1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 3. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_inet.h" 54 #include "opt_inet6.h" 55 #include "opt_ipsec.h" 56 #include "opt_tcpdebug.h" 57 58 #include <sys/param.h> 59 #include <sys/kernel.h> 60 #ifdef TCP_HHOOK 61 #include <sys/hhook.h> 62 #endif 63 #include <sys/malloc.h> 64 #include <sys/mbuf.h> 65 #include <sys/proc.h> /* for proc0 declaration */ 66 #include <sys/protosw.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77 #include <vm/uma.h> 78 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/route.h> 82 #include <net/vnet.h> 83 84 #define TCPSTATES /* for logging */ 85 86 #include <netinet/in.h> 87 #include <netinet/in_kdtrace.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/ip.h> 91 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 92 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 93 #include <netinet/ip_var.h> 94 #include <netinet/ip_options.h> 95 #include <netinet/ip6.h> 96 #include <netinet/icmp6.h> 97 #include <netinet6/in6_pcb.h> 98 #include <netinet6/in6_var.h> 99 #include <netinet6/ip6_var.h> 100 #include <netinet6/nd6.h> 101 #ifdef TCP_RFC7413 102 #include <netinet/tcp_fastopen.h> 103 #endif 104 #include <netinet/tcp.h> 105 #include <netinet/tcp_fsm.h> 106 #include <netinet/tcp_seq.h> 107 #include <netinet/tcp_timer.h> 108 #include <netinet/tcp_var.h> 109 #include <netinet6/tcp6_var.h> 110 #include <netinet/tcpip.h> 111 #include <netinet/cc/cc.h> 112 #ifdef TCPPCAP 113 #include <netinet/tcp_pcap.h> 114 #endif 115 #include <netinet/tcp_syncache.h> 116 #ifdef TCPDEBUG 117 #include <netinet/tcp_debug.h> 118 #endif /* TCPDEBUG */ 119 #ifdef TCP_OFFLOAD 120 #include <netinet/tcp_offload.h> 121 #endif 122 123 #include <netipsec/ipsec_support.h> 124 125 #include <machine/in_cksum.h> 126 127 #include <security/mac/mac_framework.h> 128 129 const int tcprexmtthresh = 3; 130 131 int tcp_log_in_vain = 0; 132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 133 &tcp_log_in_vain, 0, 134 "Log all incoming TCP segments to closed ports"); 135 136 VNET_DEFINE(int, blackhole) = 0; 137 #define V_blackhole VNET(blackhole) 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(blackhole), 0, 140 "Do not send RST on segments to closed ports"); 141 142 VNET_DEFINE(int, tcp_delack_enabled) = 1; 143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 144 &VNET_NAME(tcp_delack_enabled), 0, 145 "Delay ACK to try and piggyback it onto a data packet"); 146 147 VNET_DEFINE(int, drop_synfin) = 0; 148 #define V_drop_synfin VNET(drop_synfin) 149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 150 &VNET_NAME(drop_synfin), 0, 151 "Drop TCP packets with SYN+FIN set"); 152 153 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 155 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 156 "Use calculated pipe/in-flight bytes per RFC 6675"); 157 158 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 159 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 161 &VNET_NAME(tcp_do_rfc3042), 0, 162 "Enable RFC 3042 (Limited Transmit)"); 163 164 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 166 &VNET_NAME(tcp_do_rfc3390), 0, 167 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 168 169 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 171 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 172 "Slow-start flight size (initial congestion window) in number of segments"); 173 174 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 176 &VNET_NAME(tcp_do_rfc3465), 0, 177 "Enable RFC 3465 (Appropriate Byte Counting)"); 178 179 VNET_DEFINE(int, tcp_abc_l_var) = 2; 180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_abc_l_var), 2, 182 "Cap the max cwnd increment during slow-start to this number of segments"); 183 184 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 185 186 VNET_DEFINE(int, tcp_do_ecn) = 2; 187 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 188 &VNET_NAME(tcp_do_ecn), 0, 189 "TCP ECN support"); 190 191 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 192 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_ecn_maxretries), 0, 194 "Max retries before giving up on ECN"); 195 196 VNET_DEFINE(int, tcp_insecure_syn) = 0; 197 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_insecure_syn), 0, 200 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 201 202 VNET_DEFINE(int, tcp_insecure_rst) = 0; 203 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 205 &VNET_NAME(tcp_insecure_rst), 0, 206 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 207 208 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 209 #define V_tcp_recvspace VNET(tcp_recvspace) 210 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 211 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 212 213 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 214 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 216 &VNET_NAME(tcp_do_autorcvbuf), 0, 217 "Enable automatic receive buffer sizing"); 218 219 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 220 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_autorcvbuf_inc), 0, 223 "Incrementor step size of automatic receive buffer"); 224 225 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 226 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_autorcvbuf_max), 0, 229 "Max size of automatic receive buffer"); 230 231 VNET_DEFINE(struct inpcbhead, tcb); 232 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 233 VNET_DEFINE(struct inpcbinfo, tcbinfo); 234 235 /* 236 * TCP statistics are stored in an array of counter(9)s, which size matches 237 * size of struct tcpstat. TCP running connection count is a regular array. 238 */ 239 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 240 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 241 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 242 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 243 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 244 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 245 "TCP connection counts by TCP state"); 246 247 static void 248 tcp_vnet_init(const void *unused) 249 { 250 251 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 252 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 253 } 254 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 255 tcp_vnet_init, NULL); 256 257 #ifdef VIMAGE 258 static void 259 tcp_vnet_uninit(const void *unused) 260 { 261 262 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 263 VNET_PCPUSTAT_FREE(tcpstat); 264 } 265 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 266 tcp_vnet_uninit, NULL); 267 #endif /* VIMAGE */ 268 269 /* 270 * Kernel module interface for updating tcpstat. The argument is an index 271 * into tcpstat treated as an array. 272 */ 273 void 274 kmod_tcpstat_inc(int statnum) 275 { 276 277 counter_u64_add(VNET(tcpstat)[statnum], 1); 278 } 279 280 #ifdef TCP_HHOOK 281 /* 282 * Wrapper for the TCP established input helper hook. 283 */ 284 void 285 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 286 { 287 struct tcp_hhook_data hhook_data; 288 289 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 290 hhook_data.tp = tp; 291 hhook_data.th = th; 292 hhook_data.to = to; 293 294 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 295 tp->osd); 296 } 297 } 298 #endif 299 300 /* 301 * CC wrapper hook functions 302 */ 303 void 304 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 305 uint16_t type) 306 { 307 INP_WLOCK_ASSERT(tp->t_inpcb); 308 309 tp->ccv->nsegs = nsegs; 310 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 311 if (tp->snd_cwnd <= tp->snd_wnd) 312 tp->ccv->flags |= CCF_CWND_LIMITED; 313 else 314 tp->ccv->flags &= ~CCF_CWND_LIMITED; 315 316 if (type == CC_ACK) { 317 if (tp->snd_cwnd > tp->snd_ssthresh) { 318 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 319 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 320 if (tp->t_bytes_acked >= tp->snd_cwnd) { 321 tp->t_bytes_acked -= tp->snd_cwnd; 322 tp->ccv->flags |= CCF_ABC_SENTAWND; 323 } 324 } else { 325 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 326 tp->t_bytes_acked = 0; 327 } 328 } 329 330 if (CC_ALGO(tp)->ack_received != NULL) { 331 /* XXXLAS: Find a way to live without this */ 332 tp->ccv->curack = th->th_ack; 333 CC_ALGO(tp)->ack_received(tp->ccv, type); 334 } 335 } 336 337 void 338 cc_conn_init(struct tcpcb *tp) 339 { 340 struct hc_metrics_lite metrics; 341 struct inpcb *inp = tp->t_inpcb; 342 u_int maxseg; 343 int rtt; 344 345 INP_WLOCK_ASSERT(tp->t_inpcb); 346 347 tcp_hc_get(&inp->inp_inc, &metrics); 348 maxseg = tcp_maxseg(tp); 349 350 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 351 tp->t_srtt = rtt; 352 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 353 TCPSTAT_INC(tcps_usedrtt); 354 if (metrics.rmx_rttvar) { 355 tp->t_rttvar = metrics.rmx_rttvar; 356 TCPSTAT_INC(tcps_usedrttvar); 357 } else { 358 /* default variation is +- 1 rtt */ 359 tp->t_rttvar = 360 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 361 } 362 TCPT_RANGESET(tp->t_rxtcur, 363 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 364 tp->t_rttmin, TCPTV_REXMTMAX); 365 } 366 if (metrics.rmx_ssthresh) { 367 /* 368 * There's some sort of gateway or interface 369 * buffer limit on the path. Use this to set 370 * the slow start threshold, but set the 371 * threshold to no less than 2*mss. 372 */ 373 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 374 TCPSTAT_INC(tcps_usedssthresh); 375 } 376 377 /* 378 * Set the initial slow-start flight size. 379 * 380 * RFC5681 Section 3.1 specifies the default conservative values. 381 * RFC3390 specifies slightly more aggressive values. 382 * RFC6928 increases it to ten segments. 383 * Support for user specified value for initial flight size. 384 * 385 * If a SYN or SYN/ACK was lost and retransmitted, we have to 386 * reduce the initial CWND to one segment as congestion is likely 387 * requiring us to be cautious. 388 */ 389 if (tp->snd_cwnd == 1) 390 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 391 else if (V_tcp_initcwnd_segments) 392 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 393 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 394 else if (V_tcp_do_rfc3390) 395 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 396 else { 397 /* Per RFC5681 Section 3.1 */ 398 if (maxseg > 2190) 399 tp->snd_cwnd = 2 * maxseg; 400 else if (maxseg > 1095) 401 tp->snd_cwnd = 3 * maxseg; 402 else 403 tp->snd_cwnd = 4 * maxseg; 404 } 405 406 if (CC_ALGO(tp)->conn_init != NULL) 407 CC_ALGO(tp)->conn_init(tp->ccv); 408 } 409 410 void inline 411 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 412 { 413 u_int maxseg; 414 415 INP_WLOCK_ASSERT(tp->t_inpcb); 416 417 switch(type) { 418 case CC_NDUPACK: 419 if (!IN_FASTRECOVERY(tp->t_flags)) { 420 tp->snd_recover = tp->snd_max; 421 if (tp->t_flags & TF_ECN_PERMIT) 422 tp->t_flags |= TF_ECN_SND_CWR; 423 } 424 break; 425 case CC_ECN: 426 if (!IN_CONGRECOVERY(tp->t_flags)) { 427 TCPSTAT_INC(tcps_ecn_rcwnd); 428 tp->snd_recover = tp->snd_max; 429 if (tp->t_flags & TF_ECN_PERMIT) 430 tp->t_flags |= TF_ECN_SND_CWR; 431 } 432 break; 433 case CC_RTO: 434 maxseg = tcp_maxseg(tp); 435 tp->t_dupacks = 0; 436 tp->t_bytes_acked = 0; 437 EXIT_RECOVERY(tp->t_flags); 438 if (CC_ALGO(tp)->cong_signal == NULL) { 439 /* 440 * RFC5681 Section 3.1 441 * ssthresh = max (FlightSize / 2, 2*SMSS) eq (4) 442 */ 443 tp->snd_ssthresh = 444 max((tp->snd_max - tp->snd_una) / 2 / maxseg, 2) 445 * maxseg; 446 tp->snd_cwnd = maxseg; 447 } 448 break; 449 case CC_RTO_ERR: 450 TCPSTAT_INC(tcps_sndrexmitbad); 451 /* RTO was unnecessary, so reset everything. */ 452 tp->snd_cwnd = tp->snd_cwnd_prev; 453 tp->snd_ssthresh = tp->snd_ssthresh_prev; 454 tp->snd_recover = tp->snd_recover_prev; 455 if (tp->t_flags & TF_WASFRECOVERY) 456 ENTER_FASTRECOVERY(tp->t_flags); 457 if (tp->t_flags & TF_WASCRECOVERY) 458 ENTER_CONGRECOVERY(tp->t_flags); 459 tp->snd_nxt = tp->snd_max; 460 tp->t_flags &= ~TF_PREVVALID; 461 tp->t_badrxtwin = 0; 462 break; 463 } 464 465 if (CC_ALGO(tp)->cong_signal != NULL) { 466 if (th != NULL) 467 tp->ccv->curack = th->th_ack; 468 CC_ALGO(tp)->cong_signal(tp->ccv, type); 469 } 470 } 471 472 void inline 473 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 474 { 475 INP_WLOCK_ASSERT(tp->t_inpcb); 476 477 /* XXXLAS: KASSERT that we're in recovery? */ 478 479 if (CC_ALGO(tp)->post_recovery != NULL) { 480 tp->ccv->curack = th->th_ack; 481 CC_ALGO(tp)->post_recovery(tp->ccv); 482 } 483 /* XXXLAS: EXIT_RECOVERY ? */ 484 tp->t_bytes_acked = 0; 485 } 486 487 /* 488 * Indicate whether this ack should be delayed. We can delay the ack if 489 * following conditions are met: 490 * - There is no delayed ack timer in progress. 491 * - Our last ack wasn't a 0-sized window. We never want to delay 492 * the ack that opens up a 0-sized window. 493 * - LRO wasn't used for this segment. We make sure by checking that the 494 * segment size is not larger than the MSS. 495 */ 496 #define DELAY_ACK(tp, tlen) \ 497 ((!tcp_timer_active(tp, TT_DELACK) && \ 498 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 499 (tlen <= tp->t_maxseg) && \ 500 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 501 502 static void inline 503 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 504 { 505 INP_WLOCK_ASSERT(tp->t_inpcb); 506 507 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 508 switch (iptos & IPTOS_ECN_MASK) { 509 case IPTOS_ECN_CE: 510 tp->ccv->flags |= CCF_IPHDR_CE; 511 break; 512 case IPTOS_ECN_ECT0: 513 tp->ccv->flags &= ~CCF_IPHDR_CE; 514 break; 515 case IPTOS_ECN_ECT1: 516 tp->ccv->flags &= ~CCF_IPHDR_CE; 517 break; 518 } 519 520 if (th->th_flags & TH_CWR) 521 tp->ccv->flags |= CCF_TCPHDR_CWR; 522 else 523 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 524 525 if (tp->t_flags & TF_DELACK) 526 tp->ccv->flags |= CCF_DELACK; 527 else 528 tp->ccv->flags &= ~CCF_DELACK; 529 530 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 531 532 if (tp->ccv->flags & CCF_ACKNOW) 533 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 534 } 535 } 536 537 /* 538 * TCP input handling is split into multiple parts: 539 * tcp6_input is a thin wrapper around tcp_input for the extended 540 * ip6_protox[] call format in ip6_input 541 * tcp_input handles primary segment validation, inpcb lookup and 542 * SYN processing on listen sockets 543 * tcp_do_segment processes the ACK and text of the segment for 544 * establishing, established and closing connections 545 */ 546 #ifdef INET6 547 int 548 tcp6_input(struct mbuf **mp, int *offp, int proto) 549 { 550 struct mbuf *m = *mp; 551 struct in6_ifaddr *ia6; 552 struct ip6_hdr *ip6; 553 554 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 555 556 /* 557 * draft-itojun-ipv6-tcp-to-anycast 558 * better place to put this in? 559 */ 560 ip6 = mtod(m, struct ip6_hdr *); 561 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 562 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 563 struct ip6_hdr *ip6; 564 565 ifa_free(&ia6->ia_ifa); 566 ip6 = mtod(m, struct ip6_hdr *); 567 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 568 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 569 return (IPPROTO_DONE); 570 } 571 if (ia6) 572 ifa_free(&ia6->ia_ifa); 573 574 return (tcp_input(mp, offp, proto)); 575 } 576 #endif /* INET6 */ 577 578 int 579 tcp_input(struct mbuf **mp, int *offp, int proto) 580 { 581 struct mbuf *m = *mp; 582 struct tcphdr *th = NULL; 583 struct ip *ip = NULL; 584 struct inpcb *inp = NULL; 585 struct tcpcb *tp = NULL; 586 struct socket *so = NULL; 587 u_char *optp = NULL; 588 int off0; 589 int optlen = 0; 590 #ifdef INET 591 int len; 592 #endif 593 int tlen = 0, off; 594 int drop_hdrlen; 595 int thflags; 596 int rstreason = 0; /* For badport_bandlim accounting purposes */ 597 uint8_t iptos; 598 struct m_tag *fwd_tag = NULL; 599 #ifdef INET6 600 struct ip6_hdr *ip6 = NULL; 601 int isipv6; 602 #else 603 const void *ip6 = NULL; 604 #endif /* INET6 */ 605 struct tcpopt to; /* options in this segment */ 606 char *s = NULL; /* address and port logging */ 607 int ti_locked; 608 #ifdef TCPDEBUG 609 /* 610 * The size of tcp_saveipgen must be the size of the max ip header, 611 * now IPv6. 