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 #ifdef TCP_HHOOK 272 /* 273 * Wrapper for the TCP established input helper hook. 274 */ 275 void 276 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 277 { 278 struct tcp_hhook_data hhook_data; 279 280 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 281 hhook_data.tp = tp; 282 hhook_data.th = th; 283 hhook_data.to = to; 284 285 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 286 tp->osd); 287 } 288 } 289 #endif 290 291 /* 292 * CC wrapper hook functions 293 */ 294 void 295 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 296 uint16_t type) 297 { 298 #ifdef STATS 299 int32_t gput; 300 #endif 301 302 INP_WLOCK_ASSERT(tp->t_inpcb); 303 304 tp->ccv->nsegs = nsegs; 305 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 306 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 307 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 308 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 309 tp->ccv->flags |= CCF_CWND_LIMITED; 310 else 311 tp->ccv->flags &= ~CCF_CWND_LIMITED; 312 313 if (type == CC_ACK) { 314 #ifdef STATS 315 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 316 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 317 if (!IN_RECOVERY(tp->t_flags)) 318 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 319 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 320 if ((tp->t_flags & TF_GPUTINPROG) && 321 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 322 /* 323 * Compute goodput in bits per millisecond. 324 */ 325 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 326 max(1, tcp_ts_getticks() - tp->gput_ts); 327 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 328 gput); 329 /* 330 * XXXLAS: This is a temporary hack, and should be 331 * chained off VOI_TCP_GPUT when stats(9) grows an API 332 * to deal with chained VOIs. 333 */ 334 if (tp->t_stats_gput_prev > 0) 335 stats_voi_update_abs_s32(tp->t_stats, 336 VOI_TCP_GPUT_ND, 337 ((gput - tp->t_stats_gput_prev) * 100) / 338 tp->t_stats_gput_prev); 339 tp->t_flags &= ~TF_GPUTINPROG; 340 tp->t_stats_gput_prev = gput; 341 } 342 #endif /* STATS */ 343 if (tp->snd_cwnd > tp->snd_ssthresh) { 344 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 345 if (tp->t_bytes_acked >= tp->snd_cwnd) { 346 tp->t_bytes_acked -= tp->snd_cwnd; 347 tp->ccv->flags |= CCF_ABC_SENTAWND; 348 } 349 } else { 350 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 351 tp->t_bytes_acked = 0; 352 } 353 } 354 355 if (CC_ALGO(tp)->ack_received != NULL) { 356 /* XXXLAS: Find a way to live without this */ 357 tp->ccv->curack = th->th_ack; 358 CC_ALGO(tp)->ack_received(tp->ccv, type); 359 } 360 #ifdef STATS 361 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 362 #endif 363 } 364 365 void 366 cc_conn_init(struct tcpcb *tp) 367 { 368 struct hc_metrics_lite metrics; 369 struct inpcb *inp = tp->t_inpcb; 370 u_int maxseg; 371 int rtt; 372 373 INP_WLOCK_ASSERT(tp->t_inpcb); 374 375 tcp_hc_get(&inp->inp_inc, &metrics); 376 maxseg = tcp_maxseg(tp); 377 378 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 379 tp->t_srtt = rtt; 380 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 381 TCPSTAT_INC(tcps_usedrtt); 382 if (metrics.rmx_rttvar) { 383 tp->t_rttvar = metrics.rmx_rttvar; 384 TCPSTAT_INC(tcps_usedrttvar); 385 } else { 386 /* default variation is +- 1 rtt */ 387 tp->t_rttvar = 388 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 389 } 390 TCPT_RANGESET(tp->t_rxtcur, 391 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 392 tp->t_rttmin, TCPTV_REXMTMAX); 393 } 394 if (metrics.rmx_ssthresh) { 395 /* 396 * There's some sort of gateway or interface 397 * buffer limit on the path. Use this to set 398 * the slow start threshold, but set the 399 * threshold to no less than 2*mss. 400 */ 401 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 402 TCPSTAT_INC(tcps_usedssthresh); 403 } 404 405 /* 406 * Set the initial slow-start flight size. 407 * 408 * If a SYN or SYN/ACK was lost and retransmitted, we have to 409 * reduce the initial CWND to one segment as congestion is likely 410 * requiring us to be cautious. 411 */ 412 if (tp->snd_cwnd == 1) 413 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 414 else 415 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 416 417 if (CC_ALGO(tp)->conn_init != NULL) 418 CC_ALGO(tp)->conn_init(tp->ccv); 419 } 420 421 void inline 422 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 423 { 424 INP_WLOCK_ASSERT(tp->t_inpcb); 425 426 #ifdef STATS 427 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 428 #endif 429 430 switch(type) { 431 case CC_NDUPACK: 432 if (!IN_FASTRECOVERY(tp->t_flags)) { 433 tp->snd_recover = tp->snd_max; 434 if (tp->t_flags2 & TF2_ECN_PERMIT) 435 tp->t_flags2 |= TF2_ECN_SND_CWR; 436 } 437 break; 438 case CC_ECN: 439 if (!IN_CONGRECOVERY(tp->t_flags) || 440 /* 441 * Allow ECN reaction on ACK to CWR, if 442 * that data segment was also CE marked. 443 */ 444 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 445 EXIT_CONGRECOVERY(tp->t_flags); 446 TCPSTAT_INC(tcps_ecn_rcwnd); 447 tp->snd_recover = tp->snd_max + 1; 448 if (tp->t_flags2 & TF2_ECN_PERMIT) 449 tp->t_flags2 |= TF2_ECN_SND_CWR; 450 } 451 break; 452 case CC_RTO: 453 tp->t_dupacks = 0; 454 tp->t_bytes_acked = 0; 455 EXIT_RECOVERY(tp->t_flags); 456 if (tp->t_flags2 & TF2_ECN_PERMIT) 457 tp->t_flags2 |= TF2_ECN_SND_CWR; 458 break; 459 case CC_RTO_ERR: 460 TCPSTAT_INC(tcps_sndrexmitbad); 461 /* RTO was unnecessary, so reset everything. */ 462 tp->snd_cwnd = tp->snd_cwnd_prev; 463 tp->snd_ssthresh = tp->snd_ssthresh_prev; 464 tp->snd_recover = tp->snd_recover_prev; 465 if (tp->t_flags & TF_WASFRECOVERY) 466 ENTER_FASTRECOVERY(tp->t_flags); 467 if (tp->t_flags & TF_WASCRECOVERY) 468 ENTER_CONGRECOVERY(tp->t_flags); 469 tp->snd_nxt = tp->snd_max; 470 tp->t_flags &= ~TF_PREVVALID; 471 tp->t_badrxtwin = 0; 472 break; 473 } 474 475 if (CC_ALGO(tp)->cong_signal != NULL) { 476 if (th != NULL) 477 tp->ccv->curack = th->th_ack; 478 CC_ALGO(tp)->cong_signal(tp->ccv, type); 479 } 480 } 481 482 void inline 483 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 484 { 485 INP_WLOCK_ASSERT(tp->t_inpcb); 486 487 /* XXXLAS: KASSERT that we're in recovery? */ 488 489 if (CC_ALGO(tp)->post_recovery != NULL) { 490 tp->ccv->curack = th->th_ack; 491 CC_ALGO(tp)->post_recovery(tp->ccv); 492 } 493 /* XXXLAS: EXIT_RECOVERY ? */ 494 tp->t_bytes_acked = 0; 495 tp->sackhint.delivered_data = 0; 496 tp->sackhint.prr_out = 0; 497 } 498 499 /* 500 * Indicate whether this ack should be delayed. We can delay the ack if 501 * following conditions are met: 502 * - There is no delayed ack timer in progress. 503 * - Our last ack wasn't a 0-sized window. We never want to delay 504 * the ack that opens up a 0-sized window. 505 * - LRO wasn't used for this segment. We make sure by checking that the 506 * segment size is not larger than the MSS. 507 */ 508 #define DELAY_ACK(tp, tlen) \ 509 ((!tcp_timer_active(tp, TT_DELACK) && \ 510 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 511 (tlen <= tp->t_maxseg) && \ 512 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 513 514 void inline 515 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 516 { 517 INP_WLOCK_ASSERT(tp->t_inpcb); 518 519 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 520 switch (iptos & IPTOS_ECN_MASK) { 521 case IPTOS_ECN_CE: 522 tp->ccv->flags |= CCF_IPHDR_CE; 523 break; 524 case IPTOS_ECN_ECT0: 525 /* FALLTHROUGH */ 526 case IPTOS_ECN_ECT1: 527 /* FALLTHROUGH */ 528 case IPTOS_ECN_NOTECT: 529 tp->ccv->flags &= ~CCF_IPHDR_CE; 530 break; 531 } 532 533 if (flags & TH_CWR) 534 tp->ccv->flags |= CCF_TCPHDR_CWR; 535 else 536 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 537 538 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 539 540 if (tp->ccv->flags & CCF_ACKNOW) { 541 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 542 tp->t_flags |= TF_ACKNOW; 543 } 544 } 545 } 546 547 void inline 548 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 549 { 550 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 551 } 552 553 /* 554 * TCP input handling is split into multiple parts: 555 * tcp6_input is a thin wrapper around tcp_input for the extended 556 * ip6_protox[] call format in ip6_input 557 * tcp_input handles primary segment validation, inpcb lookup and 558 * SYN processing on listen sockets 559 * tcp_do_segment processes the ACK and text of the segment for 560 * establishing, established and closing connections 561 */ 562 #ifdef INET6 563 int 564 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 565 { 566 struct mbuf *m; 567 struct in6_ifaddr *ia6; 568 struct ip6_hdr *ip6; 569 570 m = *mp; 571 if (m->m_len < *offp + sizeof(struct tcphdr)) { 572 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 573 if (m == NULL) { 574 *mp = m; 575 TCPSTAT_INC(tcps_rcvshort); 576 return (IPPROTO_DONE); 577 } 578 } 579 580 /* 581 * draft-itojun-ipv6-tcp-to-anycast 582 * better place to put this in? 583 */ 584 ip6 = mtod(m, struct ip6_hdr *); 585 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 586 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 587 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 588 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 589 *mp = NULL; 590 return (IPPROTO_DONE); 591 } 592 593 *mp = m; 594 return (tcp_input_with_port(mp, offp, proto, port)); 595 } 596 597 int 598 tcp6_input(struct mbuf **mp, int *offp, int proto) 599 { 600 601 return(tcp6_input_with_port(mp, offp, proto, 0)); 602 } 603 #endif /* INET6 */ 604 605 int 606 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 607 { 608 struct mbuf *m = *mp; 609 struct tcphdr *th = NULL; 610 struct ip *ip = NULL; 611 struct inpcb *inp = NULL; 612 struct tcpcb *tp = NULL; 613 struct socket *so = NULL; 614 u_char *optp = NULL; 615 int off0; 616 int optlen = 0; 617 #ifdef INET 618 int len; 619 uint8_t ipttl; 620 #endif 621 int tlen = 0, off; 622 int drop_hdrlen; 623 int thflags; 624 int rstreason = 0; /* For badport_bandlim accounting purposes */ 625 int lookupflag; 626 uint8_t iptos; 627 struct m_tag *fwd_tag = NULL; 628 #ifdef INET6 629 struct ip6_hdr *ip6 = NULL; 630 int isipv6; 631 #else 632 const void *ip6 = NULL; 633 #endif /* INET6 */ 634 struct tcpopt to; /* options in this segment */ 635 char *s = NULL; /* address and port logging */ 636 #ifdef TCPDEBUG 637 /* 638 * The size of tcp_saveipgen must be the size of the max ip header, 639 * now IPv6. 640 */ 641 u_char tcp_saveipgen[IP6_HDR_LEN]; 642 struct tcphdr tcp_savetcp; 643 short ostate = 0; 644 #endif 645 646 NET_EPOCH_ASSERT(); 647 648 #ifdef INET6 649 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 650 #endif 651 652 off0 = *offp; 653 m = *mp; 654 *mp = NULL; 655 to.to_flags = 0; 656 TCPSTAT_INC(tcps_rcvtotal); 657 658 #ifdef INET6 659 if (isipv6) { 660 ip6 = mtod(m, struct ip6_hdr *); 661 th = (struct tcphdr *)((caddr_t)ip6 + off0); 662 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 663 if (port) 664 goto skip6_csum; 665 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 666 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 667 th->th_sum = m->m_pkthdr.csum_data; 668 else 669 th->th_sum = in6_cksum_pseudo(ip6, tlen, 670 IPPROTO_TCP, m->m_pkthdr.csum_data); 671 th->th_sum ^= 0xffff; 672 } else 673 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 674 if (th->th_sum) { 675 TCPSTAT_INC(tcps_rcvbadsum); 676 goto drop; 677 } 678 skip6_csum: 679 /* 680 * Be proactive about unspecified IPv6 address in source. 681 * As we use all-zero to indicate unbounded/unconnected pcb, 682 * unspecified IPv6 address can be used to confuse us. 683 * 684 * Note that packets with unspecified IPv6 destination is 685 * already dropped in ip6_input. 686 */ 687 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 688 /* XXX stat */ 689 goto drop; 690 } 691 iptos = IPV6_TRAFFIC_CLASS(ip6); 692 } 693 #endif 694 #if defined(INET) && defined(INET6) 695 else 696 #endif 697 #ifdef INET 698 { 699 /* 700 * Get IP and TCP header together in first mbuf. 701 * Note: IP leaves IP header in first mbuf. 702 */ 703 if (off0 > sizeof (struct ip)) { 704 ip_stripoptions(m); 705 off0 = sizeof(struct ip); 706 } 707 if (m->m_len < sizeof (struct tcpiphdr)) { 708 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 709 == NULL) { 710 TCPSTAT_INC(tcps_rcvshort); 711 return (IPPROTO_DONE); 712 } 713 } 714 ip = mtod(m, struct ip *); 715 th = (struct tcphdr *)((caddr_t)ip + off0); 716 tlen = ntohs(ip->ip_len) - off0; 717 718 iptos = ip->ip_tos; 719 if (port) 720 goto skip_csum; 721 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 722 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 723 th->th_sum = m->m_pkthdr.csum_data; 724 else 725 th->th_sum = in_pseudo(ip->ip_src.s_addr, 726 ip->ip_dst.s_addr, 727 htonl(m->m_pkthdr.csum_data + tlen + 728 IPPROTO_TCP)); 729 th->th_sum ^= 0xffff; 730 } else { 731 struct ipovly *ipov = (struct ipovly *)ip; 732 733 /* 734 * Checksum extended TCP header and data. 735 */ 736 len = off0 + tlen; 737 ipttl = ip->ip_ttl; 738 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 739 ipov->ih_len = htons(tlen); 740 th->th_sum = in_cksum(m, len); 741 /* Reset length for SDT probes. */ 742 ip->ip_len = htons(len); 743 /* Reset TOS bits */ 744 ip->ip_tos = iptos; 745 /* Re-initialization for later version check */ 746 ip->ip_ttl = ipttl; 747 ip->ip_v = IPVERSION; 748 ip->ip_hl = off0 >> 2; 749 } 750 skip_csum: 751 if (th->th_sum && (port == 0)) { 752 TCPSTAT_INC(tcps_rcvbadsum); 753 goto drop; 754 } 755 } 756 #endif /* INET */ 757 758 /* 759 * Check that TCP offset makes sense, 760 * pull out TCP options and adjust length. XXX 761 */ 762 off = th->th_off << 2; 763 if (off < sizeof (struct tcphdr) || off > tlen) { 764 TCPSTAT_INC(tcps_rcvbadoff); 765 goto drop; 766 } 767 tlen -= off; /* tlen is used instead of ti->ti_len */ 768 if (off > sizeof (struct tcphdr)) { 769 #ifdef INET6 770 if (isipv6) { 771 if (m->m_len < off0 + off) { 772 m = m_pullup(m, off0 + off); 773 if (m == NULL) { 774 TCPSTAT_INC(tcps_rcvshort); 775 return (IPPROTO_DONE); 776 } 777 } 778 ip6 = mtod(m, struct ip6_hdr *); 779 th = (struct tcphdr *)((caddr_t)ip6 + off0); 780 } 781 #endif 782 #if defined(INET) && defined(INET6) 783 else 784 #endif 785 #ifdef INET 786 { 787 if (m->m_len < sizeof(struct ip) + off) { 788 if ((m = m_pullup(m, sizeof (struct ip) + off)) 789 == NULL) { 790 TCPSTAT_INC(tcps_rcvshort); 791 return (IPPROTO_DONE); 792 } 793 ip = mtod(m, struct ip *); 794 th = (struct tcphdr *)((caddr_t)ip + off0); 795 } 796 } 797 #endif 798 optlen = off - sizeof (struct tcphdr); 799 optp = (u_char *)(th + 1); 800 } 801 thflags = tcp_get_flags(th); 802 803 /* 804 * Convert TCP protocol specific fields to host format. 