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 uint8_t ipttl; 558 #endif 559 int tlen = 0, off; 560 int drop_hdrlen; 561 int thflags; 562 int rstreason = 0; /* For badport_bandlim accounting purposes */ 563 uint8_t iptos; 564 struct m_tag *fwd_tag = NULL; 565 struct epoch_tracker et; 566 #ifdef INET6 567 struct ip6_hdr *ip6 = NULL; 568 int isipv6; 569 #else 570 const void *ip6 = NULL; 571 #endif /* INET6 */ 572 struct tcpopt to; /* options in this segment */ 573 char *s = NULL; /* address and port logging */ 574 int ti_locked; 575 #ifdef TCPDEBUG 576 /* 577 * The size of tcp_saveipgen must be the size of the max ip header, 578 * now IPv6. 579 */ 580 u_char tcp_saveipgen[IP6_HDR_LEN]; 581 struct tcphdr tcp_savetcp; 582 short ostate = 0; 583 #endif 584 585 #ifdef INET6 586 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 587 #endif 588 589 off0 = *offp; 590 m = *mp; 591 *mp = NULL; 592 to.to_flags = 0; 593 TCPSTAT_INC(tcps_rcvtotal); 594 595 #ifdef INET6 596 if (isipv6) { 597 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 598 599 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 600 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 601 if (m == NULL) { 602 TCPSTAT_INC(tcps_rcvshort); 603 return (IPPROTO_DONE); 604 } 605 } 606 607 ip6 = mtod(m, struct ip6_hdr *); 608 th = (struct tcphdr *)((caddr_t)ip6 + off0); 609 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 610 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 611 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 612 th->th_sum = m->m_pkthdr.csum_data; 613 else 614 th->th_sum = in6_cksum_pseudo(ip6, tlen, 615 IPPROTO_TCP, m->m_pkthdr.csum_data); 616 th->th_sum ^= 0xffff; 617 } else 618 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 619 if (th->th_sum) { 620 TCPSTAT_INC(tcps_rcvbadsum); 621 goto drop; 622 } 623 624 /* 625 * Be proactive about unspecified IPv6 address in source. 626 * As we use all-zero to indicate unbounded/unconnected pcb, 627 * unspecified IPv6 address can be used to confuse us. 628 * 629 * Note that packets with unspecified IPv6 destination is 630 * already dropped in ip6_input. 631 */ 632 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 633 /* XXX stat */ 634 goto drop; 635 } 636 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 637 } 638 #endif 639 #if defined(INET) && defined(INET6) 640 else 641 #endif 642 #ifdef INET 643 { 644 /* 645 * Get IP and TCP header together in first mbuf. 646 * Note: IP leaves IP header in first mbuf. 647 */ 648 if (off0 > sizeof (struct ip)) { 649 ip_stripoptions(m); 650 off0 = sizeof(struct ip); 651 } 652 if (m->m_len < sizeof (struct tcpiphdr)) { 653 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 654 == NULL) { 655 TCPSTAT_INC(tcps_rcvshort); 656 return (IPPROTO_DONE); 657 } 658 } 659 ip = mtod(m, struct ip *); 660 th = (struct tcphdr *)((caddr_t)ip + off0); 661 tlen = ntohs(ip->ip_len) - off0; 662 663 iptos = ip->ip_tos; 664 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 665 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 666 th->th_sum = m->m_pkthdr.csum_data; 667 else 668 th->th_sum = in_pseudo(ip->ip_src.s_addr, 669 ip->ip_dst.s_addr, 670 htonl(m->m_pkthdr.csum_data + tlen + 671 IPPROTO_TCP)); 672 th->th_sum ^= 0xffff; 673 } else { 674 struct ipovly *ipov = (struct ipovly *)ip; 675 676 /* 677 * Checksum extended TCP header and data. 678 */ 679 len = off0 + tlen; 680 ipttl = ip->ip_ttl; 681 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 682 ipov->ih_len = htons(tlen); 683 th->th_sum = in_cksum(m, len); 684 /* Reset length for SDT probes. */ 685 ip->ip_len = htons(len); 686 /* Reset TOS bits */ 687 ip->ip_tos = iptos; 688 /* Re-initialization for later version check */ 689 ip->ip_ttl = ipttl; 690 ip->ip_v = IPVERSION; 691 ip->ip_hl = off0 >> 2; 692 } 693 694 if (th->th_sum) { 695 TCPSTAT_INC(tcps_rcvbadsum); 696 goto drop; 697 } 698 } 699 #endif /* INET */ 700 701 /* 702 * Check that TCP offset makes sense, 703 * pull out TCP options and adjust length. XXX 704 */ 705 off = th->th_off << 2; 706 if (off < sizeof (struct tcphdr) || off > tlen) { 707 TCPSTAT_INC(tcps_rcvbadoff); 708 goto drop; 709 } 710 tlen -= off; /* tlen is used instead of ti->ti_len */ 711 if (off > sizeof (struct tcphdr)) { 712 #ifdef INET6 713 if (isipv6) { 714 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 715 ip6 = mtod(m, struct ip6_hdr *); 716 th = (struct tcphdr *)((caddr_t)ip6 + off0); 717 } 718 #endif 719 #if defined(INET) && defined(INET6) 720 else 721 #endif 722 #ifdef INET 723 { 724 if (m->m_len < sizeof(struct ip) + off) { 725 if ((m = m_pullup(m, sizeof (struct ip) + off)) 726 == NULL) { 727 TCPSTAT_INC(tcps_rcvshort); 728 return (IPPROTO_DONE); 729 } 730 ip = mtod(m, struct ip *); 731 th = (struct tcphdr *)((caddr_t)ip + off0); 732 } 733 } 734 #endif 735 optlen = off - sizeof (struct tcphdr); 736 optp = (u_char *)(th + 1); 737 } 738 thflags = th->th_flags; 739 740 /* 741 * Convert TCP protocol specific fields to host format. 742 */ 743 tcp_fields_to_host(th); 744 745 /* 746 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 747 */ 748 drop_hdrlen = off0 + off; 749 750 /* 751 * Locate pcb for segment; if we're likely to add or remove a 752 * connection then first acquire pcbinfo lock. There are three cases 753 * where we might discover later we need a write lock despite the 754 * flags: ACKs moving a connection out of the syncache, ACKs for a 755 * connection in TIMEWAIT and SYNs not targeting a listening socket. 756 */ 757 if ((thflags & (TH_FIN | TH_RST)) != 0) { 758 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 759 ti_locked = TI_RLOCKED; 760 } else 761 ti_locked = TI_UNLOCKED; 762 763 /* 764 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 765 */ 766 if ( 767 #ifdef INET6 768 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 769 #ifdef INET 770 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 771 #endif 772 #endif 773 #if defined(INET) && !defined(INET6) 774 (m->m_flags & M_IP_NEXTHOP) 775 #endif 776 ) 777 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 778 779 findpcb: 780 #ifdef INVARIANTS 781 if (ti_locked == TI_RLOCKED) { 782 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 783 } else { 784 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 785 } 786 #endif 787 #ifdef INET6 788 if (isipv6 && fwd_tag != NULL) { 789 struct sockaddr_in6 *next_hop6; 790 791 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 792 /* 793 * Transparently forwarded. Pretend to be the destination. 794 * Already got one like this? 795 */ 796 inp = in6_pcblookup_mbuf(&V_tcbinfo, 797 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 798 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 799 if (!inp) { 800 /* 801 * It's new. Try to find the ambushing socket. 802 * Because we've rewritten the destination address, 803 * any hardware-generated hash is ignored. 804 */ 805 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 806 th->th_sport, &next_hop6->sin6_addr, 807 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 808 th->th_dport, INPLOOKUP_WILDCARD | 809 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 810 } 811 } else if (isipv6) { 812 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 813 th->th_sport, &ip6->ip6_dst, th->th_dport, 814 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 815 m->m_pkthdr.rcvif, m); 816 } 817 #endif /* INET6 */ 818 #if defined(INET6) && defined(INET) 819 else 820 #endif 821 #ifdef INET 822 if (fwd_tag != NULL) { 823 struct sockaddr_in *next_hop; 824 825 next_hop = (struct sockaddr_in *)(fwd_tag+1); 826 /* 827 * Transparently forwarded. Pretend to be the destination. 828 * already got one like this? 829 */ 830 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 831 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 832 m->m_pkthdr.rcvif, m); 833 if (!inp) { 834 /* 835 * It's new. Try to find the ambushing socket. 836 * Because we've rewritten the destination address, 837 * any hardware-generated hash is ignored. 838 */ 839 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 840 th->th_sport, next_hop->sin_addr, 841 next_hop->sin_port ? ntohs(next_hop->sin_port) : 842 th->th_dport, INPLOOKUP_WILDCARD | 843 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 844 } 845 } else 846 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 847 th->th_sport, ip->ip_dst, th->th_dport, 848 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 849 m->m_pkthdr.rcvif, m); 850 #endif /* INET */ 851 852 /* 853 * If the INPCB does not exist then all data in the incoming 854 * segment is discarded and an appropriate RST is sent back. 855 * XXX MRT Send RST using which routing table? 856 */ 857 if (inp == NULL) { 858 /* 859 * Log communication attempts to ports that are not 860 * in use. 861 */ 862 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 863 tcp_log_in_vain == 2) { 864 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 865 log(LOG_INFO, "%s; %s: Connection attempt " 866 "to closed port\n", s, __func__); 867 } 868 /* 869 * When blackholing do not respond with a RST but 870 * completely ignore the segment and drop it. 871 */ 872 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 873 V_blackhole == 2) 874 goto dropunlock; 875 876 rstreason = BANDLIM_RST_CLOSEDPORT; 877 goto dropwithreset; 878 } 879 INP_WLOCK_ASSERT(inp); 880 /* 881 * While waiting for inp lock during the lookup, another thread 882 * can have dropped the inpcb, in which case we need to loop back 883 * and try to find a new inpcb to deliver to. 884 */ 885 if (inp->inp_flags & INP_DROPPED) { 886 INP_WUNLOCK(inp); 887 inp = NULL; 888 goto findpcb; 889 } 890 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 891 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 892 ((inp->inp_socket == NULL) || 893 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 894 inp->inp_flowid = m->m_pkthdr.flowid; 895 inp->inp_flowtype = M_HASHTYPE_GET(m); 896 } 897 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 898 #ifdef INET6 899 if (isipv6 && IPSEC_ENABLED(ipv6) && 900 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 901 goto dropunlock; 902 } 903 #ifdef INET 904 else 905 #endif 906 #endif /* INET6 */ 907 #ifdef INET 908 if (IPSEC_ENABLED(ipv4) && 909 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 910 goto dropunlock; 911 } 912 #endif /* INET */ 913 #endif /* IPSEC */ 914 915 /* 916 * Check the minimum TTL for socket. 917 */ 918 if (inp->inp_ip_minttl != 0) { 919 #ifdef INET6 920 if (isipv6) { 921 if (inp->inp_ip_minttl > ip6->ip6_hlim) 922 goto dropunlock; 923 } else 924 #endif 925 if (inp->inp_ip_minttl > ip->ip_ttl) 926 goto dropunlock; 927 } 928 929 /* 930 * A previous connection in TIMEWAIT state is supposed to catch stray 931 * or duplicate segments arriving late. If this segment was a 932 * legitimate new connection attempt, the old INPCB gets removed and 933 * we can try again to find a listening socket. 934 * 935 * At this point, due to earlier optimism, we may hold only an inpcb 936 * lock, and not the inpcbinfo write lock. If so, we need to try to 937 * acquire it, or if that fails, acquire a reference on the inpcb, 938 * drop all locks, acquire a global write lock, and then re-acquire 939 * the inpcb lock. We may at that point discover that another thread 940 * has tried to free the inpcb, in which case we need to loop back 941 * and try to find a new inpcb to deliver to. 942 * 943 * XXXRW: It may be time to rethink timewait locking. 944 */ 945 if (inp->inp_flags & INP_TIMEWAIT) { 946 if (ti_locked == TI_UNLOCKED) { 947 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 948 ti_locked = TI_RLOCKED; 949 } 950 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 951 952 if (thflags & TH_SYN) 953 tcp_dooptions(&to, optp, optlen, TO_SYN); 954 /* 955 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 956 */ 957 if (tcp_twcheck(inp, &to, th, m, tlen)) 958 goto findpcb; 959 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 960 return (IPPROTO_DONE); 961 } 962 /* 963 * The TCPCB may no longer exist if the connection is winding 964 * down or it is in the CLOSED state. Either way we drop the 965 * segment and send an appropriate response. 966 */ 967 tp = intotcpcb(inp); 968 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 969 rstreason = BANDLIM_RST_CLOSEDPORT; 970 goto dropwithreset; 971 } 972 973 #ifdef TCP_OFFLOAD 974 if (tp->t_flags & TF_TOE) { 975 tcp_offload_input(tp, m); 976 m = NULL; /* consumed by the TOE driver */ 977 goto dropunlock; 978 } 979 #endif 980 981 /* 982 * We've identified a valid inpcb, but it could be that we need an 983 * inpcbinfo write lock but don't hold it. In this case, attempt to 984 * acquire using the same strategy as the TIMEWAIT case above. If we 985 * relock, we have to jump back to 'relocked' as the connection might 986 * now be in TIMEWAIT. 