1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 __FBSDID("$FreeBSD$"); 54 55 #include "opt_inet.h" 56 #include "opt_inet6.h" 57 #include "opt_ipsec.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/arb.h> 62 #include <sys/kernel.h> 63 #ifdef TCP_HHOOK 64 #include <sys/hhook.h> 65 #endif 66 #include <sys/malloc.h> 67 #include <sys/mbuf.h> 68 #include <sys/proc.h> /* for proc0 declaration */ 69 #include <sys/protosw.h> 70 #include <sys/qmath.h> 71 #include <sys/sdt.h> 72 #include <sys/signalvar.h> 73 #include <sys/socket.h> 74 #include <sys/socketvar.h> 75 #include <sys/sysctl.h> 76 #include <sys/syslog.h> 77 #include <sys/systm.h> 78 #include <sys/stats.h> 79 80 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 81 82 #include <vm/uma.h> 83 84 #include <net/if.h> 85 #include <net/if_var.h> 86 #include <net/route.h> 87 #include <net/vnet.h> 88 89 #define TCPSTATES /* for logging */ 90 91 #include <netinet/in.h> 92 #include <netinet/in_kdtrace.h> 93 #include <netinet/in_pcb.h> 94 #include <netinet/in_systm.h> 95 #include <netinet/ip.h> 96 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 98 #include <netinet/ip_var.h> 99 #include <netinet/ip_options.h> 100 #include <netinet/ip6.h> 101 #include <netinet/icmp6.h> 102 #include <netinet6/in6_pcb.h> 103 #include <netinet6/in6_var.h> 104 #include <netinet6/ip6_var.h> 105 #include <netinet6/nd6.h> 106 #include <netinet/tcp.h> 107 #include <netinet/tcp_fsm.h> 108 #include <netinet/tcp_log_buf.h> 109 #include <netinet/tcp_seq.h> 110 #include <netinet/tcp_timer.h> 111 #include <netinet/tcp_var.h> 112 #include <netinet6/tcp6_var.h> 113 #include <netinet/tcpip.h> 114 #include <netinet/cc/cc.h> 115 #include <netinet/tcp_fastopen.h> 116 #ifdef TCPPCAP 117 #include <netinet/tcp_pcap.h> 118 #endif 119 #include <netinet/tcp_syncache.h> 120 #ifdef TCPDEBUG 121 #include <netinet/tcp_debug.h> 122 #endif /* TCPDEBUG */ 123 #ifdef TCP_OFFLOAD 124 #include <netinet/tcp_offload.h> 125 #endif 126 127 #include <netipsec/ipsec_support.h> 128 129 #include <machine/in_cksum.h> 130 131 #include <security/mac/mac_framework.h> 132 133 const int tcprexmtthresh = 3; 134 135 VNET_DEFINE(int, tcp_log_in_vain) = 0; 136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, 137 &VNET_NAME(tcp_log_in_vain), 0, 138 "Log all incoming TCP segments to closed ports"); 139 140 VNET_DEFINE(int, blackhole) = 0; 141 #define V_blackhole VNET(blackhole) 142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 143 &VNET_NAME(blackhole), 0, 144 "Do not send RST on segments to closed ports"); 145 146 VNET_DEFINE(int, tcp_delack_enabled) = 1; 147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 148 &VNET_NAME(tcp_delack_enabled), 0, 149 "Delay ACK to try and piggyback it onto a data packet"); 150 151 VNET_DEFINE(int, drop_synfin) = 0; 152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 153 &VNET_NAME(drop_synfin), 0, 154 "Drop TCP packets with SYN+FIN set"); 155 156 VNET_DEFINE(int, tcp_do_prr_conservative) = 0; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr_conservative, CTLFLAG_VNET | CTLFLAG_RW, 158 &VNET_NAME(tcp_do_prr_conservative), 0, 159 "Do conservative Proportional Rate Reduction"); 160 161 VNET_DEFINE(int, tcp_do_prr) = 1; 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, 163 &VNET_NAME(tcp_do_prr), 1, 164 "Enable Proportional Rate Reduction per RFC 6937"); 165 166 VNET_DEFINE(int, tcp_do_newcwv) = 0; 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, 168 &VNET_NAME(tcp_do_newcwv), 0, 169 "Enable New Congestion Window Validation per RFC7661"); 170 171 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 173 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 174 "Use calculated pipe/in-flight bytes per RFC 6675"); 175 176 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 178 &VNET_NAME(tcp_do_rfc3042), 0, 179 "Enable RFC 3042 (Limited Transmit)"); 180 181 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 183 &VNET_NAME(tcp_do_rfc3390), 0, 184 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 185 186 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 188 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 189 "Slow-start flight size (initial congestion window) in number of segments"); 190 191 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_do_rfc3465), 0, 194 "Enable RFC 3465 (Appropriate Byte Counting)"); 195 196 VNET_DEFINE(int, tcp_abc_l_var) = 2; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_abc_l_var), 2, 199 "Cap the max cwnd increment during slow-start to this number of segments"); 200 201 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, 202 CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 203 "TCP ECN"); 204 205 VNET_DEFINE(int, tcp_do_ecn) = 2; 206 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(tcp_do_ecn), 0, 208 "TCP ECN support"); 209 210 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 211 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_ecn_maxretries), 0, 213 "Max retries before giving up on ECN"); 214 215 VNET_DEFINE(int, tcp_insecure_syn) = 0; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_insecure_syn), 0, 218 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 219 220 VNET_DEFINE(int, tcp_insecure_rst) = 0; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_insecure_rst), 0, 223 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 224 225 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 226 #define V_tcp_recvspace VNET(tcp_recvspace) 227 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 229 230 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 232 &VNET_NAME(tcp_do_autorcvbuf), 0, 233 "Enable automatic receive buffer sizing"); 234 235 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 237 &VNET_NAME(tcp_autorcvbuf_max), 0, 238 "Max size of automatic receive buffer"); 239 240 VNET_DEFINE(struct inpcbhead, tcb); 241 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 242 VNET_DEFINE(struct inpcbinfo, tcbinfo); 243 244 /* 245 * TCP statistics are stored in an array of counter(9)s, which size matches 246 * size of struct tcpstat. TCP running connection count is a regular array. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 250 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 251 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 252 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 253 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 254 "TCP connection counts by TCP state"); 255 256 static void 257 tcp_vnet_init(const void *unused) 258 { 259 260 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 261 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 262 } 263 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 264 tcp_vnet_init, NULL); 265 266 #ifdef VIMAGE 267 static void 268 tcp_vnet_uninit(const void *unused) 269 { 270 271 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 272 VNET_PCPUSTAT_FREE(tcpstat); 273 } 274 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 275 tcp_vnet_uninit, NULL); 276 #endif /* VIMAGE */ 277 278 /* 279 * Kernel module interface for updating tcpstat. The first argument is an index 280 * into tcpstat treated as an array. 281 */ 282 void 283 kmod_tcpstat_add(int statnum, int val) 284 { 285 286 counter_u64_add(VNET(tcpstat)[statnum], val); 287 } 288 289 #ifdef TCP_HHOOK 290 /* 291 * Wrapper for the TCP established input helper hook. 292 */ 293 void 294 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 295 { 296 struct tcp_hhook_data hhook_data; 297 298 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 299 hhook_data.tp = tp; 300 hhook_data.th = th; 301 hhook_data.to = to; 302 303 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 304 tp->osd); 305 } 306 } 307 #endif 308 309 /* 310 * CC wrapper hook functions 311 */ 312 void 313 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 314 uint16_t type) 315 { 316 #ifdef STATS 317 int32_t gput; 318 #endif 319 320 INP_WLOCK_ASSERT(tp->t_inpcb); 321 322 tp->ccv->nsegs = nsegs; 323 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 324 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || 325 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && 326 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) 327 tp->ccv->flags |= CCF_CWND_LIMITED; 328 else 329 tp->ccv->flags &= ~CCF_CWND_LIMITED; 330 331 if (type == CC_ACK) { 332 #ifdef STATS 333 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, 334 ((int32_t)tp->snd_cwnd) - tp->snd_wnd); 335 if (!IN_RECOVERY(tp->t_flags)) 336 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, 337 tp->ccv->bytes_this_ack / (tcp_maxseg(tp) * nsegs)); 338 if ((tp->t_flags & TF_GPUTINPROG) && 339 SEQ_GEQ(th->th_ack, tp->gput_ack)) { 340 /* 341 * Compute goodput in bits per millisecond. 342 */ 343 gput = (((int64_t)(th->th_ack - tp->gput_seq)) << 3) / 344 max(1, tcp_ts_getticks() - tp->gput_ts); 345 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, 346 gput); 347 /* 348 * XXXLAS: This is a temporary hack, and should be 349 * chained off VOI_TCP_GPUT when stats(9) grows an API 350 * to deal with chained VOIs. 351 */ 352 if (tp->t_stats_gput_prev > 0) 353 stats_voi_update_abs_s32(tp->t_stats, 354 VOI_TCP_GPUT_ND, 355 ((gput - tp->t_stats_gput_prev) * 100) / 356 tp->t_stats_gput_prev); 357 tp->t_flags &= ~TF_GPUTINPROG; 358 tp->t_stats_gput_prev = gput; 359 } 360 #endif /* STATS */ 361 if (tp->snd_cwnd > tp->snd_ssthresh) { 362 tp->t_bytes_acked += tp->ccv->bytes_this_ack; 363 if (tp->t_bytes_acked >= tp->snd_cwnd) { 364 tp->t_bytes_acked -= tp->snd_cwnd; 365 tp->ccv->flags |= CCF_ABC_SENTAWND; 366 } 367 } else { 368 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 369 tp->t_bytes_acked = 0; 370 } 371 } 372 373 if (CC_ALGO(tp)->ack_received != NULL) { 374 /* XXXLAS: Find a way to live without this */ 375 tp->ccv->curack = th->th_ack; 376 CC_ALGO(tp)->ack_received(tp->ccv, type); 377 } 378 #ifdef STATS 379 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); 380 #endif 381 } 382 383 void 384 cc_conn_init(struct tcpcb *tp) 385 { 386 struct hc_metrics_lite metrics; 387 struct inpcb *inp = tp->t_inpcb; 388 u_int maxseg; 389 int rtt; 390 391 INP_WLOCK_ASSERT(tp->t_inpcb); 392 393 tcp_hc_get(&inp->inp_inc, &metrics); 394 maxseg = tcp_maxseg(tp); 395 396 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 397 tp->t_srtt = rtt; 398 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 399 TCPSTAT_INC(tcps_usedrtt); 400 if (metrics.rmx_rttvar) { 401 tp->t_rttvar = metrics.rmx_rttvar; 402 TCPSTAT_INC(tcps_usedrttvar); 403 } else { 404 /* default variation is +- 1 rtt */ 405 tp->t_rttvar = 406 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 407 } 408 TCPT_RANGESET(tp->t_rxtcur, 409 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 410 tp->t_rttmin, TCPTV_REXMTMAX); 411 } 412 if (metrics.rmx_ssthresh) { 413 /* 414 * There's some sort of gateway or interface 415 * buffer limit on the path. Use this to set 416 * the slow start threshold, but set the 417 * threshold to no less than 2*mss. 418 */ 419 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 420 TCPSTAT_INC(tcps_usedssthresh); 421 } 422 423 /* 424 * Set the initial slow-start flight size. 425 * 426 * If a SYN or SYN/ACK was lost and retransmitted, we have to 427 * reduce the initial CWND to one segment as congestion is likely 428 * requiring us to be cautious. 429 */ 430 if (tp->snd_cwnd == 1) 431 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 432 else 433 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 434 435 if (CC_ALGO(tp)->conn_init != NULL) 436 CC_ALGO(tp)->conn_init(tp->ccv); 437 } 438 439 void inline 440 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 441 { 442 INP_WLOCK_ASSERT(tp->t_inpcb); 443 444 #ifdef STATS 445 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); 446 #endif 447 448 switch(type) { 449 case CC_NDUPACK: 450 if (!IN_FASTRECOVERY(tp->t_flags)) { 451 tp->snd_recover = tp->snd_max; 452 if (tp->t_flags2 & TF2_ECN_PERMIT) 453 tp->t_flags2 |= TF2_ECN_SND_CWR; 454 } 455 break; 456 case CC_ECN: 457 if (!IN_CONGRECOVERY(tp->t_flags) || 458 /* 459 * Allow ECN reaction on ACK to CWR, if 460 * that data segment was also CE marked. 461 */ 462 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 463 EXIT_CONGRECOVERY(tp->t_flags); 464 TCPSTAT_INC(tcps_ecn_rcwnd); 465 tp->snd_recover = tp->snd_max + 1; 466 if (tp->t_flags2 & TF2_ECN_PERMIT) 467 tp->t_flags2 |= TF2_ECN_SND_CWR; 468 } 469 break; 470 case CC_RTO: 471 tp->t_dupacks = 0; 472 tp->t_bytes_acked = 0; 473 EXIT_RECOVERY(tp->t_flags); 474 if (tp->t_flags2 & TF2_ECN_PERMIT) 475 tp->t_flags2 |= TF2_ECN_SND_CWR; 476 break; 477 case CC_RTO_ERR: 478 TCPSTAT_INC(tcps_sndrexmitbad); 479 /* RTO was unnecessary, so reset everything. */ 480 tp->snd_cwnd = tp->snd_cwnd_prev; 481 tp->snd_ssthresh = tp->snd_ssthresh_prev; 482 tp->snd_recover = tp->snd_recover_prev; 483 if (tp->t_flags & TF_WASFRECOVERY) 484 ENTER_FASTRECOVERY(tp->t_flags); 485 if (tp->t_flags & TF_WASCRECOVERY) 486 ENTER_CONGRECOVERY(tp->t_flags); 487 tp->snd_nxt = tp->snd_max; 488 tp->t_flags &= ~TF_PREVVALID; 489 tp->t_badrxtwin = 0; 490 break; 491 } 492 493 if (CC_ALGO(tp)->cong_signal != NULL) { 494 if (th != NULL) 495 tp->ccv->curack = th->th_ack; 496 CC_ALGO(tp)->cong_signal(tp->ccv, type); 497 } 498 } 499 500 void inline 501 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 502 { 503 INP_WLOCK_ASSERT(tp->t_inpcb); 504 505 /* XXXLAS: KASSERT that we're in recovery? */ 506 507 if (CC_ALGO(tp)->post_recovery != NULL) { 508 tp->ccv->curack = th->th_ack; 509 CC_ALGO(tp)->post_recovery(tp->ccv); 510 } 511 /* XXXLAS: EXIT_RECOVERY ? */ 512 tp->t_bytes_acked = 0; 513 } 514 515 /* 516 * Indicate whether this ack should be delayed. We can delay the ack if 517 * following conditions are met: 518 * - There is no delayed ack timer in progress. 519 * - Our last ack wasn't a 0-sized window. We never want to delay 520 * the ack that opens up a 0-sized window. 