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(bool, blackhole_local) = false; 148 #define V_blackhole_local VNET(blackhole_local) 149 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 150 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 151 "Enforce net.inet.tcp.blackhole for locally originated packets"); 152 153 VNET_DEFINE(int, tcp_delack_enabled) = 1; 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 155 &VNET_NAME(tcp_delack_enabled), 0, 156 "Delay ACK to try and piggyback it onto a data packet"); 157 158 VNET_DEFINE(int, drop_synfin) = 0; 159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 160 &VNET_NAME(drop_synfin), 0, 161 "Drop TCP packets with SYN+FIN set"); 162 163 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 165 &VNET_NAME(tcp_do_prr_conservative), 0, 166 "Do conservative Proportional Rate Reduction"); 167 168 VNET_DEFINE(int, tcp_do_prr) = 1; 169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 170 &VNET_NAME(tcp_do_prr), 1, 171 "Enable Proportional Rate Reduction per RFC 6937"); 172 173 VNET_DEFINE(int, tcp_do_lrd) = 0; 174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW, 175 &VNET_NAME(tcp_do_lrd), 1, 176 "Perform Lost Retransmission Detection"); 177 178 VNET_DEFINE(int, tcp_do_newcwv) = 0; 179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 180 &VNET_NAME(tcp_do_newcwv), 0, 181 "Enable New Congestion Window Validation per RFC7661"); 182 183 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 184 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 185 &VNET_NAME(tcp_do_rfc3042), 0, 186 "Enable RFC 3042 (Limited Transmit)"); 187 188 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 190 &VNET_NAME(tcp_do_rfc3390), 0, 191 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 192 193 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 195 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 196 "Slow-start flight size (initial congestion window) in number of segments"); 197 198 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 200 &VNET_NAME(tcp_do_rfc3465), 0, 201 "Enable RFC 3465 (Appropriate Byte Counting)"); 202 203 VNET_DEFINE(int, tcp_abc_l_var) = 2; 204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 205 &VNET_NAME(tcp_abc_l_var), 2, 206 "Cap the max cwnd increment during slow-start to this number of segments"); 207 208 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 209 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 210 "TCP ECN"); 211 212 VNET_DEFINE(int, tcp_do_ecn) = 2; 213 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 214 &VNET_NAME(tcp_do_ecn), 0, 215 "TCP ECN support"); 216 217 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 218 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 219 &VNET_NAME(tcp_ecn_maxretries), 0, 220 "Max retries before giving up on ECN"); 221 222 VNET_DEFINE(int, tcp_insecure_syn) = 0; 223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 224 &VNET_NAME(tcp_insecure_syn), 0, 225 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 226 227 VNET_DEFINE(int, tcp_insecure_rst) = 0; 228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 229 &VNET_NAME(tcp_insecure_rst), 0, 230 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 231 232 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 233 #define V_tcp_recvspace VNET(tcp_recvspace) 234 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 235 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 236 237 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 238 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 239 &VNET_NAME(tcp_do_autorcvbuf), 0, 240 "Enable automatic receive buffer sizing"); 241 242 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 243 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 244 &VNET_NAME(tcp_autorcvbuf_max), 0, 245 "Max size of automatic receive buffer"); 246 247 VNET_DEFINE(struct inpcbinfo, tcbinfo); 248 249 /* 250 * TCP statistics are stored in an array of counter(9)s, which size matches 251 * size of struct tcpstat. TCP running connection count is a regular array. 252 */ 253 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 254 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 255 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 256 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 257 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 258 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 259 "TCP connection counts by TCP state"); 260 261 static void 262 tcp_vnet_init(const void *unused) 263 { 264 265 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 266 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 267 } 268 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 269 tcp_vnet_init, NULL); 270 271 #ifdef VIMAGE 272 static void 273 tcp_vnet_uninit(const void *unused) 274 { 275 276 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 277 VNET_PCPUSTAT_FREE(tcpstat); 278 } 279 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 280 tcp_vnet_uninit, NULL); 281 #endif /* VIMAGE */ 282 283 /* 284 * Kernel module interface for updating tcpstat. The first argument is an index 285 * into tcpstat treated as an array. 286 */ 287 void 288 kmod_tcpstat_add(int statnum, int val) 289 { 290 291 counter_u64_add(VNET(tcpstat)[statnum], val); 292 } 293 294 #ifdef TCP_HHOOK 295 /* 296 * Wrapper for the TCP established input helper hook. 297 */ 298 void 299 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 300 { 301 struct tcp_hhook_data hhook_data; 302 303 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 304 hhook_data.tp = tp; 305 hhook_data.th = th; 306 hhook_data.to = to; 307 308 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 309 tp->osd); 310 } 311 } 312 #endif 313 314 /* 315 * CC wrapper hook functions 316 */ 317 void 318 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 319 uint16_t type) 320 { 321 #ifdef STATS 322 int32_t gput; 323 #endif 324 325 INP_WLOCK_ASSERT(tp->t_inpcb); 326 327 tp->ccv->nsegs = nsegs; 328 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 329 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 330 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 331 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 332 tp->ccv->flags |= CCF_CWND_LIMITED; 333 else 334 tp->ccv->flags &= ~CCF_CWND_LIMITED; 335 336 if (type == CC_ACK) { 337 #ifdef STATS 338 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 339 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 340 if (!IN_RECOVERY(tp->t_flags)) 341 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 342 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 343 if ((tp->t_flags & TF_GPUTINPROG) && 344 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 345 /* 346 * Compute goodput in bits per millisecond. 347 */ 348 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) / 349 max(1, tcp_ts_getticks() - tp->gput_ts); 350 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 351 gput); 352 /* 353 * XXXLAS: This is a temporary hack, and should be 354 * chained off VOI_TCP_GPUT when stats(9) grows an API 355 * to deal with chained VOIs. 356 */ 357 if (tp->t_stats_gput_prev > 0) 358 stats_voi_update_abs_s32(tp->t_stats, 359 VOI_TCP_GPUT_ND, 360 ((gput - tp->t_stats_gput_prev) * 100) / 361 tp->t_stats_gput_prev); 362 tp->t_flags &= ~TF_GPUTINPROG; 363 tp->t_stats_gput_prev = gput; 364 } 365 #endif /* STATS */ 366 if (tp->snd_cwnd > tp->snd_ssthresh) { 367 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 368 if (tp->t_bytes_acked >= tp->snd_cwnd) { 369 tp->t_bytes_acked -= tp->snd_cwnd; 370 tp->ccv->flags |= CCF_ABC_SENTAWND; 371 } 372 } else { 373 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 374 tp->t_bytes_acked = 0; 375 } 376 } 377 378 if (CC_ALGO(tp)->ack_received != NULL) { 379 /* XXXLAS: Find a way to live without this */ 380 tp->ccv->curack = th->th_ack; 381 CC_ALGO(tp)->ack_received(tp->ccv, type); 382 } 383 #ifdef STATS 384 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 385 #endif 386 } 387 388 void 389 cc_conn_init(struct tcpcb *tp) 390 { 391 struct hc_metrics_lite metrics; 392 struct inpcb *inp = tp->t_inpcb; 393 u_int maxseg; 394 int rtt; 395 396 INP_WLOCK_ASSERT(tp->t_inpcb); 397 398 tcp_hc_get(&inp->inp_inc, &metrics); 399 maxseg = tcp_maxseg(tp); 400 401 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 402 tp->t_srtt = rtt; 403 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 404 TCPSTAT_INC(tcps_usedrtt); 405 if (metrics.rmx_rttvar) { 406 tp->t_rttvar = metrics.rmx_rttvar; 407 TCPSTAT_INC(tcps_usedrttvar); 408 } else { 409 /* default variation is +- 1 rtt */ 410 tp->t_rttvar = 411 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 412 } 413 TCPT_RANGESET(tp->t_rxtcur, 414 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 415 tp->t_rttmin, TCPTV_REXMTMAX); 416 } 417 if (metrics.rmx_ssthresh) { 418 /* 419 * There's some sort of gateway or interface 420 * buffer limit on the path. Use this to set 421 * the slow start threshold, but set the 422 * threshold to no less than 2*mss. 423 */ 424 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 425 TCPSTAT_INC(tcps_usedssthresh); 426 } 427 428 /* 429 * Set the initial slow-start flight size. 430 * 431 * If a SYN or SYN/ACK was lost and retransmitted, we have to 432 * reduce the initial CWND to one segment as congestion is likely 433 * requiring us to be cautious. 434 */ 435 if (tp->snd_cwnd == 1) 436 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 437 else 438 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 439 440 if (CC_ALGO(tp)->conn_init != NULL) 441 CC_ALGO(tp)->conn_init(tp->ccv); 442 } 443 444 void inline 445 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 446 { 447 INP_WLOCK_ASSERT(tp->t_inpcb); 448 449 #ifdef STATS 450 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 451 #endif 452 453 switch(type) { 454 case CC_NDUPACK: 455 if (!IN_FASTRECOVERY(tp->t_flags)) { 456 tp->snd_recover = tp->snd_max; 457 if (tp->t_flags2 & TF2_ECN_PERMIT) 458 tp->t_flags2 |= TF2_ECN_SND_CWR; 459 } 460 break; 461 case CC_ECN: 462 if (!IN_CONGRECOVERY(tp->t_flags) || 463 /* 464 * Allow ECN reaction on ACK to CWR, if 465 * that data segment was also CE marked. 466 */ 467 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 468 EXIT_CONGRECOVERY(tp->t_flags); 469 TCPSTAT_INC(tcps_ecn_rcwnd); 470 tp->snd_recover = tp->snd_max + 1; 471 if (tp->t_flags2 & TF2_ECN_PERMIT) 472 tp->t_flags2 |= TF2_ECN_SND_CWR; 473 } 474 break; 475 case CC_RTO: 476 tp->t_dupacks = 0; 477 tp->t_bytes_acked = 0; 478 EXIT_RECOVERY(tp->t_flags); 479 if (tp->t_flags2 & TF2_ECN_PERMIT) 480 tp->t_flags2 |= TF2_ECN_SND_CWR; 481 break; 482 case CC_RTO_ERR: 483 TCPSTAT_INC(tcps_sndrexmitbad); 484 /* RTO was unnecessary, so reset everything. */ 485 tp->snd_cwnd = tp->snd_cwnd_prev; 486 tp->snd_ssthresh = tp->snd_ssthresh_prev; 487 tp->snd_recover = tp->snd_recover_prev; 488 if (tp->t_flags & TF_WASFRECOVERY) 489 ENTER_FASTRECOVERY(tp->t_flags); 490 if (tp->t_flags & TF_WASCRECOVERY) 491 ENTER_CONGRECOVERY(tp->t_flags); 492 tp->snd_nxt = tp->snd_max; 493 tp->t_flags &= ~TF_PREVVALID; 494 tp->t_badrxtwin = 0; 495 break; 496 } 497 498 if (CC_ALGO(tp)->cong_signal != NULL) { 499 if (th != NULL) 500 tp->ccv->curack = th->th_ack; 501 CC_ALGO(tp)->cong_signal(tp->ccv, type); 502 } 503 } 504 505 void inline 506 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 507 { 508 INP_WLOCK_ASSERT(tp->t_inpcb); 509 510 /* XXXLAS: KASSERT that we're in recovery? */ 511 512 if (CC_ALGO(tp)->post_recovery != NULL) { 513 tp->ccv->curack = th->th_ack; 514 CC_ALGO(tp)->post_recovery(tp->ccv); 515 } 516 /* XXXLAS: EXIT_RECOVERY ? */ 517 tp->t_bytes_acked = 0; 518 tp->sackhint.delivered_data = 0; 519 tp->sackhint.prr_out = 0; 520 } 521 522 /* 523 * Indicate whether this ack should be delayed. We can delay the ack if 524 * following conditions are met: 525 * - There is no delayed ack timer in progress. 526 * - Our last ack wasn't a 0-sized window. We never want to delay 527 * the ack that opens up a 0-sized window. 528 * - LRO wasn't used for this segment. We make sure by checking that the 529 * segment size is not larger than the MSS. 530 */ 531 #define DELAY_ACK(tp, tlen) \ 532 ((!tcp_timer_active(tp, TT_DELACK) && \ 533 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 534 (tlen <= tp->t_maxseg) && \ 535 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 536 537 void inline 538 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 539 { 540 INP_WLOCK_ASSERT(tp->t_inpcb); 541 542 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 543 switch (iptos & IPTOS_ECN_MASK) { 544 case IPTOS_ECN_CE: 545 tp->ccv->flags |= CCF_IPHDR_CE; 546 break; 547 case IPTOS_ECN_ECT0: 548 /* FALLTHROUGH */ 549 case IPTOS_ECN_ECT1: 550 /* FALLTHROUGH */ 551 case IPTOS_ECN_NOTECT: 552 tp->ccv->flags &= ~CCF_IPHDR_CE; 553 break; 554 } 555 556 if (flags & TH_CWR) 557 tp->ccv->flags |= CCF_TCPHDR_CWR; 558 else 559 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 560 561 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 562 563 if (tp->ccv->flags & CCF_ACKNOW) { 564 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 565 tp->t_flags |= TF_ACKNOW; 566 } 567 } 568 } 569 570 void inline 571 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 572 { 573 cc_ecnpkt_handler_flags(tp, th->th_flags, iptos); 574 } 575 576 /* 577 * TCP input handling is split into multiple parts: 578 * tcp6_input is a thin wrapper around tcp_input for the extended 579 * ip6_protox[] call format in ip6_input 580 * tcp_input handles primary segment validation, inpcb lookup and 581 * SYN processing on listen sockets 582 * tcp_do_segment processes the ACK and text of the segment for 583 * establishing, established and closing connections 584 */ 585 #ifdef INET6 586 int 587 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 588 { 589 struct mbuf *m; 590 struct in6_ifaddr *ia6; 591 struct ip6_hdr *ip6; 592 593 m = *mp; 594 if (m->m_len < *offp + sizeof(struct tcphdr)) { 595 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 596 if (m == NULL) { 597 *mp = m; 598 TCPSTAT_INC(tcps_rcvshort); 599 return (IPPROTO_DONE); 600 } 601 } 602 603 /* 604 * draft-itojun-ipv6-tcp-to-anycast 605 * better place to put this in? 606 */ 607 ip6 = mtod(m, struct ip6_hdr *); 608 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 609 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 610 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 611 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 612 *mp = NULL; 613 return (IPPROTO_DONE); 614 } 615 616 *mp = m; 617 return (tcp_input_with_port(mp, offp, proto, port)); 618 } 619 620 int 621 tcp6_input(struct mbuf **mp, int *offp, int proto) 622 { 623 624 return(tcp6_input_with_port(mp, offp, proto, 0)); 625 } 626 #endif /* INET6 */ 627 628 int 629 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 630 { 631 struct mbuf *m = *mp; 632 struct tcphdr *th = NULL; 633 struct ip *ip = NULL; 634 struct inpcb *inp = NULL; 635 struct tcpcb *tp = NULL; 636 struct socket *so = NULL; 637 u_char *optp = NULL; 638 int off0; 639 int optlen = 0; 640 #ifdef INET 641 int len; 642 uint8_t ipttl; 643 #endif 644 int tlen = 0, off; 645 int drop_hdrlen; 646 int thflags; 647 int rstreason = 0; /* For badport_bandlim accounting purposes */ 648 int lookupflag; 649 uint8_t iptos; 650 struct m_tag *fwd_tag = NULL; 651 #ifdef INET6 652 struct ip6_hdr *ip6 = NULL; 653 int isipv6; 654 #else 655 const void *ip6 = NULL; 656 #endif /* INET6 */ 657 struct tcpopt to; /* options in this segment */ 658 char *s = NULL; /* address and port logging */ 659 #ifdef TCPDEBUG 660 /* 661 * The size of tcp_saveipgen must be the size of the max ip header, 662 * now IPv6. 