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_max) = 2*1024*1024; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_autorcvbuf_max), 0, 218 "Max size of automatic receive buffer"); 219 220 VNET_DEFINE(struct inpcbhead, tcb); 221 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 222 VNET_DEFINE(struct inpcbinfo, tcbinfo); 223 224 /* 225 * TCP statistics are stored in an array of counter(9)s, which size matches 226 * size of struct tcpstat. TCP running connection count is a regular array. 227 */ 228 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 229 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 230 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 231 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 232 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 233 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 234 "TCP connection counts by TCP state"); 235 236 static void 237 tcp_vnet_init(const void *unused) 238 { 239 240 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 241 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 242 } 243 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 244 tcp_vnet_init, NULL); 245 246 #ifdef VIMAGE 247 static void 248 tcp_vnet_uninit(const void *unused) 249 { 250 251 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 252 VNET_PCPUSTAT_FREE(tcpstat); 253 } 254 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 255 tcp_vnet_uninit, NULL); 256 #endif /* VIMAGE */ 257 258 /* 259 * Kernel module interface for updating tcpstat. The argument is an index 260 * into tcpstat treated as an array. 261 */ 262 void 263 kmod_tcpstat_inc(int statnum) 264 { 265 266 counter_u64_add(VNET(tcpstat)[statnum], 1); 267 } 268 269 #ifdef TCP_HHOOK 270 /* 271 * Wrapper for the TCP established input helper hook. 272 */ 273 void 274 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 275 { 276 struct tcp_hhook_data hhook_data; 277 278 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 279 hhook_data.tp = tp; 280 hhook_data.th = th; 281 hhook_data.to = to; 282 283 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 284 tp->osd); 285 } 286 } 287 #endif 288 289 /* 290 * CC wrapper hook functions 291 */ 292 void 293 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 294 uint16_t type) 295 { 296 INP_WLOCK_ASSERT(tp->t_inpcb); 297 298 tp->ccv->nsegs = nsegs; 299 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 300 if (tp->snd_cwnd <= tp->snd_wnd) 301 tp->ccv->flags |= CCF_CWND_LIMITED; 302 else 303 tp->ccv->flags &= ~CCF_CWND_LIMITED; 304 305 if (type == CC_ACK) { 306 if (tp->snd_cwnd > tp->snd_ssthresh) { 307 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 308 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 309 if (tp->t_bytes_acked >= tp->snd_cwnd) { 310 tp->t_bytes_acked -= tp->snd_cwnd; 311 tp->ccv->flags |= CCF_ABC_SENTAWND; 312 } 313 } else { 314 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 315 tp->t_bytes_acked = 0; 316 } 317 } 318 319 if (CC_ALGO(tp)->ack_received != NULL) { 320 /* XXXLAS: Find a way to live without this */ 321 tp->ccv->curack = th->th_ack; 322 CC_ALGO(tp)->ack_received(tp->ccv, type); 323 } 324 } 325 326 void 327 cc_conn_init(struct tcpcb *tp) 328 { 329 struct hc_metrics_lite metrics; 330 struct inpcb *inp = tp->t_inpcb; 331 u_int maxseg; 332 int rtt; 333 334 INP_WLOCK_ASSERT(tp->t_inpcb); 335 336 tcp_hc_get(&inp->inp_inc, &metrics); 337 maxseg = tcp_maxseg(tp); 338 339 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 340 tp->t_srtt = rtt; 341 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 342 TCPSTAT_INC(tcps_usedrtt); 343 if (metrics.rmx_rttvar) { 344 tp->t_rttvar = metrics.rmx_rttvar; 345 TCPSTAT_INC(tcps_usedrttvar); 346 } else { 347 /* default variation is +- 1 rtt */ 348 tp->t_rttvar = 349 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 350 } 351 TCPT_RANGESET(tp->t_rxtcur, 352 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 353 tp->t_rttmin, TCPTV_REXMTMAX); 354 } 355 if (metrics.rmx_ssthresh) { 356 /* 357 * There's some sort of gateway or interface 358 * buffer limit on the path. Use this to set 359 * the slow start threshold, but set the 360 * threshold to no less than 2*mss. 361 */ 362 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 363 TCPSTAT_INC(tcps_usedssthresh); 364 } 365 366 /* 367 * Set the initial slow-start flight size. 368 * 369 * If a SYN or SYN/ACK was lost and retransmitted, we have to 370 * reduce the initial CWND to one segment as congestion is likely 371 * requiring us to be cautious. 372 */ 373 if (tp->snd_cwnd == 1) 374 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 375 else 376 tp->snd_cwnd = tcp_compute_initwnd(maxseg); 377 378 if (CC_ALGO(tp)->conn_init != NULL) 379 CC_ALGO(tp)->conn_init(tp->ccv); 380 } 381 382 void inline 383 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 384 { 385 u_int maxseg; 386 387 INP_WLOCK_ASSERT(tp->t_inpcb); 388 389 switch(type) { 390 case CC_NDUPACK: 391 if (!IN_FASTRECOVERY(tp->t_flags)) { 392 tp->snd_recover = tp->snd_max; 393 if (tp->t_flags & TF_ECN_PERMIT) 394 tp->t_flags |= TF_ECN_SND_CWR; 395 } 396 break; 397 case CC_ECN: 398 if (!IN_CONGRECOVERY(tp->t_flags)) { 399 TCPSTAT_INC(tcps_ecn_rcwnd); 400 tp->snd_recover = tp->snd_max; 401 if (tp->t_flags & TF_ECN_PERMIT) 402 tp->t_flags |= TF_ECN_SND_CWR; 403 } 404 break; 405 case CC_RTO: 406 maxseg = tcp_maxseg(tp); 407 tp->t_dupacks = 0; 408 tp->t_bytes_acked = 0; 409 EXIT_RECOVERY(tp->t_flags); 410 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 411 maxseg) * maxseg; 412 tp->snd_cwnd = maxseg; 413 break; 414 case CC_RTO_ERR: 415 TCPSTAT_INC(tcps_sndrexmitbad); 416 /* RTO was unnecessary, so reset everything. */ 417 tp->snd_cwnd = tp->snd_cwnd_prev; 418 tp->snd_ssthresh = tp->snd_ssthresh_prev; 419 tp->snd_recover = tp->snd_recover_prev; 420 if (tp->t_flags & TF_WASFRECOVERY) 421 ENTER_FASTRECOVERY(tp->t_flags); 422 if (tp->t_flags & TF_WASCRECOVERY) 423 ENTER_CONGRECOVERY(tp->t_flags); 424 tp->snd_nxt = tp->snd_max; 425 tp->t_flags &= ~TF_PREVVALID; 426 tp->t_badrxtwin = 0; 427 break; 428 } 429 430 if (CC_ALGO(tp)->cong_signal != NULL) { 431 if (th != NULL) 432 tp->ccv->curack = th->th_ack; 433 CC_ALGO(tp)->cong_signal(tp->ccv, type); 434 } 435 } 436 437 void inline 438 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 439 { 440 INP_WLOCK_ASSERT(tp->t_inpcb); 441 442 /* XXXLAS: KASSERT that we're in recovery? */ 443 444 if (CC_ALGO(tp)->post_recovery != NULL) { 445 tp->ccv->curack = th->th_ack; 446 CC_ALGO(tp)->post_recovery(tp->ccv); 447 } 448 /* XXXLAS: EXIT_RECOVERY ? */ 449 tp->t_bytes_acked = 0; 450 } 451 452 /* 453 * Indicate whether this ack should be delayed. We can delay the ack if 454 * following conditions are met: 455 * - There is no delayed ack timer in progress. 456 * - Our last ack wasn't a 0-sized window. We never want to delay 457 * the ack that opens up a 0-sized window. 458 * - LRO wasn't used for this segment. We make sure by checking that the 459 * segment size is not larger than the MSS. 460 */ 461 #define DELAY_ACK(tp, tlen) \ 462 ((!tcp_timer_active(tp, TT_DELACK) && \ 463 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 464 (tlen <= tp->t_maxseg) && \ 465 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 466 467 static void inline 468 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 469 { 470 INP_WLOCK_ASSERT(tp->t_inpcb); 471 472 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 473 switch (iptos & IPTOS_ECN_MASK) { 474 case IPTOS_ECN_CE: 475 tp->ccv->flags |= CCF_IPHDR_CE; 476 break; 477 case IPTOS_ECN_ECT0: 478 tp->ccv->flags &= ~CCF_IPHDR_CE; 479 break; 480 case IPTOS_ECN_ECT1: 481 tp->ccv->flags &= ~CCF_IPHDR_CE; 482 break; 483 } 484 485 if (th->th_flags & TH_CWR) 486 tp->ccv->flags |= CCF_TCPHDR_CWR; 487 else 488 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 489 490 if (tp->t_flags & TF_DELACK) 491 tp->ccv->flags |= CCF_DELACK; 492 else 493 tp->ccv->flags &= ~CCF_DELACK; 494 495 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 496 497 if (tp->ccv->flags & CCF_ACKNOW) 498 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 499 } 500 } 501 502 /* 503 * TCP input handling is split into multiple parts: 504 * tcp6_input is a thin wrapper around tcp_input for the extended 505 * ip6_protox[] call format in ip6_input 506 * tcp_input handles primary segment validation, inpcb lookup and 507 * SYN processing on listen sockets 508 * tcp_do_segment processes the ACK and text of the segment for 509 * establishing, established and closing connections 510 */ 511 #ifdef INET6 512 int 513 tcp6_input(struct mbuf **mp, int *offp, int proto) 514 { 515 struct mbuf *m = *mp; 516 struct in6_ifaddr *ia6; 517 struct ip6_hdr *ip6; 518 519 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 520 521 /* 522 * draft-itojun-ipv6-tcp-to-anycast 523 * better place to put this in? 524 */ 525 ip6 = mtod(m, struct ip6_hdr *); 526 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 527 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 528 struct ip6_hdr *ip6; 529 530 ifa_free(&ia6->ia_ifa); 531 ip6 = mtod(m, struct ip6_hdr *); 532 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 533 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 534 return (IPPROTO_DONE); 535 } 536 if (ia6) 537 ifa_free(&ia6->ia_ifa); 538 539 return (tcp_input(mp, offp, proto)); 540 } 541 #endif /* INET6 */ 542 543 int 544 tcp_input(struct mbuf **mp, int *offp, int proto) 545 { 546 struct mbuf *m = *mp; 547 struct tcphdr *th = NULL; 548 struct ip *ip = NULL; 549 struct inpcb *inp = NULL; 550 struct tcpcb *tp = NULL; 551 struct socket *so = NULL; 552 u_char *optp = NULL; 553 int off0; 554 int optlen = 0; 555 #ifdef INET 556 int len; 557 #endif 558 int tlen = 0, off; 559 int drop_hdrlen; 560 int thflags; 561 int rstreason = 0; /* For badport_bandlim accounting purposes */ 562 uint8_t iptos; 563 struct m_tag *fwd_tag = NULL; 564 struct epoch_tracker et; 565 #ifdef INET6 566 struct ip6_hdr *ip6 = NULL; 567 int isipv6; 568 #else 569 const void *ip6 = NULL; 570 #endif /* INET6 */ 571 struct tcpopt to; /* options in this segment */ 572 char *s = NULL; /* address and port logging */ 573 int ti_locked; 574 #ifdef TCPDEBUG 575 /* 576 * The size of tcp_saveipgen must be the size of the max ip header, 577 * now IPv6. 578 */ 579 u_char tcp_saveipgen[IP6_HDR_LEN]; 580 struct tcphdr tcp_savetcp; 581 short ostate = 0; 582 #endif 583 584 #ifdef INET6 585 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 586 #endif 587 588 off0 = *offp; 589 m = *mp; 590 *mp = NULL; 591 to.to_flags = 0; 592 TCPSTAT_INC(tcps_rcvtotal); 593 594 #ifdef INET6 595 if (isipv6) { 596 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 597 598 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 599 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 600 if (m == NULL) { 601 TCPSTAT_INC(tcps_rcvshort); 602 return (IPPROTO_DONE); 603 } 604 } 605 606 ip6 = mtod(m, struct ip6_hdr *); 607 th = (struct tcphdr *)((caddr_t)ip6 + off0); 608 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 609 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 610 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 611 th->th_sum = m->m_pkthdr.csum_data; 612 else 613 th->th_sum = in6_cksum_pseudo(ip6, tlen, 614 IPPROTO_TCP, m->m_pkthdr.csum_data); 615 th->th_sum ^= 0xffff; 616 } else 617 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 618 if (th->th_sum) { 619 TCPSTAT_INC(tcps_rcvbadsum); 620 goto drop; 621 } 622 623 /* 624 * Be proactive about unspecified IPv6 address in source. 625 * As we use all-zero to indicate unbounded/unconnected pcb, 626 * unspecified IPv6 address can be used to confuse us. 627 * 628 * Note that packets with unspecified IPv6 destination is 629 * already dropped in ip6_input. 630 */ 631 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 632 /* XXX stat */ 633 goto drop; 634 } 635 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 636 } 637 #endif 638 #if defined(INET) && defined(INET6) 639 else 640 #endif 641 #ifdef INET 642 { 643 /* 644 * Get IP and TCP header together in first mbuf. 645 * Note: IP leaves IP header in first mbuf. 646 */ 647 if (off0 > sizeof (struct ip)) { 648 ip_stripoptions(m); 649 off0 = sizeof(struct ip); 650 } 651 if (m->m_len < sizeof (struct tcpiphdr)) { 652 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 653 == NULL) { 654 TCPSTAT_INC(tcps_rcvshort); 655 return (IPPROTO_DONE); 656 } 657 } 658 ip = mtod(m, struct ip *); 659 th = (struct tcphdr *)((caddr_t)ip + off0); 660 tlen = ntohs(ip->ip_len) - off0; 661 662 iptos = ip->ip_tos; 663 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 664 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 665 th->th_sum = m->m_pkthdr.csum_data; 666 else 667 th->th_sum = in_pseudo(ip->ip_src.s_addr, 668 ip->ip_dst.s_addr, 669 htonl(m->m_pkthdr.csum_data + tlen + 670 IPPROTO_TCP)); 671 th->th_sum ^= 0xffff; 672 } else { 673 struct ipovly *ipov = (struct ipovly *)ip; 674 675 /* 676 * Checksum extended TCP header and data. 677 */ 678 len = off0 + tlen; 679 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 680 ipov->ih_len = htons(tlen); 681 th->th_sum = in_cksum(m, len); 682 /* Reset length for SDT probes. */ 683 ip->ip_len = htons(len); 684 /* Reset TOS bits */ 685 ip->ip_tos = iptos; 686 /* Re-initialization for later version check */ 687 ip->ip_v = IPVERSION; 688 ip->ip_hl = off0 >> 2; 689 } 690 691 if (th->th_sum) { 692 TCPSTAT_INC(tcps_rcvbadsum); 693 goto drop; 694 } 695 } 696 #endif /* INET */ 697 698 /* 699 * Check that TCP offset makes sense, 700 * pull out TCP options and adjust length. XXX 701 */ 702 off = th->th_off << 2; 703 if (off < sizeof (struct tcphdr) || off > tlen) { 704 TCPSTAT_INC(tcps_rcvbadoff); 705 goto drop; 706 } 707 tlen -= off; /* tlen is used instead of ti->ti_len */ 708 if (off > sizeof (struct tcphdr)) { 709 #ifdef INET6 710 if (isipv6) { 711 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 712 ip6 = mtod(m, struct ip6_hdr *); 713 th = (struct tcphdr *)((caddr_t)ip6 + off0); 714 } 715 #endif 716 #if defined(INET) && defined(INET6) 717 else 718 #endif 719 #ifdef INET 720 { 721 if (m->m_len < sizeof(struct ip) + off) { 722 if ((m = m_pullup(m, sizeof (struct ip) + off)) 723 == NULL) { 724 TCPSTAT_INC(tcps_rcvshort); 725 return (IPPROTO_DONE); 726 } 727 ip = mtod(m, struct ip *); 728 th = (struct tcphdr *)((caddr_t)ip + off0); 729 } 730 } 731 #endif 732 optlen = off - sizeof (struct tcphdr); 733 optp = (u_char *)(th + 1); 734 } 735 thflags = th->th_flags; 736 737 /* 738 * Convert TCP protocol specific fields to host format. 