1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 __FBSDID("$FreeBSD$"); 54 55 #include "opt_inet.h" 56 #include "opt_inet6.h" 57 #include "opt_ipsec.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/arb.h> 62 #include <sys/kernel.h> 63 #ifdef TCP_HHOOK 64 #include <sys/hhook.h> 65 #endif 66 #include <sys/malloc.h> 67 #include <sys/mbuf.h> 68 #include <sys/proc.h> /* for proc0 declaration */ 69 #include <sys/protosw.h> 70 #include <sys/qmath.h> 71 #include <sys/sdt.h> 72 #include <sys/signalvar.h> 73 #include <sys/socket.h> 74 #include <sys/socketvar.h> 75 #include <sys/sysctl.h> 76 #include <sys/syslog.h> 77 #include <sys/systm.h> 78 #include <sys/stats.h> 79 80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 81 82 #include <vm/uma.h> 83 84 #include <net/if.h> 85 #include <net/if_var.h> 86 #include <net/route.h> 87 #include <net/vnet.h> 88 89 #define TCPSTATES /* for logging */ 90 91 #include <netinet/in.h> 92 #include <netinet/in_kdtrace.h> 93 #include <netinet/in_pcb.h> 94 #include <netinet/in_systm.h> 95 #include <netinet/ip.h> 96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 98 #include <netinet/ip_var.h> 99 #include <netinet/ip_options.h> 100 #include <netinet/ip6.h> 101 #include <netinet/icmp6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_var.h> 104 #include <netinet6/ip6_var.h> 105 #include <netinet6/nd6.h> 106 #include <netinet/tcp.h> 107 #include <netinet/tcp_fsm.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_seq.h> 110 #include <netinet/tcp_timer.h> 111 #include <netinet/tcp_var.h> 112 #include <netinet6/tcp6_var.h> 113 #include <netinet/tcpip.h> 114 #include <netinet/cc/cc.h> 115 #include <netinet/tcp_fastopen.h> 116 #ifdef TCPPCAP 117 #include <netinet/tcp_pcap.h> 118 #endif 119 #include <netinet/tcp_syncache.h> 120 #ifdef TCPDEBUG 121 #include <netinet/tcp_debug.h> 122 #endif /* TCPDEBUG */ 123 #ifdef TCP_OFFLOAD 124 #include <netinet/tcp_offload.h> 125 #endif 126 #include <netinet/tcp_ecn.h> 127 #include <netinet/udp.h> 128 129 #include <netipsec/ipsec_support.h> 130 131 #include <machine/in_cksum.h> 132 133 #include <security/mac/mac_framework.h> 134 135 const int tcprexmtthresh = 3; 136 137 VNET_DEFINE(int, tcp_log_in_vain) = 0; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(tcp_log_in_vain), 0, 140 "Log all incoming TCP segments to closed ports"); 141 142 VNET_DEFINE(int, blackhole) = 0; 143 #define V_blackhole VNET(blackhole) 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 145 &VNET_NAME(blackhole), 0, 146 "Do not send RST on segments to closed ports"); 147 148 VNET_DEFINE(bool, blackhole_local) = false; 149 #define V_blackhole_local VNET(blackhole_local) 150 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 151 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 152 "Enforce net.inet.tcp.blackhole for locally originated packets"); 153 154 VNET_DEFINE(int, tcp_delack_enabled) = 1; 155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 156 &VNET_NAME(tcp_delack_enabled), 0, 157 "Delay ACK to try and piggyback it onto a data packet"); 158 159 VNET_DEFINE(int, drop_synfin) = 0; 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 161 &VNET_NAME(drop_synfin), 0, 162 "Drop TCP packets with SYN+FIN set"); 163 164 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 166 &VNET_NAME(tcp_do_prr_conservative), 0, 167 "Do conservative Proportional Rate Reduction"); 168 169 VNET_DEFINE(int, tcp_do_prr) = 1; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 171 &VNET_NAME(tcp_do_prr), 1, 172 "Enable Proportional Rate Reduction per RFC 6937"); 173 174 VNET_DEFINE(int, tcp_do_lrd) = 0; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW, 176 &VNET_NAME(tcp_do_lrd), 1, 177 "Perform Lost Retransmission Detection"); 178 179 VNET_DEFINE(int, tcp_do_newcwv) = 0; 180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_do_newcwv), 0, 182 "Enable New Congestion Window Validation per RFC7661"); 183 184 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 186 &VNET_NAME(tcp_do_rfc3042), 0, 187 "Enable RFC 3042 (Limited Transmit)"); 188 189 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 191 &VNET_NAME(tcp_do_rfc3390), 0, 192 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 193 194 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 195 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 196 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 197 "Slow-start flight size (initial congestion window) in number of segments"); 198 199 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 201 &VNET_NAME(tcp_do_rfc3465), 0, 202 "Enable RFC 3465 (Appropriate Byte Counting)"); 203 204 VNET_DEFINE(int, tcp_abc_l_var) = 2; 205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 206 &VNET_NAME(tcp_abc_l_var), 2, 207 "Cap the max cwnd increment during slow-start to this number of segments"); 208 209 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 210 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 211 "TCP ECN"); 212 213 VNET_DEFINE(int, tcp_do_ecn) = 2; 214 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 215 &VNET_NAME(tcp_do_ecn), 0, 216 "TCP ECN support"); 217 218 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 219 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 220 &VNET_NAME(tcp_ecn_maxretries), 0, 221 "Max retries before giving up on ECN"); 222 223 VNET_DEFINE(int, tcp_insecure_syn) = 0; 224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 225 &VNET_NAME(tcp_insecure_syn), 0, 226 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 227 228 VNET_DEFINE(int, tcp_insecure_rst) = 0; 229 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 230 &VNET_NAME(tcp_insecure_rst), 0, 231 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 232 233 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 234 #define V_tcp_recvspace VNET(tcp_recvspace) 235 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 236 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 237 238 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 239 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 240 &VNET_NAME(tcp_do_autorcvbuf), 0, 241 "Enable automatic receive buffer sizing"); 242 243 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 244 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 245 &VNET_NAME(tcp_autorcvbuf_max), 0, 246 "Max size of automatic receive buffer"); 247 248 VNET_DEFINE(struct inpcbinfo, tcbinfo); 249 250 /* 251 * TCP statistics are stored in an array of counter(9)s, which size matches 252 * size of struct tcpstat. TCP running connection count is a regular array. 253 */ 254 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 255 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 256 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 257 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 258 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 259 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 260 "TCP connection counts by TCP state"); 261 262 /* 263 * Kernel module interface for updating tcpstat. The first argument is an index 264 * into tcpstat treated as an array. 265 */ 266 void 267 kmod_tcpstat_add(int statnum, int val) 268 { 269 270 counter_u64_add(VNET(tcpstat)[statnum], val); 271 } 272 273 #ifdef TCP_HHOOK 274 /* 275 * Wrapper for the TCP established input helper hook. 276 */ 277 void 278 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 279 { 280 struct tcp_hhook_data hhook_data; 281 282 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 283 hhook_data.tp = tp; 284 hhook_data.th = th; 285 hhook_data.to = to; 286 287 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 288 tp->osd); 289 } 290 } 291 #endif 292 293 /* 294 * CC wrapper hook functions 295 */ 296 void 297 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 298 uint16_t type) 299 { 300 #ifdef STATS 301 int32_t gput; 302 #endif 303 304 INP_WLOCK_ASSERT(tp->t_inpcb); 305 306 tp->ccv->nsegs = nsegs; 307 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 308 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 309 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 310 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 311 tp->ccv->flags |= CCF_CWND_LIMITED; 312 else 313 tp->ccv->flags &= ~CCF_CWND_LIMITED; 314 315 if (type == CC_ACK) { 316 #ifdef STATS 317 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 318 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 319 if (!IN_RECOVERY(tp->t_flags)) 320 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 321 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 322 if ((tp->t_flags & TF_GPUTINPROG) && 323 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 324 /* 325 * Compute goodput in bits per millisecond. 326 */ 327 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 328 max(1, tcp_ts_getticks() - tp->gput_ts); 329 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 330 gput); 331 /* 332 * XXXLAS: This is a temporary hack, and should be 333 * chained off VOI_TCP_GPUT when stats(9) grows an API 334 * to deal with chained VOIs. 335 */ 336 if (tp->t_stats_gput_prev > 0) 337 stats_voi_update_abs_s32(tp->t_stats, 338 VOI_TCP_GPUT_ND, 339 ((gput - tp->t_stats_gput_prev) * 100) / 340 tp->t_stats_gput_prev); 341 tp->t_flags &= ~TF_GPUTINPROG; 342 tp->t_stats_gput_prev = gput; 343 } 344 #endif /* STATS */ 345 if (tp->snd_cwnd > tp->snd_ssthresh) { 346 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 347 if (tp->t_bytes_acked >= tp->snd_cwnd) { 348 tp->t_bytes_acked -= tp->snd_cwnd; 349 tp->ccv->flags |= CCF_ABC_SENTAWND; 350 } 351 } else { 352 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 353 tp->t_bytes_acked = 0; 354 } 355 } 356 357 if (CC_ALGO(tp)->ack_received != NULL) { 358 /* XXXLAS: Find a way to live without this */ 359 tp->ccv->curack = th->th_ack; 360 CC_ALGO(tp)->ack_received(tp->ccv, type); 361 } 362 #ifdef STATS 363 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 364 #endif 365 } 366 367 void 368 cc_conn_init(struct tcpcb *tp) 369 { 370 struct hc_metrics_lite metrics; 371 struct inpcb *inp = tp->t_inpcb; 372 u_int maxseg; 373 int rtt; 374 375 INP_WLOCK_ASSERT(tp->t_inpcb); 376 377 tcp_hc_get(&inp->inp_inc, &metrics); 378 maxseg = tcp_maxseg(tp); 379 380 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 381 tp->t_srtt = rtt; 382 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 383 TCPSTAT_INC(tcps_usedrtt); 384 if (metrics.rmx_rttvar) { 385 tp->t_rttvar = metrics.rmx_rttvar; 386 TCPSTAT_INC(tcps_usedrttvar); 387 } else { 388 /* default variation is +- 1 rtt */ 389 tp->t_rttvar = 390 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 391 } 392 TCPT_RANGESET(tp->t_rxtcur, 393 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 394 tp->t_rttmin, TCPTV_REXMTMAX); 395 } 396 if (metrics.rmx_ssthresh) { 397 /* 398 * There's some sort of gateway or interface 399 * buffer limit on the path. Use this to set 400 * the slow start threshold, but set the 401 * threshold to no less than 2*mss. 402 */ 403 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 404 TCPSTAT_INC(tcps_usedssthresh); 405 } 406 407 /* 408 * Set the initial slow-start flight size. 409 * 410 * If a SYN or SYN/ACK was lost and retransmitted, we have to 411 * reduce the initial CWND to one segment as congestion is likely 412 * requiring us to be cautious. 413 */ 414 if (tp->snd_cwnd == 1) 415 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 416 else 417 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 418 419 if (CC_ALGO(tp)->conn_init != NULL) 420 CC_ALGO(tp)->conn_init(tp->ccv); 421 } 422 423 void inline 424 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 425 { 426 INP_WLOCK_ASSERT(tp->t_inpcb); 427 428 #ifdef STATS 429 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 430 #endif 431 432 switch(type) { 433 case CC_NDUPACK: 434 if (!IN_FASTRECOVERY(tp->t_flags)) { 435 tp->snd_recover = tp->snd_max; 436 if (tp->t_flags2 & TF2_ECN_PERMIT) 437 tp->t_flags2 |= TF2_ECN_SND_CWR; 438 } 439 break; 440 case CC_ECN: 441 if (!IN_CONGRECOVERY(tp->t_flags) || 442 /* 443 * Allow ECN reaction on ACK to CWR, if 444 * that data segment was also CE marked. 445 */ 446 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 447 EXIT_CONGRECOVERY(tp->t_flags); 448 TCPSTAT_INC(tcps_ecn_rcwnd); 449 tp->snd_recover = tp->snd_max + 1; 450 if (tp->t_flags2 & TF2_ECN_PERMIT) 451 tp->t_flags2 |= TF2_ECN_SND_CWR; 452 } 453 break; 454 case CC_RTO: 455 tp->t_dupacks = 0; 456 tp->t_bytes_acked = 0; 457 EXIT_RECOVERY(tp->t_flags); 458 if (tp->t_flags2 & TF2_ECN_PERMIT) 459 tp->t_flags2 |= TF2_ECN_SND_CWR; 460 break; 461 case CC_RTO_ERR: 462 TCPSTAT_INC(tcps_sndrexmitbad); 463 /* RTO was unnecessary, so reset everything. */ 464 tp->snd_cwnd = tp->snd_cwnd_prev; 465 tp->snd_ssthresh = tp->snd_ssthresh_prev; 466 tp->snd_recover = tp->snd_recover_prev; 467 if (tp->t_flags & TF_WASFRECOVERY) 468 ENTER_FASTRECOVERY(tp->t_flags); 469 if (tp->t_flags & TF_WASCRECOVERY) 470 ENTER_CONGRECOVERY(tp->t_flags); 471 tp->snd_nxt = tp->snd_max; 472 tp->t_flags &= ~TF_PREVVALID; 473 tp->t_badrxtwin = 0; 474 break; 475 } 476 477 if (CC_ALGO(tp)->cong_signal != NULL) { 478 if (th != NULL) 479 tp->ccv->curack = th->th_ack; 480 CC_ALGO(tp)->cong_signal(tp->ccv, type); 481 } 482 } 483 484 void inline 485 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 486 { 487 INP_WLOCK_ASSERT(tp->t_inpcb); 488 489 /* XXXLAS: KASSERT that we're in recovery? */ 490 491 if (CC_ALGO(tp)->post_recovery != NULL) { 492 tp->ccv->curack = th->th_ack; 493 CC_ALGO(tp)->post_recovery(tp->ccv); 494 } 495 /* XXXLAS: EXIT_RECOVERY ? */ 496 tp->t_bytes_acked = 0; 497 tp->sackhint.delivered_data = 0; 498 tp->sackhint.prr_out = 0; 499 } 500 501 /* 502 * Indicate whether this ack should be delayed. We can delay the ack if 503 * following conditions are met: 504 * - There is no delayed ack timer in progress. 505 * - Our last ack wasn't a 0-sized window. We never want to delay 506 * the ack that opens up a 0-sized window. 