1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 3. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_inet.h" 54 #include "opt_inet6.h" 55 #include "opt_ipsec.h" 56 #include "opt_tcpdebug.h" 57 58 #include <sys/param.h> 59 #include <sys/kernel.h> 60 #ifdef TCP_HHOOK 61 #include <sys/hhook.h> 62 #endif 63 #include <sys/malloc.h> 64 #include <sys/mbuf.h> 65 #include <sys/proc.h> /* for proc0 declaration */ 66 #include <sys/protosw.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77 #include <vm/uma.h> 78 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/route.h> 82 #include <net/vnet.h> 83 84 #define TCPSTATES /* for logging */ 85 86 #include <netinet/in.h> 87 #include <netinet/in_kdtrace.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/ip.h> 91 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 92 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 93 #include <netinet/ip_var.h> 94 #include <netinet/ip_options.h> 95 #include <netinet/ip6.h> 96 #include <netinet/icmp6.h> 97 #include <netinet6/in6_pcb.h> 98 #include <netinet6/in6_var.h> 99 #include <netinet6/ip6_var.h> 100 #include <netinet6/nd6.h> 101 #ifdef TCP_RFC7413 102 #include <netinet/tcp_fastopen.h> 103 #endif 104 #include <netinet/tcp.h> 105 #include <netinet/tcp_fsm.h> 106 #include <netinet/tcp_seq.h> 107 #include <netinet/tcp_timer.h> 108 #include <netinet/tcp_var.h> 109 #include <netinet6/tcp6_var.h> 110 #include <netinet/tcpip.h> 111 #include <netinet/cc/cc.h> 112 #ifdef TCPPCAP 113 #include <netinet/tcp_pcap.h> 114 #endif 115 #include <netinet/tcp_syncache.h> 116 #ifdef TCPDEBUG 117 #include <netinet/tcp_debug.h> 118 #endif /* TCPDEBUG */ 119 #ifdef TCP_OFFLOAD 120 #include <netinet/tcp_offload.h> 121 #endif 122 123 #include <netipsec/ipsec_support.h> 124 125 #include <machine/in_cksum.h> 126 127 #include <security/mac/mac_framework.h> 128 129 const int tcprexmtthresh = 3; 130 131 int tcp_log_in_vain = 0; 132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 133 &tcp_log_in_vain, 0, 134 "Log all incoming TCP segments to closed ports"); 135 136 VNET_DEFINE(int, blackhole) = 0; 137 #define V_blackhole VNET(blackhole) 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(blackhole), 0, 140 "Do not send RST on segments to closed ports"); 141 142 VNET_DEFINE(int, tcp_delack_enabled) = 1; 143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 144 &VNET_NAME(tcp_delack_enabled), 0, 145 "Delay ACK to try and piggyback it onto a data packet"); 146 147 VNET_DEFINE(int, drop_synfin) = 0; 148 #define V_drop_synfin VNET(drop_synfin) 149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 150 &VNET_NAME(drop_synfin), 0, 151 "Drop TCP packets with SYN+FIN set"); 152 153 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 155 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 156 "Use calculated pipe/in-flight bytes per RFC 6675"); 157 158 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 159 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 161 &VNET_NAME(tcp_do_rfc3042), 0, 162 "Enable RFC 3042 (Limited Transmit)"); 163 164 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 166 &VNET_NAME(tcp_do_rfc3390), 0, 167 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 168 169 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 171 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 172 "Slow-start flight size (initial congestion window) in number of segments"); 173 174 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 176 &VNET_NAME(tcp_do_rfc3465), 0, 177 "Enable RFC 3465 (Appropriate Byte Counting)"); 178 179 VNET_DEFINE(int, tcp_abc_l_var) = 2; 180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_abc_l_var), 2, 182 "Cap the max cwnd increment during slow-start to this number of segments"); 183 184 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 185 186 VNET_DEFINE(int, tcp_do_ecn) = 2; 187 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 188 &VNET_NAME(tcp_do_ecn), 0, 189 "TCP ECN support"); 190 191 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 192 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_ecn_maxretries), 0, 194 "Max retries before giving up on ECN"); 195 196 VNET_DEFINE(int, tcp_insecure_syn) = 0; 197 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_insecure_syn), 0, 200 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 201 202 VNET_DEFINE(int, tcp_insecure_rst) = 0; 203 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 205 &VNET_NAME(tcp_insecure_rst), 0, 206 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 207 208 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 209 #define V_tcp_recvspace VNET(tcp_recvspace) 210 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 211 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 212 213 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 214 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 216 &VNET_NAME(tcp_do_autorcvbuf), 0, 217 "Enable automatic receive buffer sizing"); 218 219 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 220 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_autorcvbuf_inc), 0, 223 "Incrementor step size of automatic receive buffer"); 224 225 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 226 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_autorcvbuf_max), 0, 229 "Max size of automatic receive buffer"); 230 231 VNET_DEFINE(struct inpcbhead, tcb); 232 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 233 VNET_DEFINE(struct inpcbinfo, tcbinfo); 234 235 /* 236 * TCP statistics are stored in an array of counter(9)s, which size matches 237 * size of struct tcpstat. TCP running connection count is a regular array. 238 */ 239 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 240 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 241 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 242 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 243 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 244 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 245 "TCP connection counts by TCP state"); 246 247 static void 248 tcp_vnet_init(const void *unused) 249 { 250 251 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 252 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 253 } 254 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 255 tcp_vnet_init, NULL); 256 257 #ifdef VIMAGE 258 static void 259 tcp_vnet_uninit(const void *unused) 260 { 261 262 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 263 VNET_PCPUSTAT_FREE(tcpstat); 264 } 265 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 266 tcp_vnet_uninit, NULL); 267 #endif /* VIMAGE */ 268 269 /* 270 * Kernel module interface for updating tcpstat. The argument is an index 271 * into tcpstat treated as an array. 272 */ 273 void 274 kmod_tcpstat_inc(int statnum) 275 { 276 277 counter_u64_add(VNET(tcpstat)[statnum], 1); 278 } 279 280 #ifdef TCP_HHOOK 281 /* 282 * Wrapper for the TCP established input helper hook. 283 */ 284 void 285 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 286 { 287 struct tcp_hhook_data hhook_data; 288 289 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 290 hhook_data.tp = tp; 291 hhook_data.th = th; 292 hhook_data.to = to; 293 294 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 295 tp->osd); 296 } 297 } 298 #endif 299 300 /* 301 * CC wrapper hook functions 302 */ 303 void 304 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 305 uint16_t type) 306 { 307 INP_WLOCK_ASSERT(tp->t_inpcb); 308 309 tp->ccv->nsegs = nsegs; 310 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 311 if (tp->snd_cwnd <= tp->snd_wnd) 312 tp->ccv->flags |= CCF_CWND_LIMITED; 313 else 314 tp->ccv->flags &= ~CCF_CWND_LIMITED; 315 316 if (type == CC_ACK) { 317 if (tp->snd_cwnd > tp->snd_ssthresh) { 318 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 319 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 320 if (tp->t_bytes_acked >= tp->snd_cwnd) { 321 tp->t_bytes_acked -= tp->snd_cwnd; 322 tp->ccv->flags |= CCF_ABC_SENTAWND; 323 } 324 } else { 325 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 326 tp->t_bytes_acked = 0; 327 } 328 } 329 330 if (CC_ALGO(tp)->ack_received != NULL) { 331 /* XXXLAS: Find a way to live without this */ 332 tp->ccv->curack = th->th_ack; 333 CC_ALGO(tp)->ack_received(tp->ccv, type); 334 } 335 } 336 337 void 338 cc_conn_init(struct tcpcb *tp) 339 { 340 struct hc_metrics_lite metrics; 341 struct inpcb *inp = tp->t_inpcb; 342 u_int maxseg; 343 int rtt; 344 345 INP_WLOCK_ASSERT(tp->t_inpcb); 346 347 tcp_hc_get(&inp->inp_inc, &metrics); 348 maxseg = tcp_maxseg(tp); 349 350 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 351 tp->t_srtt = rtt; 352 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 353 TCPSTAT_INC(tcps_usedrtt); 354 if (metrics.rmx_rttvar) { 355 tp->t_rttvar = metrics.rmx_rttvar; 356 TCPSTAT_INC(tcps_usedrttvar); 357 } else { 358 /* default variation is +- 1 rtt */ 359 tp->t_rttvar = 360 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 361 } 362 TCPT_RANGESET(tp->t_rxtcur, 363 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 364 tp->t_rttmin, TCPTV_REXMTMAX); 365 } 366 if (metrics.rmx_ssthresh) { 367 /* 368 * There's some sort of gateway or interface 369 * buffer limit on the path. Use this to set 370 * the slow start threshold, but set the 371 * threshold to no less than 2*mss. 372 */ 373 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 374 TCPSTAT_INC(tcps_usedssthresh); 375 } 376 377 /* 378 * Set the initial slow-start flight size. 379 * 380 * RFC5681 Section 3.1 specifies the default conservative values. 381 * RFC3390 specifies slightly more aggressive values. 382 * RFC6928 increases it to ten segments. 383 * Support for user specified value for initial flight size. 384 * 385 * If a SYN or SYN/ACK was lost and retransmitted, we have to 386 * reduce the initial CWND to one segment as congestion is likely 387 * requiring us to be cautious. 388 */ 389 if (tp->snd_cwnd == 1) 390 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 391 else if (V_tcp_initcwnd_segments) 392 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 393 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 394 else if (V_tcp_do_rfc3390) 395 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 396 else { 397 /* Per RFC5681 Section 3.1 */ 398 if (maxseg > 2190) 399 tp->snd_cwnd = 2 * maxseg; 400 else if (maxseg > 1095) 401 tp->snd_cwnd = 3 * maxseg; 402 else 403 tp->snd_cwnd = 4 * maxseg; 404 } 405 406 if (CC_ALGO(tp)->conn_init != NULL) 407 CC_ALGO(tp)->conn_init(tp->ccv); 408 } 409 410 void inline 411 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 412 { 413 u_int maxseg; 414 415 INP_WLOCK_ASSERT(tp->t_inpcb); 416 417 switch(type) { 418 case CC_NDUPACK: 419 if (!IN_FASTRECOVERY(tp->t_flags)) { 420 tp->snd_recover = tp->snd_max; 421 if (tp->t_flags & TF_ECN_PERMIT) 422 tp->t_flags |= TF_ECN_SND_CWR; 423 } 424 break; 425 case CC_ECN: 426 if (!IN_CONGRECOVERY(tp->t_flags)) { 427 TCPSTAT_INC(tcps_ecn_rcwnd); 428 tp->snd_recover = tp->snd_max; 429 if (tp->t_flags & TF_ECN_PERMIT) 430 tp->t_flags |= TF_ECN_SND_CWR; 431 } 432 break; 433 case CC_RTO: 434 maxseg = tcp_maxseg(tp); 435 tp->t_dupacks = 0; 436 tp->t_bytes_acked = 0; 437 EXIT_RECOVERY(tp->t_flags); 438 if (CC_ALGO(tp)->cong_signal == NULL) { 439 /* 440 * RFC5681 Section 3.1 441 * ssthresh = max (FlightSize / 2, 2*SMSS) eq (4) 442 */ 443 tp->snd_ssthresh = 444 max((tp->snd_max - tp->snd_una) / 2 / maxseg, 2) 445 * maxseg; 446 tp->snd_cwnd = maxseg; 447 } 448 break; 449 case CC_RTO_ERR: 450 TCPSTAT_INC(tcps_sndrexmitbad); 451 /* RTO was unnecessary, so reset everything. */ 452 tp->snd_cwnd = tp->snd_cwnd_prev; 453 tp->snd_ssthresh = tp->snd_ssthresh_prev; 454 tp->snd_recover = tp->snd_recover_prev; 455 if (tp->t_flags & TF_WASFRECOVERY) 456 ENTER_FASTRECOVERY(tp->t_flags); 457 if (tp->t_flags & TF_WASCRECOVERY) 458 ENTER_CONGRECOVERY(tp->t_flags); 459 tp->snd_nxt = tp->snd_max; 460 tp->t_flags &= ~TF_PREVVALID; 461 tp->t_badrxtwin = 0; 462 break; 463 } 464 465 if (CC_ALGO(tp)->cong_signal != NULL) { 466 if (th != NULL) 467 tp->ccv->curack = th->th_ack; 468 CC_ALGO(tp)->cong_signal(tp->ccv, type); 469 } 470 } 471 472 void inline 473 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 474 { 475 INP_WLOCK_ASSERT(tp->t_inpcb); 476 477 /* XXXLAS: KASSERT that we're in recovery? */ 478 479 if (CC_ALGO(tp)->post_recovery != NULL) { 480 tp->ccv->curack = th->th_ack; 481 CC_ALGO(tp)->post_recovery(tp->ccv); 482 } 483 /* XXXLAS: EXIT_RECOVERY ? */ 484 tp->t_bytes_acked = 0; 485 } 486 487 /* 488 * Indicate whether this ack should be delayed. We can delay the ack if 489 * following conditions are met: 490 * - There is no delayed ack timer in progress. 491 * - Our last ack wasn't a 0-sized window. We never want to delay 492 * the ack that opens up a 0-sized window. 493 * - LRO wasn't used for this segment. We make sure by checking that the 494 * segment size is not larger than the MSS. 495 */ 496 #define DELAY_ACK(tp, tlen) \ 497 ((!tcp_timer_active(tp, TT_DELACK) && \ 498 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 499 (tlen <= tp->t_maxseg) && \ 500 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 501 502 static void inline 503 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 504 { 505 INP_WLOCK_ASSERT(tp->t_inpcb); 506 507 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 508 switch (iptos & IPTOS_ECN_MASK) { 509 case IPTOS_ECN_CE: 510 tp->ccv->flags |= CCF_IPHDR_CE; 511 break; 512 case IPTOS_ECN_ECT0: 513 tp->ccv->flags &= ~CCF_IPHDR_CE; 514 break; 515 case IPTOS_ECN_ECT1: 516 tp->ccv->flags &= ~CCF_IPHDR_CE; 517 break; 518 } 519 520 if (th->th_flags & TH_CWR) 521 tp->ccv->flags |= CCF_TCPHDR_CWR; 522 else 523 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 524 525 if (tp->t_flags & TF_DELACK) 526 tp->ccv->flags |= CCF_DELACK; 527 else 528 tp->ccv->flags &= ~CCF_DELACK; 529 530 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 531 532 if (tp->ccv->flags & CCF_ACKNOW) 533 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 534 } 535 } 536 537 /* 538 * TCP input handling is split into multiple parts: 539 * tcp6_input is a thin wrapper around tcp_input for the extended 540 * ip6_protox[] call format in ip6_input 541 * tcp_input handles primary segment validation, inpcb lookup and 542 * SYN processing on listen sockets 543 * tcp_do_segment processes the ACK and text of the segment for 544 * establishing, established and closing connections 545 */ 546 #ifdef INET6 547 int 548 tcp6_input(struct mbuf **mp, int *offp, int proto) 549 { 550 struct mbuf *m = *mp; 551 struct in6_ifaddr *ia6; 552 struct ip6_hdr *ip6; 553 554 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 555 556 /* 557 * draft-itojun-ipv6-tcp-to-anycast 558 * better place to put this in? 559 */ 560 ip6 = mtod(m, struct ip6_hdr *); 561 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 562 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 563 struct ip6_hdr *ip6; 564 565 ifa_free(&ia6->ia_ifa); 566 ip6 = mtod(m, struct ip6_hdr *); 567 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 568 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 569 return (IPPROTO_DONE); 570 } 571 if (ia6) 572 ifa_free(&ia6->ia_ifa); 573 574 return (tcp_input(mp, offp, proto)); 575 } 576 #endif /* INET6 */ 577 578 int 579 tcp_input(struct mbuf **mp, int *offp, int proto) 580 { 581 struct mbuf *m = *mp; 582 struct tcphdr *th = NULL; 583 struct ip *ip = NULL; 584 struct inpcb *inp = NULL; 585 struct tcpcb *tp = NULL; 586 struct socket *so = NULL; 587 u_char *optp = NULL; 588 int off0; 589 int optlen = 0; 590 #ifdef INET 591 int len; 592 #endif 593 int tlen = 0, off; 594 int drop_hdrlen; 595 int thflags; 596 int rstreason = 0; /* For badport_bandlim accounting purposes */ 597 uint8_t iptos; 598 struct m_tag *fwd_tag = NULL; 599 #ifdef INET6 600 struct ip6_hdr *ip6 = NULL; 601 int isipv6; 602 #else 603 const void *ip6 = NULL; 604 #endif /* INET6 */ 605 struct tcpopt to; /* options in this segment */ 606 char *s = NULL; /* address and port logging */ 607 int ti_locked; 608 #ifdef TCPDEBUG 609 /* 610 * The size of tcp_saveipgen must be the size of the max ip header, 611 * now IPv6. 