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 #include <netinet/tcp_debug.h> 121 #ifdef TCP_OFFLOAD 122 #include <netinet/tcp_offload.h> 123 #endif 124 #include <netinet/tcp_ecn.h> 125 #include <netinet/udp.h> 126 127 #include <netipsec/ipsec_support.h> 128 129 #include <machine/in_cksum.h> 130 131 #include <security/mac/mac_framework.h> 132 133 const int tcprexmtthresh = 3; 134 135 VNET_DEFINE(int, tcp_log_in_vain) = 0; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 137 &VNET_NAME(tcp_log_in_vain), 0, 138 "Log all incoming TCP segments to closed ports"); 139 140 VNET_DEFINE(int, blackhole) = 0; 141 #define V_blackhole VNET(blackhole) 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 143 &VNET_NAME(blackhole), 0, 144 "Do not send RST on segments to closed ports"); 145 146 VNET_DEFINE(bool, blackhole_local) = false; 147 #define V_blackhole_local VNET(blackhole_local) 148 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 149 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 150 "Enforce net.inet.tcp.blackhole for locally originated packets"); 151 152 VNET_DEFINE(int, tcp_delack_enabled) = 1; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 154 &VNET_NAME(tcp_delack_enabled), 0, 155 "Delay ACK to try and piggyback it onto a data packet"); 156 157 VNET_DEFINE(int, drop_synfin) = 0; 158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 159 &VNET_NAME(drop_synfin), 0, 160 "Drop TCP packets with SYN+FIN set"); 161 162 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_prr_conservative), 0, 165 "Do conservative Proportional Rate Reduction"); 166 167 VNET_DEFINE(int, tcp_do_prr) = 1; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_prr), 1, 170 "Enable Proportional Rate Reduction per RFC 6937"); 171 172 VNET_DEFINE(int, tcp_do_lrd) = 0; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_do_lrd), 1, 175 "Perform Lost Retransmission Detection"); 176 177 VNET_DEFINE(int, tcp_do_newcwv) = 0; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 179 &VNET_NAME(tcp_do_newcwv), 0, 180 "Enable New Congestion Window Validation per RFC7661"); 181 182 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 184 &VNET_NAME(tcp_do_rfc3042), 0, 185 "Enable RFC 3042 (Limited Transmit)"); 186 187 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 189 &VNET_NAME(tcp_do_rfc3390), 0, 190 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 191 192 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 194 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 195 "Slow-start flight size (initial congestion window) in number of segments"); 196 197 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_do_rfc3465), 0, 200 "Enable RFC 3465 (Appropriate Byte Counting)"); 201 202 VNET_DEFINE(int, tcp_abc_l_var) = 2; 203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 204 &VNET_NAME(tcp_abc_l_var), 2, 205 "Cap the max cwnd increment during slow-start to this number of segments"); 206 207 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 208 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 209 "TCP ECN"); 210 211 VNET_DEFINE(int, tcp_do_ecn) = 2; 212 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(tcp_do_ecn), 0, 214 "TCP ECN support"); 215 216 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 217 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(tcp_ecn_maxretries), 0, 219 "Max retries before giving up on ECN"); 220 221 VNET_DEFINE(int, tcp_insecure_syn) = 0; 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_insecure_syn), 0, 224 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 225 226 VNET_DEFINE(int, tcp_insecure_rst) = 0; 227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_insecure_rst), 0, 229 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 230 231 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 232 #define V_tcp_recvspace VNET(tcp_recvspace) 233 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 234 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 235 236 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 238 &VNET_NAME(tcp_do_autorcvbuf), 0, 239 "Enable automatic receive buffer sizing"); 240 241 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 243 &VNET_NAME(tcp_autorcvbuf_max), 0, 244 "Max size of automatic receive buffer"); 245 246 VNET_DEFINE(struct inpcbinfo, tcbinfo); 247 248 /* 249 * TCP statistics are stored in an array of counter(9)s, which size matches 250 * size of struct tcpstat. TCP running connection count is a regular array. 251 */ 252 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 253 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 254 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 255 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 256 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 257 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 258 "TCP connection counts by TCP state"); 259 260 /* 261 * Kernel module interface for updating tcpstat. The first argument is an index 262 * into tcpstat treated as an array. 263 */ 264 void 265 kmod_tcpstat_add(int statnum, int val) 266 { 267 268 counter_u64_add(VNET(tcpstat)[statnum], val); 269 } 270 271 /* 272 * Make sure that we only start a SACK loss recovery when 273 * receiving a duplicate ACK with a SACK block, and also 274 * complete SACK loss recovery in case the other end 275 * reneges. 276 */ 277 static bool inline 278 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) 279 { 280 return ((tp->t_flags & TF_SACK_PERMIT) && 281 ((to->to_flags & TOF_SACK) || 282 (!TAILQ_EMPTY(&tp->snd_holes)))); 283 } 284 285 #ifdef TCP_HHOOK 286 /* 287 * Wrapper for the TCP established input helper hook. 288 */ 289 void 290 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 291 { 292 struct tcp_hhook_data hhook_data; 293 294 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 295 hhook_data.tp = tp; 296 hhook_data.th = th; 297 hhook_data.to = to; 298 299 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 300 tp->osd); 301 } 302 } 303 #endif 304 305 /* 306 * CC wrapper hook functions 307 */ 308 void 309 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 310 uint16_t type) 311 { 312 #ifdef STATS 313 int32_t gput; 314 #endif 315 316 INP_WLOCK_ASSERT(tp->t_inpcb); 317 318 tp->ccv->nsegs = nsegs; 319 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 320 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 321 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 322 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 323 tp->ccv->flags |= CCF_CWND_LIMITED; 324 else 325 tp->ccv->flags &= ~CCF_CWND_LIMITED; 326 327 if (type == CC_ACK) { 328 #ifdef STATS 329 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 330 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 331 if (!IN_RECOVERY(tp->t_flags)) 332 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 333 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 334 if ((tp->t_flags & TF_GPUTINPROG) && 335 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 336 /* 337 * Compute goodput in bits per millisecond. 338 */ 339 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 340 max(1, tcp_ts_getticks() - tp->gput_ts); 341 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 342 gput); 343 /* 344 * XXXLAS: This is a temporary hack, and should be 345 * chained off VOI_TCP_GPUT when stats(9) grows an API 346 * to deal with chained VOIs. 347 */ 348 if (tp->t_stats_gput_prev > 0) 349 stats_voi_update_abs_s32(tp->t_stats, 350 VOI_TCP_GPUT_ND, 351 ((gput - tp->t_stats_gput_prev) * 100) / 352 tp->t_stats_gput_prev); 353 tp->t_flags &= ~TF_GPUTINPROG; 354 tp->t_stats_gput_prev = gput; 355 } 356 #endif /* STATS */ 357 if (tp->snd_cwnd > tp->snd_ssthresh) { 358 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 359 if (tp->t_bytes_acked >= tp->snd_cwnd) { 360 tp->t_bytes_acked -= tp->snd_cwnd; 361 tp->ccv->flags |= CCF_ABC_SENTAWND; 362 } 363 } else { 364 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 365 tp->t_bytes_acked = 0; 366 } 367 } 368 369 if (CC_ALGO(tp)->ack_received != NULL) { 370 /* XXXLAS: Find a way to live without this */ 371 tp->ccv->curack = th->th_ack; 372 CC_ALGO(tp)->ack_received(tp->ccv, type); 373 } 374 #ifdef STATS 375 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 376 #endif 377 } 378 379 void 380 cc_conn_init(struct tcpcb *tp) 381 { 382 struct hc_metrics_lite metrics; 383 struct inpcb *inp = tp->t_inpcb; 384 u_int maxseg; 385 int rtt; 386 387 INP_WLOCK_ASSERT(tp->t_inpcb); 388 389 tcp_hc_get(&inp->inp_inc, &metrics); 390 maxseg = tcp_maxseg(tp); 391 392 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 393 tp->t_srtt = rtt; 394 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 395 TCPSTAT_INC(tcps_usedrtt); 396 if (metrics.rmx_rttvar) { 397 tp->t_rttvar = metrics.rmx_rttvar; 398 TCPSTAT_INC(tcps_usedrttvar); 399 } else { 400 /* default variation is +- 1 rtt */ 401 tp->t_rttvar = 402 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 403 } 404 TCPT_RANGESET(tp->t_rxtcur, 405 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 406 tp->t_rttmin, TCPTV_REXMTMAX); 407 } 408 if (metrics.rmx_ssthresh) { 409 /* 410 * There's some sort of gateway or interface 411 * buffer limit on the path. Use this to set 412 * the slow start threshold, but set the 413 * threshold to no less than 2*mss. 414 */ 415 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 416 TCPSTAT_INC(tcps_usedssthresh); 417 } 418 419 /* 420 * Set the initial slow-start flight size. 421 * 422 * If a SYN or SYN/ACK was lost and retransmitted, we have to 423 * reduce the initial CWND to one segment as congestion is likely 424 * requiring us to be cautious. 425 */ 426 if (tp->snd_cwnd == 1) 427 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 428 else 429 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 430 431 if (CC_ALGO(tp)->conn_init != NULL) 432 CC_ALGO(tp)->conn_init(tp->ccv); 433 } 434 435 void inline 436 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 437 { 438 INP_WLOCK_ASSERT(tp->t_inpcb); 439 440 #ifdef STATS 441 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 442 #endif 443 444 switch(type) { 445 case CC_NDUPACK: 446 if (!IN_FASTRECOVERY(tp->t_flags)) { 447 tp->snd_recover = tp->snd_max; 448 if (tp->t_flags2 & TF2_ECN_PERMIT) 449 tp->t_flags2 |= TF2_ECN_SND_CWR; 450 } 451 break; 452 case CC_ECN: 453 if (!IN_CONGRECOVERY(tp->t_flags) || 454 /* 455 * Allow ECN reaction on ACK to CWR, if 456 * that data segment was also CE marked. 457 */ 458 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 459 EXIT_CONGRECOVERY(tp->t_flags); 460 TCPSTAT_INC(tcps_ecn_rcwnd); 461 tp->snd_recover = tp->snd_max + 1; 462 if (tp->t_flags2 & TF2_ECN_PERMIT) 463 tp->t_flags2 |= TF2_ECN_SND_CWR; 464 } 465 break; 466 case CC_RTO: 467 tp->t_dupacks = 0; 468 tp->t_bytes_acked = 0; 469 EXIT_RECOVERY(tp->t_flags); 470 if (tp->t_flags2 & TF2_ECN_PERMIT) 471 tp->t_flags2 |= TF2_ECN_SND_CWR; 472 break; 473 case CC_RTO_ERR: 474 TCPSTAT_INC(tcps_sndrexmitbad); 475 /* RTO was unnecessary, so reset everything. */ 476 tp->snd_cwnd = tp->snd_cwnd_prev; 477 tp->snd_ssthresh = tp->snd_ssthresh_prev; 478 tp->snd_recover = tp->snd_recover_prev; 479 if (tp->t_flags & TF_WASFRECOVERY) 480 ENTER_FASTRECOVERY(tp->t_flags); 481 if (tp->t_flags & TF_WASCRECOVERY) 482 ENTER_CONGRECOVERY(tp->t_flags); 483 tp->snd_nxt = tp->snd_max; 484 tp->t_flags &= ~TF_PREVVALID; 485 tp->t_badrxtwin = 0; 486 break; 487 } 488 489 if (CC_ALGO(tp)->cong_signal != NULL) { 490 if (th != NULL) 491 tp->ccv->curack = th->th_ack; 492 CC_ALGO(tp)->cong_signal(tp->ccv, type); 493 } 494 } 495 496 void inline 497 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 498 { 499 INP_WLOCK_ASSERT(tp->t_inpcb); 500 501 /* XXXLAS: KASSERT that we're in recovery? */ 502 503 if (CC_ALGO(tp)->post_recovery != NULL) { 504 tp->ccv->curack = th->th_ack; 505 CC_ALGO(tp)->post_recovery(tp->ccv); 506 } 507 /* XXXLAS: EXIT_RECOVERY ? */ 508 tp->t_bytes_acked = 0; 509 tp->sackhint.delivered_data = 0; 510 tp->sackhint.prr_out = 0; 511 } 512 513 /* 514 * Indicate whether this ack should be delayed. We can delay the ack if 515 * following conditions are met: 516 * - There is no delayed ack timer in progress. 517 * - Our last ack wasn't a 0-sized window. We never want to delay 518 * the ack that opens up a 0-sized window. 519 * - LRO wasn't used for this segment. We make sure by checking that the 520 * segment size is not larger than the MSS. 521 */ 522 #define DELAY_ACK(tp, tlen) \ 523 ((!tcp_timer_active(tp, TT_DELACK) && \ 524 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 525 (tlen <= tp->t_maxseg) && \ 526 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 527 528 void inline 529 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 530 { 531 INP_WLOCK_ASSERT(tp->t_inpcb); 532 533 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 534 switch (iptos & IPTOS_ECN_MASK) { 535 case IPTOS_ECN_CE: 536 tp->ccv->flags |= CCF_IPHDR_CE; 537 break; 538 case IPTOS_ECN_ECT0: 539 /* FALLTHROUGH */ 540 case IPTOS_ECN_ECT1: 541 /* FALLTHROUGH */ 542 case IPTOS_ECN_NOTECT: 543 tp->ccv->flags &= ~CCF_IPHDR_CE; 544 break; 545 } 546 547 if (flags & TH_CWR) 548 tp->ccv->flags |= CCF_TCPHDR_CWR; 549 else 550 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 551 552 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 553 554 if (tp->ccv->flags & CCF_ACKNOW) { 555 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 556 tp->t_flags |= TF_ACKNOW; 557 } 558 } 559 } 560 561 void inline 562 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 563 { 564 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 565 } 566 567 /* 568 * TCP input handling is split into multiple parts: 569 * tcp6_input is a thin wrapper around tcp_input for the extended 570 * ip6_protox[] call format in ip6_input 571 * tcp_input handles primary segment validation, inpcb lookup and 572 * SYN processing on listen sockets 573 * tcp_do_segment processes the ACK and text of the segment for 574 * establishing, established and closing connections 575 */ 576 #ifdef INET6 577 int 578 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 579 { 580 struct mbuf *m; 581 struct in6_ifaddr *ia6; 582 struct ip6_hdr *ip6; 583 584 m = *mp; 585 if (m->m_len < *offp + sizeof(struct tcphdr)) { 586 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 587 if (m == NULL) { 588 *mp = m; 589 TCPSTAT_INC(tcps_rcvshort); 590 return (IPPROTO_DONE); 591 } 592 } 593 594 /* 595 * draft-itojun-ipv6-tcp-to-anycast 596 * better place to put this in? 597 */ 598 ip6 = mtod(m, struct ip6_hdr *); 599 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 600 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 601 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 602 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 603 *mp = NULL; 604 return (IPPROTO_DONE); 605 } 606 607 *mp = m; 608 return (tcp_input_with_port(mp, offp, proto, port)); 609 } 610 611 int 612 tcp6_input(struct mbuf **mp, int *offp, int proto) 613 { 614 615 return(tcp6_input_with_port(mp, offp, proto, 0)); 616 } 617 #endif /* INET6 */ 618 619 int 620 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 621 { 622 struct mbuf *m = *mp; 623 struct tcphdr *th = NULL; 624 struct ip *ip = NULL; 625 struct inpcb *inp = NULL; 626 struct tcpcb *tp = NULL; 627 struct socket *so = NULL; 628 u_char *optp = NULL; 629 int off0; 630 int optlen = 0; 631 #ifdef INET 632 int len; 633 uint8_t ipttl; 634 #endif 635 int tlen = 0, off; 636 int drop_hdrlen; 637 int thflags; 638 int rstreason = 0; /* For badport_bandlim accounting purposes */ 639 int lookupflag; 640 uint8_t iptos; 641 struct m_tag *fwd_tag = NULL; 642 #ifdef INET6 643 struct ip6_hdr *ip6 = NULL; 644 int isipv6; 645 #else 646 const void *ip6 = NULL; 647 #endif /* INET6 */ 648 struct tcpopt to; /* options in this segment */ 649 char *s = NULL; /* address and port logging */ 650 #ifdef TCPDEBUG 651 /* 652 * The size of tcp_saveipgen must be the size of the max ip header, 653 * now IPv6. 