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 * 4. 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_ipfw.h" /* for ipfw_fwd */ 54 #include "opt_inet.h" 55 #include "opt_inet6.h" 56 #include "opt_ipsec.h" 57 #include "opt_tcpdebug.h" 58 59 #include <sys/param.h> 60 #include <sys/kernel.h> 61 #include <sys/hhook.h> 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/sdt.h> 67 #include <sys/signalvar.h> 68 #include <sys/socket.h> 69 #include <sys/socketvar.h> 70 #include <sys/sysctl.h> 71 #include <sys/syslog.h> 72 #include <sys/systm.h> 73 74 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 75 76 #include <vm/uma.h> 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/route.h> 81 #include <net/vnet.h> 82 83 #define TCPSTATES /* for logging */ 84 85 #include <netinet/in.h> 86 #include <netinet/in_kdtrace.h> 87 #include <netinet/in_pcb.h> 88 #include <netinet/in_systm.h> 89 #include <netinet/ip.h> 90 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 91 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 92 #include <netinet/ip_var.h> 93 #include <netinet/ip_options.h> 94 #include <netinet/ip6.h> 95 #include <netinet/icmp6.h> 96 #include <netinet6/in6_pcb.h> 97 #include <netinet6/in6_var.h> 98 #include <netinet6/ip6_var.h> 99 #include <netinet6/nd6.h> 100 #ifdef TCP_RFC7413 101 #include <netinet/tcp_fastopen.h> 102 #endif 103 #include <netinet/tcp.h> 104 #include <netinet/tcp_fsm.h> 105 #include <netinet/tcp_seq.h> 106 #include <netinet/tcp_timer.h> 107 #include <netinet/tcp_var.h> 108 #include <netinet6/tcp6_var.h> 109 #include <netinet/tcpip.h> 110 #include <netinet/cc/cc.h> 111 #ifdef TCPPCAP 112 #include <netinet/tcp_pcap.h> 113 #endif 114 #include <netinet/tcp_syncache.h> 115 #ifdef TCPDEBUG 116 #include <netinet/tcp_debug.h> 117 #endif /* TCPDEBUG */ 118 #ifdef TCP_OFFLOAD 119 #include <netinet/tcp_offload.h> 120 #endif 121 122 #ifdef IPSEC 123 #include <netipsec/ipsec.h> 124 #include <netipsec/ipsec6.h> 125 #endif /*IPSEC*/ 126 127 #include <machine/in_cksum.h> 128 129 #include <security/mac/mac_framework.h> 130 131 const int tcprexmtthresh = 3; 132 133 int tcp_log_in_vain = 0; 134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 135 &tcp_log_in_vain, 0, 136 "Log all incoming TCP segments to closed ports"); 137 138 VNET_DEFINE(int, blackhole) = 0; 139 #define V_blackhole VNET(blackhole) 140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 141 &VNET_NAME(blackhole), 0, 142 "Do not send RST on segments to closed ports"); 143 144 VNET_DEFINE(int, tcp_delack_enabled) = 1; 145 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 146 &VNET_NAME(tcp_delack_enabled), 0, 147 "Delay ACK to try and piggyback it onto a data packet"); 148 149 VNET_DEFINE(int, drop_synfin) = 0; 150 #define V_drop_synfin VNET(drop_synfin) 151 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 152 &VNET_NAME(drop_synfin), 0, 153 "Drop TCP packets with SYN+FIN set"); 154 155 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 157 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 158 "Use calculated pipe/in-flight bytes per RFC 6675"); 159 160 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 161 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 163 &VNET_NAME(tcp_do_rfc3042), 0, 164 "Enable RFC 3042 (Limited Transmit)"); 165 166 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 168 &VNET_NAME(tcp_do_rfc3390), 0, 169 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 170 171 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 173 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 174 "Slow-start flight size (initial congestion window) in number of segments"); 175 176 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 178 &VNET_NAME(tcp_do_rfc3465), 0, 179 "Enable RFC 3465 (Appropriate Byte Counting)"); 180 181 VNET_DEFINE(int, tcp_abc_l_var) = 2; 182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 183 &VNET_NAME(tcp_abc_l_var), 2, 184 "Cap the max cwnd increment during slow-start to this number of segments"); 185 186 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 187 188 VNET_DEFINE(int, tcp_do_ecn) = 0; 189 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 190 &VNET_NAME(tcp_do_ecn), 0, 191 "TCP ECN support"); 192 193 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 194 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 195 &VNET_NAME(tcp_ecn_maxretries), 0, 196 "Max retries before giving up on ECN"); 197 198 VNET_DEFINE(int, tcp_insecure_syn) = 0; 199 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 201 &VNET_NAME(tcp_insecure_syn), 0, 202 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 203 204 VNET_DEFINE(int, tcp_insecure_rst) = 0; 205 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(tcp_insecure_rst), 0, 208 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 209 210 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 211 #define V_tcp_recvspace VNET(tcp_recvspace) 212 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 213 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 214 215 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 216 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 217 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 218 &VNET_NAME(tcp_do_autorcvbuf), 0, 219 "Enable automatic receive buffer sizing"); 220 221 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 222 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 223 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 224 &VNET_NAME(tcp_autorcvbuf_inc), 0, 225 "Incrementor step size of automatic receive buffer"); 226 227 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 228 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 229 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 230 &VNET_NAME(tcp_autorcvbuf_max), 0, 231 "Max size of automatic receive buffer"); 232 233 VNET_DEFINE(struct inpcbhead, tcb); 234 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 235 VNET_DEFINE(struct inpcbinfo, tcbinfo); 236 237 /* 238 * TCP statistics are stored in an array of counter(9)s, which size matches 239 * size of struct tcpstat. TCP running connection count is a regular array. 240 */ 241 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 242 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 243 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 244 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 245 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 246 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 247 "TCP connection counts by TCP state"); 248 249 static void 250 tcp_vnet_init(const void *unused) 251 { 252 253 COUNTER_ARRAY_ALLOC(VNET(tcps_states), TCP_NSTATES, M_WAITOK); 254 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 255 } 256 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 257 tcp_vnet_init, NULL); 258 259 #ifdef VIMAGE 260 static void 261 tcp_vnet_uninit(const void *unused) 262 { 263 264 COUNTER_ARRAY_FREE(VNET(tcps_states), TCP_NSTATES); 265 VNET_PCPUSTAT_FREE(tcpstat); 266 } 267 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 268 tcp_vnet_uninit, NULL); 269 #endif /* VIMAGE */ 270 271 /* 272 * Kernel module interface for updating tcpstat. The argument is an index 273 * into tcpstat treated as an array. 274 */ 275 void 276 kmod_tcpstat_inc(int statnum) 277 { 278 279 counter_u64_add(VNET(tcpstat)[statnum], 1); 280 } 281 282 /* 283 * Wrapper for the TCP established input helper hook. 284 */ 285 void 286 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 287 { 288 struct tcp_hhook_data hhook_data; 289 290 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 291 hhook_data.tp = tp; 292 hhook_data.th = th; 293 hhook_data.to = to; 294 295 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 296 tp->osd); 297 } 298 } 299 300 /* 301 * CC wrapper hook functions 302 */ 303 void 304 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 305 { 306 INP_WLOCK_ASSERT(tp->t_inpcb); 307 308 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 309 if (tp->snd_cwnd <= tp->snd_wnd) 310 tp->ccv->flags |= CCF_CWND_LIMITED; 311 else 312 tp->ccv->flags &= ~CCF_CWND_LIMITED; 313 314 if (type == CC_ACK) { 315 if (tp->snd_cwnd > tp->snd_ssthresh) { 316 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 317 V_tcp_abc_l_var * tcp_maxseg(tp)); 318 if (tp->t_bytes_acked >= tp->snd_cwnd) { 319 tp->t_bytes_acked -= tp->snd_cwnd; 320 tp->ccv->flags |= CCF_ABC_SENTAWND; 321 } 322 } else { 323 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 324 tp->t_bytes_acked = 0; 325 } 326 } 327 328 if (CC_ALGO(tp)->ack_received != NULL) { 329 /* XXXLAS: Find a way to live without this */ 330 tp->ccv->curack = th->th_ack; 331 CC_ALGO(tp)->ack_received(tp->ccv, type); 332 } 333 } 334 335 void 336 cc_conn_init(struct tcpcb *tp) 337 { 338 struct hc_metrics_lite metrics; 339 struct inpcb *inp = tp->t_inpcb; 340 u_int maxseg; 341 int rtt; 342 343 INP_WLOCK_ASSERT(tp->t_inpcb); 344 345 tcp_hc_get(&inp->inp_inc, &metrics); 346 maxseg = tcp_maxseg(tp); 347 348 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 349 tp->t_srtt = rtt; 350 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 351 TCPSTAT_INC(tcps_usedrtt); 352 if (metrics.rmx_rttvar) { 353 tp->t_rttvar = metrics.rmx_rttvar; 354 TCPSTAT_INC(tcps_usedrttvar); 355 } else { 356 /* default variation is +- 1 rtt */ 357 tp->t_rttvar = 358 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 359 } 360 TCPT_RANGESET(tp->t_rxtcur, 361 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 362 tp->t_rttmin, TCPTV_REXMTMAX); 363 } 364 if (metrics.rmx_ssthresh) { 365 /* 366 * There's some sort of gateway or interface 367 * buffer limit on the path. Use this to set 368 * the slow start threshhold, but set the 369 * threshold to no less than 2*mss. 370 */ 371 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 372 TCPSTAT_INC(tcps_usedssthresh); 373 } 374 375 /* 376 * Set the initial slow-start flight size. 377 * 378 * RFC5681 Section 3.1 specifies the default conservative values. 379 * RFC3390 specifies slightly more aggressive values. 380 * RFC6928 increases it to ten segments. 381 * Support for user specified value for initial flight size. 382 * 383 * If a SYN or SYN/ACK was lost and retransmitted, we have to 384 * reduce the initial CWND to one segment as congestion is likely 385 * requiring us to be cautious. 386 */ 387 if (tp->snd_cwnd == 1) 388 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 389 else if (V_tcp_initcwnd_segments) 390 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 391 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 392 else if (V_tcp_do_rfc3390) 393 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 394 else { 395 /* Per RFC5681 Section 3.1 */ 396 if (maxseg > 2190) 397 tp->snd_cwnd = 2 * maxseg; 398 else if (maxseg > 1095) 399 tp->snd_cwnd = 3 * maxseg; 400 else 401 tp->snd_cwnd = 4 * maxseg; 402 } 403 404 if (CC_ALGO(tp)->conn_init != NULL) 405 CC_ALGO(tp)->conn_init(tp->ccv); 406 } 407 408 void inline 409 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 410 { 411 u_int maxseg; 412 413 INP_WLOCK_ASSERT(tp->t_inpcb); 414 415 switch(type) { 416 case CC_NDUPACK: 417 if (!IN_FASTRECOVERY(tp->t_flags)) { 418 tp->snd_recover = tp->snd_max; 419 if (tp->t_flags & TF_ECN_PERMIT) 420 tp->t_flags |= TF_ECN_SND_CWR; 421 } 422 break; 423 case CC_ECN: 424 if (!IN_CONGRECOVERY(tp->t_flags)) { 425 TCPSTAT_INC(tcps_ecn_rcwnd); 426 tp->snd_recover = tp->snd_max; 427 if (tp->t_flags & TF_ECN_PERMIT) 428 tp->t_flags |= TF_ECN_SND_CWR; 429 } 430 break; 431 case CC_RTO: 432 maxseg = tcp_maxseg(tp); 433 tp->t_dupacks = 0; 434 tp->t_bytes_acked = 0; 435 EXIT_RECOVERY(tp->t_flags); 436 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 437 maxseg) * maxseg; 438 tp->snd_cwnd = maxseg; 439 break; 440 case CC_RTO_ERR: 441 TCPSTAT_INC(tcps_sndrexmitbad); 442 /* RTO was unnecessary, so reset everything. */ 443 tp->snd_cwnd = tp->snd_cwnd_prev; 444 tp->snd_ssthresh = tp->snd_ssthresh_prev; 445 tp->snd_recover = tp->snd_recover_prev; 446 if (tp->t_flags & TF_WASFRECOVERY) 447 ENTER_FASTRECOVERY(tp->t_flags); 448 if (tp->t_flags & TF_WASCRECOVERY) 449 ENTER_CONGRECOVERY(tp->t_flags); 450 tp->snd_nxt = tp->snd_max; 451 tp->t_flags &= ~TF_PREVVALID; 452 tp->t_badrxtwin = 0; 453 break; 454 } 455 456 if (CC_ALGO(tp)->cong_signal != NULL) { 457 if (th != NULL) 458 tp->ccv->curack = th->th_ack; 459 CC_ALGO(tp)->cong_signal(tp->ccv, type); 460 } 461 } 462 463 void inline 464 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 465 { 466 INP_WLOCK_ASSERT(tp->t_inpcb); 467 468 /* XXXLAS: KASSERT that we're in recovery? */ 469 470 if (CC_ALGO(tp)->post_recovery != NULL) { 471 tp->ccv->curack = th->th_ack; 472 CC_ALGO(tp)->post_recovery(tp->ccv); 473 } 474 /* XXXLAS: EXIT_RECOVERY ? */ 475 tp->t_bytes_acked = 0; 476 } 477 478 #ifdef TCP_SIGNATURE 479 static inline int 480 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 481 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 482 { 483 int ret; 484 485 tcp_fields_to_net(th); 486 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 487 tcp_fields_to_host(th); 488 return (ret); 489 } 490 #endif 491 492 /* 493 * Indicate whether this ack should be delayed. We can delay the ack if 494 * following conditions are met: 495 * - There is no delayed ack timer in progress. 496 * - Our last ack wasn't a 0-sized window. We never want to delay 497 * the ack that opens up a 0-sized window. 