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 <lstewart@freebsd.org> 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 #endif 580 int tlen = 0, off; 581 int drop_hdrlen; 582 int thflags; 583 int rstreason = 0; /* For badport_bandlim accounting purposes */ 584 uint8_t iptos; 585 struct m_tag *fwd_tag = NULL; 586 #ifdef INET6 587 struct ip6_hdr *ip6 = NULL; 588 int isipv6; 589 #else 590 const void *ip6 = NULL; 591 #endif /* INET6 */ 592 struct tcpopt to; /* options in this segment */ 593 char *s = NULL; /* address and port logging */ 594 int ti_locked; 595 #ifdef TCPDEBUG 596 /* 597 * The size of tcp_saveipgen must be the size of the max ip header, 598 * now IPv6. 599 */ 600 u_char tcp_saveipgen[IP6_HDR_LEN]; 601 struct tcphdr tcp_savetcp; 602 short ostate = 0; 603 #endif 604 605 #ifdef INET6 606 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 607 #endif 608 609 off0 = *offp; 610 m = *mp; 611 *mp = NULL; 612 to.to_flags = 0; 613 TCPSTAT_INC(tcps_rcvtotal); 614 615 #ifdef INET6 616 if (isipv6) { 617 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 618 619 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 620 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 621 if (m == NULL) { 622 TCPSTAT_INC(tcps_rcvshort); 623 return (IPPROTO_DONE); 624 } 625 } 626 627 ip6 = mtod(m, struct ip6_hdr *); 628 th = (struct tcphdr *)((caddr_t)ip6 + off0); 629 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 630 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 631 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 632 th->th_sum = m->m_pkthdr.csum_data; 633 else 634 th->th_sum = in6_cksum_pseudo(ip6, tlen, 635 IPPROTO_TCP, m->m_pkthdr.csum_data); 636 th->th_sum ^= 0xffff; 637 } else 638 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 639 if (th->th_sum) { 640 TCPSTAT_INC(tcps_rcvbadsum); 641 goto drop; 642 } 643 644 /* 645 * Be proactive about unspecified IPv6 address in source. 646 * As we use all-zero to indicate unbounded/unconnected pcb, 647 * unspecified IPv6 address can be used to confuse us. 648 * 649 * Note that packets with unspecified IPv6 destination is 650 * already dropped in ip6_input. 651 */ 652 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 653 /* XXX stat */ 654 goto drop; 655 } 656 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 657 } 658 #endif 659 #if defined(INET) && defined(INET6) 660 else 661 #endif 662 #ifdef INET 663 { 664 /* 665 * Get IP and TCP header together in first mbuf. 666 * Note: IP leaves IP header in first mbuf. 667 */ 668 if (off0 > sizeof (struct ip)) { 669 ip_stripoptions(m); 670 off0 = sizeof(struct ip); 671 } 672 if (m->m_len < sizeof (struct tcpiphdr)) { 673 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 674 == NULL) { 675 TCPSTAT_INC(tcps_rcvshort); 676 return (IPPROTO_DONE); 677 } 678 } 679 ip = mtod(m, struct ip *); 680 th = (struct tcphdr *)((caddr_t)ip + off0); 681 tlen = ntohs(ip->ip_len) - off0; 682 683 iptos = ip->ip_tos; 684 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 685 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 686 th->th_sum = m->m_pkthdr.csum_data; 687 else 688 th->th_sum = in_pseudo(ip->ip_src.s_addr, 689 ip->ip_dst.s_addr, 690 htonl(m->m_pkthdr.csum_data + tlen + 691 IPPROTO_TCP)); 692 th->th_sum ^= 0xffff; 693 } else { 694 struct ipovly *ipov = (struct ipovly *)ip; 695 696 /* 697 * Checksum extended TCP header and data. 698 */ 699 len = off0 + tlen; 700 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 701 ipov->ih_len = htons(tlen); 702 th->th_sum = in_cksum(m, len); 703 /* Reset length for SDT probes. */ 704 ip->ip_len = htons(len); 705 /* Reset TOS bits */ 706 ip->ip_tos = iptos; 707 /* Re-initialization for later version check */ 708 ip->ip_v = IPVERSION; 709 ip->ip_hl = off0 >> 2; 710 } 711 712 if (th->th_sum) { 713 TCPSTAT_INC(tcps_rcvbadsum); 714 goto drop; 715 } 716 } 717 #endif /* INET */ 718 719 /* 720 * Check that TCP offset makes sense, 721 * pull out TCP options and adjust length. XXX 722 */ 723 off = th->th_off << 2; 724 if (off < sizeof (struct tcphdr) || off > tlen) { 725 TCPSTAT_INC(tcps_rcvbadoff); 726 goto drop; 727 } 728 tlen -= off; /* tlen is used instead of ti->ti_len */ 729 if (off > sizeof (struct tcphdr)) { 730 #ifdef INET6 731 if (isipv6) { 732 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 733 ip6 = mtod(m, struct ip6_hdr *); 734 th = (struct tcphdr *)((caddr_t)ip6 + off0); 735 } 736 #endif 737 #if defined(INET) && defined(INET6) 738 else 739 #endif 740 #ifdef INET 741 { 742 if (m->m_len < sizeof(struct ip) + off) { 743 if ((m = m_pullup(m, sizeof (struct ip) + off)) 744 == NULL) { 745 TCPSTAT_INC(tcps_rcvshort); 746 return (IPPROTO_DONE); 747 } 748 ip = mtod(m, struct ip *); 749 th = (struct tcphdr *)((caddr_t)ip + off0); 750 } 751 } 752 #endif 753 optlen = off - sizeof (struct tcphdr); 754 optp = (u_char *)(th + 1); 755 } 756 thflags = th->th_flags; 757 758 /* 759 * Convert TCP protocol specific fields to host format. 760 */ 761 tcp_fields_to_host(th); 762 763 /* 764 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 765 */ 766 drop_hdrlen = off0 + off; 767 768 /* 769 * Locate pcb for segment; if we're likely to add or remove a 770 * connection then first acquire pcbinfo lock. There are three cases 771 * where we might discover later we need a write lock despite the 772 * flags: ACKs moving a connection out of the syncache, ACKs for a 773 * connection in TIMEWAIT and SYNs not targeting a listening socket. 774 */ 775 if ((thflags & (TH_FIN | TH_RST)) != 0) { 776 INP_INFO_RLOCK(&V_tcbinfo); 777 ti_locked = TI_RLOCKED; 778 } else 779 ti_locked = TI_UNLOCKED; 780 781 /* 782 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 783 */ 784 if ( 785 #ifdef INET6 786 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 787 #ifdef INET 788 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 789 #endif 790 #endif 791 #if defined(INET) && !defined(INET6) 792 (m->m_flags & M_IP_NEXTHOP) 793 #endif 794 ) 795 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 796 797 findpcb: 798 #ifdef INVARIANTS 799 if (ti_locked == TI_RLOCKED) { 800 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 801 } else { 802 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 803 } 804 #endif 805 #ifdef INET6 806 if (isipv6 && fwd_tag != NULL) { 807 struct sockaddr_in6 *next_hop6; 808 809 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 810 /* 811 * Transparently forwarded. Pretend to be the destination. 812 * Already got one like this? 813 */ 814 inp = in6_pcblookup_mbuf(&V_tcbinfo, 815 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 816 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 817 if (!inp) { 818 /* 819 * It's new. Try to find the ambushing socket. 820 * Because we've rewritten the destination address, 821 * any hardware-generated hash is ignored. 822 */ 823 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 824 th->th_sport, &next_hop6->sin6_addr, 825 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 826 th->th_dport, INPLOOKUP_WILDCARD | 827 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 828 } 829 } else if (isipv6) { 830 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 831 th->th_sport, &ip6->ip6_dst, th->th_dport, 832 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 833 m->m_pkthdr.rcvif, m); 834 } 835 #endif /* INET6 */ 836 #if defined(INET6) && defined(INET) 837 else 838 #endif 839 #ifdef INET 840 if (fwd_tag != NULL) { 841 struct sockaddr_in *next_hop; 842 843 next_hop = (struct sockaddr_in *)(fwd_tag+1); 844 /* 845 * Transparently forwarded. Pretend to be the destination. 846 * already got one like this? 847 */ 848 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 849 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 850 m->m_pkthdr.rcvif, m); 851 if (!inp) { 852 /* 853 * It's new. Try to find the ambushing socket. 854 * Because we've rewritten the destination address, 855 * any hardware-generated hash is ignored. 856 */ 857 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 858 th->th_sport, next_hop->sin_addr, 859 next_hop->sin_port ? ntohs(next_hop->sin_port) : 860 th->th_dport, INPLOOKUP_WILDCARD | 861 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 862 } 863 } else 864 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 865 th->th_sport, ip->ip_dst, th->th_dport, 866 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 867 m->m_pkthdr.rcvif, m); 868 #endif /* INET */ 869 870 /* 871 * If the INPCB does not exist then all data in the incoming 872 * segment is discarded and an appropriate RST is sent back. 873 * XXX MRT Send RST using which routing table? 874 */ 875 if (inp == NULL) { 876 /* 877 * Log communication attempts to ports that are not 878 * in use. 879 */ 880 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 881 tcp_log_in_vain == 2) { 882 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 883 log(LOG_INFO, "%s; %s: Connection attempt " 884 "to closed port\n", s, __func__); 885 } 886 /* 887 * When blackholing do not respond with a RST but 888 * completely ignore the segment and drop it. 889 */ 890 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 891 V_blackhole == 2) 892 goto dropunlock; 893 894 rstreason = BANDLIM_RST_CLOSEDPORT; 895 goto dropwithreset; 896 } 897 INP_WLOCK_ASSERT(inp); 898 /* 899 * While waiting for inp lock during the lookup, another thread 900 * can have dropped the inpcb, in which case we need to loop back 901 * and try to find a new inpcb to deliver to. 902 */ 903 if (inp->inp_flags & INP_DROPPED) { 904 INP_WUNLOCK(inp); 905 inp = NULL; 906 goto findpcb; 907 } 908 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 909 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 910 ((inp->inp_socket == NULL) || 911 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 912 inp->inp_flowid = m->m_pkthdr.flowid; 913 inp->inp_flowtype = M_HASHTYPE_GET(m); 914 } 915 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 916 #ifdef INET6 917 if (isipv6 && IPSEC_ENABLED(ipv6) && 918 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 919 goto dropunlock; 920 } 921 #ifdef INET 922 else 923 #endif 924 #endif /* INET6 */ 925 #ifdef INET 926 if (IPSEC_ENABLED(ipv4) && 927 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 928 goto dropunlock; 929 } 930 #endif /* INET */ 931 #endif /* IPSEC */ 932 933 /* 934 * Check the minimum TTL for socket. 935 */ 936 if (inp->inp_ip_minttl != 0) { 937 #ifdef INET6 938 if (isipv6) { 939 if (inp->inp_ip_minttl > ip6->ip6_hlim) 940 goto dropunlock; 941 } else 942 #endif 943 if (inp->inp_ip_minttl > ip->ip_ttl) 944 goto dropunlock; 945 } 946 947 /* 948 * A previous connection in TIMEWAIT state is supposed to catch stray 949 * or duplicate segments arriving late. If this segment was a 950 * legitimate new connection attempt, the old INPCB gets removed and 951 * we can try again to find a listening socket. 952 * 953 * At this point, due to earlier optimism, we may hold only an inpcb 954 * lock, and not the inpcbinfo write lock. If so, we need to try to 955 * acquire it, or if that fails, acquire a reference on the inpcb, 956 * drop all locks, acquire a global write lock, and then re-acquire 957 * the inpcb lock. We may at that point discover that another thread 958 * has tried to free the inpcb, in which case we need to loop back 959 * and try to find a new inpcb to deliver to. 960 * 961 * XXXRW: It may be time to rethink timewait locking. 962 */ 963 if (inp->inp_flags & INP_TIMEWAIT) { 964 if (ti_locked == TI_UNLOCKED) { 965 INP_INFO_RLOCK(); 966 ti_locked = TI_RLOCKED; 967 } 968 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 969 970 if (thflags & TH_SYN) 971 tcp_dooptions(&to, optp, optlen, TO_SYN); 972 /* 973 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 974 */ 975 if (tcp_twcheck(inp, &to, th, m, tlen)) 976 goto findpcb; 977 INP_INFO_RUNLOCK(&V_tcbinfo); 978 return (IPPROTO_DONE); 979 } 980 /* 981 * The TCPCB may no longer exist if the connection is winding 982 * down or it is in the CLOSED state. Either way we drop the 983 * segment and send an appropriate response. 984 */ 985 tp = intotcpcb(inp); 986 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 987 rstreason = BANDLIM_RST_CLOSEDPORT; 988 goto dropwithreset; 989 } 990 991 #ifdef TCP_OFFLOAD 992 if (tp->t_flags & TF_TOE) { 993 tcp_offload_input(tp, m); 994 m = NULL; /* consumed by the TOE driver */ 995 goto dropunlock; 996 } 997 #endif 998 999 /* 1000 * We've identified a valid inpcb, but it could be that we need an 1001 * inpcbinfo write lock but don't hold it. In this case, attempt to 1002 * acquire using the same strategy as the TIMEWAIT case above. If we 1003 * relock, we have to jump back to 'relocked' as the connection might 1004 * now be in TIMEWAIT. 