1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 3. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_inet.h" 54 #include "opt_inet6.h" 55 #include "opt_ipsec.h" 56 #include "opt_tcpdebug.h" 57 58 #include <sys/param.h> 59 #include <sys/kernel.h> 60 #ifdef TCP_HHOOK 61 #include <sys/hhook.h> 62 #endif 63 #include <sys/malloc.h> 64 #include <sys/mbuf.h> 65 #include <sys/proc.h> /* for proc0 declaration */ 66 #include <sys/protosw.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77 #include <vm/uma.h> 78 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/route.h> 82 #include <net/vnet.h> 83 84 #define TCPSTATES /* for logging */ 85 86 #include <netinet/in.h> 87 #include <netinet/in_kdtrace.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/ip.h> 91 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 92 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 93 #include <netinet/ip_var.h> 94 #include <netinet/ip_options.h> 95 #include <netinet/ip6.h> 96 #include <netinet/icmp6.h> 97 #include <netinet6/in6_pcb.h> 98 #include <netinet6/in6_var.h> 99 #include <netinet6/ip6_var.h> 100 #include <netinet6/nd6.h> 101 #ifdef TCP_RFC7413 102 #include <netinet/tcp_fastopen.h> 103 #endif 104 #include <netinet/tcp.h> 105 #include <netinet/tcp_fsm.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 #ifdef TCPPCAP 113 #include <netinet/tcp_pcap.h> 114 #endif 115 #include <netinet/tcp_syncache.h> 116 #ifdef TCPDEBUG 117 #include <netinet/tcp_debug.h> 118 #endif /* TCPDEBUG */ 119 #ifdef TCP_OFFLOAD 120 #include <netinet/tcp_offload.h> 121 #endif 122 123 #include <netipsec/ipsec_support.h> 124 125 #include <machine/in_cksum.h> 126 127 #include <security/mac/mac_framework.h> 128 129 const int tcprexmtthresh = 3; 130 131 int tcp_log_in_vain = 0; 132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 133 &tcp_log_in_vain, 0, 134 "Log all incoming TCP segments to closed ports"); 135 136 VNET_DEFINE(int, blackhole) = 0; 137 #define V_blackhole VNET(blackhole) 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(blackhole), 0, 140 "Do not send RST on segments to closed ports"); 141 142 VNET_DEFINE(int, tcp_delack_enabled) = 1; 143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 144 &VNET_NAME(tcp_delack_enabled), 0, 145 "Delay ACK to try and piggyback it onto a data packet"); 146 147 VNET_DEFINE(int, drop_synfin) = 0; 148 #define V_drop_synfin VNET(drop_synfin) 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 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 161 &VNET_NAME(tcp_do_rfc3042), 0, 162 "Enable RFC 3042 (Limited Transmit)"); 163 164 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 166 &VNET_NAME(tcp_do_rfc3390), 0, 167 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 168 169 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 171 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 172 "Slow-start flight size (initial congestion window) in number of segments"); 173 174 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 176 &VNET_NAME(tcp_do_rfc3465), 0, 177 "Enable RFC 3465 (Appropriate Byte Counting)"); 178 179 VNET_DEFINE(int, tcp_abc_l_var) = 2; 180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_abc_l_var), 2, 182 "Cap the max cwnd increment during slow-start to this number of segments"); 183 184 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 185 186 VNET_DEFINE(int, tcp_do_ecn) = 2; 187 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 188 &VNET_NAME(tcp_do_ecn), 0, 189 "TCP ECN support"); 190 191 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 192 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_ecn_maxretries), 0, 194 "Max retries before giving up on ECN"); 195 196 VNET_DEFINE(int, tcp_insecure_syn) = 0; 197 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_insecure_syn), 0, 200 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 201 202 VNET_DEFINE(int, tcp_insecure_rst) = 0; 203 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 205 &VNET_NAME(tcp_insecure_rst), 0, 206 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 207 208 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 209 #define V_tcp_recvspace VNET(tcp_recvspace) 210 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 211 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 212 213 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 214 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 216 &VNET_NAME(tcp_do_autorcvbuf), 0, 217 "Enable automatic receive buffer sizing"); 218 219 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 220 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_autorcvbuf_inc), 0, 223 "Incrementor step size of automatic receive buffer"); 224 225 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 226 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_autorcvbuf_max), 0, 229 "Max size of automatic receive buffer"); 230 231 VNET_DEFINE(struct inpcbhead, tcb); 232 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 233 VNET_DEFINE(struct inpcbinfo, tcbinfo); 234 235 /* 236 * TCP statistics are stored in an array of counter(9)s, which size matches 237 * size of struct tcpstat. TCP running connection count is a regular array. 238 */ 239 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 240 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 241 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 242 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 243 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 244 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 245 "TCP connection counts by TCP state"); 246 247 static void 248 tcp_vnet_init(const void *unused) 249 { 250 251 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 252 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 253 } 254 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 255 tcp_vnet_init, NULL); 256 257 #ifdef VIMAGE 258 static void 259 tcp_vnet_uninit(const void *unused) 260 { 261 262 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 263 VNET_PCPUSTAT_FREE(tcpstat); 264 } 265 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 266 tcp_vnet_uninit, NULL); 267 #endif /* VIMAGE */ 268 269 /* 270 * Kernel module interface for updating tcpstat. The argument is an index 271 * into tcpstat treated as an array. 272 */ 273 void 274 kmod_tcpstat_inc(int statnum) 275 { 276 277 counter_u64_add(VNET(tcpstat)[statnum], 1); 278 } 279 280 #ifdef TCP_HHOOK 281 /* 282 * Wrapper for the TCP established input helper hook. 283 */ 284 void 285 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 286 { 287 struct tcp_hhook_data hhook_data; 288 289 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 290 hhook_data.tp = tp; 291 hhook_data.th = th; 292 hhook_data.to = to; 293 294 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 295 tp->osd); 296 } 297 } 298 #endif 299 300 /* 301 * CC wrapper hook functions 302 */ 303 void 304 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 305 uint16_t type) 306 { 307 INP_WLOCK_ASSERT(tp->t_inpcb); 308 309 tp->ccv->nsegs = nsegs; 310 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 311 if (tp->snd_cwnd <= tp->snd_wnd) 312 tp->ccv->flags |= CCF_CWND_LIMITED; 313 else 314 tp->ccv->flags &= ~CCF_CWND_LIMITED; 315 316 if (type == CC_ACK) { 317 if (tp->snd_cwnd > tp->snd_ssthresh) { 318 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 319 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 320 if (tp->t_bytes_acked >= tp->snd_cwnd) { 321 tp->t_bytes_acked -= tp->snd_cwnd; 322 tp->ccv->flags |= CCF_ABC_SENTAWND; 323 } 324 } else { 325 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 326 tp->t_bytes_acked = 0; 327 } 328 } 329 330 if (CC_ALGO(tp)->ack_received != NULL) { 331 /* XXXLAS: Find a way to live without this */ 332 tp->ccv->curack = th->th_ack; 333 CC_ALGO(tp)->ack_received(tp->ccv, type); 334 } 335 } 336 337 void 338 cc_conn_init(struct tcpcb *tp) 339 { 340 struct hc_metrics_lite metrics; 341 struct inpcb *inp = tp->t_inpcb; 342 u_int maxseg; 343 int rtt; 344 345 INP_WLOCK_ASSERT(tp->t_inpcb); 346 347 tcp_hc_get(&inp->inp_inc, &metrics); 348 maxseg = tcp_maxseg(tp); 349 350 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 351 tp->t_srtt = rtt; 352 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 353 TCPSTAT_INC(tcps_usedrtt); 354 if (metrics.rmx_rttvar) { 355 tp->t_rttvar = metrics.rmx_rttvar; 356 TCPSTAT_INC(tcps_usedrttvar); 357 } else { 358 /* default variation is +- 1 rtt */ 359 tp->t_rttvar = 360 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 361 } 362 TCPT_RANGESET(tp->t_rxtcur, 363 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 364 tp->t_rttmin, TCPTV_REXMTMAX); 365 } 366 if (metrics.rmx_ssthresh) { 367 /* 368 * There's some sort of gateway or interface 369 * buffer limit on the path. Use this to set 370 * the slow start threshold, but set the 371 * threshold to no less than 2*mss. 372 */ 373 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 374 TCPSTAT_INC(tcps_usedssthresh); 375 } 376 377 /* 378 * Set the initial slow-start flight size. 379 * 380 * RFC5681 Section 3.1 specifies the default conservative values. 381 * RFC3390 specifies slightly more aggressive values. 382 * RFC6928 increases it to ten segments. 383 * Support for user specified value for initial flight size. 384 * 385 * If a SYN or SYN/ACK was lost and retransmitted, we have to 386 * reduce the initial CWND to one segment as congestion is likely 387 * requiring us to be cautious. 388 */ 389 if (tp->snd_cwnd == 1) 390 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 391 else if (V_tcp_initcwnd_segments) 392 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 393 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 394 else if (V_tcp_do_rfc3390) 395 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 396 else { 397 /* Per RFC5681 Section 3.1 */ 398 if (maxseg > 2190) 399 tp->snd_cwnd = 2 * maxseg; 400 else if (maxseg > 1095) 401 tp->snd_cwnd = 3 * maxseg; 402 else 403 tp->snd_cwnd = 4 * maxseg; 404 } 405 406 if (CC_ALGO(tp)->conn_init != NULL) 407 CC_ALGO(tp)->conn_init(tp->ccv); 408 } 409 410 void inline 411 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 412 { 413 u_int maxseg; 414 415 INP_WLOCK_ASSERT(tp->t_inpcb); 416 417 switch(type) { 418 case CC_NDUPACK: 419 if (!IN_FASTRECOVERY(tp->t_flags)) { 420 tp->snd_recover = tp->snd_max; 421 if (tp->t_flags & TF_ECN_PERMIT) 422 tp->t_flags |= TF_ECN_SND_CWR; 423 } 424 break; 425 case CC_ECN: 426 if (!IN_CONGRECOVERY(tp->t_flags)) { 427 TCPSTAT_INC(tcps_ecn_rcwnd); 428 tp->snd_recover = tp->snd_max; 429 if (tp->t_flags & TF_ECN_PERMIT) 430 tp->t_flags |= TF_ECN_SND_CWR; 431 } 432 break; 433 case CC_RTO: 434 maxseg = tcp_maxseg(tp); 435 tp->t_dupacks = 0; 436 tp->t_bytes_acked = 0; 437 EXIT_RECOVERY(tp->t_flags); 438 if (CC_ALGO(tp)->cong_signal == NULL) { 439 /* 440 * RFC5681 Section 3.1 441 * ssthresh = max (FlightSize / 2, 2*SMSS) eq (4) 442 */ 443 tp->snd_ssthresh = 444 max((tp->snd_max - tp->snd_una) / 2 / maxseg, 2) 445 * maxseg; 446 tp->snd_cwnd = maxseg; 447 } 448 break; 449 case CC_RTO_ERR: 450 TCPSTAT_INC(tcps_sndrexmitbad); 451 /* RTO was unnecessary, so reset everything. */ 452 tp->snd_cwnd = tp->snd_cwnd_prev; 453 tp->snd_ssthresh = tp->snd_ssthresh_prev; 454 tp->snd_recover = tp->snd_recover_prev; 455 if (tp->t_flags & TF_WASFRECOVERY) 456 ENTER_FASTRECOVERY(tp->t_flags); 457 if (tp->t_flags & TF_WASCRECOVERY) 458 ENTER_CONGRECOVERY(tp->t_flags); 459 tp->snd_nxt = tp->snd_max; 460 tp->t_flags &= ~TF_PREVVALID; 461 tp->t_badrxtwin = 0; 462 break; 463 } 464 465 if (CC_ALGO(tp)->cong_signal != NULL) { 466 if (th != NULL) 467 tp->ccv->curack = th->th_ack; 468 CC_ALGO(tp)->cong_signal(tp->ccv, type); 469 } 470 } 471 472 void inline 473 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 474 { 475 INP_WLOCK_ASSERT(tp->t_inpcb); 476 477 /* XXXLAS: KASSERT that we're in recovery? */ 478 479 if (CC_ALGO(tp)->post_recovery != NULL) { 480 tp->ccv->curack = th->th_ack; 481 CC_ALGO(tp)->post_recovery(tp->ccv); 482 } 483 /* XXXLAS: EXIT_RECOVERY ? */ 484 tp->t_bytes_acked = 0; 485 } 486 487 /* 488 * Indicate whether this ack should be delayed. We can delay the ack if 489 * following conditions are met: 490 * - There is no delayed ack timer in progress. 491 * - Our last ack wasn't a 0-sized window. We never want to delay 492 * the ack that opens up a 0-sized window. 493 * - LRO wasn't used for this segment. We make sure by checking that the 494 * segment size is not larger than the MSS. 495 */ 496 #define DELAY_ACK(tp, tlen) \ 497 ((!tcp_timer_active(tp, TT_DELACK) && \ 498 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 499 (tlen <= tp->t_maxseg) && \ 500 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 501 502 static void inline 503 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 504 { 505 INP_WLOCK_ASSERT(tp->t_inpcb); 506 507 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 508 switch (iptos & IPTOS_ECN_MASK) { 509 case IPTOS_ECN_CE: 510 tp->ccv->flags |= CCF_IPHDR_CE; 511 break; 512 case IPTOS_ECN_ECT0: 513 tp->ccv->flags &= ~CCF_IPHDR_CE; 514 break; 515 case IPTOS_ECN_ECT1: 516 tp->ccv->flags &= ~CCF_IPHDR_CE; 517 break; 518 } 519 520 if (th->th_flags & TH_CWR) 521 tp->ccv->flags |= CCF_TCPHDR_CWR; 522 else 523 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 524 525 if (tp->t_flags & TF_DELACK) 526 tp->ccv->flags |= CCF_DELACK; 527 else 528 tp->ccv->flags &= ~CCF_DELACK; 529 530 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 531 532 if (tp->ccv->flags & CCF_ACKNOW) 533 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 534 } 535 } 536 537 /* 538 * TCP input handling is split into multiple parts: 539 * tcp6_input is a thin wrapper around tcp_input for the extended 540 * ip6_protox[] call format in ip6_input 541 * tcp_input handles primary segment validation, inpcb lookup and 542 * SYN processing on listen sockets 543 * tcp_do_segment processes the ACK and text of the segment for 544 * establishing, established and closing connections 545 */ 546 #ifdef INET6 547 int 548 tcp6_input(struct mbuf **mp, int *offp, int proto) 549 { 550 struct mbuf *m = *mp; 551 struct in6_ifaddr *ia6; 552 struct ip6_hdr *ip6; 553 554 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 555 556 /* 557 * draft-itojun-ipv6-tcp-to-anycast 558 * better place to put this in? 559 */ 560 ip6 = mtod(m, struct ip6_hdr *); 561 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 562 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 563 struct ip6_hdr *ip6; 564 565 ifa_free(&ia6->ia_ifa); 566 ip6 = mtod(m, struct ip6_hdr *); 567 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 568 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 569 return (IPPROTO_DONE); 570 } 571 if (ia6) 572 ifa_free(&ia6->ia_ifa); 573 574 return (tcp_input(mp, offp, proto)); 575 } 576 #endif /* INET6 */ 577 578 int 579 tcp_input(struct mbuf **mp, int *offp, int proto) 580 { 581 struct mbuf *m = *mp; 582 struct tcphdr *th = NULL; 583 struct ip *ip = NULL; 584 struct inpcb *inp = NULL; 585 struct tcpcb *tp = NULL; 586 struct socket *so = NULL; 587 u_char *optp = NULL; 588 int off0; 589 int optlen = 0; 590 #ifdef INET 591 int len; 592 #endif 593 int tlen = 0, off; 594 int drop_hdrlen; 595 int thflags; 596 int rstreason = 0; /* For badport_bandlim accounting purposes */ 597 uint8_t iptos; 598 struct m_tag *fwd_tag = NULL; 599 #ifdef INET6 600 struct ip6_hdr *ip6 = NULL; 601 int isipv6; 602 #else 603 const void *ip6 = NULL; 604 #endif /* INET6 */ 605 struct tcpopt to; /* options in this segment */ 606 char *s = NULL; /* address and port logging */ 607 int ti_locked; 608 #ifdef TCPDEBUG 609 /* 610 * The size of tcp_saveipgen must be the size of the max ip header, 611 * now IPv6. 