663 */ 664 u_char tcp_saveipgen[IP6_HDR_LEN]; 665 struct tcphdr tcp_savetcp; 666 short ostate = 0; 667 #endif 668 669 NET_EPOCH_ASSERT(); 670 671 #ifdef INET6 672 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 673 #endif 674 675 off0 = *offp; 676 m = *mp; 677 *mp = NULL; 678 to.to_flags = 0; 679 TCPSTAT_INC(tcps_rcvtotal); 680 681 #ifdef INET6 682 if (isipv6) { 683 ip6 = mtod(m, struct ip6_hdr *); 684 th = (struct tcphdr *)((caddr_t)ip6 + off0); 685 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 686 if (port) 687 goto skip6_csum; 688 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 689 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 690 th->th_sum = m->m_pkthdr.csum_data; 691 else 692 th->th_sum = in6_cksum_pseudo(ip6, tlen, 693 IPPROTO_TCP, m->m_pkthdr.csum_data); 694 th->th_sum ^= 0xffff; 695 } else 696 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 697 if (th->th_sum) { 698 TCPSTAT_INC(tcps_rcvbadsum); 699 goto drop; 700 } 701 skip6_csum: 702 /* 703 * Be proactive about unspecified IPv6 address in source. 704 * As we use all-zero to indicate unbounded/unconnected pcb, 705 * unspecified IPv6 address can be used to confuse us. 706 * 707 * Note that packets with unspecified IPv6 destination is 708 * already dropped in ip6_input. 709 */ 710 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 711 /* XXX stat */ 712 goto drop; 713 } 714 iptos = IPV6_TRAFFIC_CLASS(ip6); 715 } 716 #endif 717 #if defined(INET) && defined(INET6) 718 else 719 #endif 720 #ifdef INET 721 { 722 /* 723 * Get IP and TCP header together in first mbuf. 724 * Note: IP leaves IP header in first mbuf. 725 */ 726 if (off0 > sizeof (struct ip)) { 727 ip_stripoptions(m); 728 off0 = sizeof(struct ip); 729 } 730 if (m->m_len < sizeof (struct tcpiphdr)) { 731 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 732 == NULL) { 733 TCPSTAT_INC(tcps_rcvshort); 734 return (IPPROTO_DONE); 735 } 736 } 737 ip = mtod(m, struct ip *); 738 th = (struct tcphdr *)((caddr_t)ip + off0); 739 tlen = ntohs(ip->ip_len) - off0; 740 741 iptos = ip->ip_tos; 742 if (port) 743 goto skip_csum; 744 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 745 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 746 th->th_sum = m->m_pkthdr.csum_data; 747 else 748 th->th_sum = in_pseudo(ip->ip_src.s_addr, 749 ip->ip_dst.s_addr, 750 htonl(m->m_pkthdr.csum_data + tlen + 751 IPPROTO_TCP)); 752 th->th_sum ^= 0xffff; 753 } else { 754 struct ipovly *ipov = (struct ipovly *)ip; 755 756 /* 757 * Checksum extended TCP header and data. 758 */ 759 len = off0 + tlen; 760 ipttl = ip->ip_ttl; 761 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 762 ipov->ih_len = htons(tlen); 763 th->th_sum = in_cksum(m, len); 764 /* Reset length for SDT probes. */ 765 ip->ip_len = htons(len); 766 /* Reset TOS bits */ 767 ip->ip_tos = iptos; 768 /* Re-initialization for later version check */ 769 ip->ip_ttl = ipttl; 770 ip->ip_v = IPVERSION; 771 ip->ip_hl = off0 >> 2; 772 } 773 skip_csum: 774 if (th->th_sum && (port == 0)) { 775 TCPSTAT_INC(tcps_rcvbadsum); 776 goto drop; 777 } 778 } 779 #endif /* INET */ 780 781 /* 782 * Check that TCP offset makes sense, 783 * pull out TCP options and adjust length. XXX 784 */ 785 off = th->th_off << 2; 786 if (off < sizeof (struct tcphdr) || off > tlen) { 787 TCPSTAT_INC(tcps_rcvbadoff); 788 goto drop; 789 } 790 tlen -= off; /* tlen is used instead of ti->ti_len */ 791 if (off > sizeof (struct tcphdr)) { 792 #ifdef INET6 793 if (isipv6) { 794 if (m->m_len < off0 + off) { 795 m = m_pullup(m, off0 + off); 796 if (m == NULL) { 797 TCPSTAT_INC(tcps_rcvshort); 798 return (IPPROTO_DONE); 799 } 800 } 801 ip6 = mtod(m, struct ip6_hdr *); 802 th = (struct tcphdr *)((caddr_t)ip6 + off0); 803 } 804 #endif 805 #if defined(INET) && defined(INET6) 806 else 807 #endif 808 #ifdef INET 809 { 810 if (m->m_len < sizeof(struct ip) + off) { 811 if ((m = m_pullup(m, sizeof (struct ip) + off)) 812 == NULL) { 813 TCPSTAT_INC(tcps_rcvshort); 814 return (IPPROTO_DONE); 815 } 816 ip = mtod(m, struct ip *); 817 th = (struct tcphdr *)((caddr_t)ip + off0); 818 } 819 } 820 #endif 821 optlen = off - sizeof (struct tcphdr); 822 optp = (u_char *)(th + 1); 823 } 824 thflags = th->th_flags; 825 826 /* 827 * Convert TCP protocol specific fields to host format. 828 */ 829 tcp_fields_to_host(th); 830 831 /* 832 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 833 */ 834 drop_hdrlen = off0 + off; 835 836 /* 837 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 838 */ 839 if ( 840 #ifdef INET6 841 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 842 #ifdef INET 843 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 844 #endif 845 #endif 846 #if defined(INET) && !defined(INET6) 847 (m->m_flags & M_IP_NEXTHOP) 848 #endif 849 ) 850 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 851 852 /* 853 * For initial SYN packets we don't need write lock on matching 854 * PCB, be it a listening one or a synchronized one. The packet 855 * shall not modify its state. 856 */ 857 lookupflag = (thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 858 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB; 859 findpcb: 860 #ifdef INET6 861 if (isipv6 && fwd_tag != NULL) { 862 struct sockaddr_in6 *next_hop6; 863 864 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 865 /* 866 * Transparently forwarded. Pretend to be the destination. 867 * Already got one like this? 868 */ 869 inp = in6_pcblookup_mbuf(&V_tcbinfo, 870 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 871 lookupflag, m->m_pkthdr.rcvif, m); 872 if (!inp) { 873 /* 874 * It's new. Try to find the ambushing socket. 875 * Because we've rewritten the destination address, 876 * any hardware-generated hash is ignored. 877 */ 878 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 879 th->th_sport, &next_hop6->sin6_addr, 880 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 881 th->th_dport, INPLOOKUP_WILDCARD | lookupflag, 882 m->m_pkthdr.rcvif); 883 } 884 } else if (isipv6) { 885 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 886 th->th_sport, &ip6->ip6_dst, th->th_dport, 887 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m); 888 } 889 #endif /* INET6 */ 890 #if defined(INET6) && defined(INET) 891 else 892 #endif 893 #ifdef INET 894 if (fwd_tag != NULL) { 895 struct sockaddr_in *next_hop; 896 897 next_hop = (struct sockaddr_in *)(fwd_tag+1); 898 /* 899 * Transparently forwarded. Pretend to be the destination. 900 * already got one like this? 901 */ 902 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 903 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); 904 if (!inp) { 905 /* 906 * It's new. Try to find the ambushing socket. 907 * Because we've rewritten the destination address, 908 * any hardware-generated hash is ignored. 909 */ 910 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 911 th->th_sport, next_hop->sin_addr, 912 next_hop->sin_port ? ntohs(next_hop->sin_port) : 913 th->th_dport, INPLOOKUP_WILDCARD | lookupflag, 914 m->m_pkthdr.rcvif); 915 } 916 } else 917 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 918 th->th_sport, ip->ip_dst, th->th_dport, 919 INPLOOKUP_WILDCARD | lookupflag, m->m_pkthdr.rcvif, m); 920 #endif /* INET */ 921 922 /* 923 * If the INPCB does not exist then all data in the incoming 924 * segment is discarded and an appropriate RST is sent back. 925 * XXX MRT Send RST using which routing table? 926 */ 927 if (inp == NULL) { 928 /* 929 * Log communication attempts to ports that are not 930 * in use. 931 */ 932 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 933 V_tcp_log_in_vain == 2) { 934 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 935 log(LOG_INFO, "%s; %s: Connection attempt " 936 "to closed port\n", s, __func__); 937 } 938 /* 939 * When blackholing do not respond with a RST but 940 * completely ignore the segment and drop it. 941 */ 942 if (((V_blackhole == 1 && (thflags & TH_SYN)) || 943 V_blackhole == 2) && (V_blackhole_local || 944 #ifdef INET6 945 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 946 #endif 947 #ifdef INET 948 !in_localip(ip->ip_src) 949 #else 950 true 951 #endif 952 )) 953 goto dropunlock; 954 955 rstreason = BANDLIM_RST_CLOSEDPORT; 956 goto dropwithreset; 957 } 958 INP_LOCK_ASSERT(inp); 959 /* 960 * While waiting for inp lock during the lookup, another thread 961 * can have dropped the inpcb, in which case we need to loop back 962 * and try to find a new inpcb to deliver to. 963 */ 964 if (inp->inp_flags & INP_DROPPED) { 965 INP_UNLOCK(inp); 966 inp = NULL; 967 goto findpcb; 968 } 969 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 970 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 971 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) { 972 inp->inp_flowid = m->m_pkthdr.flowid; 973 inp->inp_flowtype = M_HASHTYPE_GET(m); 974 } 975 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 976 #ifdef INET6 977 if (isipv6 && IPSEC_ENABLED(ipv6) && 978 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 979 goto dropunlock; 980 } 981 #ifdef INET 982 else 983 #endif 984 #endif /* INET6 */ 985 #ifdef INET 986 if (IPSEC_ENABLED(ipv4) && 987 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 988 goto dropunlock; 989 } 990 #endif /* INET */ 991 #endif /* IPSEC */ 992 993 /* 994 * Check the minimum TTL for socket. 995 */ 996 if (inp->inp_ip_minttl != 0) { 997 #ifdef INET6 998 if (isipv6) { 999 if (inp->inp_ip_minttl > ip6->ip6_hlim) 1000 goto dropunlock; 1001 } else 1002 #endif 1003 if (inp->inp_ip_minttl > ip->ip_ttl) 1004 goto dropunlock; 1005 } 1006 1007 /* 1008 * A previous connection in TIMEWAIT state is supposed to catch stray 1009 * or duplicate segments arriving late. If this segment was a 1010 * legitimate new connection attempt, the old INPCB gets removed and 1011 * we can try again to find a listening socket. 1012 */ 1013 if (inp->inp_flags & INP_TIMEWAIT) { 1014 tcp_dooptions(&to, optp, optlen, 1015 (thflags & TH_SYN) ? TO_SYN : 0); 1016 /* 1017 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1018 */ 1019 if (tcp_twcheck(inp, &to, th, m, tlen)) 1020 goto findpcb; 1021 return (IPPROTO_DONE); 1022 } 1023 /* 1024 * The TCPCB may no longer exist if the connection is winding 1025 * down or it is in the CLOSED state. Either way we drop the 1026 * segment and send an appropriate response. 1027 */ 1028 tp = intotcpcb(inp); 1029 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1030 rstreason = BANDLIM_RST_CLOSEDPORT; 1031 goto dropwithreset; 1032 } 1033 1034 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1035 rstreason = BANDLIM_RST_CLOSEDPORT; 1036 goto dropwithreset; 1037 } 1038 1039 #ifdef TCP_OFFLOAD 1040 if (tp->t_flags & TF_TOE) { 1041 tcp_offload_input(tp, m); 1042 m = NULL; /* consumed by the TOE driver */ 1043 goto dropunlock; 1044 } 1045 #endif 1046 1047 #ifdef MAC 1048 if (mac_inpcb_check_deliver(inp, m)) 1049 goto dropunlock; 1050 #endif 1051 so = inp->inp_socket; 1052 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1053 #ifdef TCPDEBUG 1054 if (so->so_options & SO_DEBUG) { 1055 ostate = tp->t_state; 1056 #ifdef INET6 1057 if (isipv6) { 1058 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1059 } else 1060 #endif 1061 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1062 tcp_savetcp = *th; 1063 } 1064 #endif /* TCPDEBUG */ 1065 /* 1066 * When the socket is accepting connections (the INPCB is in LISTEN 1067 * state) we look into the SYN cache if this is a new connection 1068 * attempt or the completion of a previous one. 1069 */ 1070 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1071 ("%s: so accepting but tp %p not listening", __func__, tp)); 1072 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1073 struct in_conninfo inc; 1074 1075 bzero(&inc, sizeof(inc)); 1076 #ifdef INET6 1077 if (isipv6) { 1078 inc.inc_flags |= INC_ISIPV6; 1079 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1080 inc.inc_flags |= INC_IPV6MINMTU; 1081 inc.inc6_faddr = ip6->ip6_src; 1082 inc.inc6_laddr = ip6->ip6_dst; 1083 } else 1084 #endif 1085 { 1086 inc.inc_faddr = ip->ip_src; 1087 inc.inc_laddr = ip->ip_dst; 1088 } 1089 inc.inc_fport = th->th_sport; 1090 inc.inc_lport = th->th_dport; 1091 inc.inc_fibnum = so->so_fibnum; 1092 1093 /* 1094 * Check for an existing connection attempt in syncache if 1095 * the flag is only ACK. A successful lookup creates a new 1096 * socket appended to the listen queue in SYN_RECEIVED state. 1097 */ 1098 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1099 /* 1100 * Parse the TCP options here because 1101 * syncookies need access to the reflected 1102 * timestamp. 1103 */ 1104 tcp_dooptions(&to, optp, optlen, 0); 1105 /* 1106 * NB: syncache_expand() doesn't unlock 1107 * inp and tcpinfo locks. 1108 */ 1109 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1110 if (rstreason < 0) { 1111 /* 1112 * A failing TCP MD5 signature comparison 1113 * must result in the segment being dropped 1114 * and must not produce any response back 1115 * to the sender. 1116 */ 1117 goto dropunlock; 1118 } else if (rstreason == 0) { 1119 /* 1120 * No syncache entry or ACK was not 1121 * for our SYN/ACK. Send a RST. 1122 * NB: syncache did its own logging 1123 * of the failure cause. 1124 */ 1125 rstreason = BANDLIM_RST_OPENPORT; 1126 goto dropwithreset; 1127 } 1128 tfo_socket_result: 1129 if (so == NULL) { 1130 /* 1131 * We completed the 3-way handshake 1132 * but could not allocate a socket 1133 * either due to memory shortage, 1134 * listen queue length limits or 1135 * global socket limits. Send RST 1136 * or wait and have the remote end 1137 * retransmit the ACK for another 1138 * try. 1139 */ 1140 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1141 log(LOG_DEBUG, "%s; %s: Listen socket: " 1142 "Socket allocation failed due to " 1143 "limits or memory shortage, %s\n", 1144 s, __func__, 1145 V_tcp_sc_rst_sock_fail ? 1146 "sending RST" : "try again"); 1147 if (V_tcp_sc_rst_sock_fail) { 1148 rstreason = BANDLIM_UNLIMITED; 1149 goto dropwithreset; 1150 } else 1151 goto dropunlock; 1152 } 1153 /* 1154 * Socket is created in state SYN_RECEIVED. 1155 * Unlock the listen socket, lock the newly 1156 * created socket and update the tp variable. 1157 * If we came here via jump to tfo_socket_result, 1158 * then listening socket is read-locked. 1159 */ 1160 INP_UNLOCK(inp); /* listen socket */ 1161 inp = sotoinpcb(so); 1162 /* 1163 * New connection inpcb is already locked by 1164 * syncache_expand(). 1165 */ 1166 INP_WLOCK_ASSERT(inp); 1167 tp = intotcpcb(inp); 1168 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1169 ("%s: ", __func__)); 1170 /* 1171 * Process the segment and the data it 1172 * contains. tcp_do_segment() consumes 1173 * the mbuf chain and unlocks the inpcb. 1174 */ 1175 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1176 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1177 iptos); 1178 return (IPPROTO_DONE); 1179 } 1180 /* 1181 * Segment flag validation for new connection attempts: 1182 * 1183 * Our (SYN|ACK) response was rejected. 1184 * Check with syncache and remove entry to prevent 1185 * retransmits. 1186 * 1187 * NB: syncache_chkrst does its own logging of failure 1188 * causes. 