1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 5 * The Regents of the University of California. All rights reserved. 6 * Copyright (c) 2007-2008,2010 7 * Swinburne University of Technology, Melbourne, Australia. 8 * Copyright (c) 2009-2010 Lawrence Stewart <[email protected]> 9 * Copyright (c) 2010 The FreeBSD Foundation 10 * Copyright (c) 2010-2011 Juniper Networks, Inc. 11 * All rights reserved. 12 * 13 * Portions of this software were developed at the Centre for Advanced Internet 14 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 15 * James Healy and David Hayes, made possible in part by a grant from the Cisco 16 * University Research Program Fund at Community Foundation Silicon Valley. 17 * 18 * Portions of this software were developed at the Centre for Advanced 19 * Internet Architectures, Swinburne University of Technology, Melbourne, 20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 21 * 22 * Portions of this software were developed by Robert N. M. Watson under 23 * contract to Juniper Networks, Inc. 24 * 25 * Redistribution and use in source and binary forms, with or without 26 * modification, are permitted provided that the following conditions 27 * are met: 28 * 1. Redistributions of source code must retain the above copyright 29 * notice, this list of conditions and the following disclaimer. 30 * 2. Redistributions in binary form must reproduce the above copyright 31 * notice, this list of conditions and the following disclaimer in the 32 * documentation and/or other materials provided with the distribution. 33 * 3. Neither the name of the University nor the names of its contributors 34 * may be used to endorse or promote products derived from this software 35 * without specific prior written permission. 36 * 37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 47 * SUCH DAMAGE. 48 * 49 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 50 */ 51 52 #include <sys/cdefs.h> 53 __FBSDID("$FreeBSD$"); 54 55 #include "opt_inet.h" 56 #include "opt_inet6.h" 57 #include "opt_ipsec.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/kernel.h> 62 #ifdef TCP_HHOOK 63 #include <sys/hhook.h> 64 #endif 65 #include <sys/malloc.h> 66 #include <sys/mbuf.h> 67 #include <sys/proc.h> /* for proc0 declaration */ 68 #include <sys/protosw.h> 69 #include <sys/sdt.h> 70 #include <sys/signalvar.h> 71 #include <sys/socket.h> 72 #include <sys/socketvar.h> 73 #include <sys/sysctl.h> 74 #include <sys/syslog.h> 75 #include <sys/systm.h> 76 77 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 78 79 #include <vm/uma.h> 80 81 #include <net/if.h> 82 #include <net/if_var.h> 83 #include <net/route.h> 84 #include <net/vnet.h> 85 86 #define TCPSTATES /* for logging */ 87 88 #include <netinet/in.h> 89 #include <netinet/in_kdtrace.h> 90 #include <netinet/in_pcb.h> 91 #include <netinet/in_systm.h> 92 #include <netinet/ip.h> 93 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 94 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 95 #include <netinet/ip_var.h> 96 #include <netinet/ip_options.h> 97 #include <netinet/ip6.h> 98 #include <netinet/icmp6.h> 99 #include <netinet6/in6_pcb.h> 100 #include <netinet6/in6_var.h> 101 #include <netinet6/ip6_var.h> 102 #include <netinet6/nd6.h> 103 #include <netinet/tcp.h> 104 #include <netinet/tcp_fsm.h> 105 #include <netinet/tcp_log_buf.h> 106 #include <netinet/tcp_seq.h> 107 #include <netinet/tcp_timer.h> 108 #include <netinet/tcp_var.h> 109 #include <netinet6/tcp6_var.h> 110 #include <netinet/tcpip.h> 111 #include <netinet/cc/cc.h> 112 #include <netinet/tcp_fastopen.h> 113 #ifdef TCPPCAP 114 #include <netinet/tcp_pcap.h> 115 #endif 116 #include <netinet/tcp_syncache.h> 117 #ifdef TCPDEBUG 118 #include <netinet/tcp_debug.h> 119 #endif /* TCPDEBUG */ 120 #ifdef TCP_OFFLOAD 121 #include <netinet/tcp_offload.h> 122 #endif 123 124 #include <netipsec/ipsec_support.h> 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 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 150 &VNET_NAME(drop_synfin), 0, 151 "Drop TCP packets with SYN+FIN set"); 152 153 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 155 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 156 "Use calculated pipe/in-flight bytes per RFC 6675"); 157 158 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 160 &VNET_NAME(tcp_do_rfc3042), 0, 161 "Enable RFC 3042 (Limited Transmit)"); 162 163 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 165 &VNET_NAME(tcp_do_rfc3390), 0, 166 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 167 168 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 170 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 171 "Slow-start flight size (initial congestion window) in number of segments"); 172 173 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 175 &VNET_NAME(tcp_do_rfc3465), 0, 176 "Enable RFC 3465 (Appropriate Byte Counting)"); 177 178 VNET_DEFINE(int, tcp_abc_l_var) = 2; 179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 180 &VNET_NAME(tcp_abc_l_var), 2, 181 "Cap the max cwnd increment during slow-start to this number of segments"); 182 183 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 184 185 VNET_DEFINE(int, tcp_do_ecn) = 2; 186 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 187 &VNET_NAME(tcp_do_ecn), 0, 188 "TCP ECN support"); 189 190 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 191 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_ecn_maxretries), 0, 193 "Max retries before giving up on ECN"); 194 195 VNET_DEFINE(int, tcp_insecure_syn) = 0; 196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 197 &VNET_NAME(tcp_insecure_syn), 0, 198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 199 200 VNET_DEFINE(int, tcp_insecure_rst) = 0; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 202 &VNET_NAME(tcp_insecure_rst), 0, 203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 204 205 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 206 #define V_tcp_recvspace VNET(tcp_recvspace) 207 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 208 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 209 210 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 212 &VNET_NAME(tcp_do_autorcvbuf), 0, 213 "Enable automatic receive buffer sizing"); 214 215 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 217 &VNET_NAME(tcp_autorcvbuf_inc), 0, 218 "Incrementor step size of automatic receive buffer"); 219 220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_autorcvbuf_max), 0, 223 "Max size of automatic receive buffer"); 224 225 VNET_DEFINE(struct inpcbhead, tcb); 226 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 227 VNET_DEFINE(struct inpcbinfo, tcbinfo); 228 229 /* 230 * TCP statistics are stored in an array of counter(9)s, which size matches 231 * size of struct tcpstat. TCP running connection count is a regular array. 232 */ 233 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 234 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 235 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 236 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 237 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 238 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 239 "TCP connection counts by TCP state"); 240 241 static void 242 tcp_vnet_init(const void *unused) 243 { 244 245 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 246 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 247 } 248 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 249 tcp_vnet_init, NULL); 250 251 #ifdef VIMAGE 252 static void 253 tcp_vnet_uninit(const void *unused) 254 { 255 256 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 257 VNET_PCPUSTAT_FREE(tcpstat); 258 } 259 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 260 tcp_vnet_uninit, NULL); 261 #endif /* VIMAGE */ 262 263 /* 264 * Kernel module interface for updating tcpstat. The argument is an index 265 * into tcpstat treated as an array. 266 */ 267 void 268 kmod_tcpstat_inc(int statnum) 269 { 270 271 counter_u64_add(VNET(tcpstat)[statnum], 1); 272 } 273 274 #ifdef TCP_HHOOK 275 /* 276 * Wrapper for the TCP established input helper hook. 277 */ 278 void 279 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 280 { 281 struct tcp_hhook_data hhook_data; 282 283 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 284 hhook_data.tp = tp; 285 hhook_data.th = th; 286 hhook_data.to = to; 287 288 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 289 tp->osd); 290 } 291 } 292 #endif 293 294 /* 295 * CC wrapper hook functions 296 */ 297 void 298 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 299 uint16_t type) 300 { 301 INP_WLOCK_ASSERT(tp->t_inpcb); 302 303 tp->ccv->nsegs = nsegs; 304 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 305 if (tp->snd_cwnd <= tp->snd_wnd) 306 tp->ccv->flags |= CCF_CWND_LIMITED; 307 else 308 tp->ccv->flags &= ~CCF_CWND_LIMITED; 309 310 if (type == CC_ACK) { 311 if (tp->snd_cwnd > tp->snd_ssthresh) { 312 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 313 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 314 if (tp->t_bytes_acked >= tp->snd_cwnd) { 315 tp->t_bytes_acked -= tp->snd_cwnd; 316 tp->ccv->flags |= CCF_ABC_SENTAWND; 317 } 318 } else { 319 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 320 tp->t_bytes_acked = 0; 321 } 322 } 323 324 if (CC_ALGO(tp)->ack_received != NULL) { 325 /* XXXLAS: Find a way to live without this */ 326 tp->ccv->curack = th->th_ack; 327 CC_ALGO(tp)->ack_received(tp->ccv, type); 328 } 329 } 330 331 void 332 cc_conn_init(struct tcpcb *tp) 333 { 334 struct hc_metrics_lite metrics; 335 struct inpcb *inp = tp->t_inpcb; 336 u_int maxseg; 337 int rtt; 338 339 INP_WLOCK_ASSERT(tp->t_inpcb); 340 341 tcp_hc_get(&inp->inp_inc, &metrics); 342 maxseg = tcp_maxseg(tp); 343 344 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 345 tp->t_srtt = rtt; 346 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 347 TCPSTAT_INC(tcps_usedrtt); 348 if (metrics.rmx_rttvar) { 349 tp->t_rttvar = metrics.rmx_rttvar; 350 TCPSTAT_INC(tcps_usedrttvar); 351 } else { 352 /* default variation is +- 1 rtt */ 353 tp->t_rttvar = 354 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 355 } 356 TCPT_RANGESET(tp->t_rxtcur, 357 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 358 tp->t_rttmin, TCPTV_REXMTMAX); 359 } 360 if (metrics.rmx_ssthresh) { 361 /* 362 * There's some sort of gateway or interface 363 * buffer limit on the path. Use this to set 364 * the slow start threshold, but set the 365 * threshold to no less than 2*mss. 366 */ 367 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 368 TCPSTAT_INC(tcps_usedssthresh); 369 } 370 371 /* 372 * Set the initial slow-start flight size. 373 * 374 * RFC5681 Section 3.1 specifies the default conservative values. 375 * RFC3390 specifies slightly more aggressive values. 376 * RFC6928 increases it to ten segments. 377 * Support for user specified value for initial flight size. 378 * 379 * If a SYN or SYN/ACK was lost and retransmitted, we have to 380 * reduce the initial CWND to one segment as congestion is likely 381 * requiring us to be cautious. 382 */ 383 if (tp->snd_cwnd == 1) 384 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 385 else if (V_tcp_initcwnd_segments) 386 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 387 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 388 else if (V_tcp_do_rfc3390) 389 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 390 else { 391 /* Per RFC5681 Section 3.1 */ 392 if (maxseg > 2190) 393 tp->snd_cwnd = 2 * maxseg; 394 else if (maxseg > 1095) 395 tp->snd_cwnd = 3 * maxseg; 396 else 397 tp->snd_cwnd = 4 * maxseg; 398 } 399 400 if (CC_ALGO(tp)->conn_init != NULL) 401 CC_ALGO(tp)->conn_init(tp->ccv); 402 } 403 404 void inline 405 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 406 { 407 u_int maxseg; 408 409 INP_WLOCK_ASSERT(tp->t_inpcb); 410 411 switch(type) { 412 case CC_NDUPACK: 413 if (!IN_FASTRECOVERY(tp->t_flags)) { 414 tp->snd_recover = tp->snd_max; 415 if (tp->t_flags & TF_ECN_PERMIT) 416 tp->t_flags |= TF_ECN_SND_CWR; 417 } 418 break; 419 case CC_ECN: 420 if (!IN_CONGRECOVERY(tp->t_flags)) { 421 TCPSTAT_INC(tcps_ecn_rcwnd); 422 tp->snd_recover = tp->snd_max; 423 if (tp->t_flags & TF_ECN_PERMIT) 424 tp->t_flags |= TF_ECN_SND_CWR; 425 } 426 break; 427 case CC_RTO: 428 maxseg = tcp_maxseg(tp); 429 tp->t_dupacks = 0; 430 tp->t_bytes_acked = 0; 431 EXIT_RECOVERY(tp->t_flags); 432 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 433 maxseg) * maxseg; 434 tp->snd_cwnd = maxseg; 435 break; 436 case CC_RTO_ERR: 437 TCPSTAT_INC(tcps_sndrexmitbad); 438 /* RTO was unnecessary, so reset everything. */ 439 tp->snd_cwnd = tp->snd_cwnd_prev; 440 tp->snd_ssthresh = tp->snd_ssthresh_prev; 441 tp->snd_recover = tp->snd_recover_prev; 442 if (tp->t_flags & TF_WASFRECOVERY) 443 ENTER_FASTRECOVERY(tp->t_flags); 444 if (tp->t_flags & TF_WASCRECOVERY) 445 ENTER_CONGRECOVERY(tp->t_flags); 446 tp->snd_nxt = tp->snd_max; 447 tp->t_flags &= ~TF_PREVVALID; 448 tp->t_badrxtwin = 0; 449 break; 450 } 451 452 if (CC_ALGO(tp)->cong_signal != NULL) { 453 if (th != NULL) 454 tp->ccv->curack = th->th_ack; 455 CC_ALGO(tp)->cong_signal(tp->ccv, type); 456 } 457 } 458 459 void inline 460 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 461 { 462 INP_WLOCK_ASSERT(tp->t_inpcb); 463 464 /* XXXLAS: KASSERT that we're in recovery? */ 465 466 if (CC_ALGO(tp)->post_recovery != NULL) { 467 tp->ccv->curack = th->th_ack; 468 CC_ALGO(tp)->post_recovery(tp->ccv); 469 } 470 /* XXXLAS: EXIT_RECOVERY ? */ 471 tp->t_bytes_acked = 0; 472 } 473 474 /* 475 * Indicate whether this ack should be delayed. We can delay the ack if 476 * following conditions are met: 477 * - There is no delayed ack timer in progress. 478 * - Our last ack wasn't a 0-sized window. We never want to delay 479 * the ack that opens up a 0-sized window. 