612 */ 613 u_char tcp_saveipgen[IP6_HDR_LEN]; 614 struct tcphdr tcp_savetcp; 615 short ostate = 0; 616 #endif 617 618 #ifdef INET6 619 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 620 #endif 621 622 off0 = *offp; 623 m = *mp; 624 *mp = NULL; 625 to.to_flags = 0; 626 TCPSTAT_INC(tcps_rcvtotal); 627 628 #ifdef INET6 629 if (isipv6) { 630 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 631 632 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 633 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 634 if (m == NULL) { 635 TCPSTAT_INC(tcps_rcvshort); 636 return (IPPROTO_DONE); 637 } 638 } 639 640 ip6 = mtod(m, struct ip6_hdr *); 641 th = (struct tcphdr *)((caddr_t)ip6 + off0); 642 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 643 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 644 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 645 th->th_sum = m->m_pkthdr.csum_data; 646 else 647 th->th_sum = in6_cksum_pseudo(ip6, tlen, 648 IPPROTO_TCP, m->m_pkthdr.csum_data); 649 th->th_sum ^= 0xffff; 650 } else 651 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 652 if (th->th_sum) { 653 TCPSTAT_INC(tcps_rcvbadsum); 654 goto drop; 655 } 656 657 /* 658 * Be proactive about unspecified IPv6 address in source. 659 * As we use all-zero to indicate unbounded/unconnected pcb, 660 * unspecified IPv6 address can be used to confuse us. 661 * 662 * Note that packets with unspecified IPv6 destination is 663 * already dropped in ip6_input. 664 */ 665 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 666 /* XXX stat */ 667 goto drop; 668 } 669 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 670 } 671 #endif 672 #if defined(INET) && defined(INET6) 673 else 674 #endif 675 #ifdef INET 676 { 677 /* 678 * Get IP and TCP header together in first mbuf. 679 * Note: IP leaves IP header in first mbuf. 680 */ 681 if (off0 > sizeof (struct ip)) { 682 ip_stripoptions(m); 683 off0 = sizeof(struct ip); 684 } 685 if (m->m_len < sizeof (struct tcpiphdr)) { 686 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 687 == NULL) { 688 TCPSTAT_INC(tcps_rcvshort); 689 return (IPPROTO_DONE); 690 } 691 } 692 ip = mtod(m, struct ip *); 693 th = (struct tcphdr *)((caddr_t)ip + off0); 694 tlen = ntohs(ip->ip_len) - off0; 695 696 iptos = ip->ip_tos; 697 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 698 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 699 th->th_sum = m->m_pkthdr.csum_data; 700 else 701 th->th_sum = in_pseudo(ip->ip_src.s_addr, 702 ip->ip_dst.s_addr, 703 htonl(m->m_pkthdr.csum_data + tlen + 704 IPPROTO_TCP)); 705 th->th_sum ^= 0xffff; 706 } else { 707 struct ipovly *ipov = (struct ipovly *)ip; 708 709 /* 710 * Checksum extended TCP header and data. 711 */ 712 len = off0 + tlen; 713 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 714 ipov->ih_len = htons(tlen); 715 th->th_sum = in_cksum(m, len); 716 /* Reset length for SDT probes. */ 717 ip->ip_len = htons(len); 718 /* Reset TOS bits */ 719 ip->ip_tos = iptos; 720 /* Re-initialization for later version check */ 721 ip->ip_v = IPVERSION; 722 } 723 724 if (th->th_sum) { 725 TCPSTAT_INC(tcps_rcvbadsum); 726 goto drop; 727 } 728 } 729 #endif /* INET */ 730 731 /* 732 * Check that TCP offset makes sense, 733 * pull out TCP options and adjust length. XXX 734 */ 735 off = th->th_off << 2; 736 if (off < sizeof (struct tcphdr) || off > tlen) { 737 TCPSTAT_INC(tcps_rcvbadoff); 738 goto drop; 739 } 740 tlen -= off; /* tlen is used instead of ti->ti_len */ 741 if (off > sizeof (struct tcphdr)) { 742 #ifdef INET6 743 if (isipv6) { 744 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 745 ip6 = mtod(m, struct ip6_hdr *); 746 th = (struct tcphdr *)((caddr_t)ip6 + off0); 747 } 748 #endif 749 #if defined(INET) && defined(INET6) 750 else 751 #endif 752 #ifdef INET 753 { 754 if (m->m_len < sizeof(struct ip) + off) { 755 if ((m = m_pullup(m, sizeof (struct ip) + off)) 756 == NULL) { 757 TCPSTAT_INC(tcps_rcvshort); 758 return (IPPROTO_DONE); 759 } 760 ip = mtod(m, struct ip *); 761 th = (struct tcphdr *)((caddr_t)ip + off0); 762 } 763 } 764 #endif 765 optlen = off - sizeof (struct tcphdr); 766 optp = (u_char *)(th + 1); 767 } 768 thflags = th->th_flags; 769 770 /* 771 * Convert TCP protocol specific fields to host format. 772 */ 773 tcp_fields_to_host(th); 774 775 /* 776 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 777 */ 778 drop_hdrlen = off0 + off; 779 780 /* 781 * Locate pcb for segment; if we're likely to add or remove a 782 * connection then first acquire pcbinfo lock. There are three cases 783 * where we might discover later we need a write lock despite the 784 * flags: ACKs moving a connection out of the syncache, ACKs for a 785 * connection in TIMEWAIT and SYNs not targeting a listening socket. 786 */ 787 if ((thflags & (TH_FIN | TH_RST)) != 0) { 788 INP_INFO_RLOCK(&V_tcbinfo); 789 ti_locked = TI_RLOCKED; 790 } else 791 ti_locked = TI_UNLOCKED; 792 793 /* 794 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 795 */ 796 if ( 797 #ifdef INET6 798 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 799 #ifdef INET 800 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 801 #endif 802 #endif 803 #if defined(INET) && !defined(INET6) 804 (m->m_flags & M_IP_NEXTHOP) 805 #endif 806 ) 807 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 808 809 findpcb: 810 #ifdef INVARIANTS 811 if (ti_locked == TI_RLOCKED) { 812 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 813 } else { 814 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 815 } 816 #endif 817 #ifdef INET6 818 if (isipv6 && fwd_tag != NULL) { 819 struct sockaddr_in6 *next_hop6; 820 821 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 822 /* 823 * Transparently forwarded. Pretend to be the destination. 824 * Already got one like this? 825 */ 826 inp = in6_pcblookup_mbuf(&V_tcbinfo, 827 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 828 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 829 if (!inp) { 830 /* 831 * It's new. Try to find the ambushing socket. 832 * Because we've rewritten the destination address, 833 * any hardware-generated hash is ignored. 834 */ 835 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 836 th->th_sport, &next_hop6->sin6_addr, 837 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 838 th->th_dport, INPLOOKUP_WILDCARD | 839 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 840 } 841 } else if (isipv6) { 842 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 843 th->th_sport, &ip6->ip6_dst, th->th_dport, 844 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 845 m->m_pkthdr.rcvif, m); 846 } 847 #endif /* INET6 */ 848 #if defined(INET6) && defined(INET) 849 else 850 #endif 851 #ifdef INET 852 if (fwd_tag != NULL) { 853 struct sockaddr_in *next_hop; 854 855 next_hop = (struct sockaddr_in *)(fwd_tag+1); 856 /* 857 * Transparently forwarded. Pretend to be the destination. 858 * already got one like this? 859 */ 860 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 861 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 862 m->m_pkthdr.rcvif, m); 863 if (!inp) { 864 /* 865 * It's new. Try to find the ambushing socket. 866 * Because we've rewritten the destination address, 867 * any hardware-generated hash is ignored. 868 */ 869 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 870 th->th_sport, next_hop->sin_addr, 871 next_hop->sin_port ? ntohs(next_hop->sin_port) : 872 th->th_dport, INPLOOKUP_WILDCARD | 873 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 874 } 875 } else 876 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 877 th->th_sport, ip->ip_dst, th->th_dport, 878 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 879 m->m_pkthdr.rcvif, m); 880 #endif /* INET */ 881 882 /* 883 * If the INPCB does not exist then all data in the incoming 884 * segment is discarded and an appropriate RST is sent back. 885 * XXX MRT Send RST using which routing table? 886 */ 887 if (inp == NULL) { 888 /* 889 * Log communication attempts to ports that are not 890 * in use. 891 */ 892 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 893 tcp_log_in_vain == 2) { 894 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 895 log(LOG_INFO, "%s; %s: Connection attempt " 896 "to closed port\n", s, __func__); 897 } 898 /* 899 * When blackholing do not respond with a RST but 900 * completely ignore the segment and drop it. 901 */ 902 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 903 V_blackhole == 2) 904 goto dropunlock; 905 906 rstreason = BANDLIM_RST_CLOSEDPORT; 907 goto dropwithreset; 908 } 909 INP_WLOCK_ASSERT(inp); 910 /* 911 * While waiting for inp lock during the lookup, another thread 912 * can have dropped the inpcb, in which case we need to loop back 913 * and try to find a new inpcb to deliver to. 914 */ 915 if (inp->inp_flags & INP_DROPPED) { 916 INP_WUNLOCK(inp); 917 inp = NULL; 918 goto findpcb; 919 } 920 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 921 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 922 ((inp->inp_socket == NULL) || 923 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 924 inp->inp_flowid = m->m_pkthdr.flowid; 925 inp->inp_flowtype = M_HASHTYPE_GET(m); 926 } 927 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 928 #ifdef INET6 929 if (isipv6 && IPSEC_ENABLED(ipv6) && 930 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 931 goto dropunlock; 932 } 933 #ifdef INET 934 else 935 #endif 936 #endif /* INET6 */ 937 #ifdef INET 938 if (IPSEC_ENABLED(ipv4) && 939 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 940 goto dropunlock; 941 } 942 #endif /* INET */ 943 #endif /* IPSEC */ 944 945 /* 946 * Check the minimum TTL for socket. 947 */ 948 if (inp->inp_ip_minttl != 0) { 949 #ifdef INET6 950 if (isipv6) { 951 if (inp->inp_ip_minttl > ip6->ip6_hlim) 952 goto dropunlock; 953 } else 954 #endif 955 if (inp->inp_ip_minttl > ip->ip_ttl) 956 goto dropunlock; 957 } 958 959 /* 960 * A previous connection in TIMEWAIT state is supposed to catch stray 961 * or duplicate segments arriving late. If this segment was a 962 * legitimate new connection attempt, the old INPCB gets removed and 963 * we can try again to find a listening socket. 964 * 965 * At this point, due to earlier optimism, we may hold only an inpcb 966 * lock, and not the inpcbinfo write lock. If so, we need to try to 967 * acquire it, or if that fails, acquire a reference on the inpcb, 968 * drop all locks, acquire a global write lock, and then re-acquire 969 * the inpcb lock. We may at that point discover that another thread 970 * has tried to free the inpcb, in which case we need to loop back 971 * and try to find a new inpcb to deliver to. 972 * 973 * XXXRW: It may be time to rethink timewait locking. 974 */ 975 relocked: 976 if (inp->inp_flags & INP_TIMEWAIT) { 977 if (ti_locked == TI_UNLOCKED) { 978 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 979 in_pcbref(inp); 980 INP_WUNLOCK(inp); 981 INP_INFO_RLOCK(&V_tcbinfo); 982 ti_locked = TI_RLOCKED; 983 INP_WLOCK(inp); 984 if (in_pcbrele_wlocked(inp)) { 985 inp = NULL; 986 goto findpcb; 987 } else if (inp->inp_flags & INP_DROPPED) { 988 INP_WUNLOCK(inp); 989 inp = NULL; 990 goto findpcb; 991 } 992 } else 993 ti_locked = TI_RLOCKED; 994 } 995 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 996 997 if (thflags & TH_SYN) 998 tcp_dooptions(&to, optp, optlen, TO_SYN); 999 /* 1000 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1001 */ 1002 if (tcp_twcheck(inp, &to, th, m, tlen)) 1003 goto findpcb; 1004 INP_INFO_RUNLOCK(&V_tcbinfo); 1005 return (IPPROTO_DONE); 1006 } 1007 /* 1008 * The TCPCB may no longer exist if the connection is winding 1009 * down or it is in the CLOSED state. Either way we drop the 1010 * segment and send an appropriate response. 1011 */ 1012 tp = intotcpcb(inp); 1013 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1014 rstreason = BANDLIM_RST_CLOSEDPORT; 1015 goto dropwithreset; 1016 } 1017 1018 #ifdef TCP_OFFLOAD 1019 if (tp->t_flags & TF_TOE) { 1020 tcp_offload_input(tp, m); 1021 m = NULL; /* consumed by the TOE driver */ 1022 goto dropunlock; 1023 } 1024 #endif 1025 1026 /* 1027 * We've identified a valid inpcb, but it could be that we need an 1028 * inpcbinfo write lock but don't hold it. In this case, attempt to 1029 * acquire using the same strategy as the TIMEWAIT case above. If we 1030 * relock, we have to jump back to 'relocked' as the connection might 1031 * now be in TIMEWAIT. 1032 */ 1033 #ifdef INVARIANTS 1034 if ((thflags & (TH_FIN | TH_RST)) != 0) 1035 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1036 #endif 1037 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1038 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1039 !IS_FASTOPEN(tp->t_flags)))) { 1040 if (ti_locked == TI_UNLOCKED) { 1041 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1042 in_pcbref(inp); 1043 INP_WUNLOCK(inp); 1044 INP_INFO_RLOCK(&V_tcbinfo); 1045 ti_locked = TI_RLOCKED; 1046 INP_WLOCK(inp); 1047 if (in_pcbrele_wlocked(inp)) { 1048 inp = NULL; 1049 goto findpcb; 1050 } else if (inp->inp_flags & INP_DROPPED) { 1051 INP_WUNLOCK(inp); 1052 inp = NULL; 1053 goto findpcb; 1054 } 1055 goto relocked; 1056 } else 1057 ti_locked = TI_RLOCKED; 1058 } 1059 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1060 } 1061 1062 #ifdef MAC 1063 INP_WLOCK_ASSERT(inp); 1064 if (mac_inpcb_check_deliver(inp, m)) 1065 goto dropunlock; 1066 #endif 1067 so = inp->inp_socket; 1068 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1069 #ifdef TCPDEBUG 1070 if (so->so_options & SO_DEBUG) { 1071 ostate = tp->t_state; 1072 #ifdef INET6 1073 if (isipv6) { 1074 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1075 } else 1076 #endif 1077 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1078 tcp_savetcp = *th; 1079 } 1080 #endif /* TCPDEBUG */ 1081 /* 1082 * When the socket is accepting connections (the INPCB is in LISTEN 1083 * state) we look into the SYN cache if this is a new connection 1084 * attempt or the completion of a previous one. 1085 */ 1086 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1087 ("%s: so accepting but tp %p not listening", __func__, tp)); 1088 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1089 struct in_conninfo inc; 1090 1091 bzero(&inc, sizeof(inc)); 1092 #ifdef INET6 1093 if (isipv6) { 1094 inc.inc_flags |= INC_ISIPV6; 1095 inc.inc6_faddr = ip6->ip6_src; 1096 inc.inc6_laddr = ip6->ip6_dst; 1097 } else 1098 #endif 1099 { 1100 inc.inc_faddr = ip->ip_src; 1101 inc.inc_laddr = ip->ip_dst; 1102 } 1103 inc.inc_fport = th->th_sport; 1104 inc.inc_lport = th->th_dport; 1105 inc.inc_fibnum = so->so_fibnum; 1106 1107 /* 1108 * Check for an existing connection attempt in syncache if 1109 * the flag is only ACK. A successful lookup creates a new 1110 * socket appended to the listen queue in SYN_RECEIVED state. 