1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 __FBSDID("$FreeBSD$"); 54 55 #include "opt_inet.h" 56 #include "opt_inet6.h" 57 #include "opt_ipsec.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/arb.h> 62 #include <sys/kernel.h> 63 #ifdef TCP_HHOOK 64 #include <sys/hhook.h> 65 #endif 66 #include <sys/malloc.h> 67 #include <sys/mbuf.h> 68 #include <sys/proc.h> /* for proc0 declaration */ 69 #include <sys/protosw.h> 70 #include <sys/qmath.h> 71 #include <sys/sdt.h> 72 #include <sys/signalvar.h> 73 #include <sys/socket.h> 74 #include <sys/socketvar.h> 75 #include <sys/sysctl.h> 76 #include <sys/syslog.h> 77 #include <sys/systm.h> 78 #include <sys/stats.h> 79 80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 81 82 #include <vm/uma.h> 83 84 #include <net/if.h> 85 #include <net/if_var.h> 86 #include <net/route.h> 87 #include <net/vnet.h> 88 89 #define TCPSTATES /* for logging */ 90 91 #include <netinet/in.h> 92 #include <netinet/in_kdtrace.h> 93 #include <netinet/in_pcb.h> 94 #include <netinet/in_systm.h> 95 #include <netinet/ip.h> 96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 98 #include <netinet/ip_var.h> 99 #include <netinet/ip_options.h> 100 #include <netinet/ip6.h> 101 #include <netinet/icmp6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_var.h> 104 #include <netinet6/ip6_var.h> 105 #include <netinet6/nd6.h> 106 #include <netinet/tcp.h> 107 #include <netinet/tcp_fsm.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_seq.h> 110 #include <netinet/tcp_timer.h> 111 #include <netinet/tcp_var.h> 112 #include <netinet6/tcp6_var.h> 113 #include <netinet/tcpip.h> 114 #include <netinet/cc/cc.h> 115 #include <netinet/tcp_fastopen.h> 116 #ifdef TCPPCAP 117 #include <netinet/tcp_pcap.h> 118 #endif 119 #include <netinet/tcp_syncache.h> 120 #ifdef TCPDEBUG 121 #include <netinet/tcp_debug.h> 122 #endif /* TCPDEBUG */ 123 #ifdef TCP_OFFLOAD 124 #include <netinet/tcp_offload.h> 125 #endif 126 #include <netinet/udp.h> 127 128 #include <netipsec/ipsec_support.h> 129 130 #include <machine/in_cksum.h> 131 132 #include <security/mac/mac_framework.h> 133 134 const int tcprexmtthresh = 3; 135 136 VNET_DEFINE(int, tcp_log_in_vain) = 0; 137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 138 &VNET_NAME(tcp_log_in_vain), 0, 139 "Log all incoming TCP segments to closed ports"); 140 141 VNET_DEFINE(int, blackhole) = 0; 142 #define V_blackhole VNET(blackhole) 143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 144 &VNET_NAME(blackhole), 0, 145 "Do not send RST on segments to closed ports"); 146 147 VNET_DEFINE(int, tcp_delack_enabled) = 1; 148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 149 &VNET_NAME(tcp_delack_enabled), 0, 150 "Delay ACK to try and piggyback it onto a data packet"); 151 152 VNET_DEFINE(int, drop_synfin) = 0; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 154 &VNET_NAME(drop_synfin), 0, 155 "Drop TCP packets with SYN+FIN set"); 156 157 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 159 &VNET_NAME(tcp_do_prr_conservative), 0, 160 "Do conservative Proportional Rate Reduction"); 161 162 VNET_DEFINE(int, tcp_do_prr) = 1; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_prr), 1, 165 "Enable Proportional Rate Reduction per RFC 6937"); 166 167 VNET_DEFINE(int, tcp_do_lrd) = 0; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_lrd), 1, 170 "Perform Lost Retransmission Detection"); 171 172 VNET_DEFINE(int, tcp_do_newcwv) = 0; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_do_newcwv), 0, 175 "Enable New Congestion Window Validation per RFC7661"); 176 177 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 179 &VNET_NAME(tcp_do_rfc3042), 0, 180 "Enable RFC 3042 (Limited Transmit)"); 181 182 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 184 &VNET_NAME(tcp_do_rfc3390), 0, 185 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 186 187 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 189 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 190 "Slow-start flight size (initial congestion window) in number of segments"); 191 192 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 194 &VNET_NAME(tcp_do_rfc3465), 0, 195 "Enable RFC 3465 (Appropriate Byte Counting)"); 196 197 VNET_DEFINE(int, tcp_abc_l_var) = 2; 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_abc_l_var), 2, 200 "Cap the max cwnd increment during slow-start to this number of segments"); 201 202 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 203 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 204 "TCP ECN"); 205 206 VNET_DEFINE(int, tcp_do_ecn) = 2; 207 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_do_ecn), 0, 209 "TCP ECN support"); 210 211 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 212 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(tcp_ecn_maxretries), 0, 214 "Max retries before giving up on ECN"); 215 216 VNET_DEFINE(int, tcp_insecure_syn) = 0; 217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(tcp_insecure_syn), 0, 219 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 220 221 VNET_DEFINE(int, tcp_insecure_rst) = 0; 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_insecure_rst), 0, 224 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 225 226 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 227 #define V_tcp_recvspace VNET(tcp_recvspace) 228 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 229 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 230 231 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 232 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 233 &VNET_NAME(tcp_do_autorcvbuf), 0, 234 "Enable automatic receive buffer sizing"); 235 236 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 238 &VNET_NAME(tcp_autorcvbuf_max), 0, 239 "Max size of automatic receive buffer"); 240 241 VNET_DEFINE(struct inpcbhead, tcb); 242 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 243 VNET_DEFINE(struct inpcbinfo, tcbinfo); 244 245 /* 246 * TCP statistics are stored in an array of counter(9)s, which size matches 247 * size of struct tcpstat. TCP running connection count is a regular array. 248 */ 249 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 253 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 254 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 255 "TCP connection counts by TCP state"); 256 257 static void 258 tcp_vnet_init(const void *unused) 259 { 260 261 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 262 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 263 } 264 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 265 tcp_vnet_init, NULL); 266 267 #ifdef VIMAGE 268 static void 269 tcp_vnet_uninit(const void *unused) 270 { 271 272 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 273 VNET_PCPUSTAT_FREE(tcpstat); 274 } 275 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 276 tcp_vnet_uninit, NULL); 277 #endif /* VIMAGE */ 278 279 /* 280 * Kernel module interface for updating tcpstat. The first argument is an index 281 * into tcpstat treated as an array. 282 */ 283 void 284 kmod_tcpstat_add(int statnum, int val) 285 { 286 287 counter_u64_add(VNET(tcpstat)[statnum], val); 288 } 289 290 #ifdef TCP_HHOOK 291 /* 292 * Wrapper for the TCP established input helper hook. 293 */ 294 void 295 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 296 { 297 struct tcp_hhook_data hhook_data; 298 299 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 300 hhook_data.tp = tp; 301 hhook_data.th = th; 302 hhook_data.to = to; 303 304 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 305 tp->osd); 306 } 307 } 308 #endif 309 310 /* 311 * CC wrapper hook functions 312 */ 313 void 314 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 315 uint16_t type) 316 { 317 #ifdef STATS 318 int32_t gput; 319 #endif 320 321 INP_WLOCK_ASSERT(tp->t_inpcb); 322 323 tp->ccv->nsegs = nsegs; 324 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 325 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 326 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 327 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 328 tp->ccv->flags |= CCF_CWND_LIMITED; 329 else 330 tp->ccv->flags &= ~CCF_CWND_LIMITED; 331 332 if (type == CC_ACK) { 333 #ifdef STATS 334 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 335 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 336 if (!IN_RECOVERY(tp->t_flags)) 337 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 338 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 339 if ((tp->t_flags & TF_GPUTINPROG) && 340 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 341 /* 342 * Compute goodput in bits per millisecond. 343 */ 344 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) / 345 max(1, tcp_ts_getticks() - tp->gput_ts); 346 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 347 gput); 348 /* 349 * XXXLAS: This is a temporary hack, and should be 350 * chained off VOI_TCP_GPUT when stats(9) grows an API 351 * to deal with chained VOIs. 352 */ 353 if (tp->t_stats_gput_prev > 0) 354 stats_voi_update_abs_s32(tp->t_stats, 355 VOI_TCP_GPUT_ND, 356 ((gput - tp->t_stats_gput_prev) * 100) / 357 tp->t_stats_gput_prev); 358 tp->t_flags &= ~TF_GPUTINPROG; 359 tp->t_stats_gput_prev = gput; 360 } 361 #endif /* STATS */ 362 if (tp->snd_cwnd > tp->snd_ssthresh) { 363 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 364 if (tp->t_bytes_acked >= tp->snd_cwnd) { 365 tp->t_bytes_acked -= tp->snd_cwnd; 366 tp->ccv->flags |= CCF_ABC_SENTAWND; 367 } 368 } else { 369 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 370 tp->t_bytes_acked = 0; 371 } 372 } 373 374 if (CC_ALGO(tp)->ack_received != NULL) { 375 /* XXXLAS: Find a way to live without this */ 376 tp->ccv->curack = th->th_ack; 377 CC_ALGO(tp)->ack_received(tp->ccv, type); 378 } 379 #ifdef STATS 380 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 381 #endif 382 } 383 384 void 385 cc_conn_init(struct tcpcb *tp) 386 { 387 struct hc_metrics_lite metrics; 388 struct inpcb *inp = tp->t_inpcb; 389 u_int maxseg; 390 int rtt; 391 392 INP_WLOCK_ASSERT(tp->t_inpcb); 393 394 tcp_hc_get(&inp->inp_inc, &metrics); 395 maxseg = tcp_maxseg(tp); 396 397 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 398 tp->t_srtt = rtt; 399 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 400 TCPSTAT_INC(tcps_usedrtt); 401 if (metrics.rmx_rttvar) { 402 tp->t_rttvar = metrics.rmx_rttvar; 403 TCPSTAT_INC(tcps_usedrttvar); 404 } else { 405 /* default variation is +- 1 rtt */ 406 tp->t_rttvar = 407 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 408 } 409 TCPT_RANGESET(tp->t_rxtcur, 410 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 411 tp->t_rttmin, TCPTV_REXMTMAX); 412 } 413 if (metrics.rmx_ssthresh) { 414 /* 415 * There's some sort of gateway or interface 416 * buffer limit on the path. Use this to set 417 * the slow start threshold, but set the 418 * threshold to no less than 2*mss. 419 */ 420 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 421 TCPSTAT_INC(tcps_usedssthresh); 422 } 423 424 /* 425 * Set the initial slow-start flight size. 426 * 427 * If a SYN or SYN/ACK was lost and retransmitted, we have to 428 * reduce the initial CWND to one segment as congestion is likely 429 * requiring us to be cautious. 430 */ 431 if (tp->snd_cwnd == 1) 432 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 433 else 434 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 435 436 if (CC_ALGO(tp)->conn_init != NULL) 437 CC_ALGO(tp)->conn_init(tp->ccv); 438 } 439 440 void inline 441 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 442 { 443 INP_WLOCK_ASSERT(tp->t_inpcb); 444 445 #ifdef STATS 446 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 447 #endif 448 449 switch(type) { 450 case CC_NDUPACK: 451 if (!IN_FASTRECOVERY(tp->t_flags)) { 452 tp->snd_recover = tp->snd_max; 453 if (tp->t_flags2 & TF2_ECN_PERMIT) 454 tp->t_flags2 |= TF2_ECN_SND_CWR; 455 } 456 break; 457 case CC_ECN: 458 if (!IN_CONGRECOVERY(tp->t_flags) || 459 /* 460 * Allow ECN reaction on ACK to CWR, if 461 * that data segment was also CE marked. 462 */ 463 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 464 EXIT_CONGRECOVERY(tp->t_flags); 465 TCPSTAT_INC(tcps_ecn_rcwnd); 466 tp->snd_recover = tp->snd_max + 1; 467 if (tp->t_flags2 & TF2_ECN_PERMIT) 468 tp->t_flags2 |= TF2_ECN_SND_CWR; 469 } 470 break; 471 case CC_RTO: 472 tp->t_dupacks = 0; 473 tp->t_bytes_acked = 0; 474 EXIT_RECOVERY(tp->t_flags); 475 if (tp->t_flags2 & TF2_ECN_PERMIT) 476 tp->t_flags2 |= TF2_ECN_SND_CWR; 477 break; 478 case CC_RTO_ERR: 479 TCPSTAT_INC(tcps_sndrexmitbad); 480 /* RTO was unnecessary, so reset everything. */ 481 tp->snd_cwnd = tp->snd_cwnd_prev; 482 tp->snd_ssthresh = tp->snd_ssthresh_prev; 483 tp->snd_recover = tp->snd_recover_prev; 484 if (tp->t_flags & TF_WASFRECOVERY) 485 ENTER_FASTRECOVERY(tp->t_flags); 486 if (tp->t_flags & TF_WASCRECOVERY) 487 ENTER_CONGRECOVERY(tp->t_flags); 488 tp->snd_nxt = tp->snd_max; 489 tp->t_flags &= ~TF_PREVVALID; 490 tp->t_badrxtwin = 0; 491 break; 492 } 493 494 if (CC_ALGO(tp)->cong_signal != NULL) { 495 if (th != NULL) 496 tp->ccv->curack = th->th_ack; 497 CC_ALGO(tp)->cong_signal(tp->ccv, type); 498 } 499 } 500 501 void inline 502 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 503 { 504 INP_WLOCK_ASSERT(tp->t_inpcb); 505 506 /* XXXLAS: KASSERT that we're in recovery? */ 507 508 if (CC_ALGO(tp)->post_recovery != NULL) { 509 tp->ccv->curack = th->th_ack; 510 CC_ALGO(tp)->post_recovery(tp->ccv); 511 } 512 /* XXXLAS: EXIT_RECOVERY ? */ 513 tp->t_bytes_acked = 0; 514 tp->sackhint.delivered_data = 0; 515 tp->sackhint.prr_out = 0; 516 } 517 518 /* 519 * Indicate whether this ack should be delayed. We can delay the ack if 520 * following conditions are met: 521 * - There is no delayed ack timer in progress. 522 * - Our last ack wasn't a 0-sized window. We never want to delay 523 * the ack that opens up a 0-sized window. 