612 */ 613 u_char tcp_saveipgen[IP6_HDR_LEN]; 614 struct tcphdr tcp_savetcp; 615 short ostate = 0; 616 #endif 617 618 #ifdef INET6 619 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 620 #endif 621 622 off0 = *offp; 623 m = *mp; 624 *mp = NULL; 625 to.to_flags = 0; 626 TCPSTAT_INC(tcps_rcvtotal); 627 628 #ifdef INET6 629 if (isipv6) { 630 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 631 632 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 633 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 634 if (m == NULL) { 635 TCPSTAT_INC(tcps_rcvshort); 636 return (IPPROTO_DONE); 637 } 638 } 639 640 ip6 = mtod(m, struct ip6_hdr *); 641 th = (struct tcphdr *)((caddr_t)ip6 + off0); 642 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 643 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 644 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 645 th->th_sum = m->m_pkthdr.csum_data; 646 else 647 th->th_sum = in6_cksum_pseudo(ip6, tlen, 648 IPPROTO_TCP, m->m_pkthdr.csum_data); 649 th->th_sum ^= 0xffff; 650 } else 651 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 652 if (th->th_sum) { 653 TCPSTAT_INC(tcps_rcvbadsum); 654 goto drop; 655 } 656 657 /* 658 * Be proactive about unspecified IPv6 address in source. 659 * As we use all-zero to indicate unbounded/unconnected pcb, 660 * unspecified IPv6 address can be used to confuse us. 661 * 662 * Note that packets with unspecified IPv6 destination is 663 * already dropped in ip6_input. 664 */ 665 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 666 /* XXX stat */ 667 goto drop; 668 } 669 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 670 } 671 #endif 672 #if defined(INET) && defined(INET6) 673 else 674 #endif 675 #ifdef INET 676 { 677 /* 678 * Get IP and TCP header together in first mbuf. 679 * Note: IP leaves IP header in first mbuf. 680 */ 681 if (off0 > sizeof (struct ip)) { 682 ip_stripoptions(m); 683 off0 = sizeof(struct ip); 684 } 685 if (m->m_len < sizeof (struct tcpiphdr)) { 686 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 687 == NULL) { 688 TCPSTAT_INC(tcps_rcvshort); 689 return (IPPROTO_DONE); 690 } 691 } 692 ip = mtod(m, struct ip *); 693 th = (struct tcphdr *)((caddr_t)ip + off0); 694 tlen = ntohs(ip->ip_len) - off0; 695 696 iptos = ip->ip_tos; 697 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 698 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 699 th->th_sum = m->m_pkthdr.csum_data; 700 else 701 th->th_sum = in_pseudo(ip->ip_src.s_addr, 702 ip->ip_dst.s_addr, 703 htonl(m->m_pkthdr.csum_data + tlen + 704 IPPROTO_TCP)); 705 th->th_sum ^= 0xffff; 706 } else { 707 struct ipovly *ipov = (struct ipovly *)ip; 708 709 /* 710 * Checksum extended TCP header and data. 711 */ 712 len = off0 + tlen; 713 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 714 ipov->ih_len = htons(tlen); 715 th->th_sum = in_cksum(m, len); 716 /* Reset length for SDT probes. */ 717 ip->ip_len = htons(len); 718 /* Reset TOS bits */ 719 ip->ip_tos = iptos; 720 /* Re-initialization for later version check */ 721 ip->ip_v = IPVERSION; 722 } 723 724 if (th->th_sum) { 725 TCPSTAT_INC(tcps_rcvbadsum); 726 goto drop; 727 } 728 } 729 #endif /* INET */ 730 731 /* 732 * Check that TCP offset makes sense, 733 * pull out TCP options and adjust length. XXX 734 */ 735 off = th->th_off << 2; 736 if (off < sizeof (struct tcphdr) || off > tlen) { 737 TCPSTAT_INC(tcps_rcvbadoff); 738 goto drop; 739 } 740 tlen -= off; /* tlen is used instead of ti->ti_len */ 741 if (off > sizeof (struct tcphdr)) { 742 #ifdef INET6 743 if (isipv6) { 744 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 745 ip6 = mtod(m, struct ip6_hdr *); 746 th = (struct tcphdr *)((caddr_t)ip6 + off0); 747 } 748 #endif 749 #if defined(INET) && defined(INET6) 750 else 751 #endif 752 #ifdef INET 753 { 754 if (m->m_len < sizeof(struct ip) + off) { 755 if ((m = m_pullup(m, sizeof (struct ip) + off)) 756 == NULL) { 757 TCPSTAT_INC(tcps_rcvshort); 758 return (IPPROTO_DONE); 759 } 760 ip = mtod(m, struct ip *); 761 th = (struct tcphdr *)((caddr_t)ip + off0); 762 } 763 } 764 #endif 765 optlen = off - sizeof (struct tcphdr); 766 optp = (u_char *)(th + 1); 767 } 768 thflags = th->th_flags; 769 770 /* 771 * Convert TCP protocol specific fields to host format. 772 */ 773 tcp_fields_to_host(th); 774 775 /* 776 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 777 */ 778 drop_hdrlen = off0 + off; 779 780 /* 781 * Locate pcb for segment; if we're likely to add or remove a 782 * connection then first acquire pcbinfo lock. There are three cases 783 * where we might discover later we need a write lock despite the 784 * flags: ACKs moving a connection out of the syncache, ACKs for a 785 * connection in TIMEWAIT and SYNs not targeting a listening socket. 786 */ 787 if ((thflags & (TH_FIN | TH_RST)) != 0) { 788 INP_INFO_RLOCK(&V_tcbinfo); 789 ti_locked = TI_RLOCKED; 790 } else 791 ti_locked = TI_UNLOCKED; 792 793 /* 794 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 795 */ 796 if ( 797 #ifdef INET6 798 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 799 #ifdef INET 800 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 801 #endif 802 #endif 803 #if defined(INET) && !defined(INET6) 804 (m->m_flags & M_IP_NEXTHOP) 805 #endif 806 ) 807 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 808 809 findpcb: 810 #ifdef INVARIANTS 811 if (ti_locked == TI_RLOCKED) { 812 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 813 } else { 814 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 815 } 816 #endif 817 #ifdef INET6 818 if (isipv6 && fwd_tag != NULL) { 819 struct sockaddr_in6 *next_hop6; 820 821 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 822 /* 823 * Transparently forwarded. Pretend to be the destination. 824 * Already got one like this? 825 */ 826 inp = in6_pcblookup_mbuf(&V_tcbinfo, 827 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 828 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 829 if (!inp) { 830 /* 831 * It's new. Try to find the ambushing socket. 832 * Because we've rewritten the destination address, 833 * any hardware-generated hash is ignored. 834 */ 835 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 836 th->th_sport, &next_hop6->sin6_addr, 837 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 838 th->th_dport, INPLOOKUP_WILDCARD | 839 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 840 } 841 } else if (isipv6) { 842 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 843 th->th_sport, &ip6->ip6_dst, th->th_dport, 844 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 845 m->m_pkthdr.rcvif, m); 846 } 847 #endif /* INET6 */ 848 #if defined(INET6) && defined(INET) 849 else 850 #endif 851 #ifdef INET 852 if (fwd_tag != NULL) { 853 struct sockaddr_in *next_hop; 854 855 next_hop = (struct sockaddr_in *)(fwd_tag+1); 856 /* 857 * Transparently forwarded. Pretend to be the destination. 858 * already got one like this? 859 */ 860 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 861 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 862 m->m_pkthdr.rcvif, m); 863 if (!inp) { 864 /* 865 * It's new. Try to find the ambushing socket. 866 * Because we've rewritten the destination address, 867 * any hardware-generated hash is ignored. 868 */ 869 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 870 th->th_sport, next_hop->sin_addr, 871 next_hop->sin_port ? ntohs(next_hop->sin_port) : 872 th->th_dport, INPLOOKUP_WILDCARD | 873 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 874 } 875 } else 876 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 877 th->th_sport, ip->ip_dst, th->th_dport, 878 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 879 m->m_pkthdr.rcvif, m); 880 #endif /* INET */ 881 882 /* 883 * If the INPCB does not exist then all data in the incoming 884 * segment is discarded and an appropriate RST is sent back. 885 * XXX MRT Send RST using which routing table? 886 */ 887 if (inp == NULL) { 888 /* 889 * Log communication attempts to ports that are not 890 * in use. 891 */ 892 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 893 tcp_log_in_vain == 2) { 894 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 895 log(LOG_INFO, "%s; %s: Connection attempt " 896 "to closed port\n", s, __func__); 897 } 898 /* 899 * When blackholing do not respond with a RST but 900 * completely ignore the segment and drop it. 901 */ 902 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 903 V_blackhole == 2) 904 goto dropunlock; 905 906 rstreason = BANDLIM_RST_CLOSEDPORT; 907 goto dropwithreset; 908 } 909 INP_WLOCK_ASSERT(inp); 910 /* 911 * While waiting for inp lock during the lookup, another thread 912 * can have dropped the inpcb, in which case we need to loop back 913 * and try to find a new inpcb to deliver to. 914 */ 915 if (inp->inp_flags & INP_DROPPED) { 916 INP_WUNLOCK(inp); 917 inp = NULL; 918 goto findpcb; 919 } 920 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 921 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 922 ((inp->inp_socket == NULL) || 923 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 924 inp->inp_flowid = m->m_pkthdr.flowid; 925 inp->inp_flowtype = M_HASHTYPE_GET(m); 926 } 927 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 928 #ifdef INET6 929 if (isipv6 && IPSEC_ENABLED(ipv6) && 930 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 931 goto dropunlock; 932 } 933 #ifdef INET 934 else 935 #endif 936 #endif /* INET6 */ 937 #ifdef INET 938 if (IPSEC_ENABLED(ipv4) && 939 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 940 goto dropunlock; 941 } 942 #endif /* INET */ 943 #endif /* IPSEC */ 944 945 /* 946 * Check the minimum TTL for socket. 947 */ 948 if (inp->inp_ip_minttl != 0) { 949 #ifdef INET6 950 if (isipv6) { 951 if (inp->inp_ip_minttl > ip6->ip6_hlim) 952 goto dropunlock; 953 } else 954 #endif 955 if (inp->inp_ip_minttl > ip->ip_ttl) 956 goto dropunlock; 957 } 958 959 /* 960 * A previous connection in TIMEWAIT state is supposed to catch stray 961 * or duplicate segments arriving late. If this segment was a 962 * legitimate new connection attempt, the old INPCB gets removed and 963 * we can try again to find a listening socket. 964 * 965 * At this point, due to earlier optimism, we may hold only an inpcb 966 * lock, and not the inpcbinfo write lock. If so, we need to try to 967 * acquire it, or if that fails, acquire a reference on the inpcb, 968 * drop all locks, acquire a global write lock, and then re-acquire 969 * the inpcb lock. We may at that point discover that another thread 970 * has tried to free the inpcb, in which case we need to loop back 971 * and try to find a new inpcb to deliver to. 972 * 973 * XXXRW: It may be time to rethink timewait locking. 974 */ 975 relocked: 976 if (inp->inp_flags & INP_TIMEWAIT) { 977 if (ti_locked == TI_UNLOCKED) { 978 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 979 in_pcbref(inp); 980 INP_WUNLOCK(inp); 981 INP_INFO_RLOCK(&V_tcbinfo); 982 ti_locked = TI_RLOCKED; 983 INP_WLOCK(inp); 984 if (in_pcbrele_wlocked(inp)) { 985 inp = NULL; 986 goto findpcb; 987 } else if (inp->inp_flags & INP_DROPPED) { 988 INP_WUNLOCK(inp); 989 inp = NULL; 990 goto findpcb; 991 } 992 } else 993 ti_locked = TI_RLOCKED; 994 } 995 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 996 997 if (thflags & TH_SYN) 998 tcp_dooptions(&to, optp, optlen, TO_SYN); 999 /* 1000 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1001 */ 1002 if (tcp_twcheck(inp, &to, th, m, tlen)) 1003 goto findpcb; 1004 INP_INFO_RUNLOCK(&V_tcbinfo); 1005 return (IPPROTO_DONE); 1006 } 1007 /* 1008 * The TCPCB may no longer exist if the connection is winding 1009 * down or it is in the CLOSED state. Either way we drop the 1010 * segment and send an appropriate response. 1011 */ 1012 tp = intotcpcb(inp); 1013 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1014 rstreason = BANDLIM_RST_CLOSEDPORT; 1015 goto dropwithreset; 1016 } 1017 1018 #ifdef TCP_OFFLOAD 1019 if (tp->t_flags & TF_TOE) { 1020 tcp_offload_input(tp, m); 1021 m = NULL; /* consumed by the TOE driver */ 1022 goto dropunlock; 1023 } 1024 #endif 1025 1026 /* 1027 * We've identified a valid inpcb, but it could be that we need an 1028 * inpcbinfo write lock but don't hold it. In this case, attempt to 1029 * acquire using the same strategy as the TIMEWAIT case above. If we 1030 * relock, we have to jump back to 'relocked' as the connection might 1031 * now be in TIMEWAIT. 1032 */ 1033 #ifdef INVARIANTS 1034 if ((thflags & (TH_FIN | TH_RST)) != 0) 1035 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1036 #endif 1037 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1038 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1039 !IS_FASTOPEN(tp->t_flags)))) { 1040 if (ti_locked == TI_UNLOCKED) { 1041 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1042 in_pcbref(inp); 1043 INP_WUNLOCK(inp); 1044 INP_INFO_RLOCK(&V_tcbinfo); 1045 ti_locked = TI_RLOCKED; 1046 INP_WLOCK(inp); 1047 if (in_pcbrele_wlocked(inp)) { 1048 inp = NULL; 1049 goto findpcb; 1050 } else if (inp->inp_flags & INP_DROPPED) { 1051 INP_WUNLOCK(inp); 1052 inp = NULL; 1053 goto findpcb; 1054 } 1055 goto relocked; 1056 } else 1057 ti_locked = TI_RLOCKED; 1058 } 1059 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1060 } 1061 1062 #ifdef MAC 1063 INP_WLOCK_ASSERT(inp); 1064 if (mac_inpcb_check_deliver(inp, m)) 1065 goto dropunlock; 1066 #endif 1067 so = inp->inp_socket; 1068 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1069 #ifdef TCPDEBUG 1070 if (so->so_options & SO_DEBUG) { 1071 ostate = tp->t_state; 1072 #ifdef INET6 1073 if (isipv6) { 1074 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1075 } else 1076 #endif 1077 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1078 tcp_savetcp = *th; 1079 } 1080 #endif /* TCPDEBUG */ 1081 /* 1082 * When the socket is accepting connections (the INPCB is in LISTEN 1083 * state) we look into the SYN cache if this is a new connection 1084 * attempt or the completion of a previous one. 1085 */ 1086 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1087 ("%s: so accepting but tp %p not listening", __func__, tp)); 1088 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1089 struct in_conninfo inc; 1090 1091 bzero(&inc, sizeof(inc)); 1092 #ifdef INET6 1093 if (isipv6) { 1094 inc.inc_flags |= INC_ISIPV6; 1095 inc.inc6_faddr = ip6->ip6_src; 1096 inc.inc6_laddr = ip6->ip6_dst; 1097 } else 1098 #endif 1099 { 1100 inc.inc_faddr = ip->ip_src; 1101 inc.inc_laddr = ip->ip_dst; 1102 } 1103 inc.inc_fport = th->th_sport; 1104 inc.inc_lport = th->th_dport; 1105 inc.inc_fibnum = so->so_fibnum; 1106 1107 /* 1108 * Check for an existing connection attempt in syncache if 1109 * the flag is only ACK. A successful lookup creates a new 1110 * socket appended to the listen queue in SYN_RECEIVED state. 