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 void 1490 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1491 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1492 int ti_locked) 1493 { 1494 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1495 int rstreason, todrop, win; 1496 uint32_t tiwin; 1497 uint16_t nsegs; 1498 char *s; 1499 struct in_conninfo *inc; 1500 struct mbuf *mfree; 1501 struct tcpopt to; 1502 #ifdef TCP_RFC7413 1503 int tfo_syn; 1504 #endif 1505 1506 #ifdef TCPDEBUG 1507 /* 1508 * The size of tcp_saveipgen must be the size of the max ip header, 1509 * now IPv6. 1510 */ 1511 u_char tcp_saveipgen[IP6_HDR_LEN]; 1512 struct tcphdr tcp_savetcp; 1513 short ostate = 0; 1514 #endif 1515 thflags = th->th_flags; 1516 inc = &tp->t_inpcb->inp_inc; 1517 tp->sackhint.last_sack_ack = 0; 1518 sack_changed = 0; 1519 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1520 1521 /* 1522 * If this is either a state-changing packet or current state isn't 1523 * established, we require a write lock on tcbinfo. Otherwise, we 1524 * allow the tcbinfo to be in either alocked or unlocked, as the 1525 * caller may have unnecessarily acquired a write lock due to a race. 1526 */ 1527 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1528 tp->t_state != TCPS_ESTABLISHED) { 1529 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1530 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1531 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1532 } else { 1533 #ifdef INVARIANTS 1534 if (ti_locked == TI_RLOCKED) 1535 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1536 else { 1537 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1538 "ti_locked: %d", __func__, ti_locked)); 1539 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1540 } 1541 #endif 1542 } 1543 INP_WLOCK_ASSERT(tp->t_inpcb); 1544 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1545 __func__)); 1546 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1547 __func__)); 1548 1549 #ifdef TCPPCAP 1550 /* Save segment, if requested. */ 1551 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1552 #endif 1553 1554 /* 1555 * Segment received on connection. 1556 * Reset idle time and keep-alive timer. 1557 * XXX: This should be done after segment 1558 * validation to ignore broken/spoofed segs. 1559 */ 1560 tp->t_rcvtime = ticks; 1561 1562 /* 1563 * Scale up the window into a 32-bit value. 1564 * For the SYN_SENT state the scale is zero. 1565 */ 1566 tiwin = th->th_win << tp->snd_scale; 1567 1568 /* 1569 * TCP ECN processing. 1570 */ 1571 if (tp->t_flags & TF_ECN_PERMIT) { 1572 if (thflags & TH_CWR) 1573 tp->t_flags &= ~TF_ECN_SND_ECE; 1574 switch (iptos & IPTOS_ECN_MASK) { 1575 case IPTOS_ECN_CE: 1576 tp->t_flags |= TF_ECN_SND_ECE; 1577 TCPSTAT_INC(tcps_ecn_ce); 1578 break; 1579 case IPTOS_ECN_ECT0: 1580 TCPSTAT_INC(tcps_ecn_ect0); 1581 break; 1582 case IPTOS_ECN_ECT1: 1583 TCPSTAT_INC(tcps_ecn_ect1); 1584 break; 1585 } 1586 1587 /* Process a packet differently from RFC3168. */ 1588 cc_ecnpkt_handler(tp, th, iptos); 1589 1590 /* Congestion experienced. */ 1591 if (thflags & TH_ECE) { 1592 cc_cong_signal(tp, th, CC_ECN); 1593 } 1594 } 1595 1596 /* 1597 * Parse options on any incoming segment. 1598 */ 1599 tcp_dooptions(&to, (u_char *)(th + 1), 1600 (th->th_off << 2) - sizeof(struct tcphdr), 1601 (thflags & TH_SYN) ? TO_SYN : 0); 1602 1603 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1604 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1605 (to.to_flags & TOF_SIGNATURE) == 0) { 1606 TCPSTAT_INC(tcps_sig_err_sigopt); 1607 /* XXX: should drop? */ 1608 } 1609 #endif 1610 /* 1611 * If echoed timestamp is later than the current time, 1612 * fall back to non RFC1323 RTT calculation. Normalize 1613 * timestamp if syncookies were used when this connection 1614 * was established. 1615 */ 1616 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1617 to.to_tsecr -= tp->ts_offset; 1618 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1619 to.to_tsecr = 0; 1620 } 1621 /* 1622 * If timestamps were negotiated during SYN/ACK they should 1623 * appear on every segment during this session and vice versa. 1624 */ 1625 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1626 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1627 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1628 "no action\n", s, __func__); 1629 free(s, M_TCPLOG); 1630 } 1631 } 1632 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1633 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1634 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1635 "no action\n", s, __func__); 1636 free(s, M_TCPLOG); 1637 } 1638 } 1639 1640 /* 1641 * Process options only when we get SYN/ACK back. The SYN case 1642 * for incoming connections is handled in tcp_syncache. 1643 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1644 * or <SYN,ACK>) segment itself is never scaled. 1645 * XXX this is traditional behavior, may need to be cleaned up. 1646 */ 1647 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1648 if ((to.to_flags & TOF_SCALE) && 1649 (tp->t_flags & TF_REQ_SCALE)) { 1650 tp->t_flags |= TF_RCVD_SCALE; 1651 tp->snd_scale = to.to_wscale; 1652 } 1653 /* 1654 * Initial send window. It will be updated with 1655 * the next incoming segment to the scaled value. 1656 */ 1657 tp->snd_wnd = th->th_win; 1658 if (to.to_flags & TOF_TS) { 1659 tp->t_flags |= TF_RCVD_TSTMP; 1660 tp->ts_recent = to.to_tsval; 1661 tp->ts_recent_age = tcp_ts_getticks(); 1662 } 1663 if (to.to_flags & TOF_MSS) 1664 tcp_mss(tp, to.to_mss); 1665 if ((tp->t_flags & TF_SACK_PERMIT) && 1666 (to.to_flags & TOF_SACKPERM) == 0) 1667 tp->t_flags &= ~TF_SACK_PERMIT; 1668 } 1669 1670 /* 1671 * Header prediction: check for the two common cases 1672 * of a uni-directional data xfer. If the packet has 1673 * no control flags, is in-sequence, the window didn't 1674 * change and we're not retransmitting, it's a 1675 * candidate. If the length is zero and the ack moved 1676 * forward, we're the sender side of the xfer. Just 1677 * free the data acked & wake any higher level process 1678 * that was blocked waiting for space. If the length 1679 * is non-zero and the ack didn't move, we're the 1680 * receiver side. If we're getting packets in-order 1681 * (the reassembly queue is empty), add the data to 1682 * the socket buffer and note that we need a delayed ack. 1683 * Make sure that the hidden state-flags are also off. 1684 * Since we check for TCPS_ESTABLISHED first, it can only 1685 * be TH_NEEDSYN. 1686 */ 1687 if (tp->t_state == TCPS_ESTABLISHED && 1688 th->th_seq == tp->rcv_nxt && 1689 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1690 tp->snd_nxt == tp->snd_max && 1691 tiwin && tiwin == tp->snd_wnd && 1692 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1693 LIST_EMPTY(&tp->t_segq) && 1694 ((to.to_flags & TOF_TS) == 0 || 1695 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1696 1697 /* 1698 * If last ACK falls within this segment's sequence numbers, 1699 * record the timestamp. 1700 * NOTE that the test is modified according to the latest 1701 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1702 */ 1703 if ((to.to_flags & TOF_TS) != 0 && 1704 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1705 tp->ts_recent_age = tcp_ts_getticks(); 1706 tp->ts_recent = to.to_tsval; 1707 } 1708 1709 if (tlen == 0) { 1710 if (SEQ_GT(th->th_ack, tp->snd_una) && 1711 SEQ_LEQ(th->th_ack, tp->snd_max) && 1712 !IN_RECOVERY(tp->t_flags) && 1713 (to.to_flags & TOF_SACK) == 0 && 1714 TAILQ_EMPTY(&tp->snd_holes)) { 1715 /* 1716 * This is a pure ack for outstanding data. 1717 */ 1718 if (ti_locked == TI_RLOCKED) 1719 INP_INFO_RUNLOCK(&V_tcbinfo); 1720 ti_locked = TI_UNLOCKED; 1721 1722 TCPSTAT_INC(tcps_predack); 1723 1724 /* 1725 * "bad retransmit" recovery. 1726 */ 1727 if (tp->t_rxtshift == 1 && 1728 tp->t_flags & TF_PREVVALID && 1729 (int)(ticks - tp->t_badrxtwin) < 0) { 1730 cc_cong_signal(tp, th, CC_RTO_ERR); 1731 } 1732 1733 /* 1734 * Recalculate the transmit timer / rtt. 1735 * 1736 * Some boxes send broken timestamp replies 1737 * during the SYN+ACK phase, ignore 1738 * timestamps of 0 or we could calculate a 1739 * huge RTT and blow up the retransmit timer. 1740 */ 1741 if ((to.to_flags & TOF_TS) != 0 && 1742 to.to_tsecr) { 1743 uint32_t t; 1744 1745 t = tcp_ts_getticks() - to.to_tsecr; 1746 if (!tp->t_rttlow || tp->t_rttlow > t) 1747 tp->t_rttlow = t; 1748 tcp_xmit_timer(tp, 1749 TCP_TS_TO_TICKS(t) + 1); 1750 } else if (tp->t_rtttime && 1751 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1752 if (!tp->t_rttlow || 1753 tp->t_rttlow > ticks - tp->t_rtttime) 1754 tp->t_rttlow = ticks - tp->t_rtttime; 1755 tcp_xmit_timer(tp, 1756 ticks - tp->t_rtttime); 1757 } 1758 acked = BYTES_THIS_ACK(tp, th); 1759 1760 #ifdef TCP_HHOOK 1761 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1762 hhook_run_tcp_est_in(tp, th, &to); 1763 #endif 1764 1765 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1766 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1767 sbdrop(&so->so_snd, acked); 1768 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1769 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1770 tp->snd_recover = th->th_ack - 1; 1771 1772 /* 1773 * Let the congestion control algorithm update 1774 * congestion control related information. This 1775 * typically means increasing the congestion 1776 * window. 