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