1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 __FBSDID("$FreeBSD$"); 54 55 #include "opt_inet.h" 56 #include "opt_inet6.h" 57 #include "opt_ipsec.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/kernel.h> 62 #ifdef TCP_HHOOK 63 #include <sys/hhook.h> 64 #endif 65 #include <sys/malloc.h> 66 #include <sys/mbuf.h> 67 #include <sys/proc.h> /* for proc0 declaration */ 68 #include <sys/protosw.h> 69 #include <sys/sdt.h> 70 #include <sys/signalvar.h> 71 #include <sys/socket.h> 72 #include <sys/socketvar.h> 73 #include <sys/sysctl.h> 74 #include <sys/syslog.h> 75 #include <sys/systm.h> 76 77 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 78 79 #include <vm/uma.h> 80 81 #include <net/if.h> 82 #include <net/if_var.h> 83 #include <net/route.h> 84 #include <net/vnet.h> 85 86 #define TCPSTATES /* for logging */ 87 88 #include <netinet/in.h> 89 #include <netinet/in_kdtrace.h> 90 #include <netinet/in_pcb.h> 91 #include <netinet/in_systm.h> 92 #include <netinet/ip.h> 93 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 94 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 95 #include <netinet/ip_var.h> 96 #include <netinet/ip_options.h> 97 #include <netinet/ip6.h> 98 #include <netinet/icmp6.h> 99 #include <netinet6/in6_pcb.h> 100 #include <netinet6/in6_var.h> 101 #include <netinet6/ip6_var.h> 102 #include <netinet6/nd6.h> 103 #include <netinet/tcp.h> 104 #include <netinet/tcp_fsm.h> 105 #include <netinet/tcp_log_buf.h> 106 #include <netinet/tcp_seq.h> 107 #include <netinet/tcp_timer.h> 108 #include <netinet/tcp_var.h> 109 #include <netinet6/tcp6_var.h> 110 #include <netinet/tcpip.h> 111 #include <netinet/cc/cc.h> 112 #include <netinet/tcp_fastopen.h> 113 #ifdef TCPPCAP 114 #include <netinet/tcp_pcap.h> 115 #endif 116 #include <netinet/tcp_syncache.h> 117 #ifdef TCPDEBUG 118 #include <netinet/tcp_debug.h> 119 #endif /* TCPDEBUG */ 120 #ifdef TCP_OFFLOAD 121 #include <netinet/tcp_offload.h> 122 #endif 123 124 #include <netipsec/ipsec_support.h> 125 126 #include <machine/in_cksum.h> 127 128 #include <security/mac/mac_framework.h> 129 130 const int tcprexmtthresh = 3; 131 132 int tcp_log_in_vain = 0; 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 134 &tcp_log_in_vain, 0, 135 "Log all incoming TCP segments to closed ports"); 136 137 VNET_DEFINE(int, blackhole) = 0; 138 #define V_blackhole VNET(blackhole) 139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 140 &VNET_NAME(blackhole), 0, 141 "Do not send RST on segments to closed ports"); 142 143 VNET_DEFINE(int, tcp_delack_enabled) = 1; 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 145 &VNET_NAME(tcp_delack_enabled), 0, 146 "Delay ACK to try and piggyback it onto a data packet"); 147 148 VNET_DEFINE(int, drop_synfin) = 0; 149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 150 &VNET_NAME(drop_synfin), 0, 151 "Drop TCP packets with SYN+FIN set"); 152 153 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 155 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 156 "Use calculated pipe/in-flight bytes per RFC 6675"); 157 158 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 160 &VNET_NAME(tcp_do_rfc3042), 0, 161 "Enable RFC 3042 (Limited Transmit)"); 162 163 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 165 &VNET_NAME(tcp_do_rfc3390), 0, 166 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 167 168 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 170 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 171 "Slow-start flight size (initial congestion window) in number of segments"); 172 173 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 175 &VNET_NAME(tcp_do_rfc3465), 0, 176 "Enable RFC 3465 (Appropriate Byte Counting)"); 177 178 VNET_DEFINE(int, tcp_abc_l_var) = 2; 179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 180 &VNET_NAME(tcp_abc_l_var), 2, 181 "Cap the max cwnd increment during slow-start to this number of segments"); 182 183 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 184 185 VNET_DEFINE(int, tcp_do_ecn) = 2; 186 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 187 &VNET_NAME(tcp_do_ecn), 0, 188 "TCP ECN support"); 189 190 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 191 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_ecn_maxretries), 0, 193 "Max retries before giving up on ECN"); 194 195 VNET_DEFINE(int, tcp_insecure_syn) = 0; 196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 197 &VNET_NAME(tcp_insecure_syn), 0, 198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 199 200 VNET_DEFINE(int, tcp_insecure_rst) = 0; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 202 &VNET_NAME(tcp_insecure_rst), 0, 203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 204 205 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 206 #define V_tcp_recvspace VNET(tcp_recvspace) 207 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 209 210 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_do_autorcvbuf), 0, 213 "Enable automatic receive buffer sizing"); 214 215 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_autorcvbuf_inc), 0, 218 "Incrementor step size of automatic receive buffer"); 219 220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_autorcvbuf_max), 0, 223 "Max size of automatic receive buffer"); 224 225 VNET_DEFINE(struct inpcbhead, tcb); 226 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 227 VNET_DEFINE(struct inpcbinfo, tcbinfo); 228 229 /* 230 * TCP statistics are stored in an array of counter(9)s, which size matches 231 * size of struct tcpstat. TCP running connection count is a regular array. 232 */ 233 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 234 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 235 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 236 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 237 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 238 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 239 "TCP connection counts by TCP state"); 240 241 static void 242 tcp_vnet_init(const void *unused) 243 { 244 245 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 246 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 247 } 248 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 249 tcp_vnet_init, NULL); 250 251 #ifdef VIMAGE 252 static void 253 tcp_vnet_uninit(const void *unused) 254 { 255 256 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 257 VNET_PCPUSTAT_FREE(tcpstat); 258 } 259 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 260 tcp_vnet_uninit, NULL); 261 #endif /* VIMAGE */ 262 263 /* 264 * Kernel module interface for updating tcpstat. The argument is an index 265 * into tcpstat treated as an array. 266 */ 267 void 268 kmod_tcpstat_inc(int statnum) 269 { 270 271 counter_u64_add(VNET(tcpstat)[statnum], 1); 272 } 273 274 #ifdef TCP_HHOOK 275 /* 276 * Wrapper for the TCP established input helper hook. 277 */ 278 void 279 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 280 { 281 struct tcp_hhook_data hhook_data; 282 283 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 284 hhook_data.tp = tp; 285 hhook_data.th = th; 286 hhook_data.to = to; 287 288 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 289 tp->osd); 290 } 291 } 292 #endif 293 294 /* 295 * CC wrapper hook functions 296 */ 297 void 298 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 299 uint16_t type) 300 { 301 INP_WLOCK_ASSERT(tp->t_inpcb); 302 303 tp->ccv->nsegs = nsegs; 304 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 305 if (tp->snd_cwnd <= tp->snd_wnd) 306 tp->ccv->flags |= CCF_CWND_LIMITED; 307 else 308 tp->ccv->flags &= ~CCF_CWND_LIMITED; 309 310 if (type == CC_ACK) { 311 if (tp->snd_cwnd > tp->snd_ssthresh) { 312 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 313 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 314 if (tp->t_bytes_acked >= tp->snd_cwnd) { 315 tp->t_bytes_acked -= tp->snd_cwnd; 316 tp->ccv->flags |= CCF_ABC_SENTAWND; 317 } 318 } else { 319 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 320 tp->t_bytes_acked = 0; 321 } 322 } 323 324 if (CC_ALGO(tp)->ack_received != NULL) { 325 /* XXXLAS: Find a way to live without this */ 326 tp->ccv->curack = th->th_ack; 327 CC_ALGO(tp)->ack_received(tp->ccv, type); 328 } 329 } 330 331 void 332 cc_conn_init(struct tcpcb *tp) 333 { 334 struct hc_metrics_lite metrics; 335 struct inpcb *inp = tp->t_inpcb; 336 u_int maxseg; 337 int rtt; 338 339 INP_WLOCK_ASSERT(tp->t_inpcb); 340 341 tcp_hc_get(&inp->inp_inc, &metrics); 342 maxseg = tcp_maxseg(tp); 343 344 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 345 tp->t_srtt = rtt; 346 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 347 TCPSTAT_INC(tcps_usedrtt); 348 if (metrics.rmx_rttvar) { 349 tp->t_rttvar = metrics.rmx_rttvar; 350 TCPSTAT_INC(tcps_usedrttvar); 351 } else { 352 /* default variation is +- 1 rtt */ 353 tp->t_rttvar = 354 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 355 } 356 TCPT_RANGESET(tp->t_rxtcur, 357 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 358 tp->t_rttmin, TCPTV_REXMTMAX); 359 } 360 if (metrics.rmx_ssthresh) { 361 /* 362 * There's some sort of gateway or interface 363 * buffer limit on the path. Use this to set 364 * the slow start threshold, but set the 365 * threshold to no less than 2*mss. 366 */ 367 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 368 TCPSTAT_INC(tcps_usedssthresh); 369 } 370 371 /* 372 * Set the initial slow-start flight size. 373 * 374 * RFC5681 Section 3.1 specifies the default conservative values. 375 * RFC3390 specifies slightly more aggressive values. 376 * RFC6928 increases it to ten segments. 377 * Support for user specified value for initial flight size. 378 * 379 * If a SYN or SYN/ACK was lost and retransmitted, we have to 380 * reduce the initial CWND to one segment as congestion is likely 381 * requiring us to be cautious. 382 */ 383 if (tp->snd_cwnd == 1) 384 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 385 else if (V_tcp_initcwnd_segments) 386 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 387 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 388 else if (V_tcp_do_rfc3390) 389 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 390 else { 391 /* Per RFC5681 Section 3.1 */ 392 if (maxseg > 2190) 393 tp->snd_cwnd = 2 * maxseg; 394 else if (maxseg > 1095) 395 tp->snd_cwnd = 3 * maxseg; 396 else 397 tp->snd_cwnd = 4 * maxseg; 398 } 399 400 if (CC_ALGO(tp)->conn_init != NULL) 401 CC_ALGO(tp)->conn_init(tp->ccv); 402 } 403 404 void inline 405 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 406 { 407 u_int maxseg; 408 409 INP_WLOCK_ASSERT(tp->t_inpcb); 410 411 switch(type) { 412 case CC_NDUPACK: 413 if (!IN_FASTRECOVERY(tp->t_flags)) { 414 tp->snd_recover = tp->snd_max; 415 if (tp->t_flags & TF_ECN_PERMIT) 416 tp->t_flags |= TF_ECN_SND_CWR; 417 } 418 break; 419 case CC_ECN: 420 if (!IN_CONGRECOVERY(tp->t_flags)) { 421 TCPSTAT_INC(tcps_ecn_rcwnd); 422 tp->snd_recover = tp->snd_max; 423 if (tp->t_flags & TF_ECN_PERMIT) 424 tp->t_flags |= TF_ECN_SND_CWR; 425 } 426 break; 427 case CC_RTO: 428 maxseg = tcp_maxseg(tp); 429 tp->t_dupacks = 0; 430 tp->t_bytes_acked = 0; 431 EXIT_RECOVERY(tp->t_flags); 432 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 433 maxseg) * maxseg; 434 tp->snd_cwnd = maxseg; 435 break; 436 case CC_RTO_ERR: 437 TCPSTAT_INC(tcps_sndrexmitbad); 438 /* RTO was unnecessary, so reset everything. */ 439 tp->snd_cwnd = tp->snd_cwnd_prev; 440 tp->snd_ssthresh = tp->snd_ssthresh_prev; 441 tp->snd_recover = tp->snd_recover_prev; 442 if (tp->t_flags & TF_WASFRECOVERY) 443 ENTER_FASTRECOVERY(tp->t_flags); 444 if (tp->t_flags & TF_WASCRECOVERY) 445 ENTER_CONGRECOVERY(tp->t_flags); 446 tp->snd_nxt = tp->snd_max; 447 tp->t_flags &= ~TF_PREVVALID; 448 tp->t_badrxtwin = 0; 449 break; 450 } 451 452 if (CC_ALGO(tp)->cong_signal != NULL) { 453 if (th != NULL) 454 tp->ccv->curack = th->th_ack; 455 CC_ALGO(tp)->cong_signal(tp->ccv, type); 456 } 457 } 458 459 void inline 460 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 461 { 462 INP_WLOCK_ASSERT(tp->t_inpcb); 463 464 /* XXXLAS: KASSERT that we're in recovery? */ 465 466 if (CC_ALGO(tp)->post_recovery != NULL) { 467 tp->ccv->curack = th->th_ack; 468 CC_ALGO(tp)->post_recovery(tp->ccv); 469 } 470 /* XXXLAS: EXIT_RECOVERY ? */ 471 tp->t_bytes_acked = 0; 472 } 473 474 /* 475 * Indicate whether this ack should be delayed. We can delay the ack if 476 * following conditions are met: 477 * - There is no delayed ack timer in progress. 478 * - Our last ack wasn't a 0-sized window. We never want to delay 479 * the ack that opens up a 0-sized window. 480 * - LRO wasn't used for this segment. We make sure by checking that the 481 * segment size is not larger than the MSS. 482 */ 483 #define DELAY_ACK(tp, tlen) \ 484 ((!tcp_timer_active(tp, TT_DELACK) && \ 485 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 486 (tlen <= tp->t_maxseg) && \ 487 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 488 489 static void inline 490 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 491 { 492 INP_WLOCK_ASSERT(tp->t_inpcb); 493 494 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 495 switch (iptos & IPTOS_ECN_MASK) { 496 case IPTOS_ECN_CE: 497 tp->ccv->flags |= CCF_IPHDR_CE; 498 break; 499 case IPTOS_ECN_ECT0: 500 tp->ccv->flags &= ~CCF_IPHDR_CE; 501 break; 502 case IPTOS_ECN_ECT1: 503 tp->ccv->flags &= ~CCF_IPHDR_CE; 504 break; 505 } 506 507 if (th->th_flags & TH_CWR) 508 tp->ccv->flags |= CCF_TCPHDR_CWR; 509 else 510 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 511 512 if (tp->t_flags & TF_DELACK) 513 tp->ccv->flags |= CCF_DELACK; 514 else 515 tp->ccv->flags &= ~CCF_DELACK; 516 517 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 518 519 if (tp->ccv->flags & CCF_ACKNOW) 520 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 521 } 522 } 523 524 /* 525 * TCP input handling is split into multiple parts: 526 * tcp6_input is a thin wrapper around tcp_input for the extended 527 * ip6_protox[] call format in ip6_input 528 * tcp_input handles primary segment validation, inpcb lookup and 529 * SYN processing on listen sockets 530 * tcp_do_segment processes the ACK and text of the segment for 531 * establishing, established and closing connections 532 */ 533 #ifdef INET6 534 int 535 tcp6_input(struct mbuf **mp, int *offp, int proto) 536 { 537 struct mbuf *m = *mp; 538 struct in6_ifaddr *ia6; 539 struct ip6_hdr *ip6; 540 541 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 542 543 /* 544 * draft-itojun-ipv6-tcp-to-anycast 545 * better place to put this in? 546 */ 547 ip6 = mtod(m, struct ip6_hdr *); 548 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 549 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 550 struct ip6_hdr *ip6; 551 552 ifa_free(&ia6->ia_ifa); 553 ip6 = mtod(m, struct ip6_hdr *); 554 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 555 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 556 return (IPPROTO_DONE); 557 } 558 if (ia6) 559 ifa_free(&ia6->ia_ifa); 560 561 return (tcp_input(mp, offp, proto)); 562 } 563 #endif /* INET6 */ 564 565 int 566 tcp_input(struct mbuf **mp, int *offp, int proto) 567 { 568 struct mbuf *m = *mp; 569 struct tcphdr *th = NULL; 570 struct ip *ip = NULL; 571 struct inpcb *inp = NULL; 572 struct tcpcb *tp = NULL; 573 struct socket *so = NULL; 574 u_char *optp = NULL; 575 int off0; 576 int optlen = 0; 577 #ifdef INET 578 int len; 579 #endif 580 int tlen = 0, off; 581 int drop_hdrlen; 582 int thflags; 583 int rstreason = 0; /* For badport_bandlim accounting purposes */ 584 uint8_t iptos; 585 struct m_tag *fwd_tag = NULL; 586 struct epoch_tracker et; 587 #ifdef INET6 588 struct ip6_hdr *ip6 = NULL; 589 int isipv6; 590 #else 591 const void *ip6 = NULL; 592 #endif /* INET6 */ 593 struct tcpopt to; /* options in this segment */ 594 char *s = NULL; /* address and port logging */ 595 int ti_locked; 596 #ifdef TCPDEBUG 597 /* 598 * The size of tcp_saveipgen must be the size of the max ip header, 599 * now IPv6. 