1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 __FBSDID("$FreeBSD$"); 54 55 #include "opt_inet.h" 56 #include "opt_inet6.h" 57 #include "opt_ipsec.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/kernel.h> 62 #ifdef TCP_HHOOK 63 #include <sys/hhook.h> 64 #endif 65 #include <sys/malloc.h> 66 #include <sys/mbuf.h> 67 #include <sys/proc.h> /* for proc0 declaration */ 68 #include <sys/protosw.h> 69 #include <sys/sdt.h> 70 #include <sys/signalvar.h> 71 #include <sys/socket.h> 72 #include <sys/socketvar.h> 73 #include <sys/sysctl.h> 74 #include <sys/syslog.h> 75 #include <sys/systm.h> 76 77 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 78 79 #include <vm/uma.h> 80 81 #include <net/if.h> 82 #include <net/if_var.h> 83 #include <net/route.h> 84 #include <net/vnet.h> 85 86 #define TCPSTATES /* for logging */ 87 88 #include <netinet/in.h> 89 #include <netinet/in_kdtrace.h> 90 #include <netinet/in_pcb.h> 91 #include <netinet/in_systm.h> 92 #include <netinet/ip.h> 93 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 94 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 95 #include <netinet/ip_var.h> 96 #include <netinet/ip_options.h> 97 #include <netinet/ip6.h> 98 #include <netinet/icmp6.h> 99 #include <netinet6/in6_pcb.h> 100 #include <netinet6/in6_var.h> 101 #include <netinet6/ip6_var.h> 102 #include <netinet6/nd6.h> 103 #include <netinet/tcp.h> 104 #include <netinet/tcp_fsm.h> 105 #include <netinet/tcp_log_buf.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 #include <netinet/tcp_fastopen.h> 113 #ifdef TCPPCAP 114 #include <netinet/tcp_pcap.h> 115 #endif 116 #include <netinet/tcp_syncache.h> 117 #ifdef TCPDEBUG 118 #include <netinet/tcp_debug.h> 119 #endif /* TCPDEBUG */ 120 #ifdef TCP_OFFLOAD 121 #include <netinet/tcp_offload.h> 122 #endif 123 124 #include <netipsec/ipsec_support.h> 125 126 #include <machine/in_cksum.h> 127 128 #include <security/mac/mac_framework.h> 129 130 const int tcprexmtthresh = 3; 131 132 int tcp_log_in_vain = 0; 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 134 &tcp_log_in_vain, 0, 135 "Log all incoming TCP segments to closed ports"); 136 137 VNET_DEFINE(int, blackhole) = 0; 138 #define V_blackhole VNET(blackhole) 139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 140 &VNET_NAME(blackhole), 0, 141 "Do not send RST on segments to closed ports"); 142 143 VNET_DEFINE(int, tcp_delack_enabled) = 1; 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 145 &VNET_NAME(tcp_delack_enabled), 0, 146 "Delay ACK to try and piggyback it onto a data packet"); 147 148 VNET_DEFINE(int, drop_synfin) = 0; 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 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 160 &VNET_NAME(tcp_do_rfc3042), 0, 161 "Enable RFC 3042 (Limited Transmit)"); 162 163 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 165 &VNET_NAME(tcp_do_rfc3390), 0, 166 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 167 168 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 170 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 171 "Slow-start flight size (initial congestion window) in number of segments"); 172 173 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 175 &VNET_NAME(tcp_do_rfc3465), 0, 176 "Enable RFC 3465 (Appropriate Byte Counting)"); 177 178 VNET_DEFINE(int, tcp_abc_l_var) = 2; 179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 180 &VNET_NAME(tcp_abc_l_var), 2, 181 "Cap the max cwnd increment during slow-start to this number of segments"); 182 183 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 184 185 VNET_DEFINE(int, tcp_do_ecn) = 2; 186 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 187 &VNET_NAME(tcp_do_ecn), 0, 188 "TCP ECN support"); 189 190 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 191 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_ecn_maxretries), 0, 193 "Max retries before giving up on ECN"); 194 195 VNET_DEFINE(int, tcp_insecure_syn) = 0; 196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 197 &VNET_NAME(tcp_insecure_syn), 0, 198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 199 200 VNET_DEFINE(int, tcp_insecure_rst) = 0; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 202 &VNET_NAME(tcp_insecure_rst), 0, 203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 204 205 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 206 #define V_tcp_recvspace VNET(tcp_recvspace) 207 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 209 210 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_do_autorcvbuf), 0, 213 "Enable automatic receive buffer sizing"); 214 215 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_autorcvbuf_inc), 0, 218 "Incrementor step size of automatic receive buffer"); 219 220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_autorcvbuf_max), 0, 223 "Max size of automatic receive buffer"); 224 225 VNET_DEFINE(struct inpcbhead, tcb); 226 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 227 VNET_DEFINE(struct inpcbinfo, tcbinfo); 228 229 /* 230 * TCP statistics are stored in an array of counter(9)s, which size matches 231 * size of struct tcpstat. TCP running connection count is a regular array. 232 */ 233 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 234 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 235 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 236 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 237 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 238 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 239 "TCP connection counts by TCP state"); 240 241 static void 242 tcp_vnet_init(const void *unused) 243 { 244 245 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 246 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 247 } 248 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 249 tcp_vnet_init, NULL); 250 251 #ifdef VIMAGE 252 static void 253 tcp_vnet_uninit(const void *unused) 254 { 255 256 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 257 VNET_PCPUSTAT_FREE(tcpstat); 258 } 259 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 260 tcp_vnet_uninit, NULL); 261 #endif /* VIMAGE */ 262 263 /* 264 * Kernel module interface for updating tcpstat. The argument is an index 265 * into tcpstat treated as an array. 266 */ 267 void 268 kmod_tcpstat_inc(int statnum) 269 { 270 271 counter_u64_add(VNET(tcpstat)[statnum], 1); 272 } 273 274 #ifdef TCP_HHOOK 275 /* 276 * Wrapper for the TCP established input helper hook. 277 */ 278 void 279 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 280 { 281 struct tcp_hhook_data hhook_data; 282 283 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 284 hhook_data.tp = tp; 285 hhook_data.th = th; 286 hhook_data.to = to; 287 288 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 289 tp->osd); 290 } 291 } 292 #endif 293 294 /* 295 * CC wrapper hook functions 296 */ 297 void 298 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 299 uint16_t type) 300 { 301 INP_WLOCK_ASSERT(tp->t_inpcb); 302 303 tp->ccv->nsegs = nsegs; 304 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 305 if (tp->snd_cwnd <= tp->snd_wnd) 306 tp->ccv->flags |= CCF_CWND_LIMITED; 307 else 308 tp->ccv->flags &= ~CCF_CWND_LIMITED; 309 310 if (type == CC_ACK) { 311 if (tp->snd_cwnd > tp->snd_ssthresh) { 312 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 313 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 314 if (tp->t_bytes_acked >= tp->snd_cwnd) { 315 tp->t_bytes_acked -= tp->snd_cwnd; 316 tp->ccv->flags |= CCF_ABC_SENTAWND; 317 } 318 } else { 319 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 320 tp->t_bytes_acked = 0; 321 } 322 } 323 324 if (CC_ALGO(tp)->ack_received != NULL) { 325 /* XXXLAS: Find a way to live without this */ 326 tp->ccv->curack = th->th_ack; 327 CC_ALGO(tp)->ack_received(tp->ccv, type); 328 } 329 } 330 331 void 332 cc_conn_init(struct tcpcb *tp) 333 { 334 struct hc_metrics_lite metrics; 335 struct inpcb *inp = tp->t_inpcb; 336 u_int maxseg; 337 int rtt; 338 339 INP_WLOCK_ASSERT(tp->t_inpcb); 340 341 tcp_hc_get(&inp->inp_inc, &metrics); 342 maxseg = tcp_maxseg(tp); 343 344 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 345 tp->t_srtt = rtt; 346 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 347 TCPSTAT_INC(tcps_usedrtt); 348 if (metrics.rmx_rttvar) { 349 tp->t_rttvar = metrics.rmx_rttvar; 350 TCPSTAT_INC(tcps_usedrttvar); 351 } else { 352 /* default variation is +- 1 rtt */ 353 tp->t_rttvar = 354 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 355 } 356 TCPT_RANGESET(tp->t_rxtcur, 357 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 358 tp->t_rttmin, TCPTV_REXMTMAX); 359 } 360 if (metrics.rmx_ssthresh) { 361 /* 362 * There's some sort of gateway or interface 363 * buffer limit on the path. Use this to set 364 * the slow start threshold, but set the 365 * threshold to no less than 2*mss. 366 */ 367 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 368 TCPSTAT_INC(tcps_usedssthresh); 369 } 370 371 /* 372 * Set the initial slow-start flight size. 373 * 374 * RFC5681 Section 3.1 specifies the default conservative values. 375 * RFC3390 specifies slightly more aggressive values. 376 * RFC6928 increases it to ten segments. 377 * Support for user specified value for initial flight size. 378 * 379 * If a SYN or SYN/ACK was lost and retransmitted, we have to 380 * reduce the initial CWND to one segment as congestion is likely 381 * requiring us to be cautious. 382 */ 383 if (tp->snd_cwnd == 1) 384 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 385 else if (V_tcp_initcwnd_segments) 386 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 387 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 388 else if (V_tcp_do_rfc3390) 389 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 390 else { 391 /* Per RFC5681 Section 3.1 */ 392 if (maxseg > 2190) 393 tp->snd_cwnd = 2 * maxseg; 394 else if (maxseg > 1095) 395 tp->snd_cwnd = 3 * maxseg; 396 else 397 tp->snd_cwnd = 4 * maxseg; 398 } 399 400 if (CC_ALGO(tp)->conn_init != NULL) 401 CC_ALGO(tp)->conn_init(tp->ccv); 402 } 403 404 void inline 405 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 406 { 407 u_int maxseg; 408 409 INP_WLOCK_ASSERT(tp->t_inpcb); 410 411 switch(type) { 412 case CC_NDUPACK: 413 if (!IN_FASTRECOVERY(tp->t_flags)) { 414 tp->snd_recover = tp->snd_max; 415 if (tp->t_flags & TF_ECN_PERMIT) 416 tp->t_flags |= TF_ECN_SND_CWR; 417 } 418 break; 419 case CC_ECN: 420 if (!IN_CONGRECOVERY(tp->t_flags)) { 421 TCPSTAT_INC(tcps_ecn_rcwnd); 422 tp->snd_recover = tp->snd_max; 423 if (tp->t_flags & TF_ECN_PERMIT) 424 tp->t_flags |= TF_ECN_SND_CWR; 425 } 426 break; 427 case CC_RTO: 428 maxseg = tcp_maxseg(tp); 429 tp->t_dupacks = 0; 430 tp->t_bytes_acked = 0; 431 EXIT_RECOVERY(tp->t_flags); 432 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 433 maxseg) * maxseg; 434 tp->snd_cwnd = maxseg; 435 break; 436 case CC_RTO_ERR: 437 TCPSTAT_INC(tcps_sndrexmitbad); 438 /* RTO was unnecessary, so reset everything. */ 439 tp->snd_cwnd = tp->snd_cwnd_prev; 440 tp->snd_ssthresh = tp->snd_ssthresh_prev; 441 tp->snd_recover = tp->snd_recover_prev; 442 if (tp->t_flags & TF_WASFRECOVERY) 443 ENTER_FASTRECOVERY(tp->t_flags); 444 if (tp->t_flags & TF_WASCRECOVERY) 445 ENTER_CONGRECOVERY(tp->t_flags); 446 tp->snd_nxt = tp->snd_max; 447 tp->t_flags &= ~TF_PREVVALID; 448 tp->t_badrxtwin = 0; 449 break; 450 } 451 452 if (CC_ALGO(tp)->cong_signal != NULL) { 453 if (th != NULL) 454 tp->ccv->curack = th->th_ack; 455 CC_ALGO(tp)->cong_signal(tp->ccv, type); 456 } 457 } 458 459 void inline 460 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 461 { 462 INP_WLOCK_ASSERT(tp->t_inpcb); 463 464 /* XXXLAS: KASSERT that we're in recovery? */ 465 466 if (CC_ALGO(tp)->post_recovery != NULL) { 467 tp->ccv->curack = th->th_ack; 468 CC_ALGO(tp)->post_recovery(tp->ccv); 469 } 470 /* XXXLAS: EXIT_RECOVERY ? */ 471 tp->t_bytes_acked = 0; 472 } 473 474 /* 475 * Indicate whether this ack should be delayed. We can delay the ack if 476 * following conditions are met: 477 * - There is no delayed ack timer in progress. 478 * - Our last ack wasn't a 0-sized window. We never want to delay 479 * the ack that opens up a 0-sized window. 480 * - LRO wasn't used for this segment. We make sure by checking that the 481 * segment size is not larger than the MSS. 482 */ 483 #define DELAY_ACK(tp, tlen) \ 484 ((!tcp_timer_active(tp, TT_DELACK) && \ 485 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 486 (tlen <= tp->t_maxseg) && \ 487 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 488 489 static void inline 490 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 491 { 492 INP_WLOCK_ASSERT(tp->t_inpcb); 493 494 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 495 switch (iptos & IPTOS_ECN_MASK) { 496 case IPTOS_ECN_CE: 497 tp->ccv->flags |= CCF_IPHDR_CE; 498 break; 499 case IPTOS_ECN_ECT0: 500 tp->ccv->flags &= ~CCF_IPHDR_CE; 501 break; 502 case IPTOS_ECN_ECT1: 503 tp->ccv->flags &= ~CCF_IPHDR_CE; 504 break; 505 } 506 507 if (th->th_flags & TH_CWR) 508 tp->ccv->flags |= CCF_TCPHDR_CWR; 509 else 510 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 511 512 if (tp->t_flags & TF_DELACK) 513 tp->ccv->flags |= CCF_DELACK; 514 else 515 tp->ccv->flags &= ~CCF_DELACK; 516 517 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 518 519 if (tp->ccv->flags & CCF_ACKNOW) 520 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 521 } 522 } 523 524 /* 525 * TCP input handling is split into multiple parts: 526 * tcp6_input is a thin wrapper around tcp_input for the extended 527 * ip6_protox[] call format in ip6_input 528 * tcp_input handles primary segment validation, inpcb lookup and 529 * SYN processing on listen sockets 530 * tcp_do_segment processes the ACK and text of the segment for 531 * establishing, established and closing connections 532 */ 533 #ifdef INET6 534 int 535 tcp6_input(struct mbuf **mp, int *offp, int proto) 536 { 537 struct mbuf *m = *mp; 538 struct in6_ifaddr *ia6; 539 struct ip6_hdr *ip6; 540 541 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 542 543 /* 544 * draft-itojun-ipv6-tcp-to-anycast 545 * better place to put this in? 546 */ 547 ip6 = mtod(m, struct ip6_hdr *); 548 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 549 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 550 struct ip6_hdr *ip6; 551 552 ifa_free(&ia6->ia_ifa); 553 ip6 = mtod(m, struct ip6_hdr *); 554 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 555 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 556 return (IPPROTO_DONE); 557 } 558 if (ia6) 559 ifa_free(&ia6->ia_ifa); 560 561 return (tcp_input(mp, offp, proto)); 562 } 563 #endif /* INET6 */ 564 565 int 566 tcp_input(struct mbuf **mp, int *offp, int proto) 567 { 568 struct mbuf *m = *mp; 569 struct tcphdr *th = NULL; 570 struct ip *ip = NULL; 571 struct inpcb *inp = NULL; 572 struct tcpcb *tp = NULL; 573 struct socket *so = NULL; 574 u_char *optp = NULL; 575 int off0; 576 int optlen = 0; 577 #ifdef INET 578 int len; 579 #endif 580 int tlen = 0, off; 581 int drop_hdrlen; 582 int thflags; 583 int rstreason = 0; /* For badport_bandlim accounting purposes */ 584 uint8_t iptos; 585 struct m_tag *fwd_tag = NULL; 586 struct epoch_tracker et; 587 #ifdef INET6 588 struct ip6_hdr *ip6 = NULL; 589 int isipv6; 590 #else 591 const void *ip6 = NULL; 592 #endif /* INET6 */ 593 struct tcpopt to; /* options in this segment */ 594 char *s = NULL; /* address and port logging */ 595 int ti_locked; 596 #ifdef TCPDEBUG 597 /* 598 * The size of tcp_saveipgen must be the size of the max ip header, 599 * now IPv6. 