1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 __FBSDID("$FreeBSD$"); 54 55 #include "opt_inet.h" 56 #include "opt_inet6.h" 57 #include "opt_ipsec.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/arb.h> 62 #include <sys/kernel.h> 63 #ifdef TCP_HHOOK 64 #include <sys/hhook.h> 65 #endif 66 #include <sys/malloc.h> 67 #include <sys/mbuf.h> 68 #include <sys/proc.h> /* for proc0 declaration */ 69 #include <sys/protosw.h> 70 #include <sys/qmath.h> 71 #include <sys/sdt.h> 72 #include <sys/signalvar.h> 73 #include <sys/socket.h> 74 #include <sys/socketvar.h> 75 #include <sys/sysctl.h> 76 #include <sys/syslog.h> 77 #include <sys/systm.h> 78 #include <sys/stats.h> 79 80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 81 82 #include <vm/uma.h> 83 84 #include <net/if.h> 85 #include <net/if_var.h> 86 #include <net/route.h> 87 #include <net/vnet.h> 88 89 #define TCPSTATES /* for logging */ 90 91 #include <netinet/in.h> 92 #include <netinet/in_kdtrace.h> 93 #include <netinet/in_pcb.h> 94 #include <netinet/in_systm.h> 95 #include <netinet/ip.h> 96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 98 #include <netinet/ip_var.h> 99 #include <netinet/ip_options.h> 100 #include <netinet/ip6.h> 101 #include <netinet/icmp6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_var.h> 104 #include <netinet6/ip6_var.h> 105 #include <netinet6/nd6.h> 106 #include <netinet/tcp.h> 107 #include <netinet/tcp_fsm.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_seq.h> 110 #include <netinet/tcp_timer.h> 111 #include <netinet/tcp_var.h> 112 #include <netinet6/tcp6_var.h> 113 #include <netinet/tcpip.h> 114 #include <netinet/cc/cc.h> 115 #include <netinet/tcp_fastopen.h> 116 #ifdef TCPPCAP 117 #include <netinet/tcp_pcap.h> 118 #endif 119 #include <netinet/tcp_syncache.h> 120 #ifdef TCPDEBUG 121 #include <netinet/tcp_debug.h> 122 #endif /* TCPDEBUG */ 123 #ifdef TCP_OFFLOAD 124 #include <netinet/tcp_offload.h> 125 #endif 126 127 #include <netipsec/ipsec_support.h> 128 129 #include <machine/in_cksum.h> 130 131 #include <security/mac/mac_framework.h> 132 133 const int tcprexmtthresh = 3; 134 135 VNET_DEFINE(int, tcp_log_in_vain) = 0; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 137 &VNET_NAME(tcp_log_in_vain), 0, 138 "Log all incoming TCP segments to closed ports"); 139 140 VNET_DEFINE(int, blackhole) = 0; 141 #define V_blackhole VNET(blackhole) 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 143 &VNET_NAME(blackhole), 0, 144 "Do not send RST on segments to closed ports"); 145 146 VNET_DEFINE(int, tcp_delack_enabled) = 1; 147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 148 &VNET_NAME(tcp_delack_enabled), 0, 149 "Delay ACK to try and piggyback it onto a data packet"); 150 151 VNET_DEFINE(int, drop_synfin) = 0; 152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 153 &VNET_NAME(drop_synfin), 0, 154 "Drop TCP packets with SYN+FIN set"); 155 156 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 158 &VNET_NAME(tcp_do_prr_conservative), 0, 159 "Do conservative Proportional Rate Reduction"); 160 161 VNET_DEFINE(int, tcp_do_prr) = 1; 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 163 &VNET_NAME(tcp_do_prr), 1, 164 "Enable Proportional Rate Reduction per RFC 6937"); 165 166 VNET_DEFINE(int, tcp_do_newcwv) = 0; 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 168 &VNET_NAME(tcp_do_newcwv), 0, 169 "Enable New Congestion Window Validation per RFC7661"); 170 171 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 173 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 174 "Use calculated pipe/in-flight bytes per RFC 6675"); 175 176 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 178 &VNET_NAME(tcp_do_rfc3042), 0, 179 "Enable RFC 3042 (Limited Transmit)"); 180 181 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 183 &VNET_NAME(tcp_do_rfc3390), 0, 184 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 185 186 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 188 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 189 "Slow-start flight size (initial congestion window) in number of segments"); 190 191 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_do_rfc3465), 0, 194 "Enable RFC 3465 (Appropriate Byte Counting)"); 195 196 VNET_DEFINE(int, tcp_abc_l_var) = 2; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_abc_l_var), 2, 199 "Cap the max cwnd increment during slow-start to this number of segments"); 200 201 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 202 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 203 "TCP ECN"); 204 205 VNET_DEFINE(int, tcp_do_ecn) = 2; 206 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(tcp_do_ecn), 0, 208 "TCP ECN support"); 209 210 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 211 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_ecn_maxretries), 0, 213 "Max retries before giving up on ECN"); 214 215 VNET_DEFINE(int, tcp_insecure_syn) = 0; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_insecure_syn), 0, 218 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 219 220 VNET_DEFINE(int, tcp_insecure_rst) = 0; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_insecure_rst), 0, 223 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 224 225 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 226 #define V_tcp_recvspace VNET(tcp_recvspace) 227 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 229 230 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 232 &VNET_NAME(tcp_do_autorcvbuf), 0, 233 "Enable automatic receive buffer sizing"); 234 235 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 237 &VNET_NAME(tcp_autorcvbuf_max), 0, 238 "Max size of automatic receive buffer"); 239 240 VNET_DEFINE(struct inpcbhead, tcb); 241 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 242 VNET_DEFINE(struct inpcbinfo, tcbinfo); 243 244 /* 245 * TCP statistics are stored in an array of counter(9)s, which size matches 246 * size of struct tcpstat. TCP running connection count is a regular array. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 250 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 251 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 252 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 253 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 254 "TCP connection counts by TCP state"); 255 256 static void 257 tcp_vnet_init(const void *unused) 258 { 259 260 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 261 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 262 } 263 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 264 tcp_vnet_init, NULL); 265 266 #ifdef VIMAGE 267 static void 268 tcp_vnet_uninit(const void *unused) 269 { 270 271 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 272 VNET_PCPUSTAT_FREE(tcpstat); 273 } 274 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 275 tcp_vnet_uninit, NULL); 276 #endif /* VIMAGE */ 277 278 /* 279 * Kernel module interface for updating tcpstat. The first argument is an index 280 * into tcpstat treated as an array. 281 */ 282 void 283 kmod_tcpstat_add(int statnum, int val) 284 { 285 286 counter_u64_add(VNET(tcpstat)[statnum], val); 287 } 288 289 #ifdef TCP_HHOOK 290 /* 291 * Wrapper for the TCP established input helper hook. 292 */ 293 void 294 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 295 { 296 struct tcp_hhook_data hhook_data; 297 298 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 299 hhook_data.tp = tp; 300 hhook_data.th = th; 301 hhook_data.to = to; 302 303 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 304 tp->osd); 305 } 306 } 307 #endif 308 309 /* 310 * CC wrapper hook functions 311 */ 312 void 313 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 314 uint16_t type) 315 { 316 #ifdef STATS 317 int32_t gput; 318 #endif 319 320 INP_WLOCK_ASSERT(tp->t_inpcb); 321 322 tp->ccv->nsegs = nsegs; 323 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 324 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 325 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 326 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 327 tp->ccv->flags |= CCF_CWND_LIMITED; 328 else 329 tp->ccv->flags &= ~CCF_CWND_LIMITED; 330 331 if (type == CC_ACK) { 332 #ifdef STATS 333 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 334 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 335 if (!IN_RECOVERY(tp->t_flags)) 336 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 337 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 338 if ((tp->t_flags & TF_GPUTINPROG) && 339 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 340 /* 341 * Compute goodput in bits per millisecond. 342 */ 343 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) / 344 max(1, tcp_ts_getticks() - tp->gput_ts); 345 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 346 gput); 347 /* 348 * XXXLAS: This is a temporary hack, and should be 349 * chained off VOI_TCP_GPUT when stats(9) grows an API 350 * to deal with chained VOIs. 351 */ 352 if (tp->t_stats_gput_prev > 0) 353 stats_voi_update_abs_s32(tp->t_stats, 354 VOI_TCP_GPUT_ND, 355 ((gput - tp->t_stats_gput_prev) * 100) / 356 tp->t_stats_gput_prev); 357 tp->t_flags &= ~TF_GPUTINPROG; 358 tp->t_stats_gput_prev = gput; 359 } 360 #endif /* STATS */ 361 if (tp->snd_cwnd > tp->snd_ssthresh) { 362 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 363 if (tp->t_bytes_acked >= tp->snd_cwnd) { 364 tp->t_bytes_acked -= tp->snd_cwnd; 365 tp->ccv->flags |= CCF_ABC_SENTAWND; 366 } 367 } else { 368 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 369 tp->t_bytes_acked = 0; 370 } 371 } 372 373 if (CC_ALGO(tp)->ack_received != NULL) { 374 /* XXXLAS: Find a way to live without this */ 375 tp->ccv->curack = th->th_ack; 376 CC_ALGO(tp)->ack_received(tp->ccv, type); 377 } 378 #ifdef STATS 379 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 380 #endif 381 } 382 383 void 384 cc_conn_init(struct tcpcb *tp) 385 { 386 struct hc_metrics_lite metrics; 387 struct inpcb *inp = tp->t_inpcb; 388 u_int maxseg; 389 int rtt; 390 391 INP_WLOCK_ASSERT(tp->t_inpcb); 392 393 tcp_hc_get(&inp->inp_inc, &metrics); 394 maxseg = tcp_maxseg(tp); 395 396 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 397 tp->t_srtt = rtt; 398 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 399 TCPSTAT_INC(tcps_usedrtt); 400 if (metrics.rmx_rttvar) { 401 tp->t_rttvar = metrics.rmx_rttvar; 402 TCPSTAT_INC(tcps_usedrttvar); 403 } else { 404 /* default variation is +- 1 rtt */ 405 tp->t_rttvar = 406 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 407 } 408 TCPT_RANGESET(tp->t_rxtcur, 409 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 410 tp->t_rttmin, TCPTV_REXMTMAX); 411 } 412 if (metrics.rmx_ssthresh) { 413 /* 414 * There's some sort of gateway or interface 415 * buffer limit on the path. Use this to set 416 * the slow start threshold, but set the 417 * threshold to no less than 2*mss. 418 */ 419 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 420 TCPSTAT_INC(tcps_usedssthresh); 421 } 422 423 /* 424 * Set the initial slow-start flight size. 425 * 426 * If a SYN or SYN/ACK was lost and retransmitted, we have to 427 * reduce the initial CWND to one segment as congestion is likely 428 * requiring us to be cautious. 429 */ 430 if (tp->snd_cwnd == 1) 431 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 432 else 433 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 434 435 if (CC_ALGO(tp)->conn_init != NULL) 436 CC_ALGO(tp)->conn_init(tp->ccv); 437 } 438 439 void inline 440 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 441 { 442 INP_WLOCK_ASSERT(tp->t_inpcb); 443 444 #ifdef STATS 445 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 446 #endif 447 448 switch(type) { 449 case CC_NDUPACK: 450 if (!IN_FASTRECOVERY(tp->t_flags)) { 451 tp->snd_recover = tp->snd_max; 452 if (tp->t_flags2 & TF2_ECN_PERMIT) 453 tp->t_flags2 |= TF2_ECN_SND_CWR; 454 } 455 break; 456 case CC_ECN: 457 if (!IN_CONGRECOVERY(tp->t_flags) || 458 /* 459 * Allow ECN reaction on ACK to CWR, if 460 * that data segment was also CE marked. 461 */ 462 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 463 EXIT_CONGRECOVERY(tp->t_flags); 464 TCPSTAT_INC(tcps_ecn_rcwnd); 465 tp->snd_recover = tp->snd_max + 1; 466 if (tp->t_flags2 & TF2_ECN_PERMIT) 467 tp->t_flags2 |= TF2_ECN_SND_CWR; 468 } 469 break; 470 case CC_RTO: 471 tp->t_dupacks = 0; 472 tp->t_bytes_acked = 0; 473 EXIT_RECOVERY(tp->t_flags); 474 if (tp->t_flags2 & TF2_ECN_PERMIT) 475 tp->t_flags2 |= TF2_ECN_SND_CWR; 476 break; 477 case CC_RTO_ERR: 478 TCPSTAT_INC(tcps_sndrexmitbad); 479 /* RTO was unnecessary, so reset everything. */ 480 tp->snd_cwnd = tp->snd_cwnd_prev; 481 tp->snd_ssthresh = tp->snd_ssthresh_prev; 482 tp->snd_recover = tp->snd_recover_prev; 483 if (tp->t_flags & TF_WASFRECOVERY) 484 ENTER_FASTRECOVERY(tp->t_flags); 485 if (tp->t_flags & TF_WASCRECOVERY) 486 ENTER_CONGRECOVERY(tp->t_flags); 487 tp->snd_nxt = tp->snd_max; 488 tp->t_flags &= ~TF_PREVVALID; 489 tp->t_badrxtwin = 0; 490 break; 491 } 492 493 if (CC_ALGO(tp)->cong_signal != NULL) { 494 if (th != NULL) 495 tp->ccv->curack = th->th_ack; 496 CC_ALGO(tp)->cong_signal(tp->ccv, type); 497 } 498 } 499 500 void inline 501 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 502 { 503 INP_WLOCK_ASSERT(tp->t_inpcb); 504 505 /* XXXLAS: KASSERT that we're in recovery? */ 506 507 if (CC_ALGO(tp)->post_recovery != NULL) { 508 tp->ccv->curack = th->th_ack; 509 CC_ALGO(tp)->post_recovery(tp->ccv); 510 } 511 /* XXXLAS: EXIT_RECOVERY ? */ 512 tp->t_bytes_acked = 0; 513 } 514 515 /* 516 * Indicate whether this ack should be delayed. We can delay the ack if 517 * following conditions are met: 518 * - There is no delayed ack timer in progress. 519 * - Our last ack wasn't a 0-sized window. We never want to delay 520 * the ack that opens up a 0-sized window. 