480 * - LRO wasn't used for this segment. We make sure by checking that the 481 * segment size is not larger than the MSS. 482 */ 483 #define DELAY_ACK(tp, tlen) \ 484 ((!tcp_timer_active(tp, TT_DELACK) && \ 485 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 486 (tlen <= tp->t_maxseg) && \ 487 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 488 489 static void inline 490 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 491 { 492 INP_WLOCK_ASSERT(tp->t_inpcb); 493 494 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 495 switch (iptos & IPTOS_ECN_MASK) { 496 case IPTOS_ECN_CE: 497 tp->ccv->flags |= CCF_IPHDR_CE; 498 break; 499 case IPTOS_ECN_ECT0: 500 tp->ccv->flags &= ~CCF_IPHDR_CE; 501 break; 502 case IPTOS_ECN_ECT1: 503 tp->ccv->flags &= ~CCF_IPHDR_CE; 504 break; 505 } 506 507 if (th->th_flags & TH_CWR) 508 tp->ccv->flags |= CCF_TCPHDR_CWR; 509 else 510 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 511 512 if (tp->t_flags & TF_DELACK) 513 tp->ccv->flags |= CCF_DELACK; 514 else 515 tp->ccv->flags &= ~CCF_DELACK; 516 517 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 518 519 if (tp->ccv->flags & CCF_ACKNOW) 520 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 521 } 522 } 523 524 /* 525 * TCP input handling is split into multiple parts: 526 * tcp6_input is a thin wrapper around tcp_input for the extended 527 * ip6_protox[] call format in ip6_input 528 * tcp_input handles primary segment validation, inpcb lookup and 529 * SYN processing on listen sockets 530 * tcp_do_segment processes the ACK and text of the segment for 531 * establishing, established and closing connections 532 */ 533 #ifdef INET6 534 int 535 tcp6_input(struct mbuf **mp, int *offp, int proto) 536 { 537 struct mbuf *m = *mp; 538 struct in6_ifaddr *ia6; 539 struct ip6_hdr *ip6; 540 541 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 542 543 /* 544 * draft-itojun-ipv6-tcp-to-anycast 545 * better place to put this in? 546 */ 547 ip6 = mtod(m, struct ip6_hdr *); 548 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 549 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 550 struct ip6_hdr *ip6; 551 552 ifa_free(&ia6->ia_ifa); 553 ip6 = mtod(m, struct ip6_hdr *); 554 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 555 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 556 return (IPPROTO_DONE); 557 } 558 if (ia6) 559 ifa_free(&ia6->ia_ifa); 560 561 return (tcp_input(mp, offp, proto)); 562 } 563 #endif /* INET6 */ 564 565 int 566 tcp_input(struct mbuf **mp, int *offp, int proto) 567 { 568 struct mbuf *m = *mp; 569 struct tcphdr *th = NULL; 570 struct ip *ip = NULL; 571 struct inpcb *inp = NULL; 572 struct tcpcb *tp = NULL; 573 struct socket *so = NULL; 574 u_char *optp = NULL; 575 int off0; 576 int optlen = 0; 577 #ifdef INET 578 int len; 579 uint8_t ipttl; 580 #endif 581 int tlen = 0, off; 582 int drop_hdrlen; 583 int thflags; 584 int rstreason = 0; /* For badport_bandlim accounting purposes */ 585 uint8_t iptos; 586 struct m_tag *fwd_tag = NULL; 587 struct epoch_tracker et; 588 #ifdef INET6 589 struct ip6_hdr *ip6 = NULL; 590 int isipv6; 591 #else 592 const void *ip6 = NULL; 593 #endif /* INET6 */ 594 struct tcpopt to; /* options in this segment */ 595 char *s = NULL; /* address and port logging */ 596 int ti_locked; 597 #ifdef TCPDEBUG 598 /* 599 * The size of tcp_saveipgen must be the size of the max ip header, 600 * now IPv6. 601 */ 602 u_char tcp_saveipgen[IP6_HDR_LEN]; 603 struct tcphdr tcp_savetcp; 604 short ostate = 0; 605 #endif 606 607 #ifdef INET6 608 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 609 #endif 610 611 off0 = *offp; 612 m = *mp; 613 *mp = NULL; 614 to.to_flags = 0; 615 TCPSTAT_INC(tcps_rcvtotal); 616 617 #ifdef INET6 618 if (isipv6) { 619 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 620 621 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 622 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 623 if (m == NULL) { 624 TCPSTAT_INC(tcps_rcvshort); 625 return (IPPROTO_DONE); 626 } 627 } 628 629 ip6 = mtod(m, struct ip6_hdr *); 630 th = (struct tcphdr *)((caddr_t)ip6 + off0); 631 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 632 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 633 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 634 th->th_sum = m->m_pkthdr.csum_data; 635 else 636 th->th_sum = in6_cksum_pseudo(ip6, tlen, 637 IPPROTO_TCP, m->m_pkthdr.csum_data); 638 th->th_sum ^= 0xffff; 639 } else 640 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 641 if (th->th_sum) { 642 TCPSTAT_INC(tcps_rcvbadsum); 643 goto drop; 644 } 645 646 /* 647 * Be proactive about unspecified IPv6 address in source. 648 * As we use all-zero to indicate unbounded/unconnected pcb, 649 * unspecified IPv6 address can be used to confuse us. 650 * 651 * Note that packets with unspecified IPv6 destination is 652 * already dropped in ip6_input. 653 */ 654 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 655 /* XXX stat */ 656 goto drop; 657 } 658 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 659 } 660 #endif 661 #if defined(INET) && defined(INET6) 662 else 663 #endif 664 #ifdef INET 665 { 666 /* 667 * Get IP and TCP header together in first mbuf. 668 * Note: IP leaves IP header in first mbuf. 669 */ 670 if (off0 > sizeof (struct ip)) { 671 ip_stripoptions(m); 672 off0 = sizeof(struct ip); 673 } 674 if (m->m_len < sizeof (struct tcpiphdr)) { 675 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 676 == NULL) { 677 TCPSTAT_INC(tcps_rcvshort); 678 return (IPPROTO_DONE); 679 } 680 } 681 ip = mtod(m, struct ip *); 682 th = (struct tcphdr *)((caddr_t)ip + off0); 683 tlen = ntohs(ip->ip_len) - off0; 684 685 iptos = ip->ip_tos; 686 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 687 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 688 th->th_sum = m->m_pkthdr.csum_data; 689 else 690 th->th_sum = in_pseudo(ip->ip_src.s_addr, 691 ip->ip_dst.s_addr, 692 htonl(m->m_pkthdr.csum_data + tlen + 693 IPPROTO_TCP)); 694 th->th_sum ^= 0xffff; 695 } else { 696 struct ipovly *ipov = (struct ipovly *)ip; 697 698 /* 699 * Checksum extended TCP header and data. 700 */ 701 len = off0 + tlen; 702 ipttl = ip->ip_ttl; 703 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 704 ipov->ih_len = htons(tlen); 705 th->th_sum = in_cksum(m, len); 706 /* Reset length for SDT probes. */ 707 ip->ip_len = htons(len); 708 /* Reset TOS bits */ 709 ip->ip_tos = iptos; 710 /* Re-initialization for later version check */ 711 ip->ip_ttl = ipttl; 712 ip->ip_v = IPVERSION; 713 ip->ip_hl = off0 >> 2; 714 } 715 716 if (th->th_sum) { 717 TCPSTAT_INC(tcps_rcvbadsum); 718 goto drop; 719 } 720 } 721 #endif /* INET */ 722 723 /* 724 * Check that TCP offset makes sense, 725 * pull out TCP options and adjust length. XXX 726 */ 727 off = th->th_off << 2; 728 if (off < sizeof (struct tcphdr) || off > tlen) { 729 TCPSTAT_INC(tcps_rcvbadoff); 730 goto drop; 731 } 732 tlen -= off; /* tlen is used instead of ti->ti_len */ 733 if (off > sizeof (struct tcphdr)) { 734 #ifdef INET6 735 if (isipv6) { 736 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 737 ip6 = mtod(m, struct ip6_hdr *); 738 th = (struct tcphdr *)((caddr_t)ip6 + off0); 739 } 740 #endif 741 #if defined(INET) && defined(INET6) 742 else 743 #endif 744 #ifdef INET 745 { 746 if (m->m_len < sizeof(struct ip) + off) { 747 if ((m = m_pullup(m, sizeof (struct ip) + off)) 748 == NULL) { 749 TCPSTAT_INC(tcps_rcvshort); 750 return (IPPROTO_DONE); 751 } 752 ip = mtod(m, struct ip *); 753 th = (struct tcphdr *)((caddr_t)ip + off0); 754 } 755 } 756 #endif 757 optlen = off - sizeof (struct tcphdr); 758 optp = (u_char *)(th + 1); 759 } 760 thflags = th->th_flags; 761 762 /* 763 * Convert TCP protocol specific fields to host format. 764 */ 765 tcp_fields_to_host(th); 766 767 /* 768 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 769 */ 770 drop_hdrlen = off0 + off; 771 772 /* 773 * Locate pcb for segment; if we're likely to add or remove a 774 * connection then first acquire pcbinfo lock. There are three cases 775 * where we might discover later we need a write lock despite the 776 * flags: ACKs moving a connection out of the syncache, ACKs for a 777 * connection in TIMEWAIT and SYNs not targeting a listening socket. 778 */ 779 if ((thflags & (TH_FIN | TH_RST)) != 0) { 780 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 781 ti_locked = TI_RLOCKED; 782 } else 783 ti_locked = TI_UNLOCKED; 784 785 /* 786 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 787 */ 788 if ( 789 #ifdef INET6 790 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 791 #ifdef INET 792 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 793 #endif 794 #endif 795 #if defined(INET) && !defined(INET6) 796 (m->m_flags & M_IP_NEXTHOP) 797 #endif 798 ) 799 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 800 801 findpcb: 802 #ifdef INVARIANTS 803 if (ti_locked == TI_RLOCKED) { 804 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 805 } else { 806 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 807 } 808 #endif 809 #ifdef INET6 810 if (isipv6 && fwd_tag != NULL) { 811 struct sockaddr_in6 *next_hop6; 812 813 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 814 /* 815 * Transparently forwarded. Pretend to be the destination. 816 * Already got one like this? 817 */ 818 inp = in6_pcblookup_mbuf(&V_tcbinfo, 819 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 820 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 821 if (!inp) { 822 /* 823 * It's new. Try to find the ambushing socket. 824 * Because we've rewritten the destination address, 825 * any hardware-generated hash is ignored. 826 */ 827 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 828 th->th_sport, &next_hop6->sin6_addr, 829 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 830 th->th_dport, INPLOOKUP_WILDCARD | 831 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 832 } 833 } else if (isipv6) { 834 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 835 th->th_sport, &ip6->ip6_dst, th->th_dport, 836 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 837 m->m_pkthdr.rcvif, m); 838 } 839 #endif /* INET6 */ 840 #if defined(INET6) && defined(INET) 841 else 842 #endif 843 #ifdef INET 844 if (fwd_tag != NULL) { 845 struct sockaddr_in *next_hop; 846 847 next_hop = (struct sockaddr_in *)(fwd_tag+1); 848 /* 849 * Transparently forwarded. Pretend to be the destination. 850 * already got one like this? 851 */ 852 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 853 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 854 m->m_pkthdr.rcvif, m); 855 if (!inp) { 856 /* 857 * It's new. Try to find the ambushing socket. 858 * Because we've rewritten the destination address, 859 * any hardware-generated hash is ignored. 860 */ 861 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 862 th->th_sport, next_hop->sin_addr, 863 next_hop->sin_port ? ntohs(next_hop->sin_port) : 864 th->th_dport, INPLOOKUP_WILDCARD | 865 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 866 } 867 } else 868 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 869 th->th_sport, ip->ip_dst, th->th_dport, 870 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 871 m->m_pkthdr.rcvif, m); 872 #endif /* INET */ 873 874 /* 875 * If the INPCB does not exist then all data in the incoming 876 * segment is discarded and an appropriate RST is sent back. 877 * XXX MRT Send RST using which routing table? 878 */ 879 if (inp == NULL) { 880 /* 881 * Log communication attempts to ports that are not 882 * in use. 883 */ 884 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 885 tcp_log_in_vain == 2) { 886 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 887 log(LOG_INFO, "%s; %s: Connection attempt " 888 "to closed port\n", s, __func__); 889 } 890 /* 891 * When blackholing do not respond with a RST but 892 * completely ignore the segment and drop it. 893 */ 894 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 895 V_blackhole == 2) 896 goto dropunlock; 897 898 rstreason = BANDLIM_RST_CLOSEDPORT; 899 goto dropwithreset; 900 } 901 INP_WLOCK_ASSERT(inp); 902 /* 903 * While waiting for inp lock during the lookup, another thread 904 * can have dropped the inpcb, in which case we need to loop back 905 * and try to find a new inpcb to deliver to. 906 */ 907 if (inp->inp_flags & INP_DROPPED) { 908 INP_WUNLOCK(inp); 909 inp = NULL; 910 goto findpcb; 911 } 912 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 913 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 914 ((inp->inp_socket == NULL) || 915 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 916 inp->inp_flowid = m->m_pkthdr.flowid; 917 inp->inp_flowtype = M_HASHTYPE_GET(m); 918 } 919 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 920 #ifdef INET6 921 if (isipv6 && IPSEC_ENABLED(ipv6) && 922 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 923 goto dropunlock; 924 } 925 #ifdef INET 926 else 927 #endif 928 #endif /* INET6 */ 929 #ifdef INET 930 if (IPSEC_ENABLED(ipv4) && 931 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 932 goto dropunlock; 933 } 934 #endif /* INET */ 935 #endif /* IPSEC */ 936 937 /* 938 * Check the minimum TTL for socket. 939 */ 940 if (inp->inp_ip_minttl != 0) { 941 #ifdef INET6 942 if (isipv6) { 943 if (inp->inp_ip_minttl > ip6->ip6_hlim) 944 goto dropunlock; 945 } else 946 #endif 947 if (inp->inp_ip_minttl > ip->ip_ttl) 948 goto dropunlock; 949 } 950 951 /* 952 * A previous connection in TIMEWAIT state is supposed to catch stray 953 * or duplicate segments arriving late. If this segment was a 954 * legitimate new connection attempt, the old INPCB gets removed and 955 * we can try again to find a listening socket. 956 * 957 * At this point, due to earlier optimism, we may hold only an inpcb 958 * lock, and not the inpcbinfo write lock. If so, we need to try to 959 * acquire it, or if that fails, acquire a reference on the inpcb, 960 * drop all locks, acquire a global write lock, and then re-acquire 961 * the inpcb lock. We may at that point discover that another thread 962 * has tried to free the inpcb, in which case we need to loop back 963 * and try to find a new inpcb to deliver to. 964 * 965 * XXXRW: It may be time to rethink timewait locking. 966 */ 967 if (inp->inp_flags & INP_TIMEWAIT) { 968 if (ti_locked == TI_UNLOCKED) { 969 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 970 ti_locked = TI_RLOCKED; 971 } 972 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 973 974 if (thflags & TH_SYN) 975 tcp_dooptions(&to, optp, optlen, TO_SYN); 976 /* 977 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 978 */ 979 if (tcp_twcheck(inp, &to, th, m, tlen)) 980 goto findpcb; 981 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 982 return (IPPROTO_DONE); 983 } 984 /* 985 * The TCPCB may no longer exist if the connection is winding 986 * down or it is in the CLOSED state. Either way we drop the 987 * segment and send an appropriate response. 988 */ 989 tp = intotcpcb(inp); 990 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 991 rstreason = BANDLIM_RST_CLOSEDPORT; 992 goto dropwithreset; 993 } 994 995 #ifdef TCP_OFFLOAD 996 if (tp->t_flags & TF_TOE) { 997 tcp_offload_input(tp, m); 998 m = NULL; /* consumed by the TOE driver */ 999 goto dropunlock; 1000 } 1001 #endif 1002 1003 /* 1004 * We've identified a valid inpcb, but it could be that we need an 1005 * inpcbinfo write lock but don't hold it. In this case, attempt to 1006 * acquire using the same strategy as the TIMEWAIT case above. If we 1007 * relock, we have to jump back to 'relocked' as the connection might 1008 * now be in TIMEWAIT. 