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 #include <netinet/tcp_debug.h> 121 #ifdef TCP_OFFLOAD 122 #include <netinet/tcp_offload.h> 123 #endif 124 #include <netinet/tcp_ecn.h> 125 #include <netinet/udp.h> 126 127 #include <netipsec/ipsec_support.h> 128 129 #include <machine/in_cksum.h> 130 131 #include <security/mac/mac_framework.h> 132 133 const int tcprexmtthresh = 3; 134 135 VNET_DEFINE(int, tcp_log_in_vain) = 0; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 137 &VNET_NAME(tcp_log_in_vain), 0, 138 "Log all incoming TCP segments to closed ports"); 139 140 VNET_DEFINE(int, blackhole) = 0; 141 #define V_blackhole VNET(blackhole) 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 143 &VNET_NAME(blackhole), 0, 144 "Do not send RST on segments to closed ports"); 145 146 VNET_DEFINE(bool, blackhole_local) = false; 147 #define V_blackhole_local VNET(blackhole_local) 148 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | 149 CTLFLAG_RW, &VNET_NAME(blackhole_local), false, 150 "Enforce net.inet.tcp.blackhole for locally originated packets"); 151 152 VNET_DEFINE(int, tcp_delack_enabled) = 1; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 154 &VNET_NAME(tcp_delack_enabled), 0, 155 "Delay ACK to try and piggyback it onto a data packet"); 156 157 VNET_DEFINE(int, drop_synfin) = 0; 158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 159 &VNET_NAME(drop_synfin), 0, 160 "Drop TCP packets with SYN+FIN set"); 161 162 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_prr_conservative), 0, 165 "Do conservative Proportional Rate Reduction"); 166 167 VNET_DEFINE(int, tcp_do_prr) = 1; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_prr), 1, 170 "Enable Proportional Rate Reduction per RFC 6937"); 171 172 VNET_DEFINE(int, tcp_do_lrd) = 0; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_lrd, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_do_lrd), 1, 175 "Perform Lost Retransmission Detection"); 176 177 VNET_DEFINE(int, tcp_do_newcwv) = 0; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 179 &VNET_NAME(tcp_do_newcwv), 0, 180 "Enable New Congestion Window Validation per RFC7661"); 181 182 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 184 &VNET_NAME(tcp_do_rfc3042), 0, 185 "Enable RFC 3042 (Limited Transmit)"); 186 187 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 189 &VNET_NAME(tcp_do_rfc3390), 0, 190 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 191 192 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 194 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 195 "Slow-start flight size (initial congestion window) in number of segments"); 196 197 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_do_rfc3465), 0, 200 "Enable RFC 3465 (Appropriate Byte Counting)"); 201 202 VNET_DEFINE(int, tcp_abc_l_var) = 2; 203 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 204 &VNET_NAME(tcp_abc_l_var), 2, 205 "Cap the max cwnd increment during slow-start to this number of segments"); 206 207 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 208 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 209 "TCP ECN"); 210 211 VNET_DEFINE(int, tcp_do_ecn) = 2; 212 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(tcp_do_ecn), 0, 214 "TCP ECN support"); 215 216 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 217 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(tcp_ecn_maxretries), 0, 219 "Max retries before giving up on ECN"); 220 221 VNET_DEFINE(int, tcp_insecure_syn) = 0; 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_insecure_syn), 0, 224 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 225 226 VNET_DEFINE(int, tcp_insecure_rst) = 0; 227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_insecure_rst), 0, 229 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 230 231 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 232 #define V_tcp_recvspace VNET(tcp_recvspace) 233 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 234 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 235 236 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 237 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 238 &VNET_NAME(tcp_do_autorcvbuf), 0, 239 "Enable automatic receive buffer sizing"); 240 241 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 242 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 243 &VNET_NAME(tcp_autorcvbuf_max), 0, 244 "Max size of automatic receive buffer"); 245 246 VNET_DEFINE(struct inpcbinfo, tcbinfo); 247 248 /* 249 * TCP statistics are stored in an array of counter(9)s, which size matches 250 * size of struct tcpstat. TCP running connection count is a regular array. 251 */ 252 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 253 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 254 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 255 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 256 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 257 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 258 "TCP connection counts by TCP state"); 259 260 /* 261 * Kernel module interface for updating tcpstat. The first argument is an index 262 * into tcpstat treated as an array. 263 */ 264 void 265 kmod_tcpstat_add(int statnum, int val) 266 { 267 268 counter_u64_add(VNET(tcpstat)[statnum], val); 269 } 270 271 /* 272 * Make sure that we only start a SACK loss recovery when 273 * receiving a duplicate ACK with a SACK block, and also 274 * complete SACK loss recovery in case the other end 275 * reneges. 276 */ 277 static bool inline 278 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) 279 { 280 return ((tp->t_flags & TF_SACK_PERMIT) && 281 ((to->to_flags & TOF_SACK) || 282 (!TAILQ_EMPTY(&tp->snd_holes)))); 283 } 284 285 #ifdef TCP_HHOOK 286 /* 287 * Wrapper for the TCP established input helper hook. 288 */ 289 void 290 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 291 { 292 struct tcp_hhook_data hhook_data; 293 294 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 295 hhook_data.tp = tp; 296 hhook_data.th = th; 297 hhook_data.to = to; 298 299 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 300 tp->osd); 301 } 302 } 303 #endif 304 305 /* 306 * CC wrapper hook functions 307 */ 308 void 309 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 310 uint16_t type) 311 { 312 #ifdef STATS 313 int32_t gput; 314 #endif 315 316 INP_WLOCK_ASSERT(tp->t_inpcb); 317 318 tp->ccv->nsegs = nsegs; 319 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 320 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 321 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 322 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 323 tp->ccv->flags |= CCF_CWND_LIMITED; 324 else 325 tp->ccv->flags &= ~CCF_CWND_LIMITED; 326 327 if (type == CC_ACK) { 328 #ifdef STATS 329 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 330 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 331 if (!IN_RECOVERY(tp->t_flags)) 332 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 333 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 334 if ((tp->t_flags & TF_GPUTINPROG) && 335 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 336 /* 337 * Compute goodput in bits per millisecond. 338 */ 339 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / 340 max(1, tcp_ts_getticks() - tp->gput_ts); 341 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 342 gput); 343 /* 344 * XXXLAS: This is a temporary hack, and should be 345 * chained off VOI_TCP_GPUT when stats(9) grows an API 346 * to deal with chained VOIs. 347 */ 348 if (tp->t_stats_gput_prev > 0) 349 stats_voi_update_abs_s32(tp->t_stats, 350 VOI_TCP_GPUT_ND, 351 ((gput - tp->t_stats_gput_prev) * 100) / 352 tp->t_stats_gput_prev); 353 tp->t_flags &= ~TF_GPUTINPROG; 354 tp->t_stats_gput_prev = gput; 355 } 356 #endif /* STATS */ 357 if (tp->snd_cwnd > tp->snd_ssthresh) { 358 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 359 if (tp->t_bytes_acked >= tp->snd_cwnd) { 360 tp->t_bytes_acked -= tp->snd_cwnd; 361 tp->ccv->flags |= CCF_ABC_SENTAWND; 362 } 363 } else { 364 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 365 tp->t_bytes_acked = 0; 366 } 367 } 368 369 if (CC_ALGO(tp)->ack_received != NULL) { 370 /* XXXLAS: Find a way to live without this */ 371 tp->ccv->curack = th->th_ack; 372 CC_ALGO(tp)->ack_received(tp->ccv, type); 373 } 374 #ifdef STATS 375 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 376 #endif 377 } 378 379 void 380 cc_conn_init(struct tcpcb *tp) 381 { 382 struct hc_metrics_lite metrics; 383 struct inpcb *inp = tp->t_inpcb; 384 u_int maxseg; 385 int rtt; 386 387 INP_WLOCK_ASSERT(tp->t_inpcb); 388 389 tcp_hc_get(&inp->inp_inc, &metrics); 390 maxseg = tcp_maxseg(tp); 391 392 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 393 tp->t_srtt = rtt; 394 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 395 TCPSTAT_INC(tcps_usedrtt); 396 if (metrics.rmx_rttvar) { 397 tp->t_rttvar = metrics.rmx_rttvar; 398 TCPSTAT_INC(tcps_usedrttvar); 399 } else { 400 /* default variation is +- 1 rtt */ 401 tp->t_rttvar = 402 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 403 } 404 TCPT_RANGESET(tp->t_rxtcur, 405 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 406 tp->t_rttmin, TCPTV_REXMTMAX); 407 } 408 if (metrics.rmx_ssthresh) { 409 /* 410 * There's some sort of gateway or interface 411 * buffer limit on the path. Use this to set 412 * the slow start threshold, but set the 413 * threshold to no less than 2*mss. 414 */ 415 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 416 TCPSTAT_INC(tcps_usedssthresh); 417 } 418 419 /* 420 * Set the initial slow-start flight size. 421 * 422 * If a SYN or SYN/ACK was lost and retransmitted, we have to 423 * reduce the initial CWND to one segment as congestion is likely 424 * requiring us to be cautious. 425 */ 426 if (tp->snd_cwnd == 1) 427 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 428 else 429 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 430 431 if (CC_ALGO(tp)->conn_init != NULL) 432 CC_ALGO(tp)->conn_init(tp->ccv); 433 } 434 435 void inline 436 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 437 { 438 INP_WLOCK_ASSERT(tp->t_inpcb); 439 440 #ifdef STATS 441 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 442 #endif 443 444 switch(type) { 445 case CC_NDUPACK: 446 if (!IN_FASTRECOVERY(tp->t_flags)) { 447 tp->snd_recover = tp->snd_max; 448 if (tp->t_flags2 & TF2_ECN_PERMIT) 449 tp->t_flags2 |= TF2_ECN_SND_CWR; 450 } 451 break; 452 case CC_ECN: 453 if (!IN_CONGRECOVERY(tp->t_flags) || 454 /* 455 * Allow ECN reaction on ACK to CWR, if 456 * that data segment was also CE marked. 457 */ 458 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 459 EXIT_CONGRECOVERY(tp->t_flags); 460 TCPSTAT_INC(tcps_ecn_rcwnd); 461 tp->snd_recover = tp->snd_max + 1; 462 if (tp->t_flags2 & TF2_ECN_PERMIT) 463 tp->t_flags2 |= TF2_ECN_SND_CWR; 464 } 465 break; 466 case CC_RTO: 467 tp->t_dupacks = 0; 468 tp->t_bytes_acked = 0; 469 EXIT_RECOVERY(tp->t_flags); 470 if (tp->t_flags2 & TF2_ECN_PERMIT) 471 tp->t_flags2 |= TF2_ECN_SND_CWR; 472 break; 473 case CC_RTO_ERR: 474 TCPSTAT_INC(tcps_sndrexmitbad); 475 /* RTO was unnecessary, so reset everything. */ 476 tp->snd_cwnd = tp->snd_cwnd_prev; 477 tp->snd_ssthresh = tp->snd_ssthresh_prev; 478 tp->snd_recover = tp->snd_recover_prev; 479 if (tp->t_flags & TF_WASFRECOVERY) 480 ENTER_FASTRECOVERY(tp->t_flags); 481 if (tp->t_flags & TF_WASCRECOVERY) 482 ENTER_CONGRECOVERY(tp->t_flags); 483 tp->snd_nxt = tp->snd_max; 484 tp->t_flags &= ~TF_PREVVALID; 485 tp->t_badrxtwin = 0; 486 break; 487 } 488 489 if (CC_ALGO(tp)->cong_signal != NULL) { 490 if (th != NULL) 491 tp->ccv->curack = th->th_ack; 492 CC_ALGO(tp)->cong_signal(tp->ccv, type); 493 } 494 } 495 496 void inline 497 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 498 { 499 INP_WLOCK_ASSERT(tp->t_inpcb); 500 501 /* XXXLAS: KASSERT that we're in recovery? */ 502 503 if (CC_ALGO(tp)->post_recovery != NULL) { 504 tp->ccv->curack = th->th_ack; 505 CC_ALGO(tp)->post_recovery(tp->ccv); 506 } 507 /* XXXLAS: EXIT_RECOVERY ? */ 508 tp->t_bytes_acked = 0; 509 tp->sackhint.delivered_data = 0; 510 tp->sackhint.prr_out = 0; 511 } 512 513 /* 514 * Indicate whether this ack should be delayed. We can delay the ack if 515 * following conditions are met: 516 * - There is no delayed ack timer in progress. 517 * - Our last ack wasn't a 0-sized window. We never want to delay 518 * the ack that opens up a 0-sized window. 519 * - LRO wasn't used for this segment. We make sure by checking that the 520 * segment size is not larger than the MSS. 521 */ 522 #define DELAY_ACK(tp, tlen) \ 523 ((!tcp_timer_active(tp, TT_DELACK) && \ 524 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 525 (tlen <= tp->t_maxseg) && \ 526 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 527 528 void inline 529 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) 530 { 531 INP_WLOCK_ASSERT(tp->t_inpcb); 532 533 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 534 switch (iptos & IPTOS_ECN_MASK) { 535 case IPTOS_ECN_CE: 536 tp->ccv->flags |= CCF_IPHDR_CE; 537 break; 538 case IPTOS_ECN_ECT0: 539 /* FALLTHROUGH */ 540 case IPTOS_ECN_ECT1: 541 /* FALLTHROUGH */ 542 case IPTOS_ECN_NOTECT: 543 tp->ccv->flags &= ~CCF_IPHDR_CE; 544 break; 545 } 546 547 if (flags & TH_CWR) 548 tp->ccv->flags |= CCF_TCPHDR_CWR; 549 else 550 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 551 552 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 553 554 if (tp->ccv->flags & CCF_ACKNOW) { 555 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 556 tp->t_flags |= TF_ACKNOW; 557 } 558 } 559 } 560 561 void inline 562 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 563 { 564 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); 565 } 566 567 /* 568 * TCP input handling is split into multiple parts: 569 * tcp6_input is a thin wrapper around tcp_input for the extended 570 * ip6_protox[] call format in ip6_input 571 * tcp_input handles primary segment validation, inpcb lookup and 572 * SYN processing on listen sockets 573 * tcp_do_segment processes the ACK and text of the segment for 574 * establishing, established and closing connections 575 */ 576 #ifdef INET6 577 int 578 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 579 { 580 struct mbuf *m; 581 struct in6_ifaddr *ia6; 582 struct ip6_hdr *ip6; 583 584 m = *mp; 585 if (m->m_len < *offp + sizeof(struct tcphdr)) { 586 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 587 if (m == NULL) { 588 *mp = m; 589 TCPSTAT_INC(tcps_rcvshort); 590 return (IPPROTO_DONE); 591 } 592 } 593 594 /* 595 * draft-itojun-ipv6-tcp-to-anycast 596 * better place to put this in? 597 */ 598 ip6 = mtod(m, struct ip6_hdr *); 599 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 600 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 601 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 602 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 603 *mp = NULL; 604 return (IPPROTO_DONE); 605 } 606 607 *mp = m; 608 return (tcp_input_with_port(mp, offp, proto, port)); 609 } 610 611 int 612 tcp6_input(struct mbuf **mp, int *offp, int proto) 613 { 614 615 return(tcp6_input_with_port(mp, offp, proto, 0)); 616 } 617 #endif /* INET6 */ 618 619 int 620 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) 621 { 622 struct mbuf *m = *mp; 623 struct tcphdr *th = NULL; 624 struct ip *ip = NULL; 625 struct inpcb *inp = NULL; 626 struct tcpcb *tp = NULL; 627 struct socket *so = NULL; 628 u_char *optp = NULL; 629 int off0; 630 int optlen = 0; 631 #ifdef INET 632 int len; 633 uint8_t ipttl; 634 #endif 635 int tlen = 0, off; 636 int drop_hdrlen; 637 int thflags; 638 int rstreason = 0; /* For badport_bandlim accounting purposes */ 639 int lookupflag; 640 uint8_t iptos; 641 struct m_tag *fwd_tag = NULL; 642 #ifdef INET6 643 struct ip6_hdr *ip6 = NULL; 644 int isipv6; 645 #else 646 const void *ip6 = NULL; 647 #endif /* INET6 */ 648 struct tcpopt to; /* options in this segment */ 649 char *s = NULL; /* address and port logging */ 650 #ifdef TCPDEBUG 651 /* 652 * The size of tcp_saveipgen must be the size of the max ip header, 653 * now IPv6. 