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 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1149 iptos); 1150 if (ti_locked == TI_RLOCKED) 1151 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1152 return (IPPROTO_DONE); 1153 } 1154 /* 1155 * Segment flag validation for new connection attempts: 1156 * 1157 * Our (SYN|ACK) response was rejected. 1158 * Check with syncache and remove entry to prevent 1159 * retransmits. 1160 * 1161 * NB: syncache_chkrst does its own logging of failure 1162 * causes. 1163 */ 1164 if (thflags & TH_RST) { 1165 syncache_chkrst(&inc, th); 1166 goto dropunlock; 1167 } 1168 /* 1169 * We can't do anything without SYN. 1170 */ 1171 if ((thflags & TH_SYN) == 0) { 1172 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1173 log(LOG_DEBUG, "%s; %s: Listen socket: " 1174 "SYN is missing, segment ignored\n", 1175 s, __func__); 1176 TCPSTAT_INC(tcps_badsyn); 1177 goto dropunlock; 1178 } 1179 /* 1180 * (SYN|ACK) is bogus on a listen socket. 1181 */ 1182 if (thflags & TH_ACK) { 1183 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1184 log(LOG_DEBUG, "%s; %s: Listen socket: " 1185 "SYN|ACK invalid, segment rejected\n", 1186 s, __func__); 1187 syncache_badack(&inc); /* XXX: Not needed! */ 1188 TCPSTAT_INC(tcps_badsyn); 1189 rstreason = BANDLIM_RST_OPENPORT; 1190 goto dropwithreset; 1191 } 1192 /* 1193 * If the drop_synfin option is enabled, drop all 1194 * segments with both the SYN and FIN bits set. 1195 * This prevents e.g. nmap from identifying the 1196 * TCP/IP stack. 1197 * XXX: Poor reasoning. nmap has other methods 1198 * and is constantly refining its stack detection 1199 * strategies. 1200 * XXX: This is a violation of the TCP specification 1201 * and was used by RFC1644. 1202 */ 1203 if ((thflags & TH_FIN) && V_drop_synfin) { 1204 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1205 log(LOG_DEBUG, "%s; %s: Listen socket: " 1206 "SYN|FIN segment ignored (based on " 1207 "sysctl setting)\n", s, __func__); 1208 TCPSTAT_INC(tcps_badsyn); 1209 goto dropunlock; 1210 } 1211 /* 1212 * Segment's flags are (SYN) or (SYN|FIN). 1213 * 1214 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1215 * as they do not affect the state of the TCP FSM. 1216 * The data pointed to by TH_URG and th_urp is ignored. 1217 */ 1218 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1219 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1220 KASSERT(thflags & (TH_SYN), 1221 ("%s: Listen socket: TH_SYN not set", __func__)); 1222 #ifdef INET6 1223 /* 1224 * If deprecated address is forbidden, 1225 * we do not accept SYN to deprecated interface 1226 * address to prevent any new inbound connection from 1227 * getting established. 1228 * When we do not accept SYN, we send a TCP RST, 1229 * with deprecated source address (instead of dropping 1230 * it). We compromise it as it is much better for peer 1231 * to send a RST, and RST will be the final packet 1232 * for the exchange. 1233 * 1234 * If we do not forbid deprecated addresses, we accept 1235 * the SYN packet. RFC2462 does not suggest dropping 1236 * SYN in this case. 1237 * If we decipher RFC2462 5.5.4, it says like this: 1238 * 1. use of deprecated addr with existing 1239 * communication is okay - "SHOULD continue to be 1240 * used" 1241 * 2. use of it with new communication: 1242 * (2a) "SHOULD NOT be used if alternate address 1243 * with sufficient scope is available" 1244 * (2b) nothing mentioned otherwise. 1245 * Here we fall into (2b) case as we have no choice in 1246 * our source address selection - we must obey the peer. 1247 * 1248 * The wording in RFC2462 is confusing, and there are 1249 * multiple description text for deprecated address 1250 * handling - worse, they are not exactly the same. 1251 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1252 */ 1253 if (isipv6 && !V_ip6_use_deprecated) { 1254 struct in6_ifaddr *ia6; 1255 1256 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1257 if (ia6 != NULL && 1258 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1259 ifa_free(&ia6->ia_ifa); 1260 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1261 log(LOG_DEBUG, "%s; %s: Listen socket: " 1262 "Connection attempt to deprecated " 1263 "IPv6 address rejected\n", 1264 s, __func__); 1265 rstreason = BANDLIM_RST_OPENPORT; 1266 goto dropwithreset; 1267 } 1268 if (ia6) 1269 ifa_free(&ia6->ia_ifa); 1270 } 1271 #endif /* INET6 */ 1272 /* 1273 * Basic sanity checks on incoming SYN requests: 1274 * Don't respond if the destination is a link layer 1275 * broadcast according to RFC1122 4.2.3.10, p. 104. 1276 * If it is from this socket it must be forged. 1277 * Don't respond if the source or destination is a 1278 * global or subnet broad- or multicast address. 1279 * Note that it is quite possible to receive unicast 1280 * link-layer packets with a broadcast IP address. Use 1281 * in_broadcast() to find them. 1282 */ 1283 if (m->m_flags & (M_BCAST|M_MCAST)) { 1284 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1285 log(LOG_DEBUG, "%s; %s: Listen socket: " 1286 "Connection attempt from broad- or multicast " 1287 "link layer address ignored\n", s, __func__); 1288 goto dropunlock; 1289 } 1290 #ifdef INET6 1291 if (isipv6) { 1292 if (th->th_dport == th->th_sport && 1293 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1294 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1295 log(LOG_DEBUG, "%s; %s: Listen socket: " 1296 "Connection attempt to/from self " 1297 "ignored\n", s, __func__); 1298 goto dropunlock; 1299 } 1300 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1301 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1302 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1303 log(LOG_DEBUG, "%s; %s: Listen socket: " 1304 "Connection attempt from/to multicast " 1305 "address ignored\n", s, __func__); 1306 goto dropunlock; 1307 } 1308 } 1309 #endif 1310 #if defined(INET) && defined(INET6) 1311 else 1312 #endif 1313 #ifdef INET 1314 { 1315 if (th->th_dport == th->th_sport && 1316 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1318 log(LOG_DEBUG, "%s; %s: Listen socket: " 1319 "Connection attempt from/to self " 1320 "ignored\n", s, __func__); 1321 goto dropunlock; 1322 } 1323 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1324 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1325 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1326 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1327 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1328 log(LOG_DEBUG, "%s; %s: Listen socket: " 1329 "Connection attempt from/to broad- " 1330 "or multicast address ignored\n", 1331 s, __func__); 1332 goto dropunlock; 1333 } 1334 } 1335 #endif 1336 /* 1337 * SYN appears to be valid. Create compressed TCP state 1338 * for syncache. 1339 */ 1340 #ifdef TCPDEBUG 1341 if (so->so_options & SO_DEBUG) 1342 tcp_trace(TA_INPUT, ostate, tp, 1343 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1344 #endif 1345 TCP_PROBE3(debug__input, tp, th, m); 1346 tcp_dooptions(&to, optp, optlen, TO_SYN); 1347 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1348 goto tfo_socket_result; 1349 1350 /* 1351 * Entry added to syncache and mbuf consumed. 1352 * Only the listen socket is unlocked by syncache_add(). 1353 */ 1354 if (ti_locked == TI_RLOCKED) { 1355 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1356 ti_locked = TI_UNLOCKED; 1357 } 1358 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1359 return (IPPROTO_DONE); 1360 } else if (tp->t_state == TCPS_LISTEN) { 1361 /* 1362 * When a listen socket is torn down the SO_ACCEPTCONN 1363 * flag is removed first while connections are drained 1364 * from the accept queue in a unlock/lock cycle of the 1365 * ACCEPT_LOCK, opening a race condition allowing a SYN 1366 * attempt go through unhandled. 1367 */ 1368 goto dropunlock; 1369 } 1370 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1371 if (tp->t_flags & TF_SIGNATURE) { 1372 tcp_dooptions(&to, optp, optlen, thflags); 1373 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1374 TCPSTAT_INC(tcps_sig_err_nosigopt); 1375 goto dropunlock; 1376 } 1377 if (!TCPMD5_ENABLED() || 1378 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1379 goto dropunlock; 1380 } 1381 #endif 1382 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1383 1384 /* 1385 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1386 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1387 * the inpcb, and unlocks pcbinfo. 1388 */ 1389 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1390 if (ti_locked == TI_RLOCKED) 1391 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1392 return (IPPROTO_DONE); 1393 1394 dropwithreset: 1395 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1396 1397 if (ti_locked == TI_RLOCKED) { 1398 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1399 ti_locked = TI_UNLOCKED; 1400 } 1401 #ifdef INVARIANTS 1402 else { 1403 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1404 "ti_locked: %d", __func__, ti_locked)); 1405 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1406 } 1407 #endif 1408 1409 if (inp != NULL) { 1410 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1411 INP_WUNLOCK(inp); 1412 } else 1413 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1414 m = NULL; /* mbuf chain got consumed. */ 1415 goto drop; 1416 1417 dropunlock: 1418 if (m != NULL) 1419 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1420 1421 if (ti_locked == TI_RLOCKED) { 1422 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1423 ti_locked = TI_UNLOCKED; 1424 } 1425 #ifdef INVARIANTS 1426 else { 1427 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1428 "ti_locked: %d", __func__, ti_locked)); 1429 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1430 } 1431 #endif 1432 1433 if (inp != NULL) 1434 INP_WUNLOCK(inp); 1435 1436 drop: 1437 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1438 if (s != NULL) 1439 free(s, M_TCPLOG); 1440 if (m != NULL) 1441 m_freem(m); 1442 return (IPPROTO_DONE); 1443 } 1444 1445 /* 1446 * Automatic sizing of receive socket buffer. Often the send 1447 * buffer size is not optimally adjusted to the actual network 1448 * conditions at hand (delay bandwidth product). Setting the 1449 * buffer size too small limits throughput on links with high 1450 * bandwidth and high delay (eg. trans-continental/oceanic links). 