498 * - LRO wasn't used for this segment. We make sure by checking that the 499 * segment size is not larger than the MSS. 500 */ 501 #define DELAY_ACK(tp, tlen) \ 502 ((!tcp_timer_active(tp, TT_DELACK) && \ 503 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 504 (tlen <= tp->t_maxseg) && \ 505 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 506 507 static void inline 508 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 509 { 510 INP_WLOCK_ASSERT(tp->t_inpcb); 511 512 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 513 switch (iptos & IPTOS_ECN_MASK) { 514 case IPTOS_ECN_CE: 515 tp->ccv->flags |= CCF_IPHDR_CE; 516 break; 517 case IPTOS_ECN_ECT0: 518 tp->ccv->flags &= ~CCF_IPHDR_CE; 519 break; 520 case IPTOS_ECN_ECT1: 521 tp->ccv->flags &= ~CCF_IPHDR_CE; 522 break; 523 } 524 525 if (th->th_flags & TH_CWR) 526 tp->ccv->flags |= CCF_TCPHDR_CWR; 527 else 528 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 529 530 if (tp->t_flags & TF_DELACK) 531 tp->ccv->flags |= CCF_DELACK; 532 else 533 tp->ccv->flags &= ~CCF_DELACK; 534 535 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 536 537 if (tp->ccv->flags & CCF_ACKNOW) 538 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 539 } 540 } 541 542 /* 543 * TCP input handling is split into multiple parts: 544 * tcp6_input is a thin wrapper around tcp_input for the extended 545 * ip6_protox[] call format in ip6_input 546 * tcp_input handles primary segment validation, inpcb lookup and 547 * SYN processing on listen sockets 548 * tcp_do_segment processes the ACK and text of the segment for 549 * establishing, established and closing connections 550 */ 551 #ifdef INET6 552 int 553 tcp6_input(struct mbuf **mp, int *offp, int proto) 554 { 555 struct mbuf *m = *mp; 556 struct in6_ifaddr *ia6; 557 struct ip6_hdr *ip6; 558 559 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 560 561 /* 562 * draft-itojun-ipv6-tcp-to-anycast 563 * better place to put this in? 564 */ 565 ip6 = mtod(m, struct ip6_hdr *); 566 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 567 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 568 struct ip6_hdr *ip6; 569 570 ifa_free(&ia6->ia_ifa); 571 ip6 = mtod(m, struct ip6_hdr *); 572 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 573 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 574 return (IPPROTO_DONE); 575 } 576 if (ia6) 577 ifa_free(&ia6->ia_ifa); 578 579 return (tcp_input(mp, offp, proto)); 580 } 581 #endif /* INET6 */ 582 583 int 584 tcp_input(struct mbuf **mp, int *offp, int proto) 585 { 586 struct mbuf *m = *mp; 587 struct tcphdr *th = NULL; 588 struct ip *ip = NULL; 589 struct inpcb *inp = NULL; 590 struct tcpcb *tp = NULL; 591 struct socket *so = NULL; 592 u_char *optp = NULL; 593 int off0; 594 int optlen = 0; 595 #ifdef INET 596 int len; 597 #endif 598 int tlen = 0, off; 599 int drop_hdrlen; 600 int thflags; 601 int rstreason = 0; /* For badport_bandlim accounting purposes */ 602 #ifdef TCP_SIGNATURE 603 uint8_t sig_checked = 0; 604 #endif 605 uint8_t iptos = 0; 606 struct m_tag *fwd_tag = NULL; 607 #ifdef INET6 608 struct ip6_hdr *ip6 = NULL; 609 int isipv6; 610 #else 611 const void *ip6 = NULL; 612 #endif /* INET6 */ 613 struct tcpopt to; /* options in this segment */ 614 char *s = NULL; /* address and port logging */ 615 int ti_locked; 616 #ifdef TCPDEBUG 617 /* 618 * The size of tcp_saveipgen must be the size of the max ip header, 619 * now IPv6. 620 */ 621 u_char tcp_saveipgen[IP6_HDR_LEN]; 622 struct tcphdr tcp_savetcp; 623 short ostate = 0; 624 #endif 625 626 #ifdef INET6 627 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 628 #endif 629 630 off0 = *offp; 631 m = *mp; 632 *mp = NULL; 633 to.to_flags = 0; 634 TCPSTAT_INC(tcps_rcvtotal); 635 636 #ifdef INET6 637 if (isipv6) { 638 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 639 640 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 641 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 642 if (m == NULL) { 643 TCPSTAT_INC(tcps_rcvshort); 644 return (IPPROTO_DONE); 645 } 646 } 647 648 ip6 = mtod(m, struct ip6_hdr *); 649 th = (struct tcphdr *)((caddr_t)ip6 + off0); 650 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 651 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 652 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 653 th->th_sum = m->m_pkthdr.csum_data; 654 else 655 th->th_sum = in6_cksum_pseudo(ip6, tlen, 656 IPPROTO_TCP, m->m_pkthdr.csum_data); 657 th->th_sum ^= 0xffff; 658 } else 659 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 660 if (th->th_sum) { 661 TCPSTAT_INC(tcps_rcvbadsum); 662 goto drop; 663 } 664 665 /* 666 * Be proactive about unspecified IPv6 address in source. 667 * As we use all-zero to indicate unbounded/unconnected pcb, 668 * unspecified IPv6 address can be used to confuse us. 669 * 670 * Note that packets with unspecified IPv6 destination is 671 * already dropped in ip6_input. 672 */ 673 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 674 /* XXX stat */ 675 goto drop; 676 } 677 } 678 #endif 679 #if defined(INET) && defined(INET6) 680 else 681 #endif 682 #ifdef INET 683 { 684 /* 685 * Get IP and TCP header together in first mbuf. 686 * Note: IP leaves IP header in first mbuf. 687 */ 688 if (off0 > sizeof (struct ip)) { 689 ip_stripoptions(m); 690 off0 = sizeof(struct ip); 691 } 692 if (m->m_len < sizeof (struct tcpiphdr)) { 693 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 694 == NULL) { 695 TCPSTAT_INC(tcps_rcvshort); 696 return (IPPROTO_DONE); 697 } 698 } 699 ip = mtod(m, struct ip *); 700 th = (struct tcphdr *)((caddr_t)ip + off0); 701 tlen = ntohs(ip->ip_len) - off0; 702 703 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 704 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 705 th->th_sum = m->m_pkthdr.csum_data; 706 else 707 th->th_sum = in_pseudo(ip->ip_src.s_addr, 708 ip->ip_dst.s_addr, 709 htonl(m->m_pkthdr.csum_data + tlen + 710 IPPROTO_TCP)); 711 th->th_sum ^= 0xffff; 712 } else { 713 struct ipovly *ipov = (struct ipovly *)ip; 714 715 /* 716 * Checksum extended TCP header and data. 717 */ 718 len = off0 + tlen; 719 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 720 ipov->ih_len = htons(tlen); 721 th->th_sum = in_cksum(m, len); 722 /* Reset length for SDT probes. */ 723 ip->ip_len = htons(tlen + off0); 724 } 725 726 if (th->th_sum) { 727 TCPSTAT_INC(tcps_rcvbadsum); 728 goto drop; 729 } 730 /* Re-initialization for later version check */ 731 ip->ip_v = IPVERSION; 732 } 733 #endif /* INET */ 734 735 #ifdef INET6 736 if (isipv6) 737 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 738 #endif 739 #if defined(INET) && defined(INET6) 740 else 741 #endif 742 #ifdef INET 743 iptos = ip->ip_tos; 744 #endif 745 746 /* 747 * Check that TCP offset makes sense, 748 * pull out TCP options and adjust length. XXX 749 */ 750 off = th->th_off << 2; 751 if (off < sizeof (struct tcphdr) || off > tlen) { 752 TCPSTAT_INC(tcps_rcvbadoff); 753 goto drop; 754 } 755 tlen -= off; /* tlen is used instead of ti->ti_len */ 756 if (off > sizeof (struct tcphdr)) { 757 #ifdef INET6 758 if (isipv6) { 759 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 760 ip6 = mtod(m, struct ip6_hdr *); 761 th = (struct tcphdr *)((caddr_t)ip6 + off0); 762 } 763 #endif 764 #if defined(INET) && defined(INET6) 765 else 766 #endif 767 #ifdef INET 768 { 769 if (m->m_len < sizeof(struct ip) + off) { 770 if ((m = m_pullup(m, sizeof (struct ip) + off)) 771 == NULL) { 772 TCPSTAT_INC(tcps_rcvshort); 773 return (IPPROTO_DONE); 774 } 775 ip = mtod(m, struct ip *); 776 th = (struct tcphdr *)((caddr_t)ip + off0); 777 } 778 } 779 #endif 780 optlen = off - sizeof (struct tcphdr); 781 optp = (u_char *)(th + 1); 782 } 783 thflags = th->th_flags; 784 785 /* 786 * Convert TCP protocol specific fields to host format. 787 */ 788 tcp_fields_to_host(th); 789 790 /* 791 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 792 */ 793 drop_hdrlen = off0 + off; 794 795 /* 796 * Locate pcb for segment; if we're likely to add or remove a 797 * connection then first acquire pcbinfo lock. There are three cases 798 * where we might discover later we need a write lock despite the 799 * flags: ACKs moving a connection out of the syncache, ACKs for a 800 * connection in TIMEWAIT and SYNs not targeting a listening socket. 801 */ 802 if ((thflags & (TH_FIN | TH_RST)) != 0) { 803 INP_INFO_RLOCK(&V_tcbinfo); 804 ti_locked = TI_RLOCKED; 805 } else 806 ti_locked = TI_UNLOCKED; 807 808 /* 809 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 810 */ 811 if ( 812 #ifdef INET6 813 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 814 #ifdef INET 815 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 816 #endif 817 #endif 818 #if defined(INET) && !defined(INET6) 819 (m->m_flags & M_IP_NEXTHOP) 820 #endif 821 ) 822 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 823 824 findpcb: 825 #ifdef INVARIANTS 826 if (ti_locked == TI_RLOCKED) { 827 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 828 } else { 829 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 830 } 831 #endif 832 #ifdef INET6 833 if (isipv6 && fwd_tag != NULL) { 834 struct sockaddr_in6 *next_hop6; 835 836 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 837 /* 838 * Transparently forwarded. Pretend to be the destination. 839 * Already got one like this? 840 */ 841 inp = in6_pcblookup_mbuf(&V_tcbinfo, 842 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 843 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 844 if (!inp) { 845 /* 846 * It's new. Try to find the ambushing socket. 847 * Because we've rewritten the destination address, 848 * any hardware-generated hash is ignored. 849 */ 850 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 851 th->th_sport, &next_hop6->sin6_addr, 852 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 853 th->th_dport, INPLOOKUP_WILDCARD | 854 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 855 } 856 } else if (isipv6) { 857 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 858 th->th_sport, &ip6->ip6_dst, th->th_dport, 859 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 860 m->m_pkthdr.rcvif, m); 861 } 862 #endif /* INET6 */ 863 #if defined(INET6) && defined(INET) 864 else 865 #endif 866 #ifdef INET 867 if (fwd_tag != NULL) { 868 struct sockaddr_in *next_hop; 869 870 next_hop = (struct sockaddr_in *)(fwd_tag+1); 871 /* 872 * Transparently forwarded. Pretend to be the destination. 873 * already got one like this? 874 */ 875 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 876 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 877 m->m_pkthdr.rcvif, m); 878 if (!inp) { 879 /* 880 * It's new. Try to find the ambushing socket. 881 * Because we've rewritten the destination address, 882 * any hardware-generated hash is ignored. 883 */ 884 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 885 th->th_sport, next_hop->sin_addr, 886 next_hop->sin_port ? ntohs(next_hop->sin_port) : 887 th->th_dport, INPLOOKUP_WILDCARD | 888 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 889 } 890 } else 891 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 892 th->th_sport, ip->ip_dst, th->th_dport, 893 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 894 m->m_pkthdr.rcvif, m); 895 #endif /* INET */ 896 897 /* 898 * If the INPCB does not exist then all data in the incoming 899 * segment is discarded and an appropriate RST is sent back. 900 * XXX MRT Send RST using which routing table? 901 */ 902 if (inp == NULL) { 903 /* 904 * Log communication attempts to ports that are not 905 * in use. 906 */ 907 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 908 tcp_log_in_vain == 2) { 909 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 910 log(LOG_INFO, "%s; %s: Connection attempt " 911 "to closed port\n", s, __func__); 912 } 913 /* 914 * When blackholing do not respond with a RST but 915 * completely ignore the segment and drop it. 916 */ 917 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 918 V_blackhole == 2) 919 goto dropunlock; 920 921 rstreason = BANDLIM_RST_CLOSEDPORT; 922 goto dropwithreset; 923 } 924 INP_WLOCK_ASSERT(inp); 925 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 926 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 927 ((inp->inp_socket == NULL) || 928 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 929 inp->inp_flowid = m->m_pkthdr.flowid; 930 inp->inp_flowtype = M_HASHTYPE_GET(m); 931 } 932 #ifdef IPSEC 933 #ifdef INET6 934 if (isipv6 && ipsec6_in_reject(m, inp)) { 935 goto dropunlock; 936 } else 937 #endif /* INET6 */ 938 if (ipsec4_in_reject(m, inp) != 0) { 939 goto dropunlock; 940 } 941 #endif /* IPSEC */ 942 943 /* 944 * Check the minimum TTL for socket. 945 */ 946 if (inp->inp_ip_minttl != 0) { 947 #ifdef INET6 948 if (isipv6) { 949 if (inp->inp_ip_minttl > ip6->ip6_hlim) 950 goto dropunlock; 951 } else 952 #endif 953 if (inp->inp_ip_minttl > ip->ip_ttl) 954 goto dropunlock; 955 } 956 957 /* 958 * A previous connection in TIMEWAIT state is supposed to catch stray 959 * or duplicate segments arriving late. If this segment was a 960 * legitimate new connection attempt, the old INPCB gets removed and 961 * we can try again to find a listening socket. 962 * 963 * At this point, due to earlier optimism, we may hold only an inpcb 964 * lock, and not the inpcbinfo write lock. If so, we need to try to 965 * acquire it, or if that fails, acquire a reference on the inpcb, 966 * drop all locks, acquire a global write lock, and then re-acquire 967 * the inpcb lock. We may at that point discover that another thread 968 * has tried to free the inpcb, in which case we need to loop back 969 * and try to find a new inpcb to deliver to. 970 * 971 * XXXRW: It may be time to rethink timewait locking. 972 */ 973 relocked: 974 if (inp->inp_flags & INP_TIMEWAIT) { 975 if (ti_locked == TI_UNLOCKED) { 976 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 977 in_pcbref(inp); 978 INP_WUNLOCK(inp); 979 INP_INFO_RLOCK(&V_tcbinfo); 980 ti_locked = TI_RLOCKED; 981 INP_WLOCK(inp); 982 if (in_pcbrele_wlocked(inp)) { 983 inp = NULL; 984 goto findpcb; 985 } 986 } else 987 ti_locked = TI_RLOCKED; 988 } 989 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 990 991 if (thflags & TH_SYN) 992 tcp_dooptions(&to, optp, optlen, TO_SYN); 993 /* 994 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 995 */ 996 if (tcp_twcheck(inp, &to, th, m, tlen)) 997 goto findpcb; 998 INP_INFO_RUNLOCK(&V_tcbinfo); 999 return (IPPROTO_DONE); 1000 } 1001 /* 1002 * The TCPCB may no longer exist if the connection is winding 1003 * down or it is in the CLOSED state. Either way we drop the 1004 * segment and send an appropriate response. 1005 */ 1006 tp = intotcpcb(inp); 1007 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1008 rstreason = BANDLIM_RST_CLOSEDPORT; 1009 goto dropwithreset; 1010 } 1011 1012 #ifdef TCP_OFFLOAD 1013 if (tp->t_flags & TF_TOE) { 1014 tcp_offload_input(tp, m); 1015 m = NULL; /* consumed by the TOE driver */ 1016 goto dropunlock; 1017 } 1018 #endif 1019 1020 /* 1021 * We've identified a valid inpcb, but it could be that we need an 1022 * inpcbinfo write lock but don't hold it. In this case, attempt to 1023 * acquire using the same strategy as the TIMEWAIT case above. If we 1024 * relock, we have to jump back to 'relocked' as the connection might 1025 * now be in TIMEWAIT. 1026 */ 1027 #ifdef INVARIANTS 1028 if ((thflags & (TH_FIN | TH_RST)) != 0) 1029 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1030 #endif 1031 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1032 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1033 !