654 */ 655 u_char tcp_saveipgen[IP6_HDR_LEN]; 656 struct tcphdr tcp_savetcp; 657 short ostate = 0; 658 #endif 659 660 NET_EPOCH_ASSERT(); 661 662 #ifdef INET6 663 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 664 #endif 665 666 off0 = *offp; 667 m = *mp; 668 *mp = NULL; 669 to.to_flags = 0; 670 TCPSTAT_INC(tcps_rcvtotal); 671 672 #ifdef INET6 673 if (isipv6) { 674 ip6 = mtod(m, struct ip6_hdr *); 675 th = (struct tcphdr *)((caddr_t)ip6 + off0); 676 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 677 if (port) 678 goto skip6_csum; 679 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 680 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 681 th->th_sum = m->m_pkthdr.csum_data; 682 else 683 th->th_sum = in6_cksum_pseudo(ip6, tlen, 684 IPPROTO_TCP, m->m_pkthdr.csum_data); 685 th->th_sum ^= 0xffff; 686 } else 687 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 688 if (th->th_sum) { 689 TCPSTAT_INC(tcps_rcvbadsum); 690 goto drop; 691 } 692 skip6_csum: 693 /* 694 * Be proactive about unspecified IPv6 address in source. 695 * As we use all-zero to indicate unbounded/unconnected pcb, 696 * unspecified IPv6 address can be used to confuse us. 697 * 698 * Note that packets with unspecified IPv6 destination is 699 * already dropped in ip6_input. 700 */ 701 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 702 /* XXX stat */ 703 goto drop; 704 } 705 iptos = IPV6_TRAFFIC_CLASS(ip6); 706 } 707 #endif 708 #if defined(INET) && defined(INET6) 709 else 710 #endif 711 #ifdef INET 712 { 713 /* 714 * Get IP and TCP header together in first mbuf. 715 * Note: IP leaves IP header in first mbuf. 716 */ 717 if (off0 > sizeof (struct ip)) { 718 ip_stripoptions(m); 719 off0 = sizeof(struct ip); 720 } 721 if (m->m_len < sizeof (struct tcpiphdr)) { 722 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 723 == NULL) { 724 TCPSTAT_INC(tcps_rcvshort); 725 return (IPPROTO_DONE); 726 } 727 } 728 ip = mtod(m, struct ip *); 729 th = (struct tcphdr *)((caddr_t)ip + off0); 730 tlen = ntohs(ip->ip_len) - off0; 731 732 iptos = ip->ip_tos; 733 if (port) 734 goto skip_csum; 735 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 736 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 737 th->th_sum = m->m_pkthdr.csum_data; 738 else 739 th->th_sum = in_pseudo(ip->ip_src.s_addr, 740 ip->ip_dst.s_addr, 741 htonl(m->m_pkthdr.csum_data + tlen + 742 IPPROTO_TCP)); 743 th->th_sum ^= 0xffff; 744 } else { 745 struct ipovly *ipov = (struct ipovly *)ip; 746 747 /* 748 * Checksum extended TCP header and data. 749 */ 750 len = off0 + tlen; 751 ipttl = ip->ip_ttl; 752 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 753 ipov->ih_len = htons(tlen); 754 th->th_sum = in_cksum(m, len); 755 /* Reset length for SDT probes. */ 756 ip->ip_len = htons(len); 757 /* Reset TOS bits */ 758 ip->ip_tos = iptos; 759 /* Re-initialization for later version check */ 760 ip->ip_ttl = ipttl; 761 ip->ip_v = IPVERSION; 762 ip->ip_hl = off0 >> 2; 763 } 764 skip_csum: 765 if (th->th_sum && (port == 0)) { 766 TCPSTAT_INC(tcps_rcvbadsum); 767 goto drop; 768 } 769 } 770 #endif /* INET */ 771 772 /* 773 * Check that TCP offset makes sense, 774 * pull out TCP options and adjust length. XXX 775 */ 776 off = th->th_off << 2; 777 if (off < sizeof (struct tcphdr) || off > tlen) { 778 TCPSTAT_INC(tcps_rcvbadoff); 779 goto drop; 780 } 781 tlen -= off; /* tlen is used instead of ti->ti_len */ 782 if (off > sizeof (struct tcphdr)) { 783 #ifdef INET6 784 if (isipv6) { 785 if (m->m_len < off0 + off) { 786 m = m_pullup(m, off0 + off); 787 if (m == NULL) { 788 TCPSTAT_INC(tcps_rcvshort); 789 return (IPPROTO_DONE); 790 } 791 } 792 ip6 = mtod(m, struct ip6_hdr *); 793 th = (struct tcphdr *)((caddr_t)ip6 + off0); 794 } 795 #endif 796 #if defined(INET) && defined(INET6) 797 else 798 #endif 799 #ifdef INET 800 { 801 if (m->m_len < sizeof(struct ip) + off) { 802 if ((m = m_pullup(m, sizeof (struct ip) + off)) 803 == NULL) { 804 TCPSTAT_INC(tcps_rcvshort); 805 return (IPPROTO_DONE); 806 } 807 ip = mtod(m, struct ip *); 808 th = (struct tcphdr *)((caddr_t)ip + off0); 809 } 810 } 811 #endif 812 optlen = off - sizeof (struct tcphdr); 813 optp = (u_char *)(th + 1); 814 } 815 thflags = tcp_get_flags(th); 816 817 /* 818 * Convert TCP protocol specific fields to host format. 819 */ 820 tcp_fields_to_host(th); 821 822 /* 823 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 824 */ 825 drop_hdrlen = off0 + off; 826 827 /* 828 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 829 */ 830 if ( 831 #ifdef INET6 832 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 833 #ifdef INET 834 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 835 #endif 836 #endif 837 #if defined(INET) && !defined(INET6) 838 (m->m_flags & M_IP_NEXTHOP) 839 #endif 840 ) 841 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 842 843 /* 844 * For initial SYN packets we don't need write lock on matching 845 * PCB, be it a listening one or a synchronized one. The packet 846 * shall not modify its state. 847 */ 848 lookupflag = INPLOOKUP_WILDCARD | 849 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 850 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); 851 findpcb: 852 #ifdef INET6 853 if (isipv6 && fwd_tag != NULL) { 854 struct sockaddr_in6 *next_hop6; 855 856 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 857 /* 858 * Transparently forwarded. Pretend to be the destination. 859 * Already got one like this? 860 */ 861 inp = in6_pcblookup_mbuf(&V_tcbinfo, 862 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 863 lookupflag, m->m_pkthdr.rcvif, m); 864 if (!inp) { 865 /* 866 * It's new. Try to find the ambushing socket. 867 * Because we've rewritten the destination address, 868 * any hardware-generated hash is ignored. 869 */ 870 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 871 th->th_sport, &next_hop6->sin6_addr, 872 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 873 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 874 } 875 } else if (isipv6) { 876 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 877 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 878 m->m_pkthdr.rcvif, m); 879 } 880 #endif /* INET6 */ 881 #if defined(INET6) && defined(INET) 882 else 883 #endif 884 #ifdef INET 885 if (fwd_tag != NULL) { 886 struct sockaddr_in *next_hop; 887 888 next_hop = (struct sockaddr_in *)(fwd_tag+1); 889 /* 890 * Transparently forwarded. Pretend to be the destination. 891 * already got one like this? 892 */ 893 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 894 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); 895 if (!inp) { 896 /* 897 * It's new. Try to find the ambushing socket. 898 * Because we've rewritten the destination address, 899 * any hardware-generated hash is ignored. 900 */ 901 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 902 th->th_sport, next_hop->sin_addr, 903 next_hop->sin_port ? ntohs(next_hop->sin_port) : 904 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 905 } 906 } else 907 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 908 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 909 m->m_pkthdr.rcvif, m); 910 #endif /* INET */ 911 912 /* 913 * If the INPCB does not exist then all data in the incoming 914 * segment is discarded and an appropriate RST is sent back. 915 * XXX MRT Send RST using which routing table? 916 */ 917 if (inp == NULL) { 918 if (rstreason != 0) { 919 /* We came here after second (safety) lookup. */ 920 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); 921 goto dropwithreset; 922 } 923 /* 924 * Log communication attempts to ports that are not 925 * in use. 926 */ 927 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 928 V_tcp_log_in_vain == 2) { 929 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 930 log(LOG_INFO, "%s; %s: Connection attempt " 931 "to closed port\n", s, __func__); 932 } 933 /* 934 * When blackholing do not respond with a RST but 935 * completely ignore the segment and drop it. 936 */ 937 if (((V_blackhole == 1 && (thflags & TH_SYN)) || 938 V_blackhole == 2) && (V_blackhole_local || ( 939 #ifdef INET6 940 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 941 #endif 942 #ifdef INET 943 !in_localip(ip->ip_src) 944 #else 945 true 946 #endif 947 ))) 948 goto dropunlock; 949 950 rstreason = BANDLIM_RST_CLOSEDPORT; 951 goto dropwithreset; 952 } 953 INP_LOCK_ASSERT(inp); 954 /* 955 * While waiting for inp lock during the lookup, another thread 956 * can have dropped the inpcb, in which case we need to loop back 957 * and try to find a new inpcb to deliver to. 958 */ 959 if (inp->inp_flags & INP_DROPPED) { 960 INP_UNLOCK(inp); 961 inp = NULL; 962 goto findpcb; 963 } 964 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 965 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 966 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) { 967 inp->inp_flowid = m->m_pkthdr.flowid; 968 inp->inp_flowtype = M_HASHTYPE_GET(m); 969 } 970 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 971 #ifdef INET6 972 if (isipv6 && IPSEC_ENABLED(ipv6) && 973 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 974 goto dropunlock; 975 } 976 #ifdef INET 977 else 978 #endif 979 #endif /* INET6 */ 980 #ifdef INET 981 if (IPSEC_ENABLED(ipv4) && 982 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 983 goto dropunlock; 984 } 985 #endif /* INET */ 986 #endif /* IPSEC */ 987 988 /* 989 * Check the minimum TTL for socket. 990 */ 991 if (inp->inp_ip_minttl != 0) { 992 #ifdef INET6 993 if (isipv6) { 994 if (inp->inp_ip_minttl > ip6->ip6_hlim) 995 goto dropunlock; 996 } else 997 #endif 998 if (inp->inp_ip_minttl > ip->ip_ttl) 999 goto dropunlock; 1000 } 1001 1002 /* 1003 * A previous connection in TIMEWAIT state is supposed to catch stray 1004 * or duplicate segments arriving late. If this segment was a 1005 * legitimate new connection attempt, the old INPCB gets removed and 1006 * we can try again to find a listening socket. 1007 */ 1008 if (inp->inp_flags & INP_TIMEWAIT) { 1009 tcp_dooptions(&to, optp, optlen, 1010 (thflags & TH_SYN) ? TO_SYN : 0); 1011 /* 1012 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1013 */ 1014 if (tcp_twcheck(inp, &to, th, m, tlen)) 1015 goto findpcb; 1016 return (IPPROTO_DONE); 1017 } 1018 /* 1019 * The TCPCB may no longer exist if the connection is winding 1020 * down or it is in the CLOSED state. Either way we drop the 1021 * segment and send an appropriate response. 1022 */ 1023 tp = intotcpcb(inp); 1024 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1025 rstreason = BANDLIM_RST_CLOSEDPORT; 1026 goto dropwithreset; 1027 } 1028 1029 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1030 rstreason = BANDLIM_RST_CLOSEDPORT; 1031 goto dropwithreset; 1032 } 1033 1034 #ifdef TCP_OFFLOAD 1035 if (tp->t_flags & TF_TOE) { 1036 tcp_offload_input(tp, m); 1037 m = NULL; /* consumed by the TOE driver */ 1038 goto dropunlock; 1039 } 1040 #endif 1041 1042 #ifdef MAC 1043 if (mac_inpcb_check_deliver(inp, m)) 1044 goto dropunlock; 1045 #endif 1046 so = inp->inp_socket; 1047 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1048 #ifdef TCPDEBUG 1049 if (so->so_options & SO_DEBUG) { 1050 ostate = tp->t_state; 1051 #ifdef INET6 1052 if (isipv6) { 1053 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1054 } else 1055 #endif 1056 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1057 tcp_savetcp = *th; 1058 } 1059 #endif /* TCPDEBUG */ 1060 /* 1061 * When the socket is accepting connections (the INPCB is in LISTEN 1062 * state) we look into the SYN cache if this is a new connection 1063 * attempt or the completion of a previous one. 1064 */ 1065 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1066 ("%s: so accepting but tp %p not listening", __func__, tp)); 1067 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1068 struct in_conninfo inc; 1069 1070 bzero(&inc, sizeof(inc)); 1071 #ifdef INET6 1072 if (isipv6) { 1073 inc.inc_flags |= INC_ISIPV6; 1074 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1075 inc.inc_flags |= INC_IPV6MINMTU; 1076 inc.inc6_faddr = ip6->ip6_src; 1077 inc.inc6_laddr = ip6->ip6_dst; 1078 } else 1079 #endif 1080 { 1081 inc.inc_faddr = ip->ip_src; 1082 inc.inc_laddr = ip->ip_dst; 1083 } 1084 inc.inc_fport = th->th_sport; 1085 inc.inc_lport = th->th_dport; 1086 inc.inc_fibnum = so->so_fibnum; 1087 1088 /* 1089 * Check for an existing connection attempt in syncache if 1090 * the flag is only ACK. A successful lookup creates a new 1091 * socket appended to the listen queue in SYN_RECEIVED state. 1092 */ 1093 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1094 /* 1095 * Parse the TCP options here because 1096 * syncookies need access to the reflected 1097 * timestamp. 1098 */ 1099 tcp_dooptions(&to, optp, optlen, 0); 1100 /* 1101 * NB: syncache_expand() doesn't unlock inp. 1102 */ 1103 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1104 if (rstreason < 0) { 1105 /* 1106 * A failing TCP MD5 signature comparison 1107 * must result in the segment being dropped 1108 * and must not produce any response back 1109 * to the sender. 1110 */ 1111 goto dropunlock; 1112 } else if (rstreason == 0) { 1113 /* 1114 * No syncache entry, or ACK was not for our 1115 * SYN/ACK. Do our protection against double 1116 * ACK. If peer sent us 2 ACKs, then for the 1117 * first one syncache_expand() successfully 1118 * converted syncache entry into a socket, 1119 * while we were waiting on the inpcb lock. We 1120 * don't want to sent RST for the second ACK, 1121 * so we perform second lookup without wildcard 1122 * match, hoping to find the new socket. If 1123 * the ACK is stray indeed, rstreason would 1124 * hint the above code that the lookup was a 1125 * second attempt. 1126 * 1127 * NB: syncache did its own logging 1128 * of the failure cause. 1129 */ 1130 INP_WUNLOCK(inp); 1131 rstreason = BANDLIM_RST_OPENPORT; 1132 lookupflag &= ~INPLOOKUP_WILDCARD; 1133 goto findpcb; 1134 } 1135 tfo_socket_result: 1136 if (so == NULL) { 1137 /* 1138 * We completed the 3-way handshake 1139 * but could not allocate a socket 1140 * either due to memory shortage, 1141 * listen queue length limits or 1142 * global socket limits. Send RST 1143 * or wait and have the remote end 1144 * retransmit the ACK for another 1145 * try. 1146 */ 1147 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1148 log(LOG_DEBUG, "%s; %s: Listen socket: " 1149 "Socket allocation failed due to " 1150 "limits or memory shortage, %s\n", 1151 s, __func__, 1152 V_tcp_sc_rst_sock_fail ? 1153 "sending RST" : "try again"); 1154 if (V_tcp_sc_rst_sock_fail) { 1155 rstreason = BANDLIM_UNLIMITED; 1156 goto dropwithreset; 1157 } else 1158 goto dropunlock; 1159 } 1160 /* 1161 * Socket is created in state SYN_RECEIVED. 1162 * Unlock the listen socket, lock the newly 1163 * created socket and update the tp variable. 1164 * If we came here via jump to tfo_socket_result, 1165 * then listening socket is read-locked. 