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