1777 */ 1778 cc_ack_received(tp, th, nsegs, CC_ACK); 1779 1780 tp->snd_una = th->th_ack; 1781 /* 1782 * Pull snd_wl2 up to prevent seq wrap relative 1783 * to th_ack. 1784 */ 1785 tp->snd_wl2 = th->th_ack; 1786 tp->t_dupacks = 0; 1787 m_freem(m); 1788 1789 /* 1790 * If all outstanding data are acked, stop 1791 * retransmit timer, otherwise restart timer 1792 * using current (possibly backed-off) value. 1793 * If process is waiting for space, 1794 * wakeup/selwakeup/signal. If data 1795 * are ready to send, let tcp_output 1796 * decide between more output or persist. 1797 */ 1798 #ifdef TCPDEBUG 1799 if (so->so_options & SO_DEBUG) 1800 tcp_trace(TA_INPUT, ostate, tp, 1801 (void *)tcp_saveipgen, 1802 &tcp_savetcp, 0); 1803 #endif 1804 TCP_PROBE3(debug__input, tp, th, m); 1805 if (tp->snd_una == tp->snd_max) 1806 tcp_timer_activate(tp, TT_REXMT, 0); 1807 else if (!tcp_timer_active(tp, TT_PERSIST)) 1808 tcp_timer_activate(tp, TT_REXMT, 1809 tp->t_rxtcur); 1810 sowwakeup(so); 1811 if (sbavail(&so->so_snd)) 1812 (void) tp->t_fb->tfb_tcp_output(tp); 1813 goto check_delack; 1814 } 1815 } else if (th->th_ack == tp->snd_una && 1816 tlen <= sbspace(&so->so_rcv)) { 1817 int newsize = 0; /* automatic sockbuf scaling */ 1818 1819 /* 1820 * This is a pure, in-sequence data packet with 1821 * nothing on the reassembly queue and we have enough 1822 * buffer space to take it. 1823 */ 1824 if (ti_locked == TI_RLOCKED) 1825 INP_INFO_RUNLOCK(&V_tcbinfo); 1826 ti_locked = TI_UNLOCKED; 1827 1828 /* Clean receiver SACK report if present */ 1829 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1830 tcp_clean_sackreport(tp); 1831 TCPSTAT_INC(tcps_preddat); 1832 tp->rcv_nxt += tlen; 1833 /* 1834 * Pull snd_wl1 up to prevent seq wrap relative to 1835 * th_seq. 1836 */ 1837 tp->snd_wl1 = th->th_seq; 1838 /* 1839 * Pull rcv_up up to prevent seq wrap relative to 1840 * rcv_nxt. 1841 */ 1842 tp->rcv_up = tp->rcv_nxt; 1843 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1844 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1845 #ifdef TCPDEBUG 1846 if (so->so_options & SO_DEBUG) 1847 tcp_trace(TA_INPUT, ostate, tp, 1848 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1849 #endif 1850 TCP_PROBE3(debug__input, tp, th, m); 1851 1852 /* 1853 * Automatic sizing of receive socket buffer. Often the send 1854 * buffer size is not optimally adjusted to the actual network 1855 * conditions at hand (delay bandwidth product). Setting the 1856 * buffer size too small limits throughput on links with high 1857 * bandwidth and high delay (eg. trans-continental/oceanic links). 1858 * 1859 * On the receive side the socket buffer memory is only rarely 1860 * used to any significant extent. This allows us to be much 1861 * more aggressive in scaling the receive socket buffer. For 1862 * the case that the buffer space is actually used to a large 1863 * extent and we run out of kernel memory we can simply drop 1864 * the new segments; TCP on the sender will just retransmit it 1865 * later. Setting the buffer size too big may only consume too 1866 * much kernel memory if the application doesn't read() from 1867 * the socket or packet loss or reordering makes use of the 1868 * reassembly queue. 1869 * 1870 * The criteria to step up the receive buffer one notch are: 1871 * 1. Application has not set receive buffer size with 1872 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1873 * 2. the number of bytes received during the time it takes 1874 * one timestamp to be reflected back to us (the RTT); 1875 * 3. received bytes per RTT is within seven eighth of the 1876 * current socket buffer size; 1877 * 4. receive buffer size has not hit maximal automatic size; 1878 * 1879 * This algorithm does one step per RTT at most and only if 1880 * we receive a bulk stream w/o packet losses or reorderings. 1881 * Shrinking the buffer during idle times is not necessary as 1882 * it doesn't consume any memory when idle. 1883 * 1884 * TODO: Only step up if the application is actually serving 1885 * the buffer to better manage the socket buffer resources. 1886 */ 1887 if (V_tcp_do_autorcvbuf && 1888 (to.to_flags & TOF_TS) && 1889 to.to_tsecr && 1890 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1891 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1892 to.to_tsecr - tp->rfbuf_ts < hz) { 1893 if (tp->rfbuf_cnt > 1894 (so->so_rcv.sb_hiwat / 8 * 7) && 1895 so->so_rcv.sb_hiwat < 1896 V_tcp_autorcvbuf_max) { 1897 newsize = 1898 min(so->so_rcv.sb_hiwat + 1899 V_tcp_autorcvbuf_inc, 1900 V_tcp_autorcvbuf_max); 1901 } 1902 /* Start over with next RTT. */ 1903 tp->rfbuf_ts = 0; 1904 tp->rfbuf_cnt = 0; 1905 } else 1906 tp->rfbuf_cnt += tlen; /* add up */ 1907 } 1908 1909 /* Add data to socket buffer. */ 1910 SOCKBUF_LOCK(&so->so_rcv); 1911 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1912 m_freem(m); 1913 } else { 1914 /* 1915 * Set new socket buffer size. 1916 * Give up when limit is reached. 1917 */ 1918 if (newsize) 1919 if (!sbreserve_locked(&so->so_rcv, 1920 newsize, so, NULL)) 1921 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1922 m_adj(m, drop_hdrlen); /* delayed header drop */ 1923 sbappendstream_locked(&so->so_rcv, m, 0); 1924 } 1925 /* NB: sorwakeup_locked() does an implicit unlock. */ 1926 sorwakeup_locked(so); 1927 if (DELAY_ACK(tp, tlen)) { 1928 tp->t_flags |= TF_DELACK; 1929 } else { 1930 tp->t_flags |= TF_ACKNOW; 1931 tp->t_fb->tfb_tcp_output(tp); 1932 } 1933 goto check_delack; 1934 } 1935 } 1936 1937 /* 1938 * Calculate amount of space in receive window, 1939 * and then do TCP input processing. 1940 * Receive window is amount of space in rcv queue, 1941 * but not less than advertised window. 1942 */ 1943 win = sbspace(&so->so_rcv); 1944 if (win < 0) 1945 win = 0; 1946 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1947 1948 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1949 tp->rfbuf_ts = 0; 1950 tp->rfbuf_cnt = 0; 1951 1952 switch (tp->t_state) { 1953 1954 /* 1955 * If the state is SYN_RECEIVED: 1956 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1957 */ 1958 case TCPS_SYN_RECEIVED: 1959 if ((thflags & TH_ACK) && 1960 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1961 SEQ_GT(th->th_ack, tp->snd_max))) { 1962 rstreason = BANDLIM_RST_OPENPORT; 1963 goto dropwithreset; 1964 } 1965 #ifdef TCP_RFC7413 1966 if (IS_FASTOPEN(tp->t_flags)) { 1967 /* 1968 * When a TFO connection is in SYN_RECEIVED, the 1969 * only valid packets are the initial SYN, a 1970 * retransmit/copy of the initial SYN (possibly with 1971 * a subset of the original data), a valid ACK, a 1972 * FIN, or a RST. 1973 */ 1974 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1975 rstreason = BANDLIM_RST_OPENPORT; 1976 goto dropwithreset; 1977 } else if (thflags & TH_SYN) { 1978 /* non-initial SYN is ignored */ 1979 if ((tcp_timer_active(tp, TT_DELACK) || 1980 tcp_timer_active(tp, TT_REXMT))) 1981 goto drop; 1982 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1983 goto drop; 1984 } 1985 } 1986 #endif 1987 break; 1988 1989 /* 1990 * If the state is SYN_SENT: 1991 * if seg contains an ACK, but not for our SYN, drop the input. 1992 * if seg contains a RST, then drop the connection. 1993 * if seg does not contain SYN, then drop it. 1994 * Otherwise this is an acceptable SYN segment 1995 * initialize tp->rcv_nxt and tp->irs 1996 * if seg contains ack then advance tp->snd_una 1997 * if seg contains an ECE and ECN support is enabled, the stream 1998 * is ECN capable. 1999 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2000 * arrange for segment to be acked (eventually) 2001 * continue processing rest of data/controls, beginning with URG 2002 */ 2003 case TCPS_SYN_SENT: 2004 if ((thflags & TH_ACK) && 2005 (SEQ_LEQ(th->th_ack, tp->iss) || 2006 SEQ_GT(th->th_ack, tp->snd_max))) { 2007 rstreason = BANDLIM_UNLIMITED; 2008 goto dropwithreset; 2009 } 2010 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2011 TCP_PROBE5(connect__refused, NULL, tp, 2012 m, tp, th); 2013 tp = tcp_drop(tp, ECONNREFUSED); 2014 } 2015 if (thflags & TH_RST) 2016 goto drop; 2017 if (!(thflags & TH_SYN)) 2018 goto drop; 2019 2020 tp->irs = th->th_seq; 2021 tcp_rcvseqinit(tp); 2022 if (thflags & TH_ACK) { 2023 TCPSTAT_INC(tcps_connects); 2024 soisconnected(so); 2025 #ifdef MAC 2026 mac_socketpeer_set_from_mbuf(m, so); 2027 #endif 2028 /* Do window scaling on this connection? */ 2029 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2030 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2031 tp->rcv_scale = tp->request_r_scale; 2032 } 2033 tp->rcv_adv += min(tp->rcv_wnd, 2034 TCP_MAXWIN << tp->rcv_scale); 2035 tp->snd_una++; /* SYN is acked */ 2036 /* 2037 * If there's data, delay ACK; if there's also a FIN 2038 * ACKNOW will be turned on later. 2039 */ 2040 if (DELAY_ACK(tp, tlen) && tlen != 0) 2041 tcp_timer_activate(tp, TT_DELACK, 2042 tcp_delacktime); 2043 else 2044 tp->t_flags |= TF_ACKNOW; 2045 2046 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2047 tp->t_flags |= TF_ECN_PERMIT; 2048 TCPSTAT_INC(tcps_ecn_shs); 2049 } 2050 2051 /* 2052 * Received <SYN,ACK> in SYN_SENT[*] state. 2053 * Transitions: 2054 * SYN_SENT --> ESTABLISHED 2055 * SYN_SENT* --> FIN_WAIT_1 2056 */ 2057 tp->t_starttime = ticks; 2058 if (tp->t_flags & TF_NEEDFIN) { 2059 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2060 tp->t_flags &= ~TF_NEEDFIN; 2061 thflags &= ~TH_SYN; 2062 } else { 2063 tcp_state_change(tp, TCPS_ESTABLISHED); 2064 TCP_PROBE5(connect__established, NULL, tp, 2065 m, tp, th); 2066 cc_conn_init(tp); 2067 tcp_timer_activate(tp, TT_KEEP, 2068 TP_KEEPIDLE(tp)); 2069 } 2070 } else { 2071 /* 2072 * Received initial SYN in SYN-SENT[*] state => 2073 * simultaneous open. 2074 * If it succeeds, connection is * half-synchronized. 