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 <[email protected]> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 __FBSDID("$FreeBSD$"); 54 55 #include "opt_inet.h" 56 #include "opt_inet6.h" 57 #include "opt_ipsec.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/arb.h> 62 #include <sys/kernel.h> 63 #ifdef TCP_HHOOK 64 #include <sys/hhook.h> 65 #endif 66 #include <sys/malloc.h> 67 #include <sys/mbuf.h> 68 #include <sys/proc.h> /* for proc0 declaration */ 69 #include <sys/protosw.h> 70 #include <sys/qmath.h> 71 #include <sys/sdt.h> 72 #include <sys/signalvar.h> 73 #include <sys/socket.h> 74 #include <sys/socketvar.h> 75 #include <sys/sysctl.h> 76 #include <sys/syslog.h> 77 #include <sys/systm.h> 78 #include <sys/stats.h> 79 80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 81 82 #include <vm/uma.h> 83 84 #include <net/if.h> 85 #include <net/if_var.h> 86 #include <net/route.h> 87 #include <net/vnet.h> 88 89 #define TCPSTATES /* for logging */ 90 91 #include <netinet/in.h> 92 #include <netinet/in_kdtrace.h> 93 #include <netinet/in_pcb.h> 94 #include <netinet/in_systm.h> 95 #include <netinet/ip.h> 96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 98 #include <netinet/ip_var.h> 99 #include <netinet/ip_options.h> 100 #include <netinet/ip6.h> 101 #include <netinet/icmp6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_var.h> 104 #include <netinet6/ip6_var.h> 105 #include <netinet6/nd6.h> 106 #include <netinet/tcp.h> 107 #include <netinet/tcp_fsm.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_seq.h> 110 #include <netinet/tcp_timer.h> 111 #include <netinet/tcp_var.h> 112 #include <netinet6/tcp6_var.h> 113 #include <netinet/tcpip.h> 114 #include <netinet/cc/cc.h> 115 #include <netinet/tcp_fastopen.h> 116 #ifdef TCPPCAP 117 #include <netinet/tcp_pcap.h> 118 #endif 119 #include <netinet/tcp_syncache.h> 120 #ifdef TCPDEBUG 121 #include <netinet/tcp_debug.h> 122 #endif /* TCPDEBUG */ 123 #ifdef TCP_OFFLOAD 124 #include <netinet/tcp_offload.h> 125 #endif 126 #include <netinet/udp.h> 127 128 #include <netipsec/ipsec_support.h> 129 130 #include <machine/in_cksum.h> 131 132 #include <security/mac/mac_framework.h> 133 134 const int tcprexmtthresh = 3; 135 136 VNET_DEFINE(int, tcp_log_in_vain) = 0; 137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 138 &VNET_NAME(tcp_log_in_vain), 0, 139 "Log all incoming TCP segments to closed ports"); 140 141 VNET_DEFINE(int, blackhole) = 0; 142 #define V_blackhole VNET(blackhole) 143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 144 &VNET_NAME(blackhole), 0, 145 "Do not send RST on segments to closed ports"); 146 147 VNET_DEFINE(int, tcp_delack_enabled) = 1; 148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 149 &VNET_NAME(tcp_delack_enabled), 0, 150 "Delay ACK to try and piggyback it onto a data packet"); 151 152 VNET_DEFINE(int, drop_synfin) = 0; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 154 &VNET_NAME(drop_synfin), 0, 155 "Drop TCP packets with SYN+FIN set"); 156 157 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 159 &VNET_NAME(tcp_do_prr_conservative), 0, 160 "Do conservative Proportional Rate Reduction"); 161 162 VNET_DEFINE(int, tcp_do_prr) = 1; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_prr), 1, 165 "Enable Proportional Rate Reduction per RFC 6937"); 166 167 VNET_DEFINE(int, tcp_do_newcwv) = 0; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_newcwv), 0, 170 "Enable New Congestion Window Validation per RFC7661"); 171 172 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 175 "Use calculated pipe/in-flight bytes per RFC 6675"); 176 177 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 179 &VNET_NAME(tcp_do_rfc3042), 0, 180 "Enable RFC 3042 (Limited Transmit)"); 181 182 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 184 &VNET_NAME(tcp_do_rfc3390), 0, 185 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 186 187 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 189 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 190 "Slow-start flight size (initial congestion window) in number of segments"); 191 192 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 194 &VNET_NAME(tcp_do_rfc3465), 0, 195 "Enable RFC 3465 (Appropriate Byte Counting)"); 196 197 VNET_DEFINE(int, tcp_abc_l_var) = 2; 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_abc_l_var), 2, 200 "Cap the max cwnd increment during slow-start to this number of segments"); 201 202 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 203 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 204 "TCP ECN"); 205 206 VNET_DEFINE(int, tcp_do_ecn) = 2; 207 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_do_ecn), 0, 209 "TCP ECN support"); 210 211 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 212 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(tcp_ecn_maxretries), 0, 214 "Max retries before giving up on ECN"); 215 216 VNET_DEFINE(int, tcp_insecure_syn) = 0; 217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(tcp_insecure_syn), 0, 219 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 220 221 VNET_DEFINE(int, tcp_insecure_rst) = 0; 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_insecure_rst), 0, 224 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 225 226 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 227 #define V_tcp_recvspace VNET(tcp_recvspace) 228 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 229 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 230 231 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 232 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 233 &VNET_NAME(tcp_do_autorcvbuf), 0, 234 "Enable automatic receive buffer sizing"); 235 236 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 238 &VNET_NAME(tcp_autorcvbuf_max), 0, 239 "Max size of automatic receive buffer"); 240 241 VNET_DEFINE(struct inpcbhead, tcb); 242 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 243 VNET_DEFINE(struct inpcbinfo, tcbinfo); 244 245 /* 246 * TCP statistics are stored in an array of counter(9)s, which size matches 247 * size of struct tcpstat. TCP running connection count is a regular array. 248 */ 249 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 253 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 254 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 255 "TCP connection counts by TCP state"); 256 257 static void 258 tcp_vnet_init(const void *unused) 259 { 260 261 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 262 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 263 } 264 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 265 tcp_vnet_init, NULL); 266 267 #ifdef VIMAGE 268 static void 269 tcp_vnet_uninit(const void *unused) 270 { 271 272 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 273 VNET_PCPUSTAT_FREE(tcpstat); 274 } 275 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 276 tcp_vnet_uninit, NULL); 277 #endif /* VIMAGE */ 278 279 /* 280 * Kernel module interface for updating tcpstat. The first argument is an index 281 * into tcpstat treated as an array. 282 */ 283 void 284 kmod_tcpstat_add(int statnum, int val) 285 { 286 287 counter_u64_add(VNET(tcpstat)[statnum], val); 288 } 289 290 #ifdef TCP_HHOOK 291 /* 292 * Wrapper for the TCP established input helper hook. 293 */ 294 void 295 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 296 { 297 struct tcp_hhook_data hhook_data; 298 299 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 300 hhook_data.tp = tp; 301 hhook_data.th = th; 302 hhook_data.to = to; 303 304 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 305 tp->osd); 306 } 307 } 308 #endif 309 310 /* 311 * CC wrapper hook functions 312 */ 313 void 314 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 315 uint16_t type) 316 { 317 #ifdef STATS 318 int32_t gput; 319 #endif 320 321 INP_WLOCK_ASSERT(tp->t_inpcb); 322 323 tp->ccv->nsegs = nsegs; 324 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 325 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 326 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 327 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 328 tp->ccv->flags |= CCF_CWND_LIMITED; 329 else 330 tp->ccv->flags &= ~CCF_CWND_LIMITED; 331 332 if (type == CC_ACK) { 333 #ifdef STATS 334 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 335 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 336 if (!IN_RECOVERY(tp->t_flags)) 337 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 338 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 339 if ((tp->t_flags & TF_GPUTINPROG) && 340 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 341 /* 342 * Compute goodput in bits per millisecond. 343 */ 344 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) / 345 max(1, tcp_ts_getticks() - tp->gput_ts); 346 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 347 gput); 348 /* 349 * XXXLAS: This is a temporary hack, and should be 350 * chained off VOI_TCP_GPUT when stats(9) grows an API 351 * to deal with chained VOIs. 352 */ 353 if (tp->t_stats_gput_prev > 0) 354 stats_voi_update_abs_s32(tp->t_stats, 355 VOI_TCP_GPUT_ND, 356 ((gput - tp->t_stats_gput_prev) * 100) / 357 tp->t_stats_gput_prev); 358 tp->t_flags &= ~TF_GPUTINPROG; 359 tp->t_stats_gput_prev = gput; 360 } 361 #endif /* STATS */ 362 if (tp->snd_cwnd > tp->snd_ssthresh) { 363 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 364 if (tp->t_bytes_acked >= tp->snd_cwnd) { 365 tp->t_bytes_acked -= tp->snd_cwnd; 366 tp->ccv->flags |= CCF_ABC_SENTAWND; 367 } 368 } else { 369 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 370 tp->t_bytes_acked = 0; 371 } 372 } 373 374 if (CC_ALGO(tp)->ack_received != NULL) { 375 /* XXXLAS: Find a way to live without this */ 376 tp->ccv->curack = th->th_ack; 377 CC_ALGO(tp)->ack_received(tp->ccv, type); 378 } 379 #ifdef STATS 380 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 381 #endif 382 } 383 384 void 385 cc_conn_init(struct tcpcb *tp) 386 { 387 struct hc_metrics_lite metrics; 388 struct inpcb *inp = tp->t_inpcb; 389 u_int maxseg; 390 int rtt; 391 392 INP_WLOCK_ASSERT(tp->t_inpcb); 393 394 tcp_hc_get(&inp->inp_inc, &metrics); 395 maxseg = tcp_maxseg(tp); 396 397 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 398 tp->t_srtt = rtt; 399 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 400 TCPSTAT_INC(tcps_usedrtt); 401 if (metrics.rmx_rttvar) { 402 tp->t_rttvar = metrics.rmx_rttvar; 403 TCPSTAT_INC(tcps_usedrttvar); 404 } else { 405 /* default variation is +- 1 rtt */ 406 tp->t_rttvar = 407 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 408 } 409 TCPT_RANGESET(tp->t_rxtcur, 410 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 411 tp->t_rttmin, TCPTV_REXMTMAX); 412 } 413 if (metrics.rmx_ssthresh) { 414 /* 415 * There's some sort of gateway or interface 416 * buffer limit on the path. Use this to set 417 * the slow start threshold, but set the 418 * threshold to no less than 2*mss. 419 */ 420 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 421 TCPSTAT_INC(tcps_usedssthresh); 422 } 423 424 /* 425 * Set the initial slow-start flight size. 426 * 427 * If a SYN or SYN/ACK was lost and retransmitted, we have to 428 * reduce the initial CWND to one segment as congestion is likely 429 * requiring us to be cautious. 430 */ 431 if (tp->snd_cwnd == 1) 432 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 433 else 434 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 435 436 if (CC_ALGO(tp)->conn_init != NULL) 437 CC_ALGO(tp)->conn_init(tp->ccv); 438 } 439 440 void inline 441 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 442 { 443 INP_WLOCK_ASSERT(tp->t_inpcb); 444 445 #ifdef STATS 446 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 447 #endif 448 449 switch(type) { 450 case CC_NDUPACK: 451 if (!IN_FASTRECOVERY(tp->t_flags)) { 452 tp->snd_recover = tp->snd_max; 453 if (tp->t_flags2 & TF2_ECN_PERMIT) 454 tp->t_flags2 |= TF2_ECN_SND_CWR; 455 } 456 break; 457 case CC_ECN: 458 if (!IN_CONGRECOVERY(tp->t_flags) || 459 /* 460 * Allow ECN reaction on ACK to CWR, if 461 * that data segment was also CE marked. 462 */ 463 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 464 EXIT_CONGRECOVERY(tp->t_flags); 465 TCPSTAT_INC(tcps_ecn_rcwnd); 466 tp->snd_recover = tp->snd_max + 1; 467 if (tp->t_flags2 & TF2_ECN_PERMIT) 468 tp->t_flags2 |= TF2_ECN_SND_CWR; 469 } 470 break; 471 case CC_RTO: 472 tp->t_dupacks = 0; 473 tp->t_bytes_acked = 0; 474 EXIT_RECOVERY(tp->t_flags); 475 if (tp->t_flags2 & TF2_ECN_PERMIT) 476 tp->t_flags2 |= TF2_ECN_SND_CWR; 477 break; 478 case CC_RTO_ERR: 479 TCPSTAT_INC(tcps_sndrexmitbad); 480 /* RTO was unnecessary, so reset everything. */ 481 tp->snd_cwnd = tp->snd_cwnd_prev; 482 tp->snd_ssthresh = tp->snd_ssthresh_prev; 483 tp->snd_recover = tp->snd_recover_prev; 484 if (tp->t_flags & TF_WASFRECOVERY) 485 ENTER_FASTRECOVERY(tp->t_flags); 486 if (tp->t_flags & TF_WASCRECOVERY) 487 ENTER_CONGRECOVERY(tp->t_flags); 488 tp->snd_nxt = tp->snd_max; 489 tp->t_flags &= ~TF_PREVVALID; 490 tp->t_badrxtwin = 0; 491 break; 492 } 493 494 if (CC_ALGO(tp)->cong_signal != NULL) { 495 if (th != NULL) 496 tp->ccv->curack = th->th_ack; 497 CC_ALGO(tp)->cong_signal(tp->ccv, type); 498 } 499 } 500 501 void inline 502 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 503 { 504 INP_WLOCK_ASSERT(tp->t_inpcb); 505 506 /* XXXLAS: KASSERT that we're in recovery? */ 507 508 if (CC_ALGO(tp)->post_recovery != NULL) { 509 tp->ccv->curack = th->th_ack; 510 CC_ALGO(tp)->post_recovery(tp->ccv); 511 } 512 /* XXXLAS: EXIT_RECOVERY ? */ 513 tp->t_bytes_acked = 0; 514 tp->sackhint.delivered_data = 0; 515 tp->sackhint.prr_out = 0; 516 } 517 518 /* 519 * Indicate whether this ack should be delayed. We can delay the ack if 520 * following conditions are met: 521 * - There is no delayed ack timer in progress. 522 * - Our last ack wasn't a 0-sized window. We never want to delay 523 * the ack that opens up a 0-sized window. 