1451 * 1452 * On the receive side the socket buffer memory is only rarely 1453 * used to any significant extent. This allows us to be much 1454 * more aggressive in scaling the receive socket buffer. For 1455 * the case that the buffer space is actually used to a large 1456 * extent and we run out of kernel memory we can simply drop 1457 * the new segments; TCP on the sender will just retransmit it 1458 * later. Setting the buffer size too big may only consume too 1459 * much kernel memory if the application doesn't read() from 1460 * the socket or packet loss or reordering makes use of the 1461 * reassembly queue. 1462 * 1463 * The criteria to step up the receive buffer one notch are: 1464 * 1. Application has not set receive buffer size with 1465 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1466 * 2. the number of bytes received during the time it takes 1467 * one timestamp to be reflected back to us (the RTT); 1468 * 3. received bytes per RTT is within seven eighth of the 1469 * current socket buffer size; 1470 * 4. receive buffer size has not hit maximal automatic size; 1471 * 1472 * This algorithm does one step per RTT at most and only if 1473 * we receive a bulk stream w/o packet losses or reorderings. 1474 * Shrinking the buffer during idle times is not necessary as 1475 * it doesn't consume any memory when idle. 1476 * 1477 * TODO: Only step up if the application is actually serving 1478 * the buffer to better manage the socket buffer resources. 1479 */ 1480 int 1481 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1482 struct tcpcb *tp, int tlen) 1483 { 1484 int newsize = 0; 1485 1486 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1487 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1488 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1489 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1490 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1491 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1492 newsize = min(so->so_rcv.sb_hiwat + 1493 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1494 } 1495 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1496 1497 /* Start over with next RTT. */ 1498 tp->rfbuf_ts = 0; 1499 tp->rfbuf_cnt = 0; 1500 } else { 1501 tp->rfbuf_cnt += tlen; /* add up */ 1502 } 1503 1504 return (newsize); 1505 } 1506 1507 void 1508 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1509 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1510 { 1511 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1512 int rstreason, todrop, win; 1513 uint32_t tiwin; 1514 uint16_t nsegs; 1515 char *s; 1516 struct in_conninfo *inc; 1517 struct mbuf *mfree; 1518 struct tcpopt to; 1519 int tfo_syn; 1520 1521 #ifdef TCPDEBUG 1522 /* 1523 * The size of tcp_saveipgen must be the size of the max ip header, 1524 * now IPv6. 1525 */ 1526 u_char tcp_saveipgen[IP6_HDR_LEN]; 1527 struct tcphdr tcp_savetcp; 1528 short ostate = 0; 1529 #endif 1530 thflags = th->th_flags; 1531 inc = &tp->t_inpcb->inp_inc; 1532 tp->sackhint.last_sack_ack = 0; 1533 sack_changed = 0; 1534 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1535 /* 1536 * If this is either a state-changing packet or current state isn't 1537 * established, we require a write lock on tcbinfo. Otherwise, we 1538 * allow the tcbinfo to be in either alocked or unlocked, as the 1539 * caller may have unnecessarily acquired a write lock due to a race. 1540 */ 1541 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1542 tp->t_state != TCPS_ESTABLISHED) { 1543 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1544 } 1545 INP_WLOCK_ASSERT(tp->t_inpcb); 1546 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1547 __func__)); 1548 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1549 __func__)); 1550 1551 #ifdef TCPPCAP 1552 /* Save segment, if requested. */ 1553 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1554 #endif 1555 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1556 tlen, NULL, true); 1557 1558 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1559 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1560 log(LOG_DEBUG, "%s; %s: " 1561 "SYN|FIN segment ignored (based on " 1562 "sysctl setting)\n", s, __func__); 1563 free(s, M_TCPLOG); 1564 } 1565 goto drop; 1566 } 1567 1568 /* 1569 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1570 * check SEQ.ACK first. 1571 */ 1572 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1573 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1574 rstreason = BANDLIM_UNLIMITED; 1575 goto dropwithreset; 1576 } 1577 1578 /* 1579 * Segment received on connection. 1580 * Reset idle time and keep-alive timer. 1581 * XXX: This should be done after segment 1582 * validation to ignore broken/spoofed segs. 1583 */ 1584 tp->t_rcvtime = ticks; 1585 1586 /* 1587 * Scale up the window into a 32-bit value. 1588 * For the SYN_SENT state the scale is zero. 1589 */ 1590 tiwin = th->th_win << tp->snd_scale; 1591 1592 /* 1593 * TCP ECN processing. 1594 */ 1595 if (tp->t_flags & TF_ECN_PERMIT) { 1596 if (thflags & TH_CWR) 1597 tp->t_flags &= ~TF_ECN_SND_ECE; 1598 switch (iptos & IPTOS_ECN_MASK) { 1599 case IPTOS_ECN_CE: 1600 tp->t_flags |= TF_ECN_SND_ECE; 1601 TCPSTAT_INC(tcps_ecn_ce); 1602 break; 1603 case IPTOS_ECN_ECT0: 1604 TCPSTAT_INC(tcps_ecn_ect0); 1605 break; 1606 case IPTOS_ECN_ECT1: 1607 TCPSTAT_INC(tcps_ecn_ect1); 1608 break; 1609 } 1610 1611 /* Process a packet differently from RFC3168. */ 1612 cc_ecnpkt_handler(tp, th, iptos); 1613 1614 /* Congestion experienced. */ 1615 if (thflags & TH_ECE) { 1616 cc_cong_signal(tp, th, CC_ECN); 1617 } 1618 } 1619 1620 /* 1621 * Parse options on any incoming segment. 1622 */ 1623 tcp_dooptions(&to, (u_char *)(th + 1), 1624 (th->th_off << 2) - sizeof(struct tcphdr), 1625 (thflags & TH_SYN) ? TO_SYN : 0); 1626 1627 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1628 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1629 (to.to_flags & TOF_SIGNATURE) == 0) { 1630 TCPSTAT_INC(tcps_sig_err_sigopt); 1631 /* XXX: should drop? */ 1632 } 1633 #endif 1634 /* 1635 * If echoed timestamp is later than the current time, 1636 * fall back to non RFC1323 RTT calculation. Normalize 1637 * timestamp if syncookies were used when this connection 1638 * was established. 1639 */ 1640 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1641 to.to_tsecr -= tp->ts_offset; 1642 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1643 to.to_tsecr = 0; 1644 else if (tp->t_flags & TF_PREVVALID && 1645 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1646 cc_cong_signal(tp, th, CC_RTO_ERR); 1647 } 1648 /* 1649 * Process options only when we get SYN/ACK back. The SYN case 1650 * for incoming connections is handled in tcp_syncache. 1651 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1652 * or <SYN,ACK>) segment itself is never scaled. 1653 * XXX this is traditional behavior, may need to be cleaned up. 1654 */ 1655 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1656 if ((to.to_flags & TOF_SCALE) && 1657 (tp->t_flags & TF_REQ_SCALE)) { 1658 tp->t_flags |= TF_RCVD_SCALE; 1659 tp->snd_scale = to.to_wscale; 1660 } 1661 /* 1662 * Initial send window. It will be updated with 1663 * the next incoming segment to the scaled value. 1664 */ 1665 tp->snd_wnd = th->th_win; 1666 if (to.to_flags & TOF_TS) { 1667 tp->t_flags |= TF_RCVD_TSTMP; 1668 tp->ts_recent = to.to_tsval; 1669 tp->ts_recent_age = tcp_ts_getticks(); 1670 } 1671 if (to.to_flags & TOF_MSS) 1672 tcp_mss(tp, to.to_mss); 1673 if ((tp->t_flags & TF_SACK_PERMIT) && 1674 (to.to_flags & TOF_SACKPERM) == 0) 1675 tp->t_flags &= ~TF_SACK_PERMIT; 1676 if (IS_FASTOPEN(tp->t_flags)) { 1677 if (to.to_flags & TOF_FASTOPEN) { 1678 uint16_t mss; 1679 1680 if (to.to_flags & TOF_MSS) 1681 mss = to.to_mss; 1682 else 1683 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1684 mss = TCP6_MSS; 1685 else 1686 mss = TCP_MSS; 1687 tcp_fastopen_update_cache(tp, mss, 1688 to.to_tfo_len, to.to_tfo_cookie); 1689 } else 1690 tcp_fastopen_disable_path(tp); 1691 } 1692 } 1693 1694 /* 1695 * If timestamps were negotiated during SYN/ACK they should 1696 * appear on every segment during this session and vice versa. 1697 */ 1698 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1699 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1700 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1701 "no action\n", s, __func__); 1702 free(s, M_TCPLOG); 1703 } 1704 } 1705 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1706 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1707 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1708 "no action\n", s, __func__); 1709 free(s, M_TCPLOG); 1710 } 1711 } 1712 1713 /* 1714 * Header prediction: check for the two common cases 1715 * of a uni-directional data xfer. If the packet has 1716 * no control flags, is in-sequence, the window didn't 1717 * change and we're not retransmitting, it's a 1718 * candidate. If the length is zero and the ack moved 1719 * forward, we're the sender side of the xfer. Just 1720 * free the data acked & wake any higher level process 1721 * that was blocked waiting for space. If the length 1722 * is non-zero and the ack didn't move, we're the 1723 * receiver side. If we're getting packets in-order 1724 * (the reassembly queue is empty), add the data to 1725 * the socket buffer and note that we need a delayed ack. 1726 * Make sure that the hidden state-flags are also off. 1727 * Since we check for TCPS_ESTABLISHED first, it can only 1728 * be TH_NEEDSYN. 1729 */ 1730 if (tp->t_state == TCPS_ESTABLISHED && 1731 th->th_seq == tp->rcv_nxt && 1732 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1733 tp->snd_nxt == tp->snd_max && 1734 tiwin && tiwin == tp->snd_wnd && 1735 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1736 LIST_EMPTY(&tp->t_segq) && 1737 ((to.to_flags & TOF_TS) == 0 || 1738 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1739 1740 /* 1741 * If last ACK falls within this segment's sequence numbers, 1742 * record the timestamp. 1743 * NOTE that the test is modified according to the latest 1744 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1745 */ 1746 if ((to.to_flags & TOF_TS) != 0 && 1747 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1748 tp->ts_recent_age = tcp_ts_getticks(); 1749 tp->ts_recent = to.to_tsval; 1750 } 1751 1752 if (tlen == 0) { 1753 if (SEQ_GT(th->th_ack, tp->snd_una) && 1754 SEQ_LEQ(th->th_ack, tp->snd_max) && 1755 !IN_RECOVERY(tp->t_flags) && 1756 (to.to_flags & TOF_SACK) == 0 && 1757 TAILQ_EMPTY(&tp->snd_holes)) { 1758 /* 1759 * This is a pure ack for outstanding data. 1760 */ 1761 TCPSTAT_INC(tcps_predack); 1762 1763 /* 1764 * "bad retransmit" recovery without timestamps. 