805 */ 806 tcp_fields_to_host(th); 807 808 /* 809 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 810 */ 811 drop_hdrlen = off0 + off; 812 813 /* 814 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 815 */ 816 if ( 817 #ifdef INET6 818 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 819 #ifdef INET 820 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 821 #endif 822 #endif 823 #if defined(INET) && !defined(INET6) 824 (m->m_flags & M_IP_NEXTHOP) 825 #endif 826 ) 827 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 828 829 /* 830 * For initial SYN packets we don't need write lock on matching 831 * PCB, be it a listening one or a synchronized one. The packet 832 * shall not modify its state. 833 */ 834 lookupflag = INPLOOKUP_WILDCARD | 835 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 836 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); 837 findpcb: 838 #ifdef INET6 839 if (isipv6 && fwd_tag != NULL) { 840 struct sockaddr_in6 *next_hop6; 841 842 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 843 /* 844 * Transparently forwarded. Pretend to be the destination. 845 * Already got one like this? 846 */ 847 inp = in6_pcblookup_mbuf(&V_tcbinfo, 848 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 849 lookupflag, m->m_pkthdr.rcvif, m); 850 if (!inp) { 851 /* 852 * It's new. Try to find the ambushing socket. 853 * Because we've rewritten the destination address, 854 * any hardware-generated hash is ignored. 855 */ 856 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 857 th->th_sport, &next_hop6->sin6_addr, 858 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 859 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 860 } 861 } else if (isipv6) { 862 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 863 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 864 m->m_pkthdr.rcvif, m); 865 } 866 #endif /* INET6 */ 867 #if defined(INET6) && defined(INET) 868 else 869 #endif 870 #ifdef INET 871 if (fwd_tag != NULL) { 872 struct sockaddr_in *next_hop; 873 874 next_hop = (struct sockaddr_in *)(fwd_tag+1); 875 /* 876 * Transparently forwarded. Pretend to be the destination. 877 * already got one like this? 878 */ 879 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 880 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); 881 if (!inp) { 882 /* 883 * It's new. Try to find the ambushing socket. 884 * Because we've rewritten the destination address, 885 * any hardware-generated hash is ignored. 886 */ 887 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 888 th->th_sport, next_hop->sin_addr, 889 next_hop->sin_port ? ntohs(next_hop->sin_port) : 890 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 891 } 892 } else 893 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 894 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 895 m->m_pkthdr.rcvif, m); 896 #endif /* INET */ 897 898 /* 899 * If the INPCB does not exist then all data in the incoming 900 * segment is discarded and an appropriate RST is sent back. 901 * XXX MRT Send RST using which routing table? 902 */ 903 if (inp == NULL) { 904 if (rstreason != 0) { 905 /* We came here after second (safety) lookup. */ 906 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); 907 goto dropwithreset; 908 } 909 /* 910 * Log communication attempts to ports that are not 911 * in use. 912 */ 913 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 914 V_tcp_log_in_vain == 2) { 915 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 916 log(LOG_INFO, "%s; %s: Connection attempt " 917 "to closed port\n", s, __func__); 918 } 919 /* 920 * When blackholing do not respond with a RST but 921 * completely ignore the segment and drop it. 922 */ 923 if (((V_blackhole == 1 && (thflags & TH_SYN)) || 924 V_blackhole == 2) && (V_blackhole_local || ( 925 #ifdef INET6 926 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 927 #endif 928 #ifdef INET 929 !in_localip(ip->ip_src) 930 #else 931 true 932 #endif 933 ))) 934 goto dropunlock; 935 936 rstreason = BANDLIM_RST_CLOSEDPORT; 937 goto dropwithreset; 938 } 939 INP_LOCK_ASSERT(inp); 940 /* 941 * While waiting for inp lock during the lookup, another thread 942 * can have dropped the inpcb, in which case we need to loop back 943 * and try to find a new inpcb to deliver to. 944 */ 945 if (inp->inp_flags & INP_DROPPED) { 946 INP_UNLOCK(inp); 947 inp = NULL; 948 goto findpcb; 949 } 950 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 951 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 952 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) { 953 inp->inp_flowid = m->m_pkthdr.flowid; 954 inp->inp_flowtype = M_HASHTYPE_GET(m); 955 } 956 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 957 #ifdef INET6 958 if (isipv6 && IPSEC_ENABLED(ipv6) && 959 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 960 goto dropunlock; 961 } 962 #ifdef INET 963 else 964 #endif 965 #endif /* INET6 */ 966 #ifdef INET 967 if (IPSEC_ENABLED(ipv4) && 968 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 969 goto dropunlock; 970 } 971 #endif /* INET */ 972 #endif /* IPSEC */ 973 974 /* 975 * Check the minimum TTL for socket. 976 */ 977 if (inp->inp_ip_minttl != 0) { 978 #ifdef INET6 979 if (isipv6) { 980 if (inp->inp_ip_minttl > ip6->ip6_hlim) 981 goto dropunlock; 982 } else 983 #endif 984 if (inp->inp_ip_minttl > ip->ip_ttl) 985 goto dropunlock; 986 } 987 988 /* 989 * A previous connection in TIMEWAIT state is supposed to catch stray 990 * or duplicate segments arriving late. If this segment was a 991 * legitimate new connection attempt, the old INPCB gets removed and 992 * we can try again to find a listening socket. 993 */ 994 if (inp->inp_flags & INP_TIMEWAIT) { 995 tcp_dooptions(&to, optp, optlen, 996 (thflags & TH_SYN) ? TO_SYN : 0); 997 /* 998 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 999 */ 1000 if (tcp_twcheck(inp, &to, th, m, tlen)) 1001 goto findpcb; 1002 return (IPPROTO_DONE); 1003 } 1004 /* 1005 * The TCPCB may no longer exist if the connection is winding 1006 * down or it is in the CLOSED state. Either way we drop the 1007 * segment and send an appropriate response. 1008 */ 1009 tp = intotcpcb(inp); 1010 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1011 rstreason = BANDLIM_RST_CLOSEDPORT; 1012 goto dropwithreset; 1013 } 1014 1015 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1016 rstreason = BANDLIM_RST_CLOSEDPORT; 1017 goto dropwithreset; 1018 } 1019 1020 #ifdef TCP_OFFLOAD 1021 if (tp->t_flags & TF_TOE) { 1022 tcp_offload_input(tp, m); 1023 m = NULL; /* consumed by the TOE driver */ 1024 goto dropunlock; 1025 } 1026 #endif 1027 1028 #ifdef MAC 1029 if (mac_inpcb_check_deliver(inp, m)) 1030 goto dropunlock; 1031 #endif 1032 so = inp->inp_socket; 1033 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1034 #ifdef TCPDEBUG 1035 if (so->so_options & SO_DEBUG) { 1036 ostate = tp->t_state; 1037 #ifdef INET6 1038 if (isipv6) { 1039 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1040 } else 1041 #endif 1042 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1043 tcp_savetcp = *th; 1044 } 1045 #endif /* TCPDEBUG */ 1046 /* 1047 * When the socket is accepting connections (the INPCB is in LISTEN 1048 * state) we look into the SYN cache if this is a new connection 1049 * attempt or the completion of a previous one. 1050 */ 1051 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1052 ("%s: so accepting but tp %p not listening", __func__, tp)); 1053 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1054 struct in_conninfo inc; 1055 1056 bzero(&inc, sizeof(inc)); 1057 #ifdef INET6 1058 if (isipv6) { 1059 inc.inc_flags |= INC_ISIPV6; 1060 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1061 inc.inc_flags |= INC_IPV6MINMTU; 1062 inc.inc6_faddr = ip6->ip6_src; 1063 inc.inc6_laddr = ip6->ip6_dst; 1064 } else 1065 #endif 1066 { 1067 inc.inc_faddr = ip->ip_src; 1068 inc.inc_laddr = ip->ip_dst; 1069 } 1070 inc.inc_fport = th->th_sport; 1071 inc.inc_lport = th->th_dport; 1072 inc.inc_fibnum = so->so_fibnum; 1073 1074 /* 1075 * Check for an existing connection attempt in syncache if 1076 * the flag is only ACK. A successful lookup creates a new 1077 * socket appended to the listen queue in SYN_RECEIVED state. 1078 */ 1079 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1080 /* 1081 * Parse the TCP options here because 1082 * syncookies need access to the reflected 1083 * timestamp. 1084 */ 1085 tcp_dooptions(&to, optp, optlen, 0); 1086 /* 1087 * NB: syncache_expand() doesn't unlock inp. 1088 */ 1089 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1090 if (rstreason < 0) { 1091 /* 1092 * A failing TCP MD5 signature comparison 1093 * must result in the segment being dropped 1094 * and must not produce any response back 1095 * to the sender. 1096 */ 1097 goto dropunlock; 1098 } else if (rstreason == 0) { 1099 /* 1100 * No syncache entry, or ACK was not for our 1101 * SYN/ACK. Do our protection against double 1102 * ACK. If peer sent us 2 ACKs, then for the 1103 * first one syncache_expand() successfully 1104 * converted syncache entry into a socket, 1105 * while we were waiting on the inpcb lock. We 1106 * don't want to sent RST for the second ACK, 1107 * so we perform second lookup without wildcard 1108 * match, hoping to find the new socket. If 1109 * the ACK is stray indeed, rstreason would 1110 * hint the above code that the lookup was a 1111 * second attempt. 1112 * 1113 * NB: syncache did its own logging 1114 * of the failure cause. 1115 */ 1116 INP_WUNLOCK(inp); 1117 rstreason = BANDLIM_RST_OPENPORT; 1118 lookupflag &= ~INPLOOKUP_WILDCARD; 1119 goto findpcb; 1120 } 1121 tfo_socket_result: 1122 if (so == NULL) { 1123 /* 1124 * We completed the 3-way handshake 1125 * but could not allocate a socket 1126 * either due to memory shortage, 1127 * listen queue length limits or 1128 * global socket limits. Send RST 1129 * or wait and have the remote end 1130 * retransmit the ACK for another 1131 * try. 1132 */ 1133 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1134 log(LOG_DEBUG, "%s; %s: Listen socket: " 1135 "Socket allocation failed due to " 1136 "limits or memory shortage, %s\n", 1137 s, __func__, 1138 V_tcp_sc_rst_sock_fail ? 1139 "sending RST" : "try again"); 1140 if (V_tcp_sc_rst_sock_fail) { 1141 rstreason = BANDLIM_UNLIMITED; 1142 goto dropwithreset; 1143 } else 1144 goto dropunlock; 1145 } 1146 /* 1147 * Socket is created in state SYN_RECEIVED. 1148 * Unlock the listen socket, lock the newly 1149 * created socket and update the tp variable. 1150 * If we came here via jump to tfo_socket_result, 1151 * then listening socket is read-locked. 1152 */ 1153 INP_UNLOCK(inp); /* listen socket */ 1154 inp = sotoinpcb(so); 1155 /* 1156 * New connection inpcb is already locked by 1157 * syncache_expand(). 1158 */ 1159 INP_WLOCK_ASSERT(inp); 1160 tp = intotcpcb(inp); 1161 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1162 ("%s: ", __func__)); 1163 /* 1164 * Process the segment and the data it 1165 * contains. tcp_do_segment() consumes 1166 * the mbuf chain and unlocks the inpcb. 1167 */ 1168 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1169 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1170 iptos); 1171 return (IPPROTO_DONE); 1172 } 1173 /* 1174 * Segment flag validation for new connection attempts: 1175 * 1176 * Our (SYN|ACK) response was rejected. 1177 * Check with syncache and remove entry to prevent 1178 * retransmits. 1179 * 1180 * NB: syncache_chkrst does its own logging of failure 1181 * causes. 1182 */ 1183 if (thflags & TH_RST) { 1184 syncache_chkrst(&inc, th, m, port); 1185 goto dropunlock; 1186 } 1187 /* 1188 * We can't do anything without SYN. 1189 */ 1190 if ((thflags & TH_SYN) == 0) { 1191 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1192 log(LOG_DEBUG, "%s; %s: Listen socket: " 1193 "SYN is missing, segment ignored\n", 1194 s, __func__); 1195 TCPSTAT_INC(tcps_badsyn); 1196 goto dropunlock; 1197 } 1198 /* 1199 * (SYN|ACK) is bogus on a listen socket. 1200 */ 1201 if (thflags & TH_ACK) { 1202 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1203 log(LOG_DEBUG, "%s; %s: Listen socket: " 1204 "SYN|ACK invalid, segment rejected\n", 1205 s, __func__); 1206 syncache_badack(&inc, port); /* XXX: Not needed! */ 1207 TCPSTAT_INC(tcps_badsyn); 1208 rstreason = BANDLIM_RST_OPENPORT; 1209 goto dropwithreset; 1210 } 1211 /* 1212 * If the drop_synfin option is enabled, drop all 1213 * segments with both the SYN and FIN bits set. 1214 * This prevents e.g. nmap from identifying the 1215 * TCP/IP stack. 1216 * XXX: Poor reasoning. nmap has other methods 1217 * and is constantly refining its stack detection 1218 * strategies. 1219 * XXX: This is a violation of the TCP specification 1220 * and was used by RFC1644. 1221 */ 1222 if ((thflags & TH_FIN) && V_drop_synfin) { 1223 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1224 log(LOG_DEBUG, "%s; %s: Listen socket: " 1225 "SYN|FIN segment ignored (based on " 1226 "sysctl setting)\n", s, __func__); 1227 TCPSTAT_INC(tcps_badsyn); 1228 goto dropunlock; 1229 } 1230 /* 1231 * Segment's flags are (SYN) or (SYN|FIN). 1232 * 1233 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1234 * as they do not affect the state of the TCP FSM. 1235 * The data pointed to by TH_URG and th_urp is ignored. 1236 */ 1237 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1238 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1239 KASSERT(thflags & (TH_SYN), 1240 ("%s: Listen socket: TH_SYN not set", __func__)); 1241 INP_RLOCK_ASSERT(inp); 1242 #ifdef INET6 1243 /* 1244 * If deprecated address is forbidden, 1245 * we do not accept SYN to deprecated interface 1246 * address to prevent any new inbound connection from 1247 * getting established. 1248 * When we do not accept SYN, we send a TCP RST, 1249 * with deprecated source address (instead of dropping 1250 * it). We compromise it as it is much better for peer 1251 * to send a RST, and RST will be the final packet 1252 * for the exchange. 1253 * 1254 * If we do not forbid deprecated addresses, we accept 1255 * the SYN packet. RFC2462 does not suggest dropping 1256 * SYN in this case. 1257 * If we decipher RFC2462 5.5.4, it says like this: 1258 * 1. use of deprecated addr with existing 1259 * communication is okay - "SHOULD continue to be 1260 * used" 1261 * 2. use of it with new communication: 1262 * (2a) "SHOULD NOT be used if alternate address 1263 * with sufficient scope is available" 1264 * (2b) nothing mentioned otherwise. 