739 */ 740 tcp_fields_to_host(th); 741 742 /* 743 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 744 */ 745 drop_hdrlen = off0 + off; 746 747 /* 748 * Locate pcb for segment; if we're likely to add or remove a 749 * connection then first acquire pcbinfo lock. There are three cases 750 * where we might discover later we need a write lock despite the 751 * flags: ACKs moving a connection out of the syncache, ACKs for a 752 * connection in TIMEWAIT and SYNs not targeting a listening socket. 753 */ 754 if ((thflags & (TH_FIN | TH_RST)) != 0) { 755 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 756 ti_locked = TI_RLOCKED; 757 } else 758 ti_locked = TI_UNLOCKED; 759 760 /* 761 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 762 */ 763 if ( 764 #ifdef INET6 765 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 766 #ifdef INET 767 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 768 #endif 769 #endif 770 #if defined(INET) && !defined(INET6) 771 (m->m_flags & M_IP_NEXTHOP) 772 #endif 773 ) 774 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 775 776 findpcb: 777 #ifdef INVARIANTS 778 if (ti_locked == TI_RLOCKED) { 779 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 780 } else { 781 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 782 } 783 #endif 784 #ifdef INET6 785 if (isipv6 && fwd_tag != NULL) { 786 struct sockaddr_in6 *next_hop6; 787 788 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 789 /* 790 * Transparently forwarded. Pretend to be the destination. 791 * Already got one like this? 792 */ 793 inp = in6_pcblookup_mbuf(&V_tcbinfo, 794 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 795 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 796 if (!inp) { 797 /* 798 * It's new. Try to find the ambushing socket. 799 * Because we've rewritten the destination address, 800 * any hardware-generated hash is ignored. 801 */ 802 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 803 th->th_sport, &next_hop6->sin6_addr, 804 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 805 th->th_dport, INPLOOKUP_WILDCARD | 806 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 807 } 808 } else if (isipv6) { 809 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 810 th->th_sport, &ip6->ip6_dst, th->th_dport, 811 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 812 m->m_pkthdr.rcvif, m); 813 } 814 #endif /* INET6 */ 815 #if defined(INET6) && defined(INET) 816 else 817 #endif 818 #ifdef INET 819 if (fwd_tag != NULL) { 820 struct sockaddr_in *next_hop; 821 822 next_hop = (struct sockaddr_in *)(fwd_tag+1); 823 /* 824 * Transparently forwarded. Pretend to be the destination. 825 * already got one like this? 826 */ 827 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 828 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 829 m->m_pkthdr.rcvif, m); 830 if (!inp) { 831 /* 832 * It's new. Try to find the ambushing socket. 833 * Because we've rewritten the destination address, 834 * any hardware-generated hash is ignored. 835 */ 836 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 837 th->th_sport, next_hop->sin_addr, 838 next_hop->sin_port ? ntohs(next_hop->sin_port) : 839 th->th_dport, INPLOOKUP_WILDCARD | 840 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 841 } 842 } else 843 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 844 th->th_sport, ip->ip_dst, th->th_dport, 845 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 846 m->m_pkthdr.rcvif, m); 847 #endif /* INET */ 848 849 /* 850 * If the INPCB does not exist then all data in the incoming 851 * segment is discarded and an appropriate RST is sent back. 852 * XXX MRT Send RST using which routing table? 853 */ 854 if (inp == NULL) { 855 /* 856 * Log communication attempts to ports that are not 857 * in use. 858 */ 859 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 860 tcp_log_in_vain == 2) { 861 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 862 log(LOG_INFO, "%s; %s: Connection attempt " 863 "to closed port\n", s, __func__); 864 } 865 /* 866 * When blackholing do not respond with a RST but 867 * completely ignore the segment and drop it. 868 */ 869 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 870 V_blackhole == 2) 871 goto dropunlock; 872 873 rstreason = BANDLIM_RST_CLOSEDPORT; 874 goto dropwithreset; 875 } 876 INP_WLOCK_ASSERT(inp); 877 /* 878 * While waiting for inp lock during the lookup, another thread 879 * can have dropped the inpcb, in which case we need to loop back 880 * and try to find a new inpcb to deliver to. 881 */ 882 if (inp->inp_flags & INP_DROPPED) { 883 INP_WUNLOCK(inp); 884 inp = NULL; 885 goto findpcb; 886 } 887 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 888 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 889 ((inp->inp_socket == NULL) || 890 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 891 inp->inp_flowid = m->m_pkthdr.flowid; 892 inp->inp_flowtype = M_HASHTYPE_GET(m); 893 } 894 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 895 #ifdef INET6 896 if (isipv6 && IPSEC_ENABLED(ipv6) && 897 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 898 goto dropunlock; 899 } 900 #ifdef INET 901 else 902 #endif 903 #endif /* INET6 */ 904 #ifdef INET 905 if (IPSEC_ENABLED(ipv4) && 906 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 907 goto dropunlock; 908 } 909 #endif /* INET */ 910 #endif /* IPSEC */ 911 912 /* 913 * Check the minimum TTL for socket. 914 */ 915 if (inp->inp_ip_minttl != 0) { 916 #ifdef INET6 917 if (isipv6) { 918 if (inp->inp_ip_minttl > ip6->ip6_hlim) 919 goto dropunlock; 920 } else 921 #endif 922 if (inp->inp_ip_minttl > ip->ip_ttl) 923 goto dropunlock; 924 } 925 926 /* 927 * A previous connection in TIMEWAIT state is supposed to catch stray 928 * or duplicate segments arriving late. If this segment was a 929 * legitimate new connection attempt, the old INPCB gets removed and 930 * we can try again to find a listening socket. 931 * 932 * At this point, due to earlier optimism, we may hold only an inpcb 933 * lock, and not the inpcbinfo write lock. If so, we need to try to 934 * acquire it, or if that fails, acquire a reference on the inpcb, 935 * drop all locks, acquire a global write lock, and then re-acquire 936 * the inpcb lock. We may at that point discover that another thread 937 * has tried to free the inpcb, in which case we need to loop back 938 * and try to find a new inpcb to deliver to. 939 * 940 * XXXRW: It may be time to rethink timewait locking. 941 */ 942 if (inp->inp_flags & INP_TIMEWAIT) { 943 if (ti_locked == TI_UNLOCKED) { 944 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 945 ti_locked = TI_RLOCKED; 946 } 947 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 948 949 if (thflags & TH_SYN) 950 tcp_dooptions(&to, optp, optlen, TO_SYN); 951 /* 952 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 953 */ 954 if (tcp_twcheck(inp, &to, th, m, tlen)) 955 goto findpcb; 956 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 957 return (IPPROTO_DONE); 958 } 959 /* 960 * The TCPCB may no longer exist if the connection is winding 961 * down or it is in the CLOSED state. Either way we drop the 962 * segment and send an appropriate response. 963 */ 964 tp = intotcpcb(inp); 965 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 966 rstreason = BANDLIM_RST_CLOSEDPORT; 967 goto dropwithreset; 968 } 969 970 #ifdef TCP_OFFLOAD 971 if (tp->t_flags & TF_TOE) { 972 tcp_offload_input(tp, m); 973 m = NULL; /* consumed by the TOE driver */ 974 goto dropunlock; 975 } 976 #endif 977 978 /* 979 * We've identified a valid inpcb, but it could be that we need an 980 * inpcbinfo write lock but don't hold it. In this case, attempt to 981 * acquire using the same strategy as the TIMEWAIT case above. If we 982 * relock, we have to jump back to 'relocked' as the connection might 983 * now be in TIMEWAIT. 984 */ 985 #ifdef INVARIANTS 986 if ((thflags & (TH_FIN | TH_RST)) != 0) 987 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 988 #endif 989 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 990 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 991 !IS_FASTOPEN(tp->t_flags)))) { 992 if (ti_locked == TI_UNLOCKED) { 993 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 994 ti_locked = TI_RLOCKED; 995 } 996 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 997 } 998 999 #ifdef MAC 1000 INP_WLOCK_ASSERT(inp); 1001 if (mac_inpcb_check_deliver(inp, m)) 1002 goto dropunlock; 1003 #endif 1004 so = inp->inp_socket; 1005 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1006 #ifdef TCPDEBUG 1007 if (so->so_options & SO_DEBUG) { 1008 ostate = tp->t_state; 1009 #ifdef INET6 1010 if (isipv6) { 1011 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1012 } else 1013 #endif 1014 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1015 tcp_savetcp = *th; 1016 } 1017 #endif /* TCPDEBUG */ 1018 /* 1019 * When the socket is accepting connections (the INPCB is in LISTEN 1020 * state) we look into the SYN cache if this is a new connection 1021 * attempt or the completion of a previous one. 1022 */ 1023 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1024 ("%s: so accepting but tp %p not listening", __func__, tp)); 1025 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1026 struct in_conninfo inc; 1027 1028 bzero(&inc, sizeof(inc)); 1029 #ifdef INET6 1030 if (isipv6) { 1031 inc.inc_flags |= INC_ISIPV6; 1032 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1033 inc.inc_flags |= INC_IPV6MINMTU; 1034 inc.inc6_faddr = ip6->ip6_src; 1035 inc.inc6_laddr = ip6->ip6_dst; 1036 } else 1037 #endif 1038 { 1039 inc.inc_faddr = ip->ip_src; 1040 inc.inc_laddr = ip->ip_dst; 1041 } 1042 inc.inc_fport = th->th_sport; 1043 inc.inc_lport = th->th_dport; 1044 inc.inc_fibnum = so->so_fibnum; 1045 1046 /* 1047 * Check for an existing connection attempt in syncache if 1048 * the flag is only ACK. A successful lookup creates a new 1049 * socket appended to the listen queue in SYN_RECEIVED state. 1050 */ 1051 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1052 1053 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1054 /* 1055 * Parse the TCP options here because 1056 * syncookies need access to the reflected 1057 * timestamp. 1058 */ 1059 tcp_dooptions(&to, optp, optlen, 0); 1060 /* 1061 * NB: syncache_expand() doesn't unlock 1062 * inp and tcpinfo locks. 1063 */ 1064 rstreason = syncache_expand(&inc, &to, th, &so, m); 1065 if (rstreason < 0) { 1066 /* 1067 * A failing TCP MD5 signature comparison 1068 * must result in the segment being dropped 1069 * and must not produce any response back 1070 * to the sender. 1071 */ 1072 goto dropunlock; 1073 } else if (rstreason == 0) { 1074 /* 1075 * No syncache entry or ACK was not 1076 * for our SYN/ACK. Send a RST. 1077 * NB: syncache did its own logging 1078 * of the failure cause. 1079 */ 1080 rstreason = BANDLIM_RST_OPENPORT; 1081 goto dropwithreset; 1082 } 1083 tfo_socket_result: 1084 if (so == NULL) { 1085 /* 1086 * We completed the 3-way handshake 1087 * but could not allocate a socket 1088 * either due to memory shortage, 1089 * listen queue length limits or 1090 * global socket limits. Send RST 1091 * or wait and have the remote end 1092 * retransmit the ACK for another 1093 * try. 1094 */ 1095 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1096 log(LOG_DEBUG, "%s; %s: Listen socket: " 1097 "Socket allocation failed due to " 1098 "limits or memory shortage, %s\n", 1099 s, __func__, 1100 V_tcp_sc_rst_sock_fail ? 1101 "sending RST" : "try again"); 1102 if (V_tcp_sc_rst_sock_fail) { 1103 rstreason = BANDLIM_UNLIMITED; 1104 goto dropwithreset; 1105 } else 1106 goto dropunlock; 1107 } 1108 /* 1109 * Socket is created in state SYN_RECEIVED. 1110 * Unlock the listen socket, lock the newly 1111 * created socket and update the tp variable. 1112 */ 1113 INP_WUNLOCK(inp); /* listen socket */ 1114 inp = sotoinpcb(so); 1115 /* 1116 * New connection inpcb is already locked by 1117 * syncache_expand(). 1118 */ 1119 INP_WLOCK_ASSERT(inp); 1120 tp = intotcpcb(inp); 1121 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1122 ("%s: ", __func__)); 1123 /* 1124 * Process the segment and the data it 1125 * contains. tcp_do_segment() consumes 1126 * the mbuf chain and unlocks the inpcb. 1127 */ 1128 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1129 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1130 iptos); 1131 if (ti_locked == TI_RLOCKED) 1132 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1133 return (IPPROTO_DONE); 1134 } 1135 /* 1136 * Segment flag validation for new connection attempts: 1137 * 1138 * Our (SYN|ACK) response was rejected. 1139 * Check with syncache and remove entry to prevent 1140 * retransmits. 1141 * 1142 * NB: syncache_chkrst does its own logging of failure 1143 * causes. 1144 */ 1145 if (thflags & TH_RST) { 1146 syncache_chkrst(&inc, th, m); 1147 goto dropunlock; 1148 } 1149 /* 1150 * We can't do anything without SYN. 1151 */ 1152 if ((thflags & TH_SYN) == 0) { 1153 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1154 log(LOG_DEBUG, "%s; %s: Listen socket: " 1155 "SYN is missing, segment ignored\n", 1156 s, __func__); 1157 TCPSTAT_INC(tcps_badsyn); 1158 goto dropunlock; 1159 } 1160 /* 1161 * (SYN|ACK) is bogus on a listen socket. 1162 */ 1163 if (thflags & TH_ACK) { 1164 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1165 log(LOG_DEBUG, "%s; %s: Listen socket: " 1166 "SYN|ACK invalid, segment rejected\n", 1167 s, __func__); 1168 syncache_badack(&inc); /* XXX: Not needed! */ 1169 TCPSTAT_INC(tcps_badsyn); 1170 rstreason = BANDLIM_RST_OPENPORT; 1171 goto dropwithreset; 1172 } 1173 /* 1174 * If the drop_synfin option is enabled, drop all 1175 * segments with both the SYN and FIN bits set. 1176 * This prevents e.g. nmap from identifying the 1177 * TCP/IP stack. 