507 * - LRO wasn't used for this segment. We make sure by checking that the 508 * segment size is not larger than the MSS. 509 */ 510 #define DELAY_ACK(tp, tlen) \ 511 ((!tcp_timer_active(tp, TT_DELACK) && \ 512 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 513 (tlen <= tp->t_maxseg) && \ 514 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 515 516 void inline 517 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 518 { 519 INP_WLOCK_ASSERT(tp->t_inpcb); 520 521 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 522 switch (iptos & IPTOS_ECN_MASK) { 523 case IPTOS_ECN_CE: 524 tp->ccv->flags |= CCF_IPHDR_CE; 525 break; 526 case IPTOS_ECN_ECT0: 527 /* FALLTHROUGH */ 528 case IPTOS_ECN_ECT1: 529 /* FALLTHROUGH */ 530 case IPTOS_ECN_NOTECT: 531 tp->ccv->flags &= ~CCF_IPHDR_CE; 532 break; 533 } 534 535 if (flags & TH_CWR) 536 tp->ccv->flags |= CCF_TCPHDR_CWR; 537 else 538 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 539 540 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 541 542 if (tp->ccv->flags & CCF_ACKNOW) { 543 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 544 tp->t_flags |= TF_ACKNOW; 545 } 546 } 547 } 548 549 void inline 550 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 551 { 552 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 553 } 554 555 /* 556 * TCP input handling is split into multiple parts: 557 * tcp6_input is a thin wrapper around tcp_input for the extended 558 * ip6_protox[] call format in ip6_input 559 * tcp_input handles primary segment validation, inpcb lookup and 560 * SYN processing on listen sockets 561 * tcp_do_segment processes the ACK and text of the segment for 562 * establishing, established and closing connections 563 */ 564 #ifdef INET6 565 int 566 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 567 { 568 struct mbuf *m; 569 struct in6_ifaddr *ia6; 570 struct ip6_hdr *ip6; 571 572 m = *mp; 573 if (m->m_len < *offp + sizeof(struct tcphdr)) { 574 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 575 if (m == NULL) { 576 *mp = m; 577 TCPSTAT_INC(tcps_rcvshort); 578 return (IPPROTO_DONE); 579 } 580 } 581 582 /* 583 * draft-itojun-ipv6-tcp-to-anycast 584 * better place to put this in? 585 */ 586 ip6 = mtod(m, struct ip6_hdr *); 587 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 588 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 589 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 590 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 591 *mp = NULL; 592 return (IPPROTO_DONE); 593 } 594 595 *mp = m; 596 return (tcp_input_with_port(mp, offp, proto, port)); 597 } 598 599 int 600 tcp6_input(struct mbuf **mp, int *offp, int proto) 601 { 602 603 return(tcp6_input_with_port(mp, offp, proto, 0)); 604 } 605 #endif /* INET6 */ 606 607 int 608 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 609 { 610 struct mbuf *m = *mp; 611 struct tcphdr *th = NULL; 612 struct ip *ip = NULL; 613 struct inpcb *inp = NULL; 614 struct tcpcb *tp = NULL; 615 struct socket *so = NULL; 616 u_char *optp = NULL; 617 int off0; 618 int optlen = 0; 619 #ifdef INET 620 int len; 621 uint8_t ipttl; 622 #endif 623 int tlen = 0, off; 624 int drop_hdrlen; 625 int thflags; 626 int rstreason = 0; /* For badport_bandlim accounting purposes */ 627 int lookupflag; 628 uint8_t iptos; 629 struct m_tag *fwd_tag = NULL; 630 #ifdef INET6 631 struct ip6_hdr *ip6 = NULL; 632 int isipv6; 633 #else 634 const void *ip6 = NULL; 635 #endif /* INET6 */ 636 struct tcpopt to; /* options in this segment */ 637 char *s = NULL; /* address and port logging */ 638 #ifdef TCPDEBUG 639 /* 640 * The size of tcp_saveipgen must be the size of the max ip header, 641 * now IPv6. 642 */ 643 u_char tcp_saveipgen[IP6_HDR_LEN]; 644 struct tcphdr tcp_savetcp; 645 short ostate = 0; 646 #endif 647 648 NET_EPOCH_ASSERT(); 649 650 #ifdef INET6 651 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 652 #endif 653 654 off0 = *offp; 655 m = *mp; 656 *mp = NULL; 657 to.to_flags = 0; 658 TCPSTAT_INC(tcps_rcvtotal); 659 660 #ifdef INET6 661 if (isipv6) { 662 ip6 = mtod(m, struct ip6_hdr *); 663 th = (struct tcphdr *)((caddr_t)ip6 + off0); 664 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 665 if (port) 666 goto skip6_csum; 667 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 668 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 669 th->th_sum = m->m_pkthdr.csum_data; 670 else 671 th->th_sum = in6_cksum_pseudo(ip6, tlen, 672 IPPROTO_TCP, m->m_pkthdr.csum_data); 673 th->th_sum ^= 0xffff; 674 } else 675 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 676 if (th->th_sum) { 677 TCPSTAT_INC(tcps_rcvbadsum); 678 goto drop; 679 } 680 skip6_csum: 681 /* 682 * Be proactive about unspecified IPv6 address in source. 683 * As we use all-zero to indicate unbounded/unconnected pcb, 684 * unspecified IPv6 address can be used to confuse us. 685 * 686 * Note that packets with unspecified IPv6 destination is 687 * already dropped in ip6_input. 688 */ 689 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 690 /* XXX stat */ 691 goto drop; 692 } 693 iptos = IPV6_TRAFFIC_CLASS(ip6); 694 } 695 #endif 696 #if defined(INET) && defined(INET6) 697 else 698 #endif 699 #ifdef INET 700 { 701 /* 702 * Get IP and TCP header together in first mbuf. 703 * Note: IP leaves IP header in first mbuf. 704 */ 705 if (off0 > sizeof (struct ip)) { 706 ip_stripoptions(m); 707 off0 = sizeof(struct ip); 708 } 709 if (m->m_len < sizeof (struct tcpiphdr)) { 710 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 711 == NULL) { 712 TCPSTAT_INC(tcps_rcvshort); 713 return (IPPROTO_DONE); 714 } 715 } 716 ip = mtod(m, struct ip *); 717 th = (struct tcphdr *)((caddr_t)ip + off0); 718 tlen = ntohs(ip->ip_len) - off0; 719 720 iptos = ip->ip_tos; 721 if (port) 722 goto skip_csum; 723 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 724 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 725 th->th_sum = m->m_pkthdr.csum_data; 726 else 727 th->th_sum = in_pseudo(ip->ip_src.s_addr, 728 ip->ip_dst.s_addr, 729 htonl(m->m_pkthdr.csum_data + tlen + 730 IPPROTO_TCP)); 731 th->th_sum ^= 0xffff; 732 } else { 733 struct ipovly *ipov = (struct ipovly *)ip; 734 735 /* 736 * Checksum extended TCP header and data. 737 */ 738 len = off0 + tlen; 739 ipttl = ip->ip_ttl; 740 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 741 ipov->ih_len = htons(tlen); 742 th->th_sum = in_cksum(m, len); 743 /* Reset length for SDT probes. */ 744 ip->ip_len = htons(len); 745 /* Reset TOS bits */ 746 ip->ip_tos = iptos; 747 /* Re-initialization for later version check */ 748 ip->ip_ttl = ipttl; 749 ip->ip_v = IPVERSION; 750 ip->ip_hl = off0 >> 2; 751 } 752 skip_csum: 753 if (th->th_sum && (port == 0)) { 754 TCPSTAT_INC(tcps_rcvbadsum); 755 goto drop; 756 } 757 } 758 #endif /* INET */ 759 760 /* 761 * Check that TCP offset makes sense, 762 * pull out TCP options and adjust length. XXX 763 */ 764 off = th->th_off << 2; 765 if (off < sizeof (struct tcphdr) || off > tlen) { 766 TCPSTAT_INC(tcps_rcvbadoff); 767 goto drop; 768 } 769 tlen -= off; /* tlen is used instead of ti->ti_len */ 770 if (off > sizeof (struct tcphdr)) { 771 #ifdef INET6 772 if (isipv6) { 773 if (m->m_len < off0 + off) { 774 m = m_pullup(m, off0 + off); 775 if (m == NULL) { 776 TCPSTAT_INC(tcps_rcvshort); 777 return (IPPROTO_DONE); 778 } 779 } 780 ip6 = mtod(m, struct ip6_hdr *); 781 th = (struct tcphdr *)((caddr_t)ip6 + off0); 782 } 783 #endif 784 #if defined(INET) && defined(INET6) 785 else 786 #endif 787 #ifdef INET 788 { 789 if (m->m_len < sizeof(struct ip) + off) { 790 if ((m = m_pullup(m, sizeof (struct ip) + off)) 791 == NULL) { 792 TCPSTAT_INC(tcps_rcvshort); 793 return (IPPROTO_DONE); 794 } 795 ip = mtod(m, struct ip *); 796 th = (struct tcphdr *)((caddr_t)ip + off0); 797 } 798 } 799 #endif 800 optlen = off - sizeof (struct tcphdr); 801 optp = (u_char *)(th + 1); 802 } 803 thflags = tcp_get_flags(th); 804 805 /* 806 * Convert TCP protocol specific fields to host format. 807 */ 808 tcp_fields_to_host(th); 809 810 /* 811 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 812 */ 813 drop_hdrlen = off0 + off; 814 815 /* 816 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 817 */ 818 if ( 819 #ifdef INET6 820 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 821 #ifdef INET 822 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 823 #endif 824 #endif 825 #if defined(INET) && !defined(INET6) 826 (m->m_flags & M_IP_NEXTHOP) 827 #endif 828 ) 829 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 830 831 /* 832 * For initial SYN packets we don't need write lock on matching 833 * PCB, be it a listening one or a synchronized one. The packet 834 * shall not modify its state. 835 */ 836 lookupflag = INPLOOKUP_WILDCARD | 837 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 838 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); 839 findpcb: 840 #ifdef INET6 841 if (isipv6 && fwd_tag != NULL) { 842 struct sockaddr_in6 *next_hop6; 843 844 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 845 /* 846 * Transparently forwarded. Pretend to be the destination. 847 * Already got one like this? 848 */ 849 inp = in6_pcblookup_mbuf(&V_tcbinfo, 850 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 851 lookupflag, m->m_pkthdr.rcvif, m); 852 if (!inp) { 853 /* 854 * It's new. Try to find the ambushing socket. 855 * Because we've rewritten the destination address, 856 * any hardware-generated hash is ignored. 857 */ 858 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 859 th->th_sport, &next_hop6->sin6_addr, 860 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 861 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 862 } 863 } else if (isipv6) { 864 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 865 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 866 m->m_pkthdr.rcvif, m); 867 } 868 #endif /* INET6 */ 869 #if defined(INET6) && defined(INET) 870 else 871 #endif 872 #ifdef INET 873 if (fwd_tag != NULL) { 874 struct sockaddr_in *next_hop; 875 876 next_hop = (struct sockaddr_in *)(fwd_tag+1); 877 /* 878 * Transparently forwarded. Pretend to be the destination. 879 * already got one like this? 880 */ 881 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 882 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); 883 if (!inp) { 884 /* 885 * It's new. Try to find the ambushing socket. 886 * Because we've rewritten the destination address, 887 * any hardware-generated hash is ignored. 888 */ 889 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 890 th->th_sport, next_hop->sin_addr, 891 next_hop->sin_port ? ntohs(next_hop->sin_port) : 892 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 893 } 894 } else 895 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 896 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 897 m->m_pkthdr.rcvif, m); 898 #endif /* INET */ 899 900 /* 901 * If the INPCB does not exist then all data in the incoming 902 * segment is discarded and an appropriate RST is sent back. 903 * XXX MRT Send RST using which routing table? 904 */ 905 if (inp == NULL) { 906 if (rstreason != 0) { 907 /* We came here after second (safety) lookup. */ 908 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); 909 goto dropwithreset; 910 } 911 /* 912 * Log communication attempts to ports that are not 913 * in use. 914 */ 915 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 916 V_tcp_log_in_vain == 2) { 917 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 918 log(LOG_INFO, "%s; %s: Connection attempt " 919 "to closed port\n", s, __func__); 920 } 921 /* 922 * When blackholing do not respond with a RST but 923 * completely ignore the segment and drop it. 924 */ 925 if (((V_blackhole == 1 && (thflags & TH_SYN)) || 926 V_blackhole == 2) && (V_blackhole_local || ( 927 #ifdef INET6 928 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 929 #endif 930 #ifdef INET 931 !in_localip(ip->ip_src) 932 #else 933 true 934 #endif 935 ))) 936 goto dropunlock; 937 938 rstreason = BANDLIM_RST_CLOSEDPORT; 939 goto dropwithreset; 940 } 941 INP_LOCK_ASSERT(inp); 942 /* 943 * While waiting for inp lock during the lookup, another thread 944 * can have dropped the inpcb, in which case we need to loop back 945 * and try to find a new inpcb to deliver to. 946 */ 947 if (inp->inp_flags & INP_DROPPED) { 948 INP_UNLOCK(inp); 949 inp = NULL; 950 goto findpcb; 951 } 952 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 953 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 954 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) { 955 inp->inp_flowid = m->m_pkthdr.flowid; 956 inp->inp_flowtype = M_HASHTYPE_GET(m); 957 } 958 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 959 #ifdef INET6 960 if (isipv6 && IPSEC_ENABLED(ipv6) && 961 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 962 goto dropunlock; 963 } 964 #ifdef INET 965 else 966 #endif 967 #endif /* INET6 */ 968 #ifdef INET 969 if (IPSEC_ENABLED(ipv4) && 970 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 971 goto dropunlock; 972 } 973 #endif /* INET */ 974 #endif /* IPSEC */ 975 976 /* 977 * Check the minimum TTL for socket. 978 */ 979 if (inp->inp_ip_minttl != 0) { 980 #ifdef INET6 981 if (isipv6) { 982 if (inp->inp_ip_minttl > ip6->ip6_hlim) 983 goto dropunlock; 984 } else 985 #endif 986 if (inp->inp_ip_minttl > ip->ip_ttl) 987 goto dropunlock; 988 } 989 990 /* 991 * A previous connection in TIMEWAIT state is supposed to catch stray 992 * or duplicate segments arriving late. If this segment was a 993 * legitimate new connection attempt, the old INPCB gets removed and 994 * we can try again to find a listening socket. 995 */ 996 if (inp->inp_flags & INP_TIMEWAIT) { 997 tcp_dooptions(&to, optp, optlen, 998 (thflags & TH_SYN) ? TO_SYN : 0); 999 /* 1000 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1001 */ 1002 if (tcp_twcheck(inp, &to, th, m, tlen)) 1003 goto findpcb; 1004 return (IPPROTO_DONE); 1005 } 1006 /* 1007 * The TCPCB may no longer exist if the connection is winding 1008 * down or it is in the CLOSED state. Either way we drop the 1009 * segment and send an appropriate response. 1010 */ 1011 tp = intotcpcb(inp); 1012 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1013 rstreason = BANDLIM_RST_CLOSEDPORT; 1014 goto dropwithreset; 1015 } 1016 1017 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1018 rstreason = BANDLIM_RST_CLOSEDPORT; 1019 goto dropwithreset; 1020 } 1021 1022 #ifdef TCP_OFFLOAD 1023 if (tp->t_flags & TF_TOE) { 1024 tcp_offload_input(tp, m); 1025 m = NULL; /* consumed by the TOE driver */ 1026 goto dropunlock; 1027 } 1028 #endif 1029 1030 #ifdef MAC 1031 if (mac_inpcb_check_deliver(inp, m)) 1032 goto dropunlock; 1033 #endif 1034 so = inp->inp_socket; 1035 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1036 #ifdef TCPDEBUG 1037 if (so->so_options & SO_DEBUG) { 1038 ostate = tp->t_state; 1039 #ifdef INET6 1040 if (isipv6) { 1041 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1042 } else 1043 #endif 1044 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1045 tcp_savetcp = *th; 1046 } 1047 #endif /* TCPDEBUG */ 1048 /* 1049 * When the socket is accepting connections (the INPCB is in LISTEN 1050 * state) we look into the SYN cache if this is a new connection 1051 * attempt or the completion of a previous one. 1052 */ 1053 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1054 ("%s: so accepting but tp %p not listening", __func__, tp)); 1055 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1056 struct in_conninfo inc; 1057 1058 bzero(&inc, sizeof(inc)); 1059 #ifdef INET6 1060 if (isipv6) { 1061 inc.inc_flags |= INC_ISIPV6; 1062 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1063 inc.inc_flags |= INC_IPV6MINMTU; 1064 inc.inc6_faddr = ip6->ip6_src; 1065 inc.inc6_laddr = ip6->ip6_dst; 1066 } else 1067 #endif 1068 { 1069 inc.inc_faddr = ip->ip_src; 1070 inc.inc_laddr = ip->ip_dst; 1071 } 1072 inc.inc_fport = th->th_sport; 1073 inc.inc_lport = th->th_dport; 1074 inc.inc_fibnum = so->so_fibnum; 1075 1076 /* 1077 * Check for an existing connection attempt in syncache if 1078 * the flag is only ACK. A successful lookup creates a new 1079 * socket appended to the listen queue in SYN_RECEIVED state. 1080 */ 1081 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1082 /* 1083 * Parse the TCP options here because 1084 * syncookies need access to the reflected 1085 * timestamp. 1086 */ 1087 tcp_dooptions(&to, optp, optlen, 0); 1088 /* 1089 * NB: syncache_expand() doesn't unlock inp. 1090 */ 1091 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1092 if (rstreason < 0) { 1093 /* 1094 * A failing TCP MD5 signature comparison 1095 * must result in the segment being dropped 1096 * and must not produce any response back 1097 * to the sender. 1098 */ 1099 goto dropunlock; 1100 } else if (rstreason == 0) { 1101 /* 1102 * No syncache entry, or ACK was not for our 1103 * SYN/ACK. Do our protection against double 1104 * ACK. If peer sent us 2 ACKs, then for the 1105 * first one syncache_expand() successfully 1106 * converted syncache entry into a socket, 1107 * while we were waiting on the inpcb lock. We 1108 * don't want to sent RST for the second ACK, 1109 * so we perform second lookup without wildcard 1110 * match, hoping to find the new socket. If 1111 * the ACK is stray indeed, rstreason would 1112 * hint the above code that the lookup was a 1113 * second attempt. 1114 * 1115 * NB: syncache did its own logging 1116 * of the failure cause. 