(tp->t_flags & TF_FASTOPEN)))) { 1034 if (ti_locked == TI_UNLOCKED) { 1035 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1036 in_pcbref(inp); 1037 INP_WUNLOCK(inp); 1038 INP_INFO_RLOCK(&V_tcbinfo); 1039 ti_locked = TI_RLOCKED; 1040 INP_WLOCK(inp); 1041 if (in_pcbrele_wlocked(inp)) { 1042 inp = NULL; 1043 goto findpcb; 1044 } 1045 goto relocked; 1046 } else 1047 ti_locked = TI_RLOCKED; 1048 } 1049 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1050 } 1051 1052 #ifdef MAC 1053 INP_WLOCK_ASSERT(inp); 1054 if (mac_inpcb_check_deliver(inp, m)) 1055 goto dropunlock; 1056 #endif 1057 so = inp->inp_socket; 1058 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1059 #ifdef TCPDEBUG 1060 if (so->so_options & SO_DEBUG) { 1061 ostate = tp->t_state; 1062 #ifdef INET6 1063 if (isipv6) { 1064 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1065 } else 1066 #endif 1067 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1068 tcp_savetcp = *th; 1069 } 1070 #endif /* TCPDEBUG */ 1071 /* 1072 * When the socket is accepting connections (the INPCB is in LISTEN 1073 * state) we look into the SYN cache if this is a new connection 1074 * attempt or the completion of a previous one. 1075 */ 1076 if (so->so_options & SO_ACCEPTCONN) { 1077 struct in_conninfo inc; 1078 1079 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1080 "tp not listening", __func__)); 1081 bzero(&inc, sizeof(inc)); 1082 #ifdef INET6 1083 if (isipv6) { 1084 inc.inc_flags |= INC_ISIPV6; 1085 inc.inc6_faddr = ip6->ip6_src; 1086 inc.inc6_laddr = ip6->ip6_dst; 1087 } else 1088 #endif 1089 { 1090 inc.inc_faddr = ip->ip_src; 1091 inc.inc_laddr = ip->ip_dst; 1092 } 1093 inc.inc_fport = th->th_sport; 1094 inc.inc_lport = th->th_dport; 1095 inc.inc_fibnum = so->so_fibnum; 1096 1097 /* 1098 * Check for an existing connection attempt in syncache if 1099 * the flag is only ACK. A successful lookup creates a new 1100 * socket appended to the listen queue in SYN_RECEIVED state. 1101 */ 1102 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1103 1104 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1105 /* 1106 * Parse the TCP options here because 1107 * syncookies need access to the reflected 1108 * timestamp. 1109 */ 1110 tcp_dooptions(&to, optp, optlen, 0); 1111 /* 1112 * NB: syncache_expand() doesn't unlock 1113 * inp and tcpinfo locks. 1114 */ 1115 if (!syncache_expand(&inc, &to, th, &so, m)) { 1116 /* 1117 * No syncache entry or ACK was not 1118 * for our SYN/ACK. Send a RST. 1119 * NB: syncache did its own logging 1120 * of the failure cause. 1121 */ 1122 rstreason = BANDLIM_RST_OPENPORT; 1123 goto dropwithreset; 1124 } 1125 #ifdef TCP_RFC7413 1126 new_tfo_socket: 1127 #endif 1128 if (so == NULL) { 1129 /* 1130 * We completed the 3-way handshake 1131 * but could not allocate a socket 1132 * either due to memory shortage, 1133 * listen queue length limits or 1134 * global socket limits. Send RST 1135 * or wait and have the remote end 1136 * retransmit the ACK for another 1137 * try. 1138 */ 1139 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1140 log(LOG_DEBUG, "%s; %s: Listen socket: " 1141 "Socket allocation failed due to " 1142 "limits or memory shortage, %s\n", 1143 s, __func__, 1144 V_tcp_sc_rst_sock_fail ? 1145 "sending RST" : "try again"); 1146 if (V_tcp_sc_rst_sock_fail) { 1147 rstreason = BANDLIM_UNLIMITED; 1148 goto dropwithreset; 1149 } else 1150 goto dropunlock; 1151 } 1152 /* 1153 * Socket is created in state SYN_RECEIVED. 1154 * Unlock the listen socket, lock the newly 1155 * created socket and update the tp variable. 1156 */ 1157 INP_WUNLOCK(inp); /* listen socket */ 1158 inp = sotoinpcb(so); 1159 /* 1160 * New connection inpcb is already locked by 1161 * syncache_expand(). 1162 */ 1163 INP_WLOCK_ASSERT(inp); 1164 tp = intotcpcb(inp); 1165 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1166 ("%s: ", __func__)); 1167 #ifdef TCP_SIGNATURE 1168 if (sig_checked == 0) { 1169 tcp_dooptions(&to, optp, optlen, 1170 (thflags & TH_SYN) ? TO_SYN : 0); 1171 if (!tcp_signature_verify_input(m, off0, tlen, 1172 optlen, &to, th, tp->t_flags)) { 1173 1174 /* 1175 * In SYN_SENT state if it receives an 1176 * RST, it is allowed for further 1177 * processing. 1178 */ 1179 if ((thflags & TH_RST) == 0 || 1180 (tp->t_state == TCPS_SYN_SENT) == 0) 1181 goto dropunlock; 1182 } 1183 sig_checked = 1; 1184 } 1185 #endif 1186 1187 /* 1188 * Process the segment and the data it 1189 * contains. tcp_do_segment() consumes 1190 * the mbuf chain and unlocks the inpcb. 1191 */ 1192 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1193 iptos, ti_locked); 1194 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1195 return (IPPROTO_DONE); 1196 } 1197 /* 1198 * Segment flag validation for new connection attempts: 1199 * 1200 * Our (SYN|ACK) response was rejected. 1201 * Check with syncache and remove entry to prevent 1202 * retransmits. 1203 * 1204 * NB: syncache_chkrst does its own logging of failure 1205 * causes. 1206 */ 1207 if (thflags & TH_RST) { 1208 syncache_chkrst(&inc, th); 1209 goto dropunlock; 1210 } 1211 /* 1212 * We can't do anything without SYN. 1213 */ 1214 if ((thflags & TH_SYN) == 0) { 1215 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1216 log(LOG_DEBUG, "%s; %s: Listen socket: " 1217 "SYN is missing, segment ignored\n", 1218 s, __func__); 1219 TCPSTAT_INC(tcps_badsyn); 1220 goto dropunlock; 1221 } 1222 /* 1223 * (SYN|ACK) is bogus on a listen socket. 1224 */ 1225 if (thflags & TH_ACK) { 1226 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1227 log(LOG_DEBUG, "%s; %s: Listen socket: " 1228 "SYN|ACK invalid, segment rejected\n", 1229 s, __func__); 1230 syncache_badack(&inc); /* XXX: Not needed! */ 1231 TCPSTAT_INC(tcps_badsyn); 1232 rstreason = BANDLIM_RST_OPENPORT; 1233 goto dropwithreset; 1234 } 1235 /* 1236 * If the drop_synfin option is enabled, drop all 1237 * segments with both the SYN and FIN bits set. 1238 * This prevents e.g. nmap from identifying the 1239 * TCP/IP stack. 1240 * XXX: Poor reasoning. nmap has other methods 1241 * and is constantly refining its stack detection 1242 * strategies. 1243 * XXX: This is a violation of the TCP specification 1244 * and was used by RFC1644. 1245 */ 1246 if ((thflags & TH_FIN) && V_drop_synfin) { 1247 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1248 log(LOG_DEBUG, "%s; %s: Listen socket: " 1249 "SYN|FIN segment ignored (based on " 1250 "sysctl setting)\n", s, __func__); 1251 TCPSTAT_INC(tcps_badsyn); 1252 goto dropunlock; 1253 } 1254 /* 1255 * Segment's flags are (SYN) or (SYN|FIN). 1256 * 1257 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1258 * as they do not affect the state of the TCP FSM. 1259 * The data pointed to by TH_URG and th_urp is ignored. 1260 */ 1261 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1262 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1263 KASSERT(thflags & (TH_SYN), 1264 ("%s: Listen socket: TH_SYN not set", __func__)); 1265 #ifdef INET6 1266 /* 1267 * If deprecated address is forbidden, 1268 * we do not accept SYN to deprecated interface 1269 * address to prevent any new inbound connection from 1270 * getting established. 1271 * When we do not accept SYN, we send a TCP RST, 1272 * with deprecated source address (instead of dropping 1273 * it). We compromise it as it is much better for peer 1274 * to send a RST, and RST will be the final packet 1275 * for the exchange. 1276 * 1277 * If we do not forbid deprecated addresses, we accept 1278 * the SYN packet. RFC2462 does not suggest dropping 1279 * SYN in this case. 1280 * If we decipher RFC2462 5.5.4, it says like this: 1281 * 1. use of deprecated addr with existing 1282 * communication is okay - "SHOULD continue to be 1283 * used" 1284 * 2. use of it with new communication: 1285 * (2a) "SHOULD NOT be used if alternate address 1286 * with sufficient scope is available" 1287 * (2b) nothing mentioned otherwise. 1288 * Here we fall into (2b) case as we have no choice in 1289 * our source address selection - we must obey the peer. 1290 * 1291 * The wording in RFC2462 is confusing, and there are 1292 * multiple description text for deprecated address 1293 * handling - worse, they are not exactly the same. 1294 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1295 */ 1296 if (isipv6 && !V_ip6_use_deprecated) { 1297 struct in6_ifaddr *ia6; 1298 1299 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1300 if (ia6 != NULL && 1301 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1302 ifa_free(&ia6->ia_ifa); 1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1304 log(LOG_DEBUG, "%s; %s: Listen socket: " 1305 "Connection attempt to deprecated " 1306 "IPv6 address rejected\n", 1307 s, __func__); 1308 rstreason = BANDLIM_RST_OPENPORT; 1309 goto dropwithreset; 1310 } 1311 if (ia6) 1312 ifa_free(&ia6->ia_ifa); 1313 } 1314 #endif /* INET6 */ 1315 /* 1316 * Basic sanity checks on incoming SYN requests: 1317 * Don't respond if the destination is a link layer 1318 * broadcast according to RFC1122 4.2.3.10, p. 104. 1319 * If it is from this socket it must be forged. 1320 * Don't respond if the source or destination is a 1321 * global or subnet broad- or multicast address. 1322 * Note that it is quite possible to receive unicast 1323 * link-layer packets with a broadcast IP address. Use 1324 * in_broadcast() to find them. 1325 */ 1326 if (m->m_flags & (M_BCAST|M_MCAST)) { 1327 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1328 log(LOG_DEBUG, "%s; %s: Listen socket: " 1329 "Connection attempt from broad- or multicast " 1330 "link layer address ignored\n", s, __func__); 1331 goto dropunlock; 1332 } 1333 #ifdef INET6 1334 if (isipv6) { 1335 if (th->th_dport == th->th_sport && 1336 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1337 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1338 log(LOG_DEBUG, "%s; %s: Listen socket: " 1339 "Connection attempt to/from self " 1340 "ignored\n", s, __func__); 1341 goto dropunlock; 1342 } 1343 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1344 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1345 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1346 log(LOG_DEBUG, "%s; %s: Listen socket: " 1347 "Connection attempt from/to multicast " 1348 "address ignored\n", s, __func__); 1349 goto dropunlock; 1350 } 1351 } 1352 #endif 1353 #if defined(INET) && defined(INET6) 1354 else 1355 #endif 1356 #ifdef INET 1357 { 1358 if (th->th_dport == th->th_sport && 1359 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1360 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1361 log(LOG_DEBUG, "%s; %s: Listen socket: " 1362 "Connection attempt from/to self " 1363 "ignored\n", s, __func__); 1364 goto dropunlock; 1365 } 1366 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1367 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1368 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1369 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1370 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1371 log(LOG_DEBUG, "%s; %s: Listen socket: " 1372 "Connection attempt from/to broad- " 1373 "or multicast address ignored\n", 1374 s, __func__); 1375 goto dropunlock; 1376 } 1377 } 1378 #endif 1379 /* 1380 * SYN appears to be valid. Create compressed TCP state 1381 * for syncache. 1382 */ 1383 #ifdef TCPDEBUG 1384 if (so->so_options & SO_DEBUG) 1385 tcp_trace(TA_INPUT, ostate, tp, 1386 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1387 #endif 1388 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 1389 tcp_dooptions(&to, optp, optlen, TO_SYN); 1390 #ifdef TCP_RFC7413 1391 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1392 goto new_tfo_socket; 1393 #else 1394 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1395 #endif 1396 /* 1397 * Entry added to syncache and mbuf consumed. 1398 * Only the listen socket is unlocked by syncache_add(). 1399 */ 1400 if (ti_locked == TI_RLOCKED) { 1401 INP_INFO_RUNLOCK(&V_tcbinfo); 1402 ti_locked = TI_UNLOCKED; 1403 } 1404 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1405 return (IPPROTO_DONE); 1406 } else if (tp->t_state == TCPS_LISTEN) { 1407 /* 1408 * When a listen socket is torn down the SO_ACCEPTCONN 1409 * flag is removed first while connections are drained 1410 * from the accept queue in a unlock/lock cycle of the 1411 * ACCEPT_LOCK, opening a race condition allowing a SYN 1412 * attempt go through unhandled. 1413 */ 1414 goto dropunlock; 1415 } 1416 1417 #ifdef TCP_SIGNATURE 1418 if (sig_checked == 0) { 1419 tcp_dooptions(&to, optp, optlen, 1420 (thflags & TH_SYN) ? TO_SYN : 0); 1421 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1422 th, tp->t_flags)) { 1423 1424 /* 1425 * In SYN_SENT state if it receives an RST, it is 1426 * allowed for further processing. 1427 */ 1428 if ((thflags & TH_RST) == 0 || 1429 (tp->t_state == TCPS_SYN_SENT) == 0) 1430 goto dropunlock; 1431 } 1432 sig_checked = 1; 1433 } 1434 #endif 1435 1436 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1437 1438 /* 1439 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1440 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1441 * the inpcb, and unlocks pcbinfo. 1442 */ 1443 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1444 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1445 return (IPPROTO_DONE); 1446 1447 dropwithreset: 1448 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1449 1450 if (ti_locked == TI_RLOCKED) { 1451 INP_INFO_RUNLOCK(&V_tcbinfo); 1452 ti_locked = TI_UNLOCKED; 1453 } 1454 #ifdef INVARIANTS 1455 else { 1456 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1457 "ti_locked: %d", __func__, ti_locked)); 1458 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1459 } 1460 #endif 1461 1462 if (inp != NULL) { 1463 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1464 INP_WUNLOCK(inp); 1465 } else 1466 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1467 m = NULL; /* mbuf chain got consumed. */ 1468 goto drop; 1469 1470 dropunlock: 1471 if (m != NULL) 1472 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1473 1474 if (ti_locked == TI_RLOCKED) { 1475 INP_INFO_RUNLOCK(&V_tcbinfo); 1476 ti_locked = TI_UNLOCKED; 1477 } 1478 #ifdef INVARIANTS 1479 else { 1480 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1481 "ti_locked: %d", __func__, ti_locked)); 1482 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1483 } 1484 #endif 1485 1486 if (inp != NULL) 1487 INP_WUNLOCK(inp); 1488 1489 drop: 1490 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1491 if (s != NULL) 1492 free(s, M_TCPLOG); 1493 if (m != NULL) 1494 m_freem(m); 1495 return (IPPROTO_DONE); 1496 } 1497 1498 void 1499 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1500 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1501 int ti_locked) 1502 { 1503 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1504 int rstreason, todrop, win; 1505 u_long tiwin; 1506 char *s; 1507 struct in_conninfo *inc; 1508 struct mbuf *mfree; 1509 struct tcpopt to; 1510 int tfo_syn; 1511 1512 #ifdef TCPDEBUG 1513 /* 1514 * The size of tcp_saveipgen must be the size of the max ip header, 1515 * now IPv6. 