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_seq.h> 106 #include <netinet/tcp_timer.h> 107 #include <netinet/tcp_var.h> 108 #include <netinet6/tcp6_var.h> 109 #include <netinet/tcpip.h> 110 #include <netinet/cc/cc.h> 111 #include <netinet/tcp_fastopen.h> 112 #ifdef TCPPCAP 113 #include <netinet/tcp_pcap.h> 114 #endif 115 #include <netinet/tcp_syncache.h> 116 #ifdef TCPDEBUG 117 #include <netinet/tcp_debug.h> 118 #endif /* TCPDEBUG */ 119 #ifdef TCP_OFFLOAD 120 #include <netinet/tcp_offload.h> 121 #endif 122 123 #include <netipsec/ipsec_support.h> 124 125 #include <machine/in_cksum.h> 126 127 #include <security/mac/mac_framework.h> 128 129 const int tcprexmtthresh = 3; 130 131 int tcp_log_in_vain = 0; 132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 133 &tcp_log_in_vain, 0, 134 "Log all incoming TCP segments to closed ports"); 135 136 VNET_DEFINE(int, blackhole) = 0; 137 #define V_blackhole VNET(blackhole) 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(blackhole), 0, 140 "Do not send RST on segments to closed ports"); 141 142 VNET_DEFINE(int, tcp_delack_enabled) = 1; 143 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 144 &VNET_NAME(tcp_delack_enabled), 0, 145 "Delay ACK to try and piggyback it onto a data packet"); 146 147 VNET_DEFINE(int, drop_synfin) = 0; 148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 149 &VNET_NAME(drop_synfin), 0, 150 "Drop TCP packets with SYN+FIN set"); 151 152 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 154 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 155 "Use calculated pipe/in-flight bytes per RFC 6675"); 156 157 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 159 &VNET_NAME(tcp_do_rfc3042), 0, 160 "Enable RFC 3042 (Limited Transmit)"); 161 162 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_rfc3390), 0, 165 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 166 167 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 169 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 170 "Slow-start flight size (initial congestion window) in number of segments"); 171 172 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_do_rfc3465), 0, 175 "Enable RFC 3465 (Appropriate Byte Counting)"); 176 177 VNET_DEFINE(int, tcp_abc_l_var) = 2; 178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 179 &VNET_NAME(tcp_abc_l_var), 2, 180 "Cap the max cwnd increment during slow-start to this number of segments"); 181 182 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 183 184 VNET_DEFINE(int, tcp_do_ecn) = 2; 185 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 186 &VNET_NAME(tcp_do_ecn), 0, 187 "TCP ECN support"); 188 189 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 190 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 191 &VNET_NAME(tcp_ecn_maxretries), 0, 192 "Max retries before giving up on ECN"); 193 194 VNET_DEFINE(int, tcp_insecure_syn) = 0; 195 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 196 &VNET_NAME(tcp_insecure_syn), 0, 197 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 198 199 VNET_DEFINE(int, tcp_insecure_rst) = 0; 200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 201 &VNET_NAME(tcp_insecure_rst), 0, 202 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 203 204 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 205 #define V_tcp_recvspace VNET(tcp_recvspace) 206 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 207 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 208 209 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 211 &VNET_NAME(tcp_do_autorcvbuf), 0, 212 "Enable automatic receive buffer sizing"); 213 214 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 216 &VNET_NAME(tcp_autorcvbuf_inc), 0, 217 "Incrementor step size of automatic receive buffer"); 218 219 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 221 &VNET_NAME(tcp_autorcvbuf_max), 0, 222 "Max size of automatic receive buffer"); 223 224 VNET_DEFINE(struct inpcbhead, tcb); 225 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 226 VNET_DEFINE(struct inpcbinfo, tcbinfo); 227 228 /* 229 * TCP statistics are stored in an array of counter(9)s, which size matches 230 * size of struct tcpstat. TCP running connection count is a regular array. 231 */ 232 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 233 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 234 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 235 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 236 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 237 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 238 "TCP connection counts by TCP state"); 239 240 static void 241 tcp_vnet_init(const void *unused) 242 { 243 244 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 245 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 246 } 247 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 248 tcp_vnet_init, NULL); 249 250 #ifdef VIMAGE 251 static void 252 tcp_vnet_uninit(const void *unused) 253 { 254 255 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 256 VNET_PCPUSTAT_FREE(tcpstat); 257 } 258 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 259 tcp_vnet_uninit, NULL); 260 #endif /* VIMAGE */ 261 262 /* 263 * Kernel module interface for updating tcpstat. The argument is an index 264 * into tcpstat treated as an array. 265 */ 266 void 267 kmod_tcpstat_inc(int statnum) 268 { 269 270 counter_u64_add(VNET(tcpstat)[statnum], 1); 271 } 272 273 #ifdef TCP_HHOOK 274 /* 275 * Wrapper for the TCP established input helper hook. 276 */ 277 void 278 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 279 { 280 struct tcp_hhook_data hhook_data; 281 282 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 283 hhook_data.tp = tp; 284 hhook_data.th = th; 285 hhook_data.to = to; 286 287 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 288 tp->osd); 289 } 290 } 291 #endif 292 293 /* 294 * CC wrapper hook functions 295 */ 296 void 297 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 298 uint16_t type) 299 { 300 INP_WLOCK_ASSERT(tp->t_inpcb); 301 302 tp->ccv->nsegs = nsegs; 303 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 304 if (tp->snd_cwnd <= tp->snd_wnd) 305 tp->ccv->flags |= CCF_CWND_LIMITED; 306 else 307 tp->ccv->flags &= ~CCF_CWND_LIMITED; 308 309 if (type == CC_ACK) { 310 if (tp->snd_cwnd > tp->snd_ssthresh) { 311 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 312 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 313 if (tp->t_bytes_acked >= tp->snd_cwnd) { 314 tp->t_bytes_acked -= tp->snd_cwnd; 315 tp->ccv->flags |= CCF_ABC_SENTAWND; 316 } 317 } else { 318 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 319 tp->t_bytes_acked = 0; 320 } 321 } 322 323 if (CC_ALGO(tp)->ack_received != NULL) { 324 /* XXXLAS: Find a way to live without this */ 325 tp->ccv->curack = th->th_ack; 326 CC_ALGO(tp)->ack_received(tp->ccv, type); 327 } 328 } 329 330 void 331 cc_conn_init(struct tcpcb *tp) 332 { 333 struct hc_metrics_lite metrics; 334 struct inpcb *inp = tp->t_inpcb; 335 u_int maxseg; 336 int rtt; 337 338 INP_WLOCK_ASSERT(tp->t_inpcb); 339 340 tcp_hc_get(&inp->inp_inc, &metrics); 341 maxseg = tcp_maxseg(tp); 342 343 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 344 tp->t_srtt = rtt; 345 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 346 TCPSTAT_INC(tcps_usedrtt); 347 if (metrics.rmx_rttvar) { 348 tp->t_rttvar = metrics.rmx_rttvar; 349 TCPSTAT_INC(tcps_usedrttvar); 350 } else { 351 /* default variation is +- 1 rtt */ 352 tp->t_rttvar = 353 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 354 } 355 TCPT_RANGESET(tp->t_rxtcur, 356 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 357 tp->t_rttmin, TCPTV_REXMTMAX); 358 } 359 if (metrics.rmx_ssthresh) { 360 /* 361 * There's some sort of gateway or interface 362 * buffer limit on the path. Use this to set 363 * the slow start threshold, but set the 364 * threshold to no less than 2*mss. 365 */ 366 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 367 TCPSTAT_INC(tcps_usedssthresh); 368 } 369 370 /* 371 * Set the initial slow-start flight size. 372 * 373 * RFC5681 Section 3.1 specifies the default conservative values. 374 * RFC3390 specifies slightly more aggressive values. 375 * RFC6928 increases it to ten segments. 376 * Support for user specified value for initial flight size. 377 * 378 * If a SYN or SYN/ACK was lost and retransmitted, we have to 379 * reduce the initial CWND to one segment as congestion is likely 380 * requiring us to be cautious. 381 */ 382 if (tp->snd_cwnd == 1) 383 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 384 else if (V_tcp_initcwnd_segments) 385 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 386 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 387 else if (V_tcp_do_rfc3390) 388 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 389 else { 390 /* Per RFC5681 Section 3.1 */ 391 if (maxseg > 2190) 392 tp->snd_cwnd = 2 * maxseg; 393 else if (maxseg > 1095) 394 tp->snd_cwnd = 3 * maxseg; 395 else 396 tp->snd_cwnd = 4 * maxseg; 397 } 398 399 if (CC_ALGO(tp)->conn_init != NULL) 400 CC_ALGO(tp)->conn_init(tp->ccv); 401 } 402 403 void inline 404 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 405 { 406 u_int maxseg; 407 408 INP_WLOCK_ASSERT(tp->t_inpcb); 409 410 switch(type) { 411 case CC_NDUPACK: 412 if (!IN_FASTRECOVERY(tp->t_flags)) { 413 tp->snd_recover = tp->snd_max; 414 if (tp->t_flags & TF_ECN_PERMIT) 415 tp->t_flags |= TF_ECN_SND_CWR; 416 } 417 break; 418 case CC_ECN: 419 if (!IN_CONGRECOVERY(tp->t_flags)) { 420 TCPSTAT_INC(tcps_ecn_rcwnd); 421 tp->snd_recover = tp->snd_max; 422 if (tp->t_flags & TF_ECN_PERMIT) 423 tp->t_flags |= TF_ECN_SND_CWR; 424 } 425 break; 426 case CC_RTO: 427 maxseg = tcp_maxseg(tp); 428 tp->t_dupacks = 0; 429 tp->t_bytes_acked = 0; 430 EXIT_RECOVERY(tp->t_flags); 431 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 432 maxseg) * maxseg; 433 tp->snd_cwnd = maxseg; 434 break; 435 case CC_RTO_ERR: 436 TCPSTAT_INC(tcps_sndrexmitbad); 437 /* RTO was unnecessary, so reset everything. */ 438 tp->snd_cwnd = tp->snd_cwnd_prev; 439 tp->snd_ssthresh = tp->snd_ssthresh_prev; 440 tp->snd_recover = tp->snd_recover_prev; 441 if (tp->t_flags & TF_WASFRECOVERY) 442 ENTER_FASTRECOVERY(tp->t_flags); 443 if (tp->t_flags & TF_WASCRECOVERY) 444 ENTER_CONGRECOVERY(tp->t_flags); 445 tp->snd_nxt = tp->snd_max; 446 tp->t_flags &= ~TF_PREVVALID; 447 tp->t_badrxtwin = 0; 448 break; 449 } 450 451 if (CC_ALGO(tp)->cong_signal != NULL) { 452 if (th != NULL) 453 tp->ccv->curack = th->th_ack; 454 CC_ALGO(tp)->cong_signal(tp->ccv, type); 455 } 456 } 457 458 void inline 459 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 460 { 461 INP_WLOCK_ASSERT(tp->t_inpcb); 462 463 /* XXXLAS: KASSERT that we're in recovery? */ 464 465 if (CC_ALGO(tp)->post_recovery != NULL) { 466 tp->ccv->curack = th->th_ack; 467 CC_ALGO(tp)->post_recovery(tp->ccv); 468 } 469 /* XXXLAS: EXIT_RECOVERY ? */ 470 tp->t_bytes_acked = 0; 471 } 472 473 /* 474 * Indicate whether this ack should be delayed. We can delay the ack if 475 * following conditions are met: 476 * - There is no delayed ack timer in progress. 477 * - Our last ack wasn't a 0-sized window. We never want to delay 478 * the ack that opens up a 0-sized window. 479 * - LRO wasn't used for this segment. We make sure by checking that the 480 * segment size is not larger than the MSS. 481 */ 482 #define DELAY_ACK(tp, tlen) \ 483 ((!tcp_timer_active(tp, TT_DELACK) && \ 484 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 485 (tlen <= tp->t_maxseg) && \ 486 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 487 488 static void inline 489 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 490 { 491 INP_WLOCK_ASSERT(tp->t_inpcb); 492 493 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 494 switch (iptos & IPTOS_ECN_MASK) { 495 case IPTOS_ECN_CE: 496 tp->ccv->flags |= CCF_IPHDR_CE; 497 break; 498 case IPTOS_ECN_ECT0: 499 tp->ccv->flags &= ~CCF_IPHDR_CE; 500 break; 501 case IPTOS_ECN_ECT1: 502 tp->ccv->flags &= ~CCF_IPHDR_CE; 503 break; 504 } 505 506 if (th->th_flags & TH_CWR) 507 tp->ccv->flags |= CCF_TCPHDR_CWR; 508 else 509 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 510 511 if (tp->t_flags & TF_DELACK) 512 tp->ccv->flags |= CCF_DELACK; 513 else 514 tp->ccv->flags &= ~CCF_DELACK; 515 516 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 517 518 if (tp->ccv->flags & CCF_ACKNOW) 519 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 520 } 521 } 522 523 /* 524 * TCP input handling is split into multiple parts: 525 * tcp6_input is a thin wrapper around tcp_input for the extended 526 * ip6_protox[] call format in ip6_input 527 * tcp_input handles primary segment validation, inpcb lookup and 528 * SYN processing on listen sockets 529 * tcp_do_segment processes the ACK and text of the segment for 530 * establishing, established and closing connections 531 */ 532 #ifdef INET6 533 int 534 tcp6_input(struct mbuf **mp, int *offp, int proto) 535 { 536 struct mbuf *m = *mp; 537 struct in6_ifaddr *ia6; 538 struct ip6_hdr *ip6; 539 540 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 541 542 /* 543 * draft-itojun-ipv6-tcp-to-anycast 544 * better place to put this in? 545 */ 546 ip6 = mtod(m, struct ip6_hdr *); 547 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 548 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 549 struct ip6_hdr *ip6; 550 551 ifa_free(&ia6->ia_ifa); 552 ip6 = mtod(m, struct ip6_hdr *); 553 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 554 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 555 return (IPPROTO_DONE); 556 } 557 if (ia6) 558 ifa_free(&ia6->ia_ifa); 559 560 return (tcp_input(mp, offp, proto)); 561 } 562 #endif /* INET6 */ 563 564 int 565 tcp_input(struct mbuf **mp, int *offp, int proto) 566 { 567 struct mbuf *m = *mp; 568 struct tcphdr *th = NULL; 569 struct ip *ip = NULL; 570 struct inpcb *inp = NULL; 571 struct tcpcb *tp = NULL; 572 struct socket *so = NULL; 573 u_char *optp = NULL; 574 int off0; 575 int optlen = 0; 576 #ifdef INET 577 int len; 578 #endif 579 int tlen = 0, off; 580 int drop_hdrlen; 581 int thflags; 582 int rstreason = 0; /* For badport_bandlim accounting purposes */ 583 uint8_t iptos; 584 struct m_tag *fwd_tag = NULL; 585 #ifdef INET6 586 struct ip6_hdr *ip6 = NULL; 587 int isipv6; 588 #else 589 const void *ip6 = NULL; 590 #endif /* INET6 */ 591 struct tcpopt to; /* options in this segment */ 592 char *s = NULL; /* address and port logging */ 593 int ti_locked; 594 #ifdef TCPDEBUG 595 /* 596 * The size of tcp_saveipgen must be the size of the max ip header, 597 * now IPv6. 