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