1111 */ 1112 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1113 1114 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1115 /* 1116 * Parse the TCP options here because 1117 * syncookies need access to the reflected 1118 * timestamp. 1119 */ 1120 tcp_dooptions(&to, optp, optlen, 0); 1121 /* 1122 * NB: syncache_expand() doesn't unlock 1123 * inp and tcpinfo locks. 1124 */ 1125 rstreason = syncache_expand(&inc, &to, th, &so, m); 1126 if (rstreason < 0) { 1127 /* 1128 * A failing TCP MD5 signature comparison 1129 * must result in the segment being dropped 1130 * and must not produce any response back 1131 * to the sender. 1132 */ 1133 goto dropunlock; 1134 } else if (rstreason == 0) { 1135 /* 1136 * No syncache entry or ACK was not 1137 * for our SYN/ACK. Send a RST. 1138 * NB: syncache did its own logging 1139 * of the failure cause. 1140 */ 1141 rstreason = BANDLIM_RST_OPENPORT; 1142 goto dropwithreset; 1143 } 1144 #ifdef TCP_RFC7413 1145 tfo_socket_result: 1146 #endif 1147 if (so == NULL) { 1148 /* 1149 * We completed the 3-way handshake 1150 * but could not allocate a socket 1151 * either due to memory shortage, 1152 * listen queue length limits or 1153 * global socket limits. Send RST 1154 * or wait and have the remote end 1155 * retransmit the ACK for another 1156 * try. 1157 */ 1158 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1159 log(LOG_DEBUG, "%s; %s: Listen socket: " 1160 "Socket allocation failed due to " 1161 "limits or memory shortage, %s\n", 1162 s, __func__, 1163 V_tcp_sc_rst_sock_fail ? 1164 "sending RST" : "try again"); 1165 if (V_tcp_sc_rst_sock_fail) { 1166 rstreason = BANDLIM_UNLIMITED; 1167 goto dropwithreset; 1168 } else 1169 goto dropunlock; 1170 } 1171 /* 1172 * Socket is created in state SYN_RECEIVED. 1173 * Unlock the listen socket, lock the newly 1174 * created socket and update the tp variable. 1175 */ 1176 INP_WUNLOCK(inp); /* listen socket */ 1177 inp = sotoinpcb(so); 1178 /* 1179 * New connection inpcb is already locked by 1180 * syncache_expand(). 1181 */ 1182 INP_WLOCK_ASSERT(inp); 1183 tp = intotcpcb(inp); 1184 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1185 ("%s: ", __func__)); 1186 /* 1187 * Process the segment and the data it 1188 * contains. tcp_do_segment() consumes 1189 * the mbuf chain and unlocks the inpcb. 1190 */ 1191 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1192 iptos, ti_locked); 1193 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1194 return (IPPROTO_DONE); 1195 } 1196 /* 1197 * Segment flag validation for new connection attempts: 1198 * 1199 * Our (SYN|ACK) response was rejected. 1200 * Check with syncache and remove entry to prevent 1201 * retransmits. 1202 * 1203 * NB: syncache_chkrst does its own logging of failure 1204 * causes. 1205 */ 1206 if (thflags & TH_RST) { 1207 syncache_chkrst(&inc, th); 1208 goto dropunlock; 1209 } 1210 /* 1211 * We can't do anything without SYN. 1212 */ 1213 if ((thflags & TH_SYN) == 0) { 1214 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1215 log(LOG_DEBUG, "%s; %s: Listen socket: " 1216 "SYN is missing, segment ignored\n", 1217 s, __func__); 1218 TCPSTAT_INC(tcps_badsyn); 1219 goto dropunlock; 1220 } 1221 /* 1222 * (SYN|ACK) is bogus on a listen socket. 1223 */ 1224 if (thflags & TH_ACK) { 1225 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1226 log(LOG_DEBUG, "%s; %s: Listen socket: " 1227 "SYN|ACK invalid, segment rejected\n", 1228 s, __func__); 1229 syncache_badack(&inc); /* XXX: Not needed! */ 1230 TCPSTAT_INC(tcps_badsyn); 1231 rstreason = BANDLIM_RST_OPENPORT; 1232 goto dropwithreset; 1233 } 1234 /* 1235 * If the drop_synfin option is enabled, drop all 1236 * segments with both the SYN and FIN bits set. 1237 * This prevents e.g. nmap from identifying the 1238 * TCP/IP stack. 1239 * XXX: Poor reasoning. nmap has other methods 1240 * and is constantly refining its stack detection 1241 * strategies. 1242 * XXX: This is a violation of the TCP specification 1243 * and was used by RFC1644. 1244 */ 1245 if ((thflags & TH_FIN) && V_drop_synfin) { 1246 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1247 log(LOG_DEBUG, "%s; %s: Listen socket: " 1248 "SYN|FIN segment ignored (based on " 1249 "sysctl setting)\n", s, __func__); 1250 TCPSTAT_INC(tcps_badsyn); 1251 goto dropunlock; 1252 } 1253 /* 1254 * Segment's flags are (SYN) or (SYN|FIN). 1255 * 1256 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1257 * as they do not affect the state of the TCP FSM. 1258 * The data pointed to by TH_URG and th_urp is ignored. 1259 */ 1260 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1261 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1262 KASSERT(thflags & (TH_SYN), 1263 ("%s: Listen socket: TH_SYN not set", __func__)); 1264 #ifdef INET6 1265 /* 1266 * If deprecated address is forbidden, 1267 * we do not accept SYN to deprecated interface 1268 * address to prevent any new inbound connection from 1269 * getting established. 1270 * When we do not accept SYN, we send a TCP RST, 1271 * with deprecated source address (instead of dropping 1272 * it). We compromise it as it is much better for peer 1273 * to send a RST, and RST will be the final packet 1274 * for the exchange. 1275 * 1276 * If we do not forbid deprecated addresses, we accept 1277 * the SYN packet. RFC2462 does not suggest dropping 1278 * SYN in this case. 1279 * If we decipher RFC2462 5.5.4, it says like this: 1280 * 1. use of deprecated addr with existing 1281 * communication is okay - "SHOULD continue to be 1282 * used" 1283 * 2. use of it with new communication: 1284 * (2a) "SHOULD NOT be used if alternate address 1285 * with sufficient scope is available" 1286 * (2b) nothing mentioned otherwise. 1287 * Here we fall into (2b) case as we have no choice in 1288 * our source address selection - we must obey the peer. 1289 * 1290 * The wording in RFC2462 is confusing, and there are 1291 * multiple description text for deprecated address 1292 * handling - worse, they are not exactly the same. 1293 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1294 */ 1295 if (isipv6 && !V_ip6_use_deprecated) { 1296 struct in6_ifaddr *ia6; 1297 1298 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1299 if (ia6 != NULL && 1300 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1301 ifa_free(&ia6->ia_ifa); 1302 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1303 log(LOG_DEBUG, "%s; %s: Listen socket: " 1304 "Connection attempt to deprecated " 1305 "IPv6 address rejected\n", 1306 s, __func__); 1307 rstreason = BANDLIM_RST_OPENPORT; 1308 goto dropwithreset; 1309 } 1310 if (ia6) 1311 ifa_free(&ia6->ia_ifa); 1312 } 1313 #endif /* INET6 */ 1314 /* 1315 * Basic sanity checks on incoming SYN requests: 1316 * Don't respond if the destination is a link layer 1317 * broadcast according to RFC1122 4.2.3.10, p. 104. 1318 * If it is from this socket it must be forged. 1319 * Don't respond if the source or destination is a 1320 * global or subnet broad- or multicast address. 1321 * Note that it is quite possible to receive unicast 1322 * link-layer packets with a broadcast IP address. Use 1323 * in_broadcast() to find them. 1324 */ 1325 if (m->m_flags & (M_BCAST|M_MCAST)) { 1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1327 log(LOG_DEBUG, "%s; %s: Listen socket: " 1328 "Connection attempt from broad- or multicast " 1329 "link layer address ignored\n", s, __func__); 1330 goto dropunlock; 1331 } 1332 #ifdef INET6 1333 if (isipv6) { 1334 if (th->th_dport == th->th_sport && 1335 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1336 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1337 log(LOG_DEBUG, "%s; %s: Listen socket: " 1338 "Connection attempt to/from self " 1339 "ignored\n", s, __func__); 1340 goto dropunlock; 1341 } 1342 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1343 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1344 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1345 log(LOG_DEBUG, "%s; %s: Listen socket: " 1346 "Connection attempt from/to multicast " 1347 "address ignored\n", s, __func__); 1348 goto dropunlock; 1349 } 1350 } 1351 #endif 1352 #if defined(INET) && defined(INET6) 1353 else 1354 #endif 1355 #ifdef INET 1356 { 1357 if (th->th_dport == th->th_sport && 1358 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1359 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1360 log(LOG_DEBUG, "%s; %s: Listen socket: " 1361 "Connection attempt from/to self " 1362 "ignored\n", s, __func__); 1363 goto dropunlock; 1364 } 1365 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1366 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1367 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1368 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1369 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1370 log(LOG_DEBUG, "%s; %s: Listen socket: " 1371 "Connection attempt from/to broad- " 1372 "or multicast address ignored\n", 1373 s, __func__); 1374 goto dropunlock; 1375 } 1376 } 1377 #endif 1378 /* 1379 * SYN appears to be valid. Create compressed TCP state 1380 * for syncache. 1381 */ 1382 #ifdef TCPDEBUG 1383 if (so->so_options & SO_DEBUG) 1384 tcp_trace(TA_INPUT, ostate, tp, 1385 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1386 #endif 1387 TCP_PROBE3(debug__input, tp, th, m); 1388 tcp_dooptions(&to, optp, optlen, TO_SYN); 1389 #ifdef TCP_RFC7413 1390 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1391 goto tfo_socket_result; 1392 #else 1393 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1394 #endif 1395 /* 1396 * Entry added to syncache and mbuf consumed. 1397 * Only the listen socket is unlocked by syncache_add(). 1398 */ 1399 if (ti_locked == TI_RLOCKED) { 1400 INP_INFO_RUNLOCK(&V_tcbinfo); 1401 ti_locked = TI_UNLOCKED; 1402 } 1403 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1404 return (IPPROTO_DONE); 1405 } else if (tp->t_state == TCPS_LISTEN) { 1406 /* 1407 * When a listen socket is torn down the SO_ACCEPTCONN 1408 * flag is removed first while connections are drained 1409 * from the accept queue in a unlock/lock cycle of the 1410 * ACCEPT_LOCK, opening a race condition allowing a SYN 1411 * attempt go through unhandled. 1412 */ 1413 goto dropunlock; 1414 } 1415 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1416 if (tp->t_flags & TF_SIGNATURE) { 1417 tcp_dooptions(&to, optp, optlen, thflags); 1418 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1419 TCPSTAT_INC(tcps_sig_err_nosigopt); 1420 goto dropunlock; 1421 } 1422 if (!TCPMD5_ENABLED() || 1423 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1424 goto dropunlock; 1425 } 1426 #endif 1427 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1428 1429 /* 1430 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1431 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1432 * the inpcb, and unlocks pcbinfo. 1433 */ 1434 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1435 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1436 return (IPPROTO_DONE); 1437 1438 dropwithreset: 1439 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1440 1441 if (ti_locked == TI_RLOCKED) { 1442 INP_INFO_RUNLOCK(&V_tcbinfo); 1443 ti_locked = TI_UNLOCKED; 1444 } 1445 #ifdef INVARIANTS 1446 else { 1447 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1448 "ti_locked: %d", __func__, ti_locked)); 1449 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1450 } 1451 #endif 1452 1453 if (inp != NULL) { 1454 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1455 INP_WUNLOCK(inp); 1456 } else 1457 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1458 m = NULL; /* mbuf chain got consumed. */ 1459 goto drop; 1460 1461 dropunlock: 1462 if (m != NULL) 1463 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1464 1465 if (ti_locked == TI_RLOCKED) { 1466 INP_INFO_RUNLOCK(&V_tcbinfo); 1467 ti_locked = TI_UNLOCKED; 1468 } 1469 #ifdef INVARIANTS 1470 else { 1471 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1472 "ti_locked: %d", __func__, ti_locked)); 1473 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1474 } 1475 #endif 1476 1477 if (inp != NULL) 1478 INP_WUNLOCK(inp); 1479 1480 drop: 1481 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1482 if (s != NULL) 1483 free(s, M_TCPLOG); 1484 if (m != NULL) 1485 m_freem(m); 1486 return (IPPROTO_DONE); 1487 } 1488 1489 /* 1490 * Automatic sizing of receive socket buffer. Often the send 1491 * buffer size is not optimally adjusted to the actual network 1492 * conditions at hand (delay bandwidth product). Setting the 1493 * buffer size too small limits throughput on links with high 1494 * bandwidth and high delay (eg. trans-continental/oceanic links). 1495 * 1496 * On the receive side the socket buffer memory is only rarely 1497 * used to any significant extent. This allows us to be much 1498 * more aggressive in scaling the receive socket buffer. For 1499 * the case that the buffer space is actually used to a large 1500 * extent and we run out of kernel memory we can simply drop 1501 * the new segments; TCP on the sender will just retransmit it 1502 * later. Setting the buffer size too big may only consume too 1503 * much kernel memory if the application doesn't read() from 1504 * the socket or packet loss or reordering makes use of the 1505 * reassembly queue. 1506 * 1507 * The criteria to step up the receive buffer one notch are: 1508 * 1. Application has not set receive buffer size with 1509 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1510 * 2. the number of bytes received during the time it takes 1511 * one timestamp to be reflected back to us (the RTT); 1512 * 3. received bytes per RTT is within seven eighth of the 1513 * current socket buffer size; 1514 * 4. receive buffer size has not hit maximal automatic size; 1515 * 1516 * This algorithm does one step per RTT at most and only if 1517 * we receive a bulk stream w/o packet losses or reorderings. 