521 * - LRO wasn't used for this segment. We make sure by checking that the 522 * segment size is not larger than the MSS. 523 */ 524 #define DELAY_ACK(tp, tlen) \ 525 ((!tcp_timer_active(tp, TT_DELACK) && \ 526 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 527 (tlen <= tp->t_maxseg) && \ 528 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 529 530 void inline 531 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 532 { 533 INP_WLOCK_ASSERT(tp->t_inpcb); 534 535 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 536 switch (iptos & IPTOS_ECN_MASK) { 537 case IPTOS_ECN_CE: 538 tp->ccv->flags |= CCF_IPHDR_CE; 539 break; 540 case IPTOS_ECN_ECT0: 541 /* FALLTHROUGH */ 542 case IPTOS_ECN_ECT1: 543 /* FALLTHROUGH */ 544 case IPTOS_ECN_NOTECT: 545 tp->ccv->flags &= ~CCF_IPHDR_CE; 546 break; 547 } 548 549 if (th->th_flags & TH_CWR) 550 tp->ccv->flags |= CCF_TCPHDR_CWR; 551 else 552 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 553 554 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 555 556 if (tp->ccv->flags & CCF_ACKNOW) { 557 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 558 tp->t_flags |= TF_ACKNOW; 559 } 560 } 561 } 562 563 /* 564 * TCP input handling is split into multiple parts: 565 * tcp6_input is a thin wrapper around tcp_input for the extended 566 * ip6_protox[] call format in ip6_input 567 * tcp_input handles primary segment validation, inpcb lookup and 568 * SYN processing on listen sockets 569 * tcp_do_segment processes the ACK and text of the segment for 570 * establishing, established and closing connections 571 */ 572 #ifdef INET6 573 int 574 tcp6_input(struct mbuf **mp, int *offp, int proto) 575 { 576 struct mbuf *m; 577 struct in6_ifaddr *ia6; 578 struct ip6_hdr *ip6; 579 580 m = *mp; 581 if (m->m_len < *offp + sizeof(struct tcphdr)) { 582 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 583 if (m == NULL) { 584 *mp = m; 585 TCPSTAT_INC(tcps_rcvshort); 586 return (IPPROTO_DONE); 587 } 588 } 589 590 /* 591 * draft-itojun-ipv6-tcp-to-anycast 592 * better place to put this in? 593 */ 594 ip6 = mtod(m, struct ip6_hdr *); 595 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 596 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 597 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 598 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 599 *mp = NULL; 600 return (IPPROTO_DONE); 601 } 602 603 *mp = m; 604 return (tcp_input(mp, offp, proto)); 605 } 606 #endif /* INET6 */ 607 608 int 609 tcp_input(struct mbuf **mp, int *offp, int proto) 610 { 611 struct mbuf *m = *mp; 612 struct tcphdr *th = NULL; 613 struct ip *ip = NULL; 614 struct inpcb *inp = NULL; 615 struct tcpcb *tp = NULL; 616 struct socket *so = NULL; 617 u_char *optp = NULL; 618 int off0; 619 int optlen = 0; 620 #ifdef INET 621 int len; 622 uint8_t ipttl; 623 #endif 624 int tlen = 0, off; 625 int drop_hdrlen; 626 int thflags; 627 int rstreason = 0; /* For badport_bandlim accounting purposes */ 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 (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 666 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 667 th->th_sum = m->m_pkthdr.csum_data; 668 else 669 th->th_sum = in6_cksum_pseudo(ip6, tlen, 670 IPPROTO_TCP, m->m_pkthdr.csum_data); 671 th->th_sum ^= 0xffff; 672 } else 673 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 674 if (th->th_sum) { 675 TCPSTAT_INC(tcps_rcvbadsum); 676 goto drop; 677 } 678 679 /* 680 * Be proactive about unspecified IPv6 address in source. 681 * As we use all-zero to indicate unbounded/unconnected pcb, 682 * unspecified IPv6 address can be used to confuse us. 683 * 684 * Note that packets with unspecified IPv6 destination is 685 * already dropped in ip6_input. 686 */ 687 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 688 /* XXX stat */ 689 goto drop; 690 } 691 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 692 } 693 #endif 694 #if defined(INET) && defined(INET6) 695 else 696 #endif 697 #ifdef INET 698 { 699 /* 700 * Get IP and TCP header together in first mbuf. 701 * Note: IP leaves IP header in first mbuf. 702 */ 703 if (off0 > sizeof (struct ip)) { 704 ip_stripoptions(m); 705 off0 = sizeof(struct ip); 706 } 707 if (m->m_len < sizeof (struct tcpiphdr)) { 708 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 709 == NULL) { 710 TCPSTAT_INC(tcps_rcvshort); 711 return (IPPROTO_DONE); 712 } 713 } 714 ip = mtod(m, struct ip *); 715 th = (struct tcphdr *)((caddr_t)ip + off0); 716 tlen = ntohs(ip->ip_len) - off0; 717 718 iptos = ip->ip_tos; 719 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 720 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 721 th->th_sum = m->m_pkthdr.csum_data; 722 else 723 th->th_sum = in_pseudo(ip->ip_src.s_addr, 724 ip->ip_dst.s_addr, 725 htonl(m->m_pkthdr.csum_data + tlen + 726 IPPROTO_TCP)); 727 th->th_sum ^= 0xffff; 728 } else { 729 struct ipovly *ipov = (struct ipovly *)ip; 730 731 /* 732 * Checksum extended TCP header and data. 733 */ 734 len = off0 + tlen; 735 ipttl = ip->ip_ttl; 736 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 737 ipov->ih_len = htons(tlen); 738 th->th_sum = in_cksum(m, len); 739 /* Reset length for SDT probes. */ 740 ip->ip_len = htons(len); 741 /* Reset TOS bits */ 742 ip->ip_tos = iptos; 743 /* Re-initialization for later version check */ 744 ip->ip_ttl = ipttl; 745 ip->ip_v = IPVERSION; 746 ip->ip_hl = off0 >> 2; 747 } 748 749 if (th->th_sum) { 750 TCPSTAT_INC(tcps_rcvbadsum); 751 goto drop; 752 } 753 } 754 #endif /* INET */ 755 756 /* 757 * Check that TCP offset makes sense, 758 * pull out TCP options and adjust length. XXX 759 */ 760 off = th->th_off << 2; 761 if (off < sizeof (struct tcphdr) || off > tlen) { 762 TCPSTAT_INC(tcps_rcvbadoff); 763 goto drop; 764 } 765 tlen -= off; /* tlen is used instead of ti->ti_len */ 766 if (off > sizeof (struct tcphdr)) { 767 #ifdef INET6 768 if (isipv6) { 769 if (m->m_len < off0 + off) { 770 m = m_pullup(m, off0 + off); 771 if (m == NULL) { 772 TCPSTAT_INC(tcps_rcvshort); 773 return (IPPROTO_DONE); 774 } 775 } 776 ip6 = mtod(m, struct ip6_hdr *); 777 th = (struct tcphdr *)((caddr_t)ip6 + off0); 778 } 779 #endif 780 #if defined(INET) && defined(INET6) 781 else 782 #endif 783 #ifdef INET 784 { 785 if (m->m_len < sizeof(struct ip) + off) { 786 if ((m = m_pullup(m, sizeof (struct ip) + off)) 787 == NULL) { 788 TCPSTAT_INC(tcps_rcvshort); 789 return (IPPROTO_DONE); 790 } 791 ip = mtod(m, struct ip *); 792 th = (struct tcphdr *)((caddr_t)ip + off0); 793 } 794 } 795 #endif 796 optlen = off - sizeof (struct tcphdr); 797 optp = (u_char *)(th + 1); 798 } 799 thflags = th->th_flags; 800 801 /* 802 * Convert TCP protocol specific fields to host format. 803 */ 804 tcp_fields_to_host(th); 805 806 /* 807 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 808 */ 809 drop_hdrlen = off0 + off; 810 811 /* 812 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 813 */ 814 if ( 815 #ifdef INET6 816 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 817 #ifdef INET 818 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 819 #endif 820 #endif 821 #if defined(INET) && !defined(INET6) 822 (m->m_flags & M_IP_NEXTHOP) 823 #endif 824 ) 825 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 826 827 findpcb: 828 #ifdef INET6 829 if (isipv6 && fwd_tag != NULL) { 830 struct sockaddr_in6 *next_hop6; 831 832 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 833 /* 834 * Transparently forwarded. Pretend to be the destination. 835 * Already got one like this? 836 */ 837 inp = in6_pcblookup_mbuf(&V_tcbinfo, 838 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 839 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 840 if (!inp) { 841 /* 842 * It's new. Try to find the ambushing socket. 843 * Because we've rewritten the destination address, 844 * any hardware-generated hash is ignored. 845 */ 846 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 847 th->th_sport, &next_hop6->sin6_addr, 848 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 849 th->th_dport, INPLOOKUP_WILDCARD | 850 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 851 } 852 } else if (isipv6) { 853 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 854 th->th_sport, &ip6->ip6_dst, th->th_dport, 855 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 856 m->m_pkthdr.rcvif, m); 857 } 858 #endif /* INET6 */ 859 #if defined(INET6) && defined(INET) 860 else 861 #endif 862 #ifdef INET 863 if (fwd_tag != NULL) { 864 struct sockaddr_in *next_hop; 865 866 next_hop = (struct sockaddr_in *)(fwd_tag+1); 867 /* 868 * Transparently forwarded. Pretend to be the destination. 869 * already got one like this? 870 */ 871 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 872 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 873 m->m_pkthdr.rcvif, m); 874 if (!inp) { 875 /* 876 * It's new. Try to find the ambushing socket. 877 * Because we've rewritten the destination address, 878 * any hardware-generated hash is ignored. 879 */ 880 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 881 th->th_sport, next_hop->sin_addr, 882 next_hop->sin_port ? ntohs(next_hop->sin_port) : 883 th->th_dport, INPLOOKUP_WILDCARD | 884 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 885 } 886 } else 887 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 888 th->th_sport, ip->ip_dst, th->th_dport, 889 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 890 m->m_pkthdr.rcvif, m); 891 #endif /* INET */ 892 893 /* 894 * If the INPCB does not exist then all data in the incoming 895 * segment is discarded and an appropriate RST is sent back. 896 * XXX MRT Send RST using which routing table? 897 */ 898 if (inp == NULL) { 899 /* 900 * Log communication attempts to ports that are not 901 * in use. 902 */ 903 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 904 V_tcp_log_in_vain == 2) { 905 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 906 log(LOG_INFO, "%s; %s: Connection attempt " 907 "to closed port\n", s, __func__); 908 } 909 /* 910 * When blackholing do not respond with a RST but 911 * completely ignore the segment and drop it. 912 */ 913 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 914 V_blackhole == 2) 915 goto dropunlock; 916 917 rstreason = BANDLIM_RST_CLOSEDPORT; 918 goto dropwithreset; 919 } 920 INP_WLOCK_ASSERT(inp); 921 /* 922 * While waiting for inp lock during the lookup, another thread 923 * can have dropped the inpcb, in which case we need to loop back 924 * and try to find a new inpcb to deliver to. 925 */ 926 if (inp->inp_flags & INP_DROPPED) { 927 INP_WUNLOCK(inp); 928 inp = NULL; 929 goto findpcb; 930 } 931 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 932 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 933 ((inp->inp_socket == NULL) || 934 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 935 inp->inp_flowid = m->m_pkthdr.flowid; 936 inp->inp_flowtype = M_HASHTYPE_GET(m); 937 } 938 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 939 #ifdef INET6 940 if (isipv6 && IPSEC_ENABLED(ipv6) && 941 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 942 goto dropunlock; 943 } 944 #ifdef INET 945 else 946 #endif 947 #endif /* INET6 */ 948 #ifdef INET 949 if (IPSEC_ENABLED(ipv4) && 950 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 951 goto dropunlock; 952 } 953 #endif /* INET */ 954 #endif /* IPSEC */ 955 956 /* 957 * Check the minimum TTL for socket. 958 */ 959 if (inp->inp_ip_minttl != 0) { 960 #ifdef INET6 961 if (isipv6) { 962 if (inp->inp_ip_minttl > ip6->ip6_hlim) 963 goto dropunlock; 964 } else 965 #endif 966 if (inp->inp_ip_minttl > ip->ip_ttl) 967 goto dropunlock; 968 } 969 970 /* 971 * A previous connection in TIMEWAIT state is supposed to catch stray 972 * or duplicate segments arriving late. If this segment was a 973 * legitimate new connection attempt, the old INPCB gets removed and 974 * we can try again to find a listening socket. 975 * 976 * At this point, due to earlier optimism, we may hold only an inpcb 977 * lock, and not the inpcbinfo write lock. If so, we need to try to 978 * acquire it, or if that fails, acquire a reference on the inpcb, 979 * drop all locks, acquire a global write lock, and then re-acquire 980 * the inpcb lock. We may at that point discover that another thread 981 * has tried to free the inpcb, in which case we need to loop back 982 * and try to find a new inpcb to deliver to. 983 * 984 * XXXRW: It may be time to rethink timewait locking. 985 */ 986 if (inp->inp_flags & INP_TIMEWAIT) { 987 tcp_dooptions(&to, optp, optlen, 988 (thflags & TH_SYN) ? TO_SYN : 0); 989 /* 990 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 991 */ 992 if (tcp_twcheck(inp, &to, th, m, tlen)) 993 goto findpcb; 994 return (IPPROTO_DONE); 995 } 996 /* 997 * The TCPCB may no longer exist if the connection is winding 998 * down or it is in the CLOSED state. Either way we drop the 999 * segment and send an appropriate response. 1000 */ 1001 tp = intotcpcb(inp); 1002 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1003 rstreason = BANDLIM_RST_CLOSEDPORT; 1004 goto dropwithreset; 1005 } 1006 1007 #ifdef TCP_OFFLOAD 1008 if (tp->t_flags & TF_TOE) { 1009 tcp_offload_input(tp, m); 1010 m = NULL; /* consumed by the TOE driver */ 1011 goto dropunlock; 1012 } 1013 #endif 1014 1015 #ifdef MAC 1016 INP_WLOCK_ASSERT(inp); 1017 if (mac_inpcb_check_deliver(inp, m)) 1018 goto dropunlock; 1019 #endif 1020 so = inp->inp_socket; 1021 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1022 #ifdef TCPDEBUG 1023 if (so->so_options & SO_DEBUG) { 1024 ostate = tp->t_state; 1025 #ifdef INET6 1026 if (isipv6) { 1027 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1028 } else 1029 #endif 1030 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1031 tcp_savetcp = *th; 1032 } 1033 #endif /* TCPDEBUG */ 1034 /* 1035 * When the socket is accepting connections (the INPCB is in LISTEN 1036 * state) we look into the SYN cache if this is a new connection 1037 * attempt or the completion of a previous one. 1038 */ 1039 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1040 ("%s: so accepting but tp %p not listening", __func__, tp)); 1041 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1042 struct in_conninfo inc; 1043 1044 bzero(&inc, sizeof(inc)); 1045 #ifdef INET6 1046 if (isipv6) { 1047 inc.inc_flags |= INC_ISIPV6; 1048 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1049 inc.inc_flags |= INC_IPV6MINMTU; 1050 inc.inc6_faddr = ip6->ip6_src; 1051 inc.inc6_laddr = ip6->ip6_dst; 1052 } else 1053 #endif 1054 { 1055 inc.inc_faddr = ip->ip_src; 1056 inc.inc_laddr = ip->ip_dst; 1057 } 1058 inc.inc_fport = th->th_sport; 1059 inc.inc_lport = th->th_dport; 1060 inc.inc_fibnum = so->so_fibnum; 1061 1062 /* 1063 * Check for an existing connection attempt in syncache if 1064 * the flag is only ACK. A successful lookup creates a new 1065 * socket appended to the listen queue in SYN_RECEIVED state. 1066 */ 1067 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1068 /* 1069 * Parse the TCP options here because 1070 * syncookies need access to the reflected 1071 * timestamp. 1072 */ 1073 tcp_dooptions(&to, optp, optlen, 0); 1074 /* 1075 * NB: syncache_expand() doesn't unlock 1076 * inp and tcpinfo locks. 1077 */ 1078 rstreason = syncache_expand(&inc, &to, th, &so, m); 1079 if (rstreason < 0) { 1080 /* 1081 * A failing TCP MD5 signature comparison 1082 * must result in the segment being dropped 1083 * and must not produce any response back 1084 * to the sender. 1085 */ 1086 goto dropunlock; 1087 } else if (rstreason == 0) { 1088 /* 1089 * No syncache entry or ACK was not 1090 * for our SYN/ACK. Send a RST. 1091 * NB: syncache did its own logging 1092 * of the failure cause. 1093 */ 1094 rstreason = BANDLIM_RST_OPENPORT; 1095 goto dropwithreset; 1096 } 1097 tfo_socket_result: 1098 if (so == NULL) { 1099 /* 1100 * We completed the 3-way handshake 1101 * but could not allocate a socket 1102 * either due to memory shortage, 1103 * listen queue length limits or 1104 * global socket limits. Send RST 1105 * or wait and have the remote end 1106 * retransmit the ACK for another 1107 * try. 1108 */ 1109 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1110 log(LOG_DEBUG, "%s; %s: Listen socket: " 1111 "Socket allocation failed due to " 1112 "limits or memory shortage, %s\n", 1113 s, __func__, 1114 V_tcp_sc_rst_sock_fail ? 1115 "sending RST" : "try again"); 1116 if (V_tcp_sc_rst_sock_fail) { 1117 rstreason = BANDLIM_UNLIMITED; 1118 goto dropwithreset; 1119 } else 1120 goto dropunlock; 1121 } 1122 /* 1123 * Socket is created in state SYN_RECEIVED. 1124 * Unlock the listen socket, lock the newly 1125 * created socket and update the tp variable. 