598 */ 599 u_char tcp_saveipgen[IP6_HDR_LEN]; 600 struct tcphdr tcp_savetcp; 601 short ostate = 0; 602 #endif 603 604 #ifdef INET6 605 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 606 #endif 607 608 off0 = *offp; 609 m = *mp; 610 *mp = NULL; 611 to.to_flags = 0; 612 TCPSTAT_INC(tcps_rcvtotal); 613 614 #ifdef INET6 615 if (isipv6) { 616 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 617 618 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 619 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 620 if (m == NULL) { 621 TCPSTAT_INC(tcps_rcvshort); 622 return (IPPROTO_DONE); 623 } 624 } 625 626 ip6 = mtod(m, struct ip6_hdr *); 627 th = (struct tcphdr *)((caddr_t)ip6 + off0); 628 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 629 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 630 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 631 th->th_sum = m->m_pkthdr.csum_data; 632 else 633 th->th_sum = in6_cksum_pseudo(ip6, tlen, 634 IPPROTO_TCP, m->m_pkthdr.csum_data); 635 th->th_sum ^= 0xffff; 636 } else 637 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 638 if (th->th_sum) { 639 TCPSTAT_INC(tcps_rcvbadsum); 640 goto drop; 641 } 642 643 /* 644 * Be proactive about unspecified IPv6 address in source. 645 * As we use all-zero to indicate unbounded/unconnected pcb, 646 * unspecified IPv6 address can be used to confuse us. 647 * 648 * Note that packets with unspecified IPv6 destination is 649 * already dropped in ip6_input. 650 */ 651 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 652 /* XXX stat */ 653 goto drop; 654 } 655 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 656 } 657 #endif 658 #if defined(INET) && defined(INET6) 659 else 660 #endif 661 #ifdef INET 662 { 663 /* 664 * Get IP and TCP header together in first mbuf. 665 * Note: IP leaves IP header in first mbuf. 666 */ 667 if (off0 > sizeof (struct ip)) { 668 ip_stripoptions(m); 669 off0 = sizeof(struct ip); 670 } 671 if (m->m_len < sizeof (struct tcpiphdr)) { 672 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 673 == NULL) { 674 TCPSTAT_INC(tcps_rcvshort); 675 return (IPPROTO_DONE); 676 } 677 } 678 ip = mtod(m, struct ip *); 679 th = (struct tcphdr *)((caddr_t)ip + off0); 680 tlen = ntohs(ip->ip_len) - off0; 681 682 iptos = ip->ip_tos; 683 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 684 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 685 th->th_sum = m->m_pkthdr.csum_data; 686 else 687 th->th_sum = in_pseudo(ip->ip_src.s_addr, 688 ip->ip_dst.s_addr, 689 htonl(m->m_pkthdr.csum_data + tlen + 690 IPPROTO_TCP)); 691 th->th_sum ^= 0xffff; 692 } else { 693 struct ipovly *ipov = (struct ipovly *)ip; 694 695 /* 696 * Checksum extended TCP header and data. 697 */ 698 len = off0 + tlen; 699 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 700 ipov->ih_len = htons(tlen); 701 th->th_sum = in_cksum(m, len); 702 /* Reset length for SDT probes. */ 703 ip->ip_len = htons(len); 704 /* Reset TOS bits */ 705 ip->ip_tos = iptos; 706 /* Re-initialization for later version check */ 707 ip->ip_v = IPVERSION; 708 ip->ip_hl = off0 >> 2; 709 } 710 711 if (th->th_sum) { 712 TCPSTAT_INC(tcps_rcvbadsum); 713 goto drop; 714 } 715 } 716 #endif /* INET */ 717 718 /* 719 * Check that TCP offset makes sense, 720 * pull out TCP options and adjust length. XXX 721 */ 722 off = th->th_off << 2; 723 if (off < sizeof (struct tcphdr) || off > tlen) { 724 TCPSTAT_INC(tcps_rcvbadoff); 725 goto drop; 726 } 727 tlen -= off; /* tlen is used instead of ti->ti_len */ 728 if (off > sizeof (struct tcphdr)) { 729 #ifdef INET6 730 if (isipv6) { 731 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 732 ip6 = mtod(m, struct ip6_hdr *); 733 th = (struct tcphdr *)((caddr_t)ip6 + off0); 734 } 735 #endif 736 #if defined(INET) && defined(INET6) 737 else 738 #endif 739 #ifdef INET 740 { 741 if (m->m_len < sizeof(struct ip) + off) { 742 if ((m = m_pullup(m, sizeof (struct ip) + off)) 743 == NULL) { 744 TCPSTAT_INC(tcps_rcvshort); 745 return (IPPROTO_DONE); 746 } 747 ip = mtod(m, struct ip *); 748 th = (struct tcphdr *)((caddr_t)ip + off0); 749 } 750 } 751 #endif 752 optlen = off - sizeof (struct tcphdr); 753 optp = (u_char *)(th + 1); 754 } 755 thflags = th->th_flags; 756 757 /* 758 * Convert TCP protocol specific fields to host format. 759 */ 760 tcp_fields_to_host(th); 761 762 /* 763 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 764 */ 765 drop_hdrlen = off0 + off; 766 767 /* 768 * Locate pcb for segment; if we're likely to add or remove a 769 * connection then first acquire pcbinfo lock. There are three cases 770 * where we might discover later we need a write lock despite the 771 * flags: ACKs moving a connection out of the syncache, ACKs for a 772 * connection in TIMEWAIT and SYNs not targeting a listening socket. 773 */ 774 if ((thflags & (TH_FIN | TH_RST)) != 0) { 775 INP_INFO_RLOCK(&V_tcbinfo); 776 ti_locked = TI_RLOCKED; 777 } else 778 ti_locked = TI_UNLOCKED; 779 780 /* 781 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 782 */ 783 if ( 784 #ifdef INET6 785 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 786 #ifdef INET 787 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 788 #endif 789 #endif 790 #if defined(INET) && !defined(INET6) 791 (m->m_flags & M_IP_NEXTHOP) 792 #endif 793 ) 794 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 795 796 findpcb: 797 #ifdef INVARIANTS 798 if (ti_locked == TI_RLOCKED) { 799 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 800 } else { 801 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 802 } 803 #endif 804 #ifdef INET6 805 if (isipv6 && fwd_tag != NULL) { 806 struct sockaddr_in6 *next_hop6; 807 808 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 809 /* 810 * Transparently forwarded. Pretend to be the destination. 811 * Already got one like this? 812 */ 813 inp = in6_pcblookup_mbuf(&V_tcbinfo, 814 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 815 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 816 if (!inp) { 817 /* 818 * It's new. Try to find the ambushing socket. 819 * Because we've rewritten the destination address, 820 * any hardware-generated hash is ignored. 821 */ 822 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 823 th->th_sport, &next_hop6->sin6_addr, 824 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 825 th->th_dport, INPLOOKUP_WILDCARD | 826 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 827 } 828 } else if (isipv6) { 829 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 830 th->th_sport, &ip6->ip6_dst, th->th_dport, 831 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 832 m->m_pkthdr.rcvif, m); 833 } 834 #endif /* INET6 */ 835 #if defined(INET6) && defined(INET) 836 else 837 #endif 838 #ifdef INET 839 if (fwd_tag != NULL) { 840 struct sockaddr_in *next_hop; 841 842 next_hop = (struct sockaddr_in *)(fwd_tag+1); 843 /* 844 * Transparently forwarded. Pretend to be the destination. 845 * already got one like this? 846 */ 847 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 848 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 849 m->m_pkthdr.rcvif, m); 850 if (!inp) { 851 /* 852 * It's new. Try to find the ambushing socket. 853 * Because we've rewritten the destination address, 854 * any hardware-generated hash is ignored. 855 */ 856 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 857 th->th_sport, next_hop->sin_addr, 858 next_hop->sin_port ? ntohs(next_hop->sin_port) : 859 th->th_dport, INPLOOKUP_WILDCARD | 860 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 861 } 862 } else 863 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 864 th->th_sport, ip->ip_dst, th->th_dport, 865 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 866 m->m_pkthdr.rcvif, m); 867 #endif /* INET */ 868 869 /* 870 * If the INPCB does not exist then all data in the incoming 871 * segment is discarded and an appropriate RST is sent back. 872 * XXX MRT Send RST using which routing table? 873 */ 874 if (inp == NULL) { 875 /* 876 * Log communication attempts to ports that are not 877 * in use. 878 */ 879 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 880 tcp_log_in_vain == 2) { 881 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 882 log(LOG_INFO, "%s; %s: Connection attempt " 883 "to closed port\n", s, __func__); 884 } 885 /* 886 * When blackholing do not respond with a RST but 887 * completely ignore the segment and drop it. 888 */ 889 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 890 V_blackhole == 2) 891 goto dropunlock; 892 893 rstreason = BANDLIM_RST_CLOSEDPORT; 894 goto dropwithreset; 895 } 896 INP_WLOCK_ASSERT(inp); 897 /* 898 * While waiting for inp lock during the lookup, another thread 899 * can have dropped the inpcb, in which case we need to loop back 900 * and try to find a new inpcb to deliver to. 901 */ 902 if (inp->inp_flags & INP_DROPPED) { 903 INP_WUNLOCK(inp); 904 inp = NULL; 905 goto findpcb; 906 } 907 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 908 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 909 ((inp->inp_socket == NULL) || 910 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 911 inp->inp_flowid = m->m_pkthdr.flowid; 912 inp->inp_flowtype = M_HASHTYPE_GET(m); 913 } 914 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 915 #ifdef INET6 916 if (isipv6 && IPSEC_ENABLED(ipv6) && 917 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 918 goto dropunlock; 919 } 920 #ifdef INET 921 else 922 #endif 923 #endif /* INET6 */ 924 #ifdef INET 925 if (IPSEC_ENABLED(ipv4) && 926 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 927 goto dropunlock; 928 } 929 #endif /* INET */ 930 #endif /* IPSEC */ 931 932 /* 933 * Check the minimum TTL for socket. 934 */ 935 if (inp->inp_ip_minttl != 0) { 936 #ifdef INET6 937 if (isipv6) { 938 if (inp->inp_ip_minttl > ip6->ip6_hlim) 939 goto dropunlock; 940 } else 941 #endif 942 if (inp->inp_ip_minttl > ip->ip_ttl) 943 goto dropunlock; 944 } 945 946 /* 947 * A previous connection in TIMEWAIT state is supposed to catch stray 948 * or duplicate segments arriving late. If this segment was a 949 * legitimate new connection attempt, the old INPCB gets removed and 950 * we can try again to find a listening socket. 951 * 952 * At this point, due to earlier optimism, we may hold only an inpcb 953 * lock, and not the inpcbinfo write lock. If so, we need to try to 954 * acquire it, or if that fails, acquire a reference on the inpcb, 955 * drop all locks, acquire a global write lock, and then re-acquire 956 * the inpcb lock. We may at that point discover that another thread 957 * has tried to free the inpcb, in which case we need to loop back 958 * and try to find a new inpcb to deliver to. 959 * 960 * XXXRW: It may be time to rethink timewait locking. 961 */ 962 relocked: 963 if (inp->inp_flags & INP_TIMEWAIT) { 964 if (ti_locked == TI_UNLOCKED) { 965 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 966 in_pcbref(inp); 967 INP_WUNLOCK(inp); 968 INP_INFO_RLOCK(&V_tcbinfo); 969 ti_locked = TI_RLOCKED; 970 INP_WLOCK(inp); 971 if (in_pcbrele_wlocked(inp)) { 972 inp = NULL; 973 goto findpcb; 974 } else if (inp->inp_flags & INP_DROPPED) { 975 INP_WUNLOCK(inp); 976 inp = NULL; 977 goto findpcb; 978 } 979 } else 980 ti_locked = TI_RLOCKED; 981 } 982 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 983 984 if (thflags & TH_SYN) 985 tcp_dooptions(&to, optp, optlen, TO_SYN); 986 /* 987 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 988 */ 989 if (tcp_twcheck(inp, &to, th, m, tlen)) 990 goto findpcb; 991 INP_INFO_RUNLOCK(&V_tcbinfo); 992 return (IPPROTO_DONE); 993 } 994 /* 995 * The TCPCB may no longer exist if the connection is winding 996 * down or it is in the CLOSED state. Either way we drop the 997 * segment and send an appropriate response. 998 */ 999 tp = intotcpcb(inp); 1000 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1001 rstreason = BANDLIM_RST_CLOSEDPORT; 1002 goto dropwithreset; 1003 } 1004 1005 #ifdef TCP_OFFLOAD 1006 if (tp->t_flags & TF_TOE) { 1007 tcp_offload_input(tp, m); 1008 m = NULL; /* consumed by the TOE driver */ 1009 goto dropunlock; 1010 } 1011 #endif 1012 1013 /* 1014 * We've identified a valid inpcb, but it could be that we need an 1015 * inpcbinfo write lock but don't hold it. In this case, attempt to 1016 * acquire using the same strategy as the TIMEWAIT case above. If we 1017 * relock, we have to jump back to 'relocked' as the connection might 1018 * now be in TIMEWAIT. 1019 */ 1020 #ifdef INVARIANTS 1021 if ((thflags & (TH_FIN | TH_RST)) != 0) 1022 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1023 #endif 1024 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1025 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1026 !IS_FASTOPEN(tp->t_flags)))) { 1027 if (ti_locked == TI_UNLOCKED) { 1028 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1029 in_pcbref(inp); 1030 INP_WUNLOCK(inp); 1031 INP_INFO_RLOCK(&V_tcbinfo); 1032 ti_locked = TI_RLOCKED; 1033 INP_WLOCK(inp); 1034 if (in_pcbrele_wlocked(inp)) { 1035 inp = NULL; 1036 goto findpcb; 1037 } else if (inp->inp_flags & INP_DROPPED) { 1038 INP_WUNLOCK(inp); 1039 inp = NULL; 1040 goto findpcb; 1041 } 1042 goto relocked; 1043 } else 1044 ti_locked = TI_RLOCKED; 1045 } 1046 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1047 } 1048 1049 #ifdef MAC 1050 INP_WLOCK_ASSERT(inp); 1051 if (mac_inpcb_check_deliver(inp, m)) 1052 goto dropunlock; 1053 #endif 1054 so = inp->inp_socket; 1055 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1056 #ifdef TCPDEBUG 1057 if (so->so_options & SO_DEBUG) { 1058 ostate = tp->t_state; 1059 #ifdef INET6 1060 if (isipv6) { 1061 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1062 } else 1063 #endif 1064 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1065 tcp_savetcp = *th; 1066 } 1067 #endif /* TCPDEBUG */ 1068 /* 1069 * When the socket is accepting connections (the INPCB is in LISTEN 1070 * state) we look into the SYN cache if this is a new connection 1071 * attempt or the completion of a previous one. 1072 */ 1073 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1074 ("%s: so accepting but tp %p not listening", __func__, tp)); 1075 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1076 struct in_conninfo inc; 1077 1078 bzero(&inc, sizeof(inc)); 1079 #ifdef INET6 1080 if (isipv6) { 1081 inc.inc_flags |= INC_ISIPV6; 1082 inc.inc6_faddr = ip6->ip6_src; 1083 inc.inc6_laddr = ip6->ip6_dst; 1084 } else 1085 #endif 1086 { 1087 inc.inc_faddr = ip->ip_src; 1088 inc.inc_laddr = ip->ip_dst; 1089 } 1090 inc.inc_fport = th->th_sport; 1091 inc.inc_lport = th->th_dport; 1092 inc.inc_fibnum = so->so_fibnum; 1093 1094 /* 1095 * Check for an existing connection attempt in syncache if 1096 * the flag is only ACK. A successful lookup creates a new 1097 * socket appended to the listen queue in SYN_RECEIVED state. 