600 */ 601 u_char tcp_saveipgen[IP6_HDR_LEN]; 602 struct tcphdr tcp_savetcp; 603 short ostate = 0; 604 #endif 605 606 #ifdef INET6 607 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 608 #endif 609 610 off0 = *offp; 611 m = *mp; 612 *mp = NULL; 613 to.to_flags = 0; 614 TCPSTAT_INC(tcps_rcvtotal); 615 616 #ifdef INET6 617 if (isipv6) { 618 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 619 620 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 621 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 622 if (m == NULL) { 623 TCPSTAT_INC(tcps_rcvshort); 624 return (IPPROTO_DONE); 625 } 626 } 627 628 ip6 = mtod(m, struct ip6_hdr *); 629 th = (struct tcphdr *)((caddr_t)ip6 + off0); 630 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 631 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 632 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 633 th->th_sum = m->m_pkthdr.csum_data; 634 else 635 th->th_sum = in6_cksum_pseudo(ip6, tlen, 636 IPPROTO_TCP, m->m_pkthdr.csum_data); 637 th->th_sum ^= 0xffff; 638 } else 639 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 640 if (th->th_sum) { 641 TCPSTAT_INC(tcps_rcvbadsum); 642 goto drop; 643 } 644 645 /* 646 * Be proactive about unspecified IPv6 address in source. 647 * As we use all-zero to indicate unbounded/unconnected pcb, 648 * unspecified IPv6 address can be used to confuse us. 649 * 650 * Note that packets with unspecified IPv6 destination is 651 * already dropped in ip6_input. 652 */ 653 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 654 /* XXX stat */ 655 goto drop; 656 } 657 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 658 } 659 #endif 660 #if defined(INET) && defined(INET6) 661 else 662 #endif 663 #ifdef INET 664 { 665 /* 666 * Get IP and TCP header together in first mbuf. 667 * Note: IP leaves IP header in first mbuf. 668 */ 669 if (off0 > sizeof (struct ip)) { 670 ip_stripoptions(m); 671 off0 = sizeof(struct ip); 672 } 673 if (m->m_len < sizeof (struct tcpiphdr)) { 674 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 675 == NULL) { 676 TCPSTAT_INC(tcps_rcvshort); 677 return (IPPROTO_DONE); 678 } 679 } 680 ip = mtod(m, struct ip *); 681 th = (struct tcphdr *)((caddr_t)ip + off0); 682 tlen = ntohs(ip->ip_len) - off0; 683 684 iptos = ip->ip_tos; 685 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 686 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 687 th->th_sum = m->m_pkthdr.csum_data; 688 else 689 th->th_sum = in_pseudo(ip->ip_src.s_addr, 690 ip->ip_dst.s_addr, 691 htonl(m->m_pkthdr.csum_data + tlen + 692 IPPROTO_TCP)); 693 th->th_sum ^= 0xffff; 694 } else { 695 struct ipovly *ipov = (struct ipovly *)ip; 696 697 /* 698 * Checksum extended TCP header and data. 699 */ 700 len = off0 + tlen; 701 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 702 ipov->ih_len = htons(tlen); 703 th->th_sum = in_cksum(m, len); 704 /* Reset length for SDT probes. */ 705 ip->ip_len = htons(len); 706 /* Reset TOS bits */ 707 ip->ip_tos = iptos; 708 /* Re-initialization for later version check */ 709 ip->ip_v = IPVERSION; 710 ip->ip_hl = off0 >> 2; 711 } 712 713 if (th->th_sum) { 714 TCPSTAT_INC(tcps_rcvbadsum); 715 goto drop; 716 } 717 } 718 #endif /* INET */ 719 720 /* 721 * Check that TCP offset makes sense, 722 * pull out TCP options and adjust length. XXX 723 */ 724 off = th->th_off << 2; 725 if (off < sizeof (struct tcphdr) || off > tlen) { 726 TCPSTAT_INC(tcps_rcvbadoff); 727 goto drop; 728 } 729 tlen -= off; /* tlen is used instead of ti->ti_len */ 730 if (off > sizeof (struct tcphdr)) { 731 #ifdef INET6 732 if (isipv6) { 733 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 734 ip6 = mtod(m, struct ip6_hdr *); 735 th = (struct tcphdr *)((caddr_t)ip6 + off0); 736 } 737 #endif 738 #if defined(INET) && defined(INET6) 739 else 740 #endif 741 #ifdef INET 742 { 743 if (m->m_len < sizeof(struct ip) + off) { 744 if ((m = m_pullup(m, sizeof (struct ip) + off)) 745 == NULL) { 746 TCPSTAT_INC(tcps_rcvshort); 747 return (IPPROTO_DONE); 748 } 749 ip = mtod(m, struct ip *); 750 th = (struct tcphdr *)((caddr_t)ip + off0); 751 } 752 } 753 #endif 754 optlen = off - sizeof (struct tcphdr); 755 optp = (u_char *)(th + 1); 756 } 757 thflags = th->th_flags; 758 759 /* 760 * Convert TCP protocol specific fields to host format. 761 */ 762 tcp_fields_to_host(th); 763 764 /* 765 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 766 */ 767 drop_hdrlen = off0 + off; 768 769 /* 770 * Locate pcb for segment; if we're likely to add or remove a 771 * connection then first acquire pcbinfo lock. There are three cases 772 * where we might discover later we need a write lock despite the 773 * flags: ACKs moving a connection out of the syncache, ACKs for a 774 * connection in TIMEWAIT and SYNs not targeting a listening socket. 775 */ 776 if ((thflags & (TH_FIN | TH_RST)) != 0) { 777 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 778 ti_locked = TI_RLOCKED; 779 } else 780 ti_locked = TI_UNLOCKED; 781 782 /* 783 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 784 */ 785 if ( 786 #ifdef INET6 787 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 788 #ifdef INET 789 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 790 #endif 791 #endif 792 #if defined(INET) && !defined(INET6) 793 (m->m_flags & M_IP_NEXTHOP) 794 #endif 795 ) 796 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 797 798 findpcb: 799 #ifdef INVARIANTS 800 if (ti_locked == TI_RLOCKED) { 801 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 802 } else { 803 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 804 } 805 #endif 806 #ifdef INET6 807 if (isipv6 && fwd_tag != NULL) { 808 struct sockaddr_in6 *next_hop6; 809 810 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 811 /* 812 * Transparently forwarded. Pretend to be the destination. 813 * Already got one like this? 814 */ 815 inp = in6_pcblookup_mbuf(&V_tcbinfo, 816 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 817 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 818 if (!inp) { 819 /* 820 * It's new. Try to find the ambushing socket. 821 * Because we've rewritten the destination address, 822 * any hardware-generated hash is ignored. 823 */ 824 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 825 th->th_sport, &next_hop6->sin6_addr, 826 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 827 th->th_dport, INPLOOKUP_WILDCARD | 828 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 829 } 830 } else if (isipv6) { 831 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 832 th->th_sport, &ip6->ip6_dst, th->th_dport, 833 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 834 m->m_pkthdr.rcvif, m); 835 } 836 #endif /* INET6 */ 837 #if defined(INET6) && defined(INET) 838 else 839 #endif 840 #ifdef INET 841 if (fwd_tag != NULL) { 842 struct sockaddr_in *next_hop; 843 844 next_hop = (struct sockaddr_in *)(fwd_tag+1); 845 /* 846 * Transparently forwarded. Pretend to be the destination. 847 * already got one like this? 848 */ 849 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 850 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 851 m->m_pkthdr.rcvif, m); 852 if (!inp) { 853 /* 854 * It's new. Try to find the ambushing socket. 855 * Because we've rewritten the destination address, 856 * any hardware-generated hash is ignored. 857 */ 858 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 859 th->th_sport, next_hop->sin_addr, 860 next_hop->sin_port ? ntohs(next_hop->sin_port) : 861 th->th_dport, INPLOOKUP_WILDCARD | 862 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 863 } 864 } else 865 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 866 th->th_sport, ip->ip_dst, th->th_dport, 867 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 868 m->m_pkthdr.rcvif, m); 869 #endif /* INET */ 870 871 /* 872 * If the INPCB does not exist then all data in the incoming 873 * segment is discarded and an appropriate RST is sent back. 874 * XXX MRT Send RST using which routing table? 875 */ 876 if (inp == NULL) { 877 /* 878 * Log communication attempts to ports that are not 879 * in use. 880 */ 881 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 882 tcp_log_in_vain == 2) { 883 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 884 log(LOG_INFO, "%s; %s: Connection attempt " 885 "to closed port\n", s, __func__); 886 } 887 /* 888 * When blackholing do not respond with a RST but 889 * completely ignore the segment and drop it. 890 */ 891 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 892 V_blackhole == 2) 893 goto dropunlock; 894 895 rstreason = BANDLIM_RST_CLOSEDPORT; 896 goto dropwithreset; 897 } 898 INP_WLOCK_ASSERT(inp); 899 /* 900 * While waiting for inp lock during the lookup, another thread 901 * can have dropped the inpcb, in which case we need to loop back 902 * and try to find a new inpcb to deliver to. 903 */ 904 if (inp->inp_flags & INP_DROPPED) { 905 INP_WUNLOCK(inp); 906 inp = NULL; 907 goto findpcb; 908 } 909 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 910 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 911 ((inp->inp_socket == NULL) || 912 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 913 inp->inp_flowid = m->m_pkthdr.flowid; 914 inp->inp_flowtype = M_HASHTYPE_GET(m); 915 } 916 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 917 #ifdef INET6 918 if (isipv6 && IPSEC_ENABLED(ipv6) && 919 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 920 goto dropunlock; 921 } 922 #ifdef INET 923 else 924 #endif 925 #endif /* INET6 */ 926 #ifdef INET 927 if (IPSEC_ENABLED(ipv4) && 928 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 929 goto dropunlock; 930 } 931 #endif /* INET */ 932 #endif /* IPSEC */ 933 934 /* 935 * Check the minimum TTL for socket. 936 */ 937 if (inp->inp_ip_minttl != 0) { 938 #ifdef INET6 939 if (isipv6) { 940 if (inp->inp_ip_minttl > ip6->ip6_hlim) 941 goto dropunlock; 942 } else 943 #endif 944 if (inp->inp_ip_minttl > ip->ip_ttl) 945 goto dropunlock; 946 } 947 948 /* 949 * A previous connection in TIMEWAIT state is supposed to catch stray 950 * or duplicate segments arriving late. If this segment was a 951 * legitimate new connection attempt, the old INPCB gets removed and 952 * we can try again to find a listening socket. 953 * 954 * At this point, due to earlier optimism, we may hold only an inpcb 955 * lock, and not the inpcbinfo write lock. If so, we need to try to 956 * acquire it, or if that fails, acquire a reference on the inpcb, 957 * drop all locks, acquire a global write lock, and then re-acquire 958 * the inpcb lock. We may at that point discover that another thread 959 * has tried to free the inpcb, in which case we need to loop back 960 * and try to find a new inpcb to deliver to. 961 * 962 * XXXRW: It may be time to rethink timewait locking. 963 */ 964 if (inp->inp_flags & INP_TIMEWAIT) { 965 if (ti_locked == TI_UNLOCKED) { 966 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 967 ti_locked = TI_RLOCKED; 968 } 969 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 970 971 if (thflags & TH_SYN) 972 tcp_dooptions(&to, optp, optlen, TO_SYN); 973 /* 974 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 975 */ 976 if (tcp_twcheck(inp, &to, th, m, tlen)) 977 goto findpcb; 978 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 979 return (IPPROTO_DONE); 980 } 981 /* 982 * The TCPCB may no longer exist if the connection is winding 983 * down or it is in the CLOSED state. Either way we drop the 984 * segment and send an appropriate response. 985 */ 986 tp = intotcpcb(inp); 987 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 988 rstreason = BANDLIM_RST_CLOSEDPORT; 989 goto dropwithreset; 990 } 991 992 #ifdef TCP_OFFLOAD 993 if (tp->t_flags & TF_TOE) { 994 tcp_offload_input(tp, m); 995 m = NULL; /* consumed by the TOE driver */ 996 goto dropunlock; 997 } 998 #endif 999 1000 /* 1001 * We've identified a valid inpcb, but it could be that we need an 1002 * inpcbinfo write lock but don't hold it. In this case, attempt to 1003 * acquire using the same strategy as the TIMEWAIT case above. If we 1004 * relock, we have to jump back to 'relocked' as the connection might 1005 * now be in TIMEWAIT. 1006 */ 1007 #ifdef INVARIANTS 1008 if ((thflags & (TH_FIN | TH_RST)) != 0) 1009 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1010 #endif 1011 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1012 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1013 !IS_FASTOPEN(tp->t_flags)))) { 1014 if (ti_locked == TI_UNLOCKED) { 1015 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 1016 ti_locked = TI_RLOCKED; 1017 } 1018 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1019 } 1020 1021 #ifdef MAC 1022 INP_WLOCK_ASSERT(inp); 1023 if (mac_inpcb_check_deliver(inp, m)) 1024 goto dropunlock; 1025 #endif 1026 so = inp->inp_socket; 1027 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1028 #ifdef TCPDEBUG 1029 if (so->so_options & SO_DEBUG) { 1030 ostate = tp->t_state; 1031 #ifdef INET6 1032 if (isipv6) { 1033 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1034 } else 1035 #endif 1036 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1037 tcp_savetcp = *th; 1038 } 1039 #endif /* TCPDEBUG */ 1040 /* 1041 * When the socket is accepting connections (the INPCB is in LISTEN 1042 * state) we look into the SYN cache if this is a new connection 1043 * attempt or the completion of a previous one. 1044 */ 1045 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1046 ("%s: so accepting but tp %p not listening", __func__, tp)); 1047 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1048 struct in_conninfo inc; 1049 1050 bzero(&inc, sizeof(inc)); 1051 #ifdef INET6 1052 if (isipv6) { 1053 inc.inc_flags |= INC_ISIPV6; 1054 inc.inc6_faddr = ip6->ip6_src; 1055 inc.inc6_laddr = ip6->ip6_dst; 1056 } else 1057 #endif 1058 { 1059 inc.inc_faddr = ip->ip_src; 1060 inc.inc_laddr = ip->ip_dst; 1061 } 1062 inc.inc_fport = th->th_sport; 1063 inc.inc_lport = th->th_dport; 1064 inc.inc_fibnum = so->so_fibnum; 1065 1066 /* 1067 * Check for an existing connection attempt in syncache if 1068 * the flag is only ACK. A successful lookup creates a new 1069 * socket appended to the listen queue in SYN_RECEIVED state. 