1126 */ 1127 INP_WUNLOCK(inp); /* listen socket */ 1128 inp = sotoinpcb(so); 1129 /* 1130 * New connection inpcb is already locked by 1131 * syncache_expand(). 1132 */ 1133 INP_WLOCK_ASSERT(inp); 1134 tp = intotcpcb(inp); 1135 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1136 ("%s: ", __func__)); 1137 /* 1138 * Process the segment and the data it 1139 * contains. tcp_do_segment() consumes 1140 * the mbuf chain and unlocks the inpcb. 1141 */ 1142 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1143 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1144 iptos); 1145 return (IPPROTO_DONE); 1146 } 1147 /* 1148 * Segment flag validation for new connection attempts: 1149 * 1150 * Our (SYN|ACK) response was rejected. 1151 * Check with syncache and remove entry to prevent 1152 * retransmits. 1153 * 1154 * NB: syncache_chkrst does its own logging of failure 1155 * causes. 1156 */ 1157 if (thflags & TH_RST) { 1158 syncache_chkrst(&inc, th, m); 1159 goto dropunlock; 1160 } 1161 /* 1162 * We can't do anything without SYN. 1163 */ 1164 if ((thflags & TH_SYN) == 0) { 1165 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1166 log(LOG_DEBUG, "%s; %s: Listen socket: " 1167 "SYN is missing, segment ignored\n", 1168 s, __func__); 1169 TCPSTAT_INC(tcps_badsyn); 1170 goto dropunlock; 1171 } 1172 /* 1173 * (SYN|ACK) is bogus on a listen socket. 1174 */ 1175 if (thflags & TH_ACK) { 1176 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1177 log(LOG_DEBUG, "%s; %s: Listen socket: " 1178 "SYN|ACK invalid, segment rejected\n", 1179 s, __func__); 1180 syncache_badack(&inc); /* XXX: Not needed! */ 1181 TCPSTAT_INC(tcps_badsyn); 1182 rstreason = BANDLIM_RST_OPENPORT; 1183 goto dropwithreset; 1184 } 1185 /* 1186 * If the drop_synfin option is enabled, drop all 1187 * segments with both the SYN and FIN bits set. 1188 * This prevents e.g. nmap from identifying the 1189 * TCP/IP stack. 1190 * XXX: Poor reasoning. nmap has other methods 1191 * and is constantly refining its stack detection 1192 * strategies. 1193 * XXX: This is a violation of the TCP specification 1194 * and was used by RFC1644. 1195 */ 1196 if ((thflags & TH_FIN) && V_drop_synfin) { 1197 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1198 log(LOG_DEBUG, "%s; %s: Listen socket: " 1199 "SYN|FIN segment ignored (based on " 1200 "sysctl setting)\n", s, __func__); 1201 TCPSTAT_INC(tcps_badsyn); 1202 goto dropunlock; 1203 } 1204 /* 1205 * Segment's flags are (SYN) or (SYN|FIN). 1206 * 1207 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1208 * as they do not affect the state of the TCP FSM. 1209 * The data pointed to by TH_URG and th_urp is ignored. 1210 */ 1211 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1212 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1213 KASSERT(thflags & (TH_SYN), 1214 ("%s: Listen socket: TH_SYN not set", __func__)); 1215 #ifdef INET6 1216 /* 1217 * If deprecated address is forbidden, 1218 * we do not accept SYN to deprecated interface 1219 * address to prevent any new inbound connection from 1220 * getting established. 1221 * When we do not accept SYN, we send a TCP RST, 1222 * with deprecated source address (instead of dropping 1223 * it). We compromise it as it is much better for peer 1224 * to send a RST, and RST will be the final packet 1225 * for the exchange. 1226 * 1227 * If we do not forbid deprecated addresses, we accept 1228 * the SYN packet. RFC2462 does not suggest dropping 1229 * SYN in this case. 1230 * If we decipher RFC2462 5.5.4, it says like this: 1231 * 1. use of deprecated addr with existing 1232 * communication is okay - "SHOULD continue to be 1233 * used" 1234 * 2. use of it with new communication: 1235 * (2a) "SHOULD NOT be used if alternate address 1236 * with sufficient scope is available" 1237 * (2b) nothing mentioned otherwise. 1238 * Here we fall into (2b) case as we have no choice in 1239 * our source address selection - we must obey the peer. 1240 * 1241 * The wording in RFC2462 is confusing, and there are 1242 * multiple description text for deprecated address 1243 * handling - worse, they are not exactly the same. 1244 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1245 */ 1246 if (isipv6 && !V_ip6_use_deprecated) { 1247 struct in6_ifaddr *ia6; 1248 1249 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1250 if (ia6 != NULL && 1251 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1252 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1253 log(LOG_DEBUG, "%s; %s: Listen socket: " 1254 "Connection attempt to deprecated " 1255 "IPv6 address rejected\n", 1256 s, __func__); 1257 rstreason = BANDLIM_RST_OPENPORT; 1258 goto dropwithreset; 1259 } 1260 } 1261 #endif /* INET6 */ 1262 /* 1263 * Basic sanity checks on incoming SYN requests: 1264 * Don't respond if the destination is a link layer 1265 * broadcast according to RFC1122 4.2.3.10, p. 104. 1266 * If it is from this socket it must be forged. 1267 * Don't respond if the source or destination is a 1268 * global or subnet broad- or multicast address. 1269 * Note that it is quite possible to receive unicast 1270 * link-layer packets with a broadcast IP address. Use 1271 * in_broadcast() to find them. 1272 */ 1273 if (m->m_flags & (M_BCAST|M_MCAST)) { 1274 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1275 log(LOG_DEBUG, "%s; %s: Listen socket: " 1276 "Connection attempt from broad- or multicast " 1277 "link layer address ignored\n", s, __func__); 1278 goto dropunlock; 1279 } 1280 #ifdef INET6 1281 if (isipv6) { 1282 if (th->th_dport == th->th_sport && 1283 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1284 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1285 log(LOG_DEBUG, "%s; %s: Listen socket: " 1286 "Connection attempt to/from self " 1287 "ignored\n", s, __func__); 1288 goto dropunlock; 1289 } 1290 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1291 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1292 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1293 log(LOG_DEBUG, "%s; %s: Listen socket: " 1294 "Connection attempt from/to multicast " 1295 "address ignored\n", s, __func__); 1296 goto dropunlock; 1297 } 1298 } 1299 #endif 1300 #if defined(INET) && defined(INET6) 1301 else 1302 #endif 1303 #ifdef INET 1304 { 1305 if (th->th_dport == th->th_sport && 1306 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1307 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1308 log(LOG_DEBUG, "%s; %s: Listen socket: " 1309 "Connection attempt from/to self " 1310 "ignored\n", s, __func__); 1311 goto dropunlock; 1312 } 1313 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1314 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1315 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1316 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1318 log(LOG_DEBUG, "%s; %s: Listen socket: " 1319 "Connection attempt from/to broad- " 1320 "or multicast address ignored\n", 1321 s, __func__); 1322 goto dropunlock; 1323 } 1324 } 1325 #endif 1326 /* 1327 * SYN appears to be valid. Create compressed TCP state 1328 * for syncache. 1329 */ 1330 #ifdef TCPDEBUG 1331 if (so->so_options & SO_DEBUG) 1332 tcp_trace(TA_INPUT, ostate, tp, 1333 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1334 #endif 1335 TCP_PROBE3(debug__input, tp, th, m); 1336 tcp_dooptions(&to, optp, optlen, TO_SYN); 1337 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos)) 1338 goto tfo_socket_result; 1339 1340 /* 1341 * Entry added to syncache and mbuf consumed. 1342 * Only the listen socket is unlocked by syncache_add(). 1343 */ 1344 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1345 return (IPPROTO_DONE); 1346 } else if (tp->t_state == TCPS_LISTEN) { 1347 /* 1348 * When a listen socket is torn down the SO_ACCEPTCONN 1349 * flag is removed first while connections are drained 1350 * from the accept queue in a unlock/lock cycle of the 1351 * ACCEPT_LOCK, opening a race condition allowing a SYN 1352 * attempt go through unhandled. 1353 */ 1354 goto dropunlock; 1355 } 1356 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1357 if (tp->t_flags & TF_SIGNATURE) { 1358 tcp_dooptions(&to, optp, optlen, thflags); 1359 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1360 TCPSTAT_INC(tcps_sig_err_nosigopt); 1361 goto dropunlock; 1362 } 1363 if (!TCPMD5_ENABLED() || 1364 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1365 goto dropunlock; 1366 } 1367 #endif 1368 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1369 1370 /* 1371 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1372 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1373 * the inpcb, and unlocks pcbinfo. 1374 */ 1375 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1376 return (IPPROTO_DONE); 1377 1378 dropwithreset: 1379 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1380 1381 if (inp != NULL) { 1382 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1383 INP_WUNLOCK(inp); 1384 } else 1385 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1386 m = NULL; /* mbuf chain got consumed. */ 1387 goto drop; 1388 1389 dropunlock: 1390 if (m != NULL) 1391 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1392 1393 if (inp != NULL) 1394 INP_WUNLOCK(inp); 1395 1396 drop: 1397 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1398 if (s != NULL) 1399 free(s, M_TCPLOG); 1400 if (m != NULL) 1401 m_freem(m); 1402 return (IPPROTO_DONE); 1403 } 1404 1405 /* 1406 * Automatic sizing of receive socket buffer. Often the send 1407 * buffer size is not optimally adjusted to the actual network 1408 * conditions at hand (delay bandwidth product). Setting the 1409 * buffer size too small limits throughput on links with high 1410 * bandwidth and high delay (eg. trans-continental/oceanic links). 1411 * 1412 * On the receive side the socket buffer memory is only rarely 1413 * used to any significant extent. This allows us to be much 1414 * more aggressive in scaling the receive socket buffer. For 1415 * the case that the buffer space is actually used to a large 1416 * extent and we run out of kernel memory we can simply drop 1417 * the new segments; TCP on the sender will just retransmit it 1418 * later. Setting the buffer size too big may only consume too 1419 * much kernel memory if the application doesn't read() from 1420 * the socket or packet loss or reordering makes use of the 1421 * reassembly queue. 1422 * 1423 * The criteria to step up the receive buffer one notch are: 1424 * 1. Application has not set receive buffer size with 1425 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1426 * 2. the number of bytes received during 1/2 of an sRTT 1427 * is at least 3/8 of the current socket buffer size. 1428 * 3. receive buffer size has not hit maximal automatic size; 1429 * 1430 * If all of the criteria are met we increaset the socket buffer 1431 * by a 1/2 (bounded by the max). This allows us to keep ahead 1432 * of slow-start but also makes it so our peer never gets limited 1433 * by our rwnd which we then open up causing a burst. 1434 * 1435 * This algorithm does two steps per RTT at most and only if 1436 * we receive a bulk stream w/o packet losses or reorderings. 1437 * Shrinking the buffer during idle times is not necessary as 1438 * it doesn't consume any memory when idle. 1439 * 1440 * TODO: Only step up if the application is actually serving 1441 * the buffer to better manage the socket buffer resources. 1442 */ 1443 int 1444 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1445 struct tcpcb *tp, int tlen) 1446 { 1447 int newsize = 0; 1448 1449 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1450 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1451 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1452 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1453 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1454 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1455 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1456 } 1457 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1458 1459 /* Start over with next RTT. */ 1460 tp->rfbuf_ts = 0; 1461 tp->rfbuf_cnt = 0; 1462 } else { 1463 tp->rfbuf_cnt += tlen; /* add up */ 1464 } 1465 return (newsize); 1466 } 1467 1468 void 1469 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1470 { 1471 /* 1472 * Since tp might be gone if the session entered 1473 * the TIME_WAIT state before coming here, we need 1474 * to check if the socket is still connected. 1475 */ 1476 if ((so->so_state & SS_ISCONNECTED) == 0) 1477 return; 1478 INP_LOCK_ASSERT(tp->t_inpcb); 1479 if (tp->t_flags & TF_WAKESOR) { 1480 tp->t_flags &= ~TF_WAKESOR; 1481 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 1482 sorwakeup(so); 1483 } 1484 if (tp->t_flags & TF_WAKESOW) { 1485 tp->t_flags &= ~TF_WAKESOW; 1486 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 1487 sowwakeup(so); 1488 } 1489 } 1490 1491 void 1492 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1493 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1494 { 1495 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1496 int rstreason, todrop, win, incforsyn = 0; 1497 uint32_t tiwin; 1498 uint16_t nsegs; 1499 char *s; 1500 struct in_conninfo *inc; 1501 struct mbuf *mfree; 1502 struct tcpopt to; 1503 int tfo_syn; 1504 u_int maxseg; 1505 1506 #ifdef TCPDEBUG 1507 /* 1508 * The size of tcp_saveipgen must be the size of the max ip header, 1509 * now IPv6. 1510 */ 1511 u_char tcp_saveipgen[IP6_HDR_LEN]; 1512 struct tcphdr tcp_savetcp; 1513 short ostate = 0; 1514 #endif 1515 thflags = th->th_flags; 1516 inc = &tp->t_inpcb->inp_inc; 1517 tp->sackhint.last_sack_ack = 0; 1518 sack_changed = 0; 1519 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1520 1521 NET_EPOCH_ASSERT(); 1522 INP_WLOCK_ASSERT(tp->t_inpcb); 1523 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1524 __func__)); 1525 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1526 __func__)); 1527 1528 #ifdef TCPPCAP 1529 /* Save segment, if requested. */ 1530 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1531 #endif 1532 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1533 tlen, NULL, true); 1534 1535 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1536 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1537 log(LOG_DEBUG, "%s; %s: " 1538 "SYN|FIN segment ignored (based on " 1539 "sysctl setting)\n", s, __func__); 1540 free(s, M_TCPLOG); 1541 } 1542 goto drop; 1543 } 1544 1545 /* 1546 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1547 * check SEQ.ACK first. 1548 */ 1549 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1550 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1551 rstreason = BANDLIM_UNLIMITED; 1552 goto dropwithreset; 1553 } 1554 1555 /* 1556 * Segment received on connection. 1557 * Reset idle time and keep-alive timer. 1558 * XXX: This should be done after segment 1559 * validation to ignore broken/spoofed segs. 1560 */ 1561 tp->t_rcvtime = ticks; 1562 1563 /* 1564 * Scale up the window into a 32-bit value. 1565 * For the SYN_SENT state the scale is zero. 1566 */ 1567 tiwin = th->th_win << tp->snd_scale; 1568 #ifdef STATS 1569 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1570 #endif 1571 1572 /* 1573 * TCP ECN processing. 