->t_flags & TF_PREVVALID && 1631 tp->t_badrxtwin != 0 && 1632 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 1633 cc_cong_signal(tp, th, CC_RTO_ERR); 1634 } 1635 /* 1636 * Process options only when we get SYN/ACK back. The SYN case 1637 * for incoming connections is handled in tcp_syncache. 1638 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1639 * or <SYN,ACK>) segment itself is never scaled. 1640 * XXX this is traditional behavior, may need to be cleaned up. 1641 */ 1642 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1643 /* Handle parallel SYN for ECN */ 1644 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1645 if ((to.to_flags & TOF_SCALE) && 1646 (tp->t_flags & TF_REQ_SCALE) && 1647 !(tp->t_flags & TF_NOOPT)) { 1648 tp->t_flags |= TF_RCVD_SCALE; 1649 tp->snd_scale = to.to_wscale; 1650 } else 1651 tp->t_flags &= ~TF_REQ_SCALE; 1652 /* 1653 * Initial send window. It will be updated with 1654 * the next incoming segment to the scaled value. 1655 */ 1656 tp->snd_wnd = th->th_win; 1657 if ((to.to_flags & TOF_TS) && 1658 (tp->t_flags & TF_REQ_TSTMP) && 1659 !(tp->t_flags & TF_NOOPT)) { 1660 tp->t_flags |= TF_RCVD_TSTMP; 1661 tp->ts_recent = to.to_tsval; 1662 tp->ts_recent_age = tcp_ts_getticks(); 1663 } else 1664 tp->t_flags &= ~TF_REQ_TSTMP; 1665 if (to.to_flags & TOF_MSS) 1666 tcp_mss(tp, to.to_mss); 1667 if ((tp->t_flags & TF_SACK_PERMIT) && 1668 (!(to.to_flags & TOF_SACKPERM) || 1669 (tp->t_flags & TF_NOOPT))) 1670 tp->t_flags &= ~TF_SACK_PERMIT; 1671 if (IS_FASTOPEN(tp->t_flags)) { 1672 if ((to.to_flags & TOF_FASTOPEN) && 1673 !(tp->t_flags & TF_NOOPT)) { 1674 uint16_t mss; 1675 1676 if (to.to_flags & TOF_MSS) 1677 mss = to.to_mss; 1678 else 1679 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1680 mss = TCP6_MSS; 1681 else 1682 mss = TCP_MSS; 1683 tcp_fastopen_update_cache(tp, mss, 1684 to.to_tfo_len, to.to_tfo_cookie); 1685 } else 1686 tcp_fastopen_disable_path(tp); 1687 } 1688 } 1689 1690 /* 1691 * If timestamps were negotiated during SYN/ACK and a 1692 * segment without a timestamp is received, silently drop 1693 * the segment, unless it is a RST segment or missing timestamps are 1694 * tolerated. 1695 * See section 3.2 of RFC 7323. 1696 */ 1697 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1698 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1699 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1700 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1701 "segment processed normally\n", 1702 s, __func__); 1703 free(s, M_TCPLOG); 1704 } 1705 } else { 1706 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1707 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1708 "segment silently dropped\n", s, __func__); 1709 free(s, M_TCPLOG); 1710 } 1711 goto drop; 1712 } 1713 } 1714 /* 1715 * If timestamps were not negotiated during SYN/ACK and a 1716 * segment with a timestamp is received, ignore the 1717 * timestamp and process the packet normally. 1718 * See section 3.2 of RFC 7323. 1719 */ 1720 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1721 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1722 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1723 "segment processed normally\n", s, __func__); 1724 free(s, M_TCPLOG); 1725 } 1726 } 1727 1728 /* 1729 * Header prediction: check for the two common cases 1730 * of a uni-directional data xfer. If the packet has 1731 * no control flags, is in-sequence, the window didn't 1732 * change and we're not retransmitting, it's a 1733 * candidate. If the length is zero and the ack moved 1734 * forward, we're the sender side of the xfer. Just 1735 * free the data acked & wake any higher level process 1736 * that was blocked waiting for space. If the length 1737 * is non-zero and the ack didn't move, we're the 1738 * receiver side. If we're getting packets in-order 1739 * (the reassembly queue is empty), add the data to 1740 * the socket buffer and note that we need a delayed ack. 1741 * Make sure that the hidden state-flags are also off. 1742 * Since we check for TCPS_ESTABLISHED first, it can only 1743 * be TH_NEEDSYN. 1744 */ 1745 if (tp->t_state == TCPS_ESTABLISHED && 1746 th->th_seq == tp->rcv_nxt && 1747 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1748 tp->snd_nxt == tp->snd_max && 1749 tiwin && tiwin == tp->snd_wnd && 1750 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1751 SEGQ_EMPTY(tp) && 1752 ((to.to_flags & TOF_TS) == 0 || 1753 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1754 /* 1755 * If last ACK falls within this segment's sequence numbers, 1756 * record the timestamp. 1757 * NOTE that the test is modified according to the latest 1758 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1759 */ 1760 if ((to.to_flags & TOF_TS) != 0 && 1761 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1762 tp->ts_recent_age = tcp_ts_getticks(); 1763 tp->ts_recent = to.to_tsval; 1764 } 1765 1766 if (tlen == 0) { 1767 if (SEQ_GT(th->th_ack, tp->snd_una) && 1768 SEQ_LEQ(th->th_ack, tp->snd_max) && 1769 !IN_RECOVERY(tp->t_flags) && 1770 (to.to_flags & TOF_SACK) == 0 && 1771 TAILQ_EMPTY(&tp->snd_holes)) { 1772 /* 1773 * This is a pure ack for outstanding data. 1774 */ 1775 TCPSTAT_INC(tcps_predack); 1776 1777 /* 1778 * "bad retransmit" recovery without timestamps. 1779 */ 1780 if ((to.to_flags & TOF_TS) == 0 && 1781 tp->t_rxtshift == 1 && 1782 tp->t_flags & TF_PREVVALID && 1783 tp->t_badrxtwin != 0 && 1784 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1785 cc_cong_signal(tp, th, CC_RTO_ERR); 1786 } 1787 1788 /* 1789 * Recalculate the transmit timer / rtt. 1790 * 1791 * Some boxes send broken timestamp replies 1792 * during the SYN+ACK phase, ignore 1793 * timestamps of 0 or we could calculate a 1794 * huge RTT and blow up the retransmit timer. 1795 */ 1796 if ((to.to_flags & TOF_TS) != 0 && 1797 to.to_tsecr) { 1798 uint32_t t; 1799 1800 t = tcp_ts_getticks() - to.to_tsecr; 1801 if (!tp->t_rttlow || tp->t_rttlow > t) 1802 tp->t_rttlow = t; 1803 tcp_xmit_timer(tp, 1804 TCP_TS_TO_TICKS(t) + 1); 1805 } else if (tp->t_rtttime && 1806 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1807 if (!tp->t_rttlow || 1808 tp->t_rttlow > ticks - tp->t_rtttime) 1809 tp->t_rttlow = ticks - tp->t_rtttime; 1810 tcp_xmit_timer(tp, 1811 ticks - tp->t_rtttime); 1812 } 1813 acked = BYTES_THIS_ACK(tp, th); 1814 1815 #ifdef TCP_HHOOK 1816 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1817 hhook_run_tcp_est_in(tp, th, &to); 1818 #endif 1819 1820 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1821 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1822 sbdrop(&so->so_snd, acked); 1823 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1824 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1825 tp->snd_recover = th->th_ack - 1; 1826 1827 /* 1828 * Let the congestion control algorithm update 1829 * congestion control related information. This 1830 * typically means increasing the congestion 1831 * window. 1832 */ 1833 cc_ack_received(tp, th, nsegs, CC_ACK); 1834 1835 tp->snd_una = th->th_ack; 1836 /* 1837 * Pull snd_wl2 up to prevent seq wrap relative 1838 * to th_ack. 1839 */ 1840 tp->snd_wl2 = th->th_ack; 1841 tp->t_dupacks = 0; 1842 m_freem(m); 1843 1844 /* 1845 * If all outstanding data are acked, stop 1846 * retransmit timer, otherwise restart timer 1847 * using current (possibly backed-off) value. 1848 * If process is waiting for space, 1849 * wakeup/selwakeup/signal. If data 1850 * are ready to send, let tcp_output 1851 * decide between more output or persist. 1852 */ 1853 #ifdef TCPDEBUG 1854 if (so->so_options & SO_DEBUG) 1855 tcp_trace(TA_INPUT, ostate, tp, 1856 (void *)tcp_saveipgen, 1857 &tcp_savetcp, 0); 1858 #endif 1859 TCP_PROBE3(debug__input, tp, th, m); 1860 if (tp->snd_una == tp->snd_max) 1861 tcp_timer_activate(tp, TT_REXMT, 0); 1862 else if (!tcp_timer_active(tp, TT_PERSIST)) 1863 tcp_timer_activate(tp, TT_REXMT, 1864 tp->t_rxtcur); 1865 sowwakeup(so); 1866 if (sbavail(&so->so_snd)) 1867 (void) tcp_output(tp); 1868 goto check_delack; 1869 } 1870 } else if (th->th_ack == tp->snd_una && 1871 tlen <= sbspace(&so->so_rcv)) { 1872 int newsize = 0; /* automatic sockbuf scaling */ 1873 1874 /* 1875 * This is a pure, in-sequence data packet with 1876 * nothing on the reassembly queue and we have enough 1877 * buffer space to take it. 1878 */ 1879 /* Clean receiver SACK report if present */ 1880 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1881 tcp_clean_sackreport(tp); 1882 TCPSTAT_INC(tcps_preddat); 1883 tp->rcv_nxt += tlen; 1884 if (tlen && 1885 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1886 (tp->t_fbyte_in == 0)) { 1887 tp->t_fbyte_in = ticks; 1888 if (tp->t_fbyte_in == 0) 1889 tp->t_fbyte_in = 1; 1890 if (tp->t_fbyte_out && tp->t_fbyte_in) 1891 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1892 } 1893 /* 1894 * Pull snd_wl1 up to prevent seq wrap relative to 1895 * th_seq. 1896 */ 1897 tp->snd_wl1 = th->th_seq; 1898 /* 1899 * Pull rcv_up up to prevent seq wrap relative to 1900 * rcv_nxt. 1901 */ 1902 tp->rcv_up = tp->rcv_nxt; 1903 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1904 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1905 #ifdef TCPDEBUG 1906 if (so->so_options & SO_DEBUG) 1907 tcp_trace(TA_INPUT, ostate, tp, 1908 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1909 #endif 1910 TCP_PROBE3(debug__input, tp, th, m); 1911 1912 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1913 1914 /* Add data to socket buffer. */ 1915 SOCKBUF_LOCK(&so->so_rcv); 1916 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1917 m_freem(m); 1918 } else { 1919 /* 1920 * Set new socket buffer size. 1921 * Give up when limit is reached. 