1166 */ 1167 INP_UNLOCK(inp); /* listen socket */ 1168 inp = sotoinpcb(so); 1169 /* 1170 * New connection inpcb is already locked by 1171 * syncache_expand(). 1172 */ 1173 INP_WLOCK_ASSERT(inp); 1174 tp = intotcpcb(inp); 1175 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1176 ("%s: ", __func__)); 1177 /* 1178 * Process the segment and the data it 1179 * contains. tcp_do_segment() consumes 1180 * the mbuf chain and unlocks the inpcb. 1181 */ 1182 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1183 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1184 iptos); 1185 return (IPPROTO_DONE); 1186 } 1187 /* 1188 * Segment flag validation for new connection attempts: 1189 * 1190 * Our (SYN|ACK) response was rejected. 1191 * Check with syncache and remove entry to prevent 1192 * retransmits. 1193 * 1194 * NB: syncache_chkrst does its own logging of failure 1195 * causes. 1196 */ 1197 if (thflags & TH_RST) { 1198 syncache_chkrst(&inc, th, m, port); 1199 goto dropunlock; 1200 } 1201 /* 1202 * We can't do anything without SYN. 1203 */ 1204 if ((thflags & TH_SYN) == 0) { 1205 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1206 log(LOG_DEBUG, "%s; %s: Listen socket: " 1207 "SYN is missing, segment ignored\n", 1208 s, __func__); 1209 TCPSTAT_INC(tcps_badsyn); 1210 goto dropunlock; 1211 } 1212 /* 1213 * (SYN|ACK) is bogus on a listen socket. 1214 */ 1215 if (thflags & TH_ACK) { 1216 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1217 log(LOG_DEBUG, "%s; %s: Listen socket: " 1218 "SYN|ACK invalid, segment rejected\n", 1219 s, __func__); 1220 syncache_badack(&inc, port); /* XXX: Not needed! */ 1221 TCPSTAT_INC(tcps_badsyn); 1222 rstreason = BANDLIM_RST_OPENPORT; 1223 goto dropwithreset; 1224 } 1225 /* 1226 * If the drop_synfin option is enabled, drop all 1227 * segments with both the SYN and FIN bits set. 1228 * This prevents e.g. nmap from identifying the 1229 * TCP/IP stack. 1230 * XXX: Poor reasoning. nmap has other methods 1231 * and is constantly refining its stack detection 1232 * strategies. 1233 * XXX: This is a violation of the TCP specification 1234 * and was used by RFC1644. 1235 */ 1236 if ((thflags & TH_FIN) && V_drop_synfin) { 1237 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1238 log(LOG_DEBUG, "%s; %s: Listen socket: " 1239 "SYN|FIN segment ignored (based on " 1240 "sysctl setting)\n", s, __func__); 1241 TCPSTAT_INC(tcps_badsyn); 1242 goto dropunlock; 1243 } 1244 /* 1245 * Segment's flags are (SYN) or (SYN|FIN). 1246 * 1247 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1248 * as they do not affect the state of the TCP FSM. 1249 * The data pointed to by TH_URG and th_urp is ignored. 1250 */ 1251 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1252 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1253 KASSERT(thflags & (TH_SYN), 1254 ("%s: Listen socket: TH_SYN not set", __func__)); 1255 INP_RLOCK_ASSERT(inp); 1256 #ifdef INET6 1257 /* 1258 * If deprecated address is forbidden, 1259 * we do not accept SYN to deprecated interface 1260 * address to prevent any new inbound connection from 1261 * getting established. 1262 * When we do not accept SYN, we send a TCP RST, 1263 * with deprecated source address (instead of dropping 1264 * it). We compromise it as it is much better for peer 1265 * to send a RST, and RST will be the final packet 1266 * for the exchange. 1267 * 1268 * If we do not forbid deprecated addresses, we accept 1269 * the SYN packet. RFC2462 does not suggest dropping 1270 * SYN in this case. 1271 * If we decipher RFC2462 5.5.4, it says like this: 1272 * 1. use of deprecated addr with existing 1273 * communication is okay - "SHOULD continue to be 1274 * used" 1275 * 2. use of it with new communication: 1276 * (2a) "SHOULD NOT be used if alternate address 1277 * with sufficient scope is available" 1278 * (2b) nothing mentioned otherwise. 1279 * Here we fall into (2b) case as we have no choice in 1280 * our source address selection - we must obey the peer. 1281 * 1282 * The wording in RFC2462 is confusing, and there are 1283 * multiple description text for deprecated address 1284 * handling - worse, they are not exactly the same. 1285 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1286 */ 1287 if (isipv6 && !V_ip6_use_deprecated) { 1288 struct in6_ifaddr *ia6; 1289 1290 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1291 if (ia6 != NULL && 1292 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1293 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1294 log(LOG_DEBUG, "%s; %s: Listen socket: " 1295 "Connection attempt to deprecated " 1296 "IPv6 address rejected\n", 1297 s, __func__); 1298 rstreason = BANDLIM_RST_OPENPORT; 1299 goto dropwithreset; 1300 } 1301 } 1302 #endif /* INET6 */ 1303 /* 1304 * Basic sanity checks on incoming SYN requests: 1305 * Don't respond if the destination is a link layer 1306 * broadcast according to RFC1122 4.2.3.10, p. 104. 1307 * If it is from this socket it must be forged. 1308 * Don't respond if the source or destination is a 1309 * global or subnet broad- or multicast address. 1310 * Note that it is quite possible to receive unicast 1311 * link-layer packets with a broadcast IP address. Use 1312 * in_broadcast() to find them. 1313 */ 1314 if (m->m_flags & (M_BCAST|M_MCAST)) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from broad- or multicast " 1318 "link layer address ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 #ifdef INET6 1322 if (isipv6) { 1323 if (th->th_dport == th->th_sport && 1324 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1326 log(LOG_DEBUG, "%s; %s: Listen socket: " 1327 "Connection attempt to/from self " 1328 "ignored\n", s, __func__); 1329 goto dropunlock; 1330 } 1331 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1332 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1333 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1334 log(LOG_DEBUG, "%s; %s: Listen socket: " 1335 "Connection attempt from/to multicast " 1336 "address ignored\n", s, __func__); 1337 goto dropunlock; 1338 } 1339 } 1340 #endif 1341 #if defined(INET) && defined(INET6) 1342 else 1343 #endif 1344 #ifdef INET 1345 { 1346 if (th->th_dport == th->th_sport && 1347 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1348 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1349 log(LOG_DEBUG, "%s; %s: Listen socket: " 1350 "Connection attempt from/to self " 1351 "ignored\n", s, __func__); 1352 goto dropunlock; 1353 } 1354 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1355 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1356 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1357 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1358 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1359 log(LOG_DEBUG, "%s; %s: Listen socket: " 1360 "Connection attempt from/to broad- " 1361 "or multicast address ignored\n", 1362 s, __func__); 1363 goto dropunlock; 1364 } 1365 } 1366 #endif 1367 /* 1368 * SYN appears to be valid. Create compressed TCP state 1369 * for syncache. 1370 */ 1371 #ifdef TCPDEBUG 1372 if (so->so_options & SO_DEBUG) 1373 tcp_trace(TA_INPUT, ostate, tp, 1374 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1375 #endif 1376 TCP_PROBE3(debug__input, tp, th, m); 1377 tcp_dooptions(&to, optp, optlen, TO_SYN); 1378 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1379 iptos, port)) != NULL) 1380 goto tfo_socket_result; 1381 1382 /* 1383 * Entry added to syncache and mbuf consumed. 1384 * Only the listen socket is unlocked by syncache_add(). 1385 */ 1386 return (IPPROTO_DONE); 1387 } else if (tp->t_state == TCPS_LISTEN) { 1388 /* 1389 * When a listen socket is torn down the SO_ACCEPTCONN 1390 * flag is removed first while connections are drained 1391 * from the accept queue in a unlock/lock cycle of the 1392 * ACCEPT_LOCK, opening a race condition allowing a SYN 1393 * attempt go through unhandled. 1394 */ 1395 goto dropunlock; 1396 } 1397 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1398 if (tp->t_flags & TF_SIGNATURE) { 1399 tcp_dooptions(&to, optp, optlen, thflags); 1400 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1401 TCPSTAT_INC(tcps_sig_err_nosigopt); 1402 goto dropunlock; 1403 } 1404 if (!TCPMD5_ENABLED() || 1405 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1406 goto dropunlock; 1407 } 1408 #endif 1409 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1410 1411 /* 1412 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1413 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1414 * the inpcb, and unlocks pcbinfo. 1415 * 1416 * XXXGL: in case of a pure SYN arriving on existing connection 1417 * TCP stacks won't need to modify the PCB, they would either drop 1418 * the segment silently, or send a challenge ACK. However, we try 1419 * to upgrade the lock, because calling convention for stacks is 1420 * write-lock on PCB. If upgrade fails, drop the SYN. 1421 */ 1422 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1423 goto dropunlock; 1424 1425 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1426 return (IPPROTO_DONE); 1427 1428 dropwithreset: 1429 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1430 1431 if (inp != NULL) { 1432 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1433 INP_UNLOCK(inp); 1434 } else 1435 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1436 m = NULL; /* mbuf chain got consumed. */ 1437 goto drop; 1438 1439 dropunlock: 1440 if (m != NULL) 1441 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1442 1443 if (inp != NULL) 1444 INP_UNLOCK(inp); 1445 1446 drop: 1447 if (s != NULL) 1448 free(s, M_TCPLOG); 1449 if (m != NULL) 1450 m_freem(m); 1451 return (IPPROTO_DONE); 1452 } 1453 1454 /* 1455 * Automatic sizing of receive socket buffer. Often the send 1456 * buffer size is not optimally adjusted to the actual network 1457 * conditions at hand (delay bandwidth product). Setting the 1458 * buffer size too small limits throughput on links with high 1459 * bandwidth and high delay (eg. trans-continental/oceanic links). 1460 * 1461 * On the receive side the socket buffer memory is only rarely 1462 * used to any significant extent. This allows us to be much 1463 * more aggressive in scaling the receive socket buffer. For 1464 * the case that the buffer space is actually used to a large 1465 * extent and we run out of kernel memory we can simply drop 1466 * the new segments; TCP on the sender will just retransmit it 1467 * later. Setting the buffer size too big may only consume too 1468 * much kernel memory if the application doesn't read() from 1469 * the socket or packet loss or reordering makes use of the 1470 * reassembly queue. 1471 * 1472 * The criteria to step up the receive buffer one notch are: 1473 * 1. Application has not set receive buffer size with 1474 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1475 * 2. the number of bytes received during 1/2 of an sRTT 1476 * is at least 3/8 of the current socket buffer size. 1477 * 3. receive buffer size has not hit maximal automatic size; 1478 * 1479 * If all of the criteria are met we increaset the socket buffer 1480 * by a 1/2 (bounded by the max). This allows us to keep ahead 1481 * of slow-start but also makes it so our peer never gets limited 1482 * by our rwnd which we then open up causing a burst. 1483 * 1484 * This algorithm does two steps per RTT at most and only if 1485 * we receive a bulk stream w/o packet losses or reorderings. 1486 * Shrinking the buffer during idle times is not necessary as 1487 * it doesn't consume any memory when idle. 1488 * 1489 * TODO: Only step up if the application is actually serving 1490 * the buffer to better manage the socket buffer resources. 1491 */ 1492 int 1493 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1494 struct tcpcb *tp, int tlen) 1495 { 1496 int newsize = 0; 1497 1498 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1499 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1500 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1501 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1502 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1503 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1504 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1505 } 1506 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1507 1508 /* Start over with next RTT. */ 1509 tp->rfbuf_ts = 0; 1510 tp->rfbuf_cnt = 0; 1511 } else { 1512 tp->rfbuf_cnt += tlen; /* add up */ 1513 } 1514 return (newsize); 1515 } 1516 1517 int 1518 tcp_input(struct mbuf **mp, int *offp, int proto) 1519 { 1520 return(tcp_input_with_port(mp, offp, proto, 0)); 1521 } 1522 1523 void 1524 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1525 { 1526 /* 1527 * Since tp might be gone if the session entered 1528 * the TIME_WAIT state before coming here, we need 1529 * to check if the socket is still connected. 1530 */ 1531 if (tp == NULL) { 1532 return; 1533 } 1534 if (so == NULL) { 1535 return; 1536 } 1537 INP_LOCK_ASSERT(tp->t_inpcb); 1538 if (tp->t_flags & TF_WAKESOR) { 1539 tp->t_flags &= ~TF_WAKESOR; 1540 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1541 sorwakeup_locked(so); 1542 } 1543 } 1544 1545 void 1546 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1547 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1548 { 1549 uint16_t thflags; 1550 int acked, ourfinisacked, needoutput = 0, sack_changed; 1551 int rstreason, todrop, win, incforsyn = 0; 1552 uint32_t tiwin; 1553 uint16_t nsegs; 1554 char *s; 1555 struct in_conninfo *inc; 1556 struct mbuf *mfree; 1557 struct tcpopt to; 1558 int tfo_syn; 1559 u_int maxseg; 1560 1561 #ifdef TCPDEBUG 1562 /* 1563 * The size of tcp_saveipgen must be the size of the max ip header, 1564 * now IPv6. 1565 */ 1566 u_char tcp_saveipgen[IP6_HDR_LEN]; 1567 struct tcphdr tcp_savetcp; 1568 short ostate = 0; 1569 #endif 1570 thflags = tcp_get_flags(th); 1571 inc = &tp->t_inpcb->inp_inc; 1572 tp->sackhint.last_sack_ack = 0; 1573 sack_changed = 0; 1574 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1575 1576 NET_EPOCH_ASSERT(); 1577 INP_WLOCK_ASSERT(tp->t_inpcb); 1578 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1579 __func__)); 1580 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1581 __func__)); 1582 1583 #ifdef TCPPCAP 1584 /* Save segment, if requested. */ 1585 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1586 #endif 1587 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1588 tlen, NULL, true); 1589 1590 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1591 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1592 log(LOG_DEBUG, "%s; %s: " 1593 "SYN|FIN segment ignored (based on " 1594 "sysctl setting)\n", s, __func__); 1595 free(s, M_TCPLOG); 1596 } 1597 goto drop; 1598 } 1599 1600 /* 1601 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1602 * check SEQ.ACK first. 1603 */ 1604 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1605 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1606 rstreason = BANDLIM_UNLIMITED; 1607 goto dropwithreset; 1608 } 1609 1610 /* 1611 * Segment received on connection. 1612 * Reset idle time and keep-alive timer. 1613 * XXX: This should be done after segment 1614 * validation to ignore broken/spoofed segs. 1615 */ 1616 tp->t_rcvtime = ticks; 1617 1618 /* 1619 * Scale up the window into a 32-bit value. 