1005 */ 1006 #ifdef INVARIANTS 1007 if ((thflags & (TH_FIN | TH_RST)) != 0) 1008 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1009 #endif 1010 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1011 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1012 !IS_FASTOPEN(tp->t_flags)))) { 1013 if (ti_locked == TI_UNLOCKED) { 1014 INP_INFO_RLOCK(); 1015 ti_locked = TI_RLOCKED; 1016 } 1017 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1018 } 1019 1020 #ifdef MAC 1021 INP_WLOCK_ASSERT(inp); 1022 if (mac_inpcb_check_deliver(inp, m)) 1023 goto dropunlock; 1024 #endif 1025 so = inp->inp_socket; 1026 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1027 #ifdef TCPDEBUG 1028 if (so->so_options & SO_DEBUG) { 1029 ostate = tp->t_state; 1030 #ifdef INET6 1031 if (isipv6) { 1032 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1033 } else 1034 #endif 1035 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1036 tcp_savetcp = *th; 1037 } 1038 #endif /* TCPDEBUG */ 1039 /* 1040 * When the socket is accepting connections (the INPCB is in LISTEN 1041 * state) we look into the SYN cache if this is a new connection 1042 * attempt or the completion of a previous one. 1043 */ 1044 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1045 ("%s: so accepting but tp %p not listening", __func__, tp)); 1046 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1047 struct in_conninfo inc; 1048 1049 bzero(&inc, sizeof(inc)); 1050 #ifdef INET6 1051 if (isipv6) { 1052 inc.inc_flags |= INC_ISIPV6; 1053 inc.inc6_faddr = ip6->ip6_src; 1054 inc.inc6_laddr = ip6->ip6_dst; 1055 } else 1056 #endif 1057 { 1058 inc.inc_faddr = ip->ip_src; 1059 inc.inc_laddr = ip->ip_dst; 1060 } 1061 inc.inc_fport = th->th_sport; 1062 inc.inc_lport = th->th_dport; 1063 inc.inc_fibnum = so->so_fibnum; 1064 1065 /* 1066 * Check for an existing connection attempt in syncache if 1067 * the flag is only ACK. A successful lookup creates a new 1068 * socket appended to the listen queue in SYN_RECEIVED state. 1069 */ 1070 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1071 1072 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1073 /* 1074 * Parse the TCP options here because 1075 * syncookies need access to the reflected 1076 * timestamp. 1077 */ 1078 tcp_dooptions(&to, optp, optlen, 0); 1079 /* 1080 * NB: syncache_expand() doesn't unlock 1081 * inp and tcpinfo locks. 1082 */ 1083 rstreason = syncache_expand(&inc, &to, th, &so, m); 1084 if (rstreason < 0) { 1085 /* 1086 * A failing TCP MD5 signature comparison 1087 * must result in the segment being dropped 1088 * and must not produce any response back 1089 * to the sender. 1090 */ 1091 goto dropunlock; 1092 } else if (rstreason == 0) { 1093 /* 1094 * No syncache entry or ACK was not 1095 * for our SYN/ACK. Send a RST. 1096 * NB: syncache did its own logging 1097 * of the failure cause. 1098 */ 1099 rstreason = BANDLIM_RST_OPENPORT; 1100 goto dropwithreset; 1101 } 1102 tfo_socket_result: 1103 if (so == NULL) { 1104 /* 1105 * We completed the 3-way handshake 1106 * but could not allocate a socket 1107 * either due to memory shortage, 1108 * listen queue length limits or 1109 * global socket limits. Send RST 1110 * or wait and have the remote end 1111 * retransmit the ACK for another 1112 * try. 1113 */ 1114 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1115 log(LOG_DEBUG, "%s; %s: Listen socket: " 1116 "Socket allocation failed due to " 1117 "limits or memory shortage, %s\n", 1118 s, __func__, 1119 V_tcp_sc_rst_sock_fail ? 1120 "sending RST" : "try again"); 1121 if (V_tcp_sc_rst_sock_fail) { 1122 rstreason = BANDLIM_UNLIMITED; 1123 goto dropwithreset; 1124 } else 1125 goto dropunlock; 1126 } 1127 /* 1128 * Socket is created in state SYN_RECEIVED. 1129 * Unlock the listen socket, lock the newly 1130 * created socket and update the tp variable. 1131 */ 1132 INP_WUNLOCK(inp); /* listen socket */ 1133 inp = sotoinpcb(so); 1134 /* 1135 * New connection inpcb is already locked by 1136 * syncache_expand(). 1137 */ 1138 INP_WLOCK_ASSERT(inp); 1139 tp = intotcpcb(inp); 1140 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1141 ("%s: ", __func__)); 1142 /* 1143 * Process the segment and the data it 1144 * contains. tcp_do_segment() consumes 1145 * the mbuf chain and unlocks the inpcb. 1146 */ 1147 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1148 iptos, ti_locked); 1149 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1150 return (IPPROTO_DONE); 1151 } 1152 /* 1153 * Segment flag validation for new connection attempts: 1154 * 1155 * Our (SYN|ACK) response was rejected. 1156 * Check with syncache and remove entry to prevent 1157 * retransmits. 1158 * 1159 * NB: syncache_chkrst does its own logging of failure 1160 * causes. 1161 */ 1162 if (thflags & TH_RST) { 1163 syncache_chkrst(&inc, th); 1164 goto dropunlock; 1165 } 1166 /* 1167 * We can't do anything without SYN. 1168 */ 1169 if ((thflags & TH_SYN) == 0) { 1170 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1171 log(LOG_DEBUG, "%s; %s: Listen socket: " 1172 "SYN is missing, segment ignored\n", 1173 s, __func__); 1174 TCPSTAT_INC(tcps_badsyn); 1175 goto dropunlock; 1176 } 1177 /* 1178 * (SYN|ACK) is bogus on a listen socket. 1179 */ 1180 if (thflags & TH_ACK) { 1181 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1182 log(LOG_DEBUG, "%s; %s: Listen socket: " 1183 "SYN|ACK invalid, segment rejected\n", 1184 s, __func__); 1185 syncache_badack(&inc); /* XXX: Not needed! */ 1186 TCPSTAT_INC(tcps_badsyn); 1187 rstreason = BANDLIM_RST_OPENPORT; 1188 goto dropwithreset; 1189 } 1190 /* 1191 * If the drop_synfin option is enabled, drop all 1192 * segments with both the SYN and FIN bits set. 1193 * This prevents e.g. nmap from identifying the 1194 * TCP/IP stack. 1195 * XXX: Poor reasoning. nmap has other methods 1196 * and is constantly refining its stack detection 1197 * strategies. 1198 * XXX: This is a violation of the TCP specification 1199 * and was used by RFC1644. 1200 */ 1201 if ((thflags & TH_FIN) && V_drop_synfin) { 1202 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1203 log(LOG_DEBUG, "%s; %s: Listen socket: " 1204 "SYN|FIN segment ignored (based on " 1205 "sysctl setting)\n", s, __func__); 1206 TCPSTAT_INC(tcps_badsyn); 1207 goto dropunlock; 1208 } 1209 /* 1210 * Segment's flags are (SYN) or (SYN|FIN). 1211 * 1212 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1213 * as they do not affect the state of the TCP FSM. 1214 * The data pointed to by TH_URG and th_urp is ignored. 1215 */ 1216 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1217 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1218 KASSERT(thflags & (TH_SYN), 1219 ("%s: Listen socket: TH_SYN not set", __func__)); 1220 #ifdef INET6 1221 /* 1222 * If deprecated address is forbidden, 1223 * we do not accept SYN to deprecated interface 1224 * address to prevent any new inbound connection from 1225 * getting established. 1226 * When we do not accept SYN, we send a TCP RST, 1227 * with deprecated source address (instead of dropping 1228 * it). We compromise it as it is much better for peer 1229 * to send a RST, and RST will be the final packet 1230 * for the exchange. 1231 * 1232 * If we do not forbid deprecated addresses, we accept 1233 * the SYN packet. RFC2462 does not suggest dropping 1234 * SYN in this case. 1235 * If we decipher RFC2462 5.5.4, it says like this: 1236 * 1. use of deprecated addr with existing 1237 * communication is okay - "SHOULD continue to be 1238 * used" 1239 * 2. use of it with new communication: 1240 * (2a) "SHOULD NOT be used if alternate address 1241 * with sufficient scope is available" 1242 * (2b) nothing mentioned otherwise. 1243 * Here we fall into (2b) case as we have no choice in 1244 * our source address selection - we must obey the peer. 1245 * 1246 * The wording in RFC2462 is confusing, and there are 1247 * multiple description text for deprecated address 1248 * handling - worse, they are not exactly the same. 1249 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1250 */ 1251 if (isipv6 && !V_ip6_use_deprecated) { 1252 struct in6_ifaddr *ia6; 1253 1254 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1255 if (ia6 != NULL && 1256 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1257 ifa_free(&ia6->ia_ifa); 1258 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1259 log(LOG_DEBUG, "%s; %s: Listen socket: " 1260 "Connection attempt to deprecated " 1261 "IPv6 address rejected\n", 1262 s, __func__); 1263 rstreason = BANDLIM_RST_OPENPORT; 1264 goto dropwithreset; 1265 } 1266 if (ia6) 1267 ifa_free(&ia6->ia_ifa); 1268 } 1269 #endif /* INET6 */ 1270 /* 1271 * Basic sanity checks on incoming SYN requests: 1272 * Don't respond if the destination is a link layer 1273 * broadcast according to RFC1122 4.2.3.10, p. 104. 1274 * If it is from this socket it must be forged. 1275 * Don't respond if the source or destination is a 1276 * global or subnet broad- or multicast address. 1277 * Note that it is quite possible to receive unicast 1278 * link-layer packets with a broadcast IP address. Use 1279 * in_broadcast() to find them. 1280 */ 1281 if (m->m_flags & (M_BCAST|M_MCAST)) { 1282 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1283 log(LOG_DEBUG, "%s; %s: Listen socket: " 1284 "Connection attempt from broad- or multicast " 1285 "link layer address ignored\n", s, __func__); 1286 goto dropunlock; 1287 } 1288 #ifdef INET6 1289 if (isipv6) { 1290 if (th->th_dport == th->th_sport && 1291 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1292 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1293 log(LOG_DEBUG, "%s; %s: Listen socket: " 1294 "Connection attempt to/from self " 1295 "ignored\n", s, __func__); 1296 goto dropunlock; 1297 } 1298 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1299 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1300 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1301 log(LOG_DEBUG, "%s; %s: Listen socket: " 1302 "Connection attempt from/to multicast " 1303 "address ignored\n", s, __func__); 1304 goto dropunlock; 1305 } 1306 } 1307 #endif 1308 #if defined(INET) && defined(INET6) 1309 else 1310 #endif 1311 #ifdef INET 1312 { 1313 if (th->th_dport == th->th_sport && 1314 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from/to self " 1318 "ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1322 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1323 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1324 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1326 log(LOG_DEBUG, "%s; %s: Listen socket: " 1327 "Connection attempt from/to broad- " 1328 "or multicast address ignored\n", 1329 s, __func__); 1330 goto dropunlock; 1331 } 1332 } 1333 #endif 1334 /* 1335 * SYN appears to be valid. Create compressed TCP state 1336 * for syncache. 1337 */ 1338 #ifdef TCPDEBUG 1339 if (so->so_options & SO_DEBUG) 1340 tcp_trace(TA_INPUT, ostate, tp, 1341 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1342 #endif 1343 TCP_PROBE3(debug__input, tp, th, m); 1344 tcp_dooptions(&to, optp, optlen, TO_SYN); 1345 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1346 goto tfo_socket_result; 1347 1348 /* 1349 * Entry added to syncache and mbuf consumed. 1350 * Only the listen socket is unlocked by syncache_add(). 1351 */ 1352 if (ti_locked == TI_RLOCKED) { 1353 INP_INFO_RUNLOCK(&V_tcbinfo); 1354 ti_locked = TI_UNLOCKED; 1355 } 1356 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1357 return (IPPROTO_DONE); 1358 } else if (tp->t_state == TCPS_LISTEN) { 1359 /* 1360 * When a listen socket is torn down the SO_ACCEPTCONN 1361 * flag is removed first while connections are drained 1362 * from the accept queue in a unlock/lock cycle of the 1363 * ACCEPT_LOCK, opening a race condition allowing a SYN 1364 * attempt go through unhandled. 1365 */ 1366 goto dropunlock; 1367 } 1368 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1369 if (tp->t_flags & TF_SIGNATURE) { 1370 tcp_dooptions(&to, optp, optlen, thflags); 1371 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1372 TCPSTAT_INC(tcps_sig_err_nosigopt); 1373 goto dropunlock; 1374 } 1375 if (!TCPMD5_ENABLED() || 1376 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1377 goto dropunlock; 1378 } 1379 #endif 1380 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1381 1382 /* 1383 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1384 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1385 * the inpcb, and unlocks pcbinfo. 