1922 */ 1923 if (newsize) 1924 if (!sbreserve_locked(so, SO_RCV, 1925 newsize, NULL)) 1926 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1927 m_adj(m, drop_hdrlen); /* delayed header drop */ 1928 sbappendstream_locked(&so->so_rcv, m, 0); 1929 } 1930 /* NB: sorwakeup_locked() does an implicit unlock. */ 1931 sorwakeup_locked(so); 1932 if (DELAY_ACK(tp, tlen)) { 1933 tp->t_flags |= TF_DELACK; 1934 } else { 1935 tp->t_flags |= TF_ACKNOW; 1936 tcp_output(tp); 1937 } 1938 goto check_delack; 1939 } 1940 } 1941 1942 /* 1943 * Calculate amount of space in receive window, 1944 * and then do TCP input processing. 1945 * Receive window is amount of space in rcv queue, 1946 * but not less than advertised window. 1947 */ 1948 win = sbspace(&so->so_rcv); 1949 if (win < 0) 1950 win = 0; 1951 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1952 1953 switch (tp->t_state) { 1954 /* 1955 * If the state is SYN_RECEIVED: 1956 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1957 */ 1958 case TCPS_SYN_RECEIVED: 1959 if ((thflags & TH_ACK) && 1960 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1961 SEQ_GT(th->th_ack, tp->snd_max))) { 1962 rstreason = BANDLIM_RST_OPENPORT; 1963 goto dropwithreset; 1964 } 1965 if (IS_FASTOPEN(tp->t_flags)) { 1966 /* 1967 * When a TFO connection is in SYN_RECEIVED, the 1968 * only valid packets are the initial SYN, a 1969 * retransmit/copy of the initial SYN (possibly with 1970 * a subset of the original data), a valid ACK, a 1971 * FIN, or a RST. 1972 */ 1973 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1974 rstreason = BANDLIM_RST_OPENPORT; 1975 goto dropwithreset; 1976 } else if (thflags & TH_SYN) { 1977 /* non-initial SYN is ignored */ 1978 if ((tcp_timer_active(tp, TT_DELACK) || 1979 tcp_timer_active(tp, TT_REXMT))) 1980 goto drop; 1981 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1982 goto drop; 1983 } 1984 } 1985 break; 1986 1987 /* 1988 * If the state is SYN_SENT: 1989 * if seg contains a RST with valid ACK (SEQ.ACK has already 1990 * been verified), then drop the connection. 1991 * if seg contains a RST without an ACK, drop the seg. 1992 * if seg does not contain SYN, then drop the seg. 1993 * Otherwise this is an acceptable SYN segment 1994 * initialize tp->rcv_nxt and tp->irs 1995 * if seg contains ack then advance tp->snd_una 1996 * if seg contains an ECE and ECN support is enabled, the stream 1997 * is ECN capable. 1998 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1999 * arrange for segment to be acked (eventually) 2000 * continue processing rest of data/controls, beginning with URG 2001 */ 2002 case TCPS_SYN_SENT: 2003 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2004 TCP_PROBE5(connect__refused, NULL, tp, 2005 m, tp, th); 2006 tp = tcp_drop(tp, ECONNREFUSED); 2007 } 2008 if (thflags & TH_RST) 2009 goto drop; 2010 if (!(thflags & TH_SYN)) 2011 goto drop; 2012 2013 tp->irs = th->th_seq; 2014 tcp_rcvseqinit(tp); 2015 if (thflags & TH_ACK) { 2016 int tfo_partial_ack = 0; 2017 2018 TCPSTAT_INC(tcps_connects); 2019 soisconnected(so); 2020 #ifdef MAC 2021 mac_socketpeer_set_from_mbuf(m, so); 2022 #endif 2023 /* Do window scaling on this connection? */ 2024 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2025 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2026 tp->rcv_scale = tp->request_r_scale; 2027 } 2028 tp->rcv_adv += min(tp->rcv_wnd, 2029 TCP_MAXWIN << tp->rcv_scale); 2030 tp->snd_una++; /* SYN is acked */ 2031 /* 2032 * If not all the data that was sent in the TFO SYN 2033 * has been acked, resend the remainder right away. 2034 */ 2035 if (IS_FASTOPEN(tp->t_flags) && 2036 (tp->snd_una != tp->snd_max)) { 2037 tp->snd_nxt = th->th_ack; 2038 tfo_partial_ack = 1; 2039 } 2040 /* 2041 * If there's data, delay ACK; if there's also a FIN 2042 * ACKNOW will be turned on later. 2043 */ 2044 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2045 tcp_timer_activate(tp, TT_DELACK, 2046 tcp_delacktime); 2047 else 2048 tp->t_flags |= TF_ACKNOW; 2049 2050 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2051 2052 /* 2053 * Received <SYN,ACK> in SYN_SENT[*] state. 2054 * Transitions: 2055 * SYN_SENT --> ESTABLISHED 2056 * SYN_SENT* --> FIN_WAIT_1 2057 */ 2058 tp->t_starttime = ticks; 2059 if (tp->t_flags & TF_NEEDFIN) { 2060 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2061 tp->t_flags &= ~TF_NEEDFIN; 2062 thflags &= ~TH_SYN; 2063 } else { 2064 tcp_state_change(tp, TCPS_ESTABLISHED); 2065 TCP_PROBE5(connect__established, NULL, tp, 2066 m, tp, th); 2067 cc_conn_init(tp); 2068 tcp_timer_activate(tp, TT_KEEP, 2069 TP_KEEPIDLE(tp)); 2070 } 2071 } else { 2072 /* 2073 * Received initial SYN in SYN-SENT[*] state => 2074 * simultaneous open. 2075 * If it succeeds, connection is * half-synchronized. 2076 * Otherwise, do 3-way handshake: 2077 * SYN-SENT -> SYN-RECEIVED 2078 * SYN-SENT* -> SYN-RECEIVED* 2079 */ 2080 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2081 tcp_timer_activate(tp, TT_REXMT, 0); 2082 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2083 } 2084 2085 INP_WLOCK_ASSERT(tp->t_inpcb); 2086 2087 /* 2088 * Advance th->th_seq to correspond to first data byte. 2089 * If data, trim to stay within window, 2090 * dropping FIN if necessary. 2091 */ 2092 th->th_seq++; 2093 if (tlen > tp->rcv_wnd) { 2094 todrop = tlen - tp->rcv_wnd; 2095 m_adj(m, -todrop); 2096 tlen = tp->rcv_wnd; 2097 thflags &= ~TH_FIN; 2098 TCPSTAT_INC(tcps_rcvpackafterwin); 2099 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2100 } 2101 tp->snd_wl1 = th->th_seq - 1; 2102 tp->rcv_up = th->th_seq; 2103 /* 2104 * Client side of transaction: already sent SYN and data. 2105 * If the remote host used T/TCP to validate the SYN, 2106 * our data will be ACK'd; if so, enter normal data segment 2107 * processing in the middle of step 5, ack processing. 2108 * Otherwise, goto step 6. 2109 */ 2110 if (thflags & TH_ACK) 2111 goto process_ACK; 2112 2113 goto step6; 2114 2115 /* 2116 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2117 * do normal processing. 2118 * 2119 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2120 */ 2121 case TCPS_LAST_ACK: 2122 case TCPS_CLOSING: 2123 break; /* continue normal processing */ 2124 } 2125 2126 /* 2127 * States other than LISTEN or SYN_SENT. 2128 * First check the RST flag and sequence number since reset segments 2129 * are exempt from the timestamp and connection count tests. This 2130 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2131 * below which allowed reset segments in half the sequence space 2132 * to fall though and be processed (which gives forged reset 2133 * segments with a random sequence number a 50 percent chance of 2134 * killing a connection). 2135 * Then check timestamp, if present. 2136 * Then check the connection count, if present. 2137 * Then check that at least some bytes of segment are within 2138 * receive window. If segment begins before rcv_nxt, 2139 * drop leading data (and SYN); if nothing left, just ack. 2140 */ 2141 if (thflags & TH_RST) { 2142 /* 2143 * RFC5961 Section 3.2 2144 * 2145 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2146 * - If RST is in window, we send challenge ACK. 2147 * 2148 * Note: to take into account delayed ACKs, we should 2149 * test against last_ack_sent instead of rcv_nxt. 2150 * Note 2: we handle special case of closed window, not 2151 * covered by the RFC. 2152 */ 2153 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2154 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2155 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2156 KASSERT(tp->t_state != TCPS_SYN_SENT, 2157 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2158 __func__, th, tp)); 2159 2160 if (V_tcp_insecure_rst || 2161 tp->last_ack_sent == th->th_seq) { 2162 TCPSTAT_INC(tcps_drops); 2163 /* Drop the connection. */ 2164 switch (tp->t_state) { 2165 case TCPS_SYN_RECEIVED: 2166 so->so_error = ECONNREFUSED; 2167 goto close; 2168 case TCPS_ESTABLISHED: 2169 case TCPS_FIN_WAIT_1: 2170 case TCPS_FIN_WAIT_2: 2171 case TCPS_CLOSE_WAIT: 2172 case TCPS_CLOSING: 2173 case TCPS_LAST_ACK: 2174 so->so_error = ECONNRESET; 2175 close: 2176 /* FALLTHROUGH */ 2177 default: 2178 tp = tcp_close(tp); 2179 } 2180 } else { 2181 TCPSTAT_INC(tcps_badrst); 2182 /* Send challenge ACK. */ 2183 tcp_respond(tp, mtod(m, void *), th, m, 2184 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2185 tp->last_ack_sent = tp->rcv_nxt; 2186 m = NULL; 2187 } 2188 } 2189 goto drop; 2190 } 2191 2192 /* 2193 * RFC5961 Section 4.2 2194 * Send challenge ACK for any SYN in synchronized state. 