1009 */ 1010 #ifdef INVARIANTS 1011 if ((thflags & (TH_FIN | TH_RST)) != 0) 1012 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1013 #endif 1014 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1015 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1016 !IS_FASTOPEN(tp->t_flags)))) { 1017 if (ti_locked == TI_UNLOCKED) { 1018 INP_INFO_RLOCK_ET(&V_tcbinfo, et); 1019 ti_locked = TI_RLOCKED; 1020 } 1021 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1022 } 1023 1024 #ifdef MAC 1025 INP_WLOCK_ASSERT(inp); 1026 if (mac_inpcb_check_deliver(inp, m)) 1027 goto dropunlock; 1028 #endif 1029 so = inp->inp_socket; 1030 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1031 #ifdef TCPDEBUG 1032 if (so->so_options & SO_DEBUG) { 1033 ostate = tp->t_state; 1034 #ifdef INET6 1035 if (isipv6) { 1036 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1037 } else 1038 #endif 1039 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1040 tcp_savetcp = *th; 1041 } 1042 #endif /* TCPDEBUG */ 1043 /* 1044 * When the socket is accepting connections (the INPCB is in LISTEN 1045 * state) we look into the SYN cache if this is a new connection 1046 * attempt or the completion of a previous one. 1047 */ 1048 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1049 ("%s: so accepting but tp %p not listening", __func__, tp)); 1050 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1051 struct in_conninfo inc; 1052 1053 bzero(&inc, sizeof(inc)); 1054 #ifdef INET6 1055 if (isipv6) { 1056 inc.inc_flags |= INC_ISIPV6; 1057 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) 1058 inc.inc_flags |= INC_IPV6MINMTU; 1059 inc.inc6_faddr = ip6->ip6_src; 1060 inc.inc6_laddr = ip6->ip6_dst; 1061 } else 1062 #endif 1063 { 1064 inc.inc_faddr = ip->ip_src; 1065 inc.inc_laddr = ip->ip_dst; 1066 } 1067 inc.inc_fport = th->th_sport; 1068 inc.inc_lport = th->th_dport; 1069 inc.inc_fibnum = so->so_fibnum; 1070 1071 /* 1072 * Check for an existing connection attempt in syncache if 1073 * the flag is only ACK. A successful lookup creates a new 1074 * socket appended to the listen queue in SYN_RECEIVED state. 1075 */ 1076 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1077 1078 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1079 /* 1080 * Parse the TCP options here because 1081 * syncookies need access to the reflected 1082 * timestamp. 1083 */ 1084 tcp_dooptions(&to, optp, optlen, 0); 1085 /* 1086 * NB: syncache_expand() doesn't unlock 1087 * inp and tcpinfo locks. 1088 */ 1089 rstreason = syncache_expand(&inc, &to, th, &so, m); 1090 if (rstreason < 0) { 1091 /* 1092 * A failing TCP MD5 signature comparison 1093 * must result in the segment being dropped 1094 * and must not produce any response back 1095 * to the sender. 1096 */ 1097 goto dropunlock; 1098 } else if (rstreason == 0) { 1099 /* 1100 * No syncache entry or ACK was not 1101 * for our SYN/ACK. Send a RST. 1102 * NB: syncache did its own logging 1103 * of the failure cause. 1104 */ 1105 rstreason = BANDLIM_RST_OPENPORT; 1106 goto dropwithreset; 1107 } 1108 tfo_socket_result: 1109 if (so == NULL) { 1110 /* 1111 * We completed the 3-way handshake 1112 * but could not allocate a socket 1113 * either due to memory shortage, 1114 * listen queue length limits or 1115 * global socket limits. Send RST 1116 * or wait and have the remote end 1117 * retransmit the ACK for another 1118 * try. 1119 */ 1120 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1121 log(LOG_DEBUG, "%s; %s: Listen socket: " 1122 "Socket allocation failed due to " 1123 "limits or memory shortage, %s\n", 1124 s, __func__, 1125 V_tcp_sc_rst_sock_fail ? 1126 "sending RST" : "try again"); 1127 if (V_tcp_sc_rst_sock_fail) { 1128 rstreason = BANDLIM_UNLIMITED; 1129 goto dropwithreset; 1130 } else 1131 goto dropunlock; 1132 } 1133 /* 1134 * Socket is created in state SYN_RECEIVED. 1135 * Unlock the listen socket, lock the newly 1136 * created socket and update the tp variable. 1137 */ 1138 INP_WUNLOCK(inp); /* listen socket */ 1139 inp = sotoinpcb(so); 1140 /* 1141 * New connection inpcb is already locked by 1142 * syncache_expand(). 1143 */ 1144 INP_WLOCK_ASSERT(inp); 1145 tp = intotcpcb(inp); 1146 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1147 ("%s: ", __func__)); 1148 /* 1149 * Process the segment and the data it 1150 * contains. tcp_do_segment() consumes 1151 * the mbuf chain and unlocks the inpcb. 1152 */ 1153 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1154 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1155 iptos); 1156 if (ti_locked == TI_RLOCKED) 1157 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1158 return (IPPROTO_DONE); 1159 } 1160 /* 1161 * Segment flag validation for new connection attempts: 1162 * 1163 * Our (SYN|ACK) response was rejected. 1164 * Check with syncache and remove entry to prevent 1165 * retransmits. 1166 * 1167 * NB: syncache_chkrst does its own logging of failure 1168 * causes. 1169 */ 1170 if (thflags & TH_RST) { 1171 syncache_chkrst(&inc, th, m); 1172 goto dropunlock; 1173 } 1174 /* 1175 * We can't do anything without SYN. 1176 */ 1177 if ((thflags & TH_SYN) == 0) { 1178 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1179 log(LOG_DEBUG, "%s; %s: Listen socket: " 1180 "SYN is missing, segment ignored\n", 1181 s, __func__); 1182 TCPSTAT_INC(tcps_badsyn); 1183 goto dropunlock; 1184 } 1185 /* 1186 * (SYN|ACK) is bogus on a listen socket. 1187 */ 1188 if (thflags & TH_ACK) { 1189 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1190 log(LOG_DEBUG, "%s; %s: Listen socket: " 1191 "SYN|ACK invalid, segment rejected\n", 1192 s, __func__); 1193 syncache_badack(&inc); /* XXX: Not needed! */ 1194 TCPSTAT_INC(tcps_badsyn); 1195 rstreason = BANDLIM_RST_OPENPORT; 1196 goto dropwithreset; 1197 } 1198 /* 1199 * If the drop_synfin option is enabled, drop all 1200 * segments with both the SYN and FIN bits set. 1201 * This prevents e.g. nmap from identifying the 1202 * TCP/IP stack. 1203 * XXX: Poor reasoning. nmap has other methods 1204 * and is constantly refining its stack detection 1205 * strategies. 1206 * XXX: This is a violation of the TCP specification 1207 * and was used by RFC1644. 1208 */ 1209 if ((thflags & TH_FIN) && V_drop_synfin) { 1210 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1211 log(LOG_DEBUG, "%s; %s: Listen socket: " 1212 "SYN|FIN segment ignored (based on " 1213 "sysctl setting)\n", s, __func__); 1214 TCPSTAT_INC(tcps_badsyn); 1215 goto dropunlock; 1216 } 1217 /* 1218 * Segment's flags are (SYN) or (SYN|FIN). 1219 * 1220 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1221 * as they do not affect the state of the TCP FSM. 1222 * The data pointed to by TH_URG and th_urp is ignored. 1223 */ 1224 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1225 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1226 KASSERT(thflags & (TH_SYN), 1227 ("%s: Listen socket: TH_SYN not set", __func__)); 1228 #ifdef INET6 1229 /* 1230 * If deprecated address is forbidden, 1231 * we do not accept SYN to deprecated interface 1232 * address to prevent any new inbound connection from 1233 * getting established. 1234 * When we do not accept SYN, we send a TCP RST, 1235 * with deprecated source address (instead of dropping 1236 * it). We compromise it as it is much better for peer 1237 * to send a RST, and RST will be the final packet 1238 * for the exchange. 1239 * 1240 * If we do not forbid deprecated addresses, we accept 1241 * the SYN packet. RFC2462 does not suggest dropping 1242 * SYN in this case. 1243 * If we decipher RFC2462 5.5.4, it says like this: 1244 * 1. use of deprecated addr with existing 1245 * communication is okay - "SHOULD continue to be 1246 * used" 1247 * 2. use of it with new communication: 1248 * (2a) "SHOULD NOT be used if alternate address 1249 * with sufficient scope is available" 1250 * (2b) nothing mentioned otherwise. 1251 * Here we fall into (2b) case as we have no choice in 1252 * our source address selection - we must obey the peer. 1253 * 1254 * The wording in RFC2462 is confusing, and there are 1255 * multiple description text for deprecated address 1256 * handling - worse, they are not exactly the same. 1257 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1258 */ 1259 if (isipv6 && !V_ip6_use_deprecated) { 1260 struct in6_ifaddr *ia6; 1261 1262 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1263 if (ia6 != NULL && 1264 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1265 ifa_free(&ia6->ia_ifa); 1266 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1267 log(LOG_DEBUG, "%s; %s: Listen socket: " 1268 "Connection attempt to deprecated " 1269 "IPv6 address rejected\n", 1270 s, __func__); 1271 rstreason = BANDLIM_RST_OPENPORT; 1272 goto dropwithreset; 1273 } 1274 if (ia6) 1275 ifa_free(&ia6->ia_ifa); 1276 } 1277 #endif /* INET6 */ 1278 /* 1279 * Basic sanity checks on incoming SYN requests: 1280 * Don't respond if the destination is a link layer 1281 * broadcast according to RFC1122 4.2.3.10, p. 104. 1282 * If it is from this socket it must be forged. 1283 * Don't respond if the source or destination is a 1284 * global or subnet broad- or multicast address. 1285 * Note that it is quite possible to receive unicast 1286 * link-layer packets with a broadcast IP address. Use 1287 * in_broadcast() to find them. 1288 */ 1289 if (m->m_flags & (M_BCAST|M_MCAST)) { 1290 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1291 log(LOG_DEBUG, "%s; %s: Listen socket: " 1292 "Connection attempt from broad- or multicast " 1293 "link layer address ignored\n", s, __func__); 1294 goto dropunlock; 1295 } 1296 #ifdef INET6 1297 if (isipv6) { 1298 if (th->th_dport == th->th_sport && 1299 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1300 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1301 log(LOG_DEBUG, "%s; %s: Listen socket: " 1302 "Connection attempt to/from self " 1303 "ignored\n", s, __func__); 1304 goto dropunlock; 1305 } 1306 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1307 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1309 log(LOG_DEBUG, "%s; %s: Listen socket: " 1310 "Connection attempt from/to multicast " 1311 "address ignored\n", s, __func__); 1312 goto dropunlock; 1313 } 1314 } 1315 #endif 1316 #if defined(INET) && defined(INET6) 1317 else 1318 #endif 1319 #ifdef INET 1320 { 1321 if (th->th_dport == th->th_sport && 1322 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1323 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1324 log(LOG_DEBUG, "%s; %s: Listen socket: " 1325 "Connection attempt from/to self " 1326 "ignored\n", s, __func__); 1327 goto dropunlock; 1328 } 1329 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1330 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1331 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1332 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1333 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1334 log(LOG_DEBUG, "%s; %s: Listen socket: " 1335 "Connection attempt from/to broad- " 1336 "or multicast address ignored\n", 1337 s, __func__); 1338 goto dropunlock; 1339 } 1340 } 1341 #endif 1342 /* 1343 * SYN appears to be valid. Create compressed TCP state 1344 * for syncache. 1345 */ 1346 #ifdef TCPDEBUG 1347 if (so->so_options & SO_DEBUG) 1348 tcp_trace(TA_INPUT, ostate, tp, 1349 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1350 #endif 1351 TCP_PROBE3(debug__input, tp, th, m); 1352 tcp_dooptions(&to, optp, optlen, TO_SYN); 1353 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1354 goto tfo_socket_result; 1355 1356 /* 1357 * Entry added to syncache and mbuf consumed. 1358 * Only the listen socket is unlocked by syncache_add(). 1359 */ 1360 if (ti_locked == TI_RLOCKED) { 1361 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1362 ti_locked = TI_UNLOCKED; 1363 } 1364 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1365 return (IPPROTO_DONE); 1366 } else if (tp->t_state == TCPS_LISTEN) { 1367 /* 1368 * When a listen socket is torn down the SO_ACCEPTCONN 1369 * flag is removed first while connections are drained 1370 * from the accept queue in a unlock/lock cycle of the 1371 * ACCEPT_LOCK, opening a race condition allowing a SYN 1372 * attempt go through unhandled. 1373 */ 1374 goto dropunlock; 1375 } 1376 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1377 if (tp->t_flags & TF_SIGNATURE) { 1378 tcp_dooptions(&to, optp, optlen, thflags); 1379 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1380 TCPSTAT_INC(tcps_sig_err_nosigopt); 1381 goto dropunlock; 1382 } 1383 if (!TCPMD5_ENABLED() || 1384 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1385 goto dropunlock; 1386 } 1387 #endif 1388 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1389 1390 /* 1391 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1392 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1393 * the inpcb, and unlocks pcbinfo. 1394 */ 1395 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos); 1396 if (ti_locked == TI_RLOCKED) 1397 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1398 return (IPPROTO_DONE); 1399 1400 dropwithreset: 1401 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1402 1403 if (ti_locked == TI_RLOCKED) { 1404 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1405 ti_locked = TI_UNLOCKED; 1406 } 1407 #ifdef INVARIANTS 1408 else { 1409 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1410 "ti_locked: %d", __func__, ti_locked)); 1411 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1412 } 1413 #endif 1414 1415 if (inp != NULL) { 1416 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1417 INP_WUNLOCK(inp); 1418 } else 1419 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1420 m = NULL; /* mbuf chain got consumed. */ 1421 goto drop; 1422 1423 dropunlock: 1424 if (m != NULL) 1425 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1426 1427 if (ti_locked == TI_RLOCKED) { 1428 INP_INFO_RUNLOCK_ET(&V_tcbinfo, et); 1429 ti_locked = TI_UNLOCKED; 1430 } 1431 #ifdef INVARIANTS 1432 else { 1433 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1434 "ti_locked: %d", __func__, ti_locked)); 1435 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1436 } 1437 #endif 1438 1439 if (inp != NULL) 1440 INP_WUNLOCK(inp); 1441 1442 drop: 1443 INP_INFO_WUNLOCK_ASSERT(&V_tcbinfo); 1444 if (s != NULL) 1445 free(s, M_TCPLOG); 1446 if (m != NULL) 1447 m_freem(m); 1448 return (IPPROTO_DONE); 1449 } 1450 1451 /* 1452 * Automatic sizing of receive socket buffer. Often the send 1453 * buffer size is not optimally adjusted to the actual network 1454 * conditions at hand (delay bandwidth product). Setting the 1455 * buffer size too small limits throughput on links with high 1456 * bandwidth and high delay (eg. trans-continental/oceanic links). 1457 * 1458 * On the receive side the socket buffer memory is only rarely 1459 * used to any significant extent. This allows us to be much 1460 * more aggressive in scaling the receive socket buffer. For 1461 * the case that the buffer space is actually used to a large 1462 * extent and we run out of kernel memory we can simply drop 1463 * the new segments; TCP on the sender will just retransmit it 1464 * later. Setting the buffer size too big may only consume too 1465 * much kernel memory if the application doesn't read() from 1466 * the socket or packet loss or reordering makes use of the 1467 * reassembly queue. 