1265 * Here we fall into (2b) case as we have no choice in 1266 * our source address selection - we must obey the peer. 1267 * 1268 * The wording in RFC2462 is confusing, and there are 1269 * multiple description text for deprecated address 1270 * handling - worse, they are not exactly the same. 1271 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1272 */ 1273 if (isipv6 && !V_ip6_use_deprecated) { 1274 struct in6_ifaddr *ia6; 1275 1276 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1277 if (ia6 != NULL && 1278 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1279 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1280 log(LOG_DEBUG, "%s; %s: Listen socket: " 1281 "Connection attempt to deprecated " 1282 "IPv6 address rejected\n", 1283 s, __func__); 1284 rstreason = BANDLIM_RST_OPENPORT; 1285 goto dropwithreset; 1286 } 1287 } 1288 #endif /* INET6 */ 1289 /* 1290 * Basic sanity checks on incoming SYN requests: 1291 * Don't respond if the destination is a link layer 1292 * broadcast according to RFC1122 4.2.3.10, p. 104. 1293 * If it is from this socket it must be forged. 1294 * Don't respond if the source or destination is a 1295 * global or subnet broad- or multicast address. 1296 * Note that it is quite possible to receive unicast 1297 * link-layer packets with a broadcast IP address. Use 1298 * in_broadcast() to find them. 1299 */ 1300 if (m->m_flags & (M_BCAST|M_MCAST)) { 1301 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1302 log(LOG_DEBUG, "%s; %s: Listen socket: " 1303 "Connection attempt from broad- or multicast " 1304 "link layer address ignored\n", s, __func__); 1305 goto dropunlock; 1306 } 1307 #ifdef INET6 1308 if (isipv6) { 1309 if (th->th_dport == th->th_sport && 1310 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1311 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1312 log(LOG_DEBUG, "%s; %s: Listen socket: " 1313 "Connection attempt to/from self " 1314 "ignored\n", s, __func__); 1315 goto dropunlock; 1316 } 1317 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1318 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1319 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1320 log(LOG_DEBUG, "%s; %s: Listen socket: " 1321 "Connection attempt from/to multicast " 1322 "address ignored\n", s, __func__); 1323 goto dropunlock; 1324 } 1325 } 1326 #endif 1327 #if defined(INET) && defined(INET6) 1328 else 1329 #endif 1330 #ifdef INET 1331 { 1332 if (th->th_dport == th->th_sport && 1333 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1334 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1335 log(LOG_DEBUG, "%s; %s: Listen socket: " 1336 "Connection attempt from/to self " 1337 "ignored\n", s, __func__); 1338 goto dropunlock; 1339 } 1340 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1341 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1342 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1343 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1344 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1345 log(LOG_DEBUG, "%s; %s: Listen socket: " 1346 "Connection attempt from/to broad- " 1347 "or multicast address ignored\n", 1348 s, __func__); 1349 goto dropunlock; 1350 } 1351 } 1352 #endif 1353 /* 1354 * SYN appears to be valid. Create compressed TCP state 1355 * for syncache. 1356 */ 1357 #ifdef TCPDEBUG 1358 if (so->so_options & SO_DEBUG) 1359 tcp_trace(TA_INPUT, ostate, tp, 1360 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1361 #endif 1362 TCP_PROBE3(debug__input, tp, th, m); 1363 tcp_dooptions(&to, optp, optlen, TO_SYN); 1364 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1365 iptos, port)) != NULL) 1366 goto tfo_socket_result; 1367 1368 /* 1369 * Entry added to syncache and mbuf consumed. 1370 * Only the listen socket is unlocked by syncache_add(). 1371 */ 1372 return (IPPROTO_DONE); 1373 } else if (tp->t_state == TCPS_LISTEN) { 1374 /* 1375 * When a listen socket is torn down the SO_ACCEPTCONN 1376 * flag is removed first while connections are drained 1377 * from the accept queue in a unlock/lock cycle of the 1378 * ACCEPT_LOCK, opening a race condition allowing a SYN 1379 * attempt go through unhandled. 1380 */ 1381 goto dropunlock; 1382 } 1383 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1384 if (tp->t_flags & TF_SIGNATURE) { 1385 tcp_dooptions(&to, optp, optlen, thflags); 1386 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1387 TCPSTAT_INC(tcps_sig_err_nosigopt); 1388 goto dropunlock; 1389 } 1390 if (!TCPMD5_ENABLED() || 1391 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1392 goto dropunlock; 1393 } 1394 #endif 1395 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1396 1397 /* 1398 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1399 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1400 * the inpcb, and unlocks pcbinfo. 1401 * 1402 * XXXGL: in case of a pure SYN arriving on existing connection 1403 * TCP stacks won't need to modify the PCB, they would either drop 1404 * the segment silently, or send a challenge ACK. However, we try 1405 * to upgrade the lock, because calling convention for stacks is 1406 * write-lock on PCB. If upgrade fails, drop the SYN. 1407 */ 1408 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1409 goto dropunlock; 1410 1411 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1412 return (IPPROTO_DONE); 1413 1414 dropwithreset: 1415 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1416 1417 if (inp != NULL) { 1418 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1419 INP_UNLOCK(inp); 1420 } else 1421 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1422 m = NULL; /* mbuf chain got consumed. */ 1423 goto drop; 1424 1425 dropunlock: 1426 if (m != NULL) 1427 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1428 1429 if (inp != NULL) 1430 INP_UNLOCK(inp); 1431 1432 drop: 1433 if (s != NULL) 1434 free(s, M_TCPLOG); 1435 if (m != NULL) 1436 m_freem(m); 1437 return (IPPROTO_DONE); 1438 } 1439 1440 /* 1441 * Automatic sizing of receive socket buffer. Often the send 1442 * buffer size is not optimally adjusted to the actual network 1443 * conditions at hand (delay bandwidth product). Setting the 1444 * buffer size too small limits throughput on links with high 1445 * bandwidth and high delay (eg. trans-continental/oceanic links). 1446 * 1447 * On the receive side the socket buffer memory is only rarely 1448 * used to any significant extent. This allows us to be much 1449 * more aggressive in scaling the receive socket buffer. For 1450 * the case that the buffer space is actually used to a large 1451 * extent and we run out of kernel memory we can simply drop 1452 * the new segments; TCP on the sender will just retransmit it 1453 * later. Setting the buffer size too big may only consume too 1454 * much kernel memory if the application doesn't read() from 1455 * the socket or packet loss or reordering makes use of the 1456 * reassembly queue. 1457 * 1458 * The criteria to step up the receive buffer one notch are: 1459 * 1. Application has not set receive buffer size with 1460 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1461 * 2. the number of bytes received during 1/2 of an sRTT 1462 * is at least 3/8 of the current socket buffer size. 1463 * 3. receive buffer size has not hit maximal automatic size; 1464 * 1465 * If all of the criteria are met we increaset the socket buffer 1466 * by a 1/2 (bounded by the max). This allows us to keep ahead 1467 * of slow-start but also makes it so our peer never gets limited 1468 * by our rwnd which we then open up causing a burst. 1469 * 1470 * This algorithm does two steps per RTT at most and only if 1471 * we receive a bulk stream w/o packet losses or reorderings. 1472 * Shrinking the buffer during idle times is not necessary as 1473 * it doesn't consume any memory when idle. 1474 * 1475 * TODO: Only step up if the application is actually serving 1476 * the buffer to better manage the socket buffer resources. 1477 */ 1478 int 1479 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1480 struct tcpcb *tp, int tlen) 1481 { 1482 int newsize = 0; 1483 1484 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1485 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1486 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1487 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1488 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1489 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1490 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1491 } 1492 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1493 1494 /* Start over with next RTT. */ 1495 tp->rfbuf_ts = 0; 1496 tp->rfbuf_cnt = 0; 1497 } else { 1498 tp->rfbuf_cnt += tlen; /* add up */ 1499 } 1500 return (newsize); 1501 } 1502 1503 int 1504 tcp_input(struct mbuf **mp, int *offp, int proto) 1505 { 1506 return(tcp_input_with_port(mp, offp, proto, 0)); 1507 } 1508 1509 void 1510 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1511 { 1512 /* 1513 * Since tp might be gone if the session entered 1514 * the TIME_WAIT state before coming here, we need 1515 * to check if the socket is still connected. 1516 */ 1517 if (tp == NULL) { 1518 return; 1519 } 1520 if (so == NULL) { 1521 return; 1522 } 1523 INP_LOCK_ASSERT(tp->t_inpcb); 1524 if (tp->t_flags & TF_WAKESOR) { 1525 tp->t_flags &= ~TF_WAKESOR; 1526 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1527 sorwakeup_locked(so); 1528 } 1529 } 1530 1531 void 1532 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1533 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1534 { 1535 uint16_t thflags; 1536 int acked, ourfinisacked, needoutput = 0, sack_changed; 1537 int rstreason, todrop, win, incforsyn = 0; 1538 uint32_t tiwin; 1539 uint16_t nsegs; 1540 char *s; 1541 struct in_conninfo *inc; 1542 struct mbuf *mfree; 1543 struct tcpopt to; 1544 int tfo_syn; 1545 u_int maxseg; 1546 1547 #ifdef TCPDEBUG 1548 /* 1549 * The size of tcp_saveipgen must be the size of the max ip header, 1550 * now IPv6. 1551 */ 1552 u_char tcp_saveipgen[IP6_HDR_LEN]; 1553 struct tcphdr tcp_savetcp; 1554 short ostate = 0; 1555 #endif 1556 thflags = tcp_get_flags(th); 1557 inc = &tp->t_inpcb->inp_inc; 1558 tp->sackhint.last_sack_ack = 0; 1559 sack_changed = 0; 1560 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1561 1562 NET_EPOCH_ASSERT(); 1563 INP_WLOCK_ASSERT(tp->t_inpcb); 1564 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1565 __func__)); 1566 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1567 __func__)); 1568 1569 #ifdef TCPPCAP 1570 /* Save segment, if requested. */ 1571 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1572 #endif 1573 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1574 tlen, NULL, true); 1575 1576 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1577 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1578 log(LOG_DEBUG, "%s; %s: " 1579 "SYN|FIN segment ignored (based on " 1580 "sysctl setting)\n", s, __func__); 1581 free(s, M_TCPLOG); 1582 } 1583 goto drop; 1584 } 1585 1586 /* 1587 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1588 * check SEQ.ACK first. 1589 */ 1590 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1591 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1592 rstreason = BANDLIM_UNLIMITED; 1593 goto dropwithreset; 1594 } 1595 1596 /* 1597 * Segment received on connection. 1598 * Reset idle time and keep-alive timer. 1599 * XXX: This should be done after segment 1600 * validation to ignore broken/spoofed segs. 1601 */ 1602 tp->t_rcvtime = ticks; 1603 1604 /* 1605 * Scale up the window into a 32-bit value. 1606 * For the SYN_SENT state the scale is zero. 1607 */ 1608 tiwin = th->th_win << tp->snd_scale; 1609 #ifdef STATS 1610 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1611 #endif 1612 1613 /* 1614 * TCP ECN processing. 1615 */ 1616 if (tcp_ecn_input_segment(tp, thflags, iptos)) 1617 cc_cong_signal(tp, th, CC_ECN); 1618 1619 /* 1620 * Parse options on any incoming segment. 1621 */ 1622 tcp_dooptions(&to, (u_char *)(th + 1), 1623 (th->th_off << 2) - sizeof(struct tcphdr), 1624 (thflags & TH_SYN) ? TO_SYN : 0); 1625 1626 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1627 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1628 (to.to_flags & TOF_SIGNATURE) == 0) { 1629 TCPSTAT_INC(tcps_sig_err_sigopt); 1630 /* XXX: should drop? */ 1631 } 1632 #endif 1633 /* 1634 * If echoed timestamp is later than the current time, 1635 * fall back to non RFC1323 RTT calculation. Normalize 1636 * timestamp if syncookies were used when this connection 1637 * was established. 1638 */ 1639 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1640 to.to_tsecr -= tp->ts_offset; 1641 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1642 to.to_tsecr = 0; 1643 else if (tp->t_rxtshift == 1 && 1644 tp->t_flags & TF_PREVVALID && 1645 tp->t_badrxtwin != 0 && 1646 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 1647 cc_cong_signal(tp, th, CC_RTO_ERR); 1648 } 1649 /* 1650 * Process options only when we get SYN/ACK back. The SYN case 1651 * for incoming connections is handled in tcp_syncache. 1652 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1653 * or <SYN,ACK>) segment itself is never scaled. 1654 * XXX this is traditional behavior, may need to be cleaned up. 1655 */ 1656 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1657 /* Handle parallel SYN for ECN */ 1658 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1659 if ((to.to_flags & TOF_SCALE) && 1660 (tp->t_flags & TF_REQ_SCALE) && 1661 !(tp->t_flags & TF_NOOPT)) { 1662 tp->t_flags |= TF_RCVD_SCALE; 1663 tp->snd_scale = to.to_wscale; 1664 } else 1665 tp->t_flags &= ~TF_REQ_SCALE; 1666 /* 1667 * Initial send window. It will be updated with 1668 * the next incoming segment to the scaled value. 1669 */ 1670 tp->snd_wnd = th->th_win; 1671 if ((to.to_flags & TOF_TS) && 1672 (tp->t_flags & TF_REQ_TSTMP) && 1673 !(tp->t_flags & TF_NOOPT)) { 1674 tp->t_flags |= TF_RCVD_TSTMP; 1675 tp->ts_recent = to.to_tsval; 1676 tp->ts_recent_age = tcp_ts_getticks(); 1677 } else 1678 tp->t_flags &= ~TF_REQ_TSTMP; 1679 if (to.to_flags & TOF_MSS) 1680 tcp_mss(tp, to.to_mss); 1681 if ((tp->t_flags & TF_SACK_PERMIT) && 1682 (!(to.to_flags & TOF_SACKPERM) || 1683 (tp->t_flags & TF_NOOPT))) 1684 tp->t_flags &= ~TF_SACK_PERMIT; 1685 if (IS_FASTOPEN(tp->t_flags)) { 1686 if ((to.to_flags & TOF_FASTOPEN) && 1687 !