600 */ 601 u_char tcp_saveipgen[IP6_HDR_LEN]; 602 struct tcphdr tcp_savetcp; 603 short ostate = 0; 604 #endif 605 606 #ifdef INET6 607 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 608 #endif 609 610 off0 = *offp; 611 m = *mp; 612 *mp = NULL; 613 to.to_flags = 0; 614 TCPSTAT_INC(tcps_rcvtotal); 615 616 #ifdef INET6 617 if (isipv6) { 618 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 619 620 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 621 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 622 if (m == NULL) { 623 TCPSTAT_INC(tcps_rcvshort); 624 return (IPPROTO_DONE); 625 } 626 } 627 628 ip6 = mtod(m, struct ip6_hdr *); 629 th = (struct tcphdr *)((caddr_t)ip6 + off0); 630 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 631 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 632 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 633 th->th_sum = m->m_pkthdr.csum_data; 634 else 635 th->th_sum = in6_cksum_pseudo(ip6, tlen, 636 IPPROTO_TCP, m->m_pkthdr.csum_data); 637 th->th_sum ^= 0xffff; 638 } else 639 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 640 if (th->th_sum) { 641 TCPSTAT_INC(tcps_rcvbadsum); 642 goto drop; 643 } 644 645 /* 646 * Be proactive about unspecified IPv6 address in source. 647 * As we use all-zero to indicate unbounded/unconnected pcb, 648 * unspecified IPv6 address can be used to confuse us. 649 * 650 * Note that packets with unspecified IPv6 destination is 651 * already dropped in ip6_input. 652 */ 653 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 654 /* XXX stat */ 655 goto drop; 656 } 657 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 658 } 659 #endif 660 #if defined(INET) && defined(INET6) 661 else 662 #endif 663 #ifdef INET 664 { 665 /* 666 * Get IP and TCP header together in first mbuf. 667 * Note: IP leaves IP header in first mbuf. 668 */ 669 if (off0 > sizeof (struct ip)) { 670 ip_stripoptions(m); 671 off0 = sizeof(struct ip); 672 } 673 if (m->m_len < sizeof (struct tcpiphdr)) { 674 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 675 == NULL) { 676 TCPSTAT_INC(tcps_rcvshort); 677 return (IPPROTO_DONE); 678 } 679 } 680 ip = mtod(m, struct ip *); 681 th = (struct tcphdr *)((caddr_t)ip + off0); 682 tlen = ntohs(ip->ip_len) - off0; 683 684 iptos = ip->ip_tos; 685 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 686 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 687 th->th_sum = m->m_pkthdr.csum_data; 688 else 689 th->th_sum = in_pseudo(ip->ip_src.s_addr, 690 ip->ip_dst.s_addr, 691 htonl(m->m_pkthdr.csum_data + tlen + 692 IPPROTO_TCP)); 693 th->th_sum ^= 0xffff; 694 } else { 695 struct ipovly *ipov = (struct ipovly *)ip; 696 697 /* 698 * Checksum extended TCP header and data. 699 */ 700 len = off0 + tlen; 701 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 702 ipov->ih_len = htons(tlen); 703 th->th_sum = in_cksum(m, len); 704 /* Reset length for SDT probes. */ 705 ip->ip_len = htons(len); 706 /* Reset TOS bits */ 707 ip->ip_tos = iptos; 708 /* Re-initialization for later version check */ 709 ip->ip_v = IPVERSION; 710 ip->ip_hl = off0 >> 2; 711 } 712 713 if (th->th_sum) { 714 TCPSTAT_INC(tcps_rcvbadsum); 715 goto drop; 716 } 717 } 718 #endif /* INET */ 719 720 /* 721 * Check that TCP offset makes sense, 722 * pull out TCP options and adjust length. XXX 723 */ 724 off = th->th_off << 2; 725 if (off < sizeof (struct tcphdr) || off > tlen) { 726 TCPSTAT_INC(tcps_rcvbadoff); 727 goto drop; 728 } 729 tlen -= off; /* tlen is used instead of ti->ti_len */ 730 if (off > sizeof (struct tcphdr)) { 731 #ifdef INET6 732 if (isipv6) { 733 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 734 ip6 = mtod(m, struct ip6_hdr *); 735 th = (struct tcphdr *)((caddr_t)ip6 + off0); 736 } 737 #endif 738 #if defined(INET) && defined(INET6) 739 else 740 #endif 741 #ifdef INET 742 { 743 if (m->m_len < sizeof(struct ip) + off) { 744 if ((m = m_pullup(m, sizeof (struct ip) + off)) 745 == NULL) { 746 TCPSTAT_INC(tcps_rcvshort); 747 return (IPPROTO_DONE); 748 } 749 ip = mtod(m, struct ip *); 750 th = (struct tcphdr *)((caddr_t)ip + off0); 751 } 752 } 753 #endif 754 optlen = off - sizeof (struct tcphdr); 755 optp = (u_char *)(th + 1); 756 } 757 thflags = th->th_flags; 758 759 /* 760 * Convert TCP protocol specific fields to host format. 761 */ 762 tcp_fields_to_host(th); 763 764 /* 765 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 766 */ 767 drop_hdrlen = off0 + off; 768 769 /* 770 * Locate pcb for segment; if we're likely to add or remove a 771 * connection then first acquire pcbinfo lock. There are three cases 772 * where we might discover later we need a write lock despite the 773 * flags: ACKs moving a connection out of the syncache, ACKs for a 774 * connection in TIMEWAIT and SYNs not targeting a listening socket. 775 */ 776 if ((thflags & (TH_FIN | TH_RST)) != 0) { 777 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 778 ti_locked = TI_RLOCKED; 779 } else 780 ti_locked = TI_UNLOCKED; 781 782 /* 783 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 784 */ 785 if ( 786 #ifdef INET6 787 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 788 #ifdef INET 789 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 790 #endif 791 #endif 792 #if defined(INET) && !defined(INET6) 793 (m->m_flags & M_IP_NEXTHOP) 794 #endif 795 ) 796 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 797 798 findpcb: 799 #ifdef INVARIANTS 800 if (ti_locked == TI_RLOCKED) { 801 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 802 } else { 803 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 804 } 805 #endif 806 #ifdef INET6 807 if (isipv6 && fwd_tag != NULL) { 808 struct sockaddr_in6 *next_hop6; 809 810 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 811 /* 812 * Transparently forwarded. Pretend to be the destination. 813 * Already got one like this? 814 */ 815 inp = in6_pcblookup_mbuf(&V_tcbinfo, 816 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 817 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 818 if (!inp) { 819 /* 820 * It's new. Try to find the ambushing socket. 821 * Because we've rewritten the destination address, 822 * any hardware-generated hash is ignored. 823 */ 824 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 825 th->th_sport, &next_hop6->sin6_addr, 826 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 827 th->th_dport, INPLOOKUP_WILDCARD | 828 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 829 } 830 } else if (isipv6) { 831 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 832 th->th_sport, &ip6->ip6_dst, th->th_dport, 833 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 834 m->m_pkthdr.rcvif, m); 835 } 836 #endif /* INET6 */ 837 #if defined(INET6) && defined(INET) 838 else 839 #endif 840 #ifdef INET 841 if (fwd_tag != NULL) { 842 struct sockaddr_in *next_hop; 843 844 next_hop = (struct sockaddr_in *)(fwd_tag+1); 845 /* 846 * Transparently forwarded. Pretend to be the destination. 847 * already got one like this? 848 */ 849 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 850 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 851 m->m_pkthdr.rcvif, m); 852 if (!inp) { 853 /* 854 * It's new. Try to find the ambushing socket. 855 * Because we've rewritten the destination address, 856 * any hardware-generated hash is ignored. 857 */ 858 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 859 th->th_sport, next_hop->sin_addr, 860 next_hop->sin_port ? ntohs(next_hop->sin_port) : 861 th->th_dport, INPLOOKUP_WILDCARD | 862 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 863 } 864 } else 865 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 866 th->th_sport, ip->ip_dst, th->th_dport, 867 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 868 m->m_pkthdr.rcvif, m); 869 #endif /* INET */ 870 871 /* 872 * If the INPCB does not exist then all data in the incoming 873 * segment is discarded and an appropriate RST is sent back. 874 * XXX MRT Send RST using which routing table? 875 */ 876 if (inp == NULL) { 877 /* 878 * Log communication attempts to ports that are not 879 * in use. 880 */ 881 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 882 tcp_log_in_vain == 2) { 883 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 884 log(LOG_INFO, "%s; %s: Connection attempt " 885 "to closed port\n", s, __func__); 886 } 887 /* 888 * When blackholing do not respond with a RST but 889 * completely ignore the segment and drop it. 890 */ 891 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 892 V_blackhole == 2) 893 goto dropunlock; 894 895 rstreason = BANDLIM_RST_CLOSEDPORT; 896 goto dropwithreset; 897 } 898 INP_WLOCK_ASSERT(inp); 899 /* 900 * While waiting for inp lock during the lookup, another thread 901 * can have dropped the inpcb, in which case we need to loop back 902 * and try to find a new inpcb to deliver to. 903 */ 904 if (inp->inp_flags & INP_DROPPED) { 905 INP_WUNLOCK(inp); 906 inp = NULL; 907 goto findpcb; 908 } 909 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 910 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 911 ((inp->inp_socket == NULL) || 912 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 913 inp->inp_flowid = m->m_pkthdr.flowid; 914 inp->inp_flowtype = M_HASHTYPE_GET(m); 915 } 916 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 917 #ifdef INET6 918 if (isipv6 && IPSEC_ENABLED(ipv6) && 919 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 920 goto dropunlock; 921 } 922 #ifdef INET 923 else 924 #endif 925 #endif /* INET6 */ 926 #ifdef INET 927 if (IPSEC_ENABLED(ipv4) && 928 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 929 goto dropunlock; 930 } 931 #endif /* INET */ 932 #endif /* IPSEC */ 933 934 /* 935 * Check the minimum TTL for socket. 936 */ 937 if (inp->inp_ip_minttl != 0) { 938 #ifdef INET6 939 if (isipv6) { 940 if (inp->inp_ip_minttl > ip6->ip6_hlim) 941 goto dropunlock; 942 } else 943 #endif 944 if (inp->inp_ip_minttl > ip->ip_ttl) 945 goto dropunlock; 946 } 947 948 /* 949 * A previous connection in TIMEWAIT state is supposed to catch stray 950 * or duplicate segments arriving late. If this segment was a 951 * legitimate new connection attempt, the old INPCB gets removed and 952 * we can try again to find a listening socket. 953 * 954 * At this point, due to earlier optimism, we may hold only an inpcb 955 * lock, and not the inpcbinfo write lock. If so, we need to try to 956 * acquire it, or if that fails, acquire a reference on the inpcb, 957 * drop all locks, acquire a global write lock, and then re-acquire 958 * the inpcb lock. We may at that point discover that another thread 959 * has tried to free the inpcb, in which case we need to loop back 960 * and try to find a new inpcb to deliver to. 961 * 962 * XXXRW: It may be time to rethink timewait locking. 963 */ 964 if (inp->inp_flags & INP_TIMEWAIT) { 965 if (ti_locked == TI_UNLOCKED) { 966 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 967 ti_locked = TI_RLOCKED; 968 } 969 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 970 971 if (thflags & TH_SYN) 972 tcp_dooptions(&to, optp, optlen, TO_SYN); 973 /* 974 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 975 */ 976 if (tcp_twcheck(inp, &to, th, m, tlen)) 977 goto findpcb; 978 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 979 return (IPPROTO_DONE); 980 } 981 /* 982 * The TCPCB may no longer exist if the connection is winding 983 * down or it is in the CLOSED state. Either way we drop the 984 * segment and send an appropriate response. 985 */ 986 tp = intotcpcb(inp); 987 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 988 rstreason = BANDLIM_RST_CLOSEDPORT; 989 goto dropwithreset; 990 } 991 992 #ifdef TCP_OFFLOAD 993 if (tp->t_flags & TF_TOE) { 994 tcp_offload_input(tp, m); 995 m = NULL; /* consumed by the TOE driver */ 996 goto dropunlock; 997 } 998 #endif 999 1000 /* 1001 * We've identified a valid inpcb, but it could be that we need an 1002 * inpcbinfo write lock but don't hold it. In this case, attempt to 1003 * acquire using the same strategy as the TIMEWAIT case above. If we 1004 * relock, we have to jump back to 'relocked' as the connection might 1005 * now be in TIMEWAIT. 1006 */ 1007 #ifdef INVARIANTS 1008 if ((thflags & (TH_FIN | TH_RST)) != 0) 1009 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1010 #endif 1011 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1012 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1013 !IS_FASTOPEN(tp->t_flags)))) { 1014 if (ti_locked == TI_UNLOCKED) { 1015 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 1016 ti_locked = TI_RLOCKED; 1017 } 1018 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1019 } 1020 1021 #ifdef MAC 1022 INP_WLOCK_ASSERT(inp); 1023 if (mac_inpcb_check_deliver(inp, m)) 1024 goto dropunlock; 1025 #endif 1026 so = inp->inp_socket; 1027 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1028 #ifdef TCPDEBUG 1029 if (so->so_options & SO_DEBUG) { 1030 ostate = tp->t_state; 1031 #ifdef INET6 1032 if (isipv6) { 1033 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1034 } else 1035 #endif 1036 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1037 tcp_savetcp = *th; 1038 } 1039 #endif /* TCPDEBUG */ 1040 /* 1041 * When the socket is accepting connections (the INPCB is in LISTEN 1042 * state) we look into the SYN cache if this is a new connection 1043 * attempt or the completion of a previous one. 1044 */ 1045 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1046 ("%s: so accepting but tp %p not listening", __func__, tp)); 1047 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1048 struct in_conninfo inc; 1049 1050 bzero(&inc, sizeof(inc)); 1051 #ifdef INET6 1052 if (isipv6) { 1053 inc.inc_flags |= INC_ISIPV6; 1054 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1055 inc.inc_flags |= INC_IPV6MINMTU; 1056 inc.inc6_faddr = ip6->ip6_src; 1057 inc.inc6_laddr = ip6->ip6_dst; 1058 } else 1059 #endif 1060 { 1061 inc.inc_faddr = ip->ip_src; 1062 inc.inc_laddr = ip->ip_dst; 1063 } 1064 inc.inc_fport = th->th_sport; 1065 inc.inc_lport = th->th_dport; 1066 inc.inc_fibnum = so->so_fibnum; 1067 1068 /* 1069 * Check for an existing connection attempt in syncache if 1070 * the flag is only ACK. A successful lookup creates a new 1071 * socket appended to the listen queue in SYN_RECEIVED state. 