1098 */ 1099 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1100 1101 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1102 /* 1103 * Parse the TCP options here because 1104 * syncookies need access to the reflected 1105 * timestamp. 1106 */ 1107 tcp_dooptions(&to, optp, optlen, 0); 1108 /* 1109 * NB: syncache_expand() doesn't unlock 1110 * inp and tcpinfo locks. 1111 */ 1112 rstreason = syncache_expand(&inc, &to, th, &so, m); 1113 if (rstreason < 0) { 1114 /* 1115 * A failing TCP MD5 signature comparison 1116 * must result in the segment being dropped 1117 * and must not produce any response back 1118 * to the sender. 1119 */ 1120 goto dropunlock; 1121 } else if (rstreason == 0) { 1122 /* 1123 * No syncache entry or ACK was not 1124 * for our SYN/ACK. Send a RST. 1125 * NB: syncache did its own logging 1126 * of the failure cause. 1127 */ 1128 rstreason = BANDLIM_RST_OPENPORT; 1129 goto dropwithreset; 1130 } 1131 tfo_socket_result: 1132 if (so == NULL) { 1133 /* 1134 * We completed the 3-way handshake 1135 * but could not allocate a socket 1136 * either due to memory shortage, 1137 * listen queue length limits or 1138 * global socket limits. Send RST 1139 * or wait and have the remote end 1140 * retransmit the ACK for another 1141 * try. 1142 */ 1143 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1144 log(LOG_DEBUG, "%s; %s: Listen socket: " 1145 "Socket allocation failed due to " 1146 "limits or memory shortage, %s\n", 1147 s, __func__, 1148 V_tcp_sc_rst_sock_fail ? 1149 "sending RST" : "try again"); 1150 if (V_tcp_sc_rst_sock_fail) { 1151 rstreason = BANDLIM_UNLIMITED; 1152 goto dropwithreset; 1153 } else 1154 goto dropunlock; 1155 } 1156 /* 1157 * Socket is created in state SYN_RECEIVED. 1158 * Unlock the listen socket, lock the newly 1159 * created socket and update the tp variable. 1160 */ 1161 INP_WUNLOCK(inp); /* listen socket */ 1162 inp = sotoinpcb(so); 1163 /* 1164 * New connection inpcb is already locked by 1165 * syncache_expand(). 1166 */ 1167 INP_WLOCK_ASSERT(inp); 1168 tp = intotcpcb(inp); 1169 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1170 ("%s: ", __func__)); 1171 /* 1172 * Process the segment and the data it 1173 * contains. tcp_do_segment() consumes 1174 * the mbuf chain and unlocks the inpcb. 1175 */ 1176 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1177 iptos, ti_locked); 1178 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1179 return (IPPROTO_DONE); 1180 } 1181 /* 1182 * Segment flag validation for new connection attempts: 1183 * 1184 * Our (SYN|ACK) response was rejected. 1185 * Check with syncache and remove entry to prevent 1186 * retransmits. 1187 * 1188 * NB: syncache_chkrst does its own logging of failure 1189 * causes. 1190 */ 1191 if (thflags & TH_RST) { 1192 syncache_chkrst(&inc, th); 1193 goto dropunlock; 1194 } 1195 /* 1196 * We can't do anything without SYN. 1197 */ 1198 if ((thflags & TH_SYN) == 0) { 1199 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1200 log(LOG_DEBUG, "%s; %s: Listen socket: " 1201 "SYN is missing, segment ignored\n", 1202 s, __func__); 1203 TCPSTAT_INC(tcps_badsyn); 1204 goto dropunlock; 1205 } 1206 /* 1207 * (SYN|ACK) is bogus on a listen socket. 1208 */ 1209 if (thflags & TH_ACK) { 1210 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1211 log(LOG_DEBUG, "%s; %s: Listen socket: " 1212 "SYN|ACK invalid, segment rejected\n", 1213 s, __func__); 1214 syncache_badack(&inc); /* XXX: Not needed! */ 1215 TCPSTAT_INC(tcps_badsyn); 1216 rstreason = BANDLIM_RST_OPENPORT; 1217 goto dropwithreset; 1218 } 1219 /* 1220 * If the drop_synfin option is enabled, drop all 1221 * segments with both the SYN and FIN bits set. 1222 * This prevents e.g. nmap from identifying the 1223 * TCP/IP stack. 1224 * XXX: Poor reasoning. nmap has other methods 1225 * and is constantly refining its stack detection 1226 * strategies. 1227 * XXX: This is a violation of the TCP specification 1228 * and was used by RFC1644. 1229 */ 1230 if ((thflags & TH_FIN) && V_drop_synfin) { 1231 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1232 log(LOG_DEBUG, "%s; %s: Listen socket: " 1233 "SYN|FIN segment ignored (based on " 1234 "sysctl setting)\n", s, __func__); 1235 TCPSTAT_INC(tcps_badsyn); 1236 goto dropunlock; 1237 } 1238 /* 1239 * Segment's flags are (SYN) or (SYN|FIN). 1240 * 1241 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1242 * as they do not affect the state of the TCP FSM. 1243 * The data pointed to by TH_URG and th_urp is ignored. 1244 */ 1245 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1246 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1247 KASSERT(thflags & (TH_SYN), 1248 ("%s: Listen socket: TH_SYN not set", __func__)); 1249 #ifdef INET6 1250 /* 1251 * If deprecated address is forbidden, 1252 * we do not accept SYN to deprecated interface 1253 * address to prevent any new inbound connection from 1254 * getting established. 1255 * When we do not accept SYN, we send a TCP RST, 1256 * with deprecated source address (instead of dropping 1257 * it). We compromise it as it is much better for peer 1258 * to send a RST, and RST will be the final packet 1259 * for the exchange. 1260 * 1261 * If we do not forbid deprecated addresses, we accept 1262 * the SYN packet. RFC2462 does not suggest dropping 1263 * SYN in this case. 1264 * If we decipher RFC2462 5.5.4, it says like this: 1265 * 1. use of deprecated addr with existing 1266 * communication is okay - "SHOULD continue to be 1267 * used" 1268 * 2. use of it with new communication: 1269 * (2a) "SHOULD NOT be used if alternate address 1270 * with sufficient scope is available" 1271 * (2b) nothing mentioned otherwise. 1272 * Here we fall into (2b) case as we have no choice in 1273 * our source address selection - we must obey the peer. 1274 * 1275 * The wording in RFC2462 is confusing, and there are 1276 * multiple description text for deprecated address 1277 * handling - worse, they are not exactly the same. 1278 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1279 */ 1280 if (isipv6 && !V_ip6_use_deprecated) { 1281 struct in6_ifaddr *ia6; 1282 1283 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1284 if (ia6 != NULL && 1285 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1286 ifa_free(&ia6->ia_ifa); 1287 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1288 log(LOG_DEBUG, "%s; %s: Listen socket: " 1289 "Connection attempt to deprecated " 1290 "IPv6 address rejected\n", 1291 s, __func__); 1292 rstreason = BANDLIM_RST_OPENPORT; 1293 goto dropwithreset; 1294 } 1295 if (ia6) 1296 ifa_free(&ia6->ia_ifa); 1297 } 1298 #endif /* INET6 */ 1299 /* 1300 * Basic sanity checks on incoming SYN requests: 1301 * Don't respond if the destination is a link layer 1302 * broadcast according to RFC1122 4.2.3.10, p. 104. 1303 * If it is from this socket it must be forged. 1304 * Don't respond if the source or destination is a 1305 * global or subnet broad- or multicast address. 1306 * Note that it is quite possible to receive unicast 1307 * link-layer packets with a broadcast IP address. Use 1308 * in_broadcast() to find them. 1309 */ 1310 if (m->m_flags & (M_BCAST|M_MCAST)) { 1311 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1312 log(LOG_DEBUG, "%s; %s: Listen socket: " 1313 "Connection attempt from broad- or multicast " 1314 "link layer address ignored\n", s, __func__); 1315 goto dropunlock; 1316 } 1317 #ifdef INET6 1318 if (isipv6) { 1319 if (th->th_dport == th->th_sport && 1320 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1321 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1322 log(LOG_DEBUG, "%s; %s: Listen socket: " 1323 "Connection attempt to/from self " 1324 "ignored\n", s, __func__); 1325 goto dropunlock; 1326 } 1327 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1328 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1329 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1330 log(LOG_DEBUG, "%s; %s: Listen socket: " 1331 "Connection attempt from/to multicast " 1332 "address ignored\n", s, __func__); 1333 goto dropunlock; 1334 } 1335 } 1336 #endif 1337 #if defined(INET) && defined(INET6) 1338 else 1339 #endif 1340 #ifdef INET 1341 { 1342 if (th->th_dport == th->th_sport && 1343 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1344 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1345 log(LOG_DEBUG, "%s; %s: Listen socket: " 1346 "Connection attempt from/to self " 1347 "ignored\n", s, __func__); 1348 goto dropunlock; 1349 } 1350 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1351 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1352 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1353 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1354 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1355 log(LOG_DEBUG, "%s; %s: Listen socket: " 1356 "Connection attempt from/to broad- " 1357 "or multicast address ignored\n", 1358 s, __func__); 1359 goto dropunlock; 1360 } 1361 } 1362 #endif 1363 /* 1364 * SYN appears to be valid. Create compressed TCP state 1365 * for syncache. 1366 */ 1367 #ifdef TCPDEBUG 1368 if (so->so_options & SO_DEBUG) 1369 tcp_trace(TA_INPUT, ostate, tp, 1370 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1371 #endif 1372 TCP_PROBE3(debug__input, tp, th, m); 1373 tcp_dooptions(&to, optp, optlen, TO_SYN); 1374 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1375 goto tfo_socket_result; 1376 1377 /* 1378 * Entry added to syncache and mbuf consumed. 1379 * Only the listen socket is unlocked by syncache_add(). 1380 */ 1381 if (ti_locked == TI_RLOCKED) { 1382 INP_INFO_RUNLOCK(&V_tcbinfo); 1383 ti_locked = TI_UNLOCKED; 1384 } 1385 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1386 return (IPPROTO_DONE); 1387 } else if (tp->t_state == TCPS_LISTEN) { 1388 /* 1389 * When a listen socket is torn down the SO_ACCEPTCONN 1390 * flag is removed first while connections are drained 1391 * from the accept queue in a unlock/lock cycle of the 1392 * ACCEPT_LOCK, opening a race condition allowing a SYN 1393 * attempt go through unhandled. 1394 */ 1395 goto dropunlock; 1396 } 1397 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1398 if (tp->t_flags & TF_SIGNATURE) { 1399 tcp_dooptions(&to, optp, optlen, thflags); 1400 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1401 TCPSTAT_INC(tcps_sig_err_nosigopt); 1402 goto dropunlock; 1403 } 1404 if (!TCPMD5_ENABLED() || 1405 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1406 goto dropunlock; 1407 } 1408 #endif 1409 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1410 1411 /* 1412 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1413 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1414 * the inpcb, and unlocks pcbinfo. 1415 */ 1416 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1417 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1418 return (IPPROTO_DONE); 1419 1420 dropwithreset: 1421 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1422 1423 if (ti_locked == TI_RLOCKED) { 1424 INP_INFO_RUNLOCK(&V_tcbinfo); 1425 ti_locked = TI_UNLOCKED; 1426 } 1427 #ifdef INVARIANTS 1428 else { 1429 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1430 "ti_locked: %d", __func__, ti_locked)); 1431 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1432 } 1433 #endif 1434 1435 if (inp != NULL) { 1436 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1437 INP_WUNLOCK(inp); 1438 } else 1439 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1440 m = NULL; /* mbuf chain got consumed. */ 1441 goto drop; 1442 1443 dropunlock: 1444 if (m != NULL) 1445 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1446 1447 if (ti_locked == TI_RLOCKED) { 1448 INP_INFO_RUNLOCK(&V_tcbinfo); 1449 ti_locked = TI_UNLOCKED; 1450 } 1451 #ifdef INVARIANTS 1452 else { 1453 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1454 "ti_locked: %d", __func__, ti_locked)); 1455 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1456 } 1457 #endif 1458 1459 if (inp != NULL) 1460 INP_WUNLOCK(inp); 1461 1462 drop: 1463 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1464 if (s != NULL) 1465 free(s, M_TCPLOG); 1466 if (m != NULL) 1467 m_freem(m); 1468 return (IPPROTO_DONE); 1469 } 1470 1471 /* 1472 * Automatic sizing of receive socket buffer. Often the send 1473 * buffer size is not optimally adjusted to the actual network 1474 * conditions at hand (delay bandwidth product). Setting the 1475 * buffer size too small limits throughput on links with high 1476 * bandwidth and high delay (eg. trans-continental/oceanic links). 1477 * 1478 * On the receive side the socket buffer memory is only rarely 1479 * used to any significant extent. This allows us to be much 1480 * more aggressive in scaling the receive socket buffer. For 1481 * the case that the buffer space is actually used to a large 1482 * extent and we run out of kernel memory we can simply drop 1483 * the new segments; TCP on the sender will just retransmit it 1484 * later. Setting the buffer size too big may only consume too 1485 * much kernel memory if the application doesn't read() from 1486 * the socket or packet loss or reordering makes use of the 1487 * reassembly queue. 1488 * 1489 * The criteria to step up the receive buffer one notch are: 1490 * 1. Application has not set receive buffer size with 1491 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1492 * 2. the number of bytes received during the time it takes 1493 * one timestamp to be reflected back to us (the RTT); 1494 * 3. received bytes per RTT is within seven eighth of the 1495 * current socket buffer size; 1496 * 4. receive buffer size has not hit maximal automatic size; 1497 * 1498 * This algorithm does one step per RTT at most and only if 1499 * we receive a bulk stream w/o packet losses or reorderings. 1500 * Shrinking the buffer during idle times is not necessary as 1501 * it doesn't consume any memory when idle. 1502 * 1503 * TODO: Only step up if the application is actually serving 1504 * the buffer to better manage the socket buffer resources. 1505 */ 1506 int 1507 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1508 struct tcpcb *tp, int tlen) 1509 { 1510 int newsize = 0; 1511 1512 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1513 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1514 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1515 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1516 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1517 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1518 newsize = min(so->so_rcv.sb_hiwat + 1519 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1520 } 1521 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1522 1523 /* Start over with next RTT. */ 1524 tp->rfbuf_ts = 0; 1525 tp->rfbuf_cnt = 0; 1526 } else { 1527 tp->rfbuf_cnt += tlen; /* add up */ 1528 } 1529 1530 return (newsize); 1531 } 1532 1533 void 1534 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1535 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1536 int ti_locked) 1537 { 1538 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1539 int rstreason, todrop, win; 1540 uint32_t tiwin; 1541 uint16_t nsegs; 1542 char *s; 1543 struct in_conninfo *inc; 1544 struct mbuf *mfree; 1545 struct tcpopt to; 1546 int tfo_syn; 1547 1548 #ifdef TCPDEBUG 1549 /* 1550 * The size of tcp_saveipgen must be the size of the max ip header, 1551 * now IPv6. 