521 * - LRO wasn't used for this segment. We make sure by checking that the 522 * segment size is not larger than the MSS. 523 */ 524 #define DELAY_ACK(tp, tlen) \ 525 ((!tcp_timer_active(tp, TT_DELACK) && \ 526 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 527 (tlen <= tp->t_maxseg) && \ 528 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 529 530 void inline 531 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 532 { 533 INP_WLOCK_ASSERT(tp->t_inpcb); 534 535 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 536 switch (iptos & IPTOS_ECN_MASK) { 537 case IPTOS_ECN_CE: 538 tp->ccv->flags |= CCF_IPHDR_CE; 539 break; 540 case IPTOS_ECN_ECT0: 541 /* FALLTHROUGH */ 542 case IPTOS_ECN_ECT1: 543 /* FALLTHROUGH */ 544 case IPTOS_ECN_NOTECT: 545 tp->ccv->flags &= ~CCF_IPHDR_CE; 546 break; 547 } 548 549 if (th->th_flags & TH_CWR) 550 tp->ccv->flags |= CCF_TCPHDR_CWR; 551 else 552 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 553 554 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 555 556 if (tp->ccv->flags & CCF_ACKNOW) { 557 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 558 tp->t_flags |= TF_ACKNOW; 559 } 560 } 561 } 562 563 /* 564 * TCP input handling is split into multiple parts: 565 * tcp6_input is a thin wrapper around tcp_input for the extended 566 * ip6_protox[] call format in ip6_input 567 * tcp_input handles primary segment validation, inpcb lookup and 568 * SYN processing on listen sockets 569 * tcp_do_segment processes the ACK and text of the segment for 570 * establishing, established and closing connections 571 */ 572 #ifdef INET6 573 int 574 tcp6_input(struct mbuf **mp, int *offp, int proto) 575 { 576 struct mbuf *m; 577 struct in6_ifaddr *ia6; 578 struct ip6_hdr *ip6; 579 580 m = *mp; 581 if (m->m_len < *offp + sizeof(struct tcphdr)) { 582 m = m_pullup(m, *offp + sizeof(struct tcphdr)); 583 if (m == NULL) { 584 *mp = m; 585 TCPSTAT_INC(tcps_rcvshort); 586 return (IPPROTO_DONE); 587 } 588 } 589 590 /* 591 * draft-itojun-ipv6-tcp-to-anycast 592 * better place to put this in? 593 */ 594 ip6 = mtod(m, struct ip6_hdr *); 595 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 596 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 597 ifa_free(&ia6->ia_ifa); 598 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 599 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 600 *mp = NULL; 601 return (IPPROTO_DONE); 602 } 603 if (ia6) 604 ifa_free(&ia6->ia_ifa); 605 606 *mp = m; 607 return (tcp_input(mp, offp, proto)); 608 } 609 #endif /* INET6 */ 610 611 int 612 tcp_input(struct mbuf **mp, int *offp, int proto) 613 { 614 struct mbuf *m = *mp; 615 struct tcphdr *th = NULL; 616 struct ip *ip = NULL; 617 struct inpcb *inp = NULL; 618 struct tcpcb *tp = NULL; 619 struct socket *so = NULL; 620 u_char *optp = NULL; 621 int off0; 622 int optlen = 0; 623 #ifdef INET 624 int len; 625 uint8_t ipttl; 626 #endif 627 int tlen = 0, off; 628 int drop_hdrlen; 629 int thflags; 630 int rstreason = 0; /* For badport_bandlim accounting purposes */ 631 uint8_t iptos; 632 struct m_tag *fwd_tag = NULL; 633 #ifdef INET6 634 struct ip6_hdr *ip6 = NULL; 635 int isipv6; 636 #else 637 const void *ip6 = NULL; 638 #endif /* INET6 */ 639 struct tcpopt to; /* options in this segment */ 640 char *s = NULL; /* address and port logging */ 641 #ifdef TCPDEBUG 642 /* 643 * The size of tcp_saveipgen must be the size of the max ip header, 644 * now IPv6. 645 */ 646 u_char tcp_saveipgen[IP6_HDR_LEN]; 647 struct tcphdr tcp_savetcp; 648 short ostate = 0; 649 #endif 650 651 NET_EPOCH_ASSERT(); 652 653 #ifdef INET6 654 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 655 #endif 656 657 off0 = *offp; 658 m = *mp; 659 *mp = NULL; 660 to.to_flags = 0; 661 TCPSTAT_INC(tcps_rcvtotal); 662 663 #ifdef INET6 664 if (isipv6) { 665 ip6 = mtod(m, struct ip6_hdr *); 666 th = (struct tcphdr *)((caddr_t)ip6 + off0); 667 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 668 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 669 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 670 th->th_sum = m->m_pkthdr.csum_data; 671 else 672 th->th_sum = in6_cksum_pseudo(ip6, tlen, 673 IPPROTO_TCP, m->m_pkthdr.csum_data); 674 th->th_sum ^= 0xffff; 675 } else 676 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 677 if (th->th_sum) { 678 TCPSTAT_INC(tcps_rcvbadsum); 679 goto drop; 680 } 681 682 /* 683 * Be proactive about unspecified IPv6 address in source. 684 * As we use all-zero to indicate unbounded/unconnected pcb, 685 * unspecified IPv6 address can be used to confuse us. 686 * 687 * Note that packets with unspecified IPv6 destination is 688 * already dropped in ip6_input. 689 */ 690 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 691 /* XXX stat */ 692 goto drop; 693 } 694 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 695 } 696 #endif 697 #if defined(INET) && defined(INET6) 698 else 699 #endif 700 #ifdef INET 701 { 702 /* 703 * Get IP and TCP header together in first mbuf. 704 * Note: IP leaves IP header in first mbuf. 705 */ 706 if (off0 > sizeof (struct ip)) { 707 ip_stripoptions(m); 708 off0 = sizeof(struct ip); 709 } 710 if (m->m_len < sizeof (struct tcpiphdr)) { 711 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 712 == NULL) { 713 TCPSTAT_INC(tcps_rcvshort); 714 return (IPPROTO_DONE); 715 } 716 } 717 ip = mtod(m, struct ip *); 718 th = (struct tcphdr *)((caddr_t)ip + off0); 719 tlen = ntohs(ip->ip_len) - off0; 720 721 iptos = ip->ip_tos; 722 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 723 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 724 th->th_sum = m->m_pkthdr.csum_data; 725 else 726 th->th_sum = in_pseudo(ip->ip_src.s_addr, 727 ip->ip_dst.s_addr, 728 htonl(m->m_pkthdr.csum_data + tlen + 729 IPPROTO_TCP)); 730 th->th_sum ^= 0xffff; 731 } else { 732 struct ipovly *ipov = (struct ipovly *)ip; 733 734 /* 735 * Checksum extended TCP header and data. 736 */ 737 len = off0 + tlen; 738 ipttl = ip->ip_ttl; 739 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 740 ipov->ih_len = htons(tlen); 741 th->th_sum = in_cksum(m, len); 742 /* Reset length for SDT probes. */ 743 ip->ip_len = htons(len); 744 /* Reset TOS bits */ 745 ip->ip_tos = iptos; 746 /* Re-initialization for later version check */ 747 ip->ip_ttl = ipttl; 748 ip->ip_v = IPVERSION; 749 ip->ip_hl = off0 >> 2; 750 } 751 752 if (th->th_sum) { 753 TCPSTAT_INC(tcps_rcvbadsum); 754 goto drop; 755 } 756 } 757 #endif /* INET */ 758 759 /* 760 * Check that TCP offset makes sense, 761 * pull out TCP options and adjust length. XXX 762 */ 763 off = th->th_off << 2; 764 if (off < sizeof (struct tcphdr) || off > tlen) { 765 TCPSTAT_INC(tcps_rcvbadoff); 766 goto drop; 767 } 768 tlen -= off; /* tlen is used instead of ti->ti_len */ 769 if (off > sizeof (struct tcphdr)) { 770 #ifdef INET6 771 if (isipv6) { 772 if (m->m_len < off0 + off) { 773 m = m_pullup(m, off0 + off); 774 if (m == NULL) { 775 TCPSTAT_INC(tcps_rcvshort); 776 return (IPPROTO_DONE); 777 } 778 } 779 ip6 = mtod(m, struct ip6_hdr *); 780 th = (struct tcphdr *)((caddr_t)ip6 + off0); 781 } 782 #endif 783 #if defined(INET) && defined(INET6) 784 else 785 #endif 786 #ifdef INET 787 { 788 if (m->m_len < sizeof(struct ip) + off) { 789 if ((m = m_pullup(m, sizeof (struct ip) + off)) 790 == NULL) { 791 TCPSTAT_INC(tcps_rcvshort); 792 return (IPPROTO_DONE); 793 } 794 ip = mtod(m, struct ip *); 795 th = (struct tcphdr *)((caddr_t)ip + off0); 796 } 797 } 798 #endif 799 optlen = off - sizeof (struct tcphdr); 800 optp = (u_char *)(th + 1); 801 } 802 thflags = th->th_flags; 803 804 /* 805 * Convert TCP protocol specific fields to host format. 806 */ 807 tcp_fields_to_host(th); 808 809 /* 810 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 811 */ 812 drop_hdrlen = off0 + off; 813 814 /* 815 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 816 */ 817 if ( 818 #ifdef INET6 819 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 820 #ifdef INET 821 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 822 #endif 823 #endif 824 #if defined(INET) && !defined(INET6) 825 (m->m_flags & M_IP_NEXTHOP) 826 #endif 827 ) 828 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 829 830 findpcb: 831 #ifdef INET6 832 if (isipv6 && fwd_tag != NULL) { 833 struct sockaddr_in6 *next_hop6; 834 835 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 836 /* 837 * Transparently forwarded. Pretend to be the destination. 838 * Already got one like this? 839 */ 840 inp = in6_pcblookup_mbuf(&V_tcbinfo, 841 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 842 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 843 if (!inp) { 844 /* 845 * It's new. Try to find the ambushing socket. 846 * Because we've rewritten the destination address, 847 * any hardware-generated hash is ignored. 848 */ 849 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 850 th->th_sport, &next_hop6->sin6_addr, 851 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 852 th->th_dport, INPLOOKUP_WILDCARD | 853 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 854 } 855 } else if (isipv6) { 856 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 857 th->th_sport, &ip6->ip6_dst, th->th_dport, 858 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 859 m->m_pkthdr.rcvif, m); 860 } 861 #endif /* INET6 */ 862 #if defined(INET6) && defined(INET) 863 else 864 #endif 865 #ifdef INET 866 if (fwd_tag != NULL) { 867 struct sockaddr_in *next_hop; 868 869 next_hop = (struct sockaddr_in *)(fwd_tag+1); 870 /* 871 * Transparently forwarded. Pretend to be the destination. 872 * already got one like this? 873 */ 874 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 875 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 876 m->m_pkthdr.rcvif, m); 877 if (!inp) { 878 /* 879 * It's new. Try to find the ambushing socket. 880 * Because we've rewritten the destination address, 881 * any hardware-generated hash is ignored. 882 */ 883 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 884 th->th_sport, next_hop->sin_addr, 885 next_hop->sin_port ? ntohs(next_hop->sin_port) : 886 th->th_dport, INPLOOKUP_WILDCARD | 887 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 888 } 889 } else 890 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 891 th->th_sport, ip->ip_dst, th->th_dport, 892 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 893 m->m_pkthdr.rcvif, m); 894 #endif /* INET */ 895 896 /* 897 * If the INPCB does not exist then all data in the incoming 898 * segment is discarded and an appropriate RST is sent back. 899 * XXX MRT Send RST using which routing table? 900 */ 901 if (inp == NULL) { 902 /* 903 * Log communication attempts to ports that are not 904 * in use. 905 */ 906 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 907 V_tcp_log_in_vain == 2) { 908 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 909 log(LOG_INFO, "%s; %s: Connection attempt " 910 "to closed port\n", s, __func__); 911 } 912 /* 913 * When blackholing do not respond with a RST but 914 * completely ignore the segment and drop it. 915 */ 916 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 917 V_blackhole == 2) 918 goto dropunlock; 919 920 rstreason = BANDLIM_RST_CLOSEDPORT; 921 goto dropwithreset; 922 } 923 INP_WLOCK_ASSERT(inp); 924 /* 925 * While waiting for inp lock during the lookup, another thread 926 * can have dropped the inpcb, in which case we need to loop back 927 * and try to find a new inpcb to deliver to. 928 */ 929 if (inp->inp_flags & INP_DROPPED) { 930 INP_WUNLOCK(inp); 931 inp = NULL; 932 goto findpcb; 933 } 934 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 935 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 936 ((inp->inp_socket == NULL) || 937 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 938 inp->inp_flowid = m->m_pkthdr.flowid; 939 inp->inp_flowtype = M_HASHTYPE_GET(m); 940 } 941 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 942 #ifdef INET6 943 if (isipv6 && IPSEC_ENABLED(ipv6) && 944 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 945 goto dropunlock; 946 } 947 #ifdef INET 948 else 949 #endif 950 #endif /* INET6 */ 951 #ifdef INET 952 if (IPSEC_ENABLED(ipv4) && 953 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 954 goto dropunlock; 955 } 956 #endif /* INET */ 957 #endif /* IPSEC */ 958 959 /* 960 * Check the minimum TTL for socket. 961 */ 962 if (inp->inp_ip_minttl != 0) { 963 #ifdef INET6 964 if (isipv6) { 965 if (inp->inp_ip_minttl > ip6->ip6_hlim) 966 goto dropunlock; 967 } else 968 #endif 969 if (inp->inp_ip_minttl > ip->ip_ttl) 970 goto dropunlock; 971 } 972 973 /* 974 * A previous connection in TIMEWAIT state is supposed to catch stray 975 * or duplicate segments arriving late. If this segment was a 976 * legitimate new connection attempt, the old INPCB gets removed and 977 * we can try again to find a listening socket. 978 * 979 * At this point, due to earlier optimism, we may hold only an inpcb 980 * lock, and not the inpcbinfo write lock. If so, we need to try to 981 * acquire it, or if that fails, acquire a reference on the inpcb, 982 * drop all locks, acquire a global write lock, and then re-acquire 983 * the inpcb lock. We may at that point discover that another thread 984 * has tried to free the inpcb, in which case we need to loop back 985 * and try to find a new inpcb to deliver to. 986 * 987 * XXXRW: It may be time to rethink timewait locking. 988 */ 989 if (inp->inp_flags & INP_TIMEWAIT) { 990 tcp_dooptions(&to, optp, optlen, 991 (thflags & TH_SYN) ? TO_SYN : 0); 992 /* 993 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 994 */ 995 if (tcp_twcheck(inp, &to, th, m, tlen)) 996 goto findpcb; 997 return (IPPROTO_DONE); 998 } 999 /* 1000 * The TCPCB may no longer exist if the connection is winding 1001 * down or it is in the CLOSED state. Either way we drop the 1002 * segment and send an appropriate response. 1003 */ 1004 tp = intotcpcb(inp); 1005 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1006 rstreason = BANDLIM_RST_CLOSEDPORT; 1007 goto dropwithreset; 1008 } 1009 1010 #ifdef TCP_OFFLOAD 1011 if (tp->t_flags & TF_TOE) { 1012 tcp_offload_input(tp, m); 1013 m = NULL; /* consumed by the TOE driver */ 1014 goto dropunlock; 1015 } 1016 #endif 1017 1018 #ifdef MAC 1019 INP_WLOCK_ASSERT(inp); 1020 if (mac_inpcb_check_deliver(inp, m)) 1021 goto dropunlock; 1022 #endif 1023 so = inp->inp_socket; 1024 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1025 #ifdef TCPDEBUG 1026 if (so->so_options & SO_DEBUG) { 1027 ostate = tp->t_state; 1028 #ifdef INET6 1029 if (isipv6) { 1030 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1031 } else 1032 #endif 1033 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1034 tcp_savetcp = *th; 1035 } 1036 #endif /* TCPDEBUG */ 1037 /* 1038 * When the socket is accepting connections (the INPCB is in LISTEN 1039 * state) we look into the SYN cache if this is a new connection 1040 * attempt or the completion of a previous one. 1041 */ 1042 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1043 ("%s: so accepting but tp %p not listening", __func__, tp)); 1044 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1045 struct in_conninfo inc; 1046 1047 bzero(&inc, sizeof(inc)); 1048 #ifdef INET6 1049 if (isipv6) { 1050 inc.inc_flags |= INC_ISIPV6; 1051 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1052 inc.inc_flags |= INC_IPV6MINMTU; 1053 inc.inc6_faddr = ip6->ip6_src; 1054 inc.inc6_laddr = ip6->ip6_dst; 1055 } else 1056 #endif 1057 { 1058 inc.inc_faddr = ip->ip_src; 1059 inc.inc_laddr = ip->ip_dst; 1060 } 1061 inc.inc_fport = th->th_sport; 1062 inc.inc_lport = th->th_dport; 1063 inc.inc_fibnum = so->so_fibnum; 1064 1065 /* 1066 * Check for an existing connection attempt in syncache if 1067 * the flag is only ACK. A successful lookup creates a new 1068 * socket appended to the listen queue in SYN_RECEIVED state. 1069 */ 1070 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1071 /* 1072 * Parse the TCP options here because 1073 * syncookies need access to the reflected 1074 * timestamp. 1075 */ 1076 tcp_dooptions(&to, optp, optlen, 0); 1077 /* 1078 * NB: syncache_expand() doesn't unlock 1079 * inp and tcpinfo locks. 1080 */ 1081 rstreason = syncache_expand(&inc, &to, th, &so, m); 1082 if (rstreason < 0) { 1083 /* 1084 * A failing TCP MD5 signature comparison 1085 * must result in the segment being dropped 1086 * and must not produce any response back 1087 * to the sender. 1088 */ 1089 goto dropunlock; 1090 } else if (rstreason == 0) { 1091 /* 1092 * No syncache entry or ACK was not 1093 * for our SYN/ACK. Send a RST. 1094 * NB: syncache did its own logging 1095 * of the failure cause. 1096 */ 1097 rstreason = BANDLIM_RST_OPENPORT; 1098 goto dropwithreset; 1099 } 1100 tfo_socket_result: 1101 if (so == NULL) { 1102 /* 1103 * We completed the 3-way handshake 1104 * but could not allocate a socket 1105 * either due to memory shortage, 1106 * listen queue length limits or 1107 * global socket limits. Send RST 1108 * or wait and have the remote end 1109 * retransmit the ACK for another 1110 * try. 1111 */ 1112 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1113 log(LOG_DEBUG, "%s; %s: Listen socket: " 1114 "Socket allocation failed due to " 1115 "limits or memory shortage, %s\n", 1116 s, __func__, 1117 V_tcp_sc_rst_sock_fail ? 