1386 */ 1387 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1388 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1389 return (IPPROTO_DONE); 1390 1391 dropwithreset: 1392 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1393 1394 if (ti_locked == TI_RLOCKED) { 1395 INP_INFO_RUNLOCK(&V_tcbinfo); 1396 ti_locked = TI_UNLOCKED; 1397 } 1398 #ifdef INVARIANTS 1399 else { 1400 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1401 "ti_locked: %d", __func__, ti_locked)); 1402 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1403 } 1404 #endif 1405 1406 if (inp != NULL) { 1407 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1408 INP_WUNLOCK(inp); 1409 } else 1410 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1411 m = NULL; /* mbuf chain got consumed. */ 1412 goto drop; 1413 1414 dropunlock: 1415 if (m != NULL) 1416 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1417 1418 if (ti_locked == TI_RLOCKED) { 1419 INP_INFO_RUNLOCK(&V_tcbinfo); 1420 ti_locked = TI_UNLOCKED; 1421 } 1422 #ifdef INVARIANTS 1423 else { 1424 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1425 "ti_locked: %d", __func__, ti_locked)); 1426 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1427 } 1428 #endif 1429 1430 if (inp != NULL) 1431 INP_WUNLOCK(inp); 1432 1433 drop: 1434 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1435 if (s != NULL) 1436 free(s, M_TCPLOG); 1437 if (m != NULL) 1438 m_freem(m); 1439 return (IPPROTO_DONE); 1440 } 1441 1442 /* 1443 * Automatic sizing of receive socket buffer. Often the send 1444 * buffer size is not optimally adjusted to the actual network 1445 * conditions at hand (delay bandwidth product). Setting the 1446 * buffer size too small limits throughput on links with high 1447 * bandwidth and high delay (eg. trans-continental/oceanic links). 1448 * 1449 * On the receive side the socket buffer memory is only rarely 1450 * used to any significant extent. This allows us to be much 1451 * more aggressive in scaling the receive socket buffer. For 1452 * the case that the buffer space is actually used to a large 1453 * extent and we run out of kernel memory we can simply drop 1454 * the new segments; TCP on the sender will just retransmit it 1455 * later. Setting the buffer size too big may only consume too 1456 * much kernel memory if the application doesn't read() from 1457 * the socket or packet loss or reordering makes use of the 1458 * reassembly queue. 1459 * 1460 * The criteria to step up the receive buffer one notch are: 1461 * 1. Application has not set receive buffer size with 1462 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1463 * 2. the number of bytes received during the time it takes 1464 * one timestamp to be reflected back to us (the RTT); 1465 * 3. received bytes per RTT is within seven eighth of the 1466 * current socket buffer size; 1467 * 4. receive buffer size has not hit maximal automatic size; 1468 * 1469 * This algorithm does one step per RTT at most and only if 1470 * we receive a bulk stream w/o packet losses or reorderings. 1471 * Shrinking the buffer during idle times is not necessary as 1472 * it doesn't consume any memory when idle. 1473 * 1474 * TODO: Only step up if the application is actually serving 1475 * the buffer to better manage the socket buffer resources. 1476 */ 1477 int 1478 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1479 struct tcpcb *tp, int tlen) 1480 { 1481 int newsize = 0; 1482 1483 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1484 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1485 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1486 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1487 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1488 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1489 newsize = min(so->so_rcv.sb_hiwat + 1490 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1491 } 1492 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1493 1494 /* Start over with next RTT. */ 1495 tp->rfbuf_ts = 0; 1496 tp->rfbuf_cnt = 0; 1497 } else { 1498 tp->rfbuf_cnt += tlen; /* add up */ 1499 } 1500 1501 return (newsize); 1502 } 1503 1504 void 1505 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1506 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1507 int ti_locked) 1508 { 1509 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1510 int rstreason, todrop, win; 1511 uint32_t tiwin; 1512 uint16_t nsegs; 1513 char *s; 1514 struct in_conninfo *inc; 1515 struct mbuf *mfree; 1516 struct tcpopt to; 1517 int tfo_syn; 1518 1519 #ifdef TCPDEBUG 1520 /* 1521 * The size of tcp_saveipgen must be the size of the max ip header, 1522 * now IPv6. 1523 */ 1524 u_char tcp_saveipgen[IP6_HDR_LEN]; 1525 struct tcphdr tcp_savetcp; 1526 short ostate = 0; 1527 #endif 1528 thflags = th->th_flags; 1529 inc = &tp->t_inpcb->inp_inc; 1530 tp->sackhint.last_sack_ack = 0; 1531 sack_changed = 0; 1532 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1533 1534 /* 1535 * If this is either a state-changing packet or current state isn't 1536 * established, we require a write lock on tcbinfo. Otherwise, we 1537 * allow the tcbinfo to be in either alocked or unlocked, as the 1538 * caller may have unnecessarily acquired a write lock due to a race. 1539 */ 1540 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1541 tp->t_state != TCPS_ESTABLISHED) { 1542 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1543 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1544 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1545 } else { 1546 #ifdef INVARIANTS 1547 if (ti_locked == TI_RLOCKED) 1548 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1549 else { 1550 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1551 "ti_locked: %d", __func__, ti_locked)); 1552 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1553 } 1554 #endif 1555 } 1556 INP_WLOCK_ASSERT(tp->t_inpcb); 1557 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1558 __func__)); 1559 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1560 __func__)); 1561 1562 #ifdef TCPPCAP 1563 /* Save segment, if requested. */ 1564 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1565 #endif 1566 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, 1567 tlen, NULL, true); 1568 1569 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1570 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1571 log(LOG_DEBUG, "%s; %s: " 1572 "SYN|FIN segment ignored (based on " 1573 "sysctl setting)\n", s, __func__); 1574 free(s, M_TCPLOG); 1575 } 1576 goto drop; 1577 } 1578 1579 /* 1580 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1581 * check SEQ.ACK first. 1582 */ 1583 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1584 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1585 rstreason = BANDLIM_UNLIMITED; 1586 goto dropwithreset; 1587 } 1588 1589 /* 1590 * Segment received on connection. 1591 * Reset idle time and keep-alive timer. 1592 * XXX: This should be done after segment 1593 * validation to ignore broken/spoofed segs. 1594 */ 1595 tp->t_rcvtime = ticks; 1596 1597 /* 1598 * Scale up the window into a 32-bit value. 1599 * For the SYN_SENT state the scale is zero. 1600 */ 1601 tiwin = th->th_win << tp->snd_scale; 1602 1603 /* 1604 * TCP ECN processing. 1605 */ 1606 if (tp->t_flags & TF_ECN_PERMIT) { 1607 if (thflags & TH_CWR) 1608 tp->t_flags &= ~TF_ECN_SND_ECE; 1609 switch (iptos & IPTOS_ECN_MASK) { 1610 case IPTOS_ECN_CE: 1611 tp->t_flags |= TF_ECN_SND_ECE; 1612 TCPSTAT_INC(tcps_ecn_ce); 1613 break; 1614 case IPTOS_ECN_ECT0: 1615 TCPSTAT_INC(tcps_ecn_ect0); 1616 break; 1617 case IPTOS_ECN_ECT1: 1618 TCPSTAT_INC(tcps_ecn_ect1); 1619 break; 1620 } 1621 1622 /* Process a packet differently from RFC3168. */ 1623 cc_ecnpkt_handler(tp, th, iptos); 1624 1625 /* Congestion experienced. */ 1626 if (thflags & TH_ECE) { 1627 cc_cong_signal(tp, th, CC_ECN); 1628 } 1629 } 1630 1631 /* 1632 * Parse options on any incoming segment. 1633 */ 1634 tcp_dooptions(&to, (u_char *)(th + 1), 1635 (th->th_off << 2) - sizeof(struct tcphdr), 1636 (thflags & TH_SYN) ? TO_SYN : 0); 1637 1638 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1639 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1640 (to.to_flags & TOF_SIGNATURE) == 0) { 1641 TCPSTAT_INC(tcps_sig_err_sigopt); 1642 /* XXX: should drop? */ 1643 } 1644 #endif 1645 /* 1646 * If echoed timestamp is later than the current time, 1647 * fall back to non RFC1323 RTT calculation. Normalize 1648 * timestamp if syncookies were used when this connection 1649 * was established. 1650 */ 1651 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1652 to.to_tsecr -= tp->ts_offset; 1653 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1654 to.to_tsecr = 0; 1655 else if (tp->t_flags & TF_PREVVALID && 1656 tp->t_badrxtwin != 0 && SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 1657 cc_cong_signal(tp, th, CC_RTO_ERR); 1658 } 1659 /* 1660 * Process options only when we get SYN/ACK back. The SYN case 1661 * for incoming connections is handled in tcp_syncache. 1662 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1663 * or <SYN,ACK>) segment itself is never scaled. 1664 * XXX this is traditional behavior, may need to be cleaned up. 1665 */ 1666 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1667 if ((to.to_flags & TOF_SCALE) && 1668 (tp->t_flags & TF_REQ_SCALE)) { 1669 tp->t_flags |= TF_RCVD_SCALE; 1670 tp->snd_scale = to.to_wscale; 1671 } 1672 /* 1673 * Initial send window. It will be updated with 1674 * the next incoming segment to the scaled value. 1675 */ 1676 tp->snd_wnd = th->th_win; 1677 if (to.to_flags & TOF_TS) { 1678 tp->t_flags |= TF_RCVD_TSTMP; 1679 tp->ts_recent = to.to_tsval; 1680 tp->ts_recent_age = tcp_ts_getticks(); 1681 } 1682 if (to.to_flags & TOF_MSS) 1683 tcp_mss(tp, to.to_mss); 1684 if ((tp->t_flags & TF_SACK_PERMIT) && 1685 (to.to_flags & TOF_SACKPERM) == 0) 1686 tp->t_flags &= ~TF_SACK_PERMIT; 1687 if (IS_FASTOPEN(tp->t_flags)) { 1688 if (to.to_flags & TOF_FASTOPEN) 1689 tcp_fastopen_update_cache(tp, to.to_mss, 1690 to.to_tfo_len, to.to_tfo_cookie); 1691 else 1692 tcp_fastopen_disable_path(tp); 1693 } 1694 } 1695 1696 /* 1697 * If timestamps were negotiated during SYN/ACK they should 1698 * appear on every segment during this session and vice versa. 1699 */ 1700 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1701 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1702 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1703 "no action\n", s, __func__); 1704 free(s, M_TCPLOG); 1705 } 1706 } 1707 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1708 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1709 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1710 "no action\n", s, __func__); 1711 free(s, M_TCPLOG); 1712 } 1713 } 1714 1715 /* 1716 * Header prediction: check for the two common cases 1717 * of a uni-directional data xfer. If the packet has 1718 * no control flags, is in-sequence, the window didn't 1719 * change and we're not retransmitting, it's a 1720 * candidate. If the length is zero and the ack moved 1721 * forward, we're the sender side of the xfer. Just 1722 * free the data acked & wake any higher level process 1723 * that was blocked waiting for space. If the length 1724 * is non-zero and the ack didn't move, we're the 1725 * receiver side. If we're getting packets in-order 1726 * (the reassembly queue is empty), add the data to 1727 * the socket buffer and note that we need a delayed ack. 1728 * Make sure that the hidden state-flags are also off. 1729 * Since we check for TCPS_ESTABLISHED first, it can only 1730 * be TH_NEEDSYN. 1731 */ 1732 if (tp->t_state == TCPS_ESTABLISHED && 1733 th->th_seq == tp->rcv_nxt && 1734 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1735 tp->snd_nxt == tp->snd_max && 1736 tiwin && tiwin == tp->snd_wnd && 1737 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1738 LIST_EMPTY(&tp->t_segq) && 1739 ((to.to_flags & TOF_TS) == 0 || 1740 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1741 1742 /* 1743 * If last ACK falls within this segment's sequence numbers, 1744 * record the timestamp. 1745 * NOTE that the test is modified according to the latest 1746 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1747 */ 1748 if ((to.to_flags & TOF_TS) != 0 && 1749 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1750 tp->ts_recent_age = tcp_ts_getticks(); 1751 tp->ts_recent = to.to_tsval; 1752 } 1753 1754 if (tlen == 0) { 1755 if (SEQ_GT(th->th_ack, tp->snd_una) && 1756 SEQ_LEQ(th->th_ack, tp->snd_max) && 1757 !IN_RECOVERY(tp->t_flags) && 1758 (to.to_flags & TOF_SACK) == 0 && 1759 TAILQ_EMPTY(&tp->snd_holes)) { 1760 /* 1761 * This is a pure ack for outstanding data. 1762 */ 1763 if (ti_locked == TI_RLOCKED) 1764 INP_INFO_RUNLOCK(&V_tcbinfo); 1765 ti_locked = TI_UNLOCKED; 1766 1767 TCPSTAT_INC(tcps_predack); 1768 1769 /* 1770 * "bad retransmit" recovery without timestamps. 