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 #ifdef TCP_RFC7413 104 #include <netinet/tcp_fastopen.h> 105 #endif 106 #include <netinet/tcp.h> 107 #include <netinet/tcp_fsm.h> 108 #include <netinet/tcp_seq.h> 109 #include <netinet/tcp_timer.h> 110 #include <netinet/tcp_var.h> 111 #include <netinet6/tcp6_var.h> 112 #include <netinet/tcpip.h> 113 #include <netinet/cc/cc.h> 114 #ifdef TCPPCAP 115 #include <netinet/tcp_pcap.h> 116 #endif 117 #include <netinet/tcp_syncache.h> 118 #ifdef TCPDEBUG 119 #include <netinet/tcp_debug.h> 120 #endif /* TCPDEBUG */ 121 #ifdef TCP_OFFLOAD 122 #include <netinet/tcp_offload.h> 123 #endif 124 125 #include <netipsec/ipsec_support.h> 126 127 #include <machine/in_cksum.h> 128 129 #include <security/mac/mac_framework.h> 130 131 const int tcprexmtthresh = 3; 132 133 int tcp_log_in_vain = 0; 134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 135 &tcp_log_in_vain, 0, 136 "Log all incoming TCP segments to closed ports"); 137 138 VNET_DEFINE(int, blackhole) = 0; 139 #define V_blackhole VNET(blackhole) 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 141 &VNET_NAME(blackhole), 0, 142 "Do not send RST on segments to closed ports"); 143 144 VNET_DEFINE(int, tcp_delack_enabled) = 1; 145 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 146 &VNET_NAME(tcp_delack_enabled), 0, 147 "Delay ACK to try and piggyback it onto a data packet"); 148 149 VNET_DEFINE(int, drop_synfin) = 0; 150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 151 &VNET_NAME(drop_synfin), 0, 152 "Drop TCP packets with SYN+FIN set"); 153 154 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 156 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 157 "Use calculated pipe/in-flight bytes per RFC 6675"); 158 159 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 161 &VNET_NAME(tcp_do_rfc3042), 0, 162 "Enable RFC 3042 (Limited Transmit)"); 163 164 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 166 &VNET_NAME(tcp_do_rfc3390), 0, 167 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 168 169 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 171 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 172 "Slow-start flight size (initial congestion window) in number of segments"); 173 174 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 176 &VNET_NAME(tcp_do_rfc3465), 0, 177 "Enable RFC 3465 (Appropriate Byte Counting)"); 178 179 VNET_DEFINE(int, tcp_abc_l_var) = 2; 180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_abc_l_var), 2, 182 "Cap the max cwnd increment during slow-start to this number of segments"); 183 184 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 185 186 VNET_DEFINE(int, tcp_do_ecn) = 2; 187 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 188 &VNET_NAME(tcp_do_ecn), 0, 189 "TCP ECN support"); 190 191 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 192 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_ecn_maxretries), 0, 194 "Max retries before giving up on ECN"); 195 196 VNET_DEFINE(int, tcp_insecure_syn) = 0; 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_insecure_syn), 0, 199 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 200 201 VNET_DEFINE(int, tcp_insecure_rst) = 0; 202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 203 &VNET_NAME(tcp_insecure_rst), 0, 204 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 205 206 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 207 #define V_tcp_recvspace VNET(tcp_recvspace) 208 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 209 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 210 211 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 212 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(tcp_do_autorcvbuf), 0, 214 "Enable automatic receive buffer sizing"); 215 216 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(tcp_autorcvbuf_inc), 0, 219 "Incrementor step size of automatic receive buffer"); 220 221 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_autorcvbuf_max), 0, 224 "Max size of automatic receive buffer"); 225 226 VNET_DEFINE(struct inpcbhead, tcb); 227 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 228 VNET_DEFINE(struct inpcbinfo, tcbinfo); 229 230 /* 231 * TCP statistics are stored in an array of counter(9)s, which size matches 232 * size of struct tcpstat. TCP running connection count is a regular array. 233 */ 234 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 235 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 236 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 237 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 238 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 239 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 240 "TCP connection counts by TCP state"); 241 242 static void 243 tcp_vnet_init(const void *unused) 244 { 245 246 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 247 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 248 } 249 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 250 tcp_vnet_init, NULL); 251 252 #ifdef VIMAGE 253 static void 254 tcp_vnet_uninit(const void *unused) 255 { 256 257 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 258 VNET_PCPUSTAT_FREE(tcpstat); 259 } 260 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 261 tcp_vnet_uninit, NULL); 262 #endif /* VIMAGE */ 263 264 /* 265 * Kernel module interface for updating tcpstat. The argument is an index 266 * into tcpstat treated as an array. 267 */ 268 void 269 kmod_tcpstat_inc(int statnum) 270 { 271 272 counter_u64_add(VNET(tcpstat)[statnum], 1); 273 } 274 275 #ifdef TCP_HHOOK 276 /* 277 * Wrapper for the TCP established input helper hook. 278 */ 279 void 280 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 281 { 282 struct tcp_hhook_data hhook_data; 283 284 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 285 hhook_data.tp = tp; 286 hhook_data.th = th; 287 hhook_data.to = to; 288 289 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 290 tp->osd); 291 } 292 } 293 #endif 294 295 /* 296 * CC wrapper hook functions 297 */ 298 void 299 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 300 uint16_t type) 301 { 302 INP_WLOCK_ASSERT(tp->t_inpcb); 303 304 tp->ccv->nsegs = nsegs; 305 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 306 if (tp->snd_cwnd <= tp->snd_wnd) 307 tp->ccv->flags |= CCF_CWND_LIMITED; 308 else 309 tp->ccv->flags &= ~CCF_CWND_LIMITED; 310 311 if (type == CC_ACK) { 312 if (tp->snd_cwnd > tp->snd_ssthresh) { 313 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 314 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 315 if (tp->t_bytes_acked >= tp->snd_cwnd) { 316 tp->t_bytes_acked -= tp->snd_cwnd; 317 tp->ccv->flags |= CCF_ABC_SENTAWND; 318 } 319 } else { 320 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 321 tp->t_bytes_acked = 0; 322 } 323 } 324 325 if (CC_ALGO(tp)->ack_received != NULL) { 326 /* XXXLAS: Find a way to live without this */ 327 tp->ccv->curack = th->th_ack; 328 CC_ALGO(tp)->ack_received(tp->ccv, type); 329 } 330 } 331 332 void 333 cc_conn_init(struct tcpcb *tp) 334 { 335 struct hc_metrics_lite metrics; 336 struct inpcb *inp = tp->t_inpcb; 337 u_int maxseg; 338 int rtt; 339 340 INP_WLOCK_ASSERT(tp->t_inpcb); 341 342 tcp_hc_get(&inp->inp_inc, &metrics); 343 maxseg = tcp_maxseg(tp); 344 345 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 346 tp->t_srtt = rtt; 347 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 348 TCPSTAT_INC(tcps_usedrtt); 349 if (metrics.rmx_rttvar) { 350 tp->t_rttvar = metrics.rmx_rttvar; 351 TCPSTAT_INC(tcps_usedrttvar); 352 } else { 353 /* default variation is +- 1 rtt */ 354 tp->t_rttvar = 355 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 356 } 357 TCPT_RANGESET(tp->t_rxtcur, 358 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 359 tp->t_rttmin, TCPTV_REXMTMAX); 360 } 361 if (metrics.rmx_ssthresh) { 362 /* 363 * There's some sort of gateway or interface 364 * buffer limit on the path. Use this to set 365 * the slow start threshold, but set the 366 * threshold to no less than 2*mss. 367 */ 368 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 369 TCPSTAT_INC(tcps_usedssthresh); 370 } 371 372 /* 373 * Set the initial slow-start flight size. 374 * 375 * RFC5681 Section 3.1 specifies the default conservative values. 376 * RFC3390 specifies slightly more aggressive values. 377 * RFC6928 increases it to ten segments. 378 * Support for user specified value for initial flight size. 379 * 380 * If a SYN or SYN/ACK was lost and retransmitted, we have to 381 * reduce the initial CWND to one segment as congestion is likely 382 * requiring us to be cautious. 383 */ 384 if (tp->snd_cwnd == 1) 385 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 386 else if (V_tcp_initcwnd_segments) 387 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 388 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 389 else if (V_tcp_do_rfc3390) 390 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 391 else { 392 /* Per RFC5681 Section 3.1 */ 393 if (maxseg > 2190) 394 tp->snd_cwnd = 2 * maxseg; 395 else if (maxseg > 1095) 396 tp->snd_cwnd = 3 * maxseg; 397 else 398 tp->snd_cwnd = 4 * maxseg; 399 } 400 401 if (CC_ALGO(tp)->conn_init != NULL) 402 CC_ALGO(tp)->conn_init(tp->ccv); 403 } 404 405 void inline 406 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 407 { 408 u_int maxseg; 409 410 INP_WLOCK_ASSERT(tp->t_inpcb); 411 412 switch(type) { 413 case CC_NDUPACK: 414 if (!IN_FASTRECOVERY(tp->t_flags)) { 415 tp->snd_recover = tp->snd_max; 416 if (tp->t_flags & TF_ECN_PERMIT) 417 tp->t_flags |= TF_ECN_SND_CWR; 418 } 419 break; 420 case CC_ECN: 421 if (!IN_CONGRECOVERY(tp->t_flags)) { 422 TCPSTAT_INC(tcps_ecn_rcwnd); 423 tp->snd_recover = tp->snd_max; 424 if (tp->t_flags & TF_ECN_PERMIT) 425 tp->t_flags |= TF_ECN_SND_CWR; 426 } 427 break; 428 case CC_RTO: 429 maxseg = tcp_maxseg(tp); 430 tp->t_dupacks = 0; 431 tp->t_bytes_acked = 0; 432 EXIT_RECOVERY(tp->t_flags); 433 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 434 maxseg) * maxseg; 435 tp->snd_cwnd = maxseg; 436 break; 437 case CC_RTO_ERR: 438 TCPSTAT_INC(tcps_sndrexmitbad); 439 /* RTO was unnecessary, so reset everything. */ 440 tp->snd_cwnd = tp->snd_cwnd_prev; 441 tp->snd_ssthresh = tp->snd_ssthresh_prev; 442 tp->snd_recover = tp->snd_recover_prev; 443 if (tp->t_flags & TF_WASFRECOVERY) 444 ENTER_FASTRECOVERY(tp->t_flags); 445 if (tp->t_flags & TF_WASCRECOVERY) 446 ENTER_CONGRECOVERY(tp->t_flags); 447 tp->snd_nxt = tp->snd_max; 448 tp->t_flags &= ~TF_PREVVALID; 449 tp->t_badrxtwin = 0; 450 break; 451 } 452 453 if (CC_ALGO(tp)->cong_signal != NULL) { 454 if (th != NULL) 455 tp->ccv->curack = th->th_ack; 456 CC_ALGO(tp)->cong_signal(tp->ccv, type); 457 } 458 } 459 460 void inline 461 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 462 { 463 INP_WLOCK_ASSERT(tp->t_inpcb); 464 465 /* XXXLAS: KASSERT that we're in recovery? */ 466 467 if (CC_ALGO(tp)->post_recovery != NULL) { 468 tp->ccv->curack = th->th_ack; 469 CC_ALGO(tp)->post_recovery(tp->ccv); 470 } 471 /* XXXLAS: EXIT_RECOVERY ? */ 472 tp->t_bytes_acked = 0; 473 } 474 475 /* 476 * Indicate whether this ack should be delayed. We can delay the ack if 477 * following conditions are met: 478 * - There is no delayed ack timer in progress. 479 * - Our last ack wasn't a 0-sized window. We never want to delay 480 * the ack that opens up a 0-sized window. 481 * - LRO wasn't used for this segment. We make sure by checking that the 482 * segment size is not larger than the MSS. 483 */ 484 #define DELAY_ACK(tp, tlen) \ 485 ((!tcp_timer_active(tp, TT_DELACK) && \ 486 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 487 (tlen <= tp->t_maxseg) && \ 488 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 489 490 static void inline 491 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 492 { 493 INP_WLOCK_ASSERT(tp->t_inpcb); 494 495 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 496 switch (iptos & IPTOS_ECN_MASK) { 497 case IPTOS_ECN_CE: 498 tp->ccv->flags |= CCF_IPHDR_CE; 499 break; 500 case IPTOS_ECN_ECT0: 501 tp->ccv->flags &= ~CCF_IPHDR_CE; 502 break; 503 case IPTOS_ECN_ECT1: 504 tp->ccv->flags &= ~CCF_IPHDR_CE; 505 break; 506 } 507 508 if (th->th_flags & TH_CWR) 509 tp->ccv->flags |= CCF_TCPHDR_CWR; 510 else 511 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 512 513 if (tp->t_flags & TF_DELACK) 514 tp->ccv->flags |= CCF_DELACK; 515 else 516 tp->ccv->flags &= ~CCF_DELACK; 517 518 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 519 520 if (tp->ccv->flags & CCF_ACKNOW) 521 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 522 } 523 } 524 525 /* 526 * TCP input handling is split into multiple parts: 527 * tcp6_input is a thin wrapper around tcp_input for the extended 528 * ip6_protox[] call format in ip6_input 529 * tcp_input handles primary segment validation, inpcb lookup and 530 * SYN processing on listen sockets 531 * tcp_do_segment processes the ACK and text of the segment for 532 * establishing, established and closing connections 533 */ 534 #ifdef INET6 535 int 536 tcp6_input(struct mbuf **mp, int *offp, int proto) 537 { 538 struct mbuf *m = *mp; 539 struct in6_ifaddr *ia6; 540 struct ip6_hdr *ip6; 541 542 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 543 544 /* 545 * draft-itojun-ipv6-tcp-to-anycast 546 * better place to put this in? 547 */ 548 ip6 = mtod(m, struct ip6_hdr *); 549 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 550 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 551 struct ip6_hdr *ip6; 552 553 ifa_free(&ia6->ia_ifa); 554 ip6 = mtod(m, struct ip6_hdr *); 555 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 556 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 557 return (IPPROTO_DONE); 558 } 559 if (ia6) 560 ifa_free(&ia6->ia_ifa); 561 562 return (tcp_input(mp, offp, proto)); 563 } 564 #endif /* INET6 */ 565 566 int 567 tcp_input(struct mbuf **mp, int *offp, int proto) 568 { 569 struct mbuf *m = *mp; 570 struct tcphdr *th = NULL; 571 struct ip *ip = NULL; 572 struct inpcb *inp = NULL; 573 struct tcpcb *tp = NULL; 574 struct socket *so = NULL; 575 u_char *optp = NULL; 576 int off0; 577 int optlen = 0; 578 #ifdef INET 579 int len; 580 #endif 581 int tlen = 0, off; 582 int drop_hdrlen; 583 int thflags; 584 int rstreason = 0; /* For badport_bandlim accounting purposes */ 585 uint8_t iptos; 586 struct m_tag *fwd_tag = NULL; 587 #ifdef INET6 588 struct ip6_hdr *ip6 = NULL; 589 int isipv6; 590 #else 591 const void *ip6 = NULL; 592 #endif /* INET6 */ 593 struct tcpopt to; /* options in this segment */ 594 char *s = NULL; /* address and port logging */ 595 int ti_locked; 596 #ifdef TCPDEBUG 597 /* 598 * The size of tcp_saveipgen must be the size of the max ip header, 599 * now IPv6. 