1765 */ 1766 if ((to.to_flags & TOF_TS) == 0 && 1767 tp->t_rxtshift == 1 && 1768 tp->t_flags & TF_PREVVALID && 1769 (int)(ticks - tp->t_badrxtwin) < 0) { 1770 cc_cong_signal(tp, th, CC_RTO_ERR); 1771 } 1772 1773 /* 1774 * Recalculate the transmit timer / rtt. 1775 * 1776 * Some boxes send broken timestamp replies 1777 * during the SYN+ACK phase, ignore 1778 * timestamps of 0 or we could calculate a 1779 * huge RTT and blow up the retransmit timer. 1780 */ 1781 if ((to.to_flags & TOF_TS) != 0 && 1782 to.to_tsecr) { 1783 uint32_t t; 1784 1785 t = tcp_ts_getticks() - to.to_tsecr; 1786 if (!tp->t_rttlow || tp->t_rttlow > t) 1787 tp->t_rttlow = t; 1788 tcp_xmit_timer(tp, 1789 TCP_TS_TO_TICKS(t) + 1); 1790 } else if (tp->t_rtttime && 1791 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1792 if (!tp->t_rttlow || 1793 tp->t_rttlow > ticks - tp->t_rtttime) 1794 tp->t_rttlow = ticks - tp->t_rtttime; 1795 tcp_xmit_timer(tp, 1796 ticks - tp->t_rtttime); 1797 } 1798 acked = BYTES_THIS_ACK(tp, th); 1799 1800 #ifdef TCP_HHOOK 1801 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1802 hhook_run_tcp_est_in(tp, th, &to); 1803 #endif 1804 1805 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1806 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1807 sbdrop(&so->so_snd, acked); 1808 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1809 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1810 tp->snd_recover = th->th_ack - 1; 1811 1812 /* 1813 * Let the congestion control algorithm update 1814 * congestion control related information. This 1815 * typically means increasing the congestion 1816 * window. 1817 */ 1818 cc_ack_received(tp, th, nsegs, CC_ACK); 1819 1820 tp->snd_una = th->th_ack; 1821 /* 1822 * Pull snd_wl2 up to prevent seq wrap relative 1823 * to th_ack. 1824 */ 1825 tp->snd_wl2 = th->th_ack; 1826 tp->t_dupacks = 0; 1827 m_freem(m); 1828 1829 /* 1830 * If all outstanding data are acked, stop 1831 * retransmit timer, otherwise restart timer 1832 * using current (possibly backed-off) value. 1833 * If process is waiting for space, 1834 * wakeup/selwakeup/signal. If data 1835 * are ready to send, let tcp_output 1836 * decide between more output or persist. 1837 */ 1838 #ifdef TCPDEBUG 1839 if (so->so_options & SO_DEBUG) 1840 tcp_trace(TA_INPUT, ostate, tp, 1841 (void *)tcp_saveipgen, 1842 &tcp_savetcp, 0); 1843 #endif 1844 TCP_PROBE3(debug__input, tp, th, m); 1845 if (tp->snd_una == tp->snd_max) 1846 tcp_timer_activate(tp, TT_REXMT, 0); 1847 else if (!tcp_timer_active(tp, TT_PERSIST)) 1848 tcp_timer_activate(tp, TT_REXMT, 1849 tp->t_rxtcur); 1850 sowwakeup(so); 1851 if (sbavail(&so->so_snd)) 1852 (void) tp->t_fb->tfb_tcp_output(tp); 1853 goto check_delack; 1854 } 1855 } else if (th->th_ack == tp->snd_una && 1856 tlen <= sbspace(&so->so_rcv)) { 1857 int newsize = 0; /* automatic sockbuf scaling */ 1858 1859 /* 1860 * This is a pure, in-sequence data packet with 1861 * nothing on the reassembly queue and we have enough 1862 * buffer space to take it. 1863 */ 1864 /* Clean receiver SACK report if present */ 1865 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1866 tcp_clean_sackreport(tp); 1867 TCPSTAT_INC(tcps_preddat); 1868 tp->rcv_nxt += tlen; 1869 /* 1870 * Pull snd_wl1 up to prevent seq wrap relative to 1871 * th_seq. 1872 */ 1873 tp->snd_wl1 = th->th_seq; 1874 /* 1875 * Pull rcv_up up to prevent seq wrap relative to 1876 * rcv_nxt. 1877 */ 1878 tp->rcv_up = tp->rcv_nxt; 1879 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1880 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1881 #ifdef TCPDEBUG 1882 if (so->so_options & SO_DEBUG) 1883 tcp_trace(TA_INPUT, ostate, tp, 1884 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1885 #endif 1886 TCP_PROBE3(debug__input, tp, th, m); 1887 1888 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1889 1890 /* Add data to socket buffer. */ 1891 SOCKBUF_LOCK(&so->so_rcv); 1892 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1893 m_freem(m); 1894 } else { 1895 /* 1896 * Set new socket buffer size. 1897 * Give up when limit is reached. 1898 */ 1899 if (newsize) 1900 if (!sbreserve_locked(&so->so_rcv, 1901 newsize, so, NULL)) 1902 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1903 m_adj(m, drop_hdrlen); /* delayed header drop */ 1904 sbappendstream_locked(&so->so_rcv, m, 0); 1905 } 1906 /* NB: sorwakeup_locked() does an implicit unlock. */ 1907 sorwakeup_locked(so); 1908 if (DELAY_ACK(tp, tlen)) { 1909 tp->t_flags |= TF_DELACK; 1910 } else { 1911 tp->t_flags |= TF_ACKNOW; 1912 tp->t_fb->tfb_tcp_output(tp); 1913 } 1914 goto check_delack; 1915 } 1916 } 1917 1918 /* 1919 * Calculate amount of space in receive window, 1920 * and then do TCP input processing. 1921 * Receive window is amount of space in rcv queue, 1922 * but not less than advertised window. 1923 */ 1924 win = sbspace(&so->so_rcv); 1925 if (win < 0) 1926 win = 0; 1927 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1928 1929 switch (tp->t_state) { 1930 1931 /* 1932 * If the state is SYN_RECEIVED: 1933 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1934 */ 1935 case TCPS_SYN_RECEIVED: 1936 if ((thflags & TH_ACK) && 1937 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1938 SEQ_GT(th->th_ack, tp->snd_max))) { 1939 rstreason = BANDLIM_RST_OPENPORT; 1940 goto dropwithreset; 1941 } 1942 if (IS_FASTOPEN(tp->t_flags)) { 1943 /* 1944 * When a TFO connection is in SYN_RECEIVED, the 1945 * only valid packets are the initial SYN, a 1946 * retransmit/copy of the initial SYN (possibly with 1947 * a subset of the original data), a valid ACK, a 1948 * FIN, or a RST. 1949 */ 1950 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1951 rstreason = BANDLIM_RST_OPENPORT; 1952 goto dropwithreset; 1953 } else if (thflags & TH_SYN) { 1954 /* non-initial SYN is ignored */ 1955 if ((tcp_timer_active(tp, TT_DELACK) || 1956 tcp_timer_active(tp, TT_REXMT))) 1957 goto drop; 1958 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1959 goto drop; 1960 } 1961 } 1962 break; 1963 1964 /* 1965 * If the state is SYN_SENT: 1966 * if seg contains a RST with valid ACK (SEQ.ACK has already 1967 * been verified), then drop the connection. 1968 * if seg contains a RST without an ACK, drop the seg. 1969 * if seg does not contain SYN, then drop the seg. 1970 * Otherwise this is an acceptable SYN segment 1971 * initialize tp->rcv_nxt and tp->irs 1972 * if seg contains ack then advance tp->snd_una 1973 * if seg contains an ECE and ECN support is enabled, the stream 1974 * is ECN capable. 1975 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1976 * arrange for segment to be acked (eventually) 1977 * continue processing rest of data/controls, beginning with URG 1978 */ 1979 case TCPS_SYN_SENT: 1980 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1981 TCP_PROBE5(connect__refused, NULL, tp, 1982 m, tp, th); 1983 tp = tcp_drop(tp, ECONNREFUSED); 1984 } 1985 if (thflags & TH_RST) 1986 goto drop; 1987 if (!(thflags & TH_SYN)) 1988 goto drop; 1989 1990 tp->irs = th->th_seq; 1991 tcp_rcvseqinit(tp); 1992 if (thflags & TH_ACK) { 1993 int tfo_partial_ack = 0; 1994 1995 TCPSTAT_INC(tcps_connects); 1996 soisconnected(so); 1997 #ifdef MAC 1998 mac_socketpeer_set_from_mbuf(m, so); 1999 #endif 2000 /* Do window scaling on this connection? */ 2001 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2002 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2003 tp->rcv_scale = tp->request_r_scale; 2004 } 2005 tp->rcv_adv += min(tp->rcv_wnd, 2006 TCP_MAXWIN << tp->rcv_scale); 2007 tp->snd_una++; /* SYN is acked */ 2008 /* 2009 * If not all the data that was sent in the TFO SYN 2010 * has been acked, resend the remainder right away. 2011 */ 2012 if (IS_FASTOPEN(tp->t_flags) && 2013 (tp->snd_una != tp->snd_max)) { 2014 tp->snd_nxt = th->th_ack; 2015 tfo_partial_ack = 1; 2016 } 2017 /* 2018 * If there's data, delay ACK; if there's also a FIN 2019 * ACKNOW will be turned on later. 2020 */ 2021 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2022 tcp_timer_activate(tp, TT_DELACK, 2023 tcp_delacktime); 2024 else 2025 tp->t_flags |= TF_ACKNOW; 2026 2027 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2028 tp->t_flags |= TF_ECN_PERMIT; 2029 TCPSTAT_INC(tcps_ecn_shs); 2030 } 2031 2032 /* 2033 * Received <SYN,ACK> in SYN_SENT[*] state. 2034 * Transitions: 2035 * SYN_SENT --> ESTABLISHED 2036 * SYN_SENT* --> FIN_WAIT_1 2037 */ 2038 tp->t_starttime = ticks; 2039 if (tp->t_flags & TF_NEEDFIN) { 2040 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2041 tp->t_flags &= ~TF_NEEDFIN; 2042 thflags &= ~TH_SYN; 2043 } else { 2044 tcp_state_change(tp, TCPS_ESTABLISHED); 2045 TCP_PROBE5(connect__established, NULL, tp, 2046 m, tp, th); 2047 cc_conn_init(tp); 2048 tcp_timer_activate(tp, TT_KEEP, 2049 TP_KEEPIDLE(tp)); 2050 } 2051 } else { 2052 /* 2053 * Received initial SYN in SYN-SENT[*] state => 2054 * simultaneous open. 2055 * If it succeeds, connection is * half-synchronized. 2056 * Otherwise, do 3-way handshake: 2057 * SYN-SENT -> SYN-RECEIVED 2058 * SYN-SENT* -> SYN-RECEIVED* 2059 */ 2060 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2061 tcp_timer_activate(tp, TT_REXMT, 0); 2062 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2063 } 2064 2065 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2066 INP_WLOCK_ASSERT(tp->t_inpcb); 2067 2068 /* 2069 * Advance th->th_seq to correspond to first data byte. 2070 * If data, trim to stay within window, 2071 * dropping FIN if necessary. 2072 */ 2073 th->th_seq++; 2074 if (tlen > tp->rcv_wnd) { 2075 todrop = tlen - tp->rcv_wnd; 2076 m_adj(m, -todrop); 2077 tlen = tp->rcv_wnd; 2078 thflags &= ~TH_FIN; 2079 TCPSTAT_INC(tcps_rcvpackafterwin); 2080 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2081 } 2082 tp->snd_wl1 = th->th_seq - 1; 2083 tp->rcv_up = th->th_seq; 2084 /* 2085 * Client side of transaction: already sent SYN and data. 2086 * If the remote host used T/TCP to validate the SYN, 2087 * our data will be ACK'd; if so, enter normal data segment 2088 * processing in the middle of step 5, ack processing. 2089 * Otherwise, goto step 6. 2090 */ 2091 if (thflags & TH_ACK) 2092 goto process_ACK; 2093 2094 goto step6; 2095 2096 /* 2097 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2098 * do normal processing. 2099 * 2100 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2101 */ 2102 case TCPS_LAST_ACK: 2103 case TCPS_CLOSING: 2104 break; /* continue normal processing */ 2105 } 2106 2107 /* 2108 * States other than LISTEN or SYN_SENT. 2109 * First check the RST flag and sequence number since reset segments 2110 * are exempt from the timestamp and connection count tests. This 2111 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2112 * below which allowed reset segments in half the sequence space 2113 * to fall though and be processed (which gives forged reset 2114 * segments with a random sequence number a 50 percent chance of 2115 * killing a connection). 