654 */ 655 u_char tcp_saveipgen[IP6_HDR_LEN]; 656 struct tcphdr tcp_savetcp; 657 short ostate = 0; 658 #endif 659 660 NET_EPOCH_ASSERT(); 661 662 #ifdef INET6 663 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 664 #endif 665 666 off0 = *offp; 667 m = *mp; 668 *mp = NULL; 669 to.to_flags = 0; 670 TCPSTAT_INC(tcps_rcvtotal); 671 672 #ifdef INET6 673 if (isipv6) { 674 ip6 = mtod(m, struct ip6_hdr *); 675 th = (struct tcphdr *)((caddr_t)ip6 + off0); 676 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 677 if (port) 678 goto skip6_csum; 679 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 680 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 681 th->th_sum = m->m_pkthdr.csum_data; 682 else 683 th->th_sum = in6_cksum_pseudo(ip6, tlen, 684 IPPROTO_TCP, m->m_pkthdr.csum_data); 685 th->th_sum ^= 0xffff; 686 } else 687 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 688 if (th->th_sum) { 689 TCPSTAT_INC(tcps_rcvbadsum); 690 goto drop; 691 } 692 skip6_csum: 693 /* 694 * Be proactive about unspecified IPv6 address in source. 695 * As we use all-zero to indicate unbounded/unconnected pcb, 696 * unspecified IPv6 address can be used to confuse us. 697 * 698 * Note that packets with unspecified IPv6 destination is 699 * already dropped in ip6_input. 700 */ 701 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 702 /* XXX stat */ 703 goto drop; 704 } 705 iptos = IPV6_TRAFFIC_CLASS(ip6); 706 } 707 #endif 708 #if defined(INET) && defined(INET6) 709 else 710 #endif 711 #ifdef INET 712 { 713 /* 714 * Get IP and TCP header together in first mbuf. 715 * Note: IP leaves IP header in first mbuf. 716 */ 717 if (off0 > sizeof (struct ip)) { 718 ip_stripoptions(m); 719 off0 = sizeof(struct ip); 720 } 721 if (m->m_len < sizeof (struct tcpiphdr)) { 722 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 723 == NULL) { 724 TCPSTAT_INC(tcps_rcvshort); 725 return (IPPROTO_DONE); 726 } 727 } 728 ip = mtod(m, struct ip *); 729 th = (struct tcphdr *)((caddr_t)ip + off0); 730 tlen = ntohs(ip->ip_len) - off0; 731 732 iptos = ip->ip_tos; 733 if (port) 734 goto skip_csum; 735 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 736 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 737 th->th_sum = m->m_pkthdr.csum_data; 738 else 739 th->th_sum = in_pseudo(ip->ip_src.s_addr, 740 ip->ip_dst.s_addr, 741 htonl(m->m_pkthdr.csum_data + tlen + 742 IPPROTO_TCP)); 743 th->th_sum ^= 0xffff; 744 } else { 745 struct ipovly *ipov = (struct ipovly *)ip; 746 747 /* 748 * Checksum extended TCP header and data. 749 */ 750 len = off0 + tlen; 751 ipttl = ip->ip_ttl; 752 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 753 ipov->ih_len = htons(tlen); 754 th->th_sum = in_cksum(m, len); 755 /* Reset length for SDT probes. */ 756 ip->ip_len = htons(len); 757 /* Reset TOS bits */ 758 ip->ip_tos = iptos; 759 /* Re-initialization for later version check */ 760 ip->ip_ttl = ipttl; 761 ip->ip_v = IPVERSION; 762 ip->ip_hl = off0 >> 2; 763 } 764 skip_csum: 765 if (th->th_sum && (port == 0)) { 766 TCPSTAT_INC(tcps_rcvbadsum); 767 goto drop; 768 } 769 } 770 #endif /* INET */ 771 772 /* 773 * Check that TCP offset makes sense, 774 * pull out TCP options and adjust length. XXX 775 */ 776 off = th->th_off << 2; 777 if (off < sizeof (struct tcphdr) || off > tlen) { 778 TCPSTAT_INC(tcps_rcvbadoff); 779 goto drop; 780 } 781 tlen -= off; /* tlen is used instead of ti->ti_len */ 782 if (off > sizeof (struct tcphdr)) { 783 #ifdef INET6 784 if (isipv6) { 785 if (m->m_len < off0 + off) { 786 m = m_pullup(m, off0 + off); 787 if (m == NULL) { 788 TCPSTAT_INC(tcps_rcvshort); 789 return (IPPROTO_DONE); 790 } 791 } 792 ip6 = mtod(m, struct ip6_hdr *); 793 th = (struct tcphdr *)((caddr_t)ip6 + off0); 794 } 795 #endif 796 #if defined(INET) && defined(INET6) 797 else 798 #endif 799 #ifdef INET 800 { 801 if (m->m_len < sizeof(struct ip) + off) { 802 if ((m = m_pullup(m, sizeof (struct ip) + off)) 803 == NULL) { 804 TCPSTAT_INC(tcps_rcvshort); 805 return (IPPROTO_DONE); 806 } 807 ip = mtod(m, struct ip *); 808 th = (struct tcphdr *)((caddr_t)ip + off0); 809 } 810 } 811 #endif 812 optlen = off - sizeof (struct tcphdr); 813 optp = (u_char *)(th + 1); 814 } 815 thflags = tcp_get_flags(th); 816 817 /* 818 * Convert TCP protocol specific fields to host format. 819 */ 820 tcp_fields_to_host(th); 821 822 /* 823 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 824 */ 825 drop_hdrlen = off0 + off; 826 827 /* 828 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 829 */ 830 if ( 831 #ifdef INET6 832 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 833 #ifdef INET 834 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 835 #endif 836 #endif 837 #if defined(INET) && !defined(INET6) 838 (m->m_flags & M_IP_NEXTHOP) 839 #endif 840 ) 841 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 842 843 /* 844 * For initial SYN packets we don't need write lock on matching 845 * PCB, be it a listening one or a synchronized one. The packet 846 * shall not modify its state. 847 */ 848 lookupflag = INPLOOKUP_WILDCARD | 849 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? 850 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); 851 findpcb: 852 #ifdef INET6 853 if (isipv6 && fwd_tag != NULL) { 854 struct sockaddr_in6 *next_hop6; 855 856 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 857 /* 858 * Transparently forwarded. Pretend to be the destination. 859 * Already got one like this? 860 */ 861 inp = in6_pcblookup_mbuf(&V_tcbinfo, 862 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 863 lookupflag, m->m_pkthdr.rcvif, m); 864 if (!inp) { 865 /* 866 * It's new. Try to find the ambushing socket. 867 * Because we've rewritten the destination address, 868 * any hardware-generated hash is ignored. 869 */ 870 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 871 th->th_sport, &next_hop6->sin6_addr, 872 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 873 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 874 } 875 } else if (isipv6) { 876 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 877 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, 878 m->m_pkthdr.rcvif, m); 879 } 880 #endif /* INET6 */ 881 #if defined(INET6) && defined(INET) 882 else 883 #endif 884 #ifdef INET 885 if (fwd_tag != NULL) { 886 struct sockaddr_in *next_hop; 887 888 next_hop = (struct sockaddr_in *)(fwd_tag+1); 889 /* 890 * Transparently forwarded. Pretend to be the destination. 891 * already got one like this? 892 */ 893 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 894 ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); 895 if (!inp) { 896 /* 897 * It's new. Try to find the ambushing socket. 898 * Because we've rewritten the destination address, 899 * any hardware-generated hash is ignored. 900 */ 901 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 902 th->th_sport, next_hop->sin_addr, 903 next_hop->sin_port ? ntohs(next_hop->sin_port) : 904 th->th_dport, lookupflag, m->m_pkthdr.rcvif); 905 } 906 } else 907 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 908 th->th_sport, ip->ip_dst, th->th_dport, lookupflag, 909 m->m_pkthdr.rcvif, m); 910 #endif /* INET */ 911 912 /* 913 * If the INPCB does not exist then all data in the incoming 914 * segment is discarded and an appropriate RST is sent back. 915 * XXX MRT Send RST using which routing table? 916 */ 917 if (inp == NULL) { 918 if (rstreason != 0) { 919 /* We came here after second (safety) lookup. */ 920 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); 921 goto dropwithreset; 922 } 923 /* 924 * Log communication attempts to ports that are not 925 * in use. 926 */ 927 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 928 V_tcp_log_in_vain == 2) { 929 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 930 log(LOG_INFO, "%s; %s: Connection attempt " 931 "to closed port\n", s, __func__); 932 } 933 /* 934 * When blackholing do not respond with a RST but 935 * completely ignore the segment and drop it. 936 */ 937 if (((V_blackhole == 1 && (thflags & TH_SYN)) || 938 V_blackhole == 2) && (V_blackhole_local || ( 939 #ifdef INET6 940 isipv6 ? !in6_localaddr(&ip6->ip6_src) : 941 #endif 942 #ifdef INET 943 !in_localip(ip->ip_src) 944 #else 945 true 946 #endif 947 ))) 948 goto dropunlock; 949 950 rstreason = BANDLIM_RST_CLOSEDPORT; 951 goto dropwithreset; 952 } 953 INP_LOCK_ASSERT(inp); 954 /* 955 * While waiting for inp lock during the lookup, another thread 956 * can have dropped the inpcb, in which case we need to loop back 957 * and try to find a new inpcb to deliver to. 958 */ 959 if (inp->inp_flags & INP_DROPPED) { 960 INP_UNLOCK(inp); 961 inp = NULL; 962 goto findpcb; 963 } 964 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 965 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 966 ((inp->inp_socket == NULL) || !SOLISTENING(inp->inp_socket))) { 967 inp->inp_flowid = m->m_pkthdr.flowid; 968 inp->inp_flowtype = M_HASHTYPE_GET(m); 969 } 970 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 971 #ifdef INET6 972 if (isipv6 && IPSEC_ENABLED(ipv6) && 973 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 974 goto dropunlock; 975 } 976 #ifdef INET 977 else 978 #endif 979 #endif /* INET6 */ 980 #ifdef INET 981 if (IPSEC_ENABLED(ipv4) && 982 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 983 goto dropunlock; 984 } 985 #endif /* INET */ 986 #endif /* IPSEC */ 987 988 /* 989 * Check the minimum TTL for socket. 990 */ 991 if (inp->inp_ip_minttl != 0) { 992 #ifdef INET6 993 if (isipv6) { 994 if (inp->inp_ip_minttl > ip6->ip6_hlim) 995 goto dropunlock; 996 } else 997 #endif 998 if (inp->inp_ip_minttl > ip->ip_ttl) 999 goto dropunlock; 1000 } 1001 1002 tp = intotcpcb(inp); 1003 switch (tp->t_state) { 1004 case TCPS_TIME_WAIT: 1005 /* 1006 * A previous connection in TIMEWAIT state is supposed to catch 1007 * stray or duplicate segments arriving late. If this segment 1008 * was a legitimate new connection attempt, the old INPCB gets 1009 * removed and we can try again to find a listening socket. 1010 */ 1011 tcp_dooptions(&to, optp, optlen, 1012 (thflags & TH_SYN) ? TO_SYN : 0); 1013 /* 1014 * tcp_twcheck unlocks the inp always, and frees the m if fails. 1015 */ 1016 if (tcp_twcheck(inp, &to, th, m, tlen)) 1017 goto findpcb; 1018 return (IPPROTO_DONE); 1019 case TCPS_CLOSED: 1020 /* 1021 * The TCPCB may no longer exist if the connection is winding 1022 * down or it is in the CLOSED state. Either way we drop the 1023 * segment and send an appropriate response. 1024 */ 1025 rstreason = BANDLIM_RST_CLOSEDPORT; 1026 goto dropwithreset; 1027 } 1028 1029 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { 1030 rstreason = BANDLIM_RST_CLOSEDPORT; 1031 goto dropwithreset; 1032 } 1033 1034 #ifdef TCP_OFFLOAD 1035 if (tp->t_flags & TF_TOE) { 1036 tcp_offload_input(tp, m); 1037 m = NULL; /* consumed by the TOE driver */ 1038 goto dropunlock; 1039 } 1040 #endif 1041 1042 #ifdef MAC 1043 if (mac_inpcb_check_deliver(inp, m)) 1044 goto dropunlock; 1045 #endif 1046 so = inp->inp_socket; 1047 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1048 #ifdef TCPDEBUG 1049 if (so->so_options & SO_DEBUG) { 1050 ostate = tp->t_state; 1051 #ifdef INET6 1052 if (isipv6) { 1053 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1054 } else 1055 #endif 1056 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1057 tcp_savetcp = *th; 1058 } 1059 #endif /* TCPDEBUG */ 1060 /* 1061 * When the socket is accepting connections (the INPCB is in LISTEN 1062 * state) we look into the SYN cache if this is a new connection 1063 * attempt or the completion of a previous one. 1064 */ 1065 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), 1066 ("%s: so accepting but tp %p not listening", __func__, tp)); 1067 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { 1068 struct in_conninfo inc; 1069 1070 bzero(&inc, sizeof(inc)); 1071 #ifdef INET6 1072 if (isipv6) { 1073 inc.inc_flags |= INC_ISIPV6; 1074 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1075 inc.inc_flags |= INC_IPV6MINMTU; 1076 inc.inc6_faddr = ip6->ip6_src; 1077 inc.inc6_laddr = ip6->ip6_dst; 1078 } else 1079 #endif 1080 { 1081 inc.inc_faddr = ip->ip_src; 1082 inc.inc_laddr = ip->ip_dst; 1083 } 1084 inc.inc_fport = th->th_sport; 1085 inc.inc_lport = th->th_dport; 1086 inc.inc_fibnum = so->so_fibnum; 1087 1088 /* 1089 * Check for an existing connection attempt in syncache if 1090 * the flag is only ACK. A successful lookup creates a new 1091 * socket appended to the listen queue in SYN_RECEIVED state. 1092 */ 1093 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1094 /* 1095 * Parse the TCP options here because 1096 * syncookies need access to the reflected 1097 * timestamp. 1098 */ 1099 tcp_dooptions(&to, optp, optlen, 0); 1100 /* 1101 * NB: syncache_expand() doesn't unlock inp. 1102 */ 1103 rstreason = syncache_expand(&inc, &to, th, &so, m, port); 1104 if (rstreason < 0) { 1105 /* 1106 * A failing TCP MD5 signature comparison 1107 * must result in the segment being dropped 1108 * and must not produce any response back 1109 * to the sender. 1110 */ 1111 goto dropunlock; 1112 } else if (rstreason == 0) { 1113 /* 1114 * No syncache entry, or ACK was not for our 1115 * SYN/ACK. Do our protection against double 1116 * ACK. If peer sent us 2 ACKs, then for the 1117 * first one syncache_expand() successfully 1118 * converted syncache entry into a socket, 1119 * while we were waiting on the inpcb lock. We 1120 * don't want to sent RST for the second ACK, 1121 * so we perform second lookup without wildcard 1122 * match, hoping to find the new socket. If 1123 * the ACK is stray indeed, rstreason would 1124 * hint the above code that the lookup was a 1125 * second attempt. 1126 * 1127 * NB: syncache did its own logging 1128 * of the failure cause. 1129 */ 1130 INP_WUNLOCK(inp); 1131 rstreason = BANDLIM_RST_OPENPORT; 1132 lookupflag &= ~INPLOOKUP_WILDCARD; 1133 goto findpcb; 1134 } 1135 tfo_socket_result: 1136 if (so == NULL) { 1137 /* 1138 * We completed the 3-way handshake 1139 * but could not allocate a socket 1140 * either due to memory shortage, 1141 * listen queue length limits or 1142 * global socket limits. Send RST 1143 * or wait and have the remote end 1144 * retransmit the ACK for another 1145 * try. 1146 */ 1147 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1148 log(LOG_DEBUG, "%s; %s: Listen socket: " 1149 "Socket allocation failed due to " 1150 "limits or memory shortage, %s\n", 1151 s, __func__, 1152 V_tcp_sc_rst_sock_fail ? 