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_rfc6675_pipe) = 0; 172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 173 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 174 "Use calculated pipe/in-flight bytes per RFC 6675"); 175 176 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 178 &VNET_NAME(tcp_do_rfc3042), 0, 179 "Enable RFC 3042 (Limited Transmit)"); 180 181 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 183 &VNET_NAME(tcp_do_rfc3390), 0, 184 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 185 186 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 188 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 189 "Slow-start flight size (initial congestion window) in number of segments"); 190 191 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_do_rfc3465), 0, 194 "Enable RFC 3465 (Appropriate Byte Counting)"); 195 196 VNET_DEFINE(int, tcp_abc_l_var) = 2; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_abc_l_var), 2, 199 "Cap the max cwnd increment during slow-start to this number of segments"); 200 201 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 202 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 203 "TCP ECN"); 204 205 VNET_DEFINE(int, tcp_do_ecn) = 2; 206 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(tcp_do_ecn), 0, 208 "TCP ECN support"); 209 210 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 211 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_ecn_maxretries), 0, 213 "Max retries before giving up on ECN"); 214 215 VNET_DEFINE(int, tcp_insecure_syn) = 0; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_insecure_syn), 0, 218 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 219 220 VNET_DEFINE(int, tcp_insecure_rst) = 0; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_insecure_rst), 0, 223 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 224 225 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 226 #define V_tcp_recvspace VNET(tcp_recvspace) 227 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 229 230 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 232 &VNET_NAME(tcp_do_autorcvbuf), 0, 233 "Enable automatic receive buffer sizing"); 234 235 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 237 &VNET_NAME(tcp_autorcvbuf_max), 0, 238 "Max size of automatic receive buffer"); 239 240 VNET_DEFINE(struct inpcbhead, tcb); 241 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 242 VNET_DEFINE(struct inpcbinfo, tcbinfo); 243 244 /* 245 * TCP statistics are stored in an array of counter(9)s, which size matches 246 * size of struct tcpstat. TCP running connection count is a regular array. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 250 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 251 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 252 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 253 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 254 "TCP connection counts by TCP state"); 255 256 static void 257 tcp_vnet_init(const void *unused) 258 { 259 260 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 261 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 262 } 263 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 264 tcp_vnet_init, NULL); 265 266 #ifdef VIMAGE 267 static void 268 tcp_vnet_uninit(const void *unused) 269 { 270 271 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 272 VNET_PCPUSTAT_FREE(tcpstat); 273 } 274 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 275 tcp_vnet_uninit, NULL); 276 #endif /* VIMAGE */ 277 278 /* 279 * Kernel module interface for updating tcpstat. The first argument is an index 280 * into tcpstat treated as an array. 281 */ 282 void 283 kmod_tcpstat_add(int statnum, int val) 284 { 285 286 counter_u64_add(VNET(tcpstat)[statnum], val); 287 } 288 289 #ifdef TCP_HHOOK 290 /* 291 * Wrapper for the TCP established input helper hook. 292 */ 293 void 294 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 295 { 296 struct tcp_hhook_data hhook_data; 297 298 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 299 hhook_data.tp = tp; 300 hhook_data.th = th; 301 hhook_data.to = to; 302 303 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 304 tp->osd); 305 } 306 } 307 #endif 308 309 /* 310 * CC wrapper hook functions 311 */ 312 void 313 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 314 uint16_t type) 315 { 316 #ifdef STATS 317 int32_t gput; 318 #endif 319 320 INP_WLOCK_ASSERT(tp->t_inpcb); 321 322 tp->ccv->nsegs = nsegs; 323 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 324 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 325 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 326 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 327 tp->ccv->flags |= CCF_CWND_LIMITED; 328 else 329 tp->ccv->flags &= ~CCF_CWND_LIMITED; 330 331 if (type == CC_ACK) { 332 #ifdef STATS 333 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 334 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 335 if (!IN_RECOVERY(tp->t_flags)) 336 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 337 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 338 if ((tp->t_flags & TF_GPUTINPROG) && 339 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 340 /* 341 * Compute goodput in bits per millisecond. 342 */ 343 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) / 344 max(1, tcp_ts_getticks() - tp->gput_ts); 345 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 346 gput); 347 /* 348 * XXXLAS: This is a temporary hack, and should be 349 * chained off VOI_TCP_GPUT when stats(9) grows an API 350 * to deal with chained VOIs. 351 */ 352 if (tp->t_stats_gput_prev > 0) 353 stats_voi_update_abs_s32(tp->t_stats, 354 VOI_TCP_GPUT_ND, 355 ((gput - tp->t_stats_gput_prev) * 100) / 356 tp->t_stats_gput_prev); 357 tp->t_flags &= ~TF_GPUTINPROG; 358 tp->t_stats_gput_prev = gput; 359 } 360 #endif /* STATS */ 361 if (tp->snd_cwnd > tp->snd_ssthresh) { 362 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 363 if (tp->t_bytes_acked >= tp->snd_cwnd) { 364 tp->t_bytes_acked -= tp->snd_cwnd; 365 tp->ccv->flags |= CCF_ABC_SENTAWND; 366 } 367 } else { 368 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 369 tp->t_bytes_acked = 0; 370 } 371 } 372 373 if (CC_ALGO(tp)->ack_received != NULL) { 374 /* XXXLAS: Find a way to live without this */ 375 tp->ccv->curack = th->th_ack; 376 CC_ALGO(tp)->ack_received(tp->ccv, type); 377 } 378 #ifdef STATS 379 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 380 #endif 381 } 382 383 void 384 cc_conn_init(struct tcpcb *tp) 385 { 386 struct hc_metrics_lite metrics; 387 struct inpcb *inp = tp->t_inpcb; 388 u_int maxseg; 389 int rtt; 390 391 INP_WLOCK_ASSERT(tp->t_inpcb); 392 393 tcp_hc_get(&inp->inp_inc, &metrics); 394 maxseg = tcp_maxseg(tp); 395 396 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 397 tp->t_srtt = rtt; 398 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 399 TCPSTAT_INC(tcps_usedrtt); 400 if (metrics.rmx_rttvar) { 401 tp->t_rttvar = metrics.rmx_rttvar; 402 TCPSTAT_INC(tcps_usedrttvar); 403 } else { 404 /* default variation is +- 1 rtt */ 405 tp->t_rttvar = 406 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 407 } 408 TCPT_RANGESET(tp->t_rxtcur, 409 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 410 tp->t_rttmin, TCPTV_REXMTMAX); 411 } 412 if (metrics.rmx_ssthresh) { 413 /* 414 * There's some sort of gateway or interface 415 * buffer limit on the path. Use this to set 416 * the slow start threshold, but set the 417 * threshold to no less than 2*mss. 418 */ 419 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 420 TCPSTAT_INC(tcps_usedssthresh); 421 } 422 423 /* 424 * Set the initial slow-start flight size. 425 * 426 * If a SYN or SYN/ACK was lost and retransmitted, we have to 427 * reduce the initial CWND to one segment as congestion is likely 428 * requiring us to be cautious. 429 */ 430 if (tp->snd_cwnd == 1) 431 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 432 else 433 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 434 435 if (CC_ALGO(tp)->conn_init != NULL) 436 CC_ALGO(tp)->conn_init(tp->ccv); 437 } 438 439 void inline 440 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 441 { 442 INP_WLOCK_ASSERT(tp->t_inpcb); 443 444 #ifdef STATS 445 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 446 #endif 447 448 switch(type) { 449 case CC_NDUPACK: 450 if (!IN_FASTRECOVERY(tp->t_flags)) { 451 tp->snd_recover = tp->snd_max; 452 if (tp->t_flags2 & TF2_ECN_PERMIT) 453 tp->t_flags2 |= TF2_ECN_SND_CWR; 454 } 455 break; 456 case CC_ECN: 457 if (!IN_CONGRECOVERY(tp->t_flags) || 458 /* 459 * Allow ECN reaction on ACK to CWR, if 460 * that data segment was also CE marked. 461 */ 462 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 463 EXIT_CONGRECOVERY(tp->t_flags); 464 TCPSTAT_INC(tcps_ecn_rcwnd); 465 tp->snd_recover = tp->snd_max + 1; 466 if (tp->t_flags2 & TF2_ECN_PERMIT) 467 tp->t_flags2 |= TF2_ECN_SND_CWR; 468 } 469 break; 470 case CC_RTO: 471 tp->t_dupacks = 0; 472 tp->t_bytes_acked = 0; 473 EXIT_RECOVERY(tp->t_flags); 474 if (tp->t_flags2 & TF2_ECN_PERMIT) 475 tp->t_flags2 |= TF2_ECN_SND_CWR; 476 break; 477 case CC_RTO_ERR: 478 TCPSTAT_INC(tcps_sndrexmitbad); 479 /* RTO was unnecessary, so reset everything. */ 480 tp->snd_cwnd = tp->snd_cwnd_prev; 481 tp->snd_ssthresh = tp->snd_ssthresh_prev; 482 tp->snd_recover = tp->snd_recover_prev; 483 if (tp->t_flags & TF_WASFRECOVERY) 484 ENTER_FASTRECOVERY(tp->t_flags); 485 if (tp->t_flags & TF_WASCRECOVERY) 486 ENTER_CONGRECOVERY(tp->t_flags); 487 tp->snd_nxt = tp->snd_max; 488 tp->t_flags &= ~TF_PREVVALID; 489 tp->t_badrxtwin = 0; 490 break; 491 } 492 493 if (CC_ALGO(tp)->cong_signal != NULL) { 494 if (th != NULL) 495 tp->ccv->curack = th->th_ack; 496 CC_ALGO(tp)->cong_signal(tp->ccv, type); 497 } 498 } 499 500 void inline 501 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 502 { 503 INP_WLOCK_ASSERT(tp->t_inpcb); 504 505 /* XXXLAS: KASSERT that we're in recovery? */ 506 507 if (CC_ALGO(tp)->post_recovery != NULL) { 508 tp->ccv->curack = th->th_ack; 509 CC_ALGO(tp)->post_recovery(tp->ccv); 510 } 511 /* XXXLAS: EXIT_RECOVERY ? */ 512 tp->t_bytes_acked = 0; 513 } 514 515 /* 516 * Indicate whether this ack should be delayed. We can delay the ack if 517 * following conditions are met: 518 * - There is no delayed ack timer in progress. 519 * - Our last ack wasn't a 0-sized window. We never want to delay 520 * the ack that opens up a 0-sized window. 521 * - LRO wasn't used for this segment. We make sure by checking that the 522 * segment size is not larger than the MSS. 523 */ 524 #define DELAY_ACK(tp, tlen) \ 525 ((!tcp_timer_active(tp, TT_DELACK) && \ 526 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 527 (tlen <= tp->t_maxseg) && \ 528 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 529 530 void inline 531 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 532 { 533 INP_WLOCK_ASSERT(tp->t_inpcb); 534 535 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 536 switch (iptos & IPTOS_ECN_MASK) { 537 case IPTOS_ECN_CE: 538 tp->ccv->flags |= CCF_IPHDR_CE; 539 break; 540 case IPTOS_ECN_ECT0: 541 /* FALLTHROUGH */ 542 case IPTOS_ECN_ECT1: 543 /* FALLTHROUGH */ 544 case IPTOS_ECN_NOTECT: 545 tp->ccv->flags &= ~CCF_IPHDR_CE; 546 break; 547 } 548 549 if (th->th_flags & TH_CWR) 550 tp->ccv->flags |= CCF_TCPHDR_CWR; 551 else 552 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 553 554 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 555 556 if (tp->ccv->flags & CCF_ACKNOW) { 557 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 558 tp->t_flags |= TF_ACKNOW; 559 } 560 } 561 } 562 563 /* 564 * TCP input handling is split into multiple parts: 565 * tcp6_input is a thin wrapper around tcp_input for the extended 566 * ip6_protox[] call format in ip6_input 567 * tcp_input handles primary segment validation, inpcb lookup and 568 * SYN processing on listen sockets 569 * tcp_do_segment processes the ACK and text of the segment for 570 * establishing, established and closing connections 571 */ 572 #ifdef INET6 573 int 574 tcp6_input(struct mbuf **mp, int *offp, int proto) 575 { 576 struct mbuf *m; 577 struct in6_ifaddr *ia6; 578 struct ip6_hdr *ip6; 579 580 m = *mp; 581 if (m->m_len < *offp + sizeof(struct tcphdr)) { 582 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 583 if (m == NULL) { 584 *mp = m; 585 TCPSTAT_INC(tcps_rcvshort); 586 return (IPPROTO_DONE); 587 } 588 } 589 590 /* 591 * draft-itojun-ipv6-tcp-to-anycast 592 * better place to put this in? 593 */ 594 ip6 = mtod(m, struct ip6_hdr *); 595 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 596 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 597 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 598 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 599 *mp = NULL; 600 return (IPPROTO_DONE); 601 } 602 603 *mp = m; 604 return (tcp_input(mp, offp, proto)); 605 } 606 #endif /* INET6 */ 607 608 int 609 tcp_input(struct mbuf **mp, int *offp, int proto) 610 { 611 struct mbuf *m = *mp; 612 struct tcphdr *th = NULL; 613 struct ip *ip = NULL; 614 struct inpcb *inp = NULL; 615 struct tcpcb *tp = NULL; 616 struct socket *so = NULL; 617 u_char *optp = NULL; 618 int off0; 619 int optlen = 0; 620 #ifdef INET 621 int len; 622 uint8_t ipttl; 623 #endif 624 int tlen = 0, off; 625 int drop_hdrlen; 626 int thflags; 627 int rstreason = 0; /* For badport_bandlim accounting purposes */ 628 uint8_t iptos; 629 struct m_tag *fwd_tag = NULL; 630 #ifdef INET6 631 struct ip6_hdr *ip6 = NULL; 632 int isipv6; 633 #else 634 const void *ip6 = NULL; 635 #endif /* INET6 */ 636 struct tcpopt to; /* options in this segment */ 637 char *s = NULL; /* address and port logging */ 638 #ifdef TCPDEBUG 639 /* 640 * The size of tcp_saveipgen must be the size of the max ip header, 641 * now IPv6. 642 */ 643 u_char tcp_saveipgen[IP6_HDR_LEN]; 644 struct tcphdr tcp_savetcp; 645 short ostate = 0; 646 #endif 647 648 NET_EPOCH_ASSERT(); 649 650 #ifdef INET6 651 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 652 #endif 653 654 off0 = *offp; 655 m = *mp; 656 *mp = NULL; 657 to.to_flags = 0; 658 TCPSTAT_INC(tcps_rcvtotal); 659 660 #ifdef INET6 661 if (isipv6) { 662 ip6 = mtod(m, struct ip6_hdr *); 663 th = (struct tcphdr *)((caddr_t)ip6 + off0); 664 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 665 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 666 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 667 th->th_sum = m->m_pkthdr.csum_data; 668 else 669 th->th_sum = in6_cksum_pseudo(ip6, tlen, 670 IPPROTO_TCP, m->m_pkthdr.csum_data); 671 th->th_sum ^= 0xffff; 672 } else 673 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 674 if (th->th_sum) { 675 TCPSTAT_INC(tcps_rcvbadsum); 676 goto drop; 677 } 678 679 /* 680 * Be proactive about unspecified IPv6 address in source. 