1189 */ 1190 if (thflags & TH_RST) { 1191 syncache_chkrst(&inc, th, m, port); 1192 goto dropunlock; 1193 } 1194 /* 1195 * We can't do anything without SYN. 1196 */ 1197 if ((thflags & TH_SYN) == 0) { 1198 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1199 log(LOG_DEBUG, "%s; %s: Listen socket: " 1200 "SYN is missing, segment ignored\n", 1201 s, __func__); 1202 TCPSTAT_INC(tcps_badsyn); 1203 goto dropunlock; 1204 } 1205 /* 1206 * (SYN|ACK) is bogus on a listen socket. 1207 */ 1208 if (thflags & TH_ACK) { 1209 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1210 log(LOG_DEBUG, "%s; %s: Listen socket: " 1211 "SYN|ACK invalid, segment rejected\n", 1212 s, __func__); 1213 syncache_badack(&inc, port); /* XXX: Not needed! */ 1214 TCPSTAT_INC(tcps_badsyn); 1215 rstreason = BANDLIM_RST_OPENPORT; 1216 goto dropwithreset; 1217 } 1218 /* 1219 * If the drop_synfin option is enabled, drop all 1220 * segments with both the SYN and FIN bits set. 1221 * This prevents e.g. nmap from identifying the 1222 * TCP/IP stack. 1223 * XXX: Poor reasoning. nmap has other methods 1224 * and is constantly refining its stack detection 1225 * strategies. 1226 * XXX: This is a violation of the TCP specification 1227 * and was used by RFC1644. 1228 */ 1229 if ((thflags & TH_FIN) && V_drop_synfin) { 1230 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1231 log(LOG_DEBUG, "%s; %s: Listen socket: " 1232 "SYN|FIN segment ignored (based on " 1233 "sysctl setting)\n", s, __func__); 1234 TCPSTAT_INC(tcps_badsyn); 1235 goto dropunlock; 1236 } 1237 /* 1238 * Segment's flags are (SYN) or (SYN|FIN). 1239 * 1240 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1241 * as they do not affect the state of the TCP FSM. 1242 * The data pointed to by TH_URG and th_urp is ignored. 1243 */ 1244 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1245 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1246 KASSERT(thflags & (TH_SYN), 1247 ("%s: Listen socket: TH_SYN not set", __func__)); 1248 INP_RLOCK_ASSERT(inp); 1249 #ifdef INET6 1250 /* 1251 * If deprecated address is forbidden, 1252 * we do not accept SYN to deprecated interface 1253 * address to prevent any new inbound connection from 1254 * getting established. 1255 * When we do not accept SYN, we send a TCP RST, 1256 * with deprecated source address (instead of dropping 1257 * it). We compromise it as it is much better for peer 1258 * to send a RST, and RST will be the final packet 1259 * for the exchange. 1260 * 1261 * If we do not forbid deprecated addresses, we accept 1262 * the SYN packet. RFC2462 does not suggest dropping 1263 * SYN in this case. 1264 * If we decipher RFC2462 5.5.4, it says like this: 1265 * 1. use of deprecated addr with existing 1266 * communication is okay - "SHOULD continue to be 1267 * used" 1268 * 2. use of it with new communication: 1269 * (2a) "SHOULD NOT be used if alternate address 1270 * with sufficient scope is available" 1271 * (2b) nothing mentioned otherwise. 1272 * Here we fall into (2b) case as we have no choice in 1273 * our source address selection - we must obey the peer. 1274 * 1275 * The wording in RFC2462 is confusing, and there are 1276 * multiple description text for deprecated address 1277 * handling - worse, they are not exactly the same. 1278 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1279 */ 1280 if (isipv6 && !V_ip6_use_deprecated) { 1281 struct in6_ifaddr *ia6; 1282 1283 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1284 if (ia6 != NULL && 1285 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1286 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1287 log(LOG_DEBUG, "%s; %s: Listen socket: " 1288 "Connection attempt to deprecated " 1289 "IPv6 address rejected\n", 1290 s, __func__); 1291 rstreason = BANDLIM_RST_OPENPORT; 1292 goto dropwithreset; 1293 } 1294 } 1295 #endif /* INET6 */ 1296 /* 1297 * Basic sanity checks on incoming SYN requests: 1298 * Don't respond if the destination is a link layer 1299 * broadcast according to RFC1122 4.2.3.10, p. 104. 1300 * If it is from this socket it must be forged. 1301 * Don't respond if the source or destination is a 1302 * global or subnet broad- or multicast address. 1303 * Note that it is quite possible to receive unicast 1304 * link-layer packets with a broadcast IP address. Use 1305 * in_broadcast() to find them. 1306 */ 1307 if (m->m_flags & (M_BCAST|M_MCAST)) { 1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1309 log(LOG_DEBUG, "%s; %s: Listen socket: " 1310 "Connection attempt from broad- or multicast " 1311 "link layer address ignored\n", s, __func__); 1312 goto dropunlock; 1313 } 1314 #ifdef INET6 1315 if (isipv6) { 1316 if (th->th_dport == th->th_sport && 1317 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1318 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1319 log(LOG_DEBUG, "%s; %s: Listen socket: " 1320 "Connection attempt to/from self " 1321 "ignored\n", s, __func__); 1322 goto dropunlock; 1323 } 1324 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1325 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1327 log(LOG_DEBUG, "%s; %s: Listen socket: " 1328 "Connection attempt from/to multicast " 1329 "address ignored\n", s, __func__); 1330 goto dropunlock; 1331 } 1332 } 1333 #endif 1334 #if defined(INET) && defined(INET6) 1335 else 1336 #endif 1337 #ifdef INET 1338 { 1339 if (th->th_dport == th->th_sport && 1340 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1341 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1342 log(LOG_DEBUG, "%s; %s: Listen socket: " 1343 "Connection attempt from/to self " 1344 "ignored\n", s, __func__); 1345 goto dropunlock; 1346 } 1347 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1348 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1349 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1350 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1351 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1352 log(LOG_DEBUG, "%s; %s: Listen socket: " 1353 "Connection attempt from/to broad- " 1354 "or multicast address ignored\n", 1355 s, __func__); 1356 goto dropunlock; 1357 } 1358 } 1359 #endif 1360 /* 1361 * SYN appears to be valid. Create compressed TCP state 1362 * for syncache. 1363 */ 1364 #ifdef TCPDEBUG 1365 if (so->so_options & SO_DEBUG) 1366 tcp_trace(TA_INPUT, ostate, tp, 1367 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1368 #endif 1369 TCP_PROBE3(debug__input, tp, th, m); 1370 tcp_dooptions(&to, optp, optlen, TO_SYN); 1371 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1372 iptos, port)) != NULL) 1373 goto tfo_socket_result; 1374 1375 /* 1376 * Entry added to syncache and mbuf consumed. 1377 * Only the listen socket is unlocked by syncache_add(). 1378 */ 1379 return (IPPROTO_DONE); 1380 } else if (tp->t_state == TCPS_LISTEN) { 1381 /* 1382 * When a listen socket is torn down the SO_ACCEPTCONN 1383 * flag is removed first while connections are drained 1384 * from the accept queue in a unlock/lock cycle of the 1385 * ACCEPT_LOCK, opening a race condition allowing a SYN 1386 * attempt go through unhandled. 1387 */ 1388 goto dropunlock; 1389 } 1390 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1391 if (tp->t_flags & TF_SIGNATURE) { 1392 tcp_dooptions(&to, optp, optlen, thflags); 1393 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1394 TCPSTAT_INC(tcps_sig_err_nosigopt); 1395 goto dropunlock; 1396 } 1397 if (!TCPMD5_ENABLED() || 1398 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1399 goto dropunlock; 1400 } 1401 #endif 1402 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1403 1404 /* 1405 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1406 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1407 * the inpcb, and unlocks pcbinfo. 1408 * 1409 * XXXGL: in case of a pure SYN arriving on existing connection 1410 * TCP stacks won't need to modify the PCB, they would either drop 1411 * the segment silently, or send a challenge ACK. However, we try 1412 * to upgrade the lock, because calling convention for stacks is 1413 * write-lock on PCB. If upgrade fails, drop the SYN. 1414 */ 1415 if (lookupflag == INPLOOKUP_RLOCKPCB && INP_TRY_UPGRADE(inp) == 0) 1416 goto dropunlock; 1417 1418 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1419 return (IPPROTO_DONE); 1420 1421 dropwithreset: 1422 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1423 1424 if (inp != NULL) { 1425 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1426 INP_UNLOCK(inp); 1427 } else 1428 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1429 m = NULL; /* mbuf chain got consumed. */ 1430 goto drop; 1431 1432 dropunlock: 1433 if (m != NULL) 1434 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1435 1436 if (inp != NULL) 1437 INP_UNLOCK(inp); 1438 1439 drop: 1440 if (s != NULL) 1441 free(s, M_TCPLOG); 1442 if (m != NULL) 1443 m_freem(m); 1444 return (IPPROTO_DONE); 1445 } 1446 1447 /* 1448 * Automatic sizing of receive socket buffer. Often the send 1449 * buffer size is not optimally adjusted to the actual network 1450 * conditions at hand (delay bandwidth product). Setting the 1451 * buffer size too small limits throughput on links with high 1452 * bandwidth and high delay (eg. trans-continental/oceanic links). 1453 * 1454 * On the receive side the socket buffer memory is only rarely 1455 * used to any significant extent. This allows us to be much 1456 * more aggressive in scaling the receive socket buffer. For 1457 * the case that the buffer space is actually used to a large 1458 * extent and we run out of kernel memory we can simply drop 1459 * the new segments; TCP on the sender will just retransmit it 1460 * later. Setting the buffer size too big may only consume too 1461 * much kernel memory if the application doesn't read() from 1462 * the socket or packet loss or reordering makes use of the 1463 * reassembly queue. 1464 * 1465 * The criteria to step up the receive buffer one notch are: 1466 * 1. Application has not set receive buffer size with 1467 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1468 * 2. the number of bytes received during 1/2 of an sRTT 1469 * is at least 3/8 of the current socket buffer size. 1470 * 3. receive buffer size has not hit maximal automatic size; 1471 * 1472 * If all of the criteria are met we increaset the socket buffer 1473 * by a 1/2 (bounded by the max). This allows us to keep ahead 1474 * of slow-start but also makes it so our peer never gets limited 1475 * by our rwnd which we then open up causing a burst. 1476 * 1477 * This algorithm does two steps per RTT at most and only if 1478 * we receive a bulk stream w/o packet losses or reorderings. 1479 * Shrinking the buffer during idle times is not necessary as 1480 * it doesn't consume any memory when idle. 1481 * 1482 * TODO: Only step up if the application is actually serving 1483 * the buffer to better manage the socket buffer resources. 1484 */ 1485 int 1486 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1487 struct tcpcb *tp, int tlen) 1488 { 1489 int newsize = 0; 1490 1491 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1492 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1493 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1494 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1495 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1496 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1497 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1498 } 1499 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1500 1501 /* Start over with next RTT. */ 1502 tp->rfbuf_ts = 0; 1503 tp->rfbuf_cnt = 0; 1504 } else { 1505 tp->rfbuf_cnt += tlen; /* add up */ 1506 } 1507 return (newsize); 1508 } 1509 1510 int 1511 tcp_input(struct mbuf **mp, int *offp, int proto) 1512 { 1513 return(tcp_input_with_port(mp, offp, proto, 0)); 1514 } 1515 1516 void 1517 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1518 { 1519 /* 1520 * Since tp might be gone if the session entered 1521 * the TIME_WAIT state before coming here, we need 1522 * to check if the socket is still connected. 1523 */ 1524 if (tp == NULL) { 1525 return; 1526 } 1527 if (so == NULL) { 1528 return; 1529 } 1530 INP_LOCK_ASSERT(tp->t_inpcb); 1531 if (tp->t_flags & TF_WAKESOR) { 1532 tp->t_flags &= ~TF_WAKESOR; 1533 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1534 sorwakeup_locked(so); 1535 } 1536 } 1537 1538 void 1539 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1540 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1541 { 1542 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1543 int rstreason, todrop, win, incforsyn = 0; 1544 uint32_t tiwin; 1545 uint16_t nsegs; 1546 char *s; 1547 struct in_conninfo *inc; 1548 struct mbuf *mfree; 1549 struct tcpopt to; 1550 int tfo_syn; 1551 u_int maxseg; 1552 1553 #ifdef TCPDEBUG 1554 /* 1555 * The size of tcp_saveipgen must be the size of the max ip header, 1556 * now IPv6. 1557 */ 1558 u_char tcp_saveipgen[IP6_HDR_LEN]; 1559 struct tcphdr tcp_savetcp; 1560 short ostate = 0; 1561 #endif 1562 thflags = th->th_flags; 1563 inc = &tp->t_inpcb->inp_inc; 1564 tp->sackhint.last_sack_ack = 0; 1565 sack_changed = 0; 1566 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1567 1568 NET_EPOCH_ASSERT(); 1569 INP_WLOCK_ASSERT(tp->t_inpcb); 1570 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1571 __func__)); 1572 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1573 __func__)); 1574 1575 #ifdef TCPPCAP 1576 /* Save segment, if requested. */ 1577 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1578 #endif 1579 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1580 tlen, NULL, true); 1581 1582 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1583 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1584 log(LOG_DEBUG, "%s; %s: " 1585 "SYN|FIN segment ignored (based on " 1586 "sysctl setting)\n", s, __func__); 1587 free(s, M_TCPLOG); 1588 } 1589 goto drop; 1590 } 1591 1592 /* 1593 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1594 * check SEQ.ACK first. 1595 */ 1596 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1597 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1598 rstreason = BANDLIM_UNLIMITED; 1599 goto dropwithreset; 1600 } 1601 1602 /* 1603 * Segment received on connection. 1604 * Reset idle time and keep-alive timer. 1605 * XXX: This should be done after segment 1606 * validation to ignore broken/spoofed segs. 1607 */ 1608 tp->t_rcvtime = ticks; 1609 1610 /* 1611 * Scale up the window into a 32-bit value. 1612 * For the SYN_SENT state the scale is zero. 1613 */ 1614 tiwin = th->th_win << tp->snd_scale; 1615 #ifdef STATS 1616 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1617 #endif 1618 1619 /* 1620 * TCP ECN processing. 1621 */ 1622 if (tp->t_flags2 & TF2_ECN_PERMIT) { 1623 if (thflags & TH_CWR) { 1624 tp->t_flags2 &= ~TF2_ECN_SND_ECE; 1625 tp->t_flags |= TF_ACKNOW; 1626 } 1627 switch (iptos & IPTOS_ECN_MASK) { 1628 case IPTOS_ECN_CE: 1629 tp->t_flags2 |= TF2_ECN_SND_ECE; 1630 TCPSTAT_INC(tcps_ecn_ce); 1631 break; 1632 case IPTOS_ECN_ECT0: 1633 TCPSTAT_INC(tcps_ecn_ect0); 1634 break; 1635 case IPTOS_ECN_ECT1: 1636 TCPSTAT_INC(tcps_ecn_ect1); 1637 break; 1638 } 1639 1640 /* Process a packet differently from RFC3168. */ 1641 cc_ecnpkt_handler(tp, th, iptos); 1642 1643 /* Congestion experienced. */ 1644 if (thflags & TH_ECE) { 1645 cc_cong_signal(tp, th, CC_ECN); 1646 } 1647 } 1648 1649 /* 1650 * Parse options on any incoming segment. 