1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 4. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_inet.h" 54 #include "opt_inet6.h" 55 #include "opt_ipsec.h" 56 #include "opt_tcpdebug.h" 57 58 #include <sys/param.h> 59 #include <sys/kernel.h> 60 #include <sys/hhook.h> 61 #include <sys/malloc.h> 62 #include <sys/mbuf.h> 63 #include <sys/proc.h> /* for proc0 declaration */ 64 #include <sys/protosw.h> 65 #include <sys/sdt.h> 66 #include <sys/signalvar.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/sysctl.h> 70 #include <sys/syslog.h> 71 #include <sys/systm.h> 72 73 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 74 75 #include <vm/uma.h> 76 77 #include <net/if.h> 78 #include <net/if_var.h> 79 #include <net/route.h> 80 #include <net/vnet.h> 81 82 #define TCPSTATES /* for logging */ 83 84 #include <netinet/in.h> 85 #include <netinet/in_kdtrace.h> 86 #include <netinet/in_pcb.h> 87 #include <netinet/in_systm.h> 88 #include <netinet/ip.h> 89 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 90 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 91 #include <netinet/ip_var.h> 92 #include <netinet/ip_options.h> 93 #include <netinet/ip6.h> 94 #include <netinet/icmp6.h> 95 #include <netinet6/in6_pcb.h> 96 #include <netinet6/in6_var.h> 97 #include <netinet6/ip6_var.h> 98 #include <netinet6/nd6.h> 99 #ifdef TCP_RFC7413 100 #include <netinet/tcp_fastopen.h> 101 #endif 102 #include <netinet/tcp.h> 103 #include <netinet/tcp_fsm.h> 104 #include <netinet/tcp_seq.h> 105 #include <netinet/tcp_timer.h> 106 #include <netinet/tcp_var.h> 107 #include <netinet6/tcp6_var.h> 108 #include <netinet/tcpip.h> 109 #include <netinet/cc/cc.h> 110 #ifdef TCPPCAP 111 #include <netinet/tcp_pcap.h> 112 #endif 113 #include <netinet/tcp_syncache.h> 114 #ifdef TCPDEBUG 115 #include <netinet/tcp_debug.h> 116 #endif /* TCPDEBUG */ 117 #ifdef TCP_OFFLOAD 118 #include <netinet/tcp_offload.h> 119 #endif 120 121 #ifdef IPSEC 122 #include <netipsec/ipsec.h> 123 #include <netipsec/ipsec6.h> 124 #endif /*IPSEC*/ 125 126 #include <machine/in_cksum.h> 127 128 #include <security/mac/mac_framework.h> 129 130 const int tcprexmtthresh = 3; 131 132 int tcp_log_in_vain = 0; 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 134 &tcp_log_in_vain, 0, 135 "Log all incoming TCP segments to closed ports"); 136 137 VNET_DEFINE(int, blackhole) = 0; 138 #define V_blackhole VNET(blackhole) 139 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 140 &VNET_NAME(blackhole), 0, 141 "Do not send RST on segments to closed ports"); 142 143 VNET_DEFINE(int, tcp_delack_enabled) = 1; 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 145 &VNET_NAME(tcp_delack_enabled), 0, 146 "Delay ACK to try and piggyback it onto a data packet"); 147 148 VNET_DEFINE(int, drop_synfin) = 0; 149 #define V_drop_synfin VNET(drop_synfin) 150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 151 &VNET_NAME(drop_synfin), 0, 152 "Drop TCP packets with SYN+FIN set"); 153 154 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 156 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 157 "Use calculated pipe/in-flight bytes per RFC 6675"); 158 159 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 160 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 162 &VNET_NAME(tcp_do_rfc3042), 0, 163 "Enable RFC 3042 (Limited Transmit)"); 164 165 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 167 &VNET_NAME(tcp_do_rfc3390), 0, 168 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 169 170 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 172 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 173 "Slow-start flight size (initial congestion window) in number of segments"); 174 175 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 177 &VNET_NAME(tcp_do_rfc3465), 0, 178 "Enable RFC 3465 (Appropriate Byte Counting)"); 179 180 VNET_DEFINE(int, tcp_abc_l_var) = 2; 181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 182 &VNET_NAME(tcp_abc_l_var), 2, 183 "Cap the max cwnd increment during slow-start to this number of segments"); 184 185 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 186 187 VNET_DEFINE(int, tcp_do_ecn) = 2; 188 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 189 &VNET_NAME(tcp_do_ecn), 0, 190 "TCP ECN support"); 191 192 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 193 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 194 &VNET_NAME(tcp_ecn_maxretries), 0, 195 "Max retries before giving up on ECN"); 196 197 VNET_DEFINE(int, tcp_insecure_syn) = 0; 198 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 200 &VNET_NAME(tcp_insecure_syn), 0, 201 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 202 203 VNET_DEFINE(int, tcp_insecure_rst) = 0; 204 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 206 &VNET_NAME(tcp_insecure_rst), 0, 207 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 208 209 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 210 #define V_tcp_recvspace VNET(tcp_recvspace) 211 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 213 214 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 215 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_do_autorcvbuf), 0, 218 "Enable automatic receive buffer sizing"); 219 220 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 221 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 223 &VNET_NAME(tcp_autorcvbuf_inc), 0, 224 "Incrementor step size of automatic receive buffer"); 225 226 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 227 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 229 &VNET_NAME(tcp_autorcvbuf_max), 0, 230 "Max size of automatic receive buffer"); 231 232 VNET_DEFINE(struct inpcbhead, tcb); 233 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 234 VNET_DEFINE(struct inpcbinfo, tcbinfo); 235 236 /* 237 * TCP statistics are stored in an array of counter(9)s, which size matches 238 * size of struct tcpstat. TCP running connection count is a regular array. 239 */ 240 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 241 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 242 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 243 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 244 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 245 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 246 "TCP connection counts by TCP state"); 247 248 static void 249 tcp_vnet_init(const void *unused) 250 { 251 252 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 253 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 254 } 255 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 256 tcp_vnet_init, NULL); 257 258 #ifdef VIMAGE 259 static void 260 tcp_vnet_uninit(const void *unused) 261 { 262 263 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 264 VNET_PCPUSTAT_FREE(tcpstat); 265 } 266 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 267 tcp_vnet_uninit, NULL); 268 #endif /* VIMAGE */ 269 270 /* 271 * Kernel module interface for updating tcpstat. The argument is an index 272 * into tcpstat treated as an array. 273 */ 274 void 275 kmod_tcpstat_inc(int statnum) 276 { 277 278 counter_u64_add(VNET(tcpstat)[statnum], 1); 279 } 280 281 /* 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 } 298 299 /* 300 * CC wrapper hook functions 301 */ 302 void 303 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 304 uint16_t type) 305 { 306 INP_WLOCK_ASSERT(tp->t_inpcb); 307 308 tp->ccv->nsegs = nsegs; 309 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 310 if (tp->snd_cwnd <= tp->snd_wnd) 311 tp->ccv->flags |= CCF_CWND_LIMITED; 312 else 313 tp->ccv->flags &= ~CCF_CWND_LIMITED; 314 315 if (type == CC_ACK) { 316 if (tp->snd_cwnd > tp->snd_ssthresh) { 317 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 318 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 319 if (tp->t_bytes_acked >= tp->snd_cwnd) { 320 tp->t_bytes_acked -= tp->snd_cwnd; 321 tp->ccv->flags |= CCF_ABC_SENTAWND; 322 } 323 } else { 324 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 325 tp->t_bytes_acked = 0; 326 } 327 } 328 329 if (CC_ALGO(tp)->ack_received != NULL) { 330 /* XXXLAS: Find a way to live without this */ 331 tp->ccv->curack = th->th_ack; 332 CC_ALGO(tp)->ack_received(tp->ccv, type); 333 } 334 } 335 336 void 337 cc_conn_init(struct tcpcb *tp) 338 { 339 struct hc_metrics_lite metrics; 340 struct inpcb *inp = tp->t_inpcb; 341 u_int maxseg; 342 int rtt; 343 344 INP_WLOCK_ASSERT(tp->t_inpcb); 345 346 tcp_hc_get(&inp->inp_inc, &metrics); 347 maxseg = tcp_maxseg(tp); 348 349 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 350 tp->t_srtt = rtt; 351 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 352 TCPSTAT_INC(tcps_usedrtt); 353 if (metrics.rmx_rttvar) { 354 tp->t_rttvar = metrics.rmx_rttvar; 355 TCPSTAT_INC(tcps_usedrttvar); 356 } else { 357 /* default variation is +- 1 rtt */ 358 tp->t_rttvar = 359 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 360 } 361 TCPT_RANGESET(tp->t_rxtcur, 362 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 363 tp->t_rttmin, TCPTV_REXMTMAX); 364 } 365 if (metrics.rmx_ssthresh) { 366 /* 367 * There's some sort of gateway or interface 368 * buffer limit on the path. Use this to set 369 * the slow start threshold, but set the 370 * threshold to no less than 2*mss. 371 */ 372 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 373 TCPSTAT_INC(tcps_usedssthresh); 374 } 375 376 /* 377 * Set the initial slow-start flight size. 378 * 379 * RFC5681 Section 3.1 specifies the default conservative values. 380 * RFC3390 specifies slightly more aggressive values. 381 * RFC6928 increases it to ten segments. 382 * Support for user specified value for initial flight size. 383 * 384 * If a SYN or SYN/ACK was lost and retransmitted, we have to 385 * reduce the initial CWND to one segment as congestion is likely 386 * requiring us to be cautious. 387 */ 388 if (tp->snd_cwnd == 1) 389 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 390 else if (V_tcp_initcwnd_segments) 391 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 392 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 393 else if (V_tcp_do_rfc3390) 394 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 395 else { 396 /* Per RFC5681 Section 3.1 */ 397 if (maxseg > 2190) 398 tp->snd_cwnd = 2 * maxseg; 399 else if (maxseg > 1095) 400 tp->snd_cwnd = 3 * maxseg; 401 else 402 tp->snd_cwnd = 4 * maxseg; 403 } 404 405 if (CC_ALGO(tp)->conn_init != NULL) 406 CC_ALGO(tp)->conn_init(tp->ccv); 407 } 408 409 void inline 410 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 411 { 412 u_int maxseg; 413 414 INP_WLOCK_ASSERT(tp->t_inpcb); 415 416 switch(type) { 417 case CC_NDUPACK: 418 if (!IN_FASTRECOVERY(tp->t_flags)) { 419 tp->snd_recover = tp->snd_max; 420 if (tp->t_flags & TF_ECN_PERMIT) 421 tp->t_flags |= TF_ECN_SND_CWR; 422 } 423 break; 424 case CC_ECN: 425 if (!IN_CONGRECOVERY(tp->t_flags)) { 426 TCPSTAT_INC(tcps_ecn_rcwnd); 427 tp->snd_recover = tp->snd_max; 428 if (tp->t_flags & TF_ECN_PERMIT) 429 tp->t_flags |= TF_ECN_SND_CWR; 430 } 431 break; 432 case CC_RTO: 433 maxseg = tcp_maxseg(tp); 434 tp->t_dupacks = 0; 435 tp->t_bytes_acked = 0; 436 EXIT_RECOVERY(tp->t_flags); 437 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 438 maxseg) * maxseg; 439 tp->snd_cwnd = maxseg; 440 break; 441 case CC_RTO_ERR: 442 TCPSTAT_INC(tcps_sndrexmitbad); 443 /* RTO was unnecessary, so reset everything. */ 444 tp->snd_cwnd = tp->snd_cwnd_prev; 445 tp->snd_ssthresh = tp->snd_ssthresh_prev; 446 tp->snd_recover = tp->snd_recover_prev; 447 if (tp->t_flags & TF_WASFRECOVERY) 448 ENTER_FASTRECOVERY(tp->t_flags); 449 if (tp->t_flags & TF_WASCRECOVERY) 450 ENTER_CONGRECOVERY(tp->t_flags); 451 tp->snd_nxt = tp->snd_max; 452 tp->t_flags &= ~TF_PREVVALID; 453 tp->t_badrxtwin = 0; 454 break; 455 } 456 457 if (CC_ALGO(tp)->cong_signal != NULL) { 458 if (th != NULL) 459 tp->ccv->curack = th->th_ack; 460 CC_ALGO(tp)->cong_signal(tp->ccv, type); 461 } 462 } 463 464 void inline 465 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 466 { 467 INP_WLOCK_ASSERT(tp->t_inpcb); 468 469 /* XXXLAS: KASSERT that we're in recovery? */ 470 471 if (CC_ALGO(tp)->post_recovery != NULL) { 472 tp->ccv->curack = th->th_ack; 473 CC_ALGO(tp)->post_recovery(tp->ccv); 474 } 475 /* XXXLAS: EXIT_RECOVERY ? */ 476 tp->t_bytes_acked = 0; 477 } 478 479 #ifdef TCP_SIGNATURE 480 static inline int 481 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 482 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 483 { 484 int ret; 485 486 tcp_fields_to_net(th); 487 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 488 tcp_fields_to_host(th); 489 return (ret); 490 } 491 #endif 492 493 /* 494 * Indicate whether this ack should be delayed. We can delay the ack if 495 * following conditions are met: 496 * - There is no delayed ack timer in progress. 497 * - Our last ack wasn't a 0-sized window. We never want to delay 498 * the ack that opens up a 0-sized window. 499 * - LRO wasn't used for this segment. We make sure by checking that the 500 * segment size is not larger than the MSS. 501 */ 502 #define DELAY_ACK(tp, tlen) \ 503 ((!tcp_timer_active(tp, TT_DELACK) && \ 504 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 505 (tlen <= tp->t_maxseg) && \ 506 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 507 508 static void inline 509 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 510 { 511 INP_WLOCK_ASSERT(tp->t_inpcb); 512 513 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 514 switch (iptos & IPTOS_ECN_MASK) { 515 case IPTOS_ECN_CE: 516 tp->ccv->flags |= CCF_IPHDR_CE; 517 break; 518 case IPTOS_ECN_ECT0: 519 tp->ccv->flags &= ~CCF_IPHDR_CE; 520 break; 521 case IPTOS_ECN_ECT1: 522 tp->ccv->flags &= ~CCF_IPHDR_CE; 523 break; 524 } 525 526 if (th->th_flags & TH_CWR) 527 tp->ccv->flags |= CCF_TCPHDR_CWR; 528 else 529 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 530 531 if (tp->t_flags & TF_DELACK) 532 tp->ccv->flags |= CCF_DELACK; 533 else 534 tp->ccv->flags &= ~CCF_DELACK; 535 536 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 537 538 if (tp->ccv->flags & CCF_ACKNOW) 539 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 540 } 541 } 542 543 /* 544 * TCP input handling is split into multiple parts: 545 * tcp6_input is a thin wrapper around tcp_input for the extended 546 * ip6_protox[] call format in ip6_input 547 * tcp_input handles primary segment validation, inpcb lookup and 548 * SYN processing on listen sockets 549 * tcp_do_segment processes the ACK and text of the segment for 550 * establishing, established and closing connections 551 */ 552 #ifdef INET6 553 int 554 tcp6_input(struct mbuf **mp, int *offp, int proto) 555 { 556 struct mbuf *m = *mp; 557 struct in6_ifaddr *ia6; 558 struct ip6_hdr *ip6; 559 560 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 561 562 /* 563 * draft-itojun-ipv6-tcp-to-anycast 564 * better place to put this in? 565 */ 566 ip6 = mtod(m, struct ip6_hdr *); 567 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 568 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 569 struct ip6_hdr *ip6; 570 571 ifa_free(&ia6->ia_ifa); 572 ip6 = mtod(m, struct ip6_hdr *); 573 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 574 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 575 return (IPPROTO_DONE); 576 } 577 if (ia6) 578 ifa_free(&ia6->ia_ifa); 579 580 return (tcp_input(mp, offp, proto)); 581 } 582 #endif /* INET6 */ 583 584 int 585 tcp_input(struct mbuf **mp, int *offp, int proto) 586 { 587 struct mbuf *m = *mp; 588 struct tcphdr *th = NULL; 589 struct ip *ip = NULL; 590 struct inpcb *inp = NULL; 591 struct tcpcb *tp = NULL; 592 struct socket *so = NULL; 593 u_char *optp = NULL; 594 int off0; 595 int optlen = 0; 596 #ifdef INET 597 int len; 598 #endif 599 int tlen = 0, off; 600 int drop_hdrlen; 601 int thflags; 602 int rstreason = 0; /* For badport_bandlim accounting purposes */ 603 #ifdef TCP_SIGNATURE 604 uint8_t sig_checked = 0; 605 #endif 606 uint8_t iptos = 0; 607 struct m_tag *fwd_tag = NULL; 608 #ifdef INET6 609 struct ip6_hdr *ip6 = NULL; 610 int isipv6; 611 #else 612 const void *ip6 = NULL; 613 #endif /* INET6 */ 614 struct tcpopt to; /* options in this segment */ 615 char *s = NULL; /* address and port logging */ 616 int ti_locked; 617 #ifdef TCPDEBUG 618 /* 619 * The size of tcp_saveipgen must be the size of the max ip header, 620 * now IPv6. 