1117 */ 1118 INP_WUNLOCK(inp); 1119 rstreason = BANDLIM_RST_OPENPORT; 1120 lookupflag &= ~INPLOOKUP_WILDCARD; 1121 goto findpcb; 1122 } 1123 tfo_socket_result: 1124 if (so == NULL) { 1125 /* 1126 * We completed the 3-way handshake 1127 * but could not allocate a socket 1128 * either due to memory shortage, 1129 * listen queue length limits or 1130 * global socket limits. Send RST 1131 * or wait and have the remote end 1132 * retransmit the ACK for another 1133 * try. 1134 */ 1135 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1136 log(LOG_DEBUG, "%s; %s: Listen socket: " 1137 "Socket allocation failed due to " 1138 "limits or memory shortage, %s\n", 1139 s, __func__, 1140 V_tcp_sc_rst_sock_fail ? 1141 "sending RST" : "try again"); 1142 if (V_tcp_sc_rst_sock_fail) { 1143 rstreason = BANDLIM_UNLIMITED; 1144 goto dropwithreset; 1145 } else 1146 goto dropunlock; 1147 } 1148 /* 1149 * Socket is created in state SYN_RECEIVED. 1150 * Unlock the listen socket, lock the newly 1151 * created socket and update the tp variable. 1152 * If we came here via jump to tfo_socket_result, 1153 * then listening socket is read-locked. 1154 */ 1155 INP_UNLOCK(inp); /* listen socket */ 1156 inp = sotoinpcb(so); 1157 /* 1158 * New connection inpcb is already locked by 1159 * syncache_expand(). 1160 */ 1161 INP_WLOCK_ASSERT(inp); 1162 tp = intotcpcb(inp); 1163 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1164 ("%s: ", __func__)); 1165 /* 1166 * Process the segment and the data it 1167 * contains. tcp_do_segment() consumes 1168 * the mbuf chain and unlocks the inpcb. 1169 */ 1170 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1171 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1172 iptos); 1173 return (IPPROTO_DONE); 1174 } 1175 /* 1176 * Segment flag validation for new connection attempts: 1177 * 1178 * Our (SYN|ACK) response was rejected. 1179 * Check with syncache and remove entry to prevent 1180 * retransmits. 1181 * 1182 * NB: syncache_chkrst does its own logging of failure 1183 * causes. 1184 */ 1185 if (thflags & TH_RST) { 1186 syncache_chkrst(&inc, th, m, port); 1187 goto dropunlock; 1188 } 1189 /* 1190 * We can't do anything without SYN. 1191 */ 1192 if ((thflags & TH_SYN) == 0) { 1193 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1194 log(LOG_DEBUG, "%s; %s: Listen socket: " 1195 "SYN is missing, segment ignored\n", 1196 s, __func__); 1197 TCPSTAT_INC(tcps_badsyn); 1198 goto dropunlock; 1199 } 1200 /* 1201 * (SYN|ACK) is bogus on a listen socket. 1202 */ 1203 if (thflags & TH_ACK) { 1204 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1205 log(LOG_DEBUG, "%s; %s: Listen socket: " 1206 "SYN|ACK invalid, segment rejected\n", 1207 s, __func__); 1208 syncache_badack(&inc, port); /* XXX: Not needed! */ 1209 TCPSTAT_INC(tcps_badsyn); 1210 rstreason = BANDLIM_RST_OPENPORT; 1211 goto dropwithreset; 1212 } 1213 /* 1214 * If the drop_synfin option is enabled, drop all 1215 * segments with both the SYN and FIN bits set. 1216 * This prevents e.g. nmap from identifying the 1217 * TCP/IP stack. 1218 * XXX: Poor reasoning. nmap has other methods 1219 * and is constantly refining its stack detection 1220 * strategies. 1221 * XXX: This is a violation of the TCP specification 1222 * and was used by RFC1644. 1223 */ 1224 if ((thflags & TH_FIN) && V_drop_synfin) { 1225 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1226 log(LOG_DEBUG, "%s; %s: Listen socket: " 1227 "SYN|FIN segment ignored (based on " 1228 "sysctl setting)\n", s, __func__); 1229 TCPSTAT_INC(tcps_badsyn); 1230 goto dropunlock; 1231 } 1232 /* 1233 * Segment's flags are (SYN) or (SYN|FIN). 1234 * 1235 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1236 * as they do not affect the state of the TCP FSM. 1237 * The data pointed to by TH_URG and th_urp is ignored. 1238 */ 1239 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1240 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1241 KASSERT(thflags & (TH_SYN), 1242 ("%s: Listen socket: TH_SYN not set", __func__)); 1243 INP_RLOCK_ASSERT(inp); 1244 #ifdef INET6 1245 /* 1246 * If deprecated address is forbidden, 1247 * we do not accept SYN to deprecated interface 1248 * address to prevent any new inbound connection from 1249 * getting established. 1250 * When we do not accept SYN, we send a TCP RST, 1251 * with deprecated source address (instead of dropping 1252 * it). We compromise it as it is much better for peer 1253 * to send a RST, and RST will be the final packet 1254 * for the exchange. 1255 * 1256 * If we do not forbid deprecated addresses, we accept 1257 * the SYN packet. RFC2462 does not suggest dropping 1258 * SYN in this case. 1259 * If we decipher RFC2462 5.5.4, it says like this: 1260 * 1. use of deprecated addr with existing 1261 * communication is okay - "SHOULD continue to be 1262 * used" 1263 * 2. use of it with new communication: 1264 * (2a) "SHOULD NOT be used if alternate address 1265 * with sufficient scope is available" 1266 * (2b) nothing mentioned otherwise. 1267 * Here we fall into (2b) case as we have no choice in 1268 * our source address selection - we must obey the peer. 1269 * 1270 * The wording in RFC2462 is confusing, and there are 1271 * multiple description text for deprecated address 1272 * handling - worse, they are not exactly the same. 1273 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1274 */ 1275 if (isipv6 && !V_ip6_use_deprecated) { 1276 struct in6_ifaddr *ia6; 1277 1278 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1279 if (ia6 != NULL && 1280 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1281 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1282 log(LOG_DEBUG, "%s; %s: Listen socket: " 1283 "Connection attempt to deprecated " 1284 "IPv6 address rejected\n", 1285 s, __func__); 1286 rstreason = BANDLIM_RST_OPENPORT; 1287 goto dropwithreset; 1288 } 1289 } 1290 #endif /* INET6 */ 1291 /* 1292 * Basic sanity checks on incoming SYN requests: 1293 * Don't respond if the destination is a link layer 1294 * broadcast according to RFC1122 4.2.3.10, p. 104. 1295 * If it is from this socket it must be forged. 1296 * Don't respond if the source or destination is a 1297 * global or subnet broad- or multicast address. 1298 * Note that it is quite possible to receive unicast 1299 * link-layer packets with a broadcast IP address. Use 1300 * in_broadcast() to find them. 1301 */ 1302 if (m->m_flags & (M_BCAST|M_MCAST)) { 1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1304 log(LOG_DEBUG, "%s; %s: Listen socket: " 1305 "Connection attempt from broad- or multicast " 1306 "link layer address ignored\n", s, __func__); 1307 goto dropunlock; 1308 } 1309 #ifdef INET6 1310 if (isipv6) { 1311 if (th->th_dport == th->th_sport && 1312 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1313 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1314 log(LOG_DEBUG, "%s; %s: Listen socket: " 1315 "Connection attempt to/from self " 1316 "ignored\n", s, __func__); 1317 goto dropunlock; 1318 } 1319 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1320 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1321 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1322 log(LOG_DEBUG, "%s; %s: Listen socket: " 1323 "Connection attempt from/to multicast " 1324 "address ignored\n", s, __func__); 1325 goto dropunlock; 1326 } 1327 } 1328 #endif 1329 #if defined(INET) && defined(INET6) 1330 else 1331 #endif 1332 #ifdef INET 1333 { 1334 if (th->th_dport == th->th_sport && 1335 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1336 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1337 log(LOG_DEBUG, "%s; %s: Listen socket: " 1338 "Connection attempt from/to self " 1339 "ignored\n", s, __func__); 1340 goto dropunlock; 1341 } 1342 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1343 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1344 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1345 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1346 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1347 log(LOG_DEBUG, "%s; %s: Listen socket: " 1348 "Connection attempt from/to broad- " 1349 "or multicast address ignored\n", 1350 s, __func__); 1351 goto dropunlock; 1352 } 1353 } 1354 #endif 1355 /* 1356 * SYN appears to be valid. Create compressed TCP state 1357 * for syncache. 1358 */ 1359 #ifdef TCPDEBUG 1360 if (so->so_options & SO_DEBUG) 1361 tcp_trace(TA_INPUT, ostate, tp, 1362 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1363 #endif 1364 TCP_PROBE3(debug__input, tp, th, m); 1365 tcp_dooptions(&to, optp, optlen, TO_SYN); 1366 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1367 iptos, port)) != NULL) 1368 goto tfo_socket_result; 1369 1370 /* 1371 * Entry added to syncache and mbuf consumed. 1372 * Only the listen socket is unlocked by syncache_add(). 1373 */ 1374 return (IPPROTO_DONE); 1375 } else if (tp->t_state == TCPS_LISTEN) { 1376 /* 1377 * When a listen socket is torn down the SO_ACCEPTCONN 1378 * flag is removed first while connections are drained 1379 * from the accept queue in a unlock/lock cycle of the 1380 * ACCEPT_LOCK, opening a race condition allowing a SYN 1381 * attempt go through unhandled. 1382 */ 1383 goto dropunlock; 1384 } 1385 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1386 if (tp->t_flags & TF_SIGNATURE) { 1387 tcp_dooptions(&to, optp, optlen, thflags); 1388 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1389 TCPSTAT_INC(tcps_sig_err_nosigopt); 1390 goto dropunlock; 1391 } 1392 if (!TCPMD5_ENABLED() || 1393 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1394 goto dropunlock; 1395 } 1396 #endif 1397 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1398 1399 /* 1400 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1401 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1402 * the inpcb, and unlocks pcbinfo. 1403 * 1404 * XXXGL: in case of a pure SYN arriving on existing connection 1405 * TCP stacks won't need to modify the PCB, they would either drop 1406 * the segment silently, or send a challenge ACK. However, we try 1407 * to upgrade the lock, because calling convention for stacks is 1408 * write-lock on PCB. If upgrade fails, drop the SYN. 1409 */ 1410 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1411 goto dropunlock; 1412 1413 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1414 return (IPPROTO_DONE); 1415 1416 dropwithreset: 1417 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1418 1419 if (inp != NULL) { 1420 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1421 INP_UNLOCK(inp); 1422 } else 1423 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1424 m = NULL; /* mbuf chain got consumed. */ 1425 goto drop; 1426 1427 dropunlock: 1428 if (m != NULL) 1429 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1430 1431 if (inp != NULL) 1432 INP_UNLOCK(inp); 1433 1434 drop: 1435 if (s != NULL) 1436 free(s, M_TCPLOG); 1437 if (m != NULL) 1438 m_freem(m); 1439 return (IPPROTO_DONE); 1440 } 1441 1442 /* 1443 * Automatic sizing of receive socket buffer. Often the send 1444 * buffer size is not optimally adjusted to the actual network 1445 * conditions at hand (delay bandwidth product). Setting the 1446 * buffer size too small limits throughput on links with high 1447 * bandwidth and high delay (eg. trans-continental/oceanic links). 1448 * 1449 * On the receive side the socket buffer memory is only rarely 1450 * used to any significant extent. This allows us to be much 1451 * more aggressive in scaling the receive socket buffer. For 1452 * the case that the buffer space is actually used to a large 1453 * extent and we run out of kernel memory we can simply drop 1454 * the new segments; TCP on the sender will just retransmit it 1455 * later. Setting the buffer size too big may only consume too 1456 * much kernel memory if the application doesn't read() from 1457 * the socket or packet loss or reordering makes use of the 1458 * reassembly queue. 1459 * 1460 * The criteria to step up the receive buffer one notch are: 1461 * 1. Application has not set receive buffer size with 1462 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1463 * 2. the number of bytes received during 1/2 of an sRTT 1464 * is at least 3/8 of the current socket buffer size. 1465 * 3. receive buffer size has not hit maximal automatic size; 1466 * 1467 * If all of the criteria are met we increaset the socket buffer 1468 * by a 1/2 (bounded by the max). This allows us to keep ahead 1469 * of slow-start but also makes it so our peer never gets limited 1470 * by our rwnd which we then open up causing a burst. 1471 * 1472 * This algorithm does two steps per RTT at most and only if 1473 * we receive a bulk stream w/o packet losses or reorderings. 1474 * Shrinking the buffer during idle times is not necessary as 1475 * it doesn't consume any memory when idle. 1476 * 1477 * TODO: Only step up if the application is actually serving 1478 * the buffer to better manage the socket buffer resources. 1479 */ 1480 int 1481 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1482 struct tcpcb *tp, int tlen) 1483 { 1484 int newsize = 0; 1485 1486 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1487 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1488 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1489 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1490 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1491 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1492 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1493 } 1494 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1495 1496 /* Start over with next RTT. */ 1497 tp->rfbuf_ts = 0; 1498 tp->rfbuf_cnt = 0; 1499 } else { 1500 tp->rfbuf_cnt += tlen; /* add up */ 1501 } 1502 return (newsize); 1503 } 1504 1505 int 1506 tcp_input(struct mbuf **mp, int *offp, int proto) 1507 { 1508 return(tcp_input_with_port(mp, offp, proto, 0)); 1509 } 1510 1511 void 1512 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1513 { 1514 /* 1515 * Since tp might be gone if the session entered 1516 * the TIME_WAIT state before coming here, we need 1517 * to check if the socket is still connected. 1518 */ 1519 if (tp == NULL) { 1520 return; 1521 } 1522 if (so == NULL) { 1523 return; 1524 } 1525 INP_LOCK_ASSERT(tp->t_inpcb); 1526 if (tp->t_flags & TF_WAKESOR) { 1527 tp->t_flags &= ~TF_WAKESOR; 1528 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1529 sorwakeup_locked(so); 1530 } 1531 } 1532 1533 void 1534 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1535 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1536 { 1537 uint16_t thflags; 1538 int acked, ourfinisacked, needoutput = 0, sack_changed; 1539 int rstreason, todrop, win, incforsyn = 0; 1540 uint32_t tiwin; 1541 uint16_t nsegs; 1542 char *s; 1543 struct in_conninfo *inc; 1544 struct mbuf *mfree; 1545 struct tcpopt to; 1546 int tfo_syn; 1547 u_int maxseg; 1548 1549 #ifdef TCPDEBUG 1550 /* 1551 * The size of tcp_saveipgen must be the size of the max ip header, 1552 * now IPv6. 1553 */ 1554 u_char tcp_saveipgen[IP6_HDR_LEN]; 1555 struct tcphdr tcp_savetcp; 1556 short ostate = 0; 1557 #endif 1558 thflags = tcp_get_flags(th); 1559 inc = &tp->t_inpcb->inp_inc; 1560 tp->sackhint.last_sack_ack = 0; 1561 sack_changed = 0; 1562 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1563 1564 NET_EPOCH_ASSERT(); 1565 INP_WLOCK_ASSERT(tp->t_inpcb); 1566 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1567 __func__)); 1568 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1569 __func__)); 1570 1571 #ifdef TCPPCAP 1572 /* Save segment, if requested. */ 1573 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1574 #endif 1575 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1576 tlen, NULL, true); 1577 1578 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1579 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1580 log(LOG_DEBUG, "%s; %s: " 1581 "SYN|FIN segment ignored (based on " 1582 "sysctl setting)\n", s, __func__); 1583 free(s, M_TCPLOG); 1584 } 1585 goto drop; 1586 } 1587 1588 /* 1589 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1590 * check SEQ.ACK first. 