1111 */ 1112 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1113 1114 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1115 /* 1116 * Parse the TCP options here because 1117 * syncookies need access to the reflected 1118 * timestamp. 1119 */ 1120 tcp_dooptions(&to, optp, optlen, 0); 1121 /* 1122 * NB: syncache_expand() doesn't unlock 1123 * inp and tcpinfo locks. 1124 */ 1125 rstreason = syncache_expand(&inc, &to, th, &so, m); 1126 if (rstreason < 0) { 1127 /* 1128 * A failing TCP MD5 signature comparison 1129 * must result in the segment being dropped 1130 * and must not produce any response back 1131 * to the sender. 1132 */ 1133 goto dropunlock; 1134 } else if (rstreason == 0) { 1135 /* 1136 * No syncache entry or ACK was not 1137 * for our SYN/ACK. Send a RST. 1138 * NB: syncache did its own logging 1139 * of the failure cause. 1140 */ 1141 rstreason = BANDLIM_RST_OPENPORT; 1142 goto dropwithreset; 1143 } 1144 #ifdef TCP_RFC7413 1145 tfo_socket_result: 1146 #endif 1147 if (so == NULL) { 1148 /* 1149 * We completed the 3-way handshake 1150 * but could not allocate a socket 1151 * either due to memory shortage, 1152 * listen queue length limits or 1153 * global socket limits. Send RST 1154 * or wait and have the remote end 1155 * retransmit the ACK for another 1156 * try. 1157 */ 1158 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1159 log(LOG_DEBUG, "%s; %s: Listen socket: " 1160 "Socket allocation failed due to " 1161 "limits or memory shortage, %s\n", 1162 s, __func__, 1163 V_tcp_sc_rst_sock_fail ? 1164 "sending RST" : "try again"); 1165 if (V_tcp_sc_rst_sock_fail) { 1166 rstreason = BANDLIM_UNLIMITED; 1167 goto dropwithreset; 1168 } else 1169 goto dropunlock; 1170 } 1171 /* 1172 * Socket is created in state SYN_RECEIVED. 1173 * Unlock the listen socket, lock the newly 1174 * created socket and update the tp variable. 1175 */ 1176 INP_WUNLOCK(inp); /* listen socket */ 1177 inp = sotoinpcb(so); 1178 /* 1179 * New connection inpcb is already locked by 1180 * syncache_expand(). 1181 */ 1182 INP_WLOCK_ASSERT(inp); 1183 tp = intotcpcb(inp); 1184 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1185 ("%s: ", __func__)); 1186 /* 1187 * Process the segment and the data it 1188 * contains. tcp_do_segment() consumes 1189 * the mbuf chain and unlocks the inpcb. 1190 */ 1191 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1192 iptos, ti_locked); 1193 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1194 return (IPPROTO_DONE); 1195 } 1196 /* 1197 * Segment flag validation for new connection attempts: 1198 * 1199 * Our (SYN|ACK) response was rejected. 1200 * Check with syncache and remove entry to prevent 1201 * retransmits. 1202 * 1203 * NB: syncache_chkrst does its own logging of failure 1204 * causes. 1205 */ 1206 if (thflags & TH_RST) { 1207 syncache_chkrst(&inc, th); 1208 goto dropunlock; 1209 } 1210 /* 1211 * We can't do anything without SYN. 1212 */ 1213 if ((thflags & TH_SYN) == 0) { 1214 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1215 log(LOG_DEBUG, "%s; %s: Listen socket: " 1216 "SYN is missing, segment ignored\n", 1217 s, __func__); 1218 TCPSTAT_INC(tcps_badsyn); 1219 goto dropunlock; 1220 } 1221 /* 1222 * (SYN|ACK) is bogus on a listen socket. 1223 */ 1224 if (thflags & TH_ACK) { 1225 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1226 log(LOG_DEBUG, "%s; %s: Listen socket: " 1227 "SYN|ACK invalid, segment rejected\n", 1228 s, __func__); 1229 syncache_badack(&inc); /* XXX: Not needed! */ 1230 TCPSTAT_INC(tcps_badsyn); 1231 rstreason = BANDLIM_RST_OPENPORT; 1232 goto dropwithreset; 1233 } 1234 /* 1235 * If the drop_synfin option is enabled, drop all 1236 * segments with both the SYN and FIN bits set. 1237 * This prevents e.g. nmap from identifying the 1238 * TCP/IP stack. 1239 * XXX: Poor reasoning. nmap has other methods 1240 * and is constantly refining its stack detection 1241 * strategies. 1242 * XXX: This is a violation of the TCP specification 1243 * and was used by RFC1644. 1244 */ 1245 if ((thflags & TH_FIN) && V_drop_synfin) { 1246 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1247 log(LOG_DEBUG, "%s; %s: Listen socket: " 1248 "SYN|FIN segment ignored (based on " 1249 "sysctl setting)\n", s, __func__); 1250 TCPSTAT_INC(tcps_badsyn); 1251 goto dropunlock; 1252 } 1253 /* 1254 * Segment's flags are (SYN) or (SYN|FIN). 1255 * 1256 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1257 * as they do not affect the state of the TCP FSM. 1258 * The data pointed to by TH_URG and th_urp is ignored. 1259 */ 1260 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1261 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1262 KASSERT(thflags & (TH_SYN), 1263 ("%s: Listen socket: TH_SYN not set", __func__)); 1264 #ifdef INET6 1265 /* 1266 * If deprecated address is forbidden, 1267 * we do not accept SYN to deprecated interface 1268 * address to prevent any new inbound connection from 1269 * getting established. 1270 * When we do not accept SYN, we send a TCP RST, 1271 * with deprecated source address (instead of dropping 1272 * it). We compromise it as it is much better for peer 1273 * to send a RST, and RST will be the final packet 1274 * for the exchange. 1275 * 1276 * If we do not forbid deprecated addresses, we accept 1277 * the SYN packet. RFC2462 does not suggest dropping 1278 * SYN in this case. 1279 * If we decipher RFC2462 5.5.4, it says like this: 1280 * 1. use of deprecated addr with existing 1281 * communication is okay - "SHOULD continue to be 1282 * used" 1283 * 2. use of it with new communication: 1284 * (2a) "SHOULD NOT be used if alternate address 1285 * with sufficient scope is available" 1286 * (2b) nothing mentioned otherwise. 1287 * Here we fall into (2b) case as we have no choice in 1288 * our source address selection - we must obey the peer. 1289 * 1290 * The wording in RFC2462 is confusing, and there are 1291 * multiple description text for deprecated address 1292 * handling - worse, they are not exactly the same. 1293 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1294 */ 1295 if (isipv6 && !V_ip6_use_deprecated) { 1296 struct in6_ifaddr *ia6; 1297 1298 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1299 if (ia6 != NULL && 1300 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1301 ifa_free(&ia6->ia_ifa); 1302 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1303 log(LOG_DEBUG, "%s; %s: Listen socket: " 1304 "Connection attempt to deprecated " 1305 "IPv6 address rejected\n", 1306 s, __func__); 1307 rstreason = BANDLIM_RST_OPENPORT; 1308 goto dropwithreset; 1309 } 1310 if (ia6) 1311 ifa_free(&ia6->ia_ifa); 1312 } 1313 #endif /* INET6 */ 1314 /* 1315 * Basic sanity checks on incoming SYN requests: 1316 * Don't respond if the destination is a link layer 1317 * broadcast according to RFC1122 4.2.3.10, p. 104. 1318 * If it is from this socket it must be forged. 1319 * Don't respond if the source or destination is a 1320 * global or subnet broad- or multicast address. 1321 * Note that it is quite possible to receive unicast 1322 * link-layer packets with a broadcast IP address. Use 1323 * in_broadcast() to find them. 1324 */ 1325 if (m->m_flags & (M_BCAST|M_MCAST)) { 1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1327 log(LOG_DEBUG, "%s; %s: Listen socket: " 1328 "Connection attempt from broad- or multicast " 1329 "link layer address ignored\n", s, __func__); 1330 goto dropunlock; 1331 } 1332 #ifdef INET6 1333 if (isipv6) { 1334 if (th->th_dport == th->th_sport && 1335 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1336 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1337 log(LOG_DEBUG, "%s; %s: Listen socket: " 1338 "Connection attempt to/from self " 1339 "ignored\n", s, __func__); 1340 goto dropunlock; 1341 } 1342 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1343 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1344 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1345 log(LOG_DEBUG, "%s; %s: Listen socket: " 1346 "Connection attempt from/to multicast " 1347 "address ignored\n", s, __func__); 1348 goto dropunlock; 1349 } 1350 } 1351 #endif 1352 #if defined(INET) && defined(INET6) 1353 else 1354 #endif 1355 #ifdef INET 1356 { 1357 if (th->th_dport == th->th_sport && 1358 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1359 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1360 log(LOG_DEBUG, "%s; %s: Listen socket: " 1361 "Connection attempt from/to self " 1362 "ignored\n", s, __func__); 1363 goto dropunlock; 1364 } 1365 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1366 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1367 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1368 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1369 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1370 log(LOG_DEBUG, "%s; %s: Listen socket: " 1371 "Connection attempt from/to broad- " 1372 "or multicast address ignored\n", 1373 s, __func__); 1374 goto dropunlock; 1375 } 1376 } 1377 #endif 1378 /* 1379 * SYN appears to be valid. Create compressed TCP state 1380 * for syncache. 1381 */ 1382 #ifdef TCPDEBUG 1383 if (so->so_options & SO_DEBUG) 1384 tcp_trace(TA_INPUT, ostate, tp, 1385 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1386 #endif 1387 TCP_PROBE3(debug__input, tp, th, m); 1388 tcp_dooptions(&to, optp, optlen, TO_SYN); 1389 #ifdef TCP_RFC7413 1390 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1391 goto tfo_socket_result; 1392 #else 1393 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1394 #endif 1395 /* 1396 * Entry added to syncache and mbuf consumed. 1397 * Only the listen socket is unlocked by syncache_add(). 1398 */ 1399 if (ti_locked == TI_RLOCKED) { 1400 INP_INFO_RUNLOCK(&V_tcbinfo); 1401 ti_locked = TI_UNLOCKED; 1402 } 1403 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1404 return (IPPROTO_DONE); 1405 } else if (tp->t_state == TCPS_LISTEN) { 1406 /* 1407 * When a listen socket is torn down the SO_ACCEPTCONN 1408 * flag is removed first while connections are drained 1409 * from the accept queue in a unlock/lock cycle of the 1410 * ACCEPT_LOCK, opening a race condition allowing a SYN 1411 * attempt go through unhandled. 1412 */ 1413 goto dropunlock; 1414 } 1415 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1416 if (tp->t_flags & TF_SIGNATURE) { 1417 tcp_dooptions(&to, optp, optlen, thflags); 1418 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1419 TCPSTAT_INC(tcps_sig_err_nosigopt); 1420 goto dropunlock; 1421 } 1422 if (!TCPMD5_ENABLED() || 1423 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1424 goto dropunlock; 1425 } 1426 #endif 1427 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1428 1429 /* 1430 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1431 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1432 * the inpcb, and unlocks pcbinfo. 1433 */ 1434 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1435 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1436 return (IPPROTO_DONE); 1437 1438 dropwithreset: 1439 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1440 1441 if (ti_locked == TI_RLOCKED) { 1442 INP_INFO_RUNLOCK(&V_tcbinfo); 1443 ti_locked = TI_UNLOCKED; 1444 } 1445 #ifdef INVARIANTS 1446 else { 1447 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1448 "ti_locked: %d", __func__, ti_locked)); 1449 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1450 } 1451 #endif 1452 1453 if (inp != NULL) { 1454 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1455 INP_WUNLOCK(inp); 1456 } else 1457 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1458 m = NULL; /* mbuf chain got consumed. */ 1459 goto drop; 1460 1461 dropunlock: 1462 if (m != NULL) 1463 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1464 1465 if (ti_locked == TI_RLOCKED) { 1466 INP_INFO_RUNLOCK(&V_tcbinfo); 1467 ti_locked = TI_UNLOCKED; 1468 } 1469 #ifdef INVARIANTS 1470 else { 1471 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1472 "ti_locked: %d", __func__, ti_locked)); 1473 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1474 } 1475 #endif 1476 1477 if (inp != NULL) 1478 INP_WUNLOCK(inp); 1479 1480 drop: 1481 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1482 if (s != NULL) 1483 free(s, M_TCPLOG); 1484 if (m != NULL) 1485 m_freem(m); 1486 return (IPPROTO_DONE); 1487 } 1488 1489 /* 1490 * Automatic sizing of receive socket buffer. Often the send 1491 * buffer size is not optimally adjusted to the actual network 1492 * conditions at hand (delay bandwidth product). Setting the 1493 * buffer size too small limits throughput on links with high 1494 * bandwidth and high delay (eg. trans-continental/oceanic links). 1495 * 1496 * On the receive side the socket buffer memory is only rarely 1497 * used to any significant extent. This allows us to be much 1498 * more aggressive in scaling the receive socket buffer. For 1499 * the case that the buffer space is actually used to a large 1500 * extent and we run out of kernel memory we can simply drop 1501 * the new segments; TCP on the sender will just retransmit it 1502 * later. Setting the buffer size too big may only consume too 1503 * much kernel memory if the application doesn't read() from 1504 * the socket or packet loss or reordering makes use of the 1505 * reassembly queue. 1506 * 1507 * The criteria to step up the receive buffer one notch are: 1508 * 1. Application has not set receive buffer size with 1509 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1510 * 2. the number of bytes received during the time it takes 1511 * one timestamp to be reflected back to us (the RTT); 1512 * 3. received bytes per RTT is within seven eighth of the 1513 * current socket buffer size; 1514 * 4. receive buffer size has not hit maximal automatic size; 1515 * 1516 * This algorithm does one step per RTT at most and only if 1517 * we receive a bulk stream w/o packet losses or reorderings. 1518 * Shrinking the buffer during idle times is not necessary as 1519 * it doesn't consume any memory when idle. 1520 * 1521 * TODO: Only step up if the application is actually serving 1522 * the buffer to better manage the socket buffer resources. 