1552 */ 1553 u_char tcp_saveipgen[IP6_HDR_LEN]; 1554 struct tcphdr tcp_savetcp; 1555 short ostate = 0; 1556 #endif 1557 thflags = th->th_flags; 1558 inc = &tp->t_inpcb->inp_inc; 1559 tp->sackhint.last_sack_ack = 0; 1560 sack_changed = 0; 1561 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1562 1563 /* 1564 * If this is either a state-changing packet or current state isn't 1565 * established, we require a write lock on tcbinfo. Otherwise, we 1566 * allow the tcbinfo to be in either alocked or unlocked, as the 1567 * caller may have unnecessarily acquired a write lock due to a race. 1568 */ 1569 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1570 tp->t_state != TCPS_ESTABLISHED) { 1571 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1572 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1573 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1574 } else { 1575 #ifdef INVARIANTS 1576 if (ti_locked == TI_RLOCKED) 1577 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1578 else { 1579 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1580 "ti_locked: %d", __func__, ti_locked)); 1581 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1582 } 1583 #endif 1584 } 1585 INP_WLOCK_ASSERT(tp->t_inpcb); 1586 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1587 __func__)); 1588 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1589 __func__)); 1590 1591 #ifdef TCPPCAP 1592 /* Save segment, if requested. */ 1593 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1594 #endif 1595 1596 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1597 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1598 log(LOG_DEBUG, "%s; %s: " 1599 "SYN|FIN segment ignored (based on " 1600 "sysctl setting)\n", s, __func__); 1601 free(s, M_TCPLOG); 1602 } 1603 goto drop; 1604 } 1605 1606 /* 1607 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1608 * check SEQ.ACK first. 1609 */ 1610 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1611 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1612 rstreason = BANDLIM_UNLIMITED; 1613 goto dropwithreset; 1614 } 1615 1616 /* 1617 * Segment received on connection. 1618 * Reset idle time and keep-alive timer. 1619 * XXX: This should be done after segment 1620 * validation to ignore broken/spoofed segs. 1621 */ 1622 tp->t_rcvtime = ticks; 1623 1624 /* 1625 * Scale up the window into a 32-bit value. 1626 * For the SYN_SENT state the scale is zero. 1627 */ 1628 tiwin = th->th_win << tp->snd_scale; 1629 1630 /* 1631 * TCP ECN processing. 1632 */ 1633 if (tp->t_flags & TF_ECN_PERMIT) { 1634 if (thflags & TH_CWR) 1635 tp->t_flags &= ~TF_ECN_SND_ECE; 1636 switch (iptos & IPTOS_ECN_MASK) { 1637 case IPTOS_ECN_CE: 1638 tp->t_flags |= TF_ECN_SND_ECE; 1639 TCPSTAT_INC(tcps_ecn_ce); 1640 break; 1641 case IPTOS_ECN_ECT0: 1642 TCPSTAT_INC(tcps_ecn_ect0); 1643 break; 1644 case IPTOS_ECN_ECT1: 1645 TCPSTAT_INC(tcps_ecn_ect1); 1646 break; 1647 } 1648 1649 /* Process a packet differently from RFC3168. */ 1650 cc_ecnpkt_handler(tp, th, iptos); 1651 1652 /* Congestion experienced. */ 1653 if (thflags & TH_ECE) { 1654 cc_cong_signal(tp, th, CC_ECN); 1655 } 1656 } 1657 1658 /* 1659 * Parse options on any incoming segment. 1660 */ 1661 tcp_dooptions(&to, (u_char *)(th + 1), 1662 (th->th_off << 2) - sizeof(struct tcphdr), 1663 (thflags & TH_SYN) ? TO_SYN : 0); 1664 1665 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1666 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1667 (to.to_flags & TOF_SIGNATURE) == 0) { 1668 TCPSTAT_INC(tcps_sig_err_sigopt); 1669 /* XXX: should drop? */ 1670 } 1671 #endif 1672 /* 1673 * If echoed timestamp is later than the current time, 1674 * fall back to non RFC1323 RTT calculation. Normalize 1675 * timestamp if syncookies were used when this connection 1676 * was established. 1677 */ 1678 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1679 to.to_tsecr -= tp->ts_offset; 1680 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1681 to.to_tsecr = 0; 1682 } 1683 /* 1684 * Process options only when we get SYN/ACK back. The SYN case 1685 * for incoming connections is handled in tcp_syncache. 1686 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1687 * or <SYN,ACK>) segment itself is never scaled. 1688 * XXX this is traditional behavior, may need to be cleaned up. 1689 */ 1690 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1691 if ((to.to_flags & TOF_SCALE) && 1692 (tp->t_flags & TF_REQ_SCALE)) { 1693 tp->t_flags |= TF_RCVD_SCALE; 1694 tp->snd_scale = to.to_wscale; 1695 } 1696 /* 1697 * Initial send window. It will be updated with 1698 * the next incoming segment to the scaled value. 1699 */ 1700 tp->snd_wnd = th->th_win; 1701 if (to.to_flags & TOF_TS) { 1702 tp->t_flags |= TF_RCVD_TSTMP; 1703 tp->ts_recent = to.to_tsval; 1704 tp->ts_recent_age = tcp_ts_getticks(); 1705 } 1706 if (to.to_flags & TOF_MSS) 1707 tcp_mss(tp, to.to_mss); 1708 if ((tp->t_flags & TF_SACK_PERMIT) && 1709 (to.to_flags & TOF_SACKPERM) == 0) 1710 tp->t_flags &= ~TF_SACK_PERMIT; 1711 if (IS_FASTOPEN(tp->t_flags)) { 1712 if (to.to_flags & TOF_FASTOPEN) 1713 tcp_fastopen_update_cache(tp, to.to_mss, 1714 to.to_tfo_len, to.to_tfo_cookie); 1715 else 1716 tcp_fastopen_disable_path(tp); 1717 } 1718 } 1719 1720 /* 1721 * If timestamps were negotiated during SYN/ACK they should 1722 * appear on every segment during this session and vice versa. 1723 */ 1724 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1725 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1726 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1727 "no action\n", s, __func__); 1728 free(s, M_TCPLOG); 1729 } 1730 } 1731 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1732 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1733 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1734 "no action\n", s, __func__); 1735 free(s, M_TCPLOG); 1736 } 1737 } 1738 1739 /* 1740 * Header prediction: check for the two common cases 1741 * of a uni-directional data xfer. If the packet has 1742 * no control flags, is in-sequence, the window didn't 1743 * change and we're not retransmitting, it's a 1744 * candidate. If the length is zero and the ack moved 1745 * forward, we're the sender side of the xfer. Just 1746 * free the data acked & wake any higher level process 1747 * that was blocked waiting for space. If the length 1748 * is non-zero and the ack didn't move, we're the 1749 * receiver side. If we're getting packets in-order 1750 * (the reassembly queue is empty), add the data to 1751 * the socket buffer and note that we need a delayed ack. 1752 * Make sure that the hidden state-flags are also off. 1753 * Since we check for TCPS_ESTABLISHED first, it can only 1754 * be TH_NEEDSYN. 1755 */ 1756 if (tp->t_state == TCPS_ESTABLISHED && 1757 th->th_seq == tp->rcv_nxt && 1758 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1759 tp->snd_nxt == tp->snd_max && 1760 tiwin && tiwin == tp->snd_wnd && 1761 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1762 LIST_EMPTY(&tp->t_segq) && 1763 ((to.to_flags & TOF_TS) == 0 || 1764 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1765 1766 /* 1767 * If last ACK falls within this segment's sequence numbers, 1768 * record the timestamp. 1769 * NOTE that the test is modified according to the latest 1770 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1771 */ 1772 if ((to.to_flags & TOF_TS) != 0 && 1773 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1774 tp->ts_recent_age = tcp_ts_getticks(); 1775 tp->ts_recent = to.to_tsval; 1776 } 1777 1778 if (tlen == 0) { 1779 if (SEQ_GT(th->th_ack, tp->snd_una) && 1780 SEQ_LEQ(th->th_ack, tp->snd_max) && 1781 !IN_RECOVERY(tp->t_flags) && 1782 (to.to_flags & TOF_SACK) == 0 && 1783 TAILQ_EMPTY(&tp->snd_holes)) { 1784 /* 1785 * This is a pure ack for outstanding data. 1786 */ 1787 if (ti_locked == TI_RLOCKED) 1788 INP_INFO_RUNLOCK(&V_tcbinfo); 1789 ti_locked = TI_UNLOCKED; 1790 1791 TCPSTAT_INC(tcps_predack); 1792 1793 /* 1794 * "bad retransmit" recovery. 1795 */ 1796 if (tp->t_rxtshift == 1 && 1797 tp->t_flags & TF_PREVVALID && 1798 (int)(ticks - tp->t_badrxtwin) < 0) { 1799 cc_cong_signal(tp, th, CC_RTO_ERR); 1800 } 1801 1802 /* 1803 * Recalculate the transmit timer / rtt. 1804 * 1805 * Some boxes send broken timestamp replies 1806 * during the SYN+ACK phase, ignore 1807 * timestamps of 0 or we could calculate a 1808 * huge RTT and blow up the retransmit timer. 1809 */ 1810 if ((to.to_flags & TOF_TS) != 0 && 1811 to.to_tsecr) { 1812 uint32_t t; 1813 1814 t = tcp_ts_getticks() - to.to_tsecr; 1815 if (!tp->t_rttlow || tp->t_rttlow > t) 1816 tp->t_rttlow = t; 1817 tcp_xmit_timer(tp, 1818 TCP_TS_TO_TICKS(t) + 1); 1819 } else if (tp->t_rtttime && 1820 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1821 if (!tp->t_rttlow || 1822 tp->t_rttlow > ticks - tp->t_rtttime) 1823 tp->t_rttlow = ticks - tp->t_rtttime; 1824 tcp_xmit_timer(tp, 1825 ticks - tp->t_rtttime); 1826 } 1827 acked = BYTES_THIS_ACK(tp, th); 1828 1829 #ifdef TCP_HHOOK 1830 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1831 hhook_run_tcp_est_in(tp, th, &to); 1832 #endif 1833 1834 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1835 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1836 sbdrop(&so->so_snd, acked); 1837 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1838 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1839 tp->snd_recover = th->th_ack - 1; 1840 1841 /* 1842 * Let the congestion control algorithm update 1843 * congestion control related information. This 1844 * typically means increasing the congestion 1845 * window. 1846 */ 1847 cc_ack_received(tp, th, nsegs, CC_ACK); 1848 1849 tp->snd_una = th->th_ack; 1850 /* 1851 * Pull snd_wl2 up to prevent seq wrap relative 1852 * to th_ack. 1853 */ 1854 tp->snd_wl2 = th->th_ack; 1855 tp->t_dupacks = 0; 1856 m_freem(m); 1857 1858 /* 1859 * If all outstanding data are acked, stop 1860 * retransmit timer, otherwise restart timer 1861 * using current (possibly backed-off) value. 1862 * If process is waiting for space, 1863 * wakeup/selwakeup/signal. If data 1864 * are ready to send, let tcp_output 1865 * decide between more output or persist. 1866 */ 1867 #ifdef TCPDEBUG 1868 if (so->so_options & SO_DEBUG) 1869 tcp_trace(TA_INPUT, ostate, tp, 1870 (void *)tcp_saveipgen, 1871 &tcp_savetcp, 0); 1872 #endif 1873 TCP_PROBE3(debug__input, tp, th, m); 1874 if (tp->snd_una == tp->snd_max) 1875 tcp_timer_activate(tp, TT_REXMT, 0); 1876 else if (!tcp_timer_active(tp, TT_PERSIST)) 1877 tcp_timer_activate(tp, TT_REXMT, 1878 tp->t_rxtcur); 1879 sowwakeup(so); 1880 if (sbavail(&so->so_snd)) 1881 (void) tp->t_fb->tfb_tcp_output(tp); 1882 goto check_delack; 1883 } 1884 } else if (th->th_ack == tp->snd_una && 1885 tlen <= sbspace(&so->so_rcv)) { 1886 int newsize = 0; /* automatic sockbuf scaling */ 1887 1888 /* 1889 * This is a pure, in-sequence data packet with 1890 * nothing on the reassembly queue and we have enough 1891 * buffer space to take it. 1892 */ 1893 if (ti_locked == TI_RLOCKED) 1894 INP_INFO_RUNLOCK(&V_tcbinfo); 1895 ti_locked = TI_UNLOCKED; 1896 1897 /* Clean receiver SACK report if present */ 1898 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1899 tcp_clean_sackreport(tp); 1900 TCPSTAT_INC(tcps_preddat); 1901 tp->rcv_nxt += tlen; 1902 /* 1903 * Pull snd_wl1 up to prevent seq wrap relative to 1904 * th_seq. 1905 */ 1906 tp->snd_wl1 = th->th_seq; 1907 /* 1908 * Pull rcv_up up to prevent seq wrap relative to 1909 * rcv_nxt. 1910 */ 1911 tp->rcv_up = tp->rcv_nxt; 1912 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1913 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1914 #ifdef TCPDEBUG 1915 if (so->so_options & SO_DEBUG) 1916 tcp_trace(TA_INPUT, ostate, tp, 1917 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1918 #endif 1919 TCP_PROBE3(debug__input, tp, th, m); 1920 1921 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1922 1923 /* Add data to socket buffer. */ 1924 SOCKBUF_LOCK(&so->so_rcv); 1925 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1926 m_freem(m); 1927 } else { 1928 /* 1929 * Set new socket buffer size. 1930 * Give up when limit is reached. 1931 */ 1932 if (newsize) 1933 if (!sbreserve_locked(&so->so_rcv, 1934 newsize, so, NULL)) 1935 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1936 m_adj(m, drop_hdrlen); /* delayed header drop */ 1937 sbappendstream_locked(&so->so_rcv, m, 0); 1938 } 1939 /* NB: sorwakeup_locked() does an implicit unlock. */ 1940 sorwakeup_locked(so); 1941 if (DELAY_ACK(tp, tlen)) { 1942 tp->t_flags |= TF_DELACK; 1943 } else { 1944 tp->t_flags |= TF_ACKNOW; 1945 tp->t_fb->tfb_tcp_output(tp); 1946 } 1947 goto check_delack; 1948 } 1949 } 1950 1951 /* 1952 * Calculate amount of space in receive window, 1953 * and then do TCP input processing. 1954 * Receive window is amount of space in rcv queue, 1955 * but not less than advertised window. 1956 */ 1957 win = sbspace(&so->so_rcv); 1958 if (win < 0) 1959 win = 0; 1960 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1961 1962 switch (tp->t_state) { 1963 1964 /* 1965 * If the state is SYN_RECEIVED: 1966 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1967 */ 1968 case TCPS_SYN_RECEIVED: 1969 if ((thflags & TH_ACK) && 1970 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1971 SEQ_GT(th->th_ack, tp->snd_max))) { 1972 rstreason = BANDLIM_RST_OPENPORT; 1973 goto dropwithreset; 1974 } 1975 if (IS_FASTOPEN(tp->t_flags)) { 1976 /* 1977 * When a TFO connection is in SYN_RECEIVED, the 1978 * only valid packets are the initial SYN, a 1979 * retransmit/copy of the initial SYN (possibly with 1980 * a subset of the original data), a valid ACK, a 1981 * FIN, or a RST. 1982 */ 1983 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1984 rstreason = BANDLIM_RST_OPENPORT; 1985 goto dropwithreset; 1986 } else if (thflags & TH_SYN) { 1987 /* non-initial SYN is ignored */ 1988 if ((tcp_timer_active(tp, TT_DELACK) || 1989 tcp_timer_active(tp, TT_REXMT))) 1990 goto drop; 1991 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1992 goto drop; 1993 } 1994 } 1995 break; 1996 1997 /* 1998 * If the state is SYN_SENT: 1999 * if seg contains a RST with valid ACK (SEQ.ACK has already 2000 * been verified), then drop the connection. 