(tp->t_flags & TF_NOOPT)) { 1688 uint16_t mss; 1689 1690 if (to.to_flags & TOF_MSS) 1691 mss = to.to_mss; 1692 else 1693 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1694 mss = TCP6_MSS; 1695 else 1696 mss = TCP_MSS; 1697 tcp_fastopen_update_cache(tp, mss, 1698 to.to_tfo_len, to.to_tfo_cookie); 1699 } else 1700 tcp_fastopen_disable_path(tp); 1701 } 1702 } 1703 1704 /* 1705 * If timestamps were negotiated during SYN/ACK and a 1706 * segment without a timestamp is received, silently drop 1707 * the segment, unless it is a RST segment or missing timestamps are 1708 * tolerated. 1709 * See section 3.2 of RFC 7323. 1710 */ 1711 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1712 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1713 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1714 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1715 "segment processed normally\n", 1716 s, __func__); 1717 free(s, M_TCPLOG); 1718 } 1719 } else { 1720 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1721 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1722 "segment silently dropped\n", s, __func__); 1723 free(s, M_TCPLOG); 1724 } 1725 goto drop; 1726 } 1727 } 1728 /* 1729 * If timestamps were not negotiated during SYN/ACK and a 1730 * segment with a timestamp is received, ignore the 1731 * timestamp and process the packet normally. 1732 * See section 3.2 of RFC 7323. 1733 */ 1734 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1735 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1736 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1737 "segment processed normally\n", s, __func__); 1738 free(s, M_TCPLOG); 1739 } 1740 } 1741 1742 /* 1743 * Header prediction: check for the two common cases 1744 * of a uni-directional data xfer. If the packet has 1745 * no control flags, is in-sequence, the window didn't 1746 * change and we're not retransmitting, it's a 1747 * candidate. If the length is zero and the ack moved 1748 * forward, we're the sender side of the xfer. Just 1749 * free the data acked & wake any higher level process 1750 * that was blocked waiting for space. If the length 1751 * is non-zero and the ack didn't move, we're the 1752 * receiver side. If we're getting packets in-order 1753 * (the reassembly queue is empty), add the data to 1754 * the socket buffer and note that we need a delayed ack. 1755 * Make sure that the hidden state-flags are also off. 1756 * Since we check for TCPS_ESTABLISHED first, it can only 1757 * be TH_NEEDSYN. 1758 */ 1759 if (tp->t_state == TCPS_ESTABLISHED && 1760 th->th_seq == tp->rcv_nxt && 1761 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1762 tp->snd_nxt == tp->snd_max && 1763 tiwin && tiwin == tp->snd_wnd && 1764 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1765 SEGQ_EMPTY(tp) && 1766 ((to.to_flags & TOF_TS) == 0 || 1767 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1768 /* 1769 * If last ACK falls within this segment's sequence numbers, 1770 * record the timestamp. 1771 * NOTE that the test is modified according to the latest 1772 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1773 */ 1774 if ((to.to_flags & TOF_TS) != 0 && 1775 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1776 tp->ts_recent_age = tcp_ts_getticks(); 1777 tp->ts_recent = to.to_tsval; 1778 } 1779 1780 if (tlen == 0) { 1781 if (SEQ_GT(th->th_ack, tp->snd_una) && 1782 SEQ_LEQ(th->th_ack, tp->snd_max) && 1783 !IN_RECOVERY(tp->t_flags) && 1784 (to.to_flags & TOF_SACK) == 0 && 1785 TAILQ_EMPTY(&tp->snd_holes)) { 1786 /* 1787 * This is a pure ack for outstanding data. 1788 */ 1789 TCPSTAT_INC(tcps_predack); 1790 1791 /* 1792 * "bad retransmit" recovery without timestamps. 1793 */ 1794 if ((to.to_flags & TOF_TS) == 0 && 1795 tp->t_rxtshift == 1 && 1796 tp->t_flags & TF_PREVVALID && 1797 tp->t_badrxtwin != 0 && 1798 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1799 cc_cong_signal(tp, th, CC_RTO_ERR); 1800 } 1801 1802 /* 1803 * Recalculate the transmit timer / rtt. 1804 * 1805 * Some boxes send broken timestamp replies 1806 * during the SYN+ACK phase, ignore 1807 * timestamps of 0 or we could calculate a 1808 * huge RTT and blow up the retransmit timer. 1809 */ 1810 if ((to.to_flags & TOF_TS) != 0 && 1811 to.to_tsecr) { 1812 uint32_t t; 1813 1814 t = tcp_ts_getticks() - to.to_tsecr; 1815 if (!tp->t_rttlow || tp->t_rttlow > t) 1816 tp->t_rttlow = t; 1817 tcp_xmit_timer(tp, 1818 TCP_TS_TO_TICKS(t) + 1); 1819 } else if (tp->t_rtttime && 1820 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1821 if (!tp->t_rttlow || 1822 tp->t_rttlow > ticks - tp->t_rtttime) 1823 tp->t_rttlow = ticks - tp->t_rtttime; 1824 tcp_xmit_timer(tp, 1825 ticks - tp->t_rtttime); 1826 } 1827 acked = BYTES_THIS_ACK(tp, th); 1828 1829 #ifdef TCP_HHOOK 1830 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1831 hhook_run_tcp_est_in(tp, th, &to); 1832 #endif 1833 1834 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1835 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1836 sbdrop(&so->so_snd, acked); 1837 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1838 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1839 tp->snd_recover = th->th_ack - 1; 1840 1841 /* 1842 * Let the congestion control algorithm update 1843 * congestion control related information. This 1844 * typically means increasing the congestion 1845 * window. 1846 */ 1847 cc_ack_received(tp, th, nsegs, CC_ACK); 1848 1849 tp->snd_una = th->th_ack; 1850 /* 1851 * Pull snd_wl2 up to prevent seq wrap relative 1852 * to th_ack. 1853 */ 1854 tp->snd_wl2 = th->th_ack; 1855 tp->t_dupacks = 0; 1856 m_freem(m); 1857 1858 /* 1859 * If all outstanding data are acked, stop 1860 * retransmit timer, otherwise restart timer 1861 * using current (possibly backed-off) value. 1862 * If process is waiting for space, 1863 * wakeup/selwakeup/signal. If data 1864 * are ready to send, let tcp_output 1865 * decide between more output or persist. 1866 */ 1867 #ifdef TCPDEBUG 1868 if (so->so_options & SO_DEBUG) 1869 tcp_trace(TA_INPUT, ostate, tp, 1870 (void *)tcp_saveipgen, 1871 &tcp_savetcp, 0); 1872 #endif 1873 TCP_PROBE3(debug__input, tp, th, m); 1874 if (tp->snd_una == tp->snd_max) 1875 tcp_timer_activate(tp, TT_REXMT, 0); 1876 else if (!tcp_timer_active(tp, TT_PERSIST)) 1877 tcp_timer_activate(tp, TT_REXMT, 1878 tp->t_rxtcur); 1879 sowwakeup(so); 1880 if (sbavail(&so->so_snd)) 1881 (void) tcp_output(tp); 1882 goto check_delack; 1883 } 1884 } else if (th->th_ack == tp->snd_una && 1885 tlen <= sbspace(&so->so_rcv)) { 1886 int newsize = 0; /* automatic sockbuf scaling */ 1887 1888 /* 1889 * This is a pure, in-sequence data packet with 1890 * nothing on the reassembly queue and we have enough 1891 * buffer space to take it. 1892 */ 1893 /* Clean receiver SACK report if present */ 1894 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1895 tcp_clean_sackreport(tp); 1896 TCPSTAT_INC(tcps_preddat); 1897 tp->rcv_nxt += tlen; 1898 if (tlen && 1899 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1900 (tp->t_fbyte_in == 0)) { 1901 tp->t_fbyte_in = ticks; 1902 if (tp->t_fbyte_in == 0) 1903 tp->t_fbyte_in = 1; 1904 if (tp->t_fbyte_out && tp->t_fbyte_in) 1905 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1906 } 1907 /* 1908 * Pull snd_wl1 up to prevent seq wrap relative to 1909 * th_seq. 1910 */ 1911 tp->snd_wl1 = th->th_seq; 1912 /* 1913 * Pull rcv_up up to prevent seq wrap relative to 1914 * rcv_nxt. 1915 */ 1916 tp->rcv_up = tp->rcv_nxt; 1917 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1918 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1919 #ifdef TCPDEBUG 1920 if (so->so_options & SO_DEBUG) 1921 tcp_trace(TA_INPUT, ostate, tp, 1922 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1923 #endif 1924 TCP_PROBE3(debug__input, tp, th, m); 1925 1926 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1927 1928 /* Add data to socket buffer. */ 1929 SOCKBUF_LOCK(&so->so_rcv); 1930 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1931 m_freem(m); 1932 } else { 1933 /* 1934 * Set new socket buffer size. 1935 * Give up when limit is reached. 1936 */ 1937 if (newsize) 1938 if (!sbreserve_locked(so, SO_RCV, 1939 newsize, NULL)) 1940 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1941 m_adj(m, drop_hdrlen); /* delayed header drop */ 1942 sbappendstream_locked(&so->so_rcv, m, 0); 1943 } 1944 /* NB: sorwakeup_locked() does an implicit unlock. */ 1945 sorwakeup_locked(so); 1946 if (DELAY_ACK(tp, tlen)) { 1947 tp->t_flags |= TF_DELACK; 1948 } else { 1949 tp->t_flags |= TF_ACKNOW; 1950 tcp_output(tp); 1951 } 1952 goto check_delack; 1953 } 1954 } 1955 1956 /* 1957 * Calculate amount of space in receive window, 1958 * and then do TCP input processing. 1959 * Receive window is amount of space in rcv queue, 1960 * but not less than advertised window. 1961 */ 1962 win = sbspace(&so->so_rcv); 1963 if (win < 0) 1964 win = 0; 1965 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1966 1967 switch (tp->t_state) { 1968 /* 1969 * If the state is SYN_RECEIVED: 1970 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1971 */ 1972 case TCPS_SYN_RECEIVED: 1973 if ((thflags & TH_ACK) && 1974 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1975 SEQ_GT(th->th_ack, tp->snd_max))) { 1976 rstreason = BANDLIM_RST_OPENPORT; 1977 goto dropwithreset; 1978 } 1979 if (IS_FASTOPEN(tp->t_flags)) { 1980 /* 1981 * When a TFO connection is in SYN_RECEIVED, the 1982 * only valid packets are the initial SYN, a 1983 * retransmit/copy of the initial SYN (possibly with 1984 * a subset of the original data), a valid ACK, a 1985 * FIN, or a RST. 1986 */ 1987 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1988 rstreason = BANDLIM_RST_OPENPORT; 1989 goto dropwithreset; 1990 } else if (thflags & TH_SYN) { 1991 /* non-initial SYN is ignored */ 1992 if ((tcp_timer_active(tp, TT_DELACK) || 1993 tcp_timer_active(tp, TT_REXMT))) 1994 goto drop; 1995 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1996 goto drop; 1997 } 1998 } 1999 break; 2000 2001 /* 2002 * If the state is SYN_SENT: 2003 * if seg contains a RST with valid ACK (SEQ.ACK has already 2004 * been verified), then drop the connection. 2005 * if seg contains a RST without an ACK, drop the seg. 2006 * if seg does not contain SYN, then drop the seg. 2007 * Otherwise this is an acceptable SYN segment 2008 * initialize tp->rcv_nxt and tp->irs 2009 * if seg contains ack then advance tp->snd_una 2010 * if seg contains an ECE and ECN support is enabled, the stream 2011 * is ECN capable. 2012 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2013 * arrange for segment to be acked (eventually) 2014 * continue processing rest of data/controls, beginning with URG 2015 */ 2016 case TCPS_SYN_SENT: 2017 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2018 TCP_PROBE5(connect__refused, NULL, tp, 2019 m, tp, th); 2020 tp = tcp_drop(tp, ECONNREFUSED); 2021 } 2022 if (thflags & TH_RST) 2023 goto drop; 2024 if (!(thflags & TH_SYN)) 2025 goto drop; 2026 2027 tp->irs = th->th_seq; 2028 tcp_rcvseqinit(tp); 2029 if (thflags & TH_ACK) { 2030 int tfo_partial_ack = 0; 2031 2032 TCPSTAT_INC(tcps_connects); 2033 soisconnected(so); 2034 #ifdef MAC 2035 mac_socketpeer_set_from_mbuf(m, so); 2036 #endif 2037 /* Do window scaling on this connection? */ 2038 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2039 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2040 tp->rcv_scale = tp->request_r_scale; 2041 } 2042 tp->rcv_adv += min(tp->rcv_wnd, 2043 TCP_MAXWIN << tp->rcv_scale); 2044 tp->snd_una++; /* SYN is acked */ 2045 /* 2046 * If not all the data that was sent in the TFO SYN 2047 * has been acked, resend the remainder right away. 2048 */ 2049 if (IS_FASTOPEN(tp->t_flags) && 2050 (tp->snd_una != tp->snd_max)) { 2051 tp->snd_nxt = th->th_ack; 2052 tfo_partial_ack = 1; 2053 } 2054 /* 2055 * If there's data, delay ACK; if there's also a FIN 2056 * ACKNOW will be turned on later. 2057 */ 2058 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2059 tcp_timer_activate(tp, TT_DELACK, 2060 tcp_delacktime); 2061 else 2062 tp->t_flags |= TF_ACKNOW; 2063 2064 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2065 2066 /* 2067 * Received <SYN,ACK> in SYN_SENT[*] state. 2068 * Transitions: 2069 * SYN_SENT --> ESTABLISHED 2070 * SYN_SENT* --> FIN_WAIT_1 2071 */ 2072 tp->t_starttime = ticks; 2073 if (tp->t_flags & TF_NEEDFIN) { 2074 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2075 tp->t_flags &= ~TF_NEEDFIN; 2076 thflags &= ~TH_SYN; 2077 } else { 2078 tcp_state_change(tp, TCPS_ESTABLISHED); 2079 TCP_PROBE5(connect__established, NULL, tp, 2080 m, tp, th); 2081 cc_conn_init(tp); 2082 tcp_timer_activate(tp, TT_KEEP, 2083 TP_KEEPIDLE(tp)); 2084 } 2085 } else { 2086 /* 2087 * Received initial SYN in SYN-SENT[*] state => 2088 * simultaneous open. 2089 * If it succeeds, connection is * half-synchronized. 2090 * Otherwise, do 3-way handshake: 2091 * SYN-SENT -> SYN-RECEIVED 2092 * SYN-SENT* -> SYN-RECEIVED* 2093 */ 2094 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2095 tcp_timer_activate(tp, TT_REXMT, 0); 2096 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2097 } 2098 2099 INP_WLOCK_ASSERT(tp->t_inpcb); 2100 2101 /* 2102 * Advance th->th_seq to correspond to first data byte. 2103 * If data, trim to stay within window, 2104 * dropping FIN if necessary. 2105 */ 2106 th->th_seq++; 2107 if (tlen > tp->rcv_wnd) { 2108 todrop = tlen - tp->rcv_wnd; 2109 m_adj(m, -todrop); 2110 tlen = tp->rcv_wnd; 2111 thflags &= ~TH_FIN; 2112 TCPSTAT_INC(tcps_rcvpackafterwin); 2113 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2114 } 2115 tp->snd_wl1 = th->th_seq - 1; 2116 tp->rcv_up = th->th_seq; 2117 /* 2118 * Client side of transaction: already sent SYN and data. 2119 * If the remote host used T/TCP to validate the SYN, 2120 * our data will be ACK'd; if so, enter normal data segment 2121 * processing in the middle of step 5, ack processing. 2122 * Otherwise, goto step 6. 2123 */ 2124 if (thflags & TH_ACK) 2125 goto process_ACK; 2126 2127 goto step6; 2128 2129 /* 2130 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2131 * do normal processing. 2132 * 2133 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2134 */ 2135 case TCPS_LAST_ACK: 2136 case TCPS_CLOSING: 2137 break; /* continue normal processing */ 2138 } 2139 2140 /* 2141 * States other than LISTEN or SYN_SENT. 2142 * First check the RST flag and sequence number since reset segments 2143 * are exempt from the timestamp and connection count tests. This 2144 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2145 * below which allowed reset segments in half the sequence space 2146 * to fall though and be processed (which gives forged reset 2147 * segments with a random sequence number a 50 percent chance of 2148 * killing a connection). 