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