621 */ 622 u_char tcp_saveipgen[IP6_HDR_LEN]; 623 struct tcphdr tcp_savetcp; 624 short ostate = 0; 625 #endif 626 627 #ifdef INET6 628 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 629 #endif 630 631 off0 = *offp; 632 m = *mp; 633 *mp = NULL; 634 to.to_flags = 0; 635 TCPSTAT_INC(tcps_rcvtotal); 636 637 #ifdef INET6 638 if (isipv6) { 639 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 640 641 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 642 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 643 if (m == NULL) { 644 TCPSTAT_INC(tcps_rcvshort); 645 return (IPPROTO_DONE); 646 } 647 } 648 649 ip6 = mtod(m, struct ip6_hdr *); 650 th = (struct tcphdr *)((caddr_t)ip6 + off0); 651 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 652 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 653 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 654 th->th_sum = m->m_pkthdr.csum_data; 655 else 656 th->th_sum = in6_cksum_pseudo(ip6, tlen, 657 IPPROTO_TCP, m->m_pkthdr.csum_data); 658 th->th_sum ^= 0xffff; 659 } else 660 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 661 if (th->th_sum) { 662 TCPSTAT_INC(tcps_rcvbadsum); 663 goto drop; 664 } 665 666 /* 667 * Be proactive about unspecified IPv6 address in source. 668 * As we use all-zero to indicate unbounded/unconnected pcb, 669 * unspecified IPv6 address can be used to confuse us. 670 * 671 * Note that packets with unspecified IPv6 destination is 672 * already dropped in ip6_input. 673 */ 674 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 675 /* XXX stat */ 676 goto drop; 677 } 678 } 679 #endif 680 #if defined(INET) && defined(INET6) 681 else 682 #endif 683 #ifdef INET 684 { 685 /* 686 * Get IP and TCP header together in first mbuf. 687 * Note: IP leaves IP header in first mbuf. 688 */ 689 if (off0 > sizeof (struct ip)) { 690 ip_stripoptions(m); 691 off0 = sizeof(struct ip); 692 } 693 if (m->m_len < sizeof (struct tcpiphdr)) { 694 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 695 == NULL) { 696 TCPSTAT_INC(tcps_rcvshort); 697 return (IPPROTO_DONE); 698 } 699 } 700 ip = mtod(m, struct ip *); 701 th = (struct tcphdr *)((caddr_t)ip + off0); 702 tlen = ntohs(ip->ip_len) - off0; 703 704 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 705 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 706 th->th_sum = m->m_pkthdr.csum_data; 707 else 708 th->th_sum = in_pseudo(ip->ip_src.s_addr, 709 ip->ip_dst.s_addr, 710 htonl(m->m_pkthdr.csum_data + tlen + 711 IPPROTO_TCP)); 712 th->th_sum ^= 0xffff; 713 } else { 714 struct ipovly *ipov = (struct ipovly *)ip; 715 716 /* 717 * Checksum extended TCP header and data. 718 */ 719 len = off0 + tlen; 720 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 721 ipov->ih_len = htons(tlen); 722 th->th_sum = in_cksum(m, len); 723 /* Reset length for SDT probes. */ 724 ip->ip_len = htons(tlen + off0); 725 } 726 727 if (th->th_sum) { 728 TCPSTAT_INC(tcps_rcvbadsum); 729 goto drop; 730 } 731 /* Re-initialization for later version check */ 732 ip->ip_v = IPVERSION; 733 } 734 #endif /* INET */ 735 736 #ifdef INET6 737 if (isipv6) 738 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 739 #endif 740 #if defined(INET) && defined(INET6) 741 else 742 #endif 743 #ifdef INET 744 iptos = ip->ip_tos; 745 #endif 746 747 /* 748 * Check that TCP offset makes sense, 749 * pull out TCP options and adjust length. XXX 750 */ 751 off = th->th_off << 2; 752 if (off < sizeof (struct tcphdr) || off > tlen) { 753 TCPSTAT_INC(tcps_rcvbadoff); 754 goto drop; 755 } 756 tlen -= off; /* tlen is used instead of ti->ti_len */ 757 if (off > sizeof (struct tcphdr)) { 758 #ifdef INET6 759 if (isipv6) { 760 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 761 ip6 = mtod(m, struct ip6_hdr *); 762 th = (struct tcphdr *)((caddr_t)ip6 + off0); 763 } 764 #endif 765 #if defined(INET) && defined(INET6) 766 else 767 #endif 768 #ifdef INET 769 { 770 if (m->m_len < sizeof(struct ip) + off) { 771 if ((m = m_pullup(m, sizeof (struct ip) + off)) 772 == NULL) { 773 TCPSTAT_INC(tcps_rcvshort); 774 return (IPPROTO_DONE); 775 } 776 ip = mtod(m, struct ip *); 777 th = (struct tcphdr *)((caddr_t)ip + off0); 778 } 779 } 780 #endif 781 optlen = off - sizeof (struct tcphdr); 782 optp = (u_char *)(th + 1); 783 } 784 thflags = th->th_flags; 785 786 /* 787 * Convert TCP protocol specific fields to host format. 788 */ 789 tcp_fields_to_host(th); 790 791 /* 792 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 793 */ 794 drop_hdrlen = off0 + off; 795 796 /* 797 * Locate pcb for segment; if we're likely to add or remove a 798 * connection then first acquire pcbinfo lock. There are three cases 799 * where we might discover later we need a write lock despite the 800 * flags: ACKs moving a connection out of the syncache, ACKs for a 801 * connection in TIMEWAIT and SYNs not targeting a listening socket. 802 */ 803 if ((thflags & (TH_FIN | TH_RST)) != 0) { 804 INP_INFO_RLOCK(&V_tcbinfo); 805 ti_locked = TI_RLOCKED; 806 } else 807 ti_locked = TI_UNLOCKED; 808 809 /* 810 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 811 */ 812 if ( 813 #ifdef INET6 814 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 815 #ifdef INET 816 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 817 #endif 818 #endif 819 #if defined(INET) && !defined(INET6) 820 (m->m_flags & M_IP_NEXTHOP) 821 #endif 822 ) 823 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 824 825 findpcb: 826 #ifdef INVARIANTS 827 if (ti_locked == TI_RLOCKED) { 828 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 829 } else { 830 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 831 } 832 #endif 833 #ifdef INET6 834 if (isipv6 && fwd_tag != NULL) { 835 struct sockaddr_in6 *next_hop6; 836 837 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 838 /* 839 * Transparently forwarded. Pretend to be the destination. 840 * Already got one like this? 841 */ 842 inp = in6_pcblookup_mbuf(&V_tcbinfo, 843 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 844 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 845 if (!inp) { 846 /* 847 * It's new. Try to find the ambushing socket. 848 * Because we've rewritten the destination address, 849 * any hardware-generated hash is ignored. 850 */ 851 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 852 th->th_sport, &next_hop6->sin6_addr, 853 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 854 th->th_dport, INPLOOKUP_WILDCARD | 855 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 856 } 857 } else if (isipv6) { 858 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 859 th->th_sport, &ip6->ip6_dst, th->th_dport, 860 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 861 m->m_pkthdr.rcvif, m); 862 } 863 #endif /* INET6 */ 864 #if defined(INET6) && defined(INET) 865 else 866 #endif 867 #ifdef INET 868 if (fwd_tag != NULL) { 869 struct sockaddr_in *next_hop; 870 871 next_hop = (struct sockaddr_in *)(fwd_tag+1); 872 /* 873 * Transparently forwarded. Pretend to be the destination. 874 * already got one like this? 875 */ 876 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 877 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 878 m->m_pkthdr.rcvif, m); 879 if (!inp) { 880 /* 881 * It's new. Try to find the ambushing socket. 882 * Because we've rewritten the destination address, 883 * any hardware-generated hash is ignored. 884 */ 885 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 886 th->th_sport, next_hop->sin_addr, 887 next_hop->sin_port ? ntohs(next_hop->sin_port) : 888 th->th_dport, INPLOOKUP_WILDCARD | 889 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 890 } 891 } else 892 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 893 th->th_sport, ip->ip_dst, th->th_dport, 894 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 895 m->m_pkthdr.rcvif, m); 896 #endif /* INET */ 897 898 /* 899 * If the INPCB does not exist then all data in the incoming 900 * segment is discarded and an appropriate RST is sent back. 901 * XXX MRT Send RST using which routing table? 902 */ 903 if (inp == NULL) { 904 /* 905 * Log communication attempts to ports that are not 906 * in use. 907 */ 908 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 909 tcp_log_in_vain == 2) { 910 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 911 log(LOG_INFO, "%s; %s: Connection attempt " 912 "to closed port\n", s, __func__); 913 } 914 /* 915 * When blackholing do not respond with a RST but 916 * completely ignore the segment and drop it. 917 */ 918 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 919 V_blackhole == 2) 920 goto dropunlock; 921 922 rstreason = BANDLIM_RST_CLOSEDPORT; 923 goto dropwithreset; 924 } 925 INP_WLOCK_ASSERT(inp); 926 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 927 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 928 ((inp->inp_socket == NULL) || 929 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 930 inp->inp_flowid = m->m_pkthdr.flowid; 931 inp->inp_flowtype = M_HASHTYPE_GET(m); 932 } 933 #ifdef IPSEC 934 #ifdef INET6 935 if (isipv6 && ipsec6_in_reject(m, inp)) { 936 goto dropunlock; 937 } else 938 #endif /* INET6 */ 939 if (ipsec4_in_reject(m, inp) != 0) { 940 goto dropunlock; 941 } 942 #endif /* IPSEC */ 943 944 /* 945 * Check the minimum TTL for socket. 946 */ 947 if (inp->inp_ip_minttl != 0) { 948 #ifdef INET6 949 if (isipv6) { 950 if (inp->inp_ip_minttl > ip6->ip6_hlim) 951 goto dropunlock; 952 } else 953 #endif 954 if (inp->inp_ip_minttl > ip->ip_ttl) 955 goto dropunlock; 956 } 957 958 /* 959 * A previous connection in TIMEWAIT state is supposed to catch stray 960 * or duplicate segments arriving late. If this segment was a 961 * legitimate new connection attempt, the old INPCB gets removed and 962 * we can try again to find a listening socket. 963 * 964 * At this point, due to earlier optimism, we may hold only an inpcb 965 * lock, and not the inpcbinfo write lock. If so, we need to try to 966 * acquire it, or if that fails, acquire a reference on the inpcb, 967 * drop all locks, acquire a global write lock, and then re-acquire 968 * the inpcb lock. We may at that point discover that another thread 969 * has tried to free the inpcb, in which case we need to loop back 970 * and try to find a new inpcb to deliver to. 971 * 972 * XXXRW: It may be time to rethink timewait locking. 973 */ 974 relocked: 975 if (inp->inp_flags & INP_TIMEWAIT) { 976 if (ti_locked == TI_UNLOCKED) { 977 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 978 in_pcbref(inp); 979 INP_WUNLOCK(inp); 980 INP_INFO_RLOCK(&V_tcbinfo); 981 ti_locked = TI_RLOCKED; 982 INP_WLOCK(inp); 983 if (in_pcbrele_wlocked(inp)) { 984 inp = NULL; 985 goto findpcb; 986 } 987 } else 988 ti_locked = TI_RLOCKED; 989 } 990 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 991 992 if (thflags & TH_SYN) 993 tcp_dooptions(&to, optp, optlen, TO_SYN); 994 /* 995 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 996 */ 997 if (tcp_twcheck(inp, &to, th, m, tlen)) 998 goto findpcb; 999 INP_INFO_RUNLOCK(&V_tcbinfo); 1000 return (IPPROTO_DONE); 1001 } 1002 /* 1003 * The TCPCB may no longer exist if the connection is winding 1004 * down or it is in the CLOSED state. Either way we drop the 1005 * segment and send an appropriate response. 1006 */ 1007 tp = intotcpcb(inp); 1008 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1009 rstreason = BANDLIM_RST_CLOSEDPORT; 1010 goto dropwithreset; 1011 } 1012 1013 #ifdef TCP_OFFLOAD 1014 if (tp->t_flags & TF_TOE) { 1015 tcp_offload_input(tp, m); 1016 m = NULL; /* consumed by the TOE driver */ 1017 goto dropunlock; 1018 } 1019 #endif 1020 1021 /* 1022 * We've identified a valid inpcb, but it could be that we need an 1023 * inpcbinfo write lock but don't hold it. In this case, attempt to 1024 * acquire using the same strategy as the TIMEWAIT case above. If we 1025 * relock, we have to jump back to 'relocked' as the connection might 1026 * now be in TIMEWAIT. 1027 */ 1028 #ifdef INVARIANTS 1029 if ((thflags & (TH_FIN | TH_RST)) != 0) 1030 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1031 #endif 1032 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1033 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1034 !