987 */ 988 #ifdef INVARIANTS 989 if ((thflags & (TH_FIN | TH_RST)) != 0) 990 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 991 #endif 992 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 993 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 994 !IS_FASTOPEN(tp->t_flags)))) { 995 if (ti_locked == TI_UNLOCKED) { 996 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 997 ti_locked = TI_RLOCKED; 998 } 999 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1000 } 1001 1002 #ifdef MAC 1003 INP_WLOCK_ASSERT(inp); 1004 if (mac_inpcb_check_deliver(inp, m)) 1005 goto dropunlock; 1006 #endif 1007 so = inp->inp_socket; 1008 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1009 #ifdef TCPDEBUG 1010 if (so->so_options & SO_DEBUG) { 1011 ostate = tp->t_state; 1012 #ifdef INET6 1013 if (isipv6) { 1014 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1015 } else 1016 #endif 1017 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1018 tcp_savetcp = *th; 1019 } 1020 #endif /* TCPDEBUG */ 1021 /* 1022 * When the socket is accepting connections (the INPCB is in LISTEN 1023 * state) we look into the SYN cache if this is a new connection 1024 * attempt or the completion of a previous one. 1025 */ 1026 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1027 ("%s: so accepting but tp %p not listening", __func__, tp)); 1028 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1029 struct in_conninfo inc; 1030 1031 bzero(&inc, sizeof(inc)); 1032 #ifdef INET6 1033 if (isipv6) { 1034 inc.inc_flags |= INC_ISIPV6; 1035 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1036 inc.inc_flags |= INC_IPV6MINMTU; 1037 inc.inc6_faddr = ip6->ip6_src; 1038 inc.inc6_laddr = ip6->ip6_dst; 1039 } else 1040 #endif 1041 { 1042 inc.inc_faddr = ip->ip_src; 1043 inc.inc_laddr = ip->ip_dst; 1044 } 1045 inc.inc_fport = th->th_sport; 1046 inc.inc_lport = th->th_dport; 1047 inc.inc_fibnum = so->so_fibnum; 1048 1049 /* 1050 * Check for an existing connection attempt in syncache if 1051 * the flag is only ACK. A successful lookup creates a new 1052 * socket appended to the listen queue in SYN_RECEIVED state. 1053 */ 1054 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1055 1056 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1057 /* 1058 * Parse the TCP options here because 1059 * syncookies need access to the reflected 1060 * timestamp. 1061 */ 1062 tcp_dooptions(&to, optp, optlen, 0); 1063 /* 1064 * NB: syncache_expand() doesn't unlock 1065 * inp and tcpinfo locks. 1066 */ 1067 rstreason = syncache_expand(&inc, &to, th, &so, m); 1068 if (rstreason < 0) { 1069 /* 1070 * A failing TCP MD5 signature comparison 1071 * must result in the segment being dropped 1072 * and must not produce any response back 1073 * to the sender. 1074 */ 1075 goto dropunlock; 1076 } else if (rstreason == 0) { 1077 /* 1078 * No syncache entry or ACK was not 1079 * for our SYN/ACK. Send a RST. 1080 * NB: syncache did its own logging 1081 * of the failure cause. 1082 */ 1083 rstreason = BANDLIM_RST_OPENPORT; 1084 goto dropwithreset; 1085 } 1086 tfo_socket_result: 1087 if (so == NULL) { 1088 /* 1089 * We completed the 3-way handshake 1090 * but could not allocate a socket 1091 * either due to memory shortage, 1092 * listen queue length limits or 1093 * global socket limits. Send RST 1094 * or wait and have the remote end 1095 * retransmit the ACK for another 1096 * try. 1097 */ 1098 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1099 log(LOG_DEBUG, "%s; %s: Listen socket: " 1100 "Socket allocation failed due to " 1101 "limits or memory shortage, %s\n", 1102 s, __func__, 1103 V_tcp_sc_rst_sock_fail ? 1104 "sending RST" : "try again"); 1105 if (V_tcp_sc_rst_sock_fail) { 1106 rstreason = BANDLIM_UNLIMITED; 1107 goto dropwithreset; 1108 } else 1109 goto dropunlock; 1110 } 1111 /* 1112 * Socket is created in state SYN_RECEIVED. 1113 * Unlock the listen socket, lock the newly 1114 * created socket and update the tp variable. 1115 */ 1116 INP_WUNLOCK(inp); /* listen socket */ 1117 inp = sotoinpcb(so); 1118 /* 1119 * New connection inpcb is already locked by 1120 * syncache_expand(). 1121 */ 1122 INP_WLOCK_ASSERT(inp); 1123 tp = intotcpcb(inp); 1124 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1125 ("%s: ", __func__)); 1126 /* 1127 * Process the segment and the data it 1128 * contains. tcp_do_segment() consumes 1129 * the mbuf chain and unlocks the inpcb. 1130 */ 1131 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1132 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1133 iptos); 1134 if (ti_locked == TI_RLOCKED) 1135 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1136 return (IPPROTO_DONE); 1137 } 1138 /* 1139 * Segment flag validation for new connection attempts: 1140 * 1141 * Our (SYN|ACK) response was rejected. 1142 * Check with syncache and remove entry to prevent 1143 * retransmits. 1144 * 1145 * NB: syncache_chkrst does its own logging of failure 1146 * causes. 1147 */ 1148 if (thflags & TH_RST) { 1149 syncache_chkrst(&inc, th, m); 1150 goto dropunlock; 1151 } 1152 /* 1153 * We can't do anything without SYN. 1154 */ 1155 if ((thflags & TH_SYN) == 0) { 1156 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1157 log(LOG_DEBUG, "%s; %s: Listen socket: " 1158 "SYN is missing, segment ignored\n", 1159 s, __func__); 1160 TCPSTAT_INC(tcps_badsyn); 1161 goto dropunlock; 1162 } 1163 /* 1164 * (SYN|ACK) is bogus on a listen socket. 1165 */ 1166 if (thflags & TH_ACK) { 1167 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1168 log(LOG_DEBUG, "%s; %s: Listen socket: " 1169 "SYN|ACK invalid, segment rejected\n", 1170 s, __func__); 1171 syncache_badack(&inc); /* XXX: Not needed! */ 1172 TCPSTAT_INC(tcps_badsyn); 1173 rstreason = BANDLIM_RST_OPENPORT; 1174 goto dropwithreset; 1175 } 1176 /* 1177 * If the drop_synfin option is enabled, drop all 1178 * segments with both the SYN and FIN bits set. 1179 * This prevents e.g. nmap from identifying the 1180 * TCP/IP stack. 1181 * XXX: Poor reasoning. nmap has other methods 1182 * and is constantly refining its stack detection 1183 * strategies. 1184 * XXX: This is a violation of the TCP specification 1185 * and was used by RFC1644. 1186 */ 1187 if ((thflags & TH_FIN) && V_drop_synfin) { 1188 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1189 log(LOG_DEBUG, "%s; %s: Listen socket: " 1190 "SYN|FIN segment ignored (based on " 1191 "sysctl setting)\n", s, __func__); 1192 TCPSTAT_INC(tcps_badsyn); 1193 goto dropunlock; 1194 } 1195 /* 1196 * Segment's flags are (SYN) or (SYN|FIN). 1197 * 1198 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1199 * as they do not affect the state of the TCP FSM. 1200 * The data pointed to by TH_URG and th_urp is ignored. 1201 */ 1202 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1203 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1204 KASSERT(thflags & (TH_SYN), 1205 ("%s: Listen socket: TH_SYN not set", __func__)); 1206 #ifdef INET6 1207 /* 1208 * If deprecated address is forbidden, 1209 * we do not accept SYN to deprecated interface 1210 * address to prevent any new inbound connection from 1211 * getting established. 1212 * When we do not accept SYN, we send a TCP RST, 1213 * with deprecated source address (instead of dropping 1214 * it). We compromise it as it is much better for peer 1215 * to send a RST, and RST will be the final packet 1216 * for the exchange. 1217 * 1218 * If we do not forbid deprecated addresses, we accept 1219 * the SYN packet. RFC2462 does not suggest dropping 1220 * SYN in this case. 1221 * If we decipher RFC2462 5.5.4, it says like this: 1222 * 1. use of deprecated addr with existing 1223 * communication is okay - "SHOULD continue to be 1224 * used" 1225 * 2. use of it with new communication: 1226 * (2a) "SHOULD NOT be used if alternate address 1227 * with sufficient scope is available" 1228 * (2b) nothing mentioned otherwise. 1229 * Here we fall into (2b) case as we have no choice in 1230 * our source address selection - we must obey the peer. 1231 * 1232 * The wording in RFC2462 is confusing, and there are 1233 * multiple description text for deprecated address 1234 * handling - worse, they are not exactly the same. 1235 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1236 */ 1237 if (isipv6 && !V_ip6_use_deprecated) { 1238 struct in6_ifaddr *ia6; 1239 1240 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1241 if (ia6 != NULL && 1242 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1243 ifa_free(&ia6->ia_ifa); 1244 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1245 log(LOG_DEBUG, "%s; %s: Listen socket: " 1246 "Connection attempt to deprecated " 1247 "IPv6 address rejected\n", 1248 s, __func__); 1249 rstreason = BANDLIM_RST_OPENPORT; 1250 goto dropwithreset; 1251 } 1252 if (ia6) 1253 ifa_free(&ia6->ia_ifa); 1254 } 1255 #endif /* INET6 */ 1256 /* 1257 * Basic sanity checks on incoming SYN requests: 1258 * Don't respond if the destination is a link layer 1259 * broadcast according to RFC1122 4.2.3.10, p. 104. 1260 * If it is from this socket it must be forged. 1261 * Don't respond if the source or destination is a 1262 * global or subnet broad- or multicast address. 1263 * Note that it is quite possible to receive unicast 1264 * link-layer packets with a broadcast IP address. Use 1265 * in_broadcast() to find them. 1266 */ 1267 if (m->m_flags & (M_BCAST|M_MCAST)) { 1268 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1269 log(LOG_DEBUG, "%s; %s: Listen socket: " 1270 "Connection attempt from broad- or multicast " 1271 "link layer address ignored\n", s, __func__); 1272 goto dropunlock; 1273 } 1274 #ifdef INET6 1275 if (isipv6) { 1276 if (th->th_dport == th->th_sport && 1277 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1278 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1279 log(LOG_DEBUG, "%s; %s: Listen socket: " 1280 "Connection attempt to/from self " 1281 "ignored\n", s, __func__); 1282 goto dropunlock; 1283 } 1284 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1285 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1286 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1287 log(LOG_DEBUG, "%s; %s: Listen socket: " 1288 "Connection attempt from/to multicast " 1289 "address ignored\n", s, __func__); 1290 goto dropunlock; 1291 } 1292 } 1293 #endif 1294 #if defined(INET) && defined(INET6) 1295 else 1296 #endif 1297 #ifdef INET 1298 { 1299 if (th->th_dport == th->th_sport && 1300 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1301 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1302 log(LOG_DEBUG, "%s; %s: Listen socket: " 1303 "Connection attempt from/to self " 1304 "ignored\n", s, __func__); 1305 goto dropunlock; 1306 } 1307 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1308 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1309 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1310 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1311 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1312 log(LOG_DEBUG, "%s; %s: Listen socket: " 1313 "Connection attempt from/to broad- " 1314 "or multicast address ignored\n", 1315 s, __func__); 1316 goto dropunlock; 1317 } 1318 } 1319 #endif 1320 /* 1321 * SYN appears to be valid. Create compressed TCP state 1322 * for syncache. 1323 */ 1324 #ifdef TCPDEBUG 1325 if (so->so_options & SO_DEBUG) 1326 tcp_trace(TA_INPUT, ostate, tp, 1327 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1328 #endif 1329 TCP_PROBE3(debug__input, tp, th, m); 1330 tcp_dooptions(&to, optp, optlen, TO_SYN); 1331 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1332 goto tfo_socket_result; 1333 1334 /* 1335 * Entry added to syncache and mbuf consumed. 1336 * Only the listen socket is unlocked by syncache_add(). 1337 */ 1338 if (ti_locked == TI_RLOCKED) { 1339 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1340 ti_locked = TI_UNLOCKED; 1341 } 1342 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1343 return (IPPROTO_DONE); 1344 } else if (tp->t_state == TCPS_LISTEN) { 1345 /* 1346 * When a listen socket is torn down the SO_ACCEPTCONN 1347 * flag is removed first while connections are drained 1348 * from the accept queue in a unlock/lock cycle of the 1349 * ACCEPT_LOCK, opening a race condition allowing a SYN 1350 * attempt go through unhandled. 1351 */ 1352 goto dropunlock; 1353 } 1354 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1355 if (tp->t_flags & TF_SIGNATURE) { 1356 tcp_dooptions(&to, optp, optlen, thflags); 1357 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1358 TCPSTAT_INC(tcps_sig_err_nosigopt); 1359 goto dropunlock; 1360 } 1361 if (!TCPMD5_ENABLED() || 1362 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1363 goto dropunlock; 1364 } 1365 #endif 1366 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1367 1368 /* 1369 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1370 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1371 * the inpcb, and unlocks pcbinfo. 