1468 * 1469 * The criteria to step up the receive buffer one notch are: 1470 * 1. Application has not set receive buffer size with 1471 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1472 * 2. the number of bytes received during the time it takes 1473 * one timestamp to be reflected back to us (the RTT); 1474 * 3. received bytes per RTT is within seven eighth of the 1475 * current socket buffer size; 1476 * 4. receive buffer size has not hit maximal automatic size; 1477 * 1478 * This algorithm does one step per RTT at most and only if 1479 * we receive a bulk stream w/o packet losses or reorderings. 1480 * Shrinking the buffer during idle times is not necessary as 1481 * it doesn't consume any memory when idle. 1482 * 1483 * TODO: Only step up if the application is actually serving 1484 * the buffer to better manage the socket buffer resources. 1485 */ 1486 int 1487 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1488 struct tcpcb *tp, int tlen) 1489 { 1490 int newsize = 0; 1491 1492 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1493 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1494 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1495 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1496 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1497 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1498 newsize = min(so->so_rcv.sb_hiwat + 1499 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1500 } 1501 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1502 1503 /* Start over with next RTT. */ 1504 tp->rfbuf_ts = 0; 1505 tp->rfbuf_cnt = 0; 1506 } else { 1507 tp->rfbuf_cnt += tlen; /* add up */ 1508 } 1509 return (newsize); 1510 } 1511 1512 void 1513 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1514 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos) 1515 { 1516 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1517 int rstreason, todrop, win; 1518 uint32_t tiwin; 1519 uint16_t nsegs; 1520 char *s; 1521 struct in_conninfo *inc; 1522 struct mbuf *mfree; 1523 struct tcpopt to; 1524 int tfo_syn; 1525 1526 #ifdef TCPDEBUG 1527 /* 1528 * The size of tcp_saveipgen must be the size of the max ip header, 1529 * now IPv6. 1530 */ 1531 u_char tcp_saveipgen[IP6_HDR_LEN]; 1532 struct tcphdr tcp_savetcp; 1533 short ostate = 0; 1534 #endif 1535 thflags = th->th_flags; 1536 inc = &tp->t_inpcb->inp_inc; 1537 tp->sackhint.last_sack_ack = 0; 1538 sack_changed = 0; 1539 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1540 /* 1541 * If this is either a state-changing packet or current state isn't 1542 * established, we require a write lock on tcbinfo. Otherwise, we 1543 * allow the tcbinfo to be in either alocked or unlocked, as the 1544 * caller may have unnecessarily acquired a write lock due to a race. 1545 */ 1546 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1547 tp->t_state != TCPS_ESTABLISHED) { 1548 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1549 } 1550 INP_WLOCK_ASSERT(tp->t_inpcb); 1551 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1552 __func__)); 1553 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1554 __func__)); 1555 1556 #ifdef TCPPCAP 1557 /* Save segment, if requested. */ 1558 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1559 #endif 1560 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1561 tlen, NULL, true); 1562 1563 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1564 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1565 log(LOG_DEBUG, "%s; %s: " 1566 "SYN|FIN segment ignored (based on " 1567 "sysctl setting)\n", s, __func__); 1568 free(s, M_TCPLOG); 1569 } 1570 goto drop; 1571 } 1572 1573 /* 1574 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1575 * check SEQ.ACK first. 1576 */ 1577 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1578 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1579 rstreason = BANDLIM_UNLIMITED; 1580 goto dropwithreset; 1581 } 1582 1583 /* 1584 * Segment received on connection. 1585 * Reset idle time and keep-alive timer. 1586 * XXX: This should be done after segment 1587 * validation to ignore broken/spoofed segs. 1588 */ 1589 tp->t_rcvtime = ticks; 1590 1591 /* 1592 * Scale up the window into a 32-bit value. 1593 * For the SYN_SENT state the scale is zero. 1594 */ 1595 tiwin = th->th_win << tp->snd_scale; 1596 1597 /* 1598 * TCP ECN processing. 1599 */ 1600 if (tp->t_flags & TF_ECN_PERMIT) { 1601 if (thflags & TH_CWR) 1602 tp->t_flags &= ~TF_ECN_SND_ECE; 1603 switch (iptos & IPTOS_ECN_MASK) { 1604 case IPTOS_ECN_CE: 1605 tp->t_flags |= TF_ECN_SND_ECE; 1606 TCPSTAT_INC(tcps_ecn_ce); 1607 break; 1608 case IPTOS_ECN_ECT0: 1609 TCPSTAT_INC(tcps_ecn_ect0); 1610 break; 1611 case IPTOS_ECN_ECT1: 1612 TCPSTAT_INC(tcps_ecn_ect1); 1613 break; 1614 } 1615 1616 /* Process a packet differently from RFC3168. */ 1617 cc_ecnpkt_handler(tp, th, iptos); 1618 1619 /* Congestion experienced. */ 1620 if (thflags & TH_ECE) { 1621 cc_cong_signal(tp, th, CC_ECN); 1622 } 1623 } 1624 1625 /* 1626 * Parse options on any incoming segment. 1627 */ 1628 tcp_dooptions(&to, (u_char *)(th + 1), 1629 (th->th_off << 2) - sizeof(struct tcphdr), 1630 (thflags & TH_SYN) ? TO_SYN : 0); 1631 1632 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1633 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1634 (to.to_flags & TOF_SIGNATURE) == 0) { 1635 TCPSTAT_INC(tcps_sig_err_sigopt); 1636 /* XXX: should drop? */ 1637 } 1638 #endif 1639 /* 1640 * If echoed timestamp is later than the current time, 1641 * fall back to non RFC1323 RTT calculation. Normalize 1642 * timestamp if syncookies were used when this connection 1643 * was established. 1644 */ 1645 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1646 to.to_tsecr -= tp->ts_offset; 1647 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1648 to.to_tsecr = 0; 1649 else if (tp->t_flags & TF_PREVVALID && 1650 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1651 cc_cong_signal(tp, th, CC_RTO_ERR); 1652 } 1653 /* 1654 * Process options only when we get SYN/ACK back. The SYN case 1655 * for incoming connections is handled in tcp_syncache. 1656 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1657 * or <SYN,ACK>) segment itself is never scaled. 1658 * XXX this is traditional behavior, may need to be cleaned up. 1659 */ 1660 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1661 if ((to.to_flags & TOF_SCALE) && 1662 (tp->t_flags & TF_REQ_SCALE)) { 1663 tp->t_flags |= TF_RCVD_SCALE; 1664 tp->snd_scale = to.to_wscale; 1665 } 1666 /* 1667 * Initial send window. It will be updated with 1668 * the next incoming segment to the scaled value. 1669 */ 1670 tp->snd_wnd = th->th_win; 1671 if (to.to_flags & TOF_TS) { 1672 tp->t_flags |= TF_RCVD_TSTMP; 1673 tp->ts_recent = to.to_tsval; 1674 tp->ts_recent_age = tcp_ts_getticks(); 1675 } 1676 if (to.to_flags & TOF_MSS) 1677 tcp_mss(tp, to.to_mss); 1678 if ((tp->t_flags & TF_SACK_PERMIT) && 1679 (to.to_flags & TOF_SACKPERM) == 0) 1680 tp->t_flags &= ~TF_SACK_PERMIT; 1681 if (IS_FASTOPEN(tp->t_flags)) { 1682 if (to.to_flags & TOF_FASTOPEN) { 1683 uint16_t mss; 1684 1685 if (to.to_flags & TOF_MSS) 1686 mss = to.to_mss; 1687 else 1688 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) 1689 mss = TCP6_MSS; 1690 else 1691 mss = TCP_MSS; 1692 tcp_fastopen_update_cache(tp, mss, 1693 to.to_tfo_len, to.to_tfo_cookie); 1694 } else 1695 tcp_fastopen_disable_path(tp); 1696 } 1697 } 1698 1699 /* 1700 * If timestamps were negotiated during SYN/ACK they should 1701 * appear on every segment during this session and vice versa. 1702 */ 1703 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1704 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1705 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1706 "no action\n", s, __func__); 1707 free(s, M_TCPLOG); 1708 } 1709 } 1710 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1711 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1712 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1713 "no action\n", s, __func__); 1714 free(s, M_TCPLOG); 1715 } 1716 } 1717 1718 /* 1719 * Header prediction: check for the two common cases 1720 * of a uni-directional data xfer. If the packet has 1721 * no control flags, is in-sequence, the window didn't 1722 * change and we're not retransmitting, it's a 1723 * candidate. If the length is zero and the ack moved 1724 * forward, we're the sender side of the xfer. Just 1725 * free the data acked & wake any higher level process 1726 * that was blocked waiting for space. If the length 1727 * is non-zero and the ack didn't move, we're the 1728 * receiver side. If we're getting packets in-order 1729 * (the reassembly queue is empty), add the data to 1730 * the socket buffer and note that we need a delayed ack. 1731 * Make sure that the hidden state-flags are also off. 1732 * Since we check for TCPS_ESTABLISHED first, it can only 1733 * be TH_NEEDSYN. 1734 */ 1735 if (tp->t_state == TCPS_ESTABLISHED && 1736 th->th_seq == tp->rcv_nxt && 1737 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1738 tp->snd_nxt == tp->snd_max && 1739 tiwin && tiwin == tp->snd_wnd && 1740 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1741 SEGQ_EMPTY(tp) && 1742 ((to.to_flags & TOF_TS) == 0 || 1743 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1744 1745 /* 1746 * If last ACK falls within this segment's sequence numbers, 1747 * record the timestamp. 1748 * NOTE that the test is modified according to the latest 1749 * proposal of the [email protected] list (Braden 1993/04/26). 1750 */ 1751 if ((to.to_flags & TOF_TS) != 0 && 1752 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1753 tp->ts_recent_age = tcp_ts_getticks(); 1754 tp->ts_recent = to.to_tsval; 1755 } 1756 1757 if (tlen == 0) { 1758 if (SEQ_GT(th->th_ack, tp->snd_una) && 1759 SEQ_LEQ(th->th_ack, tp->snd_max) && 1760 !IN_RECOVERY(tp->t_flags) && 1761 (to.to_flags & TOF_SACK) == 0 && 1762 TAILQ_EMPTY(&tp->snd_holes)) { 1763 /* 1764 * This is a pure ack for outstanding data. 1765 */ 1766 TCPSTAT_INC(tcps_predack); 1767 1768 /* 1769 * "bad retransmit" recovery without timestamps. 1770 */ 1771 if ((to.to_flags & TOF_TS) == 0 && 1772 tp->t_rxtshift == 1 && 1773 tp->t_flags & TF_PREVVALID && 1774 (int)(ticks - tp->t_badrxtwin) < 0) { 1775 cc_cong_signal(tp, th, CC_RTO_ERR); 1776 } 1777 1778 /* 1779 * Recalculate the transmit timer / rtt. 1780 * 1781 * Some boxes send broken timestamp replies 1782 * during the SYN+ACK phase, ignore 1783 * timestamps of 0 or we could calculate a 1784 * huge RTT and blow up the retransmit timer. 1785 */ 1786 if ((to.to_flags & TOF_TS) != 0 && 1787 to.to_tsecr) { 1788 uint32_t t; 1789 1790 t = tcp_ts_getticks() - to.to_tsecr; 1791 if (!tp->t_rttlow || tp->t_rttlow > t) 1792 tp->t_rttlow = t; 1793 tcp_xmit_timer(tp, 1794 TCP_TS_TO_TICKS(t) + 1); 1795 } else if (tp->t_rtttime && 1796 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1797 if (!tp->t_rttlow || 1798 tp->t_rttlow > ticks - tp->t_rtttime) 1799 tp->t_rttlow = ticks - tp->t_rtttime; 1800 tcp_xmit_timer(tp, 1801 ticks - tp->t_rtttime); 1802 } 1803 acked = BYTES_THIS_ACK(tp, th); 1804 1805 #ifdef TCP_HHOOK 1806 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1807 hhook_run_tcp_est_in(tp, th, &to); 1808 #endif 1809 1810 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1811 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1812 sbdrop(&so->so_snd, acked); 1813 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1814 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1815 tp->snd_recover = th->th_ack - 1; 1816 1817 /* 1818 * Let the congestion control algorithm update 1819 * congestion control related information. This 1820 * typically means increasing the congestion 1821 * window. 1822 */ 1823 cc_ack_received(tp, th, nsegs, CC_ACK); 1824 1825 tp->snd_una = th->th_ack; 1826 /* 1827 * Pull snd_wl2 up to prevent seq wrap relative 1828 * to th_ack. 1829 */ 1830 tp->snd_wl2 = th->th_ack; 1831 tp->t_dupacks = 0; 1832 m_freem(m); 1833 1834 /* 1835 * If all outstanding data are acked, stop 1836 * retransmit timer, otherwise restart timer 1837 * using current (possibly backed-off) value. 1838 * If process is waiting for space, 1839 * wakeup/selwakeup/signal. If data 1840 * are ready to send, let tcp_output 1841 * decide between more output or persist. 1842 */ 1843 #ifdef TCPDEBUG 1844 if (so->so_options & SO_DEBUG) 1845 tcp_trace(TA_INPUT, ostate, tp, 1846 (void *)tcp_saveipgen, 1847 &tcp_savetcp, 0); 1848 #endif 1849 TCP_PROBE3(debug__input, tp, th, m); 1850 if (tp->snd_una == tp->snd_max) 1851 tcp_timer_activate(tp, TT_REXMT, 0); 1852 else if (!tcp_timer_active(tp, TT_PERSIST)) 1853 tcp_timer_activate(tp, TT_REXMT, 1854 tp->t_rxtcur); 1855 sowwakeup(so); 1856 if (sbavail(&so->so_snd)) 1857 (void) tp->t_fb->tfb_tcp_output(tp); 1858 goto check_delack; 1859 } 1860 } else if (th->th_ack == tp->snd_una && 1861 tlen <= sbspace(&so->so_rcv)) { 1862 int newsize = 0; /* automatic sockbuf scaling */ 1863 1864 /* 1865 * This is a pure, in-sequence data packet with 1866 * nothing on the reassembly queue and we have enough 1867 * buffer space to take it. 1868 */ 1869 /* Clean receiver SACK report if present */ 1870 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1871 tcp_clean_sackreport(tp); 1872 TCPSTAT_INC(tcps_preddat); 1873 tp->rcv_nxt += tlen; 1874 /* 1875 * Pull snd_wl1 up to prevent seq wrap relative to 1876 * th_seq. 1877 */ 1878 tp->snd_wl1 = th->th_seq; 1879 /* 1880 * Pull rcv_up up to prevent seq wrap relative to 1881 * rcv_nxt. 1882 */ 1883 tp->rcv_up = tp->rcv_nxt; 1884 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1885 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1886 #ifdef TCPDEBUG 1887 if (so->so_options & SO_DEBUG) 1888 tcp_trace(TA_INPUT, ostate, tp, 1889 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1890 #endif 1891 TCP_PROBE3(debug__input, tp, th, m); 1892 1893 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1894 1895 /* Add data to socket buffer. */ 1896 SOCKBUF_LOCK(&so->so_rcv); 1897 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1898 m_freem(m); 1899 } else { 1900 /* 1901 * Set new socket buffer size. 1902 * Give up when limit is reached. 1903 */ 1904 if (newsize) 1905 if (!sbreserve_locked(&so->so_rcv, 1906 newsize, so, NULL)) 1907 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1908 m_adj(m, drop_hdrlen); /* delayed header drop */ 1909 sbappendstream_locked(&so->so_rcv, m, 0); 1910 } 1911 /* NB: sorwakeup_locked() does an implicit unlock. */ 1912 sorwakeup_locked(so); 1913 if (DELAY_ACK(tp, tlen)) { 1914 tp->t_flags |= TF_DELACK; 1915 } else { 1916 tp->t_flags |= TF_ACKNOW; 1917 tp->t_fb->tfb_tcp_output(tp); 1918 } 1919 goto check_delack; 1920 } 1921 } 1922 1923 /* 1924 * Calculate amount of space in receive window, 1925 * and then do TCP input processing. 1926 * Receive window is amount of space in rcv queue, 1927 * but not less than advertised window. 