524 * - LRO wasn't used for this segment. We make sure by checking that the 525 * segment size is not larger than the MSS. 526 */ 527 #define DELAY_ACK(tp, tlen) \ 528 ((!tcp_timer_active(tp, TT_DELACK) && \ 529 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 530 (tlen <= tp->t_maxseg) && \ 531 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 532 533 void inline 534 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 535 { 536 INP_WLOCK_ASSERT(tp->t_inpcb); 537 538 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 539 switch (iptos & IPTOS_ECN_MASK) { 540 case IPTOS_ECN_CE: 541 tp->ccv->flags |= CCF_IPHDR_CE; 542 break; 543 case IPTOS_ECN_ECT0: 544 /* FALLTHROUGH */ 545 case IPTOS_ECN_ECT1: 546 /* FALLTHROUGH */ 547 case IPTOS_ECN_NOTECT: 548 tp->ccv->flags &= ~CCF_IPHDR_CE; 549 break; 550 } 551 552 if (flags & TH_CWR) 553 tp->ccv->flags |= CCF_TCPHDR_CWR; 554 else 555 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 556 557 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 558 559 if (tp->ccv->flags & CCF_ACKNOW) { 560 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 561 tp->t_flags |= TF_ACKNOW; 562 } 563 } 564 } 565 566 void inline 567 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 568 { 569 cc_ecnpkt_handler_flags(tp, th->th_flags, iptos); 570 } 571 572 /* 573 * TCP input handling is split into multiple parts: 574 * tcp6_input is a thin wrapper around tcp_input for the extended 575 * ip6_protox[] call format in ip6_input 576 * tcp_input handles primary segment validation, inpcb lookup and 577 * SYN processing on listen sockets 578 * tcp_do_segment processes the ACK and text of the segment for 579 * establishing, established and closing connections 580 */ 581 #ifdef INET6 582 int 583 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 584 { 585 struct mbuf *m; 586 struct in6_ifaddr *ia6; 587 struct ip6_hdr *ip6; 588 589 m = *mp; 590 if (m->m_len < *offp + sizeof(struct tcphdr)) { 591 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 592 if (m == NULL) { 593 *mp = m; 594 TCPSTAT_INC(tcps_rcvshort); 595 return (IPPROTO_DONE); 596 } 597 } 598 599 /* 600 * draft-itojun-ipv6-tcp-to-anycast 601 * better place to put this in? 602 */ 603 ip6 = mtod(m, struct ip6_hdr *); 604 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 605 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 606 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 607 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 608 *mp = NULL; 609 return (IPPROTO_DONE); 610 } 611 612 *mp = m; 613 return (tcp_input_with_port(mp, offp, proto, port)); 614 } 615 616 int 617 tcp6_input(struct mbuf **mp, int *offp, int proto) 618 { 619 620 return(tcp6_input_with_port(mp, offp, proto, 0)); 621 } 622 #endif /* INET6 */ 623 624 int 625 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 626 { 627 struct mbuf *m = *mp; 628 struct tcphdr *th = NULL; 629 struct ip *ip = NULL; 630 struct inpcb *inp = NULL; 631 struct tcpcb *tp = NULL; 632 struct socket *so = NULL; 633 u_char *optp = NULL; 634 int off0; 635 int optlen = 0; 636 #ifdef INET 637 int len; 638 uint8_t ipttl; 639 #endif 640 int tlen = 0, off; 641 int drop_hdrlen; 642 int thflags; 643 int rstreason = 0; /* For badport_bandlim accounting purposes */ 644 uint8_t iptos; 645 struct m_tag *fwd_tag = NULL; 646 #ifdef INET6 647 struct ip6_hdr *ip6 = NULL; 648 int isipv6; 649 #else 650 const void *ip6 = NULL; 651 #endif /* INET6 */ 652 struct tcpopt to; /* options in this segment */ 653 char *s = NULL; /* address and port logging */ 654 #ifdef TCPDEBUG 655 /* 656 * The size of tcp_saveipgen must be the size of the max ip header, 657 * now IPv6. 658 */ 659 u_char tcp_saveipgen[IP6_HDR_LEN]; 660 struct tcphdr tcp_savetcp; 661 short ostate = 0; 662 #endif 663 664 NET_EPOCH_ASSERT(); 665 666 #ifdef INET6 667 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 668 #endif 669 670 off0 = *offp; 671 m = *mp; 672 *mp = NULL; 673 to.to_flags = 0; 674 TCPSTAT_INC(tcps_rcvtotal); 675 676 #ifdef INET6 677 if (isipv6) { 678 ip6 = mtod(m, struct ip6_hdr *); 679 th = (struct tcphdr *)((caddr_t)ip6 + off0); 680 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 681 if (port) 682 goto skip6_csum; 683 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 684 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 685 th->th_sum = m->m_pkthdr.csum_data; 686 else 687 th->th_sum = in6_cksum_pseudo(ip6, tlen, 688 IPPROTO_TCP, m->m_pkthdr.csum_data); 689 th->th_sum ^= 0xffff; 690 } else 691 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 692 if (th->th_sum) { 693 TCPSTAT_INC(tcps_rcvbadsum); 694 goto drop; 695 } 696 skip6_csum: 697 /* 698 * Be proactive about unspecified IPv6 address in source. 699 * As we use all-zero to indicate unbounded/unconnected pcb, 700 * unspecified IPv6 address can be used to confuse us. 701 * 702 * Note that packets with unspecified IPv6 destination is 703 * already dropped in ip6_input. 704 */ 705 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 706 /* XXX stat */ 707 goto drop; 708 } 709 iptos = IPV6_TRAFFIC_CLASS(ip6); 710 } 711 #endif 712 #if defined(INET) && defined(INET6) 713 else 714 #endif 715 #ifdef INET 716 { 717 /* 718 * Get IP and TCP header together in first mbuf. 719 * Note: IP leaves IP header in first mbuf. 720 */ 721 if (off0 > sizeof (struct ip)) { 722 ip_stripoptions(m); 723 off0 = sizeof(struct ip); 724 } 725 if (m->m_len < sizeof (struct tcpiphdr)) { 726 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 727 == NULL) { 728 TCPSTAT_INC(tcps_rcvshort); 729 return (IPPROTO_DONE); 730 } 731 } 732 ip = mtod(m, struct ip *); 733 th = (struct tcphdr *)((caddr_t)ip + off0); 734 tlen = ntohs(ip->ip_len) - off0; 735 736 iptos = ip->ip_tos; 737 if (port) 738 goto skip_csum; 739 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 740 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 741 th->th_sum = m->m_pkthdr.csum_data; 742 else 743 th->th_sum = in_pseudo(ip->ip_src.s_addr, 744 ip->ip_dst.s_addr, 745 htonl(m->m_pkthdr.csum_data + tlen + 746 IPPROTO_TCP)); 747 th->th_sum ^= 0xffff; 748 } else { 749 struct ipovly *ipov = (struct ipovly *)ip; 750 751 /* 752 * Checksum extended TCP header and data. 753 */ 754 len = off0 + tlen; 755 ipttl = ip->ip_ttl; 756 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 757 ipov->ih_len = htons(tlen); 758 th->th_sum = in_cksum(m, len); 759 /* Reset length for SDT probes. */ 760 ip->ip_len = htons(len); 761 /* Reset TOS bits */ 762 ip->ip_tos = iptos; 763 /* Re-initialization for later version check */ 764 ip->ip_ttl = ipttl; 765 ip->ip_v = IPVERSION; 766 ip->ip_hl = off0 >> 2; 767 } 768 skip_csum: 769 if (th->th_sum && (port == 0)) { 770 TCPSTAT_INC(tcps_rcvbadsum); 771 goto drop; 772 } 773 } 774 #endif /* INET */ 775 776 /* 777 * Check that TCP offset makes sense, 778 * pull out TCP options and adjust length. XXX 779 */ 780 off = th->th_off << 2; 781 if (off < sizeof (struct tcphdr) || off > tlen) { 782 TCPSTAT_INC(tcps_rcvbadoff); 783 goto drop; 784 } 785 tlen -= off; /* tlen is used instead of ti->ti_len */ 786 if (off > sizeof (struct tcphdr)) { 787 #ifdef INET6 788 if (isipv6) { 789 if (m->m_len < off0 + off) { 790 m = m_pullup(m, off0 + off); 791 if (m == NULL) { 792 TCPSTAT_INC(tcps_rcvshort); 793 return (IPPROTO_DONE); 794 } 795 } 796 ip6 = mtod(m, struct ip6_hdr *); 797 th = (struct tcphdr *)((caddr_t)ip6 + off0); 798 } 799 #endif 800 #if defined(INET) && defined(INET6) 801 else 802 #endif 803 #ifdef INET 804 { 805 if (m->m_len < sizeof(struct ip) + off) { 806 if ((m = m_pullup(m, sizeof (struct ip) + off)) 807 == NULL) { 808 TCPSTAT_INC(tcps_rcvshort); 809 return (IPPROTO_DONE); 810 } 811 ip = mtod(m, struct ip *); 812 th = (struct tcphdr *)((caddr_t)ip + off0); 813 } 814 } 815 #endif 816 optlen = off - sizeof (struct tcphdr); 817 optp = (u_char *)(th + 1); 818 } 819 thflags = th->th_flags; 820 821 /* 822 * Convert TCP protocol specific fields to host format. 823 */ 824 tcp_fields_to_host(th); 825 826 /* 827 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 828 */ 829 drop_hdrlen = off0 + off; 830 831 /* 832 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 833 */ 834 if ( 835 #ifdef INET6 836 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 837 #ifdef INET 838 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 839 #endif 840 #endif 841 #if defined(INET) && !defined(INET6) 842 (m->m_flags & M_IP_NEXTHOP) 843 #endif 844 ) 845 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 846 847 findpcb: 848 #ifdef INET6 849 if (isipv6 && fwd_tag != NULL) { 850 struct sockaddr_in6 *next_hop6; 851 852 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 853 /* 854 * Transparently forwarded. Pretend to be the destination. 855 * Already got one like this? 856 */ 857 inp = in6_pcblookup_mbuf(&V_tcbinfo, 858 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 859 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 860 if (!inp) { 861 /* 862 * It's new. Try to find the ambushing socket. 863 * Because we've rewritten the destination address, 864 * any hardware-generated hash is ignored. 865 */ 866 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 867 th->th_sport, &next_hop6->sin6_addr, 868 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 869 th->th_dport, INPLOOKUP_WILDCARD | 870 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 871 } 872 } else if (isipv6) { 873 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 874 th->th_sport, &ip6->ip6_dst, th->th_dport, 875 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 876 m->m_pkthdr.rcvif, m); 877 } 878 #endif /* INET6 */ 879 #if defined(INET6) && defined(INET) 880 else 881 #endif 882 #ifdef INET 883 if (fwd_tag != NULL) { 884 struct sockaddr_in *next_hop; 885 886 next_hop = (struct sockaddr_in *)(fwd_tag+1); 887 /* 888 * Transparently forwarded. Pretend to be the destination. 889 * already got one like this? 890 */ 891 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 892 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 893 m->m_pkthdr.rcvif, m); 894 if (!inp) { 895 /* 896 * It's new. Try to find the ambushing socket. 897 * Because we've rewritten the destination address, 898 * any hardware-generated hash is ignored. 899 */ 900 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 901 th->th_sport, next_hop->sin_addr, 902 next_hop->sin_port ? ntohs(next_hop->sin_port) : 903 th->th_dport, INPLOOKUP_WILDCARD | 904 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 905 } 906 } else 907 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 908 th->th_sport, ip->ip_dst, th->th_dport, 909 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 910 m->m_pkthdr.rcvif, m); 911 #endif /* INET */ 912 913 /* 914 * If the INPCB does not exist then all data in the incoming 915 * segment is discarded and an appropriate RST is sent back. 916 * XXX MRT Send RST using which routing table? 917 */ 918 if (inp == NULL) { 919 /* 920 * Log communication attempts to ports that are not 921 * in use. 922 */ 923 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 924 V_tcp_log_in_vain == 2) { 925 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 926 log(LOG_INFO, "%s; %s: Connection attempt " 927 "to closed port\n", s, __func__); 928 } 929 /* 930 * When blackholing do not respond with a RST but 931 * completely ignore the segment and drop it. 932 */ 933 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 934 V_blackhole == 2) 935 goto dropunlock; 936 937 rstreason = BANDLIM_RST_CLOSEDPORT; 938 goto dropwithreset; 939 } 940 INP_WLOCK_ASSERT(inp); 941 /* 942 * While waiting for inp lock during the lookup, another thread 943 * can have dropped the inpcb, in which case we need to loop back 944 * and try to find a new inpcb to deliver to. 945 */ 946 if (inp->inp_flags & INP_DROPPED) { 947 INP_WUNLOCK(inp); 948 inp = NULL; 949 goto findpcb; 950 } 951 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 952 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 953 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) { 954 inp->inp_flowid = m->m_pkthdr.flowid; 955 inp->inp_flowtype = M_HASHTYPE_GET(m); 956 } 957 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 958 #ifdef INET6 959 if (isipv6 && IPSEC_ENABLED(ipv6) && 960 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 961 goto dropunlock; 962 } 963 #ifdef INET 964 else 965 #endif 966 #endif /* INET6 */ 967 #ifdef INET 968 if (IPSEC_ENABLED(ipv4) && 969 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 970 goto dropunlock; 971 } 972 #endif /* INET */ 973 #endif /* IPSEC */ 974 975 /* 976 * Check the minimum TTL for socket. 977 */ 978 if (inp->inp_ip_minttl != 0) { 979 #ifdef INET6 980 if (isipv6) { 981 if (inp->inp_ip_minttl > ip6->ip6_hlim) 982 goto dropunlock; 983 } else 984 #endif 985 if (inp->inp_ip_minttl > ip->ip_ttl) 986 goto dropunlock; 987 } 988 989 /* 990 * A previous connection in TIMEWAIT state is supposed to catch stray 991 * or duplicate segments arriving late. If this segment was a 992 * legitimate new connection attempt, the old INPCB gets removed and 993 * we can try again to find a listening socket. 994 * 995 * At this point, due to earlier optimism, we may hold only an inpcb 996 * lock, and not the inpcbinfo write lock. If so, we need to try to 997 * acquire it, or if that fails, acquire a reference on the inpcb, 998 * drop all locks, acquire a global write lock, and then re-acquire 999 * the inpcb lock. We may at that point discover that another thread 1000 * has tried to free the inpcb, in which case we need to loop back 1001 * and try to find a new inpcb to deliver to. 1002 * 1003 * XXXRW: It may be time to rethink timewait locking. 1004 */ 1005 if (inp->inp_flags & INP_TIMEWAIT) { 1006 tcp_dooptions(&to, optp, optlen, 1007 (thflags & TH_SYN) ? TO_SYN : 0); 1008 /* 1009 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1010 */ 1011 if (tcp_twcheck(inp, &to, th, m, tlen)) 1012 goto findpcb; 1013 return (IPPROTO_DONE); 1014 } 1015 /* 1016 * The TCPCB may no longer exist if the connection is winding 1017 * down or it is in the CLOSED state. Either way we drop the 1018 * segment and send an appropriate response. 1019 */ 1020 tp = intotcpcb(inp); 1021 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1022 rstreason = BANDLIM_RST_CLOSEDPORT; 1023 goto dropwithreset; 1024 } 1025 1026 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1027 rstreason = BANDLIM_RST_CLOSEDPORT; 1028 goto dropwithreset; 1029 } 1030 1031 #ifdef TCP_OFFLOAD 1032 if (tp->t_flags & TF_TOE) { 1033 tcp_offload_input(tp, m); 1034 m = NULL; /* consumed by the TOE driver */ 1035 goto dropunlock; 1036 } 1037 #endif 1038 1039 #ifdef MAC 1040 INP_WLOCK_ASSERT(inp); 1041 if (mac_inpcb_check_deliver(inp, m)) 1042 goto dropunlock; 1043 #endif 1044 so = inp->inp_socket; 1045 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1046 #ifdef TCPDEBUG 1047 if (so->so_options & SO_DEBUG) { 1048 ostate = tp->t_state; 1049 #ifdef INET6 1050 if (isipv6) { 1051 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1052 } else 1053 #endif 1054 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1055 tcp_savetcp = *th; 1056 } 1057 #endif /* TCPDEBUG */ 1058 /* 1059 * When the socket is accepting connections (the INPCB is in LISTEN 1060 * state) we look into the SYN cache if this is a new connection 1061 * attempt or the completion of a previous one. 1062 */ 1063 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1064 ("%s: so accepting but tp %p not listening", __func__, tp)); 1065 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1066 struct in_conninfo inc; 1067 1068 bzero(&inc, sizeof(inc)); 1069 #ifdef INET6 1070 if (isipv6) { 1071 inc.inc_flags |= INC_ISIPV6; 1072 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1073 inc.inc_flags |= INC_IPV6MINMTU; 1074 inc.inc6_faddr = ip6->ip6_src; 1075 inc.inc6_laddr = ip6->ip6_dst; 1076 } else 1077 #endif 1078 { 1079 inc.inc_faddr = ip->ip_src; 1080 inc.inc_laddr = ip->ip_dst; 1081 } 1082 inc.inc_fport = th->th_sport; 1083 inc.inc_lport = th->th_dport; 1084 inc.inc_fibnum = so->so_fibnum; 1085 1086 /* 1087 * Check for an existing connection attempt in syncache if 1088 * the flag is only ACK. A successful lookup creates a new 1089 * socket appended to the listen queue in SYN_RECEIVED state. 1090 */ 1091 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1092 /* 1093 * Parse the TCP options here because 1094 * syncookies need access to the reflected 1095 * timestamp. 1096 */ 1097 tcp_dooptions(&to, optp, optlen, 0); 1098 /* 1099 * NB: syncache_expand() doesn't unlock 1100 * inp and tcpinfo locks. 1101 */ 1102 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1103 if (rstreason < 0) { 1104 /* 1105 * A failing TCP MD5 signature comparison 1106 * must result in the segment being dropped 1107 * and must not produce any response back 1108 * to the sender. 1109 */ 1110 goto dropunlock; 1111 } else if (rstreason == 0) { 1112 /* 1113 * No syncache entry or ACK was not 1114 * for our SYN/ACK. Send a RST. 1115 * NB: syncache did its own logging 1116 * of the failure cause. 1117 */ 1118 rstreason = BANDLIM_RST_OPENPORT; 1119 goto dropwithreset; 1120 } 1121 tfo_socket_result: 1122 if (so == NULL) { 1123 /* 1124 * We completed the 3-way handshake 1125 * but could not allocate a socket 1126 * either due to memory shortage, 1127 * listen queue length limits or 1128 * global socket limits. Send RST 1129 * or wait and have the remote end 1130 * retransmit the ACK for another 1131 * try. 1132 */ 1133 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1134 log(LOG_DEBUG, "%s; %s: Listen socket: " 1135 "Socket allocation failed due to " 1136 "limits or memory shortage, %s\n", 1137 s, __func__, 1138 V_tcp_sc_rst_sock_fail ? 