1372 */ 1373 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1374 if (ti_locked == TI_RLOCKED) 1375 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1376 return (IPPROTO_DONE); 1377 1378 dropwithreset: 1379 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1380 1381 if (ti_locked == TI_RLOCKED) { 1382 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1383 ti_locked = TI_UNLOCKED; 1384 } 1385 #ifdef INVARIANTS 1386 else { 1387 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1388 "ti_locked: %d", __func__, ti_locked)); 1389 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1390 } 1391 #endif 1392 1393 if (inp != NULL) { 1394 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1395 INP_WUNLOCK(inp); 1396 } else 1397 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1398 m = NULL; /* mbuf chain got consumed. */ 1399 goto drop; 1400 1401 dropunlock: 1402 if (m != NULL) 1403 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1404 1405 if (ti_locked == TI_RLOCKED) { 1406 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1407 ti_locked = TI_UNLOCKED; 1408 } 1409 #ifdef INVARIANTS 1410 else { 1411 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1412 "ti_locked: %d", __func__, ti_locked)); 1413 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1414 } 1415 #endif 1416 1417 if (inp != NULL) 1418 INP_WUNLOCK(inp); 1419 1420 drop: 1421 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1422 if (s != NULL) 1423 free(s, M_TCPLOG); 1424 if (m != NULL) 1425 m_freem(m); 1426 return (IPPROTO_DONE); 1427 } 1428 1429 /* 1430 * Automatic sizing of receive socket buffer. Often the send 1431 * buffer size is not optimally adjusted to the actual network 1432 * conditions at hand (delay bandwidth product). Setting the 1433 * buffer size too small limits throughput on links with high 1434 * bandwidth and high delay (eg. trans-continental/oceanic links). 1435 * 1436 * On the receive side the socket buffer memory is only rarely 1437 * used to any significant extent. This allows us to be much 1438 * more aggressive in scaling the receive socket buffer. For 1439 * the case that the buffer space is actually used to a large 1440 * extent and we run out of kernel memory we can simply drop 1441 * the new segments; TCP on the sender will just retransmit it 1442 * later. Setting the buffer size too big may only consume too 1443 * much kernel memory if the application doesn't read() from 1444 * the socket or packet loss or reordering makes use of the 1445 * reassembly queue. 1446 * 1447 * The criteria to step up the receive buffer one notch are: 1448 * 1. Application has not set receive buffer size with 1449 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1450 * 2. the number of bytes received during 1/2 of an sRTT 1451 * is at least 3/8 of the current socket buffer size. 1452 * 3. receive buffer size has not hit maximal automatic size; 1453 * 1454 * If all of the criteria are met we increaset the socket buffer 1455 * by a 1/2 (bounded by the max). This allows us to keep ahead 1456 * of slow-start but also makes it so our peer never gets limited 1457 * by our rwnd which we then open up causing a burst. 1458 * 1459 * This algorithm does two steps per RTT at most and only if 1460 * we receive a bulk stream w/o packet losses or reorderings. 1461 * Shrinking the buffer during idle times is not necessary as 1462 * it doesn't consume any memory when idle. 1463 * 1464 * TODO: Only step up if the application is actually serving 1465 * the buffer to better manage the socket buffer resources. 1466 */ 1467 int 1468 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1469 struct tcpcb *tp, int tlen) 1470 { 1471 int newsize = 0; 1472 1473 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1474 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1475 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1476 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { 1477 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && 1478 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1479 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); 1480 } 1481 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1482 1483 /* Start over with next RTT. */ 1484 tp->rfbuf_ts = 0; 1485 tp->rfbuf_cnt = 0; 1486 } else { 1487 tp->rfbuf_cnt += tlen; /* add up */ 1488 } 1489 return (newsize); 1490 } 1491 1492 void 1493 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1494 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1495 { 1496 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1497 int rstreason, todrop, win; 1498 uint32_t tiwin; 1499 uint16_t nsegs; 1500 char *s; 1501 struct in_conninfo *inc; 1502 struct mbuf *mfree; 1503 struct tcpopt to; 1504 int tfo_syn; 1505 1506 #ifdef TCPDEBUG 1507 /* 1508 * The size of tcp_saveipgen must be the size of the max ip header, 1509 * now IPv6. 1510 */ 1511 u_char tcp_saveipgen[IP6_HDR_LEN]; 1512 struct tcphdr tcp_savetcp; 1513 short ostate = 0; 1514 #endif 1515 thflags = th->th_flags; 1516 inc = &tp->t_inpcb->inp_inc; 1517 tp->sackhint.last_sack_ack = 0; 1518 sack_changed = 0; 1519 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1520 /* 1521 * If this is either a state-changing packet or current state isn't 1522 * established, we require a write lock on tcbinfo. Otherwise, we 1523 * allow the tcbinfo to be in either alocked or unlocked, as the 1524 * caller may have unnecessarily acquired a write lock due to a race. 1525 */ 1526 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1527 tp->t_state != TCPS_ESTABLISHED) { 1528 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1529 } 1530 INP_WLOCK_ASSERT(tp->t_inpcb); 1531 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1532 __func__)); 1533 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1534 __func__)); 1535 1536 #ifdef TCPPCAP 1537 /* Save segment, if requested. */ 1538 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1539 #endif 1540 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1541 tlen, NULL, true); 1542 1543 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1544 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1545 log(LOG_DEBUG, "%s; %s: " 1546 "SYN|FIN segment ignored (based on " 1547 "sysctl setting)\n", s, __func__); 1548 free(s, M_TCPLOG); 1549 } 1550 goto drop; 1551 } 1552 1553 /* 1554 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1555 * check SEQ.ACK first. 1556 */ 1557 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1558 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1559 rstreason = BANDLIM_UNLIMITED; 1560 goto dropwithreset; 1561 } 1562 1563 /* 1564 * Segment received on connection. 1565 * Reset idle time and keep-alive timer. 1566 * XXX: This should be done after segment 1567 * validation to ignore broken/spoofed segs. 1568 */ 1569 tp->t_rcvtime = ticks; 1570 1571 /* 1572 * Scale up the window into a 32-bit value. 1573 * For the SYN_SENT state the scale is zero. 1574 */ 1575 tiwin = th->th_win << tp->snd_scale; 1576 1577 /* 1578 * TCP ECN processing. 1579 */ 1580 if (tp->t_flags & TF_ECN_PERMIT) { 1581 if (thflags & TH_CWR) 1582 tp->t_flags &= ~TF_ECN_SND_ECE; 1583 switch (iptos & IPTOS_ECN_MASK) { 1584 case IPTOS_ECN_CE: 1585 tp->t_flags |= TF_ECN_SND_ECE; 1586 TCPSTAT_INC(tcps_ecn_ce); 1587 break; 1588 case IPTOS_ECN_ECT0: 1589 TCPSTAT_INC(tcps_ecn_ect0); 1590 break; 1591 case IPTOS_ECN_ECT1: 1592 TCPSTAT_INC(tcps_ecn_ect1); 1593 break; 1594 } 1595 1596 /* Process a packet differently from RFC3168. */ 1597 cc_ecnpkt_handler(tp, th, iptos); 1598 1599 /* Congestion experienced. */ 1600 if (thflags & TH_ECE) { 1601 cc_cong_signal(tp, th, CC_ECN); 1602 } 1603 } 1604 1605 /* 1606 * Parse options on any incoming segment. 1607 */ 1608 tcp_dooptions(&to, (u_char *)(th + 1), 1609 (th->th_off << 2) - sizeof(struct tcphdr), 1610 (thflags & TH_SYN) ? TO_SYN : 0); 1611 1612 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1613 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1614 (to.to_flags & TOF_SIGNATURE) == 0) { 1615 TCPSTAT_INC(tcps_sig_err_sigopt); 1616 /* XXX: should drop? */ 1617 } 1618 #endif 1619 /* 1620 * If echoed timestamp is later than the current time, 1621 * fall back to non RFC1323 RTT calculation. Normalize 1622 * timestamp if syncookies were used when this connection 1623 * was established. 1624 */ 1625 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1626 to.to_tsecr -= tp->ts_offset; 1627 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1628 to.to_tsecr = 0; 1629 else if (tp->t_flags & TF_PREVVALID && 1630 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1631 cc_cong_signal(tp, th, CC_RTO_ERR); 1632 } 1633 /* 1634 * Process options only when we get SYN/ACK back. The SYN case 1635 * for incoming connections is handled in tcp_syncache. 1636 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1637 * or <SYN,ACK>) segment itself is never scaled. 1638 * XXX this is traditional behavior, may need to be cleaned up. 1639 */ 1640 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1641 if ((to.to_flags & TOF_SCALE) && 1642 (tp->t_flags & TF_REQ_SCALE)) { 1643 tp->t_flags |= TF_RCVD_SCALE; 1644 tp->snd_scale = to.to_wscale; 1645 } 1646 /* 1647 * Initial send window. It will be updated with 1648 * the next incoming segment to the scaled value. 1649 */ 1650 tp->snd_wnd = th->th_win; 1651 if (to.to_flags & TOF_TS) { 1652 tp->t_flags |= TF_RCVD_TSTMP; 1653 tp->ts_recent = to.to_tsval; 1654 tp->ts_recent_age = tcp_ts_getticks(); 1655 } 1656 if (to.to_flags & TOF_MSS) 1657 tcp_mss(tp, to.to_mss); 1658 if ((tp->t_flags & TF_SACK_PERMIT) && 1659 (to.to_flags & TOF_SACKPERM) == 0) 1660 tp->t_flags &= ~TF_SACK_PERMIT; 1661 if (IS_FASTOPEN(tp->t_flags)) { 1662 if (to.to_flags & TOF_FASTOPEN) { 1663 uint16_t mss; 1664 1665 if (to.to_flags & TOF_MSS) 1666 mss = to.to_mss; 1667 else 1668 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1669 mss = TCP6_MSS; 1670 else 1671 mss = TCP_MSS; 1672 tcp_fastopen_update_cache(tp, mss, 1673 to.to_tfo_len, to.to_tfo_cookie); 1674 } else 1675 tcp_fastopen_disable_path(tp); 1676 } 1677 } 1678 1679 /* 1680 * If timestamps were negotiated during SYN/ACK they should 1681 * appear on every segment during this session and vice versa. 1682 */ 1683 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1684 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1685 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1686 "no action\n", s, __func__); 1687 free(s, M_TCPLOG); 1688 } 1689 } 1690 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1691 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1692 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1693 "no action\n", s, __func__); 1694 free(s, M_TCPLOG); 1695 } 1696 } 1697 1698 /* 1699 * Header prediction: check for the two common cases 1700 * of a uni-directional data xfer. If the packet has 1701 * no control flags, is in-sequence, the window didn't 1702 * change and we're not retransmitting, it's a 1703 * candidate. If the length is zero and the ack moved 1704 * forward, we're the sender side of the xfer. Just 1705 * free the data acked & wake any higher level process 1706 * that was blocked waiting for space. If the length 1707 * is non-zero and the ack didn't move, we're the 1708 * receiver side. If we're getting packets in-order 1709 * (the reassembly queue is empty), add the data to 1710 * the socket buffer and note that we need a delayed ack. 1711 * Make sure that the hidden state-flags are also off. 1712 * Since we check for TCPS_ESTABLISHED first, it can only 1713 * be TH_NEEDSYN. 1714 */ 1715 if (tp->t_state == TCPS_ESTABLISHED && 1716 th->th_seq == tp->rcv_nxt && 1717 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1718 tp->snd_nxt == tp->snd_max && 1719 tiwin && tiwin == tp->snd_wnd && 1720 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1721 SEGQ_EMPTY(tp) && 1722 ((to.to_flags & TOF_TS) == 0 || 1723 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1724 1725 /* 1726 * If last ACK falls within this segment's sequence numbers, 1727 * record the timestamp. 1728 * NOTE that the test is modified according to the latest 1729 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1730 */ 1731 if ((to.to_flags & TOF_TS) != 0 && 1732 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1733 tp->ts_recent_age = tcp_ts_getticks(); 1734 tp->ts_recent = to.to_tsval; 1735 } 1736 1737 if (tlen == 0) { 1738 if (SEQ_GT(th->th_ack, tp->snd_una) && 1739 SEQ_LEQ(th->th_ack, tp->snd_max) && 1740 !IN_RECOVERY(tp->t_flags) && 1741 (to.to_flags & TOF_SACK) == 0 && 1742 TAILQ_EMPTY(&tp->snd_holes)) { 1743 /* 1744 * This is a pure ack for outstanding data. 1745 */ 1746 TCPSTAT_INC(tcps_predack); 1747 1748 /* 1749 * "bad retransmit" recovery without timestamps. 1750 */ 1751 if ((to.to_flags & TOF_TS) == 0 && 1752 tp->t_rxtshift == 1 && 1753 tp->t_flags & TF_PREVVALID && 1754 (int)(ticks - tp->t_badrxtwin) < 0) { 1755 cc_cong_signal(tp, th, CC_RTO_ERR); 1756 } 1757 1758 /* 1759 * Recalculate the transmit timer / rtt. 1760 * 1761 * Some boxes send broken timestamp replies 1762 * during the SYN+ACK phase, ignore 1763 * timestamps of 0 or we could calculate a 1764 * huge RTT and blow up the retransmit timer. 