1620 * For the SYN_SENT state the scale is zero. 1621 */ 1622 tiwin = th->th_win << tp->snd_scale; 1623 #ifdef STATS 1624 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1625 #endif 1626 1627 /* 1628 * TCP ECN processing. 1629 */ 1630 if (tcp_ecn_input_segment(tp, thflags, iptos)) 1631 cc_cong_signal(tp, th, CC_ECN); 1632 1633 /* 1634 * Parse options on any incoming segment. 1635 */ 1636 tcp_dooptions(&to, (u_char *)(th + 1), 1637 (th->th_off << 2) - sizeof(struct tcphdr), 1638 (thflags & TH_SYN) ? TO_SYN : 0); 1639 1640 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1641 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1642 (to.to_flags & TOF_SIGNATURE) == 0) { 1643 TCPSTAT_INC(tcps_sig_err_sigopt); 1644 /* XXX: should drop? */ 1645 } 1646 #endif 1647 /* 1648 * If echoed timestamp is later than the current time, 1649 * fall back to non RFC1323 RTT calculation. Normalize 1650 * timestamp if syncookies were used when this connection 1651 * was established. 1652 */ 1653 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1654 to.to_tsecr -= tp->ts_offset; 1655 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1656 to.to_tsecr = 0; 1657 else if (tp->t_rxtshift == 1 && 1658 tp->t_flags & TF_PREVVALID && 1659 tp->t_badrxtwin != 0 && 1660 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 1661 cc_cong_signal(tp, th, CC_RTO_ERR); 1662 } 1663 /* 1664 * Process options only when we get SYN/ACK back. The SYN case 1665 * for incoming connections is handled in tcp_syncache. 1666 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1667 * or <SYN,ACK>) segment itself is never scaled. 1668 * XXX this is traditional behavior, may need to be cleaned up. 1669 */ 1670 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1671 /* Handle parallel SYN for ECN */ 1672 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1673 if ((to.to_flags & TOF_SCALE) && 1674 (tp->t_flags & TF_REQ_SCALE) && 1675 !(tp->t_flags & TF_NOOPT)) { 1676 tp->t_flags |= TF_RCVD_SCALE; 1677 tp->snd_scale = to.to_wscale; 1678 } else 1679 tp->t_flags &= ~TF_REQ_SCALE; 1680 /* 1681 * Initial send window. It will be updated with 1682 * the next incoming segment to the scaled value. 1683 */ 1684 tp->snd_wnd = th->th_win; 1685 if ((to.to_flags & TOF_TS) && 1686 (tp->t_flags & TF_REQ_TSTMP) && 1687 !(tp->t_flags & TF_NOOPT)) { 1688 tp->t_flags |= TF_RCVD_TSTMP; 1689 tp->ts_recent = to.to_tsval; 1690 tp->ts_recent_age = tcp_ts_getticks(); 1691 } else 1692 tp->t_flags &= ~TF_REQ_TSTMP; 1693 if (to.to_flags & TOF_MSS) 1694 tcp_mss(tp, to.to_mss); 1695 if ((tp->t_flags & TF_SACK_PERMIT) && 1696 (!(to.to_flags & TOF_SACKPERM) || 1697 (tp->t_flags & TF_NOOPT))) 1698 tp->t_flags &= ~TF_SACK_PERMIT; 1699 if (IS_FASTOPEN(tp->t_flags)) { 1700 if ((to.to_flags & TOF_FASTOPEN) && 1701 !(tp->t_flags & TF_NOOPT)) { 1702 uint16_t mss; 1703 1704 if (to.to_flags & TOF_MSS) 1705 mss = to.to_mss; 1706 else 1707 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1708 mss = TCP6_MSS; 1709 else 1710 mss = TCP_MSS; 1711 tcp_fastopen_update_cache(tp, mss, 1712 to.to_tfo_len, to.to_tfo_cookie); 1713 } else 1714 tcp_fastopen_disable_path(tp); 1715 } 1716 } 1717 1718 /* 1719 * If timestamps were negotiated during SYN/ACK and a 1720 * segment without a timestamp is received, silently drop 1721 * the segment, unless it is a RST segment or missing timestamps are 1722 * tolerated. 1723 * See section 3.2 of RFC 7323. 1724 */ 1725 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1726 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1727 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1728 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1729 "segment processed normally\n", 1730 s, __func__); 1731 free(s, M_TCPLOG); 1732 } 1733 } else { 1734 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1735 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1736 "segment silently dropped\n", s, __func__); 1737 free(s, M_TCPLOG); 1738 } 1739 goto drop; 1740 } 1741 } 1742 /* 1743 * If timestamps were not negotiated during SYN/ACK and a 1744 * segment with a timestamp is received, ignore the 1745 * timestamp and process the packet normally. 1746 * See section 3.2 of RFC 7323. 1747 */ 1748 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1749 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1750 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1751 "segment processed normally\n", s, __func__); 1752 free(s, M_TCPLOG); 1753 } 1754 } 1755 1756 /* 1757 * Header prediction: check for the two common cases 1758 * of a uni-directional data xfer. If the packet has 1759 * no control flags, is in-sequence, the window didn't 1760 * change and we're not retransmitting, it's a 1761 * candidate. If the length is zero and the ack moved 1762 * forward, we're the sender side of the xfer. Just 1763 * free the data acked & wake any higher level process 1764 * that was blocked waiting for space. If the length 1765 * is non-zero and the ack didn't move, we're the 1766 * receiver side. If we're getting packets in-order 1767 * (the reassembly queue is empty), add the data to 1768 * the socket buffer and note that we need a delayed ack. 1769 * Make sure that the hidden state-flags are also off. 1770 * Since we check for TCPS_ESTABLISHED first, it can only 1771 * be TH_NEEDSYN. 1772 */ 1773 if (tp->t_state == TCPS_ESTABLISHED && 1774 th->th_seq == tp->rcv_nxt && 1775 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1776 tp->snd_nxt == tp->snd_max && 1777 tiwin && tiwin == tp->snd_wnd && 1778 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1779 SEGQ_EMPTY(tp) && 1780 ((to.to_flags & TOF_TS) == 0 || 1781 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1782 /* 1783 * If last ACK falls within this segment's sequence numbers, 1784 * record the timestamp. 1785 * NOTE that the test is modified according to the latest 1786 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1787 */ 1788 if ((to.to_flags & TOF_TS) != 0 && 1789 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1790 tp->ts_recent_age = tcp_ts_getticks(); 1791 tp->ts_recent = to.to_tsval; 1792 } 1793 1794 if (tlen == 0) { 1795 if (SEQ_GT(th->th_ack, tp->snd_una) && 1796 SEQ_LEQ(th->th_ack, tp->snd_max) && 1797 !IN_RECOVERY(tp->t_flags) && 1798 (to.to_flags & TOF_SACK) == 0 && 1799 TAILQ_EMPTY(&tp->snd_holes)) { 1800 /* 1801 * This is a pure ack for outstanding data. 1802 */ 1803 TCPSTAT_INC(tcps_predack); 1804 1805 /* 1806 * "bad retransmit" recovery without timestamps. 1807 */ 1808 if ((to.to_flags & TOF_TS) == 0 && 1809 tp->t_rxtshift == 1 && 1810 tp->t_flags & TF_PREVVALID && 1811 tp->t_badrxtwin != 0 && 1812 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1813 cc_cong_signal(tp, th, CC_RTO_ERR); 1814 } 1815 1816 /* 1817 * Recalculate the transmit timer / rtt. 1818 * 1819 * Some boxes send broken timestamp replies 1820 * during the SYN+ACK phase, ignore 1821 * timestamps of 0 or we could calculate a 1822 * huge RTT and blow up the retransmit timer. 1823 */ 1824 if ((to.to_flags & TOF_TS) != 0 && 1825 to.to_tsecr) { 1826 uint32_t t; 1827 1828 t = tcp_ts_getticks() - to.to_tsecr; 1829 if (!tp->t_rttlow || tp->t_rttlow > t) 1830 tp->t_rttlow = t; 1831 tcp_xmit_timer(tp, 1832 TCP_TS_TO_TICKS(t) + 1); 1833 } else if (tp->t_rtttime && 1834 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1835 if (!tp->t_rttlow || 1836 tp->t_rttlow > ticks - tp->t_rtttime) 1837 tp->t_rttlow = ticks - tp->t_rtttime; 1838 tcp_xmit_timer(tp, 1839 ticks - tp->t_rtttime); 1840 } 1841 acked = BYTES_THIS_ACK(tp, th); 1842 1843 #ifdef TCP_HHOOK 1844 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1845 hhook_run_tcp_est_in(tp, th, &to); 1846 #endif 1847 1848 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1849 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1850 sbdrop(&so->so_snd, acked); 1851 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1852 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1853 tp->snd_recover = th->th_ack - 1; 1854 1855 /* 1856 * Let the congestion control algorithm update 1857 * congestion control related information. This 1858 * typically means increasing the congestion 1859 * window. 1860 */ 1861 cc_ack_received(tp, th, nsegs, CC_ACK); 1862 1863 tp->snd_una = th->th_ack; 1864 /* 1865 * Pull snd_wl2 up to prevent seq wrap relative 1866 * to th_ack. 1867 */ 1868 tp->snd_wl2 = th->th_ack; 1869 tp->t_dupacks = 0; 1870 m_freem(m); 1871 1872 /* 1873 * If all outstanding data are acked, stop 1874 * retransmit timer, otherwise restart timer 1875 * using current (possibly backed-off) value. 1876 * If process is waiting for space, 1877 * wakeup/selwakeup/signal. If data 1878 * are ready to send, let tcp_output 1879 * decide between more output or persist. 1880 */ 1881 #ifdef TCPDEBUG 1882 if (so->so_options & SO_DEBUG) 1883 tcp_trace(TA_INPUT, ostate, tp, 1884 (void *)tcp_saveipgen, 1885 &tcp_savetcp, 0); 1886 #endif 1887 TCP_PROBE3(debug__input, tp, th, m); 1888 if (tp->snd_una == tp->snd_max) 1889 tcp_timer_activate(tp, TT_REXMT, 0); 1890 else if (!tcp_timer_active(tp, TT_PERSIST)) 1891 tcp_timer_activate(tp, TT_REXMT, 1892 tp->t_rxtcur); 1893 sowwakeup(so); 1894 if (sbavail(&so->so_snd)) 1895 (void) tcp_output(tp); 1896 goto check_delack; 1897 } 1898 } else if (th->th_ack == tp->snd_una && 1899 tlen <= sbspace(&so->so_rcv)) { 1900 int newsize = 0; /* automatic sockbuf scaling */ 1901 1902 /* 1903 * This is a pure, in-sequence data packet with 1904 * nothing on the reassembly queue and we have enough 1905 * buffer space to take it. 1906 */ 1907 /* Clean receiver SACK report if present */ 1908 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1909 tcp_clean_sackreport(tp); 1910 TCPSTAT_INC(tcps_preddat); 1911 tp->rcv_nxt += tlen; 1912 if (tlen && 1913 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1914 (tp->t_fbyte_in == 0)) { 1915 tp->t_fbyte_in = ticks; 1916 if (tp->t_fbyte_in == 0) 1917 tp->t_fbyte_in = 1; 1918 if (tp->t_fbyte_out && tp->t_fbyte_in) 1919 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1920 } 1921 /* 1922 * Pull snd_wl1 up to prevent seq wrap relative to 1923 * th_seq. 1924 */ 1925 tp->snd_wl1 = th->th_seq; 1926 /* 1927 * Pull rcv_up up to prevent seq wrap relative to 1928 * rcv_nxt. 1929 */ 1930 tp->rcv_up = tp->rcv_nxt; 1931 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1932 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1933 #ifdef TCPDEBUG 1934 if (so->so_options & SO_DEBUG) 1935 tcp_trace(TA_INPUT, ostate, tp, 1936 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1937 #endif 1938 TCP_PROBE3(debug__input, tp, th, m); 1939 1940 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1941 1942 /* Add data to socket buffer. */ 1943 SOCKBUF_LOCK(&so->so_rcv); 1944 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1945 m_freem(m); 1946 } else { 1947 /* 1948 * Set new socket buffer size. 1949 * Give up when limit is reached. 1950 */ 1951 if (newsize) 1952 if (!sbreserve_locked(so, SO_RCV, 1953 newsize, NULL)) 1954 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1955 m_adj(m, drop_hdrlen); /* delayed header drop */ 1956 sbappendstream_locked(&so->so_rcv, m, 0); 1957 } 1958 /* NB: sorwakeup_locked() does an implicit unlock. */ 1959 sorwakeup_locked(so); 1960 if (DELAY_ACK(tp, tlen)) { 1961 tp->t_flags |= TF_DELACK; 1962 } else { 1963 tp->t_flags |= TF_ACKNOW; 1964 tcp_output(tp); 1965 } 1966 goto check_delack; 1967 } 1968 } 1969 1970 /* 1971 * Calculate amount of space in receive window, 1972 * and then do TCP input processing. 1973 * Receive window is amount of space in rcv queue, 1974 * but not less than advertised window. 1975 */ 1976 win = sbspace(&so->so_rcv); 1977 if (win < 0) 1978 win = 0; 1979 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1980 1981 switch (tp->t_state) { 1982 /* 1983 * If the state is SYN_RECEIVED: 1984 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1985 */ 1986 case TCPS_SYN_RECEIVED: 1987 if ((thflags & TH_ACK) && 1988 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1989 SEQ_GT(th->th_ack, tp->snd_max))) { 1990 rstreason = BANDLIM_RST_OPENPORT; 1991 goto dropwithreset; 1992 } 1993 if (IS_FASTOPEN(tp->t_flags)) { 1994 /* 1995 * When a TFO connection is in SYN_RECEIVED, the 1996 * only valid packets are the initial SYN, a 1997 * retransmit/copy of the initial SYN (possibly with 1998 * a subset of the original data), a valid ACK, a 1999 * FIN, or a RST. 2000 */ 2001 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 2002 rstreason = BANDLIM_RST_OPENPORT; 2003 goto dropwithreset; 2004 } else if (thflags & TH_SYN) { 2005 /* non-initial SYN is ignored */ 2006 if ((tcp_timer_active(tp, TT_DELACK) || 2007 tcp_timer_active(tp, TT_REXMT))) 2008 goto drop; 2009 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 2010 goto drop; 2011 } 2012 } 2013 break; 2014 2015 /* 2016 * If the state is SYN_SENT: 2017 * if seg contains a RST with valid ACK (SEQ.ACK has already 2018 * been verified), then drop the connection. 2019 * if seg contains a RST without an ACK, drop the seg. 2020 * if seg does not contain SYN, then drop the seg. 2021 * Otherwise this is an acceptable SYN segment 2022 * initialize tp->rcv_nxt and tp->irs 2023 * if seg contains ack then advance tp->snd_una 2024 * if seg contains an ECE and ECN support is enabled, the stream 2025 * is ECN capable. 2026 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2027 * arrange for segment to be acked (eventually) 2028 * continue processing rest of data/controls, beginning with URG 2029 */ 2030 case TCPS_SYN_SENT: 2031 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2032 TCP_PROBE5(connect__refused, NULL, tp, 2033 m, tp, th); 2034 tp = tcp_drop(tp, ECONNREFUSED); 2035 } 2036 if (thflags & TH_RST) 2037 goto drop; 2038 if (!(thflags & TH_SYN)) 2039 goto drop; 2040 2041 tp->irs = th->th_seq; 2042 tcp_rcvseqinit(tp); 2043 if (thflags & TH_ACK) { 2044 int tfo_partial_ack = 0; 2045 2046 TCPSTAT_INC(tcps_connects); 2047 soisconnected(so); 2048 #ifdef MAC 2049 mac_socketpeer_set_from_mbuf(m, so); 2050 #endif 2051 /* Do window scaling on this connection? */ 2052 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2053 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2054 tp->rcv_scale = tp->request_r_scale; 2055 } 2056 tp->rcv_adv += min(tp->rcv_wnd, 2057 TCP_MAXWIN << tp->rcv_scale); 2058 tp->snd_una++; /* SYN is acked */ 2059 /* 2060 * If not all the data that was sent in the TFO SYN 2061 * has been acked, resend the remainder right away. 2062 */ 2063 if (IS_FASTOPEN(tp->t_flags) && 2064 (tp->snd_una != tp->snd_max)) { 2065 tp->snd_nxt = th->th_ack; 2066 tfo_partial_ack = 1; 2067 } 2068 /* 2069 * If there's data, delay ACK; if there's also a FIN 2070 * ACKNOW will be turned on later. 2071 */ 2072 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2073 tcp_timer_activate(tp, TT_DELACK, 2074 tcp_delacktime); 2075 else 2076 tp->t_flags |= TF_ACKNOW; 2077 2078 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2079 2080 /* 2081 * Received <SYN,ACK> in SYN_SENT[*] state. 2082 * Transitions: 2083 * SYN_SENT --> ESTABLISHED 2084 * SYN_SENT* --> FIN_WAIT_1 2085 */ 2086 tp->t_starttime = ticks; 2087 if (tp->t_flags & TF_NEEDFIN) { 2088 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2089 tp->t_flags &= ~TF_NEEDFIN; 2090 thflags &= ~TH_SYN; 2091 } else { 2092 tcp_state_change(tp, TCPS_ESTABLISHED); 2093 TCP_PROBE5(connect__established, NULL, tp, 2094 m, tp, th); 2095 cc_conn_init(tp); 2096 tcp_timer_activate(tp, TT_KEEP, 2097 TP_KEEPIDLE(tp)); 2098 } 2099 } else { 2100 /* 2101 * Received initial SYN in SYN-SENT[*] state => 2102 * simultaneous open. 2103 * If it succeeds, connection is * half-synchronized. 