600 */ 601 u_char tcp_saveipgen[IP6_HDR_LEN]; 602 struct tcphdr tcp_savetcp; 603 short ostate = 0; 604 #endif 605 606 #ifdef INET6 607 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 608 #endif 609 610 off0 = *offp; 611 m = *mp; 612 *mp = NULL; 613 to.to_flags = 0; 614 TCPSTAT_INC(tcps_rcvtotal); 615 616 #ifdef INET6 617 if (isipv6) { 618 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 619 620 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 621 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 622 if (m == NULL) { 623 TCPSTAT_INC(tcps_rcvshort); 624 return (IPPROTO_DONE); 625 } 626 } 627 628 ip6 = mtod(m, struct ip6_hdr *); 629 th = (struct tcphdr *)((caddr_t)ip6 + off0); 630 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 631 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 632 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 633 th->th_sum = m->m_pkthdr.csum_data; 634 else 635 th->th_sum = in6_cksum_pseudo(ip6, tlen, 636 IPPROTO_TCP, m->m_pkthdr.csum_data); 637 th->th_sum ^= 0xffff; 638 } else 639 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 640 if (th->th_sum) { 641 TCPSTAT_INC(tcps_rcvbadsum); 642 goto drop; 643 } 644 645 /* 646 * Be proactive about unspecified IPv6 address in source. 647 * As we use all-zero to indicate unbounded/unconnected pcb, 648 * unspecified IPv6 address can be used to confuse us. 649 * 650 * Note that packets with unspecified IPv6 destination is 651 * already dropped in ip6_input. 652 */ 653 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 654 /* XXX stat */ 655 goto drop; 656 } 657 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 658 } 659 #endif 660 #if defined(INET) && defined(INET6) 661 else 662 #endif 663 #ifdef INET 664 { 665 /* 666 * Get IP and TCP header together in first mbuf. 667 * Note: IP leaves IP header in first mbuf. 668 */ 669 if (off0 > sizeof (struct ip)) { 670 ip_stripoptions(m); 671 off0 = sizeof(struct ip); 672 } 673 if (m->m_len < sizeof (struct tcpiphdr)) { 674 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 675 == NULL) { 676 TCPSTAT_INC(tcps_rcvshort); 677 return (IPPROTO_DONE); 678 } 679 } 680 ip = mtod(m, struct ip *); 681 th = (struct tcphdr *)((caddr_t)ip + off0); 682 tlen = ntohs(ip->ip_len) - off0; 683 684 iptos = ip->ip_tos; 685 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 686 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 687 th->th_sum = m->m_pkthdr.csum_data; 688 else 689 th->th_sum = in_pseudo(ip->ip_src.s_addr, 690 ip->ip_dst.s_addr, 691 htonl(m->m_pkthdr.csum_data + tlen + 692 IPPROTO_TCP)); 693 th->th_sum ^= 0xffff; 694 } else { 695 struct ipovly *ipov = (struct ipovly *)ip; 696 697 /* 698 * Checksum extended TCP header and data. 699 */ 700 len = off0 + tlen; 701 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 702 ipov->ih_len = htons(tlen); 703 th->th_sum = in_cksum(m, len); 704 /* Reset length for SDT probes. */ 705 ip->ip_len = htons(len); 706 /* Reset TOS bits */ 707 ip->ip_tos = iptos; 708 /* Re-initialization for later version check */ 709 ip->ip_v = IPVERSION; 710 ip->ip_hl = off0 >> 2; 711 } 712 713 if (th->th_sum) { 714 TCPSTAT_INC(tcps_rcvbadsum); 715 goto drop; 716 } 717 } 718 #endif /* INET */ 719 720 /* 721 * Check that TCP offset makes sense, 722 * pull out TCP options and adjust length. XXX 723 */ 724 off = th->th_off << 2; 725 if (off < sizeof (struct tcphdr) || off > tlen) { 726 TCPSTAT_INC(tcps_rcvbadoff); 727 goto drop; 728 } 729 tlen -= off; /* tlen is used instead of ti->ti_len */ 730 if (off > sizeof (struct tcphdr)) { 731 #ifdef INET6 732 if (isipv6) { 733 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 734 ip6 = mtod(m, struct ip6_hdr *); 735 th = (struct tcphdr *)((caddr_t)ip6 + off0); 736 } 737 #endif 738 #if defined(INET) && defined(INET6) 739 else 740 #endif 741 #ifdef INET 742 { 743 if (m->m_len < sizeof(struct ip) + off) { 744 if ((m = m_pullup(m, sizeof (struct ip) + off)) 745 == NULL) { 746 TCPSTAT_INC(tcps_rcvshort); 747 return (IPPROTO_DONE); 748 } 749 ip = mtod(m, struct ip *); 750 th = (struct tcphdr *)((caddr_t)ip + off0); 751 } 752 } 753 #endif 754 optlen = off - sizeof (struct tcphdr); 755 optp = (u_char *)(th + 1); 756 } 757 thflags = th->th_flags; 758 759 /* 760 * Convert TCP protocol specific fields to host format. 761 */ 762 tcp_fields_to_host(th); 763 764 /* 765 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 766 */ 767 drop_hdrlen = off0 + off; 768 769 /* 770 * Locate pcb for segment; if we're likely to add or remove a 771 * connection then first acquire pcbinfo lock. There are three cases 772 * where we might discover later we need a write lock despite the 773 * flags: ACKs moving a connection out of the syncache, ACKs for a 774 * connection in TIMEWAIT and SYNs not targeting a listening socket. 775 */ 776 if ((thflags & (TH_FIN | TH_RST)) != 0) { 777 INP_INFO_RLOCK(&V_tcbinfo); 778 ti_locked = TI_RLOCKED; 779 } else 780 ti_locked = TI_UNLOCKED; 781 782 /* 783 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 784 */ 785 if ( 786 #ifdef INET6 787 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 788 #ifdef INET 789 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 790 #endif 791 #endif 792 #if defined(INET) && !defined(INET6) 793 (m->m_flags & M_IP_NEXTHOP) 794 #endif 795 ) 796 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 797 798 findpcb: 799 #ifdef INVARIANTS 800 if (ti_locked == TI_RLOCKED) { 801 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 802 } else { 803 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 804 } 805 #endif 806 #ifdef INET6 807 if (isipv6 && fwd_tag != NULL) { 808 struct sockaddr_in6 *next_hop6; 809 810 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 811 /* 812 * Transparently forwarded. Pretend to be the destination. 813 * Already got one like this? 814 */ 815 inp = in6_pcblookup_mbuf(&V_tcbinfo, 816 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 817 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 818 if (!inp) { 819 /* 820 * It's new. Try to find the ambushing socket. 821 * Because we've rewritten the destination address, 822 * any hardware-generated hash is ignored. 823 */ 824 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 825 th->th_sport, &next_hop6->sin6_addr, 826 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 827 th->th_dport, INPLOOKUP_WILDCARD | 828 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 829 } 830 } else if (isipv6) { 831 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 832 th->th_sport, &ip6->ip6_dst, th->th_dport, 833 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 834 m->m_pkthdr.rcvif, m); 835 } 836 #endif /* INET6 */ 837 #if defined(INET6) && defined(INET) 838 else 839 #endif 840 #ifdef INET 841 if (fwd_tag != NULL) { 842 struct sockaddr_in *next_hop; 843 844 next_hop = (struct sockaddr_in *)(fwd_tag+1); 845 /* 846 * Transparently forwarded. Pretend to be the destination. 847 * already got one like this? 848 */ 849 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 850 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 851 m->m_pkthdr.rcvif, m); 852 if (!inp) { 853 /* 854 * It's new. Try to find the ambushing socket. 855 * Because we've rewritten the destination address, 856 * any hardware-generated hash is ignored. 857 */ 858 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 859 th->th_sport, next_hop->sin_addr, 860 next_hop->sin_port ? ntohs(next_hop->sin_port) : 861 th->th_dport, INPLOOKUP_WILDCARD | 862 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 863 } 864 } else 865 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 866 th->th_sport, ip->ip_dst, th->th_dport, 867 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 868 m->m_pkthdr.rcvif, m); 869 #endif /* INET */ 870 871 /* 872 * If the INPCB does not exist then all data in the incoming 873 * segment is discarded and an appropriate RST is sent back. 874 * XXX MRT Send RST using which routing table? 875 */ 876 if (inp == NULL) { 877 /* 878 * Log communication attempts to ports that are not 879 * in use. 880 */ 881 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 882 tcp_log_in_vain == 2) { 883 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 884 log(LOG_INFO, "%s; %s: Connection attempt " 885 "to closed port\n", s, __func__); 886 } 887 /* 888 * When blackholing do not respond with a RST but 889 * completely ignore the segment and drop it. 890 */ 891 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 892 V_blackhole == 2) 893 goto dropunlock; 894 895 rstreason = BANDLIM_RST_CLOSEDPORT; 896 goto dropwithreset; 897 } 898 INP_WLOCK_ASSERT(inp); 899 /* 900 * While waiting for inp lock during the lookup, another thread 901 * can have dropped the inpcb, in which case we need to loop back 902 * and try to find a new inpcb to deliver to. 903 */ 904 if (inp->inp_flags & INP_DROPPED) { 905 INP_WUNLOCK(inp); 906 inp = NULL; 907 goto findpcb; 908 } 909 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 910 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 911 ((inp->inp_socket == NULL) || 912 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 913 inp->inp_flowid = m->m_pkthdr.flowid; 914 inp->inp_flowtype = M_HASHTYPE_GET(m); 915 } 916 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 917 #ifdef INET6 918 if (isipv6 && IPSEC_ENABLED(ipv6) && 919 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 920 goto dropunlock; 921 } 922 #ifdef INET 923 else 924 #endif 925 #endif /* INET6 */ 926 #ifdef INET 927 if (IPSEC_ENABLED(ipv4) && 928 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 929 goto dropunlock; 930 } 931 #endif /* INET */ 932 #endif /* IPSEC */ 933 934 /* 935 * Check the minimum TTL for socket. 936 */ 937 if (inp->inp_ip_minttl != 0) { 938 #ifdef INET6 939 if (isipv6) { 940 if (inp->inp_ip_minttl > ip6->ip6_hlim) 941 goto dropunlock; 942 } else 943 #endif 944 if (inp->inp_ip_minttl > ip->ip_ttl) 945 goto dropunlock; 946 } 947 948 /* 949 * A previous connection in TIMEWAIT state is supposed to catch stray 950 * or duplicate segments arriving late. If this segment was a 951 * legitimate new connection attempt, the old INPCB gets removed and 952 * we can try again to find a listening socket. 953 * 954 * At this point, due to earlier optimism, we may hold only an inpcb 955 * lock, and not the inpcbinfo write lock. If so, we need to try to 956 * acquire it, or if that fails, acquire a reference on the inpcb, 957 * drop all locks, acquire a global write lock, and then re-acquire 958 * the inpcb lock. We may at that point discover that another thread 959 * has tried to free the inpcb, in which case we need to loop back 960 * and try to find a new inpcb to deliver to. 961 * 962 * XXXRW: It may be time to rethink timewait locking. 963 */ 964 relocked: 965 if (inp->inp_flags & INP_TIMEWAIT) { 966 if (ti_locked == TI_UNLOCKED) { 967 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 968 in_pcbref(inp); 969 INP_WUNLOCK(inp); 970 INP_INFO_RLOCK(&V_tcbinfo); 971 ti_locked = TI_RLOCKED; 972 INP_WLOCK(inp); 973 if (in_pcbrele_wlocked(inp)) { 974 inp = NULL; 975 goto findpcb; 976 } else if (inp->inp_flags & INP_DROPPED) { 977 INP_WUNLOCK(inp); 978 inp = NULL; 979 goto findpcb; 980 } 981 } else 982 ti_locked = TI_RLOCKED; 983 } 984 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 985 986 if (thflags & TH_SYN) 987 tcp_dooptions(&to, optp, optlen, TO_SYN); 988 /* 989 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 990 */ 991 if (tcp_twcheck(inp, &to, th, m, tlen)) 992 goto findpcb; 993 INP_INFO_RUNLOCK(&V_tcbinfo); 994 return (IPPROTO_DONE); 995 } 996 /* 997 * The TCPCB may no longer exist if the connection is winding 998 * down or it is in the CLOSED state. Either way we drop the 999 * segment and send an appropriate response. 1000 */ 1001 tp = intotcpcb(inp); 1002 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1003 rstreason = BANDLIM_RST_CLOSEDPORT; 1004 goto dropwithreset; 1005 } 1006 1007 #ifdef TCP_OFFLOAD 1008 if (tp->t_flags & TF_TOE) { 1009 tcp_offload_input(tp, m); 1010 m = NULL; /* consumed by the TOE driver */ 1011 goto dropunlock; 1012 } 1013 #endif 1014 1015 /* 1016 * We've identified a valid inpcb, but it could be that we need an 1017 * inpcbinfo write lock but don't hold it. In this case, attempt to 1018 * acquire using the same strategy as the TIMEWAIT case above. If we 1019 * relock, we have to jump back to 'relocked' as the connection might 1020 * now be in TIMEWAIT. 1021 */ 1022 #ifdef INVARIANTS 1023 if ((thflags & (TH_FIN | TH_RST)) != 0) 1024 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1025 #endif 1026 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1027 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1028 !IS_FASTOPEN(tp->t_flags)))) { 1029 if (ti_locked == TI_UNLOCKED) { 1030 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1031 in_pcbref(inp); 1032 INP_WUNLOCK(inp); 1033 INP_INFO_RLOCK(&V_tcbinfo); 1034 ti_locked = TI_RLOCKED; 1035 INP_WLOCK(inp); 1036 if (in_pcbrele_wlocked(inp)) { 1037 inp = NULL; 1038 goto findpcb; 1039 } else if (inp->inp_flags & INP_DROPPED) { 1040 INP_WUNLOCK(inp); 1041 inp = NULL; 1042 goto findpcb; 1043 } 1044 goto relocked; 1045 } else 1046 ti_locked = TI_RLOCKED; 1047 } 1048 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1049 } 1050 1051 #ifdef MAC 1052 INP_WLOCK_ASSERT(inp); 1053 if (mac_inpcb_check_deliver(inp, m)) 1054 goto dropunlock; 1055 #endif 1056 so = inp->inp_socket; 1057 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1058 #ifdef TCPDEBUG 1059 if (so->so_options & SO_DEBUG) { 1060 ostate = tp->t_state; 1061 #ifdef INET6 1062 if (isipv6) { 1063 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1064 } else 1065 #endif 1066 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1067 tcp_savetcp = *th; 1068 } 1069 #endif /* TCPDEBUG */ 1070 /* 1071 * When the socket is accepting connections (the INPCB is in LISTEN 1072 * state) we look into the SYN cache if this is a new connection 1073 * attempt or the completion of a previous one. 1074 */ 1075 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1076 ("%s: so accepting but tp %p not listening", __func__, tp)); 1077 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1078 struct in_conninfo inc; 1079 1080 bzero(&inc, sizeof(inc)); 1081 #ifdef INET6 1082 if (isipv6) { 1083 inc.inc_flags |= INC_ISIPV6; 1084 inc.inc6_faddr = ip6->ip6_src; 1085 inc.inc6_laddr = ip6->ip6_dst; 1086 } else 1087 #endif 1088 { 1089 inc.inc_faddr = ip->ip_src; 1090 inc.inc_laddr = ip->ip_dst; 1091 } 1092 inc.inc_fport = th->th_sport; 1093 inc.inc_lport = th->th_dport; 1094 inc.inc_fibnum = so->so_fibnum; 1095 1096 /* 1097 * Check for an existing connection attempt in syncache if 1098 * the flag is only ACK. A successful lookup creates a new 1099 * socket appended to the listen queue in SYN_RECEIVED state. 