1765 */ 1766 if ((to.to_flags & TOF_TS) != 0 && 1767 to.to_tsecr) { 1768 uint32_t t; 1769 1770 t = tcp_ts_getticks() - to.to_tsecr; 1771 if (!tp->t_rttlow || tp->t_rttlow > t) 1772 tp->t_rttlow = t; 1773 tcp_xmit_timer(tp, 1774 TCP_TS_TO_TICKS(t) + 1); 1775 } else if (tp->t_rtttime && 1776 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1777 if (!tp->t_rttlow || 1778 tp->t_rttlow > ticks - tp->t_rtttime) 1779 tp->t_rttlow = ticks - tp->t_rtttime; 1780 tcp_xmit_timer(tp, 1781 ticks - tp->t_rtttime); 1782 } 1783 acked = BYTES_THIS_ACK(tp, th); 1784 1785 #ifdef TCP_HHOOK 1786 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1787 hhook_run_tcp_est_in(tp, th, &to); 1788 #endif 1789 1790 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1791 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1792 sbdrop(&so->so_snd, acked); 1793 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1794 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1795 tp->snd_recover = th->th_ack - 1; 1796 1797 /* 1798 * Let the congestion control algorithm update 1799 * congestion control related information. This 1800 * typically means increasing the congestion 1801 * window. 1802 */ 1803 cc_ack_received(tp, th, nsegs, CC_ACK); 1804 1805 tp->snd_una = th->th_ack; 1806 /* 1807 * Pull snd_wl2 up to prevent seq wrap relative 1808 * to th_ack. 1809 */ 1810 tp->snd_wl2 = th->th_ack; 1811 tp->t_dupacks = 0; 1812 m_freem(m); 1813 1814 /* 1815 * If all outstanding data are acked, stop 1816 * retransmit timer, otherwise restart timer 1817 * using current (possibly backed-off) value. 1818 * If process is waiting for space, 1819 * wakeup/selwakeup/signal. If data 1820 * are ready to send, let tcp_output 1821 * decide between more output or persist. 1822 */ 1823 #ifdef TCPDEBUG 1824 if (so->so_options & SO_DEBUG) 1825 tcp_trace(TA_INPUT, ostate, tp, 1826 (void *)tcp_saveipgen, 1827 &tcp_savetcp, 0); 1828 #endif 1829 TCP_PROBE3(debug__input, tp, th, m); 1830 if (tp->snd_una == tp->snd_max) 1831 tcp_timer_activate(tp, TT_REXMT, 0); 1832 else if (!tcp_timer_active(tp, TT_PERSIST)) 1833 tcp_timer_activate(tp, TT_REXMT, 1834 tp->t_rxtcur); 1835 sowwakeup(so); 1836 if (sbavail(&so->so_snd)) 1837 (void) tp->t_fb->tfb_tcp_output(tp); 1838 goto check_delack; 1839 } 1840 } else if (th->th_ack == tp->snd_una && 1841 tlen <= sbspace(&so->so_rcv)) { 1842 int newsize = 0; /* automatic sockbuf scaling */ 1843 1844 /* 1845 * This is a pure, in-sequence data packet with 1846 * nothing on the reassembly queue and we have enough 1847 * buffer space to take it. 1848 */ 1849 /* Clean receiver SACK report if present */ 1850 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1851 tcp_clean_sackreport(tp); 1852 TCPSTAT_INC(tcps_preddat); 1853 tp->rcv_nxt += tlen; 1854 /* 1855 * Pull snd_wl1 up to prevent seq wrap relative to 1856 * th_seq. 1857 */ 1858 tp->snd_wl1 = th->th_seq; 1859 /* 1860 * Pull rcv_up up to prevent seq wrap relative to 1861 * rcv_nxt. 1862 */ 1863 tp->rcv_up = tp->rcv_nxt; 1864 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1865 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1866 #ifdef TCPDEBUG 1867 if (so->so_options & SO_DEBUG) 1868 tcp_trace(TA_INPUT, ostate, tp, 1869 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1870 #endif 1871 TCP_PROBE3(debug__input, tp, th, m); 1872 1873 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1874 1875 /* Add data to socket buffer. */ 1876 SOCKBUF_LOCK(&so->so_rcv); 1877 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1878 m_freem(m); 1879 } else { 1880 /* 1881 * Set new socket buffer size. 1882 * Give up when limit is reached. 1883 */ 1884 if (newsize) 1885 if (!sbreserve_locked(&so->so_rcv, 1886 newsize, so, NULL)) 1887 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1888 m_adj(m, drop_hdrlen); /* delayed header drop */ 1889 sbappendstream_locked(&so->so_rcv, m, 0); 1890 } 1891 /* NB: sorwakeup_locked() does an implicit unlock. */ 1892 sorwakeup_locked(so); 1893 if (DELAY_ACK(tp, tlen)) { 1894 tp->t_flags |= TF_DELACK; 1895 } else { 1896 tp->t_flags |= TF_ACKNOW; 1897 tp->t_fb->tfb_tcp_output(tp); 1898 } 1899 goto check_delack; 1900 } 1901 } 1902 1903 /* 1904 * Calculate amount of space in receive window, 1905 * and then do TCP input processing. 1906 * Receive window is amount of space in rcv queue, 1907 * but not less than advertised window. 1908 */ 1909 win = sbspace(&so->so_rcv); 1910 if (win < 0) 1911 win = 0; 1912 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1913 1914 switch (tp->t_state) { 1915 1916 /* 1917 * If the state is SYN_RECEIVED: 1918 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1919 */ 1920 case TCPS_SYN_RECEIVED: 1921 if ((thflags & TH_ACK) && 1922 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1923 SEQ_GT(th->th_ack, tp->snd_max))) { 1924 rstreason = BANDLIM_RST_OPENPORT; 1925 goto dropwithreset; 1926 } 1927 if (IS_FASTOPEN(tp->t_flags)) { 1928 /* 1929 * When a TFO connection is in SYN_RECEIVED, the 1930 * only valid packets are the initial SYN, a 1931 * retransmit/copy of the initial SYN (possibly with 1932 * a subset of the original data), a valid ACK, a 1933 * FIN, or a RST. 1934 */ 1935 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1936 rstreason = BANDLIM_RST_OPENPORT; 1937 goto dropwithreset; 1938 } else if (thflags & TH_SYN) { 1939 /* non-initial SYN is ignored */ 1940 if ((tcp_timer_active(tp, TT_DELACK) || 1941 tcp_timer_active(tp, TT_REXMT))) 1942 goto drop; 1943 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1944 goto drop; 1945 } 1946 } 1947 break; 1948 1949 /* 1950 * If the state is SYN_SENT: 1951 * if seg contains a RST with valid ACK (SEQ.ACK has already 1952 * been verified), then drop the connection. 1953 * if seg contains a RST without an ACK, drop the seg. 1954 * if seg does not contain SYN, then drop the seg. 1955 * Otherwise this is an acceptable SYN segment 1956 * initialize tp->rcv_nxt and tp->irs 1957 * if seg contains ack then advance tp->snd_una 1958 * if seg contains an ECE and ECN support is enabled, the stream 1959 * is ECN capable. 1960 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1961 * arrange for segment to be acked (eventually) 1962 * continue processing rest of data/controls, beginning with URG 1963 */ 1964 case TCPS_SYN_SENT: 1965 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1966 TCP_PROBE5(connect__refused, NULL, tp, 1967 m, tp, th); 1968 tp = tcp_drop(tp, ECONNREFUSED); 1969 } 1970 if (thflags & TH_RST) 1971 goto drop; 1972 if (!(thflags & TH_SYN)) 1973 goto drop; 1974 1975 tp->irs = th->th_seq; 1976 tcp_rcvseqinit(tp); 1977 if (thflags & TH_ACK) { 1978 int tfo_partial_ack = 0; 1979 1980 TCPSTAT_INC(tcps_connects); 1981 soisconnected(so); 1982 #ifdef MAC 1983 mac_socketpeer_set_from_mbuf(m, so); 1984 #endif 1985 /* Do window scaling on this connection? */ 1986 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1987 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1988 tp->rcv_scale = tp->request_r_scale; 1989 } 1990 tp->rcv_adv += min(tp->rcv_wnd, 1991 TCP_MAXWIN << tp->rcv_scale); 1992 tp->snd_una++; /* SYN is acked */ 1993 /* 1994 * If not all the data that was sent in the TFO SYN 1995 * has been acked, resend the remainder right away. 1996 */ 1997 if (IS_FASTOPEN(tp->t_flags) && 1998 (tp->snd_una != tp->snd_max)) { 1999 tp->snd_nxt = th->th_ack; 2000 tfo_partial_ack = 1; 2001 } 2002 /* 2003 * If there's data, delay ACK; if there's also a FIN 2004 * ACKNOW will be turned on later. 2005 */ 2006 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2007 tcp_timer_activate(tp, TT_DELACK, 2008 tcp_delacktime); 2009 else 2010 tp->t_flags |= TF_ACKNOW; 2011 2012 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2013 V_tcp_do_ecn) { 2014 tp->t_flags |= TF_ECN_PERMIT; 2015 TCPSTAT_INC(tcps_ecn_shs); 2016 } 2017 2018 /* 2019 * Received <SYN,ACK> in SYN_SENT[*] state. 2020 * Transitions: 2021 * SYN_SENT --> ESTABLISHED 2022 * SYN_SENT* --> FIN_WAIT_1 2023 */ 2024 tp->t_starttime = ticks; 2025 if (tp->t_flags & TF_NEEDFIN) { 2026 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2027 tp->t_flags &= ~TF_NEEDFIN; 2028 thflags &= ~TH_SYN; 2029 } else { 2030 tcp_state_change(tp, TCPS_ESTABLISHED); 2031 TCP_PROBE5(connect__established, NULL, tp, 2032 m, tp, th); 2033 cc_conn_init(tp); 2034 tcp_timer_activate(tp, TT_KEEP, 2035 TP_KEEPIDLE(tp)); 2036 } 2037 } else { 2038 /* 2039 * Received initial SYN in SYN-SENT[*] state => 2040 * simultaneous open. 2041 * If it succeeds, connection is * half-synchronized. 2042 * Otherwise, do 3-way handshake: 2043 * SYN-SENT -> SYN-RECEIVED 2044 * SYN-SENT* -> SYN-RECEIVED* 2045 */ 2046 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2047 tcp_timer_activate(tp, TT_REXMT, 0); 2048 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2049 } 2050 2051 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2052 INP_WLOCK_ASSERT(tp->t_inpcb); 2053 2054 /* 2055 * Advance th->th_seq to correspond to first data byte. 2056 * If data, trim to stay within window, 2057 * dropping FIN if necessary. 2058 */ 2059 th->th_seq++; 2060 if (tlen > tp->rcv_wnd) { 2061 todrop = tlen - tp->rcv_wnd; 2062 m_adj(m, -todrop); 2063 tlen = tp->rcv_wnd; 2064 thflags &= ~TH_FIN; 2065 TCPSTAT_INC(tcps_rcvpackafterwin); 2066 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2067 } 2068 tp->snd_wl1 = th->th_seq - 1; 2069 tp->rcv_up = th->th_seq; 2070 /* 2071 * Client side of transaction: already sent SYN and data. 2072 * If the remote host used T/TCP to validate the SYN, 2073 * our data will be ACK'd; if so, enter normal data segment 2074 * processing in the middle of step 5, ack processing. 2075 * Otherwise, goto step 6. 2076 */ 2077 if (thflags & TH_ACK) 2078 goto process_ACK; 2079 2080 goto step6; 2081 2082 /* 2083 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2084 * do normal processing. 2085 * 2086 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2087 */ 2088 case TCPS_LAST_ACK: 2089 case TCPS_CLOSING: 2090 break; /* continue normal processing */ 2091 } 2092 2093 /* 2094 * States other than LISTEN or SYN_SENT. 2095 * First check the RST flag and sequence number since reset segments 2096 * are exempt from the timestamp and connection count tests. This 2097 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2098 * below which allowed reset segments in half the sequence space 2099 * to fall though and be processed (which gives forged reset 2100 * segments with a random sequence number a 50 percent chance of 2101 * killing a connection). 2102 * Then check timestamp, if present. 2103 * Then check the connection count, if present. 2104 * Then check that at least some bytes of segment are within 2105 * receive window. If segment begins before rcv_nxt, 2106 * drop leading data (and SYN); if nothing left, just ack. 2107 */ 2108 if (thflags & TH_RST) { 2109 /* 2110 * RFC5961 Section 3.2 2111 * 2112 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2113 * - If RST is in window, we send challenge ACK. 2114 * 2115 * Note: to take into account delayed ACKs, we should 2116 * test against last_ack_sent instead of rcv_nxt. 2117 * Note 2: we handle special case of closed window, not 2118 * covered by the RFC. 2119 */ 2120 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2121 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2122 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2123 2124 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2125 KASSERT(tp->t_state != TCPS_SYN_SENT, 2126 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2127 __func__, th, tp)); 2128 2129 if (V_tcp_insecure_rst || 2130 tp->last_ack_sent == th->th_seq) { 2131 TCPSTAT_INC(tcps_drops); 2132 /* Drop the connection. */ 2133 switch (tp->t_state) { 2134 case TCPS_SYN_RECEIVED: 2135 so->so_error = ECONNREFUSED; 2136 goto close; 2137 case TCPS_ESTABLISHED: 2138 case TCPS_FIN_WAIT_1: 2139 case TCPS_FIN_WAIT_2: 2140 case TCPS_CLOSE_WAIT: 2141 case TCPS_CLOSING: 2142 case TCPS_LAST_ACK: 2143 so->so_error = ECONNRESET; 2144 close: 2145 /* FALLTHROUGH */ 2146 default: 2147 tp = tcp_close(tp); 2148 } 2149 } else { 2150 TCPSTAT_INC(tcps_badrst); 2151 /* Send challenge ACK. */ 2152 tcp_respond(tp, mtod(m, void *), th, m, 2153 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2154 tp->last_ack_sent = tp->rcv_nxt; 2155 m = NULL; 2156 } 2157 } 2158 goto drop; 2159 } 2160 2161 /* 2162 * RFC5961 Section 4.2 2163 * Send challenge ACK for any SYN in synchronized state. 