1118 "sending RST" : "try again"); 1119 if (V_tcp_sc_rst_sock_fail) { 1120 rstreason = BANDLIM_UNLIMITED; 1121 goto dropwithreset; 1122 } else 1123 goto dropunlock; 1124 } 1125 /* 1126 * Socket is created in state SYN_RECEIVED. 1127 * Unlock the listen socket, lock the newly 1128 * created socket and update the tp variable. 1129 */ 1130 INP_WUNLOCK(inp); /* listen socket */ 1131 inp = sotoinpcb(so); 1132 /* 1133 * New connection inpcb is already locked by 1134 * syncache_expand(). 1135 */ 1136 INP_WLOCK_ASSERT(inp); 1137 tp = intotcpcb(inp); 1138 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1139 ("%s: ", __func__)); 1140 /* 1141 * Process the segment and the data it 1142 * contains. tcp_do_segment() consumes 1143 * the mbuf chain and unlocks the inpcb. 1144 */ 1145 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1146 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1147 iptos); 1148 return (IPPROTO_DONE); 1149 } 1150 /* 1151 * Segment flag validation for new connection attempts: 1152 * 1153 * Our (SYN|ACK) response was rejected. 1154 * Check with syncache and remove entry to prevent 1155 * retransmits. 1156 * 1157 * NB: syncache_chkrst does its own logging of failure 1158 * causes. 1159 */ 1160 if (thflags & TH_RST) { 1161 syncache_chkrst(&inc, th, m); 1162 goto dropunlock; 1163 } 1164 /* 1165 * We can't do anything without SYN. 1166 */ 1167 if ((thflags & TH_SYN) == 0) { 1168 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1169 log(LOG_DEBUG, "%s; %s: Listen socket: " 1170 "SYN is missing, segment ignored\n", 1171 s, __func__); 1172 TCPSTAT_INC(tcps_badsyn); 1173 goto dropunlock; 1174 } 1175 /* 1176 * (SYN|ACK) is bogus on a listen socket. 1177 */ 1178 if (thflags & TH_ACK) { 1179 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1180 log(LOG_DEBUG, "%s; %s: Listen socket: " 1181 "SYN|ACK invalid, segment rejected\n", 1182 s, __func__); 1183 syncache_badack(&inc); /* XXX: Not needed! */ 1184 TCPSTAT_INC(tcps_badsyn); 1185 rstreason = BANDLIM_RST_OPENPORT; 1186 goto dropwithreset; 1187 } 1188 /* 1189 * If the drop_synfin option is enabled, drop all 1190 * segments with both the SYN and FIN bits set. 1191 * This prevents e.g. nmap from identifying the 1192 * TCP/IP stack. 1193 * XXX: Poor reasoning. nmap has other methods 1194 * and is constantly refining its stack detection 1195 * strategies. 1196 * XXX: This is a violation of the TCP specification 1197 * and was used by RFC1644. 1198 */ 1199 if ((thflags & TH_FIN) && V_drop_synfin) { 1200 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1201 log(LOG_DEBUG, "%s; %s: Listen socket: " 1202 "SYN|FIN segment ignored (based on " 1203 "sysctl setting)\n", s, __func__); 1204 TCPSTAT_INC(tcps_badsyn); 1205 goto dropunlock; 1206 } 1207 /* 1208 * Segment's flags are (SYN) or (SYN|FIN). 1209 * 1210 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1211 * as they do not affect the state of the TCP FSM. 1212 * The data pointed to by TH_URG and th_urp is ignored. 1213 */ 1214 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1215 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1216 KASSERT(thflags & (TH_SYN), 1217 ("%s: Listen socket: TH_SYN not set", __func__)); 1218 #ifdef INET6 1219 /* 1220 * If deprecated address is forbidden, 1221 * we do not accept SYN to deprecated interface 1222 * address to prevent any new inbound connection from 1223 * getting established. 1224 * When we do not accept SYN, we send a TCP RST, 1225 * with deprecated source address (instead of dropping 1226 * it). We compromise it as it is much better for peer 1227 * to send a RST, and RST will be the final packet 1228 * for the exchange. 1229 * 1230 * If we do not forbid deprecated addresses, we accept 1231 * the SYN packet. RFC2462 does not suggest dropping 1232 * SYN in this case. 1233 * If we decipher RFC2462 5.5.4, it says like this: 1234 * 1. use of deprecated addr with existing 1235 * communication is okay - "SHOULD continue to be 1236 * used" 1237 * 2. use of it with new communication: 1238 * (2a) "SHOULD NOT be used if alternate address 1239 * with sufficient scope is available" 1240 * (2b) nothing mentioned otherwise. 1241 * Here we fall into (2b) case as we have no choice in 1242 * our source address selection - we must obey the peer. 1243 * 1244 * The wording in RFC2462 is confusing, and there are 1245 * multiple description text for deprecated address 1246 * handling - worse, they are not exactly the same. 1247 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1248 */ 1249 if (isipv6 && !V_ip6_use_deprecated) { 1250 struct in6_ifaddr *ia6; 1251 1252 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1253 if (ia6 != NULL && 1254 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1255 ifa_free(&ia6->ia_ifa); 1256 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1257 log(LOG_DEBUG, "%s; %s: Listen socket: " 1258 "Connection attempt to deprecated " 1259 "IPv6 address rejected\n", 1260 s, __func__); 1261 rstreason = BANDLIM_RST_OPENPORT; 1262 goto dropwithreset; 1263 } 1264 if (ia6) 1265 ifa_free(&ia6->ia_ifa); 1266 } 1267 #endif /* INET6 */ 1268 /* 1269 * Basic sanity checks on incoming SYN requests: 1270 * Don't respond if the destination is a link layer 1271 * broadcast according to RFC1122 4.2.3.10, p. 104. 1272 * If it is from this socket it must be forged. 1273 * Don't respond if the source or destination is a 1274 * global or subnet broad- or multicast address. 1275 * Note that it is quite possible to receive unicast 1276 * link-layer packets with a broadcast IP address. Use 1277 * in_broadcast() to find them. 1278 */ 1279 if (m->m_flags & (M_BCAST|M_MCAST)) { 1280 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1281 log(LOG_DEBUG, "%s; %s: Listen socket: " 1282 "Connection attempt from broad- or multicast " 1283 "link layer address ignored\n", s, __func__); 1284 goto dropunlock; 1285 } 1286 #ifdef INET6 1287 if (isipv6) { 1288 if (th->th_dport == th->th_sport && 1289 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1290 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1291 log(LOG_DEBUG, "%s; %s: Listen socket: " 1292 "Connection attempt to/from self " 1293 "ignored\n", s, __func__); 1294 goto dropunlock; 1295 } 1296 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1297 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1299 log(LOG_DEBUG, "%s; %s: Listen socket: " 1300 "Connection attempt from/to multicast " 1301 "address ignored\n", s, __func__); 1302 goto dropunlock; 1303 } 1304 } 1305 #endif 1306 #if defined(INET) && defined(INET6) 1307 else 1308 #endif 1309 #ifdef INET 1310 { 1311 if (th->th_dport == th->th_sport && 1312 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1313 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1314 log(LOG_DEBUG, "%s; %s: Listen socket: " 1315 "Connection attempt from/to self " 1316 "ignored\n", s, __func__); 1317 goto dropunlock; 1318 } 1319 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1320 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1321 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1322 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1323 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1324 log(LOG_DEBUG, "%s; %s: Listen socket: " 1325 "Connection attempt from/to broad- " 1326 "or multicast address ignored\n", 1327 s, __func__); 1328 goto dropunlock; 1329 } 1330 } 1331 #endif 1332 /* 1333 * SYN appears to be valid. Create compressed TCP state 1334 * for syncache. 1335 */ 1336 #ifdef TCPDEBUG 1337 if (so->so_options & SO_DEBUG) 1338 tcp_trace(TA_INPUT, ostate, tp, 1339 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1340 #endif 1341 TCP_PROBE3(debug__input, tp, th, m); 1342 tcp_dooptions(&to, optp, optlen, TO_SYN); 1343 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL, iptos)) 1344 goto tfo_socket_result; 1345 1346 /* 1347 * Entry added to syncache and mbuf consumed. 1348 * Only the listen socket is unlocked by syncache_add(). 1349 */ 1350 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1351 return (IPPROTO_DONE); 1352 } else if (tp->t_state == TCPS_LISTEN) { 1353 /* 1354 * When a listen socket is torn down the SO_ACCEPTCONN 1355 * flag is removed first while connections are drained 1356 * from the accept queue in a unlock/lock cycle of the 1357 * ACCEPT_LOCK, opening a race condition allowing a SYN 1358 * attempt go through unhandled. 1359 */ 1360 goto dropunlock; 1361 } 1362 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1363 if (tp->t_flags & TF_SIGNATURE) { 1364 tcp_dooptions(&to, optp, optlen, thflags); 1365 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1366 TCPSTAT_INC(tcps_sig_err_nosigopt); 1367 goto dropunlock; 1368 } 1369 if (!TCPMD5_ENABLED() || 1370 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1371 goto dropunlock; 1372 } 1373 #endif 1374 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1375 1376 /* 1377 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1378 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1379 * the inpcb, and unlocks pcbinfo. 1380 */ 1381 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1382 return (IPPROTO_DONE); 1383 1384 dropwithreset: 1385 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1386 1387 if (inp != NULL) { 1388 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1389 INP_WUNLOCK(inp); 1390 } else 1391 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1392 m = NULL; /* mbuf chain got consumed. */ 1393 goto drop; 1394 1395 dropunlock: 1396 if (m != NULL) 1397 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1398 1399 if (inp != NULL) 1400 INP_WUNLOCK(inp); 1401 1402 drop: 1403 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1404 if (s != NULL) 1405 free(s, M_TCPLOG); 1406 if (m != NULL) 1407 m_freem(m); 1408 return (IPPROTO_DONE); 1409 } 1410 1411 /* 1412 * Automatic sizing of receive socket buffer. Often the send 1413 * buffer size is not optimally adjusted to the actual network 1414 * conditions at hand (delay bandwidth product). Setting the 1415 * buffer size too small limits throughput on links with high 1416 * bandwidth and high delay (eg. trans-continental/oceanic links). 1417 * 1418 * On the receive side the socket buffer memory is only rarely 1419 * used to any significant extent. This allows us to be much 1420 * more aggressive in scaling the receive socket buffer. For 1421 * the case that the buffer space is actually used to a large 1422 * extent and we run out of kernel memory we can simply drop 1423 * the new segments; TCP on the sender will just retransmit it 1424 * later. Setting the buffer size too big may only consume too 1425 * much kernel memory if the application doesn't read() from 1426 * the socket or packet loss or reordering makes use of the 1427 * reassembly queue. 1428 * 1429 * The criteria to step up the receive buffer one notch are: 1430 * 1. Application has not set receive buffer size with 1431 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1432 * 2. the number of bytes received during 1/2 of an sRTT 1433 * is at least 3/8 of the current socket buffer size. 1434 * 3. receive buffer size has not hit maximal automatic size; 1435 * 1436 * If all of the criteria are met we increaset the socket buffer 1437 * by a 1/2 (bounded by the max). This allows us to keep ahead 1438 * of slow-start but also makes it so our peer never gets limited 1439 * by our rwnd which we then open up causing a burst. 1440 * 1441 * This algorithm does two steps per RTT at most and only if 1442 * we receive a bulk stream w/o packet losses or reorderings. 1443 * Shrinking the buffer during idle times is not necessary as 1444 * it doesn't consume any memory when idle. 1445 * 1446 * TODO: Only step up if the application is actually serving 1447 * the buffer to better manage the socket buffer resources. 1448 */ 1449 int 1450 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1451 struct tcpcb *tp, int tlen) 1452 { 1453 int newsize = 0; 1454 1455 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1456 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1457 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1458 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1459 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1460 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1461 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1462 } 1463 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1464 1465 /* Start over with next RTT. */ 1466 tp->rfbuf_ts = 0; 1467 tp->rfbuf_cnt = 0; 1468 } else { 1469 tp->rfbuf_cnt += tlen; /* add up */ 1470 } 1471 return (newsize); 1472 } 1473 1474 void 1475 tcp_handle_wakeup(struct tcpcb *tp, struct socket *so) 1476 { 1477 /* 1478 * Since tp might be gone if the session entered 1479 * the TIME_WAIT state before coming here, we need 1480 * to check if the socket is still connected. 1481 */ 1482 if ((so->so_state & SS_ISCONNECTED) == 0) 1483 return; 1484 INP_LOCK_ASSERT(tp->t_inpcb); 1485 if (tp->t_flags & TF_WAKESOR) { 1486 tp->t_flags &= ~TF_WAKESOR; 1487 SOCKBUF_UNLOCK_ASSERT(&so->so_rcv); 1488 sorwakeup(so); 1489 } 1490 if (tp->t_flags & TF_WAKESOW) { 1491 tp->t_flags &= ~TF_WAKESOW; 1492 SOCKBUF_UNLOCK_ASSERT(&so->so_snd); 1493 sowwakeup(so); 1494 } 1495 } 1496 1497 void 1498 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1499 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1500 { 1501 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1502 int rstreason, todrop, win, incforsyn = 0; 1503 uint32_t tiwin; 1504 uint16_t nsegs; 1505 char *s; 1506 struct in_conninfo *inc; 1507 struct mbuf *mfree; 1508 struct tcpopt to; 1509 int tfo_syn; 1510 1511 #ifdef TCPDEBUG 1512 /* 1513 * The size of tcp_saveipgen must be the size of the max ip header, 1514 * now IPv6. 1515 */ 1516 u_char tcp_saveipgen[IP6_HDR_LEN]; 1517 struct tcphdr tcp_savetcp; 1518 short ostate = 0; 1519 #endif 1520 thflags = th->th_flags; 1521 inc = &tp->t_inpcb->inp_inc; 1522 tp->sackhint.last_sack_ack = 0; 1523 sack_changed = 0; 1524 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1525 1526 NET_EPOCH_ASSERT(); 1527 INP_WLOCK_ASSERT(tp->t_inpcb); 1528 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1529 __func__)); 1530 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1531 __func__)); 1532 1533 #ifdef TCPPCAP 1534 /* Save segment, if requested. */ 1535 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1536 #endif 1537 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1538 tlen, NULL, true); 1539 1540 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1541 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1542 log(LOG_DEBUG, "%s; %s: " 1543 "SYN|FIN segment ignored (based on " 1544 "sysctl setting)\n", s, __func__); 1545 free(s, M_TCPLOG); 1546 } 1547 goto drop; 1548 } 1549 1550 /* 1551 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1552 * check SEQ.ACK first. 1553 */ 1554 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1555 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1556 rstreason = BANDLIM_UNLIMITED; 1557 goto dropwithreset; 1558 } 1559 1560 /* 1561 * Segment received on connection. 1562 * Reset idle time and keep-alive timer. 