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_inet.h" 54 #include "opt_inet6.h" 55 #include "opt_ipsec.h" 56 #include "opt_tcpdebug.h" 57 58 #include <sys/param.h> 59 #include <sys/kernel.h> 60 #ifdef TCP_HHOOK 61 #include <sys/hhook.h> 62 #endif 63 #include <sys/malloc.h> 64 #include <sys/mbuf.h> 65 #include <sys/proc.h> /* for proc0 declaration */ 66 #include <sys/protosw.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77 #include <vm/uma.h> 78 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/route.h> 82 #include <net/vnet.h> 83 84 #define TCPSTATES /* for logging */ 85 86 #include <netinet/in.h> 87 #include <netinet/in_kdtrace.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/ip.h> 91 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 92 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 93 #include <netinet/ip_var.h> 94 #include <netinet/ip_options.h> 95 #include <netinet/ip6.h> 96 #include <netinet/icmp6.h> 97 #include <netinet6/in6_pcb.h> 98 #include <netinet6/in6_var.h> 99 #include <netinet6/ip6_var.h> 100 #include <netinet6/nd6.h> 101 #ifdef TCP_RFC7413 102 #include <netinet/tcp_fastopen.h> 103 #endif 104 #include <netinet/tcp.h> 105 #include <netinet/tcp_fsm.h> 106 #include <netinet/tcp_seq.h> 107 #include <netinet/tcp_timer.h> 108 #include <netinet/tcp_var.h> 109 #include <netinet6/tcp6_var.h> 110 #include <netinet/tcpip.h> 111 #include <netinet/cc/cc.h> 112 #ifdef TCPPCAP 113 #include <netinet/tcp_pcap.h> 114 #endif 115 #include <netinet/tcp_syncache.h> 116 #ifdef TCPDEBUG 117 #include <netinet/tcp_debug.h> 118 #endif /* TCPDEBUG */ 119 #ifdef TCP_OFFLOAD 120 #include <netinet/tcp_offload.h> 121 #endif 122 123 #ifdef IPSEC 124 #include <netipsec/ipsec.h> 125 #include <netipsec/ipsec6.h> 126 #endif /*IPSEC*/ 127 128 #include <machine/in_cksum.h> 129 130 #include <security/mac/mac_framework.h> 131 132 const int tcprexmtthresh = 3; 133 134 int tcp_log_in_vain = 0; 135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 136 &tcp_log_in_vain, 0, 137 "Log all incoming TCP segments to closed ports"); 138 139 VNET_DEFINE(int, blackhole) = 0; 140 #define V_blackhole VNET(blackhole) 141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 142 &VNET_NAME(blackhole), 0, 143 "Do not send RST on segments to closed ports"); 144 145 VNET_DEFINE(int, tcp_delack_enabled) = 1; 146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 147 &VNET_NAME(tcp_delack_enabled), 0, 148 "Delay ACK to try and piggyback it onto a data packet"); 149 150 VNET_DEFINE(int, drop_synfin) = 0; 151 #define V_drop_synfin VNET(drop_synfin) 152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 153 &VNET_NAME(drop_synfin), 0, 154 "Drop TCP packets with SYN+FIN set"); 155 156 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 158 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 159 "Use calculated pipe/in-flight bytes per RFC 6675"); 160 161 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 162 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_rfc3042), 0, 165 "Enable RFC 3042 (Limited Transmit)"); 166 167 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_rfc3390), 0, 170 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 171 172 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 174 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 175 "Slow-start flight size (initial congestion window) in number of segments"); 176 177 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 179 &VNET_NAME(tcp_do_rfc3465), 0, 180 "Enable RFC 3465 (Appropriate Byte Counting)"); 181 182 VNET_DEFINE(int, tcp_abc_l_var) = 2; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 184 &VNET_NAME(tcp_abc_l_var), 2, 185 "Cap the max cwnd increment during slow-start to this number of segments"); 186 187 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 188 189 VNET_DEFINE(int, tcp_do_ecn) = 2; 190 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 191 &VNET_NAME(tcp_do_ecn), 0, 192 "TCP ECN support"); 193 194 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 195 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 196 &VNET_NAME(tcp_ecn_maxretries), 0, 197 "Max retries before giving up on ECN"); 198 199 VNET_DEFINE(int, tcp_insecure_syn) = 0; 200 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 202 &VNET_NAME(tcp_insecure_syn), 0, 203 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 204 205 VNET_DEFINE(int, tcp_insecure_rst) = 0; 206 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_insecure_rst), 0, 209 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 210 211 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 212 #define V_tcp_recvspace VNET(tcp_recvspace) 213 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 214 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 215 216 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 217 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 219 &VNET_NAME(tcp_do_autorcvbuf), 0, 220 "Enable automatic receive buffer sizing"); 221 222 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 223 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 225 &VNET_NAME(tcp_autorcvbuf_inc), 0, 226 "Incrementor step size of automatic receive buffer"); 227 228 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 229 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 231 &VNET_NAME(tcp_autorcvbuf_max), 0, 232 "Max size of automatic receive buffer"); 233 234 VNET_DEFINE(struct inpcbhead, tcb); 235 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 236 VNET_DEFINE(struct inpcbinfo, tcbinfo); 237 238 /* 239 * TCP statistics are stored in an array of counter(9)s, which size matches 240 * size of struct tcpstat. TCP running connection count is a regular array. 241 */ 242 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 243 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 244 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 245 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 246 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 247 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 248 "TCP connection counts by TCP state"); 249 250 static void 251 tcp_vnet_init(const void *unused) 252 { 253 254 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 255 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 256 } 257 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 258 tcp_vnet_init, NULL); 259 260 #ifdef VIMAGE 261 static void 262 tcp_vnet_uninit(const void *unused) 263 { 264 265 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 266 VNET_PCPUSTAT_FREE(tcpstat); 267 } 268 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 269 tcp_vnet_uninit, NULL); 270 #endif /* VIMAGE */ 271 272 /* 273 * Kernel module interface for updating tcpstat. The argument is an index 274 * into tcpstat treated as an array. 275 */ 276 void 277 kmod_tcpstat_inc(int statnum) 278 { 279 280 counter_u64_add(VNET(tcpstat)[statnum], 1); 281 } 282 283 #ifdef TCP_HHOOK 284 /* 285 * Wrapper for the TCP established input helper hook. 286 */ 287 void 288 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 289 { 290 struct tcp_hhook_data hhook_data; 291 292 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 293 hhook_data.tp = tp; 294 hhook_data.th = th; 295 hhook_data.to = to; 296 297 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 298 tp->osd); 299 } 300 } 301 #endif 302 303 /* 304 * CC wrapper hook functions 305 */ 306 void 307 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 308 uint16_t type) 309 { 310 INP_WLOCK_ASSERT(tp->t_inpcb); 311 312 tp->ccv->nsegs = nsegs; 313 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 314 if (tp->snd_cwnd <= tp->snd_wnd) 315 tp->ccv->flags |= CCF_CWND_LIMITED; 316 else 317 tp->ccv->flags &= ~CCF_CWND_LIMITED; 318 319 if (type == CC_ACK) { 320 if (tp->snd_cwnd > tp->snd_ssthresh) { 321 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 322 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 323 if (tp->t_bytes_acked >= tp->snd_cwnd) { 324 tp->t_bytes_acked -= tp->snd_cwnd; 325 tp->ccv->flags |= CCF_ABC_SENTAWND; 326 } 327 } else { 328 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 329 tp->t_bytes_acked = 0; 330 } 331 } 332 333 if (CC_ALGO(tp)->ack_received != NULL) { 334 /* XXXLAS: Find a way to live without this */ 335 tp->ccv->curack = th->th_ack; 336 CC_ALGO(tp)->ack_received(tp->ccv, type); 337 } 338 } 339 340 void 341 cc_conn_init(struct tcpcb *tp) 342 { 343 struct hc_metrics_lite metrics; 344 struct inpcb *inp = tp->t_inpcb; 345 u_int maxseg; 346 int rtt; 347 348 INP_WLOCK_ASSERT(tp->t_inpcb); 349 350 tcp_hc_get(&inp->inp_inc, &metrics); 351 maxseg = tcp_maxseg(tp); 352 353 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 354 tp->t_srtt = rtt; 355 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 356 TCPSTAT_INC(tcps_usedrtt); 357 if (metrics.rmx_rttvar) { 358 tp->t_rttvar = metrics.rmx_rttvar; 359 TCPSTAT_INC(tcps_usedrttvar); 360 } else { 361 /* default variation is +- 1 rtt */ 362 tp->t_rttvar = 363 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 364 } 365 TCPT_RANGESET(tp->t_rxtcur, 366 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 367 tp->t_rttmin, TCPTV_REXMTMAX); 368 } 369 if (metrics.rmx_ssthresh) { 370 /* 371 * There's some sort of gateway or interface 372 * buffer limit on the path. Use this to set 373 * the slow start threshold, but set the 374 * threshold to no less than 2*mss. 375 */ 376 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 377 TCPSTAT_INC(tcps_usedssthresh); 378 } 379 380 /* 381 * Set the initial slow-start flight size. 382 * 383 * RFC5681 Section 3.1 specifies the default conservative values. 384 * RFC3390 specifies slightly more aggressive values. 385 * RFC6928 increases it to ten segments. 386 * Support for user specified value for initial flight size. 387 * 388 * If a SYN or SYN/ACK was lost and retransmitted, we have to 389 * reduce the initial CWND to one segment as congestion is likely 390 * requiring us to be cautious. 391 */ 392 if (tp->snd_cwnd == 1) 393 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 394 else if (V_tcp_initcwnd_segments) 395 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 396 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 397 else if (V_tcp_do_rfc3390) 398 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 399 else { 400 /* Per RFC5681 Section 3.1 */ 401 if (maxseg > 2190) 402 tp->snd_cwnd = 2 * maxseg; 403 else if (maxseg > 1095) 404 tp->snd_cwnd = 3 * maxseg; 405 else 406 tp->snd_cwnd = 4 * maxseg; 407 } 408 409 if (CC_ALGO(tp)->conn_init != NULL) 410 CC_ALGO(tp)->conn_init(tp->ccv); 411 } 412 413 void inline 414 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 415 { 416 u_int maxseg; 417 418 INP_WLOCK_ASSERT(tp->t_inpcb); 419 420 switch(type) { 421 case CC_NDUPACK: 422 if (!IN_FASTRECOVERY(tp->t_flags)) { 423 tp->snd_recover = tp->snd_max; 424 if (tp->t_flags & TF_ECN_PERMIT) 425 tp->t_flags |= TF_ECN_SND_CWR; 426 } 427 break; 428 case CC_ECN: 429 if (!IN_CONGRECOVERY(tp->t_flags)) { 430 TCPSTAT_INC(tcps_ecn_rcwnd); 431 tp->snd_recover = tp->snd_max; 432 if (tp->t_flags & TF_ECN_PERMIT) 433 tp->t_flags |= TF_ECN_SND_CWR; 434 } 435 break; 436 case CC_RTO: 437 maxseg = tcp_maxseg(tp); 438 tp->t_dupacks = 0; 439 tp->t_bytes_acked = 0; 440 EXIT_RECOVERY(tp->t_flags); 441 if (CC_ALGO(tp)->cong_signal == NULL) { 442 /* 443 * RFC5681 Section 3.1 444 * ssthresh = max (FlightSize / 2, 2*SMSS) eq (4) 445 */ 446 tp->snd_ssthresh = 447 max((tp->snd_max - tp->snd_una) / 2 / maxseg, 2) 448 * maxseg; 449 tp->snd_cwnd = maxseg; 450 } 451 break; 452 case CC_RTO_ERR: 453 TCPSTAT_INC(tcps_sndrexmitbad); 454 /* RTO was unnecessary, so reset everything. */ 455 tp->snd_cwnd = tp->snd_cwnd_prev; 456 tp->snd_ssthresh = tp->snd_ssthresh_prev; 457 tp->snd_recover = tp->snd_recover_prev; 458 if (tp->t_flags & TF_WASFRECOVERY) 459 ENTER_FASTRECOVERY(tp->t_flags); 460 if (tp->t_flags & TF_WASCRECOVERY) 461 ENTER_CONGRECOVERY(tp->t_flags); 462 tp->snd_nxt = tp->snd_max; 463 tp->t_flags &= ~TF_PREVVALID; 464 tp->t_badrxtwin = 0; 465 break; 466 } 467 468 if (CC_ALGO(tp)->cong_signal != NULL) { 469 if (th != NULL) 470 tp->ccv->curack = th->th_ack; 471 CC_ALGO(tp)->cong_signal(tp->ccv, type); 472 } 473 } 474 475 void inline 476 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 477 { 478 INP_WLOCK_ASSERT(tp->t_inpcb); 479 480 /* XXXLAS: KASSERT that we're in recovery? */ 481 482 if (CC_ALGO(tp)->post_recovery != NULL) { 483 tp->ccv->curack = th->th_ack; 484 CC_ALGO(tp)->post_recovery(tp->ccv); 485 } 486 /* XXXLAS: EXIT_RECOVERY ? */ 487 tp->t_bytes_acked = 0; 488 } 489 490 #ifdef TCP_SIGNATURE 491 static inline int 492 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 493 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 494 { 495 int ret; 496 497 tcp_fields_to_net(th); 498 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 499 tcp_fields_to_host(th); 500 return (ret); 501 } 502 #endif 503 504 /* 505 * Indicate whether this ack should be delayed. We can delay the ack if 506 * following conditions are met: 507 * - There is no delayed ack timer in progress. 508 * - Our last ack wasn't a 0-sized window. We never want to delay 509 * the ack that opens up a 0-sized window. 