1153 "sending RST" : "try again"); 1154 if (V_tcp_sc_rst_sock_fail) { 1155 rstreason = BANDLIM_UNLIMITED; 1156 goto dropwithreset; 1157 } else 1158 goto dropunlock; 1159 } 1160 /* 1161 * Socket is created in state SYN_RECEIVED. 1162 * Unlock the listen socket, lock the newly 1163 * created socket and update the tp variable. 1164 * If we came here via jump to tfo_socket_result, 1165 * then listening socket is read-locked. 1166 */ 1167 INP_UNLOCK(inp); /* listen socket */ 1168 inp = sotoinpcb(so); 1169 /* 1170 * New connection inpcb is already locked by 1171 * syncache_expand(). 1172 */ 1173 INP_WLOCK_ASSERT(inp); 1174 tp = intotcpcb(inp); 1175 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1176 ("%s: ", __func__)); 1177 /* 1178 * Process the segment and the data it 1179 * contains. tcp_do_segment() consumes 1180 * the mbuf chain and unlocks the inpcb. 1181 */ 1182 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1183 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1184 iptos); 1185 return (IPPROTO_DONE); 1186 } 1187 /* 1188 * Segment flag validation for new connection attempts: 1189 * 1190 * Our (SYN|ACK) response was rejected. 1191 * Check with syncache and remove entry to prevent 1192 * retransmits. 1193 * 1194 * NB: syncache_chkrst does its own logging of failure 1195 * causes. 1196 */ 1197 if (thflags & TH_RST) { 1198 syncache_chkrst(&inc, th, m, port); 1199 goto dropunlock; 1200 } 1201 /* 1202 * We can't do anything without SYN. 1203 */ 1204 if ((thflags & TH_SYN) == 0) { 1205 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1206 log(LOG_DEBUG, "%s; %s: Listen socket: " 1207 "SYN is missing, segment ignored\n", 1208 s, __func__); 1209 TCPSTAT_INC(tcps_badsyn); 1210 goto dropunlock; 1211 } 1212 /* 1213 * (SYN|ACK) is bogus on a listen socket. 1214 */ 1215 if (thflags & TH_ACK) { 1216 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1217 log(LOG_DEBUG, "%s; %s: Listen socket: " 1218 "SYN|ACK invalid, segment rejected\n", 1219 s, __func__); 1220 syncache_badack(&inc, port); /* XXX: Not needed! */ 1221 TCPSTAT_INC(tcps_badsyn); 1222 rstreason = BANDLIM_RST_OPENPORT; 1223 goto dropwithreset; 1224 } 1225 /* 1226 * If the drop_synfin option is enabled, drop all 1227 * segments with both the SYN and FIN bits set. 1228 * This prevents e.g. nmap from identifying the 1229 * TCP/IP stack. 1230 * XXX: Poor reasoning. nmap has other methods 1231 * and is constantly refining its stack detection 1232 * strategies. 1233 * XXX: This is a violation of the TCP specification 1234 * and was used by RFC1644. 1235 */ 1236 if ((thflags & TH_FIN) && V_drop_synfin) { 1237 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1238 log(LOG_DEBUG, "%s; %s: Listen socket: " 1239 "SYN|FIN segment ignored (based on " 1240 "sysctl setting)\n", s, __func__); 1241 TCPSTAT_INC(tcps_badsyn); 1242 goto dropunlock; 1243 } 1244 /* 1245 * Segment's flags are (SYN) or (SYN|FIN). 1246 * 1247 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1248 * as they do not affect the state of the TCP FSM. 1249 * The data pointed to by TH_URG and th_urp is ignored. 1250 */ 1251 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1252 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1253 KASSERT(thflags & (TH_SYN), 1254 ("%s: Listen socket: TH_SYN not set", __func__)); 1255 INP_RLOCK_ASSERT(inp); 1256 #ifdef INET6 1257 /* 1258 * If deprecated address is forbidden, 1259 * we do not accept SYN to deprecated interface 1260 * address to prevent any new inbound connection from 1261 * getting established. 1262 * When we do not accept SYN, we send a TCP RST, 1263 * with deprecated source address (instead of dropping 1264 * it). We compromise it as it is much better for peer 1265 * to send a RST, and RST will be the final packet 1266 * for the exchange. 1267 * 1268 * If we do not forbid deprecated addresses, we accept 1269 * the SYN packet. RFC2462 does not suggest dropping 1270 * SYN in this case. 1271 * If we decipher RFC2462 5.5.4, it says like this: 1272 * 1. use of deprecated addr with existing 1273 * communication is okay - "SHOULD continue to be 1274 * used" 1275 * 2. use of it with new communication: 1276 * (2a) "SHOULD NOT be used if alternate address 1277 * with sufficient scope is available" 1278 * (2b) nothing mentioned otherwise. 1279 * Here we fall into (2b) case as we have no choice in 1280 * our source address selection - we must obey the peer. 1281 * 1282 * The wording in RFC2462 is confusing, and there are 1283 * multiple description text for deprecated address 1284 * handling - worse, they are not exactly the same. 1285 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1286 */ 1287 if (isipv6 && !V_ip6_use_deprecated) { 1288 struct in6_ifaddr *ia6; 1289 1290 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); 1291 if (ia6 != NULL && 1292 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1293 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1294 log(LOG_DEBUG, "%s; %s: Listen socket: " 1295 "Connection attempt to deprecated " 1296 "IPv6 address rejected\n", 1297 s, __func__); 1298 rstreason = BANDLIM_RST_OPENPORT; 1299 goto dropwithreset; 1300 } 1301 } 1302 #endif /* INET6 */ 1303 /* 1304 * Basic sanity checks on incoming SYN requests: 1305 * Don't respond if the destination is a link layer 1306 * broadcast according to RFC1122 4.2.3.10, p. 104. 1307 * If it is from this socket it must be forged. 1308 * Don't respond if the source or destination is a 1309 * global or subnet broad- or multicast address. 1310 * Note that it is quite possible to receive unicast 1311 * link-layer packets with a broadcast IP address. Use 1312 * in_broadcast() to find them. 1313 */ 1314 if (m->m_flags & (M_BCAST|M_MCAST)) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from broad- or multicast " 1318 "link layer address ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 #ifdef INET6 1322 if (isipv6) { 1323 if (th->th_dport == th->th_sport && 1324 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1326 log(LOG_DEBUG, "%s; %s: Listen socket: " 1327 "Connection attempt to/from self " 1328 "ignored\n", s, __func__); 1329 goto dropunlock; 1330 } 1331 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1332 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1333 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1334 log(LOG_DEBUG, "%s; %s: Listen socket: " 1335 "Connection attempt from/to multicast " 1336 "address ignored\n", s, __func__); 1337 goto dropunlock; 1338 } 1339 } 1340 #endif 1341 #if defined(INET) && defined(INET6) 1342 else 1343 #endif 1344 #ifdef INET 1345 { 1346 if (th->th_dport == th->th_sport && 1347 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1348 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1349 log(LOG_DEBUG, "%s; %s: Listen socket: " 1350 "Connection attempt from/to self " 1351 "ignored\n", s, __func__); 1352 goto dropunlock; 1353 } 1354 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1355 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1356 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1357 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1358 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1359 log(LOG_DEBUG, "%s; %s: Listen socket: " 1360 "Connection attempt from/to broad- " 1361 "or multicast address ignored\n", 1362 s, __func__); 1363 goto dropunlock; 1364 } 1365 } 1366 #endif 1367 /* 1368 * SYN appears to be valid. Create compressed TCP state 1369 * for syncache. 1370 */ 1371 #ifdef TCPDEBUG 1372 if (so->so_options & SO_DEBUG) 1373 tcp_trace(TA_INPUT, ostate, tp, 1374 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1375 #endif 1376 TCP_PROBE3(debug__input, tp, th, m); 1377 tcp_dooptions(&to, optp, optlen, TO_SYN); 1378 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, 1379 iptos, port)) != NULL) 1380 goto tfo_socket_result; 1381 1382 /* 1383 * Entry added to syncache and mbuf consumed. 1384 * Only the listen socket is unlocked by syncache_add(). 1385 */ 1386 return (IPPROTO_DONE); 1387 } else if (tp->t_state == TCPS_LISTEN) { 1388 /* 1389 * When a listen socket is torn down the SO_ACCEPTCONN 1390 * flag is removed first while connections are drained 1391 * from the accept queue in a unlock/lock cycle of the 1392 * ACCEPT_LOCK, opening a race condition allowing a SYN 1393 * attempt go through unhandled. 1394 */ 1395 goto dropunlock; 1396 } 1397 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1398 if (tp->t_flags & TF_SIGNATURE) { 1399 tcp_dooptions(&to, optp, optlen, thflags); 1400 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1401 TCPSTAT_INC(tcps_sig_err_nosigopt); 1402 goto dropunlock; 1403 } 1404 if (!TCPMD5_ENABLED() || 1405 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1406 goto dropunlock; 1407 } 1408 #endif 1409 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1410 1411 /* 1412 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1413 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1414 * the inpcb, and unlocks pcbinfo. 1415 * 1416 * XXXGL: in case of a pure SYN arriving on existing connection 1417 * TCP stacks won't need to modify the PCB, they would either drop 1418 * the segment silently, or send a challenge ACK. However, we try 1419 * to upgrade the lock, because calling convention for stacks is 1420 * write-lock on PCB. If upgrade fails, drop the SYN. 1421 */ 1422 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) 1423 goto dropunlock; 1424 1425 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1426 return (IPPROTO_DONE); 1427 1428 dropwithreset: 1429 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1430 1431 if (inp != NULL) { 1432 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1433 INP_UNLOCK(inp); 1434 } else 1435 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1436 m = NULL; /* mbuf chain got consumed. */ 1437 goto drop; 1438 1439 dropunlock: 1440 if (m != NULL) 1441 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1442 1443 if (inp != NULL) 1444 INP_UNLOCK(inp); 1445 1446 drop: 1447 if (s != NULL) 1448 free(s, M_TCPLOG); 1449 if (m != NULL) 1450 m_freem(m); 1451 return (IPPROTO_DONE); 1452 } 1453 1454 /* 1455 * Automatic sizing of receive socket buffer. Often the send 1456 * buffer size is not optimally adjusted to the actual network 1457 * conditions at hand (delay bandwidth product). Setting the 1458 * buffer size too small limits throughput on links with high 1459 * bandwidth and high delay (eg. trans-continental/oceanic links). 1460 * 1461 * On the receive side the socket buffer memory is only rarely 1462 * used to any significant extent. This allows us to be much 1463 * more aggressive in scaling the receive socket buffer. For 1464 * the case that the buffer space is actually used to a large 1465 * extent and we run out of kernel memory we can simply drop 1466 * the new segments; TCP on the sender will just retransmit it 1467 * later. Setting the buffer size too big may only consume too 1468 * much kernel memory if the application doesn't read() from 1469 * the socket or packet loss or reordering makes use of the 1470 * reassembly queue. 1471 * 1472 * The criteria to step up the receive buffer one notch are: 1473 * 1. Application has not set receive buffer size with 1474 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1475 * 2. the number of bytes received during 1/2 of an sRTT 1476 * is at least 3/8 of the current socket buffer size. 1477 * 3. receive buffer size has not hit maximal automatic size; 1478 * 1479 * If all of the criteria are met we increaset the socket buffer 1480 * by a 1/2 (bounded by the max). This allows us to keep ahead 1481 * of slow-start but also makes it so our peer never gets limited 1482 * by our rwnd which we then open up causing a burst. 1483 * 1484 * This algorithm does two steps per RTT at most and only if 1485 * we receive a bulk stream w/o packet losses or reorderings. 1486 * Shrinking the buffer during idle times is not necessary as 1487 * it doesn't consume any memory when idle. 1488 * 1489 * TODO: Only step up if the application is actually serving 1490 * the buffer to better manage the socket buffer resources. 1491 */ 1492 int 1493 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1494 struct tcpcb *tp, int tlen) 1495 { 1496 int newsize = 0; 1497 1498 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1499 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1500 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1501 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1502 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1503 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1504 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1505 } 1506 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1507 1508 /* Start over with next RTT. */ 1509 tp->rfbuf_ts = 0; 1510 tp->rfbuf_cnt = 0; 1511 } else { 1512 tp->rfbuf_cnt += tlen; /* add up */ 1513 } 1514 return (newsize); 1515 } 1516 1517 int 1518 tcp_input(struct mbuf **mp, int *offp, int proto) 1519 { 1520 return(tcp_input_with_port(mp, offp, proto, 0)); 1521 } 1522 1523 void 1524 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1525 { 1526 /* 1527 * Since tp might be gone if the session entered 1528 * the TIME_WAIT state before coming here, we need 1529 * to check if the socket is still connected. 1530 */ 1531 if (tp == NULL) { 1532 return; 1533 } 1534 if (so == NULL) { 1535 return; 1536 } 1537 INP_LOCK_ASSERT(tp->t_inpcb); 1538 if (tp->t_flags & TF_WAKESOR) { 1539 tp->t_flags &= ~TF_WAKESOR; 1540 SOCKBUF_LOCK_ASSERT(&so->so_rcv); 1541 sorwakeup_locked(so); 1542 } 1543 } 1544 1545 void 1546 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1547 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1548 { 1549 uint16_t thflags; 1550 int acked, ourfinisacked, needoutput = 0, sack_changed; 1551 int rstreason, todrop, win, incforsyn = 0; 1552 uint32_t tiwin; 1553 uint16_t nsegs; 1554 char *s; 1555 struct in_conninfo *inc; 1556 struct mbuf *mfree; 1557 struct tcpopt to; 1558 int tfo_syn; 1559 u_int maxseg; 1560 1561 #ifdef TCPDEBUG 1562 /* 1563 * The size of tcp_saveipgen must be the size of the max ip header, 1564 * now IPv6. 1565 */ 1566 u_char tcp_saveipgen[IP6_HDR_LEN]; 1567 struct tcphdr tcp_savetcp; 1568 short ostate = 0; 1569 #endif 1570 thflags = tcp_get_flags(th); 1571 inc = &tp->t_inpcb->inp_inc; 1572 tp->sackhint.last_sack_ack = 0; 1573 sack_changed = 0; 1574 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1575 1576 NET_EPOCH_ASSERT(); 1577 INP_WLOCK_ASSERT(tp->t_inpcb); 1578 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1579 __func__)); 1580 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1581 __func__)); 1582 1583 #ifdef TCPPCAP 1584 /* Save segment, if requested. */ 1585 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1586 #endif 1587 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1588 tlen, NULL, true); 1589 1590 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1591 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1592 log(LOG_DEBUG, "%s; %s: " 1593 "SYN|FIN segment ignored (based on " 1594 "sysctl setting)\n", s, __func__); 1595 free(s, M_TCPLOG); 1596 } 1597 goto drop; 1598 } 1599 1600 /* 1601 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1602 * check SEQ.ACK first. 