1574 */ 1575 if (tp->t_flags2 & TF2_ECN_PERMIT) { 1576 if (thflags & TH_CWR) { 1577 tp->t_flags2 &= ~TF2_ECN_SND_ECE; 1578 tp->t_flags |= TF_ACKNOW; 1579 } 1580 switch (iptos & IPTOS_ECN_MASK) { 1581 case IPTOS_ECN_CE: 1582 tp->t_flags2 |= TF2_ECN_SND_ECE; 1583 TCPSTAT_INC(tcps_ecn_ce); 1584 break; 1585 case IPTOS_ECN_ECT0: 1586 TCPSTAT_INC(tcps_ecn_ect0); 1587 break; 1588 case IPTOS_ECN_ECT1: 1589 TCPSTAT_INC(tcps_ecn_ect1); 1590 break; 1591 } 1592 1593 /* Process a packet differently from RFC3168. */ 1594 cc_ecnpkt_handler(tp, th, iptos); 1595 1596 /* Congestion experienced. */ 1597 if (thflags & TH_ECE) { 1598 cc_cong_signal(tp, th, CC_ECN); 1599 } 1600 } 1601 1602 /* 1603 * Parse options on any incoming segment. 1604 */ 1605 tcp_dooptions(&to, (u_char *)(th + 1), 1606 (th->th_off << 2) - sizeof(struct tcphdr), 1607 (thflags & TH_SYN) ? TO_SYN : 0); 1608 1609 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1610 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1611 (to.to_flags & TOF_SIGNATURE) == 0) { 1612 TCPSTAT_INC(tcps_sig_err_sigopt); 1613 /* XXX: should drop? */ 1614 } 1615 #endif 1616 /* 1617 * If echoed timestamp is later than the current time, 1618 * fall back to non RFC1323 RTT calculation. Normalize 1619 * timestamp if syncookies were used when this connection 1620 * was established. 1621 */ 1622 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1623 to.to_tsecr -= tp->ts_offset; 1624 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1625 to.to_tsecr = 0; 1626 else if (tp->t_flags & TF_PREVVALID && 1627 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1628 cc_cong_signal(tp, th, CC_RTO_ERR); 1629 } 1630 /* 1631 * Process options only when we get SYN/ACK back. The SYN case 1632 * for incoming connections is handled in tcp_syncache. 1633 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1634 * or <SYN,ACK>) segment itself is never scaled. 1635 * XXX this is traditional behavior, may need to be cleaned up. 1636 */ 1637 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1638 /* Handle parallel SYN for ECN */ 1639 if (!(thflags & TH_ACK) && 1640 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) && 1641 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) { 1642 tp->t_flags2 |= TF2_ECN_PERMIT; 1643 tp->t_flags2 |= TF2_ECN_SND_ECE; 1644 TCPSTAT_INC(tcps_ecn_shs); 1645 } 1646 if ((to.to_flags & TOF_SCALE) && 1647 (tp->t_flags & TF_REQ_SCALE) && 1648 !(tp->t_flags & TF_NOOPT)) { 1649 tp->t_flags |= TF_RCVD_SCALE; 1650 tp->snd_scale = to.to_wscale; 1651 } else 1652 tp->t_flags &= ~TF_REQ_SCALE; 1653 /* 1654 * Initial send window. It will be updated with 1655 * the next incoming segment to the scaled value. 1656 */ 1657 tp->snd_wnd = th->th_win; 1658 if ((to.to_flags & TOF_TS) && 1659 (tp->t_flags & TF_REQ_TSTMP) && 1660 !(tp->t_flags & TF_NOOPT)) { 1661 tp->t_flags |= TF_RCVD_TSTMP; 1662 tp->ts_recent = to.to_tsval; 1663 tp->ts_recent_age = tcp_ts_getticks(); 1664 } else 1665 tp->t_flags &= ~TF_REQ_TSTMP; 1666 if (to.to_flags & TOF_MSS) 1667 tcp_mss(tp, to.to_mss); 1668 if ((tp->t_flags & TF_SACK_PERMIT) && 1669 (!(to.to_flags & TOF_SACKPERM) || 1670 (tp->t_flags & TF_NOOPT))) 1671 tp->t_flags &= ~TF_SACK_PERMIT; 1672 if (IS_FASTOPEN(tp->t_flags)) { 1673 if ((to.to_flags & TOF_FASTOPEN) && 1674 !(tp->t_flags & TF_NOOPT)) { 1675 uint16_t mss; 1676 1677 if (to.to_flags & TOF_MSS) 1678 mss = to.to_mss; 1679 else 1680 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1681 mss = TCP6_MSS; 1682 else 1683 mss = TCP_MSS; 1684 tcp_fastopen_update_cache(tp, mss, 1685 to.to_tfo_len, to.to_tfo_cookie); 1686 } else 1687 tcp_fastopen_disable_path(tp); 1688 } 1689 } 1690 1691 /* 1692 * If timestamps were negotiated during SYN/ACK and a 1693 * segment without a timestamp is received, silently drop 1694 * the segment, unless it is a RST segment or missing timestamps are 1695 * tolerated. 1696 * See section 3.2 of RFC 7323. 1697 */ 1698 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1699 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1700 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1701 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1702 "segment processed normally\n", 1703 s, __func__); 1704 free(s, M_TCPLOG); 1705 } 1706 } else { 1707 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1708 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1709 "segment silently dropped\n", s, __func__); 1710 free(s, M_TCPLOG); 1711 } 1712 goto drop; 1713 } 1714 } 1715 /* 1716 * If timestamps were not negotiated during SYN/ACK and a 1717 * segment with a timestamp is received, ignore the 1718 * timestamp and process the packet normally. 1719 * See section 3.2 of RFC 7323. 1720 */ 1721 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1722 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1723 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1724 "segment processed normally\n", s, __func__); 1725 free(s, M_TCPLOG); 1726 } 1727 } 1728 1729 /* 1730 * Header prediction: check for the two common cases 1731 * of a uni-directional data xfer. If the packet has 1732 * no control flags, is in-sequence, the window didn't 1733 * change and we're not retransmitting, it's a 1734 * candidate. If the length is zero and the ack moved 1735 * forward, we're the sender side of the xfer. Just 1736 * free the data acked & wake any higher level process 1737 * that was blocked waiting for space. If the length 1738 * is non-zero and the ack didn't move, we're the 1739 * receiver side. If we're getting packets in-order 1740 * (the reassembly queue is empty), add the data to 1741 * the socket buffer and note that we need a delayed ack. 1742 * Make sure that the hidden state-flags are also off. 1743 * Since we check for TCPS_ESTABLISHED first, it can only 1744 * be TH_NEEDSYN. 1745 */ 1746 if (tp->t_state == TCPS_ESTABLISHED && 1747 th->th_seq == tp->rcv_nxt && 1748 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1749 tp->snd_nxt == tp->snd_max && 1750 tiwin && tiwin == tp->snd_wnd && 1751 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1752 SEGQ_EMPTY(tp) && 1753 ((to.to_flags & TOF_TS) == 0 || 1754 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1755 /* 1756 * If last ACK falls within this segment's sequence numbers, 1757 * record the timestamp. 1758 * NOTE that the test is modified according to the latest 1759 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1760 */ 1761 if ((to.to_flags & TOF_TS) != 0 && 1762 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1763 tp->ts_recent_age = tcp_ts_getticks(); 1764 tp->ts_recent = to.to_tsval; 1765 } 1766 1767 if (tlen == 0) { 1768 if (SEQ_GT(th->th_ack, tp->snd_una) && 1769 SEQ_LEQ(th->th_ack, tp->snd_max) && 1770 !IN_RECOVERY(tp->t_flags) && 1771 (to.to_flags & TOF_SACK) == 0 && 1772 TAILQ_EMPTY(&tp->snd_holes)) { 1773 /* 1774 * This is a pure ack for outstanding data. 1775 */ 1776 TCPSTAT_INC(tcps_predack); 1777 1778 /* 1779 * "bad retransmit" recovery without timestamps. 1780 */ 1781 if ((to.to_flags & TOF_TS) == 0 && 1782 tp->t_rxtshift == 1 && 1783 tp->t_flags & TF_PREVVALID && 1784 (int)(ticks - tp->t_badrxtwin) < 0) { 1785 cc_cong_signal(tp, th, CC_RTO_ERR); 1786 } 1787 1788 /* 1789 * Recalculate the transmit timer / rtt. 1790 * 1791 * Some boxes send broken timestamp replies 1792 * during the SYN+ACK phase, ignore 1793 * timestamps of 0 or we could calculate a 1794 * huge RTT and blow up the retransmit timer. 1795 */ 1796 if ((to.to_flags & TOF_TS) != 0 && 1797 to.to_tsecr) { 1798 uint32_t t; 1799 1800 t = tcp_ts_getticks() - to.to_tsecr; 1801 if (!tp->t_rttlow || tp->t_rttlow > t) 1802 tp->t_rttlow = t; 1803 tcp_xmit_timer(tp, 1804 TCP_TS_TO_TICKS(t) + 1); 1805 } else if (tp->t_rtttime && 1806 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1807 if (!tp->t_rttlow || 1808 tp->t_rttlow > ticks - tp->t_rtttime) 1809 tp->t_rttlow = ticks - tp->t_rtttime; 1810 tcp_xmit_timer(tp, 1811 ticks - tp->t_rtttime); 1812 } 1813 acked = BYTES_THIS_ACK(tp, th); 1814 1815 #ifdef TCP_HHOOK 1816 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1817 hhook_run_tcp_est_in(tp, th, &to); 1818 #endif 1819 1820 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1821 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1822 sbdrop(&so->so_snd, acked); 1823 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1824 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1825 tp->snd_recover = th->th_ack - 1; 1826 1827 /* 1828 * Let the congestion control algorithm update 1829 * congestion control related information. This 1830 * typically means increasing the congestion 1831 * window. 1832 */ 1833 cc_ack_received(tp, th, nsegs, CC_ACK); 1834 1835 tp->snd_una = th->th_ack; 1836 /* 1837 * Pull snd_wl2 up to prevent seq wrap relative 1838 * to th_ack. 1839 */ 1840 tp->snd_wl2 = th->th_ack; 1841 tp->t_dupacks = 0; 1842 m_freem(m); 1843 1844 /* 1845 * If all outstanding data are acked, stop 1846 * retransmit timer, otherwise restart timer 1847 * using current (possibly backed-off) value. 1848 * If process is waiting for space, 1849 * wakeup/selwakeup/signal. If data 1850 * are ready to send, let tcp_output 1851 * decide between more output or persist. 1852 */ 1853 #ifdef TCPDEBUG 1854 if (so->so_options & SO_DEBUG) 1855 tcp_trace(TA_INPUT, ostate, tp, 1856 (void *)tcp_saveipgen, 1857 &tcp_savetcp, 0); 1858 #endif 1859 TCP_PROBE3(debug__input, tp, th, m); 1860 if (tp->snd_una == tp->snd_max) 1861 tcp_timer_activate(tp, TT_REXMT, 0); 1862 else if (!tcp_timer_active(tp, TT_PERSIST)) 1863 tcp_timer_activate(tp, TT_REXMT, 1864 tp->t_rxtcur); 1865 tp->t_flags |= TF_WAKESOW; 1866 if (sbavail(&so->so_snd)) 1867 (void) tp->t_fb->tfb_tcp_output(tp); 1868 goto check_delack; 1869 } 1870 } else if (th->th_ack == tp->snd_una && 1871 tlen <= sbspace(&so->so_rcv)) { 1872 int newsize = 0; /* automatic sockbuf scaling */ 1873 1874 /* 1875 * This is a pure, in-sequence data packet with 1876 * nothing on the reassembly queue and we have enough 1877 * buffer space to take it. 1878 */ 1879 /* Clean receiver SACK report if present */ 1880 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1881 tcp_clean_sackreport(tp); 1882 TCPSTAT_INC(tcps_preddat); 1883 tp->rcv_nxt += tlen; 1884 if (tlen && 1885 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1886 (tp->t_fbyte_in == 0)) { 1887 tp->t_fbyte_in = ticks; 1888 if (tp->t_fbyte_in == 0) 1889 tp->t_fbyte_in = 1; 1890 if (tp->t_fbyte_out && tp->t_fbyte_in) 1891 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1892 } 1893 /* 1894 * Pull snd_wl1 up to prevent seq wrap relative to 1895 * th_seq. 1896 */ 1897 tp->snd_wl1 = th->th_seq; 1898 /* 1899 * Pull rcv_up up to prevent seq wrap relative to 1900 * rcv_nxt. 1901 */ 1902 tp->rcv_up = tp->rcv_nxt; 1903 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1904 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1905 #ifdef TCPDEBUG 1906 if (so->so_options & SO_DEBUG) 1907 tcp_trace(TA_INPUT, ostate, tp, 1908 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1909 #endif 1910 TCP_PROBE3(debug__input, tp, th, m); 1911 1912 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1913 1914 /* Add data to socket buffer. */ 1915 SOCKBUF_LOCK(&so->so_rcv); 1916 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1917 m_freem(m); 1918 } else { 1919 /* 1920 * Set new socket buffer size. 1921 * Give up when limit is reached. 1922 */ 1923 if (newsize) 1924 if (!sbreserve_locked(&so->so_rcv, 1925 newsize, so, NULL)) 1926 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1927 m_adj(m, drop_hdrlen); /* delayed header drop */ 1928 sbappendstream_locked(&so->so_rcv, m, 0); 1929 } 1930 SOCKBUF_UNLOCK(&so->so_rcv); 1931 tp->t_flags |= TF_WAKESOR; 1932 if (DELAY_ACK(tp, tlen)) { 1933 tp->t_flags |= TF_DELACK; 1934 } else { 1935 tp->t_flags |= TF_ACKNOW; 1936 tp->t_fb->tfb_tcp_output(tp); 1937 } 1938 goto check_delack; 1939 } 1940 } 1941 1942 /* 1943 * Calculate amount of space in receive window, 1944 * and then do TCP input processing. 1945 * Receive window is amount of space in rcv queue, 1946 * but not less than advertised window. 1947 */ 1948 win = sbspace(&so->so_rcv); 1949 if (win < 0) 1950 win = 0; 1951 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1952 1953 switch (tp->t_state) { 1954 /* 1955 * If the state is SYN_RECEIVED: 1956 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1957 */ 1958 case TCPS_SYN_RECEIVED: 1959 if ((thflags & TH_ACK) && 1960 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1961 SEQ_GT(th->th_ack, tp->snd_max))) { 1962 rstreason = BANDLIM_RST_OPENPORT; 1963 goto dropwithreset; 1964 } 1965 if (IS_FASTOPEN(tp->t_flags)) { 1966 /* 1967 * When a TFO connection is in SYN_RECEIVED, the 1968 * only valid packets are the initial SYN, a 1969 * retransmit/copy of the initial SYN (possibly with 1970 * a subset of the original data), a valid ACK, a 1971 * FIN, or a RST. 1972 */ 1973 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1974 rstreason = BANDLIM_RST_OPENPORT; 1975 goto dropwithreset; 1976 } else if (thflags & TH_SYN) { 1977 /* non-initial SYN is ignored */ 1978 if ((tcp_timer_active(tp, TT_DELACK) || 1979 tcp_timer_active(tp, TT_REXMT))) 1980 goto drop; 1981 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1982 goto drop; 1983 } 1984 } 1985 break; 1986 1987 /* 1988 * If the state is SYN_SENT: 1989 * if seg contains a RST with valid ACK (SEQ.ACK has already 1990 * been verified), then drop the connection. 1991 * if seg contains a RST without an ACK, drop the seg. 1992 * if seg does not contain SYN, then drop the seg. 1993 * Otherwise this is an acceptable SYN segment 1994 * initialize tp->rcv_nxt and tp->irs 1995 * if seg contains ack then advance tp->snd_una 1996 * if seg contains an ECE and ECN support is enabled, the stream 1997 * is ECN capable. 1998 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1999 * arrange for segment to be acked (eventually) 2000 * continue processing rest of data/controls, beginning with URG 2001 */ 2002 case TCPS_SYN_SENT: 2003 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2004 TCP_PROBE5(connect__refused, NULL, tp, 2005 m, tp, th); 2006 tp = tcp_drop(tp, ECONNREFUSED); 2007 } 2008 if (thflags & TH_RST) 2009 goto drop; 2010 if (!(thflags & TH_SYN)) 2011 goto drop; 2012 2013 tp->irs = th->th_seq; 2014 tcp_rcvseqinit(tp); 2015 if (thflags & TH_ACK) { 2016 int tfo_partial_ack = 0; 2017 2018 TCPSTAT_INC(tcps_connects); 2019 soisconnected(so); 2020 #ifdef MAC 2021 mac_socketpeer_set_from_mbuf(m, so); 2022 #endif 2023 /* Do window scaling on this connection? */ 2024 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2025 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2026 tp->rcv_scale = tp->request_r_scale; 2027 } 2028 tp->rcv_adv += min(tp->rcv_wnd, 2029 TCP_MAXWIN << tp->rcv_scale); 2030 tp->snd_una++; /* SYN is acked */ 2031 /* 2032 * If not all the data that was sent in the TFO SYN 2033 * has been acked, resend the remainder right away. 2034 */ 2035 if (IS_FASTOPEN(tp->t_flags) && 2036 (tp->snd_una != tp->snd_max)) { 2037 tp->snd_nxt = th->th_ack; 2038 tfo_partial_ack = 1; 2039 } 2040 /* 2041 * If there's data, delay ACK; if there's also a FIN 2042 * ACKNOW will be turned on later. 