1072 */ 1073 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1074 1075 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1076 /* 1077 * Parse the TCP options here because 1078 * syncookies need access to the reflected 1079 * timestamp. 1080 */ 1081 tcp_dooptions(&to, optp, optlen, 0); 1082 /* 1083 * NB: syncache_expand() doesn't unlock 1084 * inp and tcpinfo locks. 1085 */ 1086 rstreason = syncache_expand(&inc, &to, th, &so, m); 1087 if (rstreason < 0) { 1088 /* 1089 * A failing TCP MD5 signature comparison 1090 * must result in the segment being dropped 1091 * and must not produce any response back 1092 * to the sender. 1093 */ 1094 goto dropunlock; 1095 } else if (rstreason == 0) { 1096 /* 1097 * No syncache entry or ACK was not 1098 * for our SYN/ACK. Send a RST. 1099 * NB: syncache did its own logging 1100 * of the failure cause. 1101 */ 1102 rstreason = BANDLIM_RST_OPENPORT; 1103 goto dropwithreset; 1104 } 1105 tfo_socket_result: 1106 if (so == NULL) { 1107 /* 1108 * We completed the 3-way handshake 1109 * but could not allocate a socket 1110 * either due to memory shortage, 1111 * listen queue length limits or 1112 * global socket limits. Send RST 1113 * or wait and have the remote end 1114 * retransmit the ACK for another 1115 * try. 1116 */ 1117 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1118 log(LOG_DEBUG, "%s; %s: Listen socket: " 1119 "Socket allocation failed due to " 1120 "limits or memory shortage, %s\n", 1121 s, __func__, 1122 V_tcp_sc_rst_sock_fail ? 1123 "sending RST" : "try again"); 1124 if (V_tcp_sc_rst_sock_fail) { 1125 rstreason = BANDLIM_UNLIMITED; 1126 goto dropwithreset; 1127 } else 1128 goto dropunlock; 1129 } 1130 /* 1131 * Socket is created in state SYN_RECEIVED. 1132 * Unlock the listen socket, lock the newly 1133 * created socket and update the tp variable. 1134 */ 1135 INP_WUNLOCK(inp); /* listen socket */ 1136 inp = sotoinpcb(so); 1137 /* 1138 * New connection inpcb is already locked by 1139 * syncache_expand(). 1140 */ 1141 INP_WLOCK_ASSERT(inp); 1142 tp = intotcpcb(inp); 1143 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1144 ("%s: ", __func__)); 1145 /* 1146 * Process the segment and the data it 1147 * contains. tcp_do_segment() consumes 1148 * the mbuf chain and unlocks the inpcb. 1149 */ 1150 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1151 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1152 iptos); 1153 if (ti_locked == TI_RLOCKED) 1154 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1155 return (IPPROTO_DONE); 1156 } 1157 /* 1158 * Segment flag validation for new connection attempts: 1159 * 1160 * Our (SYN|ACK) response was rejected. 1161 * Check with syncache and remove entry to prevent 1162 * retransmits. 1163 * 1164 * NB: syncache_chkrst does its own logging of failure 1165 * causes. 1166 */ 1167 if (thflags & TH_RST) { 1168 syncache_chkrst(&inc, th, m); 1169 goto dropunlock; 1170 } 1171 /* 1172 * We can't do anything without SYN. 1173 */ 1174 if ((thflags & TH_SYN) == 0) { 1175 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1176 log(LOG_DEBUG, "%s; %s: Listen socket: " 1177 "SYN is missing, segment ignored\n", 1178 s, __func__); 1179 TCPSTAT_INC(tcps_badsyn); 1180 goto dropunlock; 1181 } 1182 /* 1183 * (SYN|ACK) is bogus on a listen socket. 1184 */ 1185 if (thflags & TH_ACK) { 1186 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1187 log(LOG_DEBUG, "%s; %s: Listen socket: " 1188 "SYN|ACK invalid, segment rejected\n", 1189 s, __func__); 1190 syncache_badack(&inc); /* XXX: Not needed! */ 1191 TCPSTAT_INC(tcps_badsyn); 1192 rstreason = BANDLIM_RST_OPENPORT; 1193 goto dropwithreset; 1194 } 1195 /* 1196 * If the drop_synfin option is enabled, drop all 1197 * segments with both the SYN and FIN bits set. 1198 * This prevents e.g. nmap from identifying the 1199 * TCP/IP stack. 1200 * XXX: Poor reasoning. nmap has other methods 1201 * and is constantly refining its stack detection 1202 * strategies. 1203 * XXX: This is a violation of the TCP specification 1204 * and was used by RFC1644. 1205 */ 1206 if ((thflags & TH_FIN) && V_drop_synfin) { 1207 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1208 log(LOG_DEBUG, "%s; %s: Listen socket: " 1209 "SYN|FIN segment ignored (based on " 1210 "sysctl setting)\n", s, __func__); 1211 TCPSTAT_INC(tcps_badsyn); 1212 goto dropunlock; 1213 } 1214 /* 1215 * Segment's flags are (SYN) or (SYN|FIN). 1216 * 1217 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1218 * as they do not affect the state of the TCP FSM. 1219 * The data pointed to by TH_URG and th_urp is ignored. 1220 */ 1221 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1222 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1223 KASSERT(thflags & (TH_SYN), 1224 ("%s: Listen socket: TH_SYN not set", __func__)); 1225 #ifdef INET6 1226 /* 1227 * If deprecated address is forbidden, 1228 * we do not accept SYN to deprecated interface 1229 * address to prevent any new inbound connection from 1230 * getting established. 1231 * When we do not accept SYN, we send a TCP RST, 1232 * with deprecated source address (instead of dropping 1233 * it). We compromise it as it is much better for peer 1234 * to send a RST, and RST will be the final packet 1235 * for the exchange. 1236 * 1237 * If we do not forbid deprecated addresses, we accept 1238 * the SYN packet. RFC2462 does not suggest dropping 1239 * SYN in this case. 1240 * If we decipher RFC2462 5.5.4, it says like this: 1241 * 1. use of deprecated addr with existing 1242 * communication is okay - "SHOULD continue to be 1243 * used" 1244 * 2. use of it with new communication: 1245 * (2a) "SHOULD NOT be used if alternate address 1246 * with sufficient scope is available" 1247 * (2b) nothing mentioned otherwise. 1248 * Here we fall into (2b) case as we have no choice in 1249 * our source address selection - we must obey the peer. 1250 * 1251 * The wording in RFC2462 is confusing, and there are 1252 * multiple description text for deprecated address 1253 * handling - worse, they are not exactly the same. 1254 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1255 */ 1256 if (isipv6 && !V_ip6_use_deprecated) { 1257 struct in6_ifaddr *ia6; 1258 1259 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1260 if (ia6 != NULL && 1261 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1262 ifa_free(&ia6->ia_ifa); 1263 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1264 log(LOG_DEBUG, "%s; %s: Listen socket: " 1265 "Connection attempt to deprecated " 1266 "IPv6 address rejected\n", 1267 s, __func__); 1268 rstreason = BANDLIM_RST_OPENPORT; 1269 goto dropwithreset; 1270 } 1271 if (ia6) 1272 ifa_free(&ia6->ia_ifa); 1273 } 1274 #endif /* INET6 */ 1275 /* 1276 * Basic sanity checks on incoming SYN requests: 1277 * Don't respond if the destination is a link layer 1278 * broadcast according to RFC1122 4.2.3.10, p. 104. 1279 * If it is from this socket it must be forged. 1280 * Don't respond if the source or destination is a 1281 * global or subnet broad- or multicast address. 1282 * Note that it is quite possible to receive unicast 1283 * link-layer packets with a broadcast IP address. Use 1284 * in_broadcast() to find them. 1285 */ 1286 if (m->m_flags & (M_BCAST|M_MCAST)) { 1287 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1288 log(LOG_DEBUG, "%s; %s: Listen socket: " 1289 "Connection attempt from broad- or multicast " 1290 "link layer address ignored\n", s, __func__); 1291 goto dropunlock; 1292 } 1293 #ifdef INET6 1294 if (isipv6) { 1295 if (th->th_dport == th->th_sport && 1296 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1297 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1298 log(LOG_DEBUG, "%s; %s: Listen socket: " 1299 "Connection attempt to/from self " 1300 "ignored\n", s, __func__); 1301 goto dropunlock; 1302 } 1303 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1304 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1305 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1306 log(LOG_DEBUG, "%s; %s: Listen socket: " 1307 "Connection attempt from/to multicast " 1308 "address ignored\n", s, __func__); 1309 goto dropunlock; 1310 } 1311 } 1312 #endif 1313 #if defined(INET) && defined(INET6) 1314 else 1315 #endif 1316 #ifdef INET 1317 { 1318 if (th->th_dport == th->th_sport && 1319 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1320 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1321 log(LOG_DEBUG, "%s; %s: Listen socket: " 1322 "Connection attempt from/to self " 1323 "ignored\n", s, __func__); 1324 goto dropunlock; 1325 } 1326 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1327 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1328 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1329 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1330 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1331 log(LOG_DEBUG, "%s; %s: Listen socket: " 1332 "Connection attempt from/to broad- " 1333 "or multicast address ignored\n", 1334 s, __func__); 1335 goto dropunlock; 1336 } 1337 } 1338 #endif 1339 /* 1340 * SYN appears to be valid. Create compressed TCP state 1341 * for syncache. 1342 */ 1343 #ifdef TCPDEBUG 1344 if (so->so_options & SO_DEBUG) 1345 tcp_trace(TA_INPUT, ostate, tp, 1346 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1347 #endif 1348 TCP_PROBE3(debug__input, tp, th, m); 1349 tcp_dooptions(&to, optp, optlen, TO_SYN); 1350 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1351 goto tfo_socket_result; 1352 1353 /* 1354 * Entry added to syncache and mbuf consumed. 1355 * Only the listen socket is unlocked by syncache_add(). 1356 */ 1357 if (ti_locked == TI_RLOCKED) { 1358 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1359 ti_locked = TI_UNLOCKED; 1360 } 1361 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1362 return (IPPROTO_DONE); 1363 } else if (tp->t_state == TCPS_LISTEN) { 1364 /* 1365 * When a listen socket is torn down the SO_ACCEPTCONN 1366 * flag is removed first while connections are drained 1367 * from the accept queue in a unlock/lock cycle of the 1368 * ACCEPT_LOCK, opening a race condition allowing a SYN 1369 * attempt go through unhandled. 1370 */ 1371 goto dropunlock; 1372 } 1373 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1374 if (tp->t_flags & TF_SIGNATURE) { 1375 tcp_dooptions(&to, optp, optlen, thflags); 1376 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1377 TCPSTAT_INC(tcps_sig_err_nosigopt); 1378 goto dropunlock; 1379 } 1380 if (!TCPMD5_ENABLED() || 1381 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1382 goto dropunlock; 1383 } 1384 #endif 1385 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1386 1387 /* 1388 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1389 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1390 * the inpcb, and unlocks pcbinfo. 1391 */ 1392 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1393 if (ti_locked == TI_RLOCKED) 1394 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1395 return (IPPROTO_DONE); 1396 1397 dropwithreset: 1398 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1399 1400 if (ti_locked == TI_RLOCKED) { 1401 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1402 ti_locked = TI_UNLOCKED; 1403 } 1404 #ifdef INVARIANTS 1405 else { 1406 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1407 "ti_locked: %d", __func__, ti_locked)); 1408 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1409 } 1410 #endif 1411 1412 if (inp != NULL) { 1413 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1414 INP_WUNLOCK(inp); 1415 } else 1416 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1417 m = NULL; /* mbuf chain got consumed. */ 1418 goto drop; 1419 1420 dropunlock: 1421 if (m != NULL) 1422 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1423 1424 if (ti_locked == TI_RLOCKED) { 1425 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1426 ti_locked = TI_UNLOCKED; 1427 } 1428 #ifdef INVARIANTS 1429 else { 1430 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1431 "ti_locked: %d", __func__, ti_locked)); 1432 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1433 } 1434 #endif 1435 1436 if (inp != NULL) 1437 INP_WUNLOCK(inp); 1438 1439 drop: 1440 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1441 if (s != NULL) 1442 free(s, M_TCPLOG); 1443 if (m != NULL) 1444 m_freem(m); 1445 return (IPPROTO_DONE); 1446 } 1447 1448 /* 1449 * Automatic sizing of receive socket buffer. Often the send 1450 * buffer size is not optimally adjusted to the actual network 1451 * conditions at hand (delay bandwidth product). Setting the 1452 * buffer size too small limits throughput on links with high 1453 * bandwidth and high delay (eg. trans-continental/oceanic links). 1454 * 1455 * On the receive side the socket buffer memory is only rarely 1456 * used to any significant extent. This allows us to be much 1457 * more aggressive in scaling the receive socket buffer. For 1458 * the case that the buffer space is actually used to a large 1459 * extent and we run out of kernel memory we can simply drop 1460 * the new segments; TCP on the sender will just retransmit it 1461 * later. Setting the buffer size too big may only consume too 1462 * much kernel memory if the application doesn't read() from 1463 * the socket or packet loss or reordering makes use of the 1464 * reassembly queue. 1465 * 1466 * The criteria to step up the receive buffer one notch are: 1467 * 1. Application has not set receive buffer size with 1468 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1469 * 2. the number of bytes received during the time it takes 1470 * one timestamp to be reflected back to us (the RTT); 1471 * 3. received bytes per RTT is within seven eighth of the 1472 * current socket buffer size; 1473 * 4. receive buffer size has not hit maximal automatic size; 1474 * 1475 * This algorithm does one step per RTT at most and only if 1476 * we receive a bulk stream w/o packet losses or reorderings. 