2116 * Then check timestamp, if present. 2117 * Then check the connection count, if present. 2118 * Then check that at least some bytes of segment are within 2119 * receive window. If segment begins before rcv_nxt, 2120 * drop leading data (and SYN); if nothing left, just ack. 2121 */ 2122 if (thflags & TH_RST) { 2123 /* 2124 * RFC5961 Section 3.2 2125 * 2126 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2127 * - If RST is in window, we send challenge ACK. 2128 * 2129 * Note: to take into account delayed ACKs, we should 2130 * test against last_ack_sent instead of rcv_nxt. 2131 * Note 2: we handle special case of closed window, not 2132 * covered by the RFC. 2133 */ 2134 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2135 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2136 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2137 2138 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2139 KASSERT(tp->t_state != TCPS_SYN_SENT, 2140 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2141 __func__, th, tp)); 2142 2143 if (V_tcp_insecure_rst || 2144 tp->last_ack_sent == th->th_seq) { 2145 TCPSTAT_INC(tcps_drops); 2146 /* Drop the connection. */ 2147 switch (tp->t_state) { 2148 case TCPS_SYN_RECEIVED: 2149 so->so_error = ECONNREFUSED; 2150 goto close; 2151 case TCPS_ESTABLISHED: 2152 case TCPS_FIN_WAIT_1: 2153 case TCPS_FIN_WAIT_2: 2154 case TCPS_CLOSE_WAIT: 2155 case TCPS_CLOSING: 2156 case TCPS_LAST_ACK: 2157 so->so_error = ECONNRESET; 2158 close: 2159 /* FALLTHROUGH */ 2160 default: 2161 tp = tcp_close(tp); 2162 } 2163 } else { 2164 TCPSTAT_INC(tcps_badrst); 2165 /* Send challenge ACK. */ 2166 tcp_respond(tp, mtod(m, void *), th, m, 2167 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2168 tp->last_ack_sent = tp->rcv_nxt; 2169 m = NULL; 2170 } 2171 } 2172 goto drop; 2173 } 2174 2175 /* 2176 * RFC5961 Section 4.2 2177 * Send challenge ACK for any SYN in synchronized state. 2178 */ 2179 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2180 tp->t_state != TCPS_SYN_RECEIVED) { 2181 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2182 2183 TCPSTAT_INC(tcps_badsyn); 2184 if (V_tcp_insecure_syn && 2185 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2186 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2187 tp = tcp_drop(tp, ECONNRESET); 2188 rstreason = BANDLIM_UNLIMITED; 2189 } else { 2190 /* Send challenge ACK. */ 2191 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2192 tp->snd_nxt, TH_ACK); 2193 tp->last_ack_sent = tp->rcv_nxt; 2194 m = NULL; 2195 } 2196 goto drop; 2197 } 2198 2199 /* 2200 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2201 * and it's less than ts_recent, drop it. 2202 */ 2203 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2204 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2205 2206 /* Check to see if ts_recent is over 24 days old. */ 2207 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2208 /* 2209 * Invalidate ts_recent. If this segment updates 2210 * ts_recent, the age will be reset later and ts_recent 2211 * will get a valid value. If it does not, setting 2212 * ts_recent to zero will at least satisfy the 2213 * requirement that zero be placed in the timestamp 2214 * echo reply when ts_recent isn't valid. The 2215 * age isn't reset until we get a valid ts_recent 2216 * because we don't want out-of-order segments to be 2217 * dropped when ts_recent is old. 2218 */ 2219 tp->ts_recent = 0; 2220 } else { 2221 TCPSTAT_INC(tcps_rcvduppack); 2222 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2223 TCPSTAT_INC(tcps_pawsdrop); 2224 if (tlen) 2225 goto dropafterack; 2226 goto drop; 2227 } 2228 } 2229 2230 /* 2231 * In the SYN-RECEIVED state, validate that the packet belongs to 2232 * this connection before trimming the data to fit the receive 2233 * window. Check the sequence number versus IRS since we know 2234 * the sequence numbers haven't wrapped. This is a partial fix 2235 * for the "LAND" DoS attack. 2236 */ 2237 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2238 rstreason = BANDLIM_RST_OPENPORT; 2239 goto dropwithreset; 2240 } 2241 2242 todrop = tp->rcv_nxt - th->th_seq; 2243 if (todrop > 0) { 2244 if (thflags & TH_SYN) { 2245 thflags &= ~TH_SYN; 2246 th->th_seq++; 2247 if (th->th_urp > 1) 2248 th->th_urp--; 2249 else 2250 thflags &= ~TH_URG; 2251 todrop--; 2252 } 2253 /* 2254 * Following if statement from Stevens, vol. 2, p. 960. 2255 */ 2256 if (todrop > tlen 2257 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2258 /* 2259 * Any valid FIN must be to the left of the window. 2260 * At this point the FIN must be a duplicate or out 2261 * of sequence; drop it. 2262 */ 2263 thflags &= ~TH_FIN; 2264 2265 /* 2266 * Send an ACK to resynchronize and drop any data. 2267 * But keep on processing for RST or ACK. 2268 */ 2269 tp->t_flags |= TF_ACKNOW; 2270 todrop = tlen; 2271 TCPSTAT_INC(tcps_rcvduppack); 2272 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2273 } else { 2274 TCPSTAT_INC(tcps_rcvpartduppack); 2275 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2276 } 2277 drop_hdrlen += todrop; /* drop from the top afterwards */ 2278 th->th_seq += todrop; 2279 tlen -= todrop; 2280 if (th->th_urp > todrop) 2281 th->th_urp -= todrop; 2282 else { 2283 thflags &= ~TH_URG; 2284 th->th_urp = 0; 2285 } 2286 } 2287 2288 /* 2289 * If new data are received on a connection after the 2290 * user processes are gone, then RST the other end. 2291 */ 2292 if ((so->so_state & SS_NOFDREF) && 2293 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2294 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2295 2296 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2297 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2298 "after socket was closed, " 2299 "sending RST and removing tcpcb\n", 2300 s, __func__, tcpstates[tp->t_state], tlen); 2301 free(s, M_TCPLOG); 2302 } 2303 tp = tcp_close(tp); 2304 TCPSTAT_INC(tcps_rcvafterclose); 2305 rstreason = BANDLIM_UNLIMITED; 2306 goto dropwithreset; 2307 } 2308 2309 /* 2310 * If segment ends after window, drop trailing data 2311 * (and PUSH and FIN); if nothing left, just ACK. 2312 */ 2313 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2314 if (todrop > 0) { 2315 TCPSTAT_INC(tcps_rcvpackafterwin); 2316 if (todrop >= tlen) { 2317 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2318 /* 2319 * If window is closed can only take segments at 2320 * window edge, and have to drop data and PUSH from 2321 * incoming segments. Continue processing, but 2322 * remember to ack. Otherwise, drop segment 2323 * and ack. 2324 */ 2325 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2326 tp->t_flags |= TF_ACKNOW; 2327 TCPSTAT_INC(tcps_rcvwinprobe); 2328 } else 2329 goto dropafterack; 2330 } else 2331 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2332 m_adj(m, -todrop); 2333 tlen -= todrop; 2334 thflags &= ~(TH_PUSH|TH_FIN); 2335 } 2336 2337 /* 2338 * If last ACK falls within this segment's sequence numbers, 2339 * record its timestamp. 2340 * NOTE: 2341 * 1) That the test incorporates suggestions from the latest 2342 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2343 * 2) That updating only on newer timestamps interferes with 2344 * our earlier PAWS tests, so this check should be solely 2345 * predicated on the sequence space of this segment. 2346 * 3) That we modify the segment boundary check to be 2347 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2348 * instead of RFC1323's 2349 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2350 * This modified check allows us to overcome RFC1323's 2351 * limitations as described in Stevens TCP/IP Illustrated 2352 * Vol. 2 p.869. In such cases, we can still calculate the 2353 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2354 */ 2355 if ((to.to_flags & TOF_TS) != 0 && 2356 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2357 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2358 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2359 tp->ts_recent_age = tcp_ts_getticks(); 2360 tp->ts_recent = to.to_tsval; 2361 } 2362 2363 /* 2364 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2365 * flag is on (half-synchronized state), then queue data for 2366 * later processing; else drop segment and return. 2367 */ 2368 if ((thflags & TH_ACK) == 0) { 2369 if (tp->t_state == TCPS_SYN_RECEIVED || 2370 (tp->t_flags & TF_NEEDSYN)) { 2371 if (tp->t_state == TCPS_SYN_RECEIVED && 2372 IS_FASTOPEN(tp->t_flags)) { 2373 tp->snd_wnd = tiwin; 2374 cc_conn_init(tp); 2375 } 2376 goto step6; 2377 } else if (tp->t_flags & TF_ACKNOW) 2378 goto dropafterack; 2379 else 2380 goto drop; 2381 } 2382 2383 /* 2384 * Ack processing. 2385 */ 2386 switch (tp->t_state) { 2387 2388 /* 2389 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2390 * ESTABLISHED state and continue processing. 2391 * The ACK was checked above. 2392 */ 2393 case TCPS_SYN_RECEIVED: 2394 2395 TCPSTAT_INC(tcps_connects); 2396 soisconnected(so); 2397 /* Do window scaling? */ 2398 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2399 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2400 tp->rcv_scale = tp->request_r_scale; 2401 tp->snd_wnd = tiwin; 2402 } 2403 /* 2404 * Make transitions: 2405 * SYN-RECEIVED -> ESTABLISHED 2406 * SYN-RECEIVED* -> FIN-WAIT-1 2407 */ 2408 tp->t_starttime = ticks; 2409 if (tp->t_flags & TF_NEEDFIN) { 2410 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2411 tp->t_flags &= ~TF_NEEDFIN; 2412 } else { 2413 tcp_state_change(tp, TCPS_ESTABLISHED); 2414 TCP_PROBE5(accept__established, NULL, tp, 2415 m, tp, th); 2416 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2417 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2418 tp->t_tfo_pending = NULL; 2419 2420 /* 2421 * Account for the ACK of our SYN prior to 2422 * regular ACK processing below. 2423 */ 2424 tp->snd_una++; 2425 } 2426 /* 2427 * TFO connections call cc_conn_init() during SYN 2428 * processing. Calling it again here for such 2429 * connections is not harmless as it would undo the 2430 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2431 * is retransmitted. 