2043 */ 2044 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2045 tcp_timer_activate(tp, TT_DELACK, 2046 tcp_delacktime); 2047 else 2048 tp->t_flags |= TF_ACKNOW; 2049 2050 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2051 (V_tcp_do_ecn == 1)) { 2052 tp->t_flags2 |= TF2_ECN_PERMIT; 2053 TCPSTAT_INC(tcps_ecn_shs); 2054 } 2055 2056 /* 2057 * Received <SYN,ACK> in SYN_SENT[*] state. 2058 * Transitions: 2059 * SYN_SENT --> ESTABLISHED 2060 * SYN_SENT* --> FIN_WAIT_1 2061 */ 2062 tp->t_starttime = ticks; 2063 if (tp->t_flags & TF_NEEDFIN) { 2064 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2065 tp->t_flags &= ~TF_NEEDFIN; 2066 thflags &= ~TH_SYN; 2067 } else { 2068 tcp_state_change(tp, TCPS_ESTABLISHED); 2069 TCP_PROBE5(connect__established, NULL, tp, 2070 m, tp, th); 2071 cc_conn_init(tp); 2072 tcp_timer_activate(tp, TT_KEEP, 2073 TP_KEEPIDLE(tp)); 2074 } 2075 } else { 2076 /* 2077 * Received initial SYN in SYN-SENT[*] state => 2078 * simultaneous open. 2079 * If it succeeds, connection is * half-synchronized. 2080 * Otherwise, do 3-way handshake: 2081 * SYN-SENT -> SYN-RECEIVED 2082 * SYN-SENT* -> SYN-RECEIVED* 2083 */ 2084 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2085 tcp_timer_activate(tp, TT_REXMT, 0); 2086 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2087 } 2088 2089 INP_WLOCK_ASSERT(tp->t_inpcb); 2090 2091 /* 2092 * Advance th->th_seq to correspond to first data byte. 2093 * If data, trim to stay within window, 2094 * dropping FIN if necessary. 2095 */ 2096 th->th_seq++; 2097 if (tlen > tp->rcv_wnd) { 2098 todrop = tlen - tp->rcv_wnd; 2099 m_adj(m, -todrop); 2100 tlen = tp->rcv_wnd; 2101 thflags &= ~TH_FIN; 2102 TCPSTAT_INC(tcps_rcvpackafterwin); 2103 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2104 } 2105 tp->snd_wl1 = th->th_seq - 1; 2106 tp->rcv_up = th->th_seq; 2107 /* 2108 * Client side of transaction: already sent SYN and data. 2109 * If the remote host used T/TCP to validate the SYN, 2110 * our data will be ACK'd; if so, enter normal data segment 2111 * processing in the middle of step 5, ack processing. 2112 * Otherwise, goto step 6. 2113 */ 2114 if (thflags & TH_ACK) 2115 goto process_ACK; 2116 2117 goto step6; 2118 2119 /* 2120 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2121 * do normal processing. 2122 * 2123 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2124 */ 2125 case TCPS_LAST_ACK: 2126 case TCPS_CLOSING: 2127 break; /* continue normal processing */ 2128 } 2129 2130 /* 2131 * States other than LISTEN or SYN_SENT. 2132 * First check the RST flag and sequence number since reset segments 2133 * are exempt from the timestamp and connection count tests. This 2134 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2135 * below which allowed reset segments in half the sequence space 2136 * to fall though and be processed (which gives forged reset 2137 * segments with a random sequence number a 50 percent chance of 2138 * killing a connection). 2139 * Then check timestamp, if present. 2140 * Then check the connection count, if present. 2141 * Then check that at least some bytes of segment are within 2142 * receive window. If segment begins before rcv_nxt, 2143 * drop leading data (and SYN); if nothing left, just ack. 2144 */ 2145 if (thflags & TH_RST) { 2146 /* 2147 * RFC5961 Section 3.2 2148 * 2149 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2150 * - If RST is in window, we send challenge ACK. 2151 * 2152 * Note: to take into account delayed ACKs, we should 2153 * test against last_ack_sent instead of rcv_nxt. 2154 * Note 2: we handle special case of closed window, not 2155 * covered by the RFC. 2156 */ 2157 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2158 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2159 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2160 KASSERT(tp->t_state != TCPS_SYN_SENT, 2161 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2162 __func__, th, tp)); 2163 2164 if (V_tcp_insecure_rst || 2165 tp->last_ack_sent == th->th_seq) { 2166 TCPSTAT_INC(tcps_drops); 2167 /* Drop the connection. */ 2168 switch (tp->t_state) { 2169 case TCPS_SYN_RECEIVED: 2170 so->so_error = ECONNREFUSED; 2171 goto close; 2172 case TCPS_ESTABLISHED: 2173 case TCPS_FIN_WAIT_1: 2174 case TCPS_FIN_WAIT_2: 2175 case TCPS_CLOSE_WAIT: 2176 case TCPS_CLOSING: 2177 case TCPS_LAST_ACK: 2178 so->so_error = ECONNRESET; 2179 close: 2180 /* FALLTHROUGH */ 2181 default: 2182 tp = tcp_close(tp); 2183 } 2184 } else { 2185 TCPSTAT_INC(tcps_badrst); 2186 /* Send challenge ACK. */ 2187 tcp_respond(tp, mtod(m, void *), th, m, 2188 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2189 tp->last_ack_sent = tp->rcv_nxt; 2190 m = NULL; 2191 } 2192 } 2193 goto drop; 2194 } 2195 2196 /* 2197 * RFC5961 Section 4.2 2198 * Send challenge ACK for any SYN in synchronized state. 2199 */ 2200 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2201 tp->t_state != TCPS_SYN_RECEIVED) { 2202 TCPSTAT_INC(tcps_badsyn); 2203 if (V_tcp_insecure_syn && 2204 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2205 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2206 tp = tcp_drop(tp, ECONNRESET); 2207 rstreason = BANDLIM_UNLIMITED; 2208 } else { 2209 /* Send challenge ACK. */ 2210 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2211 tp->snd_nxt, TH_ACK); 2212 tp->last_ack_sent = tp->rcv_nxt; 2213 m = NULL; 2214 } 2215 goto drop; 2216 } 2217 2218 /* 2219 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2220 * and it's less than ts_recent, drop it. 2221 */ 2222 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2223 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2224 /* Check to see if ts_recent is over 24 days old. */ 2225 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2226 /* 2227 * Invalidate ts_recent. If this segment updates 2228 * ts_recent, the age will be reset later and ts_recent 2229 * will get a valid value. If it does not, setting 2230 * ts_recent to zero will at least satisfy the 2231 * requirement that zero be placed in the timestamp 2232 * echo reply when ts_recent isn't valid. The 2233 * age isn't reset until we get a valid ts_recent 2234 * because we don't want out-of-order segments to be 2235 * dropped when ts_recent is old. 2236 */ 2237 tp->ts_recent = 0; 2238 } else { 2239 TCPSTAT_INC(tcps_rcvduppack); 2240 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2241 TCPSTAT_INC(tcps_pawsdrop); 2242 if (tlen) 2243 goto dropafterack; 2244 goto drop; 2245 } 2246 } 2247 2248 /* 2249 * In the SYN-RECEIVED state, validate that the packet belongs to 2250 * this connection before trimming the data to fit the receive 2251 * window. Check the sequence number versus IRS since we know 2252 * the sequence numbers haven't wrapped. This is a partial fix 2253 * for the "LAND" DoS attack. 2254 */ 2255 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2256 rstreason = BANDLIM_RST_OPENPORT; 2257 goto dropwithreset; 2258 } 2259 2260 todrop = tp->rcv_nxt - th->th_seq; 2261 if (todrop > 0) { 2262 if (thflags & TH_SYN) { 2263 thflags &= ~TH_SYN; 2264 th->th_seq++; 2265 if (th->th_urp > 1) 2266 th->th_urp--; 2267 else 2268 thflags &= ~TH_URG; 2269 todrop--; 2270 } 2271 /* 2272 * Following if statement from Stevens, vol. 2, p. 960. 2273 */ 2274 if (todrop > tlen 2275 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2276 /* 2277 * Any valid FIN must be to the left of the window. 2278 * At this point the FIN must be a duplicate or out 2279 * of sequence; drop it. 2280 */ 2281 thflags &= ~TH_FIN; 2282 2283 /* 2284 * Send an ACK to resynchronize and drop any data. 2285 * But keep on processing for RST or ACK. 2286 */ 2287 tp->t_flags |= TF_ACKNOW; 2288 todrop = tlen; 2289 TCPSTAT_INC(tcps_rcvduppack); 2290 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2291 } else { 2292 TCPSTAT_INC(tcps_rcvpartduppack); 2293 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2294 } 2295 /* 2296 * DSACK - add SACK block for dropped range 2297 */ 2298 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2299 tcp_update_sack_list(tp, th->th_seq, 2300 th->th_seq + todrop); 2301 /* 2302 * ACK now, as the next in-sequence segment 2303 * will clear the DSACK block again 2304 */ 2305 tp->t_flags |= TF_ACKNOW; 2306 } 2307 drop_hdrlen += todrop; /* drop from the top afterwards */ 2308 th->th_seq += todrop; 2309 tlen -= todrop; 2310 if (th->th_urp > todrop) 2311 th->th_urp -= todrop; 2312 else { 2313 thflags &= ~TH_URG; 2314 th->th_urp = 0; 2315 } 2316 } 2317 2318 /* 2319 * If new data are received on a connection after the 2320 * user processes are gone, then RST the other end. 2321 */ 2322 if ((so->so_state & SS_NOFDREF) && 2323 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2324 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2325 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2326 "after socket was closed, " 2327 "sending RST and removing tcpcb\n", 2328 s, __func__, tcpstates[tp->t_state], tlen); 2329 free(s, M_TCPLOG); 2330 } 2331 tp = tcp_close(tp); 2332 TCPSTAT_INC(tcps_rcvafterclose); 2333 rstreason = BANDLIM_UNLIMITED; 2334 goto dropwithreset; 2335 } 2336 2337 /* 2338 * If segment ends after window, drop trailing data 2339 * (and PUSH and FIN); if nothing left, just ACK. 2340 */ 2341 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2342 if (todrop > 0) { 2343 TCPSTAT_INC(tcps_rcvpackafterwin); 2344 if (todrop >= tlen) { 2345 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2346 /* 2347 * If window is closed can only take segments at 2348 * window edge, and have to drop data and PUSH from 2349 * incoming segments. Continue processing, but 2350 * remember to ack. Otherwise, drop segment 2351 * and ack. 2352 */ 2353 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2354 tp->t_flags |= TF_ACKNOW; 2355 TCPSTAT_INC(tcps_rcvwinprobe); 2356 } else 2357 goto dropafterack; 2358 } else 2359 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2360 m_adj(m, -todrop); 2361 tlen -= todrop; 2362 thflags &= ~(TH_PUSH|TH_FIN); 2363 } 2364 2365 /* 2366 * If last ACK falls within this segment's sequence numbers, 2367 * record its timestamp. 2368 * NOTE: 2369 * 1) That the test incorporates suggestions from the latest 2370 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2371 * 2) That updating only on newer timestamps interferes with 2372 * our earlier PAWS tests, so this check should be solely 2373 * predicated on the sequence space of this segment. 2374 * 3) That we modify the segment boundary check to be 2375 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2376 * instead of RFC1323's 2377 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2378 * This modified check allows us to overcome RFC1323's 2379 * limitations as described in Stevens TCP/IP Illustrated 2380 * Vol. 2 p.869. In such cases, we can still calculate the 2381 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2382 */ 2383 if ((to.to_flags & TOF_TS) != 0 && 2384 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2385 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2386 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2387 tp->ts_recent_age = tcp_ts_getticks(); 2388 tp->ts_recent = to.to_tsval; 2389 } 2390 2391 /* 2392 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2393 * flag is on (half-synchronized state), then queue data for 2394 * later processing; else drop segment and return. 2395 */ 2396 if ((thflags & TH_ACK) == 0) { 2397 if (tp->t_state == TCPS_SYN_RECEIVED || 2398 (tp->t_flags & TF_NEEDSYN)) { 2399 if (tp->t_state == TCPS_SYN_RECEIVED && 2400 IS_FASTOPEN(tp->t_flags)) { 2401 tp->snd_wnd = tiwin; 2402 cc_conn_init(tp); 2403 } 2404 goto step6; 2405 } else if (tp->t_flags & TF_ACKNOW) 2406 goto dropafterack; 2407 else 2408 goto drop; 2409 } 2410 2411 /* 2412 * Ack processing. 2413 */ 2414 switch (tp->t_state) { 2415 /* 2416 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2417 * ESTABLISHED state and continue processing. 2418 * The ACK was checked above. 2419 */ 2420 case TCPS_SYN_RECEIVED: 2421 2422 TCPSTAT_INC(tcps_connects); 2423 soisconnected(so); 2424 /* Do window scaling? */ 2425 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2426 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2427 tp->rcv_scale = tp->request_r_scale; 2428 } 2429 tp->snd_wnd = tiwin; 2430 /* 2431 * Make transitions: 2432 * SYN-RECEIVED -> ESTABLISHED 2433 * SYN-RECEIVED* -> FIN-WAIT-1 2434 */ 2435 tp->t_starttime = ticks; 2436 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2437 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2438 tp->t_tfo_pending = NULL; 2439 } 2440 if (tp->t_flags & TF_NEEDFIN) { 2441 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2442 tp->t_flags &= ~TF_NEEDFIN; 2443 } else { 2444 tcp_state_change(tp, TCPS_ESTABLISHED); 2445 TCP_PROBE5(accept__established, NULL, tp, 2446 m, tp, th); 2447 /* 2448 * TFO connections call cc_conn_init() during SYN 2449 * processing. Calling it again here for such 2450 * connections is not harmless as it would undo the 2451 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2452 * is retransmitted. 2453 */ 2454 if (!IS_FASTOPEN(tp->t_flags)) 2455 cc_conn_init(tp); 2456 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2457 } 2458 /* 2459 * Account for the ACK of our SYN prior to 2460 * regular ACK processing below, except for 2461 * simultaneous SYN, which is handled later. 2462 */ 2463 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2464 incforsyn = 1; 2465 /* 2466 * If segment contains data or ACK, will call tcp_reass() 2467 * later; if not, do so now to pass queued data to user. 2468 */ 2469 if (tlen == 0 && (thflags & TH_FIN) == 0) 2470 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2471 (struct mbuf *)0); 2472 tp->snd_wl1 = th->th_seq - 1; 2473 /* FALLTHROUGH */ 2474 2475 /* 2476 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2477 * ACKs. If the ack is in the range 2478 * tp->snd_una < th->th_ack <= tp->snd_max 2479 * then advance tp->snd_una to th->th_ack and drop 2480 * data from the retransmission queue. If this ACK reflects 2481 * more up to date window information we update our window information. 2482 */ 2483 case TCPS_ESTABLISHED: 2484 case TCPS_FIN_WAIT_1: 2485 case TCPS_FIN_WAIT_2: 2486 case TCPS_CLOSE_WAIT: 2487 case TCPS_CLOSING: 2488 case TCPS_LAST_ACK: 2489 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2490 TCPSTAT_INC(tcps_rcvacktoomuch); 2491 goto dropafterack; 2492 } 2493 if ((tp->t_flags & TF_SACK_PERMIT) && 2494 ((to.to_flags & TOF_SACK) || 2495 !TAILQ_EMPTY(&tp->snd_holes))) 2496 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2497 else 2498 /* 2499 * Reset the value so that previous (valid) value 2500 * from the last ack with SACK doesn't get used. 2501 */ 2502 tp->sackhint.sacked_bytes = 0; 2503 2504 #ifdef TCP_HHOOK 2505 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2506 hhook_run_tcp_est_in(tp, th, &to); 2507 #endif 2508 2509 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2510 maxseg = tcp_maxseg(tp); 2511 if (tlen == 0 && 2512 (tiwin == tp->snd_wnd || 2513 (tp->t_flags & TF_SACK_PERMIT))) { 2514 /* 2515 * If this is the first time we've seen a 2516 * FIN from the remote, this is not a 2517 * duplicate and it needs to be processed 2518 * normally. This happens during a 2519 * simultaneous close. 