1477 * Shrinking the buffer during idle times is not necessary as 1478 * it doesn't consume any memory when idle. 1479 * 1480 * TODO: Only step up if the application is actually serving 1481 * the buffer to better manage the socket buffer resources. 1482 */ 1483 int 1484 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1485 struct tcpcb *tp, int tlen) 1486 { 1487 int newsize = 0; 1488 1489 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1490 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1491 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1492 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1493 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1494 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1495 newsize = min(so->so_rcv.sb_hiwat + 1496 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1497 } 1498 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1499 1500 /* Start over with next RTT. */ 1501 tp->rfbuf_ts = 0; 1502 tp->rfbuf_cnt = 0; 1503 } else { 1504 tp->rfbuf_cnt += tlen; /* add up */ 1505 } 1506 1507 return (newsize); 1508 } 1509 1510 void 1511 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1512 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1513 { 1514 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1515 int rstreason, todrop, win; 1516 uint32_t tiwin; 1517 uint16_t nsegs; 1518 char *s; 1519 struct in_conninfo *inc; 1520 struct mbuf *mfree; 1521 struct tcpopt to; 1522 int tfo_syn; 1523 1524 #ifdef TCPDEBUG 1525 /* 1526 * The size of tcp_saveipgen must be the size of the max ip header, 1527 * now IPv6. 1528 */ 1529 u_char tcp_saveipgen[IP6_HDR_LEN]; 1530 struct tcphdr tcp_savetcp; 1531 short ostate = 0; 1532 #endif 1533 thflags = th->th_flags; 1534 inc = &tp->t_inpcb->inp_inc; 1535 tp->sackhint.last_sack_ack = 0; 1536 sack_changed = 0; 1537 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1538 /* 1539 * If this is either a state-changing packet or current state isn't 1540 * established, we require a write lock on tcbinfo. Otherwise, we 1541 * allow the tcbinfo to be in either alocked or unlocked, as the 1542 * caller may have unnecessarily acquired a write lock due to a race. 1543 */ 1544 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1545 tp->t_state != TCPS_ESTABLISHED) { 1546 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1547 } 1548 INP_WLOCK_ASSERT(tp->t_inpcb); 1549 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1550 __func__)); 1551 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1552 __func__)); 1553 1554 #ifdef TCPPCAP 1555 /* Save segment, if requested. */ 1556 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1557 #endif 1558 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1559 tlen, NULL, true); 1560 1561 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1562 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1563 log(LOG_DEBUG, "%s; %s: " 1564 "SYN|FIN segment ignored (based on " 1565 "sysctl setting)\n", s, __func__); 1566 free(s, M_TCPLOG); 1567 } 1568 goto drop; 1569 } 1570 1571 /* 1572 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1573 * check SEQ.ACK first. 1574 */ 1575 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1576 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1577 rstreason = BANDLIM_UNLIMITED; 1578 goto dropwithreset; 1579 } 1580 1581 /* 1582 * Segment received on connection. 1583 * Reset idle time and keep-alive timer. 1584 * XXX: This should be done after segment 1585 * validation to ignore broken/spoofed segs. 1586 */ 1587 tp->t_rcvtime = ticks; 1588 1589 /* 1590 * Scale up the window into a 32-bit value. 1591 * For the SYN_SENT state the scale is zero. 1592 */ 1593 tiwin = th->th_win << tp->snd_scale; 1594 1595 /* 1596 * TCP ECN processing. 1597 */ 1598 if (tp->t_flags & TF_ECN_PERMIT) { 1599 if (thflags & TH_CWR) 1600 tp->t_flags &= ~TF_ECN_SND_ECE; 1601 switch (iptos & IPTOS_ECN_MASK) { 1602 case IPTOS_ECN_CE: 1603 tp->t_flags |= TF_ECN_SND_ECE; 1604 TCPSTAT_INC(tcps_ecn_ce); 1605 break; 1606 case IPTOS_ECN_ECT0: 1607 TCPSTAT_INC(tcps_ecn_ect0); 1608 break; 1609 case IPTOS_ECN_ECT1: 1610 TCPSTAT_INC(tcps_ecn_ect1); 1611 break; 1612 } 1613 1614 /* Process a packet differently from RFC3168. */ 1615 cc_ecnpkt_handler(tp, th, iptos); 1616 1617 /* Congestion experienced. */ 1618 if (thflags & TH_ECE) { 1619 cc_cong_signal(tp, th, CC_ECN); 1620 } 1621 } 1622 1623 /* 1624 * Parse options on any incoming segment. 1625 */ 1626 tcp_dooptions(&to, (u_char *)(th + 1), 1627 (th->th_off << 2) - sizeof(struct tcphdr), 1628 (thflags & TH_SYN) ? TO_SYN : 0); 1629 1630 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1631 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1632 (to.to_flags & TOF_SIGNATURE) == 0) { 1633 TCPSTAT_INC(tcps_sig_err_sigopt); 1634 /* XXX: should drop? */ 1635 } 1636 #endif 1637 /* 1638 * If echoed timestamp is later than the current time, 1639 * fall back to non RFC1323 RTT calculation. Normalize 1640 * timestamp if syncookies were used when this connection 1641 * was established. 1642 */ 1643 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1644 to.to_tsecr -= tp->ts_offset; 1645 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1646 to.to_tsecr = 0; 1647 else if (tp->t_flags & TF_PREVVALID && 1648 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1649 cc_cong_signal(tp, th, CC_RTO_ERR); 1650 } 1651 /* 1652 * Process options only when we get SYN/ACK back. The SYN case 1653 * for incoming connections is handled in tcp_syncache. 1654 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1655 * or <SYN,ACK>) segment itself is never scaled. 1656 * XXX this is traditional behavior, may need to be cleaned up. 1657 */ 1658 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1659 if ((to.to_flags & TOF_SCALE) && 1660 (tp->t_flags & TF_REQ_SCALE)) { 1661 tp->t_flags |= TF_RCVD_SCALE; 1662 tp->snd_scale = to.to_wscale; 1663 } 1664 /* 1665 * Initial send window. It will be updated with 1666 * the next incoming segment to the scaled value. 1667 */ 1668 tp->snd_wnd = th->th_win; 1669 if (to.to_flags & TOF_TS) { 1670 tp->t_flags |= TF_RCVD_TSTMP; 1671 tp->ts_recent = to.to_tsval; 1672 tp->ts_recent_age = tcp_ts_getticks(); 1673 } 1674 if (to.to_flags & TOF_MSS) 1675 tcp_mss(tp, to.to_mss); 1676 if ((tp->t_flags & TF_SACK_PERMIT) && 1677 (to.to_flags & TOF_SACKPERM) == 0) 1678 tp->t_flags &= ~TF_SACK_PERMIT; 1679 if (IS_FASTOPEN(tp->t_flags)) { 1680 if (to.to_flags & TOF_FASTOPEN) { 1681 uint16_t mss; 1682 1683 if (to.to_flags & TOF_MSS) 1684 mss = to.to_mss; 1685 else 1686 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1687 mss = TCP6_MSS; 1688 else 1689 mss = TCP_MSS; 1690 tcp_fastopen_update_cache(tp, mss, 1691 to.to_tfo_len, to.to_tfo_cookie); 1692 } else 1693 tcp_fastopen_disable_path(tp); 1694 } 1695 } 1696 1697 /* 1698 * If timestamps were negotiated during SYN/ACK they should 1699 * appear on every segment during this session and vice versa. 1700 */ 1701 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1702 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1703 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1704 "no action\n", s, __func__); 1705 free(s, M_TCPLOG); 1706 } 1707 } 1708 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1709 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1710 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1711 "no action\n", s, __func__); 1712 free(s, M_TCPLOG); 1713 } 1714 } 1715 1716 /* 1717 * Header prediction: check for the two common cases 1718 * of a uni-directional data xfer. If the packet has 1719 * no control flags, is in-sequence, the window didn't 1720 * change and we're not retransmitting, it's a 1721 * candidate. If the length is zero and the ack moved 1722 * forward, we're the sender side of the xfer. Just 1723 * free the data acked & wake any higher level process 1724 * that was blocked waiting for space. If the length 1725 * is non-zero and the ack didn't move, we're the 1726 * receiver side. If we're getting packets in-order 1727 * (the reassembly queue is empty), add the data to 1728 * the socket buffer and note that we need a delayed ack. 1729 * Make sure that the hidden state-flags are also off. 1730 * Since we check for TCPS_ESTABLISHED first, it can only 1731 * be TH_NEEDSYN. 1732 */ 1733 if (tp->t_state == TCPS_ESTABLISHED && 1734 th->th_seq == tp->rcv_nxt && 1735 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1736 tp->snd_nxt == tp->snd_max && 1737 tiwin && tiwin == tp->snd_wnd && 1738 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1739 SEGQ_EMPTY(tp) && 1740 ((to.to_flags & TOF_TS) == 0 || 1741 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1742 1743 /* 1744 * If last ACK falls within this segment's sequence numbers, 1745 * record the timestamp. 1746 * NOTE that the test is modified according to the latest 1747 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1748 */ 1749 if ((to.to_flags & TOF_TS) != 0 && 1750 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1751 tp->ts_recent_age = tcp_ts_getticks(); 1752 tp->ts_recent = to.to_tsval; 1753 } 1754 1755 if (tlen == 0) { 1756 if (SEQ_GT(th->th_ack, tp->snd_una) && 1757 SEQ_LEQ(th->th_ack, tp->snd_max) && 1758 !IN_RECOVERY(tp->t_flags) && 1759 (to.to_flags & TOF_SACK) == 0 && 1760 TAILQ_EMPTY(&tp->snd_holes)) { 1761 /* 1762 * This is a pure ack for outstanding data. 1763 */ 1764 TCPSTAT_INC(tcps_predack); 1765 1766 /* 1767 * "bad retransmit" recovery without timestamps. 1768 */ 1769 if ((to.to_flags & TOF_TS) == 0 && 1770 tp->t_rxtshift == 1 && 1771 tp->t_flags & TF_PREVVALID && 1772 (int)(ticks - tp->t_badrxtwin) < 0) { 1773 cc_cong_signal(tp, th, CC_RTO_ERR); 1774 } 1775 1776 /* 1777 * Recalculate the transmit timer / rtt. 1778 * 1779 * Some boxes send broken timestamp replies 1780 * during the SYN+ACK phase, ignore 1781 * timestamps of 0 or we could calculate a 1782 * huge RTT and blow up the retransmit timer. 1783 */ 1784 if ((to.to_flags & TOF_TS) != 0 && 1785 to.to_tsecr) { 1786 uint32_t t; 1787 1788 t = tcp_ts_getticks() - to.to_tsecr; 1789 if (!tp->t_rttlow || tp->t_rttlow > t) 1790 tp->t_rttlow = t; 1791 tcp_xmit_timer(tp, 1792 TCP_TS_TO_TICKS(t) + 1); 1793 } else if (tp->t_rtttime && 1794 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1795 if (!tp->t_rttlow || 1796 tp->t_rttlow > ticks - tp->t_rtttime) 1797 tp->t_rttlow = ticks - tp->t_rtttime; 1798 tcp_xmit_timer(tp, 1799 ticks - tp->t_rtttime); 1800 } 1801 acked = BYTES_THIS_ACK(tp, th); 1802 1803 #ifdef TCP_HHOOK 1804 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1805 hhook_run_tcp_est_in(tp, th, &to); 1806 #endif 1807 1808 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1809 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1810 sbdrop(&so->so_snd, acked); 1811 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1812 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1813 tp->snd_recover = th->th_ack - 1; 1814 1815 /* 1816 * Let the congestion control algorithm update 1817 * congestion control related information. This 1818 * typically means increasing the congestion 1819 * window. 1820 */ 1821 cc_ack_received(tp, th, nsegs, CC_ACK); 1822 1823 tp->snd_una = th->th_ack; 1824 /* 1825 * Pull snd_wl2 up to prevent seq wrap relative 1826 * to th_ack. 1827 */ 1828 tp->snd_wl2 = th->th_ack; 1829 tp->t_dupacks = 0; 1830 m_freem(m); 1831 1832 /* 1833 * If all outstanding data are acked, stop 1834 * retransmit timer, otherwise restart timer 1835 * using current (possibly backed-off) value. 1836 * If process is waiting for space, 1837 * wakeup/selwakeup/signal. If data 1838 * are ready to send, let tcp_output 1839 * decide between more output or persist. 1840 */ 1841 #ifdef TCPDEBUG 1842 if (so->so_options & SO_DEBUG) 1843 tcp_trace(TA_INPUT, ostate, tp, 1844 (void *)tcp_saveipgen, 1845 &tcp_savetcp, 0); 1846 #endif 1847 TCP_PROBE3(debug__input, tp, th, m); 1848 if (tp->snd_una == tp->snd_max) 1849 tcp_timer_activate(tp, TT_REXMT, 0); 1850 else if (!tcp_timer_active(tp, TT_PERSIST)) 1851 tcp_timer_activate(tp, TT_REXMT, 1852 tp->t_rxtcur); 1853 sowwakeup(so); 1854 if (sbavail(&so->so_snd)) 1855 (void) tp->t_fb->tfb_tcp_output(tp); 1856 goto check_delack; 1857 } 1858 } else if (th->th_ack == tp->snd_una && 1859 tlen <= sbspace(&so->so_rcv)) { 1860 int newsize = 0; /* automatic sockbuf scaling */ 1861 1862 /* 1863 * This is a pure, in-sequence data packet with 1864 * nothing on the reassembly queue and we have enough 1865 * buffer space to take it. 1866 */ 1867 /* Clean receiver SACK report if present */ 1868 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1869 tcp_clean_sackreport(tp); 1870 TCPSTAT_INC(tcps_preddat); 1871 tp->rcv_nxt += tlen; 1872 /* 1873 * Pull snd_wl1 up to prevent seq wrap relative to 1874 * th_seq. 1875 */ 1876 tp->snd_wl1 = th->th_seq; 1877 /* 1878 * Pull rcv_up up to prevent seq wrap relative to 1879 * rcv_nxt. 1880 */ 1881 tp->rcv_up = tp->rcv_nxt; 1882 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1883 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1884 #ifdef TCPDEBUG 1885 if (so->so_options & SO_DEBUG) 1886 tcp_trace(TA_INPUT, ostate, tp, 1887 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1888 #endif 1889 TCP_PROBE3(debug__input, tp, th, m); 1890 1891 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1892 1893 /* Add data to socket buffer. */ 1894 SOCKBUF_LOCK(&so->so_rcv); 1895 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1896 m_freem(m); 1897 } else { 1898 /* 1899 * Set new socket buffer size. 1900 * Give up when limit is reached. 1901 */ 1902 if (newsize) 1903 if (!sbreserve_locked(&so->so_rcv, 1904 newsize, so, NULL)) 1905 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1906 m_adj(m, drop_hdrlen); /* delayed header drop */ 1907 sbappendstream_locked(&so->so_rcv, m, 0); 1908 } 1909 /* NB: sorwakeup_locked() does an implicit unlock. */ 1910 sorwakeup_locked(so); 1911 if (DELAY_ACK(tp, tlen)) { 1912 tp->t_flags |= TF_DELACK; 1913 } else { 1914 tp->t_flags |= TF_ACKNOW; 1915 tp->t_fb->tfb_tcp_output(tp); 1916 } 1917 goto check_delack; 1918 } 1919 } 1920 1921 /* 1922 * Calculate amount of space in receive window, 1923 * and then do TCP input processing. 1924 * Receive window is amount of space in rcv queue, 1925 * but not less than advertised window. 