524 * - LRO wasn't used for this segment. We make sure by checking that the 525 * segment size is not larger than the MSS. 526 */ 527 #define DELAY_ACK(tp, tlen) \ 528 ((!tcp_timer_active(tp, TT_DELACK) && \ 529 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 530 (tlen <= tp->t_maxseg) && \ 531 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 532 533 void inline 534 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 535 { 536 INP_WLOCK_ASSERT(tp->t_inpcb); 537 538 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 539 switch (iptos & IPTOS_ECN_MASK) { 540 case IPTOS_ECN_CE: 541 tp->ccv->flags |= CCF_IPHDR_CE; 542 break; 543 case IPTOS_ECN_ECT0: 544 /* FALLTHROUGH */ 545 case IPTOS_ECN_ECT1: 546 /* FALLTHROUGH */ 547 case IPTOS_ECN_NOTECT: 548 tp->ccv->flags &= ~CCF_IPHDR_CE; 549 break; 550 } 551 552 if (flags & TH_CWR) 553 tp->ccv->flags |= CCF_TCPHDR_CWR; 554 else 555 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 556 557 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 558 559 if (tp->ccv->flags & CCF_ACKNOW) { 560 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 561 tp->t_flags |= TF_ACKNOW; 562 } 563 } 564 } 565 566 void inline 567 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 568 { 569 cc_ecnpkt_handler_flags(tp, th->th_flags, iptos); 570 } 571 572 /* 573 * TCP input handling is split into multiple parts: 574 * tcp6_input is a thin wrapper around tcp_input for the extended 575 * ip6_protox[] call format in ip6_input 576 * tcp_input handles primary segment validation, inpcb lookup and 577 * SYN processing on listen sockets 578 * tcp_do_segment processes the ACK and text of the segment for 579 * establishing, established and closing connections 580 */ 581 #ifdef INET6 582 int 583 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 584 { 585 struct mbuf *m; 586 struct in6_ifaddr *ia6; 587 struct ip6_hdr *ip6; 588 589 m = *mp; 590 if (m->m_len < *offp + sizeof(struct tcphdr)) { 591 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 592 if (m == NULL) { 593 *mp = m; 594 TCPSTAT_INC(tcps_rcvshort); 595 return (IPPROTO_DONE); 596 } 597 } 598 599 /* 600 * draft-itojun-ipv6-tcp-to-anycast 601 * better place to put this in? 602 */ 603 ip6 = mtod(m, struct ip6_hdr *); 604 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 605 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 606 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 607 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 608 *mp = NULL; 609 return (IPPROTO_DONE); 610 } 611 612 *mp = m; 613 return (tcp_input_with_port(mp, offp, proto, port)); 614 } 615 616 int 617 tcp6_input(struct mbuf **mp, int *offp, int proto) 618 { 619 620 return(tcp6_input_with_port(mp, offp, proto, 0)); 621 } 622 #endif /* INET6 */ 623 624 int 625 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 626 { 627 struct mbuf *m = *mp; 628 struct tcphdr *th = NULL; 629 struct ip *ip = NULL; 630 struct inpcb *inp = NULL; 631 struct tcpcb *tp = NULL; 632 struct socket *so = NULL; 633 u_char *optp = NULL; 634 int off0; 635 int optlen = 0; 636 #ifdef INET 637 int len; 638 uint8_t ipttl; 639 #endif 640 int tlen = 0, off; 641 int drop_hdrlen; 642 int thflags; 643 int rstreason = 0; /* For badport_bandlim accounting purposes */ 644 int lookupflag; 645 uint8_t iptos; 646 struct m_tag *fwd_tag = NULL; 647 #ifdef INET6 648 struct ip6_hdr *ip6 = NULL; 649 int isipv6; 650 #else 651 const void *ip6 = NULL; 652 #endif /* INET6 */ 653 struct tcpopt to; /* options in this segment */ 654 char *s = NULL; /* address and port logging */ 655 #ifdef TCPDEBUG 656 /* 657 * The size of tcp_saveipgen must be the size of the max ip header, 658 * now IPv6. 659 */ 660 u_char tcp_saveipgen[IP6_HDR_LEN]; 661 struct tcphdr tcp_savetcp; 662 short ostate = 0; 663 #endif 664 665 NET_EPOCH_ASSERT(); 666 667 #ifdef INET6 668 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 669 #endif 670 671 off0 = *offp; 672 m = *mp; 673 *mp = NULL; 674 to.to_flags = 0; 675 TCPSTAT_INC(tcps_rcvtotal); 676 677 #ifdef INET6 678 if (isipv6) { 679 ip6 = mtod(m, struct ip6_hdr *); 680 th = (struct tcphdr *)((caddr_t)ip6 + off0); 681 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 682 if (port) 683 goto skip6_csum; 684 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 685 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 686 th->th_sum = m->m_pkthdr.csum_data; 687 else 688 th->th_sum = in6_cksum_pseudo(ip6, tlen, 689 IPPROTO_TCP, m->m_pkthdr.csum_data); 690 th->th_sum ^= 0xffff; 691 } else 692 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 693 if (th->th_sum) { 694 TCPSTAT_INC(tcps_rcvbadsum); 695 goto drop; 696 } 697 skip6_csum: 698 /* 699 * Be proactive about unspecified IPv6 address in source. 700 * As we use all-zero to indicate unbounded/unconnected pcb, 701 * unspecified IPv6 address can be used to confuse us. 702 * 703 * Note that packets with unspecified IPv6 destination is 704 * already dropped in ip6_input. 705 */ 706 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 707 /* XXX stat */ 708 goto drop; 709 } 710 iptos = IPV6_TRAFFIC_CLASS(ip6); 711 } 712 #endif 713 #if defined(INET) && defined(INET6) 714 else 715 #endif 716 #ifdef INET 717 { 718 /* 719 * Get IP and TCP header together in first mbuf. 720 * Note: IP leaves IP header in first mbuf. 721 */ 722 if (off0 > sizeof (struct ip)) { 723 ip_stripoptions(m); 724 off0 = sizeof(struct ip); 725 } 726 if (m->m_len < sizeof (struct tcpiphdr)) { 727 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 728 == NULL) { 729 TCPSTAT_INC(tcps_rcvshort); 730 return (IPPROTO_DONE); 731 } 732 } 733 ip = mtod(m, struct ip *); 734 th = (struct tcphdr *)((caddr_t)ip + off0); 735 tlen = ntohs(ip->ip_len) - off0; 736 737 iptos = ip->ip_tos; 738 if (port) 739 goto skip_csum; 740 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 741 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 742 th->th_sum = m->m_pkthdr.csum_data; 743 else 744 th->th_sum = in_pseudo(ip->ip_src.s_addr, 745 ip->ip_dst.s_addr, 746 htonl(m->m_pkthdr.csum_data + tlen + 747 IPPROTO_TCP)); 748 th->th_sum ^= 0xffff; 749 } else { 750 struct ipovly *ipov = (struct ipovly *)ip; 751 752 /* 753 * Checksum extended TCP header and data. 754 */ 755 len = off0 + tlen; 756 ipttl = ip->ip_ttl; 757 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 758 ipov->ih_len = htons(tlen); 759 th->th_sum = in_cksum(m, len); 760 /* Reset length for SDT probes. */ 761 ip->ip_len = htons(len); 762 /* Reset TOS bits */ 763 ip->ip_tos = iptos; 764 /* Re-initialization for later version check */ 765 ip->ip_ttl = ipttl; 766 ip->ip_v = IPVERSION; 767 ip->ip_hl = off0 >> 2; 768 } 769 skip_csum: 770 if (th->th_sum && (port == 0)) { 771 TCPSTAT_INC(tcps_rcvbadsum); 772 goto drop; 773 } 774 } 775 #endif /* INET */ 776 777 /* 778 * Check that TCP offset makes sense, 779 * pull out TCP options and adjust length. XXX 780 */ 781 off = th->th_off << 2; 782 if (off < sizeof (struct tcphdr) || off > tlen) { 783 TCPSTAT_INC(tcps_rcvbadoff); 784 goto drop; 785 } 786 tlen -= off; /* tlen is used instead of ti->ti_len */ 787 if (off > sizeof (struct tcphdr)) { 788 #ifdef INET6 789 if (isipv6) { 790 if (m->m_len < off0 + off) { 791 m = m_pullup(m, off0 + off); 792 if (m == NULL) { 793 TCPSTAT_INC(tcps_rcvshort); 794 return (IPPROTO_DONE); 795 } 796 } 797 ip6 = mtod(m, struct ip6_hdr *); 798 th = (struct tcphdr *)((caddr_t)ip6 + off0); 799 } 800 #endif 801 #if defined(INET) && defined(INET6) 802 else 803 #endif 804 #ifdef INET 805 { 806 if (m->m_len < sizeof(struct ip) + off) { 807 if ((m = m_pullup(m, sizeof (struct ip) + off)) 808 == NULL) { 809 TCPSTAT_INC(tcps_rcvshort); 810 return (IPPROTO_DONE); 811 } 812 ip = mtod(m, struct ip *); 813 th = (struct tcphdr *)((caddr_t)ip + off0); 814 } 815 } 816 #endif 817 optlen = off - sizeof (struct tcphdr); 818 optp = (u_char *)(th + 1); 819 } 820 thflags = th->th_flags; 821 822 /* 823 * Convert TCP protocol specific fields to host format. 824 */ 825 tcp_fields_to_host(th); 826 827 /* 828 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 829 */ 830 drop_hdrlen = off0 + off; 831 832 /* 833 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 834 */ 835 if ( 836 #ifdef INET6 837 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 838 #ifdef INET 839 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 840 #endif 841 #endif 842 #if defined(INET) && !defined(INET6) 843 (m->m_flags & M_IP_NEXTHOP) 844 #endif 845 ) 846 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 847 848 /* 849 * For initial SYN packets we don't need write lock on matching 850 * PCB, be it a listening one or a synchronized one. The packet 851 * shall not modify its state. 852 */ 853 lookupflag = (thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 854 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB; 855 findpcb: 856 #ifdef INET6 857 if (isipv6 && fwd_tag != NULL) { 858 struct sockaddr_in6 *next_hop6; 859 860 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 861 /* 862 * Transparently forwarded. Pretend to be the destination. 863 * Already got one like this? 864 */ 865 inp = in6_pcblookup_mbuf(&V_tcbinfo, 866 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 867 lookupflag, m->m_pkthdr.rcvif, m); 868 if (!inp) { 869 /* 870 * It's new. Try to find the ambushing socket. 871 * Because we've rewritten the destination address, 872 * any hardware-generated hash is ignored. 873 */ 874 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 875 th->th_sport, &next_hop6->sin6_addr, 876 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 877 th->th_dport, INPLOOKUP_WILDCARD | lookupflag, 878 m->m_pkthdr.rcvif); 879 } 880 } else if (isipv6) { 881 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 882 th->th_sport, &ip6->ip6_dst, th->th_dport, 883 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m); 884 } 885 #endif /* INET6 */ 886 #if defined(INET6) && defined(INET) 887 else 888 #endif 889 #ifdef INET 890 if (fwd_tag != NULL) { 891 struct sockaddr_in *next_hop; 892 893 next_hop = (struct sockaddr_in *)(fwd_tag+1); 894 /* 895 * Transparently forwarded. Pretend to be the destination. 896 * already got one like this? 897 */ 898 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 899 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); 900 if (!inp) { 901 /* 902 * It's new. Try to find the ambushing socket. 903 * Because we've rewritten the destination address, 904 * any hardware-generated hash is ignored. 905 */ 906 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 907 th->th_sport, next_hop->sin_addr, 908 next_hop->sin_port ? ntohs(next_hop->sin_port) : 909 th->th_dport, INPLOOKUP_WILDCARD | lookupflag, 910 m->m_pkthdr.rcvif); 911 } 912 } else 913 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 914 th->th_sport, ip->ip_dst, th->th_dport, 915 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m); 916 #endif /* INET */ 917 918 /* 919 * If the INPCB does not exist then all data in the incoming 920 * segment is discarded and an appropriate RST is sent back. 921 * XXX MRT Send RST using which routing table? 922 */ 923 if (inp == NULL) { 924 /* 925 * Log communication attempts to ports that are not 926 * in use. 927 */ 928 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 929 V_tcp_log_in_vain == 2) { 930 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 931 log(LOG_INFO, "%s; %s: Connection attempt " 932 "to closed port\n", s, __func__); 933 } 934 /* 935 * When blackholing do not respond with a RST but 936 * completely ignore the segment and drop it. 937 */ 938 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 939 V_blackhole == 2) 940 goto dropunlock; 941 942 rstreason = BANDLIM_RST_CLOSEDPORT; 943 goto dropwithreset; 944 } 945 INP_LOCK_ASSERT(inp); 946 /* 947 * While waiting for inp lock during the lookup, another thread 948 * can have dropped the inpcb, in which case we need to loop back 949 * and try to find a new inpcb to deliver to. 950 */ 951 if (inp->inp_flags & INP_DROPPED) { 952 INP_UNLOCK(inp); 953 inp = NULL; 954 goto findpcb; 955 } 956 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 957 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 958 ((inp->inp_socket == NULL) || 959 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 960 inp->inp_flowid = m->m_pkthdr.flowid; 961 inp->inp_flowtype = M_HASHTYPE_GET(m); 962 } 963 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 964 #ifdef INET6 965 if (isipv6 && IPSEC_ENABLED(ipv6) && 966 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 967 goto dropunlock; 968 } 969 #ifdef INET 970 else 971 #endif 972 #endif /* INET6 */ 973 #ifdef INET 974 if (IPSEC_ENABLED(ipv4) && 975 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 976 goto dropunlock; 977 } 978 #endif /* INET */ 979 #endif /* IPSEC */ 980 981 /* 982 * Check the minimum TTL for socket. 983 */ 984 if (inp->inp_ip_minttl != 0) { 985 #ifdef INET6 986 if (isipv6) { 987 if (inp->inp_ip_minttl > ip6->ip6_hlim) 988 goto dropunlock; 989 } else 990 #endif 991 if (inp->inp_ip_minttl > ip->ip_ttl) 992 goto dropunlock; 993 } 994 995 /* 996 * A previous connection in TIMEWAIT state is supposed to catch stray 997 * or duplicate segments arriving late. If this segment was a 998 * legitimate new connection attempt, the old INPCB gets removed and 999 * we can try again to find a listening socket. 1000 */ 1001 if (inp->inp_flags & INP_TIMEWAIT) { 1002 tcp_dooptions(&to, optp, optlen, 1003 (thflags & TH_SYN) ? TO_SYN : 0); 1004 /* 1005 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1006 */ 1007 if (tcp_twcheck(inp, &to, th, m, tlen)) 1008 goto findpcb; 1009 return (IPPROTO_DONE); 1010 } 1011 /* 1012 * The TCPCB may no longer exist if the connection is winding 1013 * down or it is in the CLOSED state. Either way we drop the 1014 * segment and send an appropriate response. 1015 */ 1016 tp = intotcpcb(inp); 1017 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1018 rstreason = BANDLIM_RST_CLOSEDPORT; 1019 goto dropwithreset; 1020 } 1021 1022 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1023 rstreason = BANDLIM_RST_CLOSEDPORT; 1024 goto dropwithreset; 1025 } 1026 1027 #ifdef TCP_OFFLOAD 1028 if (tp->t_flags & TF_TOE) { 1029 tcp_offload_input(tp, m); 1030 m = NULL; /* consumed by the TOE driver */ 1031 goto dropunlock; 1032 } 1033 #endif 1034 1035 #ifdef MAC 1036 if (mac_inpcb_check_deliver(inp, m)) 1037 goto dropunlock; 1038 #endif 1039 so = inp->inp_socket; 1040 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1041 #ifdef TCPDEBUG 1042 if (so->so_options & SO_DEBUG) { 1043 ostate = tp->t_state; 1044 #ifdef INET6 1045 if (isipv6) { 1046 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1047 } else 1048 #endif 1049 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1050 tcp_savetcp = *th; 1051 } 1052 #endif /* TCPDEBUG */ 1053 /* 1054 * When the socket is accepting connections (the INPCB is in LISTEN 1055 * state) we look into the SYN cache if this is a new connection 1056 * attempt or the completion of a previous one. 1057 */ 1058 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1059 ("%s: so accepting but tp %p not listening", __func__, tp)); 1060 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1061 struct in_conninfo inc; 1062 1063 bzero(&inc, sizeof(inc)); 1064 #ifdef INET6 1065 if (isipv6) { 1066 inc.inc_flags |= INC_ISIPV6; 1067 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1068 inc.inc_flags |= INC_IPV6MINMTU; 1069 inc.inc6_faddr = ip6->ip6_src; 1070 inc.inc6_laddr = ip6->ip6_dst; 1071 } else 1072 #endif 1073 { 1074 inc.inc_faddr = ip->ip_src; 1075 inc.inc_laddr = ip->ip_dst; 1076 } 1077 inc.inc_fport = th->th_sport; 1078 inc.inc_lport = th->th_dport; 1079 inc.inc_fibnum = so->so_fibnum; 1080 1081 /* 1082 * Check for an existing connection attempt in syncache if 1083 * the flag is only ACK. A successful lookup creates a new 1084 * socket appended to the listen queue in SYN_RECEIVED state. 1085 */ 1086 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1087 /* 1088 * Parse the TCP options here because 1089 * syncookies need access to the reflected 1090 * timestamp. 1091 */ 1092 tcp_dooptions(&to, optp, optlen, 0); 1093 /* 1094 * NB: syncache_expand() doesn't unlock 1095 * inp and tcpinfo locks. 1096 */ 1097 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1098 if (rstreason < 0) { 1099 /* 1100 * A failing TCP MD5 signature comparison 1101 * must result in the segment being dropped 1102 * and must not produce any response back 1103 * to the sender. 1104 */ 1105 goto dropunlock; 1106 } else if (rstreason == 0) { 1107 /* 1108 * No syncache entry or ACK was not 1109 * for our SYN/ACK. Send a RST. 1110 * NB: syncache did its own logging 1111 * of the failure cause. 1112 */ 1113 rstreason = BANDLIM_RST_OPENPORT; 1114 goto dropwithreset; 1115 } 1116 tfo_socket_result: 1117 if (so == NULL) { 1118 /* 1119 * We completed the 3-way handshake 1120 * but could not allocate a socket 1121 * either due to memory shortage, 1122 * listen queue length limits or 1123 * global socket limits. Send RST 1124 * or wait and have the remote end 1125 * retransmit the ACK for another 1126 * try. 1127 */ 1128 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1129 log(LOG_DEBUG, "%s; %s: Listen socket: " 1130 "Socket allocation failed due to " 1131 "limits or memory shortage, %s\n", 1132 s, __func__, 1133 V_tcp_sc_rst_sock_fail ? 