(tp->t_flags & TF_FASTOPEN)))) { 1035 if (ti_locked == TI_UNLOCKED) { 1036 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1037 in_pcbref(inp); 1038 INP_WUNLOCK(inp); 1039 INP_INFO_RLOCK(&V_tcbinfo); 1040 ti_locked = TI_RLOCKED; 1041 INP_WLOCK(inp); 1042 if (in_pcbrele_wlocked(inp)) { 1043 inp = NULL; 1044 goto findpcb; 1045 } 1046 goto relocked; 1047 } else 1048 ti_locked = TI_RLOCKED; 1049 } 1050 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1051 } 1052 1053 #ifdef MAC 1054 INP_WLOCK_ASSERT(inp); 1055 if (mac_inpcb_check_deliver(inp, m)) 1056 goto dropunlock; 1057 #endif 1058 so = inp->inp_socket; 1059 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1060 #ifdef TCPDEBUG 1061 if (so->so_options & SO_DEBUG) { 1062 ostate = tp->t_state; 1063 #ifdef INET6 1064 if (isipv6) { 1065 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1066 } else 1067 #endif 1068 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1069 tcp_savetcp = *th; 1070 } 1071 #endif /* TCPDEBUG */ 1072 /* 1073 * When the socket is accepting connections (the INPCB is in LISTEN 1074 * state) we look into the SYN cache if this is a new connection 1075 * attempt or the completion of a previous one. 1076 */ 1077 if (so->so_options & SO_ACCEPTCONN) { 1078 struct in_conninfo inc; 1079 1080 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1081 "tp not listening", __func__)); 1082 bzero(&inc, sizeof(inc)); 1083 #ifdef INET6 1084 if (isipv6) { 1085 inc.inc_flags |= INC_ISIPV6; 1086 inc.inc6_faddr = ip6->ip6_src; 1087 inc.inc6_laddr = ip6->ip6_dst; 1088 } else 1089 #endif 1090 { 1091 inc.inc_faddr = ip->ip_src; 1092 inc.inc_laddr = ip->ip_dst; 1093 } 1094 inc.inc_fport = th->th_sport; 1095 inc.inc_lport = th->th_dport; 1096 inc.inc_fibnum = so->so_fibnum; 1097 1098 /* 1099 * Check for an existing connection attempt in syncache if 1100 * the flag is only ACK. A successful lookup creates a new 1101 * socket appended to the listen queue in SYN_RECEIVED state. 1102 */ 1103 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1104 1105 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1106 /* 1107 * Parse the TCP options here because 1108 * syncookies need access to the reflected 1109 * timestamp. 1110 */ 1111 tcp_dooptions(&to, optp, optlen, 0); 1112 /* 1113 * NB: syncache_expand() doesn't unlock 1114 * inp and tcpinfo locks. 1115 */ 1116 if (!syncache_expand(&inc, &to, th, &so, m)) { 1117 /* 1118 * No syncache entry or ACK was not 1119 * for our SYN/ACK. Send a RST. 1120 * NB: syncache did its own logging 1121 * of the failure cause. 1122 */ 1123 rstreason = BANDLIM_RST_OPENPORT; 1124 goto dropwithreset; 1125 } 1126 #ifdef TCP_RFC7413 1127 new_tfo_socket: 1128 #endif 1129 if (so == NULL) { 1130 /* 1131 * We completed the 3-way handshake 1132 * but could not allocate a socket 1133 * either due to memory shortage, 1134 * listen queue length limits or 1135 * global socket limits. Send RST 1136 * or wait and have the remote end 1137 * retransmit the ACK for another 1138 * try. 1139 */ 1140 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1141 log(LOG_DEBUG, "%s; %s: Listen socket: " 1142 "Socket allocation failed due to " 1143 "limits or memory shortage, %s\n", 1144 s, __func__, 1145 V_tcp_sc_rst_sock_fail ? 1146 "sending RST" : "try again"); 1147 if (V_tcp_sc_rst_sock_fail) { 1148 rstreason = BANDLIM_UNLIMITED; 1149 goto dropwithreset; 1150 } else 1151 goto dropunlock; 1152 } 1153 /* 1154 * Socket is created in state SYN_RECEIVED. 1155 * Unlock the listen socket, lock the newly 1156 * created socket and update the tp variable. 1157 */ 1158 INP_WUNLOCK(inp); /* listen socket */ 1159 inp = sotoinpcb(so); 1160 /* 1161 * New connection inpcb is already locked by 1162 * syncache_expand(). 1163 */ 1164 INP_WLOCK_ASSERT(inp); 1165 tp = intotcpcb(inp); 1166 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1167 ("%s: ", __func__)); 1168 #ifdef TCP_SIGNATURE 1169 if (sig_checked == 0) { 1170 tcp_dooptions(&to, optp, optlen, 1171 (thflags & TH_SYN) ? TO_SYN : 0); 1172 if (!tcp_signature_verify_input(m, off0, tlen, 1173 optlen, &to, th, tp->t_flags)) { 1174 1175 /* 1176 * In SYN_SENT state if it receives an 1177 * RST, it is allowed for further 1178 * processing. 1179 */ 1180 if ((thflags & TH_RST) == 0 || 1181 (tp->t_state == TCPS_SYN_SENT) == 0) 1182 goto dropunlock; 1183 } 1184 sig_checked = 1; 1185 } 1186 #endif 1187 1188 /* 1189 * Process the segment and the data it 1190 * contains. tcp_do_segment() consumes 1191 * the mbuf chain and unlocks the inpcb. 1192 */ 1193 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1194 iptos, ti_locked); 1195 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1196 return (IPPROTO_DONE); 1197 } 1198 /* 1199 * Segment flag validation for new connection attempts: 1200 * 1201 * Our (SYN|ACK) response was rejected. 1202 * Check with syncache and remove entry to prevent 1203 * retransmits. 1204 * 1205 * NB: syncache_chkrst does its own logging of failure 1206 * causes. 1207 */ 1208 if (thflags & TH_RST) { 1209 syncache_chkrst(&inc, th); 1210 goto dropunlock; 1211 } 1212 /* 1213 * We can't do anything without SYN. 1214 */ 1215 if ((thflags & TH_SYN) == 0) { 1216 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1217 log(LOG_DEBUG, "%s; %s: Listen socket: " 1218 "SYN is missing, segment ignored\n", 1219 s, __func__); 1220 TCPSTAT_INC(tcps_badsyn); 1221 goto dropunlock; 1222 } 1223 /* 1224 * (SYN|ACK) is bogus on a listen socket. 1225 */ 1226 if (thflags & TH_ACK) { 1227 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1228 log(LOG_DEBUG, "%s; %s: Listen socket: " 1229 "SYN|ACK invalid, segment rejected\n", 1230 s, __func__); 1231 syncache_badack(&inc); /* XXX: Not needed! */ 1232 TCPSTAT_INC(tcps_badsyn); 1233 rstreason = BANDLIM_RST_OPENPORT; 1234 goto dropwithreset; 1235 } 1236 /* 1237 * If the drop_synfin option is enabled, drop all 1238 * segments with both the SYN and FIN bits set. 1239 * This prevents e.g. nmap from identifying the 1240 * TCP/IP stack. 1241 * XXX: Poor reasoning. nmap has other methods 1242 * and is constantly refining its stack detection 1243 * strategies. 1244 * XXX: This is a violation of the TCP specification 1245 * and was used by RFC1644. 1246 */ 1247 if ((thflags & TH_FIN) && V_drop_synfin) { 1248 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1249 log(LOG_DEBUG, "%s; %s: Listen socket: " 1250 "SYN|FIN segment ignored (based on " 1251 "sysctl setting)\n", s, __func__); 1252 TCPSTAT_INC(tcps_badsyn); 1253 goto dropunlock; 1254 } 1255 /* 1256 * Segment's flags are (SYN) or (SYN|FIN). 1257 * 1258 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1259 * as they do not affect the state of the TCP FSM. 1260 * The data pointed to by TH_URG and th_urp is ignored. 1261 */ 1262 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1263 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1264 KASSERT(thflags & (TH_SYN), 1265 ("%s: Listen socket: TH_SYN not set", __func__)); 1266 #ifdef INET6 1267 /* 1268 * If deprecated address is forbidden, 1269 * we do not accept SYN to deprecated interface 1270 * address to prevent any new inbound connection from 1271 * getting established. 1272 * When we do not accept SYN, we send a TCP RST, 1273 * with deprecated source address (instead of dropping 1274 * it). We compromise it as it is much better for peer 1275 * to send a RST, and RST will be the final packet 1276 * for the exchange. 1277 * 1278 * If we do not forbid deprecated addresses, we accept 1279 * the SYN packet. RFC2462 does not suggest dropping 1280 * SYN in this case. 1281 * If we decipher RFC2462 5.5.4, it says like this: 1282 * 1. use of deprecated addr with existing 1283 * communication is okay - "SHOULD continue to be 1284 * used" 1285 * 2. use of it with new communication: 1286 * (2a) "SHOULD NOT be used if alternate address 1287 * with sufficient scope is available" 1288 * (2b) nothing mentioned otherwise. 1289 * Here we fall into (2b) case as we have no choice in 1290 * our source address selection - we must obey the peer. 1291 * 1292 * The wording in RFC2462 is confusing, and there are 1293 * multiple description text for deprecated address 1294 * handling - worse, they are not exactly the same. 1295 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1296 */ 1297 if (isipv6 && !V_ip6_use_deprecated) { 1298 struct in6_ifaddr *ia6; 1299 1300 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1301 if (ia6 != NULL && 1302 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1303 ifa_free(&ia6->ia_ifa); 1304 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1305 log(LOG_DEBUG, "%s; %s: Listen socket: " 1306 "Connection attempt to deprecated " 1307 "IPv6 address rejected\n", 1308 s, __func__); 1309 rstreason = BANDLIM_RST_OPENPORT; 1310 goto dropwithreset; 1311 } 1312 if (ia6) 1313 ifa_free(&ia6->ia_ifa); 1314 } 1315 #endif /* INET6 */ 1316 /* 1317 * Basic sanity checks on incoming SYN requests: 1318 * Don't respond if the destination is a link layer 1319 * broadcast according to RFC1122 4.2.3.10, p. 104. 1320 * If it is from this socket it must be forged. 1321 * Don't respond if the source or destination is a 1322 * global or subnet broad- or multicast address. 1323 * Note that it is quite possible to receive unicast 1324 * link-layer packets with a broadcast IP address. Use 1325 * in_broadcast() to find them. 1326 */ 1327 if (m->m_flags & (M_BCAST|M_MCAST)) { 1328 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1329 log(LOG_DEBUG, "%s; %s: Listen socket: " 1330 "Connection attempt from broad- or multicast " 1331 "link layer address ignored\n", s, __func__); 1332 goto dropunlock; 1333 } 1334 #ifdef INET6 1335 if (isipv6) { 1336 if (th->th_dport == th->th_sport && 1337 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1338 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1339 log(LOG_DEBUG, "%s; %s: Listen socket: " 1340 "Connection attempt to/from self " 1341 "ignored\n", s, __func__); 1342 goto dropunlock; 1343 } 1344 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1345 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1346 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1347 log(LOG_DEBUG, "%s; %s: Listen socket: " 1348 "Connection attempt from/to multicast " 1349 "address ignored\n", s, __func__); 1350 goto dropunlock; 1351 } 1352 } 1353 #endif 1354 #if defined(INET) && defined(INET6) 1355 else 1356 #endif 1357 #ifdef INET 1358 { 1359 if (th->th_dport == th->th_sport && 1360 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1361 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1362 log(LOG_DEBUG, "%s; %s: Listen socket: " 1363 "Connection attempt from/to self " 1364 "ignored\n", s, __func__); 1365 goto dropunlock; 1366 } 1367 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1368 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1369 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1370 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1371 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1372 log(LOG_DEBUG, "%s; %s: Listen socket: " 1373 "Connection attempt from/to broad- " 1374 "or multicast address ignored\n", 1375 s, __func__); 1376 goto dropunlock; 1377 } 1378 } 1379 #endif 1380 /* 1381 * SYN appears to be valid. Create compressed TCP state 1382 * for syncache. 1383 */ 1384 #ifdef TCPDEBUG 1385 if (so->so_options & SO_DEBUG) 1386 tcp_trace(TA_INPUT, ostate, tp, 1387 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1388 #endif 1389 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 1390 tcp_dooptions(&to, optp, optlen, TO_SYN); 1391 #ifdef TCP_RFC7413 1392 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1393 goto new_tfo_socket; 1394 #else 1395 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1396 #endif 1397 /* 1398 * Entry added to syncache and mbuf consumed. 1399 * Only the listen socket is unlocked by syncache_add(). 1400 */ 1401 if (ti_locked == TI_RLOCKED) { 1402 INP_INFO_RUNLOCK(&V_tcbinfo); 1403 ti_locked = TI_UNLOCKED; 1404 } 1405 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1406 return (IPPROTO_DONE); 1407 } else if (tp->t_state == TCPS_LISTEN) { 1408 /* 1409 * When a listen socket is torn down the SO_ACCEPTCONN 1410 * flag is removed first while connections are drained 1411 * from the accept queue in a unlock/lock cycle of the 1412 * ACCEPT_LOCK, opening a race condition allowing a SYN 1413 * attempt go through unhandled. 1414 */ 1415 goto dropunlock; 1416 } 1417 1418 #ifdef TCP_SIGNATURE 1419 if (sig_checked == 0) { 1420 tcp_dooptions(&to, optp, optlen, 1421 (thflags & TH_SYN) ? TO_SYN : 0); 1422 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1423 th, tp->t_flags)) { 1424 1425 /* 1426 * In SYN_SENT state if it receives an RST, it is 1427 * allowed for further processing. 1428 */ 1429 if ((thflags & TH_RST) == 0 || 1430 (tp->t_state == TCPS_SYN_SENT) == 0) 1431 goto dropunlock; 1432 } 1433 sig_checked = 1; 1434 } 1435 #endif 1436 1437 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1438 1439 /* 1440 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1441 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1442 * the inpcb, and unlocks pcbinfo. 1443 */ 1444 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1445 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1446 return (IPPROTO_DONE); 1447 1448 dropwithreset: 1449 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1450 1451 if (ti_locked == TI_RLOCKED) { 1452 INP_INFO_RUNLOCK(&V_tcbinfo); 1453 ti_locked = TI_UNLOCKED; 1454 } 1455 #ifdef INVARIANTS 1456 else { 1457 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1458 "ti_locked: %d", __func__, ti_locked)); 1459 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1460 } 1461 #endif 1462 1463 if (inp != NULL) { 1464 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1465 INP_WUNLOCK(inp); 1466 } else 1467 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1468 m = NULL; /* mbuf chain got consumed. */ 1469 goto drop; 1470 1471 dropunlock: 1472 if (m != NULL) 1473 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1474 1475 if (ti_locked == TI_RLOCKED) { 1476 INP_INFO_RUNLOCK(&V_tcbinfo); 1477 ti_locked = TI_UNLOCKED; 1478 } 1479 #ifdef INVARIANTS 1480 else { 1481 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1482 "ti_locked: %d", __func__, ti_locked)); 1483 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1484 } 1485 #endif 1486 1487 if (inp != NULL) 1488 INP_WUNLOCK(inp); 1489 1490 drop: 1491 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1492 if (s != NULL) 1493 free(s, M_TCPLOG); 1494 if (m != NULL) 1495 m_freem(m); 1496 return (IPPROTO_DONE); 1497 } 1498 1499 void 1500 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1501 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1502 int ti_locked) 1503 { 1504 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1505 int rstreason, todrop, win; 1506 u_long tiwin; 1507 uint16_t nsegs; 1508 char *s; 1509 struct in_conninfo *inc; 1510 struct mbuf *mfree; 1511 struct tcpopt to; 1512 int tfo_syn; 1513 1514 #ifdef TCPDEBUG 1515 /* 1516 * The size of tcp_saveipgen must be the size of the max ip header, 1517 * now IPv6. 