1070 */ 1071 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1072 1073 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1074 /* 1075 * Parse the TCP options here because 1076 * syncookies need access to the reflected 1077 * timestamp. 1078 */ 1079 tcp_dooptions(&to, optp, optlen, 0); 1080 /* 1081 * NB: syncache_expand() doesn't unlock 1082 * inp and tcpinfo locks. 1083 */ 1084 rstreason = syncache_expand(&inc, &to, th, &so, m); 1085 if (rstreason < 0) { 1086 /* 1087 * A failing TCP MD5 signature comparison 1088 * must result in the segment being dropped 1089 * and must not produce any response back 1090 * to the sender. 1091 */ 1092 goto dropunlock; 1093 } else if (rstreason == 0) { 1094 /* 1095 * No syncache entry or ACK was not 1096 * for our SYN/ACK. Send a RST. 1097 * NB: syncache did its own logging 1098 * of the failure cause. 1099 */ 1100 rstreason = BANDLIM_RST_OPENPORT; 1101 goto dropwithreset; 1102 } 1103 tfo_socket_result: 1104 if (so == NULL) { 1105 /* 1106 * We completed the 3-way handshake 1107 * but could not allocate a socket 1108 * either due to memory shortage, 1109 * listen queue length limits or 1110 * global socket limits. Send RST 1111 * or wait and have the remote end 1112 * retransmit the ACK for another 1113 * try. 1114 */ 1115 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1116 log(LOG_DEBUG, "%s; %s: Listen socket: " 1117 "Socket allocation failed due to " 1118 "limits or memory shortage, %s\n", 1119 s, __func__, 1120 V_tcp_sc_rst_sock_fail ? 1121 "sending RST" : "try again"); 1122 if (V_tcp_sc_rst_sock_fail) { 1123 rstreason = BANDLIM_UNLIMITED; 1124 goto dropwithreset; 1125 } else 1126 goto dropunlock; 1127 } 1128 /* 1129 * Socket is created in state SYN_RECEIVED. 1130 * Unlock the listen socket, lock the newly 1131 * created socket and update the tp variable. 1132 */ 1133 INP_WUNLOCK(inp); /* listen socket */ 1134 inp = sotoinpcb(so); 1135 /* 1136 * New connection inpcb is already locked by 1137 * syncache_expand(). 1138 */ 1139 INP_WLOCK_ASSERT(inp); 1140 tp = intotcpcb(inp); 1141 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1142 ("%s: ", __func__)); 1143 /* 1144 * Process the segment and the data it 1145 * contains. tcp_do_segment() consumes 1146 * the mbuf chain and unlocks the inpcb. 1147 */ 1148 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1149 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1150 iptos); 1151 if (ti_locked == TI_RLOCKED) 1152 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1153 return (IPPROTO_DONE); 1154 } 1155 /* 1156 * Segment flag validation for new connection attempts: 1157 * 1158 * Our (SYN|ACK) response was rejected. 1159 * Check with syncache and remove entry to prevent 1160 * retransmits. 1161 * 1162 * NB: syncache_chkrst does its own logging of failure 1163 * causes. 1164 */ 1165 if (thflags & TH_RST) { 1166 syncache_chkrst(&inc, th); 1167 goto dropunlock; 1168 } 1169 /* 1170 * We can't do anything without SYN. 1171 */ 1172 if ((thflags & TH_SYN) == 0) { 1173 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1174 log(LOG_DEBUG, "%s; %s: Listen socket: " 1175 "SYN is missing, segment ignored\n", 1176 s, __func__); 1177 TCPSTAT_INC(tcps_badsyn); 1178 goto dropunlock; 1179 } 1180 /* 1181 * (SYN|ACK) is bogus on a listen socket. 1182 */ 1183 if (thflags & TH_ACK) { 1184 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1185 log(LOG_DEBUG, "%s; %s: Listen socket: " 1186 "SYN|ACK invalid, segment rejected\n", 1187 s, __func__); 1188 syncache_badack(&inc); /* XXX: Not needed! */ 1189 TCPSTAT_INC(tcps_badsyn); 1190 rstreason = BANDLIM_RST_OPENPORT; 1191 goto dropwithreset; 1192 } 1193 /* 1194 * If the drop_synfin option is enabled, drop all 1195 * segments with both the SYN and FIN bits set. 1196 * This prevents e.g. nmap from identifying the 1197 * TCP/IP stack. 1198 * XXX: Poor reasoning. nmap has other methods 1199 * and is constantly refining its stack detection 1200 * strategies. 1201 * XXX: This is a violation of the TCP specification 1202 * and was used by RFC1644. 1203 */ 1204 if ((thflags & TH_FIN) && V_drop_synfin) { 1205 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1206 log(LOG_DEBUG, "%s; %s: Listen socket: " 1207 "SYN|FIN segment ignored (based on " 1208 "sysctl setting)\n", s, __func__); 1209 TCPSTAT_INC(tcps_badsyn); 1210 goto dropunlock; 1211 } 1212 /* 1213 * Segment's flags are (SYN) or (SYN|FIN). 1214 * 1215 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1216 * as they do not affect the state of the TCP FSM. 1217 * The data pointed to by TH_URG and th_urp is ignored. 1218 */ 1219 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1220 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1221 KASSERT(thflags & (TH_SYN), 1222 ("%s: Listen socket: TH_SYN not set", __func__)); 1223 #ifdef INET6 1224 /* 1225 * If deprecated address is forbidden, 1226 * we do not accept SYN to deprecated interface 1227 * address to prevent any new inbound connection from 1228 * getting established. 1229 * When we do not accept SYN, we send a TCP RST, 1230 * with deprecated source address (instead of dropping 1231 * it). We compromise it as it is much better for peer 1232 * to send a RST, and RST will be the final packet 1233 * for the exchange. 1234 * 1235 * If we do not forbid deprecated addresses, we accept 1236 * the SYN packet. RFC2462 does not suggest dropping 1237 * SYN in this case. 1238 * If we decipher RFC2462 5.5.4, it says like this: 1239 * 1. use of deprecated addr with existing 1240 * communication is okay - "SHOULD continue to be 1241 * used" 1242 * 2. use of it with new communication: 1243 * (2a) "SHOULD NOT be used if alternate address 1244 * with sufficient scope is available" 1245 * (2b) nothing mentioned otherwise. 1246 * Here we fall into (2b) case as we have no choice in 1247 * our source address selection - we must obey the peer. 1248 * 1249 * The wording in RFC2462 is confusing, and there are 1250 * multiple description text for deprecated address 1251 * handling - worse, they are not exactly the same. 1252 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1253 */ 1254 if (isipv6 && !V_ip6_use_deprecated) { 1255 struct in6_ifaddr *ia6; 1256 1257 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1258 if (ia6 != NULL && 1259 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1260 ifa_free(&ia6->ia_ifa); 1261 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1262 log(LOG_DEBUG, "%s; %s: Listen socket: " 1263 "Connection attempt to deprecated " 1264 "IPv6 address rejected\n", 1265 s, __func__); 1266 rstreason = BANDLIM_RST_OPENPORT; 1267 goto dropwithreset; 1268 } 1269 if (ia6) 1270 ifa_free(&ia6->ia_ifa); 1271 } 1272 #endif /* INET6 */ 1273 /* 1274 * Basic sanity checks on incoming SYN requests: 1275 * Don't respond if the destination is a link layer 1276 * broadcast according to RFC1122 4.2.3.10, p. 104. 1277 * If it is from this socket it must be forged. 1278 * Don't respond if the source or destination is a 1279 * global or subnet broad- or multicast address. 1280 * Note that it is quite possible to receive unicast 1281 * link-layer packets with a broadcast IP address. Use 1282 * in_broadcast() to find them. 1283 */ 1284 if (m->m_flags & (M_BCAST|M_MCAST)) { 1285 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1286 log(LOG_DEBUG, "%s; %s: Listen socket: " 1287 "Connection attempt from broad- or multicast " 1288 "link layer address ignored\n", s, __func__); 1289 goto dropunlock; 1290 } 1291 #ifdef INET6 1292 if (isipv6) { 1293 if (th->th_dport == th->th_sport && 1294 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1295 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1296 log(LOG_DEBUG, "%s; %s: Listen socket: " 1297 "Connection attempt to/from self " 1298 "ignored\n", s, __func__); 1299 goto dropunlock; 1300 } 1301 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1302 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1304 log(LOG_DEBUG, "%s; %s: Listen socket: " 1305 "Connection attempt from/to multicast " 1306 "address ignored\n", s, __func__); 1307 goto dropunlock; 1308 } 1309 } 1310 #endif 1311 #if defined(INET) && defined(INET6) 1312 else 1313 #endif 1314 #ifdef INET 1315 { 1316 if (th->th_dport == th->th_sport && 1317 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1318 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1319 log(LOG_DEBUG, "%s; %s: Listen socket: " 1320 "Connection attempt from/to self " 1321 "ignored\n", s, __func__); 1322 goto dropunlock; 1323 } 1324 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1325 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1326 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1327 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1328 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1329 log(LOG_DEBUG, "%s; %s: Listen socket: " 1330 "Connection attempt from/to broad- " 1331 "or multicast address ignored\n", 1332 s, __func__); 1333 goto dropunlock; 1334 } 1335 } 1336 #endif 1337 /* 1338 * SYN appears to be valid. Create compressed TCP state 1339 * for syncache. 1340 */ 1341 #ifdef TCPDEBUG 1342 if (so->so_options & SO_DEBUG) 1343 tcp_trace(TA_INPUT, ostate, tp, 1344 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1345 #endif 1346 TCP_PROBE3(debug__input, tp, th, m); 1347 tcp_dooptions(&to, optp, optlen, TO_SYN); 1348 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1349 goto tfo_socket_result; 1350 1351 /* 1352 * Entry added to syncache and mbuf consumed. 1353 * Only the listen socket is unlocked by syncache_add(). 1354 */ 1355 if (ti_locked == TI_RLOCKED) { 1356 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1357 ti_locked = TI_UNLOCKED; 1358 } 1359 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1360 return (IPPROTO_DONE); 1361 } else if (tp->t_state == TCPS_LISTEN) { 1362 /* 1363 * When a listen socket is torn down the SO_ACCEPTCONN 1364 * flag is removed first while connections are drained 1365 * from the accept queue in a unlock/lock cycle of the 1366 * ACCEPT_LOCK, opening a race condition allowing a SYN 1367 * attempt go through unhandled. 1368 */ 1369 goto dropunlock; 1370 } 1371 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1372 if (tp->t_flags & TF_SIGNATURE) { 1373 tcp_dooptions(&to, optp, optlen, thflags); 1374 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1375 TCPSTAT_INC(tcps_sig_err_nosigopt); 1376 goto dropunlock; 1377 } 1378 if (!TCPMD5_ENABLED() || 1379 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1380 goto dropunlock; 1381 } 1382 #endif 1383 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1384 1385 /* 1386 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1387 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1388 * the inpcb, and unlocks pcbinfo. 1389 */ 1390 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1391 if (ti_locked == TI_RLOCKED) 1392 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1393 return (IPPROTO_DONE); 1394 1395 dropwithreset: 1396 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1397 1398 if (ti_locked == TI_RLOCKED) { 1399 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1400 ti_locked = TI_UNLOCKED; 1401 } 1402 #ifdef INVARIANTS 1403 else { 1404 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1405 "ti_locked: %d", __func__, ti_locked)); 1406 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1407 } 1408 #endif 1409 1410 if (inp != NULL) { 1411 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1412 INP_WUNLOCK(inp); 1413 } else 1414 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1415 m = NULL; /* mbuf chain got consumed. */ 1416 goto drop; 1417 1418 dropunlock: 1419 if (m != NULL) 1420 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1421 1422 if (ti_locked == TI_RLOCKED) { 1423 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1424 ti_locked = TI_UNLOCKED; 1425 } 1426 #ifdef INVARIANTS 1427 else { 1428 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1429 "ti_locked: %d", __func__, ti_locked)); 1430 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1431 } 1432 #endif 1433 1434 if (inp != NULL) 1435 INP_WUNLOCK(inp); 1436 1437 drop: 1438 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1439 if (s != NULL) 1440 free(s, M_TCPLOG); 1441 if (m != NULL) 1442 m_freem(m); 1443 return (IPPROTO_DONE); 1444 } 1445 1446 /* 1447 * Automatic sizing of receive socket buffer. Often the send 1448 * buffer size is not optimally adjusted to the actual network 1449 * conditions at hand (delay bandwidth product). Setting the 1450 * buffer size too small limits throughput on links with high 1451 * bandwidth and high delay (eg. trans-continental/oceanic links). 1452 * 1453 * On the receive side the socket buffer memory is only rarely 1454 * used to any significant extent. This allows us to be much 1455 * more aggressive in scaling the receive socket buffer. For 1456 * the case that the buffer space is actually used to a large 1457 * extent and we run out of kernel memory we can simply drop 1458 * the new segments; TCP on the sender will just retransmit it 1459 * later. Setting the buffer size too big may only consume too 1460 * much kernel memory if the application doesn't read() from 1461 * the socket or packet loss or reordering makes use of the 1462 * reassembly queue. 1463 * 1464 * The criteria to step up the receive buffer one notch are: 1465 * 1. Application has not set receive buffer size with 1466 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1467 * 2. the number of bytes received during the time it takes 1468 * one timestamp to be reflected back to us (the RTT); 1469 * 3. received bytes per RTT is within seven eighth of the 1470 * current socket buffer size; 1471 * 4. receive buffer size has not hit maximal automatic size; 1472 * 1473 * This algorithm does one step per RTT at most and only if 1474 * we receive a bulk stream w/o packet losses or reorderings. 1475 * Shrinking the buffer during idle times is not necessary as 1476 * it doesn't consume any memory when idle. 1477 * 1478 * TODO: Only step up if the application is actually serving 1479 * the buffer to better manage the socket buffer resources. 