1178 * XXX: Poor reasoning. nmap has other methods 1179 * and is constantly refining its stack detection 1180 * strategies. 1181 * XXX: This is a violation of the TCP specification 1182 * and was used by RFC1644. 1183 */ 1184 if ((thflags & TH_FIN) && V_drop_synfin) { 1185 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1186 log(LOG_DEBUG, "%s; %s: Listen socket: " 1187 "SYN|FIN segment ignored (based on " 1188 "sysctl setting)\n", s, __func__); 1189 TCPSTAT_INC(tcps_badsyn); 1190 goto dropunlock; 1191 } 1192 /* 1193 * Segment's flags are (SYN) or (SYN|FIN). 1194 * 1195 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1196 * as they do not affect the state of the TCP FSM. 1197 * The data pointed to by TH_URG and th_urp is ignored. 1198 */ 1199 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1200 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1201 KASSERT(thflags & (TH_SYN), 1202 ("%s: Listen socket: TH_SYN not set", __func__)); 1203 #ifdef INET6 1204 /* 1205 * If deprecated address is forbidden, 1206 * we do not accept SYN to deprecated interface 1207 * address to prevent any new inbound connection from 1208 * getting established. 1209 * When we do not accept SYN, we send a TCP RST, 1210 * with deprecated source address (instead of dropping 1211 * it). We compromise it as it is much better for peer 1212 * to send a RST, and RST will be the final packet 1213 * for the exchange. 1214 * 1215 * If we do not forbid deprecated addresses, we accept 1216 * the SYN packet. RFC2462 does not suggest dropping 1217 * SYN in this case. 1218 * If we decipher RFC2462 5.5.4, it says like this: 1219 * 1. use of deprecated addr with existing 1220 * communication is okay - "SHOULD continue to be 1221 * used" 1222 * 2. use of it with new communication: 1223 * (2a) "SHOULD NOT be used if alternate address 1224 * with sufficient scope is available" 1225 * (2b) nothing mentioned otherwise. 1226 * Here we fall into (2b) case as we have no choice in 1227 * our source address selection - we must obey the peer. 1228 * 1229 * The wording in RFC2462 is confusing, and there are 1230 * multiple description text for deprecated address 1231 * handling - worse, they are not exactly the same. 1232 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1233 */ 1234 if (isipv6 && !V_ip6_use_deprecated) { 1235 struct in6_ifaddr *ia6; 1236 1237 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1238 if (ia6 != NULL && 1239 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1240 ifa_free(&ia6->ia_ifa); 1241 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1242 log(LOG_DEBUG, "%s; %s: Listen socket: " 1243 "Connection attempt to deprecated " 1244 "IPv6 address rejected\n", 1245 s, __func__); 1246 rstreason = BANDLIM_RST_OPENPORT; 1247 goto dropwithreset; 1248 } 1249 if (ia6) 1250 ifa_free(&ia6->ia_ifa); 1251 } 1252 #endif /* INET6 */ 1253 /* 1254 * Basic sanity checks on incoming SYN requests: 1255 * Don't respond if the destination is a link layer 1256 * broadcast according to RFC1122 4.2.3.10, p. 104. 1257 * If it is from this socket it must be forged. 1258 * Don't respond if the source or destination is a 1259 * global or subnet broad- or multicast address. 1260 * Note that it is quite possible to receive unicast 1261 * link-layer packets with a broadcast IP address. Use 1262 * in_broadcast() to find them. 1263 */ 1264 if (m->m_flags & (M_BCAST|M_MCAST)) { 1265 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1266 log(LOG_DEBUG, "%s; %s: Listen socket: " 1267 "Connection attempt from broad- or multicast " 1268 "link layer address ignored\n", s, __func__); 1269 goto dropunlock; 1270 } 1271 #ifdef INET6 1272 if (isipv6) { 1273 if (th->th_dport == th->th_sport && 1274 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1275 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1276 log(LOG_DEBUG, "%s; %s: Listen socket: " 1277 "Connection attempt to/from self " 1278 "ignored\n", s, __func__); 1279 goto dropunlock; 1280 } 1281 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1282 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1283 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1284 log(LOG_DEBUG, "%s; %s: Listen socket: " 1285 "Connection attempt from/to multicast " 1286 "address ignored\n", s, __func__); 1287 goto dropunlock; 1288 } 1289 } 1290 #endif 1291 #if defined(INET) && defined(INET6) 1292 else 1293 #endif 1294 #ifdef INET 1295 { 1296 if (th->th_dport == th->th_sport && 1297 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1299 log(LOG_DEBUG, "%s; %s: Listen socket: " 1300 "Connection attempt from/to self " 1301 "ignored\n", s, __func__); 1302 goto dropunlock; 1303 } 1304 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1305 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1306 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1307 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1309 log(LOG_DEBUG, "%s; %s: Listen socket: " 1310 "Connection attempt from/to broad- " 1311 "or multicast address ignored\n", 1312 s, __func__); 1313 goto dropunlock; 1314 } 1315 } 1316 #endif 1317 /* 1318 * SYN appears to be valid. Create compressed TCP state 1319 * for syncache. 1320 */ 1321 #ifdef TCPDEBUG 1322 if (so->so_options & SO_DEBUG) 1323 tcp_trace(TA_INPUT, ostate, tp, 1324 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1325 #endif 1326 TCP_PROBE3(debug__input, tp, th, m); 1327 tcp_dooptions(&to, optp, optlen, TO_SYN); 1328 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1329 goto tfo_socket_result; 1330 1331 /* 1332 * Entry added to syncache and mbuf consumed. 1333 * Only the listen socket is unlocked by syncache_add(). 1334 */ 1335 if (ti_locked == TI_RLOCKED) { 1336 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1337 ti_locked = TI_UNLOCKED; 1338 } 1339 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1340 return (IPPROTO_DONE); 1341 } else if (tp->t_state == TCPS_LISTEN) { 1342 /* 1343 * When a listen socket is torn down the SO_ACCEPTCONN 1344 * flag is removed first while connections are drained 1345 * from the accept queue in a unlock/lock cycle of the 1346 * ACCEPT_LOCK, opening a race condition allowing a SYN 1347 * attempt go through unhandled. 1348 */ 1349 goto dropunlock; 1350 } 1351 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1352 if (tp->t_flags & TF_SIGNATURE) { 1353 tcp_dooptions(&to, optp, optlen, thflags); 1354 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1355 TCPSTAT_INC(tcps_sig_err_nosigopt); 1356 goto dropunlock; 1357 } 1358 if (!TCPMD5_ENABLED() || 1359 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1360 goto dropunlock; 1361 } 1362 #endif 1363 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1364 1365 /* 1366 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1367 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1368 * the inpcb, and unlocks pcbinfo. 1369 */ 1370 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1371 if (ti_locked == TI_RLOCKED) 1372 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1373 return (IPPROTO_DONE); 1374 1375 dropwithreset: 1376 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1377 1378 if (ti_locked == TI_RLOCKED) { 1379 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1380 ti_locked = TI_UNLOCKED; 1381 } 1382 #ifdef INVARIANTS 1383 else { 1384 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1385 "ti_locked: %d", __func__, ti_locked)); 1386 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1387 } 1388 #endif 1389 1390 if (inp != NULL) { 1391 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1392 INP_WUNLOCK(inp); 1393 } else 1394 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1395 m = NULL; /* mbuf chain got consumed. */ 1396 goto drop; 1397 1398 dropunlock: 1399 if (m != NULL) 1400 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1401 1402 if (ti_locked == TI_RLOCKED) { 1403 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1404 ti_locked = TI_UNLOCKED; 1405 } 1406 #ifdef INVARIANTS 1407 else { 1408 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1409 "ti_locked: %d", __func__, ti_locked)); 1410 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1411 } 1412 #endif 1413 1414 if (inp != NULL) 1415 INP_WUNLOCK(inp); 1416 1417 drop: 1418 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1419 if (s != NULL) 1420 free(s, M_TCPLOG); 1421 if (m != NULL) 1422 m_freem(m); 1423 return (IPPROTO_DONE); 1424 } 1425 1426 /* 1427 * Automatic sizing of receive socket buffer. Often the send 1428 * buffer size is not optimally adjusted to the actual network 1429 * conditions at hand (delay bandwidth product). Setting the 1430 * buffer size too small limits throughput on links with high 1431 * bandwidth and high delay (eg. trans-continental/oceanic links). 1432 * 1433 * On the receive side the socket buffer memory is only rarely 1434 * used to any significant extent. This allows us to be much 1435 * more aggressive in scaling the receive socket buffer. For 1436 * the case that the buffer space is actually used to a large 1437 * extent and we run out of kernel memory we can simply drop 1438 * the new segments; TCP on the sender will just retransmit it 1439 * later. Setting the buffer size too big may only consume too 1440 * much kernel memory if the application doesn't read() from 1441 * the socket or packet loss or reordering makes use of the 1442 * reassembly queue. 1443 * 1444 * The criteria to step up the receive buffer one notch are: 1445 * 1. Application has not set receive buffer size with 1446 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1447 * 2. the number of bytes received during 1/2 of an sRTT 1448 * is at least 3/8 of the current socket buffer size. 1449 * 3. receive buffer size has not hit maximal automatic size; 1450 * 1451 * If all of the criteria are met we increaset the socket buffer 1452 * by a 1/2 (bounded by the max). This allows us to keep ahead 1453 * of slow-start but also makes it so our peer never gets limited 1454 * by our rwnd which we then open up causing a burst. 1455 * 1456 * This algorithm does two steps per RTT at most and only if 1457 * we receive a bulk stream w/o packet losses or reorderings. 1458 * Shrinking the buffer during idle times is not necessary as 1459 * it doesn't consume any memory when idle. 1460 * 1461 * TODO: Only step up if the application is actually serving 1462 * the buffer to better manage the socket buffer resources. 1463 */ 1464 int 1465 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1466 struct tcpcb *tp, int tlen) 1467 { 1468 int newsize = 0; 1469 1470 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1471 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1472 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1473 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1474 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1475 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1476 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1477 } 1478 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1479 1480 /* Start over with next RTT. */ 1481 tp->rfbuf_ts = 0; 1482 tp->rfbuf_cnt = 0; 1483 } else { 1484 tp->rfbuf_cnt += tlen; /* add up */ 1485 } 1486 1487 return (newsize); 1488 } 1489 1490 void 1491 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1492 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1493 { 1494 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1495 int rstreason, todrop, win; 1496 uint32_t tiwin; 1497 uint16_t nsegs; 1498 char *s; 1499 struct in_conninfo *inc; 1500 struct mbuf *mfree; 1501 struct tcpopt to; 1502 int tfo_syn; 1503 1504 #ifdef TCPDEBUG 1505 /* 1506 * The size of tcp_saveipgen must be the size of the max ip header, 1507 * now IPv6. 1508 */ 1509 u_char tcp_saveipgen[IP6_HDR_LEN]; 1510 struct tcphdr tcp_savetcp; 1511 short ostate = 0; 1512 #endif 1513 thflags = th->th_flags; 1514 inc = &tp->t_inpcb->inp_inc; 1515 tp->sackhint.last_sack_ack = 0; 1516 sack_changed = 0; 1517 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1518 /* 1519 * If this is either a state-changing packet or current state isn't 1520 * established, we require a write lock on tcbinfo. Otherwise, we 1521 * allow the tcbinfo to be in either alocked or unlocked, as the 1522 * caller may have unnecessarily acquired a write lock due to a race. 1523 */ 1524 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1525 tp->t_state != TCPS_ESTABLISHED) { 1526 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1527 } 1528 INP_WLOCK_ASSERT(tp->t_inpcb); 1529 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1530 __func__)); 1531 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1532 __func__)); 1533 1534 #ifdef TCPPCAP 1535 /* Save segment, if requested. */ 1536 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1537 #endif 1538 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1539 tlen, NULL, true); 1540 1541 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1542 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1543 log(LOG_DEBUG, "%s; %s: " 1544 "SYN|FIN segment ignored (based on " 1545 "sysctl setting)\n", s, __func__); 1546 free(s, M_TCPLOG); 1547 } 1548 goto drop; 1549 } 1550 1551 /* 1552 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1553 * check SEQ.ACK first. 1554 */ 1555 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1556 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1557 rstreason = BANDLIM_UNLIMITED; 1558 goto dropwithreset; 1559 } 1560 1561 /* 1562 * Segment received on connection. 1563 * Reset idle time and keep-alive timer. 1564 * XXX: This should be done after segment 1565 * validation to ignore broken/spoofed segs. 1566 */ 1567 tp->t_rcvtime = ticks; 1568 1569 /* 1570 * Scale up the window into a 32-bit value. 1571 * For the SYN_SENT state the scale is zero. 1572 */ 1573 tiwin = th->th_win << tp->snd_scale; 1574 1575 /* 1576 * TCP ECN processing. 