2001 * if seg contains a RST without an ACK, drop the seg. 2002 * if seg does not contain SYN, then drop the seg. 2003 * Otherwise this is an acceptable SYN segment 2004 * initialize tp->rcv_nxt and tp->irs 2005 * if seg contains ack then advance tp->snd_una 2006 * if seg contains an ECE and ECN support is enabled, the stream 2007 * is ECN capable. 2008 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2009 * arrange for segment to be acked (eventually) 2010 * continue processing rest of data/controls, beginning with URG 2011 */ 2012 case TCPS_SYN_SENT: 2013 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2014 TCP_PROBE5(connect__refused, NULL, tp, 2015 m, tp, th); 2016 tp = tcp_drop(tp, ECONNREFUSED); 2017 } 2018 if (thflags & TH_RST) 2019 goto drop; 2020 if (!(thflags & TH_SYN)) 2021 goto drop; 2022 2023 tp->irs = th->th_seq; 2024 tcp_rcvseqinit(tp); 2025 if (thflags & TH_ACK) { 2026 int tfo_partial_ack = 0; 2027 2028 TCPSTAT_INC(tcps_connects); 2029 soisconnected(so); 2030 #ifdef MAC 2031 mac_socketpeer_set_from_mbuf(m, so); 2032 #endif 2033 /* Do window scaling on this connection? */ 2034 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2035 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2036 tp->rcv_scale = tp->request_r_scale; 2037 } 2038 tp->rcv_adv += min(tp->rcv_wnd, 2039 TCP_MAXWIN << tp->rcv_scale); 2040 tp->snd_una++; /* SYN is acked */ 2041 /* 2042 * If not all the data that was sent in the TFO SYN 2043 * has been acked, resend the remainder right away. 2044 */ 2045 if (IS_FASTOPEN(tp->t_flags) && 2046 (tp->snd_una != tp->snd_max)) { 2047 tp->snd_nxt = th->th_ack; 2048 tfo_partial_ack = 1; 2049 } 2050 /* 2051 * If there's data, delay ACK; if there's also a FIN 2052 * ACKNOW will be turned on later. 2053 */ 2054 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) 2055 tcp_timer_activate(tp, TT_DELACK, 2056 tcp_delacktime); 2057 else 2058 tp->t_flags |= TF_ACKNOW; 2059 2060 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2061 tp->t_flags |= TF_ECN_PERMIT; 2062 TCPSTAT_INC(tcps_ecn_shs); 2063 } 2064 2065 /* 2066 * Received <SYN,ACK> in SYN_SENT[*] state. 2067 * Transitions: 2068 * SYN_SENT --> ESTABLISHED 2069 * SYN_SENT* --> FIN_WAIT_1 2070 */ 2071 tp->t_starttime = ticks; 2072 if (tp->t_flags & TF_NEEDFIN) { 2073 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2074 tp->t_flags &= ~TF_NEEDFIN; 2075 thflags &= ~TH_SYN; 2076 } else { 2077 tcp_state_change(tp, TCPS_ESTABLISHED); 2078 TCP_PROBE5(connect__established, NULL, tp, 2079 m, tp, th); 2080 cc_conn_init(tp); 2081 tcp_timer_activate(tp, TT_KEEP, 2082 TP_KEEPIDLE(tp)); 2083 } 2084 } else { 2085 /* 2086 * Received initial SYN in SYN-SENT[*] state => 2087 * simultaneous open. 2088 * If it succeeds, connection is * half-synchronized. 2089 * Otherwise, do 3-way handshake: 2090 * SYN-SENT -> SYN-RECEIVED 2091 * SYN-SENT* -> SYN-RECEIVED* 2092 */ 2093 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2094 tcp_timer_activate(tp, TT_REXMT, 0); 2095 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2096 } 2097 2098 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2099 "ti_locked %d", __func__, ti_locked)); 2100 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2101 INP_WLOCK_ASSERT(tp->t_inpcb); 2102 2103 /* 2104 * Advance th->th_seq to correspond to first data byte. 2105 * If data, trim to stay within window, 2106 * dropping FIN if necessary. 2107 */ 2108 th->th_seq++; 2109 if (tlen > tp->rcv_wnd) { 2110 todrop = tlen - tp->rcv_wnd; 2111 m_adj(m, -todrop); 2112 tlen = tp->rcv_wnd; 2113 thflags &= ~TH_FIN; 2114 TCPSTAT_INC(tcps_rcvpackafterwin); 2115 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2116 } 2117 tp->snd_wl1 = th->th_seq - 1; 2118 tp->rcv_up = th->th_seq; 2119 /* 2120 * Client side of transaction: already sent SYN and data. 2121 * If the remote host used T/TCP to validate the SYN, 2122 * our data will be ACK'd; if so, enter normal data segment 2123 * processing in the middle of step 5, ack processing. 2124 * Otherwise, goto step 6. 2125 */ 2126 if (thflags & TH_ACK) 2127 goto process_ACK; 2128 2129 goto step6; 2130 2131 /* 2132 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2133 * do normal processing. 2134 * 2135 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2136 */ 2137 case TCPS_LAST_ACK: 2138 case TCPS_CLOSING: 2139 break; /* continue normal processing */ 2140 } 2141 2142 /* 2143 * States other than LISTEN or SYN_SENT. 2144 * First check the RST flag and sequence number since reset segments 2145 * are exempt from the timestamp and connection count tests. This 2146 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2147 * below which allowed reset segments in half the sequence space 2148 * to fall though and be processed (which gives forged reset 2149 * segments with a random sequence number a 50 percent chance of 2150 * killing a connection). 2151 * Then check timestamp, if present. 2152 * Then check the connection count, if present. 2153 * Then check that at least some bytes of segment are within 2154 * receive window. If segment begins before rcv_nxt, 2155 * drop leading data (and SYN); if nothing left, just ack. 2156 */ 2157 if (thflags & TH_RST) { 2158 /* 2159 * RFC5961 Section 3.2 2160 * 2161 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2162 * - If RST is in window, we send challenge ACK. 2163 * 2164 * Note: to take into account delayed ACKs, we should 2165 * test against last_ack_sent instead of rcv_nxt. 2166 * Note 2: we handle special case of closed window, not 2167 * covered by the RFC. 2168 */ 2169 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2170 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2171 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2172 2173 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2174 KASSERT(ti_locked == TI_RLOCKED, 2175 ("%s: TH_RST ti_locked %d, th %p tp %p", 2176 __func__, ti_locked, th, tp)); 2177 KASSERT(tp->t_state != TCPS_SYN_SENT, 2178 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2179 __func__, th, tp)); 2180 2181 if (V_tcp_insecure_rst || 2182 tp->last_ack_sent == th->th_seq) { 2183 TCPSTAT_INC(tcps_drops); 2184 /* Drop the connection. */ 2185 switch (tp->t_state) { 2186 case TCPS_SYN_RECEIVED: 2187 so->so_error = ECONNREFUSED; 2188 goto close; 2189 case TCPS_ESTABLISHED: 2190 case TCPS_FIN_WAIT_1: 2191 case TCPS_FIN_WAIT_2: 2192 case TCPS_CLOSE_WAIT: 2193 case TCPS_CLOSING: 2194 case TCPS_LAST_ACK: 2195 so->so_error = ECONNRESET; 2196 close: 2197 /* FALLTHROUGH */ 2198 default: 2199 tp = tcp_close(tp); 2200 } 2201 } else { 2202 TCPSTAT_INC(tcps_badrst); 2203 /* Send challenge ACK. */ 2204 tcp_respond(tp, mtod(m, void *), th, m, 2205 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2206 tp->last_ack_sent = tp->rcv_nxt; 2207 m = NULL; 2208 } 2209 } 2210 goto drop; 2211 } 2212 2213 /* 2214 * RFC5961 Section 4.2 2215 * Send challenge ACK for any SYN in synchronized state. 2216 */ 2217 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2218 tp->t_state != TCPS_SYN_RECEIVED) { 2219 KASSERT(ti_locked == TI_RLOCKED, 2220 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2221 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2222 2223 TCPSTAT_INC(tcps_badsyn); 2224 if (V_tcp_insecure_syn && 2225 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2226 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2227 tp = tcp_drop(tp, ECONNRESET); 2228 rstreason = BANDLIM_UNLIMITED; 2229 } else { 2230 /* Send challenge ACK. */ 2231 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2232 tp->snd_nxt, TH_ACK); 2233 tp->last_ack_sent = tp->rcv_nxt; 2234 m = NULL; 2235 } 2236 goto drop; 2237 } 2238 2239 /* 2240 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2241 * and it's less than ts_recent, drop it. 2242 */ 2243 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2244 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2245 2246 /* Check to see if ts_recent is over 24 days old. */ 2247 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2248 /* 2249 * Invalidate ts_recent. If this segment updates 2250 * ts_recent, the age will be reset later and ts_recent 2251 * will get a valid value. If it does not, setting 2252 * ts_recent to zero will at least satisfy the 2253 * requirement that zero be placed in the timestamp 2254 * echo reply when ts_recent isn't valid. The 2255 * age isn't reset until we get a valid ts_recent 2256 * because we don't want out-of-order segments to be 2257 * dropped when ts_recent is old. 2258 */ 2259 tp->ts_recent = 0; 2260 } else { 2261 TCPSTAT_INC(tcps_rcvduppack); 2262 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2263 TCPSTAT_INC(tcps_pawsdrop); 2264 if (tlen) 2265 goto dropafterack; 2266 goto drop; 2267 } 2268 } 2269 2270 /* 2271 * In the SYN-RECEIVED state, validate that the packet belongs to 2272 * this connection before trimming the data to fit the receive 2273 * window. Check the sequence number versus IRS since we know 2274 * the sequence numbers haven't wrapped. This is a partial fix 2275 * for the "LAND" DoS attack. 2276 */ 2277 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2278 rstreason = BANDLIM_RST_OPENPORT; 2279 goto dropwithreset; 2280 } 2281 2282 todrop = tp->rcv_nxt - th->th_seq; 2283 if (todrop > 0) { 2284 if (thflags & TH_SYN) { 2285 thflags &= ~TH_SYN; 2286 th->th_seq++; 2287 if (th->th_urp > 1) 2288 th->th_urp--; 2289 else 2290 thflags &= ~TH_URG; 2291 todrop--; 2292 } 2293 /* 2294 * Following if statement from Stevens, vol. 2, p. 960. 2295 */ 2296 if (todrop > tlen 2297 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2298 /* 2299 * Any valid FIN must be to the left of the window. 2300 * At this point the FIN must be a duplicate or out 2301 * of sequence; drop it. 2302 */ 2303 thflags &= ~TH_FIN; 2304 2305 /* 2306 * Send an ACK to resynchronize and drop any data. 2307 * But keep on processing for RST or ACK. 2308 */ 2309 tp->t_flags |= TF_ACKNOW; 2310 todrop = tlen; 2311 TCPSTAT_INC(tcps_rcvduppack); 2312 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2313 } else { 2314 TCPSTAT_INC(tcps_rcvpartduppack); 2315 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2316 } 2317 drop_hdrlen += todrop; /* drop from the top afterwards */ 2318 th->th_seq += todrop; 2319 tlen -= todrop; 2320 if (th->th_urp > todrop) 2321 th->th_urp -= todrop; 2322 else { 2323 thflags &= ~TH_URG; 2324 th->th_urp = 0; 2325 } 2326 } 2327 2328 /* 2329 * If new data are received on a connection after the 2330 * user processes are gone, then RST the other end. 2331 */ 2332 if ((so->so_state & SS_NOFDREF) && 2333 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2334 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2335 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2336 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2337 2338 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2339 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2340 "after socket was closed, " 2341 "sending RST and removing tcpcb\n", 2342 s, __func__, tcpstates[tp->t_state], tlen); 2343 free(s, M_TCPLOG); 2344 } 2345 tp = tcp_close(tp); 2346 TCPSTAT_INC(tcps_rcvafterclose); 2347 rstreason = BANDLIM_UNLIMITED; 2348 goto dropwithreset; 2349 } 2350 2351 /* 2352 * If segment ends after window, drop trailing data 2353 * (and PUSH and FIN); if nothing left, just ACK. 2354 */ 2355 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2356 if (todrop > 0) { 2357 TCPSTAT_INC(tcps_rcvpackafterwin); 2358 if (todrop >= tlen) { 2359 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2360 /* 2361 * If window is closed can only take segments at 2362 * window edge, and have to drop data and PUSH from 2363 * incoming segments. Continue processing, but 2364 * remember to ack. Otherwise, drop segment 2365 * and ack. 2366 */ 2367 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2368 tp->t_flags |= TF_ACKNOW; 2369 TCPSTAT_INC(tcps_rcvwinprobe); 2370 } else 2371 goto dropafterack; 2372 } else 2373 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2374 m_adj(m, -todrop); 2375 tlen -= todrop; 2376 thflags &= ~(TH_PUSH|TH_FIN); 2377 } 2378 2379 /* 2380 * If last ACK falls within this segment's sequence numbers, 2381 * record its timestamp. 2382 * NOTE: 2383 * 1) That the test incorporates suggestions from the latest 2384 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2385 * 2) That updating only on newer timestamps interferes with 2386 * our earlier PAWS tests, so this check should be solely 2387 * predicated on the sequence space of this segment. 2388 * 3) That we modify the segment boundary check to be 2389 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2390 * instead of RFC1323's 2391 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2392 * This modified check allows us to overcome RFC1323's 2393 * limitations as described in Stevens TCP/IP Illustrated 2394 * Vol. 2 p.869. In such cases, we can still calculate the 2395 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2396 */ 2397 if ((to.to_flags & TOF_TS) != 0 && 2398 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2399 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2400 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2401 tp->ts_recent_age = tcp_ts_getticks(); 2402 tp->ts_recent = to.to_tsval; 2403 } 2404 2405 /* 2406 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2407 * flag is on (half-synchronized state), then queue data for 2408 * later processing; else drop segment and return. 2409 */ 2410 if ((thflags & TH_ACK) == 0) { 2411 if (tp->t_state == TCPS_SYN_RECEIVED || 2412 (tp->t_flags & TF_NEEDSYN)) { 2413 if (tp->t_state == TCPS_SYN_RECEIVED && 2414 IS_FASTOPEN(tp->t_flags)) { 2415 tp->snd_wnd = tiwin; 2416 cc_conn_init(tp); 2417 } 2418 goto step6; 2419 } else if (tp->t_flags & TF_ACKNOW) 2420 goto dropafterack; 2421 else 2422 goto drop; 2423 } 2424 2425 /* 2426 * Ack processing. 2427 */ 2428 switch (tp->t_state) { 2429 2430 /* 2431 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2432 * ESTABLISHED state and continue processing. 2433 * The ACK was checked above. 2434 */ 2435 case TCPS_SYN_RECEIVED: 2436 2437 TCPSTAT_INC(tcps_connects); 2438 soisconnected(so); 2439 /* Do window scaling? */ 2440 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2441 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2442 tp->rcv_scale = tp->request_r_scale; 2443 tp->snd_wnd = tiwin; 2444 } 2445 /* 2446 * Make transitions: 2447 * SYN-RECEIVED -> ESTABLISHED 2448 * SYN-RECEIVED* -> FIN-WAIT-1 2449 */ 2450 tp->t_starttime = ticks; 2451 if (tp->t_flags & TF_NEEDFIN) { 2452 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2453 tp->t_flags &= ~TF_NEEDFIN; 2454 } else { 2455 tcp_state_change(tp, TCPS_ESTABLISHED); 2456 TCP_PROBE5(accept__established, NULL, tp, 2457 m, tp, th); 2458 if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { 2459 tcp_fastopen_decrement_counter(tp->t_tfo_pending); 2460 tp->t_tfo_pending = NULL; 2461 2462 /* 2463 * Account for the ACK of our SYN prior to 2464 * regular ACK processing below. 