1518 */ 1519 u_char tcp_saveipgen[IP6_HDR_LEN]; 1520 struct tcphdr tcp_savetcp; 1521 short ostate = 0; 1522 #endif 1523 thflags = th->th_flags; 1524 inc = &tp->t_inpcb->inp_inc; 1525 tp->sackhint.last_sack_ack = 0; 1526 sack_changed = 0; 1527 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1528 1529 /* 1530 * If this is either a state-changing packet or current state isn't 1531 * established, we require a write lock on tcbinfo. Otherwise, we 1532 * allow the tcbinfo to be in either alocked or unlocked, as the 1533 * caller may have unnecessarily acquired a write lock due to a race. 1534 */ 1535 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1536 tp->t_state != TCPS_ESTABLISHED) { 1537 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1538 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1539 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1540 } else { 1541 #ifdef INVARIANTS 1542 if (ti_locked == TI_RLOCKED) 1543 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1544 else { 1545 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1546 "ti_locked: %d", __func__, ti_locked)); 1547 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1548 } 1549 #endif 1550 } 1551 INP_WLOCK_ASSERT(tp->t_inpcb); 1552 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1553 __func__)); 1554 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1555 __func__)); 1556 1557 #ifdef TCPPCAP 1558 /* Save segment, if requested. */ 1559 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1560 #endif 1561 1562 /* 1563 * Segment received on connection. 1564 * Reset idle time and keep-alive timer. 1565 * XXX: This should be done after segment 1566 * validation to ignore broken/spoofed segs. 1567 */ 1568 tp->t_rcvtime = ticks; 1569 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1570 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1571 1572 /* 1573 * Scale up the window into a 32-bit value. 1574 * For the SYN_SENT state the scale is zero. 1575 */ 1576 tiwin = th->th_win << tp->snd_scale; 1577 1578 /* 1579 * TCP ECN processing. 1580 */ 1581 if (tp->t_flags & TF_ECN_PERMIT) { 1582 if (thflags & TH_CWR) 1583 tp->t_flags &= ~TF_ECN_SND_ECE; 1584 switch (iptos & IPTOS_ECN_MASK) { 1585 case IPTOS_ECN_CE: 1586 tp->t_flags |= TF_ECN_SND_ECE; 1587 TCPSTAT_INC(tcps_ecn_ce); 1588 break; 1589 case IPTOS_ECN_ECT0: 1590 TCPSTAT_INC(tcps_ecn_ect0); 1591 break; 1592 case IPTOS_ECN_ECT1: 1593 TCPSTAT_INC(tcps_ecn_ect1); 1594 break; 1595 } 1596 1597 /* Process a packet differently from RFC3168. */ 1598 cc_ecnpkt_handler(tp, th, iptos); 1599 1600 /* Congestion experienced. */ 1601 if (thflags & TH_ECE) { 1602 cc_cong_signal(tp, th, CC_ECN); 1603 } 1604 } 1605 1606 /* 1607 * Parse options on any incoming segment. 1608 */ 1609 tcp_dooptions(&to, (u_char *)(th + 1), 1610 (th->th_off << 2) - sizeof(struct tcphdr), 1611 (thflags & TH_SYN) ? TO_SYN : 0); 1612 1613 /* 1614 * If echoed timestamp is later than the current time, 1615 * fall back to non RFC1323 RTT calculation. Normalize 1616 * timestamp if syncookies were used when this connection 1617 * was established. 1618 */ 1619 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1620 to.to_tsecr -= tp->ts_offset; 1621 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1622 to.to_tsecr = 0; 1623 } 1624 /* 1625 * If timestamps were negotiated during SYN/ACK they should 1626 * appear on every segment during this session and vice versa. 1627 */ 1628 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1629 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1630 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1631 "no action\n", s, __func__); 1632 free(s, M_TCPLOG); 1633 } 1634 } 1635 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1636 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1637 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1638 "no action\n", s, __func__); 1639 free(s, M_TCPLOG); 1640 } 1641 } 1642 1643 /* 1644 * Process options only when we get SYN/ACK back. The SYN case 1645 * for incoming connections is handled in tcp_syncache. 1646 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1647 * or <SYN,ACK>) segment itself is never scaled. 1648 * XXX this is traditional behavior, may need to be cleaned up. 1649 */ 1650 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1651 if ((to.to_flags & TOF_SCALE) && 1652 (tp->t_flags & TF_REQ_SCALE)) { 1653 tp->t_flags |= TF_RCVD_SCALE; 1654 tp->snd_scale = to.to_wscale; 1655 } 1656 /* 1657 * Initial send window. It will be updated with 1658 * the next incoming segment to the scaled value. 1659 */ 1660 tp->snd_wnd = th->th_win; 1661 if (to.to_flags & TOF_TS) { 1662 tp->t_flags |= TF_RCVD_TSTMP; 1663 tp->ts_recent = to.to_tsval; 1664 tp->ts_recent_age = tcp_ts_getticks(); 1665 } 1666 if (to.to_flags & TOF_MSS) 1667 tcp_mss(tp, to.to_mss); 1668 if ((tp->t_flags & TF_SACK_PERMIT) && 1669 (to.to_flags & TOF_SACKPERM) == 0) 1670 tp->t_flags &= ~TF_SACK_PERMIT; 1671 } 1672 1673 /* 1674 * Header prediction: check for the two common cases 1675 * of a uni-directional data xfer. If the packet has 1676 * no control flags, is in-sequence, the window didn't 1677 * change and we're not retransmitting, it's a 1678 * candidate. If the length is zero and the ack moved 1679 * forward, we're the sender side of the xfer. Just 1680 * free the data acked & wake any higher level process 1681 * that was blocked waiting for space. If the length 1682 * is non-zero and the ack didn't move, we're the 1683 * receiver side. If we're getting packets in-order 1684 * (the reassembly queue is empty), add the data to 1685 * the socket buffer and note that we need a delayed ack. 1686 * Make sure that the hidden state-flags are also off. 1687 * Since we check for TCPS_ESTABLISHED first, it can only 1688 * be TH_NEEDSYN. 1689 */ 1690 if (tp->t_state == TCPS_ESTABLISHED && 1691 th->th_seq == tp->rcv_nxt && 1692 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1693 tp->snd_nxt == tp->snd_max && 1694 tiwin && tiwin == tp->snd_wnd && 1695 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1696 LIST_EMPTY(&tp->t_segq) && 1697 ((to.to_flags & TOF_TS) == 0 || 1698 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1699 1700 /* 1701 * If last ACK falls within this segment's sequence numbers, 1702 * record the timestamp. 1703 * NOTE that the test is modified according to the latest 1704 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1705 */ 1706 if ((to.to_flags & TOF_TS) != 0 && 1707 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1708 tp->ts_recent_age = tcp_ts_getticks(); 1709 tp->ts_recent = to.to_tsval; 1710 } 1711 1712 if (tlen == 0) { 1713 if (SEQ_GT(th->th_ack, tp->snd_una) && 1714 SEQ_LEQ(th->th_ack, tp->snd_max) && 1715 !IN_RECOVERY(tp->t_flags) && 1716 (to.to_flags & TOF_SACK) == 0 && 1717 TAILQ_EMPTY(&tp->snd_holes)) { 1718 /* 1719 * This is a pure ack for outstanding data. 1720 */ 1721 if (ti_locked == TI_RLOCKED) 1722 INP_INFO_RUNLOCK(&V_tcbinfo); 1723 ti_locked = TI_UNLOCKED; 1724 1725 TCPSTAT_INC(tcps_predack); 1726 1727 /* 1728 * "bad retransmit" recovery. 1729 */ 1730 if (tp->t_rxtshift == 1 && 1731 tp->t_flags & TF_PREVVALID && 1732 (int)(ticks - tp->t_badrxtwin) < 0) { 1733 cc_cong_signal(tp, th, CC_RTO_ERR); 1734 } 1735 1736 /* 1737 * Recalculate the transmit timer / rtt. 1738 * 1739 * Some boxes send broken timestamp replies 1740 * during the SYN+ACK phase, ignore 1741 * timestamps of 0 or we could calculate a 1742 * huge RTT and blow up the retransmit timer. 1743 */ 1744 if ((to.to_flags & TOF_TS) != 0 && 1745 to.to_tsecr) { 1746 u_int t; 1747 1748 t = tcp_ts_getticks() - to.to_tsecr; 1749 if (!tp->t_rttlow || tp->t_rttlow > t) 1750 tp->t_rttlow = t; 1751 tcp_xmit_timer(tp, 1752 TCP_TS_TO_TICKS(t) + 1); 1753 } else if (tp->t_rtttime && 1754 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1755 if (!tp->t_rttlow || 1756 tp->t_rttlow > ticks - tp->t_rtttime) 1757 tp->t_rttlow = ticks - tp->t_rtttime; 1758 tcp_xmit_timer(tp, 1759 ticks - tp->t_rtttime); 1760 } 1761 acked = BYTES_THIS_ACK(tp, th); 1762 1763 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1764 hhook_run_tcp_est_in(tp, th, &to); 1765 1766 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1767 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1768 sbdrop(&so->so_snd, acked); 1769 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1770 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1771 tp->snd_recover = th->th_ack - 1; 1772 1773 /* 1774 * Let the congestion control algorithm update 1775 * congestion control related information. This 1776 * typically means increasing the congestion 1777 * window. 1778 */ 1779 cc_ack_received(tp, th, nsegs, CC_ACK); 1780 1781 tp->snd_una = th->th_ack; 1782 /* 1783 * Pull snd_wl2 up to prevent seq wrap relative 1784 * to th_ack. 1785 */ 1786 tp->snd_wl2 = th->th_ack; 1787 tp->t_dupacks = 0; 1788 m_freem(m); 1789 1790 /* 1791 * If all outstanding data are acked, stop 1792 * retransmit timer, otherwise restart timer 1793 * using current (possibly backed-off) value. 1794 * If process is waiting for space, 1795 * wakeup/selwakeup/signal. If data 1796 * are ready to send, let tcp_output 1797 * decide between more output or persist. 1798 */ 1799 #ifdef TCPDEBUG 1800 if (so->so_options & SO_DEBUG) 1801 tcp_trace(TA_INPUT, ostate, tp, 1802 (void *)tcp_saveipgen, 1803 &tcp_savetcp, 0); 1804 #endif 1805 TCP_PROBE3(debug__input, tp, th, 1806 mtod(m, const char *)); 1807 if (tp->snd_una == tp->snd_max) 1808 tcp_timer_activate(tp, TT_REXMT, 0); 1809 else if (!tcp_timer_active(tp, TT_PERSIST)) 1810 tcp_timer_activate(tp, TT_REXMT, 1811 tp->t_rxtcur); 1812 sowwakeup(so); 1813 if (sbavail(&so->so_snd)) 1814 (void) tp->t_fb->tfb_tcp_output(tp); 1815 goto check_delack; 1816 } 1817 } else if (th->th_ack == tp->snd_una && 1818 tlen <= sbspace(&so->so_rcv)) { 1819 int newsize = 0; /* automatic sockbuf scaling */ 1820 1821 /* 1822 * This is a pure, in-sequence data packet with 1823 * nothing on the reassembly queue and we have enough 1824 * buffer space to take it. 1825 */ 1826 if (ti_locked == TI_RLOCKED) 1827 INP_INFO_RUNLOCK(&V_tcbinfo); 1828 ti_locked = TI_UNLOCKED; 1829 1830 /* Clean receiver SACK report if present */ 1831 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1832 tcp_clean_sackreport(tp); 1833 TCPSTAT_INC(tcps_preddat); 1834 tp->rcv_nxt += tlen; 1835 /* 1836 * Pull snd_wl1 up to prevent seq wrap relative to 1837 * th_seq. 1838 */ 1839 tp->snd_wl1 = th->th_seq; 1840 /* 1841 * Pull rcv_up up to prevent seq wrap relative to 1842 * rcv_nxt. 1843 */ 1844 tp->rcv_up = tp->rcv_nxt; 1845 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1846 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1847 #ifdef TCPDEBUG 1848 if (so->so_options & SO_DEBUG) 1849 tcp_trace(TA_INPUT, ostate, tp, 1850 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1851 #endif 1852 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 1853 1854 /* 1855 * Automatic sizing of receive socket buffer. Often the send 1856 * buffer size is not optimally adjusted to the actual network 1857 * conditions at hand (delay bandwidth product). Setting the 1858 * buffer size too small limits throughput on links with high 1859 * bandwidth and high delay (eg. trans-continental/oceanic links). 1860 * 1861 * On the receive side the socket buffer memory is only rarely 1862 * used to any significant extent. This allows us to be much 1863 * more aggressive in scaling the receive socket buffer. For 1864 * the case that the buffer space is actually used to a large 1865 * extent and we run out of kernel memory we can simply drop 1866 * the new segments; TCP on the sender will just retransmit it 1867 * later. Setting the buffer size too big may only consume too 1868 * much kernel memory if the application doesn't read() from 1869 * the socket or packet loss or reordering makes use of the 1870 * reassembly queue. 1871 * 1872 * The criteria to step up the receive buffer one notch are: 1873 * 1. Application has not set receive buffer size with 1874 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1875 * 2. the number of bytes received during the time it takes 1876 * one timestamp to be reflected back to us (the RTT); 1877 * 3. received bytes per RTT is within seven eighth of the 1878 * current socket buffer size; 1879 * 4. receive buffer size has not hit maximal automatic size; 1880 * 1881 * This algorithm does one step per RTT at most and only if 1882 * we receive a bulk stream w/o packet losses or reorderings. 1883 * Shrinking the buffer during idle times is not necessary as 1884 * it doesn't consume any memory when idle. 1885 * 1886 * TODO: Only step up if the application is actually serving 1887 * the buffer to better manage the socket buffer resources. 1888 */ 1889 if (V_tcp_do_autorcvbuf && 1890 (to.to_flags & TOF_TS) && 1891 to.to_tsecr && 1892 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1893 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1894 to.to_tsecr - tp->rfbuf_ts < hz) { 1895 if (tp->rfbuf_cnt > 1896 (so->so_rcv.sb_hiwat / 8 * 7) && 1897 so->so_rcv.sb_hiwat < 1898 V_tcp_autorcvbuf_max) { 1899 newsize = 1900 min(so->so_rcv.sb_hiwat + 1901 V_tcp_autorcvbuf_inc, 1902 V_tcp_autorcvbuf_max); 1903 } 1904 /* Start over with next RTT. */ 1905 tp->rfbuf_ts = 0; 1906 tp->rfbuf_cnt = 0; 1907 } else 1908 tp->rfbuf_cnt += tlen; /* add up */ 1909 } 1910 1911 /* Add data to socket buffer. */ 1912 SOCKBUF_LOCK(&so->so_rcv); 1913 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1914 m_freem(m); 1915 } else { 1916 /* 1917 * Set new socket buffer size. 1918 * Give up when limit is reached. 1919 */ 1920 if (newsize) 1921 if (!sbreserve_locked(&so->so_rcv, 1922 newsize, so, NULL)) 1923 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1924 m_adj(m, drop_hdrlen); /* delayed header drop */ 1925 sbappendstream_locked(&so->so_rcv, m, 0); 1926 } 1927 /* NB: sorwakeup_locked() does an implicit unlock. */ 1928 sorwakeup_locked(so); 1929 if (DELAY_ACK(tp, tlen)) { 1930 tp->t_flags |= TF_DELACK; 1931 } else { 1932 tp->t_flags |= TF_ACKNOW; 1933 tp->t_fb->tfb_tcp_output(tp); 1934 } 1935 goto check_delack; 1936 } 1937 } 1938 1939 /* 1940 * Calculate amount of space in receive window, 1941 * and then do TCP input processing. 1942 * Receive window is amount of space in rcv queue, 1943 * but not less than advertised window. 1944 */ 1945 win = sbspace(&so->so_rcv); 1946 if (win < 0) 1947 win = 0; 1948 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1949 1950 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1951 tp->rfbuf_ts = 0; 1952 tp->rfbuf_cnt = 0; 1953 1954 switch (tp->t_state) { 1955 1956 /* 1957 * If the state is SYN_RECEIVED: 1958 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1959 */ 1960 case TCPS_SYN_RECEIVED: 1961 if ((thflags & TH_ACK) && 1962 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1963 SEQ_GT(th->th_ack, tp->snd_max))) { 1964 rstreason = BANDLIM_RST_OPENPORT; 1965 goto dropwithreset; 1966 } 1967 #ifdef TCP_RFC7413 1968 if (tp->t_flags & TF_FASTOPEN) { 1969 /* 1970 * When a TFO connection is in SYN_RECEIVED, the 1971 * only valid packets are the initial SYN, a 1972 * retransmit/copy of the initial SYN (possibly with 1973 * a subset of the original data), a valid ACK, a 1974 * FIN, or a RST. 1975 */ 1976 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1977 rstreason = BANDLIM_RST_OPENPORT; 1978 goto dropwithreset; 1979 } else if (thflags & TH_SYN) { 1980 /* non-initial SYN is ignored */ 1981 if ((tcp_timer_active(tp, TT_DELACK) || 1982 tcp_timer_active(tp, TT_REXMT))) 1983 goto drop; 1984 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1985 goto drop; 1986 } 1987 } 1988 #endif 1989 break; 1990 1991 /* 1992 * If the state is SYN_SENT: 1993 * if seg contains an ACK, but not for our SYN, drop the input. 