2075 * Otherwise, do 3-way handshake: 2076 * SYN-SENT -> SYN-RECEIVED 2077 * SYN-SENT* -> SYN-RECEIVED* 2078 */ 2079 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2080 tcp_timer_activate(tp, TT_REXMT, 0); 2081 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2082 } 2083 2084 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2085 "ti_locked %d", __func__, ti_locked)); 2086 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2087 INP_WLOCK_ASSERT(tp->t_inpcb); 2088 2089 /* 2090 * Advance th->th_seq to correspond to first data byte. 2091 * If data, trim to stay within window, 2092 * dropping FIN if necessary. 2093 */ 2094 th->th_seq++; 2095 if (tlen > tp->rcv_wnd) { 2096 todrop = tlen - tp->rcv_wnd; 2097 m_adj(m, -todrop); 2098 tlen = tp->rcv_wnd; 2099 thflags &= ~TH_FIN; 2100 TCPSTAT_INC(tcps_rcvpackafterwin); 2101 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2102 } 2103 tp->snd_wl1 = th->th_seq - 1; 2104 tp->rcv_up = th->th_seq; 2105 /* 2106 * Client side of transaction: already sent SYN and data. 2107 * If the remote host used T/TCP to validate the SYN, 2108 * our data will be ACK'd; if so, enter normal data segment 2109 * processing in the middle of step 5, ack processing. 2110 * Otherwise, goto step 6. 2111 */ 2112 if (thflags & TH_ACK) 2113 goto process_ACK; 2114 2115 goto step6; 2116 2117 /* 2118 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2119 * do normal processing. 2120 * 2121 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2122 */ 2123 case TCPS_LAST_ACK: 2124 case TCPS_CLOSING: 2125 break; /* continue normal processing */ 2126 } 2127 2128 /* 2129 * States other than LISTEN or SYN_SENT. 2130 * First check the RST flag and sequence number since reset segments 2131 * are exempt from the timestamp and connection count tests. This 2132 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2133 * below which allowed reset segments in half the sequence space 2134 * to fall though and be processed (which gives forged reset 2135 * segments with a random sequence number a 50 percent chance of 2136 * killing a connection). 2137 * Then check timestamp, if present. 2138 * Then check the connection count, if present. 2139 * Then check that at least some bytes of segment are within 2140 * receive window. If segment begins before rcv_nxt, 2141 * drop leading data (and SYN); if nothing left, just ack. 2142 */ 2143 if (thflags & TH_RST) { 2144 /* 2145 * RFC5961 Section 3.2 2146 * 2147 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2148 * - If RST is in window, we send challenge ACK. 2149 * 2150 * Note: to take into account delayed ACKs, we should 2151 * test against last_ack_sent instead of rcv_nxt. 2152 * Note 2: we handle special case of closed window, not 2153 * covered by the RFC. 2154 */ 2155 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2156 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2157 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2158 2159 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2160 KASSERT(ti_locked == TI_RLOCKED, 2161 ("%s: TH_RST ti_locked %d, th %p tp %p", 2162 __func__, ti_locked, th, tp)); 2163 KASSERT(tp->t_state != TCPS_SYN_SENT, 2164 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2165 __func__, th, tp)); 2166 2167 if (V_tcp_insecure_rst || 2168 tp->last_ack_sent == th->th_seq) { 2169 TCPSTAT_INC(tcps_drops); 2170 /* Drop the connection. */ 2171 switch (tp->t_state) { 2172 case TCPS_SYN_RECEIVED: 2173 so->so_error = ECONNREFUSED; 2174 goto close; 2175 case TCPS_ESTABLISHED: 2176 case TCPS_FIN_WAIT_1: 2177 case TCPS_FIN_WAIT_2: 2178 case TCPS_CLOSE_WAIT: 2179 case TCPS_CLOSING: 2180 case TCPS_LAST_ACK: 2181 so->so_error = ECONNRESET; 2182 close: 2183 /* FALLTHROUGH */ 2184 default: 2185 tp = tcp_close(tp); 2186 } 2187 } else { 2188 TCPSTAT_INC(tcps_badrst); 2189 /* Send challenge ACK. */ 2190 tcp_respond(tp, mtod(m, void *), th, m, 2191 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2192 tp->last_ack_sent = tp->rcv_nxt; 2193 m = NULL; 2194 } 2195 } 2196 goto drop; 2197 } 2198 2199 /* 2200 * RFC5961 Section 4.2 2201 * Send challenge ACK for any SYN in synchronized state. 2202 */ 2203 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2204 tp->t_state != TCPS_SYN_RECEIVED) { 2205 KASSERT(ti_locked == TI_RLOCKED, 2206 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2207 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2208 2209 TCPSTAT_INC(tcps_badsyn); 2210 if (V_tcp_insecure_syn && 2211 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2212 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2213 tp = tcp_drop(tp, ECONNRESET); 2214 rstreason = BANDLIM_UNLIMITED; 2215 } else { 2216 /* Send challenge ACK. */ 2217 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2218 tp->snd_nxt, TH_ACK); 2219 tp->last_ack_sent = tp->rcv_nxt; 2220 m = NULL; 2221 } 2222 goto drop; 2223 } 2224 2225 /* 2226 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2227 * and it's less than ts_recent, drop it. 2228 */ 2229 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2230 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2231 2232 /* Check to see if ts_recent is over 24 days old. */ 2233 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2234 /* 2235 * Invalidate ts_recent. If this segment updates 2236 * ts_recent, the age will be reset later and ts_recent 2237 * will get a valid value. If it does not, setting 2238 * ts_recent to zero will at least satisfy the 2239 * requirement that zero be placed in the timestamp 2240 * echo reply when ts_recent isn't valid. The 2241 * age isn't reset until we get a valid ts_recent 2242 * because we don't want out-of-order segments to be 2243 * dropped when ts_recent is old. 2244 */ 2245 tp->ts_recent = 0; 2246 } else { 2247 TCPSTAT_INC(tcps_rcvduppack); 2248 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2249 TCPSTAT_INC(tcps_pawsdrop); 2250 if (tlen) 2251 goto dropafterack; 2252 goto drop; 2253 } 2254 } 2255 2256 /* 2257 * In the SYN-RECEIVED state, validate that the packet belongs to 2258 * this connection before trimming the data to fit the receive 2259 * window. Check the sequence number versus IRS since we know 2260 * the sequence numbers haven't wrapped. This is a partial fix 2261 * for the "LAND" DoS attack. 2262 */ 2263 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2264 rstreason = BANDLIM_RST_OPENPORT; 2265 goto dropwithreset; 2266 } 2267 2268 todrop = tp->rcv_nxt - th->th_seq; 2269 if (todrop > 0) { 2270 if (thflags & TH_SYN) { 2271 thflags &= ~TH_SYN; 2272 th->th_seq++; 2273 if (th->th_urp > 1) 2274 th->th_urp--; 2275 else 2276 thflags &= ~TH_URG; 2277 todrop--; 2278 } 2279 /* 2280 * Following if statement from Stevens, vol. 2, p. 960. 2281 */ 2282 if (todrop > tlen 2283 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2284 /* 2285 * Any valid FIN must be to the left of the window. 2286 * At this point the FIN must be a duplicate or out 2287 * of sequence; drop it. 2288 */ 2289 thflags &= ~TH_FIN; 2290 2291 /* 2292 * Send an ACK to resynchronize and drop any data. 2293 * But keep on processing for RST or ACK. 2294 */ 2295 tp->t_flags |= TF_ACKNOW; 2296 todrop = tlen; 2297 TCPSTAT_INC(tcps_rcvduppack); 2298 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2299 } else { 2300 TCPSTAT_INC(tcps_rcvpartduppack); 2301 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2302 } 2303 drop_hdrlen += todrop; /* drop from the top afterwards */ 2304 th->th_seq += todrop; 2305 tlen -= todrop; 2306 if (th->th_urp > todrop) 2307 th->th_urp -= todrop; 2308 else { 2309 thflags &= ~TH_URG; 2310 th->th_urp = 0; 2311 } 2312 } 2313 2314 /* 2315 * If new data are received on a connection after the 2316 * user processes are gone, then RST the other end. 2317 */ 2318 if ((so->so_state & SS_NOFDREF) && 2319 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2320 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2321 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2322 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2323 2324 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2325 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2326 "after socket was closed, " 2327 "sending RST and removing tcpcb\n", 2328 s, __func__, tcpstates[tp->t_state], tlen); 2329 free(s, M_TCPLOG); 2330 } 2331 tp = tcp_close(tp); 2332 TCPSTAT_INC(tcps_rcvafterclose); 2333 rstreason = BANDLIM_UNLIMITED; 2334 goto dropwithreset; 2335 } 2336 2337 /* 2338 * If segment ends after window, drop trailing data 2339 * (and PUSH and FIN); if nothing left, just ACK. 2340 */ 2341 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2342 if (todrop > 0) { 2343 TCPSTAT_INC(tcps_rcvpackafterwin); 2344 if (todrop >= tlen) { 2345 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2346 /* 2347 * If window is closed can only take segments at 2348 * window edge, and have to drop data and PUSH from 2349 * incoming segments. Continue processing, but 2350 * remember to ack. Otherwise, drop segment 2351 * and ack. 2352 */ 2353 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2354 tp->t_flags |= TF_ACKNOW; 2355 TCPSTAT_INC(tcps_rcvwinprobe); 2356 } else 2357 goto dropafterack; 2358 } else 2359 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2360 m_adj(m, -todrop); 2361 tlen -= todrop; 2362 thflags &= ~(TH_PUSH|TH_FIN); 2363 } 2364 2365 /* 2366 * If last ACK falls within this segment's sequence numbers, 2367 * record its timestamp. 