1563 * XXX: This should be done after segment 1564 * validation to ignore broken/spoofed segs. 1565 */ 1566 tp->t_rcvtime = ticks; 1567 1568 /* 1569 * Scale up the window into a 32-bit value. 1570 * For the SYN_SENT state the scale is zero. 1571 */ 1572 tiwin = th->th_win << tp->snd_scale; 1573 #ifdef STATS 1574 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); 1575 #endif 1576 1577 /* 1578 * TCP ECN processing. 1579 */ 1580 if (tp->t_flags2 & TF2_ECN_PERMIT) { 1581 if (thflags & TH_CWR) { 1582 tp->t_flags2 &= ~TF2_ECN_SND_ECE; 1583 tp->t_flags |= TF_ACKNOW; 1584 } 1585 switch (iptos & IPTOS_ECN_MASK) { 1586 case IPTOS_ECN_CE: 1587 tp->t_flags2 |= TF2_ECN_SND_ECE; 1588 TCPSTAT_INC(tcps_ecn_ce); 1589 break; 1590 case IPTOS_ECN_ECT0: 1591 TCPSTAT_INC(tcps_ecn_ect0); 1592 break; 1593 case IPTOS_ECN_ECT1: 1594 TCPSTAT_INC(tcps_ecn_ect1); 1595 break; 1596 } 1597 1598 /* Process a packet differently from RFC3168. */ 1599 cc_ecnpkt_handler(tp, th, iptos); 1600 1601 /* Congestion experienced. */ 1602 if (thflags & TH_ECE) { 1603 cc_cong_signal(tp, th, CC_ECN); 1604 } 1605 } 1606 1607 /* 1608 * Parse options on any incoming segment. 1609 */ 1610 tcp_dooptions(&to, (u_char *)(th + 1), 1611 (th->th_off << 2) - sizeof(struct tcphdr), 1612 (thflags & TH_SYN) ? TO_SYN : 0); 1613 1614 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1615 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1616 (to.to_flags & TOF_SIGNATURE) == 0) { 1617 TCPSTAT_INC(tcps_sig_err_sigopt); 1618 /* XXX: should drop? */ 1619 } 1620 #endif 1621 /* 1622 * If echoed timestamp is later than the current time, 1623 * fall back to non RFC1323 RTT calculation. Normalize 1624 * timestamp if syncookies were used when this connection 1625 * was established. 1626 */ 1627 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1628 to.to_tsecr -= tp->ts_offset; 1629 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1630 to.to_tsecr = 0; 1631 else if (tp->t_flags & TF_PREVVALID && 1632 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1633 cc_cong_signal(tp, th, CC_RTO_ERR); 1634 } 1635 /* 1636 * Process options only when we get SYN/ACK back. The SYN case 1637 * for incoming connections is handled in tcp_syncache. 1638 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1639 * or <SYN,ACK>) segment itself is never scaled. 1640 * XXX this is traditional behavior, may need to be cleaned up. 1641 */ 1642 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1643 /* Handle parallel SYN for ECN */ 1644 if (!(thflags & TH_ACK) && 1645 ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) && 1646 ((V_tcp_do_ecn == 1) || (V_tcp_do_ecn == 2))) { 1647 tp->t_flags2 |= TF2_ECN_PERMIT; 1648 tp->t_flags2 |= TF2_ECN_SND_ECE; 1649 TCPSTAT_INC(tcps_ecn_shs); 1650 } 1651 if ((to.to_flags & TOF_SCALE) && 1652 (tp->t_flags & TF_REQ_SCALE)) { 1653 tp->t_flags |= TF_RCVD_SCALE; 1654 tp->snd_scale = to.to_wscale; 1655 } else 1656 tp->t_flags &= ~TF_REQ_SCALE; 1657 /* 1658 * Initial send window. It will be updated with 1659 * the next incoming segment to the scaled value. 1660 */ 1661 tp->snd_wnd = th->th_win; 1662 if ((to.to_flags & TOF_TS) && 1663 (tp->t_flags & TF_REQ_TSTMP)) { 1664 tp->t_flags |= TF_RCVD_TSTMP; 1665 tp->ts_recent = to.to_tsval; 1666 tp->ts_recent_age = tcp_ts_getticks(); 1667 } else 1668 tp->t_flags &= ~TF_REQ_TSTMP; 1669 if (to.to_flags & TOF_MSS) 1670 tcp_mss(tp, to.to_mss); 1671 if ((tp->t_flags & TF_SACK_PERMIT) && 1672 (to.to_flags & TOF_SACKPERM) == 0) 1673 tp->t_flags &= ~TF_SACK_PERMIT; 1674 if (IS_FASTOPEN(tp->t_flags)) { 1675 if (to.to_flags & TOF_FASTOPEN) { 1676 uint16_t mss; 1677 1678 if (to.to_flags & TOF_MSS) 1679 mss = to.to_mss; 1680 else 1681 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1682 mss = TCP6_MSS; 1683 else 1684 mss = TCP_MSS; 1685 tcp_fastopen_update_cache(tp, mss, 1686 to.to_tfo_len, to.to_tfo_cookie); 1687 } else 1688 tcp_fastopen_disable_path(tp); 1689 } 1690 } 1691 1692 /* 1693 * If timestamps were negotiated during SYN/ACK and a 1694 * segment without a timestamp is received, silently drop 1695 * the segment. 1696 * See section 3.2 of RFC 7323. 1697 */ 1698 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1699 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1700 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1701 "segment silently dropped\n", s, __func__); 1702 free(s, M_TCPLOG); 1703 } 1704 goto drop; 1705 } 1706 /* 1707 * If timestamps were not negotiated during SYN/ACK and a 1708 * segment with a timestamp is received, ignore the 1709 * timestamp and process the packet normally. 1710 * See section 3.2 of RFC 7323. 1711 */ 1712 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1713 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1714 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1715 "segment processed normally\n", s, __func__); 1716 free(s, M_TCPLOG); 1717 } 1718 } 1719 1720 /* 1721 * Header prediction: check for the two common cases 1722 * of a uni-directional data xfer. If the packet has 1723 * no control flags, is in-sequence, the window didn't 1724 * change and we're not retransmitting, it's a 1725 * candidate. If the length is zero and the ack moved 1726 * forward, we're the sender side of the xfer. Just 1727 * free the data acked & wake any higher level process 1728 * that was blocked waiting for space. If the length 1729 * is non-zero and the ack didn't move, we're the 1730 * receiver side. If we're getting packets in-order 1731 * (the reassembly queue is empty), add the data to 1732 * the socket buffer and note that we need a delayed ack. 1733 * Make sure that the hidden state-flags are also off. 1734 * Since we check for TCPS_ESTABLISHED first, it can only 1735 * be TH_NEEDSYN. 1736 */ 1737 if (tp->t_state == TCPS_ESTABLISHED && 1738 th->th_seq == tp->rcv_nxt && 1739 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1740 tp->snd_nxt == tp->snd_max && 1741 tiwin && tiwin == tp->snd_wnd && 1742 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1743 SEGQ_EMPTY(tp) && 1744 ((to.to_flags & TOF_TS) == 0 || 1745 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1746 /* 1747 * If last ACK falls within this segment's sequence numbers, 1748 * record the timestamp. 1749 * NOTE that the test is modified according to the latest 1750 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1751 */ 1752 if ((to.to_flags & TOF_TS) != 0 && 1753 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1754 tp->ts_recent_age = tcp_ts_getticks(); 1755 tp->ts_recent = to.to_tsval; 1756 } 1757 1758 if (tlen == 0) { 1759 if (SEQ_GT(th->th_ack, tp->snd_una) && 1760 SEQ_LEQ(th->th_ack, tp->snd_max) && 1761 !IN_RECOVERY(tp->t_flags) && 1762 (to.to_flags & TOF_SACK) == 0 && 1763 TAILQ_EMPTY(&tp->snd_holes)) { 1764 /* 1765 * This is a pure ack for outstanding data. 1766 */ 1767 TCPSTAT_INC(tcps_predack); 1768 1769 /* 1770 * "bad retransmit" recovery without timestamps. 1771 */ 1772 if ((to.to_flags & TOF_TS) == 0 && 1773 tp->t_rxtshift == 1 && 1774 tp->t_flags & TF_PREVVALID && 1775 (int)(ticks - tp->t_badrxtwin) < 0) { 1776 cc_cong_signal(tp, th, CC_RTO_ERR); 1777 } 1778 1779 /* 1780 * Recalculate the transmit timer / rtt. 1781 * 1782 * Some boxes send broken timestamp replies 1783 * during the SYN+ACK phase, ignore 1784 * timestamps of 0 or we could calculate a 1785 * huge RTT and blow up the retransmit timer. 1786 */ 1787 if ((to.to_flags & TOF_TS) != 0 && 1788 to.to_tsecr) { 1789 uint32_t t; 1790 1791 t = tcp_ts_getticks() - to.to_tsecr; 1792 if (!tp->t_rttlow || tp->t_rttlow > t) 1793 tp->t_rttlow = t; 1794 tcp_xmit_timer(tp, 1795 TCP_TS_TO_TICKS(t) + 1); 1796 } else if (tp->t_rtttime && 1797 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1798 if (!tp->t_rttlow || 1799 tp->t_rttlow > ticks - tp->t_rtttime) 1800 tp->t_rttlow = ticks - tp->t_rtttime; 1801 tcp_xmit_timer(tp, 1802 ticks - tp->t_rtttime); 1803 } 1804 acked = BYTES_THIS_ACK(tp, th); 1805 1806 #ifdef TCP_HHOOK 1807 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1808 hhook_run_tcp_est_in(tp, th, &to); 1809 #endif 1810 1811 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1812 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1813 sbdrop(&so->so_snd, acked); 1814 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1815 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1816 tp->snd_recover = th->th_ack - 1; 1817 1818 /* 1819 * Let the congestion control algorithm update 1820 * congestion control related information. This 1821 * typically means increasing the congestion 1822 * window. 1823 */ 1824 cc_ack_received(tp, th, nsegs, CC_ACK); 1825 1826 tp->snd_una = th->th_ack; 1827 /* 1828 * Pull snd_wl2 up to prevent seq wrap relative 1829 * to th_ack. 1830 */ 1831 tp->snd_wl2 = th->th_ack; 1832 tp->t_dupacks = 0; 1833 m_freem(m); 1834 1835 /* 1836 * If all outstanding data are acked, stop 1837 * retransmit timer, otherwise restart timer 1838 * using current (possibly backed-off) value. 1839 * If process is waiting for space, 1840 * wakeup/selwakeup/signal. If data 1841 * are ready to send, let tcp_output 1842 * decide between more output or persist. 1843 */ 1844 #ifdef TCPDEBUG 1845 if (so->so_options & SO_DEBUG) 1846 tcp_trace(TA_INPUT, ostate, tp, 1847 (void *)tcp_saveipgen, 1848 &tcp_savetcp, 0); 1849 #endif 1850 TCP_PROBE3(debug__input, tp, th, m); 1851 if (tp->snd_una == tp->snd_max) 1852 tcp_timer_activate(tp, TT_REXMT, 0); 1853 else if (!tcp_timer_active(tp, TT_PERSIST)) 1854 tcp_timer_activate(tp, TT_REXMT, 1855 tp->t_rxtcur); 1856 tp->t_flags |= TF_WAKESOW; 1857 if (sbavail(&so->so_snd)) 1858 (void) tp->t_fb->tfb_tcp_output(tp); 1859 goto check_delack; 1860 } 1861 } else if (th->th_ack == tp->snd_una && 1862 tlen <= sbspace(&so->so_rcv)) { 1863 int newsize = 0; /* automatic sockbuf scaling */ 1864 1865 /* 1866 * This is a pure, in-sequence data packet with 1867 * nothing on the reassembly queue and we have enough 1868 * buffer space to take it. 1869 */ 1870 /* Clean receiver SACK report if present */ 1871 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1872 tcp_clean_sackreport(tp); 1873 TCPSTAT_INC(tcps_preddat); 1874 tp->rcv_nxt += tlen; 1875 if (tlen && 1876 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 1877 (tp->t_fbyte_in == 0)) { 1878 tp->t_fbyte_in = ticks; 1879 if (tp->t_fbyte_in == 0) 1880 tp->t_fbyte_in = 1; 1881 if (tp->t_fbyte_out && tp->t_fbyte_in) 1882 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 1883 } 1884 /* 1885 * Pull snd_wl1 up to prevent seq wrap relative to 1886 * th_seq. 1887 */ 1888 tp->snd_wl1 = th->th_seq; 1889 /* 1890 * Pull rcv_up up to prevent seq wrap relative to 1891 * rcv_nxt. 1892 */ 1893 tp->rcv_up = tp->rcv_nxt; 1894 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1895 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1896 #ifdef TCPDEBUG 1897 if (so->so_options & SO_DEBUG) 1898 tcp_trace(TA_INPUT, ostate, tp, 1899 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1900 #endif 1901 TCP_PROBE3(debug__input, tp, th, m); 1902 1903 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1904 1905 /* Add data to socket buffer. */ 1906 SOCKBUF_LOCK(&so->so_rcv); 1907 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1908 m_freem(m); 1909 } else { 1910 /* 1911 * Set new socket buffer size. 1912 * Give up when limit is reached. 1913 */ 1914 if (newsize) 1915 if (!sbreserve_locked(&so->so_rcv, 1916 newsize, so, NULL)) 1917 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1918 m_adj(m, drop_hdrlen); /* delayed header drop */ 1919 sbappendstream_locked(&so->so_rcv, m, 0); 1920 } 1921 SOCKBUF_UNLOCK(&so->so_rcv); 1922 tp->t_flags |= TF_WAKESOR; 1923 if (DELAY_ACK(tp, tlen)) { 1924 tp->t_flags |= TF_DELACK; 1925 } else { 1926 tp->t_flags |= TF_ACKNOW; 1927 tp->t_fb->tfb_tcp_output(tp); 1928 } 1929 goto check_delack; 1930 } 1931 } 1932 1933 /* 1934 * Calculate amount of space in receive window, 1935 * and then do TCP input processing. 1936 * Receive window is amount of space in rcv queue, 1937 * but not less than advertised window. 1938 */ 1939 win = sbspace(&so->so_rcv); 1940 if (win < 0) 1941 win = 0; 1942 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1943 1944 switch (tp->t_state) { 1945 /* 1946 * If the state is SYN_RECEIVED: 1947 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1948 */ 1949 case TCPS_SYN_RECEIVED: 1950 if ((thflags & TH_ACK) && 1951 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1952 SEQ_GT(th->th_ack, tp->snd_max))) { 1953 rstreason = BANDLIM_RST_OPENPORT; 1954 goto dropwithreset; 1955 } 1956 if (IS_FASTOPEN(tp->t_flags)) { 1957 /* 1958 * When a TFO connection is in SYN_RECEIVED, the 1959 * only valid packets are the initial SYN, a 1960 * retransmit/copy of the initial SYN (possibly with 1961 * a subset of the original data), a valid ACK, a 1962 * FIN, or a RST. 1963 */ 1964 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1965 rstreason = BANDLIM_RST_OPENPORT; 1966 goto dropwithreset; 1967 } else if (thflags & TH_SYN) { 1968 /* non-initial SYN is ignored */ 1969 if ((tcp_timer_active(tp, TT_DELACK) || 1970 tcp_timer_active(tp, TT_REXMT))) 1971 goto drop; 1972 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1973 goto drop; 1974 } 1975 } 1976 break; 1977 1978 /* 1979 * If the state is SYN_SENT: 1980 * if seg contains a RST with valid ACK (SEQ.ACK has already 1981 * been verified), then drop the connection. 1982 * if seg contains a RST without an ACK, drop the seg. 1983 * if seg does not contain SYN, then drop the seg. 1984 * Otherwise this is an acceptable SYN segment 1985 * initialize tp->rcv_nxt and tp->irs 1986 * if seg contains ack then advance tp->snd_una 1987 * if seg contains an ECE and ECN support is enabled, the stream 1988 * is ECN capable. 1989 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1990 * arrange for segment to be acked (eventually) 1991 * continue processing rest of data/controls, beginning with URG 1992 */ 1993 case TCPS_SYN_SENT: 1994 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1995 TCP_PROBE5(connect__refused, NULL, tp, 1996 m, tp, th); 1997 tp = tcp_drop(tp, ECONNREFUSED); 1998 } 1999 if (thflags & TH_RST) 2000 goto drop; 2001 if (!(thflags & TH_SYN)) 2002 goto drop; 2003 2004 tp->irs = th->th_seq; 2005 tcp_rcvseqinit(tp); 2006 if (thflags & TH_ACK) { 2007 int tfo_partial_ack = 0; 2008 2009 TCPSTAT_INC(tcps_connects); 2010 soisconnected(so); 2011 #ifdef MAC 2012 mac_socketpeer_set_from_mbuf(m, so); 2013 #endif 2014 /* Do window scaling on this connection? */ 2015 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2016 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2017 tp->rcv_scale = tp->request_r_scale; 2018 } 2019 tp->rcv_adv += min(tp->rcv_wnd, 2020 TCP_MAXWIN << tp->rcv_scale); 2021 tp->snd_una++; /* SYN is acked */ 2022 /* 2023 * If not all the data that was sent in the TFO SYN 2024 * has been acked, resend the remainder right away. 2025 */ 2026 if (IS_FASTOPEN(tp->t_flags) && 2027 (tp->snd_una != tp->snd_max)) { 2028 tp->snd_nxt = th->th_ack; 2029 tfo_partial_ack = 1; 2030 } 2031 /* 2032 * If there's data, delay ACK; if there's also a FIN 2033 * ACKNOW will be turned on later. 2034 */ 2035 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2036 tcp_timer_activate(tp, TT_DELACK, 2037 tcp_delacktime); 2038 else 2039 tp->t_flags |= TF_ACKNOW; 2040 2041 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2042 (V_tcp_do_ecn == 1)) { 2043 tp->t_flags2 |= TF2_ECN_PERMIT; 2044 TCPSTAT_INC(tcps_ecn_shs); 2045 } 2046 2047 /* 2048 * Received <SYN,ACK> in SYN_SENT[*] state. 