1651 */ 1652 tcp_dooptions(&to, (u_char *)(th + 1), 1653 (th->th_off << 2) - sizeof(struct tcphdr), 1654 (thflags & TH_SYN) ? TO_SYN : 0); 1655 1656 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1657 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1658 (to.to_flags & TOF_SIGNATURE) == 0) { 1659 TCPSTAT_INC(tcps_sig_err_sigopt); 1660 /* XXX: should drop? */ 1661 } 1662 #endif 1663 /* 1664 * If echoed timestamp is later than the current time, 1665 * fall back to non RFC1323 RTT calculation. Normalize 1666 * timestamp if syncookies were used when this connection 1667 * was established. 1668 */ 1669 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1670 to.to_tsecr -= tp->ts_offset; 1671 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1672 to.to_tsecr = 0; 1673 else if (tp->t_flags & TF_PREVVALID && 1674 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1675 cc_cong_signal(tp, th, CC_RTO_ERR); 1676 } 1677 /* 1678 * Process options only when we get SYN/ACK back. The SYN case 1679 * for incoming connections is handled in tcp_syncache. 1680 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1681 * or <SYN,ACK>) segment itself is never scaled. 1682 * XXX this is traditional behavior, may need to be cleaned up. 1683 */ 1684 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1685 /* Handle parallel SYN for ECN */ 1686 if (!(thflags & TH_ACK) && 1687 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) && 1688 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) { 1689 tp->t_flags2 |= TF2_ECN_PERMIT; 1690 tp->t_flags2 |= TF2_ECN_SND_ECE; 1691 TCPSTAT_INC(tcps_ecn_shs); 1692 } 1693 if ((to.to_flags & TOF_SCALE) && 1694 (tp->t_flags & TF_REQ_SCALE) && 1695 !(tp->t_flags & TF_NOOPT)) { 1696 tp->t_flags |= TF_RCVD_SCALE; 1697 tp->snd_scale = to.to_wscale; 1698 } else 1699 tp->t_flags &= ~TF_REQ_SCALE; 1700 /* 1701 * Initial send window. It will be updated with 1702 * the next incoming segment to the scaled value. 1703 */ 1704 tp->snd_wnd = th->th_win; 1705 if ((to.to_flags & TOF_TS) && 1706 (tp->t_flags & TF_REQ_TSTMP) && 1707 !(tp->t_flags & TF_NOOPT)) { 1708 tp->t_flags |= TF_RCVD_TSTMP; 1709 tp->ts_recent = to.to_tsval; 1710 tp->ts_recent_age = tcp_ts_getticks(); 1711 } else 1712 tp->t_flags &= ~TF_REQ_TSTMP; 1713 if (to.to_flags & TOF_MSS) 1714 tcp_mss(tp, to.to_mss); 1715 if ((tp->t_flags & TF_SACK_PERMIT) && 1716 (!(to.to_flags & TOF_SACKPERM) || 1717 (tp->t_flags & TF_NOOPT))) 1718 tp->t_flags &= ~TF_SACK_PERMIT; 1719 if (IS_FASTOPEN(tp->t_flags)) { 1720 if ((to.to_flags & TOF_FASTOPEN) && 1721 !(tp->t_flags & TF_NOOPT)) { 1722 uint16_t mss; 1723 1724 if (to.to_flags & TOF_MSS) 1725 mss = to.to_mss; 1726 else 1727 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1728 mss = TCP6_MSS; 1729 else 1730 mss = TCP_MSS; 1731 tcp_fastopen_update_cache(tp, mss, 1732 to.to_tfo_len, to.to_tfo_cookie); 1733 } else 1734 tcp_fastopen_disable_path(tp); 1735 } 1736 } 1737 1738 /* 1739 * If timestamps were negotiated during SYN/ACK and a 1740 * segment without a timestamp is received, silently drop 1741 * the segment, unless it is a RST segment or missing timestamps are 1742 * tolerated. 1743 * See section 3.2 of RFC 7323. 1744 */ 1745 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1746 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1747 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1748 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1749 "segment processed normally\n", 1750 s, __func__); 1751 free(s, M_TCPLOG); 1752 } 1753 } else { 1754 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1755 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1756 "segment silently dropped\n", s, __func__); 1757 free(s, M_TCPLOG); 1758 } 1759 goto drop; 1760 } 1761 } 1762 /* 1763 * If timestamps were not negotiated during SYN/ACK and a 1764 * segment with a timestamp is received, ignore the 1765 * timestamp and process the packet normally. 1766 * See section 3.2 of RFC 7323. 1767 */ 1768 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1769 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1770 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1771 "segment processed normally\n", s, __func__); 1772 free(s, M_TCPLOG); 1773 } 1774 } 1775 1776 /* 1777 * Header prediction: check for the two common cases 1778 * of a uni-directional data xfer. If the packet has 1779 * no control flags, is in-sequence, the window didn't 1780 * change and we're not retransmitting, it's a 1781 * candidate. If the length is zero and the ack moved 1782 * forward, we're the sender side of the xfer. Just 1783 * free the data acked & wake any higher level process 1784 * that was blocked waiting for space. If the length 1785 * is non-zero and the ack didn't move, we're the 1786 * receiver side. If we're getting packets in-order 1787 * (the reassembly queue is empty), add the data to 1788 * the socket buffer and note that we need a delayed ack. 1789 * Make sure that the hidden state-flags are also off. 1790 * Since we check for TCPS_ESTABLISHED first, it can only 1791 * be TH_NEEDSYN. 1792 */ 1793 if (tp->t_state == TCPS_ESTABLISHED && 1794 th->th_seq == tp->rcv_nxt && 1795 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1796 tp->snd_nxt == tp->snd_max && 1797 tiwin && tiwin == tp->snd_wnd && 1798 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1799 SEGQ_EMPTY(tp) && 1800 ((to.to_flags & TOF_TS) == 0 || 1801 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1802 /* 1803 * If last ACK falls within this segment's sequence numbers, 1804 * record the timestamp. 1805 * NOTE that the test is modified according to the latest 1806 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1807 */ 1808 if ((to.to_flags & TOF_TS) != 0 && 1809 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1810 tp->ts_recent_age = tcp_ts_getticks(); 1811 tp->ts_recent = to.to_tsval; 1812 } 1813 1814 if (tlen == 0) { 1815 if (SEQ_GT(th->th_ack, tp->snd_una) && 1816 SEQ_LEQ(th->th_ack, tp->snd_max) && 1817 !IN_RECOVERY(tp->t_flags) && 1818 (to.to_flags & TOF_SACK) == 0 && 1819 TAILQ_EMPTY(&tp->snd_holes)) { 1820 /* 1821 * This is a pure ack for outstanding data. 1822 */ 1823 TCPSTAT_INC(tcps_predack); 1824 1825 /* 1826 * "bad retransmit" recovery without timestamps. 1827 */ 1828 if ((to.to_flags & TOF_TS) == 0 && 1829 tp->t_rxtshift == 1 && 1830 tp->t_flags & TF_PREVVALID && 1831 (int)(ticks - tp->t_badrxtwin) < 0) { 1832 cc_cong_signal(tp, th, CC_RTO_ERR); 1833 } 1834 1835 /* 1836 * Recalculate the transmit timer / rtt. 1837 * 1838 * Some boxes send broken timestamp replies 1839 * during the SYN+ACK phase, ignore 1840 * timestamps of 0 or we could calculate a 1841 * huge RTT and blow up the retransmit timer. 1842 */ 1843 if ((to.to_flags & TOF_TS) != 0 && 1844 to.to_tsecr) { 1845 uint32_t t; 1846 1847 t = tcp_ts_getticks() - to.to_tsecr; 1848 if (!tp->t_rttlow || tp->t_rttlow > t) 1849 tp->t_rttlow = t; 1850 tcp_xmit_timer(tp, 1851 TCP_TS_TO_TICKS(t) + 1); 1852 } else if (tp->t_rtttime && 1853 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1854 if (!tp->t_rttlow || 1855 tp->t_rttlow > ticks - tp->t_rtttime) 1856 tp->t_rttlow = ticks - tp->t_rtttime; 1857 tcp_xmit_timer(tp, 1858 ticks - tp->t_rtttime); 1859 } 1860 acked = BYTES_THIS_ACK(tp, th); 1861 1862 #ifdef TCP_HHOOK 1863 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1864 hhook_run_tcp_est_in(tp, th, &to); 1865 #endif 1866 1867 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1868 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1869 sbdrop(&so->so_snd, acked); 1870 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1871 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1872 tp->snd_recover = th->th_ack - 1; 1873 1874 /* 1875 * Let the congestion control algorithm update 1876 * congestion control related information. This 1877 * typically means increasing the congestion 1878 * window. 1879 */ 1880 cc_ack_received(tp, th, nsegs, CC_ACK); 1881 1882 tp->snd_una = th->th_ack; 1883 /* 1884 * Pull snd_wl2 up to prevent seq wrap relative 1885 * to th_ack. 1886 */ 1887 tp->snd_wl2 = th->th_ack; 1888 tp->t_dupacks = 0; 1889 m_freem(m); 1890 1891 /* 1892 * If all outstanding data are acked, stop 1893 * retransmit timer, otherwise restart timer 1894 * using current (possibly backed-off) value. 1895 * If process is waiting for space, 1896 * wakeup/selwakeup/signal. If data 1897 * are ready to send, let tcp_output 1898 * decide between more output or persist. 1899 */ 1900 #ifdef TCPDEBUG 1901 if (so->so_options & SO_DEBUG) 1902 tcp_trace(TA_INPUT, ostate, tp, 1903 (void *)tcp_saveipgen, 1904 &tcp_savetcp, 0); 1905 #endif 1906 TCP_PROBE3(debug__input, tp, th, m); 1907 if (tp->snd_una == tp->snd_max) 1908 tcp_timer_activate(tp, TT_REXMT, 0); 1909 else if (!tcp_timer_active(tp, TT_PERSIST)) 1910 tcp_timer_activate(tp, TT_REXMT, 1911 tp->t_rxtcur); 1912 sowwakeup(so); 1913 if (sbavail(&so->so_snd)) 1914 (void) tp->t_fb->tfb_tcp_output(tp); 1915 goto check_delack; 1916 } 1917 } else if (th->th_ack == tp->snd_una && 1918 tlen <= sbspace(&so->so_rcv)) { 1919 int newsize = 0; /* automatic sockbuf scaling */ 1920 1921 /* 1922 * This is a pure, in-sequence data packet with 1923 * nothing on the reassembly queue and we have enough 1924 * buffer space to take it. 1925 */ 1926 /* Clean receiver SACK report if present */ 1927 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1928 tcp_clean_sackreport(tp); 1929 TCPSTAT_INC(tcps_preddat); 1930 tp->rcv_nxt += tlen; 1931 if (tlen && 1932 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1933 (tp->t_fbyte_in == 0)) { 1934 tp->t_fbyte_in = ticks; 1935 if (tp->t_fbyte_in == 0) 1936 tp->t_fbyte_in = 1; 1937 if (tp->t_fbyte_out && tp->t_fbyte_in) 1938 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1939 } 1940 /* 1941 * Pull snd_wl1 up to prevent seq wrap relative to 1942 * th_seq. 1943 */ 1944 tp->snd_wl1 = th->th_seq; 1945 /* 1946 * Pull rcv_up up to prevent seq wrap relative to 1947 * rcv_nxt. 1948 */ 1949 tp->rcv_up = tp->rcv_nxt; 1950 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1951 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1952 #ifdef TCPDEBUG 1953 if (so->so_options & SO_DEBUG) 1954 tcp_trace(TA_INPUT, ostate, tp, 1955 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1956 #endif 1957 TCP_PROBE3(debug__input, tp, th, m); 1958 1959 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1960 1961 /* Add data to socket buffer. */ 1962 SOCKBUF_LOCK(&so->so_rcv); 1963 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1964 m_freem(m); 1965 } else { 1966 /* 1967 * Set new socket buffer size. 1968 * Give up when limit is reached. 1969 */ 1970 if (newsize) 1971 if (!sbreserve_locked(&so->so_rcv, 1972 newsize, so, NULL)) 1973 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1974 m_adj(m, drop_hdrlen); /* delayed header drop */ 1975 sbappendstream_locked(&so->so_rcv, m, 0); 1976 } 1977 /* NB: sorwakeup_locked() does an implicit unlock. */ 1978 sorwakeup_locked(so); 1979 if (DELAY_ACK(tp, tlen)) { 1980 tp->t_flags |= TF_DELACK; 1981 } else { 1982 tp->t_flags |= TF_ACKNOW; 1983 tp->t_fb->tfb_tcp_output(tp); 1984 } 1985 goto check_delack; 1986 } 1987 } 1988 1989 /* 1990 * Calculate amount of space in receive window, 1991 * and then do TCP input processing. 1992 * Receive window is amount of space in rcv queue, 1993 * but not less than advertised window. 1994 */ 1995 win = sbspace(&so->so_rcv); 1996 if (win < 0) 1997 win = 0; 1998 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1999 2000 switch (tp->t_state) { 2001 /* 2002 * If the state is SYN_RECEIVED: 2003 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 2004 */ 2005 case TCPS_SYN_RECEIVED: 2006 if ((thflags & TH_ACK) && 2007 (SEQ_LEQ(th->th_ack, tp->snd_una) || 2008 SEQ_GT(th->th_ack, tp->snd_max))) { 2009 rstreason = BANDLIM_RST_OPENPORT; 2010 goto dropwithreset; 2011 } 2012 if (IS_FASTOPEN(tp->t_flags)) { 2013 /* 2014 * When a TFO connection is in SYN_RECEIVED, the 2015 * only valid packets are the initial SYN, a 2016 * retransmit/copy of the initial SYN (possibly with 2017 * a subset of the original data), a valid ACK, a 2018 * FIN, or a RST. 2019 */ 2020 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 2021 rstreason = BANDLIM_RST_OPENPORT; 2022 goto dropwithreset; 2023 } else if (thflags & TH_SYN) { 2024 /* non-initial SYN is ignored */ 2025 if ((tcp_timer_active(tp, TT_DELACK) || 2026 tcp_timer_active(tp, TT_REXMT))) 2027 goto drop; 2028 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 2029 goto drop; 2030 } 2031 } 2032 break; 2033 2034 /* 2035 * If the state is SYN_SENT: 2036 * if seg contains a RST with valid ACK (SEQ.ACK has already 2037 * been verified), then drop the connection. 2038 * if seg contains a RST without an ACK, drop the seg. 2039 * if seg does not contain SYN, then drop the seg. 2040 * Otherwise this is an acceptable SYN segment 2041 * initialize tp->rcv_nxt and tp->irs 2042 * if seg contains ack then advance tp->snd_una 2043 * if seg contains an ECE and ECN support is enabled, the stream 2044 * is ECN capable. 2045 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2046 * arrange for segment to be acked (eventually) 2047 * continue processing rest of data/controls, beginning with URG 2048 */ 2049 case TCPS_SYN_SENT: 2050 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2051 TCP_PROBE5(connect__refused, NULL, tp, 2052 m, tp, th); 2053 tp = tcp_drop(tp, ECONNREFUSED); 2054 } 2055 if (thflags & TH_RST) 2056 goto drop; 2057 if (!(thflags & TH_SYN)) 2058 goto drop; 2059 2060 tp->irs = th->th_seq; 2061 tcp_rcvseqinit(tp); 2062 if (thflags & TH_ACK) { 2063 int tfo_partial_ack = 0; 2064 2065 TCPSTAT_INC(tcps_connects); 2066 soisconnected(so); 2067 #ifdef MAC 2068 mac_socketpeer_set_from_mbuf(m, so); 2069 #endif 2070 /* Do window scaling on this connection? */ 2071 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2072 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2073 tp->rcv_scale = tp->request_r_scale; 2074 } 2075 tp->rcv_adv += min(tp->rcv_wnd, 2076 TCP_MAXWIN << tp->rcv_scale); 2077 tp->snd_una++; /* SYN is acked */ 2078 /* 2079 * If not all the data that was sent in the TFO SYN 2080 * has been acked, resend the remainder right away. 2081 */ 2082 if (IS_FASTOPEN(tp->t_flags) && 2083 (tp->snd_una != tp->snd_max)) { 2084 tp->snd_nxt = th->th_ack; 2085 tfo_partial_ack = 1; 2086 } 2087 /* 2088 * If there's data, delay ACK; if there's also a FIN 2089 * ACKNOW will be turned on later. 2090 */ 2091 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2092 tcp_timer_activate(tp, TT_DELACK, 2093 tcp_delacktime); 2094 else 2095 tp->t_flags |= TF_ACKNOW; 2096 2097 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2098 (V_tcp_do_ecn == 1)) { 2099 tp->t_flags2 |= TF2_ECN_PERMIT; 2100 TCPSTAT_INC(tcps_ecn_shs); 2101 } 2102 2103 /* 2104 * Received <SYN,ACK> in SYN_SENT[*] state. 2105 * Transitions: 2106 * SYN_SENT --> ESTABLISHED 2107 * SYN_SENT* --> FIN_WAIT_1 2108 */ 2109 tp->t_starttime = ticks; 2110 if (tp->t_flags & TF_NEEDFIN) { 2111 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2112 tp->t_flags &= ~TF_NEEDFIN; 2113 thflags &= ~TH_SYN; 2114 } else { 2115 tcp_state_change(tp, TCPS_ESTABLISHED); 2116 TCP_PROBE5(connect__established, NULL, tp, 2117 m, tp, th); 2118 cc_conn_init(tp); 2119 tcp_timer_activate(tp, TT_KEEP, 2120 TP_KEEPIDLE(tp)); 2121 } 2122 } else { 2123 /* 2124 * Received initial SYN in SYN-SENT[*] state => 2125 * simultaneous open. 2126 * If it succeeds, connection is * half-synchronized. 