1577 */ 1578 if (tp->t_flags & TF_ECN_PERMIT) { 1579 if (thflags & TH_CWR) 1580 tp->t_flags &= ~TF_ECN_SND_ECE; 1581 switch (iptos & IPTOS_ECN_MASK) { 1582 case IPTOS_ECN_CE: 1583 tp->t_flags |= TF_ECN_SND_ECE; 1584 TCPSTAT_INC(tcps_ecn_ce); 1585 break; 1586 case IPTOS_ECN_ECT0: 1587 TCPSTAT_INC(tcps_ecn_ect0); 1588 break; 1589 case IPTOS_ECN_ECT1: 1590 TCPSTAT_INC(tcps_ecn_ect1); 1591 break; 1592 } 1593 1594 /* Process a packet differently from RFC3168. */ 1595 cc_ecnpkt_handler(tp, th, iptos); 1596 1597 /* Congestion experienced. */ 1598 if (thflags & TH_ECE) { 1599 cc_cong_signal(tp, th, CC_ECN); 1600 } 1601 } 1602 1603 /* 1604 * Parse options on any incoming segment. 1605 */ 1606 tcp_dooptions(&to, (u_char *)(th + 1), 1607 (th->th_off << 2) - sizeof(struct tcphdr), 1608 (thflags & TH_SYN) ? TO_SYN : 0); 1609 1610 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1611 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1612 (to.to_flags & TOF_SIGNATURE) == 0) { 1613 TCPSTAT_INC(tcps_sig_err_sigopt); 1614 /* XXX: should drop? */ 1615 } 1616 #endif 1617 /* 1618 * If echoed timestamp is later than the current time, 1619 * fall back to non RFC1323 RTT calculation. Normalize 1620 * timestamp if syncookies were used when this connection 1621 * was established. 1622 */ 1623 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1624 to.to_tsecr -= tp->ts_offset; 1625 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1626 to.to_tsecr = 0; 1627 else if (tp->t_flags & TF_PREVVALID && 1628 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1629 cc_cong_signal(tp, th, CC_RTO_ERR); 1630 } 1631 /* 1632 * Process options only when we get SYN/ACK back. The SYN case 1633 * for incoming connections is handled in tcp_syncache. 1634 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1635 * or <SYN,ACK>) segment itself is never scaled. 1636 * XXX this is traditional behavior, may need to be cleaned up. 1637 */ 1638 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1639 if ((to.to_flags & TOF_SCALE) && 1640 (tp->t_flags & TF_REQ_SCALE)) { 1641 tp->t_flags |= TF_RCVD_SCALE; 1642 tp->snd_scale = to.to_wscale; 1643 } 1644 /* 1645 * Initial send window. It will be updated with 1646 * the next incoming segment to the scaled value. 1647 */ 1648 tp->snd_wnd = th->th_win; 1649 if (to.to_flags & TOF_TS) { 1650 tp->t_flags |= TF_RCVD_TSTMP; 1651 tp->ts_recent = to.to_tsval; 1652 tp->ts_recent_age = tcp_ts_getticks(); 1653 } 1654 if (to.to_flags & TOF_MSS) 1655 tcp_mss(tp, to.to_mss); 1656 if ((tp->t_flags & TF_SACK_PERMIT) && 1657 (to.to_flags & TOF_SACKPERM) == 0) 1658 tp->t_flags &= ~TF_SACK_PERMIT; 1659 if (IS_FASTOPEN(tp->t_flags)) { 1660 if (to.to_flags & TOF_FASTOPEN) { 1661 uint16_t mss; 1662 1663 if (to.to_flags & TOF_MSS) 1664 mss = to.to_mss; 1665 else 1666 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1667 mss = TCP6_MSS; 1668 else 1669 mss = TCP_MSS; 1670 tcp_fastopen_update_cache(tp, mss, 1671 to.to_tfo_len, to.to_tfo_cookie); 1672 } else 1673 tcp_fastopen_disable_path(tp); 1674 } 1675 } 1676 1677 /* 1678 * If timestamps were negotiated during SYN/ACK they should 1679 * appear on every segment during this session and vice versa. 1680 */ 1681 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1682 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1683 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1684 "no action\n", s, __func__); 1685 free(s, M_TCPLOG); 1686 } 1687 } 1688 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1689 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1690 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1691 "no action\n", s, __func__); 1692 free(s, M_TCPLOG); 1693 } 1694 } 1695 1696 /* 1697 * Header prediction: check for the two common cases 1698 * of a uni-directional data xfer. If the packet has 1699 * no control flags, is in-sequence, the window didn't 1700 * change and we're not retransmitting, it's a 1701 * candidate. If the length is zero and the ack moved 1702 * forward, we're the sender side of the xfer. Just 1703 * free the data acked & wake any higher level process 1704 * that was blocked waiting for space. If the length 1705 * is non-zero and the ack didn't move, we're the 1706 * receiver side. If we're getting packets in-order 1707 * (the reassembly queue is empty), add the data to 1708 * the socket buffer and note that we need a delayed ack. 1709 * Make sure that the hidden state-flags are also off. 1710 * Since we check for TCPS_ESTABLISHED first, it can only 1711 * be TH_NEEDSYN. 1712 */ 1713 if (tp->t_state == TCPS_ESTABLISHED && 1714 th->th_seq == tp->rcv_nxt && 1715 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1716 tp->snd_nxt == tp->snd_max && 1717 tiwin && tiwin == tp->snd_wnd && 1718 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1719 SEGQ_EMPTY(tp) && 1720 ((to.to_flags & TOF_TS) == 0 || 1721 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1722 1723 /* 1724 * If last ACK falls within this segment's sequence numbers, 1725 * record the timestamp. 1726 * NOTE that the test is modified according to the latest 1727 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1728 */ 1729 if ((to.to_flags & TOF_TS) != 0 && 1730 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1731 tp->ts_recent_age = tcp_ts_getticks(); 1732 tp->ts_recent = to.to_tsval; 1733 } 1734 1735 if (tlen == 0) { 1736 if (SEQ_GT(th->th_ack, tp->snd_una) && 1737 SEQ_LEQ(th->th_ack, tp->snd_max) && 1738 !IN_RECOVERY(tp->t_flags) && 1739 (to.to_flags & TOF_SACK) == 0 && 1740 TAILQ_EMPTY(&tp->snd_holes)) { 1741 /* 1742 * This is a pure ack for outstanding data. 1743 */ 1744 TCPSTAT_INC(tcps_predack); 1745 1746 /* 1747 * "bad retransmit" recovery without timestamps. 1748 */ 1749 if ((to.to_flags & TOF_TS) == 0 && 1750 tp->t_rxtshift == 1 && 1751 tp->t_flags & TF_PREVVALID && 1752 (int)(ticks - tp->t_badrxtwin) < 0) { 1753 cc_cong_signal(tp, th, CC_RTO_ERR); 1754 } 1755 1756 /* 1757 * Recalculate the transmit timer / rtt. 1758 * 1759 * Some boxes send broken timestamp replies 1760 * during the SYN+ACK phase, ignore 1761 * timestamps of 0 or we could calculate a 1762 * huge RTT and blow up the retransmit timer. 1763 */ 1764 if ((to.to_flags & TOF_TS) != 0 && 1765 to.to_tsecr) { 1766 uint32_t t; 1767 1768 t = tcp_ts_getticks() - to.to_tsecr; 1769 if (!tp->t_rttlow || tp->t_rttlow > t) 1770 tp->t_rttlow = t; 1771 tcp_xmit_timer(tp, 1772 TCP_TS_TO_TICKS(t) + 1); 1773 } else if (tp->t_rtttime && 1774 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1775 if (!tp->t_rttlow || 1776 tp->t_rttlow > ticks - tp->t_rtttime) 1777 tp->t_rttlow = ticks - tp->t_rtttime; 1778 tcp_xmit_timer(tp, 1779 ticks - tp->t_rtttime); 1780 } 1781 acked = BYTES_THIS_ACK(tp, th); 1782 1783 #ifdef TCP_HHOOK 1784 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1785 hhook_run_tcp_est_in(tp, th, &to); 1786 #endif 1787 1788 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1789 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1790 sbdrop(&so->so_snd, acked); 1791 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1792 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1793 tp->snd_recover = th->th_ack - 1; 1794 1795 /* 1796 * Let the congestion control algorithm update 1797 * congestion control related information. This 1798 * typically means increasing the congestion 1799 * window. 1800 */ 1801 cc_ack_received(tp, th, nsegs, CC_ACK); 1802 1803 tp->snd_una = th->th_ack; 1804 /* 1805 * Pull snd_wl2 up to prevent seq wrap relative 1806 * to th_ack. 1807 */ 1808 tp->snd_wl2 = th->th_ack; 1809 tp->t_dupacks = 0; 1810 m_freem(m); 1811 1812 /* 1813 * If all outstanding data are acked, stop 1814 * retransmit timer, otherwise restart timer 1815 * using current (possibly backed-off) value. 1816 * If process is waiting for space, 1817 * wakeup/selwakeup/signal. If data 1818 * are ready to send, let tcp_output 1819 * decide between more output or persist. 1820 */ 1821 #ifdef TCPDEBUG 1822 if (so->so_options & SO_DEBUG) 1823 tcp_trace(TA_INPUT, ostate, tp, 1824 (void *)tcp_saveipgen, 1825 &tcp_savetcp, 0); 1826 #endif 1827 TCP_PROBE3(debug__input, tp, th, m); 1828 if (tp->snd_una == tp->snd_max) 1829 tcp_timer_activate(tp, TT_REXMT, 0); 1830 else if (!tcp_timer_active(tp, TT_PERSIST)) 1831 tcp_timer_activate(tp, TT_REXMT, 1832 tp->t_rxtcur); 1833 sowwakeup(so); 1834 if (sbavail(&so->so_snd)) 1835 (void) tp->t_fb->tfb_tcp_output(tp); 1836 goto check_delack; 1837 } 1838 } else if (th->th_ack == tp->snd_una && 1839 tlen <= sbspace(&so->so_rcv)) { 1840 int newsize = 0; /* automatic sockbuf scaling */ 1841 1842 /* 1843 * This is a pure, in-sequence data packet with 1844 * nothing on the reassembly queue and we have enough 1845 * buffer space to take it. 1846 */ 1847 /* Clean receiver SACK report if present */ 1848 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1849 tcp_clean_sackreport(tp); 1850 TCPSTAT_INC(tcps_preddat); 1851 tp->rcv_nxt += tlen; 1852 /* 1853 * Pull snd_wl1 up to prevent seq wrap relative to 1854 * th_seq. 1855 */ 1856 tp->snd_wl1 = th->th_seq; 1857 /* 1858 * Pull rcv_up up to prevent seq wrap relative to 1859 * rcv_nxt. 1860 */ 1861 tp->rcv_up = tp->rcv_nxt; 1862 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1863 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1864 #ifdef TCPDEBUG 1865 if (so->so_options & SO_DEBUG) 1866 tcp_trace(TA_INPUT, ostate, tp, 1867 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1868 #endif 1869 TCP_PROBE3(debug__input, tp, th, m); 1870 1871 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1872 1873 /* Add data to socket buffer. */ 1874 SOCKBUF_LOCK(&so->so_rcv); 1875 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1876 m_freem(m); 1877 } else { 1878 /* 1879 * Set new socket buffer size. 1880 * Give up when limit is reached. 1881 */ 1882 if (newsize) 1883 if (!sbreserve_locked(&so->so_rcv, 1884 newsize, so, NULL)) 1885 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1886 m_adj(m, drop_hdrlen); /* delayed header drop */ 1887 sbappendstream_locked(&so->so_rcv, m, 0); 1888 } 1889 /* NB: sorwakeup_locked() does an implicit unlock. */ 1890 sorwakeup_locked(so); 1891 if (DELAY_ACK(tp, tlen)) { 1892 tp->t_flags |= TF_DELACK; 1893 } else { 1894 tp->t_flags |= TF_ACKNOW; 1895 tp->t_fb->tfb_tcp_output(tp); 1896 } 1897 goto check_delack; 1898 } 1899 } 1900 1901 /* 1902 * Calculate amount of space in receive window, 1903 * and then do TCP input processing. 1904 * Receive window is amount of space in rcv queue, 1905 * but not less than advertised window. 1906 */ 1907 win = sbspace(&so->so_rcv); 1908 if (win < 0) 1909 win = 0; 1910 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1911 1912 switch (tp->t_state) { 1913 1914 /* 1915 * If the state is SYN_RECEIVED: 1916 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1917 */ 1918 case TCPS_SYN_RECEIVED: 1919 if ((thflags & TH_ACK) && 1920 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1921 SEQ_GT(th->th_ack, tp->snd_max))) { 1922 rstreason = BANDLIM_RST_OPENPORT; 1923 goto dropwithreset; 1924 } 1925 if (IS_FASTOPEN(tp->t_flags)) { 1926 /* 1927 * When a TFO connection is in SYN_RECEIVED, the 1928 * only valid packets are the initial SYN, a 1929 * retransmit/copy of the initial SYN (possibly with 1930 * a subset of the original data), a valid ACK, a 1931 * FIN, or a RST. 1932 */ 1933 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1934 rstreason = BANDLIM_RST_OPENPORT; 1935 goto dropwithreset; 1936 } else if (thflags & TH_SYN) { 1937 /* non-initial SYN is ignored */ 1938 if ((tcp_timer_active(tp, TT_DELACK) || 1939 tcp_timer_active(tp, TT_REXMT))) 1940 goto drop; 1941 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1942 goto drop; 1943 } 1944 } 1945 break; 1946 1947 /* 1948 * If the state is SYN_SENT: 1949 * if seg contains a RST with valid ACK (SEQ.ACK has already 1950 * been verified), then drop the connection. 1951 * if seg contains a RST without an ACK, drop the seg. 1952 * if seg does not contain SYN, then drop the seg. 1953 * Otherwise this is an acceptable SYN segment 1954 * initialize tp->rcv_nxt and tp->irs 1955 * if seg contains ack then advance tp->snd_una 1956 * if seg contains an ECE and ECN support is enabled, the stream 1957 * is ECN capable. 1958 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1959 * arrange for segment to be acked (eventually) 1960 * continue processing rest of data/controls, beginning with URG 1961 */ 1962 case TCPS_SYN_SENT: 1963 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1964 TCP_PROBE5(connect__refused, NULL, tp, 1965 m, tp, th); 1966 tp = tcp_drop(tp, ECONNREFUSED); 1967 } 1968 if (thflags & TH_RST) 1969 goto drop; 1970 if (!(thflags & TH_SYN)) 1971 goto drop; 1972 1973 tp->irs = th->th_seq; 1974 tcp_rcvseqinit(tp); 1975 if (thflags & TH_ACK) { 1976 int tfo_partial_ack = 0; 1977 1978 TCPSTAT_INC(tcps_connects); 1979 soisconnected(so); 1980 #ifdef MAC 1981 mac_socketpeer_set_from_mbuf(m, so); 1982 #endif 1983 /* Do window scaling on this connection? */ 1984 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1985 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1986 tp->rcv_scale = tp->request_r_scale; 1987 } 1988 tp->rcv_adv += min(tp->rcv_wnd, 1989 TCP_MAXWIN << tp->rcv_scale); 1990 tp->snd_una++; /* SYN is acked */ 1991 /* 1992 * If not all the data that was sent in the TFO SYN 1993 * has been acked, resend the remainder right away. 1994 */ 1995 if (IS_FASTOPEN(tp->t_flags) && 1996 (tp->snd_una != tp->snd_max)) { 1997 tp->snd_nxt = th->th_ack; 1998 tfo_partial_ack = 1; 1999 } 2000 /* 2001 * If there's data, delay ACK; if there's also a FIN 2002 * ACKNOW will be turned on later. 2003 */ 2004 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2005 tcp_timer_activate(tp, TT_DELACK, 2006 tcp_delacktime); 2007 else 2008 tp->t_flags |= TF_ACKNOW; 2009 2010 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2011 V_tcp_do_ecn) { 2012 tp->t_flags |= TF_ECN_PERMIT; 2013 TCPSTAT_INC(tcps_ecn_shs); 2014 } 2015 2016 /* 2017 * Received <SYN,ACK> in SYN_SENT[*] state. 2018 * Transitions: 2019 * SYN_SENT --> ESTABLISHED 2020 * SYN_SENT* --> FIN_WAIT_1 2021 */ 2022 tp->t_starttime = ticks; 2023 if (tp->t_flags & TF_NEEDFIN) { 2024 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2025 tp->t_flags &= ~TF_NEEDFIN; 2026 thflags &= ~TH_SYN; 2027 } else { 2028 tcp_state_change(tp, TCPS_ESTABLISHED); 2029 TCP_PROBE5(connect__established, NULL, tp, 2030 m, tp, th); 2031 cc_conn_init(tp); 2032 tcp_timer_activate(tp, TT_KEEP, 2033 TP_KEEPIDLE(tp)); 2034 } 2035 } else { 2036 /* 2037 * Received initial SYN in SYN-SENT[*] state => 2038 * simultaneous open. 2039 * If it succeeds, connection is * half-synchronized. 