1480 */ 1481 int 1482 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1483 struct tcpcb *tp, int tlen) 1484 { 1485 int newsize = 0; 1486 1487 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1488 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1489 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1490 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1491 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1492 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1493 newsize = min(so->so_rcv.sb_hiwat + 1494 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1495 } 1496 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1497 1498 /* Start over with next RTT. */ 1499 tp->rfbuf_ts = 0; 1500 tp->rfbuf_cnt = 0; 1501 } else { 1502 tp->rfbuf_cnt += tlen; /* add up */ 1503 } 1504 1505 return (newsize); 1506 } 1507 1508 void 1509 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1510 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1511 { 1512 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1513 int rstreason, todrop, win; 1514 uint32_t tiwin; 1515 uint16_t nsegs; 1516 char *s; 1517 struct in_conninfo *inc; 1518 struct mbuf *mfree; 1519 struct tcpopt to; 1520 int tfo_syn; 1521 1522 #ifdef TCPDEBUG 1523 /* 1524 * The size of tcp_saveipgen must be the size of the max ip header, 1525 * now IPv6. 1526 */ 1527 u_char tcp_saveipgen[IP6_HDR_LEN]; 1528 struct tcphdr tcp_savetcp; 1529 short ostate = 0; 1530 #endif 1531 thflags = th->th_flags; 1532 inc = &tp->t_inpcb->inp_inc; 1533 tp->sackhint.last_sack_ack = 0; 1534 sack_changed = 0; 1535 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1536 /* 1537 * If this is either a state-changing packet or current state isn't 1538 * established, we require a write lock on tcbinfo. Otherwise, we 1539 * allow the tcbinfo to be in either alocked or unlocked, as the 1540 * caller may have unnecessarily acquired a write lock due to a race. 1541 */ 1542 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1543 tp->t_state != TCPS_ESTABLISHED) { 1544 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1545 } 1546 INP_WLOCK_ASSERT(tp->t_inpcb); 1547 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1548 __func__)); 1549 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1550 __func__)); 1551 1552 #ifdef TCPPCAP 1553 /* Save segment, if requested. */ 1554 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1555 #endif 1556 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1557 tlen, NULL, true); 1558 1559 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1560 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1561 log(LOG_DEBUG, "%s; %s: " 1562 "SYN|FIN segment ignored (based on " 1563 "sysctl setting)\n", s, __func__); 1564 free(s, M_TCPLOG); 1565 } 1566 goto drop; 1567 } 1568 1569 /* 1570 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1571 * check SEQ.ACK first. 1572 */ 1573 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1574 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1575 rstreason = BANDLIM_UNLIMITED; 1576 goto dropwithreset; 1577 } 1578 1579 /* 1580 * Segment received on connection. 1581 * Reset idle time and keep-alive timer. 1582 * XXX: This should be done after segment 1583 * validation to ignore broken/spoofed segs. 1584 */ 1585 tp->t_rcvtime = ticks; 1586 1587 /* 1588 * Scale up the window into a 32-bit value. 1589 * For the SYN_SENT state the scale is zero. 1590 */ 1591 tiwin = th->th_win << tp->snd_scale; 1592 1593 /* 1594 * TCP ECN processing. 1595 */ 1596 if (tp->t_flags & TF_ECN_PERMIT) { 1597 if (thflags & TH_CWR) 1598 tp->t_flags &= ~TF_ECN_SND_ECE; 1599 switch (iptos & IPTOS_ECN_MASK) { 1600 case IPTOS_ECN_CE: 1601 tp->t_flags |= TF_ECN_SND_ECE; 1602 TCPSTAT_INC(tcps_ecn_ce); 1603 break; 1604 case IPTOS_ECN_ECT0: 1605 TCPSTAT_INC(tcps_ecn_ect0); 1606 break; 1607 case IPTOS_ECN_ECT1: 1608 TCPSTAT_INC(tcps_ecn_ect1); 1609 break; 1610 } 1611 1612 /* Process a packet differently from RFC3168. */ 1613 cc_ecnpkt_handler(tp, th, iptos); 1614 1615 /* Congestion experienced. */ 1616 if (thflags & TH_ECE) { 1617 cc_cong_signal(tp, th, CC_ECN); 1618 } 1619 } 1620 1621 /* 1622 * Parse options on any incoming segment. 1623 */ 1624 tcp_dooptions(&to, (u_char *)(th + 1), 1625 (th->th_off << 2) - sizeof(struct tcphdr), 1626 (thflags & TH_SYN) ? TO_SYN : 0); 1627 1628 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1629 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1630 (to.to_flags & TOF_SIGNATURE) == 0) { 1631 TCPSTAT_INC(tcps_sig_err_sigopt); 1632 /* XXX: should drop? */ 1633 } 1634 #endif 1635 /* 1636 * If echoed timestamp is later than the current time, 1637 * fall back to non RFC1323 RTT calculation. Normalize 1638 * timestamp if syncookies were used when this connection 1639 * was established. 1640 */ 1641 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1642 to.to_tsecr -= tp->ts_offset; 1643 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1644 to.to_tsecr = 0; 1645 else if (tp->t_flags & TF_PREVVALID && 1646 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1647 cc_cong_signal(tp, th, CC_RTO_ERR); 1648 } 1649 /* 1650 * Process options only when we get SYN/ACK back. The SYN case 1651 * for incoming connections is handled in tcp_syncache. 1652 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1653 * or <SYN,ACK>) segment itself is never scaled. 1654 * XXX this is traditional behavior, may need to be cleaned up. 1655 */ 1656 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1657 if ((to.to_flags & TOF_SCALE) && 1658 (tp->t_flags & TF_REQ_SCALE)) { 1659 tp->t_flags |= TF_RCVD_SCALE; 1660 tp->snd_scale = to.to_wscale; 1661 } 1662 /* 1663 * Initial send window. It will be updated with 1664 * the next incoming segment to the scaled value. 1665 */ 1666 tp->snd_wnd = th->th_win; 1667 if (to.to_flags & TOF_TS) { 1668 tp->t_flags |= TF_RCVD_TSTMP; 1669 tp->ts_recent = to.to_tsval; 1670 tp->ts_recent_age = tcp_ts_getticks(); 1671 } 1672 if (to.to_flags & TOF_MSS) 1673 tcp_mss(tp, to.to_mss); 1674 if ((tp->t_flags & TF_SACK_PERMIT) && 1675 (to.to_flags & TOF_SACKPERM) == 0) 1676 tp->t_flags &= ~TF_SACK_PERMIT; 1677 if (IS_FASTOPEN(tp->t_flags)) { 1678 if (to.to_flags & TOF_FASTOPEN) { 1679 uint16_t mss; 1680 1681 if (to.to_flags & TOF_MSS) 1682 mss = to.to_mss; 1683 else 1684 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1685 mss = TCP6_MSS; 1686 else 1687 mss = TCP_MSS; 1688 tcp_fastopen_update_cache(tp, mss, 1689 to.to_tfo_len, to.to_tfo_cookie); 1690 } else 1691 tcp_fastopen_disable_path(tp); 1692 } 1693 } 1694 1695 /* 1696 * If timestamps were negotiated during SYN/ACK they should 1697 * appear on every segment during this session and vice versa. 1698 */ 1699 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1700 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1701 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1702 "no action\n", s, __func__); 1703 free(s, M_TCPLOG); 1704 } 1705 } 1706 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1707 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1708 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1709 "no action\n", s, __func__); 1710 free(s, M_TCPLOG); 1711 } 1712 } 1713 1714 /* 1715 * Header prediction: check for the two common cases 1716 * of a uni-directional data xfer. If the packet has 1717 * no control flags, is in-sequence, the window didn't 1718 * change and we're not retransmitting, it's a 1719 * candidate. If the length is zero and the ack moved 1720 * forward, we're the sender side of the xfer. Just 1721 * free the data acked & wake any higher level process 1722 * that was blocked waiting for space. If the length 1723 * is non-zero and the ack didn't move, we're the 1724 * receiver side. If we're getting packets in-order 1725 * (the reassembly queue is empty), add the data to 1726 * the socket buffer and note that we need a delayed ack. 1727 * Make sure that the hidden state-flags are also off. 1728 * Since we check for TCPS_ESTABLISHED first, it can only 1729 * be TH_NEEDSYN. 1730 */ 1731 if (tp->t_state == TCPS_ESTABLISHED && 1732 th->th_seq == tp->rcv_nxt && 1733 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1734 tp->snd_nxt == tp->snd_max && 1735 tiwin && tiwin == tp->snd_wnd && 1736 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1737 LIST_EMPTY(&tp->t_segq) && 1738 ((to.to_flags & TOF_TS) == 0 || 1739 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1740 1741 /* 1742 * If last ACK falls within this segment's sequence numbers, 1743 * record the timestamp. 1744 * NOTE that the test is modified according to the latest 1745 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1746 */ 1747 if ((to.to_flags & TOF_TS) != 0 && 1748 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1749 tp->ts_recent_age = tcp_ts_getticks(); 1750 tp->ts_recent = to.to_tsval; 1751 } 1752 1753 if (tlen == 0) { 1754 if (SEQ_GT(th->th_ack, tp->snd_una) && 1755 SEQ_LEQ(th->th_ack, tp->snd_max) && 1756 !IN_RECOVERY(tp->t_flags) && 1757 (to.to_flags & TOF_SACK) == 0 && 1758 TAILQ_EMPTY(&tp->snd_holes)) { 1759 /* 1760 * This is a pure ack for outstanding data. 1761 */ 1762 TCPSTAT_INC(tcps_predack); 1763 1764 /* 1765 * "bad retransmit" recovery without timestamps. 1766 */ 1767 if ((to.to_flags & TOF_TS) == 0 && 1768 tp->t_rxtshift == 1 && 1769 tp->t_flags & TF_PREVVALID && 1770 (int)(ticks - tp->t_badrxtwin) < 0) { 1771 cc_cong_signal(tp, th, CC_RTO_ERR); 1772 } 1773 1774 /* 1775 * Recalculate the transmit timer / rtt. 1776 * 1777 * Some boxes send broken timestamp replies 1778 * during the SYN+ACK phase, ignore 1779 * timestamps of 0 or we could calculate a 1780 * huge RTT and blow up the retransmit timer. 1781 */ 1782 if ((to.to_flags & TOF_TS) != 0 && 1783 to.to_tsecr) { 1784 uint32_t t; 1785 1786 t = tcp_ts_getticks() - to.to_tsecr; 1787 if (!tp->t_rttlow || tp->t_rttlow > t) 1788 tp->t_rttlow = t; 1789 tcp_xmit_timer(tp, 1790 TCP_TS_TO_TICKS(t) + 1); 1791 } else if (tp->t_rtttime && 1792 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1793 if (!tp->t_rttlow || 1794 tp->t_rttlow > ticks - tp->t_rtttime) 1795 tp->t_rttlow = ticks - tp->t_rtttime; 1796 tcp_xmit_timer(tp, 1797 ticks - tp->t_rtttime); 1798 } 1799 acked = BYTES_THIS_ACK(tp, th); 1800 1801 #ifdef TCP_HHOOK 1802 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1803 hhook_run_tcp_est_in(tp, th, &to); 1804 #endif 1805 1806 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1807 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1808 sbdrop(&so->so_snd, acked); 1809 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1810 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1811 tp->snd_recover = th->th_ack - 1; 1812 1813 /* 1814 * Let the congestion control algorithm update 1815 * congestion control related information. This 1816 * typically means increasing the congestion 1817 * window. 1818 */ 1819 cc_ack_received(tp, th, nsegs, CC_ACK); 1820 1821 tp->snd_una = th->th_ack; 1822 /* 1823 * Pull snd_wl2 up to prevent seq wrap relative 1824 * to th_ack. 1825 */ 1826 tp->snd_wl2 = th->th_ack; 1827 tp->t_dupacks = 0; 1828 m_freem(m); 1829 1830 /* 1831 * If all outstanding data are acked, stop 1832 * retransmit timer, otherwise restart timer 1833 * using current (possibly backed-off) value. 1834 * If process is waiting for space, 1835 * wakeup/selwakeup/signal. If data 1836 * are ready to send, let tcp_output 1837 * decide between more output or persist. 1838 */ 1839 #ifdef TCPDEBUG 1840 if (so->so_options & SO_DEBUG) 1841 tcp_trace(TA_INPUT, ostate, tp, 1842 (void *)tcp_saveipgen, 1843 &tcp_savetcp, 0); 1844 #endif 1845 TCP_PROBE3(debug__input, tp, th, m); 1846 if (tp->snd_una == tp->snd_max) 1847 tcp_timer_activate(tp, TT_REXMT, 0); 1848 else if (!tcp_timer_active(tp, TT_PERSIST)) 1849 tcp_timer_activate(tp, TT_REXMT, 1850 tp->t_rxtcur); 1851 sowwakeup(so); 1852 if (sbavail(&so->so_snd)) 1853 (void) tp->t_fb->tfb_tcp_output(tp); 1854 goto check_delack; 1855 } 1856 } else if (th->th_ack == tp->snd_una && 1857 tlen <= sbspace(&so->so_rcv)) { 1858 int newsize = 0; /* automatic sockbuf scaling */ 1859 1860 /* 1861 * This is a pure, in-sequence data packet with 1862 * nothing on the reassembly queue and we have enough 1863 * buffer space to take it. 1864 */ 1865 /* Clean receiver SACK report if present */ 1866 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1867 tcp_clean_sackreport(tp); 1868 TCPSTAT_INC(tcps_preddat); 1869 tp->rcv_nxt += tlen; 1870 /* 1871 * Pull snd_wl1 up to prevent seq wrap relative to 1872 * th_seq. 1873 */ 1874 tp->snd_wl1 = th->th_seq; 1875 /* 1876 * Pull rcv_up up to prevent seq wrap relative to 1877 * rcv_nxt. 1878 */ 1879 tp->rcv_up = tp->rcv_nxt; 1880 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1881 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1882 #ifdef TCPDEBUG 1883 if (so->so_options & SO_DEBUG) 1884 tcp_trace(TA_INPUT, ostate, tp, 1885 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1886 #endif 1887 TCP_PROBE3(debug__input, tp, th, m); 1888 1889 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1890 1891 /* Add data to socket buffer. */ 1892 SOCKBUF_LOCK(&so->so_rcv); 1893 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1894 m_freem(m); 1895 } else { 1896 /* 1897 * Set new socket buffer size. 1898 * Give up when limit is reached. 1899 */ 1900 if (newsize) 1901 if (!sbreserve_locked(&so->so_rcv, 1902 newsize, so, NULL)) 1903 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1904 m_adj(m, drop_hdrlen); /* delayed header drop */ 1905 sbappendstream_locked(&so->so_rcv, m, 0); 1906 } 1907 /* NB: sorwakeup_locked() does an implicit unlock. */ 1908 sorwakeup_locked(so); 1909 if (DELAY_ACK(tp, tlen)) { 1910 tp->t_flags |= TF_DELACK; 1911 } else { 1912 tp->t_flags |= TF_ACKNOW; 1913 tp->t_fb->tfb_tcp_output(tp); 1914 } 1915 goto check_delack; 1916 } 1917 } 1918 1919 /* 1920 * Calculate amount of space in receive window, 1921 * and then do TCP input processing. 1922 * Receive window is amount of space in rcv queue, 1923 * but not less than advertised window. 