612 */ 613 u_char tcp_saveipgen[IP6_HDR_LEN]; 614 struct tcphdr tcp_savetcp; 615 short ostate = 0; 616 #endif 617 618 #ifdef INET6 619 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 620 #endif 621 622 off0 = *offp; 623 m = *mp; 624 *mp = NULL; 625 to.to_flags = 0; 626 TCPSTAT_INC(tcps_rcvtotal); 627 628 #ifdef INET6 629 if (isipv6) { 630 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 631 632 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 633 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 634 if (m == NULL) { 635 TCPSTAT_INC(tcps_rcvshort); 636 return (IPPROTO_DONE); 637 } 638 } 639 640 ip6 = mtod(m, struct ip6_hdr *); 641 th = (struct tcphdr *)((caddr_t)ip6 + off0); 642 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 643 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 644 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 645 th->th_sum = m->m_pkthdr.csum_data; 646 else 647 th->th_sum = in6_cksum_pseudo(ip6, tlen, 648 IPPROTO_TCP, m->m_pkthdr.csum_data); 649 th->th_sum ^= 0xffff; 650 } else 651 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 652 if (th->th_sum) { 653 TCPSTAT_INC(tcps_rcvbadsum); 654 goto drop; 655 } 656 657 /* 658 * Be proactive about unspecified IPv6 address in source. 659 * As we use all-zero to indicate unbounded/unconnected pcb, 660 * unspecified IPv6 address can be used to confuse us. 661 * 662 * Note that packets with unspecified IPv6 destination is 663 * already dropped in ip6_input. 664 */ 665 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 666 /* XXX stat */ 667 goto drop; 668 } 669 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 670 } 671 #endif 672 #if defined(INET) && defined(INET6) 673 else 674 #endif 675 #ifdef INET 676 { 677 /* 678 * Get IP and TCP header together in first mbuf. 679 * Note: IP leaves IP header in first mbuf. 680 */ 681 if (off0 > sizeof (struct ip)) { 682 ip_stripoptions(m); 683 off0 = sizeof(struct ip); 684 } 685 if (m->m_len < sizeof (struct tcpiphdr)) { 686 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 687 == NULL) { 688 TCPSTAT_INC(tcps_rcvshort); 689 return (IPPROTO_DONE); 690 } 691 } 692 ip = mtod(m, struct ip *); 693 th = (struct tcphdr *)((caddr_t)ip + off0); 694 tlen = ntohs(ip->ip_len) - off0; 695 696 iptos = ip->ip_tos; 697 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 698 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 699 th->th_sum = m->m_pkthdr.csum_data; 700 else 701 th->th_sum = in_pseudo(ip->ip_src.s_addr, 702 ip->ip_dst.s_addr, 703 htonl(m->m_pkthdr.csum_data + tlen + 704 IPPROTO_TCP)); 705 th->th_sum ^= 0xffff; 706 } else { 707 struct ipovly *ipov = (struct ipovly *)ip; 708 709 /* 710 * Checksum extended TCP header and data. 711 */ 712 len = off0 + tlen; 713 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 714 ipov->ih_len = htons(tlen); 715 th->th_sum = in_cksum(m, len); 716 /* Reset length for SDT probes. */ 717 ip->ip_len = htons(len); 718 /* Reset TOS bits */ 719 ip->ip_tos = iptos; 720 /* Re-initialization for later version check */ 721 ip->ip_v = IPVERSION; 722 } 723 724 if (th->th_sum) { 725 TCPSTAT_INC(tcps_rcvbadsum); 726 goto drop; 727 } 728 } 729 #endif /* INET */ 730 731 /* 732 * Check that TCP offset makes sense, 733 * pull out TCP options and adjust length. XXX 734 */ 735 off = th->th_off << 2; 736 if (off < sizeof (struct tcphdr) || off > tlen) { 737 TCPSTAT_INC(tcps_rcvbadoff); 738 goto drop; 739 } 740 tlen -= off; /* tlen is used instead of ti->ti_len */ 741 if (off > sizeof (struct tcphdr)) { 742 #ifdef INET6 743 if (isipv6) { 744 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 745 ip6 = mtod(m, struct ip6_hdr *); 746 th = (struct tcphdr *)((caddr_t)ip6 + off0); 747 } 748 #endif 749 #if defined(INET) && defined(INET6) 750 else 751 #endif 752 #ifdef INET 753 { 754 if (m->m_len < sizeof(struct ip) + off) { 755 if ((m = m_pullup(m, sizeof (struct ip) + off)) 756 == NULL) { 757 TCPSTAT_INC(tcps_rcvshort); 758 return (IPPROTO_DONE); 759 } 760 ip = mtod(m, struct ip *); 761 th = (struct tcphdr *)((caddr_t)ip + off0); 762 } 763 } 764 #endif 765 optlen = off - sizeof (struct tcphdr); 766 optp = (u_char *)(th + 1); 767 } 768 thflags = th->th_flags; 769 770 /* 771 * Convert TCP protocol specific fields to host format. 772 */ 773 tcp_fields_to_host(th); 774 775 /* 776 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 777 */ 778 drop_hdrlen = off0 + off; 779 780 /* 781 * Locate pcb for segment; if we're likely to add or remove a 782 * connection then first acquire pcbinfo lock. There are three cases 783 * where we might discover later we need a write lock despite the 784 * flags: ACKs moving a connection out of the syncache, ACKs for a 785 * connection in TIMEWAIT and SYNs not targeting a listening socket. 786 */ 787 if ((thflags & (TH_FIN | TH_RST)) != 0) { 788 INP_INFO_RLOCK(&V_tcbinfo); 789 ti_locked = TI_RLOCKED; 790 } else 791 ti_locked = TI_UNLOCKED; 792 793 /* 794 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 795 */ 796 if ( 797 #ifdef INET6 798 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 799 #ifdef INET 800 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 801 #endif 802 #endif 803 #if defined(INET) && !defined(INET6) 804 (m->m_flags & M_IP_NEXTHOP) 805 #endif 806 ) 807 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 808 809 findpcb: 810 #ifdef INVARIANTS 811 if (ti_locked == TI_RLOCKED) { 812 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 813 } else { 814 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 815 } 816 #endif 817 #ifdef INET6 818 if (isipv6 && fwd_tag != NULL) { 819 struct sockaddr_in6 *next_hop6; 820 821 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 822 /* 823 * Transparently forwarded. Pretend to be the destination. 824 * Already got one like this? 825 */ 826 inp = in6_pcblookup_mbuf(&V_tcbinfo, 827 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 828 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 829 if (!inp) { 830 /* 831 * It's new. Try to find the ambushing socket. 832 * Because we've rewritten the destination address, 833 * any hardware-generated hash is ignored. 834 */ 835 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 836 th->th_sport, &next_hop6->sin6_addr, 837 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 838 th->th_dport, INPLOOKUP_WILDCARD | 839 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 840 } 841 } else if (isipv6) { 842 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 843 th->th_sport, &ip6->ip6_dst, th->th_dport, 844 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 845 m->m_pkthdr.rcvif, m); 846 } 847 #endif /* INET6 */ 848 #if defined(INET6) && defined(INET) 849 else 850 #endif 851 #ifdef INET 852 if (fwd_tag != NULL) { 853 struct sockaddr_in *next_hop; 854 855 next_hop = (struct sockaddr_in *)(fwd_tag+1); 856 /* 857 * Transparently forwarded. Pretend to be the destination. 858 * already got one like this? 859 */ 860 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 861 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 862 m->m_pkthdr.rcvif, m); 863 if (!inp) { 864 /* 865 * It's new. Try to find the ambushing socket. 866 * Because we've rewritten the destination address, 867 * any hardware-generated hash is ignored. 868 */ 869 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 870 th->th_sport, next_hop->sin_addr, 871 next_hop->sin_port ? ntohs(next_hop->sin_port) : 872 th->th_dport, INPLOOKUP_WILDCARD | 873 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 874 } 875 } else 876 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 877 th->th_sport, ip->ip_dst, th->th_dport, 878 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 879 m->m_pkthdr.rcvif, m); 880 #endif /* INET */ 881 882 /* 883 * If the INPCB does not exist then all data in the incoming 884 * segment is discarded and an appropriate RST is sent back. 885 * XXX MRT Send RST using which routing table? 886 */ 887 if (inp == NULL) { 888 /* 889 * Log communication attempts to ports that are not 890 * in use. 891 */ 892 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 893 tcp_log_in_vain == 2) { 894 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 895 log(LOG_INFO, "%s; %s: Connection attempt " 896 "to closed port\n", s, __func__); 897 } 898 /* 899 * When blackholing do not respond with a RST but 900 * completely ignore the segment and drop it. 901 */ 902 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 903 V_blackhole == 2) 904 goto dropunlock; 905 906 rstreason = BANDLIM_RST_CLOSEDPORT; 907 goto dropwithreset; 908 } 909 INP_WLOCK_ASSERT(inp); 910 /* 911 * While waiting for inp lock during the lookup, another thread 912 * can have dropped the inpcb, in which case we need to loop back 913 * and try to find a new inpcb to deliver to. 914 */ 915 if (inp->inp_flags & INP_DROPPED) { 916 INP_WUNLOCK(inp); 917 inp = NULL; 918 goto findpcb; 919 } 920 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 921 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 922 ((inp->inp_socket == NULL) || 923 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 924 inp->inp_flowid = m->m_pkthdr.flowid; 925 inp->inp_flowtype = M_HASHTYPE_GET(m); 926 } 927 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 928 #ifdef INET6 929 if (isipv6 && IPSEC_ENABLED(ipv6) && 930 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 931 goto dropunlock; 932 } 933 #ifdef INET 934 else 935 #endif 936 #endif /* INET6 */ 937 #ifdef INET 938 if (IPSEC_ENABLED(ipv4) && 939 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 940 goto dropunlock; 941 } 942 #endif /* INET */ 943 #endif /* IPSEC */ 944 945 /* 946 * Check the minimum TTL for socket. 947 */ 948 if (inp->inp_ip_minttl != 0) { 949 #ifdef INET6 950 if (isipv6) { 951 if (inp->inp_ip_minttl > ip6->ip6_hlim) 952 goto dropunlock; 953 } else 954 #endif 955 if (inp->inp_ip_minttl > ip->ip_ttl) 956 goto dropunlock; 957 } 958 959 /* 960 * A previous connection in TIMEWAIT state is supposed to catch stray 961 * or duplicate segments arriving late. If this segment was a 962 * legitimate new connection attempt, the old INPCB gets removed and 963 * we can try again to find a listening socket. 964 * 965 * At this point, due to earlier optimism, we may hold only an inpcb 966 * lock, and not the inpcbinfo write lock. If so, we need to try to 967 * acquire it, or if that fails, acquire a reference on the inpcb, 968 * drop all locks, acquire a global write lock, and then re-acquire 969 * the inpcb lock. We may at that point discover that another thread 970 * has tried to free the inpcb, in which case we need to loop back 971 * and try to find a new inpcb to deliver to. 972 * 973 * XXXRW: It may be time to rethink timewait locking. 974 */ 975 relocked: 976 if (inp->inp_flags & INP_TIMEWAIT) { 977 if (ti_locked == TI_UNLOCKED) { 978 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 979 in_pcbref(inp); 980 INP_WUNLOCK(inp); 981 INP_INFO_RLOCK(&V_tcbinfo); 982 ti_locked = TI_RLOCKED; 983 INP_WLOCK(inp); 984 if (in_pcbrele_wlocked(inp)) { 985 inp = NULL; 986 goto findpcb; 987 } else if (inp->inp_flags & INP_DROPPED) { 988 INP_WUNLOCK(inp); 989 inp = NULL; 990 goto findpcb; 991 } 992 } else 993 ti_locked = TI_RLOCKED; 994 } 995 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 996 997 if (thflags & TH_SYN) 998 tcp_dooptions(&to, optp, optlen, TO_SYN); 999 /* 1000 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 1001 */ 1002 if (tcp_twcheck(inp, &to, th, m, tlen)) 1003 goto findpcb; 1004 INP_INFO_RUNLOCK(&V_tcbinfo); 1005 return (IPPROTO_DONE); 1006 } 1007 /* 1008 * The TCPCB may no longer exist if the connection is winding 1009 * down or it is in the CLOSED state. Either way we drop the 1010 * segment and send an appropriate response. 1011 */ 1012 tp = intotcpcb(inp); 1013 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1014 rstreason = BANDLIM_RST_CLOSEDPORT; 1015 goto dropwithreset; 1016 } 1017 1018 #ifdef TCP_OFFLOAD 1019 if (tp->t_flags & TF_TOE) { 1020 tcp_offload_input(tp, m); 1021 m = NULL; /* consumed by the TOE driver */ 1022 goto dropunlock; 1023 } 1024 #endif 1025 1026 /* 1027 * We've identified a valid inpcb, but it could be that we need an 1028 * inpcbinfo write lock but don't hold it. In this case, attempt to 1029 * acquire using the same strategy as the TIMEWAIT case above. If we 1030 * relock, we have to jump back to 'relocked' as the connection might 1031 * now be in TIMEWAIT. 