1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 3. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_inet.h" 54 #include "opt_inet6.h" 55 #include "opt_ipsec.h" 56 #include "opt_tcpdebug.h" 57 58 #include <sys/param.h> 59 #include <sys/kernel.h> 60 #ifdef TCP_HHOOK 61 #include <sys/hhook.h> 62 #endif 63 #include <sys/malloc.h> 64 #include <sys/mbuf.h> 65 #include <sys/proc.h> /* for proc0 declaration */ 66 #include <sys/protosw.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77 #include <vm/uma.h> 78 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/route.h> 82 #include <net/vnet.h> 83 84 #define TCPSTATES /* for logging */ 85 86 #include <netinet/in.h> 87 #include <netinet/in_kdtrace.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/ip.h> 91 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 92 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 93 #include <netinet/ip_var.h> 94 #include <netinet/ip_options.h> 95 #include <netinet/ip6.h> 96 #include <netinet/icmp6.h> 97 #include <netinet6/in6_pcb.h> 98 #include <netinet6/in6_var.h> 99 #include <netinet6/ip6_var.h> 100 #include <netinet6/nd6.h> 101 #ifdef TCP_RFC7413 102 #include <netinet/tcp_fastopen.h> 103 #endif 104 #include <netinet/tcp.h> 105 #include <netinet/tcp_fsm.h> 106 #include <netinet/tcp_seq.h> 107 #include <netinet/tcp_timer.h> 108 #include <netinet/tcp_var.h> 109 #include <netinet6/tcp6_var.h> 110 #include <netinet/tcpip.h> 111 #include <netinet/cc/cc.h> 112 #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 #define V_drop_synfin VNET(drop_synfin) 149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 150 &VNET_NAME(drop_synfin), 0, 151 "Drop TCP packets with SYN+FIN set"); 152 153 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0; 154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW, 155 &VNET_NAME(tcp_do_rfc6675_pipe), 0, 156 "Use calculated pipe/in-flight bytes per RFC 6675"); 157 158 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 159 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 161 &VNET_NAME(tcp_do_rfc3042), 0, 162 "Enable RFC 3042 (Limited Transmit)"); 163 164 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 166 &VNET_NAME(tcp_do_rfc3390), 0, 167 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 168 169 VNET_DEFINE(int, tcp_initcwnd_segments) = 10; 170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, 171 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, 172 "Slow-start flight size (initial congestion window) in number of segments"); 173 174 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 176 &VNET_NAME(tcp_do_rfc3465), 0, 177 "Enable RFC 3465 (Appropriate Byte Counting)"); 178 179 VNET_DEFINE(int, tcp_abc_l_var) = 2; 180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_abc_l_var), 2, 182 "Cap the max cwnd increment during slow-start to this number of segments"); 183 184 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 185 186 VNET_DEFINE(int, tcp_do_ecn) = 2; 187 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 188 &VNET_NAME(tcp_do_ecn), 0, 189 "TCP ECN support"); 190 191 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 192 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 193 &VNET_NAME(tcp_ecn_maxretries), 0, 194 "Max retries before giving up on ECN"); 195 196 VNET_DEFINE(int, tcp_insecure_syn) = 0; 197 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 199 &VNET_NAME(tcp_insecure_syn), 0, 200 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 201 202 VNET_DEFINE(int, tcp_insecure_rst) = 0; 203 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 205 &VNET_NAME(tcp_insecure_rst), 0, 206 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 207 208 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 209 #define V_tcp_recvspace VNET(tcp_recvspace) 210 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 211 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 212 213 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 214 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 216 &VNET_NAME(tcp_do_autorcvbuf), 0, 217 "Enable automatic receive buffer sizing"); 218 219 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 220 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 222 &VNET_NAME(tcp_autorcvbuf_inc), 0, 223 "Incrementor step size of automatic receive buffer"); 224 225 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 226 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 228 &VNET_NAME(tcp_autorcvbuf_max), 0, 229 "Max size of automatic receive buffer"); 230 231 VNET_DEFINE(struct inpcbhead, tcb); 232 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 233 VNET_DEFINE(struct inpcbinfo, tcbinfo); 234 235 /* 236 * TCP statistics are stored in an array of counter(9)s, which size matches 237 * size of struct tcpstat. TCP running connection count is a regular array. 238 */ 239 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 240 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 241 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 242 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); 243 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | 244 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, 245 "TCP connection counts by TCP state"); 246 247 static void 248 tcp_vnet_init(const void *unused) 249 { 250 251 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK); 252 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK); 253 } 254 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 255 tcp_vnet_init, NULL); 256 257 #ifdef VIMAGE 258 static void 259 tcp_vnet_uninit(const void *unused) 260 { 261 262 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES); 263 VNET_PCPUSTAT_FREE(tcpstat); 264 } 265 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY, 266 tcp_vnet_uninit, NULL); 267 #endif /* VIMAGE */ 268 269 /* 270 * Kernel module interface for updating tcpstat. The argument is an index 271 * into tcpstat treated as an array. 272 */ 273 void 274 kmod_tcpstat_inc(int statnum) 275 { 276 277 counter_u64_add(VNET(tcpstat)[statnum], 1); 278 } 279 280 #ifdef TCP_HHOOK 281 /* 282 * Wrapper for the TCP established input helper hook. 283 */ 284 void 285 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 286 { 287 struct tcp_hhook_data hhook_data; 288 289 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 290 hhook_data.tp = tp; 291 hhook_data.th = th; 292 hhook_data.to = to; 293 294 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 295 tp->osd); 296 } 297 } 298 #endif 299 300 /* 301 * CC wrapper hook functions 302 */ 303 void 304 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, 305 uint16_t type) 306 { 307 INP_WLOCK_ASSERT(tp->t_inpcb); 308 309 tp->ccv->nsegs = nsegs; 310 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 311 if (tp->snd_cwnd <= tp->snd_wnd) 312 tp->ccv->flags |= CCF_CWND_LIMITED; 313 else 314 tp->ccv->flags &= ~CCF_CWND_LIMITED; 315 316 if (type == CC_ACK) { 317 if (tp->snd_cwnd > tp->snd_ssthresh) { 318 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 319 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp)); 320 if (tp->t_bytes_acked >= tp->snd_cwnd) { 321 tp->t_bytes_acked -= tp->snd_cwnd; 322 tp->ccv->flags |= CCF_ABC_SENTAWND; 323 } 324 } else { 325 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 326 tp->t_bytes_acked = 0; 327 } 328 } 329 330 if (CC_ALGO(tp)->ack_received != NULL) { 331 /* XXXLAS: Find a way to live without this */ 332 tp->ccv->curack = th->th_ack; 333 CC_ALGO(tp)->ack_received(tp->ccv, type); 334 } 335 } 336 337 void 338 cc_conn_init(struct tcpcb *tp) 339 { 340 struct hc_metrics_lite metrics; 341 struct inpcb *inp = tp->t_inpcb; 342 u_int maxseg; 343 int rtt; 344 345 INP_WLOCK_ASSERT(tp->t_inpcb); 346 347 tcp_hc_get(&inp->inp_inc, &metrics); 348 maxseg = tcp_maxseg(tp); 349 350 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 351 tp->t_srtt = rtt; 352 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 353 TCPSTAT_INC(tcps_usedrtt); 354 if (metrics.rmx_rttvar) { 355 tp->t_rttvar = metrics.rmx_rttvar; 356 TCPSTAT_INC(tcps_usedrttvar); 357 } else { 358 /* default variation is +- 1 rtt */ 359 tp->t_rttvar = 360 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 361 } 362 TCPT_RANGESET(tp->t_rxtcur, 363 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 364 tp->t_rttmin, TCPTV_REXMTMAX); 365 } 366 if (metrics.rmx_ssthresh) { 367 /* 368 * There's some sort of gateway or interface 369 * buffer limit on the path. Use this to set 370 * the slow start threshold, but set the 371 * threshold to no less than 2*mss. 372 */ 373 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); 374 TCPSTAT_INC(tcps_usedssthresh); 375 } 376 377 /* 378 * Set the initial slow-start flight size. 379 * 380 * RFC5681 Section 3.1 specifies the default conservative values. 