1994 * if seg contains a RST, then drop the connection. 1995 * if seg does not contain SYN, then drop it. 1996 * Otherwise this is an acceptable SYN segment 1997 * initialize tp->rcv_nxt and tp->irs 1998 * if seg contains ack then advance tp->snd_una 1999 * if seg contains an ECE and ECN support is enabled, the stream 2000 * is ECN capable. 2001 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2002 * arrange for segment to be acked (eventually) 2003 * continue processing rest of data/controls, beginning with URG 2004 */ 2005 case TCPS_SYN_SENT: 2006 if ((thflags & TH_ACK) && 2007 (SEQ_LEQ(th->th_ack, tp->iss) || 2008 SEQ_GT(th->th_ack, tp->snd_max))) { 2009 rstreason = BANDLIM_UNLIMITED; 2010 goto dropwithreset; 2011 } 2012 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2013 TCP_PROBE5(connect__refused, NULL, tp, 2014 mtod(m, const char *), tp, th); 2015 tp = tcp_drop(tp, ECONNREFUSED); 2016 } 2017 if (thflags & TH_RST) 2018 goto drop; 2019 if (!(thflags & TH_SYN)) 2020 goto drop; 2021 2022 tp->irs = th->th_seq; 2023 tcp_rcvseqinit(tp); 2024 if (thflags & TH_ACK) { 2025 TCPSTAT_INC(tcps_connects); 2026 soisconnected(so); 2027 #ifdef MAC 2028 mac_socketpeer_set_from_mbuf(m, so); 2029 #endif 2030 /* Do window scaling on this connection? */ 2031 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2032 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2033 tp->rcv_scale = tp->request_r_scale; 2034 } 2035 tp->rcv_adv += imin(tp->rcv_wnd, 2036 TCP_MAXWIN << tp->rcv_scale); 2037 tp->snd_una++; /* SYN is acked */ 2038 /* 2039 * If there's data, delay ACK; if there's also a FIN 2040 * ACKNOW will be turned on later. 2041 */ 2042 if (DELAY_ACK(tp, tlen) && tlen != 0) 2043 tcp_timer_activate(tp, TT_DELACK, 2044 tcp_delacktime); 2045 else 2046 tp->t_flags |= TF_ACKNOW; 2047 2048 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2049 tp->t_flags |= TF_ECN_PERMIT; 2050 TCPSTAT_INC(tcps_ecn_shs); 2051 } 2052 2053 /* 2054 * Received <SYN,ACK> in SYN_SENT[*] state. 2055 * Transitions: 2056 * SYN_SENT --> ESTABLISHED 2057 * SYN_SENT* --> FIN_WAIT_1 2058 */ 2059 tp->t_starttime = ticks; 2060 if (tp->t_flags & TF_NEEDFIN) { 2061 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2062 tp->t_flags &= ~TF_NEEDFIN; 2063 thflags &= ~TH_SYN; 2064 } else { 2065 tcp_state_change(tp, TCPS_ESTABLISHED); 2066 TCP_PROBE5(connect__established, NULL, tp, 2067 mtod(m, const char *), tp, th); 2068 cc_conn_init(tp); 2069 tcp_timer_activate(tp, TT_KEEP, 2070 TP_KEEPIDLE(tp)); 2071 } 2072 } else { 2073 /* 2074 * Received initial SYN in SYN-SENT[*] state => 2075 * simultaneous open. 2076 * If it succeeds, connection is * half-synchronized. 2077 * Otherwise, do 3-way handshake: 2078 * SYN-SENT -> SYN-RECEIVED 2079 * SYN-SENT* -> SYN-RECEIVED* 2080 */ 2081 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2082 tcp_timer_activate(tp, TT_REXMT, 0); 2083 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2084 } 2085 2086 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2087 "ti_locked %d", __func__, ti_locked)); 2088 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2089 INP_WLOCK_ASSERT(tp->t_inpcb); 2090 2091 /* 2092 * Advance th->th_seq to correspond to first data byte. 2093 * If data, trim to stay within window, 2094 * dropping FIN if necessary. 2095 */ 2096 th->th_seq++; 2097 if (tlen > tp->rcv_wnd) { 2098 todrop = tlen - tp->rcv_wnd; 2099 m_adj(m, -todrop); 2100 tlen = tp->rcv_wnd; 2101 thflags &= ~TH_FIN; 2102 TCPSTAT_INC(tcps_rcvpackafterwin); 2103 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2104 } 2105 tp->snd_wl1 = th->th_seq - 1; 2106 tp->rcv_up = th->th_seq; 2107 /* 2108 * Client side of transaction: already sent SYN and data. 2109 * If the remote host used T/TCP to validate the SYN, 2110 * our data will be ACK'd; if so, enter normal data segment 2111 * processing in the middle of step 5, ack processing. 2112 * Otherwise, goto step 6. 2113 */ 2114 if (thflags & TH_ACK) 2115 goto process_ACK; 2116 2117 goto step6; 2118 2119 /* 2120 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2121 * do normal processing. 2122 * 2123 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2124 */ 2125 case TCPS_LAST_ACK: 2126 case TCPS_CLOSING: 2127 break; /* continue normal processing */ 2128 } 2129 2130 /* 2131 * States other than LISTEN or SYN_SENT. 2132 * First check the RST flag and sequence number since reset segments 2133 * are exempt from the timestamp and connection count tests. This 2134 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2135 * below which allowed reset segments in half the sequence space 2136 * to fall though and be processed (which gives forged reset 2137 * segments with a random sequence number a 50 percent chance of 2138 * killing a connection). 2139 * Then check timestamp, if present. 2140 * Then check the connection count, if present. 2141 * Then check that at least some bytes of segment are within 2142 * receive window. If segment begins before rcv_nxt, 2143 * drop leading data (and SYN); if nothing left, just ack. 2144 */ 2145 if (thflags & TH_RST) { 2146 /* 2147 * RFC5961 Section 3.2 2148 * 2149 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2150 * - If RST is in window, we send challenge ACK. 2151 * 2152 * Note: to take into account delayed ACKs, we should 2153 * test against last_ack_sent instead of rcv_nxt. 2154 * Note 2: we handle special case of closed window, not 2155 * covered by the RFC. 2156 */ 2157 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2158 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2159 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2160 2161 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2162 KASSERT(ti_locked == TI_RLOCKED, 2163 ("%s: TH_RST ti_locked %d, th %p tp %p", 2164 __func__, ti_locked, th, tp)); 2165 KASSERT(tp->t_state != TCPS_SYN_SENT, 2166 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2167 __func__, th, tp)); 2168 2169 if (V_tcp_insecure_rst || 2170 tp->last_ack_sent == th->th_seq) { 2171 TCPSTAT_INC(tcps_drops); 2172 /* Drop the connection. */ 2173 switch (tp->t_state) { 2174 case TCPS_SYN_RECEIVED: 2175 so->so_error = ECONNREFUSED; 2176 goto close; 2177 case TCPS_ESTABLISHED: 2178 case TCPS_FIN_WAIT_1: 2179 case TCPS_FIN_WAIT_2: 2180 case TCPS_CLOSE_WAIT: 2181 so->so_error = ECONNRESET; 2182 close: 2183 tcp_state_change(tp, TCPS_CLOSED); 2184 /* FALLTHROUGH */ 2185 default: 2186 tp = tcp_close(tp); 2187 } 2188 } else { 2189 TCPSTAT_INC(tcps_badrst); 2190 /* Send challenge ACK. */ 2191 tcp_respond(tp, mtod(m, void *), th, m, 2192 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2193 tp->last_ack_sent = tp->rcv_nxt; 2194 m = NULL; 2195 } 2196 } 2197 goto drop; 2198 } 2199 2200 /* 2201 * RFC5961 Section 4.2 2202 * Send challenge ACK for any SYN in synchronized state. 2203 */ 2204 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2205 tp->t_state != TCPS_SYN_RECEIVED) { 2206 KASSERT(ti_locked == TI_RLOCKED, 2207 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2208 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2209 2210 TCPSTAT_INC(tcps_badsyn); 2211 if (V_tcp_insecure_syn && 2212 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2213 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2214 tp = tcp_drop(tp, ECONNRESET); 2215 rstreason = BANDLIM_UNLIMITED; 2216 } else { 2217 /* Send challenge ACK. */ 2218 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2219 tp->snd_nxt, TH_ACK); 2220 tp->last_ack_sent = tp->rcv_nxt; 2221 m = NULL; 2222 } 2223 goto drop; 2224 } 2225 2226 /* 2227 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2228 * and it's less than ts_recent, drop it. 2229 */ 2230 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2231 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2232 2233 /* Check to see if ts_recent is over 24 days old. */ 2234 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2235 /* 2236 * Invalidate ts_recent. If this segment updates 2237 * ts_recent, the age will be reset later and ts_recent 2238 * will get a valid value. If it does not, setting 2239 * ts_recent to zero will at least satisfy the 2240 * requirement that zero be placed in the timestamp 2241 * echo reply when ts_recent isn't valid. The 2242 * age isn't reset until we get a valid ts_recent 2243 * because we don't want out-of-order segments to be 2244 * dropped when ts_recent is old. 2245 */ 2246 tp->ts_recent = 0; 2247 } else { 2248 TCPSTAT_INC(tcps_rcvduppack); 2249 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2250 TCPSTAT_INC(tcps_pawsdrop); 2251 if (tlen) 2252 goto dropafterack; 2253 goto drop; 2254 } 2255 } 2256 2257 /* 2258 * In the SYN-RECEIVED state, validate that the packet belongs to 2259 * this connection before trimming the data to fit the receive 2260 * window. Check the sequence number versus IRS since we know 2261 * the sequence numbers haven't wrapped. This is a partial fix 2262 * for the "LAND" DoS attack. 2263 */ 2264 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2265 rstreason = BANDLIM_RST_OPENPORT; 2266 goto dropwithreset; 2267 } 2268 2269 todrop = tp->rcv_nxt - th->th_seq; 2270 if (todrop > 0) { 2271 if (thflags & TH_SYN) { 2272 thflags &= ~TH_SYN; 2273 th->th_seq++; 2274 if (th->th_urp > 1) 2275 th->th_urp--; 2276 else 2277 thflags &= ~TH_URG; 2278 todrop--; 2279 } 2280 /* 2281 * Following if statement from Stevens, vol. 2, p. 960. 2282 */ 2283 if (todrop > tlen 2284 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2285 /* 2286 * Any valid FIN must be to the left of the window. 2287 * At this point the FIN must be a duplicate or out 2288 * of sequence; drop it. 2289 */ 2290 thflags &= ~TH_FIN; 2291 2292 /* 2293 * Send an ACK to resynchronize and drop any data. 2294 * But keep on processing for RST or ACK. 2295 */ 2296 tp->t_flags |= TF_ACKNOW; 2297 todrop = tlen; 2298 TCPSTAT_INC(tcps_rcvduppack); 2299 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2300 } else { 2301 TCPSTAT_INC(tcps_rcvpartduppack); 2302 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2303 } 2304 drop_hdrlen += todrop; /* drop from the top afterwards */ 2305 th->th_seq += todrop; 2306 tlen -= todrop; 2307 if (th->th_urp > todrop) 2308 th->th_urp -= todrop; 2309 else { 2310 thflags &= ~TH_URG; 2311 th->th_urp = 0; 2312 } 2313 } 2314 2315 /* 2316 * If new data are received on a connection after the 2317 * user processes are gone, then RST the other end. 2318 */ 2319 if ((so->so_state & SS_NOFDREF) && 2320 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2321 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2322 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2323 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2324 2325 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2326 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2327 "after socket was closed, " 2328 "sending RST and removing tcpcb\n", 2329 s, __func__, tcpstates[tp->t_state], tlen); 2330 free(s, M_TCPLOG); 2331 } 2332 tp = tcp_close(tp); 2333 TCPSTAT_INC(tcps_rcvafterclose); 2334 rstreason = BANDLIM_UNLIMITED; 2335 goto dropwithreset; 2336 } 2337 2338 /* 2339 * If segment ends after window, drop trailing data 2340 * (and PUSH and FIN); if nothing left, just ACK. 2341 */ 2342 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2343 if (todrop > 0) { 2344 TCPSTAT_INC(tcps_rcvpackafterwin); 2345 if (todrop >= tlen) { 2346 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2347 /* 2348 * If window is closed can only take segments at 2349 * window edge, and have to drop data and PUSH from 2350 * incoming segments. Continue processing, but 2351 * remember to ack. Otherwise, drop segment 2352 * and ack. 2353 */ 2354 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2355 tp->t_flags |= TF_ACKNOW; 2356 TCPSTAT_INC(tcps_rcvwinprobe); 2357 } else 2358 goto dropafterack; 2359 } else 2360 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2361 m_adj(m, -todrop); 2362 tlen -= todrop; 2363 thflags &= ~(TH_PUSH|TH_FIN); 2364 } 2365 2366 /* 2367 * If last ACK falls within this segment's sequence numbers, 2368 * record its timestamp. 2369 * NOTE: 2370 * 1) That the test incorporates suggestions from the latest 2371 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2372 * 2) That updating only on newer timestamps interferes with 2373 * our earlier PAWS tests, so this check should be solely 2374 * predicated on the sequence space of this segment. 2375 * 3) That we modify the segment boundary check to be 2376 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2377 * instead of RFC1323's 2378 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2379 * This modified check allows us to overcome RFC1323's 2380 * limitations as described in Stevens TCP/IP Illustrated 2381 * Vol. 2 p.869. In such cases, we can still calculate the 2382 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2383 */ 2384 if ((to.to_flags & TOF_TS) != 0 && 2385 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2386 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2387 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2388 tp->ts_recent_age = tcp_ts_getticks(); 2389 tp->ts_recent = to.to_tsval; 2390 } 2391 2392 /* 2393 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2394 * flag is on (half-synchronized state), then queue data for 2395 * later processing; else drop segment and return. 2396 */ 2397 if ((thflags & TH_ACK) == 0) { 2398 if (tp->t_state == TCPS_SYN_RECEIVED || 2399 (tp->t_flags & TF_NEEDSYN)) { 2400 #ifdef TCP_RFC7413 2401 if (tp->t_state == TCPS_SYN_RECEIVED && 2402 tp->t_flags & TF_FASTOPEN) { 2403 tp->snd_wnd = tiwin; 2404 cc_conn_init(tp); 2405 } 2406 #endif 2407 goto step6; 2408 } else if (tp->t_flags & TF_ACKNOW) 2409 goto dropafterack; 2410 else 2411 goto drop; 2412 } 2413 2414 /* 2415 * Ack processing. 2416 */ 2417 switch (tp->t_state) { 2418 2419 /* 2420 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2421 * ESTABLISHED state and continue processing. 2422 * The ACK was checked above. 