2465 */ 2466 tp->snd_una++; 2467 } 2468 /* 2469 * TFO connections call cc_conn_init() during SYN 2470 * processing. Calling it again here for such 2471 * connections is not harmless as it would undo the 2472 * snd_cwnd reduction that occurs when a TFO SYN|ACK 2473 * is retransmitted. 2474 */ 2475 if (!IS_FASTOPEN(tp->t_flags)) 2476 cc_conn_init(tp); 2477 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2478 } 2479 /* 2480 * If segment contains data or ACK, will call tcp_reass() 2481 * later; if not, do so now to pass queued data to user. 2482 */ 2483 if (tlen == 0 && (thflags & TH_FIN) == 0) 2484 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2485 (struct mbuf *)0); 2486 tp->snd_wl1 = th->th_seq - 1; 2487 /* FALLTHROUGH */ 2488 2489 /* 2490 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2491 * ACKs. If the ack is in the range 2492 * tp->snd_una < th->th_ack <= tp->snd_max 2493 * then advance tp->snd_una to th->th_ack and drop 2494 * data from the retransmission queue. If this ACK reflects 2495 * more up to date window information we update our window information. 2496 */ 2497 case TCPS_ESTABLISHED: 2498 case TCPS_FIN_WAIT_1: 2499 case TCPS_FIN_WAIT_2: 2500 case TCPS_CLOSE_WAIT: 2501 case TCPS_CLOSING: 2502 case TCPS_LAST_ACK: 2503 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2504 TCPSTAT_INC(tcps_rcvacktoomuch); 2505 goto dropafterack; 2506 } 2507 if ((tp->t_flags & TF_SACK_PERMIT) && 2508 ((to.to_flags & TOF_SACK) || 2509 !TAILQ_EMPTY(&tp->snd_holes))) 2510 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2511 else 2512 /* 2513 * Reset the value so that previous (valid) value 2514 * from the last ack with SACK doesn't get used. 2515 */ 2516 tp->sackhint.sacked_bytes = 0; 2517 2518 #ifdef TCP_HHOOK 2519 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2520 hhook_run_tcp_est_in(tp, th, &to); 2521 #endif 2522 2523 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2524 u_int maxseg; 2525 2526 maxseg = tcp_maxseg(tp); 2527 if (tlen == 0 && 2528 (tiwin == tp->snd_wnd || 2529 (tp->t_flags & TF_SACK_PERMIT))) { 2530 /* 2531 * If this is the first time we've seen a 2532 * FIN from the remote, this is not a 2533 * duplicate and it needs to be processed 2534 * normally. This happens during a 2535 * simultaneous close. 2536 */ 2537 if ((thflags & TH_FIN) && 2538 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2539 tp->t_dupacks = 0; 2540 break; 2541 } 2542 TCPSTAT_INC(tcps_rcvdupack); 2543 /* 2544 * If we have outstanding data (other than 2545 * a window probe), this is a completely 2546 * duplicate ack (ie, window info didn't 2547 * change and FIN isn't set), 2548 * the ack is the biggest we've 2549 * seen and we've seen exactly our rexmt 2550 * threshold of them, assume a packet 2551 * has been dropped and retransmit it. 2552 * Kludge snd_nxt & the congestion 2553 * window so we send only this one 2554 * packet. 2555 * 2556 * We know we're losing at the current 2557 * window size so do congestion avoidance 2558 * (set ssthresh to half the current window 2559 * and pull our congestion window back to 2560 * the new ssthresh). 2561 * 2562 * Dup acks mean that packets have left the 2563 * network (they're now cached at the receiver) 2564 * so bump cwnd by the amount in the receiver 2565 * to keep a constant cwnd packets in the 2566 * network. 2567 * 2568 * When using TCP ECN, notify the peer that 2569 * we reduced the cwnd. 2570 */ 2571 /* 2572 * Following 2 kinds of acks should not affect 2573 * dupack counting: 2574 * 1) Old acks 2575 * 2) Acks with SACK but without any new SACK 2576 * information in them. These could result from 2577 * any anomaly in the network like a switch 2578 * duplicating packets or a possible DoS attack. 2579 */ 2580 if (th->th_ack != tp->snd_una || 2581 ((tp->t_flags & TF_SACK_PERMIT) && 2582 !sack_changed)) 2583 break; 2584 else if (!tcp_timer_active(tp, TT_REXMT)) 2585 tp->t_dupacks = 0; 2586 else if (++tp->t_dupacks > tcprexmtthresh || 2587 IN_FASTRECOVERY(tp->t_flags)) { 2588 cc_ack_received(tp, th, nsegs, 2589 CC_DUPACK); 2590 if ((tp->t_flags & TF_SACK_PERMIT) && 2591 IN_FASTRECOVERY(tp->t_flags)) { 2592 int awnd; 2593 2594 /* 2595 * Compute the amount of data in flight first. 2596 * We can inject new data into the pipe iff 2597 * we have less than 1/2 the original window's 2598 * worth of data in flight. 2599 */ 2600 if (V_tcp_do_rfc6675_pipe) 2601 awnd = tcp_compute_pipe(tp); 2602 else 2603 awnd = (tp->snd_nxt - tp->snd_fack) + 2604 tp->sackhint.sack_bytes_rexmit; 2605 2606 if (awnd < tp->snd_ssthresh) { 2607 tp->snd_cwnd += maxseg; 2608 if (tp->snd_cwnd > tp->snd_ssthresh) 2609 tp->snd_cwnd = tp->snd_ssthresh; 2610 } 2611 } else 2612 tp->snd_cwnd += maxseg; 2613 (void) tp->t_fb->tfb_tcp_output(tp); 2614 goto drop; 2615 } else if (tp->t_dupacks == tcprexmtthresh) { 2616 tcp_seq onxt = tp->snd_nxt; 2617 2618 /* 2619 * If we're doing sack, check to 2620 * see if we're already in sack 2621 * recovery. If we're not doing sack, 2622 * check to see if we're in newreno 2623 * recovery. 2624 */ 2625 if (tp->t_flags & TF_SACK_PERMIT) { 2626 if (IN_FASTRECOVERY(tp->t_flags)) { 2627 tp->t_dupacks = 0; 2628 break; 2629 } 2630 } else { 2631 if (SEQ_LEQ(th->th_ack, 2632 tp->snd_recover)) { 2633 tp->t_dupacks = 0; 2634 break; 2635 } 2636 } 2637 /* Congestion signal before ack. */ 2638 cc_cong_signal(tp, th, CC_NDUPACK); 2639 cc_ack_received(tp, th, nsegs, 2640 CC_DUPACK); 2641 tcp_timer_activate(tp, TT_REXMT, 0); 2642 tp->t_rtttime = 0; 2643 if (tp->t_flags & TF_SACK_PERMIT) { 2644 TCPSTAT_INC( 2645 tcps_sack_recovery_episode); 2646 tp->sack_newdata = tp->snd_nxt; 2647 tp->snd_cwnd = maxseg; 2648 (void) tp->t_fb->tfb_tcp_output(tp); 2649 goto drop; 2650 } 2651 tp->snd_nxt = th->th_ack; 2652 tp->snd_cwnd = maxseg; 2653 (void) tp->t_fb->tfb_tcp_output(tp); 2654 KASSERT(tp->snd_limited <= 2, 2655 ("%s: tp->snd_limited too big", 2656 __func__)); 2657 tp->snd_cwnd = tp->snd_ssthresh + 2658 maxseg * 2659 (tp->t_dupacks - tp->snd_limited); 2660 if (SEQ_GT(onxt, tp->snd_nxt)) 2661 tp->snd_nxt = onxt; 2662 goto drop; 2663 } else if (V_tcp_do_rfc3042) { 2664 /* 2665 * Process first and second duplicate 2666 * ACKs. Each indicates a segment 2667 * leaving the network, creating room 2668 * for more. Make sure we can send a 2669 * packet on reception of each duplicate 2670 * ACK by increasing snd_cwnd by one 2671 * segment. Restore the original 2672 * snd_cwnd after packet transmission. 2673 */ 2674 cc_ack_received(tp, th, nsegs, 2675 CC_DUPACK); 2676 uint32_t oldcwnd = tp->snd_cwnd; 2677 tcp_seq oldsndmax = tp->snd_max; 2678 u_int sent; 2679 int avail; 2680 2681 KASSERT(tp->t_dupacks == 1 || 2682 tp->t_dupacks == 2, 2683 ("%s: dupacks not 1 or 2", 2684 __func__)); 2685 if (tp->t_dupacks == 1) 2686 tp->snd_limited = 0; 2687 tp->snd_cwnd = 2688 (tp->snd_nxt - tp->snd_una) + 2689 (tp->t_dupacks - tp->snd_limited) * 2690 maxseg; 2691 /* 2692 * Only call tcp_output when there 2693 * is new data available to be sent. 2694 * Otherwise we would send pure ACKs. 2695 */ 2696 SOCKBUF_LOCK(&so->so_snd); 2697 avail = sbavail(&so->so_snd) - 2698 (tp->snd_nxt - tp->snd_una); 2699 SOCKBUF_UNLOCK(&so->so_snd); 2700 if (avail > 0) 2701 (void) tp->t_fb->tfb_tcp_output(tp); 2702 sent = tp->snd_max - oldsndmax; 2703 if (sent > maxseg) { 2704 KASSERT((tp->t_dupacks == 2 && 2705 tp->snd_limited == 0) || 2706 (sent == maxseg + 1 && 2707 tp->t_flags & TF_SENTFIN), 2708 ("%s: sent too much", 2709 __func__)); 2710 tp->snd_limited = 2; 2711 } else if (sent > 0) 2712 ++tp->snd_limited; 2713 tp->snd_cwnd = oldcwnd; 2714 goto drop; 2715 } 2716 } 2717 break; 2718 } else { 2719 /* 2720 * This ack is advancing the left edge, reset the 2721 * counter. 2722 */ 2723 tp->t_dupacks = 0; 2724 /* 2725 * If this ack also has new SACK info, increment the 2726 * counter as per rfc6675. 2727 */ 2728 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2729 tp->t_dupacks++; 2730 } 2731 2732 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2733 ("%s: th_ack <= snd_una", __func__)); 2734 2735 /* 2736 * If the congestion window was inflated to account 2737 * for the other side's cached packets, retract it. 2738 */ 2739 if (IN_FASTRECOVERY(tp->t_flags)) { 2740 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2741 if (tp->t_flags & TF_SACK_PERMIT) 2742 tcp_sack_partialack(tp, th); 2743 else 2744 tcp_newreno_partial_ack(tp, th); 2745 } else 2746 cc_post_recovery(tp, th); 2747 } 2748 /* 2749 * If we reach this point, ACK is not a duplicate, 2750 * i.e., it ACKs something we sent. 2751 */ 2752 if (tp->t_flags & TF_NEEDSYN) { 2753 /* 2754 * T/TCP: Connection was half-synchronized, and our 2755 * SYN has been ACK'd (so connection is now fully 2756 * synchronized). Go to non-starred state, 2757 * increment snd_una for ACK of SYN, and check if 2758 * we can do window scaling. 2759 */ 2760 tp->t_flags &= ~TF_NEEDSYN; 2761 tp->snd_una++; 2762 /* Do window scaling? */ 2763 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2764 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2765 tp->rcv_scale = tp->request_r_scale; 2766 /* Send window already scaled. */ 2767 } 2768 } 2769 2770 process_ACK: 2771 INP_WLOCK_ASSERT(tp->t_inpcb); 2772 2773 acked = BYTES_THIS_ACK(tp, th); 2774 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2775 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2776 tp->snd_una, th->th_ack, tp, m)); 2777 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2778 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2779 2780 /* 2781 * If we just performed our first retransmit, and the ACK 2782 * arrives within our recovery window, then it was a mistake 2783 * to do the retransmit in the first place. Recover our 2784 * original cwnd and ssthresh, and proceed to transmit where 2785 * we left off. 2786 */ 2787 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2788 (int)(ticks - tp->t_badrxtwin) < 0) 2789 cc_cong_signal(tp, th, CC_RTO_ERR); 2790 2791 /* 2792 * If we have a timestamp reply, update smoothed 2793 * round trip time. If no timestamp is present but 2794 * transmit timer is running and timed sequence 2795 * number was acked, update smoothed round trip time. 2796 * Since we now have an rtt measurement, cancel the 2797 * timer backoff (cf., Phil Karn's retransmit alg.). 2798 * Recompute the initial retransmit timer. 2799 * 2800 * Some boxes send broken timestamp replies 2801 * during the SYN+ACK phase, ignore 2802 * timestamps of 0 or we could calculate a 2803 * huge RTT and blow up the retransmit timer. 2804 */ 2805 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2806 uint32_t t; 2807 2808 t = tcp_ts_getticks() - to.to_tsecr; 2809 if (!tp->t_rttlow || tp->t_rttlow > t) 2810 tp->t_rttlow = t; 2811 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2812 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2813 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2814 tp->t_rttlow = ticks - tp->t_rtttime; 2815 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2816 } 2817 2818 /* 2819 * If all outstanding data is acked, stop retransmit 2820 * timer and remember to restart (more output or persist). 2821 * If there is more data to be acked, restart retransmit 2822 * timer, using current (possibly backed-off) value. 2823 */ 2824 if (th->th_ack == tp->snd_max) { 2825 tcp_timer_activate(tp, TT_REXMT, 0); 2826 needoutput = 1; 2827 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2828 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2829 2830 /* 2831 * If no data (only SYN) was ACK'd, 2832 * skip rest of ACK processing. 2833 */ 2834 if (acked == 0) 2835 goto step6; 2836 2837 /* 2838 * Let the congestion control algorithm update congestion 2839 * control related information. This typically means increasing 2840 * the congestion window. 2841 */ 2842 cc_ack_received(tp, th, nsegs, CC_ACK); 2843 2844 SOCKBUF_LOCK(&so->so_snd); 2845 if (acked > sbavail(&so->so_snd)) { 2846 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2847 tp->snd_wnd -= sbavail(&so->so_snd); 2848 else 2849 tp->snd_wnd = 0; 2850 mfree = sbcut_locked(&so->so_snd, 2851 (int)sbavail(&so->so_snd)); 2852 ourfinisacked = 1; 2853 } else { 2854 mfree = sbcut_locked(&so->so_snd, acked); 2855 if (tp->snd_wnd >= (uint32_t) acked) 2856 tp->snd_wnd -= acked; 2857 else 2858 tp->snd_wnd = 0; 2859 ourfinisacked = 0; 2860 } 2861 /* NB: sowwakeup_locked() does an implicit unlock. */ 2862 sowwakeup_locked(so); 2863 m_freem(mfree); 2864 /* Detect una wraparound. */ 2865 if (!IN_RECOVERY(tp->t_flags) && 2866 SEQ_GT(tp->snd_una, tp->snd_recover) && 2867 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2868 tp->snd_recover = th->th_ack - 1; 2869 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2870 if (IN_RECOVERY(tp->t_flags) && 2871 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2872 EXIT_RECOVERY(tp->t_flags); 2873 } 2874 tp->snd_una = th->th_ack; 2875 if (tp->t_flags & TF_SACK_PERMIT) { 2876 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2877 tp->snd_recover = tp->snd_una; 2878 } 2879 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2880 tp->snd_nxt = tp->snd_una; 2881 2882 switch (tp->t_state) { 2883 2884 /* 2885 * In FIN_WAIT_1 STATE in addition to the processing 2886 * for the ESTABLISHED state if our FIN is now acknowledged 2887 * then enter FIN_WAIT_2. 2888 */ 2889 case TCPS_FIN_WAIT_1: 2890 if (ourfinisacked) { 2891 /* 2892 * If we can't receive any more 2893 * data, then closing user can proceed. 2894 * Starting the timer is contrary to the 2895 * specification, but if we don't get a FIN 2896 * we'll hang forever. 2897 * 2898 * XXXjl: 2899 * we should release the tp also, and use a 2900 * compressed state. 2901 */ 2902 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2903 soisdisconnected(so); 2904 tcp_timer_activate(tp, TT_2MSL, 2905 (tcp_fast_finwait2_recycle ? 2906 tcp_finwait2_timeout : 2907 TP_MAXIDLE(tp))); 2908 } 2909 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2910 } 2911 break; 2912 2913 /* 2914 * In CLOSING STATE in addition to the processing for 2915 * the ESTABLISHED state if the ACK acknowledges our FIN 2916 * then enter the TIME-WAIT state, otherwise ignore 2917 * the segment. 