2164 */ 2165 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2166 tp->t_state != TCPS_SYN_RECEIVED) { 2167 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2168 2169 TCPSTAT_INC(tcps_badsyn); 2170 if (V_tcp_insecure_syn && 2171 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2172 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2173 tp = tcp_drop(tp, ECONNRESET); 2174 rstreason = BANDLIM_UNLIMITED; 2175 } else { 2176 /* Send challenge ACK. */ 2177 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2178 tp->snd_nxt, TH_ACK); 2179 tp->last_ack_sent = tp->rcv_nxt; 2180 m = NULL; 2181 } 2182 goto drop; 2183 } 2184 2185 /* 2186 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2187 * and it's less than ts_recent, drop it. 2188 */ 2189 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2190 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2191 2192 /* Check to see if ts_recent is over 24 days old. */ 2193 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2194 /* 2195 * Invalidate ts_recent. If this segment updates 2196 * ts_recent, the age will be reset later and ts_recent 2197 * will get a valid value. If it does not, setting 2198 * ts_recent to zero will at least satisfy the 2199 * requirement that zero be placed in the timestamp 2200 * echo reply when ts_recent isn't valid. The 2201 * age isn't reset until we get a valid ts_recent 2202 * because we don't want out-of-order segments to be 2203 * dropped when ts_recent is old. 2204 */ 2205 tp->ts_recent = 0; 2206 } else { 2207 TCPSTAT_INC(tcps_rcvduppack); 2208 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2209 TCPSTAT_INC(tcps_pawsdrop); 2210 if (tlen) 2211 goto dropafterack; 2212 goto drop; 2213 } 2214 } 2215 2216 /* 2217 * In the SYN-RECEIVED state, validate that the packet belongs to 2218 * this connection before trimming the data to fit the receive 2219 * window. Check the sequence number versus IRS since we know 2220 * the sequence numbers haven't wrapped. This is a partial fix 2221 * for the "LAND" DoS attack. 2222 */ 2223 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2224 rstreason = BANDLIM_RST_OPENPORT; 2225 goto dropwithreset; 2226 } 2227 2228 todrop = tp->rcv_nxt - th->th_seq; 2229 if (todrop > 0) { 2230 if (thflags & TH_SYN) { 2231 thflags &= ~TH_SYN; 2232 th->th_seq++; 2233 if (th->th_urp > 1) 2234 th->th_urp--; 2235 else 2236 thflags &= ~TH_URG; 2237 todrop--; 2238 } 2239 /* 2240 * Following if statement from Stevens, vol. 2, p. 960. 2241 */ 2242 if (todrop > tlen 2243 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2244 /* 2245 * Any valid FIN must be to the left of the window. 2246 * At this point the FIN must be a duplicate or out 2247 * of sequence; drop it. 2248 */ 2249 thflags &= ~TH_FIN; 2250 2251 /* 2252 * Send an ACK to resynchronize and drop any data. 2253 * But keep on processing for RST or ACK. 2254 */ 2255 tp->t_flags |= TF_ACKNOW; 2256 todrop = tlen; 2257 TCPSTAT_INC(tcps_rcvduppack); 2258 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2259 } else { 2260 TCPSTAT_INC(tcps_rcvpartduppack); 2261 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2262 } 2263 /* 2264 * DSACK - add SACK block for dropped range 2265 */ 2266 if (tp->t_flags & TF_SACK_PERMIT) { 2267 tcp_update_sack_list(tp, th->th_seq, 2268 th->th_seq + todrop); 2269 /* 2270 * ACK now, as the next in-sequence segment 2271 * will clear the DSACK block again 2272 */ 2273 tp->t_flags |= TF_ACKNOW; 2274 } 2275 drop_hdrlen += todrop; /* drop from the top afterwards */ 2276 th->th_seq += todrop; 2277 tlen -= todrop; 2278 if (th->th_urp > todrop) 2279 th->th_urp -= todrop; 2280 else { 2281 thflags &= ~TH_URG; 2282 th->th_urp = 0; 2283 } 2284 } 2285 2286 /* 2287 * If new data are received on a connection after the 2288 * user processes are gone, then RST the other end. 2289 */ 2290 if ((so->so_state & SS_NOFDREF) && 2291 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2292 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2293 2294 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2295 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2296 "after socket was closed, " 2297 "sending RST and removing tcpcb\n", 2298 s, __func__, tcpstates[tp->t_state], tlen); 2299 free(s, M_TCPLOG); 2300 } 2301 tp = tcp_close(tp); 2302 TCPSTAT_INC(tcps_rcvafterclose); 2303 rstreason = BANDLIM_UNLIMITED; 2304 goto dropwithreset; 2305 } 2306 2307 /* 2308 * If segment ends after window, drop trailing data 2309 * (and PUSH and FIN); if nothing left, just ACK. 2310 */ 2311 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2312 if (todrop > 0) { 2313 TCPSTAT_INC(tcps_rcvpackafterwin); 2314 if (todrop >= tlen) { 2315 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2316 /* 2317 * If window is closed can only take segments at 2318 * window edge, and have to drop data and PUSH from 2319 * incoming segments. Continue processing, but 2320 * remember to ack. Otherwise, drop segment 2321 * and ack. 2322 */ 2323 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2324 tp->t_flags |= TF_ACKNOW; 2325 TCPSTAT_INC(tcps_rcvwinprobe); 2326 } else 2327 goto dropafterack; 2328 } else 2329 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2330 m_adj(m, -todrop); 2331 tlen -= todrop; 2332 thflags &= ~(TH_PUSH|TH_FIN); 2333 } 2334 2335 /* 2336 * If last ACK falls within this segment's sequence numbers, 2337 * record its timestamp. 2338 * NOTE: 2339 * 1) That the test incorporates suggestions from the latest 2340 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2341 * 2) That updating only on newer timestamps interferes with 2342 * our earlier PAWS tests, so this check should be solely 2343 * predicated on the sequence space of this segment. 2344 * 3) That we modify the segment boundary check to be 2345 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2346 * instead of RFC1323's 2347 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2348 * This modified check allows us to overcome RFC1323's 2349 * limitations as described in Stevens TCP/IP Illustrated 2350 * Vol. 2 p.869. In such cases, we can still calculate the 2351 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2352 */ 2353 if ((to.to_flags & TOF_TS) != 0 && 2354 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2355 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2356 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2357 tp->ts_recent_age = tcp_ts_getticks(); 2358 tp->ts_recent = to.to_tsval; 2359 } 2360 2361 /* 2362 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2363 * flag is on (half-synchronized state), then queue data for 2364 * later processing; else drop segment and return. 2365 */ 2366 if ((thflags & TH_ACK) == 0) { 2367 if (tp->t_state == TCPS_SYN_RECEIVED || 2368 (tp->t_flags & TF_NEEDSYN)) { 2369 if (tp->t_state == TCPS_SYN_RECEIVED && 2370 IS_FASTOPEN(tp->t_flags)) { 2371 tp->snd_wnd = tiwin; 2372 cc_conn_init(tp); 2373 } 2374 goto step6; 2375 } else if (tp->t_flags & TF_ACKNOW) 2376 goto dropafterack; 2377 else 2378 goto drop; 2379 } 2380 2381 /* 2382 * Ack processing. 2383 */ 2384 switch (tp->t_state) { 2385 2386 /* 2387 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2388 * ESTABLISHED state and continue processing. 2389 * The ACK was checked above. 2390 */ 2391 case TCPS_SYN_RECEIVED: 2392 2393 TCPSTAT_INC(tcps_connects); 2394 soisconnected(so); 2395 /* Do window scaling? */ 2396 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2397 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2398 tp->rcv_scale = tp->request_r_scale; 2399 } 2400 tp->snd_wnd = tiwin; 2401 /* 2402 * Make transitions: 2403 * SYN-RECEIVED -> ESTABLISHED 2404 * SYN-RECEIVED* -> FIN-WAIT-1 2405 */ 2406 tp->t_starttime = ticks; 2407 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2408 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2409 tp->t_tfo_pending = NULL; 2410 2411 /* 2412 * Account for the ACK of our SYN prior to 2413 * regular ACK processing below. 2414 */ 2415 tp->snd_una++; 2416 } 2417 if (tp->t_flags & TF_NEEDFIN) { 2418 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2419 tp->t_flags &= ~TF_NEEDFIN; 2420 } else { 2421 tcp_state_change(tp, TCPS_ESTABLISHED); 2422 TCP_PROBE5(accept__established, NULL, tp, 2423 m, tp, th); 2424 /* 2425 * TFO connections call cc_conn_init() during SYN 2426 * processing. Calling it again here for such 2427 * connections is not harmless as it would undo the 2428 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2429 * is retransmitted. 2430 */ 2431 if (!IS_FASTOPEN(tp->t_flags)) 2432 cc_conn_init(tp); 2433 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2434 } 2435 /* 2436 * If segment contains data or ACK, will call tcp_reass() 2437 * later; if not, do so now to pass queued data to user. 2438 */ 2439 if (tlen == 0 && (thflags & TH_FIN) == 0) 2440 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2441 (struct mbuf *)0); 2442 tp->snd_wl1 = th->th_seq - 1; 2443 /* FALLTHROUGH */ 2444 2445 /* 2446 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2447 * ACKs. If the ack is in the range 2448 * tp->snd_una < th->th_ack <= tp->snd_max 2449 * then advance tp->snd_una to th->th_ack and drop 2450 * data from the retransmission queue. If this ACK reflects 2451 * more up to date window information we update our window information. 2452 */ 2453 case TCPS_ESTABLISHED: 2454 case TCPS_FIN_WAIT_1: 2455 case TCPS_FIN_WAIT_2: 2456 case TCPS_CLOSE_WAIT: 2457 case TCPS_CLOSING: 2458 case TCPS_LAST_ACK: 2459 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2460 TCPSTAT_INC(tcps_rcvacktoomuch); 2461 goto dropafterack; 2462 } 2463 if ((tp->t_flags & TF_SACK_PERMIT) && 2464 ((to.to_flags & TOF_SACK) || 2465 !TAILQ_EMPTY(&tp->snd_holes))) 2466 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2467 else 2468 /* 2469 * Reset the value so that previous (valid) value 2470 * from the last ack with SACK doesn't get used. 2471 */ 2472 tp->sackhint.sacked_bytes = 0; 2473 2474 #ifdef TCP_HHOOK 2475 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2476 hhook_run_tcp_est_in(tp, th, &to); 2477 #endif 2478 2479 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2480 u_int maxseg; 2481 2482 maxseg = tcp_maxseg(tp); 2483 if (tlen == 0 && 2484 (tiwin == tp->snd_wnd || 2485 (tp->t_flags & TF_SACK_PERMIT))) { 2486 /* 2487 * If this is the first time we've seen a 2488 * FIN from the remote, this is not a 2489 * duplicate and it needs to be processed 2490 * normally. This happens during a 2491 * simultaneous close. 2492 */ 2493 if ((thflags & TH_FIN) && 2494 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2495 tp->t_dupacks = 0; 2496 break; 2497 } 2498 TCPSTAT_INC(tcps_rcvdupack); 2499 /* 2500 * If we have outstanding data (other than 2501 * a window probe), this is a completely 2502 * duplicate ack (ie, window info didn't 2503 * change and FIN isn't set), 2504 * the ack is the biggest we've 2505 * seen and we've seen exactly our rexmt 2506 * threshold of them, assume a packet 2507 * has been dropped and retransmit it. 2508 * Kludge snd_nxt & the congestion 2509 * window so we send only this one 2510 * packet. 2511 * 2512 * We know we're losing at the current 2513 * window size so do congestion avoidance 2514 * (set ssthresh to half the current window 2515 * and pull our congestion window back to 2516 * the new ssthresh). 2517 * 2518 * Dup acks mean that packets have left the 2519 * network (they're now cached at the receiver) 2520 * so bump cwnd by the amount in the receiver 2521 * to keep a constant cwnd packets in the 2522 * network. 2523 * 2524 * When using TCP ECN, notify the peer that 2525 * we reduced the cwnd. 2526 */ 2527 /* 2528 * Following 2 kinds of acks should not affect 2529 * dupack counting: 2530 * 1) Old acks 2531 * 2) Acks with SACK but without any new SACK 2532 * information in them. These could result from 2533 * any anomaly in the network like a switch 2534 * duplicating packets or a possible DoS attack. 2535 */ 2536 if (th->th_ack != tp->snd_una || 2537 ((tp->t_flags & TF_SACK_PERMIT) && 2538 !sack_changed)) 2539 break; 2540 else if (!tcp_timer_active(tp, TT_REXMT)) 2541 tp->t_dupacks = 0; 2542 else if (++tp->t_dupacks > tcprexmtthresh || 2543 IN_FASTRECOVERY(tp->t_flags)) { 2544 cc_ack_received(tp, th, nsegs, 2545 CC_DUPACK); 2546 if ((tp->t_flags & TF_SACK_PERMIT) && 2547 IN_FASTRECOVERY(tp->t_flags)) { 2548 int awnd; 2549 2550 /* 2551 * Compute the amount of data in flight first. 2552 * We can inject new data into the pipe iff 2553 * we have less than 1/2 the original window's 2554 * worth of data in flight. 2555 */ 2556 if (V_tcp_do_rfc6675_pipe) 2557 awnd = tcp_compute_pipe(tp); 2558 else 2559 awnd = (tp->snd_nxt - tp->snd_fack) + 2560 tp->sackhint.sack_bytes_rexmit; 2561 2562 if (awnd < tp->snd_ssthresh) { 2563 tp->snd_cwnd += maxseg; 2564 if (tp->snd_cwnd > tp->snd_ssthresh) 2565 tp->snd_cwnd = tp->snd_ssthresh; 2566 } 2567 } else 2568 tp->snd_cwnd += maxseg; 2569 (void) tp->t_fb->tfb_tcp_output(tp); 2570 goto drop; 2571 } else if (tp->t_dupacks == tcprexmtthresh) { 2572 tcp_seq onxt = tp->snd_nxt; 2573 2574 /* 2575 * If we're doing sack, check to 2576 * see if we're already in sack 2577 * recovery. If we're not doing sack, 2578 * check to see if we're in newreno 2579 * recovery. 