1926 */ 1927 win = sbspace(&so->so_rcv); 1928 if (win < 0) 1929 win = 0; 1930 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1931 1932 switch (tp->t_state) { 1933 1934 /* 1935 * If the state is SYN_RECEIVED: 1936 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1937 */ 1938 case TCPS_SYN_RECEIVED: 1939 if ((thflags & TH_ACK) && 1940 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1941 SEQ_GT(th->th_ack, tp->snd_max))) { 1942 rstreason = BANDLIM_RST_OPENPORT; 1943 goto dropwithreset; 1944 } 1945 if (IS_FASTOPEN(tp->t_flags)) { 1946 /* 1947 * When a TFO connection is in SYN_RECEIVED, the 1948 * only valid packets are the initial SYN, a 1949 * retransmit/copy of the initial SYN (possibly with 1950 * a subset of the original data), a valid ACK, a 1951 * FIN, or a RST. 1952 */ 1953 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1954 rstreason = BANDLIM_RST_OPENPORT; 1955 goto dropwithreset; 1956 } else if (thflags & TH_SYN) { 1957 /* non-initial SYN is ignored */ 1958 if ((tcp_timer_active(tp, TT_DELACK) || 1959 tcp_timer_active(tp, TT_REXMT))) 1960 goto drop; 1961 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1962 goto drop; 1963 } 1964 } 1965 break; 1966 1967 /* 1968 * If the state is SYN_SENT: 1969 * if seg contains a RST with valid ACK (SEQ.ACK has already 1970 * been verified), then drop the connection. 1971 * if seg contains a RST without an ACK, drop the seg. 1972 * if seg does not contain SYN, then drop the seg. 1973 * Otherwise this is an acceptable SYN segment 1974 * initialize tp->rcv_nxt and tp->irs 1975 * if seg contains ack then advance tp->snd_una 1976 * if seg contains an ECE and ECN support is enabled, the stream 1977 * is ECN capable. 1978 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1979 * arrange for segment to be acked (eventually) 1980 * continue processing rest of data/controls, beginning with URG 1981 */ 1982 case TCPS_SYN_SENT: 1983 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1984 TCP_PROBE5(connect__refused, NULL, tp, 1985 m, tp, th); 1986 tp = tcp_drop(tp, ECONNREFUSED); 1987 } 1988 if (thflags & TH_RST) 1989 goto drop; 1990 if (!(thflags & TH_SYN)) 1991 goto drop; 1992 1993 tp->irs = th->th_seq; 1994 tcp_rcvseqinit(tp); 1995 if (thflags & TH_ACK) { 1996 int tfo_partial_ack = 0; 1997 1998 TCPSTAT_INC(tcps_connects); 1999 soisconnected(so); 2000 #ifdef MAC 2001 mac_socketpeer_set_from_mbuf(m, so); 2002 #endif 2003 /* Do window scaling on this connection? */ 2004 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2005 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2006 tp->rcv_scale = tp->request_r_scale; 2007 } 2008 tp->rcv_adv += min(tp->rcv_wnd, 2009 TCP_MAXWIN << tp->rcv_scale); 2010 tp->snd_una++; /* SYN is acked */ 2011 /* 2012 * If not all the data that was sent in the TFO SYN 2013 * has been acked, resend the remainder right away. 2014 */ 2015 if (IS_FASTOPEN(tp->t_flags) && 2016 (tp->snd_una != tp->snd_max)) { 2017 tp->snd_nxt = th->th_ack; 2018 tfo_partial_ack = 1; 2019 } 2020 /* 2021 * If there's data, delay ACK; if there's also a FIN 2022 * ACKNOW will be turned on later. 2023 */ 2024 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2025 tcp_timer_activate(tp, TT_DELACK, 2026 tcp_delacktime); 2027 else 2028 tp->t_flags |= TF_ACKNOW; 2029 2030 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2031 tp->t_flags |= TF_ECN_PERMIT; 2032 TCPSTAT_INC(tcps_ecn_shs); 2033 } 2034 2035 /* 2036 * Received <SYN,ACK> in SYN_SENT[*] state. 2037 * Transitions: 2038 * SYN_SENT --> ESTABLISHED 2039 * SYN_SENT* --> FIN_WAIT_1 2040 */ 2041 tp->t_starttime = ticks; 2042 if (tp->t_flags & TF_NEEDFIN) { 2043 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2044 tp->t_flags &= ~TF_NEEDFIN; 2045 thflags &= ~TH_SYN; 2046 } else { 2047 tcp_state_change(tp, TCPS_ESTABLISHED); 2048 TCP_PROBE5(connect__established, NULL, tp, 2049 m, tp, th); 2050 cc_conn_init(tp); 2051 tcp_timer_activate(tp, TT_KEEP, 2052 TP_KEEPIDLE(tp)); 2053 } 2054 } else { 2055 /* 2056 * Received initial SYN in SYN-SENT[*] state => 2057 * simultaneous open. 2058 * If it succeeds, connection is * half-synchronized. 2059 * Otherwise, do 3-way handshake: 2060 * SYN-SENT -> SYN-RECEIVED 2061 * SYN-SENT* -> SYN-RECEIVED* 2062 */ 2063 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2064 tcp_timer_activate(tp, TT_REXMT, 0); 2065 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2066 } 2067 2068 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2069 INP_WLOCK_ASSERT(tp->t_inpcb); 2070 2071 /* 2072 * Advance th->th_seq to correspond to first data byte. 2073 * If data, trim to stay within window, 2074 * dropping FIN if necessary. 2075 */ 2076 th->th_seq++; 2077 if (tlen > tp->rcv_wnd) { 2078 todrop = tlen - tp->rcv_wnd; 2079 m_adj(m, -todrop); 2080 tlen = tp->rcv_wnd; 2081 thflags &= ~TH_FIN; 2082 TCPSTAT_INC(tcps_rcvpackafterwin); 2083 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2084 } 2085 tp->snd_wl1 = th->th_seq - 1; 2086 tp->rcv_up = th->th_seq; 2087 /* 2088 * Client side of transaction: already sent SYN and data. 2089 * If the remote host used T/TCP to validate the SYN, 2090 * our data will be ACK'd; if so, enter normal data segment 2091 * processing in the middle of step 5, ack processing. 2092 * Otherwise, goto step 6. 2093 */ 2094 if (thflags & TH_ACK) 2095 goto process_ACK; 2096 2097 goto step6; 2098 2099 /* 2100 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2101 * do normal processing. 2102 * 2103 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2104 */ 2105 case TCPS_LAST_ACK: 2106 case TCPS_CLOSING: 2107 break; /* continue normal processing */ 2108 } 2109 2110 /* 2111 * States other than LISTEN or SYN_SENT. 2112 * First check the RST flag and sequence number since reset segments 2113 * are exempt from the timestamp and connection count tests. This 2114 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2115 * below which allowed reset segments in half the sequence space 2116 * to fall though and be processed (which gives forged reset 2117 * segments with a random sequence number a 50 percent chance of 2118 * killing a connection). 2119 * Then check timestamp, if present. 2120 * Then check the connection count, if present. 2121 * Then check that at least some bytes of segment are within 2122 * receive window. If segment begins before rcv_nxt, 2123 * drop leading data (and SYN); if nothing left, just ack. 2124 */ 2125 if (thflags & TH_RST) { 2126 /* 2127 * RFC5961 Section 3.2 2128 * 2129 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2130 * - If RST is in window, we send challenge ACK. 2131 * 2132 * Note: to take into account delayed ACKs, we should 2133 * test against last_ack_sent instead of rcv_nxt. 2134 * Note 2: we handle special case of closed window, not 2135 * covered by the RFC. 2136 */ 2137 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2138 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2139 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2140 2141 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2142 KASSERT(tp->t_state != TCPS_SYN_SENT, 2143 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2144 __func__, th, tp)); 2145 2146 if (V_tcp_insecure_rst || 2147 tp->last_ack_sent == th->th_seq) { 2148 TCPSTAT_INC(tcps_drops); 2149 /* Drop the connection. */ 2150 switch (tp->t_state) { 2151 case TCPS_SYN_RECEIVED: 2152 so->so_error = ECONNREFUSED; 2153 goto close; 2154 case TCPS_ESTABLISHED: 2155 case TCPS_FIN_WAIT_1: 2156 case TCPS_FIN_WAIT_2: 2157 case TCPS_CLOSE_WAIT: 2158 case TCPS_CLOSING: 2159 case TCPS_LAST_ACK: 2160 so->so_error = ECONNRESET; 2161 close: 2162 /* FALLTHROUGH */ 2163 default: 2164 tp = tcp_close(tp); 2165 } 2166 } else { 2167 TCPSTAT_INC(tcps_badrst); 2168 /* Send challenge ACK. */ 2169 tcp_respond(tp, mtod(m, void *), th, m, 2170 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2171 tp->last_ack_sent = tp->rcv_nxt; 2172 m = NULL; 2173 } 2174 } 2175 goto drop; 2176 } 2177 2178 /* 2179 * RFC5961 Section 4.2 2180 * Send challenge ACK for any SYN in synchronized state. 2181 */ 2182 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2183 tp->t_state != TCPS_SYN_RECEIVED) { 2184 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2185 2186 TCPSTAT_INC(tcps_badsyn); 2187 if (V_tcp_insecure_syn && 2188 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2189 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2190 tp = tcp_drop(tp, ECONNRESET); 2191 rstreason = BANDLIM_UNLIMITED; 2192 } else { 2193 /* Send challenge ACK. */ 2194 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2195 tp->snd_nxt, TH_ACK); 2196 tp->last_ack_sent = tp->rcv_nxt; 2197 m = NULL; 2198 } 2199 goto drop; 2200 } 2201 2202 /* 2203 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2204 * and it's less than ts_recent, drop it. 2205 */ 2206 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2207 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2208 2209 /* Check to see if ts_recent is over 24 days old. */ 2210 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2211 /* 2212 * Invalidate ts_recent. If this segment updates 2213 * ts_recent, the age will be reset later and ts_recent 2214 * will get a valid value. If it does not, setting 2215 * ts_recent to zero will at least satisfy the 2216 * requirement that zero be placed in the timestamp 2217 * echo reply when ts_recent isn't valid. The 2218 * age isn't reset until we get a valid ts_recent 2219 * because we don't want out-of-order segments to be 2220 * dropped when ts_recent is old. 2221 */ 2222 tp->ts_recent = 0; 2223 } else { 2224 TCPSTAT_INC(tcps_rcvduppack); 2225 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2226 TCPSTAT_INC(tcps_pawsdrop); 2227 if (tlen) 2228 goto dropafterack; 2229 goto drop; 2230 } 2231 } 2232 2233 /* 2234 * In the SYN-RECEIVED state, validate that the packet belongs to 2235 * this connection before trimming the data to fit the receive 2236 * window. Check the sequence number versus IRS since we know 2237 * the sequence numbers haven't wrapped. This is a partial fix 2238 * for the "LAND" DoS attack. 2239 */ 2240 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2241 rstreason = BANDLIM_RST_OPENPORT; 2242 goto dropwithreset; 2243 } 2244 2245 todrop = tp->rcv_nxt - th->th_seq; 2246 if (todrop > 0) { 2247 if (thflags & TH_SYN) { 2248 thflags &= ~TH_SYN; 2249 th->th_seq++; 2250 if (th->th_urp > 1) 2251 th->th_urp--; 2252 else 2253 thflags &= ~TH_URG; 2254 todrop--; 2255 } 2256 /* 2257 * Following if statement from Stevens, vol. 2, p. 960. 2258 */ 2259 if (todrop > tlen 2260 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2261 /* 2262 * Any valid FIN must be to the left of the window. 2263 * At this point the FIN must be a duplicate or out 2264 * of sequence; drop it. 2265 */ 2266 thflags &= ~TH_FIN; 2267 2268 /* 2269 * Send an ACK to resynchronize and drop any data. 2270 * But keep on processing for RST or ACK. 2271 */ 2272 tp->t_flags |= TF_ACKNOW; 2273 todrop = tlen; 2274 TCPSTAT_INC(tcps_rcvduppack); 2275 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2276 } else { 2277 TCPSTAT_INC(tcps_rcvpartduppack); 2278 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2279 } 2280 drop_hdrlen += todrop; /* drop from the top afterwards */ 2281 th->th_seq += todrop; 2282 tlen -= todrop; 2283 if (th->th_urp > todrop) 2284 th->th_urp -= todrop; 2285 else { 2286 thflags &= ~TH_URG; 2287 th->th_urp = 0; 2288 } 2289 } 2290 2291 /* 2292 * If new data are received on a connection after the 2293 * user processes are gone, then RST the other end. 2294 */ 2295 if ((so->so_state & SS_NOFDREF) && 2296 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2297 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2298 2299 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2300 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2301 "after socket was closed, " 2302 "sending RST and removing tcpcb\n", 2303 s, __func__, tcpstates[tp->t_state], tlen); 2304 free(s, M_TCPLOG); 2305 } 2306 tp = tcp_close(tp); 2307 TCPSTAT_INC(tcps_rcvafterclose); 2308 rstreason = BANDLIM_UNLIMITED; 2309 goto dropwithreset; 2310 } 2311 2312 /* 2313 * If segment ends after window, drop trailing data 2314 * (and PUSH and FIN); if nothing left, just ACK. 2315 */ 2316 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2317 if (todrop > 0) { 2318 TCPSTAT_INC(tcps_rcvpackafterwin); 2319 if (todrop >= tlen) { 2320 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2321 /* 2322 * If window is closed can only take segments at 2323 * window edge, and have to drop data and PUSH from 2324 * incoming segments. Continue processing, but 2325 * remember to ack. Otherwise, drop segment 2326 * and ack. 2327 */ 2328 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2329 tp->t_flags |= TF_ACKNOW; 2330 TCPSTAT_INC(tcps_rcvwinprobe); 2331 } else 2332 goto dropafterack; 2333 } else 2334 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2335 m_adj(m, -todrop); 2336 tlen -= todrop; 2337 thflags &= ~(TH_PUSH|TH_FIN); 2338 } 2339 2340 /* 2341 * If last ACK falls within this segment's sequence numbers, 2342 * record its timestamp. 2343 * NOTE: 2344 * 1) That the test incorporates suggestions from the latest 2345 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2346 * 2) That updating only on newer timestamps interferes with 2347 * our earlier PAWS tests, so this check should be solely 2348 * predicated on the sequence space of this segment. 2349 * 3) That we modify the segment boundary check to be 2350 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2351 * instead of RFC1323's 2352 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2353 * This modified check allows us to overcome RFC1323's 2354 * limitations as described in Stevens TCP/IP Illustrated 2355 * Vol. 2 p.869. In such cases, we can still calculate the 2356 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2357 */ 2358 if ((to.to_flags & TOF_TS) != 0 && 2359 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2360 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2361 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2362 tp->ts_recent_age = tcp_ts_getticks(); 2363 tp->ts_recent = to.to_tsval; 2364 } 2365 2366 /* 2367 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2368 * flag is on (half-synchronized state), then queue data for 2369 * later processing; else drop segment and return. 2370 */ 2371 if ((thflags & TH_ACK) == 0) { 2372 if (tp->t_state == TCPS_SYN_RECEIVED || 2373 (tp->t_flags & TF_NEEDSYN)) { 2374 if (tp->t_state == TCPS_SYN_RECEIVED && 2375 IS_FASTOPEN(tp->t_flags)) { 2376 tp->snd_wnd = tiwin; 2377 cc_conn_init(tp); 2378 } 2379 goto step6; 2380 } else if (tp->t_flags & TF_ACKNOW) 2381 goto dropafterack; 2382 else 2383 goto drop; 2384 } 2385 2386 /* 2387 * Ack processing. 