510 * - LRO wasn't used for this segment. We make sure by checking that the 511 * segment size is not larger than the MSS. 512 */ 513 #define DELAY_ACK(tp, tlen) \ 514 ((!tcp_timer_active(tp, TT_DELACK) && \ 515 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 516 (tlen <= tp->t_maxseg) && \ 517 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 518 519 static void inline 520 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 521 { 522 INP_WLOCK_ASSERT(tp->t_inpcb); 523 524 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 525 switch (iptos & IPTOS_ECN_MASK) { 526 case IPTOS_ECN_CE: 527 tp->ccv->flags |= CCF_IPHDR_CE; 528 break; 529 case IPTOS_ECN_ECT0: 530 tp->ccv->flags &= ~CCF_IPHDR_CE; 531 break; 532 case IPTOS_ECN_ECT1: 533 tp->ccv->flags &= ~CCF_IPHDR_CE; 534 break; 535 } 536 537 if (th->th_flags & TH_CWR) 538 tp->ccv->flags |= CCF_TCPHDR_CWR; 539 else 540 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 541 542 if (tp->t_flags & TF_DELACK) 543 tp->ccv->flags |= CCF_DELACK; 544 else 545 tp->ccv->flags &= ~CCF_DELACK; 546 547 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 548 549 if (tp->ccv->flags & CCF_ACKNOW) 550 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 551 } 552 } 553 554 /* 555 * TCP input handling is split into multiple parts: 556 * tcp6_input is a thin wrapper around tcp_input for the extended 557 * ip6_protox[] call format in ip6_input 558 * tcp_input handles primary segment validation, inpcb lookup and 559 * SYN processing on listen sockets 560 * tcp_do_segment processes the ACK and text of the segment for 561 * establishing, established and closing connections 562 */ 563 #ifdef INET6 564 int 565 tcp6_input(struct mbuf **mp, int *offp, int proto) 566 { 567 struct mbuf *m = *mp; 568 struct in6_ifaddr *ia6; 569 struct ip6_hdr *ip6; 570 571 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 572 573 /* 574 * draft-itojun-ipv6-tcp-to-anycast 575 * better place to put this in? 576 */ 577 ip6 = mtod(m, struct ip6_hdr *); 578 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 579 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 580 struct ip6_hdr *ip6; 581 582 ifa_free(&ia6->ia_ifa); 583 ip6 = mtod(m, struct ip6_hdr *); 584 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 585 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 586 return (IPPROTO_DONE); 587 } 588 if (ia6) 589 ifa_free(&ia6->ia_ifa); 590 591 return (tcp_input(mp, offp, proto)); 592 } 593 #endif /* INET6 */ 594 595 int 596 tcp_input(struct mbuf **mp, int *offp, int proto) 597 { 598 struct mbuf *m = *mp; 599 struct tcphdr *th = NULL; 600 struct ip *ip = NULL; 601 struct inpcb *inp = NULL; 602 struct tcpcb *tp = NULL; 603 struct socket *so = NULL; 604 u_char *optp = NULL; 605 int off0; 606 int optlen = 0; 607 #ifdef INET 608 int len; 609 #endif 610 int tlen = 0, off; 611 int drop_hdrlen; 612 int thflags; 613 int rstreason = 0; /* For badport_bandlim accounting purposes */ 614 #ifdef TCP_SIGNATURE 615 uint8_t sig_checked = 0; 616 #endif 617 uint8_t iptos; 618 struct m_tag *fwd_tag = NULL; 619 #ifdef INET6 620 struct ip6_hdr *ip6 = NULL; 621 int isipv6; 622 #else 623 const void *ip6 = NULL; 624 #endif /* INET6 */ 625 struct tcpopt to; /* options in this segment */ 626 char *s = NULL; /* address and port logging */ 627 int ti_locked; 628 #ifdef TCPDEBUG 629 /* 630 * The size of tcp_saveipgen must be the size of the max ip header, 631 * now IPv6. 632 */ 633 u_char tcp_saveipgen[IP6_HDR_LEN]; 634 struct tcphdr tcp_savetcp; 635 short ostate = 0; 636 #endif 637 638 #ifdef INET6 639 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 640 #endif 641 642 off0 = *offp; 643 m = *mp; 644 *mp = NULL; 645 to.to_flags = 0; 646 TCPSTAT_INC(tcps_rcvtotal); 647 648 #ifdef INET6 649 if (isipv6) { 650 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 651 652 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 653 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 654 if (m == NULL) { 655 TCPSTAT_INC(tcps_rcvshort); 656 return (IPPROTO_DONE); 657 } 658 } 659 660 ip6 = mtod(m, struct ip6_hdr *); 661 th = (struct tcphdr *)((caddr_t)ip6 + off0); 662 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 663 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 664 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 665 th->th_sum = m->m_pkthdr.csum_data; 666 else 667 th->th_sum = in6_cksum_pseudo(ip6, tlen, 668 IPPROTO_TCP, m->m_pkthdr.csum_data); 669 th->th_sum ^= 0xffff; 670 } else 671 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 672 if (th->th_sum) { 673 TCPSTAT_INC(tcps_rcvbadsum); 674 goto drop; 675 } 676 677 /* 678 * Be proactive about unspecified IPv6 address in source. 679 * As we use all-zero to indicate unbounded/unconnected pcb, 680 * unspecified IPv6 address can be used to confuse us. 681 * 682 * Note that packets with unspecified IPv6 destination is 683 * already dropped in ip6_input. 684 */ 685 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 686 /* XXX stat */ 687 goto drop; 688 } 689 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 690 } 691 #endif 692 #if defined(INET) && defined(INET6) 693 else 694 #endif 695 #ifdef INET 696 { 697 /* 698 * Get IP and TCP header together in first mbuf. 699 * Note: IP leaves IP header in first mbuf. 700 */ 701 if (off0 > sizeof (struct ip)) { 702 ip_stripoptions(m); 703 off0 = sizeof(struct ip); 704 } 705 if (m->m_len < sizeof (struct tcpiphdr)) { 706 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 707 == NULL) { 708 TCPSTAT_INC(tcps_rcvshort); 709 return (IPPROTO_DONE); 710 } 711 } 712 ip = mtod(m, struct ip *); 713 th = (struct tcphdr *)((caddr_t)ip + off0); 714 tlen = ntohs(ip->ip_len) - off0; 715 716 iptos = ip->ip_tos; 717 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 718 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 719 th->th_sum = m->m_pkthdr.csum_data; 720 else 721 th->th_sum = in_pseudo(ip->ip_src.s_addr, 722 ip->ip_dst.s_addr, 723 htonl(m->m_pkthdr.csum_data + tlen + 724 IPPROTO_TCP)); 725 th->th_sum ^= 0xffff; 726 } else { 727 struct ipovly *ipov = (struct ipovly *)ip; 728 729 /* 730 * Checksum extended TCP header and data. 731 */ 732 len = off0 + tlen; 733 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 734 ipov->ih_len = htons(tlen); 735 th->th_sum = in_cksum(m, len); 736 /* Reset length for SDT probes. */ 737 ip->ip_len = htons(len); 738 /* Reset TOS bits */ 739 ip->ip_tos = iptos; 740 /* Re-initialization for later version check */ 741 ip->ip_v = IPVERSION; 742 } 743 744 if (th->th_sum) { 745 TCPSTAT_INC(tcps_rcvbadsum); 746 goto drop; 747 } 748 } 749 #endif /* INET */ 750 751 /* 752 * Check that TCP offset makes sense, 753 * pull out TCP options and adjust length. XXX 754 */ 755 off = th->th_off << 2; 756 if (off < sizeof (struct tcphdr) || off > tlen) { 757 TCPSTAT_INC(tcps_rcvbadoff); 758 goto drop; 759 } 760 tlen -= off; /* tlen is used instead of ti->ti_len */ 761 if (off > sizeof (struct tcphdr)) { 762 #ifdef INET6 763 if (isipv6) { 764 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 765 ip6 = mtod(m, struct ip6_hdr *); 766 th = (struct tcphdr *)((caddr_t)ip6 + off0); 767 } 768 #endif 769 #if defined(INET) && defined(INET6) 770 else 771 #endif 772 #ifdef INET 773 { 774 if (m->m_len < sizeof(struct ip) + off) { 775 if ((m = m_pullup(m, sizeof (struct ip) + off)) 776 == NULL) { 777 TCPSTAT_INC(tcps_rcvshort); 778 return (IPPROTO_DONE); 779 } 780 ip = mtod(m, struct ip *); 781 th = (struct tcphdr *)((caddr_t)ip + off0); 782 } 783 } 784 #endif 785 optlen = off - sizeof (struct tcphdr); 786 optp = (u_char *)(th + 1); 787 } 788 thflags = th->th_flags; 789 790 /* 791 * Convert TCP protocol specific fields to host format. 792 */ 793 tcp_fields_to_host(th); 794 795 /* 796 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 797 */ 798 drop_hdrlen = off0 + off; 799 800 /* 801 * Locate pcb for segment; if we're likely to add or remove a 802 * connection then first acquire pcbinfo lock. There are three cases 803 * where we might discover later we need a write lock despite the 804 * flags: ACKs moving a connection out of the syncache, ACKs for a 805 * connection in TIMEWAIT and SYNs not targeting a listening socket. 806 */ 807 if ((thflags & (TH_FIN | TH_RST)) != 0) { 808 INP_INFO_RLOCK(&V_tcbinfo); 809 ti_locked = TI_RLOCKED; 810 } else 811 ti_locked = TI_UNLOCKED; 812 813 /* 814 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 815 */ 816 if ( 817 #ifdef INET6 818 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 819 #ifdef INET 820 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 821 #endif 822 #endif 823 #if defined(INET) && !defined(INET6) 824 (m->m_flags & M_IP_NEXTHOP) 825 #endif 826 ) 827 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 828 829 findpcb: 830 #ifdef INVARIANTS 831 if (ti_locked == TI_RLOCKED) { 832 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 833 } else { 834 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 835 } 836 #endif 837 #ifdef INET6 838 if (isipv6 && fwd_tag != NULL) { 839 struct sockaddr_in6 *next_hop6; 840 841 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 842 /* 843 * Transparently forwarded. Pretend to be the destination. 844 * Already got one like this? 845 */ 846 inp = in6_pcblookup_mbuf(&V_tcbinfo, 847 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 848 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 849 if (!inp) { 850 /* 851 * It's new. Try to find the ambushing socket. 852 * Because we've rewritten the destination address, 853 * any hardware-generated hash is ignored. 854 */ 855 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 856 th->th_sport, &next_hop6->sin6_addr, 857 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 858 th->th_dport, INPLOOKUP_WILDCARD | 859 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 860 } 861 } else if (isipv6) { 862 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 863 th->th_sport, &ip6->ip6_dst, th->th_dport, 864 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 865 m->m_pkthdr.rcvif, m); 866 } 867 #endif /* INET6 */ 868 #if defined(INET6) && defined(INET) 869 else 870 #endif 871 #ifdef INET 872 if (fwd_tag != NULL) { 873 struct sockaddr_in *next_hop; 874 875 next_hop = (struct sockaddr_in *)(fwd_tag+1); 876 /* 877 * Transparently forwarded. Pretend to be the destination. 878 * already got one like this? 879 */ 880 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 881 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 882 m->m_pkthdr.rcvif, m); 883 if (!inp) { 884 /* 885 * It's new. Try to find the ambushing socket. 886 * Because we've rewritten the destination address, 887 * any hardware-generated hash is ignored. 888 */ 889 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 890 th->th_sport, next_hop->sin_addr, 891 next_hop->sin_port ? ntohs(next_hop->sin_port) : 892 th->th_dport, INPLOOKUP_WILDCARD | 893 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 894 } 895 } else 896 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 897 th->th_sport, ip->ip_dst, th->th_dport, 898 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 899 m->m_pkthdr.rcvif, m); 900 #endif /* INET */ 901 902 /* 903 * If the INPCB does not exist then all data in the incoming 904 * segment is discarded and an appropriate RST is sent back. 905 * XXX MRT Send RST using which routing table? 906 */ 907 if (inp == NULL) { 908 /* 909 * Log communication attempts to ports that are not 910 * in use. 911 */ 912 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 913 tcp_log_in_vain == 2) { 914 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 915 log(LOG_INFO, "%s; %s: Connection attempt " 916 "to closed port\n", s, __func__); 917 } 918 /* 919 * When blackholing do not respond with a RST but 920 * completely ignore the segment and drop it. 921 */ 922 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 923 V_blackhole == 2) 924 goto dropunlock; 925 926 rstreason = BANDLIM_RST_CLOSEDPORT; 927 goto dropwithreset; 928 } 929 INP_WLOCK_ASSERT(inp); 930 /* 931 * While waiting for inp lock during the lookup, another thread 932 * can have dropped the inpcb, in which case we need to loop back 933 * and try to find a new inpcb to deliver to. 934 */ 935 if (inp->inp_flags & INP_DROPPED) { 936 INP_WUNLOCK(inp); 937 inp = NULL; 938 goto findpcb; 939 } 940 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 941 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 942 ((inp->inp_socket == NULL) || 943 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 944 inp->inp_flowid = m->m_pkthdr.flowid; 945 inp->inp_flowtype = M_HASHTYPE_GET(m); 946 } 947 #ifdef IPSEC 948 #ifdef INET6 949 if (isipv6 && ipsec6_in_reject(m, inp)) { 950 goto dropunlock; 951 } else 952 #endif /* INET6 */ 953 if (ipsec4_in_reject(m, inp) != 0) { 954 goto dropunlock; 955 } 956 #endif /* IPSEC */ 957 958 /* 959 * Check the minimum TTL for socket. 960 */ 961 if (inp->inp_ip_minttl != 0) { 962 #ifdef INET6 963 if (isipv6) { 964 if (inp->inp_ip_minttl > ip6->ip6_hlim) 965 goto dropunlock; 966 } else 967 #endif 968 if (inp->inp_ip_minttl > ip->ip_ttl) 969 goto dropunlock; 970 } 971 972 /* 973 * A previous connection in TIMEWAIT state is supposed to catch stray 974 * or duplicate segments arriving late. If this segment was a 975 * legitimate new connection attempt, the old INPCB gets removed and 976 * we can try again to find a listening socket. 977 * 978 * At this point, due to earlier optimism, we may hold only an inpcb 979 * lock, and not the inpcbinfo write lock. If so, we need to try to 980 * acquire it, or if that fails, acquire a reference on the inpcb, 981 * drop all locks, acquire a global write lock, and then re-acquire 982 * the inpcb lock. We may at that point discover that another thread 983 * has tried to free the inpcb, in which case we need to loop back 984 * and try to find a new inpcb to deliver to. 985 * 986 * XXXRW: It may be time to rethink timewait locking. 987 */ 988 relocked: 989 if (inp->inp_flags & INP_TIMEWAIT) { 990 if (ti_locked == TI_UNLOCKED) { 991 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 992 in_pcbref(inp); 993 INP_WUNLOCK(inp); 994 INP_INFO_RLOCK(&V_tcbinfo); 995 ti_locked = TI_RLOCKED; 996 INP_WLOCK(inp); 997 if (in_pcbrele_wlocked(inp)) { 998 inp = NULL; 999 goto findpcb; 1000 } else if (inp->inp_flags & INP_DROPPED) { 1001 INP_WUNLOCK(inp); 1002 inp = NULL; 1003 goto findpcb; 1004 } 1005 } else 1006 ti_locked = TI_RLOCKED; 1007 } 1008 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1009 1010 if (thflags & TH_SYN) 1011 tcp_dooptions(&to, optp, optlen, TO_SYN); 1012 /* 1013 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1014 */ 1015 if (tcp_twcheck(inp, &to, th, m, tlen)) 1016 goto findpcb; 1017 INP_INFO_RUNLOCK(&V_tcbinfo); 1018 return (IPPROTO_DONE); 1019 } 1020 /* 1021 * The TCPCB may no longer exist if the connection is winding 1022 * down or it is in the CLOSED state. Either way we drop the 1023 * segment and send an appropriate response. 1024 */ 1025 tp = intotcpcb(inp); 1026 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1027 rstreason = BANDLIM_RST_CLOSEDPORT; 1028 goto dropwithreset; 1029 } 1030 1031 #ifdef TCP_OFFLOAD 1032 if (tp->t_flags & TF_TOE) { 1033 tcp_offload_input(tp, m); 1034 m = NULL; /* consumed by the TOE driver */ 1035 goto dropunlock; 1036 } 1037 #endif 1038 1039 /* 1040 * We've identified a valid inpcb, but it could be that we need an 1041 * inpcbinfo write lock but don't hold it. In this case, attempt to 1042 * acquire using the same strategy as the TIMEWAIT case above. If we 1043 * relock, we have to jump back to 'relocked' as the connection might 1044 * now be in TIMEWAIT. 