1516 */ 1517 u_char tcp_saveipgen[IP6_HDR_LEN]; 1518 struct tcphdr tcp_savetcp; 1519 short ostate = 0; 1520 #endif 1521 thflags = th->th_flags; 1522 inc = &tp->t_inpcb->inp_inc; 1523 tp->sackhint.last_sack_ack = 0; 1524 sack_changed = 0; 1525 1526 /* 1527 * If this is either a state-changing packet or current state isn't 1528 * established, we require a write lock on tcbinfo. Otherwise, we 1529 * allow the tcbinfo to be in either alocked or unlocked, as the 1530 * caller may have unnecessarily acquired a write lock due to a race. 1531 */ 1532 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1533 tp->t_state != TCPS_ESTABLISHED) { 1534 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1535 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1536 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1537 } else { 1538 #ifdef INVARIANTS 1539 if (ti_locked == TI_RLOCKED) 1540 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1541 else { 1542 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1543 "ti_locked: %d", __func__, ti_locked)); 1544 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1545 } 1546 #endif 1547 } 1548 INP_WLOCK_ASSERT(tp->t_inpcb); 1549 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1550 __func__)); 1551 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1552 __func__)); 1553 1554 #ifdef TCPPCAP 1555 /* Save segment, if requested. */ 1556 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1557 #endif 1558 1559 /* 1560 * Segment received on connection. 1561 * Reset idle time and keep-alive timer. 1562 * XXX: This should be done after segment 1563 * validation to ignore broken/spoofed segs. 1564 */ 1565 tp->t_rcvtime = ticks; 1566 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1567 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1568 1569 /* 1570 * Scale up the window into a 32-bit value. 1571 * For the SYN_SENT state the scale is zero. 1572 */ 1573 tiwin = th->th_win << tp->snd_scale; 1574 1575 /* 1576 * TCP ECN processing. 1577 */ 1578 if (tp->t_flags & TF_ECN_PERMIT) { 1579 if (thflags & TH_CWR) 1580 tp->t_flags &= ~TF_ECN_SND_ECE; 1581 switch (iptos & IPTOS_ECN_MASK) { 1582 case IPTOS_ECN_CE: 1583 tp->t_flags |= TF_ECN_SND_ECE; 1584 TCPSTAT_INC(tcps_ecn_ce); 1585 break; 1586 case IPTOS_ECN_ECT0: 1587 TCPSTAT_INC(tcps_ecn_ect0); 1588 break; 1589 case IPTOS_ECN_ECT1: 1590 TCPSTAT_INC(tcps_ecn_ect1); 1591 break; 1592 } 1593 1594 /* Process a packet differently from RFC3168. */ 1595 cc_ecnpkt_handler(tp, th, iptos); 1596 1597 /* Congestion experienced. */ 1598 if (thflags & TH_ECE) { 1599 cc_cong_signal(tp, th, CC_ECN); 1600 } 1601 } 1602 1603 /* 1604 * Parse options on any incoming segment. 1605 */ 1606 tcp_dooptions(&to, (u_char *)(th + 1), 1607 (th->th_off << 2) - sizeof(struct tcphdr), 1608 (thflags & TH_SYN) ? TO_SYN : 0); 1609 1610 /* 1611 * If echoed timestamp is later than the current time, 1612 * fall back to non RFC1323 RTT calculation. Normalize 1613 * timestamp if syncookies were used when this connection 1614 * was established. 1615 */ 1616 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1617 to.to_tsecr -= tp->ts_offset; 1618 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1619 to.to_tsecr = 0; 1620 } 1621 /* 1622 * If timestamps were negotiated during SYN/ACK they should 1623 * appear on every segment during this session and vice versa. 1624 */ 1625 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1626 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1627 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1628 "no action\n", s, __func__); 1629 free(s, M_TCPLOG); 1630 } 1631 } 1632 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1633 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1634 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1635 "no action\n", s, __func__); 1636 free(s, M_TCPLOG); 1637 } 1638 } 1639 1640 /* 1641 * Process options only when we get SYN/ACK back. The SYN case 1642 * for incoming connections is handled in tcp_syncache. 1643 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1644 * or <SYN,ACK>) segment itself is never scaled. 1645 * XXX this is traditional behavior, may need to be cleaned up. 1646 */ 1647 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1648 if ((to.to_flags & TOF_SCALE) && 1649 (tp->t_flags & TF_REQ_SCALE)) { 1650 tp->t_flags |= TF_RCVD_SCALE; 1651 tp->snd_scale = to.to_wscale; 1652 } 1653 /* 1654 * Initial send window. It will be updated with 1655 * the next incoming segment to the scaled value. 1656 */ 1657 tp->snd_wnd = th->th_win; 1658 if (to.to_flags & TOF_TS) { 1659 tp->t_flags |= TF_RCVD_TSTMP; 1660 tp->ts_recent = to.to_tsval; 1661 tp->ts_recent_age = tcp_ts_getticks(); 1662 } 1663 if (to.to_flags & TOF_MSS) 1664 tcp_mss(tp, to.to_mss); 1665 if ((tp->t_flags & TF_SACK_PERMIT) && 1666 (to.to_flags & TOF_SACKPERM) == 0) 1667 tp->t_flags &= ~TF_SACK_PERMIT; 1668 } 1669 1670 /* 1671 * Header prediction: check for the two common cases 1672 * of a uni-directional data xfer. If the packet has 1673 * no control flags, is in-sequence, the window didn't 1674 * change and we're not retransmitting, it's a 1675 * candidate. If the length is zero and the ack moved 1676 * forward, we're the sender side of the xfer. Just 1677 * free the data acked & wake any higher level process 1678 * that was blocked waiting for space. If the length 1679 * is non-zero and the ack didn't move, we're the 1680 * receiver side. If we're getting packets in-order 1681 * (the reassembly queue is empty), add the data to 1682 * the socket buffer and note that we need a delayed ack. 1683 * Make sure that the hidden state-flags are also off. 1684 * Since we check for TCPS_ESTABLISHED first, it can only 1685 * be TH_NEEDSYN. 1686 */ 1687 if (tp->t_state == TCPS_ESTABLISHED && 1688 th->th_seq == tp->rcv_nxt && 1689 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1690 tp->snd_nxt == tp->snd_max && 1691 tiwin && tiwin == tp->snd_wnd && 1692 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1693 LIST_EMPTY(&tp->t_segq) && 1694 ((to.to_flags & TOF_TS) == 0 || 1695 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1696 1697 /* 1698 * If last ACK falls within this segment's sequence numbers, 1699 * record the timestamp. 1700 * NOTE that the test is modified according to the latest 1701 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1702 */ 1703 if ((to.to_flags & TOF_TS) != 0 && 1704 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1705 tp->ts_recent_age = tcp_ts_getticks(); 1706 tp->ts_recent = to.to_tsval; 1707 } 1708 1709 if (tlen == 0) { 1710 if (SEQ_GT(th->th_ack, tp->snd_una) && 1711 SEQ_LEQ(th->th_ack, tp->snd_max) && 1712 !IN_RECOVERY(tp->t_flags) && 1713 (to.to_flags & TOF_SACK) == 0 && 1714 TAILQ_EMPTY(&tp->snd_holes)) { 1715 /* 1716 * This is a pure ack for outstanding data. 1717 */ 1718 if (ti_locked == TI_RLOCKED) 1719 INP_INFO_RUNLOCK(&V_tcbinfo); 1720 ti_locked = TI_UNLOCKED; 1721 1722 TCPSTAT_INC(tcps_predack); 1723 1724 /* 1725 * "bad retransmit" recovery. 1726 */ 1727 if (tp->t_rxtshift == 1 && 1728 tp->t_flags & TF_PREVVALID && 1729 (int)(ticks - tp->t_badrxtwin) < 0) { 1730 cc_cong_signal(tp, th, CC_RTO_ERR); 1731 } 1732 1733 /* 1734 * Recalculate the transmit timer / rtt. 1735 * 1736 * Some boxes send broken timestamp replies 1737 * during the SYN+ACK phase, ignore 1738 * timestamps of 0 or we could calculate a 1739 * huge RTT and blow up the retransmit timer. 1740 */ 1741 if ((to.to_flags & TOF_TS) != 0 && 1742 to.to_tsecr) { 1743 u_int t; 1744 1745 t = tcp_ts_getticks() - to.to_tsecr; 1746 if (!tp->t_rttlow || tp->t_rttlow > t) 1747 tp->t_rttlow = t; 1748 tcp_xmit_timer(tp, 1749 TCP_TS_TO_TICKS(t) + 1); 1750 } else if (tp->t_rtttime && 1751 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1752 if (!tp->t_rttlow || 1753 tp->t_rttlow > ticks - tp->t_rtttime) 1754 tp->t_rttlow = ticks - tp->t_rtttime; 1755 tcp_xmit_timer(tp, 1756 ticks - tp->t_rtttime); 1757 } 1758 acked = BYTES_THIS_ACK(tp, th); 1759 1760 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1761 hhook_run_tcp_est_in(tp, th, &to); 1762 1763 TCPSTAT_INC(tcps_rcvackpack); 1764 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1765 sbdrop(&so->so_snd, acked); 1766 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1767 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1768 tp->snd_recover = th->th_ack - 1; 1769 1770 /* 1771 * Let the congestion control algorithm update 1772 * congestion control related information. This 1773 * typically means increasing the congestion 1774 * window. 1775 */ 1776 cc_ack_received(tp, th, CC_ACK); 1777 1778 tp->snd_una = th->th_ack; 1779 /* 1780 * Pull snd_wl2 up to prevent seq wrap relative 1781 * to th_ack. 1782 */ 1783 tp->snd_wl2 = th->th_ack; 1784 tp->t_dupacks = 0; 1785 m_freem(m); 1786 1787 /* 1788 * If all outstanding data are acked, stop 1789 * retransmit timer, otherwise restart timer 1790 * using current (possibly backed-off) value. 1791 * If process is waiting for space, 1792 * wakeup/selwakeup/signal. If data 1793 * are ready to send, let tcp_output 1794 * decide between more output or persist. 1795 */ 1796 #ifdef TCPDEBUG 1797 if (so->so_options & SO_DEBUG) 1798 tcp_trace(TA_INPUT, ostate, tp, 1799 (void *)tcp_saveipgen, 1800 &tcp_savetcp, 0); 1801 #endif 1802 TCP_PROBE3(debug__input, tp, th, 1803 mtod(m, const char *)); 1804 if (tp->snd_una == tp->snd_max) 1805 tcp_timer_activate(tp, TT_REXMT, 0); 1806 else if (!tcp_timer_active(tp, TT_PERSIST)) 1807 tcp_timer_activate(tp, TT_REXMT, 1808 tp->t_rxtcur); 1809 sowwakeup(so); 1810 if (sbavail(&so->so_snd)) 1811 (void) tp->t_fb->tfb_tcp_output(tp); 1812 goto check_delack; 1813 } 1814 } else if (th->th_ack == tp->snd_una && 1815 tlen <= sbspace(&so->so_rcv)) { 1816 int newsize = 0; /* automatic sockbuf scaling */ 1817 1818 /* 1819 * This is a pure, in-sequence data packet with 1820 * nothing on the reassembly queue and we have enough 1821 * buffer space to take it. 1822 */ 1823 if (ti_locked == TI_RLOCKED) 1824 INP_INFO_RUNLOCK(&V_tcbinfo); 1825 ti_locked = TI_UNLOCKED; 1826 1827 /* Clean receiver SACK report if present */ 1828 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1829 tcp_clean_sackreport(tp); 1830 TCPSTAT_INC(tcps_preddat); 1831 tp->rcv_nxt += tlen; 1832 /* 1833 * Pull snd_wl1 up to prevent seq wrap relative to 1834 * th_seq. 1835 */ 1836 tp->snd_wl1 = th->th_seq; 1837 /* 1838 * Pull rcv_up up to prevent seq wrap relative to 1839 * rcv_nxt. 1840 */ 1841 tp->rcv_up = tp->rcv_nxt; 1842 TCPSTAT_INC(tcps_rcvpack); 1843 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1844 #ifdef TCPDEBUG 1845 if (so->so_options & SO_DEBUG) 1846 tcp_trace(TA_INPUT, ostate, tp, 1847 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1848 #endif 1849 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 1850 1851 /* 1852 * Automatic sizing of receive socket buffer. Often the send 1853 * buffer size is not optimally adjusted to the actual network 1854 * conditions at hand (delay bandwidth product). Setting the 1855 * buffer size too small limits throughput on links with high 1856 * bandwidth and high delay (eg. trans-continental/oceanic links). 1857 * 1858 * On the receive side the socket buffer memory is only rarely 1859 * used to any significant extent. This allows us to be much 1860 * more aggressive in scaling the receive socket buffer. For 1861 * the case that the buffer space is actually used to a large 1862 * extent and we run out of kernel memory we can simply drop 1863 * the new segments; TCP on the sender will just retransmit it 1864 * later. Setting the buffer size too big may only consume too 1865 * much kernel memory if the application doesn't read() from 1866 * the socket or packet loss or reordering makes use of the 1867 * reassembly queue. 1868 * 1869 * The criteria to step up the receive buffer one notch are: 1870 * 1. Application has not set receive buffer size with 1871 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1872 * 2. the number of bytes received during the time it takes 1873 * one timestamp to be reflected back to us (the RTT); 1874 * 3. received bytes per RTT is within seven eighth of the 1875 * current socket buffer size; 1876 * 4. receive buffer size has not hit maximal automatic size; 1877 * 1878 * This algorithm does one step per RTT at most and only if 1879 * we receive a bulk stream w/o packet losses or reorderings. 1880 * Shrinking the buffer during idle times is not necessary as 1881 * it doesn't consume any memory when idle. 1882 * 1883 * TODO: Only step up if the application is actually serving 1884 * the buffer to better manage the socket buffer resources. 1885 */ 1886 if (V_tcp_do_autorcvbuf && 1887 (to.to_flags & TOF_TS) && 1888 to.to_tsecr && 1889 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1890 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1891 to.to_tsecr - tp->rfbuf_ts < hz) { 1892 if (tp->rfbuf_cnt > 1893 (so->so_rcv.sb_hiwat / 8 * 7) && 1894 so->so_rcv.sb_hiwat < 1895 V_tcp_autorcvbuf_max) { 1896 newsize = 1897 min(so->so_rcv.sb_hiwat + 1898 V_tcp_autorcvbuf_inc, 1899 V_tcp_autorcvbuf_max); 1900 } 1901 /* Start over with next RTT. */ 1902 tp->rfbuf_ts = 0; 1903 tp->rfbuf_cnt = 0; 1904 } else 1905 tp->rfbuf_cnt += tlen; /* add up */ 1906 } 1907 1908 /* Add data to socket buffer. */ 1909 SOCKBUF_LOCK(&so->so_rcv); 1910 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1911 m_freem(m); 1912 } else { 1913 /* 1914 * Set new socket buffer size. 1915 * Give up when limit is reached. 1916 */ 1917 if (newsize) 1918 if (!sbreserve_locked(&so->so_rcv, 1919 newsize, so, NULL)) 1920 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1921 m_adj(m, drop_hdrlen); /* delayed header drop */ 1922 sbappendstream_locked(&so->so_rcv, m, 0); 1923 } 1924 /* NB: sorwakeup_locked() does an implicit unlock. */ 1925 sorwakeup_locked(so); 1926 if (DELAY_ACK(tp, tlen)) { 1927 tp->t_flags |= TF_DELACK; 1928 } else { 1929 tp->t_flags |= TF_ACKNOW; 1930 tp->t_fb->tfb_tcp_output(tp); 1931 } 1932 goto check_delack; 1933 } 1934 } 1935 1936 /* 1937 * Calculate amount of space in receive window, 1938 * and then do TCP input processing. 1939 * Receive window is amount of space in rcv queue, 1940 * but not less than advertised window. 