2104 * Otherwise, do 3-way handshake: 2105 * SYN-SENT -> SYN-RECEIVED 2106 * SYN-SENT* -> SYN-RECEIVED* 2107 */ 2108 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2109 tcp_timer_activate(tp, TT_REXMT, 0); 2110 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2111 } 2112 2113 INP_WLOCK_ASSERT(tp->t_inpcb); 2114 2115 /* 2116 * Advance th->th_seq to correspond to first data byte. 2117 * If data, trim to stay within window, 2118 * dropping FIN if necessary. 2119 */ 2120 th->th_seq++; 2121 if (tlen > tp->rcv_wnd) { 2122 todrop = tlen - tp->rcv_wnd; 2123 m_adj(m, -todrop); 2124 tlen = tp->rcv_wnd; 2125 thflags &= ~TH_FIN; 2126 TCPSTAT_INC(tcps_rcvpackafterwin); 2127 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2128 } 2129 tp->snd_wl1 = th->th_seq - 1; 2130 tp->rcv_up = th->th_seq; 2131 /* 2132 * Client side of transaction: already sent SYN and data. 2133 * If the remote host used T/TCP to validate the SYN, 2134 * our data will be ACK'd; if so, enter normal data segment 2135 * processing in the middle of step 5, ack processing. 2136 * Otherwise, goto step 6. 2137 */ 2138 if (thflags & TH_ACK) 2139 goto process_ACK; 2140 2141 goto step6; 2142 } 2143 2144 /* 2145 * States other than LISTEN or SYN_SENT. 2146 * First check the RST flag and sequence number since reset segments 2147 * are exempt from the timestamp and connection count tests. This 2148 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2149 * below which allowed reset segments in half the sequence space 2150 * to fall though and be processed (which gives forged reset 2151 * segments with a random sequence number a 50 percent chance of 2152 * killing a connection). 2153 * Then check timestamp, if present. 2154 * Then check the connection count, if present. 2155 * Then check that at least some bytes of segment are within 2156 * receive window. If segment begins before rcv_nxt, 2157 * drop leading data (and SYN); if nothing left, just ack. 2158 */ 2159 if (thflags & TH_RST) { 2160 /* 2161 * RFC5961 Section 3.2 2162 * 2163 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2164 * - If RST is in window, we send challenge ACK. 2165 * 2166 * Note: to take into account delayed ACKs, we should 2167 * test against last_ack_sent instead of rcv_nxt. 2168 * Note 2: we handle special case of closed window, not 2169 * covered by the RFC. 2170 */ 2171 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2172 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2173 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2174 KASSERT(tp->t_state != TCPS_SYN_SENT, 2175 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2176 __func__, th, tp)); 2177 2178 if (V_tcp_insecure_rst || 2179 tp->last_ack_sent == th->th_seq) { 2180 TCPSTAT_INC(tcps_drops); 2181 /* Drop the connection. */ 2182 switch (tp->t_state) { 2183 case TCPS_SYN_RECEIVED: 2184 so->so_error = ECONNREFUSED; 2185 goto close; 2186 case TCPS_ESTABLISHED: 2187 case TCPS_FIN_WAIT_1: 2188 case TCPS_FIN_WAIT_2: 2189 case TCPS_CLOSE_WAIT: 2190 case TCPS_CLOSING: 2191 case TCPS_LAST_ACK: 2192 so->so_error = ECONNRESET; 2193 close: 2194 /* FALLTHROUGH */ 2195 default: 2196 tp = tcp_close(tp); 2197 } 2198 } else { 2199 TCPSTAT_INC(tcps_badrst); 2200 /* Send challenge ACK. */ 2201 tcp_respond(tp, mtod(m, void *), th, m, 2202 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2203 tp->last_ack_sent = tp->rcv_nxt; 2204 m = NULL; 2205 } 2206 } 2207 goto drop; 2208 } 2209 2210 /* 2211 * RFC5961 Section 4.2 2212 * Send challenge ACK for any SYN in synchronized state. 2213 */ 2214 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2215 tp->t_state != TCPS_SYN_RECEIVED) { 2216 TCPSTAT_INC(tcps_badsyn); 2217 if (V_tcp_insecure_syn && 2218 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2219 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2220 tp = tcp_drop(tp, ECONNRESET); 2221 rstreason = BANDLIM_UNLIMITED; 2222 } else { 2223 /* Send challenge ACK. */ 2224 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2225 tp->snd_nxt, TH_ACK); 2226 tp->last_ack_sent = tp->rcv_nxt; 2227 m = NULL; 2228 } 2229 goto drop; 2230 } 2231 2232 /* 2233 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2234 * and it's less than ts_recent, drop it. 2235 */ 2236 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2237 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2238 /* Check to see if ts_recent is over 24 days old. */ 2239 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2240 /* 2241 * Invalidate ts_recent. If this segment updates 2242 * ts_recent, the age will be reset later and ts_recent 2243 * will get a valid value. If it does not, setting 2244 * ts_recent to zero will at least satisfy the 2245 * requirement that zero be placed in the timestamp 2246 * echo reply when ts_recent isn't valid. The 2247 * age isn't reset until we get a valid ts_recent 2248 * because we don't want out-of-order segments to be 2249 * dropped when ts_recent is old. 2250 */ 2251 tp->ts_recent = 0; 2252 } else { 2253 TCPSTAT_INC(tcps_rcvduppack); 2254 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2255 TCPSTAT_INC(tcps_pawsdrop); 2256 if (tlen) 2257 goto dropafterack; 2258 goto drop; 2259 } 2260 } 2261 2262 /* 2263 * In the SYN-RECEIVED state, validate that the packet belongs to 2264 * this connection before trimming the data to fit the receive 2265 * window. Check the sequence number versus IRS since we know 2266 * the sequence numbers haven't wrapped. This is a partial fix 2267 * for the "LAND" DoS attack. 2268 */ 2269 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2270 rstreason = BANDLIM_RST_OPENPORT; 2271 goto dropwithreset; 2272 } 2273 2274 todrop = tp->rcv_nxt - th->th_seq; 2275 if (todrop > 0) { 2276 if (thflags & TH_SYN) { 2277 thflags &= ~TH_SYN; 2278 th->th_seq++; 2279 if (th->th_urp > 1) 2280 th->th_urp--; 2281 else 2282 thflags &= ~TH_URG; 2283 todrop--; 2284 } 2285 /* 2286 * Following if statement from Stevens, vol. 2, p. 960. 2287 */ 2288 if (todrop > tlen 2289 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2290 /* 2291 * Any valid FIN must be to the left of the window. 2292 * At this point the FIN must be a duplicate or out 2293 * of sequence; drop it. 2294 */ 2295 thflags &= ~TH_FIN; 2296 2297 /* 2298 * Send an ACK to resynchronize and drop any data. 2299 * But keep on processing for RST or ACK. 2300 */ 2301 tp->t_flags |= TF_ACKNOW; 2302 todrop = tlen; 2303 TCPSTAT_INC(tcps_rcvduppack); 2304 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2305 } else { 2306 TCPSTAT_INC(tcps_rcvpartduppack); 2307 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2308 } 2309 /* 2310 * DSACK - add SACK block for dropped range 2311 */ 2312 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2313 tcp_update_sack_list(tp, th->th_seq, 2314 th->th_seq + todrop); 2315 /* 2316 * ACK now, as the next in-sequence segment 2317 * will clear the DSACK block again 2318 */ 2319 tp->t_flags |= TF_ACKNOW; 2320 } 2321 drop_hdrlen += todrop; /* drop from the top afterwards */ 2322 th->th_seq += todrop; 2323 tlen -= todrop; 2324 if (th->th_urp > todrop) 2325 th->th_urp -= todrop; 2326 else { 2327 thflags &= ~TH_URG; 2328 th->th_urp = 0; 2329 } 2330 } 2331 2332 /* 2333 * If new data are received on a connection after the 2334 * user processes are gone, then RST the other end. 2335 */ 2336 if ((tp->t_flags & TF_CLOSED) && tlen) { 2337 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2338 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2339 "after socket was closed, " 2340 "sending RST and removing tcpcb\n", 2341 s, __func__, tcpstates[tp->t_state], tlen); 2342 free(s, M_TCPLOG); 2343 } 2344 tp = tcp_close(tp); 2345 TCPSTAT_INC(tcps_rcvafterclose); 2346 rstreason = BANDLIM_UNLIMITED; 2347 goto dropwithreset; 2348 } 2349 2350 /* 2351 * If segment ends after window, drop trailing data 2352 * (and PUSH and FIN); if nothing left, just ACK. 2353 */ 2354 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2355 if (todrop > 0) { 2356 TCPSTAT_INC(tcps_rcvpackafterwin); 2357 if (todrop >= tlen) { 2358 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2359 /* 2360 * If window is closed can only take segments at 2361 * window edge, and have to drop data and PUSH from 2362 * incoming segments. Continue processing, but 2363 * remember to ack. Otherwise, drop segment 2364 * and ack. 2365 */ 2366 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2367 tp->t_flags |= TF_ACKNOW; 2368 TCPSTAT_INC(tcps_rcvwinprobe); 2369 } else 2370 goto dropafterack; 2371 } else 2372 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2373 m_adj(m, -todrop); 2374 tlen -= todrop; 2375 thflags &= ~(TH_PUSH|TH_FIN); 2376 } 2377 2378 /* 2379 * If last ACK falls within this segment's sequence numbers, 2380 * record its timestamp. 2381 * NOTE: 2382 * 1) That the test incorporates suggestions from the latest 2383 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2384 * 2) That updating only on newer timestamps interferes with 2385 * our earlier PAWS tests, so this check should be solely 2386 * predicated on the sequence space of this segment. 2387 * 3) That we modify the segment boundary check to be 2388 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2389 * instead of RFC1323's 2390 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2391 * This modified check allows us to overcome RFC1323's 2392 * limitations as described in Stevens TCP/IP Illustrated 2393 * Vol. 2 p.869. In such cases, we can still calculate the 2394 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2395 */ 2396 if ((to.to_flags & TOF_TS) != 0 && 2397 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2398 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2399 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2400 tp->ts_recent_age = tcp_ts_getticks(); 2401 tp->ts_recent = to.to_tsval; 2402 } 2403 2404 /* 2405 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2406 * flag is on (half-synchronized state), then queue data for 2407 * later processing; else drop segment and return. 2408 */ 2409 if ((thflags & TH_ACK) == 0) { 2410 if (tp->t_state == TCPS_SYN_RECEIVED || 2411 (tp->t_flags & TF_NEEDSYN)) { 2412 if (tp->t_state == TCPS_SYN_RECEIVED && 2413 IS_FASTOPEN(tp->t_flags)) { 2414 tp->snd_wnd = tiwin; 2415 cc_conn_init(tp); 2416 } 2417 goto step6; 2418 } else if (tp->t_flags & TF_ACKNOW) 2419 goto dropafterack; 2420 else 2421 goto drop; 2422 } 2423 2424 /* 2425 * Ack processing. 2426 */ 2427 switch (tp->t_state) { 2428 /* 2429 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2430 * ESTABLISHED state and continue processing. 2431 * The ACK was checked above. 2432 */ 2433 case TCPS_SYN_RECEIVED: 2434 2435 TCPSTAT_INC(tcps_connects); 2436 soisconnected(so); 2437 /* Do window scaling? */ 2438 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2439 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2440 tp->rcv_scale = tp->request_r_scale; 2441 } 2442 tp->snd_wnd = tiwin; 2443 /* 2444 * Make transitions: 2445 * SYN-RECEIVED -> ESTABLISHED 2446 * SYN-RECEIVED* -> FIN-WAIT-1 2447 */ 2448 tp->t_starttime = ticks; 2449 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2450 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2451 tp->t_tfo_pending = NULL; 2452 } 2453 if (tp->t_flags & TF_NEEDFIN) { 2454 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2455 tp->t_flags &= ~TF_NEEDFIN; 2456 } else { 2457 tcp_state_change(tp, TCPS_ESTABLISHED); 2458 TCP_PROBE5(accept__established, NULL, tp, 2459 m, tp, th); 2460 /* 2461 * TFO connections call cc_conn_init() during SYN 2462 * processing. Calling it again here for such 2463 * connections is not harmless as it would undo the 2464 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2465 * is retransmitted. 2466 */ 2467 if (!IS_FASTOPEN(tp->t_flags)) 2468 cc_conn_init(tp); 2469 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2470 } 2471 /* 2472 * Account for the ACK of our SYN prior to 2473 * regular ACK processing below, except for 2474 * simultaneous SYN, which is handled later. 2475 */ 2476 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2477 incforsyn = 1; 2478 /* 2479 * If segment contains data or ACK, will call tcp_reass() 2480 * later; if not, do so now to pass queued data to user. 2481 */ 2482 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2483 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2484 (struct mbuf *)0); 2485 tcp_handle_wakeup(tp, so); 2486 } 2487 tp->snd_wl1 = th->th_seq - 1; 2488 /* FALLTHROUGH */ 2489 2490 /* 2491 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2492 * ACKs. If the ack is in the range 2493 * tp->snd_una < th->th_ack <= tp->snd_max 2494 * then advance tp->snd_una to th->th_ack and drop 2495 * data from the retransmission queue. If this ACK reflects 2496 * more up to date window information we update our window information. 2497 */ 2498 case TCPS_ESTABLISHED: 2499 case TCPS_FIN_WAIT_1: 2500 case TCPS_FIN_WAIT_2: 2501 case TCPS_CLOSE_WAIT: 2502 case TCPS_CLOSING: 2503 case TCPS_LAST_ACK: 2504 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2505 TCPSTAT_INC(tcps_rcvacktoomuch); 2506 goto dropafterack; 2507 } 2508 if (tcp_is_sack_recovery(tp, &to)) { 2509 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && 2510 (tp->t_flags & TF_LRD)) { 2511 tcp_sack_lost_retransmission(tp, th); 2512 } 2513 } else 2514 /* 2515 * Reset the value so that previous (valid) value 2516 * from the last ack with SACK doesn't get used. 2517 */ 2518 tp->sackhint.sacked_bytes = 0; 2519 2520 #ifdef TCP_HHOOK 2521 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2522 hhook_run_tcp_est_in(tp, th, &to); 2523 #endif 2524 2525 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2526 maxseg = tcp_maxseg(tp); 2527 if (tlen == 0 && 2528 (tiwin == tp->snd_wnd || 2529 (tp->t_flags & TF_SACK_PERMIT))) { 2530 /* 2531 * If this is the first time we've seen a 2532 * FIN from the remote, this is not a 2533 * duplicate and it needs to be processed 2534 * normally. This happens during a 2535 * simultaneous close. 2536 */ 2537 if ((thflags & TH_FIN) && 2538 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2539 tp->t_dupacks = 0; 2540 break; 2541 } 2542 TCPSTAT_INC(tcps_rcvdupack); 2543 /* 2544 * If we have outstanding data (other than 2545 * a window probe), this is a completely 2546 * duplicate ack (ie, window info didn't 2547 * change and FIN isn't set), 2548 * the ack is the biggest we've 2549 * seen and we've seen exactly our rexmt 2550 * threshold of them, assume a packet 2551 * has been dropped and retransmit it. 2552 * Kludge snd_nxt & the congestion 2553 * window so we send only this one 2554 * packet. 2555 * 2556 * We know we're losing at the current 2557 * window size so do congestion avoidance 2558 * (set ssthresh to half the current window 2559 * and pull our congestion window back to 2560 * the new ssthresh). 2561 * 2562 * Dup acks mean that packets have left the 2563 * network (they're now cached at the receiver) 2564 * so bump cwnd by the amount in the receiver 2565 * to keep a constant cwnd packets in the 2566 * network. 2567 * 2568 * When using TCP ECN, notify the peer that 2569 * we reduced the cwnd. 2570 */ 2571 /* 2572 * Following 2 kinds of acks should not affect 2573 * dupack counting: 2574 * 1) Old acks 2575 * 2) Acks with SACK but without any new SACK 2576 * information in them. These could result from 2577 * any anomaly in the network like a switch 2578 * duplicating packets or a possible DoS attack. 