2368 * NOTE: 2369 * 1) That the test incorporates suggestions from the latest 2370 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2371 * 2) That updating only on newer timestamps interferes with 2372 * our earlier PAWS tests, so this check should be solely 2373 * predicated on the sequence space of this segment. 2374 * 3) That we modify the segment boundary check to be 2375 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2376 * instead of RFC1323's 2377 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2378 * This modified check allows us to overcome RFC1323's 2379 * limitations as described in Stevens TCP/IP Illustrated 2380 * Vol. 2 p.869. In such cases, we can still calculate the 2381 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2382 */ 2383 if ((to.to_flags & TOF_TS) != 0 && 2384 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2385 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2386 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2387 tp->ts_recent_age = tcp_ts_getticks(); 2388 tp->ts_recent = to.to_tsval; 2389 } 2390 2391 /* 2392 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2393 * flag is on (half-synchronized state), then queue data for 2394 * later processing; else drop segment and return. 2395 */ 2396 if ((thflags & TH_ACK) == 0) { 2397 if (tp->t_state == TCPS_SYN_RECEIVED || 2398 (tp->t_flags & TF_NEEDSYN)) { 2399 #ifdef TCP_RFC7413 2400 if (tp->t_state == TCPS_SYN_RECEIVED && 2401 IS_FASTOPEN(tp->t_flags)) { 2402 tp->snd_wnd = tiwin; 2403 cc_conn_init(tp); 2404 } 2405 #endif 2406 goto step6; 2407 } else if (tp->t_flags & TF_ACKNOW) 2408 goto dropafterack; 2409 else 2410 goto drop; 2411 } 2412 2413 /* 2414 * Ack processing. 2415 */ 2416 switch (tp->t_state) { 2417 2418 /* 2419 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2420 * ESTABLISHED state and continue processing. 2421 * The ACK was checked above. 2422 */ 2423 case TCPS_SYN_RECEIVED: 2424 2425 TCPSTAT_INC(tcps_connects); 2426 soisconnected(so); 2427 /* Do window scaling? */ 2428 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2429 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2430 tp->rcv_scale = tp->request_r_scale; 2431 tp->snd_wnd = tiwin; 2432 } 2433 /* 2434 * Make transitions: 2435 * SYN-RECEIVED -> ESTABLISHED 2436 * SYN-RECEIVED* -> FIN-WAIT-1 2437 */ 2438 tp->t_starttime = ticks; 2439 if (tp->t_flags & TF_NEEDFIN) { 2440 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2441 tp->t_flags &= ~TF_NEEDFIN; 2442 } else { 2443 tcp_state_change(tp, TCPS_ESTABLISHED); 2444 TCP_PROBE5(accept__established, NULL, tp, 2445 m, tp, th); 2446 #ifdef TCP_RFC7413 2447 if (tp->t_tfo_pending) { 2448 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2449 tp->t_tfo_pending = NULL; 2450 2451 /* 2452 * Account for the ACK of our SYN prior to 2453 * regular ACK processing below. 2454 */ 2455 tp->snd_una++; 2456 } 2457 /* 2458 * TFO connections call cc_conn_init() during SYN 2459 * processing. Calling it again here for such 2460 * connections is not harmless as it would undo the 2461 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2462 * is retransmitted. 2463 */ 2464 if (!IS_FASTOPEN(tp->t_flags)) 2465 #endif 2466 cc_conn_init(tp); 2467 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2468 } 2469 /* 2470 * If segment contains data or ACK, will call tcp_reass() 2471 * later; if not, do so now to pass queued data to user. 2472 */ 2473 if (tlen == 0 && (thflags & TH_FIN) == 0) 2474 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2475 (struct mbuf *)0); 2476 tp->snd_wl1 = th->th_seq - 1; 2477 /* FALLTHROUGH */ 2478 2479 /* 2480 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2481 * ACKs. If the ack is in the range 2482 * tp->snd_una < th->th_ack <= tp->snd_max 2483 * then advance tp->snd_una to th->th_ack and drop 2484 * data from the retransmission queue. If this ACK reflects 2485 * more up to date window information we update our window information. 2486 */ 2487 case TCPS_ESTABLISHED: 2488 case TCPS_FIN_WAIT_1: 2489 case TCPS_FIN_WAIT_2: 2490 case TCPS_CLOSE_WAIT: 2491 case TCPS_CLOSING: 2492 case TCPS_LAST_ACK: 2493 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2494 TCPSTAT_INC(tcps_rcvacktoomuch); 2495 goto dropafterack; 2496 } 2497 if ((tp->t_flags & TF_SACK_PERMIT) && 2498 ((to.to_flags & TOF_SACK) || 2499 !TAILQ_EMPTY(&tp->snd_holes))) 2500 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2501 else 2502 /* 2503 * Reset the value so that previous (valid) value 2504 * from the last ack with SACK doesn't get used. 2505 */ 2506 tp->sackhint.sacked_bytes = 0; 2507 2508 #ifdef TCP_HHOOK 2509 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2510 hhook_run_tcp_est_in(tp, th, &to); 2511 #endif 2512 2513 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2514 u_int maxseg; 2515 2516 maxseg = tcp_maxseg(tp); 2517 if (tlen == 0 && 2518 (tiwin == tp->snd_wnd || 2519 (tp->t_flags & TF_SACK_PERMIT))) { 2520 /* 2521 * If this is the first time we've seen a 2522 * FIN from the remote, this is not a 2523 * duplicate and it needs to be processed 2524 * normally. This happens during a 2525 * simultaneous close. 2526 */ 2527 if ((thflags & TH_FIN) && 2528 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2529 tp->t_dupacks = 0; 2530 break; 2531 } 2532 TCPSTAT_INC(tcps_rcvdupack); 2533 /* 2534 * If we have outstanding data (other than 2535 * a window probe), this is a completely 2536 * duplicate ack (ie, window info didn't 2537 * change and FIN isn't set), 2538 * the ack is the biggest we've 2539 * seen and we've seen exactly our rexmt 2540 * threshold of them, assume a packet 2541 * has been dropped and retransmit it. 2542 * Kludge snd_nxt & the congestion 2543 * window so we send only this one 2544 * packet. 2545 * 2546 * We know we're losing at the current 2547 * window size so do congestion avoidance 2548 * (set ssthresh to half the current window 2549 * and pull our congestion window back to 2550 * the new ssthresh). 2551 * 2552 * Dup acks mean that packets have left the 2553 * network (they're now cached at the receiver) 2554 * so bump cwnd by the amount in the receiver 2555 * to keep a constant cwnd packets in the 2556 * network. 2557 * 2558 * When using TCP ECN, notify the peer that 2559 * we reduced the cwnd. 2560 */ 2561 /* 2562 * Following 2 kinds of acks should not affect 2563 * dupack counting: 2564 * 1) Old acks 2565 * 2) Acks with SACK but without any new SACK 2566 * information in them. These could result from 2567 * any anomaly in the network like a switch 2568 * duplicating packets or a possible DoS attack. 2569 */ 2570 if (th->th_ack != tp->snd_una || 2571 ((tp->t_flags & TF_SACK_PERMIT) && 2572 !sack_changed)) 2573 break; 2574 else if (!tcp_timer_active(tp, TT_REXMT)) 2575 tp->t_dupacks = 0; 2576 else if (++tp->t_dupacks > tcprexmtthresh || 2577 IN_FASTRECOVERY(tp->t_flags)) { 2578 cc_ack_received(tp, th, nsegs, 2579 CC_DUPACK); 2580 if ((tp->t_flags & TF_SACK_PERMIT) && 2581 IN_FASTRECOVERY(tp->t_flags)) { 2582 int awnd; 2583 2584 /* 2585 * Compute the amount of data in flight first. 2586 * We can inject new data into the pipe iff 2587 * we have less than 1/2 the original window's 2588 * worth of data in flight. 2589 */ 2590 if (V_tcp_do_rfc6675_pipe) 2591 awnd = tcp_compute_pipe(tp); 2592 else 2593 awnd = (tp->snd_nxt - tp->snd_fack) + 2594 tp->sackhint.sack_bytes_rexmit; 2595 2596 if (awnd < tp->snd_ssthresh) { 2597 tp->snd_cwnd += maxseg; 2598 /* 2599 * RFC5681 Section 3.2 talks about cwnd 2600 * inflation on additional dupacks and 2601 * deflation on recovering from loss. 2602 * 2603 * We keep cwnd into check so that 2604 * we don't have to 'deflate' it when we 2605 * get out of recovery. 2606 */ 2607 if (tp->snd_cwnd > tp->snd_ssthresh) 2608 tp->snd_cwnd = tp->snd_ssthresh; 2609 } 2610 } else 2611 tp->snd_cwnd += maxseg; 2612 (void) tp->t_fb->tfb_tcp_output(tp); 2613 goto drop; 2614 } else if (tp->t_dupacks == tcprexmtthresh) { 2615 tcp_seq onxt = tp->snd_nxt; 2616 2617 /* 2618 * If we're doing sack, check to 2619 * see if we're already in sack 2620 * recovery. If we're not doing sack, 2621 * check to see if we're in newreno 2622 * recovery. 2623 */ 2624 if (tp->t_flags & TF_SACK_PERMIT) { 2625 if (IN_FASTRECOVERY(tp->t_flags)) { 2626 tp->t_dupacks = 0; 2627 break; 2628 } 2629 } else { 2630 if (SEQ_LEQ(th->th_ack, 2631 tp->snd_recover)) { 2632 tp->t_dupacks = 0; 2633 break; 2634 } 2635 } 2636 /* Congestion signal before ack. */ 2637 cc_cong_signal(tp, th, CC_NDUPACK); 2638 cc_ack_received(tp, th, nsegs, 2639 CC_DUPACK); 2640 tcp_timer_activate(tp, TT_REXMT, 0); 2641 tp->t_rtttime = 0; 2642 if (tp->t_flags & TF_SACK_PERMIT) { 2643 TCPSTAT_INC( 2644 tcps_sack_recovery_episode); 2645 tp->sack_newdata = tp->snd_nxt; 2646 if (CC_ALGO(tp)->cong_signal == NULL) 2647 tp->snd_cwnd = maxseg; 2648 (void) tp->t_fb->tfb_tcp_output(tp); 2649 goto drop; 2650 } 2651 tp->snd_nxt = th->th_ack; 2652 if (CC_ALGO(tp)->cong_signal == NULL) 2653 tp->snd_cwnd = maxseg; 2654 (void) tp->t_fb->tfb_tcp_output(tp); 2655 KASSERT(tp->snd_limited <= 2, 2656 ("%s: tp->snd_limited too big", 2657 __func__)); 2658 if (CC_ALGO(tp)->cong_signal == NULL) 2659 tp->snd_cwnd = tp->snd_ssthresh + 2660 maxseg * 2661 (tp->t_dupacks - tp->snd_limited); 2662 if (SEQ_GT(onxt, tp->snd_nxt)) 2663 tp->snd_nxt = onxt; 2664 goto drop; 2665 } else if (V_tcp_do_rfc3042) { 2666 /* 2667 * Process first and second duplicate 2668 * ACKs. Each indicates a segment 2669 * leaving the network, creating room 2670 * for more. Make sure we can send a 2671 * packet on reception of each duplicate 2672 * ACK by increasing snd_cwnd by one 2673 * segment. Restore the original 2674 * snd_cwnd after packet transmission. 