2149 * Then check timestamp, if present. 2150 * Then check the connection count, if present. 2151 * Then check that at least some bytes of segment are within 2152 * receive window. If segment begins before rcv_nxt, 2153 * drop leading data (and SYN); if nothing left, just ack. 2154 */ 2155 if (thflags & TH_RST) { 2156 /* 2157 * RFC5961 Section 3.2 2158 * 2159 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2160 * - If RST is in window, we send challenge ACK. 2161 * 2162 * Note: to take into account delayed ACKs, we should 2163 * test against last_ack_sent instead of rcv_nxt. 2164 * Note 2: we handle special case of closed window, not 2165 * covered by the RFC. 2166 */ 2167 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2168 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2169 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2170 KASSERT(tp->t_state != TCPS_SYN_SENT, 2171 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2172 __func__, th, tp)); 2173 2174 if (V_tcp_insecure_rst || 2175 tp->last_ack_sent == th->th_seq) { 2176 TCPSTAT_INC(tcps_drops); 2177 /* Drop the connection. */ 2178 switch (tp->t_state) { 2179 case TCPS_SYN_RECEIVED: 2180 so->so_error = ECONNREFUSED; 2181 goto close; 2182 case TCPS_ESTABLISHED: 2183 case TCPS_FIN_WAIT_1: 2184 case TCPS_FIN_WAIT_2: 2185 case TCPS_CLOSE_WAIT: 2186 case TCPS_CLOSING: 2187 case TCPS_LAST_ACK: 2188 so->so_error = ECONNRESET; 2189 close: 2190 /* FALLTHROUGH */ 2191 default: 2192 tp = tcp_close(tp); 2193 } 2194 } else { 2195 TCPSTAT_INC(tcps_badrst); 2196 /* Send challenge ACK. */ 2197 tcp_respond(tp, mtod(m, void *), th, m, 2198 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2199 tp->last_ack_sent = tp->rcv_nxt; 2200 m = NULL; 2201 } 2202 } 2203 goto drop; 2204 } 2205 2206 /* 2207 * RFC5961 Section 4.2 2208 * Send challenge ACK for any SYN in synchronized state. 2209 */ 2210 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2211 tp->t_state != TCPS_SYN_RECEIVED) { 2212 TCPSTAT_INC(tcps_badsyn); 2213 if (V_tcp_insecure_syn && 2214 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2215 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2216 tp = tcp_drop(tp, ECONNRESET); 2217 rstreason = BANDLIM_UNLIMITED; 2218 } else { 2219 /* Send challenge ACK. */ 2220 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2221 tp->snd_nxt, TH_ACK); 2222 tp->last_ack_sent = tp->rcv_nxt; 2223 m = NULL; 2224 } 2225 goto drop; 2226 } 2227 2228 /* 2229 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2230 * and it's less than ts_recent, drop it. 2231 */ 2232 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2233 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2234 /* Check to see if ts_recent is over 24 days old. */ 2235 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2236 /* 2237 * Invalidate ts_recent. If this segment updates 2238 * ts_recent, the age will be reset later and ts_recent 2239 * will get a valid value. If it does not, setting 2240 * ts_recent to zero will at least satisfy the 2241 * requirement that zero be placed in the timestamp 2242 * echo reply when ts_recent isn't valid. The 2243 * age isn't reset until we get a valid ts_recent 2244 * because we don't want out-of-order segments to be 2245 * dropped when ts_recent is old. 2246 */ 2247 tp->ts_recent = 0; 2248 } else { 2249 TCPSTAT_INC(tcps_rcvduppack); 2250 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2251 TCPSTAT_INC(tcps_pawsdrop); 2252 if (tlen) 2253 goto dropafterack; 2254 goto drop; 2255 } 2256 } 2257 2258 /* 2259 * In the SYN-RECEIVED state, validate that the packet belongs to 2260 * this connection before trimming the data to fit the receive 2261 * window. Check the sequence number versus IRS since we know 2262 * the sequence numbers haven't wrapped. This is a partial fix 2263 * for the "LAND" DoS attack. 2264 */ 2265 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2266 rstreason = BANDLIM_RST_OPENPORT; 2267 goto dropwithreset; 2268 } 2269 2270 todrop = tp->rcv_nxt - th->th_seq; 2271 if (todrop > 0) { 2272 if (thflags & TH_SYN) { 2273 thflags &= ~TH_SYN; 2274 th->th_seq++; 2275 if (th->th_urp > 1) 2276 th->th_urp--; 2277 else 2278 thflags &= ~TH_URG; 2279 todrop--; 2280 } 2281 /* 2282 * Following if statement from Stevens, vol. 2, p. 960. 2283 */ 2284 if (todrop > tlen 2285 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2286 /* 2287 * Any valid FIN must be to the left of the window. 2288 * At this point the FIN must be a duplicate or out 2289 * of sequence; drop it. 2290 */ 2291 thflags &= ~TH_FIN; 2292 2293 /* 2294 * Send an ACK to resynchronize and drop any data. 2295 * But keep on processing for RST or ACK. 2296 */ 2297 tp->t_flags |= TF_ACKNOW; 2298 todrop = tlen; 2299 TCPSTAT_INC(tcps_rcvduppack); 2300 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2301 } else { 2302 TCPSTAT_INC(tcps_rcvpartduppack); 2303 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2304 } 2305 /* 2306 * DSACK - add SACK block for dropped range 2307 */ 2308 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2309 tcp_update_sack_list(tp, th->th_seq, 2310 th->th_seq + todrop); 2311 /* 2312 * ACK now, as the next in-sequence segment 2313 * will clear the DSACK block again 2314 */ 2315 tp->t_flags |= TF_ACKNOW; 2316 } 2317 drop_hdrlen += todrop; /* drop from the top afterwards */ 2318 th->th_seq += todrop; 2319 tlen -= todrop; 2320 if (th->th_urp > todrop) 2321 th->th_urp -= todrop; 2322 else { 2323 thflags &= ~TH_URG; 2324 th->th_urp = 0; 2325 } 2326 } 2327 2328 /* 2329 * If new data are received on a connection after the 2330 * user processes are gone, then RST the other end. 2331 */ 2332 if ((tp->t_flags & TF_CLOSED) && tlen) { 2333 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2334 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2335 "after socket was closed, " 2336 "sending RST and removing tcpcb\n", 2337 s, __func__, tcpstates[tp->t_state], tlen); 2338 free(s, M_TCPLOG); 2339 } 2340 tp = tcp_close(tp); 2341 TCPSTAT_INC(tcps_rcvafterclose); 2342 rstreason = BANDLIM_UNLIMITED; 2343 goto dropwithreset; 2344 } 2345 2346 /* 2347 * If segment ends after window, drop trailing data 2348 * (and PUSH and FIN); if nothing left, just ACK. 2349 */ 2350 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2351 if (todrop > 0) { 2352 TCPSTAT_INC(tcps_rcvpackafterwin); 2353 if (todrop >= tlen) { 2354 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2355 /* 2356 * If window is closed can only take segments at 2357 * window edge, and have to drop data and PUSH from 2358 * incoming segments. Continue processing, but 2359 * remember to ack. Otherwise, drop segment 2360 * and ack. 2361 */ 2362 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2363 tp->t_flags |= TF_ACKNOW; 2364 TCPSTAT_INC(tcps_rcvwinprobe); 2365 } else 2366 goto dropafterack; 2367 } else 2368 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2369 m_adj(m, -todrop); 2370 tlen -= todrop; 2371 thflags &= ~(TH_PUSH|TH_FIN); 2372 } 2373 2374 /* 2375 * If last ACK falls within this segment's sequence numbers, 2376 * record its timestamp. 2377 * NOTE: 2378 * 1) That the test incorporates suggestions from the latest 2379 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2380 * 2) That updating only on newer timestamps interferes with 2381 * our earlier PAWS tests, so this check should be solely 2382 * predicated on the sequence space of this segment. 2383 * 3) That we modify the segment boundary check to be 2384 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2385 * instead of RFC1323's 2386 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2387 * This modified check allows us to overcome RFC1323's 2388 * limitations as described in Stevens TCP/IP Illustrated 2389 * Vol. 2 p.869. In such cases, we can still calculate the 2390 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2391 */ 2392 if ((to.to_flags & TOF_TS) != 0 && 2393 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2394 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2395 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2396 tp->ts_recent_age = tcp_ts_getticks(); 2397 tp->ts_recent = to.to_tsval; 2398 } 2399 2400 /* 2401 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2402 * flag is on (half-synchronized state), then queue data for 2403 * later processing; else drop segment and return. 2404 */ 2405 if ((thflags & TH_ACK) == 0) { 2406 if (tp->t_state == TCPS_SYN_RECEIVED || 2407 (tp->t_flags & TF_NEEDSYN)) { 2408 if (tp->t_state == TCPS_SYN_RECEIVED && 2409 IS_FASTOPEN(tp->t_flags)) { 2410 tp->snd_wnd = tiwin; 2411 cc_conn_init(tp); 2412 } 2413 goto step6; 2414 } else if (tp->t_flags & TF_ACKNOW) 2415 goto dropafterack; 2416 else 2417 goto drop; 2418 } 2419 2420 /* 2421 * Ack processing. 2422 */ 2423 switch (tp->t_state) { 2424 /* 2425 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2426 * ESTABLISHED state and continue processing. 2427 * The ACK was checked above. 2428 */ 2429 case TCPS_SYN_RECEIVED: 2430 2431 TCPSTAT_INC(tcps_connects); 2432 soisconnected(so); 2433 /* Do window scaling? */ 2434 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2435 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2436 tp->rcv_scale = tp->request_r_scale; 2437 } 2438 tp->snd_wnd = tiwin; 2439 /* 2440 * Make transitions: 2441 * SYN-RECEIVED -> ESTABLISHED 2442 * SYN-RECEIVED* -> FIN-WAIT-1 2443 */ 2444 tp->t_starttime = ticks; 2445 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2446 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2447 tp->t_tfo_pending = NULL; 2448 } 2449 if (tp->t_flags & TF_NEEDFIN) { 2450 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2451 tp->t_flags &= ~TF_NEEDFIN; 2452 } else { 2453 tcp_state_change(tp, TCPS_ESTABLISHED); 2454 TCP_PROBE5(accept__established, NULL, tp, 2455 m, tp, th); 2456 /* 2457 * TFO connections call cc_conn_init() during SYN 2458 * processing. Calling it again here for such 2459 * connections is not harmless as it would undo the 2460 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2461 * is retransmitted. 2462 */ 2463 if (!IS_FASTOPEN(tp->t_flags)) 2464 cc_conn_init(tp); 2465 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2466 } 2467 /* 2468 * Account for the ACK of our SYN prior to 2469 * regular ACK processing below, except for 2470 * simultaneous SYN, which is handled later. 2471 */ 2472 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2473 incforsyn = 1; 2474 /* 2475 * If segment contains data or ACK, will call tcp_reass() 2476 * later; if not, do so now to pass queued data to user. 2477 */ 2478 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2479 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2480 (struct mbuf *)0); 2481 tcp_handle_wakeup(tp, so); 2482 } 2483 tp->snd_wl1 = th->th_seq - 1; 2484 /* FALLTHROUGH */ 2485 2486 /* 2487 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2488 * ACKs. If the ack is in the range 2489 * tp->snd_una < th->th_ack <= tp->snd_max 2490 * then advance tp->snd_una to th->th_ack and drop 2491 * data from the retransmission queue. If this ACK reflects 2492 * more up to date window information we update our window information. 2493 */ 2494 case TCPS_ESTABLISHED: 2495 case TCPS_FIN_WAIT_1: 2496 case TCPS_FIN_WAIT_2: 2497 case TCPS_CLOSE_WAIT: 2498 case TCPS_CLOSING: 2499 case TCPS_LAST_ACK: 2500 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2501 TCPSTAT_INC(tcps_rcvacktoomuch); 2502 goto dropafterack; 2503 } 2504 if ((tp->t_flags & TF_SACK_PERMIT) && 2505 ((to.to_flags & TOF_SACK) || 2506 !TAILQ_EMPTY(&tp->snd_holes))) { 2507 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && 2508 (tp->t_flags & TF_LRD)) { 2509 tcp_sack_lost_retransmission(tp, th); 2510 } 2511 } else 2512 /* 2513 * Reset the value so that previous (valid) value 2514 * from the last ack with SACK doesn't get used. 2515 */ 2516 tp->sackhint.sacked_bytes = 0; 2517 2518 #ifdef TCP_HHOOK 2519 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2520 hhook_run_tcp_est_in(tp, th, &to); 2521 #endif 2522 2523 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2524 maxseg = tcp_maxseg(tp); 2525 if (tlen == 0 && 2526 (tiwin == tp->snd_wnd || 2527 (tp->t_flags & TF_SACK_PERMIT))) { 2528 /* 2529 * If this is the first time we've seen a 2530 * FIN from the remote, this is not a 2531 * duplicate and it needs to be processed 2532 * normally. This happens during a 2533 * simultaneous close. 2534 */ 2535 if ((thflags & TH_FIN) && 2536 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2537 tp->t_dupacks = 0; 2538 break; 2539 } 2540 TCPSTAT_INC(tcps_rcvdupack); 2541 /* 2542 * If we have outstanding data (other than 2543 * a window probe), this is a completely 2544 * duplicate ack (ie, window info didn't 2545 * change and FIN isn't set), 2546 * the ack is the biggest we've 2547 * seen and we've seen exactly our rexmt 2548 * threshold of them, assume a packet 2549 * has been dropped and retransmit it. 2550 * Kludge snd_nxt & the congestion 2551 * window so we send only this one 2552 * packet. 2553 * 2554 * We know we're losing at the current 2555 * window size so do congestion avoidance 2556 * (set ssthresh to half the current window 2557 * and pull our congestion window back to 2558 * the new ssthresh). 2559 * 2560 * Dup acks mean that packets have left the 2561 * network (they're now cached at the receiver) 2562 * so bump cwnd by the amount in the receiver 2563 * to keep a constant cwnd packets in the 2564 * network. 2565 * 2566 * When using TCP ECN, notify the peer that 2567 * we reduced the cwnd. 2568 */ 2569 /* 2570 * Following 2 kinds of acks should not affect 2571 * dupack counting: 2572 * 1) Old acks 2573 * 2) Acks with SACK but without any new SACK 2574 * information in them. These could result from 2575 * any anomaly in the network like a switch 2576 * duplicating packets or a possible DoS attack. 