2040 * Otherwise, do 3-way handshake: 2041 * SYN-SENT -> SYN-RECEIVED 2042 * SYN-SENT* -> SYN-RECEIVED* 2043 */ 2044 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2045 tcp_timer_activate(tp, TT_REXMT, 0); 2046 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2047 } 2048 2049 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2050 INP_WLOCK_ASSERT(tp->t_inpcb); 2051 2052 /* 2053 * Advance th->th_seq to correspond to first data byte. 2054 * If data, trim to stay within window, 2055 * dropping FIN if necessary. 2056 */ 2057 th->th_seq++; 2058 if (tlen > tp->rcv_wnd) { 2059 todrop = tlen - tp->rcv_wnd; 2060 m_adj(m, -todrop); 2061 tlen = tp->rcv_wnd; 2062 thflags &= ~TH_FIN; 2063 TCPSTAT_INC(tcps_rcvpackafterwin); 2064 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2065 } 2066 tp->snd_wl1 = th->th_seq - 1; 2067 tp->rcv_up = th->th_seq; 2068 /* 2069 * Client side of transaction: already sent SYN and data. 2070 * If the remote host used T/TCP to validate the SYN, 2071 * our data will be ACK'd; if so, enter normal data segment 2072 * processing in the middle of step 5, ack processing. 2073 * Otherwise, goto step 6. 2074 */ 2075 if (thflags & TH_ACK) 2076 goto process_ACK; 2077 2078 goto step6; 2079 2080 /* 2081 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2082 * do normal processing. 2083 * 2084 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2085 */ 2086 case TCPS_LAST_ACK: 2087 case TCPS_CLOSING: 2088 break; /* continue normal processing */ 2089 } 2090 2091 /* 2092 * States other than LISTEN or SYN_SENT. 2093 * First check the RST flag and sequence number since reset segments 2094 * are exempt from the timestamp and connection count tests. This 2095 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2096 * below which allowed reset segments in half the sequence space 2097 * to fall though and be processed (which gives forged reset 2098 * segments with a random sequence number a 50 percent chance of 2099 * killing a connection). 2100 * Then check timestamp, if present. 2101 * Then check the connection count, if present. 2102 * Then check that at least some bytes of segment are within 2103 * receive window. If segment begins before rcv_nxt, 2104 * drop leading data (and SYN); if nothing left, just ack. 2105 */ 2106 if (thflags & TH_RST) { 2107 /* 2108 * RFC5961 Section 3.2 2109 * 2110 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2111 * - If RST is in window, we send challenge ACK. 2112 * 2113 * Note: to take into account delayed ACKs, we should 2114 * test against last_ack_sent instead of rcv_nxt. 2115 * Note 2: we handle special case of closed window, not 2116 * covered by the RFC. 2117 */ 2118 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2119 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2120 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2121 2122 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2123 KASSERT(tp->t_state != TCPS_SYN_SENT, 2124 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2125 __func__, th, tp)); 2126 2127 if (V_tcp_insecure_rst || 2128 tp->last_ack_sent == th->th_seq) { 2129 TCPSTAT_INC(tcps_drops); 2130 /* Drop the connection. */ 2131 switch (tp->t_state) { 2132 case TCPS_SYN_RECEIVED: 2133 so->so_error = ECONNREFUSED; 2134 goto close; 2135 case TCPS_ESTABLISHED: 2136 case TCPS_FIN_WAIT_1: 2137 case TCPS_FIN_WAIT_2: 2138 case TCPS_CLOSE_WAIT: 2139 case TCPS_CLOSING: 2140 case TCPS_LAST_ACK: 2141 so->so_error = ECONNRESET; 2142 close: 2143 /* FALLTHROUGH */ 2144 default: 2145 tp = tcp_close(tp); 2146 } 2147 } else { 2148 TCPSTAT_INC(tcps_badrst); 2149 /* Send challenge ACK. */ 2150 tcp_respond(tp, mtod(m, void *), th, m, 2151 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2152 tp->last_ack_sent = tp->rcv_nxt; 2153 m = NULL; 2154 } 2155 } 2156 goto drop; 2157 } 2158 2159 /* 2160 * RFC5961 Section 4.2 2161 * Send challenge ACK for any SYN in synchronized state. 2162 */ 2163 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2164 tp->t_state != TCPS_SYN_RECEIVED) { 2165 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2166 2167 TCPSTAT_INC(tcps_badsyn); 2168 if (V_tcp_insecure_syn && 2169 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2170 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2171 tp = tcp_drop(tp, ECONNRESET); 2172 rstreason = BANDLIM_UNLIMITED; 2173 } else { 2174 /* Send challenge ACK. */ 2175 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2176 tp->snd_nxt, TH_ACK); 2177 tp->last_ack_sent = tp->rcv_nxt; 2178 m = NULL; 2179 } 2180 goto drop; 2181 } 2182 2183 /* 2184 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2185 * and it's less than ts_recent, drop it. 2186 */ 2187 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2188 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2189 2190 /* Check to see if ts_recent is over 24 days old. */ 2191 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2192 /* 2193 * Invalidate ts_recent. If this segment updates 2194 * ts_recent, the age will be reset later and ts_recent 2195 * will get a valid value. If it does not, setting 2196 * ts_recent to zero will at least satisfy the 2197 * requirement that zero be placed in the timestamp 2198 * echo reply when ts_recent isn't valid. The 2199 * age isn't reset until we get a valid ts_recent 2200 * because we don't want out-of-order segments to be 2201 * dropped when ts_recent is old. 2202 */ 2203 tp->ts_recent = 0; 2204 } else { 2205 TCPSTAT_INC(tcps_rcvduppack); 2206 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2207 TCPSTAT_INC(tcps_pawsdrop); 2208 if (tlen) 2209 goto dropafterack; 2210 goto drop; 2211 } 2212 } 2213 2214 /* 2215 * In the SYN-RECEIVED state, validate that the packet belongs to 2216 * this connection before trimming the data to fit the receive 2217 * window. Check the sequence number versus IRS since we know 2218 * the sequence numbers haven't wrapped. This is a partial fix 2219 * for the "LAND" DoS attack. 2220 */ 2221 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2222 rstreason = BANDLIM_RST_OPENPORT; 2223 goto dropwithreset; 2224 } 2225 2226 todrop = tp->rcv_nxt - th->th_seq; 2227 if (todrop > 0) { 2228 if (thflags & TH_SYN) { 2229 thflags &= ~TH_SYN; 2230 th->th_seq++; 2231 if (th->th_urp > 1) 2232 th->th_urp--; 2233 else 2234 thflags &= ~TH_URG; 2235 todrop--; 2236 } 2237 /* 2238 * Following if statement from Stevens, vol. 2, p. 960. 2239 */ 2240 if (todrop > tlen 2241 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2242 /* 2243 * Any valid FIN must be to the left of the window. 2244 * At this point the FIN must be a duplicate or out 2245 * of sequence; drop it. 2246 */ 2247 thflags &= ~TH_FIN; 2248 2249 /* 2250 * Send an ACK to resynchronize and drop any data. 2251 * But keep on processing for RST or ACK. 2252 */ 2253 tp->t_flags |= TF_ACKNOW; 2254 todrop = tlen; 2255 TCPSTAT_INC(tcps_rcvduppack); 2256 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2257 } else { 2258 TCPSTAT_INC(tcps_rcvpartduppack); 2259 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2260 } 2261 /* 2262 * DSACK - add SACK block for dropped range 2263 */ 2264 if (tp->t_flags & TF_SACK_PERMIT) { 2265 tcp_update_sack_list(tp, th->th_seq, th->th_seq+tlen); 2266 /* 2267 * ACK now, as the next in-sequence segment 2268 * will clear the DSACK block again 2269 */ 2270 tp->t_flags |= TF_ACKNOW; 2271 } 2272 drop_hdrlen += todrop; /* drop from the top afterwards */ 2273 th->th_seq += todrop; 2274 tlen -= todrop; 2275 if (th->th_urp > todrop) 2276 th->th_urp -= todrop; 2277 else { 2278 thflags &= ~TH_URG; 2279 th->th_urp = 0; 2280 } 2281 } 2282 2283 /* 2284 * If new data are received on a connection after the 2285 * user processes are gone, then RST the other end. 2286 */ 2287 if ((so->so_state & SS_NOFDREF) && 2288 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2289 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2290 2291 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2292 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2293 "after socket was closed, " 2294 "sending RST and removing tcpcb\n", 2295 s, __func__, tcpstates[tp->t_state], tlen); 2296 free(s, M_TCPLOG); 2297 } 2298 tp = tcp_close(tp); 2299 TCPSTAT_INC(tcps_rcvafterclose); 2300 rstreason = BANDLIM_UNLIMITED; 2301 goto dropwithreset; 2302 } 2303 2304 /* 2305 * If segment ends after window, drop trailing data 2306 * (and PUSH and FIN); if nothing left, just ACK. 2307 */ 2308 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2309 if (todrop > 0) { 2310 TCPSTAT_INC(tcps_rcvpackafterwin); 2311 if (todrop >= tlen) { 2312 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2313 /* 2314 * If window is closed can only take segments at 2315 * window edge, and have to drop data and PUSH from 2316 * incoming segments. Continue processing, but 2317 * remember to ack. Otherwise, drop segment 2318 * and ack. 2319 */ 2320 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2321 tp->t_flags |= TF_ACKNOW; 2322 TCPSTAT_INC(tcps_rcvwinprobe); 2323 } else 2324 goto dropafterack; 2325 } else 2326 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2327 m_adj(m, -todrop); 2328 tlen -= todrop; 2329 thflags &= ~(TH_PUSH|TH_FIN); 2330 } 2331 2332 /* 2333 * If last ACK falls within this segment's sequence numbers, 2334 * record its timestamp. 2335 * NOTE: 2336 * 1) That the test incorporates suggestions from the latest 2337 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2338 * 2) That updating only on newer timestamps interferes with 2339 * our earlier PAWS tests, so this check should be solely 2340 * predicated on the sequence space of this segment. 2341 * 3) That we modify the segment boundary check to be 2342 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2343 * instead of RFC1323's 2344 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2345 * This modified check allows us to overcome RFC1323's 2346 * limitations as described in Stevens TCP/IP Illustrated 2347 * Vol. 2 p.869. In such cases, we can still calculate the 2348 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2349 */ 2350 if ((to.to_flags & TOF_TS) != 0 && 2351 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2352 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2353 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2354 tp->ts_recent_age = tcp_ts_getticks(); 2355 tp->ts_recent = to.to_tsval; 2356 } 2357 2358 /* 2359 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2360 * flag is on (half-synchronized state), then queue data for 2361 * later processing; else drop segment and return. 2362 */ 2363 if ((thflags & TH_ACK) == 0) { 2364 if (tp->t_state == TCPS_SYN_RECEIVED || 2365 (tp->t_flags & TF_NEEDSYN)) { 2366 if (tp->t_state == TCPS_SYN_RECEIVED && 2367 IS_FASTOPEN(tp->t_flags)) { 2368 tp->snd_wnd = tiwin; 2369 cc_conn_init(tp); 2370 } 2371 goto step6; 2372 } else if (tp->t_flags & TF_ACKNOW) 2373 goto dropafterack; 2374 else 2375 goto drop; 2376 } 2377 2378 /* 2379 * Ack processing. 2380 */ 2381 switch (tp->t_state) { 2382 2383 /* 2384 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2385 * ESTABLISHED state and continue processing. 2386 * The ACK was checked above. 2387 */ 2388 case TCPS_SYN_RECEIVED: 2389 2390 TCPSTAT_INC(tcps_connects); 2391 soisconnected(so); 2392 /* Do window scaling? */ 2393 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2394 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2395 tp->rcv_scale = tp->request_r_scale; 2396 } 2397 tp->snd_wnd = tiwin; 2398 /* 2399 * Make transitions: 2400 * SYN-RECEIVED -> ESTABLISHED 2401 * SYN-RECEIVED* -> FIN-WAIT-1 2402 */ 2403 tp->t_starttime = ticks; 2404 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2405 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2406 tp->t_tfo_pending = NULL; 2407 2408 /* 2409 * Account for the ACK of our SYN prior to 2410 * regular ACK processing below. 2411 */ 2412 tp->snd_una++; 2413 } 2414 if (tp->t_flags & TF_NEEDFIN) { 2415 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2416 tp->t_flags &= ~TF_NEEDFIN; 2417 } else { 2418 tcp_state_change(tp, TCPS_ESTABLISHED); 2419 TCP_PROBE5(accept__established, NULL, tp, 2420 m, tp, th); 2421 /* 2422 * TFO connections call cc_conn_init() during SYN 2423 * processing. Calling it again here for such 2424 * connections is not harmless as it would undo the 2425 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2426 * is retransmitted. 2427 */ 2428 if (!IS_FASTOPEN(tp->t_flags)) 2429 cc_conn_init(tp); 2430 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2431 } 2432 /* 2433 * If segment contains data or ACK, will call tcp_reass() 2434 * later; if not, do so now to pass queued data to user. 2435 */ 2436 if (tlen == 0 && (thflags & TH_FIN) == 0) 2437 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2438 (struct mbuf *)0); 2439 tp->snd_wl1 = th->th_seq - 1; 2440 /* FALLTHROUGH */ 2441 2442 /* 2443 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2444 * ACKs. If the ack is in the range 2445 * tp->snd_una < th->th_ack <= tp->snd_max 2446 * then advance tp->snd_una to th->th_ack and drop 2447 * data from the retransmission queue. If this ACK reflects 2448 * more up to date window information we update our window information. 2449 */ 2450 case TCPS_ESTABLISHED: 2451 case TCPS_FIN_WAIT_1: 2452 case TCPS_FIN_WAIT_2: 2453 case TCPS_CLOSE_WAIT: 2454 case TCPS_CLOSING: 2455 case TCPS_LAST_ACK: 2456 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2457 TCPSTAT_INC(tcps_rcvacktoomuch); 2458 goto dropafterack; 2459 } 2460 if ((tp->t_flags & TF_SACK_PERMIT) && 2461 ((to.to_flags & TOF_SACK) || 2462 !TAILQ_EMPTY(&tp->snd_holes))) 2463 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2464 else 2465 /* 2466 * Reset the value so that previous (valid) value 2467 * from the last ack with SACK doesn't get used. 2468 */ 2469 tp->sackhint.sacked_bytes = 0; 2470 2471 #ifdef TCP_HHOOK 2472 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2473 hhook_run_tcp_est_in(tp, th, &to); 2474 #endif 2475 2476 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2477 u_int maxseg; 2478 2479 maxseg = tcp_maxseg(tp); 2480 if (tlen == 0 && 2481 (tiwin == tp->snd_wnd || 2482 (tp->t_flags & TF_SACK_PERMIT))) { 2483 /* 2484 * If this is the first time we've seen a 2485 * FIN from the remote, this is not a 2486 * duplicate and it needs to be processed 2487 * normally. This happens during a 2488 * simultaneous close. 