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