1139 "sending RST" : "try again"); 1140 if (V_tcp_sc_rst_sock_fail) { 1141 rstreason = BANDLIM_UNLIMITED; 1142 goto dropwithreset; 1143 } else 1144 goto dropunlock; 1145 } 1146 /* 1147 * Socket is created in state SYN_RECEIVED. 1148 * Unlock the listen socket, lock the newly 1149 * created socket and update the tp variable. 1150 */ 1151 INP_WUNLOCK(inp); /* listen socket */ 1152 inp = sotoinpcb(so); 1153 /* 1154 * New connection inpcb is already locked by 1155 * syncache_expand(). 1156 */ 1157 INP_WLOCK_ASSERT(inp); 1158 tp = intotcpcb(inp); 1159 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1160 ("%s: ", __func__)); 1161 /* 1162 * Process the segment and the data it 1163 * contains. tcp_do_segment() consumes 1164 * the mbuf chain and unlocks the inpcb. 1165 */ 1166 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1167 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1168 iptos); 1169 return (IPPROTO_DONE); 1170 } 1171 /* 1172 * Segment flag validation for new connection attempts: 1173 * 1174 * Our (SYN|ACK) response was rejected. 1175 * Check with syncache and remove entry to prevent 1176 * retransmits. 1177 * 1178 * NB: syncache_chkrst does its own logging of failure 1179 * causes. 1180 */ 1181 if (thflags & TH_RST) { 1182 syncache_chkrst(&inc, th, m, port); 1183 goto dropunlock; 1184 } 1185 /* 1186 * We can't do anything without SYN. 1187 */ 1188 if ((thflags & TH_SYN) == 0) { 1189 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1190 log(LOG_DEBUG, "%s; %s: Listen socket: " 1191 "SYN is missing, segment ignored\n", 1192 s, __func__); 1193 TCPSTAT_INC(tcps_badsyn); 1194 goto dropunlock; 1195 } 1196 /* 1197 * (SYN|ACK) is bogus on a listen socket. 1198 */ 1199 if (thflags & TH_ACK) { 1200 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1201 log(LOG_DEBUG, "%s; %s: Listen socket: " 1202 "SYN|ACK invalid, segment rejected\n", 1203 s, __func__); 1204 syncache_badack(&inc, port); /* XXX: Not needed! */ 1205 TCPSTAT_INC(tcps_badsyn); 1206 rstreason = BANDLIM_RST_OPENPORT; 1207 goto dropwithreset; 1208 } 1209 /* 1210 * If the drop_synfin option is enabled, drop all 1211 * segments with both the SYN and FIN bits set. 1212 * This prevents e.g. nmap from identifying the 1213 * TCP/IP stack. 1214 * XXX: Poor reasoning. nmap has other methods 1215 * and is constantly refining its stack detection 1216 * strategies. 1217 * XXX: This is a violation of the TCP specification 1218 * and was used by RFC1644. 1219 */ 1220 if ((thflags & TH_FIN) && V_drop_synfin) { 1221 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1222 log(LOG_DEBUG, "%s; %s: Listen socket: " 1223 "SYN|FIN segment ignored (based on " 1224 "sysctl setting)\n", s, __func__); 1225 TCPSTAT_INC(tcps_badsyn); 1226 goto dropunlock; 1227 } 1228 /* 1229 * Segment's flags are (SYN) or (SYN|FIN). 1230 * 1231 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1232 * as they do not affect the state of the TCP FSM. 1233 * The data pointed to by TH_URG and th_urp is ignored. 1234 */ 1235 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1236 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1237 KASSERT(thflags & (TH_SYN), 1238 ("%s: Listen socket: TH_SYN not set", __func__)); 1239 #ifdef INET6 1240 /* 1241 * If deprecated address is forbidden, 1242 * we do not accept SYN to deprecated interface 1243 * address to prevent any new inbound connection from 1244 * getting established. 1245 * When we do not accept SYN, we send a TCP RST, 1246 * with deprecated source address (instead of dropping 1247 * it). We compromise it as it is much better for peer 1248 * to send a RST, and RST will be the final packet 1249 * for the exchange. 1250 * 1251 * If we do not forbid deprecated addresses, we accept 1252 * the SYN packet. RFC2462 does not suggest dropping 1253 * SYN in this case. 1254 * If we decipher RFC2462 5.5.4, it says like this: 1255 * 1. use of deprecated addr with existing 1256 * communication is okay - "SHOULD continue to be 1257 * used" 1258 * 2. use of it with new communication: 1259 * (2a) "SHOULD NOT be used if alternate address 1260 * with sufficient scope is available" 1261 * (2b) nothing mentioned otherwise. 1262 * Here we fall into (2b) case as we have no choice in 1263 * our source address selection - we must obey the peer. 1264 * 1265 * The wording in RFC2462 is confusing, and there are 1266 * multiple description text for deprecated address 1267 * handling - worse, they are not exactly the same. 1268 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1269 */ 1270 if (isipv6 && !V_ip6_use_deprecated) { 1271 struct in6_ifaddr *ia6; 1272 1273 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1274 if (ia6 != NULL && 1275 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1276 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1277 log(LOG_DEBUG, "%s; %s: Listen socket: " 1278 "Connection attempt to deprecated " 1279 "IPv6 address rejected\n", 1280 s, __func__); 1281 rstreason = BANDLIM_RST_OPENPORT; 1282 goto dropwithreset; 1283 } 1284 } 1285 #endif /* INET6 */ 1286 /* 1287 * Basic sanity checks on incoming SYN requests: 1288 * Don't respond if the destination is a link layer 1289 * broadcast according to RFC1122 4.2.3.10, p. 104. 1290 * If it is from this socket it must be forged. 1291 * Don't respond if the source or destination is a 1292 * global or subnet broad- or multicast address. 1293 * Note that it is quite possible to receive unicast 1294 * link-layer packets with a broadcast IP address. Use 1295 * in_broadcast() to find them. 1296 */ 1297 if (m->m_flags & (M_BCAST|M_MCAST)) { 1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1299 log(LOG_DEBUG, "%s; %s: Listen socket: " 1300 "Connection attempt from broad- or multicast " 1301 "link layer address ignored\n", s, __func__); 1302 goto dropunlock; 1303 } 1304 #ifdef INET6 1305 if (isipv6) { 1306 if (th->th_dport == th->th_sport && 1307 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1309 log(LOG_DEBUG, "%s; %s: Listen socket: " 1310 "Connection attempt to/from self " 1311 "ignored\n", s, __func__); 1312 goto dropunlock; 1313 } 1314 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1315 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1316 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1317 log(LOG_DEBUG, "%s; %s: Listen socket: " 1318 "Connection attempt from/to multicast " 1319 "address ignored\n", s, __func__); 1320 goto dropunlock; 1321 } 1322 } 1323 #endif 1324 #if defined(INET) && defined(INET6) 1325 else 1326 #endif 1327 #ifdef INET 1328 { 1329 if (th->th_dport == th->th_sport && 1330 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1331 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1332 log(LOG_DEBUG, "%s; %s: Listen socket: " 1333 "Connection attempt from/to self " 1334 "ignored\n", s, __func__); 1335 goto dropunlock; 1336 } 1337 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1338 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1339 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1340 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1341 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1342 log(LOG_DEBUG, "%s; %s: Listen socket: " 1343 "Connection attempt from/to broad- " 1344 "or multicast address ignored\n", 1345 s, __func__); 1346 goto dropunlock; 1347 } 1348 } 1349 #endif 1350 /* 1351 * SYN appears to be valid. Create compressed TCP state 1352 * for syncache. 1353 */ 1354 #ifdef TCPDEBUG 1355 if (so->so_options & SO_DEBUG) 1356 tcp_trace(TA_INPUT, ostate, tp, 1357 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1358 #endif 1359 TCP_PROBE3(debug__input, tp, th, m); 1360 tcp_dooptions(&to, optp, optlen, TO_SYN); 1361 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos, 1362 port)) 1363 goto tfo_socket_result; 1364 1365 /* 1366 * Entry added to syncache and mbuf consumed. 1367 * Only the listen socket is unlocked by syncache_add(). 1368 */ 1369 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1370 return (IPPROTO_DONE); 1371 } else if (tp->t_state == TCPS_LISTEN) { 1372 /* 1373 * When a listen socket is torn down the SO_ACCEPTCONN 1374 * flag is removed first while connections are drained 1375 * from the accept queue in a unlock/lock cycle of the 1376 * ACCEPT_LOCK, opening a race condition allowing a SYN 1377 * attempt go through unhandled. 1378 */ 1379 goto dropunlock; 1380 } 1381 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1382 if (tp->t_flags & TF_SIGNATURE) { 1383 tcp_dooptions(&to, optp, optlen, thflags); 1384 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1385 TCPSTAT_INC(tcps_sig_err_nosigopt); 1386 goto dropunlock; 1387 } 1388 if (!TCPMD5_ENABLED() || 1389 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1390 goto dropunlock; 1391 } 1392 #endif 1393 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1394 1395 /* 1396 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1397 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1398 * the inpcb, and unlocks pcbinfo. 1399 */ 1400 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1401 return (IPPROTO_DONE); 1402 1403 dropwithreset: 1404 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1405 1406 if (inp != NULL) { 1407 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1408 INP_WUNLOCK(inp); 1409 } else 1410 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1411 m = NULL; /* mbuf chain got consumed. */ 1412 goto drop; 1413 1414 dropunlock: 1415 if (m != NULL) 1416 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1417 1418 if (inp != NULL) 1419 INP_WUNLOCK(inp); 1420 1421 drop: 1422 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1423 if (s != NULL) 1424 free(s, M_TCPLOG); 1425 if (m != NULL) 1426 m_freem(m); 1427 return (IPPROTO_DONE); 1428 } 1429 1430 /* 1431 * Automatic sizing of receive socket buffer. Often the send 1432 * buffer size is not optimally adjusted to the actual network 1433 * conditions at hand (delay bandwidth product). Setting the 1434 * buffer size too small limits throughput on links with high 1435 * bandwidth and high delay (eg. trans-continental/oceanic links). 1436 * 1437 * On the receive side the socket buffer memory is only rarely 1438 * used to any significant extent. This allows us to be much 1439 * more aggressive in scaling the receive socket buffer. For 1440 * the case that the buffer space is actually used to a large 1441 * extent and we run out of kernel memory we can simply drop 1442 * the new segments; TCP on the sender will just retransmit it 1443 * later. Setting the buffer size too big may only consume too 1444 * much kernel memory if the application doesn't read() from 1445 * the socket or packet loss or reordering makes use of the 1446 * reassembly queue. 1447 * 1448 * The criteria to step up the receive buffer one notch are: 1449 * 1. Application has not set receive buffer size with 1450 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1451 * 2. the number of bytes received during 1/2 of an sRTT 1452 * is at least 3/8 of the current socket buffer size. 1453 * 3. receive buffer size has not hit maximal automatic size; 1454 * 1455 * If all of the criteria are met we increaset the socket buffer 1456 * by a 1/2 (bounded by the max). This allows us to keep ahead 1457 * of slow-start but also makes it so our peer never gets limited 1458 * by our rwnd which we then open up causing a burst. 1459 * 1460 * This algorithm does two steps per RTT at most and only if 1461 * we receive a bulk stream w/o packet losses or reorderings. 1462 * Shrinking the buffer during idle times is not necessary as 1463 * it doesn't consume any memory when idle. 1464 * 1465 * TODO: Only step up if the application is actually serving 1466 * the buffer to better manage the socket buffer resources. 1467 */ 1468 int 1469 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1470 struct tcpcb *tp, int tlen) 1471 { 1472 int newsize = 0; 1473 1474 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1475 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1476 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1477 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1478 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1479 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1480 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1481 } 1482 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1483 1484 /* Start over with next RTT. */ 1485 tp->rfbuf_ts = 0; 1486 tp->rfbuf_cnt = 0; 1487 } else { 1488 tp->rfbuf_cnt += tlen; /* add up */ 1489 } 1490 return (newsize); 1491 } 1492 1493 int 1494 tcp_input(struct mbuf **mp, int *offp, int proto) 1495 { 1496 return(tcp_input_with_port(mp, offp, proto, 0)); 1497 } 1498 1499 void 1500 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1501 { 1502 /* 1503 * Since tp might be gone if the session entered 1504 * the TIME_WAIT state before coming here, we need 1505 * to check if the socket is still connected. 1506 */ 1507 if (tp == NULL) { 1508 return; 1509 } 1510 if (so == NULL) { 1511 return; 1512 } 1513 INP_LOCK_ASSERT(tp->t_inpcb); 1514 if (tp->t_flags & TF_WAKESOR) { 1515 tp->t_flags &= ~TF_WAKESOR; 1516 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1517 sorwakeup_locked(so); 1518 } 1519 } 1520 1521 void 1522 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1523 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1524 { 1525 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1526 int rstreason, todrop, win, incforsyn = 0; 1527 uint32_t tiwin; 1528 uint16_t nsegs; 1529 char *s; 1530 struct in_conninfo *inc; 1531 struct mbuf *mfree; 1532 struct tcpopt to; 1533 int tfo_syn; 1534 u_int maxseg; 1535 1536 #ifdef TCPDEBUG 1537 /* 1538 * The size of tcp_saveipgen must be the size of the max ip header, 1539 * now IPv6. 1540 */ 1541 u_char tcp_saveipgen[IP6_HDR_LEN]; 1542 struct tcphdr tcp_savetcp; 1543 short ostate = 0; 1544 #endif 1545 thflags = th->th_flags; 1546 inc = &tp->t_inpcb->inp_inc; 1547 tp->sackhint.last_sack_ack = 0; 1548 sack_changed = 0; 1549 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1550 1551 NET_EPOCH_ASSERT(); 1552 INP_WLOCK_ASSERT(tp->t_inpcb); 1553 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1554 __func__)); 1555 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1556 __func__)); 1557 1558 #ifdef TCPPCAP 1559 /* Save segment, if requested. */ 1560 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1561 #endif 1562 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1563 tlen, NULL, true); 1564 1565 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1566 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1567 log(LOG_DEBUG, "%s; %s: " 1568 "SYN|FIN segment ignored (based on " 1569 "sysctl setting)\n", s, __func__); 1570 free(s, M_TCPLOG); 1571 } 1572 goto drop; 1573 } 1574 1575 /* 1576 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1577 * check SEQ.ACK first. 1578 */ 1579 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1580 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1581 rstreason = BANDLIM_UNLIMITED; 1582 goto dropwithreset; 1583 } 1584 1585 /* 1586 * Segment received on connection. 1587 * Reset idle time and keep-alive timer. 1588 * XXX: This should be done after segment 1589 * validation to ignore broken/spoofed segs. 1590 */ 1591 tp->t_rcvtime = ticks; 1592 1593 /* 1594 * Scale up the window into a 32-bit value. 1595 * For the SYN_SENT state the scale is zero. 1596 */ 1597 tiwin = th->th_win << tp->snd_scale; 1598 #ifdef STATS 1599 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1600 #endif 1601 1602 /* 1603 * TCP ECN processing. 