2127 * Otherwise, do 3-way handshake: 2128 * SYN-SENT -> SYN-RECEIVED 2129 * SYN-SENT* -> SYN-RECEIVED* 2130 */ 2131 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2132 tcp_timer_activate(tp, TT_REXMT, 0); 2133 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2134 } 2135 2136 INP_WLOCK_ASSERT(tp->t_inpcb); 2137 2138 /* 2139 * Advance th->th_seq to correspond to first data byte. 2140 * If data, trim to stay within window, 2141 * dropping FIN if necessary. 2142 */ 2143 th->th_seq++; 2144 if (tlen > tp->rcv_wnd) { 2145 todrop = tlen - tp->rcv_wnd; 2146 m_adj(m, -todrop); 2147 tlen = tp->rcv_wnd; 2148 thflags &= ~TH_FIN; 2149 TCPSTAT_INC(tcps_rcvpackafterwin); 2150 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2151 } 2152 tp->snd_wl1 = th->th_seq - 1; 2153 tp->rcv_up = th->th_seq; 2154 /* 2155 * Client side of transaction: already sent SYN and data. 2156 * If the remote host used T/TCP to validate the SYN, 2157 * our data will be ACK'd; if so, enter normal data segment 2158 * processing in the middle of step 5, ack processing. 2159 * Otherwise, goto step 6. 2160 */ 2161 if (thflags & TH_ACK) 2162 goto process_ACK; 2163 2164 goto step6; 2165 2166 /* 2167 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2168 * do normal processing. 2169 * 2170 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2171 */ 2172 case TCPS_LAST_ACK: 2173 case TCPS_CLOSING: 2174 break; /* continue normal processing */ 2175 } 2176 2177 /* 2178 * States other than LISTEN or SYN_SENT. 2179 * First check the RST flag and sequence number since reset segments 2180 * are exempt from the timestamp and connection count tests. This 2181 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2182 * below which allowed reset segments in half the sequence space 2183 * to fall though and be processed (which gives forged reset 2184 * segments with a random sequence number a 50 percent chance of 2185 * killing a connection). 2186 * Then check timestamp, if present. 2187 * Then check the connection count, if present. 2188 * Then check that at least some bytes of segment are within 2189 * receive window. If segment begins before rcv_nxt, 2190 * drop leading data (and SYN); if nothing left, just ack. 2191 */ 2192 if (thflags & TH_RST) { 2193 /* 2194 * RFC5961 Section 3.2 2195 * 2196 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2197 * - If RST is in window, we send challenge ACK. 2198 * 2199 * Note: to take into account delayed ACKs, we should 2200 * test against last_ack_sent instead of rcv_nxt. 2201 * Note 2: we handle special case of closed window, not 2202 * covered by the RFC. 2203 */ 2204 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2205 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2206 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2207 KASSERT(tp->t_state != TCPS_SYN_SENT, 2208 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2209 __func__, th, tp)); 2210 2211 if (V_tcp_insecure_rst || 2212 tp->last_ack_sent == th->th_seq) { 2213 TCPSTAT_INC(tcps_drops); 2214 /* Drop the connection. */ 2215 switch (tp->t_state) { 2216 case TCPS_SYN_RECEIVED: 2217 so->so_error = ECONNREFUSED; 2218 goto close; 2219 case TCPS_ESTABLISHED: 2220 case TCPS_FIN_WAIT_1: 2221 case TCPS_FIN_WAIT_2: 2222 case TCPS_CLOSE_WAIT: 2223 case TCPS_CLOSING: 2224 case TCPS_LAST_ACK: 2225 so->so_error = ECONNRESET; 2226 close: 2227 /* FALLTHROUGH */ 2228 default: 2229 tp = tcp_close(tp); 2230 } 2231 } else { 2232 TCPSTAT_INC(tcps_badrst); 2233 /* Send challenge ACK. */ 2234 tcp_respond(tp, mtod(m, void *), th, m, 2235 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2236 tp->last_ack_sent = tp->rcv_nxt; 2237 m = NULL; 2238 } 2239 } 2240 goto drop; 2241 } 2242 2243 /* 2244 * RFC5961 Section 4.2 2245 * Send challenge ACK for any SYN in synchronized state. 2246 */ 2247 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2248 tp->t_state != TCPS_SYN_RECEIVED) { 2249 TCPSTAT_INC(tcps_badsyn); 2250 if (V_tcp_insecure_syn && 2251 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2252 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2253 tp = tcp_drop(tp, ECONNRESET); 2254 rstreason = BANDLIM_UNLIMITED; 2255 } else { 2256 /* Send challenge ACK. */ 2257 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2258 tp->snd_nxt, TH_ACK); 2259 tp->last_ack_sent = tp->rcv_nxt; 2260 m = NULL; 2261 } 2262 goto drop; 2263 } 2264 2265 /* 2266 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2267 * and it's less than ts_recent, drop it. 2268 */ 2269 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2270 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2271 /* Check to see if ts_recent is over 24 days old. */ 2272 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2273 /* 2274 * Invalidate ts_recent. If this segment updates 2275 * ts_recent, the age will be reset later and ts_recent 2276 * will get a valid value. If it does not, setting 2277 * ts_recent to zero will at least satisfy the 2278 * requirement that zero be placed in the timestamp 2279 * echo reply when ts_recent isn't valid. The 2280 * age isn't reset until we get a valid ts_recent 2281 * because we don't want out-of-order segments to be 2282 * dropped when ts_recent is old. 2283 */ 2284 tp->ts_recent = 0; 2285 } else { 2286 TCPSTAT_INC(tcps_rcvduppack); 2287 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2288 TCPSTAT_INC(tcps_pawsdrop); 2289 if (tlen) 2290 goto dropafterack; 2291 goto drop; 2292 } 2293 } 2294 2295 /* 2296 * In the SYN-RECEIVED state, validate that the packet belongs to 2297 * this connection before trimming the data to fit the receive 2298 * window. Check the sequence number versus IRS since we know 2299 * the sequence numbers haven't wrapped. This is a partial fix 2300 * for the "LAND" DoS attack. 2301 */ 2302 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2303 rstreason = BANDLIM_RST_OPENPORT; 2304 goto dropwithreset; 2305 } 2306 2307 todrop = tp->rcv_nxt - th->th_seq; 2308 if (todrop > 0) { 2309 if (thflags & TH_SYN) { 2310 thflags &= ~TH_SYN; 2311 th->th_seq++; 2312 if (th->th_urp > 1) 2313 th->th_urp--; 2314 else 2315 thflags &= ~TH_URG; 2316 todrop--; 2317 } 2318 /* 2319 * Following if statement from Stevens, vol. 2, p. 960. 2320 */ 2321 if (todrop > tlen 2322 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2323 /* 2324 * Any valid FIN must be to the left of the window. 2325 * At this point the FIN must be a duplicate or out 2326 * of sequence; drop it. 2327 */ 2328 thflags &= ~TH_FIN; 2329 2330 /* 2331 * Send an ACK to resynchronize and drop any data. 2332 * But keep on processing for RST or ACK. 2333 */ 2334 tp->t_flags |= TF_ACKNOW; 2335 todrop = tlen; 2336 TCPSTAT_INC(tcps_rcvduppack); 2337 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2338 } else { 2339 TCPSTAT_INC(tcps_rcvpartduppack); 2340 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2341 } 2342 /* 2343 * DSACK - add SACK block for dropped range 2344 */ 2345 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2346 tcp_update_sack_list(tp, th->th_seq, 2347 th->th_seq + todrop); 2348 /* 2349 * ACK now, as the next in-sequence segment 2350 * will clear the DSACK block again 2351 */ 2352 tp->t_flags |= TF_ACKNOW; 2353 } 2354 drop_hdrlen += todrop; /* drop from the top afterwards */ 2355 th->th_seq += todrop; 2356 tlen -= todrop; 2357 if (th->th_urp > todrop) 2358 th->th_urp -= todrop; 2359 else { 2360 thflags &= ~TH_URG; 2361 th->th_urp = 0; 2362 } 2363 } 2364 2365 /* 2366 * If new data are received on a connection after the 2367 * user processes are gone, then RST the other end. 2368 */ 2369 if ((so->so_state & SS_NOFDREF) && 2370 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2371 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2372 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2373 "after socket was closed, " 2374 "sending RST and removing tcpcb\n", 2375 s, __func__, tcpstates[tp->t_state], tlen); 2376 free(s, M_TCPLOG); 2377 } 2378 tp = tcp_close(tp); 2379 TCPSTAT_INC(tcps_rcvafterclose); 2380 rstreason = BANDLIM_UNLIMITED; 2381 goto dropwithreset; 2382 } 2383 2384 /* 2385 * If segment ends after window, drop trailing data 2386 * (and PUSH and FIN); if nothing left, just ACK. 2387 */ 2388 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2389 if (todrop > 0) { 2390 TCPSTAT_INC(tcps_rcvpackafterwin); 2391 if (todrop >= tlen) { 2392 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2393 /* 2394 * If window is closed can only take segments at 2395 * window edge, and have to drop data and PUSH from 2396 * incoming segments. Continue processing, but 2397 * remember to ack. Otherwise, drop segment 2398 * and ack. 2399 */ 2400 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2401 tp->t_flags |= TF_ACKNOW; 2402 TCPSTAT_INC(tcps_rcvwinprobe); 2403 } else 2404 goto dropafterack; 2405 } else 2406 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2407 m_adj(m, -todrop); 2408 tlen -= todrop; 2409 thflags &= ~(TH_PUSH|TH_FIN); 2410 } 2411 2412 /* 2413 * If last ACK falls within this segment's sequence numbers, 2414 * record its timestamp. 2415 * NOTE: 2416 * 1) That the test incorporates suggestions from the latest 2417 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2418 * 2) That updating only on newer timestamps interferes with 2419 * our earlier PAWS tests, so this check should be solely 2420 * predicated on the sequence space of this segment. 2421 * 3) That we modify the segment boundary check to be 2422 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2423 * instead of RFC1323's 2424 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2425 * This modified check allows us to overcome RFC1323's 2426 * limitations as described in Stevens TCP/IP Illustrated 2427 * Vol. 2 p.869. In such cases, we can still calculate the 2428 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2429 */ 2430 if ((to.to_flags & TOF_TS) != 0 && 2431 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2432 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2433 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2434 tp->ts_recent_age = tcp_ts_getticks(); 2435 tp->ts_recent = to.to_tsval; 2436 } 2437 2438 /* 2439 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2440 * flag is on (half-synchronized state), then queue data for 2441 * later processing; else drop segment and return. 2442 */ 2443 if ((thflags & TH_ACK) == 0) { 2444 if (tp->t_state == TCPS_SYN_RECEIVED || 2445 (tp->t_flags & TF_NEEDSYN)) { 2446 if (tp->t_state == TCPS_SYN_RECEIVED && 2447 IS_FASTOPEN(tp->t_flags)) { 2448 tp->snd_wnd = tiwin; 2449 cc_conn_init(tp); 2450 } 2451 goto step6; 2452 } else if (tp->t_flags & TF_ACKNOW) 2453 goto dropafterack; 2454 else 2455 goto drop; 2456 } 2457 2458 /* 2459 * Ack processing. 2460 */ 2461 switch (tp->t_state) { 2462 /* 2463 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2464 * ESTABLISHED state and continue processing. 2465 * The ACK was checked above. 2466 */ 2467 case TCPS_SYN_RECEIVED: 2468 2469 TCPSTAT_INC(tcps_connects); 2470 soisconnected(so); 2471 /* Do window scaling? */ 2472 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2473 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2474 tp->rcv_scale = tp->request_r_scale; 2475 } 2476 tp->snd_wnd = tiwin; 2477 /* 2478 * Make transitions: 2479 * SYN-RECEIVED -> ESTABLISHED 2480 * SYN-RECEIVED* -> FIN-WAIT-1 2481 */ 2482 tp->t_starttime = ticks; 2483 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2484 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2485 tp->t_tfo_pending = NULL; 2486 } 2487 if (tp->t_flags & TF_NEEDFIN) { 2488 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2489 tp->t_flags &= ~TF_NEEDFIN; 2490 } else { 2491 tcp_state_change(tp, TCPS_ESTABLISHED); 2492 TCP_PROBE5(accept__established, NULL, tp, 2493 m, tp, th); 2494 /* 2495 * TFO connections call cc_conn_init() during SYN 2496 * processing. Calling it again here for such 2497 * connections is not harmless as it would undo the 2498 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2499 * is retransmitted. 2500 */ 2501 if (!IS_FASTOPEN(tp->t_flags)) 2502 cc_conn_init(tp); 2503 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2504 } 2505 /* 2506 * Account for the ACK of our SYN prior to 2507 * regular ACK processing below, except for 2508 * simultaneous SYN, which is handled later. 2509 */ 2510 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2511 incforsyn = 1; 2512 /* 2513 * If segment contains data or ACK, will call tcp_reass() 2514 * later; if not, do so now to pass queued data to user. 2515 */ 2516 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2517 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2518 (struct mbuf *)0); 2519 tcp_handle_wakeup(tp, so); 2520 } 2521 tp->snd_wl1 = th->th_seq - 1; 2522 /* FALLTHROUGH */ 2523 2524 /* 2525 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2526 * ACKs. If the ack is in the range 2527 * tp->snd_una < th->th_ack <= tp->snd_max 2528 * then advance tp->snd_una to th->th_ack and drop 2529 * data from the retransmission queue. If this ACK reflects 2530 * more up to date window information we update our window information. 2531 */ 2532 case TCPS_ESTABLISHED: 2533 case TCPS_FIN_WAIT_1: 2534 case TCPS_FIN_WAIT_2: 2535 case TCPS_CLOSE_WAIT: 2536 case TCPS_CLOSING: 2537 case TCPS_LAST_ACK: 2538 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2539 TCPSTAT_INC(tcps_rcvacktoomuch); 2540 goto dropafterack; 2541 } 2542 if ((tp->t_flags & TF_SACK_PERMIT) && 2543 ((to.to_flags & TOF_SACK) || 2544 !TAILQ_EMPTY(&tp->snd_holes))) { 2545 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && 2546 (tp->t_flags & TF_LRD)) { 2547 tcp_sack_lost_retransmission(tp, th); 2548 } 2549 } else 2550 /* 2551 * Reset the value so that previous (valid) value 2552 * from the last ack with SACK doesn't get used. 2553 */ 2554 tp->sackhint.sacked_bytes = 0; 2555 2556 #ifdef TCP_HHOOK 2557 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2558 hhook_run_tcp_est_in(tp, th, &to); 2559 #endif 2560 2561 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2562 maxseg = tcp_maxseg(tp); 2563 if (tlen == 0 && 2564 (tiwin == tp->snd_wnd || 2565 (tp->t_flags & TF_SACK_PERMIT))) { 2566 /* 2567 * If this is the first time we've seen a 2568 * FIN from the remote, this is not a 2569 * duplicate and it needs to be processed 2570 * normally. This happens during a 2571 * simultaneous close. 2572 */ 2573 if ((thflags & TH_FIN) && 2574 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2575 tp->t_dupacks = 0; 2576 break; 2577 } 2578 TCPSTAT_INC(tcps_rcvdupack); 2579 /* 2580 * If we have outstanding data (other than 2581 * a window probe), this is a completely 2582 * duplicate ack (ie, window info didn't 2583 * change and FIN isn't set), 2584 * the ack is the biggest we've 2585 * seen and we've seen exactly our rexmt 2586 * threshold of them, assume a packet 2587 * has been dropped and retransmit it. 2588 * Kludge snd_nxt & the congestion 2589 * window so we send only this one 2590 * packet. 2591 * 2592 * We know we're losing at the current 2593 * window size so do congestion avoidance 2594 * (set ssthresh to half the current window 2595 * and pull our congestion window back to 2596 * the new ssthresh). 2597 * 2598 * Dup acks mean that packets have left the 2599 * network (they're now cached at the receiver) 2600 * so bump cwnd by the amount in the receiver 2601 * to keep a constant cwnd packets in the 2602 * network. 2603 * 2604 * When using TCP ECN, notify the peer that 2605 * we reduced the cwnd. 2606 */ 2607 /* 2608 * Following 2 kinds of acks should not affect 2609 * dupack counting: 2610 * 1) Old acks 2611 * 2) Acks with SACK but without any new SACK 2612 * information in them. These could result from 2613 * any anomaly in the network like a switch 2614 * duplicating packets or a possible DoS attack. 