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