1941 */ 1942 win = sbspace(&so->so_rcv); 1943 if (win < 0) 1944 win = 0; 1945 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1946 1947 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1948 tp->rfbuf_ts = 0; 1949 tp->rfbuf_cnt = 0; 1950 1951 switch (tp->t_state) { 1952 1953 /* 1954 * If the state is SYN_RECEIVED: 1955 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1956 */ 1957 case TCPS_SYN_RECEIVED: 1958 if ((thflags & TH_ACK) && 1959 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1960 SEQ_GT(th->th_ack, tp->snd_max))) { 1961 rstreason = BANDLIM_RST_OPENPORT; 1962 goto dropwithreset; 1963 } 1964 #ifdef TCP_RFC7413 1965 if (tp->t_flags & TF_FASTOPEN) { 1966 /* 1967 * When a TFO connection is in SYN_RECEIVED, the 1968 * only valid packets are the initial SYN, a 1969 * retransmit/copy of the initial SYN (possibly with 1970 * a subset of the original data), a valid ACK, a 1971 * FIN, or a RST. 1972 */ 1973 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1974 rstreason = BANDLIM_RST_OPENPORT; 1975 goto dropwithreset; 1976 } else if (thflags & TH_SYN) { 1977 /* non-initial SYN is ignored */ 1978 if ((tcp_timer_active(tp, TT_DELACK) || 1979 tcp_timer_active(tp, TT_REXMT))) 1980 goto drop; 1981 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1982 goto drop; 1983 } 1984 } 1985 #endif 1986 break; 1987 1988 /* 1989 * If the state is SYN_SENT: 1990 * if seg contains an ACK, but not for our SYN, drop the input. 1991 * if seg contains a RST, then drop the connection. 1992 * if seg does not contain SYN, then drop it. 1993 * Otherwise this is an acceptable SYN segment 1994 * initialize tp->rcv_nxt and tp->irs 1995 * if seg contains ack then advance tp->snd_una 1996 * if seg contains an ECE and ECN support is enabled, the stream 1997 * is ECN capable. 1998 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1999 * arrange for segment to be acked (eventually) 2000 * continue processing rest of data/controls, beginning with URG 2001 */ 2002 case TCPS_SYN_SENT: 2003 if ((thflags & TH_ACK) && 2004 (SEQ_LEQ(th->th_ack, tp->iss) || 2005 SEQ_GT(th->th_ack, tp->snd_max))) { 2006 rstreason = BANDLIM_UNLIMITED; 2007 goto dropwithreset; 2008 } 2009 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2010 TCP_PROBE5(connect__refused, NULL, tp, 2011 mtod(m, const char *), tp, th); 2012 tp = tcp_drop(tp, ECONNREFUSED); 2013 } 2014 if (thflags & TH_RST) 2015 goto drop; 2016 if (!(thflags & TH_SYN)) 2017 goto drop; 2018 2019 tp->irs = th->th_seq; 2020 tcp_rcvseqinit(tp); 2021 if (thflags & TH_ACK) { 2022 TCPSTAT_INC(tcps_connects); 2023 soisconnected(so); 2024 #ifdef MAC 2025 mac_socketpeer_set_from_mbuf(m, so); 2026 #endif 2027 /* Do window scaling on this connection? */ 2028 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2029 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2030 tp->rcv_scale = tp->request_r_scale; 2031 } 2032 tp->rcv_adv += imin(tp->rcv_wnd, 2033 TCP_MAXWIN << tp->rcv_scale); 2034 tp->snd_una++; /* SYN is acked */ 2035 /* 2036 * If there's data, delay ACK; if there's also a FIN 2037 * ACKNOW will be turned on later. 2038 */ 2039 if (DELAY_ACK(tp, tlen) && tlen != 0) 2040 tcp_timer_activate(tp, TT_DELACK, 2041 tcp_delacktime); 2042 else 2043 tp->t_flags |= TF_ACKNOW; 2044 2045 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2046 tp->t_flags |= TF_ECN_PERMIT; 2047 TCPSTAT_INC(tcps_ecn_shs); 2048 } 2049 2050 /* 2051 * Received <SYN,ACK> in SYN_SENT[*] state. 2052 * Transitions: 2053 * SYN_SENT --> ESTABLISHED 2054 * SYN_SENT* --> FIN_WAIT_1 2055 */ 2056 tp->t_starttime = ticks; 2057 if (tp->t_flags & TF_NEEDFIN) { 2058 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2059 tp->t_flags &= ~TF_NEEDFIN; 2060 thflags &= ~TH_SYN; 2061 } else { 2062 tcp_state_change(tp, TCPS_ESTABLISHED); 2063 TCP_PROBE5(connect__established, NULL, tp, 2064 mtod(m, const char *), tp, th); 2065 cc_conn_init(tp); 2066 tcp_timer_activate(tp, TT_KEEP, 2067 TP_KEEPIDLE(tp)); 2068 } 2069 } else { 2070 /* 2071 * Received initial SYN in SYN-SENT[*] state => 2072 * simultaneous open. 2073 * If it succeeds, connection is * half-synchronized. 2074 * Otherwise, do 3-way handshake: 2075 * SYN-SENT -> SYN-RECEIVED 2076 * SYN-SENT* -> SYN-RECEIVED* 2077 */ 2078 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2079 tcp_timer_activate(tp, TT_REXMT, 0); 2080 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2081 } 2082 2083 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2084 "ti_locked %d", __func__, ti_locked)); 2085 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2086 INP_WLOCK_ASSERT(tp->t_inpcb); 2087 2088 /* 2089 * Advance th->th_seq to correspond to first data byte. 2090 * If data, trim to stay within window, 2091 * dropping FIN if necessary. 2092 */ 2093 th->th_seq++; 2094 if (tlen > tp->rcv_wnd) { 2095 todrop = tlen - tp->rcv_wnd; 2096 m_adj(m, -todrop); 2097 tlen = tp->rcv_wnd; 2098 thflags &= ~TH_FIN; 2099 TCPSTAT_INC(tcps_rcvpackafterwin); 2100 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2101 } 2102 tp->snd_wl1 = th->th_seq - 1; 2103 tp->rcv_up = th->th_seq; 2104 /* 2105 * Client side of transaction: already sent SYN and data. 2106 * If the remote host used T/TCP to validate the SYN, 2107 * our data will be ACK'd; if so, enter normal data segment 2108 * processing in the middle of step 5, ack processing. 2109 * Otherwise, goto step 6. 2110 */ 2111 if (thflags & TH_ACK) 2112 goto process_ACK; 2113 2114 goto step6; 2115 2116 /* 2117 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2118 * do normal processing. 2119 * 2120 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2121 */ 2122 case TCPS_LAST_ACK: 2123 case TCPS_CLOSING: 2124 break; /* continue normal processing */ 2125 } 2126 2127 /* 2128 * States other than LISTEN or SYN_SENT. 2129 * First check the RST flag and sequence number since reset segments 2130 * are exempt from the timestamp and connection count tests. This 2131 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2132 * below which allowed reset segments in half the sequence space 2133 * to fall though and be processed (which gives forged reset 2134 * segments with a random sequence number a 50 percent chance of 2135 * killing a connection). 2136 * Then check timestamp, if present. 2137 * Then check the connection count, if present. 2138 * Then check that at least some bytes of segment are within 2139 * receive window. If segment begins before rcv_nxt, 2140 * drop leading data (and SYN); if nothing left, just ack. 2141 */ 2142 if (thflags & TH_RST) { 2143 /* 2144 * RFC5961 Section 3.2 2145 * 2146 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2147 * - If RST is in window, we send challenge ACK. 2148 * 2149 * Note: to take into account delayed ACKs, we should 2150 * test against last_ack_sent instead of rcv_nxt. 2151 * Note 2: we handle special case of closed window, not 2152 * covered by the RFC. 2153 */ 2154 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2155 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2156 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2157 2158 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2159 KASSERT(ti_locked == TI_RLOCKED, 2160 ("%s: TH_RST ti_locked %d, th %p tp %p", 2161 __func__, ti_locked, th, tp)); 2162 KASSERT(tp->t_state != TCPS_SYN_SENT, 2163 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2164 __func__, th, tp)); 2165 2166 if (V_tcp_insecure_rst || 2167 tp->last_ack_sent == th->th_seq) { 2168 TCPSTAT_INC(tcps_drops); 2169 /* Drop the connection. */ 2170 switch (tp->t_state) { 2171 case TCPS_SYN_RECEIVED: 2172 so->so_error = ECONNREFUSED; 2173 goto close; 2174 case TCPS_ESTABLISHED: 2175 case TCPS_FIN_WAIT_1: 2176 case TCPS_FIN_WAIT_2: 2177 case TCPS_CLOSE_WAIT: 2178 so->so_error = ECONNRESET; 2179 close: 2180 tcp_state_change(tp, TCPS_CLOSED); 2181 /* FALLTHROUGH */ 2182 default: 2183 tp = tcp_close(tp); 2184 } 2185 } else { 2186 TCPSTAT_INC(tcps_badrst); 2187 /* Send challenge ACK. */ 2188 tcp_respond(tp, mtod(m, void *), th, m, 2189 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2190 tp->last_ack_sent = tp->rcv_nxt; 2191 m = NULL; 2192 } 2193 } 2194 goto drop; 2195 } 2196 2197 /* 2198 * RFC5961 Section 4.2 2199 * Send challenge ACK for any SYN in synchronized state. 2200 */ 2201 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2202 tp->t_state != TCPS_SYN_RECEIVED) { 2203 KASSERT(ti_locked == TI_RLOCKED, 2204 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2205 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2206 2207 TCPSTAT_INC(tcps_badsyn); 2208 if (V_tcp_insecure_syn && 2209 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2210 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2211 tp = tcp_drop(tp, ECONNRESET); 2212 rstreason = BANDLIM_UNLIMITED; 2213 } else { 2214 /* Send challenge ACK. */ 2215 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2216 tp->snd_nxt, TH_ACK); 2217 tp->last_ack_sent = tp->rcv_nxt; 2218 m = NULL; 2219 } 2220 goto drop; 2221 } 2222 2223 /* 2224 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2225 * and it's less than ts_recent, drop it. 2226 */ 2227 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2228 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2229 2230 /* Check to see if ts_recent is over 24 days old. */ 2231 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2232 /* 2233 * Invalidate ts_recent. If this segment updates 2234 * ts_recent, the age will be reset later and ts_recent 2235 * will get a valid value. If it does not, setting 2236 * ts_recent to zero will at least satisfy the 2237 * requirement that zero be placed in the timestamp 2238 * echo reply when ts_recent isn't valid. The 2239 * age isn't reset until we get a valid ts_recent 2240 * because we don't want out-of-order segments to be 2241 * dropped when ts_recent is old. 2242 */ 2243 tp->ts_recent = 0; 2244 } else { 2245 TCPSTAT_INC(tcps_rcvduppack); 2246 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2247 TCPSTAT_INC(tcps_pawsdrop); 2248 if (tlen) 2249 goto dropafterack; 2250 goto drop; 2251 } 2252 } 2253 2254 /* 2255 * In the SYN-RECEIVED state, validate that the packet belongs to 2256 * this connection before trimming the data to fit the receive 2257 * window. Check the sequence number versus IRS since we know 2258 * the sequence numbers haven't wrapped. This is a partial fix 2259 * for the "LAND" DoS attack. 2260 */ 2261 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2262 rstreason = BANDLIM_RST_OPENPORT; 2263 goto dropwithreset; 2264 } 2265 2266 todrop = tp->rcv_nxt - th->th_seq; 2267 if (todrop > 0) { 2268 if (thflags & TH_SYN) { 2269 thflags &= ~TH_SYN; 2270 th->th_seq++; 2271 if (th->th_urp > 1) 2272 th->th_urp--; 2273 else 2274 thflags &= ~TH_URG; 2275 todrop--; 2276 } 2277 /* 2278 * Following if statement from Stevens, vol. 2, p. 960. 2279 */ 2280 if (todrop > tlen 2281 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2282 /* 2283 * Any valid FIN must be to the left of the window. 2284 * At this point the FIN must be a duplicate or out 2285 * of sequence; drop it. 2286 */ 2287 thflags &= ~TH_FIN; 2288 2289 /* 2290 * Send an ACK to resynchronize and drop any data. 2291 * But keep on processing for RST or ACK. 2292 */ 2293 tp->t_flags |= TF_ACKNOW; 2294 todrop = tlen; 2295 TCPSTAT_INC(tcps_rcvduppack); 2296 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2297 } else { 2298 TCPSTAT_INC(tcps_rcvpartduppack); 2299 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2300 } 2301 drop_hdrlen += todrop; /* drop from the top afterwards */ 2302 th->th_seq += todrop; 2303 tlen -= todrop; 2304 if (th->th_urp > todrop) 2305 th->th_urp -= todrop; 2306 else { 2307 thflags &= ~TH_URG; 2308 th->th_urp = 0; 2309 } 2310 } 2311 2312 /* 2313 * If new data are received on a connection after the 2314 * user processes are gone, then RST the other end. 2315 */ 2316 if ((so->so_state & SS_NOFDREF) && 2317 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2318 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2319 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2320 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2321 2322 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2323 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2324 "after socket was closed, " 2325 "sending RST and removing tcpcb\n", 2326 s, __func__, tcpstates[tp->t_state], tlen); 2327 free(s, M_TCPLOG); 2328 } 2329 tp = tcp_close(tp); 2330 TCPSTAT_INC(tcps_rcvafterclose); 2331 rstreason = BANDLIM_UNLIMITED; 2332 goto dropwithreset; 2333 } 2334 2335 /* 2336 * If segment ends after window, drop trailing data 2337 * (and PUSH and FIN); if nothing left, just ACK. 2338 */ 2339 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2340 if (todrop > 0) { 2341 TCPSTAT_INC(tcps_rcvpackafterwin); 2342 if (todrop >= tlen) { 2343 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2344 /* 2345 * If window is closed can only take segments at 2346 * window edge, and have to drop data and PUSH from 2347 * incoming segments. Continue processing, but 2348 * remember to ack. Otherwise, drop segment 2349 * and ack. 2350 */ 2351 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2352 tp->t_flags |= TF_ACKNOW; 2353 TCPSTAT_INC(tcps_rcvwinprobe); 2354 } else 2355 goto dropafterack; 2356 } else 2357 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2358 m_adj(m, -todrop); 2359 tlen -= todrop; 2360 thflags &= ~(TH_PUSH|TH_FIN); 2361 } 2362 2363 /* 2364 * If last ACK falls within this segment's sequence numbers, 2365 * record its timestamp. 2366 * NOTE: 2367 * 1) That the test incorporates suggestions from the latest 2368 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2369 * 2) That updating only on newer timestamps interferes with 2370 * our earlier PAWS tests, so this check should be solely 2371 * predicated on the sequence space of this segment. 2372 * 3) That we modify the segment boundary check to be 2373 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2374 * instead of RFC1323's 2375 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2376 * This modified check allows us to overcome RFC1323's 2377 * limitations as described in Stevens TCP/IP Illustrated 2378 * Vol. 2 p.869. In such cases, we can still calculate the 2379 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2380 */ 2381 if ((to.to_flags & TOF_TS) != 0 && 2382 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2383 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2384 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2385 tp->ts_recent_age = tcp_ts_getticks(); 2386 tp->ts_recent = to.to_tsval; 2387 } 2388 2389 /* 2390 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2391 * flag is on (half-synchronized state), then queue data for 2392 * later processing; else drop segment and return. 2393 */ 2394 if ((thflags & TH_ACK) == 0) { 2395 if (tp->t_state == TCPS_SYN_RECEIVED || 2396 (tp->t_flags & TF_NEEDSYN)) { 2397 #ifdef TCP_RFC7413 2398 if (tp->t_state == TCPS_SYN_RECEIVED && 2399 tp->t_flags & TF_FASTOPEN) { 2400 tp->snd_wnd = tiwin; 2401 cc_conn_init(tp); 2402 } 2403 #endif 2404 goto step6; 2405 } else if (tp->t_flags & TF_ACKNOW) 2406 goto dropafterack; 2407 else 2408 goto drop; 2409 } 2410 2411 /* 2412 * Ack processing. 