1100 */ 1101 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1102 1103 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1104 /* 1105 * Parse the TCP options here because 1106 * syncookies need access to the reflected 1107 * timestamp. 1108 */ 1109 tcp_dooptions(&to, optp, optlen, 0); 1110 /* 1111 * NB: syncache_expand() doesn't unlock 1112 * inp and tcpinfo locks. 1113 */ 1114 rstreason = syncache_expand(&inc, &to, th, &so, m); 1115 if (rstreason < 0) { 1116 /* 1117 * A failing TCP MD5 signature comparison 1118 * must result in the segment being dropped 1119 * and must not produce any response back 1120 * to the sender. 1121 */ 1122 goto dropunlock; 1123 } else if (rstreason == 0) { 1124 /* 1125 * No syncache entry or ACK was not 1126 * for our SYN/ACK. Send a RST. 1127 * NB: syncache did its own logging 1128 * of the failure cause. 1129 */ 1130 rstreason = BANDLIM_RST_OPENPORT; 1131 goto dropwithreset; 1132 } 1133 #ifdef TCP_RFC7413 1134 tfo_socket_result: 1135 #endif 1136 if (so == NULL) { 1137 /* 1138 * We completed the 3-way handshake 1139 * but could not allocate a socket 1140 * either due to memory shortage, 1141 * listen queue length limits or 1142 * global socket limits. Send RST 1143 * or wait and have the remote end 1144 * retransmit the ACK for another 1145 * try. 1146 */ 1147 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1148 log(LOG_DEBUG, "%s; %s: Listen socket: " 1149 "Socket allocation failed due to " 1150 "limits or memory shortage, %s\n", 1151 s, __func__, 1152 V_tcp_sc_rst_sock_fail ? 1153 "sending RST" : "try again"); 1154 if (V_tcp_sc_rst_sock_fail) { 1155 rstreason = BANDLIM_UNLIMITED; 1156 goto dropwithreset; 1157 } else 1158 goto dropunlock; 1159 } 1160 /* 1161 * Socket is created in state SYN_RECEIVED. 1162 * Unlock the listen socket, lock the newly 1163 * created socket and update the tp variable. 1164 */ 1165 INP_WUNLOCK(inp); /* listen socket */ 1166 inp = sotoinpcb(so); 1167 /* 1168 * New connection inpcb is already locked by 1169 * syncache_expand(). 1170 */ 1171 INP_WLOCK_ASSERT(inp); 1172 tp = intotcpcb(inp); 1173 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1174 ("%s: ", __func__)); 1175 /* 1176 * Process the segment and the data it 1177 * contains. tcp_do_segment() consumes 1178 * the mbuf chain and unlocks the inpcb. 1179 */ 1180 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1181 iptos, ti_locked); 1182 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1183 return (IPPROTO_DONE); 1184 } 1185 /* 1186 * Segment flag validation for new connection attempts: 1187 * 1188 * Our (SYN|ACK) response was rejected. 1189 * Check with syncache and remove entry to prevent 1190 * retransmits. 1191 * 1192 * NB: syncache_chkrst does its own logging of failure 1193 * causes. 1194 */ 1195 if (thflags & TH_RST) { 1196 syncache_chkrst(&inc, th); 1197 goto dropunlock; 1198 } 1199 /* 1200 * We can't do anything without SYN. 1201 */ 1202 if ((thflags & TH_SYN) == 0) { 1203 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1204 log(LOG_DEBUG, "%s; %s: Listen socket: " 1205 "SYN is missing, segment ignored\n", 1206 s, __func__); 1207 TCPSTAT_INC(tcps_badsyn); 1208 goto dropunlock; 1209 } 1210 /* 1211 * (SYN|ACK) is bogus on a listen socket. 1212 */ 1213 if (thflags & TH_ACK) { 1214 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1215 log(LOG_DEBUG, "%s; %s: Listen socket: " 1216 "SYN|ACK invalid, segment rejected\n", 1217 s, __func__); 1218 syncache_badack(&inc); /* XXX: Not needed! */ 1219 TCPSTAT_INC(tcps_badsyn); 1220 rstreason = BANDLIM_RST_OPENPORT; 1221 goto dropwithreset; 1222 } 1223 /* 1224 * If the drop_synfin option is enabled, drop all 1225 * segments with both the SYN and FIN bits set. 1226 * This prevents e.g. nmap from identifying the 1227 * TCP/IP stack. 1228 * XXX: Poor reasoning. nmap has other methods 1229 * and is constantly refining its stack detection 1230 * strategies. 1231 * XXX: This is a violation of the TCP specification 1232 * and was used by RFC1644. 1233 */ 1234 if ((thflags & TH_FIN) && V_drop_synfin) { 1235 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1236 log(LOG_DEBUG, "%s; %s: Listen socket: " 1237 "SYN|FIN segment ignored (based on " 1238 "sysctl setting)\n", s, __func__); 1239 TCPSTAT_INC(tcps_badsyn); 1240 goto dropunlock; 1241 } 1242 /* 1243 * Segment's flags are (SYN) or (SYN|FIN). 1244 * 1245 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1246 * as they do not affect the state of the TCP FSM. 1247 * The data pointed to by TH_URG and th_urp is ignored. 1248 */ 1249 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1250 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1251 KASSERT(thflags & (TH_SYN), 1252 ("%s: Listen socket: TH_SYN not set", __func__)); 1253 #ifdef INET6 1254 /* 1255 * If deprecated address is forbidden, 1256 * we do not accept SYN to deprecated interface 1257 * address to prevent any new inbound connection from 1258 * getting established. 1259 * When we do not accept SYN, we send a TCP RST, 1260 * with deprecated source address (instead of dropping 1261 * it). We compromise it as it is much better for peer 1262 * to send a RST, and RST will be the final packet 1263 * for the exchange. 1264 * 1265 * If we do not forbid deprecated addresses, we accept 1266 * the SYN packet. RFC2462 does not suggest dropping 1267 * SYN in this case. 1268 * If we decipher RFC2462 5.5.4, it says like this: 1269 * 1. use of deprecated addr with existing 1270 * communication is okay - "SHOULD continue to be 1271 * used" 1272 * 2. use of it with new communication: 1273 * (2a) "SHOULD NOT be used if alternate address 1274 * with sufficient scope is available" 1275 * (2b) nothing mentioned otherwise. 1276 * Here we fall into (2b) case as we have no choice in 1277 * our source address selection - we must obey the peer. 1278 * 1279 * The wording in RFC2462 is confusing, and there are 1280 * multiple description text for deprecated address 1281 * handling - worse, they are not exactly the same. 1282 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1283 */ 1284 if (isipv6 && !V_ip6_use_deprecated) { 1285 struct in6_ifaddr *ia6; 1286 1287 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1288 if (ia6 != NULL && 1289 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1290 ifa_free(&ia6->ia_ifa); 1291 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1292 log(LOG_DEBUG, "%s; %s: Listen socket: " 1293 "Connection attempt to deprecated " 1294 "IPv6 address rejected\n", 1295 s, __func__); 1296 rstreason = BANDLIM_RST_OPENPORT; 1297 goto dropwithreset; 1298 } 1299 if (ia6) 1300 ifa_free(&ia6->ia_ifa); 1301 } 1302 #endif /* INET6 */ 1303 /* 1304 * Basic sanity checks on incoming SYN requests: 1305 * Don't respond if the destination is a link layer 1306 * broadcast according to RFC1122 4.2.3.10, p. 104. 1307 * If it is from this socket it must be forged. 1308 * Don't respond if the source or destination is a 1309 * global or subnet broad- or multicast address. 1310 * Note that it is quite possible to receive unicast 1311 * link-layer packets with a broadcast IP address. Use 1312 * in_broadcast() to find them. 1313 */ 1314 if (m->m_flags & (M_BCAST|M_MCAST)) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from broad- or multicast " 1318 "link layer address ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 #ifdef INET6 1322 if (isipv6) { 1323 if (th->th_dport == th->th_sport && 1324 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1326 log(LOG_DEBUG, "%s; %s: Listen socket: " 1327 "Connection attempt to/from self " 1328 "ignored\n", s, __func__); 1329 goto dropunlock; 1330 } 1331 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1332 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1333 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1334 log(LOG_DEBUG, "%s; %s: Listen socket: " 1335 "Connection attempt from/to multicast " 1336 "address ignored\n", s, __func__); 1337 goto dropunlock; 1338 } 1339 } 1340 #endif 1341 #if defined(INET) && defined(INET6) 1342 else 1343 #endif 1344 #ifdef INET 1345 { 1346 if (th->th_dport == th->th_sport && 1347 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1348 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1349 log(LOG_DEBUG, "%s; %s: Listen socket: " 1350 "Connection attempt from/to self " 1351 "ignored\n", s, __func__); 1352 goto dropunlock; 1353 } 1354 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1355 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1356 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1357 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1358 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1359 log(LOG_DEBUG, "%s; %s: Listen socket: " 1360 "Connection attempt from/to broad- " 1361 "or multicast address ignored\n", 1362 s, __func__); 1363 goto dropunlock; 1364 } 1365 } 1366 #endif 1367 /* 1368 * SYN appears to be valid. Create compressed TCP state 1369 * for syncache. 1370 */ 1371 #ifdef TCPDEBUG 1372 if (so->so_options & SO_DEBUG) 1373 tcp_trace(TA_INPUT, ostate, tp, 1374 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1375 #endif 1376 TCP_PROBE3(debug__input, tp, th, m); 1377 tcp_dooptions(&to, optp, optlen, TO_SYN); 1378 #ifdef TCP_RFC7413 1379 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1380 goto tfo_socket_result; 1381 #else 1382 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1383 #endif 1384 /* 1385 * Entry added to syncache and mbuf consumed. 1386 * Only the listen socket is unlocked by syncache_add(). 1387 */ 1388 if (ti_locked == TI_RLOCKED) { 1389 INP_INFO_RUNLOCK(&V_tcbinfo); 1390 ti_locked = TI_UNLOCKED; 1391 } 1392 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1393 return (IPPROTO_DONE); 1394 } else if (tp->t_state == TCPS_LISTEN) { 1395 /* 1396 * When a listen socket is torn down the SO_ACCEPTCONN 1397 * flag is removed first while connections are drained 1398 * from the accept queue in a unlock/lock cycle of the 1399 * ACCEPT_LOCK, opening a race condition allowing a SYN 1400 * attempt go through unhandled. 1401 */ 1402 goto dropunlock; 1403 } 1404 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1405 if (tp->t_flags & TF_SIGNATURE) { 1406 tcp_dooptions(&to, optp, optlen, thflags); 1407 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1408 TCPSTAT_INC(tcps_sig_err_nosigopt); 1409 goto dropunlock; 1410 } 1411 if (!TCPMD5_ENABLED() || 1412 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1413 goto dropunlock; 1414 } 1415 #endif 1416 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1417 1418 /* 1419 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1420 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1421 * the inpcb, and unlocks pcbinfo. 1422 */ 1423 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1424 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1425 return (IPPROTO_DONE); 1426 1427 dropwithreset: 1428 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1429 1430 if (ti_locked == TI_RLOCKED) { 1431 INP_INFO_RUNLOCK(&V_tcbinfo); 1432 ti_locked = TI_UNLOCKED; 1433 } 1434 #ifdef INVARIANTS 1435 else { 1436 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1437 "ti_locked: %d", __func__, ti_locked)); 1438 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1439 } 1440 #endif 1441 1442 if (inp != NULL) { 1443 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1444 INP_WUNLOCK(inp); 1445 } else 1446 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1447 m = NULL; /* mbuf chain got consumed. */ 1448 goto drop; 1449 1450 dropunlock: 1451 if (m != NULL) 1452 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1453 1454 if (ti_locked == TI_RLOCKED) { 1455 INP_INFO_RUNLOCK(&V_tcbinfo); 1456 ti_locked = TI_UNLOCKED; 1457 } 1458 #ifdef INVARIANTS 1459 else { 1460 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1461 "ti_locked: %d", __func__, ti_locked)); 1462 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1463 } 1464 #endif 1465 1466 if (inp != NULL) 1467 INP_WUNLOCK(inp); 1468 1469 drop: 1470 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1471 if (s != NULL) 1472 free(s, M_TCPLOG); 1473 if (m != NULL) 1474 m_freem(m); 1475 return (IPPROTO_DONE); 1476 } 1477 1478 /* 1479 * Automatic sizing of receive socket buffer. Often the send 1480 * buffer size is not optimally adjusted to the actual network 1481 * conditions at hand (delay bandwidth product). Setting the 1482 * buffer size too small limits throughput on links with high 1483 * bandwidth and high delay (eg. trans-continental/oceanic links). 1484 * 1485 * On the receive side the socket buffer memory is only rarely 1486 * used to any significant extent. This allows us to be much 1487 * more aggressive in scaling the receive socket buffer. For 1488 * the case that the buffer space is actually used to a large 1489 * extent and we run out of kernel memory we can simply drop 1490 * the new segments; TCP on the sender will just retransmit it 1491 * later. Setting the buffer size too big may only consume too 1492 * much kernel memory if the application doesn't read() from 1493 * the socket or packet loss or reordering makes use of the 1494 * reassembly queue. 1495 * 1496 * The criteria to step up the receive buffer one notch are: 1497 * 1. Application has not set receive buffer size with 1498 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1499 * 2. the number of bytes received during the time it takes 1500 * one timestamp to be reflected back to us (the RTT); 1501 * 3. received bytes per RTT is within seven eighth of the 1502 * current socket buffer size; 1503 * 4. receive buffer size has not hit maximal automatic size; 1504 * 1505 * This algorithm does one step per RTT at most and only if 1506 * we receive a bulk stream w/o packet losses or reorderings. 1507 * Shrinking the buffer during idle times is not necessary as 1508 * it doesn't consume any memory when idle. 1509 * 1510 * TODO: Only step up if the application is actually serving 1511 * the buffer to better manage the socket buffer resources. 