1032 */ 1033 #ifdef INVARIANTS 1034 if ((thflags & (TH_FIN | TH_RST)) != 0) 1035 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1036 #endif 1037 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1038 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1039 !IS_FASTOPEN(tp->t_flags)))) { 1040 if (ti_locked == TI_UNLOCKED) { 1041 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1042 in_pcbref(inp); 1043 INP_WUNLOCK(inp); 1044 INP_INFO_RLOCK(&V_tcbinfo); 1045 ti_locked = TI_RLOCKED; 1046 INP_WLOCK(inp); 1047 if (in_pcbrele_wlocked(inp)) { 1048 inp = NULL; 1049 goto findpcb; 1050 } else if (inp->inp_flags & INP_DROPPED) { 1051 INP_WUNLOCK(inp); 1052 inp = NULL; 1053 goto findpcb; 1054 } 1055 goto relocked; 1056 } else 1057 ti_locked = TI_RLOCKED; 1058 } 1059 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1060 } 1061 1062 #ifdef MAC 1063 INP_WLOCK_ASSERT(inp); 1064 if (mac_inpcb_check_deliver(inp, m)) 1065 goto dropunlock; 1066 #endif 1067 so = inp->inp_socket; 1068 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1069 #ifdef TCPDEBUG 1070 if (so->so_options & SO_DEBUG) { 1071 ostate = tp->t_state; 1072 #ifdef INET6 1073 if (isipv6) { 1074 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1075 } else 1076 #endif 1077 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1078 tcp_savetcp = *th; 1079 } 1080 #endif /* TCPDEBUG */ 1081 /* 1082 * When the socket is accepting connections (the INPCB is in LISTEN 1083 * state) we look into the SYN cache if this is a new connection 1084 * attempt or the completion of a previous one. 1085 */ 1086 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1087 ("%s: so accepting but tp %p not listening", __func__, tp)); 1088 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1089 struct in_conninfo inc; 1090 1091 bzero(&inc, sizeof(inc)); 1092 #ifdef INET6 1093 if (isipv6) { 1094 inc.inc_flags |= INC_ISIPV6; 1095 inc.inc6_faddr = ip6->ip6_src; 1096 inc.inc6_laddr = ip6->ip6_dst; 1097 } else 1098 #endif 1099 { 1100 inc.inc_faddr = ip->ip_src; 1101 inc.inc_laddr = ip->ip_dst; 1102 } 1103 inc.inc_fport = th->th_sport; 1104 inc.inc_lport = th->th_dport; 1105 inc.inc_fibnum = so->so_fibnum; 1106 1107 /* 1108 * Check for an existing connection attempt in syncache if 1109 * the flag is only ACK. A successful lookup creates a new 1110 * socket appended to the listen queue in SYN_RECEIVED state. 1111 */ 1112 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1113 1114 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1115 /* 1116 * Parse the TCP options here because 1117 * syncookies need access to the reflected 1118 * timestamp. 1119 */ 1120 tcp_dooptions(&to, optp, optlen, 0); 1121 /* 1122 * NB: syncache_expand() doesn't unlock 1123 * inp and tcpinfo locks. 1124 */ 1125 rstreason = syncache_expand(&inc, &to, th, &so, m); 1126 if (rstreason < 0) { 1127 /* 1128 * A failing TCP MD5 signature comparison 1129 * must result in the segment being dropped 1130 * and must not produce any response back 1131 * to the sender. 1132 */ 1133 goto dropunlock; 1134 } else if (rstreason == 0) { 1135 /* 1136 * No syncache entry or ACK was not 1137 * for our SYN/ACK. Send a RST. 1138 * NB: syncache did its own logging 1139 * of the failure cause. 1140 */ 1141 rstreason = BANDLIM_RST_OPENPORT; 1142 goto dropwithreset; 1143 } 1144 #ifdef TCP_RFC7413 1145 tfo_socket_result: 1146 #endif 1147 if (so == NULL) { 1148 /* 1149 * We completed the 3-way handshake 1150 * but could not allocate a socket 1151 * either due to memory shortage, 1152 * listen queue length limits or 1153 * global socket limits. Send RST 1154 * or wait and have the remote end 1155 * retransmit the ACK for another 1156 * try. 1157 */ 1158 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1159 log(LOG_DEBUG, "%s; %s: Listen socket: " 1160 "Socket allocation failed due to " 1161 "limits or memory shortage, %s\n", 1162 s, __func__, 1163 V_tcp_sc_rst_sock_fail ? 1164 "sending RST" : "try again"); 1165 if (V_tcp_sc_rst_sock_fail) { 1166 rstreason = BANDLIM_UNLIMITED; 1167 goto dropwithreset; 1168 } else 1169 goto dropunlock; 1170 } 1171 /* 1172 * Socket is created in state SYN_RECEIVED. 1173 * Unlock the listen socket, lock the newly 1174 * created socket and update the tp variable. 1175 */ 1176 INP_WUNLOCK(inp); /* listen socket */ 1177 inp = sotoinpcb(so); 1178 /* 1179 * New connection inpcb is already locked by 1180 * syncache_expand(). 1181 */ 1182 INP_WLOCK_ASSERT(inp); 1183 tp = intotcpcb(inp); 1184 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1185 ("%s: ", __func__)); 1186 /* 1187 * Process the segment and the data it 1188 * contains. tcp_do_segment() consumes 1189 * the mbuf chain and unlocks the inpcb. 1190 */ 1191 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1192 iptos, ti_locked); 1193 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1194 return (IPPROTO_DONE); 1195 } 1196 /* 1197 * Segment flag validation for new connection attempts: 1198 * 1199 * Our (SYN|ACK) response was rejected. 1200 * Check with syncache and remove entry to prevent 1201 * retransmits. 1202 * 1203 * NB: syncache_chkrst does its own logging of failure 1204 * causes. 1205 */ 1206 if (thflags & TH_RST) { 1207 syncache_chkrst(&inc, th); 1208 goto dropunlock; 1209 } 1210 /* 1211 * We can't do anything without SYN. 1212 */ 1213 if ((thflags & TH_SYN) == 0) { 1214 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1215 log(LOG_DEBUG, "%s; %s: Listen socket: " 1216 "SYN is missing, segment ignored\n", 1217 s, __func__); 1218 TCPSTAT_INC(tcps_badsyn); 1219 goto dropunlock; 1220 } 1221 /* 1222 * (SYN|ACK) is bogus on a listen socket. 1223 */ 1224 if (thflags & TH_ACK) { 1225 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1226 log(LOG_DEBUG, "%s; %s: Listen socket: " 1227 "SYN|ACK invalid, segment rejected\n", 1228 s, __func__); 1229 syncache_badack(&inc); /* XXX: Not needed! */ 1230 TCPSTAT_INC(tcps_badsyn); 1231 rstreason = BANDLIM_RST_OPENPORT; 1232 goto dropwithreset; 1233 } 1234 /* 1235 * If the drop_synfin option is enabled, drop all 1236 * segments with both the SYN and FIN bits set. 1237 * This prevents e.g. nmap from identifying the 1238 * TCP/IP stack. 1239 * XXX: Poor reasoning. nmap has other methods 1240 * and is constantly refining its stack detection 1241 * strategies. 1242 * XXX: This is a violation of the TCP specification 1243 * and was used by RFC1644. 1244 */ 1245 if ((thflags & TH_FIN) && V_drop_synfin) { 1246 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1247 log(LOG_DEBUG, "%s; %s: Listen socket: " 1248 "SYN|FIN segment ignored (based on " 1249 "sysctl setting)\n", s, __func__); 1250 TCPSTAT_INC(tcps_badsyn); 1251 goto dropunlock; 1252 } 1253 /* 1254 * Segment's flags are (SYN) or (SYN|FIN). 1255 * 1256 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1257 * as they do not affect the state of the TCP FSM. 1258 * The data pointed to by TH_URG and th_urp is ignored. 1259 */ 1260 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1261 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1262 KASSERT(thflags & (TH_SYN), 1263 ("%s: Listen socket: TH_SYN not set", __func__)); 1264 #ifdef INET6 1265 /* 1266 * If deprecated address is forbidden, 1267 * we do not accept SYN to deprecated interface 1268 * address to prevent any new inbound connection from 1269 * getting established. 1270 * When we do not accept SYN, we send a TCP RST, 1271 * with deprecated source address (instead of dropping 1272 * it). We compromise it as it is much better for peer 1273 * to send a RST, and RST will be the final packet 1274 * for the exchange. 1275 * 1276 * If we do not forbid deprecated addresses, we accept 1277 * the SYN packet. RFC2462 does not suggest dropping 1278 * SYN in this case. 1279 * If we decipher RFC2462 5.5.4, it says like this: 1280 * 1. use of deprecated addr with existing 1281 * communication is okay - "SHOULD continue to be 1282 * used" 1283 * 2. use of it with new communication: 1284 * (2a) "SHOULD NOT be used if alternate address 1285 * with sufficient scope is available" 1286 * (2b) nothing mentioned otherwise. 1287 * Here we fall into (2b) case as we have no choice in 1288 * our source address selection - we must obey the peer. 1289 * 1290 * The wording in RFC2462 is confusing, and there are 1291 * multiple description text for deprecated address 1292 * handling - worse, they are not exactly the same. 1293 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1294 */ 1295 if (isipv6 && !V_ip6_use_deprecated) { 1296 struct in6_ifaddr *ia6; 1297 1298 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1299 if (ia6 != NULL && 1300 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1301 ifa_free(&ia6->ia_ifa); 1302 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1303 log(LOG_DEBUG, "%s; %s: Listen socket: " 1304 "Connection attempt to deprecated " 1305 "IPv6 address rejected\n", 1306 s, __func__); 1307 rstreason = BANDLIM_RST_OPENPORT; 1308 goto dropwithreset; 1309 } 1310 if (ia6) 1311 ifa_free(&ia6->ia_ifa); 1312 } 1313 #endif /* INET6 */ 1314 /* 1315 * Basic sanity checks on incoming SYN requests: 1316 * Don't respond if the destination is a link layer 1317 * broadcast according to RFC1122 4.2.3.10, p. 104. 1318 * If it is from this socket it must be forged. 1319 * Don't respond if the source or destination is a 1320 * global or subnet broad- or multicast address. 1321 * Note that it is quite possible to receive unicast 1322 * link-layer packets with a broadcast IP address. Use 1323 * in_broadcast() to find them. 1324 */ 1325 if (m->m_flags & (M_BCAST|M_MCAST)) { 1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1327 log(LOG_DEBUG, "%s; %s: Listen socket: " 1328 "Connection attempt from broad- or multicast " 1329 "link layer address ignored\n", s, __func__); 1330 goto dropunlock; 1331 } 1332 #ifdef INET6 1333 if (isipv6) { 1334 if (th->th_dport == th->th_sport && 1335 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1336 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1337 log(LOG_DEBUG, "%s; %s: Listen socket: " 1338 "Connection attempt to/from self " 1339 "ignored\n", s, __func__); 1340 goto dropunlock; 1341 } 1342 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1343 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1344 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1345 log(LOG_DEBUG, "%s; %s: Listen socket: " 1346 "Connection attempt from/to multicast " 1347 "address ignored\n", s, __func__); 1348 goto dropunlock; 1349 } 1350 } 1351 #endif 1352 #if defined(INET) && defined(INET6) 1353 else 1354 #endif 1355 #ifdef INET 1356 { 1357 if (th->th_dport == th->th_sport && 1358 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1359 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1360 log(LOG_DEBUG, "%s; %s: Listen socket: " 1361 "Connection attempt from/to self " 1362 "ignored\n", s, __func__); 1363 goto dropunlock; 1364 } 1365 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1366 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1367 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1368 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1369 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1370 log(LOG_DEBUG, "%s; %s: Listen socket: " 1371 "Connection attempt from/to broad- " 1372 "or multicast address ignored\n", 1373 s, __func__); 1374 goto dropunlock; 1375 } 1376 } 1377 #endif 1378 /* 1379 * SYN appears to be valid. Create compressed TCP state 1380 * for syncache. 1381 */ 1382 #ifdef TCPDEBUG 1383 if (so->so_options & SO_DEBUG) 1384 tcp_trace(TA_INPUT, ostate, tp, 1385 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1386 #endif 1387 TCP_PROBE3(debug__input, tp, th, m); 1388 tcp_dooptions(&to, optp, optlen, TO_SYN); 1389 #ifdef TCP_RFC7413 1390 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1391 goto tfo_socket_result; 1392 #else 1393 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1394 #endif 1395 /* 1396 * Entry added to syncache and mbuf consumed. 1397 * Only the listen socket is unlocked by syncache_add(). 1398 */ 1399 if (ti_locked == TI_RLOCKED) { 1400 INP_INFO_RUNLOCK(&V_tcbinfo); 1401 ti_locked = TI_UNLOCKED; 1402 } 1403 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1404 return (IPPROTO_DONE); 1405 } else if (tp->t_state == TCPS_LISTEN) { 1406 /* 1407 * When a listen socket is torn down the SO_ACCEPTCONN 1408 * flag is removed first while connections are drained 1409 * from the accept queue in a unlock/lock cycle of the 1410 * ACCEPT_LOCK, opening a race condition allowing a SYN 1411 * attempt go through unhandled. 1412 */ 1413 goto dropunlock; 1414 } 1415 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1416 if (tp->t_flags & TF_SIGNATURE) { 1417 tcp_dooptions(&to, optp, optlen, thflags); 1418 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1419 TCPSTAT_INC(tcps_sig_err_nosigopt); 1420 goto dropunlock; 1421 } 1422 if (!TCPMD5_ENABLED() || 1423 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1424 goto dropunlock; 1425 } 1426 #endif 1427 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1428 1429 /* 1430 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1431 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1432 * the inpcb, and unlocks pcbinfo. 1433 */ 1434 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1435 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1436 return (IPPROTO_DONE); 1437 1438 dropwithreset: 1439 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1440 1441 if (ti_locked == TI_RLOCKED) { 1442 INP_INFO_RUNLOCK(&V_tcbinfo); 1443 ti_locked = TI_UNLOCKED; 1444 } 1445 #ifdef INVARIANTS 1446 else { 1447 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1448 "ti_locked: %d", __func__, ti_locked)); 1449 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1450 } 1451 #endif 1452 1453 if (inp != NULL) { 1454 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1455 INP_WUNLOCK(inp); 1456 } else 1457 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1458 m = NULL; /* mbuf chain got consumed. */ 1459 goto drop; 1460 1461 dropunlock: 1462 if (m != NULL) 1463 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1464 1465 if (ti_locked == TI_RLOCKED) { 1466 INP_INFO_RUNLOCK(&V_tcbinfo); 1467 ti_locked = TI_UNLOCKED; 1468 } 1469 #ifdef INVARIANTS 1470 else { 1471 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1472 "ti_locked: %d", __func__, ti_locked)); 1473 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1474 } 1475 #endif 1476 1477 if (inp != NULL) 1478 INP_WUNLOCK(inp); 1479 1480 drop: 1481 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1482 if (s != NULL) 1483 free(s, M_TCPLOG); 1484 if (m != NULL) 1485 m_freem(m); 1486 return (IPPROTO_DONE); 1487 } 1488 1489 /* 1490 * Automatic sizing of receive socket buffer. Often the send 1491 * buffer size is not optimally adjusted to the actual network 1492 * conditions at hand (delay bandwidth product). Setting the 1493 * buffer size too small limits throughput on links with high 1494 * bandwidth and high delay (eg. trans-continental/oceanic links). 1495 * 1496 * On the receive side the socket buffer memory is only rarely 1497 * used to any significant extent. This allows us to be much 1498 * more aggressive in scaling the receive socket buffer. For 1499 * the case that the buffer space is actually used to a large 1500 * extent and we run out of kernel memory we can simply drop 1501 * the new segments; TCP on the sender will just retransmit it 1502 * later. Setting the buffer size too big may only consume too 1503 * much kernel memory if the application doesn't read() from 1504 * the socket or packet loss or reordering makes use of the 1505 * reassembly queue. 1506 * 1507 * The criteria to step up the receive buffer one notch are: 1508 * 1. Application has not set receive buffer size with 1509 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1510 * 2. the number of bytes received during the time it takes 1511 * one timestamp to be reflected back to us (the RTT); 1512 * 3. received bytes per RTT is within seven eighth of the 1513 * current socket buffer size; 1514 * 4. receive buffer size has not hit maximal automatic size; 1515 * 1516 * This algorithm does one step per RTT at most and only if 1517 * we receive a bulk stream w/o packet losses or reorderings. 