1603 */ 1604 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1605 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1606 rstreason = BANDLIM_UNLIMITED; 1607 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 1608 goto dropwithreset; 1609 } 1610 1611 /* 1612 * Segment received on connection. 1613 * Reset idle time and keep-alive timer. 1614 * XXX: This should be done after segment 1615 * validation to ignore broken/spoofed segs. 1616 */ 1617 if (tp->t_idle_reduce && 1618 (tp->snd_max == tp->snd_una) && 1619 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) 1620 cc_after_idle(tp); 1621 tp->t_rcvtime = ticks; 1622 1623 if (thflags & TH_FIN) 1624 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); 1625 /* 1626 * Scale up the window into a 32-bit value. 1627 * For the SYN_SENT state the scale is zero. 1628 */ 1629 tiwin = th->th_win << tp->snd_scale; 1630 #ifdef STATS 1631 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1632 #endif 1633 1634 /* 1635 * TCP ECN processing. 1636 */ 1637 if (tcp_ecn_input_segment(tp, thflags, iptos)) 1638 cc_cong_signal(tp, th, CC_ECN); 1639 1640 /* 1641 * Parse options on any incoming segment. 1642 */ 1643 tcp_dooptions(&to, (u_char *)(th + 1), 1644 (th->th_off << 2) - sizeof(struct tcphdr), 1645 (thflags & TH_SYN) ? TO_SYN : 0); 1646 1647 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1648 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1649 (to.to_flags & TOF_SIGNATURE) == 0) { 1650 TCPSTAT_INC(tcps_sig_err_sigopt); 1651 /* XXX: should drop? */ 1652 } 1653 #endif 1654 /* 1655 * If echoed timestamp is later than the current time, 1656 * fall back to non RFC1323 RTT calculation. Normalize 1657 * timestamp if syncookies were used when this connection 1658 * was established. 1659 */ 1660 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1661 to.to_tsecr -= tp->ts_offset; 1662 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1663 to.to_tsecr = 0; 1664 else if (tp->t_rxtshift == 1 && 1665 tp->t_flags & TF_PREVVALID && 1666 tp->t_badrxtwin != 0 && 1667 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 1668 cc_cong_signal(tp, th, CC_RTO_ERR); 1669 } 1670 /* 1671 * Process options only when we get SYN/ACK back. The SYN case 1672 * for incoming connections is handled in tcp_syncache. 1673 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1674 * or <SYN,ACK>) segment itself is never scaled. 1675 * XXX this is traditional behavior, may need to be cleaned up. 1676 */ 1677 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1678 /* Handle parallel SYN for ECN */ 1679 tcp_ecn_input_parallel_syn(tp, thflags, iptos); 1680 if ((to.to_flags & TOF_SCALE) && 1681 (tp->t_flags & TF_REQ_SCALE) && 1682 !(tp->t_flags & TF_NOOPT)) { 1683 tp->t_flags |= TF_RCVD_SCALE; 1684 tp->snd_scale = to.to_wscale; 1685 } else 1686 tp->t_flags &= ~TF_REQ_SCALE; 1687 /* 1688 * Initial send window. It will be updated with 1689 * the next incoming segment to the scaled value. 1690 */ 1691 tp->snd_wnd = th->th_win; 1692 if ((to.to_flags & TOF_TS) && 1693 (tp->t_flags & TF_REQ_TSTMP) && 1694 !(tp->t_flags & TF_NOOPT)) { 1695 tp->t_flags |= TF_RCVD_TSTMP; 1696 tp->ts_recent = to.to_tsval; 1697 tp->ts_recent_age = tcp_ts_getticks(); 1698 } else 1699 tp->t_flags &= ~TF_REQ_TSTMP; 1700 if (to.to_flags & TOF_MSS) 1701 tcp_mss(tp, to.to_mss); 1702 if ((tp->t_flags & TF_SACK_PERMIT) && 1703 (!(to.to_flags & TOF_SACKPERM) || 1704 (tp->t_flags & TF_NOOPT))) 1705 tp->t_flags &= ~TF_SACK_PERMIT; 1706 if (IS_FASTOPEN(tp->t_flags)) { 1707 if ((to.to_flags & TOF_FASTOPEN) && 1708 !(tp->t_flags & TF_NOOPT)) { 1709 uint16_t mss; 1710 1711 if (to.to_flags & TOF_MSS) 1712 mss = to.to_mss; 1713 else 1714 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1715 mss = TCP6_MSS; 1716 else 1717 mss = TCP_MSS; 1718 tcp_fastopen_update_cache(tp, mss, 1719 to.to_tfo_len, to.to_tfo_cookie); 1720 } else 1721 tcp_fastopen_disable_path(tp); 1722 } 1723 } 1724 1725 /* 1726 * If timestamps were negotiated during SYN/ACK and a 1727 * segment without a timestamp is received, silently drop 1728 * the segment, unless it is a RST segment or missing timestamps are 1729 * tolerated. 1730 * See section 3.2 of RFC 7323. 1731 */ 1732 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1733 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { 1734 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1735 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1736 "segment processed normally\n", 1737 s, __func__); 1738 free(s, M_TCPLOG); 1739 } 1740 } else { 1741 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1742 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1743 "segment silently dropped\n", s, __func__); 1744 free(s, M_TCPLOG); 1745 } 1746 goto drop; 1747 } 1748 } 1749 /* 1750 * If timestamps were not negotiated during SYN/ACK and a 1751 * segment with a timestamp is received, ignore the 1752 * timestamp and process the packet normally. 1753 * See section 3.2 of RFC 7323. 1754 */ 1755 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1756 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1757 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1758 "segment processed normally\n", s, __func__); 1759 free(s, M_TCPLOG); 1760 } 1761 } 1762 1763 /* 1764 * Header prediction: check for the two common cases 1765 * of a uni-directional data xfer. If the packet has 1766 * no control flags, is in-sequence, the window didn't 1767 * change and we're not retransmitting, it's a 1768 * candidate. If the length is zero and the ack moved 1769 * forward, we're the sender side of the xfer. Just 1770 * free the data acked & wake any higher level process 1771 * that was blocked waiting for space. If the length 1772 * is non-zero and the ack didn't move, we're the 1773 * receiver side. If we're getting packets in-order 1774 * (the reassembly queue is empty), add the data to 1775 * the socket buffer and note that we need a delayed ack. 1776 * Make sure that the hidden state-flags are also off. 1777 * Since we check for TCPS_ESTABLISHED first, it can only 1778 * be TH_NEEDSYN. 1779 */ 1780 if (tp->t_state == TCPS_ESTABLISHED && 1781 th->th_seq == tp->rcv_nxt && 1782 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1783 tp->snd_nxt == tp->snd_max && 1784 tiwin && tiwin == tp->snd_wnd && 1785 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1786 SEGQ_EMPTY(tp) && 1787 ((to.to_flags & TOF_TS) == 0 || 1788 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1789 /* 1790 * If last ACK falls within this segment's sequence numbers, 1791 * record the timestamp. 1792 * NOTE that the test is modified according to the latest 1793 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1794 */ 1795 if ((to.to_flags & TOF_TS) != 0 && 1796 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1797 tp->ts_recent_age = tcp_ts_getticks(); 1798 tp->ts_recent = to.to_tsval; 1799 } 1800 1801 if (tlen == 0) { 1802 if (SEQ_GT(th->th_ack, tp->snd_una) && 1803 SEQ_LEQ(th->th_ack, tp->snd_max) && 1804 !IN_RECOVERY(tp->t_flags) && 1805 (to.to_flags & TOF_SACK) == 0 && 1806 TAILQ_EMPTY(&tp->snd_holes)) { 1807 /* 1808 * This is a pure ack for outstanding data. 1809 */ 1810 TCPSTAT_INC(tcps_predack); 1811 1812 /* 1813 * "bad retransmit" recovery without timestamps. 1814 */ 1815 if ((to.to_flags & TOF_TS) == 0 && 1816 tp->t_rxtshift == 1 && 1817 tp->t_flags & TF_PREVVALID && 1818 tp->t_badrxtwin != 0 && 1819 TSTMP_LT(ticks, tp->t_badrxtwin)) { 1820 cc_cong_signal(tp, th, CC_RTO_ERR); 1821 } 1822 1823 /* 1824 * Recalculate the transmit timer / rtt. 1825 * 1826 * Some boxes send broken timestamp replies 1827 * during the SYN+ACK phase, ignore 1828 * timestamps of 0 or we could calculate a 1829 * huge RTT and blow up the retransmit timer. 1830 */ 1831 if ((to.to_flags & TOF_TS) != 0 && 1832 to.to_tsecr) { 1833 uint32_t t; 1834 1835 t = tcp_ts_getticks() - to.to_tsecr; 1836 if (!tp->t_rttlow || tp->t_rttlow > t) 1837 tp->t_rttlow = t; 1838 tcp_xmit_timer(tp, 1839 TCP_TS_TO_TICKS(t) + 1); 1840 } else if (tp->t_rtttime && 1841 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1842 if (!tp->t_rttlow || 1843 tp->t_rttlow > ticks - tp->t_rtttime) 1844 tp->t_rttlow = ticks - tp->t_rtttime; 1845 tcp_xmit_timer(tp, 1846 ticks - tp->t_rtttime); 1847 } 1848 acked = BYTES_THIS_ACK(tp, th); 1849 1850 #ifdef TCP_HHOOK 1851 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1852 hhook_run_tcp_est_in(tp, th, &to); 1853 #endif 1854 1855 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1856 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1857 sbdrop(&so->so_snd, acked); 1858 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1859 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1860 tp->snd_recover = th->th_ack - 1; 1861 1862 /* 1863 * Let the congestion control algorithm update 1864 * congestion control related information. This 1865 * typically means increasing the congestion 1866 * window. 1867 */ 1868 cc_ack_received(tp, th, nsegs, CC_ACK); 1869 1870 tp->snd_una = th->th_ack; 1871 /* 1872 * Pull snd_wl2 up to prevent seq wrap relative 1873 * to th_ack. 1874 */ 1875 tp->snd_wl2 = th->th_ack; 1876 tp->t_dupacks = 0; 1877 m_freem(m); 1878 1879 /* 1880 * If all outstanding data are acked, stop 1881 * retransmit timer, otherwise restart timer 1882 * using current (possibly backed-off) value. 1883 * If process is waiting for space, 1884 * wakeup/selwakeup/signal. If data 1885 * are ready to send, let tcp_output 1886 * decide between more output or persist. 1887 */ 1888 #ifdef TCPDEBUG 1889 if (so->so_options & SO_DEBUG) 1890 tcp_trace(TA_INPUT, ostate, tp, 1891 (void *)tcp_saveipgen, 1892 &tcp_savetcp, 0); 1893 #endif 1894 TCP_PROBE3(debug__input, tp, th, m); 1895 /* 1896 * Clear t_acktime if remote side has ACKd 1897 * all data in the socket buffer. 1898 * Otherwise, update t_acktime if we received 1899 * a sufficiently large ACK. 1900 */ 1901 if (sbavail(&so->so_snd) == 0) 1902 tp->t_acktime = 0; 1903 else if (acked > 1) 1904 tp->t_acktime = ticks; 1905 if (tp->snd_una == tp->snd_max) 1906 tcp_timer_activate(tp, TT_REXMT, 0); 1907 else if (!tcp_timer_active(tp, TT_PERSIST)) 1908 tcp_timer_activate(tp, TT_REXMT, 1909 TP_RXTCUR(tp)); 1910 sowwakeup(so); 1911 if (sbavail(&so->so_snd)) 1912 (void) tcp_output(tp); 1913 goto check_delack; 1914 } 1915 } else if (th->th_ack == tp->snd_una && 1916 tlen <= sbspace(&so->so_rcv)) { 1917 int newsize = 0; /* automatic sockbuf scaling */ 1918 1919 /* 1920 * This is a pure, in-sequence data packet with 1921 * nothing on the reassembly queue and we have enough 1922 * buffer space to take it. 1923 */ 1924 /* Clean receiver SACK report if present */ 1925 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1926 tcp_clean_sackreport(tp); 1927 TCPSTAT_INC(tcps_preddat); 1928 tp->rcv_nxt += tlen; 1929 if (tlen && 1930 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1931 (tp->t_fbyte_in == 0)) { 1932 tp->t_fbyte_in = ticks; 1933 if (tp->t_fbyte_in == 0) 1934 tp->t_fbyte_in = 1; 1935 if (tp->t_fbyte_out && tp->t_fbyte_in) 1936 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1937 } 1938 /* 1939 * Pull snd_wl1 up to prevent seq wrap relative to 1940 * th_seq. 1941 */ 1942 tp->snd_wl1 = th->th_seq; 1943 /* 1944 * Pull rcv_up up to prevent seq wrap relative to 1945 * rcv_nxt. 1946 */ 1947 tp->rcv_up = tp->rcv_nxt; 1948 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1949 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1950 #ifdef TCPDEBUG 1951 if (so->so_options & SO_DEBUG) 1952 tcp_trace(TA_INPUT, ostate, tp, 1953 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1954 #endif 1955 TCP_PROBE3(debug__input, tp, th, m); 1956 1957 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1958 1959 /* Add data to socket buffer. */ 1960 SOCKBUF_LOCK(&so->so_rcv); 1961 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1962 m_freem(m); 1963 } else { 1964 /* 1965 * Set new socket buffer size. 1966 * Give up when limit is reached. 1967 */ 1968 if (newsize) 1969 if (!sbreserve_locked(so, SO_RCV, 1970 newsize, NULL)) 1971 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1972 m_adj(m, drop_hdrlen); /* delayed header drop */ 1973 sbappendstream_locked(&so->so_rcv, m, 0); 1974 } 1975 /* NB: sorwakeup_locked() does an implicit unlock. */ 1976 sorwakeup_locked(so); 1977 if (DELAY_ACK(tp, tlen)) { 1978 tp->t_flags |= TF_DELACK; 1979 } else { 1980 tp->t_flags |= TF_ACKNOW; 1981 tcp_output(tp); 1982 } 1983 goto check_delack; 1984 } 1985 } 1986 1987 /* 1988 * Calculate amount of space in receive window, 1989 * and then do TCP input processing. 1990 * Receive window is amount of space in rcv queue, 1991 * but not less than advertised window. 1992 */ 1993 win = sbspace(&so->so_rcv); 1994 if (win < 0) 1995 win = 0; 1996 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1997 1998 switch (tp->t_state) { 1999 /* 2000 * If the state is SYN_RECEIVED: 2001 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 2002 */ 2003 case TCPS_SYN_RECEIVED: 2004 if ((thflags & TH_ACK) && 2005 (SEQ_LEQ(th->th_ack, tp->snd_una) || 2006 SEQ_GT(th->th_ack, tp->snd_max))) { 2007 rstreason = BANDLIM_RST_OPENPORT; 2008 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2009 goto dropwithreset; 2010 } 2011 if (IS_FASTOPEN(tp->t_flags)) { 2012 /* 2013 * When a TFO connection is in SYN_RECEIVED, the 2014 * only valid packets are the initial SYN, a 2015 * retransmit/copy of the initial SYN (possibly with 2016 * a subset of the original data), a valid ACK, a 2017 * FIN, or a RST. 2018 */ 2019 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 2020 rstreason = BANDLIM_RST_OPENPORT; 2021 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 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 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2054 tp = tcp_drop(tp, ECONNREFUSED); 2055 } 2056 if (thflags & TH_RST) 2057 goto drop; 2058 if (!(thflags & TH_SYN)) 2059 goto drop; 2060 2061 tp->irs = th->th_seq; 2062 tcp_rcvseqinit(tp); 2063 if (thflags & TH_ACK) { 2064 int tfo_partial_ack = 0; 2065 2066 TCPSTAT_INC(tcps_connects); 2067 soisconnected(so); 2068 #ifdef MAC 2069 mac_socketpeer_set_from_mbuf(m, so); 2070 #endif 2071 /* Do window scaling on this connection? */ 2072 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2073 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2074 tp->rcv_scale = tp->request_r_scale; 2075 } 2076 tp->rcv_adv += min(tp->rcv_wnd, 2077 TCP_MAXWIN << tp->rcv_scale); 2078 tp->snd_una++; /* SYN is acked */ 2079 /* 2080 * If not all the data that was sent in the TFO SYN 2081 * has been acked, resend the remainder right away. 2082 */ 2083 if (IS_FASTOPEN(tp->t_flags) && 2084 (tp->snd_una != tp->snd_max)) { 2085 tp->snd_nxt = th->th_ack; 2086 tfo_partial_ack = 1; 2087 } 2088 /* 2089 * If there's data, delay ACK; if there's also a FIN 2090 * ACKNOW will be turned on later. 2091 */ 2092 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2093 tcp_timer_activate(tp, TT_DELACK, 2094 tcp_delacktime); 2095 else 2096 tp->t_flags |= TF_ACKNOW; 2097 2098 tcp_ecn_input_syn_sent(tp, thflags, iptos); 2099 2100 /* 2101 * Received <SYN,ACK> in SYN_SENT[*] state. 