1134 "sending RST" : "try again"); 1135 if (V_tcp_sc_rst_sock_fail) { 1136 rstreason = BANDLIM_UNLIMITED; 1137 goto dropwithreset; 1138 } else 1139 goto dropunlock; 1140 } 1141 /* 1142 * Socket is created in state SYN_RECEIVED. 1143 * Unlock the listen socket, lock the newly 1144 * created socket and update the tp variable. 1145 * If we came here via jump to tfo_socket_result, 1146 * then listening socket is read-locked. 1147 */ 1148 INP_UNLOCK(inp); /* listen socket */ 1149 inp = sotoinpcb(so); 1150 /* 1151 * New connection inpcb is already locked by 1152 * syncache_expand(). 1153 */ 1154 INP_WLOCK_ASSERT(inp); 1155 tp = intotcpcb(inp); 1156 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1157 ("%s: ", __func__)); 1158 /* 1159 * Process the segment and the data it 1160 * contains. tcp_do_segment() consumes 1161 * the mbuf chain and unlocks the inpcb. 1162 */ 1163 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1164 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1165 iptos); 1166 return (IPPROTO_DONE); 1167 } 1168 /* 1169 * Segment flag validation for new connection attempts: 1170 * 1171 * Our (SYN|ACK) response was rejected. 1172 * Check with syncache and remove entry to prevent 1173 * retransmits. 1174 * 1175 * NB: syncache_chkrst does its own logging of failure 1176 * causes. 1177 */ 1178 if (thflags & TH_RST) { 1179 syncache_chkrst(&inc, th, m, port); 1180 goto dropunlock; 1181 } 1182 /* 1183 * We can't do anything without SYN. 1184 */ 1185 if ((thflags & TH_SYN) == 0) { 1186 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1187 log(LOG_DEBUG, "%s; %s: Listen socket: " 1188 "SYN is missing, segment ignored\n", 1189 s, __func__); 1190 TCPSTAT_INC(tcps_badsyn); 1191 goto dropunlock; 1192 } 1193 /* 1194 * (SYN|ACK) is bogus on a listen socket. 1195 */ 1196 if (thflags & TH_ACK) { 1197 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1198 log(LOG_DEBUG, "%s; %s: Listen socket: " 1199 "SYN|ACK invalid, segment rejected\n", 1200 s, __func__); 1201 syncache_badack(&inc, port); /* XXX: Not needed! */ 1202 TCPSTAT_INC(tcps_badsyn); 1203 rstreason = BANDLIM_RST_OPENPORT; 1204 goto dropwithreset; 1205 } 1206 /* 1207 * If the drop_synfin option is enabled, drop all 1208 * segments with both the SYN and FIN bits set. 1209 * This prevents e.g. nmap from identifying the 1210 * TCP/IP stack. 1211 * XXX: Poor reasoning. nmap has other methods 1212 * and is constantly refining its stack detection 1213 * strategies. 1214 * XXX: This is a violation of the TCP specification 1215 * and was used by RFC1644. 1216 */ 1217 if ((thflags & TH_FIN) && V_drop_synfin) { 1218 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1219 log(LOG_DEBUG, "%s; %s: Listen socket: " 1220 "SYN|FIN segment ignored (based on " 1221 "sysctl setting)\n", s, __func__); 1222 TCPSTAT_INC(tcps_badsyn); 1223 goto dropunlock; 1224 } 1225 /* 1226 * Segment's flags are (SYN) or (SYN|FIN). 1227 * 1228 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1229 * as they do not affect the state of the TCP FSM. 1230 * The data pointed to by TH_URG and th_urp is ignored. 1231 */ 1232 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1233 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1234 KASSERT(thflags & (TH_SYN), 1235 ("%s: Listen socket: TH_SYN not set", __func__)); 1236 INP_RLOCK_ASSERT(inp); 1237 #ifdef INET6 1238 /* 1239 * If deprecated address is forbidden, 1240 * we do not accept SYN to deprecated interface 1241 * address to prevent any new inbound connection from 1242 * getting established. 1243 * When we do not accept SYN, we send a TCP RST, 1244 * with deprecated source address (instead of dropping 1245 * it). We compromise it as it is much better for peer 1246 * to send a RST, and RST will be the final packet 1247 * for the exchange. 1248 * 1249 * If we do not forbid deprecated addresses, we accept 1250 * the SYN packet. RFC2462 does not suggest dropping 1251 * SYN in this case. 1252 * If we decipher RFC2462 5.5.4, it says like this: 1253 * 1. use of deprecated addr with existing 1254 * communication is okay - "SHOULD continue to be 1255 * used" 1256 * 2. use of it with new communication: 1257 * (2a) "SHOULD NOT be used if alternate address 1258 * with sufficient scope is available" 1259 * (2b) nothing mentioned otherwise. 1260 * Here we fall into (2b) case as we have no choice in 1261 * our source address selection - we must obey the peer. 1262 * 1263 * The wording in RFC2462 is confusing, and there are 1264 * multiple description text for deprecated address 1265 * handling - worse, they are not exactly the same. 1266 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1267 */ 1268 if (isipv6 && !V_ip6_use_deprecated) { 1269 struct in6_ifaddr *ia6; 1270 1271 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1272 if (ia6 != NULL && 1273 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1274 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1275 log(LOG_DEBUG, "%s; %s: Listen socket: " 1276 "Connection attempt to deprecated " 1277 "IPv6 address rejected\n", 1278 s, __func__); 1279 rstreason = BANDLIM_RST_OPENPORT; 1280 goto dropwithreset; 1281 } 1282 } 1283 #endif /* INET6 */ 1284 /* 1285 * Basic sanity checks on incoming SYN requests: 1286 * Don't respond if the destination is a link layer 1287 * broadcast according to RFC1122 4.2.3.10, p. 104. 1288 * If it is from this socket it must be forged. 1289 * Don't respond if the source or destination is a 1290 * global or subnet broad- or multicast address. 1291 * Note that it is quite possible to receive unicast 1292 * link-layer packets with a broadcast IP address. Use 1293 * in_broadcast() to find them. 1294 */ 1295 if (m->m_flags & (M_BCAST|M_MCAST)) { 1296 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1297 log(LOG_DEBUG, "%s; %s: Listen socket: " 1298 "Connection attempt from broad- or multicast " 1299 "link layer address ignored\n", s, __func__); 1300 goto dropunlock; 1301 } 1302 #ifdef INET6 1303 if (isipv6) { 1304 if (th->th_dport == th->th_sport && 1305 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1306 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1307 log(LOG_DEBUG, "%s; %s: Listen socket: " 1308 "Connection attempt to/from self " 1309 "ignored\n", s, __func__); 1310 goto dropunlock; 1311 } 1312 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1313 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1314 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1315 log(LOG_DEBUG, "%s; %s: Listen socket: " 1316 "Connection attempt from/to multicast " 1317 "address ignored\n", s, __func__); 1318 goto dropunlock; 1319 } 1320 } 1321 #endif 1322 #if defined(INET) && defined(INET6) 1323 else 1324 #endif 1325 #ifdef INET 1326 { 1327 if (th->th_dport == th->th_sport && 1328 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1329 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1330 log(LOG_DEBUG, "%s; %s: Listen socket: " 1331 "Connection attempt from/to self " 1332 "ignored\n", s, __func__); 1333 goto dropunlock; 1334 } 1335 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1336 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1337 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1338 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1339 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1340 log(LOG_DEBUG, "%s; %s: Listen socket: " 1341 "Connection attempt from/to broad- " 1342 "or multicast address ignored\n", 1343 s, __func__); 1344 goto dropunlock; 1345 } 1346 } 1347 #endif 1348 /* 1349 * SYN appears to be valid. Create compressed TCP state 1350 * for syncache. 1351 */ 1352 #ifdef TCPDEBUG 1353 if (so->so_options & SO_DEBUG) 1354 tcp_trace(TA_INPUT, ostate, tp, 1355 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1356 #endif 1357 TCP_PROBE3(debug__input, tp, th, m); 1358 tcp_dooptions(&to, optp, optlen, TO_SYN); 1359 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1360 iptos, port)) != NULL) 1361 goto tfo_socket_result; 1362 1363 /* 1364 * Entry added to syncache and mbuf consumed. 1365 * Only the listen socket is unlocked by syncache_add(). 1366 */ 1367 return (IPPROTO_DONE); 1368 } else if (tp->t_state == TCPS_LISTEN) { 1369 /* 1370 * When a listen socket is torn down the SO_ACCEPTCONN 1371 * flag is removed first while connections are drained 1372 * from the accept queue in a unlock/lock cycle of the 1373 * ACCEPT_LOCK, opening a race condition allowing a SYN 1374 * attempt go through unhandled. 1375 */ 1376 goto dropunlock; 1377 } 1378 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1379 if (tp->t_flags & TF_SIGNATURE) { 1380 tcp_dooptions(&to, optp, optlen, thflags); 1381 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1382 TCPSTAT_INC(tcps_sig_err_nosigopt); 1383 goto dropunlock; 1384 } 1385 if (!TCPMD5_ENABLED() || 1386 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1387 goto dropunlock; 1388 } 1389 #endif 1390 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1391 1392 /* 1393 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1394 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1395 * the inpcb, and unlocks pcbinfo. 1396 * 1397 * XXXGL: in case of a pure SYN arriving on existing connection 1398 * TCP stacks won't need to modify the PCB, they would either drop 1399 * the segment silently, or send a challenge ACK. However, we try 1400 * to upgrade the lock, because calling convention for stacks is 1401 * write-lock on PCB. If upgrade fails, drop the SYN. 1402 */ 1403 if (lookupflag == INPLOOKUP_RLOCKPCB && INP_TRY_UPGRADE(inp) == 0) 1404 goto dropunlock; 1405 1406 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1407 return (IPPROTO_DONE); 1408 1409 dropwithreset: 1410 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1411 1412 if (inp != NULL) { 1413 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1414 INP_UNLOCK(inp); 1415 } else 1416 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1417 m = NULL; /* mbuf chain got consumed. */ 1418 goto drop; 1419 1420 dropunlock: 1421 if (m != NULL) 1422 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1423 1424 if (inp != NULL) 1425 INP_UNLOCK(inp); 1426 1427 drop: 1428 if (s != NULL) 1429 free(s, M_TCPLOG); 1430 if (m != NULL) 1431 m_freem(m); 1432 return (IPPROTO_DONE); 1433 } 1434 1435 /* 1436 * Automatic sizing of receive socket buffer. Often the send 1437 * buffer size is not optimally adjusted to the actual network 1438 * conditions at hand (delay bandwidth product). Setting the 1439 * buffer size too small limits throughput on links with high 1440 * bandwidth and high delay (eg. trans-continental/oceanic links). 1441 * 1442 * On the receive side the socket buffer memory is only rarely 1443 * used to any significant extent. This allows us to be much 1444 * more aggressive in scaling the receive socket buffer. For 1445 * the case that the buffer space is actually used to a large 1446 * extent and we run out of kernel memory we can simply drop 1447 * the new segments; TCP on the sender will just retransmit it 1448 * later. Setting the buffer size too big may only consume too 1449 * much kernel memory if the application doesn't read() from 1450 * the socket or packet loss or reordering makes use of the 1451 * reassembly queue. 1452 * 1453 * The criteria to step up the receive buffer one notch are: 1454 * 1. Application has not set receive buffer size with 1455 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1456 * 2. the number of bytes received during 1/2 of an sRTT 1457 * is at least 3/8 of the current socket buffer size. 1458 * 3. receive buffer size has not hit maximal automatic size; 1459 * 1460 * If all of the criteria are met we increaset the socket buffer 1461 * by a 1/2 (bounded by the max). This allows us to keep ahead 1462 * of slow-start but also makes it so our peer never gets limited 1463 * by our rwnd which we then open up causing a burst. 1464 * 1465 * This algorithm does two steps per RTT at most and only if 1466 * we receive a bulk stream w/o packet losses or reorderings. 1467 * Shrinking the buffer during idle times is not necessary as 1468 * it doesn't consume any memory when idle. 1469 * 1470 * TODO: Only step up if the application is actually serving 1471 * the buffer to better manage the socket buffer resources. 1472 */ 1473 int 1474 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1475 struct tcpcb *tp, int tlen) 1476 { 1477 int newsize = 0; 1478 1479 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1480 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1481 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1482 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1483 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1484 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1485 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1486 } 1487 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1488 1489 /* Start over with next RTT. */ 1490 tp->rfbuf_ts = 0; 1491 tp->rfbuf_cnt = 0; 1492 } else { 1493 tp->rfbuf_cnt += tlen; /* add up */ 1494 } 1495 return (newsize); 1496 } 1497 1498 int 1499 tcp_input(struct mbuf **mp, int *offp, int proto) 1500 { 1501 return(tcp_input_with_port(mp, offp, proto, 0)); 1502 } 1503 1504 void 1505 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1506 { 1507 /* 1508 * Since tp might be gone if the session entered 1509 * the TIME_WAIT state before coming here, we need 1510 * to check if the socket is still connected. 1511 */ 1512 if ((so->so_state & SS_ISCONNECTED) == 0) 1513 return; 1514 INP_LOCK_ASSERT(tp->t_inpcb); 1515 if (tp->t_flags & TF_WAKESOR) { 1516 tp->t_flags &= ~TF_WAKESOR; 1517 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 1518 sorwakeup(so); 1519 } 1520 if (tp->t_flags & TF_WAKESOW) { 1521 tp->t_flags &= ~TF_WAKESOW; 1522 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 1523 sowwakeup(so); 1524 } 1525 } 1526 1527 void 1528 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1529 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1530 { 1531 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1532 int rstreason, todrop, win, incforsyn = 0; 1533 uint32_t tiwin; 1534 uint16_t nsegs; 1535 char *s; 1536 struct in_conninfo *inc; 1537 struct mbuf *mfree; 1538 struct tcpopt to; 1539 int tfo_syn; 1540 u_int maxseg; 1541 1542 #ifdef TCPDEBUG 1543 /* 1544 * The size of tcp_saveipgen must be the size of the max ip header, 1545 * now IPv6. 1546 */ 1547 u_char tcp_saveipgen[IP6_HDR_LEN]; 1548 struct tcphdr tcp_savetcp; 1549 short ostate = 0; 1550 #endif 1551 thflags = th->th_flags; 1552 inc = &tp->t_inpcb->inp_inc; 1553 tp->sackhint.last_sack_ack = 0; 1554 sack_changed = 0; 1555 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1556 1557 NET_EPOCH_ASSERT(); 1558 INP_WLOCK_ASSERT(tp->t_inpcb); 1559 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1560 __func__)); 1561 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1562 __func__)); 1563 1564 #ifdef TCPPCAP 1565 /* Save segment, if requested. */ 1566 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1567 #endif 1568 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1569 tlen, NULL, true); 1570 1571 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1572 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1573 log(LOG_DEBUG, "%s; %s: " 1574 "SYN|FIN segment ignored (based on " 1575 "sysctl setting)\n", s, __func__); 1576 free(s, M_TCPLOG); 1577 } 1578 goto drop; 1579 } 1580 1581 /* 1582 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1583 * check SEQ.ACK first. 1584 */ 1585 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1586 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1587 rstreason = BANDLIM_UNLIMITED; 1588 goto dropwithreset; 1589 } 1590 1591 /* 1592 * Segment received on connection. 