1604 */ 1605 if (tp->t_flags2 & TF2_ECN_PERMIT) { 1606 if (thflags & TH_CWR) { 1607 tp->t_flags2 &= ~TF2_ECN_SND_ECE; 1608 tp->t_flags |= TF_ACKNOW; 1609 } 1610 switch (iptos & IPTOS_ECN_MASK) { 1611 case IPTOS_ECN_CE: 1612 tp->t_flags2 |= TF2_ECN_SND_ECE; 1613 TCPSTAT_INC(tcps_ecn_ce); 1614 break; 1615 case IPTOS_ECN_ECT0: 1616 TCPSTAT_INC(tcps_ecn_ect0); 1617 break; 1618 case IPTOS_ECN_ECT1: 1619 TCPSTAT_INC(tcps_ecn_ect1); 1620 break; 1621 } 1622 1623 /* Process a packet differently from RFC3168. */ 1624 cc_ecnpkt_handler(tp, th, iptos); 1625 1626 /* Congestion experienced. */ 1627 if (thflags & TH_ECE) { 1628 cc_cong_signal(tp, th, CC_ECN); 1629 } 1630 } 1631 1632 /* 1633 * Parse options on any incoming segment. 1634 */ 1635 tcp_dooptions(&to, (u_char *)(th + 1), 1636 (th->th_off << 2) - sizeof(struct tcphdr), 1637 (thflags & TH_SYN) ? TO_SYN : 0); 1638 1639 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1640 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1641 (to.to_flags & TOF_SIGNATURE) == 0) { 1642 TCPSTAT_INC(tcps_sig_err_sigopt); 1643 /* XXX: should drop? */ 1644 } 1645 #endif 1646 /* 1647 * If echoed timestamp is later than the current time, 1648 * fall back to non RFC1323 RTT calculation. Normalize 1649 * timestamp if syncookies were used when this connection 1650 * was established. 1651 */ 1652 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1653 to.to_tsecr -= tp->ts_offset; 1654 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1655 to.to_tsecr = 0; 1656 else if (tp->t_rxtshift == 1 && 1657 tp->t_flags & TF_PREVVALID && 1658 tp->t_badrxtwin != 0 && 1659 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 1660 cc_cong_signal(tp, th, CC_RTO_ERR); 1661 } 1662 /* 1663 * Process options only when we get SYN/ACK back. The SYN case 1664 * for incoming connections is handled in tcp_syncache. 1665 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1666 * or <SYN,ACK>) segment itself is never scaled. 1667 * XXX this is traditional behavior, may need to be cleaned up. 1668 */ 1669 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1670 /* Handle parallel SYN for ECN */ 1671 if (!(thflags & TH_ACK) && 1672 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) && 1673 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) { 1674 tp->t_flags2 |= TF2_ECN_PERMIT; 1675 tp->t_flags2 |= TF2_ECN_SND_ECE; 1676 TCPSTAT_INC(tcps_ecn_shs); 1677 } 1678 if ((to.to_flags & TOF_SCALE) && 1679 (tp->t_flags & TF_REQ_SCALE) && 1680 !(tp->t_flags & TF_NOOPT)) { 1681 tp->t_flags |= TF_RCVD_SCALE; 1682 tp->snd_scale = to.to_wscale; 1683 } else 1684 tp->t_flags &= ~TF_REQ_SCALE; 1685 /* 1686 * Initial send window. It will be updated with 1687 * the next incoming segment to the scaled value. 1688 */ 1689 tp->snd_wnd = th->th_win; 1690 if ((to.to_flags & TOF_TS) && 1691 (tp->t_flags & TF_REQ_TSTMP) && 1692 !(tp->t_flags & TF_NOOPT)) { 1693 tp->t_flags |= TF_RCVD_TSTMP; 1694 tp->ts_recent = to.to_tsval; 1695 tp->ts_recent_age = tcp_ts_getticks(); 1696 } else 1697 tp->t_flags &= ~TF_REQ_TSTMP; 1698 if (to.to_flags & TOF_MSS) 1699 tcp_mss(tp, to.to_mss); 1700 if ((tp->t_flags & TF_SACK_PERMIT) && 1701 (!(to.to_flags & TOF_SACKPERM) || 1702 (tp->t_flags & TF_NOOPT))) 1703 tp->t_flags &= ~TF_SACK_PERMIT; 1704 if (IS_FASTOPEN(tp->t_flags)) { 1705 if ((to.to_flags & TOF_FASTOPEN) && 1706 !(tp->t_flags & TF_NOOPT)) { 1707 uint16_t mss; 1708 1709 if (to.to_flags & TOF_MSS) 1710 mss = to.to_mss; 1711 else 1712 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1713 mss = TCP6_MSS; 1714 else 1715 mss = TCP_MSS; 1716 tcp_fastopen_update_cache(tp, mss, 1717 to.to_tfo_len, to.to_tfo_cookie); 1718 } else 1719 tcp_fastopen_disable_path(tp); 1720 } 1721 } 1722 1723 /* 1724 * If timestamps were negotiated during SYN/ACK and a 1725 * segment without a timestamp is received, silently drop 1726 * the segment, unless it is a RST segment or missing timestamps are 1727 * tolerated. 1728 * See section 3.2 of RFC 7323. 1729 */ 1730 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1731 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1732 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1733 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1734 "segment processed normally\n", 1735 s, __func__); 1736 free(s, M_TCPLOG); 1737 } 1738 } else { 1739 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1740 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1741 "segment silently dropped\n", s, __func__); 1742 free(s, M_TCPLOG); 1743 } 1744 goto drop; 1745 } 1746 } 1747 /* 1748 * If timestamps were not negotiated during SYN/ACK and a 1749 * segment with a timestamp is received, ignore the 1750 * timestamp and process the packet normally. 1751 * See section 3.2 of RFC 7323. 1752 */ 1753 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1754 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1755 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1756 "segment processed normally\n", s, __func__); 1757 free(s, M_TCPLOG); 1758 } 1759 } 1760 1761 /* 1762 * Header prediction: check for the two common cases 1763 * of a uni-directional data xfer. If the packet has 1764 * no control flags, is in-sequence, the window didn't 1765 * change and we're not retransmitting, it's a 1766 * candidate. If the length is zero and the ack moved 1767 * forward, we're the sender side of the xfer. Just 1768 * free the data acked & wake any higher level process 1769 * that was blocked waiting for space. If the length 1770 * is non-zero and the ack didn't move, we're the 1771 * receiver side. If we're getting packets in-order 1772 * (the reassembly queue is empty), add the data to 1773 * the socket buffer and note that we need a delayed ack. 1774 * Make sure that the hidden state-flags are also off. 1775 * Since we check for TCPS_ESTABLISHED first, it can only 1776 * be TH_NEEDSYN. 1777 */ 1778 if (tp->t_state == TCPS_ESTABLISHED && 1779 th->th_seq == tp->rcv_nxt && 1780 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1781 tp->snd_nxt == tp->snd_max && 1782 tiwin && tiwin == tp->snd_wnd && 1783 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1784 SEGQ_EMPTY(tp) && 1785 ((to.to_flags & TOF_TS) == 0 || 1786 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1787 /* 1788 * If last ACK falls within this segment's sequence numbers, 1789 * record the timestamp. 1790 * NOTE that the test is modified according to the latest 1791 * proposal of the [email protected] list (Braden 1993/04/26). 1792 */ 1793 if ((to.to_flags & TOF_TS) != 0 && 1794 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1795 tp->ts_recent_age = tcp_ts_getticks(); 1796 tp->ts_recent = to.to_tsval; 1797 } 1798 1799 if (tlen == 0) { 1800 if (SEQ_GT(th->th_ack, tp->snd_una) && 1801 SEQ_LEQ(th->th_ack, tp->snd_max) && 1802 !IN_RECOVERY(tp->t_flags) && 1803 (to.to_flags & TOF_SACK) == 0 && 1804 TAILQ_EMPTY(&tp->snd_holes)) { 1805 /* 1806 * This is a pure ack for outstanding data. 1807 */ 1808 TCPSTAT_INC(tcps_predack); 1809 1810 /* 1811 * "bad retransmit" recovery without timestamps. 1812 */ 1813 if ((to.to_flags & TOF_TS) == 0 && 1814 tp->t_rxtshift == 1 && 1815 tp->t_flags & TF_PREVVALID && 1816 tp->t_badrxtwin != 0 && 1817 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1818 cc_cong_signal(tp, th, CC_RTO_ERR); 1819 } 1820 1821 /* 1822 * Recalculate the transmit timer / rtt. 1823 * 1824 * Some boxes send broken timestamp replies 1825 * during the SYN+ACK phase, ignore 1826 * timestamps of 0 or we could calculate a 1827 * huge RTT and blow up the retransmit timer. 1828 */ 1829 if ((to.to_flags & TOF_TS) != 0 && 1830 to.to_tsecr) { 1831 uint32_t t; 1832 1833 t = tcp_ts_getticks() - to.to_tsecr; 1834 if (!tp->t_rttlow || tp->t_rttlow > t) 1835 tp->t_rttlow = t; 1836 tcp_xmit_timer(tp, 1837 TCP_TS_TO_TICKS(t) + 1); 1838 } else if (tp->t_rtttime && 1839 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1840 if (!tp->t_rttlow || 1841 tp->t_rttlow > ticks - tp->t_rtttime) 1842 tp->t_rttlow = ticks - tp->t_rtttime; 1843 tcp_xmit_timer(tp, 1844 ticks - tp->t_rtttime); 1845 } 1846 acked = BYTES_THIS_ACK(tp, th); 1847 1848 #ifdef TCP_HHOOK 1849 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1850 hhook_run_tcp_est_in(tp, th, &to); 1851 #endif 1852 1853 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1854 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1855 sbdrop(&so->so_snd, acked); 1856 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1857 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1858 tp->snd_recover = th->th_ack - 1; 1859 1860 /* 1861 * Let the congestion control algorithm update 1862 * congestion control related information. This 1863 * typically means increasing the congestion 1864 * window. 1865 */ 1866 cc_ack_received(tp, th, nsegs, CC_ACK); 1867 1868 tp->snd_una = th->th_ack; 1869 /* 1870 * Pull snd_wl2 up to prevent seq wrap relative 1871 * to th_ack. 1872 */ 1873 tp->snd_wl2 = th->th_ack; 1874 tp->t_dupacks = 0; 1875 m_freem(m); 1876 1877 /* 1878 * If all outstanding data are acked, stop 1879 * retransmit timer, otherwise restart timer 1880 * using current (possibly backed-off) value. 1881 * If process is waiting for space, 1882 * wakeup/selwakeup/signal. If data 1883 * are ready to send, let tcp_output 1884 * decide between more output or persist. 1885 */ 1886 #ifdef TCPDEBUG 1887 if (so->so_options & SO_DEBUG) 1888 tcp_trace(TA_INPUT, ostate, tp, 1889 (void *)tcp_saveipgen, 1890 &tcp_savetcp, 0); 1891 #endif 1892 TCP_PROBE3(debug__input, tp, th, m); 1893 if (tp->snd_una == tp->snd_max) 1894 tcp_timer_activate(tp, TT_REXMT, 0); 1895 else if (!tcp_timer_active(tp, TT_PERSIST)) 1896 tcp_timer_activate(tp, TT_REXMT, 1897 tp->t_rxtcur); 1898 sowwakeup(so); 1899 if (sbavail(&so->so_snd)) 1900 (void) tp->t_fb->tfb_tcp_output(tp); 1901 goto check_delack; 1902 } 1903 } else if (th->th_ack == tp->snd_una && 1904 tlen <= sbspace(&so->so_rcv)) { 1905 int newsize = 0; /* automatic sockbuf scaling */ 1906 1907 /* 1908 * This is a pure, in-sequence data packet with 1909 * nothing on the reassembly queue and we have enough 1910 * buffer space to take it. 1911 */ 1912 /* Clean receiver SACK report if present */ 1913 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1914 tcp_clean_sackreport(tp); 1915 TCPSTAT_INC(tcps_preddat); 1916 tp->rcv_nxt += tlen; 1917 if (tlen && 1918 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1919 (tp->t_fbyte_in == 0)) { 1920 tp->t_fbyte_in = ticks; 1921 if (tp->t_fbyte_in == 0) 1922 tp->t_fbyte_in = 1; 1923 if (tp->t_fbyte_out && tp->t_fbyte_in) 1924 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1925 } 1926 /* 1927 * Pull snd_wl1 up to prevent seq wrap relative to 1928 * th_seq. 1929 */ 1930 tp->snd_wl1 = th->th_seq; 1931 /* 1932 * Pull rcv_up up to prevent seq wrap relative to 1933 * rcv_nxt. 1934 */ 1935 tp->rcv_up = tp->rcv_nxt; 1936 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1937 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1938 #ifdef TCPDEBUG 1939 if (so->so_options & SO_DEBUG) 1940 tcp_trace(TA_INPUT, ostate, tp, 1941 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1942 #endif 1943 TCP_PROBE3(debug__input, tp, th, m); 1944 1945 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1946 1947 /* Add data to socket buffer. */ 1948 SOCKBUF_LOCK(&so->so_rcv); 1949 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1950 m_freem(m); 1951 } else { 1952 /* 1953 * Set new socket buffer size. 1954 * Give up when limit is reached. 1955 */ 1956 if (newsize) 1957 if (!sbreserve_locked(&so->so_rcv, 1958 newsize, so, NULL)) 1959 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1960 m_adj(m, drop_hdrlen); /* delayed header drop */ 1961 sbappendstream_locked(&so->so_rcv, m, 0); 1962 } 1963 /* NB: sorwakeup_locked() does an implicit unlock. */ 1964 sorwakeup_locked(so); 1965 if (DELAY_ACK(tp, tlen)) { 1966 tp->t_flags |= TF_DELACK; 1967 } else { 1968 tp->t_flags |= TF_ACKNOW; 1969 tp->t_fb->tfb_tcp_output(tp); 1970 } 1971 goto check_delack; 1972 } 1973 } 1974 1975 /* 1976 * Calculate amount of space in receive window, 1977 * and then do TCP input processing. 1978 * Receive window is amount of space in rcv queue, 1979 * but not less than advertised window. 1980 */ 1981 win = sbspace(&so->so_rcv); 1982 if (win < 0) 1983 win = 0; 1984 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1985 1986 switch (tp->t_state) { 1987 /* 1988 * If the state is SYN_RECEIVED: 1989 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1990 */ 1991 case TCPS_SYN_RECEIVED: 1992 if ((thflags & TH_ACK) && 1993 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1994 SEQ_GT(th->th_ack, tp->snd_max))) { 1995 rstreason = BANDLIM_RST_OPENPORT; 1996 goto dropwithreset; 1997 } 1998 if (IS_FASTOPEN(tp->t_flags)) { 1999 /* 2000 * When a TFO connection is in SYN_RECEIVED, the 2001 * only valid packets are the initial SYN, a 2002 * retransmit/copy of the initial SYN (possibly with 2003 * a subset of the original data), a valid ACK, a 2004 * FIN, or a RST. 2005 */ 2006 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 2007 rstreason = BANDLIM_RST_OPENPORT; 2008 goto dropwithreset; 2009 } else if (thflags & TH_SYN) { 2010 /* non-initial SYN is ignored */ 2011 if ((tcp_timer_active(tp, TT_DELACK) || 2012 tcp_timer_active(tp, TT_REXMT))) 2013 goto drop; 2014 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 2015 goto drop; 2016 } 2017 } 2018 break; 2019 2020 /* 2021 * If the state is SYN_SENT: 2022 * if seg contains a RST with valid ACK (SEQ.ACK has already 2023 * been verified), then drop the connection. 2024 * if seg contains a RST without an ACK, drop the seg. 2025 * if seg does not contain SYN, then drop the seg. 2026 * Otherwise this is an acceptable SYN segment 2027 * initialize tp->rcv_nxt and tp->irs 2028 * if seg contains ack then advance tp->snd_una 2029 * if seg contains an ECE and ECN support is enabled, the stream 2030 * is ECN capable. 2031 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2032 * arrange for segment to be acked (eventually) 2033 * continue processing rest of data/controls, beginning with URG 2034 */ 2035 case TCPS_SYN_SENT: 2036 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2037 TCP_PROBE5(connect__refused, NULL, tp, 2038 m, tp, th); 2039 tp = tcp_drop(tp, ECONNREFUSED); 2040 } 2041 if (thflags & TH_RST) 2042 goto drop; 2043 if (!(thflags & TH_SYN)) 2044 goto drop; 2045 2046 tp->irs = th->th_seq; 2047 tcp_rcvseqinit(tp); 2048 if (thflags & TH_ACK) { 2049 int tfo_partial_ack = 0; 2050 2051 TCPSTAT_INC(tcps_connects); 2052 soisconnected(so); 2053 #ifdef MAC 2054 mac_socketpeer_set_from_mbuf(m, so); 2055 #endif 2056 /* Do window scaling on this connection? */ 2057 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2058 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2059 tp->rcv_scale = tp->request_r_scale; 2060 } 2061 tp->rcv_adv += min(tp->rcv_wnd, 2062 TCP_MAXWIN << tp->rcv_scale); 2063 tp->snd_una++; /* SYN is acked */ 2064 /* 2065 * If not all the data that was sent in the TFO SYN 2066 * has been acked, resend the remainder right away. 2067 */ 2068 if (IS_FASTOPEN(tp->t_flags) && 2069 (tp->snd_una != tp->snd_max)) { 2070 tp->snd_nxt = th->th_ack; 2071 tfo_partial_ack = 1; 2072 } 2073 /* 2074 * If there's data, delay ACK; if there's also a FIN 2075 * ACKNOW will be turned on later. 2076 */ 2077 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2078 tcp_timer_activate(tp, TT_DELACK, 2079 tcp_delacktime); 2080 else 2081 tp->t_flags |= TF_ACKNOW; 2082 2083 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2084 (V_tcp_do_ecn == 1)) { 2085 tp->t_flags2 |= TF2_ECN_PERMIT; 2086 TCPSTAT_INC(tcps_ecn_shs); 2087 } 2088 2089 /* 2090 * Received <SYN,ACK> in SYN_SENT[*] state. 