2520 */ 2521 if ((thflags & TH_FIN) && 2522 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2523 tp->t_dupacks = 0; 2524 break; 2525 } 2526 TCPSTAT_INC(tcps_rcvdupack); 2527 /* 2528 * If we have outstanding data (other than 2529 * a window probe), this is a completely 2530 * duplicate ack (ie, window info didn't 2531 * change and FIN isn't set), 2532 * the ack is the biggest we've 2533 * seen and we've seen exactly our rexmt 2534 * threshold of them, assume a packet 2535 * has been dropped and retransmit it. 2536 * Kludge snd_nxt & the congestion 2537 * window so we send only this one 2538 * packet. 2539 * 2540 * We know we're losing at the current 2541 * window size so do congestion avoidance 2542 * (set ssthresh to half the current window 2543 * and pull our congestion window back to 2544 * the new ssthresh). 2545 * 2546 * Dup acks mean that packets have left the 2547 * network (they're now cached at the receiver) 2548 * so bump cwnd by the amount in the receiver 2549 * to keep a constant cwnd packets in the 2550 * network. 2551 * 2552 * When using TCP ECN, notify the peer that 2553 * we reduced the cwnd. 2554 */ 2555 /* 2556 * Following 2 kinds of acks should not affect 2557 * dupack counting: 2558 * 1) Old acks 2559 * 2) Acks with SACK but without any new SACK 2560 * information in them. These could result from 2561 * any anomaly in the network like a switch 2562 * duplicating packets or a possible DoS attack. 2563 */ 2564 if (th->th_ack != tp->snd_una || 2565 ((tp->t_flags & TF_SACK_PERMIT) && 2566 !sack_changed)) 2567 break; 2568 else if (!tcp_timer_active(tp, TT_REXMT)) 2569 tp->t_dupacks = 0; 2570 else if (++tp->t_dupacks > tcprexmtthresh || 2571 IN_FASTRECOVERY(tp->t_flags)) { 2572 cc_ack_received(tp, th, nsegs, 2573 CC_DUPACK); 2574 if (V_tcp_do_prr && 2575 IN_FASTRECOVERY(tp->t_flags) && 2576 (tp->t_flags & TF_SACK_PERMIT)) { 2577 int snd_cnt = 0, limit = 0; 2578 int del_data = 0, pipe = 0; 2579 /* 2580 * In a duplicate ACK del_data is only the 2581 * diff_in_sack. If no SACK is used del_data 2582 * will be 0. Pipe is the amount of data we 2583 * estimate to be in the network. 2584 */ 2585 del_data = tp->sackhint.delivered_data; 2586 if (V_tcp_do_rfc6675_pipe) 2587 pipe = tcp_compute_pipe(tp); 2588 else 2589 pipe = (tp->snd_nxt - tp->snd_fack) + 2590 tp->sackhint.sack_bytes_rexmit; 2591 tp->sackhint.prr_delivered += del_data; 2592 if (pipe >= tp->snd_ssthresh) { 2593 if (tp->sackhint.recover_fs == 0) 2594 tp->sackhint.recover_fs = 2595 imax(1, tp->snd_nxt - tp->snd_una); 2596 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 2597 tp->snd_ssthresh, tp->sackhint.recover_fs) - 2598 (tp->sackhint.sack_bytes_rexmit + 2599 (tp->snd_nxt - tp->snd_recover)); 2600 } else { 2601 if (V_tcp_do_prr_conservative) 2602 limit = tp->sackhint.prr_delivered - 2603 (tp->sackhint.sack_bytes_rexmit + 2604 (tp->snd_nxt - tp->snd_recover)); 2605 else 2606 limit = imax(tp->sackhint.prr_delivered - 2607 (tp->sackhint.sack_bytes_rexmit + 2608 (tp->snd_nxt - tp->snd_recover)), 2609 del_data) + maxseg; 2610 snd_cnt = imin(tp->snd_ssthresh - pipe, limit); 2611 } 2612 snd_cnt = imax(snd_cnt, 0) / maxseg; 2613 /* 2614 * Send snd_cnt new data into the network in 2615 * response to this ACK. If there is a going 2616 * to be a SACK retransmission, adjust snd_cwnd 2617 * accordingly. 2618 */ 2619 tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover + 2620 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg)); 2621 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2622 IN_FASTRECOVERY(tp->t_flags)) { 2623 int awnd; 2624 2625 /* 2626 * Compute the amount of data in flight first. 2627 * We can inject new data into the pipe iff 2628 * we have less than 1/2 the original window's 2629 * worth of data in flight. 2630 */ 2631 if (V_tcp_do_rfc6675_pipe) 2632 awnd = tcp_compute_pipe(tp); 2633 else 2634 awnd = (tp->snd_nxt - tp->snd_fack) + 2635 tp->sackhint.sack_bytes_rexmit; 2636 2637 if (awnd < tp->snd_ssthresh) { 2638 tp->snd_cwnd += maxseg; 2639 if (tp->snd_cwnd > tp->snd_ssthresh) 2640 tp->snd_cwnd = tp->snd_ssthresh; 2641 } 2642 } else 2643 tp->snd_cwnd += maxseg; 2644 (void) tp->t_fb->tfb_tcp_output(tp); 2645 goto drop; 2646 } else if (tp->t_dupacks == tcprexmtthresh || 2647 (tp->t_flags & TF_SACK_PERMIT && 2648 V_tcp_do_rfc6675_pipe && 2649 tp->sackhint.sacked_bytes > 2650 (tcprexmtthresh - 1) * maxseg)) { 2651 enter_recovery: 2652 /* 2653 * Above is the RFC6675 trigger condition of 2654 * more than (dupthresh-1)*maxseg sacked data. 2655 * If the count of holes in the 2656 * scoreboard is >= dupthresh, we could 2657 * also enter loss recovery, but don't 2658 * have that value readily available. 2659 */ 2660 tp->t_dupacks = tcprexmtthresh; 2661 tcp_seq onxt = tp->snd_nxt; 2662 2663 /* 2664 * If we're doing sack, or prr, check 2665 * to see if we're already in sack 2666 * recovery. If we're not doing sack, 2667 * check to see if we're in newreno 2668 * recovery. 2669 */ 2670 if (V_tcp_do_prr || 2671 (tp->t_flags & TF_SACK_PERMIT)) { 2672 if (IN_FASTRECOVERY(tp->t_flags)) { 2673 tp->t_dupacks = 0; 2674 break; 2675 } 2676 } else { 2677 if (SEQ_LEQ(th->th_ack, 2678 tp->snd_recover)) { 2679 tp->t_dupacks = 0; 2680 break; 2681 } 2682 } 2683 /* Congestion signal before ack. */ 2684 cc_cong_signal(tp, th, CC_NDUPACK); 2685 cc_ack_received(tp, th, nsegs, 2686 CC_DUPACK); 2687 tcp_timer_activate(tp, TT_REXMT, 0); 2688 tp->t_rtttime = 0; 2689 if (V_tcp_do_prr) { 2690 /* 2691 * snd_ssthresh is already updated by 2692 * cc_cong_signal. 2693 */ 2694 tp->sackhint.prr_delivered = 2695 tp->sackhint.sacked_bytes; 2696 tp->sackhint.sack_bytes_rexmit = 0; 2697 tp->sackhint.recover_fs = max(1, 2698 tp->snd_nxt - tp->snd_una); 2699 } 2700 if (tp->t_flags & TF_SACK_PERMIT) { 2701 TCPSTAT_INC( 2702 tcps_sack_recovery_episode); 2703 tp->snd_recover = tp->snd_nxt; 2704 tp->snd_cwnd = maxseg; 2705 (void) tp->t_fb->tfb_tcp_output(tp); 2706 if (SEQ_GT(th->th_ack, tp->snd_una)) 2707 goto resume_partialack; 2708 goto drop; 2709 } 2710 tp->snd_nxt = th->th_ack; 2711 tp->snd_cwnd = maxseg; 2712 (void) tp->t_fb->tfb_tcp_output(tp); 2713 KASSERT(tp->snd_limited <= 2, 2714 ("%s: tp->snd_limited too big", 2715 __func__)); 2716 tp->snd_cwnd = tp->snd_ssthresh + 2717 maxseg * 2718 (tp->t_dupacks - tp->snd_limited); 2719 if (SEQ_GT(onxt, tp->snd_nxt)) 2720 tp->snd_nxt = onxt; 2721 goto drop; 2722 } else if (V_tcp_do_rfc3042) { 2723 /* 2724 * Process first and second duplicate 2725 * ACKs. Each indicates a segment 2726 * leaving the network, creating room 2727 * for more. Make sure we can send a 2728 * packet on reception of each duplicate 2729 * ACK by increasing snd_cwnd by one 2730 * segment. Restore the original 2731 * snd_cwnd after packet transmission. 2732 */ 2733 cc_ack_received(tp, th, nsegs, 2734 CC_DUPACK); 2735 uint32_t oldcwnd = tp->snd_cwnd; 2736 tcp_seq oldsndmax = tp->snd_max; 2737 u_int sent; 2738 int avail; 2739 2740 KASSERT(tp->t_dupacks == 1 || 2741 tp->t_dupacks == 2, 2742 ("%s: dupacks not 1 or 2", 2743 __func__)); 2744 if (tp->t_dupacks == 1) 2745 tp->snd_limited = 0; 2746 tp->snd_cwnd = 2747 (tp->snd_nxt - tp->snd_una) + 2748 (tp->t_dupacks - tp->snd_limited) * 2749 maxseg; 2750 /* 2751 * Only call tcp_output when there 2752 * is new data available to be sent. 2753 * Otherwise we would send pure ACKs. 2754 */ 2755 SOCKBUF_LOCK(&so->so_snd); 2756 avail = sbavail(&so->so_snd) - 2757 (tp->snd_nxt - tp->snd_una); 2758 SOCKBUF_UNLOCK(&so->so_snd); 2759 if (avail > 0) 2760 (void) tp->t_fb->tfb_tcp_output(tp); 2761 sent = tp->snd_max - oldsndmax; 2762 if (sent > maxseg) { 2763 KASSERT((tp->t_dupacks == 2 && 2764 tp->snd_limited == 0) || 2765 (sent == maxseg + 1 && 2766 tp->t_flags & TF_SENTFIN), 2767 ("%s: sent too much", 2768 __func__)); 2769 tp->snd_limited = 2; 2770 } else if (sent > 0) 2771 ++tp->snd_limited; 2772 tp->snd_cwnd = oldcwnd; 2773 goto drop; 2774 } 2775 } 2776 break; 2777 } else { 2778 /* 2779 * This ack is advancing the left edge, reset the 2780 * counter. 2781 */ 2782 tp->t_dupacks = 0; 2783 /* 2784 * If this ack also has new SACK info, increment the 2785 * counter as per rfc6675. The variable 2786 * sack_changed tracks all changes to the SACK 2787 * scoreboard, including when partial ACKs without 2788 * SACK options are received, and clear the scoreboard 2789 * from the left side. Such partial ACKs should not be 2790 * counted as dupacks here. 2791 */ 2792 if ((tp->t_flags & TF_SACK_PERMIT) && 2793 (to.to_flags & TOF_SACK) && 2794 sack_changed) { 2795 tp->t_dupacks++; 2796 /* limit overhead by setting maxseg last */ 2797 if (!IN_FASTRECOVERY(tp->t_flags) && 2798 (tp->sackhint.sacked_bytes > 2799 ((tcprexmtthresh - 1) * 2800 (maxseg = tcp_maxseg(tp))))) { 2801 goto enter_recovery; 2802 } 2803 } 2804 } 2805 2806 resume_partialack: 2807 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2808 ("%s: th_ack <= snd_una", __func__)); 2809 2810 /* 2811 * If the congestion window was inflated to account 2812 * for the other side's cached packets, retract it. 2813 */ 2814 if (IN_FASTRECOVERY(tp->t_flags)) { 2815 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2816 if (tp->t_flags & TF_SACK_PERMIT) 2817 if (V_tcp_do_prr && to.to_flags & TOF_SACK) 2818 tcp_prr_partialack(tp, th); 2819 else 2820 tcp_sack_partialack(tp, th); 2821 else 2822 tcp_newreno_partial_ack(tp, th); 2823 } else 2824 cc_post_recovery(tp, th); 2825 } 2826 /* 2827 * If we reach this point, ACK is not a duplicate, 2828 * i.e., it ACKs something we sent. 2829 */ 2830 if (tp->t_flags & TF_NEEDSYN) { 2831 /* 2832 * T/TCP: Connection was half-synchronized, and our 2833 * SYN has been ACK'd (so connection is now fully 2834 * synchronized). Go to non-starred state, 2835 * increment snd_una for ACK of SYN, and check if 2836 * we can do window scaling. 2837 */ 2838 tp->t_flags &= ~TF_NEEDSYN; 2839 tp->snd_una++; 2840 /* Do window scaling? */ 2841 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2842 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2843 tp->rcv_scale = tp->request_r_scale; 2844 /* Send window already scaled. */ 2845 } 2846 } 2847 2848 process_ACK: 2849 INP_WLOCK_ASSERT(tp->t_inpcb); 2850 2851 /* 2852 * Adjust for the SYN bit in sequence space, 2853 * but don't account for it in cwnd calculations. 2854 * This is for the SYN_RECEIVED, non-simultaneous 2855 * SYN case. SYN_SENT and simultaneous SYN are 2856 * treated elsewhere. 2857 */ 2858 if (incforsyn) 2859 tp->snd_una++; 2860 acked = BYTES_THIS_ACK(tp, th); 2861 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2862 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2863 tp->snd_una, th->th_ack, tp, m)); 2864 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2865 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2866 2867 /* 2868 * If we just performed our first retransmit, and the ACK 2869 * arrives within our recovery window, then it was a mistake 2870 * to do the retransmit in the first place. Recover our 2871 * original cwnd and ssthresh, and proceed to transmit where 2872 * we left off. 2873 */ 2874 if (tp->t_rxtshift == 1 && 2875 tp->t_flags & TF_PREVVALID && 2876 tp->t_badrxtwin && 2877 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2878 cc_cong_signal(tp, th, CC_RTO_ERR); 2879 2880 /* 2881 * If we have a timestamp reply, update smoothed 2882 * round trip time. If no timestamp is present but 2883 * transmit timer is running and timed sequence 2884 * number was acked, update smoothed round trip time. 2885 * Since we now have an rtt measurement, cancel the 2886 * timer backoff (cf., Phil Karn's retransmit alg.). 2887 * Recompute the initial retransmit timer. 2888 * 2889 * Some boxes send broken timestamp replies 2890 * during the SYN+ACK phase, ignore 2891 * timestamps of 0 or we could calculate a 2892 * huge RTT and blow up the retransmit timer. 2893 */ 2894 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2895 uint32_t t; 2896 2897 t = tcp_ts_getticks() - to.to_tsecr; 2898 if (!tp->t_rttlow || tp->t_rttlow > t) 2899 tp->t_rttlow = t; 2900 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2901 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2902 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2903 tp->t_rttlow = ticks - tp->t_rtttime; 2904 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2905 } 2906 2907 /* 2908 * If all outstanding data is acked, stop retransmit 2909 * timer and remember to restart (more output or persist). 2910 * If there is more data to be acked, restart retransmit 2911 * timer, using current (possibly backed-off) value. 2912 */ 2913 if (th->th_ack == tp->snd_max) { 2914 tcp_timer_activate(tp, TT_REXMT, 0); 2915 needoutput = 1; 2916 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2917 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2918 2919 /* 2920 * If no data (only SYN) was ACK'd, 2921 * skip rest of ACK processing. 2922 */ 2923 if (acked == 0) 2924 goto step6; 2925 2926 /* 2927 * Let the congestion control algorithm update congestion 2928 * control related information. This typically means increasing 2929 * the congestion window. 2930 */ 2931 cc_ack_received(tp, th, nsegs, CC_ACK); 2932 2933 SOCKBUF_LOCK(&so->so_snd); 2934 if (acked > sbavail(&so->so_snd)) { 2935 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2936 tp->snd_wnd -= sbavail(&so->so_snd); 2937 else 2938 tp->snd_wnd = 0; 2939 mfree = sbcut_locked(&so->so_snd, 2940 (int)sbavail(&so->so_snd)); 2941 ourfinisacked = 1; 2942 } else { 2943 mfree = sbcut_locked(&so->so_snd, acked); 2944 if (tp->snd_wnd >= (uint32_t) acked) 2945 tp->snd_wnd -= acked; 2946 else 2947 tp->snd_wnd = 0; 2948 ourfinisacked = 0; 2949 } 2950 SOCKBUF_UNLOCK(&so->so_snd); 2951 tp->t_flags |= TF_WAKESOW; 2952 m_freem(mfree); 2953 /* Detect una wraparound. */ 2954 if (!IN_RECOVERY(tp->t_flags) && 2955 SEQ_GT(tp->snd_una, tp->snd_recover) && 2956 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2957 tp->snd_recover = th->th_ack - 1; 2958 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2959 if (IN_RECOVERY(tp->t_flags) && 2960 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2961 EXIT_RECOVERY(tp->t_flags); 2962 } 2963 tp->snd_una = th->th_ack; 2964 if (tp->t_flags & TF_SACK_PERMIT) { 2965 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2966 tp->snd_recover = tp->snd_una; 2967 } 2968 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2969 tp->snd_nxt = tp->snd_una; 2970 2971 switch (tp->t_state) { 2972 /* 2973 * In FIN_WAIT_1 STATE in addition to the processing 2974 * for the ESTABLISHED state if our FIN is now acknowledged 2975 * then enter FIN_WAIT_2. 2976 */ 2977 case TCPS_FIN_WAIT_1: 2978 if (ourfinisacked) { 2979 /* 2980 * If we can't receive any more 2981 * data, then closing user can proceed. 2982 * Starting the timer is contrary to the 2983 * specification, but if we don't get a FIN 2984 * we'll hang forever. 2985 * 2986 * XXXjl: 2987 * we should release the tp also, and use a 2988 * compressed state. 2989 */ 2990 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2991 soisdisconnected(so); 2992 tcp_timer_activate(tp, TT_2MSL, 2993 (tcp_fast_finwait2_recycle ? 2994 tcp_finwait2_timeout : 2995 TP_MAXIDLE(tp))); 2996 } 2997 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2998 } 2999 break; 3000 3001 /* 3002 * In CLOSING STATE in addition to the processing for 3003 * the ESTABLISHED state if the ACK acknowledges our FIN 3004 * then enter the TIME-WAIT state, otherwise ignore 3005 * the segment. 