2195 */ 2196 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2197 tp->t_state != TCPS_SYN_RECEIVED) { 2198 TCPSTAT_INC(tcps_badsyn); 2199 if (V_tcp_insecure_syn && 2200 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2201 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2202 tp = tcp_drop(tp, ECONNRESET); 2203 rstreason = BANDLIM_UNLIMITED; 2204 } else { 2205 /* Send challenge ACK. */ 2206 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2207 tp->snd_nxt, TH_ACK); 2208 tp->last_ack_sent = tp->rcv_nxt; 2209 m = NULL; 2210 } 2211 goto drop; 2212 } 2213 2214 /* 2215 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2216 * and it's less than ts_recent, drop it. 2217 */ 2218 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2219 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2220 /* Check to see if ts_recent is over 24 days old. */ 2221 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2222 /* 2223 * Invalidate ts_recent. If this segment updates 2224 * ts_recent, the age will be reset later and ts_recent 2225 * will get a valid value. If it does not, setting 2226 * ts_recent to zero will at least satisfy the 2227 * requirement that zero be placed in the timestamp 2228 * echo reply when ts_recent isn't valid. The 2229 * age isn't reset until we get a valid ts_recent 2230 * because we don't want out-of-order segments to be 2231 * dropped when ts_recent is old. 2232 */ 2233 tp->ts_recent = 0; 2234 } else { 2235 TCPSTAT_INC(tcps_rcvduppack); 2236 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2237 TCPSTAT_INC(tcps_pawsdrop); 2238 if (tlen) 2239 goto dropafterack; 2240 goto drop; 2241 } 2242 } 2243 2244 /* 2245 * In the SYN-RECEIVED state, validate that the packet belongs to 2246 * this connection before trimming the data to fit the receive 2247 * window. Check the sequence number versus IRS since we know 2248 * the sequence numbers haven't wrapped. This is a partial fix 2249 * for the "LAND" DoS attack. 2250 */ 2251 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2252 rstreason = BANDLIM_RST_OPENPORT; 2253 goto dropwithreset; 2254 } 2255 2256 todrop = tp->rcv_nxt - th->th_seq; 2257 if (todrop > 0) { 2258 if (thflags & TH_SYN) { 2259 thflags &= ~TH_SYN; 2260 th->th_seq++; 2261 if (th->th_urp > 1) 2262 th->th_urp--; 2263 else 2264 thflags &= ~TH_URG; 2265 todrop--; 2266 } 2267 /* 2268 * Following if statement from Stevens, vol. 2, p. 960. 2269 */ 2270 if (todrop > tlen 2271 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2272 /* 2273 * Any valid FIN must be to the left of the window. 2274 * At this point the FIN must be a duplicate or out 2275 * of sequence; drop it. 2276 */ 2277 thflags &= ~TH_FIN; 2278 2279 /* 2280 * Send an ACK to resynchronize and drop any data. 2281 * But keep on processing for RST or ACK. 2282 */ 2283 tp->t_flags |= TF_ACKNOW; 2284 todrop = tlen; 2285 TCPSTAT_INC(tcps_rcvduppack); 2286 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2287 } else { 2288 TCPSTAT_INC(tcps_rcvpartduppack); 2289 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2290 } 2291 /* 2292 * DSACK - add SACK block for dropped range 2293 */ 2294 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2295 tcp_update_sack_list(tp, th->th_seq, 2296 th->th_seq + todrop); 2297 /* 2298 * ACK now, as the next in-sequence segment 2299 * will clear the DSACK block again 2300 */ 2301 tp->t_flags |= TF_ACKNOW; 2302 } 2303 drop_hdrlen += todrop; /* drop from the top afterwards */ 2304 th->th_seq += todrop; 2305 tlen -= todrop; 2306 if (th->th_urp > todrop) 2307 th->th_urp -= todrop; 2308 else { 2309 thflags &= ~TH_URG; 2310 th->th_urp = 0; 2311 } 2312 } 2313 2314 /* 2315 * If new data are received on a connection after the 2316 * user processes are gone, then RST the other end. 2317 */ 2318 if ((tp->t_flags & TF_CLOSED) && tlen) { 2319 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2320 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2321 "after socket was closed, " 2322 "sending RST and removing tcpcb\n", 2323 s, __func__, tcpstates[tp->t_state], tlen); 2324 free(s, M_TCPLOG); 2325 } 2326 tp = tcp_close(tp); 2327 TCPSTAT_INC(tcps_rcvafterclose); 2328 rstreason = BANDLIM_UNLIMITED; 2329 goto dropwithreset; 2330 } 2331 2332 /* 2333 * If segment ends after window, drop trailing data 2334 * (and PUSH and FIN); if nothing left, just ACK. 2335 */ 2336 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2337 if (todrop > 0) { 2338 TCPSTAT_INC(tcps_rcvpackafterwin); 2339 if (todrop >= tlen) { 2340 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2341 /* 2342 * If window is closed can only take segments at 2343 * window edge, and have to drop data and PUSH from 2344 * incoming segments. Continue processing, but 2345 * remember to ack. Otherwise, drop segment 2346 * and ack. 2347 */ 2348 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2349 tp->t_flags |= TF_ACKNOW; 2350 TCPSTAT_INC(tcps_rcvwinprobe); 2351 } else 2352 goto dropafterack; 2353 } else 2354 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2355 m_adj(m, -todrop); 2356 tlen -= todrop; 2357 thflags &= ~(TH_PUSH|TH_FIN); 2358 } 2359 2360 /* 2361 * If last ACK falls within this segment's sequence numbers, 2362 * record its timestamp. 2363 * NOTE: 2364 * 1) That the test incorporates suggestions from the latest 2365 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2366 * 2) That updating only on newer timestamps interferes with 2367 * our earlier PAWS tests, so this check should be solely 2368 * predicated on the sequence space of this segment. 2369 * 3) That we modify the segment boundary check to be 2370 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2371 * instead of RFC1323's 2372 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2373 * This modified check allows us to overcome RFC1323's 2374 * limitations as described in Stevens TCP/IP Illustrated 2375 * Vol. 2 p.869. In such cases, we can still calculate the 2376 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2377 */ 2378 if ((to.to_flags & TOF_TS) != 0 && 2379 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2380 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2381 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2382 tp->ts_recent_age = tcp_ts_getticks(); 2383 tp->ts_recent = to.to_tsval; 2384 } 2385 2386 /* 2387 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2388 * flag is on (half-synchronized state), then queue data for 2389 * later processing; else drop segment and return. 2390 */ 2391 if ((thflags & TH_ACK) == 0) { 2392 if (tp->t_state == TCPS_SYN_RECEIVED || 2393 (tp->t_flags & TF_NEEDSYN)) { 2394 if (tp->t_state == TCPS_SYN_RECEIVED && 2395 IS_FASTOPEN(tp->t_flags)) { 2396 tp->snd_wnd = tiwin; 2397 cc_conn_init(tp); 2398 } 2399 goto step6; 2400 } else if (tp->t_flags & TF_ACKNOW) 2401 goto dropafterack; 2402 else 2403 goto drop; 2404 } 2405 2406 /* 2407 * Ack processing. 2408 */ 2409 switch (tp->t_state) { 2410 /* 2411 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2412 * ESTABLISHED state and continue processing. 2413 * The ACK was checked above. 2414 */ 2415 case TCPS_SYN_RECEIVED: 2416 2417 TCPSTAT_INC(tcps_connects); 2418 soisconnected(so); 2419 /* Do window scaling? */ 2420 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2421 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2422 tp->rcv_scale = tp->request_r_scale; 2423 } 2424 tp->snd_wnd = tiwin; 2425 /* 2426 * Make transitions: 2427 * SYN-RECEIVED -> ESTABLISHED 2428 * SYN-RECEIVED* -> FIN-WAIT-1 2429 */ 2430 tp->t_starttime = ticks; 2431 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2432 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2433 tp->t_tfo_pending = NULL; 2434 } 2435 if (tp->t_flags & TF_NEEDFIN) { 2436 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2437 tp->t_flags &= ~TF_NEEDFIN; 2438 } else { 2439 tcp_state_change(tp, TCPS_ESTABLISHED); 2440 TCP_PROBE5(accept__established, NULL, tp, 2441 m, tp, th); 2442 /* 2443 * TFO connections call cc_conn_init() during SYN 2444 * processing. Calling it again here for such 2445 * connections is not harmless as it would undo the 2446 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2447 * is retransmitted. 2448 */ 2449 if (!IS_FASTOPEN(tp->t_flags)) 2450 cc_conn_init(tp); 2451 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2452 } 2453 /* 2454 * Account for the ACK of our SYN prior to 2455 * regular ACK processing below, except for 2456 * simultaneous SYN, which is handled later. 2457 */ 2458 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2459 incforsyn = 1; 2460 /* 2461 * If segment contains data or ACK, will call tcp_reass() 2462 * later; if not, do so now to pass queued data to user. 2463 */ 2464 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2465 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2466 (struct mbuf *)0); 2467 tcp_handle_wakeup(tp, so); 2468 } 2469 tp->snd_wl1 = th->th_seq - 1; 2470 /* FALLTHROUGH */ 2471 2472 /* 2473 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2474 * ACKs. If the ack is in the range 2475 * tp->snd_una < th->th_ack <= tp->snd_max 2476 * then advance tp->snd_una to th->th_ack and drop 2477 * data from the retransmission queue. If this ACK reflects 2478 * more up to date window information we update our window information. 2479 */ 2480 case TCPS_ESTABLISHED: 2481 case TCPS_FIN_WAIT_1: 2482 case TCPS_FIN_WAIT_2: 2483 case TCPS_CLOSE_WAIT: 2484 case TCPS_CLOSING: 2485 case TCPS_LAST_ACK: 2486 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2487 TCPSTAT_INC(tcps_rcvacktoomuch); 2488 goto dropafterack; 2489 } 2490 if ((tp->t_flags & TF_SACK_PERMIT) && 2491 ((to.to_flags & TOF_SACK) || 2492 !TAILQ_EMPTY(&tp->snd_holes))) { 2493 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && 2494 (tp->t_flags & TF_LRD)) { 2495 tcp_sack_lost_retransmission(tp, th); 2496 } 2497 } else 2498 /* 2499 * Reset the value so that previous (valid) value 2500 * from the last ack with SACK doesn't get used. 2501 */ 2502 tp->sackhint.sacked_bytes = 0; 2503 2504 #ifdef TCP_HHOOK 2505 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2506 hhook_run_tcp_est_in(tp, th, &to); 2507 #endif 2508 2509 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2510 maxseg = tcp_maxseg(tp); 2511 if (tlen == 0 && 2512 (tiwin == tp->snd_wnd || 2513 (tp->t_flags & TF_SACK_PERMIT))) { 2514 /* 2515 * If this is the first time we've seen a 2516 * FIN from the remote, this is not a 2517 * duplicate and it needs to be processed 2518 * normally. This happens during a 2519 * simultaneous close. 