381 * RFC3390 specifies slightly more aggressive values. 382 * RFC6928 increases it to ten segments. 383 * Support for user specified value for initial flight size. 384 * 385 * If a SYN or SYN/ACK was lost and retransmitted, we have to 386 * reduce the initial CWND to one segment as congestion is likely 387 * requiring us to be cautious. 388 */ 389 if (tp->snd_cwnd == 1) 390 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ 391 else if (V_tcp_initcwnd_segments) 392 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg, 393 max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); 394 else if (V_tcp_do_rfc3390) 395 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380)); 396 else { 397 /* Per RFC5681 Section 3.1 */ 398 if (maxseg > 2190) 399 tp->snd_cwnd = 2 * maxseg; 400 else if (maxseg > 1095) 401 tp->snd_cwnd = 3 * maxseg; 402 else 403 tp->snd_cwnd = 4 * maxseg; 404 } 405 406 if (CC_ALGO(tp)->conn_init != NULL) 407 CC_ALGO(tp)->conn_init(tp->ccv); 408 } 409 410 void inline 411 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 412 { 413 u_int maxseg; 414 415 INP_WLOCK_ASSERT(tp->t_inpcb); 416 417 switch(type) { 418 case CC_NDUPACK: 419 if (!IN_FASTRECOVERY(tp->t_flags)) { 420 tp->snd_recover = tp->snd_max; 421 if (tp->t_flags & TF_ECN_PERMIT) 422 tp->t_flags |= TF_ECN_SND_CWR; 423 } 424 break; 425 case CC_ECN: 426 if (!IN_CONGRECOVERY(tp->t_flags)) { 427 TCPSTAT_INC(tcps_ecn_rcwnd); 428 tp->snd_recover = tp->snd_max; 429 if (tp->t_flags & TF_ECN_PERMIT) 430 tp->t_flags |= TF_ECN_SND_CWR; 431 } 432 break; 433 case CC_RTO: 434 maxseg = tcp_maxseg(tp); 435 tp->t_dupacks = 0; 436 tp->t_bytes_acked = 0; 437 EXIT_RECOVERY(tp->t_flags); 438 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 439 maxseg) * maxseg; 440 tp->snd_cwnd = maxseg; 441 break; 442 case CC_RTO_ERR: 443 TCPSTAT_INC(tcps_sndrexmitbad); 444 /* RTO was unnecessary, so reset everything. */ 445 tp->snd_cwnd = tp->snd_cwnd_prev; 446 tp->snd_ssthresh = tp->snd_ssthresh_prev; 447 tp->snd_recover = tp->snd_recover_prev; 448 if (tp->t_flags & TF_WASFRECOVERY) 449 ENTER_FASTRECOVERY(tp->t_flags); 450 if (tp->t_flags & TF_WASCRECOVERY) 451 ENTER_CONGRECOVERY(tp->t_flags); 452 tp->snd_nxt = tp->snd_max; 453 tp->t_flags &= ~TF_PREVVALID; 454 tp->t_badrxtwin = 0; 455 break; 456 } 457 458 if (CC_ALGO(tp)->cong_signal != NULL) { 459 if (th != NULL) 460 tp->ccv->curack = th->th_ack; 461 CC_ALGO(tp)->cong_signal(tp->ccv, type); 462 } 463 } 464 465 void inline 466 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 467 { 468 INP_WLOCK_ASSERT(tp->t_inpcb); 469 470 /* XXXLAS: KASSERT that we're in recovery? */ 471 472 if (CC_ALGO(tp)->post_recovery != NULL) { 473 tp->ccv->curack = th->th_ack; 474 CC_ALGO(tp)->post_recovery(tp->ccv); 475 } 476 /* XXXLAS: EXIT_RECOVERY ? */ 477 tp->t_bytes_acked = 0; 478 } 479 480 /* 481 * Indicate whether this ack should be delayed. We can delay the ack if 482 * following conditions are met: 483 * - There is no delayed ack timer in progress. 484 * - Our last ack wasn't a 0-sized window. We never want to delay 485 * the ack that opens up a 0-sized window. 486 * - LRO wasn't used for this segment. We make sure by checking that the 487 * segment size is not larger than the MSS. 488 */ 489 #define DELAY_ACK(tp, tlen) \ 490 ((!tcp_timer_active(tp, TT_DELACK) && \ 491 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 492 (tlen <= tp->t_maxseg) && \ 493 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 494 495 static void inline 496 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 497 { 498 INP_WLOCK_ASSERT(tp->t_inpcb); 499 500 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 501 switch (iptos & IPTOS_ECN_MASK) { 502 case IPTOS_ECN_CE: 503 tp->ccv->flags |= CCF_IPHDR_CE; 504 break; 505 case IPTOS_ECN_ECT0: 506 tp->ccv->flags &= ~CCF_IPHDR_CE; 507 break; 508 case IPTOS_ECN_ECT1: 509 tp->ccv->flags &= ~CCF_IPHDR_CE; 510 break; 511 } 512 513 if (th->th_flags & TH_CWR) 514 tp->ccv->flags |= CCF_TCPHDR_CWR; 515 else 516 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 517 518 if (tp->t_flags & TF_DELACK) 519 tp->ccv->flags |= CCF_DELACK; 520 else 521 tp->ccv->flags &= ~CCF_DELACK; 522 523 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 524 525 if (tp->ccv->flags & CCF_ACKNOW) 526 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 527 } 528 } 529 530 /* 531 * TCP input handling is split into multiple parts: 532 * tcp6_input is a thin wrapper around tcp_input for the extended 533 * ip6_protox[] call format in ip6_input 534 * tcp_input handles primary segment validation, inpcb lookup and 535 * SYN processing on listen sockets 536 * tcp_do_segment processes the ACK and text of the segment for 537 * establishing, established and closing connections 538 */ 539 #ifdef INET6 540 int 541 tcp6_input(struct mbuf **mp, int *offp, int proto) 542 { 543 struct mbuf *m = *mp; 544 struct in6_ifaddr *ia6; 545 struct ip6_hdr *ip6; 546 547 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 548 549 /* 550 * draft-itojun-ipv6-tcp-to-anycast 551 * better place to put this in? 552 */ 553 ip6 = mtod(m, struct ip6_hdr *); 554 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 555 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 556 struct ip6_hdr *ip6; 557 558 ifa_free(&ia6->ia_ifa); 559 ip6 = mtod(m, struct ip6_hdr *); 560 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 561 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 562 return (IPPROTO_DONE); 563 } 564 if (ia6) 565 ifa_free(&ia6->ia_ifa); 566 567 return (tcp_input(mp, offp, proto)); 568 } 569 #endif /* INET6 */ 570 571 int 572 tcp_input(struct mbuf **mp, int *offp, int proto) 573 { 574 struct mbuf *m = *mp; 575 struct tcphdr *th = NULL; 576 struct ip *ip = NULL; 577 struct inpcb *inp = NULL; 578 struct tcpcb *tp = NULL; 579 struct socket *so = NULL; 580 u_char *optp = NULL; 581 int off0; 582 int optlen = 0; 583 #ifdef INET 584 int len; 585 #endif 586 int tlen = 0, off; 587 int drop_hdrlen; 588 int thflags; 589 int rstreason = 0; /* For badport_bandlim accounting purposes */ 590 uint8_t iptos; 591 struct m_tag *fwd_tag = NULL; 592 #ifdef INET6 593 struct ip6_hdr *ip6 = NULL; 594 int isipv6; 595 #else 596 const void *ip6 = NULL; 597 #endif /* INET6 */ 598 struct tcpopt to; /* options in this segment */ 599 char *s = NULL; /* address and port logging */ 600 int ti_locked; 601 #ifdef TCPDEBUG 602 /* 603 * The size of tcp_saveipgen must be the size of the max ip header, 604 * now IPv6. 