2413 */ 2414 switch (tp->t_state) { 2415 2416 /* 2417 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2418 * ESTABLISHED state and continue processing. 2419 * The ACK was checked above. 2420 */ 2421 case TCPS_SYN_RECEIVED: 2422 2423 TCPSTAT_INC(tcps_connects); 2424 soisconnected(so); 2425 /* Do window scaling? */ 2426 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2427 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2428 tp->rcv_scale = tp->request_r_scale; 2429 tp->snd_wnd = tiwin; 2430 } 2431 /* 2432 * Make transitions: 2433 * SYN-RECEIVED -> ESTABLISHED 2434 * SYN-RECEIVED* -> FIN-WAIT-1 2435 */ 2436 tp->t_starttime = ticks; 2437 if (tp->t_flags & TF_NEEDFIN) { 2438 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2439 tp->t_flags &= ~TF_NEEDFIN; 2440 } else { 2441 tcp_state_change(tp, TCPS_ESTABLISHED); 2442 TCP_PROBE5(accept__established, NULL, tp, 2443 mtod(m, const char *), tp, th); 2444 #ifdef TCP_RFC7413 2445 if (tp->t_tfo_pending) { 2446 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2447 tp->t_tfo_pending = NULL; 2448 2449 /* 2450 * Account for the ACK of our SYN prior to 2451 * regular ACK processing below. 2452 */ 2453 tp->snd_una++; 2454 } 2455 /* 2456 * TFO connections call cc_conn_init() during SYN 2457 * processing. Calling it again here for such 2458 * connections is not harmless as it would undo the 2459 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2460 * is retransmitted. 2461 */ 2462 if (!(tp->t_flags & TF_FASTOPEN)) 2463 #endif 2464 cc_conn_init(tp); 2465 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2466 } 2467 /* 2468 * If segment contains data or ACK, will call tcp_reass() 2469 * later; if not, do so now to pass queued data to user. 2470 */ 2471 if (tlen == 0 && (thflags & TH_FIN) == 0) 2472 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2473 (struct mbuf *)0); 2474 tp->snd_wl1 = th->th_seq - 1; 2475 /* FALLTHROUGH */ 2476 2477 /* 2478 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2479 * ACKs. If the ack is in the range 2480 * tp->snd_una < th->th_ack <= tp->snd_max 2481 * then advance tp->snd_una to th->th_ack and drop 2482 * data from the retransmission queue. If this ACK reflects 2483 * more up to date window information we update our window information. 2484 */ 2485 case TCPS_ESTABLISHED: 2486 case TCPS_FIN_WAIT_1: 2487 case TCPS_FIN_WAIT_2: 2488 case TCPS_CLOSE_WAIT: 2489 case TCPS_CLOSING: 2490 case TCPS_LAST_ACK: 2491 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2492 TCPSTAT_INC(tcps_rcvacktoomuch); 2493 goto dropafterack; 2494 } 2495 if ((tp->t_flags & TF_SACK_PERMIT) && 2496 ((to.to_flags & TOF_SACK) || 2497 !TAILQ_EMPTY(&tp->snd_holes))) 2498 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2499 else 2500 /* 2501 * Reset the value so that previous (valid) value 2502 * from the last ack with SACK doesn't get used. 2503 */ 2504 tp->sackhint.sacked_bytes = 0; 2505 2506 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2507 hhook_run_tcp_est_in(tp, th, &to); 2508 2509 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2510 u_int maxseg; 2511 2512 maxseg = tcp_maxseg(tp); 2513 if (tlen == 0 && 2514 (tiwin == tp->snd_wnd || 2515 (tp->t_flags & TF_SACK_PERMIT))) { 2516 /* 2517 * If this is the first time we've seen a 2518 * FIN from the remote, this is not a 2519 * duplicate and it needs to be processed 2520 * normally. This happens during a 2521 * simultaneous close. 2522 */ 2523 if ((thflags & TH_FIN) && 2524 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2525 tp->t_dupacks = 0; 2526 break; 2527 } 2528 TCPSTAT_INC(tcps_rcvdupack); 2529 /* 2530 * If we have outstanding data (other than 2531 * a window probe), this is a completely 2532 * duplicate ack (ie, window info didn't 2533 * change and FIN isn't set), 2534 * the ack is the biggest we've 2535 * seen and we've seen exactly our rexmt 2536 * threshhold of them, assume a packet 2537 * has been dropped and retransmit it. 2538 * Kludge snd_nxt & the congestion 2539 * window so we send only this one 2540 * packet. 2541 * 2542 * We know we're losing at the current 2543 * window size so do congestion avoidance 2544 * (set ssthresh to half the current window 2545 * and pull our congestion window back to 2546 * the new ssthresh). 2547 * 2548 * Dup acks mean that packets have left the 2549 * network (they're now cached at the receiver) 2550 * so bump cwnd by the amount in the receiver 2551 * to keep a constant cwnd packets in the 2552 * network. 2553 * 2554 * When using TCP ECN, notify the peer that 2555 * we reduced the cwnd. 2556 */ 2557 /* 2558 * Following 2 kinds of acks should not affect 2559 * dupack counting: 2560 * 1) Old acks 2561 * 2) Acks with SACK but without any new SACK 2562 * information in them. These could result from 2563 * any anomaly in the network like a switch 2564 * duplicating packets or a possible DoS attack. 2565 */ 2566 if (th->th_ack != tp->snd_una || 2567 ((tp->t_flags & TF_SACK_PERMIT) && 2568 !sack_changed)) 2569 break; 2570 else if (!tcp_timer_active(tp, TT_REXMT)) 2571 tp->t_dupacks = 0; 2572 else if (++tp->t_dupacks > tcprexmtthresh || 2573 IN_FASTRECOVERY(tp->t_flags)) { 2574 cc_ack_received(tp, th, CC_DUPACK); 2575 if ((tp->t_flags & TF_SACK_PERMIT) && 2576 IN_FASTRECOVERY(tp->t_flags)) { 2577 int awnd; 2578 2579 /* 2580 * Compute the amount of data in flight first. 2581 * We can inject new data into the pipe iff 2582 * we have less than 1/2 the original window's 2583 * worth of data in flight. 2584 */ 2585 if (V_tcp_do_rfc6675_pipe) 2586 awnd = tcp_compute_pipe(tp); 2587 else 2588 awnd = (tp->snd_nxt - tp->snd_fack) + 2589 tp->sackhint.sack_bytes_rexmit; 2590 2591 if (awnd < tp->snd_ssthresh) { 2592 tp->snd_cwnd += maxseg; 2593 if (tp->snd_cwnd > tp->snd_ssthresh) 2594 tp->snd_cwnd = tp->snd_ssthresh; 2595 } 2596 } else 2597 tp->snd_cwnd += maxseg; 2598 (void) tp->t_fb->tfb_tcp_output(tp); 2599 goto drop; 2600 } else if (tp->t_dupacks == tcprexmtthresh) { 2601 tcp_seq onxt = tp->snd_nxt; 2602 2603 /* 2604 * If we're doing sack, check to 2605 * see if we're already in sack 2606 * recovery. If we're not doing sack, 2607 * check to see if we're in newreno 2608 * recovery. 2609 */ 2610 if (tp->t_flags & TF_SACK_PERMIT) { 2611 if (IN_FASTRECOVERY(tp->t_flags)) { 2612 tp->t_dupacks = 0; 2613 break; 2614 } 2615 } else { 2616 if (SEQ_LEQ(th->th_ack, 2617 tp->snd_recover)) { 2618 tp->t_dupacks = 0; 2619 break; 2620 } 2621 } 2622 /* Congestion signal before ack. */ 2623 cc_cong_signal(tp, th, CC_NDUPACK); 2624 cc_ack_received(tp, th, CC_DUPACK); 2625 tcp_timer_activate(tp, TT_REXMT, 0); 2626 tp->t_rtttime = 0; 2627 if (tp->t_flags & TF_SACK_PERMIT) { 2628 TCPSTAT_INC( 2629 tcps_sack_recovery_episode); 2630 tp->sack_newdata = tp->snd_nxt; 2631 tp->snd_cwnd = maxseg; 2632 (void) tp->t_fb->tfb_tcp_output(tp); 2633 goto drop; 2634 } 2635 tp->snd_nxt = th->th_ack; 2636 tp->snd_cwnd = maxseg; 2637 (void) tp->t_fb->tfb_tcp_output(tp); 2638 KASSERT(tp->snd_limited <= 2, 2639 ("%s: tp->snd_limited too big", 2640 __func__)); 2641 tp->snd_cwnd = tp->snd_ssthresh + 2642 maxseg * 2643 (tp->t_dupacks - tp->snd_limited); 2644 if (SEQ_GT(onxt, tp->snd_nxt)) 2645 tp->snd_nxt = onxt; 2646 goto drop; 2647 } else if (V_tcp_do_rfc3042) { 2648 /* 2649 * Process first and second duplicate 2650 * ACKs. Each indicates a segment 2651 * leaving the network, creating room 2652 * for more. Make sure we can send a 2653 * packet on reception of each duplicate 2654 * ACK by increasing snd_cwnd by one 2655 * segment. Restore the original 2656 * snd_cwnd after packet transmission. 2657 */ 2658 cc_ack_received(tp, th, CC_DUPACK); 2659 u_long oldcwnd = tp->snd_cwnd; 2660 tcp_seq oldsndmax = tp->snd_max; 2661 u_int sent; 2662 int avail; 2663 2664 KASSERT(tp->t_dupacks == 1 || 2665 tp->t_dupacks == 2, 2666 ("%s: dupacks not 1 or 2", 2667 __func__)); 2668 if (tp->t_dupacks == 1) 2669 tp->snd_limited = 0; 2670 tp->snd_cwnd = 2671 (tp->snd_nxt - tp->snd_una) + 2672 (tp->t_dupacks - tp->snd_limited) * 2673 maxseg; 2674 /* 2675 * Only call tcp_output when there 2676 * is new data available to be sent. 2677 * Otherwise we would send pure ACKs. 2678 */ 2679 SOCKBUF_LOCK(&so->so_snd); 2680 avail = sbavail(&so->so_snd) - 2681 (tp->snd_nxt - tp->snd_una); 2682 SOCKBUF_UNLOCK(&so->so_snd); 2683 if (avail > 0) 2684 (void) tp->t_fb->tfb_tcp_output(tp); 2685 sent = tp->snd_max - oldsndmax; 2686 if (sent > maxseg) { 2687 KASSERT((tp->t_dupacks == 2 && 2688 tp->snd_limited == 0) || 2689 (sent == maxseg + 1 && 2690 tp->t_flags & TF_SENTFIN), 2691 ("%s: sent too much", 2692 __func__)); 2693 tp->snd_limited = 2; 2694 } else if (sent > 0) 2695 ++tp->snd_limited; 2696 tp->snd_cwnd = oldcwnd; 2697 goto drop; 2698 } 2699 } 2700 break; 2701 } else { 2702 /* 2703 * This ack is advancing the left edge, reset the 2704 * counter. 2705 */ 2706 tp->t_dupacks = 0; 2707 /* 2708 * If this ack also has new SACK info, increment the 2709 * counter as per rfc6675. 2710 */ 2711 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2712 tp->t_dupacks++; 2713 } 2714 2715 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2716 ("%s: th_ack <= snd_una", __func__)); 2717 2718 /* 2719 * If the congestion window was inflated to account 2720 * for the other side's cached packets, retract it. 2721 */ 2722 if (IN_FASTRECOVERY(tp->t_flags)) { 2723 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2724 if (tp->t_flags & TF_SACK_PERMIT) 2725 tcp_sack_partialack(tp, th); 2726 else 2727 tcp_newreno_partial_ack(tp, th); 2728 } else 2729 cc_post_recovery(tp, th); 2730 } 2731 /* 2732 * If we reach this point, ACK is not a duplicate, 2733 * i.e., it ACKs something we sent. 2734 */ 2735 if (tp->t_flags & TF_NEEDSYN) { 2736 /* 2737 * T/TCP: Connection was half-synchronized, and our 2738 * SYN has been ACK'd (so connection is now fully 2739 * synchronized). Go to non-starred state, 2740 * increment snd_una for ACK of SYN, and check if 2741 * we can do window scaling. 2742 */ 2743 tp->t_flags &= ~TF_NEEDSYN; 2744 tp->snd_una++; 2745 /* Do window scaling? */ 2746 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2747 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2748 tp->rcv_scale = tp->request_r_scale; 2749 /* Send window already scaled. */ 2750 } 2751 } 2752 2753 process_ACK: 2754 INP_WLOCK_ASSERT(tp->t_inpcb); 2755 2756 acked = BYTES_THIS_ACK(tp, th); 2757 TCPSTAT_INC(tcps_rcvackpack); 2758 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2759 2760 /* 2761 * If we just performed our first retransmit, and the ACK 2762 * arrives within our recovery window, then it was a mistake 2763 * to do the retransmit in the first place. Recover our 2764 * original cwnd and ssthresh, and proceed to transmit where 2765 * we left off. 2766 */ 2767 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2768 (int)(ticks - tp->t_badrxtwin) < 0) 2769 cc_cong_signal(tp, th, CC_RTO_ERR); 2770 2771 /* 2772 * If we have a timestamp reply, update smoothed 2773 * round trip time. If no timestamp is present but 2774 * transmit timer is running and timed sequence 2775 * number was acked, update smoothed round trip time. 2776 * Since we now have an rtt measurement, cancel the 2777 * timer backoff (cf., Phil Karn's retransmit alg.). 2778 * Recompute the initial retransmit timer. 2779 * 2780 * Some boxes send broken timestamp replies 2781 * during the SYN+ACK phase, ignore 2782 * timestamps of 0 or we could calculate a 2783 * huge RTT and blow up the retransmit timer. 2784 */ 2785 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2786 u_int t; 2787 2788 t = tcp_ts_getticks() - to.to_tsecr; 2789 if (!tp->t_rttlow || tp->t_rttlow > t) 2790 tp->t_rttlow = t; 2791 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2792 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2793 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2794 tp->t_rttlow = ticks - tp->t_rtttime; 2795 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2796 } 2797 2798 /* 2799 * If all outstanding data is acked, stop retransmit 2800 * timer and remember to restart (more output or persist). 2801 * If there is more data to be acked, restart retransmit 2802 * timer, using current (possibly backed-off) value. 2803 */ 2804 if (th->th_ack == tp->snd_max) { 2805 tcp_timer_activate(tp, TT_REXMT, 0); 2806 needoutput = 1; 2807 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2808 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2809 2810 /* 2811 * If no data (only SYN) was ACK'd, 2812 * skip rest of ACK processing. 2813 */ 2814 if (acked == 0) 2815 goto step6; 2816 2817 /* 2818 * Let the congestion control algorithm update congestion 2819 * control related information. This typically means increasing 2820 * the congestion window. 2821 */ 2822 cc_ack_received(tp, th, CC_ACK); 2823 2824 SOCKBUF_LOCK(&so->so_snd); 2825 if (acked > sbavail(&so->so_snd)) { 2826 tp->snd_wnd -= sbavail(&so->so_snd); 2827 mfree = sbcut_locked(&so->so_snd, 2828 (int)sbavail(&so->so_snd)); 2829 ourfinisacked = 1; 2830 } else { 2831 mfree = sbcut_locked(&so->so_snd, acked); 2832 tp->snd_wnd -= acked; 2833 ourfinisacked = 0; 2834 } 2835 /* NB: sowwakeup_locked() does an implicit unlock. */ 2836 sowwakeup_locked(so); 2837 m_freem(mfree); 2838 /* Detect una wraparound. */ 2839 if (!IN_RECOVERY(tp->t_flags) && 2840 SEQ_GT(tp->snd_una, tp->snd_recover) && 2841 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2842 tp->snd_recover = th->th_ack - 1; 2843 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2844 if (IN_RECOVERY(tp->t_flags) && 2845 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2846 EXIT_RECOVERY(tp->t_flags); 2847 } 2848 tp->snd_una = th->th_ack; 2849 if (tp->t_flags & TF_SACK_PERMIT) { 2850 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2851 tp->snd_recover = tp->snd_una; 2852 } 2853 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2854 tp->snd_nxt = tp->snd_una; 2855 2856 switch (tp->t_state) { 2857 2858 /* 2859 * In FIN_WAIT_1 STATE in addition to the processing 2860 * for the ESTABLISHED state if our FIN is now acknowledged 2861 * then enter FIN_WAIT_2. 2862 */ 2863 case TCPS_FIN_WAIT_1: 2864 if (ourfinisacked) { 2865 /* 2866 * If we can't receive any more 2867 * data, then closing user can proceed. 2868 * Starting the timer is contrary to the 2869 * specification, but if we don't get a FIN 2870 * we'll hang forever. 2871 * 2872 * XXXjl: 2873 * we should release the tp also, and use a 2874 * compressed state. 2875 */ 2876 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2877 soisdisconnected(so); 2878 tcp_timer_activate(tp, TT_2MSL, 2879 (tcp_fast_finwait2_recycle ? 2880 tcp_finwait2_timeout : 2881 TP_MAXIDLE(tp))); 2882 } 2883 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2884 } 2885 break; 2886 2887 /* 2888 * In CLOSING STATE in addition to the processing for 2889 * the ESTABLISHED state if the ACK acknowledges our FIN 2890 * then enter the TIME-WAIT state, otherwise ignore 2891 * the segment. 