2091 * Transitions: 2092 * SYN_SENT --> ESTABLISHED 2093 * SYN_SENT* --> FIN_WAIT_1 2094 */ 2095 tp->t_starttime = ticks; 2096 if (tp->t_flags & TF_NEEDFIN) { 2097 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2098 tp->t_flags &= ~TF_NEEDFIN; 2099 thflags &= ~TH_SYN; 2100 } else { 2101 tcp_state_change(tp, TCPS_ESTABLISHED); 2102 TCP_PROBE5(connect__established, NULL, tp, 2103 m, tp, th); 2104 cc_conn_init(tp); 2105 tcp_timer_activate(tp, TT_KEEP, 2106 TP_KEEPIDLE(tp)); 2107 } 2108 } else { 2109 /* 2110 * Received initial SYN in SYN-SENT[*] state => 2111 * simultaneous open. 2112 * If it succeeds, connection is * half-synchronized. 2113 * Otherwise, do 3-way handshake: 2114 * SYN-SENT -> SYN-RECEIVED 2115 * SYN-SENT* -> SYN-RECEIVED* 2116 */ 2117 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2118 tcp_timer_activate(tp, TT_REXMT, 0); 2119 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2120 } 2121 2122 INP_WLOCK_ASSERT(tp->t_inpcb); 2123 2124 /* 2125 * Advance th->th_seq to correspond to first data byte. 2126 * If data, trim to stay within window, 2127 * dropping FIN if necessary. 2128 */ 2129 th->th_seq++; 2130 if (tlen > tp->rcv_wnd) { 2131 todrop = tlen - tp->rcv_wnd; 2132 m_adj(m, -todrop); 2133 tlen = tp->rcv_wnd; 2134 thflags &= ~TH_FIN; 2135 TCPSTAT_INC(tcps_rcvpackafterwin); 2136 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2137 } 2138 tp->snd_wl1 = th->th_seq - 1; 2139 tp->rcv_up = th->th_seq; 2140 /* 2141 * Client side of transaction: already sent SYN and data. 2142 * If the remote host used T/TCP to validate the SYN, 2143 * our data will be ACK'd; if so, enter normal data segment 2144 * processing in the middle of step 5, ack processing. 2145 * Otherwise, goto step 6. 2146 */ 2147 if (thflags & TH_ACK) 2148 goto process_ACK; 2149 2150 goto step6; 2151 2152 /* 2153 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2154 * do normal processing. 2155 * 2156 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2157 */ 2158 case TCPS_LAST_ACK: 2159 case TCPS_CLOSING: 2160 break; /* continue normal processing */ 2161 } 2162 2163 /* 2164 * States other than LISTEN or SYN_SENT. 2165 * First check the RST flag and sequence number since reset segments 2166 * are exempt from the timestamp and connection count tests. This 2167 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2168 * below which allowed reset segments in half the sequence space 2169 * to fall though and be processed (which gives forged reset 2170 * segments with a random sequence number a 50 percent chance of 2171 * killing a connection). 2172 * Then check timestamp, if present. 2173 * Then check the connection count, if present. 2174 * Then check that at least some bytes of segment are within 2175 * receive window. If segment begins before rcv_nxt, 2176 * drop leading data (and SYN); if nothing left, just ack. 2177 */ 2178 if (thflags & TH_RST) { 2179 /* 2180 * RFC5961 Section 3.2 2181 * 2182 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2183 * - If RST is in window, we send challenge ACK. 2184 * 2185 * Note: to take into account delayed ACKs, we should 2186 * test against last_ack_sent instead of rcv_nxt. 2187 * Note 2: we handle special case of closed window, not 2188 * covered by the RFC. 2189 */ 2190 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2191 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2192 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2193 KASSERT(tp->t_state != TCPS_SYN_SENT, 2194 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2195 __func__, th, tp)); 2196 2197 if (V_tcp_insecure_rst || 2198 tp->last_ack_sent == th->th_seq) { 2199 TCPSTAT_INC(tcps_drops); 2200 /* Drop the connection. */ 2201 switch (tp->t_state) { 2202 case TCPS_SYN_RECEIVED: 2203 so->so_error = ECONNREFUSED; 2204 goto close; 2205 case TCPS_ESTABLISHED: 2206 case TCPS_FIN_WAIT_1: 2207 case TCPS_FIN_WAIT_2: 2208 case TCPS_CLOSE_WAIT: 2209 case TCPS_CLOSING: 2210 case TCPS_LAST_ACK: 2211 so->so_error = ECONNRESET; 2212 close: 2213 /* FALLTHROUGH */ 2214 default: 2215 tp = tcp_close(tp); 2216 } 2217 } else { 2218 TCPSTAT_INC(tcps_badrst); 2219 /* Send challenge ACK. */ 2220 tcp_respond(tp, mtod(m, void *), th, m, 2221 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2222 tp->last_ack_sent = tp->rcv_nxt; 2223 m = NULL; 2224 } 2225 } 2226 goto drop; 2227 } 2228 2229 /* 2230 * RFC5961 Section 4.2 2231 * Send challenge ACK for any SYN in synchronized state. 2232 */ 2233 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2234 tp->t_state != TCPS_SYN_RECEIVED) { 2235 TCPSTAT_INC(tcps_badsyn); 2236 if (V_tcp_insecure_syn && 2237 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2238 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2239 tp = tcp_drop(tp, ECONNRESET); 2240 rstreason = BANDLIM_UNLIMITED; 2241 } else { 2242 /* Send challenge ACK. */ 2243 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2244 tp->snd_nxt, TH_ACK); 2245 tp->last_ack_sent = tp->rcv_nxt; 2246 m = NULL; 2247 } 2248 goto drop; 2249 } 2250 2251 /* 2252 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2253 * and it's less than ts_recent, drop it. 2254 */ 2255 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2256 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2257 /* Check to see if ts_recent is over 24 days old. */ 2258 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2259 /* 2260 * Invalidate ts_recent. If this segment updates 2261 * ts_recent, the age will be reset later and ts_recent 2262 * will get a valid value. If it does not, setting 2263 * ts_recent to zero will at least satisfy the 2264 * requirement that zero be placed in the timestamp 2265 * echo reply when ts_recent isn't valid. The 2266 * age isn't reset until we get a valid ts_recent 2267 * because we don't want out-of-order segments to be 2268 * dropped when ts_recent is old. 2269 */ 2270 tp->ts_recent = 0; 2271 } else { 2272 TCPSTAT_INC(tcps_rcvduppack); 2273 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2274 TCPSTAT_INC(tcps_pawsdrop); 2275 if (tlen) 2276 goto dropafterack; 2277 goto drop; 2278 } 2279 } 2280 2281 /* 2282 * In the SYN-RECEIVED state, validate that the packet belongs to 2283 * this connection before trimming the data to fit the receive 2284 * window. Check the sequence number versus IRS since we know 2285 * the sequence numbers haven't wrapped. This is a partial fix 2286 * for the "LAND" DoS attack. 2287 */ 2288 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2289 rstreason = BANDLIM_RST_OPENPORT; 2290 goto dropwithreset; 2291 } 2292 2293 todrop = tp->rcv_nxt - th->th_seq; 2294 if (todrop > 0) { 2295 if (thflags & TH_SYN) { 2296 thflags &= ~TH_SYN; 2297 th->th_seq++; 2298 if (th->th_urp > 1) 2299 th->th_urp--; 2300 else 2301 thflags &= ~TH_URG; 2302 todrop--; 2303 } 2304 /* 2305 * Following if statement from Stevens, vol. 2, p. 960. 2306 */ 2307 if (todrop > tlen 2308 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2309 /* 2310 * Any valid FIN must be to the left of the window. 2311 * At this point the FIN must be a duplicate or out 2312 * of sequence; drop it. 2313 */ 2314 thflags &= ~TH_FIN; 2315 2316 /* 2317 * Send an ACK to resynchronize and drop any data. 2318 * But keep on processing for RST or ACK. 2319 */ 2320 tp->t_flags |= TF_ACKNOW; 2321 todrop = tlen; 2322 TCPSTAT_INC(tcps_rcvduppack); 2323 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2324 } else { 2325 TCPSTAT_INC(tcps_rcvpartduppack); 2326 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2327 } 2328 /* 2329 * DSACK - add SACK block for dropped range 2330 */ 2331 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2332 tcp_update_sack_list(tp, th->th_seq, 2333 th->th_seq + todrop); 2334 /* 2335 * ACK now, as the next in-sequence segment 2336 * will clear the DSACK block again 2337 */ 2338 tp->t_flags |= TF_ACKNOW; 2339 } 2340 drop_hdrlen += todrop; /* drop from the top afterwards */ 2341 th->th_seq += todrop; 2342 tlen -= todrop; 2343 if (th->th_urp > todrop) 2344 th->th_urp -= todrop; 2345 else { 2346 thflags &= ~TH_URG; 2347 th->th_urp = 0; 2348 } 2349 } 2350 2351 /* 2352 * If new data are received on a connection after the 2353 * user processes are gone, then RST the other end. 2354 */ 2355 if ((so->so_state & SS_NOFDREF) && 2356 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2357 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2358 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2359 "after socket was closed, " 2360 "sending RST and removing tcpcb\n", 2361 s, __func__, tcpstates[tp->t_state], tlen); 2362 free(s, M_TCPLOG); 2363 } 2364 tp = tcp_close(tp); 2365 TCPSTAT_INC(tcps_rcvafterclose); 2366 rstreason = BANDLIM_UNLIMITED; 2367 goto dropwithreset; 2368 } 2369 2370 /* 2371 * If segment ends after window, drop trailing data 2372 * (and PUSH and FIN); if nothing left, just ACK. 2373 */ 2374 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2375 if (todrop > 0) { 2376 TCPSTAT_INC(tcps_rcvpackafterwin); 2377 if (todrop >= tlen) { 2378 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2379 /* 2380 * If window is closed can only take segments at 2381 * window edge, and have to drop data and PUSH from 2382 * incoming segments. Continue processing, but 2383 * remember to ack. Otherwise, drop segment 2384 * and ack. 2385 */ 2386 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2387 tp->t_flags |= TF_ACKNOW; 2388 TCPSTAT_INC(tcps_rcvwinprobe); 2389 } else 2390 goto dropafterack; 2391 } else 2392 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2393 m_adj(m, -todrop); 2394 tlen -= todrop; 2395 thflags &= ~(TH_PUSH|TH_FIN); 2396 } 2397 2398 /* 2399 * If last ACK falls within this segment's sequence numbers, 2400 * record its timestamp. 2401 * NOTE: 2402 * 1) That the test incorporates suggestions from the latest 2403 * proposal of the [email protected] list (Braden 1993/04/26). 2404 * 2) That updating only on newer timestamps interferes with 2405 * our earlier PAWS tests, so this check should be solely 2406 * predicated on the sequence space of this segment. 2407 * 3) That we modify the segment boundary check to be 2408 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2409 * instead of RFC1323's 2410 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2411 * This modified check allows us to overcome RFC1323's 2412 * limitations as described in Stevens TCP/IP Illustrated 2413 * Vol. 2 p.869. In such cases, we can still calculate the 2414 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2415 */ 2416 if ((to.to_flags & TOF_TS) != 0 && 2417 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2418 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2419 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2420 tp->ts_recent_age = tcp_ts_getticks(); 2421 tp->ts_recent = to.to_tsval; 2422 } 2423 2424 /* 2425 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2426 * flag is on (half-synchronized state), then queue data for 2427 * later processing; else drop segment and return. 2428 */ 2429 if ((thflags & TH_ACK) == 0) { 2430 if (tp->t_state == TCPS_SYN_RECEIVED || 2431 (tp->t_flags & TF_NEEDSYN)) { 2432 if (tp->t_state == TCPS_SYN_RECEIVED && 2433 IS_FASTOPEN(tp->t_flags)) { 2434 tp->snd_wnd = tiwin; 2435 cc_conn_init(tp); 2436 } 2437 goto step6; 2438 } else if (tp->t_flags & TF_ACKNOW) 2439 goto dropafterack; 2440 else 2441 goto drop; 2442 } 2443 2444 /* 2445 * Ack processing. 2446 */ 2447 switch (tp->t_state) { 2448 /* 2449 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2450 * ESTABLISHED state and continue processing. 2451 * The ACK was checked above. 2452 */ 2453 case TCPS_SYN_RECEIVED: 2454 2455 TCPSTAT_INC(tcps_connects); 2456 soisconnected(so); 2457 /* Do window scaling? */ 2458 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2459 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2460 tp->rcv_scale = tp->request_r_scale; 2461 } 2462 tp->snd_wnd = tiwin; 2463 /* 2464 * Make transitions: 2465 * SYN-RECEIVED -> ESTABLISHED 2466 * SYN-RECEIVED* -> FIN-WAIT-1 2467 */ 2468 tp->t_starttime = ticks; 2469 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2470 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2471 tp->t_tfo_pending = NULL; 2472 } 2473 if (tp->t_flags & TF_NEEDFIN) { 2474 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2475 tp->t_flags &= ~TF_NEEDFIN; 2476 } else { 2477 tcp_state_change(tp, TCPS_ESTABLISHED); 2478 TCP_PROBE5(accept__established, NULL, tp, 2479 m, tp, th); 2480 /* 2481 * TFO connections call cc_conn_init() during SYN 2482 * processing. Calling it again here for such 2483 * connections is not harmless as it would undo the 2484 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2485 * is retransmitted. 2486 */ 2487 if (!IS_FASTOPEN(tp->t_flags)) 2488 cc_conn_init(tp); 2489 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2490 } 2491 /* 2492 * Account for the ACK of our SYN prior to 2493 * regular ACK processing below, except for 2494 * simultaneous SYN, which is handled later. 2495 */ 2496 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2497 incforsyn = 1; 2498 /* 2499 * If segment contains data or ACK, will call tcp_reass() 2500 * later; if not, do so now to pass queued data to user. 2501 */ 2502 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2503 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2504 (struct mbuf *)0); 2505 tcp_handle_wakeup(tp, so); 2506 } 2507 tp->snd_wl1 = th->th_seq - 1; 2508 /* FALLTHROUGH */ 2509 2510 /* 2511 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2512 * ACKs. If the ack is in the range 2513 * tp->snd_una < th->th_ack <= tp->snd_max 2514 * then advance tp->snd_una to th->th_ack and drop 2515 * data from the retransmission queue. If this ACK reflects 2516 * more up to date window information we update our window information. 2517 */ 2518 case TCPS_ESTABLISHED: 2519 case TCPS_FIN_WAIT_1: 2520 case TCPS_FIN_WAIT_2: 2521 case TCPS_CLOSE_WAIT: 2522 case TCPS_CLOSING: 2523 case TCPS_LAST_ACK: 2524 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2525 TCPSTAT_INC(tcps_rcvacktoomuch); 2526 goto dropafterack; 2527 } 2528 if ((tp->t_flags & TF_SACK_PERMIT) && 2529 ((to.to_flags & TOF_SACK) || 2530 !TAILQ_EMPTY(&tp->snd_holes))) 2531 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2532 else 2533 /* 2534 * Reset the value so that previous (valid) value 2535 * from the last ack with SACK doesn't get used. 2536 */ 2537 tp->sackhint.sacked_bytes = 0; 2538 2539 #ifdef TCP_HHOOK 2540 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2541 hhook_run_tcp_est_in(tp, th, &to); 2542 #endif 2543 2544 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2545 maxseg = tcp_maxseg(tp); 2546 if (tlen == 0 && 2547 (tiwin == tp->snd_wnd || 2548 (tp->t_flags & TF_SACK_PERMIT))) { 2549 /* 2550 * If this is the first time we've seen a 2551 * FIN from the remote, this is not a 2552 * duplicate and it needs to be processed 2553 * normally. This happens during a 2554 * simultaneous close. 2555 */ 2556 if ((thflags & TH_FIN) && 2557 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2558 tp->t_dupacks = 0; 2559 break; 2560 } 2561 TCPSTAT_INC(tcps_rcvdupack); 2562 /* 2563 * If we have outstanding data (other than 2564 * a window probe), this is a completely 2565 * duplicate ack (ie, window info didn't 2566 * change and FIN isn't set), 2567 * the ack is the biggest we've 2568 * seen and we've seen exactly our rexmt 2569 * threshold of them, assume a packet 2570 * has been dropped and retransmit it. 2571 * Kludge snd_nxt & the congestion 2572 * window so we send only this one 2573 * packet. 2574 * 2575 * We know we're losing at the current 2576 * window size so do congestion avoidance 2577 * (set ssthresh to half the current window 2578 * and pull our congestion window back to 2579 * the new ssthresh). 