1593 * Reset idle time and keep-alive timer. 1594 * XXX: This should be done after segment 1595 * validation to ignore broken/spoofed segs. 1596 */ 1597 tp->t_rcvtime = ticks; 1598 1599 /* 1600 * Scale up the window into a 32-bit value. 1601 * For the SYN_SENT state the scale is zero. 1602 */ 1603 tiwin = th->th_win << tp->snd_scale; 1604 #ifdef STATS 1605 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1606 #endif 1607 1608 /* 1609 * TCP ECN processing. 1610 */ 1611 if (tp->t_flags2 & TF2_ECN_PERMIT) { 1612 if (thflags & TH_CWR) { 1613 tp->t_flags2 &= ~TF2_ECN_SND_ECE; 1614 tp->t_flags |= TF_ACKNOW; 1615 } 1616 switch (iptos & IPTOS_ECN_MASK) { 1617 case IPTOS_ECN_CE: 1618 tp->t_flags2 |= TF2_ECN_SND_ECE; 1619 TCPSTAT_INC(tcps_ecn_ce); 1620 break; 1621 case IPTOS_ECN_ECT0: 1622 TCPSTAT_INC(tcps_ecn_ect0); 1623 break; 1624 case IPTOS_ECN_ECT1: 1625 TCPSTAT_INC(tcps_ecn_ect1); 1626 break; 1627 } 1628 1629 /* Process a packet differently from RFC3168. */ 1630 cc_ecnpkt_handler(tp, th, iptos); 1631 1632 /* Congestion experienced. */ 1633 if (thflags & TH_ECE) { 1634 cc_cong_signal(tp, th, CC_ECN); 1635 } 1636 } 1637 1638 /* 1639 * Parse options on any incoming segment. 1640 */ 1641 tcp_dooptions(&to, (u_char *)(th + 1), 1642 (th->th_off << 2) - sizeof(struct tcphdr), 1643 (thflags & TH_SYN) ? TO_SYN : 0); 1644 1645 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1646 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1647 (to.to_flags & TOF_SIGNATURE) == 0) { 1648 TCPSTAT_INC(tcps_sig_err_sigopt); 1649 /* XXX: should drop? */ 1650 } 1651 #endif 1652 /* 1653 * If echoed timestamp is later than the current time, 1654 * fall back to non RFC1323 RTT calculation. Normalize 1655 * timestamp if syncookies were used when this connection 1656 * was established. 1657 */ 1658 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1659 to.to_tsecr -= tp->ts_offset; 1660 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1661 to.to_tsecr = 0; 1662 else if (tp->t_flags & TF_PREVVALID && 1663 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1664 cc_cong_signal(tp, th, CC_RTO_ERR); 1665 } 1666 /* 1667 * Process options only when we get SYN/ACK back. The SYN case 1668 * for incoming connections is handled in tcp_syncache. 1669 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1670 * or <SYN,ACK>) segment itself is never scaled. 1671 * XXX this is traditional behavior, may need to be cleaned up. 1672 */ 1673 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1674 /* Handle parallel SYN for ECN */ 1675 if (!(thflags & TH_ACK) && 1676 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) && 1677 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) { 1678 tp->t_flags2 |= TF2_ECN_PERMIT; 1679 tp->t_flags2 |= TF2_ECN_SND_ECE; 1680 TCPSTAT_INC(tcps_ecn_shs); 1681 } 1682 if ((to.to_flags & TOF_SCALE) && 1683 (tp->t_flags & TF_REQ_SCALE) && 1684 !(tp->t_flags & TF_NOOPT)) { 1685 tp->t_flags |= TF_RCVD_SCALE; 1686 tp->snd_scale = to.to_wscale; 1687 } else 1688 tp->t_flags &= ~TF_REQ_SCALE; 1689 /* 1690 * Initial send window. It will be updated with 1691 * the next incoming segment to the scaled value. 1692 */ 1693 tp->snd_wnd = th->th_win; 1694 if ((to.to_flags & TOF_TS) && 1695 (tp->t_flags & TF_REQ_TSTMP) && 1696 !(tp->t_flags & TF_NOOPT)) { 1697 tp->t_flags |= TF_RCVD_TSTMP; 1698 tp->ts_recent = to.to_tsval; 1699 tp->ts_recent_age = tcp_ts_getticks(); 1700 } else 1701 tp->t_flags &= ~TF_REQ_TSTMP; 1702 if (to.to_flags & TOF_MSS) 1703 tcp_mss(tp, to.to_mss); 1704 if ((tp->t_flags & TF_SACK_PERMIT) && 1705 (!(to.to_flags & TOF_SACKPERM) || 1706 (tp->t_flags & TF_NOOPT))) 1707 tp->t_flags &= ~TF_SACK_PERMIT; 1708 if (IS_FASTOPEN(tp->t_flags)) { 1709 if ((to.to_flags & TOF_FASTOPEN) && 1710 !(tp->t_flags & TF_NOOPT)) { 1711 uint16_t mss; 1712 1713 if (to.to_flags & TOF_MSS) 1714 mss = to.to_mss; 1715 else 1716 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1717 mss = TCP6_MSS; 1718 else 1719 mss = TCP_MSS; 1720 tcp_fastopen_update_cache(tp, mss, 1721 to.to_tfo_len, to.to_tfo_cookie); 1722 } else 1723 tcp_fastopen_disable_path(tp); 1724 } 1725 } 1726 1727 /* 1728 * If timestamps were negotiated during SYN/ACK and a 1729 * segment without a timestamp is received, silently drop 1730 * the segment, unless it is a RST segment or missing timestamps are 1731 * tolerated. 1732 * See section 3.2 of RFC 7323. 1733 */ 1734 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1735 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1736 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1737 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1738 "segment processed normally\n", 1739 s, __func__); 1740 free(s, M_TCPLOG); 1741 } 1742 } else { 1743 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1744 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1745 "segment silently dropped\n", s, __func__); 1746 free(s, M_TCPLOG); 1747 } 1748 goto drop; 1749 } 1750 } 1751 /* 1752 * If timestamps were not negotiated during SYN/ACK and a 1753 * segment with a timestamp is received, ignore the 1754 * timestamp and process the packet normally. 1755 * See section 3.2 of RFC 7323. 1756 */ 1757 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1758 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1759 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1760 "segment processed normally\n", s, __func__); 1761 free(s, M_TCPLOG); 1762 } 1763 } 1764 1765 /* 1766 * Header prediction: check for the two common cases 1767 * of a uni-directional data xfer. If the packet has 1768 * no control flags, is in-sequence, the window didn't 1769 * change and we're not retransmitting, it's a 1770 * candidate. If the length is zero and the ack moved 1771 * forward, we're the sender side of the xfer. Just 1772 * free the data acked & wake any higher level process 1773 * that was blocked waiting for space. If the length 1774 * is non-zero and the ack didn't move, we're the 1775 * receiver side. If we're getting packets in-order 1776 * (the reassembly queue is empty), add the data to 1777 * the socket buffer and note that we need a delayed ack. 1778 * Make sure that the hidden state-flags are also off. 1779 * Since we check for TCPS_ESTABLISHED first, it can only 1780 * be TH_NEEDSYN. 1781 */ 1782 if (tp->t_state == TCPS_ESTABLISHED && 1783 th->th_seq == tp->rcv_nxt && 1784 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1785 tp->snd_nxt == tp->snd_max && 1786 tiwin && tiwin == tp->snd_wnd && 1787 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1788 SEGQ_EMPTY(tp) && 1789 ((to.to_flags & TOF_TS) == 0 || 1790 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1791 /* 1792 * If last ACK falls within this segment's sequence numbers, 1793 * record the timestamp. 1794 * NOTE that the test is modified according to the latest 1795 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1796 */ 1797 if ((to.to_flags & TOF_TS) != 0 && 1798 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1799 tp->ts_recent_age = tcp_ts_getticks(); 1800 tp->ts_recent = to.to_tsval; 1801 } 1802 1803 if (tlen == 0) { 1804 if (SEQ_GT(th->th_ack, tp->snd_una) && 1805 SEQ_LEQ(th->th_ack, tp->snd_max) && 1806 !IN_RECOVERY(tp->t_flags) && 1807 (to.to_flags & TOF_SACK) == 0 && 1808 TAILQ_EMPTY(&tp->snd_holes)) { 1809 /* 1810 * This is a pure ack for outstanding data. 1811 */ 1812 TCPSTAT_INC(tcps_predack); 1813 1814 /* 1815 * "bad retransmit" recovery without timestamps. 1816 */ 1817 if ((to.to_flags & TOF_TS) == 0 && 1818 tp->t_rxtshift == 1 && 1819 tp->t_flags & TF_PREVVALID && 1820 (int)(ticks - tp->t_badrxtwin) < 0) { 1821 cc_cong_signal(tp, th, CC_RTO_ERR); 1822 } 1823 1824 /* 1825 * Recalculate the transmit timer / rtt. 1826 * 1827 * Some boxes send broken timestamp replies 1828 * during the SYN+ACK phase, ignore 1829 * timestamps of 0 or we could calculate a 1830 * huge RTT and blow up the retransmit timer. 1831 */ 1832 if ((to.to_flags & TOF_TS) != 0 && 1833 to.to_tsecr) { 1834 uint32_t t; 1835 1836 t = tcp_ts_getticks() - to.to_tsecr; 1837 if (!tp->t_rttlow || tp->t_rttlow > t) 1838 tp->t_rttlow = t; 1839 tcp_xmit_timer(tp, 1840 TCP_TS_TO_TICKS(t) + 1); 1841 } else if (tp->t_rtttime && 1842 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1843 if (!tp->t_rttlow || 1844 tp->t_rttlow > ticks - tp->t_rtttime) 1845 tp->t_rttlow = ticks - tp->t_rtttime; 1846 tcp_xmit_timer(tp, 1847 ticks - tp->t_rtttime); 1848 } 1849 acked = BYTES_THIS_ACK(tp, th); 1850 1851 #ifdef TCP_HHOOK 1852 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1853 hhook_run_tcp_est_in(tp, th, &to); 1854 #endif 1855 1856 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1857 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1858 sbdrop(&so->so_snd, acked); 1859 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1860 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1861 tp->snd_recover = th->th_ack - 1; 1862 1863 /* 1864 * Let the congestion control algorithm update 1865 * congestion control related information. This 1866 * typically means increasing the congestion 1867 * window. 1868 */ 1869 cc_ack_received(tp, th, nsegs, CC_ACK); 1870 1871 tp->snd_una = th->th_ack; 1872 /* 1873 * Pull snd_wl2 up to prevent seq wrap relative 1874 * to th_ack. 1875 */ 1876 tp->snd_wl2 = th->th_ack; 1877 tp->t_dupacks = 0; 1878 m_freem(m); 1879 1880 /* 1881 * If all outstanding data are acked, stop 1882 * retransmit timer, otherwise restart timer 1883 * using current (possibly backed-off) value. 1884 * If process is waiting for space, 1885 * wakeup/selwakeup/signal. If data 1886 * are ready to send, let tcp_output 1887 * decide between more output or persist. 1888 */ 1889 #ifdef TCPDEBUG 1890 if (so->so_options & SO_DEBUG) 1891 tcp_trace(TA_INPUT, ostate, tp, 1892 (void *)tcp_saveipgen, 1893 &tcp_savetcp, 0); 1894 #endif 1895 TCP_PROBE3(debug__input, tp, th, m); 1896 if (tp->snd_una == tp->snd_max) 1897 tcp_timer_activate(tp, TT_REXMT, 0); 1898 else if (!tcp_timer_active(tp, TT_PERSIST)) 1899 tcp_timer_activate(tp, TT_REXMT, 1900 tp->t_rxtcur); 1901 tp->t_flags |= TF_WAKESOW; 1902 if (sbavail(&so->so_snd)) 1903 (void) tp->t_fb->tfb_tcp_output(tp); 1904 goto check_delack; 1905 } 1906 } else if (th->th_ack == tp->snd_una && 1907 tlen <= sbspace(&so->so_rcv)) { 1908 int newsize = 0; /* automatic sockbuf scaling */ 1909 1910 /* 1911 * This is a pure, in-sequence data packet with 1912 * nothing on the reassembly queue and we have enough 1913 * buffer space to take it. 1914 */ 1915 /* Clean receiver SACK report if present */ 1916 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1917 tcp_clean_sackreport(tp); 1918 TCPSTAT_INC(tcps_preddat); 1919 tp->rcv_nxt += tlen; 1920 if (tlen && 1921 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1922 (tp->t_fbyte_in == 0)) { 1923 tp->t_fbyte_in = ticks; 1924 if (tp->t_fbyte_in == 0) 1925 tp->t_fbyte_in = 1; 1926 if (tp->t_fbyte_out && tp->t_fbyte_in) 1927 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1928 } 1929 /* 1930 * Pull snd_wl1 up to prevent seq wrap relative to 1931 * th_seq. 1932 */ 1933 tp->snd_wl1 = th->th_seq; 1934 /* 1935 * Pull rcv_up up to prevent seq wrap relative to 1936 * rcv_nxt. 1937 */ 1938 tp->rcv_up = tp->rcv_nxt; 1939 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1940 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1941 #ifdef TCPDEBUG 1942 if (so->so_options & SO_DEBUG) 1943 tcp_trace(TA_INPUT, ostate, tp, 1944 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1945 #endif 1946 TCP_PROBE3(debug__input, tp, th, m); 1947 1948 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1949 1950 /* Add data to socket buffer. */ 1951 SOCKBUF_LOCK(&so->so_rcv); 1952 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1953 m_freem(m); 1954 } else { 1955 /* 1956 * Set new socket buffer size. 1957 * Give up when limit is reached. 1958 */ 1959 if (newsize) 1960 if (!sbreserve_locked(&so->so_rcv, 1961 newsize, so, NULL)) 1962 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1963 m_adj(m, drop_hdrlen); /* delayed header drop */ 1964 sbappendstream_locked(&so->so_rcv, m, 0); 1965 } 1966 SOCKBUF_UNLOCK(&so->so_rcv); 1967 tp->t_flags |= TF_WAKESOR; 1968 if (DELAY_ACK(tp, tlen)) { 1969 tp->t_flags |= TF_DELACK; 1970 } else { 1971 tp->t_flags |= TF_ACKNOW; 1972 tp->t_fb->tfb_tcp_output(tp); 1973 } 1974 goto check_delack; 1975 } 1976 } 1977 1978 /* 1979 * Calculate amount of space in receive window, 1980 * and then do TCP input processing. 1981 * Receive window is amount of space in rcv queue, 1982 * but not less than advertised window. 1983 */ 1984 win = sbspace(&so->so_rcv); 1985 if (win < 0) 1986 win = 0; 1987 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1988 1989 switch (tp->t_state) { 1990 /* 1991 * If the state is SYN_RECEIVED: 1992 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1993 */ 1994 case TCPS_SYN_RECEIVED: 1995 if ((thflags & TH_ACK) && 1996 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1997 SEQ_GT(th->th_ack, tp->snd_max))) { 1998 rstreason = BANDLIM_RST_OPENPORT; 1999 goto dropwithreset; 2000 } 2001 if (IS_FASTOPEN(tp->t_flags)) { 2002 /* 2003 * When a TFO connection is in SYN_RECEIVED, the 2004 * only valid packets are the initial SYN, a 2005 * retransmit/copy of the initial SYN (possibly with 2006 * a subset of the original data), a valid ACK, a 2007 * FIN, or a RST. 2008 */ 2009 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 2010 rstreason = BANDLIM_RST_OPENPORT; 2011 goto dropwithreset; 2012 } else if (thflags & TH_SYN) { 2013 /* non-initial SYN is ignored */ 2014 if ((tcp_timer_active(tp, TT_DELACK) || 2015 tcp_timer_active(tp, TT_REXMT))) 2016 goto drop; 2017 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 2018 goto drop; 2019 } 2020 } 2021 break; 2022 2023 /* 2024 * If the state is SYN_SENT: 2025 * if seg contains a RST with valid ACK (SEQ.ACK has already 2026 * been verified), then drop the connection. 2027 * if seg contains a RST without an ACK, drop the seg. 2028 * if seg does not contain SYN, then drop the seg. 2029 * Otherwise this is an acceptable SYN segment 2030 * initialize tp->rcv_nxt and tp->irs 2031 * if seg contains ack then advance tp->snd_una 2032 * if seg contains an ECE and ECN support is enabled, the stream 2033 * is ECN capable. 2034 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2035 * arrange for segment to be acked (eventually) 2036 * continue processing rest of data/controls, beginning with URG 2037 */ 2038 case TCPS_SYN_SENT: 2039 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2040 TCP_PROBE5(connect__refused, NULL, tp, 2041 m, tp, th); 2042 tp = tcp_drop(tp, ECONNREFUSED); 2043 } 2044 if (thflags & TH_RST) 2045 goto drop; 2046 if (!(thflags & TH_SYN)) 2047 goto drop; 2048 2049 tp->irs = th->th_seq; 2050 tcp_rcvseqinit(tp); 2051 if (thflags & TH_ACK) { 2052 int tfo_partial_ack = 0; 2053 2054 TCPSTAT_INC(tcps_connects); 2055 soisconnected(so); 2056 #ifdef MAC 2057 mac_socketpeer_set_from_mbuf(m, so); 2058 #endif 2059 /* Do window scaling on this connection? */ 2060 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2061 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2062 tp->rcv_scale = tp->request_r_scale; 2063 } 2064 tp->rcv_adv += min(tp->rcv_wnd, 2065 TCP_MAXWIN << tp->rcv_scale); 2066 tp->snd_una++; /* SYN is acked */ 2067 /* 2068 * If not all the data that was sent in the TFO SYN 2069 * has been acked, resend the remainder right away. 2070 */ 2071 if (IS_FASTOPEN(tp->t_flags) && 2072 (tp->snd_una != tp->snd_max)) { 2073 tp->snd_nxt = th->th_ack; 2074 tfo_partial_ack = 1; 2075 } 2076 /* 2077 * If there's data, delay ACK; if there's also a FIN 2078 * ACKNOW will be turned on later. 