2892 */ 2893 case TCPS_CLOSING: 2894 if (ourfinisacked) { 2895 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2896 tcp_twstart(tp); 2897 INP_INFO_RUNLOCK(&V_tcbinfo); 2898 m_freem(m); 2899 return; 2900 } 2901 break; 2902 2903 /* 2904 * In LAST_ACK, we may still be waiting for data to drain 2905 * and/or to be acked, as well as for the ack of our FIN. 2906 * If our FIN is now acknowledged, delete the TCB, 2907 * enter the closed state and return. 2908 */ 2909 case TCPS_LAST_ACK: 2910 if (ourfinisacked) { 2911 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2912 tp = tcp_close(tp); 2913 goto drop; 2914 } 2915 break; 2916 } 2917 } 2918 2919 step6: 2920 INP_WLOCK_ASSERT(tp->t_inpcb); 2921 2922 /* 2923 * Update window information. 2924 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2925 */ 2926 if ((thflags & TH_ACK) && 2927 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2928 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2929 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2930 /* keep track of pure window updates */ 2931 if (tlen == 0 && 2932 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2933 TCPSTAT_INC(tcps_rcvwinupd); 2934 tp->snd_wnd = tiwin; 2935 tp->snd_wl1 = th->th_seq; 2936 tp->snd_wl2 = th->th_ack; 2937 if (tp->snd_wnd > tp->max_sndwnd) 2938 tp->max_sndwnd = tp->snd_wnd; 2939 needoutput = 1; 2940 } 2941 2942 /* 2943 * Process segments with URG. 2944 */ 2945 if ((thflags & TH_URG) && th->th_urp && 2946 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2947 /* 2948 * This is a kludge, but if we receive and accept 2949 * random urgent pointers, we'll crash in 2950 * soreceive. It's hard to imagine someone 2951 * actually wanting to send this much urgent data. 2952 */ 2953 SOCKBUF_LOCK(&so->so_rcv); 2954 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2955 th->th_urp = 0; /* XXX */ 2956 thflags &= ~TH_URG; /* XXX */ 2957 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2958 goto dodata; /* XXX */ 2959 } 2960 /* 2961 * If this segment advances the known urgent pointer, 2962 * then mark the data stream. This should not happen 2963 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2964 * a FIN has been received from the remote side. 2965 * In these states we ignore the URG. 2966 * 2967 * According to RFC961 (Assigned Protocols), 2968 * the urgent pointer points to the last octet 2969 * of urgent data. We continue, however, 2970 * to consider it to indicate the first octet 2971 * of data past the urgent section as the original 2972 * spec states (in one of two places). 2973 */ 2974 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2975 tp->rcv_up = th->th_seq + th->th_urp; 2976 so->so_oobmark = sbavail(&so->so_rcv) + 2977 (tp->rcv_up - tp->rcv_nxt) - 1; 2978 if (so->so_oobmark == 0) 2979 so->so_rcv.sb_state |= SBS_RCVATMARK; 2980 sohasoutofband(so); 2981 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2982 } 2983 SOCKBUF_UNLOCK(&so->so_rcv); 2984 /* 2985 * Remove out of band data so doesn't get presented to user. 2986 * This can happen independent of advancing the URG pointer, 2987 * but if two URG's are pending at once, some out-of-band 2988 * data may creep in... ick. 2989 */ 2990 if (th->th_urp <= (u_long)tlen && 2991 !(so->so_options & SO_OOBINLINE)) { 2992 /* hdr drop is delayed */ 2993 tcp_pulloutofband(so, th, m, drop_hdrlen); 2994 } 2995 } else { 2996 /* 2997 * If no out of band data is expected, 2998 * pull receive urgent pointer along 2999 * with the receive window. 3000 */ 3001 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3002 tp->rcv_up = tp->rcv_nxt; 3003 } 3004 dodata: /* XXX */ 3005 INP_WLOCK_ASSERT(tp->t_inpcb); 3006 3007 /* 3008 * Process the segment text, merging it into the TCP sequencing queue, 3009 * and arranging for acknowledgment of receipt if necessary. 3010 * This process logically involves adjusting tp->rcv_wnd as data 3011 * is presented to the user (this happens in tcp_usrreq.c, 3012 * case PRU_RCVD). If a FIN has already been received on this 3013 * connection then we just ignore the text. 3014 */ 3015 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3016 (tp->t_flags & TF_FASTOPEN)); 3017 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3018 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3019 tcp_seq save_start = th->th_seq; 3020 m_adj(m, drop_hdrlen); /* delayed header drop */ 3021 /* 3022 * Insert segment which includes th into TCP reassembly queue 3023 * with control block tp. Set thflags to whether reassembly now 3024 * includes a segment with FIN. This handles the common case 3025 * inline (segment is the next to be received on an established 3026 * connection, and the queue is empty), avoiding linkage into 3027 * and removal from the queue and repetition of various 3028 * conversions. 3029 * Set DELACK for segments received in order, but ack 3030 * immediately when segments are out of order (so 3031 * fast retransmit can work). 3032 */ 3033 if (th->th_seq == tp->rcv_nxt && 3034 LIST_EMPTY(&tp->t_segq) && 3035 (TCPS_HAVEESTABLISHED(tp->t_state) || 3036 tfo_syn)) { 3037 if (DELAY_ACK(tp, tlen) || tfo_syn) 3038 tp->t_flags |= TF_DELACK; 3039 else 3040 tp->t_flags |= TF_ACKNOW; 3041 tp->rcv_nxt += tlen; 3042 thflags = th->th_flags & TH_FIN; 3043 TCPSTAT_INC(tcps_rcvpack); 3044 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3045 SOCKBUF_LOCK(&so->so_rcv); 3046 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3047 m_freem(m); 3048 else 3049 sbappendstream_locked(&so->so_rcv, m, 0); 3050 /* NB: sorwakeup_locked() does an implicit unlock. */ 3051 sorwakeup_locked(so); 3052 } else { 3053 /* 3054 * XXX: Due to the header drop above "th" is 3055 * theoretically invalid by now. Fortunately 3056 * m_adj() doesn't actually frees any mbufs 3057 * when trimming from the head. 3058 */ 3059 thflags = tcp_reass(tp, th, &tlen, m); 3060 tp->t_flags |= TF_ACKNOW; 3061 } 3062 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3063 tcp_update_sack_list(tp, save_start, save_start + tlen); 3064 #if 0 3065 /* 3066 * Note the amount of data that peer has sent into 3067 * our window, in order to estimate the sender's 3068 * buffer size. 3069 * XXX: Unused. 3070 */ 3071 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3072 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3073 else 3074 len = so->so_rcv.sb_hiwat; 3075 #endif 3076 } else { 3077 m_freem(m); 3078 thflags &= ~TH_FIN; 3079 } 3080 3081 /* 3082 * If FIN is received ACK the FIN and let the user know 3083 * that the connection is closing. 3084 */ 3085 if (thflags & TH_FIN) { 3086 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3087 socantrcvmore(so); 3088 /* 3089 * If connection is half-synchronized 3090 * (ie NEEDSYN flag on) then delay ACK, 3091 * so it may be piggybacked when SYN is sent. 3092 * Otherwise, since we received a FIN then no 3093 * more input can be expected, send ACK now. 3094 */ 3095 if (tp->t_flags & TF_NEEDSYN) 3096 tp->t_flags |= TF_DELACK; 3097 else 3098 tp->t_flags |= TF_ACKNOW; 3099 tp->rcv_nxt++; 3100 } 3101 switch (tp->t_state) { 3102 3103 /* 3104 * In SYN_RECEIVED and ESTABLISHED STATES 3105 * enter the CLOSE_WAIT state. 3106 */ 3107 case TCPS_SYN_RECEIVED: 3108 tp->t_starttime = ticks; 3109 /* FALLTHROUGH */ 3110 case TCPS_ESTABLISHED: 3111 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3112 break; 3113 3114 /* 3115 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3116 * enter the CLOSING state. 3117 */ 3118 case TCPS_FIN_WAIT_1: 3119 tcp_state_change(tp, TCPS_CLOSING); 3120 break; 3121 3122 /* 3123 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3124 * starting the time-wait timer, turning off the other 3125 * standard timers. 3126 */ 3127 case TCPS_FIN_WAIT_2: 3128 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3129 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3130 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3131 ti_locked)); 3132 3133 tcp_twstart(tp); 3134 INP_INFO_RUNLOCK(&V_tcbinfo); 3135 return; 3136 } 3137 } 3138 if (ti_locked == TI_RLOCKED) 3139 INP_INFO_RUNLOCK(&V_tcbinfo); 3140 ti_locked = TI_UNLOCKED; 3141 3142 #ifdef TCPDEBUG 3143 if (so->so_options & SO_DEBUG) 3144 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3145 &tcp_savetcp, 0); 3146 #endif 3147 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 3148 3149 /* 3150 * Return any desired output. 3151 */ 3152 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3153 (void) tp->t_fb->tfb_tcp_output(tp); 3154 3155 check_delack: 3156 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3157 __func__, ti_locked)); 3158 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3159 INP_WLOCK_ASSERT(tp->t_inpcb); 3160 3161 if (tp->t_flags & TF_DELACK) { 3162 tp->t_flags &= ~TF_DELACK; 3163 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3164 } 3165 INP_WUNLOCK(tp->t_inpcb); 3166 return; 3167 3168 dropafterack: 3169 /* 3170 * Generate an ACK dropping incoming segment if it occupies 3171 * sequence space, where the ACK reflects our state. 3172 * 3173 * We can now skip the test for the RST flag since all 3174 * paths to this code happen after packets containing 3175 * RST have been dropped. 3176 * 3177 * In the SYN-RECEIVED state, don't send an ACK unless the 3178 * segment we received passes the SYN-RECEIVED ACK test. 3179 * If it fails send a RST. This breaks the loop in the 3180 * "LAND" DoS attack, and also prevents an ACK storm 3181 * between two listening ports that have been sent forged 3182 * SYN segments, each with the source address of the other. 3183 */ 3184 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3185 (SEQ_GT(tp->snd_una, th->th_ack) || 3186 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3187 rstreason = BANDLIM_RST_OPENPORT; 3188 goto dropwithreset; 3189 } 3190 #ifdef TCPDEBUG 3191 if (so->so_options & SO_DEBUG) 3192 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3193 &tcp_savetcp, 0); 3194 #endif 3195 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 3196 if (ti_locked == TI_RLOCKED) 3197 INP_INFO_RUNLOCK(&V_tcbinfo); 3198 ti_locked = TI_UNLOCKED; 3199 3200 tp->t_flags |= TF_ACKNOW; 3201 (void) tp->t_fb->tfb_tcp_output(tp); 3202 INP_WUNLOCK(tp->t_inpcb); 3203 m_freem(m); 3204 return; 3205 3206 dropwithreset: 3207 if (ti_locked == TI_RLOCKED) 3208 INP_INFO_RUNLOCK(&V_tcbinfo); 3209 ti_locked = TI_UNLOCKED; 3210 3211 if (tp != NULL) { 3212 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3213 INP_WUNLOCK(tp->t_inpcb); 3214 } else 3215 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3216 return; 3217 3218 drop: 3219 if (ti_locked == TI_RLOCKED) { 3220 INP_INFO_RUNLOCK(&V_tcbinfo); 3221 ti_locked = TI_UNLOCKED; 3222 } 3223 #ifdef INVARIANTS 3224 else 3225 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3226 #endif 3227 3228 /* 3229 * Drop space held by incoming segment and return. 3230 */ 3231 #ifdef TCPDEBUG 3232 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3233 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3234 &tcp_savetcp, 0); 3235 #endif 3236 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 3237 if (tp != NULL) 3238 INP_WUNLOCK(tp->t_inpcb); 3239 m_freem(m); 3240 } 3241 3242 /* 3243 * Issue RST and make ACK acceptable to originator of segment. 3244 * The mbuf must still include the original packet header. 3245 * tp may be NULL. 3246 */ 3247 void 3248 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3249 int tlen, int rstreason) 3250 { 3251 #ifdef INET 3252 struct ip *ip; 3253 #endif 3254 #ifdef INET6 3255 struct ip6_hdr *ip6; 3256 #endif 3257 3258 if (tp != NULL) { 3259 INP_WLOCK_ASSERT(tp->t_inpcb); 3260 } 3261 3262 /* Don't bother if destination was broadcast/multicast. */ 3263 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3264 goto drop; 3265 #ifdef INET6 3266 if (mtod(m, struct ip *)->ip_v == 6) { 3267 ip6 = mtod(m, struct ip6_hdr *); 3268 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3269 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3270 goto drop; 3271 /* IPv6 anycast check is done at tcp6_input() */ 3272 } 3273 #endif 3274 #if defined(INET) && defined(INET6) 3275 else 3276 #endif 3277 #ifdef INET 3278 { 3279 ip = mtod(m, struct ip *); 3280 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3281 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3282 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3283 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3284 goto drop; 3285 } 3286 #endif 3287 3288 /* Perform bandwidth limiting. */ 3289 if (badport_bandlim(rstreason) < 0) 3290 goto drop; 3291 3292 /* tcp_respond consumes the mbuf chain. */ 3293 if (th->th_flags & TH_ACK) { 3294 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3295 th->th_ack, TH_RST); 3296 } else { 3297 if (th->th_flags & TH_SYN) 3298 tlen++; 3299 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3300 (tcp_seq)0, TH_RST|TH_ACK); 3301 } 3302 return; 3303 drop: 3304 m_freem(m); 3305 } 3306 3307 /* 3308 * Parse TCP options and place in tcpopt. 3309 */ 3310 void 3311 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3312 { 3313 int opt, optlen; 3314 3315 to->to_flags = 0; 3316 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3317 opt = cp[0]; 3318 if (opt == TCPOPT_EOL) 3319 break; 3320 if (opt == TCPOPT_NOP) 3321 optlen = 1; 3322 else { 3323 if (cnt < 2) 3324 break; 3325 optlen = cp[1]; 3326 if (optlen < 2 || optlen > cnt) 3327 break; 3328 } 3329 switch (opt) { 3330 case TCPOPT_MAXSEG: 3331 if (optlen != TCPOLEN_MAXSEG) 3332 continue; 3333 if (!(flags & TO_SYN)) 3334 continue; 3335 to->to_flags |= TOF_MSS; 3336 bcopy((char *)cp + 2, 3337 (char *)&to->to_mss, sizeof(to->to_mss)); 3338 to->to_mss = ntohs(to->to_mss); 3339 break; 3340 case TCPOPT_WINDOW: 3341 if (optlen != TCPOLEN_WINDOW) 3342 continue; 3343 if (!(flags & TO_SYN)) 3344 continue; 3345 to->to_flags |= TOF_SCALE; 3346 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3347 break; 3348 case TCPOPT_TIMESTAMP: 3349 if (optlen != TCPOLEN_TIMESTAMP) 3350 continue; 3351 to->to_flags |= TOF_TS; 3352 bcopy((char *)cp + 2, 3353 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3354 to->to_tsval = ntohl(to->to_tsval); 3355 bcopy((char *)cp + 6, 3356 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3357 to->to_tsecr = ntohl(to->to_tsecr); 3358 break; 3359 #ifdef TCP_SIGNATURE 3360 /* 3361 * XXX In order to reply to a host which has set the 3362 * TCP_SIGNATURE option in its initial SYN, we have to 3363 * record the fact that the option was observed here 3364 * for the syncache code to perform the correct response. 