2432 */ 2433 if (!IS_FASTOPEN(tp->t_flags)) 2434 cc_conn_init(tp); 2435 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2436 } 2437 /* 2438 * If segment contains data or ACK, will call tcp_reass() 2439 * later; if not, do so now to pass queued data to user. 2440 */ 2441 if (tlen == 0 && (thflags & TH_FIN) == 0) 2442 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2443 (struct mbuf *)0); 2444 tp->snd_wl1 = th->th_seq - 1; 2445 /* FALLTHROUGH */ 2446 2447 /* 2448 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2449 * ACKs. If the ack is in the range 2450 * tp->snd_una < th->th_ack <= tp->snd_max 2451 * then advance tp->snd_una to th->th_ack and drop 2452 * data from the retransmission queue. If this ACK reflects 2453 * more up to date window information we update our window information. 2454 */ 2455 case TCPS_ESTABLISHED: 2456 case TCPS_FIN_WAIT_1: 2457 case TCPS_FIN_WAIT_2: 2458 case TCPS_CLOSE_WAIT: 2459 case TCPS_CLOSING: 2460 case TCPS_LAST_ACK: 2461 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2462 TCPSTAT_INC(tcps_rcvacktoomuch); 2463 goto dropafterack; 2464 } 2465 if ((tp->t_flags & TF_SACK_PERMIT) && 2466 ((to.to_flags & TOF_SACK) || 2467 !TAILQ_EMPTY(&tp->snd_holes))) 2468 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2469 else 2470 /* 2471 * Reset the value so that previous (valid) value 2472 * from the last ack with SACK doesn't get used. 2473 */ 2474 tp->sackhint.sacked_bytes = 0; 2475 2476 #ifdef TCP_HHOOK 2477 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2478 hhook_run_tcp_est_in(tp, th, &to); 2479 #endif 2480 2481 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2482 u_int maxseg; 2483 2484 maxseg = tcp_maxseg(tp); 2485 if (tlen == 0 && 2486 (tiwin == tp->snd_wnd || 2487 (tp->t_flags & TF_SACK_PERMIT))) { 2488 /* 2489 * If this is the first time we've seen a 2490 * FIN from the remote, this is not a 2491 * duplicate and it needs to be processed 2492 * normally. This happens during a 2493 * simultaneous close. 2494 */ 2495 if ((thflags & TH_FIN) && 2496 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2497 tp->t_dupacks = 0; 2498 break; 2499 } 2500 TCPSTAT_INC(tcps_rcvdupack); 2501 /* 2502 * If we have outstanding data (other than 2503 * a window probe), this is a completely 2504 * duplicate ack (ie, window info didn't 2505 * change and FIN isn't set), 2506 * the ack is the biggest we've 2507 * seen and we've seen exactly our rexmt 2508 * threshold of them, assume a packet 2509 * has been dropped and retransmit it. 2510 * Kludge snd_nxt & the congestion 2511 * window so we send only this one 2512 * packet. 2513 * 2514 * We know we're losing at the current 2515 * window size so do congestion avoidance 2516 * (set ssthresh to half the current window 2517 * and pull our congestion window back to 2518 * the new ssthresh). 2519 * 2520 * Dup acks mean that packets have left the 2521 * network (they're now cached at the receiver) 2522 * so bump cwnd by the amount in the receiver 2523 * to keep a constant cwnd packets in the 2524 * network. 2525 * 2526 * When using TCP ECN, notify the peer that 2527 * we reduced the cwnd. 2528 */ 2529 /* 2530 * Following 2 kinds of acks should not affect 2531 * dupack counting: 2532 * 1) Old acks 2533 * 2) Acks with SACK but without any new SACK 2534 * information in them. These could result from 2535 * any anomaly in the network like a switch 2536 * duplicating packets or a possible DoS attack. 2537 */ 2538 if (th->th_ack != tp->snd_una || 2539 ((tp->t_flags & TF_SACK_PERMIT) && 2540 !sack_changed)) 2541 break; 2542 else if (!tcp_timer_active(tp, TT_REXMT)) 2543 tp->t_dupacks = 0; 2544 else if (++tp->t_dupacks > tcprexmtthresh || 2545 IN_FASTRECOVERY(tp->t_flags)) { 2546 cc_ack_received(tp, th, nsegs, 2547 CC_DUPACK); 2548 if ((tp->t_flags & TF_SACK_PERMIT) && 2549 IN_FASTRECOVERY(tp->t_flags)) { 2550 int awnd; 2551 2552 /* 2553 * Compute the amount of data in flight first. 2554 * We can inject new data into the pipe iff 2555 * we have less than 1/2 the original window's 2556 * worth of data in flight. 2557 */ 2558 if (V_tcp_do_rfc6675_pipe) 2559 awnd = tcp_compute_pipe(tp); 2560 else 2561 awnd = (tp->snd_nxt - tp->snd_fack) + 2562 tp->sackhint.sack_bytes_rexmit; 2563 2564 if (awnd < tp->snd_ssthresh) { 2565 tp->snd_cwnd += maxseg; 2566 if (tp->snd_cwnd > tp->snd_ssthresh) 2567 tp->snd_cwnd = tp->snd_ssthresh; 2568 } 2569 } else 2570 tp->snd_cwnd += maxseg; 2571 (void) tp->t_fb->tfb_tcp_output(tp); 2572 goto drop; 2573 } else if (tp->t_dupacks == tcprexmtthresh) { 2574 tcp_seq onxt = tp->snd_nxt; 2575 2576 /* 2577 * If we're doing sack, check to 2578 * see if we're already in sack 2579 * recovery. If we're not doing sack, 2580 * check to see if we're in newreno 2581 * recovery. 2582 */ 2583 if (tp->t_flags & TF_SACK_PERMIT) { 2584 if (IN_FASTRECOVERY(tp->t_flags)) { 2585 tp->t_dupacks = 0; 2586 break; 2587 } 2588 } else { 2589 if (SEQ_LEQ(th->th_ack, 2590 tp->snd_recover)) { 2591 tp->t_dupacks = 0; 2592 break; 2593 } 2594 } 2595 /* Congestion signal before ack. */ 2596 cc_cong_signal(tp, th, CC_NDUPACK); 2597 cc_ack_received(tp, th, nsegs, 2598 CC_DUPACK); 2599 tcp_timer_activate(tp, TT_REXMT, 0); 2600 tp->t_rtttime = 0; 2601 if (tp->t_flags & TF_SACK_PERMIT) { 2602 TCPSTAT_INC( 2603 tcps_sack_recovery_episode); 2604 tp->sack_newdata = tp->snd_nxt; 2605 tp->snd_cwnd = maxseg; 2606 (void) tp->t_fb->tfb_tcp_output(tp); 2607 goto drop; 2608 } 2609 tp->snd_nxt = th->th_ack; 2610 tp->snd_cwnd = maxseg; 2611 (void) tp->t_fb->tfb_tcp_output(tp); 2612 KASSERT(tp->snd_limited <= 2, 2613 ("%s: tp->snd_limited too big", 2614 __func__)); 2615 tp->snd_cwnd = tp->snd_ssthresh + 2616 maxseg * 2617 (tp->t_dupacks - tp->snd_limited); 2618 if (SEQ_GT(onxt, tp->snd_nxt)) 2619 tp->snd_nxt = onxt; 2620 goto drop; 2621 } else if (V_tcp_do_rfc3042) { 2622 /* 2623 * Process first and second duplicate 2624 * ACKs. Each indicates a segment 2625 * leaving the network, creating room 2626 * for more. Make sure we can send a 2627 * packet on reception of each duplicate 2628 * ACK by increasing snd_cwnd by one 2629 * segment. Restore the original 2630 * snd_cwnd after packet transmission. 2631 */ 2632 cc_ack_received(tp, th, nsegs, 2633 CC_DUPACK); 2634 uint32_t oldcwnd = tp->snd_cwnd; 2635 tcp_seq oldsndmax = tp->snd_max; 2636 u_int sent; 2637 int avail; 2638 2639 KASSERT(tp->t_dupacks == 1 || 2640 tp->t_dupacks == 2, 2641 ("%s: dupacks not 1 or 2", 2642 __func__)); 2643 if (tp->t_dupacks == 1) 2644 tp->snd_limited = 0; 2645 tp->snd_cwnd = 2646 (tp->snd_nxt - tp->snd_una) + 2647 (tp->t_dupacks - tp->snd_limited) * 2648 maxseg; 2649 /* 2650 * Only call tcp_output when there 2651 * is new data available to be sent. 2652 * Otherwise we would send pure ACKs. 2653 */ 2654 SOCKBUF_LOCK(&so->so_snd); 2655 avail = sbavail(&so->so_snd) - 2656 (tp->snd_nxt - tp->snd_una); 2657 SOCKBUF_UNLOCK(&so->so_snd); 2658 if (avail > 0) 2659 (void) tp->t_fb->tfb_tcp_output(tp); 2660 sent = tp->snd_max - oldsndmax; 2661 if (sent > maxseg) { 2662 KASSERT((tp->t_dupacks == 2 && 2663 tp->snd_limited == 0) || 2664 (sent == maxseg + 1 && 2665 tp->t_flags & TF_SENTFIN), 2666 ("%s: sent too much", 2667 __func__)); 2668 tp->snd_limited = 2; 2669 } else if (sent > 0) 2670 ++tp->snd_limited; 2671 tp->snd_cwnd = oldcwnd; 2672 goto drop; 2673 } 2674 } 2675 break; 2676 } else { 2677 /* 2678 * This ack is advancing the left edge, reset the 2679 * counter. 2680 */ 2681 tp->t_dupacks = 0; 2682 /* 2683 * If this ack also has new SACK info, increment the 2684 * counter as per rfc6675. 2685 */ 2686 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2687 tp->t_dupacks++; 2688 } 2689 2690 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2691 ("%s: th_ack <= snd_una", __func__)); 2692 2693 /* 2694 * If the congestion window was inflated to account 2695 * for the other side's cached packets, retract it. 2696 */ 2697 if (IN_FASTRECOVERY(tp->t_flags)) { 2698 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2699 if (tp->t_flags & TF_SACK_PERMIT) 2700 tcp_sack_partialack(tp, th); 2701 else 2702 tcp_newreno_partial_ack(tp, th); 2703 } else 2704 cc_post_recovery(tp, th); 2705 } 2706 /* 2707 * If we reach this point, ACK is not a duplicate, 2708 * i.e., it ACKs something we sent. 2709 */ 2710 if (tp->t_flags & TF_NEEDSYN) { 2711 /* 2712 * T/TCP: Connection was half-synchronized, and our 2713 * SYN has been ACK'd (so connection is now fully 2714 * synchronized). Go to non-starred state, 2715 * increment snd_una for ACK of SYN, and check if 2716 * we can do window scaling. 2717 */ 2718 tp->t_flags &= ~TF_NEEDSYN; 2719 tp->snd_una++; 2720 /* Do window scaling? */ 2721 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2722 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2723 tp->rcv_scale = tp->request_r_scale; 2724 /* Send window already scaled. */ 2725 } 2726 } 2727 2728 process_ACK: 2729 INP_WLOCK_ASSERT(tp->t_inpcb); 2730 2731 acked = BYTES_THIS_ACK(tp, th); 2732 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2733 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2734 tp->snd_una, th->th_ack, tp, m)); 2735 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2736 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2737 2738 /* 2739 * If we just performed our first retransmit, and the ACK 2740 * arrives within our recovery window, then it was a mistake 2741 * to do the retransmit in the first place. Recover our 2742 * original cwnd and ssthresh, and proceed to transmit where 2743 * we left off. 2744 */ 2745 if (tp->t_rxtshift == 1 && 2746 tp->t_flags & TF_PREVVALID && 2747 tp->t_badrxtwin && 2748 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2749 cc_cong_signal(tp, th, CC_RTO_ERR); 2750 2751 /* 2752 * If we have a timestamp reply, update smoothed 2753 * round trip time. If no timestamp is present but 2754 * transmit timer is running and timed sequence 2755 * number was acked, update smoothed round trip time. 2756 * Since we now have an rtt measurement, cancel the 2757 * timer backoff (cf., Phil Karn's retransmit alg.). 2758 * Recompute the initial retransmit timer. 2759 * 2760 * Some boxes send broken timestamp replies 2761 * during the SYN+ACK phase, ignore 2762 * timestamps of 0 or we could calculate a 2763 * huge RTT and blow up the retransmit timer. 2764 */ 2765 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2766 uint32_t t; 2767 2768 t = tcp_ts_getticks() - to.to_tsecr; 2769 if (!tp->t_rttlow || tp->t_rttlow > t) 2770 tp->t_rttlow = t; 2771 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2772 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2773 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2774 tp->t_rttlow = ticks - tp->t_rtttime; 2775 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2776 } 2777 2778 /* 2779 * If all outstanding data is acked, stop retransmit 2780 * timer and remember to restart (more output or persist). 