1518 * Shrinking the buffer during idle times is not necessary as 1519 * it doesn't consume any memory when idle. 1520 * 1521 * TODO: Only step up if the application is actually serving 1522 * the buffer to better manage the socket buffer resources. 1523 */ 1524 int 1525 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1526 struct tcpcb *tp, int tlen) 1527 { 1528 int newsize = 0; 1529 1530 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1531 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1532 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1533 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1534 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1535 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1536 newsize = min(so->so_rcv.sb_hiwat + 1537 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1538 } 1539 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1540 1541 /* Start over with next RTT. */ 1542 tp->rfbuf_ts = 0; 1543 tp->rfbuf_cnt = 0; 1544 } else { 1545 tp->rfbuf_cnt += tlen; /* add up */ 1546 } 1547 1548 return (newsize); 1549 } 1550 1551 void 1552 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1553 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1554 int ti_locked) 1555 { 1556 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1557 int rstreason, todrop, win; 1558 uint32_t tiwin; 1559 uint16_t nsegs; 1560 char *s; 1561 struct in_conninfo *inc; 1562 struct mbuf *mfree; 1563 struct tcpopt to; 1564 #ifdef TCP_RFC7413 1565 int tfo_syn; 1566 #endif 1567 1568 #ifdef TCPDEBUG 1569 /* 1570 * The size of tcp_saveipgen must be the size of the max ip header, 1571 * now IPv6. 1572 */ 1573 u_char tcp_saveipgen[IP6_HDR_LEN]; 1574 struct tcphdr tcp_savetcp; 1575 short ostate = 0; 1576 #endif 1577 thflags = th->th_flags; 1578 inc = &tp->t_inpcb->inp_inc; 1579 tp->sackhint.last_sack_ack = 0; 1580 sack_changed = 0; 1581 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1582 1583 /* 1584 * If this is either a state-changing packet or current state isn't 1585 * established, we require a write lock on tcbinfo. Otherwise, we 1586 * allow the tcbinfo to be in either alocked or unlocked, as the 1587 * caller may have unnecessarily acquired a write lock due to a race. 1588 */ 1589 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1590 tp->t_state != TCPS_ESTABLISHED) { 1591 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1592 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1593 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1594 } else { 1595 #ifdef INVARIANTS 1596 if (ti_locked == TI_RLOCKED) 1597 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1598 else { 1599 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1600 "ti_locked: %d", __func__, ti_locked)); 1601 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1602 } 1603 #endif 1604 } 1605 INP_WLOCK_ASSERT(tp->t_inpcb); 1606 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1607 __func__)); 1608 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1609 __func__)); 1610 1611 #ifdef TCPPCAP 1612 /* Save segment, if requested. */ 1613 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1614 #endif 1615 1616 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1617 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1618 log(LOG_DEBUG, "%s; %s: " 1619 "SYN|FIN segment ignored (based on " 1620 "sysctl setting)\n", s, __func__); 1621 free(s, M_TCPLOG); 1622 } 1623 goto drop; 1624 } 1625 1626 /* 1627 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1628 * check SEQ.ACK first. 1629 */ 1630 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1631 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1632 rstreason = BANDLIM_UNLIMITED; 1633 goto dropwithreset; 1634 } 1635 1636 /* 1637 * Segment received on connection. 1638 * Reset idle time and keep-alive timer. 1639 * XXX: This should be done after segment 1640 * validation to ignore broken/spoofed segs. 1641 */ 1642 tp->t_rcvtime = ticks; 1643 1644 /* 1645 * Scale up the window into a 32-bit value. 1646 * For the SYN_SENT state the scale is zero. 1647 */ 1648 tiwin = th->th_win << tp->snd_scale; 1649 1650 /* 1651 * TCP ECN processing. 1652 */ 1653 if (tp->t_flags & TF_ECN_PERMIT) { 1654 if (thflags & TH_CWR) 1655 tp->t_flags &= ~TF_ECN_SND_ECE; 1656 switch (iptos & IPTOS_ECN_MASK) { 1657 case IPTOS_ECN_CE: 1658 tp->t_flags |= TF_ECN_SND_ECE; 1659 TCPSTAT_INC(tcps_ecn_ce); 1660 break; 1661 case IPTOS_ECN_ECT0: 1662 TCPSTAT_INC(tcps_ecn_ect0); 1663 break; 1664 case IPTOS_ECN_ECT1: 1665 TCPSTAT_INC(tcps_ecn_ect1); 1666 break; 1667 } 1668 1669 /* Process a packet differently from RFC3168. */ 1670 cc_ecnpkt_handler(tp, th, iptos); 1671 1672 /* Congestion experienced. */ 1673 if (thflags & TH_ECE) { 1674 cc_cong_signal(tp, th, CC_ECN); 1675 } 1676 } 1677 1678 /* 1679 * Parse options on any incoming segment. 1680 */ 1681 tcp_dooptions(&to, (u_char *)(th + 1), 1682 (th->th_off << 2) - sizeof(struct tcphdr), 1683 (thflags & TH_SYN) ? TO_SYN : 0); 1684 1685 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1686 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1687 (to.to_flags & TOF_SIGNATURE) == 0) { 1688 TCPSTAT_INC(tcps_sig_err_sigopt); 1689 /* XXX: should drop? */ 1690 } 1691 #endif 1692 /* 1693 * If echoed timestamp is later than the current time, 1694 * fall back to non RFC1323 RTT calculation. Normalize 1695 * timestamp if syncookies were used when this connection 1696 * was established. 1697 */ 1698 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1699 to.to_tsecr -= tp->ts_offset; 1700 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1701 to.to_tsecr = 0; 1702 } 1703 /* 1704 * If timestamps were negotiated during SYN/ACK they should 1705 * appear on every segment during this session and vice versa. 1706 */ 1707 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1708 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1709 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1710 "no action\n", s, __func__); 1711 free(s, M_TCPLOG); 1712 } 1713 } 1714 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1715 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1716 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1717 "no action\n", s, __func__); 1718 free(s, M_TCPLOG); 1719 } 1720 } 1721 1722 /* 1723 * Process options only when we get SYN/ACK back. The SYN case 1724 * for incoming connections is handled in tcp_syncache. 1725 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1726 * or <SYN,ACK>) segment itself is never scaled. 1727 * XXX this is traditional behavior, may need to be cleaned up. 1728 */ 1729 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1730 if ((to.to_flags & TOF_SCALE) && 1731 (tp->t_flags & TF_REQ_SCALE)) { 1732 tp->t_flags |= TF_RCVD_SCALE; 1733 tp->snd_scale = to.to_wscale; 1734 } 1735 /* 1736 * Initial send window. It will be updated with 1737 * the next incoming segment to the scaled value. 1738 */ 1739 tp->snd_wnd = th->th_win; 1740 if (to.to_flags & TOF_TS) { 1741 tp->t_flags |= TF_RCVD_TSTMP; 1742 tp->ts_recent = to.to_tsval; 1743 tp->ts_recent_age = tcp_ts_getticks(); 1744 } 1745 if (to.to_flags & TOF_MSS) 1746 tcp_mss(tp, to.to_mss); 1747 if ((tp->t_flags & TF_SACK_PERMIT) && 1748 (to.to_flags & TOF_SACKPERM) == 0) 1749 tp->t_flags &= ~TF_SACK_PERMIT; 1750 } 1751 1752 /* 1753 * Header prediction: check for the two common cases 1754 * of a uni-directional data xfer. If the packet has 1755 * no control flags, is in-sequence, the window didn't 1756 * change and we're not retransmitting, it's a 1757 * candidate. If the length is zero and the ack moved 1758 * forward, we're the sender side of the xfer. Just 1759 * free the data acked & wake any higher level process 1760 * that was blocked waiting for space. If the length 1761 * is non-zero and the ack didn't move, we're the 1762 * receiver side. If we're getting packets in-order 1763 * (the reassembly queue is empty), add the data to 1764 * the socket buffer and note that we need a delayed ack. 1765 * Make sure that the hidden state-flags are also off. 1766 * Since we check for TCPS_ESTABLISHED first, it can only 1767 * be TH_NEEDSYN. 1768 */ 1769 if (tp->t_state == TCPS_ESTABLISHED && 1770 th->th_seq == tp->rcv_nxt && 1771 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1772 tp->snd_nxt == tp->snd_max && 1773 tiwin && tiwin == tp->snd_wnd && 1774 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1775 LIST_EMPTY(&tp->t_segq) && 1776 ((to.to_flags & TOF_TS) == 0 || 1777 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1778 1779 /* 1780 * If last ACK falls within this segment's sequence numbers, 1781 * record the timestamp. 1782 * NOTE that the test is modified according to the latest 1783 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1784 */ 1785 if ((to.to_flags & TOF_TS) != 0 && 1786 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1787 tp->ts_recent_age = tcp_ts_getticks(); 1788 tp->ts_recent = to.to_tsval; 1789 } 1790 1791 if (tlen == 0) { 1792 if (SEQ_GT(th->th_ack, tp->snd_una) && 1793 SEQ_LEQ(th->th_ack, tp->snd_max) && 1794 !IN_RECOVERY(tp->t_flags) && 1795 (to.to_flags & TOF_SACK) == 0 && 1796 TAILQ_EMPTY(&tp->snd_holes)) { 1797 /* 1798 * This is a pure ack for outstanding data. 1799 */ 1800 if (ti_locked == TI_RLOCKED) 1801 INP_INFO_RUNLOCK(&V_tcbinfo); 1802 ti_locked = TI_UNLOCKED; 1803 1804 TCPSTAT_INC(tcps_predack); 1805 1806 /* 1807 * "bad retransmit" recovery. 1808 */ 1809 if (tp->t_rxtshift == 1 && 1810 tp->t_flags & TF_PREVVALID && 1811 (int)(ticks - tp->t_badrxtwin) < 0) { 1812 cc_cong_signal(tp, th, CC_RTO_ERR); 1813 } 1814 1815 /* 1816 * Recalculate the transmit timer / rtt. 1817 * 1818 * Some boxes send broken timestamp replies 1819 * during the SYN+ACK phase, ignore 1820 * timestamps of 0 or we could calculate a 1821 * huge RTT and blow up the retransmit timer. 1822 */ 1823 if ((to.to_flags & TOF_TS) != 0 && 1824 to.to_tsecr) { 1825 uint32_t t; 1826 1827 t = tcp_ts_getticks() - to.to_tsecr; 1828 if (!tp->t_rttlow || tp->t_rttlow > t) 1829 tp->t_rttlow = t; 1830 tcp_xmit_timer(tp, 1831 TCP_TS_TO_TICKS(t) + 1); 1832 } else if (tp->t_rtttime && 1833 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1834 if (!tp->t_rttlow || 1835 tp->t_rttlow > ticks - tp->t_rtttime) 1836 tp->t_rttlow = ticks - tp->t_rtttime; 1837 tcp_xmit_timer(tp, 1838 ticks - tp->t_rtttime); 1839 } 1840 acked = BYTES_THIS_ACK(tp, th); 1841 1842 #ifdef TCP_HHOOK 1843 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1844 hhook_run_tcp_est_in(tp, th, &to); 1845 #endif 1846 1847 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1848 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1849 sbdrop(&so->so_snd, acked); 1850 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1851 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1852 tp->snd_recover = th->th_ack - 1; 1853 1854 /* 1855 * Let the congestion control algorithm update 1856 * congestion control related information. This 1857 * typically means increasing the congestion 1858 * window. 1859 */ 1860 cc_ack_received(tp, th, nsegs, CC_ACK); 1861 1862 tp->snd_una = th->th_ack; 1863 /* 1864 * Pull snd_wl2 up to prevent seq wrap relative 1865 * to th_ack. 1866 */ 1867 tp->snd_wl2 = th->th_ack; 1868 tp->t_dupacks = 0; 1869 m_freem(m); 1870 1871 /* 1872 * If all outstanding data are acked, stop 1873 * retransmit timer, otherwise restart timer 1874 * using current (possibly backed-off) value. 1875 * If process is waiting for space, 1876 * wakeup/selwakeup/signal. If data 1877 * are ready to send, let tcp_output 1878 * decide between more output or persist. 1879 */ 1880 #ifdef TCPDEBUG 1881 if (so->so_options & SO_DEBUG) 1882 tcp_trace(TA_INPUT, ostate, tp, 1883 (void *)tcp_saveipgen, 1884 &tcp_savetcp, 0); 1885 #endif 1886 TCP_PROBE3(debug__input, tp, th, m); 1887 if (tp->snd_una == tp->snd_max) 1888 tcp_timer_activate(tp, TT_REXMT, 0); 1889 else if (!tcp_timer_active(tp, TT_PERSIST)) 1890 tcp_timer_activate(tp, TT_REXMT, 1891 tp->t_rxtcur); 1892 sowwakeup(so); 1893 if (sbavail(&so->so_snd)) 1894 (void) tp->t_fb->tfb_tcp_output(tp); 1895 goto check_delack; 1896 } 1897 } else if (th->th_ack == tp->snd_una && 1898 tlen <= sbspace(&so->so_rcv)) { 1899 int newsize = 0; /* automatic sockbuf scaling */ 1900 1901 /* 1902 * This is a pure, in-sequence data packet with 1903 * nothing on the reassembly queue and we have enough 1904 * buffer space to take it. 1905 */ 1906 if (ti_locked == TI_RLOCKED) 1907 INP_INFO_RUNLOCK(&V_tcbinfo); 1908 ti_locked = TI_UNLOCKED; 1909 1910 /* Clean receiver SACK report if present */ 1911 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1912 tcp_clean_sackreport(tp); 1913 TCPSTAT_INC(tcps_preddat); 1914 tp->rcv_nxt += tlen; 1915 /* 1916 * Pull snd_wl1 up to prevent seq wrap relative to 1917 * th_seq. 1918 */ 1919 tp->snd_wl1 = th->th_seq; 1920 /* 1921 * Pull rcv_up up to prevent seq wrap relative to 1922 * rcv_nxt. 1923 */ 1924 tp->rcv_up = tp->rcv_nxt; 1925 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1926 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1927 #ifdef TCPDEBUG 1928 if (so->so_options & SO_DEBUG) 1929 tcp_trace(TA_INPUT, ostate, tp, 1930 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1931 #endif 1932 TCP_PROBE3(debug__input, tp, th, m); 1933 1934 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1935 1936 /* Add data to socket buffer. */ 1937 SOCKBUF_LOCK(&so->so_rcv); 1938 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1939 m_freem(m); 1940 } else { 1941 /* 1942 * Set new socket buffer size. 1943 * Give up when limit is reached. 1944 */ 1945 if (newsize) 1946 if (!sbreserve_locked(&so->so_rcv, 1947 newsize, so, NULL)) 1948 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1949 m_adj(m, drop_hdrlen); /* delayed header drop */ 1950 sbappendstream_locked(&so->so_rcv, m, 0); 1951 } 1952 /* NB: sorwakeup_locked() does an implicit unlock. */ 1953 sorwakeup_locked(so); 1954 if (DELAY_ACK(tp, tlen)) { 1955 tp->t_flags |= TF_DELACK; 1956 } else { 1957 tp->t_flags |= TF_ACKNOW; 1958 tp->t_fb->tfb_tcp_output(tp); 1959 } 1960 goto check_delack; 1961 } 1962 } 1963 1964 /* 1965 * Calculate amount of space in receive window, 1966 * and then do TCP input processing. 1967 * Receive window is amount of space in rcv queue, 1968 * but not less than advertised window. 