1518 * Shrinking the buffer during idle times is not necessary as 1519 * it doesn't consume any memory when idle. 1520 * 1521 * TODO: Only step up if the application is actually serving 1522 * the buffer to better manage the socket buffer resources. 1523 */ 1524 int 1525 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1526 struct tcpcb *tp, int tlen) 1527 { 1528 int newsize = 0; 1529 1530 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1531 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1532 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1533 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1534 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1535 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1536 newsize = min(so->so_rcv.sb_hiwat + 1537 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1538 } 1539 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1540 1541 /* Start over with next RTT. */ 1542 tp->rfbuf_ts = 0; 1543 tp->rfbuf_cnt = 0; 1544 } else { 1545 tp->rfbuf_cnt += tlen; /* add up */ 1546 } 1547 1548 return (newsize); 1549 } 1550 1551 void 1552 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1553 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1554 int ti_locked) 1555 { 1556 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1557 int rstreason, todrop, win; 1558 uint32_t tiwin; 1559 uint16_t nsegs; 1560 char *s; 1561 struct in_conninfo *inc; 1562 struct mbuf *mfree; 1563 struct tcpopt to; 1564 #ifdef TCP_RFC7413 1565 int tfo_syn; 1566 #endif 1567 1568 #ifdef TCPDEBUG 1569 /* 1570 * The size of tcp_saveipgen must be the size of the max ip header, 1571 * now IPv6. 1572 */ 1573 u_char tcp_saveipgen[IP6_HDR_LEN]; 1574 struct tcphdr tcp_savetcp; 1575 short ostate = 0; 1576 #endif 1577 thflags = th->th_flags; 1578 inc = &tp->t_inpcb->inp_inc; 1579 tp->sackhint.last_sack_ack = 0; 1580 sack_changed = 0; 1581 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1582 1583 /* 1584 * If this is either a state-changing packet or current state isn't 1585 * established, we require a write lock on tcbinfo. Otherwise, we 1586 * allow the tcbinfo to be in either alocked or unlocked, as the 1587 * caller may have unnecessarily acquired a write lock due to a race. 1588 */ 1589 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1590 tp->t_state != TCPS_ESTABLISHED) { 1591 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1592 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1593 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1594 } else { 1595 #ifdef INVARIANTS 1596 if (ti_locked == TI_RLOCKED) 1597 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1598 else { 1599 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1600 "ti_locked: %d", __func__, ti_locked)); 1601 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1602 } 1603 #endif 1604 } 1605 INP_WLOCK_ASSERT(tp->t_inpcb); 1606 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1607 __func__)); 1608 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1609 __func__)); 1610 1611 #ifdef TCPPCAP 1612 /* Save segment, if requested. */ 1613 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1614 #endif 1615 1616 /* 1617 * Segment received on connection. 1618 * Reset idle time and keep-alive timer. 1619 * XXX: This should be done after segment 1620 * validation to ignore broken/spoofed segs. 1621 */ 1622 tp->t_rcvtime = ticks; 1623 1624 /* 1625 * Scale up the window into a 32-bit value. 1626 * For the SYN_SENT state the scale is zero. 1627 */ 1628 tiwin = th->th_win << tp->snd_scale; 1629 1630 /* 1631 * TCP ECN processing. 1632 */ 1633 if (tp->t_flags & TF_ECN_PERMIT) { 1634 if (thflags & TH_CWR) 1635 tp->t_flags &= ~TF_ECN_SND_ECE; 1636 switch (iptos & IPTOS_ECN_MASK) { 1637 case IPTOS_ECN_CE: 1638 tp->t_flags |= TF_ECN_SND_ECE; 1639 TCPSTAT_INC(tcps_ecn_ce); 1640 break; 1641 case IPTOS_ECN_ECT0: 1642 TCPSTAT_INC(tcps_ecn_ect0); 1643 break; 1644 case IPTOS_ECN_ECT1: 1645 TCPSTAT_INC(tcps_ecn_ect1); 1646 break; 1647 } 1648 1649 /* Process a packet differently from RFC3168. */ 1650 cc_ecnpkt_handler(tp, th, iptos); 1651 1652 /* Congestion experienced. */ 1653 if (thflags & TH_ECE) { 1654 cc_cong_signal(tp, th, CC_ECN); 1655 } 1656 } 1657 1658 /* 1659 * Parse options on any incoming segment. 1660 */ 1661 tcp_dooptions(&to, (u_char *)(th + 1), 1662 (th->th_off << 2) - sizeof(struct tcphdr), 1663 (thflags & TH_SYN) ? TO_SYN : 0); 1664 1665 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1666 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1667 (to.to_flags & TOF_SIGNATURE) == 0) { 1668 TCPSTAT_INC(tcps_sig_err_sigopt); 1669 /* XXX: should drop? */ 1670 } 1671 #endif 1672 /* 1673 * If echoed timestamp is later than the current time, 1674 * fall back to non RFC1323 RTT calculation. Normalize 1675 * timestamp if syncookies were used when this connection 1676 * was established. 1677 */ 1678 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1679 to.to_tsecr -= tp->ts_offset; 1680 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1681 to.to_tsecr = 0; 1682 } 1683 /* 1684 * If timestamps were negotiated during SYN/ACK they should 1685 * appear on every segment during this session and vice versa. 1686 */ 1687 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1688 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1689 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1690 "no action\n", s, __func__); 1691 free(s, M_TCPLOG); 1692 } 1693 } 1694 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1695 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1696 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1697 "no action\n", s, __func__); 1698 free(s, M_TCPLOG); 1699 } 1700 } 1701 1702 /* 1703 * Process options only when we get SYN/ACK back. The SYN case 1704 * for incoming connections is handled in tcp_syncache. 1705 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1706 * or <SYN,ACK>) segment itself is never scaled. 1707 * XXX this is traditional behavior, may need to be cleaned up. 1708 */ 1709 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1710 if ((to.to_flags & TOF_SCALE) && 1711 (tp->t_flags & TF_REQ_SCALE)) { 1712 tp->t_flags |= TF_RCVD_SCALE; 1713 tp->snd_scale = to.to_wscale; 1714 } 1715 /* 1716 * Initial send window. It will be updated with 1717 * the next incoming segment to the scaled value. 1718 */ 1719 tp->snd_wnd = th->th_win; 1720 if (to.to_flags & TOF_TS) { 1721 tp->t_flags |= TF_RCVD_TSTMP; 1722 tp->ts_recent = to.to_tsval; 1723 tp->ts_recent_age = tcp_ts_getticks(); 1724 } 1725 if (to.to_flags & TOF_MSS) 1726 tcp_mss(tp, to.to_mss); 1727 if ((tp->t_flags & TF_SACK_PERMIT) && 1728 (to.to_flags & TOF_SACKPERM) == 0) 1729 tp->t_flags &= ~TF_SACK_PERMIT; 1730 } 1731 1732 /* 1733 * Header prediction: check for the two common cases 1734 * of a uni-directional data xfer. If the packet has 1735 * no control flags, is in-sequence, the window didn't 1736 * change and we're not retransmitting, it's a 1737 * candidate. If the length is zero and the ack moved 1738 * forward, we're the sender side of the xfer. Just 1739 * free the data acked & wake any higher level process 1740 * that was blocked waiting for space. If the length 1741 * is non-zero and the ack didn't move, we're the 1742 * receiver side. If we're getting packets in-order 1743 * (the reassembly queue is empty), add the data to 1744 * the socket buffer and note that we need a delayed ack. 1745 * Make sure that the hidden state-flags are also off. 1746 * Since we check for TCPS_ESTABLISHED first, it can only 1747 * be TH_NEEDSYN. 1748 */ 1749 if (tp->t_state == TCPS_ESTABLISHED && 1750 th->th_seq == tp->rcv_nxt && 1751 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1752 tp->snd_nxt == tp->snd_max && 1753 tiwin && tiwin == tp->snd_wnd && 1754 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1755 LIST_EMPTY(&tp->t_segq) && 1756 ((to.to_flags & TOF_TS) == 0 || 1757 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1758 1759 /* 1760 * If last ACK falls within this segment's sequence numbers, 1761 * record the timestamp. 1762 * NOTE that the test is modified according to the latest 1763 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1764 */ 1765 if ((to.to_flags & TOF_TS) != 0 && 1766 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1767 tp->ts_recent_age = tcp_ts_getticks(); 1768 tp->ts_recent = to.to_tsval; 1769 } 1770 1771 if (tlen == 0) { 1772 if (SEQ_GT(th->th_ack, tp->snd_una) && 1773 SEQ_LEQ(th->th_ack, tp->snd_max) && 1774 !IN_RECOVERY(tp->t_flags) && 1775 (to.to_flags & TOF_SACK) == 0 && 1776 TAILQ_EMPTY(&tp->snd_holes)) { 1777 /* 1778 * This is a pure ack for outstanding data. 1779 */ 1780 if (ti_locked == TI_RLOCKED) 1781 INP_INFO_RUNLOCK(&V_tcbinfo); 1782 ti_locked = TI_UNLOCKED; 1783 1784 TCPSTAT_INC(tcps_predack); 1785 1786 /* 1787 * "bad retransmit" recovery. 1788 */ 1789 if (tp->t_rxtshift == 1 && 1790 tp->t_flags & TF_PREVVALID && 1791 (int)(ticks - tp->t_badrxtwin) < 0) { 1792 cc_cong_signal(tp, th, CC_RTO_ERR); 1793 } 1794 1795 /* 1796 * Recalculate the transmit timer / rtt. 1797 * 1798 * Some boxes send broken timestamp replies 1799 * during the SYN+ACK phase, ignore 1800 * timestamps of 0 or we could calculate a 1801 * huge RTT and blow up the retransmit timer. 1802 */ 1803 if ((to.to_flags & TOF_TS) != 0 && 1804 to.to_tsecr) { 1805 uint32_t t; 1806 1807 t = tcp_ts_getticks() - to.to_tsecr; 1808 if (!tp->t_rttlow || tp->t_rttlow > t) 1809 tp->t_rttlow = t; 1810 tcp_xmit_timer(tp, 1811 TCP_TS_TO_TICKS(t) + 1); 1812 } else if (tp->t_rtttime && 1813 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1814 if (!tp->t_rttlow || 1815 tp->t_rttlow > ticks - tp->t_rtttime) 1816 tp->t_rttlow = ticks - tp->t_rtttime; 1817 tcp_xmit_timer(tp, 1818 ticks - tp->t_rtttime); 1819 } 1820 acked = BYTES_THIS_ACK(tp, th); 1821 1822 #ifdef TCP_HHOOK 1823 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1824 hhook_run_tcp_est_in(tp, th, &to); 1825 #endif 1826 1827 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1828 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1829 sbdrop(&so->so_snd, acked); 1830 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1831 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1832 tp->snd_recover = th->th_ack - 1; 1833 1834 /* 1835 * Let the congestion control algorithm update 1836 * congestion control related information. This 1837 * typically means increasing the congestion 1838 * window. 1839 */ 1840 cc_ack_received(tp, th, nsegs, CC_ACK); 1841 1842 tp->snd_una = th->th_ack; 1843 /* 1844 * Pull snd_wl2 up to prevent seq wrap relative 1845 * to th_ack. 1846 */ 1847 tp->snd_wl2 = th->th_ack; 1848 tp->t_dupacks = 0; 1849 m_freem(m); 1850 1851 /* 1852 * If all outstanding data are acked, stop 1853 * retransmit timer, otherwise restart timer 1854 * using current (possibly backed-off) value. 1855 * If process is waiting for space, 1856 * wakeup/selwakeup/signal. If data 1857 * are ready to send, let tcp_output 1858 * decide between more output or persist. 1859 */ 1860 #ifdef TCPDEBUG 1861 if (so->so_options & SO_DEBUG) 1862 tcp_trace(TA_INPUT, ostate, tp, 1863 (void *)tcp_saveipgen, 1864 &tcp_savetcp, 0); 1865 #endif 1866 TCP_PROBE3(debug__input, tp, th, m); 1867 if (tp->snd_una == tp->snd_max) 1868 tcp_timer_activate(tp, TT_REXMT, 0); 1869 else if (!tcp_timer_active(tp, TT_PERSIST)) 1870 tcp_timer_activate(tp, TT_REXMT, 1871 tp->t_rxtcur); 1872 sowwakeup(so); 1873 if (sbavail(&so->so_snd)) 1874 (void) tp->t_fb->tfb_tcp_output(tp); 1875 goto check_delack; 1876 } 1877 } else if (th->th_ack == tp->snd_una && 1878 tlen <= sbspace(&so->so_rcv)) { 1879 int newsize = 0; /* automatic sockbuf scaling */ 1880 1881 /* 1882 * This is a pure, in-sequence data packet with 1883 * nothing on the reassembly queue and we have enough 1884 * buffer space to take it. 1885 */ 1886 if (ti_locked == TI_RLOCKED) 1887 INP_INFO_RUNLOCK(&V_tcbinfo); 1888 ti_locked = TI_UNLOCKED; 1889 1890 /* Clean receiver SACK report if present */ 1891 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1892 tcp_clean_sackreport(tp); 1893 TCPSTAT_INC(tcps_preddat); 1894 tp->rcv_nxt += tlen; 1895 /* 1896 * Pull snd_wl1 up to prevent seq wrap relative to 1897 * th_seq. 1898 */ 1899 tp->snd_wl1 = th->th_seq; 1900 /* 1901 * Pull rcv_up up to prevent seq wrap relative to 1902 * rcv_nxt. 1903 */ 1904 tp->rcv_up = tp->rcv_nxt; 1905 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1906 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1907 #ifdef TCPDEBUG 1908 if (so->so_options & SO_DEBUG) 1909 tcp_trace(TA_INPUT, ostate, tp, 1910 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1911 #endif 1912 TCP_PROBE3(debug__input, tp, th, m); 1913 1914 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1915 1916 /* Add data to socket buffer. */ 1917 SOCKBUF_LOCK(&so->so_rcv); 1918 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1919 m_freem(m); 1920 } else { 1921 /* 1922 * Set new socket buffer size. 1923 * Give up when limit is reached. 1924 */ 1925 if (newsize) 1926 if (!sbreserve_locked(&so->so_rcv, 1927 newsize, so, NULL)) 1928 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1929 m_adj(m, drop_hdrlen); /* delayed header drop */ 1930 sbappendstream_locked(&so->so_rcv, m, 0); 1931 } 1932 /* NB: sorwakeup_locked() does an implicit unlock. */ 1933 sorwakeup_locked(so); 1934 if (DELAY_ACK(tp, tlen)) { 1935 tp->t_flags |= TF_DELACK; 1936 } else { 1937 tp->t_flags |= TF_ACKNOW; 1938 tp->t_fb->tfb_tcp_output(tp); 1939 } 1940 goto check_delack; 1941 } 1942 } 1943 1944 /* 1945 * Calculate amount of space in receive window, 1946 * and then do TCP input processing. 1947 * Receive window is amount of space in rcv queue, 1948 * but not less than advertised window. 1949 */ 1950 win = sbspace(&so->so_rcv); 1951 if (win < 0) 1952 win = 0; 1953 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1954 1955 switch (tp->t_state) { 1956 1957 /* 1958 * If the state is SYN_RECEIVED: 1959 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1960 */ 1961 case TCPS_SYN_RECEIVED: 1962 if ((thflags & TH_ACK) && 1963 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1964 SEQ_GT(th->th_ack, tp->snd_max))) { 1965 rstreason = BANDLIM_RST_OPENPORT; 1966 goto dropwithreset; 1967 } 1968 #ifdef TCP_RFC7413 1969 if (IS_FASTOPEN(tp->t_flags)) { 1970 /* 1971 * When a TFO connection is in SYN_RECEIVED, the 1972 * only valid packets are the initial SYN, a 1973 * retransmit/copy of the initial SYN (possibly with 1974 * a subset of the original data), a valid ACK, a 1975 * FIN, or a RST. 1976 */ 1977 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1978 rstreason = BANDLIM_RST_OPENPORT; 1979 goto dropwithreset; 1980 } else if (thflags & TH_SYN) { 1981 /* non-initial SYN is ignored */ 1982 if ((tcp_timer_active(tp, TT_DELACK) || 1983 tcp_timer_active(tp, TT_REXMT))) 1984 goto drop; 1985 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1986 goto drop; 1987 } 1988 } 1989 #endif 1990 break; 1991 1992 /* 1993 * If the state is SYN_SENT: 1994 * if seg contains an ACK, but not for our SYN, drop the input. 