1591 */ 1592 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1593 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1594 rstreason = BANDLIM_UNLIMITED; 1595 goto dropwithreset; 1596 } 1597 1598 /* 1599 * Segment received on connection. 1600 * Reset idle time and keep-alive timer. 1601 * XXX: This should be done after segment 1602 * validation to ignore broken/spoofed segs. 1603 */ 1604 tp->t_rcvtime = ticks; 1605 1606 /* 1607 * Scale up the window into a 32-bit value. 1608 * For the SYN_SENT state the scale is zero. 1609 */ 1610 tiwin = th->th_win << tp->snd_scale; 1611 #ifdef STATS 1612 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1613 #endif 1614 1615 /* 1616 * TCP ECN processing. 1617 */ 1618 if (tcp_ecn_input_segment(tp, thflags, iptos)) 1619 cc_cong_signal(tp, th, CC_ECN); 1620 1621 /* 1622 * Parse options on any incoming segment. 1623 */ 1624 tcp_dooptions(&to, (u_char *)(th + 1), 1625 (th->th_off << 2) - sizeof(struct tcphdr), 1626 (thflags & TH_SYN) ? TO_SYN : 0); 1627 1628 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1629 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1630 (to.to_flags & TOF_SIGNATURE) == 0) { 1631 TCPSTAT_INC(tcps_sig_err_sigopt); 1632 /* XXX: should drop? */ 1633 } 1634 #endif 1635 /* 1636 * If echoed timestamp is later than the current time, 1637 * fall back to non RFC1323 RTT calculation. Normalize 1638 * timestamp if syncookies were used when this connection 1639 * was established. 1640 */ 1641 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1642 to.to_tsecr -= tp->ts_offset; 1643 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1644 to.to_tsecr = 0; 1645 else if (tp->t_rxtshift == 1 && 1646 tp->t_flags & TF_PREVVALID && 1647 tp->t_badrxtwin != 0 && 1648 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 1649 cc_cong_signal(tp, th, CC_RTO_ERR); 1650 } 1651 /* 1652 * Process options only when we get SYN/ACK back. The SYN case 1653 * for incoming connections is handled in tcp_syncache. 1654 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1655 * or <SYN,ACK>) segment itself is never scaled. 1656 * XXX this is traditional behavior, may need to be cleaned up. 1657 */ 1658 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1659 /* Handle parallel SYN for ECN */ 1660 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1661 if ((to.to_flags & TOF_SCALE) && 1662 (tp->t_flags & TF_REQ_SCALE) && 1663 !(tp->t_flags & TF_NOOPT)) { 1664 tp->t_flags |= TF_RCVD_SCALE; 1665 tp->snd_scale = to.to_wscale; 1666 } else 1667 tp->t_flags &= ~TF_REQ_SCALE; 1668 /* 1669 * Initial send window. It will be updated with 1670 * the next incoming segment to the scaled value. 1671 */ 1672 tp->snd_wnd = th->th_win; 1673 if ((to.to_flags & TOF_TS) && 1674 (tp->t_flags & TF_REQ_TSTMP) && 1675 !(tp->t_flags & TF_NOOPT)) { 1676 tp->t_flags |= TF_RCVD_TSTMP; 1677 tp->ts_recent = to.to_tsval; 1678 tp->ts_recent_age = tcp_ts_getticks(); 1679 } else 1680 tp->t_flags &= ~TF_REQ_TSTMP; 1681 if (to.to_flags & TOF_MSS) 1682 tcp_mss(tp, to.to_mss); 1683 if ((tp->t_flags & TF_SACK_PERMIT) && 1684 (!(to.to_flags & TOF_SACKPERM) || 1685 (tp->t_flags & TF_NOOPT))) 1686 tp->t_flags &= ~TF_SACK_PERMIT; 1687 if (IS_FASTOPEN(tp->t_flags)) { 1688 if ((to.to_flags & TOF_FASTOPEN) && 1689 !(tp->t_flags & TF_NOOPT)) { 1690 uint16_t mss; 1691 1692 if (to.to_flags & TOF_MSS) 1693 mss = to.to_mss; 1694 else 1695 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1696 mss = TCP6_MSS; 1697 else 1698 mss = TCP_MSS; 1699 tcp_fastopen_update_cache(tp, mss, 1700 to.to_tfo_len, to.to_tfo_cookie); 1701 } else 1702 tcp_fastopen_disable_path(tp); 1703 } 1704 } 1705 1706 /* 1707 * If timestamps were negotiated during SYN/ACK and a 1708 * segment without a timestamp is received, silently drop 1709 * the segment, unless it is a RST segment or missing timestamps are 1710 * tolerated. 1711 * See section 3.2 of RFC 7323. 1712 */ 1713 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1714 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1715 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1716 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1717 "segment processed normally\n", 1718 s, __func__); 1719 free(s, M_TCPLOG); 1720 } 1721 } else { 1722 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1723 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1724 "segment silently dropped\n", s, __func__); 1725 free(s, M_TCPLOG); 1726 } 1727 goto drop; 1728 } 1729 } 1730 /* 1731 * If timestamps were not negotiated during SYN/ACK and a 1732 * segment with a timestamp is received, ignore the 1733 * timestamp and process the packet normally. 1734 * See section 3.2 of RFC 7323. 1735 */ 1736 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1737 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1738 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1739 "segment processed normally\n", s, __func__); 1740 free(s, M_TCPLOG); 1741 } 1742 } 1743 1744 /* 1745 * Header prediction: check for the two common cases 1746 * of a uni-directional data xfer. If the packet has 1747 * no control flags, is in-sequence, the window didn't 1748 * change and we're not retransmitting, it's a 1749 * candidate. If the length is zero and the ack moved 1750 * forward, we're the sender side of the xfer. Just 1751 * free the data acked & wake any higher level process 1752 * that was blocked waiting for space. If the length 1753 * is non-zero and the ack didn't move, we're the 1754 * receiver side. If we're getting packets in-order 1755 * (the reassembly queue is empty), add the data to 1756 * the socket buffer and note that we need a delayed ack. 1757 * Make sure that the hidden state-flags are also off. 1758 * Since we check for TCPS_ESTABLISHED first, it can only 1759 * be TH_NEEDSYN. 1760 */ 1761 if (tp->t_state == TCPS_ESTABLISHED && 1762 th->th_seq == tp->rcv_nxt && 1763 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1764 tp->snd_nxt == tp->snd_max && 1765 tiwin && tiwin == tp->snd_wnd && 1766 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1767 SEGQ_EMPTY(tp) && 1768 ((to.to_flags & TOF_TS) == 0 || 1769 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1770 /* 1771 * If last ACK falls within this segment's sequence numbers, 1772 * record the timestamp. 1773 * NOTE that the test is modified according to the latest 1774 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1775 */ 1776 if ((to.to_flags & TOF_TS) != 0 && 1777 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1778 tp->ts_recent_age = tcp_ts_getticks(); 1779 tp->ts_recent = to.to_tsval; 1780 } 1781 1782 if (tlen == 0) { 1783 if (SEQ_GT(th->th_ack, tp->snd_una) && 1784 SEQ_LEQ(th->th_ack, tp->snd_max) && 1785 !IN_RECOVERY(tp->t_flags) && 1786 (to.to_flags & TOF_SACK) == 0 && 1787 TAILQ_EMPTY(&tp->snd_holes)) { 1788 /* 1789 * This is a pure ack for outstanding data. 1790 */ 1791 TCPSTAT_INC(tcps_predack); 1792 1793 /* 1794 * "bad retransmit" recovery without timestamps. 1795 */ 1796 if ((to.to_flags & TOF_TS) == 0 && 1797 tp->t_rxtshift == 1 && 1798 tp->t_flags & TF_PREVVALID && 1799 tp->t_badrxtwin != 0 && 1800 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1801 cc_cong_signal(tp, th, CC_RTO_ERR); 1802 } 1803 1804 /* 1805 * Recalculate the transmit timer / rtt. 1806 * 1807 * Some boxes send broken timestamp replies 1808 * during the SYN+ACK phase, ignore 1809 * timestamps of 0 or we could calculate a 1810 * huge RTT and blow up the retransmit timer. 1811 */ 1812 if ((to.to_flags & TOF_TS) != 0 && 1813 to.to_tsecr) { 1814 uint32_t t; 1815 1816 t = tcp_ts_getticks() - to.to_tsecr; 1817 if (!tp->t_rttlow || tp->t_rttlow > t) 1818 tp->t_rttlow = t; 1819 tcp_xmit_timer(tp, 1820 TCP_TS_TO_TICKS(t) + 1); 1821 } else if (tp->t_rtttime && 1822 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1823 if (!tp->t_rttlow || 1824 tp->t_rttlow > ticks - tp->t_rtttime) 1825 tp->t_rttlow = ticks - tp->t_rtttime; 1826 tcp_xmit_timer(tp, 1827 ticks - tp->t_rtttime); 1828 } 1829 acked = BYTES_THIS_ACK(tp, th); 1830 1831 #ifdef TCP_HHOOK 1832 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1833 hhook_run_tcp_est_in(tp, th, &to); 1834 #endif 1835 1836 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1837 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1838 sbdrop(&so->so_snd, acked); 1839 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1840 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1841 tp->snd_recover = th->th_ack - 1; 1842 1843 /* 1844 * Let the congestion control algorithm update 1845 * congestion control related information. This 1846 * typically means increasing the congestion 1847 * window. 1848 */ 1849 cc_ack_received(tp, th, nsegs, CC_ACK); 1850 1851 tp->snd_una = th->th_ack; 1852 /* 1853 * Pull snd_wl2 up to prevent seq wrap relative 1854 * to th_ack. 1855 */ 1856 tp->snd_wl2 = th->th_ack; 1857 tp->t_dupacks = 0; 1858 m_freem(m); 1859 1860 /* 1861 * If all outstanding data are acked, stop 1862 * retransmit timer, otherwise restart timer 1863 * using current (possibly backed-off) value. 1864 * If process is waiting for space, 1865 * wakeup/selwakeup/signal. If data 1866 * are ready to send, let tcp_output 1867 * decide between more output or persist. 1868 */ 1869 #ifdef TCPDEBUG 1870 if (so->so_options & SO_DEBUG) 1871 tcp_trace(TA_INPUT, ostate, tp, 1872 (void *)tcp_saveipgen, 1873 &tcp_savetcp, 0); 1874 #endif 1875 TCP_PROBE3(debug__input, tp, th, m); 1876 if (tp->snd_una == tp->snd_max) 1877 tcp_timer_activate(tp, TT_REXMT, 0); 1878 else if (!tcp_timer_active(tp, TT_PERSIST)) 1879 tcp_timer_activate(tp, TT_REXMT, 1880 tp->t_rxtcur); 1881 sowwakeup(so); 1882 if (sbavail(&so->so_snd)) 1883 (void) tcp_output(tp); 1884 goto check_delack; 1885 } 1886 } else if (th->th_ack == tp->snd_una && 1887 tlen <= sbspace(&so->so_rcv)) { 1888 int newsize = 0; /* automatic sockbuf scaling */ 1889 1890 /* 1891 * This is a pure, in-sequence data packet with 1892 * nothing on the reassembly queue and we have enough 1893 * buffer space to take it. 1894 */ 1895 /* Clean receiver SACK report if present */ 1896 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1897 tcp_clean_sackreport(tp); 1898 TCPSTAT_INC(tcps_preddat); 1899 tp->rcv_nxt += tlen; 1900 if (tlen && 1901 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1902 (tp->t_fbyte_in == 0)) { 1903 tp->t_fbyte_in = ticks; 1904 if (tp->t_fbyte_in == 0) 1905 tp->t_fbyte_in = 1; 1906 if (tp->t_fbyte_out && tp->t_fbyte_in) 1907 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1908 } 1909 /* 1910 * Pull snd_wl1 up to prevent seq wrap relative to 1911 * th_seq. 1912 */ 1913 tp->snd_wl1 = th->th_seq; 1914 /* 1915 * Pull rcv_up up to prevent seq wrap relative to 1916 * rcv_nxt. 1917 */ 1918 tp->rcv_up = tp->rcv_nxt; 1919 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1920 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1921 #ifdef TCPDEBUG 1922 if (so->so_options & SO_DEBUG) 1923 tcp_trace(TA_INPUT, ostate, tp, 1924 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1925 #endif 1926 TCP_PROBE3(debug__input, tp, th, m); 1927 1928 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1929 1930 /* Add data to socket buffer. */ 1931 SOCKBUF_LOCK(&so->so_rcv); 1932 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1933 m_freem(m); 1934 } else { 1935 /* 1936 * Set new socket buffer size. 1937 * Give up when limit is reached. 1938 */ 1939 if (newsize) 1940 if (!sbreserve_locked(so, SO_RCV, 1941 newsize, NULL)) 1942 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1943 m_adj(m, drop_hdrlen); /* delayed header drop */ 1944 sbappendstream_locked(&so->so_rcv, m, 0); 1945 } 1946 /* NB: sorwakeup_locked() does an implicit unlock. */ 1947 sorwakeup_locked(so); 1948 if (DELAY_ACK(tp, tlen)) { 1949 tp->t_flags |= TF_DELACK; 1950 } else { 1951 tp->t_flags |= TF_ACKNOW; 1952 tcp_output(tp); 1953 } 1954 goto check_delack; 1955 } 1956 } 1957 1958 /* 1959 * Calculate amount of space in receive window, 1960 * and then do TCP input processing. 1961 * Receive window is amount of space in rcv queue, 1962 * but not less than advertised window. 1963 */ 1964 win = sbspace(&so->so_rcv); 1965 if (win < 0) 1966 win = 0; 1967 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1968 1969 switch (tp->t_state) { 1970 /* 1971 * If the state is SYN_RECEIVED: 1972 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1973 */ 1974 case TCPS_SYN_RECEIVED: 1975 if ((thflags & TH_ACK) && 1976 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1977 SEQ_GT(th->th_ack, tp->snd_max))) { 1978 rstreason = BANDLIM_RST_OPENPORT; 1979 goto dropwithreset; 1980 } 1981 if (IS_FASTOPEN(tp->t_flags)) { 1982 /* 1983 * When a TFO connection is in SYN_RECEIVED, the 1984 * only valid packets are the initial SYN, a 1985 * retransmit/copy of the initial SYN (possibly with 1986 * a subset of the original data), a valid ACK, a 1987 * FIN, or a RST. 1988 */ 1989 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1990 rstreason = BANDLIM_RST_OPENPORT; 1991 goto dropwithreset; 1992 } else if (thflags & TH_SYN) { 1993 /* non-initial SYN is ignored */ 1994 if ((tcp_timer_active(tp, TT_DELACK) || 1995 tcp_timer_active(tp, TT_REXMT))) 1996 goto drop; 1997 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1998 goto drop; 1999 } 2000 } 2001 break; 2002 2003 /* 2004 * If the state is SYN_SENT: 2005 * if seg contains a RST with valid ACK (SEQ.ACK has already 2006 * been verified), then drop the connection. 2007 * if seg contains a RST without an ACK, drop the seg. 2008 * if seg does not contain SYN, then drop the seg. 2009 * Otherwise this is an acceptable SYN segment 2010 * initialize tp->rcv_nxt and tp->irs 2011 * if seg contains ack then advance tp->snd_una 2012 * if seg contains an ECE and ECN support is enabled, the stream 2013 * is ECN capable. 2014 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2015 * arrange for segment to be acked (eventually) 2016 * continue processing rest of data/controls, beginning with URG 2017 */ 2018 case TCPS_SYN_SENT: 2019 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2020 TCP_PROBE5(connect__refused, NULL, tp, 2021 m, tp, th); 2022 tp = tcp_drop(tp, ECONNREFUSED); 2023 } 2024 if (thflags & TH_RST) 2025 goto drop; 2026 if (!(thflags & TH_SYN)) 2027 goto drop; 2028 2029 tp->irs = th->th_seq; 2030 tcp_rcvseqinit(tp); 2031 if (thflags & TH_ACK) { 2032 int tfo_partial_ack = 0; 2033 2034 TCPSTAT_INC(tcps_connects); 2035 soisconnected(so); 2036 #ifdef MAC 2037 mac_socketpeer_set_from_mbuf(m, so); 2038 #endif 2039 /* Do window scaling on this connection? */ 2040 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2041 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2042 tp->rcv_scale = tp->request_r_scale; 2043 } 2044 tp->rcv_adv += min(tp->rcv_wnd, 2045 TCP_MAXWIN << tp->rcv_scale); 2046 tp->snd_una++; /* SYN is acked */ 2047 /* 2048 * If not all the data that was sent in the TFO SYN 2049 * has been acked, resend the remainder right away. 2050 */ 2051 if (IS_FASTOPEN(tp->t_flags) && 2052 (tp->snd_una != tp->snd_max)) { 2053 tp->snd_nxt = th->th_ack; 2054 tfo_partial_ack = 1; 2055 } 2056 /* 2057 * If there's data, delay ACK; if there's also a FIN 2058 * ACKNOW will be turned on later. 2059 */ 2060 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2061 tcp_timer_activate(tp, TT_DELACK, 2062 tcp_delacktime); 2063 else 2064 tp->t_flags |= TF_ACKNOW; 2065 2066 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2067 2068 /* 2069 * Received <SYN,ACK> in SYN_SENT[*] state. 