2580 */ 2581 if (tp->t_flags & TF_SACK_PERMIT) { 2582 if (IN_FASTRECOVERY(tp->t_flags)) { 2583 tp->t_dupacks = 0; 2584 break; 2585 } 2586 } else { 2587 if (SEQ_LEQ(th->th_ack, 2588 tp->snd_recover)) { 2589 tp->t_dupacks = 0; 2590 break; 2591 } 2592 } 2593 /* Congestion signal before ack. */ 2594 cc_cong_signal(tp, th, CC_NDUPACK); 2595 cc_ack_received(tp, th, nsegs, 2596 CC_DUPACK); 2597 tcp_timer_activate(tp, TT_REXMT, 0); 2598 tp->t_rtttime = 0; 2599 if (tp->t_flags & TF_SACK_PERMIT) { 2600 TCPSTAT_INC( 2601 tcps_sack_recovery_episode); 2602 tp->sack_newdata = tp->snd_nxt; 2603 tp->snd_cwnd = maxseg; 2604 (void) tp->t_fb->tfb_tcp_output(tp); 2605 goto drop; 2606 } 2607 tp->snd_nxt = th->th_ack; 2608 tp->snd_cwnd = maxseg; 2609 (void) tp->t_fb->tfb_tcp_output(tp); 2610 KASSERT(tp->snd_limited <= 2, 2611 ("%s: tp->snd_limited too big", 2612 __func__)); 2613 tp->snd_cwnd = tp->snd_ssthresh + 2614 maxseg * 2615 (tp->t_dupacks - tp->snd_limited); 2616 if (SEQ_GT(onxt, tp->snd_nxt)) 2617 tp->snd_nxt = onxt; 2618 goto drop; 2619 } else if (V_tcp_do_rfc3042) { 2620 /* 2621 * Process first and second duplicate 2622 * ACKs. Each indicates a segment 2623 * leaving the network, creating room 2624 * for more. Make sure we can send a 2625 * packet on reception of each duplicate 2626 * ACK by increasing snd_cwnd by one 2627 * segment. Restore the original 2628 * snd_cwnd after packet transmission. 2629 */ 2630 cc_ack_received(tp, th, nsegs, 2631 CC_DUPACK); 2632 uint32_t oldcwnd = tp->snd_cwnd; 2633 tcp_seq oldsndmax = tp->snd_max; 2634 u_int sent; 2635 int avail; 2636 2637 KASSERT(tp->t_dupacks == 1 || 2638 tp->t_dupacks == 2, 2639 ("%s: dupacks not 1 or 2", 2640 __func__)); 2641 if (tp->t_dupacks == 1) 2642 tp->snd_limited = 0; 2643 tp->snd_cwnd = 2644 (tp->snd_nxt - tp->snd_una) + 2645 (tp->t_dupacks - tp->snd_limited) * 2646 maxseg; 2647 /* 2648 * Only call tcp_output when there 2649 * is new data available to be sent. 2650 * Otherwise we would send pure ACKs. 2651 */ 2652 SOCKBUF_LOCK(&so->so_snd); 2653 avail = sbavail(&so->so_snd) - 2654 (tp->snd_nxt - tp->snd_una); 2655 SOCKBUF_UNLOCK(&so->so_snd); 2656 if (avail > 0) 2657 (void) tp->t_fb->tfb_tcp_output(tp); 2658 sent = tp->snd_max - oldsndmax; 2659 if (sent > maxseg) { 2660 KASSERT((tp->t_dupacks == 2 && 2661 tp->snd_limited == 0) || 2662 (sent == maxseg + 1 && 2663 tp->t_flags & TF_SENTFIN), 2664 ("%s: sent too much", 2665 __func__)); 2666 tp->snd_limited = 2; 2667 } else if (sent > 0) 2668 ++tp->snd_limited; 2669 tp->snd_cwnd = oldcwnd; 2670 goto drop; 2671 } 2672 } 2673 break; 2674 } else { 2675 /* 2676 * This ack is advancing the left edge, reset the 2677 * counter. 2678 */ 2679 tp->t_dupacks = 0; 2680 /* 2681 * If this ack also has new SACK info, increment the 2682 * counter as per rfc6675. 2683 */ 2684 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2685 tp->t_dupacks++; 2686 } 2687 2688 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2689 ("%s: th_ack <= snd_una", __func__)); 2690 2691 /* 2692 * If the congestion window was inflated to account 2693 * for the other side's cached packets, retract it. 2694 */ 2695 if (IN_FASTRECOVERY(tp->t_flags)) { 2696 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2697 if (tp->t_flags & TF_SACK_PERMIT) 2698 tcp_sack_partialack(tp, th); 2699 else 2700 tcp_newreno_partial_ack(tp, th); 2701 } else 2702 cc_post_recovery(tp, th); 2703 } 2704 /* 2705 * If we reach this point, ACK is not a duplicate, 2706 * i.e., it ACKs something we sent. 2707 */ 2708 if (tp->t_flags & TF_NEEDSYN) { 2709 /* 2710 * T/TCP: Connection was half-synchronized, and our 2711 * SYN has been ACK'd (so connection is now fully 2712 * synchronized). Go to non-starred state, 2713 * increment snd_una for ACK of SYN, and check if 2714 * we can do window scaling. 2715 */ 2716 tp->t_flags &= ~TF_NEEDSYN; 2717 tp->snd_una++; 2718 /* Do window scaling? */ 2719 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2720 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2721 tp->rcv_scale = tp->request_r_scale; 2722 /* Send window already scaled. */ 2723 } 2724 } 2725 2726 process_ACK: 2727 INP_WLOCK_ASSERT(tp->t_inpcb); 2728 2729 acked = BYTES_THIS_ACK(tp, th); 2730 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2731 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2732 tp->snd_una, th->th_ack, tp, m)); 2733 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2734 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2735 2736 /* 2737 * If we just performed our first retransmit, and the ACK 2738 * arrives within our recovery window, then it was a mistake 2739 * to do the retransmit in the first place. Recover our 2740 * original cwnd and ssthresh, and proceed to transmit where 2741 * we left off. 2742 */ 2743 if (tp->t_rxtshift == 1 && 2744 tp->t_flags & TF_PREVVALID && 2745 tp->t_badrxtwin && 2746 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2747 cc_cong_signal(tp, th, CC_RTO_ERR); 2748 2749 /* 2750 * If we have a timestamp reply, update smoothed 2751 * round trip time. If no timestamp is present but 2752 * transmit timer is running and timed sequence 2753 * number was acked, update smoothed round trip time. 2754 * Since we now have an rtt measurement, cancel the 2755 * timer backoff (cf., Phil Karn's retransmit alg.). 2756 * Recompute the initial retransmit timer. 2757 * 2758 * Some boxes send broken timestamp replies 2759 * during the SYN+ACK phase, ignore 2760 * timestamps of 0 or we could calculate a 2761 * huge RTT and blow up the retransmit timer. 2762 */ 2763 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2764 uint32_t t; 2765 2766 t = tcp_ts_getticks() - to.to_tsecr; 2767 if (!tp->t_rttlow || tp->t_rttlow > t) 2768 tp->t_rttlow = t; 2769 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2770 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2771 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2772 tp->t_rttlow = ticks - tp->t_rtttime; 2773 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2774 } 2775 2776 /* 2777 * If all outstanding data is acked, stop retransmit 2778 * timer and remember to restart (more output or persist). 2779 * If there is more data to be acked, restart retransmit 2780 * timer, using current (possibly backed-off) value. 2781 */ 2782 if (th->th_ack == tp->snd_max) { 2783 tcp_timer_activate(tp, TT_REXMT, 0); 2784 needoutput = 1; 2785 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2786 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2787 2788 /* 2789 * If no data (only SYN) was ACK'd, 2790 * skip rest of ACK processing. 2791 */ 2792 if (acked == 0) 2793 goto step6; 2794 2795 /* 2796 * Let the congestion control algorithm update congestion 2797 * control related information. This typically means increasing 2798 * the congestion window. 2799 */ 2800 cc_ack_received(tp, th, nsegs, CC_ACK); 2801 2802 SOCKBUF_LOCK(&so->so_snd); 2803 if (acked > sbavail(&so->so_snd)) { 2804 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2805 tp->snd_wnd -= sbavail(&so->so_snd); 2806 else 2807 tp->snd_wnd = 0; 2808 mfree = sbcut_locked(&so->so_snd, 2809 (int)sbavail(&so->so_snd)); 2810 ourfinisacked = 1; 2811 } else { 2812 mfree = sbcut_locked(&so->so_snd, acked); 2813 if (tp->snd_wnd >= (uint32_t) acked) 2814 tp->snd_wnd -= acked; 2815 else 2816 tp->snd_wnd = 0; 2817 ourfinisacked = 0; 2818 } 2819 /* NB: sowwakeup_locked() does an implicit unlock. */ 2820 sowwakeup_locked(so); 2821 m_freem(mfree); 2822 /* Detect una wraparound. */ 2823 if (!IN_RECOVERY(tp->t_flags) && 2824 SEQ_GT(tp->snd_una, tp->snd_recover) && 2825 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2826 tp->snd_recover = th->th_ack - 1; 2827 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2828 if (IN_RECOVERY(tp->t_flags) && 2829 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2830 EXIT_RECOVERY(tp->t_flags); 2831 } 2832 tp->snd_una = th->th_ack; 2833 if (tp->t_flags & TF_SACK_PERMIT) { 2834 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2835 tp->snd_recover = tp->snd_una; 2836 } 2837 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2838 tp->snd_nxt = tp->snd_una; 2839 2840 switch (tp->t_state) { 2841 2842 /* 2843 * In FIN_WAIT_1 STATE in addition to the processing 2844 * for the ESTABLISHED state if our FIN is now acknowledged 2845 * then enter FIN_WAIT_2. 2846 */ 2847 case TCPS_FIN_WAIT_1: 2848 if (ourfinisacked) { 2849 /* 2850 * If we can't receive any more 2851 * data, then closing user can proceed. 2852 * Starting the timer is contrary to the 2853 * specification, but if we don't get a FIN 2854 * we'll hang forever. 2855 * 2856 * XXXjl: 2857 * we should release the tp also, and use a 2858 * compressed state. 2859 */ 2860 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2861 soisdisconnected(so); 2862 tcp_timer_activate(tp, TT_2MSL, 2863 (tcp_fast_finwait2_recycle ? 2864 tcp_finwait2_timeout : 2865 TP_MAXIDLE(tp))); 2866 } 2867 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2868 } 2869 break; 2870 2871 /* 2872 * In CLOSING STATE in addition to the processing for 2873 * the ESTABLISHED state if the ACK acknowledges our FIN 2874 * then enter the TIME-WAIT state, otherwise ignore 2875 * the segment. 2876 */ 2877 case TCPS_CLOSING: 2878 if (ourfinisacked) { 2879 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2880 tcp_twstart(tp); 2881 m_freem(m); 2882 return; 2883 } 2884 break; 2885 2886 /* 2887 * In LAST_ACK, we may still be waiting for data to drain 2888 * and/or to be acked, as well as for the ack of our FIN. 2889 * If our FIN is now acknowledged, delete the TCB, 2890 * enter the closed state and return. 2891 */ 2892 case TCPS_LAST_ACK: 2893 if (ourfinisacked) { 2894 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2895 tp = tcp_close(tp); 2896 goto drop; 2897 } 2898 break; 2899 } 2900 } 2901 2902 step6: 2903 INP_WLOCK_ASSERT(tp->t_inpcb); 2904 2905 /* 2906 * Update window information. 2907 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2908 */ 2909 if ((thflags & TH_ACK) && 2910 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2911 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2912 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2913 /* keep track of pure window updates */ 2914 if (tlen == 0 && 2915 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2916 TCPSTAT_INC(tcps_rcvwinupd); 2917 tp->snd_wnd = tiwin; 2918 tp->snd_wl1 = th->th_seq; 2919 tp->snd_wl2 = th->th_ack; 2920 if (tp->snd_wnd > tp->max_sndwnd) 2921 tp->max_sndwnd = tp->snd_wnd; 2922 needoutput = 1; 2923 } 2924 2925 /* 2926 * Process segments with URG. 2927 */ 2928 if ((thflags & TH_URG) && th->th_urp && 2929 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2930 /* 2931 * This is a kludge, but if we receive and accept 2932 * random urgent pointers, we'll crash in 2933 * soreceive. It's hard to imagine someone 2934 * actually wanting to send this much urgent data. 2935 */ 2936 SOCKBUF_LOCK(&so->so_rcv); 2937 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2938 th->th_urp = 0; /* XXX */ 2939 thflags &= ~TH_URG; /* XXX */ 2940 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2941 goto dodata; /* XXX */ 2942 } 2943 /* 2944 * If this segment advances the known urgent pointer, 2945 * then mark the data stream. This should not happen 2946 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2947 * a FIN has been received from the remote side. 2948 * In these states we ignore the URG. 2949 * 2950 * According to RFC961 (Assigned Protocols), 2951 * the urgent pointer points to the last octet 2952 * of urgent data. We continue, however, 2953 * to consider it to indicate the first octet 2954 * of data past the urgent section as the original 2955 * spec states (in one of two places). 2956 */ 2957 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2958 tp->rcv_up = th->th_seq + th->th_urp; 2959 so->so_oobmark = sbavail(&so->so_rcv) + 2960 (tp->rcv_up - tp->rcv_nxt) - 1; 2961 if (so->so_oobmark == 0) 2962 so->so_rcv.sb_state |= SBS_RCVATMARK; 2963 sohasoutofband(so); 2964 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2965 } 2966 SOCKBUF_UNLOCK(&so->so_rcv); 2967 /* 2968 * Remove out of band data so doesn't get presented to user. 2969 * This can happen independent of advancing the URG pointer, 2970 * but if two URG's are pending at once, some out-of-band 2971 * data may creep in... ick. 2972 */ 2973 if (th->th_urp <= (uint32_t)tlen && 2974 !(so->so_options & SO_OOBINLINE)) { 2975 /* hdr drop is delayed */ 2976 tcp_pulloutofband(so, th, m, drop_hdrlen); 2977 } 2978 } else { 2979 /* 2980 * If no out of band data is expected, 2981 * pull receive urgent pointer along 2982 * with the receive window. 2983 */ 2984 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2985 tp->rcv_up = tp->rcv_nxt; 2986 } 2987 dodata: /* XXX */ 2988 INP_WLOCK_ASSERT(tp->t_inpcb); 2989 2990 /* 2991 * Process the segment text, merging it into the TCP sequencing queue, 2992 * and arranging for acknowledgment of receipt if necessary. 2993 * This process logically involves adjusting tp->rcv_wnd as data 2994 * is presented to the user (this happens in tcp_usrreq.c, 2995 * case PRU_RCVD). If a FIN has already been received on this 2996 * connection then we just ignore the text. 2997 */ 2998 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 2999 IS_FASTOPEN(tp->t_flags)); 3000 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3001 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3002 tcp_seq save_start = th->th_seq; 3003 tcp_seq save_rnxt = tp->rcv_nxt; 3004 int save_tlen = tlen; 3005 m_adj(m, drop_hdrlen); /* delayed header drop */ 3006 /* 3007 * Insert segment which includes th into TCP reassembly queue 3008 * with control block tp. Set thflags to whether reassembly now 3009 * includes a segment with FIN. This handles the common case 3010 * inline (segment is the next to be received on an established 3011 * connection, and the queue is empty), avoiding linkage into 3012 * and removal from the queue and repetition of various 3013 * conversions. 