1924 */ 1925 win = sbspace(&so->so_rcv); 1926 if (win < 0) 1927 win = 0; 1928 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1929 1930 switch (tp->t_state) { 1931 1932 /* 1933 * If the state is SYN_RECEIVED: 1934 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1935 */ 1936 case TCPS_SYN_RECEIVED: 1937 if ((thflags & TH_ACK) && 1938 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1939 SEQ_GT(th->th_ack, tp->snd_max))) { 1940 rstreason = BANDLIM_RST_OPENPORT; 1941 goto dropwithreset; 1942 } 1943 if (IS_FASTOPEN(tp->t_flags)) { 1944 /* 1945 * When a TFO connection is in SYN_RECEIVED, the 1946 * only valid packets are the initial SYN, a 1947 * retransmit/copy of the initial SYN (possibly with 1948 * a subset of the original data), a valid ACK, a 1949 * FIN, or a RST. 1950 */ 1951 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1952 rstreason = BANDLIM_RST_OPENPORT; 1953 goto dropwithreset; 1954 } else if (thflags & TH_SYN) { 1955 /* non-initial SYN is ignored */ 1956 if ((tcp_timer_active(tp, TT_DELACK) || 1957 tcp_timer_active(tp, TT_REXMT))) 1958 goto drop; 1959 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1960 goto drop; 1961 } 1962 } 1963 break; 1964 1965 /* 1966 * If the state is SYN_SENT: 1967 * if seg contains a RST with valid ACK (SEQ.ACK has already 1968 * been verified), then drop the connection. 1969 * if seg contains a RST without an ACK, drop the seg. 1970 * if seg does not contain SYN, then drop the seg. 1971 * Otherwise this is an acceptable SYN segment 1972 * initialize tp->rcv_nxt and tp->irs 1973 * if seg contains ack then advance tp->snd_una 1974 * if seg contains an ECE and ECN support is enabled, the stream 1975 * is ECN capable. 1976 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1977 * arrange for segment to be acked (eventually) 1978 * continue processing rest of data/controls, beginning with URG 1979 */ 1980 case TCPS_SYN_SENT: 1981 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1982 TCP_PROBE5(connect__refused, NULL, tp, 1983 m, tp, th); 1984 tp = tcp_drop(tp, ECONNREFUSED); 1985 } 1986 if (thflags & TH_RST) 1987 goto drop; 1988 if (!(thflags & TH_SYN)) 1989 goto drop; 1990 1991 tp->irs = th->th_seq; 1992 tcp_rcvseqinit(tp); 1993 if (thflags & TH_ACK) { 1994 int tfo_partial_ack = 0; 1995 1996 TCPSTAT_INC(tcps_connects); 1997 soisconnected(so); 1998 #ifdef MAC 1999 mac_socketpeer_set_from_mbuf(m, so); 2000 #endif 2001 /* Do window scaling on this connection? */ 2002 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2003 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2004 tp->rcv_scale = tp->request_r_scale; 2005 } 2006 tp->rcv_adv += min(tp->rcv_wnd, 2007 TCP_MAXWIN << tp->rcv_scale); 2008 tp->snd_una++; /* SYN is acked */ 2009 /* 2010 * If not all the data that was sent in the TFO SYN 2011 * has been acked, resend the remainder right away. 2012 */ 2013 if (IS_FASTOPEN(tp->t_flags) && 2014 (tp->snd_una != tp->snd_max)) { 2015 tp->snd_nxt = th->th_ack; 2016 tfo_partial_ack = 1; 2017 } 2018 /* 2019 * If there's data, delay ACK; if there's also a FIN 2020 * ACKNOW will be turned on later. 2021 */ 2022 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2023 tcp_timer_activate(tp, TT_DELACK, 2024 tcp_delacktime); 2025 else 2026 tp->t_flags |= TF_ACKNOW; 2027 2028 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2029 tp->t_flags |= TF_ECN_PERMIT; 2030 TCPSTAT_INC(tcps_ecn_shs); 2031 } 2032 2033 /* 2034 * Received <SYN,ACK> in SYN_SENT[*] state. 2035 * Transitions: 2036 * SYN_SENT --> ESTABLISHED 2037 * SYN_SENT* --> FIN_WAIT_1 2038 */ 2039 tp->t_starttime = ticks; 2040 if (tp->t_flags & TF_NEEDFIN) { 2041 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2042 tp->t_flags &= ~TF_NEEDFIN; 2043 thflags &= ~TH_SYN; 2044 } else { 2045 tcp_state_change(tp, TCPS_ESTABLISHED); 2046 TCP_PROBE5(connect__established, NULL, tp, 2047 m, tp, th); 2048 cc_conn_init(tp); 2049 tcp_timer_activate(tp, TT_KEEP, 2050 TP_KEEPIDLE(tp)); 2051 } 2052 } else { 2053 /* 2054 * Received initial SYN in SYN-SENT[*] state => 2055 * simultaneous open. 2056 * If it succeeds, connection is * half-synchronized. 2057 * Otherwise, do 3-way handshake: 2058 * SYN-SENT -> SYN-RECEIVED 2059 * SYN-SENT* -> SYN-RECEIVED* 2060 */ 2061 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2062 tcp_timer_activate(tp, TT_REXMT, 0); 2063 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2064 } 2065 2066 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2067 INP_WLOCK_ASSERT(tp->t_inpcb); 2068 2069 /* 2070 * Advance th->th_seq to correspond to first data byte. 2071 * If data, trim to stay within window, 2072 * dropping FIN if necessary. 2073 */ 2074 th->th_seq++; 2075 if (tlen > tp->rcv_wnd) { 2076 todrop = tlen - tp->rcv_wnd; 2077 m_adj(m, -todrop); 2078 tlen = tp->rcv_wnd; 2079 thflags &= ~TH_FIN; 2080 TCPSTAT_INC(tcps_rcvpackafterwin); 2081 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2082 } 2083 tp->snd_wl1 = th->th_seq - 1; 2084 tp->rcv_up = th->th_seq; 2085 /* 2086 * Client side of transaction: already sent SYN and data. 2087 * If the remote host used T/TCP to validate the SYN, 2088 * our data will be ACK'd; if so, enter normal data segment 2089 * processing in the middle of step 5, ack processing. 2090 * Otherwise, goto step 6. 2091 */ 2092 if (thflags & TH_ACK) 2093 goto process_ACK; 2094 2095 goto step6; 2096 2097 /* 2098 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2099 * do normal processing. 2100 * 2101 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2102 */ 2103 case TCPS_LAST_ACK: 2104 case TCPS_CLOSING: 2105 break; /* continue normal processing */ 2106 } 2107 2108 /* 2109 * States other than LISTEN or SYN_SENT. 2110 * First check the RST flag and sequence number since reset segments 2111 * are exempt from the timestamp and connection count tests. This 2112 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2113 * below which allowed reset segments in half the sequence space 2114 * to fall though and be processed (which gives forged reset 2115 * segments with a random sequence number a 50 percent chance of 2116 * killing a connection). 2117 * Then check timestamp, if present. 2118 * Then check the connection count, if present. 2119 * Then check that at least some bytes of segment are within 2120 * receive window. If segment begins before rcv_nxt, 2121 * drop leading data (and SYN); if nothing left, just ack. 2122 */ 2123 if (thflags & TH_RST) { 2124 /* 2125 * RFC5961 Section 3.2 2126 * 2127 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2128 * - If RST is in window, we send challenge ACK. 2129 * 2130 * Note: to take into account delayed ACKs, we should 2131 * test against last_ack_sent instead of rcv_nxt. 2132 * Note 2: we handle special case of closed window, not 2133 * covered by the RFC. 2134 */ 2135 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2136 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2137 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2138 2139 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2140 KASSERT(tp->t_state != TCPS_SYN_SENT, 2141 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2142 __func__, th, tp)); 2143 2144 if (V_tcp_insecure_rst || 2145 tp->last_ack_sent == th->th_seq) { 2146 TCPSTAT_INC(tcps_drops); 2147 /* Drop the connection. */ 2148 switch (tp->t_state) { 2149 case TCPS_SYN_RECEIVED: 2150 so->so_error = ECONNREFUSED; 2151 goto close; 2152 case TCPS_ESTABLISHED: 2153 case TCPS_FIN_WAIT_1: 2154 case TCPS_FIN_WAIT_2: 2155 case TCPS_CLOSE_WAIT: 2156 case TCPS_CLOSING: 2157 case TCPS_LAST_ACK: 2158 so->so_error = ECONNRESET; 2159 close: 2160 /* FALLTHROUGH */ 2161 default: 2162 tp = tcp_close(tp); 2163 } 2164 } else { 2165 TCPSTAT_INC(tcps_badrst); 2166 /* Send challenge ACK. */ 2167 tcp_respond(tp, mtod(m, void *), th, m, 2168 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2169 tp->last_ack_sent = tp->rcv_nxt; 2170 m = NULL; 2171 } 2172 } 2173 goto drop; 2174 } 2175 2176 /* 2177 * RFC5961 Section 4.2 2178 * Send challenge ACK for any SYN in synchronized state. 2179 */ 2180 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2181 tp->t_state != TCPS_SYN_RECEIVED) { 2182 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2183 2184 TCPSTAT_INC(tcps_badsyn); 2185 if (V_tcp_insecure_syn && 2186 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2187 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2188 tp = tcp_drop(tp, ECONNRESET); 2189 rstreason = BANDLIM_UNLIMITED; 2190 } else { 2191 /* Send challenge ACK. */ 2192 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2193 tp->snd_nxt, TH_ACK); 2194 tp->last_ack_sent = tp->rcv_nxt; 2195 m = NULL; 2196 } 2197 goto drop; 2198 } 2199 2200 /* 2201 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2202 * and it's less than ts_recent, drop it. 2203 */ 2204 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2205 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2206 2207 /* Check to see if ts_recent is over 24 days old. */ 2208 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2209 /* 2210 * Invalidate ts_recent. If this segment updates 2211 * ts_recent, the age will be reset later and ts_recent 2212 * will get a valid value. If it does not, setting 2213 * ts_recent to zero will at least satisfy the 2214 * requirement that zero be placed in the timestamp 2215 * echo reply when ts_recent isn't valid. The 2216 * age isn't reset until we get a valid ts_recent 2217 * because we don't want out-of-order segments to be 2218 * dropped when ts_recent is old. 2219 */ 2220 tp->ts_recent = 0; 2221 } else { 2222 TCPSTAT_INC(tcps_rcvduppack); 2223 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2224 TCPSTAT_INC(tcps_pawsdrop); 2225 if (tlen) 2226 goto dropafterack; 2227 goto drop; 2228 } 2229 } 2230 2231 /* 2232 * In the SYN-RECEIVED state, validate that the packet belongs to 2233 * this connection before trimming the data to fit the receive 2234 * window. Check the sequence number versus IRS since we know 2235 * the sequence numbers haven't wrapped. This is a partial fix 2236 * for the "LAND" DoS attack. 2237 */ 2238 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2239 rstreason = BANDLIM_RST_OPENPORT; 2240 goto dropwithreset; 2241 } 2242 2243 todrop = tp->rcv_nxt - th->th_seq; 2244 if (todrop > 0) { 2245 if (thflags & TH_SYN) { 2246 thflags &= ~TH_SYN; 2247 th->th_seq++; 2248 if (th->th_urp > 1) 2249 th->th_urp--; 2250 else 2251 thflags &= ~TH_URG; 2252 todrop--; 2253 } 2254 /* 2255 * Following if statement from Stevens, vol. 2, p. 960. 2256 */ 2257 if (todrop > tlen 2258 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2259 /* 2260 * Any valid FIN must be to the left of the window. 2261 * At this point the FIN must be a duplicate or out 2262 * of sequence; drop it. 2263 */ 2264 thflags &= ~TH_FIN; 2265 2266 /* 2267 * Send an ACK to resynchronize and drop any data. 2268 * But keep on processing for RST or ACK. 2269 */ 2270 tp->t_flags |= TF_ACKNOW; 2271 todrop = tlen; 2272 TCPSTAT_INC(tcps_rcvduppack); 2273 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2274 } else { 2275 TCPSTAT_INC(tcps_rcvpartduppack); 2276 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2277 } 2278 drop_hdrlen += todrop; /* drop from the top afterwards */ 2279 th->th_seq += todrop; 2280 tlen -= todrop; 2281 if (th->th_urp > todrop) 2282 th->th_urp -= todrop; 2283 else { 2284 thflags &= ~TH_URG; 2285 th->th_urp = 0; 2286 } 2287 } 2288 2289 /* 2290 * If new data are received on a connection after the 2291 * user processes are gone, then RST the other end. 2292 */ 2293 if ((so->so_state & SS_NOFDREF) && 2294 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2295 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2296 2297 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2298 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2299 "after socket was closed, " 2300 "sending RST and removing tcpcb\n", 2301 s, __func__, tcpstates[tp->t_state], tlen); 2302 free(s, M_TCPLOG); 2303 } 2304 tp = tcp_close(tp); 2305 TCPSTAT_INC(tcps_rcvafterclose); 2306 rstreason = BANDLIM_UNLIMITED; 2307 goto dropwithreset; 2308 } 2309 2310 /* 2311 * If segment ends after window, drop trailing data 2312 * (and PUSH and FIN); if nothing left, just ACK. 2313 */ 2314 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2315 if (todrop > 0) { 2316 TCPSTAT_INC(tcps_rcvpackafterwin); 2317 if (todrop >= tlen) { 2318 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2319 /* 2320 * If window is closed can only take segments at 2321 * window edge, and have to drop data and PUSH from 2322 * incoming segments. Continue processing, but 2323 * remember to ack. Otherwise, drop segment 2324 * and ack. 2325 */ 2326 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2327 tp->t_flags |= TF_ACKNOW; 2328 TCPSTAT_INC(tcps_rcvwinprobe); 2329 } else 2330 goto dropafterack; 2331 } else 2332 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2333 m_adj(m, -todrop); 2334 tlen -= todrop; 2335 thflags &= ~(TH_PUSH|TH_FIN); 2336 } 2337 2338 /* 2339 * If last ACK falls within this segment's sequence numbers, 2340 * record its timestamp. 2341 * NOTE: 2342 * 1) That the test incorporates suggestions from the latest 2343 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2344 * 2) That updating only on newer timestamps interferes with 2345 * our earlier PAWS tests, so this check should be solely 2346 * predicated on the sequence space of this segment. 2347 * 3) That we modify the segment boundary check to be 2348 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2349 * instead of RFC1323's 2350 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2351 * This modified check allows us to overcome RFC1323's 2352 * limitations as described in Stevens TCP/IP Illustrated 2353 * Vol. 2 p.869. In such cases, we can still calculate the 2354 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2355 */ 2356 if ((to.to_flags & TOF_TS) != 0 && 2357 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2358 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2359 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2360 tp->ts_recent_age = tcp_ts_getticks(); 2361 tp->ts_recent = to.to_tsval; 2362 } 2363 2364 /* 2365 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2366 * flag is on (half-synchronized state), then queue data for 2367 * later processing; else drop segment and return. 2368 */ 2369 if ((thflags & TH_ACK) == 0) { 2370 if (tp->t_state == TCPS_SYN_RECEIVED || 2371 (tp->t_flags & TF_NEEDSYN)) { 2372 if (tp->t_state == TCPS_SYN_RECEIVED && 2373 IS_FASTOPEN(tp->t_flags)) { 2374 tp->snd_wnd = tiwin; 2375 cc_conn_init(tp); 2376 } 2377 goto step6; 2378 } else if (tp->t_flags & TF_ACKNOW) 2379 goto dropafterack; 2380 else 2381 goto drop; 2382 } 2383 2384 /* 2385 * Ack processing. 2386 */ 2387 switch (tp->t_state) { 2388 2389 /* 2390 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2391 * ESTABLISHED state and continue processing. 2392 * The ACK was checked above. 