2049 * Transitions: 2050 * SYN_SENT --> ESTABLISHED 2051 * SYN_SENT* --> FIN_WAIT_1 2052 */ 2053 tp->t_starttime = ticks; 2054 if (tp->t_flags & TF_NEEDFIN) { 2055 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2056 tp->t_flags &= ~TF_NEEDFIN; 2057 thflags &= ~TH_SYN; 2058 } else { 2059 tcp_state_change(tp, TCPS_ESTABLISHED); 2060 TCP_PROBE5(connect__established, NULL, tp, 2061 m, tp, th); 2062 cc_conn_init(tp); 2063 tcp_timer_activate(tp, TT_KEEP, 2064 TP_KEEPIDLE(tp)); 2065 } 2066 } else { 2067 /* 2068 * Received initial SYN in SYN-SENT[*] state => 2069 * simultaneous open. 2070 * If it succeeds, connection is * half-synchronized. 2071 * Otherwise, do 3-way handshake: 2072 * SYN-SENT -> SYN-RECEIVED 2073 * SYN-SENT* -> SYN-RECEIVED* 2074 */ 2075 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2076 tcp_timer_activate(tp, TT_REXMT, 0); 2077 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2078 } 2079 2080 INP_WLOCK_ASSERT(tp->t_inpcb); 2081 2082 /* 2083 * Advance th->th_seq to correspond to first data byte. 2084 * If data, trim to stay within window, 2085 * dropping FIN if necessary. 2086 */ 2087 th->th_seq++; 2088 if (tlen > tp->rcv_wnd) { 2089 todrop = tlen - tp->rcv_wnd; 2090 m_adj(m, -todrop); 2091 tlen = tp->rcv_wnd; 2092 thflags &= ~TH_FIN; 2093 TCPSTAT_INC(tcps_rcvpackafterwin); 2094 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2095 } 2096 tp->snd_wl1 = th->th_seq - 1; 2097 tp->rcv_up = th->th_seq; 2098 /* 2099 * Client side of transaction: already sent SYN and data. 2100 * If the remote host used T/TCP to validate the SYN, 2101 * our data will be ACK'd; if so, enter normal data segment 2102 * processing in the middle of step 5, ack processing. 2103 * Otherwise, goto step 6. 2104 */ 2105 if (thflags & TH_ACK) 2106 goto process_ACK; 2107 2108 goto step6; 2109 2110 /* 2111 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2112 * do normal processing. 2113 * 2114 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2115 */ 2116 case TCPS_LAST_ACK: 2117 case TCPS_CLOSING: 2118 break; /* continue normal processing */ 2119 } 2120 2121 /* 2122 * States other than LISTEN or SYN_SENT. 2123 * First check the RST flag and sequence number since reset segments 2124 * are exempt from the timestamp and connection count tests. This 2125 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2126 * below which allowed reset segments in half the sequence space 2127 * to fall though and be processed (which gives forged reset 2128 * segments with a random sequence number a 50 percent chance of 2129 * killing a connection). 2130 * Then check timestamp, if present. 2131 * Then check the connection count, if present. 2132 * Then check that at least some bytes of segment are within 2133 * receive window. If segment begins before rcv_nxt, 2134 * drop leading data (and SYN); if nothing left, just ack. 2135 */ 2136 if (thflags & TH_RST) { 2137 /* 2138 * RFC5961 Section 3.2 2139 * 2140 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2141 * - If RST is in window, we send challenge ACK. 2142 * 2143 * Note: to take into account delayed ACKs, we should 2144 * test against last_ack_sent instead of rcv_nxt. 2145 * Note 2: we handle special case of closed window, not 2146 * covered by the RFC. 2147 */ 2148 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2149 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2150 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2151 KASSERT(tp->t_state != TCPS_SYN_SENT, 2152 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2153 __func__, th, tp)); 2154 2155 if (V_tcp_insecure_rst || 2156 tp->last_ack_sent == th->th_seq) { 2157 TCPSTAT_INC(tcps_drops); 2158 /* Drop the connection. */ 2159 switch (tp->t_state) { 2160 case TCPS_SYN_RECEIVED: 2161 so->so_error = ECONNREFUSED; 2162 goto close; 2163 case TCPS_ESTABLISHED: 2164 case TCPS_FIN_WAIT_1: 2165 case TCPS_FIN_WAIT_2: 2166 case TCPS_CLOSE_WAIT: 2167 case TCPS_CLOSING: 2168 case TCPS_LAST_ACK: 2169 so->so_error = ECONNRESET; 2170 close: 2171 /* FALLTHROUGH */ 2172 default: 2173 tp = tcp_close(tp); 2174 } 2175 } else { 2176 TCPSTAT_INC(tcps_badrst); 2177 /* Send challenge ACK. */ 2178 tcp_respond(tp, mtod(m, void *), th, m, 2179 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2180 tp->last_ack_sent = tp->rcv_nxt; 2181 m = NULL; 2182 } 2183 } 2184 goto drop; 2185 } 2186 2187 /* 2188 * RFC5961 Section 4.2 2189 * Send challenge ACK for any SYN in synchronized state. 2190 */ 2191 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2192 tp->t_state != TCPS_SYN_RECEIVED) { 2193 TCPSTAT_INC(tcps_badsyn); 2194 if (V_tcp_insecure_syn && 2195 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2196 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2197 tp = tcp_drop(tp, ECONNRESET); 2198 rstreason = BANDLIM_UNLIMITED; 2199 } else { 2200 /* Send challenge ACK. */ 2201 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2202 tp->snd_nxt, TH_ACK); 2203 tp->last_ack_sent = tp->rcv_nxt; 2204 m = NULL; 2205 } 2206 goto drop; 2207 } 2208 2209 /* 2210 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2211 * and it's less than ts_recent, drop it. 2212 */ 2213 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2214 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2215 /* Check to see if ts_recent is over 24 days old. */ 2216 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2217 /* 2218 * Invalidate ts_recent. If this segment updates 2219 * ts_recent, the age will be reset later and ts_recent 2220 * will get a valid value. If it does not, setting 2221 * ts_recent to zero will at least satisfy the 2222 * requirement that zero be placed in the timestamp 2223 * echo reply when ts_recent isn't valid. The 2224 * age isn't reset until we get a valid ts_recent 2225 * because we don't want out-of-order segments to be 2226 * dropped when ts_recent is old. 2227 */ 2228 tp->ts_recent = 0; 2229 } else { 2230 TCPSTAT_INC(tcps_rcvduppack); 2231 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2232 TCPSTAT_INC(tcps_pawsdrop); 2233 if (tlen) 2234 goto dropafterack; 2235 goto drop; 2236 } 2237 } 2238 2239 /* 2240 * In the SYN-RECEIVED state, validate that the packet belongs to 2241 * this connection before trimming the data to fit the receive 2242 * window. Check the sequence number versus IRS since we know 2243 * the sequence numbers haven't wrapped. This is a partial fix 2244 * for the "LAND" DoS attack. 2245 */ 2246 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2247 rstreason = BANDLIM_RST_OPENPORT; 2248 goto dropwithreset; 2249 } 2250 2251 todrop = tp->rcv_nxt - th->th_seq; 2252 if (todrop > 0) { 2253 if (thflags & TH_SYN) { 2254 thflags &= ~TH_SYN; 2255 th->th_seq++; 2256 if (th->th_urp > 1) 2257 th->th_urp--; 2258 else 2259 thflags &= ~TH_URG; 2260 todrop--; 2261 } 2262 /* 2263 * Following if statement from Stevens, vol. 2, p. 960. 2264 */ 2265 if (todrop > tlen 2266 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2267 /* 2268 * Any valid FIN must be to the left of the window. 2269 * At this point the FIN must be a duplicate or out 2270 * of sequence; drop it. 2271 */ 2272 thflags &= ~TH_FIN; 2273 2274 /* 2275 * Send an ACK to resynchronize and drop any data. 2276 * But keep on processing for RST or ACK. 2277 */ 2278 tp->t_flags |= TF_ACKNOW; 2279 todrop = tlen; 2280 TCPSTAT_INC(tcps_rcvduppack); 2281 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2282 } else { 2283 TCPSTAT_INC(tcps_rcvpartduppack); 2284 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2285 } 2286 /* 2287 * DSACK - add SACK block for dropped range 2288 */ 2289 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { 2290 tcp_update_sack_list(tp, th->th_seq, 2291 th->th_seq + todrop); 2292 /* 2293 * ACK now, as the next in-sequence segment 2294 * will clear the DSACK block again 2295 */ 2296 tp->t_flags |= TF_ACKNOW; 2297 } 2298 drop_hdrlen += todrop; /* drop from the top afterwards */ 2299 th->th_seq += todrop; 2300 tlen -= todrop; 2301 if (th->th_urp > todrop) 2302 th->th_urp -= todrop; 2303 else { 2304 thflags &= ~TH_URG; 2305 th->th_urp = 0; 2306 } 2307 } 2308 2309 /* 2310 * If new data are received on a connection after the 2311 * user processes are gone, then RST the other end. 2312 */ 2313 if ((so->so_state & SS_NOFDREF) && 2314 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2315 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2316 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2317 "after socket was closed, " 2318 "sending RST and removing tcpcb\n", 2319 s, __func__, tcpstates[tp->t_state], tlen); 2320 free(s, M_TCPLOG); 2321 } 2322 tp = tcp_close(tp); 2323 TCPSTAT_INC(tcps_rcvafterclose); 2324 rstreason = BANDLIM_UNLIMITED; 2325 goto dropwithreset; 2326 } 2327 2328 /* 2329 * If segment ends after window, drop trailing data 2330 * (and PUSH and FIN); if nothing left, just ACK. 2331 */ 2332 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2333 if (todrop > 0) { 2334 TCPSTAT_INC(tcps_rcvpackafterwin); 2335 if (todrop >= tlen) { 2336 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2337 /* 2338 * If window is closed can only take segments at 2339 * window edge, and have to drop data and PUSH from 2340 * incoming segments. Continue processing, but 2341 * remember to ack. Otherwise, drop segment 2342 * and ack. 2343 */ 2344 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2345 tp->t_flags |= TF_ACKNOW; 2346 TCPSTAT_INC(tcps_rcvwinprobe); 2347 } else 2348 goto dropafterack; 2349 } else 2350 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2351 m_adj(m, -todrop); 2352 tlen -= todrop; 2353 thflags &= ~(TH_PUSH|TH_FIN); 2354 } 2355 2356 /* 2357 * If last ACK falls within this segment's sequence numbers, 2358 * record its timestamp. 2359 * NOTE: 2360 * 1) That the test incorporates suggestions from the latest 2361 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2362 * 2) That updating only on newer timestamps interferes with 2363 * our earlier PAWS tests, so this check should be solely 2364 * predicated on the sequence space of this segment. 2365 * 3) That we modify the segment boundary check to be 2366 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2367 * instead of RFC1323's 2368 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2369 * This modified check allows us to overcome RFC1323's 2370 * limitations as described in Stevens TCP/IP Illustrated 2371 * Vol. 2 p.869. In such cases, we can still calculate the 2372 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2373 */ 2374 if ((to.to_flags & TOF_TS) != 0 && 2375 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2376 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2377 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2378 tp->ts_recent_age = tcp_ts_getticks(); 2379 tp->ts_recent = to.to_tsval; 2380 } 2381 2382 /* 2383 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2384 * flag is on (half-synchronized state), then queue data for 2385 * later processing; else drop segment and return. 2386 */ 2387 if ((thflags & TH_ACK) == 0) { 2388 if (tp->t_state == TCPS_SYN_RECEIVED || 2389 (tp->t_flags & TF_NEEDSYN)) { 2390 if (tp->t_state == TCPS_SYN_RECEIVED && 2391 IS_FASTOPEN(tp->t_flags)) { 2392 tp->snd_wnd = tiwin; 2393 cc_conn_init(tp); 2394 } 2395 goto step6; 2396 } else if (tp->t_flags & TF_ACKNOW) 2397 goto dropafterack; 2398 else 2399 goto drop; 2400 } 2401 2402 /* 2403 * Ack processing. 2404 */ 2405 switch (tp->t_state) { 2406 /* 2407 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2408 * ESTABLISHED state and continue processing. 2409 * The ACK was checked above. 2410 */ 2411 case TCPS_SYN_RECEIVED: 2412 2413 TCPSTAT_INC(tcps_connects); 2414 soisconnected(so); 2415 /* Do window scaling? */ 2416 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2417 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2418 tp->rcv_scale = tp->request_r_scale; 2419 } 2420 tp->snd_wnd = tiwin; 2421 /* 2422 * Make transitions: 2423 * SYN-RECEIVED -> ESTABLISHED 2424 * SYN-RECEIVED* -> FIN-WAIT-1 2425 */ 2426 tp->t_starttime = ticks; 2427 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2428 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2429 tp->t_tfo_pending = NULL; 2430 } 2431 if (tp->t_flags & TF_NEEDFIN) { 2432 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2433 tp->t_flags &= ~TF_NEEDFIN; 2434 } else { 2435 tcp_state_change(tp, TCPS_ESTABLISHED); 2436 TCP_PROBE5(accept__established, NULL, tp, 2437 m, tp, th); 2438 /* 2439 * TFO connections call cc_conn_init() during SYN 2440 * processing. Calling it again here for such 2441 * connections is not harmless as it would undo the 2442 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2443 * is retransmitted. 2444 */ 2445 if (!IS_FASTOPEN(tp->t_flags)) 2446 cc_conn_init(tp); 2447 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2448 } 2449 /* 2450 * Account for the ACK of our SYN prior to 2451 * regular ACK processing below, except for 2452 * simultaneous SYN, which is handled later. 2453 */ 2454 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) 2455 incforsyn = 1; 2456 /* 2457 * If segment contains data or ACK, will call tcp_reass() 2458 * later; if not, do so now to pass queued data to user. 2459 */ 2460 if (tlen == 0 && (thflags & TH_FIN) == 0) 2461 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2462 (struct mbuf *)0); 2463 tp->snd_wl1 = th->th_seq - 1; 2464 /* FALLTHROUGH */ 2465 2466 /* 2467 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2468 * ACKs. If the ack is in the range 2469 * tp->snd_una < th->th_ack <= tp->snd_max 2470 * then advance tp->snd_una to th->th_ack and drop 2471 * data from the retransmission queue. If this ACK reflects 2472 * more up to date window information we update our window information. 2473 */ 2474 case TCPS_ESTABLISHED: 2475 case TCPS_FIN_WAIT_1: 2476 case TCPS_FIN_WAIT_2: 2477 case TCPS_CLOSE_WAIT: 2478 case TCPS_CLOSING: 2479 case TCPS_LAST_ACK: 2480 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2481 TCPSTAT_INC(tcps_rcvacktoomuch); 2482 goto dropafterack; 2483 } 2484 if ((tp->t_flags & TF_SACK_PERMIT) && 2485 ((to.to_flags & TOF_SACK) || 2486 !TAILQ_EMPTY(&tp->snd_holes))) 2487 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2488 else 2489 /* 2490 * Reset the value so that previous (valid) value 2491 * from the last ack with SACK doesn't get used. 2492 */ 2493 tp->sackhint.sacked_bytes = 0; 2494 2495 #ifdef TCP_HHOOK 2496 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2497 hhook_run_tcp_est_in(tp, th, &to); 2498 #endif 2499 2500 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2501 u_int maxseg; 2502 2503 maxseg = tcp_maxseg(tp); 2504 if (tlen == 0 && 2505 (tiwin == tp->snd_wnd || 2506 (tp->t_flags & TF_SACK_PERMIT))) { 2507 /* 2508 * If this is the first time we've seen a 2509 * FIN from the remote, this is not a 2510 * duplicate and it needs to be processed 2511 * normally. This happens during a 2512 * simultaneous close. 2513 */ 2514 if ((thflags & TH_FIN) && 2515 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2516 tp->t_dupacks = 0; 2517 break; 2518 } 2519 TCPSTAT_INC(tcps_rcvdupack); 2520 /* 2521 * If we have outstanding data (other than 2522 * a window probe), this is a completely 2523 * duplicate ack (ie, window info didn't 2524 * change and FIN isn't set), 2525 * the ack is the biggest we've 2526 * seen and we've seen exactly our rexmt 2527 * threshold of them, assume a packet 2528 * has been dropped and retransmit it. 2529 * Kludge snd_nxt & the congestion 2530 * window so we send only this one 2531 * packet. 