681 * As we use all-zero to indicate unbounded/unconnected pcb, 682 * unspecified IPv6 address can be used to confuse us. 683 * 684 * Note that packets with unspecified IPv6 destination is 685 * already dropped in ip6_input. 686 */ 687 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 688 /* XXX stat */ 689 goto drop; 690 } 691 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 692 } 693 #endif 694 #if defined(INET) && defined(INET6) 695 else 696 #endif 697 #ifdef INET 698 { 699 /* 700 * Get IP and TCP header together in first mbuf. 701 * Note: IP leaves IP header in first mbuf. 702 */ 703 if (off0 > sizeof (struct ip)) { 704 ip_stripoptions(m); 705 off0 = sizeof(struct ip); 706 } 707 if (m->m_len < sizeof (struct tcpiphdr)) { 708 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 709 == NULL) { 710 TCPSTAT_INC(tcps_rcvshort); 711 return (IPPROTO_DONE); 712 } 713 } 714 ip = mtod(m, struct ip *); 715 th = (struct tcphdr *)((caddr_t)ip + off0); 716 tlen = ntohs(ip->ip_len) - off0; 717 718 iptos = ip->ip_tos; 719 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 720 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 721 th->th_sum = m->m_pkthdr.csum_data; 722 else 723 th->th_sum = in_pseudo(ip->ip_src.s_addr, 724 ip->ip_dst.s_addr, 725 htonl(m->m_pkthdr.csum_data + tlen + 726 IPPROTO_TCP)); 727 th->th_sum ^= 0xffff; 728 } else { 729 struct ipovly *ipov = (struct ipovly *)ip; 730 731 /* 732 * Checksum extended TCP header and data. 733 */ 734 len = off0 + tlen; 735 ipttl = ip->ip_ttl; 736 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 737 ipov->ih_len = htons(tlen); 738 th->th_sum = in_cksum(m, len); 739 /* Reset length for SDT probes. */ 740 ip->ip_len = htons(len); 741 /* Reset TOS bits */ 742 ip->ip_tos = iptos; 743 /* Re-initialization for later version check */ 744 ip->ip_ttl = ipttl; 745 ip->ip_v = IPVERSION; 746 ip->ip_hl = off0 >> 2; 747 } 748 749 if (th->th_sum) { 750 TCPSTAT_INC(tcps_rcvbadsum); 751 goto drop; 752 } 753 } 754 #endif /* INET */ 755 756 /* 757 * Check that TCP offset makes sense, 758 * pull out TCP options and adjust length. XXX 759 */ 760 off = th->th_off << 2; 761 if (off < sizeof (struct tcphdr) || off > tlen) { 762 TCPSTAT_INC(tcps_rcvbadoff); 763 goto drop; 764 } 765 tlen -= off; /* tlen is used instead of ti->ti_len */ 766 if (off > sizeof (struct tcphdr)) { 767 #ifdef INET6 768 if (isipv6) { 769 if (m->m_len < off0 + off) { 770 m = m_pullup(m, off0 + off); 771 if (m == NULL) { 772 TCPSTAT_INC(tcps_rcvshort); 773 return (IPPROTO_DONE); 774 } 775 } 776 ip6 = mtod(m, struct ip6_hdr *); 777 th = (struct tcphdr *)((caddr_t)ip6 + off0); 778 } 779 #endif 780 #if defined(INET) && defined(INET6) 781 else 782 #endif 783 #ifdef INET 784 { 785 if (m->m_len < sizeof(struct ip) + off) { 786 if ((m = m_pullup(m, sizeof (struct ip) + off)) 787 == NULL) { 788 TCPSTAT_INC(tcps_rcvshort); 789 return (IPPROTO_DONE); 790 } 791 ip = mtod(m, struct ip *); 792 th = (struct tcphdr *)((caddr_t)ip + off0); 793 } 794 } 795 #endif 796 optlen = off - sizeof (struct tcphdr); 797 optp = (u_char *)(th + 1); 798 } 799 thflags = th->th_flags; 800 801 /* 802 * Convert TCP protocol specific fields to host format. 803 */ 804 tcp_fields_to_host(th); 805 806 /* 807 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 808 */ 809 drop_hdrlen = off0 + off; 810 811 /* 812 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 813 */ 814 if ( 815 #ifdef INET6 816 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 817 #ifdef INET 818 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 819 #endif 820 #endif 821 #if defined(INET) && !defined(INET6) 822 (m->m_flags & M_IP_NEXTHOP) 823 #endif 824 ) 825 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 826 827 findpcb: 828 #ifdef INET6 829 if (isipv6 && fwd_tag != NULL) { 830 struct sockaddr_in6 *next_hop6; 831 832 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 833 /* 834 * Transparently forwarded. Pretend to be the destination. 835 * Already got one like this? 836 */ 837 inp = in6_pcblookup_mbuf(&V_tcbinfo, 838 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 839 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 840 if (!inp) { 841 /* 842 * It's new. Try to find the ambushing socket. 843 * Because we've rewritten the destination address, 844 * any hardware-generated hash is ignored. 845 */ 846 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 847 th->th_sport, &next_hop6->sin6_addr, 848 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 849 th->th_dport, INPLOOKUP_WILDCARD | 850 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 851 } 852 } else if (isipv6) { 853 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 854 th->th_sport, &ip6->ip6_dst, th->th_dport, 855 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 856 m->m_pkthdr.rcvif, m); 857 } 858 #endif /* INET6 */ 859 #if defined(INET6) && defined(INET) 860 else 861 #endif 862 #ifdef INET 863 if (fwd_tag != NULL) { 864 struct sockaddr_in *next_hop; 865 866 next_hop = (struct sockaddr_in *)(fwd_tag+1); 867 /* 868 * Transparently forwarded. Pretend to be the destination. 869 * already got one like this? 870 */ 871 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 872 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 873 m->m_pkthdr.rcvif, m); 874 if (!inp) { 875 /* 876 * It's new. Try to find the ambushing socket. 877 * Because we've rewritten the destination address, 878 * any hardware-generated hash is ignored. 879 */ 880 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 881 th->th_sport, next_hop->sin_addr, 882 next_hop->sin_port ? ntohs(next_hop->sin_port) : 883 th->th_dport, INPLOOKUP_WILDCARD | 884 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 885 } 886 } else 887 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 888 th->th_sport, ip->ip_dst, th->th_dport, 889 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 890 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_WLOCK_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_WUNLOCK(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 INP_WLOCK_ASSERT(inp); 1017 if (mac_inpcb_check_deliver(inp, m)) 1018 goto dropunlock; 1019 #endif 1020 so = inp->inp_socket; 1021 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1022 #ifdef TCPDEBUG 1023 if (so->so_options & SO_DEBUG) { 1024 ostate = tp->t_state; 1025 #ifdef INET6 1026 if (isipv6) { 1027 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1028 } else 1029 #endif 1030 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1031 tcp_savetcp = *th; 1032 } 1033 #endif /* TCPDEBUG */ 1034 /* 1035 * When the socket is accepting connections (the INPCB is in LISTEN 1036 * state) we look into the SYN cache if this is a new connection 1037 * attempt or the completion of a previous one. 1038 */ 1039 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1040 ("%s: so accepting but tp %p not listening", __func__, tp)); 1041 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1042 struct in_conninfo inc; 1043 1044 bzero(&inc, sizeof(inc)); 1045 #ifdef INET6 1046 if (isipv6) { 1047 inc.inc_flags |= INC_ISIPV6; 1048 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1049 inc.inc_flags |= INC_IPV6MINMTU; 1050 inc.inc6_faddr = ip6->ip6_src; 1051 inc.inc6_laddr = ip6->ip6_dst; 1052 } else 1053 #endif 1054 { 1055 inc.inc_faddr = ip->ip_src; 1056 inc.inc_laddr = ip->ip_dst; 1057 } 1058 inc.inc_fport = th->th_sport; 1059 inc.inc_lport = th->th_dport; 1060 inc.inc_fibnum = so->so_fibnum; 1061 1062 /* 1063 * Check for an existing connection attempt in syncache if 1064 * the flag is only ACK. A successful lookup creates a new 1065 * socket appended to the listen queue in SYN_RECEIVED state. 1066 */ 1067 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1068 /* 1069 * Parse the TCP options here because 1070 * syncookies need access to the reflected 1071 * timestamp. 1072 */ 1073 tcp_dooptions(&to, optp, optlen, 0); 1074 /* 1075 * NB: syncache_expand() doesn't unlock 1076 * inp and tcpinfo locks. 1077 */ 1078 rstreason = syncache_expand(&inc, &to, th, &so, m); 1079 if (rstreason < 0) { 1080 /* 1081 * A failing TCP MD5 signature comparison 1082 * must result in the segment being dropped 1083 * and must not produce any response back 1084 * to the sender. 1085 */ 1086 goto dropunlock; 1087 } else if (rstreason == 0) { 1088 /* 1089 * No syncache entry or ACK was not 1090 * for our SYN/ACK. Send a RST. 1091 * NB: syncache did its own logging 1092 * of the failure cause. 1093 */ 1094 rstreason = BANDLIM_RST_OPENPORT; 1095 goto dropwithreset; 1096 } 1097 tfo_socket_result: 1098 if (so == NULL) { 1099 /* 1100 * We completed the 3-way handshake 1101 * but could not allocate a socket 1102 * either due to memory shortage, 1103 * listen queue length limits or 1104 * global socket limits. Send RST 1105 * or wait and have the remote end 1106 * retransmit the ACK for another 1107 * try. 1108 */ 1109 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1110 log(LOG_DEBUG, "%s; %s: Listen socket: " 1111 "Socket allocation failed due to " 1112 "limits or memory shortage, %s\n", 1113 s, __func__, 1114 V_tcp_sc_rst_sock_fail ? 1115 "sending RST" : "try again"); 1116 if (V_tcp_sc_rst_sock_fail) { 1117 rstreason = BANDLIM_UNLIMITED; 1118 goto dropwithreset; 1119 } else 1120 goto dropunlock; 1121 } 1122 /* 1123 * Socket is created in state SYN_RECEIVED. 1124 * Unlock the listen socket, lock the newly 1125 * created socket and update the tp variable. 1126 */ 1127 INP_WUNLOCK(inp); /* listen socket */ 1128 inp = sotoinpcb(so); 1129 /* 1130 * New connection inpcb is already locked by 1131 * syncache_expand(). 1132 */ 1133 INP_WLOCK_ASSERT(inp); 1134 tp = intotcpcb(inp); 1135 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1136 ("%s: ", __func__)); 1137 /* 1138 * Process the segment and the data it 1139 * contains. tcp_do_segment() consumes 1140 * the mbuf chain and unlocks the inpcb. 1141 */ 1142 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1143 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1144 iptos); 1145 return (IPPROTO_DONE); 1146 } 1147 /* 1148 * Segment flag validation for new connection attempts: 1149 * 1150 * Our (SYN|ACK) response was rejected. 1151 * Check with syncache and remove entry to prevent 1152 * retransmits. 1153 * 1154 * NB: syncache_chkrst does its own logging of failure 1155 * causes. 1156 */ 1157 if (thflags & TH_RST) { 1158 syncache_chkrst(&inc, th, m); 1159 goto dropunlock; 1160 } 1161 /* 1162 * We can't do anything without SYN. 1163 */ 1164 if ((thflags & TH_SYN) == 0) { 1165 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1166 log(LOG_DEBUG, "%s; %s: Listen socket: " 1167 "SYN is missing, segment ignored\n", 1168 s, __func__); 1169 TCPSTAT_INC(tcps_badsyn); 1170 goto dropunlock; 1171 } 1172 /* 1173 * (SYN|ACK) is bogus on a listen socket. 1174 */ 1175 if (thflags & TH_ACK) { 1176 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1177 log(LOG_DEBUG, "%s; %s: Listen socket: " 1178 "SYN|ACK invalid, segment rejected\n", 1179 s, __func__); 1180 syncache_badack(&inc); /* XXX: Not needed! */ 1181 TCPSTAT_INC(tcps_badsyn); 1182 rstreason = BANDLIM_RST_OPENPORT; 1183 goto dropwithreset; 1184 } 1185 /* 1186 * If the drop_synfin option is enabled, drop all 1187 * segments with both the SYN and FIN bits set. 1188 * This prevents e.g. nmap from identifying the 1189 * TCP/IP stack. 1190 * XXX: Poor reasoning. nmap has other methods 1191 * and is constantly refining its stack detection 1192 * strategies. 1193 * XXX: This is a violation of the TCP specification 1194 * and was used by RFC1644. 1195 */ 1196 if ((thflags & TH_FIN) && V_drop_synfin) { 1197 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1198 log(LOG_DEBUG, "%s; %s: Listen socket: " 1199 "SYN|FIN segment ignored (based on " 1200 "sysctl setting)\n", s, __func__); 1201 TCPSTAT_INC(tcps_badsyn); 1202 goto dropunlock; 1203 } 1204 /* 1205 * Segment's flags are (SYN) or (SYN|FIN). 1206 * 1207 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1208 * as they do not affect the state of the TCP FSM. 1209 * The data pointed to by TH_URG and th_urp is ignored. 1210 */ 1211 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1212 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1213 KASSERT(thflags & (TH_SYN), 1214 ("%s: Listen socket: TH_SYN not set", __func__)); 1215 #ifdef INET6 1216 /* 1217 * If deprecated address is forbidden, 1218 * we do not accept SYN to deprecated interface 1219 * address to prevent any new inbound connection from 1220 * getting established. 1221 * When we do not accept SYN, we send a TCP RST, 1222 * with deprecated source address (instead of dropping 1223 * it). We compromise it as it is much better for peer 1224 * to send a RST, and RST will be the final packet 1225 * for the exchange. 1226 * 1227 * If we do not forbid deprecated addresses, we accept 1228 * the SYN packet. RFC2462 does not suggest dropping 1229 * SYN in this case. 1230 * If we decipher RFC2462 5.5.4, it says like this: 1231 * 1. use of deprecated addr with existing 1232 * communication is okay - "SHOULD continue to be 1233 * used" 1234 * 2. use of it with new communication: 1235 * (2a) "SHOULD NOT be used if alternate address 1236 * with sufficient scope is available" 1237 * (2b) nothing mentioned otherwise. 1238 * Here we fall into (2b) case as we have no choice in 1239 * our source address selection - we must obey the peer. 1240 * 1241 * The wording in RFC2462 is confusing, and there are 1242 * multiple description text for deprecated address 1243 * handling - worse, they are not exactly the same. 1244 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1245 */ 1246 if (isipv6 && !V_ip6_use_deprecated) { 1247 struct in6_ifaddr *ia6; 1248 1249 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1250 if (ia6 != NULL && 1251 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1252 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1253 log(LOG_DEBUG, "%s; %s: Listen socket: " 1254 "Connection attempt to deprecated " 1255 "IPv6 address rejected\n", 1256 s, __func__); 1257 rstreason = BANDLIM_RST_OPENPORT; 1258 goto dropwithreset; 1259 } 1260 } 1261 #endif /* INET6 */ 1262 /* 1263 * Basic sanity checks on incoming SYN requests: 1264 * Don't respond if the destination is a link layer 1265 * broadcast according to RFC1122 4.2.3.10, p. 104. 1266 * If it is from this socket it must be forged. 