1523 */ 1524 int 1525 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1526 struct tcpcb *tp, int tlen) 1527 { 1528 int newsize = 0; 1529 1530 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1531 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1532 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1533 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1534 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1535 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1536 newsize = min(so->so_rcv.sb_hiwat + 1537 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1538 } 1539 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1540 1541 /* Start over with next RTT. */ 1542 tp->rfbuf_ts = 0; 1543 tp->rfbuf_cnt = 0; 1544 } else { 1545 tp->rfbuf_cnt += tlen; /* add up */ 1546 } 1547 1548 return (newsize); 1549 } 1550 1551 void 1552 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1553 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1554 int ti_locked) 1555 { 1556 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1557 int rstreason, todrop, win; 1558 uint32_t tiwin; 1559 uint16_t nsegs; 1560 char *s; 1561 struct in_conninfo *inc; 1562 struct mbuf *mfree; 1563 struct tcpopt to; 1564 #ifdef TCP_RFC7413 1565 int tfo_syn; 1566 #endif 1567 1568 #ifdef TCPDEBUG 1569 /* 1570 * The size of tcp_saveipgen must be the size of the max ip header, 1571 * now IPv6. 1572 */ 1573 u_char tcp_saveipgen[IP6_HDR_LEN]; 1574 struct tcphdr tcp_savetcp; 1575 short ostate = 0; 1576 #endif 1577 thflags = th->th_flags; 1578 inc = &tp->t_inpcb->inp_inc; 1579 tp->sackhint.last_sack_ack = 0; 1580 sack_changed = 0; 1581 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1582 1583 /* 1584 * If this is either a state-changing packet or current state isn't 1585 * established, we require a write lock on tcbinfo. Otherwise, we 1586 * allow the tcbinfo to be in either alocked or unlocked, as the 1587 * caller may have unnecessarily acquired a write lock due to a race. 1588 */ 1589 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1590 tp->t_state != TCPS_ESTABLISHED) { 1591 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1592 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1593 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1594 } else { 1595 #ifdef INVARIANTS 1596 if (ti_locked == TI_RLOCKED) 1597 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1598 else { 1599 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1600 "ti_locked: %d", __func__, ti_locked)); 1601 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1602 } 1603 #endif 1604 } 1605 INP_WLOCK_ASSERT(tp->t_inpcb); 1606 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1607 __func__)); 1608 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1609 __func__)); 1610 1611 #ifdef TCPPCAP 1612 /* Save segment, if requested. */ 1613 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1614 #endif 1615 1616 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1617 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1618 log(LOG_DEBUG, "%s; %s: " 1619 "SYN|FIN segment ignored (based on " 1620 "sysctl setting)\n", s, __func__); 1621 free(s, M_TCPLOG); 1622 } 1623 goto drop; 1624 } 1625 1626 /* 1627 * Segment received on connection. 1628 * Reset idle time and keep-alive timer. 1629 * XXX: This should be done after segment 1630 * validation to ignore broken/spoofed segs. 1631 */ 1632 tp->t_rcvtime = ticks; 1633 1634 /* 1635 * Scale up the window into a 32-bit value. 1636 * For the SYN_SENT state the scale is zero. 1637 */ 1638 tiwin = th->th_win << tp->snd_scale; 1639 1640 /* 1641 * TCP ECN processing. 1642 */ 1643 if (tp->t_flags & TF_ECN_PERMIT) { 1644 if (thflags & TH_CWR) 1645 tp->t_flags &= ~TF_ECN_SND_ECE; 1646 switch (iptos & IPTOS_ECN_MASK) { 1647 case IPTOS_ECN_CE: 1648 tp->t_flags |= TF_ECN_SND_ECE; 1649 TCPSTAT_INC(tcps_ecn_ce); 1650 break; 1651 case IPTOS_ECN_ECT0: 1652 TCPSTAT_INC(tcps_ecn_ect0); 1653 break; 1654 case IPTOS_ECN_ECT1: 1655 TCPSTAT_INC(tcps_ecn_ect1); 1656 break; 1657 } 1658 1659 /* Process a packet differently from RFC3168. */ 1660 cc_ecnpkt_handler(tp, th, iptos); 1661 1662 /* Congestion experienced. */ 1663 if (thflags & TH_ECE) { 1664 cc_cong_signal(tp, th, CC_ECN); 1665 } 1666 } 1667 1668 /* 1669 * Parse options on any incoming segment. 1670 */ 1671 tcp_dooptions(&to, (u_char *)(th + 1), 1672 (th->th_off << 2) - sizeof(struct tcphdr), 1673 (thflags & TH_SYN) ? TO_SYN : 0); 1674 1675 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1676 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1677 (to.to_flags & TOF_SIGNATURE) == 0) { 1678 TCPSTAT_INC(tcps_sig_err_sigopt); 1679 /* XXX: should drop? */ 1680 } 1681 #endif 1682 /* 1683 * If echoed timestamp is later than the current time, 1684 * fall back to non RFC1323 RTT calculation. Normalize 1685 * timestamp if syncookies were used when this connection 1686 * was established. 1687 */ 1688 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1689 to.to_tsecr -= tp->ts_offset; 1690 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1691 to.to_tsecr = 0; 1692 } 1693 /* 1694 * If timestamps were negotiated during SYN/ACK they should 1695 * appear on every segment during this session and vice versa. 1696 */ 1697 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1698 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1699 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1700 "no action\n", s, __func__); 1701 free(s, M_TCPLOG); 1702 } 1703 } 1704 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1705 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1706 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1707 "no action\n", s, __func__); 1708 free(s, M_TCPLOG); 1709 } 1710 } 1711 1712 /* 1713 * Process options only when we get SYN/ACK back. The SYN case 1714 * for incoming connections is handled in tcp_syncache. 1715 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1716 * or <SYN,ACK>) segment itself is never scaled. 1717 * XXX this is traditional behavior, may need to be cleaned up. 1718 */ 1719 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1720 if ((to.to_flags & TOF_SCALE) && 1721 (tp->t_flags & TF_REQ_SCALE)) { 1722 tp->t_flags |= TF_RCVD_SCALE; 1723 tp->snd_scale = to.to_wscale; 1724 } 1725 /* 1726 * Initial send window. It will be updated with 1727 * the next incoming segment to the scaled value. 1728 */ 1729 tp->snd_wnd = th->th_win; 1730 if (to.to_flags & TOF_TS) { 1731 tp->t_flags |= TF_RCVD_TSTMP; 1732 tp->ts_recent = to.to_tsval; 1733 tp->ts_recent_age = tcp_ts_getticks(); 1734 } 1735 if (to.to_flags & TOF_MSS) 1736 tcp_mss(tp, to.to_mss); 1737 if ((tp->t_flags & TF_SACK_PERMIT) && 1738 (to.to_flags & TOF_SACKPERM) == 0) 1739 tp->t_flags &= ~TF_SACK_PERMIT; 1740 } 1741 1742 /* 1743 * Header prediction: check for the two common cases 1744 * of a uni-directional data xfer. If the packet has 1745 * no control flags, is in-sequence, the window didn't 1746 * change and we're not retransmitting, it's a 1747 * candidate. If the length is zero and the ack moved 1748 * forward, we're the sender side of the xfer. Just 1749 * free the data acked & wake any higher level process 1750 * that was blocked waiting for space. If the length 1751 * is non-zero and the ack didn't move, we're the 1752 * receiver side. If we're getting packets in-order 1753 * (the reassembly queue is empty), add the data to 1754 * the socket buffer and note that we need a delayed ack. 1755 * Make sure that the hidden state-flags are also off. 1756 * Since we check for TCPS_ESTABLISHED first, it can only 1757 * be TH_NEEDSYN. 1758 */ 1759 if (tp->t_state == TCPS_ESTABLISHED && 1760 th->th_seq == tp->rcv_nxt && 1761 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1762 tp->snd_nxt == tp->snd_max && 1763 tiwin && tiwin == tp->snd_wnd && 1764 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1765 LIST_EMPTY(&tp->t_segq) && 1766 ((to.to_flags & TOF_TS) == 0 || 1767 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1768 1769 /* 1770 * If last ACK falls within this segment's sequence numbers, 1771 * record the timestamp. 1772 * NOTE that the test is modified according to the latest 1773 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1774 */ 1775 if ((to.to_flags & TOF_TS) != 0 && 1776 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1777 tp->ts_recent_age = tcp_ts_getticks(); 1778 tp->ts_recent = to.to_tsval; 1779 } 1780 1781 if (tlen == 0) { 1782 if (SEQ_GT(th->th_ack, tp->snd_una) && 1783 SEQ_LEQ(th->th_ack, tp->snd_max) && 1784 !IN_RECOVERY(tp->t_flags) && 1785 (to.to_flags & TOF_SACK) == 0 && 1786 TAILQ_EMPTY(&tp->snd_holes)) { 1787 /* 1788 * This is a pure ack for outstanding data. 1789 */ 1790 if (ti_locked == TI_RLOCKED) 1791 INP_INFO_RUNLOCK(&V_tcbinfo); 1792 ti_locked = TI_UNLOCKED; 1793 1794 TCPSTAT_INC(tcps_predack); 1795 1796 /* 1797 * "bad retransmit" recovery. 1798 */ 1799 if (tp->t_rxtshift == 1 && 1800 tp->t_flags & TF_PREVVALID && 1801 (int)(ticks - tp->t_badrxtwin) < 0) { 1802 cc_cong_signal(tp, th, CC_RTO_ERR); 1803 } 1804 1805 /* 1806 * Recalculate the transmit timer / rtt. 1807 * 1808 * Some boxes send broken timestamp replies 1809 * during the SYN+ACK phase, ignore 1810 * timestamps of 0 or we could calculate a 1811 * huge RTT and blow up the retransmit timer. 1812 */ 1813 if ((to.to_flags & TOF_TS) != 0 && 1814 to.to_tsecr) { 1815 uint32_t t; 1816 1817 t = tcp_ts_getticks() - to.to_tsecr; 1818 if (!tp->t_rttlow || tp->t_rttlow > t) 1819 tp->t_rttlow = t; 1820 tcp_xmit_timer(tp, 1821 TCP_TS_TO_TICKS(t) + 1); 1822 } else if (tp->t_rtttime && 1823 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1824 if (!tp->t_rttlow || 1825 tp->t_rttlow > ticks - tp->t_rtttime) 1826 tp->t_rttlow = ticks - tp->t_rtttime; 1827 tcp_xmit_timer(tp, 1828 ticks - tp->t_rtttime); 1829 } 1830 acked = BYTES_THIS_ACK(tp, th); 1831 1832 #ifdef TCP_HHOOK 1833 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1834 hhook_run_tcp_est_in(tp, th, &to); 1835 #endif 1836 1837 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1838 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1839 sbdrop(&so->so_snd, acked); 1840 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1841 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1842 tp->snd_recover = th->th_ack - 1; 1843 1844 /* 1845 * Let the congestion control algorithm update 1846 * congestion control related information. This 1847 * typically means increasing the congestion 1848 * window. 1849 */ 1850 cc_ack_received(tp, th, nsegs, CC_ACK); 1851 1852 tp->snd_una = th->th_ack; 1853 /* 1854 * Pull snd_wl2 up to prevent seq wrap relative 1855 * to th_ack. 1856 */ 1857 tp->snd_wl2 = th->th_ack; 1858 tp->t_dupacks = 0; 1859 m_freem(m); 1860 1861 /* 1862 * If all outstanding data are acked, stop 1863 * retransmit timer, otherwise restart timer 1864 * using current (possibly backed-off) value. 1865 * If process is waiting for space, 1866 * wakeup/selwakeup/signal. If data 1867 * are ready to send, let tcp_output 1868 * decide between more output or persist. 1869 */ 1870 #ifdef TCPDEBUG 1871 if (so->so_options & SO_DEBUG) 1872 tcp_trace(TA_INPUT, ostate, tp, 1873 (void *)tcp_saveipgen, 1874 &tcp_savetcp, 0); 1875 #endif 1876 TCP_PROBE3(debug__input, tp, th, m); 1877 if (tp->snd_una == tp->snd_max) 1878 tcp_timer_activate(tp, TT_REXMT, 0); 1879 else if (!tcp_timer_active(tp, TT_PERSIST)) 1880 tcp_timer_activate(tp, TT_REXMT, 1881 tp->t_rxtcur); 1882 sowwakeup(so); 1883 if (sbavail(&so->so_snd)) 1884 (void) tp->t_fb->tfb_tcp_output(tp); 1885 goto check_delack; 1886 } 1887 } else if (th->th_ack == tp->snd_una && 1888 tlen <= sbspace(&so->so_rcv)) { 1889 int newsize = 0; /* automatic sockbuf scaling */ 1890 1891 /* 1892 * This is a pure, in-sequence data packet with 1893 * nothing on the reassembly queue and we have enough 1894 * buffer space to take it. 1895 */ 1896 if (ti_locked == TI_RLOCKED) 1897 INP_INFO_RUNLOCK(&V_tcbinfo); 1898 ti_locked = TI_UNLOCKED; 1899 1900 /* Clean receiver SACK report if present */ 1901 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1902 tcp_clean_sackreport(tp); 1903 TCPSTAT_INC(tcps_preddat); 1904 tp->rcv_nxt += tlen; 1905 /* 1906 * Pull snd_wl1 up to prevent seq wrap relative to 1907 * th_seq. 1908 */ 1909 tp->snd_wl1 = th->th_seq; 1910 /* 1911 * Pull rcv_up up to prevent seq wrap relative to 1912 * rcv_nxt. 1913 */ 1914 tp->rcv_up = tp->rcv_nxt; 1915 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1916 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1917 #ifdef TCPDEBUG 1918 if (so->so_options & SO_DEBUG) 1919 tcp_trace(TA_INPUT, ostate, tp, 1920 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1921 #endif 1922 TCP_PROBE3(debug__input, tp, th, m); 1923 1924 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1925 1926 /* Add data to socket buffer. */ 1927 SOCKBUF_LOCK(&so->so_rcv); 1928 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1929 m_freem(m); 1930 } else { 1931 /* 1932 * Set new socket buffer size. 1933 * Give up when limit is reached. 1934 */ 1935 if (newsize) 1936 if (!sbreserve_locked(&so->so_rcv, 1937 newsize, so, NULL)) 1938 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1939 m_adj(m, drop_hdrlen); /* delayed header drop */ 1940 sbappendstream_locked(&so->so_rcv, m, 0); 1941 } 1942 /* NB: sorwakeup_locked() does an implicit unlock. */ 1943 sorwakeup_locked(so); 1944 if (DELAY_ACK(tp, tlen)) { 1945 tp->t_flags |= TF_DELACK; 1946 } else { 1947 tp->t_flags |= TF_ACKNOW; 1948 tp->t_fb->tfb_tcp_output(tp); 1949 } 1950 goto check_delack; 1951 } 1952 } 1953 1954 /* 1955 * Calculate amount of space in receive window, 1956 * and then do TCP input processing. 1957 * Receive window is amount of space in rcv queue, 1958 * but not less than advertised window. 1959 */ 1960 win = sbspace(&so->so_rcv); 1961 if (win < 0) 1962 win = 0; 1963 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1964 1965 switch (tp->t_state) { 1966 1967 /* 1968 * If the state is SYN_RECEIVED: 1969 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1970 */ 1971 case TCPS_SYN_RECEIVED: 1972 if ((thflags & TH_ACK) && 1973 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1974 SEQ_GT(th->th_ack, tp->snd_max))) { 1975 rstreason = BANDLIM_RST_OPENPORT; 1976 goto dropwithreset; 1977 } 1978 #ifdef TCP_RFC7413 1979 if (IS_FASTOPEN(tp->t_flags)) { 1980 /* 1981 * When a TFO connection is in SYN_RECEIVED, the 1982 * only valid packets are the initial SYN, a 1983 * retransmit/copy of the initial SYN (possibly with 1984 * a subset of the original data), a valid ACK, a 1985 * FIN, or a RST. 1986 */ 1987 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1988 rstreason = BANDLIM_RST_OPENPORT; 1989 goto dropwithreset; 1990 } else if (thflags & TH_SYN) { 1991 /* non-initial SYN is ignored */ 1992 if ((tcp_timer_active(tp, TT_DELACK) || 1993 tcp_timer_active(tp, TT_REXMT))) 1994 goto drop; 1995 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1996 goto drop; 1997 } 1998 } 1999 #endif 2000 break; 2001 2002 /* 2003 * If the state is SYN_SENT: 2004 * if seg contains an ACK, but not for our SYN, drop the input. 