605 */ 606 u_char tcp_saveipgen[IP6_HDR_LEN]; 607 struct tcphdr tcp_savetcp; 608 short ostate = 0; 609 #endif 610 611 #ifdef INET6 612 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 613 #endif 614 615 off0 = *offp; 616 m = *mp; 617 *mp = NULL; 618 to.to_flags = 0; 619 TCPSTAT_INC(tcps_rcvtotal); 620 621 #ifdef INET6 622 if (isipv6) { 623 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 624 625 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 626 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 627 if (m == NULL) { 628 TCPSTAT_INC(tcps_rcvshort); 629 return (IPPROTO_DONE); 630 } 631 } 632 633 ip6 = mtod(m, struct ip6_hdr *); 634 th = (struct tcphdr *)((caddr_t)ip6 + off0); 635 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 636 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 637 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 638 th->th_sum = m->m_pkthdr.csum_data; 639 else 640 th->th_sum = in6_cksum_pseudo(ip6, tlen, 641 IPPROTO_TCP, m->m_pkthdr.csum_data); 642 th->th_sum ^= 0xffff; 643 } else 644 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 645 if (th->th_sum) { 646 TCPSTAT_INC(tcps_rcvbadsum); 647 goto drop; 648 } 649 650 /* 651 * Be proactive about unspecified IPv6 address in source. 652 * As we use all-zero to indicate unbounded/unconnected pcb, 653 * unspecified IPv6 address can be used to confuse us. 654 * 655 * Note that packets with unspecified IPv6 destination is 656 * already dropped in ip6_input. 657 */ 658 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 659 /* XXX stat */ 660 goto drop; 661 } 662 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 663 } 664 #endif 665 #if defined(INET) && defined(INET6) 666 else 667 #endif 668 #ifdef INET 669 { 670 /* 671 * Get IP and TCP header together in first mbuf. 672 * Note: IP leaves IP header in first mbuf. 673 */ 674 if (off0 > sizeof (struct ip)) { 675 ip_stripoptions(m); 676 off0 = sizeof(struct ip); 677 } 678 if (m->m_len < sizeof (struct tcpiphdr)) { 679 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 680 == NULL) { 681 TCPSTAT_INC(tcps_rcvshort); 682 return (IPPROTO_DONE); 683 } 684 } 685 ip = mtod(m, struct ip *); 686 th = (struct tcphdr *)((caddr_t)ip + off0); 687 tlen = ntohs(ip->ip_len) - off0; 688 689 iptos = ip->ip_tos; 690 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 691 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 692 th->th_sum = m->m_pkthdr.csum_data; 693 else 694 th->th_sum = in_pseudo(ip->ip_src.s_addr, 695 ip->ip_dst.s_addr, 696 htonl(m->m_pkthdr.csum_data + tlen + 697 IPPROTO_TCP)); 698 th->th_sum ^= 0xffff; 699 } else { 700 struct ipovly *ipov = (struct ipovly *)ip; 701 702 /* 703 * Checksum extended TCP header and data. 704 */ 705 len = off0 + tlen; 706 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 707 ipov->ih_len = htons(tlen); 708 th->th_sum = in_cksum(m, len); 709 /* Reset length for SDT probes. */ 710 ip->ip_len = htons(len); 711 /* Reset TOS bits */ 712 ip->ip_tos = iptos; 713 /* Re-initialization for later version check */ 714 ip->ip_v = IPVERSION; 715 } 716 717 if (th->th_sum) { 718 TCPSTAT_INC(tcps_rcvbadsum); 719 goto drop; 720 } 721 } 722 #endif /* INET */ 723 724 /* 725 * Check that TCP offset makes sense, 726 * pull out TCP options and adjust length. XXX 727 */ 728 off = th->th_off << 2; 729 if (off < sizeof (struct tcphdr) || off > tlen) { 730 TCPSTAT_INC(tcps_rcvbadoff); 731 goto drop; 732 } 733 tlen -= off; /* tlen is used instead of ti->ti_len */ 734 if (off > sizeof (struct tcphdr)) { 735 #ifdef INET6 736 if (isipv6) { 737 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 738 ip6 = mtod(m, struct ip6_hdr *); 739 th = (struct tcphdr *)((caddr_t)ip6 + off0); 740 } 741 #endif 742 #if defined(INET) && defined(INET6) 743 else 744 #endif 745 #ifdef INET 746 { 747 if (m->m_len < sizeof(struct ip) + off) { 748 if ((m = m_pullup(m, sizeof (struct ip) + off)) 749 == NULL) { 750 TCPSTAT_INC(tcps_rcvshort); 751 return (IPPROTO_DONE); 752 } 753 ip = mtod(m, struct ip *); 754 th = (struct tcphdr *)((caddr_t)ip + off0); 755 } 756 } 757 #endif 758 optlen = off - sizeof (struct tcphdr); 759 optp = (u_char *)(th + 1); 760 } 761 thflags = th->th_flags; 762 763 /* 764 * Convert TCP protocol specific fields to host format. 765 */ 766 tcp_fields_to_host(th); 767 768 /* 769 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 770 */ 771 drop_hdrlen = off0 + off; 772 773 /* 774 * Locate pcb for segment; if we're likely to add or remove a 775 * connection then first acquire pcbinfo lock. There are three cases 776 * where we might discover later we need a write lock despite the 777 * flags: ACKs moving a connection out of the syncache, ACKs for a 778 * connection in TIMEWAIT and SYNs not targeting a listening socket. 779 */ 780 if ((thflags & (TH_FIN | TH_RST)) != 0) { 781 INP_INFO_RLOCK(&V_tcbinfo); 782 ti_locked = TI_RLOCKED; 783 } else 784 ti_locked = TI_UNLOCKED; 785 786 /* 787 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 788 */ 789 if ( 790 #ifdef INET6 791 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 792 #ifdef INET 793 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 794 #endif 795 #endif 796 #if defined(INET) && !defined(INET6) 797 (m->m_flags & M_IP_NEXTHOP) 798 #endif 799 ) 800 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 801 802 findpcb: 803 #ifdef INVARIANTS 804 if (ti_locked == TI_RLOCKED) { 805 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 806 } else { 807 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 808 } 809 #endif 810 #ifdef INET6 811 if (isipv6 && fwd_tag != NULL) { 812 struct sockaddr_in6 *next_hop6; 813 814 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 815 /* 816 * Transparently forwarded. Pretend to be the destination. 817 * Already got one like this? 818 */ 819 inp = in6_pcblookup_mbuf(&V_tcbinfo, 820 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 821 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 822 if (!inp) { 823 /* 824 * It's new. Try to find the ambushing socket. 825 * Because we've rewritten the destination address, 826 * any hardware-generated hash is ignored. 827 */ 828 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 829 th->th_sport, &next_hop6->sin6_addr, 830 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 831 th->th_dport, INPLOOKUP_WILDCARD | 832 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 833 } 834 } else if (isipv6) { 835 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 836 th->th_sport, &ip6->ip6_dst, th->th_dport, 837 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 838 m->m_pkthdr.rcvif, m); 839 } 840 #endif /* INET6 */ 841 #if defined(INET6) && defined(INET) 842 else 843 #endif 844 #ifdef INET 845 if (fwd_tag != NULL) { 846 struct sockaddr_in *next_hop; 847 848 next_hop = (struct sockaddr_in *)(fwd_tag+1); 849 /* 850 * Transparently forwarded. Pretend to be the destination. 851 * already got one like this? 852 */ 853 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 854 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 855 m->m_pkthdr.rcvif, m); 856 if (!inp) { 857 /* 858 * It's new. Try to find the ambushing socket. 859 * Because we've rewritten the destination address, 860 * any hardware-generated hash is ignored. 861 */ 862 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 863 th->th_sport, next_hop->sin_addr, 864 next_hop->sin_port ? ntohs(next_hop->sin_port) : 865 th->th_dport, INPLOOKUP_WILDCARD | 866 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 867 } 868 } else 869 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 870 th->th_sport, ip->ip_dst, th->th_dport, 871 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 872 m->m_pkthdr.rcvif, m); 873 #endif /* INET */ 874 875 /* 876 * If the INPCB does not exist then all data in the incoming 877 * segment is discarded and an appropriate RST is sent back. 878 * XXX MRT Send RST using which routing table? 