2532 * 2533 * We know we're losing at the current 2534 * window size so do congestion avoidance 2535 * (set ssthresh to half the current window 2536 * and pull our congestion window back to 2537 * the new ssthresh). 2538 * 2539 * Dup acks mean that packets have left the 2540 * network (they're now cached at the receiver) 2541 * so bump cwnd by the amount in the receiver 2542 * to keep a constant cwnd packets in the 2543 * network. 2544 * 2545 * When using TCP ECN, notify the peer that 2546 * we reduced the cwnd. 2547 */ 2548 /* 2549 * Following 2 kinds of acks should not affect 2550 * dupack counting: 2551 * 1) Old acks 2552 * 2) Acks with SACK but without any new SACK 2553 * information in them. These could result from 2554 * any anomaly in the network like a switch 2555 * duplicating packets or a possible DoS attack. 2556 */ 2557 if (th->th_ack != tp->snd_una || 2558 ((tp->t_flags & TF_SACK_PERMIT) && 2559 !sack_changed)) 2560 break; 2561 else if (!tcp_timer_active(tp, TT_REXMT)) 2562 tp->t_dupacks = 0; 2563 else if (++tp->t_dupacks > tcprexmtthresh || 2564 IN_FASTRECOVERY(tp->t_flags)) { 2565 cc_ack_received(tp, th, nsegs, 2566 CC_DUPACK); 2567 if (V_tcp_do_prr && 2568 IN_FASTRECOVERY(tp->t_flags) && 2569 (tp->t_flags & TF_SACK_PERMIT)) { 2570 long snd_cnt = 0, limit = 0; 2571 long del_data = 0, pipe = 0; 2572 /* 2573 * In a duplicate ACK del_data is only the 2574 * diff_in_sack. If no SACK is used del_data 2575 * will be 0. Pipe is the amount of data we 2576 * estimate to be in the network. 2577 */ 2578 del_data = tp->sackhint.delivered_data; 2579 pipe = (tp->snd_nxt - tp->snd_fack) + 2580 tp->sackhint.sack_bytes_rexmit; 2581 tp->sackhint.prr_delivered += del_data; 2582 if (pipe > tp->snd_ssthresh) { 2583 snd_cnt = (tp->sackhint.prr_delivered * 2584 tp->snd_ssthresh / 2585 tp->sackhint.recover_fs) + 2586 1 - tp->sackhint.sack_bytes_rexmit; 2587 } else { 2588 if (V_tcp_do_prr_conservative) 2589 limit = tp->sackhint.prr_delivered - 2590 tp->sackhint.sack_bytes_rexmit; 2591 else 2592 if ((tp->sackhint.prr_delivered - 2593 tp->sackhint.sack_bytes_rexmit) > 2594 del_data) 2595 limit = tp->sackhint.prr_delivered - 2596 tp->sackhint.sack_bytes_rexmit + 2597 maxseg; 2598 else 2599 limit = del_data + maxseg; 2600 if ((tp->snd_ssthresh - pipe) < limit) 2601 snd_cnt = tp->snd_ssthresh - pipe; 2602 else 2603 snd_cnt = limit; 2604 } 2605 snd_cnt = max((snd_cnt / maxseg), 0); 2606 /* 2607 * Send snd_cnt new data into the network in 2608 * response to this ACK. If there is a going 2609 * to be a SACK retransmission, adjust snd_cwnd 2610 * accordingly. 2611 */ 2612 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 2613 tp->sackhint.sack_bytes_rexmit + 2614 (snd_cnt * maxseg); 2615 } else if ((tp->t_flags & TF_SACK_PERMIT) && 2616 IN_FASTRECOVERY(tp->t_flags)) { 2617 int awnd; 2618 2619 /* 2620 * Compute the amount of data in flight first. 2621 * We can inject new data into the pipe iff 2622 * we have less than 1/2 the original window's 2623 * worth of data in flight. 2624 */ 2625 if (V_tcp_do_rfc6675_pipe) 2626 awnd = tcp_compute_pipe(tp); 2627 else 2628 awnd = (tp->snd_nxt - tp->snd_fack) + 2629 tp->sackhint.sack_bytes_rexmit; 2630 2631 if (awnd < tp->snd_ssthresh) { 2632 tp->snd_cwnd += maxseg; 2633 if (tp->snd_cwnd > tp->snd_ssthresh) 2634 tp->snd_cwnd = tp->snd_ssthresh; 2635 } 2636 } else 2637 tp->snd_cwnd += maxseg; 2638 (void) tp->t_fb->tfb_tcp_output(tp); 2639 goto drop; 2640 } else if (tp->t_dupacks == tcprexmtthresh) { 2641 tcp_seq onxt = tp->snd_nxt; 2642 2643 /* 2644 * If we're doing sack, or prr, check 2645 * to see if we're already in sack 2646 * recovery. If we're not doing sack, 2647 * check to see if we're in newreno 2648 * recovery. 2649 */ 2650 if (V_tcp_do_prr || 2651 (tp->t_flags & TF_SACK_PERMIT)) { 2652 if (IN_FASTRECOVERY(tp->t_flags)) { 2653 tp->t_dupacks = 0; 2654 break; 2655 } 2656 } else { 2657 if (SEQ_LEQ(th->th_ack, 2658 tp->snd_recover)) { 2659 tp->t_dupacks = 0; 2660 break; 2661 } 2662 } 2663 /* Congestion signal before ack. */ 2664 cc_cong_signal(tp, th, CC_NDUPACK); 2665 cc_ack_received(tp, th, nsegs, 2666 CC_DUPACK); 2667 tcp_timer_activate(tp, TT_REXMT, 0); 2668 tp->t_rtttime = 0; 2669 if (V_tcp_do_prr) { 2670 /* 2671 * snd_ssthresh is already updated by 2672 * cc_cong_signal. 2673 */ 2674 tp->sackhint.prr_delivered = 0; 2675 tp->sackhint.sack_bytes_rexmit = 0; 2676 if (!(tp->sackhint.recover_fs = tp->snd_nxt - tp->snd_una)) 2677 tp->sackhint.recover_fs = 1; 2678 } 2679 if (tp->t_flags & TF_SACK_PERMIT) { 2680 TCPSTAT_INC( 2681 tcps_sack_recovery_episode); 2682 tp->snd_recover = tp->snd_nxt; 2683 tp->snd_cwnd = maxseg; 2684 (void) tp->t_fb->tfb_tcp_output(tp); 2685 goto drop; 2686 } 2687 tp->snd_nxt = th->th_ack; 2688 tp->snd_cwnd = maxseg; 2689 (void) tp->t_fb->tfb_tcp_output(tp); 2690 KASSERT(tp->snd_limited <= 2, 2691 ("%s: tp->snd_limited too big", 2692 __func__)); 2693 tp->snd_cwnd = tp->snd_ssthresh + 2694 maxseg * 2695 (tp->t_dupacks - tp->snd_limited); 2696 if (SEQ_GT(onxt, tp->snd_nxt)) 2697 tp->snd_nxt = onxt; 2698 goto drop; 2699 } else if (V_tcp_do_rfc3042) { 2700 /* 2701 * Process first and second duplicate 2702 * ACKs. Each indicates a segment 2703 * leaving the network, creating room 2704 * for more. Make sure we can send a 2705 * packet on reception of each duplicate 2706 * ACK by increasing snd_cwnd by one 2707 * segment. Restore the original 2708 * snd_cwnd after packet transmission. 2709 */ 2710 cc_ack_received(tp, th, nsegs, 2711 CC_DUPACK); 2712 uint32_t oldcwnd = tp->snd_cwnd; 2713 tcp_seq oldsndmax = tp->snd_max; 2714 u_int sent; 2715 int avail; 2716 2717 KASSERT(tp->t_dupacks == 1 || 2718 tp->t_dupacks == 2, 2719 ("%s: dupacks not 1 or 2", 2720 __func__)); 2721 if (tp->t_dupacks == 1) 2722 tp->snd_limited = 0; 2723 tp->snd_cwnd = 2724 (tp->snd_nxt - tp->snd_una) + 2725 (tp->t_dupacks - tp->snd_limited) * 2726 maxseg; 2727 /* 2728 * Only call tcp_output when there 2729 * is new data available to be sent. 2730 * Otherwise we would send pure ACKs. 2731 */ 2732 SOCKBUF_LOCK(&so->so_snd); 2733 avail = sbavail(&so->so_snd) - 2734 (tp->snd_nxt - tp->snd_una); 2735 SOCKBUF_UNLOCK(&so->so_snd); 2736 if (avail > 0) 2737 (void) tp->t_fb->tfb_tcp_output(tp); 2738 sent = tp->snd_max - oldsndmax; 2739 if (sent > maxseg) { 2740 KASSERT((tp->t_dupacks == 2 && 2741 tp->snd_limited == 0) || 2742 (sent == maxseg + 1 && 2743 tp->t_flags & TF_SENTFIN), 2744 ("%s: sent too much", 2745 __func__)); 2746 tp->snd_limited = 2; 2747 } else if (sent > 0) 2748 ++tp->snd_limited; 2749 tp->snd_cwnd = oldcwnd; 2750 goto drop; 2751 } 2752 } 2753 break; 2754 } else { 2755 /* 2756 * This ack is advancing the left edge, reset the 2757 * counter. 2758 */ 2759 tp->t_dupacks = 0; 2760 /* 2761 * If this ack also has new SACK info, increment the 2762 * counter as per rfc6675. The variable 2763 * sack_changed tracks all changes to the SACK 2764 * scoreboard, including when partial ACKs without 2765 * SACK options are received, and clear the scoreboard 2766 * from the left side. Such partial ACKs should not be 2767 * counted as dupacks here. 2768 */ 2769 if ((tp->t_flags & TF_SACK_PERMIT) && 2770 (to.to_flags & TOF_SACK) && 2771 sack_changed) 2772 tp->t_dupacks++; 2773 } 2774 2775 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2776 ("%s: th_ack <= snd_una", __func__)); 2777 2778 /* 2779 * If the congestion window was inflated to account 2780 * for the other side's cached packets, retract it. 2781 */ 2782 if (IN_FASTRECOVERY(tp->t_flags)) { 2783 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2784 if (tp->t_flags & TF_SACK_PERMIT) 2785 if (V_tcp_do_prr) 2786 tcp_prr_partialack(tp, th); 2787 else 2788 tcp_sack_partialack(tp, th); 2789 else 2790 tcp_newreno_partial_ack(tp, th); 2791 } else 2792 cc_post_recovery(tp, th); 2793 } 2794 /* 2795 * If we reach this point, ACK is not a duplicate, 2796 * i.e., it ACKs something we sent. 2797 */ 2798 if (tp->t_flags & TF_NEEDSYN) { 2799 /* 2800 * T/TCP: Connection was half-synchronized, and our 2801 * SYN has been ACK'd (so connection is now fully 2802 * synchronized). Go to non-starred state, 2803 * increment snd_una for ACK of SYN, and check if 2804 * we can do window scaling. 2805 */ 2806 tp->t_flags &= ~TF_NEEDSYN; 2807 tp->snd_una++; 2808 /* Do window scaling? */ 2809 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2810 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2811 tp->rcv_scale = tp->request_r_scale; 2812 /* Send window already scaled. */ 2813 } 2814 } 2815 2816 process_ACK: 2817 INP_WLOCK_ASSERT(tp->t_inpcb); 2818 2819 /* 2820 * Adjust for the SYN bit in sequence space, 2821 * but don't account for it in cwnd calculations. 2822 * This is for the SYN_RECEIVED, non-simultaneous 2823 * SYN case. SYN_SENT and simultaneous SYN are 2824 * treated elsewhere. 2825 */ 2826 if (incforsyn) 2827 tp->snd_una++; 2828 acked = BYTES_THIS_ACK(tp, th); 2829 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2830 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2831 tp->snd_una, th->th_ack, tp, m)); 2832 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2833 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2834 2835 /* 2836 * If we just performed our first retransmit, and the ACK 2837 * arrives within our recovery window, then it was a mistake 2838 * to do the retransmit in the first place. Recover our 2839 * original cwnd and ssthresh, and proceed to transmit where 2840 * we left off. 2841 */ 2842 if (tp->t_rxtshift == 1 && 2843 tp->t_flags & TF_PREVVALID && 2844 tp->t_badrxtwin && 2845 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2846 cc_cong_signal(tp, th, CC_RTO_ERR); 2847 2848 /* 2849 * If we have a timestamp reply, update smoothed 2850 * round trip time. If no timestamp is present but 2851 * transmit timer is running and timed sequence 2852 * number was acked, update smoothed round trip time. 2853 * Since we now have an rtt measurement, cancel the 2854 * timer backoff (cf., Phil Karn's retransmit alg.). 2855 * Recompute the initial retransmit timer. 2856 * 2857 * Some boxes send broken timestamp replies 2858 * during the SYN+ACK phase, ignore 2859 * timestamps of 0 or we could calculate a 2860 * huge RTT and blow up the retransmit timer. 2861 */ 2862 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2863 uint32_t t; 2864 2865 t = tcp_ts_getticks() - to.to_tsecr; 2866 if (!tp->t_rttlow || tp->t_rttlow > t) 2867 tp->t_rttlow = t; 2868 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2869 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2870 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2871 tp->t_rttlow = ticks - tp->t_rtttime; 2872 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2873 } 2874 2875 /* 2876 * If all outstanding data is acked, stop retransmit 2877 * timer and remember to restart (more output or persist). 2878 * If there is more data to be acked, restart retransmit 2879 * timer, using current (possibly backed-off) value. 2880 */ 2881 if (th->th_ack == tp->snd_max) { 2882 tcp_timer_activate(tp, TT_REXMT, 0); 2883 needoutput = 1; 2884 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2885 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2886 2887 /* 2888 * If no data (only SYN) was ACK'd, 2889 * skip rest of ACK processing. 2890 */ 2891 if (acked == 0) 2892 goto step6; 2893 2894 /* 2895 * Let the congestion control algorithm update congestion 2896 * control related information. This typically means increasing 2897 * the congestion window. 2898 */ 2899 cc_ack_received(tp, th, nsegs, CC_ACK); 2900 2901 SOCKBUF_LOCK(&so->so_snd); 2902 if (acked > sbavail(&so->so_snd)) { 2903 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2904 tp->snd_wnd -= sbavail(&so->so_snd); 2905 else 2906 tp->snd_wnd = 0; 2907 mfree = sbcut_locked(&so->so_snd, 2908 (int)sbavail(&so->so_snd)); 2909 ourfinisacked = 1; 2910 } else { 2911 mfree = sbcut_locked(&so->so_snd, acked); 2912 if (tp->snd_wnd >= (uint32_t) acked) 2913 tp->snd_wnd -= acked; 2914 else 2915 tp->snd_wnd = 0; 2916 ourfinisacked = 0; 2917 } 2918 SOCKBUF_UNLOCK(&so->so_snd); 2919 tp->t_flags |= TF_WAKESOW; 2920 m_freem(mfree); 2921 /* Detect una wraparound. */ 2922 if (!IN_RECOVERY(tp->t_flags) && 2923 SEQ_GT(tp->snd_una, tp->snd_recover) && 2924 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2925 tp->snd_recover = th->th_ack - 1; 2926 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2927 if (IN_RECOVERY(tp->t_flags) && 2928 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2929 EXIT_RECOVERY(tp->t_flags); 2930 } 2931 tp->snd_una = th->th_ack; 2932 if (tp->t_flags & TF_SACK_PERMIT) { 2933 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2934 tp->snd_recover = tp->snd_una; 2935 } 2936 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2937 tp->snd_nxt = tp->snd_una; 2938 2939 switch (tp->t_state) { 2940 /* 2941 * In FIN_WAIT_1 STATE in addition to the processing 2942 * for the ESTABLISHED state if our FIN is now acknowledged 2943 * then enter FIN_WAIT_2. 2944 */ 2945 case TCPS_FIN_WAIT_1: 2946 if (ourfinisacked) { 2947 /* 2948 * If we can't receive any more 2949 * data, then closing user can proceed. 2950 * Starting the timer is contrary to the 2951 * specification, but if we don't get a FIN 2952 * we'll hang forever. 2953 * 2954 * XXXjl: 2955 * we should release the tp also, and use a 2956 * compressed state. 2957 */ 2958 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2959 soisdisconnected(so); 2960 tcp_timer_activate(tp, TT_2MSL, 2961 (tcp_fast_finwait2_recycle ? 2962 tcp_finwait2_timeout : 2963 TP_MAXIDLE(tp))); 2964 } 2965 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2966 } 2967 break; 2968 2969 /* 2970 * In CLOSING STATE in addition to the processing for 2971 * the ESTABLISHED state if the ACK acknowledges our FIN 2972 * then enter the TIME-WAIT state, otherwise ignore 2973 * the segment. 2974 */ 2975 case TCPS_CLOSING: 2976 if (ourfinisacked) { 2977 tcp_twstart(tp); 2978 m_freem(m); 2979 return; 2980 } 2981 break; 2982 2983 /* 2984 * In LAST_ACK, we may still be waiting for data to drain 2985 * and/or to be acked, as well as for the ack of our FIN. 2986 * If our FIN is now acknowledged, delete the TCB, 2987 * enter the closed state and return. 2988 */ 2989 case TCPS_LAST_ACK: 2990 if (ourfinisacked) { 2991 tp = tcp_close(tp); 2992 goto drop; 2993 } 2994 break; 2995 } 2996 } 2997 2998 step6: 2999 INP_WLOCK_ASSERT(tp->t_inpcb); 3000 3001 /* 3002 * Update window information. 3003 * Don't look at window if no ACK: TAC's send garbage on first SYN. 3004 */ 3005 if ((thflags & TH_ACK) && 3006 (SEQ_LT(tp->snd_wl1, th->th_seq) || 3007 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 3008 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 3009 /* keep track of pure window updates */ 3010 if (tlen == 0 && 3011 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 3012 TCPSTAT_INC(tcps_rcvwinupd); 3013 tp->snd_wnd = tiwin; 3014 tp->snd_wl1 = th->th_seq; 3015 tp->snd_wl2 = th->th_ack; 3016 if (tp->snd_wnd > tp->max_sndwnd) 3017 tp->max_sndwnd = tp->snd_wnd; 3018 needoutput = 1; 3019 } 3020 3021 /* 3022 * Process segments with URG. 3023 */ 3024 if ((thflags & TH_URG) && th->th_urp && 3025 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3026 /* 3027 * This is a kludge, but if we receive and accept 3028 * random urgent pointers, we'll crash in 3029 * soreceive. It's hard to imagine someone 3030 * actually wanting to send this much urgent data. 3031 */ 3032 SOCKBUF_LOCK(&so->so_rcv); 3033 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3034 th->th_urp = 0; /* XXX */ 3035 thflags &= ~TH_URG; /* XXX */ 3036 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3037 goto dodata; /* XXX */ 3038 } 3039 /* 3040 * If this segment advances the known urgent pointer, 3041 * then mark the data stream. This should not happen 3042 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3043 * a FIN has been received from the remote side. 3044 * In these states we ignore the URG. 3045 * 3046 * According to RFC961 (Assigned Protocols), 3047 * the urgent pointer points to the last octet 3048 * of urgent data. We continue, however, 3049 * to consider it to indicate the first octet 3050 * of data past the urgent section as the original 3051 * spec states (in one of two places). 