2070 * Transitions: 2071 * SYN_SENT --> ESTABLISHED 2072 * SYN_SENT* --> FIN_WAIT_1 2073 */ 2074 tp->t_starttime = ticks; 2075 if (tp->t_flags & TF_NEEDFIN) { 2076 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2077 tp->t_flags &= ~TF_NEEDFIN; 2078 thflags &= ~TH_SYN; 2079 } else { 2080 tcp_state_change(tp, TCPS_ESTABLISHED); 2081 TCP_PROBE5(connect__established, NULL, tp, 2082 m, tp, th); 2083 cc_conn_init(tp); 2084 tcp_timer_activate(tp, TT_KEEP, 2085 TP_KEEPIDLE(tp)); 2086 } 2087 } else { 2088 /* 2089 * Received initial SYN in SYN-SENT[*] state => 2090 * simultaneous open. 2091 * If it succeeds, connection is * half-synchronized. 2092 * Otherwise, do 3-way handshake: 2093 * SYN-SENT -> SYN-RECEIVED 2094 * SYN-SENT* -> SYN-RECEIVED* 2095 */ 2096 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2097 tcp_timer_activate(tp, TT_REXMT, 0); 2098 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2099 } 2100 2101 INP_WLOCK_ASSERT(tp->t_inpcb); 2102 2103 /* 2104 * Advance th->th_seq to correspond to first data byte. 2105 * If data, trim to stay within window, 2106 * dropping FIN if necessary. 2107 */ 2108 th->th_seq++; 2109 if (tlen > tp->rcv_wnd) { 2110 todrop = tlen - tp->rcv_wnd; 2111 m_adj(m, -todrop); 2112 tlen = tp->rcv_wnd; 2113 thflags &= ~TH_FIN; 2114 TCPSTAT_INC(tcps_rcvpackafterwin); 2115 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2116 } 2117 tp->snd_wl1 = th->th_seq - 1; 2118 tp->rcv_up = th->th_seq; 2119 /* 2120 * Client side of transaction: already sent SYN and data. 2121 * If the remote host used T/TCP to validate the SYN, 2122 * our data will be ACK'd; if so, enter normal data segment 2123 * processing in the middle of step 5, ack processing. 2124 * Otherwise, goto step 6. 2125 */ 2126 if (thflags & TH_ACK) 2127 goto process_ACK; 2128 2129 goto step6; 2130 2131 /* 2132 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2133 * do normal processing. 2134 * 2135 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2136 */ 2137 case TCPS_LAST_ACK: 2138 case TCPS_CLOSING: 2139 break; /* continue normal processing */ 2140 } 2141 2142 /* 2143 * States other than LISTEN or SYN_SENT. 2144 * First check the RST flag and sequence number since reset segments 2145 * are exempt from the timestamp and connection count tests. This 2146 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2147 * below which allowed reset segments in half the sequence space 2148 * to fall though and be processed (which gives forged reset 2149 * segments with a random sequence number a 50 percent chance of 2150 * killing a connection). 2151 * Then check timestamp, if present. 2152 * Then check the connection count, if present. 2153 * Then check that at least some bytes of segment are within 2154 * receive window. If segment begins before rcv_nxt, 2155 * drop leading data (and SYN); if nothing left, just ack. 2156 */ 2157 if (thflags & TH_RST) { 2158 /* 2159 * RFC5961 Section 3.2 2160 * 2161 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2162 * - If RST is in window, we send challenge ACK. 2163 * 2164 * Note: to take into account delayed ACKs, we should 2165 * test against last_ack_sent instead of rcv_nxt. 2166 * Note 2: we handle special case of closed window, not 2167 * covered by the RFC. 2168 */ 2169 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2170 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2171 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2172 KASSERT(tp->t_state != TCPS_SYN_SENT, 2173 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2174 __func__, th, tp)); 2175 2176 if (V_tcp_insecure_rst || 2177 tp->last_ack_sent == th->th_seq) { 2178 TCPSTAT_INC(tcps_drops); 2179 /* Drop the connection. */ 2180 switch (tp->t_state) { 2181 case TCPS_SYN_RECEIVED: 2182 so->so_error = ECONNREFUSED; 2183 goto close; 2184 case TCPS_ESTABLISHED: 2185 case TCPS_FIN_WAIT_1: 2186 case TCPS_FIN_WAIT_2: 2187 case TCPS_CLOSE_WAIT: 2188 case TCPS_CLOSING: 2189 case TCPS_LAST_ACK: 2190 so->so_error = ECONNRESET; 2191 close: 2192 /* FALLTHROUGH */ 2193 default: 2194 tp = tcp_close(tp); 2195 } 2196 } else { 2197 TCPSTAT_INC(tcps_badrst); 2198 /* Send challenge ACK. */ 2199 tcp_respond(tp, mtod(m, void *), th, m, 2200 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2201 tp->last_ack_sent = tp->rcv_nxt; 2202 m = NULL; 2203 } 2204 } 2205 goto drop; 2206 } 2207 2208 /* 2209 * RFC5961 Section 4.2 2210 * Send challenge ACK for any SYN in synchronized state. 2211 */ 2212 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2213 tp->t_state != TCPS_SYN_RECEIVED) { 2214 TCPSTAT_INC(tcps_badsyn); 2215 if (V_tcp_insecure_syn && 2216 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2217 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2218 tp = tcp_drop(tp, ECONNRESET); 2219 rstreason = BANDLIM_UNLIMITED; 2220 } else { 2221 /* Send challenge ACK. */ 2222 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2223 tp->snd_nxt, TH_ACK); 2224 tp->last_ack_sent = tp->rcv_nxt; 2225 m = NULL; 2226 } 2227 goto drop; 2228 } 2229 2230 /* 2231 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2232 * and it's less than ts_recent, drop it. 2233 */ 2234 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2235 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2236 /* Check to see if ts_recent is over 24 days old. */ 2237 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2238 /* 2239 * Invalidate ts_recent. If this segment updates 2240 * ts_recent, the age will be reset later and ts_recent 2241 * will get a valid value. If it does not, setting 2242 * ts_recent to zero will at least satisfy the 2243 * requirement that zero be placed in the timestamp 2244 * echo reply when ts_recent isn't valid. The 2245 * age isn't reset until we get a valid ts_recent 2246 * because we don't want out-of-order segments to be 2247 * dropped when ts_recent is old. 2248 */ 2249 tp->ts_recent = 0; 2250 } else { 2251 TCPSTAT_INC(tcps_rcvduppack); 2252 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2253 TCPSTAT_INC(tcps_pawsdrop); 2254 if (tlen) 2255 goto dropafterack; 2256 goto drop; 2257 } 2258 } 2259 2260 /* 2261 * In the SYN-RECEIVED state, validate that the packet belongs to 2262 * this connection before trimming the data to fit the receive 2263 * window. Check the sequence number versus IRS since we know 2264 * the sequence numbers haven't wrapped. This is a partial fix 2265 * for the "LAND" DoS attack. 2266 */ 2267 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2268 rstreason = BANDLIM_RST_OPENPORT; 2269 goto dropwithreset; 2270 } 2271 2272 todrop = tp->rcv_nxt - th->th_seq; 2273 if (todrop > 0) { 2274 if (thflags & TH_SYN) { 2275 thflags &= ~TH_SYN; 2276 th->th_seq++; 2277 if (th->th_urp > 1) 2278 th->th_urp--; 2279 else 2280 thflags &= ~TH_URG; 2281 todrop--; 2282 } 2283 /* 2284 * Following if statement from Stevens, vol. 2, p. 960. 2285 */ 2286 if (todrop > tlen 2287 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2288 /* 2289 * Any valid FIN must be to the left of the window. 2290 * At this point the FIN must be a duplicate or out 2291 * of sequence; drop it. 2292 */ 2293 thflags &= ~TH_FIN; 2294 2295 /* 2296 * Send an ACK to resynchronize and drop any data. 2297 * But keep on processing for RST or ACK. 2298 */ 2299 tp->t_flags |= TF_ACKNOW; 2300 todrop = tlen; 2301 TCPSTAT_INC(tcps_rcvduppack); 2302 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2303 } else { 2304 TCPSTAT_INC(tcps_rcvpartduppack); 2305 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2306 } 2307 /* 2308 * DSACK - add SACK block for dropped range 2309 */ 2310 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2311 tcp_update_sack_list(tp, th->th_seq, 2312 th->th_seq + todrop); 2313 /* 2314 * ACK now, as the next in-sequence segment 2315 * will clear the DSACK block again 2316 */ 2317 tp->t_flags |= TF_ACKNOW; 2318 } 2319 drop_hdrlen += todrop; /* drop from the top afterwards */ 2320 th->th_seq += todrop; 2321 tlen -= todrop; 2322 if (th->th_urp > todrop) 2323 th->th_urp -= todrop; 2324 else { 2325 thflags &= ~TH_URG; 2326 th->th_urp = 0; 2327 } 2328 } 2329 2330 /* 2331 * If new data are received on a connection after the 2332 * user processes are gone, then RST the other end. 2333 */ 2334 if ((tp->t_flags & TF_CLOSED) && tlen) { 2335 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2336 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2337 "after socket was closed, " 2338 "sending RST and removing tcpcb\n", 2339 s, __func__, tcpstates[tp->t_state], tlen); 2340 free(s, M_TCPLOG); 2341 } 2342 tp = tcp_close(tp); 2343 TCPSTAT_INC(tcps_rcvafterclose); 2344 rstreason = BANDLIM_UNLIMITED; 2345 goto dropwithreset; 2346 } 2347 2348 /* 2349 * If segment ends after window, drop trailing data 2350 * (and PUSH and FIN); if nothing left, just ACK. 2351 */ 2352 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2353 if (todrop > 0) { 2354 TCPSTAT_INC(tcps_rcvpackafterwin); 2355 if (todrop >= tlen) { 2356 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2357 /* 2358 * If window is closed can only take segments at 2359 * window edge, and have to drop data and PUSH from 2360 * incoming segments. Continue processing, but 2361 * remember to ack. Otherwise, drop segment 2362 * and ack. 2363 */ 2364 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2365 tp->t_flags |= TF_ACKNOW; 2366 TCPSTAT_INC(tcps_rcvwinprobe); 2367 } else 2368 goto dropafterack; 2369 } else 2370 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2371 m_adj(m, -todrop); 2372 tlen -= todrop; 2373 thflags &= ~(TH_PUSH|TH_FIN); 2374 } 2375 2376 /* 2377 * If last ACK falls within this segment's sequence numbers, 2378 * record its timestamp. 2379 * NOTE: 2380 * 1) That the test incorporates suggestions from the latest 2381 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2382 * 2) That updating only on newer timestamps interferes with 2383 * our earlier PAWS tests, so this check should be solely 2384 * predicated on the sequence space of this segment. 2385 * 3) That we modify the segment boundary check to be 2386 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2387 * instead of RFC1323's 2388 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2389 * This modified check allows us to overcome RFC1323's 2390 * limitations as described in Stevens TCP/IP Illustrated 2391 * Vol. 2 p.869. In such cases, we can still calculate the 2392 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2393 */ 2394 if ((to.to_flags & TOF_TS) != 0 && 2395 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2396 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2397 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2398 tp->ts_recent_age = tcp_ts_getticks(); 2399 tp->ts_recent = to.to_tsval; 2400 } 2401 2402 /* 2403 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2404 * flag is on (half-synchronized state), then queue data for 2405 * later processing; else drop segment and return. 2406 */ 2407 if ((thflags & TH_ACK) == 0) { 2408 if (tp->t_state == TCPS_SYN_RECEIVED || 2409 (tp->t_flags & TF_NEEDSYN)) { 2410 if (tp->t_state == TCPS_SYN_RECEIVED && 2411 IS_FASTOPEN(tp->t_flags)) { 2412 tp->snd_wnd = tiwin; 2413 cc_conn_init(tp); 2414 } 2415 goto step6; 2416 } else if (tp->t_flags & TF_ACKNOW) 2417 goto dropafterack; 2418 else 2419 goto drop; 2420 } 2421 2422 /* 2423 * Ack processing. 2424 */ 2425 switch (tp->t_state) { 2426 /* 2427 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2428 * ESTABLISHED state and continue processing. 2429 * The ACK was checked above. 2430 */ 2431 case TCPS_SYN_RECEIVED: 2432 2433 TCPSTAT_INC(tcps_connects); 2434 soisconnected(so); 2435 /* Do window scaling? */ 2436 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2437 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2438 tp->rcv_scale = tp->request_r_scale; 2439 } 2440 tp->snd_wnd = tiwin; 2441 /* 2442 * Make transitions: 2443 * SYN-RECEIVED -> ESTABLISHED 2444 * SYN-RECEIVED* -> FIN-WAIT-1 2445 */ 2446 tp->t_starttime = ticks; 2447 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2448 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2449 tp->t_tfo_pending = NULL; 2450 } 2451 if (tp->t_flags & TF_NEEDFIN) { 2452 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2453 tp->t_flags &= ~TF_NEEDFIN; 2454 } else { 2455 tcp_state_change(tp, TCPS_ESTABLISHED); 2456 TCP_PROBE5(accept__established, NULL, tp, 2457 m, tp, th); 2458 /* 2459 * TFO connections call cc_conn_init() during SYN 2460 * processing. Calling it again here for such 2461 * connections is not harmless as it would undo the 2462 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2463 * is retransmitted. 2464 */ 2465 if (!IS_FASTOPEN(tp->t_flags)) 2466 cc_conn_init(tp); 2467 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2468 } 2469 /* 2470 * Account for the ACK of our SYN prior to 2471 * regular ACK processing below, except for 2472 * simultaneous SYN, which is handled later. 2473 */ 2474 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2475 incforsyn = 1; 2476 /* 2477 * If segment contains data or ACK, will call tcp_reass() 2478 * later; if not, do so now to pass queued data to user. 2479 */ 2480 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2481 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2482 (struct mbuf *)0); 2483 tcp_handle_wakeup(tp, so); 2484 } 2485 tp->snd_wl1 = th->th_seq - 1; 2486 /* FALLTHROUGH */ 2487 2488 /* 2489 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2490 * ACKs. If the ack is in the range 2491 * tp->snd_una < th->th_ack <= tp->snd_max 2492 * then advance tp->snd_una to th->th_ack and drop 2493 * data from the retransmission queue. If this ACK reflects 2494 * more up to date window information we update our window information. 2495 */ 2496 case TCPS_ESTABLISHED: 2497 case TCPS_FIN_WAIT_1: 2498 case TCPS_FIN_WAIT_2: 2499 case TCPS_CLOSE_WAIT: 2500 case TCPS_CLOSING: 2501 case TCPS_LAST_ACK: 2502 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2503 TCPSTAT_INC(tcps_rcvacktoomuch); 2504 goto dropafterack; 2505 } 2506 if ((tp->t_flags & TF_SACK_PERMIT) && 2507 ((to.to_flags & TOF_SACK) || 2508 !TAILQ_EMPTY(&tp->snd_holes))) { 2509 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && 2510 (tp->t_flags & TF_LRD)) { 2511 tcp_sack_lost_retransmission(tp, th); 2512 } 2513 } else 2514 /* 2515 * Reset the value so that previous (valid) value 2516 * from the last ack with SACK doesn't get used. 2517 */ 2518 tp->sackhint.sacked_bytes = 0; 2519 2520 #ifdef TCP_HHOOK 2521 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2522 hhook_run_tcp_est_in(tp, th, &to); 2523 #endif 2524 2525 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2526 maxseg = tcp_maxseg(tp); 2527 if (tlen == 0 && 2528 (tiwin == tp->snd_wnd || 2529 (tp->t_flags & TF_SACK_PERMIT))) { 2530 /* 2531 * If this is the first time we've seen a 2532 * FIN from the remote, this is not a 2533 * duplicate and it needs to be processed 2534 * normally. This happens during a 2535 * simultaneous close. 2536 */ 2537 if ((thflags & TH_FIN) && 2538 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2539 tp->t_dupacks = 0; 2540 break; 2541 } 2542 TCPSTAT_INC(tcps_rcvdupack); 2543 /* 2544 * If we have outstanding data (other than 2545 * a window probe), this is a completely 2546 * duplicate ack (ie, window info didn't 2547 * change and FIN isn't set), 2548 * the ack is the biggest we've 2549 * seen and we've seen exactly our rexmt 2550 * threshold of them, assume a packet 2551 * has been dropped and retransmit it. 2552 * Kludge snd_nxt & the congestion 2553 * window so we send only this one 2554 * packet. 