2079 */ 2080 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2081 tcp_timer_activate(tp, TT_DELACK, 2082 tcp_delacktime); 2083 else 2084 tp->t_flags |= TF_ACKNOW; 2085 2086 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2087 (V_tcp_do_ecn == 1)) { 2088 tp->t_flags2 |= TF2_ECN_PERMIT; 2089 TCPSTAT_INC(tcps_ecn_shs); 2090 } 2091 2092 /* 2093 * Received <SYN,ACK> in SYN_SENT[*] state. 2094 * Transitions: 2095 * SYN_SENT --> ESTABLISHED 2096 * SYN_SENT* --> FIN_WAIT_1 2097 */ 2098 tp->t_starttime = ticks; 2099 if (tp->t_flags & TF_NEEDFIN) { 2100 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2101 tp->t_flags &= ~TF_NEEDFIN; 2102 thflags &= ~TH_SYN; 2103 } else { 2104 tcp_state_change(tp, TCPS_ESTABLISHED); 2105 TCP_PROBE5(connect__established, NULL, tp, 2106 m, tp, th); 2107 cc_conn_init(tp); 2108 tcp_timer_activate(tp, TT_KEEP, 2109 TP_KEEPIDLE(tp)); 2110 } 2111 } else { 2112 /* 2113 * Received initial SYN in SYN-SENT[*] state => 2114 * simultaneous open. 2115 * If it succeeds, connection is * half-synchronized. 2116 * Otherwise, do 3-way handshake: 2117 * SYN-SENT -> SYN-RECEIVED 2118 * SYN-SENT* -> SYN-RECEIVED* 2119 */ 2120 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2121 tcp_timer_activate(tp, TT_REXMT, 0); 2122 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2123 } 2124 2125 INP_WLOCK_ASSERT(tp->t_inpcb); 2126 2127 /* 2128 * Advance th->th_seq to correspond to first data byte. 2129 * If data, trim to stay within window, 2130 * dropping FIN if necessary. 2131 */ 2132 th->th_seq++; 2133 if (tlen > tp->rcv_wnd) { 2134 todrop = tlen - tp->rcv_wnd; 2135 m_adj(m, -todrop); 2136 tlen = tp->rcv_wnd; 2137 thflags &= ~TH_FIN; 2138 TCPSTAT_INC(tcps_rcvpackafterwin); 2139 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2140 } 2141 tp->snd_wl1 = th->th_seq - 1; 2142 tp->rcv_up = th->th_seq; 2143 /* 2144 * Client side of transaction: already sent SYN and data. 2145 * If the remote host used T/TCP to validate the SYN, 2146 * our data will be ACK'd; if so, enter normal data segment 2147 * processing in the middle of step 5, ack processing. 2148 * Otherwise, goto step 6. 2149 */ 2150 if (thflags & TH_ACK) 2151 goto process_ACK; 2152 2153 goto step6; 2154 2155 /* 2156 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2157 * do normal processing. 2158 * 2159 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2160 */ 2161 case TCPS_LAST_ACK: 2162 case TCPS_CLOSING: 2163 break; /* continue normal processing */ 2164 } 2165 2166 /* 2167 * States other than LISTEN or SYN_SENT. 2168 * First check the RST flag and sequence number since reset segments 2169 * are exempt from the timestamp and connection count tests. This 2170 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2171 * below which allowed reset segments in half the sequence space 2172 * to fall though and be processed (which gives forged reset 2173 * segments with a random sequence number a 50 percent chance of 2174 * killing a connection). 2175 * Then check timestamp, if present. 2176 * Then check the connection count, if present. 2177 * Then check that at least some bytes of segment are within 2178 * receive window. If segment begins before rcv_nxt, 2179 * drop leading data (and SYN); if nothing left, just ack. 2180 */ 2181 if (thflags & TH_RST) { 2182 /* 2183 * RFC5961 Section 3.2 2184 * 2185 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2186 * - If RST is in window, we send challenge ACK. 2187 * 2188 * Note: to take into account delayed ACKs, we should 2189 * test against last_ack_sent instead of rcv_nxt. 2190 * Note 2: we handle special case of closed window, not 2191 * covered by the RFC. 2192 */ 2193 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2194 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2195 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2196 KASSERT(tp->t_state != TCPS_SYN_SENT, 2197 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2198 __func__, th, tp)); 2199 2200 if (V_tcp_insecure_rst || 2201 tp->last_ack_sent == th->th_seq) { 2202 TCPSTAT_INC(tcps_drops); 2203 /* Drop the connection. */ 2204 switch (tp->t_state) { 2205 case TCPS_SYN_RECEIVED: 2206 so->so_error = ECONNREFUSED; 2207 goto close; 2208 case TCPS_ESTABLISHED: 2209 case TCPS_FIN_WAIT_1: 2210 case TCPS_FIN_WAIT_2: 2211 case TCPS_CLOSE_WAIT: 2212 case TCPS_CLOSING: 2213 case TCPS_LAST_ACK: 2214 so->so_error = ECONNRESET; 2215 close: 2216 /* FALLTHROUGH */ 2217 default: 2218 tp = tcp_close(tp); 2219 } 2220 } else { 2221 TCPSTAT_INC(tcps_badrst); 2222 /* Send challenge ACK. */ 2223 tcp_respond(tp, mtod(m, void *), th, m, 2224 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2225 tp->last_ack_sent = tp->rcv_nxt; 2226 m = NULL; 2227 } 2228 } 2229 goto drop; 2230 } 2231 2232 /* 2233 * RFC5961 Section 4.2 2234 * Send challenge ACK for any SYN in synchronized state. 2235 */ 2236 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2237 tp->t_state != TCPS_SYN_RECEIVED) { 2238 TCPSTAT_INC(tcps_badsyn); 2239 if (V_tcp_insecure_syn && 2240 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2241 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2242 tp = tcp_drop(tp, ECONNRESET); 2243 rstreason = BANDLIM_UNLIMITED; 2244 } else { 2245 /* Send challenge ACK. */ 2246 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2247 tp->snd_nxt, TH_ACK); 2248 tp->last_ack_sent = tp->rcv_nxt; 2249 m = NULL; 2250 } 2251 goto drop; 2252 } 2253 2254 /* 2255 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2256 * and it's less than ts_recent, drop it. 2257 */ 2258 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2259 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2260 /* Check to see if ts_recent is over 24 days old. */ 2261 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2262 /* 2263 * Invalidate ts_recent. If this segment updates 2264 * ts_recent, the age will be reset later and ts_recent 2265 * will get a valid value. If it does not, setting 2266 * ts_recent to zero will at least satisfy the 2267 * requirement that zero be placed in the timestamp 2268 * echo reply when ts_recent isn't valid. The 2269 * age isn't reset until we get a valid ts_recent 2270 * because we don't want out-of-order segments to be 2271 * dropped when ts_recent is old. 2272 */ 2273 tp->ts_recent = 0; 2274 } else { 2275 TCPSTAT_INC(tcps_rcvduppack); 2276 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2277 TCPSTAT_INC(tcps_pawsdrop); 2278 if (tlen) 2279 goto dropafterack; 2280 goto drop; 2281 } 2282 } 2283 2284 /* 2285 * In the SYN-RECEIVED state, validate that the packet belongs to 2286 * this connection before trimming the data to fit the receive 2287 * window. Check the sequence number versus IRS since we know 2288 * the sequence numbers haven't wrapped. This is a partial fix 2289 * for the "LAND" DoS attack. 2290 */ 2291 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2292 rstreason = BANDLIM_RST_OPENPORT; 2293 goto dropwithreset; 2294 } 2295 2296 todrop = tp->rcv_nxt - th->th_seq; 2297 if (todrop > 0) { 2298 if (thflags & TH_SYN) { 2299 thflags &= ~TH_SYN; 2300 th->th_seq++; 2301 if (th->th_urp > 1) 2302 th->th_urp--; 2303 else 2304 thflags &= ~TH_URG; 2305 todrop--; 2306 } 2307 /* 2308 * Following if statement from Stevens, vol. 2, p. 960. 2309 */ 2310 if (todrop > tlen 2311 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2312 /* 2313 * Any valid FIN must be to the left of the window. 2314 * At this point the FIN must be a duplicate or out 2315 * of sequence; drop it. 2316 */ 2317 thflags &= ~TH_FIN; 2318 2319 /* 2320 * Send an ACK to resynchronize and drop any data. 2321 * But keep on processing for RST or ACK. 2322 */ 2323 tp->t_flags |= TF_ACKNOW; 2324 todrop = tlen; 2325 TCPSTAT_INC(tcps_rcvduppack); 2326 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2327 } else { 2328 TCPSTAT_INC(tcps_rcvpartduppack); 2329 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2330 } 2331 /* 2332 * DSACK - add SACK block for dropped range 2333 */ 2334 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2335 tcp_update_sack_list(tp, th->th_seq, 2336 th->th_seq + todrop); 2337 /* 2338 * ACK now, as the next in-sequence segment 2339 * will clear the DSACK block again 2340 */ 2341 tp->t_flags |= TF_ACKNOW; 2342 } 2343 drop_hdrlen += todrop; /* drop from the top afterwards */ 2344 th->th_seq += todrop; 2345 tlen -= todrop; 2346 if (th->th_urp > todrop) 2347 th->th_urp -= todrop; 2348 else { 2349 thflags &= ~TH_URG; 2350 th->th_urp = 0; 2351 } 2352 } 2353 2354 /* 2355 * If new data are received on a connection after the 2356 * user processes are gone, then RST the other end. 2357 */ 2358 if ((so->so_state & SS_NOFDREF) && 2359 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2360 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2361 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2362 "after socket was closed, " 2363 "sending RST and removing tcpcb\n", 2364 s, __func__, tcpstates[tp->t_state], tlen); 2365 free(s, M_TCPLOG); 2366 } 2367 tp = tcp_close(tp); 2368 TCPSTAT_INC(tcps_rcvafterclose); 2369 rstreason = BANDLIM_UNLIMITED; 2370 goto dropwithreset; 2371 } 2372 2373 /* 2374 * If segment ends after window, drop trailing data 2375 * (and PUSH and FIN); if nothing left, just ACK. 2376 */ 2377 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2378 if (todrop > 0) { 2379 TCPSTAT_INC(tcps_rcvpackafterwin); 2380 if (todrop >= tlen) { 2381 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2382 /* 2383 * If window is closed can only take segments at 2384 * window edge, and have to drop data and PUSH from 2385 * incoming segments. Continue processing, but 2386 * remember to ack. Otherwise, drop segment 2387 * and ack. 2388 */ 2389 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2390 tp->t_flags |= TF_ACKNOW; 2391 TCPSTAT_INC(tcps_rcvwinprobe); 2392 } else 2393 goto dropafterack; 2394 } else 2395 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2396 m_adj(m, -todrop); 2397 tlen -= todrop; 2398 thflags &= ~(TH_PUSH|TH_FIN); 2399 } 2400 2401 /* 2402 * If last ACK falls within this segment's sequence numbers, 2403 * record its timestamp. 2404 * NOTE: 2405 * 1) That the test incorporates suggestions from the latest 2406 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2407 * 2) That updating only on newer timestamps interferes with 2408 * our earlier PAWS tests, so this check should be solely 2409 * predicated on the sequence space of this segment. 2410 * 3) That we modify the segment boundary check to be 2411 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2412 * instead of RFC1323's 2413 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2414 * This modified check allows us to overcome RFC1323's 2415 * limitations as described in Stevens TCP/IP Illustrated 2416 * Vol. 2 p.869. In such cases, we can still calculate the 2417 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2418 */ 2419 if ((to.to_flags & TOF_TS) != 0 && 2420 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2421 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2422 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2423 tp->ts_recent_age = tcp_ts_getticks(); 2424 tp->ts_recent = to.to_tsval; 2425 } 2426 2427 /* 2428 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2429 * flag is on (half-synchronized state), then queue data for 2430 * later processing; else drop segment and return. 2431 */ 2432 if ((thflags & TH_ACK) == 0) { 2433 if (tp->t_state == TCPS_SYN_RECEIVED || 2434 (tp->t_flags & TF_NEEDSYN)) { 2435 if (tp->t_state == TCPS_SYN_RECEIVED && 2436 IS_FASTOPEN(tp->t_flags)) { 2437 tp->snd_wnd = tiwin; 2438 cc_conn_init(tp); 2439 } 2440 goto step6; 2441 } else if (tp->t_flags & TF_ACKNOW) 2442 goto dropafterack; 2443 else 2444 goto drop; 2445 } 2446 2447 /* 2448 * Ack processing. 2449 */ 2450 switch (tp->t_state) { 2451 /* 2452 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2453 * ESTABLISHED state and continue processing. 2454 * The ACK was checked above. 2455 */ 2456 case TCPS_SYN_RECEIVED: 2457 2458 TCPSTAT_INC(tcps_connects); 2459 soisconnected(so); 2460 /* Do window scaling? */ 2461 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2462 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2463 tp->rcv_scale = tp->request_r_scale; 2464 } 2465 tp->snd_wnd = tiwin; 2466 /* 2467 * Make transitions: 2468 * SYN-RECEIVED -> ESTABLISHED 2469 * SYN-RECEIVED* -> FIN-WAIT-1 2470 */ 2471 tp->t_starttime = ticks; 2472 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2473 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2474 tp->t_tfo_pending = NULL; 2475 } 2476 if (tp->t_flags & TF_NEEDFIN) { 2477 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2478 tp->t_flags &= ~TF_NEEDFIN; 2479 } else { 2480 tcp_state_change(tp, TCPS_ESTABLISHED); 2481 TCP_PROBE5(accept__established, NULL, tp, 2482 m, tp, th); 2483 /* 2484 * TFO connections call cc_conn_init() during SYN 2485 * processing. Calling it again here for such 2486 * connections is not harmless as it would undo the 2487 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2488 * is retransmitted. 2489 */ 2490 if (!IS_FASTOPEN(tp->t_flags)) 2491 cc_conn_init(tp); 2492 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2493 } 2494 /* 2495 * Account for the ACK of our SYN prior to 2496 * regular ACK processing below, except for 2497 * simultaneous SYN, which is handled later. 2498 */ 2499 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2500 incforsyn = 1; 2501 /* 2502 * If segment contains data or ACK, will call tcp_reass() 2503 * later; if not, do so now to pass queued data to user. 2504 */ 2505 if (tlen == 0 && (thflags & TH_FIN) == 0) 2506 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2507 (struct mbuf *)0); 2508 tp->snd_wl1 = th->th_seq - 1; 2509 /* FALLTHROUGH */ 2510 2511 /* 2512 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2513 * ACKs. If the ack is in the range 2514 * tp->snd_una < th->th_ack <= tp->snd_max 2515 * then advance tp->snd_una to th->th_ack and drop 2516 * data from the retransmission queue. If this ACK reflects 2517 * more up to date window information we update our window information. 2518 */ 2519 case TCPS_ESTABLISHED: 2520 case TCPS_FIN_WAIT_1: 2521 case TCPS_FIN_WAIT_2: 2522 case TCPS_CLOSE_WAIT: 2523 case TCPS_CLOSING: 2524 case TCPS_LAST_ACK: 2525 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2526 TCPSTAT_INC(tcps_rcvacktoomuch); 2527 goto dropafterack; 2528 } 2529 if ((tp->t_flags & TF_SACK_PERMIT) && 2530 ((to.to_flags & TOF_SACK) || 2531 !TAILQ_EMPTY(&tp->snd_holes))) { 2532 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && 2533 (tp->t_flags & TF_LRD)) { 2534 tcp_sack_lost_retransmission(tp, th); 2535 } 2536 } else 2537 /* 2538 * Reset the value so that previous (valid) value 2539 * from the last ack with SACK doesn't get used. 2540 */ 2541 tp->sackhint.sacked_bytes = 0; 2542 2543 #ifdef TCP_HHOOK 2544 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2545 hhook_run_tcp_est_in(tp, th, &to); 2546 #endif 2547 2548 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2549 maxseg = tcp_maxseg(tp); 2550 if (tlen == 0 && 2551 (tiwin == tp->snd_wnd || 2552 (tp->t_flags & TF_SACK_PERMIT))) { 2553 /* 2554 * If this is the first time we've seen a 2555 * FIN from the remote, this is not a 2556 * duplicate and it needs to be processed 2557 * normally. This happens during a 2558 * simultaneous close. 