3052 */ 3053 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3054 tp->rcv_up = th->th_seq + th->th_urp; 3055 so->so_oobmark = sbavail(&so->so_rcv) + 3056 (tp->rcv_up - tp->rcv_nxt) - 1; 3057 if (so->so_oobmark == 0) 3058 so->so_rcv.sb_state |= SBS_RCVATMARK; 3059 sohasoutofband(so); 3060 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3061 } 3062 SOCKBUF_UNLOCK(&so->so_rcv); 3063 /* 3064 * Remove out of band data so doesn't get presented to user. 3065 * This can happen independent of advancing the URG pointer, 3066 * but if two URG's are pending at once, some out-of-band 3067 * data may creep in... ick. 3068 */ 3069 if (th->th_urp <= (uint32_t)tlen && 3070 !(so->so_options & SO_OOBINLINE)) { 3071 /* hdr drop is delayed */ 3072 tcp_pulloutofband(so, th, m, drop_hdrlen); 3073 } 3074 } else { 3075 /* 3076 * If no out of band data is expected, 3077 * pull receive urgent pointer along 3078 * with the receive window. 3079 */ 3080 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3081 tp->rcv_up = tp->rcv_nxt; 3082 } 3083 dodata: /* XXX */ 3084 INP_WLOCK_ASSERT(tp->t_inpcb); 3085 3086 /* 3087 * Process the segment text, merging it into the TCP sequencing queue, 3088 * and arranging for acknowledgment of receipt if necessary. 3089 * This process logically involves adjusting tp->rcv_wnd as data 3090 * is presented to the user (this happens in tcp_usrreq.c, 3091 * case PRU_RCVD). If a FIN has already been received on this 3092 * connection then we just ignore the text. 3093 */ 3094 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3095 IS_FASTOPEN(tp->t_flags)); 3096 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && 3097 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3098 tcp_seq save_start = th->th_seq; 3099 tcp_seq save_rnxt = tp->rcv_nxt; 3100 int save_tlen = tlen; 3101 m_adj(m, drop_hdrlen); /* delayed header drop */ 3102 /* 3103 * Insert segment which includes th into TCP reassembly queue 3104 * with control block tp. Set thflags to whether reassembly now 3105 * includes a segment with FIN. This handles the common case 3106 * inline (segment is the next to be received on an established 3107 * connection, and the queue is empty), avoiding linkage into 3108 * and removal from the queue and repetition of various 3109 * conversions. 3110 * Set DELACK for segments received in order, but ack 3111 * immediately when segments are out of order (so 3112 * fast retransmit can work). 3113 */ 3114 if (th->th_seq == tp->rcv_nxt && 3115 SEGQ_EMPTY(tp) && 3116 (TCPS_HAVEESTABLISHED(tp->t_state) || 3117 tfo_syn)) { 3118 if (DELAY_ACK(tp, tlen) || tfo_syn) 3119 tp->t_flags |= TF_DELACK; 3120 else 3121 tp->t_flags |= TF_ACKNOW; 3122 tp->rcv_nxt += tlen; 3123 if (tlen && 3124 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && 3125 (tp->t_fbyte_in == 0)) { 3126 tp->t_fbyte_in = ticks; 3127 if (tp->t_fbyte_in == 0) 3128 tp->t_fbyte_in = 1; 3129 if (tp->t_fbyte_out && tp->t_fbyte_in) 3130 tp->t_flags2 |= TF2_FBYTES_COMPLETE; 3131 } 3132 thflags = th->th_flags & TH_FIN; 3133 TCPSTAT_INC(tcps_rcvpack); 3134 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3135 SOCKBUF_LOCK(&so->so_rcv); 3136 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3137 m_freem(m); 3138 else 3139 sbappendstream_locked(&so->so_rcv, m, 0); 3140 SOCKBUF_UNLOCK(&so->so_rcv); 3141 tp->t_flags |= TF_WAKESOR; 3142 } else { 3143 /* 3144 * XXX: Due to the header drop above "th" is 3145 * theoretically invalid by now. Fortunately 3146 * m_adj() doesn't actually frees any mbufs 3147 * when trimming from the head. 3148 */ 3149 tcp_seq temp = save_start; 3150 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3151 tp->t_flags |= TF_ACKNOW; 3152 } 3153 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) { 3154 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3155 /* 3156 * DSACK actually handled in the fastpath 3157 * above. 3158 */ 3159 tcp_update_sack_list(tp, save_start, 3160 save_start + save_tlen); 3161 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3162 if ((tp->rcv_numsacks >= 1) && 3163 (tp->sackblks[0].end == save_start)) { 3164 /* 3165 * Partial overlap, recorded at todrop 3166 * above. 3167 */ 3168 tcp_update_sack_list(tp, 3169 tp->sackblks[0].start, 3170 tp->sackblks[0].end); 3171 } else { 3172 tcp_update_dsack_list(tp, save_start, 3173 save_start + save_tlen); 3174 } 3175 } else if (tlen >= save_tlen) { 3176 /* Update of sackblks. */ 3177 tcp_update_dsack_list(tp, save_start, 3178 save_start + save_tlen); 3179 } else if (tlen > 0) { 3180 tcp_update_dsack_list(tp, save_start, 3181 save_start + tlen); 3182 } 3183 } 3184 #if 0 3185 /* 3186 * Note the amount of data that peer has sent into 3187 * our window, in order to estimate the sender's 3188 * buffer size. 3189 * XXX: Unused. 3190 */ 3191 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3192 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3193 else 3194 len = so->so_rcv.sb_hiwat; 3195 #endif 3196 } else { 3197 m_freem(m); 3198 thflags &= ~TH_FIN; 3199 } 3200 3201 /* 3202 * If FIN is received ACK the FIN and let the user know 3203 * that the connection is closing. 3204 */ 3205 if (thflags & TH_FIN) { 3206 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3207 socantrcvmore(so); 3208 /* The socket upcall is handled by socantrcvmore. */ 3209 tp->t_flags &= ~TF_WAKESOR; 3210 /* 3211 * If connection is half-synchronized 3212 * (ie NEEDSYN flag on) then delay ACK, 3213 * so it may be piggybacked when SYN is sent. 3214 * Otherwise, since we received a FIN then no 3215 * more input can be expected, send ACK now. 3216 */ 3217 if (tp->t_flags & TF_NEEDSYN) 3218 tp->t_flags |= TF_DELACK; 3219 else 3220 tp->t_flags |= TF_ACKNOW; 3221 tp->rcv_nxt++; 3222 } 3223 switch (tp->t_state) { 3224 /* 3225 * In SYN_RECEIVED and ESTABLISHED STATES 3226 * enter the CLOSE_WAIT state. 3227 */ 3228 case TCPS_SYN_RECEIVED: 3229 tp->t_starttime = ticks; 3230 /* FALLTHROUGH */ 3231 case TCPS_ESTABLISHED: 3232 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3233 break; 3234 3235 /* 3236 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3237 * enter the CLOSING state. 3238 */ 3239 case TCPS_FIN_WAIT_1: 3240 tcp_state_change(tp, TCPS_CLOSING); 3241 break; 3242 3243 /* 3244 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3245 * starting the time-wait timer, turning off the other 3246 * standard timers. 3247 */ 3248 case TCPS_FIN_WAIT_2: 3249 tcp_twstart(tp); 3250 return; 3251 } 3252 } 3253 #ifdef TCPDEBUG 3254 if (so->so_options & SO_DEBUG) 3255 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3256 &tcp_savetcp, 0); 3257 #endif 3258 TCP_PROBE3(debug__input, tp, th, m); 3259 3260 /* 3261 * Return any desired output. 3262 */ 3263 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3264 (void) tp->t_fb->tfb_tcp_output(tp); 3265 3266 check_delack: 3267 INP_WLOCK_ASSERT(tp->t_inpcb); 3268 3269 if (tp->t_flags & TF_DELACK) { 3270 tp->t_flags &= ~TF_DELACK; 3271 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3272 } 3273 tcp_handle_wakeup(tp, so); 3274 INP_WUNLOCK(tp->t_inpcb); 3275 return; 3276 3277 dropafterack: 3278 /* 3279 * Generate an ACK dropping incoming segment if it occupies 3280 * sequence space, where the ACK reflects our state. 3281 * 3282 * We can now skip the test for the RST flag since all 3283 * paths to this code happen after packets containing 3284 * RST have been dropped. 3285 * 3286 * In the SYN-RECEIVED state, don't send an ACK unless the 3287 * segment we received passes the SYN-RECEIVED ACK test. 3288 * If it fails send a RST. This breaks the loop in the 3289 * "LAND" DoS attack, and also prevents an ACK storm 3290 * between two listening ports that have been sent forged 3291 * SYN segments, each with the source address of the other. 3292 */ 3293 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3294 (SEQ_GT(tp->snd_una, th->th_ack) || 3295 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3296 rstreason = BANDLIM_RST_OPENPORT; 3297 goto dropwithreset; 3298 } 3299 #ifdef TCPDEBUG 3300 if (so->so_options & SO_DEBUG) 3301 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3302 &tcp_savetcp, 0); 3303 #endif 3304 TCP_PROBE3(debug__input, tp, th, m); 3305 tp->t_flags |= TF_ACKNOW; 3306 (void) tp->t_fb->tfb_tcp_output(tp); 3307 tcp_handle_wakeup(tp, so); 3308 INP_WUNLOCK(tp->t_inpcb); 3309 m_freem(m); 3310 return; 3311 3312 dropwithreset: 3313 if (tp != NULL) { 3314 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3315 tcp_handle_wakeup(tp, so); 3316 INP_WUNLOCK(tp->t_inpcb); 3317 } else 3318 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3319 return; 3320 3321 drop: 3322 /* 3323 * Drop space held by incoming segment and return. 3324 */ 3325 #ifdef TCPDEBUG 3326 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3327 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3328 &tcp_savetcp, 0); 3329 #endif 3330 TCP_PROBE3(debug__input, tp, th, m); 3331 if (tp != NULL) { 3332 tcp_handle_wakeup(tp, so); 3333 INP_WUNLOCK(tp->t_inpcb); 3334 } 3335 m_freem(m); 3336 } 3337 3338 /* 3339 * Issue RST and make ACK acceptable to originator of segment. 3340 * The mbuf must still include the original packet header. 3341 * tp may be NULL. 3342 */ 3343 void 3344 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3345 int tlen, int rstreason) 3346 { 3347 #ifdef INET 3348 struct ip *ip; 3349 #endif 3350 #ifdef INET6 3351 struct ip6_hdr *ip6; 3352 #endif 3353 3354 if (tp != NULL) { 3355 INP_WLOCK_ASSERT(tp->t_inpcb); 3356 } 3357 3358 /* Don't bother if destination was broadcast/multicast. */ 3359 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3360 goto drop; 3361 #ifdef INET6 3362 if (mtod(m, struct ip *)->ip_v == 6) { 3363 ip6 = mtod(m, struct ip6_hdr *); 3364 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3365 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3366 goto drop; 3367 /* IPv6 anycast check is done at tcp6_input() */ 3368 } 3369 #endif 3370 #if defined(INET) && defined(INET6) 3371 else 3372 #endif 3373 #ifdef INET 3374 { 3375 ip = mtod(m, struct ip *); 3376 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3377 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3378 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3379 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3380 goto drop; 3381 } 3382 #endif 3383 3384 /* Perform bandwidth limiting. */ 3385 if (badport_bandlim(rstreason) < 0) 3386 goto drop; 3387 3388 /* tcp_respond consumes the mbuf chain. */ 3389 if (th->th_flags & TH_ACK) { 3390 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3391 th->th_ack, TH_RST); 3392 } else { 3393 if (th->th_flags & TH_SYN) 3394 tlen++; 3395 if (th->th_flags & TH_FIN) 3396 tlen++; 3397 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3398 (tcp_seq)0, TH_RST|TH_ACK); 3399 } 3400 return; 3401 drop: 3402 m_freem(m); 3403 } 3404 3405 /* 3406 * Parse TCP options and place in tcpopt. 3407 */ 3408 void 3409 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3410 { 3411 int opt, optlen; 3412 3413 to->to_flags = 0; 3414 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3415 opt = cp[0]; 3416 if (opt == TCPOPT_EOL) 3417 break; 3418 if (opt == TCPOPT_NOP) 3419 optlen = 1; 3420 else { 3421 if (cnt < 2) 3422 break; 3423 optlen = cp[1]; 3424 if (optlen < 2 || optlen > cnt) 3425 break; 3426 } 3427 switch (opt) { 3428 case TCPOPT_MAXSEG: 3429 if (optlen != TCPOLEN_MAXSEG) 3430 continue; 3431 if (!(flags & TO_SYN)) 3432 continue; 3433 to->to_flags |= TOF_MSS; 3434 bcopy((char *)cp + 2, 3435 (char *)&to->to_mss, sizeof(to->to_mss)); 3436 to->to_mss = ntohs(to->to_mss); 3437 break; 3438 case TCPOPT_WINDOW: 3439 if (optlen != TCPOLEN_WINDOW) 3440 continue; 3441 if (!(flags & TO_SYN)) 3442 continue; 3443 to->to_flags |= TOF_SCALE; 3444 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3445 break; 3446 case TCPOPT_TIMESTAMP: 3447 if (optlen != TCPOLEN_TIMESTAMP) 3448 continue; 3449 to->to_flags |= TOF_TS; 3450 bcopy((char *)cp + 2, 3451 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3452 to->to_tsval = ntohl(to->to_tsval); 3453 bcopy((char *)cp + 6, 3454 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3455 to->to_tsecr = ntohl(to->to_tsecr); 3456 break; 3457 case TCPOPT_SIGNATURE: 3458 /* 3459 * In order to reply to a host which has set the 3460 * TCP_SIGNATURE option in its initial SYN, we have 3461 * to record the fact that the option was observed 3462 * here for the syncache code to perform the correct 3463 * response. 3464 */ 3465 if (optlen != TCPOLEN_SIGNATURE) 3466 continue; 3467 to->to_flags |= TOF_SIGNATURE; 3468 to->to_signature = cp + 2; 3469 break; 3470 case TCPOPT_SACK_PERMITTED: 3471 if (optlen != TCPOLEN_SACK_PERMITTED) 3472 continue; 3473 if (!(flags & TO_SYN)) 3474 continue; 3475 if (!V_tcp_do_sack) 3476 continue; 3477 to->to_flags |= TOF_SACKPERM; 3478 break; 3479 case TCPOPT_SACK: 3480 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3481 continue; 3482 if (flags & TO_SYN) 3483 continue; 3484 to->to_flags |= TOF_SACK; 3485 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3486 to->to_sacks = cp + 2; 3487 TCPSTAT_INC(tcps_sack_rcv_blocks); 3488 break; 3489 case TCPOPT_FAST_OPEN: 3490 /* 3491 * Cookie length validation is performed by the 3492 * server side cookie checking code or the client 3493 * side cookie cache update code. 3494 */ 3495 if (!(flags & TO_SYN)) 3496 continue; 3497 if (!V_tcp_fastopen_client_enable && 3498 !V_tcp_fastopen_server_enable) 3499 continue; 3500 to->to_flags |= TOF_FASTOPEN; 3501 to->to_tfo_len = optlen - 2; 3502 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3503 break; 3504 default: 3505 continue; 3506 } 3507 } 3508 } 3509 3510 /* 3511 * Pull out of band byte out of a segment so 3512 * it doesn't appear in the user's data queue. 3513 * It is still reflected in the segment length for 3514 * sequencing purposes. 3515 */ 3516 void 3517 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3518 int off) 3519 { 3520 int cnt = off + th->th_urp - 1; 3521 3522 while (cnt >= 0) { 3523 if (m->m_len > cnt) { 3524 char *cp = mtod(m, caddr_t) + cnt; 3525 struct tcpcb *tp = sototcpcb(so); 3526 3527 INP_WLOCK_ASSERT(tp->t_inpcb); 3528 3529 tp->t_iobc = *cp; 3530 tp->t_oobflags |= TCPOOB_HAVEDATA; 3531 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3532 m->m_len--; 3533 if (m->m_flags & M_PKTHDR) 3534 m->m_pkthdr.len--; 3535 return; 3536 } 3537 cnt -= m->m_len; 3538 m = m->m_next; 3539 if (m == NULL) 3540 break; 3541 } 3542 panic("tcp_pulloutofband"); 3543 } 3544 3545 /* 3546 * Collect new round-trip time estimate 3547 * and update averages and current timeout. 