2423 */ 2424 case TCPS_SYN_RECEIVED: 2425 2426 TCPSTAT_INC(tcps_connects); 2427 soisconnected(so); 2428 /* Do window scaling? */ 2429 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2430 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2431 tp->rcv_scale = tp->request_r_scale; 2432 tp->snd_wnd = tiwin; 2433 } 2434 /* 2435 * Make transitions: 2436 * SYN-RECEIVED -> ESTABLISHED 2437 * SYN-RECEIVED* -> FIN-WAIT-1 2438 */ 2439 tp->t_starttime = ticks; 2440 if (tp->t_flags & TF_NEEDFIN) { 2441 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2442 tp->t_flags &= ~TF_NEEDFIN; 2443 } else { 2444 tcp_state_change(tp, TCPS_ESTABLISHED); 2445 TCP_PROBE5(accept__established, NULL, tp, 2446 mtod(m, const char *), tp, th); 2447 #ifdef TCP_RFC7413 2448 if (tp->t_tfo_pending) { 2449 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2450 tp->t_tfo_pending = NULL; 2451 2452 /* 2453 * Account for the ACK of our SYN prior to 2454 * regular ACK processing below. 2455 */ 2456 tp->snd_una++; 2457 } 2458 /* 2459 * TFO connections call cc_conn_init() during SYN 2460 * processing. Calling it again here for such 2461 * connections is not harmless as it would undo the 2462 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2463 * is retransmitted. 2464 */ 2465 if (!(tp->t_flags & TF_FASTOPEN)) 2466 #endif 2467 cc_conn_init(tp); 2468 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2469 } 2470 /* 2471 * If segment contains data or ACK, will call tcp_reass() 2472 * later; if not, do so now to pass queued data to user. 2473 */ 2474 if (tlen == 0 && (thflags & TH_FIN) == 0) 2475 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2476 (struct mbuf *)0); 2477 tp->snd_wl1 = th->th_seq - 1; 2478 /* FALLTHROUGH */ 2479 2480 /* 2481 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2482 * ACKs. If the ack is in the range 2483 * tp->snd_una < th->th_ack <= tp->snd_max 2484 * then advance tp->snd_una to th->th_ack and drop 2485 * data from the retransmission queue. If this ACK reflects 2486 * more up to date window information we update our window information. 2487 */ 2488 case TCPS_ESTABLISHED: 2489 case TCPS_FIN_WAIT_1: 2490 case TCPS_FIN_WAIT_2: 2491 case TCPS_CLOSE_WAIT: 2492 case TCPS_CLOSING: 2493 case TCPS_LAST_ACK: 2494 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2495 TCPSTAT_INC(tcps_rcvacktoomuch); 2496 goto dropafterack; 2497 } 2498 if ((tp->t_flags & TF_SACK_PERMIT) && 2499 ((to.to_flags & TOF_SACK) || 2500 !TAILQ_EMPTY(&tp->snd_holes))) 2501 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2502 else 2503 /* 2504 * Reset the value so that previous (valid) value 2505 * from the last ack with SACK doesn't get used. 2506 */ 2507 tp->sackhint.sacked_bytes = 0; 2508 2509 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2510 hhook_run_tcp_est_in(tp, th, &to); 2511 2512 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2513 u_int maxseg; 2514 2515 maxseg = tcp_maxseg(tp); 2516 if (tlen == 0 && 2517 (tiwin == tp->snd_wnd || 2518 (tp->t_flags & TF_SACK_PERMIT))) { 2519 /* 2520 * If this is the first time we've seen a 2521 * FIN from the remote, this is not a 2522 * duplicate and it needs to be processed 2523 * normally. This happens during a 2524 * simultaneous close. 2525 */ 2526 if ((thflags & TH_FIN) && 2527 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2528 tp->t_dupacks = 0; 2529 break; 2530 } 2531 TCPSTAT_INC(tcps_rcvdupack); 2532 /* 2533 * If we have outstanding data (other than 2534 * a window probe), this is a completely 2535 * duplicate ack (ie, window info didn't 2536 * change and FIN isn't set), 2537 * the ack is the biggest we've 2538 * seen and we've seen exactly our rexmt 2539 * threshold of them, assume a packet 2540 * has been dropped and retransmit it. 2541 * Kludge snd_nxt & the congestion 2542 * window so we send only this one 2543 * packet. 2544 * 2545 * We know we're losing at the current 2546 * window size so do congestion avoidance 2547 * (set ssthresh to half the current window 2548 * and pull our congestion window back to 2549 * the new ssthresh). 2550 * 2551 * Dup acks mean that packets have left the 2552 * network (they're now cached at the receiver) 2553 * so bump cwnd by the amount in the receiver 2554 * to keep a constant cwnd packets in the 2555 * network. 2556 * 2557 * When using TCP ECN, notify the peer that 2558 * we reduced the cwnd. 2559 */ 2560 /* 2561 * Following 2 kinds of acks should not affect 2562 * dupack counting: 2563 * 1) Old acks 2564 * 2) Acks with SACK but without any new SACK 2565 * information in them. These could result from 2566 * any anomaly in the network like a switch 2567 * duplicating packets or a possible DoS attack. 2568 */ 2569 if (th->th_ack != tp->snd_una || 2570 ((tp->t_flags & TF_SACK_PERMIT) && 2571 !sack_changed)) 2572 break; 2573 else if (!tcp_timer_active(tp, TT_REXMT)) 2574 tp->t_dupacks = 0; 2575 else if (++tp->t_dupacks > tcprexmtthresh || 2576 IN_FASTRECOVERY(tp->t_flags)) { 2577 cc_ack_received(tp, th, nsegs, 2578 CC_DUPACK); 2579 if ((tp->t_flags & TF_SACK_PERMIT) && 2580 IN_FASTRECOVERY(tp->t_flags)) { 2581 int awnd; 2582 2583 /* 2584 * Compute the amount of data in flight first. 2585 * We can inject new data into the pipe iff 2586 * we have less than 1/2 the original window's 2587 * worth of data in flight. 2588 */ 2589 if (V_tcp_do_rfc6675_pipe) 2590 awnd = tcp_compute_pipe(tp); 2591 else 2592 awnd = (tp->snd_nxt - tp->snd_fack) + 2593 tp->sackhint.sack_bytes_rexmit; 2594 2595 if (awnd < tp->snd_ssthresh) { 2596 tp->snd_cwnd += maxseg; 2597 if (tp->snd_cwnd > tp->snd_ssthresh) 2598 tp->snd_cwnd = tp->snd_ssthresh; 2599 } 2600 } else 2601 tp->snd_cwnd += maxseg; 2602 (void) tp->t_fb->tfb_tcp_output(tp); 2603 goto drop; 2604 } else if (tp->t_dupacks == tcprexmtthresh) { 2605 tcp_seq onxt = tp->snd_nxt; 2606 2607 /* 2608 * If we're doing sack, check to 2609 * see if we're already in sack 2610 * recovery. If we're not doing sack, 2611 * check to see if we're in newreno 2612 * recovery. 2613 */ 2614 if (tp->t_flags & TF_SACK_PERMIT) { 2615 if (IN_FASTRECOVERY(tp->t_flags)) { 2616 tp->t_dupacks = 0; 2617 break; 2618 } 2619 } else { 2620 if (SEQ_LEQ(th->th_ack, 2621 tp->snd_recover)) { 2622 tp->t_dupacks = 0; 2623 break; 2624 } 2625 } 2626 /* Congestion signal before ack. */ 2627 cc_cong_signal(tp, th, CC_NDUPACK); 2628 cc_ack_received(tp, th, nsegs, 2629 CC_DUPACK); 2630 tcp_timer_activate(tp, TT_REXMT, 0); 2631 tp->t_rtttime = 0; 2632 if (tp->t_flags & TF_SACK_PERMIT) { 2633 TCPSTAT_INC( 2634 tcps_sack_recovery_episode); 2635 tp->sack_newdata = tp->snd_nxt; 2636 tp->snd_cwnd = maxseg; 2637 (void) tp->t_fb->tfb_tcp_output(tp); 2638 goto drop; 2639 } 2640 tp->snd_nxt = th->th_ack; 2641 tp->snd_cwnd = maxseg; 2642 (void) tp->t_fb->tfb_tcp_output(tp); 2643 KASSERT(tp->snd_limited <= 2, 2644 ("%s: tp->snd_limited too big", 2645 __func__)); 2646 tp->snd_cwnd = tp->snd_ssthresh + 2647 maxseg * 2648 (tp->t_dupacks - tp->snd_limited); 2649 if (SEQ_GT(onxt, tp->snd_nxt)) 2650 tp->snd_nxt = onxt; 2651 goto drop; 2652 } else if (V_tcp_do_rfc3042) { 2653 /* 2654 * Process first and second duplicate 2655 * ACKs. Each indicates a segment 2656 * leaving the network, creating room 2657 * for more. Make sure we can send a 2658 * packet on reception of each duplicate 2659 * ACK by increasing snd_cwnd by one 2660 * segment. Restore the original 2661 * snd_cwnd after packet transmission. 2662 */ 2663 cc_ack_received(tp, th, nsegs, 2664 CC_DUPACK); 2665 u_long oldcwnd = tp->snd_cwnd; 2666 tcp_seq oldsndmax = tp->snd_max; 2667 u_int sent; 2668 int avail; 2669 2670 KASSERT(tp->t_dupacks == 1 || 2671 tp->t_dupacks == 2, 2672 ("%s: dupacks not 1 or 2", 2673 __func__)); 2674 if (tp->t_dupacks == 1) 2675 tp->snd_limited = 0; 2676 tp->snd_cwnd = 2677 (tp->snd_nxt - tp->snd_una) + 2678 (tp->t_dupacks - tp->snd_limited) * 2679 maxseg; 2680 /* 2681 * Only call tcp_output when there 2682 * is new data available to be sent. 2683 * Otherwise we would send pure ACKs. 2684 */ 2685 SOCKBUF_LOCK(&so->so_snd); 2686 avail = sbavail(&so->so_snd) - 2687 (tp->snd_nxt - tp->snd_una); 2688 SOCKBUF_UNLOCK(&so->so_snd); 2689 if (avail > 0) 2690 (void) tp->t_fb->tfb_tcp_output(tp); 2691 sent = tp->snd_max - oldsndmax; 2692 if (sent > maxseg) { 2693 KASSERT((tp->t_dupacks == 2 && 2694 tp->snd_limited == 0) || 2695 (sent == maxseg + 1 && 2696 tp->t_flags & TF_SENTFIN), 2697 ("%s: sent too much", 2698 __func__)); 2699 tp->snd_limited = 2; 2700 } else if (sent > 0) 2701 ++tp->snd_limited; 2702 tp->snd_cwnd = oldcwnd; 2703 goto drop; 2704 } 2705 } 2706 break; 2707 } else { 2708 /* 2709 * This ack is advancing the left edge, reset the 2710 * counter. 2711 */ 2712 tp->t_dupacks = 0; 2713 /* 2714 * If this ack also has new SACK info, increment the 2715 * counter as per rfc6675. 2716 */ 2717 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2718 tp->t_dupacks++; 2719 } 2720 2721 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2722 ("%s: th_ack <= snd_una", __func__)); 2723 2724 /* 2725 * If the congestion window was inflated to account 2726 * for the other side's cached packets, retract it. 2727 */ 2728 if (IN_FASTRECOVERY(tp->t_flags)) { 2729 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2730 if (tp->t_flags & TF_SACK_PERMIT) 2731 tcp_sack_partialack(tp, th); 2732 else 2733 tcp_newreno_partial_ack(tp, th); 2734 } else 2735 cc_post_recovery(tp, th); 2736 } 2737 /* 2738 * If we reach this point, ACK is not a duplicate, 2739 * i.e., it ACKs something we sent. 2740 */ 2741 if (tp->t_flags & TF_NEEDSYN) { 2742 /* 2743 * T/TCP: Connection was half-synchronized, and our 2744 * SYN has been ACK'd (so connection is now fully 2745 * synchronized). Go to non-starred state, 2746 * increment snd_una for ACK of SYN, and check if 2747 * we can do window scaling. 2748 */ 2749 tp->t_flags &= ~TF_NEEDSYN; 2750 tp->snd_una++; 2751 /* Do window scaling? */ 2752 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2753 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2754 tp->rcv_scale = tp->request_r_scale; 2755 /* Send window already scaled. */ 2756 } 2757 } 2758 2759 process_ACK: 2760 INP_WLOCK_ASSERT(tp->t_inpcb); 2761 2762 acked = BYTES_THIS_ACK(tp, th); 2763 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2764 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2765 tp->snd_una, th->th_ack, tp, m)); 2766 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2767 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2768 2769 /* 2770 * If we just performed our first retransmit, and the ACK 2771 * arrives within our recovery window, then it was a mistake 2772 * to do the retransmit in the first place. Recover our 2773 * original cwnd and ssthresh, and proceed to transmit where 2774 * we left off. 2775 */ 2776 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2777 (int)(ticks - tp->t_badrxtwin) < 0) 2778 cc_cong_signal(tp, th, CC_RTO_ERR); 2779 2780 /* 2781 * If we have a timestamp reply, update smoothed 2782 * round trip time. If no timestamp is present but 2783 * transmit timer is running and timed sequence 2784 * number was acked, update smoothed round trip time. 2785 * Since we now have an rtt measurement, cancel the 2786 * timer backoff (cf., Phil Karn's retransmit alg.). 2787 * Recompute the initial retransmit timer. 2788 * 2789 * Some boxes send broken timestamp replies 2790 * during the SYN+ACK phase, ignore 2791 * timestamps of 0 or we could calculate a 2792 * huge RTT and blow up the retransmit timer. 2793 */ 2794 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2795 u_int t; 2796 2797 t = tcp_ts_getticks() - to.to_tsecr; 2798 if (!tp->t_rttlow || tp->t_rttlow > t) 2799 tp->t_rttlow = t; 2800 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2801 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2802 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2803 tp->t_rttlow = ticks - tp->t_rtttime; 2804 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2805 } 2806 2807 /* 2808 * If all outstanding data is acked, stop retransmit 2809 * timer and remember to restart (more output or persist). 2810 * If there is more data to be acked, restart retransmit 2811 * timer, using current (possibly backed-off) value. 2812 */ 2813 if (th->th_ack == tp->snd_max) { 2814 tcp_timer_activate(tp, TT_REXMT, 0); 2815 needoutput = 1; 2816 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2817 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2818 2819 /* 2820 * If no data (only SYN) was ACK'd, 2821 * skip rest of ACK processing. 2822 */ 2823 if (acked == 0) 2824 goto step6; 2825 2826 /* 2827 * Let the congestion control algorithm update congestion 2828 * control related information. This typically means increasing 2829 * the congestion window. 2830 */ 2831 cc_ack_received(tp, th, nsegs, CC_ACK); 2832 2833 SOCKBUF_LOCK(&so->so_snd); 2834 if (acked > sbavail(&so->so_snd)) { 2835 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2836 tp->snd_wnd -= sbavail(&so->so_snd); 2837 else 2838 tp->snd_wnd = 0; 2839 mfree = sbcut_locked(&so->so_snd, 2840 (int)sbavail(&so->so_snd)); 2841 ourfinisacked = 1; 2842 } else { 2843 mfree = sbcut_locked(&so->so_snd, acked); 2844 if (tp->snd_wnd >= (u_long) acked) 2845 tp->snd_wnd -= acked; 2846 else 2847 tp->snd_wnd = 0; 2848 ourfinisacked = 0; 2849 } 2850 /* NB: sowwakeup_locked() does an implicit unlock. */ 2851 sowwakeup_locked(so); 2852 m_freem(mfree); 2853 /* Detect una wraparound. */ 2854 if (!IN_RECOVERY(tp->t_flags) && 2855 SEQ_GT(tp->snd_una, tp->snd_recover) && 2856 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2857 tp->snd_recover = th->th_ack - 1; 2858 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2859 if (IN_RECOVERY(tp->t_flags) && 2860 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2861 EXIT_RECOVERY(tp->t_flags); 2862 } 2863 tp->snd_una = th->th_ack; 2864 if (tp->t_flags & TF_SACK_PERMIT) { 2865 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2866 tp->snd_recover = tp->snd_una; 2867 } 2868 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2869 tp->snd_nxt = tp->snd_una; 2870 2871 switch (tp->t_state) { 2872 2873 /* 2874 * In FIN_WAIT_1 STATE in addition to the processing 2875 * for the ESTABLISHED state if our FIN is now acknowledged 2876 * then enter FIN_WAIT_2. 2877 */ 2878 case TCPS_FIN_WAIT_1: 2879 if (ourfinisacked) { 2880 /* 2881 * If we can't receive any more 2882 * data, then closing user can proceed. 2883 * Starting the timer is contrary to the 2884 * specification, but if we don't get a FIN 2885 * we'll hang forever. 2886 * 2887 * XXXjl: 2888 * we should release the tp also, and use a 2889 * compressed state. 2890 */ 2891 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2892 soisdisconnected(so); 2893 tcp_timer_activate(tp, TT_2MSL, 2894 (tcp_fast_finwait2_recycle ? 2895 tcp_finwait2_timeout : 2896 TP_MAXIDLE(tp))); 2897 } 2898 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2899 } 2900 break; 2901 2902 /* 2903 * In CLOSING STATE in addition to the processing for 2904 * the ESTABLISHED state if the ACK acknowledges our FIN 2905 * then enter the TIME-WAIT state, otherwise ignore 2906 * the segment. 