1995 * if seg contains a RST, then drop the connection. 1996 * if seg does not contain SYN, then drop it. 1997 * Otherwise this is an acceptable SYN segment 1998 * initialize tp->rcv_nxt and tp->irs 1999 * if seg contains ack then advance tp->snd_una 2000 * if seg contains an ECE and ECN support is enabled, the stream 2001 * is ECN capable. 2002 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2003 * arrange for segment to be acked (eventually) 2004 * continue processing rest of data/controls, beginning with URG 2005 */ 2006 case TCPS_SYN_SENT: 2007 if ((thflags & TH_ACK) && 2008 (SEQ_LEQ(th->th_ack, tp->iss) || 2009 SEQ_GT(th->th_ack, tp->snd_max))) { 2010 rstreason = BANDLIM_UNLIMITED; 2011 goto dropwithreset; 2012 } 2013 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2014 TCP_PROBE5(connect__refused, NULL, tp, 2015 m, tp, th); 2016 tp = tcp_drop(tp, ECONNREFUSED); 2017 } 2018 if (thflags & TH_RST) 2019 goto drop; 2020 if (!(thflags & TH_SYN)) 2021 goto drop; 2022 2023 tp->irs = th->th_seq; 2024 tcp_rcvseqinit(tp); 2025 if (thflags & TH_ACK) { 2026 TCPSTAT_INC(tcps_connects); 2027 soisconnected(so); 2028 #ifdef MAC 2029 mac_socketpeer_set_from_mbuf(m, so); 2030 #endif 2031 /* Do window scaling on this connection? */ 2032 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2033 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2034 tp->rcv_scale = tp->request_r_scale; 2035 } 2036 tp->rcv_adv += min(tp->rcv_wnd, 2037 TCP_MAXWIN << tp->rcv_scale); 2038 tp->snd_una++; /* SYN is acked */ 2039 /* 2040 * If there's data, delay ACK; if there's also a FIN 2041 * ACKNOW will be turned on later. 2042 */ 2043 if (DELAY_ACK(tp, tlen) && tlen != 0) 2044 tcp_timer_activate(tp, TT_DELACK, 2045 tcp_delacktime); 2046 else 2047 tp->t_flags |= TF_ACKNOW; 2048 2049 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2050 tp->t_flags |= TF_ECN_PERMIT; 2051 TCPSTAT_INC(tcps_ecn_shs); 2052 } 2053 2054 /* 2055 * Received <SYN,ACK> in SYN_SENT[*] state. 2056 * Transitions: 2057 * SYN_SENT --> ESTABLISHED 2058 * SYN_SENT* --> FIN_WAIT_1 2059 */ 2060 tp->t_starttime = ticks; 2061 if (tp->t_flags & TF_NEEDFIN) { 2062 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2063 tp->t_flags &= ~TF_NEEDFIN; 2064 thflags &= ~TH_SYN; 2065 } else { 2066 tcp_state_change(tp, TCPS_ESTABLISHED); 2067 TCP_PROBE5(connect__established, NULL, tp, 2068 m, tp, th); 2069 cc_conn_init(tp); 2070 tcp_timer_activate(tp, TT_KEEP, 2071 TP_KEEPIDLE(tp)); 2072 } 2073 } else { 2074 /* 2075 * Received initial SYN in SYN-SENT[*] state => 2076 * simultaneous open. 2077 * If it succeeds, connection is * half-synchronized. 2078 * Otherwise, do 3-way handshake: 2079 * SYN-SENT -> SYN-RECEIVED 2080 * SYN-SENT* -> SYN-RECEIVED* 2081 */ 2082 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2083 tcp_timer_activate(tp, TT_REXMT, 0); 2084 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2085 } 2086 2087 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2088 "ti_locked %d", __func__, ti_locked)); 2089 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2090 INP_WLOCK_ASSERT(tp->t_inpcb); 2091 2092 /* 2093 * Advance th->th_seq to correspond to first data byte. 2094 * If data, trim to stay within window, 2095 * dropping FIN if necessary. 2096 */ 2097 th->th_seq++; 2098 if (tlen > tp->rcv_wnd) { 2099 todrop = tlen - tp->rcv_wnd; 2100 m_adj(m, -todrop); 2101 tlen = tp->rcv_wnd; 2102 thflags &= ~TH_FIN; 2103 TCPSTAT_INC(tcps_rcvpackafterwin); 2104 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2105 } 2106 tp->snd_wl1 = th->th_seq - 1; 2107 tp->rcv_up = th->th_seq; 2108 /* 2109 * Client side of transaction: already sent SYN and data. 2110 * If the remote host used T/TCP to validate the SYN, 2111 * our data will be ACK'd; if so, enter normal data segment 2112 * processing in the middle of step 5, ack processing. 2113 * Otherwise, goto step 6. 2114 */ 2115 if (thflags & TH_ACK) 2116 goto process_ACK; 2117 2118 goto step6; 2119 2120 /* 2121 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2122 * do normal processing. 2123 * 2124 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2125 */ 2126 case TCPS_LAST_ACK: 2127 case TCPS_CLOSING: 2128 break; /* continue normal processing */ 2129 } 2130 2131 /* 2132 * States other than LISTEN or SYN_SENT. 2133 * First check the RST flag and sequence number since reset segments 2134 * are exempt from the timestamp and connection count tests. This 2135 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2136 * below which allowed reset segments in half the sequence space 2137 * to fall though and be processed (which gives forged reset 2138 * segments with a random sequence number a 50 percent chance of 2139 * killing a connection). 2140 * Then check timestamp, if present. 2141 * Then check the connection count, if present. 2142 * Then check that at least some bytes of segment are within 2143 * receive window. If segment begins before rcv_nxt, 2144 * drop leading data (and SYN); if nothing left, just ack. 2145 */ 2146 if (thflags & TH_RST) { 2147 /* 2148 * RFC5961 Section 3.2 2149 * 2150 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2151 * - If RST is in window, we send challenge ACK. 2152 * 2153 * Note: to take into account delayed ACKs, we should 2154 * test against last_ack_sent instead of rcv_nxt. 2155 * Note 2: we handle special case of closed window, not 2156 * covered by the RFC. 2157 */ 2158 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2159 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2160 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2161 2162 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2163 KASSERT(ti_locked == TI_RLOCKED, 2164 ("%s: TH_RST ti_locked %d, th %p tp %p", 2165 __func__, ti_locked, th, tp)); 2166 KASSERT(tp->t_state != TCPS_SYN_SENT, 2167 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2168 __func__, th, tp)); 2169 2170 if (V_tcp_insecure_rst || 2171 tp->last_ack_sent == th->th_seq) { 2172 TCPSTAT_INC(tcps_drops); 2173 /* Drop the connection. */ 2174 switch (tp->t_state) { 2175 case TCPS_SYN_RECEIVED: 2176 so->so_error = ECONNREFUSED; 2177 goto close; 2178 case TCPS_ESTABLISHED: 2179 case TCPS_FIN_WAIT_1: 2180 case TCPS_FIN_WAIT_2: 2181 case TCPS_CLOSE_WAIT: 2182 case TCPS_CLOSING: 2183 case TCPS_LAST_ACK: 2184 so->so_error = ECONNRESET; 2185 close: 2186 /* FALLTHROUGH */ 2187 default: 2188 tp = tcp_close(tp); 2189 } 2190 } else { 2191 TCPSTAT_INC(tcps_badrst); 2192 /* Send challenge ACK. */ 2193 tcp_respond(tp, mtod(m, void *), th, m, 2194 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2195 tp->last_ack_sent = tp->rcv_nxt; 2196 m = NULL; 2197 } 2198 } 2199 goto drop; 2200 } 2201 2202 /* 2203 * RFC5961 Section 4.2 2204 * Send challenge ACK for any SYN in synchronized state. 2205 */ 2206 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2207 tp->t_state != TCPS_SYN_RECEIVED) { 2208 KASSERT(ti_locked == TI_RLOCKED, 2209 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2210 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2211 2212 TCPSTAT_INC(tcps_badsyn); 2213 if (V_tcp_insecure_syn && 2214 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2215 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2216 tp = tcp_drop(tp, ECONNRESET); 2217 rstreason = BANDLIM_UNLIMITED; 2218 } else { 2219 /* Send challenge ACK. */ 2220 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2221 tp->snd_nxt, TH_ACK); 2222 tp->last_ack_sent = tp->rcv_nxt; 2223 m = NULL; 2224 } 2225 goto drop; 2226 } 2227 2228 /* 2229 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2230 * and it's less than ts_recent, drop it. 2231 */ 2232 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2233 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2234 2235 /* Check to see if ts_recent is over 24 days old. */ 2236 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2237 /* 2238 * Invalidate ts_recent. If this segment updates 2239 * ts_recent, the age will be reset later and ts_recent 2240 * will get a valid value. If it does not, setting 2241 * ts_recent to zero will at least satisfy the 2242 * requirement that zero be placed in the timestamp 2243 * echo reply when ts_recent isn't valid. The 2244 * age isn't reset until we get a valid ts_recent 2245 * because we don't want out-of-order segments to be 2246 * dropped when ts_recent is old. 2247 */ 2248 tp->ts_recent = 0; 2249 } else { 2250 TCPSTAT_INC(tcps_rcvduppack); 2251 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2252 TCPSTAT_INC(tcps_pawsdrop); 2253 if (tlen) 2254 goto dropafterack; 2255 goto drop; 2256 } 2257 } 2258 2259 /* 2260 * In the SYN-RECEIVED state, validate that the packet belongs to 2261 * this connection before trimming the data to fit the receive 2262 * window. Check the sequence number versus IRS since we know 2263 * the sequence numbers haven't wrapped. This is a partial fix 2264 * for the "LAND" DoS attack. 2265 */ 2266 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2267 rstreason = BANDLIM_RST_OPENPORT; 2268 goto dropwithreset; 2269 } 2270 2271 todrop = tp->rcv_nxt - th->th_seq; 2272 if (todrop > 0) { 2273 if (thflags & TH_SYN) { 2274 thflags &= ~TH_SYN; 2275 th->th_seq++; 2276 if (th->th_urp > 1) 2277 th->th_urp--; 2278 else 2279 thflags &= ~TH_URG; 2280 todrop--; 2281 } 2282 /* 2283 * Following if statement from Stevens, vol. 2, p. 960. 2284 */ 2285 if (todrop > tlen 2286 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2287 /* 2288 * Any valid FIN must be to the left of the window. 2289 * At this point the FIN must be a duplicate or out 2290 * of sequence; drop it. 2291 */ 2292 thflags &= ~TH_FIN; 2293 2294 /* 2295 * Send an ACK to resynchronize and drop any data. 2296 * But keep on processing for RST or ACK. 2297 */ 2298 tp->t_flags |= TF_ACKNOW; 2299 todrop = tlen; 2300 TCPSTAT_INC(tcps_rcvduppack); 2301 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2302 } else { 2303 TCPSTAT_INC(tcps_rcvpartduppack); 2304 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2305 } 2306 drop_hdrlen += todrop; /* drop from the top afterwards */ 2307 th->th_seq += todrop; 2308 tlen -= todrop; 2309 if (th->th_urp > todrop) 2310 th->th_urp -= todrop; 2311 else { 2312 thflags &= ~TH_URG; 2313 th->th_urp = 0; 2314 } 2315 } 2316 2317 /* 2318 * If new data are received on a connection after the 2319 * user processes are gone, then RST the other end. 2320 */ 2321 if ((so->so_state & SS_NOFDREF) && 2322 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2323 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2324 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2325 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2326 2327 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2328 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2329 "after socket was closed, " 2330 "sending RST and removing tcpcb\n", 2331 s, __func__, tcpstates[tp->t_state], tlen); 2332 free(s, M_TCPLOG); 2333 } 2334 tp = tcp_close(tp); 2335 TCPSTAT_INC(tcps_rcvafterclose); 2336 rstreason = BANDLIM_UNLIMITED; 2337 goto dropwithreset; 2338 } 2339 2340 /* 2341 * If segment ends after window, drop trailing data 2342 * (and PUSH and FIN); if nothing left, just ACK. 2343 */ 2344 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2345 if (todrop > 0) { 2346 TCPSTAT_INC(tcps_rcvpackafterwin); 2347 if (todrop >= tlen) { 2348 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2349 /* 2350 * If window is closed can only take segments at 2351 * window edge, and have to drop data and PUSH from 2352 * incoming segments. Continue processing, but 2353 * remember to ack. Otherwise, drop segment 2354 * and ack. 2355 */ 2356 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2357 tp->t_flags |= TF_ACKNOW; 2358 TCPSTAT_INC(tcps_rcvwinprobe); 2359 } else 2360 goto dropafterack; 2361 } else 2362 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2363 m_adj(m, -todrop); 2364 tlen -= todrop; 2365 thflags &= ~(TH_PUSH|TH_FIN); 2366 } 2367 2368 /* 2369 * If last ACK falls within this segment's sequence numbers, 2370 * record its timestamp. 2371 * NOTE: 2372 * 1) That the test incorporates suggestions from the latest 2373 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2374 * 2) That updating only on newer timestamps interferes with 2375 * our earlier PAWS tests, so this check should be solely 2376 * predicated on the sequence space of this segment. 2377 * 3) That we modify the segment boundary check to be 2378 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2379 * instead of RFC1323's 2380 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2381 * This modified check allows us to overcome RFC1323's 2382 * limitations as described in Stevens TCP/IP Illustrated 2383 * Vol. 2 p.869. In such cases, we can still calculate the 2384 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2385 */ 2386 if ((to.to_flags & TOF_TS) != 0 && 2387 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2388 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2389 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2390 tp->ts_recent_age = tcp_ts_getticks(); 2391 tp->ts_recent = to.to_tsval; 2392 } 2393 2394 /* 2395 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2396 * flag is on (half-synchronized state), then queue data for 2397 * later processing; else drop segment and return. 2398 */ 2399 if ((thflags & TH_ACK) == 0) { 2400 if (tp->t_state == TCPS_SYN_RECEIVED || 2401 (tp->t_flags & TF_NEEDSYN)) { 2402 #ifdef TCP_RFC7413 2403 if (tp->t_state == TCPS_SYN_RECEIVED && 2404 IS_FASTOPEN(tp->t_flags)) { 2405 tp->snd_wnd = tiwin; 2406 cc_conn_init(tp); 2407 } 2408 #endif 2409 goto step6; 2410 } else if (tp->t_flags & TF_ACKNOW) 2411 goto dropafterack; 2412 else 2413 goto drop; 2414 } 2415 2416 /* 2417 * Ack processing. 2418 */ 2419 switch (tp->t_state) { 2420 2421 /* 2422 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2423 * ESTABLISHED state and continue processing. 2424 * The ACK was checked above. 2425 */ 2426 case TCPS_SYN_RECEIVED: 2427 2428 TCPSTAT_INC(tcps_connects); 2429 soisconnected(so); 2430 /* Do window scaling? */ 2431 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2432 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2433 tp->rcv_scale = tp->request_r_scale; 2434 tp->snd_wnd = tiwin; 2435 } 2436 /* 2437 * Make transitions: 2438 * SYN-RECEIVED -> ESTABLISHED 2439 * SYN-RECEIVED* -> FIN-WAIT-1 2440 */ 2441 tp->t_starttime = ticks; 2442 if (tp->t_flags & TF_NEEDFIN) { 2443 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2444 tp->t_flags &= ~TF_NEEDFIN; 2445 } else { 2446 tcp_state_change(tp, TCPS_ESTABLISHED); 2447 TCP_PROBE5(accept__established, NULL, tp, 2448 m, tp, th); 2449 #ifdef TCP_RFC7413 2450 if (tp->t_tfo_pending) { 2451 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2452 tp->t_tfo_pending = NULL; 2453 2454 /* 2455 * Account for the ACK of our SYN prior to 2456 * regular ACK processing below. 