3548 */ 3549 void 3550 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3551 { 3552 int delta; 3553 3554 INP_WLOCK_ASSERT(tp->t_inpcb); 3555 3556 TCPSTAT_INC(tcps_rttupdated); 3557 tp->t_rttupdated++; 3558 #ifdef STATS 3559 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, 3560 imax(0, rtt * 1000 / hz)); 3561 #endif 3562 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3563 /* 3564 * srtt is stored as fixed point with 5 bits after the 3565 * binary point (i.e., scaled by 8). The following magic 3566 * is equivalent to the smoothing algorithm in rfc793 with 3567 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3568 * point). Adjust rtt to origin 0. 3569 */ 3570 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3571 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3572 3573 if ((tp->t_srtt += delta) <= 0) 3574 tp->t_srtt = 1; 3575 3576 /* 3577 * We accumulate a smoothed rtt variance (actually, a 3578 * smoothed mean difference), then set the retransmit 3579 * timer to smoothed rtt + 4 times the smoothed variance. 3580 * rttvar is stored as fixed point with 4 bits after the 3581 * binary point (scaled by 16). The following is 3582 * equivalent to rfc793 smoothing with an alpha of .75 3583 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3584 * rfc793's wired-in beta. 3585 */ 3586 if (delta < 0) 3587 delta = -delta; 3588 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3589 if ((tp->t_rttvar += delta) <= 0) 3590 tp->t_rttvar = 1; 3591 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3592 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3593 } else { 3594 /* 3595 * No rtt measurement yet - use the unsmoothed rtt. 3596 * Set the variance to half the rtt (so our first 3597 * retransmit happens at 3*rtt). 3598 */ 3599 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3600 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3601 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3602 } 3603 tp->t_rtttime = 0; 3604 tp->t_rxtshift = 0; 3605 3606 /* 3607 * the retransmit should happen at rtt + 4 * rttvar. 3608 * Because of the way we do the smoothing, srtt and rttvar 3609 * will each average +1/2 tick of bias. When we compute 3610 * the retransmit timer, we want 1/2 tick of rounding and 3611 * 1 extra tick because of +-1/2 tick uncertainty in the 3612 * firing of the timer. The bias will give us exactly the 3613 * 1.5 tick we need. But, because the bias is 3614 * statistical, we have to test that we don't drop below 3615 * the minimum feasible timer (which is 2 ticks). 3616 */ 3617 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3618 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3619 3620 /* 3621 * We received an ack for a packet that wasn't retransmitted; 3622 * it is probably safe to discard any error indications we've 3623 * received recently. This isn't quite right, but close enough 3624 * for now (a route might have failed after we sent a segment, 3625 * and the return path might not be symmetrical). 3626 */ 3627 tp->t_softerror = 0; 3628 } 3629 3630 /* 3631 * Determine a reasonable value for maxseg size. 3632 * If the route is known, check route for mtu. 3633 * If none, use an mss that can be handled on the outgoing interface 3634 * without forcing IP to fragment. If no route is found, route has no mtu, 3635 * or the destination isn't local, use a default, hopefully conservative 3636 * size (usually 512 or the default IP max size, but no more than the mtu 3637 * of the interface), as we can't discover anything about intervening 3638 * gateways or networks. We also initialize the congestion/slow start 3639 * window to be a single segment if the destination isn't local. 3640 * While looking at the routing entry, we also initialize other path-dependent 3641 * parameters from pre-set or cached values in the routing entry. 3642 * 3643 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3644 * IP options, e.g. IPSEC data, since length of this data may vary, and 3645 * thus it is calculated for every segment separately in tcp_output(). 3646 * 3647 * NOTE that this routine is only called when we process an incoming 3648 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3649 * settings are handled in tcp_mssopt(). 3650 */ 3651 void 3652 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3653 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3654 { 3655 int mss = 0; 3656 uint32_t maxmtu = 0; 3657 struct inpcb *inp = tp->t_inpcb; 3658 struct hc_metrics_lite metrics; 3659 #ifdef INET6 3660 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3661 size_t min_protoh = isipv6 ? 3662 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3663 sizeof (struct tcpiphdr); 3664 #else 3665 const size_t min_protoh = sizeof(struct tcpiphdr); 3666 #endif 3667 3668 INP_WLOCK_ASSERT(tp->t_inpcb); 3669 3670 if (mtuoffer != -1) { 3671 KASSERT(offer == -1, ("%s: conflict", __func__)); 3672 offer = mtuoffer - min_protoh; 3673 } 3674 3675 /* Initialize. */ 3676 #ifdef INET6 3677 if (isipv6) { 3678 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3679 tp->t_maxseg = V_tcp_v6mssdflt; 3680 } 3681 #endif 3682 #if defined(INET) && defined(INET6) 3683 else 3684 #endif 3685 #ifdef INET 3686 { 3687 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3688 tp->t_maxseg = V_tcp_mssdflt; 3689 } 3690 #endif 3691 3692 /* 3693 * No route to sender, stay with default mss and return. 3694 */ 3695 if (maxmtu == 0) { 3696 /* 3697 * In case we return early we need to initialize metrics 3698 * to a defined state as tcp_hc_get() would do for us 3699 * if there was no cache hit. 3700 */ 3701 if (metricptr != NULL) 3702 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3703 return; 3704 } 3705 3706 /* What have we got? */ 3707 switch (offer) { 3708 case 0: 3709 /* 3710 * Offer == 0 means that there was no MSS on the SYN 3711 * segment, in this case we use tcp_mssdflt as 3712 * already assigned to t_maxseg above. 3713 */ 3714 offer = tp->t_maxseg; 3715 break; 3716 3717 case -1: 3718 /* 3719 * Offer == -1 means that we didn't receive SYN yet. 3720 */ 3721 /* FALLTHROUGH */ 3722 3723 default: 3724 /* 3725 * Prevent DoS attack with too small MSS. Round up 3726 * to at least minmss. 3727 */ 3728 offer = max(offer, V_tcp_minmss); 3729 } 3730 3731 /* 3732 * rmx information is now retrieved from tcp_hostcache. 3733 */ 3734 tcp_hc_get(&inp->inp_inc, &metrics); 3735 if (metricptr != NULL) 3736 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3737 3738 /* 3739 * If there's a discovered mtu in tcp hostcache, use it. 3740 * Else, use the link mtu. 3741 */ 3742 if (metrics.rmx_mtu) 3743 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3744 else { 3745 #ifdef INET6 3746 if (isipv6) { 3747 mss = maxmtu - min_protoh; 3748 if (!V_path_mtu_discovery && 3749 !in6_localaddr(&inp->in6p_faddr)) 3750 mss = min(mss, V_tcp_v6mssdflt); 3751 } 3752 #endif 3753 #if defined(INET) && defined(INET6) 3754 else 3755 #endif 3756 #ifdef INET 3757 { 3758 mss = maxmtu - min_protoh; 3759 if (!V_path_mtu_discovery && 3760 !in_localaddr(inp->inp_faddr)) 3761 mss = min(mss, V_tcp_mssdflt); 3762 } 3763 #endif 3764 /* 3765 * XXX - The above conditional (mss = maxmtu - min_protoh) 3766 * probably violates the TCP spec. 3767 * The problem is that, since we don't know the 3768 * other end's MSS, we are supposed to use a conservative 3769 * default. But, if we do that, then MTU discovery will 3770 * never actually take place, because the conservative 3771 * default is much less than the MTUs typically seen 3772 * on the Internet today. For the moment, we'll sweep 3773 * this under the carpet. 3774 * 3775 * The conservative default might not actually be a problem 3776 * if the only case this occurs is when sending an initial 3777 * SYN with options and data to a host we've never talked 3778 * to before. Then, they will reply with an MSS value which 3779 * will get recorded and the new parameters should get 3780 * recomputed. For Further Study. 3781 */ 3782 } 3783 mss = min(mss, offer); 3784 3785 /* 3786 * Sanity check: make sure that maxseg will be large 3787 * enough to allow some data on segments even if the 3788 * all the option space is used (40bytes). Otherwise 3789 * funny things may happen in tcp_output. 3790 * 3791 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3792 */ 3793 mss = max(mss, 64); 3794 3795 tp->t_maxseg = mss; 3796 } 3797 3798 void 3799 tcp_mss(struct tcpcb *tp, int offer) 3800 { 3801 int mss; 3802 uint32_t bufsize; 3803 struct inpcb *inp; 3804 struct socket *so; 3805 struct hc_metrics_lite metrics; 3806 struct tcp_ifcap cap; 3807 3808 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3809 3810 bzero(&cap, sizeof(cap)); 3811 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3812 3813 mss = tp->t_maxseg; 3814 inp = tp->t_inpcb; 3815 3816 /* 3817 * If there's a pipesize, change the socket buffer to that size, 3818 * don't change if sb_hiwat is different than default (then it 3819 * has been changed on purpose with setsockopt). 3820 * Make the socket buffers an integral number of mss units; 3821 * if the mss is larger than the socket buffer, decrease the mss. 3822 */ 3823 so = inp->inp_socket; 3824 SOCKBUF_LOCK(&so->so_snd); 3825 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3826 bufsize = metrics.rmx_sendpipe; 3827 else 3828 bufsize = so->so_snd.sb_hiwat; 3829 if (bufsize < mss) 3830 mss = bufsize; 3831 else { 3832 bufsize = roundup(bufsize, mss); 3833 if (bufsize > sb_max) 3834 bufsize = sb_max; 3835 if (bufsize > so->so_snd.sb_hiwat) 3836 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3837 } 3838 SOCKBUF_UNLOCK(&so->so_snd); 3839 /* 3840 * Sanity check: make sure that maxseg will be large 3841 * enough to allow some data on segments even if the 3842 * all the option space is used (40bytes). Otherwise 3843 * funny things may happen in tcp_output. 3844 * 3845 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3846 */ 3847 tp->t_maxseg = max(mss, 64); 3848 3849 SOCKBUF_LOCK(&so->so_rcv); 3850 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3851 bufsize = metrics.rmx_recvpipe; 3852 else 3853 bufsize = so->so_rcv.sb_hiwat; 3854 if (bufsize > mss) { 3855 bufsize = roundup(bufsize, mss); 3856 if (bufsize > sb_max) 3857 bufsize = sb_max; 3858 if (bufsize > so->so_rcv.sb_hiwat) 3859 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3860 } 3861 SOCKBUF_UNLOCK(&so->so_rcv); 3862 3863 /* Check the interface for TSO capabilities. */ 3864 if (cap.ifcap & CSUM_TSO) { 3865 tp->t_flags |= TF_TSO; 3866 tp->t_tsomax = cap.tsomax; 3867 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3868 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3869 } 3870 } 3871 3872 /* 3873 * Determine the MSS option to send on an outgoing SYN. 3874 */ 3875 int 3876 tcp_mssopt(struct in_conninfo *inc) 3877 { 3878 int mss = 0; 3879 uint32_t thcmtu = 0; 3880 uint32_t maxmtu = 0; 3881 size_t min_protoh; 3882 3883 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3884 3885 #ifdef INET6 3886 if (inc->inc_flags & INC_ISIPV6) { 3887 mss = V_tcp_v6mssdflt; 3888 maxmtu = tcp_maxmtu6(inc, NULL); 3889 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3890 } 3891 #endif 3892 #if defined(INET) && defined(INET6) 3893 else 3894 #endif 3895 #ifdef INET 3896 { 3897 mss = V_tcp_mssdflt; 3898 maxmtu = tcp_maxmtu(inc, NULL); 3899 min_protoh = sizeof(struct tcpiphdr); 3900 } 3901 #endif 3902 #if defined(INET6) || defined(INET) 3903 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3904 #endif 3905 3906 if (maxmtu && thcmtu) 3907 mss = min(maxmtu, thcmtu) - min_protoh; 3908 else if (maxmtu || thcmtu) 3909 mss = max(maxmtu, thcmtu) - min_protoh; 3910 3911 return (mss); 3912 } 3913 3914 void 3915 tcp_prr_partialack(struct tcpcb *tp, struct tcphdr *th) 3916 { 3917 long snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; 3918 int maxseg = tcp_maxseg(tp); 3919 3920 INP_WLOCK_ASSERT(tp->t_inpcb); 3921 3922 tcp_timer_activate(tp, TT_REXMT, 0); 3923 tp->t_rtttime = 0; 3924 /* 3925 * Compute the amount of data that this ACK is indicating 3926 * (del_data) and an estimate of how many bytes are in the 3927 * network. 3928 */ 3929 if (SEQ_GEQ(th->th_ack, tp->snd_una)) 3930 del_data = BYTES_THIS_ACK(tp, th); 3931 del_data += tp->sackhint.delivered_data; 3932 pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; 3933 tp->sackhint.prr_delivered += del_data; 3934 /* 3935 * Proportional Rate Reduction 3936 */ 3937 if (pipe > tp->snd_ssthresh) 3938 snd_cnt = (tp->sackhint.prr_delivered * tp->snd_ssthresh / tp->sackhint.recover_fs) - 3939 tp->sackhint.sack_bytes_rexmit; 3940 else { 3941 if (V_tcp_do_prr_conservative) 3942 limit = tp->sackhint.prr_delivered - tp->sackhint.sack_bytes_rexmit; 3943 else 3944 if ((tp->sackhint.prr_delivered - tp->sackhint.sack_bytes_rexmit) > del_data) 3945 limit = tp->sackhint.prr_delivered - tp->sackhint.sack_bytes_rexmit + maxseg; 3946 else 3947 limit = del_data + maxseg; 3948 snd_cnt = min((tp->snd_ssthresh - pipe), limit); 3949 } 3950 snd_cnt = max((snd_cnt / maxseg), 0); 3951 /* 3952 * Send snd_cnt new data into the network in response to this ack. 3953 * If there is going to be a SACK retransmission, adjust snd_cwnd 3954 * accordingly. 3955 */ 3956 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + 3957 tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg); 3958 tp->t_flags |= TF_ACKNOW; 3959 (void) tcp_output(tp); 3960 } 3961 3962 /* 3963 * On a partial ack arrives, force the retransmission of the 3964 * next unacknowledged segment. Do not clear tp->t_dupacks. 3965 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3966 * be started again. 3967 */ 3968 void 3969 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3970 { 3971 tcp_seq onxt = tp->snd_nxt; 3972 uint32_t ocwnd = tp->snd_cwnd; 3973 u_int maxseg = tcp_maxseg(tp); 3974 3975 INP_WLOCK_ASSERT(tp->t_inpcb); 3976 3977 tcp_timer_activate(tp, TT_REXMT, 0); 3978 tp->t_rtttime = 0; 3979 tp->snd_nxt = th->th_ack; 3980 /* 3981 * Set snd_cwnd to one segment beyond acknowledged offset. 3982 * (tp->snd_una has not yet been updated when this function is called.) 3983 */ 3984 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3985 tp->t_flags |= TF_ACKNOW; 3986 (void) tp->t_fb->tfb_tcp_output(tp); 3987 tp->snd_cwnd = ocwnd; 3988 if (SEQ_GT(onxt, tp->snd_nxt)) 3989 tp->snd_nxt = onxt; 3990 /* 3991 * Partial window deflation. Relies on fact that tp->snd_una 3992 * not updated yet. 3993 */ 3994 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3995 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3996 else 3997 tp->snd_cwnd = 0; 3998 tp->snd_cwnd += maxseg; 3999 } 4000 4001 int 4002 tcp_compute_pipe(struct tcpcb *tp) 4003 { 4004 return (tp->snd_max - tp->snd_una + 4005 tp->sackhint.sack_bytes_rexmit - 4006 tp->sackhint.sacked_bytes); 4007 } 4008 4009 uint32_t 4010 tcp_compute_initwnd(uint32_t maxseg) 4011 { 4012 /* 4013 * Calculate the Initial Window, also used as Restart Window 4014 * 4015 * RFC5681 Section 3.1 specifies the default conservative values. 4016 * RFC3390 specifies slightly more aggressive values. 4017 * RFC6928 increases it to ten segments. 4018 * Support for user specified value for initial flight size. 4019 */ 4020 if (V_tcp_initcwnd_segments) 4021 return min(V_tcp_initcwnd_segments * maxseg, 4022 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 4023 else if (V_tcp_do_rfc3390) 4024 return min(4 * maxseg, max(2 * maxseg, 4380)); 4025 else { 4026 /* Per RFC5681 Section 3.1 */ 4027 if (maxseg > 2190) 4028 return (2 * maxseg); 4029 else if (maxseg > 1095) 4030 return (3 * maxseg); 4031 else 4032 return (4 * maxseg); 4033 } 4034 } 4035