1771 */ 1772 if ((to.to_flags & TOF_TS) == 0 && 1773 tp->t_rxtshift == 1 && 1774 tp->t_flags & TF_PREVVALID && 1775 (int)(ticks - tp->t_badrxtwin) < 0) { 1776 cc_cong_signal(tp, th, CC_RTO_ERR); 1777 } 1778 1779 /* 1780 * Recalculate the transmit timer / rtt. 1781 * 1782 * Some boxes send broken timestamp replies 1783 * during the SYN+ACK phase, ignore 1784 * timestamps of 0 or we could calculate a 1785 * huge RTT and blow up the retransmit timer. 1786 */ 1787 if ((to.to_flags & TOF_TS) != 0 && 1788 to.to_tsecr) { 1789 uint32_t t; 1790 1791 t = tcp_ts_getticks() - to.to_tsecr; 1792 if (!tp->t_rttlow || tp->t_rttlow > t) 1793 tp->t_rttlow = t; 1794 tcp_xmit_timer(tp, 1795 TCP_TS_TO_TICKS(t) + 1); 1796 } else if (tp->t_rtttime && 1797 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1798 if (!tp->t_rttlow || 1799 tp->t_rttlow > ticks - tp->t_rtttime) 1800 tp->t_rttlow = ticks - tp->t_rtttime; 1801 tcp_xmit_timer(tp, 1802 ticks - tp->t_rtttime); 1803 } 1804 acked = BYTES_THIS_ACK(tp, th); 1805 1806 #ifdef TCP_HHOOK 1807 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1808 hhook_run_tcp_est_in(tp, th, &to); 1809 #endif 1810 1811 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1812 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1813 sbdrop(&so->so_snd, acked); 1814 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1815 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1816 tp->snd_recover = th->th_ack - 1; 1817 1818 /* 1819 * Let the congestion control algorithm update 1820 * congestion control related information. This 1821 * typically means increasing the congestion 1822 * window. 1823 */ 1824 cc_ack_received(tp, th, nsegs, CC_ACK); 1825 1826 tp->snd_una = th->th_ack; 1827 /* 1828 * Pull snd_wl2 up to prevent seq wrap relative 1829 * to th_ack. 1830 */ 1831 tp->snd_wl2 = th->th_ack; 1832 tp->t_dupacks = 0; 1833 m_freem(m); 1834 1835 /* 1836 * If all outstanding data are acked, stop 1837 * retransmit timer, otherwise restart timer 1838 * using current (possibly backed-off) value. 1839 * If process is waiting for space, 1840 * wakeup/selwakeup/signal. If data 1841 * are ready to send, let tcp_output 1842 * decide between more output or persist. 1843 */ 1844 #ifdef TCPDEBUG 1845 if (so->so_options & SO_DEBUG) 1846 tcp_trace(TA_INPUT, ostate, tp, 1847 (void *)tcp_saveipgen, 1848 &tcp_savetcp, 0); 1849 #endif 1850 TCP_PROBE3(debug__input, tp, th, m); 1851 if (tp->snd_una == tp->snd_max) 1852 tcp_timer_activate(tp, TT_REXMT, 0); 1853 else if (!tcp_timer_active(tp, TT_PERSIST)) 1854 tcp_timer_activate(tp, TT_REXMT, 1855 tp->t_rxtcur); 1856 sowwakeup(so); 1857 if (sbavail(&so->so_snd)) 1858 (void) tp->t_fb->tfb_tcp_output(tp); 1859 goto check_delack; 1860 } 1861 } else if (th->th_ack == tp->snd_una && 1862 tlen <= sbspace(&so->so_rcv)) { 1863 int newsize = 0; /* automatic sockbuf scaling */ 1864 1865 /* 1866 * This is a pure, in-sequence data packet with 1867 * nothing on the reassembly queue and we have enough 1868 * buffer space to take it. 1869 */ 1870 if (ti_locked == TI_RLOCKED) 1871 INP_INFO_RUNLOCK(&V_tcbinfo); 1872 ti_locked = TI_UNLOCKED; 1873 1874 /* Clean receiver SACK report if present */ 1875 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1876 tcp_clean_sackreport(tp); 1877 TCPSTAT_INC(tcps_preddat); 1878 tp->rcv_nxt += tlen; 1879 /* 1880 * Pull snd_wl1 up to prevent seq wrap relative to 1881 * th_seq. 1882 */ 1883 tp->snd_wl1 = th->th_seq; 1884 /* 1885 * Pull rcv_up up to prevent seq wrap relative to 1886 * rcv_nxt. 1887 */ 1888 tp->rcv_up = tp->rcv_nxt; 1889 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1890 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1891 #ifdef TCPDEBUG 1892 if (so->so_options & SO_DEBUG) 1893 tcp_trace(TA_INPUT, ostate, tp, 1894 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1895 #endif 1896 TCP_PROBE3(debug__input, tp, th, m); 1897 1898 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1899 1900 /* Add data to socket buffer. */ 1901 SOCKBUF_LOCK(&so->so_rcv); 1902 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1903 m_freem(m); 1904 } else { 1905 /* 1906 * Set new socket buffer size. 1907 * Give up when limit is reached. 1908 */ 1909 if (newsize) 1910 if (!sbreserve_locked(&so->so_rcv, 1911 newsize, so, NULL)) 1912 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1913 m_adj(m, drop_hdrlen); /* delayed header drop */ 1914 sbappendstream_locked(&so->so_rcv, m, 0); 1915 } 1916 /* NB: sorwakeup_locked() does an implicit unlock. */ 1917 sorwakeup_locked(so); 1918 if (DELAY_ACK(tp, tlen)) { 1919 tp->t_flags |= TF_DELACK; 1920 } else { 1921 tp->t_flags |= TF_ACKNOW; 1922 tp->t_fb->tfb_tcp_output(tp); 1923 } 1924 goto check_delack; 1925 } 1926 } 1927 1928 /* 1929 * Calculate amount of space in receive window, 1930 * and then do TCP input processing. 1931 * Receive window is amount of space in rcv queue, 1932 * but not less than advertised window. 1933 */ 1934 win = sbspace(&so->so_rcv); 1935 if (win < 0) 1936 win = 0; 1937 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1938 1939 switch (tp->t_state) { 1940 1941 /* 1942 * If the state is SYN_RECEIVED: 1943 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1944 */ 1945 case TCPS_SYN_RECEIVED: 1946 if ((thflags & TH_ACK) && 1947 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1948 SEQ_GT(th->th_ack, tp->snd_max))) { 1949 rstreason = BANDLIM_RST_OPENPORT; 1950 goto dropwithreset; 1951 } 1952 if (IS_FASTOPEN(tp->t_flags)) { 1953 /* 1954 * When a TFO connection is in SYN_RECEIVED, the 1955 * only valid packets are the initial SYN, a 1956 * retransmit/copy of the initial SYN (possibly with 1957 * a subset of the original data), a valid ACK, a 1958 * FIN, or a RST. 1959 */ 1960 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1961 rstreason = BANDLIM_RST_OPENPORT; 1962 goto dropwithreset; 1963 } else if (thflags & TH_SYN) { 1964 /* non-initial SYN is ignored */ 1965 if ((tcp_timer_active(tp, TT_DELACK) || 1966 tcp_timer_active(tp, TT_REXMT))) 1967 goto drop; 1968 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1969 goto drop; 1970 } 1971 } 1972 break; 1973 1974 /* 1975 * If the state is SYN_SENT: 1976 * if seg contains a RST with valid ACK (SEQ.ACK has already 1977 * been verified), then drop the connection. 1978 * if seg contains a RST without an ACK, drop the seg. 1979 * if seg does not contain SYN, then drop the seg. 1980 * Otherwise this is an acceptable SYN segment 1981 * initialize tp->rcv_nxt and tp->irs 1982 * if seg contains ack then advance tp->snd_una 1983 * if seg contains an ECE and ECN support is enabled, the stream 1984 * is ECN capable. 1985 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1986 * arrange for segment to be acked (eventually) 1987 * continue processing rest of data/controls, beginning with URG 1988 */ 1989 case TCPS_SYN_SENT: 1990 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1991 TCP_PROBE5(connect__refused, NULL, tp, 1992 m, tp, th); 1993 tp = tcp_drop(tp, ECONNREFUSED); 1994 } 1995 if (thflags & TH_RST) 1996 goto drop; 1997 if (!(thflags & TH_SYN)) 1998 goto drop; 1999 2000 tp->irs = th->th_seq; 2001 tcp_rcvseqinit(tp); 2002 if (thflags & TH_ACK) { 2003 int tfo_partial_ack = 0; 2004 2005 TCPSTAT_INC(tcps_connects); 2006 soisconnected(so); 2007 #ifdef MAC 2008 mac_socketpeer_set_from_mbuf(m, so); 2009 #endif 2010 /* Do window scaling on this connection? */ 2011 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2012 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2013 tp->rcv_scale = tp->request_r_scale; 2014 } 2015 tp->rcv_adv += min(tp->rcv_wnd, 2016 TCP_MAXWIN << tp->rcv_scale); 2017 tp->snd_una++; /* SYN is acked */ 2018 /* 2019 * If not all the data that was sent in the TFO SYN 2020 * has been acked, resend the remainder right away. 2021 */ 2022 if (IS_FASTOPEN(tp->t_flags) && 2023 (tp->snd_una != tp->snd_max)) { 2024 tp->snd_nxt = th->th_ack; 2025 tfo_partial_ack = 1; 2026 } 2027 /* 2028 * If there's data, delay ACK; if there's also a FIN 2029 * ACKNOW will be turned on later. 2030 */ 2031 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2032 tcp_timer_activate(tp, TT_DELACK, 2033 tcp_delacktime); 2034 else 2035 tp->t_flags |= TF_ACKNOW; 2036 2037 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2038 tp->t_flags |= TF_ECN_PERMIT; 2039 TCPSTAT_INC(tcps_ecn_shs); 2040 } 2041 2042 /* 2043 * Received <SYN,ACK> in SYN_SENT[*] state. 2044 * Transitions: 2045 * SYN_SENT --> ESTABLISHED 2046 * SYN_SENT* --> FIN_WAIT_1 2047 */ 2048 tp->t_starttime = ticks; 2049 if (tp->t_flags & TF_NEEDFIN) { 2050 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2051 tp->t_flags &= ~TF_NEEDFIN; 2052 thflags &= ~TH_SYN; 2053 } else { 2054 tcp_state_change(tp, TCPS_ESTABLISHED); 2055 TCP_PROBE5(connect__established, NULL, tp, 2056 m, tp, th); 2057 cc_conn_init(tp); 2058 tcp_timer_activate(tp, TT_KEEP, 2059 TP_KEEPIDLE(tp)); 2060 } 2061 } else { 2062 /* 2063 * Received initial SYN in SYN-SENT[*] state => 2064 * simultaneous open. 2065 * If it succeeds, connection is * half-synchronized. 2066 * Otherwise, do 3-way handshake: 2067 * SYN-SENT -> SYN-RECEIVED 2068 * SYN-SENT* -> SYN-RECEIVED* 2069 */ 2070 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2071 tcp_timer_activate(tp, TT_REXMT, 0); 2072 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2073 } 2074 2075 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2076 "ti_locked %d", __func__, ti_locked)); 2077 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2078 INP_WLOCK_ASSERT(tp->t_inpcb); 2079 2080 /* 2081 * Advance th->th_seq to correspond to first data byte. 2082 * If data, trim to stay within window, 2083 * dropping FIN if necessary. 2084 */ 2085 th->th_seq++; 2086 if (tlen > tp->rcv_wnd) { 2087 todrop = tlen - tp->rcv_wnd; 2088 m_adj(m, -todrop); 2089 tlen = tp->rcv_wnd; 2090 thflags &= ~TH_FIN; 2091 TCPSTAT_INC(tcps_rcvpackafterwin); 2092 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2093 } 2094 tp->snd_wl1 = th->th_seq - 1; 2095 tp->rcv_up = th->th_seq; 2096 /* 2097 * Client side of transaction: already sent SYN and data. 2098 * If the remote host used T/TCP to validate the SYN, 2099 * our data will be ACK'd; if so, enter normal data segment 2100 * processing in the middle of step 5, ack processing. 2101 * Otherwise, goto step 6. 2102 */ 2103 if (thflags & TH_ACK) 2104 goto process_ACK; 2105 2106 goto step6; 2107 2108 /* 2109 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2110 * do normal processing. 2111 * 2112 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2113 */ 2114 case TCPS_LAST_ACK: 2115 case TCPS_CLOSING: 2116 break; /* continue normal processing */ 2117 } 2118 2119 /* 2120 * States other than LISTEN or SYN_SENT. 2121 * First check the RST flag and sequence number since reset segments 2122 * are exempt from the timestamp and connection count tests. This 2123 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2124 * below which allowed reset segments in half the sequence space 2125 * to fall though and be processed (which gives forged reset 2126 * segments with a random sequence number a 50 percent chance of 2127 * killing a connection). 2128 * Then check timestamp, if present. 2129 * Then check the connection count, if present. 2130 * Then check that at least some bytes of segment are within 2131 * receive window. If segment begins before rcv_nxt, 2132 * drop leading data (and SYN); if nothing left, just ack. 2133 */ 2134 if (thflags & TH_RST) { 2135 /* 2136 * RFC5961 Section 3.2 2137 * 2138 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2139 * - If RST is in window, we send challenge ACK. 2140 * 2141 * Note: to take into account delayed ACKs, we should 2142 * test against last_ack_sent instead of rcv_nxt. 2143 * Note 2: we handle special case of closed window, not 2144 * covered by the RFC. 2145 */ 2146 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2147 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2148 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2149 2150 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2151 KASSERT(ti_locked == TI_RLOCKED, 2152 ("%s: TH_RST ti_locked %d, th %p tp %p", 2153 __func__, ti_locked, th, tp)); 2154 KASSERT(tp->t_state != TCPS_SYN_SENT, 2155 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2156 __func__, th, tp)); 2157 2158 if (V_tcp_insecure_rst || 2159 tp->last_ack_sent == th->th_seq) { 2160 TCPSTAT_INC(tcps_drops); 2161 /* Drop the connection. */ 2162 switch (tp->t_state) { 2163 case TCPS_SYN_RECEIVED: 2164 so->so_error = ECONNREFUSED; 2165 goto close; 2166 case TCPS_ESTABLISHED: 2167 case TCPS_FIN_WAIT_1: 2168 case TCPS_FIN_WAIT_2: 2169 case TCPS_CLOSE_WAIT: 2170 case TCPS_CLOSING: 2171 case TCPS_LAST_ACK: 2172 so->so_error = ECONNRESET; 2173 close: 2174 /* FALLTHROUGH */ 2175 default: 2176 tp = tcp_close(tp); 2177 } 2178 } else { 2179 TCPSTAT_INC(tcps_badrst); 2180 /* Send challenge ACK. */ 2181 tcp_respond(tp, mtod(m, void *), th, m, 2182 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2183 tp->last_ack_sent = tp->rcv_nxt; 2184 m = NULL; 2185 } 2186 } 2187 goto drop; 2188 } 2189 2190 /* 2191 * RFC5961 Section 4.2 2192 * Send challenge ACK for any SYN in synchronized state. 