1969 */ 1970 win = sbspace(&so->so_rcv); 1971 if (win < 0) 1972 win = 0; 1973 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1974 1975 switch (tp->t_state) { 1976 1977 /* 1978 * If the state is SYN_RECEIVED: 1979 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1980 */ 1981 case TCPS_SYN_RECEIVED: 1982 if ((thflags & TH_ACK) && 1983 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1984 SEQ_GT(th->th_ack, tp->snd_max))) { 1985 rstreason = BANDLIM_RST_OPENPORT; 1986 goto dropwithreset; 1987 } 1988 #ifdef TCP_RFC7413 1989 if (IS_FASTOPEN(tp->t_flags)) { 1990 /* 1991 * When a TFO connection is in SYN_RECEIVED, the 1992 * only valid packets are the initial SYN, a 1993 * retransmit/copy of the initial SYN (possibly with 1994 * a subset of the original data), a valid ACK, a 1995 * FIN, or a RST. 1996 */ 1997 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1998 rstreason = BANDLIM_RST_OPENPORT; 1999 goto dropwithreset; 2000 } else if (thflags & TH_SYN) { 2001 /* non-initial SYN is ignored */ 2002 if ((tcp_timer_active(tp, TT_DELACK) || 2003 tcp_timer_active(tp, TT_REXMT))) 2004 goto drop; 2005 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 2006 goto drop; 2007 } 2008 } 2009 #endif 2010 break; 2011 2012 /* 2013 * If the state is SYN_SENT: 2014 * if seg contains a RST, then drop the connection. 2015 * if seg does not contain SYN, then drop it. 2016 * Otherwise this is an acceptable SYN segment 2017 * initialize tp->rcv_nxt and tp->irs 2018 * if seg contains ack then advance tp->snd_una 2019 * if seg contains an ECE and ECN support is enabled, the stream 2020 * is ECN capable. 2021 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2022 * arrange for segment to be acked (eventually) 2023 * continue processing rest of data/controls, beginning with URG 2024 */ 2025 case TCPS_SYN_SENT: 2026 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2027 TCP_PROBE5(connect__refused, NULL, tp, 2028 m, tp, th); 2029 tp = tcp_drop(tp, ECONNREFUSED); 2030 } 2031 if (thflags & TH_RST) 2032 goto drop; 2033 if (!(thflags & TH_SYN)) 2034 goto drop; 2035 2036 tp->irs = th->th_seq; 2037 tcp_rcvseqinit(tp); 2038 if (thflags & TH_ACK) { 2039 TCPSTAT_INC(tcps_connects); 2040 soisconnected(so); 2041 #ifdef MAC 2042 mac_socketpeer_set_from_mbuf(m, so); 2043 #endif 2044 /* Do window scaling on this connection? */ 2045 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2046 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2047 tp->rcv_scale = tp->request_r_scale; 2048 } 2049 tp->rcv_adv += min(tp->rcv_wnd, 2050 TCP_MAXWIN << tp->rcv_scale); 2051 tp->snd_una++; /* SYN is acked */ 2052 /* 2053 * If there's data, delay ACK; if there's also a FIN 2054 * ACKNOW will be turned on later. 2055 */ 2056 if (DELAY_ACK(tp, tlen) && tlen != 0) 2057 tcp_timer_activate(tp, TT_DELACK, 2058 tcp_delacktime); 2059 else 2060 tp->t_flags |= TF_ACKNOW; 2061 2062 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2063 tp->t_flags |= TF_ECN_PERMIT; 2064 TCPSTAT_INC(tcps_ecn_shs); 2065 } 2066 2067 /* 2068 * Received <SYN,ACK> in SYN_SENT[*] state. 2069 * Transitions: 2070 * SYN_SENT --> ESTABLISHED 2071 * SYN_SENT* --> FIN_WAIT_1 2072 */ 2073 tp->t_starttime = ticks; 2074 if (tp->t_flags & TF_NEEDFIN) { 2075 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2076 tp->t_flags &= ~TF_NEEDFIN; 2077 thflags &= ~TH_SYN; 2078 } else { 2079 tcp_state_change(tp, TCPS_ESTABLISHED); 2080 TCP_PROBE5(connect__established, NULL, tp, 2081 m, tp, th); 2082 cc_conn_init(tp); 2083 tcp_timer_activate(tp, TT_KEEP, 2084 TP_KEEPIDLE(tp)); 2085 } 2086 } else { 2087 /* 2088 * Received initial SYN in SYN-SENT[*] state => 2089 * simultaneous open. 2090 * If it succeeds, connection is * half-synchronized. 2091 * Otherwise, do 3-way handshake: 2092 * SYN-SENT -> SYN-RECEIVED 2093 * SYN-SENT* -> SYN-RECEIVED* 2094 */ 2095 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2096 tcp_timer_activate(tp, TT_REXMT, 0); 2097 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2098 } 2099 2100 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2101 "ti_locked %d", __func__, ti_locked)); 2102 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2103 INP_WLOCK_ASSERT(tp->t_inpcb); 2104 2105 /* 2106 * Advance th->th_seq to correspond to first data byte. 2107 * If data, trim to stay within window, 2108 * dropping FIN if necessary. 2109 */ 2110 th->th_seq++; 2111 if (tlen > tp->rcv_wnd) { 2112 todrop = tlen - tp->rcv_wnd; 2113 m_adj(m, -todrop); 2114 tlen = tp->rcv_wnd; 2115 thflags &= ~TH_FIN; 2116 TCPSTAT_INC(tcps_rcvpackafterwin); 2117 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2118 } 2119 tp->snd_wl1 = th->th_seq - 1; 2120 tp->rcv_up = th->th_seq; 2121 /* 2122 * Client side of transaction: already sent SYN and data. 2123 * If the remote host used T/TCP to validate the SYN, 2124 * our data will be ACK'd; if so, enter normal data segment 2125 * processing in the middle of step 5, ack processing. 2126 * Otherwise, goto step 6. 2127 */ 2128 if (thflags & TH_ACK) 2129 goto process_ACK; 2130 2131 goto step6; 2132 2133 /* 2134 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2135 * do normal processing. 2136 * 2137 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2138 */ 2139 case TCPS_LAST_ACK: 2140 case TCPS_CLOSING: 2141 break; /* continue normal processing */ 2142 } 2143 2144 /* 2145 * States other than LISTEN or SYN_SENT. 2146 * First check the RST flag and sequence number since reset segments 2147 * are exempt from the timestamp and connection count tests. This 2148 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2149 * below which allowed reset segments in half the sequence space 2150 * to fall though and be processed (which gives forged reset 2151 * segments with a random sequence number a 50 percent chance of 2152 * killing a connection). 2153 * Then check timestamp, if present. 2154 * Then check the connection count, if present. 2155 * Then check that at least some bytes of segment are within 2156 * receive window. If segment begins before rcv_nxt, 2157 * drop leading data (and SYN); if nothing left, just ack. 2158 */ 2159 if (thflags & TH_RST) { 2160 /* 2161 * RFC5961 Section 3.2 2162 * 2163 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2164 * - If RST is in window, we send challenge ACK. 2165 * 2166 * Note: to take into account delayed ACKs, we should 2167 * test against last_ack_sent instead of rcv_nxt. 2168 * Note 2: we handle special case of closed window, not 2169 * covered by the RFC. 2170 */ 2171 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2172 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2173 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2174 2175 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2176 KASSERT(ti_locked == TI_RLOCKED, 2177 ("%s: TH_RST ti_locked %d, th %p tp %p", 2178 __func__, ti_locked, th, tp)); 2179 KASSERT(tp->t_state != TCPS_SYN_SENT, 2180 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2181 __func__, th, tp)); 2182 2183 if (V_tcp_insecure_rst || 2184 tp->last_ack_sent == th->th_seq) { 2185 TCPSTAT_INC(tcps_drops); 2186 /* Drop the connection. */ 2187 switch (tp->t_state) { 2188 case TCPS_SYN_RECEIVED: 2189 so->so_error = ECONNREFUSED; 2190 goto close; 2191 case TCPS_ESTABLISHED: 2192 case TCPS_FIN_WAIT_1: 2193 case TCPS_FIN_WAIT_2: 2194 case TCPS_CLOSE_WAIT: 2195 case TCPS_CLOSING: 2196 case TCPS_LAST_ACK: 2197 so->so_error = ECONNRESET; 2198 close: 2199 /* FALLTHROUGH */ 2200 default: 2201 tp = tcp_close(tp); 2202 } 2203 } else { 2204 TCPSTAT_INC(tcps_badrst); 2205 /* Send challenge ACK. */ 2206 tcp_respond(tp, mtod(m, void *), th, m, 2207 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2208 tp->last_ack_sent = tp->rcv_nxt; 2209 m = NULL; 2210 } 2211 } 2212 goto drop; 2213 } 2214 2215 /* 2216 * RFC5961 Section 4.2 2217 * Send challenge ACK for any SYN in synchronized state. 2218 */ 2219 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2220 tp->t_state != TCPS_SYN_RECEIVED) { 2221 KASSERT(ti_locked == TI_RLOCKED, 2222 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2223 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2224 2225 TCPSTAT_INC(tcps_badsyn); 2226 if (V_tcp_insecure_syn && 2227 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2228 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2229 tp = tcp_drop(tp, ECONNRESET); 2230 rstreason = BANDLIM_UNLIMITED; 2231 } else { 2232 /* Send challenge ACK. */ 2233 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2234 tp->snd_nxt, TH_ACK); 2235 tp->last_ack_sent = tp->rcv_nxt; 2236 m = NULL; 2237 } 2238 goto drop; 2239 } 2240 2241 /* 2242 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2243 * and it's less than ts_recent, drop it. 2244 */ 2245 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2246 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2247 2248 /* Check to see if ts_recent is over 24 days old. */ 2249 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2250 /* 2251 * Invalidate ts_recent. If this segment updates 2252 * ts_recent, the age will be reset later and ts_recent 2253 * will get a valid value. If it does not, setting 2254 * ts_recent to zero will at least satisfy the 2255 * requirement that zero be placed in the timestamp 2256 * echo reply when ts_recent isn't valid. The 2257 * age isn't reset until we get a valid ts_recent 2258 * because we don't want out-of-order segments to be 2259 * dropped when ts_recent is old. 2260 */ 2261 tp->ts_recent = 0; 2262 } else { 2263 TCPSTAT_INC(tcps_rcvduppack); 2264 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2265 TCPSTAT_INC(tcps_pawsdrop); 2266 if (tlen) 2267 goto dropafterack; 2268 goto drop; 2269 } 2270 } 2271 2272 /* 2273 * In the SYN-RECEIVED state, validate that the packet belongs to 2274 * this connection before trimming the data to fit the receive 2275 * window. Check the sequence number versus IRS since we know 2276 * the sequence numbers haven't wrapped. This is a partial fix 2277 * for the "LAND" DoS attack. 2278 */ 2279 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2280 rstreason = BANDLIM_RST_OPENPORT; 2281 goto dropwithreset; 2282 } 2283 2284 todrop = tp->rcv_nxt - th->th_seq; 2285 if (todrop > 0) { 2286 if (thflags & TH_SYN) { 2287 thflags &= ~TH_SYN; 2288 th->th_seq++; 2289 if (th->th_urp > 1) 2290 th->th_urp--; 2291 else 2292 thflags &= ~TH_URG; 2293 todrop--; 2294 } 2295 /* 2296 * Following if statement from Stevens, vol. 2, p. 960. 2297 */ 2298 if (todrop > tlen 2299 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2300 /* 2301 * Any valid FIN must be to the left of the window. 2302 * At this point the FIN must be a duplicate or out 2303 * of sequence; drop it. 2304 */ 2305 thflags &= ~TH_FIN; 2306 2307 /* 2308 * Send an ACK to resynchronize and drop any data. 2309 * But keep on processing for RST or ACK. 2310 */ 2311 tp->t_flags |= TF_ACKNOW; 2312 todrop = tlen; 2313 TCPSTAT_INC(tcps_rcvduppack); 2314 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2315 } else { 2316 TCPSTAT_INC(tcps_rcvpartduppack); 2317 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2318 } 2319 drop_hdrlen += todrop; /* drop from the top afterwards */ 2320 th->th_seq += todrop; 2321 tlen -= todrop; 2322 if (th->th_urp > todrop) 2323 th->th_urp -= todrop; 2324 else { 2325 thflags &= ~TH_URG; 2326 th->th_urp = 0; 2327 } 2328 } 2329 2330 /* 2331 * If new data are received on a connection after the 2332 * user processes are gone, then RST the other end. 2333 */ 2334 if ((so->so_state & SS_NOFDREF) && 2335 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2336 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2337 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2338 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2339 2340 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2341 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2342 "after socket was closed, " 2343 "sending RST and removing tcpcb\n", 2344 s, __func__, tcpstates[tp->t_state], tlen); 2345 free(s, M_TCPLOG); 2346 } 2347 tp = tcp_close(tp); 2348 TCPSTAT_INC(tcps_rcvafterclose); 2349 rstreason = BANDLIM_UNLIMITED; 2350 goto dropwithreset; 2351 } 2352 2353 /* 2354 * If segment ends after window, drop trailing data 2355 * (and PUSH and FIN); if nothing left, just ACK. 2356 */ 2357 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2358 if (todrop > 0) { 2359 TCPSTAT_INC(tcps_rcvpackafterwin); 2360 if (todrop >= tlen) { 2361 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2362 /* 2363 * If window is closed can only take segments at 2364 * window edge, and have to drop data and PUSH from 2365 * incoming segments. Continue processing, but 2366 * remember to ack. Otherwise, drop segment 2367 * and ack. 2368 */ 2369 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2370 tp->t_flags |= TF_ACKNOW; 2371 TCPSTAT_INC(tcps_rcvwinprobe); 2372 } else 2373 goto dropafterack; 2374 } else 2375 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2376 m_adj(m, -todrop); 2377 tlen -= todrop; 2378 thflags &= ~(TH_PUSH|TH_FIN); 2379 } 2380 2381 /* 2382 * If last ACK falls within this segment's sequence numbers, 2383 * record its timestamp. 2384 * NOTE: 2385 * 1) That the test incorporates suggestions from the latest 2386 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2387 * 2) That updating only on newer timestamps interferes with 2388 * our earlier PAWS tests, so this check should be solely 2389 * predicated on the sequence space of this segment. 2390 * 3) That we modify the segment boundary check to be 2391 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2392 * instead of RFC1323's 2393 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2394 * This modified check allows us to overcome RFC1323's 2395 * limitations as described in Stevens TCP/IP Illustrated 2396 * Vol. 2 p.869. In such cases, we can still calculate the 2397 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2398 */ 2399 if ((to.to_flags & TOF_TS) != 0 && 2400 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2401 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2402 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2403 tp->ts_recent_age = tcp_ts_getticks(); 2404 tp->ts_recent = to.to_tsval; 2405 } 2406 2407 /* 2408 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2409 * flag is on (half-synchronized state), then queue data for 2410 * later processing; else drop segment and return. 2411 */ 2412 if ((thflags & TH_ACK) == 0) { 2413 if (tp->t_state == TCPS_SYN_RECEIVED || 2414 (tp->t_flags & TF_NEEDSYN)) { 2415 #ifdef TCP_RFC7413 2416 if (tp->t_state == TCPS_SYN_RECEIVED && 2417 IS_FASTOPEN(tp->t_flags)) { 2418 tp->snd_wnd = tiwin; 2419 cc_conn_init(tp); 2420 } 2421 #endif 2422 goto step6; 2423 } else if (tp->t_flags & TF_ACKNOW) 2424 goto dropafterack; 2425 else 2426 goto drop; 2427 } 2428 2429 /* 2430 * Ack processing. 2431 */ 2432 switch (tp->t_state) { 2433 2434 /* 2435 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2436 * ESTABLISHED state and continue processing. 2437 * The ACK was checked above. 