1045 */ 1046 #ifdef INVARIANTS 1047 if ((thflags & (TH_FIN | TH_RST)) != 0) 1048 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1049 #endif 1050 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1051 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1052 !IS_FASTOPEN(tp->t_flags)))) { 1053 if (ti_locked == TI_UNLOCKED) { 1054 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1055 in_pcbref(inp); 1056 INP_WUNLOCK(inp); 1057 INP_INFO_RLOCK(&V_tcbinfo); 1058 ti_locked = TI_RLOCKED; 1059 INP_WLOCK(inp); 1060 if (in_pcbrele_wlocked(inp)) { 1061 inp = NULL; 1062 goto findpcb; 1063 } else if (inp->inp_flags & INP_DROPPED) { 1064 INP_WUNLOCK(inp); 1065 inp = NULL; 1066 goto findpcb; 1067 } 1068 goto relocked; 1069 } else 1070 ti_locked = TI_RLOCKED; 1071 } 1072 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1073 } 1074 1075 #ifdef MAC 1076 INP_WLOCK_ASSERT(inp); 1077 if (mac_inpcb_check_deliver(inp, m)) 1078 goto dropunlock; 1079 #endif 1080 so = inp->inp_socket; 1081 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1082 #ifdef TCPDEBUG 1083 if (so->so_options & SO_DEBUG) { 1084 ostate = tp->t_state; 1085 #ifdef INET6 1086 if (isipv6) { 1087 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1088 } else 1089 #endif 1090 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1091 tcp_savetcp = *th; 1092 } 1093 #endif /* TCPDEBUG */ 1094 /* 1095 * When the socket is accepting connections (the INPCB is in LISTEN 1096 * state) we look into the SYN cache if this is a new connection 1097 * attempt or the completion of a previous one. 1098 */ 1099 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1100 ("%s: so accepting but tp %p not listening", __func__, tp)); 1101 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1102 struct in_conninfo inc; 1103 1104 bzero(&inc, sizeof(inc)); 1105 #ifdef INET6 1106 if (isipv6) { 1107 inc.inc_flags |= INC_ISIPV6; 1108 inc.inc6_faddr = ip6->ip6_src; 1109 inc.inc6_laddr = ip6->ip6_dst; 1110 } else 1111 #endif 1112 { 1113 inc.inc_faddr = ip->ip_src; 1114 inc.inc_laddr = ip->ip_dst; 1115 } 1116 inc.inc_fport = th->th_sport; 1117 inc.inc_lport = th->th_dport; 1118 inc.inc_fibnum = so->so_fibnum; 1119 1120 /* 1121 * Check for an existing connection attempt in syncache if 1122 * the flag is only ACK. A successful lookup creates a new 1123 * socket appended to the listen queue in SYN_RECEIVED state. 1124 */ 1125 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1126 1127 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1128 /* 1129 * Parse the TCP options here because 1130 * syncookies need access to the reflected 1131 * timestamp. 1132 */ 1133 tcp_dooptions(&to, optp, optlen, 0); 1134 /* 1135 * NB: syncache_expand() doesn't unlock 1136 * inp and tcpinfo locks. 1137 */ 1138 if (!syncache_expand(&inc, &to, th, &so, m)) { 1139 /* 1140 * No syncache entry or ACK was not 1141 * for our SYN/ACK. Send a RST. 1142 * NB: syncache did its own logging 1143 * of the failure cause. 1144 */ 1145 rstreason = BANDLIM_RST_OPENPORT; 1146 goto dropwithreset; 1147 } 1148 #ifdef TCP_RFC7413 1149 tfo_socket_result: 1150 #endif 1151 if (so == NULL) { 1152 /* 1153 * We completed the 3-way handshake 1154 * but could not allocate a socket 1155 * either due to memory shortage, 1156 * listen queue length limits or 1157 * global socket limits. Send RST 1158 * or wait and have the remote end 1159 * retransmit the ACK for another 1160 * try. 1161 */ 1162 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1163 log(LOG_DEBUG, "%s; %s: Listen socket: " 1164 "Socket allocation failed due to " 1165 "limits or memory shortage, %s\n", 1166 s, __func__, 1167 V_tcp_sc_rst_sock_fail ? 1168 "sending RST" : "try again"); 1169 if (V_tcp_sc_rst_sock_fail) { 1170 rstreason = BANDLIM_UNLIMITED; 1171 goto dropwithreset; 1172 } else 1173 goto dropunlock; 1174 } 1175 /* 1176 * Socket is created in state SYN_RECEIVED. 1177 * Unlock the listen socket, lock the newly 1178 * created socket and update the tp variable. 1179 */ 1180 INP_WUNLOCK(inp); /* listen socket */ 1181 inp = sotoinpcb(so); 1182 /* 1183 * New connection inpcb is already locked by 1184 * syncache_expand(). 1185 */ 1186 INP_WLOCK_ASSERT(inp); 1187 tp = intotcpcb(inp); 1188 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1189 ("%s: ", __func__)); 1190 #ifdef TCP_SIGNATURE 1191 if (sig_checked == 0) { 1192 tcp_dooptions(&to, optp, optlen, 1193 (thflags & TH_SYN) ? TO_SYN : 0); 1194 if (!tcp_signature_verify_input(m, off0, tlen, 1195 optlen, &to, th, tp->t_flags)) { 1196 1197 /* 1198 * In SYN_SENT state if it receives an 1199 * RST, it is allowed for further 1200 * processing. 1201 */ 1202 if ((thflags & TH_RST) == 0 || 1203 (tp->t_state == TCPS_SYN_SENT) == 0) 1204 goto dropunlock; 1205 } 1206 sig_checked = 1; 1207 } 1208 #endif 1209 1210 /* 1211 * Process the segment and the data it 1212 * contains. tcp_do_segment() consumes 1213 * the mbuf chain and unlocks the inpcb. 1214 */ 1215 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1216 iptos, ti_locked); 1217 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1218 return (IPPROTO_DONE); 1219 } 1220 /* 1221 * Segment flag validation for new connection attempts: 1222 * 1223 * Our (SYN|ACK) response was rejected. 1224 * Check with syncache and remove entry to prevent 1225 * retransmits. 1226 * 1227 * NB: syncache_chkrst does its own logging of failure 1228 * causes. 1229 */ 1230 if (thflags & TH_RST) { 1231 syncache_chkrst(&inc, th); 1232 goto dropunlock; 1233 } 1234 /* 1235 * We can't do anything without SYN. 1236 */ 1237 if ((thflags & TH_SYN) == 0) { 1238 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1239 log(LOG_DEBUG, "%s; %s: Listen socket: " 1240 "SYN is missing, segment ignored\n", 1241 s, __func__); 1242 TCPSTAT_INC(tcps_badsyn); 1243 goto dropunlock; 1244 } 1245 /* 1246 * (SYN|ACK) is bogus on a listen socket. 1247 */ 1248 if (thflags & TH_ACK) { 1249 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1250 log(LOG_DEBUG, "%s; %s: Listen socket: " 1251 "SYN|ACK invalid, segment rejected\n", 1252 s, __func__); 1253 syncache_badack(&inc); /* XXX: Not needed! */ 1254 TCPSTAT_INC(tcps_badsyn); 1255 rstreason = BANDLIM_RST_OPENPORT; 1256 goto dropwithreset; 1257 } 1258 /* 1259 * If the drop_synfin option is enabled, drop all 1260 * segments with both the SYN and FIN bits set. 1261 * This prevents e.g. nmap from identifying the 1262 * TCP/IP stack. 1263 * XXX: Poor reasoning. nmap has other methods 1264 * and is constantly refining its stack detection 1265 * strategies. 1266 * XXX: This is a violation of the TCP specification 1267 * and was used by RFC1644. 1268 */ 1269 if ((thflags & TH_FIN) && V_drop_synfin) { 1270 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1271 log(LOG_DEBUG, "%s; %s: Listen socket: " 1272 "SYN|FIN segment ignored (based on " 1273 "sysctl setting)\n", s, __func__); 1274 TCPSTAT_INC(tcps_badsyn); 1275 goto dropunlock; 1276 } 1277 /* 1278 * Segment's flags are (SYN) or (SYN|FIN). 1279 * 1280 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1281 * as they do not affect the state of the TCP FSM. 1282 * The data pointed to by TH_URG and th_urp is ignored. 1283 */ 1284 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1285 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1286 KASSERT(thflags & (TH_SYN), 1287 ("%s: Listen socket: TH_SYN not set", __func__)); 1288 #ifdef INET6 1289 /* 1290 * If deprecated address is forbidden, 1291 * we do not accept SYN to deprecated interface 1292 * address to prevent any new inbound connection from 1293 * getting established. 1294 * When we do not accept SYN, we send a TCP RST, 1295 * with deprecated source address (instead of dropping 1296 * it). We compromise it as it is much better for peer 1297 * to send a RST, and RST will be the final packet 1298 * for the exchange. 1299 * 1300 * If we do not forbid deprecated addresses, we accept 1301 * the SYN packet. RFC2462 does not suggest dropping 1302 * SYN in this case. 1303 * If we decipher RFC2462 5.5.4, it says like this: 1304 * 1. use of deprecated addr with existing 1305 * communication is okay - "SHOULD continue to be 1306 * used" 1307 * 2. use of it with new communication: 1308 * (2a) "SHOULD NOT be used if alternate address 1309 * with sufficient scope is available" 1310 * (2b) nothing mentioned otherwise. 1311 * Here we fall into (2b) case as we have no choice in 1312 * our source address selection - we must obey the peer. 1313 * 1314 * The wording in RFC2462 is confusing, and there are 1315 * multiple description text for deprecated address 1316 * handling - worse, they are not exactly the same. 1317 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1318 */ 1319 if (isipv6 && !V_ip6_use_deprecated) { 1320 struct in6_ifaddr *ia6; 1321 1322 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1323 if (ia6 != NULL && 1324 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1325 ifa_free(&ia6->ia_ifa); 1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1327 log(LOG_DEBUG, "%s; %s: Listen socket: " 1328 "Connection attempt to deprecated " 1329 "IPv6 address rejected\n", 1330 s, __func__); 1331 rstreason = BANDLIM_RST_OPENPORT; 1332 goto dropwithreset; 1333 } 1334 if (ia6) 1335 ifa_free(&ia6->ia_ifa); 1336 } 1337 #endif /* INET6 */ 1338 /* 1339 * Basic sanity checks on incoming SYN requests: 1340 * Don't respond if the destination is a link layer 1341 * broadcast according to RFC1122 4.2.3.10, p. 104. 1342 * If it is from this socket it must be forged. 1343 * Don't respond if the source or destination is a 1344 * global or subnet broad- or multicast address. 1345 * Note that it is quite possible to receive unicast 1346 * link-layer packets with a broadcast IP address. Use 1347 * in_broadcast() to find them. 1348 */ 1349 if (m->m_flags & (M_BCAST|M_MCAST)) { 1350 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1351 log(LOG_DEBUG, "%s; %s: Listen socket: " 1352 "Connection attempt from broad- or multicast " 1353 "link layer address ignored\n", s, __func__); 1354 goto dropunlock; 1355 } 1356 #ifdef INET6 1357 if (isipv6) { 1358 if (th->th_dport == th->th_sport && 1359 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1360 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1361 log(LOG_DEBUG, "%s; %s: Listen socket: " 1362 "Connection attempt to/from self " 1363 "ignored\n", s, __func__); 1364 goto dropunlock; 1365 } 1366 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1367 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1368 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1369 log(LOG_DEBUG, "%s; %s: Listen socket: " 1370 "Connection attempt from/to multicast " 1371 "address ignored\n", s, __func__); 1372 goto dropunlock; 1373 } 1374 } 1375 #endif 1376 #if defined(INET) && defined(INET6) 1377 else 1378 #endif 1379 #ifdef INET 1380 { 1381 if (th->th_dport == th->th_sport && 1382 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1383 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1384 log(LOG_DEBUG, "%s; %s: Listen socket: " 1385 "Connection attempt from/to self " 1386 "ignored\n", s, __func__); 1387 goto dropunlock; 1388 } 1389 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1390 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1391 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1392 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1393 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1394 log(LOG_DEBUG, "%s; %s: Listen socket: " 1395 "Connection attempt from/to broad- " 1396 "or multicast address ignored\n", 1397 s, __func__); 1398 goto dropunlock; 1399 } 1400 } 1401 #endif 1402 /* 1403 * SYN appears to be valid. Create compressed TCP state 1404 * for syncache. 1405 */ 1406 #ifdef TCPDEBUG 1407 if (so->so_options & SO_DEBUG) 1408 tcp_trace(TA_INPUT, ostate, tp, 1409 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1410 #endif 1411 TCP_PROBE3(debug__input, tp, th, m); 1412 tcp_dooptions(&to, optp, optlen, TO_SYN); 1413 #ifdef TCP_RFC7413 1414 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1415 goto tfo_socket_result; 1416 #else 1417 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1418 #endif 1419 /* 1420 * Entry added to syncache and mbuf consumed. 1421 * Only the listen socket is unlocked by syncache_add(). 1422 */ 1423 if (ti_locked == TI_RLOCKED) { 1424 INP_INFO_RUNLOCK(&V_tcbinfo); 1425 ti_locked = TI_UNLOCKED; 1426 } 1427 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1428 return (IPPROTO_DONE); 1429 } else if (tp->t_state == TCPS_LISTEN) { 1430 /* 1431 * When a listen socket is torn down the SO_ACCEPTCONN 1432 * flag is removed first while connections are drained 1433 * from the accept queue in a unlock/lock cycle of the 1434 * ACCEPT_LOCK, opening a race condition allowing a SYN 1435 * attempt go through unhandled. 1436 */ 1437 goto dropunlock; 1438 } 1439 1440 #ifdef TCP_SIGNATURE 1441 if (sig_checked == 0) { 1442 tcp_dooptions(&to, optp, optlen, 1443 (thflags & TH_SYN) ? TO_SYN : 0); 1444 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1445 th, tp->t_flags)) { 1446 1447 /* 1448 * In SYN_SENT state if it receives an RST, it is 1449 * allowed for further processing. 1450 */ 1451 if ((thflags & TH_RST) == 0 || 1452 (tp->t_state == TCPS_SYN_SENT) == 0) 1453 goto dropunlock; 1454 } 1455 sig_checked = 1; 1456 } 1457 #endif 1458 1459 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1460 1461 /* 1462 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1463 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1464 * the inpcb, and unlocks pcbinfo. 1465 */ 1466 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1467 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1468 return (IPPROTO_DONE); 1469 1470 dropwithreset: 1471 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1472 1473 if (ti_locked == TI_RLOCKED) { 1474 INP_INFO_RUNLOCK(&V_tcbinfo); 1475 ti_locked = TI_UNLOCKED; 1476 } 1477 #ifdef INVARIANTS 1478 else { 1479 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1480 "ti_locked: %d", __func__, ti_locked)); 1481 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1482 } 1483 #endif 1484 1485 if (inp != NULL) { 1486 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1487 INP_WUNLOCK(inp); 1488 } else 1489 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1490 m = NULL; /* mbuf chain got consumed. */ 1491 goto drop; 1492 1493 dropunlock: 1494 if (m != NULL) 1495 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1496 1497 if (ti_locked == TI_RLOCKED) { 1498 INP_INFO_RUNLOCK(&V_tcbinfo); 1499 ti_locked = TI_UNLOCKED; 1500 } 1501 #ifdef INVARIANTS 1502 else { 1503 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1504 "ti_locked: %d", __func__, ti_locked)); 1505 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1506 } 1507 #endif 1508 1509 if (inp != NULL) 1510 INP_WUNLOCK(inp); 1511 1512 drop: 1513 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1514 if (s != NULL) 1515 free(s, M_TCPLOG); 1516 if (m != NULL) 1517 m_freem(m); 1518 return (IPPROTO_DONE); 1519 } 1520 1521 void 1522 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1523 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1524 int ti_locked) 1525 { 1526 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1527 int rstreason, todrop, win; 1528 uint32_t tiwin; 1529 uint16_t nsegs; 1530 char *s; 1531 struct in_conninfo *inc; 1532 struct mbuf *mfree; 1533 struct tcpopt to; 1534 #ifdef TCP_RFC7413 1535 int tfo_syn; 1536 #endif 1537 1538 #ifdef TCPDEBUG 1539 /* 1540 * The size of tcp_saveipgen must be the size of the max ip header, 1541 * now IPv6. 