2393 */ 2394 case TCPS_SYN_RECEIVED: 2395 2396 TCPSTAT_INC(tcps_connects); 2397 soisconnected(so); 2398 /* Do window scaling? */ 2399 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2400 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2401 tp->rcv_scale = tp->request_r_scale; 2402 tp->snd_wnd = tiwin; 2403 } 2404 /* 2405 * Make transitions: 2406 * SYN-RECEIVED -> ESTABLISHED 2407 * SYN-RECEIVED* -> FIN-WAIT-1 2408 */ 2409 tp->t_starttime = ticks; 2410 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2411 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2412 tp->t_tfo_pending = NULL; 2413 2414 /* 2415 * Account for the ACK of our SYN prior to 2416 * regular ACK processing below. 2417 */ 2418 tp->snd_una++; 2419 } 2420 if (tp->t_flags & TF_NEEDFIN) { 2421 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2422 tp->t_flags &= ~TF_NEEDFIN; 2423 } else { 2424 tcp_state_change(tp, TCPS_ESTABLISHED); 2425 TCP_PROBE5(accept__established, NULL, tp, 2426 m, tp, th); 2427 /* 2428 * TFO connections call cc_conn_init() during SYN 2429 * processing. Calling it again here for such 2430 * connections is not harmless as it would undo the 2431 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2432 * is retransmitted. 2433 */ 2434 if (!IS_FASTOPEN(tp->t_flags)) 2435 cc_conn_init(tp); 2436 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2437 } 2438 /* 2439 * If segment contains data or ACK, will call tcp_reass() 2440 * later; if not, do so now to pass queued data to user. 2441 */ 2442 if (tlen == 0 && (thflags & TH_FIN) == 0) 2443 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2444 (struct mbuf *)0); 2445 tp->snd_wl1 = th->th_seq - 1; 2446 /* FALLTHROUGH */ 2447 2448 /* 2449 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2450 * ACKs. If the ack is in the range 2451 * tp->snd_una < th->th_ack <= tp->snd_max 2452 * then advance tp->snd_una to th->th_ack and drop 2453 * data from the retransmission queue. If this ACK reflects 2454 * more up to date window information we update our window information. 2455 */ 2456 case TCPS_ESTABLISHED: 2457 case TCPS_FIN_WAIT_1: 2458 case TCPS_FIN_WAIT_2: 2459 case TCPS_CLOSE_WAIT: 2460 case TCPS_CLOSING: 2461 case TCPS_LAST_ACK: 2462 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2463 TCPSTAT_INC(tcps_rcvacktoomuch); 2464 goto dropafterack; 2465 } 2466 if ((tp->t_flags & TF_SACK_PERMIT) && 2467 ((to.to_flags & TOF_SACK) || 2468 !TAILQ_EMPTY(&tp->snd_holes))) 2469 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2470 else 2471 /* 2472 * Reset the value so that previous (valid) value 2473 * from the last ack with SACK doesn't get used. 2474 */ 2475 tp->sackhint.sacked_bytes = 0; 2476 2477 #ifdef TCP_HHOOK 2478 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2479 hhook_run_tcp_est_in(tp, th, &to); 2480 #endif 2481 2482 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2483 u_int maxseg; 2484 2485 maxseg = tcp_maxseg(tp); 2486 if (tlen == 0 && 2487 (tiwin == tp->snd_wnd || 2488 (tp->t_flags & TF_SACK_PERMIT))) { 2489 /* 2490 * If this is the first time we've seen a 2491 * FIN from the remote, this is not a 2492 * duplicate and it needs to be processed 2493 * normally. This happens during a 2494 * simultaneous close. 2495 */ 2496 if ((thflags & TH_FIN) && 2497 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2498 tp->t_dupacks = 0; 2499 break; 2500 } 2501 TCPSTAT_INC(tcps_rcvdupack); 2502 /* 2503 * If we have outstanding data (other than 2504 * a window probe), this is a completely 2505 * duplicate ack (ie, window info didn't 2506 * change and FIN isn't set), 2507 * the ack is the biggest we've 2508 * seen and we've seen exactly our rexmt 2509 * threshold of them, assume a packet 2510 * has been dropped and retransmit it. 2511 * Kludge snd_nxt & the congestion 2512 * window so we send only this one 2513 * packet. 2514 * 2515 * We know we're losing at the current 2516 * window size so do congestion avoidance 2517 * (set ssthresh to half the current window 2518 * and pull our congestion window back to 2519 * the new ssthresh). 2520 * 2521 * Dup acks mean that packets have left the 2522 * network (they're now cached at the receiver) 2523 * so bump cwnd by the amount in the receiver 2524 * to keep a constant cwnd packets in the 2525 * network. 2526 * 2527 * When using TCP ECN, notify the peer that 2528 * we reduced the cwnd. 2529 */ 2530 /* 2531 * Following 2 kinds of acks should not affect 2532 * dupack counting: 2533 * 1) Old acks 2534 * 2) Acks with SACK but without any new SACK 2535 * information in them. These could result from 2536 * any anomaly in the network like a switch 2537 * duplicating packets or a possible DoS attack. 2538 */ 2539 if (th->th_ack != tp->snd_una || 2540 ((tp->t_flags & TF_SACK_PERMIT) && 2541 !sack_changed)) 2542 break; 2543 else if (!tcp_timer_active(tp, TT_REXMT)) 2544 tp->t_dupacks = 0; 2545 else if (++tp->t_dupacks > tcprexmtthresh || 2546 IN_FASTRECOVERY(tp->t_flags)) { 2547 cc_ack_received(tp, th, nsegs, 2548 CC_DUPACK); 2549 if ((tp->t_flags & TF_SACK_PERMIT) && 2550 IN_FASTRECOVERY(tp->t_flags)) { 2551 int awnd; 2552 2553 /* 2554 * Compute the amount of data in flight first. 2555 * We can inject new data into the pipe iff 2556 * we have less than 1/2 the original window's 2557 * worth of data in flight. 2558 */ 2559 if (V_tcp_do_rfc6675_pipe) 2560 awnd = tcp_compute_pipe(tp); 2561 else 2562 awnd = (tp->snd_nxt - tp->snd_fack) + 2563 tp->sackhint.sack_bytes_rexmit; 2564 2565 if (awnd < tp->snd_ssthresh) { 2566 tp->snd_cwnd += maxseg; 2567 if (tp->snd_cwnd > tp->snd_ssthresh) 2568 tp->snd_cwnd = tp->snd_ssthresh; 2569 } 2570 } else 2571 tp->snd_cwnd += maxseg; 2572 (void) tp->t_fb->tfb_tcp_output(tp); 2573 goto drop; 2574 } else if (tp->t_dupacks == tcprexmtthresh) { 2575 tcp_seq onxt = tp->snd_nxt; 2576 2577 /* 2578 * If we're doing sack, check to 2579 * see if we're already in sack 2580 * recovery. If we're not doing sack, 2581 * check to see if we're in newreno 2582 * recovery. 2583 */ 2584 if (tp->t_flags & TF_SACK_PERMIT) { 2585 if (IN_FASTRECOVERY(tp->t_flags)) { 2586 tp->t_dupacks = 0; 2587 break; 2588 } 2589 } else { 2590 if (SEQ_LEQ(th->th_ack, 2591 tp->snd_recover)) { 2592 tp->t_dupacks = 0; 2593 break; 2594 } 2595 } 2596 /* Congestion signal before ack. */ 2597 cc_cong_signal(tp, th, CC_NDUPACK); 2598 cc_ack_received(tp, th, nsegs, 2599 CC_DUPACK); 2600 tcp_timer_activate(tp, TT_REXMT, 0); 2601 tp->t_rtttime = 0; 2602 if (tp->t_flags & TF_SACK_PERMIT) { 2603 TCPSTAT_INC( 2604 tcps_sack_recovery_episode); 2605 tp->sack_newdata = tp->snd_nxt; 2606 tp->snd_cwnd = maxseg; 2607 (void) tp->t_fb->tfb_tcp_output(tp); 2608 goto drop; 2609 } 2610 tp->snd_nxt = th->th_ack; 2611 tp->snd_cwnd = maxseg; 2612 (void) tp->t_fb->tfb_tcp_output(tp); 2613 KASSERT(tp->snd_limited <= 2, 2614 ("%s: tp->snd_limited too big", 2615 __func__)); 2616 tp->snd_cwnd = tp->snd_ssthresh + 2617 maxseg * 2618 (tp->t_dupacks - tp->snd_limited); 2619 if (SEQ_GT(onxt, tp->snd_nxt)) 2620 tp->snd_nxt = onxt; 2621 goto drop; 2622 } else if (V_tcp_do_rfc3042) { 2623 /* 2624 * Process first and second duplicate 2625 * ACKs. Each indicates a segment 2626 * leaving the network, creating room 2627 * for more. Make sure we can send a 2628 * packet on reception of each duplicate 2629 * ACK by increasing snd_cwnd by one 2630 * segment. Restore the original 2631 * snd_cwnd after packet transmission. 2632 */ 2633 cc_ack_received(tp, th, nsegs, 2634 CC_DUPACK); 2635 uint32_t oldcwnd = tp->snd_cwnd; 2636 tcp_seq oldsndmax = tp->snd_max; 2637 u_int sent; 2638 int avail; 2639 2640 KASSERT(tp->t_dupacks == 1 || 2641 tp->t_dupacks == 2, 2642 ("%s: dupacks not 1 or 2", 2643 __func__)); 2644 if (tp->t_dupacks == 1) 2645 tp->snd_limited = 0; 2646 tp->snd_cwnd = 2647 (tp->snd_nxt - tp->snd_una) + 2648 (tp->t_dupacks - tp->snd_limited) * 2649 maxseg; 2650 /* 2651 * Only call tcp_output when there 2652 * is new data available to be sent. 2653 * Otherwise we would send pure ACKs. 2654 */ 2655 SOCKBUF_LOCK(&so->so_snd); 2656 avail = sbavail(&so->so_snd) - 2657 (tp->snd_nxt - tp->snd_una); 2658 SOCKBUF_UNLOCK(&so->so_snd); 2659 if (avail > 0) 2660 (void) tp->t_fb->tfb_tcp_output(tp); 2661 sent = tp->snd_max - oldsndmax; 2662 if (sent > maxseg) { 2663 KASSERT((tp->t_dupacks == 2 && 2664 tp->snd_limited == 0) || 2665 (sent == maxseg + 1 && 2666 tp->t_flags & TF_SENTFIN), 2667 ("%s: sent too much", 2668 __func__)); 2669 tp->snd_limited = 2; 2670 } else if (sent > 0) 2671 ++tp->snd_limited; 2672 tp->snd_cwnd = oldcwnd; 2673 goto drop; 2674 } 2675 } 2676 break; 2677 } else { 2678 /* 2679 * This ack is advancing the left edge, reset the 2680 * counter. 2681 */ 2682 tp->t_dupacks = 0; 2683 /* 2684 * If this ack also has new SACK info, increment the 2685 * counter as per rfc6675. 2686 */ 2687 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2688 tp->t_dupacks++; 2689 } 2690 2691 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2692 ("%s: th_ack <= snd_una", __func__)); 2693 2694 /* 2695 * If the congestion window was inflated to account 2696 * for the other side's cached packets, retract it. 2697 */ 2698 if (IN_FASTRECOVERY(tp->t_flags)) { 2699 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2700 if (tp->t_flags & TF_SACK_PERMIT) 2701 tcp_sack_partialack(tp, th); 2702 else 2703 tcp_newreno_partial_ack(tp, th); 2704 } else 2705 cc_post_recovery(tp, th); 2706 } 2707 /* 2708 * If we reach this point, ACK is not a duplicate, 2709 * i.e., it ACKs something we sent. 2710 */ 2711 if (tp->t_flags & TF_NEEDSYN) { 2712 /* 2713 * T/TCP: Connection was half-synchronized, and our 2714 * SYN has been ACK'd (so connection is now fully 2715 * synchronized). Go to non-starred state, 2716 * increment snd_una for ACK of SYN, and check if 2717 * we can do window scaling. 2718 */ 2719 tp->t_flags &= ~TF_NEEDSYN; 2720 tp->snd_una++; 2721 /* Do window scaling? */ 2722 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2723 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2724 tp->rcv_scale = tp->request_r_scale; 2725 /* Send window already scaled. */ 2726 } 2727 } 2728 2729 process_ACK: 2730 INP_WLOCK_ASSERT(tp->t_inpcb); 2731 2732 acked = BYTES_THIS_ACK(tp, th); 2733 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2734 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2735 tp->snd_una, th->th_ack, tp, m)); 2736 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2737 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2738 2739 /* 2740 * If we just performed our first retransmit, and the ACK 2741 * arrives within our recovery window, then it was a mistake 2742 * to do the retransmit in the first place. Recover our 2743 * original cwnd and ssthresh, and proceed to transmit where 2744 * we left off. 2745 */ 2746 if (tp->t_rxtshift == 1 && 2747 tp->t_flags & TF_PREVVALID && 2748 tp->t_badrxtwin && 2749 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2750 cc_cong_signal(tp, th, CC_RTO_ERR); 2751 2752 /* 2753 * If we have a timestamp reply, update smoothed 2754 * round trip time. If no timestamp is present but 2755 * transmit timer is running and timed sequence 2756 * number was acked, update smoothed round trip time. 2757 * Since we now have an rtt measurement, cancel the 2758 * timer backoff (cf., Phil Karn's retransmit alg.). 2759 * Recompute the initial retransmit timer. 2760 * 2761 * Some boxes send broken timestamp replies 2762 * during the SYN+ACK phase, ignore 2763 * timestamps of 0 or we could calculate a 2764 * huge RTT and blow up the retransmit timer. 2765 */ 2766 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2767 uint32_t t; 2768 2769 t = tcp_ts_getticks() - to.to_tsecr; 2770 if (!tp->t_rttlow || tp->t_rttlow > t) 2771 tp->t_rttlow = t; 2772 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2773 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2774 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2775 tp->t_rttlow = ticks - tp->t_rtttime; 2776 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2777 } 2778 2779 /* 2780 * If all outstanding data is acked, stop retransmit 2781 * timer and remember to restart (more output or persist). 2782 * If there is more data to be acked, restart retransmit 2783 * timer, using current (possibly backed-off) value. 2784 */ 2785 if (th->th_ack == tp->snd_max) { 2786 tcp_timer_activate(tp, TT_REXMT, 0); 2787 needoutput = 1; 2788 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2789 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2790 2791 /* 2792 * If no data (only SYN) was ACK'd, 2793 * skip rest of ACK processing. 2794 */ 2795 if (acked == 0) 2796 goto step6; 2797 2798 /* 2799 * Let the congestion control algorithm update congestion 2800 * control related information. This typically means increasing 2801 * the congestion window. 2802 */ 2803 cc_ack_received(tp, th, nsegs, CC_ACK); 2804 2805 SOCKBUF_LOCK(&so->so_snd); 2806 if (acked > sbavail(&so->so_snd)) { 2807 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2808 tp->snd_wnd -= sbavail(&so->so_snd); 2809 else 2810 tp->snd_wnd = 0; 2811 mfree = sbcut_locked(&so->so_snd, 2812 (int)sbavail(&so->so_snd)); 2813 ourfinisacked = 1; 2814 } else { 2815 mfree = sbcut_locked(&so->so_snd, acked); 2816 if (tp->snd_wnd >= (uint32_t) acked) 2817 tp->snd_wnd -= acked; 2818 else 2819 tp->snd_wnd = 0; 2820 ourfinisacked = 0; 2821 } 2822 /* NB: sowwakeup_locked() does an implicit unlock. */ 2823 sowwakeup_locked(so); 2824 m_freem(mfree); 2825 /* Detect una wraparound. */ 2826 if (!IN_RECOVERY(tp->t_flags) && 2827 SEQ_GT(tp->snd_una, tp->snd_recover) && 2828 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2829 tp->snd_recover = th->th_ack - 1; 2830 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2831 if (IN_RECOVERY(tp->t_flags) && 2832 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2833 EXIT_RECOVERY(tp->t_flags); 2834 } 2835 tp->snd_una = th->th_ack; 2836 if (tp->t_flags & TF_SACK_PERMIT) { 2837 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2838 tp->snd_recover = tp->snd_una; 2839 } 2840 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2841 tp->snd_nxt = tp->snd_una; 2842 2843 switch (tp->t_state) { 2844 2845 /* 2846 * In FIN_WAIT_1 STATE in addition to the processing 2847 * for the ESTABLISHED state if our FIN is now acknowledged 2848 * then enter FIN_WAIT_2. 2849 */ 2850 case TCPS_FIN_WAIT_1: 2851 if (ourfinisacked) { 2852 /* 2853 * If we can't receive any more 2854 * data, then closing user can proceed. 2855 * Starting the timer is contrary to the 2856 * specification, but if we don't get a FIN 2857 * we'll hang forever. 2858 * 2859 * XXXjl: 2860 * we should release the tp also, and use a 2861 * compressed state. 