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