2102 * Transitions: 2103 * SYN_SENT --> ESTABLISHED 2104 * SYN_SENT* --> FIN_WAIT_1 2105 */ 2106 tp->t_starttime = ticks; 2107 if (tp->t_flags & TF_NEEDFIN) { 2108 tp->t_acktime = ticks; 2109 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2110 tp->t_flags &= ~TF_NEEDFIN; 2111 thflags &= ~TH_SYN; 2112 } else { 2113 tcp_state_change(tp, TCPS_ESTABLISHED); 2114 TCP_PROBE5(connect__established, NULL, tp, 2115 m, tp, th); 2116 cc_conn_init(tp); 2117 tcp_timer_activate(tp, TT_KEEP, 2118 TP_KEEPIDLE(tp)); 2119 } 2120 } else { 2121 /* 2122 * Received initial SYN in SYN-SENT[*] state => 2123 * simultaneous open. 2124 * If it succeeds, connection is * half-synchronized. 2125 * Otherwise, do 3-way handshake: 2126 * SYN-SENT -> SYN-RECEIVED 2127 * SYN-SENT* -> SYN-RECEIVED* 2128 */ 2129 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); 2130 tcp_timer_activate(tp, TT_REXMT, 0); 2131 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2132 } 2133 2134 INP_WLOCK_ASSERT(tp->t_inpcb); 2135 2136 /* 2137 * Advance th->th_seq to correspond to first data byte. 2138 * If data, trim to stay within window, 2139 * dropping FIN if necessary. 2140 */ 2141 th->th_seq++; 2142 if (tlen > tp->rcv_wnd) { 2143 todrop = tlen - tp->rcv_wnd; 2144 m_adj(m, -todrop); 2145 tlen = tp->rcv_wnd; 2146 thflags &= ~TH_FIN; 2147 TCPSTAT_INC(tcps_rcvpackafterwin); 2148 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2149 } 2150 tp->snd_wl1 = th->th_seq - 1; 2151 tp->rcv_up = th->th_seq; 2152 /* 2153 * Client side of transaction: already sent SYN and data. 2154 * If the remote host used T/TCP to validate the SYN, 2155 * our data will be ACK'd; if so, enter normal data segment 2156 * processing in the middle of step 5, ack processing. 2157 * Otherwise, goto step 6. 2158 */ 2159 if (thflags & TH_ACK) 2160 goto process_ACK; 2161 2162 goto step6; 2163 } 2164 2165 /* 2166 * States other than LISTEN or SYN_SENT. 2167 * First check the RST flag and sequence number since reset segments 2168 * are exempt from the timestamp and connection count tests. This 2169 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2170 * below which allowed reset segments in half the sequence space 2171 * to fall though and be processed (which gives forged reset 2172 * segments with a random sequence number a 50 percent chance of 2173 * killing a connection). 2174 * Then check timestamp, if present. 2175 * Then check the connection count, if present. 2176 * Then check that at least some bytes of segment are within 2177 * receive window. If segment begins before rcv_nxt, 2178 * drop leading data (and SYN); if nothing left, just ack. 2179 */ 2180 if (thflags & TH_RST) { 2181 /* 2182 * RFC5961 Section 3.2 2183 * 2184 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2185 * - If RST is in window, we send challenge ACK. 2186 * 2187 * Note: to take into account delayed ACKs, we should 2188 * test against last_ack_sent instead of rcv_nxt. 2189 * Note 2: we handle special case of closed window, not 2190 * covered by the RFC. 2191 */ 2192 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2193 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2194 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2195 KASSERT(tp->t_state != TCPS_SYN_SENT, 2196 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2197 __func__, th, tp)); 2198 2199 if (V_tcp_insecure_rst || 2200 tp->last_ack_sent == th->th_seq) { 2201 TCPSTAT_INC(tcps_drops); 2202 /* Drop the connection. */ 2203 switch (tp->t_state) { 2204 case TCPS_SYN_RECEIVED: 2205 so->so_error = ECONNREFUSED; 2206 goto close; 2207 case TCPS_ESTABLISHED: 2208 case TCPS_FIN_WAIT_1: 2209 case TCPS_FIN_WAIT_2: 2210 case TCPS_CLOSE_WAIT: 2211 case TCPS_CLOSING: 2212 case TCPS_LAST_ACK: 2213 so->so_error = ECONNRESET; 2214 close: 2215 /* FALLTHROUGH */ 2216 default: 2217 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); 2218 tp = tcp_close(tp); 2219 } 2220 } else { 2221 TCPSTAT_INC(tcps_badrst); 2222 /* Send challenge ACK. */ 2223 tcp_respond(tp, mtod(m, void *), th, m, 2224 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2225 tp->last_ack_sent = tp->rcv_nxt; 2226 m = NULL; 2227 } 2228 } 2229 goto drop; 2230 } 2231 2232 /* 2233 * RFC5961 Section 4.2 2234 * Send challenge ACK for any SYN in synchronized state. 2235 */ 2236 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2237 tp->t_state != TCPS_SYN_RECEIVED) { 2238 TCPSTAT_INC(tcps_badsyn); 2239 if (V_tcp_insecure_syn && 2240 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2241 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2242 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2243 tp = tcp_drop(tp, ECONNRESET); 2244 rstreason = BANDLIM_UNLIMITED; 2245 } else { 2246 /* Send challenge ACK. */ 2247 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2248 tp->snd_nxt, TH_ACK); 2249 tp->last_ack_sent = tp->rcv_nxt; 2250 m = NULL; 2251 } 2252 goto drop; 2253 } 2254 2255 /* 2256 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2257 * and it's less than ts_recent, drop it. 2258 */ 2259 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2260 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2261 /* Check to see if ts_recent is over 24 days old. */ 2262 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2263 /* 2264 * Invalidate ts_recent. If this segment updates 2265 * ts_recent, the age will be reset later and ts_recent 2266 * will get a valid value. If it does not, setting 2267 * ts_recent to zero will at least satisfy the 2268 * requirement that zero be placed in the timestamp 2269 * echo reply when ts_recent isn't valid. The 2270 * age isn't reset until we get a valid ts_recent 2271 * because we don't want out-of-order segments to be 2272 * dropped when ts_recent is old. 2273 */ 2274 tp->ts_recent = 0; 2275 } else { 2276 TCPSTAT_INC(tcps_rcvduppack); 2277 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2278 TCPSTAT_INC(tcps_pawsdrop); 2279 if (tlen) 2280 goto dropafterack; 2281 goto drop; 2282 } 2283 } 2284 2285 /* 2286 * In the SYN-RECEIVED state, validate that the packet belongs to 2287 * this connection before trimming the data to fit the receive 2288 * window. Check the sequence number versus IRS since we know 2289 * the sequence numbers haven't wrapped. This is a partial fix 2290 * for the "LAND" DoS attack. 2291 */ 2292 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2293 rstreason = BANDLIM_RST_OPENPORT; 2294 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 2295 goto dropwithreset; 2296 } 2297 2298 todrop = tp->rcv_nxt - th->th_seq; 2299 if (todrop > 0) { 2300 if (thflags & TH_SYN) { 2301 thflags &= ~TH_SYN; 2302 th->th_seq++; 2303 if (th->th_urp > 1) 2304 th->th_urp--; 2305 else 2306 thflags &= ~TH_URG; 2307 todrop--; 2308 } 2309 /* 2310 * Following if statement from Stevens, vol. 2, p. 960. 2311 */ 2312 if (todrop > tlen 2313 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2314 /* 2315 * Any valid FIN must be to the left of the window. 2316 * At this point the FIN must be a duplicate or out 2317 * of sequence; drop it. 2318 */ 2319 thflags &= ~TH_FIN; 2320 2321 /* 2322 * Send an ACK to resynchronize and drop any data. 2323 * But keep on processing for RST or ACK. 2324 */ 2325 tp->t_flags |= TF_ACKNOW; 2326 todrop = tlen; 2327 TCPSTAT_INC(tcps_rcvduppack); 2328 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2329 } else { 2330 TCPSTAT_INC(tcps_rcvpartduppack); 2331 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2332 } 2333 /* 2334 * DSACK - add SACK block for dropped range 2335 */ 2336 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2337 tcp_update_sack_list(tp, th->th_seq, 2338 th->th_seq + todrop); 2339 /* 2340 * ACK now, as the next in-sequence segment 2341 * will clear the DSACK block again 2342 */ 2343 tp->t_flags |= TF_ACKNOW; 2344 } 2345 drop_hdrlen += todrop; /* drop from the top afterwards */ 2346 th->th_seq += todrop; 2347 tlen -= todrop; 2348 if (th->th_urp > todrop) 2349 th->th_urp -= todrop; 2350 else { 2351 thflags &= ~TH_URG; 2352 th->th_urp = 0; 2353 } 2354 } 2355 2356 /* 2357 * If new data are received on a connection after the 2358 * user processes are gone, then RST the other end. 2359 */ 2360 if ((tp->t_flags & TF_CLOSED) && tlen) { 2361 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2362 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2363 "after socket was closed, " 2364 "sending RST and removing tcpcb\n", 2365 s, __func__, tcpstates[tp->t_state], tlen); 2366 free(s, M_TCPLOG); 2367 } 2368 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); 2369 /* tcp_close will kill the inp pre-log the Reset */ 2370 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); 2371 tp = tcp_close(tp); 2372 TCPSTAT_INC(tcps_rcvafterclose); 2373 rstreason = BANDLIM_UNLIMITED; 2374 goto dropwithreset; 2375 } 2376 2377 /* 2378 * If segment ends after window, drop trailing data 2379 * (and PUSH and FIN); if nothing left, just ACK. 2380 */ 2381 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2382 if (todrop > 0) { 2383 TCPSTAT_INC(tcps_rcvpackafterwin); 2384 if (todrop >= tlen) { 2385 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2386 /* 2387 * If window is closed can only take segments at 2388 * window edge, and have to drop data and PUSH from 2389 * incoming segments. Continue processing, but 2390 * remember to ack. Otherwise, drop segment 2391 * and ack. 2392 */ 2393 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2394 tp->t_flags |= TF_ACKNOW; 2395 TCPSTAT_INC(tcps_rcvwinprobe); 2396 } else 2397 goto dropafterack; 2398 } else 2399 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2400 m_adj(m, -todrop); 2401 tlen -= todrop; 2402 thflags &= ~(TH_PUSH|TH_FIN); 2403 } 2404 2405 /* 2406 * If last ACK falls within this segment's sequence numbers, 2407 * record its timestamp. 2408 * NOTE: 2409 * 1) That the test incorporates suggestions from the latest 2410 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2411 * 2) That updating only on newer timestamps interferes with 2412 * our earlier PAWS tests, so this check should be solely 2413 * predicated on the sequence space of this segment. 2414 * 3) That we modify the segment boundary check to be 2415 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2416 * instead of RFC1323's 2417 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2418 * This modified check allows us to overcome RFC1323's 2419 * limitations as described in Stevens TCP/IP Illustrated 2420 * Vol. 2 p.869. In such cases, we can still calculate the 2421 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2422 */ 2423 if ((to.to_flags & TOF_TS) != 0 && 2424 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2425 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2426 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2427 tp->ts_recent_age = tcp_ts_getticks(); 2428 tp->ts_recent = to.to_tsval; 2429 } 2430 2431 /* 2432 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2433 * flag is on (half-synchronized state), then queue data for 2434 * later processing; else drop segment and return. 2435 */ 2436 if ((thflags & TH_ACK) == 0) { 2437 if (tp->t_state == TCPS_SYN_RECEIVED || 2438 (tp->t_flags & TF_NEEDSYN)) { 2439 if (tp->t_state == TCPS_SYN_RECEIVED && 2440 IS_FASTOPEN(tp->t_flags)) { 2441 tp->snd_wnd = tiwin; 2442 cc_conn_init(tp); 2443 } 2444 goto step6; 2445 } else if (tp->t_flags & TF_ACKNOW) 2446 goto dropafterack; 2447 else 2448 goto drop; 2449 } 2450 2451 /* 2452 * Ack processing. 2453 */ 2454 switch (tp->t_state) { 2455 /* 2456 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2457 * ESTABLISHED state and continue processing. 2458 * The ACK was checked above. 2459 */ 2460 case TCPS_SYN_RECEIVED: 2461 2462 TCPSTAT_INC(tcps_connects); 2463 if (tp->t_flags & TF_SONOTCONN) { 2464 /* 2465 * Usually SYN_RECEIVED had been created from a LISTEN, 2466 * and solisten_enqueue() has already marked the socket 2467 * layer as connected. If it didn't, which can happen 2468 * only with an accept_filter(9), then the tp is marked 2469 * with TF_SONOTCONN. The other reason for this mark 2470 * to be set is a simultaneous open, a SYN_RECEIVED 2471 * that had been created from SYN_SENT. 2472 */ 2473 tp->t_flags &= ~TF_SONOTCONN; 2474 soisconnected(so); 2475 } 2476 /* Do window scaling? */ 2477 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2478 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2479 tp->rcv_scale = tp->request_r_scale; 2480 } 2481 tp->snd_wnd = tiwin; 2482 /* 2483 * Make transitions: 2484 * SYN-RECEIVED -> ESTABLISHED 2485 * SYN-RECEIVED* -> FIN-WAIT-1 2486 */ 2487 tp->t_starttime = ticks; 2488 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2489 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2490 tp->t_tfo_pending = NULL; 2491 } 2492 if (tp->t_flags & TF_NEEDFIN) { 2493 tp->t_acktime = ticks; 2494 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2495 tp->t_flags &= ~TF_NEEDFIN; 2496 } else { 2497 tcp_state_change(tp, TCPS_ESTABLISHED); 2498 TCP_PROBE5(accept__established, NULL, tp, 2499 m, tp, th); 2500 /* 2501 * TFO connections call cc_conn_init() during SYN 2502 * processing. Calling it again here for such 2503 * connections is not harmless as it would undo the 2504 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2505 * is retransmitted. 2506 */ 2507 if (!IS_FASTOPEN(tp->t_flags)) 2508 cc_conn_init(tp); 2509 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2510 } 2511 /* 2512 * Account for the ACK of our SYN prior to 2513 * regular ACK processing below, except for 2514 * simultaneous SYN, which is handled later. 2515 */ 2516 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2517 incforsyn = 1; 2518 /* 2519 * If segment contains data or ACK, will call tcp_reass() 2520 * later; if not, do so now to pass queued data to user. 2521 */ 2522 if (tlen == 0 && (thflags & TH_FIN) == 0) { 2523 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2524 (struct mbuf *)0); 2525 tcp_handle_wakeup(tp, so); 2526 } 2527 tp->snd_wl1 = th->th_seq - 1; 2528 /* FALLTHROUGH */ 2529 2530 /* 2531 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2532 * ACKs. If the ack is in the range 2533 * tp->snd_una < th->th_ack <= tp->snd_max 2534 * then advance tp->snd_una to th->th_ack and drop 2535 * data from the retransmission queue. If this ACK reflects 2536 * more up to date window information we update our window information. 2537 */ 2538 case TCPS_ESTABLISHED: 2539 case TCPS_FIN_WAIT_1: 2540 case TCPS_FIN_WAIT_2: 2541 case TCPS_CLOSE_WAIT: 2542 case TCPS_CLOSING: 2543 case TCPS_LAST_ACK: 2544 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2545 TCPSTAT_INC(tcps_rcvacktoomuch); 2546 goto dropafterack; 2547 } 2548 if (tcp_is_sack_recovery(tp, &to)) { 2549 if (((sack_changed = tcp_sack_doack(tp, &to, th->th_ack)) != 0) && 2550 (tp->t_flags & TF_LRD)) { 2551 tcp_sack_lost_retransmission(tp, th); 2552 } 2553 } else 2554 /* 2555 * Reset the value so that previous (valid) value 2556 * from the last ack with SACK doesn't get used. 2557 */ 2558 tp->sackhint.sacked_bytes = 0; 2559 2560 #ifdef TCP_HHOOK 2561 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2562 hhook_run_tcp_est_in(tp, th, &to); 2563 #endif 2564 2565 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2566 maxseg = tcp_maxseg(tp); 2567 if (tlen == 0 && 2568 (tiwin == tp->snd_wnd || 2569 (tp->t_flags & TF_SACK_PERMIT))) { 2570 /* 2571 * If this is the first time we've seen a 2572 * FIN from the remote, this is not a 2573 * duplicate and it needs to be processed 2574 * normally. This happens during a 2575 * simultaneous close. 