2489 */ 2490 if ((thflags & TH_FIN) && 2491 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2492 tp->t_dupacks = 0; 2493 break; 2494 } 2495 TCPSTAT_INC(tcps_rcvdupack); 2496 /* 2497 * If we have outstanding data (other than 2498 * a window probe), this is a completely 2499 * duplicate ack (ie, window info didn't 2500 * change and FIN isn't set), 2501 * the ack is the biggest we've 2502 * seen and we've seen exactly our rexmt 2503 * threshold of them, assume a packet 2504 * has been dropped and retransmit it. 2505 * Kludge snd_nxt & the congestion 2506 * window so we send only this one 2507 * packet. 2508 * 2509 * We know we're losing at the current 2510 * window size so do congestion avoidance 2511 * (set ssthresh to half the current window 2512 * and pull our congestion window back to 2513 * the new ssthresh). 2514 * 2515 * Dup acks mean that packets have left the 2516 * network (they're now cached at the receiver) 2517 * so bump cwnd by the amount in the receiver 2518 * to keep a constant cwnd packets in the 2519 * network. 2520 * 2521 * When using TCP ECN, notify the peer that 2522 * we reduced the cwnd. 2523 */ 2524 /* 2525 * Following 2 kinds of acks should not affect 2526 * dupack counting: 2527 * 1) Old acks 2528 * 2) Acks with SACK but without any new SACK 2529 * information in them. These could result from 2530 * any anomaly in the network like a switch 2531 * duplicating packets or a possible DoS attack. 2532 */ 2533 if (th->th_ack != tp->snd_una || 2534 ((tp->t_flags & TF_SACK_PERMIT) && 2535 !sack_changed)) 2536 break; 2537 else if (!tcp_timer_active(tp, TT_REXMT)) 2538 tp->t_dupacks = 0; 2539 else if (++tp->t_dupacks > tcprexmtthresh || 2540 IN_FASTRECOVERY(tp->t_flags)) { 2541 cc_ack_received(tp, th, nsegs, 2542 CC_DUPACK); 2543 if ((tp->t_flags & TF_SACK_PERMIT) && 2544 IN_FASTRECOVERY(tp->t_flags)) { 2545 int awnd; 2546 2547 /* 2548 * Compute the amount of data in flight first. 2549 * We can inject new data into the pipe iff 2550 * we have less than 1/2 the original window's 2551 * worth of data in flight. 2552 */ 2553 if (V_tcp_do_rfc6675_pipe) 2554 awnd = tcp_compute_pipe(tp); 2555 else 2556 awnd = (tp->snd_nxt - tp->snd_fack) + 2557 tp->sackhint.sack_bytes_rexmit; 2558 2559 if (awnd < tp->snd_ssthresh) { 2560 tp->snd_cwnd += maxseg; 2561 if (tp->snd_cwnd > tp->snd_ssthresh) 2562 tp->snd_cwnd = tp->snd_ssthresh; 2563 } 2564 } else 2565 tp->snd_cwnd += maxseg; 2566 (void) tp->t_fb->tfb_tcp_output(tp); 2567 goto drop; 2568 } else if (tp->t_dupacks == tcprexmtthresh) { 2569 tcp_seq onxt = tp->snd_nxt; 2570 2571 /* 2572 * If we're doing sack, check to 2573 * see if we're already in sack 2574 * recovery. If we're not doing sack, 2575 * check to see if we're in newreno 2576 * recovery. 2577 */ 2578 if (tp->t_flags & TF_SACK_PERMIT) { 2579 if (IN_FASTRECOVERY(tp->t_flags)) { 2580 tp->t_dupacks = 0; 2581 break; 2582 } 2583 } else { 2584 if (SEQ_LEQ(th->th_ack, 2585 tp->snd_recover)) { 2586 tp->t_dupacks = 0; 2587 break; 2588 } 2589 } 2590 /* Congestion signal before ack. */ 2591 cc_cong_signal(tp, th, CC_NDUPACK); 2592 cc_ack_received(tp, th, nsegs, 2593 CC_DUPACK); 2594 tcp_timer_activate(tp, TT_REXMT, 0); 2595 tp->t_rtttime = 0; 2596 if (tp->t_flags & TF_SACK_PERMIT) { 2597 TCPSTAT_INC( 2598 tcps_sack_recovery_episode); 2599 tp->sack_newdata = tp->snd_nxt; 2600 tp->snd_cwnd = maxseg; 2601 (void) tp->t_fb->tfb_tcp_output(tp); 2602 goto drop; 2603 } 2604 tp->snd_nxt = th->th_ack; 2605 tp->snd_cwnd = maxseg; 2606 (void) tp->t_fb->tfb_tcp_output(tp); 2607 KASSERT(tp->snd_limited <= 2, 2608 ("%s: tp->snd_limited too big", 2609 __func__)); 2610 tp->snd_cwnd = tp->snd_ssthresh + 2611 maxseg * 2612 (tp->t_dupacks - tp->snd_limited); 2613 if (SEQ_GT(onxt, tp->snd_nxt)) 2614 tp->snd_nxt = onxt; 2615 goto drop; 2616 } else if (V_tcp_do_rfc3042) { 2617 /* 2618 * Process first and second duplicate 2619 * ACKs. Each indicates a segment 2620 * leaving the network, creating room 2621 * for more. Make sure we can send a 2622 * packet on reception of each duplicate 2623 * ACK by increasing snd_cwnd by one 2624 * segment. Restore the original 2625 * snd_cwnd after packet transmission. 2626 */ 2627 cc_ack_received(tp, th, nsegs, 2628 CC_DUPACK); 2629 uint32_t oldcwnd = tp->snd_cwnd; 2630 tcp_seq oldsndmax = tp->snd_max; 2631 u_int sent; 2632 int avail; 2633 2634 KASSERT(tp->t_dupacks == 1 || 2635 tp->t_dupacks == 2, 2636 ("%s: dupacks not 1 or 2", 2637 __func__)); 2638 if (tp->t_dupacks == 1) 2639 tp->snd_limited = 0; 2640 tp->snd_cwnd = 2641 (tp->snd_nxt - tp->snd_una) + 2642 (tp->t_dupacks - tp->snd_limited) * 2643 maxseg; 2644 /* 2645 * Only call tcp_output when there 2646 * is new data available to be sent. 2647 * Otherwise we would send pure ACKs. 2648 */ 2649 SOCKBUF_LOCK(&so->so_snd); 2650 avail = sbavail(&so->so_snd) - 2651 (tp->snd_nxt - tp->snd_una); 2652 SOCKBUF_UNLOCK(&so->so_snd); 2653 if (avail > 0) 2654 (void) tp->t_fb->tfb_tcp_output(tp); 2655 sent = tp->snd_max - oldsndmax; 2656 if (sent > maxseg) { 2657 KASSERT((tp->t_dupacks == 2 && 2658 tp->snd_limited == 0) || 2659 (sent == maxseg + 1 && 2660 tp->t_flags & TF_SENTFIN), 2661 ("%s: sent too much", 2662 __func__)); 2663 tp->snd_limited = 2; 2664 } else if (sent > 0) 2665 ++tp->snd_limited; 2666 tp->snd_cwnd = oldcwnd; 2667 goto drop; 2668 } 2669 } 2670 break; 2671 } else { 2672 /* 2673 * This ack is advancing the left edge, reset the 2674 * counter. 2675 */ 2676 tp->t_dupacks = 0; 2677 /* 2678 * If this ack also has new SACK info, increment the 2679 * counter as per rfc6675. 2680 */ 2681 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2682 tp->t_dupacks++; 2683 } 2684 2685 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2686 ("%s: th_ack <= snd_una", __func__)); 2687 2688 /* 2689 * If the congestion window was inflated to account 2690 * for the other side's cached packets, retract it. 2691 */ 2692 if (IN_FASTRECOVERY(tp->t_flags)) { 2693 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2694 if (tp->t_flags & TF_SACK_PERMIT) 2695 tcp_sack_partialack(tp, th); 2696 else 2697 tcp_newreno_partial_ack(tp, th); 2698 } else 2699 cc_post_recovery(tp, th); 2700 } 2701 /* 2702 * If we reach this point, ACK is not a duplicate, 2703 * i.e., it ACKs something we sent. 2704 */ 2705 if (tp->t_flags & TF_NEEDSYN) { 2706 /* 2707 * T/TCP: Connection was half-synchronized, and our 2708 * SYN has been ACK'd (so connection is now fully 2709 * synchronized). Go to non-starred state, 2710 * increment snd_una for ACK of SYN, and check if 2711 * we can do window scaling. 2712 */ 2713 tp->t_flags &= ~TF_NEEDSYN; 2714 tp->snd_una++; 2715 /* Do window scaling? */ 2716 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2717 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2718 tp->rcv_scale = tp->request_r_scale; 2719 /* Send window already scaled. */ 2720 } 2721 } 2722 2723 process_ACK: 2724 INP_WLOCK_ASSERT(tp->t_inpcb); 2725 2726 acked = BYTES_THIS_ACK(tp, th); 2727 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2728 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2729 tp->snd_una, th->th_ack, tp, m)); 2730 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2731 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2732 2733 /* 2734 * If we just performed our first retransmit, and the ACK 2735 * arrives within our recovery window, then it was a mistake 2736 * to do the retransmit in the first place. Recover our 2737 * original cwnd and ssthresh, and proceed to transmit where 2738 * we left off. 2739 */ 2740 if (tp->t_rxtshift == 1 && 2741 tp->t_flags & TF_PREVVALID && 2742 tp->t_badrxtwin && 2743 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2744 cc_cong_signal(tp, th, CC_RTO_ERR); 2745 2746 /* 2747 * If we have a timestamp reply, update smoothed 2748 * round trip time. If no timestamp is present but 2749 * transmit timer is running and timed sequence 2750 * number was acked, update smoothed round trip time. 2751 * Since we now have an rtt measurement, cancel the 2752 * timer backoff (cf., Phil Karn's retransmit alg.). 2753 * Recompute the initial retransmit timer. 2754 * 2755 * Some boxes send broken timestamp replies 2756 * during the SYN+ACK phase, ignore 2757 * timestamps of 0 or we could calculate a 2758 * huge RTT and blow up the retransmit timer. 2759 */ 2760 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2761 uint32_t t; 2762 2763 t = tcp_ts_getticks() - to.to_tsecr; 2764 if (!tp->t_rttlow || tp->t_rttlow > t) 2765 tp->t_rttlow = t; 2766 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2767 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2768 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2769 tp->t_rttlow = ticks - tp->t_rtttime; 2770 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2771 } 2772 2773 /* 2774 * If all outstanding data is acked, stop retransmit 2775 * timer and remember to restart (more output or persist). 2776 * If there is more data to be acked, restart retransmit 2777 * timer, using current (possibly backed-off) value. 2778 */ 2779 if (th->th_ack == tp->snd_max) { 2780 tcp_timer_activate(tp, TT_REXMT, 0); 2781 needoutput = 1; 2782 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2783 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2784 2785 /* 2786 * If no data (only SYN) was ACK'd, 2787 * skip rest of ACK processing. 2788 */ 2789 if (acked == 0) 2790 goto step6; 2791 2792 /* 2793 * Let the congestion control algorithm update congestion 2794 * control related information. This typically means increasing 2795 * the congestion window. 2796 */ 2797 cc_ack_received(tp, th, nsegs, CC_ACK); 2798 2799 SOCKBUF_LOCK(&so->so_snd); 2800 if (acked > sbavail(&so->so_snd)) { 2801 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2802 tp->snd_wnd -= sbavail(&so->so_snd); 2803 else 2804 tp->snd_wnd = 0; 2805 mfree = sbcut_locked(&so->so_snd, 2806 (int)sbavail(&so->so_snd)); 2807 ourfinisacked = 1; 2808 } else { 2809 mfree = sbcut_locked(&so->so_snd, acked); 2810 if (tp->snd_wnd >= (uint32_t) acked) 2811 tp->snd_wnd -= acked; 2812 else 2813 tp->snd_wnd = 0; 2814 ourfinisacked = 0; 2815 } 2816 /* NB: sowwakeup_locked() does an implicit unlock. */ 2817 sowwakeup_locked(so); 2818 m_freem(mfree); 2819 /* Detect una wraparound. */ 2820 if (!IN_RECOVERY(tp->t_flags) && 2821 SEQ_GT(tp->snd_una, tp->snd_recover) && 2822 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2823 tp->snd_recover = th->th_ack - 1; 2824 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2825 if (IN_RECOVERY(tp->t_flags) && 2826 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2827 EXIT_RECOVERY(tp->t_flags); 2828 } 2829 tp->snd_una = th->th_ack; 2830 if (tp->t_flags & TF_SACK_PERMIT) { 2831 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2832 tp->snd_recover = tp->snd_una; 2833 } 2834 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2835 tp->snd_nxt = tp->snd_una; 2836 2837 switch (tp->t_state) { 2838 2839 /* 2840 * In FIN_WAIT_1 STATE in addition to the processing 2841 * for the ESTABLISHED state if our FIN is now acknowledged 2842 * then enter FIN_WAIT_2. 2843 */ 2844 case TCPS_FIN_WAIT_1: 2845 if (ourfinisacked) { 2846 /* 2847 * If we can't receive any more 2848 * data, then closing user can proceed. 2849 * Starting the timer is contrary to the 2850 * specification, but if we don't get a FIN 2851 * we'll hang forever. 2852 * 2853 * XXXjl: 2854 * we should release the tp also, and use a 2855 * compressed state. 2856 */ 2857 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2858 soisdisconnected(so); 2859 tcp_timer_activate(tp, TT_2MSL, 2860 (tcp_fast_finwait2_recycle ? 2861 tcp_finwait2_timeout : 2862 TP_MAXIDLE(tp))); 2863 } 2864 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2865 } 2866 break; 2867 2868 /* 2869 * In CLOSING STATE in addition to the processing for 2870 * the ESTABLISHED state if the ACK acknowledges our FIN 2871 * then enter the TIME-WAIT state, otherwise ignore 2872 * the segment. 2873 */ 2874 case TCPS_CLOSING: 2875 if (ourfinisacked) { 2876 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2877 tcp_twstart(tp); 2878 m_freem(m); 2879 return; 2880 } 2881 break; 2882 2883 /* 2884 * In LAST_ACK, we may still be waiting for data to drain 2885 * and/or to be acked, as well as for the ack of our FIN. 2886 * If our FIN is now acknowledged, delete the TCB, 2887 * enter the closed state and return. 2888 */ 2889 case TCPS_LAST_ACK: 2890 if (ourfinisacked) { 2891 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2892 tp = tcp_close(tp); 2893 goto drop; 2894 } 2895 break; 2896 } 2897 } 2898 2899 step6: 2900 INP_WLOCK_ASSERT(tp->t_inpcb); 2901 2902 /* 2903 * Update window information. 2904 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2905 */ 2906 if ((thflags & TH_ACK) && 2907 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2908 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2909 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2910 /* keep track of pure window updates */ 2911 if (tlen == 0 && 2912 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2913 TCPSTAT_INC(tcps_rcvwinupd); 2914 tp->snd_wnd = tiwin; 2915 tp->snd_wl1 = th->th_seq; 2916 tp->snd_wl2 = th->th_ack; 2917 if (tp->snd_wnd > tp->max_sndwnd) 2918 tp->max_sndwnd = tp->snd_wnd; 2919 needoutput = 1; 2920 } 2921 2922 /* 2923 * Process segments with URG. 2924 */ 2925 if ((thflags & TH_URG) && th->th_urp && 2926 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2927 /* 2928 * This is a kludge, but if we receive and accept 2929 * random urgent pointers, we'll crash in 2930 * soreceive. It's hard to imagine someone 2931 * actually wanting to send this much urgent data. 2932 */ 2933 SOCKBUF_LOCK(&so->so_rcv); 2934 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2935 th->th_urp = 0; /* XXX */ 2936 thflags &= ~TH_URG; /* XXX */ 2937 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2938 goto dodata; /* XXX */ 2939 } 2940 /* 2941 * If this segment advances the known urgent pointer, 2942 * then mark the data stream. This should not happen 2943 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2944 * a FIN has been received from the remote side. 2945 * In these states we ignore the URG. 2946 * 2947 * According to RFC961 (Assigned Protocols), 2948 * the urgent pointer points to the last octet 2949 * of urgent data. We continue, however, 2950 * to consider it to indicate the first octet 2951 * of data past the urgent section as the original 2952 * spec states (in one of two places). 