2781 * If there is more data to be acked, restart retransmit 2782 * timer, using current (possibly backed-off) value. 2783 */ 2784 if (th->th_ack == tp->snd_max) { 2785 tcp_timer_activate(tp, TT_REXMT, 0); 2786 needoutput = 1; 2787 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2788 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2789 2790 /* 2791 * If no data (only SYN) was ACK'd, 2792 * skip rest of ACK processing. 2793 */ 2794 if (acked == 0) 2795 goto step6; 2796 2797 /* 2798 * Let the congestion control algorithm update congestion 2799 * control related information. This typically means increasing 2800 * the congestion window. 2801 */ 2802 cc_ack_received(tp, th, nsegs, CC_ACK); 2803 2804 SOCKBUF_LOCK(&so->so_snd); 2805 if (acked > sbavail(&so->so_snd)) { 2806 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2807 tp->snd_wnd -= sbavail(&so->so_snd); 2808 else 2809 tp->snd_wnd = 0; 2810 mfree = sbcut_locked(&so->so_snd, 2811 (int)sbavail(&so->so_snd)); 2812 ourfinisacked = 1; 2813 } else { 2814 mfree = sbcut_locked(&so->so_snd, acked); 2815 if (tp->snd_wnd >= (uint32_t) acked) 2816 tp->snd_wnd -= acked; 2817 else 2818 tp->snd_wnd = 0; 2819 ourfinisacked = 0; 2820 } 2821 /* NB: sowwakeup_locked() does an implicit unlock. */ 2822 sowwakeup_locked(so); 2823 m_freem(mfree); 2824 /* Detect una wraparound. */ 2825 if (!IN_RECOVERY(tp->t_flags) && 2826 SEQ_GT(tp->snd_una, tp->snd_recover) && 2827 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2828 tp->snd_recover = th->th_ack - 1; 2829 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2830 if (IN_RECOVERY(tp->t_flags) && 2831 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2832 EXIT_RECOVERY(tp->t_flags); 2833 } 2834 tp->snd_una = th->th_ack; 2835 if (tp->t_flags & TF_SACK_PERMIT) { 2836 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2837 tp->snd_recover = tp->snd_una; 2838 } 2839 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2840 tp->snd_nxt = tp->snd_una; 2841 2842 switch (tp->t_state) { 2843 2844 /* 2845 * In FIN_WAIT_1 STATE in addition to the processing 2846 * for the ESTABLISHED state if our FIN is now acknowledged 2847 * then enter FIN_WAIT_2. 2848 */ 2849 case TCPS_FIN_WAIT_1: 2850 if (ourfinisacked) { 2851 /* 2852 * If we can't receive any more 2853 * data, then closing user can proceed. 2854 * Starting the timer is contrary to the 2855 * specification, but if we don't get a FIN 2856 * we'll hang forever. 2857 * 2858 * XXXjl: 2859 * we should release the tp also, and use a 2860 * compressed state. 2861 */ 2862 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2863 soisdisconnected(so); 2864 tcp_timer_activate(tp, TT_2MSL, 2865 (tcp_fast_finwait2_recycle ? 2866 tcp_finwait2_timeout : 2867 TP_MAXIDLE(tp))); 2868 } 2869 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2870 } 2871 break; 2872 2873 /* 2874 * In CLOSING STATE in addition to the processing for 2875 * the ESTABLISHED state if the ACK acknowledges our FIN 2876 * then enter the TIME-WAIT state, otherwise ignore 2877 * the segment. 2878 */ 2879 case TCPS_CLOSING: 2880 if (ourfinisacked) { 2881 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2882 tcp_twstart(tp); 2883 m_freem(m); 2884 return; 2885 } 2886 break; 2887 2888 /* 2889 * In LAST_ACK, we may still be waiting for data to drain 2890 * and/or to be acked, as well as for the ack of our FIN. 2891 * If our FIN is now acknowledged, delete the TCB, 2892 * enter the closed state and return. 2893 */ 2894 case TCPS_LAST_ACK: 2895 if (ourfinisacked) { 2896 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2897 tp = tcp_close(tp); 2898 goto drop; 2899 } 2900 break; 2901 } 2902 } 2903 2904 step6: 2905 INP_WLOCK_ASSERT(tp->t_inpcb); 2906 2907 /* 2908 * Update window information. 2909 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2910 */ 2911 if ((thflags & TH_ACK) && 2912 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2913 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2914 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2915 /* keep track of pure window updates */ 2916 if (tlen == 0 && 2917 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2918 TCPSTAT_INC(tcps_rcvwinupd); 2919 tp->snd_wnd = tiwin; 2920 tp->snd_wl1 = th->th_seq; 2921 tp->snd_wl2 = th->th_ack; 2922 if (tp->snd_wnd > tp->max_sndwnd) 2923 tp->max_sndwnd = tp->snd_wnd; 2924 needoutput = 1; 2925 } 2926 2927 /* 2928 * Process segments with URG. 2929 */ 2930 if ((thflags & TH_URG) && th->th_urp && 2931 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2932 /* 2933 * This is a kludge, but if we receive and accept 2934 * random urgent pointers, we'll crash in 2935 * soreceive. It's hard to imagine someone 2936 * actually wanting to send this much urgent data. 2937 */ 2938 SOCKBUF_LOCK(&so->so_rcv); 2939 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2940 th->th_urp = 0; /* XXX */ 2941 thflags &= ~TH_URG; /* XXX */ 2942 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2943 goto dodata; /* XXX */ 2944 } 2945 /* 2946 * If this segment advances the known urgent pointer, 2947 * then mark the data stream. This should not happen 2948 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2949 * a FIN has been received from the remote side. 2950 * In these states we ignore the URG. 2951 * 2952 * According to RFC961 (Assigned Protocols), 2953 * the urgent pointer points to the last octet 2954 * of urgent data. We continue, however, 2955 * to consider it to indicate the first octet 2956 * of data past the urgent section as the original 2957 * spec states (in one of two places). 2958 */ 2959 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2960 tp->rcv_up = th->th_seq + th->th_urp; 2961 so->so_oobmark = sbavail(&so->so_rcv) + 2962 (tp->rcv_up - tp->rcv_nxt) - 1; 2963 if (so->so_oobmark == 0) 2964 so->so_rcv.sb_state |= SBS_RCVATMARK; 2965 sohasoutofband(so); 2966 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2967 } 2968 SOCKBUF_UNLOCK(&so->so_rcv); 2969 /* 2970 * Remove out of band data so doesn't get presented to user. 2971 * This can happen independent of advancing the URG pointer, 2972 * but if two URG's are pending at once, some out-of-band 2973 * data may creep in... ick. 2974 */ 2975 if (th->th_urp <= (uint32_t)tlen && 2976 !(so->so_options & SO_OOBINLINE)) { 2977 /* hdr drop is delayed */ 2978 tcp_pulloutofband(so, th, m, drop_hdrlen); 2979 } 2980 } else { 2981 /* 2982 * If no out of band data is expected, 2983 * pull receive urgent pointer along 2984 * with the receive window. 2985 */ 2986 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2987 tp->rcv_up = tp->rcv_nxt; 2988 } 2989 dodata: /* XXX */ 2990 INP_WLOCK_ASSERT(tp->t_inpcb); 2991 2992 /* 2993 * Process the segment text, merging it into the TCP sequencing queue, 2994 * and arranging for acknowledgment of receipt if necessary. 2995 * This process logically involves adjusting tp->rcv_wnd as data 2996 * is presented to the user (this happens in tcp_usrreq.c, 2997 * case PRU_RCVD). If a FIN has already been received on this 2998 * connection then we just ignore the text. 2999 */ 3000 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3001 IS_FASTOPEN(tp->t_flags)); 3002 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3003 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3004 tcp_seq save_start = th->th_seq; 3005 m_adj(m, drop_hdrlen); /* delayed header drop */ 3006 /* 3007 * Insert segment which includes th into TCP reassembly queue 3008 * with control block tp. Set thflags to whether reassembly now 3009 * includes a segment with FIN. This handles the common case 3010 * inline (segment is the next to be received on an established 3011 * connection, and the queue is empty), avoiding linkage into 3012 * and removal from the queue and repetition of various 3013 * conversions. 3014 * Set DELACK for segments received in order, but ack 3015 * immediately when segments are out of order (so 3016 * fast retransmit can work). 3017 */ 3018 if (th->th_seq == tp->rcv_nxt && 3019 LIST_EMPTY(&tp->t_segq) && 3020 (TCPS_HAVEESTABLISHED(tp->t_state) || 3021 tfo_syn)) { 3022 if (DELAY_ACK(tp, tlen) || tfo_syn) 3023 tp->t_flags |= TF_DELACK; 3024 else 3025 tp->t_flags |= TF_ACKNOW; 3026 tp->rcv_nxt += tlen; 3027 thflags = th->th_flags & TH_FIN; 3028 TCPSTAT_INC(tcps_rcvpack); 3029 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3030 SOCKBUF_LOCK(&so->so_rcv); 3031 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3032 m_freem(m); 3033 else 3034 sbappendstream_locked(&so->so_rcv, m, 0); 3035 /* NB: sorwakeup_locked() does an implicit unlock. */ 3036 sorwakeup_locked(so); 3037 } else { 3038 /* 3039 * XXX: Due to the header drop above "th" is 3040 * theoretically invalid by now. Fortunately 3041 * m_adj() doesn't actually frees any mbufs 3042 * when trimming from the head. 3043 */ 3044 thflags = tcp_reass(tp, th, &tlen, m); 3045 tp->t_flags |= TF_ACKNOW; 3046 } 3047 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3048 tcp_update_sack_list(tp, save_start, save_start + tlen); 3049 #if 0 3050 /* 3051 * Note the amount of data that peer has sent into 3052 * our window, in order to estimate the sender's 3053 * buffer size. 3054 * XXX: Unused. 3055 */ 3056 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3057 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3058 else 3059 len = so->so_rcv.sb_hiwat; 3060 #endif 3061 } else { 3062 m_freem(m); 3063 thflags &= ~TH_FIN; 3064 } 3065 3066 /* 3067 * If FIN is received ACK the FIN and let the user know 3068 * that the connection is closing. 3069 */ 3070 if (thflags & TH_FIN) { 3071 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3072 socantrcvmore(so); 3073 /* 3074 * If connection is half-synchronized 3075 * (ie NEEDSYN flag on) then delay ACK, 3076 * so it may be piggybacked when SYN is sent. 3077 * Otherwise, since we received a FIN then no 3078 * more input can be expected, send ACK now. 3079 */ 3080 if (tp->t_flags & TF_NEEDSYN) 3081 tp->t_flags |= TF_DELACK; 3082 else 3083 tp->t_flags |= TF_ACKNOW; 3084 tp->rcv_nxt++; 3085 } 3086 switch (tp->t_state) { 3087 3088 /* 3089 * In SYN_RECEIVED and ESTABLISHED STATES 3090 * enter the CLOSE_WAIT state. 3091 */ 3092 case TCPS_SYN_RECEIVED: 3093 tp->t_starttime = ticks; 3094 /* FALLTHROUGH */ 3095 case TCPS_ESTABLISHED: 3096 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3097 break; 3098 3099 /* 3100 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3101 * enter the CLOSING state. 3102 */ 3103 case TCPS_FIN_WAIT_1: 3104 tcp_state_change(tp, TCPS_CLOSING); 3105 break; 3106 3107 /* 3108 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3109 * starting the time-wait timer, turning off the other 3110 * standard timers. 