879 */ 880 if (inp == NULL) { 881 /* 882 * Log communication attempts to ports that are not 883 * in use. 884 */ 885 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 886 tcp_log_in_vain == 2) { 887 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 888 log(LOG_INFO, "%s; %s: Connection attempt " 889 "to closed port\n", s, __func__); 890 } 891 /* 892 * When blackholing do not respond with a RST but 893 * completely ignore the segment and drop it. 894 */ 895 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 896 V_blackhole == 2) 897 goto dropunlock; 898 899 rstreason = BANDLIM_RST_CLOSEDPORT; 900 goto dropwithreset; 901 } 902 INP_WLOCK_ASSERT(inp); 903 /* 904 * While waiting for inp lock during the lookup, another thread 905 * can have dropped the inpcb, in which case we need to loop back 906 * and try to find a new inpcb to deliver to. 907 */ 908 if (inp->inp_flags & INP_DROPPED) { 909 INP_WUNLOCK(inp); 910 inp = NULL; 911 goto findpcb; 912 } 913 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 914 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 915 ((inp->inp_socket == NULL) || 916 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 917 inp->inp_flowid = m->m_pkthdr.flowid; 918 inp->inp_flowtype = M_HASHTYPE_GET(m); 919 } 920 #if defined(IPSEC) || defined(IPSEC_SUPPORT) 921 #ifdef INET6 922 if (isipv6 && IPSEC_ENABLED(ipv6) && 923 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { 924 goto dropunlock; 925 } 926 #ifdef INET 927 else 928 #endif 929 #endif /* INET6 */ 930 #ifdef INET 931 if (IPSEC_ENABLED(ipv4) && 932 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { 933 goto dropunlock; 934 } 935 #endif /* INET */ 936 #endif /* IPSEC */ 937 938 /* 939 * Check the minimum TTL for socket. 940 */ 941 if (inp->inp_ip_minttl != 0) { 942 #ifdef INET6 943 if (isipv6) { 944 if (inp->inp_ip_minttl > ip6->ip6_hlim) 945 goto dropunlock; 946 } else 947 #endif 948 if (inp->inp_ip_minttl > ip->ip_ttl) 949 goto dropunlock; 950 } 951 952 /* 953 * A previous connection in TIMEWAIT state is supposed to catch stray 954 * or duplicate segments arriving late. If this segment was a 955 * legitimate new connection attempt, the old INPCB gets removed and 956 * we can try again to find a listening socket. 957 * 958 * At this point, due to earlier optimism, we may hold only an inpcb 959 * lock, and not the inpcbinfo write lock. If so, we need to try to 960 * acquire it, or if that fails, acquire a reference on the inpcb, 961 * drop all locks, acquire a global write lock, and then re-acquire 962 * the inpcb lock. We may at that point discover that another thread 963 * has tried to free the inpcb, in which case we need to loop back 964 * and try to find a new inpcb to deliver to. 965 * 966 * XXXRW: It may be time to rethink timewait locking. 967 */ 968 relocked: 969 if (inp->inp_flags & INP_TIMEWAIT) { 970 if (ti_locked == TI_UNLOCKED) { 971 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 972 in_pcbref(inp); 973 INP_WUNLOCK(inp); 974 INP_INFO_RLOCK(&V_tcbinfo); 975 ti_locked = TI_RLOCKED; 976 INP_WLOCK(inp); 977 if (in_pcbrele_wlocked(inp)) { 978 inp = NULL; 979 goto findpcb; 980 } else if (inp->inp_flags & INP_DROPPED) { 981 INP_WUNLOCK(inp); 982 inp = NULL; 983 goto findpcb; 984 } 985 } else 986 ti_locked = TI_RLOCKED; 987 } 988 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 989 990 if (thflags & TH_SYN) 991 tcp_dooptions(&to, optp, optlen, TO_SYN); 992 /* 993 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 994 */ 995 if (tcp_twcheck(inp, &to, th, m, tlen)) 996 goto findpcb; 997 INP_INFO_RUNLOCK(&V_tcbinfo); 998 return (IPPROTO_DONE); 999 } 1000 /* 1001 * The TCPCB may no longer exist if the connection is winding 1002 * down or it is in the CLOSED state. Either way we drop the 1003 * segment and send an appropriate response. 1004 */ 1005 tp = intotcpcb(inp); 1006 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1007 rstreason = BANDLIM_RST_CLOSEDPORT; 1008 goto dropwithreset; 1009 } 1010 1011 #ifdef TCP_OFFLOAD 1012 if (tp->t_flags & TF_TOE) { 1013 tcp_offload_input(tp, m); 1014 m = NULL; /* consumed by the TOE driver */ 1015 goto dropunlock; 1016 } 1017 #endif 1018 1019 /* 1020 * We've identified a valid inpcb, but it could be that we need an 1021 * inpcbinfo write lock but don't hold it. In this case, attempt to 1022 * acquire using the same strategy as the TIMEWAIT case above. If we 1023 * relock, we have to jump back to 'relocked' as the connection might 1024 * now be in TIMEWAIT. 1025 */ 1026 #ifdef INVARIANTS 1027 if ((thflags & (TH_FIN | TH_RST)) != 0) 1028 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1029 #endif 1030 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1031 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) && 1032 !IS_FASTOPEN(tp->t_flags)))) { 1033 if (ti_locked == TI_UNLOCKED) { 1034 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1035 in_pcbref(inp); 1036 INP_WUNLOCK(inp); 1037 INP_INFO_RLOCK(&V_tcbinfo); 1038 ti_locked = TI_RLOCKED; 1039 INP_WLOCK(inp); 1040 if (in_pcbrele_wlocked(inp)) { 1041 inp = NULL; 1042 goto findpcb; 1043 } else if (inp->inp_flags & INP_DROPPED) { 1044 INP_WUNLOCK(inp); 1045 inp = NULL; 1046 goto findpcb; 1047 } 1048 goto relocked; 1049 } else 1050 ti_locked = TI_RLOCKED; 1051 } 1052 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1053 } 1054 1055 #ifdef MAC 1056 INP_WLOCK_ASSERT(inp); 1057 if (mac_inpcb_check_deliver(inp, m)) 1058 goto dropunlock; 1059 #endif 1060 so = inp->inp_socket; 1061 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1062 #ifdef TCPDEBUG 1063 if (so->so_options & SO_DEBUG) { 1064 ostate = tp->t_state; 1065 #ifdef INET6 1066 if (isipv6) { 1067 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1068 } else 1069 #endif 1070 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1071 tcp_savetcp = *th; 1072 } 1073 #endif /* TCPDEBUG */ 1074 /* 1075 * When the socket is accepting connections (the INPCB is in LISTEN 1076 * state) we look into the SYN cache if this is a new connection 1077 * attempt or the completion of a previous one. 1078 */ 1079 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN), 1080 ("%s: so accepting but tp %p not listening", __func__, tp)); 1081 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) { 1082 struct in_conninfo inc; 1083 1084 bzero(&inc, sizeof(inc)); 1085 #ifdef INET6 1086 if (isipv6) { 1087 inc.inc_flags |= INC_ISIPV6; 1088 inc.inc6_faddr = ip6->ip6_src; 1089 inc.inc6_laddr = ip6->ip6_dst; 1090 } else 1091 #endif 1092 { 1093 inc.inc_faddr = ip->ip_src; 1094 inc.inc_laddr = ip->ip_dst; 1095 } 1096 inc.inc_fport = th->th_sport; 1097 inc.inc_lport = th->th_dport; 1098 inc.inc_fibnum = so->so_fibnum; 1099 1100 /* 1101 * Check for an existing connection attempt in syncache if 1102 * the flag is only ACK. A successful lookup creates a new 1103 * socket appended to the listen queue in SYN_RECEIVED state. 1104 */ 1105 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1106 1107 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1108 /* 1109 * Parse the TCP options here because 1110 * syncookies need access to the reflected 1111 * timestamp. 1112 */ 1113 tcp_dooptions(&to, optp, optlen, 0); 1114 /* 1115 * NB: syncache_expand() doesn't unlock 1116 * inp and tcpinfo locks. 1117 */ 1118 rstreason = syncache_expand(&inc, &to, th, &so, m); 1119 if (rstreason < 0) { 1120 /* 1121 * A failing TCP MD5 signature comparison 1122 * must result in the segment being dropped 1123 * and must not produce any response back 1124 * to the sender. 1125 */ 1126 goto dropunlock; 1127 } else if (rstreason == 0) { 1128 /* 1129 * No syncache entry or ACK was not 1130 * for our SYN/ACK. Send a RST. 1131 * NB: syncache did its own logging 1132 * of the failure cause. 1133 */ 1134 rstreason = BANDLIM_RST_OPENPORT; 1135 goto dropwithreset; 1136 } 1137 #ifdef TCP_RFC7413 1138 tfo_socket_result: 1139 #endif 1140 if (so == NULL) { 1141 /* 1142 * We completed the 3-way handshake 1143 * but could not allocate a socket 1144 * either due to memory shortage, 1145 * listen queue length limits or 1146 * global socket limits. Send RST 1147 * or wait and have the remote end 1148 * retransmit the ACK for another 1149 * try. 1150 */ 1151 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1152 log(LOG_DEBUG, "%s; %s: Listen socket: " 1153 "Socket allocation failed due to " 1154 "limits or memory shortage, %s\n", 1155 s, __func__, 1156 V_tcp_sc_rst_sock_fail ? 