1512 */ 1513 int 1514 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1515 struct tcpcb *tp, int tlen) 1516 { 1517 int newsize = 0; 1518 1519 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1520 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1521 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1522 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1523 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1524 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1525 newsize = min(so->so_rcv.sb_hiwat + 1526 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1527 } 1528 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1529 1530 /* Start over with next RTT. */ 1531 tp->rfbuf_ts = 0; 1532 tp->rfbuf_cnt = 0; 1533 } else { 1534 tp->rfbuf_cnt += tlen; /* add up */ 1535 } 1536 1537 return (newsize); 1538 } 1539 1540 void 1541 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1542 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1543 int ti_locked) 1544 { 1545 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1546 int rstreason, todrop, win; 1547 uint32_t tiwin; 1548 uint16_t nsegs; 1549 char *s; 1550 struct in_conninfo *inc; 1551 struct mbuf *mfree; 1552 struct tcpopt to; 1553 #ifdef TCP_RFC7413 1554 int tfo_syn; 1555 #endif 1556 1557 #ifdef TCPDEBUG 1558 /* 1559 * The size of tcp_saveipgen must be the size of the max ip header, 1560 * now IPv6. 1561 */ 1562 u_char tcp_saveipgen[IP6_HDR_LEN]; 1563 struct tcphdr tcp_savetcp; 1564 short ostate = 0; 1565 #endif 1566 thflags = th->th_flags; 1567 inc = &tp->t_inpcb->inp_inc; 1568 tp->sackhint.last_sack_ack = 0; 1569 sack_changed = 0; 1570 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1571 1572 /* 1573 * If this is either a state-changing packet or current state isn't 1574 * established, we require a write lock on tcbinfo. Otherwise, we 1575 * allow the tcbinfo to be in either alocked or unlocked, as the 1576 * caller may have unnecessarily acquired a write lock due to a race. 1577 */ 1578 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1579 tp->t_state != TCPS_ESTABLISHED) { 1580 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1581 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1582 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1583 } else { 1584 #ifdef INVARIANTS 1585 if (ti_locked == TI_RLOCKED) 1586 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1587 else { 1588 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1589 "ti_locked: %d", __func__, ti_locked)); 1590 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1591 } 1592 #endif 1593 } 1594 INP_WLOCK_ASSERT(tp->t_inpcb); 1595 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1596 __func__)); 1597 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1598 __func__)); 1599 1600 #ifdef TCPPCAP 1601 /* Save segment, if requested. */ 1602 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1603 #endif 1604 1605 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1606 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1607 log(LOG_DEBUG, "%s; %s: " 1608 "SYN|FIN segment ignored (based on " 1609 "sysctl setting)\n", s, __func__); 1610 free(s, M_TCPLOG); 1611 } 1612 goto drop; 1613 } 1614 1615 /* 1616 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1617 * check SEQ.ACK first. 1618 */ 1619 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1620 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1621 rstreason = BANDLIM_UNLIMITED; 1622 goto dropwithreset; 1623 } 1624 1625 /* 1626 * Segment received on connection. 1627 * Reset idle time and keep-alive timer. 1628 * XXX: This should be done after segment 1629 * validation to ignore broken/spoofed segs. 1630 */ 1631 tp->t_rcvtime = ticks; 1632 1633 /* 1634 * Scale up the window into a 32-bit value. 1635 * For the SYN_SENT state the scale is zero. 1636 */ 1637 tiwin = th->th_win << tp->snd_scale; 1638 1639 /* 1640 * TCP ECN processing. 1641 */ 1642 if (tp->t_flags & TF_ECN_PERMIT) { 1643 if (thflags & TH_CWR) 1644 tp->t_flags &= ~TF_ECN_SND_ECE; 1645 switch (iptos & IPTOS_ECN_MASK) { 1646 case IPTOS_ECN_CE: 1647 tp->t_flags |= TF_ECN_SND_ECE; 1648 TCPSTAT_INC(tcps_ecn_ce); 1649 break; 1650 case IPTOS_ECN_ECT0: 1651 TCPSTAT_INC(tcps_ecn_ect0); 1652 break; 1653 case IPTOS_ECN_ECT1: 1654 TCPSTAT_INC(tcps_ecn_ect1); 1655 break; 1656 } 1657 1658 /* Process a packet differently from RFC3168. */ 1659 cc_ecnpkt_handler(tp, th, iptos); 1660 1661 /* Congestion experienced. */ 1662 if (thflags & TH_ECE) { 1663 cc_cong_signal(tp, th, CC_ECN); 1664 } 1665 } 1666 1667 /* 1668 * Parse options on any incoming segment. 1669 */ 1670 tcp_dooptions(&to, (u_char *)(th + 1), 1671 (th->th_off << 2) - sizeof(struct tcphdr), 1672 (thflags & TH_SYN) ? TO_SYN : 0); 1673 1674 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1675 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1676 (to.to_flags & TOF_SIGNATURE) == 0) { 1677 TCPSTAT_INC(tcps_sig_err_sigopt); 1678 /* XXX: should drop? */ 1679 } 1680 #endif 1681 /* 1682 * If echoed timestamp is later than the current time, 1683 * fall back to non RFC1323 RTT calculation. Normalize 1684 * timestamp if syncookies were used when this connection 1685 * was established. 1686 */ 1687 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1688 to.to_tsecr -= tp->ts_offset; 1689 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1690 to.to_tsecr = 0; 1691 } 1692 /* 1693 * Process options only when we get SYN/ACK back. The SYN case 1694 * for incoming connections is handled in tcp_syncache. 1695 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1696 * or <SYN,ACK>) segment itself is never scaled. 1697 * XXX this is traditional behavior, may need to be cleaned up. 1698 */ 1699 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1700 if ((to.to_flags & TOF_SCALE) && 1701 (tp->t_flags & TF_REQ_SCALE)) { 1702 tp->t_flags |= TF_RCVD_SCALE; 1703 tp->snd_scale = to.to_wscale; 1704 } 1705 /* 1706 * Initial send window. It will be updated with 1707 * the next incoming segment to the scaled value. 1708 */ 1709 tp->snd_wnd = th->th_win; 1710 if (to.to_flags & TOF_TS) { 1711 tp->t_flags |= TF_RCVD_TSTMP; 1712 tp->ts_recent = to.to_tsval; 1713 tp->ts_recent_age = tcp_ts_getticks(); 1714 } 1715 if (to.to_flags & TOF_MSS) 1716 tcp_mss(tp, to.to_mss); 1717 if ((tp->t_flags & TF_SACK_PERMIT) && 1718 (to.to_flags & TOF_SACKPERM) == 0) 1719 tp->t_flags &= ~TF_SACK_PERMIT; 1720 } 1721 1722 /* 1723 * If timestamps were negotiated during SYN/ACK they should 1724 * appear on every segment during this session and vice versa. 1725 */ 1726 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1727 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1728 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1729 "no action\n", s, __func__); 1730 free(s, M_TCPLOG); 1731 } 1732 } 1733 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1734 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1735 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1736 "no action\n", s, __func__); 1737 free(s, M_TCPLOG); 1738 } 1739 } 1740 1741 /* 1742 * Header prediction: check for the two common cases 1743 * of a uni-directional data xfer. If the packet has 1744 * no control flags, is in-sequence, the window didn't 1745 * change and we're not retransmitting, it's a 1746 * candidate. If the length is zero and the ack moved 1747 * forward, we're the sender side of the xfer. Just 1748 * free the data acked & wake any higher level process 1749 * that was blocked waiting for space. If the length 1750 * is non-zero and the ack didn't move, we're the 1751 * receiver side. If we're getting packets in-order 1752 * (the reassembly queue is empty), add the data to 1753 * the socket buffer and note that we need a delayed ack. 1754 * Make sure that the hidden state-flags are also off. 1755 * Since we check for TCPS_ESTABLISHED first, it can only 1756 * be TH_NEEDSYN. 1757 */ 1758 if (tp->t_state == TCPS_ESTABLISHED && 1759 th->th_seq == tp->rcv_nxt && 1760 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1761 tp->snd_nxt == tp->snd_max && 1762 tiwin && tiwin == tp->snd_wnd && 1763 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1764 LIST_EMPTY(&tp->t_segq) && 1765 ((to.to_flags & TOF_TS) == 0 || 1766 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1767 1768 /* 1769 * If last ACK falls within this segment's sequence numbers, 1770 * record the timestamp. 1771 * NOTE that the test is modified according to the latest 1772 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1773 */ 1774 if ((to.to_flags & TOF_TS) != 0 && 1775 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1776 tp->ts_recent_age = tcp_ts_getticks(); 1777 tp->ts_recent = to.to_tsval; 1778 } 1779 1780 if (tlen == 0) { 1781 if (SEQ_GT(th->th_ack, tp->snd_una) && 1782 SEQ_LEQ(th->th_ack, tp->snd_max) && 1783 !IN_RECOVERY(tp->t_flags) && 1784 (to.to_flags & TOF_SACK) == 0 && 1785 TAILQ_EMPTY(&tp->snd_holes)) { 1786 /* 1787 * This is a pure ack for outstanding data. 1788 */ 1789 if (ti_locked == TI_RLOCKED) 1790 INP_INFO_RUNLOCK(&V_tcbinfo); 1791 ti_locked = TI_UNLOCKED; 1792 1793 TCPSTAT_INC(tcps_predack); 1794 1795 /* 1796 * "bad retransmit" recovery. 1797 */ 1798 if (tp->t_rxtshift == 1 && 1799 tp->t_flags & TF_PREVVALID && 1800 (int)(ticks - tp->t_badrxtwin) < 0) { 1801 cc_cong_signal(tp, th, CC_RTO_ERR); 1802 } 1803 1804 /* 1805 * Recalculate the transmit timer / rtt. 1806 * 1807 * Some boxes send broken timestamp replies 1808 * during the SYN+ACK phase, ignore 1809 * timestamps of 0 or we could calculate a 1810 * huge RTT and blow up the retransmit timer. 1811 */ 1812 if ((to.to_flags & TOF_TS) != 0 && 1813 to.to_tsecr) { 1814 uint32_t t; 1815 1816 t = tcp_ts_getticks() - to.to_tsecr; 1817 if (!tp->t_rttlow || tp->t_rttlow > t) 1818 tp->t_rttlow = t; 1819 tcp_xmit_timer(tp, 1820 TCP_TS_TO_TICKS(t) + 1); 1821 } else if (tp->t_rtttime && 1822 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1823 if (!tp->t_rttlow || 1824 tp->t_rttlow > ticks - tp->t_rtttime) 1825 tp->t_rttlow = ticks - tp->t_rtttime; 1826 tcp_xmit_timer(tp, 1827 ticks - tp->t_rtttime); 1828 } 1829 acked = BYTES_THIS_ACK(tp, th); 1830 1831 #ifdef TCP_HHOOK 1832 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1833 hhook_run_tcp_est_in(tp, th, &to); 1834 #endif 1835 1836 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1837 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1838 sbdrop(&so->so_snd, acked); 1839 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1840 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1841 tp->snd_recover = th->th_ack - 1; 1842 1843 /* 1844 * Let the congestion control algorithm update 1845 * congestion control related information. This 1846 * typically means increasing the congestion 1847 * window. 1848 */ 1849 cc_ack_received(tp, th, nsegs, CC_ACK); 1850 1851 tp->snd_una = th->th_ack; 1852 /* 1853 * Pull snd_wl2 up to prevent seq wrap relative 1854 * to th_ack. 1855 */ 1856 tp->snd_wl2 = th->th_ack; 1857 tp->t_dupacks = 0; 1858 m_freem(m); 1859 1860 /* 1861 * If all outstanding data are acked, stop 1862 * retransmit timer, otherwise restart timer 1863 * using current (possibly backed-off) value. 1864 * If process is waiting for space, 1865 * wakeup/selwakeup/signal. If data 1866 * are ready to send, let tcp_output 1867 * decide between more output or persist. 1868 */ 1869 #ifdef TCPDEBUG 1870 if (so->so_options & SO_DEBUG) 1871 tcp_trace(TA_INPUT, ostate, tp, 1872 (void *)tcp_saveipgen, 1873 &tcp_savetcp, 0); 1874 #endif 1875 TCP_PROBE3(debug__input, tp, th, m); 1876 if (tp->snd_una == tp->snd_max) 1877 tcp_timer_activate(tp, TT_REXMT, 0); 1878 else if (!tcp_timer_active(tp, TT_PERSIST)) 1879 tcp_timer_activate(tp, TT_REXMT, 1880 tp->t_rxtcur); 1881 sowwakeup(so); 1882 if (sbavail(&so->so_snd)) 1883 (void) tp->t_fb->tfb_tcp_output(tp); 1884 goto check_delack; 1885 } 1886 } else if (th->th_ack == tp->snd_una && 1887 tlen <= sbspace(&so->so_rcv)) { 1888 int newsize = 0; /* automatic sockbuf scaling */ 1889 1890 /* 1891 * This is a pure, in-sequence data packet with 1892 * nothing on the reassembly queue and we have enough 1893 * buffer space to take it. 1894 */ 1895 if (ti_locked == TI_RLOCKED) 1896 INP_INFO_RUNLOCK(&V_tcbinfo); 1897 ti_locked = TI_UNLOCKED; 1898 1899 /* Clean receiver SACK report if present */ 1900 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1901 tcp_clean_sackreport(tp); 1902 TCPSTAT_INC(tcps_preddat); 1903 tp->rcv_nxt += tlen; 1904 /* 1905 * Pull snd_wl1 up to prevent seq wrap relative to 1906 * th_seq. 1907 */ 1908 tp->snd_wl1 = th->th_seq; 1909 /* 1910 * Pull rcv_up up to prevent seq wrap relative to 1911 * rcv_nxt. 1912 */ 1913 tp->rcv_up = tp->rcv_nxt; 1914 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1915 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1916 #ifdef TCPDEBUG 1917 if (so->so_options & SO_DEBUG) 1918 tcp_trace(TA_INPUT, ostate, tp, 1919 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1920 #endif 1921 TCP_PROBE3(debug__input, tp, th, m); 1922 1923 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1924 1925 /* Add data to socket buffer. */ 1926 SOCKBUF_LOCK(&so->so_rcv); 1927 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1928 m_freem(m); 1929 } else { 1930 /* 1931 * Set new socket buffer size. 1932 * Give up when limit is reached. 1933 */ 1934 if (newsize) 1935 if (!sbreserve_locked(&so->so_rcv, 1936 newsize, so, NULL)) 1937 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1938 m_adj(m, drop_hdrlen); /* delayed header drop */ 1939 sbappendstream_locked(&so->so_rcv, m, 0); 1940 } 1941 /* NB: sorwakeup_locked() does an implicit unlock. */ 1942 sorwakeup_locked(so); 1943 if (DELAY_ACK(tp, tlen)) { 1944 tp->t_flags |= TF_DELACK; 1945 } else { 1946 tp->t_flags |= TF_ACKNOW; 1947 tp->t_fb->tfb_tcp_output(tp); 1948 } 1949 goto check_delack; 1950 } 1951 } 1952 1953 /* 1954 * Calculate amount of space in receive window, 1955 * and then do TCP input processing. 1956 * Receive window is amount of space in rcv queue, 1957 * but not less than advertised window. 1958 */ 1959 win = sbspace(&so->so_rcv); 1960 if (win < 0) 1961 win = 0; 1962 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1963 1964 switch (tp->t_state) { 1965 1966 /* 1967 * If the state is SYN_RECEIVED: 1968 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1969 */ 1970 case TCPS_SYN_RECEIVED: 1971 if ((thflags & TH_ACK) && 1972 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1973 SEQ_GT(th->th_ack, tp->snd_max))) { 1974 rstreason = BANDLIM_RST_OPENPORT; 1975 goto dropwithreset; 1976 } 1977 #ifdef TCP_RFC7413 1978 if (IS_FASTOPEN(tp->t_flags)) { 1979 /* 1980 * When a TFO connection is in SYN_RECEIVED, the 1981 * only valid packets are the initial SYN, a 1982 * retransmit/copy of the initial SYN (possibly with 1983 * a subset of the original data), a valid ACK, a 1984 * FIN, or a RST. 1985 */ 1986 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1987 rstreason = BANDLIM_RST_OPENPORT; 1988 goto dropwithreset; 1989 } else if (thflags & TH_SYN) { 1990 /* non-initial SYN is ignored */ 1991 if ((tcp_timer_active(tp, TT_DELACK) || 1992 tcp_timer_active(tp, TT_REXMT))) 1993 goto drop; 1994 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1995 goto drop; 1996 } 1997 } 1998 #endif 1999 break; 2000 2001 /* 2002 * If the state is SYN_SENT: 2003 * if seg contains a RST with valid ACK (SEQ.ACK has already 2004 * been verified), then drop the connection. 