1542 */ 1543 u_char tcp_saveipgen[IP6_HDR_LEN]; 1544 struct tcphdr tcp_savetcp; 1545 short ostate = 0; 1546 #endif 1547 thflags = th->th_flags; 1548 inc = &tp->t_inpcb->inp_inc; 1549 tp->sackhint.last_sack_ack = 0; 1550 sack_changed = 0; 1551 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1552 1553 /* 1554 * If this is either a state-changing packet or current state isn't 1555 * established, we require a write lock on tcbinfo. Otherwise, we 1556 * allow the tcbinfo to be in either alocked or unlocked, as the 1557 * caller may have unnecessarily acquired a write lock due to a race. 1558 */ 1559 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1560 tp->t_state != TCPS_ESTABLISHED) { 1561 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1562 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1563 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1564 } else { 1565 #ifdef INVARIANTS 1566 if (ti_locked == TI_RLOCKED) 1567 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1568 else { 1569 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1570 "ti_locked: %d", __func__, ti_locked)); 1571 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1572 } 1573 #endif 1574 } 1575 INP_WLOCK_ASSERT(tp->t_inpcb); 1576 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1577 __func__)); 1578 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1579 __func__)); 1580 1581 #ifdef TCPPCAP 1582 /* Save segment, if requested. */ 1583 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1584 #endif 1585 1586 /* 1587 * Segment received on connection. 1588 * Reset idle time and keep-alive timer. 1589 * XXX: This should be done after segment 1590 * validation to ignore broken/spoofed segs. 1591 */ 1592 tp->t_rcvtime = ticks; 1593 1594 /* 1595 * Scale up the window into a 32-bit value. 1596 * For the SYN_SENT state the scale is zero. 1597 */ 1598 tiwin = th->th_win << tp->snd_scale; 1599 1600 /* 1601 * TCP ECN processing. 1602 */ 1603 if (tp->t_flags & TF_ECN_PERMIT) { 1604 if (thflags & TH_CWR) 1605 tp->t_flags &= ~TF_ECN_SND_ECE; 1606 switch (iptos & IPTOS_ECN_MASK) { 1607 case IPTOS_ECN_CE: 1608 tp->t_flags |= TF_ECN_SND_ECE; 1609 TCPSTAT_INC(tcps_ecn_ce); 1610 break; 1611 case IPTOS_ECN_ECT0: 1612 TCPSTAT_INC(tcps_ecn_ect0); 1613 break; 1614 case IPTOS_ECN_ECT1: 1615 TCPSTAT_INC(tcps_ecn_ect1); 1616 break; 1617 } 1618 1619 /* Process a packet differently from RFC3168. */ 1620 cc_ecnpkt_handler(tp, th, iptos); 1621 1622 /* Congestion experienced. */ 1623 if (thflags & TH_ECE) { 1624 cc_cong_signal(tp, th, CC_ECN); 1625 } 1626 } 1627 1628 /* 1629 * Parse options on any incoming segment. 1630 */ 1631 tcp_dooptions(&to, (u_char *)(th + 1), 1632 (th->th_off << 2) - sizeof(struct tcphdr), 1633 (thflags & TH_SYN) ? TO_SYN : 0); 1634 1635 /* 1636 * If echoed timestamp is later than the current time, 1637 * fall back to non RFC1323 RTT calculation. Normalize 1638 * timestamp if syncookies were used when this connection 1639 * was established. 1640 */ 1641 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1642 to.to_tsecr -= tp->ts_offset; 1643 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1644 to.to_tsecr = 0; 1645 } 1646 /* 1647 * If timestamps were negotiated during SYN/ACK they should 1648 * appear on every segment during this session and vice versa. 1649 */ 1650 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1651 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1652 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1653 "no action\n", s, __func__); 1654 free(s, M_TCPLOG); 1655 } 1656 } 1657 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1658 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1659 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1660 "no action\n", s, __func__); 1661 free(s, M_TCPLOG); 1662 } 1663 } 1664 1665 /* 1666 * Process options only when we get SYN/ACK back. The SYN case 1667 * for incoming connections is handled in tcp_syncache. 1668 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1669 * or <SYN,ACK>) segment itself is never scaled. 1670 * XXX this is traditional behavior, may need to be cleaned up. 1671 */ 1672 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1673 if ((to.to_flags & TOF_SCALE) && 1674 (tp->t_flags & TF_REQ_SCALE)) { 1675 tp->t_flags |= TF_RCVD_SCALE; 1676 tp->snd_scale = to.to_wscale; 1677 } 1678 /* 1679 * Initial send window. It will be updated with 1680 * the next incoming segment to the scaled value. 1681 */ 1682 tp->snd_wnd = th->th_win; 1683 if (to.to_flags & TOF_TS) { 1684 tp->t_flags |= TF_RCVD_TSTMP; 1685 tp->ts_recent = to.to_tsval; 1686 tp->ts_recent_age = tcp_ts_getticks(); 1687 } 1688 if (to.to_flags & TOF_MSS) 1689 tcp_mss(tp, to.to_mss); 1690 if ((tp->t_flags & TF_SACK_PERMIT) && 1691 (to.to_flags & TOF_SACKPERM) == 0) 1692 tp->t_flags &= ~TF_SACK_PERMIT; 1693 } 1694 1695 /* 1696 * Header prediction: check for the two common cases 1697 * of a uni-directional data xfer. If the packet has 1698 * no control flags, is in-sequence, the window didn't 1699 * change and we're not retransmitting, it's a 1700 * candidate. If the length is zero and the ack moved 1701 * forward, we're the sender side of the xfer. Just 1702 * free the data acked & wake any higher level process 1703 * that was blocked waiting for space. If the length 1704 * is non-zero and the ack didn't move, we're the 1705 * receiver side. If we're getting packets in-order 1706 * (the reassembly queue is empty), add the data to 1707 * the socket buffer and note that we need a delayed ack. 1708 * Make sure that the hidden state-flags are also off. 1709 * Since we check for TCPS_ESTABLISHED first, it can only 1710 * be TH_NEEDSYN. 1711 */ 1712 if (tp->t_state == TCPS_ESTABLISHED && 1713 th->th_seq == tp->rcv_nxt && 1714 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1715 tp->snd_nxt == tp->snd_max && 1716 tiwin && tiwin == tp->snd_wnd && 1717 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1718 LIST_EMPTY(&tp->t_segq) && 1719 ((to.to_flags & TOF_TS) == 0 || 1720 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1721 1722 /* 1723 * If last ACK falls within this segment's sequence numbers, 1724 * record the timestamp. 1725 * NOTE that the test is modified according to the latest 1726 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1727 */ 1728 if ((to.to_flags & TOF_TS) != 0 && 1729 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1730 tp->ts_recent_age = tcp_ts_getticks(); 1731 tp->ts_recent = to.to_tsval; 1732 } 1733 1734 if (tlen == 0) { 1735 if (SEQ_GT(th->th_ack, tp->snd_una) && 1736 SEQ_LEQ(th->th_ack, tp->snd_max) && 1737 !IN_RECOVERY(tp->t_flags) && 1738 (to.to_flags & TOF_SACK) == 0 && 1739 TAILQ_EMPTY(&tp->snd_holes)) { 1740 /* 1741 * This is a pure ack for outstanding data. 1742 */ 1743 if (ti_locked == TI_RLOCKED) 1744 INP_INFO_RUNLOCK(&V_tcbinfo); 1745 ti_locked = TI_UNLOCKED; 1746 1747 TCPSTAT_INC(tcps_predack); 1748 1749 /* 1750 * "bad retransmit" recovery. 1751 */ 1752 if (tp->t_rxtshift == 1 && 1753 tp->t_flags & TF_PREVVALID && 1754 (int)(ticks - tp->t_badrxtwin) < 0) { 1755 cc_cong_signal(tp, th, CC_RTO_ERR); 1756 } 1757 1758 /* 1759 * Recalculate the transmit timer / rtt. 1760 * 1761 * Some boxes send broken timestamp replies 1762 * during the SYN+ACK phase, ignore 1763 * timestamps of 0 or we could calculate a 1764 * huge RTT and blow up the retransmit timer. 1765 */ 1766 if ((to.to_flags & TOF_TS) != 0 && 1767 to.to_tsecr) { 1768 uint32_t t; 1769 1770 t = tcp_ts_getticks() - to.to_tsecr; 1771 if (!tp->t_rttlow || tp->t_rttlow > t) 1772 tp->t_rttlow = t; 1773 tcp_xmit_timer(tp, 1774 TCP_TS_TO_TICKS(t) + 1); 1775 } else if (tp->t_rtttime && 1776 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1777 if (!tp->t_rttlow || 1778 tp->t_rttlow > ticks - tp->t_rtttime) 1779 tp->t_rttlow = ticks - tp->t_rtttime; 1780 tcp_xmit_timer(tp, 1781 ticks - tp->t_rtttime); 1782 } 1783 acked = BYTES_THIS_ACK(tp, th); 1784 1785 #ifdef TCP_HHOOK 1786 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1787 hhook_run_tcp_est_in(tp, th, &to); 1788 #endif 1789 1790 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1791 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1792 sbdrop(&so->so_snd, acked); 1793 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1794 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1795 tp->snd_recover = th->th_ack - 1; 1796 1797 /* 1798 * Let the congestion control algorithm update 1799 * congestion control related information. This 1800 * typically means increasing the congestion 1801 * window. 1802 */ 1803 cc_ack_received(tp, th, nsegs, CC_ACK); 1804 1805 tp->snd_una = th->th_ack; 1806 /* 1807 * Pull snd_wl2 up to prevent seq wrap relative 1808 * to th_ack. 1809 */ 1810 tp->snd_wl2 = th->th_ack; 1811 tp->t_dupacks = 0; 1812 m_freem(m); 1813 1814 /* 1815 * If all outstanding data are acked, stop 1816 * retransmit timer, otherwise restart timer 1817 * using current (possibly backed-off) value. 1818 * If process is waiting for space, 1819 * wakeup/selwakeup/signal. If data 1820 * are ready to send, let tcp_output 1821 * decide between more output or persist. 1822 */ 1823 #ifdef TCPDEBUG 1824 if (so->so_options & SO_DEBUG) 1825 tcp_trace(TA_INPUT, ostate, tp, 1826 (void *)tcp_saveipgen, 1827 &tcp_savetcp, 0); 1828 #endif 1829 TCP_PROBE3(debug__input, tp, th, m); 1830 if (tp->snd_una == tp->snd_max) 1831 tcp_timer_activate(tp, TT_REXMT, 0); 1832 else if (!tcp_timer_active(tp, TT_PERSIST)) 1833 tcp_timer_activate(tp, TT_REXMT, 1834 tp->t_rxtcur); 1835 sowwakeup(so); 1836 if (sbavail(&so->so_snd)) 1837 (void) tp->t_fb->tfb_tcp_output(tp); 1838 goto check_delack; 1839 } 1840 } else if (th->th_ack == tp->snd_una && 1841 tlen <= sbspace(&so->so_rcv)) { 1842 int newsize = 0; /* automatic sockbuf scaling */ 1843 1844 /* 1845 * This is a pure, in-sequence data packet with 1846 * nothing on the reassembly queue and we have enough 1847 * buffer space to take it. 1848 */ 1849 if (ti_locked == TI_RLOCKED) 1850 INP_INFO_RUNLOCK(&V_tcbinfo); 1851 ti_locked = TI_UNLOCKED; 1852 1853 /* Clean receiver SACK report if present */ 1854 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1855 tcp_clean_sackreport(tp); 1856 TCPSTAT_INC(tcps_preddat); 1857 tp->rcv_nxt += tlen; 1858 /* 1859 * Pull snd_wl1 up to prevent seq wrap relative to 1860 * th_seq. 1861 */ 1862 tp->snd_wl1 = th->th_seq; 1863 /* 1864 * Pull rcv_up up to prevent seq wrap relative to 1865 * rcv_nxt. 1866 */ 1867 tp->rcv_up = tp->rcv_nxt; 1868 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1869 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1870 #ifdef TCPDEBUG 1871 if (so->so_options & SO_DEBUG) 1872 tcp_trace(TA_INPUT, ostate, tp, 1873 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1874 #endif 1875 TCP_PROBE3(debug__input, tp, th, m); 1876 1877 /* 1878 * Automatic sizing of receive socket buffer. Often the send 1879 * buffer size is not optimally adjusted to the actual network 1880 * conditions at hand (delay bandwidth product). Setting the 1881 * buffer size too small limits throughput on links with high 1882 * bandwidth and high delay (eg. trans-continental/oceanic links). 1883 * 1884 * On the receive side the socket buffer memory is only rarely 1885 * used to any significant extent. This allows us to be much 1886 * more aggressive in scaling the receive socket buffer. For 1887 * the case that the buffer space is actually used to a large 1888 * extent and we run out of kernel memory we can simply drop 1889 * the new segments; TCP on the sender will just retransmit it 1890 * later. Setting the buffer size too big may only consume too 1891 * much kernel memory if the application doesn't read() from 1892 * the socket or packet loss or reordering makes use of the 1893 * reassembly queue. 1894 * 1895 * The criteria to step up the receive buffer one notch are: 1896 * 1. Application has not set receive buffer size with 1897 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1898 * 2. the number of bytes received during the time it takes 1899 * one timestamp to be reflected back to us (the RTT); 1900 * 3. received bytes per RTT is within seven eighth of the 1901 * current socket buffer size; 1902 * 4. receive buffer size has not hit maximal automatic size; 1903 * 1904 * This algorithm does one step per RTT at most and only if 1905 * we receive a bulk stream w/o packet losses or reorderings. 1906 * Shrinking the buffer during idle times is not necessary as 1907 * it doesn't consume any memory when idle. 1908 * 1909 * TODO: Only step up if the application is actually serving 1910 * the buffer to better manage the socket buffer resources. 1911 */ 1912 if (V_tcp_do_autorcvbuf && 1913 (to.to_flags & TOF_TS) && 1914 to.to_tsecr && 1915 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1916 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1917 to.to_tsecr - tp->rfbuf_ts < hz) { 1918 if (tp->rfbuf_cnt > 1919 (so->so_rcv.sb_hiwat / 8 * 7) && 1920 so->so_rcv.sb_hiwat < 1921 V_tcp_autorcvbuf_max) { 1922 newsize = 1923 min(so->so_rcv.sb_hiwat + 1924 V_tcp_autorcvbuf_inc, 1925 V_tcp_autorcvbuf_max); 1926 } 1927 /* Start over with next RTT. */ 1928 tp->rfbuf_ts = 0; 1929 tp->rfbuf_cnt = 0; 1930 } else 1931 tp->rfbuf_cnt += tlen; /* add up */ 1932 } 1933 1934 /* Add data to socket buffer. */ 1935 SOCKBUF_LOCK(&so->so_rcv); 1936 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1937 m_freem(m); 1938 } else { 1939 /* 1940 * Set new socket buffer size. 1941 * Give up when limit is reached. 1942 */ 1943 if (newsize) 1944 if (!sbreserve_locked(&so->so_rcv, 1945 newsize, so, NULL)) 1946 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1947 m_adj(m, drop_hdrlen); /* delayed header drop */ 1948 sbappendstream_locked(&so->so_rcv, m, 0); 1949 } 1950 /* NB: sorwakeup_locked() does an implicit unlock. */ 1951 sorwakeup_locked(so); 1952 if (DELAY_ACK(tp, tlen)) { 1953 tp->t_flags |= TF_DELACK; 1954 } else { 1955 tp->t_flags |= TF_ACKNOW; 1956 tp->t_fb->tfb_tcp_output(tp); 1957 } 1958 goto check_delack; 1959 } 1960 } 1961 1962 /* 1963 * Calculate amount of space in receive window, 1964 * and then do TCP input processing. 1965 * Receive window is amount of space in rcv queue, 1966 * but not less than advertised window. 1967 */ 1968 win = sbspace(&so->so_rcv); 1969 if (win < 0) 1970 win = 0; 1971 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1972 1973 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1974 tp->rfbuf_ts = 0; 1975 tp->rfbuf_cnt = 0; 1976 1977 switch (tp->t_state) { 1978 1979 /* 1980 * If the state is SYN_RECEIVED: 1981 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1982 */ 1983 case TCPS_SYN_RECEIVED: 1984 if ((thflags & TH_ACK) && 1985 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1986 SEQ_GT(th->th_ack, tp->snd_max))) { 1987 rstreason = BANDLIM_RST_OPENPORT; 1988 goto dropwithreset; 1989 } 1990 #ifdef TCP_RFC7413 1991 if (IS_FASTOPEN(tp->t_flags)) { 1992 /* 1993 * When a TFO connection is in SYN_RECEIVED, the 1994 * only valid packets are the initial SYN, a 1995 * retransmit/copy of the initial SYN (possibly with 1996 * a subset of the original data), a valid ACK, a 1997 * FIN, or a RST. 1998 */ 1999 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 2000 rstreason = BANDLIM_RST_OPENPORT; 2001 goto dropwithreset; 2002 } else if (thflags & TH_SYN) { 2003 /* non-initial SYN is ignored */ 2004 if ((tcp_timer_active(tp, TT_DELACK) || 2005 tcp_timer_active(tp, TT_REXMT))) 2006 goto drop; 2007 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 2008 goto drop; 2009 } 2010 } 2011 #endif 2012 break; 2013 2014 /* 2015 * If the state is SYN_SENT: 2016 * if seg contains an ACK, but not for our SYN, drop the input. 