2457 */ 2458 tp->snd_una++; 2459 } 2460 /* 2461 * TFO connections call cc_conn_init() during SYN 2462 * processing. Calling it again here for such 2463 * connections is not harmless as it would undo the 2464 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2465 * is retransmitted. 2466 */ 2467 if (!IS_FASTOPEN(tp->t_flags)) 2468 #endif 2469 cc_conn_init(tp); 2470 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2471 } 2472 /* 2473 * If segment contains data or ACK, will call tcp_reass() 2474 * later; if not, do so now to pass queued data to user. 2475 */ 2476 if (tlen == 0 && (thflags & TH_FIN) == 0) 2477 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2478 (struct mbuf *)0); 2479 tp->snd_wl1 = th->th_seq - 1; 2480 /* FALLTHROUGH */ 2481 2482 /* 2483 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2484 * ACKs. If the ack is in the range 2485 * tp->snd_una < th->th_ack <= tp->snd_max 2486 * then advance tp->snd_una to th->th_ack and drop 2487 * data from the retransmission queue. If this ACK reflects 2488 * more up to date window information we update our window information. 2489 */ 2490 case TCPS_ESTABLISHED: 2491 case TCPS_FIN_WAIT_1: 2492 case TCPS_FIN_WAIT_2: 2493 case TCPS_CLOSE_WAIT: 2494 case TCPS_CLOSING: 2495 case TCPS_LAST_ACK: 2496 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2497 TCPSTAT_INC(tcps_rcvacktoomuch); 2498 goto dropafterack; 2499 } 2500 if ((tp->t_flags & TF_SACK_PERMIT) && 2501 ((to.to_flags & TOF_SACK) || 2502 !TAILQ_EMPTY(&tp->snd_holes))) 2503 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2504 else 2505 /* 2506 * Reset the value so that previous (valid) value 2507 * from the last ack with SACK doesn't get used. 2508 */ 2509 tp->sackhint.sacked_bytes = 0; 2510 2511 #ifdef TCP_HHOOK 2512 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2513 hhook_run_tcp_est_in(tp, th, &to); 2514 #endif 2515 2516 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2517 u_int maxseg; 2518 2519 maxseg = tcp_maxseg(tp); 2520 if (tlen == 0 && 2521 (tiwin == tp->snd_wnd || 2522 (tp->t_flags & TF_SACK_PERMIT))) { 2523 /* 2524 * If this is the first time we've seen a 2525 * FIN from the remote, this is not a 2526 * duplicate and it needs to be processed 2527 * normally. This happens during a 2528 * simultaneous close. 2529 */ 2530 if ((thflags & TH_FIN) && 2531 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2532 tp->t_dupacks = 0; 2533 break; 2534 } 2535 TCPSTAT_INC(tcps_rcvdupack); 2536 /* 2537 * If we have outstanding data (other than 2538 * a window probe), this is a completely 2539 * duplicate ack (ie, window info didn't 2540 * change and FIN isn't set), 2541 * the ack is the biggest we've 2542 * seen and we've seen exactly our rexmt 2543 * threshold of them, assume a packet 2544 * has been dropped and retransmit it. 2545 * Kludge snd_nxt & the congestion 2546 * window so we send only this one 2547 * packet. 2548 * 2549 * We know we're losing at the current 2550 * window size so do congestion avoidance 2551 * (set ssthresh to half the current window 2552 * and pull our congestion window back to 2553 * the new ssthresh). 2554 * 2555 * Dup acks mean that packets have left the 2556 * network (they're now cached at the receiver) 2557 * so bump cwnd by the amount in the receiver 2558 * to keep a constant cwnd packets in the 2559 * network. 2560 * 2561 * When using TCP ECN, notify the peer that 2562 * we reduced the cwnd. 2563 */ 2564 /* 2565 * Following 2 kinds of acks should not affect 2566 * dupack counting: 2567 * 1) Old acks 2568 * 2) Acks with SACK but without any new SACK 2569 * information in them. These could result from 2570 * any anomaly in the network like a switch 2571 * duplicating packets or a possible DoS attack. 2572 */ 2573 if (th->th_ack != tp->snd_una || 2574 ((tp->t_flags & TF_SACK_PERMIT) && 2575 !sack_changed)) 2576 break; 2577 else if (!tcp_timer_active(tp, TT_REXMT)) 2578 tp->t_dupacks = 0; 2579 else if (++tp->t_dupacks > tcprexmtthresh || 2580 IN_FASTRECOVERY(tp->t_flags)) { 2581 cc_ack_received(tp, th, nsegs, 2582 CC_DUPACK); 2583 if ((tp->t_flags & TF_SACK_PERMIT) && 2584 IN_FASTRECOVERY(tp->t_flags)) { 2585 int awnd; 2586 2587 /* 2588 * Compute the amount of data in flight first. 2589 * We can inject new data into the pipe iff 2590 * we have less than 1/2 the original window's 2591 * worth of data in flight. 2592 */ 2593 if (V_tcp_do_rfc6675_pipe) 2594 awnd = tcp_compute_pipe(tp); 2595 else 2596 awnd = (tp->snd_nxt - tp->snd_fack) + 2597 tp->sackhint.sack_bytes_rexmit; 2598 2599 if (awnd < tp->snd_ssthresh) { 2600 tp->snd_cwnd += maxseg; 2601 /* 2602 * RFC5681 Section 3.2 talks about cwnd 2603 * inflation on additional dupacks and 2604 * deflation on recovering from loss. 2605 * 2606 * We keep cwnd into check so that 2607 * we don't have to 'deflate' it when we 2608 * get out of recovery. 2609 */ 2610 if (tp->snd_cwnd > tp->snd_ssthresh) 2611 tp->snd_cwnd = tp->snd_ssthresh; 2612 } 2613 } else 2614 tp->snd_cwnd += maxseg; 2615 (void) tp->t_fb->tfb_tcp_output(tp); 2616 goto drop; 2617 } else if (tp->t_dupacks == tcprexmtthresh) { 2618 tcp_seq onxt = tp->snd_nxt; 2619 2620 /* 2621 * If we're doing sack, check to 2622 * see if we're already in sack 2623 * recovery. If we're not doing sack, 2624 * check to see if we're in newreno 2625 * recovery. 2626 */ 2627 if (tp->t_flags & TF_SACK_PERMIT) { 2628 if (IN_FASTRECOVERY(tp->t_flags)) { 2629 tp->t_dupacks = 0; 2630 break; 2631 } 2632 } else { 2633 if (SEQ_LEQ(th->th_ack, 2634 tp->snd_recover)) { 2635 tp->t_dupacks = 0; 2636 break; 2637 } 2638 } 2639 /* Congestion signal before ack. */ 2640 cc_cong_signal(tp, th, CC_NDUPACK); 2641 cc_ack_received(tp, th, nsegs, 2642 CC_DUPACK); 2643 tcp_timer_activate(tp, TT_REXMT, 0); 2644 tp->t_rtttime = 0; 2645 if (tp->t_flags & TF_SACK_PERMIT) { 2646 TCPSTAT_INC( 2647 tcps_sack_recovery_episode); 2648 tp->sack_newdata = tp->snd_nxt; 2649 if (CC_ALGO(tp)->cong_signal == NULL) 2650 tp->snd_cwnd = maxseg; 2651 (void) tp->t_fb->tfb_tcp_output(tp); 2652 goto drop; 2653 } 2654 tp->snd_nxt = th->th_ack; 2655 if (CC_ALGO(tp)->cong_signal == NULL) 2656 tp->snd_cwnd = maxseg; 2657 (void) tp->t_fb->tfb_tcp_output(tp); 2658 KASSERT(tp->snd_limited <= 2, 2659 ("%s: tp->snd_limited too big", 2660 __func__)); 2661 if (CC_ALGO(tp)->cong_signal == NULL) 2662 tp->snd_cwnd = tp->snd_ssthresh + 2663 maxseg * 2664 (tp->t_dupacks - tp->snd_limited); 2665 if (SEQ_GT(onxt, tp->snd_nxt)) 2666 tp->snd_nxt = onxt; 2667 goto drop; 2668 } else if (V_tcp_do_rfc3042) { 2669 /* 2670 * Process first and second duplicate 2671 * ACKs. Each indicates a segment 2672 * leaving the network, creating room 2673 * for more. Make sure we can send a 2674 * packet on reception of each duplicate 2675 * ACK by increasing snd_cwnd by one 2676 * segment. Restore the original 2677 * snd_cwnd after packet transmission. 2678 */ 2679 cc_ack_received(tp, th, nsegs, 2680 CC_DUPACK); 2681 uint32_t oldcwnd = tp->snd_cwnd; 2682 tcp_seq oldsndmax = tp->snd_max; 2683 u_int sent; 2684 int avail; 2685 2686 KASSERT(tp->t_dupacks == 1 || 2687 tp->t_dupacks == 2, 2688 ("%s: dupacks not 1 or 2", 2689 __func__)); 2690 if (tp->t_dupacks == 1) 2691 tp->snd_limited = 0; 2692 tp->snd_cwnd = 2693 (tp->snd_nxt - tp->snd_una) + 2694 (tp->t_dupacks - tp->snd_limited) * 2695 maxseg; 2696 /* 2697 * Only call tcp_output when there 2698 * is new data available to be sent. 2699 * Otherwise we would send pure ACKs. 2700 */ 2701 SOCKBUF_LOCK(&so->so_snd); 2702 avail = sbavail(&so->so_snd) - 2703 (tp->snd_nxt - tp->snd_una); 2704 SOCKBUF_UNLOCK(&so->so_snd); 2705 if (avail > 0) 2706 (void) tp->t_fb->tfb_tcp_output(tp); 2707 sent = tp->snd_max - oldsndmax; 2708 if (sent > maxseg) { 2709 KASSERT((tp->t_dupacks == 2 && 2710 tp->snd_limited == 0) || 2711 (sent == maxseg + 1 && 2712 tp->t_flags & TF_SENTFIN), 2713 ("%s: sent too much", 2714 __func__)); 2715 tp->snd_limited = 2; 2716 } else if (sent > 0) 2717 ++tp->snd_limited; 2718 tp->snd_cwnd = oldcwnd; 2719 goto drop; 2720 } 2721 } 2722 break; 2723 } else { 2724 /* 2725 * This ack is advancing the left edge, reset the 2726 * counter. 2727 */ 2728 tp->t_dupacks = 0; 2729 /* 2730 * If this ack also has new SACK info, increment the 2731 * counter as per rfc6675. 2732 */ 2733 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2734 tp->t_dupacks++; 2735 } 2736 2737 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2738 ("%s: th_ack <= snd_una", __func__)); 2739 2740 /* 2741 * If the congestion window was inflated to account 2742 * for the other side's cached packets, retract it. 2743 */ 2744 if (IN_FASTRECOVERY(tp->t_flags)) { 2745 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2746 if (tp->t_flags & TF_SACK_PERMIT) 2747 tcp_sack_partialack(tp, th); 2748 else 2749 tcp_newreno_partial_ack(tp, th); 2750 } else 2751 cc_post_recovery(tp, th); 2752 } 2753 /* 2754 * If we reach this point, ACK is not a duplicate, 2755 * i.e., it ACKs something we sent. 2756 */ 2757 if (tp->t_flags & TF_NEEDSYN) { 2758 /* 2759 * T/TCP: Connection was half-synchronized, and our 2760 * SYN has been ACK'd (so connection is now fully 2761 * synchronized). Go to non-starred state, 2762 * increment snd_una for ACK of SYN, and check if 2763 * we can do window scaling. 2764 */ 2765 tp->t_flags &= ~TF_NEEDSYN; 2766 tp->snd_una++; 2767 /* Do window scaling? */ 2768 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2769 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2770 tp->rcv_scale = tp->request_r_scale; 2771 /* Send window already scaled. */ 2772 } 2773 } 2774 2775 process_ACK: 2776 INP_WLOCK_ASSERT(tp->t_inpcb); 2777 2778 acked = BYTES_THIS_ACK(tp, th); 2779 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2780 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2781 tp->snd_una, th->th_ack, tp, m)); 2782 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2783 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2784 2785 /* 2786 * If we just performed our first retransmit, and the ACK 2787 * arrives within our recovery window, then it was a mistake 2788 * to do the retransmit in the first place. Recover our 2789 * original cwnd and ssthresh, and proceed to transmit where 2790 * we left off. 2791 */ 2792 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2793 (int)(ticks - tp->t_badrxtwin) < 0) 2794 cc_cong_signal(tp, th, CC_RTO_ERR); 2795 2796 /* 2797 * If we have a timestamp reply, update smoothed 2798 * round trip time. If no timestamp is present but 2799 * transmit timer is running and timed sequence 2800 * number was acked, update smoothed round trip time. 2801 * Since we now have an rtt measurement, cancel the 2802 * timer backoff (cf., Phil Karn's retransmit alg.). 2803 * Recompute the initial retransmit timer. 2804 * 2805 * Some boxes send broken timestamp replies 2806 * during the SYN+ACK phase, ignore 2807 * timestamps of 0 or we could calculate a 2808 * huge RTT and blow up the retransmit timer. 2809 */ 2810 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2811 uint32_t t; 2812 2813 t = tcp_ts_getticks() - to.to_tsecr; 2814 if (!tp->t_rttlow || tp->t_rttlow > t) 2815 tp->t_rttlow = t; 2816 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2817 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2818 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2819 tp->t_rttlow = ticks - tp->t_rtttime; 2820 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2821 } 2822 2823 /* 2824 * If all outstanding data is acked, stop retransmit 2825 * timer and remember to restart (more output or persist). 2826 * If there is more data to be acked, restart retransmit 2827 * timer, using current (possibly backed-off) value. 2828 */ 2829 if (th->th_ack == tp->snd_max) { 2830 tcp_timer_activate(tp, TT_REXMT, 0); 2831 needoutput = 1; 2832 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2833 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2834 2835 /* 2836 * If no data (only SYN) was ACK'd, 2837 * skip rest of ACK processing. 2838 */ 2839 if (acked == 0) 2840 goto step6; 2841 2842 /* 2843 * Let the congestion control algorithm update congestion 2844 * control related information. This typically means increasing 2845 * the congestion window. 2846 */ 2847 cc_ack_received(tp, th, nsegs, CC_ACK); 2848 2849 SOCKBUF_LOCK(&so->so_snd); 2850 if (acked > sbavail(&so->so_snd)) { 2851 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2852 tp->snd_wnd -= sbavail(&so->so_snd); 2853 else 2854 tp->snd_wnd = 0; 2855 mfree = sbcut_locked(&so->so_snd, 2856 (int)sbavail(&so->so_snd)); 2857 ourfinisacked = 1; 2858 } else { 2859 mfree = sbcut_locked(&so->so_snd, acked); 2860 if (tp->snd_wnd >= (uint32_t) acked) 2861 tp->snd_wnd -= acked; 2862 else 2863 tp->snd_wnd = 0; 2864 ourfinisacked = 0; 2865 } 2866 /* NB: sowwakeup_locked() does an implicit unlock. */ 2867 sowwakeup_locked(so); 2868 m_freem(mfree); 2869 /* Detect una wraparound. */ 2870 if (!IN_RECOVERY(tp->t_flags) && 2871 SEQ_GT(tp->snd_una, tp->snd_recover) && 2872 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2873 tp->snd_recover = th->th_ack - 1; 2874 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2875 if (IN_RECOVERY(tp->t_flags) && 2876 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2877 EXIT_RECOVERY(tp->t_flags); 2878 } 2879 tp->snd_una = th->th_ack; 2880 if (tp->t_flags & TF_SACK_PERMIT) { 2881 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2882 tp->snd_recover = tp->snd_una; 2883 } 2884 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2885 tp->snd_nxt = tp->snd_una; 2886 2887 switch (tp->t_state) { 2888 2889 /* 2890 * In FIN_WAIT_1 STATE in addition to the processing 2891 * for the ESTABLISHED state if our FIN is now acknowledged 2892 * then enter FIN_WAIT_2. 2893 */ 2894 case TCPS_FIN_WAIT_1: 2895 if (ourfinisacked) { 2896 /* 2897 * If we can't receive any more 2898 * data, then closing user can proceed. 2899 * Starting the timer is contrary to the 2900 * specification, but if we don't get a FIN 2901 * we'll hang forever. 2902 * 2903 * XXXjl: 2904 * we should release the tp also, and use a 2905 * compressed state. 2906 */ 2907 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2908 soisdisconnected(so); 2909 tcp_timer_activate(tp, TT_2MSL, 2910 (tcp_fast_finwait2_recycle ? 2911 tcp_finwait2_timeout : 2912 TP_MAXIDLE(tp))); 2913 } 2914 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2915 } 2916 break; 2917 2918 /* 2919 * In CLOSING STATE in addition to the processing for 2920 * the ESTABLISHED state if the ACK acknowledges our FIN 2921 * then enter the TIME-WAIT state, otherwise ignore 2922 * the segment. 