1267 * Don't respond if the source or destination is a 1268 * global or subnet broad- or multicast address. 1269 * Note that it is quite possible to receive unicast 1270 * link-layer packets with a broadcast IP address. Use 1271 * in_broadcast() to find them. 1272 */ 1273 if (m->m_flags & (M_BCAST|M_MCAST)) { 1274 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1275 log(LOG_DEBUG, "%s; %s: Listen socket: " 1276 "Connection attempt from broad- or multicast " 1277 "link layer address ignored\n", s, __func__); 1278 goto dropunlock; 1279 } 1280 #ifdef INET6 1281 if (isipv6) { 1282 if (th->th_dport == th->th_sport && 1283 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1284 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1285 log(LOG_DEBUG, "%s; %s: Listen socket: " 1286 "Connection attempt to/from self " 1287 "ignored\n", s, __func__); 1288 goto dropunlock; 1289 } 1290 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1291 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1292 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1293 log(LOG_DEBUG, "%s; %s: Listen socket: " 1294 "Connection attempt from/to multicast " 1295 "address ignored\n", s, __func__); 1296 goto dropunlock; 1297 } 1298 } 1299 #endif 1300 #if defined(INET) && defined(INET6) 1301 else 1302 #endif 1303 #ifdef INET 1304 { 1305 if (th->th_dport == th->th_sport && 1306 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1307 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1308 log(LOG_DEBUG, "%s; %s: Listen socket: " 1309 "Connection attempt from/to self " 1310 "ignored\n", s, __func__); 1311 goto dropunlock; 1312 } 1313 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1314 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1315 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1316 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1318 log(LOG_DEBUG, "%s; %s: Listen socket: " 1319 "Connection attempt from/to broad- " 1320 "or multicast address ignored\n", 1321 s, __func__); 1322 goto dropunlock; 1323 } 1324 } 1325 #endif 1326 /* 1327 * SYN appears to be valid. Create compressed TCP state 1328 * for syncache. 1329 */ 1330 #ifdef TCPDEBUG 1331 if (so->so_options & SO_DEBUG) 1332 tcp_trace(TA_INPUT, ostate, tp, 1333 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1334 #endif 1335 TCP_PROBE3(debug__input, tp, th, m); 1336 tcp_dooptions(&to, optp, optlen, TO_SYN); 1337 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos)) 1338 goto tfo_socket_result; 1339 1340 /* 1341 * Entry added to syncache and mbuf consumed. 1342 * Only the listen socket is unlocked by syncache_add(). 1343 */ 1344 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1345 return (IPPROTO_DONE); 1346 } else if (tp->t_state == TCPS_LISTEN) { 1347 /* 1348 * When a listen socket is torn down the SO_ACCEPTCONN 1349 * flag is removed first while connections are drained 1350 * from the accept queue in a unlock/lock cycle of the 1351 * ACCEPT_LOCK, opening a race condition allowing a SYN 1352 * attempt go through unhandled. 1353 */ 1354 goto dropunlock; 1355 } 1356 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1357 if (tp->t_flags & TF_SIGNATURE) { 1358 tcp_dooptions(&to, optp, optlen, thflags); 1359 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1360 TCPSTAT_INC(tcps_sig_err_nosigopt); 1361 goto dropunlock; 1362 } 1363 if (!TCPMD5_ENABLED() || 1364 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1365 goto dropunlock; 1366 } 1367 #endif 1368 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1369 1370 /* 1371 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1372 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1373 * the inpcb, and unlocks pcbinfo. 1374 */ 1375 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1376 return (IPPROTO_DONE); 1377 1378 dropwithreset: 1379 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1380 1381 if (inp != NULL) { 1382 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1383 INP_WUNLOCK(inp); 1384 } else 1385 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1386 m = NULL; /* mbuf chain got consumed. */ 1387 goto drop; 1388 1389 dropunlock: 1390 if (m != NULL) 1391 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1392 1393 if (inp != NULL) 1394 INP_WUNLOCK(inp); 1395 1396 drop: 1397 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1398 if (s != NULL) 1399 free(s, M_TCPLOG); 1400 if (m != NULL) 1401 m_freem(m); 1402 return (IPPROTO_DONE); 1403 } 1404 1405 /* 1406 * Automatic sizing of receive socket buffer. Often the send 1407 * buffer size is not optimally adjusted to the actual network 1408 * conditions at hand (delay bandwidth product). Setting the 1409 * buffer size too small limits throughput on links with high 1410 * bandwidth and high delay (eg. trans-continental/oceanic links). 1411 * 1412 * On the receive side the socket buffer memory is only rarely 1413 * used to any significant extent. This allows us to be much 1414 * more aggressive in scaling the receive socket buffer. For 1415 * the case that the buffer space is actually used to a large 1416 * extent and we run out of kernel memory we can simply drop 1417 * the new segments; TCP on the sender will just retransmit it 1418 * later. Setting the buffer size too big may only consume too 1419 * much kernel memory if the application doesn't read() from 1420 * the socket or packet loss or reordering makes use of the 1421 * reassembly queue. 1422 * 1423 * The criteria to step up the receive buffer one notch are: 1424 * 1. Application has not set receive buffer size with 1425 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1426 * 2. the number of bytes received during 1/2 of an sRTT 1427 * is at least 3/8 of the current socket buffer size. 1428 * 3. receive buffer size has not hit maximal automatic size; 1429 * 1430 * If all of the criteria are met we increaset the socket buffer 1431 * by a 1/2 (bounded by the max). This allows us to keep ahead 1432 * of slow-start but also makes it so our peer never gets limited 1433 * by our rwnd which we then open up causing a burst. 1434 * 1435 * This algorithm does two steps per RTT at most and only if 1436 * we receive a bulk stream w/o packet losses or reorderings. 1437 * Shrinking the buffer during idle times is not necessary as 1438 * it doesn't consume any memory when idle. 1439 * 1440 * TODO: Only step up if the application is actually serving 1441 * the buffer to better manage the socket buffer resources. 1442 */ 1443 int 1444 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1445 struct tcpcb *tp, int tlen) 1446 { 1447 int newsize = 0; 1448 1449 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1450 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1451 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1452 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1453 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1454 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1455 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1456 } 1457 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1458 1459 /* Start over with next RTT. */ 1460 tp->rfbuf_ts = 0; 1461 tp->rfbuf_cnt = 0; 1462 } else { 1463 tp->rfbuf_cnt += tlen; /* add up */ 1464 } 1465 return (newsize); 1466 } 1467 1468 void 1469 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1470 { 1471 /* 1472 * Since tp might be gone if the session entered 1473 * the TIME_WAIT state before coming here, we need 1474 * to check if the socket is still connected. 1475 */ 1476 if ((so->so_state & SS_ISCONNECTED) == 0) 1477 return; 1478 INP_LOCK_ASSERT(tp->t_inpcb); 1479 if (tp->t_flags & TF_WAKESOR) { 1480 tp->t_flags &= ~TF_WAKESOR; 1481 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 1482 sorwakeup(so); 1483 } 1484 if (tp->t_flags & TF_WAKESOW) { 1485 tp->t_flags &= ~TF_WAKESOW; 1486 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 1487 sowwakeup(so); 1488 } 1489 } 1490 1491 void 1492 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1493 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1494 { 1495 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1496 int rstreason, todrop, win, incforsyn = 0; 1497 uint32_t tiwin; 1498 uint16_t nsegs; 1499 char *s; 1500 struct in_conninfo *inc; 1501 struct mbuf *mfree; 1502 struct tcpopt to; 1503 int tfo_syn; 1504 1505 #ifdef TCPDEBUG 1506 /* 1507 * The size of tcp_saveipgen must be the size of the max ip header, 1508 * now IPv6. 1509 */ 1510 u_char tcp_saveipgen[IP6_HDR_LEN]; 1511 struct tcphdr tcp_savetcp; 1512 short ostate = 0; 1513 #endif 1514 thflags = th->th_flags; 1515 inc = &tp->t_inpcb->inp_inc; 1516 tp->sackhint.last_sack_ack = 0; 1517 sack_changed = 0; 1518 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1519 1520 NET_EPOCH_ASSERT(); 1521 INP_WLOCK_ASSERT(tp->t_inpcb); 1522 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1523 __func__)); 1524 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1525 __func__)); 1526 1527 #ifdef TCPPCAP 1528 /* Save segment, if requested. */ 1529 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1530 #endif 1531 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1532 tlen, NULL, true); 1533 1534 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1535 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1536 log(LOG_DEBUG, "%s; %s: " 1537 "SYN|FIN segment ignored (based on " 1538 "sysctl setting)\n", s, __func__); 1539 free(s, M_TCPLOG); 1540 } 1541 goto drop; 1542 } 1543 1544 /* 1545 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1546 * check SEQ.ACK first. 1547 */ 1548 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1549 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1550 rstreason = BANDLIM_UNLIMITED; 1551 goto dropwithreset; 1552 } 1553 1554 /* 1555 * Segment received on connection. 1556 * Reset idle time and keep-alive timer. 1557 * XXX: This should be done after segment 1558 * validation to ignore broken/spoofed segs. 1559 */ 1560 tp->t_rcvtime = ticks; 1561 1562 /* 1563 * Scale up the window into a 32-bit value. 1564 * For the SYN_SENT state the scale is zero. 1565 */ 1566 tiwin = th->th_win << tp->snd_scale; 1567 #ifdef STATS 1568 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1569 #endif 1570 1571 /* 1572 * TCP ECN processing. 1573 */ 1574 if (tp->t_flags2 & TF2_ECN_PERMIT) { 1575 if (thflags & TH_CWR) { 1576 tp->t_flags2 &= ~TF2_ECN_SND_ECE; 1577 tp->t_flags |= TF_ACKNOW; 1578 } 1579 switch (iptos & IPTOS_ECN_MASK) { 1580 case IPTOS_ECN_CE: 1581 tp->t_flags2 |= TF2_ECN_SND_ECE; 1582 TCPSTAT_INC(tcps_ecn_ce); 1583 break; 1584 case IPTOS_ECN_ECT0: 1585 TCPSTAT_INC(tcps_ecn_ect0); 1586 break; 1587 case IPTOS_ECN_ECT1: 1588 TCPSTAT_INC(tcps_ecn_ect1); 1589 break; 1590 } 1591 1592 /* Process a packet differently from RFC3168. */ 1593 cc_ecnpkt_handler(tp, th, iptos); 1594 1595 /* Congestion experienced. */ 1596 if (thflags & TH_ECE) { 1597 cc_cong_signal(tp, th, CC_ECN); 1598 } 1599 } 1600 1601 /* 1602 * Parse options on any incoming segment. 1603 */ 1604 tcp_dooptions(&to, (u_char *)(th + 1), 1605 (th->th_off << 2) - sizeof(struct tcphdr), 1606 (thflags & TH_SYN) ? TO_SYN : 0); 1607 1608 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1609 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1610 (to.to_flags & TOF_SIGNATURE) == 0) { 1611 TCPSTAT_INC(tcps_sig_err_sigopt); 1612 /* XXX: should drop? */ 1613 } 1614 #endif 1615 /* 1616 * If echoed timestamp is later than the current time, 1617 * fall back to non RFC1323 RTT calculation. Normalize 1618 * timestamp if syncookies were used when this connection 1619 * was established. 1620 */ 1621 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1622 to.to_tsecr -= tp->ts_offset; 1623 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1624 to.to_tsecr = 0; 1625 else if (tp->t_flags & TF_PREVVALID && 1626 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1627 cc_cong_signal(tp, th, CC_RTO_ERR); 1628 } 1629 /* 1630 * Process options only when we get SYN/ACK back. The SYN case 1631 * for incoming connections is handled in tcp_syncache. 1632 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1633 * or <SYN,ACK>) segment itself is never scaled. 1634 * XXX this is traditional behavior, may need to be cleaned up. 1635 */ 1636 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1637 /* Handle parallel SYN for ECN */ 1638 if (!(thflags & TH_ACK) && 1639 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) && 1640 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) { 1641 tp->t_flags2 |= TF2_ECN_PERMIT; 1642 tp->t_flags2 |= TF2_ECN_SND_ECE; 1643 TCPSTAT_INC(tcps_ecn_shs); 1644 } 1645 if ((to.to_flags & TOF_SCALE) && 1646 (tp->t_flags & TF_REQ_SCALE)) { 1647 tp->t_flags |= TF_RCVD_SCALE; 1648 tp->snd_scale = to.to_wscale; 1649 } else 1650 tp->t_flags &= ~TF_REQ_SCALE; 1651 /* 1652 * Initial send window. It will be updated with 1653 * the next incoming segment to the scaled value. 1654 */ 1655 tp->snd_wnd = th->th_win; 1656 if ((to.to_flags & TOF_TS) && 1657 (tp->t_flags & TF_REQ_TSTMP)) { 1658 tp->t_flags |= TF_RCVD_TSTMP; 1659 tp->ts_recent = to.to_tsval; 1660 tp->ts_recent_age = tcp_ts_getticks(); 1661 } else 1662 tp->t_flags &= ~TF_REQ_TSTMP; 1663 if (to.to_flags & TOF_MSS) 1664 tcp_mss(tp, to.to_mss); 1665 if ((tp->t_flags & TF_SACK_PERMIT) && 1666 (to.to_flags & TOF_SACKPERM) == 0) 1667 tp->t_flags &= ~TF_SACK_PERMIT; 1668 if (IS_FASTOPEN(tp->t_flags)) { 1669 if (to.to_flags & TOF_FASTOPEN) { 1670 uint16_t mss; 1671 1672 if (to.to_flags & TOF_MSS) 1673 mss = to.to_mss; 1674 else 1675 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1676 mss = TCP6_MSS; 1677 else 1678 mss = TCP_MSS; 1679 tcp_fastopen_update_cache(tp, mss, 1680 to.to_tfo_len, to.to_tfo_cookie); 1681 } else 1682 tcp_fastopen_disable_path(tp); 1683 } 1684 } 1685 1686 /* 1687 * If timestamps were negotiated during SYN/ACK and a 1688 * segment without a timestamp is received, silently drop 1689 * the segment, unless it is a RST segment or missing timestamps are 1690 * tolerated. 1691 * See section 3.2 of RFC 7323. 1692 */ 1693 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1694 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1695 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1696 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1697 "segment processed normally\n", 1698 s, __func__); 1699 free(s, M_TCPLOG); 1700 } 1701 } else { 1702 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1703 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1704 "segment silently dropped\n", s, __func__); 1705 free(s, M_TCPLOG); 1706 } 1707 goto drop; 1708 } 1709 } 1710 /* 1711 * If timestamps were not negotiated during SYN/ACK and a 1712 * segment with a timestamp is received, ignore the 1713 * timestamp and process the packet normally. 