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