2580 * 2581 * Dup acks mean that packets have left the 2582 * network (they're now cached at the receiver) 2583 * so bump cwnd by the amount in the receiver 2584 * to keep a constant cwnd packets in the 2585 * network. 2586 * 2587 * When using TCP ECN, notify the peer that 2588 * we reduced the cwnd. 2589 */ 2590 /* 2591 * Following 2 kinds of acks should not affect 2592 * dupack counting: 2593 * 1) Old acks 2594 * 2) Acks with SACK but without any new SACK 2595 * information in them. These could result from 2596 * any anomaly in the network like a switch 2597 * duplicating packets or a possible DoS attack. 2598 */ 2599 if (th->th_ack != tp->snd_una || 2600 ((tp->t_flags & TF_SACK_PERMIT) && 2601 (to.to_flags & TOF_SACK) && 2602 !sack_changed)) 2603 break; 2604 else if (!tcp_timer_active(tp, TT_REXMT)) 2605 tp->t_dupacks = 0; 2606 else if (++tp->t_dupacks > tcprexmtthresh || 2607 IN_FASTRECOVERY(tp->t_flags)) { 2608 cc_ack_received(tp, th, nsegs, 2609 CC_DUPACK); 2610 if (V_tcp_do_prr && 2611 IN_FASTRECOVERY(tp->t_flags)) { 2612 tcp_do_prr_ack(tp, th, &to); 2613 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2614 (to.to_flags & TOF_SACK) && 2615 IN_FASTRECOVERY(tp->t_flags)) { 2616 int awnd; 2617 2618 /* 2619 * Compute the amount of data in flight first. 2620 * We can inject new data into the pipe iff 2621 * we have less than 1/2 the original window's 2622 * worth of data in flight. 2623 */ 2624 if (V_tcp_do_rfc6675_pipe) 2625 awnd = tcp_compute_pipe(tp); 2626 else 2627 awnd = (tp->snd_nxt - tp->snd_fack) + 2628 tp->sackhint.sack_bytes_rexmit; 2629 2630 if (awnd < tp->snd_ssthresh) { 2631 tp->snd_cwnd += maxseg; 2632 if (tp->snd_cwnd > tp->snd_ssthresh) 2633 tp->snd_cwnd = tp->snd_ssthresh; 2634 } 2635 } else 2636 tp->snd_cwnd += maxseg; 2637 (void) tp->t_fb->tfb_tcp_output(tp); 2638 goto drop; 2639 } else if (tp->t_dupacks == tcprexmtthresh || 2640 (tp->t_flags & TF_SACK_PERMIT && 2641 V_tcp_do_rfc6675_pipe && 2642 tp->sackhint.sacked_bytes > 2643 (tcprexmtthresh - 1) * maxseg)) { 2644 enter_recovery: 2645 /* 2646 * Above is the RFC6675 trigger condition of 2647 * more than (dupthresh-1)*maxseg sacked data. 2648 * If the count of holes in the 2649 * scoreboard is >= dupthresh, we could 2650 * also enter loss recovery, but don't 2651 * have that value readily available. 2652 */ 2653 tp->t_dupacks = tcprexmtthresh; 2654 tcp_seq onxt = tp->snd_nxt; 2655 2656 /* 2657 * If we're doing sack, or prr, check 2658 * to see if we're already in sack 2659 * recovery. If we're not doing sack, 2660 * check to see if we're in newreno 2661 * recovery. 2662 */ 2663 if (V_tcp_do_prr || 2664 (tp->t_flags & TF_SACK_PERMIT)) { 2665 if (IN_FASTRECOVERY(tp->t_flags)) { 2666 tp->t_dupacks = 0; 2667 break; 2668 } 2669 } else { 2670 if (SEQ_LEQ(th->th_ack, 2671 tp->snd_recover)) { 2672 tp->t_dupacks = 0; 2673 break; 2674 } 2675 } 2676 /* Congestion signal before ack. */ 2677 cc_cong_signal(tp, th, CC_NDUPACK); 2678 cc_ack_received(tp, th, nsegs, 2679 CC_DUPACK); 2680 tcp_timer_activate(tp, TT_REXMT, 0); 2681 tp->t_rtttime = 0; 2682 if (V_tcp_do_prr) { 2683 /* 2684 * snd_ssthresh is already updated by 2685 * cc_cong_signal. 2686 */ 2687 if ((tp->t_flags & TF_SACK_PERMIT) && 2688 (to.to_flags & TOF_SACK)) { 2689 tp->sackhint.prr_delivered = 2690 tp->sackhint.sacked_bytes; 2691 } else { 2692 tp->sackhint.prr_delivered = 2693 imin(tp->snd_max - tp->snd_una, 2694 imin(INT_MAX / 65536, 2695 tp->t_dupacks) * maxseg); 2696 } 2697 tp->sackhint.recover_fs = max(1, 2698 tp->snd_nxt - tp->snd_una); 2699 } 2700 if ((tp->t_flags & TF_SACK_PERMIT) && 2701 (to.to_flags & TOF_SACK)) { 2702 TCPSTAT_INC( 2703 tcps_sack_recovery_episode); 2704 tp->snd_recover = tp->snd_nxt; 2705 tp->snd_cwnd = maxseg; 2706 (void) tp->t_fb->tfb_tcp_output(tp); 2707 if (SEQ_GT(th->th_ack, tp->snd_una)) 2708 goto resume_partialack; 2709 goto drop; 2710 } 2711 tp->snd_nxt = th->th_ack; 2712 tp->snd_cwnd = maxseg; 2713 (void) tp->t_fb->tfb_tcp_output(tp); 2714 KASSERT(tp->snd_limited <= 2, 2715 ("%s: tp->snd_limited too big", 2716 __func__)); 2717 tp->snd_cwnd = tp->snd_ssthresh + 2718 maxseg * 2719 (tp->t_dupacks - tp->snd_limited); 2720 if (SEQ_GT(onxt, tp->snd_nxt)) 2721 tp->snd_nxt = onxt; 2722 goto drop; 2723 } else if (V_tcp_do_rfc3042) { 2724 /* 2725 * Process first and second duplicate 2726 * ACKs. Each indicates a segment 2727 * leaving the network, creating room 2728 * for more. Make sure we can send a 2729 * packet on reception of each duplicate 2730 * ACK by increasing snd_cwnd by one 2731 * segment. Restore the original 2732 * snd_cwnd after packet transmission. 2733 */ 2734 cc_ack_received(tp, th, nsegs, 2735 CC_DUPACK); 2736 uint32_t oldcwnd = tp->snd_cwnd; 2737 tcp_seq oldsndmax = tp->snd_max; 2738 u_int sent; 2739 int avail; 2740 2741 KASSERT(tp->t_dupacks == 1 || 2742 tp->t_dupacks == 2, 2743 ("%s: dupacks not 1 or 2", 2744 __func__)); 2745 if (tp->t_dupacks == 1) 2746 tp->snd_limited = 0; 2747 tp->snd_cwnd = 2748 (tp->snd_nxt - tp->snd_una) + 2749 (tp->t_dupacks - tp->snd_limited) * 2750 maxseg; 2751 /* 2752 * Only call tcp_output when there 2753 * is new data available to be sent. 2754 * Otherwise we would send pure ACKs. 2755 */ 2756 SOCKBUF_LOCK(&so->so_snd); 2757 avail = sbavail(&so->so_snd) - 2758 (tp->snd_nxt - tp->snd_una); 2759 SOCKBUF_UNLOCK(&so->so_snd); 2760 if (avail > 0) 2761 (void) tp->t_fb->tfb_tcp_output(tp); 2762 sent = tp->snd_max - oldsndmax; 2763 if (sent > maxseg) { 2764 KASSERT((tp->t_dupacks == 2 && 2765 tp->snd_limited == 0) || 2766 (sent == maxseg + 1 && 2767 tp->t_flags & TF_SENTFIN), 2768 ("%s: sent too much", 2769 __func__)); 2770 tp->snd_limited = 2; 2771 } else if (sent > 0) 2772 ++tp->snd_limited; 2773 tp->snd_cwnd = oldcwnd; 2774 goto drop; 2775 } 2776 } 2777 break; 2778 } else { 2779 /* 2780 * This ack is advancing the left edge, reset the 2781 * counter. 2782 */ 2783 tp->t_dupacks = 0; 2784 /* 2785 * If this ack also has new SACK info, increment the 2786 * counter as per rfc6675. The variable 2787 * sack_changed tracks all changes to the SACK 2788 * scoreboard, including when partial ACKs without 2789 * SACK options are received, and clear the scoreboard 2790 * from the left side. Such partial ACKs should not be 2791 * counted as dupacks here. 2792 */ 2793 if ((tp->t_flags & TF_SACK_PERMIT) && 2794 (to.to_flags & TOF_SACK) && 2795 sack_changed) { 2796 tp->t_dupacks++; 2797 /* limit overhead by setting maxseg last */ 2798 if (!IN_FASTRECOVERY(tp->t_flags) && 2799 (tp->sackhint.sacked_bytes > 2800 ((tcprexmtthresh - 1) * 2801 (maxseg = tcp_maxseg(tp))))) { 2802 goto enter_recovery; 2803 } 2804 } 2805 } 2806 2807 resume_partialack: 2808 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2809 ("%s: th_ack <= snd_una", __func__)); 2810 2811 /* 2812 * If the congestion window was inflated to account 2813 * for the other side's cached packets, retract it. 2814 */ 2815 if (IN_FASTRECOVERY(tp->t_flags)) { 2816 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2817 if (tp->t_flags & TF_SACK_PERMIT) 2818 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2819 tcp_timer_activate(tp, TT_REXMT, 0); 2820 tp->t_rtttime = 0; 2821 tcp_do_prr_ack(tp, th, &to); 2822 tp->t_flags |= TF_ACKNOW; 2823 (void) tcp_output(tp); 2824 } else 2825 tcp_sack_partialack(tp, th); 2826 else 2827 tcp_newreno_partial_ack(tp, th); 2828 } else 2829 cc_post_recovery(tp, th); 2830 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2831 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2832 if (V_tcp_do_prr) { 2833 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2834 tp->snd_fack = th->th_ack; 2835 tcp_do_prr_ack(tp, th, &to); 2836 (void) tcp_output(tp); 2837 } 2838 } else 2839 cc_post_recovery(tp, th); 2840 } 2841 /* 2842 * If we reach this point, ACK is not a duplicate, 2843 * i.e., it ACKs something we sent. 2844 */ 2845 if (tp->t_flags & TF_NEEDSYN) { 2846 /* 2847 * T/TCP: Connection was half-synchronized, and our 2848 * SYN has been ACK'd (so connection is now fully 2849 * synchronized). Go to non-starred state, 2850 * increment snd_una for ACK of SYN, and check if 2851 * we can do window scaling. 2852 */ 2853 tp->t_flags &= ~TF_NEEDSYN; 2854 tp->snd_una++; 2855 /* Do window scaling? */ 2856 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2857 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2858 tp->rcv_scale = tp->request_r_scale; 2859 /* Send window already scaled. */ 2860 } 2861 } 2862 2863 process_ACK: 2864 INP_WLOCK_ASSERT(tp->t_inpcb); 2865 2866 /* 2867 * Adjust for the SYN bit in sequence space, 2868 * but don't account for it in cwnd calculations. 2869 * This is for the SYN_RECEIVED, non-simultaneous 2870 * SYN case. SYN_SENT and simultaneous SYN are 2871 * treated elsewhere. 2872 */ 2873 if (incforsyn) 2874 tp->snd_una++; 2875 acked = BYTES_THIS_ACK(tp, th); 2876 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2877 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2878 tp->snd_una, th->th_ack, tp, m)); 2879 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2880 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2881 2882 /* 2883 * If we just performed our first retransmit, and the ACK 2884 * arrives within our recovery window, then it was a mistake 2885 * to do the retransmit in the first place. Recover our 2886 * original cwnd and ssthresh, and proceed to transmit where 2887 * we left off. 2888 */ 2889 if (tp->t_rxtshift == 1 && 2890 tp->t_flags & TF_PREVVALID && 2891 tp->t_badrxtwin != 0 && 2892 to.to_flags & TOF_TS && 2893 to.to_tsecr != 0 && 2894 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2895 cc_cong_signal(tp, th, CC_RTO_ERR); 2896 2897 /* 2898 * If we have a timestamp reply, update smoothed 2899 * round trip time. If no timestamp is present but 2900 * transmit timer is running and timed sequence 2901 * number was acked, update smoothed round trip time. 2902 * Since we now have an rtt measurement, cancel the 2903 * timer backoff (cf., Phil Karn's retransmit alg.). 2904 * Recompute the initial retransmit timer. 2905 * 2906 * Some boxes send broken timestamp replies 2907 * during the SYN+ACK phase, ignore 2908 * timestamps of 0 or we could calculate a 2909 * huge RTT and blow up the retransmit timer. 2910 */ 2911 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2912 uint32_t t; 2913 2914 t = tcp_ts_getticks() - to.to_tsecr; 2915 if (!tp->t_rttlow || tp->t_rttlow > t) 2916 tp->t_rttlow = t; 2917 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2918 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2919 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2920 tp->t_rttlow = ticks - tp->t_rtttime; 2921 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2922 } 2923 2924 /* 2925 * If all outstanding data is acked, stop retransmit 2926 * timer and remember to restart (more output or persist). 2927 * If there is more data to be acked, restart retransmit 2928 * timer, using current (possibly backed-off) value. 2929 */ 2930 if (th->th_ack == tp->snd_max) { 2931 tcp_timer_activate(tp, TT_REXMT, 0); 2932 needoutput = 1; 2933 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2934 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2935 2936 /* 2937 * If no data (only SYN) was ACK'd, 2938 * skip rest of ACK processing. 2939 */ 2940 if (acked == 0) 2941 goto step6; 2942 2943 /* 2944 * Let the congestion control algorithm update congestion 2945 * control related information. This typically means increasing 2946 * the congestion window. 2947 */ 2948 cc_ack_received(tp, th, nsegs, CC_ACK); 2949 2950 SOCKBUF_LOCK(&so->so_snd); 2951 if (acked > sbavail(&so->so_snd)) { 2952 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2953 tp->snd_wnd -= sbavail(&so->so_snd); 2954 else 2955 tp->snd_wnd = 0; 2956 mfree = sbcut_locked(&so->so_snd, 2957 (int)sbavail(&so->so_snd)); 2958 ourfinisacked = 1; 2959 } else { 2960 mfree = sbcut_locked(&so->so_snd, acked); 2961 if (tp->snd_wnd >= (uint32_t) acked) 2962 tp->snd_wnd -= acked; 2963 else 2964 tp->snd_wnd = 0; 2965 ourfinisacked = 0; 2966 } 2967 /* NB: sowwakeup_locked() does an implicit unlock. */ 2968 sowwakeup_locked(so); 2969 m_freem(mfree); 2970 /* Detect una wraparound. */ 2971 if (!IN_RECOVERY(tp->t_flags) && 2972 SEQ_GT(tp->snd_una, tp->snd_recover) && 2973 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2974 tp->snd_recover = th->th_ack - 1; 2975 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2976 if (IN_RECOVERY(tp->t_flags) && 2977 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2978 EXIT_RECOVERY(tp->t_flags); 2979 } 2980 tp->snd_una = th->th_ack; 2981 if (tp->t_flags & TF_SACK_PERMIT) { 2982 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2983 tp->snd_recover = tp->snd_una; 2984 } 2985 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2986 tp->snd_nxt = tp->snd_una; 2987 2988 switch (tp->t_state) { 2989 /* 2990 * In FIN_WAIT_1 STATE in addition to the processing 2991 * for the ESTABLISHED state if our FIN is now acknowledged 2992 * then enter FIN_WAIT_2. 2993 */ 2994 case TCPS_FIN_WAIT_1: 2995 if (ourfinisacked) { 2996 /* 2997 * If we can't receive any more 2998 * data, then closing user can proceed. 2999 * Starting the timer is contrary to the 3000 * specification, but if we don't get a FIN 3001 * we'll hang forever. 3002 * 3003 * XXXjl: 3004 * we should release the tp also, and use a 3005 * compressed state. 3006 */ 3007 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3008 soisdisconnected(so); 3009 tcp_timer_activate(tp, TT_2MSL, 3010 (tcp_fast_finwait2_recycle ? 3011 tcp_finwait2_timeout : 3012 TP_MAXIDLE(tp))); 3013 } 3014 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3015 } 3016 break; 3017 3018 /* 3019 * In CLOSING STATE in addition to the processing for 3020 * the ESTABLISHED state if the ACK acknowledges our FIN 3021 * then enter the TIME-WAIT state, otherwise ignore 3022 * the segment. 3023 */ 3024 case TCPS_CLOSING: 3025 if (ourfinisacked) { 3026 tcp_twstart(tp); 3027 m_freem(m); 3028 return; 3029 } 3030 break; 3031 3032 /* 3033 * In LAST_ACK, we may still be waiting for data to drain 3034 * and/or to be acked, as well as for the ack of our FIN. 3035 * If our FIN is now acknowledged, delete the TCB, 3036 * enter the closed state and return. 3037 */ 3038 case TCPS_LAST_ACK: 3039 if (ourfinisacked) { 3040 tp = tcp_close(tp); 3041 goto drop; 3042 } 3043 break; 3044 } 3045 } 3046 3047 step6: 3048 INP_WLOCK_ASSERT(tp->t_inpcb); 3049 3050 /* 3051 * Update window information. 3052 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3053 */ 3054 if ((thflags & TH_ACK) && 3055 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3056 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3057 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3058 /* keep track of pure window updates */ 3059 if (tlen == 0 && 3060 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3061 TCPSTAT_INC(tcps_rcvwinupd); 3062 tp->snd_wnd = tiwin; 3063 tp->snd_wl1 = th->th_seq; 3064 tp->snd_wl2 = th->th_ack; 3065 if (tp->snd_wnd > tp->max_sndwnd) 3066 tp->max_sndwnd = tp->snd_wnd; 3067 needoutput = 1; 3068 } 3069 3070 /* 3071 * Process segments with URG. 3072 */ 3073 if ((thflags & TH_URG) && th->th_urp && 3074 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3075 /* 3076 * This is a kludge, but if we receive and accept 3077 * random urgent pointers, we'll crash in 3078 * soreceive. It's hard to imagine someone 3079 * actually wanting to send this much urgent data. 3080 */ 3081 SOCKBUF_LOCK(&so->so_rcv); 3082 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3083 th->th_urp = 0; /* XXX */ 3084 thflags &= ~TH_URG; /* XXX */ 3085 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3086 goto dodata; /* XXX */ 3087 } 3088 /* 3089 * If this segment advances the known urgent pointer, 3090 * then mark the data stream. This should not happen 3091 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3092 * a FIN has been received from the remote side. 3093 * In these states we ignore the URG. 3094 * 3095 * According to RFC961 (Assigned Protocols), 3096 * the urgent pointer points to the last octet 3097 * of urgent data. We continue, however, 3098 * to consider it to indicate the first octet 3099 * of data past the urgent section as the original 3100 * spec states (in one of two places). 