2005 * if seg contains a RST without an ACK, drop the seg. 2006 * if seg does not contain SYN, then drop the seg. 2007 * Otherwise this is an acceptable SYN segment 2008 * initialize tp->rcv_nxt and tp->irs 2009 * if seg contains ack then advance tp->snd_una 2010 * if seg contains an ECE and ECN support is enabled, the stream 2011 * is ECN capable. 2012 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2013 * arrange for segment to be acked (eventually) 2014 * continue processing rest of data/controls, beginning with URG 2015 */ 2016 case TCPS_SYN_SENT: 2017 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2018 TCP_PROBE5(connect__refused, NULL, tp, 2019 m, tp, th); 2020 tp = tcp_drop(tp, ECONNREFUSED); 2021 } 2022 if (thflags & TH_RST) 2023 goto drop; 2024 if (!(thflags & TH_SYN)) 2025 goto drop; 2026 2027 tp->irs = th->th_seq; 2028 tcp_rcvseqinit(tp); 2029 if (thflags & TH_ACK) { 2030 TCPSTAT_INC(tcps_connects); 2031 soisconnected(so); 2032 #ifdef MAC 2033 mac_socketpeer_set_from_mbuf(m, so); 2034 #endif 2035 /* Do window scaling on this connection? */ 2036 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2037 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2038 tp->rcv_scale = tp->request_r_scale; 2039 } 2040 tp->rcv_adv += min(tp->rcv_wnd, 2041 TCP_MAXWIN << tp->rcv_scale); 2042 tp->snd_una++; /* SYN is acked */ 2043 /* 2044 * If there's data, delay ACK; if there's also a FIN 2045 * ACKNOW will be turned on later. 2046 */ 2047 if (DELAY_ACK(tp, tlen) && tlen != 0) 2048 tcp_timer_activate(tp, TT_DELACK, 2049 tcp_delacktime); 2050 else 2051 tp->t_flags |= TF_ACKNOW; 2052 2053 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2054 tp->t_flags |= TF_ECN_PERMIT; 2055 TCPSTAT_INC(tcps_ecn_shs); 2056 } 2057 2058 /* 2059 * Received <SYN,ACK> in SYN_SENT[*] state. 2060 * Transitions: 2061 * SYN_SENT --> ESTABLISHED 2062 * SYN_SENT* --> FIN_WAIT_1 2063 */ 2064 tp->t_starttime = ticks; 2065 if (tp->t_flags & TF_NEEDFIN) { 2066 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2067 tp->t_flags &= ~TF_NEEDFIN; 2068 thflags &= ~TH_SYN; 2069 } else { 2070 tcp_state_change(tp, TCPS_ESTABLISHED); 2071 TCP_PROBE5(connect__established, NULL, tp, 2072 m, tp, th); 2073 cc_conn_init(tp); 2074 tcp_timer_activate(tp, TT_KEEP, 2075 TP_KEEPIDLE(tp)); 2076 } 2077 } else { 2078 /* 2079 * Received initial SYN in SYN-SENT[*] state => 2080 * simultaneous open. 2081 * If it succeeds, connection is * half-synchronized. 2082 * Otherwise, do 3-way handshake: 2083 * SYN-SENT -> SYN-RECEIVED 2084 * SYN-SENT* -> SYN-RECEIVED* 2085 */ 2086 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2087 tcp_timer_activate(tp, TT_REXMT, 0); 2088 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2089 } 2090 2091 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2092 "ti_locked %d", __func__, ti_locked)); 2093 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2094 INP_WLOCK_ASSERT(tp->t_inpcb); 2095 2096 /* 2097 * Advance th->th_seq to correspond to first data byte. 2098 * If data, trim to stay within window, 2099 * dropping FIN if necessary. 2100 */ 2101 th->th_seq++; 2102 if (tlen > tp->rcv_wnd) { 2103 todrop = tlen - tp->rcv_wnd; 2104 m_adj(m, -todrop); 2105 tlen = tp->rcv_wnd; 2106 thflags &= ~TH_FIN; 2107 TCPSTAT_INC(tcps_rcvpackafterwin); 2108 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2109 } 2110 tp->snd_wl1 = th->th_seq - 1; 2111 tp->rcv_up = th->th_seq; 2112 /* 2113 * Client side of transaction: already sent SYN and data. 2114 * If the remote host used T/TCP to validate the SYN, 2115 * our data will be ACK'd; if so, enter normal data segment 2116 * processing in the middle of step 5, ack processing. 2117 * Otherwise, goto step 6. 2118 */ 2119 if (thflags & TH_ACK) 2120 goto process_ACK; 2121 2122 goto step6; 2123 2124 /* 2125 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2126 * do normal processing. 2127 * 2128 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2129 */ 2130 case TCPS_LAST_ACK: 2131 case TCPS_CLOSING: 2132 break; /* continue normal processing */ 2133 } 2134 2135 /* 2136 * States other than LISTEN or SYN_SENT. 2137 * First check the RST flag and sequence number since reset segments 2138 * are exempt from the timestamp and connection count tests. This 2139 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2140 * below which allowed reset segments in half the sequence space 2141 * to fall though and be processed (which gives forged reset 2142 * segments with a random sequence number a 50 percent chance of 2143 * killing a connection). 2144 * Then check timestamp, if present. 2145 * Then check the connection count, if present. 2146 * Then check that at least some bytes of segment are within 2147 * receive window. If segment begins before rcv_nxt, 2148 * drop leading data (and SYN); if nothing left, just ack. 2149 */ 2150 if (thflags & TH_RST) { 2151 /* 2152 * RFC5961 Section 3.2 2153 * 2154 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2155 * - If RST is in window, we send challenge ACK. 2156 * 2157 * Note: to take into account delayed ACKs, we should 2158 * test against last_ack_sent instead of rcv_nxt. 2159 * Note 2: we handle special case of closed window, not 2160 * covered by the RFC. 2161 */ 2162 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2163 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2164 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2165 2166 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2167 KASSERT(ti_locked == TI_RLOCKED, 2168 ("%s: TH_RST ti_locked %d, th %p tp %p", 2169 __func__, ti_locked, th, tp)); 2170 KASSERT(tp->t_state != TCPS_SYN_SENT, 2171 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2172 __func__, th, tp)); 2173 2174 if (V_tcp_insecure_rst || 2175 tp->last_ack_sent == th->th_seq) { 2176 TCPSTAT_INC(tcps_drops); 2177 /* Drop the connection. */ 2178 switch (tp->t_state) { 2179 case TCPS_SYN_RECEIVED: 2180 so->so_error = ECONNREFUSED; 2181 goto close; 2182 case TCPS_ESTABLISHED: 2183 case TCPS_FIN_WAIT_1: 2184 case TCPS_FIN_WAIT_2: 2185 case TCPS_CLOSE_WAIT: 2186 case TCPS_CLOSING: 2187 case TCPS_LAST_ACK: 2188 so->so_error = ECONNRESET; 2189 close: 2190 /* FALLTHROUGH */ 2191 default: 2192 tp = tcp_close(tp); 2193 } 2194 } else { 2195 TCPSTAT_INC(tcps_badrst); 2196 /* Send challenge ACK. */ 2197 tcp_respond(tp, mtod(m, void *), th, m, 2198 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2199 tp->last_ack_sent = tp->rcv_nxt; 2200 m = NULL; 2201 } 2202 } 2203 goto drop; 2204 } 2205 2206 /* 2207 * RFC5961 Section 4.2 2208 * Send challenge ACK for any SYN in synchronized state. 2209 */ 2210 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2211 tp->t_state != TCPS_SYN_RECEIVED) { 2212 KASSERT(ti_locked == TI_RLOCKED, 2213 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2214 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2215 2216 TCPSTAT_INC(tcps_badsyn); 2217 if (V_tcp_insecure_syn && 2218 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2219 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2220 tp = tcp_drop(tp, ECONNRESET); 2221 rstreason = BANDLIM_UNLIMITED; 2222 } else { 2223 /* Send challenge ACK. */ 2224 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2225 tp->snd_nxt, TH_ACK); 2226 tp->last_ack_sent = tp->rcv_nxt; 2227 m = NULL; 2228 } 2229 goto drop; 2230 } 2231 2232 /* 2233 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2234 * and it's less than ts_recent, drop it. 2235 */ 2236 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2237 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2238 2239 /* Check to see if ts_recent is over 24 days old. */ 2240 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2241 /* 2242 * Invalidate ts_recent. If this segment updates 2243 * ts_recent, the age will be reset later and ts_recent 2244 * will get a valid value. If it does not, setting 2245 * ts_recent to zero will at least satisfy the 2246 * requirement that zero be placed in the timestamp 2247 * echo reply when ts_recent isn't valid. The 2248 * age isn't reset until we get a valid ts_recent 2249 * because we don't want out-of-order segments to be 2250 * dropped when ts_recent is old. 2251 */ 2252 tp->ts_recent = 0; 2253 } else { 2254 TCPSTAT_INC(tcps_rcvduppack); 2255 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2256 TCPSTAT_INC(tcps_pawsdrop); 2257 if (tlen) 2258 goto dropafterack; 2259 goto drop; 2260 } 2261 } 2262 2263 /* 2264 * In the SYN-RECEIVED state, validate that the packet belongs to 2265 * this connection before trimming the data to fit the receive 2266 * window. Check the sequence number versus IRS since we know 2267 * the sequence numbers haven't wrapped. This is a partial fix 2268 * for the "LAND" DoS attack. 2269 */ 2270 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2271 rstreason = BANDLIM_RST_OPENPORT; 2272 goto dropwithreset; 2273 } 2274 2275 todrop = tp->rcv_nxt - th->th_seq; 2276 if (todrop > 0) { 2277 if (thflags & TH_SYN) { 2278 thflags &= ~TH_SYN; 2279 th->th_seq++; 2280 if (th->th_urp > 1) 2281 th->th_urp--; 2282 else 2283 thflags &= ~TH_URG; 2284 todrop--; 2285 } 2286 /* 2287 * Following if statement from Stevens, vol. 2, p. 960. 2288 */ 2289 if (todrop > tlen 2290 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2291 /* 2292 * Any valid FIN must be to the left of the window. 2293 * At this point the FIN must be a duplicate or out 2294 * of sequence; drop it. 2295 */ 2296 thflags &= ~TH_FIN; 2297 2298 /* 2299 * Send an ACK to resynchronize and drop any data. 2300 * But keep on processing for RST or ACK. 2301 */ 2302 tp->t_flags |= TF_ACKNOW; 2303 todrop = tlen; 2304 TCPSTAT_INC(tcps_rcvduppack); 2305 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2306 } else { 2307 TCPSTAT_INC(tcps_rcvpartduppack); 2308 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2309 } 2310 drop_hdrlen += todrop; /* drop from the top afterwards */ 2311 th->th_seq += todrop; 2312 tlen -= todrop; 2313 if (th->th_urp > todrop) 2314 th->th_urp -= todrop; 2315 else { 2316 thflags &= ~TH_URG; 2317 th->th_urp = 0; 2318 } 2319 } 2320 2321 /* 2322 * If new data are received on a connection after the 2323 * user processes are gone, then RST the other end. 2324 */ 2325 if ((so->so_state & SS_NOFDREF) && 2326 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2327 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2328 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2329 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2330 2331 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2332 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2333 "after socket was closed, " 2334 "sending RST and removing tcpcb\n", 2335 s, __func__, tcpstates[tp->t_state], tlen); 2336 free(s, M_TCPLOG); 2337 } 2338 tp = tcp_close(tp); 2339 TCPSTAT_INC(tcps_rcvafterclose); 2340 rstreason = BANDLIM_UNLIMITED; 2341 goto dropwithreset; 2342 } 2343 2344 /* 2345 * If segment ends after window, drop trailing data 2346 * (and PUSH and FIN); if nothing left, just ACK. 2347 */ 2348 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2349 if (todrop > 0) { 2350 TCPSTAT_INC(tcps_rcvpackafterwin); 2351 if (todrop >= tlen) { 2352 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2353 /* 2354 * If window is closed can only take segments at 2355 * window edge, and have to drop data and PUSH from 2356 * incoming segments. Continue processing, but 2357 * remember to ack. Otherwise, drop segment 2358 * and ack. 2359 */ 2360 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2361 tp->t_flags |= TF_ACKNOW; 2362 TCPSTAT_INC(tcps_rcvwinprobe); 2363 } else 2364 goto dropafterack; 2365 } else 2366 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2367 m_adj(m, -todrop); 2368 tlen -= todrop; 2369 thflags &= ~(TH_PUSH|TH_FIN); 2370 } 2371 2372 /* 2373 * If last ACK falls within this segment's sequence numbers, 2374 * record its timestamp. 2375 * NOTE: 2376 * 1) That the test incorporates suggestions from the latest 2377 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2378 * 2) That updating only on newer timestamps interferes with 2379 * our earlier PAWS tests, so this check should be solely 2380 * predicated on the sequence space of this segment. 2381 * 3) That we modify the segment boundary check to be 2382 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2383 * instead of RFC1323's 2384 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2385 * This modified check allows us to overcome RFC1323's 2386 * limitations as described in Stevens TCP/IP Illustrated 2387 * Vol. 2 p.869. In such cases, we can still calculate the 2388 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2389 */ 2390 if ((to.to_flags & TOF_TS) != 0 && 2391 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2392 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2393 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2394 tp->ts_recent_age = tcp_ts_getticks(); 2395 tp->ts_recent = to.to_tsval; 2396 } 2397 2398 /* 2399 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2400 * flag is on (half-synchronized state), then queue data for 2401 * later processing; else drop segment and return. 2402 */ 2403 if ((thflags & TH_ACK) == 0) { 2404 if (tp->t_state == TCPS_SYN_RECEIVED || 2405 (tp->t_flags & TF_NEEDSYN)) { 2406 #ifdef TCP_RFC7413 2407 if (tp->t_state == TCPS_SYN_RECEIVED && 2408 IS_FASTOPEN(tp->t_flags)) { 2409 tp->snd_wnd = tiwin; 2410 cc_conn_init(tp); 2411 } 2412 #endif 2413 goto step6; 2414 } else if (tp->t_flags & TF_ACKNOW) 2415 goto dropafterack; 2416 else 2417 goto drop; 2418 } 2419 2420 /* 2421 * Ack processing. 2422 */ 2423 switch (tp->t_state) { 2424 2425 /* 2426 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2427 * ESTABLISHED state and continue processing. 2428 * The ACK was checked above. 2429 */ 2430 case TCPS_SYN_RECEIVED: 2431 2432 TCPSTAT_INC(tcps_connects); 2433 soisconnected(so); 2434 /* Do window scaling? */ 2435 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2436 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2437 tp->rcv_scale = tp->request_r_scale; 2438 tp->snd_wnd = tiwin; 2439 } 2440 /* 2441 * Make transitions: 2442 * SYN-RECEIVED -> ESTABLISHED 2443 * SYN-RECEIVED* -> FIN-WAIT-1 2444 */ 2445 tp->t_starttime = ticks; 2446 if (tp->t_flags & TF_NEEDFIN) { 2447 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2448 tp->t_flags &= ~TF_NEEDFIN; 2449 } else { 2450 tcp_state_change(tp, TCPS_ESTABLISHED); 2451 TCP_PROBE5(accept__established, NULL, tp, 2452 m, tp, th); 2453 #ifdef TCP_RFC7413 2454 if (tp->t_tfo_pending) { 2455 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2456 tp->t_tfo_pending = NULL; 2457 2458 /* 2459 * Account for the ACK of our SYN prior to 2460 * regular ACK processing below. 