2438 */ 2439 case TCPS_SYN_RECEIVED: 2440 2441 TCPSTAT_INC(tcps_connects); 2442 soisconnected(so); 2443 /* Do window scaling? */ 2444 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2445 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2446 tp->rcv_scale = tp->request_r_scale; 2447 tp->snd_wnd = tiwin; 2448 } 2449 /* 2450 * Make transitions: 2451 * SYN-RECEIVED -> ESTABLISHED 2452 * SYN-RECEIVED* -> FIN-WAIT-1 2453 */ 2454 tp->t_starttime = ticks; 2455 if (tp->t_flags & TF_NEEDFIN) { 2456 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2457 tp->t_flags &= ~TF_NEEDFIN; 2458 } else { 2459 tcp_state_change(tp, TCPS_ESTABLISHED); 2460 TCP_PROBE5(accept__established, NULL, tp, 2461 m, tp, th); 2462 #ifdef TCP_RFC7413 2463 if (tp->t_tfo_pending) { 2464 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2465 tp->t_tfo_pending = NULL; 2466 2467 /* 2468 * Account for the ACK of our SYN prior to 2469 * regular ACK processing below. 2470 */ 2471 tp->snd_una++; 2472 } 2473 /* 2474 * TFO connections call cc_conn_init() during SYN 2475 * processing. Calling it again here for such 2476 * connections is not harmless as it would undo the 2477 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2478 * is retransmitted. 2479 */ 2480 if (!IS_FASTOPEN(tp->t_flags)) 2481 #endif 2482 cc_conn_init(tp); 2483 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2484 } 2485 /* 2486 * If segment contains data or ACK, will call tcp_reass() 2487 * later; if not, do so now to pass queued data to user. 2488 */ 2489 if (tlen == 0 && (thflags & TH_FIN) == 0) 2490 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2491 (struct mbuf *)0); 2492 tp->snd_wl1 = th->th_seq - 1; 2493 /* FALLTHROUGH */ 2494 2495 /* 2496 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2497 * ACKs. If the ack is in the range 2498 * tp->snd_una < th->th_ack <= tp->snd_max 2499 * then advance tp->snd_una to th->th_ack and drop 2500 * data from the retransmission queue. If this ACK reflects 2501 * more up to date window information we update our window information. 2502 */ 2503 case TCPS_ESTABLISHED: 2504 case TCPS_FIN_WAIT_1: 2505 case TCPS_FIN_WAIT_2: 2506 case TCPS_CLOSE_WAIT: 2507 case TCPS_CLOSING: 2508 case TCPS_LAST_ACK: 2509 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2510 TCPSTAT_INC(tcps_rcvacktoomuch); 2511 goto dropafterack; 2512 } 2513 if ((tp->t_flags & TF_SACK_PERMIT) && 2514 ((to.to_flags & TOF_SACK) || 2515 !TAILQ_EMPTY(&tp->snd_holes))) 2516 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2517 else 2518 /* 2519 * Reset the value so that previous (valid) value 2520 * from the last ack with SACK doesn't get used. 2521 */ 2522 tp->sackhint.sacked_bytes = 0; 2523 2524 #ifdef TCP_HHOOK 2525 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2526 hhook_run_tcp_est_in(tp, th, &to); 2527 #endif 2528 2529 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2530 u_int maxseg; 2531 2532 maxseg = tcp_maxseg(tp); 2533 if (tlen == 0 && 2534 (tiwin == tp->snd_wnd || 2535 (tp->t_flags & TF_SACK_PERMIT))) { 2536 /* 2537 * If this is the first time we've seen a 2538 * FIN from the remote, this is not a 2539 * duplicate and it needs to be processed 2540 * normally. This happens during a 2541 * simultaneous close. 2542 */ 2543 if ((thflags & TH_FIN) && 2544 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2545 tp->t_dupacks = 0; 2546 break; 2547 } 2548 TCPSTAT_INC(tcps_rcvdupack); 2549 /* 2550 * If we have outstanding data (other than 2551 * a window probe), this is a completely 2552 * duplicate ack (ie, window info didn't 2553 * change and FIN isn't set), 2554 * the ack is the biggest we've 2555 * seen and we've seen exactly our rexmt 2556 * threshold of them, assume a packet 2557 * has been dropped and retransmit it. 2558 * Kludge snd_nxt & the congestion 2559 * window so we send only this one 2560 * packet. 2561 * 2562 * We know we're losing at the current 2563 * window size so do congestion avoidance 2564 * (set ssthresh to half the current window 2565 * and pull our congestion window back to 2566 * the new ssthresh). 2567 * 2568 * Dup acks mean that packets have left the 2569 * network (they're now cached at the receiver) 2570 * so bump cwnd by the amount in the receiver 2571 * to keep a constant cwnd packets in the 2572 * network. 2573 * 2574 * When using TCP ECN, notify the peer that 2575 * we reduced the cwnd. 2576 */ 2577 /* 2578 * Following 2 kinds of acks should not affect 2579 * dupack counting: 2580 * 1) Old acks 2581 * 2) Acks with SACK but without any new SACK 2582 * information in them. These could result from 2583 * any anomaly in the network like a switch 2584 * duplicating packets or a possible DoS attack. 2585 */ 2586 if (th->th_ack != tp->snd_una || 2587 ((tp->t_flags & TF_SACK_PERMIT) && 2588 !sack_changed)) 2589 break; 2590 else if (!tcp_timer_active(tp, TT_REXMT)) 2591 tp->t_dupacks = 0; 2592 else if (++tp->t_dupacks > tcprexmtthresh || 2593 IN_FASTRECOVERY(tp->t_flags)) { 2594 cc_ack_received(tp, th, nsegs, 2595 CC_DUPACK); 2596 if ((tp->t_flags & TF_SACK_PERMIT) && 2597 IN_FASTRECOVERY(tp->t_flags)) { 2598 int awnd; 2599 2600 /* 2601 * Compute the amount of data in flight first. 2602 * We can inject new data into the pipe iff 2603 * we have less than 1/2 the original window's 2604 * worth of data in flight. 2605 */ 2606 if (V_tcp_do_rfc6675_pipe) 2607 awnd = tcp_compute_pipe(tp); 2608 else 2609 awnd = (tp->snd_nxt - tp->snd_fack) + 2610 tp->sackhint.sack_bytes_rexmit; 2611 2612 if (awnd < tp->snd_ssthresh) { 2613 tp->snd_cwnd += maxseg; 2614 /* 2615 * RFC5681 Section 3.2 talks about cwnd 2616 * inflation on additional dupacks and 2617 * deflation on recovering from loss. 2618 * 2619 * We keep cwnd into check so that 2620 * we don't have to 'deflate' it when we 2621 * get out of recovery. 2622 */ 2623 if (tp->snd_cwnd > tp->snd_ssthresh) 2624 tp->snd_cwnd = tp->snd_ssthresh; 2625 } 2626 } else 2627 tp->snd_cwnd += maxseg; 2628 (void) tp->t_fb->tfb_tcp_output(tp); 2629 goto drop; 2630 } else if (tp->t_dupacks == tcprexmtthresh) { 2631 tcp_seq onxt = tp->snd_nxt; 2632 2633 /* 2634 * If we're doing sack, check to 2635 * see if we're already in sack 2636 * recovery. If we're not doing sack, 2637 * check to see if we're in newreno 2638 * recovery. 2639 */ 2640 if (tp->t_flags & TF_SACK_PERMIT) { 2641 if (IN_FASTRECOVERY(tp->t_flags)) { 2642 tp->t_dupacks = 0; 2643 break; 2644 } 2645 } else { 2646 if (SEQ_LEQ(th->th_ack, 2647 tp->snd_recover)) { 2648 tp->t_dupacks = 0; 2649 break; 2650 } 2651 } 2652 /* Congestion signal before ack. */ 2653 cc_cong_signal(tp, th, CC_NDUPACK); 2654 cc_ack_received(tp, th, nsegs, 2655 CC_DUPACK); 2656 tcp_timer_activate(tp, TT_REXMT, 0); 2657 tp->t_rtttime = 0; 2658 if (tp->t_flags & TF_SACK_PERMIT) { 2659 TCPSTAT_INC( 2660 tcps_sack_recovery_episode); 2661 tp->sack_newdata = tp->snd_nxt; 2662 if (CC_ALGO(tp)->cong_signal == NULL) 2663 tp->snd_cwnd = maxseg; 2664 (void) tp->t_fb->tfb_tcp_output(tp); 2665 goto drop; 2666 } 2667 tp->snd_nxt = th->th_ack; 2668 if (CC_ALGO(tp)->cong_signal == NULL) 2669 tp->snd_cwnd = maxseg; 2670 (void) tp->t_fb->tfb_tcp_output(tp); 2671 KASSERT(tp->snd_limited <= 2, 2672 ("%s: tp->snd_limited too big", 2673 __func__)); 2674 if (CC_ALGO(tp)->cong_signal == NULL) 2675 tp->snd_cwnd = tp->snd_ssthresh + 2676 maxseg * 2677 (tp->t_dupacks - tp->snd_limited); 2678 if (SEQ_GT(onxt, tp->snd_nxt)) 2679 tp->snd_nxt = onxt; 2680 goto drop; 2681 } else if (V_tcp_do_rfc3042) { 2682 /* 2683 * Process first and second duplicate 2684 * ACKs. Each indicates a segment 2685 * leaving the network, creating room 2686 * for more. Make sure we can send a 2687 * packet on reception of each duplicate 2688 * ACK by increasing snd_cwnd by one 2689 * segment. Restore the original 2690 * snd_cwnd after packet transmission. 2691 */ 2692 cc_ack_received(tp, th, nsegs, 2693 CC_DUPACK); 2694 uint32_t oldcwnd = tp->snd_cwnd; 2695 tcp_seq oldsndmax = tp->snd_max; 2696 u_int sent; 2697 int avail; 2698 2699 KASSERT(tp->t_dupacks == 1 || 2700 tp->t_dupacks == 2, 2701 ("%s: dupacks not 1 or 2", 2702 __func__)); 2703 if (tp->t_dupacks == 1) 2704 tp->snd_limited = 0; 2705 tp->snd_cwnd = 2706 (tp->snd_nxt - tp->snd_una) + 2707 (tp->t_dupacks - tp->snd_limited) * 2708 maxseg; 2709 /* 2710 * Only call tcp_output when there 2711 * is new data available to be sent. 2712 * Otherwise we would send pure ACKs. 2713 */ 2714 SOCKBUF_LOCK(&so->so_snd); 2715 avail = sbavail(&so->so_snd) - 2716 (tp->snd_nxt - tp->snd_una); 2717 SOCKBUF_UNLOCK(&so->so_snd); 2718 if (avail > 0) 2719 (void) tp->t_fb->tfb_tcp_output(tp); 2720 sent = tp->snd_max - oldsndmax; 2721 if (sent > maxseg) { 2722 KASSERT((tp->t_dupacks == 2 && 2723 tp->snd_limited == 0) || 2724 (sent == maxseg + 1 && 2725 tp->t_flags & TF_SENTFIN), 2726 ("%s: sent too much", 2727 __func__)); 2728 tp->snd_limited = 2; 2729 } else if (sent > 0) 2730 ++tp->snd_limited; 2731 tp->snd_cwnd = oldcwnd; 2732 goto drop; 2733 } 2734 } 2735 break; 2736 } else { 2737 /* 2738 * This ack is advancing the left edge, reset the 2739 * counter. 2740 */ 2741 tp->t_dupacks = 0; 2742 /* 2743 * If this ack also has new SACK info, increment the 2744 * counter as per rfc6675. 2745 */ 2746 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2747 tp->t_dupacks++; 2748 } 2749 2750 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2751 ("%s: th_ack <= snd_una", __func__)); 2752 2753 /* 2754 * If the congestion window was inflated to account 2755 * for the other side's cached packets, retract it. 2756 */ 2757 if (IN_FASTRECOVERY(tp->t_flags)) { 2758 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2759 if (tp->t_flags & TF_SACK_PERMIT) 2760 tcp_sack_partialack(tp, th); 2761 else 2762 tcp_newreno_partial_ack(tp, th); 2763 } else 2764 cc_post_recovery(tp, th); 2765 } 2766 /* 2767 * If we reach this point, ACK is not a duplicate, 2768 * i.e., it ACKs something we sent. 2769 */ 2770 if (tp->t_flags & TF_NEEDSYN) { 2771 /* 2772 * T/TCP: Connection was half-synchronized, and our 2773 * SYN has been ACK'd (so connection is now fully 2774 * synchronized). Go to non-starred state, 2775 * increment snd_una for ACK of SYN, and check if 2776 * we can do window scaling. 2777 */ 2778 tp->t_flags &= ~TF_NEEDSYN; 2779 tp->snd_una++; 2780 /* Do window scaling? */ 2781 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2782 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2783 tp->rcv_scale = tp->request_r_scale; 2784 /* Send window already scaled. */ 2785 } 2786 } 2787 2788 process_ACK: 2789 INP_WLOCK_ASSERT(tp->t_inpcb); 2790 2791 acked = BYTES_THIS_ACK(tp, th); 2792 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2793 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2794 tp->snd_una, th->th_ack, tp, m)); 2795 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2796 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2797 2798 /* 2799 * If we just performed our first retransmit, and the ACK 2800 * arrives within our recovery window, then it was a mistake 2801 * to do the retransmit in the first place. Recover our 2802 * original cwnd and ssthresh, and proceed to transmit where 2803 * we left off. 2804 */ 2805 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2806 (int)(ticks - tp->t_badrxtwin) < 0) 2807 cc_cong_signal(tp, th, CC_RTO_ERR); 2808 2809 /* 2810 * If we have a timestamp reply, update smoothed 2811 * round trip time. If no timestamp is present but 2812 * transmit timer is running and timed sequence 2813 * number was acked, update smoothed round trip time. 2814 * Since we now have an rtt measurement, cancel the 2815 * timer backoff (cf., Phil Karn's retransmit alg.). 2816 * Recompute the initial retransmit timer. 2817 * 2818 * Some boxes send broken timestamp replies 2819 * during the SYN+ACK phase, ignore 2820 * timestamps of 0 or we could calculate a 2821 * huge RTT and blow up the retransmit timer. 2822 */ 2823 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2824 uint32_t t; 2825 2826 t = tcp_ts_getticks() - to.to_tsecr; 2827 if (!tp->t_rttlow || tp->t_rttlow > t) 2828 tp->t_rttlow = t; 2829 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2830 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2831 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2832 tp->t_rttlow = ticks - tp->t_rtttime; 2833 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2834 } 2835 2836 /* 2837 * If all outstanding data is acked, stop retransmit 2838 * timer and remember to restart (more output or persist). 2839 * If there is more data to be acked, restart retransmit 2840 * timer, using current (possibly backed-off) value. 2841 */ 2842 if (th->th_ack == tp->snd_max) { 2843 tcp_timer_activate(tp, TT_REXMT, 0); 2844 needoutput = 1; 2845 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2846 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2847 2848 /* 2849 * If no data (only SYN) was ACK'd, 2850 * skip rest of ACK processing. 2851 */ 2852 if (acked == 0) 2853 goto step6; 2854 2855 /* 2856 * Let the congestion control algorithm update congestion 2857 * control related information. This typically means increasing 2858 * the congestion window. 2859 */ 2860 cc_ack_received(tp, th, nsegs, CC_ACK); 2861 2862 SOCKBUF_LOCK(&so->so_snd); 2863 if (acked > sbavail(&so->so_snd)) { 2864 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2865 tp->snd_wnd -= sbavail(&so->so_snd); 2866 else 2867 tp->snd_wnd = 0; 2868 mfree = sbcut_locked(&so->so_snd, 2869 (int)sbavail(&so->so_snd)); 2870 ourfinisacked = 1; 2871 } else { 2872 mfree = sbcut_locked(&so->so_snd, acked); 2873 if (tp->snd_wnd >= (uint32_t) acked) 2874 tp->snd_wnd -= acked; 2875 else 2876 tp->snd_wnd = 0; 2877 ourfinisacked = 0; 2878 } 2879 /* NB: sowwakeup_locked() does an implicit unlock. */ 2880 sowwakeup_locked(so); 2881 m_freem(mfree); 2882 /* Detect una wraparound. */ 2883 if (!IN_RECOVERY(tp->t_flags) && 2884 SEQ_GT(tp->snd_una, tp->snd_recover) && 2885 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2886 tp->snd_recover = th->th_ack - 1; 2887 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2888 if (IN_RECOVERY(tp->t_flags) && 2889 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2890 EXIT_RECOVERY(tp->t_flags); 2891 } 2892 tp->snd_una = th->th_ack; 2893 if (tp->t_flags & TF_SACK_PERMIT) { 2894 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2895 tp->snd_recover = tp->snd_una; 2896 } 2897 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2898 tp->snd_nxt = tp->snd_una; 2899 2900 switch (tp->t_state) { 2901 2902 /* 2903 * In FIN_WAIT_1 STATE in addition to the processing 2904 * for the ESTABLISHED state if our FIN is now acknowledged 2905 * then enter FIN_WAIT_2. 2906 */ 2907 case TCPS_FIN_WAIT_1: 2908 if (ourfinisacked) { 2909 /* 2910 * If we can't receive any more 2911 * data, then closing user can proceed. 2912 * Starting the timer is contrary to the 2913 * specification, but if we don't get a FIN 2914 * we'll hang forever. 2915 * 2916 * XXXjl: 2917 * we should release the tp also, and use a 2918 * compressed state. 2919 */ 2920 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2921 soisdisconnected(so); 2922 tcp_timer_activate(tp, TT_2MSL, 2923 (tcp_fast_finwait2_recycle ? 2924 tcp_finwait2_timeout : 2925 TP_MAXIDLE(tp))); 2926 } 2927 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2928 } 2929 break; 2930 2931 /* 2932 * In CLOSING STATE in addition to the processing for 2933 * the ESTABLISHED state if the ACK acknowledges our FIN 2934 * then enter the TIME-WAIT state, otherwise ignore 2935 * the segment. 