1928 */ 1929 win = sbspace(&so->so_rcv); 1930 if (win < 0) 1931 win = 0; 1932 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1933 1934 switch (tp->t_state) { 1935 1936 /* 1937 * If the state is SYN_RECEIVED: 1938 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1939 */ 1940 case TCPS_SYN_RECEIVED: 1941 if ((thflags & TH_ACK) && 1942 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1943 SEQ_GT(th->th_ack, tp->snd_max))) { 1944 rstreason = BANDLIM_RST_OPENPORT; 1945 goto dropwithreset; 1946 } 1947 if (IS_FASTOPEN(tp->t_flags)) { 1948 /* 1949 * When a TFO connection is in SYN_RECEIVED, the 1950 * only valid packets are the initial SYN, a 1951 * retransmit/copy of the initial SYN (possibly with 1952 * a subset of the original data), a valid ACK, a 1953 * FIN, or a RST. 1954 */ 1955 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1956 rstreason = BANDLIM_RST_OPENPORT; 1957 goto dropwithreset; 1958 } else if (thflags & TH_SYN) { 1959 /* non-initial SYN is ignored */ 1960 if ((tcp_timer_active(tp, TT_DELACK) || 1961 tcp_timer_active(tp, TT_REXMT))) 1962 goto drop; 1963 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1964 goto drop; 1965 } 1966 } 1967 break; 1968 1969 /* 1970 * If the state is SYN_SENT: 1971 * if seg contains a RST with valid ACK (SEQ.ACK has already 1972 * been verified), then drop the connection. 1973 * if seg contains a RST without an ACK, drop the seg. 1974 * if seg does not contain SYN, then drop the seg. 1975 * Otherwise this is an acceptable SYN segment 1976 * initialize tp->rcv_nxt and tp->irs 1977 * if seg contains ack then advance tp->snd_una 1978 * if seg contains an ECE and ECN support is enabled, the stream 1979 * is ECN capable. 1980 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1981 * arrange for segment to be acked (eventually) 1982 * continue processing rest of data/controls, beginning with URG 1983 */ 1984 case TCPS_SYN_SENT: 1985 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1986 TCP_PROBE5(connect__refused, NULL, tp, 1987 m, 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 int tfo_partial_ack = 0; 1999 2000 TCPSTAT_INC(tcps_connects); 2001 soisconnected(so); 2002 #ifdef MAC 2003 mac_socketpeer_set_from_mbuf(m, so); 2004 #endif 2005 /* Do window scaling on this connection? */ 2006 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2007 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2008 tp->rcv_scale = tp->request_r_scale; 2009 } 2010 tp->rcv_adv += min(tp->rcv_wnd, 2011 TCP_MAXWIN << tp->rcv_scale); 2012 tp->snd_una++; /* SYN is acked */ 2013 /* 2014 * If not all the data that was sent in the TFO SYN 2015 * has been acked, resend the remainder right away. 2016 */ 2017 if (IS_FASTOPEN(tp->t_flags) && 2018 (tp->snd_una != tp->snd_max)) { 2019 tp->snd_nxt = th->th_ack; 2020 tfo_partial_ack = 1; 2021 } 2022 /* 2023 * If there's data, delay ACK; if there's also a FIN 2024 * ACKNOW will be turned on later. 2025 */ 2026 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2027 tcp_timer_activate(tp, TT_DELACK, 2028 tcp_delacktime); 2029 else 2030 tp->t_flags |= TF_ACKNOW; 2031 2032 if (((thflags & (TH_CWR | TH_ECE)) == TH_ECE) && 2033 V_tcp_do_ecn) { 2034 tp->t_flags |= TF_ECN_PERMIT; 2035 TCPSTAT_INC(tcps_ecn_shs); 2036 } 2037 2038 /* 2039 * Received <SYN,ACK> in SYN_SENT[*] state. 2040 * Transitions: 2041 * SYN_SENT --> ESTABLISHED 2042 * SYN_SENT* --> FIN_WAIT_1 2043 */ 2044 tp->t_starttime = ticks; 2045 if (tp->t_flags & TF_NEEDFIN) { 2046 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2047 tp->t_flags &= ~TF_NEEDFIN; 2048 thflags &= ~TH_SYN; 2049 } else { 2050 tcp_state_change(tp, TCPS_ESTABLISHED); 2051 TCP_PROBE5(connect__established, NULL, tp, 2052 m, tp, th); 2053 cc_conn_init(tp); 2054 tcp_timer_activate(tp, TT_KEEP, 2055 TP_KEEPIDLE(tp)); 2056 } 2057 } else { 2058 /* 2059 * Received initial SYN in SYN-SENT[*] state => 2060 * simultaneous open. 2061 * If it succeeds, connection is * half-synchronized. 2062 * Otherwise, do 3-way handshake: 2063 * SYN-SENT -> SYN-RECEIVED 2064 * SYN-SENT* -> SYN-RECEIVED* 2065 */ 2066 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2067 tcp_timer_activate(tp, TT_REXMT, 0); 2068 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2069 } 2070 2071 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2072 INP_WLOCK_ASSERT(tp->t_inpcb); 2073 2074 /* 2075 * Advance th->th_seq to correspond to first data byte. 2076 * If data, trim to stay within window, 2077 * dropping FIN if necessary. 2078 */ 2079 th->th_seq++; 2080 if (tlen > tp->rcv_wnd) { 2081 todrop = tlen - tp->rcv_wnd; 2082 m_adj(m, -todrop); 2083 tlen = tp->rcv_wnd; 2084 thflags &= ~TH_FIN; 2085 TCPSTAT_INC(tcps_rcvpackafterwin); 2086 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2087 } 2088 tp->snd_wl1 = th->th_seq - 1; 2089 tp->rcv_up = th->th_seq; 2090 /* 2091 * Client side of transaction: already sent SYN and data. 2092 * If the remote host used T/TCP to validate the SYN, 2093 * our data will be ACK'd; if so, enter normal data segment 2094 * processing in the middle of step 5, ack processing. 2095 * Otherwise, goto step 6. 2096 */ 2097 if (thflags & TH_ACK) 2098 goto process_ACK; 2099 2100 goto step6; 2101 2102 /* 2103 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2104 * do normal processing. 2105 * 2106 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2107 */ 2108 case TCPS_LAST_ACK: 2109 case TCPS_CLOSING: 2110 break; /* continue normal processing */ 2111 } 2112 2113 /* 2114 * States other than LISTEN or SYN_SENT. 2115 * First check the RST flag and sequence number since reset segments 2116 * are exempt from the timestamp and connection count tests. This 2117 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2118 * below which allowed reset segments in half the sequence space 2119 * to fall though and be processed (which gives forged reset 2120 * segments with a random sequence number a 50 percent chance of 2121 * killing a connection). 2122 * Then check timestamp, if present. 2123 * Then check the connection count, if present. 2124 * Then check that at least some bytes of segment are within 2125 * receive window. If segment begins before rcv_nxt, 2126 * drop leading data (and SYN); if nothing left, just ack. 2127 */ 2128 if (thflags & TH_RST) { 2129 /* 2130 * RFC5961 Section 3.2 2131 * 2132 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2133 * - If RST is in window, we send challenge ACK. 2134 * 2135 * Note: to take into account delayed ACKs, we should 2136 * test against last_ack_sent instead of rcv_nxt. 2137 * Note 2: we handle special case of closed window, not 2138 * covered by the RFC. 2139 */ 2140 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2141 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2142 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2143 2144 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2145 KASSERT(tp->t_state != TCPS_SYN_SENT, 2146 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2147 __func__, th, tp)); 2148 2149 if (V_tcp_insecure_rst || 2150 tp->last_ack_sent == th->th_seq) { 2151 TCPSTAT_INC(tcps_drops); 2152 /* Drop the connection. */ 2153 switch (tp->t_state) { 2154 case TCPS_SYN_RECEIVED: 2155 so->so_error = ECONNREFUSED; 2156 goto close; 2157 case TCPS_ESTABLISHED: 2158 case TCPS_FIN_WAIT_1: 2159 case TCPS_FIN_WAIT_2: 2160 case TCPS_CLOSE_WAIT: 2161 case TCPS_CLOSING: 2162 case TCPS_LAST_ACK: 2163 so->so_error = ECONNRESET; 2164 close: 2165 /* FALLTHROUGH */ 2166 default: 2167 tp = tcp_close(tp); 2168 } 2169 } else { 2170 TCPSTAT_INC(tcps_badrst); 2171 /* Send challenge ACK. */ 2172 tcp_respond(tp, mtod(m, void *), th, m, 2173 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2174 tp->last_ack_sent = tp->rcv_nxt; 2175 m = NULL; 2176 } 2177 } 2178 goto drop; 2179 } 2180 2181 /* 2182 * RFC5961 Section 4.2 2183 * Send challenge ACK for any SYN in synchronized state. 2184 */ 2185 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2186 tp->t_state != TCPS_SYN_RECEIVED) { 2187 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2188 2189 TCPSTAT_INC(tcps_badsyn); 2190 if (V_tcp_insecure_syn && 2191 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2192 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2193 tp = tcp_drop(tp, ECONNRESET); 2194 rstreason = BANDLIM_UNLIMITED; 2195 } else { 2196 /* Send challenge ACK. */ 2197 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2198 tp->snd_nxt, TH_ACK); 2199 tp->last_ack_sent = tp->rcv_nxt; 2200 m = NULL; 2201 } 2202 goto drop; 2203 } 2204 2205 /* 2206 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2207 * and it's less than ts_recent, drop it. 2208 */ 2209 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2210 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2211 2212 /* Check to see if ts_recent is over 24 days old. */ 2213 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2214 /* 2215 * Invalidate ts_recent. If this segment updates 2216 * ts_recent, the age will be reset later and ts_recent 2217 * will get a valid value. If it does not, setting 2218 * ts_recent to zero will at least satisfy the 2219 * requirement that zero be placed in the timestamp 2220 * echo reply when ts_recent isn't valid. The 2221 * age isn't reset until we get a valid ts_recent 2222 * because we don't want out-of-order segments to be 2223 * dropped when ts_recent is old. 2224 */ 2225 tp->ts_recent = 0; 2226 } else { 2227 TCPSTAT_INC(tcps_rcvduppack); 2228 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2229 TCPSTAT_INC(tcps_pawsdrop); 2230 if (tlen) 2231 goto dropafterack; 2232 goto drop; 2233 } 2234 } 2235 2236 /* 2237 * In the SYN-RECEIVED state, validate that the packet belongs to 2238 * this connection before trimming the data to fit the receive 2239 * window. Check the sequence number versus IRS since we know 2240 * the sequence numbers haven't wrapped. This is a partial fix 2241 * for the "LAND" DoS attack. 2242 */ 2243 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2244 rstreason = BANDLIM_RST_OPENPORT; 2245 goto dropwithreset; 2246 } 2247 2248 todrop = tp->rcv_nxt - th->th_seq; 2249 if (todrop > 0) { 2250 if (thflags & TH_SYN) { 2251 thflags &= ~TH_SYN; 2252 th->th_seq++; 2253 if (th->th_urp > 1) 2254 th->th_urp--; 2255 else 2256 thflags &= ~TH_URG; 2257 todrop--; 2258 } 2259 /* 2260 * Following if statement from Stevens, vol. 2, p. 960. 2261 */ 2262 if (todrop > tlen 2263 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2264 /* 2265 * Any valid FIN must be to the left of the window. 2266 * At this point the FIN must be a duplicate or out 2267 * of sequence; drop it. 2268 */ 2269 thflags &= ~TH_FIN; 2270 2271 /* 2272 * Send an ACK to resynchronize and drop any data. 2273 * But keep on processing for RST or ACK. 2274 */ 2275 tp->t_flags |= TF_ACKNOW; 2276 todrop = tlen; 2277 TCPSTAT_INC(tcps_rcvduppack); 2278 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2279 } else { 2280 TCPSTAT_INC(tcps_rcvpartduppack); 2281 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2282 } 2283 /* 2284 * DSACK - add SACK block for dropped range 2285 */ 2286 if (tp->t_flags & TF_SACK_PERMIT) { 2287 tcp_update_sack_list(tp, th->th_seq, 2288 th->th_seq + todrop); 2289 /* 2290 * ACK now, as the next in-sequence segment 2291 * will clear the DSACK block again 2292 */ 2293 tp->t_flags |= TF_ACKNOW; 2294 } 2295 drop_hdrlen += todrop; /* drop from the top afterwards */ 2296 th->th_seq += todrop; 2297 tlen -= todrop; 2298 if (th->th_urp > todrop) 2299 th->th_urp -= todrop; 2300 else { 2301 thflags &= ~TH_URG; 2302 th->th_urp = 0; 2303 } 2304 } 2305 2306 /* 2307 * If new data are received on a connection after the 2308 * user processes are gone, then RST the other end. 2309 */ 2310 if ((so->so_state & SS_NOFDREF) && 2311 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2312 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2313 2314 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2315 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2316 "after socket was closed, " 2317 "sending RST and removing tcpcb\n", 2318 s, __func__, tcpstates[tp->t_state], tlen); 2319 free(s, M_TCPLOG); 2320 } 2321 tp = tcp_close(tp); 2322 TCPSTAT_INC(tcps_rcvafterclose); 2323 rstreason = BANDLIM_UNLIMITED; 2324 goto dropwithreset; 2325 } 2326 2327 /* 2328 * If segment ends after window, drop trailing data 2329 * (and PUSH and FIN); if nothing left, just ACK. 2330 */ 2331 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2332 if (todrop > 0) { 2333 TCPSTAT_INC(tcps_rcvpackafterwin); 2334 if (todrop >= tlen) { 2335 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2336 /* 2337 * If window is closed can only take segments at 2338 * window edge, and have to drop data and PUSH from 2339 * incoming segments. Continue processing, but 2340 * remember to ack. Otherwise, drop segment 2341 * and ack. 2342 */ 2343 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2344 tp->t_flags |= TF_ACKNOW; 2345 TCPSTAT_INC(tcps_rcvwinprobe); 2346 } else 2347 goto dropafterack; 2348 } else 2349 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2350 m_adj(m, -todrop); 2351 tlen -= todrop; 2352 thflags &= ~(TH_PUSH|TH_FIN); 2353 } 2354 2355 /* 2356 * If last ACK falls within this segment's sequence numbers, 2357 * record its timestamp. 2358 * NOTE: 2359 * 1) That the test incorporates suggestions from the latest 2360 * proposal of the [email protected] list (Braden 1993/04/26). 2361 * 2) That updating only on newer timestamps interferes with 2362 * our earlier PAWS tests, so this check should be solely 2363 * predicated on the sequence space of this segment. 2364 * 3) That we modify the segment boundary check to be 2365 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2366 * instead of RFC1323's 2367 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2368 * This modified check allows us to overcome RFC1323's 2369 * limitations as described in Stevens TCP/IP Illustrated 2370 * Vol. 2 p.869. In such cases, we can still calculate the 2371 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2372 */ 2373 if ((to.to_flags & TOF_TS) != 0 && 2374 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2375 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2376 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2377 tp->ts_recent_age = tcp_ts_getticks(); 2378 tp->ts_recent = to.to_tsval; 2379 } 2380 2381 /* 2382 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2383 * flag is on (half-synchronized state), then queue data for 2384 * later processing; else drop segment and return. 2385 */ 2386 if ((thflags & TH_ACK) == 0) { 2387 if (tp->t_state == TCPS_SYN_RECEIVED || 2388 (tp->t_flags & TF_NEEDSYN)) { 2389 if (tp->t_state == TCPS_SYN_RECEIVED && 2390 IS_FASTOPEN(tp->t_flags)) { 2391 tp->snd_wnd = tiwin; 2392 cc_conn_init(tp); 2393 } 2394 goto step6; 2395 } else if (tp->t_flags & TF_ACKNOW) 2396 goto dropafterack; 2397 else 2398 goto drop; 2399 } 2400 2401 /* 2402 * Ack processing. 2403 */ 2404 switch (tp->t_state) { 2405 2406 /* 2407 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2408 * ESTABLISHED state and continue processing. 