2388 */ 2389 switch (tp->t_state) { 2390 2391 /* 2392 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2393 * ESTABLISHED state and continue processing. 2394 * The ACK was checked above. 2395 */ 2396 case TCPS_SYN_RECEIVED: 2397 2398 TCPSTAT_INC(tcps_connects); 2399 soisconnected(so); 2400 /* Do window scaling? */ 2401 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2402 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2403 tp->rcv_scale = tp->request_r_scale; 2404 tp->snd_wnd = tiwin; 2405 } 2406 /* 2407 * Make transitions: 2408 * SYN-RECEIVED -> ESTABLISHED 2409 * SYN-RECEIVED* -> FIN-WAIT-1 2410 */ 2411 tp->t_starttime = ticks; 2412 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2413 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2414 tp->t_tfo_pending = NULL; 2415 2416 /* 2417 * Account for the ACK of our SYN prior to 2418 * regular ACK processing below. 2419 */ 2420 tp->snd_una++; 2421 } 2422 if (tp->t_flags & TF_NEEDFIN) { 2423 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2424 tp->t_flags &= ~TF_NEEDFIN; 2425 } else { 2426 tcp_state_change(tp, TCPS_ESTABLISHED); 2427 TCP_PROBE5(accept__established, NULL, tp, 2428 m, tp, th); 2429 /* 2430 * TFO connections call cc_conn_init() during SYN 2431 * processing. Calling it again here for such 2432 * connections is not harmless as it would undo the 2433 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2434 * is retransmitted. 2435 */ 2436 if (!IS_FASTOPEN(tp->t_flags)) 2437 cc_conn_init(tp); 2438 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2439 } 2440 /* 2441 * If segment contains data or ACK, will call tcp_reass() 2442 * later; if not, do so now to pass queued data to user. 2443 */ 2444 if (tlen == 0 && (thflags & TH_FIN) == 0) 2445 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2446 (struct mbuf *)0); 2447 tp->snd_wl1 = th->th_seq - 1; 2448 /* FALLTHROUGH */ 2449 2450 /* 2451 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2452 * ACKs. If the ack is in the range 2453 * tp->snd_una < th->th_ack <= tp->snd_max 2454 * then advance tp->snd_una to th->th_ack and drop 2455 * data from the retransmission queue. If this ACK reflects 2456 * more up to date window information we update our window information. 2457 */ 2458 case TCPS_ESTABLISHED: 2459 case TCPS_FIN_WAIT_1: 2460 case TCPS_FIN_WAIT_2: 2461 case TCPS_CLOSE_WAIT: 2462 case TCPS_CLOSING: 2463 case TCPS_LAST_ACK: 2464 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2465 TCPSTAT_INC(tcps_rcvacktoomuch); 2466 goto dropafterack; 2467 } 2468 if ((tp->t_flags & TF_SACK_PERMIT) && 2469 ((to.to_flags & TOF_SACK) || 2470 !TAILQ_EMPTY(&tp->snd_holes))) 2471 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2472 else 2473 /* 2474 * Reset the value so that previous (valid) value 2475 * from the last ack with SACK doesn't get used. 2476 */ 2477 tp->sackhint.sacked_bytes = 0; 2478 2479 #ifdef TCP_HHOOK 2480 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2481 hhook_run_tcp_est_in(tp, th, &to); 2482 #endif 2483 2484 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2485 u_int maxseg; 2486 2487 maxseg = tcp_maxseg(tp); 2488 if (tlen == 0 && 2489 (tiwin == tp->snd_wnd || 2490 (tp->t_flags & TF_SACK_PERMIT))) { 2491 /* 2492 * If this is the first time we've seen a 2493 * FIN from the remote, this is not a 2494 * duplicate and it needs to be processed 2495 * normally. This happens during a 2496 * simultaneous close. 2497 */ 2498 if ((thflags & TH_FIN) && 2499 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2500 tp->t_dupacks = 0; 2501 break; 2502 } 2503 TCPSTAT_INC(tcps_rcvdupack); 2504 /* 2505 * If we have outstanding data (other than 2506 * a window probe), this is a completely 2507 * duplicate ack (ie, window info didn't 2508 * change and FIN isn't set), 2509 * the ack is the biggest we've 2510 * seen and we've seen exactly our rexmt 2511 * threshold of them, assume a packet 2512 * has been dropped and retransmit it. 2513 * Kludge snd_nxt & the congestion 2514 * window so we send only this one 2515 * packet. 2516 * 2517 * We know we're losing at the current 2518 * window size so do congestion avoidance 2519 * (set ssthresh to half the current window 2520 * and pull our congestion window back to 2521 * the new ssthresh). 2522 * 2523 * Dup acks mean that packets have left the 2524 * network (they're now cached at the receiver) 2525 * so bump cwnd by the amount in the receiver 2526 * to keep a constant cwnd packets in the 2527 * network. 2528 * 2529 * When using TCP ECN, notify the peer that 2530 * we reduced the cwnd. 2531 */ 2532 /* 2533 * Following 2 kinds of acks should not affect 2534 * dupack counting: 2535 * 1) Old acks 2536 * 2) Acks with SACK but without any new SACK 2537 * information in them. These could result from 2538 * any anomaly in the network like a switch 2539 * duplicating packets or a possible DoS attack. 2540 */ 2541 if (th->th_ack != tp->snd_una || 2542 ((tp->t_flags & TF_SACK_PERMIT) && 2543 !sack_changed)) 2544 break; 2545 else if (!tcp_timer_active(tp, TT_REXMT)) 2546 tp->t_dupacks = 0; 2547 else if (++tp->t_dupacks > tcprexmtthresh || 2548 IN_FASTRECOVERY(tp->t_flags)) { 2549 cc_ack_received(tp, th, nsegs, 2550 CC_DUPACK); 2551 if ((tp->t_flags & TF_SACK_PERMIT) && 2552 IN_FASTRECOVERY(tp->t_flags)) { 2553 int awnd; 2554 2555 /* 2556 * Compute the amount of data in flight first. 2557 * We can inject new data into the pipe iff 2558 * we have less than 1/2 the original window's 2559 * worth of data in flight. 2560 */ 2561 if (V_tcp_do_rfc6675_pipe) 2562 awnd = tcp_compute_pipe(tp); 2563 else 2564 awnd = (tp->snd_nxt - tp->snd_fack) + 2565 tp->sackhint.sack_bytes_rexmit; 2566 2567 if (awnd < tp->snd_ssthresh) { 2568 tp->snd_cwnd += maxseg; 2569 if (tp->snd_cwnd > tp->snd_ssthresh) 2570 tp->snd_cwnd = tp->snd_ssthresh; 2571 } 2572 } else 2573 tp->snd_cwnd += maxseg; 2574 (void) tp->t_fb->tfb_tcp_output(tp); 2575 goto drop; 2576 } else if (tp->t_dupacks == tcprexmtthresh) { 2577 tcp_seq onxt = tp->snd_nxt; 2578 2579 /* 2580 * If we're doing sack, check to 2581 * see if we're already in sack 2582 * recovery. If we're not doing sack, 2583 * check to see if we're in newreno 2584 * recovery. 2585 */ 2586 if (tp->t_flags & TF_SACK_PERMIT) { 2587 if (IN_FASTRECOVERY(tp->t_flags)) { 2588 tp->t_dupacks = 0; 2589 break; 2590 } 2591 } else { 2592 if (SEQ_LEQ(th->th_ack, 2593 tp->snd_recover)) { 2594 tp->t_dupacks = 0; 2595 break; 2596 } 2597 } 2598 /* Congestion signal before ack. */ 2599 cc_cong_signal(tp, th, CC_NDUPACK); 2600 cc_ack_received(tp, th, nsegs, 2601 CC_DUPACK); 2602 tcp_timer_activate(tp, TT_REXMT, 0); 2603 tp->t_rtttime = 0; 2604 if (tp->t_flags & TF_SACK_PERMIT) { 2605 TCPSTAT_INC( 2606 tcps_sack_recovery_episode); 2607 tp->sack_newdata = tp->snd_nxt; 2608 tp->snd_cwnd = maxseg; 2609 (void) tp->t_fb->tfb_tcp_output(tp); 2610 goto drop; 2611 } 2612 tp->snd_nxt = th->th_ack; 2613 tp->snd_cwnd = maxseg; 2614 (void) tp->t_fb->tfb_tcp_output(tp); 2615 KASSERT(tp->snd_limited <= 2, 2616 ("%s: tp->snd_limited too big", 2617 __func__)); 2618 tp->snd_cwnd = tp->snd_ssthresh + 2619 maxseg * 2620 (tp->t_dupacks - tp->snd_limited); 2621 if (SEQ_GT(onxt, tp->snd_nxt)) 2622 tp->snd_nxt = onxt; 2623 goto drop; 2624 } else if (V_tcp_do_rfc3042) { 2625 /* 2626 * Process first and second duplicate 2627 * ACKs. Each indicates a segment 2628 * leaving the network, creating room 2629 * for more. Make sure we can send a 2630 * packet on reception of each duplicate 2631 * ACK by increasing snd_cwnd by one 2632 * segment. Restore the original 2633 * snd_cwnd after packet transmission. 2634 */ 2635 cc_ack_received(tp, th, nsegs, 2636 CC_DUPACK); 2637 uint32_t oldcwnd = tp->snd_cwnd; 2638 tcp_seq oldsndmax = tp->snd_max; 2639 u_int sent; 2640 int avail; 2641 2642 KASSERT(tp->t_dupacks == 1 || 2643 tp->t_dupacks == 2, 2644 ("%s: dupacks not 1 or 2", 2645 __func__)); 2646 if (tp->t_dupacks == 1) 2647 tp->snd_limited = 0; 2648 tp->snd_cwnd = 2649 (tp->snd_nxt - tp->snd_una) + 2650 (tp->t_dupacks - tp->snd_limited) * 2651 maxseg; 2652 /* 2653 * Only call tcp_output when there 2654 * is new data available to be sent. 2655 * Otherwise we would send pure ACKs. 2656 */ 2657 SOCKBUF_LOCK(&so->so_snd); 2658 avail = sbavail(&so->so_snd) - 2659 (tp->snd_nxt - tp->snd_una); 2660 SOCKBUF_UNLOCK(&so->so_snd); 2661 if (avail > 0) 2662 (void) tp->t_fb->tfb_tcp_output(tp); 2663 sent = tp->snd_max - oldsndmax; 2664 if (sent > maxseg) { 2665 KASSERT((tp->t_dupacks == 2 && 2666 tp->snd_limited == 0) || 2667 (sent == maxseg + 1 && 2668 tp->t_flags & TF_SENTFIN), 2669 ("%s: sent too much", 2670 __func__)); 2671 tp->snd_limited = 2; 2672 } else if (sent > 0) 2673 ++tp->snd_limited; 2674 tp->snd_cwnd = oldcwnd; 2675 goto drop; 2676 } 2677 } 2678 break; 2679 } else { 2680 /* 2681 * This ack is advancing the left edge, reset the 2682 * counter. 2683 */ 2684 tp->t_dupacks = 0; 2685 /* 2686 * If this ack also has new SACK info, increment the 2687 * counter as per rfc6675. 2688 */ 2689 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2690 tp->t_dupacks++; 2691 } 2692 2693 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2694 ("%s: th_ack <= snd_una", __func__)); 2695 2696 /* 2697 * If the congestion window was inflated to account 2698 * for the other side's cached packets, retract it. 2699 */ 2700 if (IN_FASTRECOVERY(tp->t_flags)) { 2701 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2702 if (tp->t_flags & TF_SACK_PERMIT) 2703 tcp_sack_partialack(tp, th); 2704 else 2705 tcp_newreno_partial_ack(tp, th); 2706 } else 2707 cc_post_recovery(tp, th); 2708 } 2709 /* 2710 * If we reach this point, ACK is not a duplicate, 2711 * i.e., it ACKs something we sent. 2712 */ 2713 if (tp->t_flags & TF_NEEDSYN) { 2714 /* 2715 * T/TCP: Connection was half-synchronized, and our 2716 * SYN has been ACK'd (so connection is now fully 2717 * synchronized). Go to non-starred state, 2718 * increment snd_una for ACK of SYN, and check if 2719 * we can do window scaling. 2720 */ 2721 tp->t_flags &= ~TF_NEEDSYN; 2722 tp->snd_una++; 2723 /* Do window scaling? */ 2724 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2725 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2726 tp->rcv_scale = tp->request_r_scale; 2727 /* Send window already scaled. */ 2728 } 2729 } 2730 2731 process_ACK: 2732 INP_WLOCK_ASSERT(tp->t_inpcb); 2733 2734 acked = BYTES_THIS_ACK(tp, th); 2735 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2736 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2737 tp->snd_una, th->th_ack, tp, m)); 2738 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2739 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2740 2741 /* 2742 * If we just performed our first retransmit, and the ACK 2743 * arrives within our recovery window, then it was a mistake 2744 * to do the retransmit in the first place. Recover our 2745 * original cwnd and ssthresh, and proceed to transmit where 2746 * we left off. 2747 */ 2748 if (tp->t_rxtshift == 1 && 2749 tp->t_flags & TF_PREVVALID && 2750 tp->t_badrxtwin && 2751 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2752 cc_cong_signal(tp, th, CC_RTO_ERR); 2753 2754 /* 2755 * If we have a timestamp reply, update smoothed 2756 * round trip time. If no timestamp is present but 2757 * transmit timer is running and timed sequence 2758 * number was acked, update smoothed round trip time. 2759 * Since we now have an rtt measurement, cancel the 2760 * timer backoff (cf., Phil Karn's retransmit alg.). 2761 * Recompute the initial retransmit timer. 2762 * 2763 * Some boxes send broken timestamp replies 2764 * during the SYN+ACK phase, ignore 2765 * timestamps of 0 or we could calculate a 2766 * huge RTT and blow up the retransmit timer. 2767 */ 2768 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2769 uint32_t t; 2770 2771 t = tcp_ts_getticks() - to.to_tsecr; 2772 if (!tp->t_rttlow || tp->t_rttlow > t) 2773 tp->t_rttlow = t; 2774 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2775 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2776 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2777 tp->t_rttlow = ticks - tp->t_rtttime; 2778 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2779 } 2780 2781 /* 2782 * If all outstanding data is acked, stop retransmit 2783 * timer and remember to restart (more output or persist). 2784 * If there is more data to be acked, restart retransmit 2785 * timer, using current (possibly backed-off) value. 2786 */ 2787 if (th->th_ack == tp->snd_max) { 2788 tcp_timer_activate(tp, TT_REXMT, 0); 2789 needoutput = 1; 2790 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2791 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2792 2793 /* 2794 * If no data (only SYN) was ACK'd, 2795 * skip rest of ACK processing. 2796 */ 2797 if (acked == 0) 2798 goto step6; 2799 2800 /* 2801 * Let the congestion control algorithm update congestion 2802 * control related information. This typically means increasing 2803 * the congestion window. 2804 */ 2805 cc_ack_received(tp, th, nsegs, CC_ACK); 2806 2807 SOCKBUF_LOCK(&so->so_snd); 2808 if (acked > sbavail(&so->so_snd)) { 2809 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2810 tp->snd_wnd -= sbavail(&so->so_snd); 2811 else 2812 tp->snd_wnd = 0; 2813 mfree = sbcut_locked(&so->so_snd, 2814 (int)sbavail(&so->so_snd)); 2815 ourfinisacked = 1; 2816 } else { 2817 mfree = sbcut_locked(&so->so_snd, acked); 2818 if (tp->snd_wnd >= (uint32_t) acked) 2819 tp->snd_wnd -= acked; 2820 else 2821 tp->snd_wnd = 0; 2822 ourfinisacked = 0; 2823 } 2824 /* NB: sowwakeup_locked() does an implicit unlock. */ 2825 sowwakeup_locked(so); 2826 m_freem(mfree); 2827 /* Detect una wraparound. */ 2828 if (!IN_RECOVERY(tp->t_flags) && 2829 SEQ_GT(tp->snd_una, tp->snd_recover) && 2830 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2831 tp->snd_recover = th->th_ack - 1; 2832 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2833 if (IN_RECOVERY(tp->t_flags) && 2834 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2835 EXIT_RECOVERY(tp->t_flags); 2836 } 2837 tp->snd_una = th->th_ack; 2838 if (tp->t_flags & TF_SACK_PERMIT) { 2839 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2840 tp->snd_recover = tp->snd_una; 2841 } 2842 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2843 tp->snd_nxt = tp->snd_una; 2844 2845 switch (tp->t_state) { 2846 2847 /* 2848 * In FIN_WAIT_1 STATE in addition to the processing 2849 * for the ESTABLISHED state if our FIN is now acknowledged 2850 * then enter FIN_WAIT_2. 2851 */ 2852 case TCPS_FIN_WAIT_1: 2853 if (ourfinisacked) { 2854 /* 2855 * If we can't receive any more 2856 * data, then closing user can proceed. 2857 * Starting the timer is contrary to the 2858 * specification, but if we don't get a FIN 2859 * we'll hang forever. 