2579 */ 2580 if (th->th_ack != tp->snd_una || 2581 (tcp_is_sack_recovery(tp, &to) && 2582 !sack_changed)) 2583 break; 2584 else if (!tcp_timer_active(tp, TT_REXMT)) 2585 tp->t_dupacks = 0; 2586 else if (++tp->t_dupacks > tcprexmtthresh || 2587 IN_FASTRECOVERY(tp->t_flags)) { 2588 cc_ack_received(tp, th, nsegs, 2589 CC_DUPACK); 2590 if (V_tcp_do_prr && 2591 IN_FASTRECOVERY(tp->t_flags)) { 2592 tcp_do_prr_ack(tp, th, &to); 2593 } else if (tcp_is_sack_recovery(tp, &to) && 2594 IN_FASTRECOVERY(tp->t_flags)) { 2595 int awnd; 2596 2597 /* 2598 * Compute the amount of data in flight first. 2599 * We can inject new data into the pipe iff 2600 * we have less than 1/2 the original window's 2601 * worth of data in flight. 2602 */ 2603 if (V_tcp_do_newsack) 2604 awnd = tcp_compute_pipe(tp); 2605 else 2606 awnd = (tp->snd_nxt - tp->snd_fack) + 2607 tp->sackhint.sack_bytes_rexmit; 2608 2609 if (awnd < tp->snd_ssthresh) { 2610 tp->snd_cwnd += maxseg; 2611 if (tp->snd_cwnd > tp->snd_ssthresh) 2612 tp->snd_cwnd = tp->snd_ssthresh; 2613 } 2614 } else 2615 tp->snd_cwnd += maxseg; 2616 (void) tcp_output(tp); 2617 goto drop; 2618 } else if (tp->t_dupacks == tcprexmtthresh || 2619 (tp->t_flags & TF_SACK_PERMIT && 2620 V_tcp_do_newsack && 2621 tp->sackhint.sacked_bytes > 2622 (tcprexmtthresh - 1) * maxseg)) { 2623 enter_recovery: 2624 /* 2625 * Above is the RFC6675 trigger condition of 2626 * more than (dupthresh-1)*maxseg sacked data. 2627 * If the count of holes in the 2628 * scoreboard is >= dupthresh, we could 2629 * also enter loss recovery, but don't 2630 * have that value readily available. 2631 */ 2632 tp->t_dupacks = tcprexmtthresh; 2633 tcp_seq onxt = tp->snd_nxt; 2634 2635 /* 2636 * If we're doing sack, or prr, check 2637 * to see if we're already in sack 2638 * recovery. If we're not doing sack, 2639 * check to see if we're in newreno 2640 * recovery. 2641 */ 2642 if (V_tcp_do_prr || 2643 (tp->t_flags & TF_SACK_PERMIT)) { 2644 if (IN_FASTRECOVERY(tp->t_flags)) { 2645 tp->t_dupacks = 0; 2646 break; 2647 } 2648 } else { 2649 if (SEQ_LEQ(th->th_ack, 2650 tp->snd_recover)) { 2651 tp->t_dupacks = 0; 2652 break; 2653 } 2654 } 2655 /* Congestion signal before ack. */ 2656 cc_cong_signal(tp, th, CC_NDUPACK); 2657 cc_ack_received(tp, th, nsegs, 2658 CC_DUPACK); 2659 tcp_timer_activate(tp, TT_REXMT, 0); 2660 tp->t_rtttime = 0; 2661 if (V_tcp_do_prr) { 2662 /* 2663 * snd_ssthresh is already updated by 2664 * cc_cong_signal. 2665 */ 2666 if (tcp_is_sack_recovery(tp, &to)) { 2667 tp->sackhint.prr_delivered = 2668 tp->sackhint.sacked_bytes; 2669 } else { 2670 tp->sackhint.prr_delivered = 2671 imin(tp->snd_max - tp->snd_una, 2672 imin(INT_MAX / 65536, 2673 tp->t_dupacks) * maxseg); 2674 } 2675 tp->sackhint.recover_fs = max(1, 2676 tp->snd_nxt - tp->snd_una); 2677 } 2678 if (tcp_is_sack_recovery(tp, &to)) { 2679 TCPSTAT_INC( 2680 tcps_sack_recovery_episode); 2681 tp->snd_recover = tp->snd_nxt; 2682 tp->snd_cwnd = maxseg; 2683 (void) tcp_output(tp); 2684 if (SEQ_GT(th->th_ack, tp->snd_una)) 2685 goto resume_partialack; 2686 goto drop; 2687 } 2688 tp->snd_nxt = th->th_ack; 2689 tp->snd_cwnd = maxseg; 2690 (void) tcp_output(tp); 2691 KASSERT(tp->snd_limited <= 2, 2692 ("%s: tp->snd_limited too big", 2693 __func__)); 2694 tp->snd_cwnd = tp->snd_ssthresh + 2695 maxseg * 2696 (tp->t_dupacks - tp->snd_limited); 2697 if (SEQ_GT(onxt, tp->snd_nxt)) 2698 tp->snd_nxt = onxt; 2699 goto drop; 2700 } else if (V_tcp_do_rfc3042) { 2701 /* 2702 * Process first and second duplicate 2703 * ACKs. Each indicates a segment 2704 * leaving the network, creating room 2705 * for more. Make sure we can send a 2706 * packet on reception of each duplicate 2707 * ACK by increasing snd_cwnd by one 2708 * segment. Restore the original 2709 * snd_cwnd after packet transmission. 2710 */ 2711 cc_ack_received(tp, th, nsegs, 2712 CC_DUPACK); 2713 uint32_t oldcwnd = tp->snd_cwnd; 2714 tcp_seq oldsndmax = tp->snd_max; 2715 u_int sent; 2716 int avail; 2717 2718 KASSERT(tp->t_dupacks == 1 || 2719 tp->t_dupacks == 2, 2720 ("%s: dupacks not 1 or 2", 2721 __func__)); 2722 if (tp->t_dupacks == 1) 2723 tp->snd_limited = 0; 2724 tp->snd_cwnd = 2725 (tp->snd_nxt - tp->snd_una) + 2726 (tp->t_dupacks - tp->snd_limited) * 2727 maxseg; 2728 /* 2729 * Only call tcp_output when there 2730 * is new data available to be sent. 2731 * Otherwise we would send pure ACKs. 2732 */ 2733 SOCKBUF_LOCK(&so->so_snd); 2734 avail = sbavail(&so->so_snd) - 2735 (tp->snd_nxt - tp->snd_una); 2736 SOCKBUF_UNLOCK(&so->so_snd); 2737 if (avail > 0) 2738 (void) tcp_output(tp); 2739 sent = tp->snd_max - oldsndmax; 2740 if (sent > maxseg) { 2741 KASSERT((tp->t_dupacks == 2 && 2742 tp->snd_limited == 0) || 2743 (sent == maxseg + 1 && 2744 tp->t_flags & TF_SENTFIN), 2745 ("%s: sent too much", 2746 __func__)); 2747 tp->snd_limited = 2; 2748 } else if (sent > 0) 2749 ++tp->snd_limited; 2750 tp->snd_cwnd = oldcwnd; 2751 goto drop; 2752 } 2753 } 2754 break; 2755 } else { 2756 /* 2757 * This ack is advancing the left edge, reset the 2758 * counter. 2759 */ 2760 tp->t_dupacks = 0; 2761 /* 2762 * If this ack also has new SACK info, increment the 2763 * counter as per rfc6675. The variable 2764 * sack_changed tracks all changes to the SACK 2765 * scoreboard, including when partial ACKs without 2766 * SACK options are received, and clear the scoreboard 2767 * from the left side. Such partial ACKs should not be 2768 * counted as dupacks here. 2769 */ 2770 if (tcp_is_sack_recovery(tp, &to) && 2771 sack_changed) { 2772 tp->t_dupacks++; 2773 /* limit overhead by setting maxseg last */ 2774 if (!IN_FASTRECOVERY(tp->t_flags) && 2775 (tp->sackhint.sacked_bytes > 2776 ((tcprexmtthresh - 1) * 2777 (maxseg = tcp_maxseg(tp))))) { 2778 goto enter_recovery; 2779 } 2780 } 2781 } 2782 2783 resume_partialack: 2784 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2785 ("%s: th_ack <= snd_una", __func__)); 2786 2787 /* 2788 * If the congestion window was inflated to account 2789 * for the other side's cached packets, retract it. 2790 */ 2791 if (IN_FASTRECOVERY(tp->t_flags)) { 2792 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2793 if (tp->t_flags & TF_SACK_PERMIT) 2794 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2795 tcp_timer_activate(tp, TT_REXMT, 0); 2796 tp->t_rtttime = 0; 2797 tcp_do_prr_ack(tp, th, &to); 2798 tp->t_flags |= TF_ACKNOW; 2799 (void) tcp_output(tp); 2800 } else 2801 tcp_sack_partialack(tp, th); 2802 else 2803 tcp_newreno_partial_ack(tp, th); 2804 } else 2805 cc_post_recovery(tp, th); 2806 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2807 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2808 if (V_tcp_do_prr) { 2809 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2810 tp->snd_fack = th->th_ack; 2811 tcp_do_prr_ack(tp, th, &to); 2812 (void) tcp_output(tp); 2813 } 2814 } else 2815 cc_post_recovery(tp, th); 2816 } 2817 /* 2818 * If we reach this point, ACK is not a duplicate, 2819 * i.e., it ACKs something we sent. 2820 */ 2821 if (tp->t_flags & TF_NEEDSYN) { 2822 /* 2823 * T/TCP: Connection was half-synchronized, and our 2824 * SYN has been ACK'd (so connection is now fully 2825 * synchronized). Go to non-starred state, 2826 * increment snd_una for ACK of SYN, and check if 2827 * we can do window scaling. 2828 */ 2829 tp->t_flags &= ~TF_NEEDSYN; 2830 tp->snd_una++; 2831 /* Do window scaling? */ 2832 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2833 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2834 tp->rcv_scale = tp->request_r_scale; 2835 /* Send window already scaled. */ 2836 } 2837 } 2838 2839 process_ACK: 2840 INP_WLOCK_ASSERT(tp->t_inpcb); 2841 2842 /* 2843 * Adjust for the SYN bit in sequence space, 2844 * but don't account for it in cwnd calculations. 2845 * This is for the SYN_RECEIVED, non-simultaneous 2846 * SYN case. SYN_SENT and simultaneous SYN are 2847 * treated elsewhere. 2848 */ 2849 if (incforsyn) 2850 tp->snd_una++; 2851 acked = BYTES_THIS_ACK(tp, th); 2852 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2853 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2854 tp->snd_una, th->th_ack, tp, m)); 2855 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2856 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2857 2858 /* 2859 * If we just performed our first retransmit, and the ACK 2860 * arrives within our recovery window, then it was a mistake 2861 * to do the retransmit in the first place. Recover our 2862 * original cwnd and ssthresh, and proceed to transmit where 2863 * we left off. 2864 */ 2865 if (tp->t_rxtshift == 1 && 2866 tp->t_flags & TF_PREVVALID && 2867 tp->t_badrxtwin != 0 && 2868 to.to_flags & TOF_TS && 2869 to.to_tsecr != 0 && 2870 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2871 cc_cong_signal(tp, th, CC_RTO_ERR); 2872 2873 /* 2874 * If we have a timestamp reply, update smoothed 2875 * round trip time. If no timestamp is present but 2876 * transmit timer is running and timed sequence 2877 * number was acked, update smoothed round trip time. 2878 * Since we now have an rtt measurement, cancel the 2879 * timer backoff (cf., Phil Karn's retransmit alg.). 2880 * Recompute the initial retransmit timer. 2881 * 2882 * Some boxes send broken timestamp replies 2883 * during the SYN+ACK phase, ignore 2884 * timestamps of 0 or we could calculate a 2885 * huge RTT and blow up the retransmit timer. 2886 */ 2887 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2888 uint32_t t; 2889 2890 t = tcp_ts_getticks() - to.to_tsecr; 2891 if (!tp->t_rttlow || tp->t_rttlow > t) 2892 tp->t_rttlow = t; 2893 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2894 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2895 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2896 tp->t_rttlow = ticks - tp->t_rtttime; 2897 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2898 } 2899 2900 /* 2901 * If all outstanding data is acked, stop retransmit 2902 * timer and remember to restart (more output or persist). 2903 * If there is more data to be acked, restart retransmit 2904 * timer, using current (possibly backed-off) value. 2905 */ 2906 if (th->th_ack == tp->snd_max) { 2907 tcp_timer_activate(tp, TT_REXMT, 0); 2908 needoutput = 1; 2909 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2910 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2911 2912 /* 2913 * If no data (only SYN) was ACK'd, 2914 * skip rest of ACK processing. 2915 */ 2916 if (acked == 0) 2917 goto step6; 2918 2919 /* 2920 * Let the congestion control algorithm update congestion 2921 * control related information. This typically means increasing 2922 * the congestion window. 2923 */ 2924 cc_ack_received(tp, th, nsegs, CC_ACK); 2925 2926 SOCKBUF_LOCK(&so->so_snd); 2927 if (acked > sbavail(&so->so_snd)) { 2928 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2929 tp->snd_wnd -= sbavail(&so->so_snd); 2930 else 2931 tp->snd_wnd = 0; 2932 mfree = sbcut_locked(&so->so_snd, 2933 (int)sbavail(&so->so_snd)); 2934 ourfinisacked = 1; 2935 } else { 2936 mfree = sbcut_locked(&so->so_snd, acked); 2937 if (tp->snd_wnd >= (uint32_t) acked) 2938 tp->snd_wnd -= acked; 2939 else 2940 tp->snd_wnd = 0; 2941 ourfinisacked = 0; 2942 } 2943 /* NB: sowwakeup_locked() does an implicit unlock. */ 2944 sowwakeup_locked(so); 2945 m_freem(mfree); 2946 /* Detect una wraparound. */ 2947 if (!IN_RECOVERY(tp->t_flags) && 2948 SEQ_GT(tp->snd_una, tp->snd_recover) && 2949 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2950 tp->snd_recover = th->th_ack - 1; 2951 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2952 if (IN_RECOVERY(tp->t_flags) && 2953 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2954 EXIT_RECOVERY(tp->t_flags); 2955 } 2956 tp->snd_una = th->th_ack; 2957 if (tp->t_flags & TF_SACK_PERMIT) { 2958 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2959 tp->snd_recover = tp->snd_una; 2960 } 2961 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2962 tp->snd_nxt = tp->snd_una; 2963 2964 switch (tp->t_state) { 2965 /* 2966 * In FIN_WAIT_1 STATE in addition to the processing 2967 * for the ESTABLISHED state if our FIN is now acknowledged 2968 * then enter FIN_WAIT_2. 2969 */ 2970 case TCPS_FIN_WAIT_1: 2971 if (ourfinisacked) { 2972 /* 2973 * If we can't receive any more 2974 * data, then closing user can proceed. 2975 * Starting the timer is contrary to the 2976 * specification, but if we don't get a FIN 2977 * we'll hang forever. 2978 * 2979 * XXXjl: 2980 * we should release the tp also, and use a 2981 * compressed state. 2982 */ 2983 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2984 soisdisconnected(so); 2985 tcp_timer_activate(tp, TT_2MSL, 2986 (tcp_fast_finwait2_recycle ? 2987 tcp_finwait2_timeout : 2988 TP_MAXIDLE(tp))); 2989 } 2990 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2991 } 2992 break; 2993 2994 /* 2995 * In CLOSING STATE in addition to the processing for 2996 * the ESTABLISHED state if the ACK acknowledges our FIN 2997 * then enter the TIME-WAIT state, otherwise ignore 2998 * the segment. 2999 */ 3000 case TCPS_CLOSING: 3001 if (ourfinisacked) { 3002 tcp_twstart(tp); 3003 m_freem(m); 3004 return; 3005 } 3006 break; 3007 3008 /* 3009 * In LAST_ACK, we may still be waiting for data to drain 3010 * and/or to be acked, as well as for the ack of our FIN. 3011 * If our FIN is now acknowledged, delete the TCB, 3012 * enter the closed state and return. 3013 */ 3014 case TCPS_LAST_ACK: 3015 if (ourfinisacked) { 3016 tp = tcp_close(tp); 3017 goto drop; 3018 } 3019 break; 3020 } 3021 } 3022 3023 step6: 3024 INP_WLOCK_ASSERT(tp->t_inpcb); 3025 3026 /* 3027 * Update window information. 3028 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3029 */ 3030 if ((thflags & TH_ACK) && 3031 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3032 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3033 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3034 /* keep track of pure window updates */ 3035 if (tlen == 0 && 3036 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3037 TCPSTAT_INC(tcps_rcvwinupd); 3038 tp->snd_wnd = tiwin; 3039 tp->snd_wl1 = th->th_seq; 3040 tp->snd_wl2 = th->th_ack; 3041 if (tp->snd_wnd > tp->max_sndwnd) 3042 tp->max_sndwnd = tp->snd_wnd; 3043 needoutput = 1; 3044 } 3045 3046 /* 3047 * Process segments with URG. 3048 */ 3049 if ((thflags & TH_URG) && th->th_urp && 3050 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3051 /* 3052 * This is a kludge, but if we receive and accept 3053 * random urgent pointers, we'll crash in 3054 * soreceive. It's hard to imagine someone 3055 * actually wanting to send this much urgent data. 3056 */ 3057 SOCKBUF_LOCK(&so->so_rcv); 3058 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3059 th->th_urp = 0; /* XXX */ 3060 thflags &= ~TH_URG; /* XXX */ 3061 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3062 goto dodata; /* XXX */ 3063 } 3064 /* 3065 * If this segment advances the known urgent pointer, 3066 * then mark the data stream. This should not happen 3067 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3068 * a FIN has been received from the remote side. 3069 * In these states we ignore the URG. 3070 * 3071 * According to RFC961 (Assigned Protocols), 3072 * the urgent pointer points to the last octet 3073 * of urgent data. We continue, however, 3074 * to consider it to indicate the first octet 3075 * of data past the urgent section as the original 3076 * spec states (in one of two places). 