2005 * if seg contains a RST, then drop the connection. 2006 * if seg does not contain SYN, then drop it. 2007 * Otherwise this is an acceptable SYN segment 2008 * initialize tp->rcv_nxt and tp->irs 2009 * if seg contains ack then advance tp->snd_una 2010 * if seg contains an ECE and ECN support is enabled, the stream 2011 * is ECN capable. 2012 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2013 * arrange for segment to be acked (eventually) 2014 * continue processing rest of data/controls, beginning with URG 2015 */ 2016 case TCPS_SYN_SENT: 2017 if ((thflags & TH_ACK) && 2018 (SEQ_LEQ(th->th_ack, tp->iss) || 2019 SEQ_GT(th->th_ack, tp->snd_max))) { 2020 rstreason = BANDLIM_UNLIMITED; 2021 goto dropwithreset; 2022 } 2023 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2024 TCP_PROBE5(connect__refused, NULL, tp, 2025 m, tp, th); 2026 tp = tcp_drop(tp, ECONNREFUSED); 2027 } 2028 if (thflags & TH_RST) 2029 goto drop; 2030 if (!(thflags & TH_SYN)) 2031 goto drop; 2032 2033 tp->irs = th->th_seq; 2034 tcp_rcvseqinit(tp); 2035 if (thflags & TH_ACK) { 2036 TCPSTAT_INC(tcps_connects); 2037 soisconnected(so); 2038 #ifdef MAC 2039 mac_socketpeer_set_from_mbuf(m, so); 2040 #endif 2041 /* Do window scaling on this connection? */ 2042 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2043 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2044 tp->rcv_scale = tp->request_r_scale; 2045 } 2046 tp->rcv_adv += min(tp->rcv_wnd, 2047 TCP_MAXWIN << tp->rcv_scale); 2048 tp->snd_una++; /* SYN is acked */ 2049 /* 2050 * If there's data, delay ACK; if there's also a FIN 2051 * ACKNOW will be turned on later. 2052 */ 2053 if (DELAY_ACK(tp, tlen) && tlen != 0) 2054 tcp_timer_activate(tp, TT_DELACK, 2055 tcp_delacktime); 2056 else 2057 tp->t_flags |= TF_ACKNOW; 2058 2059 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2060 tp->t_flags |= TF_ECN_PERMIT; 2061 TCPSTAT_INC(tcps_ecn_shs); 2062 } 2063 2064 /* 2065 * Received <SYN,ACK> in SYN_SENT[*] state. 2066 * Transitions: 2067 * SYN_SENT --> ESTABLISHED 2068 * SYN_SENT* --> FIN_WAIT_1 2069 */ 2070 tp->t_starttime = ticks; 2071 if (tp->t_flags & TF_NEEDFIN) { 2072 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2073 tp->t_flags &= ~TF_NEEDFIN; 2074 thflags &= ~TH_SYN; 2075 } else { 2076 tcp_state_change(tp, TCPS_ESTABLISHED); 2077 TCP_PROBE5(connect__established, NULL, tp, 2078 m, tp, th); 2079 cc_conn_init(tp); 2080 tcp_timer_activate(tp, TT_KEEP, 2081 TP_KEEPIDLE(tp)); 2082 } 2083 } else { 2084 /* 2085 * Received initial SYN in SYN-SENT[*] state => 2086 * simultaneous open. 2087 * If it succeeds, connection is * half-synchronized. 2088 * Otherwise, do 3-way handshake: 2089 * SYN-SENT -> SYN-RECEIVED 2090 * SYN-SENT* -> SYN-RECEIVED* 2091 */ 2092 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2093 tcp_timer_activate(tp, TT_REXMT, 0); 2094 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2095 } 2096 2097 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2098 "ti_locked %d", __func__, ti_locked)); 2099 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2100 INP_WLOCK_ASSERT(tp->t_inpcb); 2101 2102 /* 2103 * Advance th->th_seq to correspond to first data byte. 2104 * If data, trim to stay within window, 2105 * dropping FIN if necessary. 2106 */ 2107 th->th_seq++; 2108 if (tlen > tp->rcv_wnd) { 2109 todrop = tlen - tp->rcv_wnd; 2110 m_adj(m, -todrop); 2111 tlen = tp->rcv_wnd; 2112 thflags &= ~TH_FIN; 2113 TCPSTAT_INC(tcps_rcvpackafterwin); 2114 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2115 } 2116 tp->snd_wl1 = th->th_seq - 1; 2117 tp->rcv_up = th->th_seq; 2118 /* 2119 * Client side of transaction: already sent SYN and data. 2120 * If the remote host used T/TCP to validate the SYN, 2121 * our data will be ACK'd; if so, enter normal data segment 2122 * processing in the middle of step 5, ack processing. 2123 * Otherwise, goto step 6. 2124 */ 2125 if (thflags & TH_ACK) 2126 goto process_ACK; 2127 2128 goto step6; 2129 2130 /* 2131 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2132 * do normal processing. 2133 * 2134 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2135 */ 2136 case TCPS_LAST_ACK: 2137 case TCPS_CLOSING: 2138 break; /* continue normal processing */ 2139 } 2140 2141 /* 2142 * States other than LISTEN or SYN_SENT. 2143 * First check the RST flag and sequence number since reset segments 2144 * are exempt from the timestamp and connection count tests. This 2145 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2146 * below which allowed reset segments in half the sequence space 2147 * to fall though and be processed (which gives forged reset 2148 * segments with a random sequence number a 50 percent chance of 2149 * killing a connection). 2150 * Then check timestamp, if present. 2151 * Then check the connection count, if present. 2152 * Then check that at least some bytes of segment are within 2153 * receive window. If segment begins before rcv_nxt, 2154 * drop leading data (and SYN); if nothing left, just ack. 2155 */ 2156 if (thflags & TH_RST) { 2157 /* 2158 * RFC5961 Section 3.2 2159 * 2160 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2161 * - If RST is in window, we send challenge ACK. 2162 * 2163 * Note: to take into account delayed ACKs, we should 2164 * test against last_ack_sent instead of rcv_nxt. 2165 * Note 2: we handle special case of closed window, not 2166 * covered by the RFC. 2167 */ 2168 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2169 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2170 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2171 2172 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2173 KASSERT(ti_locked == TI_RLOCKED, 2174 ("%s: TH_RST ti_locked %d, th %p tp %p", 2175 __func__, ti_locked, th, tp)); 2176 KASSERT(tp->t_state != TCPS_SYN_SENT, 2177 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2178 __func__, th, tp)); 2179 2180 if (V_tcp_insecure_rst || 2181 tp->last_ack_sent == th->th_seq) { 2182 TCPSTAT_INC(tcps_drops); 2183 /* Drop the connection. */ 2184 switch (tp->t_state) { 2185 case TCPS_SYN_RECEIVED: 2186 so->so_error = ECONNREFUSED; 2187 goto close; 2188 case TCPS_ESTABLISHED: 2189 case TCPS_FIN_WAIT_1: 2190 case TCPS_FIN_WAIT_2: 2191 case TCPS_CLOSE_WAIT: 2192 case TCPS_CLOSING: 2193 case TCPS_LAST_ACK: 2194 so->so_error = ECONNRESET; 2195 close: 2196 /* FALLTHROUGH */ 2197 default: 2198 tp = tcp_close(tp); 2199 } 2200 } else { 2201 TCPSTAT_INC(tcps_badrst); 2202 /* Send challenge ACK. */ 2203 tcp_respond(tp, mtod(m, void *), th, m, 2204 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2205 tp->last_ack_sent = tp->rcv_nxt; 2206 m = NULL; 2207 } 2208 } 2209 goto drop; 2210 } 2211 2212 /* 2213 * RFC5961 Section 4.2 2214 * Send challenge ACK for any SYN in synchronized state. 2215 */ 2216 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2217 tp->t_state != TCPS_SYN_RECEIVED) { 2218 KASSERT(ti_locked == TI_RLOCKED, 2219 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2220 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2221 2222 TCPSTAT_INC(tcps_badsyn); 2223 if (V_tcp_insecure_syn && 2224 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2225 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2226 tp = tcp_drop(tp, ECONNRESET); 2227 rstreason = BANDLIM_UNLIMITED; 2228 } else { 2229 /* Send challenge ACK. */ 2230 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2231 tp->snd_nxt, TH_ACK); 2232 tp->last_ack_sent = tp->rcv_nxt; 2233 m = NULL; 2234 } 2235 goto drop; 2236 } 2237 2238 /* 2239 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2240 * and it's less than ts_recent, drop it. 2241 */ 2242 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2243 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2244 2245 /* Check to see if ts_recent is over 24 days old. */ 2246 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2247 /* 2248 * Invalidate ts_recent. If this segment updates 2249 * ts_recent, the age will be reset later and ts_recent 2250 * will get a valid value. If it does not, setting 2251 * ts_recent to zero will at least satisfy the 2252 * requirement that zero be placed in the timestamp 2253 * echo reply when ts_recent isn't valid. The 2254 * age isn't reset until we get a valid ts_recent 2255 * because we don't want out-of-order segments to be 2256 * dropped when ts_recent is old. 2257 */ 2258 tp->ts_recent = 0; 2259 } else { 2260 TCPSTAT_INC(tcps_rcvduppack); 2261 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2262 TCPSTAT_INC(tcps_pawsdrop); 2263 if (tlen) 2264 goto dropafterack; 2265 goto drop; 2266 } 2267 } 2268 2269 /* 2270 * In the SYN-RECEIVED state, validate that the packet belongs to 2271 * this connection before trimming the data to fit the receive 2272 * window. Check the sequence number versus IRS since we know 2273 * the sequence numbers haven't wrapped. This is a partial fix 2274 * for the "LAND" DoS attack. 2275 */ 2276 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2277 rstreason = BANDLIM_RST_OPENPORT; 2278 goto dropwithreset; 2279 } 2280 2281 todrop = tp->rcv_nxt - th->th_seq; 2282 if (todrop > 0) { 2283 if (thflags & TH_SYN) { 2284 thflags &= ~TH_SYN; 2285 th->th_seq++; 2286 if (th->th_urp > 1) 2287 th->th_urp--; 2288 else 2289 thflags &= ~TH_URG; 2290 todrop--; 2291 } 2292 /* 2293 * Following if statement from Stevens, vol. 2, p. 960. 2294 */ 2295 if (todrop > tlen 2296 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2297 /* 2298 * Any valid FIN must be to the left of the window. 2299 * At this point the FIN must be a duplicate or out 2300 * of sequence; drop it. 2301 */ 2302 thflags &= ~TH_FIN; 2303 2304 /* 2305 * Send an ACK to resynchronize and drop any data. 2306 * But keep on processing for RST or ACK. 2307 */ 2308 tp->t_flags |= TF_ACKNOW; 2309 todrop = tlen; 2310 TCPSTAT_INC(tcps_rcvduppack); 2311 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2312 } else { 2313 TCPSTAT_INC(tcps_rcvpartduppack); 2314 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2315 } 2316 drop_hdrlen += todrop; /* drop from the top afterwards */ 2317 th->th_seq += todrop; 2318 tlen -= todrop; 2319 if (th->th_urp > todrop) 2320 th->th_urp -= todrop; 2321 else { 2322 thflags &= ~TH_URG; 2323 th->th_urp = 0; 2324 } 2325 } 2326 2327 /* 2328 * If new data are received on a connection after the 2329 * user processes are gone, then RST the other end. 2330 */ 2331 if ((so->so_state & SS_NOFDREF) && 2332 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2333 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2334 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2335 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2336 2337 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2338 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2339 "after socket was closed, " 2340 "sending RST and removing tcpcb\n", 2341 s, __func__, tcpstates[tp->t_state], tlen); 2342 free(s, M_TCPLOG); 2343 } 2344 tp = tcp_close(tp); 2345 TCPSTAT_INC(tcps_rcvafterclose); 2346 rstreason = BANDLIM_UNLIMITED; 2347 goto dropwithreset; 2348 } 2349 2350 /* 2351 * If segment ends after window, drop trailing data 2352 * (and PUSH and FIN); if nothing left, just ACK. 2353 */ 2354 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2355 if (todrop > 0) { 2356 TCPSTAT_INC(tcps_rcvpackafterwin); 2357 if (todrop >= tlen) { 2358 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2359 /* 2360 * If window is closed can only take segments at 2361 * window edge, and have to drop data and PUSH from 2362 * incoming segments. Continue processing, but 2363 * remember to ack. Otherwise, drop segment 2364 * and ack. 2365 */ 2366 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2367 tp->t_flags |= TF_ACKNOW; 2368 TCPSTAT_INC(tcps_rcvwinprobe); 2369 } else 2370 goto dropafterack; 2371 } else 2372 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2373 m_adj(m, -todrop); 2374 tlen -= todrop; 2375 thflags &= ~(TH_PUSH|TH_FIN); 2376 } 2377 2378 /* 2379 * If last ACK falls within this segment's sequence numbers, 2380 * record its timestamp. 2381 * NOTE: 2382 * 1) That the test incorporates suggestions from the latest 2383 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2384 * 2) That updating only on newer timestamps interferes with 2385 * our earlier PAWS tests, so this check should be solely 2386 * predicated on the sequence space of this segment. 2387 * 3) That we modify the segment boundary check to be 2388 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2389 * instead of RFC1323's 2390 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2391 * This modified check allows us to overcome RFC1323's 2392 * limitations as described in Stevens TCP/IP Illustrated 2393 * Vol. 2 p.869. In such cases, we can still calculate the 2394 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2395 */ 2396 if ((to.to_flags & TOF_TS) != 0 && 2397 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2398 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2399 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2400 tp->ts_recent_age = tcp_ts_getticks(); 2401 tp->ts_recent = to.to_tsval; 2402 } 2403 2404 /* 2405 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2406 * flag is on (half-synchronized state), then queue data for 2407 * later processing; else drop segment and return. 2408 */ 2409 if ((thflags & TH_ACK) == 0) { 2410 if (tp->t_state == TCPS_SYN_RECEIVED || 2411 (tp->t_flags & TF_NEEDSYN)) { 2412 #ifdef TCP_RFC7413 2413 if (tp->t_state == TCPS_SYN_RECEIVED && 2414 IS_FASTOPEN(tp->t_flags)) { 2415 tp->snd_wnd = tiwin; 2416 cc_conn_init(tp); 2417 } 2418 #endif 2419 goto step6; 2420 } else if (tp->t_flags & TF_ACKNOW) 2421 goto dropafterack; 2422 else 2423 goto drop; 2424 } 2425 2426 /* 2427 * Ack processing. 2428 */ 2429 switch (tp->t_state) { 2430 2431 /* 2432 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2433 * ESTABLISHED state and continue processing. 2434 * The ACK was checked above. 2435 */ 2436 case TCPS_SYN_RECEIVED: 2437 2438 TCPSTAT_INC(tcps_connects); 2439 soisconnected(so); 2440 /* Do window scaling? */ 2441 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2442 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2443 tp->rcv_scale = tp->request_r_scale; 2444 tp->snd_wnd = tiwin; 2445 } 2446 /* 2447 * Make transitions: 2448 * SYN-RECEIVED -> ESTABLISHED 2449 * SYN-RECEIVED* -> FIN-WAIT-1 2450 */ 2451 tp->t_starttime = ticks; 2452 if (tp->t_flags & TF_NEEDFIN) { 2453 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2454 tp->t_flags &= ~TF_NEEDFIN; 2455 } else { 2456 tcp_state_change(tp, TCPS_ESTABLISHED); 2457 TCP_PROBE5(accept__established, NULL, tp, 2458 m, tp, th); 2459 #ifdef TCP_RFC7413 2460 if (tp->t_tfo_pending) { 2461 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2462 tp->t_tfo_pending = NULL; 2463 2464 /* 2465 * Account for the ACK of our SYN prior to 2466 * regular ACK processing below. 