2675 */ 2676 cc_ack_received(tp, th, nsegs, 2677 CC_DUPACK); 2678 uint32_t oldcwnd = tp->snd_cwnd; 2679 tcp_seq oldsndmax = tp->snd_max; 2680 u_int sent; 2681 int avail; 2682 2683 KASSERT(tp->t_dupacks == 1 || 2684 tp->t_dupacks == 2, 2685 ("%s: dupacks not 1 or 2", 2686 __func__)); 2687 if (tp->t_dupacks == 1) 2688 tp->snd_limited = 0; 2689 tp->snd_cwnd = 2690 (tp->snd_nxt - tp->snd_una) + 2691 (tp->t_dupacks - tp->snd_limited) * 2692 maxseg; 2693 /* 2694 * Only call tcp_output when there 2695 * is new data available to be sent. 2696 * Otherwise we would send pure ACKs. 2697 */ 2698 SOCKBUF_LOCK(&so->so_snd); 2699 avail = sbavail(&so->so_snd) - 2700 (tp->snd_nxt - tp->snd_una); 2701 SOCKBUF_UNLOCK(&so->so_snd); 2702 if (avail > 0) 2703 (void) tp->t_fb->tfb_tcp_output(tp); 2704 sent = tp->snd_max - oldsndmax; 2705 if (sent > maxseg) { 2706 KASSERT((tp->t_dupacks == 2 && 2707 tp->snd_limited == 0) || 2708 (sent == maxseg + 1 && 2709 tp->t_flags & TF_SENTFIN), 2710 ("%s: sent too much", 2711 __func__)); 2712 tp->snd_limited = 2; 2713 } else if (sent > 0) 2714 ++tp->snd_limited; 2715 tp->snd_cwnd = oldcwnd; 2716 goto drop; 2717 } 2718 } 2719 break; 2720 } else { 2721 /* 2722 * This ack is advancing the left edge, reset the 2723 * counter. 2724 */ 2725 tp->t_dupacks = 0; 2726 /* 2727 * If this ack also has new SACK info, increment the 2728 * counter as per rfc6675. 2729 */ 2730 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2731 tp->t_dupacks++; 2732 } 2733 2734 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2735 ("%s: th_ack <= snd_una", __func__)); 2736 2737 /* 2738 * If the congestion window was inflated to account 2739 * for the other side's cached packets, retract it. 2740 */ 2741 if (IN_FASTRECOVERY(tp->t_flags)) { 2742 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2743 if (tp->t_flags & TF_SACK_PERMIT) 2744 tcp_sack_partialack(tp, th); 2745 else 2746 tcp_newreno_partial_ack(tp, th); 2747 } else 2748 cc_post_recovery(tp, th); 2749 } 2750 /* 2751 * If we reach this point, ACK is not a duplicate, 2752 * i.e., it ACKs something we sent. 2753 */ 2754 if (tp->t_flags & TF_NEEDSYN) { 2755 /* 2756 * T/TCP: Connection was half-synchronized, and our 2757 * SYN has been ACK'd (so connection is now fully 2758 * synchronized). Go to non-starred state, 2759 * increment snd_una for ACK of SYN, and check if 2760 * we can do window scaling. 2761 */ 2762 tp->t_flags &= ~TF_NEEDSYN; 2763 tp->snd_una++; 2764 /* Do window scaling? */ 2765 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2766 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2767 tp->rcv_scale = tp->request_r_scale; 2768 /* Send window already scaled. */ 2769 } 2770 } 2771 2772 process_ACK: 2773 INP_WLOCK_ASSERT(tp->t_inpcb); 2774 2775 acked = BYTES_THIS_ACK(tp, th); 2776 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2777 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2778 tp->snd_una, th->th_ack, tp, m)); 2779 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2780 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2781 2782 /* 2783 * If we just performed our first retransmit, and the ACK 2784 * arrives within our recovery window, then it was a mistake 2785 * to do the retransmit in the first place. Recover our 2786 * original cwnd and ssthresh, and proceed to transmit where 2787 * we left off. 2788 */ 2789 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2790 (int)(ticks - tp->t_badrxtwin) < 0) 2791 cc_cong_signal(tp, th, CC_RTO_ERR); 2792 2793 /* 2794 * If we have a timestamp reply, update smoothed 2795 * round trip time. If no timestamp is present but 2796 * transmit timer is running and timed sequence 2797 * number was acked, update smoothed round trip time. 2798 * Since we now have an rtt measurement, cancel the 2799 * timer backoff (cf., Phil Karn's retransmit alg.). 2800 * Recompute the initial retransmit timer. 2801 * 2802 * Some boxes send broken timestamp replies 2803 * during the SYN+ACK phase, ignore 2804 * timestamps of 0 or we could calculate a 2805 * huge RTT and blow up the retransmit timer. 2806 */ 2807 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2808 uint32_t t; 2809 2810 t = tcp_ts_getticks() - to.to_tsecr; 2811 if (!tp->t_rttlow || tp->t_rttlow > t) 2812 tp->t_rttlow = t; 2813 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2814 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2815 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2816 tp->t_rttlow = ticks - tp->t_rtttime; 2817 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2818 } 2819 2820 /* 2821 * If all outstanding data is acked, stop retransmit 2822 * timer and remember to restart (more output or persist). 2823 * If there is more data to be acked, restart retransmit 2824 * timer, using current (possibly backed-off) value. 2825 */ 2826 if (th->th_ack == tp->snd_max) { 2827 tcp_timer_activate(tp, TT_REXMT, 0); 2828 needoutput = 1; 2829 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2830 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2831 2832 /* 2833 * If no data (only SYN) was ACK'd, 2834 * skip rest of ACK processing. 2835 */ 2836 if (acked == 0) 2837 goto step6; 2838 2839 /* 2840 * Let the congestion control algorithm update congestion 2841 * control related information. This typically means increasing 2842 * the congestion window. 2843 */ 2844 cc_ack_received(tp, th, nsegs, CC_ACK); 2845 2846 SOCKBUF_LOCK(&so->so_snd); 2847 if (acked > sbavail(&so->so_snd)) { 2848 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2849 tp->snd_wnd -= sbavail(&so->so_snd); 2850 else 2851 tp->snd_wnd = 0; 2852 mfree = sbcut_locked(&so->so_snd, 2853 (int)sbavail(&so->so_snd)); 2854 ourfinisacked = 1; 2855 } else { 2856 mfree = sbcut_locked(&so->so_snd, acked); 2857 if (tp->snd_wnd >= (uint32_t) acked) 2858 tp->snd_wnd -= acked; 2859 else 2860 tp->snd_wnd = 0; 2861 ourfinisacked = 0; 2862 } 2863 /* NB: sowwakeup_locked() does an implicit unlock. */ 2864 sowwakeup_locked(so); 2865 m_freem(mfree); 2866 /* Detect una wraparound. */ 2867 if (!IN_RECOVERY(tp->t_flags) && 2868 SEQ_GT(tp->snd_una, tp->snd_recover) && 2869 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2870 tp->snd_recover = th->th_ack - 1; 2871 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2872 if (IN_RECOVERY(tp->t_flags) && 2873 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2874 EXIT_RECOVERY(tp->t_flags); 2875 } 2876 tp->snd_una = th->th_ack; 2877 if (tp->t_flags & TF_SACK_PERMIT) { 2878 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2879 tp->snd_recover = tp->snd_una; 2880 } 2881 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2882 tp->snd_nxt = tp->snd_una; 2883 2884 switch (tp->t_state) { 2885 2886 /* 2887 * In FIN_WAIT_1 STATE in addition to the processing 2888 * for the ESTABLISHED state if our FIN is now acknowledged 2889 * then enter FIN_WAIT_2. 2890 */ 2891 case TCPS_FIN_WAIT_1: 2892 if (ourfinisacked) { 2893 /* 2894 * If we can't receive any more 2895 * data, then closing user can proceed. 2896 * Starting the timer is contrary to the 2897 * specification, but if we don't get a FIN 2898 * we'll hang forever. 2899 * 2900 * XXXjl: 2901 * we should release the tp also, and use a 2902 * compressed state. 2903 */ 2904 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2905 soisdisconnected(so); 2906 tcp_timer_activate(tp, TT_2MSL, 2907 (tcp_fast_finwait2_recycle ? 2908 tcp_finwait2_timeout : 2909 TP_MAXIDLE(tp))); 2910 } 2911 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2912 } 2913 break; 2914 2915 /* 2916 * In CLOSING STATE in addition to the processing for 2917 * the ESTABLISHED state if the ACK acknowledges our FIN 2918 * then enter the TIME-WAIT state, otherwise ignore 2919 * the segment. 2920 */ 2921 case TCPS_CLOSING: 2922 if (ourfinisacked) { 2923 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2924 tcp_twstart(tp); 2925 INP_INFO_RUNLOCK(&V_tcbinfo); 2926 m_freem(m); 2927 return; 2928 } 2929 break; 2930 2931 /* 2932 * In LAST_ACK, we may still be waiting for data to drain 2933 * and/or to be acked, as well as for the ack of our FIN. 2934 * If our FIN is now acknowledged, delete the TCB, 2935 * enter the closed state and return. 2936 */ 2937 case TCPS_LAST_ACK: 2938 if (ourfinisacked) { 2939 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2940 tp = tcp_close(tp); 2941 goto drop; 2942 } 2943 break; 2944 } 2945 } 2946 2947 step6: 2948 INP_WLOCK_ASSERT(tp->t_inpcb); 2949 2950 /* 2951 * Update window information. 