2907 */ 2908 case TCPS_CLOSING: 2909 if (ourfinisacked) { 2910 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2911 tcp_twstart(tp); 2912 INP_INFO_RUNLOCK(&V_tcbinfo); 2913 m_freem(m); 2914 return; 2915 } 2916 break; 2917 2918 /* 2919 * In LAST_ACK, we may still be waiting for data to drain 2920 * and/or to be acked, as well as for the ack of our FIN. 2921 * If our FIN is now acknowledged, delete the TCB, 2922 * enter the closed state and return. 2923 */ 2924 case TCPS_LAST_ACK: 2925 if (ourfinisacked) { 2926 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2927 tp = tcp_close(tp); 2928 goto drop; 2929 } 2930 break; 2931 } 2932 } 2933 2934 step6: 2935 INP_WLOCK_ASSERT(tp->t_inpcb); 2936 2937 /* 2938 * Update window information. 2939 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2940 */ 2941 if ((thflags & TH_ACK) && 2942 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2943 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2944 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2945 /* keep track of pure window updates */ 2946 if (tlen == 0 && 2947 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2948 TCPSTAT_INC(tcps_rcvwinupd); 2949 tp->snd_wnd = tiwin; 2950 tp->snd_wl1 = th->th_seq; 2951 tp->snd_wl2 = th->th_ack; 2952 if (tp->snd_wnd > tp->max_sndwnd) 2953 tp->max_sndwnd = tp->snd_wnd; 2954 needoutput = 1; 2955 } 2956 2957 /* 2958 * Process segments with URG. 2959 */ 2960 if ((thflags & TH_URG) && th->th_urp && 2961 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2962 /* 2963 * This is a kludge, but if we receive and accept 2964 * random urgent pointers, we'll crash in 2965 * soreceive. It's hard to imagine someone 2966 * actually wanting to send this much urgent data. 2967 */ 2968 SOCKBUF_LOCK(&so->so_rcv); 2969 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2970 th->th_urp = 0; /* XXX */ 2971 thflags &= ~TH_URG; /* XXX */ 2972 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2973 goto dodata; /* XXX */ 2974 } 2975 /* 2976 * If this segment advances the known urgent pointer, 2977 * then mark the data stream. This should not happen 2978 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2979 * a FIN has been received from the remote side. 2980 * In these states we ignore the URG. 2981 * 2982 * According to RFC961 (Assigned Protocols), 2983 * the urgent pointer points to the last octet 2984 * of urgent data. We continue, however, 2985 * to consider it to indicate the first octet 2986 * of data past the urgent section as the original 2987 * spec states (in one of two places). 2988 */ 2989 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2990 tp->rcv_up = th->th_seq + th->th_urp; 2991 so->so_oobmark = sbavail(&so->so_rcv) + 2992 (tp->rcv_up - tp->rcv_nxt) - 1; 2993 if (so->so_oobmark == 0) 2994 so->so_rcv.sb_state |= SBS_RCVATMARK; 2995 sohasoutofband(so); 2996 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2997 } 2998 SOCKBUF_UNLOCK(&so->so_rcv); 2999 /* 3000 * Remove out of band data so doesn't get presented to user. 3001 * This can happen independent of advancing the URG pointer, 3002 * but if two URG's are pending at once, some out-of-band 3003 * data may creep in... ick. 3004 */ 3005 if (th->th_urp <= (u_long)tlen && 3006 !(so->so_options & SO_OOBINLINE)) { 3007 /* hdr drop is delayed */ 3008 tcp_pulloutofband(so, th, m, drop_hdrlen); 3009 } 3010 } else { 3011 /* 3012 * If no out of band data is expected, 3013 * pull receive urgent pointer along 3014 * with the receive window. 3015 */ 3016 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3017 tp->rcv_up = tp->rcv_nxt; 3018 } 3019 dodata: /* XXX */ 3020 INP_WLOCK_ASSERT(tp->t_inpcb); 3021 3022 /* 3023 * Process the segment text, merging it into the TCP sequencing queue, 3024 * and arranging for acknowledgment of receipt if necessary. 3025 * This process logically involves adjusting tp->rcv_wnd as data 3026 * is presented to the user (this happens in tcp_usrreq.c, 3027 * case PRU_RCVD). If a FIN has already been received on this 3028 * connection then we just ignore the text. 3029 */ 3030 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3031 (tp->t_flags & TF_FASTOPEN)); 3032 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3033 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3034 tcp_seq save_start = th->th_seq; 3035 m_adj(m, drop_hdrlen); /* delayed header drop */ 3036 /* 3037 * Insert segment which includes th into TCP reassembly queue 3038 * with control block tp. Set thflags to whether reassembly now 3039 * includes a segment with FIN. This handles the common case 3040 * inline (segment is the next to be received on an established 3041 * connection, and the queue is empty), avoiding linkage into 3042 * and removal from the queue and repetition of various 3043 * conversions. 3044 * Set DELACK for segments received in order, but ack 3045 * immediately when segments are out of order (so 3046 * fast retransmit can work). 3047 */ 3048 if (th->th_seq == tp->rcv_nxt && 3049 LIST_EMPTY(&tp->t_segq) && 3050 (TCPS_HAVEESTABLISHED(tp->t_state) || 3051 tfo_syn)) { 3052 if (DELAY_ACK(tp, tlen) || tfo_syn) 3053 tp->t_flags |= TF_DELACK; 3054 else 3055 tp->t_flags |= TF_ACKNOW; 3056 tp->rcv_nxt += tlen; 3057 thflags = th->th_flags & TH_FIN; 3058 TCPSTAT_INC(tcps_rcvpack); 3059 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3060 SOCKBUF_LOCK(&so->so_rcv); 3061 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3062 m_freem(m); 3063 else 3064 sbappendstream_locked(&so->so_rcv, m, 0); 3065 /* NB: sorwakeup_locked() does an implicit unlock. */ 3066 sorwakeup_locked(so); 3067 } else { 3068 /* 3069 * XXX: Due to the header drop above "th" is 3070 * theoretically invalid by now. Fortunately 3071 * m_adj() doesn't actually frees any mbufs 3072 * when trimming from the head. 3073 */ 3074 thflags = tcp_reass(tp, th, &tlen, m); 3075 tp->t_flags |= TF_ACKNOW; 3076 } 3077 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3078 tcp_update_sack_list(tp, save_start, save_start + tlen); 3079 #if 0 3080 /* 3081 * Note the amount of data that peer has sent into 3082 * our window, in order to estimate the sender's 3083 * buffer size. 3084 * XXX: Unused. 3085 */ 3086 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3087 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3088 else 3089 len = so->so_rcv.sb_hiwat; 3090 #endif 3091 } else { 3092 m_freem(m); 3093 thflags &= ~TH_FIN; 3094 } 3095 3096 /* 3097 * If FIN is received ACK the FIN and let the user know 3098 * that the connection is closing. 3099 */ 3100 if (thflags & TH_FIN) { 3101 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3102 socantrcvmore(so); 3103 /* 3104 * If connection is half-synchronized 3105 * (ie NEEDSYN flag on) then delay ACK, 3106 * so it may be piggybacked when SYN is sent. 3107 * Otherwise, since we received a FIN then no 3108 * more input can be expected, send ACK now. 3109 */ 3110 if (tp->t_flags & TF_NEEDSYN) 3111 tp->t_flags |= TF_DELACK; 3112 else 3113 tp->t_flags |= TF_ACKNOW; 3114 tp->rcv_nxt++; 3115 } 3116 switch (tp->t_state) { 3117 3118 /* 3119 * In SYN_RECEIVED and ESTABLISHED STATES 3120 * enter the CLOSE_WAIT state. 3121 */ 3122 case TCPS_SYN_RECEIVED: 3123 tp->t_starttime = ticks; 3124 /* FALLTHROUGH */ 3125 case TCPS_ESTABLISHED: 3126 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3127 break; 3128 3129 /* 3130 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3131 * enter the CLOSING state. 3132 */ 3133 case TCPS_FIN_WAIT_1: 3134 tcp_state_change(tp, TCPS_CLOSING); 3135 break; 3136 3137 /* 3138 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3139 * starting the time-wait timer, turning off the other 3140 * standard timers. 3141 */ 3142 case TCPS_FIN_WAIT_2: 3143 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3144 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3145 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3146 ti_locked)); 3147 3148 tcp_twstart(tp); 3149 INP_INFO_RUNLOCK(&V_tcbinfo); 3150 return; 3151 } 3152 } 3153 if (ti_locked == TI_RLOCKED) 3154 INP_INFO_RUNLOCK(&V_tcbinfo); 3155 ti_locked = TI_UNLOCKED; 3156 3157 #ifdef TCPDEBUG 3158 if (so->so_options & SO_DEBUG) 3159 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3160 &tcp_savetcp, 0); 3161 #endif 3162 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 3163 3164 /* 3165 * Return any desired output. 3166 */ 3167 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3168 (void) tp->t_fb->tfb_tcp_output(tp); 3169 3170 check_delack: 3171 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3172 __func__, ti_locked)); 3173 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3174 INP_WLOCK_ASSERT(tp->t_inpcb); 3175 3176 if (tp->t_flags & TF_DELACK) { 3177 tp->t_flags &= ~TF_DELACK; 3178 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3179 } 3180 INP_WUNLOCK(tp->t_inpcb); 3181 return; 3182 3183 dropafterack: 3184 /* 3185 * Generate an ACK dropping incoming segment if it occupies 3186 * sequence space, where the ACK reflects our state. 3187 * 3188 * We can now skip the test for the RST flag since all 3189 * paths to this code happen after packets containing 3190 * RST have been dropped. 3191 * 3192 * In the SYN-RECEIVED state, don't send an ACK unless the 3193 * segment we received passes the SYN-RECEIVED ACK test. 3194 * If it fails send a RST. This breaks the loop in the 3195 * "LAND" DoS attack, and also prevents an ACK storm 3196 * between two listening ports that have been sent forged 3197 * SYN segments, each with the source address of the other. 3198 */ 3199 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3200 (SEQ_GT(tp->snd_una, th->th_ack) || 3201 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3202 rstreason = BANDLIM_RST_OPENPORT; 3203 goto dropwithreset; 3204 } 3205 #ifdef TCPDEBUG 3206 if (so->so_options & SO_DEBUG) 3207 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3208 &tcp_savetcp, 0); 3209 #endif 3210 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 3211 if (ti_locked == TI_RLOCKED) 3212 INP_INFO_RUNLOCK(&V_tcbinfo); 3213 ti_locked = TI_UNLOCKED; 3214 3215 tp->t_flags |= TF_ACKNOW; 3216 (void) tp->t_fb->tfb_tcp_output(tp); 3217 INP_WUNLOCK(tp->t_inpcb); 3218 m_freem(m); 3219 return; 3220 3221 dropwithreset: 3222 if (ti_locked == TI_RLOCKED) 3223 INP_INFO_RUNLOCK(&V_tcbinfo); 3224 ti_locked = TI_UNLOCKED; 3225 3226 if (tp != NULL) { 3227 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3228 INP_WUNLOCK(tp->t_inpcb); 3229 } else 3230 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3231 return; 3232 3233 drop: 3234 if (ti_locked == TI_RLOCKED) { 3235 INP_INFO_RUNLOCK(&V_tcbinfo); 3236 ti_locked = TI_UNLOCKED; 3237 } 3238 #ifdef INVARIANTS 3239 else 3240 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3241 #endif 3242 3243 /* 3244 * Drop space held by incoming segment and return. 3245 */ 3246 #ifdef TCPDEBUG 3247 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3248 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3249 &tcp_savetcp, 0); 3250 #endif 3251 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 3252 if (tp != NULL) 3253 INP_WUNLOCK(tp->t_inpcb); 3254 m_freem(m); 3255 } 3256 3257 /* 3258 * Issue RST and make ACK acceptable to originator of segment. 3259 * The mbuf must still include the original packet header. 3260 * tp may be NULL. 3261 */ 3262 void 3263 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3264 int tlen, int rstreason) 3265 { 3266 #ifdef INET 3267 struct ip *ip; 3268 #endif 3269 #ifdef INET6 3270 struct ip6_hdr *ip6; 3271 #endif 3272 3273 if (tp != NULL) { 3274 INP_WLOCK_ASSERT(tp->t_inpcb); 3275 } 3276 3277 /* Don't bother if destination was broadcast/multicast. */ 3278 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3279 goto drop; 3280 #ifdef INET6 3281 if (mtod(m, struct ip *)->ip_v == 6) { 3282 ip6 = mtod(m, struct ip6_hdr *); 3283 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3284 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3285 goto drop; 3286 /* IPv6 anycast check is done at tcp6_input() */ 3287 } 3288 #endif 3289 #if defined(INET) && defined(INET6) 3290 else 3291 #endif 3292 #ifdef INET 3293 { 3294 ip = mtod(m, struct ip *); 3295 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3296 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3297 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3298 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3299 goto drop; 3300 } 3301 #endif 3302 3303 /* Perform bandwidth limiting. */ 3304 if (badport_bandlim(rstreason) < 0) 3305 goto drop; 3306 3307 /* tcp_respond consumes the mbuf chain. */ 3308 if (th->th_flags & TH_ACK) { 3309 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3310 th->th_ack, TH_RST); 3311 } else { 3312 if (th->th_flags & TH_SYN) 3313 tlen++; 3314 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3315 (tcp_seq)0, TH_RST|TH_ACK); 3316 } 3317 return; 3318 drop: 3319 m_freem(m); 3320 } 3321 3322 /* 3323 * Parse TCP options and place in tcpopt. 3324 */ 3325 void 3326 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3327 { 3328 int opt, optlen; 3329 3330 to->to_flags = 0; 3331 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3332 opt = cp[0]; 3333 if (opt == TCPOPT_EOL) 3334 break; 3335 if (opt == TCPOPT_NOP) 3336 optlen = 1; 3337 else { 3338 if (cnt < 2) 3339 break; 3340 optlen = cp[1]; 3341 if (optlen < 2 || optlen > cnt) 3342 break; 3343 } 3344 switch (opt) { 3345 case TCPOPT_MAXSEG: 3346 if (optlen != TCPOLEN_MAXSEG) 3347 continue; 3348 if (!(flags & TO_SYN)) 3349 continue; 3350 to->to_flags |= TOF_MSS; 3351 bcopy((char *)cp + 2, 3352 (char *)&to->to_mss, sizeof(to->to_mss)); 3353 to->to_mss = ntohs(to->to_mss); 3354 break; 3355 case TCPOPT_WINDOW: 3356 if (optlen != TCPOLEN_WINDOW) 3357 continue; 3358 if (!(flags & TO_SYN)) 3359 continue; 3360 to->to_flags |= TOF_SCALE; 3361 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3362 break; 3363 case TCPOPT_TIMESTAMP: 3364 if (optlen != TCPOLEN_TIMESTAMP) 3365 continue; 3366 to->to_flags |= TOF_TS; 3367 bcopy((char *)cp + 2, 3368 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3369 to->to_tsval = ntohl(to->to_tsval); 3370 bcopy((char *)cp + 6, 3371 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3372 to->to_tsecr = ntohl(to->to_tsecr); 3373 break; 3374 #ifdef TCP_SIGNATURE 3375 /* 3376 * XXX In order to reply to a host which has set the 3377 * TCP_SIGNATURE option in its initial SYN, we have to 3378 * record the fact that the option was observed here 3379 * for the syncache code to perform the correct response. 