2576 */ 2577 if ((thflags & TH_FIN) && 2578 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2579 tp->t_dupacks = 0; 2580 break; 2581 } 2582 TCPSTAT_INC(tcps_rcvdupack); 2583 /* 2584 * If we have outstanding data (other than 2585 * a window probe), this is a completely 2586 * duplicate ack (ie, window info didn't 2587 * change and FIN isn't set), 2588 * the ack is the biggest we've 2589 * seen and we've seen exactly our rexmt 2590 * threshold of them, assume a packet 2591 * has been dropped and retransmit it. 2592 * Kludge snd_nxt & the congestion 2593 * window so we send only this one 2594 * packet. 2595 * 2596 * We know we're losing at the current 2597 * window size so do congestion avoidance 2598 * (set ssthresh to half the current window 2599 * and pull our congestion window back to 2600 * the new ssthresh). 2601 * 2602 * Dup acks mean that packets have left the 2603 * network (they're now cached at the receiver) 2604 * so bump cwnd by the amount in the receiver 2605 * to keep a constant cwnd packets in the 2606 * network. 2607 * 2608 * When using TCP ECN, notify the peer that 2609 * we reduced the cwnd. 2610 */ 2611 /* 2612 * Following 2 kinds of acks should not affect 2613 * dupack counting: 2614 * 1) Old acks 2615 * 2) Acks with SACK but without any new SACK 2616 * information in them. These could result from 2617 * any anomaly in the network like a switch 2618 * duplicating packets or a possible DoS attack. 2619 */ 2620 if (th->th_ack != tp->snd_una || 2621 (tcp_is_sack_recovery(tp, &to) && 2622 !sack_changed)) 2623 break; 2624 else if (!tcp_timer_active(tp, TT_REXMT)) 2625 tp->t_dupacks = 0; 2626 else if (++tp->t_dupacks > tcprexmtthresh || 2627 IN_FASTRECOVERY(tp->t_flags)) { 2628 cc_ack_received(tp, th, nsegs, 2629 CC_DUPACK); 2630 if (V_tcp_do_prr && 2631 IN_FASTRECOVERY(tp->t_flags)) { 2632 tcp_do_prr_ack(tp, th, &to); 2633 } else if (tcp_is_sack_recovery(tp, &to) && 2634 IN_FASTRECOVERY(tp->t_flags)) { 2635 int awnd; 2636 2637 /* 2638 * Compute the amount of data in flight first. 2639 * We can inject new data into the pipe iff 2640 * we have less than 1/2 the original window's 2641 * worth of data in flight. 2642 */ 2643 if (V_tcp_do_newsack) 2644 awnd = tcp_compute_pipe(tp); 2645 else 2646 awnd = (tp->snd_nxt - tp->snd_fack) + 2647 tp->sackhint.sack_bytes_rexmit; 2648 2649 if (awnd < tp->snd_ssthresh) { 2650 tp->snd_cwnd += maxseg; 2651 if (tp->snd_cwnd > tp->snd_ssthresh) 2652 tp->snd_cwnd = tp->snd_ssthresh; 2653 } 2654 } else 2655 tp->snd_cwnd += maxseg; 2656 (void) tcp_output(tp); 2657 goto drop; 2658 } else if (tp->t_dupacks == tcprexmtthresh || 2659 (tp->t_flags & TF_SACK_PERMIT && 2660 V_tcp_do_newsack && 2661 tp->sackhint.sacked_bytes > 2662 (tcprexmtthresh - 1) * maxseg)) { 2663 enter_recovery: 2664 /* 2665 * Above is the RFC6675 trigger condition of 2666 * more than (dupthresh-1)*maxseg sacked data. 2667 * If the count of holes in the 2668 * scoreboard is >= dupthresh, we could 2669 * also enter loss recovery, but don't 2670 * have that value readily available. 2671 */ 2672 tp->t_dupacks = tcprexmtthresh; 2673 tcp_seq onxt = tp->snd_nxt; 2674 2675 /* 2676 * If we're doing sack, or prr, check 2677 * to see if we're already in sack 2678 * recovery. If we're not doing sack, 2679 * check to see if we're in newreno 2680 * recovery. 2681 */ 2682 if (V_tcp_do_prr || 2683 (tp->t_flags & TF_SACK_PERMIT)) { 2684 if (IN_FASTRECOVERY(tp->t_flags)) { 2685 tp->t_dupacks = 0; 2686 break; 2687 } 2688 } else { 2689 if (SEQ_LEQ(th->th_ack, 2690 tp->snd_recover)) { 2691 tp->t_dupacks = 0; 2692 break; 2693 } 2694 } 2695 /* Congestion signal before ack. */ 2696 cc_cong_signal(tp, th, CC_NDUPACK); 2697 cc_ack_received(tp, th, nsegs, 2698 CC_DUPACK); 2699 tcp_timer_activate(tp, TT_REXMT, 0); 2700 tp->t_rtttime = 0; 2701 if (V_tcp_do_prr) { 2702 /* 2703 * snd_ssthresh is already updated by 2704 * cc_cong_signal. 2705 */ 2706 if (tcp_is_sack_recovery(tp, &to)) { 2707 tp->sackhint.prr_delivered = 2708 tp->sackhint.sacked_bytes; 2709 } else { 2710 tp->sackhint.prr_delivered = 2711 imin(tp->snd_max - tp->snd_una, 2712 imin(INT_MAX / 65536, 2713 tp->t_dupacks) * maxseg); 2714 } 2715 tp->sackhint.recover_fs = max(1, 2716 tp->snd_nxt - tp->snd_una); 2717 } 2718 if (tcp_is_sack_recovery(tp, &to)) { 2719 TCPSTAT_INC( 2720 tcps_sack_recovery_episode); 2721 tp->snd_recover = tp->snd_nxt; 2722 tp->snd_cwnd = maxseg; 2723 (void) 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) 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 * or we need to send an ACK. 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 || tp->t_flags & TF_ACKNOW) 2778 (void) 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 (tcp_is_sack_recovery(tp, &to) && 2811 sack_changed) { 2812 tp->t_dupacks++; 2813 /* limit overhead by setting maxseg last */ 2814 if (!IN_FASTRECOVERY(tp->t_flags) && 2815 (tp->sackhint.sacked_bytes > 2816 ((tcprexmtthresh - 1) * 2817 (maxseg = tcp_maxseg(tp))))) { 2818 goto enter_recovery; 2819 } 2820 } 2821 } 2822 2823 resume_partialack: 2824 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2825 ("%s: th_ack <= snd_una", __func__)); 2826 2827 /* 2828 * If the congestion window was inflated to account 2829 * for the other side's cached packets, retract it. 2830 */ 2831 if (IN_FASTRECOVERY(tp->t_flags)) { 2832 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2833 if (tp->t_flags & TF_SACK_PERMIT) 2834 if (V_tcp_do_prr && to.to_flags & TOF_SACK) { 2835 tcp_timer_activate(tp, TT_REXMT, 0); 2836 tp->t_rtttime = 0; 2837 tcp_do_prr_ack(tp, th, &to); 2838 tp->t_flags |= TF_ACKNOW; 2839 (void) tcp_output(tp); 2840 } else 2841 tcp_sack_partialack(tp, th); 2842 else 2843 tcp_newreno_partial_ack(tp, th); 2844 } else 2845 cc_post_recovery(tp, th); 2846 } else if (IN_CONGRECOVERY(tp->t_flags)) { 2847 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2848 if (V_tcp_do_prr) { 2849 tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); 2850 tp->snd_fack = th->th_ack; 2851 tcp_do_prr_ack(tp, th, &to); 2852 (void) tcp_output(tp); 2853 } 2854 } else 2855 cc_post_recovery(tp, th); 2856 } 2857 /* 2858 * If we reach this point, ACK is not a duplicate, 2859 * i.e., it ACKs something we sent. 2860 */ 2861 if (tp->t_flags & TF_NEEDSYN) { 2862 /* 2863 * T/TCP: Connection was half-synchronized, and our 2864 * SYN has been ACK'd (so connection is now fully 2865 * synchronized). Go to non-starred state, 2866 * increment snd_una for ACK of SYN, and check if 2867 * we can do window scaling. 2868 */ 2869 tp->t_flags &= ~TF_NEEDSYN; 2870 tp->snd_una++; 2871 /* Do window scaling? */ 2872 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2873 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2874 tp->rcv_scale = tp->request_r_scale; 2875 /* Send window already scaled. */ 2876 } 2877 } 2878 2879 process_ACK: 2880 INP_WLOCK_ASSERT(tp->t_inpcb); 2881 2882 /* 2883 * Adjust for the SYN bit in sequence space, 2884 * but don't account for it in cwnd calculations. 2885 * This is for the SYN_RECEIVED, non-simultaneous 2886 * SYN case. SYN_SENT and simultaneous SYN are 2887 * treated elsewhere. 2888 */ 2889 if (incforsyn) 2890 tp->snd_una++; 2891 acked = BYTES_THIS_ACK(tp, th); 2892 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2893 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2894 tp->snd_una, th->th_ack, tp, m)); 2895 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2896 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2897 2898 /* 2899 * If we just performed our first retransmit, and the ACK 2900 * arrives within our recovery window, then it was a mistake 2901 * to do the retransmit in the first place. Recover our 2902 * original cwnd and ssthresh, and proceed to transmit where 2903 * we left off. 2904 */ 2905 if (tp->t_rxtshift == 1 && 2906 tp->t_flags & TF_PREVVALID && 2907 tp->t_badrxtwin != 0 && 2908 to.to_flags & TOF_TS && 2909 to.to_tsecr != 0 && 2910 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) 2911 cc_cong_signal(tp, th, CC_RTO_ERR); 2912 2913 /* 2914 * If we have a timestamp reply, update smoothed 2915 * round trip time. If no timestamp is present but 2916 * transmit timer is running and timed sequence 2917 * number was acked, update smoothed round trip time. 2918 * Since we now have an rtt measurement, cancel the 2919 * timer backoff (cf., Phil Karn's retransmit alg.). 2920 * Recompute the initial retransmit timer. 2921 * 2922 * Some boxes send broken timestamp replies 2923 * during the SYN+ACK phase, ignore 2924 * timestamps of 0 or we could calculate a 2925 * huge RTT and blow up the retransmit timer. 2926 */ 2927 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2928 uint32_t t; 2929 2930 t = tcp_ts_getticks() - to.to_tsecr; 2931 if (!tp->t_rttlow || tp->t_rttlow > t) 2932 tp->t_rttlow = t; 2933 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2934 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2935 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2936 tp->t_rttlow = ticks - tp->t_rtttime; 2937 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2938 } 2939 2940 SOCKBUF_LOCK(&so->so_snd); 2941 /* 2942 * Clear t_acktime if remote side has ACKd all data in the 2943 * socket buffer and FIN (if applicable). 2944 * Otherwise, update t_acktime if we received a sufficiently 2945 * large ACK. 2946 */ 2947 if ((tp->t_state <= TCPS_CLOSE_WAIT && 2948 acked == sbavail(&so->so_snd)) || 2949 acked > sbavail(&so->so_snd)) 2950 tp->t_acktime = 0; 2951 else if (acked > 1) 2952 tp->t_acktime = ticks; 2953 2954 /* 2955 * If all outstanding data is acked, stop retransmit 2956 * timer and remember to restart (more output or persist). 2957 * If there is more data to be acked, restart retransmit 2958 * timer, using current (possibly backed-off) value. 2959 */ 2960 if (th->th_ack == tp->snd_max) { 2961 tcp_timer_activate(tp, TT_REXMT, 0); 2962 needoutput = 1; 2963 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2964 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); 2965 2966 /* 2967 * If no data (only SYN) was ACK'd, 2968 * skip rest of ACK processing. 2969 */ 2970 if (acked == 0) { 2971 SOCKBUF_UNLOCK(&so->so_snd); 2972 goto step6; 2973 } 2974 2975 /* 2976 * Let the congestion control algorithm update congestion 2977 * control related information. This typically means increasing 2978 * the congestion window. 2979 */ 2980 cc_ack_received(tp, th, nsegs, CC_ACK); 2981 2982 if (acked > sbavail(&so->so_snd)) { 2983 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2984 tp->snd_wnd -= sbavail(&so->so_snd); 2985 else 2986 tp->snd_wnd = 0; 2987 mfree = sbcut_locked(&so->so_snd, 2988 (int)sbavail(&so->so_snd)); 2989 ourfinisacked = 1; 2990 } else { 2991 mfree = sbcut_locked(&so->so_snd, acked); 2992 if (tp->snd_wnd >= (uint32_t) acked) 2993 tp->snd_wnd -= acked; 2994 else 2995 tp->snd_wnd = 0; 2996 ourfinisacked = 0; 2997 } 2998 /* NB: sowwakeup_locked() does an implicit unlock. */ 2999 sowwakeup_locked(so); 3000 m_freem(mfree); 3001 /* Detect una wraparound. */ 3002 if (!IN_RECOVERY(tp->t_flags) && 3003 SEQ_GT(tp->snd_una, tp->snd_recover) && 3004 SEQ_LEQ(th->th_ack, tp->snd_recover)) 3005 tp->snd_recover = th->th_ack - 1; 3006 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 3007 if (IN_RECOVERY(tp->t_flags) && 3008 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 3009 EXIT_RECOVERY(tp->t_flags); 3010 } 3011 tp->snd_una = th->th_ack; 3012 if (tp->t_flags & TF_SACK_PERMIT) { 3013 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 3014 tp->snd_recover = tp->snd_una; 3015 } 3016 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 3017 tp->snd_nxt = tp->snd_una; 3018 3019 switch (tp->t_state) { 3020 /* 3021 * In FIN_WAIT_1 STATE in addition to the processing 3022 * for the ESTABLISHED state if our FIN is now acknowledged 3023 * then enter FIN_WAIT_2. 3024 */ 3025 case TCPS_FIN_WAIT_1: 3026 if (ourfinisacked) { 3027 /* 3028 * If we can't receive any more 3029 * data, then closing user can proceed. 3030 * Starting the timer is contrary to the 3031 * specification, but if we don't get a FIN 3032 * we'll hang forever. 3033 */ 3034 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 3035 soisdisconnected(so); 3036 tcp_timer_activate(tp, TT_2MSL, 3037 (tcp_fast_finwait2_recycle ? 3038 tcp_finwait2_timeout : 3039 TP_MAXIDLE(tp))); 3040 } 3041 tcp_state_change(tp, TCPS_FIN_WAIT_2); 3042 } 3043 break; 3044 3045 /* 3046 * In CLOSING STATE in addition to the processing for 3047 * the ESTABLISHED state if the ACK acknowledges our FIN 3048 * then enter the TIME-WAIT state, otherwise ignore 3049 * the segment. 3050 */ 3051 case TCPS_CLOSING: 3052 if (ourfinisacked) { 3053 tcp_twstart(tp); 3054 m_freem(m); 3055 return; 3056 } 3057 break; 3058 3059 /* 3060 * In LAST_ACK, we may still be waiting for data to drain 3061 * and/or to be acked, as well as for the ack of our FIN. 3062 * If our FIN is now acknowledged, delete the TCB, 3063 * enter the closed state and return. 3064 */ 3065 case TCPS_LAST_ACK: 3066 if (ourfinisacked) { 3067 tp = tcp_close(tp); 3068 goto drop; 3069 } 3070 break; 3071 } 3072 } 3073 3074 step6: 3075 INP_WLOCK_ASSERT(tp->t_inpcb); 3076 3077 /* 3078 * Update window information. 3079 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3080 */ 3081 if ((thflags & TH_ACK) && 3082 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3083 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3084 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3085 /* keep track of pure window updates */ 3086 if (tlen == 0 && 3087 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3088 TCPSTAT_INC(tcps_rcvwinupd); 3089 tp->snd_wnd = tiwin; 3090 tp->snd_wl1 = th->th_seq; 3091 tp->snd_wl2 = th->th_ack; 3092 if (tp->snd_wnd > tp->max_sndwnd) 3093 tp->max_sndwnd = tp->snd_wnd; 3094 needoutput = 1; 3095 } 3096 3097 /* 3098 * Process segments with URG. 3099 */ 3100 if ((thflags & TH_URG) && th->th_urp && 3101 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3102 /* 3103 * This is a kludge, but if we receive and accept 3104 * random urgent pointers, we'll crash in 3105 * soreceive. It's hard to imagine someone 3106 * actually wanting to send this much urgent data. 3107 */ 3108 SOCKBUF_LOCK(&so->so_rcv); 3109 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3110 th->th_urp = 0; /* XXX */ 3111 thflags &= ~TH_URG; /* XXX */ 3112 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3113 goto dodata; /* XXX */ 3114 } 3115 /* 3116 * If this segment advances the known urgent pointer, 3117 * then mark the data stream. This should not happen 3118 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3119 * a FIN has been received from the remote side. 3120 * In these states we ignore the URG. 3121 * 3122 * According to RFC961 (Assigned Protocols), 3123 * the urgent pointer points to the last octet 3124 * of urgent data. We continue, however, 3125 * to consider it to indicate the first octet 3126 * of data past the urgent section as the original 3127 * spec states (in one of two places). 