2193 */ 2194 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2195 tp->t_state != TCPS_SYN_RECEIVED) { 2196 KASSERT(ti_locked == TI_RLOCKED, 2197 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2198 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2199 2200 TCPSTAT_INC(tcps_badsyn); 2201 if (V_tcp_insecure_syn && 2202 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2203 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2204 tp = tcp_drop(tp, ECONNRESET); 2205 rstreason = BANDLIM_UNLIMITED; 2206 } else { 2207 /* Send challenge ACK. */ 2208 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2209 tp->snd_nxt, TH_ACK); 2210 tp->last_ack_sent = tp->rcv_nxt; 2211 m = NULL; 2212 } 2213 goto drop; 2214 } 2215 2216 /* 2217 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2218 * and it's less than ts_recent, drop it. 2219 */ 2220 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2221 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2222 2223 /* Check to see if ts_recent is over 24 days old. */ 2224 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2225 /* 2226 * Invalidate ts_recent. If this segment updates 2227 * ts_recent, the age will be reset later and ts_recent 2228 * will get a valid value. If it does not, setting 2229 * ts_recent to zero will at least satisfy the 2230 * requirement that zero be placed in the timestamp 2231 * echo reply when ts_recent isn't valid. The 2232 * age isn't reset until we get a valid ts_recent 2233 * because we don't want out-of-order segments to be 2234 * dropped when ts_recent is old. 2235 */ 2236 tp->ts_recent = 0; 2237 } else { 2238 TCPSTAT_INC(tcps_rcvduppack); 2239 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2240 TCPSTAT_INC(tcps_pawsdrop); 2241 if (tlen) 2242 goto dropafterack; 2243 goto drop; 2244 } 2245 } 2246 2247 /* 2248 * In the SYN-RECEIVED state, validate that the packet belongs to 2249 * this connection before trimming the data to fit the receive 2250 * window. Check the sequence number versus IRS since we know 2251 * the sequence numbers haven't wrapped. This is a partial fix 2252 * for the "LAND" DoS attack. 2253 */ 2254 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2255 rstreason = BANDLIM_RST_OPENPORT; 2256 goto dropwithreset; 2257 } 2258 2259 todrop = tp->rcv_nxt - th->th_seq; 2260 if (todrop > 0) { 2261 if (thflags & TH_SYN) { 2262 thflags &= ~TH_SYN; 2263 th->th_seq++; 2264 if (th->th_urp > 1) 2265 th->th_urp--; 2266 else 2267 thflags &= ~TH_URG; 2268 todrop--; 2269 } 2270 /* 2271 * Following if statement from Stevens, vol. 2, p. 960. 2272 */ 2273 if (todrop > tlen 2274 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2275 /* 2276 * Any valid FIN must be to the left of the window. 2277 * At this point the FIN must be a duplicate or out 2278 * of sequence; drop it. 2279 */ 2280 thflags &= ~TH_FIN; 2281 2282 /* 2283 * Send an ACK to resynchronize and drop any data. 2284 * But keep on processing for RST or ACK. 2285 */ 2286 tp->t_flags |= TF_ACKNOW; 2287 todrop = tlen; 2288 TCPSTAT_INC(tcps_rcvduppack); 2289 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2290 } else { 2291 TCPSTAT_INC(tcps_rcvpartduppack); 2292 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2293 } 2294 drop_hdrlen += todrop; /* drop from the top afterwards */ 2295 th->th_seq += todrop; 2296 tlen -= todrop; 2297 if (th->th_urp > todrop) 2298 th->th_urp -= todrop; 2299 else { 2300 thflags &= ~TH_URG; 2301 th->th_urp = 0; 2302 } 2303 } 2304 2305 /* 2306 * If new data are received on a connection after the 2307 * user processes are gone, then RST the other end. 2308 */ 2309 if ((so->so_state & SS_NOFDREF) && 2310 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2311 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2312 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2313 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2314 2315 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2316 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2317 "after socket was closed, " 2318 "sending RST and removing tcpcb\n", 2319 s, __func__, tcpstates[tp->t_state], tlen); 2320 free(s, M_TCPLOG); 2321 } 2322 tp = tcp_close(tp); 2323 TCPSTAT_INC(tcps_rcvafterclose); 2324 rstreason = BANDLIM_UNLIMITED; 2325 goto dropwithreset; 2326 } 2327 2328 /* 2329 * If segment ends after window, drop trailing data 2330 * (and PUSH and FIN); if nothing left, just ACK. 2331 */ 2332 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2333 if (todrop > 0) { 2334 TCPSTAT_INC(tcps_rcvpackafterwin); 2335 if (todrop >= tlen) { 2336 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2337 /* 2338 * If window is closed can only take segments at 2339 * window edge, and have to drop data and PUSH from 2340 * incoming segments. Continue processing, but 2341 * remember to ack. Otherwise, drop segment 2342 * and ack. 2343 */ 2344 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2345 tp->t_flags |= TF_ACKNOW; 2346 TCPSTAT_INC(tcps_rcvwinprobe); 2347 } else 2348 goto dropafterack; 2349 } else 2350 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2351 m_adj(m, -todrop); 2352 tlen -= todrop; 2353 thflags &= ~(TH_PUSH|TH_FIN); 2354 } 2355 2356 /* 2357 * If last ACK falls within this segment's sequence numbers, 2358 * record its timestamp. 2359 * NOTE: 2360 * 1) That the test incorporates suggestions from the latest 2361 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2362 * 2) That updating only on newer timestamps interferes with 2363 * our earlier PAWS tests, so this check should be solely 2364 * predicated on the sequence space of this segment. 2365 * 3) That we modify the segment boundary check to be 2366 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2367 * instead of RFC1323's 2368 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2369 * This modified check allows us to overcome RFC1323's 2370 * limitations as described in Stevens TCP/IP Illustrated 2371 * Vol. 2 p.869. In such cases, we can still calculate the 2372 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2373 */ 2374 if ((to.to_flags & TOF_TS) != 0 && 2375 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2376 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2377 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2378 tp->ts_recent_age = tcp_ts_getticks(); 2379 tp->ts_recent = to.to_tsval; 2380 } 2381 2382 /* 2383 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2384 * flag is on (half-synchronized state), then queue data for 2385 * later processing; else drop segment and return. 2386 */ 2387 if ((thflags & TH_ACK) == 0) { 2388 if (tp->t_state == TCPS_SYN_RECEIVED || 2389 (tp->t_flags & TF_NEEDSYN)) { 2390 if (tp->t_state == TCPS_SYN_RECEIVED && 2391 IS_FASTOPEN(tp->t_flags)) { 2392 tp->snd_wnd = tiwin; 2393 cc_conn_init(tp); 2394 } 2395 goto step6; 2396 } else if (tp->t_flags & TF_ACKNOW) 2397 goto dropafterack; 2398 else 2399 goto drop; 2400 } 2401 2402 /* 2403 * Ack processing. 2404 */ 2405 switch (tp->t_state) { 2406 2407 /* 2408 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2409 * ESTABLISHED state and continue processing. 2410 * The ACK was checked above. 2411 */ 2412 case TCPS_SYN_RECEIVED: 2413 2414 TCPSTAT_INC(tcps_connects); 2415 soisconnected(so); 2416 /* Do window scaling? */ 2417 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2418 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2419 tp->rcv_scale = tp->request_r_scale; 2420 tp->snd_wnd = tiwin; 2421 } 2422 /* 2423 * Make transitions: 2424 * SYN-RECEIVED -> ESTABLISHED 2425 * SYN-RECEIVED* -> FIN-WAIT-1 2426 */ 2427 tp->t_starttime = ticks; 2428 if (tp->t_flags & TF_NEEDFIN) { 2429 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2430 tp->t_flags &= ~TF_NEEDFIN; 2431 } else { 2432 tcp_state_change(tp, TCPS_ESTABLISHED); 2433 TCP_PROBE5(accept__established, NULL, tp, 2434 m, tp, th); 2435 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2436 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2437 tp->t_tfo_pending = NULL; 2438 2439 /* 2440 * Account for the ACK of our SYN prior to 2441 * regular ACK processing below. 2442 */ 2443 tp->snd_una++; 2444 } 2445 /* 2446 * TFO connections call cc_conn_init() during SYN 2447 * processing. Calling it again here for such 2448 * connections is not harmless as it would undo the 2449 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2450 * is retransmitted. 2451 */ 2452 if (!IS_FASTOPEN(tp->t_flags)) 2453 cc_conn_init(tp); 2454 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2455 } 2456 /* 2457 * If segment contains data or ACK, will call tcp_reass() 2458 * later; if not, do so now to pass queued data to user. 2459 */ 2460 if (tlen == 0 && (thflags & TH_FIN) == 0) 2461 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2462 (struct mbuf *)0); 2463 tp->snd_wl1 = th->th_seq - 1; 2464 /* FALLTHROUGH */ 2465 2466 /* 2467 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2468 * ACKs. If the ack is in the range 2469 * tp->snd_una < th->th_ack <= tp->snd_max 2470 * then advance tp->snd_una to th->th_ack and drop 2471 * data from the retransmission queue. If this ACK reflects 2472 * more up to date window information we update our window information. 2473 */ 2474 case TCPS_ESTABLISHED: 2475 case TCPS_FIN_WAIT_1: 2476 case TCPS_FIN_WAIT_2: 2477 case TCPS_CLOSE_WAIT: 2478 case TCPS_CLOSING: 2479 case TCPS_LAST_ACK: 2480 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2481 TCPSTAT_INC(tcps_rcvacktoomuch); 2482 goto dropafterack; 2483 } 2484 if ((tp->t_flags & TF_SACK_PERMIT) && 2485 ((to.to_flags & TOF_SACK) || 2486 !TAILQ_EMPTY(&tp->snd_holes))) 2487 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2488 else 2489 /* 2490 * Reset the value so that previous (valid) value 2491 * from the last ack with SACK doesn't get used. 2492 */ 2493 tp->sackhint.sacked_bytes = 0; 2494 2495 #ifdef TCP_HHOOK 2496 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2497 hhook_run_tcp_est_in(tp, th, &to); 2498 #endif 2499 2500 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2501 u_int maxseg; 2502 2503 maxseg = tcp_maxseg(tp); 2504 if (tlen == 0 && 2505 (tiwin == tp->snd_wnd || 2506 (tp->t_flags & TF_SACK_PERMIT))) { 2507 /* 2508 * If this is the first time we've seen a 2509 * FIN from the remote, this is not a 2510 * duplicate and it needs to be processed 2511 * normally. This happens during a 2512 * simultaneous close. 2513 */ 2514 if ((thflags & TH_FIN) && 2515 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2516 tp->t_dupacks = 0; 2517 break; 2518 } 2519 TCPSTAT_INC(tcps_rcvdupack); 2520 /* 2521 * If we have outstanding data (other than 2522 * a window probe), this is a completely 2523 * duplicate ack (ie, window info didn't 2524 * change and FIN isn't set), 2525 * the ack is the biggest we've 2526 * seen and we've seen exactly our rexmt 2527 * threshold of them, assume a packet 2528 * has been dropped and retransmit it. 2529 * Kludge snd_nxt & the congestion 2530 * window so we send only this one 2531 * packet. 2532 * 2533 * We know we're losing at the current 2534 * window size so do congestion avoidance 2535 * (set ssthresh to half the current window 2536 * and pull our congestion window back to 2537 * the new ssthresh). 2538 * 2539 * Dup acks mean that packets have left the 2540 * network (they're now cached at the receiver) 2541 * so bump cwnd by the amount in the receiver 2542 * to keep a constant cwnd packets in the 2543 * network. 2544 * 2545 * When using TCP ECN, notify the peer that 2546 * we reduced the cwnd. 2547 */ 2548 /* 2549 * Following 2 kinds of acks should not affect 2550 * dupack counting: 2551 * 1) Old acks 2552 * 2) Acks with SACK but without any new SACK 2553 * information in them. These could result from 2554 * any anomaly in the network like a switch 2555 * duplicating packets or a possible DoS attack. 2556 */ 2557 if (th->th_ack != tp->snd_una || 2558 ((tp->t_flags & TF_SACK_PERMIT) && 2559 !sack_changed)) 2560 break; 2561 else if (!tcp_timer_active(tp, TT_REXMT)) 2562 tp->t_dupacks = 0; 2563 else if (++tp->t_dupacks > tcprexmtthresh || 2564 IN_FASTRECOVERY(tp->t_flags)) { 2565 cc_ack_received(tp, th, nsegs, 2566 CC_DUPACK); 2567 if ((tp->t_flags & TF_SACK_PERMIT) && 2568 IN_FASTRECOVERY(tp->t_flags)) { 2569 int awnd; 2570 2571 /* 2572 * Compute the amount of data in flight first. 