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