2409 * The ACK was checked above. 2410 */ 2411 case TCPS_SYN_RECEIVED: 2412 2413 TCPSTAT_INC(tcps_connects); 2414 soisconnected(so); 2415 /* Do window scaling? */ 2416 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2417 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2418 tp->rcv_scale = tp->request_r_scale; 2419 } 2420 tp->snd_wnd = tiwin; 2421 /* 2422 * Make transitions: 2423 * SYN-RECEIVED -> ESTABLISHED 2424 * SYN-RECEIVED* -> FIN-WAIT-1 2425 */ 2426 tp->t_starttime = ticks; 2427 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2428 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2429 tp->t_tfo_pending = NULL; 2430 2431 /* 2432 * Account for the ACK of our SYN prior to 2433 * regular ACK processing below. 2434 */ 2435 tp->snd_una++; 2436 } 2437 if (tp->t_flags & TF_NEEDFIN) { 2438 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2439 tp->t_flags &= ~TF_NEEDFIN; 2440 } else { 2441 tcp_state_change(tp, TCPS_ESTABLISHED); 2442 TCP_PROBE5(accept__established, NULL, tp, 2443 m, tp, th); 2444 /* 2445 * TFO connections call cc_conn_init() during SYN 2446 * processing. Calling it again here for such 2447 * connections is not harmless as it would undo the 2448 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2449 * is retransmitted. 2450 */ 2451 if (!IS_FASTOPEN(tp->t_flags)) 2452 cc_conn_init(tp); 2453 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2454 } 2455 /* 2456 * If segment contains data or ACK, will call tcp_reass() 2457 * later; if not, do so now to pass queued data to user. 2458 */ 2459 if (tlen == 0 && (thflags & TH_FIN) == 0) 2460 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, 2461 (struct mbuf *)0); 2462 tp->snd_wl1 = th->th_seq - 1; 2463 /* FALLTHROUGH */ 2464 2465 /* 2466 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2467 * ACKs. If the ack is in the range 2468 * tp->snd_una < th->th_ack <= tp->snd_max 2469 * then advance tp->snd_una to th->th_ack and drop 2470 * data from the retransmission queue. If this ACK reflects 2471 * more up to date window information we update our window information. 2472 */ 2473 case TCPS_ESTABLISHED: 2474 case TCPS_FIN_WAIT_1: 2475 case TCPS_FIN_WAIT_2: 2476 case TCPS_CLOSE_WAIT: 2477 case TCPS_CLOSING: 2478 case TCPS_LAST_ACK: 2479 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2480 TCPSTAT_INC(tcps_rcvacktoomuch); 2481 goto dropafterack; 2482 } 2483 if ((tp->t_flags & TF_SACK_PERMIT) && 2484 ((to.to_flags & TOF_SACK) || 2485 !TAILQ_EMPTY(&tp->snd_holes))) 2486 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2487 else 2488 /* 2489 * Reset the value so that previous (valid) value 2490 * from the last ack with SACK doesn't get used. 2491 */ 2492 tp->sackhint.sacked_bytes = 0; 2493 2494 #ifdef TCP_HHOOK 2495 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2496 hhook_run_tcp_est_in(tp, th, &to); 2497 #endif 2498 2499 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2500 u_int maxseg; 2501 2502 maxseg = tcp_maxseg(tp); 2503 if (tlen == 0 && 2504 (tiwin == tp->snd_wnd || 2505 (tp->t_flags & TF_SACK_PERMIT))) { 2506 /* 2507 * If this is the first time we've seen a 2508 * FIN from the remote, this is not a 2509 * duplicate and it needs to be processed 2510 * normally. This happens during a 2511 * simultaneous close. 2512 */ 2513 if ((thflags & TH_FIN) && 2514 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2515 tp->t_dupacks = 0; 2516 break; 2517 } 2518 TCPSTAT_INC(tcps_rcvdupack); 2519 /* 2520 * If we have outstanding data (other than 2521 * a window probe), this is a completely 2522 * duplicate ack (ie, window info didn't 2523 * change and FIN isn't set), 2524 * the ack is the biggest we've 2525 * seen and we've seen exactly our rexmt 2526 * threshold of them, assume a packet 2527 * has been dropped and retransmit it. 2528 * Kludge snd_nxt & the congestion 2529 * window so we send only this one 2530 * packet. 2531 * 2532 * We know we're losing at the current 2533 * window size so do congestion avoidance 2534 * (set ssthresh to half the current window 2535 * and pull our congestion window back to 2536 * the new ssthresh). 2537 * 2538 * Dup acks mean that packets have left the 2539 * network (they're now cached at the receiver) 2540 * so bump cwnd by the amount in the receiver 2541 * to keep a constant cwnd packets in the 2542 * network. 2543 * 2544 * When using TCP ECN, notify the peer that 2545 * we reduced the cwnd. 2546 */ 2547 /* 2548 * Following 2 kinds of acks should not affect 2549 * dupack counting: 2550 * 1) Old acks 2551 * 2) Acks with SACK but without any new SACK 2552 * information in them. These could result from 2553 * any anomaly in the network like a switch 2554 * duplicating packets or a possible DoS attack. 2555 */ 2556 if (th->th_ack != tp->snd_una || 2557 ((tp->t_flags & TF_SACK_PERMIT) && 2558 !sack_changed)) 2559 break; 2560 else if (!tcp_timer_active(tp, TT_REXMT)) 2561 tp->t_dupacks = 0; 2562 else if (++tp->t_dupacks > tcprexmtthresh || 2563 IN_FASTRECOVERY(tp->t_flags)) { 2564 cc_ack_received(tp, th, nsegs, 2565 CC_DUPACK); 2566 if ((tp->t_flags & TF_SACK_PERMIT) && 2567 IN_FASTRECOVERY(tp->t_flags)) { 2568 int awnd; 2569 2570 /* 2571 * Compute the amount of data in flight first. 2572 * We can inject new data into the pipe iff 2573 * we have less than 1/2 the original window's 2574 * worth of data in flight. 2575 */ 2576 if (V_tcp_do_rfc6675_pipe) 2577 awnd = tcp_compute_pipe(tp); 2578 else 2579 awnd = (tp->snd_nxt - tp->snd_fack) + 2580 tp->sackhint.sack_bytes_rexmit; 2581 2582 if (awnd < tp->snd_ssthresh) { 2583 tp->snd_cwnd += maxseg; 2584 if (tp->snd_cwnd > tp->snd_ssthresh) 2585 tp->snd_cwnd = tp->snd_ssthresh; 2586 } 2587 } else 2588 tp->snd_cwnd += maxseg; 2589 (void) tp->t_fb->tfb_tcp_output(tp); 2590 goto drop; 2591 } else if (tp->t_dupacks == tcprexmtthresh) { 2592 tcp_seq onxt = tp->snd_nxt; 2593 2594 /* 2595 * If we're doing sack, check to 2596 * see if we're already in sack 2597 * recovery. If we're not doing sack, 2598 * check to see if we're in newreno 2599 * recovery. 2600 */ 2601 if (tp->t_flags & TF_SACK_PERMIT) { 2602 if (IN_FASTRECOVERY(tp->t_flags)) { 2603 tp->t_dupacks = 0; 2604 break; 2605 } 2606 } else { 2607 if (SEQ_LEQ(th->th_ack, 2608 tp->snd_recover)) { 2609 tp->t_dupacks = 0; 2610 break; 2611 } 2612 } 2613 /* Congestion signal before ack. */ 2614 cc_cong_signal(tp, th, CC_NDUPACK); 2615 cc_ack_received(tp, th, nsegs, 2616 CC_DUPACK); 2617 tcp_timer_activate(tp, TT_REXMT, 0); 2618 tp->t_rtttime = 0; 2619 if (tp->t_flags & TF_SACK_PERMIT) { 2620 TCPSTAT_INC( 2621 tcps_sack_recovery_episode); 2622 tp->sack_newdata = tp->snd_nxt; 2623 tp->snd_cwnd = maxseg; 2624 (void) tp->t_fb->tfb_tcp_output(tp); 2625 goto drop; 2626 } 2627 tp->snd_nxt = th->th_ack; 2628 tp->snd_cwnd = maxseg; 2629 (void) tp->t_fb->tfb_tcp_output(tp); 2630 KASSERT(tp->snd_limited <= 2, 2631 ("%s: tp->snd_limited too big", 2632 __func__)); 2633 tp->snd_cwnd = tp->snd_ssthresh + 2634 maxseg * 2635 (tp->t_dupacks - tp->snd_limited); 2636 if (SEQ_GT(onxt, tp->snd_nxt)) 2637 tp->snd_nxt = onxt; 2638 goto drop; 2639 } else if (V_tcp_do_rfc3042) { 2640 /* 2641 * Process first and second duplicate 2642 * ACKs. Each indicates a segment 2643 * leaving the network, creating room 2644 * for more. Make sure we can send a 2645 * packet on reception of each duplicate 2646 * ACK by increasing snd_cwnd by one 2647 * segment. Restore the original 2648 * snd_cwnd after packet transmission. 2649 */ 2650 cc_ack_received(tp, th, nsegs, 2651 CC_DUPACK); 2652 uint32_t oldcwnd = tp->snd_cwnd; 2653 tcp_seq oldsndmax = tp->snd_max; 2654 u_int sent; 2655 int avail; 2656 2657 KASSERT(tp->t_dupacks == 1 || 2658 tp->t_dupacks == 2, 2659 ("%s: dupacks not 1 or 2", 2660 __func__)); 2661 if (tp->t_dupacks == 1) 2662 tp->snd_limited = 0; 2663 tp->snd_cwnd = 2664 (tp->snd_nxt - tp->snd_una) + 2665 (tp->t_dupacks - tp->snd_limited) * 2666 maxseg; 2667 /* 2668 * Only call tcp_output when there 2669 * is new data available to be sent. 2670 * Otherwise we would send pure ACKs. 2671 */ 2672 SOCKBUF_LOCK(&so->so_snd); 2673 avail = sbavail(&so->so_snd) - 2674 (tp->snd_nxt - tp->snd_una); 2675 SOCKBUF_UNLOCK(&so->so_snd); 2676 if (avail > 0) 2677 (void) tp->t_fb->tfb_tcp_output(tp); 2678 sent = tp->snd_max - oldsndmax; 2679 if (sent > maxseg) { 2680 KASSERT((tp->t_dupacks == 2 && 2681 tp->snd_limited == 0) || 2682 (sent == maxseg + 1 && 2683 tp->t_flags & TF_SENTFIN), 2684 ("%s: sent too much", 2685 __func__)); 2686 tp->snd_limited = 2; 2687 } else if (sent > 0) 2688 ++tp->snd_limited; 2689 tp->snd_cwnd = oldcwnd; 2690 goto drop; 2691 } 2692 } 2693 break; 2694 } else { 2695 /* 2696 * This ack is advancing the left edge, reset the 2697 * counter. 2698 */ 2699 tp->t_dupacks = 0; 2700 /* 2701 * If this ack also has new SACK info, increment the 2702 * counter as per rfc6675. 2703 */ 2704 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2705 tp->t_dupacks++; 2706 } 2707 2708 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2709 ("%s: th_ack <= snd_una", __func__)); 2710 2711 /* 2712 * If the congestion window was inflated to account 2713 * for the other side's cached packets, retract it. 2714 */ 2715 if (IN_FASTRECOVERY(tp->t_flags)) { 2716 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2717 if (tp->t_flags & TF_SACK_PERMIT) 2718 tcp_sack_partialack(tp, th); 2719 else 2720 tcp_newreno_partial_ack(tp, th); 2721 } else 2722 cc_post_recovery(tp, th); 2723 } 2724 /* 2725 * If we reach this point, ACK is not a duplicate, 2726 * i.e., it ACKs something we sent. 2727 */ 2728 if (tp->t_flags & TF_NEEDSYN) { 2729 /* 2730 * T/TCP: Connection was half-synchronized, and our 2731 * SYN has been ACK'd (so connection is now fully 2732 * synchronized). Go to non-starred state, 2733 * increment snd_una for ACK of SYN, and check if 2734 * we can do window scaling. 2735 */ 2736 tp->t_flags &= ~TF_NEEDSYN; 2737 tp->snd_una++; 2738 /* Do window scaling? */ 2739 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2740 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2741 tp->rcv_scale = tp->request_r_scale; 2742 /* Send window already scaled. */ 2743 } 2744 } 2745 2746 process_ACK: 2747 INP_WLOCK_ASSERT(tp->t_inpcb); 2748 2749 acked = BYTES_THIS_ACK(tp, th); 2750 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2751 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2752 tp->snd_una, th->th_ack, tp, m)); 2753 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2754 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2755 2756 /* 2757 * If we just performed our first retransmit, and the ACK 2758 * arrives within our recovery window, then it was a mistake 2759 * to do the retransmit in the first place. Recover our 2760 * original cwnd and ssthresh, and proceed to transmit where 2761 * we left off. 2762 */ 2763 if (tp->t_rxtshift == 1 && 2764 tp->t_flags & TF_PREVVALID && 2765 tp->t_badrxtwin && 2766 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2767 cc_cong_signal(tp, th, CC_RTO_ERR); 2768 2769 /* 2770 * If we have a timestamp reply, update smoothed 2771 * round trip time. If no timestamp is present but 2772 * transmit timer is running and timed sequence 2773 * number was acked, update smoothed round trip time. 2774 * Since we now have an rtt measurement, cancel the 2775 * timer backoff (cf., Phil Karn's retransmit alg.). 2776 * Recompute the initial retransmit timer. 2777 * 2778 * Some boxes send broken timestamp replies 2779 * during the SYN+ACK phase, ignore 2780 * timestamps of 0 or we could calculate a 2781 * huge RTT and blow up the retransmit timer. 2782 */ 2783 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2784 uint32_t t; 2785 2786 t = tcp_ts_getticks() - to.to_tsecr; 2787 if (!tp->t_rttlow || tp->t_rttlow > t) 2788 tp->t_rttlow = t; 2789 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2790 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2791 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2792 tp->t_rttlow = ticks - tp->t_rtttime; 2793 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2794 } 2795 2796 /* 2797 * If all outstanding data is acked, stop retransmit 2798 * timer and remember to restart (more output or persist). 2799 * If there is more data to be acked, restart retransmit 2800 * timer, using current (possibly backed-off) value. 2801 */ 2802 if (th->th_ack == tp->snd_max) { 2803 tcp_timer_activate(tp, TT_REXMT, 0); 2804 needoutput = 1; 2805 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2806 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2807 2808 /* 2809 * If no data (only SYN) was ACK'd, 2810 * skip rest of ACK processing. 2811 */ 2812 if (acked == 0) 2813 goto step6; 2814 2815 /* 2816 * Let the congestion control algorithm update congestion 2817 * control related information. This typically means increasing 2818 * the congestion window. 2819 */ 2820 cc_ack_received(tp, th, nsegs, CC_ACK); 2821 2822 SOCKBUF_LOCK(&so->so_snd); 2823 if (acked > sbavail(&so->so_snd)) { 2824 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2825 tp->snd_wnd -= sbavail(&so->so_snd); 2826 else 2827 tp->snd_wnd = 0; 2828 mfree = sbcut_locked(&so->so_snd, 2829 (int)sbavail(&so->so_snd)); 2830 ourfinisacked = 1; 2831 } else { 2832 mfree = sbcut_locked(&so->so_snd, acked); 2833 if (tp->snd_wnd >= (uint32_t) acked) 2834 tp->snd_wnd -= acked; 2835 else 2836 tp->snd_wnd = 0; 2837 ourfinisacked = 0; 2838 } 2839 /* NB: sowwakeup_locked() does an implicit unlock. */ 2840 sowwakeup_locked(so); 2841 m_freem(mfree); 2842 /* Detect una wraparound. */ 2843 if (!IN_RECOVERY(tp->t_flags) && 2844 SEQ_GT(tp->snd_una, tp->snd_recover) && 2845 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2846 tp->snd_recover = th->th_ack - 1; 2847 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2848 if (IN_RECOVERY(tp->t_flags) && 2849 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2850 EXIT_RECOVERY(tp->t_flags); 2851 } 2852 tp->snd_una = th->th_ack; 2853 if (tp->t_flags & TF_SACK_PERMIT) { 2854 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2855 tp->snd_recover = tp->snd_una; 2856 } 2857 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2858 tp->snd_nxt = tp->snd_una; 2859 2860 switch (tp->t_state) { 2861 2862 /* 2863 * In FIN_WAIT_1 STATE in addition to the processing 2864 * for the ESTABLISHED state if our FIN is now acknowledged 2865 * then enter FIN_WAIT_2. 2866 */ 2867 case TCPS_FIN_WAIT_1: 2868 if (ourfinisacked) { 2869 /* 2870 * If we can't receive any more 2871 * data, then closing user can proceed. 2872 * Starting the timer is contrary to the 2873 * specification, but if we don't get a FIN 2874 * we'll hang forever. 2875 * 2876 * XXXjl: 2877 * we should release the tp also, and use a 2878 * compressed state. 2879 */ 2880 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2881 soisdisconnected(so); 2882 tcp_timer_activate(tp, TT_2MSL, 2883 (tcp_fast_finwait2_recycle ? 