1714 * See section 3.2 of RFC 7323. 1715 */ 1716 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1717 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1718 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1719 "segment processed normally\n", s, __func__); 1720 free(s, M_TCPLOG); 1721 } 1722 } 1723 1724 /* 1725 * Header prediction: check for the two common cases 1726 * of a uni-directional data xfer. If the packet has 1727 * no control flags, is in-sequence, the window didn't 1728 * change and we're not retransmitting, it's a 1729 * candidate. If the length is zero and the ack moved 1730 * forward, we're the sender side of the xfer. Just 1731 * free the data acked & wake any higher level process 1732 * that was blocked waiting for space. If the length 1733 * is non-zero and the ack didn't move, we're the 1734 * receiver side. If we're getting packets in-order 1735 * (the reassembly queue is empty), add the data to 1736 * the socket buffer and note that we need a delayed ack. 1737 * Make sure that the hidden state-flags are also off. 1738 * Since we check for TCPS_ESTABLISHED first, it can only 1739 * be TH_NEEDSYN. 1740 */ 1741 if (tp->t_state == TCPS_ESTABLISHED && 1742 th->th_seq == tp->rcv_nxt && 1743 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1744 tp->snd_nxt == tp->snd_max && 1745 tiwin && tiwin == tp->snd_wnd && 1746 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1747 SEGQ_EMPTY(tp) && 1748 ((to.to_flags & TOF_TS) == 0 || 1749 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1750 /* 1751 * If last ACK falls within this segment's sequence numbers, 1752 * record the timestamp. 1753 * NOTE that the test is modified according to the latest 1754 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1755 */ 1756 if ((to.to_flags & TOF_TS) != 0 && 1757 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1758 tp->ts_recent_age = tcp_ts_getticks(); 1759 tp->ts_recent = to.to_tsval; 1760 } 1761 1762 if (tlen == 0) { 1763 if (SEQ_GT(th->th_ack, tp->snd_una) && 1764 SEQ_LEQ(th->th_ack, tp->snd_max) && 1765 !IN_RECOVERY(tp->t_flags) && 1766 (to.to_flags & TOF_SACK) == 0 && 1767 TAILQ_EMPTY(&tp->snd_holes)) { 1768 /* 1769 * This is a pure ack for outstanding data. 1770 */ 1771 TCPSTAT_INC(tcps_predack); 1772 1773 /* 1774 * "bad retransmit" recovery without timestamps. 1775 */ 1776 if ((to.to_flags & TOF_TS) == 0 && 1777 tp->t_rxtshift == 1 && 1778 tp->t_flags & TF_PREVVALID && 1779 (int)(ticks - tp->t_badrxtwin) < 0) { 1780 cc_cong_signal(tp, th, CC_RTO_ERR); 1781 } 1782 1783 /* 1784 * Recalculate the transmit timer / rtt. 1785 * 1786 * Some boxes send broken timestamp replies 1787 * during the SYN+ACK phase, ignore 1788 * timestamps of 0 or we could calculate a 1789 * huge RTT and blow up the retransmit timer. 1790 */ 1791 if ((to.to_flags & TOF_TS) != 0 && 1792 to.to_tsecr) { 1793 uint32_t t; 1794 1795 t = tcp_ts_getticks() - to.to_tsecr; 1796 if (!tp->t_rttlow || tp->t_rttlow > t) 1797 tp->t_rttlow = t; 1798 tcp_xmit_timer(tp, 1799 TCP_TS_TO_TICKS(t) + 1); 1800 } else if (tp->t_rtttime && 1801 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1802 if (!tp->t_rttlow || 1803 tp->t_rttlow > ticks - tp->t_rtttime) 1804 tp->t_rttlow = ticks - tp->t_rtttime; 1805 tcp_xmit_timer(tp, 1806 ticks - tp->t_rtttime); 1807 } 1808 acked = BYTES_THIS_ACK(tp, th); 1809 1810 #ifdef TCP_HHOOK 1811 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1812 hhook_run_tcp_est_in(tp, th, &to); 1813 #endif 1814 1815 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1816 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1817 sbdrop(&so->so_snd, acked); 1818 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1819 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1820 tp->snd_recover = th->th_ack - 1; 1821 1822 /* 1823 * Let the congestion control algorithm update 1824 * congestion control related information. This 1825 * typically means increasing the congestion 1826 * window. 1827 */ 1828 cc_ack_received(tp, th, nsegs, CC_ACK); 1829 1830 tp->snd_una = th->th_ack; 1831 /* 1832 * Pull snd_wl2 up to prevent seq wrap relative 1833 * to th_ack. 1834 */ 1835 tp->snd_wl2 = th->th_ack; 1836 tp->t_dupacks = 0; 1837 m_freem(m); 1838 1839 /* 1840 * If all outstanding data are acked, stop 1841 * retransmit timer, otherwise restart timer 1842 * using current (possibly backed-off) value. 1843 * If process is waiting for space, 1844 * wakeup/selwakeup/signal. If data 1845 * are ready to send, let tcp_output 1846 * decide between more output or persist. 1847 */ 1848 #ifdef TCPDEBUG 1849 if (so->so_options & SO_DEBUG) 1850 tcp_trace(TA_INPUT, ostate, tp, 1851 (void *)tcp_saveipgen, 1852 &tcp_savetcp, 0); 1853 #endif 1854 TCP_PROBE3(debug__input, tp, th, m); 1855 if (tp->snd_una == tp->snd_max) 1856 tcp_timer_activate(tp, TT_REXMT, 0); 1857 else if (!tcp_timer_active(tp, TT_PERSIST)) 1858 tcp_timer_activate(tp, TT_REXMT, 1859 tp->t_rxtcur); 1860 tp->t_flags |= TF_WAKESOW; 1861 if (sbavail(&so->so_snd)) 1862 (void) tp->t_fb->tfb_tcp_output(tp); 1863 goto check_delack; 1864 } 1865 } else if (th->th_ack == tp->snd_una && 1866 tlen <= sbspace(&so->so_rcv)) { 1867 int newsize = 0; /* automatic sockbuf scaling */ 1868 1869 /* 1870 * This is a pure, in-sequence data packet with 1871 * nothing on the reassembly queue and we have enough 1872 * buffer space to take it. 1873 */ 1874 /* Clean receiver SACK report if present */ 1875 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1876 tcp_clean_sackreport(tp); 1877 TCPSTAT_INC(tcps_preddat); 1878 tp->rcv_nxt += tlen; 1879 if (tlen && 1880 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1881 (tp->t_fbyte_in == 0)) { 1882 tp->t_fbyte_in = ticks; 1883 if (tp->t_fbyte_in == 0) 1884 tp->t_fbyte_in = 1; 1885 if (tp->t_fbyte_out && tp->t_fbyte_in) 1886 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1887 } 1888 /* 1889 * Pull snd_wl1 up to prevent seq wrap relative to 1890 * th_seq. 1891 */ 1892 tp->snd_wl1 = th->th_seq; 1893 /* 1894 * Pull rcv_up up to prevent seq wrap relative to 1895 * rcv_nxt. 1896 */ 1897 tp->rcv_up = tp->rcv_nxt; 1898 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1899 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1900 #ifdef TCPDEBUG 1901 if (so->so_options & SO_DEBUG) 1902 tcp_trace(TA_INPUT, ostate, tp, 1903 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1904 #endif 1905 TCP_PROBE3(debug__input, tp, th, m); 1906 1907 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1908 1909 /* Add data to socket buffer. */ 1910 SOCKBUF_LOCK(&so->so_rcv); 1911 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1912 m_freem(m); 1913 } else { 1914 /* 1915 * Set new socket buffer size. 1916 * Give up when limit is reached. 1917 */ 1918 if (newsize) 1919 if (!sbreserve_locked(&so->so_rcv, 1920 newsize, so, NULL)) 1921 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1922 m_adj(m, drop_hdrlen); /* delayed header drop */ 1923 sbappendstream_locked(&so->so_rcv, m, 0); 1924 } 1925 SOCKBUF_UNLOCK(&so->so_rcv); 1926 tp->t_flags |= TF_WAKESOR; 1927 if (DELAY_ACK(tp, tlen)) { 1928 tp->t_flags |= TF_DELACK; 1929 } else { 1930 tp->t_flags |= TF_ACKNOW; 1931 tp->t_fb->tfb_tcp_output(tp); 1932 } 1933 goto check_delack; 1934 } 1935 } 1936 1937 /* 1938 * Calculate amount of space in receive window, 1939 * and then do TCP input processing. 1940 * Receive window is amount of space in rcv queue, 1941 * but not less than advertised window. 1942 */ 1943 win = sbspace(&so->so_rcv); 1944 if (win < 0) 1945 win = 0; 1946 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1947 1948 switch (tp->t_state) { 1949 /* 1950 * If the state is SYN_RECEIVED: 1951 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1952 */ 1953 case TCPS_SYN_RECEIVED: 1954 if ((thflags & TH_ACK) && 1955 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1956 SEQ_GT(th->th_ack, tp->snd_max))) { 1957 rstreason = BANDLIM_RST_OPENPORT; 1958 goto dropwithreset; 1959 } 1960 if (IS_FASTOPEN(tp->t_flags)) { 1961 /* 1962 * When a TFO connection is in SYN_RECEIVED, the 1963 * only valid packets are the initial SYN, a 1964 * retransmit/copy of the initial SYN (possibly with 1965 * a subset of the original data), a valid ACK, a 1966 * FIN, or a RST. 1967 */ 1968 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1969 rstreason = BANDLIM_RST_OPENPORT; 1970 goto dropwithreset; 1971 } else if (thflags & TH_SYN) { 1972 /* non-initial SYN is ignored */ 1973 if ((tcp_timer_active(tp, TT_DELACK) || 1974 tcp_timer_active(tp, TT_REXMT))) 1975 goto drop; 1976 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1977 goto drop; 1978 } 1979 } 1980 break; 1981 1982 /* 1983 * If the state is SYN_SENT: 1984 * if seg contains a RST with valid ACK (SEQ.ACK has already 1985 * been verified), then drop the connection. 1986 * if seg contains a RST without an ACK, drop the seg. 1987 * if seg does not contain SYN, then drop the seg. 1988 * Otherwise this is an acceptable SYN segment 1989 * initialize tp->rcv_nxt and tp->irs 1990 * if seg contains ack then advance tp->snd_una 1991 * if seg contains an ECE and ECN support is enabled, the stream 1992 * is ECN capable. 1993 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1994 * arrange for segment to be acked (eventually) 1995 * continue processing rest of data/controls, beginning with URG 1996 */ 1997 case TCPS_SYN_SENT: 1998 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1999 TCP_PROBE5(connect__refused, NULL, tp, 2000 m, tp, th); 2001 tp = tcp_drop(tp, ECONNREFUSED); 2002 } 2003 if (thflags & TH_RST) 2004 goto drop; 2005 if (!(thflags & TH_SYN)) 2006 goto drop; 2007 2008 tp->irs = th->th_seq; 2009 tcp_rcvseqinit(tp); 2010 if (thflags & TH_ACK) { 2011 int tfo_partial_ack = 0; 2012 2013 TCPSTAT_INC(tcps_connects); 2014 soisconnected(so); 2015 #ifdef MAC 2016 mac_socketpeer_set_from_mbuf(m, so); 2017 #endif 2018 /* Do window scaling on this connection? */ 2019 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2020 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2021 tp->rcv_scale = tp->request_r_scale; 2022 } 2023 tp->rcv_adv += min(tp->rcv_wnd, 2024 TCP_MAXWIN << tp->rcv_scale); 2025 tp->snd_una++; /* SYN is acked */ 2026 /* 2027 * If not all the data that was sent in the TFO SYN 2028 * has been acked, resend the remainder right away. 2029 */ 2030 if (IS_FASTOPEN(tp->t_flags) && 2031 (tp->snd_una != tp->snd_max)) { 2032 tp->snd_nxt = th->th_ack; 2033 tfo_partial_ack = 1; 2034 } 2035 /* 2036 * If there's data, delay ACK; if there's also a FIN 2037 * ACKNOW will be turned on later. 2038 */ 2039 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2040 tcp_timer_activate(tp, TT_DELACK, 2041 tcp_delacktime); 2042 else 2043 tp->t_flags |= TF_ACKNOW; 2044 2045 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2046 (V_tcp_do_ecn == 1)) { 2047 tp->t_flags2 |= TF2_ECN_PERMIT; 2048 TCPSTAT_INC(tcps_ecn_shs); 2049 } 2050 2051 /* 2052 * Received <SYN,ACK> in SYN_SENT[*] state. 2053 * Transitions: 2054 * SYN_SENT --> ESTABLISHED 2055 * SYN_SENT* --> FIN_WAIT_1 2056 */ 2057 tp->t_starttime = ticks; 2058 if (tp->t_flags & TF_NEEDFIN) { 2059 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2060 tp->t_flags &= ~TF_NEEDFIN; 2061 thflags &= ~TH_SYN; 2062 } else { 2063 tcp_state_change(tp, TCPS_ESTABLISHED); 2064 TCP_PROBE5(connect__established, NULL, tp, 2065 m, tp, th); 2066 cc_conn_init(tp); 2067 tcp_timer_activate(tp, TT_KEEP, 2068 TP_KEEPIDLE(tp)); 2069 } 2070 } else { 2071 /* 2072 * Received initial SYN in SYN-SENT[*] state => 2073 * simultaneous open. 2074 * If it succeeds, connection is * half-synchronized. 2075 * Otherwise, do 3-way handshake: 2076 * SYN-SENT -> SYN-RECEIVED 2077 * SYN-SENT* -> SYN-RECEIVED* 2078 */ 2079 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2080 tcp_timer_activate(tp, TT_REXMT, 0); 2081 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2082 } 2083 2084 INP_WLOCK_ASSERT(tp->t_inpcb); 2085 2086 /* 2087 * Advance th->th_seq to correspond to first data byte. 2088 * If data, trim to stay within window, 2089 * dropping FIN if necessary. 2090 */ 2091 th->th_seq++; 2092 if (tlen > tp->rcv_wnd) { 2093 todrop = tlen - tp->rcv_wnd; 2094 m_adj(m, -todrop); 2095 tlen = tp->rcv_wnd; 2096 thflags &= ~TH_FIN; 2097 TCPSTAT_INC(tcps_rcvpackafterwin); 2098 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2099 } 2100 tp->snd_wl1 = th->th_seq - 1; 2101 tp->rcv_up = th->th_seq; 2102 /* 2103 * Client side of transaction: already sent SYN and data. 2104 * If the remote host used T/TCP to validate the SYN, 2105 * our data will be ACK'd; if so, enter normal data segment 2106 * processing in the middle of step 5, ack processing. 2107 * Otherwise, goto step 6. 2108 */ 2109 if (thflags & TH_ACK) 2110 goto process_ACK; 2111 2112 goto step6; 2113 2114 /* 2115 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2116 * do normal processing. 2117 * 2118 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2119 */ 2120 case TCPS_LAST_ACK: 2121 case TCPS_CLOSING: 2122 break; /* continue normal processing */ 2123 } 2124 2125 /* 2126 * States other than LISTEN or SYN_SENT. 2127 * First check the RST flag and sequence number since reset segments 2128 * are exempt from the timestamp and connection count tests. This 2129 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2130 * below which allowed reset segments in half the sequence space 2131 * to fall though and be processed (which gives forged reset 2132 * segments with a random sequence number a 50 percent chance of 2133 * killing a connection). 2134 * Then check timestamp, if present. 2135 * Then check the connection count, if present. 