2573 * We can inject new data into the pipe iff 2574 * we have less than 1/2 the original window's 2575 * worth of data in flight. 2576 */ 2577 if (V_tcp_do_rfc6675_pipe) 2578 awnd = tcp_compute_pipe(tp); 2579 else 2580 awnd = (tp->snd_nxt - tp->snd_fack) + 2581 tp->sackhint.sack_bytes_rexmit; 2582 2583 if (awnd < tp->snd_ssthresh) { 2584 tp->snd_cwnd += maxseg; 2585 if (tp->snd_cwnd > tp->snd_ssthresh) 2586 tp->snd_cwnd = tp->snd_ssthresh; 2587 } 2588 } else 2589 tp->snd_cwnd += maxseg; 2590 (void) tp->t_fb->tfb_tcp_output(tp); 2591 goto drop; 2592 } else if (tp->t_dupacks == tcprexmtthresh) { 2593 tcp_seq onxt = tp->snd_nxt; 2594 2595 /* 2596 * If we're doing sack, check to 2597 * see if we're already in sack 2598 * recovery. If we're not doing sack, 2599 * check to see if we're in newreno 2600 * recovery. 2601 */ 2602 if (tp->t_flags & TF_SACK_PERMIT) { 2603 if (IN_FASTRECOVERY(tp->t_flags)) { 2604 tp->t_dupacks = 0; 2605 break; 2606 } 2607 } else { 2608 if (SEQ_LEQ(th->th_ack, 2609 tp->snd_recover)) { 2610 tp->t_dupacks = 0; 2611 break; 2612 } 2613 } 2614 /* Congestion signal before ack. */ 2615 cc_cong_signal(tp, th, CC_NDUPACK); 2616 cc_ack_received(tp, th, nsegs, 2617 CC_DUPACK); 2618 tcp_timer_activate(tp, TT_REXMT, 0); 2619 tp->t_rtttime = 0; 2620 if (tp->t_flags & TF_SACK_PERMIT) { 2621 TCPSTAT_INC( 2622 tcps_sack_recovery_episode); 2623 tp->sack_newdata = tp->snd_nxt; 2624 tp->snd_cwnd = maxseg; 2625 (void) tp->t_fb->tfb_tcp_output(tp); 2626 goto drop; 2627 } 2628 tp->snd_nxt = th->th_ack; 2629 tp->snd_cwnd = maxseg; 2630 (void) tp->t_fb->tfb_tcp_output(tp); 2631 KASSERT(tp->snd_limited <= 2, 2632 ("%s: tp->snd_limited too big", 2633 __func__)); 2634 tp->snd_cwnd = tp->snd_ssthresh + 2635 maxseg * 2636 (tp->t_dupacks - tp->snd_limited); 2637 if (SEQ_GT(onxt, tp->snd_nxt)) 2638 tp->snd_nxt = onxt; 2639 goto drop; 2640 } else if (V_tcp_do_rfc3042) { 2641 /* 2642 * Process first and second duplicate 2643 * ACKs. Each indicates a segment 2644 * leaving the network, creating room 2645 * for more. Make sure we can send a 2646 * packet on reception of each duplicate 2647 * ACK by increasing snd_cwnd by one 2648 * segment. Restore the original 2649 * snd_cwnd after packet transmission. 2650 */ 2651 cc_ack_received(tp, th, nsegs, 2652 CC_DUPACK); 2653 uint32_t oldcwnd = tp->snd_cwnd; 2654 tcp_seq oldsndmax = tp->snd_max; 2655 u_int sent; 2656 int avail; 2657 2658 KASSERT(tp->t_dupacks == 1 || 2659 tp->t_dupacks == 2, 2660 ("%s: dupacks not 1 or 2", 2661 __func__)); 2662 if (tp->t_dupacks == 1) 2663 tp->snd_limited = 0; 2664 tp->snd_cwnd = 2665 (tp->snd_nxt - tp->snd_una) + 2666 (tp->t_dupacks - tp->snd_limited) * 2667 maxseg; 2668 /* 2669 * Only call tcp_output when there 2670 * is new data available to be sent. 2671 * Otherwise we would send pure ACKs. 2672 */ 2673 SOCKBUF_LOCK(&so->so_snd); 2674 avail = sbavail(&so->so_snd) - 2675 (tp->snd_nxt - tp->snd_una); 2676 SOCKBUF_UNLOCK(&so->so_snd); 2677 if (avail > 0) 2678 (void) tp->t_fb->tfb_tcp_output(tp); 2679 sent = tp->snd_max - oldsndmax; 2680 if (sent > maxseg) { 2681 KASSERT((tp->t_dupacks == 2 && 2682 tp->snd_limited == 0) || 2683 (sent == maxseg + 1 && 2684 tp->t_flags & TF_SENTFIN), 2685 ("%s: sent too much", 2686 __func__)); 2687 tp->snd_limited = 2; 2688 } else if (sent > 0) 2689 ++tp->snd_limited; 2690 tp->snd_cwnd = oldcwnd; 2691 goto drop; 2692 } 2693 } 2694 break; 2695 } else { 2696 /* 2697 * This ack is advancing the left edge, reset the 2698 * counter. 2699 */ 2700 tp->t_dupacks = 0; 2701 /* 2702 * If this ack also has new SACK info, increment the 2703 * counter as per rfc6675. 2704 */ 2705 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2706 tp->t_dupacks++; 2707 } 2708 2709 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2710 ("%s: th_ack <= snd_una", __func__)); 2711 2712 /* 2713 * If the congestion window was inflated to account 2714 * for the other side's cached packets, retract it. 2715 */ 2716 if (IN_FASTRECOVERY(tp->t_flags)) { 2717 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2718 if (tp->t_flags & TF_SACK_PERMIT) 2719 tcp_sack_partialack(tp, th); 2720 else 2721 tcp_newreno_partial_ack(tp, th); 2722 } else 2723 cc_post_recovery(tp, th); 2724 } 2725 /* 2726 * If we reach this point, ACK is not a duplicate, 2727 * i.e., it ACKs something we sent. 2728 */ 2729 if (tp->t_flags & TF_NEEDSYN) { 2730 /* 2731 * T/TCP: Connection was half-synchronized, and our 2732 * SYN has been ACK'd (so connection is now fully 2733 * synchronized). Go to non-starred state, 2734 * increment snd_una for ACK of SYN, and check if 2735 * we can do window scaling. 2736 */ 2737 tp->t_flags &= ~TF_NEEDSYN; 2738 tp->snd_una++; 2739 /* Do window scaling? */ 2740 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2741 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2742 tp->rcv_scale = tp->request_r_scale; 2743 /* Send window already scaled. */ 2744 } 2745 } 2746 2747 process_ACK: 2748 INP_WLOCK_ASSERT(tp->t_inpcb); 2749 2750 acked = BYTES_THIS_ACK(tp, th); 2751 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2752 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2753 tp->snd_una, th->th_ack, tp, m)); 2754 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2755 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2756 2757 /* 2758 * If we just performed our first retransmit, and the ACK 2759 * arrives within our recovery window, then it was a mistake 2760 * to do the retransmit in the first place. Recover our 2761 * original cwnd and ssthresh, and proceed to transmit where 2762 * we left off. 2763 */ 2764 if (tp->t_rxtshift == 1 && 2765 tp->t_flags & TF_PREVVALID && 2766 tp->t_badrxtwin && 2767 SEQ_LT(to.to_tsecr, tp->t_badrxtwin)) 2768 cc_cong_signal(tp, th, CC_RTO_ERR); 2769 2770 /* 2771 * If we have a timestamp reply, update smoothed 2772 * round trip time. If no timestamp is present but 2773 * transmit timer is running and timed sequence 2774 * number was acked, update smoothed round trip time. 2775 * Since we now have an rtt measurement, cancel the 2776 * timer backoff (cf., Phil Karn's retransmit alg.). 2777 * Recompute the initial retransmit timer. 2778 * 2779 * Some boxes send broken timestamp replies 2780 * during the SYN+ACK phase, ignore 2781 * timestamps of 0 or we could calculate a 2782 * huge RTT and blow up the retransmit timer. 2783 */ 2784 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2785 uint32_t t; 2786 2787 t = tcp_ts_getticks() - to.to_tsecr; 2788 if (!tp->t_rttlow || tp->t_rttlow > t) 2789 tp->t_rttlow = t; 2790 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2791 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2792 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2793 tp->t_rttlow = ticks - tp->t_rtttime; 2794 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2795 } 2796 2797 /* 2798 * If all outstanding data is acked, stop retransmit 2799 * timer and remember to restart (more output or persist). 2800 * If there is more data to be acked, restart retransmit 2801 * timer, using current (possibly backed-off) value. 2802 */ 2803 if (th->th_ack == tp->snd_max) { 2804 tcp_timer_activate(tp, TT_REXMT, 0); 2805 needoutput = 1; 2806 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2807 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2808 2809 /* 2810 * If no data (only SYN) was ACK'd, 2811 * skip rest of ACK processing. 2812 */ 2813 if (acked == 0) 2814 goto step6; 2815 2816 /* 2817 * Let the congestion control algorithm update congestion 2818 * control related information. This typically means increasing 2819 * the congestion window. 2820 */ 2821 cc_ack_received(tp, th, nsegs, CC_ACK); 2822 2823 SOCKBUF_LOCK(&so->so_snd); 2824 if (acked > sbavail(&so->so_snd)) { 2825 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2826 tp->snd_wnd -= sbavail(&so->so_snd); 2827 else 2828 tp->snd_wnd = 0; 2829 mfree = sbcut_locked(&so->so_snd, 2830 (int)sbavail(&so->so_snd)); 2831 ourfinisacked = 1; 2832 } else { 2833 mfree = sbcut_locked(&so->so_snd, acked); 2834 if (tp->snd_wnd >= (uint32_t) acked) 2835 tp->snd_wnd -= acked; 2836 else 2837 tp->snd_wnd = 0; 2838 ourfinisacked = 0; 2839 } 2840 /* NB: sowwakeup_locked() does an implicit unlock. */ 2841 sowwakeup_locked(so); 2842 m_freem(mfree); 2843 /* Detect una wraparound. */ 2844 if (!IN_RECOVERY(tp->t_flags) && 2845 SEQ_GT(tp->snd_una, tp->snd_recover) && 2846 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2847 tp->snd_recover = th->th_ack - 1; 2848 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2849 if (IN_RECOVERY(tp->t_flags) && 2850 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2851 EXIT_RECOVERY(tp->t_flags); 2852 } 2853 tp->snd_una = th->th_ack; 2854 if (tp->t_flags & TF_SACK_PERMIT) { 2855 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2856 tp->snd_recover = tp->snd_una; 2857 } 2858 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2859 tp->snd_nxt = tp->snd_una; 2860 2861 switch (tp->t_state) { 2862 2863 /* 2864 * In FIN_WAIT_1 STATE in addition to the processing 2865 * for the ESTABLISHED state if our FIN is now acknowledged 2866 * then enter FIN_WAIT_2. 2867 */ 2868 case TCPS_FIN_WAIT_1: 2869 if (ourfinisacked) { 2870 /* 2871 * If we can't receive any more 2872 * data, then closing user can proceed. 2873 * Starting the timer is contrary to the 2874 * specification, but if we don't get a FIN 2875 * we'll hang forever. 2876 * 2877 * XXXjl: 2878 * we should release the tp also, and use a 2879 * compressed state. 2880 */ 2881 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2882 soisdisconnected(so); 2883 tcp_timer_activate(tp, TT_2MSL, 2884 (tcp_fast_finwait2_recycle ? 2885 tcp_finwait2_timeout : 2886 TP_MAXIDLE(tp))); 2887 } 2888 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2889 } 2890 break; 2891 2892 /* 2893 * In CLOSING STATE in addition to the processing for 2894 * the ESTABLISHED state if the ACK acknowledges our FIN 2895 * then enter the TIME-WAIT state, otherwise ignore 2896 * the segment. 2897 */ 2898 case TCPS_CLOSING: 2899 if (ourfinisacked) { 2900 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2901 tcp_twstart(tp); 2902 INP_INFO_RUNLOCK(&V_tcbinfo); 2903 m_freem(m); 2904 return; 2905 } 2906 break; 2907 2908 /* 2909 * In LAST_ACK, we may still be waiting for data to drain 2910 * and/or to be acked, as well as for the ack of our FIN. 2911 * If our FIN is now acknowledged, delete the TCB, 2912 * enter the closed state and return. 2913 */ 2914 case TCPS_LAST_ACK: 2915 if (ourfinisacked) { 2916 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2917 tp = tcp_close(tp); 2918 goto drop; 2919 } 2920 break; 2921 } 2922 } 2923 2924 step6: 2925 INP_WLOCK_ASSERT(tp->t_inpcb); 2926 2927 /* 2928 * Update window information. 2929 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2930 */ 2931 if ((thflags & TH_ACK) && 2932 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2933 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2934 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2935 /* keep track of pure window updates */ 2936 if (tlen == 0 && 2937 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2938 TCPSTAT_INC(tcps_rcvwinupd); 2939 tp->snd_wnd = tiwin; 2940 tp->snd_wl1 = th->th_seq; 2941 tp->snd_wl2 = th->th_ack; 2942 if (tp->snd_wnd > tp->max_sndwnd) 2943 tp->max_sndwnd = tp->snd_wnd; 2944 needoutput = 1; 2945 } 2946 2947 /* 2948 * Process segments with URG. 2949 */ 2950 if ((thflags & TH_URG) && th->th_urp && 2951 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2952 /* 2953 * This is a kludge, but if we receive and accept 2954 * random urgent pointers, we'll crash in 2955 * soreceive. It's hard to imagine someone 2956 * actually wanting to send this much urgent data. 2957 */ 2958 SOCKBUF_LOCK(&so->so_rcv); 2959 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2960 th->th_urp = 0; /* XXX */ 2961 thflags &= ~TH_URG; /* XXX */ 2962 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2963 goto dodata; /* XXX */ 2964 } 2965 /* 2966 * If this segment advances the known urgent pointer, 2967 * then mark the data stream. This should not happen 2968 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2969 * a FIN has been received from the remote side. 2970 * In these states we ignore the URG. 2971 * 2972 * According to RFC961 (Assigned Protocols), 2973 * the urgent pointer points to the last octet 2974 * of urgent data. We continue, however, 2975 * to consider it to indicate the first octet 2976 * of data past the urgent section as the original 2977 * spec states (in one of two places). 