2017 * if seg contains a RST, then drop the connection. 2018 * if seg does not contain SYN, then drop it. 2019 * Otherwise this is an acceptable SYN segment 2020 * initialize tp->rcv_nxt and tp->irs 2021 * if seg contains ack then advance tp->snd_una 2022 * if seg contains an ECE and ECN support is enabled, the stream 2023 * is ECN capable. 2024 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2025 * arrange for segment to be acked (eventually) 2026 * continue processing rest of data/controls, beginning with URG 2027 */ 2028 case TCPS_SYN_SENT: 2029 if ((thflags & TH_ACK) && 2030 (SEQ_LEQ(th->th_ack, tp->iss) || 2031 SEQ_GT(th->th_ack, tp->snd_max))) { 2032 rstreason = BANDLIM_UNLIMITED; 2033 goto dropwithreset; 2034 } 2035 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2036 TCP_PROBE5(connect__refused, NULL, tp, 2037 m, tp, th); 2038 tp = tcp_drop(tp, ECONNREFUSED); 2039 } 2040 if (thflags & TH_RST) 2041 goto drop; 2042 if (!(thflags & TH_SYN)) 2043 goto drop; 2044 2045 tp->irs = th->th_seq; 2046 tcp_rcvseqinit(tp); 2047 if (thflags & TH_ACK) { 2048 TCPSTAT_INC(tcps_connects); 2049 soisconnected(so); 2050 #ifdef MAC 2051 mac_socketpeer_set_from_mbuf(m, so); 2052 #endif 2053 /* Do window scaling on this connection? */ 2054 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2055 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2056 tp->rcv_scale = tp->request_r_scale; 2057 } 2058 tp->rcv_adv += min(tp->rcv_wnd, 2059 TCP_MAXWIN << tp->rcv_scale); 2060 tp->snd_una++; /* SYN is acked */ 2061 /* 2062 * If there's data, delay ACK; if there's also a FIN 2063 * ACKNOW will be turned on later. 2064 */ 2065 if (DELAY_ACK(tp, tlen) && tlen != 0) 2066 tcp_timer_activate(tp, TT_DELACK, 2067 tcp_delacktime); 2068 else 2069 tp->t_flags |= TF_ACKNOW; 2070 2071 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2072 tp->t_flags |= TF_ECN_PERMIT; 2073 TCPSTAT_INC(tcps_ecn_shs); 2074 } 2075 2076 /* 2077 * Received <SYN,ACK> in SYN_SENT[*] state. 2078 * Transitions: 2079 * SYN_SENT --> ESTABLISHED 2080 * SYN_SENT* --> FIN_WAIT_1 2081 */ 2082 tp->t_starttime = ticks; 2083 if (tp->t_flags & TF_NEEDFIN) { 2084 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2085 tp->t_flags &= ~TF_NEEDFIN; 2086 thflags &= ~TH_SYN; 2087 } else { 2088 tcp_state_change(tp, TCPS_ESTABLISHED); 2089 TCP_PROBE5(connect__established, NULL, tp, 2090 m, tp, th); 2091 cc_conn_init(tp); 2092 tcp_timer_activate(tp, TT_KEEP, 2093 TP_KEEPIDLE(tp)); 2094 } 2095 } else { 2096 /* 2097 * Received initial SYN in SYN-SENT[*] state => 2098 * simultaneous open. 2099 * If it succeeds, connection is * half-synchronized. 2100 * Otherwise, do 3-way handshake: 2101 * SYN-SENT -> SYN-RECEIVED 2102 * SYN-SENT* -> SYN-RECEIVED* 2103 */ 2104 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2105 tcp_timer_activate(tp, TT_REXMT, 0); 2106 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2107 } 2108 2109 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2110 "ti_locked %d", __func__, ti_locked)); 2111 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2112 INP_WLOCK_ASSERT(tp->t_inpcb); 2113 2114 /* 2115 * Advance th->th_seq to correspond to first data byte. 2116 * If data, trim to stay within window, 2117 * dropping FIN if necessary. 2118 */ 2119 th->th_seq++; 2120 if (tlen > tp->rcv_wnd) { 2121 todrop = tlen - tp->rcv_wnd; 2122 m_adj(m, -todrop); 2123 tlen = tp->rcv_wnd; 2124 thflags &= ~TH_FIN; 2125 TCPSTAT_INC(tcps_rcvpackafterwin); 2126 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2127 } 2128 tp->snd_wl1 = th->th_seq - 1; 2129 tp->rcv_up = th->th_seq; 2130 /* 2131 * Client side of transaction: already sent SYN and data. 2132 * If the remote host used T/TCP to validate the SYN, 2133 * our data will be ACK'd; if so, enter normal data segment 2134 * processing in the middle of step 5, ack processing. 2135 * Otherwise, goto step 6. 2136 */ 2137 if (thflags & TH_ACK) 2138 goto process_ACK; 2139 2140 goto step6; 2141 2142 /* 2143 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2144 * do normal processing. 2145 * 2146 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2147 */ 2148 case TCPS_LAST_ACK: 2149 case TCPS_CLOSING: 2150 break; /* continue normal processing */ 2151 } 2152 2153 /* 2154 * States other than LISTEN or SYN_SENT. 2155 * First check the RST flag and sequence number since reset segments 2156 * are exempt from the timestamp and connection count tests. This 2157 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2158 * below which allowed reset segments in half the sequence space 2159 * to fall though and be processed (which gives forged reset 2160 * segments with a random sequence number a 50 percent chance of 2161 * killing a connection). 2162 * Then check timestamp, if present. 2163 * Then check the connection count, if present. 2164 * Then check that at least some bytes of segment are within 2165 * receive window. If segment begins before rcv_nxt, 2166 * drop leading data (and SYN); if nothing left, just ack. 2167 */ 2168 if (thflags & TH_RST) { 2169 /* 2170 * RFC5961 Section 3.2 2171 * 2172 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2173 * - If RST is in window, we send challenge ACK. 2174 * 2175 * Note: to take into account delayed ACKs, we should 2176 * test against last_ack_sent instead of rcv_nxt. 2177 * Note 2: we handle special case of closed window, not 2178 * covered by the RFC. 2179 */ 2180 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2181 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2182 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2183 2184 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2185 KASSERT(ti_locked == TI_RLOCKED, 2186 ("%s: TH_RST ti_locked %d, th %p tp %p", 2187 __func__, ti_locked, th, tp)); 2188 KASSERT(tp->t_state != TCPS_SYN_SENT, 2189 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2190 __func__, th, tp)); 2191 2192 if (V_tcp_insecure_rst || 2193 tp->last_ack_sent == th->th_seq) { 2194 TCPSTAT_INC(tcps_drops); 2195 /* Drop the connection. */ 2196 switch (tp->t_state) { 2197 case TCPS_SYN_RECEIVED: 2198 so->so_error = ECONNREFUSED; 2199 goto close; 2200 case TCPS_ESTABLISHED: 2201 case TCPS_FIN_WAIT_1: 2202 case TCPS_FIN_WAIT_2: 2203 case TCPS_CLOSE_WAIT: 2204 case TCPS_CLOSING: 2205 case TCPS_LAST_ACK: 2206 so->so_error = ECONNRESET; 2207 close: 2208 /* FALLTHROUGH */ 2209 default: 2210 tp = tcp_close(tp); 2211 } 2212 } else { 2213 TCPSTAT_INC(tcps_badrst); 2214 /* Send challenge ACK. */ 2215 tcp_respond(tp, mtod(m, void *), th, m, 2216 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2217 tp->last_ack_sent = tp->rcv_nxt; 2218 m = NULL; 2219 } 2220 } 2221 goto drop; 2222 } 2223 2224 /* 2225 * RFC5961 Section 4.2 2226 * Send challenge ACK for any SYN in synchronized state. 2227 */ 2228 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2229 tp->t_state != TCPS_SYN_RECEIVED) { 2230 KASSERT(ti_locked == TI_RLOCKED, 2231 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2232 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2233 2234 TCPSTAT_INC(tcps_badsyn); 2235 if (V_tcp_insecure_syn && 2236 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2237 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2238 tp = tcp_drop(tp, ECONNRESET); 2239 rstreason = BANDLIM_UNLIMITED; 2240 } else { 2241 /* Send challenge ACK. */ 2242 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2243 tp->snd_nxt, TH_ACK); 2244 tp->last_ack_sent = tp->rcv_nxt; 2245 m = NULL; 2246 } 2247 goto drop; 2248 } 2249 2250 /* 2251 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2252 * and it's less than ts_recent, drop it. 2253 */ 2254 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2255 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2256 2257 /* Check to see if ts_recent is over 24 days old. */ 2258 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2259 /* 2260 * Invalidate ts_recent. If this segment updates 2261 * ts_recent, the age will be reset later and ts_recent 2262 * will get a valid value. If it does not, setting 2263 * ts_recent to zero will at least satisfy the 2264 * requirement that zero be placed in the timestamp 2265 * echo reply when ts_recent isn't valid. The 2266 * age isn't reset until we get a valid ts_recent 2267 * because we don't want out-of-order segments to be 2268 * dropped when ts_recent is old. 2269 */ 2270 tp->ts_recent = 0; 2271 } else { 2272 TCPSTAT_INC(tcps_rcvduppack); 2273 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2274 TCPSTAT_INC(tcps_pawsdrop); 2275 if (tlen) 2276 goto dropafterack; 2277 goto drop; 2278 } 2279 } 2280 2281 /* 2282 * In the SYN-RECEIVED state, validate that the packet belongs to 2283 * this connection before trimming the data to fit the receive 2284 * window. Check the sequence number versus IRS since we know 2285 * the sequence numbers haven't wrapped. This is a partial fix 2286 * for the "LAND" DoS attack. 2287 */ 2288 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2289 rstreason = BANDLIM_RST_OPENPORT; 2290 goto dropwithreset; 2291 } 2292 2293 todrop = tp->rcv_nxt - th->th_seq; 2294 if (todrop > 0) { 2295 if (thflags & TH_SYN) { 2296 thflags &= ~TH_SYN; 2297 th->th_seq++; 2298 if (th->th_urp > 1) 2299 th->th_urp--; 2300 else 2301 thflags &= ~TH_URG; 2302 todrop--; 2303 } 2304 /* 2305 * Following if statement from Stevens, vol. 2, p. 960. 2306 */ 2307 if (todrop > tlen 2308 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2309 /* 2310 * Any valid FIN must be to the left of the window. 2311 * At this point the FIN must be a duplicate or out 2312 * of sequence; drop it. 2313 */ 2314 thflags &= ~TH_FIN; 2315 2316 /* 2317 * Send an ACK to resynchronize and drop any data. 2318 * But keep on processing for RST or ACK. 2319 */ 2320 tp->t_flags |= TF_ACKNOW; 2321 todrop = tlen; 2322 TCPSTAT_INC(tcps_rcvduppack); 2323 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2324 } else { 2325 TCPSTAT_INC(tcps_rcvpartduppack); 2326 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2327 } 2328 drop_hdrlen += todrop; /* drop from the top afterwards */ 2329 th->th_seq += todrop; 2330 tlen -= todrop; 2331 if (th->th_urp > todrop) 2332 th->th_urp -= todrop; 2333 else { 2334 thflags &= ~TH_URG; 2335 th->th_urp = 0; 2336 } 2337 } 2338 2339 /* 2340 * If new data are received on a connection after the 2341 * user processes are gone, then RST the other end. 2342 */ 2343 if ((so->so_state & SS_NOFDREF) && 2344 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2345 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2346 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2347 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2348 2349 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2350 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2351 "after socket was closed, " 2352 "sending RST and removing tcpcb\n", 2353 s, __func__, tcpstates[tp->t_state], tlen); 2354 free(s, M_TCPLOG); 2355 } 2356 tp = tcp_close(tp); 2357 TCPSTAT_INC(tcps_rcvafterclose); 2358 rstreason = BANDLIM_UNLIMITED; 2359 goto dropwithreset; 2360 } 2361 2362 /* 2363 * If segment ends after window, drop trailing data 2364 * (and PUSH and FIN); if nothing left, just ACK. 2365 */ 2366 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2367 if (todrop > 0) { 2368 TCPSTAT_INC(tcps_rcvpackafterwin); 2369 if (todrop >= tlen) { 2370 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2371 /* 2372 * If window is closed can only take segments at 2373 * window edge, and have to drop data and PUSH from 2374 * incoming segments. Continue processing, but 2375 * remember to ack. Otherwise, drop segment 2376 * and ack. 2377 */ 2378 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2379 tp->t_flags |= TF_ACKNOW; 2380 TCPSTAT_INC(tcps_rcvwinprobe); 2381 } else 2382 goto dropafterack; 2383 } else 2384 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2385 m_adj(m, -todrop); 2386 tlen -= todrop; 2387 thflags &= ~(TH_PUSH|TH_FIN); 2388 } 2389 2390 /* 2391 * If last ACK falls within this segment's sequence numbers, 2392 * record its timestamp. 2393 * NOTE: 2394 * 1) That the test incorporates suggestions from the latest 2395 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2396 * 2) That updating only on newer timestamps interferes with 2397 * our earlier PAWS tests, so this check should be solely 2398 * predicated on the sequence space of this segment. 2399 * 3) That we modify the segment boundary check to be 2400 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2401 * instead of RFC1323's 2402 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2403 * This modified check allows us to overcome RFC1323's 2404 * limitations as described in Stevens TCP/IP Illustrated 2405 * Vol. 2 p.869. In such cases, we can still calculate the 2406 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2407 */ 2408 if ((to.to_flags & TOF_TS) != 0 && 2409 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2410 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2411 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2412 tp->ts_recent_age = tcp_ts_getticks(); 2413 tp->ts_recent = to.to_tsval; 2414 } 2415 2416 /* 2417 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2418 * flag is on (half-synchronized state), then queue data for 2419 * later processing; else drop segment and return. 2420 */ 2421 if ((thflags & TH_ACK) == 0) { 2422 if (tp->t_state == TCPS_SYN_RECEIVED || 2423 (tp->t_flags & TF_NEEDSYN)) { 2424 #ifdef TCP_RFC7413 2425 if (tp->t_state == TCPS_SYN_RECEIVED && 2426 IS_FASTOPEN(tp->t_flags)) { 2427 tp->snd_wnd = tiwin; 2428 cc_conn_init(tp); 2429 } 2430 #endif 2431 goto step6; 2432 } else if (tp->t_flags & TF_ACKNOW) 2433 goto dropafterack; 2434 else 2435 goto drop; 2436 } 2437 2438 /* 2439 * Ack processing. 2440 */ 2441 switch (tp->t_state) { 2442 2443 /* 2444 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2445 * ESTABLISHED state and continue processing. 2446 * The ACK was checked above. 2447 */ 2448 case TCPS_SYN_RECEIVED: 2449 2450 TCPSTAT_INC(tcps_connects); 2451 soisconnected(so); 2452 /* Do window scaling? */ 2453 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2454 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2455 tp->rcv_scale = tp->request_r_scale; 2456 tp->snd_wnd = tiwin; 2457 } 2458 /* 2459 * Make transitions: 2460 * SYN-RECEIVED -> ESTABLISHED 2461 * SYN-RECEIVED* -> FIN-WAIT-1 2462 */ 2463 tp->t_starttime = ticks; 2464 if (tp->t_flags & TF_NEEDFIN) { 2465 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2466 tp->t_flags &= ~TF_NEEDFIN; 2467 } else { 2468 tcp_state_change(tp, TCPS_ESTABLISHED); 2469 TCP_PROBE5(accept__established, NULL, tp, 2470 m, tp, th); 2471 #ifdef TCP_RFC7413 2472 if (tp->t_tfo_pending) { 2473 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2474 tp->t_tfo_pending = NULL; 2475 2476 /* 2477 * Account for the ACK of our SYN prior to 2478 * regular ACK processing below. 