3365 */ 3366 case TCPOPT_SIGNATURE: 3367 if (optlen != TCPOLEN_SIGNATURE) 3368 continue; 3369 to->to_flags |= TOF_SIGNATURE; 3370 to->to_signature = cp + 2; 3371 break; 3372 #endif 3373 case TCPOPT_SACK_PERMITTED: 3374 if (optlen != TCPOLEN_SACK_PERMITTED) 3375 continue; 3376 if (!(flags & TO_SYN)) 3377 continue; 3378 if (!V_tcp_do_sack) 3379 continue; 3380 to->to_flags |= TOF_SACKPERM; 3381 break; 3382 case TCPOPT_SACK: 3383 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3384 continue; 3385 if (flags & TO_SYN) 3386 continue; 3387 to->to_flags |= TOF_SACK; 3388 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3389 to->to_sacks = cp + 2; 3390 TCPSTAT_INC(tcps_sack_rcv_blocks); 3391 break; 3392 #ifdef TCP_RFC7413 3393 case TCPOPT_FAST_OPEN: 3394 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3395 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3396 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3397 continue; 3398 if (!(flags & TO_SYN)) 3399 continue; 3400 if (!V_tcp_fastopen_enabled) 3401 continue; 3402 to->to_flags |= TOF_FASTOPEN; 3403 to->to_tfo_len = optlen - 2; 3404 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3405 break; 3406 #endif 3407 default: 3408 continue; 3409 } 3410 } 3411 } 3412 3413 /* 3414 * Pull out of band byte out of a segment so 3415 * it doesn't appear in the user's data queue. 3416 * It is still reflected in the segment length for 3417 * sequencing purposes. 3418 */ 3419 void 3420 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3421 int off) 3422 { 3423 int cnt = off + th->th_urp - 1; 3424 3425 while (cnt >= 0) { 3426 if (m->m_len > cnt) { 3427 char *cp = mtod(m, caddr_t) + cnt; 3428 struct tcpcb *tp = sototcpcb(so); 3429 3430 INP_WLOCK_ASSERT(tp->t_inpcb); 3431 3432 tp->t_iobc = *cp; 3433 tp->t_oobflags |= TCPOOB_HAVEDATA; 3434 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3435 m->m_len--; 3436 if (m->m_flags & M_PKTHDR) 3437 m->m_pkthdr.len--; 3438 return; 3439 } 3440 cnt -= m->m_len; 3441 m = m->m_next; 3442 if (m == NULL) 3443 break; 3444 } 3445 panic("tcp_pulloutofband"); 3446 } 3447 3448 /* 3449 * Collect new round-trip time estimate 3450 * and update averages and current timeout. 3451 */ 3452 void 3453 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3454 { 3455 int delta; 3456 3457 INP_WLOCK_ASSERT(tp->t_inpcb); 3458 3459 TCPSTAT_INC(tcps_rttupdated); 3460 tp->t_rttupdated++; 3461 if (tp->t_srtt != 0) { 3462 /* 3463 * srtt is stored as fixed point with 5 bits after the 3464 * binary point (i.e., scaled by 8). The following magic 3465 * is equivalent to the smoothing algorithm in rfc793 with 3466 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3467 * point). Adjust rtt to origin 0. 3468 */ 3469 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3470 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3471 3472 if ((tp->t_srtt += delta) <= 0) 3473 tp->t_srtt = 1; 3474 3475 /* 3476 * We accumulate a smoothed rtt variance (actually, a 3477 * smoothed mean difference), then set the retransmit 3478 * timer to smoothed rtt + 4 times the smoothed variance. 3479 * rttvar is stored as fixed point with 4 bits after the 3480 * binary point (scaled by 16). The following is 3481 * equivalent to rfc793 smoothing with an alpha of .75 3482 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3483 * rfc793's wired-in beta. 3484 */ 3485 if (delta < 0) 3486 delta = -delta; 3487 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3488 if ((tp->t_rttvar += delta) <= 0) 3489 tp->t_rttvar = 1; 3490 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3491 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3492 } else { 3493 /* 3494 * No rtt measurement yet - use the unsmoothed rtt. 3495 * Set the variance to half the rtt (so our first 3496 * retransmit happens at 3*rtt). 3497 */ 3498 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3499 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3500 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3501 } 3502 tp->t_rtttime = 0; 3503 tp->t_rxtshift = 0; 3504 3505 /* 3506 * the retransmit should happen at rtt + 4 * rttvar. 3507 * Because of the way we do the smoothing, srtt and rttvar 3508 * will each average +1/2 tick of bias. When we compute 3509 * the retransmit timer, we want 1/2 tick of rounding and 3510 * 1 extra tick because of +-1/2 tick uncertainty in the 3511 * firing of the timer. The bias will give us exactly the 3512 * 1.5 tick we need. But, because the bias is 3513 * statistical, we have to test that we don't drop below 3514 * the minimum feasible timer (which is 2 ticks). 3515 */ 3516 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3517 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3518 3519 /* 3520 * We received an ack for a packet that wasn't retransmitted; 3521 * it is probably safe to discard any error indications we've 3522 * received recently. This isn't quite right, but close enough 3523 * for now (a route might have failed after we sent a segment, 3524 * and the return path might not be symmetrical). 3525 */ 3526 tp->t_softerror = 0; 3527 } 3528 3529 /* 3530 * Determine a reasonable value for maxseg size. 3531 * If the route is known, check route for mtu. 3532 * If none, use an mss that can be handled on the outgoing interface 3533 * without forcing IP to fragment. If no route is found, route has no mtu, 3534 * or the destination isn't local, use a default, hopefully conservative 3535 * size (usually 512 or the default IP max size, but no more than the mtu 3536 * of the interface), as we can't discover anything about intervening 3537 * gateways or networks. We also initialize the congestion/slow start 3538 * window to be a single segment if the destination isn't local. 3539 * While looking at the routing entry, we also initialize other path-dependent 3540 * parameters from pre-set or cached values in the routing entry. 3541 * 3542 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3543 * IP options, e.g. IPSEC data, since length of this data may vary, and 3544 * thus it is calculated for every segment separately in tcp_output(). 3545 * 3546 * NOTE that this routine is only called when we process an incoming 3547 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3548 * settings are handled in tcp_mssopt(). 3549 */ 3550 void 3551 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3552 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3553 { 3554 int mss = 0; 3555 u_long maxmtu = 0; 3556 struct inpcb *inp = tp->t_inpcb; 3557 struct hc_metrics_lite metrics; 3558 #ifdef INET6 3559 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3560 size_t min_protoh = isipv6 ? 3561 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3562 sizeof (struct tcpiphdr); 3563 #else 3564 const size_t min_protoh = sizeof(struct tcpiphdr); 3565 #endif 3566 3567 INP_WLOCK_ASSERT(tp->t_inpcb); 3568 3569 if (mtuoffer != -1) { 3570 KASSERT(offer == -1, ("%s: conflict", __func__)); 3571 offer = mtuoffer - min_protoh; 3572 } 3573 3574 /* Initialize. */ 3575 #ifdef INET6 3576 if (isipv6) { 3577 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3578 tp->t_maxseg = V_tcp_v6mssdflt; 3579 } 3580 #endif 3581 #if defined(INET) && defined(INET6) 3582 else 3583 #endif 3584 #ifdef INET 3585 { 3586 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3587 tp->t_maxseg = V_tcp_mssdflt; 3588 } 3589 #endif 3590 3591 /* 3592 * No route to sender, stay with default mss and return. 3593 */ 3594 if (maxmtu == 0) { 3595 /* 3596 * In case we return early we need to initialize metrics 3597 * to a defined state as tcp_hc_get() would do for us 3598 * if there was no cache hit. 3599 */ 3600 if (metricptr != NULL) 3601 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3602 return; 3603 } 3604 3605 /* What have we got? */ 3606 switch (offer) { 3607 case 0: 3608 /* 3609 * Offer == 0 means that there was no MSS on the SYN 3610 * segment, in this case we use tcp_mssdflt as 3611 * already assigned to t_maxseg above. 3612 */ 3613 offer = tp->t_maxseg; 3614 break; 3615 3616 case -1: 3617 /* 3618 * Offer == -1 means that we didn't receive SYN yet. 3619 */ 3620 /* FALLTHROUGH */ 3621 3622 default: 3623 /* 3624 * Prevent DoS attack with too small MSS. Round up 3625 * to at least minmss. 3626 */ 3627 offer = max(offer, V_tcp_minmss); 3628 } 3629 3630 /* 3631 * rmx information is now retrieved from tcp_hostcache. 3632 */ 3633 tcp_hc_get(&inp->inp_inc, &metrics); 3634 if (metricptr != NULL) 3635 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3636 3637 /* 3638 * If there's a discovered mtu in tcp hostcache, use it. 3639 * Else, use the link mtu. 3640 */ 3641 if (metrics.rmx_mtu) 3642 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3643 else { 3644 #ifdef INET6 3645 if (isipv6) { 3646 mss = maxmtu - min_protoh; 3647 if (!V_path_mtu_discovery && 3648 !in6_localaddr(&inp->in6p_faddr)) 3649 mss = min(mss, V_tcp_v6mssdflt); 3650 } 3651 #endif 3652 #if defined(INET) && defined(INET6) 3653 else 3654 #endif 3655 #ifdef INET 3656 { 3657 mss = maxmtu - min_protoh; 3658 if (!V_path_mtu_discovery && 3659 !in_localaddr(inp->inp_faddr)) 3660 mss = min(mss, V_tcp_mssdflt); 3661 } 3662 #endif 3663 /* 3664 * XXX - The above conditional (mss = maxmtu - min_protoh) 3665 * probably violates the TCP spec. 3666 * The problem is that, since we don't know the 3667 * other end's MSS, we are supposed to use a conservative 3668 * default. But, if we do that, then MTU discovery will 3669 * never actually take place, because the conservative 3670 * default is much less than the MTUs typically seen 3671 * on the Internet today. For the moment, we'll sweep 3672 * this under the carpet. 3673 * 3674 * The conservative default might not actually be a problem 3675 * if the only case this occurs is when sending an initial 3676 * SYN with options and data to a host we've never talked 3677 * to before. Then, they will reply with an MSS value which 3678 * will get recorded and the new parameters should get 3679 * recomputed. For Further Study. 3680 */ 3681 } 3682 mss = min(mss, offer); 3683 3684 /* 3685 * Sanity check: make sure that maxseg will be large 3686 * enough to allow some data on segments even if the 3687 * all the option space is used (40bytes). Otherwise 3688 * funny things may happen in tcp_output. 3689 * 3690 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3691 */ 3692 mss = max(mss, 64); 3693 3694 tp->t_maxseg = mss; 3695 } 3696 3697 void 3698 tcp_mss(struct tcpcb *tp, int offer) 3699 { 3700 int mss; 3701 u_long bufsize; 3702 struct inpcb *inp; 3703 struct socket *so; 3704 struct hc_metrics_lite metrics; 3705 struct tcp_ifcap cap; 3706 3707 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3708 3709 bzero(&cap, sizeof(cap)); 3710 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3711 3712 mss = tp->t_maxseg; 3713 inp = tp->t_inpcb; 3714 3715 /* 3716 * If there's a pipesize, change the socket buffer to that size, 3717 * don't change if sb_hiwat is different than default (then it 3718 * has been changed on purpose with setsockopt). 3719 * Make the socket buffers an integral number of mss units; 3720 * if the mss is larger than the socket buffer, decrease the mss. 3721 */ 3722 so = inp->inp_socket; 3723 SOCKBUF_LOCK(&so->so_snd); 3724 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3725 bufsize = metrics.rmx_sendpipe; 3726 else 3727 bufsize = so->so_snd.sb_hiwat; 3728 if (bufsize < mss) 3729 mss = bufsize; 3730 else { 3731 bufsize = roundup(bufsize, mss); 3732 if (bufsize > sb_max) 3733 bufsize = sb_max; 3734 if (bufsize > so->so_snd.sb_hiwat) 3735 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3736 } 3737 SOCKBUF_UNLOCK(&so->so_snd); 3738 tp->t_maxseg = mss; 3739 3740 SOCKBUF_LOCK(&so->so_rcv); 3741 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3742 bufsize = metrics.rmx_recvpipe; 3743 else 3744 bufsize = so->so_rcv.sb_hiwat; 3745 if (bufsize > mss) { 3746 bufsize = roundup(bufsize, mss); 3747 if (bufsize > sb_max) 3748 bufsize = sb_max; 3749 if (bufsize > so->so_rcv.sb_hiwat) 3750 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3751 } 3752 SOCKBUF_UNLOCK(&so->so_rcv); 3753 3754 /* Check the interface for TSO capabilities. */ 3755 if (cap.ifcap & CSUM_TSO) { 3756 tp->t_flags |= TF_TSO; 3757 tp->t_tsomax = cap.tsomax; 3758 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3759 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3760 } 3761 } 3762 3763 /* 3764 * Determine the MSS option to send on an outgoing SYN. 3765 */ 3766 int 3767 tcp_mssopt(struct in_conninfo *inc) 3768 { 3769 int mss = 0; 3770 u_long maxmtu = 0; 3771 u_long thcmtu = 0; 3772 size_t min_protoh; 3773 3774 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3775 3776 #ifdef INET6 3777 if (inc->inc_flags & INC_ISIPV6) { 3778 mss = V_tcp_v6mssdflt; 3779 maxmtu = tcp_maxmtu6(inc, NULL); 3780 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3781 } 3782 #endif 3783 #if defined(INET) && defined(INET6) 3784 else 3785 #endif 3786 #ifdef INET 3787 { 3788 mss = V_tcp_mssdflt; 3789 maxmtu = tcp_maxmtu(inc, NULL); 3790 min_protoh = sizeof(struct tcpiphdr); 3791 } 3792 #endif 3793 #if defined(INET6) || defined(INET) 3794 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3795 #endif 3796 3797 if (maxmtu && thcmtu) 3798 mss = min(maxmtu, thcmtu) - min_protoh; 3799 else if (maxmtu || thcmtu) 3800 mss = max(maxmtu, thcmtu) - min_protoh; 3801 3802 return (mss); 3803 } 3804 3805 3806 /* 3807 * On a partial ack arrives, force the retransmission of the 3808 * next unacknowledged segment. Do not clear tp->t_dupacks. 3809 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3810 * be started again. 3811 */ 3812 void 3813 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3814 { 3815 tcp_seq onxt = tp->snd_nxt; 3816 u_long ocwnd = tp->snd_cwnd; 3817 u_int maxseg = tcp_maxseg(tp); 3818 3819 INP_WLOCK_ASSERT(tp->t_inpcb); 3820 3821 tcp_timer_activate(tp, TT_REXMT, 0); 3822 tp->t_rtttime = 0; 3823 tp->snd_nxt = th->th_ack; 3824 /* 3825 * Set snd_cwnd to one segment beyond acknowledged offset. 3826 * (tp->snd_una has not yet been updated when this function is called.) 3827 */ 3828 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3829 tp->t_flags |= TF_ACKNOW; 3830 (void) tp->t_fb->tfb_tcp_output(tp); 3831 tp->snd_cwnd = ocwnd; 3832 if (SEQ_GT(onxt, tp->snd_nxt)) 3833 tp->snd_nxt = onxt; 3834 /* 3835 * Partial window deflation. Relies on fact that tp->snd_una 3836 * not updated yet. 3837 */ 3838 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3839 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3840 else 3841 tp->snd_cwnd = 0; 3842 tp->snd_cwnd += maxseg; 3843 } 3844 3845 int 3846 tcp_compute_pipe(struct tcpcb *tp) 3847 { 3848 return (tp->snd_max - tp->snd_una + 3849 tp->sackhint.sack_bytes_rexmit - 3850 tp->sackhint.sacked_bytes); 3851 } 3852