2978 */ 2979 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2980 tp->rcv_up = th->th_seq + th->th_urp; 2981 so->so_oobmark = sbavail(&so->so_rcv) + 2982 (tp->rcv_up - tp->rcv_nxt) - 1; 2983 if (so->so_oobmark == 0) 2984 so->so_rcv.sb_state |= SBS_RCVATMARK; 2985 sohasoutofband(so); 2986 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2987 } 2988 SOCKBUF_UNLOCK(&so->so_rcv); 2989 /* 2990 * Remove out of band data so doesn't get presented to user. 2991 * This can happen independent of advancing the URG pointer, 2992 * but if two URG's are pending at once, some out-of-band 2993 * data may creep in... ick. 2994 */ 2995 if (th->th_urp <= (uint32_t)tlen && 2996 !(so->so_options & SO_OOBINLINE)) { 2997 /* hdr drop is delayed */ 2998 tcp_pulloutofband(so, th, m, drop_hdrlen); 2999 } 3000 } else { 3001 /* 3002 * If no out of band data is expected, 3003 * pull receive urgent pointer along 3004 * with the receive window. 3005 */ 3006 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3007 tp->rcv_up = tp->rcv_nxt; 3008 } 3009 dodata: /* XXX */ 3010 INP_WLOCK_ASSERT(tp->t_inpcb); 3011 3012 /* 3013 * Process the segment text, merging it into the TCP sequencing queue, 3014 * and arranging for acknowledgment of receipt if necessary. 3015 * This process logically involves adjusting tp->rcv_wnd as data 3016 * is presented to the user (this happens in tcp_usrreq.c, 3017 * case PRU_RCVD). If a FIN has already been received on this 3018 * connection then we just ignore the text. 3019 */ 3020 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3021 IS_FASTOPEN(tp->t_flags)); 3022 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3023 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3024 tcp_seq save_start = th->th_seq; 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 LIST_EMPTY(&tp->t_segq) && 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 thflags = tcp_reass(tp, th, &tlen, m); 3065 tp->t_flags |= TF_ACKNOW; 3066 } 3067 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3068 tcp_update_sack_list(tp, save_start, save_start + tlen); 3069 #if 0 3070 /* 3071 * Note the amount of data that peer has sent into 3072 * our window, in order to estimate the sender's 3073 * buffer size. 3074 * XXX: Unused. 3075 */ 3076 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3077 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3078 else 3079 len = so->so_rcv.sb_hiwat; 3080 #endif 3081 } else { 3082 m_freem(m); 3083 thflags &= ~TH_FIN; 3084 } 3085 3086 /* 3087 * If FIN is received ACK the FIN and let the user know 3088 * that the connection is closing. 3089 */ 3090 if (thflags & TH_FIN) { 3091 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3092 socantrcvmore(so); 3093 /* 3094 * If connection is half-synchronized 3095 * (ie NEEDSYN flag on) then delay ACK, 3096 * so it may be piggybacked when SYN is sent. 3097 * Otherwise, since we received a FIN then no 3098 * more input can be expected, send ACK now. 3099 */ 3100 if (tp->t_flags & TF_NEEDSYN) 3101 tp->t_flags |= TF_DELACK; 3102 else 3103 tp->t_flags |= TF_ACKNOW; 3104 tp->rcv_nxt++; 3105 } 3106 switch (tp->t_state) { 3107 3108 /* 3109 * In SYN_RECEIVED and ESTABLISHED STATES 3110 * enter the CLOSE_WAIT state. 3111 */ 3112 case TCPS_SYN_RECEIVED: 3113 tp->t_starttime = ticks; 3114 /* FALLTHROUGH */ 3115 case TCPS_ESTABLISHED: 3116 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3117 break; 3118 3119 /* 3120 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3121 * enter the CLOSING state. 3122 */ 3123 case TCPS_FIN_WAIT_1: 3124 tcp_state_change(tp, TCPS_CLOSING); 3125 break; 3126 3127 /* 3128 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3129 * starting the time-wait timer, turning off the other 3130 * standard timers. 3131 */ 3132 case TCPS_FIN_WAIT_2: 3133 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3134 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3135 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3136 ti_locked)); 3137 3138 tcp_twstart(tp); 3139 INP_INFO_RUNLOCK(&V_tcbinfo); 3140 return; 3141 } 3142 } 3143 if (ti_locked == TI_RLOCKED) 3144 INP_INFO_RUNLOCK(&V_tcbinfo); 3145 ti_locked = TI_UNLOCKED; 3146 3147 #ifdef TCPDEBUG 3148 if (so->so_options & SO_DEBUG) 3149 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3150 &tcp_savetcp, 0); 3151 #endif 3152 TCP_PROBE3(debug__input, tp, th, m); 3153 3154 /* 3155 * Return any desired output. 3156 */ 3157 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3158 (void) tp->t_fb->tfb_tcp_output(tp); 3159 3160 check_delack: 3161 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3162 __func__, ti_locked)); 3163 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3164 INP_WLOCK_ASSERT(tp->t_inpcb); 3165 3166 if (tp->t_flags & TF_DELACK) { 3167 tp->t_flags &= ~TF_DELACK; 3168 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3169 } 3170 INP_WUNLOCK(tp->t_inpcb); 3171 return; 3172 3173 dropafterack: 3174 /* 3175 * Generate an ACK dropping incoming segment if it occupies 3176 * sequence space, where the ACK reflects our state. 3177 * 3178 * We can now skip the test for the RST flag since all 3179 * paths to this code happen after packets containing 3180 * RST have been dropped. 3181 * 3182 * In the SYN-RECEIVED state, don't send an ACK unless the 3183 * segment we received passes the SYN-RECEIVED ACK test. 3184 * If it fails send a RST. This breaks the loop in the 3185 * "LAND" DoS attack, and also prevents an ACK storm 3186 * between two listening ports that have been sent forged 3187 * SYN segments, each with the source address of the other. 3188 */ 3189 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3190 (SEQ_GT(tp->snd_una, th->th_ack) || 3191 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3192 rstreason = BANDLIM_RST_OPENPORT; 3193 goto dropwithreset; 3194 } 3195 #ifdef TCPDEBUG 3196 if (so->so_options & SO_DEBUG) 3197 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3198 &tcp_savetcp, 0); 3199 #endif 3200 TCP_PROBE3(debug__input, tp, th, m); 3201 if (ti_locked == TI_RLOCKED) 3202 INP_INFO_RUNLOCK(&V_tcbinfo); 3203 ti_locked = TI_UNLOCKED; 3204 3205 tp->t_flags |= TF_ACKNOW; 3206 (void) tp->t_fb->tfb_tcp_output(tp); 3207 INP_WUNLOCK(tp->t_inpcb); 3208 m_freem(m); 3209 return; 3210 3211 dropwithreset: 3212 if (ti_locked == TI_RLOCKED) 3213 INP_INFO_RUNLOCK(&V_tcbinfo); 3214 ti_locked = TI_UNLOCKED; 3215 3216 if (tp != NULL) { 3217 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3218 INP_WUNLOCK(tp->t_inpcb); 3219 } else 3220 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3221 return; 3222 3223 drop: 3224 if (ti_locked == TI_RLOCKED) { 3225 INP_INFO_RUNLOCK(&V_tcbinfo); 3226 ti_locked = TI_UNLOCKED; 3227 } 3228 #ifdef INVARIANTS 3229 else 3230 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3231 #endif 3232 3233 /* 3234 * Drop space held by incoming segment and return. 3235 */ 3236 #ifdef TCPDEBUG 3237 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3238 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3239 &tcp_savetcp, 0); 3240 #endif 3241 TCP_PROBE3(debug__input, tp, th, m); 3242 if (tp != NULL) 3243 INP_WUNLOCK(tp->t_inpcb); 3244 m_freem(m); 3245 } 3246 3247 /* 3248 * Issue RST and make ACK acceptable to originator of segment. 3249 * The mbuf must still include the original packet header. 3250 * tp may be NULL. 3251 */ 3252 void 3253 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3254 int tlen, int rstreason) 3255 { 3256 #ifdef INET 3257 struct ip *ip; 3258 #endif 3259 #ifdef INET6 3260 struct ip6_hdr *ip6; 3261 #endif 3262 3263 if (tp != NULL) { 3264 INP_WLOCK_ASSERT(tp->t_inpcb); 3265 } 3266 3267 /* Don't bother if destination was broadcast/multicast. */ 3268 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3269 goto drop; 3270 #ifdef INET6 3271 if (mtod(m, struct ip *)->ip_v == 6) { 3272 ip6 = mtod(m, struct ip6_hdr *); 3273 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3274 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3275 goto drop; 3276 /* IPv6 anycast check is done at tcp6_input() */ 3277 } 3278 #endif 3279 #if defined(INET) && defined(INET6) 3280 else 3281 #endif 3282 #ifdef INET 3283 { 3284 ip = mtod(m, struct ip *); 3285 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3286 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3287 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3288 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3289 goto drop; 3290 } 3291 #endif 3292 3293 /* Perform bandwidth limiting. */ 3294 if (badport_bandlim(rstreason) < 0) 3295 goto drop; 3296 3297 /* tcp_respond consumes the mbuf chain. */ 3298 if (th->th_flags & TH_ACK) { 3299 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3300 th->th_ack, TH_RST); 3301 } else { 3302 if (th->th_flags & TH_SYN) 3303 tlen++; 3304 if (th->th_flags & TH_FIN) 3305 tlen++; 3306 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3307 (tcp_seq)0, TH_RST|TH_ACK); 3308 } 3309 return; 3310 drop: 3311 m_freem(m); 3312 } 3313 3314 /* 3315 * Parse TCP options and place in tcpopt. 3316 */ 3317 void 3318 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3319 { 3320 int opt, optlen; 3321 3322 to->to_flags = 0; 3323 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3324 opt = cp[0]; 3325 if (opt == TCPOPT_EOL) 3326 break; 3327 if (opt == TCPOPT_NOP) 3328 optlen = 1; 3329 else { 3330 if (cnt < 2) 3331 break; 3332 optlen = cp[1]; 3333 if (optlen < 2 || optlen > cnt) 3334 break; 3335 } 3336 switch (opt) { 3337 case TCPOPT_MAXSEG: 3338 if (optlen != TCPOLEN_MAXSEG) 3339 continue; 3340 if (!(flags & TO_SYN)) 3341 continue; 3342 to->to_flags |= TOF_MSS; 3343 bcopy((char *)cp + 2, 3344 (char *)&to->to_mss, sizeof(to->to_mss)); 3345 to->to_mss = ntohs(to->to_mss); 3346 break; 3347 case TCPOPT_WINDOW: 3348 if (optlen != TCPOLEN_WINDOW) 3349 continue; 3350 if (!(flags & TO_SYN)) 3351 continue; 3352 to->to_flags |= TOF_SCALE; 3353 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3354 break; 3355 case TCPOPT_TIMESTAMP: 3356 if (optlen != TCPOLEN_TIMESTAMP) 3357 continue; 3358 to->to_flags |= TOF_TS; 3359 bcopy((char *)cp + 2, 3360 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3361 to->to_tsval = ntohl(to->to_tsval); 3362 bcopy((char *)cp + 6, 3363 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3364 to->to_tsecr = ntohl(to->to_tsecr); 3365 break; 3366 case TCPOPT_SIGNATURE: 3367 /* 3368 * In order to reply to a host which has set the 3369 * TCP_SIGNATURE option in its initial SYN, we have 3370 * to record the fact that the option was observed 3371 * here for the syncache code to perform the correct 3372 * response. 3373 */ 3374 if (optlen != TCPOLEN_SIGNATURE) 3375 continue; 3376 to->to_flags |= TOF_SIGNATURE; 3377 to->to_signature = cp + 2; 3378 break; 3379 case TCPOPT_SACK_PERMITTED: 3380 if (optlen != TCPOLEN_SACK_PERMITTED) 3381 continue; 3382 if (!(flags & TO_SYN)) 3383 continue; 3384 if (!V_tcp_do_sack) 3385 continue; 3386 to->to_flags |= TOF_SACKPERM; 3387 break; 3388 case TCPOPT_SACK: 3389 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3390 continue; 3391 if (flags & TO_SYN) 3392 continue; 3393 to->to_flags |= TOF_SACK; 3394 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3395 to->to_sacks = cp + 2; 3396 TCPSTAT_INC(tcps_sack_rcv_blocks); 3397 break; 3398 case TCPOPT_FAST_OPEN: 3399 /* 3400 * Cookie length validation is performed by the 3401 * server side cookie checking code or the client 3402 * side cookie cache update code. 3403 */ 3404 if (!(flags & TO_SYN)) 3405 continue; 3406 if (!V_tcp_fastopen_client_enable && 3407 !V_tcp_fastopen_server_enable) 3408 continue; 3409 to->to_flags |= TOF_FASTOPEN; 3410 to->to_tfo_len = optlen - 2; 3411 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3412 break; 3413 default: 3414 continue; 3415 } 3416 } 3417 } 3418 3419 /* 3420 * Pull out of band byte out of a segment so 3421 * it doesn't appear in the user's data queue. 3422 * It is still reflected in the segment length for 3423 * sequencing purposes. 3424 */ 3425 void 3426 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3427 int off) 3428 { 3429 int cnt = off + th->th_urp - 1; 3430 3431 while (cnt >= 0) { 3432 if (m->m_len > cnt) { 3433 char *cp = mtod(m, caddr_t) + cnt; 3434 struct tcpcb *tp = sototcpcb(so); 3435 3436 INP_WLOCK_ASSERT(tp->t_inpcb); 3437 3438 tp->t_iobc = *cp; 3439 tp->t_oobflags |= TCPOOB_HAVEDATA; 3440 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3441 m->m_len--; 3442 if (m->m_flags & M_PKTHDR) 3443 m->m_pkthdr.len--; 3444 return; 3445 } 3446 cnt -= m->m_len; 3447 m = m->m_next; 3448 if (m == NULL) 3449 break; 3450 } 3451 panic("tcp_pulloutofband"); 3452 } 3453 3454 /* 3455 * Collect new round-trip time estimate 3456 * and update averages and current timeout. 