3101 */ 3102 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3103 tp->rcv_up = th->th_seq + th->th_urp; 3104 so->so_oobmark = sbavail(&so->so_rcv) + 3105 (tp->rcv_up - tp->rcv_nxt) - 1; 3106 if (so->so_oobmark == 0) 3107 so->so_rcv.sb_state |= SBS_RCVATMARK; 3108 sohasoutofband(so); 3109 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3110 } 3111 SOCKBUF_UNLOCK(&so->so_rcv); 3112 /* 3113 * Remove out of band data so doesn't get presented to user. 3114 * This can happen independent of advancing the URG pointer, 3115 * but if two URG's are pending at once, some out-of-band 3116 * data may creep in... ick. 3117 */ 3118 if (th->th_urp <= (uint32_t)tlen && 3119 !(so->so_options & SO_OOBINLINE)) { 3120 /* hdr drop is delayed */ 3121 tcp_pulloutofband(so, th, m, drop_hdrlen); 3122 } 3123 } else { 3124 /* 3125 * If no out of band data is expected, 3126 * pull receive urgent pointer along 3127 * with the receive window. 3128 */ 3129 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3130 tp->rcv_up = tp->rcv_nxt; 3131 } 3132 dodata: /* XXX */ 3133 INP_WLOCK_ASSERT(tp->t_inpcb); 3134 3135 /* 3136 * Process the segment text, merging it into the TCP sequencing queue, 3137 * and arranging for acknowledgment of receipt if necessary. 3138 * This process logically involves adjusting tp->rcv_wnd as data 3139 * is presented to the user (this happens in tcp_usrreq.c, 3140 * case PRU_RCVD). If a FIN has already been received on this 3141 * connection then we just ignore the text. 3142 */ 3143 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3144 IS_FASTOPEN(tp->t_flags)); 3145 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3146 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3147 tcp_seq save_start = th->th_seq; 3148 tcp_seq save_rnxt = tp->rcv_nxt; 3149 int save_tlen = tlen; 3150 m_adj(m, drop_hdrlen); /* delayed header drop */ 3151 /* 3152 * Insert segment which includes th into TCP reassembly queue 3153 * with control block tp. Set thflags to whether reassembly now 3154 * includes a segment with FIN. This handles the common case 3155 * inline (segment is the next to be received on an established 3156 * connection, and the queue is empty), avoiding linkage into 3157 * and removal from the queue and repetition of various 3158 * conversions. 3159 * Set DELACK for segments received in order, but ack 3160 * immediately when segments are out of order (so 3161 * fast retransmit can work). 3162 */ 3163 if (th->th_seq == tp->rcv_nxt && 3164 SEGQ_EMPTY(tp) && 3165 (TCPS_HAVEESTABLISHED(tp->t_state) || 3166 tfo_syn)) { 3167 if (DELAY_ACK(tp, tlen) || tfo_syn) 3168 tp->t_flags |= TF_DELACK; 3169 else 3170 tp->t_flags |= TF_ACKNOW; 3171 tp->rcv_nxt += tlen; 3172 if (tlen && 3173 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3174 (tp->t_fbyte_in == 0)) { 3175 tp->t_fbyte_in = ticks; 3176 if (tp->t_fbyte_in == 0) 3177 tp->t_fbyte_in = 1; 3178 if (tp->t_fbyte_out && tp->t_fbyte_in) 3179 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3180 } 3181 thflags = th->th_flags & TH_FIN; 3182 TCPSTAT_INC(tcps_rcvpack); 3183 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3184 SOCKBUF_LOCK(&so->so_rcv); 3185 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3186 m_freem(m); 3187 else 3188 sbappendstream_locked(&so->so_rcv, m, 0); 3189 tp->t_flags |= TF_WAKESOR; 3190 } else { 3191 /* 3192 * XXX: Due to the header drop above "th" is 3193 * theoretically invalid by now. Fortunately 3194 * m_adj() doesn't actually frees any mbufs 3195 * when trimming from the head. 3196 */ 3197 tcp_seq temp = save_start; 3198 3199 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3200 tp->t_flags |= TF_ACKNOW; 3201 } 3202 if ((tp->t_flags & TF_SACK_PERMIT) && 3203 (save_tlen > 0) && 3204 TCPS_HAVEESTABLISHED(tp->t_state)) { 3205 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3206 /* 3207 * DSACK actually handled in the fastpath 3208 * above. 3209 */ 3210 tcp_update_sack_list(tp, save_start, 3211 save_start + save_tlen); 3212 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3213 if ((tp->rcv_numsacks >= 1) && 3214 (tp->sackblks[0].end == save_start)) { 3215 /* 3216 * Partial overlap, recorded at todrop 3217 * above. 3218 */ 3219 tcp_update_sack_list(tp, 3220 tp->sackblks[0].start, 3221 tp->sackblks[0].end); 3222 } else { 3223 tcp_update_dsack_list(tp, save_start, 3224 save_start + save_tlen); 3225 } 3226 } else if (tlen >= save_tlen) { 3227 /* Update of sackblks. */ 3228 tcp_update_dsack_list(tp, save_start, 3229 save_start + save_tlen); 3230 } else if (tlen > 0) { 3231 tcp_update_dsack_list(tp, save_start, 3232 save_start + tlen); 3233 } 3234 } 3235 tcp_handle_wakeup(tp, so); 3236 #if 0 3237 /* 3238 * Note the amount of data that peer has sent into 3239 * our window, in order to estimate the sender's 3240 * buffer size. 3241 * XXX: Unused. 3242 */ 3243 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3244 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3245 else 3246 len = so->so_rcv.sb_hiwat; 3247 #endif 3248 } else { 3249 m_freem(m); 3250 thflags &= ~TH_FIN; 3251 } 3252 3253 /* 3254 * If FIN is received ACK the FIN and let the user know 3255 * that the connection is closing. 3256 */ 3257 if (thflags & TH_FIN) { 3258 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3259 /* The socket upcall is handled by socantrcvmore. */ 3260 socantrcvmore(so); 3261 /* 3262 * If connection is half-synchronized 3263 * (ie NEEDSYN flag on) then delay ACK, 3264 * so it may be piggybacked when SYN is sent. 3265 * Otherwise, since we received a FIN then no 3266 * more input can be expected, send ACK now. 3267 */ 3268 if (tp->t_flags & TF_NEEDSYN) 3269 tp->t_flags |= TF_DELACK; 3270 else 3271 tp->t_flags |= TF_ACKNOW; 3272 tp->rcv_nxt++; 3273 } 3274 switch (tp->t_state) { 3275 /* 3276 * In SYN_RECEIVED and ESTABLISHED STATES 3277 * enter the CLOSE_WAIT state. 3278 */ 3279 case TCPS_SYN_RECEIVED: 3280 tp->t_starttime = ticks; 3281 /* FALLTHROUGH */ 3282 case TCPS_ESTABLISHED: 3283 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3284 break; 3285 3286 /* 3287 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3288 * enter the CLOSING state. 3289 */ 3290 case TCPS_FIN_WAIT_1: 3291 tcp_state_change(tp, TCPS_CLOSING); 3292 break; 3293 3294 /* 3295 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3296 * starting the time-wait timer, turning off the other 3297 * standard timers. 3298 */ 3299 case TCPS_FIN_WAIT_2: 3300 tcp_twstart(tp); 3301 return; 3302 } 3303 } 3304 #ifdef TCPDEBUG 3305 if (so->so_options & SO_DEBUG) 3306 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3307 &tcp_savetcp, 0); 3308 #endif 3309 TCP_PROBE3(debug__input, tp, th, m); 3310 3311 /* 3312 * Return any desired output. 3313 */ 3314 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3315 (void) tp->t_fb->tfb_tcp_output(tp); 3316 3317 check_delack: 3318 INP_WLOCK_ASSERT(tp->t_inpcb); 3319 3320 if (tp->t_flags & TF_DELACK) { 3321 tp->t_flags &= ~TF_DELACK; 3322 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3323 } 3324 INP_WUNLOCK(tp->t_inpcb); 3325 return; 3326 3327 dropafterack: 3328 /* 3329 * Generate an ACK dropping incoming segment if it occupies 3330 * sequence space, where the ACK reflects our state. 3331 * 3332 * We can now skip the test for the RST flag since all 3333 * paths to this code happen after packets containing 3334 * RST have been dropped. 3335 * 3336 * In the SYN-RECEIVED state, don't send an ACK unless the 3337 * segment we received passes the SYN-RECEIVED ACK test. 3338 * If it fails send a RST. This breaks the loop in the 3339 * "LAND" DoS attack, and also prevents an ACK storm 3340 * between two listening ports that have been sent forged 3341 * SYN segments, each with the source address of the other. 3342 */ 3343 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3344 (SEQ_GT(tp->snd_una, th->th_ack) || 3345 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3346 rstreason = BANDLIM_RST_OPENPORT; 3347 goto dropwithreset; 3348 } 3349 #ifdef TCPDEBUG 3350 if (so->so_options & SO_DEBUG) 3351 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3352 &tcp_savetcp, 0); 3353 #endif 3354 TCP_PROBE3(debug__input, tp, th, m); 3355 tp->t_flags |= TF_ACKNOW; 3356 (void) tp->t_fb->tfb_tcp_output(tp); 3357 INP_WUNLOCK(tp->t_inpcb); 3358 m_freem(m); 3359 return; 3360 3361 dropwithreset: 3362 if (tp != NULL) { 3363 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3364 INP_WUNLOCK(tp->t_inpcb); 3365 } else 3366 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3367 return; 3368 3369 drop: 3370 /* 3371 * Drop space held by incoming segment and return. 3372 */ 3373 #ifdef TCPDEBUG 3374 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3375 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3376 &tcp_savetcp, 0); 3377 #endif 3378 TCP_PROBE3(debug__input, tp, th, m); 3379 if (tp != NULL) { 3380 INP_WUNLOCK(tp->t_inpcb); 3381 } 3382 m_freem(m); 3383 } 3384 3385 /* 3386 * Issue RST and make ACK acceptable to originator of segment. 3387 * The mbuf must still include the original packet header. 3388 * tp may be NULL. 3389 */ 3390 void 3391 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3392 int tlen, int rstreason) 3393 { 3394 #ifdef INET 3395 struct ip *ip; 3396 #endif 3397 #ifdef INET6 3398 struct ip6_hdr *ip6; 3399 #endif 3400 3401 if (tp != NULL) { 3402 INP_WLOCK_ASSERT(tp->t_inpcb); 3403 } 3404 3405 /* Don't bother if destination was broadcast/multicast. */ 3406 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3407 goto drop; 3408 #ifdef INET6 3409 if (mtod(m, struct ip *)->ip_v == 6) { 3410 ip6 = mtod(m, struct ip6_hdr *); 3411 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3412 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3413 goto drop; 3414 /* IPv6 anycast check is done at tcp6_input() */ 3415 } 3416 #endif 3417 #if defined(INET) && defined(INET6) 3418 else 3419 #endif 3420 #ifdef INET 3421 { 3422 ip = mtod(m, struct ip *); 3423 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3424 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3425 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3426 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3427 goto drop; 3428 } 3429 #endif 3430 3431 /* Perform bandwidth limiting. */ 3432 if (badport_bandlim(rstreason) < 0) 3433 goto drop; 3434 3435 /* tcp_respond consumes the mbuf chain. */ 3436 if (th->th_flags & TH_ACK) { 3437 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3438 th->th_ack, TH_RST); 3439 } else { 3440 if (th->th_flags & TH_SYN) 3441 tlen++; 3442 if (th->th_flags & TH_FIN) 3443 tlen++; 3444 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3445 (tcp_seq)0, TH_RST|TH_ACK); 3446 } 3447 return; 3448 drop: 3449 m_freem(m); 3450 } 3451 3452 /* 3453 * Parse TCP options and place in tcpopt. 3454 */ 3455 void 3456 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3457 { 3458 int opt, optlen; 3459 3460 to->to_flags = 0; 3461 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3462 opt = cp[0]; 3463 if (opt == TCPOPT_EOL) 3464 break; 3465 if (opt == TCPOPT_NOP) 3466 optlen = 1; 3467 else { 3468 if (cnt < 2) 3469 break; 3470 optlen = cp[1]; 3471 if (optlen < 2 || optlen > cnt) 3472 break; 3473 } 3474 switch (opt) { 3475 case TCPOPT_MAXSEG: 3476 if (optlen != TCPOLEN_MAXSEG) 3477 continue; 3478 if (!(flags & TO_SYN)) 3479 continue; 3480 to->to_flags |= TOF_MSS; 3481 bcopy((char *)cp + 2, 3482 (char *)&to->to_mss, sizeof(to->to_mss)); 3483 to->to_mss = ntohs(to->to_mss); 3484 break; 3485 case TCPOPT_WINDOW: 3486 if (optlen != TCPOLEN_WINDOW) 3487 continue; 3488 if (!(flags & TO_SYN)) 3489 continue; 3490 to->to_flags |= TOF_SCALE; 3491 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3492 break; 3493 case TCPOPT_TIMESTAMP: 3494 if (optlen != TCPOLEN_TIMESTAMP) 3495 continue; 3496 to->to_flags |= TOF_TS; 3497 bcopy((char *)cp + 2, 3498 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3499 to->to_tsval = ntohl(to->to_tsval); 3500 bcopy((char *)cp + 6, 3501 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3502 to->to_tsecr = ntohl(to->to_tsecr); 3503 break; 3504 case TCPOPT_SIGNATURE: 3505 /* 3506 * In order to reply to a host which has set the 3507 * TCP_SIGNATURE option in its initial SYN, we have 3508 * to record the fact that the option was observed 3509 * here for the syncache code to perform the correct 3510 * response. 3511 */ 3512 if (optlen != TCPOLEN_SIGNATURE) 3513 continue; 3514 to->to_flags |= TOF_SIGNATURE; 3515 to->to_signature = cp + 2; 3516 break; 3517 case TCPOPT_SACK_PERMITTED: 3518 if (optlen != TCPOLEN_SACK_PERMITTED) 3519 continue; 3520 if (!(flags & TO_SYN)) 3521 continue; 3522 if (!V_tcp_do_sack) 3523 continue; 3524 to->to_flags |= TOF_SACKPERM; 3525 break; 3526 case TCPOPT_SACK: 3527 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3528 continue; 3529 if (flags & TO_SYN) 3530 continue; 3531 to->to_flags |= TOF_SACK; 3532 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3533 to->to_sacks = cp + 2; 3534 TCPSTAT_INC(tcps_sack_rcv_blocks); 3535 break; 3536 case TCPOPT_FAST_OPEN: 3537 /* 3538 * Cookie length validation is performed by the 3539 * server side cookie checking code or the client 3540 * side cookie cache update code. 3541 */ 3542 if (!(flags & TO_SYN)) 3543 continue; 3544 if (!V_tcp_fastopen_client_enable && 3545 !V_tcp_fastopen_server_enable) 3546 continue; 3547 to->to_flags |= TOF_FASTOPEN; 3548 to->to_tfo_len = optlen - 2; 3549 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3550 break; 3551 default: 3552 continue; 3553 } 3554 } 3555 } 3556 3557 /* 3558 * Pull out of band byte out of a segment so 3559 * it doesn't appear in the user's data queue. 3560 * It is still reflected in the segment length for 3561 * sequencing purposes. 3562 */ 3563 void 3564 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3565 int off) 3566 { 3567 int cnt = off + th->th_urp - 1; 3568 3569 while (cnt >= 0) { 3570 if (m->m_len > cnt) { 3571 char *cp = mtod(m, caddr_t) + cnt; 3572 struct tcpcb *tp = sototcpcb(so); 3573 3574 INP_WLOCK_ASSERT(tp->t_inpcb); 3575 3576 tp->t_iobc = *cp; 3577 tp->t_oobflags |= TCPOOB_HAVEDATA; 3578 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3579 m->m_len--; 3580 if (m->m_flags & M_PKTHDR) 3581 m->m_pkthdr.len--; 3582 return; 3583 } 3584 cnt -= m->m_len; 3585 m = m->m_next; 3586 if (m == NULL) 3587 break; 3588 } 3589 panic("tcp_pulloutofband"); 3590 } 3591 3592 /* 3593 * Collect new round-trip time estimate 3594 * and update averages and current timeout. 3595 */ 3596 void 3597 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3598 { 3599 int delta; 3600 3601 INP_WLOCK_ASSERT(tp->t_inpcb); 3602 3603 TCPSTAT_INC(tcps_rttupdated); 3604 tp->t_rttupdated++; 3605 #ifdef STATS 3606 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3607 imax(0, rtt * 1000 / hz)); 3608 #endif 3609 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3610 /* 3611 * srtt is stored as fixed point with 5 bits after the 3612 * binary point (i.e., scaled by 8). The following magic 3613 * is equivalent to the smoothing algorithm in rfc793 with 3614 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3615 * point). Adjust rtt to origin 0. 