2862 */ 2863 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2864 soisdisconnected(so); 2865 tcp_timer_activate(tp, TT_2MSL, 2866 (tcp_fast_finwait2_recycle ? 2867 tcp_finwait2_timeout : 2868 TP_MAXIDLE(tp))); 2869 } 2870 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2871 } 2872 break; 2873 2874 /* 2875 * In CLOSING STATE in addition to the processing for 2876 * the ESTABLISHED state if the ACK acknowledges our FIN 2877 * then enter the TIME-WAIT state, otherwise ignore 2878 * the segment. 2879 */ 2880 case TCPS_CLOSING: 2881 if (ourfinisacked) { 2882 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2883 tcp_twstart(tp); 2884 m_freem(m); 2885 return; 2886 } 2887 break; 2888 2889 /* 2890 * In LAST_ACK, we may still be waiting for data to drain 2891 * and/or to be acked, as well as for the ack of our FIN. 2892 * If our FIN is now acknowledged, delete the TCB, 2893 * enter the closed state and return. 2894 */ 2895 case TCPS_LAST_ACK: 2896 if (ourfinisacked) { 2897 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2898 tp = tcp_close(tp); 2899 goto drop; 2900 } 2901 break; 2902 } 2903 } 2904 2905 step6: 2906 INP_WLOCK_ASSERT(tp->t_inpcb); 2907 2908 /* 2909 * Update window information. 2910 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2911 */ 2912 if ((thflags & TH_ACK) && 2913 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2914 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2915 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2916 /* keep track of pure window updates */ 2917 if (tlen == 0 && 2918 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2919 TCPSTAT_INC(tcps_rcvwinupd); 2920 tp->snd_wnd = tiwin; 2921 tp->snd_wl1 = th->th_seq; 2922 tp->snd_wl2 = th->th_ack; 2923 if (tp->snd_wnd > tp->max_sndwnd) 2924 tp->max_sndwnd = tp->snd_wnd; 2925 needoutput = 1; 2926 } 2927 2928 /* 2929 * Process segments with URG. 2930 */ 2931 if ((thflags & TH_URG) && th->th_urp && 2932 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2933 /* 2934 * This is a kludge, but if we receive and accept 2935 * random urgent pointers, we'll crash in 2936 * soreceive. It's hard to imagine someone 2937 * actually wanting to send this much urgent data. 2938 */ 2939 SOCKBUF_LOCK(&so->so_rcv); 2940 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2941 th->th_urp = 0; /* XXX */ 2942 thflags &= ~TH_URG; /* XXX */ 2943 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2944 goto dodata; /* XXX */ 2945 } 2946 /* 2947 * If this segment advances the known urgent pointer, 2948 * then mark the data stream. This should not happen 2949 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2950 * a FIN has been received from the remote side. 2951 * In these states we ignore the URG. 2952 * 2953 * According to RFC961 (Assigned Protocols), 2954 * the urgent pointer points to the last octet 2955 * of urgent data. We continue, however, 2956 * to consider it to indicate the first octet 2957 * of data past the urgent section as the original 2958 * spec states (in one of two places). 2959 */ 2960 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2961 tp->rcv_up = th->th_seq + th->th_urp; 2962 so->so_oobmark = sbavail(&so->so_rcv) + 2963 (tp->rcv_up - tp->rcv_nxt) - 1; 2964 if (so->so_oobmark == 0) 2965 so->so_rcv.sb_state |= SBS_RCVATMARK; 2966 sohasoutofband(so); 2967 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2968 } 2969 SOCKBUF_UNLOCK(&so->so_rcv); 2970 /* 2971 * Remove out of band data so doesn't get presented to user. 2972 * This can happen independent of advancing the URG pointer, 2973 * but if two URG's are pending at once, some out-of-band 2974 * data may creep in... ick. 2975 */ 2976 if (th->th_urp <= (uint32_t)tlen && 2977 !(so->so_options & SO_OOBINLINE)) { 2978 /* hdr drop is delayed */ 2979 tcp_pulloutofband(so, th, m, drop_hdrlen); 2980 } 2981 } else { 2982 /* 2983 * If no out of band data is expected, 2984 * pull receive urgent pointer along 2985 * with the receive window. 2986 */ 2987 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2988 tp->rcv_up = tp->rcv_nxt; 2989 } 2990 dodata: /* XXX */ 2991 INP_WLOCK_ASSERT(tp->t_inpcb); 2992 2993 /* 2994 * Process the segment text, merging it into the TCP sequencing queue, 2995 * and arranging for acknowledgment of receipt if necessary. 2996 * This process logically involves adjusting tp->rcv_wnd as data 2997 * is presented to the user (this happens in tcp_usrreq.c, 2998 * case PRU_RCVD). If a FIN has already been received on this 2999 * connection then we just ignore the text. 3000 */ 3001 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3002 IS_FASTOPEN(tp->t_flags)); 3003 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3004 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3005 tcp_seq save_start = th->th_seq; 3006 m_adj(m, drop_hdrlen); /* delayed header drop */ 3007 /* 3008 * Insert segment which includes th into TCP reassembly queue 3009 * with control block tp. Set thflags to whether reassembly now 3010 * includes a segment with FIN. This handles the common case 3011 * inline (segment is the next to be received on an established 3012 * connection, and the queue is empty), avoiding linkage into 3013 * and removal from the queue and repetition of various 3014 * conversions. 3015 * Set DELACK for segments received in order, but ack 3016 * immediately when segments are out of order (so 3017 * fast retransmit can work). 3018 */ 3019 if (th->th_seq == tp->rcv_nxt && 3020 LIST_EMPTY(&tp->t_segq) && 3021 (TCPS_HAVEESTABLISHED(tp->t_state) || 3022 tfo_syn)) { 3023 if (DELAY_ACK(tp, tlen) || tfo_syn) 3024 tp->t_flags |= TF_DELACK; 3025 else 3026 tp->t_flags |= TF_ACKNOW; 3027 tp->rcv_nxt += tlen; 3028 thflags = th->th_flags & TH_FIN; 3029 TCPSTAT_INC(tcps_rcvpack); 3030 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3031 SOCKBUF_LOCK(&so->so_rcv); 3032 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3033 m_freem(m); 3034 else 3035 sbappendstream_locked(&so->so_rcv, m, 0); 3036 /* NB: sorwakeup_locked() does an implicit unlock. */ 3037 sorwakeup_locked(so); 3038 } else { 3039 /* 3040 * XXX: Due to the header drop above "th" is 3041 * theoretically invalid by now. Fortunately 3042 * m_adj() doesn't actually frees any mbufs 3043 * when trimming from the head. 3044 */ 3045 thflags = tcp_reass(tp, th, &tlen, m); 3046 tp->t_flags |= TF_ACKNOW; 3047 } 3048 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3049 tcp_update_sack_list(tp, save_start, save_start + tlen); 3050 #if 0 3051 /* 3052 * Note the amount of data that peer has sent into 3053 * our window, in order to estimate the sender's 3054 * buffer size. 3055 * XXX: Unused. 3056 */ 3057 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3058 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3059 else 3060 len = so->so_rcv.sb_hiwat; 3061 #endif 3062 } else { 3063 m_freem(m); 3064 thflags &= ~TH_FIN; 3065 } 3066 3067 /* 3068 * If FIN is received ACK the FIN and let the user know 3069 * that the connection is closing. 3070 */ 3071 if (thflags & TH_FIN) { 3072 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3073 socantrcvmore(so); 3074 /* 3075 * If connection is half-synchronized 3076 * (ie NEEDSYN flag on) then delay ACK, 3077 * so it may be piggybacked when SYN is sent. 3078 * Otherwise, since we received a FIN then no 3079 * more input can be expected, send ACK now. 3080 */ 3081 if (tp->t_flags & TF_NEEDSYN) 3082 tp->t_flags |= TF_DELACK; 3083 else 3084 tp->t_flags |= TF_ACKNOW; 3085 tp->rcv_nxt++; 3086 } 3087 switch (tp->t_state) { 3088 3089 /* 3090 * In SYN_RECEIVED and ESTABLISHED STATES 3091 * enter the CLOSE_WAIT state. 3092 */ 3093 case TCPS_SYN_RECEIVED: 3094 tp->t_starttime = ticks; 3095 /* FALLTHROUGH */ 3096 case TCPS_ESTABLISHED: 3097 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3098 break; 3099 3100 /* 3101 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3102 * enter the CLOSING state. 3103 */ 3104 case TCPS_FIN_WAIT_1: 3105 tcp_state_change(tp, TCPS_CLOSING); 3106 break; 3107 3108 /* 3109 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3110 * starting the time-wait timer, turning off the other 3111 * standard timers. 3112 */ 3113 case TCPS_FIN_WAIT_2: 3114 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3115 3116 tcp_twstart(tp); 3117 return; 3118 } 3119 } 3120 #ifdef TCPDEBUG 3121 if (so->so_options & SO_DEBUG) 3122 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3123 &tcp_savetcp, 0); 3124 #endif 3125 TCP_PROBE3(debug__input, tp, th, m); 3126 3127 /* 3128 * Return any desired output. 3129 */ 3130 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3131 (void) tp->t_fb->tfb_tcp_output(tp); 3132 3133 check_delack: 3134 INP_WLOCK_ASSERT(tp->t_inpcb); 3135 3136 if (tp->t_flags & TF_DELACK) { 3137 tp->t_flags &= ~TF_DELACK; 3138 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3139 } 3140 INP_WUNLOCK(tp->t_inpcb); 3141 return; 3142 3143 dropafterack: 3144 /* 3145 * Generate an ACK dropping incoming segment if it occupies 3146 * sequence space, where the ACK reflects our state. 3147 * 3148 * We can now skip the test for the RST flag since all 3149 * paths to this code happen after packets containing 3150 * RST have been dropped. 3151 * 3152 * In the SYN-RECEIVED state, don't send an ACK unless the 3153 * segment we received passes the SYN-RECEIVED ACK test. 3154 * If it fails send a RST. This breaks the loop in the 3155 * "LAND" DoS attack, and also prevents an ACK storm 3156 * between two listening ports that have been sent forged 3157 * SYN segments, each with the source address of the other. 3158 */ 3159 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3160 (SEQ_GT(tp->snd_una, th->th_ack) || 3161 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3162 rstreason = BANDLIM_RST_OPENPORT; 3163 goto dropwithreset; 3164 } 3165 #ifdef TCPDEBUG 3166 if (so->so_options & SO_DEBUG) 3167 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3168 &tcp_savetcp, 0); 3169 #endif 3170 TCP_PROBE3(debug__input, tp, th, m); 3171 tp->t_flags |= TF_ACKNOW; 3172 (void) tp->t_fb->tfb_tcp_output(tp); 3173 INP_WUNLOCK(tp->t_inpcb); 3174 m_freem(m); 3175 return; 3176 3177 dropwithreset: 3178 if (tp != NULL) { 3179 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3180 INP_WUNLOCK(tp->t_inpcb); 3181 } else 3182 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3183 return; 3184 3185 drop: 3186 /* 3187 * Drop space held by incoming segment and return. 3188 */ 3189 #ifdef TCPDEBUG 3190 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3191 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3192 &tcp_savetcp, 0); 3193 #endif 3194 TCP_PROBE3(debug__input, tp, th, m); 3195 if (tp != NULL) 3196 INP_WUNLOCK(tp->t_inpcb); 3197 m_freem(m); 3198 } 3199 3200 /* 3201 * Issue RST and make ACK acceptable to originator of segment. 3202 * The mbuf must still include the original packet header. 3203 * tp may be NULL. 3204 */ 3205 void 3206 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3207 int tlen, int rstreason) 3208 { 3209 #ifdef INET 3210 struct ip *ip; 3211 #endif 3212 #ifdef INET6 3213 struct ip6_hdr *ip6; 3214 #endif 3215 3216 if (tp != NULL) { 3217 INP_WLOCK_ASSERT(tp->t_inpcb); 3218 } 3219 3220 /* Don't bother if destination was broadcast/multicast. */ 3221 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3222 goto drop; 3223 #ifdef INET6 3224 if (mtod(m, struct ip *)->ip_v == 6) { 3225 ip6 = mtod(m, struct ip6_hdr *); 3226 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3227 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3228 goto drop; 3229 /* IPv6 anycast check is done at tcp6_input() */ 3230 } 3231 #endif 3232 #if defined(INET) && defined(INET6) 3233 else 3234 #endif 3235 #ifdef INET 3236 { 3237 ip = mtod(m, struct ip *); 3238 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3239 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3240 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3241 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3242 goto drop; 3243 } 3244 #endif 3245 3246 /* Perform bandwidth limiting. */ 3247 if (badport_bandlim(rstreason) < 0) 3248 goto drop; 3249 3250 /* tcp_respond consumes the mbuf chain. */ 3251 if (th->th_flags & TH_ACK) { 3252 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3253 th->th_ack, TH_RST); 3254 } else { 3255 if (th->th_flags & TH_SYN) 3256 tlen++; 3257 if (th->th_flags & TH_FIN) 3258 tlen++; 3259 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3260 (tcp_seq)0, TH_RST|TH_ACK); 3261 } 3262 return; 3263 drop: 3264 m_freem(m); 3265 } 3266 3267 /* 3268 * Parse TCP options and place in tcpopt. 3269 */ 3270 void 3271 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3272 { 3273 int opt, optlen; 3274 3275 to->to_flags = 0; 3276 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3277 opt = cp[0]; 3278 if (opt == TCPOPT_EOL) 3279 break; 3280 if (opt == TCPOPT_NOP) 3281 optlen = 1; 3282 else { 3283 if (cnt < 2) 3284 break; 3285 optlen = cp[1]; 3286 if (optlen < 2 || optlen > cnt) 3287 break; 3288 } 3289 switch (opt) { 3290 case TCPOPT_MAXSEG: 3291 if (optlen != TCPOLEN_MAXSEG) 3292 continue; 3293 if (!(flags & TO_SYN)) 3294 continue; 3295 to->to_flags |= TOF_MSS; 3296 bcopy((char *)cp + 2, 3297 (char *)&to->to_mss, sizeof(to->to_mss)); 3298 to->to_mss = ntohs(to->to_mss); 3299 break; 3300 case TCPOPT_WINDOW: 3301 if (optlen != TCPOLEN_WINDOW) 3302 continue; 3303 if (!(flags & TO_SYN)) 3304 continue; 3305 to->to_flags |= TOF_SCALE; 3306 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3307 break; 3308 case TCPOPT_TIMESTAMP: 3309 if (optlen != TCPOLEN_TIMESTAMP) 3310 continue; 3311 to->to_flags |= TOF_TS; 3312 bcopy((char *)cp + 2, 3313 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3314 to->to_tsval = ntohl(to->to_tsval); 3315 bcopy((char *)cp + 6, 3316 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3317 to->to_tsecr = ntohl(to->to_tsecr); 3318 break; 3319 case TCPOPT_SIGNATURE: 3320 /* 3321 * In order to reply to a host which has set the 3322 * TCP_SIGNATURE option in its initial SYN, we have 3323 * to record the fact that the option was observed 3324 * here for the syncache code to perform the correct 3325 * response. 3326 */ 3327 if (optlen != TCPOLEN_SIGNATURE) 3328 continue; 3329 to->to_flags |= TOF_SIGNATURE; 3330 to->to_signature = cp + 2; 3331 break; 3332 case TCPOPT_SACK_PERMITTED: 3333 if (optlen != TCPOLEN_SACK_PERMITTED) 3334 continue; 3335 if (!