2467 */ 2468 tp->snd_una++; 2469 } 2470 /* 2471 * TFO connections call cc_conn_init() during SYN 2472 * processing. Calling it again here for such 2473 * connections is not harmless as it would undo the 2474 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2475 * is retransmitted. 2476 */ 2477 if (!IS_FASTOPEN(tp->t_flags)) 2478 #endif 2479 cc_conn_init(tp); 2480 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2481 } 2482 /* 2483 * If segment contains data or ACK, will call tcp_reass() 2484 * later; if not, do so now to pass queued data to user. 2485 */ 2486 if (tlen == 0 && (thflags & TH_FIN) == 0) 2487 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2488 (struct mbuf *)0); 2489 tp->snd_wl1 = th->th_seq - 1; 2490 /* FALLTHROUGH */ 2491 2492 /* 2493 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2494 * ACKs. If the ack is in the range 2495 * tp->snd_una < th->th_ack <= tp->snd_max 2496 * then advance tp->snd_una to th->th_ack and drop 2497 * data from the retransmission queue. If this ACK reflects 2498 * more up to date window information we update our window information. 2499 */ 2500 case TCPS_ESTABLISHED: 2501 case TCPS_FIN_WAIT_1: 2502 case TCPS_FIN_WAIT_2: 2503 case TCPS_CLOSE_WAIT: 2504 case TCPS_CLOSING: 2505 case TCPS_LAST_ACK: 2506 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2507 TCPSTAT_INC(tcps_rcvacktoomuch); 2508 goto dropafterack; 2509 } 2510 if ((tp->t_flags & TF_SACK_PERMIT) && 2511 ((to.to_flags & TOF_SACK) || 2512 !TAILQ_EMPTY(&tp->snd_holes))) 2513 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2514 else 2515 /* 2516 * Reset the value so that previous (valid) value 2517 * from the last ack with SACK doesn't get used. 2518 */ 2519 tp->sackhint.sacked_bytes = 0; 2520 2521 #ifdef TCP_HHOOK 2522 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2523 hhook_run_tcp_est_in(tp, th, &to); 2524 #endif 2525 2526 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2527 u_int maxseg; 2528 2529 maxseg = tcp_maxseg(tp); 2530 if (tlen == 0 && 2531 (tiwin == tp->snd_wnd || 2532 (tp->t_flags & TF_SACK_PERMIT))) { 2533 /* 2534 * If this is the first time we've seen a 2535 * FIN from the remote, this is not a 2536 * duplicate and it needs to be processed 2537 * normally. This happens during a 2538 * simultaneous close. 2539 */ 2540 if ((thflags & TH_FIN) && 2541 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2542 tp->t_dupacks = 0; 2543 break; 2544 } 2545 TCPSTAT_INC(tcps_rcvdupack); 2546 /* 2547 * If we have outstanding data (other than 2548 * a window probe), this is a completely 2549 * duplicate ack (ie, window info didn't 2550 * change and FIN isn't set), 2551 * the ack is the biggest we've 2552 * seen and we've seen exactly our rexmt 2553 * threshold of them, assume a packet 2554 * has been dropped and retransmit it. 2555 * Kludge snd_nxt & the congestion 2556 * window so we send only this one 2557 * packet. 2558 * 2559 * We know we're losing at the current 2560 * window size so do congestion avoidance 2561 * (set ssthresh to half the current window 2562 * and pull our congestion window back to 2563 * the new ssthresh). 2564 * 2565 * Dup acks mean that packets have left the 2566 * network (they're now cached at the receiver) 2567 * so bump cwnd by the amount in the receiver 2568 * to keep a constant cwnd packets in the 2569 * network. 2570 * 2571 * When using TCP ECN, notify the peer that 2572 * we reduced the cwnd. 2573 */ 2574 /* 2575 * Following 2 kinds of acks should not affect 2576 * dupack counting: 2577 * 1) Old acks 2578 * 2) Acks with SACK but without any new SACK 2579 * information in them. These could result from 2580 * any anomaly in the network like a switch 2581 * duplicating packets or a possible DoS attack. 2582 */ 2583 if (th->th_ack != tp->snd_una || 2584 ((tp->t_flags & TF_SACK_PERMIT) && 2585 !sack_changed)) 2586 break; 2587 else if (!tcp_timer_active(tp, TT_REXMT)) 2588 tp->t_dupacks = 0; 2589 else if (++tp->t_dupacks > tcprexmtthresh || 2590 IN_FASTRECOVERY(tp->t_flags)) { 2591 cc_ack_received(tp, th, nsegs, 2592 CC_DUPACK); 2593 if ((tp->t_flags & TF_SACK_PERMIT) && 2594 IN_FASTRECOVERY(tp->t_flags)) { 2595 int awnd; 2596 2597 /* 2598 * Compute the amount of data in flight first. 2599 * We can inject new data into the pipe iff 2600 * we have less than 1/2 the original window's 2601 * worth of data in flight. 2602 */ 2603 if (V_tcp_do_rfc6675_pipe) 2604 awnd = tcp_compute_pipe(tp); 2605 else 2606 awnd = (tp->snd_nxt - tp->snd_fack) + 2607 tp->sackhint.sack_bytes_rexmit; 2608 2609 if (awnd < tp->snd_ssthresh) { 2610 tp->snd_cwnd += maxseg; 2611 /* 2612 * RFC5681 Section 3.2 talks about cwnd 2613 * inflation on additional dupacks and 2614 * deflation on recovering from loss. 2615 * 2616 * We keep cwnd into check so that 2617 * we don't have to 'deflate' it when we 2618 * get out of recovery. 2619 */ 2620 if (tp->snd_cwnd > tp->snd_ssthresh) 2621 tp->snd_cwnd = tp->snd_ssthresh; 2622 } 2623 } else 2624 tp->snd_cwnd += maxseg; 2625 (void) tp->t_fb->tfb_tcp_output(tp); 2626 goto drop; 2627 } else if (tp->t_dupacks == tcprexmtthresh) { 2628 tcp_seq onxt = tp->snd_nxt; 2629 2630 /* 2631 * If we're doing sack, check to 2632 * see if we're already in sack 2633 * recovery. If we're not doing sack, 2634 * check to see if we're in newreno 2635 * recovery. 2636 */ 2637 if (tp->t_flags & TF_SACK_PERMIT) { 2638 if (IN_FASTRECOVERY(tp->t_flags)) { 2639 tp->t_dupacks = 0; 2640 break; 2641 } 2642 } else { 2643 if (SEQ_LEQ(th->th_ack, 2644 tp->snd_recover)) { 2645 tp->t_dupacks = 0; 2646 break; 2647 } 2648 } 2649 /* Congestion signal before ack. */ 2650 cc_cong_signal(tp, th, CC_NDUPACK); 2651 cc_ack_received(tp, th, nsegs, 2652 CC_DUPACK); 2653 tcp_timer_activate(tp, TT_REXMT, 0); 2654 tp->t_rtttime = 0; 2655 if (tp->t_flags & TF_SACK_PERMIT) { 2656 TCPSTAT_INC( 2657 tcps_sack_recovery_episode); 2658 tp->sack_newdata = tp->snd_nxt; 2659 if (CC_ALGO(tp)->cong_signal == NULL) 2660 tp->snd_cwnd = maxseg; 2661 (void) tp->t_fb->tfb_tcp_output(tp); 2662 goto drop; 2663 } 2664 tp->snd_nxt = th->th_ack; 2665 if (CC_ALGO(tp)->cong_signal == NULL) 2666 tp->snd_cwnd = maxseg; 2667 (void) tp->t_fb->tfb_tcp_output(tp); 2668 KASSERT(tp->snd_limited <= 2, 2669 ("%s: tp->snd_limited too big", 2670 __func__)); 2671 if (CC_ALGO(tp)->cong_signal == NULL) 2672 tp->snd_cwnd = tp->snd_ssthresh + 2673 maxseg * 2674 (tp->t_dupacks - tp->snd_limited); 2675 if (SEQ_GT(onxt, tp->snd_nxt)) 2676 tp->snd_nxt = onxt; 2677 goto drop; 2678 } else if (V_tcp_do_rfc3042) { 2679 /* 2680 * Process first and second duplicate 2681 * ACKs. Each indicates a segment 2682 * leaving the network, creating room 2683 * for more. Make sure we can send a 2684 * packet on reception of each duplicate 2685 * ACK by increasing snd_cwnd by one 2686 * segment. Restore the original 2687 * snd_cwnd after packet transmission. 2688 */ 2689 cc_ack_received(tp, th, nsegs, 2690 CC_DUPACK); 2691 uint32_t oldcwnd = tp->snd_cwnd; 2692 tcp_seq oldsndmax = tp->snd_max; 2693 u_int sent; 2694 int avail; 2695 2696 KASSERT(tp->t_dupacks == 1 || 2697 tp->t_dupacks == 2, 2698 ("%s: dupacks not 1 or 2", 2699 __func__)); 2700 if (tp->t_dupacks == 1) 2701 tp->snd_limited = 0; 2702 tp->snd_cwnd = 2703 (tp->snd_nxt - tp->snd_una) + 2704 (tp->t_dupacks - tp->snd_limited) * 2705 maxseg; 2706 /* 2707 * Only call tcp_output when there 2708 * is new data available to be sent. 2709 * Otherwise we would send pure ACKs. 2710 */ 2711 SOCKBUF_LOCK(&so->so_snd); 2712 avail = sbavail(&so->so_snd) - 2713 (tp->snd_nxt - tp->snd_una); 2714 SOCKBUF_UNLOCK(&so->so_snd); 2715 if (avail > 0) 2716 (void) tp->t_fb->tfb_tcp_output(tp); 2717 sent = tp->snd_max - oldsndmax; 2718 if (sent > maxseg) { 2719 KASSERT((tp->t_dupacks == 2 && 2720 tp->snd_limited == 0) || 2721 (sent == maxseg + 1 && 2722 tp->t_flags & TF_SENTFIN), 2723 ("%s: sent too much", 2724 __func__)); 2725 tp->snd_limited = 2; 2726 } else if (sent > 0) 2727 ++tp->snd_limited; 2728 tp->snd_cwnd = oldcwnd; 2729 goto drop; 2730 } 2731 } 2732 break; 2733 } else { 2734 /* 2735 * This ack is advancing the left edge, reset the 2736 * counter. 2737 */ 2738 tp->t_dupacks = 0; 2739 /* 2740 * If this ack also has new SACK info, increment the 2741 * counter as per rfc6675. 2742 */ 2743 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2744 tp->t_dupacks++; 2745 } 2746 2747 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2748 ("%s: th_ack <= snd_una", __func__)); 2749 2750 /* 2751 * If the congestion window was inflated to account 2752 * for the other side's cached packets, retract it. 2753 */ 2754 if (IN_FASTRECOVERY(tp->t_flags)) { 2755 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2756 if (tp->t_flags & TF_SACK_PERMIT) 2757 tcp_sack_partialack(tp, th); 2758 else 2759 tcp_newreno_partial_ack(tp, th); 2760 } else 2761 cc_post_recovery(tp, th); 2762 } 2763 /* 2764 * If we reach this point, ACK is not a duplicate, 2765 * i.e., it ACKs something we sent. 2766 */ 2767 if (tp->t_flags & TF_NEEDSYN) { 2768 /* 2769 * T/TCP: Connection was half-synchronized, and our 2770 * SYN has been ACK'd (so connection is now fully 2771 * synchronized). Go to non-starred state, 2772 * increment snd_una for ACK of SYN, and check if 2773 * we can do window scaling. 2774 */ 2775 tp->t_flags &= ~TF_NEEDSYN; 2776 tp->snd_una++; 2777 /* Do window scaling? */ 2778 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2779 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2780 tp->rcv_scale = tp->request_r_scale; 2781 /* Send window already scaled. */ 2782 } 2783 } 2784 2785 process_ACK: 2786 INP_WLOCK_ASSERT(tp->t_inpcb); 2787 2788 acked = BYTES_THIS_ACK(tp, th); 2789 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2790 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2791 tp->snd_una, th->th_ack, tp, m)); 2792 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2793 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2794 2795 /* 2796 * If we just performed our first retransmit, and the ACK 2797 * arrives within our recovery window, then it was a mistake 2798 * to do the retransmit in the first place. Recover our 2799 * original cwnd and ssthresh, and proceed to transmit where 2800 * we left off. 2801 */ 2802 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2803 (int)(ticks - tp->t_badrxtwin) < 0) 2804 cc_cong_signal(tp, th, CC_RTO_ERR); 2805 2806 /* 2807 * If we have a timestamp reply, update smoothed 2808 * round trip time. If no timestamp is present but 2809 * transmit timer is running and timed sequence 2810 * number was acked, update smoothed round trip time. 2811 * Since we now have an rtt measurement, cancel the 2812 * timer backoff (cf., Phil Karn's retransmit alg.). 2813 * Recompute the initial retransmit timer. 2814 * 2815 * Some boxes send broken timestamp replies 2816 * during the SYN+ACK phase, ignore 2817 * timestamps of 0 or we could calculate a 2818 * huge RTT and blow up the retransmit timer. 2819 */ 2820 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2821 uint32_t t; 2822 2823 t = tcp_ts_getticks() - to.to_tsecr; 2824 if (!tp->t_rttlow || tp->t_rttlow > t) 2825 tp->t_rttlow = t; 2826 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2827 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2828 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2829 tp->t_rttlow = ticks - tp->t_rtttime; 2830 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2831 } 2832 2833 /* 2834 * If all outstanding data is acked, stop retransmit 2835 * timer and remember to restart (more output or persist). 2836 * If there is more data to be acked, restart retransmit 2837 * timer, using current (possibly backed-off) value. 2838 */ 2839 if (th->th_ack == tp->snd_max) { 2840 tcp_timer_activate(tp, TT_REXMT, 0); 2841 needoutput = 1; 2842 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2843 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2844 2845 /* 2846 * If no data (only SYN) was ACK'd, 2847 * skip rest of ACK processing. 2848 */ 2849 if (acked == 0) 2850 goto step6; 2851 2852 /* 2853 * Let the congestion control algorithm update congestion 2854 * control related information. This typically means increasing 2855 * the congestion window. 2856 */ 2857 cc_ack_received(tp, th, nsegs, CC_ACK); 2858 2859 SOCKBUF_LOCK(&so->so_snd); 2860 if (acked > sbavail(&so->so_snd)) { 2861 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2862 tp->snd_wnd -= sbavail(&so->so_snd); 2863 else 2864 tp->snd_wnd = 0; 2865 mfree = sbcut_locked(&so->so_snd, 2866 (int)sbavail(&so->so_snd)); 2867 ourfinisacked = 1; 2868 } else { 2869 mfree = sbcut_locked(&so->so_snd, acked); 2870 if (tp->snd_wnd >= (uint32_t) acked) 2871 tp->snd_wnd -= acked; 2872 else 2873 tp->snd_wnd = 0; 2874 ourfinisacked = 0; 2875 } 2876 /* NB: sowwakeup_locked() does an implicit unlock. */ 2877 sowwakeup_locked(so); 2878 m_freem(mfree); 2879 /* Detect una wraparound. */ 2880 if (!IN_RECOVERY(tp->t_flags) && 2881 SEQ_GT(tp->snd_una, tp->snd_recover) && 2882 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2883 tp->snd_recover = th->th_ack - 1; 2884 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2885 if (IN_RECOVERY(tp->t_flags) && 2886 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2887 EXIT_RECOVERY(tp->t_flags); 2888 } 2889 tp->snd_una = th->th_ack; 2890 if (tp->t_flags & TF_SACK_PERMIT) { 2891 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2892 tp->snd_recover = tp->snd_una; 2893 } 2894 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2895 tp->snd_nxt = tp->snd_una; 2896 2897 switch (tp->t_state) { 2898 2899 /* 2900 * In FIN_WAIT_1 STATE in addition to the processing 2901 * for the ESTABLISHED state if our FIN is now acknowledged 2902 * then enter FIN_WAIT_2. 2903 */ 2904 case TCPS_FIN_WAIT_1: 2905 if (ourfinisacked) { 2906 /* 2907 * If we can't receive any more 2908 * data, then closing user can proceed. 2909 * Starting the timer is contrary to the 2910 * specification, but if we don't get a FIN 2911 * we'll hang forever. 2912 * 2913 * XXXjl: 2914 * we should release the tp also, and use a 2915 * compressed state. 2916 */ 2917 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2918 soisdisconnected(so); 2919 tcp_timer_activate(tp, TT_2MSL, 2920 (tcp_fast_finwait2_recycle ? 2921 tcp_finwait2_timeout : 2922 TP_MAXIDLE(tp))); 2923 } 2924 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2925 } 2926 break; 2927 2928 /* 2929 * In CLOSING STATE in addition to the processing for 2930 * the ESTABLISHED state if the ACK acknowledges our FIN 2931 * then enter the TIME-WAIT state, otherwise ignore 2932 * the segment. 