2520 */ 2521 if ((thflags & TH_FIN) && 2522 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2523 tp->t_dupacks = 0; 2524 break; 2525 } 2526 TCPSTAT_INC(tcps_rcvdupack); 2527 /* 2528 * If we have outstanding data (other than 2529 * a window probe), this is a completely 2530 * duplicate ack (ie, window info didn't 2531 * change and FIN isn't set), 2532 * the ack is the biggest we've 2533 * seen and we've seen exactly our rexmt 2534 * threshold of them, assume a packet 2535 * has been dropped and retransmit it. 2536 * Kludge snd_nxt & the congestion 2537 * window so we send only this one 2538 * packet. 2539 * 2540 * We know we're losing at the current 2541 * window size so do congestion avoidance 2542 * (set ssthresh to half the current window 2543 * and pull our congestion window back to 2544 * the new ssthresh). 2545 * 2546 * Dup acks mean that packets have left the 2547 * network (they're now cached at the receiver) 2548 * so bump cwnd by the amount in the receiver 2549 * to keep a constant cwnd packets in the 2550 * network. 2551 * 2552 * When using TCP ECN, notify the peer that 2553 * we reduced the cwnd. 2554 */ 2555 /* 2556 * Following 2 kinds of acks should not affect 2557 * dupack counting: 2558 * 1) Old acks 2559 * 2) Acks with SACK but without any new SACK 2560 * information in them. These could result from 2561 * any anomaly in the network like a switch 2562 * duplicating packets or a possible DoS attack. 2563 */ 2564 if (th->th_ack != tp->snd_una || 2565 ((tp->t_flags & TF_SACK_PERMIT) && 2566 (to.to_flags & TOF_SACK) && 2567 !sack_changed)) 2568 break; 2569 else if (!tcp_timer_active(tp, TT_REXMT)) 2570 tp->t_dupacks = 0; 2571 else if (++tp->t_dupacks > tcprexmtthresh || 2572 IN_FASTRECOVERY(tp->t_flags)) { 2573 cc_ack_received(tp, th, nsegs, 2574 CC_DUPACK); 2575 if (V_tcp_do_prr && 2576 IN_FASTRECOVERY(tp->t_flags)) { 2577 tcp_do_prr_ack(tp, th, &to); 2578 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2579 (to.to_flags & TOF_SACK) && 2580 IN_FASTRECOVERY(tp->t_flags)) { 2581 int awnd; 2582 2583 /* 2584 * Compute the amount of data in flight first. 2585 * We can inject new data into the pipe iff 2586 * we have less than 1/2 the original window's 2587 * worth of data in flight. 2588 */ 2589 if (V_tcp_do_newsack) 2590 awnd = tcp_compute_pipe(tp); 2591 else 2592 awnd = (tp->snd_nxt - tp->snd_fack) + 2593 tp->sackhint.sack_bytes_rexmit; 2594 2595 if (awnd < tp->snd_ssthresh) { 2596 tp->snd_cwnd += maxseg; 2597 if (tp->snd_cwnd > tp->snd_ssthresh) 2598 tp->snd_cwnd = tp->snd_ssthresh; 2599 } 2600 } else 2601 tp->snd_cwnd += maxseg; 2602 (void) tcp_output(tp); 2603 goto drop; 2604 } else if (tp->t_dupacks == tcprexmtthresh || 2605 (tp->t_flags & TF_SACK_PERMIT && 2606 V_tcp_do_newsack && 2607 tp->sackhint.sacked_bytes > 2608 (tcprexmtthresh - 1) * maxseg)) { 2609 enter_recovery: 2610 /* 2611 * Above is the RFC6675 trigger condition of 2612 * more than (dupthresh-1)*maxseg sacked data. 2613 * If the count of holes in the 2614 * scoreboard is >= dupthresh, we could 2615 * also enter loss recovery, but don't 2616 * have that value readily available. 2617 */ 2618 tp->t_dupacks = tcprexmtthresh; 2619 tcp_seq onxt = tp->snd_nxt; 2620 2621 /* 2622 * If we're doing sack, or prr, check 2623 * to see if we're already in sack 2624 * recovery. If we're not doing sack, 2625 * check to see if we're in newreno 2626 * recovery. 2627 */ 2628 if (V_tcp_do_prr || 2629 (tp->t_flags & TF_SACK_PERMIT)) { 2630 if (IN_FASTRECOVERY(tp->t_flags)) { 2631 tp->t_dupacks = 0; 2632 break; 2633 } 2634 } else { 2635 if (SEQ_LEQ(th->th_ack, 2636 tp->snd_recover)) { 2637 tp->t_dupacks = 0; 2638 break; 2639 } 2640 } 2641 /* Congestion signal before ack. */ 2642 cc_cong_signal(tp, th, CC_NDUPACK); 2643 cc_ack_received(tp, th, nsegs, 2644 CC_DUPACK); 2645 tcp_timer_activate(tp, TT_REXMT, 0); 2646 tp->t_rtttime = 0; 2647 if (V_tcp_do_prr) { 2648 /* 2649 * snd_ssthresh is already updated by 2650 * cc_cong_signal. 2651 */ 2652 if ((tp->t_flags & TF_SACK_PERMIT) && 2653 (to.to_flags & TOF_SACK)) { 2654 tp->sackhint.prr_delivered = 2655 tp->sackhint.sacked_bytes; 2656 } else { 2657 tp->sackhint.prr_delivered = 2658 imin(tp->snd_max - tp->snd_una, 2659 imin(INT_MAX / 65536, 2660 tp->t_dupacks) * maxseg); 2661 } 2662 tp->sackhint.recover_fs = max(1, 2663 tp->snd_nxt - tp->snd_una); 2664 } 2665 if ((tp->t_flags & TF_SACK_PERMIT) && 2666 (to.to_flags & TOF_SACK)) { 2667 TCPSTAT_INC( 2668 tcps_sack_recovery_episode); 2669 tp->snd_recover = tp->snd_nxt; 2670 tp->snd_cwnd = maxseg; 2671 (void) tcp_output(tp); 2672 if (SEQ_GT(th->th_ack, tp->snd_una)) 2673 goto resume_partialack; 2674 goto drop; 2675 } 2676 tp->snd_nxt = th->th_ack; 2677 tp->snd_cwnd = maxseg; 2678 (void) tcp_output(tp); 2679 KASSERT(tp->snd_limited <= 2, 2680 ("%s: tp->snd_limited too big", 2681 __func__)); 2682 tp->snd_cwnd = tp->snd_ssthresh + 2683 maxseg * 2684 (tp->t_dupacks - tp->snd_limited); 2685 if (SEQ_GT(onxt, tp->snd_nxt)) 2686 tp->snd_nxt = onxt; 2687 goto drop; 2688 } else if (V_tcp_do_rfc3042) { 2689 /* 2690 * Process first and second duplicate 2691 * ACKs. Each indicates a segment 2692 * leaving the network, creating room 2693 * for more. Make sure we can send a 2694 * packet on reception of each duplicate 2695 * ACK by increasing snd_cwnd by one 2696 * segment. Restore the original 2697 * snd_cwnd after packet transmission. 2698 */ 2699 cc_ack_received(tp, th, nsegs, 2700 CC_DUPACK); 2701 uint32_t oldcwnd = tp->snd_cwnd; 2702 tcp_seq oldsndmax = tp->snd_max; 2703 u_int sent; 2704 int avail; 2705 2706 KASSERT(tp->t_dupacks == 1 || 2707 tp->t_dupacks == 2, 2708 ("%s: dupacks not 1 or 2", 2709 __func__)); 2710 if (tp->t_dupacks == 1) 2711 tp->snd_limited = 0; 2712 tp->snd_cwnd = 2713 (tp->snd_nxt - tp->snd_una) + 2714 (tp->t_dupacks - tp->snd_limited) * 2715 maxseg; 2716 /* 2717 * Only call tcp_output when there 2718 * is new data available to be sent. 2719 * Otherwise we would send pure ACKs. 2720 */ 2721 SOCKBUF_LOCK(&so->so_snd); 2722 avail = sbavail(&so->so_snd) - 2723 (tp->snd_nxt - tp->snd_una); 2724 SOCKBUF_UNLOCK(&so->so_snd); 2725 if (avail > 0) 2726 (void) tcp_output(tp); 2727 sent = tp->snd_max - oldsndmax; 2728 if (sent > maxseg) { 2729 KASSERT((tp->t_dupacks == 2 && 2730 tp->snd_limited == 0) || 2731 (sent == maxseg + 1 && 2732 tp->t_flags & TF_SENTFIN), 2733 ("%s: sent too much", 2734 __func__)); 2735 tp->snd_limited = 2; 2736 } else if (sent > 0) 2737 ++tp->snd_limited; 2738 tp->snd_cwnd = oldcwnd; 2739 goto drop; 2740 } 2741 } 2742 break; 2743 } else { 2744 /* 2745 * This ack is advancing the left edge, reset the 2746 * counter. 2747 */ 2748 tp->t_dupacks = 0; 2749 /* 2750 * If this ack also has new SACK info, increment the 2751 * counter as per rfc6675. The variable 2752 * sack_changed tracks all changes to the SACK 2753 * scoreboard, including when partial ACKs without 2754 * SACK options are received, and clear the scoreboard 2755 * from the left side. Such partial ACKs should not be 2756 * counted as dupacks here. 2757 */ 2758 if ((tp->t_flags & TF_SACK_PERMIT) && 2759 (to.to_flags & TOF_SACK) && 2760 sack_changed) { 2761 tp->t_dupacks++; 2762 /* limit overhead by setting maxseg last */ 2763 if (!IN_FASTRECOVERY(tp->t_flags) && 2764 (tp->sackhint.sacked_bytes > 2765 ((tcprexmtthresh - 1) * 2766 (maxseg = tcp_maxseg(tp))))) { 2767 goto enter_recovery; 2768 } 2769 } 2770 } 2771 2772 resume_partialack: 2773 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2774 ("%s: th_ack <= snd_una", __func__)); 2775 2776 /* 2777 * If the congestion window was inflated to account 2778 * for the other side's cached packets, retract it. 2779 */ 2780 if (IN_FASTRECOVERY(tp->t_flags)) { 2781 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2782 if (tp->t_flags & TF_SACK_PERMIT) 2783 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2784 tcp_timer_activate(tp, TT_REXMT, 0); 2785 tp->t_rtttime = 0; 2786 tcp_do_prr_ack(tp, th, &to); 2787 tp->t_flags |= TF_ACKNOW; 2788 (void) tcp_output(tp); 2789 } else 2790 tcp_sack_partialack(tp, th); 2791 else 2792 tcp_newreno_partial_ack(tp, th); 2793 } else 2794 cc_post_recovery(tp, th); 2795 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2796 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2797 if (V_tcp_do_prr) { 2798 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2799 tp->snd_fack = th->th_ack; 2800 tcp_do_prr_ack(tp, th, &to); 2801 (void) tcp_output(tp); 2802 } 2803 } else 2804 cc_post_recovery(tp, th); 2805 } 2806 /* 2807 * If we reach this point, ACK is not a duplicate, 2808 * i.e., it ACKs something we sent. 2809 */ 2810 if (tp->t_flags & TF_NEEDSYN) { 2811 /* 2812 * T/TCP: Connection was half-synchronized, and our 2813 * SYN has been ACK'd (so connection is now fully 2814 * synchronized). Go to non-starred state, 2815 * increment snd_una for ACK of SYN, and check if 2816 * we can do window scaling. 