3457 */ 3458 void 3459 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3460 { 3461 int delta; 3462 3463 INP_WLOCK_ASSERT(tp->t_inpcb); 3464 3465 TCPSTAT_INC(tcps_rttupdated); 3466 tp->t_rttupdated++; 3467 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3468 /* 3469 * srtt is stored as fixed point with 5 bits after the 3470 * binary point (i.e., scaled by 8). The following magic 3471 * is equivalent to the smoothing algorithm in rfc793 with 3472 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3473 * point). Adjust rtt to origin 0. 3474 */ 3475 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3476 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3477 3478 if ((tp->t_srtt += delta) <= 0) 3479 tp->t_srtt = 1; 3480 3481 /* 3482 * We accumulate a smoothed rtt variance (actually, a 3483 * smoothed mean difference), then set the retransmit 3484 * timer to smoothed rtt + 4 times the smoothed variance. 3485 * rttvar is stored as fixed point with 4 bits after the 3486 * binary point (scaled by 16). The following is 3487 * equivalent to rfc793 smoothing with an alpha of .75 3488 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3489 * rfc793's wired-in beta. 3490 */ 3491 if (delta < 0) 3492 delta = -delta; 3493 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3494 if ((tp->t_rttvar += delta) <= 0) 3495 tp->t_rttvar = 1; 3496 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3497 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3498 } else { 3499 /* 3500 * No rtt measurement yet - use the unsmoothed rtt. 3501 * Set the variance to half the rtt (so our first 3502 * retransmit happens at 3*rtt). 3503 */ 3504 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3505 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3506 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3507 } 3508 tp->t_rtttime = 0; 3509 tp->t_rxtshift = 0; 3510 3511 /* 3512 * the retransmit should happen at rtt + 4 * rttvar. 3513 * Because of the way we do the smoothing, srtt and rttvar 3514 * will each average +1/2 tick of bias. When we compute 3515 * the retransmit timer, we want 1/2 tick of rounding and 3516 * 1 extra tick because of +-1/2 tick uncertainty in the 3517 * firing of the timer. The bias will give us exactly the 3518 * 1.5 tick we need. But, because the bias is 3519 * statistical, we have to test that we don't drop below 3520 * the minimum feasible timer (which is 2 ticks). 3521 */ 3522 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3523 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3524 3525 /* 3526 * We received an ack for a packet that wasn't retransmitted; 3527 * it is probably safe to discard any error indications we've 3528 * received recently. This isn't quite right, but close enough 3529 * for now (a route might have failed after we sent a segment, 3530 * and the return path might not be symmetrical). 3531 */ 3532 tp->t_softerror = 0; 3533 } 3534 3535 /* 3536 * Determine a reasonable value for maxseg size. 3537 * If the route is known, check route for mtu. 3538 * If none, use an mss that can be handled on the outgoing interface 3539 * without forcing IP to fragment. If no route is found, route has no mtu, 3540 * or the destination isn't local, use a default, hopefully conservative 3541 * size (usually 512 or the default IP max size, but no more than the mtu 3542 * of the interface), as we can't discover anything about intervening 3543 * gateways or networks. We also initialize the congestion/slow start 3544 * window to be a single segment if the destination isn't local. 3545 * While looking at the routing entry, we also initialize other path-dependent 3546 * parameters from pre-set or cached values in the routing entry. 3547 * 3548 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3549 * IP options, e.g. IPSEC data, since length of this data may vary, and 3550 * thus it is calculated for every segment separately in tcp_output(). 3551 * 3552 * NOTE that this routine is only called when we process an incoming 3553 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3554 * settings are handled in tcp_mssopt(). 3555 */ 3556 void 3557 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3558 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3559 { 3560 int mss = 0; 3561 uint32_t maxmtu = 0; 3562 struct inpcb *inp = tp->t_inpcb; 3563 struct hc_metrics_lite metrics; 3564 #ifdef INET6 3565 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3566 size_t min_protoh = isipv6 ? 3567 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3568 sizeof (struct tcpiphdr); 3569 #else 3570 const size_t min_protoh = sizeof(struct tcpiphdr); 3571 #endif 3572 3573 INP_WLOCK_ASSERT(tp->t_inpcb); 3574 3575 if (mtuoffer != -1) { 3576 KASSERT(offer == -1, ("%s: conflict", __func__)); 3577 offer = mtuoffer - min_protoh; 3578 } 3579 3580 /* Initialize. */ 3581 #ifdef INET6 3582 if (isipv6) { 3583 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3584 tp->t_maxseg = V_tcp_v6mssdflt; 3585 } 3586 #endif 3587 #if defined(INET) && defined(INET6) 3588 else 3589 #endif 3590 #ifdef INET 3591 { 3592 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3593 tp->t_maxseg = V_tcp_mssdflt; 3594 } 3595 #endif 3596 3597 /* 3598 * No route to sender, stay with default mss and return. 3599 */ 3600 if (maxmtu == 0) { 3601 /* 3602 * In case we return early we need to initialize metrics 3603 * to a defined state as tcp_hc_get() would do for us 3604 * if there was no cache hit. 3605 */ 3606 if (metricptr != NULL) 3607 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3608 return; 3609 } 3610 3611 /* What have we got? */ 3612 switch (offer) { 3613 case 0: 3614 /* 3615 * Offer == 0 means that there was no MSS on the SYN 3616 * segment, in this case we use tcp_mssdflt as 3617 * already assigned to t_maxseg above. 3618 */ 3619 offer = tp->t_maxseg; 3620 break; 3621 3622 case -1: 3623 /* 3624 * Offer == -1 means that we didn't receive SYN yet. 3625 */ 3626 /* FALLTHROUGH */ 3627 3628 default: 3629 /* 3630 * Prevent DoS attack with too small MSS. Round up 3631 * to at least minmss. 3632 */ 3633 offer = max(offer, V_tcp_minmss); 3634 } 3635 3636 /* 3637 * rmx information is now retrieved from tcp_hostcache. 3638 */ 3639 tcp_hc_get(&inp->inp_inc, &metrics); 3640 if (metricptr != NULL) 3641 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3642 3643 /* 3644 * If there's a discovered mtu in tcp hostcache, use it. 3645 * Else, use the link mtu. 3646 */ 3647 if (metrics.rmx_mtu) 3648 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3649 else { 3650 #ifdef INET6 3651 if (isipv6) { 3652 mss = maxmtu - min_protoh; 3653 if (!V_path_mtu_discovery && 3654 !in6_localaddr(&inp->in6p_faddr)) 3655 mss = min(mss, V_tcp_v6mssdflt); 3656 } 3657 #endif 3658 #if defined(INET) && defined(INET6) 3659 else 3660 #endif 3661 #ifdef INET 3662 { 3663 mss = maxmtu - min_protoh; 3664 if (!V_path_mtu_discovery && 3665 !in_localaddr(inp->inp_faddr)) 3666 mss = min(mss, V_tcp_mssdflt); 3667 } 3668 #endif 3669 /* 3670 * XXX - The above conditional (mss = maxmtu - min_protoh) 3671 * probably violates the TCP spec. 3672 * The problem is that, since we don't know the 3673 * other end's MSS, we are supposed to use a conservative 3674 * default. But, if we do that, then MTU discovery will 3675 * never actually take place, because the conservative 3676 * default is much less than the MTUs typically seen 3677 * on the Internet today. For the moment, we'll sweep 3678 * this under the carpet. 3679 * 3680 * The conservative default might not actually be a problem 3681 * if the only case this occurs is when sending an initial 3682 * SYN with options and data to a host we've never talked 3683 * to before. Then, they will reply with an MSS value which 3684 * will get recorded and the new parameters should get 3685 * recomputed. For Further Study. 3686 */ 3687 } 3688 mss = min(mss, offer); 3689 3690 /* 3691 * Sanity check: make sure that maxseg will be large 3692 * enough to allow some data on segments even if the 3693 * all the option space is used (40bytes). Otherwise 3694 * funny things may happen in tcp_output. 3695 * 3696 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3697 */ 3698 mss = max(mss, 64); 3699 3700 tp->t_maxseg = mss; 3701 } 3702 3703 void 3704 tcp_mss(struct tcpcb *tp, int offer) 3705 { 3706 int mss; 3707 uint32_t bufsize; 3708 struct inpcb *inp; 3709 struct socket *so; 3710 struct hc_metrics_lite metrics; 3711 struct tcp_ifcap cap; 3712 3713 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3714 3715 bzero(&cap, sizeof(cap)); 3716 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3717 3718 mss = tp->t_maxseg; 3719 inp = tp->t_inpcb; 3720 3721 /* 3722 * If there's a pipesize, change the socket buffer to that size, 3723 * don't change if sb_hiwat is different than default (then it 3724 * has been changed on purpose with setsockopt). 3725 * Make the socket buffers an integral number of mss units; 3726 * if the mss is larger than the socket buffer, decrease the mss. 3727 */ 3728 so = inp->inp_socket; 3729 SOCKBUF_LOCK(&so->so_snd); 3730 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3731 bufsize = metrics.rmx_sendpipe; 3732 else 3733 bufsize = so->so_snd.sb_hiwat; 3734 if (bufsize < mss) 3735 mss = bufsize; 3736 else { 3737 bufsize = roundup(bufsize, mss); 3738 if (bufsize > sb_max) 3739 bufsize = sb_max; 3740 if (bufsize > so->so_snd.sb_hiwat) 3741 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3742 } 3743 SOCKBUF_UNLOCK(&so->so_snd); 3744 /* 3745 * Sanity check: make sure that maxseg will be large 3746 * enough to allow some data on segments even if the 3747 * all the option space is used (40bytes). Otherwise 3748 * funny things may happen in tcp_output. 3749 * 3750 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3751 */ 3752 tp->t_maxseg = max(mss, 64); 3753 3754 SOCKBUF_LOCK(&so->so_rcv); 3755 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3756 bufsize = metrics.rmx_recvpipe; 3757 else 3758 bufsize = so->so_rcv.sb_hiwat; 3759 if (bufsize > mss) { 3760 bufsize = roundup(bufsize, mss); 3761 if (bufsize > sb_max) 3762 bufsize = sb_max; 3763 if (bufsize > so->so_rcv.sb_hiwat) 3764 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3765 } 3766 SOCKBUF_UNLOCK(&so->so_rcv); 3767 3768 /* Check the interface for TSO capabilities. */ 3769 if (cap.ifcap & CSUM_TSO) { 3770 tp->t_flags |= TF_TSO; 3771 tp->t_tsomax = cap.tsomax; 3772 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3773 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3774 } 3775 } 3776 3777 /* 3778 * Determine the MSS option to send on an outgoing SYN. 3779 */ 3780 int 3781 tcp_mssopt(struct in_conninfo *inc) 3782 { 3783 int mss = 0; 3784 uint32_t thcmtu = 0; 3785 uint32_t maxmtu = 0; 3786 size_t min_protoh; 3787 3788 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3789 3790 #ifdef INET6 3791 if (inc->inc_flags & INC_ISIPV6) { 3792 mss = V_tcp_v6mssdflt; 3793 maxmtu = tcp_maxmtu6(inc, NULL); 3794 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3795 } 3796 #endif 3797 #if defined(INET) && defined(INET6) 3798 else 3799 #endif 3800 #ifdef INET 3801 { 3802 mss = V_tcp_mssdflt; 3803 maxmtu = tcp_maxmtu(inc, NULL); 3804 min_protoh = sizeof(struct tcpiphdr); 3805 } 3806 #endif 3807 #if defined(INET6) || defined(INET) 3808 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3809 #endif 3810 3811 if (maxmtu && thcmtu) 3812 mss = min(maxmtu, thcmtu) - min_protoh; 3813 else if (maxmtu || thcmtu) 3814 mss = max(maxmtu, thcmtu) - min_protoh; 3815 3816 return (mss); 3817 } 3818 3819 3820 /* 3821 * On a partial ack arrives, force the retransmission of the 3822 * next unacknowledged segment. Do not clear tp->t_dupacks. 3823 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3824 * be started again. 3825 */ 3826 void 3827 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3828 { 3829 tcp_seq onxt = tp->snd_nxt; 3830 uint32_t ocwnd = tp->snd_cwnd; 3831 u_int maxseg = tcp_maxseg(tp); 3832 3833 INP_WLOCK_ASSERT(tp->t_inpcb); 3834 3835 tcp_timer_activate(tp, TT_REXMT, 0); 3836 tp->t_rtttime = 0; 3837 tp->snd_nxt = th->th_ack; 3838 /* 3839 * Set snd_cwnd to one segment beyond acknowledged offset. 3840 * (tp->snd_una has not yet been updated when this function is called.) 3841 */ 3842 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3843 tp->t_flags |= TF_ACKNOW; 3844 (void) tp->t_fb->tfb_tcp_output(tp); 3845 tp->snd_cwnd = ocwnd; 3846 if (SEQ_GT(onxt, tp->snd_nxt)) 3847 tp->snd_nxt = onxt; 3848 /* 3849 * Partial window deflation. Relies on fact that tp->snd_una 3850 * not updated yet. 3851 */ 3852 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3853 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3854 else 3855 tp->snd_cwnd = 0; 3856 tp->snd_cwnd += maxseg; 3857 } 3858 3859 int 3860 tcp_compute_pipe(struct tcpcb *tp) 3861 { 3862 return (tp->snd_max - tp->snd_una + 3863 tp->sackhint.sack_bytes_rexmit - 3864 tp->sackhint.sacked_bytes); 3865 } 3866