1157 "sending RST" : "try again"); 1158 if (V_tcp_sc_rst_sock_fail) { 1159 rstreason = BANDLIM_UNLIMITED; 1160 goto dropwithreset; 1161 } else 1162 goto dropunlock; 1163 } 1164 /* 1165 * Socket is created in state SYN_RECEIVED. 1166 * Unlock the listen socket, lock the newly 1167 * created socket and update the tp variable. 1168 */ 1169 INP_WUNLOCK(inp); /* listen socket */ 1170 inp = sotoinpcb(so); 1171 /* 1172 * New connection inpcb is already locked by 1173 * syncache_expand(). 1174 */ 1175 INP_WLOCK_ASSERT(inp); 1176 tp = intotcpcb(inp); 1177 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1178 ("%s: ", __func__)); 1179 /* 1180 * Process the segment and the data it 1181 * contains. tcp_do_segment() consumes 1182 * the mbuf chain and unlocks the inpcb. 1183 */ 1184 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1185 iptos, ti_locked); 1186 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1187 return (IPPROTO_DONE); 1188 } 1189 /* 1190 * Segment flag validation for new connection attempts: 1191 * 1192 * Our (SYN|ACK) response was rejected. 1193 * Check with syncache and remove entry to prevent 1194 * retransmits. 1195 * 1196 * NB: syncache_chkrst does its own logging of failure 1197 * causes. 1198 */ 1199 if (thflags & TH_RST) { 1200 syncache_chkrst(&inc, th); 1201 goto dropunlock; 1202 } 1203 /* 1204 * We can't do anything without SYN. 1205 */ 1206 if ((thflags & TH_SYN) == 0) { 1207 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1208 log(LOG_DEBUG, "%s; %s: Listen socket: " 1209 "SYN is missing, segment ignored\n", 1210 s, __func__); 1211 TCPSTAT_INC(tcps_badsyn); 1212 goto dropunlock; 1213 } 1214 /* 1215 * (SYN|ACK) is bogus on a listen socket. 1216 */ 1217 if (thflags & TH_ACK) { 1218 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1219 log(LOG_DEBUG, "%s; %s: Listen socket: " 1220 "SYN|ACK invalid, segment rejected\n", 1221 s, __func__); 1222 syncache_badack(&inc); /* XXX: Not needed! */ 1223 TCPSTAT_INC(tcps_badsyn); 1224 rstreason = BANDLIM_RST_OPENPORT; 1225 goto dropwithreset; 1226 } 1227 /* 1228 * If the drop_synfin option is enabled, drop all 1229 * segments with both the SYN and FIN bits set. 1230 * This prevents e.g. nmap from identifying the 1231 * TCP/IP stack. 1232 * XXX: Poor reasoning. nmap has other methods 1233 * and is constantly refining its stack detection 1234 * strategies. 1235 * XXX: This is a violation of the TCP specification 1236 * and was used by RFC1644. 1237 */ 1238 if ((thflags & TH_FIN) && V_drop_synfin) { 1239 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1240 log(LOG_DEBUG, "%s; %s: Listen socket: " 1241 "SYN|FIN segment ignored (based on " 1242 "sysctl setting)\n", s, __func__); 1243 TCPSTAT_INC(tcps_badsyn); 1244 goto dropunlock; 1245 } 1246 /* 1247 * Segment's flags are (SYN) or (SYN|FIN). 1248 * 1249 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1250 * as they do not affect the state of the TCP FSM. 1251 * The data pointed to by TH_URG and th_urp is ignored. 1252 */ 1253 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1254 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1255 KASSERT(thflags & (TH_SYN), 1256 ("%s: Listen socket: TH_SYN not set", __func__)); 1257 #ifdef INET6 1258 /* 1259 * If deprecated address is forbidden, 1260 * we do not accept SYN to deprecated interface 1261 * address to prevent any new inbound connection from 1262 * getting established. 1263 * When we do not accept SYN, we send a TCP RST, 1264 * with deprecated source address (instead of dropping 1265 * it). We compromise it as it is much better for peer 1266 * to send a RST, and RST will be the final packet 1267 * for the exchange. 1268 * 1269 * If we do not forbid deprecated addresses, we accept 1270 * the SYN packet. RFC2462 does not suggest dropping 1271 * SYN in this case. 1272 * If we decipher RFC2462 5.5.4, it says like this: 1273 * 1. use of deprecated addr with existing 1274 * communication is okay - "SHOULD continue to be 1275 * used" 1276 * 2. use of it with new communication: 1277 * (2a) "SHOULD NOT be used if alternate address 1278 * with sufficient scope is available" 1279 * (2b) nothing mentioned otherwise. 1280 * Here we fall into (2b) case as we have no choice in 1281 * our source address selection - we must obey the peer. 1282 * 1283 * The wording in RFC2462 is confusing, and there are 1284 * multiple description text for deprecated address 1285 * handling - worse, they are not exactly the same. 1286 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1287 */ 1288 if (isipv6 && !V_ip6_use_deprecated) { 1289 struct in6_ifaddr *ia6; 1290 1291 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1292 if (ia6 != NULL && 1293 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1294 ifa_free(&ia6->ia_ifa); 1295 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1296 log(LOG_DEBUG, "%s; %s: Listen socket: " 1297 "Connection attempt to deprecated " 1298 "IPv6 address rejected\n", 1299 s, __func__); 1300 rstreason = BANDLIM_RST_OPENPORT; 1301 goto dropwithreset; 1302 } 1303 if (ia6) 1304 ifa_free(&ia6->ia_ifa); 1305 } 1306 #endif /* INET6 */ 1307 /* 1308 * Basic sanity checks on incoming SYN requests: 1309 * Don't respond if the destination is a link layer 1310 * broadcast according to RFC1122 4.2.3.10, p. 104. 1311 * If it is from this socket it must be forged. 1312 * Don't respond if the source or destination is a 1313 * global or subnet broad- or multicast address. 1314 * Note that it is quite possible to receive unicast 1315 * link-layer packets with a broadcast IP address. Use 1316 * in_broadcast() to find them. 1317 */ 1318 if (m->m_flags & (M_BCAST|M_MCAST)) { 1319 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1320 log(LOG_DEBUG, "%s; %s: Listen socket: " 1321 "Connection attempt from broad- or multicast " 1322 "link layer address ignored\n", s, __func__); 1323 goto dropunlock; 1324 } 1325 #ifdef INET6 1326 if (isipv6) { 1327 if (th->th_dport == th->th_sport && 1328 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1329 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1330 log(LOG_DEBUG, "%s; %s: Listen socket: " 1331 "Connection attempt to/from self " 1332 "ignored\n", s, __func__); 1333 goto dropunlock; 1334 } 1335 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1336 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1337 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1338 log(LOG_DEBUG, "%s; %s: Listen socket: " 1339 "Connection attempt from/to multicast " 1340 "address ignored\n", s, __func__); 1341 goto dropunlock; 1342 } 1343 } 1344 #endif 1345 #if defined(INET) && defined(INET6) 1346 else 1347 #endif 1348 #ifdef INET 1349 { 1350 if (th->th_dport == th->th_sport && 1351 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1352 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1353 log(LOG_DEBUG, "%s; %s: Listen socket: " 1354 "Connection attempt from/to self " 1355 "ignored\n", s, __func__); 1356 goto dropunlock; 1357 } 1358 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1359 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1360 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1361 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1362 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1363 log(LOG_DEBUG, "%s; %s: Listen socket: " 1364 "Connection attempt from/to broad- " 1365 "or multicast address ignored\n", 1366 s, __func__); 1367 goto dropunlock; 1368 } 1369 } 1370 #endif 1371 /* 1372 * SYN appears to be valid. Create compressed TCP state 1373 * for syncache. 1374 */ 1375 #ifdef TCPDEBUG 1376 if (so->so_options & SO_DEBUG) 1377 tcp_trace(TA_INPUT, ostate, tp, 1378 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1379 #endif 1380 TCP_PROBE3(debug__input, tp, th, m); 1381 tcp_dooptions(&to, optp, optlen, TO_SYN); 1382 #ifdef TCP_RFC7413 1383 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL)) 1384 goto tfo_socket_result; 1385 #else 1386 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1387 #endif 1388 /* 1389 * Entry added to syncache and mbuf consumed. 