2952 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2953 */ 2954 if ((thflags & TH_ACK) && 2955 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2956 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2957 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2958 /* keep track of pure window updates */ 2959 if (tlen == 0 && 2960 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2961 TCPSTAT_INC(tcps_rcvwinupd); 2962 tp->snd_wnd = tiwin; 2963 tp->snd_wl1 = th->th_seq; 2964 tp->snd_wl2 = th->th_ack; 2965 if (tp->snd_wnd > tp->max_sndwnd) 2966 tp->max_sndwnd = tp->snd_wnd; 2967 needoutput = 1; 2968 } 2969 2970 /* 2971 * Process segments with URG. 2972 */ 2973 if ((thflags & TH_URG) && th->th_urp && 2974 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2975 /* 2976 * This is a kludge, but if we receive and accept 2977 * random urgent pointers, we'll crash in 2978 * soreceive. It's hard to imagine someone 2979 * actually wanting to send this much urgent data. 2980 */ 2981 SOCKBUF_LOCK(&so->so_rcv); 2982 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2983 th->th_urp = 0; /* XXX */ 2984 thflags &= ~TH_URG; /* XXX */ 2985 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2986 goto dodata; /* XXX */ 2987 } 2988 /* 2989 * If this segment advances the known urgent pointer, 2990 * then mark the data stream. This should not happen 2991 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2992 * a FIN has been received from the remote side. 2993 * In these states we ignore the URG. 2994 * 2995 * According to RFC961 (Assigned Protocols), 2996 * the urgent pointer points to the last octet 2997 * of urgent data. We continue, however, 2998 * to consider it to indicate the first octet 2999 * of data past the urgent section as the original 3000 * spec states (in one of two places). 3001 */ 3002 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3003 tp->rcv_up = th->th_seq + th->th_urp; 3004 so->so_oobmark = sbavail(&so->so_rcv) + 3005 (tp->rcv_up - tp->rcv_nxt) - 1; 3006 if (so->so_oobmark == 0) 3007 so->so_rcv.sb_state |= SBS_RCVATMARK; 3008 sohasoutofband(so); 3009 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3010 } 3011 SOCKBUF_UNLOCK(&so->so_rcv); 3012 /* 3013 * Remove out of band data so doesn't get presented to user. 3014 * This can happen independent of advancing the URG pointer, 3015 * but if two URG's are pending at once, some out-of-band 3016 * data may creep in... ick. 3017 */ 3018 if (th->th_urp <= (uint32_t)tlen && 3019 !(so->so_options & SO_OOBINLINE)) { 3020 /* hdr drop is delayed */ 3021 tcp_pulloutofband(so, th, m, drop_hdrlen); 3022 } 3023 } else { 3024 /* 3025 * If no out of band data is expected, 3026 * pull receive urgent pointer along 3027 * with the receive window. 3028 */ 3029 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3030 tp->rcv_up = tp->rcv_nxt; 3031 } 3032 dodata: /* XXX */ 3033 INP_WLOCK_ASSERT(tp->t_inpcb); 3034 3035 /* 3036 * Process the segment text, merging it into the TCP sequencing queue, 3037 * and arranging for acknowledgment of receipt if necessary. 3038 * This process logically involves adjusting tp->rcv_wnd as data 3039 * is presented to the user (this happens in tcp_usrreq.c, 3040 * case PRU_RCVD). If a FIN has already been received on this 3041 * connection then we just ignore the text. 3042 */ 3043 #ifdef TCP_RFC7413 3044 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3045 IS_FASTOPEN(tp->t_flags)); 3046 #else 3047 #define tfo_syn (false) 3048 #endif 3049 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3050 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3051 tcp_seq save_start = th->th_seq; 3052 m_adj(m, drop_hdrlen); /* delayed header drop */ 3053 /* 3054 * Insert segment which includes th into TCP reassembly queue 3055 * with control block tp. Set thflags to whether reassembly now 3056 * includes a segment with FIN. This handles the common case 3057 * inline (segment is the next to be received on an established 3058 * connection, and the queue is empty), avoiding linkage into 3059 * and removal from the queue and repetition of various 3060 * conversions. 3061 * Set DELACK for segments received in order, but ack 3062 * immediately when segments are out of order (so 3063 * fast retransmit can work). 3064 */ 3065 if (th->th_seq == tp->rcv_nxt && 3066 LIST_EMPTY(&tp->t_segq) && 3067 (TCPS_HAVEESTABLISHED(tp->t_state) || 3068 tfo_syn)) { 3069 if (DELAY_ACK(tp, tlen) || tfo_syn) 3070 tp->t_flags |= TF_DELACK; 3071 else 3072 tp->t_flags |= TF_ACKNOW; 3073 tp->rcv_nxt += tlen; 3074 thflags = th->th_flags & TH_FIN; 3075 TCPSTAT_INC(tcps_rcvpack); 3076 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3077 SOCKBUF_LOCK(&so->so_rcv); 3078 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3079 m_freem(m); 3080 else 3081 sbappendstream_locked(&so->so_rcv, m, 0); 3082 /* NB: sorwakeup_locked() does an implicit unlock. */ 3083 sorwakeup_locked(so); 3084 } else { 3085 /* 3086 * XXX: Due to the header drop above "th" is 3087 * theoretically invalid by now. Fortunately 3088 * m_adj() doesn't actually frees any mbufs 3089 * when trimming from the head. 3090 */ 3091 thflags = tcp_reass(tp, th, &tlen, m); 3092 tp->t_flags |= TF_ACKNOW; 3093 } 3094 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3095 tcp_update_sack_list(tp, save_start, save_start + tlen); 3096 #if 0 3097 /* 3098 * Note the amount of data that peer has sent into 3099 * our window, in order to estimate the sender's 3100 * buffer size. 3101 * XXX: Unused. 3102 */ 3103 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3104 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3105 else 3106 len = so->so_rcv.sb_hiwat; 3107 #endif 3108 } else { 3109 m_freem(m); 3110 thflags &= ~TH_FIN; 3111 } 3112 3113 /* 3114 * If FIN is received ACK the FIN and let the user know 3115 * that the connection is closing. 3116 */ 3117 if (thflags & TH_FIN) { 3118 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3119 socantrcvmore(so); 3120 /* 3121 * If connection is half-synchronized 3122 * (ie NEEDSYN flag on) then delay ACK, 3123 * so it may be piggybacked when SYN is sent. 3124 * Otherwise, since we received a FIN then no 3125 * more input can be expected, send ACK now. 3126 */ 3127 if (tp->t_flags & TF_NEEDSYN) 3128 tp->t_flags |= TF_DELACK; 3129 else 3130 tp->t_flags |= TF_ACKNOW; 3131 tp->rcv_nxt++; 3132 } 3133 switch (tp->t_state) { 3134 3135 /* 3136 * In SYN_RECEIVED and ESTABLISHED STATES 3137 * enter the CLOSE_WAIT state. 3138 */ 3139 case TCPS_SYN_RECEIVED: 3140 tp->t_starttime = ticks; 3141 /* FALLTHROUGH */ 3142 case TCPS_ESTABLISHED: 3143 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3144 break; 3145 3146 /* 3147 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3148 * enter the CLOSING state. 3149 */ 3150 case TCPS_FIN_WAIT_1: 3151 tcp_state_change(tp, TCPS_CLOSING); 3152 break; 3153 3154 /* 3155 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3156 * starting the time-wait timer, turning off the other 3157 * standard timers. 3158 */ 3159 case TCPS_FIN_WAIT_2: 3160 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3161 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3162 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3163 ti_locked)); 3164 3165 tcp_twstart(tp); 3166 INP_INFO_RUNLOCK(&V_tcbinfo); 3167 return; 3168 } 3169 } 3170 if (ti_locked == TI_RLOCKED) 3171 INP_INFO_RUNLOCK(&V_tcbinfo); 3172 ti_locked = TI_UNLOCKED; 3173 3174 #ifdef TCPDEBUG 3175 if (so->so_options & SO_DEBUG) 3176 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3177 &tcp_savetcp, 0); 3178 #endif 3179 TCP_PROBE3(debug__input, tp, th, m); 3180 3181 /* 3182 * Return any desired output. 3183 */ 3184 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3185 (void) tp->t_fb->tfb_tcp_output(tp); 3186 3187 check_delack: 3188 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3189 __func__, ti_locked)); 3190 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3191 INP_WLOCK_ASSERT(tp->t_inpcb); 3192 3193 if (tp->t_flags & TF_DELACK) { 3194 tp->t_flags &= ~TF_DELACK; 3195 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3196 } 3197 INP_WUNLOCK(tp->t_inpcb); 3198 return; 3199 3200 dropafterack: 3201 /* 3202 * Generate an ACK dropping incoming segment if it occupies 3203 * sequence space, where the ACK reflects our state. 3204 * 3205 * We can now skip the test for the RST flag since all 3206 * paths to this code happen after packets containing 3207 * RST have been dropped. 3208 * 3209 * In the SYN-RECEIVED state, don't send an ACK unless the 3210 * segment we received passes the SYN-RECEIVED ACK test. 3211 * If it fails send a RST. This breaks the loop in the 3212 * "LAND" DoS attack, and also prevents an ACK storm 3213 * between two listening ports that have been sent forged 3214 * SYN segments, each with the source address of the other. 3215 */ 3216 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3217 (SEQ_GT(tp->snd_una, th->th_ack) || 3218 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3219 rstreason = BANDLIM_RST_OPENPORT; 3220 goto dropwithreset; 3221 } 3222 #ifdef TCPDEBUG 3223 if (so->so_options & SO_DEBUG) 3224 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3225 &tcp_savetcp, 0); 3226 #endif 3227 TCP_PROBE3(debug__input, tp, th, m); 3228 if (ti_locked == TI_RLOCKED) 3229 INP_INFO_RUNLOCK(&V_tcbinfo); 3230 ti_locked = TI_UNLOCKED; 3231 3232 tp->t_flags |= TF_ACKNOW; 3233 (void) tp->t_fb->tfb_tcp_output(tp); 3234 INP_WUNLOCK(tp->t_inpcb); 3235 m_freem(m); 3236 return; 3237 3238 dropwithreset: 3239 if (ti_locked == TI_RLOCKED) 3240 INP_INFO_RUNLOCK(&V_tcbinfo); 3241 ti_locked = TI_UNLOCKED; 3242 3243 if (tp != NULL) { 3244 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3245 INP_WUNLOCK(tp->t_inpcb); 3246 } else 3247 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3248 return; 3249 3250 drop: 3251 if (ti_locked == TI_RLOCKED) { 3252 INP_INFO_RUNLOCK(&V_tcbinfo); 3253 ti_locked = TI_UNLOCKED; 3254 } 3255 #ifdef INVARIANTS 3256 else 3257 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3258 #endif 3259 3260 /* 3261 * Drop space held by incoming segment and return. 