2136 * Then check that at least some bytes of segment are within 2137 * receive window. If segment begins before rcv_nxt, 2138 * drop leading data (and SYN); if nothing left, just ack. 2139 */ 2140 if (thflags & TH_RST) { 2141 /* 2142 * RFC5961 Section 3.2 2143 * 2144 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2145 * - If RST is in window, we send challenge ACK. 2146 * 2147 * Note: to take into account delayed ACKs, we should 2148 * test against last_ack_sent instead of rcv_nxt. 2149 * Note 2: we handle special case of closed window, not 2150 * covered by the RFC. 2151 */ 2152 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2153 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2154 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2155 KASSERT(tp->t_state != TCPS_SYN_SENT, 2156 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2157 __func__, th, tp)); 2158 2159 if (V_tcp_insecure_rst || 2160 tp->last_ack_sent == th->th_seq) { 2161 TCPSTAT_INC(tcps_drops); 2162 /* Drop the connection. */ 2163 switch (tp->t_state) { 2164 case TCPS_SYN_RECEIVED: 2165 so->so_error = ECONNREFUSED; 2166 goto close; 2167 case TCPS_ESTABLISHED: 2168 case TCPS_FIN_WAIT_1: 2169 case TCPS_FIN_WAIT_2: 2170 case TCPS_CLOSE_WAIT: 2171 case TCPS_CLOSING: 2172 case TCPS_LAST_ACK: 2173 so->so_error = ECONNRESET; 2174 close: 2175 /* FALLTHROUGH */ 2176 default: 2177 tp = tcp_close(tp); 2178 } 2179 } else { 2180 TCPSTAT_INC(tcps_badrst); 2181 /* Send challenge ACK. */ 2182 tcp_respond(tp, mtod(m, void *), th, m, 2183 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2184 tp->last_ack_sent = tp->rcv_nxt; 2185 m = NULL; 2186 } 2187 } 2188 goto drop; 2189 } 2190 2191 /* 2192 * RFC5961 Section 4.2 2193 * Send challenge ACK for any SYN in synchronized state. 2194 */ 2195 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2196 tp->t_state != TCPS_SYN_RECEIVED) { 2197 TCPSTAT_INC(tcps_badsyn); 2198 if (V_tcp_insecure_syn && 2199 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2200 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2201 tp = tcp_drop(tp, ECONNRESET); 2202 rstreason = BANDLIM_UNLIMITED; 2203 } else { 2204 /* Send challenge ACK. */ 2205 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2206 tp->snd_nxt, TH_ACK); 2207 tp->last_ack_sent = tp->rcv_nxt; 2208 m = NULL; 2209 } 2210 goto drop; 2211 } 2212 2213 /* 2214 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2215 * and it's less than ts_recent, drop it. 2216 */ 2217 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2218 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2219 /* Check to see if ts_recent is over 24 days old. */ 2220 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2221 /* 2222 * Invalidate ts_recent. If this segment updates 2223 * ts_recent, the age will be reset later and ts_recent 2224 * will get a valid value. If it does not, setting 2225 * ts_recent to zero will at least satisfy the 2226 * requirement that zero be placed in the timestamp 2227 * echo reply when ts_recent isn't valid. The 2228 * age isn't reset until we get a valid ts_recent 2229 * because we don't want out-of-order segments to be 2230 * dropped when ts_recent is old. 2231 */ 2232 tp->ts_recent = 0; 2233 } else { 2234 TCPSTAT_INC(tcps_rcvduppack); 2235 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2236 TCPSTAT_INC(tcps_pawsdrop); 2237 if (tlen) 2238 goto dropafterack; 2239 goto drop; 2240 } 2241 } 2242 2243 /* 2244 * In the SYN-RECEIVED state, validate that the packet belongs to 2245 * this connection before trimming the data to fit the receive 2246 * window. Check the sequence number versus IRS since we know 2247 * the sequence numbers haven't wrapped. This is a partial fix 2248 * for the "LAND" DoS attack. 2249 */ 2250 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2251 rstreason = BANDLIM_RST_OPENPORT; 2252 goto dropwithreset; 2253 } 2254 2255 todrop = tp->rcv_nxt - th->th_seq; 2256 if (todrop > 0) { 2257 if (thflags & TH_SYN) { 2258 thflags &= ~TH_SYN; 2259 th->th_seq++; 2260 if (th->th_urp > 1) 2261 th->th_urp--; 2262 else 2263 thflags &= ~TH_URG; 2264 todrop--; 2265 } 2266 /* 2267 * Following if statement from Stevens, vol. 2, p. 960. 2268 */ 2269 if (todrop > tlen 2270 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2271 /* 2272 * Any valid FIN must be to the left of the window. 2273 * At this point the FIN must be a duplicate or out 2274 * of sequence; drop it. 2275 */ 2276 thflags &= ~TH_FIN; 2277 2278 /* 2279 * Send an ACK to resynchronize and drop any data. 2280 * But keep on processing for RST or ACK. 2281 */ 2282 tp->t_flags |= TF_ACKNOW; 2283 todrop = tlen; 2284 TCPSTAT_INC(tcps_rcvduppack); 2285 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2286 } else { 2287 TCPSTAT_INC(tcps_rcvpartduppack); 2288 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2289 } 2290 /* 2291 * DSACK - add SACK block for dropped range 2292 */ 2293 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2294 tcp_update_sack_list(tp, th->th_seq, 2295 th->th_seq + todrop); 2296 /* 2297 * ACK now, as the next in-sequence segment 2298 * will clear the DSACK block again 2299 */ 2300 tp->t_flags |= TF_ACKNOW; 2301 } 2302 drop_hdrlen += todrop; /* drop from the top afterwards */ 2303 th->th_seq += todrop; 2304 tlen -= todrop; 2305 if (th->th_urp > todrop) 2306 th->th_urp -= todrop; 2307 else { 2308 thflags &= ~TH_URG; 2309 th->th_urp = 0; 2310 } 2311 } 2312 2313 /* 2314 * If new data are received on a connection after the 2315 * user processes are gone, then RST the other end. 2316 */ 2317 if ((so->so_state & SS_NOFDREF) && 2318 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2319 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2320 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2321 "after socket was closed, " 2322 "sending RST and removing tcpcb\n", 2323 s, __func__, tcpstates[tp->t_state], tlen); 2324 free(s, M_TCPLOG); 2325 } 2326 tp = tcp_close(tp); 2327 TCPSTAT_INC(tcps_rcvafterclose); 2328 rstreason = BANDLIM_UNLIMITED; 2329 goto dropwithreset; 2330 } 2331 2332 /* 2333 * If segment ends after window, drop trailing data 2334 * (and PUSH and FIN); if nothing left, just ACK. 2335 */ 2336 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2337 if (todrop > 0) { 2338 TCPSTAT_INC(tcps_rcvpackafterwin); 2339 if (todrop >= tlen) { 2340 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2341 /* 2342 * If window is closed can only take segments at 2343 * window edge, and have to drop data and PUSH from 2344 * incoming segments. Continue processing, but 2345 * remember to ack. Otherwise, drop segment 2346 * and ack. 2347 */ 2348 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2349 tp->t_flags |= TF_ACKNOW; 2350 TCPSTAT_INC(tcps_rcvwinprobe); 2351 } else 2352 goto dropafterack; 2353 } else 2354 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2355 m_adj(m, -todrop); 2356 tlen -= todrop; 2357 thflags &= ~(TH_PUSH|TH_FIN); 2358 } 2359 2360 /* 2361 * If last ACK falls within this segment's sequence numbers, 2362 * record its timestamp. 2363 * NOTE: 2364 * 1) That the test incorporates suggestions from the latest 2365 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2366 * 2) That updating only on newer timestamps interferes with 2367 * our earlier PAWS tests, so this check should be solely 2368 * predicated on the sequence space of this segment. 2369 * 3) That we modify the segment boundary check to be 2370 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2371 * instead of RFC1323's 2372 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2373 * This modified check allows us to overcome RFC1323's 2374 * limitations as described in Stevens TCP/IP Illustrated 2375 * Vol. 2 p.869. In such cases, we can still calculate the 2376 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2377 */ 2378 if ((to.to_flags & TOF_TS) != 0 && 2379 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2380 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2381 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2382 tp->ts_recent_age = tcp_ts_getticks(); 2383 tp->ts_recent = to.to_tsval; 2384 } 2385 2386 /* 2387 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2388 * flag is on (half-synchronized state), then queue data for 2389 * later processing; else drop segment and return. 2390 */ 2391 if ((thflags & TH_ACK) == 0) { 2392 if (tp->t_state == TCPS_SYN_RECEIVED || 2393 (tp->t_flags & TF_NEEDSYN)) { 2394 if (tp->t_state == TCPS_SYN_RECEIVED && 2395 IS_FASTOPEN(tp->t_flags)) { 2396 tp->snd_wnd = tiwin; 2397 cc_conn_init(tp); 2398 } 2399 goto step6; 2400 } else if (tp->t_flags & TF_ACKNOW) 2401 goto dropafterack; 2402 else 2403 goto drop; 2404 } 2405 2406 /* 2407 * Ack processing. 2408 */ 2409 switch (tp->t_state) { 2410 /* 2411 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2412 * ESTABLISHED state and continue processing. 2413 * The ACK was checked above. 2414 */ 2415 case TCPS_SYN_RECEIVED: 2416 2417 TCPSTAT_INC(tcps_connects); 2418 soisconnected(so); 2419 /* Do window scaling? */ 2420 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2421 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2422 tp->rcv_scale = tp->request_r_scale; 2423 } 2424 tp->snd_wnd = tiwin; 2425 /* 2426 * Make transitions: 2427 * SYN-RECEIVED -> ESTABLISHED 2428 * SYN-RECEIVED* -> FIN-WAIT-1 2429 */ 2430 tp->t_starttime = ticks; 2431 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2432 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2433 tp->t_tfo_pending = NULL; 2434 } 2435 if (tp->t_flags & TF_NEEDFIN) { 2436 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2437 tp->t_flags &= ~TF_NEEDFIN; 2438 } else { 2439 tcp_state_change(tp, TCPS_ESTABLISHED); 2440 TCP_PROBE5(accept__established, NULL, tp, 2441 m, tp, th); 2442 /* 2443 * TFO connections call cc_conn_init() during SYN 2444 * processing. Calling it again here for such 2445 * connections is not harmless as it would undo the 2446 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2447 * is retransmitted. 2448 */ 2449 if (!IS_FASTOPEN(tp->t_flags)) 2450 cc_conn_init(tp); 2451 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2452 } 2453 /* 2454 * Account for the ACK of our SYN prior to 2455 * regular ACK processing below, except for 2456 * simultaneous SYN, which is handled later. 2457 */ 2458 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2459 incforsyn = 1; 2460 /* 2461 * If segment contains data or ACK, will call tcp_reass() 2462 * later; if not, do so now to pass queued data to user. 2463 */ 2464 if (tlen == 0 && (thflags & TH_FIN) == 0) 2465 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2466 (struct mbuf *)0); 2467 tp->snd_wl1 = th->th_seq - 1; 2468 /* FALLTHROUGH */ 2469 2470 /* 2471 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2472 * ACKs. If the ack is in the range 2473 * tp->snd_una < th->th_ack <= tp->snd_max 2474 * then advance tp->snd_una to th->th_ack and drop 2475 * data from the retransmission queue. If this ACK reflects 2476 * more up to date window information we update our window information. 2477 */ 2478 case TCPS_ESTABLISHED: 2479 case TCPS_FIN_WAIT_1: 2480 case TCPS_FIN_WAIT_2: 2481 case TCPS_CLOSE_WAIT: 2482 case TCPS_CLOSING: 2483 case TCPS_LAST_ACK: 2484 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2485 TCPSTAT_INC(tcps_rcvacktoomuch); 2486 goto dropafterack; 2487 } 2488 if ((tp->t_flags & TF_SACK_PERMIT) && 2489 ((to.to_flags & TOF_SACK) || 2490 !TAILQ_EMPTY(&tp->snd_holes))) 2491 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2492 else 2493 /* 2494 * Reset the value so that previous (valid) value 2495 * from the last ack with SACK doesn't get used. 2496 */ 2497 tp->sackhint.sacked_bytes = 0; 2498 2499 #ifdef TCP_HHOOK 2500 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2501 hhook_run_tcp_est_in(tp, th, &to); 2502 #endif 2503 2504 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2505 u_int maxseg; 2506 2507 maxseg = tcp_maxseg(tp); 2508 if (tlen == 0 && 2509 (tiwin == tp->snd_wnd || 2510 (tp->t_flags & TF_SACK_PERMIT))) { 2511 /* 2512 * If this is the first time we've seen a 2513 * FIN from the remote, this is not a 2514 * duplicate and it needs to be processed 2515 * normally. This happens during a 2516 * simultaneous close. 2517 */ 2518 if ((thflags & TH_FIN) && 2519 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2520 tp->t_dupacks = 0; 2521 break; 2522 } 2523 TCPSTAT_INC(tcps_rcvdupack); 2524 /* 2525 * If we have outstanding data (other than 2526 * a window probe), this is a completely 2527 * duplicate ack (ie, window info didn't 2528 * change and FIN isn't set), 2529 * the ack is the biggest we've 2530 * seen and we've seen exactly our rexmt 2531 * threshold of them, assume a packet 2532 * has been dropped and retransmit it. 2533 * Kludge snd_nxt & the congestion 2534 * window so we send only this one 2535 * packet. 2536 * 2537 * We know we're losing at the current 2538 * window size so do congestion avoidance 2539 * (set ssthresh to half the current window 2540 * and pull our congestion window back to 2541 * the new ssthresh). 2542 * 2543 * Dup acks mean that packets have left the 2544 * network (they're now cached at the receiver) 2545 * so bump cwnd by the amount in the receiver 2546 * to keep a constant cwnd packets in the 2547 * network. 2548 * 2549 * When using TCP ECN, notify the peer that 2550 * we reduced the cwnd. 2551 */ 2552 /* 2553 * Following 2 kinds of acks should not affect 2554 * dupack counting: 2555 * 1) Old acks 2556 * 2) Acks with SACK but without any new SACK 2557 * information in them. These could result from 2558 * any anomaly in the network like a switch 2559 * duplicating packets or a possible DoS attack. 2560 */ 2561 if (th->th_ack != tp->snd_una || 2562 ((tp->t_flags & TF_SACK_PERMIT) && 2563 !sack_changed)) 2564 break; 2565 else if (!tcp_timer_active(tp, TT_REXMT)) 2566 tp->t_dupacks = 0; 2567 else if (++tp->t_dupacks > tcprexmtthresh || 2568 IN_FASTRECOVERY(tp->t_flags)) { 2569 cc_ack_received(tp, th, nsegs, 2570 CC_DUPACK); 2571 if (V_tcp_do_prr && 2572 IN_FASTRECOVERY(tp->t_flags) && 2573 (tp->t_flags & TF_SACK_PERMIT)) { 2574 long snd_cnt = 0, limit = 0; 2575 long del_data = 0, pipe = 0; 2576 /* 2577 * In a duplicate ACK del_data is only the 2578 * diff_in_sack. If no SACK is used del_data 2579 * will be 0. Pipe is the amount of data we 2580 * estimate to be in the network. 2581 */ 2582 del_data = tp->sackhint.delivered_data; 2583 pipe = (tp->snd_nxt - tp->snd_fack) + 2584 tp->sackhint.sack_bytes_rexmit; 2585 tp->sackhint.prr_delivered += del_data; 2586 if (pipe > tp->snd_ssthresh) { 2587 if (tp->sackhint.recover_fs == 0) 2588 tp->sackhint.recover_fs = 2589 max(1, tp->snd_nxt - tp->snd_una); 2590 snd_cnt = (tp->sackhint.prr_delivered * 2591 tp->snd_ssthresh / 2592 tp->sackhint.recover_fs) + 2593 1 - tp->sackhint.sack_bytes_rexmit; 2594 } else { 2595 if (V_tcp_do_prr_conservative) 2596 limit = tp->sackhint.prr_delivered - 2597 tp->sackhint.sack_bytes_rexmit; 2598 else 2599 if ((tp->sackhint.prr_delivered - 2600 tp->sackhint.sack_bytes_rexmit) > 2601 del_data) 2602 limit = tp->sackhint.prr_delivered - 2603 tp->sackhint.sack_bytes_rexmit + 2604 maxseg; 2605 else 2606 limit = del_data + maxseg; 2607 if ((tp->snd_ssthresh - pipe) < limit) 2608 snd_cnt = tp->snd_ssthresh - pipe; 2609 else 2610 snd_cnt = limit; 2611 } 2612 snd_cnt = max((snd_cnt / maxseg), 0); 2613 /* 2614 * Send snd_cnt new data into the network in 2615 * response to this ACK. If there is a going 2616 * to be a SACK retransmission, adjust snd_cwnd 2617 * accordingly. 