2559 */ 2560 if ((thflags & TH_FIN) && 2561 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2562 tp->t_dupacks = 0; 2563 break; 2564 } 2565 TCPSTAT_INC(tcps_rcvdupack); 2566 /* 2567 * If we have outstanding data (other than 2568 * a window probe), this is a completely 2569 * duplicate ack (ie, window info didn't 2570 * change and FIN isn't set), 2571 * the ack is the biggest we've 2572 * seen and we've seen exactly our rexmt 2573 * threshold of them, assume a packet 2574 * has been dropped and retransmit it. 2575 * Kludge snd_nxt & the congestion 2576 * window so we send only this one 2577 * packet. 2578 * 2579 * We know we're losing at the current 2580 * window size so do congestion avoidance 2581 * (set ssthresh to half the current window 2582 * and pull our congestion window back to 2583 * the new ssthresh). 2584 * 2585 * Dup acks mean that packets have left the 2586 * network (they're now cached at the receiver) 2587 * so bump cwnd by the amount in the receiver 2588 * to keep a constant cwnd packets in the 2589 * network. 2590 * 2591 * When using TCP ECN, notify the peer that 2592 * we reduced the cwnd. 2593 */ 2594 /* 2595 * Following 2 kinds of acks should not affect 2596 * dupack counting: 2597 * 1) Old acks 2598 * 2) Acks with SACK but without any new SACK 2599 * information in them. These could result from 2600 * any anomaly in the network like a switch 2601 * duplicating packets or a possible DoS attack. 2602 */ 2603 if (th->th_ack != tp->snd_una || 2604 ((tp->t_flags & TF_SACK_PERMIT) && 2605 (to.to_flags & TOF_SACK) && 2606 !sack_changed)) 2607 break; 2608 else if (!tcp_timer_active(tp, TT_REXMT)) 2609 tp->t_dupacks = 0; 2610 else if (++tp->t_dupacks > tcprexmtthresh || 2611 IN_FASTRECOVERY(tp->t_flags)) { 2612 cc_ack_received(tp, th, nsegs, 2613 CC_DUPACK); 2614 if (V_tcp_do_prr && 2615 IN_FASTRECOVERY(tp->t_flags) && 2616 (tp->t_flags & TF_SACK_PERMIT)) { 2617 tcp_do_prr_ack(tp, th, &to); 2618 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2619 (to.to_flags & TOF_SACK) && 2620 IN_FASTRECOVERY(tp->t_flags)) { 2621 int awnd; 2622 2623 /* 2624 * Compute the amount of data in flight first. 2625 * We can inject new data into the pipe iff 2626 * we have less than 1/2 the original window's 2627 * worth of data in flight. 2628 */ 2629 if (V_tcp_do_newsack) 2630 awnd = tcp_compute_pipe(tp); 2631 else 2632 awnd = (tp->snd_nxt - tp->snd_fack) + 2633 tp->sackhint.sack_bytes_rexmit; 2634 2635 if (awnd < tp->snd_ssthresh) { 2636 tp->snd_cwnd += maxseg; 2637 if (tp->snd_cwnd > tp->snd_ssthresh) 2638 tp->snd_cwnd = tp->snd_ssthresh; 2639 } 2640 } else 2641 tp->snd_cwnd += maxseg; 2642 (void) tp->t_fb->tfb_tcp_output(tp); 2643 goto drop; 2644 } else if (tp->t_dupacks == tcprexmtthresh || 2645 (tp->t_flags & TF_SACK_PERMIT && 2646 V_tcp_do_newsack && 2647 tp->sackhint.sacked_bytes > 2648 (tcprexmtthresh - 1) * maxseg)) { 2649 enter_recovery: 2650 /* 2651 * Above is the RFC6675 trigger condition of 2652 * more than (dupthresh-1)*maxseg sacked data. 2653 * If the count of holes in the 2654 * scoreboard is >= dupthresh, we could 2655 * also enter loss recovery, but don't 2656 * have that value readily available. 2657 */ 2658 tp->t_dupacks = tcprexmtthresh; 2659 tcp_seq onxt = tp->snd_nxt; 2660 2661 /* 2662 * If we're doing sack, or prr, check 2663 * to see if we're already in sack 2664 * recovery. If we're not doing sack, 2665 * check to see if we're in newreno 2666 * recovery. 2667 */ 2668 if (V_tcp_do_prr || 2669 (tp->t_flags & TF_SACK_PERMIT)) { 2670 if (IN_FASTRECOVERY(tp->t_flags)) { 2671 tp->t_dupacks = 0; 2672 break; 2673 } 2674 } else { 2675 if (SEQ_LEQ(th->th_ack, 2676 tp->snd_recover)) { 2677 tp->t_dupacks = 0; 2678 break; 2679 } 2680 } 2681 /* Congestion signal before ack. */ 2682 cc_cong_signal(tp, th, CC_NDUPACK); 2683 cc_ack_received(tp, th, nsegs, 2684 CC_DUPACK); 2685 tcp_timer_activate(tp, TT_REXMT, 0); 2686 tp->t_rtttime = 0; 2687 if (V_tcp_do_prr) { 2688 /* 2689 * snd_ssthresh is already updated by 2690 * cc_cong_signal. 2691 */ 2692 tp->sackhint.prr_delivered = 2693 tp->sackhint.sacked_bytes; 2694 tp->sackhint.recover_fs = max(1, 2695 tp->snd_nxt - tp->snd_una); 2696 } 2697 if ((tp->t_flags & TF_SACK_PERMIT) && 2698 (to.to_flags & TOF_SACK)) { 2699 TCPSTAT_INC( 2700 tcps_sack_recovery_episode); 2701 tp->snd_recover = tp->snd_nxt; 2702 tp->snd_cwnd = maxseg; 2703 (void) tp->t_fb->tfb_tcp_output(tp); 2704 if (SEQ_GT(th->th_ack, tp->snd_una)) 2705 goto resume_partialack; 2706 goto drop; 2707 } 2708 tp->snd_nxt = th->th_ack; 2709 tp->snd_cwnd = maxseg; 2710 (void) tp->t_fb->tfb_tcp_output(tp); 2711 KASSERT(tp->snd_limited <= 2, 2712 ("%s: tp->snd_limited too big", 2713 __func__)); 2714 tp->snd_cwnd = tp->snd_ssthresh + 2715 maxseg * 2716 (tp->t_dupacks - tp->snd_limited); 2717 if (SEQ_GT(onxt, tp->snd_nxt)) 2718 tp->snd_nxt = onxt; 2719 goto drop; 2720 } else if (V_tcp_do_rfc3042) { 2721 /* 2722 * Process first and second duplicate 2723 * ACKs. Each indicates a segment 2724 * leaving the network, creating room 2725 * for more. Make sure we can send a 2726 * packet on reception of each duplicate 2727 * ACK by increasing snd_cwnd by one 2728 * segment. Restore the original 2729 * snd_cwnd after packet transmission. 2730 */ 2731 cc_ack_received(tp, th, nsegs, 2732 CC_DUPACK); 2733 uint32_t oldcwnd = tp->snd_cwnd; 2734 tcp_seq oldsndmax = tp->snd_max; 2735 u_int sent; 2736 int avail; 2737 2738 KASSERT(tp->t_dupacks == 1 || 2739 tp->t_dupacks == 2, 2740 ("%s: dupacks not 1 or 2", 2741 __func__)); 2742 if (tp->t_dupacks == 1) 2743 tp->snd_limited = 0; 2744 tp->snd_cwnd = 2745 (tp->snd_nxt - tp->snd_una) + 2746 (tp->t_dupacks - tp->snd_limited) * 2747 maxseg; 2748 /* 2749 * Only call tcp_output when there 2750 * is new data available to be sent. 2751 * Otherwise we would send pure ACKs. 2752 */ 2753 SOCKBUF_LOCK(&so->so_snd); 2754 avail = sbavail(&so->so_snd) - 2755 (tp->snd_nxt - tp->snd_una); 2756 SOCKBUF_UNLOCK(&so->so_snd); 2757 if (avail > 0) 2758 (void) tp->t_fb->tfb_tcp_output(tp); 2759 sent = tp->snd_max - oldsndmax; 2760 if (sent > maxseg) { 2761 KASSERT((tp->t_dupacks == 2 && 2762 tp->snd_limited == 0) || 2763 (sent == maxseg + 1 && 2764 tp->t_flags & TF_SENTFIN), 2765 ("%s: sent too much", 2766 __func__)); 2767 tp->snd_limited = 2; 2768 } else if (sent > 0) 2769 ++tp->snd_limited; 2770 tp->snd_cwnd = oldcwnd; 2771 goto drop; 2772 } 2773 } 2774 break; 2775 } else { 2776 /* 2777 * This ack is advancing the left edge, reset the 2778 * counter. 2779 */ 2780 tp->t_dupacks = 0; 2781 /* 2782 * If this ack also has new SACK info, increment the 2783 * counter as per rfc6675. The variable 2784 * sack_changed tracks all changes to the SACK 2785 * scoreboard, including when partial ACKs without 2786 * SACK options are received, and clear the scoreboard 2787 * from the left side. Such partial ACKs should not be 2788 * counted as dupacks here. 2789 */ 2790 if ((tp->t_flags & TF_SACK_PERMIT) && 2791 (to.to_flags & TOF_SACK) && 2792 sack_changed) { 2793 tp->t_dupacks++; 2794 /* limit overhead by setting maxseg last */ 2795 if (!IN_FASTRECOVERY(tp->t_flags) && 2796 (tp->sackhint.sacked_bytes > 2797 ((tcprexmtthresh - 1) * 2798 (maxseg = tcp_maxseg(tp))))) { 2799 goto enter_recovery; 2800 } 2801 } 2802 } 2803 2804 resume_partialack: 2805 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2806 ("%s: th_ack <= snd_una", __func__)); 2807 2808 /* 2809 * If the congestion window was inflated to account 2810 * for the other side's cached packets, retract it. 2811 */ 2812 if (IN_FASTRECOVERY(tp->t_flags)) { 2813 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2814 if (tp->t_flags & TF_SACK_PERMIT) 2815 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2816 tcp_timer_activate(tp, TT_REXMT, 0); 2817 tp->t_rtttime = 0; 2818 tcp_do_prr_ack(tp, th, &to); 2819 tp->t_flags |= TF_ACKNOW; 2820 (void) tcp_output(tp); 2821 } else 2822 tcp_sack_partialack(tp, th); 2823 else 2824 tcp_newreno_partial_ack(tp, th); 2825 } else 2826 cc_post_recovery(tp, th); 2827 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2828 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2829 if (V_tcp_do_prr) { 2830 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2831 tp->snd_fack = th->th_ack; 2832 tcp_do_prr_ack(tp, th, &to); 2833 (void) tcp_output(tp); 2834 } 2835 } else 2836 cc_post_recovery(tp, th); 2837 } 2838 /* 2839 * If we reach this point, ACK is not a duplicate, 2840 * i.e., it ACKs something we sent. 2841 */ 2842 if (tp->t_flags & TF_NEEDSYN) { 2843 /* 2844 * T/TCP: Connection was half-synchronized, and our 2845 * SYN has been ACK'd (so connection is now fully 2846 * synchronized). Go to non-starred state, 2847 * increment snd_una for ACK of SYN, and check if 2848 * we can do window scaling. 2849 */ 2850 tp->t_flags &= ~TF_NEEDSYN; 2851 tp->snd_una++; 2852 /* Do window scaling? */ 2853 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2854 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2855 tp->rcv_scale = tp->request_r_scale; 2856 /* Send window already scaled. */ 2857 } 2858 } 2859 2860 process_ACK: 2861 INP_WLOCK_ASSERT(tp->t_inpcb); 2862 2863 /* 2864 * Adjust for the SYN bit in sequence space, 2865 * but don't account for it in cwnd calculations. 2866 * This is for the SYN_RECEIVED, non-simultaneous 2867 * SYN case. SYN_SENT and simultaneous SYN are 2868 * treated elsewhere. 2869 */ 2870 if (incforsyn) 2871 tp->snd_una++; 2872 acked = BYTES_THIS_ACK(tp, th); 2873 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2874 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2875 tp->snd_una, th->th_ack, tp, m)); 2876 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2877 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2878 2879 /* 2880 * If we just performed our first retransmit, and the ACK 2881 * arrives within our recovery window, then it was a mistake 2882 * to do the retransmit in the first place. Recover our 2883 * original cwnd and ssthresh, and proceed to transmit where 2884 * we left off. 2885 */ 2886 if (tp->t_rxtshift == 1 && 2887 tp->t_flags & TF_PREVVALID && 2888 tp->t_badrxtwin && 2889 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2890 cc_cong_signal(tp, th, CC_RTO_ERR); 2891 2892 /* 2893 * If we have a timestamp reply, update smoothed 2894 * round trip time. If no timestamp is present but 2895 * transmit timer is running and timed sequence 2896 * number was acked, update smoothed round trip time. 2897 * Since we now have an rtt measurement, cancel the 2898 * timer backoff (cf., Phil Karn's retransmit alg.). 2899 * Recompute the initial retransmit timer. 2900 * 2901 * Some boxes send broken timestamp replies 2902 * during the SYN+ACK phase, ignore 2903 * timestamps of 0 or we could calculate a 2904 * huge RTT and blow up the retransmit timer. 2905 */ 2906 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2907 uint32_t t; 2908 2909 t = tcp_ts_getticks() - to.to_tsecr; 2910 if (!tp->t_rttlow || tp->t_rttlow > t) 2911 tp->t_rttlow = t; 2912 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2913 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2914 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2915 tp->t_rttlow = ticks - tp->t_rtttime; 2916 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2917 } 2918 2919 /* 2920 * If all outstanding data is acked, stop retransmit 2921 * timer and remember to restart (more output or persist). 2922 * If there is more data to be acked, restart retransmit 2923 * timer, using current (possibly backed-off) value. 2924 */ 2925 if (th->th_ack == tp->snd_max) { 2926 tcp_timer_activate(tp, TT_REXMT, 0); 2927 needoutput = 1; 2928 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2929 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2930 2931 /* 2932 * If no data (only SYN) was ACK'd, 2933 * skip rest of ACK processing. 2934 */ 2935 if (acked == 0) 2936 goto step6; 2937 2938 /* 2939 * Let the congestion control algorithm update congestion 2940 * control related information. This typically means increasing 2941 * the congestion window. 2942 */ 2943 cc_ack_received(tp, th, nsegs, CC_ACK); 2944 2945 SOCKBUF_LOCK(&so->so_snd); 2946 if (acked > sbavail(&so->so_snd)) { 2947 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2948 tp->snd_wnd -= sbavail(&so->so_snd); 2949 else 2950 tp->snd_wnd = 0; 2951 mfree = sbcut_locked(&so->so_snd, 2952 (int)sbavail(&so->so_snd)); 2953 ourfinisacked = 1; 2954 } else { 2955 mfree = sbcut_locked(&so->so_snd, acked); 2956 if (tp->snd_wnd >= (uint32_t) acked) 2957 tp->snd_wnd -= acked; 2958 else 2959 tp->snd_wnd = 0; 2960 ourfinisacked = 0; 2961 } 2962 SOCKBUF_UNLOCK(&so->so_snd); 2963 tp->t_flags |= TF_WAKESOW; 2964 m_freem(mfree); 2965 /* Detect una wraparound. */ 2966 if (!IN_RECOVERY(tp->t_flags) && 2967 SEQ_GT(tp->snd_una, tp->snd_recover) && 2968 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2969 tp->snd_recover = th->th_ack - 1; 2970 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2971 if (IN_RECOVERY(tp->t_flags) && 2972 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2973 EXIT_RECOVERY(tp->t_flags); 2974 } 2975 tp->snd_una = th->th_ack; 2976 if (tp->t_flags & TF_SACK_PERMIT) { 2977 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2978 tp->snd_recover = tp->snd_una; 2979 } 2980 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2981 tp->snd_nxt = tp->snd_una; 2982 2983 switch (tp->t_state) { 2984 /* 2985 * In FIN_WAIT_1 STATE in addition to the processing 2986 * for the ESTABLISHED state if our FIN is now acknowledged 2987 * then enter FIN_WAIT_2. 2988 */ 2989 case TCPS_FIN_WAIT_1: 2990 if (ourfinisacked) { 2991 /* 2992 * If we can't receive any more 2993 * data, then closing user can proceed. 2994 * Starting the timer is contrary to the 2995 * specification, but if we don't get a FIN 2996 * we'll hang forever. 2997 * 2998 * XXXjl: 2999 * we should release the tp also, and use a 3000 * compressed state. 3001 */ 3002 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3003 soisdisconnected(so); 3004 tcp_timer_activate(tp, TT_2MSL, 3005 (tcp_fast_finwait2_recycle ? 3006 tcp_finwait2_timeout : 3007 TP_MAXIDLE(tp))); 3008 } 3009 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3010 } 3011 break; 3012 3013 /* 3014 * In CLOSING STATE in addition to the processing for 3015 * the ESTABLISHED state if the ACK acknowledges our FIN 3016 * then enter the TIME-WAIT state, otherwise ignore 3017 * the segment. 3018 */ 3019 case TCPS_CLOSING: 3020 if (ourfinisacked) { 3021 tcp_twstart(tp); 3022 m_freem(m); 3023 return; 3024 } 3025 break; 3026 3027 /* 3028 * In LAST_ACK, we may still be waiting for data to drain 3029 * and/or to be acked, as well as for the ack of our FIN. 3030 * If our FIN is now acknowledged, delete the TCB, 3031 * enter the closed state and return. 3032 */ 3033 case TCPS_LAST_ACK: 3034 if (ourfinisacked) { 3035 tp = tcp_close(tp); 3036 goto drop; 3037 } 3038 break; 3039 } 3040 } 3041 3042 step6: 3043 INP_WLOCK_ASSERT(tp->t_inpcb); 3044 3045 /* 3046 * Update window information. 