2923 */ 2924 case TCPS_CLOSING: 2925 if (ourfinisacked) { 2926 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2927 tcp_twstart(tp); 2928 INP_INFO_RUNLOCK(&V_tcbinfo); 2929 m_freem(m); 2930 return; 2931 } 2932 break; 2933 2934 /* 2935 * In LAST_ACK, we may still be waiting for data to drain 2936 * and/or to be acked, as well as for the ack of our FIN. 2937 * If our FIN is now acknowledged, delete the TCB, 2938 * enter the closed state and return. 2939 */ 2940 case TCPS_LAST_ACK: 2941 if (ourfinisacked) { 2942 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2943 tp = tcp_close(tp); 2944 goto drop; 2945 } 2946 break; 2947 } 2948 } 2949 2950 step6: 2951 INP_WLOCK_ASSERT(tp->t_inpcb); 2952 2953 /* 2954 * Update window information. 2955 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2956 */ 2957 if ((thflags & TH_ACK) && 2958 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2959 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2960 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2961 /* keep track of pure window updates */ 2962 if (tlen == 0 && 2963 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2964 TCPSTAT_INC(tcps_rcvwinupd); 2965 tp->snd_wnd = tiwin; 2966 tp->snd_wl1 = th->th_seq; 2967 tp->snd_wl2 = th->th_ack; 2968 if (tp->snd_wnd > tp->max_sndwnd) 2969 tp->max_sndwnd = tp->snd_wnd; 2970 needoutput = 1; 2971 } 2972 2973 /* 2974 * Process segments with URG. 2975 */ 2976 if ((thflags & TH_URG) && th->th_urp && 2977 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2978 /* 2979 * This is a kludge, but if we receive and accept 2980 * random urgent pointers, we'll crash in 2981 * soreceive. It's hard to imagine someone 2982 * actually wanting to send this much urgent data. 2983 */ 2984 SOCKBUF_LOCK(&so->so_rcv); 2985 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2986 th->th_urp = 0; /* XXX */ 2987 thflags &= ~TH_URG; /* XXX */ 2988 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2989 goto dodata; /* XXX */ 2990 } 2991 /* 2992 * If this segment advances the known urgent pointer, 2993 * then mark the data stream. This should not happen 2994 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2995 * a FIN has been received from the remote side. 2996 * In these states we ignore the URG. 2997 * 2998 * According to RFC961 (Assigned Protocols), 2999 * the urgent pointer points to the last octet 3000 * of urgent data. We continue, however, 3001 * to consider it to indicate the first octet 3002 * of data past the urgent section as the original 3003 * spec states (in one of two places). 3004 */ 3005 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3006 tp->rcv_up = th->th_seq + th->th_urp; 3007 so->so_oobmark = sbavail(&so->so_rcv) + 3008 (tp->rcv_up - tp->rcv_nxt) - 1; 3009 if (so->so_oobmark == 0) 3010 so->so_rcv.sb_state |= SBS_RCVATMARK; 3011 sohasoutofband(so); 3012 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3013 } 3014 SOCKBUF_UNLOCK(&so->so_rcv); 3015 /* 3016 * Remove out of band data so doesn't get presented to user. 3017 * This can happen independent of advancing the URG pointer, 3018 * but if two URG's are pending at once, some out-of-band 3019 * data may creep in... ick. 3020 */ 3021 if (th->th_urp <= (uint32_t)tlen && 3022 !(so->so_options & SO_OOBINLINE)) { 3023 /* hdr drop is delayed */ 3024 tcp_pulloutofband(so, th, m, drop_hdrlen); 3025 } 3026 } else { 3027 /* 3028 * If no out of band data is expected, 3029 * pull receive urgent pointer along 3030 * with the receive window. 3031 */ 3032 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3033 tp->rcv_up = tp->rcv_nxt; 3034 } 3035 dodata: /* XXX */ 3036 INP_WLOCK_ASSERT(tp->t_inpcb); 3037 3038 /* 3039 * Process the segment text, merging it into the TCP sequencing queue, 3040 * and arranging for acknowledgment of receipt if necessary. 3041 * This process logically involves adjusting tp->rcv_wnd as data 3042 * is presented to the user (this happens in tcp_usrreq.c, 3043 * case PRU_RCVD). If a FIN has already been received on this 3044 * connection then we just ignore the text. 3045 */ 3046 #ifdef TCP_RFC7413 3047 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3048 IS_FASTOPEN(tp->t_flags)); 3049 #else 3050 #define tfo_syn (false) 3051 #endif 3052 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3053 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3054 tcp_seq save_start = th->th_seq; 3055 m_adj(m, drop_hdrlen); /* delayed header drop */ 3056 /* 3057 * Insert segment which includes th into TCP reassembly queue 3058 * with control block tp. Set thflags to whether reassembly now 3059 * includes a segment with FIN. This handles the common case 3060 * inline (segment is the next to be received on an established 3061 * connection, and the queue is empty), avoiding linkage into 3062 * and removal from the queue and repetition of various 3063 * conversions. 3064 * Set DELACK for segments received in order, but ack 3065 * immediately when segments are out of order (so 3066 * fast retransmit can work). 3067 */ 3068 if (th->th_seq == tp->rcv_nxt && 3069 LIST_EMPTY(&tp->t_segq) && 3070 (TCPS_HAVEESTABLISHED(tp->t_state) || 3071 tfo_syn)) { 3072 if (DELAY_ACK(tp, tlen) || tfo_syn) 3073 tp->t_flags |= TF_DELACK; 3074 else 3075 tp->t_flags |= TF_ACKNOW; 3076 tp->rcv_nxt += tlen; 3077 thflags = th->th_flags & TH_FIN; 3078 TCPSTAT_INC(tcps_rcvpack); 3079 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3080 SOCKBUF_LOCK(&so->so_rcv); 3081 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3082 m_freem(m); 3083 else 3084 sbappendstream_locked(&so->so_rcv, m, 0); 3085 /* NB: sorwakeup_locked() does an implicit unlock. */ 3086 sorwakeup_locked(so); 3087 } else { 3088 /* 3089 * XXX: Due to the header drop above "th" is 3090 * theoretically invalid by now. Fortunately 3091 * m_adj() doesn't actually frees any mbufs 3092 * when trimming from the head. 3093 */ 3094 thflags = tcp_reass(tp, th, &tlen, m); 3095 tp->t_flags |= TF_ACKNOW; 3096 } 3097 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3098 tcp_update_sack_list(tp, save_start, save_start + tlen); 3099 #if 0 3100 /* 3101 * Note the amount of data that peer has sent into 3102 * our window, in order to estimate the sender's 3103 * buffer size. 3104 * XXX: Unused. 3105 */ 3106 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3107 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3108 else 3109 len = so->so_rcv.sb_hiwat; 3110 #endif 3111 } else { 3112 m_freem(m); 3113 thflags &= ~TH_FIN; 3114 } 3115 3116 /* 3117 * If FIN is received ACK the FIN and let the user know 3118 * that the connection is closing. 3119 */ 3120 if (thflags & TH_FIN) { 3121 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3122 socantrcvmore(so); 3123 /* 3124 * If connection is half-synchronized 3125 * (ie NEEDSYN flag on) then delay ACK, 3126 * so it may be piggybacked when SYN is sent. 3127 * Otherwise, since we received a FIN then no 3128 * more input can be expected, send ACK now. 3129 */ 3130 if (tp->t_flags & TF_NEEDSYN) 3131 tp->t_flags |= TF_DELACK; 3132 else 3133 tp->t_flags |= TF_ACKNOW; 3134 tp->rcv_nxt++; 3135 } 3136 switch (tp->t_state) { 3137 3138 /* 3139 * In SYN_RECEIVED and ESTABLISHED STATES 3140 * enter the CLOSE_WAIT state. 3141 */ 3142 case TCPS_SYN_RECEIVED: 3143 tp->t_starttime = ticks; 3144 /* FALLTHROUGH */ 3145 case TCPS_ESTABLISHED: 3146 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3147 break; 3148 3149 /* 3150 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3151 * enter the CLOSING state. 3152 */ 3153 case TCPS_FIN_WAIT_1: 3154 tcp_state_change(tp, TCPS_CLOSING); 3155 break; 3156 3157 /* 3158 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3159 * starting the time-wait timer, turning off the other 3160 * standard timers. 3161 */ 3162 case TCPS_FIN_WAIT_2: 3163 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3164 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3165 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3166 ti_locked)); 3167 3168 tcp_twstart(tp); 3169 INP_INFO_RUNLOCK(&V_tcbinfo); 3170 return; 3171 } 3172 } 3173 if (ti_locked == TI_RLOCKED) 3174 INP_INFO_RUNLOCK(&V_tcbinfo); 3175 ti_locked = TI_UNLOCKED; 3176 3177 #ifdef TCPDEBUG 3178 if (so->so_options & SO_DEBUG) 3179 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3180 &tcp_savetcp, 0); 3181 #endif 3182 TCP_PROBE3(debug__input, tp, th, m); 3183 3184 /* 3185 * Return any desired output. 3186 */ 3187 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3188 (void) tp->t_fb->tfb_tcp_output(tp); 3189 3190 check_delack: 3191 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3192 __func__, ti_locked)); 3193 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3194 INP_WLOCK_ASSERT(tp->t_inpcb); 3195 3196 if (tp->t_flags & TF_DELACK) { 3197 tp->t_flags &= ~TF_DELACK; 3198 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3199 } 3200 INP_WUNLOCK(tp->t_inpcb); 3201 return; 3202 3203 dropafterack: 3204 /* 3205 * Generate an ACK dropping incoming segment if it occupies 3206 * sequence space, where the ACK reflects our state. 3207 * 3208 * We can now skip the test for the RST flag since all 3209 * paths to this code happen after packets containing 3210 * RST have been dropped. 3211 * 3212 * In the SYN-RECEIVED state, don't send an ACK unless the 3213 * segment we received passes the SYN-RECEIVED ACK test. 3214 * If it fails send a RST. This breaks the loop in the 3215 * "LAND" DoS attack, and also prevents an ACK storm 3216 * between two listening ports that have been sent forged 3217 * SYN segments, each with the source address of the other. 3218 */ 3219 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3220 (SEQ_GT(tp->snd_una, th->th_ack) || 3221 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3222 rstreason = BANDLIM_RST_OPENPORT; 3223 goto dropwithreset; 3224 } 3225 #ifdef TCPDEBUG 3226 if (so->so_options & SO_DEBUG) 3227 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3228 &tcp_savetcp, 0); 3229 #endif 3230 TCP_PROBE3(debug__input, tp, th, m); 3231 if (ti_locked == TI_RLOCKED) 3232 INP_INFO_RUNLOCK(&V_tcbinfo); 3233 ti_locked = TI_UNLOCKED; 3234 3235 tp->t_flags |= TF_ACKNOW; 3236 (void) tp->t_fb->tfb_tcp_output(tp); 3237 INP_WUNLOCK(tp->t_inpcb); 3238 m_freem(m); 3239 return; 3240 3241 dropwithreset: 3242 if (ti_locked == TI_RLOCKED) 3243 INP_INFO_RUNLOCK(&V_tcbinfo); 3244 ti_locked = TI_UNLOCKED; 3245 3246 if (tp != NULL) { 3247 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3248 INP_WUNLOCK(tp->t_inpcb); 3249 } else 3250 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3251 return; 3252 3253 drop: 3254 if (ti_locked == TI_RLOCKED) { 3255 INP_INFO_RUNLOCK(&V_tcbinfo); 3256 ti_locked = TI_UNLOCKED; 3257 } 3258 #ifdef INVARIANTS 3259 else 3260 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3261 #endif 3262 3263 /* 3264 * Drop space held by incoming segment and return. 3265 */ 3266 #ifdef TCPDEBUG 3267 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3268 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3269 &tcp_savetcp, 0); 3270 #endif 3271 TCP_PROBE3(debug__input, tp, th, m); 3272 if (tp != NULL) 3273 INP_WUNLOCK(tp->t_inpcb); 3274 m_freem(m); 3275 #ifndef TCP_RFC7413 3276 #undef tfo_syn 3277 #endif 3278 } 3279 3280 /* 3281 * Issue RST and make ACK acceptable to originator of segment. 3282 * The mbuf must still include the original packet header. 3283 * tp may be NULL. 3284 */ 3285 void 3286 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3287 int tlen, int rstreason) 3288 { 3289 #ifdef INET 3290 struct ip *ip; 3291 #endif 3292 #ifdef INET6 3293 struct ip6_hdr *ip6; 3294 #endif 3295 3296 if (tp != NULL) { 3297 INP_WLOCK_ASSERT(tp->t_inpcb); 3298 } 3299 3300 /* Don't bother if destination was broadcast/multicast. */ 3301 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3302 goto drop; 3303 #ifdef INET6 3304 if (mtod(m, struct ip *)->ip_v == 6) { 3305 ip6 = mtod(m, struct ip6_hdr *); 3306 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3307 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3308 goto drop; 3309 /* IPv6 anycast check is done at tcp6_input() */ 3310 } 3311 #endif 3312 #if defined(INET) && defined(INET6) 3313 else 3314 #endif 3315 #ifdef INET 3316 { 3317 ip = mtod(m, struct ip *); 3318 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3319 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3320 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3321 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3322 goto drop; 3323 } 3324 #endif 3325 3326 /* Perform bandwidth limiting. */ 3327 if (badport_bandlim(rstreason) < 0) 3328 goto drop; 3329 3330 /* tcp_respond consumes the mbuf chain. */ 3331 if (th->th_flags & TH_ACK) { 3332 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3333 th->th_ack, TH_RST); 3334 } else { 3335 if (th->th_flags & TH_SYN) 3336 tlen++; 3337 if (th->th_flags & TH_FIN) 3338 tlen++; 3339 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3340 (tcp_seq)0, TH_RST|TH_ACK); 3341 } 3342 return; 3343 drop: 3344 m_freem(m); 3345 } 3346 3347 /* 3348 * Parse TCP options and place in tcpopt. 3349 */ 3350 void 3351 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3352 { 3353 int opt, optlen; 3354 3355 to->to_flags = 0; 3356 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3357 opt = cp[0]; 3358 if (opt == TCPOPT_EOL) 3359 break; 3360 if (opt == TCPOPT_NOP) 3361 optlen = 1; 3362 else { 3363 if (cnt < 2) 3364 break; 3365 optlen = cp[1]; 3366 if (optlen < 2 || optlen > cnt) 3367 break; 3368 } 3369 switch (opt) { 3370 case TCPOPT_MAXSEG: 3371 if (optlen != TCPOLEN_MAXSEG) 3372 continue; 3373 if (!(flags & TO_SYN)) 3374 continue; 3375 to->to_flags |= TOF_MSS; 3376 bcopy((char *)cp + 2, 3377 (char *)&to->to_mss, sizeof(to->to_mss)); 3378 to->to_mss = ntohs(to->to_mss); 3379 break; 3380 case TCPOPT_WINDOW: 3381 if (optlen != TCPOLEN_WINDOW) 3382 continue; 3383 if (!(flags & TO_SYN)) 3384 continue; 3385 to->to_flags |= TOF_SCALE; 3386 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3387 break; 3388 case TCPOPT_TIMESTAMP: 3389 if (optlen != TCPOLEN_TIMESTAMP) 3390 continue; 3391 to->to_flags |= TOF_TS; 3392 bcopy((char *)cp + 2, 3393 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3394 to->to_tsval = ntohl(to->to_tsval); 3395 bcopy((char *)cp + 6, 3396 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3397 to->to_tsecr = ntohl(to->to_tsecr); 3398 break; 3399 case TCPOPT_SIGNATURE: 3400 /* 3401 * In order to reply to a host which has set the 3402 * TCP_SIGNATURE option in its initial SYN, we have 3403 * to record the fact that the option was observed 3404 * here for the syncache code to perform the correct 3405 * response. 3406 */ 3407 if (optlen != TCPOLEN_SIGNATURE) 3408 continue; 3409 to->to_flags |= TOF_SIGNATURE; 3410 to->to_signature = cp + 2; 3411 break; 3412 case TCPOPT_SACK_PERMITTED: 3413 if (optlen != TCPOLEN_SACK_PERMITTED) 3414 continue; 3415 if (!