3006 */ 3007 case TCPS_CLOSING: 3008 if (ourfinisacked) { 3009 tcp_twstart(tp); 3010 m_freem(m); 3011 return; 3012 } 3013 break; 3014 3015 /* 3016 * In LAST_ACK, we may still be waiting for data to drain 3017 * and/or to be acked, as well as for the ack of our FIN. 3018 * If our FIN is now acknowledged, delete the TCB, 3019 * enter the closed state and return. 3020 */ 3021 case TCPS_LAST_ACK: 3022 if (ourfinisacked) { 3023 tp = tcp_close(tp); 3024 goto drop; 3025 } 3026 break; 3027 } 3028 } 3029 3030 step6: 3031 INP_WLOCK_ASSERT(tp->t_inpcb); 3032 3033 /* 3034 * Update window information. 3035 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3036 */ 3037 if ((thflags & TH_ACK) && 3038 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3039 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3040 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3041 /* keep track of pure window updates */ 3042 if (tlen == 0 && 3043 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3044 TCPSTAT_INC(tcps_rcvwinupd); 3045 tp->snd_wnd = tiwin; 3046 tp->snd_wl1 = th->th_seq; 3047 tp->snd_wl2 = th->th_ack; 3048 if (tp->snd_wnd > tp->max_sndwnd) 3049 tp->max_sndwnd = tp->snd_wnd; 3050 needoutput = 1; 3051 } 3052 3053 /* 3054 * Process segments with URG. 3055 */ 3056 if ((thflags & TH_URG) && th->th_urp && 3057 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3058 /* 3059 * This is a kludge, but if we receive and accept 3060 * random urgent pointers, we'll crash in 3061 * soreceive. It's hard to imagine someone 3062 * actually wanting to send this much urgent data. 3063 */ 3064 SOCKBUF_LOCK(&so->so_rcv); 3065 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3066 th->th_urp = 0; /* XXX */ 3067 thflags &= ~TH_URG; /* XXX */ 3068 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3069 goto dodata; /* XXX */ 3070 } 3071 /* 3072 * If this segment advances the known urgent pointer, 3073 * then mark the data stream. This should not happen 3074 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3075 * a FIN has been received from the remote side. 3076 * In these states we ignore the URG. 3077 * 3078 * According to RFC961 (Assigned Protocols), 3079 * the urgent pointer points to the last octet 3080 * of urgent data. We continue, however, 3081 * to consider it to indicate the first octet 3082 * of data past the urgent section as the original 3083 * spec states (in one of two places). 3084 */ 3085 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3086 tp->rcv_up = th->th_seq + th->th_urp; 3087 so->so_oobmark = sbavail(&so->so_rcv) + 3088 (tp->rcv_up - tp->rcv_nxt) - 1; 3089 if (so->so_oobmark == 0) 3090 so->so_rcv.sb_state |= SBS_RCVATMARK; 3091 sohasoutofband(so); 3092 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3093 } 3094 SOCKBUF_UNLOCK(&so->so_rcv); 3095 /* 3096 * Remove out of band data so doesn't get presented to user. 3097 * This can happen independent of advancing the URG pointer, 3098 * but if two URG's are pending at once, some out-of-band 3099 * data may creep in... ick. 3100 */ 3101 if (th->th_urp <= (uint32_t)tlen && 3102 !(so->so_options & SO_OOBINLINE)) { 3103 /* hdr drop is delayed */ 3104 tcp_pulloutofband(so, th, m, drop_hdrlen); 3105 } 3106 } else { 3107 /* 3108 * If no out of band data is expected, 3109 * pull receive urgent pointer along 3110 * with the receive window. 3111 */ 3112 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3113 tp->rcv_up = tp->rcv_nxt; 3114 } 3115 dodata: /* XXX */ 3116 INP_WLOCK_ASSERT(tp->t_inpcb); 3117 3118 /* 3119 * Process the segment text, merging it into the TCP sequencing queue, 3120 * and arranging for acknowledgment of receipt if necessary. 3121 * This process logically involves adjusting tp->rcv_wnd as data 3122 * is presented to the user (this happens in tcp_usrreq.c, 3123 * case PRU_RCVD). If a FIN has already been received on this 3124 * connection then we just ignore the text. 3125 */ 3126 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3127 IS_FASTOPEN(tp->t_flags)); 3128 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3129 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3130 tcp_seq save_start = th->th_seq; 3131 tcp_seq save_rnxt = tp->rcv_nxt; 3132 int save_tlen = tlen; 3133 m_adj(m, drop_hdrlen); /* delayed header drop */ 3134 /* 3135 * Insert segment which includes th into TCP reassembly queue 3136 * with control block tp. Set thflags to whether reassembly now 3137 * includes a segment with FIN. This handles the common case 3138 * inline (segment is the next to be received on an established 3139 * connection, and the queue is empty), avoiding linkage into 3140 * and removal from the queue and repetition of various 3141 * conversions. 3142 * Set DELACK for segments received in order, but ack 3143 * immediately when segments are out of order (so 3144 * fast retransmit can work). 3145 */ 3146 if (th->th_seq == tp->rcv_nxt && 3147 SEGQ_EMPTY(tp) && 3148 (TCPS_HAVEESTABLISHED(tp->t_state) || 3149 tfo_syn)) { 3150 if (DELAY_ACK(tp, tlen) || tfo_syn) 3151 tp->t_flags |= TF_DELACK; 3152 else 3153 tp->t_flags |= TF_ACKNOW; 3154 tp->rcv_nxt += tlen; 3155 if (tlen && 3156 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3157 (tp->t_fbyte_in == 0)) { 3158 tp->t_fbyte_in = ticks; 3159 if (tp->t_fbyte_in == 0) 3160 tp->t_fbyte_in = 1; 3161 if (tp->t_fbyte_out && tp->t_fbyte_in) 3162 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3163 } 3164 thflags = th->th_flags & TH_FIN; 3165 TCPSTAT_INC(tcps_rcvpack); 3166 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3167 SOCKBUF_LOCK(&so->so_rcv); 3168 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3169 m_freem(m); 3170 else 3171 sbappendstream_locked(&so->so_rcv, m, 0); 3172 SOCKBUF_UNLOCK(&so->so_rcv); 3173 tp->t_flags |= TF_WAKESOR; 3174 } else { 3175 /* 3176 * XXX: Due to the header drop above "th" is 3177 * theoretically invalid by now. Fortunately 3178 * m_adj() doesn't actually frees any mbufs 3179 * when trimming from the head. 3180 */ 3181 tcp_seq temp = save_start; 3182 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3183 tp->t_flags |= TF_ACKNOW; 3184 } 3185 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) { 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 #if 0 3217 /* 3218 * Note the amount of data that peer has sent into 3219 * our window, in order to estimate the sender's 3220 * buffer size. 3221 * XXX: Unused. 3222 */ 3223 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3224 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3225 else 3226 len = so->so_rcv.sb_hiwat; 3227 #endif 3228 } else { 3229 m_freem(m); 3230 thflags &= ~TH_FIN; 3231 } 3232 3233 /* 3234 * If FIN is received ACK the FIN and let the user know 3235 * that the connection is closing. 3236 */ 3237 if (thflags & TH_FIN) { 3238 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3239 socantrcvmore(so); 3240 /* The socket upcall is handled by socantrcvmore. */ 3241 tp->t_flags &= ~TF_WAKESOR; 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) tp->t_fb->tfb_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 tcp_handle_wakeup(tp, so); 3306 INP_WUNLOCK(tp->t_inpcb); 3307 return; 3308 3309 dropafterack: 3310 /* 3311 * Generate an ACK dropping incoming segment if it occupies 3312 * sequence space, where the ACK reflects our state. 3313 * 3314 * We can now skip the test for the RST flag since all 3315 * paths to this code happen after packets containing 3316 * RST have been dropped. 3317 * 3318 * In the SYN-RECEIVED state, don't send an ACK unless the 3319 * segment we received passes the SYN-RECEIVED ACK test. 3320 * If it fails send a RST. This breaks the loop in the 3321 * "LAND" DoS attack, and also prevents an ACK storm 3322 * between two listening ports that have been sent forged 3323 * SYN segments, each with the source address of the other. 3324 */ 3325 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3326 (SEQ_GT(tp->snd_una, th->th_ack) || 3327 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3328 rstreason = BANDLIM_RST_OPENPORT; 3329 goto dropwithreset; 3330 } 3331 #ifdef TCPDEBUG 3332 if (so->so_options & SO_DEBUG) 3333 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3334 &tcp_savetcp, 0); 3335 #endif 3336 TCP_PROBE3(debug__input, tp, th, m); 3337 tp->t_flags |= TF_ACKNOW; 3338 (void) tp->t_fb->tfb_tcp_output(tp); 3339 tcp_handle_wakeup(tp, so); 3340 INP_WUNLOCK(tp->t_inpcb); 3341 m_freem(m); 3342 return; 3343 3344 dropwithreset: 3345 if (tp != NULL) { 3346 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3347 tcp_handle_wakeup(tp, so); 3348 INP_WUNLOCK(tp->t_inpcb); 3349 } else 3350 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3351 return; 3352 3353 drop: 3354 /* 3355 * Drop space held by incoming segment and return. 3356 */ 3357 #ifdef TCPDEBUG 3358 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3359 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3360 &tcp_savetcp, 0); 3361 #endif 3362 TCP_PROBE3(debug__input, tp, th, m); 3363 if (tp != NULL) { 3364 tcp_handle_wakeup(tp, so); 3365 INP_WUNLOCK(tp->t_inpcb); 3366 } 3367 m_freem(m); 3368 } 3369 3370 /* 3371 * Issue RST and make ACK acceptable to originator of segment. 3372 * The mbuf must still include the original packet header. 3373 * tp may be NULL. 3374 */ 3375 void 3376 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3377 int tlen, int rstreason) 3378 { 3379 #ifdef INET 3380 struct ip *ip; 3381 #endif 3382 #ifdef INET6 3383 struct ip6_hdr *ip6; 3384 #endif 3385 3386 if (tp != NULL) { 3387 INP_WLOCK_ASSERT(tp->t_inpcb); 3388 } 3389 3390 /* Don't bother if destination was broadcast/multicast. */ 3391 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3392 goto drop; 3393 #ifdef INET6 3394 if (mtod(m, struct ip *)->ip_v == 6) { 3395 ip6 = mtod(m, struct ip6_hdr *); 3396 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3397 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3398 goto drop; 3399 /* IPv6 anycast check is done at tcp6_input() */ 3400 } 3401 #endif 3402 #if defined(INET) && defined(INET6) 3403 else 3404 #endif 3405 #ifdef INET 3406 { 3407 ip = mtod(m, struct ip *); 3408 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3409 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3410 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3411 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3412 goto drop; 3413 } 3414 #endif 3415 3416 /* Perform bandwidth limiting. */ 3417 if (badport_bandlim(rstreason) < 0) 3418 goto drop; 3419 3420 /* tcp_respond consumes the mbuf chain. */ 3421 if (th->th_flags & TH_ACK) { 3422 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3423 th->th_ack, TH_RST); 3424 } else { 3425 if (th->th_flags & TH_SYN) 3426 tlen++; 3427 if (th->th_flags & TH_FIN) 3428 tlen++; 3429 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3430 (tcp_seq)0, TH_RST|TH_ACK); 3431 } 3432 return; 3433 drop: 3434 m_freem(m); 3435 } 3436 3437 /* 3438 * Parse TCP options and place in tcpopt. 3439 */ 3440 void 3441 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3442 { 3443 int opt, optlen; 3444 3445 to->to_flags = 0; 3446 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3447 opt = cp[0]; 3448 if (opt == TCPOPT_EOL) 3449 break; 3450 if (opt == TCPOPT_NOP) 3451 optlen = 1; 3452 else { 3453 if (cnt < 2) 3454 break; 3455 optlen = cp[1]; 3456 if (optlen < 2 || optlen > cnt) 3457 break; 3458 } 3459 switch (opt) { 3460 case TCPOPT_MAXSEG: 3461 if (optlen != TCPOLEN_MAXSEG) 3462 continue; 3463 if (!(flags & TO_SYN)) 3464 continue; 3465 to->to_flags |= TOF_MSS; 3466 bcopy((char *)cp + 2, 3467 (char *)&to->to_mss, sizeof(to->to_mss)); 3468 to->to_mss = ntohs(to->to_mss); 3469 break; 3470 case TCPOPT_WINDOW: 3471 if (optlen != TCPOLEN_WINDOW) 3472 continue; 3473 if (!(flags & TO_SYN)) 3474 continue; 3475 to->to_flags |= TOF_SCALE; 3476 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3477 break; 3478 case TCPOPT_TIMESTAMP: 3479 if (optlen != TCPOLEN_TIMESTAMP) 3480 continue; 3481 to->to_flags |= TOF_TS; 3482 bcopy((char *)cp + 2, 3483 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3484 to->to_tsval = ntohl(to->to_tsval); 3485 bcopy((char *)cp + 6, 3486 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3487 to->to_tsecr = ntohl(to->to_tsecr); 3488 break; 3489 case TCPOPT_SIGNATURE: 3490 /* 3491 * In order to reply to a host which has set the 3492 * TCP_SIGNATURE option in its initial SYN, we have 3493 * to record the fact that the option was observed 3494 * here for the syncache code to perform the correct 3495 * response. 3496 */ 3497 if (optlen != TCPOLEN_SIGNATURE) 3498 continue; 3499 to->to_flags |= TOF_SIGNATURE; 3500 to->to_signature = cp + 2; 3501 break; 3502 case TCPOPT_SACK_PERMITTED: 3503 if (optlen != TCPOLEN_SACK_PERMITTED) 3504 continue; 3505 if (!(flags & TO_SYN)) 3506 continue; 3507 if (!V_tcp_do_sack) 3508 continue; 3509 to->to_flags |= TOF_SACKPERM; 3510 break; 3511 case TCPOPT_SACK: 3512 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3513 continue; 3514 if (flags & TO_SYN) 3515 continue; 3516 to->to_flags |= TOF_SACK; 3517 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3518 to->to_sacks = cp + 2; 3519 TCPSTAT_INC(tcps_sack_rcv_blocks); 3520 break; 3521 case TCPOPT_FAST_OPEN: 3522 /* 3523 * Cookie length validation is performed by the 3524 * server side cookie checking code or the client 3525 * side cookie cache update code. 3526 */ 3527 if (!(flags & TO_SYN)) 3528 continue; 3529 if (!V_tcp_fastopen_client_enable && 3530 !V_tcp_fastopen_server_enable) 3531 continue; 3532 to->to_flags |= TOF_FASTOPEN; 3533 to->to_tfo_len = optlen - 2; 3534 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3535 break; 3536 default: 3537 continue; 3538 } 3539 } 3540 } 3541 3542 /* 3543 * Pull out of band byte out of a segment so 3544 * it doesn't appear in the user's data queue. 3545 * It is still reflected in the segment length for 3546 * sequencing purposes. 3547 */ 3548 void 3549 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3550 int off) 3551 { 3552 int cnt = off + th->th_urp - 1; 3553 3554 while (cnt >= 0) { 3555 if (m->m_len > cnt) { 3556 char *cp = mtod(m, caddr_t) + cnt; 3557 struct tcpcb *tp = sototcpcb(so); 3558 3559 INP_WLOCK_ASSERT(tp->t_inpcb); 3560 3561 tp->t_iobc = *cp; 3562 tp->t_oobflags |= TCPOOB_HAVEDATA; 3563 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3564 m->m_len--; 3565 if (m->m_flags & M_PKTHDR) 3566 m->m_pkthdr.len--; 3567 return; 3568 } 3569 cnt -= m->m_len; 3570 m = m->m_next; 3571 if (m == NULL) 3572 break; 3573 } 3574 panic("tcp_pulloutofband"); 3575 } 3576 3577 /* 3578 * Collect new round-trip time estimate 3579 * and update averages and current timeout. 3580 */ 3581 void 3582 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3583 { 3584 int delta; 3585 3586 INP_WLOCK_ASSERT(tp->t_inpcb); 3587 3588 TCPSTAT_INC(tcps_rttupdated); 3589 tp->t_rttupdated++; 3590 #ifdef STATS 3591 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3592 imax(0, rtt * 1000 / hz)); 3593 #endif 3594 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3595 /* 3596 * srtt is stored as fixed point with 5 bits after the 3597 * binary point (i.e., scaled by 8). The following magic 3598 * is equivalent to the smoothing algorithm in rfc793 with 3599 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3600 * point). Adjust rtt to origin 0. 