2918 */ 2919 case TCPS_CLOSING: 2920 if (ourfinisacked) { 2921 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2922 tcp_twstart(tp); 2923 INP_INFO_RUNLOCK(&V_tcbinfo); 2924 m_freem(m); 2925 return; 2926 } 2927 break; 2928 2929 /* 2930 * In LAST_ACK, we may still be waiting for data to drain 2931 * and/or to be acked, as well as for the ack of our FIN. 2932 * If our FIN is now acknowledged, delete the TCB, 2933 * enter the closed state and return. 2934 */ 2935 case TCPS_LAST_ACK: 2936 if (ourfinisacked) { 2937 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2938 tp = tcp_close(tp); 2939 goto drop; 2940 } 2941 break; 2942 } 2943 } 2944 2945 step6: 2946 INP_WLOCK_ASSERT(tp->t_inpcb); 2947 2948 /* 2949 * Update window information. 2950 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2951 */ 2952 if ((thflags & TH_ACK) && 2953 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2954 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2955 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2956 /* keep track of pure window updates */ 2957 if (tlen == 0 && 2958 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2959 TCPSTAT_INC(tcps_rcvwinupd); 2960 tp->snd_wnd = tiwin; 2961 tp->snd_wl1 = th->th_seq; 2962 tp->snd_wl2 = th->th_ack; 2963 if (tp->snd_wnd > tp->max_sndwnd) 2964 tp->max_sndwnd = tp->snd_wnd; 2965 needoutput = 1; 2966 } 2967 2968 /* 2969 * Process segments with URG. 2970 */ 2971 if ((thflags & TH_URG) && th->th_urp && 2972 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2973 /* 2974 * This is a kludge, but if we receive and accept 2975 * random urgent pointers, we'll crash in 2976 * soreceive. It's hard to imagine someone 2977 * actually wanting to send this much urgent data. 2978 */ 2979 SOCKBUF_LOCK(&so->so_rcv); 2980 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2981 th->th_urp = 0; /* XXX */ 2982 thflags &= ~TH_URG; /* XXX */ 2983 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2984 goto dodata; /* XXX */ 2985 } 2986 /* 2987 * If this segment advances the known urgent pointer, 2988 * then mark the data stream. This should not happen 2989 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2990 * a FIN has been received from the remote side. 2991 * In these states we ignore the URG. 2992 * 2993 * According to RFC961 (Assigned Protocols), 2994 * the urgent pointer points to the last octet 2995 * of urgent data. We continue, however, 2996 * to consider it to indicate the first octet 2997 * of data past the urgent section as the original 2998 * spec states (in one of two places). 2999 */ 3000 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3001 tp->rcv_up = th->th_seq + th->th_urp; 3002 so->so_oobmark = sbavail(&so->so_rcv) + 3003 (tp->rcv_up - tp->rcv_nxt) - 1; 3004 if (so->so_oobmark == 0) 3005 so->so_rcv.sb_state |= SBS_RCVATMARK; 3006 sohasoutofband(so); 3007 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3008 } 3009 SOCKBUF_UNLOCK(&so->so_rcv); 3010 /* 3011 * Remove out of band data so doesn't get presented to user. 3012 * This can happen independent of advancing the URG pointer, 3013 * but if two URG's are pending at once, some out-of-band 3014 * data may creep in... ick. 3015 */ 3016 if (th->th_urp <= (uint32_t)tlen && 3017 !(so->so_options & SO_OOBINLINE)) { 3018 /* hdr drop is delayed */ 3019 tcp_pulloutofband(so, th, m, drop_hdrlen); 3020 } 3021 } else { 3022 /* 3023 * If no out of band data is expected, 3024 * pull receive urgent pointer along 3025 * with the receive window. 3026 */ 3027 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3028 tp->rcv_up = tp->rcv_nxt; 3029 } 3030 dodata: /* XXX */ 3031 INP_WLOCK_ASSERT(tp->t_inpcb); 3032 3033 /* 3034 * Process the segment text, merging it into the TCP sequencing queue, 3035 * and arranging for acknowledgment of receipt if necessary. 3036 * This process logically involves adjusting tp->rcv_wnd as data 3037 * is presented to the user (this happens in tcp_usrreq.c, 3038 * case PRU_RCVD). If a FIN has already been received on this 3039 * connection then we just ignore the text. 3040 */ 3041 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3042 IS_FASTOPEN(tp->t_flags)); 3043 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3044 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3045 tcp_seq save_start = th->th_seq; 3046 m_adj(m, drop_hdrlen); /* delayed header drop */ 3047 /* 3048 * Insert segment which includes th into TCP reassembly queue 3049 * with control block tp. Set thflags to whether reassembly now 3050 * includes a segment with FIN. This handles the common case 3051 * inline (segment is the next to be received on an established 3052 * connection, and the queue is empty), avoiding linkage into 3053 * and removal from the queue and repetition of various 3054 * conversions. 3055 * Set DELACK for segments received in order, but ack 3056 * immediately when segments are out of order (so 3057 * fast retransmit can work). 3058 */ 3059 if (th->th_seq == tp->rcv_nxt && 3060 LIST_EMPTY(&tp->t_segq) && 3061 (TCPS_HAVEESTABLISHED(tp->t_state) || 3062 tfo_syn)) { 3063 if (DELAY_ACK(tp, tlen) || tfo_syn) 3064 tp->t_flags |= TF_DELACK; 3065 else 3066 tp->t_flags |= TF_ACKNOW; 3067 tp->rcv_nxt += tlen; 3068 thflags = th->th_flags & TH_FIN; 3069 TCPSTAT_INC(tcps_rcvpack); 3070 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3071 SOCKBUF_LOCK(&so->so_rcv); 3072 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3073 m_freem(m); 3074 else 3075 sbappendstream_locked(&so->so_rcv, m, 0); 3076 /* NB: sorwakeup_locked() does an implicit unlock. */ 3077 sorwakeup_locked(so); 3078 } else { 3079 /* 3080 * XXX: Due to the header drop above "th" is 3081 * theoretically invalid by now. Fortunately 3082 * m_adj() doesn't actually frees any mbufs 3083 * when trimming from the head. 3084 */ 3085 thflags = tcp_reass(tp, th, &tlen, m); 3086 tp->t_flags |= TF_ACKNOW; 3087 } 3088 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3089 tcp_update_sack_list(tp, save_start, save_start + tlen); 3090 #if 0 3091 /* 3092 * Note the amount of data that peer has sent into 3093 * our window, in order to estimate the sender's 3094 * buffer size. 3095 * XXX: Unused. 3096 */ 3097 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3098 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3099 else 3100 len = so->so_rcv.sb_hiwat; 3101 #endif 3102 } else { 3103 m_freem(m); 3104 thflags &= ~TH_FIN; 3105 } 3106 3107 /* 3108 * If FIN is received ACK the FIN and let the user know 3109 * that the connection is closing. 3110 */ 3111 if (thflags & TH_FIN) { 3112 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3113 socantrcvmore(so); 3114 /* 3115 * If connection is half-synchronized 3116 * (ie NEEDSYN flag on) then delay ACK, 3117 * so it may be piggybacked when SYN is sent. 3118 * Otherwise, since we received a FIN then no 3119 * more input can be expected, send ACK now. 3120 */ 3121 if (tp->t_flags & TF_NEEDSYN) 3122 tp->t_flags |= TF_DELACK; 3123 else 3124 tp->t_flags |= TF_ACKNOW; 3125 tp->rcv_nxt++; 3126 } 3127 switch (tp->t_state) { 3128 3129 /* 3130 * In SYN_RECEIVED and ESTABLISHED STATES 3131 * enter the CLOSE_WAIT state. 3132 */ 3133 case TCPS_SYN_RECEIVED: 3134 tp->t_starttime = ticks; 3135 /* FALLTHROUGH */ 3136 case TCPS_ESTABLISHED: 3137 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3138 break; 3139 3140 /* 3141 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3142 * enter the CLOSING state. 3143 */ 3144 case TCPS_FIN_WAIT_1: 3145 tcp_state_change(tp, TCPS_CLOSING); 3146 break; 3147 3148 /* 3149 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3150 * starting the time-wait timer, turning off the other 3151 * standard timers. 3152 */ 3153 case TCPS_FIN_WAIT_2: 3154 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3155 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3156 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3157 ti_locked)); 3158 3159 tcp_twstart(tp); 3160 INP_INFO_RUNLOCK(&V_tcbinfo); 3161 return; 3162 } 3163 } 3164 if (ti_locked == TI_RLOCKED) 3165 INP_INFO_RUNLOCK(&V_tcbinfo); 3166 ti_locked = TI_UNLOCKED; 3167 3168 #ifdef TCPDEBUG 3169 if (so->so_options & SO_DEBUG) 3170 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3171 &tcp_savetcp, 0); 3172 #endif 3173 TCP_PROBE3(debug__input, tp, th, m); 3174 3175 /* 3176 * Return any desired output. 3177 */ 3178 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3179 (void) tp->t_fb->tfb_tcp_output(tp); 3180 3181 check_delack: 3182 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3183 __func__, ti_locked)); 3184 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3185 INP_WLOCK_ASSERT(tp->t_inpcb); 3186 3187 if (tp->t_flags & TF_DELACK) { 3188 tp->t_flags &= ~TF_DELACK; 3189 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3190 } 3191 INP_WUNLOCK(tp->t_inpcb); 3192 return; 3193 3194 dropafterack: 3195 /* 3196 * Generate an ACK dropping incoming segment if it occupies 3197 * sequence space, where the ACK reflects our state. 3198 * 3199 * We can now skip the test for the RST flag since all 3200 * paths to this code happen after packets containing 3201 * RST have been dropped. 3202 * 3203 * In the SYN-RECEIVED state, don't send an ACK unless the 3204 * segment we received passes the SYN-RECEIVED ACK test. 3205 * If it fails send a RST. This breaks the loop in the 3206 * "LAND" DoS attack, and also prevents an ACK storm 3207 * between two listening ports that have been sent forged 3208 * SYN segments, each with the source address of the other. 3209 */ 3210 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3211 (SEQ_GT(tp->snd_una, th->th_ack) || 3212 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3213 rstreason = BANDLIM_RST_OPENPORT; 3214 goto dropwithreset; 3215 } 3216 #ifdef TCPDEBUG 3217 if (so->so_options & SO_DEBUG) 3218 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3219 &tcp_savetcp, 0); 3220 #endif 3221 TCP_PROBE3(debug__input, tp, th, m); 3222 if (ti_locked == TI_RLOCKED) 3223 INP_INFO_RUNLOCK(&V_tcbinfo); 3224 ti_locked = TI_UNLOCKED; 3225 3226 tp->t_flags |= TF_ACKNOW; 3227 (void) tp->t_fb->tfb_tcp_output(tp); 3228 INP_WUNLOCK(tp->t_inpcb); 3229 m_freem(m); 3230 return; 3231 3232 dropwithreset: 3233 if (ti_locked == TI_RLOCKED) 3234 INP_INFO_RUNLOCK(&V_tcbinfo); 3235 ti_locked = TI_UNLOCKED; 3236 3237 if (tp != NULL) { 3238 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3239 INP_WUNLOCK(tp->t_inpcb); 3240 } else 3241 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3242 return; 3243 3244 drop: 3245 if (ti_locked == TI_RLOCKED) { 3246 INP_INFO_RUNLOCK(&V_tcbinfo); 3247 ti_locked = TI_UNLOCKED; 3248 } 3249 #ifdef INVARIANTS 3250 else 3251 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3252 #endif 3253 3254 /* 3255 * Drop space held by incoming segment and return. 3256 */ 3257 #ifdef TCPDEBUG 3258 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3259 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3260 &tcp_savetcp, 0); 3261 #endif 3262 TCP_PROBE3(debug__input, tp, th, m); 3263 if (tp != NULL) 3264 INP_WUNLOCK(tp->t_inpcb); 3265 m_freem(m); 3266 } 3267 3268 /* 3269 * Issue RST and make ACK acceptable to originator of segment. 3270 * The mbuf must still include the original packet header. 3271 * tp may be NULL. 3272 */ 3273 void 3274 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3275 int tlen, int rstreason) 3276 { 3277 #ifdef INET 3278 struct ip *ip; 3279 #endif 3280 #ifdef INET6 3281 struct ip6_hdr *ip6; 3282 #endif 3283 3284 if (tp != NULL) { 3285 INP_WLOCK_ASSERT(tp->t_inpcb); 3286 } 3287 3288 /* Don't bother if destination was broadcast/multicast. */ 3289 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3290 goto drop; 3291 #ifdef INET6 3292 if (mtod(m, struct ip *)->ip_v == 6) { 3293 ip6 = mtod(m, struct ip6_hdr *); 3294 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3295 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3296 goto drop; 3297 /* IPv6 anycast check is done at tcp6_input() */ 3298 } 3299 #endif 3300 #if defined(INET) && defined(INET6) 3301 else 3302 #endif 3303 #ifdef INET 3304 { 3305 ip = mtod(m, struct ip *); 3306 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3307 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3308 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3309 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3310 goto drop; 3311 } 3312 #endif 3313 3314 /* Perform bandwidth limiting. */ 3315 if (badport_bandlim(rstreason) < 0) 3316 goto drop; 3317 3318 /* tcp_respond consumes the mbuf chain. */ 3319 if (th->th_flags & TH_ACK) { 3320 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3321 th->th_ack, TH_RST); 3322 } else { 3323 if (th->th_flags & TH_SYN) 3324 tlen++; 3325 if (th->th_flags & TH_FIN) 3326 tlen++; 3327 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3328 (tcp_seq)0, TH_RST|TH_ACK); 3329 } 3330 return; 3331 drop: 3332 m_freem(m); 3333 } 3334 3335 /* 3336 * Parse TCP options and place in tcpopt. 3337 */ 3338 void 3339 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3340 { 3341 int opt, optlen; 3342 3343 to->to_flags = 0; 3344 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3345 opt = cp[0]; 3346 if (opt == TCPOPT_EOL) 3347 break; 3348 if (opt == TCPOPT_NOP) 3349 optlen = 1; 3350 else { 3351 if (cnt < 2) 3352 break; 3353 optlen = cp[1]; 3354 if (optlen < 2 || optlen > cnt) 3355 break; 3356 } 3357 switch (opt) { 3358 case TCPOPT_MAXSEG: 3359 if (optlen != TCPOLEN_MAXSEG) 3360 continue; 3361 if (!(flags & TO_SYN)) 3362 continue; 3363 to->to_flags |= TOF_MSS; 3364 bcopy((char *)cp + 2, 3365 (char *)&to->to_mss, sizeof(to->to_mss)); 3366 to->to_mss = ntohs(to->to_mss); 3367 break; 3368 case TCPOPT_WINDOW: 3369 if (optlen != TCPOLEN_WINDOW) 3370 continue; 3371 if (!(flags & TO_SYN)) 3372 continue; 3373 to->to_flags |= TOF_SCALE; 3374 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3375 break; 3376 case TCPOPT_TIMESTAMP: 3377 if (optlen != TCPOLEN_TIMESTAMP) 3378 continue; 3379 to->to_flags |= TOF_TS; 3380 bcopy((char *)cp + 2, 3381 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3382 to->to_tsval = ntohl(to->to_tsval); 3383 bcopy((char *)cp + 6, 3384 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3385 to->to_tsecr = ntohl(to->to_tsecr); 3386 break; 3387 case TCPOPT_SIGNATURE: 3388 /* 3389 * In order to reply to a host which has set the 3390 * TCP_SIGNATURE option in its initial SYN, we have 3391 * to record the fact that the option was observed 3392 * here for the syncache code to perform the correct 3393 * response. 3394 */ 3395 if (optlen != TCPOLEN_SIGNATURE) 3396 continue; 3397 to->to_flags |= TOF_SIGNATURE; 3398 to->to_signature = cp + 2; 3399 break; 3400 case TCPOPT_SACK_PERMITTED: 3401 if (optlen != TCPOLEN_SACK_PERMITTED) 3402 continue; 3403 if (!