2618 */ 2619 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 2620 tp->sackhint.sack_bytes_rexmit + 2621 (snd_cnt * maxseg); 2622 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2623 IN_FASTRECOVERY(tp->t_flags)) { 2624 int awnd; 2625 2626 /* 2627 * Compute the amount of data in flight first. 2628 * We can inject new data into the pipe iff 2629 * we have less than 1/2 the original window's 2630 * worth of data in flight. 2631 */ 2632 if (V_tcp_do_rfc6675_pipe) 2633 awnd = tcp_compute_pipe(tp); 2634 else 2635 awnd = (tp->snd_nxt - tp->snd_fack) + 2636 tp->sackhint.sack_bytes_rexmit; 2637 2638 if (awnd < tp->snd_ssthresh) { 2639 tp->snd_cwnd += maxseg; 2640 if (tp->snd_cwnd > tp->snd_ssthresh) 2641 tp->snd_cwnd = tp->snd_ssthresh; 2642 } 2643 } else 2644 tp->snd_cwnd += maxseg; 2645 (void) tp->t_fb->tfb_tcp_output(tp); 2646 goto drop; 2647 } else if (tp->t_dupacks == tcprexmtthresh) { 2648 tcp_seq onxt = tp->snd_nxt; 2649 2650 /* 2651 * If we're doing sack, or prr, check 2652 * to see if we're already in sack 2653 * recovery. If we're not doing sack, 2654 * check to see if we're in newreno 2655 * recovery. 2656 */ 2657 if (V_tcp_do_prr || 2658 (tp->t_flags & TF_SACK_PERMIT)) { 2659 if (IN_FASTRECOVERY(tp->t_flags)) { 2660 tp->t_dupacks = 0; 2661 break; 2662 } 2663 } else { 2664 if (SEQ_LEQ(th->th_ack, 2665 tp->snd_recover)) { 2666 tp->t_dupacks = 0; 2667 break; 2668 } 2669 } 2670 /* Congestion signal before ack. */ 2671 cc_cong_signal(tp, th, CC_NDUPACK); 2672 cc_ack_received(tp, th, nsegs, 2673 CC_DUPACK); 2674 tcp_timer_activate(tp, TT_REXMT, 0); 2675 tp->t_rtttime = 0; 2676 if (V_tcp_do_prr) { 2677 /* 2678 * snd_ssthresh is already updated by 2679 * cc_cong_signal. 2680 */ 2681 tp->sackhint.prr_delivered = 0; 2682 tp->sackhint.sack_bytes_rexmit = 0; 2683 tp->sackhint.recover_fs = max(1, 2684 tp->snd_nxt - tp->snd_una); 2685 } 2686 if (tp->t_flags & TF_SACK_PERMIT) { 2687 TCPSTAT_INC( 2688 tcps_sack_recovery_episode); 2689 tp->snd_recover = tp->snd_nxt; 2690 tp->snd_cwnd = maxseg; 2691 (void) tp->t_fb->tfb_tcp_output(tp); 2692 goto drop; 2693 } 2694 tp->snd_nxt = th->th_ack; 2695 tp->snd_cwnd = maxseg; 2696 (void) tp->t_fb->tfb_tcp_output(tp); 2697 KASSERT(tp->snd_limited <= 2, 2698 ("%s: tp->snd_limited too big", 2699 __func__)); 2700 tp->snd_cwnd = tp->snd_ssthresh + 2701 maxseg * 2702 (tp->t_dupacks - tp->snd_limited); 2703 if (SEQ_GT(onxt, tp->snd_nxt)) 2704 tp->snd_nxt = onxt; 2705 goto drop; 2706 } else if (V_tcp_do_rfc3042) { 2707 /* 2708 * Process first and second duplicate 2709 * ACKs. Each indicates a segment 2710 * leaving the network, creating room 2711 * for more. Make sure we can send a 2712 * packet on reception of each duplicate 2713 * ACK by increasing snd_cwnd by one 2714 * segment. Restore the original 2715 * snd_cwnd after packet transmission. 2716 */ 2717 cc_ack_received(tp, th, nsegs, 2718 CC_DUPACK); 2719 uint32_t oldcwnd = tp->snd_cwnd; 2720 tcp_seq oldsndmax = tp->snd_max; 2721 u_int sent; 2722 int avail; 2723 2724 KASSERT(tp->t_dupacks == 1 || 2725 tp->t_dupacks == 2, 2726 ("%s: dupacks not 1 or 2", 2727 __func__)); 2728 if (tp->t_dupacks == 1) 2729 tp->snd_limited = 0; 2730 tp->snd_cwnd = 2731 (tp->snd_nxt - tp->snd_una) + 2732 (tp->t_dupacks - tp->snd_limited) * 2733 maxseg; 2734 /* 2735 * Only call tcp_output when there 2736 * is new data available to be sent. 2737 * Otherwise we would send pure ACKs. 2738 */ 2739 SOCKBUF_LOCK(&so->so_snd); 2740 avail = sbavail(&so->so_snd) - 2741 (tp->snd_nxt - tp->snd_una); 2742 SOCKBUF_UNLOCK(&so->so_snd); 2743 if (avail > 0) 2744 (void) tp->t_fb->tfb_tcp_output(tp); 2745 sent = tp->snd_max - oldsndmax; 2746 if (sent > maxseg) { 2747 KASSERT((tp->t_dupacks == 2 && 2748 tp->snd_limited == 0) || 2749 (sent == maxseg + 1 && 2750 tp->t_flags & TF_SENTFIN), 2751 ("%s: sent too much", 2752 __func__)); 2753 tp->snd_limited = 2; 2754 } else if (sent > 0) 2755 ++tp->snd_limited; 2756 tp->snd_cwnd = oldcwnd; 2757 goto drop; 2758 } 2759 } 2760 break; 2761 } else { 2762 /* 2763 * This ack is advancing the left edge, reset the 2764 * counter. 2765 */ 2766 tp->t_dupacks = 0; 2767 /* 2768 * If this ack also has new SACK info, increment the 2769 * counter as per rfc6675. The variable 2770 * sack_changed tracks all changes to the SACK 2771 * scoreboard, including when partial ACKs without 2772 * SACK options are received, and clear the scoreboard 2773 * from the left side. Such partial ACKs should not be 2774 * counted as dupacks here. 2775 */ 2776 if ((tp->t_flags & TF_SACK_PERMIT) && 2777 (to.to_flags & TOF_SACK) && 2778 sack_changed) 2779 tp->t_dupacks++; 2780 } 2781 2782 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2783 ("%s: th_ack <= snd_una", __func__)); 2784 2785 /* 2786 * If the congestion window was inflated to account 2787 * for the other side's cached packets, retract it. 2788 */ 2789 if (IN_FASTRECOVERY(tp->t_flags)) { 2790 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2791 if (tp->t_flags & TF_SACK_PERMIT) 2792 if (V_tcp_do_prr) 2793 tcp_prr_partialack(tp, th); 2794 else 2795 tcp_sack_partialack(tp, th); 2796 else 2797 tcp_newreno_partial_ack(tp, th); 2798 } else 2799 cc_post_recovery(tp, th); 2800 } 2801 /* 2802 * If we reach this point, ACK is not a duplicate, 2803 * i.e., it ACKs something we sent. 2804 */ 2805 if (tp->t_flags & TF_NEEDSYN) { 2806 /* 2807 * T/TCP: Connection was half-synchronized, and our 2808 * SYN has been ACK'd (so connection is now fully 2809 * synchronized). Go to non-starred state, 2810 * increment snd_una for ACK of SYN, and check if 2811 * we can do window scaling. 2812 */ 2813 tp->t_flags &= ~TF_NEEDSYN; 2814 tp->snd_una++; 2815 /* Do window scaling? */ 2816 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2817 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2818 tp->rcv_scale = tp->request_r_scale; 2819 /* Send window already scaled. */ 2820 } 2821 } 2822 2823 process_ACK: 2824 INP_WLOCK_ASSERT(tp->t_inpcb); 2825 2826 /* 2827 * Adjust for the SYN bit in sequence space, 2828 * but don't account for it in cwnd calculations. 2829 * This is for the SYN_RECEIVED, non-simultaneous 2830 * SYN case. SYN_SENT and simultaneous SYN are 2831 * treated elsewhere. 2832 */ 2833 if (incforsyn) 2834 tp->snd_una++; 2835 acked = BYTES_THIS_ACK(tp, th); 2836 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2837 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2838 tp->snd_una, th->th_ack, tp, m)); 2839 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2840 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2841 2842 /* 2843 * If we just performed our first retransmit, and the ACK 2844 * arrives within our recovery window, then it was a mistake 2845 * to do the retransmit in the first place. Recover our 2846 * original cwnd and ssthresh, and proceed to transmit where 2847 * we left off. 2848 */ 2849 if (tp->t_rxtshift == 1 && 2850 tp->t_flags & TF_PREVVALID && 2851 tp->t_badrxtwin && 2852 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2853 cc_cong_signal(tp, th, CC_RTO_ERR); 2854 2855 /* 2856 * If we have a timestamp reply, update smoothed 2857 * round trip time. If no timestamp is present but 2858 * transmit timer is running and timed sequence 2859 * number was acked, update smoothed round trip time. 2860 * Since we now have an rtt measurement, cancel the 2861 * timer backoff (cf., Phil Karn's retransmit alg.). 2862 * Recompute the initial retransmit timer. 2863 * 2864 * Some boxes send broken timestamp replies 2865 * during the SYN+ACK phase, ignore 2866 * timestamps of 0 or we could calculate a 2867 * huge RTT and blow up the retransmit timer. 2868 */ 2869 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2870 uint32_t t; 2871 2872 t = tcp_ts_getticks() - to.to_tsecr; 2873 if (!tp->t_rttlow || tp->t_rttlow > t) 2874 tp->t_rttlow = t; 2875 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2876 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2877 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2878 tp->t_rttlow = ticks - tp->t_rtttime; 2879 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2880 } 2881 2882 /* 2883 * If all outstanding data is acked, stop retransmit 2884 * timer and remember to restart (more output or persist). 2885 * If there is more data to be acked, restart retransmit 2886 * timer, using current (possibly backed-off) value. 2887 */ 2888 if (th->th_ack == tp->snd_max) { 2889 tcp_timer_activate(tp, TT_REXMT, 0); 2890 needoutput = 1; 2891 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2892 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2893 2894 /* 2895 * If no data (only SYN) was ACK'd, 2896 * skip rest of ACK processing. 2897 */ 2898 if (acked == 0) 2899 goto step6; 2900 2901 /* 2902 * Let the congestion control algorithm update congestion 2903 * control related information. This typically means increasing 2904 * the congestion window. 2905 */ 2906 cc_ack_received(tp, th, nsegs, CC_ACK); 2907 2908 SOCKBUF_LOCK(&so->so_snd); 2909 if (acked > sbavail(&so->so_snd)) { 2910 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2911 tp->snd_wnd -= sbavail(&so->so_snd); 2912 else 2913 tp->snd_wnd = 0; 2914 mfree = sbcut_locked(&so->so_snd, 2915 (int)sbavail(&so->so_snd)); 2916 ourfinisacked = 1; 2917 } else { 2918 mfree = sbcut_locked(&so->so_snd, acked); 2919 if (tp->snd_wnd >= (uint32_t) acked) 2920 tp->snd_wnd -= acked; 2921 else 2922 tp->snd_wnd = 0; 2923 ourfinisacked = 0; 2924 } 2925 SOCKBUF_UNLOCK(&so->so_snd); 2926 tp->t_flags |= TF_WAKESOW; 2927 m_freem(mfree); 2928 /* Detect una wraparound. */ 2929 if (!IN_RECOVERY(tp->t_flags) && 2930 SEQ_GT(tp->snd_una, tp->snd_recover) && 2931 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2932 tp->snd_recover = th->th_ack - 1; 2933 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2934 if (IN_RECOVERY(tp->t_flags) && 2935 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2936 EXIT_RECOVERY(tp->t_flags); 2937 } 2938 tp->snd_una = th->th_ack; 2939 if (tp->t_flags & TF_SACK_PERMIT) { 2940 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2941 tp->snd_recover = tp->snd_una; 2942 } 2943 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2944 tp->snd_nxt = tp->snd_una; 2945 2946 switch (tp->t_state) { 2947 /* 2948 * In FIN_WAIT_1 STATE in addition to the processing 2949 * for the ESTABLISHED state if our FIN is now acknowledged 2950 * then enter FIN_WAIT_2. 2951 */ 2952 case TCPS_FIN_WAIT_1: 2953 if (ourfinisacked) { 2954 /* 2955 * If we can't receive any more 2956 * data, then closing user can proceed. 2957 * Starting the timer is contrary to the 2958 * specification, but if we don't get a FIN 2959 * we'll hang forever. 2960 * 2961 * XXXjl: 2962 * we should release the tp also, and use a 2963 * compressed state. 2964 */ 2965 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2966 soisdisconnected(so); 2967 tcp_timer_activate(tp, TT_2MSL, 2968 (tcp_fast_finwait2_recycle ? 2969 tcp_finwait2_timeout : 2970 TP_MAXIDLE(tp))); 2971 } 2972 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2973 } 2974 break; 2975 2976 /* 2977 * In CLOSING STATE in addition to the processing for 2978 * the ESTABLISHED state if the ACK acknowledges our FIN 2979 * then enter the TIME-WAIT state, otherwise ignore 2980 * the segment. 2981 */ 2982 case TCPS_CLOSING: 2983 if (ourfinisacked) { 2984 tcp_twstart(tp); 2985 m_freem(m); 2986 return; 2987 } 2988 break; 2989 2990 /* 2991 * In LAST_ACK, we may still be waiting for data to drain 2992 * and/or to be acked, as well as for the ack of our FIN. 2993 * If our FIN is now acknowledged, delete the TCB, 2994 * enter the closed state and return. 2995 */ 2996 case TCPS_LAST_ACK: 2997 if (ourfinisacked) { 2998 tp = tcp_close(tp); 2999 goto drop; 3000 } 3001 break; 3002 } 3003 } 3004 3005 step6: 3006 INP_WLOCK_ASSERT(tp->t_inpcb); 3007 3008 /* 3009 * Update window information. 3010 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3011 */ 3012 if ((thflags & TH_ACK) && 3013 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3014 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3015 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3016 /* keep track of pure window updates */ 3017 if (tlen == 0 && 3018 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3019 TCPSTAT_INC(tcps_rcvwinupd); 3020 tp->snd_wnd = tiwin; 3021 tp->snd_wl1 = th->th_seq; 3022 tp->snd_wl2 = th->th_ack; 3023 if (tp->snd_wnd > tp->max_sndwnd) 3024 tp->max_sndwnd = tp->snd_wnd; 3025 needoutput = 1; 3026 } 3027 3028 /* 3029 * Process segments with URG. 3030 */ 3031 if ((thflags & TH_URG) && th->th_urp && 3032 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3033 /* 3034 * This is a kludge, but if we receive and accept 3035 * random urgent pointers, we'll crash in 3036 * soreceive. It's hard to imagine someone 3037 * actually wanting to send this much urgent data. 3038 */ 3039 SOCKBUF_LOCK(&so->so_rcv); 3040 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3041 th->th_urp = 0; /* XXX */ 3042 thflags &= ~TH_URG; /* XXX */ 3043 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3044 goto dodata; /* XXX */ 3045 } 3046 /* 3047 * If this segment advances the known urgent pointer, 3048 * then mark the data stream. This should not happen 3049 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3050 * a FIN has been received from the remote side. 3051 * In these states we ignore the URG. 3052 * 3053 * According to RFC961 (Assigned Protocols), 3054 * the urgent pointer points to the last octet 3055 * of urgent data. We continue, however, 3056 * to consider it to indicate the first octet 3057 * of data past the urgent section as the original 3058 * spec states (in one of two places). 3059 */ 3060 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3061 tp->rcv_up = th->th_seq + th->th_urp; 3062 so->so_oobmark = sbavail(&so->so_rcv) + 3063 (tp->rcv_up - tp->rcv_nxt) - 1; 3064 if (so->so_oobmark == 0) 3065 so->so_rcv.sb_state |= SBS_RCVATMARK; 3066 sohasoutofband(so); 3067 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3068 } 3069 SOCKBUF_UNLOCK(&so->so_rcv); 3070 /* 3071 * Remove out of band data so doesn't get presented to user. 3072 * This can happen independent of advancing the URG pointer, 3073 * but if two URG's are pending at once, some out-of-band 3074 * data may creep in... ick. 3075 */ 3076 if (th->th_urp <= (uint32_t)tlen && 3077 !(so->so_options & SO_OOBINLINE)) { 3078 /* hdr drop is delayed */ 3079 tcp_pulloutofband(so, th, m, drop_hdrlen); 3080 } 3081 } else { 3082 /* 3083 * If no out of band data is expected, 3084 * pull receive urgent pointer along 3085 * with the receive window. 