3014 * Set DELACK for segments received in order, but ack 3015 * immediately when segments are out of order (so 3016 * fast retransmit can work). 3017 */ 3018 if (th->th_seq == tp->rcv_nxt && 3019 SEGQ_EMPTY(tp) && 3020 (TCPS_HAVEESTABLISHED(tp->t_state) || 3021 tfo_syn)) { 3022 if (DELAY_ACK(tp, tlen) || tfo_syn) 3023 tp->t_flags |= TF_DELACK; 3024 else 3025 tp->t_flags |= TF_ACKNOW; 3026 tp->rcv_nxt += tlen; 3027 thflags = th->th_flags & TH_FIN; 3028 TCPSTAT_INC(tcps_rcvpack); 3029 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3030 SOCKBUF_LOCK(&so->so_rcv); 3031 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3032 m_freem(m); 3033 else 3034 sbappendstream_locked(&so->so_rcv, m, 0); 3035 /* NB: sorwakeup_locked() does an implicit unlock. */ 3036 sorwakeup_locked(so); 3037 } else { 3038 /* 3039 * XXX: Due to the header drop above "th" is 3040 * theoretically invalid by now. Fortunately 3041 * m_adj() doesn't actually frees any mbufs 3042 * when trimming from the head. 3043 */ 3044 tcp_seq temp = save_start; 3045 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3046 tp->t_flags |= TF_ACKNOW; 3047 } 3048 if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0)) { 3049 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { 3050 /* 3051 * DSACK actually handled in the fastpath 3052 * above. 3053 */ 3054 tcp_update_sack_list(tp, save_start, 3055 save_start + save_tlen); 3056 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3057 if ((tp->rcv_numsacks >= 1) && 3058 (tp->sackblks[0].end == save_start)) { 3059 /* 3060 * Partial overlap, recorded at todrop 3061 * above. 3062 */ 3063 tcp_update_sack_list(tp, 3064 tp->sackblks[0].start, 3065 tp->sackblks[0].end); 3066 } else { 3067 tcp_update_dsack_list(tp, save_start, 3068 save_start + save_tlen); 3069 } 3070 } else if (tlen >= save_tlen) { 3071 /* Update of sackblks. */ 3072 tcp_update_dsack_list(tp, save_start, 3073 save_start + save_tlen); 3074 } else if (tlen > 0) { 3075 tcp_update_dsack_list(tp, save_start, 3076 save_start + tlen); 3077 } 3078 } 3079 #if 0 3080 /* 3081 * Note the amount of data that peer has sent into 3082 * our window, in order to estimate the sender's 3083 * buffer size. 3084 * XXX: Unused. 3085 */ 3086 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3087 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3088 else 3089 len = so->so_rcv.sb_hiwat; 3090 #endif 3091 } else { 3092 m_freem(m); 3093 thflags &= ~TH_FIN; 3094 } 3095 3096 /* 3097 * If FIN is received ACK the FIN and let the user know 3098 * that the connection is closing. 3099 */ 3100 if (thflags & TH_FIN) { 3101 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3102 socantrcvmore(so); 3103 /* 3104 * If connection is half-synchronized 3105 * (ie NEEDSYN flag on) then delay ACK, 3106 * so it may be piggybacked when SYN is sent. 3107 * Otherwise, since we received a FIN then no 3108 * more input can be expected, send ACK now. 3109 */ 3110 if (tp->t_flags & TF_NEEDSYN) 3111 tp->t_flags |= TF_DELACK; 3112 else 3113 tp->t_flags |= TF_ACKNOW; 3114 tp->rcv_nxt++; 3115 } 3116 switch (tp->t_state) { 3117 3118 /* 3119 * In SYN_RECEIVED and ESTABLISHED STATES 3120 * enter the CLOSE_WAIT state. 3121 */ 3122 case TCPS_SYN_RECEIVED: 3123 tp->t_starttime = ticks; 3124 /* FALLTHROUGH */ 3125 case TCPS_ESTABLISHED: 3126 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3127 break; 3128 3129 /* 3130 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3131 * enter the CLOSING state. 3132 */ 3133 case TCPS_FIN_WAIT_1: 3134 tcp_state_change(tp, TCPS_CLOSING); 3135 break; 3136 3137 /* 3138 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3139 * starting the time-wait timer, turning off the other 3140 * standard timers. 3141 */ 3142 case TCPS_FIN_WAIT_2: 3143 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3144 3145 tcp_twstart(tp); 3146 return; 3147 } 3148 } 3149 #ifdef TCPDEBUG 3150 if (so->so_options & SO_DEBUG) 3151 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3152 &tcp_savetcp, 0); 3153 #endif 3154 TCP_PROBE3(debug__input, tp, th, m); 3155 3156 /* 3157 * Return any desired output. 3158 */ 3159 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3160 (void) tp->t_fb->tfb_tcp_output(tp); 3161 3162 check_delack: 3163 INP_WLOCK_ASSERT(tp->t_inpcb); 3164 3165 if (tp->t_flags & TF_DELACK) { 3166 tp->t_flags &= ~TF_DELACK; 3167 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3168 } 3169 INP_WUNLOCK(tp->t_inpcb); 3170 return; 3171 3172 dropafterack: 3173 /* 3174 * Generate an ACK dropping incoming segment if it occupies 3175 * sequence space, where the ACK reflects our state. 3176 * 3177 * We can now skip the test for the RST flag since all 3178 * paths to this code happen after packets containing 3179 * RST have been dropped. 3180 * 3181 * In the SYN-RECEIVED state, don't send an ACK unless the 3182 * segment we received passes the SYN-RECEIVED ACK test. 3183 * If it fails send a RST. This breaks the loop in the 3184 * "LAND" DoS attack, and also prevents an ACK storm 3185 * between two listening ports that have been sent forged 3186 * SYN segments, each with the source address of the other. 3187 */ 3188 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3189 (SEQ_GT(tp->snd_una, th->th_ack) || 3190 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3191 rstreason = BANDLIM_RST_OPENPORT; 3192 goto dropwithreset; 3193 } 3194 #ifdef TCPDEBUG 3195 if (so->so_options & SO_DEBUG) 3196 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3197 &tcp_savetcp, 0); 3198 #endif 3199 TCP_PROBE3(debug__input, tp, th, m); 3200 tp->t_flags |= TF_ACKNOW; 3201 (void) tp->t_fb->tfb_tcp_output(tp); 3202 INP_WUNLOCK(tp->t_inpcb); 3203 m_freem(m); 3204 return; 3205 3206 dropwithreset: 3207 if (tp != NULL) { 3208 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3209 INP_WUNLOCK(tp->t_inpcb); 3210 } else 3211 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3212 return; 3213 3214 drop: 3215 /* 3216 * Drop space held by incoming segment and return. 3217 */ 3218 #ifdef TCPDEBUG 3219 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3220 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3221 &tcp_savetcp, 0); 3222 #endif 3223 TCP_PROBE3(debug__input, tp, th, m); 3224 if (tp != NULL) 3225 INP_WUNLOCK(tp->t_inpcb); 3226 m_freem(m); 3227 } 3228 3229 /* 3230 * Issue RST and make ACK acceptable to originator of segment. 3231 * The mbuf must still include the original packet header. 3232 * tp may be NULL. 3233 */ 3234 void 3235 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3236 int tlen, int rstreason) 3237 { 3238 #ifdef INET 3239 struct ip *ip; 3240 #endif 3241 #ifdef INET6 3242 struct ip6_hdr *ip6; 3243 #endif 3244 3245 if (tp != NULL) { 3246 INP_WLOCK_ASSERT(tp->t_inpcb); 3247 } 3248 3249 /* Don't bother if destination was broadcast/multicast. */ 3250 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3251 goto drop; 3252 #ifdef INET6 3253 if (mtod(m, struct ip *)->ip_v == 6) { 3254 ip6 = mtod(m, struct ip6_hdr *); 3255 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3256 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3257 goto drop; 3258 /* IPv6 anycast check is done at tcp6_input() */ 3259 } 3260 #endif 3261 #if defined(INET) && defined(INET6) 3262 else 3263 #endif 3264 #ifdef INET 3265 { 3266 ip = mtod(m, struct ip *); 3267 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3268 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3269 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3270 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3271 goto drop; 3272 } 3273 #endif 3274 3275 /* Perform bandwidth limiting. */ 3276 if (badport_bandlim(rstreason) < 0) 3277 goto drop; 3278 3279 /* tcp_respond consumes the mbuf chain. */ 3280 if (th->th_flags & TH_ACK) { 3281 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3282 th->th_ack, TH_RST); 3283 } else { 3284 if (th->th_flags & TH_SYN) 3285 tlen++; 3286 if (th->th_flags & TH_FIN) 3287 tlen++; 3288 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3289 (tcp_seq)0, TH_RST|TH_ACK); 3290 } 3291 return; 3292 drop: 3293 m_freem(m); 3294 } 3295 3296 /* 3297 * Parse TCP options and place in tcpopt. 3298 */ 3299 void 3300 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3301 { 3302 int opt, optlen; 3303 3304 to->to_flags = 0; 3305 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3306 opt = cp[0]; 3307 if (opt == TCPOPT_EOL) 3308 break; 3309 if (opt == TCPOPT_NOP) 3310 optlen = 1; 3311 else { 3312 if (cnt < 2) 3313 break; 3314 optlen = cp[1]; 3315 if (optlen < 2 || optlen > cnt) 3316 break; 3317 } 3318 switch (opt) { 3319 case TCPOPT_MAXSEG: 3320 if (optlen != TCPOLEN_MAXSEG) 3321 continue; 3322 if (!(flags & TO_SYN)) 3323 continue; 3324 to->to_flags |= TOF_MSS; 3325 bcopy((char *)cp + 2, 3326 (char *)&to->to_mss, sizeof(to->to_mss)); 3327 to->to_mss = ntohs(to->to_mss); 3328 break; 3329 case TCPOPT_WINDOW: 3330 if (optlen != TCPOLEN_WINDOW) 3331 continue; 3332 if (!(flags & TO_SYN)) 3333 continue; 3334 to->to_flags |= TOF_SCALE; 3335 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3336 break; 3337 case TCPOPT_TIMESTAMP: 3338 if (optlen != TCPOLEN_TIMESTAMP) 3339 continue; 3340 to->to_flags |= TOF_TS; 3341 bcopy((char *)cp + 2, 3342 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3343 to->to_tsval = ntohl(to->to_tsval); 3344 bcopy((char *)cp + 6, 3345 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3346 to->to_tsecr = ntohl(to->to_tsecr); 3347 break; 3348 case TCPOPT_SIGNATURE: 3349 /* 3350 * In order to reply to a host which has set the 3351 * TCP_SIGNATURE option in its initial SYN, we have 3352 * to record the fact that the option was observed 3353 * here for the syncache code to perform the correct 3354 * response. 3355 */ 3356 if (optlen != TCPOLEN_SIGNATURE) 3357 continue; 3358 to->to_flags |= TOF_SIGNATURE; 3359 to->to_signature = cp + 2; 3360 break; 3361 case TCPOPT_SACK_PERMITTED: 3362 if (optlen != TCPOLEN_SACK_PERMITTED) 3363 continue; 3364 if (!(flags & TO_SYN)) 3365 continue; 3366 if (!V_tcp_do_sack) 3367 continue; 3368 to->to_flags |= TOF_SACKPERM; 3369 break; 3370 case TCPOPT_SACK: 3371 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3372 continue; 3373 if (flags & TO_SYN) 3374 continue; 3375 to->to_flags |= TOF_SACK; 3376 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3377 to->to_sacks = cp + 2; 3378 TCPSTAT_INC(tcps_sack_rcv_blocks); 3379 break; 3380 case TCPOPT_FAST_OPEN: 3381 /* 3382 * Cookie length validation is performed by the 3383 * server side cookie checking code or the client 3384 * side cookie cache update code. 3385 */ 3386 if (!(flags & TO_SYN)) 3387 continue; 3388 if (!V_tcp_fastopen_client_enable && 3389 !V_tcp_fastopen_server_enable) 3390 continue; 3391 to->to_flags |= TOF_FASTOPEN; 3392 to->to_tfo_len = optlen - 2; 3393 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3394 break; 3395 default: 3396 continue; 3397 } 3398 } 3399 } 3400 3401 /* 3402 * Pull out of band byte out of a segment so 3403 * it doesn't appear in the user's data queue. 3404 * It is still reflected in the segment length for 3405 * sequencing purposes. 3406 */ 3407 void 3408 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3409 int off) 3410 { 3411 int cnt = off + th->th_urp - 1; 3412 3413 while (cnt >= 0) { 3414 if (m->m_len > cnt) { 3415 char *cp = mtod(m, caddr_t) + cnt; 3416 struct tcpcb *tp = sototcpcb(so); 3417 3418 INP_WLOCK_ASSERT(tp->t_inpcb); 3419 3420 tp->t_iobc = *cp; 3421 tp->t_oobflags |= TCPOOB_HAVEDATA; 3422 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3423 m->m_len--; 3424 if (m->m_flags & M_PKTHDR) 3425 m->m_pkthdr.len--; 3426 return; 3427 } 3428 cnt -= m->m_len; 3429 m = m->m_next; 3430 if (m == NULL) 3431 break; 3432 } 3433 panic("tcp_pulloutofband"); 3434 } 3435 3436 /* 3437 * Collect new round-trip time estimate 3438 * and update averages and current timeout. 3439 */ 3440 void 3441 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3442 { 3443 int delta; 3444 3445 INP_WLOCK_ASSERT(tp->t_inpcb); 3446 3447 TCPSTAT_INC(tcps_rttupdated); 3448 tp->t_rttupdated++; 3449 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3450 /* 3451 * srtt is stored as fixed point with 5 bits after the 3452 * binary point (i.e., scaled by 8). The following magic 3453 * is equivalent to the smoothing algorithm in rfc793 with 3454 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3455 * point). Adjust rtt to origin 0. 3456 */ 3457 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3458 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3459 3460 if ((tp->t_srtt += delta) <= 0) 3461 tp->t_srtt = 1; 3462 3463 /* 3464 * We accumulate a smoothed rtt variance (actually, a 3465 * smoothed mean difference), then set the retransmit 3466 * timer to smoothed rtt + 4 times the smoothed variance. 