(flags & TO_SYN)) 3404 continue; 3405 if (!V_tcp_do_sack) 3406 continue; 3407 to->to_flags |= TOF_SACKPERM; 3408 break; 3409 case TCPOPT_SACK: 3410 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3411 continue; 3412 if (flags & TO_SYN) 3413 continue; 3414 to->to_flags |= TOF_SACK; 3415 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3416 to->to_sacks = cp + 2; 3417 TCPSTAT_INC(tcps_sack_rcv_blocks); 3418 break; 3419 case TCPOPT_FAST_OPEN: 3420 /* 3421 * Cookie length validation is performed by the 3422 * server side cookie checking code or the client 3423 * side cookie cache update code. 3424 */ 3425 if (!(flags & TO_SYN)) 3426 continue; 3427 if (!V_tcp_fastopen_client_enable && 3428 !V_tcp_fastopen_server_enable) 3429 continue; 3430 to->to_flags |= TOF_FASTOPEN; 3431 to->to_tfo_len = optlen - 2; 3432 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3433 break; 3434 default: 3435 continue; 3436 } 3437 } 3438 } 3439 3440 /* 3441 * Pull out of band byte out of a segment so 3442 * it doesn't appear in the user's data queue. 3443 * It is still reflected in the segment length for 3444 * sequencing purposes. 3445 */ 3446 void 3447 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3448 int off) 3449 { 3450 int cnt = off + th->th_urp - 1; 3451 3452 while (cnt >= 0) { 3453 if (m->m_len > cnt) { 3454 char *cp = mtod(m, caddr_t) + cnt; 3455 struct tcpcb *tp = sototcpcb(so); 3456 3457 INP_WLOCK_ASSERT(tp->t_inpcb); 3458 3459 tp->t_iobc = *cp; 3460 tp->t_oobflags |= TCPOOB_HAVEDATA; 3461 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3462 m->m_len--; 3463 if (m->m_flags & M_PKTHDR) 3464 m->m_pkthdr.len--; 3465 return; 3466 } 3467 cnt -= m->m_len; 3468 m = m->m_next; 3469 if (m == NULL) 3470 break; 3471 } 3472 panic("tcp_pulloutofband"); 3473 } 3474 3475 /* 3476 * Collect new round-trip time estimate 3477 * and update averages and current timeout. 3478 */ 3479 void 3480 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3481 { 3482 int delta; 3483 3484 INP_WLOCK_ASSERT(tp->t_inpcb); 3485 3486 TCPSTAT_INC(tcps_rttupdated); 3487 tp->t_rttupdated++; 3488 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3489 /* 3490 * srtt is stored as fixed point with 5 bits after the 3491 * binary point (i.e., scaled by 8). The following magic 3492 * is equivalent to the smoothing algorithm in rfc793 with 3493 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3494 * point). Adjust rtt to origin 0. 3495 */ 3496 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3497 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3498 3499 if ((tp->t_srtt += delta) <= 0) 3500 tp->t_srtt = 1; 3501 3502 /* 3503 * We accumulate a smoothed rtt variance (actually, a 3504 * smoothed mean difference), then set the retransmit 3505 * timer to smoothed rtt + 4 times the smoothed variance. 3506 * rttvar is stored as fixed point with 4 bits after the 3507 * binary point (scaled by 16). The following is 3508 * equivalent to rfc793 smoothing with an alpha of .75 3509 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3510 * rfc793's wired-in beta. 3511 */ 3512 if (delta < 0) 3513 delta = -delta; 3514 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3515 if ((tp->t_rttvar += delta) <= 0) 3516 tp->t_rttvar = 1; 3517 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3518 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3519 } else { 3520 /* 3521 * No rtt measurement yet - use the unsmoothed rtt. 3522 * Set the variance to half the rtt (so our first 3523 * retransmit happens at 3*rtt). 3524 */ 3525 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3526 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3527 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3528 } 3529 tp->t_rtttime = 0; 3530 tp->t_rxtshift = 0; 3531 3532 /* 3533 * the retransmit should happen at rtt + 4 * rttvar. 3534 * Because of the way we do the smoothing, srtt and rttvar 3535 * will each average +1/2 tick of bias. When we compute 3536 * the retransmit timer, we want 1/2 tick of rounding and 3537 * 1 extra tick because of +-1/2 tick uncertainty in the 3538 * firing of the timer. The bias will give us exactly the 3539 * 1.5 tick we need. But, because the bias is 3540 * statistical, we have to test that we don't drop below 3541 * the minimum feasible timer (which is 2 ticks). 3542 */ 3543 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3544 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3545 3546 /* 3547 * We received an ack for a packet that wasn't retransmitted; 3548 * it is probably safe to discard any error indications we've 3549 * received recently. This isn't quite right, but close enough 3550 * for now (a route might have failed after we sent a segment, 3551 * and the return path might not be symmetrical). 3552 */ 3553 tp->t_softerror = 0; 3554 } 3555 3556 /* 3557 * Determine a reasonable value for maxseg size. 3558 * If the route is known, check route for mtu. 3559 * If none, use an mss that can be handled on the outgoing interface 3560 * without forcing IP to fragment. If no route is found, route has no mtu, 3561 * or the destination isn't local, use a default, hopefully conservative 3562 * size (usually 512 or the default IP max size, but no more than the mtu 3563 * of the interface), as we can't discover anything about intervening 3564 * gateways or networks. We also initialize the congestion/slow start 3565 * window to be a single segment if the destination isn't local. 3566 * While looking at the routing entry, we also initialize other path-dependent 3567 * parameters from pre-set or cached values in the routing entry. 3568 * 3569 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3570 * IP options, e.g. IPSEC data, since length of this data may vary, and 3571 * thus it is calculated for every segment separately in tcp_output(). 3572 * 3573 * NOTE that this routine is only called when we process an incoming 3574 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3575 * settings are handled in tcp_mssopt(). 3576 */ 3577 void 3578 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3579 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3580 { 3581 int mss = 0; 3582 uint32_t maxmtu = 0; 3583 struct inpcb *inp = tp->t_inpcb; 3584 struct hc_metrics_lite metrics; 3585 #ifdef INET6 3586 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3587 size_t min_protoh = isipv6 ? 3588 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3589 sizeof (struct tcpiphdr); 3590 #else 3591 const size_t min_protoh = sizeof(struct tcpiphdr); 3592 #endif 3593 3594 INP_WLOCK_ASSERT(tp->t_inpcb); 3595 3596 if (mtuoffer != -1) { 3597 KASSERT(offer == -1, ("%s: conflict", __func__)); 3598 offer = mtuoffer - min_protoh; 3599 } 3600 3601 /* Initialize. */ 3602 #ifdef INET6 3603 if (isipv6) { 3604 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3605 tp->t_maxseg = V_tcp_v6mssdflt; 3606 } 3607 #endif 3608 #if defined(INET) && defined(INET6) 3609 else 3610 #endif 3611 #ifdef INET 3612 { 3613 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3614 tp->t_maxseg = V_tcp_mssdflt; 3615 } 3616 #endif 3617 3618 /* 3619 * No route to sender, stay with default mss and return. 3620 */ 3621 if (maxmtu == 0) { 3622 /* 3623 * In case we return early we need to initialize metrics 3624 * to a defined state as tcp_hc_get() would do for us 3625 * if there was no cache hit. 3626 */ 3627 if (metricptr != NULL) 3628 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3629 return; 3630 } 3631 3632 /* What have we got? */ 3633 switch (offer) { 3634 case 0: 3635 /* 3636 * Offer == 0 means that there was no MSS on the SYN 3637 * segment, in this case we use tcp_mssdflt as 3638 * already assigned to t_maxseg above. 3639 */ 3640 offer = tp->t_maxseg; 3641 break; 3642 3643 case -1: 3644 /* 3645 * Offer == -1 means that we didn't receive SYN yet. 3646 */ 3647 /* FALLTHROUGH */ 3648 3649 default: 3650 /* 3651 * Prevent DoS attack with too small MSS. Round up 3652 * to at least minmss. 3653 */ 3654 offer = max(offer, V_tcp_minmss); 3655 } 3656 3657 /* 3658 * rmx information is now retrieved from tcp_hostcache. 3659 */ 3660 tcp_hc_get(&inp->inp_inc, &metrics); 3661 if (metricptr != NULL) 3662 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3663 3664 /* 3665 * If there's a discovered mtu in tcp hostcache, use it. 3666 * Else, use the link mtu. 3667 */ 3668 if (metrics.rmx_mtu) 3669 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3670 else { 3671 #ifdef INET6 3672 if (isipv6) { 3673 mss = maxmtu - min_protoh; 3674 if (!V_path_mtu_discovery && 3675 !in6_localaddr(&inp->in6p_faddr)) 3676 mss = min(mss, V_tcp_v6mssdflt); 3677 } 3678 #endif 3679 #if defined(INET) && defined(INET6) 3680 else 3681 #endif 3682 #ifdef INET 3683 { 3684 mss = maxmtu - min_protoh; 3685 if (!V_path_mtu_discovery && 3686 !in_localaddr(inp->inp_faddr)) 3687 mss = min(mss, V_tcp_mssdflt); 3688 } 3689 #endif 3690 /* 3691 * XXX - The above conditional (mss = maxmtu - min_protoh) 3692 * probably violates the TCP spec. 3693 * The problem is that, since we don't know the 3694 * other end's MSS, we are supposed to use a conservative 3695 * default. But, if we do that, then MTU discovery will 3696 * never actually take place, because the conservative 3697 * default is much less than the MTUs typically seen 3698 * on the Internet today. For the moment, we'll sweep 3699 * this under the carpet. 3700 * 3701 * The conservative default might not actually be a problem 3702 * if the only case this occurs is when sending an initial 3703 * SYN with options and data to a host we've never talked 3704 * to before. Then, they will reply with an MSS value which 3705 * will get recorded and the new parameters should get 3706 * recomputed. For Further Study. 3707 */ 3708 } 3709 mss = min(mss, offer); 3710 3711 /* 3712 * Sanity check: make sure that maxseg will be large 3713 * enough to allow some data on segments even if the 3714 * all the option space is used (40bytes). Otherwise 3715 * funny things may happen in tcp_output. 3716 * 3717 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3718 */ 3719 mss = max(mss, 64); 3720 3721 tp->t_maxseg = mss; 3722 } 3723 3724 void 3725 tcp_mss(struct tcpcb *tp, int offer) 3726 { 3727 int mss; 3728 uint32_t bufsize; 3729 struct inpcb *inp; 3730 struct socket *so; 3731 struct hc_metrics_lite metrics; 3732 struct tcp_ifcap cap; 3733 3734 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3735 3736 bzero(&cap, sizeof(cap)); 3737 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3738 3739 mss = tp->t_maxseg; 3740 inp = tp->t_inpcb; 3741 3742 /* 3743 * If there's a pipesize, change the socket buffer to that size, 3744 * don't change if sb_hiwat is different than default (then it 3745 * has been changed on purpose with setsockopt). 3746 * Make the socket buffers an integral number of mss units; 3747 * if the mss is larger than the socket buffer, decrease the mss. 3748 */ 3749 so = inp->inp_socket; 3750 SOCKBUF_LOCK(&so->so_snd); 3751 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3752 bufsize = metrics.rmx_sendpipe; 3753 else 3754 bufsize = so->so_snd.sb_hiwat; 3755 if (bufsize < mss) 3756 mss = bufsize; 3757 else { 3758 bufsize = roundup(bufsize, mss); 3759 if (bufsize > sb_max) 3760 bufsize = sb_max; 3761 if (bufsize > so->so_snd.sb_hiwat) 3762 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3763 } 3764 SOCKBUF_UNLOCK(&so->so_snd); 3765 /* 3766 * Sanity check: make sure that maxseg will be large 3767 * enough to allow some data on segments even if the 3768 * all the option space is used (40bytes). Otherwise 3769 * funny things may happen in tcp_output. 3770 * 3771 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3772 */ 3773 tp->t_maxseg = max(mss, 64); 3774 3775 SOCKBUF_LOCK(&so->so_rcv); 3776 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3777 bufsize = metrics.rmx_recvpipe; 3778 else 3779 bufsize = so->so_rcv.sb_hiwat; 3780 if (bufsize > mss) { 3781 bufsize = roundup(bufsize, mss); 3782 if (bufsize > sb_max) 3783 bufsize = sb_max; 3784 if (bufsize > so->so_rcv.sb_hiwat) 3785 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3786 } 3787 SOCKBUF_UNLOCK(&so->so_rcv); 3788 3789 /* Check the interface for TSO capabilities. */ 3790 if (cap.ifcap & CSUM_TSO) { 3791 tp->t_flags |= TF_TSO; 3792 tp->t_tsomax = cap.tsomax; 3793 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3794 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3795 } 3796 } 3797 3798 /* 3799 * Determine the MSS option to send on an outgoing SYN. 3800 */ 3801 int 3802 tcp_mssopt(struct in_conninfo *inc) 3803 { 3804 int mss = 0; 3805 uint32_t thcmtu = 0; 3806 uint32_t maxmtu = 0; 3807 size_t min_protoh; 3808 3809 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3810 3811 #ifdef INET6 3812 if (inc->inc_flags & INC_ISIPV6) { 3813 mss = V_tcp_v6mssdflt; 3814 maxmtu = tcp_maxmtu6(inc, NULL); 3815 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3816 } 3817 #endif 3818 #if defined(INET) && defined(INET6) 3819 else 3820 #endif 3821 #ifdef INET 3822 { 3823 mss = V_tcp_mssdflt; 3824 maxmtu = tcp_maxmtu(inc, NULL); 3825 min_protoh = sizeof(struct tcpiphdr); 3826 } 3827 #endif 3828 #if defined(INET6) || defined(INET) 3829 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3830 #endif 3831 3832 if (maxmtu && thcmtu) 3833 mss = min(maxmtu, thcmtu) - min_protoh; 3834 else if (maxmtu || thcmtu) 3835 mss = max(maxmtu, thcmtu) - min_protoh; 3836 3837 return (mss); 3838 } 3839 3840 3841 /* 3842 * On a partial ack arrives, force the retransmission of the 3843 * next unacknowledged segment. Do not clear tp->t_dupacks. 3844 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3845 * be started again. 3846 */ 3847 void 3848 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3849 { 3850 tcp_seq onxt = tp->snd_nxt; 3851 uint32_t ocwnd = tp->snd_cwnd; 3852 u_int maxseg = tcp_maxseg(tp); 3853 3854 INP_WLOCK_ASSERT(tp->t_inpcb); 3855 3856 tcp_timer_activate(tp, TT_REXMT, 0); 3857 tp->t_rtttime = 0; 3858 tp->snd_nxt = th->th_ack; 3859 /* 3860 * Set snd_cwnd to one segment beyond acknowledged offset. 3861 * (tp->snd_una has not yet been updated when this function is called.) 3862 */ 3863 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3864 tp->t_flags |= TF_ACKNOW; 3865 (void) tp->t_fb->tfb_tcp_output(tp); 3866 tp->snd_cwnd = ocwnd; 3867 if (SEQ_GT(onxt, tp->snd_nxt)) 3868 tp->snd_nxt = onxt; 3869 /* 3870 * Partial window deflation. Relies on fact that tp->snd_una 3871 * not updated yet. 3872 */ 3873 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3874 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3875 else 3876 tp->snd_cwnd = 0; 3877 tp->snd_cwnd += maxseg; 3878 } 3879 3880 int 3881 tcp_compute_pipe(struct tcpcb *tp) 3882 { 3883 return (tp->snd_max - tp->snd_una + 3884 tp->sackhint.sack_bytes_rexmit - 3885 tp->sackhint.sacked_bytes); 3886 } 3887