3601 */ 3602 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3603 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3604 3605 if ((tp->t_srtt += delta) <= 0) 3606 tp->t_srtt = 1; 3607 3608 /* 3609 * We accumulate a smoothed rtt variance (actually, a 3610 * smoothed mean difference), then set the retransmit 3611 * timer to smoothed rtt + 4 times the smoothed variance. 3612 * rttvar is stored as fixed point with 4 bits after the 3613 * binary point (scaled by 16). The following is 3614 * equivalent to rfc793 smoothing with an alpha of .75 3615 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3616 * rfc793's wired-in beta. 3617 */ 3618 if (delta < 0) 3619 delta = -delta; 3620 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3621 if ((tp->t_rttvar += delta) <= 0) 3622 tp->t_rttvar = 1; 3623 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3624 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3625 } else { 3626 /* 3627 * No rtt measurement yet - use the unsmoothed rtt. 3628 * Set the variance to half the rtt (so our first 3629 * retransmit happens at 3*rtt). 3630 */ 3631 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3632 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3633 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3634 } 3635 tp->t_rtttime = 0; 3636 tp->t_rxtshift = 0; 3637 3638 /* 3639 * the retransmit should happen at rtt + 4 * rttvar. 3640 * Because of the way we do the smoothing, srtt and rttvar 3641 * will each average +1/2 tick of bias. When we compute 3642 * the retransmit timer, we want 1/2 tick of rounding and 3643 * 1 extra tick because of +-1/2 tick uncertainty in the 3644 * firing of the timer. The bias will give us exactly the 3645 * 1.5 tick we need. But, because the bias is 3646 * statistical, we have to test that we don't drop below 3647 * the minimum feasible timer (which is 2 ticks). 3648 */ 3649 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3650 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3651 3652 /* 3653 * We received an ack for a packet that wasn't retransmitted; 3654 * it is probably safe to discard any error indications we've 3655 * received recently. This isn't quite right, but close enough 3656 * for now (a route might have failed after we sent a segment, 3657 * and the return path might not be symmetrical). 3658 */ 3659 tp->t_softerror = 0; 3660 } 3661 3662 /* 3663 * Determine a reasonable value for maxseg size. 3664 * If the route is known, check route for mtu. 3665 * If none, use an mss that can be handled on the outgoing interface 3666 * without forcing IP to fragment. If no route is found, route has no mtu, 3667 * or the destination isn't local, use a default, hopefully conservative 3668 * size (usually 512 or the default IP max size, but no more than the mtu 3669 * of the interface), as we can't discover anything about intervening 3670 * gateways or networks. We also initialize the congestion/slow start 3671 * window to be a single segment if the destination isn't local. 3672 * While looking at the routing entry, we also initialize other path-dependent 3673 * parameters from pre-set or cached values in the routing entry. 3674 * 3675 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3676 * IP options, e.g. IPSEC data, since length of this data may vary, and 3677 * thus it is calculated for every segment separately in tcp_output(). 3678 * 3679 * NOTE that this routine is only called when we process an incoming 3680 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3681 * settings are handled in tcp_mssopt(). 3682 */ 3683 void 3684 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3685 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3686 { 3687 int mss = 0; 3688 uint32_t maxmtu = 0; 3689 struct inpcb *inp = tp->t_inpcb; 3690 struct hc_metrics_lite metrics; 3691 #ifdef INET6 3692 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3693 size_t min_protoh = isipv6 ? 3694 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3695 sizeof (struct tcpiphdr); 3696 #else 3697 const size_t min_protoh = sizeof(struct tcpiphdr); 3698 #endif 3699 3700 INP_WLOCK_ASSERT(tp->t_inpcb); 3701 3702 if (mtuoffer != -1) { 3703 KASSERT(offer == -1, ("%s: conflict", __func__)); 3704 offer = mtuoffer - min_protoh; 3705 } 3706 3707 /* Initialize. */ 3708 #ifdef INET6 3709 if (isipv6) { 3710 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3711 tp->t_maxseg = V_tcp_v6mssdflt; 3712 } 3713 #endif 3714 #if defined(INET) && defined(INET6) 3715 else 3716 #endif 3717 #ifdef INET 3718 { 3719 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3720 tp->t_maxseg = V_tcp_mssdflt; 3721 } 3722 #endif 3723 3724 /* 3725 * No route to sender, stay with default mss and return. 3726 */ 3727 if (maxmtu == 0) { 3728 /* 3729 * In case we return early we need to initialize metrics 3730 * to a defined state as tcp_hc_get() would do for us 3731 * if there was no cache hit. 3732 */ 3733 if (metricptr != NULL) 3734 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3735 return; 3736 } 3737 3738 /* What have we got? */ 3739 switch (offer) { 3740 case 0: 3741 /* 3742 * Offer == 0 means that there was no MSS on the SYN 3743 * segment, in this case we use tcp_mssdflt as 3744 * already assigned to t_maxseg above. 3745 */ 3746 offer = tp->t_maxseg; 3747 break; 3748 3749 case -1: 3750 /* 3751 * Offer == -1 means that we didn't receive SYN yet. 3752 */ 3753 /* FALLTHROUGH */ 3754 3755 default: 3756 /* 3757 * Prevent DoS attack with too small MSS. Round up 3758 * to at least minmss. 3759 */ 3760 offer = max(offer, V_tcp_minmss); 3761 } 3762 3763 /* 3764 * rmx information is now retrieved from tcp_hostcache. 3765 */ 3766 tcp_hc_get(&inp->inp_inc, &metrics); 3767 if (metricptr != NULL) 3768 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3769 3770 /* 3771 * If there's a discovered mtu in tcp hostcache, use it. 3772 * Else, use the link mtu. 3773 */ 3774 if (metrics.rmx_mtu) 3775 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3776 else { 3777 #ifdef INET6 3778 if (isipv6) { 3779 mss = maxmtu - min_protoh; 3780 if (!V_path_mtu_discovery && 3781 !in6_localaddr(&inp->in6p_faddr)) 3782 mss = min(mss, V_tcp_v6mssdflt); 3783 } 3784 #endif 3785 #if defined(INET) && defined(INET6) 3786 else 3787 #endif 3788 #ifdef INET 3789 { 3790 mss = maxmtu - min_protoh; 3791 if (!V_path_mtu_discovery && 3792 !in_localaddr(inp->inp_faddr)) 3793 mss = min(mss, V_tcp_mssdflt); 3794 } 3795 #endif 3796 /* 3797 * XXX - The above conditional (mss = maxmtu - min_protoh) 3798 * probably violates the TCP spec. 3799 * The problem is that, since we don't know the 3800 * other end's MSS, we are supposed to use a conservative 3801 * default. But, if we do that, then MTU discovery will 3802 * never actually take place, because the conservative 3803 * default is much less than the MTUs typically seen 3804 * on the Internet today. For the moment, we'll sweep 3805 * this under the carpet. 3806 * 3807 * The conservative default might not actually be a problem 3808 * if the only case this occurs is when sending an initial 3809 * SYN with options and data to a host we've never talked 3810 * to before. Then, they will reply with an MSS value which 3811 * will get recorded and the new parameters should get 3812 * recomputed. For Further Study. 3813 */ 3814 } 3815 mss = min(mss, offer); 3816 3817 /* 3818 * Sanity check: make sure that maxseg will be large 3819 * enough to allow some data on segments even if the 3820 * all the option space is used (40bytes). Otherwise 3821 * funny things may happen in tcp_output. 3822 * 3823 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3824 */ 3825 mss = max(mss, 64); 3826 3827 tp->t_maxseg = mss; 3828 } 3829 3830 void 3831 tcp_mss(struct tcpcb *tp, int offer) 3832 { 3833 int mss; 3834 uint32_t bufsize; 3835 struct inpcb *inp; 3836 struct socket *so; 3837 struct hc_metrics_lite metrics; 3838 struct tcp_ifcap cap; 3839 3840 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3841 3842 bzero(&cap, sizeof(cap)); 3843 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3844 3845 mss = tp->t_maxseg; 3846 inp = tp->t_inpcb; 3847 3848 /* 3849 * If there's a pipesize, change the socket buffer to that size, 3850 * don't change if sb_hiwat is different than default (then it 3851 * has been changed on purpose with setsockopt). 3852 * Make the socket buffers an integral number of mss units; 3853 * if the mss is larger than the socket buffer, decrease the mss. 3854 */ 3855 so = inp->inp_socket; 3856 SOCKBUF_LOCK(&so->so_snd); 3857 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3858 bufsize = metrics.rmx_sendpipe; 3859 else 3860 bufsize = so->so_snd.sb_hiwat; 3861 if (bufsize < mss) 3862 mss = bufsize; 3863 else { 3864 bufsize = roundup(bufsize, mss); 3865 if (bufsize > sb_max) 3866 bufsize = sb_max; 3867 if (bufsize > so->so_snd.sb_hiwat) 3868 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3869 } 3870 SOCKBUF_UNLOCK(&so->so_snd); 3871 /* 3872 * Sanity check: make sure that maxseg will be large 3873 * enough to allow some data on segments even if the 3874 * all the option space is used (40bytes). Otherwise 3875 * funny things may happen in tcp_output. 3876 * 3877 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3878 */ 3879 tp->t_maxseg = max(mss, 64); 3880 3881 SOCKBUF_LOCK(&so->so_rcv); 3882 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3883 bufsize = metrics.rmx_recvpipe; 3884 else 3885 bufsize = so->so_rcv.sb_hiwat; 3886 if (bufsize > mss) { 3887 bufsize = roundup(bufsize, mss); 3888 if (bufsize > sb_max) 3889 bufsize = sb_max; 3890 if (bufsize > so->so_rcv.sb_hiwat) 3891 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3892 } 3893 SOCKBUF_UNLOCK(&so->so_rcv); 3894 3895 /* Check the interface for TSO capabilities. */ 3896 if (cap.ifcap & CSUM_TSO) { 3897 tp->t_flags |= TF_TSO; 3898 tp->t_tsomax = cap.tsomax; 3899 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3900 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3901 } 3902 } 3903 3904 /* 3905 * Determine the MSS option to send on an outgoing SYN. 3906 */ 3907 int 3908 tcp_mssopt(struct in_conninfo *inc) 3909 { 3910 int mss = 0; 3911 uint32_t thcmtu = 0; 3912 uint32_t maxmtu = 0; 3913 size_t min_protoh; 3914 3915 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3916 3917 #ifdef INET6 3918 if (inc->inc_flags & INC_ISIPV6) { 3919 mss = V_tcp_v6mssdflt; 3920 maxmtu = tcp_maxmtu6(inc, NULL); 3921 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3922 } 3923 #endif 3924 #if defined(INET) && defined(INET6) 3925 else 3926 #endif 3927 #ifdef INET 3928 { 3929 mss = V_tcp_mssdflt; 3930 maxmtu = tcp_maxmtu(inc, NULL); 3931 min_protoh = sizeof(struct tcpiphdr); 3932 } 3933 #endif 3934 #if defined(INET6) || defined(INET) 3935 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3936 #endif 3937 3938 if (maxmtu && thcmtu) 3939 mss = min(maxmtu, thcmtu) - min_protoh; 3940 else if (maxmtu || thcmtu) 3941 mss = max(maxmtu, thcmtu) - min_protoh; 3942 3943 return (mss); 3944 } 3945 3946 void 3947 tcp_prr_partialack(struct tcpcb *tp, struct tcphdr *th) 3948 { 3949 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3950 int maxseg = tcp_maxseg(tp); 3951 3952 INP_WLOCK_ASSERT(tp->t_inpcb); 3953 3954 tcp_timer_activate(tp, TT_REXMT, 0); 3955 tp->t_rtttime = 0; 3956 /* 3957 * Compute the amount of data that this ACK is indicating 3958 * (del_data) and an estimate of how many bytes are in the 3959 * network. 3960 */ 3961 del_data = tp->sackhint.delivered_data; 3962 if (V_tcp_do_rfc6675_pipe) 3963 pipe = tcp_compute_pipe(tp); 3964 else 3965 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; 3966 tp->sackhint.prr_delivered += del_data; 3967 /* 3968 * Proportional Rate Reduction 3969 */ 3970 if (pipe >= tp->snd_ssthresh) { 3971 if (tp->sackhint.recover_fs == 0) 3972 tp->sackhint.recover_fs = 3973 imax(1, tp->snd_nxt - tp->snd_una); 3974 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3975 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3976 (tp->sackhint.sack_bytes_rexmit + 3977 (tp->snd_nxt - tp->snd_recover)); 3978 } else { 3979 if (V_tcp_do_prr_conservative) 3980 limit = tp->sackhint.prr_delivered - 3981 (tp->sackhint.sack_bytes_rexmit + 3982 (tp->snd_nxt - tp->snd_recover)); 3983 else 3984 limit = imax(tp->sackhint.prr_delivered - 3985 (tp->sackhint.sack_bytes_rexmit + 3986 (tp->snd_nxt - tp->snd_recover)), 3987 del_data) + maxseg; 3988 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 3989 } 3990 snd_cnt = imax(snd_cnt, 0) / maxseg; 3991 /* 3992 * Send snd_cnt new data into the network in response to this ack. 3993 * If there is going to be a SACK retransmission, adjust snd_cwnd 3994 * accordingly. 3995 */ 3996 tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover + 3997 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg)); 3998 tp->t_flags |= TF_ACKNOW; 3999 (void) tcp_output(tp); 4000 } 4001 4002 /* 4003 * On a partial ack arrives, force the retransmission of the 4004 * next unacknowledged segment. Do not clear tp->t_dupacks. 4005 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4006 * be started again. 4007 */ 4008 void 4009 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4010 { 4011 tcp_seq onxt = tp->snd_nxt; 4012 uint32_t ocwnd = tp->snd_cwnd; 4013 u_int maxseg = tcp_maxseg(tp); 4014 4015 INP_WLOCK_ASSERT(tp->t_inpcb); 4016 4017 tcp_timer_activate(tp, TT_REXMT, 0); 4018 tp->t_rtttime = 0; 4019 tp->snd_nxt = th->th_ack; 4020 /* 4021 * Set snd_cwnd to one segment beyond acknowledged offset. 4022 * (tp->snd_una has not yet been updated when this function is called.) 4023 */ 4024 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4025 tp->t_flags |= TF_ACKNOW; 4026 (void) tp->t_fb->tfb_tcp_output(tp); 4027 tp->snd_cwnd = ocwnd; 4028 if (SEQ_GT(onxt, tp->snd_nxt)) 4029 tp->snd_nxt = onxt; 4030 /* 4031 * Partial window deflation. Relies on fact that tp->snd_una 4032 * not updated yet. 4033 */ 4034 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4035 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4036 else 4037 tp->snd_cwnd = 0; 4038 tp->snd_cwnd += maxseg; 4039 } 4040 4041 int 4042 tcp_compute_pipe(struct tcpcb *tp) 4043 { 4044 return (tp->snd_max - tp->snd_una + 4045 tp->sackhint.sack_bytes_rexmit - 4046 tp->sackhint.sacked_bytes); 4047 } 4048 4049 uint32_t 4050 tcp_compute_initwnd(uint32_t maxseg) 4051 { 4052 /* 4053 * Calculate the Initial Window, also used as Restart Window 4054 * 4055 * RFC5681 Section 3.1 specifies the default conservative values. 4056 * RFC3390 specifies slightly more aggressive values. 4057 * RFC6928 increases it to ten segments. 4058 * Support for user specified value for initial flight size. 4059 */ 4060 if (V_tcp_initcwnd_segments) 4061 return min(V_tcp_initcwnd_segments * maxseg, 4062 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4063 else if (V_tcp_do_rfc3390) 4064 return min(4 * maxseg, max(2 * maxseg, 4380)); 4065 else { 4066 /* Per RFC5681 Section 3.1 */ 4067 if (maxseg > 2190) 4068 return (2 * maxseg); 4069 else if (maxseg > 1095) 4070 return (3 * maxseg); 4071 else 4072 return (4 * maxseg); 4073 } 4074 } 4075