2933 */ 2934 case TCPS_CLOSING: 2935 if (ourfinisacked) { 2936 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2937 tcp_twstart(tp); 2938 INP_INFO_RUNLOCK(&V_tcbinfo); 2939 m_freem(m); 2940 return; 2941 } 2942 break; 2943 2944 /* 2945 * In LAST_ACK, we may still be waiting for data to drain 2946 * and/or to be acked, as well as for the ack of our FIN. 2947 * If our FIN is now acknowledged, delete the TCB, 2948 * enter the closed state and return. 2949 */ 2950 case TCPS_LAST_ACK: 2951 if (ourfinisacked) { 2952 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2953 tp = tcp_close(tp); 2954 goto drop; 2955 } 2956 break; 2957 } 2958 } 2959 2960 step6: 2961 INP_WLOCK_ASSERT(tp->t_inpcb); 2962 2963 /* 2964 * Update window information. 2965 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2966 */ 2967 if ((thflags & TH_ACK) && 2968 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2969 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2970 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2971 /* keep track of pure window updates */ 2972 if (tlen == 0 && 2973 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2974 TCPSTAT_INC(tcps_rcvwinupd); 2975 tp->snd_wnd = tiwin; 2976 tp->snd_wl1 = th->th_seq; 2977 tp->snd_wl2 = th->th_ack; 2978 if (tp->snd_wnd > tp->max_sndwnd) 2979 tp->max_sndwnd = tp->snd_wnd; 2980 needoutput = 1; 2981 } 2982 2983 /* 2984 * Process segments with URG. 2985 */ 2986 if ((thflags & TH_URG) && th->th_urp && 2987 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2988 /* 2989 * This is a kludge, but if we receive and accept 2990 * random urgent pointers, we'll crash in 2991 * soreceive. It's hard to imagine someone 2992 * actually wanting to send this much urgent data. 2993 */ 2994 SOCKBUF_LOCK(&so->so_rcv); 2995 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2996 th->th_urp = 0; /* XXX */ 2997 thflags &= ~TH_URG; /* XXX */ 2998 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2999 goto dodata; /* XXX */ 3000 } 3001 /* 3002 * If this segment advances the known urgent pointer, 3003 * then mark the data stream. This should not happen 3004 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3005 * a FIN has been received from the remote side. 3006 * In these states we ignore the URG. 3007 * 3008 * According to RFC961 (Assigned Protocols), 3009 * the urgent pointer points to the last octet 3010 * of urgent data. We continue, however, 3011 * to consider it to indicate the first octet 3012 * of data past the urgent section as the original 3013 * spec states (in one of two places). 3014 */ 3015 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3016 tp->rcv_up = th->th_seq + th->th_urp; 3017 so->so_oobmark = sbavail(&so->so_rcv) + 3018 (tp->rcv_up - tp->rcv_nxt) - 1; 3019 if (so->so_oobmark == 0) 3020 so->so_rcv.sb_state |= SBS_RCVATMARK; 3021 sohasoutofband(so); 3022 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3023 } 3024 SOCKBUF_UNLOCK(&so->so_rcv); 3025 /* 3026 * Remove out of band data so doesn't get presented to user. 3027 * This can happen independent of advancing the URG pointer, 3028 * but if two URG's are pending at once, some out-of-band 3029 * data may creep in... ick. 3030 */ 3031 if (th->th_urp <= (uint32_t)tlen && 3032 !(so->so_options & SO_OOBINLINE)) { 3033 /* hdr drop is delayed */ 3034 tcp_pulloutofband(so, th, m, drop_hdrlen); 3035 } 3036 } else { 3037 /* 3038 * If no out of band data is expected, 3039 * pull receive urgent pointer along 3040 * with the receive window. 3041 */ 3042 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3043 tp->rcv_up = tp->rcv_nxt; 3044 } 3045 dodata: /* XXX */ 3046 INP_WLOCK_ASSERT(tp->t_inpcb); 3047 3048 /* 3049 * Process the segment text, merging it into the TCP sequencing queue, 3050 * and arranging for acknowledgment of receipt if necessary. 3051 * This process logically involves adjusting tp->rcv_wnd as data 3052 * is presented to the user (this happens in tcp_usrreq.c, 3053 * case PRU_RCVD). If a FIN has already been received on this 3054 * connection then we just ignore the text. 3055 */ 3056 #ifdef TCP_RFC7413 3057 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3058 IS_FASTOPEN(tp->t_flags)); 3059 #else 3060 #define tfo_syn (false) 3061 #endif 3062 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3063 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3064 tcp_seq save_start = th->th_seq; 3065 m_adj(m, drop_hdrlen); /* delayed header drop */ 3066 /* 3067 * Insert segment which includes th into TCP reassembly queue 3068 * with control block tp. Set thflags to whether reassembly now 3069 * includes a segment with FIN. This handles the common case 3070 * inline (segment is the next to be received on an established 3071 * connection, and the queue is empty), avoiding linkage into 3072 * and removal from the queue and repetition of various 3073 * conversions. 3074 * Set DELACK for segments received in order, but ack 3075 * immediately when segments are out of order (so 3076 * fast retransmit can work). 3077 */ 3078 if (th->th_seq == tp->rcv_nxt && 3079 LIST_EMPTY(&tp->t_segq) && 3080 (TCPS_HAVEESTABLISHED(tp->t_state) || 3081 tfo_syn)) { 3082 if (DELAY_ACK(tp, tlen) || tfo_syn) 3083 tp->t_flags |= TF_DELACK; 3084 else 3085 tp->t_flags |= TF_ACKNOW; 3086 tp->rcv_nxt += tlen; 3087 thflags = th->th_flags & TH_FIN; 3088 TCPSTAT_INC(tcps_rcvpack); 3089 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3090 SOCKBUF_LOCK(&so->so_rcv); 3091 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3092 m_freem(m); 3093 else 3094 sbappendstream_locked(&so->so_rcv, m, 0); 3095 /* NB: sorwakeup_locked() does an implicit unlock. */ 3096 sorwakeup_locked(so); 3097 } else { 3098 /* 3099 * XXX: Due to the header drop above "th" is 3100 * theoretically invalid by now. Fortunately 3101 * m_adj() doesn't actually frees any mbufs 3102 * when trimming from the head. 3103 */ 3104 thflags = tcp_reass(tp, th, &tlen, m); 3105 tp->t_flags |= TF_ACKNOW; 3106 } 3107 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3108 tcp_update_sack_list(tp, save_start, save_start + tlen); 3109 #if 0 3110 /* 3111 * Note the amount of data that peer has sent into 3112 * our window, in order to estimate the sender's 3113 * buffer size. 3114 * XXX: Unused. 3115 */ 3116 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3117 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3118 else 3119 len = so->so_rcv.sb_hiwat; 3120 #endif 3121 } else { 3122 m_freem(m); 3123 thflags &= ~TH_FIN; 3124 } 3125 3126 /* 3127 * If FIN is received ACK the FIN and let the user know 3128 * that the connection is closing. 3129 */ 3130 if (thflags & TH_FIN) { 3131 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3132 socantrcvmore(so); 3133 /* 3134 * If connection is half-synchronized 3135 * (ie NEEDSYN flag on) then delay ACK, 3136 * so it may be piggybacked when SYN is sent. 3137 * Otherwise, since we received a FIN then no 3138 * more input can be expected, send ACK now. 3139 */ 3140 if (tp->t_flags & TF_NEEDSYN) 3141 tp->t_flags |= TF_DELACK; 3142 else 3143 tp->t_flags |= TF_ACKNOW; 3144 tp->rcv_nxt++; 3145 } 3146 switch (tp->t_state) { 3147 3148 /* 3149 * In SYN_RECEIVED and ESTABLISHED STATES 3150 * enter the CLOSE_WAIT state. 3151 */ 3152 case TCPS_SYN_RECEIVED: 3153 tp->t_starttime = ticks; 3154 /* FALLTHROUGH */ 3155 case TCPS_ESTABLISHED: 3156 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3157 break; 3158 3159 /* 3160 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3161 * enter the CLOSING state. 3162 */ 3163 case TCPS_FIN_WAIT_1: 3164 tcp_state_change(tp, TCPS_CLOSING); 3165 break; 3166 3167 /* 3168 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3169 * starting the time-wait timer, turning off the other 3170 * standard timers. 3171 */ 3172 case TCPS_FIN_WAIT_2: 3173 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3174 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3175 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3176 ti_locked)); 3177 3178 tcp_twstart(tp); 3179 INP_INFO_RUNLOCK(&V_tcbinfo); 3180 return; 3181 } 3182 } 3183 if (ti_locked == TI_RLOCKED) 3184 INP_INFO_RUNLOCK(&V_tcbinfo); 3185 ti_locked = TI_UNLOCKED; 3186 3187 #ifdef TCPDEBUG 3188 if (so->so_options & SO_DEBUG) 3189 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3190 &tcp_savetcp, 0); 3191 #endif 3192 TCP_PROBE3(debug__input, tp, th, m); 3193 3194 /* 3195 * Return any desired output. 3196 */ 3197 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3198 (void) tp->t_fb->tfb_tcp_output(tp); 3199 3200 check_delack: 3201 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3202 __func__, ti_locked)); 3203 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3204 INP_WLOCK_ASSERT(tp->t_inpcb); 3205 3206 if (tp->t_flags & TF_DELACK) { 3207 tp->t_flags &= ~TF_DELACK; 3208 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3209 } 3210 INP_WUNLOCK(tp->t_inpcb); 3211 return; 3212 3213 dropafterack: 3214 /* 3215 * Generate an ACK dropping incoming segment if it occupies 3216 * sequence space, where the ACK reflects our state. 3217 * 3218 * We can now skip the test for the RST flag since all 3219 * paths to this code happen after packets containing 3220 * RST have been dropped. 3221 * 3222 * In the SYN-RECEIVED state, don't send an ACK unless the 3223 * segment we received passes the SYN-RECEIVED ACK test. 3224 * If it fails send a RST. This breaks the loop in the 3225 * "LAND" DoS attack, and also prevents an ACK storm 3226 * between two listening ports that have been sent forged 3227 * SYN segments, each with the source address of the other. 3228 */ 3229 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3230 (SEQ_GT(tp->snd_una, th->th_ack) || 3231 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3232 rstreason = BANDLIM_RST_OPENPORT; 3233 goto dropwithreset; 3234 } 3235 #ifdef TCPDEBUG 3236 if (so->so_options & SO_DEBUG) 3237 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3238 &tcp_savetcp, 0); 3239 #endif 3240 TCP_PROBE3(debug__input, tp, th, m); 3241 if (ti_locked == TI_RLOCKED) 3242 INP_INFO_RUNLOCK(&V_tcbinfo); 3243 ti_locked = TI_UNLOCKED; 3244 3245 tp->t_flags |= TF_ACKNOW; 3246 (void) tp->t_fb->tfb_tcp_output(tp); 3247 INP_WUNLOCK(tp->t_inpcb); 3248 m_freem(m); 3249 return; 3250 3251 dropwithreset: 3252 if (ti_locked == TI_RLOCKED) 3253 INP_INFO_RUNLOCK(&V_tcbinfo); 3254 ti_locked = TI_UNLOCKED; 3255 3256 if (tp != NULL) { 3257 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3258 INP_WUNLOCK(tp->t_inpcb); 3259 } else 3260 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3261 return; 3262 3263 drop: 3264 if (ti_locked == TI_RLOCKED) { 3265 INP_INFO_RUNLOCK(&V_tcbinfo); 3266 ti_locked = TI_UNLOCKED; 3267 } 3268 #ifdef INVARIANTS 3269 else 3270 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3271 #endif 3272 3273 /* 3274 * Drop space held by incoming segment and return. 3275 */ 3276 #ifdef TCPDEBUG 3277 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3278 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3279 &tcp_savetcp, 0); 3280 #endif 3281 TCP_PROBE3(debug__input, tp, th, m); 3282 if (tp != NULL) 3283 INP_WUNLOCK(tp->t_inpcb); 3284 m_freem(m); 3285 #ifndef TCP_RFC7413 3286 #undef tfo_syn 3287 #endif 3288 } 3289 3290 /* 3291 * Issue RST and make ACK acceptable to originator of segment. 3292 * The mbuf must still include the original packet header. 3293 * tp may be NULL. 3294 */ 3295 void 3296 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3297 int tlen, int rstreason) 3298 { 3299 #ifdef INET 3300 struct ip *ip; 3301 #endif 3302 #ifdef INET6 3303 struct ip6_hdr *ip6; 3304 #endif 3305 3306 if (tp != NULL) { 3307 INP_WLOCK_ASSERT(tp->t_inpcb); 3308 } 3309 3310 /* Don't bother if destination was broadcast/multicast. */ 3311 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3312 goto drop; 3313 #ifdef INET6 3314 if (mtod(m, struct ip *)->ip_v == 6) { 3315 ip6 = mtod(m, struct ip6_hdr *); 3316 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3317 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3318 goto drop; 3319 /* IPv6 anycast check is done at tcp6_input() */ 3320 } 3321 #endif 3322 #if defined(INET) && defined(INET6) 3323 else 3324 #endif 3325 #ifdef INET 3326 { 3327 ip = mtod(m, struct ip *); 3328 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3329 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3330 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3331 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3332 goto drop; 3333 } 3334 #endif 3335 3336 /* Perform bandwidth limiting. */ 3337 if (badport_bandlim(rstreason) < 0) 3338 goto drop; 3339 3340 /* tcp_respond consumes the mbuf chain. */ 3341 if (th->th_flags & TH_ACK) { 3342 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3343 th->th_ack, TH_RST); 3344 } else { 3345 if (th->th_flags & TH_SYN) 3346 tlen++; 3347 if (th->th_flags & TH_FIN) 3348 tlen++; 3349 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3350 (tcp_seq)0, TH_RST|TH_ACK); 3351 } 3352 return; 3353 drop: 3354 m_freem(m); 3355 } 3356 3357 /* 3358 * Parse TCP options and place in tcpopt. 3359 */ 3360 void 3361 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3362 { 3363 int opt, optlen; 3364 3365 to->to_flags = 0; 3366 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3367 opt = cp[0]; 3368 if (opt == TCPOPT_EOL) 3369 break; 3370 if (opt == TCPOPT_NOP) 3371 optlen = 1; 3372 else { 3373 if (cnt < 2) 3374 break; 3375 optlen = cp[1]; 3376 if (optlen < 2 || optlen > cnt) 3377 break; 3378 } 3379 switch (opt) { 3380 case TCPOPT_MAXSEG: 3381 if (optlen != TCPOLEN_MAXSEG) 3382 continue; 3383 if (!(flags & TO_SYN)) 3384 continue; 3385 to->to_flags |= TOF_MSS; 3386 bcopy((char *)cp + 2, 3387 (char *)&to->to_mss, sizeof(to->to_mss)); 3388 to->to_mss = ntohs(to->to_mss); 3389 break; 3390 case TCPOPT_WINDOW: 3391 if (optlen != TCPOLEN_WINDOW) 3392 continue; 3393 if (!(flags & TO_SYN)) 3394 continue; 3395 to->to_flags |= TOF_SCALE; 3396 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3397 break; 3398 case TCPOPT_TIMESTAMP: 3399 if (optlen != TCPOLEN_TIMESTAMP) 3400 continue; 3401 to->to_flags |= TOF_TS; 3402 bcopy((char *)cp + 2, 3403 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3404 to->to_tsval = ntohl(to->to_tsval); 3405 bcopy((char *)cp + 6, 3406 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3407 to->to_tsecr = ntohl(to->to_tsecr); 3408 break; 3409 case TCPOPT_SIGNATURE: 3410 /* 3411 * In order to reply to a host which has set the 3412 * TCP_SIGNATURE option in its initial SYN, we have 3413 * to record the fact that the option was observed 3414 * here for the syncache code to perform the correct 3415 * response. 3416 */ 3417 if (optlen != TCPOLEN_SIGNATURE) 3418 continue; 3419 to->to_flags |= TOF_SIGNATURE; 3420 to->to_signature = cp + 2; 3421 break; 3422 case TCPOPT_SACK_PERMITTED: 3423 if (optlen != TCPOLEN_SACK_PERMITTED) 3424 continue; 3425 if (!