2615 */ 2616 if (th->th_ack != tp->snd_una || 2617 ((tp->t_flags & TF_SACK_PERMIT) && 2618 (to.to_flags & TOF_SACK) && 2619 !sack_changed)) 2620 break; 2621 else if (!tcp_timer_active(tp, TT_REXMT)) 2622 tp->t_dupacks = 0; 2623 else if (++tp->t_dupacks > tcprexmtthresh || 2624 IN_FASTRECOVERY(tp->t_flags)) { 2625 cc_ack_received(tp, th, nsegs, 2626 CC_DUPACK); 2627 if (V_tcp_do_prr && 2628 IN_FASTRECOVERY(tp->t_flags)) { 2629 tcp_do_prr_ack(tp, th, &to); 2630 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2631 (to.to_flags & TOF_SACK) && 2632 IN_FASTRECOVERY(tp->t_flags)) { 2633 int awnd; 2634 2635 /* 2636 * Compute the amount of data in flight first. 2637 * We can inject new data into the pipe iff 2638 * we have less than 1/2 the original window's 2639 * worth of data in flight. 2640 */ 2641 if (V_tcp_do_newsack) 2642 awnd = tcp_compute_pipe(tp); 2643 else 2644 awnd = (tp->snd_nxt - tp->snd_fack) + 2645 tp->sackhint.sack_bytes_rexmit; 2646 2647 if (awnd < tp->snd_ssthresh) { 2648 tp->snd_cwnd += maxseg; 2649 if (tp->snd_cwnd > tp->snd_ssthresh) 2650 tp->snd_cwnd = tp->snd_ssthresh; 2651 } 2652 } else 2653 tp->snd_cwnd += maxseg; 2654 (void) tp->t_fb->tfb_tcp_output(tp); 2655 goto drop; 2656 } else if (tp->t_dupacks == tcprexmtthresh || 2657 (tp->t_flags & TF_SACK_PERMIT && 2658 V_tcp_do_newsack && 2659 tp->sackhint.sacked_bytes > 2660 (tcprexmtthresh - 1) * maxseg)) { 2661 enter_recovery: 2662 /* 2663 * Above is the RFC6675 trigger condition of 2664 * more than (dupthresh-1)*maxseg sacked data. 2665 * If the count of holes in the 2666 * scoreboard is >= dupthresh, we could 2667 * also enter loss recovery, but don't 2668 * have that value readily available. 2669 */ 2670 tp->t_dupacks = tcprexmtthresh; 2671 tcp_seq onxt = tp->snd_nxt; 2672 2673 /* 2674 * If we're doing sack, or prr, check 2675 * to see if we're already in sack 2676 * recovery. If we're not doing sack, 2677 * check to see if we're in newreno 2678 * recovery. 2679 */ 2680 if (V_tcp_do_prr || 2681 (tp->t_flags & TF_SACK_PERMIT)) { 2682 if (IN_FASTRECOVERY(tp->t_flags)) { 2683 tp->t_dupacks = 0; 2684 break; 2685 } 2686 } else { 2687 if (SEQ_LEQ(th->th_ack, 2688 tp->snd_recover)) { 2689 tp->t_dupacks = 0; 2690 break; 2691 } 2692 } 2693 /* Congestion signal before ack. */ 2694 cc_cong_signal(tp, th, CC_NDUPACK); 2695 cc_ack_received(tp, th, nsegs, 2696 CC_DUPACK); 2697 tcp_timer_activate(tp, TT_REXMT, 0); 2698 tp->t_rtttime = 0; 2699 if (V_tcp_do_prr) { 2700 /* 2701 * snd_ssthresh is already updated by 2702 * cc_cong_signal. 2703 */ 2704 if ((tp->t_flags & TF_SACK_PERMIT) && 2705 (to.to_flags & TOF_SACK)) { 2706 tp->sackhint.prr_delivered = 2707 tp->sackhint.sacked_bytes; 2708 } else { 2709 tp->sackhint.prr_delivered = 2710 imin(tp->snd_max - tp->snd_una, 2711 imin(INT_MAX / 65536, 2712 tp->t_dupacks) * maxseg); 2713 } 2714 tp->sackhint.recover_fs = max(1, 2715 tp->snd_nxt - tp->snd_una); 2716 } 2717 if ((tp->t_flags & TF_SACK_PERMIT) && 2718 (to.to_flags & TOF_SACK)) { 2719 TCPSTAT_INC( 2720 tcps_sack_recovery_episode); 2721 tp->snd_recover = tp->snd_nxt; 2722 tp->snd_cwnd = maxseg; 2723 (void) tp->t_fb->tfb_tcp_output(tp); 2724 if (SEQ_GT(th->th_ack, tp->snd_una)) 2725 goto resume_partialack; 2726 goto drop; 2727 } 2728 tp->snd_nxt = th->th_ack; 2729 tp->snd_cwnd = maxseg; 2730 (void) tp->t_fb->tfb_tcp_output(tp); 2731 KASSERT(tp->snd_limited <= 2, 2732 ("%s: tp->snd_limited too big", 2733 __func__)); 2734 tp->snd_cwnd = tp->snd_ssthresh + 2735 maxseg * 2736 (tp->t_dupacks - tp->snd_limited); 2737 if (SEQ_GT(onxt, tp->snd_nxt)) 2738 tp->snd_nxt = onxt; 2739 goto drop; 2740 } else if (V_tcp_do_rfc3042) { 2741 /* 2742 * Process first and second duplicate 2743 * ACKs. Each indicates a segment 2744 * leaving the network, creating room 2745 * for more. Make sure we can send a 2746 * packet on reception of each duplicate 2747 * ACK by increasing snd_cwnd by one 2748 * segment. Restore the original 2749 * snd_cwnd after packet transmission. 2750 */ 2751 cc_ack_received(tp, th, nsegs, 2752 CC_DUPACK); 2753 uint32_t oldcwnd = tp->snd_cwnd; 2754 tcp_seq oldsndmax = tp->snd_max; 2755 u_int sent; 2756 int avail; 2757 2758 KASSERT(tp->t_dupacks == 1 || 2759 tp->t_dupacks == 2, 2760 ("%s: dupacks not 1 or 2", 2761 __func__)); 2762 if (tp->t_dupacks == 1) 2763 tp->snd_limited = 0; 2764 tp->snd_cwnd = 2765 (tp->snd_nxt - tp->snd_una) + 2766 (tp->t_dupacks - tp->snd_limited) * 2767 maxseg; 2768 /* 2769 * Only call tcp_output when there 2770 * is new data available to be sent. 2771 * Otherwise we would send pure ACKs. 2772 */ 2773 SOCKBUF_LOCK(&so->so_snd); 2774 avail = sbavail(&so->so_snd) - 2775 (tp->snd_nxt - tp->snd_una); 2776 SOCKBUF_UNLOCK(&so->so_snd); 2777 if (avail > 0) 2778 (void) tp->t_fb->tfb_tcp_output(tp); 2779 sent = tp->snd_max - oldsndmax; 2780 if (sent > maxseg) { 2781 KASSERT((tp->t_dupacks == 2 && 2782 tp->snd_limited == 0) || 2783 (sent == maxseg + 1 && 2784 tp->t_flags & TF_SENTFIN), 2785 ("%s: sent too much", 2786 __func__)); 2787 tp->snd_limited = 2; 2788 } else if (sent > 0) 2789 ++tp->snd_limited; 2790 tp->snd_cwnd = oldcwnd; 2791 goto drop; 2792 } 2793 } 2794 break; 2795 } else { 2796 /* 2797 * This ack is advancing the left edge, reset the 2798 * counter. 2799 */ 2800 tp->t_dupacks = 0; 2801 /* 2802 * If this ack also has new SACK info, increment the 2803 * counter as per rfc6675. The variable 2804 * sack_changed tracks all changes to the SACK 2805 * scoreboard, including when partial ACKs without 2806 * SACK options are received, and clear the scoreboard 2807 * from the left side. Such partial ACKs should not be 2808 * counted as dupacks here. 2809 */ 2810 if ((tp->t_flags & TF_SACK_PERMIT) && 2811 (to.to_flags & TOF_SACK) && 2812 sack_changed) { 2813 tp->t_dupacks++; 2814 /* limit overhead by setting maxseg last */ 2815 if (!IN_FASTRECOVERY(tp->t_flags) && 2816 (tp->sackhint.sacked_bytes > 2817 ((tcprexmtthresh - 1) * 2818 (maxseg = tcp_maxseg(tp))))) { 2819 goto enter_recovery; 2820 } 2821 } 2822 } 2823 2824 resume_partialack: 2825 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2826 ("%s: th_ack <= snd_una", __func__)); 2827 2828 /* 2829 * If the congestion window was inflated to account 2830 * for the other side's cached packets, retract it. 2831 */ 2832 if (IN_FASTRECOVERY(tp->t_flags)) { 2833 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2834 if (tp->t_flags & TF_SACK_PERMIT) 2835 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2836 tcp_timer_activate(tp, TT_REXMT, 0); 2837 tp->t_rtttime = 0; 2838 tcp_do_prr_ack(tp, th, &to); 2839 tp->t_flags |= TF_ACKNOW; 2840 (void) tcp_output(tp); 2841 } else 2842 tcp_sack_partialack(tp, th); 2843 else 2844 tcp_newreno_partial_ack(tp, th); 2845 } else 2846 cc_post_recovery(tp, th); 2847 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2848 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2849 if (V_tcp_do_prr) { 2850 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2851 tp->snd_fack = th->th_ack; 2852 tcp_do_prr_ack(tp, th, &to); 2853 (void) tcp_output(tp); 2854 } 2855 } else 2856 cc_post_recovery(tp, th); 2857 } 2858 /* 2859 * If we reach this point, ACK is not a duplicate, 2860 * i.e., it ACKs something we sent. 2861 */ 2862 if (tp->t_flags & TF_NEEDSYN) { 2863 /* 2864 * T/TCP: Connection was half-synchronized, and our 2865 * SYN has been ACK'd (so connection is now fully 2866 * synchronized). Go to non-starred state, 2867 * increment snd_una for ACK of SYN, and check if 2868 * we can do window scaling. 2869 */ 2870 tp->t_flags &= ~TF_NEEDSYN; 2871 tp->snd_una++; 2872 /* Do window scaling? */ 2873 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2874 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2875 tp->rcv_scale = tp->request_r_scale; 2876 /* Send window already scaled. */ 2877 } 2878 } 2879 2880 process_ACK: 2881 INP_WLOCK_ASSERT(tp->t_inpcb); 2882 2883 /* 2884 * Adjust for the SYN bit in sequence space, 2885 * but don't account for it in cwnd calculations. 2886 * This is for the SYN_RECEIVED, non-simultaneous 2887 * SYN case. SYN_SENT and simultaneous SYN are 2888 * treated elsewhere. 2889 */ 2890 if (incforsyn) 2891 tp->snd_una++; 2892 acked = BYTES_THIS_ACK(tp, th); 2893 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2894 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2895 tp->snd_una, th->th_ack, tp, m)); 2896 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2897 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2898 2899 /* 2900 * If we just performed our first retransmit, and the ACK 2901 * arrives within our recovery window, then it was a mistake 2902 * to do the retransmit in the first place. Recover our 2903 * original cwnd and ssthresh, and proceed to transmit where 2904 * we left off. 2905 */ 2906 if (tp->t_rxtshift == 1 && 2907 tp->t_flags & TF_PREVVALID && 2908 tp->t_badrxtwin && 2909 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2910 cc_cong_signal(tp, th, CC_RTO_ERR); 2911 2912 /* 2913 * If we have a timestamp reply, update smoothed 2914 * round trip time. If no timestamp is present but 2915 * transmit timer is running and timed sequence 2916 * number was acked, update smoothed round trip time. 2917 * Since we now have an rtt measurement, cancel the 2918 * timer backoff (cf., Phil Karn's retransmit alg.). 2919 * Recompute the initial retransmit timer. 2920 * 2921 * Some boxes send broken timestamp replies 2922 * during the SYN+ACK phase, ignore 2923 * timestamps of 0 or we could calculate a 2924 * huge RTT and blow up the retransmit timer. 2925 */ 2926 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2927 uint32_t t; 2928 2929 t = tcp_ts_getticks() - to.to_tsecr; 2930 if (!tp->t_rttlow || tp->t_rttlow > t) 2931 tp->t_rttlow = t; 2932 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2933 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2934 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2935 tp->t_rttlow = ticks - tp->t_rtttime; 2936 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2937 } 2938 2939 /* 2940 * If all outstanding data is acked, stop retransmit 2941 * timer and remember to restart (more output or persist). 2942 * If there is more data to be acked, restart retransmit 2943 * timer, using current (possibly backed-off) value. 2944 */ 2945 if (th->th_ack == tp->snd_max) { 2946 tcp_timer_activate(tp, TT_REXMT, 0); 2947 needoutput = 1; 2948 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2949 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2950 2951 /* 2952 * If no data (only SYN) was ACK'd, 2953 * skip rest of ACK processing. 2954 */ 2955 if (acked == 0) 2956 goto step6; 2957 2958 /* 2959 * Let the congestion control algorithm update congestion 2960 * control related information. This typically means increasing 2961 * the congestion window. 2962 */ 2963 cc_ack_received(tp, th, nsegs, CC_ACK); 2964 2965 SOCKBUF_LOCK(&so->so_snd); 2966 if (acked > sbavail(&so->so_snd)) { 2967 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2968 tp->snd_wnd -= sbavail(&so->so_snd); 2969 else 2970 tp->snd_wnd = 0; 2971 mfree = sbcut_locked(&so->so_snd, 2972 (int)sbavail(&so->so_snd)); 2973 ourfinisacked = 1; 2974 } else { 2975 mfree = sbcut_locked(&so->so_snd, acked); 2976 if (tp->snd_wnd >= (uint32_t) acked) 2977 tp->snd_wnd -= acked; 2978 else 2979 tp->snd_wnd = 0; 2980 ourfinisacked = 0; 2981 } 2982 /* NB: sowwakeup_locked() does an implicit unlock. */ 2983 sowwakeup_locked(so); 2984 m_freem(mfree); 2985 /* Detect una wraparound. */ 2986 if (!IN_RECOVERY(tp->t_flags) && 2987 SEQ_GT(tp->snd_una, tp->snd_recover) && 2988 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2989 tp->snd_recover = th->th_ack - 1; 2990 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2991 if (IN_RECOVERY(tp->t_flags) && 2992 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2993 EXIT_RECOVERY(tp->t_flags); 2994 } 2995 tp->snd_una = th->th_ack; 2996 if (tp->t_flags & TF_SACK_PERMIT) { 2997 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2998 tp->snd_recover = tp->snd_una; 2999 } 3000 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 3001 tp->snd_nxt = tp->snd_una; 3002 3003 switch (tp->t_state) { 3004 /* 3005 * In FIN_WAIT_1 STATE in addition to the processing 3006 * for the ESTABLISHED state if our FIN is now acknowledged 3007 * then enter FIN_WAIT_2. 3008 */ 3009 case TCPS_FIN_WAIT_1: 3010 if (ourfinisacked) { 3011 /* 3012 * If we can't receive any more 3013 * data, then closing user can proceed. 3014 * Starting the timer is contrary to the 3015 * specification, but if we don't get a FIN 3016 * we'll hang forever. 3017 * 3018 * XXXjl: 3019 * we should release the tp also, and use a 3020 * compressed state. 3021 */ 3022 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3023 soisdisconnected(so); 3024 tcp_timer_activate(tp, TT_2MSL, 3025 (tcp_fast_finwait2_recycle ? 3026 tcp_finwait2_timeout : 3027 TP_MAXIDLE(tp))); 3028 } 3029 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3030 } 3031 break; 3032 3033 /* 3034 * In CLOSING STATE in addition to the processing for 3035 * the ESTABLISHED state if the ACK acknowledges our FIN 3036 * then enter the TIME-WAIT state, otherwise ignore 3037 * the segment. 3038 */ 3039 case TCPS_CLOSING: 3040 if (ourfinisacked) { 3041 tcp_twstart(tp); 3042 m_freem(m); 3043 return; 3044 } 3045 break; 3046 3047 /* 3048 * In LAST_ACK, we may still be waiting for data to drain 3049 * and/or to be acked, as well as for the ack of our FIN. 3050 * If our FIN is now acknowledged, delete the TCB, 3051 * enter the closed state and return. 3052 */ 3053 case TCPS_LAST_ACK: 3054 if (ourfinisacked) { 3055 tp = tcp_close(tp); 3056 goto drop; 3057 } 3058 break; 3059 } 3060 } 3061 3062 step6: 3063 INP_WLOCK_ASSERT(tp->t_inpcb); 3064 3065 /* 3066 * Update window information. 3067 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3068 */ 3069 if ((thflags & TH_ACK) && 3070 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3071 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3072 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3073 /* keep track of pure window updates */ 3074 if (tlen == 0 && 3075 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3076 TCPSTAT_INC(tcps_rcvwinupd); 3077 tp->snd_wnd = tiwin; 3078 tp->snd_wl1 = th->th_seq; 3079 tp->snd_wl2 = th->th_ack; 3080 if (tp->snd_wnd > tp->max_sndwnd) 3081 tp->max_sndwnd = tp->snd_wnd; 3082 needoutput = 1; 3083 } 3084 3085 /* 3086 * Process segments with URG. 3087 */ 3088 if ((thflags & TH_URG) && th->th_urp && 3089 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3090 /* 3091 * This is a kludge, but if we receive and accept 3092 * random urgent pointers, we'll crash in 3093 * soreceive. It's hard to imagine someone 3094 * actually wanting to send this much urgent data. 3095 */ 3096 SOCKBUF_LOCK(&so->so_rcv); 3097 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3098 th->th_urp = 0; /* XXX */ 3099 thflags &= ~TH_URG; /* XXX */ 3100 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3101 goto dodata; /* XXX */ 3102 } 3103 /* 3104 * If this segment advances the known urgent pointer, 3105 * then mark the data stream. This should not happen 3106 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3107 * a FIN has been received from the remote side. 3108 * In these states we ignore the URG. 3109 * 3110 * According to RFC961 (Assigned Protocols), 3111 * the urgent pointer points to the last octet 3112 * of urgent data. We continue, however, 3113 * to consider it to indicate the first octet 3114 * of data past the urgent section as the original 3115 * spec states (in one of two places). 