(flags & TO_SYN)) 3336 continue; 3337 if (!V_tcp_do_sack) 3338 continue; 3339 to->to_flags |= TOF_SACKPERM; 3340 break; 3341 case TCPOPT_SACK: 3342 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3343 continue; 3344 if (flags & TO_SYN) 3345 continue; 3346 to->to_flags |= TOF_SACK; 3347 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3348 to->to_sacks = cp + 2; 3349 TCPSTAT_INC(tcps_sack_rcv_blocks); 3350 break; 3351 case TCPOPT_FAST_OPEN: 3352 /* 3353 * Cookie length validation is performed by the 3354 * server side cookie checking code or the client 3355 * side cookie cache update code. 3356 */ 3357 if (!(flags & TO_SYN)) 3358 continue; 3359 if (!V_tcp_fastopen_client_enable && 3360 !V_tcp_fastopen_server_enable) 3361 continue; 3362 to->to_flags |= TOF_FASTOPEN; 3363 to->to_tfo_len = optlen - 2; 3364 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3365 break; 3366 default: 3367 continue; 3368 } 3369 } 3370 } 3371 3372 /* 3373 * Pull out of band byte out of a segment so 3374 * it doesn't appear in the user's data queue. 3375 * It is still reflected in the segment length for 3376 * sequencing purposes. 3377 */ 3378 void 3379 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3380 int off) 3381 { 3382 int cnt = off + th->th_urp - 1; 3383 3384 while (cnt >= 0) { 3385 if (m->m_len > cnt) { 3386 char *cp = mtod(m, caddr_t) + cnt; 3387 struct tcpcb *tp = sototcpcb(so); 3388 3389 INP_WLOCK_ASSERT(tp->t_inpcb); 3390 3391 tp->t_iobc = *cp; 3392 tp->t_oobflags |= TCPOOB_HAVEDATA; 3393 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3394 m->m_len--; 3395 if (m->m_flags & M_PKTHDR) 3396 m->m_pkthdr.len--; 3397 return; 3398 } 3399 cnt -= m->m_len; 3400 m = m->m_next; 3401 if (m == NULL) 3402 break; 3403 } 3404 panic("tcp_pulloutofband"); 3405 } 3406 3407 /* 3408 * Collect new round-trip time estimate 3409 * and update averages and current timeout. 3410 */ 3411 void 3412 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3413 { 3414 int delta; 3415 3416 INP_WLOCK_ASSERT(tp->t_inpcb); 3417 3418 TCPSTAT_INC(tcps_rttupdated); 3419 tp->t_rttupdated++; 3420 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3421 /* 3422 * srtt is stored as fixed point with 5 bits after the 3423 * binary point (i.e., scaled by 8). The following magic 3424 * is equivalent to the smoothing algorithm in rfc793 with 3425 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3426 * point). Adjust rtt to origin 0. 3427 */ 3428 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3429 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3430 3431 if ((tp->t_srtt += delta) <= 0) 3432 tp->t_srtt = 1; 3433 3434 /* 3435 * We accumulate a smoothed rtt variance (actually, a 3436 * smoothed mean difference), then set the retransmit 3437 * timer to smoothed rtt + 4 times the smoothed variance. 3438 * rttvar is stored as fixed point with 4 bits after the 3439 * binary point (scaled by 16). The following is 3440 * equivalent to rfc793 smoothing with an alpha of .75 3441 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3442 * rfc793's wired-in beta. 3443 */ 3444 if (delta < 0) 3445 delta = -delta; 3446 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3447 if ((tp->t_rttvar += delta) <= 0) 3448 tp->t_rttvar = 1; 3449 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3450 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3451 } else { 3452 /* 3453 * No rtt measurement yet - use the unsmoothed rtt. 3454 * Set the variance to half the rtt (so our first 3455 * retransmit happens at 3*rtt). 3456 */ 3457 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3458 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3459 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3460 } 3461 tp->t_rtttime = 0; 3462 tp->t_rxtshift = 0; 3463 3464 /* 3465 * the retransmit should happen at rtt + 4 * rttvar. 3466 * Because of the way we do the smoothing, srtt and rttvar 3467 * will each average +1/2 tick of bias. When we compute 3468 * the retransmit timer, we want 1/2 tick of rounding and 3469 * 1 extra tick because of +-1/2 tick uncertainty in the 3470 * firing of the timer. The bias will give us exactly the 3471 * 1.5 tick we need. But, because the bias is 3472 * statistical, we have to test that we don't drop below 3473 * the minimum feasible timer (which is 2 ticks). 3474 */ 3475 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3476 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3477 3478 /* 3479 * We received an ack for a packet that wasn't retransmitted; 3480 * it is probably safe to discard any error indications we've 3481 * received recently. This isn't quite right, but close enough 3482 * for now (a route might have failed after we sent a segment, 3483 * and the return path might not be symmetrical). 3484 */ 3485 tp->t_softerror = 0; 3486 } 3487 3488 /* 3489 * Determine a reasonable value for maxseg size. 3490 * If the route is known, check route for mtu. 3491 * If none, use an mss that can be handled on the outgoing interface 3492 * without forcing IP to fragment. If no route is found, route has no mtu, 3493 * or the destination isn't local, use a default, hopefully conservative 3494 * size (usually 512 or the default IP max size, but no more than the mtu 3495 * of the interface), as we can't discover anything about intervening 3496 * gateways or networks. We also initialize the congestion/slow start 3497 * window to be a single segment if the destination isn't local. 3498 * While looking at the routing entry, we also initialize other path-dependent 3499 * parameters from pre-set or cached values in the routing entry. 3500 * 3501 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3502 * IP options, e.g. IPSEC data, since length of this data may vary, and 3503 * thus it is calculated for every segment separately in tcp_output(). 3504 * 3505 * NOTE that this routine is only called when we process an incoming 3506 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3507 * settings are handled in tcp_mssopt(). 3508 */ 3509 void 3510 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3511 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3512 { 3513 int mss = 0; 3514 uint32_t maxmtu = 0; 3515 struct inpcb *inp = tp->t_inpcb; 3516 struct hc_metrics_lite metrics; 3517 #ifdef INET6 3518 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3519 size_t min_protoh = isipv6 ? 3520 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3521 sizeof (struct tcpiphdr); 3522 #else 3523 const size_t min_protoh = sizeof(struct tcpiphdr); 3524 #endif 3525 3526 INP_WLOCK_ASSERT(tp->t_inpcb); 3527 3528 if (mtuoffer != -1) { 3529 KASSERT(offer == -1, ("%s: conflict", __func__)); 3530 offer = mtuoffer - min_protoh; 3531 } 3532 3533 /* Initialize. */ 3534 #ifdef INET6 3535 if (isipv6) { 3536 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3537 tp->t_maxseg = V_tcp_v6mssdflt; 3538 } 3539 #endif 3540 #if defined(INET) && defined(INET6) 3541 else 3542 #endif 3543 #ifdef INET 3544 { 3545 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3546 tp->t_maxseg = V_tcp_mssdflt; 3547 } 3548 #endif 3549 3550 /* 3551 * No route to sender, stay with default mss and return. 3552 */ 3553 if (maxmtu == 0) { 3554 /* 3555 * In case we return early we need to initialize metrics 3556 * to a defined state as tcp_hc_get() would do for us 3557 * if there was no cache hit. 3558 */ 3559 if (metricptr != NULL) 3560 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3561 return; 3562 } 3563 3564 /* What have we got? */ 3565 switch (offer) { 3566 case 0: 3567 /* 3568 * Offer == 0 means that there was no MSS on the SYN 3569 * segment, in this case we use tcp_mssdflt as 3570 * already assigned to t_maxseg above. 3571 */ 3572 offer = tp->t_maxseg; 3573 break; 3574 3575 case -1: 3576 /* 3577 * Offer == -1 means that we didn't receive SYN yet. 3578 */ 3579 /* FALLTHROUGH */ 3580 3581 default: 3582 /* 3583 * Prevent DoS attack with too small MSS. Round up 3584 * to at least minmss. 3585 */ 3586 offer = max(offer, V_tcp_minmss); 3587 } 3588 3589 /* 3590 * rmx information is now retrieved from tcp_hostcache. 3591 */ 3592 tcp_hc_get(&inp->inp_inc, &metrics); 3593 if (metricptr != NULL) 3594 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3595 3596 /* 3597 * If there's a discovered mtu in tcp hostcache, use it. 3598 * Else, use the link mtu. 3599 */ 3600 if (metrics.rmx_mtu) 3601 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3602 else { 3603 #ifdef INET6 3604 if (isipv6) { 3605 mss = maxmtu - min_protoh; 3606 if (!V_path_mtu_discovery && 3607 !in6_localaddr(&inp->in6p_faddr)) 3608 mss = min(mss, V_tcp_v6mssdflt); 3609 } 3610 #endif 3611 #if defined(INET) && defined(INET6) 3612 else 3613 #endif 3614 #ifdef INET 3615 { 3616 mss = maxmtu - min_protoh; 3617 if (!V_path_mtu_discovery && 3618 !in_localaddr(inp->inp_faddr)) 3619 mss = min(mss, V_tcp_mssdflt); 3620 } 3621 #endif 3622 /* 3623 * XXX - The above conditional (mss = maxmtu - min_protoh) 3624 * probably violates the TCP spec. 3625 * The problem is that, since we don't know the 3626 * other end's MSS, we are supposed to use a conservative 3627 * default. But, if we do that, then MTU discovery will 3628 * never actually take place, because the conservative 3629 * default is much less than the MTUs typically seen 3630 * on the Internet today. For the moment, we'll sweep 3631 * this under the carpet. 3632 * 3633 * The conservative default might not actually be a problem 3634 * if the only case this occurs is when sending an initial 3635 * SYN with options and data to a host we've never talked 3636 * to before. Then, they will reply with an MSS value which 3637 * will get recorded and the new parameters should get 3638 * recomputed. For Further Study. 3639 */ 3640 } 3641 mss = min(mss, offer); 3642 3643 /* 3644 * Sanity check: make sure that maxseg will be large 3645 * enough to allow some data on segments even if the 3646 * all the option space is used (40bytes). Otherwise 3647 * funny things may happen in tcp_output. 3648 * 3649 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3650 */ 3651 mss = max(mss, 64); 3652 3653 tp->t_maxseg = mss; 3654 } 3655 3656 void 3657 tcp_mss(struct tcpcb *tp, int offer) 3658 { 3659 int mss; 3660 uint32_t bufsize; 3661 struct inpcb *inp; 3662 struct socket *so; 3663 struct hc_metrics_lite metrics; 3664 struct tcp_ifcap cap; 3665 3666 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3667 3668 bzero(&cap, sizeof(cap)); 3669 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3670 3671 mss = tp->t_maxseg; 3672 inp = tp->t_inpcb; 3673 3674 /* 3675 * If there's a pipesize, change the socket buffer to that size, 3676 * don't change if sb_hiwat is different than default (then it 3677 * has been changed on purpose with setsockopt). 3678 * Make the socket buffers an integral number of mss units; 3679 * if the mss is larger than the socket buffer, decrease the mss. 3680 */ 3681 so = inp->inp_socket; 3682 SOCKBUF_LOCK(&so->so_snd); 3683 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3684 bufsize = metrics.rmx_sendpipe; 3685 else 3686 bufsize = so->so_snd.sb_hiwat; 3687 if (bufsize < mss) 3688 mss = bufsize; 3689 else { 3690 bufsize = roundup(bufsize, mss); 3691 if (bufsize > sb_max) 3692 bufsize = sb_max; 3693 if (bufsize > so->so_snd.sb_hiwat) 3694 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3695 } 3696 SOCKBUF_UNLOCK(&so->so_snd); 3697 /* 3698 * Sanity check: make sure that maxseg will be large 3699 * enough to allow some data on segments even if the 3700 * all the option space is used (40bytes). Otherwise 3701 * funny things may happen in tcp_output. 3702 * 3703 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3704 */ 3705 tp->t_maxseg = max(mss, 64); 3706 3707 SOCKBUF_LOCK(&so->so_rcv); 3708 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3709 bufsize = metrics.rmx_recvpipe; 3710 else 3711 bufsize = so->so_rcv.sb_hiwat; 3712 if (bufsize > mss) { 3713 bufsize = roundup(bufsize, mss); 3714 if (bufsize > sb_max) 3715 bufsize = sb_max; 3716 if (bufsize > so->so_rcv.sb_hiwat) 3717 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3718 } 3719 SOCKBUF_UNLOCK(&so->so_rcv); 3720 3721 /* Check the interface for TSO capabilities. */ 3722 if (cap.ifcap & CSUM_TSO) { 3723 tp->t_flags |= TF_TSO; 3724 tp->t_tsomax = cap.tsomax; 3725 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3726 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3727 } 3728 } 3729 3730 /* 3731 * Determine the MSS option to send on an outgoing SYN. 3732 */ 3733 int 3734 tcp_mssopt(struct in_conninfo *inc) 3735 { 3736 int mss = 0; 3737 uint32_t thcmtu = 0; 3738 uint32_t maxmtu = 0; 3739 size_t min_protoh; 3740 3741 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3742 3743 #ifdef INET6 3744 if (inc->inc_flags & INC_ISIPV6) { 3745 mss = V_tcp_v6mssdflt; 3746 maxmtu = tcp_maxmtu6(inc, NULL); 3747 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3748 } 3749 #endif 3750 #if defined(INET) && defined(INET6) 3751 else 3752 #endif 3753 #ifdef INET 3754 { 3755 mss = V_tcp_mssdflt; 3756 maxmtu = tcp_maxmtu(inc, NULL); 3757 min_protoh = sizeof(struct tcpiphdr); 3758 } 3759 #endif 3760 #if defined(INET6) || defined(INET) 3761 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3762 #endif 3763 3764 if (maxmtu && thcmtu) 3765 mss = min(maxmtu, thcmtu) - min_protoh; 3766 else if (maxmtu || thcmtu) 3767 mss = max(maxmtu, thcmtu) - min_protoh; 3768 3769 return (mss); 3770 } 3771 3772 3773 /* 3774 * On a partial ack arrives, force the retransmission of the 3775 * next unacknowledged segment. Do not clear tp->t_dupacks. 3776 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3777 * be started again. 3778 */ 3779 void 3780 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3781 { 3782 tcp_seq onxt = tp->snd_nxt; 3783 uint32_t ocwnd = tp->snd_cwnd; 3784 u_int maxseg = tcp_maxseg(tp); 3785 3786 INP_WLOCK_ASSERT(tp->t_inpcb); 3787 3788 tcp_timer_activate(tp, TT_REXMT, 0); 3789 tp->t_rtttime = 0; 3790 tp->snd_nxt = th->th_ack; 3791 /* 3792 * Set snd_cwnd to one segment beyond acknowledged offset. 3793 * (tp->snd_una has not yet been updated when this function is called.) 3794 */ 3795 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3796 tp->t_flags |= TF_ACKNOW; 3797 (void) tp->t_fb->tfb_tcp_output(tp); 3798 tp->snd_cwnd = ocwnd; 3799 if (SEQ_GT(onxt, tp->snd_nxt)) 3800 tp->snd_nxt = onxt; 3801 /* 3802 * Partial window deflation. Relies on fact that tp->snd_una 3803 * not updated yet. 3804 */ 3805 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3806 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3807 else 3808 tp->snd_cwnd = 0; 3809 tp->snd_cwnd += maxseg; 3810 } 3811 3812 int 3813 tcp_compute_pipe(struct tcpcb *tp) 3814 { 3815 return (tp->snd_max - tp->snd_una + 3816 tp->sackhint.sack_bytes_rexmit - 3817 tp->sackhint.sacked_bytes); 3818 } 3819