2860 * 2861 * XXXjl: 2862 * we should release the tp also, and use a 2863 * compressed state. 2864 */ 2865 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2866 soisdisconnected(so); 2867 tcp_timer_activate(tp, TT_2MSL, 2868 (tcp_fast_finwait2_recycle ? 2869 tcp_finwait2_timeout : 2870 TP_MAXIDLE(tp))); 2871 } 2872 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2873 } 2874 break; 2875 2876 /* 2877 * In CLOSING STATE in addition to the processing for 2878 * the ESTABLISHED state if the ACK acknowledges our FIN 2879 * then enter the TIME-WAIT state, otherwise ignore 2880 * the segment. 2881 */ 2882 case TCPS_CLOSING: 2883 if (ourfinisacked) { 2884 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2885 tcp_twstart(tp); 2886 m_freem(m); 2887 return; 2888 } 2889 break; 2890 2891 /* 2892 * In LAST_ACK, we may still be waiting for data to drain 2893 * and/or to be acked, as well as for the ack of our FIN. 2894 * If our FIN is now acknowledged, delete the TCB, 2895 * enter the closed state and return. 2896 */ 2897 case TCPS_LAST_ACK: 2898 if (ourfinisacked) { 2899 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2900 tp = tcp_close(tp); 2901 goto drop; 2902 } 2903 break; 2904 } 2905 } 2906 2907 step6: 2908 INP_WLOCK_ASSERT(tp->t_inpcb); 2909 2910 /* 2911 * Update window information. 2912 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2913 */ 2914 if ((thflags & TH_ACK) && 2915 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2916 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2917 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2918 /* keep track of pure window updates */ 2919 if (tlen == 0 && 2920 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2921 TCPSTAT_INC(tcps_rcvwinupd); 2922 tp->snd_wnd = tiwin; 2923 tp->snd_wl1 = th->th_seq; 2924 tp->snd_wl2 = th->th_ack; 2925 if (tp->snd_wnd > tp->max_sndwnd) 2926 tp->max_sndwnd = tp->snd_wnd; 2927 needoutput = 1; 2928 } 2929 2930 /* 2931 * Process segments with URG. 2932 */ 2933 if ((thflags & TH_URG) && th->th_urp && 2934 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2935 /* 2936 * This is a kludge, but if we receive and accept 2937 * random urgent pointers, we'll crash in 2938 * soreceive. It's hard to imagine someone 2939 * actually wanting to send this much urgent data. 2940 */ 2941 SOCKBUF_LOCK(&so->so_rcv); 2942 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2943 th->th_urp = 0; /* XXX */ 2944 thflags &= ~TH_URG; /* XXX */ 2945 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2946 goto dodata; /* XXX */ 2947 } 2948 /* 2949 * If this segment advances the known urgent pointer, 2950 * then mark the data stream. This should not happen 2951 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2952 * a FIN has been received from the remote side. 2953 * In these states we ignore the URG. 2954 * 2955 * According to RFC961 (Assigned Protocols), 2956 * the urgent pointer points to the last octet 2957 * of urgent data. We continue, however, 2958 * to consider it to indicate the first octet 2959 * of data past the urgent section as the original 2960 * spec states (in one of two places). 2961 */ 2962 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2963 tp->rcv_up = th->th_seq + th->th_urp; 2964 so->so_oobmark = sbavail(&so->so_rcv) + 2965 (tp->rcv_up - tp->rcv_nxt) - 1; 2966 if (so->so_oobmark == 0) 2967 so->so_rcv.sb_state |= SBS_RCVATMARK; 2968 sohasoutofband(so); 2969 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2970 } 2971 SOCKBUF_UNLOCK(&so->so_rcv); 2972 /* 2973 * Remove out of band data so doesn't get presented to user. 2974 * This can happen independent of advancing the URG pointer, 2975 * but if two URG's are pending at once, some out-of-band 2976 * data may creep in... ick. 2977 */ 2978 if (th->th_urp <= (uint32_t)tlen && 2979 !(so->so_options & SO_OOBINLINE)) { 2980 /* hdr drop is delayed */ 2981 tcp_pulloutofband(so, th, m, drop_hdrlen); 2982 } 2983 } else { 2984 /* 2985 * If no out of band data is expected, 2986 * pull receive urgent pointer along 2987 * with the receive window. 2988 */ 2989 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2990 tp->rcv_up = tp->rcv_nxt; 2991 } 2992 dodata: /* XXX */ 2993 INP_WLOCK_ASSERT(tp->t_inpcb); 2994 2995 /* 2996 * Process the segment text, merging it into the TCP sequencing queue, 2997 * and arranging for acknowledgment of receipt if necessary. 2998 * This process logically involves adjusting tp->rcv_wnd as data 2999 * is presented to the user (this happens in tcp_usrreq.c, 3000 * case PRU_RCVD). If a FIN has already been received on this 3001 * connection then we just ignore the text. 3002 */ 3003 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3004 IS_FASTOPEN(tp->t_flags)); 3005 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3006 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3007 tcp_seq save_start = th->th_seq; 3008 m_adj(m, drop_hdrlen); /* delayed header drop */ 3009 /* 3010 * Insert segment which includes th into TCP reassembly queue 3011 * with control block tp. Set thflags to whether reassembly now 3012 * includes a segment with FIN. This handles the common case 3013 * inline (segment is the next to be received on an established 3014 * connection, and the queue is empty), avoiding linkage into 3015 * and removal from the queue and repetition of various 3016 * conversions. 3017 * Set DELACK for segments received in order, but ack 3018 * immediately when segments are out of order (so 3019 * fast retransmit can work). 3020 */ 3021 if (th->th_seq == tp->rcv_nxt && 3022 SEGQ_EMPTY(tp) && 3023 (TCPS_HAVEESTABLISHED(tp->t_state) || 3024 tfo_syn)) { 3025 if (DELAY_ACK(tp, tlen) || tfo_syn) 3026 tp->t_flags |= TF_DELACK; 3027 else 3028 tp->t_flags |= TF_ACKNOW; 3029 tp->rcv_nxt += tlen; 3030 thflags = th->th_flags & TH_FIN; 3031 TCPSTAT_INC(tcps_rcvpack); 3032 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3033 SOCKBUF_LOCK(&so->so_rcv); 3034 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3035 m_freem(m); 3036 else 3037 sbappendstream_locked(&so->so_rcv, m, 0); 3038 /* NB: sorwakeup_locked() does an implicit unlock. */ 3039 sorwakeup_locked(so); 3040 } else { 3041 /* 3042 * XXX: Due to the header drop above "th" is 3043 * theoretically invalid by now. Fortunately 3044 * m_adj() doesn't actually frees any mbufs 3045 * when trimming from the head. 3046 */ 3047 thflags = tcp_reass(tp, th, &save_start, &tlen, m); 3048 tp->t_flags |= TF_ACKNOW; 3049 } 3050 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3051 tcp_update_sack_list(tp, save_start, save_start + tlen); 3052 #if 0 3053 /* 3054 * Note the amount of data that peer has sent into 3055 * our window, in order to estimate the sender's 3056 * buffer size. 3057 * XXX: Unused. 3058 */ 3059 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3060 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3061 else 3062 len = so->so_rcv.sb_hiwat; 3063 #endif 3064 } else { 3065 m_freem(m); 3066 thflags &= ~TH_FIN; 3067 } 3068 3069 /* 3070 * If FIN is received ACK the FIN and let the user know 3071 * that the connection is closing. 3072 */ 3073 if (thflags & TH_FIN) { 3074 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3075 socantrcvmore(so); 3076 /* 3077 * If connection is half-synchronized 3078 * (ie NEEDSYN flag on) then delay ACK, 3079 * so it may be piggybacked when SYN is sent. 3080 * Otherwise, since we received a FIN then no 3081 * more input can be expected, send ACK now. 3082 */ 3083 if (tp->t_flags & TF_NEEDSYN) 3084 tp->t_flags |= TF_DELACK; 3085 else 3086 tp->t_flags |= TF_ACKNOW; 3087 tp->rcv_nxt++; 3088 } 3089 switch (tp->t_state) { 3090 3091 /* 3092 * In SYN_RECEIVED and ESTABLISHED STATES 3093 * enter the CLOSE_WAIT state. 3094 */ 3095 case TCPS_SYN_RECEIVED: 3096 tp->t_starttime = ticks; 3097 /* FALLTHROUGH */ 3098 case TCPS_ESTABLISHED: 3099 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3100 break; 3101 3102 /* 3103 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3104 * enter the CLOSING state. 3105 */ 3106 case TCPS_FIN_WAIT_1: 3107 tcp_state_change(tp, TCPS_CLOSING); 3108 break; 3109 3110 /* 3111 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3112 * starting the time-wait timer, turning off the other 3113 * standard timers. 3114 */ 3115 case TCPS_FIN_WAIT_2: 3116 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3117 3118 tcp_twstart(tp); 3119 return; 3120 } 3121 } 3122 #ifdef TCPDEBUG 3123 if (so->so_options & SO_DEBUG) 3124 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3125 &tcp_savetcp, 0); 3126 #endif 3127 TCP_PROBE3(debug__input, tp, th, m); 3128 3129 /* 3130 * Return any desired output. 3131 */ 3132 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3133 (void) tp->t_fb->tfb_tcp_output(tp); 3134 3135 check_delack: 3136 INP_WLOCK_ASSERT(tp->t_inpcb); 3137 3138 if (tp->t_flags & TF_DELACK) { 3139 tp->t_flags &= ~TF_DELACK; 3140 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3141 } 3142 INP_WUNLOCK(tp->t_inpcb); 3143 return; 3144 3145 dropafterack: 3146 /* 3147 * Generate an ACK dropping incoming segment if it occupies 3148 * sequence space, where the ACK reflects our state. 3149 * 3150 * We can now skip the test for the RST flag since all 3151 * paths to this code happen after packets containing 3152 * RST have been dropped. 3153 * 3154 * In the SYN-RECEIVED state, don't send an ACK unless the 3155 * segment we received passes the SYN-RECEIVED ACK test. 3156 * If it fails send a RST. This breaks the loop in the 3157 * "LAND" DoS attack, and also prevents an ACK storm 3158 * between two listening ports that have been sent forged 3159 * SYN segments, each with the source address of the other. 3160 */ 3161 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3162 (SEQ_GT(tp->snd_una, th->th_ack) || 3163 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3164 rstreason = BANDLIM_RST_OPENPORT; 3165 goto dropwithreset; 3166 } 3167 #ifdef TCPDEBUG 3168 if (so->so_options & SO_DEBUG) 3169 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3170 &tcp_savetcp, 0); 3171 #endif 3172 TCP_PROBE3(debug__input, tp, th, m); 3173 tp->t_flags |= TF_ACKNOW; 3174 (void) tp->t_fb->tfb_tcp_output(tp); 3175 INP_WUNLOCK(tp->t_inpcb); 3176 m_freem(m); 3177 return; 3178 3179 dropwithreset: 3180 if (tp != NULL) { 3181 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3182 INP_WUNLOCK(tp->t_inpcb); 3183 } else 3184 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3185 return; 3186 3187 drop: 3188 /* 3189 * Drop space held by incoming segment and return. 3190 */ 3191 #ifdef TCPDEBUG 3192 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3193 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3194 &tcp_savetcp, 0); 3195 #endif 3196 TCP_PROBE3(debug__input, tp, th, m); 3197 if (tp != NULL) 3198 INP_WUNLOCK(tp->t_inpcb); 3199 m_freem(m); 3200 } 3201 3202 /* 3203 * Issue RST and make ACK acceptable to originator of segment. 3204 * The mbuf must still include the original packet header. 3205 * tp may be NULL. 3206 */ 3207 void 3208 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3209 int tlen, int rstreason) 3210 { 3211 #ifdef INET 3212 struct ip *ip; 3213 #endif 3214 #ifdef INET6 3215 struct ip6_hdr *ip6; 3216 #endif 3217 3218 if (tp != NULL) { 3219 INP_WLOCK_ASSERT(tp->t_inpcb); 3220 } 3221 3222 /* Don't bother if destination was broadcast/multicast. */ 3223 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3224 goto drop; 3225 #ifdef INET6 3226 if (mtod(m, struct ip *)->ip_v == 6) { 3227 ip6 = mtod(m, struct ip6_hdr *); 3228 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3229 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3230 goto drop; 3231 /* IPv6 anycast check is done at tcp6_input() */ 3232 } 3233 #endif 3234 #if defined(INET) && defined(INET6) 3235 else 3236 #endif 3237 #ifdef INET 3238 { 3239 ip = mtod(m, struct ip *); 3240 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3241 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3242 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3243 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3244 goto drop; 3245 } 3246 #endif 3247 3248 /* Perform bandwidth limiting. */ 3249 if (badport_bandlim(rstreason) < 0) 3250 goto drop; 3251 3252 /* tcp_respond consumes the mbuf chain. */ 3253 if (th->th_flags & TH_ACK) { 3254 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3255 th->th_ack, TH_RST); 3256 } else { 3257 if (th->th_flags & TH_SYN) 3258 tlen++; 3259 if (th->th_flags & TH_FIN) 3260 tlen++; 3261 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3262 (tcp_seq)0, TH_RST|TH_ACK); 3263 } 3264 return; 3265 drop: 3266 m_freem(m); 3267 } 3268 3269 /* 3270 * Parse TCP options and place in tcpopt. 3271 */ 3272 void 3273 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3274 { 3275 int opt, optlen; 3276 3277 to->to_flags = 0; 3278 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3279 opt = cp[0]; 3280 if (opt == TCPOPT_EOL) 3281 break; 3282 if (opt == TCPOPT_NOP) 3283 optlen = 1; 3284 else { 3285 if (cnt < 2) 3286 break; 3287 optlen = cp[1]; 3288 if (optlen < 2 || optlen > cnt) 3289 break; 3290 } 3291 switch (opt) { 3292 case TCPOPT_MAXSEG: 3293 if (optlen != TCPOLEN_MAXSEG) 3294 continue; 3295 if (!(flags & TO_SYN)) 3296 continue; 3297 to->to_flags |= TOF_MSS; 3298 bcopy((char *)cp + 2, 3299 (char *)&to->to_mss, sizeof(to->to_mss)); 3300 to->to_mss = ntohs(to->to_mss); 3301 break; 3302 case TCPOPT_WINDOW: 3303 if (optlen != TCPOLEN_WINDOW) 3304 continue; 3305 if (!(flags & TO_SYN)) 3306 continue; 3307 to->to_flags |= TOF_SCALE; 3308 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3309 break; 3310 case TCPOPT_TIMESTAMP: 3311 if (optlen != TCPOLEN_TIMESTAMP) 3312 continue; 3313 to->to_flags |= TOF_TS; 3314 bcopy((char *)cp + 2, 3315 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3316 to->to_tsval = ntohl(to->to_tsval); 3317 bcopy((char *)cp + 6, 3318 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3319 to->to_tsecr = ntohl(to->to_tsecr); 3320 break; 3321 case TCPOPT_SIGNATURE: 3322 /* 3323 * In order to reply to a host which has set the 3324 * TCP_SIGNATURE option in its initial SYN, we have 3325 * to record the fact that the option was observed 3326 * here for the syncache code to perform the correct 3327 * response. 3328 */ 3329 if (optlen != TCPOLEN_SIGNATURE) 3330 continue; 3331 to->to_flags |= TOF_SIGNATURE; 3332 to->to_signature = cp + 2; 3333 break; 3334 case TCPOPT_SACK_PERMITTED: 3335 if (optlen != TCPOLEN_SACK_PERMITTED) 3336 continue; 3337 if (!