2577 */ 2578 if (th->th_ack != tp->snd_una || 2579 ((tp->t_flags & TF_SACK_PERMIT) && 2580 (to.to_flags & TOF_SACK) && 2581 !sack_changed)) 2582 break; 2583 else if (!tcp_timer_active(tp, TT_REXMT)) 2584 tp->t_dupacks = 0; 2585 else if (++tp->t_dupacks > tcprexmtthresh || 2586 IN_FASTRECOVERY(tp->t_flags)) { 2587 cc_ack_received(tp, th, nsegs, 2588 CC_DUPACK); 2589 if (V_tcp_do_prr && 2590 IN_FASTRECOVERY(tp->t_flags)) { 2591 tcp_do_prr_ack(tp, th, &to); 2592 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2593 (to.to_flags & TOF_SACK) && 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 ((tp->t_flags & TF_SACK_PERMIT) && 2667 (to.to_flags & TOF_SACK)) { 2668 tp->sackhint.prr_delivered = 2669 tp->sackhint.sacked_bytes; 2670 } else { 2671 tp->sackhint.prr_delivered = 2672 imin(tp->snd_max - tp->snd_una, 2673 imin(INT_MAX / 65536, 2674 tp->t_dupacks) * maxseg); 2675 } 2676 tp->sackhint.recover_fs = max(1, 2677 tp->snd_nxt - tp->snd_una); 2678 } 2679 if ((tp->t_flags & TF_SACK_PERMIT) && 2680 (to.to_flags & TOF_SACK)) { 2681 TCPSTAT_INC( 2682 tcps_sack_recovery_episode); 2683 tp->snd_recover = tp->snd_nxt; 2684 tp->snd_cwnd = maxseg; 2685 (void) tcp_output(tp); 2686 if (SEQ_GT(th->th_ack, tp->snd_una)) 2687 goto resume_partialack; 2688 goto drop; 2689 } 2690 tp->snd_nxt = th->th_ack; 2691 tp->snd_cwnd = maxseg; 2692 (void) tcp_output(tp); 2693 KASSERT(tp->snd_limited <= 2, 2694 ("%s: tp->snd_limited too big", 2695 __func__)); 2696 tp->snd_cwnd = tp->snd_ssthresh + 2697 maxseg * 2698 (tp->t_dupacks - tp->snd_limited); 2699 if (SEQ_GT(onxt, tp->snd_nxt)) 2700 tp->snd_nxt = onxt; 2701 goto drop; 2702 } else if (V_tcp_do_rfc3042) { 2703 /* 2704 * Process first and second duplicate 2705 * ACKs. Each indicates a segment 2706 * leaving the network, creating room 2707 * for more. Make sure we can send a 2708 * packet on reception of each duplicate 2709 * ACK by increasing snd_cwnd by one 2710 * segment. Restore the original 2711 * snd_cwnd after packet transmission. 2712 */ 2713 cc_ack_received(tp, th, nsegs, 2714 CC_DUPACK); 2715 uint32_t oldcwnd = tp->snd_cwnd; 2716 tcp_seq oldsndmax = tp->snd_max; 2717 u_int sent; 2718 int avail; 2719 2720 KASSERT(tp->t_dupacks == 1 || 2721 tp->t_dupacks == 2, 2722 ("%s: dupacks not 1 or 2", 2723 __func__)); 2724 if (tp->t_dupacks == 1) 2725 tp->snd_limited = 0; 2726 tp->snd_cwnd = 2727 (tp->snd_nxt - tp->snd_una) + 2728 (tp->t_dupacks - tp->snd_limited) * 2729 maxseg; 2730 /* 2731 * Only call tcp_output when there 2732 * is new data available to be sent. 2733 * Otherwise we would send pure ACKs. 2734 */ 2735 SOCKBUF_LOCK(&so->so_snd); 2736 avail = sbavail(&so->so_snd) - 2737 (tp->snd_nxt - tp->snd_una); 2738 SOCKBUF_UNLOCK(&so->so_snd); 2739 if (avail > 0) 2740 (void) tcp_output(tp); 2741 sent = tp->snd_max - oldsndmax; 2742 if (sent > maxseg) { 2743 KASSERT((tp->t_dupacks == 2 && 2744 tp->snd_limited == 0) || 2745 (sent == maxseg + 1 && 2746 tp->t_flags & TF_SENTFIN), 2747 ("%s: sent too much", 2748 __func__)); 2749 tp->snd_limited = 2; 2750 } else if (sent > 0) 2751 ++tp->snd_limited; 2752 tp->snd_cwnd = oldcwnd; 2753 goto drop; 2754 } 2755 } 2756 break; 2757 } else { 2758 /* 2759 * This ack is advancing the left edge, reset the 2760 * counter. 2761 */ 2762 tp->t_dupacks = 0; 2763 /* 2764 * If this ack also has new SACK info, increment the 2765 * counter as per rfc6675. The variable 2766 * sack_changed tracks all changes to the SACK 2767 * scoreboard, including when partial ACKs without 2768 * SACK options are received, and clear the scoreboard 2769 * from the left side. Such partial ACKs should not be 2770 * counted as dupacks here. 2771 */ 2772 if ((tp->t_flags & TF_SACK_PERMIT) && 2773 (to.to_flags & TOF_SACK) && 2774 sack_changed) { 2775 tp->t_dupacks++; 2776 /* limit overhead by setting maxseg last */ 2777 if (!IN_FASTRECOVERY(tp->t_flags) && 2778 (tp->sackhint.sacked_bytes > 2779 ((tcprexmtthresh - 1) * 2780 (maxseg = tcp_maxseg(tp))))) { 2781 goto enter_recovery; 2782 } 2783 } 2784 } 2785 2786 resume_partialack: 2787 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2788 ("%s: th_ack <= snd_una", __func__)); 2789 2790 /* 2791 * If the congestion window was inflated to account 2792 * for the other side's cached packets, retract it. 2793 */ 2794 if (IN_FASTRECOVERY(tp->t_flags)) { 2795 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2796 if (tp->t_flags & TF_SACK_PERMIT) 2797 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2798 tcp_timer_activate(tp, TT_REXMT, 0); 2799 tp->t_rtttime = 0; 2800 tcp_do_prr_ack(tp, th, &to); 2801 tp->t_flags |= TF_ACKNOW; 2802 (void) tcp_output(tp); 2803 } else 2804 tcp_sack_partialack(tp, th); 2805 else 2806 tcp_newreno_partial_ack(tp, th); 2807 } else 2808 cc_post_recovery(tp, th); 2809 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2810 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2811 if (V_tcp_do_prr) { 2812 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2813 tp->snd_fack = th->th_ack; 2814 tcp_do_prr_ack(tp, th, &to); 2815 (void) tcp_output(tp); 2816 } 2817 } else 2818 cc_post_recovery(tp, th); 2819 } 2820 /* 2821 * If we reach this point, ACK is not a duplicate, 2822 * i.e., it ACKs something we sent. 2823 */ 2824 if (tp->t_flags & TF_NEEDSYN) { 2825 /* 2826 * T/TCP: Connection was half-synchronized, and our 2827 * SYN has been ACK'd (so connection is now fully 2828 * synchronized). Go to non-starred state, 2829 * increment snd_una for ACK of SYN, and check if 2830 * we can do window scaling. 2831 */ 2832 tp->t_flags &= ~TF_NEEDSYN; 2833 tp->snd_una++; 2834 /* Do window scaling? */ 2835 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2836 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2837 tp->rcv_scale = tp->request_r_scale; 2838 /* Send window already scaled. */ 2839 } 2840 } 2841 2842 process_ACK: 2843 INP_WLOCK_ASSERT(tp->t_inpcb); 2844 2845 /* 2846 * Adjust for the SYN bit in sequence space, 2847 * but don't account for it in cwnd calculations. 2848 * This is for the SYN_RECEIVED, non-simultaneous 2849 * SYN case. SYN_SENT and simultaneous SYN are 2850 * treated elsewhere. 2851 */ 2852 if (incforsyn) 2853 tp->snd_una++; 2854 acked = BYTES_THIS_ACK(tp, th); 2855 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2856 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2857 tp->snd_una, th->th_ack, tp, m)); 2858 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2859 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2860 2861 /* 2862 * If we just performed our first retransmit, and the ACK 2863 * arrives within our recovery window, then it was a mistake 2864 * to do the retransmit in the first place. Recover our 2865 * original cwnd and ssthresh, and proceed to transmit where 2866 * we left off. 2867 */ 2868 if (tp->t_rxtshift == 1 && 2869 tp->t_flags & TF_PREVVALID && 2870 tp->t_badrxtwin != 0 && 2871 to.to_flags & TOF_TS && 2872 to.to_tsecr != 0 && 2873 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2874 cc_cong_signal(tp, th, CC_RTO_ERR); 2875 2876 /* 2877 * If we have a timestamp reply, update smoothed 2878 * round trip time. If no timestamp is present but 2879 * transmit timer is running and timed sequence 2880 * number was acked, update smoothed round trip time. 2881 * Since we now have an rtt measurement, cancel the 2882 * timer backoff (cf., Phil Karn's retransmit alg.). 2883 * Recompute the initial retransmit timer. 2884 * 2885 * Some boxes send broken timestamp replies 2886 * during the SYN+ACK phase, ignore 2887 * timestamps of 0 or we could calculate a 2888 * huge RTT and blow up the retransmit timer. 2889 */ 2890 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2891 uint32_t t; 2892 2893 t = tcp_ts_getticks() - to.to_tsecr; 2894 if (!tp->t_rttlow || tp->t_rttlow > t) 2895 tp->t_rttlow = t; 2896 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2897 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2898 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2899 tp->t_rttlow = ticks - tp->t_rtttime; 2900 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2901 } 2902 2903 /* 2904 * If all outstanding data is acked, stop retransmit 2905 * timer and remember to restart (more output or persist). 2906 * If there is more data to be acked, restart retransmit 2907 * timer, using current (possibly backed-off) value. 2908 */ 2909 if (th->th_ack == tp->snd_max) { 2910 tcp_timer_activate(tp, TT_REXMT, 0); 2911 needoutput = 1; 2912 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2913 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2914 2915 /* 2916 * If no data (only SYN) was ACK'd, 2917 * skip rest of ACK processing. 2918 */ 2919 if (acked == 0) 2920 goto step6; 2921 2922 /* 2923 * Let the congestion control algorithm update congestion 2924 * control related information. This typically means increasing 2925 * the congestion window. 2926 */ 2927 cc_ack_received(tp, th, nsegs, CC_ACK); 2928 2929 SOCKBUF_LOCK(&so->so_snd); 2930 if (acked > sbavail(&so->so_snd)) { 2931 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2932 tp->snd_wnd -= sbavail(&so->so_snd); 2933 else 2934 tp->snd_wnd = 0; 2935 mfree = sbcut_locked(&so->so_snd, 2936 (int)sbavail(&so->so_snd)); 2937 ourfinisacked = 1; 2938 } else { 2939 mfree = sbcut_locked(&so->so_snd, acked); 2940 if (tp->snd_wnd >= (uint32_t) acked) 2941 tp->snd_wnd -= acked; 2942 else 2943 tp->snd_wnd = 0; 2944 ourfinisacked = 0; 2945 } 2946 /* NB: sowwakeup_locked() does an implicit unlock. */ 2947 sowwakeup_locked(so); 2948 m_freem(mfree); 2949 /* Detect una wraparound. */ 2950 if (!IN_RECOVERY(tp->t_flags) && 2951 SEQ_GT(tp->snd_una, tp->snd_recover) && 2952 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2953 tp->snd_recover = th->th_ack - 1; 2954 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2955 if (IN_RECOVERY(tp->t_flags) && 2956 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2957 EXIT_RECOVERY(tp->t_flags); 2958 } 2959 tp->snd_una = th->th_ack; 2960 if (tp->t_flags & TF_SACK_PERMIT) { 2961 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2962 tp->snd_recover = tp->snd_una; 2963 } 2964 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2965 tp->snd_nxt = tp->snd_una; 2966 2967 switch (tp->t_state) { 2968 /* 2969 * In FIN_WAIT_1 STATE in addition to the processing 2970 * for the ESTABLISHED state if our FIN is now acknowledged 2971 * then enter FIN_WAIT_2. 2972 */ 2973 case TCPS_FIN_WAIT_1: 2974 if (ourfinisacked) { 2975 /* 2976 * If we can't receive any more 2977 * data, then closing user can proceed. 2978 * Starting the timer is contrary to the 2979 * specification, but if we don't get a FIN 2980 * we'll hang forever. 2981 * 2982 * XXXjl: 2983 * we should release the tp also, and use a 2984 * compressed state. 2985 */ 2986 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2987 soisdisconnected(so); 2988 tcp_timer_activate(tp, TT_2MSL, 2989 (tcp_fast_finwait2_recycle ? 2990 tcp_finwait2_timeout : 2991 TP_MAXIDLE(tp))); 2992 } 2993 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2994 } 2995 break; 2996 2997 /* 2998 * In CLOSING STATE in addition to the processing for 2999 * the ESTABLISHED state if the ACK acknowledges our FIN 3000 * then enter the TIME-WAIT state, otherwise ignore 3001 * the segment. 3002 */ 3003 case TCPS_CLOSING: 3004 if (ourfinisacked) { 3005 tcp_twstart(tp); 3006 m_freem(m); 3007 return; 3008 } 3009 break; 3010 3011 /* 3012 * In LAST_ACK, we may still be waiting for data to drain 3013 * and/or to be acked, as well as for the ack of our FIN. 3014 * If our FIN is now acknowledged, delete the TCB, 3015 * enter the closed state and return. 3016 */ 3017 case TCPS_LAST_ACK: 3018 if (ourfinisacked) { 3019 tp = tcp_close(tp); 3020 goto drop; 3021 } 3022 break; 3023 } 3024 } 3025 3026 step6: 3027 INP_WLOCK_ASSERT(tp->t_inpcb); 3028 3029 /* 3030 * Update window information. 3031 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3032 */ 3033 if ((thflags & TH_ACK) && 3034 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3035 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3036 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3037 /* keep track of pure window updates */ 3038 if (tlen == 0 && 3039 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3040 TCPSTAT_INC(tcps_rcvwinupd); 3041 tp->snd_wnd = tiwin; 3042 tp->snd_wl1 = th->th_seq; 3043 tp->snd_wl2 = th->th_ack; 3044 if (tp->snd_wnd > tp->max_sndwnd) 3045 tp->max_sndwnd = tp->snd_wnd; 3046 needoutput = 1; 3047 } 3048 3049 /* 3050 * Process segments with URG. 3051 */ 3052 if ((thflags & TH_URG) && th->th_urp && 3053 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3054 /* 3055 * This is a kludge, but if we receive and accept 3056 * random urgent pointers, we'll crash in 3057 * soreceive. It's hard to imagine someone 3058 * actually wanting to send this much urgent data. 3059 */ 3060 SOCKBUF_LOCK(&so->so_rcv); 3061 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3062 th->th_urp = 0; /* XXX */ 3063 thflags &= ~TH_URG; /* XXX */ 3064 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3065 goto dodata; /* XXX */ 3066 } 3067 /* 3068 * If this segment advances the known urgent pointer, 3069 * then mark the data stream. This should not happen 3070 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3071 * a FIN has been received from the remote side. 3072 * In these states we ignore the URG. 3073 * 3074 * According to RFC961 (Assigned Protocols), 3075 * the urgent pointer points to the last octet 3076 * of urgent data. We continue, however, 3077 * to consider it to indicate the first octet 3078 * of data past the urgent section as the original 3079 * spec states (in one of two places). 3080 */ 3081 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3082 tp->rcv_up = th->th_seq + th->th_urp; 3083 so->so_oobmark = sbavail(&so->so_rcv) + 3084 (tp->rcv_up - tp->rcv_nxt) - 1; 3085 if (so->so_oobmark == 0) 3086 so->so_rcv.sb_state |= SBS_RCVATMARK; 3087 sohasoutofband(so); 3088 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3089 } 3090 SOCKBUF_UNLOCK(&so->so_rcv); 3091 /* 3092 * Remove out of band data so doesn't get presented to user. 3093 * This can happen independent of advancing the URG pointer, 3094 * but if two URG's are pending at once, some out-of-band 3095 * data may creep in... ick. 3096 */ 3097 if (th->th_urp <= (uint32_t)tlen && 3098 !(so->so_options & SO_OOBINLINE)) { 3099 /* hdr drop is delayed */ 3100 tcp_pulloutofband(so, th, m, drop_hdrlen); 3101 } 3102 } else { 3103 /* 3104 * If no out of band data is expected, 3105 * pull receive urgent pointer along 3106 * with the receive window. 3107 */ 3108 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3109 tp->rcv_up = tp->rcv_nxt; 3110 } 3111 dodata: /* XXX */ 3112 INP_WLOCK_ASSERT(tp->t_inpcb); 3113 3114 /* 3115 * Process the segment text, merging it into the TCP sequencing queue, 3116 * and arranging for acknowledgment of receipt if necessary. 