2555 * 2556 * We know we're losing at the current 2557 * window size so do congestion avoidance 2558 * (set ssthresh to half the current window 2559 * and pull our congestion window back to 2560 * the new ssthresh). 2561 * 2562 * Dup acks mean that packets have left the 2563 * network (they're now cached at the receiver) 2564 * so bump cwnd by the amount in the receiver 2565 * to keep a constant cwnd packets in the 2566 * network. 2567 * 2568 * When using TCP ECN, notify the peer that 2569 * we reduced the cwnd. 2570 */ 2571 /* 2572 * Following 2 kinds of acks should not affect 2573 * dupack counting: 2574 * 1) Old acks 2575 * 2) Acks with SACK but without any new SACK 2576 * information in them. These could result from 2577 * any anomaly in the network like a switch 2578 * duplicating packets or a possible DoS attack. 2579 */ 2580 if (th->th_ack != tp->snd_una || 2581 ((tp->t_flags & TF_SACK_PERMIT) && 2582 (to.to_flags & TOF_SACK) && 2583 !sack_changed)) 2584 break; 2585 else if (!tcp_timer_active(tp, TT_REXMT)) 2586 tp->t_dupacks = 0; 2587 else if (++tp->t_dupacks > tcprexmtthresh || 2588 IN_FASTRECOVERY(tp->t_flags)) { 2589 cc_ack_received(tp, th, nsegs, 2590 CC_DUPACK); 2591 if (V_tcp_do_prr && 2592 IN_FASTRECOVERY(tp->t_flags)) { 2593 tcp_do_prr_ack(tp, th, &to); 2594 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2595 (to.to_flags & TOF_SACK) && 2596 IN_FASTRECOVERY(tp->t_flags)) { 2597 int awnd; 2598 2599 /* 2600 * Compute the amount of data in flight first. 2601 * We can inject new data into the pipe iff 2602 * we have less than 1/2 the original window's 2603 * worth of data in flight. 2604 */ 2605 if (V_tcp_do_newsack) 2606 awnd = tcp_compute_pipe(tp); 2607 else 2608 awnd = (tp->snd_nxt - tp->snd_fack) + 2609 tp->sackhint.sack_bytes_rexmit; 2610 2611 if (awnd < tp->snd_ssthresh) { 2612 tp->snd_cwnd += maxseg; 2613 if (tp->snd_cwnd > tp->snd_ssthresh) 2614 tp->snd_cwnd = tp->snd_ssthresh; 2615 } 2616 } else 2617 tp->snd_cwnd += maxseg; 2618 (void) tcp_output(tp); 2619 goto drop; 2620 } else if (tp->t_dupacks == tcprexmtthresh || 2621 (tp->t_flags & TF_SACK_PERMIT && 2622 V_tcp_do_newsack && 2623 tp->sackhint.sacked_bytes > 2624 (tcprexmtthresh - 1) * maxseg)) { 2625 enter_recovery: 2626 /* 2627 * Above is the RFC6675 trigger condition of 2628 * more than (dupthresh-1)*maxseg sacked data. 2629 * If the count of holes in the 2630 * scoreboard is >= dupthresh, we could 2631 * also enter loss recovery, but don't 2632 * have that value readily available. 2633 */ 2634 tp->t_dupacks = tcprexmtthresh; 2635 tcp_seq onxt = tp->snd_nxt; 2636 2637 /* 2638 * If we're doing sack, or prr, check 2639 * to see if we're already in sack 2640 * recovery. If we're not doing sack, 2641 * check to see if we're in newreno 2642 * recovery. 2643 */ 2644 if (V_tcp_do_prr || 2645 (tp->t_flags & TF_SACK_PERMIT)) { 2646 if (IN_FASTRECOVERY(tp->t_flags)) { 2647 tp->t_dupacks = 0; 2648 break; 2649 } 2650 } else { 2651 if (SEQ_LEQ(th->th_ack, 2652 tp->snd_recover)) { 2653 tp->t_dupacks = 0; 2654 break; 2655 } 2656 } 2657 /* Congestion signal before ack. */ 2658 cc_cong_signal(tp, th, CC_NDUPACK); 2659 cc_ack_received(tp, th, nsegs, 2660 CC_DUPACK); 2661 tcp_timer_activate(tp, TT_REXMT, 0); 2662 tp->t_rtttime = 0; 2663 if (V_tcp_do_prr) { 2664 /* 2665 * snd_ssthresh is already updated by 2666 * cc_cong_signal. 2667 */ 2668 if ((tp->t_flags & TF_SACK_PERMIT) && 2669 (to.to_flags & TOF_SACK)) { 2670 tp->sackhint.prr_delivered = 2671 tp->sackhint.sacked_bytes; 2672 } else { 2673 tp->sackhint.prr_delivered = 2674 imin(tp->snd_max - tp->snd_una, 2675 imin(INT_MAX / 65536, 2676 tp->t_dupacks) * maxseg); 2677 } 2678 tp->sackhint.recover_fs = max(1, 2679 tp->snd_nxt - tp->snd_una); 2680 } 2681 if ((tp->t_flags & TF_SACK_PERMIT) && 2682 (to.to_flags & TOF_SACK)) { 2683 TCPSTAT_INC( 2684 tcps_sack_recovery_episode); 2685 tp->snd_recover = tp->snd_nxt; 2686 tp->snd_cwnd = maxseg; 2687 (void) tcp_output(tp); 2688 if (SEQ_GT(th->th_ack, tp->snd_una)) 2689 goto resume_partialack; 2690 goto drop; 2691 } 2692 tp->snd_nxt = th->th_ack; 2693 tp->snd_cwnd = maxseg; 2694 (void) tcp_output(tp); 2695 KASSERT(tp->snd_limited <= 2, 2696 ("%s: tp->snd_limited too big", 2697 __func__)); 2698 tp->snd_cwnd = tp->snd_ssthresh + 2699 maxseg * 2700 (tp->t_dupacks - tp->snd_limited); 2701 if (SEQ_GT(onxt, tp->snd_nxt)) 2702 tp->snd_nxt = onxt; 2703 goto drop; 2704 } else if (V_tcp_do_rfc3042) { 2705 /* 2706 * Process first and second duplicate 2707 * ACKs. Each indicates a segment 2708 * leaving the network, creating room 2709 * for more. Make sure we can send a 2710 * packet on reception of each duplicate 2711 * ACK by increasing snd_cwnd by one 2712 * segment. Restore the original 2713 * snd_cwnd after packet transmission. 2714 */ 2715 cc_ack_received(tp, th, nsegs, 2716 CC_DUPACK); 2717 uint32_t oldcwnd = tp->snd_cwnd; 2718 tcp_seq oldsndmax = tp->snd_max; 2719 u_int sent; 2720 int avail; 2721 2722 KASSERT(tp->t_dupacks == 1 || 2723 tp->t_dupacks == 2, 2724 ("%s: dupacks not 1 or 2", 2725 __func__)); 2726 if (tp->t_dupacks == 1) 2727 tp->snd_limited = 0; 2728 tp->snd_cwnd = 2729 (tp->snd_nxt - tp->snd_una) + 2730 (tp->t_dupacks - tp->snd_limited) * 2731 maxseg; 2732 /* 2733 * Only call tcp_output when there 2734 * is new data available to be sent. 2735 * Otherwise we would send pure ACKs. 2736 */ 2737 SOCKBUF_LOCK(&so->so_snd); 2738 avail = sbavail(&so->so_snd) - 2739 (tp->snd_nxt - tp->snd_una); 2740 SOCKBUF_UNLOCK(&so->so_snd); 2741 if (avail > 0) 2742 (void) tcp_output(tp); 2743 sent = tp->snd_max - oldsndmax; 2744 if (sent > maxseg) { 2745 KASSERT((tp->t_dupacks == 2 && 2746 tp->snd_limited == 0) || 2747 (sent == maxseg + 1 && 2748 tp->t_flags & TF_SENTFIN), 2749 ("%s: sent too much", 2750 __func__)); 2751 tp->snd_limited = 2; 2752 } else if (sent > 0) 2753 ++tp->snd_limited; 2754 tp->snd_cwnd = oldcwnd; 2755 goto drop; 2756 } 2757 } 2758 break; 2759 } else { 2760 /* 2761 * This ack is advancing the left edge, reset the 2762 * counter. 2763 */ 2764 tp->t_dupacks = 0; 2765 /* 2766 * If this ack also has new SACK info, increment the 2767 * counter as per rfc6675. The variable 2768 * sack_changed tracks all changes to the SACK 2769 * scoreboard, including when partial ACKs without 2770 * SACK options are received, and clear the scoreboard 2771 * from the left side. Such partial ACKs should not be 2772 * counted as dupacks here. 2773 */ 2774 if ((tp->t_flags & TF_SACK_PERMIT) && 2775 (to.to_flags & TOF_SACK) && 2776 sack_changed) { 2777 tp->t_dupacks++; 2778 /* limit overhead by setting maxseg last */ 2779 if (!IN_FASTRECOVERY(tp->t_flags) && 2780 (tp->sackhint.sacked_bytes > 2781 ((tcprexmtthresh - 1) * 2782 (maxseg = tcp_maxseg(tp))))) { 2783 goto enter_recovery; 2784 } 2785 } 2786 } 2787 2788 resume_partialack: 2789 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2790 ("%s: th_ack <= snd_una", __func__)); 2791 2792 /* 2793 * If the congestion window was inflated to account 2794 * for the other side's cached packets, retract it. 2795 */ 2796 if (IN_FASTRECOVERY(tp->t_flags)) { 2797 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2798 if (tp->t_flags & TF_SACK_PERMIT) 2799 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2800 tcp_timer_activate(tp, TT_REXMT, 0); 2801 tp->t_rtttime = 0; 2802 tcp_do_prr_ack(tp, th, &to); 2803 tp->t_flags |= TF_ACKNOW; 2804 (void) tcp_output(tp); 2805 } else 2806 tcp_sack_partialack(tp, th); 2807 else 2808 tcp_newreno_partial_ack(tp, th); 2809 } else 2810 cc_post_recovery(tp, th); 2811 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2812 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2813 if (V_tcp_do_prr) { 2814 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2815 tp->snd_fack = th->th_ack; 2816 tcp_do_prr_ack(tp, th, &to); 2817 (void) tcp_output(tp); 2818 } 2819 } else 2820 cc_post_recovery(tp, th); 2821 } 2822 /* 2823 * If we reach this point, ACK is not a duplicate, 2824 * i.e., it ACKs something we sent. 2825 */ 2826 if (tp->t_flags & TF_NEEDSYN) { 2827 /* 2828 * T/TCP: Connection was half-synchronized, and our 2829 * SYN has been ACK'd (so connection is now fully 2830 * synchronized). Go to non-starred state, 2831 * increment snd_una for ACK of SYN, and check if 2832 * we can do window scaling. 2833 */ 2834 tp->t_flags &= ~TF_NEEDSYN; 2835 tp->snd_una++; 2836 /* Do window scaling? */ 2837 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2838 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2839 tp->rcv_scale = tp->request_r_scale; 2840 /* Send window already scaled. */ 2841 } 2842 } 2843 2844 process_ACK: 2845 INP_WLOCK_ASSERT(tp->t_inpcb); 2846 2847 /* 2848 * Adjust for the SYN bit in sequence space, 2849 * but don't account for it in cwnd calculations. 2850 * This is for the SYN_RECEIVED, non-simultaneous 2851 * SYN case. SYN_SENT and simultaneous SYN are 2852 * treated elsewhere. 2853 */ 2854 if (incforsyn) 2855 tp->snd_una++; 2856 acked = BYTES_THIS_ACK(tp, th); 2857 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2858 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2859 tp->snd_una, th->th_ack, tp, m)); 2860 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2861 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2862 2863 /* 2864 * If we just performed our first retransmit, and the ACK 2865 * arrives within our recovery window, then it was a mistake 2866 * to do the retransmit in the first place. Recover our 2867 * original cwnd and ssthresh, and proceed to transmit where 2868 * we left off. 2869 */ 2870 if (tp->t_rxtshift == 1 && 2871 tp->t_flags & TF_PREVVALID && 2872 tp->t_badrxtwin != 0 && 2873 to.to_flags & TOF_TS && 2874 to.to_tsecr != 0 && 2875 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2876 cc_cong_signal(tp, th, CC_RTO_ERR); 2877 2878 /* 2879 * If we have a timestamp reply, update smoothed 2880 * round trip time. If no timestamp is present but 2881 * transmit timer is running and timed sequence 2882 * number was acked, update smoothed round trip time. 2883 * Since we now have an rtt measurement, cancel the 2884 * timer backoff (cf., Phil Karn's retransmit alg.). 2885 * Recompute the initial retransmit timer. 2886 * 2887 * Some boxes send broken timestamp replies 2888 * during the SYN+ACK phase, ignore 2889 * timestamps of 0 or we could calculate a 2890 * huge RTT and blow up the retransmit timer. 2891 */ 2892 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2893 uint32_t t; 2894 2895 t = tcp_ts_getticks() - to.to_tsecr; 2896 if (!tp->t_rttlow || tp->t_rttlow > t) 2897 tp->t_rttlow = t; 2898 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2899 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2900 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2901 tp->t_rttlow = ticks - tp->t_rtttime; 2902 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2903 } 2904 2905 /* 2906 * If all outstanding data is acked, stop retransmit 2907 * timer and remember to restart (more output or persist). 2908 * If there is more data to be acked, restart retransmit 2909 * timer, using current (possibly backed-off) value. 2910 */ 2911 if (th->th_ack == tp->snd_max) { 2912 tcp_timer_activate(tp, TT_REXMT, 0); 2913 needoutput = 1; 2914 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2915 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2916 2917 /* 2918 * If no data (only SYN) was ACK'd, 2919 * skip rest of ACK processing. 2920 */ 2921 if (acked == 0) 2922 goto step6; 2923 2924 /* 2925 * Let the congestion control algorithm update congestion 2926 * control related information. This typically means increasing 2927 * the congestion window. 2928 */ 2929 cc_ack_received(tp, th, nsegs, CC_ACK); 2930 2931 SOCKBUF_LOCK(&so->so_snd); 2932 if (acked > sbavail(&so->so_snd)) { 2933 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2934 tp->snd_wnd -= sbavail(&so->so_snd); 2935 else 2936 tp->snd_wnd = 0; 2937 mfree = sbcut_locked(&so->so_snd, 2938 (int)sbavail(&so->so_snd)); 2939 ourfinisacked = 1; 2940 } else { 2941 mfree = sbcut_locked(&so->so_snd, acked); 2942 if (tp->snd_wnd >= (uint32_t) acked) 2943 tp->snd_wnd -= acked; 2944 else 2945 tp->snd_wnd = 0; 2946 ourfinisacked = 0; 2947 } 2948 /* NB: sowwakeup_locked() does an implicit unlock. */ 2949 sowwakeup_locked(so); 2950 m_freem(mfree); 2951 /* Detect una wraparound. */ 2952 if (!IN_RECOVERY(tp->t_flags) && 2953 SEQ_GT(tp->snd_una, tp->snd_recover) && 2954 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2955 tp->snd_recover = th->th_ack - 1; 2956 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2957 if (IN_RECOVERY(tp->t_flags) && 2958 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2959 EXIT_RECOVERY(tp->t_flags); 2960 } 2961 tp->snd_una = th->th_ack; 2962 if (tp->t_flags & TF_SACK_PERMIT) { 2963 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2964 tp->snd_recover = tp->snd_una; 2965 } 2966 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2967 tp->snd_nxt = tp->snd_una; 2968 2969 switch (tp->t_state) { 2970 /* 2971 * In FIN_WAIT_1 STATE in addition to the processing 2972 * for the ESTABLISHED state if our FIN is now acknowledged 2973 * then enter FIN_WAIT_2. 2974 */ 2975 case TCPS_FIN_WAIT_1: 2976 if (ourfinisacked) { 2977 /* 2978 * If we can't receive any more 2979 * data, then closing user can proceed. 2980 * Starting the timer is contrary to the 2981 * specification, but if we don't get a FIN 2982 * we'll hang forever. 2983 * 2984 * XXXjl: 2985 * we should release the tp also, and use a 2986 * compressed state. 2987 */ 2988 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2989 soisdisconnected(so); 2990 tcp_timer_activate(tp, TT_2MSL, 2991 (tcp_fast_finwait2_recycle ? 2992 tcp_finwait2_timeout : 2993 TP_MAXIDLE(tp))); 2994 } 2995 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2996 } 2997 break; 2998 2999 /* 3000 * In CLOSING STATE in addition to the processing for 3001 * the ESTABLISHED state if the ACK acknowledges our FIN 3002 * then enter the TIME-WAIT state, otherwise ignore 3003 * the segment. 3004 */ 3005 case TCPS_CLOSING: 3006 if (ourfinisacked) { 3007 tcp_twstart(tp); 3008 m_freem(m); 3009 return; 3010 } 3011 break; 3012 3013 /* 3014 * In LAST_ACK, we may still be waiting for data to drain 3015 * and/or to be acked, as well as for the ack of our FIN. 3016 * If our FIN is now acknowledged, delete the TCB, 3017 * enter the closed state and return. 3018 */ 3019 case TCPS_LAST_ACK: 3020 if (ourfinisacked) { 3021 tp = tcp_close(tp); 3022 goto drop; 3023 } 3024 break; 3025 } 3026 } 3027 3028 step6: 3029 INP_WLOCK_ASSERT(tp->t_inpcb); 3030 3031 /* 3032 * Update window information. 3033 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3034 */ 3035 if ((thflags & TH_ACK) && 3036 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3037 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3038 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3039 /* keep track of pure window updates */ 3040 if (tlen == 0 && 3041 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3042 TCPSTAT_INC(tcps_rcvwinupd); 3043 tp->snd_wnd = tiwin; 3044 tp->snd_wl1 = th->th_seq; 3045 tp->snd_wl2 = th->th_ack; 3046 if (tp->snd_wnd > tp->max_sndwnd) 3047 tp->max_sndwnd = tp->snd_wnd; 3048 needoutput = 1; 3049 } 3050 3051 /* 3052 * Process segments with URG. 3053 */ 3054 if ((thflags & TH_URG) && th->th_urp && 3055 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3056 /* 3057 * This is a kludge, but if we receive and accept 3058 * random urgent pointers, we'll crash in 3059 * soreceive. It's hard to imagine someone 3060 * actually wanting to send this much urgent data. 3061 */ 3062 SOCKBUF_LOCK(&so->so_rcv); 3063 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3064 th->th_urp = 0; /* XXX */ 3065 thflags &= ~TH_URG; /* XXX */ 3066 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3067 goto dodata; /* XXX */ 3068 } 3069 /* 3070 * If this segment advances the known urgent pointer, 3071 * then mark the data stream. This should not happen 3072 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3073 * a FIN has been received from the remote side. 3074 * In these states we ignore the URG. 3075 * 3076 * According to RFC961 (Assigned Protocols), 3077 * the urgent pointer points to the last octet 3078 * of urgent data. We continue, however, 3079 * to consider it to indicate the first octet 3080 * of data past the urgent section as the original 3081 * spec states (in one of two places). 