3086 */ 3087 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3088 tp->rcv_up = tp->rcv_nxt; 3089 } 3090 dodata: /* XXX */ 3091 INP_WLOCK_ASSERT(tp->t_inpcb); 3092 3093 /* 3094 * Process the segment text, merging it into the TCP sequencing queue, 3095 * and arranging for acknowledgment of receipt if necessary. 3096 * This process logically involves adjusting tp->rcv_wnd as data 3097 * is presented to the user (this happens in tcp_usrreq.c, 3098 * case PRU_RCVD). If a FIN has already been received on this 3099 * connection then we just ignore the text. 3100 */ 3101 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3102 IS_FASTOPEN(tp->t_flags)); 3103 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3104 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3105 tcp_seq save_start = th->th_seq; 3106 tcp_seq save_rnxt = tp->rcv_nxt; 3107 int save_tlen = tlen; 3108 m_adj(m, drop_hdrlen); /* delayed header drop */ 3109 /* 3110 * Insert segment which includes th into TCP reassembly queue 3111 * with control block tp. Set thflags to whether reassembly now 3112 * includes a segment with FIN. This handles the common case 3113 * inline (segment is the next to be received on an established 3114 * connection, and the queue is empty), avoiding linkage into 3115 * and removal from the queue and repetition of various 3116 * conversions. 3117 * Set DELACK for segments received in order, but ack 3118 * immediately when segments are out of order (so 3119 * fast retransmit can work). 3120 */ 3121 if (th->th_seq == tp->rcv_nxt && 3122 SEGQ_EMPTY(tp) && 3123 (TCPS_HAVEESTABLISHED(tp->t_state) || 3124 tfo_syn)) { 3125 if (DELAY_ACK(tp, tlen) || tfo_syn) 3126 tp->t_flags |= TF_DELACK; 3127 else 3128 tp->t_flags |= TF_ACKNOW; 3129 tp->rcv_nxt += tlen; 3130 if (tlen && 3131 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3132 (tp->t_fbyte_in == 0)) { 3133 tp->t_fbyte_in = ticks; 3134 if (tp->t_fbyte_in == 0) 3135 tp->t_fbyte_in = 1; 3136 if (tp->t_fbyte_out && tp->t_fbyte_in) 3137 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3138 } 3139 thflags = th->th_flags & TH_FIN; 3140 TCPSTAT_INC(tcps_rcvpack); 3141 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3142 SOCKBUF_LOCK(&so->so_rcv); 3143 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3144 m_freem(m); 3145 else 3146 sbappendstream_locked(&so->so_rcv, m, 0); 3147 SOCKBUF_UNLOCK(&so->so_rcv); 3148 tp->t_flags |= TF_WAKESOR; 3149 } else { 3150 /* 3151 * XXX: Due to the header drop above "th" is 3152 * theoretically invalid by now. Fortunately 3153 * m_adj() doesn't actually frees any mbufs 3154 * when trimming from the head. 3155 */ 3156 tcp_seq temp = save_start; 3157 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3158 tp->t_flags |= TF_ACKNOW; 3159 } 3160 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) { 3161 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3162 /* 3163 * DSACK actually handled in the fastpath 3164 * above. 3165 */ 3166 tcp_update_sack_list(tp, save_start, 3167 save_start + save_tlen); 3168 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3169 if ((tp->rcv_numsacks >= 1) && 3170 (tp->sackblks[0].end == save_start)) { 3171 /* 3172 * Partial overlap, recorded at todrop 3173 * above. 3174 */ 3175 tcp_update_sack_list(tp, 3176 tp->sackblks[0].start, 3177 tp->sackblks[0].end); 3178 } else { 3179 tcp_update_dsack_list(tp, save_start, 3180 save_start + save_tlen); 3181 } 3182 } else if (tlen >= save_tlen) { 3183 /* Update of sackblks. */ 3184 tcp_update_dsack_list(tp, save_start, 3185 save_start + save_tlen); 3186 } else if (tlen > 0) { 3187 tcp_update_dsack_list(tp, save_start, 3188 save_start + tlen); 3189 } 3190 } 3191 #if 0 3192 /* 3193 * Note the amount of data that peer has sent into 3194 * our window, in order to estimate the sender's 3195 * buffer size. 3196 * XXX: Unused. 3197 */ 3198 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3199 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3200 else 3201 len = so->so_rcv.sb_hiwat; 3202 #endif 3203 } else { 3204 m_freem(m); 3205 thflags &= ~TH_FIN; 3206 } 3207 3208 /* 3209 * If FIN is received ACK the FIN and let the user know 3210 * that the connection is closing. 3211 */ 3212 if (thflags & TH_FIN) { 3213 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3214 socantrcvmore(so); 3215 /* The socket upcall is handled by socantrcvmore. */ 3216 tp->t_flags &= ~TF_WAKESOR; 3217 /* 3218 * If connection is half-synchronized 3219 * (ie NEEDSYN flag on) then delay ACK, 3220 * so it may be piggybacked when SYN is sent. 3221 * Otherwise, since we received a FIN then no 3222 * more input can be expected, send ACK now. 3223 */ 3224 if (tp->t_flags & TF_NEEDSYN) 3225 tp->t_flags |= TF_DELACK; 3226 else 3227 tp->t_flags |= TF_ACKNOW; 3228 tp->rcv_nxt++; 3229 } 3230 switch (tp->t_state) { 3231 /* 3232 * In SYN_RECEIVED and ESTABLISHED STATES 3233 * enter the CLOSE_WAIT state. 3234 */ 3235 case TCPS_SYN_RECEIVED: 3236 tp->t_starttime = ticks; 3237 /* FALLTHROUGH */ 3238 case TCPS_ESTABLISHED: 3239 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3240 break; 3241 3242 /* 3243 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3244 * enter the CLOSING state. 3245 */ 3246 case TCPS_FIN_WAIT_1: 3247 tcp_state_change(tp, TCPS_CLOSING); 3248 break; 3249 3250 /* 3251 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3252 * starting the time-wait timer, turning off the other 3253 * standard timers. 3254 */ 3255 case TCPS_FIN_WAIT_2: 3256 tcp_twstart(tp); 3257 return; 3258 } 3259 } 3260 #ifdef TCPDEBUG 3261 if (so->so_options & SO_DEBUG) 3262 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3263 &tcp_savetcp, 0); 3264 #endif 3265 TCP_PROBE3(debug__input, tp, th, m); 3266 3267 /* 3268 * Return any desired output. 3269 */ 3270 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3271 (void) tp->t_fb->tfb_tcp_output(tp); 3272 3273 check_delack: 3274 INP_WLOCK_ASSERT(tp->t_inpcb); 3275 3276 if (tp->t_flags & TF_DELACK) { 3277 tp->t_flags &= ~TF_DELACK; 3278 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3279 } 3280 tcp_handle_wakeup(tp, so); 3281 INP_WUNLOCK(tp->t_inpcb); 3282 return; 3283 3284 dropafterack: 3285 /* 3286 * Generate an ACK dropping incoming segment if it occupies 3287 * sequence space, where the ACK reflects our state. 3288 * 3289 * We can now skip the test for the RST flag since all 3290 * paths to this code happen after packets containing 3291 * RST have been dropped. 3292 * 3293 * In the SYN-RECEIVED state, don't send an ACK unless the 3294 * segment we received passes the SYN-RECEIVED ACK test. 3295 * If it fails send a RST. This breaks the loop in the 3296 * "LAND" DoS attack, and also prevents an ACK storm 3297 * between two listening ports that have been sent forged 3298 * SYN segments, each with the source address of the other. 3299 */ 3300 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3301 (SEQ_GT(tp->snd_una, th->th_ack) || 3302 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3303 rstreason = BANDLIM_RST_OPENPORT; 3304 goto dropwithreset; 3305 } 3306 #ifdef TCPDEBUG 3307 if (so->so_options & SO_DEBUG) 3308 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3309 &tcp_savetcp, 0); 3310 #endif 3311 TCP_PROBE3(debug__input, tp, th, m); 3312 tp->t_flags |= TF_ACKNOW; 3313 (void) tp->t_fb->tfb_tcp_output(tp); 3314 tcp_handle_wakeup(tp, so); 3315 INP_WUNLOCK(tp->t_inpcb); 3316 m_freem(m); 3317 return; 3318 3319 dropwithreset: 3320 if (tp != NULL) { 3321 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3322 tcp_handle_wakeup(tp, so); 3323 INP_WUNLOCK(tp->t_inpcb); 3324 } else 3325 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3326 return; 3327 3328 drop: 3329 /* 3330 * Drop space held by incoming segment and return. 3331 */ 3332 #ifdef TCPDEBUG 3333 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3334 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3335 &tcp_savetcp, 0); 3336 #endif 3337 TCP_PROBE3(debug__input, tp, th, m); 3338 if (tp != NULL) { 3339 tcp_handle_wakeup(tp, so); 3340 INP_WUNLOCK(tp->t_inpcb); 3341 } 3342 m_freem(m); 3343 } 3344 3345 /* 3346 * Issue RST and make ACK acceptable to originator of segment. 3347 * The mbuf must still include the original packet header. 3348 * tp may be NULL. 3349 */ 3350 void 3351 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3352 int tlen, int rstreason) 3353 { 3354 #ifdef INET 3355 struct ip *ip; 3356 #endif 3357 #ifdef INET6 3358 struct ip6_hdr *ip6; 3359 #endif 3360 3361 if (tp != NULL) { 3362 INP_WLOCK_ASSERT(tp->t_inpcb); 3363 } 3364 3365 /* Don't bother if destination was broadcast/multicast. */ 3366 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3367 goto drop; 3368 #ifdef INET6 3369 if (mtod(m, struct ip *)->ip_v == 6) { 3370 ip6 = mtod(m, struct ip6_hdr *); 3371 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3372 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3373 goto drop; 3374 /* IPv6 anycast check is done at tcp6_input() */ 3375 } 3376 #endif 3377 #if defined(INET) && defined(INET6) 3378 else 3379 #endif 3380 #ifdef INET 3381 { 3382 ip = mtod(m, struct ip *); 3383 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3384 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3385 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3386 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3387 goto drop; 3388 } 3389 #endif 3390 3391 /* Perform bandwidth limiting. */ 3392 if (badport_bandlim(rstreason) < 0) 3393 goto drop; 3394 3395 /* tcp_respond consumes the mbuf chain. */ 3396 if (th->th_flags & TH_ACK) { 3397 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3398 th->th_ack, TH_RST); 3399 } else { 3400 if (th->th_flags & TH_SYN) 3401 tlen++; 3402 if (th->th_flags & TH_FIN) 3403 tlen++; 3404 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3405 (tcp_seq)0, TH_RST|TH_ACK); 3406 } 3407 return; 3408 drop: 3409 m_freem(m); 3410 } 3411 3412 /* 3413 * Parse TCP options and place in tcpopt. 3414 */ 3415 void 3416 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3417 { 3418 int opt, optlen; 3419 3420 to->to_flags = 0; 3421 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3422 opt = cp[0]; 3423 if (opt == TCPOPT_EOL) 3424 break; 3425 if (opt == TCPOPT_NOP) 3426 optlen = 1; 3427 else { 3428 if (cnt < 2) 3429 break; 3430 optlen = cp[1]; 3431 if (optlen < 2 || optlen > cnt) 3432 break; 3433 } 3434 switch (opt) { 3435 case TCPOPT_MAXSEG: 3436 if (optlen != TCPOLEN_MAXSEG) 3437 continue; 3438 if (!(flags & TO_SYN)) 3439 continue; 3440 to->to_flags |= TOF_MSS; 3441 bcopy((char *)cp + 2, 3442 (char *)&to->to_mss, sizeof(to->to_mss)); 3443 to->to_mss = ntohs(to->to_mss); 3444 break; 3445 case TCPOPT_WINDOW: 3446 if (optlen != TCPOLEN_WINDOW) 3447 continue; 3448 if (!(flags & TO_SYN)) 3449 continue; 3450 to->to_flags |= TOF_SCALE; 3451 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3452 break; 3453 case TCPOPT_TIMESTAMP: 3454 if (optlen != TCPOLEN_TIMESTAMP) 3455 continue; 3456 to->to_flags |= TOF_TS; 3457 bcopy((char *)cp + 2, 3458 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3459 to->to_tsval = ntohl(to->to_tsval); 3460 bcopy((char *)cp + 6, 3461 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3462 to->to_tsecr = ntohl(to->to_tsecr); 3463 break; 3464 case TCPOPT_SIGNATURE: 3465 /* 3466 * In order to reply to a host which has set the 3467 * TCP_SIGNATURE option in its initial SYN, we have 3468 * to record the fact that the option was observed 3469 * here for the syncache code to perform the correct 3470 * response. 3471 */ 3472 if (optlen != TCPOLEN_SIGNATURE) 3473 continue; 3474 to->to_flags |= TOF_SIGNATURE; 3475 to->to_signature = cp + 2; 3476 break; 3477 case TCPOPT_SACK_PERMITTED: 3478 if (optlen != TCPOLEN_SACK_PERMITTED) 3479 continue; 3480 if (!(flags & TO_SYN)) 3481 continue; 3482 if (!V_tcp_do_sack) 3483 continue; 3484 to->to_flags |= TOF_SACKPERM; 3485 break; 3486 case TCPOPT_SACK: 3487 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3488 continue; 3489 if (flags & TO_SYN) 3490 continue; 3491 to->to_flags |= TOF_SACK; 3492 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3493 to->to_sacks = cp + 2; 3494 TCPSTAT_INC(tcps_sack_rcv_blocks); 3495 break; 3496 case TCPOPT_FAST_OPEN: 3497 /* 3498 * Cookie length validation is performed by the 3499 * server side cookie checking code or the client 3500 * side cookie cache update code. 3501 */ 3502 if (!(flags & TO_SYN)) 3503 continue; 3504 if (!V_tcp_fastopen_client_enable && 3505 !V_tcp_fastopen_server_enable) 3506 continue; 3507 to->to_flags |= TOF_FASTOPEN; 3508 to->to_tfo_len = optlen - 2; 3509 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3510 break; 3511 default: 3512 continue; 3513 } 3514 } 3515 } 3516 3517 /* 3518 * Pull out of band byte out of a segment so 3519 * it doesn't appear in the user's data queue. 3520 * It is still reflected in the segment length for 3521 * sequencing purposes. 3522 */ 3523 void 3524 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3525 int off) 3526 { 3527 int cnt = off + th->th_urp - 1; 3528 3529 while (cnt >= 0) { 3530 if (m->m_len > cnt) { 3531 char *cp = mtod(m, caddr_t) + cnt; 3532 struct tcpcb *tp = sototcpcb(so); 3533 3534 INP_WLOCK_ASSERT(tp->t_inpcb); 3535 3536 tp->t_iobc = *cp; 3537 tp->t_oobflags |= TCPOOB_HAVEDATA; 3538 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3539 m->m_len--; 3540 if (m->m_flags & M_PKTHDR) 3541 m->m_pkthdr.len--; 3542 return; 3543 } 3544 cnt -= m->m_len; 3545 m = m->m_next; 3546 if (m == NULL) 3547 break; 3548 } 3549 panic("tcp_pulloutofband"); 3550 } 3551 3552 /* 3553 * Collect new round-trip time estimate 3554 * and update averages and current timeout. 3555 */ 3556 void 3557 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3558 { 3559 int delta; 3560 3561 INP_WLOCK_ASSERT(tp->t_inpcb); 3562 3563 TCPSTAT_INC(tcps_rttupdated); 3564 tp->t_rttupdated++; 3565 #ifdef STATS 3566 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3567 imax(0, rtt * 1000 / hz)); 3568 #endif 3569 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3570 /* 3571 * srtt is stored as fixed point with 5 bits after the 3572 * binary point (i.e., scaled by 8). The following magic 3573 * is equivalent to the smoothing algorithm in rfc793 with 3574 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3575 * point). Adjust rtt to origin 0. 3576 */ 3577 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3578 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3579 3580 if ((tp->t_srtt += delta) <= 0) 3581 tp->t_srtt = 1; 3582 3583 /* 3584 * We accumulate a smoothed rtt variance (actually, a 3585 * smoothed mean difference), then set the retransmit 3586 * timer to smoothed rtt + 4 times the smoothed variance. 