3077 */ 3078 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3079 tp->rcv_up = th->th_seq + th->th_urp; 3080 so->so_oobmark = sbavail(&so->so_rcv) + 3081 (tp->rcv_up - tp->rcv_nxt) - 1; 3082 if (so->so_oobmark == 0) 3083 so->so_rcv.sb_state |= SBS_RCVATMARK; 3084 sohasoutofband(so); 3085 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3086 } 3087 SOCKBUF_UNLOCK(&so->so_rcv); 3088 /* 3089 * Remove out of band data so doesn't get presented to user. 3090 * This can happen independent of advancing the URG pointer, 3091 * but if two URG's are pending at once, some out-of-band 3092 * data may creep in... ick. 3093 */ 3094 if (th->th_urp <= (uint32_t)tlen && 3095 !(so->so_options & SO_OOBINLINE)) { 3096 /* hdr drop is delayed */ 3097 tcp_pulloutofband(so, th, m, drop_hdrlen); 3098 } 3099 } else { 3100 /* 3101 * If no out of band data is expected, 3102 * pull receive urgent pointer along 3103 * with the receive window. 3104 */ 3105 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3106 tp->rcv_up = tp->rcv_nxt; 3107 } 3108 dodata: /* XXX */ 3109 INP_WLOCK_ASSERT(tp->t_inpcb); 3110 3111 /* 3112 * Process the segment text, merging it into the TCP sequencing queue, 3113 * and arranging for acknowledgment of receipt if necessary. 3114 * This process logically involves adjusting tp->rcv_wnd as data 3115 * is presented to the user (this happens in tcp_usrreq.c, 3116 * case PRU_RCVD). If a FIN has already been received on this 3117 * connection then we just ignore the text. 3118 */ 3119 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3120 IS_FASTOPEN(tp->t_flags)); 3121 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3122 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3123 tcp_seq save_start = th->th_seq; 3124 tcp_seq save_rnxt = tp->rcv_nxt; 3125 int save_tlen = tlen; 3126 m_adj(m, drop_hdrlen); /* delayed header drop */ 3127 /* 3128 * Insert segment which includes th into TCP reassembly queue 3129 * with control block tp. Set thflags to whether reassembly now 3130 * includes a segment with FIN. This handles the common case 3131 * inline (segment is the next to be received on an established 3132 * connection, and the queue is empty), avoiding linkage into 3133 * and removal from the queue and repetition of various 3134 * conversions. 3135 * Set DELACK for segments received in order, but ack 3136 * immediately when segments are out of order (so 3137 * fast retransmit can work). 3138 */ 3139 if (th->th_seq == tp->rcv_nxt && 3140 SEGQ_EMPTY(tp) && 3141 (TCPS_HAVEESTABLISHED(tp->t_state) || 3142 tfo_syn)) { 3143 if (DELAY_ACK(tp, tlen) || tfo_syn) 3144 tp->t_flags |= TF_DELACK; 3145 else 3146 tp->t_flags |= TF_ACKNOW; 3147 tp->rcv_nxt += tlen; 3148 if (tlen && 3149 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3150 (tp->t_fbyte_in == 0)) { 3151 tp->t_fbyte_in = ticks; 3152 if (tp->t_fbyte_in == 0) 3153 tp->t_fbyte_in = 1; 3154 if (tp->t_fbyte_out && tp->t_fbyte_in) 3155 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3156 } 3157 thflags = tcp_get_flags(th) & TH_FIN; 3158 TCPSTAT_INC(tcps_rcvpack); 3159 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3160 SOCKBUF_LOCK(&so->so_rcv); 3161 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3162 m_freem(m); 3163 else 3164 sbappendstream_locked(&so->so_rcv, m, 0); 3165 tp->t_flags |= TF_WAKESOR; 3166 } else { 3167 /* 3168 * XXX: Due to the header drop above "th" is 3169 * theoretically invalid by now. Fortunately 3170 * m_adj() doesn't actually frees any mbufs 3171 * when trimming from the head. 3172 */ 3173 tcp_seq temp = save_start; 3174 3175 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3176 tp->t_flags |= TF_ACKNOW; 3177 } 3178 if ((tp->t_flags & TF_SACK_PERMIT) && 3179 (save_tlen > 0) && 3180 TCPS_HAVEESTABLISHED(tp->t_state)) { 3181 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3182 /* 3183 * DSACK actually handled in the fastpath 3184 * above. 3185 */ 3186 tcp_update_sack_list(tp, save_start, 3187 save_start + save_tlen); 3188 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3189 if ((tp->rcv_numsacks >= 1) && 3190 (tp->sackblks[0].end == save_start)) { 3191 /* 3192 * Partial overlap, recorded at todrop 3193 * above. 3194 */ 3195 tcp_update_sack_list(tp, 3196 tp->sackblks[0].start, 3197 tp->sackblks[0].end); 3198 } else { 3199 tcp_update_dsack_list(tp, save_start, 3200 save_start + save_tlen); 3201 } 3202 } else if (tlen >= save_tlen) { 3203 /* Update of sackblks. */ 3204 tcp_update_dsack_list(tp, save_start, 3205 save_start + save_tlen); 3206 } else if (tlen > 0) { 3207 tcp_update_dsack_list(tp, save_start, 3208 save_start + tlen); 3209 } 3210 } 3211 tcp_handle_wakeup(tp, so); 3212 #if 0 3213 /* 3214 * Note the amount of data that peer has sent into 3215 * our window, in order to estimate the sender's 3216 * buffer size. 3217 * XXX: Unused. 3218 */ 3219 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3220 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3221 else 3222 len = so->so_rcv.sb_hiwat; 3223 #endif 3224 } else { 3225 m_freem(m); 3226 thflags &= ~TH_FIN; 3227 } 3228 3229 /* 3230 * If FIN is received ACK the FIN and let the user know 3231 * that the connection is closing. 3232 */ 3233 if (thflags & TH_FIN) { 3234 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3235 /* The socket upcall is handled by socantrcvmore. */ 3236 socantrcvmore(so); 3237 /* 3238 * If connection is half-synchronized 3239 * (ie NEEDSYN flag on) then delay ACK, 3240 * so it may be piggybacked when SYN is sent. 3241 * Otherwise, since we received a FIN then no 3242 * more input can be expected, send ACK now. 3243 */ 3244 if (tp->t_flags & TF_NEEDSYN) 3245 tp->t_flags |= TF_DELACK; 3246 else 3247 tp->t_flags |= TF_ACKNOW; 3248 tp->rcv_nxt++; 3249 } 3250 switch (tp->t_state) { 3251 /* 3252 * In SYN_RECEIVED and ESTABLISHED STATES 3253 * enter the CLOSE_WAIT state. 3254 */ 3255 case TCPS_SYN_RECEIVED: 3256 tp->t_starttime = ticks; 3257 /* FALLTHROUGH */ 3258 case TCPS_ESTABLISHED: 3259 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3260 break; 3261 3262 /* 3263 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3264 * enter the CLOSING state. 3265 */ 3266 case TCPS_FIN_WAIT_1: 3267 tcp_state_change(tp, TCPS_CLOSING); 3268 break; 3269 3270 /* 3271 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3272 * starting the time-wait timer, turning off the other 3273 * standard timers. 3274 */ 3275 case TCPS_FIN_WAIT_2: 3276 tcp_twstart(tp); 3277 return; 3278 } 3279 } 3280 #ifdef TCPDEBUG 3281 if (so->so_options & SO_DEBUG) 3282 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3283 &tcp_savetcp, 0); 3284 #endif 3285 TCP_PROBE3(debug__input, tp, th, m); 3286 3287 /* 3288 * Return any desired output. 3289 */ 3290 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3291 (void) tcp_output(tp); 3292 3293 check_delack: 3294 INP_WLOCK_ASSERT(tp->t_inpcb); 3295 3296 if (tp->t_flags & TF_DELACK) { 3297 tp->t_flags &= ~TF_DELACK; 3298 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3299 } 3300 INP_WUNLOCK(tp->t_inpcb); 3301 return; 3302 3303 dropafterack: 3304 /* 3305 * Generate an ACK dropping incoming segment if it occupies 3306 * sequence space, where the ACK reflects our state. 3307 * 3308 * We can now skip the test for the RST flag since all 3309 * paths to this code happen after packets containing 3310 * RST have been dropped. 3311 * 3312 * In the SYN-RECEIVED state, don't send an ACK unless the 3313 * segment we received passes the SYN-RECEIVED ACK test. 3314 * If it fails send a RST. This breaks the loop in the 3315 * "LAND" DoS attack, and also prevents an ACK storm 3316 * between two listening ports that have been sent forged 3317 * SYN segments, each with the source address of the other. 3318 */ 3319 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3320 (SEQ_GT(tp->snd_una, th->th_ack) || 3321 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3322 rstreason = BANDLIM_RST_OPENPORT; 3323 goto dropwithreset; 3324 } 3325 #ifdef TCPDEBUG 3326 if (so->so_options & SO_DEBUG) 3327 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3328 &tcp_savetcp, 0); 3329 #endif 3330 TCP_PROBE3(debug__input, tp, th, m); 3331 tp->t_flags |= TF_ACKNOW; 3332 (void) tcp_output(tp); 3333 INP_WUNLOCK(tp->t_inpcb); 3334 m_freem(m); 3335 return; 3336 3337 dropwithreset: 3338 if (tp != NULL) { 3339 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3340 INP_WUNLOCK(tp->t_inpcb); 3341 } else 3342 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3343 return; 3344 3345 drop: 3346 /* 3347 * Drop space held by incoming segment and return. 3348 */ 3349 #ifdef TCPDEBUG 3350 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3351 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3352 &tcp_savetcp, 0); 3353 #endif 3354 TCP_PROBE3(debug__input, tp, th, m); 3355 if (tp != NULL) { 3356 INP_WUNLOCK(tp->t_inpcb); 3357 } 3358 m_freem(m); 3359 } 3360 3361 /* 3362 * Issue RST and make ACK acceptable to originator of segment. 3363 * The mbuf must still include the original packet header. 3364 * tp may be NULL. 3365 */ 3366 void 3367 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3368 int tlen, int rstreason) 3369 { 3370 #ifdef INET 3371 struct ip *ip; 3372 #endif 3373 #ifdef INET6 3374 struct ip6_hdr *ip6; 3375 #endif 3376 3377 if (tp != NULL) { 3378 INP_LOCK_ASSERT(tp->t_inpcb); 3379 } 3380 3381 /* Don't bother if destination was broadcast/multicast. */ 3382 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3383 goto drop; 3384 #ifdef INET6 3385 if (mtod(m, struct ip *)->ip_v == 6) { 3386 ip6 = mtod(m, struct ip6_hdr *); 3387 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3388 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3389 goto drop; 3390 /* IPv6 anycast check is done at tcp6_input() */ 3391 } 3392 #endif 3393 #if defined(INET) && defined(INET6) 3394 else 3395 #endif 3396 #ifdef INET 3397 { 3398 ip = mtod(m, struct ip *); 3399 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3400 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3401 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3402 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3403 goto drop; 3404 } 3405 #endif 3406 3407 /* Perform bandwidth limiting. */ 3408 if (badport_bandlim(rstreason) < 0) 3409 goto drop; 3410 3411 /* tcp_respond consumes the mbuf chain. */ 3412 if (tcp_get_flags(th) & TH_ACK) { 3413 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3414 th->th_ack, TH_RST); 3415 } else { 3416 if (tcp_get_flags(th) & TH_SYN) 3417 tlen++; 3418 if (tcp_get_flags(th) & TH_FIN) 3419 tlen++; 3420 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3421 (tcp_seq)0, TH_RST|TH_ACK); 3422 } 3423 return; 3424 drop: 3425 m_freem(m); 3426 } 3427 3428 /* 3429 * Parse TCP options and place in tcpopt. 3430 */ 3431 void 3432 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3433 { 3434 int opt, optlen; 3435 3436 to->to_flags = 0; 3437 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3438 opt = cp[0]; 3439 if (opt == TCPOPT_EOL) 3440 break; 3441 if (opt == TCPOPT_NOP) 3442 optlen = 1; 3443 else { 3444 if (cnt < 2) 3445 break; 3446 optlen = cp[1]; 3447 if (optlen < 2 || optlen > cnt) 3448 break; 3449 } 3450 switch (opt) { 3451 case TCPOPT_MAXSEG: 3452 if (optlen != TCPOLEN_MAXSEG) 3453 continue; 3454 if (!(flags & TO_SYN)) 3455 continue; 3456 to->to_flags |= TOF_MSS; 3457 bcopy((char *)cp + 2, 3458 (char *)&to->to_mss, sizeof(to->to_mss)); 3459 to->to_mss = ntohs(to->to_mss); 3460 break; 3461 case TCPOPT_WINDOW: 3462 if (optlen != TCPOLEN_WINDOW) 3463 continue; 3464 if (!(flags & TO_SYN)) 3465 continue; 3466 to->to_flags |= TOF_SCALE; 3467 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3468 break; 3469 case TCPOPT_TIMESTAMP: 3470 if (optlen != TCPOLEN_TIMESTAMP) 3471 continue; 3472 to->to_flags |= TOF_TS; 3473 bcopy((char *)cp + 2, 3474 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3475 to->to_tsval = ntohl(to->to_tsval); 3476 bcopy((char *)cp + 6, 3477 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3478 to->to_tsecr = ntohl(to->to_tsecr); 3479 break; 3480 case TCPOPT_SIGNATURE: 3481 /* 3482 * In order to reply to a host which has set the 3483 * TCP_SIGNATURE option in its initial SYN, we have 3484 * to record the fact that the option was observed 3485 * here for the syncache code to perform the correct 3486 * response. 3487 */ 3488 if (optlen != TCPOLEN_SIGNATURE) 3489 continue; 3490 to->to_flags |= TOF_SIGNATURE; 3491 to->to_signature = cp + 2; 3492 break; 3493 case TCPOPT_SACK_PERMITTED: 3494 if (optlen != TCPOLEN_SACK_PERMITTED) 3495 continue; 3496 if (!(flags & TO_SYN)) 3497 continue; 3498 if (!V_tcp_do_sack) 3499 continue; 3500 to->to_flags |= TOF_SACKPERM; 3501 break; 3502 case TCPOPT_SACK: 3503 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3504 continue; 3505 if (flags & TO_SYN) 3506 continue; 3507 to->to_flags |= TOF_SACK; 3508 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3509 to->to_sacks = cp + 2; 3510 TCPSTAT_INC(tcps_sack_rcv_blocks); 3511 break; 3512 case TCPOPT_FAST_OPEN: 3513 /* 3514 * Cookie length validation is performed by the 3515 * server side cookie checking code or the client 3516 * side cookie cache update code. 3517 */ 3518 if (!(flags & TO_SYN)) 3519 continue; 3520 if (!V_tcp_fastopen_client_enable && 3521 !V_tcp_fastopen_server_enable) 3522 continue; 3523 to->to_flags |= TOF_FASTOPEN; 3524 to->to_tfo_len = optlen - 2; 3525 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3526 break; 3527 default: 3528 continue; 3529 } 3530 } 3531 } 3532 3533 /* 3534 * Pull out of band byte out of a segment so 3535 * it doesn't appear in the user's data queue. 3536 * It is still reflected in the segment length for 3537 * sequencing purposes. 3538 */ 3539 void 3540 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3541 int off) 3542 { 3543 int cnt = off + th->th_urp - 1; 3544 3545 while (cnt >= 0) { 3546 if (m->m_len > cnt) { 3547 char *cp = mtod(m, caddr_t) + cnt; 3548 struct tcpcb *tp = sototcpcb(so); 3549 3550 INP_WLOCK_ASSERT(tp->t_inpcb); 3551 3552 tp->t_iobc = *cp; 3553 tp->t_oobflags |= TCPOOB_HAVEDATA; 3554 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3555 m->m_len--; 3556 if (m->m_flags & M_PKTHDR) 3557 m->m_pkthdr.len--; 3558 return; 3559 } 3560 cnt -= m->m_len; 3561 m = m->m_next; 3562 if (m == NULL) 3563 break; 3564 } 3565 panic("tcp_pulloutofband"); 3566 } 3567 3568 /* 3569 * Collect new round-trip time estimate 3570 * and update averages and current timeout. 3571 */ 3572 void 3573 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3574 { 3575 int delta; 3576 3577 INP_WLOCK_ASSERT(tp->t_inpcb); 3578 3579 TCPSTAT_INC(tcps_rttupdated); 3580 tp->t_rttupdated++; 3581 #ifdef STATS 3582 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3583 imax(0, rtt * 1000 / hz)); 3584 #endif 3585 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3586 /* 3587 * srtt is stored as fixed point with 5 bits after the 3588 * binary point (i.e., scaled by 8). The following magic 3589 * is equivalent to the smoothing algorithm in rfc793 with 3590 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3591 * point). Adjust rtt to origin 0. 