3116 */ 3117 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3118 tp->rcv_up = th->th_seq + th->th_urp; 3119 so->so_oobmark = sbavail(&so->so_rcv) + 3120 (tp->rcv_up - tp->rcv_nxt) - 1; 3121 if (so->so_oobmark == 0) 3122 so->so_rcv.sb_state |= SBS_RCVATMARK; 3123 sohasoutofband(so); 3124 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3125 } 3126 SOCKBUF_UNLOCK(&so->so_rcv); 3127 /* 3128 * Remove out of band data so doesn't get presented to user. 3129 * This can happen independent of advancing the URG pointer, 3130 * but if two URG's are pending at once, some out-of-band 3131 * data may creep in... ick. 3132 */ 3133 if (th->th_urp <= (uint32_t)tlen && 3134 !(so->so_options & SO_OOBINLINE)) { 3135 /* hdr drop is delayed */ 3136 tcp_pulloutofband(so, th, m, drop_hdrlen); 3137 } 3138 } else { 3139 /* 3140 * If no out of band data is expected, 3141 * pull receive urgent pointer along 3142 * with the receive window. 3143 */ 3144 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3145 tp->rcv_up = tp->rcv_nxt; 3146 } 3147 dodata: /* XXX */ 3148 INP_WLOCK_ASSERT(tp->t_inpcb); 3149 3150 /* 3151 * Process the segment text, merging it into the TCP sequencing queue, 3152 * and arranging for acknowledgment of receipt if necessary. 3153 * This process logically involves adjusting tp->rcv_wnd as data 3154 * is presented to the user (this happens in tcp_usrreq.c, 3155 * case PRU_RCVD). If a FIN has already been received on this 3156 * connection then we just ignore the text. 3157 */ 3158 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3159 IS_FASTOPEN(tp->t_flags)); 3160 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3161 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3162 tcp_seq save_start = th->th_seq; 3163 tcp_seq save_rnxt = tp->rcv_nxt; 3164 int save_tlen = tlen; 3165 m_adj(m, drop_hdrlen); /* delayed header drop */ 3166 /* 3167 * Insert segment which includes th into TCP reassembly queue 3168 * with control block tp. Set thflags to whether reassembly now 3169 * includes a segment with FIN. This handles the common case 3170 * inline (segment is the next to be received on an established 3171 * connection, and the queue is empty), avoiding linkage into 3172 * and removal from the queue and repetition of various 3173 * conversions. 3174 * Set DELACK for segments received in order, but ack 3175 * immediately when segments are out of order (so 3176 * fast retransmit can work). 3177 */ 3178 if (th->th_seq == tp->rcv_nxt && 3179 SEGQ_EMPTY(tp) && 3180 (TCPS_HAVEESTABLISHED(tp->t_state) || 3181 tfo_syn)) { 3182 if (DELAY_ACK(tp, tlen) || tfo_syn) 3183 tp->t_flags |= TF_DELACK; 3184 else 3185 tp->t_flags |= TF_ACKNOW; 3186 tp->rcv_nxt += tlen; 3187 if (tlen && 3188 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3189 (tp->t_fbyte_in == 0)) { 3190 tp->t_fbyte_in = ticks; 3191 if (tp->t_fbyte_in == 0) 3192 tp->t_fbyte_in = 1; 3193 if (tp->t_fbyte_out && tp->t_fbyte_in) 3194 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3195 } 3196 thflags = th->th_flags & TH_FIN; 3197 TCPSTAT_INC(tcps_rcvpack); 3198 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3199 SOCKBUF_LOCK(&so->so_rcv); 3200 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3201 m_freem(m); 3202 else 3203 sbappendstream_locked(&so->so_rcv, m, 0); 3204 tp->t_flags |= TF_WAKESOR; 3205 } else { 3206 /* 3207 * XXX: Due to the header drop above "th" is 3208 * theoretically invalid by now. Fortunately 3209 * m_adj() doesn't actually frees any mbufs 3210 * when trimming from the head. 3211 */ 3212 tcp_seq temp = save_start; 3213 3214 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3215 tp->t_flags |= TF_ACKNOW; 3216 } 3217 if ((tp->t_flags & TF_SACK_PERMIT) && 3218 (save_tlen > 0) && 3219 TCPS_HAVEESTABLISHED(tp->t_state)) { 3220 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3221 /* 3222 * DSACK actually handled in the fastpath 3223 * above. 3224 */ 3225 tcp_update_sack_list(tp, save_start, 3226 save_start + save_tlen); 3227 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3228 if ((tp->rcv_numsacks >= 1) && 3229 (tp->sackblks[0].end == save_start)) { 3230 /* 3231 * Partial overlap, recorded at todrop 3232 * above. 3233 */ 3234 tcp_update_sack_list(tp, 3235 tp->sackblks[0].start, 3236 tp->sackblks[0].end); 3237 } else { 3238 tcp_update_dsack_list(tp, save_start, 3239 save_start + save_tlen); 3240 } 3241 } else if (tlen >= save_tlen) { 3242 /* Update of sackblks. */ 3243 tcp_update_dsack_list(tp, save_start, 3244 save_start + save_tlen); 3245 } else if (tlen > 0) { 3246 tcp_update_dsack_list(tp, save_start, 3247 save_start + tlen); 3248 } 3249 } 3250 tcp_handle_wakeup(tp, so); 3251 #if 0 3252 /* 3253 * Note the amount of data that peer has sent into 3254 * our window, in order to estimate the sender's 3255 * buffer size. 3256 * XXX: Unused. 3257 */ 3258 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3259 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3260 else 3261 len = so->so_rcv.sb_hiwat; 3262 #endif 3263 } else { 3264 m_freem(m); 3265 thflags &= ~TH_FIN; 3266 } 3267 3268 /* 3269 * If FIN is received ACK the FIN and let the user know 3270 * that the connection is closing. 3271 */ 3272 if (thflags & TH_FIN) { 3273 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3274 /* The socket upcall is handled by socantrcvmore. */ 3275 socantrcvmore(so); 3276 /* 3277 * If connection is half-synchronized 3278 * (ie NEEDSYN flag on) then delay ACK, 3279 * so it may be piggybacked when SYN is sent. 3280 * Otherwise, since we received a FIN then no 3281 * more input can be expected, send ACK now. 3282 */ 3283 if (tp->t_flags & TF_NEEDSYN) 3284 tp->t_flags |= TF_DELACK; 3285 else 3286 tp->t_flags |= TF_ACKNOW; 3287 tp->rcv_nxt++; 3288 } 3289 switch (tp->t_state) { 3290 /* 3291 * In SYN_RECEIVED and ESTABLISHED STATES 3292 * enter the CLOSE_WAIT state. 3293 */ 3294 case TCPS_SYN_RECEIVED: 3295 tp->t_starttime = ticks; 3296 /* FALLTHROUGH */ 3297 case TCPS_ESTABLISHED: 3298 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3299 break; 3300 3301 /* 3302 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3303 * enter the CLOSING state. 3304 */ 3305 case TCPS_FIN_WAIT_1: 3306 tcp_state_change(tp, TCPS_CLOSING); 3307 break; 3308 3309 /* 3310 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3311 * starting the time-wait timer, turning off the other 3312 * standard timers. 3313 */ 3314 case TCPS_FIN_WAIT_2: 3315 tcp_twstart(tp); 3316 return; 3317 } 3318 } 3319 #ifdef TCPDEBUG 3320 if (so->so_options & SO_DEBUG) 3321 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3322 &tcp_savetcp, 0); 3323 #endif 3324 TCP_PROBE3(debug__input, tp, th, m); 3325 3326 /* 3327 * Return any desired output. 3328 */ 3329 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3330 (void) tp->t_fb->tfb_tcp_output(tp); 3331 3332 check_delack: 3333 INP_WLOCK_ASSERT(tp->t_inpcb); 3334 3335 if (tp->t_flags & TF_DELACK) { 3336 tp->t_flags &= ~TF_DELACK; 3337 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3338 } 3339 INP_WUNLOCK(tp->t_inpcb); 3340 return; 3341 3342 dropafterack: 3343 /* 3344 * Generate an ACK dropping incoming segment if it occupies 3345 * sequence space, where the ACK reflects our state. 3346 * 3347 * We can now skip the test for the RST flag since all 3348 * paths to this code happen after packets containing 3349 * RST have been dropped. 3350 * 3351 * In the SYN-RECEIVED state, don't send an ACK unless the 3352 * segment we received passes the SYN-RECEIVED ACK test. 3353 * If it fails send a RST. This breaks the loop in the 3354 * "LAND" DoS attack, and also prevents an ACK storm 3355 * between two listening ports that have been sent forged 3356 * SYN segments, each with the source address of the other. 3357 */ 3358 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3359 (SEQ_GT(tp->snd_una, th->th_ack) || 3360 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3361 rstreason = BANDLIM_RST_OPENPORT; 3362 goto dropwithreset; 3363 } 3364 #ifdef TCPDEBUG 3365 if (so->so_options & SO_DEBUG) 3366 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3367 &tcp_savetcp, 0); 3368 #endif 3369 TCP_PROBE3(debug__input, tp, th, m); 3370 tp->t_flags |= TF_ACKNOW; 3371 (void) tp->t_fb->tfb_tcp_output(tp); 3372 INP_WUNLOCK(tp->t_inpcb); 3373 m_freem(m); 3374 return; 3375 3376 dropwithreset: 3377 if (tp != NULL) { 3378 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3379 INP_WUNLOCK(tp->t_inpcb); 3380 } else 3381 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3382 return; 3383 3384 drop: 3385 /* 3386 * Drop space held by incoming segment and return. 3387 */ 3388 #ifdef TCPDEBUG 3389 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3390 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3391 &tcp_savetcp, 0); 3392 #endif 3393 TCP_PROBE3(debug__input, tp, th, m); 3394 if (tp != NULL) { 3395 INP_WUNLOCK(tp->t_inpcb); 3396 } 3397 m_freem(m); 3398 } 3399 3400 /* 3401 * Issue RST and make ACK acceptable to originator of segment. 3402 * The mbuf must still include the original packet header. 3403 * tp may be NULL. 3404 */ 3405 void 3406 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3407 int tlen, int rstreason) 3408 { 3409 #ifdef INET 3410 struct ip *ip; 3411 #endif 3412 #ifdef INET6 3413 struct ip6_hdr *ip6; 3414 #endif 3415 3416 if (tp != NULL) { 3417 INP_LOCK_ASSERT(tp->t_inpcb); 3418 } 3419 3420 /* Don't bother if destination was broadcast/multicast. */ 3421 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3422 goto drop; 3423 #ifdef INET6 3424 if (mtod(m, struct ip *)->ip_v == 6) { 3425 ip6 = mtod(m, struct ip6_hdr *); 3426 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3427 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3428 goto drop; 3429 /* IPv6 anycast check is done at tcp6_input() */ 3430 } 3431 #endif 3432 #if defined(INET) && defined(INET6) 3433 else 3434 #endif 3435 #ifdef INET 3436 { 3437 ip = mtod(m, struct ip *); 3438 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3439 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3440 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3441 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3442 goto drop; 3443 } 3444 #endif 3445 3446 /* Perform bandwidth limiting. */ 3447 if (badport_bandlim(rstreason) < 0) 3448 goto drop; 3449 3450 /* tcp_respond consumes the mbuf chain. */ 3451 if (th->th_flags & TH_ACK) { 3452 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3453 th->th_ack, TH_RST); 3454 } else { 3455 if (th->th_flags & TH_SYN) 3456 tlen++; 3457 if (th->th_flags & TH_FIN) 3458 tlen++; 3459 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3460 (tcp_seq)0, TH_RST|TH_ACK); 3461 } 3462 return; 3463 drop: 3464 m_freem(m); 3465 } 3466 3467 /* 3468 * Parse TCP options and place in tcpopt. 3469 */ 3470 void 3471 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3472 { 3473 int opt, optlen; 3474 3475 to->to_flags = 0; 3476 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3477 opt = cp[0]; 3478 if (opt == TCPOPT_EOL) 3479 break; 3480 if (opt == TCPOPT_NOP) 3481 optlen = 1; 3482 else { 3483 if (cnt < 2) 3484 break; 3485 optlen = cp[1]; 3486 if (optlen < 2 || optlen > cnt) 3487 break; 3488 } 3489 switch (opt) { 3490 case TCPOPT_MAXSEG: 3491 if (optlen != TCPOLEN_MAXSEG) 3492 continue; 3493 if (!(flags & TO_SYN)) 3494 continue; 3495 to->to_flags |= TOF_MSS; 3496 bcopy((char *)cp + 2, 3497 (char *)&to->to_mss, sizeof(to->to_mss)); 3498 to->to_mss = ntohs(to->to_mss); 3499 break; 3500 case TCPOPT_WINDOW: 3501 if (optlen != TCPOLEN_WINDOW) 3502 continue; 3503 if (!(flags & TO_SYN)) 3504 continue; 3505 to->to_flags |= TOF_SCALE; 3506 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3507 break; 3508 case TCPOPT_TIMESTAMP: 3509 if (optlen != TCPOLEN_TIMESTAMP) 3510 continue; 3511 to->to_flags |= TOF_TS; 3512 bcopy((char *)cp + 2, 3513 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3514 to->to_tsval = ntohl(to->to_tsval); 3515 bcopy((char *)cp + 6, 3516 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3517 to->to_tsecr = ntohl(to->to_tsecr); 3518 break; 3519 case TCPOPT_SIGNATURE: 3520 /* 3521 * In order to reply to a host which has set the 3522 * TCP_SIGNATURE option in its initial SYN, we have 3523 * to record the fact that the option was observed 3524 * here for the syncache code to perform the correct 3525 * response. 3526 */ 3527 if (optlen != TCPOLEN_SIGNATURE) 3528 continue; 3529 to->to_flags |= TOF_SIGNATURE; 3530 to->to_signature = cp + 2; 3531 break; 3532 case TCPOPT_SACK_PERMITTED: 3533 if (optlen != TCPOLEN_SACK_PERMITTED) 3534 continue; 3535 if (!(flags & TO_SYN)) 3536 continue; 3537 if (!V_tcp_do_sack) 3538 continue; 3539 to->to_flags |= TOF_SACKPERM; 3540 break; 3541 case TCPOPT_SACK: 3542 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3543 continue; 3544 if (flags & TO_SYN) 3545 continue; 3546 to->to_flags |= TOF_SACK; 3547 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3548 to->to_sacks = cp + 2; 3549 TCPSTAT_INC(tcps_sack_rcv_blocks); 3550 break; 3551 case TCPOPT_FAST_OPEN: 3552 /* 3553 * Cookie length validation is performed by the 3554 * server side cookie checking code or the client 3555 * side cookie cache update code. 3556 */ 3557 if (!(flags & TO_SYN)) 3558 continue; 3559 if (!V_tcp_fastopen_client_enable && 3560 !V_tcp_fastopen_server_enable) 3561 continue; 3562 to->to_flags |= TOF_FASTOPEN; 3563 to->to_tfo_len = optlen - 2; 3564 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3565 break; 3566 default: 3567 continue; 3568 } 3569 } 3570 } 3571 3572 /* 3573 * Pull out of band byte out of a segment so 3574 * it doesn't appear in the user's data queue. 3575 * It is still reflected in the segment length for 3576 * sequencing purposes. 3577 */ 3578 void 3579 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3580 int off) 3581 { 3582 int cnt = off + th->th_urp - 1; 3583 3584 while (cnt >= 0) { 3585 if (m->m_len > cnt) { 3586 char *cp = mtod(m, caddr_t) + cnt; 3587 struct tcpcb *tp = sototcpcb(so); 3588 3589 INP_WLOCK_ASSERT(tp->t_inpcb); 3590 3591 tp->t_iobc = *cp; 3592 tp->t_oobflags |= TCPOOB_HAVEDATA; 3593 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3594 m->m_len--; 3595 if (m->m_flags & M_PKTHDR) 3596 m->m_pkthdr.len--; 3597 return; 3598 } 3599 cnt -= m->m_len; 3600 m = m->m_next; 3601 if (m == NULL) 3602 break; 3603 } 3604 panic("tcp_pulloutofband"); 3605 } 3606 3607 /* 3608 * Collect new round-trip time estimate 3609 * and update averages and current timeout. 3610 */ 3611 void 3612 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3613 { 3614 int delta; 3615 3616 INP_WLOCK_ASSERT(tp->t_inpcb); 3617 3618 TCPSTAT_INC(tcps_rttupdated); 3619 tp->t_rttupdated++; 3620 #ifdef STATS 3621 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3622 imax(0, rtt * 1000 / hz)); 3623 #endif 3624 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3625 /* 3626 * srtt is stored as fixed point with 5 bits after the 3627 * binary point (i.e., scaled by 8). The following magic 3628 * is equivalent to the smoothing algorithm in rfc793 with 3629 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3630 * point). Adjust rtt to origin 0. 