3047 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3048 */ 3049 if ((thflags & TH_ACK) && 3050 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3051 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3052 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3053 /* keep track of pure window updates */ 3054 if (tlen == 0 && 3055 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3056 TCPSTAT_INC(tcps_rcvwinupd); 3057 tp->snd_wnd = tiwin; 3058 tp->snd_wl1 = th->th_seq; 3059 tp->snd_wl2 = th->th_ack; 3060 if (tp->snd_wnd > tp->max_sndwnd) 3061 tp->max_sndwnd = tp->snd_wnd; 3062 needoutput = 1; 3063 } 3064 3065 /* 3066 * Process segments with URG. 3067 */ 3068 if ((thflags & TH_URG) && th->th_urp && 3069 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3070 /* 3071 * This is a kludge, but if we receive and accept 3072 * random urgent pointers, we'll crash in 3073 * soreceive. It's hard to imagine someone 3074 * actually wanting to send this much urgent data. 3075 */ 3076 SOCKBUF_LOCK(&so->so_rcv); 3077 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3078 th->th_urp = 0; /* XXX */ 3079 thflags &= ~TH_URG; /* XXX */ 3080 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3081 goto dodata; /* XXX */ 3082 } 3083 /* 3084 * If this segment advances the known urgent pointer, 3085 * then mark the data stream. This should not happen 3086 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3087 * a FIN has been received from the remote side. 3088 * In these states we ignore the URG. 3089 * 3090 * According to RFC961 (Assigned Protocols), 3091 * the urgent pointer points to the last octet 3092 * of urgent data. We continue, however, 3093 * to consider it to indicate the first octet 3094 * of data past the urgent section as the original 3095 * spec states (in one of two places). 3096 */ 3097 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3098 tp->rcv_up = th->th_seq + th->th_urp; 3099 so->so_oobmark = sbavail(&so->so_rcv) + 3100 (tp->rcv_up - tp->rcv_nxt) - 1; 3101 if (so->so_oobmark == 0) 3102 so->so_rcv.sb_state |= SBS_RCVATMARK; 3103 sohasoutofband(so); 3104 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3105 } 3106 SOCKBUF_UNLOCK(&so->so_rcv); 3107 /* 3108 * Remove out of band data so doesn't get presented to user. 3109 * This can happen independent of advancing the URG pointer, 3110 * but if two URG's are pending at once, some out-of-band 3111 * data may creep in... ick. 3112 */ 3113 if (th->th_urp <= (uint32_t)tlen && 3114 !(so->so_options & SO_OOBINLINE)) { 3115 /* hdr drop is delayed */ 3116 tcp_pulloutofband(so, th, m, drop_hdrlen); 3117 } 3118 } else { 3119 /* 3120 * If no out of band data is expected, 3121 * pull receive urgent pointer along 3122 * with the receive window. 3123 */ 3124 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3125 tp->rcv_up = tp->rcv_nxt; 3126 } 3127 dodata: /* XXX */ 3128 INP_WLOCK_ASSERT(tp->t_inpcb); 3129 3130 /* 3131 * Process the segment text, merging it into the TCP sequencing queue, 3132 * and arranging for acknowledgment of receipt if necessary. 3133 * This process logically involves adjusting tp->rcv_wnd as data 3134 * is presented to the user (this happens in tcp_usrreq.c, 3135 * case PRU_RCVD). If a FIN has already been received on this 3136 * connection then we just ignore the text. 3137 */ 3138 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3139 IS_FASTOPEN(tp->t_flags)); 3140 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3141 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3142 tcp_seq save_start = th->th_seq; 3143 tcp_seq save_rnxt = tp->rcv_nxt; 3144 int save_tlen = tlen; 3145 m_adj(m, drop_hdrlen); /* delayed header drop */ 3146 /* 3147 * Insert segment which includes th into TCP reassembly queue 3148 * with control block tp. Set thflags to whether reassembly now 3149 * includes a segment with FIN. This handles the common case 3150 * inline (segment is the next to be received on an established 3151 * connection, and the queue is empty), avoiding linkage into 3152 * and removal from the queue and repetition of various 3153 * conversions. 3154 * Set DELACK for segments received in order, but ack 3155 * immediately when segments are out of order (so 3156 * fast retransmit can work). 3157 */ 3158 if (th->th_seq == tp->rcv_nxt && 3159 SEGQ_EMPTY(tp) && 3160 (TCPS_HAVEESTABLISHED(tp->t_state) || 3161 tfo_syn)) { 3162 if (DELAY_ACK(tp, tlen) || tfo_syn) 3163 tp->t_flags |= TF_DELACK; 3164 else 3165 tp->t_flags |= TF_ACKNOW; 3166 tp->rcv_nxt += tlen; 3167 if (tlen && 3168 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3169 (tp->t_fbyte_in == 0)) { 3170 tp->t_fbyte_in = ticks; 3171 if (tp->t_fbyte_in == 0) 3172 tp->t_fbyte_in = 1; 3173 if (tp->t_fbyte_out && tp->t_fbyte_in) 3174 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3175 } 3176 thflags = th->th_flags & TH_FIN; 3177 TCPSTAT_INC(tcps_rcvpack); 3178 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3179 SOCKBUF_LOCK(&so->so_rcv); 3180 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3181 m_freem(m); 3182 else 3183 sbappendstream_locked(&so->so_rcv, m, 0); 3184 SOCKBUF_UNLOCK(&so->so_rcv); 3185 tp->t_flags |= TF_WAKESOR; 3186 } else { 3187 /* 3188 * XXX: Due to the header drop above "th" is 3189 * theoretically invalid by now. Fortunately 3190 * m_adj() doesn't actually frees any mbufs 3191 * when trimming from the head. 3192 */ 3193 tcp_seq temp = save_start; 3194 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3195 tp->t_flags |= TF_ACKNOW; 3196 } 3197 if ((tp->t_flags & TF_SACK_PERMIT) && 3198 (save_tlen > 0) && 3199 TCPS_HAVEESTABLISHED(tp->t_state)) { 3200 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3201 /* 3202 * DSACK actually handled in the fastpath 3203 * above. 3204 */ 3205 tcp_update_sack_list(tp, save_start, 3206 save_start + save_tlen); 3207 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3208 if ((tp->rcv_numsacks >= 1) && 3209 (tp->sackblks[0].end == save_start)) { 3210 /* 3211 * Partial overlap, recorded at todrop 3212 * above. 3213 */ 3214 tcp_update_sack_list(tp, 3215 tp->sackblks[0].start, 3216 tp->sackblks[0].end); 3217 } else { 3218 tcp_update_dsack_list(tp, save_start, 3219 save_start + save_tlen); 3220 } 3221 } else if (tlen >= save_tlen) { 3222 /* Update of sackblks. */ 3223 tcp_update_dsack_list(tp, save_start, 3224 save_start + save_tlen); 3225 } else if (tlen > 0) { 3226 tcp_update_dsack_list(tp, save_start, 3227 save_start + tlen); 3228 } 3229 } 3230 #if 0 3231 /* 3232 * Note the amount of data that peer has sent into 3233 * our window, in order to estimate the sender's 3234 * buffer size. 3235 * XXX: Unused. 3236 */ 3237 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3238 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3239 else 3240 len = so->so_rcv.sb_hiwat; 3241 #endif 3242 } else { 3243 m_freem(m); 3244 thflags &= ~TH_FIN; 3245 } 3246 3247 /* 3248 * If FIN is received ACK the FIN and let the user know 3249 * that the connection is closing. 3250 */ 3251 if (thflags & TH_FIN) { 3252 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3253 socantrcvmore(so); 3254 /* The socket upcall is handled by socantrcvmore. */ 3255 tp->t_flags &= ~TF_WAKESOR; 3256 /* 3257 * If connection is half-synchronized 3258 * (ie NEEDSYN flag on) then delay ACK, 3259 * so it may be piggybacked when SYN is sent. 3260 * Otherwise, since we received a FIN then no 3261 * more input can be expected, send ACK now. 3262 */ 3263 if (tp->t_flags & TF_NEEDSYN) 3264 tp->t_flags |= TF_DELACK; 3265 else 3266 tp->t_flags |= TF_ACKNOW; 3267 tp->rcv_nxt++; 3268 } 3269 switch (tp->t_state) { 3270 /* 3271 * In SYN_RECEIVED and ESTABLISHED STATES 3272 * enter the CLOSE_WAIT state. 3273 */ 3274 case TCPS_SYN_RECEIVED: 3275 tp->t_starttime = ticks; 3276 /* FALLTHROUGH */ 3277 case TCPS_ESTABLISHED: 3278 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3279 break; 3280 3281 /* 3282 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3283 * enter the CLOSING state. 3284 */ 3285 case TCPS_FIN_WAIT_1: 3286 tcp_state_change(tp, TCPS_CLOSING); 3287 break; 3288 3289 /* 3290 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3291 * starting the time-wait timer, turning off the other 3292 * standard timers. 3293 */ 3294 case TCPS_FIN_WAIT_2: 3295 tcp_twstart(tp); 3296 return; 3297 } 3298 } 3299 #ifdef TCPDEBUG 3300 if (so->so_options & SO_DEBUG) 3301 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3302 &tcp_savetcp, 0); 3303 #endif 3304 TCP_PROBE3(debug__input, tp, th, m); 3305 3306 /* 3307 * Return any desired output. 3308 */ 3309 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3310 (void) tp->t_fb->tfb_tcp_output(tp); 3311 3312 check_delack: 3313 INP_WLOCK_ASSERT(tp->t_inpcb); 3314 3315 if (tp->t_flags & TF_DELACK) { 3316 tp->t_flags &= ~TF_DELACK; 3317 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3318 } 3319 tcp_handle_wakeup(tp, so); 3320 INP_WUNLOCK(tp->t_inpcb); 3321 return; 3322 3323 dropafterack: 3324 /* 3325 * Generate an ACK dropping incoming segment if it occupies 3326 * sequence space, where the ACK reflects our state. 3327 * 3328 * We can now skip the test for the RST flag since all 3329 * paths to this code happen after packets containing 3330 * RST have been dropped. 3331 * 3332 * In the SYN-RECEIVED state, don't send an ACK unless the 3333 * segment we received passes the SYN-RECEIVED ACK test. 3334 * If it fails send a RST. This breaks the loop in the 3335 * "LAND" DoS attack, and also prevents an ACK storm 3336 * between two listening ports that have been sent forged 3337 * SYN segments, each with the source address of the other. 3338 */ 3339 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3340 (SEQ_GT(tp->snd_una, th->th_ack) || 3341 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3342 rstreason = BANDLIM_RST_OPENPORT; 3343 goto dropwithreset; 3344 } 3345 #ifdef TCPDEBUG 3346 if (so->so_options & SO_DEBUG) 3347 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3348 &tcp_savetcp, 0); 3349 #endif 3350 TCP_PROBE3(debug__input, tp, th, m); 3351 tp->t_flags |= TF_ACKNOW; 3352 (void) tp->t_fb->tfb_tcp_output(tp); 3353 tcp_handle_wakeup(tp, so); 3354 INP_WUNLOCK(tp->t_inpcb); 3355 m_freem(m); 3356 return; 3357 3358 dropwithreset: 3359 if (tp != NULL) { 3360 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3361 tcp_handle_wakeup(tp, so); 3362 INP_WUNLOCK(tp->t_inpcb); 3363 } else 3364 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3365 return; 3366 3367 drop: 3368 /* 3369 * Drop space held by incoming segment and return. 3370 */ 3371 #ifdef TCPDEBUG 3372 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3373 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3374 &tcp_savetcp, 0); 3375 #endif 3376 TCP_PROBE3(debug__input, tp, th, m); 3377 if (tp != NULL) { 3378 tcp_handle_wakeup(tp, so); 3379 INP_WUNLOCK(tp->t_inpcb); 3380 } 3381 m_freem(m); 3382 } 3383 3384 /* 3385 * Issue RST and make ACK acceptable to originator of segment. 3386 * The mbuf must still include the original packet header. 3387 * tp may be NULL. 3388 */ 3389 void 3390 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3391 int tlen, int rstreason) 3392 { 3393 #ifdef INET 3394 struct ip *ip; 3395 #endif 3396 #ifdef INET6 3397 struct ip6_hdr *ip6; 3398 #endif 3399 3400 if (tp != NULL) { 3401 INP_LOCK_ASSERT(tp->t_inpcb); 3402 } 3403 3404 /* Don't bother if destination was broadcast/multicast. */ 3405 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3406 goto drop; 3407 #ifdef INET6 3408 if (mtod(m, struct ip *)->ip_v == 6) { 3409 ip6 = mtod(m, struct ip6_hdr *); 3410 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3411 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3412 goto drop; 3413 /* IPv6 anycast check is done at tcp6_input() */ 3414 } 3415 #endif 3416 #if defined(INET) && defined(INET6) 3417 else 3418 #endif 3419 #ifdef INET 3420 { 3421 ip = mtod(m, struct ip *); 3422 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3423 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3424 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3425 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3426 goto drop; 3427 } 3428 #endif 3429 3430 /* Perform bandwidth limiting. */ 3431 if (badport_bandlim(rstreason) < 0) 3432 goto drop; 3433 3434 /* tcp_respond consumes the mbuf chain. */ 3435 if (th->th_flags & TH_ACK) { 3436 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3437 th->th_ack, TH_RST); 3438 } else { 3439 if (th->th_flags & TH_SYN) 3440 tlen++; 3441 if (th->th_flags & TH_FIN) 3442 tlen++; 3443 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3444 (tcp_seq)0, TH_RST|TH_ACK); 3445 } 3446 return; 3447 drop: 3448 m_freem(m); 3449 } 3450 3451 /* 3452 * Parse TCP options and place in tcpopt. 3453 */ 3454 void 3455 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3456 { 3457 int opt, optlen; 3458 3459 to->to_flags = 0; 3460 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3461 opt = cp[0]; 3462 if (opt == TCPOPT_EOL) 3463 break; 3464 if (opt == TCPOPT_NOP) 3465 optlen = 1; 3466 else { 3467 if (cnt < 2) 3468 break; 3469 optlen = cp[1]; 3470 if (optlen < 2 || optlen > cnt) 3471 break; 3472 } 3473 switch (opt) { 3474 case TCPOPT_MAXSEG: 3475 if (optlen != TCPOLEN_MAXSEG) 3476 continue; 3477 if (!(flags & TO_SYN)) 3478 continue; 3479 to->to_flags |= TOF_MSS; 3480 bcopy((char *)cp + 2, 3481 (char *)&to->to_mss, sizeof(to->to_mss)); 3482 to->to_mss = ntohs(to->to_mss); 3483 break; 3484 case TCPOPT_WINDOW: 3485 if (optlen != TCPOLEN_WINDOW) 3486 continue; 3487 if (!(flags & TO_SYN)) 3488 continue; 3489 to->to_flags |= TOF_SCALE; 3490 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3491 break; 3492 case TCPOPT_TIMESTAMP: 3493 if (optlen != TCPOLEN_TIMESTAMP) 3494 continue; 3495 to->to_flags |= TOF_TS; 3496 bcopy((char *)cp + 2, 3497 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3498 to->to_tsval = ntohl(to->to_tsval); 3499 bcopy((char *)cp + 6, 3500 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3501 to->to_tsecr = ntohl(to->to_tsecr); 3502 break; 3503 case TCPOPT_SIGNATURE: 3504 /* 3505 * In order to reply to a host which has set the 3506 * TCP_SIGNATURE option in its initial SYN, we have 3507 * to record the fact that the option was observed 3508 * here for the syncache code to perform the correct 3509 * response. 3510 */ 3511 if (optlen != TCPOLEN_SIGNATURE) 3512 continue; 3513 to->to_flags |= TOF_SIGNATURE; 3514 to->to_signature = cp + 2; 3515 break; 3516 case TCPOPT_SACK_PERMITTED: 3517 if (optlen != TCPOLEN_SACK_PERMITTED) 3518 continue; 3519 if (!(flags & TO_SYN)) 3520 continue; 3521 if (!V_tcp_do_sack) 3522 continue; 3523 to->to_flags |= TOF_SACKPERM; 3524 break; 3525 case TCPOPT_SACK: 3526 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3527 continue; 3528 if (flags & TO_SYN) 3529 continue; 3530 to->to_flags |= TOF_SACK; 3531 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3532 to->to_sacks = cp + 2; 3533 TCPSTAT_INC(tcps_sack_rcv_blocks); 3534 break; 3535 case TCPOPT_FAST_OPEN: 3536 /* 3537 * Cookie length validation is performed by the 3538 * server side cookie checking code or the client 3539 * side cookie cache update code. 3540 */ 3541 if (!(flags & TO_SYN)) 3542 continue; 3543 if (!V_tcp_fastopen_client_enable && 3544 !V_tcp_fastopen_server_enable) 3545 continue; 3546 to->to_flags |= TOF_FASTOPEN; 3547 to->to_tfo_len = optlen - 2; 3548 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3549 break; 3550 default: 3551 continue; 3552 } 3553 } 3554 } 3555 3556 /* 3557 * Pull out of band byte out of a segment so 3558 * it doesn't appear in the user's data queue. 3559 * It is still reflected in the segment length for 3560 * sequencing purposes. 3561 */ 3562 void 3563 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3564 int off) 3565 { 3566 int cnt = off + th->th_urp - 1; 3567 3568 while (cnt >= 0) { 3569 if (m->m_len > cnt) { 3570 char *cp = mtod(m, caddr_t) + cnt; 3571 struct tcpcb *tp = sototcpcb(so); 3572 3573 INP_WLOCK_ASSERT(tp->t_inpcb); 3574 3575 tp->t_iobc = *cp; 3576 tp->t_oobflags |= TCPOOB_HAVEDATA; 3577 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3578 m->m_len--; 3579 if (m->m_flags & M_PKTHDR) 3580 m->m_pkthdr.len--; 3581 return; 3582 } 3583 cnt -= m->m_len; 3584 m = m->m_next; 3585 if (m == NULL) 3586 break; 3587 } 3588 panic("tcp_pulloutofband"); 3589 } 3590 3591 /* 3592 * Collect new round-trip time estimate 3593 * and update averages and current timeout. 