2461 */ 2462 tp->snd_una++; 2463 } 2464 /* 2465 * TFO connections call cc_conn_init() during SYN 2466 * processing. Calling it again here for such 2467 * connections is not harmless as it would undo the 2468 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2469 * is retransmitted. 2470 */ 2471 if (!IS_FASTOPEN(tp->t_flags)) 2472 #endif 2473 cc_conn_init(tp); 2474 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2475 } 2476 /* 2477 * If segment contains data or ACK, will call tcp_reass() 2478 * later; if not, do so now to pass queued data to user. 2479 */ 2480 if (tlen == 0 && (thflags & TH_FIN) == 0) 2481 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2482 (struct mbuf *)0); 2483 tp->snd_wl1 = th->th_seq - 1; 2484 /* FALLTHROUGH */ 2485 2486 /* 2487 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2488 * ACKs. If the ack is in the range 2489 * tp->snd_una < th->th_ack <= tp->snd_max 2490 * then advance tp->snd_una to th->th_ack and drop 2491 * data from the retransmission queue. If this ACK reflects 2492 * more up to date window information we update our window information. 2493 */ 2494 case TCPS_ESTABLISHED: 2495 case TCPS_FIN_WAIT_1: 2496 case TCPS_FIN_WAIT_2: 2497 case TCPS_CLOSE_WAIT: 2498 case TCPS_CLOSING: 2499 case TCPS_LAST_ACK: 2500 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2501 TCPSTAT_INC(tcps_rcvacktoomuch); 2502 goto dropafterack; 2503 } 2504 if ((tp->t_flags & TF_SACK_PERMIT) && 2505 ((to.to_flags & TOF_SACK) || 2506 !TAILQ_EMPTY(&tp->snd_holes))) 2507 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2508 else 2509 /* 2510 * Reset the value so that previous (valid) value 2511 * from the last ack with SACK doesn't get used. 2512 */ 2513 tp->sackhint.sacked_bytes = 0; 2514 2515 #ifdef TCP_HHOOK 2516 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2517 hhook_run_tcp_est_in(tp, th, &to); 2518 #endif 2519 2520 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2521 u_int maxseg; 2522 2523 maxseg = tcp_maxseg(tp); 2524 if (tlen == 0 && 2525 (tiwin == tp->snd_wnd || 2526 (tp->t_flags & TF_SACK_PERMIT))) { 2527 /* 2528 * If this is the first time we've seen a 2529 * FIN from the remote, this is not a 2530 * duplicate and it needs to be processed 2531 * normally. This happens during a 2532 * simultaneous close. 2533 */ 2534 if ((thflags & TH_FIN) && 2535 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2536 tp->t_dupacks = 0; 2537 break; 2538 } 2539 TCPSTAT_INC(tcps_rcvdupack); 2540 /* 2541 * If we have outstanding data (other than 2542 * a window probe), this is a completely 2543 * duplicate ack (ie, window info didn't 2544 * change and FIN isn't set), 2545 * the ack is the biggest we've 2546 * seen and we've seen exactly our rexmt 2547 * threshold of them, assume a packet 2548 * has been dropped and retransmit it. 2549 * Kludge snd_nxt & the congestion 2550 * window so we send only this one 2551 * packet. 2552 * 2553 * We know we're losing at the current 2554 * window size so do congestion avoidance 2555 * (set ssthresh to half the current window 2556 * and pull our congestion window back to 2557 * the new ssthresh). 2558 * 2559 * Dup acks mean that packets have left the 2560 * network (they're now cached at the receiver) 2561 * so bump cwnd by the amount in the receiver 2562 * to keep a constant cwnd packets in the 2563 * network. 2564 * 2565 * When using TCP ECN, notify the peer that 2566 * we reduced the cwnd. 2567 */ 2568 /* 2569 * Following 2 kinds of acks should not affect 2570 * dupack counting: 2571 * 1) Old acks 2572 * 2) Acks with SACK but without any new SACK 2573 * information in them. These could result from 2574 * any anomaly in the network like a switch 2575 * duplicating packets or a possible DoS attack. 2576 */ 2577 if (th->th_ack != tp->snd_una || 2578 ((tp->t_flags & TF_SACK_PERMIT) && 2579 !sack_changed)) 2580 break; 2581 else if (!tcp_timer_active(tp, TT_REXMT)) 2582 tp->t_dupacks = 0; 2583 else if (++tp->t_dupacks > tcprexmtthresh || 2584 IN_FASTRECOVERY(tp->t_flags)) { 2585 cc_ack_received(tp, th, nsegs, 2586 CC_DUPACK); 2587 if ((tp->t_flags & TF_SACK_PERMIT) && 2588 IN_FASTRECOVERY(tp->t_flags)) { 2589 int awnd; 2590 2591 /* 2592 * Compute the amount of data in flight first. 2593 * We can inject new data into the pipe iff 2594 * we have less than 1/2 the original window's 2595 * worth of data in flight. 2596 */ 2597 if (V_tcp_do_rfc6675_pipe) 2598 awnd = tcp_compute_pipe(tp); 2599 else 2600 awnd = (tp->snd_nxt - tp->snd_fack) + 2601 tp->sackhint.sack_bytes_rexmit; 2602 2603 if (awnd < tp->snd_ssthresh) { 2604 tp->snd_cwnd += maxseg; 2605 if (tp->snd_cwnd > tp->snd_ssthresh) 2606 tp->snd_cwnd = tp->snd_ssthresh; 2607 } 2608 } else 2609 tp->snd_cwnd += maxseg; 2610 (void) tp->t_fb->tfb_tcp_output(tp); 2611 goto drop; 2612 } else if (tp->t_dupacks == tcprexmtthresh) { 2613 tcp_seq onxt = tp->snd_nxt; 2614 2615 /* 2616 * If we're doing sack, check to 2617 * see if we're already in sack 2618 * recovery. If we're not doing sack, 2619 * check to see if we're in newreno 2620 * recovery. 2621 */ 2622 if (tp->t_flags & TF_SACK_PERMIT) { 2623 if (IN_FASTRECOVERY(tp->t_flags)) { 2624 tp->t_dupacks = 0; 2625 break; 2626 } 2627 } else { 2628 if (SEQ_LEQ(th->th_ack, 2629 tp->snd_recover)) { 2630 tp->t_dupacks = 0; 2631 break; 2632 } 2633 } 2634 /* Congestion signal before ack. */ 2635 cc_cong_signal(tp, th, CC_NDUPACK); 2636 cc_ack_received(tp, th, nsegs, 2637 CC_DUPACK); 2638 tcp_timer_activate(tp, TT_REXMT, 0); 2639 tp->t_rtttime = 0; 2640 if (tp->t_flags & TF_SACK_PERMIT) { 2641 TCPSTAT_INC( 2642 tcps_sack_recovery_episode); 2643 tp->sack_newdata = tp->snd_nxt; 2644 tp->snd_cwnd = maxseg; 2645 (void) tp->t_fb->tfb_tcp_output(tp); 2646 goto drop; 2647 } 2648 tp->snd_nxt = th->th_ack; 2649 tp->snd_cwnd = maxseg; 2650 (void) tp->t_fb->tfb_tcp_output(tp); 2651 KASSERT(tp->snd_limited <= 2, 2652 ("%s: tp->snd_limited too big", 2653 __func__)); 2654 tp->snd_cwnd = tp->snd_ssthresh + 2655 maxseg * 2656 (tp->t_dupacks - tp->snd_limited); 2657 if (SEQ_GT(onxt, tp->snd_nxt)) 2658 tp->snd_nxt = onxt; 2659 goto drop; 2660 } else if (V_tcp_do_rfc3042) { 2661 /* 2662 * Process first and second duplicate 2663 * ACKs. Each indicates a segment 2664 * leaving the network, creating room 2665 * for more. Make sure we can send a 2666 * packet on reception of each duplicate 2667 * ACK by increasing snd_cwnd by one 2668 * segment. Restore the original 2669 * snd_cwnd after packet transmission. 2670 */ 2671 cc_ack_received(tp, th, nsegs, 2672 CC_DUPACK); 2673 uint32_t oldcwnd = tp->snd_cwnd; 2674 tcp_seq oldsndmax = tp->snd_max; 2675 u_int sent; 2676 int avail; 2677 2678 KASSERT(tp->t_dupacks == 1 || 2679 tp->t_dupacks == 2, 2680 ("%s: dupacks not 1 or 2", 2681 __func__)); 2682 if (tp->t_dupacks == 1) 2683 tp->snd_limited = 0; 2684 tp->snd_cwnd = 2685 (tp->snd_nxt - tp->snd_una) + 2686 (tp->t_dupacks - tp->snd_limited) * 2687 maxseg; 2688 /* 2689 * Only call tcp_output when there 2690 * is new data available to be sent. 2691 * Otherwise we would send pure ACKs. 2692 */ 2693 SOCKBUF_LOCK(&so->so_snd); 2694 avail = sbavail(&so->so_snd) - 2695 (tp->snd_nxt - tp->snd_una); 2696 SOCKBUF_UNLOCK(&so->so_snd); 2697 if (avail > 0) 2698 (void) tp->t_fb->tfb_tcp_output(tp); 2699 sent = tp->snd_max - oldsndmax; 2700 if (sent > maxseg) { 2701 KASSERT((tp->t_dupacks == 2 && 2702 tp->snd_limited == 0) || 2703 (sent == maxseg + 1 && 2704 tp->t_flags & TF_SENTFIN), 2705 ("%s: sent too much", 2706 __func__)); 2707 tp->snd_limited = 2; 2708 } else if (sent > 0) 2709 ++tp->snd_limited; 2710 tp->snd_cwnd = oldcwnd; 2711 goto drop; 2712 } 2713 } 2714 break; 2715 } else { 2716 /* 2717 * This ack is advancing the left edge, reset the 2718 * counter. 2719 */ 2720 tp->t_dupacks = 0; 2721 /* 2722 * If this ack also has new SACK info, increment the 2723 * counter as per rfc6675. 2724 */ 2725 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2726 tp->t_dupacks++; 2727 } 2728 2729 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2730 ("%s: th_ack <= snd_una", __func__)); 2731 2732 /* 2733 * If the congestion window was inflated to account 2734 * for the other side's cached packets, retract it. 2735 */ 2736 if (IN_FASTRECOVERY(tp->t_flags)) { 2737 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2738 if (tp->t_flags & TF_SACK_PERMIT) 2739 tcp_sack_partialack(tp, th); 2740 else 2741 tcp_newreno_partial_ack(tp, th); 2742 } else 2743 cc_post_recovery(tp, th); 2744 } 2745 /* 2746 * If we reach this point, ACK is not a duplicate, 2747 * i.e., it ACKs something we sent. 2748 */ 2749 if (tp->t_flags & TF_NEEDSYN) { 2750 /* 2751 * T/TCP: Connection was half-synchronized, and our 2752 * SYN has been ACK'd (so connection is now fully 2753 * synchronized). Go to non-starred state, 2754 * increment snd_una for ACK of SYN, and check if 2755 * we can do window scaling. 2756 */ 2757 tp->t_flags &= ~TF_NEEDSYN; 2758 tp->snd_una++; 2759 /* Do window scaling? */ 2760 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2761 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2762 tp->rcv_scale = tp->request_r_scale; 2763 /* Send window already scaled. */ 2764 } 2765 } 2766 2767 process_ACK: 2768 INP_WLOCK_ASSERT(tp->t_inpcb); 2769 2770 acked = BYTES_THIS_ACK(tp, th); 2771 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2772 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2773 tp->snd_una, th->th_ack, tp, m)); 2774 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2775 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2776 2777 /* 2778 * If we just performed our first retransmit, and the ACK 2779 * arrives within our recovery window, then it was a mistake 2780 * to do the retransmit in the first place. Recover our 2781 * original cwnd and ssthresh, and proceed to transmit where 2782 * we left off. 2783 */ 2784 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2785 (int)(ticks - tp->t_badrxtwin) < 0) 2786 cc_cong_signal(tp, th, CC_RTO_ERR); 2787 2788 /* 2789 * If we have a timestamp reply, update smoothed 2790 * round trip time. If no timestamp is present but 2791 * transmit timer is running and timed sequence 2792 * number was acked, update smoothed round trip time. 2793 * Since we now have an rtt measurement, cancel the 2794 * timer backoff (cf., Phil Karn's retransmit alg.). 2795 * Recompute the initial retransmit timer. 2796 * 2797 * Some boxes send broken timestamp replies 2798 * during the SYN+ACK phase, ignore 2799 * timestamps of 0 or we could calculate a 2800 * huge RTT and blow up the retransmit timer. 2801 */ 2802 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2803 uint32_t t; 2804 2805 t = tcp_ts_getticks() - to.to_tsecr; 2806 if (!tp->t_rttlow || tp->t_rttlow > t) 2807 tp->t_rttlow = t; 2808 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2809 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2810 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2811 tp->t_rttlow = ticks - tp->t_rtttime; 2812 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2813 } 2814 2815 /* 2816 * If all outstanding data is acked, stop retransmit 2817 * timer and remember to restart (more output or persist). 2818 * If there is more data to be acked, restart retransmit 2819 * timer, using current (possibly backed-off) value. 2820 */ 2821 if (th->th_ack == tp->snd_max) { 2822 tcp_timer_activate(tp, TT_REXMT, 0); 2823 needoutput = 1; 2824 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2825 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2826 2827 /* 2828 * If no data (only SYN) was ACK'd, 2829 * skip rest of ACK processing. 2830 */ 2831 if (acked == 0) 2832 goto step6; 2833 2834 /* 2835 * Let the congestion control algorithm update congestion 2836 * control related information. This typically means increasing 2837 * the congestion window. 2838 */ 2839 cc_ack_received(tp, th, nsegs, CC_ACK); 2840 2841 SOCKBUF_LOCK(&so->so_snd); 2842 if (acked > sbavail(&so->so_snd)) { 2843 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2844 tp->snd_wnd -= sbavail(&so->so_snd); 2845 else 2846 tp->snd_wnd = 0; 2847 mfree = sbcut_locked(&so->so_snd, 2848 (int)sbavail(&so->so_snd)); 2849 ourfinisacked = 1; 2850 } else { 2851 mfree = sbcut_locked(&so->so_snd, acked); 2852 if (tp->snd_wnd >= (uint32_t) acked) 2853 tp->snd_wnd -= acked; 2854 else 2855 tp->snd_wnd = 0; 2856 ourfinisacked = 0; 2857 } 2858 /* NB: sowwakeup_locked() does an implicit unlock. */ 2859 sowwakeup_locked(so); 2860 m_freem(mfree); 2861 /* Detect una wraparound. */ 2862 if (!IN_RECOVERY(tp->t_flags) && 2863 SEQ_GT(tp->snd_una, tp->snd_recover) && 2864 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2865 tp->snd_recover = th->th_ack - 1; 2866 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2867 if (IN_RECOVERY(tp->t_flags) && 2868 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2869 EXIT_RECOVERY(tp->t_flags); 2870 } 2871 tp->snd_una = th->th_ack; 2872 if (tp->t_flags & TF_SACK_PERMIT) { 2873 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2874 tp->snd_recover = tp->snd_una; 2875 } 2876 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2877 tp->snd_nxt = tp->snd_una; 2878 2879 switch (tp->t_state) { 2880 2881 /* 2882 * In FIN_WAIT_1 STATE in addition to the processing 2883 * for the ESTABLISHED state if our FIN is now acknowledged 2884 * then enter FIN_WAIT_2. 2885 */ 2886 case TCPS_FIN_WAIT_1: 2887 if (ourfinisacked) { 2888 /* 2889 * If we can't receive any more 2890 * data, then closing user can proceed. 2891 * Starting the timer is contrary to the 2892 * specification, but if we don't get a FIN 2893 * we'll hang forever. 2894 * 2895 * XXXjl: 2896 * we should release the tp also, and use a 2897 * compressed state. 2898 */ 2899 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2900 soisdisconnected(so); 2901 tcp_timer_activate(tp, TT_2MSL, 2902 (tcp_fast_finwait2_recycle ? 2903 tcp_finwait2_timeout : 2904 TP_MAXIDLE(tp))); 2905 } 2906 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2907 } 2908 break; 2909 2910 /* 2911 * In CLOSING STATE in addition to the processing for 2912 * the ESTABLISHED state if the ACK acknowledges our FIN 2913 * then enter the TIME-WAIT state, otherwise ignore 2914 * the segment. 2915 */ 2916 case TCPS_CLOSING: 2917 if (ourfinisacked) { 2918 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2919 tcp_twstart(tp); 2920 INP_INFO_RUNLOCK(&V_tcbinfo); 2921 m_freem(m); 2922 return; 2923 } 2924 break; 2925 2926 /* 2927 * In LAST_ACK, we may still be waiting for data to drain 2928 * and/or to be acked, as well as for the ack of our FIN. 2929 * If our FIN is now acknowledged, delete the TCB, 2930 * enter the closed state and return. 