3467 * rttvar is stored as fixed point with 4 bits after the 3468 * binary point (scaled by 16). The following is 3469 * equivalent to rfc793 smoothing with an alpha of .75 3470 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3471 * rfc793's wired-in beta. 3472 */ 3473 if (delta < 0) 3474 delta = -delta; 3475 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3476 if ((tp->t_rttvar += delta) <= 0) 3477 tp->t_rttvar = 1; 3478 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3479 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3480 } else { 3481 /* 3482 * No rtt measurement yet - use the unsmoothed rtt. 3483 * Set the variance to half the rtt (so our first 3484 * retransmit happens at 3*rtt). 3485 */ 3486 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3487 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3488 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3489 } 3490 tp->t_rtttime = 0; 3491 tp->t_rxtshift = 0; 3492 3493 /* 3494 * the retransmit should happen at rtt + 4 * rttvar. 3495 * Because of the way we do the smoothing, srtt and rttvar 3496 * will each average +1/2 tick of bias. When we compute 3497 * the retransmit timer, we want 1/2 tick of rounding and 3498 * 1 extra tick because of +-1/2 tick uncertainty in the 3499 * firing of the timer. The bias will give us exactly the 3500 * 1.5 tick we need. But, because the bias is 3501 * statistical, we have to test that we don't drop below 3502 * the minimum feasible timer (which is 2 ticks). 3503 */ 3504 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3505 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3506 3507 /* 3508 * We received an ack for a packet that wasn't retransmitted; 3509 * it is probably safe to discard any error indications we've 3510 * received recently. This isn't quite right, but close enough 3511 * for now (a route might have failed after we sent a segment, 3512 * and the return path might not be symmetrical). 3513 */ 3514 tp->t_softerror = 0; 3515 } 3516 3517 /* 3518 * Determine a reasonable value for maxseg size. 3519 * If the route is known, check route for mtu. 3520 * If none, use an mss that can be handled on the outgoing interface 3521 * without forcing IP to fragment. If no route is found, route has no mtu, 3522 * or the destination isn't local, use a default, hopefully conservative 3523 * size (usually 512 or the default IP max size, but no more than the mtu 3524 * of the interface), as we can't discover anything about intervening 3525 * gateways or networks. We also initialize the congestion/slow start 3526 * window to be a single segment if the destination isn't local. 3527 * While looking at the routing entry, we also initialize other path-dependent 3528 * parameters from pre-set or cached values in the routing entry. 3529 * 3530 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3531 * IP options, e.g. IPSEC data, since length of this data may vary, and 3532 * thus it is calculated for every segment separately in tcp_output(). 3533 * 3534 * NOTE that this routine is only called when we process an incoming 3535 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3536 * settings are handled in tcp_mssopt(). 3537 */ 3538 void 3539 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3540 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3541 { 3542 int mss = 0; 3543 uint32_t maxmtu = 0; 3544 struct inpcb *inp = tp->t_inpcb; 3545 struct hc_metrics_lite metrics; 3546 #ifdef INET6 3547 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3548 size_t min_protoh = isipv6 ? 3549 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3550 sizeof (struct tcpiphdr); 3551 #else 3552 const size_t min_protoh = sizeof(struct tcpiphdr); 3553 #endif 3554 3555 INP_WLOCK_ASSERT(tp->t_inpcb); 3556 3557 if (mtuoffer != -1) { 3558 KASSERT(offer == -1, ("%s: conflict", __func__)); 3559 offer = mtuoffer - min_protoh; 3560 } 3561 3562 /* Initialize. */ 3563 #ifdef INET6 3564 if (isipv6) { 3565 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3566 tp->t_maxseg = V_tcp_v6mssdflt; 3567 } 3568 #endif 3569 #if defined(INET) && defined(INET6) 3570 else 3571 #endif 3572 #ifdef INET 3573 { 3574 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3575 tp->t_maxseg = V_tcp_mssdflt; 3576 } 3577 #endif 3578 3579 /* 3580 * No route to sender, stay with default mss and return. 3581 */ 3582 if (maxmtu == 0) { 3583 /* 3584 * In case we return early we need to initialize metrics 3585 * to a defined state as tcp_hc_get() would do for us 3586 * if there was no cache hit. 3587 */ 3588 if (metricptr != NULL) 3589 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3590 return; 3591 } 3592 3593 /* What have we got? */ 3594 switch (offer) { 3595 case 0: 3596 /* 3597 * Offer == 0 means that there was no MSS on the SYN 3598 * segment, in this case we use tcp_mssdflt as 3599 * already assigned to t_maxseg above. 3600 */ 3601 offer = tp->t_maxseg; 3602 break; 3603 3604 case -1: 3605 /* 3606 * Offer == -1 means that we didn't receive SYN yet. 3607 */ 3608 /* FALLTHROUGH */ 3609 3610 default: 3611 /* 3612 * Prevent DoS attack with too small MSS. Round up 3613 * to at least minmss. 3614 */ 3615 offer = max(offer, V_tcp_minmss); 3616 } 3617 3618 /* 3619 * rmx information is now retrieved from tcp_hostcache. 3620 */ 3621 tcp_hc_get(&inp->inp_inc, &metrics); 3622 if (metricptr != NULL) 3623 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3624 3625 /* 3626 * If there's a discovered mtu in tcp hostcache, use it. 3627 * Else, use the link mtu. 3628 */ 3629 if (metrics.rmx_mtu) 3630 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3631 else { 3632 #ifdef INET6 3633 if (isipv6) { 3634 mss = maxmtu - min_protoh; 3635 if (!V_path_mtu_discovery && 3636 !in6_localaddr(&inp->in6p_faddr)) 3637 mss = min(mss, V_tcp_v6mssdflt); 3638 } 3639 #endif 3640 #if defined(INET) && defined(INET6) 3641 else 3642 #endif 3643 #ifdef INET 3644 { 3645 mss = maxmtu - min_protoh; 3646 if (!V_path_mtu_discovery && 3647 !in_localaddr(inp->inp_faddr)) 3648 mss = min(mss, V_tcp_mssdflt); 3649 } 3650 #endif 3651 /* 3652 * XXX - The above conditional (mss = maxmtu - min_protoh) 3653 * probably violates the TCP spec. 3654 * The problem is that, since we don't know the 3655 * other end's MSS, we are supposed to use a conservative 3656 * default. But, if we do that, then MTU discovery will 3657 * never actually take place, because the conservative 3658 * default is much less than the MTUs typically seen 3659 * on the Internet today. For the moment, we'll sweep 3660 * this under the carpet. 3661 * 3662 * The conservative default might not actually be a problem 3663 * if the only case this occurs is when sending an initial 3664 * SYN with options and data to a host we've never talked 3665 * to before. Then, they will reply with an MSS value which 3666 * will get recorded and the new parameters should get 3667 * recomputed. For Further Study. 3668 */ 3669 } 3670 mss = min(mss, offer); 3671 3672 /* 3673 * Sanity check: make sure that maxseg will be large 3674 * enough to allow some data on segments even if the 3675 * all the option space is used (40bytes). Otherwise 3676 * funny things may happen in tcp_output. 3677 * 3678 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3679 */ 3680 mss = max(mss, 64); 3681 3682 tp->t_maxseg = mss; 3683 } 3684 3685 void 3686 tcp_mss(struct tcpcb *tp, int offer) 3687 { 3688 int mss; 3689 uint32_t bufsize; 3690 struct inpcb *inp; 3691 struct socket *so; 3692 struct hc_metrics_lite metrics; 3693 struct tcp_ifcap cap; 3694 3695 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3696 3697 bzero(&cap, sizeof(cap)); 3698 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3699 3700 mss = tp->t_maxseg; 3701 inp = tp->t_inpcb; 3702 3703 /* 3704 * If there's a pipesize, change the socket buffer to that size, 3705 * don't change if sb_hiwat is different than default (then it 3706 * has been changed on purpose with setsockopt). 3707 * Make the socket buffers an integral number of mss units; 3708 * if the mss is larger than the socket buffer, decrease the mss. 3709 */ 3710 so = inp->inp_socket; 3711 SOCKBUF_LOCK(&so->so_snd); 3712 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3713 bufsize = metrics.rmx_sendpipe; 3714 else 3715 bufsize = so->so_snd.sb_hiwat; 3716 if (bufsize < mss) 3717 mss = bufsize; 3718 else { 3719 bufsize = roundup(bufsize, mss); 3720 if (bufsize > sb_max) 3721 bufsize = sb_max; 3722 if (bufsize > so->so_snd.sb_hiwat) 3723 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3724 } 3725 SOCKBUF_UNLOCK(&so->so_snd); 3726 /* 3727 * Sanity check: make sure that maxseg will be large 3728 * enough to allow some data on segments even if the 3729 * all the option space is used (40bytes). Otherwise 3730 * funny things may happen in tcp_output. 3731 * 3732 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3733 */ 3734 tp->t_maxseg = max(mss, 64); 3735 3736 SOCKBUF_LOCK(&so->so_rcv); 3737 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3738 bufsize = metrics.rmx_recvpipe; 3739 else 3740 bufsize = so->so_rcv.sb_hiwat; 3741 if (bufsize > mss) { 3742 bufsize = roundup(bufsize, mss); 3743 if (bufsize > sb_max) 3744 bufsize = sb_max; 3745 if (bufsize > so->so_rcv.sb_hiwat) 3746 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3747 } 3748 SOCKBUF_UNLOCK(&so->so_rcv); 3749 3750 /* Check the interface for TSO capabilities. */ 3751 if (cap.ifcap & CSUM_TSO) { 3752 tp->t_flags |= TF_TSO; 3753 tp->t_tsomax = cap.tsomax; 3754 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3755 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3756 } 3757 } 3758 3759 /* 3760 * Determine the MSS option to send on an outgoing SYN. 3761 */ 3762 int 3763 tcp_mssopt(struct in_conninfo *inc) 3764 { 3765 int mss = 0; 3766 uint32_t thcmtu = 0; 3767 uint32_t maxmtu = 0; 3768 size_t min_protoh; 3769 3770 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3771 3772 #ifdef INET6 3773 if (inc->inc_flags & INC_ISIPV6) { 3774 mss = V_tcp_v6mssdflt; 3775 maxmtu = tcp_maxmtu6(inc, NULL); 3776 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3777 } 3778 #endif 3779 #if defined(INET) && defined(INET6) 3780 else 3781 #endif 3782 #ifdef INET 3783 { 3784 mss = V_tcp_mssdflt; 3785 maxmtu = tcp_maxmtu(inc, NULL); 3786 min_protoh = sizeof(struct tcpiphdr); 3787 } 3788 #endif 3789 #if defined(INET6) || defined(INET) 3790 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3791 #endif 3792 3793 if (maxmtu && thcmtu) 3794 mss = min(maxmtu, thcmtu) - min_protoh; 3795 else if (maxmtu || thcmtu) 3796 mss = max(maxmtu, thcmtu) - min_protoh; 3797 3798 return (mss); 3799 } 3800 3801 3802 /* 3803 * On a partial ack arrives, force the retransmission of the 3804 * next unacknowledged segment. Do not clear tp->t_dupacks. 3805 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3806 * be started again. 3807 */ 3808 void 3809 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3810 { 3811 tcp_seq onxt = tp->snd_nxt; 3812 uint32_t ocwnd = tp->snd_cwnd; 3813 u_int maxseg = tcp_maxseg(tp); 3814 3815 INP_WLOCK_ASSERT(tp->t_inpcb); 3816 3817 tcp_timer_activate(tp, TT_REXMT, 0); 3818 tp->t_rtttime = 0; 3819 tp->snd_nxt = th->th_ack; 3820 /* 3821 * Set snd_cwnd to one segment beyond acknowledged offset. 3822 * (tp->snd_una has not yet been updated when this function is called.) 3823 */ 3824 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3825 tp->t_flags |= TF_ACKNOW; 3826 (void) tp->t_fb->tfb_tcp_output(tp); 3827 tp->snd_cwnd = ocwnd; 3828 if (SEQ_GT(onxt, tp->snd_nxt)) 3829 tp->snd_nxt = onxt; 3830 /* 3831 * Partial window deflation. Relies on fact that tp->snd_una 3832 * not updated yet. 3833 */ 3834 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3835 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3836 else 3837 tp->snd_cwnd = 0; 3838 tp->snd_cwnd += maxseg; 3839 } 3840 3841 int 3842 tcp_compute_pipe(struct tcpcb *tp) 3843 { 3844 return (tp->snd_max - tp->snd_una + 3845 tp->sackhint.sack_bytes_rexmit - 3846 tp->sackhint.sacked_bytes); 3847 } 3848 3849 uint32_t 3850 tcp_compute_initwnd(uint32_t maxseg) 3851 { 3852 /* 3853 * Calculate the Initial Window, also used as Restart Window 3854 * 3855 * RFC5681 Section 3.1 specifies the default conservative values. 3856 * RFC3390 specifies slightly more aggressive values. 3857 * RFC6928 increases it to ten segments. 3858 * Support for user specified value for initial flight size. 3859 */ 3860 if (V_tcp_initcwnd_segments) 3861 return min(V_tcp_initcwnd_segments * maxseg, 3862 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 3863 else if (V_tcp_do_rfc3390) 3864 return min(4 * maxseg, max(2 * maxseg, 4380)); 3865 else { 3866 /* Per RFC5681 Section 3.1 */ 3867 if (maxseg > 2190) 3868 return (2 * maxseg); 3869 else if (maxseg > 1095) 3870 return (3 * maxseg); 3871 else 3872 return (4 * maxseg); 3873 } 3874 } 3875