2817 */ 2818 tp->t_flags &= ~TF_NEEDSYN; 2819 tp->snd_una++; 2820 /* Do window scaling? */ 2821 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2822 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2823 tp->rcv_scale = tp->request_r_scale; 2824 /* Send window already scaled. */ 2825 } 2826 } 2827 2828 process_ACK: 2829 INP_WLOCK_ASSERT(tp->t_inpcb); 2830 2831 /* 2832 * Adjust for the SYN bit in sequence space, 2833 * but don't account for it in cwnd calculations. 2834 * This is for the SYN_RECEIVED, non-simultaneous 2835 * SYN case. SYN_SENT and simultaneous SYN are 2836 * treated elsewhere. 2837 */ 2838 if (incforsyn) 2839 tp->snd_una++; 2840 acked = BYTES_THIS_ACK(tp, th); 2841 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2842 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2843 tp->snd_una, th->th_ack, tp, m)); 2844 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2845 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2846 2847 /* 2848 * If we just performed our first retransmit, and the ACK 2849 * arrives within our recovery window, then it was a mistake 2850 * to do the retransmit in the first place. Recover our 2851 * original cwnd and ssthresh, and proceed to transmit where 2852 * we left off. 2853 */ 2854 if (tp->t_rxtshift == 1 && 2855 tp->t_flags & TF_PREVVALID && 2856 tp->t_badrxtwin != 0 && 2857 to.to_flags & TOF_TS && 2858 to.to_tsecr != 0 && 2859 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2860 cc_cong_signal(tp, th, CC_RTO_ERR); 2861 2862 /* 2863 * If we have a timestamp reply, update smoothed 2864 * round trip time. If no timestamp is present but 2865 * transmit timer is running and timed sequence 2866 * number was acked, update smoothed round trip time. 2867 * Since we now have an rtt measurement, cancel the 2868 * timer backoff (cf., Phil Karn's retransmit alg.). 2869 * Recompute the initial retransmit timer. 2870 * 2871 * Some boxes send broken timestamp replies 2872 * during the SYN+ACK phase, ignore 2873 * timestamps of 0 or we could calculate a 2874 * huge RTT and blow up the retransmit timer. 2875 */ 2876 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2877 uint32_t t; 2878 2879 t = tcp_ts_getticks() - to.to_tsecr; 2880 if (!tp->t_rttlow || tp->t_rttlow > t) 2881 tp->t_rttlow = t; 2882 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2883 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2884 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2885 tp->t_rttlow = ticks - tp->t_rtttime; 2886 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2887 } 2888 2889 /* 2890 * If all outstanding data is acked, stop retransmit 2891 * timer and remember to restart (more output or persist). 2892 * If there is more data to be acked, restart retransmit 2893 * timer, using current (possibly backed-off) value. 2894 */ 2895 if (th->th_ack == tp->snd_max) { 2896 tcp_timer_activate(tp, TT_REXMT, 0); 2897 needoutput = 1; 2898 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2899 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2900 2901 /* 2902 * If no data (only SYN) was ACK'd, 2903 * skip rest of ACK processing. 2904 */ 2905 if (acked == 0) 2906 goto step6; 2907 2908 /* 2909 * Let the congestion control algorithm update congestion 2910 * control related information. This typically means increasing 2911 * the congestion window. 2912 */ 2913 cc_ack_received(tp, th, nsegs, CC_ACK); 2914 2915 SOCKBUF_LOCK(&so->so_snd); 2916 if (acked > sbavail(&so->so_snd)) { 2917 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2918 tp->snd_wnd -= sbavail(&so->so_snd); 2919 else 2920 tp->snd_wnd = 0; 2921 mfree = sbcut_locked(&so->so_snd, 2922 (int)sbavail(&so->so_snd)); 2923 ourfinisacked = 1; 2924 } else { 2925 mfree = sbcut_locked(&so->so_snd, acked); 2926 if (tp->snd_wnd >= (uint32_t) acked) 2927 tp->snd_wnd -= acked; 2928 else 2929 tp->snd_wnd = 0; 2930 ourfinisacked = 0; 2931 } 2932 /* NB: sowwakeup_locked() does an implicit unlock. */ 2933 sowwakeup_locked(so); 2934 m_freem(mfree); 2935 /* Detect una wraparound. */ 2936 if (!IN_RECOVERY(tp->t_flags) && 2937 SEQ_GT(tp->snd_una, tp->snd_recover) && 2938 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2939 tp->snd_recover = th->th_ack - 1; 2940 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2941 if (IN_RECOVERY(tp->t_flags) && 2942 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2943 EXIT_RECOVERY(tp->t_flags); 2944 } 2945 tp->snd_una = th->th_ack; 2946 if (tp->t_flags & TF_SACK_PERMIT) { 2947 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2948 tp->snd_recover = tp->snd_una; 2949 } 2950 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2951 tp->snd_nxt = tp->snd_una; 2952 2953 switch (tp->t_state) { 2954 /* 2955 * In FIN_WAIT_1 STATE in addition to the processing 2956 * for the ESTABLISHED state if our FIN is now acknowledged 2957 * then enter FIN_WAIT_2. 2958 */ 2959 case TCPS_FIN_WAIT_1: 2960 if (ourfinisacked) { 2961 /* 2962 * If we can't receive any more 2963 * data, then closing user can proceed. 2964 * Starting the timer is contrary to the 2965 * specification, but if we don't get a FIN 2966 * we'll hang forever. 2967 * 2968 * XXXjl: 2969 * we should release the tp also, and use a 2970 * compressed state. 2971 */ 2972 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2973 soisdisconnected(so); 2974 tcp_timer_activate(tp, TT_2MSL, 2975 (tcp_fast_finwait2_recycle ? 2976 tcp_finwait2_timeout : 2977 TP_MAXIDLE(tp))); 2978 } 2979 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2980 } 2981 break; 2982 2983 /* 2984 * In CLOSING STATE in addition to the processing for 2985 * the ESTABLISHED state if the ACK acknowledges our FIN 2986 * then enter the TIME-WAIT state, otherwise ignore 2987 * the segment. 2988 */ 2989 case TCPS_CLOSING: 2990 if (ourfinisacked) { 2991 tcp_twstart(tp); 2992 m_freem(m); 2993 return; 2994 } 2995 break; 2996 2997 /* 2998 * In LAST_ACK, we may still be waiting for data to drain 2999 * and/or to be acked, as well as for the ack of our FIN. 3000 * If our FIN is now acknowledged, delete the TCB, 3001 * enter the closed state and return. 3002 */ 3003 case TCPS_LAST_ACK: 3004 if (ourfinisacked) { 3005 tp = tcp_close(tp); 3006 goto drop; 3007 } 3008 break; 3009 } 3010 } 3011 3012 step6: 3013 INP_WLOCK_ASSERT(tp->t_inpcb); 3014 3015 /* 3016 * Update window information. 3017 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3018 */ 3019 if ((thflags & TH_ACK) && 3020 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3021 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3022 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3023 /* keep track of pure window updates */ 3024 if (tlen == 0 && 3025 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3026 TCPSTAT_INC(tcps_rcvwinupd); 3027 tp->snd_wnd = tiwin; 3028 tp->snd_wl1 = th->th_seq; 3029 tp->snd_wl2 = th->th_ack; 3030 if (tp->snd_wnd > tp->max_sndwnd) 3031 tp->max_sndwnd = tp->snd_wnd; 3032 needoutput = 1; 3033 } 3034 3035 /* 3036 * Process segments with URG. 3037 */ 3038 if ((thflags & TH_URG) && th->th_urp && 3039 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3040 /* 3041 * This is a kludge, but if we receive and accept 3042 * random urgent pointers, we'll crash in 3043 * soreceive. It's hard to imagine someone 3044 * actually wanting to send this much urgent data. 3045 */ 3046 SOCKBUF_LOCK(&so->so_rcv); 3047 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3048 th->th_urp = 0; /* XXX */ 3049 thflags &= ~TH_URG; /* XXX */ 3050 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3051 goto dodata; /* XXX */ 3052 } 3053 /* 3054 * If this segment advances the known urgent pointer, 3055 * then mark the data stream. This should not happen 3056 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3057 * a FIN has been received from the remote side. 3058 * In these states we ignore the URG. 3059 * 3060 * According to RFC961 (Assigned Protocols), 3061 * the urgent pointer points to the last octet 3062 * of urgent data. We continue, however, 3063 * to consider it to indicate the first octet 3064 * of data past the urgent section as the original 3065 * spec states (in one of two places). 3066 */ 3067 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3068 tp->rcv_up = th->th_seq + th->th_urp; 3069 so->so_oobmark = sbavail(&so->so_rcv) + 3070 (tp->rcv_up - tp->rcv_nxt) - 1; 3071 if (so->so_oobmark == 0) 3072 so->so_rcv.sb_state |= SBS_RCVATMARK; 3073 sohasoutofband(so); 3074 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3075 } 3076 SOCKBUF_UNLOCK(&so->so_rcv); 3077 /* 3078 * Remove out of band data so doesn't get presented to user. 3079 * This can happen independent of advancing the URG pointer, 3080 * but if two URG's are pending at once, some out-of-band 3081 * data may creep in... ick. 3082 */ 3083 if (th->th_urp <= (uint32_t)tlen && 3084 !(so->so_options & SO_OOBINLINE)) { 3085 /* hdr drop is delayed */ 3086 tcp_pulloutofband(so, th, m, drop_hdrlen); 3087 } 3088 } else { 3089 /* 3090 * If no out of band data is expected, 3091 * pull receive urgent pointer along 3092 * with the receive window. 3093 */ 3094 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3095 tp->rcv_up = tp->rcv_nxt; 3096 } 3097 dodata: /* XXX */ 3098 INP_WLOCK_ASSERT(tp->t_inpcb); 3099 3100 /* 3101 * Process the segment text, merging it into the TCP sequencing queue, 3102 * and arranging for acknowledgment of receipt if necessary. 