2479 */ 2480 tp->snd_una++; 2481 } 2482 /* 2483 * TFO connections call cc_conn_init() during SYN 2484 * processing. Calling it again here for such 2485 * connections is not harmless as it would undo the 2486 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2487 * is retransmitted. 2488 */ 2489 if (!IS_FASTOPEN(tp->t_flags)) 2490 #endif 2491 cc_conn_init(tp); 2492 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2493 } 2494 /* 2495 * If segment contains data or ACK, will call tcp_reass() 2496 * later; if not, do so now to pass queued data to user. 2497 */ 2498 if (tlen == 0 && (thflags & TH_FIN) == 0) 2499 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2500 (struct mbuf *)0); 2501 tp->snd_wl1 = th->th_seq - 1; 2502 /* FALLTHROUGH */ 2503 2504 /* 2505 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2506 * ACKs. If the ack is in the range 2507 * tp->snd_una < th->th_ack <= tp->snd_max 2508 * then advance tp->snd_una to th->th_ack and drop 2509 * data from the retransmission queue. If this ACK reflects 2510 * more up to date window information we update our window information. 2511 */ 2512 case TCPS_ESTABLISHED: 2513 case TCPS_FIN_WAIT_1: 2514 case TCPS_FIN_WAIT_2: 2515 case TCPS_CLOSE_WAIT: 2516 case TCPS_CLOSING: 2517 case TCPS_LAST_ACK: 2518 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2519 TCPSTAT_INC(tcps_rcvacktoomuch); 2520 goto dropafterack; 2521 } 2522 if ((tp->t_flags & TF_SACK_PERMIT) && 2523 ((to.to_flags & TOF_SACK) || 2524 !TAILQ_EMPTY(&tp->snd_holes))) 2525 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2526 else 2527 /* 2528 * Reset the value so that previous (valid) value 2529 * from the last ack with SACK doesn't get used. 2530 */ 2531 tp->sackhint.sacked_bytes = 0; 2532 2533 #ifdef TCP_HHOOK 2534 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2535 hhook_run_tcp_est_in(tp, th, &to); 2536 #endif 2537 2538 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2539 u_int maxseg; 2540 2541 maxseg = tcp_maxseg(tp); 2542 if (tlen == 0 && 2543 (tiwin == tp->snd_wnd || 2544 (tp->t_flags & TF_SACK_PERMIT))) { 2545 /* 2546 * If this is the first time we've seen a 2547 * FIN from the remote, this is not a 2548 * duplicate and it needs to be processed 2549 * normally. This happens during a 2550 * simultaneous close. 2551 */ 2552 if ((thflags & TH_FIN) && 2553 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2554 tp->t_dupacks = 0; 2555 break; 2556 } 2557 TCPSTAT_INC(tcps_rcvdupack); 2558 /* 2559 * If we have outstanding data (other than 2560 * a window probe), this is a completely 2561 * duplicate ack (ie, window info didn't 2562 * change and FIN isn't set), 2563 * the ack is the biggest we've 2564 * seen and we've seen exactly our rexmt 2565 * threshold of them, assume a packet 2566 * has been dropped and retransmit it. 2567 * Kludge snd_nxt & the congestion 2568 * window so we send only this one 2569 * packet. 2570 * 2571 * We know we're losing at the current 2572 * window size so do congestion avoidance 2573 * (set ssthresh to half the current window 2574 * and pull our congestion window back to 2575 * the new ssthresh). 2576 * 2577 * Dup acks mean that packets have left the 2578 * network (they're now cached at the receiver) 2579 * so bump cwnd by the amount in the receiver 2580 * to keep a constant cwnd packets in the 2581 * network. 2582 * 2583 * When using TCP ECN, notify the peer that 2584 * we reduced the cwnd. 2585 */ 2586 /* 2587 * Following 2 kinds of acks should not affect 2588 * dupack counting: 2589 * 1) Old acks 2590 * 2) Acks with SACK but without any new SACK 2591 * information in them. These could result from 2592 * any anomaly in the network like a switch 2593 * duplicating packets or a possible DoS attack. 2594 */ 2595 if (th->th_ack != tp->snd_una || 2596 ((tp->t_flags & TF_SACK_PERMIT) && 2597 !sack_changed)) 2598 break; 2599 else if (!tcp_timer_active(tp, TT_REXMT)) 2600 tp->t_dupacks = 0; 2601 else if (++tp->t_dupacks > tcprexmtthresh || 2602 IN_FASTRECOVERY(tp->t_flags)) { 2603 cc_ack_received(tp, th, nsegs, 2604 CC_DUPACK); 2605 if ((tp->t_flags & TF_SACK_PERMIT) && 2606 IN_FASTRECOVERY(tp->t_flags)) { 2607 int awnd; 2608 2609 /* 2610 * Compute the amount of data in flight first. 2611 * We can inject new data into the pipe iff 2612 * we have less than 1/2 the original window's 2613 * worth of data in flight. 2614 */ 2615 if (V_tcp_do_rfc6675_pipe) 2616 awnd = tcp_compute_pipe(tp); 2617 else 2618 awnd = (tp->snd_nxt - tp->snd_fack) + 2619 tp->sackhint.sack_bytes_rexmit; 2620 2621 if (awnd < tp->snd_ssthresh) { 2622 tp->snd_cwnd += maxseg; 2623 /* 2624 * RFC5681 Section 3.2 talks about cwnd 2625 * inflation on additional dupacks and 2626 * deflation on recovering from loss. 2627 * 2628 * We keep cwnd into check so that 2629 * we don't have to 'deflate' it when we 2630 * get out of recovery. 2631 */ 2632 if (tp->snd_cwnd > tp->snd_ssthresh) 2633 tp->snd_cwnd = tp->snd_ssthresh; 2634 } 2635 } else 2636 tp->snd_cwnd += maxseg; 2637 (void) tp->t_fb->tfb_tcp_output(tp); 2638 goto drop; 2639 } else if (tp->t_dupacks == tcprexmtthresh) { 2640 tcp_seq onxt = tp->snd_nxt; 2641 2642 /* 2643 * If we're doing sack, check to 2644 * see if we're already in sack 2645 * recovery. If we're not doing sack, 2646 * check to see if we're in newreno 2647 * recovery. 2648 */ 2649 if (tp->t_flags & TF_SACK_PERMIT) { 2650 if (IN_FASTRECOVERY(tp->t_flags)) { 2651 tp->t_dupacks = 0; 2652 break; 2653 } 2654 } else { 2655 if (SEQ_LEQ(th->th_ack, 2656 tp->snd_recover)) { 2657 tp->t_dupacks = 0; 2658 break; 2659 } 2660 } 2661 /* Congestion signal before ack. */ 2662 cc_cong_signal(tp, th, CC_NDUPACK); 2663 cc_ack_received(tp, th, nsegs, 2664 CC_DUPACK); 2665 tcp_timer_activate(tp, TT_REXMT, 0); 2666 tp->t_rtttime = 0; 2667 if (tp->t_flags & TF_SACK_PERMIT) { 2668 TCPSTAT_INC( 2669 tcps_sack_recovery_episode); 2670 tp->sack_newdata = tp->snd_nxt; 2671 if (CC_ALGO(tp)->cong_signal == NULL) 2672 tp->snd_cwnd = maxseg; 2673 (void) tp->t_fb->tfb_tcp_output(tp); 2674 goto drop; 2675 } 2676 tp->snd_nxt = th->th_ack; 2677 if (CC_ALGO(tp)->cong_signal == NULL) 2678 tp->snd_cwnd = maxseg; 2679 (void) tp->t_fb->tfb_tcp_output(tp); 2680 KASSERT(tp->snd_limited <= 2, 2681 ("%s: tp->snd_limited too big", 2682 __func__)); 2683 if (CC_ALGO(tp)->cong_signal == NULL) 2684 tp->snd_cwnd = tp->snd_ssthresh + 2685 maxseg * 2686 (tp->t_dupacks - tp->snd_limited); 2687 if (SEQ_GT(onxt, tp->snd_nxt)) 2688 tp->snd_nxt = onxt; 2689 goto drop; 2690 } else if (V_tcp_do_rfc3042) { 2691 /* 2692 * Process first and second duplicate 2693 * ACKs. Each indicates a segment 2694 * leaving the network, creating room 2695 * for more. Make sure we can send a 2696 * packet on reception of each duplicate 2697 * ACK by increasing snd_cwnd by one 2698 * segment. Restore the original 2699 * snd_cwnd after packet transmission. 2700 */ 2701 cc_ack_received(tp, th, nsegs, 2702 CC_DUPACK); 2703 uint32_t oldcwnd = tp->snd_cwnd; 2704 tcp_seq oldsndmax = tp->snd_max; 2705 u_int sent; 2706 int avail; 2707 2708 KASSERT(tp->t_dupacks == 1 || 2709 tp->t_dupacks == 2, 2710 ("%s: dupacks not 1 or 2", 2711 __func__)); 2712 if (tp->t_dupacks == 1) 2713 tp->snd_limited = 0; 2714 tp->snd_cwnd = 2715 (tp->snd_nxt - tp->snd_una) + 2716 (tp->t_dupacks - tp->snd_limited) * 2717 maxseg; 2718 /* 2719 * Only call tcp_output when there 2720 * is new data available to be sent. 2721 * Otherwise we would send pure ACKs. 2722 */ 2723 SOCKBUF_LOCK(&so->so_snd); 2724 avail = sbavail(&so->so_snd) - 2725 (tp->snd_nxt - tp->snd_una); 2726 SOCKBUF_UNLOCK(&so->so_snd); 2727 if (avail > 0) 2728 (void) tp->t_fb->tfb_tcp_output(tp); 2729 sent = tp->snd_max - oldsndmax; 2730 if (sent > maxseg) { 2731 KASSERT((tp->t_dupacks == 2 && 2732 tp->snd_limited == 0) || 2733 (sent == maxseg + 1 && 2734 tp->t_flags & TF_SENTFIN), 2735 ("%s: sent too much", 2736 __func__)); 2737 tp->snd_limited = 2; 2738 } else if (sent > 0) 2739 ++tp->snd_limited; 2740 tp->snd_cwnd = oldcwnd; 2741 goto drop; 2742 } 2743 } 2744 break; 2745 } else { 2746 /* 2747 * This ack is advancing the left edge, reset the 2748 * counter. 2749 */ 2750 tp->t_dupacks = 0; 2751 /* 2752 * If this ack also has new SACK info, increment the 2753 * counter as per rfc6675. 2754 */ 2755 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2756 tp->t_dupacks++; 2757 } 2758 2759 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2760 ("%s: th_ack <= snd_una", __func__)); 2761 2762 /* 2763 * If the congestion window was inflated to account 2764 * for the other side's cached packets, retract it. 2765 */ 2766 if (IN_FASTRECOVERY(tp->t_flags)) { 2767 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2768 if (tp->t_flags & TF_SACK_PERMIT) 2769 tcp_sack_partialack(tp, th); 2770 else 2771 tcp_newreno_partial_ack(tp, th); 2772 } else 2773 cc_post_recovery(tp, th); 2774 } 2775 /* 2776 * If we reach this point, ACK is not a duplicate, 2777 * i.e., it ACKs something we sent. 2778 */ 2779 if (tp->t_flags & TF_NEEDSYN) { 2780 /* 2781 * T/TCP: Connection was half-synchronized, and our 2782 * SYN has been ACK'd (so connection is now fully 2783 * synchronized). Go to non-starred state, 2784 * increment snd_una for ACK of SYN, and check if 2785 * we can do window scaling. 2786 */ 2787 tp->t_flags &= ~TF_NEEDSYN; 2788 tp->snd_una++; 2789 /* Do window scaling? */ 2790 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2791 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2792 tp->rcv_scale = tp->request_r_scale; 2793 /* Send window already scaled. */ 2794 } 2795 } 2796 2797 process_ACK: 2798 INP_WLOCK_ASSERT(tp->t_inpcb); 2799 2800 acked = BYTES_THIS_ACK(tp, th); 2801 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2802 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2803 tp->snd_una, th->th_ack, tp, m)); 2804 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2805 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2806 2807 /* 2808 * If we just performed our first retransmit, and the ACK 2809 * arrives within our recovery window, then it was a mistake 2810 * to do the retransmit in the first place. Recover our 2811 * original cwnd and ssthresh, and proceed to transmit where 2812 * we left off. 2813 */ 2814 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2815 (int)(ticks - tp->t_badrxtwin) < 0) 2816 cc_cong_signal(tp, th, CC_RTO_ERR); 2817 2818 /* 2819 * If we have a timestamp reply, update smoothed 2820 * round trip time. If no timestamp is present but 2821 * transmit timer is running and timed sequence 2822 * number was acked, update smoothed round trip time. 2823 * Since we now have an rtt measurement, cancel the 2824 * timer backoff (cf., Phil Karn's retransmit alg.). 2825 * Recompute the initial retransmit timer. 2826 * 2827 * Some boxes send broken timestamp replies 2828 * during the SYN+ACK phase, ignore 2829 * timestamps of 0 or we could calculate a 2830 * huge RTT and blow up the retransmit timer. 2831 */ 2832 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2833 uint32_t t; 2834 2835 t = tcp_ts_getticks() - to.to_tsecr; 2836 if (!tp->t_rttlow || tp->t_rttlow > t) 2837 tp->t_rttlow = t; 2838 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2839 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2840 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2841 tp->t_rttlow = ticks - tp->t_rtttime; 2842 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2843 } 2844 2845 /* 2846 * If all outstanding data is acked, stop retransmit 2847 * timer and remember to restart (more output or persist). 2848 * If there is more data to be acked, restart retransmit 2849 * timer, using current (possibly backed-off) value. 2850 */ 2851 if (th->th_ack == tp->snd_max) { 2852 tcp_timer_activate(tp, TT_REXMT, 0); 2853 needoutput = 1; 2854 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2855 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2856 2857 /* 2858 * If no data (only SYN) was ACK'd, 2859 * skip rest of ACK processing. 2860 */ 2861 if (acked == 0) 2862 goto step6; 2863 2864 /* 2865 * Let the congestion control algorithm update congestion 2866 * control related information. This typically means increasing 2867 * the congestion window. 2868 */ 2869 cc_ack_received(tp, th, nsegs, CC_ACK); 2870 2871 SOCKBUF_LOCK(&so->so_snd); 2872 if (acked > sbavail(&so->so_snd)) { 2873 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2874 tp->snd_wnd -= sbavail(&so->so_snd); 2875 else 2876 tp->snd_wnd = 0; 2877 mfree = sbcut_locked(&so->so_snd, 2878 (int)sbavail(&so->so_snd)); 2879 ourfinisacked = 1; 2880 } else { 2881 mfree = sbcut_locked(&so->so_snd, acked); 2882 if (tp->snd_wnd >= (uint32_t) acked) 2883 tp->snd_wnd -= acked; 2884 else 2885 tp->snd_wnd = 0; 2886 ourfinisacked = 0; 2887 } 2888 /* NB: sowwakeup_locked() does an implicit unlock. */ 2889 sowwakeup_locked(so); 2890 m_freem(mfree); 2891 /* Detect una wraparound. */ 2892 if (!IN_RECOVERY(tp->t_flags) && 2893 SEQ_GT(tp->snd_una, tp->snd_recover) && 2894 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2895 tp->snd_recover = th->th_ack - 1; 2896 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2897 if (IN_RECOVERY(tp->t_flags) && 2898 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2899 EXIT_RECOVERY(tp->t_flags); 2900 } 2901 tp->snd_una = th->th_ack; 2902 if (tp->t_flags & TF_SACK_PERMIT) { 2903 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2904 tp->snd_recover = tp->snd_una; 2905 } 2906 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2907 tp->snd_nxt = tp->snd_una; 2908 2909 switch (tp->t_state) { 2910 2911 /* 2912 * In FIN_WAIT_1 STATE in addition to the processing 2913 * for the ESTABLISHED state if our FIN is now acknowledged 2914 * then enter FIN_WAIT_2. 2915 */ 2916 case TCPS_FIN_WAIT_1: 2917 if (ourfinisacked) { 2918 /* 2919 * If we can't receive any more 2920 * data, then closing user can proceed. 2921 * Starting the timer is contrary to the 2922 * specification, but if we don't get a FIN 2923 * we'll hang forever. 2924 * 2925 * XXXjl: 2926 * we should release the tp also, and use a 2927 * compressed state. 2928 */ 2929 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2930 soisdisconnected(so); 2931 tcp_timer_activate(tp, TT_2MSL, 2932 (tcp_fast_finwait2_recycle ? 2933 tcp_finwait2_timeout : 2934 TP_MAXIDLE(tp))); 2935 } 2936 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2937 } 2938 break; 2939 2940 /* 2941 * In CLOSING STATE in addition to the processing for 2942 * the ESTABLISHED state if the ACK acknowledges our FIN 2943 * then enter the TIME-WAIT state, otherwise ignore 2944 * the segment. 