3082 */ 3083 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3084 tp->rcv_up = th->th_seq + th->th_urp; 3085 so->so_oobmark = sbavail(&so->so_rcv) + 3086 (tp->rcv_up - tp->rcv_nxt) - 1; 3087 if (so->so_oobmark == 0) 3088 so->so_rcv.sb_state |= SBS_RCVATMARK; 3089 sohasoutofband(so); 3090 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3091 } 3092 SOCKBUF_UNLOCK(&so->so_rcv); 3093 /* 3094 * Remove out of band data so doesn't get presented to user. 3095 * This can happen independent of advancing the URG pointer, 3096 * but if two URG's are pending at once, some out-of-band 3097 * data may creep in... ick. 3098 */ 3099 if (th->th_urp <= (uint32_t)tlen && 3100 !(so->so_options & SO_OOBINLINE)) { 3101 /* hdr drop is delayed */ 3102 tcp_pulloutofband(so, th, m, drop_hdrlen); 3103 } 3104 } else { 3105 /* 3106 * If no out of band data is expected, 3107 * pull receive urgent pointer along 3108 * with the receive window. 3109 */ 3110 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3111 tp->rcv_up = tp->rcv_nxt; 3112 } 3113 dodata: /* XXX */ 3114 INP_WLOCK_ASSERT(tp->t_inpcb); 3115 3116 /* 3117 * Process the segment text, merging it into the TCP sequencing queue, 3118 * and arranging for acknowledgment of receipt if necessary. 3119 * This process logically involves adjusting tp->rcv_wnd as data 3120 * is presented to the user (this happens in tcp_usrreq.c, 3121 * case PRU_RCVD). If a FIN has already been received on this 3122 * connection then we just ignore the text. 3123 */ 3124 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3125 IS_FASTOPEN(tp->t_flags)); 3126 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3127 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3128 tcp_seq save_start = th->th_seq; 3129 tcp_seq save_rnxt = tp->rcv_nxt; 3130 int save_tlen = tlen; 3131 m_adj(m, drop_hdrlen); /* delayed header drop */ 3132 /* 3133 * Insert segment which includes th into TCP reassembly queue 3134 * with control block tp. Set thflags to whether reassembly now 3135 * includes a segment with FIN. This handles the common case 3136 * inline (segment is the next to be received on an established 3137 * connection, and the queue is empty), avoiding linkage into 3138 * and removal from the queue and repetition of various 3139 * conversions. 3140 * Set DELACK for segments received in order, but ack 3141 * immediately when segments are out of order (so 3142 * fast retransmit can work). 3143 */ 3144 if (th->th_seq == tp->rcv_nxt && 3145 SEGQ_EMPTY(tp) && 3146 (TCPS_HAVEESTABLISHED(tp->t_state) || 3147 tfo_syn)) { 3148 if (DELAY_ACK(tp, tlen) || tfo_syn) 3149 tp->t_flags |= TF_DELACK; 3150 else 3151 tp->t_flags |= TF_ACKNOW; 3152 tp->rcv_nxt += tlen; 3153 if (tlen && 3154 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3155 (tp->t_fbyte_in == 0)) { 3156 tp->t_fbyte_in = ticks; 3157 if (tp->t_fbyte_in == 0) 3158 tp->t_fbyte_in = 1; 3159 if (tp->t_fbyte_out && tp->t_fbyte_in) 3160 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3161 } 3162 thflags = tcp_get_flags(th) & TH_FIN; 3163 TCPSTAT_INC(tcps_rcvpack); 3164 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3165 SOCKBUF_LOCK(&so->so_rcv); 3166 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3167 m_freem(m); 3168 else 3169 sbappendstream_locked(&so->so_rcv, m, 0); 3170 tp->t_flags |= TF_WAKESOR; 3171 } else { 3172 /* 3173 * XXX: Due to the header drop above "th" is 3174 * theoretically invalid by now. Fortunately 3175 * m_adj() doesn't actually frees any mbufs 3176 * when trimming from the head. 3177 */ 3178 tcp_seq temp = save_start; 3179 3180 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3181 tp->t_flags |= TF_ACKNOW; 3182 } 3183 if ((tp->t_flags & TF_SACK_PERMIT) && 3184 (save_tlen > 0) && 3185 TCPS_HAVEESTABLISHED(tp->t_state)) { 3186 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3187 /* 3188 * DSACK actually handled in the fastpath 3189 * above. 3190 */ 3191 tcp_update_sack_list(tp, save_start, 3192 save_start + save_tlen); 3193 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3194 if ((tp->rcv_numsacks >= 1) && 3195 (tp->sackblks[0].end == save_start)) { 3196 /* 3197 * Partial overlap, recorded at todrop 3198 * above. 3199 */ 3200 tcp_update_sack_list(tp, 3201 tp->sackblks[0].start, 3202 tp->sackblks[0].end); 3203 } else { 3204 tcp_update_dsack_list(tp, save_start, 3205 save_start + save_tlen); 3206 } 3207 } else if (tlen >= save_tlen) { 3208 /* Update of sackblks. */ 3209 tcp_update_dsack_list(tp, save_start, 3210 save_start + save_tlen); 3211 } else if (tlen > 0) { 3212 tcp_update_dsack_list(tp, save_start, 3213 save_start + tlen); 3214 } 3215 } 3216 tcp_handle_wakeup(tp, so); 3217 #if 0 3218 /* 3219 * Note the amount of data that peer has sent into 3220 * our window, in order to estimate the sender's 3221 * buffer size. 3222 * XXX: Unused. 3223 */ 3224 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3225 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3226 else 3227 len = so->so_rcv.sb_hiwat; 3228 #endif 3229 } else { 3230 m_freem(m); 3231 thflags &= ~TH_FIN; 3232 } 3233 3234 /* 3235 * If FIN is received ACK the FIN and let the user know 3236 * that the connection is closing. 3237 */ 3238 if (thflags & TH_FIN) { 3239 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3240 /* The socket upcall is handled by socantrcvmore. */ 3241 socantrcvmore(so); 3242 /* 3243 * If connection is half-synchronized 3244 * (ie NEEDSYN flag on) then delay ACK, 3245 * so it may be piggybacked when SYN is sent. 3246 * Otherwise, since we received a FIN then no 3247 * more input can be expected, send ACK now. 3248 */ 3249 if (tp->t_flags & TF_NEEDSYN) 3250 tp->t_flags |= TF_DELACK; 3251 else 3252 tp->t_flags |= TF_ACKNOW; 3253 tp->rcv_nxt++; 3254 } 3255 switch (tp->t_state) { 3256 /* 3257 * In SYN_RECEIVED and ESTABLISHED STATES 3258 * enter the CLOSE_WAIT state. 3259 */ 3260 case TCPS_SYN_RECEIVED: 3261 tp->t_starttime = ticks; 3262 /* FALLTHROUGH */ 3263 case TCPS_ESTABLISHED: 3264 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3265 break; 3266 3267 /* 3268 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3269 * enter the CLOSING state. 3270 */ 3271 case TCPS_FIN_WAIT_1: 3272 tcp_state_change(tp, TCPS_CLOSING); 3273 break; 3274 3275 /* 3276 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3277 * starting the time-wait timer, turning off the other 3278 * standard timers. 3279 */ 3280 case TCPS_FIN_WAIT_2: 3281 tcp_twstart(tp); 3282 return; 3283 } 3284 } 3285 #ifdef TCPDEBUG 3286 if (so->so_options & SO_DEBUG) 3287 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3288 &tcp_savetcp, 0); 3289 #endif 3290 TCP_PROBE3(debug__input, tp, th, m); 3291 3292 /* 3293 * Return any desired output. 3294 */ 3295 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3296 (void) tcp_output(tp); 3297 3298 check_delack: 3299 INP_WLOCK_ASSERT(tp->t_inpcb); 3300 3301 if (tp->t_flags & TF_DELACK) { 3302 tp->t_flags &= ~TF_DELACK; 3303 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3304 } 3305 INP_WUNLOCK(tp->t_inpcb); 3306 return; 3307 3308 dropafterack: 3309 /* 3310 * Generate an ACK dropping incoming segment if it occupies 3311 * sequence space, where the ACK reflects our state. 3312 * 3313 * We can now skip the test for the RST flag since all 3314 * paths to this code happen after packets containing 3315 * RST have been dropped. 3316 * 3317 * In the SYN-RECEIVED state, don't send an ACK unless the 3318 * segment we received passes the SYN-RECEIVED ACK test. 3319 * If it fails send a RST. This breaks the loop in the 3320 * "LAND" DoS attack, and also prevents an ACK storm 3321 * between two listening ports that have been sent forged 3322 * SYN segments, each with the source address of the other. 3323 */ 3324 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3325 (SEQ_GT(tp->snd_una, th->th_ack) || 3326 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3327 rstreason = BANDLIM_RST_OPENPORT; 3328 goto dropwithreset; 3329 } 3330 #ifdef TCPDEBUG 3331 if (so->so_options & SO_DEBUG) 3332 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3333 &tcp_savetcp, 0); 3334 #endif 3335 TCP_PROBE3(debug__input, tp, th, m); 3336 tp->t_flags |= TF_ACKNOW; 3337 (void) tcp_output(tp); 3338 INP_WUNLOCK(tp->t_inpcb); 3339 m_freem(m); 3340 return; 3341 3342 dropwithreset: 3343 if (tp != NULL) { 3344 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3345 INP_WUNLOCK(tp->t_inpcb); 3346 } else 3347 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3348 return; 3349 3350 drop: 3351 /* 3352 * Drop space held by incoming segment and return. 3353 */ 3354 #ifdef TCPDEBUG 3355 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3356 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3357 &tcp_savetcp, 0); 3358 #endif 3359 TCP_PROBE3(debug__input, tp, th, m); 3360 if (tp != NULL) { 3361 INP_WUNLOCK(tp->t_inpcb); 3362 } 3363 m_freem(m); 3364 } 3365 3366 /* 3367 * Issue RST and make ACK acceptable to originator of segment. 3368 * The mbuf must still include the original packet header. 3369 * tp may be NULL. 3370 */ 3371 void 3372 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3373 int tlen, int rstreason) 3374 { 3375 #ifdef INET 3376 struct ip *ip; 3377 #endif 3378 #ifdef INET6 3379 struct ip6_hdr *ip6; 3380 #endif 3381 3382 if (tp != NULL) { 3383 INP_LOCK_ASSERT(tp->t_inpcb); 3384 } 3385 3386 /* Don't bother if destination was broadcast/multicast. */ 3387 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3388 goto drop; 3389 #ifdef INET6 3390 if (mtod(m, struct ip *)->ip_v == 6) { 3391 ip6 = mtod(m, struct ip6_hdr *); 3392 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3393 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3394 goto drop; 3395 /* IPv6 anycast check is done at tcp6_input() */ 3396 } 3397 #endif 3398 #if defined(INET) && defined(INET6) 3399 else 3400 #endif 3401 #ifdef INET 3402 { 3403 ip = mtod(m, struct ip *); 3404 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3405 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3406 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3407 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3408 goto drop; 3409 } 3410 #endif 3411 3412 /* Perform bandwidth limiting. */ 3413 if (badport_bandlim(rstreason) < 0) 3414 goto drop; 3415 3416 /* tcp_respond consumes the mbuf chain. */ 3417 if (tcp_get_flags(th) & TH_ACK) { 3418 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3419 th->th_ack, TH_RST); 3420 } else { 3421 if (tcp_get_flags(th) & TH_SYN) 3422 tlen++; 3423 if (tcp_get_flags(th) & TH_FIN) 3424 tlen++; 3425 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3426 (tcp_seq)0, TH_RST|TH_ACK); 3427 } 3428 return; 3429 drop: 3430 m_freem(m); 3431 } 3432 3433 /* 3434 * Parse TCP options and place in tcpopt. 3435 */ 3436 void 3437 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3438 { 3439 int opt, optlen; 3440 3441 to->to_flags = 0; 3442 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3443 opt = cp[0]; 3444 if (opt == TCPOPT_EOL) 3445 break; 3446 if (opt == TCPOPT_NOP) 3447 optlen = 1; 3448 else { 3449 if (cnt < 2) 3450 break; 3451 optlen = cp[1]; 3452 if (optlen < 2 || optlen > cnt) 3453 break; 3454 } 3455 switch (opt) { 3456 case TCPOPT_MAXSEG: 3457 if (optlen != TCPOLEN_MAXSEG) 3458 continue; 3459 if (!(flags & TO_SYN)) 3460 continue; 3461 to->to_flags |= TOF_MSS; 3462 bcopy((char *)cp + 2, 3463 (char *)&to->to_mss, sizeof(to->to_mss)); 3464 to->to_mss = ntohs(to->to_mss); 3465 break; 3466 case TCPOPT_WINDOW: 3467 if (optlen != TCPOLEN_WINDOW) 3468 continue; 3469 if (!(flags & TO_SYN)) 3470 continue; 3471 to->to_flags |= TOF_SCALE; 3472 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3473 break; 3474 case TCPOPT_TIMESTAMP: 3475 if (optlen != TCPOLEN_TIMESTAMP) 3476 continue; 3477 to->to_flags |= TOF_TS; 3478 bcopy((char *)cp + 2, 3479 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3480 to->to_tsval = ntohl(to->to_tsval); 3481 bcopy((char *)cp + 6, 3482 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3483 to->to_tsecr = ntohl(to->to_tsecr); 3484 break; 3485 case TCPOPT_SIGNATURE: 3486 /* 3487 * In order to reply to a host which has set the 3488 * TCP_SIGNATURE option in its initial SYN, we have 3489 * to record the fact that the option was observed 3490 * here for the syncache code to perform the correct 3491 * response. 3492 */ 3493 if (optlen != TCPOLEN_SIGNATURE) 3494 continue; 3495 to->to_flags |= TOF_SIGNATURE; 3496 to->to_signature = cp + 2; 3497 break; 3498 case TCPOPT_SACK_PERMITTED: 3499 if (optlen != TCPOLEN_SACK_PERMITTED) 3500 continue; 3501 if (!(flags & TO_SYN)) 3502 continue; 3503 if (!V_tcp_do_sack) 3504 continue; 3505 to->to_flags |= TOF_SACKPERM; 3506 break; 3507 case TCPOPT_SACK: 3508 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3509 continue; 3510 if (flags & TO_SYN) 3511 continue; 3512 to->to_flags |= TOF_SACK; 3513 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3514 to->to_sacks = cp + 2; 3515 TCPSTAT_INC(tcps_sack_rcv_blocks); 3516 break; 3517 case TCPOPT_FAST_OPEN: 3518 /* 3519 * Cookie length validation is performed by the 3520 * server side cookie checking code or the client 3521 * side cookie cache update code. 3522 */ 3523 if (!(flags & TO_SYN)) 3524 continue; 3525 if (!V_tcp_fastopen_client_enable && 3526 !V_tcp_fastopen_server_enable) 3527 continue; 3528 to->to_flags |= TOF_FASTOPEN; 3529 to->to_tfo_len = optlen - 2; 3530 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3531 break; 3532 default: 3533 continue; 3534 } 3535 } 3536 } 3537 3538 /* 3539 * Pull out of band byte out of a segment so 3540 * it doesn't appear in the user's data queue. 3541 * It is still reflected in the segment length for 3542 * sequencing purposes. 3543 */ 3544 void 3545 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3546 int off) 3547 { 3548 int cnt = off + th->th_urp - 1; 3549 3550 while (cnt >= 0) { 3551 if (m->m_len > cnt) { 3552 char *cp = mtod(m, caddr_t) + cnt; 3553 struct tcpcb *tp = sototcpcb(so); 3554 3555 INP_WLOCK_ASSERT(tp->t_inpcb); 3556 3557 tp->t_iobc = *cp; 3558 tp->t_oobflags |= TCPOOB_HAVEDATA; 3559 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3560 m->m_len--; 3561 if (m->m_flags & M_PKTHDR) 3562 m->m_pkthdr.len--; 3563 return; 3564 } 3565 cnt -= m->m_len; 3566 m = m->m_next; 3567 if (m == NULL) 3568 break; 3569 } 3570 panic("tcp_pulloutofband"); 3571 } 3572 3573 /* 3574 * Collect new round-trip time estimate 3575 * and update averages and current timeout. 3576 */ 3577 void 3578 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3579 { 3580 int delta; 3581 3582 INP_WLOCK_ASSERT(tp->t_inpcb); 3583 3584 TCPSTAT_INC(tcps_rttupdated); 3585 tp->t_rttupdated++; 3586 #ifdef STATS 3587 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3588 imax(0, rtt * 1000 / hz)); 3589 #endif 3590 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3591 /* 3592 * srtt is stored as fixed point with 5 bits after the 3593 * binary point (i.e., scaled by 8). The following magic 3594 * is equivalent to the smoothing algorithm in rfc793 with 3595 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3596 * point). Adjust rtt to origin 0. 