3117 * This process logically involves adjusting tp->rcv_wnd as data 3118 * is presented to the user (this happens in tcp_usrreq.c, 3119 * case PRU_RCVD). If a FIN has already been received on this 3120 * connection then we just ignore the text. 3121 */ 3122 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3123 IS_FASTOPEN(tp->t_flags)); 3124 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3125 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3126 tcp_seq save_start = th->th_seq; 3127 tcp_seq save_rnxt = tp->rcv_nxt; 3128 int save_tlen = tlen; 3129 m_adj(m, drop_hdrlen); /* delayed header drop */ 3130 /* 3131 * Insert segment which includes th into TCP reassembly queue 3132 * with control block tp. Set thflags to whether reassembly now 3133 * includes a segment with FIN. This handles the common case 3134 * inline (segment is the next to be received on an established 3135 * connection, and the queue is empty), avoiding linkage into 3136 * and removal from the queue and repetition of various 3137 * conversions. 3138 * Set DELACK for segments received in order, but ack 3139 * immediately when segments are out of order (so 3140 * fast retransmit can work). 3141 */ 3142 if (th->th_seq == tp->rcv_nxt && 3143 SEGQ_EMPTY(tp) && 3144 (TCPS_HAVEESTABLISHED(tp->t_state) || 3145 tfo_syn)) { 3146 if (DELAY_ACK(tp, tlen) || tfo_syn) 3147 tp->t_flags |= TF_DELACK; 3148 else 3149 tp->t_flags |= TF_ACKNOW; 3150 tp->rcv_nxt += tlen; 3151 if (tlen && 3152 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3153 (tp->t_fbyte_in == 0)) { 3154 tp->t_fbyte_in = ticks; 3155 if (tp->t_fbyte_in == 0) 3156 tp->t_fbyte_in = 1; 3157 if (tp->t_fbyte_out && tp->t_fbyte_in) 3158 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3159 } 3160 thflags = tcp_get_flags(th) & TH_FIN; 3161 TCPSTAT_INC(tcps_rcvpack); 3162 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3163 SOCKBUF_LOCK(&so->so_rcv); 3164 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3165 m_freem(m); 3166 else 3167 sbappendstream_locked(&so->so_rcv, m, 0); 3168 tp->t_flags |= TF_WAKESOR; 3169 } else { 3170 /* 3171 * XXX: Due to the header drop above "th" is 3172 * theoretically invalid by now. Fortunately 3173 * m_adj() doesn't actually frees any mbufs 3174 * when trimming from the head. 3175 */ 3176 tcp_seq temp = save_start; 3177 3178 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3179 tp->t_flags |= TF_ACKNOW; 3180 } 3181 if ((tp->t_flags & TF_SACK_PERMIT) && 3182 (save_tlen > 0) && 3183 TCPS_HAVEESTABLISHED(tp->t_state)) { 3184 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3185 /* 3186 * DSACK actually handled in the fastpath 3187 * above. 3188 */ 3189 tcp_update_sack_list(tp, save_start, 3190 save_start + save_tlen); 3191 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3192 if ((tp->rcv_numsacks >= 1) && 3193 (tp->sackblks[0].end == save_start)) { 3194 /* 3195 * Partial overlap, recorded at todrop 3196 * above. 3197 */ 3198 tcp_update_sack_list(tp, 3199 tp->sackblks[0].start, 3200 tp->sackblks[0].end); 3201 } else { 3202 tcp_update_dsack_list(tp, save_start, 3203 save_start + save_tlen); 3204 } 3205 } else if (tlen >= save_tlen) { 3206 /* Update of sackblks. */ 3207 tcp_update_dsack_list(tp, save_start, 3208 save_start + save_tlen); 3209 } else if (tlen > 0) { 3210 tcp_update_dsack_list(tp, save_start, 3211 save_start + tlen); 3212 } 3213 } 3214 tcp_handle_wakeup(tp, so); 3215 #if 0 3216 /* 3217 * Note the amount of data that peer has sent into 3218 * our window, in order to estimate the sender's 3219 * buffer size. 3220 * XXX: Unused. 3221 */ 3222 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3223 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3224 else 3225 len = so->so_rcv.sb_hiwat; 3226 #endif 3227 } else { 3228 m_freem(m); 3229 thflags &= ~TH_FIN; 3230 } 3231 3232 /* 3233 * If FIN is received ACK the FIN and let the user know 3234 * that the connection is closing. 3235 */ 3236 if (thflags & TH_FIN) { 3237 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3238 /* The socket upcall is handled by socantrcvmore. */ 3239 socantrcvmore(so); 3240 /* 3241 * If connection is half-synchronized 3242 * (ie NEEDSYN flag on) then delay ACK, 3243 * so it may be piggybacked when SYN is sent. 3244 * Otherwise, since we received a FIN then no 3245 * more input can be expected, send ACK now. 3246 */ 3247 if (tp->t_flags & TF_NEEDSYN) 3248 tp->t_flags |= TF_DELACK; 3249 else 3250 tp->t_flags |= TF_ACKNOW; 3251 tp->rcv_nxt++; 3252 } 3253 switch (tp->t_state) { 3254 /* 3255 * In SYN_RECEIVED and ESTABLISHED STATES 3256 * enter the CLOSE_WAIT state. 3257 */ 3258 case TCPS_SYN_RECEIVED: 3259 tp->t_starttime = ticks; 3260 /* FALLTHROUGH */ 3261 case TCPS_ESTABLISHED: 3262 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3263 break; 3264 3265 /* 3266 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3267 * enter the CLOSING state. 3268 */ 3269 case TCPS_FIN_WAIT_1: 3270 tcp_state_change(tp, TCPS_CLOSING); 3271 break; 3272 3273 /* 3274 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3275 * starting the time-wait timer, turning off the other 3276 * standard timers. 3277 */ 3278 case TCPS_FIN_WAIT_2: 3279 tcp_twstart(tp); 3280 return; 3281 } 3282 } 3283 #ifdef TCPDEBUG 3284 if (so->so_options & SO_DEBUG) 3285 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3286 &tcp_savetcp, 0); 3287 #endif 3288 TCP_PROBE3(debug__input, tp, th, m); 3289 3290 /* 3291 * Return any desired output. 3292 */ 3293 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3294 (void) tcp_output(tp); 3295 3296 check_delack: 3297 INP_WLOCK_ASSERT(tp->t_inpcb); 3298 3299 if (tp->t_flags & TF_DELACK) { 3300 tp->t_flags &= ~TF_DELACK; 3301 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3302 } 3303 INP_WUNLOCK(tp->t_inpcb); 3304 return; 3305 3306 dropafterack: 3307 /* 3308 * Generate an ACK dropping incoming segment if it occupies 3309 * sequence space, where the ACK reflects our state. 3310 * 3311 * We can now skip the test for the RST flag since all 3312 * paths to this code happen after packets containing 3313 * RST have been dropped. 3314 * 3315 * In the SYN-RECEIVED state, don't send an ACK unless the 3316 * segment we received passes the SYN-RECEIVED ACK test. 3317 * If it fails send a RST. This breaks the loop in the 3318 * "LAND" DoS attack, and also prevents an ACK storm 3319 * between two listening ports that have been sent forged 3320 * SYN segments, each with the source address of the other. 3321 */ 3322 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3323 (SEQ_GT(tp->snd_una, th->th_ack) || 3324 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3325 rstreason = BANDLIM_RST_OPENPORT; 3326 goto dropwithreset; 3327 } 3328 #ifdef TCPDEBUG 3329 if (so->so_options & SO_DEBUG) 3330 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3331 &tcp_savetcp, 0); 3332 #endif 3333 TCP_PROBE3(debug__input, tp, th, m); 3334 tp->t_flags |= TF_ACKNOW; 3335 (void) tcp_output(tp); 3336 INP_WUNLOCK(tp->t_inpcb); 3337 m_freem(m); 3338 return; 3339 3340 dropwithreset: 3341 if (tp != NULL) { 3342 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3343 INP_WUNLOCK(tp->t_inpcb); 3344 } else 3345 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3346 return; 3347 3348 drop: 3349 /* 3350 * Drop space held by incoming segment and return. 3351 */ 3352 #ifdef TCPDEBUG 3353 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3354 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3355 &tcp_savetcp, 0); 3356 #endif 3357 TCP_PROBE3(debug__input, tp, th, m); 3358 if (tp != NULL) { 3359 INP_WUNLOCK(tp->t_inpcb); 3360 } 3361 m_freem(m); 3362 } 3363 3364 /* 3365 * Issue RST and make ACK acceptable to originator of segment. 3366 * The mbuf must still include the original packet header. 3367 * tp may be NULL. 3368 */ 3369 void 3370 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3371 int tlen, int rstreason) 3372 { 3373 #ifdef INET 3374 struct ip *ip; 3375 #endif 3376 #ifdef INET6 3377 struct ip6_hdr *ip6; 3378 #endif 3379 3380 if (tp != NULL) { 3381 INP_LOCK_ASSERT(tp->t_inpcb); 3382 } 3383 3384 /* Don't bother if destination was broadcast/multicast. */ 3385 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3386 goto drop; 3387 #ifdef INET6 3388 if (mtod(m, struct ip *)->ip_v == 6) { 3389 ip6 = mtod(m, struct ip6_hdr *); 3390 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3391 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3392 goto drop; 3393 /* IPv6 anycast check is done at tcp6_input() */ 3394 } 3395 #endif 3396 #if defined(INET) && defined(INET6) 3397 else 3398 #endif 3399 #ifdef INET 3400 { 3401 ip = mtod(m, struct ip *); 3402 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3403 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3404 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3405 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3406 goto drop; 3407 } 3408 #endif 3409 3410 /* Perform bandwidth limiting. */ 3411 if (badport_bandlim(rstreason) < 0) 3412 goto drop; 3413 3414 /* tcp_respond consumes the mbuf chain. */ 3415 if (tcp_get_flags(th) & TH_ACK) { 3416 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3417 th->th_ack, TH_RST); 3418 } else { 3419 if (tcp_get_flags(th) & TH_SYN) 3420 tlen++; 3421 if (tcp_get_flags(th) & TH_FIN) 3422 tlen++; 3423 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3424 (tcp_seq)0, TH_RST|TH_ACK); 3425 } 3426 return; 3427 drop: 3428 m_freem(m); 3429 } 3430 3431 /* 3432 * Parse TCP options and place in tcpopt. 3433 */ 3434 void 3435 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3436 { 3437 int opt, optlen; 3438 3439 to->to_flags = 0; 3440 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3441 opt = cp[0]; 3442 if (opt == TCPOPT_EOL) 3443 break; 3444 if (opt == TCPOPT_NOP) 3445 optlen = 1; 3446 else { 3447 if (cnt < 2) 3448 break; 3449 optlen = cp[1]; 3450 if (optlen < 2 || optlen > cnt) 3451 break; 3452 } 3453 switch (opt) { 3454 case TCPOPT_MAXSEG: 3455 if (optlen != TCPOLEN_MAXSEG) 3456 continue; 3457 if (!(flags & TO_SYN)) 3458 continue; 3459 to->to_flags |= TOF_MSS; 3460 bcopy((char *)cp + 2, 3461 (char *)&to->to_mss, sizeof(to->to_mss)); 3462 to->to_mss = ntohs(to->to_mss); 3463 break; 3464 case TCPOPT_WINDOW: 3465 if (optlen != TCPOLEN_WINDOW) 3466 continue; 3467 if (!(flags & TO_SYN)) 3468 continue; 3469 to->to_flags |= TOF_SCALE; 3470 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3471 break; 3472 case TCPOPT_TIMESTAMP: 3473 if (optlen != TCPOLEN_TIMESTAMP) 3474 continue; 3475 to->to_flags |= TOF_TS; 3476 bcopy((char *)cp + 2, 3477 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3478 to->to_tsval = ntohl(to->to_tsval); 3479 bcopy((char *)cp + 6, 3480 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3481 to->to_tsecr = ntohl(to->to_tsecr); 3482 break; 3483 case TCPOPT_SIGNATURE: 3484 /* 3485 * In order to reply to a host which has set the 3486 * TCP_SIGNATURE option in its initial SYN, we have 3487 * to record the fact that the option was observed 3488 * here for the syncache code to perform the correct 3489 * response. 3490 */ 3491 if (optlen != TCPOLEN_SIGNATURE) 3492 continue; 3493 to->to_flags |= TOF_SIGNATURE; 3494 to->to_signature = cp + 2; 3495 break; 3496 case TCPOPT_SACK_PERMITTED: 3497 if (optlen != TCPOLEN_SACK_PERMITTED) 3498 continue; 3499 if (!(flags & TO_SYN)) 3500 continue; 3501 if (!V_tcp_do_sack) 3502 continue; 3503 to->to_flags |= TOF_SACKPERM; 3504 break; 3505 case TCPOPT_SACK: 3506 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3507 continue; 3508 if (flags & TO_SYN) 3509 continue; 3510 to->to_flags |= TOF_SACK; 3511 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3512 to->to_sacks = cp + 2; 3513 TCPSTAT_INC(tcps_sack_rcv_blocks); 3514 break; 3515 case TCPOPT_FAST_OPEN: 3516 /* 3517 * Cookie length validation is performed by the 3518 * server side cookie checking code or the client 3519 * side cookie cache update code. 3520 */ 3521 if (!(flags & TO_SYN)) 3522 continue; 3523 if (!V_tcp_fastopen_client_enable && 3524 !V_tcp_fastopen_server_enable) 3525 continue; 3526 to->to_flags |= TOF_FASTOPEN; 3527 to->to_tfo_len = optlen - 2; 3528 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3529 break; 3530 default: 3531 continue; 3532 } 3533 } 3534 } 3535 3536 /* 3537 * Pull out of band byte out of a segment so 3538 * it doesn't appear in the user's data queue. 3539 * It is still reflected in the segment length for 3540 * sequencing purposes. 3541 */ 3542 void 3543 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3544 int off) 3545 { 3546 int cnt = off + th->th_urp - 1; 3547 3548 while (cnt >= 0) { 3549 if (m->m_len > cnt) { 3550 char *cp = mtod(m, caddr_t) + cnt; 3551 struct tcpcb *tp = sototcpcb(so); 3552 3553 INP_WLOCK_ASSERT(tp->t_inpcb); 3554 3555 tp->t_iobc = *cp; 3556 tp->t_oobflags |= TCPOOB_HAVEDATA; 3557 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3558 m->m_len--; 3559 if (m->m_flags & M_PKTHDR) 3560 m->m_pkthdr.len--; 3561 return; 3562 } 3563 cnt -= m->m_len; 3564 m = m->m_next; 3565 if (m == NULL) 3566 break; 3567 } 3568 panic("tcp_pulloutofband"); 3569 } 3570 3571 /* 3572 * Collect new round-trip time estimate 3573 * and update averages and current timeout. 3574 */ 3575 void 3576 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3577 { 3578 int delta; 3579 3580 INP_WLOCK_ASSERT(tp->t_inpcb); 3581 3582 TCPSTAT_INC(tcps_rttupdated); 3583 tp->t_rttupdated++; 3584 #ifdef STATS 3585 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3586 imax(0, rtt * 1000 / hz)); 3587 #endif 3588 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3589 /* 3590 * srtt is stored as fixed point with 5 bits after the 3591 * binary point (i.e., scaled by 8). The following magic 3592 * is equivalent to the smoothing algorithm in rfc793 with 3593 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3594 * point). Adjust rtt to origin 0. 3595 */ 3596 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3597 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3598 3599 if ((tp->t_srtt += delta) <= 0) 3600 tp->t_srtt = 1; 3601 3602 /* 3603 * We accumulate a smoothed rtt variance (actually, a 3604 * smoothed mean difference), then set the retransmit 3605 * timer to smoothed rtt + 4 times the smoothed variance. 3606 * rttvar is stored as fixed point with 4 bits after the 3607 * binary point (scaled by 16). The following is 3608 * equivalent to rfc793 smoothing with an alpha of .75 3609 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3610 * rfc793's wired-in beta. 3611 */ 3612 if (delta < 0) 3613 delta = -delta; 3614 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3615 if ((tp->t_rttvar += delta) <= 0) 3616 tp->t_rttvar = 1; 3617 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3618 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3619 } else { 3620 /* 3621 * No rtt measurement yet - use the unsmoothed rtt. 3622 * Set the variance to half the rtt (so our first 3623 * retransmit happens at 3*rtt). 