2945 */ 2946 case TCPS_CLOSING: 2947 if (ourfinisacked) { 2948 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2949 tcp_twstart(tp); 2950 INP_INFO_RUNLOCK(&V_tcbinfo); 2951 m_freem(m); 2952 return; 2953 } 2954 break; 2955 2956 /* 2957 * In LAST_ACK, we may still be waiting for data to drain 2958 * and/or to be acked, as well as for the ack of our FIN. 2959 * If our FIN is now acknowledged, delete the TCB, 2960 * enter the closed state and return. 2961 */ 2962 case TCPS_LAST_ACK: 2963 if (ourfinisacked) { 2964 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2965 tp = tcp_close(tp); 2966 goto drop; 2967 } 2968 break; 2969 } 2970 } 2971 2972 step6: 2973 INP_WLOCK_ASSERT(tp->t_inpcb); 2974 2975 /* 2976 * Update window information. 2977 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2978 */ 2979 if ((thflags & TH_ACK) && 2980 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2981 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2982 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2983 /* keep track of pure window updates */ 2984 if (tlen == 0 && 2985 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2986 TCPSTAT_INC(tcps_rcvwinupd); 2987 tp->snd_wnd = tiwin; 2988 tp->snd_wl1 = th->th_seq; 2989 tp->snd_wl2 = th->th_ack; 2990 if (tp->snd_wnd > tp->max_sndwnd) 2991 tp->max_sndwnd = tp->snd_wnd; 2992 needoutput = 1; 2993 } 2994 2995 /* 2996 * Process segments with URG. 2997 */ 2998 if ((thflags & TH_URG) && th->th_urp && 2999 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3000 /* 3001 * This is a kludge, but if we receive and accept 3002 * random urgent pointers, we'll crash in 3003 * soreceive. It's hard to imagine someone 3004 * actually wanting to send this much urgent data. 3005 */ 3006 SOCKBUF_LOCK(&so->so_rcv); 3007 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 3008 th->th_urp = 0; /* XXX */ 3009 thflags &= ~TH_URG; /* XXX */ 3010 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 3011 goto dodata; /* XXX */ 3012 } 3013 /* 3014 * If this segment advances the known urgent pointer, 3015 * then mark the data stream. This should not happen 3016 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 3017 * a FIN has been received from the remote side. 3018 * In these states we ignore the URG. 3019 * 3020 * According to RFC961 (Assigned Protocols), 3021 * the urgent pointer points to the last octet 3022 * of urgent data. We continue, however, 3023 * to consider it to indicate the first octet 3024 * of data past the urgent section as the original 3025 * spec states (in one of two places). 3026 */ 3027 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3028 tp->rcv_up = th->th_seq + th->th_urp; 3029 so->so_oobmark = sbavail(&so->so_rcv) + 3030 (tp->rcv_up - tp->rcv_nxt) - 1; 3031 if (so->so_oobmark == 0) 3032 so->so_rcv.sb_state |= SBS_RCVATMARK; 3033 sohasoutofband(so); 3034 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3035 } 3036 SOCKBUF_UNLOCK(&so->so_rcv); 3037 /* 3038 * Remove out of band data so doesn't get presented to user. 3039 * This can happen independent of advancing the URG pointer, 3040 * but if two URG's are pending at once, some out-of-band 3041 * data may creep in... ick. 3042 */ 3043 if (th->th_urp <= (uint32_t)tlen && 3044 !(so->so_options & SO_OOBINLINE)) { 3045 /* hdr drop is delayed */ 3046 tcp_pulloutofband(so, th, m, drop_hdrlen); 3047 } 3048 } else { 3049 /* 3050 * If no out of band data is expected, 3051 * pull receive urgent pointer along 3052 * with the receive window. 3053 */ 3054 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3055 tp->rcv_up = tp->rcv_nxt; 3056 } 3057 dodata: /* XXX */ 3058 INP_WLOCK_ASSERT(tp->t_inpcb); 3059 3060 /* 3061 * Process the segment text, merging it into the TCP sequencing queue, 3062 * and arranging for acknowledgment of receipt if necessary. 3063 * This process logically involves adjusting tp->rcv_wnd as data 3064 * is presented to the user (this happens in tcp_usrreq.c, 3065 * case PRU_RCVD). If a FIN has already been received on this 3066 * connection then we just ignore the text. 3067 */ 3068 #ifdef TCP_RFC7413 3069 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3070 IS_FASTOPEN(tp->t_flags)); 3071 #else 3072 #define tfo_syn (false) 3073 #endif 3074 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3075 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3076 tcp_seq save_start = th->th_seq; 3077 m_adj(m, drop_hdrlen); /* delayed header drop */ 3078 /* 3079 * Insert segment which includes th into TCP reassembly queue 3080 * with control block tp. Set thflags to whether reassembly now 3081 * includes a segment with FIN. This handles the common case 3082 * inline (segment is the next to be received on an established 3083 * connection, and the queue is empty), avoiding linkage into 3084 * and removal from the queue and repetition of various 3085 * conversions. 3086 * Set DELACK for segments received in order, but ack 3087 * immediately when segments are out of order (so 3088 * fast retransmit can work). 3089 */ 3090 if (th->th_seq == tp->rcv_nxt && 3091 LIST_EMPTY(&tp->t_segq) && 3092 (TCPS_HAVEESTABLISHED(tp->t_state) || 3093 tfo_syn)) { 3094 if (DELAY_ACK(tp, tlen) || tfo_syn) 3095 tp->t_flags |= TF_DELACK; 3096 else 3097 tp->t_flags |= TF_ACKNOW; 3098 tp->rcv_nxt += tlen; 3099 thflags = th->th_flags & TH_FIN; 3100 TCPSTAT_INC(tcps_rcvpack); 3101 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3102 SOCKBUF_LOCK(&so->so_rcv); 3103 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3104 m_freem(m); 3105 else 3106 sbappendstream_locked(&so->so_rcv, m, 0); 3107 /* NB: sorwakeup_locked() does an implicit unlock. */ 3108 sorwakeup_locked(so); 3109 } else { 3110 /* 3111 * XXX: Due to the header drop above "th" is 3112 * theoretically invalid by now. Fortunately 3113 * m_adj() doesn't actually frees any mbufs 3114 * when trimming from the head. 3115 */ 3116 thflags = tcp_reass(tp, th, &tlen, m); 3117 tp->t_flags |= TF_ACKNOW; 3118 } 3119 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3120 tcp_update_sack_list(tp, save_start, save_start + tlen); 3121 #if 0 3122 /* 3123 * Note the amount of data that peer has sent into 3124 * our window, in order to estimate the sender's 3125 * buffer size. 3126 * XXX: Unused. 3127 */ 3128 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3129 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3130 else 3131 len = so->so_rcv.sb_hiwat; 3132 #endif 3133 } else { 3134 m_freem(m); 3135 thflags &= ~TH_FIN; 3136 } 3137 3138 /* 3139 * If FIN is received ACK the FIN and let the user know 3140 * that the connection is closing. 3141 */ 3142 if (thflags & TH_FIN) { 3143 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3144 socantrcvmore(so); 3145 /* 3146 * If connection is half-synchronized 3147 * (ie NEEDSYN flag on) then delay ACK, 3148 * so it may be piggybacked when SYN is sent. 3149 * Otherwise, since we received a FIN then no 3150 * more input can be expected, send ACK now. 3151 */ 3152 if (tp->t_flags & TF_NEEDSYN) 3153 tp->t_flags |= TF_DELACK; 3154 else 3155 tp->t_flags |= TF_ACKNOW; 3156 tp->rcv_nxt++; 3157 } 3158 switch (tp->t_state) { 3159 3160 /* 3161 * In SYN_RECEIVED and ESTABLISHED STATES 3162 * enter the CLOSE_WAIT state. 3163 */ 3164 case TCPS_SYN_RECEIVED: 3165 tp->t_starttime = ticks; 3166 /* FALLTHROUGH */ 3167 case TCPS_ESTABLISHED: 3168 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3169 break; 3170 3171 /* 3172 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3173 * enter the CLOSING state. 3174 */ 3175 case TCPS_FIN_WAIT_1: 3176 tcp_state_change(tp, TCPS_CLOSING); 3177 break; 3178 3179 /* 3180 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3181 * starting the time-wait timer, turning off the other 3182 * standard timers. 3183 */ 3184 case TCPS_FIN_WAIT_2: 3185 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3186 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3187 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3188 ti_locked)); 3189 3190 tcp_twstart(tp); 3191 INP_INFO_RUNLOCK(&V_tcbinfo); 3192 return; 3193 } 3194 } 3195 if (ti_locked == TI_RLOCKED) 3196 INP_INFO_RUNLOCK(&V_tcbinfo); 3197 ti_locked = TI_UNLOCKED; 3198 3199 #ifdef TCPDEBUG 3200 if (so->so_options & SO_DEBUG) 3201 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3202 &tcp_savetcp, 0); 3203 #endif 3204 TCP_PROBE3(debug__input, tp, th, m); 3205 3206 /* 3207 * Return any desired output. 3208 */ 3209 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3210 (void) tp->t_fb->tfb_tcp_output(tp); 3211 3212 check_delack: 3213 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3214 __func__, ti_locked)); 3215 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3216 INP_WLOCK_ASSERT(tp->t_inpcb); 3217 3218 if (tp->t_flags & TF_DELACK) { 3219 tp->t_flags &= ~TF_DELACK; 3220 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3221 } 3222 INP_WUNLOCK(tp->t_inpcb); 3223 return; 3224 3225 dropafterack: 3226 /* 3227 * Generate an ACK dropping incoming segment if it occupies 3228 * sequence space, where the ACK reflects our state. 3229 * 3230 * We can now skip the test for the RST flag since all 3231 * paths to this code happen after packets containing 3232 * RST have been dropped. 3233 * 3234 * In the SYN-RECEIVED state, don't send an ACK unless the 3235 * segment we received passes the SYN-RECEIVED ACK test. 3236 * If it fails send a RST. This breaks the loop in the 3237 * "LAND" DoS attack, and also prevents an ACK storm 3238 * between two listening ports that have been sent forged 3239 * SYN segments, each with the source address of the other. 3240 */ 3241 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3242 (SEQ_GT(tp->snd_una, th->th_ack) || 3243 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3244 rstreason = BANDLIM_RST_OPENPORT; 3245 goto dropwithreset; 3246 } 3247 #ifdef TCPDEBUG 3248 if (so->so_options & SO_DEBUG) 3249 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3250 &tcp_savetcp, 0); 3251 #endif 3252 TCP_PROBE3(debug__input, tp, th, m); 3253 if (ti_locked == TI_RLOCKED) 3254 INP_INFO_RUNLOCK(&V_tcbinfo); 3255 ti_locked = TI_UNLOCKED; 3256 3257 tp->t_flags |= TF_ACKNOW; 3258 (void) tp->t_fb->tfb_tcp_output(tp); 3259 INP_WUNLOCK(tp->t_inpcb); 3260 m_freem(m); 3261 return; 3262 3263 dropwithreset: 3264 if (ti_locked == TI_RLOCKED) 3265 INP_INFO_RUNLOCK(&V_tcbinfo); 3266 ti_locked = TI_UNLOCKED; 3267 3268 if (tp != NULL) { 3269 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3270 INP_WUNLOCK(tp->t_inpcb); 3271 } else 3272 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3273 return; 3274 3275 drop: 3276 if (ti_locked == TI_RLOCKED) { 3277 INP_INFO_RUNLOCK(&V_tcbinfo); 3278 ti_locked = TI_UNLOCKED; 3279 } 3280 #ifdef INVARIANTS 3281 else 3282 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3283 #endif 3284 3285 /* 3286 * Drop space held by incoming segment and return. 3287 */ 3288 #ifdef TCPDEBUG 3289 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3290 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3291 &tcp_savetcp, 0); 3292 #endif 3293 TCP_PROBE3(debug__input, tp, th, m); 3294 if (tp != NULL) 3295 INP_WUNLOCK(tp->t_inpcb); 3296 m_freem(m); 3297 #ifndef TCP_RFC7413 3298 #undef tfo_syn 3299 #endif 3300 } 3301 3302 /* 3303 * Issue RST and make ACK acceptable to originator of segment. 3304 * The mbuf must still include the original packet header. 3305 * tp may be NULL. 3306 */ 3307 void 3308 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3309 int tlen, int rstreason) 3310 { 3311 #ifdef INET 3312 struct ip *ip; 3313 #endif 3314 #ifdef INET6 3315 struct ip6_hdr *ip6; 3316 #endif 3317 3318 if (tp != NULL) { 3319 INP_WLOCK_ASSERT(tp->t_inpcb); 3320 } 3321 3322 /* Don't bother if destination was broadcast/multicast. */ 3323 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3324 goto drop; 3325 #ifdef INET6 3326 if (mtod(m, struct ip *)->ip_v == 6) { 3327 ip6 = mtod(m, struct ip6_hdr *); 3328 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3329 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3330 goto drop; 3331 /* IPv6 anycast check is done at tcp6_input() */ 3332 } 3333 #endif 3334 #if defined(INET) && defined(INET6) 3335 else 3336 #endif 3337 #ifdef INET 3338 { 3339 ip = mtod(m, struct ip *); 3340 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3341 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3342 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3343 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3344 goto drop; 3345 } 3346 #endif 3347 3348 /* Perform bandwidth limiting. */ 3349 if (badport_bandlim(rstreason) < 0) 3350 goto drop; 3351 3352 /* tcp_respond consumes the mbuf chain. */ 3353 if (th->th_flags & TH_ACK) { 3354 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3355 th->th_ack, TH_RST); 3356 } else { 3357 if (th->th_flags & TH_SYN) 3358 tlen++; 3359 if (th->th_flags & TH_FIN) 3360 tlen++; 3361 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3362 (tcp_seq)0, TH_RST|TH_ACK); 3363 } 3364 return; 3365 drop: 3366 m_freem(m); 3367 } 3368 3369 /* 3370 * Parse TCP options and place in tcpopt. 3371 */ 3372 void 3373 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3374 { 3375 int opt, optlen; 3376 3377 to->to_flags = 0; 3378 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3379 opt = cp[0]; 3380 if (opt == TCPOPT_EOL) 3381 break; 3382 if (opt == TCPOPT_NOP) 3383 optlen = 1; 3384 else { 3385 if (cnt < 2) 3386 break; 3387 optlen = cp[1]; 3388 if (optlen < 2 || optlen > cnt) 3389 break; 3390 } 3391 switch (opt) { 3392 case TCPOPT_MAXSEG: 3393 if (optlen != TCPOLEN_MAXSEG) 3394 continue; 3395 if (!(flags & TO_SYN)) 3396 continue; 3397 to->to_flags |= TOF_MSS; 3398 bcopy((char *)cp + 2, 3399 (char *)&to->to_mss, sizeof(to->to_mss)); 3400 to->to_mss = ntohs(to->to_mss); 3401 break; 3402 case TCPOPT_WINDOW: 3403 if (optlen != TCPOLEN_WINDOW) 3404 continue; 3405 if (!(flags & TO_SYN)) 3406 continue; 3407 to->to_flags |= TOF_SCALE; 3408 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3409 break; 3410 case TCPOPT_TIMESTAMP: 3411 if (optlen != TCPOLEN_TIMESTAMP) 3412 continue; 3413 to->to_flags |= TOF_TS; 3414 bcopy((char *)cp + 2, 3415 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3416 to->to_tsval = ntohl(to->to_tsval); 3417 bcopy((char *)cp + 6, 3418 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3419 to->to_tsecr = ntohl(to->to_tsecr); 3420 break; 3421 #ifdef TCP_SIGNATURE 3422 /* 3423 * XXX In order to reply to a host which has set the 3424 * TCP_SIGNATURE option in its initial SYN, we have to 3425 * record the fact that the option was observed here 3426 * for the syncache code to perform the correct response. 3427 */ 3428 case TCPOPT_SIGNATURE: 3429 if (optlen != TCPOLEN_SIGNATURE) 3430 continue; 3431 to->to_flags |= TOF_SIGNATURE; 3432 to->to_signature = cp + 2; 3433 break; 3434 #endif 3435 case TCPOPT_SACK_PERMITTED: 3436 if (optlen != TCPOLEN_SACK_PERMITTED) 3437 continue; 3438 if (!