(flags & TO_SYN)) 3338 continue; 3339 if (!V_tcp_do_sack) 3340 continue; 3341 to->to_flags |= TOF_SACKPERM; 3342 break; 3343 case TCPOPT_SACK: 3344 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3345 continue; 3346 if (flags & TO_SYN) 3347 continue; 3348 to->to_flags |= TOF_SACK; 3349 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3350 to->to_sacks = cp + 2; 3351 TCPSTAT_INC(tcps_sack_rcv_blocks); 3352 break; 3353 case TCPOPT_FAST_OPEN: 3354 /* 3355 * Cookie length validation is performed by the 3356 * server side cookie checking code or the client 3357 * side cookie cache update code. 3358 */ 3359 if (!(flags & TO_SYN)) 3360 continue; 3361 if (!V_tcp_fastopen_client_enable && 3362 !V_tcp_fastopen_server_enable) 3363 continue; 3364 to->to_flags |= TOF_FASTOPEN; 3365 to->to_tfo_len = optlen - 2; 3366 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3367 break; 3368 default: 3369 continue; 3370 } 3371 } 3372 } 3373 3374 /* 3375 * Pull out of band byte out of a segment so 3376 * it doesn't appear in the user's data queue. 3377 * It is still reflected in the segment length for 3378 * sequencing purposes. 3379 */ 3380 void 3381 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3382 int off) 3383 { 3384 int cnt = off + th->th_urp - 1; 3385 3386 while (cnt >= 0) { 3387 if (m->m_len > cnt) { 3388 char *cp = mtod(m, caddr_t) + cnt; 3389 struct tcpcb *tp = sototcpcb(so); 3390 3391 INP_WLOCK_ASSERT(tp->t_inpcb); 3392 3393 tp->t_iobc = *cp; 3394 tp->t_oobflags |= TCPOOB_HAVEDATA; 3395 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3396 m->m_len--; 3397 if (m->m_flags & M_PKTHDR) 3398 m->m_pkthdr.len--; 3399 return; 3400 } 3401 cnt -= m->m_len; 3402 m = m->m_next; 3403 if (m == NULL) 3404 break; 3405 } 3406 panic("tcp_pulloutofband"); 3407 } 3408 3409 /* 3410 * Collect new round-trip time estimate 3411 * and update averages and current timeout. 3412 */ 3413 void 3414 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3415 { 3416 int delta; 3417 3418 INP_WLOCK_ASSERT(tp->t_inpcb); 3419 3420 TCPSTAT_INC(tcps_rttupdated); 3421 tp->t_rttupdated++; 3422 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3423 /* 3424 * srtt is stored as fixed point with 5 bits after the 3425 * binary point (i.e., scaled by 8). The following magic 3426 * is equivalent to the smoothing algorithm in rfc793 with 3427 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3428 * point). Adjust rtt to origin 0. 3429 */ 3430 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3431 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3432 3433 if ((tp->t_srtt += delta) <= 0) 3434 tp->t_srtt = 1; 3435 3436 /* 3437 * We accumulate a smoothed rtt variance (actually, a 3438 * smoothed mean difference), then set the retransmit 3439 * timer to smoothed rtt + 4 times the smoothed variance. 3440 * rttvar is stored as fixed point with 4 bits after the 3441 * binary point (scaled by 16). The following is 3442 * equivalent to rfc793 smoothing with an alpha of .75 3443 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3444 * rfc793's wired-in beta. 3445 */ 3446 if (delta < 0) 3447 delta = -delta; 3448 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3449 if ((tp->t_rttvar += delta) <= 0) 3450 tp->t_rttvar = 1; 3451 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3452 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3453 } else { 3454 /* 3455 * No rtt measurement yet - use the unsmoothed rtt. 3456 * Set the variance to half the rtt (so our first 3457 * retransmit happens at 3*rtt). 3458 */ 3459 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3460 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3461 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3462 } 3463 tp->t_rtttime = 0; 3464 tp->t_rxtshift = 0; 3465 3466 /* 3467 * the retransmit should happen at rtt + 4 * rttvar. 3468 * Because of the way we do the smoothing, srtt and rttvar 3469 * will each average +1/2 tick of bias. When we compute 3470 * the retransmit timer, we want 1/2 tick of rounding and 3471 * 1 extra tick because of +-1/2 tick uncertainty in the 3472 * firing of the timer. The bias will give us exactly the 3473 * 1.5 tick we need. But, because the bias is 3474 * statistical, we have to test that we don't drop below 3475 * the minimum feasible timer (which is 2 ticks). 3476 */ 3477 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3478 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3479 3480 /* 3481 * We received an ack for a packet that wasn't retransmitted; 3482 * it is probably safe to discard any error indications we've 3483 * received recently. This isn't quite right, but close enough 3484 * for now (a route might have failed after we sent a segment, 3485 * and the return path might not be symmetrical). 3486 */ 3487 tp->t_softerror = 0; 3488 } 3489 3490 /* 3491 * Determine a reasonable value for maxseg size. 3492 * If the route is known, check route for mtu. 3493 * If none, use an mss that can be handled on the outgoing interface 3494 * without forcing IP to fragment. If no route is found, route has no mtu, 3495 * or the destination isn't local, use a default, hopefully conservative 3496 * size (usually 512 or the default IP max size, but no more than the mtu 3497 * of the interface), as we can't discover anything about intervening 3498 * gateways or networks. We also initialize the congestion/slow start 3499 * window to be a single segment if the destination isn't local. 3500 * While looking at the routing entry, we also initialize other path-dependent 3501 * parameters from pre-set or cached values in the routing entry. 3502 * 3503 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3504 * IP options, e.g. IPSEC data, since length of this data may vary, and 3505 * thus it is calculated for every segment separately in tcp_output(). 3506 * 3507 * NOTE that this routine is only called when we process an incoming 3508 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3509 * settings are handled in tcp_mssopt(). 3510 */ 3511 void 3512 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3513 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3514 { 3515 int mss = 0; 3516 uint32_t maxmtu = 0; 3517 struct inpcb *inp = tp->t_inpcb; 3518 struct hc_metrics_lite metrics; 3519 #ifdef INET6 3520 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3521 size_t min_protoh = isipv6 ? 3522 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3523 sizeof (struct tcpiphdr); 3524 #else 3525 const size_t min_protoh = sizeof(struct tcpiphdr); 3526 #endif 3527 3528 INP_WLOCK_ASSERT(tp->t_inpcb); 3529 3530 if (mtuoffer != -1) { 3531 KASSERT(offer == -1, ("%s: conflict", __func__)); 3532 offer = mtuoffer - min_protoh; 3533 } 3534 3535 /* Initialize. */ 3536 #ifdef INET6 3537 if (isipv6) { 3538 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3539 tp->t_maxseg = V_tcp_v6mssdflt; 3540 } 3541 #endif 3542 #if defined(INET) && defined(INET6) 3543 else 3544 #endif 3545 #ifdef INET 3546 { 3547 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3548 tp->t_maxseg = V_tcp_mssdflt; 3549 } 3550 #endif 3551 3552 /* 3553 * No route to sender, stay with default mss and return. 3554 */ 3555 if (maxmtu == 0) { 3556 /* 3557 * In case we return early we need to initialize metrics 3558 * to a defined state as tcp_hc_get() would do for us 3559 * if there was no cache hit. 3560 */ 3561 if (metricptr != NULL) 3562 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3563 return; 3564 } 3565 3566 /* What have we got? */ 3567 switch (offer) { 3568 case 0: 3569 /* 3570 * Offer == 0 means that there was no MSS on the SYN 3571 * segment, in this case we use tcp_mssdflt as 3572 * already assigned to t_maxseg above. 3573 */ 3574 offer = tp->t_maxseg; 3575 break; 3576 3577 case -1: 3578 /* 3579 * Offer == -1 means that we didn't receive SYN yet. 3580 */ 3581 /* FALLTHROUGH */ 3582 3583 default: 3584 /* 3585 * Prevent DoS attack with too small MSS. Round up 3586 * to at least minmss. 3587 */ 3588 offer = max(offer, V_tcp_minmss); 3589 } 3590 3591 /* 3592 * rmx information is now retrieved from tcp_hostcache. 3593 */ 3594 tcp_hc_get(&inp->inp_inc, &metrics); 3595 if (metricptr != NULL) 3596 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3597 3598 /* 3599 * If there's a discovered mtu in tcp hostcache, use it. 3600 * Else, use the link mtu. 3601 */ 3602 if (metrics.rmx_mtu) 3603 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3604 else { 3605 #ifdef INET6 3606 if (isipv6) { 3607 mss = maxmtu - min_protoh; 3608 if (!V_path_mtu_discovery && 3609 !in6_localaddr(&inp->in6p_faddr)) 3610 mss = min(mss, V_tcp_v6mssdflt); 3611 } 3612 #endif 3613 #if defined(INET) && defined(INET6) 3614 else 3615 #endif 3616 #ifdef INET 3617 { 3618 mss = maxmtu - min_protoh; 3619 if (!V_path_mtu_discovery && 3620 !in_localaddr(inp->inp_faddr)) 3621 mss = min(mss, V_tcp_mssdflt); 3622 } 3623 #endif 3624 /* 3625 * XXX - The above conditional (mss = maxmtu - min_protoh) 3626 * probably violates the TCP spec. 3627 * The problem is that, since we don't know the 3628 * other end's MSS, we are supposed to use a conservative 3629 * default. But, if we do that, then MTU discovery will 3630 * never actually take place, because the conservative 3631 * default is much less than the MTUs typically seen 3632 * on the Internet today. For the moment, we'll sweep 3633 * this under the carpet. 3634 * 3635 * The conservative default might not actually be a problem 3636 * if the only case this occurs is when sending an initial 3637 * SYN with options and data to a host we've never talked 3638 * to before. Then, they will reply with an MSS value which 3639 * will get recorded and the new parameters should get 3640 * recomputed. For Further Study. 3641 */ 3642 } 3643 mss = min(mss, offer); 3644 3645 /* 3646 * Sanity check: make sure that maxseg will be large 3647 * enough to allow some data on segments even if the 3648 * all the option space is used (40bytes). Otherwise 3649 * funny things may happen in tcp_output. 3650 * 3651 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3652 */ 3653 mss = max(mss, 64); 3654 3655 tp->t_maxseg = mss; 3656 } 3657 3658 void 3659 tcp_mss(struct tcpcb *tp, int offer) 3660 { 3661 int mss; 3662 uint32_t bufsize; 3663 struct inpcb *inp; 3664 struct socket *so; 3665 struct hc_metrics_lite metrics; 3666 struct tcp_ifcap cap; 3667 3668 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3669 3670 bzero(&cap, sizeof(cap)); 3671 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3672 3673 mss = tp->t_maxseg; 3674 inp = tp->t_inpcb; 3675 3676 /* 3677 * If there's a pipesize, change the socket buffer to that size, 3678 * don't change if sb_hiwat is different than default (then it 3679 * has been changed on purpose with setsockopt). 3680 * Make the socket buffers an integral number of mss units; 3681 * if the mss is larger than the socket buffer, decrease the mss. 3682 */ 3683 so = inp->inp_socket; 3684 SOCKBUF_LOCK(&so->so_snd); 3685 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3686 bufsize = metrics.rmx_sendpipe; 3687 else 3688 bufsize = so->so_snd.sb_hiwat; 3689 if (bufsize < mss) 3690 mss = bufsize; 3691 else { 3692 bufsize = roundup(bufsize, mss); 3693 if (bufsize > sb_max) 3694 bufsize = sb_max; 3695 if (bufsize > so->so_snd.sb_hiwat) 3696 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3697 } 3698 SOCKBUF_UNLOCK(&so->so_snd); 3699 /* 3700 * Sanity check: make sure that maxseg will be large 3701 * enough to allow some data on segments even if the 3702 * all the option space is used (40bytes). Otherwise 3703 * funny things may happen in tcp_output. 3704 * 3705 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3706 */ 3707 tp->t_maxseg = max(mss, 64); 3708 3709 SOCKBUF_LOCK(&so->so_rcv); 3710 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3711 bufsize = metrics.rmx_recvpipe; 3712 else 3713 bufsize = so->so_rcv.sb_hiwat; 3714 if (bufsize > mss) { 3715 bufsize = roundup(bufsize, mss); 3716 if (bufsize > sb_max) 3717 bufsize = sb_max; 3718 if (bufsize > so->so_rcv.sb_hiwat) 3719 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3720 } 3721 SOCKBUF_UNLOCK(&so->so_rcv); 3722 3723 /* Check the interface for TSO capabilities. */ 3724 if (cap.ifcap & CSUM_TSO) { 3725 tp->t_flags |= TF_TSO; 3726 tp->t_tsomax = cap.tsomax; 3727 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3728 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3729 } 3730 } 3731 3732 /* 3733 * Determine the MSS option to send on an outgoing SYN. 3734 */ 3735 int 3736 tcp_mssopt(struct in_conninfo *inc) 3737 { 3738 int mss = 0; 3739 uint32_t thcmtu = 0; 3740 uint32_t maxmtu = 0; 3741 size_t min_protoh; 3742 3743 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3744 3745 #ifdef INET6 3746 if (inc->inc_flags & INC_ISIPV6) { 3747 mss = V_tcp_v6mssdflt; 3748 maxmtu = tcp_maxmtu6(inc, NULL); 3749 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3750 } 3751 #endif 3752 #if defined(INET) && defined(INET6) 3753 else 3754 #endif 3755 #ifdef INET 3756 { 3757 mss = V_tcp_mssdflt; 3758 maxmtu = tcp_maxmtu(inc, NULL); 3759 min_protoh = sizeof(struct tcpiphdr); 3760 } 3761 #endif 3762 #if defined(INET6) || defined(INET) 3763 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3764 #endif 3765 3766 if (maxmtu && thcmtu) 3767 mss = min(maxmtu, thcmtu) - min_protoh; 3768 else if (maxmtu || thcmtu) 3769 mss = max(maxmtu, thcmtu) - min_protoh; 3770 3771 return (mss); 3772 } 3773 3774 3775 /* 3776 * On a partial ack arrives, force the retransmission of the 3777 * next unacknowledged segment. Do not clear tp->t_dupacks. 3778 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3779 * be started again. 3780 */ 3781 void 3782 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3783 { 3784 tcp_seq onxt = tp->snd_nxt; 3785 uint32_t ocwnd = tp->snd_cwnd; 3786 u_int maxseg = tcp_maxseg(tp); 3787 3788 INP_WLOCK_ASSERT(tp->t_inpcb); 3789 3790 tcp_timer_activate(tp, TT_REXMT, 0); 3791 tp->t_rtttime = 0; 3792 tp->snd_nxt = th->th_ack; 3793 /* 3794 * Set snd_cwnd to one segment beyond acknowledged offset. 3795 * (tp->snd_una has not yet been updated when this function is called.) 3796 */ 3797 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3798 tp->t_flags |= TF_ACKNOW; 3799 (void) tp->t_fb->tfb_tcp_output(tp); 3800 tp->snd_cwnd = ocwnd; 3801 if (SEQ_GT(onxt, tp->snd_nxt)) 3802 tp->snd_nxt = onxt; 3803 /* 3804 * Partial window deflation. Relies on fact that tp->snd_una 3805 * not updated yet. 3806 */ 3807 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3808 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3809 else 3810 tp->snd_cwnd = 0; 3811 tp->snd_cwnd += maxseg; 3812 } 3813 3814 int 3815 tcp_compute_pipe(struct tcpcb *tp) 3816 { 3817 return (tp->snd_max - tp->snd_una + 3818 tp->sackhint.sack_bytes_rexmit - 3819 tp->sackhint.sacked_bytes); 3820 } 3821