2953 */ 2954 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2955 tp->rcv_up = th->th_seq + th->th_urp; 2956 so->so_oobmark = sbavail(&so->so_rcv) + 2957 (tp->rcv_up - tp->rcv_nxt) - 1; 2958 if (so->so_oobmark == 0) 2959 so->so_rcv.sb_state |= SBS_RCVATMARK; 2960 sohasoutofband(so); 2961 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2962 } 2963 SOCKBUF_UNLOCK(&so->so_rcv); 2964 /* 2965 * Remove out of band data so doesn't get presented to user. 2966 * This can happen independent of advancing the URG pointer, 2967 * but if two URG's are pending at once, some out-of-band 2968 * data may creep in... ick. 2969 */ 2970 if (th->th_urp <= (uint32_t)tlen && 2971 !(so->so_options & SO_OOBINLINE)) { 2972 /* hdr drop is delayed */ 2973 tcp_pulloutofband(so, th, m, drop_hdrlen); 2974 } 2975 } else { 2976 /* 2977 * If no out of band data is expected, 2978 * pull receive urgent pointer along 2979 * with the receive window. 2980 */ 2981 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2982 tp->rcv_up = tp->rcv_nxt; 2983 } 2984 dodata: /* XXX */ 2985 INP_WLOCK_ASSERT(tp->t_inpcb); 2986 2987 /* 2988 * Process the segment text, merging it into the TCP sequencing queue, 2989 * and arranging for acknowledgment of receipt if necessary. 2990 * This process logically involves adjusting tp->rcv_wnd as data 2991 * is presented to the user (this happens in tcp_usrreq.c, 2992 * case PRU_RCVD). If a FIN has already been received on this 2993 * connection then we just ignore the text. 2994 */ 2995 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 2996 IS_FASTOPEN(tp->t_flags)); 2997 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 2998 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2999 tcp_seq save_start = th->th_seq; 3000 tcp_seq save_rnxt = tp->rcv_nxt; 3001 int save_tlen = tlen; 3002 m_adj(m, drop_hdrlen); /* delayed header drop */ 3003 /* 3004 * Insert segment which includes th into TCP reassembly queue 3005 * with control block tp. Set thflags to whether reassembly now 3006 * includes a segment with FIN. This handles the common case 3007 * inline (segment is the next to be received on an established 3008 * connection, and the queue is empty), avoiding linkage into 3009 * and removal from the queue and repetition of various 3010 * conversions. 3011 * Set DELACK for segments received in order, but ack 3012 * immediately when segments are out of order (so 3013 * fast retransmit can work). 3014 */ 3015 if (th->th_seq == tp->rcv_nxt && 3016 SEGQ_EMPTY(tp) && 3017 (TCPS_HAVEESTABLISHED(tp->t_state) || 3018 tfo_syn)) { 3019 if (DELAY_ACK(tp, tlen) || tfo_syn) 3020 tp->t_flags |= TF_DELACK; 3021 else 3022 tp->t_flags |= TF_ACKNOW; 3023 tp->rcv_nxt += tlen; 3024 thflags = th->th_flags & TH_FIN; 3025 TCPSTAT_INC(tcps_rcvpack); 3026 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3027 SOCKBUF_LOCK(&so->so_rcv); 3028 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3029 m_freem(m); 3030 else 3031 sbappendstream_locked(&so->so_rcv, m, 0); 3032 /* NB: sorwakeup_locked() does an implicit unlock. */ 3033 sorwakeup_locked(so); 3034 } else { 3035 /* 3036 * XXX: Due to the header drop above "th" is 3037 * theoretically invalid by now. Fortunately 3038 * m_adj() doesn't actually frees any mbufs 3039 * when trimming from the head. 3040 */ 3041 tcp_seq temp = save_start; 3042 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3043 tp->t_flags |= TF_ACKNOW; 3044 } 3045 if (tp->t_flags & TF_SACK_PERMIT) { 3046 if (((tlen == 0) && (save_tlen > 0) && 3047 (SEQ_LT(save_start, save_rnxt)))) { 3048 /* 3049 * DSACK actually handled in the fastpath 3050 * above. 3051 */ 3052 tcp_update_sack_list(tp, save_start, save_start + save_tlen); 3053 } else 3054 if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3055 /* 3056 * Cleaning sackblks by using zero length 3057 * update. 3058 */ 3059 tcp_update_sack_list(tp, save_start, save_start); 3060 } else 3061 if ((tlen > 0) && (tlen >= save_tlen)) { 3062 /* Update of sackblks. */ 3063 tcp_update_sack_list(tp, save_start, save_start + save_tlen); 3064 } else 3065 if (tlen > 0) { 3066 tcp_update_sack_list(tp, save_start, save_start+tlen); 3067 } 3068 } 3069 #if 0 3070 /* 3071 * Note the amount of data that peer has sent into 3072 * our window, in order to estimate the sender's 3073 * buffer size. 3074 * XXX: Unused. 3075 */ 3076 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3077 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3078 else 3079 len = so->so_rcv.sb_hiwat; 3080 #endif 3081 } else { 3082 m_freem(m); 3083 thflags &= ~TH_FIN; 3084 } 3085 3086 /* 3087 * If FIN is received ACK the FIN and let the user know 3088 * that the connection is closing. 3089 */ 3090 if (thflags & TH_FIN) { 3091 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3092 socantrcvmore(so); 3093 /* 3094 * If connection is half-synchronized 3095 * (ie NEEDSYN flag on) then delay ACK, 3096 * so it may be piggybacked when SYN is sent. 3097 * Otherwise, since we received a FIN then no 3098 * more input can be expected, send ACK now. 3099 */ 3100 if (tp->t_flags & TF_NEEDSYN) 3101 tp->t_flags |= TF_DELACK; 3102 else 3103 tp->t_flags |= TF_ACKNOW; 3104 tp->rcv_nxt++; 3105 } 3106 switch (tp->t_state) { 3107 3108 /* 3109 * In SYN_RECEIVED and ESTABLISHED STATES 3110 * enter the CLOSE_WAIT state. 3111 */ 3112 case TCPS_SYN_RECEIVED: 3113 tp->t_starttime = ticks; 3114 /* FALLTHROUGH */ 3115 case TCPS_ESTABLISHED: 3116 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3117 break; 3118 3119 /* 3120 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3121 * enter the CLOSING state. 3122 */ 3123 case TCPS_FIN_WAIT_1: 3124 tcp_state_change(tp, TCPS_CLOSING); 3125 break; 3126 3127 /* 3128 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3129 * starting the time-wait timer, turning off the other 3130 * standard timers. 3131 */ 3132 case TCPS_FIN_WAIT_2: 3133 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3134 3135 tcp_twstart(tp); 3136 return; 3137 } 3138 } 3139 #ifdef TCPDEBUG 3140 if (so->so_options & SO_DEBUG) 3141 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3142 &tcp_savetcp, 0); 3143 #endif 3144 TCP_PROBE3(debug__input, tp, th, m); 3145 3146 /* 3147 * Return any desired output. 3148 */ 3149 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3150 (void) tp->t_fb->tfb_tcp_output(tp); 3151 3152 check_delack: 3153 INP_WLOCK_ASSERT(tp->t_inpcb); 3154 3155 if (tp->t_flags & TF_DELACK) { 3156 tp->t_flags &= ~TF_DELACK; 3157 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3158 } 3159 INP_WUNLOCK(tp->t_inpcb); 3160 return; 3161 3162 dropafterack: 3163 /* 3164 * Generate an ACK dropping incoming segment if it occupies 3165 * sequence space, where the ACK reflects our state. 3166 * 3167 * We can now skip the test for the RST flag since all 3168 * paths to this code happen after packets containing 3169 * RST have been dropped. 3170 * 3171 * In the SYN-RECEIVED state, don't send an ACK unless the 3172 * segment we received passes the SYN-RECEIVED ACK test. 3173 * If it fails send a RST. This breaks the loop in the 3174 * "LAND" DoS attack, and also prevents an ACK storm 3175 * between two listening ports that have been sent forged 3176 * SYN segments, each with the source address of the other. 3177 */ 3178 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3179 (SEQ_GT(tp->snd_una, th->th_ack) || 3180 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3181 rstreason = BANDLIM_RST_OPENPORT; 3182 goto dropwithreset; 3183 } 3184 #ifdef TCPDEBUG 3185 if (so->so_options & SO_DEBUG) 3186 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3187 &tcp_savetcp, 0); 3188 #endif 3189 TCP_PROBE3(debug__input, tp, th, m); 3190 tp->t_flags |= TF_ACKNOW; 3191 (void) tp->t_fb->tfb_tcp_output(tp); 3192 INP_WUNLOCK(tp->t_inpcb); 3193 m_freem(m); 3194 return; 3195 3196 dropwithreset: 3197 if (tp != NULL) { 3198 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3199 INP_WUNLOCK(tp->t_inpcb); 3200 } else 3201 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3202 return; 3203 3204 drop: 3205 /* 3206 * Drop space held by incoming segment and return. 3207 */ 3208 #ifdef TCPDEBUG 3209 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3210 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3211 &tcp_savetcp, 0); 3212 #endif 3213 TCP_PROBE3(debug__input, tp, th, m); 3214 if (tp != NULL) 3215 INP_WUNLOCK(tp->t_inpcb); 3216 m_freem(m); 3217 } 3218 3219 /* 3220 * Issue RST and make ACK acceptable to originator of segment. 3221 * The mbuf must still include the original packet header. 3222 * tp may be NULL. 3223 */ 3224 void 3225 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3226 int tlen, int rstreason) 3227 { 3228 #ifdef INET 3229 struct ip *ip; 3230 #endif 3231 #ifdef INET6 3232 struct ip6_hdr *ip6; 3233 #endif 3234 3235 if (tp != NULL) { 3236 INP_WLOCK_ASSERT(tp->t_inpcb); 3237 } 3238 3239 /* Don't bother if destination was broadcast/multicast. */ 3240 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3241 goto drop; 3242 #ifdef INET6 3243 if (mtod(m, struct ip *)->ip_v == 6) { 3244 ip6 = mtod(m, struct ip6_hdr *); 3245 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3246 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3247 goto drop; 3248 /* IPv6 anycast check is done at tcp6_input() */ 3249 } 3250 #endif 3251 #if defined(INET) && defined(INET6) 3252 else 3253 #endif 3254 #ifdef INET 3255 { 3256 ip = mtod(m, struct ip *); 3257 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3258 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3259 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3260 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3261 goto drop; 3262 } 3263 #endif 3264 3265 /* Perform bandwidth limiting. */ 3266 if (badport_bandlim(rstreason) < 0) 3267 goto drop; 3268 3269 /* tcp_respond consumes the mbuf chain. */ 3270 if (th->th_flags & TH_ACK) { 3271 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3272 th->th_ack, TH_RST); 3273 } else { 3274 if (th->th_flags & TH_SYN) 3275 tlen++; 3276 if (th->th_flags & TH_FIN) 3277 tlen++; 3278 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3279 (tcp_seq)0, TH_RST|TH_ACK); 3280 } 3281 return; 3282 drop: 3283 m_freem(m); 3284 } 3285 3286 /* 3287 * Parse TCP options and place in tcpopt. 3288 */ 3289 void 3290 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3291 { 3292 int opt, optlen; 3293 3294 to->to_flags = 0; 3295 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3296 opt = cp[0]; 3297 if (opt == TCPOPT_EOL) 3298 break; 3299 if (opt == TCPOPT_NOP) 3300 optlen = 1; 3301 else { 3302 if (cnt < 2) 3303 break; 3304 optlen = cp[1]; 3305 if (optlen < 2 || optlen > cnt) 3306 break; 3307 } 3308 switch (opt) { 3309 case TCPOPT_MAXSEG: 3310 if (optlen != TCPOLEN_MAXSEG) 3311 continue; 3312 if (!(flags & TO_SYN)) 3313 continue; 3314 to->to_flags |= TOF_MSS; 3315 bcopy((char *)cp + 2, 3316 (char *)&to->to_mss, sizeof(to->to_mss)); 3317 to->to_mss = ntohs(to->to_mss); 3318 break; 3319 case TCPOPT_WINDOW: 3320 if (optlen != TCPOLEN_WINDOW) 3321 continue; 3322 if (!(flags & TO_SYN)) 3323 continue; 3324 to->to_flags |= TOF_SCALE; 3325 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3326 break; 3327 case TCPOPT_TIMESTAMP: 3328 if (optlen != TCPOLEN_TIMESTAMP) 3329 continue; 3330 to->to_flags |= TOF_TS; 3331 bcopy((char *)cp + 2, 3332 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3333 to->to_tsval = ntohl(to->to_tsval); 3334 bcopy((char *)cp + 6, 3335 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3336 to->to_tsecr = ntohl(to->to_tsecr); 3337 break; 3338 case TCPOPT_SIGNATURE: 3339 /* 3340 * In order to reply to a host which has set the 3341 * TCP_SIGNATURE option in its initial SYN, we have 3342 * to record the fact that the option was observed 3343 * here for the syncache code to perform the correct 3344 * response. 3345 */ 3346 if (optlen != TCPOLEN_SIGNATURE) 3347 continue; 3348 to->to_flags |= TOF_SIGNATURE; 3349 to->to_signature = cp + 2; 3350 break; 3351 case TCPOPT_SACK_PERMITTED: 3352 if (optlen != TCPOLEN_SACK_PERMITTED) 3353 continue; 3354 if (!(flags & TO_SYN)) 3355 continue; 3356 if (!V_tcp_do_sack) 3357 continue; 3358 to->to_flags |= TOF_SACKPERM; 3359 break; 3360 case TCPOPT_SACK: 3361 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3362 continue; 3363 if (flags & TO_SYN) 3364 continue; 3365 to->to_flags |= TOF_SACK; 3366 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3367 to->to_sacks = cp + 2; 3368 TCPSTAT_INC(tcps_sack_rcv_blocks); 3369 break; 3370 case TCPOPT_FAST_OPEN: 3371 /* 3372 * Cookie length validation is performed by the 3373 * server side cookie checking code or the client 3374 * side cookie cache update code. 3375 */ 3376 if (!(flags & TO_SYN)) 3377 continue; 3378 if (!V_tcp_fastopen_client_enable && 3379 !V_tcp_fastopen_server_enable) 3380 continue; 3381 to->to_flags |= TOF_FASTOPEN; 3382 to->to_tfo_len = optlen - 2; 3383 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3384 break; 3385 default: 3386 continue; 3387 } 3388 } 3389 } 3390 3391 /* 3392 * Pull out of band byte out of a segment so 3393 * it doesn't appear in the user's data queue. 3394 * It is still reflected in the segment length for 3395 * sequencing purposes. 