3103 * This process logically involves adjusting tp->rcv_wnd as data 3104 * is presented to the user (this happens in tcp_usrreq.c, 3105 * case PRU_RCVD). If a FIN has already been received on this 3106 * connection then we just ignore the text. 3107 */ 3108 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3109 IS_FASTOPEN(tp->t_flags)); 3110 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3111 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3112 tcp_seq save_start = th->th_seq; 3113 tcp_seq save_rnxt = tp->rcv_nxt; 3114 int save_tlen = tlen; 3115 m_adj(m, drop_hdrlen); /* delayed header drop */ 3116 /* 3117 * Insert segment which includes th into TCP reassembly queue 3118 * with control block tp. Set thflags to whether reassembly now 3119 * includes a segment with FIN. This handles the common case 3120 * inline (segment is the next to be received on an established 3121 * connection, and the queue is empty), avoiding linkage into 3122 * and removal from the queue and repetition of various 3123 * conversions. 3124 * Set DELACK for segments received in order, but ack 3125 * immediately when segments are out of order (so 3126 * fast retransmit can work). 3127 */ 3128 if (th->th_seq == tp->rcv_nxt && 3129 SEGQ_EMPTY(tp) && 3130 (TCPS_HAVEESTABLISHED(tp->t_state) || 3131 tfo_syn)) { 3132 if (DELAY_ACK(tp, tlen) || tfo_syn) 3133 tp->t_flags |= TF_DELACK; 3134 else 3135 tp->t_flags |= TF_ACKNOW; 3136 tp->rcv_nxt += tlen; 3137 if (tlen && 3138 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3139 (tp->t_fbyte_in == 0)) { 3140 tp->t_fbyte_in = ticks; 3141 if (tp->t_fbyte_in == 0) 3142 tp->t_fbyte_in = 1; 3143 if (tp->t_fbyte_out && tp->t_fbyte_in) 3144 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3145 } 3146 thflags = tcp_get_flags(th) & TH_FIN; 3147 TCPSTAT_INC(tcps_rcvpack); 3148 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3149 SOCKBUF_LOCK(&so->so_rcv); 3150 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3151 m_freem(m); 3152 else 3153 sbappendstream_locked(&so->so_rcv, m, 0); 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 3164 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3165 tp->t_flags |= TF_ACKNOW; 3166 } 3167 if ((tp->t_flags & TF_SACK_PERMIT) && 3168 (save_tlen > 0) && 3169 TCPS_HAVEESTABLISHED(tp->t_state)) { 3170 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3171 /* 3172 * DSACK actually handled in the fastpath 3173 * above. 3174 */ 3175 tcp_update_sack_list(tp, save_start, 3176 save_start + save_tlen); 3177 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3178 if ((tp->rcv_numsacks >= 1) && 3179 (tp->sackblks[0].end == save_start)) { 3180 /* 3181 * Partial overlap, recorded at todrop 3182 * above. 3183 */ 3184 tcp_update_sack_list(tp, 3185 tp->sackblks[0].start, 3186 tp->sackblks[0].end); 3187 } else { 3188 tcp_update_dsack_list(tp, save_start, 3189 save_start + save_tlen); 3190 } 3191 } else if (tlen >= save_tlen) { 3192 /* Update of sackblks. */ 3193 tcp_update_dsack_list(tp, save_start, 3194 save_start + save_tlen); 3195 } else if (tlen > 0) { 3196 tcp_update_dsack_list(tp, save_start, 3197 save_start + tlen); 3198 } 3199 } 3200 tcp_handle_wakeup(tp, so); 3201 #if 0 3202 /* 3203 * Note the amount of data that peer has sent into 3204 * our window, in order to estimate the sender's 3205 * buffer size. 3206 * XXX: Unused. 3207 */ 3208 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3209 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3210 else 3211 len = so->so_rcv.sb_hiwat; 3212 #endif 3213 } else { 3214 m_freem(m); 3215 thflags &= ~TH_FIN; 3216 } 3217 3218 /* 3219 * If FIN is received ACK the FIN and let the user know 3220 * that the connection is closing. 3221 */ 3222 if (thflags & TH_FIN) { 3223 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3224 /* The socket upcall is handled by socantrcvmore. */ 3225 socantrcvmore(so); 3226 /* 3227 * If connection is half-synchronized 3228 * (ie NEEDSYN flag on) then delay ACK, 3229 * so it may be piggybacked when SYN is sent. 3230 * Otherwise, since we received a FIN then no 3231 * more input can be expected, send ACK now. 3232 */ 3233 if (tp->t_flags & TF_NEEDSYN) 3234 tp->t_flags |= TF_DELACK; 3235 else 3236 tp->t_flags |= TF_ACKNOW; 3237 tp->rcv_nxt++; 3238 } 3239 switch (tp->t_state) { 3240 /* 3241 * In SYN_RECEIVED and ESTABLISHED STATES 3242 * enter the CLOSE_WAIT state. 3243 */ 3244 case TCPS_SYN_RECEIVED: 3245 tp->t_starttime = ticks; 3246 /* FALLTHROUGH */ 3247 case TCPS_ESTABLISHED: 3248 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3249 break; 3250 3251 /* 3252 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3253 * enter the CLOSING state. 3254 */ 3255 case TCPS_FIN_WAIT_1: 3256 tcp_state_change(tp, TCPS_CLOSING); 3257 break; 3258 3259 /* 3260 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3261 * starting the time-wait timer, turning off the other 3262 * standard timers. 3263 */ 3264 case TCPS_FIN_WAIT_2: 3265 tcp_twstart(tp); 3266 return; 3267 } 3268 } 3269 #ifdef TCPDEBUG 3270 if (so->so_options & SO_DEBUG) 3271 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3272 &tcp_savetcp, 0); 3273 #endif 3274 TCP_PROBE3(debug__input, tp, th, m); 3275 3276 /* 3277 * Return any desired output. 3278 */ 3279 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3280 (void) tcp_output(tp); 3281 3282 check_delack: 3283 INP_WLOCK_ASSERT(tp->t_inpcb); 3284 3285 if (tp->t_flags & TF_DELACK) { 3286 tp->t_flags &= ~TF_DELACK; 3287 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3288 } 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) tcp_output(tp); 3322 INP_WUNLOCK(tp->t_inpcb); 3323 m_freem(m); 3324 return; 3325 3326 dropwithreset: 3327 if (tp != NULL) { 3328 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3329 INP_WUNLOCK(tp->t_inpcb); 3330 } else 3331 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3332 return; 3333 3334 drop: 3335 /* 3336 * Drop space held by incoming segment and return. 3337 */ 3338 #ifdef TCPDEBUG 3339 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3340 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3341 &tcp_savetcp, 0); 3342 #endif 3343 TCP_PROBE3(debug__input, tp, th, m); 3344 if (tp != NULL) { 3345 INP_WUNLOCK(tp->t_inpcb); 3346 } 3347 m_freem(m); 3348 } 3349 3350 /* 3351 * Issue RST and make ACK acceptable to originator of segment. 3352 * The mbuf must still include the original packet header. 3353 * tp may be NULL. 3354 */ 3355 void 3356 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3357 int tlen, int rstreason) 3358 { 3359 #ifdef INET 3360 struct ip *ip; 3361 #endif 3362 #ifdef INET6 3363 struct ip6_hdr *ip6; 3364 #endif 3365 3366 if (tp != NULL) { 3367 INP_LOCK_ASSERT(tp->t_inpcb); 3368 } 3369 3370 /* Don't bother if destination was broadcast/multicast. */ 3371 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3372 goto drop; 3373 #ifdef INET6 3374 if (mtod(m, struct ip *)->ip_v == 6) { 3375 ip6 = mtod(m, struct ip6_hdr *); 3376 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3377 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3378 goto drop; 3379 /* IPv6 anycast check is done at tcp6_input() */ 3380 } 3381 #endif 3382 #if defined(INET) && defined(INET6) 3383 else 3384 #endif 3385 #ifdef INET 3386 { 3387 ip = mtod(m, struct ip *); 3388 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3389 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3390 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3391 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3392 goto drop; 3393 } 3394 #endif 3395 3396 /* Perform bandwidth limiting. */ 3397 if (badport_bandlim(rstreason) < 0) 3398 goto drop; 3399 3400 /* tcp_respond consumes the mbuf chain. */ 3401 if (tcp_get_flags(th) & TH_ACK) { 3402 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3403 th->th_ack, TH_RST); 3404 } else { 3405 if (tcp_get_flags(th) & TH_SYN) 3406 tlen++; 3407 if (tcp_get_flags(th) & TH_FIN) 3408 tlen++; 3409 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3410 (tcp_seq)0, TH_RST|TH_ACK); 3411 } 3412 return; 3413 drop: 3414 m_freem(m); 3415 } 3416 3417 /* 3418 * Parse TCP options and place in tcpopt. 3419 */ 3420 void 3421 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3422 { 3423 int opt, optlen; 3424 3425 to->to_flags = 0; 3426 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3427 opt = cp[0]; 3428 if (opt == TCPOPT_EOL) 3429 break; 3430 if (opt == TCPOPT_NOP) 3431 optlen = 1; 3432 else { 3433 if (cnt < 2) 3434 break; 3435 optlen = cp[1]; 3436 if (optlen < 2 || optlen > cnt) 3437 break; 3438 } 3439 switch (opt) { 3440 case TCPOPT_MAXSEG: 3441 if (optlen != TCPOLEN_MAXSEG) 3442 continue; 3443 if (!(flags & TO_SYN)) 3444 continue; 3445 to->to_flags |= TOF_MSS; 3446 bcopy((char *)cp + 2, 3447 (char *)&to->to_mss, sizeof(to->to_mss)); 3448 to->to_mss = ntohs(to->to_mss); 3449 break; 3450 case TCPOPT_WINDOW: 3451 if (optlen != TCPOLEN_WINDOW) 3452 continue; 3453 if (!(flags & TO_SYN)) 3454 continue; 3455 to->to_flags |= TOF_SCALE; 3456 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3457 break; 3458 case TCPOPT_TIMESTAMP: 3459 if (optlen != TCPOLEN_TIMESTAMP) 3460 continue; 3461 to->to_flags |= TOF_TS; 3462 bcopy((char *)cp + 2, 3463 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3464 to->to_tsval = ntohl(to->to_tsval); 3465 bcopy((char *)cp + 6, 3466 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3467 to->to_tsecr = ntohl(to->to_tsecr); 3468 break; 3469 case TCPOPT_SIGNATURE: 3470 /* 3471 * In order to reply to a host which has set the 3472 * TCP_SIGNATURE option in its initial SYN, we have 3473 * to record the fact that the option was observed 3474 * here for the syncache code to perform the correct 3475 * response. 