(flags & TO_SYN)) 3426 continue; 3427 if (!V_tcp_do_sack) 3428 continue; 3429 to->to_flags |= TOF_SACKPERM; 3430 break; 3431 case TCPOPT_SACK: 3432 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3433 continue; 3434 if (flags & TO_SYN) 3435 continue; 3436 to->to_flags |= TOF_SACK; 3437 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3438 to->to_sacks = cp + 2; 3439 TCPSTAT_INC(tcps_sack_rcv_blocks); 3440 break; 3441 #ifdef TCP_RFC7413 3442 case TCPOPT_FAST_OPEN: 3443 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3444 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3445 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3446 continue; 3447 if (!(flags & TO_SYN)) 3448 continue; 3449 if (!V_tcp_fastopen_enabled) 3450 continue; 3451 to->to_flags |= TOF_FASTOPEN; 3452 to->to_tfo_len = optlen - 2; 3453 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3454 break; 3455 #endif 3456 default: 3457 continue; 3458 } 3459 } 3460 } 3461 3462 /* 3463 * Pull out of band byte out of a segment so 3464 * it doesn't appear in the user's data queue. 3465 * It is still reflected in the segment length for 3466 * sequencing purposes. 3467 */ 3468 void 3469 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3470 int off) 3471 { 3472 int cnt = off + th->th_urp - 1; 3473 3474 while (cnt >= 0) { 3475 if (m->m_len > cnt) { 3476 char *cp = mtod(m, caddr_t) + cnt; 3477 struct tcpcb *tp = sototcpcb(so); 3478 3479 INP_WLOCK_ASSERT(tp->t_inpcb); 3480 3481 tp->t_iobc = *cp; 3482 tp->t_oobflags |= TCPOOB_HAVEDATA; 3483 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3484 m->m_len--; 3485 if (m->m_flags & M_PKTHDR) 3486 m->m_pkthdr.len--; 3487 return; 3488 } 3489 cnt -= m->m_len; 3490 m = m->m_next; 3491 if (m == NULL) 3492 break; 3493 } 3494 panic("tcp_pulloutofband"); 3495 } 3496 3497 /* 3498 * Collect new round-trip time estimate 3499 * and update averages and current timeout. 3500 */ 3501 void 3502 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3503 { 3504 int delta; 3505 3506 INP_WLOCK_ASSERT(tp->t_inpcb); 3507 3508 TCPSTAT_INC(tcps_rttupdated); 3509 tp->t_rttupdated++; 3510 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3511 /* 3512 * srtt is stored as fixed point with 5 bits after the 3513 * binary point (i.e., scaled by 8). The following magic 3514 * is equivalent to the smoothing algorithm in rfc793 with 3515 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3516 * point). Adjust rtt to origin 0. 3517 */ 3518 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3519 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3520 3521 if ((tp->t_srtt += delta) <= 0) 3522 tp->t_srtt = 1; 3523 3524 /* 3525 * We accumulate a smoothed rtt variance (actually, a 3526 * smoothed mean difference), then set the retransmit 3527 * timer to smoothed rtt + 4 times the smoothed variance. 3528 * rttvar is stored as fixed point with 4 bits after the 3529 * binary point (scaled by 16). The following is 3530 * equivalent to rfc793 smoothing with an alpha of .75 3531 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3532 * rfc793's wired-in beta. 3533 */ 3534 if (delta < 0) 3535 delta = -delta; 3536 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3537 if ((tp->t_rttvar += delta) <= 0) 3538 tp->t_rttvar = 1; 3539 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3540 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3541 } else { 3542 /* 3543 * No rtt measurement yet - use the unsmoothed rtt. 3544 * Set the variance to half the rtt (so our first 3545 * retransmit happens at 3*rtt). 3546 */ 3547 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3548 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3549 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3550 } 3551 tp->t_rtttime = 0; 3552 tp->t_rxtshift = 0; 3553 3554 /* 3555 * the retransmit should happen at rtt + 4 * rttvar. 3556 * Because of the way we do the smoothing, srtt and rttvar 3557 * will each average +1/2 tick of bias. When we compute 3558 * the retransmit timer, we want 1/2 tick of rounding and 3559 * 1 extra tick because of +-1/2 tick uncertainty in the 3560 * firing of the timer. The bias will give us exactly the 3561 * 1.5 tick we need. But, because the bias is 3562 * statistical, we have to test that we don't drop below 3563 * the minimum feasible timer (which is 2 ticks). 3564 */ 3565 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3566 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3567 3568 /* 3569 * We received an ack for a packet that wasn't retransmitted; 3570 * it is probably safe to discard any error indications we've 3571 * received recently. This isn't quite right, but close enough 3572 * for now (a route might have failed after we sent a segment, 3573 * and the return path might not be symmetrical). 3574 */ 3575 tp->t_softerror = 0; 3576 } 3577 3578 /* 3579 * Determine a reasonable value for maxseg size. 3580 * If the route is known, check route for mtu. 3581 * If none, use an mss that can be handled on the outgoing interface 3582 * without forcing IP to fragment. If no route is found, route has no mtu, 3583 * or the destination isn't local, use a default, hopefully conservative 3584 * size (usually 512 or the default IP max size, but no more than the mtu 3585 * of the interface), as we can't discover anything about intervening 3586 * gateways or networks. We also initialize the congestion/slow start 3587 * window to be a single segment if the destination isn't local. 3588 * While looking at the routing entry, we also initialize other path-dependent 3589 * parameters from pre-set or cached values in the routing entry. 3590 * 3591 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3592 * IP options, e.g. IPSEC data, since length of this data may vary, and 3593 * thus it is calculated for every segment separately in tcp_output(). 3594 * 3595 * NOTE that this routine is only called when we process an incoming 3596 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3597 * settings are handled in tcp_mssopt(). 3598 */ 3599 void 3600 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3601 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3602 { 3603 int mss = 0; 3604 uint32_t maxmtu = 0; 3605 struct inpcb *inp = tp->t_inpcb; 3606 struct hc_metrics_lite metrics; 3607 #ifdef INET6 3608 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3609 size_t min_protoh = isipv6 ? 3610 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3611 sizeof (struct tcpiphdr); 3612 #else 3613 const size_t min_protoh = sizeof(struct tcpiphdr); 3614 #endif 3615 3616 INP_WLOCK_ASSERT(tp->t_inpcb); 3617 3618 if (mtuoffer != -1) { 3619 KASSERT(offer == -1, ("%s: conflict", __func__)); 3620 offer = mtuoffer - min_protoh; 3621 } 3622 3623 /* Initialize. */ 3624 #ifdef INET6 3625 if (isipv6) { 3626 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3627 tp->t_maxseg = V_tcp_v6mssdflt; 3628 } 3629 #endif 3630 #if defined(INET) && defined(INET6) 3631 else 3632 #endif 3633 #ifdef INET 3634 { 3635 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3636 tp->t_maxseg = V_tcp_mssdflt; 3637 } 3638 #endif 3639 3640 /* 3641 * No route to sender, stay with default mss and return. 3642 */ 3643 if (maxmtu == 0) { 3644 /* 3645 * In case we return early we need to initialize metrics 3646 * to a defined state as tcp_hc_get() would do for us 3647 * if there was no cache hit. 3648 */ 3649 if (metricptr != NULL) 3650 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3651 return; 3652 } 3653 3654 /* What have we got? */ 3655 switch (offer) { 3656 case 0: 3657 /* 3658 * Offer == 0 means that there was no MSS on the SYN 3659 * segment, in this case we use tcp_mssdflt as 3660 * already assigned to t_maxseg above. 3661 */ 3662 offer = tp->t_maxseg; 3663 break; 3664 3665 case -1: 3666 /* 3667 * Offer == -1 means that we didn't receive SYN yet. 3668 */ 3669 /* FALLTHROUGH */ 3670 3671 default: 3672 /* 3673 * Prevent DoS attack with too small MSS. Round up 3674 * to at least minmss. 3675 */ 3676 offer = max(offer, V_tcp_minmss); 3677 } 3678 3679 /* 3680 * rmx information is now retrieved from tcp_hostcache. 3681 */ 3682 tcp_hc_get(&inp->inp_inc, &metrics); 3683 if (metricptr != NULL) 3684 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3685 3686 /* 3687 * If there's a discovered mtu in tcp hostcache, use it. 3688 * Else, use the link mtu. 3689 */ 3690 if (metrics.rmx_mtu) 3691 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3692 else { 3693 #ifdef INET6 3694 if (isipv6) { 3695 mss = maxmtu - min_protoh; 3696 if (!V_path_mtu_discovery && 3697 !in6_localaddr(&inp->in6p_faddr)) 3698 mss = min(mss, V_tcp_v6mssdflt); 3699 } 3700 #endif 3701 #if defined(INET) && defined(INET6) 3702 else 3703 #endif 3704 #ifdef INET 3705 { 3706 mss = maxmtu - min_protoh; 3707 if (!V_path_mtu_discovery && 3708 !in_localaddr(inp->inp_faddr)) 3709 mss = min(mss, V_tcp_mssdflt); 3710 } 3711 #endif 3712 /* 3713 * XXX - The above conditional (mss = maxmtu - min_protoh) 3714 * probably violates the TCP spec. 3715 * The problem is that, since we don't know the 3716 * other end's MSS, we are supposed to use a conservative 3717 * default. But, if we do that, then MTU discovery will 3718 * never actually take place, because the conservative 3719 * default is much less than the MTUs typically seen 3720 * on the Internet today. For the moment, we'll sweep 3721 * this under the carpet. 3722 * 3723 * The conservative default might not actually be a problem 3724 * if the only case this occurs is when sending an initial 3725 * SYN with options and data to a host we've never talked 3726 * to before. Then, they will reply with an MSS value which 3727 * will get recorded and the new parameters should get 3728 * recomputed. For Further Study. 3729 */ 3730 } 3731 mss = min(mss, offer); 3732 3733 /* 3734 * Sanity check: make sure that maxseg will be large 3735 * enough to allow some data on segments even if the 3736 * all the option space is used (40bytes). Otherwise 3737 * funny things may happen in tcp_output. 3738 * 3739 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3740 */ 3741 mss = max(mss, 64); 3742 3743 tp->t_maxseg = mss; 3744 } 3745 3746 void 3747 tcp_mss(struct tcpcb *tp, int offer) 3748 { 3749 int mss; 3750 uint32_t bufsize; 3751 struct inpcb *inp; 3752 struct socket *so; 3753 struct hc_metrics_lite metrics; 3754 struct tcp_ifcap cap; 3755 3756 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3757 3758 bzero(&cap, sizeof(cap)); 3759 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3760 3761 mss = tp->t_maxseg; 3762 inp = tp->t_inpcb; 3763 3764 /* 3765 * If there's a pipesize, change the socket buffer to that size, 3766 * don't change if sb_hiwat is different than default (then it 3767 * has been changed on purpose with setsockopt). 3768 * Make the socket buffers an integral number of mss units; 3769 * if the mss is larger than the socket buffer, decrease the mss. 3770 */ 3771 so = inp->inp_socket; 3772 SOCKBUF_LOCK(&so->so_snd); 3773 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3774 bufsize = metrics.rmx_sendpipe; 3775 else 3776 bufsize = so->so_snd.sb_hiwat; 3777 if (bufsize < mss) 3778 mss = bufsize; 3779 else { 3780 bufsize = roundup(bufsize, mss); 3781 if (bufsize > sb_max) 3782 bufsize = sb_max; 3783 if (bufsize > so->so_snd.sb_hiwat) 3784 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3785 } 3786 SOCKBUF_UNLOCK(&so->so_snd); 3787 /* 3788 * Sanity check: make sure that maxseg will be large 3789 * enough to allow some data on segments even if the 3790 * all the option space is used (40bytes). Otherwise 3791 * funny things may happen in tcp_output. 3792 * 3793 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3794 */ 3795 tp->t_maxseg = max(mss, 64); 3796 3797 SOCKBUF_LOCK(&so->so_rcv); 3798 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3799 bufsize = metrics.rmx_recvpipe; 3800 else 3801 bufsize = so->so_rcv.sb_hiwat; 3802 if (bufsize > mss) { 3803 bufsize = roundup(bufsize, mss); 3804 if (bufsize > sb_max) 3805 bufsize = sb_max; 3806 if (bufsize > so->so_rcv.sb_hiwat) 3807 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3808 } 3809 SOCKBUF_UNLOCK(&so->so_rcv); 3810 3811 /* Check the interface for TSO capabilities. */ 3812 if (cap.ifcap & CSUM_TSO) { 3813 tp->t_flags |= TF_TSO; 3814 tp->t_tsomax = cap.tsomax; 3815 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3816 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3817 } 3818 } 3819 3820 /* 3821 * Determine the MSS option to send on an outgoing SYN. 3822 */ 3823 int 3824 tcp_mssopt(struct in_conninfo *inc) 3825 { 3826 int mss = 0; 3827 uint32_t thcmtu = 0; 3828 uint32_t maxmtu = 0; 3829 size_t min_protoh; 3830 3831 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3832 3833 #ifdef INET6 3834 if (inc->inc_flags & INC_ISIPV6) { 3835 mss = V_tcp_v6mssdflt; 3836 maxmtu = tcp_maxmtu6(inc, NULL); 3837 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3838 } 3839 #endif 3840 #if defined(INET) && defined(INET6) 3841 else 3842 #endif 3843 #ifdef INET 3844 { 3845 mss = V_tcp_mssdflt; 3846 maxmtu = tcp_maxmtu(inc, NULL); 3847 min_protoh = sizeof(struct tcpiphdr); 3848 } 3849 #endif 3850 #if defined(INET6) || defined(INET) 3851 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3852 #endif 3853 3854 if (maxmtu && thcmtu) 3855 mss = min(maxmtu, thcmtu) - min_protoh; 3856 else if (maxmtu || thcmtu) 3857 mss = max(maxmtu, thcmtu) - min_protoh; 3858 3859 return (mss); 3860 } 3861 3862 3863 /* 3864 * On a partial ack arrives, force the retransmission of the 3865 * next unacknowledged segment. Do not clear tp->t_dupacks. 3866 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3867 * be started again. 3868 */ 3869 void 3870 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3871 { 3872 tcp_seq onxt = tp->snd_nxt; 3873 uint32_t ocwnd = tp->snd_cwnd; 3874 u_int maxseg = tcp_maxseg(tp); 3875 3876 INP_WLOCK_ASSERT(tp->t_inpcb); 3877 3878 tcp_timer_activate(tp, TT_REXMT, 0); 3879 tp->t_rtttime = 0; 3880 tp->snd_nxt = th->th_ack; 3881 /* 3882 * Set snd_cwnd to one segment beyond acknowledged offset. 3883 * (tp->snd_una has not yet been updated when this function is called.) 3884 */ 3885 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3886 tp->t_flags |= TF_ACKNOW; 3887 (void) tp->t_fb->tfb_tcp_output(tp); 3888 tp->snd_cwnd = ocwnd; 3889 if (SEQ_GT(onxt, tp->snd_nxt)) 3890 tp->snd_nxt = onxt; 3891 /* 3892 * Partial window deflation. Relies on fact that tp->snd_una 3893 * not updated yet. 3894 */ 3895 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3896 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3897 else 3898 tp->snd_cwnd = 0; 3899 tp->snd_cwnd += maxseg; 3900 } 3901 3902 int 3903 tcp_compute_pipe(struct tcpcb *tp) 3904 { 3905 return (tp->snd_max - tp->snd_una + 3906 tp->sackhint.sack_bytes_rexmit - 3907 tp->sackhint.sacked_bytes); 3908 } 3909