3128 */ 3129 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3130 tp->rcv_up = th->th_seq + th->th_urp; 3131 so->so_oobmark = sbavail(&so->so_rcv) + 3132 (tp->rcv_up - tp->rcv_nxt) - 1; 3133 if (so->so_oobmark == 0) 3134 so->so_rcv.sb_state |= SBS_RCVATMARK; 3135 sohasoutofband(so); 3136 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3137 } 3138 SOCKBUF_UNLOCK(&so->so_rcv); 3139 /* 3140 * Remove out of band data so doesn't get presented to user. 3141 * This can happen independent of advancing the URG pointer, 3142 * but if two URG's are pending at once, some out-of-band 3143 * data may creep in... ick. 3144 */ 3145 if (th->th_urp <= (uint32_t)tlen && 3146 !(so->so_options & SO_OOBINLINE)) { 3147 /* hdr drop is delayed */ 3148 tcp_pulloutofband(so, th, m, drop_hdrlen); 3149 } 3150 } else { 3151 /* 3152 * If no out of band data is expected, 3153 * pull receive urgent pointer along 3154 * with the receive window. 3155 */ 3156 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3157 tp->rcv_up = tp->rcv_nxt; 3158 } 3159 dodata: /* XXX */ 3160 INP_WLOCK_ASSERT(tp->t_inpcb); 3161 3162 /* 3163 * Process the segment text, merging it into the TCP sequencing queue, 3164 * and arranging for acknowledgment of receipt if necessary. 3165 * This process logically involves adjusting tp->rcv_wnd as data 3166 * is presented to the user (this happens in tcp_usrreq.c, 3167 * case PRU_RCVD). If a FIN has already been received on this 3168 * connection then we just ignore the text. 3169 */ 3170 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3171 IS_FASTOPEN(tp->t_flags)); 3172 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3173 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3174 tcp_seq save_start = th->th_seq; 3175 tcp_seq save_rnxt = tp->rcv_nxt; 3176 int save_tlen = tlen; 3177 m_adj(m, drop_hdrlen); /* delayed header drop */ 3178 /* 3179 * Insert segment which includes th into TCP reassembly queue 3180 * with control block tp. Set thflags to whether reassembly now 3181 * includes a segment with FIN. This handles the common case 3182 * inline (segment is the next to be received on an established 3183 * connection, and the queue is empty), avoiding linkage into 3184 * and removal from the queue and repetition of various 3185 * conversions. 3186 * Set DELACK for segments received in order, but ack 3187 * immediately when segments are out of order (so 3188 * fast retransmit can work). 3189 */ 3190 if (th->th_seq == tp->rcv_nxt && 3191 SEGQ_EMPTY(tp) && 3192 (TCPS_HAVEESTABLISHED(tp->t_state) || 3193 tfo_syn)) { 3194 if (DELAY_ACK(tp, tlen) || tfo_syn) 3195 tp->t_flags |= TF_DELACK; 3196 else 3197 tp->t_flags |= TF_ACKNOW; 3198 tp->rcv_nxt += tlen; 3199 if (tlen && 3200 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3201 (tp->t_fbyte_in == 0)) { 3202 tp->t_fbyte_in = ticks; 3203 if (tp->t_fbyte_in == 0) 3204 tp->t_fbyte_in = 1; 3205 if (tp->t_fbyte_out && tp->t_fbyte_in) 3206 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3207 } 3208 thflags = tcp_get_flags(th) & TH_FIN; 3209 TCPSTAT_INC(tcps_rcvpack); 3210 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3211 SOCKBUF_LOCK(&so->so_rcv); 3212 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3213 m_freem(m); 3214 else 3215 sbappendstream_locked(&so->so_rcv, m, 0); 3216 tp->t_flags |= TF_WAKESOR; 3217 } else { 3218 /* 3219 * XXX: Due to the header drop above "th" is 3220 * theoretically invalid by now. Fortunately 3221 * m_adj() doesn't actually frees any mbufs 3222 * when trimming from the head. 3223 */ 3224 tcp_seq temp = save_start; 3225 3226 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3227 tp->t_flags |= TF_ACKNOW; 3228 } 3229 if ((tp->t_flags & TF_SACK_PERMIT) && 3230 (save_tlen > 0) && 3231 TCPS_HAVEESTABLISHED(tp->t_state)) { 3232 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3233 /* 3234 * DSACK actually handled in the fastpath 3235 * above. 3236 */ 3237 tcp_update_sack_list(tp, save_start, 3238 save_start + save_tlen); 3239 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3240 if ((tp->rcv_numsacks >= 1) && 3241 (tp->sackblks[0].end == save_start)) { 3242 /* 3243 * Partial overlap, recorded at todrop 3244 * above. 3245 */ 3246 tcp_update_sack_list(tp, 3247 tp->sackblks[0].start, 3248 tp->sackblks[0].end); 3249 } else { 3250 tcp_update_dsack_list(tp, save_start, 3251 save_start + save_tlen); 3252 } 3253 } else if (tlen >= save_tlen) { 3254 /* Update of sackblks. */ 3255 tcp_update_dsack_list(tp, save_start, 3256 save_start + save_tlen); 3257 } else if (tlen > 0) { 3258 tcp_update_dsack_list(tp, save_start, 3259 save_start + tlen); 3260 } 3261 } 3262 tcp_handle_wakeup(tp, so); 3263 #if 0 3264 /* 3265 * Note the amount of data that peer has sent into 3266 * our window, in order to estimate the sender's 3267 * buffer size. 3268 * XXX: Unused. 3269 */ 3270 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3271 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3272 else 3273 len = so->so_rcv.sb_hiwat; 3274 #endif 3275 } else { 3276 m_freem(m); 3277 thflags &= ~TH_FIN; 3278 } 3279 3280 /* 3281 * If FIN is received ACK the FIN and let the user know 3282 * that the connection is closing. 3283 */ 3284 if (thflags & TH_FIN) { 3285 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3286 /* The socket upcall is handled by socantrcvmore. */ 3287 socantrcvmore(so); 3288 /* 3289 * If connection is half-synchronized 3290 * (ie NEEDSYN flag on) then delay ACK, 3291 * so it may be piggybacked when SYN is sent. 3292 * Otherwise, since we received a FIN then no 3293 * more input can be expected, send ACK now. 3294 */ 3295 if (tp->t_flags & TF_NEEDSYN) 3296 tp->t_flags |= TF_DELACK; 3297 else 3298 tp->t_flags |= TF_ACKNOW; 3299 tp->rcv_nxt++; 3300 } 3301 switch (tp->t_state) { 3302 /* 3303 * In SYN_RECEIVED and ESTABLISHED STATES 3304 * enter the CLOSE_WAIT state. 3305 */ 3306 case TCPS_SYN_RECEIVED: 3307 tp->t_starttime = ticks; 3308 /* FALLTHROUGH */ 3309 case TCPS_ESTABLISHED: 3310 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3311 break; 3312 3313 /* 3314 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3315 * enter the CLOSING state. 3316 */ 3317 case TCPS_FIN_WAIT_1: 3318 tcp_state_change(tp, TCPS_CLOSING); 3319 break; 3320 3321 /* 3322 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3323 * starting the time-wait timer, turning off the other 3324 * standard timers. 3325 */ 3326 case TCPS_FIN_WAIT_2: 3327 tcp_twstart(tp); 3328 return; 3329 } 3330 } 3331 #ifdef TCPDEBUG 3332 if (so->so_options & SO_DEBUG) 3333 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3334 &tcp_savetcp, 0); 3335 #endif 3336 TCP_PROBE3(debug__input, tp, th, m); 3337 3338 /* 3339 * Return any desired output. 3340 */ 3341 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3342 (void) tcp_output(tp); 3343 3344 check_delack: 3345 INP_WLOCK_ASSERT(tp->t_inpcb); 3346 3347 if (tp->t_flags & TF_DELACK) { 3348 tp->t_flags &= ~TF_DELACK; 3349 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3350 } 3351 INP_WUNLOCK(tp->t_inpcb); 3352 return; 3353 3354 dropafterack: 3355 /* 3356 * Generate an ACK dropping incoming segment if it occupies 3357 * sequence space, where the ACK reflects our state. 3358 * 3359 * We can now skip the test for the RST flag since all 3360 * paths to this code happen after packets containing 3361 * RST have been dropped. 3362 * 3363 * In the SYN-RECEIVED state, don't send an ACK unless the 3364 * segment we received passes the SYN-RECEIVED ACK test. 3365 * If it fails send a RST. This breaks the loop in the 3366 * "LAND" DoS attack, and also prevents an ACK storm 3367 * between two listening ports that have been sent forged 3368 * SYN segments, each with the source address of the other. 3369 */ 3370 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3371 (SEQ_GT(tp->snd_una, th->th_ack) || 3372 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3373 rstreason = BANDLIM_RST_OPENPORT; 3374 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); 3375 goto dropwithreset; 3376 } 3377 #ifdef TCPDEBUG 3378 if (so->so_options & SO_DEBUG) 3379 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3380 &tcp_savetcp, 0); 3381 #endif 3382 TCP_PROBE3(debug__input, tp, th, m); 3383 tp->t_flags |= TF_ACKNOW; 3384 (void) tcp_output(tp); 3385 INP_WUNLOCK(tp->t_inpcb); 3386 m_freem(m); 3387 return; 3388 3389 dropwithreset: 3390 if (tp != NULL) { 3391 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3392 INP_WUNLOCK(tp->t_inpcb); 3393 } else 3394 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3395 return; 3396 3397 drop: 3398 /* 3399 * Drop space held by incoming segment and return. 3400 */ 3401 #ifdef TCPDEBUG 3402 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3403 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3404 &tcp_savetcp, 0); 3405 #endif 3406 TCP_PROBE3(debug__input, tp, th, m); 3407 if (tp != NULL) { 3408 INP_WUNLOCK(tp->t_inpcb); 3409 } 3410 m_freem(m); 3411 } 3412 3413 /* 3414 * Issue RST and make ACK acceptable to originator of segment. 3415 * The mbuf must still include the original packet header. 3416 * tp may be NULL. 3417 */ 3418 void 3419 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3420 int tlen, int rstreason) 3421 { 3422 #ifdef INET 3423 struct ip *ip; 3424 #endif 3425 #ifdef INET6 3426 struct ip6_hdr *ip6; 3427 #endif 3428 3429 if (tp != NULL) { 3430 INP_LOCK_ASSERT(tp->t_inpcb); 3431 } 3432 3433 /* Don't bother if destination was broadcast/multicast. */ 3434 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3435 goto drop; 3436 #ifdef INET6 3437 if (mtod(m, struct ip *)->ip_v == 6) { 3438 ip6 = mtod(m, struct ip6_hdr *); 3439 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3440 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3441 goto drop; 3442 /* IPv6 anycast check is done at tcp6_input() */ 3443 } 3444 #endif 3445 #if defined(INET) && defined(INET6) 3446 else 3447 #endif 3448 #ifdef INET 3449 { 3450 ip = mtod(m, struct ip *); 3451 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3452 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3453 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3454 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3455 goto drop; 3456 } 3457 #endif 3458 3459 /* Perform bandwidth limiting. */ 3460 if (badport_bandlim(rstreason) < 0) 3461 goto drop; 3462 3463 /* tcp_respond consumes the mbuf chain. */ 3464 if (tcp_get_flags(th) & TH_ACK) { 3465 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3466 th->th_ack, TH_RST); 3467 } else { 3468 if (tcp_get_flags(th) & TH_SYN) 3469 tlen++; 3470 if (tcp_get_flags(th) & TH_FIN) 3471 tlen++; 3472 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3473 (tcp_seq)0, TH_RST|TH_ACK); 3474 } 3475 return; 3476 drop: 3477 m_freem(m); 3478 } 3479 3480 /* 3481 * Parse TCP options and place in tcpopt. 3482 */ 3483 void 3484 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3485 { 3486 int opt, optlen; 3487 3488 to->to_flags = 0; 3489 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3490 opt = cp[0]; 3491 if (opt == TCPOPT_EOL) 3492 break; 3493 if (opt == TCPOPT_NOP) 3494 optlen = 1; 3495 else { 3496 if (cnt < 2) 3497 break; 3498 optlen = cp[1]; 3499 if (optlen < 2 || optlen > cnt) 3500 break; 3501 } 3502 switch (opt) { 3503 case TCPOPT_MAXSEG: 3504 if (optlen != TCPOLEN_MAXSEG) 3505 continue; 3506 if (!(flags & TO_SYN)) 3507 continue; 3508 to->to_flags |= TOF_MSS; 3509 bcopy((char *)cp + 2, 3510 (char *)&to->to_mss, sizeof(to->to_mss)); 3511 to->to_mss = ntohs(to->to_mss); 3512 break; 3513 case TCPOPT_WINDOW: 3514 if (optlen != TCPOLEN_WINDOW) 3515 continue; 3516 if (!(flags & TO_SYN)) 3517 continue; 3518 to->to_flags |= TOF_SCALE; 3519 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3520 break; 3521 case TCPOPT_TIMESTAMP: 3522 if (optlen != TCPOLEN_TIMESTAMP) 3523 continue; 3524 to->to_flags |= TOF_TS; 3525 bcopy((char *)cp + 2, 3526 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3527 to->to_tsval = ntohl(to->to_tsval); 3528 bcopy((char *)cp + 6, 3529 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3530 to->to_tsecr = ntohl(to->to_tsecr); 3531 break; 3532 case TCPOPT_SIGNATURE: 3533 /* 3534 * In order to reply to a host which has set the 3535 * TCP_SIGNATURE option in its initial SYN, we have 3536 * to record the fact that the option was observed 3537 * here for the syncache code to perform the correct 3538 * response. 3539 */ 3540 if (optlen != TCPOLEN_SIGNATURE) 3541 continue; 3542 to->to_flags |= TOF_SIGNATURE; 3543 to->to_signature = cp + 2; 3544 break; 3545 case TCPOPT_SACK_PERMITTED: 3546 if (optlen != TCPOLEN_SACK_PERMITTED) 3547 continue; 3548 if (!(flags & TO_SYN)) 3549 continue; 3550 if (!V_tcp_do_sack) 3551 continue; 3552 to->to_flags |= TOF_SACKPERM; 3553 break; 3554 case TCPOPT_SACK: 3555 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3556 continue; 3557 if (flags & TO_SYN) 3558 continue; 3559 to->to_flags |= TOF_SACK; 3560 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3561 to->to_sacks = cp + 2; 3562 TCPSTAT_INC(tcps_sack_rcv_blocks); 3563 break; 3564 case TCPOPT_FAST_OPEN: 3565 /* 3566 * Cookie length validation is performed by the 3567 * server side cookie checking code or the client 3568 * side cookie cache update code. 3569 */ 3570 if (!(flags & TO_SYN)) 3571 continue; 3572 if (!V_tcp_fastopen_client_enable && 3573 !V_tcp_fastopen_server_enable) 3574 continue; 3575 to->to_flags |= TOF_FASTOPEN; 3576 to->to_tfo_len = optlen - 2; 3577 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3578 break; 3579 default: 3580 continue; 3581 } 3582 } 3583 } 3584 3585 /* 3586 * Pull out of band byte out of a segment so 3587 * it doesn't appear in the user's data queue. 3588 * It is still reflected in the segment length for 3589 * sequencing purposes. 3590 */ 3591 void 3592 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3593 int off) 3594 { 3595 int cnt = off + th->th_urp - 1; 3596 3597 while (cnt >= 0) { 3598 if (m->m_len > cnt) { 3599 char *cp = mtod(m, caddr_t) + cnt; 3600 struct tcpcb *tp = sototcpcb(so); 3601 3602 INP_WLOCK_ASSERT(tp->t_inpcb); 3603 3604 tp->t_iobc = *cp; 3605 tp->t_oobflags |= TCPOOB_HAVEDATA; 3606 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3607 m->m_len--; 3608 if (m->m_flags & M_PKTHDR) 3609 m->m_pkthdr.len--; 3610 return; 3611 } 3612 cnt -= m->m_len; 3613 m = m->m_next; 3614 if (m == NULL) 3615 break; 3616 } 3617 panic("tcp_pulloutofband"); 3618 } 3619 3620 /* 3621 * Collect new round-trip time estimate 3622 * and update averages and current timeout. 3623 */ 3624 void 3625 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3626 { 3627 int delta; 3628 3629 INP_WLOCK_ASSERT(tp->t_inpcb); 3630 3631 TCPSTAT_INC(tcps_rttupdated); 3632 tp->t_rttupdated++; 3633 #ifdef STATS 3634 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3635 imax(0, rtt * 1000 / hz)); 3636 #endif 3637 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3638 /* 3639 * srtt is stored as fixed point with 5 bits after the 3640 * binary point (i.e., scaled by 8). The following magic 3641 * is equivalent to the smoothing algorithm in rfc793 with 3642 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3643 * point). Adjust rtt to origin 0. 3644 */ 3645 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3646 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3647 3648 if ((tp->t_srtt += delta) <= 0) 3649 tp->t_srtt = 1; 3650 3651 /* 3652 * We accumulate a smoothed rtt variance (actually, a 3653 * smoothed mean difference), then set the retransmit 3654 * timer to smoothed rtt + 4 times the smoothed variance. 