3111 */ 3112 case TCPS_FIN_WAIT_2: 3113 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3114 3115 tcp_twstart(tp); 3116 return; 3117 } 3118 } 3119 #ifdef TCPDEBUG 3120 if (so->so_options & SO_DEBUG) 3121 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3122 &tcp_savetcp, 0); 3123 #endif 3124 TCP_PROBE3(debug__input, tp, th, m); 3125 3126 /* 3127 * Return any desired output. 3128 */ 3129 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3130 (void) tp->t_fb->tfb_tcp_output(tp); 3131 3132 check_delack: 3133 INP_WLOCK_ASSERT(tp->t_inpcb); 3134 3135 if (tp->t_flags & TF_DELACK) { 3136 tp->t_flags &= ~TF_DELACK; 3137 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3138 } 3139 INP_WUNLOCK(tp->t_inpcb); 3140 return; 3141 3142 dropafterack: 3143 /* 3144 * Generate an ACK dropping incoming segment if it occupies 3145 * sequence space, where the ACK reflects our state. 3146 * 3147 * We can now skip the test for the RST flag since all 3148 * paths to this code happen after packets containing 3149 * RST have been dropped. 3150 * 3151 * In the SYN-RECEIVED state, don't send an ACK unless the 3152 * segment we received passes the SYN-RECEIVED ACK test. 3153 * If it fails send a RST. This breaks the loop in the 3154 * "LAND" DoS attack, and also prevents an ACK storm 3155 * between two listening ports that have been sent forged 3156 * SYN segments, each with the source address of the other. 3157 */ 3158 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3159 (SEQ_GT(tp->snd_una, th->th_ack) || 3160 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3161 rstreason = BANDLIM_RST_OPENPORT; 3162 goto dropwithreset; 3163 } 3164 #ifdef TCPDEBUG 3165 if (so->so_options & SO_DEBUG) 3166 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3167 &tcp_savetcp, 0); 3168 #endif 3169 TCP_PROBE3(debug__input, tp, th, m); 3170 tp->t_flags |= TF_ACKNOW; 3171 (void) tp->t_fb->tfb_tcp_output(tp); 3172 INP_WUNLOCK(tp->t_inpcb); 3173 m_freem(m); 3174 return; 3175 3176 dropwithreset: 3177 if (tp != NULL) { 3178 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3179 INP_WUNLOCK(tp->t_inpcb); 3180 } else 3181 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3182 return; 3183 3184 drop: 3185 /* 3186 * Drop space held by incoming segment and return. 3187 */ 3188 #ifdef TCPDEBUG 3189 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3190 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3191 &tcp_savetcp, 0); 3192 #endif 3193 TCP_PROBE3(debug__input, tp, th, m); 3194 if (tp != NULL) 3195 INP_WUNLOCK(tp->t_inpcb); 3196 m_freem(m); 3197 } 3198 3199 /* 3200 * Issue RST and make ACK acceptable to originator of segment. 3201 * The mbuf must still include the original packet header. 3202 * tp may be NULL. 3203 */ 3204 void 3205 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3206 int tlen, int rstreason) 3207 { 3208 #ifdef INET 3209 struct ip *ip; 3210 #endif 3211 #ifdef INET6 3212 struct ip6_hdr *ip6; 3213 #endif 3214 3215 if (tp != NULL) { 3216 INP_WLOCK_ASSERT(tp->t_inpcb); 3217 } 3218 3219 /* Don't bother if destination was broadcast/multicast. */ 3220 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3221 goto drop; 3222 #ifdef INET6 3223 if (mtod(m, struct ip *)->ip_v == 6) { 3224 ip6 = mtod(m, struct ip6_hdr *); 3225 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3226 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3227 goto drop; 3228 /* IPv6 anycast check is done at tcp6_input() */ 3229 } 3230 #endif 3231 #if defined(INET) && defined(INET6) 3232 else 3233 #endif 3234 #ifdef INET 3235 { 3236 ip = mtod(m, struct ip *); 3237 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3238 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3239 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3240 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3241 goto drop; 3242 } 3243 #endif 3244 3245 /* Perform bandwidth limiting. */ 3246 if (badport_bandlim(rstreason) < 0) 3247 goto drop; 3248 3249 /* tcp_respond consumes the mbuf chain. */ 3250 if (th->th_flags & TH_ACK) { 3251 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3252 th->th_ack, TH_RST); 3253 } else { 3254 if (th->th_flags & TH_SYN) 3255 tlen++; 3256 if (th->th_flags & TH_FIN) 3257 tlen++; 3258 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3259 (tcp_seq)0, TH_RST|TH_ACK); 3260 } 3261 return; 3262 drop: 3263 m_freem(m); 3264 } 3265 3266 /* 3267 * Parse TCP options and place in tcpopt. 3268 */ 3269 void 3270 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3271 { 3272 int opt, optlen; 3273 3274 to->to_flags = 0; 3275 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3276 opt = cp[0]; 3277 if (opt == TCPOPT_EOL) 3278 break; 3279 if (opt == TCPOPT_NOP) 3280 optlen = 1; 3281 else { 3282 if (cnt < 2) 3283 break; 3284 optlen = cp[1]; 3285 if (optlen < 2 || optlen > cnt) 3286 break; 3287 } 3288 switch (opt) { 3289 case TCPOPT_MAXSEG: 3290 if (optlen != TCPOLEN_MAXSEG) 3291 continue; 3292 if (!(flags & TO_SYN)) 3293 continue; 3294 to->to_flags |= TOF_MSS; 3295 bcopy((char *)cp + 2, 3296 (char *)&to->to_mss, sizeof(to->to_mss)); 3297 to->to_mss = ntohs(to->to_mss); 3298 break; 3299 case TCPOPT_WINDOW: 3300 if (optlen != TCPOLEN_WINDOW) 3301 continue; 3302 if (!(flags & TO_SYN)) 3303 continue; 3304 to->to_flags |= TOF_SCALE; 3305 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3306 break; 3307 case TCPOPT_TIMESTAMP: 3308 if (optlen != TCPOLEN_TIMESTAMP) 3309 continue; 3310 to->to_flags |= TOF_TS; 3311 bcopy((char *)cp + 2, 3312 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3313 to->to_tsval = ntohl(to->to_tsval); 3314 bcopy((char *)cp + 6, 3315 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3316 to->to_tsecr = ntohl(to->to_tsecr); 3317 break; 3318 case TCPOPT_SIGNATURE: 3319 /* 3320 * In order to reply to a host which has set the 3321 * TCP_SIGNATURE option in its initial SYN, we have 3322 * to record the fact that the option was observed 3323 * here for the syncache code to perform the correct 3324 * response. 3325 */ 3326 if (optlen != TCPOLEN_SIGNATURE) 3327 continue; 3328 to->to_flags |= TOF_SIGNATURE; 3329 to->to_signature = cp + 2; 3330 break; 3331 case TCPOPT_SACK_PERMITTED: 3332 if (optlen != TCPOLEN_SACK_PERMITTED) 3333 continue; 3334 if (!(flags & TO_SYN)) 3335 continue; 3336 if (!V_tcp_do_sack) 3337 continue; 3338 to->to_flags |= TOF_SACKPERM; 3339 break; 3340 case TCPOPT_SACK: 3341 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3342 continue; 3343 if (flags & TO_SYN) 3344 continue; 3345 to->to_flags |= TOF_SACK; 3346 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3347 to->to_sacks = cp + 2; 3348 TCPSTAT_INC(tcps_sack_rcv_blocks); 3349 break; 3350 case TCPOPT_FAST_OPEN: 3351 /* 3352 * Cookie length validation is performed by the 3353 * server side cookie checking code or the client 3354 * side cookie cache update code. 3355 */ 3356 if (!(flags & TO_SYN)) 3357 continue; 3358 if (!V_tcp_fastopen_client_enable && 3359 !V_tcp_fastopen_server_enable) 3360 continue; 3361 to->to_flags |= TOF_FASTOPEN; 3362 to->to_tfo_len = optlen - 2; 3363 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3364 break; 3365 default: 3366 continue; 3367 } 3368 } 3369 } 3370 3371 /* 3372 * Pull out of band byte out of a segment so 3373 * it doesn't appear in the user's data queue. 3374 * It is still reflected in the segment length for 3375 * sequencing purposes. 3376 */ 3377 void 3378 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3379 int off) 3380 { 3381 int cnt = off + th->th_urp - 1; 3382 3383 while (cnt >= 0) { 3384 if (m->m_len > cnt) { 3385 char *cp = mtod(m, caddr_t) + cnt; 3386 struct tcpcb *tp = sototcpcb(so); 3387 3388 INP_WLOCK_ASSERT(tp->t_inpcb); 3389 3390 tp->t_iobc = *cp; 3391 tp->t_oobflags |= TCPOOB_HAVEDATA; 3392 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3393 m->m_len--; 3394 if (m->m_flags & M_PKTHDR) 3395 m->m_pkthdr.len--; 3396 return; 3397 } 3398 cnt -= m->m_len; 3399 m = m->m_next; 3400 if (m == NULL) 3401 break; 3402 } 3403 panic("tcp_pulloutofband"); 3404 } 3405 3406 /* 3407 * Collect new round-trip time estimate 3408 * and update averages and current timeout. 3409 */ 3410 void 3411 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3412 { 3413 int delta; 3414 3415 INP_WLOCK_ASSERT(tp->t_inpcb); 3416 3417 TCPSTAT_INC(tcps_rttupdated); 3418 tp->t_rttupdated++; 3419 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3420 /* 3421 * srtt is stored as fixed point with 5 bits after the 3422 * binary point (i.e., scaled by 8). The following magic 3423 * is equivalent to the smoothing algorithm in rfc793 with 3424 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3425 * point). Adjust rtt to origin 0. 3426 */ 3427 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3428 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3429 3430 if ((tp->t_srtt += delta) <= 0) 3431 tp->t_srtt = 1; 3432 3433 /* 3434 * We accumulate a smoothed rtt variance (actually, a 3435 * smoothed mean difference), then set the retransmit 3436 * timer to smoothed rtt + 4 times the smoothed variance. 3437 * rttvar is stored as fixed point with 4 bits after the 3438 * binary point (scaled by 16). The following is 3439 * equivalent to rfc793 smoothing with an alpha of .75 3440 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3441 * rfc793's wired-in beta. 3442 */ 3443 if (delta < 0) 3444 delta = -delta; 3445 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3446 if ((tp->t_rttvar += delta) <= 0) 3447 tp->t_rttvar = 1; 3448 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3449 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3450 } else { 3451 /* 3452 * No rtt measurement yet - use the unsmoothed rtt. 3453 * Set the variance to half the rtt (so our first 3454 * retransmit happens at 3*rtt). 3455 */ 3456 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3457 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3458 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3459 } 3460 tp->t_rtttime = 0; 3461 tp->t_rxtshift = 0; 3462 3463 /* 3464 * the retransmit should happen at rtt + 4 * rttvar. 3465 * Because of the way we do the smoothing, srtt and rttvar 3466 * will each average +1/2 tick of bias. When we compute 3467 * the retransmit timer, we want 1/2 tick of rounding and 3468 * 1 extra tick because of +-1/2 tick uncertainty in the 3469 * firing of the timer. The bias will give us exactly the 3470 * 1.5 tick we need. But, because the bias is 3471 * statistical, we have to test that we don't drop below 3472 * the minimum feasible timer (which is 2 ticks). 