3262 */ 3263 #ifdef TCPDEBUG 3264 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3265 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3266 &tcp_savetcp, 0); 3267 #endif 3268 TCP_PROBE3(debug__input, tp, th, m); 3269 if (tp != NULL) 3270 INP_WUNLOCK(tp->t_inpcb); 3271 m_freem(m); 3272 #ifndef TCP_RFC7413 3273 #undef tfo_syn 3274 #endif 3275 } 3276 3277 /* 3278 * Issue RST and make ACK acceptable to originator of segment. 3279 * The mbuf must still include the original packet header. 3280 * tp may be NULL. 3281 */ 3282 void 3283 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3284 int tlen, int rstreason) 3285 { 3286 #ifdef INET 3287 struct ip *ip; 3288 #endif 3289 #ifdef INET6 3290 struct ip6_hdr *ip6; 3291 #endif 3292 3293 if (tp != NULL) { 3294 INP_WLOCK_ASSERT(tp->t_inpcb); 3295 } 3296 3297 /* Don't bother if destination was broadcast/multicast. */ 3298 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3299 goto drop; 3300 #ifdef INET6 3301 if (mtod(m, struct ip *)->ip_v == 6) { 3302 ip6 = mtod(m, struct ip6_hdr *); 3303 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3304 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3305 goto drop; 3306 /* IPv6 anycast check is done at tcp6_input() */ 3307 } 3308 #endif 3309 #if defined(INET) && defined(INET6) 3310 else 3311 #endif 3312 #ifdef INET 3313 { 3314 ip = mtod(m, struct ip *); 3315 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3316 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3317 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3318 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3319 goto drop; 3320 } 3321 #endif 3322 3323 /* Perform bandwidth limiting. */ 3324 if (badport_bandlim(rstreason) < 0) 3325 goto drop; 3326 3327 /* tcp_respond consumes the mbuf chain. */ 3328 if (th->th_flags & TH_ACK) { 3329 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3330 th->th_ack, TH_RST); 3331 } else { 3332 if (th->th_flags & TH_SYN) 3333 tlen++; 3334 if (th->th_flags & TH_FIN) 3335 tlen++; 3336 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3337 (tcp_seq)0, TH_RST|TH_ACK); 3338 } 3339 return; 3340 drop: 3341 m_freem(m); 3342 } 3343 3344 /* 3345 * Parse TCP options and place in tcpopt. 3346 */ 3347 void 3348 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3349 { 3350 int opt, optlen; 3351 3352 to->to_flags = 0; 3353 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3354 opt = cp[0]; 3355 if (opt == TCPOPT_EOL) 3356 break; 3357 if (opt == TCPOPT_NOP) 3358 optlen = 1; 3359 else { 3360 if (cnt < 2) 3361 break; 3362 optlen = cp[1]; 3363 if (optlen < 2 || optlen > cnt) 3364 break; 3365 } 3366 switch (opt) { 3367 case TCPOPT_MAXSEG: 3368 if (optlen != TCPOLEN_MAXSEG) 3369 continue; 3370 if (!(flags & TO_SYN)) 3371 continue; 3372 to->to_flags |= TOF_MSS; 3373 bcopy((char *)cp + 2, 3374 (char *)&to->to_mss, sizeof(to->to_mss)); 3375 to->to_mss = ntohs(to->to_mss); 3376 break; 3377 case TCPOPT_WINDOW: 3378 if (optlen != TCPOLEN_WINDOW) 3379 continue; 3380 if (!(flags & TO_SYN)) 3381 continue; 3382 to->to_flags |= TOF_SCALE; 3383 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3384 break; 3385 case TCPOPT_TIMESTAMP: 3386 if (optlen != TCPOLEN_TIMESTAMP) 3387 continue; 3388 to->to_flags |= TOF_TS; 3389 bcopy((char *)cp + 2, 3390 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3391 to->to_tsval = ntohl(to->to_tsval); 3392 bcopy((char *)cp + 6, 3393 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3394 to->to_tsecr = ntohl(to->to_tsecr); 3395 break; 3396 case TCPOPT_SIGNATURE: 3397 /* 3398 * In order to reply to a host which has set the 3399 * TCP_SIGNATURE option in its initial SYN, we have 3400 * to record the fact that the option was observed 3401 * here for the syncache code to perform the correct 3402 * response. 3403 */ 3404 if (optlen != TCPOLEN_SIGNATURE) 3405 continue; 3406 to->to_flags |= TOF_SIGNATURE; 3407 to->to_signature = cp + 2; 3408 break; 3409 case TCPOPT_SACK_PERMITTED: 3410 if (optlen != TCPOLEN_SACK_PERMITTED) 3411 continue; 3412 if (!(flags & TO_SYN)) 3413 continue; 3414 if (!V_tcp_do_sack) 3415 continue; 3416 to->to_flags |= TOF_SACKPERM; 3417 break; 3418 case TCPOPT_SACK: 3419 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3420 continue; 3421 if (flags & TO_SYN) 3422 continue; 3423 to->to_flags |= TOF_SACK; 3424 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3425 to->to_sacks = cp + 2; 3426 TCPSTAT_INC(tcps_sack_rcv_blocks); 3427 break; 3428 #ifdef TCP_RFC7413 3429 case TCPOPT_FAST_OPEN: 3430 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3431 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3432 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3433 continue; 3434 if (!(flags & TO_SYN)) 3435 continue; 3436 if (!V_tcp_fastopen_enabled) 3437 continue; 3438 to->to_flags |= TOF_FASTOPEN; 3439 to->to_tfo_len = optlen - 2; 3440 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3441 break; 3442 #endif 3443 default: 3444 continue; 3445 } 3446 } 3447 } 3448 3449 /* 3450 * Pull out of band byte out of a segment so 3451 * it doesn't appear in the user's data queue. 3452 * It is still reflected in the segment length for 3453 * sequencing purposes. 3454 */ 3455 void 3456 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3457 int off) 3458 { 3459 int cnt = off + th->th_urp - 1; 3460 3461 while (cnt >= 0) { 3462 if (m->m_len > cnt) { 3463 char *cp = mtod(m, caddr_t) + cnt; 3464 struct tcpcb *tp = sototcpcb(so); 3465 3466 INP_WLOCK_ASSERT(tp->t_inpcb); 3467 3468 tp->t_iobc = *cp; 3469 tp->t_oobflags |= TCPOOB_HAVEDATA; 3470 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3471 m->m_len--; 3472 if (m->m_flags & M_PKTHDR) 3473 m->m_pkthdr.len--; 3474 return; 3475 } 3476 cnt -= m->m_len; 3477 m = m->m_next; 3478 if (m == NULL) 3479 break; 3480 } 3481 panic("tcp_pulloutofband"); 3482 } 3483 3484 /* 3485 * Collect new round-trip time estimate 3486 * and update averages and current timeout. 3487 */ 3488 void 3489 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3490 { 3491 int delta; 3492 3493 INP_WLOCK_ASSERT(tp->t_inpcb); 3494 3495 TCPSTAT_INC(tcps_rttupdated); 3496 tp->t_rttupdated++; 3497 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3498 /* 3499 * srtt is stored as fixed point with 5 bits after the 3500 * binary point (i.e., scaled by 8). The following magic 3501 * is equivalent to the smoothing algorithm in rfc793 with 3502 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3503 * point). Adjust rtt to origin 0. 3504 */ 3505 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3506 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3507 3508 if ((tp->t_srtt += delta) <= 0) 3509 tp->t_srtt = 1; 3510 3511 /* 3512 * We accumulate a smoothed rtt variance (actually, a 3513 * smoothed mean difference), then set the retransmit 3514 * timer to smoothed rtt + 4 times the smoothed variance. 3515 * rttvar is stored as fixed point with 4 bits after the 3516 * binary point (scaled by 16). The following is 3517 * equivalent to rfc793 smoothing with an alpha of .75 3518 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3519 * rfc793's wired-in beta. 3520 */ 3521 if (delta < 0) 3522 delta = -delta; 3523 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3524 if ((tp->t_rttvar += delta) <= 0) 3525 tp->t_rttvar = 1; 3526 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3527 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3528 } else { 3529 /* 3530 * No rtt measurement yet - use the unsmoothed rtt. 3531 * Set the variance to half the rtt (so our first 3532 * retransmit happens at 3*rtt). 3533 */ 3534 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3535 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3536 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3537 } 3538 tp->t_rtttime = 0; 3539 tp->t_rxtshift = 0; 3540 3541 /* 3542 * the retransmit should happen at rtt + 4 * rttvar. 3543 * Because of the way we do the smoothing, srtt and rttvar 3544 * will each average +1/2 tick of bias. When we compute 3545 * the retransmit timer, we want 1/2 tick of rounding and 3546 * 1 extra tick because of +-1/2 tick uncertainty in the 3547 * firing of the timer. The bias will give us exactly the 3548 * 1.5 tick we need. But, because the bias is 3549 * statistical, we have to test that we don't drop below 3550 * the minimum feasible timer (which is 2 ticks). 3551 */ 3552 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3553 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3554 3555 /* 3556 * We received an ack for a packet that wasn't retransmitted; 3557 * it is probably safe to discard any error indications we've 3558 * received recently. This isn't quite right, but close enough 3559 * for now (a route might have failed after we sent a segment, 3560 * and the return path might not be symmetrical). 3561 */ 3562 tp->t_softerror = 0; 3563 } 3564 3565 /* 3566 * Determine a reasonable value for maxseg size. 3567 * If the route is known, check route for mtu. 3568 * If none, use an mss that can be handled on the outgoing interface 3569 * without forcing IP to fragment. If no route is found, route has no mtu, 3570 * or the destination isn't local, use a default, hopefully conservative 3571 * size (usually 512 or the default IP max size, but no more than the mtu 3572 * of the interface), as we can't discover anything about intervening 3573 * gateways or networks. We also initialize the congestion/slow start 3574 * window to be a single segment if the destination isn't local. 