(flags & TO_SYN)) 3439 continue; 3440 if (!V_tcp_do_sack) 3441 continue; 3442 to->to_flags |= TOF_SACKPERM; 3443 break; 3444 case TCPOPT_SACK: 3445 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3446 continue; 3447 if (flags & TO_SYN) 3448 continue; 3449 to->to_flags |= TOF_SACK; 3450 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3451 to->to_sacks = cp + 2; 3452 TCPSTAT_INC(tcps_sack_rcv_blocks); 3453 break; 3454 #ifdef TCP_RFC7413 3455 case TCPOPT_FAST_OPEN: 3456 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3457 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3458 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3459 continue; 3460 if (!(flags & TO_SYN)) 3461 continue; 3462 if (!V_tcp_fastopen_enabled) 3463 continue; 3464 to->to_flags |= TOF_FASTOPEN; 3465 to->to_tfo_len = optlen - 2; 3466 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3467 break; 3468 #endif 3469 default: 3470 continue; 3471 } 3472 } 3473 } 3474 3475 /* 3476 * Pull out of band byte out of a segment so 3477 * it doesn't appear in the user's data queue. 3478 * It is still reflected in the segment length for 3479 * sequencing purposes. 3480 */ 3481 void 3482 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3483 int off) 3484 { 3485 int cnt = off + th->th_urp - 1; 3486 3487 while (cnt >= 0) { 3488 if (m->m_len > cnt) { 3489 char *cp = mtod(m, caddr_t) + cnt; 3490 struct tcpcb *tp = sototcpcb(so); 3491 3492 INP_WLOCK_ASSERT(tp->t_inpcb); 3493 3494 tp->t_iobc = *cp; 3495 tp->t_oobflags |= TCPOOB_HAVEDATA; 3496 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3497 m->m_len--; 3498 if (m->m_flags & M_PKTHDR) 3499 m->m_pkthdr.len--; 3500 return; 3501 } 3502 cnt -= m->m_len; 3503 m = m->m_next; 3504 if (m == NULL) 3505 break; 3506 } 3507 panic("tcp_pulloutofband"); 3508 } 3509 3510 /* 3511 * Collect new round-trip time estimate 3512 * and update averages and current timeout. 3513 */ 3514 void 3515 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3516 { 3517 int delta; 3518 3519 INP_WLOCK_ASSERT(tp->t_inpcb); 3520 3521 TCPSTAT_INC(tcps_rttupdated); 3522 tp->t_rttupdated++; 3523 if (tp->t_srtt != 0) { 3524 /* 3525 * srtt is stored as fixed point with 5 bits after the 3526 * binary point (i.e., scaled by 8). The following magic 3527 * is equivalent to the smoothing algorithm in rfc793 with 3528 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3529 * point). Adjust rtt to origin 0. 3530 */ 3531 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3532 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3533 3534 if ((tp->t_srtt += delta) <= 0) 3535 tp->t_srtt = 1; 3536 3537 /* 3538 * We accumulate a smoothed rtt variance (actually, a 3539 * smoothed mean difference), then set the retransmit 3540 * timer to smoothed rtt + 4 times the smoothed variance. 3541 * rttvar is stored as fixed point with 4 bits after the 3542 * binary point (scaled by 16). The following is 3543 * equivalent to rfc793 smoothing with an alpha of .75 3544 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3545 * rfc793's wired-in beta. 3546 */ 3547 if (delta < 0) 3548 delta = -delta; 3549 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3550 if ((tp->t_rttvar += delta) <= 0) 3551 tp->t_rttvar = 1; 3552 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3553 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3554 } else { 3555 /* 3556 * No rtt measurement yet - use the unsmoothed rtt. 3557 * Set the variance to half the rtt (so our first 3558 * retransmit happens at 3*rtt). 3559 */ 3560 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3561 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3562 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3563 } 3564 tp->t_rtttime = 0; 3565 tp->t_rxtshift = 0; 3566 3567 /* 3568 * the retransmit should happen at rtt + 4 * rttvar. 3569 * Because of the way we do the smoothing, srtt and rttvar 3570 * will each average +1/2 tick of bias. When we compute 3571 * the retransmit timer, we want 1/2 tick of rounding and 3572 * 1 extra tick because of +-1/2 tick uncertainty in the 3573 * firing of the timer. The bias will give us exactly the 3574 * 1.5 tick we need. But, because the bias is 3575 * statistical, we have to test that we don't drop below 3576 * the minimum feasible timer (which is 2 ticks). 3577 */ 3578 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3579 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3580 3581 /* 3582 * We received an ack for a packet that wasn't retransmitted; 3583 * it is probably safe to discard any error indications we've 3584 * received recently. This isn't quite right, but close enough 3585 * for now (a route might have failed after we sent a segment, 3586 * and the return path might not be symmetrical). 3587 */ 3588 tp->t_softerror = 0; 3589 } 3590 3591 /* 3592 * Determine a reasonable value for maxseg size. 3593 * If the route is known, check route for mtu. 3594 * If none, use an mss that can be handled on the outgoing interface 3595 * without forcing IP to fragment. If no route is found, route has no mtu, 3596 * or the destination isn't local, use a default, hopefully conservative 3597 * size (usually 512 or the default IP max size, but no more than the mtu 3598 * of the interface), as we can't discover anything about intervening 3599 * gateways or networks. We also initialize the congestion/slow start 3600 * window to be a single segment if the destination isn't local. 3601 * While looking at the routing entry, we also initialize other path-dependent 3602 * parameters from pre-set or cached values in the routing entry. 3603 * 3604 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3605 * IP options, e.g. IPSEC data, since length of this data may vary, and 3606 * thus it is calculated for every segment separately in tcp_output(). 3607 * 3608 * NOTE that this routine is only called when we process an incoming 3609 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3610 * settings are handled in tcp_mssopt(). 3611 */ 3612 void 3613 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3614 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3615 { 3616 int mss = 0; 3617 uint32_t maxmtu = 0; 3618 struct inpcb *inp = tp->t_inpcb; 3619 struct hc_metrics_lite metrics; 3620 #ifdef INET6 3621 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3622 size_t min_protoh = isipv6 ? 3623 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3624 sizeof (struct tcpiphdr); 3625 #else 3626 const size_t min_protoh = sizeof(struct tcpiphdr); 3627 #endif 3628 3629 INP_WLOCK_ASSERT(tp->t_inpcb); 3630 3631 if (mtuoffer != -1) { 3632 KASSERT(offer == -1, ("%s: conflict", __func__)); 3633 offer = mtuoffer - min_protoh; 3634 } 3635 3636 /* Initialize. */ 3637 #ifdef INET6 3638 if (isipv6) { 3639 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3640 tp->t_maxseg = V_tcp_v6mssdflt; 3641 } 3642 #endif 3643 #if defined(INET) && defined(INET6) 3644 else 3645 #endif 3646 #ifdef INET 3647 { 3648 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3649 tp->t_maxseg = V_tcp_mssdflt; 3650 } 3651 #endif 3652 3653 /* 3654 * No route to sender, stay with default mss and return. 3655 */ 3656 if (maxmtu == 0) { 3657 /* 3658 * In case we return early we need to initialize metrics 3659 * to a defined state as tcp_hc_get() would do for us 3660 * if there was no cache hit. 3661 */ 3662 if (metricptr != NULL) 3663 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3664 return; 3665 } 3666 3667 /* What have we got? */ 3668 switch (offer) { 3669 case 0: 3670 /* 3671 * Offer == 0 means that there was no MSS on the SYN 3672 * segment, in this case we use tcp_mssdflt as 3673 * already assigned to t_maxseg above. 3674 */ 3675 offer = tp->t_maxseg; 3676 break; 3677 3678 case -1: 3679 /* 3680 * Offer == -1 means that we didn't receive SYN yet. 3681 */ 3682 /* FALLTHROUGH */ 3683 3684 default: 3685 /* 3686 * Prevent DoS attack with too small MSS. Round up 3687 * to at least minmss. 3688 */ 3689 offer = max(offer, V_tcp_minmss); 3690 } 3691 3692 /* 3693 * rmx information is now retrieved from tcp_hostcache. 3694 */ 3695 tcp_hc_get(&inp->inp_inc, &metrics); 3696 if (metricptr != NULL) 3697 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3698 3699 /* 3700 * If there's a discovered mtu in tcp hostcache, use it. 3701 * Else, use the link mtu. 3702 */ 3703 if (metrics.rmx_mtu) 3704 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3705 else { 3706 #ifdef INET6 3707 if (isipv6) { 3708 mss = maxmtu - min_protoh; 3709 if (!V_path_mtu_discovery && 3710 !in6_localaddr(&inp->in6p_faddr)) 3711 mss = min(mss, V_tcp_v6mssdflt); 3712 } 3713 #endif 3714 #if defined(INET) && defined(INET6) 3715 else 3716 #endif 3717 #ifdef INET 3718 { 3719 mss = maxmtu - min_protoh; 3720 if (!V_path_mtu_discovery && 3721 !in_localaddr(inp->inp_faddr)) 3722 mss = min(mss, V_tcp_mssdflt); 3723 } 3724 #endif 3725 /* 3726 * XXX - The above conditional (mss = maxmtu - min_protoh) 3727 * probably violates the TCP spec. 3728 * The problem is that, since we don't know the 3729 * other end's MSS, we are supposed to use a conservative 3730 * default. But, if we do that, then MTU discovery will 3731 * never actually take place, because the conservative 3732 * default is much less than the MTUs typically seen 3733 * on the Internet today. For the moment, we'll sweep 3734 * this under the carpet. 3735 * 3736 * The conservative default might not actually be a problem 3737 * if the only case this occurs is when sending an initial 3738 * SYN with options and data to a host we've never talked 3739 * to before. Then, they will reply with an MSS value which 3740 * will get recorded and the new parameters should get 3741 * recomputed. For Further Study. 3742 */ 3743 } 3744 mss = min(mss, offer); 3745 3746 /* 3747 * Sanity check: make sure that maxseg will be large 3748 * enough to allow some data on segments even if the 3749 * all the option space is used (40bytes). Otherwise 3750 * funny things may happen in tcp_output. 3751 * 3752 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3753 */ 3754 mss = max(mss, 64); 3755 3756 tp->t_maxseg = mss; 3757 } 3758 3759 void 3760 tcp_mss(struct tcpcb *tp, int offer) 3761 { 3762 int mss; 3763 uint32_t bufsize; 3764 struct inpcb *inp; 3765 struct socket *so; 3766 struct hc_metrics_lite metrics; 3767 struct tcp_ifcap cap; 3768 3769 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3770 3771 bzero(&cap, sizeof(cap)); 3772 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3773 3774 mss = tp->t_maxseg; 3775 inp = tp->t_inpcb; 3776 3777 /* 3778 * If there's a pipesize, change the socket buffer to that size, 3779 * don't change if sb_hiwat is different than default (then it 3780 * has been changed on purpose with setsockopt). 3781 * Make the socket buffers an integral number of mss units; 3782 * if the mss is larger than the socket buffer, decrease the mss. 3783 */ 3784 so = inp->inp_socket; 3785 SOCKBUF_LOCK(&so->so_snd); 3786 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3787 bufsize = metrics.rmx_sendpipe; 3788 else 3789 bufsize = so->so_snd.sb_hiwat; 3790 if (bufsize < mss) 3791 mss = bufsize; 3792 else { 3793 bufsize = roundup(bufsize, mss); 3794 if (bufsize > sb_max) 3795 bufsize = sb_max; 3796 if (bufsize > so->so_snd.sb_hiwat) 3797 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3798 } 3799 SOCKBUF_UNLOCK(&so->so_snd); 3800 /* 3801 * Sanity check: make sure that maxseg will be large 3802 * enough to allow some data on segments even if the 3803 * all the option space is used (40bytes). Otherwise 3804 * funny things may happen in tcp_output. 3805 * 3806 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3807 */ 3808 tp->t_maxseg = max(mss, 64); 3809 3810 SOCKBUF_LOCK(&so->so_rcv); 3811 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3812 bufsize = metrics.rmx_recvpipe; 3813 else 3814 bufsize = so->so_rcv.sb_hiwat; 3815 if (bufsize > mss) { 3816 bufsize = roundup(bufsize, mss); 3817 if (bufsize > sb_max) 3818 bufsize = sb_max; 3819 if (bufsize > so->so_rcv.sb_hiwat) 3820 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3821 } 3822 SOCKBUF_UNLOCK(&so->so_rcv); 3823 3824 /* Check the interface for TSO capabilities. */ 3825 if (cap.ifcap & CSUM_TSO) { 3826 tp->t_flags |= TF_TSO; 3827 tp->t_tsomax = cap.tsomax; 3828 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3829 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3830 } 3831 } 3832 3833 /* 3834 * Determine the MSS option to send on an outgoing SYN. 3835 */ 3836 int 3837 tcp_mssopt(struct in_conninfo *inc) 3838 { 3839 int mss = 0; 3840 uint32_t thcmtu = 0; 3841 uint32_t maxmtu = 0; 3842 size_t min_protoh; 3843 3844 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3845 3846 #ifdef INET6 3847 if (inc->inc_flags & INC_ISIPV6) { 3848 mss = V_tcp_v6mssdflt; 3849 maxmtu = tcp_maxmtu6(inc, NULL); 3850 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3851 } 3852 #endif 3853 #if defined(INET) && defined(INET6) 3854 else 3855 #endif 3856 #ifdef INET 3857 { 3858 mss = V_tcp_mssdflt; 3859 maxmtu = tcp_maxmtu(inc, NULL); 3860 min_protoh = sizeof(struct tcpiphdr); 3861 } 3862 #endif 3863 #if defined(INET6) || defined(INET) 3864 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3865 #endif 3866 3867 if (maxmtu && thcmtu) 3868 mss = min(maxmtu, thcmtu) - min_protoh; 3869 else if (maxmtu || thcmtu) 3870 mss = max(maxmtu, thcmtu) - min_protoh; 3871 3872 return (mss); 3873 } 3874 3875 3876 /* 3877 * On a partial ack arrives, force the retransmission of the 3878 * next unacknowledged segment. Do not clear tp->t_dupacks. 3879 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3880 * be started again. 3881 */ 3882 void 3883 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3884 { 3885 tcp_seq onxt = tp->snd_nxt; 3886 uint32_t ocwnd = tp->snd_cwnd; 3887 u_int maxseg = tcp_maxseg(tp); 3888 3889 INP_WLOCK_ASSERT(tp->t_inpcb); 3890 3891 tcp_timer_activate(tp, TT_REXMT, 0); 3892 tp->t_rtttime = 0; 3893 tp->snd_nxt = th->th_ack; 3894 /* 3895 * Set snd_cwnd to one segment beyond acknowledged offset. 3896 * (tp->snd_una has not yet been updated when this function is called.) 3897 */ 3898 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3899 tp->t_flags |= TF_ACKNOW; 3900 (void) tp->t_fb->tfb_tcp_output(tp); 3901 tp->snd_cwnd = ocwnd; 3902 if (SEQ_GT(onxt, tp->snd_nxt)) 3903 tp->snd_nxt = onxt; 3904 /* 3905 * Partial window deflation. Relies on fact that tp->snd_una 3906 * not updated yet. 3907 */ 3908 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3909 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3910 else 3911 tp->snd_cwnd = 0; 3912 tp->snd_cwnd += maxseg; 3913 } 3914 3915 int 3916 tcp_compute_pipe(struct tcpcb *tp) 3917 { 3918 return (tp->snd_max - tp->snd_una + 3919 tp->sackhint.sack_bytes_rexmit - 3920 tp->sackhint.sacked_bytes); 3921 } 3922