3473 */ 3474 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3475 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3476 3477 /* 3478 * We received an ack for a packet that wasn't retransmitted; 3479 * it is probably safe to discard any error indications we've 3480 * received recently. This isn't quite right, but close enough 3481 * for now (a route might have failed after we sent a segment, 3482 * and the return path might not be symmetrical). 3483 */ 3484 tp->t_softerror = 0; 3485 } 3486 3487 /* 3488 * Determine a reasonable value for maxseg size. 3489 * If the route is known, check route for mtu. 3490 * If none, use an mss that can be handled on the outgoing interface 3491 * without forcing IP to fragment. If no route is found, route has no mtu, 3492 * or the destination isn't local, use a default, hopefully conservative 3493 * size (usually 512 or the default IP max size, but no more than the mtu 3494 * of the interface), as we can't discover anything about intervening 3495 * gateways or networks. We also initialize the congestion/slow start 3496 * window to be a single segment if the destination isn't local. 3497 * While looking at the routing entry, we also initialize other path-dependent 3498 * parameters from pre-set or cached values in the routing entry. 3499 * 3500 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3501 * IP options, e.g. IPSEC data, since length of this data may vary, and 3502 * thus it is calculated for every segment separately in tcp_output(). 3503 * 3504 * NOTE that this routine is only called when we process an incoming 3505 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3506 * settings are handled in tcp_mssopt(). 3507 */ 3508 void 3509 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3510 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3511 { 3512 int mss = 0; 3513 uint32_t maxmtu = 0; 3514 struct inpcb *inp = tp->t_inpcb; 3515 struct hc_metrics_lite metrics; 3516 #ifdef INET6 3517 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3518 size_t min_protoh = isipv6 ? 3519 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3520 sizeof (struct tcpiphdr); 3521 #else 3522 const size_t min_protoh = sizeof(struct tcpiphdr); 3523 #endif 3524 3525 INP_WLOCK_ASSERT(tp->t_inpcb); 3526 3527 if (mtuoffer != -1) { 3528 KASSERT(offer == -1, ("%s: conflict", __func__)); 3529 offer = mtuoffer - min_protoh; 3530 } 3531 3532 /* Initialize. */ 3533 #ifdef INET6 3534 if (isipv6) { 3535 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3536 tp->t_maxseg = V_tcp_v6mssdflt; 3537 } 3538 #endif 3539 #if defined(INET) && defined(INET6) 3540 else 3541 #endif 3542 #ifdef INET 3543 { 3544 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3545 tp->t_maxseg = V_tcp_mssdflt; 3546 } 3547 #endif 3548 3549 /* 3550 * No route to sender, stay with default mss and return. 3551 */ 3552 if (maxmtu == 0) { 3553 /* 3554 * In case we return early we need to initialize metrics 3555 * to a defined state as tcp_hc_get() would do for us 3556 * if there was no cache hit. 3557 */ 3558 if (metricptr != NULL) 3559 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3560 return; 3561 } 3562 3563 /* What have we got? */ 3564 switch (offer) { 3565 case 0: 3566 /* 3567 * Offer == 0 means that there was no MSS on the SYN 3568 * segment, in this case we use tcp_mssdflt as 3569 * already assigned to t_maxseg above. 3570 */ 3571 offer = tp->t_maxseg; 3572 break; 3573 3574 case -1: 3575 /* 3576 * Offer == -1 means that we didn't receive SYN yet. 3577 */ 3578 /* FALLTHROUGH */ 3579 3580 default: 3581 /* 3582 * Prevent DoS attack with too small MSS. Round up 3583 * to at least minmss. 3584 */ 3585 offer = max(offer, V_tcp_minmss); 3586 } 3587 3588 /* 3589 * rmx information is now retrieved from tcp_hostcache. 3590 */ 3591 tcp_hc_get(&inp->inp_inc, &metrics); 3592 if (metricptr != NULL) 3593 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3594 3595 /* 3596 * If there's a discovered mtu in tcp hostcache, use it. 3597 * Else, use the link mtu. 3598 */ 3599 if (metrics.rmx_mtu) 3600 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3601 else { 3602 #ifdef INET6 3603 if (isipv6) { 3604 mss = maxmtu - min_protoh; 3605 if (!V_path_mtu_discovery && 3606 !in6_localaddr(&inp->in6p_faddr)) 3607 mss = min(mss, V_tcp_v6mssdflt); 3608 } 3609 #endif 3610 #if defined(INET) && defined(INET6) 3611 else 3612 #endif 3613 #ifdef INET 3614 { 3615 mss = maxmtu - min_protoh; 3616 if (!V_path_mtu_discovery && 3617 !in_localaddr(inp->inp_faddr)) 3618 mss = min(mss, V_tcp_mssdflt); 3619 } 3620 #endif 3621 /* 3622 * XXX - The above conditional (mss = maxmtu - min_protoh) 3623 * probably violates the TCP spec. 3624 * The problem is that, since we don't know the 3625 * other end's MSS, we are supposed to use a conservative 3626 * default. But, if we do that, then MTU discovery will 3627 * never actually take place, because the conservative 3628 * default is much less than the MTUs typically seen 3629 * on the Internet today. For the moment, we'll sweep 3630 * this under the carpet. 3631 * 3632 * The conservative default might not actually be a problem 3633 * if the only case this occurs is when sending an initial 3634 * SYN with options and data to a host we've never talked 3635 * to before. Then, they will reply with an MSS value which 3636 * will get recorded and the new parameters should get 3637 * recomputed. For Further Study. 3638 */ 3639 } 3640 mss = min(mss, offer); 3641 3642 /* 3643 * Sanity check: make sure that maxseg will be large 3644 * enough to allow some data on segments even if the 3645 * all the option space is used (40bytes). Otherwise 3646 * funny things may happen in tcp_output. 3647 * 3648 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3649 */ 3650 mss = max(mss, 64); 3651 3652 tp->t_maxseg = mss; 3653 } 3654 3655 void 3656 tcp_mss(struct tcpcb *tp, int offer) 3657 { 3658 int mss; 3659 uint32_t bufsize; 3660 struct inpcb *inp; 3661 struct socket *so; 3662 struct hc_metrics_lite metrics; 3663 struct tcp_ifcap cap; 3664 3665 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3666 3667 bzero(&cap, sizeof(cap)); 3668 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3669 3670 mss = tp->t_maxseg; 3671 inp = tp->t_inpcb; 3672 3673 /* 3674 * If there's a pipesize, change the socket buffer to that size, 3675 * don't change if sb_hiwat is different than default (then it 3676 * has been changed on purpose with setsockopt). 3677 * Make the socket buffers an integral number of mss units; 3678 * if the mss is larger than the socket buffer, decrease the mss. 3679 */ 3680 so = inp->inp_socket; 3681 SOCKBUF_LOCK(&so->so_snd); 3682 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3683 bufsize = metrics.rmx_sendpipe; 3684 else 3685 bufsize = so->so_snd.sb_hiwat; 3686 if (bufsize < mss) 3687 mss = bufsize; 3688 else { 3689 bufsize = roundup(bufsize, mss); 3690 if (bufsize > sb_max) 3691 bufsize = sb_max; 3692 if (bufsize > so->so_snd.sb_hiwat) 3693 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3694 } 3695 SOCKBUF_UNLOCK(&so->so_snd); 3696 /* 3697 * Sanity check: make sure that maxseg will be large 3698 * enough to allow some data on segments even if the 3699 * all the option space is used (40bytes). Otherwise 3700 * funny things may happen in tcp_output. 3701 * 3702 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3703 */ 3704 tp->t_maxseg = max(mss, 64); 3705 3706 SOCKBUF_LOCK(&so->so_rcv); 3707 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3708 bufsize = metrics.rmx_recvpipe; 3709 else 3710 bufsize = so->so_rcv.sb_hiwat; 3711 if (bufsize > mss) { 3712 bufsize = roundup(bufsize, mss); 3713 if (bufsize > sb_max) 3714 bufsize = sb_max; 3715 if (bufsize > so->so_rcv.sb_hiwat) 3716 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3717 } 3718 SOCKBUF_UNLOCK(&so->so_rcv); 3719 3720 /* Check the interface for TSO capabilities. */ 3721 if (cap.ifcap & CSUM_TSO) { 3722 tp->t_flags |= TF_TSO; 3723 tp->t_tsomax = cap.tsomax; 3724 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3725 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3726 } 3727 } 3728 3729 /* 3730 * Determine the MSS option to send on an outgoing SYN. 3731 */ 3732 int 3733 tcp_mssopt(struct in_conninfo *inc) 3734 { 3735 int mss = 0; 3736 uint32_t thcmtu = 0; 3737 uint32_t maxmtu = 0; 3738 size_t min_protoh; 3739 3740 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3741 3742 #ifdef INET6 3743 if (inc->inc_flags & INC_ISIPV6) { 3744 mss = V_tcp_v6mssdflt; 3745 maxmtu = tcp_maxmtu6(inc, NULL); 3746 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3747 } 3748 #endif 3749 #if defined(INET) && defined(INET6) 3750 else 3751 #endif 3752 #ifdef INET 3753 { 3754 mss = V_tcp_mssdflt; 3755 maxmtu = tcp_maxmtu(inc, NULL); 3756 min_protoh = sizeof(struct tcpiphdr); 3757 } 3758 #endif 3759 #if defined(INET6) || defined(INET) 3760 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3761 #endif 3762 3763 if (maxmtu && thcmtu) 3764 mss = min(maxmtu, thcmtu) - min_protoh; 3765 else if (maxmtu || thcmtu) 3766 mss = max(maxmtu, thcmtu) - min_protoh; 3767 3768 return (mss); 3769 } 3770 3771 3772 /* 3773 * On a partial ack arrives, force the retransmission of the 3774 * next unacknowledged segment. Do not clear tp->t_dupacks. 3775 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3776 * be started again. 3777 */ 3778 void 3779 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3780 { 3781 tcp_seq onxt = tp->snd_nxt; 3782 uint32_t ocwnd = tp->snd_cwnd; 3783 u_int maxseg = tcp_maxseg(tp); 3784 3785 INP_WLOCK_ASSERT(tp->t_inpcb); 3786 3787 tcp_timer_activate(tp, TT_REXMT, 0); 3788 tp->t_rtttime = 0; 3789 tp->snd_nxt = th->th_ack; 3790 /* 3791 * Set snd_cwnd to one segment beyond acknowledged offset. 3792 * (tp->snd_una has not yet been updated when this function is called.) 3793 */ 3794 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3795 tp->t_flags |= TF_ACKNOW; 3796 (void) tp->t_fb->tfb_tcp_output(tp); 3797 tp->snd_cwnd = ocwnd; 3798 if (SEQ_GT(onxt, tp->snd_nxt)) 3799 tp->snd_nxt = onxt; 3800 /* 3801 * Partial window deflation. Relies on fact that tp->snd_una 3802 * not updated yet. 3803 */ 3804 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3805 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3806 else 3807 tp->snd_cwnd = 0; 3808 tp->snd_cwnd += maxseg; 3809 } 3810 3811 int 3812 tcp_compute_pipe(struct tcpcb *tp) 3813 { 3814 return (tp->snd_max - tp->snd_una + 3815 tp->sackhint.sack_bytes_rexmit - 3816 tp->sackhint.sacked_bytes); 3817 } 3818