2936 */ 2937 case TCPS_CLOSING: 2938 if (ourfinisacked) { 2939 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2940 tcp_twstart(tp); 2941 INP_INFO_RUNLOCK(&V_tcbinfo); 2942 m_freem(m); 2943 return; 2944 } 2945 break; 2946 2947 /* 2948 * In LAST_ACK, we may still be waiting for data to drain 2949 * and/or to be acked, as well as for the ack of our FIN. 2950 * If our FIN is now acknowledged, delete the TCB, 2951 * enter the closed state and return. 2952 */ 2953 case TCPS_LAST_ACK: 2954 if (ourfinisacked) { 2955 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2956 tp = tcp_close(tp); 2957 goto drop; 2958 } 2959 break; 2960 } 2961 } 2962 2963 step6: 2964 INP_WLOCK_ASSERT(tp->t_inpcb); 2965 2966 /* 2967 * Update window information. 2968 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2969 */ 2970 if ((thflags & TH_ACK) && 2971 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2972 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2973 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2974 /* keep track of pure window updates */ 2975 if (tlen == 0 && 2976 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2977 TCPSTAT_INC(tcps_rcvwinupd); 2978 tp->snd_wnd = tiwin; 2979 tp->snd_wl1 = th->th_seq; 2980 tp->snd_wl2 = th->th_ack; 2981 if (tp->snd_wnd > tp->max_sndwnd) 2982 tp->max_sndwnd = tp->snd_wnd; 2983 needoutput = 1; 2984 } 2985 2986 /* 2987 * Process segments with URG. 2988 */ 2989 if ((thflags & TH_URG) && th->th_urp && 2990 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2991 /* 2992 * This is a kludge, but if we receive and accept 2993 * random urgent pointers, we'll crash in 2994 * soreceive. It's hard to imagine someone 2995 * actually wanting to send this much urgent data. 2996 */ 2997 SOCKBUF_LOCK(&so->so_rcv); 2998 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2999 th->th_urp = 0; /* XXX */ 3000 thflags &= ~TH_URG; /* XXX */ 3001 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3002 goto dodata; /* XXX */ 3003 } 3004 /* 3005 * If this segment advances the known urgent pointer, 3006 * then mark the data stream. This should not happen 3007 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3008 * a FIN has been received from the remote side. 3009 * In these states we ignore the URG. 3010 * 3011 * According to RFC961 (Assigned Protocols), 3012 * the urgent pointer points to the last octet 3013 * of urgent data. We continue, however, 3014 * to consider it to indicate the first octet 3015 * of data past the urgent section as the original 3016 * spec states (in one of two places). 3017 */ 3018 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3019 tp->rcv_up = th->th_seq + th->th_urp; 3020 so->so_oobmark = sbavail(&so->so_rcv) + 3021 (tp->rcv_up - tp->rcv_nxt) - 1; 3022 if (so->so_oobmark == 0) 3023 so->so_rcv.sb_state |= SBS_RCVATMARK; 3024 sohasoutofband(so); 3025 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3026 } 3027 SOCKBUF_UNLOCK(&so->so_rcv); 3028 /* 3029 * Remove out of band data so doesn't get presented to user. 3030 * This can happen independent of advancing the URG pointer, 3031 * but if two URG's are pending at once, some out-of-band 3032 * data may creep in... ick. 3033 */ 3034 if (th->th_urp <= (uint32_t)tlen && 3035 !(so->so_options & SO_OOBINLINE)) { 3036 /* hdr drop is delayed */ 3037 tcp_pulloutofband(so, th, m, drop_hdrlen); 3038 } 3039 } else { 3040 /* 3041 * If no out of band data is expected, 3042 * pull receive urgent pointer along 3043 * with the receive window. 3044 */ 3045 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3046 tp->rcv_up = tp->rcv_nxt; 3047 } 3048 dodata: /* XXX */ 3049 INP_WLOCK_ASSERT(tp->t_inpcb); 3050 3051 /* 3052 * Process the segment text, merging it into the TCP sequencing queue, 3053 * and arranging for acknowledgment of receipt if necessary. 3054 * This process logically involves adjusting tp->rcv_wnd as data 3055 * is presented to the user (this happens in tcp_usrreq.c, 3056 * case PRU_RCVD). If a FIN has already been received on this 3057 * connection then we just ignore the text. 3058 */ 3059 #ifdef TCP_RFC7413 3060 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3061 IS_FASTOPEN(tp->t_flags)); 3062 #else 3063 #define tfo_syn (false) 3064 #endif 3065 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3066 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3067 tcp_seq save_start = th->th_seq; 3068 m_adj(m, drop_hdrlen); /* delayed header drop */ 3069 /* 3070 * Insert segment which includes th into TCP reassembly queue 3071 * with control block tp. Set thflags to whether reassembly now 3072 * includes a segment with FIN. This handles the common case 3073 * inline (segment is the next to be received on an established 3074 * connection, and the queue is empty), avoiding linkage into 3075 * and removal from the queue and repetition of various 3076 * conversions. 3077 * Set DELACK for segments received in order, but ack 3078 * immediately when segments are out of order (so 3079 * fast retransmit can work). 3080 */ 3081 if (th->th_seq == tp->rcv_nxt && 3082 LIST_EMPTY(&tp->t_segq) && 3083 (TCPS_HAVEESTABLISHED(tp->t_state) || 3084 tfo_syn)) { 3085 if (DELAY_ACK(tp, tlen) || tfo_syn) 3086 tp->t_flags |= TF_DELACK; 3087 else 3088 tp->t_flags |= TF_ACKNOW; 3089 tp->rcv_nxt += tlen; 3090 thflags = th->th_flags & TH_FIN; 3091 TCPSTAT_INC(tcps_rcvpack); 3092 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3093 SOCKBUF_LOCK(&so->so_rcv); 3094 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3095 m_freem(m); 3096 else 3097 sbappendstream_locked(&so->so_rcv, m, 0); 3098 /* NB: sorwakeup_locked() does an implicit unlock. */ 3099 sorwakeup_locked(so); 3100 } else { 3101 /* 3102 * XXX: Due to the header drop above "th" is 3103 * theoretically invalid by now. Fortunately 3104 * m_adj() doesn't actually frees any mbufs 3105 * when trimming from the head. 3106 */ 3107 thflags = tcp_reass(tp, th, &tlen, m); 3108 tp->t_flags |= TF_ACKNOW; 3109 } 3110 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3111 tcp_update_sack_list(tp, save_start, save_start + tlen); 3112 #if 0 3113 /* 3114 * Note the amount of data that peer has sent into 3115 * our window, in order to estimate the sender's 3116 * buffer size. 3117 * XXX: Unused. 3118 */ 3119 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3120 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3121 else 3122 len = so->so_rcv.sb_hiwat; 3123 #endif 3124 } else { 3125 m_freem(m); 3126 thflags &= ~TH_FIN; 3127 } 3128 3129 /* 3130 * If FIN is received ACK the FIN and let the user know 3131 * that the connection is closing. 3132 */ 3133 if (thflags & TH_FIN) { 3134 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3135 socantrcvmore(so); 3136 /* 3137 * If connection is half-synchronized 3138 * (ie NEEDSYN flag on) then delay ACK, 3139 * so it may be piggybacked when SYN is sent. 3140 * Otherwise, since we received a FIN then no 3141 * more input can be expected, send ACK now. 3142 */ 3143 if (tp->t_flags & TF_NEEDSYN) 3144 tp->t_flags |= TF_DELACK; 3145 else 3146 tp->t_flags |= TF_ACKNOW; 3147 tp->rcv_nxt++; 3148 } 3149 switch (tp->t_state) { 3150 3151 /* 3152 * In SYN_RECEIVED and ESTABLISHED STATES 3153 * enter the CLOSE_WAIT state. 3154 */ 3155 case TCPS_SYN_RECEIVED: 3156 tp->t_starttime = ticks; 3157 /* FALLTHROUGH */ 3158 case TCPS_ESTABLISHED: 3159 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3160 break; 3161 3162 /* 3163 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3164 * enter the CLOSING state. 3165 */ 3166 case TCPS_FIN_WAIT_1: 3167 tcp_state_change(tp, TCPS_CLOSING); 3168 break; 3169 3170 /* 3171 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3172 * starting the time-wait timer, turning off the other 3173 * standard timers. 3174 */ 3175 case TCPS_FIN_WAIT_2: 3176 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3177 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3178 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3179 ti_locked)); 3180 3181 tcp_twstart(tp); 3182 INP_INFO_RUNLOCK(&V_tcbinfo); 3183 return; 3184 } 3185 } 3186 if (ti_locked == TI_RLOCKED) 3187 INP_INFO_RUNLOCK(&V_tcbinfo); 3188 ti_locked = TI_UNLOCKED; 3189 3190 #ifdef TCPDEBUG 3191 if (so->so_options & SO_DEBUG) 3192 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3193 &tcp_savetcp, 0); 3194 #endif 3195 TCP_PROBE3(debug__input, tp, th, m); 3196 3197 /* 3198 * Return any desired output. 3199 */ 3200 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3201 (void) tp->t_fb->tfb_tcp_output(tp); 3202 3203 check_delack: 3204 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3205 __func__, ti_locked)); 3206 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3207 INP_WLOCK_ASSERT(tp->t_inpcb); 3208 3209 if (tp->t_flags & TF_DELACK) { 3210 tp->t_flags &= ~TF_DELACK; 3211 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3212 } 3213 INP_WUNLOCK(tp->t_inpcb); 3214 return; 3215 3216 dropafterack: 3217 /* 3218 * Generate an ACK dropping incoming segment if it occupies 3219 * sequence space, where the ACK reflects our state. 3220 * 3221 * We can now skip the test for the RST flag since all 3222 * paths to this code happen after packets containing 3223 * RST have been dropped. 3224 * 3225 * In the SYN-RECEIVED state, don't send an ACK unless the 3226 * segment we received passes the SYN-RECEIVED ACK test. 3227 * If it fails send a RST. This breaks the loop in the 3228 * "LAND" DoS attack, and also prevents an ACK storm 3229 * between two listening ports that have been sent forged 3230 * SYN segments, each with the source address of the other. 3231 */ 3232 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3233 (SEQ_GT(tp->snd_una, th->th_ack) || 3234 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3235 rstreason = BANDLIM_RST_OPENPORT; 3236 goto dropwithreset; 3237 } 3238 #ifdef TCPDEBUG 3239 if (so->so_options & SO_DEBUG) 3240 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3241 &tcp_savetcp, 0); 3242 #endif 3243 TCP_PROBE3(debug__input, tp, th, m); 3244 if (ti_locked == TI_RLOCKED) 3245 INP_INFO_RUNLOCK(&V_tcbinfo); 3246 ti_locked = TI_UNLOCKED; 3247 3248 tp->t_flags |= TF_ACKNOW; 3249 (void) tp->t_fb->tfb_tcp_output(tp); 3250 INP_WUNLOCK(tp->t_inpcb); 3251 m_freem(m); 3252 return; 3253 3254 dropwithreset: 3255 if (ti_locked == TI_RLOCKED) 3256 INP_INFO_RUNLOCK(&V_tcbinfo); 3257 ti_locked = TI_UNLOCKED; 3258 3259 if (tp != NULL) { 3260 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3261 INP_WUNLOCK(tp->t_inpcb); 3262 } else 3263 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3264 return; 3265 3266 drop: 3267 if (ti_locked == TI_RLOCKED) { 3268 INP_INFO_RUNLOCK(&V_tcbinfo); 3269 ti_locked = TI_UNLOCKED; 3270 } 3271 #ifdef INVARIANTS 3272 else 3273 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3274 #endif 3275 3276 /* 3277 * Drop space held by incoming segment and return. 3278 */ 3279 #ifdef TCPDEBUG 3280 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3281 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3282 &tcp_savetcp, 0); 3283 #endif 3284 TCP_PROBE3(debug__input, tp, th, m); 3285 if (tp != NULL) 3286 INP_WUNLOCK(tp->t_inpcb); 3287 m_freem(m); 3288 #ifndef TCP_RFC7413 3289 #undef tfo_syn 3290 #endif 3291 } 3292 3293 /* 3294 * Issue RST and make ACK acceptable to originator of segment. 3295 * The mbuf must still include the original packet header. 3296 * tp may be NULL. 3297 */ 3298 void 3299 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3300 int tlen, int rstreason) 3301 { 3302 #ifdef INET 3303 struct ip *ip; 3304 #endif 3305 #ifdef INET6 3306 struct ip6_hdr *ip6; 3307 #endif 3308 3309 if (tp != NULL) { 3310 INP_WLOCK_ASSERT(tp->t_inpcb); 3311 } 3312 3313 /* Don't bother if destination was broadcast/multicast. */ 3314 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3315 goto drop; 3316 #ifdef INET6 3317 if (mtod(m, struct ip *)->ip_v == 6) { 3318 ip6 = mtod(m, struct ip6_hdr *); 3319 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3320 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3321 goto drop; 3322 /* IPv6 anycast check is done at tcp6_input() */ 3323 } 3324 #endif 3325 #if defined(INET) && defined(INET6) 3326 else 3327 #endif 3328 #ifdef INET 3329 { 3330 ip = mtod(m, struct ip *); 3331 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3332 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3333 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3334 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3335 goto drop; 3336 } 3337 #endif 3338 3339 /* Perform bandwidth limiting. */ 3340 if (badport_bandlim(rstreason) < 0) 3341 goto drop; 3342 3343 /* tcp_respond consumes the mbuf chain. */ 3344 if (th->th_flags & TH_ACK) { 3345 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3346 th->th_ack, TH_RST); 3347 } else { 3348 if (th->th_flags & TH_SYN) 3349 tlen++; 3350 if (th->th_flags & TH_FIN) 3351 tlen++; 3352 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3353 (tcp_seq)0, TH_RST|TH_ACK); 3354 } 3355 return; 3356 drop: 3357 m_freem(m); 3358 } 3359 3360 /* 3361 * Parse TCP options and place in tcpopt. 3362 */ 3363 void 3364 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3365 { 3366 int opt, optlen; 3367 3368 to->to_flags = 0; 3369 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3370 opt = cp[0]; 3371 if (opt == TCPOPT_EOL) 3372 break; 3373 if (opt == TCPOPT_NOP) 3374 optlen = 1; 3375 else { 3376 if (cnt < 2) 3377 break; 3378 optlen = cp[1]; 3379 if (optlen < 2 || optlen > cnt) 3380 break; 3381 } 3382 switch (opt) { 3383 case TCPOPT_MAXSEG: 3384 if (optlen != TCPOLEN_MAXSEG) 3385 continue; 3386 if (!(flags & TO_SYN)) 3387 continue; 3388 to->to_flags |= TOF_MSS; 3389 bcopy((char *)cp + 2, 3390 (char *)&to->to_mss, sizeof(to->to_mss)); 3391 to->to_mss = ntohs(to->to_mss); 3392 break; 3393 case TCPOPT_WINDOW: 3394 if (optlen != TCPOLEN_WINDOW) 3395 continue; 3396 if (!(flags & TO_SYN)) 3397 continue; 3398 to->to_flags |= TOF_SCALE; 3399 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3400 break; 3401 case TCPOPT_TIMESTAMP: 3402 if (optlen != TCPOLEN_TIMESTAMP) 3403 continue; 3404 to->to_flags |= TOF_TS; 3405 bcopy((char *)cp + 2, 3406 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3407 to->to_tsval = ntohl(to->to_tsval); 3408 bcopy((char *)cp + 6, 3409 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3410 to->to_tsecr = ntohl(to->to_tsecr); 3411 break; 3412 case TCPOPT_SIGNATURE: 3413 /* 3414 * In order to reply to a host which has set the 3415 * TCP_SIGNATURE option in its initial SYN, we have 3416 * to record the fact that the option was observed 3417 * here for the syncache code to perform the correct 3418 * response. 3419 */ 3420 if (optlen != TCPOLEN_SIGNATURE) 3421 continue; 3422 to->to_flags |= TOF_SIGNATURE; 3423 to->to_signature = cp + 2; 3424 break; 3425 case TCPOPT_SACK_PERMITTED: 3426 if (optlen != TCPOLEN_SACK_PERMITTED) 3427 continue; 3428 if (!