3594 */ 3595 void 3596 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3597 { 3598 int delta; 3599 3600 INP_WLOCK_ASSERT(tp->t_inpcb); 3601 3602 TCPSTAT_INC(tcps_rttupdated); 3603 tp->t_rttupdated++; 3604 #ifdef STATS 3605 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3606 imax(0, rtt * 1000 / hz)); 3607 #endif 3608 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3609 /* 3610 * srtt is stored as fixed point with 5 bits after the 3611 * binary point (i.e., scaled by 8). The following magic 3612 * is equivalent to the smoothing algorithm in rfc793 with 3613 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3614 * point). Adjust rtt to origin 0. 3615 */ 3616 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3617 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3618 3619 if ((tp->t_srtt += delta) <= 0) 3620 tp->t_srtt = 1; 3621 3622 /* 3623 * We accumulate a smoothed rtt variance (actually, a 3624 * smoothed mean difference), then set the retransmit 3625 * timer to smoothed rtt + 4 times the smoothed variance. 3626 * rttvar is stored as fixed point with 4 bits after the 3627 * binary point (scaled by 16). The following is 3628 * equivalent to rfc793 smoothing with an alpha of .75 3629 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3630 * rfc793's wired-in beta. 3631 */ 3632 if (delta < 0) 3633 delta = -delta; 3634 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3635 if ((tp->t_rttvar += delta) <= 0) 3636 tp->t_rttvar = 1; 3637 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3638 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3639 } else { 3640 /* 3641 * No rtt measurement yet - use the unsmoothed rtt. 3642 * Set the variance to half the rtt (so our first 3643 * retransmit happens at 3*rtt). 3644 */ 3645 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3646 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3647 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3648 } 3649 tp->t_rtttime = 0; 3650 tp->t_rxtshift = 0; 3651 3652 /* 3653 * the retransmit should happen at rtt + 4 * rttvar. 3654 * Because of the way we do the smoothing, srtt and rttvar 3655 * will each average +1/2 tick of bias. When we compute 3656 * the retransmit timer, we want 1/2 tick of rounding and 3657 * 1 extra tick because of +-1/2 tick uncertainty in the 3658 * firing of the timer. The bias will give us exactly the 3659 * 1.5 tick we need. But, because the bias is 3660 * statistical, we have to test that we don't drop below 3661 * the minimum feasible timer (which is 2 ticks). 3662 */ 3663 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3664 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3665 3666 /* 3667 * We received an ack for a packet that wasn't retransmitted; 3668 * it is probably safe to discard any error indications we've 3669 * received recently. This isn't quite right, but close enough 3670 * for now (a route might have failed after we sent a segment, 3671 * and the return path might not be symmetrical). 3672 */ 3673 tp->t_softerror = 0; 3674 } 3675 3676 /* 3677 * Determine a reasonable value for maxseg size. 3678 * If the route is known, check route for mtu. 3679 * If none, use an mss that can be handled on the outgoing interface 3680 * without forcing IP to fragment. If no route is found, route has no mtu, 3681 * or the destination isn't local, use a default, hopefully conservative 3682 * size (usually 512 or the default IP max size, but no more than the mtu 3683 * of the interface), as we can't discover anything about intervening 3684 * gateways or networks. We also initialize the congestion/slow start 3685 * window to be a single segment if the destination isn't local. 3686 * While looking at the routing entry, we also initialize other path-dependent 3687 * parameters from pre-set or cached values in the routing entry. 3688 * 3689 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3690 * IP options, e.g. IPSEC data, since length of this data may vary, and 3691 * thus it is calculated for every segment separately in tcp_output(). 3692 * 3693 * NOTE that this routine is only called when we process an incoming 3694 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3695 * settings are handled in tcp_mssopt(). 3696 */ 3697 void 3698 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3699 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3700 { 3701 int mss = 0; 3702 uint32_t maxmtu = 0; 3703 struct inpcb *inp = tp->t_inpcb; 3704 struct hc_metrics_lite metrics; 3705 #ifdef INET6 3706 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3707 size_t min_protoh = isipv6 ? 3708 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3709 sizeof (struct tcpiphdr); 3710 #else 3711 size_t min_protoh = sizeof(struct tcpiphdr); 3712 #endif 3713 3714 INP_WLOCK_ASSERT(tp->t_inpcb); 3715 3716 if (tp->t_port) 3717 min_protoh += V_tcp_udp_tunneling_overhead; 3718 if (mtuoffer != -1) { 3719 KASSERT(offer == -1, ("%s: conflict", __func__)); 3720 offer = mtuoffer - min_protoh; 3721 } 3722 3723 /* Initialize. */ 3724 #ifdef INET6 3725 if (isipv6) { 3726 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3727 tp->t_maxseg = V_tcp_v6mssdflt; 3728 } 3729 #endif 3730 #if defined(INET) && defined(INET6) 3731 else 3732 #endif 3733 #ifdef INET 3734 { 3735 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3736 tp->t_maxseg = V_tcp_mssdflt; 3737 } 3738 #endif 3739 3740 /* 3741 * No route to sender, stay with default mss and return. 3742 */ 3743 if (maxmtu == 0) { 3744 /* 3745 * In case we return early we need to initialize metrics 3746 * to a defined state as tcp_hc_get() would do for us 3747 * if there was no cache hit. 3748 */ 3749 if (metricptr != NULL) 3750 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3751 return; 3752 } 3753 3754 /* What have we got? */ 3755 switch (offer) { 3756 case 0: 3757 /* 3758 * Offer == 0 means that there was no MSS on the SYN 3759 * segment, in this case we use tcp_mssdflt as 3760 * already assigned to t_maxseg above. 3761 */ 3762 offer = tp->t_maxseg; 3763 break; 3764 3765 case -1: 3766 /* 3767 * Offer == -1 means that we didn't receive SYN yet. 3768 */ 3769 /* FALLTHROUGH */ 3770 3771 default: 3772 /* 3773 * Prevent DoS attack with too small MSS. Round up 3774 * to at least minmss. 3775 */ 3776 offer = max(offer, V_tcp_minmss); 3777 } 3778 3779 /* 3780 * rmx information is now retrieved from tcp_hostcache. 3781 */ 3782 tcp_hc_get(&inp->inp_inc, &metrics); 3783 if (metricptr != NULL) 3784 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3785 3786 /* 3787 * If there's a discovered mtu in tcp hostcache, use it. 3788 * Else, use the link mtu. 3789 */ 3790 if (metrics.rmx_mtu) 3791 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3792 else { 3793 #ifdef INET6 3794 if (isipv6) { 3795 mss = maxmtu - min_protoh; 3796 if (!V_path_mtu_discovery && 3797 !in6_localaddr(&inp->in6p_faddr)) 3798 mss = min(mss, V_tcp_v6mssdflt); 3799 } 3800 #endif 3801 #if defined(INET) && defined(INET6) 3802 else 3803 #endif 3804 #ifdef INET 3805 { 3806 mss = maxmtu - min_protoh; 3807 if (!V_path_mtu_discovery && 3808 !in_localaddr(inp->inp_faddr)) 3809 mss = min(mss, V_tcp_mssdflt); 3810 } 3811 #endif 3812 /* 3813 * XXX - The above conditional (mss = maxmtu - min_protoh) 3814 * probably violates the TCP spec. 3815 * The problem is that, since we don't know the 3816 * other end's MSS, we are supposed to use a conservative 3817 * default. But, if we do that, then MTU discovery will 3818 * never actually take place, because the conservative 3819 * default is much less than the MTUs typically seen 3820 * on the Internet today. For the moment, we'll sweep 3821 * this under the carpet. 3822 * 3823 * The conservative default might not actually be a problem 3824 * if the only case this occurs is when sending an initial 3825 * SYN with options and data to a host we've never talked 3826 * to before. Then, they will reply with an MSS value which 3827 * will get recorded and the new parameters should get 3828 * recomputed. For Further Study. 3829 */ 3830 } 3831 mss = min(mss, offer); 3832 3833 /* 3834 * Sanity check: make sure that maxseg will be large 3835 * enough to allow some data on segments even if the 3836 * all the option space is used (40bytes). Otherwise 3837 * funny things may happen in tcp_output. 3838 * 3839 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3840 */ 3841 mss = max(mss, 64); 3842 3843 tp->t_maxseg = mss; 3844 } 3845 3846 void 3847 tcp_mss(struct tcpcb *tp, int offer) 3848 { 3849 int mss; 3850 uint32_t bufsize; 3851 struct inpcb *inp; 3852 struct socket *so; 3853 struct hc_metrics_lite metrics; 3854 struct tcp_ifcap cap; 3855 3856 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3857 3858 bzero(&cap, sizeof(cap)); 3859 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3860 3861 mss = tp->t_maxseg; 3862 inp = tp->t_inpcb; 3863 3864 /* 3865 * If there's a pipesize, change the socket buffer to that size, 3866 * don't change if sb_hiwat is different than default (then it 3867 * has been changed on purpose with setsockopt). 3868 * Make the socket buffers an integral number of mss units; 3869 * if the mss is larger than the socket buffer, decrease the mss. 3870 */ 3871 so = inp->inp_socket; 3872 SOCKBUF_LOCK(&so->so_snd); 3873 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3874 bufsize = metrics.rmx_sendpipe; 3875 else 3876 bufsize = so->so_snd.sb_hiwat; 3877 if (bufsize < mss) 3878 mss = bufsize; 3879 else { 3880 bufsize = roundup(bufsize, mss); 3881 if (bufsize > sb_max) 3882 bufsize = sb_max; 3883 if (bufsize > so->so_snd.sb_hiwat) 3884 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3885 } 3886 SOCKBUF_UNLOCK(&so->so_snd); 3887 /* 3888 * Sanity check: make sure that maxseg will be large 3889 * enough to allow some data on segments even if the 3890 * all the option space is used (40bytes). Otherwise 3891 * funny things may happen in tcp_output. 3892 * 3893 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3894 */ 3895 tp->t_maxseg = max(mss, 64); 3896 3897 SOCKBUF_LOCK(&so->so_rcv); 3898 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3899 bufsize = metrics.rmx_recvpipe; 3900 else 3901 bufsize = so->so_rcv.sb_hiwat; 3902 if (bufsize > mss) { 3903 bufsize = roundup(bufsize, mss); 3904 if (bufsize > sb_max) 3905 bufsize = sb_max; 3906 if (bufsize > so->so_rcv.sb_hiwat) 3907 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3908 } 3909 SOCKBUF_UNLOCK(&so->so_rcv); 3910 3911 /* Check the interface for TSO capabilities. */ 3912 if (cap.ifcap & CSUM_TSO) { 3913 tp->t_flags |= TF_TSO; 3914 tp->t_tsomax = cap.tsomax; 3915 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3916 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3917 } 3918 } 3919 3920 /* 3921 * Determine the MSS option to send on an outgoing SYN. 3922 */ 3923 int 3924 tcp_mssopt(struct in_conninfo *inc) 3925 { 3926 int mss = 0; 3927 uint32_t thcmtu = 0; 3928 uint32_t maxmtu = 0; 3929 size_t min_protoh; 3930 3931 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3932 3933 #ifdef INET6 3934 if (inc->inc_flags & INC_ISIPV6) { 3935 mss = V_tcp_v6mssdflt; 3936 maxmtu = tcp_maxmtu6(inc, NULL); 3937 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3938 } 3939 #endif 3940 #if defined(INET) && defined(INET6) 3941 else 3942 #endif 3943 #ifdef INET 3944 { 3945 mss = V_tcp_mssdflt; 3946 maxmtu = tcp_maxmtu(inc, NULL); 3947 min_protoh = sizeof(struct tcpiphdr); 3948 } 3949 #endif 3950 #if defined(INET6) || defined(INET) 3951 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3952 #endif 3953 3954 if (maxmtu && thcmtu) 3955 mss = min(maxmtu, thcmtu) - min_protoh; 3956 else if (maxmtu || thcmtu) 3957 mss = max(maxmtu, thcmtu) - min_protoh; 3958 3959 return (mss); 3960 } 3961 3962 void 3963 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3964 { 3965 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3966 int maxseg = tcp_maxseg(tp); 3967 3968 INP_WLOCK_ASSERT(tp->t_inpcb); 3969 3970 /* 3971 * Compute the amount of data that this ACK is indicating 3972 * (del_data) and an estimate of how many bytes are in the 3973 * network. 3974 */ 3975 del_data = tp->sackhint.delivered_data; 3976 if (V_tcp_do_newsack) 3977 pipe = tcp_compute_pipe(tp); 3978 else 3979 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; 3980 tp->sackhint.prr_delivered += del_data; 3981 /* 3982 * Proportional Rate Reduction 3983 */ 3984 if (pipe >= tp->snd_ssthresh) { 3985 if (tp->sackhint.recover_fs == 0) 3986 tp->sackhint.recover_fs = 3987 imax(1, tp->snd_nxt - tp->snd_una); 3988 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 3989 tp->snd_ssthresh, tp->sackhint.recover_fs) - 3990 tp->sackhint.prr_out; 3991 } else { 3992 if (V_tcp_do_prr_conservative) 3993 limit = tp->sackhint.prr_delivered - 3994 tp->sackhint.prr_out; 3995 else 3996 limit = imax(tp->sackhint.prr_delivered - 3997 tp->sackhint.prr_out, del_data) + 3998 maxseg; 3999 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 4000 } 4001 snd_cnt = imax(snd_cnt, 0) / maxseg; 4002 /* 4003 * Send snd_cnt new data into the network in response to this ack. 4004 * If there is going to be a SACK retransmission, adjust snd_cwnd 4005 * accordingly. 4006 */ 4007 if (IN_FASTRECOVERY(tp->t_flags)) { 4008 tp->snd_cwnd = imax(maxseg, tp->snd_nxt - tp->snd_recover + 4009 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg)); 4010 } else if (IN_CONGRECOVERY(tp->t_flags)) 4011 tp->snd_cwnd = imax(maxseg, pipe - del_data + 4012 (snd_cnt * maxseg)); 4013 } 4014 4015 /* 4016 * On a partial ack arrives, force the retransmission of the 4017 * next unacknowledged segment. Do not clear tp->t_dupacks. 4018 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4019 * be started again. 4020 */ 4021 void 4022 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4023 { 4024 tcp_seq onxt = tp->snd_nxt; 4025 uint32_t ocwnd = tp->snd_cwnd; 4026 u_int maxseg = tcp_maxseg(tp); 4027 4028 INP_WLOCK_ASSERT(tp->t_inpcb); 4029 4030 tcp_timer_activate(tp, TT_REXMT, 0); 4031 tp->t_rtttime = 0; 4032 tp->snd_nxt = th->th_ack; 4033 /* 4034 * Set snd_cwnd to one segment beyond acknowledged offset. 4035 * (tp->snd_una has not yet been updated when this function is called.) 4036 */ 4037 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4038 tp->t_flags |= TF_ACKNOW; 4039 (void) tp->t_fb->tfb_tcp_output(tp); 4040 tp->snd_cwnd = ocwnd; 4041 if (SEQ_GT(onxt, tp->snd_nxt)) 4042 tp->snd_nxt = onxt; 4043 /* 4044 * Partial window deflation. Relies on fact that tp->snd_una 4045 * not updated yet. 4046 */ 4047 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4048 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4049 else 4050 tp->snd_cwnd = 0; 4051 tp->snd_cwnd += maxseg; 4052 } 4053 4054 int 4055 tcp_compute_pipe(struct tcpcb *tp) 4056 { 4057 return (tp->snd_max - tp->snd_una + 4058 tp->sackhint.sack_bytes_rexmit - 4059 tp->sackhint.sacked_bytes); 4060 } 4061 4062 uint32_t 4063 tcp_compute_initwnd(uint32_t maxseg) 4064 { 4065 /* 4066 * Calculate the Initial Window, also used as Restart Window 4067 * 4068 * RFC5681 Section 3.1 specifies the default conservative values. 4069 * RFC3390 specifies slightly more aggressive values. 4070 * RFC6928 increases it to ten segments. 4071 * Support for user specified value for initial flight size. 4072 */ 4073 if (V_tcp_initcwnd_segments) 4074 return min(V_tcp_initcwnd_segments * maxseg, 4075 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4076 else if (V_tcp_do_rfc3390) 4077 return min(4 * maxseg, max(2 * maxseg, 4380)); 4078 else { 4079 /* Per RFC5681 Section 3.1 */ 4080 if (maxseg > 2190) 4081 return (2 * maxseg); 4082 else if (maxseg > 1095) 4083 return (3 * maxseg); 4084 else 4085 return (4 * maxseg); 4086 } 4087 } 4088