3587 * rttvar is stored as fixed point with 4 bits after the 3588 * binary point (scaled by 16). The following is 3589 * equivalent to rfc793 smoothing with an alpha of .75 3590 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3591 * rfc793's wired-in beta. 3592 */ 3593 if (delta < 0) 3594 delta = -delta; 3595 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3596 if ((tp->t_rttvar += delta) <= 0) 3597 tp->t_rttvar = 1; 3598 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3599 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3600 } else { 3601 /* 3602 * No rtt measurement yet - use the unsmoothed rtt. 3603 * Set the variance to half the rtt (so our first 3604 * retransmit happens at 3*rtt). 3605 */ 3606 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3607 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3608 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3609 } 3610 tp->t_rtttime = 0; 3611 tp->t_rxtshift = 0; 3612 3613 /* 3614 * the retransmit should happen at rtt + 4 * rttvar. 3615 * Because of the way we do the smoothing, srtt and rttvar 3616 * will each average +1/2 tick of bias. When we compute 3617 * the retransmit timer, we want 1/2 tick of rounding and 3618 * 1 extra tick because of +-1/2 tick uncertainty in the 3619 * firing of the timer. The bias will give us exactly the 3620 * 1.5 tick we need. But, because the bias is 3621 * statistical, we have to test that we don't drop below 3622 * the minimum feasible timer (which is 2 ticks). 3623 */ 3624 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3625 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3626 3627 /* 3628 * We received an ack for a packet that wasn't retransmitted; 3629 * it is probably safe to discard any error indications we've 3630 * received recently. This isn't quite right, but close enough 3631 * for now (a route might have failed after we sent a segment, 3632 * and the return path might not be symmetrical). 3633 */ 3634 tp->t_softerror = 0; 3635 } 3636 3637 /* 3638 * Determine a reasonable value for maxseg size. 3639 * If the route is known, check route for mtu. 3640 * If none, use an mss that can be handled on the outgoing interface 3641 * without forcing IP to fragment. If no route is found, route has no mtu, 3642 * or the destination isn't local, use a default, hopefully conservative 3643 * size (usually 512 or the default IP max size, but no more than the mtu 3644 * of the interface), as we can't discover anything about intervening 3645 * gateways or networks. We also initialize the congestion/slow start 3646 * window to be a single segment if the destination isn't local. 3647 * While looking at the routing entry, we also initialize other path-dependent 3648 * parameters from pre-set or cached values in the routing entry. 3649 * 3650 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3651 * IP options, e.g. IPSEC data, since length of this data may vary, and 3652 * thus it is calculated for every segment separately in tcp_output(). 3653 * 3654 * NOTE that this routine is only called when we process an incoming 3655 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3656 * settings are handled in tcp_mssopt(). 3657 */ 3658 void 3659 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3660 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3661 { 3662 int mss = 0; 3663 uint32_t maxmtu = 0; 3664 struct inpcb *inp = tp->t_inpcb; 3665 struct hc_metrics_lite metrics; 3666 #ifdef INET6 3667 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3668 size_t min_protoh = isipv6 ? 3669 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3670 sizeof (struct tcpiphdr); 3671 #else 3672 const size_t min_protoh = sizeof(struct tcpiphdr); 3673 #endif 3674 3675 INP_WLOCK_ASSERT(tp->t_inpcb); 3676 3677 if (mtuoffer != -1) { 3678 KASSERT(offer == -1, ("%s: conflict", __func__)); 3679 offer = mtuoffer - min_protoh; 3680 } 3681 3682 /* Initialize. */ 3683 #ifdef INET6 3684 if (isipv6) { 3685 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3686 tp->t_maxseg = V_tcp_v6mssdflt; 3687 } 3688 #endif 3689 #if defined(INET) && defined(INET6) 3690 else 3691 #endif 3692 #ifdef INET 3693 { 3694 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3695 tp->t_maxseg = V_tcp_mssdflt; 3696 } 3697 #endif 3698 3699 /* 3700 * No route to sender, stay with default mss and return. 3701 */ 3702 if (maxmtu == 0) { 3703 /* 3704 * In case we return early we need to initialize metrics 3705 * to a defined state as tcp_hc_get() would do for us 3706 * if there was no cache hit. 3707 */ 3708 if (metricptr != NULL) 3709 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3710 return; 3711 } 3712 3713 /* What have we got? */ 3714 switch (offer) { 3715 case 0: 3716 /* 3717 * Offer == 0 means that there was no MSS on the SYN 3718 * segment, in this case we use tcp_mssdflt as 3719 * already assigned to t_maxseg above. 3720 */ 3721 offer = tp->t_maxseg; 3722 break; 3723 3724 case -1: 3725 /* 3726 * Offer == -1 means that we didn't receive SYN yet. 3727 */ 3728 /* FALLTHROUGH */ 3729 3730 default: 3731 /* 3732 * Prevent DoS attack with too small MSS. Round up 3733 * to at least minmss. 3734 */ 3735 offer = max(offer, V_tcp_minmss); 3736 } 3737 3738 /* 3739 * rmx information is now retrieved from tcp_hostcache. 3740 */ 3741 tcp_hc_get(&inp->inp_inc, &metrics); 3742 if (metricptr != NULL) 3743 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3744 3745 /* 3746 * If there's a discovered mtu in tcp hostcache, use it. 3747 * Else, use the link mtu. 3748 */ 3749 if (metrics.rmx_mtu) 3750 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3751 else { 3752 #ifdef INET6 3753 if (isipv6) { 3754 mss = maxmtu - min_protoh; 3755 if (!V_path_mtu_discovery && 3756 !in6_localaddr(&inp->in6p_faddr)) 3757 mss = min(mss, V_tcp_v6mssdflt); 3758 } 3759 #endif 3760 #if defined(INET) && defined(INET6) 3761 else 3762 #endif 3763 #ifdef INET 3764 { 3765 mss = maxmtu - min_protoh; 3766 if (!V_path_mtu_discovery && 3767 !in_localaddr(inp->inp_faddr)) 3768 mss = min(mss, V_tcp_mssdflt); 3769 } 3770 #endif 3771 /* 3772 * XXX - The above conditional (mss = maxmtu - min_protoh) 3773 * probably violates the TCP spec. 3774 * The problem is that, since we don't know the 3775 * other end's MSS, we are supposed to use a conservative 3776 * default. But, if we do that, then MTU discovery will 3777 * never actually take place, because the conservative 3778 * default is much less than the MTUs typically seen 3779 * on the Internet today. For the moment, we'll sweep 3780 * this under the carpet. 3781 * 3782 * The conservative default might not actually be a problem 3783 * if the only case this occurs is when sending an initial 3784 * SYN with options and data to a host we've never talked 3785 * to before. Then, they will reply with an MSS value which 3786 * will get recorded and the new parameters should get 3787 * recomputed. For Further Study. 3788 */ 3789 } 3790 mss = min(mss, offer); 3791 3792 /* 3793 * Sanity check: make sure that maxseg will be large 3794 * enough to allow some data on segments even if the 3795 * all the option space is used (40bytes). Otherwise 3796 * funny things may happen in tcp_output. 3797 * 3798 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3799 */ 3800 mss = max(mss, 64); 3801 3802 tp->t_maxseg = mss; 3803 } 3804 3805 void 3806 tcp_mss(struct tcpcb *tp, int offer) 3807 { 3808 int mss; 3809 uint32_t bufsize; 3810 struct inpcb *inp; 3811 struct socket *so; 3812 struct hc_metrics_lite metrics; 3813 struct tcp_ifcap cap; 3814 3815 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3816 3817 bzero(&cap, sizeof(cap)); 3818 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3819 3820 mss = tp->t_maxseg; 3821 inp = tp->t_inpcb; 3822 3823 /* 3824 * If there's a pipesize, change the socket buffer to that size, 3825 * don't change if sb_hiwat is different than default (then it 3826 * has been changed on purpose with setsockopt). 3827 * Make the socket buffers an integral number of mss units; 3828 * if the mss is larger than the socket buffer, decrease the mss. 3829 */ 3830 so = inp->inp_socket; 3831 SOCKBUF_LOCK(&so->so_snd); 3832 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3833 bufsize = metrics.rmx_sendpipe; 3834 else 3835 bufsize = so->so_snd.sb_hiwat; 3836 if (bufsize < mss) 3837 mss = bufsize; 3838 else { 3839 bufsize = roundup(bufsize, mss); 3840 if (bufsize > sb_max) 3841 bufsize = sb_max; 3842 if (bufsize > so->so_snd.sb_hiwat) 3843 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3844 } 3845 SOCKBUF_UNLOCK(&so->so_snd); 3846 /* 3847 * Sanity check: make sure that maxseg will be large 3848 * enough to allow some data on segments even if the 3849 * all the option space is used (40bytes). Otherwise 3850 * funny things may happen in tcp_output. 3851 * 3852 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3853 */ 3854 tp->t_maxseg = max(mss, 64); 3855 3856 SOCKBUF_LOCK(&so->so_rcv); 3857 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3858 bufsize = metrics.rmx_recvpipe; 3859 else 3860 bufsize = so->so_rcv.sb_hiwat; 3861 if (bufsize > mss) { 3862 bufsize = roundup(bufsize, mss); 3863 if (bufsize > sb_max) 3864 bufsize = sb_max; 3865 if (bufsize > so->so_rcv.sb_hiwat) 3866 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3867 } 3868 SOCKBUF_UNLOCK(&so->so_rcv); 3869 3870 /* Check the interface for TSO capabilities. */ 3871 if (cap.ifcap & CSUM_TSO) { 3872 tp->t_flags |= TF_TSO; 3873 tp->t_tsomax = cap.tsomax; 3874 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3875 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3876 } 3877 } 3878 3879 /* 3880 * Determine the MSS option to send on an outgoing SYN. 3881 */ 3882 int 3883 tcp_mssopt(struct in_conninfo *inc) 3884 { 3885 int mss = 0; 3886 uint32_t thcmtu = 0; 3887 uint32_t maxmtu = 0; 3888 size_t min_protoh; 3889 3890 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3891 3892 #ifdef INET6 3893 if (inc->inc_flags & INC_ISIPV6) { 3894 mss = V_tcp_v6mssdflt; 3895 maxmtu = tcp_maxmtu6(inc, NULL); 3896 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3897 } 3898 #endif 3899 #if defined(INET) && defined(INET6) 3900 else 3901 #endif 3902 #ifdef INET 3903 { 3904 mss = V_tcp_mssdflt; 3905 maxmtu = tcp_maxmtu(inc, NULL); 3906 min_protoh = sizeof(struct tcpiphdr); 3907 } 3908 #endif 3909 #if defined(INET6) || defined(INET) 3910 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3911 #endif 3912 3913 if (maxmtu && thcmtu) 3914 mss = min(maxmtu, thcmtu) - min_protoh; 3915 else if (maxmtu || thcmtu) 3916 mss = max(maxmtu, thcmtu) - min_protoh; 3917 3918 return (mss); 3919 } 3920 3921 void 3922 tcp_prr_partialack(struct tcpcb *tp, struct tcphdr *th) 3923 { 3924 long snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3925 int maxseg = tcp_maxseg(tp); 3926 3927 INP_WLOCK_ASSERT(tp->t_inpcb); 3928 3929 tcp_timer_activate(tp, TT_REXMT, 0); 3930 tp->t_rtttime = 0; 3931 /* 3932 * Compute the amount of data that this ACK is indicating 3933 * (del_data) and an estimate of how many bytes are in the 3934 * network. 3935 */ 3936 if (SEQ_GEQ(th->th_ack, tp->snd_una)) 3937 del_data = BYTES_THIS_ACK(tp, th); 3938 del_data += tp->sackhint.delivered_data; 3939 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; 3940 tp->sackhint.prr_delivered += del_data; 3941 /* 3942 * Proportional Rate Reduction 3943 */ 3944 if (pipe > tp->snd_ssthresh) { 3945 if (tp->sackhint.recover_fs == 0) 3946 tp->sackhint.recover_fs = 3947 max(1, tp->snd_nxt - tp->snd_una); 3948 snd_cnt = (tp->sackhint.prr_delivered * tp->snd_ssthresh / 3949 tp->sackhint.recover_fs) - tp->sackhint.sack_bytes_rexmit; 3950 } else { 3951 if (V_tcp_do_prr_conservative) 3952 limit = tp->sackhint.prr_delivered - 3953 tp->sackhint.sack_bytes_rexmit; 3954 else 3955 if ((tp->sackhint.prr_delivered - 3956 tp->sackhint.sack_bytes_rexmit) > del_data) 3957 limit = tp->sackhint.prr_delivered - 3958 tp->sackhint.sack_bytes_rexmit + maxseg; 3959 else 3960 limit = del_data + maxseg; 3961 snd_cnt = min((tp->snd_ssthresh - pipe), limit); 3962 } 3963 snd_cnt = max((snd_cnt / maxseg), 0); 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 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 3970 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg); 3971 tp->t_flags |= TF_ACKNOW; 3972 (void) tcp_output(tp); 3973 } 3974 3975 /* 3976 * On a partial ack arrives, force the retransmission of the 3977 * next unacknowledged segment. Do not clear tp->t_dupacks. 3978 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3979 * be started again. 3980 */ 3981 void 3982 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3983 { 3984 tcp_seq onxt = tp->snd_nxt; 3985 uint32_t ocwnd = tp->snd_cwnd; 3986 u_int maxseg = tcp_maxseg(tp); 3987 3988 INP_WLOCK_ASSERT(tp->t_inpcb); 3989 3990 tcp_timer_activate(tp, TT_REXMT, 0); 3991 tp->t_rtttime = 0; 3992 tp->snd_nxt = th->th_ack; 3993 /* 3994 * Set snd_cwnd to one segment beyond acknowledged offset. 3995 * (tp->snd_una has not yet been updated when this function is called.) 3996 */ 3997 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3998 tp->t_flags |= TF_ACKNOW; 3999 (void) tp->t_fb->tfb_tcp_output(tp); 4000 tp->snd_cwnd = ocwnd; 4001 if (SEQ_GT(onxt, tp->snd_nxt)) 4002 tp->snd_nxt = onxt; 4003 /* 4004 * Partial window deflation. Relies on fact that tp->snd_una 4005 * not updated yet. 4006 */ 4007 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4008 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4009 else 4010 tp->snd_cwnd = 0; 4011 tp->snd_cwnd += maxseg; 4012 } 4013 4014 int 4015 tcp_compute_pipe(struct tcpcb *tp) 4016 { 4017 return (tp->snd_max - tp->snd_una + 4018 tp->sackhint.sack_bytes_rexmit - 4019 tp->sackhint.sacked_bytes); 4020 } 4021 4022 uint32_t 4023 tcp_compute_initwnd(uint32_t maxseg) 4024 { 4025 /* 4026 * Calculate the Initial Window, also used as Restart Window 4027 * 4028 * RFC5681 Section 3.1 specifies the default conservative values. 4029 * RFC3390 specifies slightly more aggressive values. 4030 * RFC6928 increases it to ten segments. 4031 * Support for user specified value for initial flight size. 4032 */ 4033 if (V_tcp_initcwnd_segments) 4034 return min(V_tcp_initcwnd_segments * maxseg, 4035 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4036 else if (V_tcp_do_rfc3390) 4037 return min(4 * maxseg, max(2 * maxseg, 4380)); 4038 else { 4039 /* Per RFC5681 Section 3.1 */ 4040 if (maxseg > 2190) 4041 return (2 * maxseg); 4042 else if (maxseg > 1095) 4043 return (3 * maxseg); 4044 else 4045 return (4 * maxseg); 4046 } 4047 } 4048