(flags & TO_SYN)) 3416 continue; 3417 if (!V_tcp_do_sack) 3418 continue; 3419 to->to_flags |= TOF_SACKPERM; 3420 break; 3421 case TCPOPT_SACK: 3422 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3423 continue; 3424 if (flags & TO_SYN) 3425 continue; 3426 to->to_flags |= TOF_SACK; 3427 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3428 to->to_sacks = cp + 2; 3429 TCPSTAT_INC(tcps_sack_rcv_blocks); 3430 break; 3431 #ifdef TCP_RFC7413 3432 case TCPOPT_FAST_OPEN: 3433 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3434 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3435 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3436 continue; 3437 if (!(flags & TO_SYN)) 3438 continue; 3439 if (!V_tcp_fastopen_enabled) 3440 continue; 3441 to->to_flags |= TOF_FASTOPEN; 3442 to->to_tfo_len = optlen - 2; 3443 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3444 break; 3445 #endif 3446 default: 3447 continue; 3448 } 3449 } 3450 } 3451 3452 /* 3453 * Pull out of band byte out of a segment so 3454 * it doesn't appear in the user's data queue. 3455 * It is still reflected in the segment length for 3456 * sequencing purposes. 3457 */ 3458 void 3459 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3460 int off) 3461 { 3462 int cnt = off + th->th_urp - 1; 3463 3464 while (cnt >= 0) { 3465 if (m->m_len > cnt) { 3466 char *cp = mtod(m, caddr_t) + cnt; 3467 struct tcpcb *tp = sototcpcb(so); 3468 3469 INP_WLOCK_ASSERT(tp->t_inpcb); 3470 3471 tp->t_iobc = *cp; 3472 tp->t_oobflags |= TCPOOB_HAVEDATA; 3473 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3474 m->m_len--; 3475 if (m->m_flags & M_PKTHDR) 3476 m->m_pkthdr.len--; 3477 return; 3478 } 3479 cnt -= m->m_len; 3480 m = m->m_next; 3481 if (m == NULL) 3482 break; 3483 } 3484 panic("tcp_pulloutofband"); 3485 } 3486 3487 /* 3488 * Collect new round-trip time estimate 3489 * and update averages and current timeout. 3490 */ 3491 void 3492 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3493 { 3494 int delta; 3495 3496 INP_WLOCK_ASSERT(tp->t_inpcb); 3497 3498 TCPSTAT_INC(tcps_rttupdated); 3499 tp->t_rttupdated++; 3500 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3501 /* 3502 * srtt is stored as fixed point with 5 bits after the 3503 * binary point (i.e., scaled by 8). The following magic 3504 * is equivalent to the smoothing algorithm in rfc793 with 3505 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3506 * point). Adjust rtt to origin 0. 3507 */ 3508 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3509 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3510 3511 if ((tp->t_srtt += delta) <= 0) 3512 tp->t_srtt = 1; 3513 3514 /* 3515 * We accumulate a smoothed rtt variance (actually, a 3516 * smoothed mean difference), then set the retransmit 3517 * timer to smoothed rtt + 4 times the smoothed variance. 3518 * rttvar is stored as fixed point with 4 bits after the 3519 * binary point (scaled by 16). The following is 3520 * equivalent to rfc793 smoothing with an alpha of .75 3521 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3522 * rfc793's wired-in beta. 3523 */ 3524 if (delta < 0) 3525 delta = -delta; 3526 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3527 if ((tp->t_rttvar += delta) <= 0) 3528 tp->t_rttvar = 1; 3529 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3530 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3531 } else { 3532 /* 3533 * No rtt measurement yet - use the unsmoothed rtt. 3534 * Set the variance to half the rtt (so our first 3535 * retransmit happens at 3*rtt). 3536 */ 3537 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3538 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3539 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3540 } 3541 tp->t_rtttime = 0; 3542 tp->t_rxtshift = 0; 3543 3544 /* 3545 * the retransmit should happen at rtt + 4 * rttvar. 3546 * Because of the way we do the smoothing, srtt and rttvar 3547 * will each average +1/2 tick of bias. When we compute 3548 * the retransmit timer, we want 1/2 tick of rounding and 3549 * 1 extra tick because of +-1/2 tick uncertainty in the 3550 * firing of the timer. The bias will give us exactly the 3551 * 1.5 tick we need. But, because the bias is 3552 * statistical, we have to test that we don't drop below 3553 * the minimum feasible timer (which is 2 ticks). 3554 */ 3555 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3556 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3557 3558 /* 3559 * We received an ack for a packet that wasn't retransmitted; 3560 * it is probably safe to discard any error indications we've 3561 * received recently. This isn't quite right, but close enough 3562 * for now (a route might have failed after we sent a segment, 3563 * and the return path might not be symmetrical). 3564 */ 3565 tp->t_softerror = 0; 3566 } 3567 3568 /* 3569 * Determine a reasonable value for maxseg size. 3570 * If the route is known, check route for mtu. 3571 * If none, use an mss that can be handled on the outgoing interface 3572 * without forcing IP to fragment. If no route is found, route has no mtu, 3573 * or the destination isn't local, use a default, hopefully conservative 3574 * size (usually 512 or the default IP max size, but no more than the mtu 3575 * of the interface), as we can't discover anything about intervening 3576 * gateways or networks. We also initialize the congestion/slow start 3577 * window to be a single segment if the destination isn't local. 3578 * While looking at the routing entry, we also initialize other path-dependent 3579 * parameters from pre-set or cached values in the routing entry. 3580 * 3581 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3582 * IP options, e.g. IPSEC data, since length of this data may vary, and 3583 * thus it is calculated for every segment separately in tcp_output(). 3584 * 3585 * NOTE that this routine is only called when we process an incoming 3586 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3587 * settings are handled in tcp_mssopt(). 3588 */ 3589 void 3590 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3591 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3592 { 3593 int mss = 0; 3594 uint32_t maxmtu = 0; 3595 struct inpcb *inp = tp->t_inpcb; 3596 struct hc_metrics_lite metrics; 3597 #ifdef INET6 3598 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3599 size_t min_protoh = isipv6 ? 3600 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3601 sizeof (struct tcpiphdr); 3602 #else 3603 const size_t min_protoh = sizeof(struct tcpiphdr); 3604 #endif 3605 3606 INP_WLOCK_ASSERT(tp->t_inpcb); 3607 3608 if (mtuoffer != -1) { 3609 KASSERT(offer == -1, ("%s: conflict", __func__)); 3610 offer = mtuoffer - min_protoh; 3611 } 3612 3613 /* Initialize. */ 3614 #ifdef INET6 3615 if (isipv6) { 3616 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3617 tp->t_maxseg = V_tcp_v6mssdflt; 3618 } 3619 #endif 3620 #if defined(INET) && defined(INET6) 3621 else 3622 #endif 3623 #ifdef INET 3624 { 3625 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3626 tp->t_maxseg = V_tcp_mssdflt; 3627 } 3628 #endif 3629 3630 /* 3631 * No route to sender, stay with default mss and return. 3632 */ 3633 if (maxmtu == 0) { 3634 /* 3635 * In case we return early we need to initialize metrics 3636 * to a defined state as tcp_hc_get() would do for us 3637 * if there was no cache hit. 3638 */ 3639 if (metricptr != NULL) 3640 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3641 return; 3642 } 3643 3644 /* What have we got? */ 3645 switch (offer) { 3646 case 0: 3647 /* 3648 * Offer == 0 means that there was no MSS on the SYN 3649 * segment, in this case we use tcp_mssdflt as 3650 * already assigned to t_maxseg above. 3651 */ 3652 offer = tp->t_maxseg; 3653 break; 3654 3655 case -1: 3656 /* 3657 * Offer == -1 means that we didn't receive SYN yet. 3658 */ 3659 /* FALLTHROUGH */ 3660 3661 default: 3662 /* 3663 * Prevent DoS attack with too small MSS. Round up 3664 * to at least minmss. 3665 */ 3666 offer = max(offer, V_tcp_minmss); 3667 } 3668 3669 /* 3670 * rmx information is now retrieved from tcp_hostcache. 3671 */ 3672 tcp_hc_get(&inp->inp_inc, &metrics); 3673 if (metricptr != NULL) 3674 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3675 3676 /* 3677 * If there's a discovered mtu in tcp hostcache, use it. 3678 * Else, use the link mtu. 3679 */ 3680 if (metrics.rmx_mtu) 3681 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3682 else { 3683 #ifdef INET6 3684 if (isipv6) { 3685 mss = maxmtu - min_protoh; 3686 if (!V_path_mtu_discovery && 3687 !in6_localaddr(&inp->in6p_faddr)) 3688 mss = min(mss, V_tcp_v6mssdflt); 3689 } 3690 #endif 3691 #if defined(INET) && defined(INET6) 3692 else 3693 #endif 3694 #ifdef INET 3695 { 3696 mss = maxmtu - min_protoh; 3697 if (!V_path_mtu_discovery && 3698 !in_localaddr(inp->inp_faddr)) 3699 mss = min(mss, V_tcp_mssdflt); 3700 } 3701 #endif 3702 /* 3703 * XXX - The above conditional (mss = maxmtu - min_protoh) 3704 * probably violates the TCP spec. 3705 * The problem is that, since we don't know the 3706 * other end's MSS, we are supposed to use a conservative 3707 * default. But, if we do that, then MTU discovery will 3708 * never actually take place, because the conservative 3709 * default is much less than the MTUs typically seen 3710 * on the Internet today. For the moment, we'll sweep 3711 * this under the carpet. 3712 * 3713 * The conservative default might not actually be a problem 3714 * if the only case this occurs is when sending an initial 3715 * SYN with options and data to a host we've never talked 3716 * to before. Then, they will reply with an MSS value which 3717 * will get recorded and the new parameters should get 3718 * recomputed. For Further Study. 3719 */ 3720 } 3721 mss = min(mss, offer); 3722 3723 /* 3724 * Sanity check: make sure that maxseg will be large 3725 * enough to allow some data on segments even if the 3726 * all the option space is used (40bytes). Otherwise 3727 * funny things may happen in tcp_output. 3728 * 3729 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3730 */ 3731 mss = max(mss, 64); 3732 3733 tp->t_maxseg = mss; 3734 } 3735 3736 void 3737 tcp_mss(struct tcpcb *tp, int offer) 3738 { 3739 int mss; 3740 uint32_t bufsize; 3741 struct inpcb *inp; 3742 struct socket *so; 3743 struct hc_metrics_lite metrics; 3744 struct tcp_ifcap cap; 3745 3746 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3747 3748 bzero(&cap, sizeof(cap)); 3749 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3750 3751 mss = tp->t_maxseg; 3752 inp = tp->t_inpcb; 3753 3754 /* 3755 * If there's a pipesize, change the socket buffer to that size, 3756 * don't change if sb_hiwat is different than default (then it 3757 * has been changed on purpose with setsockopt). 3758 * Make the socket buffers an integral number of mss units; 3759 * if the mss is larger than the socket buffer, decrease the mss. 3760 */ 3761 so = inp->inp_socket; 3762 SOCKBUF_LOCK(&so->so_snd); 3763 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3764 bufsize = metrics.rmx_sendpipe; 3765 else 3766 bufsize = so->so_snd.sb_hiwat; 3767 if (bufsize < mss) 3768 mss = bufsize; 3769 else { 3770 bufsize = roundup(bufsize, mss); 3771 if (bufsize > sb_max) 3772 bufsize = sb_max; 3773 if (bufsize > so->so_snd.sb_hiwat) 3774 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3775 } 3776 SOCKBUF_UNLOCK(&so->so_snd); 3777 /* 3778 * Sanity check: make sure that maxseg will be large 3779 * enough to allow some data on segments even if the 3780 * all the option space is used (40bytes). Otherwise 3781 * funny things may happen in tcp_output. 3782 * 3783 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3784 */ 3785 tp->t_maxseg = max(mss, 64); 3786 3787 SOCKBUF_LOCK(&so->so_rcv); 3788 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3789 bufsize = metrics.rmx_recvpipe; 3790 else 3791 bufsize = so->so_rcv.sb_hiwat; 3792 if (bufsize > mss) { 3793 bufsize = roundup(bufsize, mss); 3794 if (bufsize > sb_max) 3795 bufsize = sb_max; 3796 if (bufsize > so->so_rcv.sb_hiwat) 3797 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3798 } 3799 SOCKBUF_UNLOCK(&so->so_rcv); 3800 3801 /* Check the interface for TSO capabilities. */ 3802 if (cap.ifcap & CSUM_TSO) { 3803 tp->t_flags |= TF_TSO; 3804 tp->t_tsomax = cap.tsomax; 3805 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3806 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3807 } 3808 } 3809 3810 /* 3811 * Determine the MSS option to send on an outgoing SYN. 3812 */ 3813 int 3814 tcp_mssopt(struct in_conninfo *inc) 3815 { 3816 int mss = 0; 3817 uint32_t thcmtu = 0; 3818 uint32_t maxmtu = 0; 3819 size_t min_protoh; 3820 3821 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3822 3823 #ifdef INET6 3824 if (inc->inc_flags & INC_ISIPV6) { 3825 mss = V_tcp_v6mssdflt; 3826 maxmtu = tcp_maxmtu6(inc, NULL); 3827 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3828 } 3829 #endif 3830 #if defined(INET) && defined(INET6) 3831 else 3832 #endif 3833 #ifdef INET 3834 { 3835 mss = V_tcp_mssdflt; 3836 maxmtu = tcp_maxmtu(inc, NULL); 3837 min_protoh = sizeof(struct tcpiphdr); 3838 } 3839 #endif 3840 #if defined(INET6) || defined(INET) 3841 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3842 #endif 3843 3844 if (maxmtu && thcmtu) 3845 mss = min(maxmtu, thcmtu) - min_protoh; 3846 else if (maxmtu || thcmtu) 3847 mss = max(maxmtu, thcmtu) - min_protoh; 3848 3849 return (mss); 3850 } 3851 3852 3853 /* 3854 * On a partial ack arrives, force the retransmission of the 3855 * next unacknowledged segment. Do not clear tp->t_dupacks. 3856 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3857 * be started again. 3858 */ 3859 void 3860 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3861 { 3862 tcp_seq onxt = tp->snd_nxt; 3863 uint32_t ocwnd = tp->snd_cwnd; 3864 u_int maxseg = tcp_maxseg(tp); 3865 3866 INP_WLOCK_ASSERT(tp->t_inpcb); 3867 3868 tcp_timer_activate(tp, TT_REXMT, 0); 3869 tp->t_rtttime = 0; 3870 tp->snd_nxt = th->th_ack; 3871 /* 3872 * Set snd_cwnd to one segment beyond acknowledged offset. 3873 * (tp->snd_una has not yet been updated when this function is called.) 3874 */ 3875 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3876 tp->t_flags |= TF_ACKNOW; 3877 (void) tp->t_fb->tfb_tcp_output(tp); 3878 tp->snd_cwnd = ocwnd; 3879 if (SEQ_GT(onxt, tp->snd_nxt)) 3880 tp->snd_nxt = onxt; 3881 /* 3882 * Partial window deflation. Relies on fact that tp->snd_una 3883 * not updated yet. 3884 */ 3885 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3886 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3887 else 3888 tp->snd_cwnd = 0; 3889 tp->snd_cwnd += maxseg; 3890 } 3891 3892 int 3893 tcp_compute_pipe(struct tcpcb *tp) 3894 { 3895 return (tp->snd_max - tp->snd_una + 3896 tp->sackhint.sack_bytes_rexmit - 3897 tp->sackhint.sacked_bytes); 3898 } 3899