2931 */ 2932 case TCPS_LAST_ACK: 2933 if (ourfinisacked) { 2934 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2935 tp = tcp_close(tp); 2936 goto drop; 2937 } 2938 break; 2939 } 2940 } 2941 2942 step6: 2943 INP_WLOCK_ASSERT(tp->t_inpcb); 2944 2945 /* 2946 * Update window information. 2947 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2948 */ 2949 if ((thflags & TH_ACK) && 2950 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2951 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2952 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2953 /* keep track of pure window updates */ 2954 if (tlen == 0 && 2955 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2956 TCPSTAT_INC(tcps_rcvwinupd); 2957 tp->snd_wnd = tiwin; 2958 tp->snd_wl1 = th->th_seq; 2959 tp->snd_wl2 = th->th_ack; 2960 if (tp->snd_wnd > tp->max_sndwnd) 2961 tp->max_sndwnd = tp->snd_wnd; 2962 needoutput = 1; 2963 } 2964 2965 /* 2966 * Process segments with URG. 2967 */ 2968 if ((thflags & TH_URG) && th->th_urp && 2969 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2970 /* 2971 * This is a kludge, but if we receive and accept 2972 * random urgent pointers, we'll crash in 2973 * soreceive. It's hard to imagine someone 2974 * actually wanting to send this much urgent data. 2975 */ 2976 SOCKBUF_LOCK(&so->so_rcv); 2977 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2978 th->th_urp = 0; /* XXX */ 2979 thflags &= ~TH_URG; /* XXX */ 2980 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2981 goto dodata; /* XXX */ 2982 } 2983 /* 2984 * If this segment advances the known urgent pointer, 2985 * then mark the data stream. This should not happen 2986 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2987 * a FIN has been received from the remote side. 2988 * In these states we ignore the URG. 2989 * 2990 * According to RFC961 (Assigned Protocols), 2991 * the urgent pointer points to the last octet 2992 * of urgent data. We continue, however, 2993 * to consider it to indicate the first octet 2994 * of data past the urgent section as the original 2995 * spec states (in one of two places). 2996 */ 2997 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2998 tp->rcv_up = th->th_seq + th->th_urp; 2999 so->so_oobmark = sbavail(&so->so_rcv) + 3000 (tp->rcv_up - tp->rcv_nxt) - 1; 3001 if (so->so_oobmark == 0) 3002 so->so_rcv.sb_state |= SBS_RCVATMARK; 3003 sohasoutofband(so); 3004 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3005 } 3006 SOCKBUF_UNLOCK(&so->so_rcv); 3007 /* 3008 * Remove out of band data so doesn't get presented to user. 3009 * This can happen independent of advancing the URG pointer, 3010 * but if two URG's are pending at once, some out-of-band 3011 * data may creep in... ick. 3012 */ 3013 if (th->th_urp <= (uint32_t)tlen && 3014 !(so->so_options & SO_OOBINLINE)) { 3015 /* hdr drop is delayed */ 3016 tcp_pulloutofband(so, th, m, drop_hdrlen); 3017 } 3018 } else { 3019 /* 3020 * If no out of band data is expected, 3021 * pull receive urgent pointer along 3022 * with the receive window. 3023 */ 3024 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3025 tp->rcv_up = tp->rcv_nxt; 3026 } 3027 dodata: /* XXX */ 3028 INP_WLOCK_ASSERT(tp->t_inpcb); 3029 3030 /* 3031 * Process the segment text, merging it into the TCP sequencing queue, 3032 * and arranging for acknowledgment of receipt if necessary. 3033 * This process logically involves adjusting tp->rcv_wnd as data 3034 * is presented to the user (this happens in tcp_usrreq.c, 3035 * case PRU_RCVD). If a FIN has already been received on this 3036 * connection then we just ignore the text. 3037 */ 3038 #ifdef TCP_RFC7413 3039 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3040 IS_FASTOPEN(tp->t_flags)); 3041 #else 3042 #define tfo_syn (false) 3043 #endif 3044 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3045 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3046 tcp_seq save_start = th->th_seq; 3047 m_adj(m, drop_hdrlen); /* delayed header drop */ 3048 /* 3049 * Insert segment which includes th into TCP reassembly queue 3050 * with control block tp. Set thflags to whether reassembly now 3051 * includes a segment with FIN. This handles the common case 3052 * inline (segment is the next to be received on an established 3053 * connection, and the queue is empty), avoiding linkage into 3054 * and removal from the queue and repetition of various 3055 * conversions. 3056 * Set DELACK for segments received in order, but ack 3057 * immediately when segments are out of order (so 3058 * fast retransmit can work). 3059 */ 3060 if (th->th_seq == tp->rcv_nxt && 3061 LIST_EMPTY(&tp->t_segq) && 3062 (TCPS_HAVEESTABLISHED(tp->t_state) || 3063 tfo_syn)) { 3064 if (DELAY_ACK(tp, tlen) || tfo_syn) 3065 tp->t_flags |= TF_DELACK; 3066 else 3067 tp->t_flags |= TF_ACKNOW; 3068 tp->rcv_nxt += tlen; 3069 thflags = th->th_flags & TH_FIN; 3070 TCPSTAT_INC(tcps_rcvpack); 3071 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3072 SOCKBUF_LOCK(&so->so_rcv); 3073 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3074 m_freem(m); 3075 else 3076 sbappendstream_locked(&so->so_rcv, m, 0); 3077 /* NB: sorwakeup_locked() does an implicit unlock. */ 3078 sorwakeup_locked(so); 3079 } else { 3080 /* 3081 * XXX: Due to the header drop above "th" is 3082 * theoretically invalid by now. Fortunately 3083 * m_adj() doesn't actually frees any mbufs 3084 * when trimming from the head. 3085 */ 3086 thflags = tcp_reass(tp, th, &tlen, m); 3087 tp->t_flags |= TF_ACKNOW; 3088 } 3089 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3090 tcp_update_sack_list(tp, save_start, save_start + tlen); 3091 #if 0 3092 /* 3093 * Note the amount of data that peer has sent into 3094 * our window, in order to estimate the sender's 3095 * buffer size. 3096 * XXX: Unused. 3097 */ 3098 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3099 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3100 else 3101 len = so->so_rcv.sb_hiwat; 3102 #endif 3103 } else { 3104 m_freem(m); 3105 thflags &= ~TH_FIN; 3106 } 3107 3108 /* 3109 * If FIN is received ACK the FIN and let the user know 3110 * that the connection is closing. 3111 */ 3112 if (thflags & TH_FIN) { 3113 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3114 socantrcvmore(so); 3115 /* 3116 * If connection is half-synchronized 3117 * (ie NEEDSYN flag on) then delay ACK, 3118 * so it may be piggybacked when SYN is sent. 3119 * Otherwise, since we received a FIN then no 3120 * more input can be expected, send ACK now. 3121 */ 3122 if (tp->t_flags & TF_NEEDSYN) 3123 tp->t_flags |= TF_DELACK; 3124 else 3125 tp->t_flags |= TF_ACKNOW; 3126 tp->rcv_nxt++; 3127 } 3128 switch (tp->t_state) { 3129 3130 /* 3131 * In SYN_RECEIVED and ESTABLISHED STATES 3132 * enter the CLOSE_WAIT state. 3133 */ 3134 case TCPS_SYN_RECEIVED: 3135 tp->t_starttime = ticks; 3136 /* FALLTHROUGH */ 3137 case TCPS_ESTABLISHED: 3138 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3139 break; 3140 3141 /* 3142 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3143 * enter the CLOSING state. 3144 */ 3145 case TCPS_FIN_WAIT_1: 3146 tcp_state_change(tp, TCPS_CLOSING); 3147 break; 3148 3149 /* 3150 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3151 * starting the time-wait timer, turning off the other 3152 * standard timers. 3153 */ 3154 case TCPS_FIN_WAIT_2: 3155 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3156 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3157 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3158 ti_locked)); 3159 3160 tcp_twstart(tp); 3161 INP_INFO_RUNLOCK(&V_tcbinfo); 3162 return; 3163 } 3164 } 3165 if (ti_locked == TI_RLOCKED) 3166 INP_INFO_RUNLOCK(&V_tcbinfo); 3167 ti_locked = TI_UNLOCKED; 3168 3169 #ifdef TCPDEBUG 3170 if (so->so_options & SO_DEBUG) 3171 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3172 &tcp_savetcp, 0); 3173 #endif 3174 TCP_PROBE3(debug__input, tp, th, m); 3175 3176 /* 3177 * Return any desired output. 3178 */ 3179 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3180 (void) tp->t_fb->tfb_tcp_output(tp); 3181 3182 check_delack: 3183 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3184 __func__, ti_locked)); 3185 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3186 INP_WLOCK_ASSERT(tp->t_inpcb); 3187 3188 if (tp->t_flags & TF_DELACK) { 3189 tp->t_flags &= ~TF_DELACK; 3190 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3191 } 3192 INP_WUNLOCK(tp->t_inpcb); 3193 return; 3194 3195 dropafterack: 3196 /* 3197 * Generate an ACK dropping incoming segment if it occupies 3198 * sequence space, where the ACK reflects our state. 3199 * 3200 * We can now skip the test for the RST flag since all 3201 * paths to this code happen after packets containing 3202 * RST have been dropped. 3203 * 3204 * In the SYN-RECEIVED state, don't send an ACK unless the 3205 * segment we received passes the SYN-RECEIVED ACK test. 3206 * If it fails send a RST. This breaks the loop in the 3207 * "LAND" DoS attack, and also prevents an ACK storm 3208 * between two listening ports that have been sent forged 3209 * SYN segments, each with the source address of the other. 3210 */ 3211 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3212 (SEQ_GT(tp->snd_una, th->th_ack) || 3213 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3214 rstreason = BANDLIM_RST_OPENPORT; 3215 goto dropwithreset; 3216 } 3217 #ifdef TCPDEBUG 3218 if (so->so_options & SO_DEBUG) 3219 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3220 &tcp_savetcp, 0); 3221 #endif 3222 TCP_PROBE3(debug__input, tp, th, m); 3223 if (ti_locked == TI_RLOCKED) 3224 INP_INFO_RUNLOCK(&V_tcbinfo); 3225 ti_locked = TI_UNLOCKED; 3226 3227 tp->t_flags |= TF_ACKNOW; 3228 (void) tp->t_fb->tfb_tcp_output(tp); 3229 INP_WUNLOCK(tp->t_inpcb); 3230 m_freem(m); 3231 return; 3232 3233 dropwithreset: 3234 if (ti_locked == TI_RLOCKED) 3235 INP_INFO_RUNLOCK(&V_tcbinfo); 3236 ti_locked = TI_UNLOCKED; 3237 3238 if (tp != NULL) { 3239 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3240 INP_WUNLOCK(tp->t_inpcb); 3241 } else 3242 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3243 return; 3244 3245 drop: 3246 if (ti_locked == TI_RLOCKED) { 3247 INP_INFO_RUNLOCK(&V_tcbinfo); 3248 ti_locked = TI_UNLOCKED; 3249 } 3250 #ifdef INVARIANTS 3251 else 3252 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3253 #endif 3254 3255 /* 3256 * Drop space held by incoming segment and return. 3257 */ 3258 #ifdef TCPDEBUG 3259 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3260 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3261 &tcp_savetcp, 0); 3262 #endif 3263 TCP_PROBE3(debug__input, tp, th, m); 3264 if (tp != NULL) 3265 INP_WUNLOCK(tp->t_inpcb); 3266 m_freem(m); 3267 #ifndef TCP_RFC7413 3268 #undef tfo_syn 3269 #endif 3270 } 3271 3272 /* 3273 * Issue RST and make ACK acceptable to originator of segment. 3274 * The mbuf must still include the original packet header. 3275 * tp may be NULL. 3276 */ 3277 void 3278 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3279 int tlen, int rstreason) 3280 { 3281 #ifdef INET 3282 struct ip *ip; 3283 #endif 3284 #ifdef INET6 3285 struct ip6_hdr *ip6; 3286 #endif 3287 3288 if (tp != NULL) { 3289 INP_WLOCK_ASSERT(tp->t_inpcb); 3290 } 3291 3292 /* Don't bother if destination was broadcast/multicast. */ 3293 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3294 goto drop; 3295 #ifdef INET6 3296 if (mtod(m, struct ip *)->ip_v == 6) { 3297 ip6 = mtod(m, struct ip6_hdr *); 3298 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3299 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3300 goto drop; 3301 /* IPv6 anycast check is done at tcp6_input() */ 3302 } 3303 #endif 3304 #if defined(INET) && defined(INET6) 3305 else 3306 #endif 3307 #ifdef INET 3308 { 3309 ip = mtod(m, struct ip *); 3310 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3311 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3312 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3313 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3314 goto drop; 3315 } 3316 #endif 3317 3318 /* Perform bandwidth limiting. */ 3319 if (badport_bandlim(rstreason) < 0) 3320 goto drop; 3321 3322 /* tcp_respond consumes the mbuf chain. */ 3323 if (th->th_flags & TH_ACK) { 3324 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3325 th->th_ack, TH_RST); 3326 } else { 3327 if (th->th_flags & TH_SYN) 3328 tlen++; 3329 if (th->th_flags & TH_FIN) 3330 tlen++; 3331 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3332 (tcp_seq)0, TH_RST|TH_ACK); 3333 } 3334 return; 3335 drop: 3336 m_freem(m); 3337 } 3338 3339 /* 3340 * Parse TCP options and place in tcpopt. 3341 */ 3342 void 3343 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3344 { 3345 int opt, optlen; 3346 3347 to->to_flags = 0; 3348 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3349 opt = cp[0]; 3350 if (opt == TCPOPT_EOL) 3351 break; 3352 if (opt == TCPOPT_NOP) 3353 optlen = 1; 3354 else { 3355 if (cnt < 2) 3356 break; 3357 optlen = cp[1]; 3358 if (optlen < 2 || optlen > cnt) 3359 break; 3360 } 3361 switch (opt) { 3362 case TCPOPT_MAXSEG: 3363 if (optlen != TCPOLEN_MAXSEG) 3364 continue; 3365 if (!(flags & TO_SYN)) 3366 continue; 3367 to->to_flags |= TOF_MSS; 3368 bcopy((char *)cp + 2, 3369 (char *)&to->to_mss, sizeof(to->to_mss)); 3370 to->to_mss = ntohs(to->to_mss); 3371 break; 3372 case TCPOPT_WINDOW: 3373 if (optlen != TCPOLEN_WINDOW) 3374 continue; 3375 if (!(flags & TO_SYN)) 3376 continue; 3377 to->to_flags |= TOF_SCALE; 3378 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3379 break; 3380 case TCPOPT_TIMESTAMP: 3381 if (optlen != TCPOLEN_TIMESTAMP) 3382 continue; 3383 to->to_flags |= TOF_TS; 3384 bcopy((char *)cp + 2, 3385 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3386 to->to_tsval = ntohl(to->to_tsval); 3387 bcopy((char *)cp + 6, 3388 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3389 to->to_tsecr = ntohl(to->to_tsecr); 3390 break; 3391 case TCPOPT_SIGNATURE: 3392 /* 3393 * In order to reply to a host which has set the 3394 * TCP_SIGNATURE option in its initial SYN, we have 3395 * to record the fact that the option was observed 3396 * here for the syncache code to perform the correct 3397 * response. 3398 */ 3399 if (optlen != TCPOLEN_SIGNATURE) 3400 continue; 3401 to->to_flags |= TOF_SIGNATURE; 3402 to->to_signature = cp + 2; 3403 break; 3404 case TCPOPT_SACK_PERMITTED: 3405 if (optlen != TCPOLEN_SACK_PERMITTED) 3406 continue; 3407 if (!