3597 */ 3598 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3599 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3600 3601 if ((tp->t_srtt += delta) <= 0) 3602 tp->t_srtt = 1; 3603 3604 /* 3605 * We accumulate a smoothed rtt variance (actually, a 3606 * smoothed mean difference), then set the retransmit 3607 * timer to smoothed rtt + 4 times the smoothed variance. 3608 * rttvar is stored as fixed point with 4 bits after the 3609 * binary point (scaled by 16). The following is 3610 * equivalent to rfc793 smoothing with an alpha of .75 3611 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3612 * rfc793's wired-in beta. 3613 */ 3614 if (delta < 0) 3615 delta = -delta; 3616 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3617 if ((tp->t_rttvar += delta) <= 0) 3618 tp->t_rttvar = 1; 3619 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3620 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3621 } else { 3622 /* 3623 * No rtt measurement yet - use the unsmoothed rtt. 3624 * Set the variance to half the rtt (so our first 3625 * retransmit happens at 3*rtt). 3626 */ 3627 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3628 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3629 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3630 } 3631 tp->t_rtttime = 0; 3632 tp->t_rxtshift = 0; 3633 3634 /* 3635 * the retransmit should happen at rtt + 4 * rttvar. 3636 * Because of the way we do the smoothing, srtt and rttvar 3637 * will each average +1/2 tick of bias. When we compute 3638 * the retransmit timer, we want 1/2 tick of rounding and 3639 * 1 extra tick because of +-1/2 tick uncertainty in the 3640 * firing of the timer. The bias will give us exactly the 3641 * 1.5 tick we need. But, because the bias is 3642 * statistical, we have to test that we don't drop below 3643 * the minimum feasible timer (which is 2 ticks). 3644 */ 3645 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3646 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3647 3648 /* 3649 * We received an ack for a packet that wasn't retransmitted; 3650 * it is probably safe to discard any error indications we've 3651 * received recently. This isn't quite right, but close enough 3652 * for now (a route might have failed after we sent a segment, 3653 * and the return path might not be symmetrical). 3654 */ 3655 tp->t_softerror = 0; 3656 } 3657 3658 /* 3659 * Determine a reasonable value for maxseg size. 3660 * If the route is known, check route for mtu. 3661 * If none, use an mss that can be handled on the outgoing interface 3662 * without forcing IP to fragment. If no route is found, route has no mtu, 3663 * or the destination isn't local, use a default, hopefully conservative 3664 * size (usually 512 or the default IP max size, but no more than the mtu 3665 * of the interface), as we can't discover anything about intervening 3666 * gateways or networks. We also initialize the congestion/slow start 3667 * window to be a single segment if the destination isn't local. 3668 * While looking at the routing entry, we also initialize other path-dependent 3669 * parameters from pre-set or cached values in the routing entry. 3670 * 3671 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3672 * IP options, e.g. IPSEC data, since length of this data may vary, and 3673 * thus it is calculated for every segment separately in tcp_output(). 3674 * 3675 * NOTE that this routine is only called when we process an incoming 3676 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3677 * settings are handled in tcp_mssopt(). 3678 */ 3679 void 3680 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3681 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3682 { 3683 int mss = 0; 3684 uint32_t maxmtu = 0; 3685 struct inpcb *inp = tp->t_inpcb; 3686 struct hc_metrics_lite metrics; 3687 #ifdef INET6 3688 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3689 size_t min_protoh = isipv6 ? 3690 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3691 sizeof (struct tcpiphdr); 3692 #else 3693 size_t min_protoh = sizeof(struct tcpiphdr); 3694 #endif 3695 3696 INP_WLOCK_ASSERT(tp->t_inpcb); 3697 3698 if (tp->t_port) 3699 min_protoh += V_tcp_udp_tunneling_overhead; 3700 if (mtuoffer != -1) { 3701 KASSERT(offer == -1, ("%s: conflict", __func__)); 3702 offer = mtuoffer - min_protoh; 3703 } 3704 3705 /* Initialize. */ 3706 #ifdef INET6 3707 if (isipv6) { 3708 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3709 tp->t_maxseg = V_tcp_v6mssdflt; 3710 } 3711 #endif 3712 #if defined(INET) && defined(INET6) 3713 else 3714 #endif 3715 #ifdef INET 3716 { 3717 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3718 tp->t_maxseg = V_tcp_mssdflt; 3719 } 3720 #endif 3721 3722 /* 3723 * No route to sender, stay with default mss and return. 3724 */ 3725 if (maxmtu == 0) { 3726 /* 3727 * In case we return early we need to initialize metrics 3728 * to a defined state as tcp_hc_get() would do for us 3729 * if there was no cache hit. 3730 */ 3731 if (metricptr != NULL) 3732 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3733 return; 3734 } 3735 3736 /* What have we got? */ 3737 switch (offer) { 3738 case 0: 3739 /* 3740 * Offer == 0 means that there was no MSS on the SYN 3741 * segment, in this case we use tcp_mssdflt as 3742 * already assigned to t_maxseg above. 3743 */ 3744 offer = tp->t_maxseg; 3745 break; 3746 3747 case -1: 3748 /* 3749 * Offer == -1 means that we didn't receive SYN yet. 3750 */ 3751 /* FALLTHROUGH */ 3752 3753 default: 3754 /* 3755 * Prevent DoS attack with too small MSS. Round up 3756 * to at least minmss. 3757 */ 3758 offer = max(offer, V_tcp_minmss); 3759 } 3760 3761 /* 3762 * rmx information is now retrieved from tcp_hostcache. 3763 */ 3764 tcp_hc_get(&inp->inp_inc, &metrics); 3765 if (metricptr != NULL) 3766 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3767 3768 /* 3769 * If there's a discovered mtu in tcp hostcache, use it. 3770 * Else, use the link mtu. 3771 */ 3772 if (metrics.rmx_mtu) 3773 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3774 else { 3775 #ifdef INET6 3776 if (isipv6) { 3777 mss = maxmtu - min_protoh; 3778 if (!V_path_mtu_discovery && 3779 !in6_localaddr(&inp->in6p_faddr)) 3780 mss = min(mss, V_tcp_v6mssdflt); 3781 } 3782 #endif 3783 #if defined(INET) && defined(INET6) 3784 else 3785 #endif 3786 #ifdef INET 3787 { 3788 mss = maxmtu - min_protoh; 3789 if (!V_path_mtu_discovery && 3790 !in_localaddr(inp->inp_faddr)) 3791 mss = min(mss, V_tcp_mssdflt); 3792 } 3793 #endif 3794 /* 3795 * XXX - The above conditional (mss = maxmtu - min_protoh) 3796 * probably violates the TCP spec. 3797 * The problem is that, since we don't know the 3798 * other end's MSS, we are supposed to use a conservative 3799 * default. But, if we do that, then MTU discovery will 3800 * never actually take place, because the conservative 3801 * default is much less than the MTUs typically seen 3802 * on the Internet today. For the moment, we'll sweep 3803 * this under the carpet. 3804 * 3805 * The conservative default might not actually be a problem 3806 * if the only case this occurs is when sending an initial 3807 * SYN with options and data to a host we've never talked 3808 * to before. Then, they will reply with an MSS value which 3809 * will get recorded and the new parameters should get 3810 * recomputed. For Further Study. 3811 */ 3812 } 3813 mss = min(mss, offer); 3814 3815 /* 3816 * Sanity check: make sure that maxseg will be large 3817 * enough to allow some data on segments even if the 3818 * all the option space is used (40bytes). Otherwise 3819 * funny things may happen in tcp_output. 3820 * 3821 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3822 */ 3823 mss = max(mss, 64); 3824 3825 tp->t_maxseg = mss; 3826 } 3827 3828 void 3829 tcp_mss(struct tcpcb *tp, int offer) 3830 { 3831 int mss; 3832 uint32_t bufsize; 3833 struct inpcb *inp; 3834 struct socket *so; 3835 struct hc_metrics_lite metrics; 3836 struct tcp_ifcap cap; 3837 3838 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3839 3840 bzero(&cap, sizeof(cap)); 3841 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3842 3843 mss = tp->t_maxseg; 3844 inp = tp->t_inpcb; 3845 3846 /* 3847 * If there's a pipesize, change the socket buffer to that size, 3848 * don't change if sb_hiwat is different than default (then it 3849 * has been changed on purpose with setsockopt). 3850 * Make the socket buffers an integral number of mss units; 3851 * if the mss is larger than the socket buffer, decrease the mss. 3852 */ 3853 so = inp->inp_socket; 3854 SOCKBUF_LOCK(&so->so_snd); 3855 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3856 bufsize = metrics.rmx_sendpipe; 3857 else 3858 bufsize = so->so_snd.sb_hiwat; 3859 if (bufsize < mss) 3860 mss = bufsize; 3861 else { 3862 bufsize = roundup(bufsize, mss); 3863 if (bufsize > sb_max) 3864 bufsize = sb_max; 3865 if (bufsize > so->so_snd.sb_hiwat) 3866 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3867 } 3868 SOCKBUF_UNLOCK(&so->so_snd); 3869 /* 3870 * Sanity check: make sure that maxseg will be large 3871 * enough to allow some data on segments even if the 3872 * all the option space is used (40bytes). Otherwise 3873 * funny things may happen in tcp_output. 3874 * 3875 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3876 */ 3877 tp->t_maxseg = max(mss, 64); 3878 3879 SOCKBUF_LOCK(&so->so_rcv); 3880 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3881 bufsize = metrics.rmx_recvpipe; 3882 else 3883 bufsize = so->so_rcv.sb_hiwat; 3884 if (bufsize > mss) { 3885 bufsize = roundup(bufsize, mss); 3886 if (bufsize > sb_max) 3887 bufsize = sb_max; 3888 if (bufsize > so->so_rcv.sb_hiwat) 3889 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3890 } 3891 SOCKBUF_UNLOCK(&so->so_rcv); 3892 3893 /* Check the interface for TSO capabilities. */ 3894 if (cap.ifcap & CSUM_TSO) { 3895 tp->t_flags |= TF_TSO; 3896 tp->t_tsomax = cap.tsomax; 3897 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3898 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3899 } 3900 } 3901 3902 /* 3903 * Determine the MSS option to send on an outgoing SYN. 3904 */ 3905 int 3906 tcp_mssopt(struct in_conninfo *inc) 3907 { 3908 int mss = 0; 3909 uint32_t thcmtu = 0; 3910 uint32_t maxmtu = 0; 3911 size_t min_protoh; 3912 3913 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3914 3915 #ifdef INET6 3916 if (inc->inc_flags & INC_ISIPV6) { 3917 mss = V_tcp_v6mssdflt; 3918 maxmtu = tcp_maxmtu6(inc, NULL); 3919 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3920 } 3921 #endif 3922 #if defined(INET) && defined(INET6) 3923 else 3924 #endif 3925 #ifdef INET 3926 { 3927 mss = V_tcp_mssdflt; 3928 maxmtu = tcp_maxmtu(inc, NULL); 3929 min_protoh = sizeof(struct tcpiphdr); 3930 } 3931 #endif 3932 #if defined(INET6) || defined(INET) 3933 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3934 #endif 3935 3936 if (maxmtu && thcmtu) 3937 mss = min(maxmtu, thcmtu) - min_protoh; 3938 else if (maxmtu || thcmtu) 3939 mss = max(maxmtu, thcmtu) - min_protoh; 3940 3941 return (mss); 3942 } 3943 3944 void 3945 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3946 { 3947 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3948 int maxseg = tcp_maxseg(tp); 3949 3950 INP_WLOCK_ASSERT(tp->t_inpcb); 3951 3952 /* 3953 * Compute the amount of data that this ACK is indicating 3954 * (del_data) and an estimate of how many bytes are in the 3955 * network. 3956 */ 3957 if (((tp->t_flags & TF_SACK_PERMIT) && 3958 (to->to_flags & TOF_SACK)) || 3959 (IN_CONGRECOVERY(tp->t_flags) && 3960 !IN_FASTRECOVERY(tp->t_flags))) { 3961 del_data = tp->sackhint.delivered_data; 3962 if (V_tcp_do_newsack) 3963 pipe = tcp_compute_pipe(tp); 3964 else 3965 pipe = (tp->snd_nxt - tp->snd_fack) + 3966 tp->sackhint.sack_bytes_rexmit; 3967 } else { 3968 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 3969 tp->snd_recover - tp->snd_una)) 3970 del_data = maxseg; 3971 pipe = imax(0, tp->snd_max - tp->snd_una - 3972 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 3973 } 3974 tp->sackhint.prr_delivered += del_data; 3975 /* 3976 * Proportional Rate Reduction 3977 */ 3978 if (pipe >= tp->snd_ssthresh) { 3979 if (tp->sackhint.recover_fs == 0) 3980 tp->sackhint.recover_fs = 3981 imax(1, tp->snd_nxt - tp->snd_una); 3982 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3983 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3984 tp->sackhint.prr_out; 3985 } else { 3986 if (V_tcp_do_prr_conservative || (del_data == 0)) 3987 limit = tp->sackhint.prr_delivered - 3988 tp->sackhint.prr_out; 3989 else 3990 limit = imax(tp->sackhint.prr_delivered - 3991 tp->sackhint.prr_out, del_data) + 3992 maxseg; 3993 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 3994 } 3995 snd_cnt = imax(snd_cnt, 0) / maxseg; 3996 /* 3997 * Send snd_cnt new data into the network in response to this ack. 3998 * If there is going to be a SACK retransmission, adjust snd_cwnd 3999 * accordingly. 4000 */ 4001 if (IN_FASTRECOVERY(tp->t_flags)) { 4002 if ((tp->t_flags & TF_SACK_PERMIT) && 4003 (to->to_flags & TOF_SACK)) { 4004 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 4005 tp->sackhint.sack_bytes_rexmit + 4006 (snd_cnt * maxseg); 4007 } else { 4008 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4009 (snd_cnt * maxseg); 4010 } 4011 } else if (IN_CONGRECOVERY(tp->t_flags)) 4012 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4013 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4014 } 4015 4016 /* 4017 * On a partial ack arrives, force the retransmission of the 4018 * next unacknowledged segment. Do not clear tp->t_dupacks. 4019 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4020 * be started again. 4021 */ 4022 void 4023 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4024 { 4025 tcp_seq onxt = tp->snd_nxt; 4026 uint32_t ocwnd = tp->snd_cwnd; 4027 u_int maxseg = tcp_maxseg(tp); 4028 4029 INP_WLOCK_ASSERT(tp->t_inpcb); 4030 4031 tcp_timer_activate(tp, TT_REXMT, 0); 4032 tp->t_rtttime = 0; 4033 tp->snd_nxt = th->th_ack; 4034 /* 4035 * Set snd_cwnd to one segment beyond acknowledged offset. 4036 * (tp->snd_una has not yet been updated when this function is called.) 4037 */ 4038 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4039 tp->t_flags |= TF_ACKNOW; 4040 (void) tcp_output(tp); 4041 tp->snd_cwnd = ocwnd; 4042 if (SEQ_GT(onxt, tp->snd_nxt)) 4043 tp->snd_nxt = onxt; 4044 /* 4045 * Partial window deflation. Relies on fact that tp->snd_una 4046 * not updated yet. 4047 */ 4048 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4049 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4050 else 4051 tp->snd_cwnd = 0; 4052 tp->snd_cwnd += maxseg; 4053 } 4054 4055 int 4056 tcp_compute_pipe(struct tcpcb *tp) 4057 { 4058 return (tp->snd_max - tp->snd_una + 4059 tp->sackhint.sack_bytes_rexmit - 4060 tp->sackhint.sacked_bytes); 4061 } 4062 4063 uint32_t 4064 tcp_compute_initwnd(uint32_t maxseg) 4065 { 4066 /* 4067 * Calculate the Initial Window, also used as Restart Window 4068 * 4069 * RFC5681 Section 3.1 specifies the default conservative values. 4070 * RFC3390 specifies slightly more aggressive values. 4071 * RFC6928 increases it to ten segments. 4072 * Support for user specified value for initial flight size. 4073 */ 4074 if (V_tcp_initcwnd_segments) 4075 return min(V_tcp_initcwnd_segments * maxseg, 4076 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4077 else if (V_tcp_do_rfc3390) 4078 return min(4 * maxseg, max(2 * maxseg, 4380)); 4079 else { 4080 /* Per RFC5681 Section 3.1 */ 4081 if (maxseg > 2190) 4082 return (2 * maxseg); 4083 else if (maxseg > 1095) 4084 return (3 * maxseg); 4085 else 4086 return (4 * maxseg); 4087 } 4088 } 4089