3631 */ 3632 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3633 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3634 3635 if ((tp->t_srtt += delta) <= 0) 3636 tp->t_srtt = 1; 3637 3638 /* 3639 * We accumulate a smoothed rtt variance (actually, a 3640 * smoothed mean difference), then set the retransmit 3641 * timer to smoothed rtt + 4 times the smoothed variance. 3642 * rttvar is stored as fixed point with 4 bits after the 3643 * binary point (scaled by 16). The following is 3644 * equivalent to rfc793 smoothing with an alpha of .75 3645 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3646 * rfc793's wired-in beta. 3647 */ 3648 if (delta < 0) 3649 delta = -delta; 3650 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3651 if ((tp->t_rttvar += delta) <= 0) 3652 tp->t_rttvar = 1; 3653 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3654 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3655 } else { 3656 /* 3657 * No rtt measurement yet - use the unsmoothed rtt. 3658 * Set the variance to half the rtt (so our first 3659 * retransmit happens at 3*rtt). 3660 */ 3661 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3662 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3663 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3664 } 3665 tp->t_rtttime = 0; 3666 tp->t_rxtshift = 0; 3667 3668 /* 3669 * the retransmit should happen at rtt + 4 * rttvar. 3670 * Because of the way we do the smoothing, srtt and rttvar 3671 * will each average +1/2 tick of bias. When we compute 3672 * the retransmit timer, we want 1/2 tick of rounding and 3673 * 1 extra tick because of +-1/2 tick uncertainty in the 3674 * firing of the timer. The bias will give us exactly the 3675 * 1.5 tick we need. But, because the bias is 3676 * statistical, we have to test that we don't drop below 3677 * the minimum feasible timer (which is 2 ticks). 3678 */ 3679 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3680 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3681 3682 /* 3683 * We received an ack for a packet that wasn't retransmitted; 3684 * it is probably safe to discard any error indications we've 3685 * received recently. This isn't quite right, but close enough 3686 * for now (a route might have failed after we sent a segment, 3687 * and the return path might not be symmetrical). 3688 */ 3689 tp->t_softerror = 0; 3690 } 3691 3692 /* 3693 * Determine a reasonable value for maxseg size. 3694 * If the route is known, check route for mtu. 3695 * If none, use an mss that can be handled on the outgoing interface 3696 * without forcing IP to fragment. If no route is found, route has no mtu, 3697 * or the destination isn't local, use a default, hopefully conservative 3698 * size (usually 512 or the default IP max size, but no more than the mtu 3699 * of the interface), as we can't discover anything about intervening 3700 * gateways or networks. We also initialize the congestion/slow start 3701 * window to be a single segment if the destination isn't local. 3702 * While looking at the routing entry, we also initialize other path-dependent 3703 * parameters from pre-set or cached values in the routing entry. 3704 * 3705 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3706 * IP options, e.g. IPSEC data, since length of this data may vary, and 3707 * thus it is calculated for every segment separately in tcp_output(). 3708 * 3709 * NOTE that this routine is only called when we process an incoming 3710 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3711 * settings are handled in tcp_mssopt(). 3712 */ 3713 void 3714 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3715 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3716 { 3717 int mss = 0; 3718 uint32_t maxmtu = 0; 3719 struct inpcb *inp = tp->t_inpcb; 3720 struct hc_metrics_lite metrics; 3721 #ifdef INET6 3722 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3723 size_t min_protoh = isipv6 ? 3724 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3725 sizeof (struct tcpiphdr); 3726 #else 3727 size_t min_protoh = sizeof(struct tcpiphdr); 3728 #endif 3729 3730 INP_WLOCK_ASSERT(tp->t_inpcb); 3731 3732 if (tp->t_port) 3733 min_protoh += V_tcp_udp_tunneling_overhead; 3734 if (mtuoffer != -1) { 3735 KASSERT(offer == -1, ("%s: conflict", __func__)); 3736 offer = mtuoffer - min_protoh; 3737 } 3738 3739 /* Initialize. */ 3740 #ifdef INET6 3741 if (isipv6) { 3742 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3743 tp->t_maxseg = V_tcp_v6mssdflt; 3744 } 3745 #endif 3746 #if defined(INET) && defined(INET6) 3747 else 3748 #endif 3749 #ifdef INET 3750 { 3751 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3752 tp->t_maxseg = V_tcp_mssdflt; 3753 } 3754 #endif 3755 3756 /* 3757 * No route to sender, stay with default mss and return. 3758 */ 3759 if (maxmtu == 0) { 3760 /* 3761 * In case we return early we need to initialize metrics 3762 * to a defined state as tcp_hc_get() would do for us 3763 * if there was no cache hit. 3764 */ 3765 if (metricptr != NULL) 3766 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3767 return; 3768 } 3769 3770 /* What have we got? */ 3771 switch (offer) { 3772 case 0: 3773 /* 3774 * Offer == 0 means that there was no MSS on the SYN 3775 * segment, in this case we use tcp_mssdflt as 3776 * already assigned to t_maxseg above. 3777 */ 3778 offer = tp->t_maxseg; 3779 break; 3780 3781 case -1: 3782 /* 3783 * Offer == -1 means that we didn't receive SYN yet. 3784 */ 3785 /* FALLTHROUGH */ 3786 3787 default: 3788 /* 3789 * Prevent DoS attack with too small MSS. Round up 3790 * to at least minmss. 3791 */ 3792 offer = max(offer, V_tcp_minmss); 3793 } 3794 3795 /* 3796 * rmx information is now retrieved from tcp_hostcache. 3797 */ 3798 tcp_hc_get(&inp->inp_inc, &metrics); 3799 if (metricptr != NULL) 3800 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3801 3802 /* 3803 * If there's a discovered mtu in tcp hostcache, use it. 3804 * Else, use the link mtu. 3805 */ 3806 if (metrics.rmx_mtu) 3807 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3808 else { 3809 #ifdef INET6 3810 if (isipv6) { 3811 mss = maxmtu - min_protoh; 3812 if (!V_path_mtu_discovery && 3813 !in6_localaddr(&inp->in6p_faddr)) 3814 mss = min(mss, V_tcp_v6mssdflt); 3815 } 3816 #endif 3817 #if defined(INET) && defined(INET6) 3818 else 3819 #endif 3820 #ifdef INET 3821 { 3822 mss = maxmtu - min_protoh; 3823 if (!V_path_mtu_discovery && 3824 !in_localaddr(inp->inp_faddr)) 3825 mss = min(mss, V_tcp_mssdflt); 3826 } 3827 #endif 3828 /* 3829 * XXX - The above conditional (mss = maxmtu - min_protoh) 3830 * probably violates the TCP spec. 3831 * The problem is that, since we don't know the 3832 * other end's MSS, we are supposed to use a conservative 3833 * default. But, if we do that, then MTU discovery will 3834 * never actually take place, because the conservative 3835 * default is much less than the MTUs typically seen 3836 * on the Internet today. For the moment, we'll sweep 3837 * this under the carpet. 3838 * 3839 * The conservative default might not actually be a problem 3840 * if the only case this occurs is when sending an initial 3841 * SYN with options and data to a host we've never talked 3842 * to before. Then, they will reply with an MSS value which 3843 * will get recorded and the new parameters should get 3844 * recomputed. For Further Study. 3845 */ 3846 } 3847 mss = min(mss, offer); 3848 3849 /* 3850 * Sanity check: make sure that maxseg will be large 3851 * enough to allow some data on segments even if the 3852 * all the option space is used (40bytes). Otherwise 3853 * funny things may happen in tcp_output. 3854 * 3855 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3856 */ 3857 mss = max(mss, 64); 3858 3859 tp->t_maxseg = mss; 3860 } 3861 3862 void 3863 tcp_mss(struct tcpcb *tp, int offer) 3864 { 3865 int mss; 3866 uint32_t bufsize; 3867 struct inpcb *inp; 3868 struct socket *so; 3869 struct hc_metrics_lite metrics; 3870 struct tcp_ifcap cap; 3871 3872 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3873 3874 bzero(&cap, sizeof(cap)); 3875 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3876 3877 mss = tp->t_maxseg; 3878 inp = tp->t_inpcb; 3879 3880 /* 3881 * If there's a pipesize, change the socket buffer to that size, 3882 * don't change if sb_hiwat is different than default (then it 3883 * has been changed on purpose with setsockopt). 3884 * Make the socket buffers an integral number of mss units; 3885 * if the mss is larger than the socket buffer, decrease the mss. 3886 */ 3887 so = inp->inp_socket; 3888 SOCKBUF_LOCK(&so->so_snd); 3889 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3890 bufsize = metrics.rmx_sendpipe; 3891 else 3892 bufsize = so->so_snd.sb_hiwat; 3893 if (bufsize < mss) 3894 mss = bufsize; 3895 else { 3896 bufsize = roundup(bufsize, mss); 3897 if (bufsize > sb_max) 3898 bufsize = sb_max; 3899 if (bufsize > so->so_snd.sb_hiwat) 3900 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3901 } 3902 SOCKBUF_UNLOCK(&so->so_snd); 3903 /* 3904 * Sanity check: make sure that maxseg will be large 3905 * enough to allow some data on segments even if the 3906 * all the option space is used (40bytes). Otherwise 3907 * funny things may happen in tcp_output. 3908 * 3909 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3910 */ 3911 tp->t_maxseg = max(mss, 64); 3912 3913 SOCKBUF_LOCK(&so->so_rcv); 3914 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3915 bufsize = metrics.rmx_recvpipe; 3916 else 3917 bufsize = so->so_rcv.sb_hiwat; 3918 if (bufsize > mss) { 3919 bufsize = roundup(bufsize, mss); 3920 if (bufsize > sb_max) 3921 bufsize = sb_max; 3922 if (bufsize > so->so_rcv.sb_hiwat) 3923 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3924 } 3925 SOCKBUF_UNLOCK(&so->so_rcv); 3926 3927 /* Check the interface for TSO capabilities. */ 3928 if (cap.ifcap & CSUM_TSO) { 3929 tp->t_flags |= TF_TSO; 3930 tp->t_tsomax = cap.tsomax; 3931 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3932 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3933 } 3934 } 3935 3936 /* 3937 * Determine the MSS option to send on an outgoing SYN. 3938 */ 3939 int 3940 tcp_mssopt(struct in_conninfo *inc) 3941 { 3942 int mss = 0; 3943 uint32_t thcmtu = 0; 3944 uint32_t maxmtu = 0; 3945 size_t min_protoh; 3946 3947 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3948 3949 #ifdef INET6 3950 if (inc->inc_flags & INC_ISIPV6) { 3951 mss = V_tcp_v6mssdflt; 3952 maxmtu = tcp_maxmtu6(inc, NULL); 3953 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3954 } 3955 #endif 3956 #if defined(INET) && defined(INET6) 3957 else 3958 #endif 3959 #ifdef INET 3960 { 3961 mss = V_tcp_mssdflt; 3962 maxmtu = tcp_maxmtu(inc, NULL); 3963 min_protoh = sizeof(struct tcpiphdr); 3964 } 3965 #endif 3966 #if defined(INET6) || defined(INET) 3967 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3968 #endif 3969 3970 if (maxmtu && thcmtu) 3971 mss = min(maxmtu, thcmtu) - min_protoh; 3972 else if (maxmtu || thcmtu) 3973 mss = max(maxmtu, thcmtu) - min_protoh; 3974 3975 return (mss); 3976 } 3977 3978 void 3979 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3980 { 3981 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3982 int maxseg = tcp_maxseg(tp); 3983 3984 INP_WLOCK_ASSERT(tp->t_inpcb); 3985 3986 /* 3987 * Compute the amount of data that this ACK is indicating 3988 * (del_data) and an estimate of how many bytes are in the 3989 * network. 3990 */ 3991 if (((tp->t_flags & TF_SACK_PERMIT) && 3992 (to->to_flags & TOF_SACK)) || 3993 (IN_CONGRECOVERY(tp->t_flags) && 3994 !IN_FASTRECOVERY(tp->t_flags))) { 3995 del_data = tp->sackhint.delivered_data; 3996 if (V_tcp_do_newsack) 3997 pipe = tcp_compute_pipe(tp); 3998 else 3999 pipe = (tp->snd_nxt - tp->snd_fack) + 4000 tp->sackhint.sack_bytes_rexmit; 4001 } else { 4002 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 4003 tp->snd_recover - tp->snd_una)) 4004 del_data = maxseg; 4005 pipe = imax(0, tp->snd_max - tp->snd_una - 4006 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 4007 } 4008 tp->sackhint.prr_delivered += del_data; 4009 /* 4010 * Proportional Rate Reduction 4011 */ 4012 if (pipe >= tp->snd_ssthresh) { 4013 if (tp->sackhint.recover_fs == 0) 4014 tp->sackhint.recover_fs = 4015 imax(1, tp->snd_nxt - tp->snd_una); 4016 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 4017 tp->snd_ssthresh, tp->sackhint.recover_fs) - 4018 tp->sackhint.prr_out; 4019 } else { 4020 if (V_tcp_do_prr_conservative || (del_data == 0)) 4021 limit = tp->sackhint.prr_delivered - 4022 tp->sackhint.prr_out; 4023 else 4024 limit = imax(tp->sackhint.prr_delivered - 4025 tp->sackhint.prr_out, del_data) + 4026 maxseg; 4027 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 4028 } 4029 snd_cnt = imax(snd_cnt, 0) / maxseg; 4030 /* 4031 * Send snd_cnt new data into the network in response to this ack. 4032 * If there is going to be a SACK retransmission, adjust snd_cwnd 4033 * accordingly. 4034 */ 4035 if (IN_FASTRECOVERY(tp->t_flags)) { 4036 if ((tp->t_flags & TF_SACK_PERMIT) && 4037 (to->to_flags & TOF_SACK)) { 4038 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 4039 tp->sackhint.sack_bytes_rexmit + 4040 (snd_cnt * maxseg); 4041 } else { 4042 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4043 (snd_cnt * maxseg); 4044 } 4045 } else if (IN_CONGRECOVERY(tp->t_flags)) 4046 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4047 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4048 } 4049 4050 /* 4051 * On a partial ack arrives, force the retransmission of the 4052 * next unacknowledged segment. Do not clear tp->t_dupacks. 4053 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4054 * be started again. 4055 */ 4056 void 4057 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4058 { 4059 tcp_seq onxt = tp->snd_nxt; 4060 uint32_t ocwnd = tp->snd_cwnd; 4061 u_int maxseg = tcp_maxseg(tp); 4062 4063 INP_WLOCK_ASSERT(tp->t_inpcb); 4064 4065 tcp_timer_activate(tp, TT_REXMT, 0); 4066 tp->t_rtttime = 0; 4067 tp->snd_nxt = th->th_ack; 4068 /* 4069 * Set snd_cwnd to one segment beyond acknowledged offset. 4070 * (tp->snd_una has not yet been updated when this function is called.) 4071 */ 4072 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4073 tp->t_flags |= TF_ACKNOW; 4074 (void) tp->t_fb->tfb_tcp_output(tp); 4075 tp->snd_cwnd = ocwnd; 4076 if (SEQ_GT(onxt, tp->snd_nxt)) 4077 tp->snd_nxt = onxt; 4078 /* 4079 * Partial window deflation. Relies on fact that tp->snd_una 4080 * not updated yet. 4081 */ 4082 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4083 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4084 else 4085 tp->snd_cwnd = 0; 4086 tp->snd_cwnd += maxseg; 4087 } 4088 4089 int 4090 tcp_compute_pipe(struct tcpcb *tp) 4091 { 4092 return (tp->snd_max - tp->snd_una + 4093 tp->sackhint.sack_bytes_rexmit - 4094 tp->sackhint.sacked_bytes); 4095 } 4096 4097 uint32_t 4098 tcp_compute_initwnd(uint32_t maxseg) 4099 { 4100 /* 4101 * Calculate the Initial Window, also used as Restart Window 4102 * 4103 * RFC5681 Section 3.1 specifies the default conservative values. 4104 * RFC3390 specifies slightly more aggressive values. 4105 * RFC6928 increases it to ten segments. 4106 * Support for user specified value for initial flight size. 4107 */ 4108 if (V_tcp_initcwnd_segments) 4109 return min(V_tcp_initcwnd_segments * maxseg, 4110 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4111 else if (V_tcp_do_rfc3390) 4112 return min(4 * maxseg, max(2 * maxseg, 4380)); 4113 else { 4114 /* Per RFC5681 Section 3.1 */ 4115 if (maxseg > 2190) 4116 return (2 * maxseg); 4117 else if (maxseg > 1095) 4118 return (3 * maxseg); 4119 else 4120 return (4 * maxseg); 4121 } 4122 } 4123