3476 */ 3477 if (optlen != TCPOLEN_SIGNATURE) 3478 continue; 3479 to->to_flags |= TOF_SIGNATURE; 3480 to->to_signature = cp + 2; 3481 break; 3482 case TCPOPT_SACK_PERMITTED: 3483 if (optlen != TCPOLEN_SACK_PERMITTED) 3484 continue; 3485 if (!(flags & TO_SYN)) 3486 continue; 3487 if (!V_tcp_do_sack) 3488 continue; 3489 to->to_flags |= TOF_SACKPERM; 3490 break; 3491 case TCPOPT_SACK: 3492 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3493 continue; 3494 if (flags & TO_SYN) 3495 continue; 3496 to->to_flags |= TOF_SACK; 3497 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3498 to->to_sacks = cp + 2; 3499 TCPSTAT_INC(tcps_sack_rcv_blocks); 3500 break; 3501 case TCPOPT_FAST_OPEN: 3502 /* 3503 * Cookie length validation is performed by the 3504 * server side cookie checking code or the client 3505 * side cookie cache update code. 3506 */ 3507 if (!(flags & TO_SYN)) 3508 continue; 3509 if (!V_tcp_fastopen_client_enable && 3510 !V_tcp_fastopen_server_enable) 3511 continue; 3512 to->to_flags |= TOF_FASTOPEN; 3513 to->to_tfo_len = optlen - 2; 3514 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3515 break; 3516 default: 3517 continue; 3518 } 3519 } 3520 } 3521 3522 /* 3523 * Pull out of band byte out of a segment so 3524 * it doesn't appear in the user's data queue. 3525 * It is still reflected in the segment length for 3526 * sequencing purposes. 3527 */ 3528 void 3529 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3530 int off) 3531 { 3532 int cnt = off + th->th_urp - 1; 3533 3534 while (cnt >= 0) { 3535 if (m->m_len > cnt) { 3536 char *cp = mtod(m, caddr_t) + cnt; 3537 struct tcpcb *tp = sototcpcb(so); 3538 3539 INP_WLOCK_ASSERT(tp->t_inpcb); 3540 3541 tp->t_iobc = *cp; 3542 tp->t_oobflags |= TCPOOB_HAVEDATA; 3543 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3544 m->m_len--; 3545 if (m->m_flags & M_PKTHDR) 3546 m->m_pkthdr.len--; 3547 return; 3548 } 3549 cnt -= m->m_len; 3550 m = m->m_next; 3551 if (m == NULL) 3552 break; 3553 } 3554 panic("tcp_pulloutofband"); 3555 } 3556 3557 /* 3558 * Collect new round-trip time estimate 3559 * and update averages and current timeout. 3560 */ 3561 void 3562 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3563 { 3564 int delta; 3565 3566 INP_WLOCK_ASSERT(tp->t_inpcb); 3567 3568 TCPSTAT_INC(tcps_rttupdated); 3569 tp->t_rttupdated++; 3570 #ifdef STATS 3571 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3572 imax(0, rtt * 1000 / hz)); 3573 #endif 3574 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3575 /* 3576 * srtt is stored as fixed point with 5 bits after the 3577 * binary point (i.e., scaled by 8). The following magic 3578 * is equivalent to the smoothing algorithm in rfc793 with 3579 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3580 * point). Adjust rtt to origin 0. 3581 */ 3582 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3583 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3584 3585 if ((tp->t_srtt += delta) <= 0) 3586 tp->t_srtt = 1; 3587 3588 /* 3589 * We accumulate a smoothed rtt variance (actually, a 3590 * smoothed mean difference), then set the retransmit 3591 * timer to smoothed rtt + 4 times the smoothed variance. 3592 * rttvar is stored as fixed point with 4 bits after the 3593 * binary point (scaled by 16). The following is 3594 * equivalent to rfc793 smoothing with an alpha of .75 3595 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3596 * rfc793's wired-in beta. 3597 */ 3598 if (delta < 0) 3599 delta = -delta; 3600 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3601 if ((tp->t_rttvar += delta) <= 0) 3602 tp->t_rttvar = 1; 3603 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3604 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3605 } else { 3606 /* 3607 * No rtt measurement yet - use the unsmoothed rtt. 3608 * Set the variance to half the rtt (so our first 3609 * retransmit happens at 3*rtt). 3610 */ 3611 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3612 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3613 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3614 } 3615 tp->t_rtttime = 0; 3616 tp->t_rxtshift = 0; 3617 3618 /* 3619 * the retransmit should happen at rtt + 4 * rttvar. 3620 * Because of the way we do the smoothing, srtt and rttvar 3621 * will each average +1/2 tick of bias. When we compute 3622 * the retransmit timer, we want 1/2 tick of rounding and 3623 * 1 extra tick because of +-1/2 tick uncertainty in the 3624 * firing of the timer. The bias will give us exactly the 3625 * 1.5 tick we need. But, because the bias is 3626 * statistical, we have to test that we don't drop below 3627 * the minimum feasible timer (which is 2 ticks). 3628 */ 3629 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3630 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3631 3632 /* 3633 * We received an ack for a packet that wasn't retransmitted; 3634 * it is probably safe to discard any error indications we've 3635 * received recently. This isn't quite right, but close enough 3636 * for now (a route might have failed after we sent a segment, 3637 * and the return path might not be symmetrical). 3638 */ 3639 tp->t_softerror = 0; 3640 } 3641 3642 /* 3643 * Determine a reasonable value for maxseg size. 3644 * If the route is known, check route for mtu. 3645 * If none, use an mss that can be handled on the outgoing interface 3646 * without forcing IP to fragment. If no route is found, route has no mtu, 3647 * or the destination isn't local, use a default, hopefully conservative 3648 * size (usually 512 or the default IP max size, but no more than the mtu 3649 * of the interface), as we can't discover anything about intervening 3650 * gateways or networks. We also initialize the congestion/slow start 3651 * window to be a single segment if the destination isn't local. 3652 * While looking at the routing entry, we also initialize other path-dependent 3653 * parameters from pre-set or cached values in the routing entry. 3654 * 3655 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3656 * IP options, e.g. IPSEC data, since length of this data may vary, and 3657 * thus it is calculated for every segment separately in tcp_output(). 3658 * 3659 * NOTE that this routine is only called when we process an incoming 3660 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3661 * settings are handled in tcp_mssopt(). 3662 */ 3663 void 3664 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3665 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3666 { 3667 int mss = 0; 3668 uint32_t maxmtu = 0; 3669 struct inpcb *inp = tp->t_inpcb; 3670 struct hc_metrics_lite metrics; 3671 #ifdef INET6 3672 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3673 size_t min_protoh = isipv6 ? 3674 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3675 sizeof (struct tcpiphdr); 3676 #else 3677 size_t min_protoh = sizeof(struct tcpiphdr); 3678 #endif 3679 3680 INP_WLOCK_ASSERT(tp->t_inpcb); 3681 3682 if (tp->t_port) 3683 min_protoh += V_tcp_udp_tunneling_overhead; 3684 if (mtuoffer != -1) { 3685 KASSERT(offer == -1, ("%s: conflict", __func__)); 3686 offer = mtuoffer - min_protoh; 3687 } 3688 3689 /* Initialize. */ 3690 #ifdef INET6 3691 if (isipv6) { 3692 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3693 tp->t_maxseg = V_tcp_v6mssdflt; 3694 } 3695 #endif 3696 #if defined(INET) && defined(INET6) 3697 else 3698 #endif 3699 #ifdef INET 3700 { 3701 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3702 tp->t_maxseg = V_tcp_mssdflt; 3703 } 3704 #endif 3705 3706 /* 3707 * No route to sender, stay with default mss and return. 3708 */ 3709 if (maxmtu == 0) { 3710 /* 3711 * In case we return early we need to initialize metrics 3712 * to a defined state as tcp_hc_get() would do for us 3713 * if there was no cache hit. 3714 */ 3715 if (metricptr != NULL) 3716 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3717 return; 3718 } 3719 3720 /* What have we got? */ 3721 switch (offer) { 3722 case 0: 3723 /* 3724 * Offer == 0 means that there was no MSS on the SYN 3725 * segment, in this case we use tcp_mssdflt as 3726 * already assigned to t_maxseg above. 3727 */ 3728 offer = tp->t_maxseg; 3729 break; 3730 3731 case -1: 3732 /* 3733 * Offer == -1 means that we didn't receive SYN yet. 3734 */ 3735 /* FALLTHROUGH */ 3736 3737 default: 3738 /* 3739 * Prevent DoS attack with too small MSS. Round up 3740 * to at least minmss. 3741 */ 3742 offer = max(offer, V_tcp_minmss); 3743 } 3744 3745 /* 3746 * rmx information is now retrieved from tcp_hostcache. 