1390 * Only the listen socket is unlocked by syncache_add(). 1391 */ 1392 if (ti_locked == TI_RLOCKED) { 1393 INP_INFO_RUNLOCK(&V_tcbinfo); 1394 ti_locked = TI_UNLOCKED; 1395 } 1396 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1397 return (IPPROTO_DONE); 1398 } else if (tp->t_state == TCPS_LISTEN) { 1399 /* 1400 * When a listen socket is torn down the SO_ACCEPTCONN 1401 * flag is removed first while connections are drained 1402 * from the accept queue in a unlock/lock cycle of the 1403 * ACCEPT_LOCK, opening a race condition allowing a SYN 1404 * attempt go through unhandled. 1405 */ 1406 goto dropunlock; 1407 } 1408 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1409 if (tp->t_flags & TF_SIGNATURE) { 1410 tcp_dooptions(&to, optp, optlen, thflags); 1411 if ((to.to_flags & TOF_SIGNATURE) == 0) { 1412 TCPSTAT_INC(tcps_sig_err_nosigopt); 1413 goto dropunlock; 1414 } 1415 if (!TCPMD5_ENABLED() || 1416 TCPMD5_INPUT(m, th, to.to_signature) != 0) 1417 goto dropunlock; 1418 } 1419 #endif 1420 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1421 1422 /* 1423 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1424 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1425 * the inpcb, and unlocks pcbinfo. 1426 */ 1427 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1428 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1429 return (IPPROTO_DONE); 1430 1431 dropwithreset: 1432 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1433 1434 if (ti_locked == TI_RLOCKED) { 1435 INP_INFO_RUNLOCK(&V_tcbinfo); 1436 ti_locked = TI_UNLOCKED; 1437 } 1438 #ifdef INVARIANTS 1439 else { 1440 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1441 "ti_locked: %d", __func__, ti_locked)); 1442 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1443 } 1444 #endif 1445 1446 if (inp != NULL) { 1447 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1448 INP_WUNLOCK(inp); 1449 } else 1450 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1451 m = NULL; /* mbuf chain got consumed. */ 1452 goto drop; 1453 1454 dropunlock: 1455 if (m != NULL) 1456 TCP_PROBE5(receive, NULL, tp, m, tp, th); 1457 1458 if (ti_locked == TI_RLOCKED) { 1459 INP_INFO_RUNLOCK(&V_tcbinfo); 1460 ti_locked = TI_UNLOCKED; 1461 } 1462 #ifdef INVARIANTS 1463 else { 1464 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1465 "ti_locked: %d", __func__, ti_locked)); 1466 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1467 } 1468 #endif 1469 1470 if (inp != NULL) 1471 INP_WUNLOCK(inp); 1472 1473 drop: 1474 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1475 if (s != NULL) 1476 free(s, M_TCPLOG); 1477 if (m != NULL) 1478 m_freem(m); 1479 return (IPPROTO_DONE); 1480 } 1481 1482 /* 1483 * Automatic sizing of receive socket buffer. Often the send 1484 * buffer size is not optimally adjusted to the actual network 1485 * conditions at hand (delay bandwidth product). Setting the 1486 * buffer size too small limits throughput on links with high 1487 * bandwidth and high delay (eg. trans-continental/oceanic links). 1488 * 1489 * On the receive side the socket buffer memory is only rarely 1490 * used to any significant extent. This allows us to be much 1491 * more aggressive in scaling the receive socket buffer. For 1492 * the case that the buffer space is actually used to a large 1493 * extent and we run out of kernel memory we can simply drop 1494 * the new segments; TCP on the sender will just retransmit it 1495 * later. Setting the buffer size too big may only consume too 1496 * much kernel memory if the application doesn't read() from 1497 * the socket or packet loss or reordering makes use of the 1498 * reassembly queue. 1499 * 1500 * The criteria to step up the receive buffer one notch are: 1501 * 1. Application has not set receive buffer size with 1502 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1503 * 2. the number of bytes received during the time it takes 1504 * one timestamp to be reflected back to us (the RTT); 1505 * 3. received bytes per RTT is within seven eighth of the 1506 * current socket buffer size; 1507 * 4. receive buffer size has not hit maximal automatic size; 1508 * 1509 * This algorithm does one step per RTT at most and only if 1510 * we receive a bulk stream w/o packet losses or reorderings. 1511 * Shrinking the buffer during idle times is not necessary as 1512 * it doesn't consume any memory when idle. 1513 * 1514 * TODO: Only step up if the application is actually serving 1515 * the buffer to better manage the socket buffer resources. 1516 */ 1517 int 1518 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, 1519 struct tcpcb *tp, int tlen) 1520 { 1521 int newsize = 0; 1522 1523 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && 1524 tp->t_srtt != 0 && tp->rfbuf_ts != 0 && 1525 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > 1526 (tp->t_srtt >> TCP_RTT_SHIFT)) { 1527 if (tp->rfbuf_cnt > (so->so_rcv.sb_hiwat / 8 * 7) && 1528 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { 1529 newsize = min(so->so_rcv.sb_hiwat + 1530 V_tcp_autorcvbuf_inc, V_tcp_autorcvbuf_max); 1531 } 1532 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); 1533 1534 /* Start over with next RTT. */ 1535 tp->rfbuf_ts = 0; 1536 tp->rfbuf_cnt = 0; 1537 } else { 1538 tp->rfbuf_cnt += tlen; /* add up */ 1539 } 1540 1541 return (newsize); 1542 } 1543 1544 void 1545 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1546 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1547 int ti_locked) 1548 { 1549 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed; 1550 int rstreason, todrop, win; 1551 uint32_t tiwin; 1552 uint16_t nsegs; 1553 char *s; 1554 struct in_conninfo *inc; 1555 struct mbuf *mfree; 1556 struct tcpopt to; 1557 #ifdef TCP_RFC7413 1558 int tfo_syn; 1559 #endif 1560 1561 #ifdef TCPDEBUG 1562 /* 1563 * The size of tcp_saveipgen must be the size of the max ip header, 1564 * now IPv6. 1565 */ 1566 u_char tcp_saveipgen[IP6_HDR_LEN]; 1567 struct tcphdr tcp_savetcp; 1568 short ostate = 0; 1569 #endif 1570 thflags = th->th_flags; 1571 inc = &tp->t_inpcb->inp_inc; 1572 tp->sackhint.last_sack_ack = 0; 1573 sack_changed = 0; 1574 nsegs = max(1, m->m_pkthdr.lro_nsegs); 1575 1576 /* 1577 * If this is either a state-changing packet or current state isn't 1578 * established, we require a write lock on tcbinfo. Otherwise, we 1579 * allow the tcbinfo to be in either alocked or unlocked, as the 1580 * caller may have unnecessarily acquired a write lock due to a race. 1581 */ 1582 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1583 tp->t_state != TCPS_ESTABLISHED) { 1584 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1585 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1586 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1587 } else { 1588 #ifdef INVARIANTS 1589 if (ti_locked == TI_RLOCKED) 1590 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1591 else { 1592 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1593 "ti_locked: %d", __func__, ti_locked)); 1594 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1595 } 1596 #endif 1597 } 1598 INP_WLOCK_ASSERT(tp->t_inpcb); 1599 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1600 __func__)); 1601 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1602 __func__)); 1603 1604 #ifdef TCPPCAP 1605 /* Save segment, if requested. */ 1606 tcp_pcap_add(th, m, &(tp->t_inpkts)); 1607 #endif 1608 1609 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 1610 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1611 log(LOG_DEBUG, "%s; %s: " 1612 "SYN|FIN segment ignored (based on " 1613 "sysctl setting)\n", s, __func__); 1614 free(s, M_TCPLOG); 1615 } 1616 goto drop; 1617 } 1618 1619 /* 1620 * If a segment with the ACK-bit set arrives in the SYN-SENT state 1621 * check SEQ.ACK first. 