3575 * While looking at the routing entry, we also initialize other path-dependent 3576 * parameters from pre-set or cached values in the routing entry. 3577 * 3578 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3579 * IP options, e.g. IPSEC data, since length of this data may vary, and 3580 * thus it is calculated for every segment separately in tcp_output(). 3581 * 3582 * NOTE that this routine is only called when we process an incoming 3583 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3584 * settings are handled in tcp_mssopt(). 3585 */ 3586 void 3587 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3588 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3589 { 3590 int mss = 0; 3591 uint32_t maxmtu = 0; 3592 struct inpcb *inp = tp->t_inpcb; 3593 struct hc_metrics_lite metrics; 3594 #ifdef INET6 3595 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3596 size_t min_protoh = isipv6 ? 3597 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3598 sizeof (struct tcpiphdr); 3599 #else 3600 const size_t min_protoh = sizeof(struct tcpiphdr); 3601 #endif 3602 3603 INP_WLOCK_ASSERT(tp->t_inpcb); 3604 3605 if (mtuoffer != -1) { 3606 KASSERT(offer == -1, ("%s: conflict", __func__)); 3607 offer = mtuoffer - min_protoh; 3608 } 3609 3610 /* Initialize. */ 3611 #ifdef INET6 3612 if (isipv6) { 3613 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3614 tp->t_maxseg = V_tcp_v6mssdflt; 3615 } 3616 #endif 3617 #if defined(INET) && defined(INET6) 3618 else 3619 #endif 3620 #ifdef INET 3621 { 3622 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3623 tp->t_maxseg = V_tcp_mssdflt; 3624 } 3625 #endif 3626 3627 /* 3628 * No route to sender, stay with default mss and return. 3629 */ 3630 if (maxmtu == 0) { 3631 /* 3632 * In case we return early we need to initialize metrics 3633 * to a defined state as tcp_hc_get() would do for us 3634 * if there was no cache hit. 3635 */ 3636 if (metricptr != NULL) 3637 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3638 return; 3639 } 3640 3641 /* What have we got? */ 3642 switch (offer) { 3643 case 0: 3644 /* 3645 * Offer == 0 means that there was no MSS on the SYN 3646 * segment, in this case we use tcp_mssdflt as 3647 * already assigned to t_maxseg above. 3648 */ 3649 offer = tp->t_maxseg; 3650 break; 3651 3652 case -1: 3653 /* 3654 * Offer == -1 means that we didn't receive SYN yet. 3655 */ 3656 /* FALLTHROUGH */ 3657 3658 default: 3659 /* 3660 * Prevent DoS attack with too small MSS. Round up 3661 * to at least minmss. 3662 */ 3663 offer = max(offer, V_tcp_minmss); 3664 } 3665 3666 /* 3667 * rmx information is now retrieved from tcp_hostcache. 3668 */ 3669 tcp_hc_get(&inp->inp_inc, &metrics); 3670 if (metricptr != NULL) 3671 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3672 3673 /* 3674 * If there's a discovered mtu in tcp hostcache, use it. 3675 * Else, use the link mtu. 3676 */ 3677 if (metrics.rmx_mtu) 3678 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3679 else { 3680 #ifdef INET6 3681 if (isipv6) { 3682 mss = maxmtu - min_protoh; 3683 if (!V_path_mtu_discovery && 3684 !in6_localaddr(&inp->in6p_faddr)) 3685 mss = min(mss, V_tcp_v6mssdflt); 3686 } 3687 #endif 3688 #if defined(INET) && defined(INET6) 3689 else 3690 #endif 3691 #ifdef INET 3692 { 3693 mss = maxmtu - min_protoh; 3694 if (!V_path_mtu_discovery && 3695 !in_localaddr(inp->inp_faddr)) 3696 mss = min(mss, V_tcp_mssdflt); 3697 } 3698 #endif 3699 /* 3700 * XXX - The above conditional (mss = maxmtu - min_protoh) 3701 * probably violates the TCP spec. 3702 * The problem is that, since we don't know the 3703 * other end's MSS, we are supposed to use a conservative 3704 * default. But, if we do that, then MTU discovery will 3705 * never actually take place, because the conservative 3706 * default is much less than the MTUs typically seen 3707 * on the Internet today. For the moment, we'll sweep 3708 * this under the carpet. 3709 * 3710 * The conservative default might not actually be a problem 3711 * if the only case this occurs is when sending an initial 3712 * SYN with options and data to a host we've never talked 3713 * to before. Then, they will reply with an MSS value which 3714 * will get recorded and the new parameters should get 3715 * recomputed. For Further Study. 3716 */ 3717 } 3718 mss = min(mss, offer); 3719 3720 /* 3721 * Sanity check: make sure that maxseg will be large 3722 * enough to allow some data on segments even if the 3723 * all the option space is used (40bytes). Otherwise 3724 * funny things may happen in tcp_output. 3725 * 3726 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3727 */ 3728 mss = max(mss, 64); 3729 3730 tp->t_maxseg = mss; 3731 } 3732 3733 void 3734 tcp_mss(struct tcpcb *tp, int offer) 3735 { 3736 int mss; 3737 uint32_t bufsize; 3738 struct inpcb *inp; 3739 struct socket *so; 3740 struct hc_metrics_lite metrics; 3741 struct tcp_ifcap cap; 3742 3743 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3744 3745 bzero(&cap, sizeof(cap)); 3746 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3747 3748 mss = tp->t_maxseg; 3749 inp = tp->t_inpcb; 3750 3751 /* 3752 * If there's a pipesize, change the socket buffer to that size, 3753 * don't change if sb_hiwat is different than default (then it 3754 * has been changed on purpose with setsockopt). 3755 * Make the socket buffers an integral number of mss units; 3756 * if the mss is larger than the socket buffer, decrease the mss. 3757 */ 3758 so = inp->inp_socket; 3759 SOCKBUF_LOCK(&so->so_snd); 3760 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3761 bufsize = metrics.rmx_sendpipe; 3762 else 3763 bufsize = so->so_snd.sb_hiwat; 3764 if (bufsize < mss) 3765 mss = bufsize; 3766 else { 3767 bufsize = roundup(bufsize, mss); 3768 if (bufsize > sb_max) 3769 bufsize = sb_max; 3770 if (bufsize > so->so_snd.sb_hiwat) 3771 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3772 } 3773 SOCKBUF_UNLOCK(&so->so_snd); 3774 /* 3775 * Sanity check: make sure that maxseg will be large 3776 * enough to allow some data on segments even if the 3777 * all the option space is used (40bytes). Otherwise 3778 * funny things may happen in tcp_output. 3779 * 3780 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3781 */ 3782 tp->t_maxseg = max(mss, 64); 3783 3784 SOCKBUF_LOCK(&so->so_rcv); 3785 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3786 bufsize = metrics.rmx_recvpipe; 3787 else 3788 bufsize = so->so_rcv.sb_hiwat; 3789 if (bufsize > mss) { 3790 bufsize = roundup(bufsize, mss); 3791 if (bufsize > sb_max) 3792 bufsize = sb_max; 3793 if (bufsize > so->so_rcv.sb_hiwat) 3794 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3795 } 3796 SOCKBUF_UNLOCK(&so->so_rcv); 3797 3798 /* Check the interface for TSO capabilities. */ 3799 if (cap.ifcap & CSUM_TSO) { 3800 tp->t_flags |= TF_TSO; 3801 tp->t_tsomax = cap.tsomax; 3802 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3803 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3804 } 3805 } 3806 3807 /* 3808 * Determine the MSS option to send on an outgoing SYN. 3809 */ 3810 int 3811 tcp_mssopt(struct in_conninfo *inc) 3812 { 3813 int mss = 0; 3814 uint32_t thcmtu = 0; 3815 uint32_t maxmtu = 0; 3816 size_t min_protoh; 3817 3818 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3819 3820 #ifdef INET6 3821 if (inc->inc_flags & INC_ISIPV6) { 3822 mss = V_tcp_v6mssdflt; 3823 maxmtu = tcp_maxmtu6(inc, NULL); 3824 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3825 } 3826 #endif 3827 #if defined(INET) && defined(INET6) 3828 else 3829 #endif 3830 #ifdef INET 3831 { 3832 mss = V_tcp_mssdflt; 3833 maxmtu = tcp_maxmtu(inc, NULL); 3834 min_protoh = sizeof(struct tcpiphdr); 3835 } 3836 #endif 3837 #if defined(INET6) || defined(INET) 3838 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3839 #endif 3840 3841 if (maxmtu && thcmtu) 3842 mss = min(maxmtu, thcmtu) - min_protoh; 3843 else if (maxmtu || thcmtu) 3844 mss = max(maxmtu, thcmtu) - min_protoh; 3845 3846 return (mss); 3847 } 3848 3849 3850 /* 3851 * On a partial ack arrives, force the retransmission of the 3852 * next unacknowledged segment. Do not clear tp->t_dupacks. 3853 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3854 * be started again. 3855 */ 3856 void 3857 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3858 { 3859 tcp_seq onxt = tp->snd_nxt; 3860 uint32_t ocwnd = tp->snd_cwnd; 3861 u_int maxseg = tcp_maxseg(tp); 3862 3863 INP_WLOCK_ASSERT(tp->t_inpcb); 3864 3865 tcp_timer_activate(tp, TT_REXMT, 0); 3866 tp->t_rtttime = 0; 3867 tp->snd_nxt = th->th_ack; 3868 /* 3869 * Set snd_cwnd to one segment beyond acknowledged offset. 3870 * (tp->snd_una has not yet been updated when this function is called.) 3871 */ 3872 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3873 tp->t_flags |= TF_ACKNOW; 3874 (void) tp->t_fb->tfb_tcp_output(tp); 3875 tp->snd_cwnd = ocwnd; 3876 if (SEQ_GT(onxt, tp->snd_nxt)) 3877 tp->snd_nxt = onxt; 3878 /* 3879 * Partial window deflation. Relies on fact that tp->snd_una 3880 * not updated yet. 3881 */ 3882 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3883 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3884 else 3885 tp->snd_cwnd = 0; 3886 tp->snd_cwnd += maxseg; 3887 } 3888 3889 int 3890 tcp_compute_pipe(struct tcpcb *tp) 3891 { 3892 return (tp->snd_max - tp->snd_una + 3893 tp->sackhint.sack_bytes_rexmit - 3894 tp->sackhint.sacked_bytes); 3895 } 3896