3747 */ 3748 tcp_hc_get(&inp->inp_inc, &metrics); 3749 if (metricptr != NULL) 3750 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3751 3752 /* 3753 * If there's a discovered mtu in tcp hostcache, use it. 3754 * Else, use the link mtu. 3755 */ 3756 if (metrics.rmx_mtu) 3757 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3758 else { 3759 #ifdef INET6 3760 if (isipv6) { 3761 mss = maxmtu - min_protoh; 3762 if (!V_path_mtu_discovery && 3763 !in6_localaddr(&inp->in6p_faddr)) 3764 mss = min(mss, V_tcp_v6mssdflt); 3765 } 3766 #endif 3767 #if defined(INET) && defined(INET6) 3768 else 3769 #endif 3770 #ifdef INET 3771 { 3772 mss = maxmtu - min_protoh; 3773 if (!V_path_mtu_discovery && 3774 !in_localaddr(inp->inp_faddr)) 3775 mss = min(mss, V_tcp_mssdflt); 3776 } 3777 #endif 3778 /* 3779 * XXX - The above conditional (mss = maxmtu - min_protoh) 3780 * probably violates the TCP spec. 3781 * The problem is that, since we don't know the 3782 * other end's MSS, we are supposed to use a conservative 3783 * default. But, if we do that, then MTU discovery will 3784 * never actually take place, because the conservative 3785 * default is much less than the MTUs typically seen 3786 * on the Internet today. For the moment, we'll sweep 3787 * this under the carpet. 3788 * 3789 * The conservative default might not actually be a problem 3790 * if the only case this occurs is when sending an initial 3791 * SYN with options and data to a host we've never talked 3792 * to before. Then, they will reply with an MSS value which 3793 * will get recorded and the new parameters should get 3794 * recomputed. For Further Study. 3795 */ 3796 } 3797 mss = min(mss, offer); 3798 3799 /* 3800 * Sanity check: make sure that maxseg will be large 3801 * enough to allow some data on segments even if the 3802 * all the option space is used (40bytes). Otherwise 3803 * funny things may happen in tcp_output. 3804 * 3805 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3806 */ 3807 mss = max(mss, 64); 3808 3809 tp->t_maxseg = mss; 3810 } 3811 3812 void 3813 tcp_mss(struct tcpcb *tp, int offer) 3814 { 3815 int mss; 3816 uint32_t bufsize; 3817 struct inpcb *inp; 3818 struct socket *so; 3819 struct hc_metrics_lite metrics; 3820 struct tcp_ifcap cap; 3821 3822 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3823 3824 bzero(&cap, sizeof(cap)); 3825 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3826 3827 mss = tp->t_maxseg; 3828 inp = tp->t_inpcb; 3829 3830 /* 3831 * If there's a pipesize, change the socket buffer to that size, 3832 * don't change if sb_hiwat is different than default (then it 3833 * has been changed on purpose with setsockopt). 3834 * Make the socket buffers an integral number of mss units; 3835 * if the mss is larger than the socket buffer, decrease the mss. 3836 */ 3837 so = inp->inp_socket; 3838 SOCKBUF_LOCK(&so->so_snd); 3839 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3840 bufsize = metrics.rmx_sendpipe; 3841 else 3842 bufsize = so->so_snd.sb_hiwat; 3843 if (bufsize < mss) 3844 mss = bufsize; 3845 else { 3846 bufsize = roundup(bufsize, mss); 3847 if (bufsize > sb_max) 3848 bufsize = sb_max; 3849 if (bufsize > so->so_snd.sb_hiwat) 3850 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3851 } 3852 SOCKBUF_UNLOCK(&so->so_snd); 3853 /* 3854 * Sanity check: make sure that maxseg will be large 3855 * enough to allow some data on segments even if the 3856 * all the option space is used (40bytes). Otherwise 3857 * funny things may happen in tcp_output. 3858 * 3859 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3860 */ 3861 tp->t_maxseg = max(mss, 64); 3862 3863 SOCKBUF_LOCK(&so->so_rcv); 3864 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3865 bufsize = metrics.rmx_recvpipe; 3866 else 3867 bufsize = so->so_rcv.sb_hiwat; 3868 if (bufsize > mss) { 3869 bufsize = roundup(bufsize, mss); 3870 if (bufsize > sb_max) 3871 bufsize = sb_max; 3872 if (bufsize > so->so_rcv.sb_hiwat) 3873 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3874 } 3875 SOCKBUF_UNLOCK(&so->so_rcv); 3876 3877 /* Check the interface for TSO capabilities. */ 3878 if (cap.ifcap & CSUM_TSO) { 3879 tp->t_flags |= TF_TSO; 3880 tp->t_tsomax = cap.tsomax; 3881 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3882 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3883 } 3884 } 3885 3886 /* 3887 * Determine the MSS option to send on an outgoing SYN. 3888 */ 3889 int 3890 tcp_mssopt(struct in_conninfo *inc) 3891 { 3892 int mss = 0; 3893 uint32_t thcmtu = 0; 3894 uint32_t maxmtu = 0; 3895 size_t min_protoh; 3896 3897 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3898 3899 #ifdef INET6 3900 if (inc->inc_flags & INC_ISIPV6) { 3901 mss = V_tcp_v6mssdflt; 3902 maxmtu = tcp_maxmtu6(inc, NULL); 3903 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3904 } 3905 #endif 3906 #if defined(INET) && defined(INET6) 3907 else 3908 #endif 3909 #ifdef INET 3910 { 3911 mss = V_tcp_mssdflt; 3912 maxmtu = tcp_maxmtu(inc, NULL); 3913 min_protoh = sizeof(struct tcpiphdr); 3914 } 3915 #endif 3916 #if defined(INET6) || defined(INET) 3917 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3918 #endif 3919 3920 if (maxmtu && thcmtu) 3921 mss = min(maxmtu, thcmtu) - min_protoh; 3922 else if (maxmtu || thcmtu) 3923 mss = max(maxmtu, thcmtu) - min_protoh; 3924 3925 return (mss); 3926 } 3927 3928 void 3929 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3930 { 3931 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3932 int maxseg = tcp_maxseg(tp); 3933 3934 INP_WLOCK_ASSERT(tp->t_inpcb); 3935 3936 /* 3937 * Compute the amount of data that this ACK is indicating 3938 * (del_data) and an estimate of how many bytes are in the 3939 * network. 3940 */ 3941 if (((tp->t_flags & TF_SACK_PERMIT) && 3942 (to->to_flags & TOF_SACK)) || 3943 (IN_CONGRECOVERY(tp->t_flags) && 3944 !IN_FASTRECOVERY(tp->t_flags))) { 3945 del_data = tp->sackhint.delivered_data; 3946 if (V_tcp_do_newsack) 3947 pipe = tcp_compute_pipe(tp); 3948 else 3949 pipe = (tp->snd_nxt - tp->snd_fack) + 3950 tp->sackhint.sack_bytes_rexmit; 3951 } else { 3952 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 3953 tp->snd_recover - tp->snd_una)) 3954 del_data = maxseg; 3955 pipe = imax(0, tp->snd_max - tp->snd_una - 3956 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 3957 } 3958 tp->sackhint.prr_delivered += del_data; 3959 /* 3960 * Proportional Rate Reduction 3961 */ 3962 if (pipe >= tp->snd_ssthresh) { 3963 if (tp->sackhint.recover_fs == 0) 3964 tp->sackhint.recover_fs = 3965 imax(1, tp->snd_nxt - tp->snd_una); 3966 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3967 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3968 tp->sackhint.prr_out; 3969 } else { 3970 if (V_tcp_do_prr_conservative || (del_data == 0)) 3971 limit = tp->sackhint.prr_delivered - 3972 tp->sackhint.prr_out; 3973 else 3974 limit = imax(tp->sackhint.prr_delivered - 3975 tp->sackhint.prr_out, del_data) + 3976 maxseg; 3977 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 3978 } 3979 snd_cnt = imax(snd_cnt, 0) / maxseg; 3980 /* 3981 * Send snd_cnt new data into the network in response to this ack. 3982 * If there is going to be a SACK retransmission, adjust snd_cwnd 3983 * accordingly. 3984 */ 3985 if (IN_FASTRECOVERY(tp->t_flags)) { 3986 if ((tp->t_flags & TF_SACK_PERMIT) && 3987 (to->to_flags & TOF_SACK)) { 3988 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 3989 tp->sackhint.sack_bytes_rexmit + 3990 (snd_cnt * maxseg); 3991 } else { 3992 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 3993 (snd_cnt * maxseg); 3994 } 3995 } else if (IN_CONGRECOVERY(tp->t_flags)) 3996 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 3997 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 3998 } 3999 4000 /* 4001 * On a partial ack arrives, force the retransmission of the 4002 * next unacknowledged segment. Do not clear tp->t_dupacks. 4003 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4004 * be started again. 4005 */ 4006 void 4007 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4008 { 4009 tcp_seq onxt = tp->snd_nxt; 4010 uint32_t ocwnd = tp->snd_cwnd; 4011 u_int maxseg = tcp_maxseg(tp); 4012 4013 INP_WLOCK_ASSERT(tp->t_inpcb); 4014 4015 tcp_timer_activate(tp, TT_REXMT, 0); 4016 tp->t_rtttime = 0; 4017 tp->snd_nxt = th->th_ack; 4018 /* 4019 * Set snd_cwnd to one segment beyond acknowledged offset. 4020 * (tp->snd_una has not yet been updated when this function is called.) 4021 */ 4022 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4023 tp->t_flags |= TF_ACKNOW; 4024 (void) tcp_output(tp); 4025 tp->snd_cwnd = ocwnd; 4026 if (SEQ_GT(onxt, tp->snd_nxt)) 4027 tp->snd_nxt = onxt; 4028 /* 4029 * Partial window deflation. Relies on fact that tp->snd_una 4030 * not updated yet. 4031 */ 4032 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4033 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4034 else 4035 tp->snd_cwnd = 0; 4036 tp->snd_cwnd += maxseg; 4037 } 4038 4039 int 4040 tcp_compute_pipe(struct tcpcb *tp) 4041 { 4042 return (tp->snd_max - tp->snd_una + 4043 tp->sackhint.sack_bytes_rexmit - 4044 tp->sackhint.sacked_bytes); 4045 } 4046 4047 uint32_t 4048 tcp_compute_initwnd(uint32_t maxseg) 4049 { 4050 /* 4051 * Calculate the Initial Window, also used as Restart Window 4052 * 4053 * RFC5681 Section 3.1 specifies the default conservative values. 4054 * RFC3390 specifies slightly more aggressive values. 4055 * RFC6928 increases it to ten segments. 4056 * Support for user specified value for initial flight size. 4057 */ 4058 if (V_tcp_initcwnd_segments) 4059 return min(V_tcp_initcwnd_segments * maxseg, 4060 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4061 else if (V_tcp_do_rfc3390) 4062 return min(4 * maxseg, max(2 * maxseg, 4380)); 4063 else { 4064 /* Per RFC5681 Section 3.1 */ 4065 if (maxseg > 2190) 4066 return (2 * maxseg); 4067 else if (maxseg > 1095) 4068 return (3 * maxseg); 4069 else 4070 return (4 * maxseg); 4071 } 4072 } 4073