1622 */ 1623 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && 1624 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { 1625 rstreason = BANDLIM_UNLIMITED; 1626 goto dropwithreset; 1627 } 1628 1629 /* 1630 * Segment received on connection. 1631 * Reset idle time and keep-alive timer. 1632 * XXX: This should be done after segment 1633 * validation to ignore broken/spoofed segs. 1634 */ 1635 tp->t_rcvtime = ticks; 1636 1637 /* 1638 * Scale up the window into a 32-bit value. 1639 * For the SYN_SENT state the scale is zero. 1640 */ 1641 tiwin = th->th_win << tp->snd_scale; 1642 1643 /* 1644 * TCP ECN processing. 1645 */ 1646 if (tp->t_flags & TF_ECN_PERMIT) { 1647 if (thflags & TH_CWR) 1648 tp->t_flags &= ~TF_ECN_SND_ECE; 1649 switch (iptos & IPTOS_ECN_MASK) { 1650 case IPTOS_ECN_CE: 1651 tp->t_flags |= TF_ECN_SND_ECE; 1652 TCPSTAT_INC(tcps_ecn_ce); 1653 break; 1654 case IPTOS_ECN_ECT0: 1655 TCPSTAT_INC(tcps_ecn_ect0); 1656 break; 1657 case IPTOS_ECN_ECT1: 1658 TCPSTAT_INC(tcps_ecn_ect1); 1659 break; 1660 } 1661 1662 /* Process a packet differently from RFC3168. */ 1663 cc_ecnpkt_handler(tp, th, iptos); 1664 1665 /* Congestion experienced. */ 1666 if (thflags & TH_ECE) { 1667 cc_cong_signal(tp, th, CC_ECN); 1668 } 1669 } 1670 1671 /* 1672 * Parse options on any incoming segment. 1673 */ 1674 tcp_dooptions(&to, (u_char *)(th + 1), 1675 (th->th_off << 2) - sizeof(struct tcphdr), 1676 (thflags & TH_SYN) ? TO_SYN : 0); 1677 1678 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) 1679 if ((tp->t_flags & TF_SIGNATURE) != 0 && 1680 (to.to_flags & TOF_SIGNATURE) == 0) { 1681 TCPSTAT_INC(tcps_sig_err_sigopt); 1682 /* XXX: should drop? */ 1683 } 1684 #endif 1685 /* 1686 * If echoed timestamp is later than the current time, 1687 * fall back to non RFC1323 RTT calculation. Normalize 1688 * timestamp if syncookies were used when this connection 1689 * was established. 1690 */ 1691 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1692 to.to_tsecr -= tp->ts_offset; 1693 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1694 to.to_tsecr = 0; 1695 } 1696 /* 1697 * If timestamps were negotiated during SYN/ACK they should 1698 * appear on every segment during this session and vice versa. 1699 */ 1700 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1701 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1702 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1703 "no action\n", s, __func__); 1704 free(s, M_TCPLOG); 1705 } 1706 } 1707 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1708 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1709 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1710 "no action\n", s, __func__); 1711 free(s, M_TCPLOG); 1712 } 1713 } 1714 1715 /* 1716 * Process options only when we get SYN/ACK back. The SYN case 1717 * for incoming connections is handled in tcp_syncache. 1718 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1719 * or <SYN,ACK>) segment itself is never scaled. 1720 * XXX this is traditional behavior, may need to be cleaned up. 1721 */ 1722 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1723 if ((to.to_flags & TOF_SCALE) && 1724 (tp->t_flags & TF_REQ_SCALE)) { 1725 tp->t_flags |= TF_RCVD_SCALE; 1726 tp->snd_scale = to.to_wscale; 1727 } 1728 /* 1729 * Initial send window. It will be updated with 1730 * the next incoming segment to the scaled value. 1731 */ 1732 tp->snd_wnd = th->th_win; 1733 if (to.to_flags & TOF_TS) { 1734 tp->t_flags |= TF_RCVD_TSTMP; 1735 tp->ts_recent = to.to_tsval; 1736 tp->ts_recent_age = tcp_ts_getticks(); 1737 } 1738 if (to.to_flags & TOF_MSS) 1739 tcp_mss(tp, to.to_mss); 1740 if ((tp->t_flags & TF_SACK_PERMIT) && 1741 (to.to_flags & TOF_SACKPERM) == 0) 1742 tp->t_flags &= ~TF_SACK_PERMIT; 1743 } 1744 1745 /* 1746 * Header prediction: check for the two common cases 1747 * of a uni-directional data xfer. If the packet has 1748 * no control flags, is in-sequence, the window didn't 1749 * change and we're not retransmitting, it's a 1750 * candidate. If the length is zero and the ack moved 1751 * forward, we're the sender side of the xfer. Just 1752 * free the data acked & wake any higher level process 1753 * that was blocked waiting for space. If the length 1754 * is non-zero and the ack didn't move, we're the 1755 * receiver side. If we're getting packets in-order 1756 * (the reassembly queue is empty), add the data to 1757 * the socket buffer and note that we need a delayed ack. 1758 * Make sure that the hidden state-flags are also off. 1759 * Since we check for TCPS_ESTABLISHED first, it can only 1760 * be TH_NEEDSYN. 1761 */ 1762 if (tp->t_state == TCPS_ESTABLISHED && 1763 th->th_seq == tp->rcv_nxt && 1764 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1765 tp->snd_nxt == tp->snd_max && 1766 tiwin && tiwin == tp->snd_wnd && 1767 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1768 LIST_EMPTY(&tp->t_segq) && 1769 ((to.to_flags & TOF_TS) == 0 || 1770 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1771 1772 /* 1773 * If last ACK falls within this segment's sequence numbers, 1774 * record the timestamp. 1775 * NOTE that the test is modified according to the latest 1776 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1777 */ 1778 if ((to.to_flags & TOF_TS) != 0 && 1779 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1780 tp->ts_recent_age = tcp_ts_getticks(); 1781 tp->ts_recent = to.to_tsval; 1782 } 1783 1784 if (tlen == 0) { 1785 if (SEQ_GT(th->th_ack, tp->snd_una) && 1786 SEQ_LEQ(th->th_ack, tp->snd_max) && 1787 !IN_RECOVERY(tp->t_flags) && 1788 (to.to_flags & TOF_SACK) == 0 && 1789 TAILQ_EMPTY(&tp->snd_holes)) { 1790 /* 1791 * This is a pure ack for outstanding data. 1792 */ 1793 if (ti_locked == TI_RLOCKED) 1794 INP_INFO_RUNLOCK(&V_tcbinfo); 1795 ti_locked = TI_UNLOCKED; 1796 1797 TCPSTAT_INC(tcps_predack); 1798 1799 /* 1800 * "bad retransmit" recovery. 1801 */ 1802 if (tp->t_rxtshift == 1 && 1803 tp->t_flags & TF_PREVVALID && 1804 (int)(ticks - tp->t_badrxtwin) < 0) { 1805 cc_cong_signal(tp, th, CC_RTO_ERR); 1806 } 1807 1808 /* 1809 * Recalculate the transmit timer / rtt. 1810 * 1811 * Some boxes send broken timestamp replies 1812 * during the SYN+ACK phase, ignore 1813 * timestamps of 0 or we could calculate a 1814 * huge RTT and blow up the retransmit timer. 1815 */ 1816 if ((to.to_flags & TOF_TS) != 0 && 1817 to.to_tsecr) { 1818 uint32_t t; 1819 1820 t = tcp_ts_getticks() - to.to_tsecr; 1821 if (!tp->t_rttlow || tp->t_rttlow > t) 1822 tp->t_rttlow = t; 1823 tcp_xmit_timer(tp, 1824 TCP_TS_TO_TICKS(t) + 1); 1825 } else if (tp->t_rtttime && 1826 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1827 if (!tp->t_rttlow || 1828 tp->t_rttlow > ticks - tp->t_rtttime) 1829 tp->t_rttlow = ticks - tp->t_rtttime; 1830 tcp_xmit_timer(tp, 1831 ticks - tp->t_rtttime); 1832 } 1833 acked = BYTES_THIS_ACK(tp, th); 1834 1835 #ifdef TCP_HHOOK 1836 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1837 hhook_run_tcp_est_in(tp, th, &to); 1838 #endif 1839 1840 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 1841 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1842 sbdrop(&so->so_snd, acked); 1843 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1844 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1845 tp->snd_recover = th->th_ack - 1; 1846 1847 /* 1848 * Let the congestion control algorithm update 1849 * congestion control related information. This 1850 * typically means increasing the congestion 1851 * window. 1852 */ 1853 cc_ack_received(tp, th, nsegs, CC_ACK); 1854 1855 tp->snd_una = th->th_ack; 1856 /* 1857 * Pull snd_wl2 up to prevent seq wrap relative 1858 * to th_ack. 1859 */ 1860 tp->snd_wl2 = th->th_ack; 1861 tp->t_dupacks = 0; 1862 m_freem(m); 1863 1864 /* 1865 * If all outstanding data are acked, stop 1866 * retransmit timer, otherwise restart timer 1867 * using current (possibly backed-off) value. 