3592 */ 3593 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3594 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3595 3596 if ((tp->t_srtt += delta) <= 0) 3597 tp->t_srtt = 1; 3598 3599 /* 3600 * We accumulate a smoothed rtt variance (actually, a 3601 * smoothed mean difference), then set the retransmit 3602 * timer to smoothed rtt + 4 times the smoothed variance. 3603 * rttvar is stored as fixed point with 4 bits after the 3604 * binary point (scaled by 16). The following is 3605 * equivalent to rfc793 smoothing with an alpha of .75 3606 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3607 * rfc793's wired-in beta. 3608 */ 3609 if (delta < 0) 3610 delta = -delta; 3611 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3612 if ((tp->t_rttvar += delta) <= 0) 3613 tp->t_rttvar = 1; 3614 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3615 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3616 } else { 3617 /* 3618 * No rtt measurement yet - use the unsmoothed rtt. 3619 * Set the variance to half the rtt (so our first 3620 * retransmit happens at 3*rtt). 3621 */ 3622 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3623 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3624 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3625 } 3626 tp->t_rtttime = 0; 3627 tp->t_rxtshift = 0; 3628 3629 /* 3630 * the retransmit should happen at rtt + 4 * rttvar. 3631 * Because of the way we do the smoothing, srtt and rttvar 3632 * will each average +1/2 tick of bias. When we compute 3633 * the retransmit timer, we want 1/2 tick of rounding and 3634 * 1 extra tick because of +-1/2 tick uncertainty in the 3635 * firing of the timer. The bias will give us exactly the 3636 * 1.5 tick we need. But, because the bias is 3637 * statistical, we have to test that we don't drop below 3638 * the minimum feasible timer (which is 2 ticks). 3639 */ 3640 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3641 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3642 3643 /* 3644 * We received an ack for a packet that wasn't retransmitted; 3645 * it is probably safe to discard any error indications we've 3646 * received recently. This isn't quite right, but close enough 3647 * for now (a route might have failed after we sent a segment, 3648 * and the return path might not be symmetrical). 3649 */ 3650 tp->t_softerror = 0; 3651 } 3652 3653 /* 3654 * Determine a reasonable value for maxseg size. 3655 * If the route is known, check route for mtu. 3656 * If none, use an mss that can be handled on the outgoing interface 3657 * without forcing IP to fragment. If no route is found, route has no mtu, 3658 * or the destination isn't local, use a default, hopefully conservative 3659 * size (usually 512 or the default IP max size, but no more than the mtu 3660 * of the interface), as we can't discover anything about intervening 3661 * gateways or networks. We also initialize the congestion/slow start 3662 * window to be a single segment if the destination isn't local. 3663 * While looking at the routing entry, we also initialize other path-dependent 3664 * parameters from pre-set or cached values in the routing entry. 3665 * 3666 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3667 * IP options, e.g. IPSEC data, since length of this data may vary, and 3668 * thus it is calculated for every segment separately in tcp_output(). 3669 * 3670 * NOTE that this routine is only called when we process an incoming 3671 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3672 * settings are handled in tcp_mssopt(). 3673 */ 3674 void 3675 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3676 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3677 { 3678 int mss = 0; 3679 uint32_t maxmtu = 0; 3680 struct inpcb *inp = tp->t_inpcb; 3681 struct hc_metrics_lite metrics; 3682 #ifdef INET6 3683 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3684 size_t min_protoh = isipv6 ? 3685 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3686 sizeof (struct tcpiphdr); 3687 #else 3688 size_t min_protoh = sizeof(struct tcpiphdr); 3689 #endif 3690 3691 INP_WLOCK_ASSERT(tp->t_inpcb); 3692 3693 if (tp->t_port) 3694 min_protoh += V_tcp_udp_tunneling_overhead; 3695 if (mtuoffer != -1) { 3696 KASSERT(offer == -1, ("%s: conflict", __func__)); 3697 offer = mtuoffer - min_protoh; 3698 } 3699 3700 /* Initialize. */ 3701 #ifdef INET6 3702 if (isipv6) { 3703 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3704 tp->t_maxseg = V_tcp_v6mssdflt; 3705 } 3706 #endif 3707 #if defined(INET) && defined(INET6) 3708 else 3709 #endif 3710 #ifdef INET 3711 { 3712 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3713 tp->t_maxseg = V_tcp_mssdflt; 3714 } 3715 #endif 3716 3717 /* 3718 * No route to sender, stay with default mss and return. 3719 */ 3720 if (maxmtu == 0) { 3721 /* 3722 * In case we return early we need to initialize metrics 3723 * to a defined state as tcp_hc_get() would do for us 3724 * if there was no cache hit. 3725 */ 3726 if (metricptr != NULL) 3727 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3728 return; 3729 } 3730 3731 /* What have we got? */ 3732 switch (offer) { 3733 case 0: 3734 /* 3735 * Offer == 0 means that there was no MSS on the SYN 3736 * segment, in this case we use tcp_mssdflt as 3737 * already assigned to t_maxseg above. 3738 */ 3739 offer = tp->t_maxseg; 3740 break; 3741 3742 case -1: 3743 /* 3744 * Offer == -1 means that we didn't receive SYN yet. 3745 */ 3746 /* FALLTHROUGH */ 3747 3748 default: 3749 /* 3750 * Prevent DoS attack with too small MSS. Round up 3751 * to at least minmss. 3752 */ 3753 offer = max(offer, V_tcp_minmss); 3754 } 3755 3756 /* 3757 * rmx information is now retrieved from tcp_hostcache. 3758 */ 3759 tcp_hc_get(&inp->inp_inc, &metrics); 3760 if (metricptr != NULL) 3761 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3762 3763 /* 3764 * If there's a discovered mtu in tcp hostcache, use it. 3765 * Else, use the link mtu. 3766 */ 3767 if (metrics.rmx_mtu) 3768 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3769 else { 3770 #ifdef INET6 3771 if (isipv6) { 3772 mss = maxmtu - min_protoh; 3773 if (!V_path_mtu_discovery && 3774 !in6_localaddr(&inp->in6p_faddr)) 3775 mss = min(mss, V_tcp_v6mssdflt); 3776 } 3777 #endif 3778 #if defined(INET) && defined(INET6) 3779 else 3780 #endif 3781 #ifdef INET 3782 { 3783 mss = maxmtu - min_protoh; 3784 if (!V_path_mtu_discovery && 3785 !in_localaddr(inp->inp_faddr)) 3786 mss = min(mss, V_tcp_mssdflt); 3787 } 3788 #endif 3789 /* 3790 * XXX - The above conditional (mss = maxmtu - min_protoh) 3791 * probably violates the TCP spec. 3792 * The problem is that, since we don't know the 3793 * other end's MSS, we are supposed to use a conservative 3794 * default. But, if we do that, then MTU discovery will 3795 * never actually take place, because the conservative 3796 * default is much less than the MTUs typically seen 3797 * on the Internet today. For the moment, we'll sweep 3798 * this under the carpet. 3799 * 3800 * The conservative default might not actually be a problem 3801 * if the only case this occurs is when sending an initial 3802 * SYN with options and data to a host we've never talked 3803 * to before. Then, they will reply with an MSS value which 3804 * will get recorded and the new parameters should get 3805 * recomputed. For Further Study. 3806 */ 3807 } 3808 mss = min(mss, offer); 3809 3810 /* 3811 * Sanity check: make sure that maxseg will be large 3812 * enough to allow some data on segments even if the 3813 * all the option space is used (40bytes). Otherwise 3814 * funny things may happen in tcp_output. 3815 * 3816 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3817 */ 3818 mss = max(mss, 64); 3819 3820 tp->t_maxseg = mss; 3821 } 3822 3823 void 3824 tcp_mss(struct tcpcb *tp, int offer) 3825 { 3826 int mss; 3827 uint32_t bufsize; 3828 struct inpcb *inp; 3829 struct socket *so; 3830 struct hc_metrics_lite metrics; 3831 struct tcp_ifcap cap; 3832 3833 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3834 3835 bzero(&cap, sizeof(cap)); 3836 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3837 3838 mss = tp->t_maxseg; 3839 inp = tp->t_inpcb; 3840 3841 /* 3842 * If there's a pipesize, change the socket buffer to that size, 3843 * don't change if sb_hiwat is different than default (then it 3844 * has been changed on purpose with setsockopt). 3845 * Make the socket buffers an integral number of mss units; 3846 * if the mss is larger than the socket buffer, decrease the mss. 3847 */ 3848 so = inp->inp_socket; 3849 SOCKBUF_LOCK(&so->so_snd); 3850 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3851 bufsize = metrics.rmx_sendpipe; 3852 else 3853 bufsize = so->so_snd.sb_hiwat; 3854 if (bufsize < mss) 3855 mss = bufsize; 3856 else { 3857 bufsize = roundup(bufsize, mss); 3858 if (bufsize > sb_max) 3859 bufsize = sb_max; 3860 if (bufsize > so->so_snd.sb_hiwat) 3861 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3862 } 3863 SOCKBUF_UNLOCK(&so->so_snd); 3864 /* 3865 * Sanity check: make sure that maxseg will be large 3866 * enough to allow some data on segments even if the 3867 * all the option space is used (40bytes). Otherwise 3868 * funny things may happen in tcp_output. 3869 * 3870 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3871 */ 3872 tp->t_maxseg = max(mss, 64); 3873 3874 SOCKBUF_LOCK(&so->so_rcv); 3875 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3876 bufsize = metrics.rmx_recvpipe; 3877 else 3878 bufsize = so->so_rcv.sb_hiwat; 3879 if (bufsize > mss) { 3880 bufsize = roundup(bufsize, mss); 3881 if (bufsize > sb_max) 3882 bufsize = sb_max; 3883 if (bufsize > so->so_rcv.sb_hiwat) 3884 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3885 } 3886 SOCKBUF_UNLOCK(&so->so_rcv); 3887 3888 /* Check the interface for TSO capabilities. */ 3889 if (cap.ifcap & CSUM_TSO) { 3890 tp->t_flags |= TF_TSO; 3891 tp->t_tsomax = cap.tsomax; 3892 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3893 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3894 } 3895 } 3896 3897 /* 3898 * Determine the MSS option to send on an outgoing SYN. 3899 */ 3900 int 3901 tcp_mssopt(struct in_conninfo *inc) 3902 { 3903 int mss = 0; 3904 uint32_t thcmtu = 0; 3905 uint32_t maxmtu = 0; 3906 size_t min_protoh; 3907 3908 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3909 3910 #ifdef INET6 3911 if (inc->inc_flags & INC_ISIPV6) { 3912 mss = V_tcp_v6mssdflt; 3913 maxmtu = tcp_maxmtu6(inc, NULL); 3914 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3915 } 3916 #endif 3917 #if defined(INET) && defined(INET6) 3918 else 3919 #endif 3920 #ifdef INET 3921 { 3922 mss = V_tcp_mssdflt; 3923 maxmtu = tcp_maxmtu(inc, NULL); 3924 min_protoh = sizeof(struct tcpiphdr); 3925 } 3926 #endif 3927 #if defined(INET6) || defined(INET) 3928 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3929 #endif 3930 3931 if (maxmtu && thcmtu) 3932 mss = min(maxmtu, thcmtu) - min_protoh; 3933 else if (maxmtu || thcmtu) 3934 mss = max(maxmtu, thcmtu) - min_protoh; 3935 3936 return (mss); 3937 } 3938 3939 void 3940 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3941 { 3942 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3943 int maxseg = tcp_maxseg(tp); 3944 3945 INP_WLOCK_ASSERT(tp->t_inpcb); 3946 3947 /* 3948 * Compute the amount of data that this ACK is indicating 3949 * (del_data) and an estimate of how many bytes are in the 3950 * network. 3951 */ 3952 if (tcp_is_sack_recovery(tp, to) || 3953 (IN_CONGRECOVERY(tp->t_flags) && 3954 !IN_FASTRECOVERY(tp->t_flags))) { 3955 del_data = tp->sackhint.delivered_data; 3956 if (V_tcp_do_newsack) 3957 pipe = tcp_compute_pipe(tp); 3958 else 3959 pipe = (tp->snd_nxt - tp->snd_fack) + 3960 tp->sackhint.sack_bytes_rexmit; 3961 } else { 3962 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 3963 tp->snd_recover - tp->snd_una)) 3964 del_data = maxseg; 3965 pipe = imax(0, tp->snd_max - tp->snd_una - 3966 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 3967 } 3968 tp->sackhint.prr_delivered += del_data; 3969 /* 3970 * Proportional Rate Reduction 3971 */ 3972 if (pipe >= tp->snd_ssthresh) { 3973 if (tp->sackhint.recover_fs == 0) 3974 tp->sackhint.recover_fs = 3975 imax(1, tp->snd_nxt - tp->snd_una); 3976 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3977 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3978 tp->sackhint.prr_out; 3979 } else { 3980 if (V_tcp_do_prr_conservative || (del_data == 0)) 3981 limit = tp->sackhint.prr_delivered - 3982 tp->sackhint.prr_out; 3983 else 3984 limit = imax(tp->sackhint.prr_delivered - 3985 tp->sackhint.prr_out, del_data) + 3986 maxseg; 3987 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 3988 } 3989 snd_cnt = imax(snd_cnt, 0) / maxseg; 3990 /* 3991 * Send snd_cnt new data into the network in response to this ack. 3992 * If there is going to be a SACK retransmission, adjust snd_cwnd 3993 * accordingly. 3994 */ 3995 if (IN_FASTRECOVERY(tp->t_flags)) { 3996 if (tcp_is_sack_recovery(tp, to)) { 3997 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 3998 tp->sackhint.sack_bytes_rexmit + 3999 (snd_cnt * maxseg); 4000 } else { 4001 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4002 (snd_cnt * maxseg); 4003 } 4004 } else if (IN_CONGRECOVERY(tp->t_flags)) 4005 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4006 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4007 } 4008 4009 /* 4010 * On a partial ack arrives, force the retransmission of the 4011 * next unacknowledged segment. Do not clear tp->t_dupacks. 4012 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4013 * be started again. 4014 */ 4015 void 4016 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4017 { 4018 tcp_seq onxt = tp->snd_nxt; 4019 uint32_t ocwnd = tp->snd_cwnd; 4020 u_int maxseg = tcp_maxseg(tp); 4021 4022 INP_WLOCK_ASSERT(tp->t_inpcb); 4023 4024 tcp_timer_activate(tp, TT_REXMT, 0); 4025 tp->t_rtttime = 0; 4026 tp->snd_nxt = th->th_ack; 4027 /* 4028 * Set snd_cwnd to one segment beyond acknowledged offset. 4029 * (tp->snd_una has not yet been updated when this function is called.) 4030 */ 4031 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4032 tp->t_flags |= TF_ACKNOW; 4033 (void) tcp_output(tp); 4034 tp->snd_cwnd = ocwnd; 4035 if (SEQ_GT(onxt, tp->snd_nxt)) 4036 tp->snd_nxt = onxt; 4037 /* 4038 * Partial window deflation. Relies on fact that tp->snd_una 4039 * not updated yet. 4040 */ 4041 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4042 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4043 else 4044 tp->snd_cwnd = 0; 4045 tp->snd_cwnd += maxseg; 4046 } 4047 4048 int 4049 tcp_compute_pipe(struct tcpcb *tp) 4050 { 4051 return (tp->snd_max - tp->snd_una + 4052 tp->sackhint.sack_bytes_rexmit - 4053 tp->sackhint.sacked_bytes); 4054 } 4055 4056 uint32_t 4057 tcp_compute_initwnd(uint32_t maxseg) 4058 { 4059 /* 4060 * Calculate the Initial Window, also used as Restart Window 4061 * 4062 * RFC5681 Section 3.1 specifies the default conservative values. 4063 * RFC3390 specifies slightly more aggressive values. 4064 * RFC6928 increases it to ten segments. 4065 * Support for user specified value for initial flight size. 4066 */ 4067 if (V_tcp_initcwnd_segments) 4068 return min(V_tcp_initcwnd_segments * maxseg, 4069 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4070 else if (V_tcp_do_rfc3390) 4071 return min(4 * maxseg, max(2 * maxseg, 4380)); 4072 else { 4073 /* Per RFC5681 Section 3.1 */ 4074 if (maxseg > 2190) 4075 return (2 * maxseg); 4076 else if (maxseg > 1095) 4077 return (3 * maxseg); 4078 else 4079 return (4 * maxseg); 4080 } 4081 } 4082