(flags & TO_SYN)) 3408 continue; 3409 if (!V_tcp_do_sack) 3410 continue; 3411 to->to_flags |= TOF_SACKPERM; 3412 break; 3413 case TCPOPT_SACK: 3414 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3415 continue; 3416 if (flags & TO_SYN) 3417 continue; 3418 to->to_flags |= TOF_SACK; 3419 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3420 to->to_sacks = cp + 2; 3421 TCPSTAT_INC(tcps_sack_rcv_blocks); 3422 break; 3423 #ifdef TCP_RFC7413 3424 case TCPOPT_FAST_OPEN: 3425 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3426 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3427 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3428 continue; 3429 if (!(flags & TO_SYN)) 3430 continue; 3431 if (!V_tcp_fastopen_enabled) 3432 continue; 3433 to->to_flags |= TOF_FASTOPEN; 3434 to->to_tfo_len = optlen - 2; 3435 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3436 break; 3437 #endif 3438 default: 3439 continue; 3440 } 3441 } 3442 } 3443 3444 /* 3445 * Pull out of band byte out of a segment so 3446 * it doesn't appear in the user's data queue. 3447 * It is still reflected in the segment length for 3448 * sequencing purposes. 3449 */ 3450 void 3451 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3452 int off) 3453 { 3454 int cnt = off + th->th_urp - 1; 3455 3456 while (cnt >= 0) { 3457 if (m->m_len > cnt) { 3458 char *cp = mtod(m, caddr_t) + cnt; 3459 struct tcpcb *tp = sototcpcb(so); 3460 3461 INP_WLOCK_ASSERT(tp->t_inpcb); 3462 3463 tp->t_iobc = *cp; 3464 tp->t_oobflags |= TCPOOB_HAVEDATA; 3465 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3466 m->m_len--; 3467 if (m->m_flags & M_PKTHDR) 3468 m->m_pkthdr.len--; 3469 return; 3470 } 3471 cnt -= m->m_len; 3472 m = m->m_next; 3473 if (m == NULL) 3474 break; 3475 } 3476 panic("tcp_pulloutofband"); 3477 } 3478 3479 /* 3480 * Collect new round-trip time estimate 3481 * and update averages and current timeout. 3482 */ 3483 void 3484 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3485 { 3486 int delta; 3487 3488 INP_WLOCK_ASSERT(tp->t_inpcb); 3489 3490 TCPSTAT_INC(tcps_rttupdated); 3491 tp->t_rttupdated++; 3492 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3493 /* 3494 * srtt is stored as fixed point with 5 bits after the 3495 * binary point (i.e., scaled by 8). The following magic 3496 * is equivalent to the smoothing algorithm in rfc793 with 3497 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3498 * point). Adjust rtt to origin 0. 3499 */ 3500 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3501 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3502 3503 if ((tp->t_srtt += delta) <= 0) 3504 tp->t_srtt = 1; 3505 3506 /* 3507 * We accumulate a smoothed rtt variance (actually, a 3508 * smoothed mean difference), then set the retransmit 3509 * timer to smoothed rtt + 4 times the smoothed variance. 3510 * rttvar is stored as fixed point with 4 bits after the 3511 * binary point (scaled by 16). The following is 3512 * equivalent to rfc793 smoothing with an alpha of .75 3513 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3514 * rfc793's wired-in beta. 3515 */ 3516 if (delta < 0) 3517 delta = -delta; 3518 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3519 if ((tp->t_rttvar += delta) <= 0) 3520 tp->t_rttvar = 1; 3521 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3522 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3523 } else { 3524 /* 3525 * No rtt measurement yet - use the unsmoothed rtt. 3526 * Set the variance to half the rtt (so our first 3527 * retransmit happens at 3*rtt). 3528 */ 3529 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3530 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3531 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3532 } 3533 tp->t_rtttime = 0; 3534 tp->t_rxtshift = 0; 3535 3536 /* 3537 * the retransmit should happen at rtt + 4 * rttvar. 3538 * Because of the way we do the smoothing, srtt and rttvar 3539 * will each average +1/2 tick of bias. When we compute 3540 * the retransmit timer, we want 1/2 tick of rounding and 3541 * 1 extra tick because of +-1/2 tick uncertainty in the 3542 * firing of the timer. The bias will give us exactly the 3543 * 1.5 tick we need. But, because the bias is 3544 * statistical, we have to test that we don't drop below 3545 * the minimum feasible timer (which is 2 ticks). 3546 */ 3547 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3548 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3549 3550 /* 3551 * We received an ack for a packet that wasn't retransmitted; 3552 * it is probably safe to discard any error indications we've 3553 * received recently. This isn't quite right, but close enough 3554 * for now (a route might have failed after we sent a segment, 3555 * and the return path might not be symmetrical). 3556 */ 3557 tp->t_softerror = 0; 3558 } 3559 3560 /* 3561 * Determine a reasonable value for maxseg size. 3562 * If the route is known, check route for mtu. 3563 * If none, use an mss that can be handled on the outgoing interface 3564 * without forcing IP to fragment. If no route is found, route has no mtu, 3565 * or the destination isn't local, use a default, hopefully conservative 3566 * size (usually 512 or the default IP max size, but no more than the mtu 3567 * of the interface), as we can't discover anything about intervening 3568 * gateways or networks. We also initialize the congestion/slow start 3569 * window to be a single segment if the destination isn't local. 3570 * While looking at the routing entry, we also initialize other path-dependent 3571 * parameters from pre-set or cached values in the routing entry. 3572 * 3573 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3574 * IP options, e.g. IPSEC data, since length of this data may vary, and 3575 * thus it is calculated for every segment separately in tcp_output(). 3576 * 3577 * NOTE that this routine is only called when we process an incoming 3578 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3579 * settings are handled in tcp_mssopt(). 3580 */ 3581 void 3582 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3583 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3584 { 3585 int mss = 0; 3586 uint32_t maxmtu = 0; 3587 struct inpcb *inp = tp->t_inpcb; 3588 struct hc_metrics_lite metrics; 3589 #ifdef INET6 3590 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3591 size_t min_protoh = isipv6 ? 3592 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3593 sizeof (struct tcpiphdr); 3594 #else 3595 const size_t min_protoh = sizeof(struct tcpiphdr); 3596 #endif 3597 3598 INP_WLOCK_ASSERT(tp->t_inpcb); 3599 3600 if (mtuoffer != -1) { 3601 KASSERT(offer == -1, ("%s: conflict", __func__)); 3602 offer = mtuoffer - min_protoh; 3603 } 3604 3605 /* Initialize. */ 3606 #ifdef INET6 3607 if (isipv6) { 3608 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3609 tp->t_maxseg = V_tcp_v6mssdflt; 3610 } 3611 #endif 3612 #if defined(INET) && defined(INET6) 3613 else 3614 #endif 3615 #ifdef INET 3616 { 3617 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3618 tp->t_maxseg = V_tcp_mssdflt; 3619 } 3620 #endif 3621 3622 /* 3623 * No route to sender, stay with default mss and return. 3624 */ 3625 if (maxmtu == 0) { 3626 /* 3627 * In case we return early we need to initialize metrics 3628 * to a defined state as tcp_hc_get() would do for us 3629 * if there was no cache hit. 3630 */ 3631 if (metricptr != NULL) 3632 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3633 return; 3634 } 3635 3636 /* What have we got? */ 3637 switch (offer) { 3638 case 0: 3639 /* 3640 * Offer == 0 means that there was no MSS on the SYN 3641 * segment, in this case we use tcp_mssdflt as 3642 * already assigned to t_maxseg above. 3643 */ 3644 offer = tp->t_maxseg; 3645 break; 3646 3647 case -1: 3648 /* 3649 * Offer == -1 means that we didn't receive SYN yet. 3650 */ 3651 /* FALLTHROUGH */ 3652 3653 default: 3654 /* 3655 * Prevent DoS attack with too small MSS. Round up 3656 * to at least minmss. 3657 */ 3658 offer = max(offer, V_tcp_minmss); 3659 } 3660 3661 /* 3662 * rmx information is now retrieved from tcp_hostcache. 3663 */ 3664 tcp_hc_get(&inp->inp_inc, &metrics); 3665 if (metricptr != NULL) 3666 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3667 3668 /* 3669 * If there's a discovered mtu in tcp hostcache, use it. 3670 * Else, use the link mtu. 3671 */ 3672 if (metrics.rmx_mtu) 3673 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3674 else { 3675 #ifdef INET6 3676 if (isipv6) { 3677 mss = maxmtu - min_protoh; 3678 if (!V_path_mtu_discovery && 3679 !in6_localaddr(&inp->in6p_faddr)) 3680 mss = min(mss, V_tcp_v6mssdflt); 3681 } 3682 #endif 3683 #if defined(INET) && defined(INET6) 3684 else 3685 #endif 3686 #ifdef INET 3687 { 3688 mss = maxmtu - min_protoh; 3689 if (!V_path_mtu_discovery && 3690 !in_localaddr(inp->inp_faddr)) 3691 mss = min(mss, V_tcp_mssdflt); 3692 } 3693 #endif 3694 /* 3695 * XXX - The above conditional (mss = maxmtu - min_protoh) 3696 * probably violates the TCP spec. 3697 * The problem is that, since we don't know the 3698 * other end's MSS, we are supposed to use a conservative 3699 * default. But, if we do that, then MTU discovery will 3700 * never actually take place, because the conservative 3701 * default is much less than the MTUs typically seen 3702 * on the Internet today. For the moment, we'll sweep 3703 * this under the carpet. 3704 * 3705 * The conservative default might not actually be a problem 3706 * if the only case this occurs is when sending an initial 3707 * SYN with options and data to a host we've never talked 3708 * to before. Then, they will reply with an MSS value which 3709 * will get recorded and the new parameters should get 3710 * recomputed. For Further Study. 3711 */ 3712 } 3713 mss = min(mss, offer); 3714 3715 /* 3716 * Sanity check: make sure that maxseg will be large 3717 * enough to allow some data on segments even if the 3718 * all the option space is used (40bytes). Otherwise 3719 * funny things may happen in tcp_output. 3720 * 3721 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3722 */ 3723 mss = max(mss, 64); 3724 3725 tp->t_maxseg = mss; 3726 } 3727 3728 void 3729 tcp_mss(struct tcpcb *tp, int offer) 3730 { 3731 int mss; 3732 uint32_t bufsize; 3733 struct inpcb *inp; 3734 struct socket *so; 3735 struct hc_metrics_lite metrics; 3736 struct tcp_ifcap cap; 3737 3738 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3739 3740 bzero(&cap, sizeof(cap)); 3741 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3742 3743 mss = tp->t_maxseg; 3744 inp = tp->t_inpcb; 3745 3746 /* 3747 * If there's a pipesize, change the socket buffer to that size, 3748 * don't change if sb_hiwat is different than default (then it 3749 * has been changed on purpose with setsockopt). 3750 * Make the socket buffers an integral number of mss units; 3751 * if the mss is larger than the socket buffer, decrease the mss. 3752 */ 3753 so = inp->inp_socket; 3754 SOCKBUF_LOCK(&so->so_snd); 3755 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3756 bufsize = metrics.rmx_sendpipe; 3757 else 3758 bufsize = so->so_snd.sb_hiwat; 3759 if (bufsize < mss) 3760 mss = bufsize; 3761 else { 3762 bufsize = roundup(bufsize, mss); 3763 if (bufsize > sb_max) 3764 bufsize = sb_max; 3765 if (bufsize > so->so_snd.sb_hiwat) 3766 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3767 } 3768 SOCKBUF_UNLOCK(&so->so_snd); 3769 /* 3770 * Sanity check: make sure that maxseg will be large 3771 * enough to allow some data on segments even if the 3772 * all the option space is used (40bytes). Otherwise 3773 * funny things may happen in tcp_output. 3774 * 3775 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3776 */ 3777 tp->t_maxseg = max(mss, 64); 3778 3779 SOCKBUF_LOCK(&so->so_rcv); 3780 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3781 bufsize = metrics.rmx_recvpipe; 3782 else 3783 bufsize = so->so_rcv.sb_hiwat; 3784 if (bufsize > mss) { 3785 bufsize = roundup(bufsize, mss); 3786 if (bufsize > sb_max) 3787 bufsize = sb_max; 3788 if (bufsize > so->so_rcv.sb_hiwat) 3789 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3790 } 3791 SOCKBUF_UNLOCK(&so->so_rcv); 3792 3793 /* Check the interface for TSO capabilities. */ 3794 if (cap.ifcap & CSUM_TSO) { 3795 tp->t_flags |= TF_TSO; 3796 tp->t_tsomax = cap.tsomax; 3797 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3798 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3799 } 3800 } 3801 3802 /* 3803 * Determine the MSS option to send on an outgoing SYN. 3804 */ 3805 int 3806 tcp_mssopt(struct in_conninfo *inc) 3807 { 3808 int mss = 0; 3809 uint32_t thcmtu = 0; 3810 uint32_t maxmtu = 0; 3811 size_t min_protoh; 3812 3813 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3814 3815 #ifdef INET6 3816 if (inc->inc_flags & INC_ISIPV6) { 3817 mss = V_tcp_v6mssdflt; 3818 maxmtu = tcp_maxmtu6(inc, NULL); 3819 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3820 } 3821 #endif 3822 #if defined(INET) && defined(INET6) 3823 else 3824 #endif 3825 #ifdef INET 3826 { 3827 mss = V_tcp_mssdflt; 3828 maxmtu = tcp_maxmtu(inc, NULL); 3829 min_protoh = sizeof(struct tcpiphdr); 3830 } 3831 #endif 3832 #if defined(INET6) || defined(INET) 3833 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3834 #endif 3835 3836 if (maxmtu && thcmtu) 3837 mss = min(maxmtu, thcmtu) - min_protoh; 3838 else if (maxmtu || thcmtu) 3839 mss = max(maxmtu, thcmtu) - min_protoh; 3840 3841 return (mss); 3842 } 3843 3844 3845 /* 3846 * On a partial ack arrives, force the retransmission of the 3847 * next unacknowledged segment. Do not clear tp->t_dupacks. 3848 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3849 * be started again. 3850 */ 3851 void 3852 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3853 { 3854 tcp_seq onxt = tp->snd_nxt; 3855 uint32_t ocwnd = tp->snd_cwnd; 3856 u_int maxseg = tcp_maxseg(tp); 3857 3858 INP_WLOCK_ASSERT(tp->t_inpcb); 3859 3860 tcp_timer_activate(tp, TT_REXMT, 0); 3861 tp->t_rtttime = 0; 3862 tp->snd_nxt = th->th_ack; 3863 /* 3864 * Set snd_cwnd to one segment beyond acknowledged offset. 3865 * (tp->snd_una has not yet been updated when this function is called.) 3866 */ 3867 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3868 tp->t_flags |= TF_ACKNOW; 3869 (void) tp->t_fb->tfb_tcp_output(tp); 3870 tp->snd_cwnd = ocwnd; 3871 if (SEQ_GT(onxt, tp->snd_nxt)) 3872 tp->snd_nxt = onxt; 3873 /* 3874 * Partial window deflation. Relies on fact that tp->snd_una 3875 * not updated yet. 3876 */ 3877 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3878 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3879 else 3880 tp->snd_cwnd = 0; 3881 tp->snd_cwnd += maxseg; 3882 } 3883 3884 int 3885 tcp_compute_pipe(struct tcpcb *tp) 3886 { 3887 return (tp->snd_max - tp->snd_una + 3888 tp->sackhint.sack_bytes_rexmit - 3889 tp->sackhint.sacked_bytes); 3890 } 3891