(flags & TO_SYN)) 3429 continue; 3430 if (!V_tcp_do_sack) 3431 continue; 3432 to->to_flags |= TOF_SACKPERM; 3433 break; 3434 case TCPOPT_SACK: 3435 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3436 continue; 3437 if (flags & TO_SYN) 3438 continue; 3439 to->to_flags |= TOF_SACK; 3440 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3441 to->to_sacks = cp + 2; 3442 TCPSTAT_INC(tcps_sack_rcv_blocks); 3443 break; 3444 #ifdef TCP_RFC7413 3445 case TCPOPT_FAST_OPEN: 3446 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3447 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3448 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3449 continue; 3450 if (!(flags & TO_SYN)) 3451 continue; 3452 if (!V_tcp_fastopen_enabled) 3453 continue; 3454 to->to_flags |= TOF_FASTOPEN; 3455 to->to_tfo_len = optlen - 2; 3456 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3457 break; 3458 #endif 3459 default: 3460 continue; 3461 } 3462 } 3463 } 3464 3465 /* 3466 * Pull out of band byte out of a segment so 3467 * it doesn't appear in the user's data queue. 3468 * It is still reflected in the segment length for 3469 * sequencing purposes. 3470 */ 3471 void 3472 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3473 int off) 3474 { 3475 int cnt = off + th->th_urp - 1; 3476 3477 while (cnt >= 0) { 3478 if (m->m_len > cnt) { 3479 char *cp = mtod(m, caddr_t) + cnt; 3480 struct tcpcb *tp = sototcpcb(so); 3481 3482 INP_WLOCK_ASSERT(tp->t_inpcb); 3483 3484 tp->t_iobc = *cp; 3485 tp->t_oobflags |= TCPOOB_HAVEDATA; 3486 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3487 m->m_len--; 3488 if (m->m_flags & M_PKTHDR) 3489 m->m_pkthdr.len--; 3490 return; 3491 } 3492 cnt -= m->m_len; 3493 m = m->m_next; 3494 if (m == NULL) 3495 break; 3496 } 3497 panic("tcp_pulloutofband"); 3498 } 3499 3500 /* 3501 * Collect new round-trip time estimate 3502 * and update averages and current timeout. 3503 */ 3504 void 3505 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3506 { 3507 int delta; 3508 3509 INP_WLOCK_ASSERT(tp->t_inpcb); 3510 3511 TCPSTAT_INC(tcps_rttupdated); 3512 tp->t_rttupdated++; 3513 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3514 /* 3515 * srtt is stored as fixed point with 5 bits after the 3516 * binary point (i.e., scaled by 8). The following magic 3517 * is equivalent to the smoothing algorithm in rfc793 with 3518 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3519 * point). Adjust rtt to origin 0. 3520 */ 3521 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3522 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3523 3524 if ((tp->t_srtt += delta) <= 0) 3525 tp->t_srtt = 1; 3526 3527 /* 3528 * We accumulate a smoothed rtt variance (actually, a 3529 * smoothed mean difference), then set the retransmit 3530 * timer to smoothed rtt + 4 times the smoothed variance. 3531 * rttvar is stored as fixed point with 4 bits after the 3532 * binary point (scaled by 16). The following is 3533 * equivalent to rfc793 smoothing with an alpha of .75 3534 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3535 * rfc793's wired-in beta. 3536 */ 3537 if (delta < 0) 3538 delta = -delta; 3539 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3540 if ((tp->t_rttvar += delta) <= 0) 3541 tp->t_rttvar = 1; 3542 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3543 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3544 } else { 3545 /* 3546 * No rtt measurement yet - use the unsmoothed rtt. 3547 * Set the variance to half the rtt (so our first 3548 * retransmit happens at 3*rtt). 3549 */ 3550 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3551 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3552 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3553 } 3554 tp->t_rtttime = 0; 3555 tp->t_rxtshift = 0; 3556 3557 /* 3558 * the retransmit should happen at rtt + 4 * rttvar. 3559 * Because of the way we do the smoothing, srtt and rttvar 3560 * will each average +1/2 tick of bias. When we compute 3561 * the retransmit timer, we want 1/2 tick of rounding and 3562 * 1 extra tick because of +-1/2 tick uncertainty in the 3563 * firing of the timer. The bias will give us exactly the 3564 * 1.5 tick we need. But, because the bias is 3565 * statistical, we have to test that we don't drop below 3566 * the minimum feasible timer (which is 2 ticks). 3567 */ 3568 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3569 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3570 3571 /* 3572 * We received an ack for a packet that wasn't retransmitted; 3573 * it is probably safe to discard any error indications we've 3574 * received recently. This isn't quite right, but close enough 3575 * for now (a route might have failed after we sent a segment, 3576 * and the return path might not be symmetrical). 3577 */ 3578 tp->t_softerror = 0; 3579 } 3580 3581 /* 3582 * Determine a reasonable value for maxseg size. 3583 * If the route is known, check route for mtu. 3584 * If none, use an mss that can be handled on the outgoing interface 3585 * without forcing IP to fragment. If no route is found, route has no mtu, 3586 * or the destination isn't local, use a default, hopefully conservative 3587 * size (usually 512 or the default IP max size, but no more than the mtu 3588 * of the interface), as we can't discover anything about intervening 3589 * gateways or networks. We also initialize the congestion/slow start 3590 * window to be a single segment if the destination isn't local. 3591 * While looking at the routing entry, we also initialize other path-dependent 3592 * parameters from pre-set or cached values in the routing entry. 3593 * 3594 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3595 * IP options, e.g. IPSEC data, since length of this data may vary, and 3596 * thus it is calculated for every segment separately in tcp_output(). 3597 * 3598 * NOTE that this routine is only called when we process an incoming 3599 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3600 * settings are handled in tcp_mssopt(). 3601 */ 3602 void 3603 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3604 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3605 { 3606 int mss = 0; 3607 uint32_t maxmtu = 0; 3608 struct inpcb *inp = tp->t_inpcb; 3609 struct hc_metrics_lite metrics; 3610 #ifdef INET6 3611 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3612 size_t min_protoh = isipv6 ? 3613 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3614 sizeof (struct tcpiphdr); 3615 #else 3616 const size_t min_protoh = sizeof(struct tcpiphdr); 3617 #endif 3618 3619 INP_WLOCK_ASSERT(tp->t_inpcb); 3620 3621 if (mtuoffer != -1) { 3622 KASSERT(offer == -1, ("%s: conflict", __func__)); 3623 offer = mtuoffer - min_protoh; 3624 } 3625 3626 /* Initialize. */ 3627 #ifdef INET6 3628 if (isipv6) { 3629 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3630 tp->t_maxseg = V_tcp_v6mssdflt; 3631 } 3632 #endif 3633 #if defined(INET) && defined(INET6) 3634 else 3635 #endif 3636 #ifdef INET 3637 { 3638 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3639 tp->t_maxseg = V_tcp_mssdflt; 3640 } 3641 #endif 3642 3643 /* 3644 * No route to sender, stay with default mss and return. 3645 */ 3646 if (maxmtu == 0) { 3647 /* 3648 * In case we return early we need to initialize metrics 3649 * to a defined state as tcp_hc_get() would do for us 3650 * if there was no cache hit. 3651 */ 3652 if (metricptr != NULL) 3653 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3654 return; 3655 } 3656 3657 /* What have we got? */ 3658 switch (offer) { 3659 case 0: 3660 /* 3661 * Offer == 0 means that there was no MSS on the SYN 3662 * segment, in this case we use tcp_mssdflt as 3663 * already assigned to t_maxseg above. 3664 */ 3665 offer = tp->t_maxseg; 3666 break; 3667 3668 case -1: 3669 /* 3670 * Offer == -1 means that we didn't receive SYN yet. 3671 */ 3672 /* FALLTHROUGH */ 3673 3674 default: 3675 /* 3676 * Prevent DoS attack with too small MSS. Round up 3677 * to at least minmss. 3678 */ 3679 offer = max(offer, V_tcp_minmss); 3680 } 3681 3682 /* 3683 * rmx information is now retrieved from tcp_hostcache. 3684 */ 3685 tcp_hc_get(&inp->inp_inc, &metrics); 3686 if (metricptr != NULL) 3687 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3688 3689 /* 3690 * If there's a discovered mtu in tcp hostcache, use it. 3691 * Else, use the link mtu. 3692 */ 3693 if (metrics.rmx_mtu) 3694 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3695 else { 3696 #ifdef INET6 3697 if (isipv6) { 3698 mss = maxmtu - min_protoh; 3699 if (!V_path_mtu_discovery && 3700 !in6_localaddr(&inp->in6p_faddr)) 3701 mss = min(mss, V_tcp_v6mssdflt); 3702 } 3703 #endif 3704 #if defined(INET) && defined(INET6) 3705 else 3706 #endif 3707 #ifdef INET 3708 { 3709 mss = maxmtu - min_protoh; 3710 if (!V_path_mtu_discovery && 3711 !in_localaddr(inp->inp_faddr)) 3712 mss = min(mss, V_tcp_mssdflt); 3713 } 3714 #endif 3715 /* 3716 * XXX - The above conditional (mss = maxmtu - min_protoh) 3717 * probably violates the TCP spec. 3718 * The problem is that, since we don't know the 3719 * other end's MSS, we are supposed to use a conservative 3720 * default. But, if we do that, then MTU discovery will 3721 * never actually take place, because the conservative 3722 * default is much less than the MTUs typically seen 3723 * on the Internet today. For the moment, we'll sweep 3724 * this under the carpet. 3725 * 3726 * The conservative default might not actually be a problem 3727 * if the only case this occurs is when sending an initial 3728 * SYN with options and data to a host we've never talked 3729 * to before. Then, they will reply with an MSS value which 3730 * will get recorded and the new parameters should get 3731 * recomputed. For Further Study. 3732 */ 3733 } 3734 mss = min(mss, offer); 3735 3736 /* 3737 * Sanity check: make sure that maxseg will be large 3738 * enough to allow some data on segments even if the 3739 * all the option space is used (40bytes). Otherwise 3740 * funny things may happen in tcp_output. 3741 * 3742 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3743 */ 3744 mss = max(mss, 64); 3745 3746 tp->t_maxseg = mss; 3747 } 3748 3749 void 3750 tcp_mss(struct tcpcb *tp, int offer) 3751 { 3752 int mss; 3753 uint32_t bufsize; 3754 struct inpcb *inp; 3755 struct socket *so; 3756 struct hc_metrics_lite metrics; 3757 struct tcp_ifcap cap; 3758 3759 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3760 3761 bzero(&cap, sizeof(cap)); 3762 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3763 3764 mss = tp->t_maxseg; 3765 inp = tp->t_inpcb; 3766 3767 /* 3768 * If there's a pipesize, change the socket buffer to that size, 3769 * don't change if sb_hiwat is different than default (then it 3770 * has been changed on purpose with setsockopt). 3771 * Make the socket buffers an integral number of mss units; 3772 * if the mss is larger than the socket buffer, decrease the mss. 3773 */ 3774 so = inp->inp_socket; 3775 SOCKBUF_LOCK(&so->so_snd); 3776 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3777 bufsize = metrics.rmx_sendpipe; 3778 else 3779 bufsize = so->so_snd.sb_hiwat; 3780 if (bufsize < mss) 3781 mss = bufsize; 3782 else { 3783 bufsize = roundup(bufsize, mss); 3784 if (bufsize > sb_max) 3785 bufsize = sb_max; 3786 if (bufsize > so->so_snd.sb_hiwat) 3787 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3788 } 3789 SOCKBUF_UNLOCK(&so->so_snd); 3790 /* 3791 * Sanity check: make sure that maxseg will be large 3792 * enough to allow some data on segments even if the 3793 * all the option space is used (40bytes). Otherwise 3794 * funny things may happen in tcp_output. 3795 * 3796 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3797 */ 3798 tp->t_maxseg = max(mss, 64); 3799 3800 SOCKBUF_LOCK(&so->so_rcv); 3801 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3802 bufsize = metrics.rmx_recvpipe; 3803 else 3804 bufsize = so->so_rcv.sb_hiwat; 3805 if (bufsize > mss) { 3806 bufsize = roundup(bufsize, mss); 3807 if (bufsize > sb_max) 3808 bufsize = sb_max; 3809 if (bufsize > so->so_rcv.sb_hiwat) 3810 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3811 } 3812 SOCKBUF_UNLOCK(&so->so_rcv); 3813 3814 /* Check the interface for TSO capabilities. */ 3815 if (cap.ifcap & CSUM_TSO) { 3816 tp->t_flags |= TF_TSO; 3817 tp->t_tsomax = cap.tsomax; 3818 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3819 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3820 } 3821 } 3822 3823 /* 3824 * Determine the MSS option to send on an outgoing SYN. 3825 */ 3826 int 3827 tcp_mssopt(struct in_conninfo *inc) 3828 { 3829 int mss = 0; 3830 uint32_t thcmtu = 0; 3831 uint32_t maxmtu = 0; 3832 size_t min_protoh; 3833 3834 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3835 3836 #ifdef INET6 3837 if (inc->inc_flags & INC_ISIPV6) { 3838 mss = V_tcp_v6mssdflt; 3839 maxmtu = tcp_maxmtu6(inc, NULL); 3840 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3841 } 3842 #endif 3843 #if defined(INET) && defined(INET6) 3844 else 3845 #endif 3846 #ifdef INET 3847 { 3848 mss = V_tcp_mssdflt; 3849 maxmtu = tcp_maxmtu(inc, NULL); 3850 min_protoh = sizeof(struct tcpiphdr); 3851 } 3852 #endif 3853 #if defined(INET6) || defined(INET) 3854 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3855 #endif 3856 3857 if (maxmtu && thcmtu) 3858 mss = min(maxmtu, thcmtu) - min_protoh; 3859 else if (maxmtu || thcmtu) 3860 mss = max(maxmtu, thcmtu) - min_protoh; 3861 3862 return (mss); 3863 } 3864 3865 3866 /* 3867 * On a partial ack arrives, force the retransmission of the 3868 * next unacknowledged segment. Do not clear tp->t_dupacks. 3869 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3870 * be started again. 3871 */ 3872 void 3873 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3874 { 3875 tcp_seq onxt = tp->snd_nxt; 3876 uint32_t ocwnd = tp->snd_cwnd; 3877 u_int maxseg = tcp_maxseg(tp); 3878 3879 INP_WLOCK_ASSERT(tp->t_inpcb); 3880 3881 tcp_timer_activate(tp, TT_REXMT, 0); 3882 tp->t_rtttime = 0; 3883 tp->snd_nxt = th->th_ack; 3884 /* 3885 * Set snd_cwnd to one segment beyond acknowledged offset. 3886 * (tp->snd_una has not yet been updated when this function is called.) 3887 */ 3888 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3889 tp->t_flags |= TF_ACKNOW; 3890 (void) tp->t_fb->tfb_tcp_output(tp); 3891 tp->snd_cwnd = ocwnd; 3892 if (SEQ_GT(onxt, tp->snd_nxt)) 3893 tp->snd_nxt = onxt; 3894 /* 3895 * Partial window deflation. Relies on fact that tp->snd_una 3896 * not updated yet. 3897 */ 3898 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3899 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3900 else 3901 tp->snd_cwnd = 0; 3902 tp->snd_cwnd += maxseg; 3903 } 3904 3905 int 3906 tcp_compute_pipe(struct tcpcb *tp) 3907 { 3908 return (tp->snd_max - tp->snd_una + 3909 tp->sackhint.sack_bytes_rexmit - 3910 tp->sackhint.sacked_bytes); 3911 } 3912