3616 */ 3617 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3618 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3619 3620 if ((tp->t_srtt += delta) <= 0) 3621 tp->t_srtt = 1; 3622 3623 /* 3624 * We accumulate a smoothed rtt variance (actually, a 3625 * smoothed mean difference), then set the retransmit 3626 * timer to smoothed rtt + 4 times the smoothed variance. 3627 * rttvar is stored as fixed point with 4 bits after the 3628 * binary point (scaled by 16). The following is 3629 * equivalent to rfc793 smoothing with an alpha of .75 3630 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3631 * rfc793's wired-in beta. 3632 */ 3633 if (delta < 0) 3634 delta = -delta; 3635 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3636 if ((tp->t_rttvar += delta) <= 0) 3637 tp->t_rttvar = 1; 3638 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3639 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3640 } else { 3641 /* 3642 * No rtt measurement yet - use the unsmoothed rtt. 3643 * Set the variance to half the rtt (so our first 3644 * retransmit happens at 3*rtt). 3645 */ 3646 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3647 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3648 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3649 } 3650 tp->t_rtttime = 0; 3651 tp->t_rxtshift = 0; 3652 3653 /* 3654 * the retransmit should happen at rtt + 4 * rttvar. 3655 * Because of the way we do the smoothing, srtt and rttvar 3656 * will each average +1/2 tick of bias. When we compute 3657 * the retransmit timer, we want 1/2 tick of rounding and 3658 * 1 extra tick because of +-1/2 tick uncertainty in the 3659 * firing of the timer. The bias will give us exactly the 3660 * 1.5 tick we need. But, because the bias is 3661 * statistical, we have to test that we don't drop below 3662 * the minimum feasible timer (which is 2 ticks). 3663 */ 3664 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3665 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3666 3667 /* 3668 * We received an ack for a packet that wasn't retransmitted; 3669 * it is probably safe to discard any error indications we've 3670 * received recently. This isn't quite right, but close enough 3671 * for now (a route might have failed after we sent a segment, 3672 * and the return path might not be symmetrical). 3673 */ 3674 tp->t_softerror = 0; 3675 } 3676 3677 /* 3678 * Determine a reasonable value for maxseg size. 3679 * If the route is known, check route for mtu. 3680 * If none, use an mss that can be handled on the outgoing interface 3681 * without forcing IP to fragment. If no route is found, route has no mtu, 3682 * or the destination isn't local, use a default, hopefully conservative 3683 * size (usually 512 or the default IP max size, but no more than the mtu 3684 * of the interface), as we can't discover anything about intervening 3685 * gateways or networks. We also initialize the congestion/slow start 3686 * window to be a single segment if the destination isn't local. 3687 * While looking at the routing entry, we also initialize other path-dependent 3688 * parameters from pre-set or cached values in the routing entry. 3689 * 3690 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3691 * IP options, e.g. IPSEC data, since length of this data may vary, and 3692 * thus it is calculated for every segment separately in tcp_output(). 3693 * 3694 * NOTE that this routine is only called when we process an incoming 3695 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3696 * settings are handled in tcp_mssopt(). 3697 */ 3698 void 3699 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3700 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3701 { 3702 int mss = 0; 3703 uint32_t maxmtu = 0; 3704 struct inpcb *inp = tp->t_inpcb; 3705 struct hc_metrics_lite metrics; 3706 #ifdef INET6 3707 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3708 size_t min_protoh = isipv6 ? 3709 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3710 sizeof (struct tcpiphdr); 3711 #else 3712 size_t min_protoh = sizeof(struct tcpiphdr); 3713 #endif 3714 3715 INP_WLOCK_ASSERT(tp->t_inpcb); 3716 3717 if (tp->t_port) 3718 min_protoh += V_tcp_udp_tunneling_overhead; 3719 if (mtuoffer != -1) { 3720 KASSERT(offer == -1, ("%s: conflict", __func__)); 3721 offer = mtuoffer - min_protoh; 3722 } 3723 3724 /* Initialize. */ 3725 #ifdef INET6 3726 if (isipv6) { 3727 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3728 tp->t_maxseg = V_tcp_v6mssdflt; 3729 } 3730 #endif 3731 #if defined(INET) && defined(INET6) 3732 else 3733 #endif 3734 #ifdef INET 3735 { 3736 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3737 tp->t_maxseg = V_tcp_mssdflt; 3738 } 3739 #endif 3740 3741 /* 3742 * No route to sender, stay with default mss and return. 3743 */ 3744 if (maxmtu == 0) { 3745 /* 3746 * In case we return early we need to initialize metrics 3747 * to a defined state as tcp_hc_get() would do for us 3748 * if there was no cache hit. 3749 */ 3750 if (metricptr != NULL) 3751 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3752 return; 3753 } 3754 3755 /* What have we got? */ 3756 switch (offer) { 3757 case 0: 3758 /* 3759 * Offer == 0 means that there was no MSS on the SYN 3760 * segment, in this case we use tcp_mssdflt as 3761 * already assigned to t_maxseg above. 3762 */ 3763 offer = tp->t_maxseg; 3764 break; 3765 3766 case -1: 3767 /* 3768 * Offer == -1 means that we didn't receive SYN yet. 3769 */ 3770 /* FALLTHROUGH */ 3771 3772 default: 3773 /* 3774 * Prevent DoS attack with too small MSS. Round up 3775 * to at least minmss. 3776 */ 3777 offer = max(offer, V_tcp_minmss); 3778 } 3779 3780 /* 3781 * rmx information is now retrieved from tcp_hostcache. 3782 */ 3783 tcp_hc_get(&inp->inp_inc, &metrics); 3784 if (metricptr != NULL) 3785 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3786 3787 /* 3788 * If there's a discovered mtu in tcp hostcache, use it. 3789 * Else, use the link mtu. 3790 */ 3791 if (metrics.rmx_mtu) 3792 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3793 else { 3794 #ifdef INET6 3795 if (isipv6) { 3796 mss = maxmtu - min_protoh; 3797 if (!V_path_mtu_discovery && 3798 !in6_localaddr(&inp->in6p_faddr)) 3799 mss = min(mss, V_tcp_v6mssdflt); 3800 } 3801 #endif 3802 #if defined(INET) && defined(INET6) 3803 else 3804 #endif 3805 #ifdef INET 3806 { 3807 mss = maxmtu - min_protoh; 3808 if (!V_path_mtu_discovery && 3809 !in_localaddr(inp->inp_faddr)) 3810 mss = min(mss, V_tcp_mssdflt); 3811 } 3812 #endif 3813 /* 3814 * XXX - The above conditional (mss = maxmtu - min_protoh) 3815 * probably violates the TCP spec. 3816 * The problem is that, since we don't know the 3817 * other end's MSS, we are supposed to use a conservative 3818 * default. But, if we do that, then MTU discovery will 3819 * never actually take place, because the conservative 3820 * default is much less than the MTUs typically seen 3821 * on the Internet today. For the moment, we'll sweep 3822 * this under the carpet. 3823 * 3824 * The conservative default might not actually be a problem 3825 * if the only case this occurs is when sending an initial 3826 * SYN with options and data to a host we've never talked 3827 * to before. Then, they will reply with an MSS value which 3828 * will get recorded and the new parameters should get 3829 * recomputed. For Further Study. 3830 */ 3831 } 3832 mss = min(mss, offer); 3833 3834 /* 3835 * Sanity check: make sure that maxseg will be large 3836 * enough to allow some data on segments even if the 3837 * all the option space is used (40bytes). Otherwise 3838 * funny things may happen in tcp_output. 3839 * 3840 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3841 */ 3842 mss = max(mss, 64); 3843 3844 tp->t_maxseg = mss; 3845 } 3846 3847 void 3848 tcp_mss(struct tcpcb *tp, int offer) 3849 { 3850 int mss; 3851 uint32_t bufsize; 3852 struct inpcb *inp; 3853 struct socket *so; 3854 struct hc_metrics_lite metrics; 3855 struct tcp_ifcap cap; 3856 3857 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3858 3859 bzero(&cap, sizeof(cap)); 3860 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3861 3862 mss = tp->t_maxseg; 3863 inp = tp->t_inpcb; 3864 3865 /* 3866 * If there's a pipesize, change the socket buffer to that size, 3867 * don't change if sb_hiwat is different than default (then it 3868 * has been changed on purpose with setsockopt). 3869 * Make the socket buffers an integral number of mss units; 3870 * if the mss is larger than the socket buffer, decrease the mss. 3871 */ 3872 so = inp->inp_socket; 3873 SOCKBUF_LOCK(&so->so_snd); 3874 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3875 bufsize = metrics.rmx_sendpipe; 3876 else 3877 bufsize = so->so_snd.sb_hiwat; 3878 if (bufsize < mss) 3879 mss = bufsize; 3880 else { 3881 bufsize = roundup(bufsize, mss); 3882 if (bufsize > sb_max) 3883 bufsize = sb_max; 3884 if (bufsize > so->so_snd.sb_hiwat) 3885 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3886 } 3887 SOCKBUF_UNLOCK(&so->so_snd); 3888 /* 3889 * Sanity check: make sure that maxseg will be large 3890 * enough to allow some data on segments even if the 3891 * all the option space is used (40bytes). Otherwise 3892 * funny things may happen in tcp_output. 3893 * 3894 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3895 */ 3896 tp->t_maxseg = max(mss, 64); 3897 3898 SOCKBUF_LOCK(&so->so_rcv); 3899 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3900 bufsize = metrics.rmx_recvpipe; 3901 else 3902 bufsize = so->so_rcv.sb_hiwat; 3903 if (bufsize > mss) { 3904 bufsize = roundup(bufsize, mss); 3905 if (bufsize > sb_max) 3906 bufsize = sb_max; 3907 if (bufsize > so->so_rcv.sb_hiwat) 3908 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3909 } 3910 SOCKBUF_UNLOCK(&so->so_rcv); 3911 3912 /* Check the interface for TSO capabilities. */ 3913 if (cap.ifcap & CSUM_TSO) { 3914 tp->t_flags |= TF_TSO; 3915 tp->t_tsomax = cap.tsomax; 3916 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3917 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3918 } 3919 } 3920 3921 /* 3922 * Determine the MSS option to send on an outgoing SYN. 3923 */ 3924 int 3925 tcp_mssopt(struct in_conninfo *inc) 3926 { 3927 int mss = 0; 3928 uint32_t thcmtu = 0; 3929 uint32_t maxmtu = 0; 3930 size_t min_protoh; 3931 3932 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3933 3934 #ifdef INET6 3935 if (inc->inc_flags & INC_ISIPV6) { 3936 mss = V_tcp_v6mssdflt; 3937 maxmtu = tcp_maxmtu6(inc, NULL); 3938 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3939 } 3940 #endif 3941 #if defined(INET) && defined(INET6) 3942 else 3943 #endif 3944 #ifdef INET 3945 { 3946 mss = V_tcp_mssdflt; 3947 maxmtu = tcp_maxmtu(inc, NULL); 3948 min_protoh = sizeof(struct tcpiphdr); 3949 } 3950 #endif 3951 #if defined(INET6) || defined(INET) 3952 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3953 #endif 3954 3955 if (maxmtu && thcmtu) 3956 mss = min(maxmtu, thcmtu) - min_protoh; 3957 else if (maxmtu || thcmtu) 3958 mss = max(maxmtu, thcmtu) - min_protoh; 3959 3960 return (mss); 3961 } 3962 3963 void 3964 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3965 { 3966 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3967 int maxseg = tcp_maxseg(tp); 3968 3969 INP_WLOCK_ASSERT(tp->t_inpcb); 3970 3971 /* 3972 * Compute the amount of data that this ACK is indicating 3973 * (del_data) and an estimate of how many bytes are in the 3974 * network. 3975 */ 3976 if (((tp->t_flags & TF_SACK_PERMIT) && 3977 (to->to_flags & TOF_SACK)) || 3978 (IN_CONGRECOVERY(tp->t_flags) && 3979 !IN_FASTRECOVERY(tp->t_flags))) { 3980 del_data = tp->sackhint.delivered_data; 3981 if (V_tcp_do_rfc6675_pipe) 3982 pipe = tcp_compute_pipe(tp); 3983 else 3984 pipe = (tp->snd_nxt - tp->snd_fack) + 3985 tp->sackhint.sack_bytes_rexmit; 3986 } else { 3987 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 3988 tp->snd_recover - tp->snd_una)) 3989 del_data = maxseg; 3990 pipe = imax(0, tp->snd_max - tp->snd_una - 3991 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 3992 } 3993 tp->sackhint.prr_delivered += del_data; 3994 /* 3995 * Proportional Rate Reduction 3996 */ 3997 if (pipe >= tp->snd_ssthresh) { 3998 if (tp->sackhint.recover_fs == 0) 3999 tp->sackhint.recover_fs = 4000 imax(1, tp->snd_nxt - tp->snd_una); 4001 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 4002 tp->snd_ssthresh, tp->sackhint.recover_fs) - 4003 tp->sackhint.prr_out; 4004 } else { 4005 if (V_tcp_do_prr_conservative || (del_data == 0)) 4006 limit = tp->sackhint.prr_delivered - 4007 tp->sackhint.prr_out; 4008 else 4009 limit = imax(tp->sackhint.prr_delivered - 4010 tp->sackhint.prr_out, del_data) + 4011 maxseg; 4012 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 4013 } 4014 snd_cnt = imax(snd_cnt, 0) / maxseg; 4015 /* 4016 * Send snd_cnt new data into the network in response to this ack. 4017 * If there is going to be a SACK retransmission, adjust snd_cwnd 4018 * accordingly. 4019 */ 4020 if (IN_FASTRECOVERY(tp->t_flags)) { 4021 if ((tp->t_flags & TF_SACK_PERMIT) && 4022 (to->to_flags & TOF_SACK)) { 4023 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 4024 tp->sackhint.sack_bytes_rexmit + 4025 (snd_cnt * maxseg); 4026 } else { 4027 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4028 (snd_cnt * maxseg); 4029 } 4030 } else if (IN_CONGRECOVERY(tp->t_flags)) 4031 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4032 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4033 } 4034 4035 /* 4036 * On a partial ack arrives, force the retransmission of the 4037 * next unacknowledged segment. Do not clear tp->t_dupacks. 4038 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4039 * be started again. 4040 */ 4041 void 4042 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4043 { 4044 tcp_seq onxt = tp->snd_nxt; 4045 uint32_t ocwnd = tp->snd_cwnd; 4046 u_int maxseg = tcp_maxseg(tp); 4047 4048 INP_WLOCK_ASSERT(tp->t_inpcb); 4049 4050 tcp_timer_activate(tp, TT_REXMT, 0); 4051 tp->t_rtttime = 0; 4052 tp->snd_nxt = th->th_ack; 4053 /* 4054 * Set snd_cwnd to one segment beyond acknowledged offset. 4055 * (tp->snd_una has not yet been updated when this function is called.) 4056 */ 4057 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4058 tp->t_flags |= TF_ACKNOW; 4059 (void) tp->t_fb->tfb_tcp_output(tp); 4060 tp->snd_cwnd = ocwnd; 4061 if (SEQ_GT(onxt, tp->snd_nxt)) 4062 tp->snd_nxt = onxt; 4063 /* 4064 * Partial window deflation. Relies on fact that tp->snd_una 4065 * not updated yet. 4066 */ 4067 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4068 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4069 else 4070 tp->snd_cwnd = 0; 4071 tp->snd_cwnd += maxseg; 4072 } 4073 4074 int 4075 tcp_compute_pipe(struct tcpcb *tp) 4076 { 4077 return (tp->snd_max - tp->snd_una + 4078 tp->sackhint.sack_bytes_rexmit - 4079 tp->sackhint.sacked_bytes); 4080 } 4081 4082 uint32_t 4083 tcp_compute_initwnd(uint32_t maxseg) 4084 { 4085 /* 4086 * Calculate the Initial Window, also used as Restart Window 4087 * 4088 * RFC5681 Section 3.1 specifies the default conservative values. 4089 * RFC3390 specifies slightly more aggressive values. 4090 * RFC6928 increases it to ten segments. 4091 * Support for user specified value for initial flight size. 4092 */ 4093 if (V_tcp_initcwnd_segments) 4094 return min(V_tcp_initcwnd_segments * maxseg, 4095 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4096 else if (V_tcp_do_rfc3390) 4097 return min(4 * maxseg, max(2 * maxseg, 4380)); 4098 else { 4099 /* Per RFC5681 Section 3.1 */ 4100 if (maxseg > 2190) 4101 return (2 * maxseg); 4102 else if (maxseg > 1095) 4103 return (3 * maxseg); 4104 else 4105 return (4 * maxseg); 4106 } 4107 } 4108