2884 tcp_finwait2_timeout : 2885 TP_MAXIDLE(tp))); 2886 } 2887 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2888 } 2889 break; 2890 2891 /* 2892 * In CLOSING STATE in addition to the processing for 2893 * the ESTABLISHED state if the ACK acknowledges our FIN 2894 * then enter the TIME-WAIT state, otherwise ignore 2895 * the segment. 2896 */ 2897 case TCPS_CLOSING: 2898 if (ourfinisacked) { 2899 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2900 tcp_twstart(tp); 2901 m_freem(m); 2902 return; 2903 } 2904 break; 2905 2906 /* 2907 * In LAST_ACK, we may still be waiting for data to drain 2908 * and/or to be acked, as well as for the ack of our FIN. 2909 * If our FIN is now acknowledged, delete the TCB, 2910 * enter the closed state and return. 2911 */ 2912 case TCPS_LAST_ACK: 2913 if (ourfinisacked) { 2914 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2915 tp = tcp_close(tp); 2916 goto drop; 2917 } 2918 break; 2919 } 2920 } 2921 2922 step6: 2923 INP_WLOCK_ASSERT(tp->t_inpcb); 2924 2925 /* 2926 * Update window information. 2927 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2928 */ 2929 if ((thflags & TH_ACK) && 2930 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2931 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2932 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2933 /* keep track of pure window updates */ 2934 if (tlen == 0 && 2935 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2936 TCPSTAT_INC(tcps_rcvwinupd); 2937 tp->snd_wnd = tiwin; 2938 tp->snd_wl1 = th->th_seq; 2939 tp->snd_wl2 = th->th_ack; 2940 if (tp->snd_wnd > tp->max_sndwnd) 2941 tp->max_sndwnd = tp->snd_wnd; 2942 needoutput = 1; 2943 } 2944 2945 /* 2946 * Process segments with URG. 2947 */ 2948 if ((thflags & TH_URG) && th->th_urp && 2949 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2950 /* 2951 * This is a kludge, but if we receive and accept 2952 * random urgent pointers, we'll crash in 2953 * soreceive. It's hard to imagine someone 2954 * actually wanting to send this much urgent data. 2955 */ 2956 SOCKBUF_LOCK(&so->so_rcv); 2957 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2958 th->th_urp = 0; /* XXX */ 2959 thflags &= ~TH_URG; /* XXX */ 2960 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2961 goto dodata; /* XXX */ 2962 } 2963 /* 2964 * If this segment advances the known urgent pointer, 2965 * then mark the data stream. This should not happen 2966 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2967 * a FIN has been received from the remote side. 2968 * In these states we ignore the URG. 2969 * 2970 * According to RFC961 (Assigned Protocols), 2971 * the urgent pointer points to the last octet 2972 * of urgent data. We continue, however, 2973 * to consider it to indicate the first octet 2974 * of data past the urgent section as the original 2975 * spec states (in one of two places). 2976 */ 2977 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2978 tp->rcv_up = th->th_seq + th->th_urp; 2979 so->so_oobmark = sbavail(&so->so_rcv) + 2980 (tp->rcv_up - tp->rcv_nxt) - 1; 2981 if (so->so_oobmark == 0) 2982 so->so_rcv.sb_state |= SBS_RCVATMARK; 2983 sohasoutofband(so); 2984 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2985 } 2986 SOCKBUF_UNLOCK(&so->so_rcv); 2987 /* 2988 * Remove out of band data so doesn't get presented to user. 2989 * This can happen independent of advancing the URG pointer, 2990 * but if two URG's are pending at once, some out-of-band 2991 * data may creep in... ick. 2992 */ 2993 if (th->th_urp <= (uint32_t)tlen && 2994 !(so->so_options & SO_OOBINLINE)) { 2995 /* hdr drop is delayed */ 2996 tcp_pulloutofband(so, th, m, drop_hdrlen); 2997 } 2998 } else { 2999 /* 3000 * If no out of band data is expected, 3001 * pull receive urgent pointer along 3002 * with the receive window. 3003 */ 3004 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3005 tp->rcv_up = tp->rcv_nxt; 3006 } 3007 dodata: /* XXX */ 3008 INP_WLOCK_ASSERT(tp->t_inpcb); 3009 3010 /* 3011 * Process the segment text, merging it into the TCP sequencing queue, 3012 * and arranging for acknowledgment of receipt if necessary. 3013 * This process logically involves adjusting tp->rcv_wnd as data 3014 * is presented to the user (this happens in tcp_usrreq.c, 3015 * case PRU_RCVD). If a FIN has already been received on this 3016 * connection then we just ignore the text. 3017 */ 3018 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3019 IS_FASTOPEN(tp->t_flags)); 3020 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3021 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3022 tcp_seq save_start = th->th_seq; 3023 tcp_seq save_rnxt = tp->rcv_nxt; 3024 int save_tlen = tlen; 3025 m_adj(m, drop_hdrlen); /* delayed header drop */ 3026 /* 3027 * Insert segment which includes th into TCP reassembly queue 3028 * with control block tp. Set thflags to whether reassembly now 3029 * includes a segment with FIN. This handles the common case 3030 * inline (segment is the next to be received on an established 3031 * connection, and the queue is empty), avoiding linkage into 3032 * and removal from the queue and repetition of various 3033 * conversions. 3034 * Set DELACK for segments received in order, but ack 3035 * immediately when segments are out of order (so 3036 * fast retransmit can work). 3037 */ 3038 if (th->th_seq == tp->rcv_nxt && 3039 SEGQ_EMPTY(tp) && 3040 (TCPS_HAVEESTABLISHED(tp->t_state) || 3041 tfo_syn)) { 3042 if (DELAY_ACK(tp, tlen) || tfo_syn) 3043 tp->t_flags |= TF_DELACK; 3044 else 3045 tp->t_flags |= TF_ACKNOW; 3046 tp->rcv_nxt += tlen; 3047 thflags = th->th_flags & TH_FIN; 3048 TCPSTAT_INC(tcps_rcvpack); 3049 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3050 SOCKBUF_LOCK(&so->so_rcv); 3051 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3052 m_freem(m); 3053 else 3054 sbappendstream_locked(&so->so_rcv, m, 0); 3055 /* NB: sorwakeup_locked() does an implicit unlock. */ 3056 sorwakeup_locked(so); 3057 } else { 3058 /* 3059 * XXX: Due to the header drop above "th" is 3060 * theoretically invalid by now. Fortunately 3061 * m_adj() doesn't actually frees any mbufs 3062 * when trimming from the head. 3063 */ 3064 tcp_seq temp = save_start; 3065 thflags = tcp_reass(tp, th, &temp, &tlen, m); 3066 tp->t_flags |= TF_ACKNOW; 3067 } 3068 if (tp->t_flags & TF_SACK_PERMIT) { 3069 if (((tlen == 0) && (save_tlen > 0) && 3070 (SEQ_LT(save_start, save_rnxt)))) { 3071 /* 3072 * DSACK actually handled in the fastpath 3073 * above. 3074 */ 3075 tcp_update_sack_list(tp, save_start, 3076 save_start + save_tlen); 3077 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { 3078 /* 3079 * Cleaning sackblks by using zero length 3080 * update. 3081 */ 3082 if ((tp->rcv_numsacks >= 1) && 3083 (tp->sackblks[0].end == save_start)) { 3084 /* partial overlap, recorded at todrop above */ 3085 tcp_update_sack_list(tp, tp->sackblks[0].start, 3086 tp->sackblks[0].end); 3087 } else { 3088 tcp_update_dsack_list(tp, save_start, 3089 save_start + save_tlen); 3090 } 3091 } else if ((tlen > 0) && (tlen >= save_tlen)) { 3092 /* Update of sackblks. */ 3093 tcp_update_dsack_list(tp, save_start, 3094 save_start + save_tlen); 3095 } else if (tlen > 0) { 3096 tcp_update_dsack_list(tp, save_start, 3097 save_start + tlen); 3098 } 3099 } 3100 #if 0 3101 /* 3102 * Note the amount of data that peer has sent into 3103 * our window, in order to estimate the sender's 3104 * buffer size. 3105 * XXX: Unused. 3106 */ 3107 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3108 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3109 else 3110 len = so->so_rcv.sb_hiwat; 3111 #endif 3112 } else { 3113 m_freem(m); 3114 thflags &= ~TH_FIN; 3115 } 3116 3117 /* 3118 * If FIN is received ACK the FIN and let the user know 3119 * that the connection is closing. 3120 */ 3121 if (thflags & TH_FIN) { 3122 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3123 socantrcvmore(so); 3124 /* 3125 * If connection is half-synchronized 3126 * (ie NEEDSYN flag on) then delay ACK, 3127 * so it may be piggybacked when SYN is sent. 3128 * Otherwise, since we received a FIN then no 3129 * more input can be expected, send ACK now. 3130 */ 3131 if (tp->t_flags & TF_NEEDSYN) 3132 tp->t_flags |= TF_DELACK; 3133 else 3134 tp->t_flags |= TF_ACKNOW; 3135 tp->rcv_nxt++; 3136 } 3137 switch (tp->t_state) { 3138 3139 /* 3140 * In SYN_RECEIVED and ESTABLISHED STATES 3141 * enter the CLOSE_WAIT state. 3142 */ 3143 case TCPS_SYN_RECEIVED: 3144 tp->t_starttime = ticks; 3145 /* FALLTHROUGH */ 3146 case TCPS_ESTABLISHED: 3147 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3148 break; 3149 3150 /* 3151 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3152 * enter the CLOSING state. 3153 */ 3154 case TCPS_FIN_WAIT_1: 3155 tcp_state_change(tp, TCPS_CLOSING); 3156 break; 3157 3158 /* 3159 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3160 * starting the time-wait timer, turning off the other 3161 * standard timers. 3162 */ 3163 case TCPS_FIN_WAIT_2: 3164 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3165 3166 tcp_twstart(tp); 3167 return; 3168 } 3169 } 3170 #ifdef TCPDEBUG 3171 if (so->so_options & SO_DEBUG) 3172 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3173 &tcp_savetcp, 0); 3174 #endif 3175 TCP_PROBE3(debug__input, tp, th, m); 3176 3177 /* 3178 * Return any desired output. 3179 */ 3180 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3181 (void) tp->t_fb->tfb_tcp_output(tp); 3182 3183 check_delack: 3184 INP_WLOCK_ASSERT(tp->t_inpcb); 3185 3186 if (tp->t_flags & TF_DELACK) { 3187 tp->t_flags &= ~TF_DELACK; 3188 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3189 } 3190 INP_WUNLOCK(tp->t_inpcb); 3191 return; 3192 3193 dropafterack: 3194 /* 3195 * Generate an ACK dropping incoming segment if it occupies 3196 * sequence space, where the ACK reflects our state. 3197 * 3198 * We can now skip the test for the RST flag since all 3199 * paths to this code happen after packets containing 3200 * RST have been dropped. 3201 * 3202 * In the SYN-RECEIVED state, don't send an ACK unless the 3203 * segment we received passes the SYN-RECEIVED ACK test. 3204 * If it fails send a RST. This breaks the loop in the 3205 * "LAND" DoS attack, and also prevents an ACK storm 3206 * between two listening ports that have been sent forged 3207 * SYN segments, each with the source address of the other. 3208 */ 3209 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3210 (SEQ_GT(tp->snd_una, th->th_ack) || 3211 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3212 rstreason = BANDLIM_RST_OPENPORT; 3213 goto dropwithreset; 3214 } 3215 #ifdef TCPDEBUG 3216 if (so->so_options & SO_DEBUG) 3217 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3218 &tcp_savetcp, 0); 3219 #endif 3220 TCP_PROBE3(debug__input, tp, th, m); 3221 tp->t_flags |= TF_ACKNOW; 3222 (void) tp->t_fb->tfb_tcp_output(tp); 3223 INP_WUNLOCK(tp->t_inpcb); 3224 m_freem(m); 3225 return; 3226 3227 dropwithreset: 3228 if (tp != NULL) { 3229 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3230 INP_WUNLOCK(tp->t_inpcb); 3231 } else 3232 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3233 return; 3234 3235 drop: 3236 /* 3237 * Drop space held by incoming segment and return. 3238 */ 3239 #ifdef TCPDEBUG 3240 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3241 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3242 &tcp_savetcp, 0); 3243 #endif 3244 TCP_PROBE3(debug__input, tp, th, m); 3245 if (tp != NULL) 3246 INP_WUNLOCK(tp->t_inpcb); 3247 m_freem(m); 3248 } 3249 3250 /* 3251 * Issue RST and make ACK acceptable to originator of segment. 3252 * The mbuf must still include the original packet header. 3253 * tp may be NULL. 3254 */ 3255 void 3256 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3257 int tlen, int rstreason) 3258 { 3259 #ifdef INET 3260 struct ip *ip; 3261 #endif 3262 #ifdef INET6 3263 struct ip6_hdr *ip6; 3264 #endif 3265 3266 if (tp != NULL) { 3267 INP_WLOCK_ASSERT(tp->t_inpcb); 3268 } 3269 3270 /* Don't bother if destination was broadcast/multicast. */ 3271 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3272 goto drop; 3273 #ifdef INET6 3274 if (mtod(m, struct ip *)->ip_v == 6) { 3275 ip6 = mtod(m, struct ip6_hdr *); 3276 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3277 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3278 goto drop; 3279 /* IPv6 anycast check is done at tcp6_input() */ 3280 } 3281 #endif 3282 #if defined(INET) && defined(INET6) 3283 else 3284 #endif 3285 #ifdef INET 3286 { 3287 ip = mtod(m, struct ip *); 3288 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3289 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3290 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3291 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3292 goto drop; 3293 } 3294 #endif 3295 3296 /* Perform bandwidth limiting. */ 3297 if (badport_bandlim(rstreason) < 0) 3298 goto drop; 3299 3300 /* tcp_respond consumes the mbuf chain. */ 3301 if (th->th_flags & TH_ACK) { 3302 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3303 th->th_ack, TH_RST); 3304 } else { 3305 if (th->th_flags & TH_SYN) 3306 tlen++; 3307 if (th->th_flags & TH_FIN) 3308 tlen++; 3309 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3310 (tcp_seq)0, TH_RST|TH_ACK); 3311 } 3312 return; 3313 drop: 3314 m_freem(m); 3315 } 3316 3317 /* 3318 * Parse TCP options and place in tcpopt. 3319 */ 3320 void 3321 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3322 { 3323 int opt, optlen; 3324 3325 to->to_flags = 0; 3326 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3327 opt = cp[0]; 3328 if (opt == TCPOPT_EOL) 3329 break; 3330 if (opt == TCPOPT_NOP) 3331 optlen = 1; 3332 else { 3333 if (cnt < 2) 3334 break; 3335 optlen = cp[1]; 3336 if (optlen < 2 || optlen > cnt) 3337 break; 3338 } 3339 switch (opt) { 3340 case TCPOPT_MAXSEG: 3341 if (optlen != TCPOLEN_MAXSEG) 3342 continue; 3343 if (!(flags & TO_SYN)) 3344 continue; 3345 to->to_flags |= TOF_MSS; 3346 bcopy((char *)cp + 2, 3347 (char *)&to->to_mss, sizeof(to->to_mss)); 3348 to->to_mss = ntohs(to->to_mss); 3349 break; 3350 case TCPOPT_WINDOW: 3351 if (optlen != TCPOLEN_WINDOW) 3352 continue; 3353 if (!(flags & TO_SYN)) 3354 continue; 3355 to->to_flags |= TOF_SCALE; 3356 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3357 break; 3358 case TCPOPT_TIMESTAMP: 3359 if (optlen != TCPOLEN_TIMESTAMP) 3360 continue; 3361 to->to_flags |= TOF_TS; 3362 bcopy((char *)cp + 2, 3363 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3364 to->to_tsval = ntohl(to->to_tsval); 3365 bcopy((char *)cp + 6, 3366 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3367 to->to_tsecr = ntohl(to->to_tsecr); 3368 break; 3369 case TCPOPT_SIGNATURE: 3370 /* 3371 * In order to reply to a host which has set the 3372 * TCP_SIGNATURE option in its initial SYN, we have 3373 * to record the fact that the option was observed 3374 * here for the syncache code to perform the correct 3375 * response. 