3380 */ 3381 case TCPOPT_SIGNATURE: 3382 if (optlen != TCPOLEN_SIGNATURE) 3383 continue; 3384 to->to_flags |= TOF_SIGNATURE; 3385 to->to_signature = cp + 2; 3386 break; 3387 #endif 3388 case TCPOPT_SACK_PERMITTED: 3389 if (optlen != TCPOLEN_SACK_PERMITTED) 3390 continue; 3391 if (!(flags & TO_SYN)) 3392 continue; 3393 if (!V_tcp_do_sack) 3394 continue; 3395 to->to_flags |= TOF_SACKPERM; 3396 break; 3397 case TCPOPT_SACK: 3398 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3399 continue; 3400 if (flags & TO_SYN) 3401 continue; 3402 to->to_flags |= TOF_SACK; 3403 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3404 to->to_sacks = cp + 2; 3405 TCPSTAT_INC(tcps_sack_rcv_blocks); 3406 break; 3407 #ifdef TCP_RFC7413 3408 case TCPOPT_FAST_OPEN: 3409 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3410 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3411 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3412 continue; 3413 if (!(flags & TO_SYN)) 3414 continue; 3415 if (!V_tcp_fastopen_enabled) 3416 continue; 3417 to->to_flags |= TOF_FASTOPEN; 3418 to->to_tfo_len = optlen - 2; 3419 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3420 break; 3421 #endif 3422 default: 3423 continue; 3424 } 3425 } 3426 } 3427 3428 /* 3429 * Pull out of band byte out of a segment so 3430 * it doesn't appear in the user's data queue. 3431 * It is still reflected in the segment length for 3432 * sequencing purposes. 3433 */ 3434 void 3435 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3436 int off) 3437 { 3438 int cnt = off + th->th_urp - 1; 3439 3440 while (cnt >= 0) { 3441 if (m->m_len > cnt) { 3442 char *cp = mtod(m, caddr_t) + cnt; 3443 struct tcpcb *tp = sototcpcb(so); 3444 3445 INP_WLOCK_ASSERT(tp->t_inpcb); 3446 3447 tp->t_iobc = *cp; 3448 tp->t_oobflags |= TCPOOB_HAVEDATA; 3449 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3450 m->m_len--; 3451 if (m->m_flags & M_PKTHDR) 3452 m->m_pkthdr.len--; 3453 return; 3454 } 3455 cnt -= m->m_len; 3456 m = m->m_next; 3457 if (m == NULL) 3458 break; 3459 } 3460 panic("tcp_pulloutofband"); 3461 } 3462 3463 /* 3464 * Collect new round-trip time estimate 3465 * and update averages and current timeout. 3466 */ 3467 void 3468 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3469 { 3470 int delta; 3471 3472 INP_WLOCK_ASSERT(tp->t_inpcb); 3473 3474 TCPSTAT_INC(tcps_rttupdated); 3475 tp->t_rttupdated++; 3476 if (tp->t_srtt != 0) { 3477 /* 3478 * srtt is stored as fixed point with 5 bits after the 3479 * binary point (i.e., scaled by 8). The following magic 3480 * is equivalent to the smoothing algorithm in rfc793 with 3481 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3482 * point). Adjust rtt to origin 0. 3483 */ 3484 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3485 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3486 3487 if ((tp->t_srtt += delta) <= 0) 3488 tp->t_srtt = 1; 3489 3490 /* 3491 * We accumulate a smoothed rtt variance (actually, a 3492 * smoothed mean difference), then set the retransmit 3493 * timer to smoothed rtt + 4 times the smoothed variance. 3494 * rttvar is stored as fixed point with 4 bits after the 3495 * binary point (scaled by 16). The following is 3496 * equivalent to rfc793 smoothing with an alpha of .75 3497 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3498 * rfc793's wired-in beta. 3499 */ 3500 if (delta < 0) 3501 delta = -delta; 3502 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3503 if ((tp->t_rttvar += delta) <= 0) 3504 tp->t_rttvar = 1; 3505 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3506 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3507 } else { 3508 /* 3509 * No rtt measurement yet - use the unsmoothed rtt. 3510 * Set the variance to half the rtt (so our first 3511 * retransmit happens at 3*rtt). 3512 */ 3513 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3514 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3515 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3516 } 3517 tp->t_rtttime = 0; 3518 tp->t_rxtshift = 0; 3519 3520 /* 3521 * the retransmit should happen at rtt + 4 * rttvar. 3522 * Because of the way we do the smoothing, srtt and rttvar 3523 * will each average +1/2 tick of bias. When we compute 3524 * the retransmit timer, we want 1/2 tick of rounding and 3525 * 1 extra tick because of +-1/2 tick uncertainty in the 3526 * firing of the timer. The bias will give us exactly the 3527 * 1.5 tick we need. But, because the bias is 3528 * statistical, we have to test that we don't drop below 3529 * the minimum feasible timer (which is 2 ticks). 3530 */ 3531 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3532 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3533 3534 /* 3535 * We received an ack for a packet that wasn't retransmitted; 3536 * it is probably safe to discard any error indications we've 3537 * received recently. This isn't quite right, but close enough 3538 * for now (a route might have failed after we sent a segment, 3539 * and the return path might not be symmetrical). 3540 */ 3541 tp->t_softerror = 0; 3542 } 3543 3544 /* 3545 * Determine a reasonable value for maxseg size. 3546 * If the route is known, check route for mtu. 3547 * If none, use an mss that can be handled on the outgoing interface 3548 * without forcing IP to fragment. If no route is found, route has no mtu, 3549 * or the destination isn't local, use a default, hopefully conservative 3550 * size (usually 512 or the default IP max size, but no more than the mtu 3551 * of the interface), as we can't discover anything about intervening 3552 * gateways or networks. We also initialize the congestion/slow start 3553 * window to be a single segment if the destination isn't local. 3554 * While looking at the routing entry, we also initialize other path-dependent 3555 * parameters from pre-set or cached values in the routing entry. 3556 * 3557 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3558 * IP options, e.g. IPSEC data, since length of this data may vary, and 3559 * thus it is calculated for every segment separately in tcp_output(). 3560 * 3561 * NOTE that this routine is only called when we process an incoming 3562 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3563 * settings are handled in tcp_mssopt(). 3564 */ 3565 void 3566 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3567 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3568 { 3569 int mss = 0; 3570 u_long maxmtu = 0; 3571 struct inpcb *inp = tp->t_inpcb; 3572 struct hc_metrics_lite metrics; 3573 #ifdef INET6 3574 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3575 size_t min_protoh = isipv6 ? 3576 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3577 sizeof (struct tcpiphdr); 3578 #else 3579 const size_t min_protoh = sizeof(struct tcpiphdr); 3580 #endif 3581 3582 INP_WLOCK_ASSERT(tp->t_inpcb); 3583 3584 if (mtuoffer != -1) { 3585 KASSERT(offer == -1, ("%s: conflict", __func__)); 3586 offer = mtuoffer - min_protoh; 3587 } 3588 3589 /* Initialize. */ 3590 #ifdef INET6 3591 if (isipv6) { 3592 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3593 tp->t_maxseg = V_tcp_v6mssdflt; 3594 } 3595 #endif 3596 #if defined(INET) && defined(INET6) 3597 else 3598 #endif 3599 #ifdef INET 3600 { 3601 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3602 tp->t_maxseg = V_tcp_mssdflt; 3603 } 3604 #endif 3605 3606 /* 3607 * No route to sender, stay with default mss and return. 3608 */ 3609 if (maxmtu == 0) { 3610 /* 3611 * In case we return early we need to initialize metrics 3612 * to a defined state as tcp_hc_get() would do for us 3613 * if there was no cache hit. 3614 */ 3615 if (metricptr != NULL) 3616 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3617 return; 3618 } 3619 3620 /* What have we got? */ 3621 switch (offer) { 3622 case 0: 3623 /* 3624 * Offer == 0 means that there was no MSS on the SYN 3625 * segment, in this case we use tcp_mssdflt as 3626 * already assigned to t_maxseg above. 3627 */ 3628 offer = tp->t_maxseg; 3629 break; 3630 3631 case -1: 3632 /* 3633 * Offer == -1 means that we didn't receive SYN yet. 3634 */ 3635 /* FALLTHROUGH */ 3636 3637 default: 3638 /* 3639 * Prevent DoS attack with too small MSS. Round up 3640 * to at least minmss. 3641 */ 3642 offer = max(offer, V_tcp_minmss); 3643 } 3644 3645 /* 3646 * rmx information is now retrieved from tcp_hostcache. 3647 */ 3648 tcp_hc_get(&inp->inp_inc, &metrics); 3649 if (metricptr != NULL) 3650 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3651 3652 /* 3653 * If there's a discovered mtu in tcp hostcache, use it. 3654 * Else, use the link mtu. 3655 */ 3656 if (metrics.rmx_mtu) 3657 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3658 else { 3659 #ifdef INET6 3660 if (isipv6) { 3661 mss = maxmtu - min_protoh; 3662 if (!V_path_mtu_discovery && 3663 !in6_localaddr(&inp->in6p_faddr)) 3664 mss = min(mss, V_tcp_v6mssdflt); 3665 } 3666 #endif 3667 #if defined(INET) && defined(INET6) 3668 else 3669 #endif 3670 #ifdef INET 3671 { 3672 mss = maxmtu - min_protoh; 3673 if (!V_path_mtu_discovery && 3674 !in_localaddr(inp->inp_faddr)) 3675 mss = min(mss, V_tcp_mssdflt); 3676 } 3677 #endif 3678 /* 3679 * XXX - The above conditional (mss = maxmtu - min_protoh) 3680 * probably violates the TCP spec. 3681 * The problem is that, since we don't know the 3682 * other end's MSS, we are supposed to use a conservative 3683 * default. But, if we do that, then MTU discovery will 3684 * never actually take place, because the conservative 3685 * default is much less than the MTUs typically seen 3686 * on the Internet today. For the moment, we'll sweep 3687 * this under the carpet. 3688 * 3689 * The conservative default might not actually be a problem 3690 * if the only case this occurs is when sending an initial 3691 * SYN with options and data to a host we've never talked 3692 * to before. Then, they will reply with an MSS value which 3693 * will get recorded and the new parameters should get 3694 * recomputed. For Further Study. 3695 */ 3696 } 3697 mss = min(mss, offer); 3698 3699 /* 3700 * Sanity check: make sure that maxseg will be large 3701 * enough to allow some data on segments even if the 3702 * all the option space is used (40bytes). Otherwise 3703 * funny things may happen in tcp_output. 3704 * 3705 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3706 */ 3707 mss = max(mss, 64); 3708 3709 tp->t_maxseg = mss; 3710 } 3711 3712 void 3713 tcp_mss(struct tcpcb *tp, int offer) 3714 { 3715 int mss; 3716 u_long bufsize; 3717 struct inpcb *inp; 3718 struct socket *so; 3719 struct hc_metrics_lite metrics; 3720 struct tcp_ifcap cap; 3721 3722 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3723 3724 bzero(&cap, sizeof(cap)); 3725 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3726 3727 mss = tp->t_maxseg; 3728 inp = tp->t_inpcb; 3729 3730 /* 3731 * If there's a pipesize, change the socket buffer to that size, 3732 * don't change if sb_hiwat is different than default (then it 3733 * has been changed on purpose with setsockopt). 3734 * Make the socket buffers an integral number of mss units; 3735 * if the mss is larger than the socket buffer, decrease the mss. 3736 */ 3737 so = inp->inp_socket; 3738 SOCKBUF_LOCK(&so->so_snd); 3739 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3740 bufsize = metrics.rmx_sendpipe; 3741 else 3742 bufsize = so->so_snd.sb_hiwat; 3743 if (bufsize < mss) 3744 mss = bufsize; 3745 else { 3746 bufsize = roundup(bufsize, mss); 3747 if (bufsize > sb_max) 3748 bufsize = sb_max; 3749 if (bufsize > so->so_snd.sb_hiwat) 3750 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3751 } 3752 SOCKBUF_UNLOCK(&so->so_snd); 3753 tp->t_maxseg = mss; 3754 3755 SOCKBUF_LOCK(&so->so_rcv); 3756 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3757 bufsize = metrics.rmx_recvpipe; 3758 else 3759 bufsize = so->so_rcv.sb_hiwat; 3760 if (bufsize > mss) { 3761 bufsize = roundup(bufsize, mss); 3762 if (bufsize > sb_max) 3763 bufsize = sb_max; 3764 if (bufsize > so->so_rcv.sb_hiwat) 3765 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3766 } 3767 SOCKBUF_UNLOCK(&so->so_rcv); 3768 3769 /* Check the interface for TSO capabilities. */ 3770 if (cap.ifcap & CSUM_TSO) { 3771 tp->t_flags |= TF_TSO; 3772 tp->t_tsomax = cap.tsomax; 3773 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3774 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3775 } 3776 } 3777 3778 /* 3779 * Determine the MSS option to send on an outgoing SYN. 3780 */ 3781 int 3782 tcp_mssopt(struct in_conninfo *inc) 3783 { 3784 int mss = 0; 3785 u_long maxmtu = 0; 3786 u_long thcmtu = 0; 3787 size_t min_protoh; 3788 3789 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3790 3791 #ifdef INET6 3792 if (inc->inc_flags & INC_ISIPV6) { 3793 mss = V_tcp_v6mssdflt; 3794 maxmtu = tcp_maxmtu6(inc, NULL); 3795 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3796 } 3797 #endif 3798 #if defined(INET) && defined(INET6) 3799 else 3800 #endif 3801 #ifdef INET 3802 { 3803 mss = V_tcp_mssdflt; 3804 maxmtu = tcp_maxmtu(inc, NULL); 3805 min_protoh = sizeof(struct tcpiphdr); 3806 } 3807 #endif 3808 #if defined(INET6) || defined(INET) 3809 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3810 #endif 3811 3812 if (maxmtu && thcmtu) 3813 mss = min(maxmtu, thcmtu) - min_protoh; 3814 else if (maxmtu || thcmtu) 3815 mss = max(maxmtu, thcmtu) - min_protoh; 3816 3817 return (mss); 3818 } 3819 3820 3821 /* 3822 * On a partial ack arrives, force the retransmission of the 3823 * next unacknowledged segment. Do not clear tp->t_dupacks. 3824 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3825 * be started again. 3826 */ 3827 void 3828 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3829 { 3830 tcp_seq onxt = tp->snd_nxt; 3831 u_long ocwnd = tp->snd_cwnd; 3832 u_int maxseg = tcp_maxseg(tp); 3833 3834 INP_WLOCK_ASSERT(tp->t_inpcb); 3835 3836 tcp_timer_activate(tp, TT_REXMT, 0); 3837 tp->t_rtttime = 0; 3838 tp->snd_nxt = th->th_ack; 3839 /* 3840 * Set snd_cwnd to one segment beyond acknowledged offset. 3841 * (tp->snd_una has not yet been updated when this function is called.) 3842 */ 3843 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3844 tp->t_flags |= TF_ACKNOW; 3845 (void) tp->t_fb->tfb_tcp_output(tp); 3846 tp->snd_cwnd = ocwnd; 3847 if (SEQ_GT(onxt, tp->snd_nxt)) 3848 tp->snd_nxt = onxt; 3849 /* 3850 * Partial window deflation. Relies on fact that tp->snd_una 3851 * not updated yet. 3852 */ 3853 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3854 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3855 else 3856 tp->snd_cwnd = 0; 3857 tp->snd_cwnd += maxseg; 3858 } 3859 3860 int 3861 tcp_compute_pipe(struct tcpcb *tp) 3862 { 3863 return (tp->snd_max - tp->snd_una + 3864 tp->sackhint.sack_bytes_rexmit - 3865 tp->sackhint.sacked_bytes); 3866 } 3867