3396 */ 3397 void 3398 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3399 int off) 3400 { 3401 int cnt = off + th->th_urp - 1; 3402 3403 while (cnt >= 0) { 3404 if (m->m_len > cnt) { 3405 char *cp = mtod(m, caddr_t) + cnt; 3406 struct tcpcb *tp = sototcpcb(so); 3407 3408 INP_WLOCK_ASSERT(tp->t_inpcb); 3409 3410 tp->t_iobc = *cp; 3411 tp->t_oobflags |= TCPOOB_HAVEDATA; 3412 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3413 m->m_len--; 3414 if (m->m_flags & M_PKTHDR) 3415 m->m_pkthdr.len--; 3416 return; 3417 } 3418 cnt -= m->m_len; 3419 m = m->m_next; 3420 if (m == NULL) 3421 break; 3422 } 3423 panic("tcp_pulloutofband"); 3424 } 3425 3426 /* 3427 * Collect new round-trip time estimate 3428 * and update averages and current timeout. 3429 */ 3430 void 3431 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3432 { 3433 int delta; 3434 3435 INP_WLOCK_ASSERT(tp->t_inpcb); 3436 3437 TCPSTAT_INC(tcps_rttupdated); 3438 tp->t_rttupdated++; 3439 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3440 /* 3441 * srtt is stored as fixed point with 5 bits after the 3442 * binary point (i.e., scaled by 8). The following magic 3443 * is equivalent to the smoothing algorithm in rfc793 with 3444 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3445 * point). Adjust rtt to origin 0. 3446 */ 3447 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3448 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3449 3450 if ((tp->t_srtt += delta) <= 0) 3451 tp->t_srtt = 1; 3452 3453 /* 3454 * We accumulate a smoothed rtt variance (actually, a 3455 * smoothed mean difference), then set the retransmit 3456 * timer to smoothed rtt + 4 times the smoothed variance. 3457 * rttvar is stored as fixed point with 4 bits after the 3458 * binary point (scaled by 16). The following is 3459 * equivalent to rfc793 smoothing with an alpha of .75 3460 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3461 * rfc793's wired-in beta. 3462 */ 3463 if (delta < 0) 3464 delta = -delta; 3465 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3466 if ((tp->t_rttvar += delta) <= 0) 3467 tp->t_rttvar = 1; 3468 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3469 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3470 } else { 3471 /* 3472 * No rtt measurement yet - use the unsmoothed rtt. 3473 * Set the variance to half the rtt (so our first 3474 * retransmit happens at 3*rtt). 3475 */ 3476 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3477 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3478 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3479 } 3480 tp->t_rtttime = 0; 3481 tp->t_rxtshift = 0; 3482 3483 /* 3484 * the retransmit should happen at rtt + 4 * rttvar. 3485 * Because of the way we do the smoothing, srtt and rttvar 3486 * will each average +1/2 tick of bias. When we compute 3487 * the retransmit timer, we want 1/2 tick of rounding and 3488 * 1 extra tick because of +-1/2 tick uncertainty in the 3489 * firing of the timer. The bias will give us exactly the 3490 * 1.5 tick we need. But, because the bias is 3491 * statistical, we have to test that we don't drop below 3492 * the minimum feasible timer (which is 2 ticks). 3493 */ 3494 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3495 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3496 3497 /* 3498 * We received an ack for a packet that wasn't retransmitted; 3499 * it is probably safe to discard any error indications we've 3500 * received recently. This isn't quite right, but close enough 3501 * for now (a route might have failed after we sent a segment, 3502 * and the return path might not be symmetrical). 3503 */ 3504 tp->t_softerror = 0; 3505 } 3506 3507 /* 3508 * Determine a reasonable value for maxseg size. 3509 * If the route is known, check route for mtu. 3510 * If none, use an mss that can be handled on the outgoing interface 3511 * without forcing IP to fragment. If no route is found, route has no mtu, 3512 * or the destination isn't local, use a default, hopefully conservative 3513 * size (usually 512 or the default IP max size, but no more than the mtu 3514 * of the interface), as we can't discover anything about intervening 3515 * gateways or networks. We also initialize the congestion/slow start 3516 * window to be a single segment if the destination isn't local. 3517 * While looking at the routing entry, we also initialize other path-dependent 3518 * parameters from pre-set or cached values in the routing entry. 3519 * 3520 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3521 * IP options, e.g. IPSEC data, since length of this data may vary, and 3522 * thus it is calculated for every segment separately in tcp_output(). 3523 * 3524 * NOTE that this routine is only called when we process an incoming 3525 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3526 * settings are handled in tcp_mssopt(). 3527 */ 3528 void 3529 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3530 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3531 { 3532 int mss = 0; 3533 uint32_t maxmtu = 0; 3534 struct inpcb *inp = tp->t_inpcb; 3535 struct hc_metrics_lite metrics; 3536 #ifdef INET6 3537 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3538 size_t min_protoh = isipv6 ? 3539 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3540 sizeof (struct tcpiphdr); 3541 #else 3542 const size_t min_protoh = sizeof(struct tcpiphdr); 3543 #endif 3544 3545 INP_WLOCK_ASSERT(tp->t_inpcb); 3546 3547 if (mtuoffer != -1) { 3548 KASSERT(offer == -1, ("%s: conflict", __func__)); 3549 offer = mtuoffer - min_protoh; 3550 } 3551 3552 /* Initialize. */ 3553 #ifdef INET6 3554 if (isipv6) { 3555 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3556 tp->t_maxseg = V_tcp_v6mssdflt; 3557 } 3558 #endif 3559 #if defined(INET) && defined(INET6) 3560 else 3561 #endif 3562 #ifdef INET 3563 { 3564 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3565 tp->t_maxseg = V_tcp_mssdflt; 3566 } 3567 #endif 3568 3569 /* 3570 * No route to sender, stay with default mss and return. 3571 */ 3572 if (maxmtu == 0) { 3573 /* 3574 * In case we return early we need to initialize metrics 3575 * to a defined state as tcp_hc_get() would do for us 3576 * if there was no cache hit. 3577 */ 3578 if (metricptr != NULL) 3579 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3580 return; 3581 } 3582 3583 /* What have we got? */ 3584 switch (offer) { 3585 case 0: 3586 /* 3587 * Offer == 0 means that there was no MSS on the SYN 3588 * segment, in this case we use tcp_mssdflt as 3589 * already assigned to t_maxseg above. 3590 */ 3591 offer = tp->t_maxseg; 3592 break; 3593 3594 case -1: 3595 /* 3596 * Offer == -1 means that we didn't receive SYN yet. 3597 */ 3598 /* FALLTHROUGH */ 3599 3600 default: 3601 /* 3602 * Prevent DoS attack with too small MSS. Round up 3603 * to at least minmss. 3604 */ 3605 offer = max(offer, V_tcp_minmss); 3606 } 3607 3608 /* 3609 * rmx information is now retrieved from tcp_hostcache. 3610 */ 3611 tcp_hc_get(&inp->inp_inc, &metrics); 3612 if (metricptr != NULL) 3613 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3614 3615 /* 3616 * If there's a discovered mtu in tcp hostcache, use it. 3617 * Else, use the link mtu. 3618 */ 3619 if (metrics.rmx_mtu) 3620 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3621 else { 3622 #ifdef INET6 3623 if (isipv6) { 3624 mss = maxmtu - min_protoh; 3625 if (!V_path_mtu_discovery && 3626 !in6_localaddr(&inp->in6p_faddr)) 3627 mss = min(mss, V_tcp_v6mssdflt); 3628 } 3629 #endif 3630 #if defined(INET) && defined(INET6) 3631 else 3632 #endif 3633 #ifdef INET 3634 { 3635 mss = maxmtu - min_protoh; 3636 if (!V_path_mtu_discovery && 3637 !in_localaddr(inp->inp_faddr)) 3638 mss = min(mss, V_tcp_mssdflt); 3639 } 3640 #endif 3641 /* 3642 * XXX - The above conditional (mss = maxmtu - min_protoh) 3643 * probably violates the TCP spec. 3644 * The problem is that, since we don't know the 3645 * other end's MSS, we are supposed to use a conservative 3646 * default. But, if we do that, then MTU discovery will 3647 * never actually take place, because the conservative 3648 * default is much less than the MTUs typically seen 3649 * on the Internet today. For the moment, we'll sweep 3650 * this under the carpet. 3651 * 3652 * The conservative default might not actually be a problem 3653 * if the only case this occurs is when sending an initial 3654 * SYN with options and data to a host we've never talked 3655 * to before. Then, they will reply with an MSS value which 3656 * will get recorded and the new parameters should get 3657 * recomputed. For Further Study. 3658 */ 3659 } 3660 mss = min(mss, offer); 3661 3662 /* 3663 * Sanity check: make sure that maxseg will be large 3664 * enough to allow some data on segments even if the 3665 * all the option space is used (40bytes). Otherwise 3666 * funny things may happen in tcp_output. 3667 * 3668 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3669 */ 3670 mss = max(mss, 64); 3671 3672 tp->t_maxseg = mss; 3673 } 3674 3675 void 3676 tcp_mss(struct tcpcb *tp, int offer) 3677 { 3678 int mss; 3679 uint32_t bufsize; 3680 struct inpcb *inp; 3681 struct socket *so; 3682 struct hc_metrics_lite metrics; 3683 struct tcp_ifcap cap; 3684 3685 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3686 3687 bzero(&cap, sizeof(cap)); 3688 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3689 3690 mss = tp->t_maxseg; 3691 inp = tp->t_inpcb; 3692 3693 /* 3694 * If there's a pipesize, change the socket buffer to that size, 3695 * don't change if sb_hiwat is different than default (then it 3696 * has been changed on purpose with setsockopt). 3697 * Make the socket buffers an integral number of mss units; 3698 * if the mss is larger than the socket buffer, decrease the mss. 3699 */ 3700 so = inp->inp_socket; 3701 SOCKBUF_LOCK(&so->so_snd); 3702 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3703 bufsize = metrics.rmx_sendpipe; 3704 else 3705 bufsize = so->so_snd.sb_hiwat; 3706 if (bufsize < mss) 3707 mss = bufsize; 3708 else { 3709 bufsize = roundup(bufsize, mss); 3710 if (bufsize > sb_max) 3711 bufsize = sb_max; 3712 if (bufsize > so->so_snd.sb_hiwat) 3713 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3714 } 3715 SOCKBUF_UNLOCK(&so->so_snd); 3716 /* 3717 * Sanity check: make sure that maxseg will be large 3718 * enough to allow some data on segments even if the 3719 * all the option space is used (40bytes). Otherwise 3720 * funny things may happen in tcp_output. 3721 * 3722 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3723 */ 3724 tp->t_maxseg = max(mss, 64); 3725 3726 SOCKBUF_LOCK(&so->so_rcv); 3727 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3728 bufsize = metrics.rmx_recvpipe; 3729 else 3730 bufsize = so->so_rcv.sb_hiwat; 3731 if (bufsize > mss) { 3732 bufsize = roundup(bufsize, mss); 3733 if (bufsize > sb_max) 3734 bufsize = sb_max; 3735 if (bufsize > so->so_rcv.sb_hiwat) 3736 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3737 } 3738 SOCKBUF_UNLOCK(&so->so_rcv); 3739 3740 /* Check the interface for TSO capabilities. */ 3741 if (cap.ifcap & CSUM_TSO) { 3742 tp->t_flags |= TF_TSO; 3743 tp->t_tsomax = cap.tsomax; 3744 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3745 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3746 } 3747 } 3748 3749 /* 3750 * Determine the MSS option to send on an outgoing SYN. 3751 */ 3752 int 3753 tcp_mssopt(struct in_conninfo *inc) 3754 { 3755 int mss = 0; 3756 uint32_t thcmtu = 0; 3757 uint32_t maxmtu = 0; 3758 size_t min_protoh; 3759 3760 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3761 3762 #ifdef INET6 3763 if (inc->inc_flags & INC_ISIPV6) { 3764 mss = V_tcp_v6mssdflt; 3765 maxmtu = tcp_maxmtu6(inc, NULL); 3766 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3767 } 3768 #endif 3769 #if defined(INET) && defined(INET6) 3770 else 3771 #endif 3772 #ifdef INET 3773 { 3774 mss = V_tcp_mssdflt; 3775 maxmtu = tcp_maxmtu(inc, NULL); 3776 min_protoh = sizeof(struct tcpiphdr); 3777 } 3778 #endif 3779 #if defined(INET6) || defined(INET) 3780 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3781 #endif 3782 3783 if (maxmtu && thcmtu) 3784 mss = min(maxmtu, thcmtu) - min_protoh; 3785 else if (maxmtu || thcmtu) 3786 mss = max(maxmtu, thcmtu) - min_protoh; 3787 3788 return (mss); 3789 } 3790 3791 3792 /* 3793 * On a partial ack arrives, force the retransmission of the 3794 * next unacknowledged segment. Do not clear tp->t_dupacks. 3795 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3796 * be started again. 3797 */ 3798 void 3799 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3800 { 3801 tcp_seq onxt = tp->snd_nxt; 3802 uint32_t ocwnd = tp->snd_cwnd; 3803 u_int maxseg = tcp_maxseg(tp); 3804 3805 INP_WLOCK_ASSERT(tp->t_inpcb); 3806 3807 tcp_timer_activate(tp, TT_REXMT, 0); 3808 tp->t_rtttime = 0; 3809 tp->snd_nxt = th->th_ack; 3810 /* 3811 * Set snd_cwnd to one segment beyond acknowledged offset. 3812 * (tp->snd_una has not yet been updated when this function is called.) 3813 */ 3814 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3815 tp->t_flags |= TF_ACKNOW; 3816 (void) tp->t_fb->tfb_tcp_output(tp); 3817 tp->snd_cwnd = ocwnd; 3818 if (SEQ_GT(onxt, tp->snd_nxt)) 3819 tp->snd_nxt = onxt; 3820 /* 3821 * Partial window deflation. Relies on fact that tp->snd_una 3822 * not updated yet. 3823 */ 3824 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3825 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3826 else 3827 tp->snd_cwnd = 0; 3828 tp->snd_cwnd += maxseg; 3829 } 3830 3831 int 3832 tcp_compute_pipe(struct tcpcb *tp) 3833 { 3834 return (tp->snd_max - tp->snd_una + 3835 tp->sackhint.sack_bytes_rexmit - 3836 tp->sackhint.sacked_bytes); 3837 } 3838 3839 uint32_t 3840 tcp_compute_initwnd(uint32_t maxseg) 3841 { 3842 /* 3843 * Calculate the Initial Window, also used as Restart Window 3844 * 3845 * RFC5681 Section 3.1 specifies the default conservative values. 3846 * RFC3390 specifies slightly more aggressive values. 3847 * RFC6928 increases it to ten segments. 3848 * Support for user specified value for initial flight size. 3849 */ 3850 if (V_tcp_initcwnd_segments) 3851 return min(V_tcp_initcwnd_segments * maxseg, 3852 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 3853 else if (V_tcp_do_rfc3390) 3854 return min(4 * maxseg, max(2 * maxseg, 4380)); 3855 else { 3856 /* Per RFC5681 Section 3.1 */ 3857 if (maxseg > 2190) 3858 return (2 * maxseg); 3859 else if (maxseg > 1095) 3860 return (3 * maxseg); 3861 else 3862 return (4 * maxseg); 3863 } 3864 } 3865