3624 */ 3625 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3626 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3627 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3628 } 3629 tp->t_rtttime = 0; 3630 tp->t_rxtshift = 0; 3631 3632 /* 3633 * the retransmit should happen at rtt + 4 * rttvar. 3634 * Because of the way we do the smoothing, srtt and rttvar 3635 * will each average +1/2 tick of bias. When we compute 3636 * the retransmit timer, we want 1/2 tick of rounding and 3637 * 1 extra tick because of +-1/2 tick uncertainty in the 3638 * firing of the timer. The bias will give us exactly the 3639 * 1.5 tick we need. But, because the bias is 3640 * statistical, we have to test that we don't drop below 3641 * the minimum feasible timer (which is 2 ticks). 3642 */ 3643 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3644 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3645 3646 /* 3647 * We received an ack for a packet that wasn't retransmitted; 3648 * it is probably safe to discard any error indications we've 3649 * received recently. This isn't quite right, but close enough 3650 * for now (a route might have failed after we sent a segment, 3651 * and the return path might not be symmetrical). 3652 */ 3653 tp->t_softerror = 0; 3654 } 3655 3656 /* 3657 * Determine a reasonable value for maxseg size. 3658 * If the route is known, check route for mtu. 3659 * If none, use an mss that can be handled on the outgoing interface 3660 * without forcing IP to fragment. If no route is found, route has no mtu, 3661 * or the destination isn't local, use a default, hopefully conservative 3662 * size (usually 512 or the default IP max size, but no more than the mtu 3663 * of the interface), as we can't discover anything about intervening 3664 * gateways or networks. We also initialize the congestion/slow start 3665 * window to be a single segment if the destination isn't local. 3666 * While looking at the routing entry, we also initialize other path-dependent 3667 * parameters from pre-set or cached values in the routing entry. 3668 * 3669 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3670 * IP options, e.g. IPSEC data, since length of this data may vary, and 3671 * thus it is calculated for every segment separately in tcp_output(). 3672 * 3673 * NOTE that this routine is only called when we process an incoming 3674 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3675 * settings are handled in tcp_mssopt(). 3676 */ 3677 void 3678 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3679 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3680 { 3681 int mss = 0; 3682 uint32_t maxmtu = 0; 3683 struct inpcb *inp = tp->t_inpcb; 3684 struct hc_metrics_lite metrics; 3685 #ifdef INET6 3686 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3687 size_t min_protoh = isipv6 ? 3688 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3689 sizeof (struct tcpiphdr); 3690 #else 3691 size_t min_protoh = sizeof(struct tcpiphdr); 3692 #endif 3693 3694 INP_WLOCK_ASSERT(tp->t_inpcb); 3695 3696 if (tp->t_port) 3697 min_protoh += V_tcp_udp_tunneling_overhead; 3698 if (mtuoffer != -1) { 3699 KASSERT(offer == -1, ("%s: conflict", __func__)); 3700 offer = mtuoffer - min_protoh; 3701 } 3702 3703 /* Initialize. */ 3704 #ifdef INET6 3705 if (isipv6) { 3706 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3707 tp->t_maxseg = V_tcp_v6mssdflt; 3708 } 3709 #endif 3710 #if defined(INET) && defined(INET6) 3711 else 3712 #endif 3713 #ifdef INET 3714 { 3715 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3716 tp->t_maxseg = V_tcp_mssdflt; 3717 } 3718 #endif 3719 3720 /* 3721 * No route to sender, stay with default mss and return. 3722 */ 3723 if (maxmtu == 0) { 3724 /* 3725 * In case we return early we need to initialize metrics 3726 * to a defined state as tcp_hc_get() would do for us 3727 * if there was no cache hit. 3728 */ 3729 if (metricptr != NULL) 3730 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3731 return; 3732 } 3733 3734 /* What have we got? */ 3735 switch (offer) { 3736 case 0: 3737 /* 3738 * Offer == 0 means that there was no MSS on the SYN 3739 * segment, in this case we use tcp_mssdflt as 3740 * already assigned to t_maxseg above. 3741 */ 3742 offer = tp->t_maxseg; 3743 break; 3744 3745 case -1: 3746 /* 3747 * Offer == -1 means that we didn't receive SYN yet. 3748 */ 3749 /* FALLTHROUGH */ 3750 3751 default: 3752 /* 3753 * Prevent DoS attack with too small MSS. Round up 3754 * to at least minmss. 3755 */ 3756 offer = max(offer, V_tcp_minmss); 3757 } 3758 3759 /* 3760 * rmx information is now retrieved from tcp_hostcache. 3761 */ 3762 tcp_hc_get(&inp->inp_inc, &metrics); 3763 if (metricptr != NULL) 3764 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3765 3766 /* 3767 * If there's a discovered mtu in tcp hostcache, use it. 3768 * Else, use the link mtu. 3769 */ 3770 if (metrics.rmx_mtu) 3771 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3772 else { 3773 #ifdef INET6 3774 if (isipv6) { 3775 mss = maxmtu - min_protoh; 3776 if (!V_path_mtu_discovery && 3777 !in6_localaddr(&inp->in6p_faddr)) 3778 mss = min(mss, V_tcp_v6mssdflt); 3779 } 3780 #endif 3781 #if defined(INET) && defined(INET6) 3782 else 3783 #endif 3784 #ifdef INET 3785 { 3786 mss = maxmtu - min_protoh; 3787 if (!V_path_mtu_discovery && 3788 !in_localaddr(inp->inp_faddr)) 3789 mss = min(mss, V_tcp_mssdflt); 3790 } 3791 #endif 3792 /* 3793 * XXX - The above conditional (mss = maxmtu - min_protoh) 3794 * probably violates the TCP spec. 3795 * The problem is that, since we don't know the 3796 * other end's MSS, we are supposed to use a conservative 3797 * default. But, if we do that, then MTU discovery will 3798 * never actually take place, because the conservative 3799 * default is much less than the MTUs typically seen 3800 * on the Internet today. For the moment, we'll sweep 3801 * this under the carpet. 3802 * 3803 * The conservative default might not actually be a problem 3804 * if the only case this occurs is when sending an initial 3805 * SYN with options and data to a host we've never talked 3806 * to before. Then, they will reply with an MSS value which 3807 * will get recorded and the new parameters should get 3808 * recomputed. For Further Study. 3809 */ 3810 } 3811 mss = min(mss, offer); 3812 3813 /* 3814 * Sanity check: make sure that maxseg will be large 3815 * enough to allow some data on segments even if the 3816 * all the option space is used (40bytes). Otherwise 3817 * funny things may happen in tcp_output. 3818 * 3819 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3820 */ 3821 mss = max(mss, 64); 3822 3823 tp->t_maxseg = mss; 3824 } 3825 3826 void 3827 tcp_mss(struct tcpcb *tp, int offer) 3828 { 3829 int mss; 3830 uint32_t bufsize; 3831 struct inpcb *inp; 3832 struct socket *so; 3833 struct hc_metrics_lite metrics; 3834 struct tcp_ifcap cap; 3835 3836 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3837 3838 bzero(&cap, sizeof(cap)); 3839 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3840 3841 mss = tp->t_maxseg; 3842 inp = tp->t_inpcb; 3843 3844 /* 3845 * If there's a pipesize, change the socket buffer to that size, 3846 * don't change if sb_hiwat is different than default (then it 3847 * has been changed on purpose with setsockopt). 3848 * Make the socket buffers an integral number of mss units; 3849 * if the mss is larger than the socket buffer, decrease the mss. 3850 */ 3851 so = inp->inp_socket; 3852 SOCKBUF_LOCK(&so->so_snd); 3853 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3854 bufsize = metrics.rmx_sendpipe; 3855 else 3856 bufsize = so->so_snd.sb_hiwat; 3857 if (bufsize < mss) 3858 mss = bufsize; 3859 else { 3860 bufsize = roundup(bufsize, mss); 3861 if (bufsize > sb_max) 3862 bufsize = sb_max; 3863 if (bufsize > so->so_snd.sb_hiwat) 3864 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3865 } 3866 SOCKBUF_UNLOCK(&so->so_snd); 3867 /* 3868 * Sanity check: make sure that maxseg will be large 3869 * enough to allow some data on segments even if the 3870 * all the option space is used (40bytes). Otherwise 3871 * funny things may happen in tcp_output. 3872 * 3873 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3874 */ 3875 tp->t_maxseg = max(mss, 64); 3876 3877 SOCKBUF_LOCK(&so->so_rcv); 3878 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3879 bufsize = metrics.rmx_recvpipe; 3880 else 3881 bufsize = so->so_rcv.sb_hiwat; 3882 if (bufsize > mss) { 3883 bufsize = roundup(bufsize, mss); 3884 if (bufsize > sb_max) 3885 bufsize = sb_max; 3886 if (bufsize > so->so_rcv.sb_hiwat) 3887 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3888 } 3889 SOCKBUF_UNLOCK(&so->so_rcv); 3890 3891 /* Check the interface for TSO capabilities. */ 3892 if (cap.ifcap & CSUM_TSO) { 3893 tp->t_flags |= TF_TSO; 3894 tp->t_tsomax = cap.tsomax; 3895 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3896 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3897 } 3898 } 3899 3900 /* 3901 * Determine the MSS option to send on an outgoing SYN. 3902 */ 3903 int 3904 tcp_mssopt(struct in_conninfo *inc) 3905 { 3906 int mss = 0; 3907 uint32_t thcmtu = 0; 3908 uint32_t maxmtu = 0; 3909 size_t min_protoh; 3910 3911 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3912 3913 #ifdef INET6 3914 if (inc->inc_flags & INC_ISIPV6) { 3915 mss = V_tcp_v6mssdflt; 3916 maxmtu = tcp_maxmtu6(inc, NULL); 3917 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3918 } 3919 #endif 3920 #if defined(INET) && defined(INET6) 3921 else 3922 #endif 3923 #ifdef INET 3924 { 3925 mss = V_tcp_mssdflt; 3926 maxmtu = tcp_maxmtu(inc, NULL); 3927 min_protoh = sizeof(struct tcpiphdr); 3928 } 3929 #endif 3930 #if defined(INET6) || defined(INET) 3931 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3932 #endif 3933 3934 if (maxmtu && thcmtu) 3935 mss = min(maxmtu, thcmtu) - min_protoh; 3936 else if (maxmtu || thcmtu) 3937 mss = max(maxmtu, thcmtu) - min_protoh; 3938 3939 return (mss); 3940 } 3941 3942 void 3943 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3944 { 3945 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3946 int maxseg = tcp_maxseg(tp); 3947 3948 INP_WLOCK_ASSERT(tp->t_inpcb); 3949 3950 /* 3951 * Compute the amount of data that this ACK is indicating 3952 * (del_data) and an estimate of how many bytes are in the 3953 * network. 3954 */ 3955 if (((tp->t_flags & TF_SACK_PERMIT) && 3956 (to->to_flags & TOF_SACK)) || 3957 (IN_CONGRECOVERY(tp->t_flags) && 3958 !IN_FASTRECOVERY(tp->t_flags))) { 3959 del_data = tp->sackhint.delivered_data; 3960 if (V_tcp_do_newsack) 3961 pipe = tcp_compute_pipe(tp); 3962 else 3963 pipe = (tp->snd_nxt - tp->snd_fack) + 3964 tp->sackhint.sack_bytes_rexmit; 3965 } else { 3966 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 3967 tp->snd_recover - tp->snd_una)) 3968 del_data = maxseg; 3969 pipe = imax(0, tp->snd_max - tp->snd_una - 3970 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 3971 } 3972 tp->sackhint.prr_delivered += del_data; 3973 /* 3974 * Proportional Rate Reduction 3975 */ 3976 if (pipe >= tp->snd_ssthresh) { 3977 if (tp->sackhint.recover_fs == 0) 3978 tp->sackhint.recover_fs = 3979 imax(1, tp->snd_nxt - tp->snd_una); 3980 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3981 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3982 tp->sackhint.prr_out; 3983 } else { 3984 if (V_tcp_do_prr_conservative || (del_data == 0)) 3985 limit = tp->sackhint.prr_delivered - 3986 tp->sackhint.prr_out; 3987 else 3988 limit = imax(tp->sackhint.prr_delivered - 3989 tp->sackhint.prr_out, del_data) + 3990 maxseg; 3991 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 3992 } 3993 snd_cnt = imax(snd_cnt, 0) / maxseg; 3994 /* 3995 * Send snd_cnt new data into the network in response to this ack. 3996 * If there is going to be a SACK retransmission, adjust snd_cwnd 3997 * accordingly. 3998 */ 3999 if (IN_FASTRECOVERY(tp->t_flags)) { 4000 if ((tp->t_flags & TF_SACK_PERMIT) && 4001 (to->to_flags & TOF_SACK)) { 4002 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 4003 tp->sackhint.sack_bytes_rexmit + 4004 (snd_cnt * maxseg); 4005 } else { 4006 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4007 (snd_cnt * maxseg); 4008 } 4009 } else if (IN_CONGRECOVERY(tp->t_flags)) 4010 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4011 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4012 } 4013 4014 /* 4015 * On a partial ack arrives, force the retransmission of the 4016 * next unacknowledged segment. Do not clear tp->t_dupacks. 4017 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4018 * be started again. 4019 */ 4020 void 4021 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4022 { 4023 tcp_seq onxt = tp->snd_nxt; 4024 uint32_t ocwnd = tp->snd_cwnd; 4025 u_int maxseg = tcp_maxseg(tp); 4026 4027 INP_WLOCK_ASSERT(tp->t_inpcb); 4028 4029 tcp_timer_activate(tp, TT_REXMT, 0); 4030 tp->t_rtttime = 0; 4031 tp->snd_nxt = th->th_ack; 4032 /* 4033 * Set snd_cwnd to one segment beyond acknowledged offset. 4034 * (tp->snd_una has not yet been updated when this function is called.) 4035 */ 4036 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4037 tp->t_flags |= TF_ACKNOW; 4038 (void) tcp_output(tp); 4039 tp->snd_cwnd = ocwnd; 4040 if (SEQ_GT(onxt, tp->snd_nxt)) 4041 tp->snd_nxt = onxt; 4042 /* 4043 * Partial window deflation. Relies on fact that tp->snd_una 4044 * not updated yet. 4045 */ 4046 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4047 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4048 else 4049 tp->snd_cwnd = 0; 4050 tp->snd_cwnd += maxseg; 4051 } 4052 4053 int 4054 tcp_compute_pipe(struct tcpcb *tp) 4055 { 4056 return (tp->snd_max - tp->snd_una + 4057 tp->sackhint.sack_bytes_rexmit - 4058 tp->sackhint.sacked_bytes); 4059 } 4060 4061 uint32_t 4062 tcp_compute_initwnd(uint32_t maxseg) 4063 { 4064 /* 4065 * Calculate the Initial Window, also used as Restart Window 4066 * 4067 * RFC5681 Section 3.1 specifies the default conservative values. 4068 * RFC3390 specifies slightly more aggressive values. 4069 * RFC6928 increases it to ten segments. 4070 * Support for user specified value for initial flight size. 4071 */ 4072 if (V_tcp_initcwnd_segments) 4073 return min(V_tcp_initcwnd_segments * maxseg, 4074 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4075 else if (V_tcp_do_rfc3390) 4076 return min(4 * maxseg, max(2 * maxseg, 4380)); 4077 else { 4078 /* Per RFC5681 Section 3.1 */ 4079 if (maxseg > 2190) 4080 return (2 * maxseg); 4081 else if (maxseg > 1095) 4082 return (3 * maxseg); 4083 else 4084 return (4 * maxseg); 4085 } 4086 } 4087