3376 */ 3377 if (optlen != TCPOLEN_SIGNATURE) 3378 continue; 3379 to->to_flags |= TOF_SIGNATURE; 3380 to->to_signature = cp + 2; 3381 break; 3382 case TCPOPT_SACK_PERMITTED: 3383 if (optlen != TCPOLEN_SACK_PERMITTED) 3384 continue; 3385 if (!(flags & TO_SYN)) 3386 continue; 3387 if (!V_tcp_do_sack) 3388 continue; 3389 to->to_flags |= TOF_SACKPERM; 3390 break; 3391 case TCPOPT_SACK: 3392 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3393 continue; 3394 if (flags & TO_SYN) 3395 continue; 3396 to->to_flags |= TOF_SACK; 3397 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3398 to->to_sacks = cp + 2; 3399 TCPSTAT_INC(tcps_sack_rcv_blocks); 3400 break; 3401 case TCPOPT_FAST_OPEN: 3402 /* 3403 * Cookie length validation is performed by the 3404 * server side cookie checking code or the client 3405 * side cookie cache update code. 3406 */ 3407 if (!(flags & TO_SYN)) 3408 continue; 3409 if (!V_tcp_fastopen_client_enable && 3410 !V_tcp_fastopen_server_enable) 3411 continue; 3412 to->to_flags |= TOF_FASTOPEN; 3413 to->to_tfo_len = optlen - 2; 3414 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3415 break; 3416 default: 3417 continue; 3418 } 3419 } 3420 } 3421 3422 /* 3423 * Pull out of band byte out of a segment so 3424 * it doesn't appear in the user's data queue. 3425 * It is still reflected in the segment length for 3426 * sequencing purposes. 3427 */ 3428 void 3429 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3430 int off) 3431 { 3432 int cnt = off + th->th_urp - 1; 3433 3434 while (cnt >= 0) { 3435 if (m->m_len > cnt) { 3436 char *cp = mtod(m, caddr_t) + cnt; 3437 struct tcpcb *tp = sototcpcb(so); 3438 3439 INP_WLOCK_ASSERT(tp->t_inpcb); 3440 3441 tp->t_iobc = *cp; 3442 tp->t_oobflags |= TCPOOB_HAVEDATA; 3443 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3444 m->m_len--; 3445 if (m->m_flags & M_PKTHDR) 3446 m->m_pkthdr.len--; 3447 return; 3448 } 3449 cnt -= m->m_len; 3450 m = m->m_next; 3451 if (m == NULL) 3452 break; 3453 } 3454 panic("tcp_pulloutofband"); 3455 } 3456 3457 /* 3458 * Collect new round-trip time estimate 3459 * and update averages and current timeout. 3460 */ 3461 void 3462 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3463 { 3464 int delta; 3465 3466 INP_WLOCK_ASSERT(tp->t_inpcb); 3467 3468 TCPSTAT_INC(tcps_rttupdated); 3469 tp->t_rttupdated++; 3470 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3471 /* 3472 * srtt is stored as fixed point with 5 bits after the 3473 * binary point (i.e., scaled by 8). The following magic 3474 * is equivalent to the smoothing algorithm in rfc793 with 3475 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3476 * point). Adjust rtt to origin 0. 3477 */ 3478 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3479 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3480 3481 if ((tp->t_srtt += delta) <= 0) 3482 tp->t_srtt = 1; 3483 3484 /* 3485 * We accumulate a smoothed rtt variance (actually, a 3486 * smoothed mean difference), then set the retransmit 3487 * timer to smoothed rtt + 4 times the smoothed variance. 3488 * rttvar is stored as fixed point with 4 bits after the 3489 * binary point (scaled by 16). The following is 3490 * equivalent to rfc793 smoothing with an alpha of .75 3491 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3492 * rfc793's wired-in beta. 3493 */ 3494 if (delta < 0) 3495 delta = -delta; 3496 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3497 if ((tp->t_rttvar += delta) <= 0) 3498 tp->t_rttvar = 1; 3499 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3500 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3501 } else { 3502 /* 3503 * No rtt measurement yet - use the unsmoothed rtt. 3504 * Set the variance to half the rtt (so our first 3505 * retransmit happens at 3*rtt). 3506 */ 3507 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3508 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3509 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3510 } 3511 tp->t_rtttime = 0; 3512 tp->t_rxtshift = 0; 3513 3514 /* 3515 * the retransmit should happen at rtt + 4 * rttvar. 3516 * Because of the way we do the smoothing, srtt and rttvar 3517 * will each average +1/2 tick of bias. When we compute 3518 * the retransmit timer, we want 1/2 tick of rounding and 3519 * 1 extra tick because of +-1/2 tick uncertainty in the 3520 * firing of the timer. The bias will give us exactly the 3521 * 1.5 tick we need. But, because the bias is 3522 * statistical, we have to test that we don't drop below 3523 * the minimum feasible timer (which is 2 ticks). 3524 */ 3525 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3526 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3527 3528 /* 3529 * We received an ack for a packet that wasn't retransmitted; 3530 * it is probably safe to discard any error indications we've 3531 * received recently. This isn't quite right, but close enough 3532 * for now (a route might have failed after we sent a segment, 3533 * and the return path might not be symmetrical). 3534 */ 3535 tp->t_softerror = 0; 3536 } 3537 3538 /* 3539 * Determine a reasonable value for maxseg size. 3540 * If the route is known, check route for mtu. 3541 * If none, use an mss that can be handled on the outgoing interface 3542 * without forcing IP to fragment. If no route is found, route has no mtu, 3543 * or the destination isn't local, use a default, hopefully conservative 3544 * size (usually 512 or the default IP max size, but no more than the mtu 3545 * of the interface), as we can't discover anything about intervening 3546 * gateways or networks. We also initialize the congestion/slow start 3547 * window to be a single segment if the destination isn't local. 3548 * While looking at the routing entry, we also initialize other path-dependent 3549 * parameters from pre-set or cached values in the routing entry. 3550 * 3551 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3552 * IP options, e.g. IPSEC data, since length of this data may vary, and 3553 * thus it is calculated for every segment separately in tcp_output(). 3554 * 3555 * NOTE that this routine is only called when we process an incoming 3556 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3557 * settings are handled in tcp_mssopt(). 3558 */ 3559 void 3560 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3561 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3562 { 3563 int mss = 0; 3564 uint32_t maxmtu = 0; 3565 struct inpcb *inp = tp->t_inpcb; 3566 struct hc_metrics_lite metrics; 3567 #ifdef INET6 3568 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3569 size_t min_protoh = isipv6 ? 3570 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3571 sizeof (struct tcpiphdr); 3572 #else 3573 const size_t min_protoh = sizeof(struct tcpiphdr); 3574 #endif 3575 3576 INP_WLOCK_ASSERT(tp->t_inpcb); 3577 3578 if (mtuoffer != -1) { 3579 KASSERT(offer == -1, ("%s: conflict", __func__)); 3580 offer = mtuoffer - min_protoh; 3581 } 3582 3583 /* Initialize. */ 3584 #ifdef INET6 3585 if (isipv6) { 3586 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3587 tp->t_maxseg = V_tcp_v6mssdflt; 3588 } 3589 #endif 3590 #if defined(INET) && defined(INET6) 3591 else 3592 #endif 3593 #ifdef INET 3594 { 3595 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3596 tp->t_maxseg = V_tcp_mssdflt; 3597 } 3598 #endif 3599 3600 /* 3601 * No route to sender, stay with default mss and return. 3602 */ 3603 if (maxmtu == 0) { 3604 /* 3605 * In case we return early we need to initialize metrics 3606 * to a defined state as tcp_hc_get() would do for us 3607 * if there was no cache hit. 3608 */ 3609 if (metricptr != NULL) 3610 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3611 return; 3612 } 3613 3614 /* What have we got? */ 3615 switch (offer) { 3616 case 0: 3617 /* 3618 * Offer == 0 means that there was no MSS on the SYN 3619 * segment, in this case we use tcp_mssdflt as 3620 * already assigned to t_maxseg above. 3621 */ 3622 offer = tp->t_maxseg; 3623 break; 3624 3625 case -1: 3626 /* 3627 * Offer == -1 means that we didn't receive SYN yet. 3628 */ 3629 /* FALLTHROUGH */ 3630 3631 default: 3632 /* 3633 * Prevent DoS attack with too small MSS. Round up 3634 * to at least minmss. 3635 */ 3636 offer = max(offer, V_tcp_minmss); 3637 } 3638 3639 /* 3640 * rmx information is now retrieved from tcp_hostcache. 3641 */ 3642 tcp_hc_get(&inp->inp_inc, &metrics); 3643 if (metricptr != NULL) 3644 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3645 3646 /* 3647 * If there's a discovered mtu in tcp hostcache, use it. 3648 * Else, use the link mtu. 3649 */ 3650 if (metrics.rmx_mtu) 3651 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3652 else { 3653 #ifdef INET6 3654 if (isipv6) { 3655 mss = maxmtu - min_protoh; 3656 if (!V_path_mtu_discovery && 3657 !in6_localaddr(&inp->in6p_faddr)) 3658 mss = min(mss, V_tcp_v6mssdflt); 3659 } 3660 #endif 3661 #if defined(INET) && defined(INET6) 3662 else 3663 #endif 3664 #ifdef INET 3665 { 3666 mss = maxmtu - min_protoh; 3667 if (!V_path_mtu_discovery && 3668 !in_localaddr(inp->inp_faddr)) 3669 mss = min(mss, V_tcp_mssdflt); 3670 } 3671 #endif 3672 /* 3673 * XXX - The above conditional (mss = maxmtu - min_protoh) 3674 * probably violates the TCP spec. 3675 * The problem is that, since we don't know the 3676 * other end's MSS, we are supposed to use a conservative 3677 * default. But, if we do that, then MTU discovery will 3678 * never actually take place, because the conservative 3679 * default is much less than the MTUs typically seen 3680 * on the Internet today. For the moment, we'll sweep 3681 * this under the carpet. 3682 * 3683 * The conservative default might not actually be a problem 3684 * if the only case this occurs is when sending an initial 3685 * SYN with options and data to a host we've never talked 3686 * to before. Then, they will reply with an MSS value which 3687 * will get recorded and the new parameters should get 3688 * recomputed. For Further Study. 3689 */ 3690 } 3691 mss = min(mss, offer); 3692 3693 /* 3694 * Sanity check: make sure that maxseg will be large 3695 * enough to allow some data on segments even if the 3696 * all the option space is used (40bytes). Otherwise 3697 * funny things may happen in tcp_output. 3698 * 3699 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3700 */ 3701 mss = max(mss, 64); 3702 3703 tp->t_maxseg = mss; 3704 } 3705 3706 void 3707 tcp_mss(struct tcpcb *tp, int offer) 3708 { 3709 int mss; 3710 uint32_t bufsize; 3711 struct inpcb *inp; 3712 struct socket *so; 3713 struct hc_metrics_lite metrics; 3714 struct tcp_ifcap cap; 3715 3716 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3717 3718 bzero(&cap, sizeof(cap)); 3719 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3720 3721 mss = tp->t_maxseg; 3722 inp = tp->t_inpcb; 3723 3724 /* 3725 * If there's a pipesize, change the socket buffer to that size, 3726 * don't change if sb_hiwat is different than default (then it 3727 * has been changed on purpose with setsockopt). 3728 * Make the socket buffers an integral number of mss units; 3729 * if the mss is larger than the socket buffer, decrease the mss. 3730 */ 3731 so = inp->inp_socket; 3732 SOCKBUF_LOCK(&so->so_snd); 3733 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3734 bufsize = metrics.rmx_sendpipe; 3735 else 3736 bufsize = so->so_snd.sb_hiwat; 3737 if (bufsize < mss) 3738 mss = bufsize; 3739 else { 3740 bufsize = roundup(bufsize, mss); 3741 if (bufsize > sb_max) 3742 bufsize = sb_max; 3743 if (bufsize > so->so_snd.sb_hiwat) 3744 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3745 } 3746 SOCKBUF_UNLOCK(&so->so_snd); 3747 /* 3748 * Sanity check: make sure that maxseg will be large 3749 * enough to allow some data on segments even if the 3750 * all the option space is used (40bytes). Otherwise 3751 * funny things may happen in tcp_output. 3752 * 3753 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3754 */ 3755 tp->t_maxseg = max(mss, 64); 3756 3757 SOCKBUF_LOCK(&so->so_rcv); 3758 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3759 bufsize = metrics.rmx_recvpipe; 3760 else 3761 bufsize = so->so_rcv.sb_hiwat; 3762 if (bufsize > mss) { 3763 bufsize = roundup(bufsize, mss); 3764 if (bufsize > sb_max) 3765 bufsize = sb_max; 3766 if (bufsize > so->so_rcv.sb_hiwat) 3767 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3768 } 3769 SOCKBUF_UNLOCK(&so->so_rcv); 3770 3771 /* Check the interface for TSO capabilities. */ 3772 if (cap.ifcap & CSUM_TSO) { 3773 tp->t_flags |= TF_TSO; 3774 tp->t_tsomax = cap.tsomax; 3775 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3776 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3777 } 3778 } 3779 3780 /* 3781 * Determine the MSS option to send on an outgoing SYN. 3782 */ 3783 int 3784 tcp_mssopt(struct in_conninfo *inc) 3785 { 3786 int mss = 0; 3787 uint32_t thcmtu = 0; 3788 uint32_t maxmtu = 0; 3789 size_t min_protoh; 3790 3791 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3792 3793 #ifdef INET6 3794 if (inc->inc_flags & INC_ISIPV6) { 3795 mss = V_tcp_v6mssdflt; 3796 maxmtu = tcp_maxmtu6(inc, NULL); 3797 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3798 } 3799 #endif 3800 #if defined(INET) && defined(INET6) 3801 else 3802 #endif 3803 #ifdef INET 3804 { 3805 mss = V_tcp_mssdflt; 3806 maxmtu = tcp_maxmtu(inc, NULL); 3807 min_protoh = sizeof(struct tcpiphdr); 3808 } 3809 #endif 3810 #if defined(INET6) || defined(INET) 3811 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3812 #endif 3813 3814 if (maxmtu && thcmtu) 3815 mss = min(maxmtu, thcmtu) - min_protoh; 3816 else if (maxmtu || thcmtu) 3817 mss = max(maxmtu, thcmtu) - min_protoh; 3818 3819 return (mss); 3820 } 3821 3822 3823 /* 3824 * On a partial ack arrives, force the retransmission of the 3825 * next unacknowledged segment. Do not clear tp->t_dupacks. 3826 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3827 * be started again. 3828 */ 3829 void 3830 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3831 { 3832 tcp_seq onxt = tp->snd_nxt; 3833 uint32_t ocwnd = tp->snd_cwnd; 3834 u_int maxseg = tcp_maxseg(tp); 3835 3836 INP_WLOCK_ASSERT(tp->t_inpcb); 3837 3838 tcp_timer_activate(tp, TT_REXMT, 0); 3839 tp->t_rtttime = 0; 3840 tp->snd_nxt = th->th_ack; 3841 /* 3842 * Set snd_cwnd to one segment beyond acknowledged offset. 3843 * (tp->snd_una has not yet been updated when this function is called.) 3844 */ 3845 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3846 tp->t_flags |= TF_ACKNOW; 3847 (void) tp->t_fb->tfb_tcp_output(tp); 3848 tp->snd_cwnd = ocwnd; 3849 if (SEQ_GT(onxt, tp->snd_nxt)) 3850 tp->snd_nxt = onxt; 3851 /* 3852 * Partial window deflation. Relies on fact that tp->snd_una 3853 * not updated yet. 3854 */ 3855 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3856 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3857 else 3858 tp->snd_cwnd = 0; 3859 tp->snd_cwnd += maxseg; 3860 } 3861 3862 int 3863 tcp_compute_pipe(struct tcpcb *tp) 3864 { 3865 return (tp->snd_max - tp->snd_una + 3866 tp->sackhint.sack_bytes_rexmit - 3867 tp->sackhint.sacked_bytes); 3868 } 3869