3655 * rttvar is stored as fixed point with 4 bits after the 3656 * binary point (scaled by 16). The following is 3657 * equivalent to rfc793 smoothing with an alpha of .75 3658 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3659 * rfc793's wired-in beta. 3660 */ 3661 if (delta < 0) 3662 delta = -delta; 3663 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3664 if ((tp->t_rttvar += delta) <= 0) 3665 tp->t_rttvar = 1; 3666 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3667 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3668 } else { 3669 /* 3670 * No rtt measurement yet - use the unsmoothed rtt. 3671 * Set the variance to half the rtt (so our first 3672 * retransmit happens at 3*rtt). 3673 */ 3674 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3675 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3676 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3677 } 3678 tp->t_rtttime = 0; 3679 tp->t_rxtshift = 0; 3680 3681 /* 3682 * the retransmit should happen at rtt + 4 * rttvar. 3683 * Because of the way we do the smoothing, srtt and rttvar 3684 * will each average +1/2 tick of bias. When we compute 3685 * the retransmit timer, we want 1/2 tick of rounding and 3686 * 1 extra tick because of +-1/2 tick uncertainty in the 3687 * firing of the timer. The bias will give us exactly the 3688 * 1.5 tick we need. But, because the bias is 3689 * statistical, we have to test that we don't drop below 3690 * the minimum feasible timer (which is 2 ticks). 3691 */ 3692 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3693 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3694 3695 /* 3696 * We received an ack for a packet that wasn't retransmitted; 3697 * it is probably safe to discard any error indications we've 3698 * received recently. This isn't quite right, but close enough 3699 * for now (a route might have failed after we sent a segment, 3700 * and the return path might not be symmetrical). 3701 */ 3702 tp->t_softerror = 0; 3703 } 3704 3705 /* 3706 * Determine a reasonable value for maxseg size. 3707 * If the route is known, check route for mtu. 3708 * If none, use an mss that can be handled on the outgoing interface 3709 * without forcing IP to fragment. If no route is found, route has no mtu, 3710 * or the destination isn't local, use a default, hopefully conservative 3711 * size (usually 512 or the default IP max size, but no more than the mtu 3712 * of the interface), as we can't discover anything about intervening 3713 * gateways or networks. We also initialize the congestion/slow start 3714 * window to be a single segment if the destination isn't local. 3715 * While looking at the routing entry, we also initialize other path-dependent 3716 * parameters from pre-set or cached values in the routing entry. 3717 * 3718 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3719 * IP options, e.g. IPSEC data, since length of this data may vary, and 3720 * thus it is calculated for every segment separately in tcp_output(). 3721 * 3722 * NOTE that this routine is only called when we process an incoming 3723 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3724 * settings are handled in tcp_mssopt(). 3725 */ 3726 void 3727 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3728 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3729 { 3730 int mss = 0; 3731 uint32_t maxmtu = 0; 3732 struct inpcb *inp = tp->t_inpcb; 3733 struct hc_metrics_lite metrics; 3734 #ifdef INET6 3735 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3736 size_t min_protoh = isipv6 ? 3737 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3738 sizeof (struct tcpiphdr); 3739 #else 3740 size_t min_protoh = sizeof(struct tcpiphdr); 3741 #endif 3742 3743 INP_WLOCK_ASSERT(tp->t_inpcb); 3744 3745 if (tp->t_port) 3746 min_protoh += V_tcp_udp_tunneling_overhead; 3747 if (mtuoffer != -1) { 3748 KASSERT(offer == -1, ("%s: conflict", __func__)); 3749 offer = mtuoffer - min_protoh; 3750 } 3751 3752 /* Initialize. */ 3753 #ifdef INET6 3754 if (isipv6) { 3755 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3756 tp->t_maxseg = V_tcp_v6mssdflt; 3757 } 3758 #endif 3759 #if defined(INET) && defined(INET6) 3760 else 3761 #endif 3762 #ifdef INET 3763 { 3764 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3765 tp->t_maxseg = V_tcp_mssdflt; 3766 } 3767 #endif 3768 3769 /* 3770 * No route to sender, stay with default mss and return. 3771 */ 3772 if (maxmtu == 0) { 3773 /* 3774 * In case we return early we need to initialize metrics 3775 * to a defined state as tcp_hc_get() would do for us 3776 * if there was no cache hit. 3777 */ 3778 if (metricptr != NULL) 3779 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3780 return; 3781 } 3782 3783 /* What have we got? */ 3784 switch (offer) { 3785 case 0: 3786 /* 3787 * Offer == 0 means that there was no MSS on the SYN 3788 * segment, in this case we use tcp_mssdflt as 3789 * already assigned to t_maxseg above. 3790 */ 3791 offer = tp->t_maxseg; 3792 break; 3793 3794 case -1: 3795 /* 3796 * Offer == -1 means that we didn't receive SYN yet. 3797 */ 3798 /* FALLTHROUGH */ 3799 3800 default: 3801 /* 3802 * Prevent DoS attack with too small MSS. Round up 3803 * to at least minmss. 3804 */ 3805 offer = max(offer, V_tcp_minmss); 3806 } 3807 3808 /* 3809 * rmx information is now retrieved from tcp_hostcache. 3810 */ 3811 tcp_hc_get(&inp->inp_inc, &metrics); 3812 if (metricptr != NULL) 3813 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3814 3815 /* 3816 * If there's a discovered mtu in tcp hostcache, use it. 3817 * Else, use the link mtu. 3818 */ 3819 if (metrics.rmx_mtu) 3820 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3821 else { 3822 #ifdef INET6 3823 if (isipv6) { 3824 mss = maxmtu - min_protoh; 3825 if (!V_path_mtu_discovery && 3826 !in6_localaddr(&inp->in6p_faddr)) 3827 mss = min(mss, V_tcp_v6mssdflt); 3828 } 3829 #endif 3830 #if defined(INET) && defined(INET6) 3831 else 3832 #endif 3833 #ifdef INET 3834 { 3835 mss = maxmtu - min_protoh; 3836 if (!V_path_mtu_discovery && 3837 !in_localaddr(inp->inp_faddr)) 3838 mss = min(mss, V_tcp_mssdflt); 3839 } 3840 #endif 3841 /* 3842 * XXX - The above conditional (mss = maxmtu - min_protoh) 3843 * probably violates the TCP spec. 3844 * The problem is that, since we don't know the 3845 * other end's MSS, we are supposed to use a conservative 3846 * default. But, if we do that, then MTU discovery will 3847 * never actually take place, because the conservative 3848 * default is much less than the MTUs typically seen 3849 * on the Internet today. For the moment, we'll sweep 3850 * this under the carpet. 3851 * 3852 * The conservative default might not actually be a problem 3853 * if the only case this occurs is when sending an initial 3854 * SYN with options and data to a host we've never talked 3855 * to before. Then, they will reply with an MSS value which 3856 * will get recorded and the new parameters should get 3857 * recomputed. For Further Study. 3858 */ 3859 } 3860 mss = min(mss, offer); 3861 3862 /* 3863 * Sanity check: make sure that maxseg will be large 3864 * enough to allow some data on segments even if the 3865 * all the option space is used (40bytes). Otherwise 3866 * funny things may happen in tcp_output. 3867 * 3868 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3869 */ 3870 mss = max(mss, 64); 3871 3872 tp->t_maxseg = mss; 3873 } 3874 3875 void 3876 tcp_mss(struct tcpcb *tp, int offer) 3877 { 3878 int mss; 3879 uint32_t bufsize; 3880 struct inpcb *inp; 3881 struct socket *so; 3882 struct hc_metrics_lite metrics; 3883 struct tcp_ifcap cap; 3884 3885 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3886 3887 bzero(&cap, sizeof(cap)); 3888 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3889 3890 mss = tp->t_maxseg; 3891 inp = tp->t_inpcb; 3892 3893 /* 3894 * If there's a pipesize, change the socket buffer to that size, 3895 * don't change if sb_hiwat is different than default (then it 3896 * has been changed on purpose with setsockopt). 3897 * Make the socket buffers an integral number of mss units; 3898 * if the mss is larger than the socket buffer, decrease the mss. 3899 */ 3900 so = inp->inp_socket; 3901 SOCKBUF_LOCK(&so->so_snd); 3902 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3903 bufsize = metrics.rmx_sendpipe; 3904 else 3905 bufsize = so->so_snd.sb_hiwat; 3906 if (bufsize < mss) 3907 mss = bufsize; 3908 else { 3909 bufsize = roundup(bufsize, mss); 3910 if (bufsize > sb_max) 3911 bufsize = sb_max; 3912 if (bufsize > so->so_snd.sb_hiwat) 3913 (void)sbreserve_locked(so, SO_SND, bufsize, NULL); 3914 } 3915 SOCKBUF_UNLOCK(&so->so_snd); 3916 /* 3917 * Sanity check: make sure that maxseg will be large 3918 * enough to allow some data on segments even if the 3919 * all the option space is used (40bytes). Otherwise 3920 * funny things may happen in tcp_output. 3921 * 3922 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3923 */ 3924 tp->t_maxseg = max(mss, 64); 3925 3926 SOCKBUF_LOCK(&so->so_rcv); 3927 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3928 bufsize = metrics.rmx_recvpipe; 3929 else 3930 bufsize = so->so_rcv.sb_hiwat; 3931 if (bufsize > mss) { 3932 bufsize = roundup(bufsize, mss); 3933 if (bufsize > sb_max) 3934 bufsize = sb_max; 3935 if (bufsize > so->so_rcv.sb_hiwat) 3936 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); 3937 } 3938 SOCKBUF_UNLOCK(&so->so_rcv); 3939 3940 /* Check the interface for TSO capabilities. */ 3941 if (cap.ifcap & CSUM_TSO) { 3942 tp->t_flags |= TF_TSO; 3943 tp->t_tsomax = cap.tsomax; 3944 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3945 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3946 } 3947 } 3948 3949 /* 3950 * Determine the MSS option to send on an outgoing SYN. 3951 */ 3952 int 3953 tcp_mssopt(struct in_conninfo *inc) 3954 { 3955 int mss = 0; 3956 uint32_t thcmtu = 0; 3957 uint32_t maxmtu = 0; 3958 size_t min_protoh; 3959 3960 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3961 3962 #ifdef INET6 3963 if (inc->inc_flags & INC_ISIPV6) { 3964 mss = V_tcp_v6mssdflt; 3965 maxmtu = tcp_maxmtu6(inc, NULL); 3966 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3967 } 3968 #endif 3969 #if defined(INET) && defined(INET6) 3970 else 3971 #endif 3972 #ifdef INET 3973 { 3974 mss = V_tcp_mssdflt; 3975 maxmtu = tcp_maxmtu(inc, NULL); 3976 min_protoh = sizeof(struct tcpiphdr); 3977 } 3978 #endif 3979 #if defined(INET6) || defined(INET) 3980 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3981 #endif 3982 3983 if (maxmtu && thcmtu) 3984 mss = min(maxmtu, thcmtu) - min_protoh; 3985 else if (maxmtu || thcmtu) 3986 mss = max(maxmtu, thcmtu) - min_protoh; 3987 3988 return (mss); 3989 } 3990 3991 void 3992 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 3993 { 3994 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3995 int maxseg = tcp_maxseg(tp); 3996 3997 INP_WLOCK_ASSERT(tp->t_inpcb); 3998 3999 /* 4000 * Compute the amount of data that this ACK is indicating 4001 * (del_data) and an estimate of how many bytes are in the 4002 * network. 4003 */ 4004 if (tcp_is_sack_recovery(tp, to) || 4005 (IN_CONGRECOVERY(tp->t_flags) && 4006 !IN_FASTRECOVERY(tp->t_flags))) { 4007 del_data = tp->sackhint.delivered_data; 4008 if (V_tcp_do_newsack) 4009 pipe = tcp_compute_pipe(tp); 4010 else 4011 pipe = (tp->snd_nxt - tp->snd_fack) + 4012 tp->sackhint.sack_bytes_rexmit; 4013 } else { 4014 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + 4015 tp->snd_recover - tp->snd_una)) 4016 del_data = maxseg; 4017 pipe = imax(0, tp->snd_max - tp->snd_una - 4018 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); 4019 } 4020 tp->sackhint.prr_delivered += del_data; 4021 /* 4022 * Proportional Rate Reduction 4023 */ 4024 if (pipe >= tp->snd_ssthresh) { 4025 if (tp->sackhint.recover_fs == 0) 4026 tp->sackhint.recover_fs = 4027 imax(1, tp->snd_nxt - tp->snd_una); 4028 snd_cnt = howmany((long)tp->sackhint.prr_delivered * 4029 tp->snd_ssthresh, tp->sackhint.recover_fs) - 4030 tp->sackhint.prr_out; 4031 } else { 4032 if (V_tcp_do_prr_conservative || (del_data == 0)) 4033 limit = tp->sackhint.prr_delivered - 4034 tp->sackhint.prr_out; 4035 else 4036 limit = imax(tp->sackhint.prr_delivered - 4037 tp->sackhint.prr_out, del_data) + 4038 maxseg; 4039 snd_cnt = imin((tp->snd_ssthresh - pipe), limit); 4040 } 4041 snd_cnt = imax(snd_cnt, 0) / maxseg; 4042 /* 4043 * Send snd_cnt new data into the network in response to this ack. 4044 * If there is going to be a SACK retransmission, adjust snd_cwnd 4045 * accordingly. 4046 */ 4047 if (IN_FASTRECOVERY(tp->t_flags)) { 4048 if (tcp_is_sack_recovery(tp, to)) { 4049 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 4050 tp->sackhint.sack_bytes_rexmit + 4051 (snd_cnt * maxseg); 4052 } else { 4053 tp->snd_cwnd = (tp->snd_max - tp->snd_una) + 4054 (snd_cnt * maxseg); 4055 } 4056 } else if (IN_CONGRECOVERY(tp->t_flags)) 4057 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); 4058 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); 4059 } 4060 4061 /* 4062 * On a partial ack arrives, force the retransmission of the 4063 * next unacknowledged segment. Do not clear tp->t_dupacks. 4064 * By setting snd_nxt to ti_ack, this forces retransmission timer to 4065 * be started again. 4066 */ 4067 void 4068 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 4069 { 4070 tcp_seq onxt = tp->snd_nxt; 4071 uint32_t ocwnd = tp->snd_cwnd; 4072 u_int maxseg = tcp_maxseg(tp); 4073 4074 INP_WLOCK_ASSERT(tp->t_inpcb); 4075 4076 tcp_timer_activate(tp, TT_REXMT, 0); 4077 tp->t_rtttime = 0; 4078 tp->snd_nxt = th->th_ack; 4079 /* 4080 * Set snd_cwnd to one segment beyond acknowledged offset. 4081 * (tp->snd_una has not yet been updated when this function is called.) 4082 */ 4083 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 4084 tp->t_flags |= TF_ACKNOW; 4085 (void) tcp_output(tp); 4086 tp->snd_cwnd = ocwnd; 4087 if (SEQ_GT(onxt, tp->snd_nxt)) 4088 tp->snd_nxt = onxt; 4089 /* 4090 * Partial window deflation. Relies on fact that tp->snd_una 4091 * not updated yet. 4092 */ 4093 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 4094 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 4095 else 4096 tp->snd_cwnd = 0; 4097 tp->snd_cwnd += maxseg; 4098 } 4099 4100 int 4101 tcp_compute_pipe(struct tcpcb *tp) 4102 { 4103 if (tp->t_fb->tfb_compute_pipe == NULL) { 4104 return (tp->snd_max - tp->snd_una + 4105 tp->sackhint.sack_bytes_rexmit - 4106 tp->sackhint.sacked_bytes); 4107 } else { 4108 return((*tp->t_fb->tfb_compute_pipe)(tp)); 4109 } 4110 } 4111 4112 uint32_t 4113 tcp_compute_initwnd(uint32_t maxseg) 4114 { 4115 /* 4116 * Calculate the Initial Window, also used as Restart Window 4117 * 4118 * RFC5681 Section 3.1 specifies the default conservative values. 4119 * RFC3390 specifies slightly more aggressive values. 4120 * RFC6928 increases it to ten segments. 4121 * Support for user specified value for initial flight size. 4122 */ 4123 if (V_tcp_initcwnd_segments) 4124 return min(V_tcp_initcwnd_segments * maxseg, 4125 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4126 else if (V_tcp_do_rfc3390) 4127 return min(4 * maxseg, max(2 * maxseg, 4380)); 4128 else { 4129 /* Per RFC5681 Section 3.1 */ 4130 if (maxseg > 2190) 4131 return (2 * maxseg); 4132 else if (maxseg > 1095) 4133 return (3 * maxseg); 4134 else 4135 return (4 * maxseg); 4136 } 4137 } 4138