1868 * If process is waiting for space, 1869 * wakeup/selwakeup/signal. If data 1870 * are ready to send, let tcp_output 1871 * decide between more output or persist. 1872 */ 1873 #ifdef TCPDEBUG 1874 if (so->so_options & SO_DEBUG) 1875 tcp_trace(TA_INPUT, ostate, tp, 1876 (void *)tcp_saveipgen, 1877 &tcp_savetcp, 0); 1878 #endif 1879 TCP_PROBE3(debug__input, tp, th, m); 1880 if (tp->snd_una == tp->snd_max) 1881 tcp_timer_activate(tp, TT_REXMT, 0); 1882 else if (!tcp_timer_active(tp, TT_PERSIST)) 1883 tcp_timer_activate(tp, TT_REXMT, 1884 tp->t_rxtcur); 1885 sowwakeup(so); 1886 if (sbavail(&so->so_snd)) 1887 (void) tp->t_fb->tfb_tcp_output(tp); 1888 goto check_delack; 1889 } 1890 } else if (th->th_ack == tp->snd_una && 1891 tlen <= sbspace(&so->so_rcv)) { 1892 int newsize = 0; /* automatic sockbuf scaling */ 1893 1894 /* 1895 * This is a pure, in-sequence data packet with 1896 * nothing on the reassembly queue and we have enough 1897 * buffer space to take it. 1898 */ 1899 if (ti_locked == TI_RLOCKED) 1900 INP_INFO_RUNLOCK(&V_tcbinfo); 1901 ti_locked = TI_UNLOCKED; 1902 1903 /* Clean receiver SACK report if present */ 1904 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1905 tcp_clean_sackreport(tp); 1906 TCPSTAT_INC(tcps_preddat); 1907 tp->rcv_nxt += tlen; 1908 /* 1909 * Pull snd_wl1 up to prevent seq wrap relative to 1910 * th_seq. 1911 */ 1912 tp->snd_wl1 = th->th_seq; 1913 /* 1914 * Pull rcv_up up to prevent seq wrap relative to 1915 * rcv_nxt. 1916 */ 1917 tp->rcv_up = tp->rcv_nxt; 1918 TCPSTAT_ADD(tcps_rcvpack, nsegs); 1919 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1920 #ifdef TCPDEBUG 1921 if (so->so_options & SO_DEBUG) 1922 tcp_trace(TA_INPUT, ostate, tp, 1923 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1924 #endif 1925 TCP_PROBE3(debug__input, tp, th, m); 1926 1927 newsize = tcp_autorcvbuf(m, th, so, tp, tlen); 1928 1929 /* Add data to socket buffer. */ 1930 SOCKBUF_LOCK(&so->so_rcv); 1931 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1932 m_freem(m); 1933 } else { 1934 /* 1935 * Set new socket buffer size. 1936 * Give up when limit is reached. 1937 */ 1938 if (newsize) 1939 if (!sbreserve_locked(&so->so_rcv, 1940 newsize, so, NULL)) 1941 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1942 m_adj(m, drop_hdrlen); /* delayed header drop */ 1943 sbappendstream_locked(&so->so_rcv, m, 0); 1944 } 1945 /* NB: sorwakeup_locked() does an implicit unlock. */ 1946 sorwakeup_locked(so); 1947 if (DELAY_ACK(tp, tlen)) { 1948 tp->t_flags |= TF_DELACK; 1949 } else { 1950 tp->t_flags |= TF_ACKNOW; 1951 tp->t_fb->tfb_tcp_output(tp); 1952 } 1953 goto check_delack; 1954 } 1955 } 1956 1957 /* 1958 * Calculate amount of space in receive window, 1959 * and then do TCP input processing. 1960 * Receive window is amount of space in rcv queue, 1961 * but not less than advertised window. 1962 */ 1963 win = sbspace(&so->so_rcv); 1964 if (win < 0) 1965 win = 0; 1966 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1967 1968 switch (tp->t_state) { 1969 1970 /* 1971 * If the state is SYN_RECEIVED: 1972 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1973 */ 1974 case TCPS_SYN_RECEIVED: 1975 if ((thflags & TH_ACK) && 1976 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1977 SEQ_GT(th->th_ack, tp->snd_max))) { 1978 rstreason = BANDLIM_RST_OPENPORT; 1979 goto dropwithreset; 1980 } 1981 #ifdef TCP_RFC7413 1982 if (IS_FASTOPEN(tp->t_flags)) { 1983 /* 1984 * When a TFO connection is in SYN_RECEIVED, the 1985 * only valid packets are the initial SYN, a 1986 * retransmit/copy of the initial SYN (possibly with 1987 * a subset of the original data), a valid ACK, a 1988 * FIN, or a RST. 1989 */ 1990 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { 1991 rstreason = BANDLIM_RST_OPENPORT; 1992 goto dropwithreset; 1993 } else if (thflags & TH_SYN) { 1994 /* non-initial SYN is ignored */ 1995 if ((tcp_timer_active(tp, TT_DELACK) || 1996 tcp_timer_active(tp, TT_REXMT))) 1997 goto drop; 1998 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { 1999 goto drop; 2000 } 2001 } 2002 #endif 2003 break; 2004 2005 /* 2006 * If the state is SYN_SENT: 2007 * if seg contains a RST with valid ACK (SEQ.ACK has already 2008 * been verified), then drop the connection. 2009 * if seg contains a RST without an ACK, drop the seg. 2010 * if seg does not contain SYN, then drop the seg. 2011 * Otherwise this is an acceptable SYN segment 2012 * initialize tp->rcv_nxt and tp->irs 2013 * if seg contains ack then advance tp->snd_una 2014 * if seg contains an ECE and ECN support is enabled, the stream 2015 * is ECN capable. 2016 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 2017 * arrange for segment to be acked (eventually) 2018 * continue processing rest of data/controls, beginning with URG 2019 */ 2020 case TCPS_SYN_SENT: 2021 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 2022 TCP_PROBE5(connect__refused, NULL, tp, 2023 m, tp, th); 2024 tp = tcp_drop(tp, ECONNREFUSED); 2025 } 2026 if (thflags & TH_RST) 2027 goto drop; 2028 if (!(thflags & TH_SYN)) 2029 goto drop; 2030 2031 tp->irs = th->th_seq; 2032 tcp_rcvseqinit(tp); 2033 if (thflags & TH_ACK) { 2034 TCPSTAT_INC(tcps_connects); 2035 soisconnected(so); 2036 #ifdef MAC 2037 mac_socketpeer_set_from_mbuf(m, so); 2038 #endif 2039 /* Do window scaling on this connection? */ 2040 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2041 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2042 tp->rcv_scale = tp->request_r_scale; 2043 } 2044 tp->rcv_adv += min(tp->rcv_wnd, 2045 TCP_MAXWIN << tp->rcv_scale); 2046 tp->snd_una++; /* SYN is acked */ 2047 /* 2048 * If there's data, delay ACK; if there's also a FIN 2049 * ACKNOW will be turned on later. 2050 */ 2051 if (DELAY_ACK(tp, tlen) && tlen != 0) 2052 tcp_timer_activate(tp, TT_DELACK, 2053 tcp_delacktime); 2054 else 2055 tp->t_flags |= TF_ACKNOW; 2056 2057 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2058 tp->t_flags |= TF_ECN_PERMIT; 2059 TCPSTAT_INC(tcps_ecn_shs); 2060 } 2061 2062 /* 2063 * Received <SYN,ACK> in SYN_SENT[*] state. 2064 * Transitions: 2065 * SYN_SENT --> ESTABLISHED 2066 * SYN_SENT* --> FIN_WAIT_1 2067 */ 2068 tp->t_starttime = ticks; 2069 if (tp->t_flags & TF_NEEDFIN) { 2070 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2071 tp->t_flags &= ~TF_NEEDFIN; 2072 thflags &= ~TH_SYN; 2073 } else { 2074 tcp_state_change(tp, TCPS_ESTABLISHED); 2075 TCP_PROBE5(connect__established, NULL, tp, 2076 m, tp, th); 2077 cc_conn_init(tp); 2078 tcp_timer_activate(tp, TT_KEEP, 2079 TP_KEEPIDLE(tp)); 2080 } 2081 } else { 2082 /* 2083 * Received initial SYN in SYN-SENT[*] state => 2084 * simultaneous open. 2085 * If it succeeds, connection is * half-synchronized. 2086 * Otherwise, do 3-way handshake: 2087 * SYN-SENT -> SYN-RECEIVED 2088 * SYN-SENT* -> SYN-RECEIVED* 2089 */ 2090 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2091 tcp_timer_activate(tp, TT_REXMT, 0); 2092 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2093 } 2094 2095 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2096 "ti_locked %d", __func__, ti_locked)); 2097 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2098 INP_WLOCK_ASSERT(tp->t_inpcb); 2099 2100 /* 2101 * Advance th->th_seq to correspond to first data byte. 2102 * If data, trim to stay within window, 2103 * dropping FIN if necessary. 2104 */ 2105 th->th_seq++; 2106 if (tlen > tp->rcv_wnd) { 2107 todrop = tlen - tp->rcv_wnd; 2108 m_adj(m, -todrop); 2109 tlen = tp->rcv_wnd; 2110 thflags &= ~TH_FIN; 2111 TCPSTAT_INC(tcps_rcvpackafterwin); 2112 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2113 } 2114 tp->snd_wl1 = th->th_seq - 1; 2115 tp->rcv_up = th->th_seq; 2116 /* 2117 * Client side of transaction: already sent SYN and data. 2118 * If the remote host used T/TCP to validate the SYN, 2119 * our data will be ACK'd; if so, enter normal data segment 2120 * processing in the middle of step 5, ack processing. 2121 * Otherwise, goto step 6. 2122 */ 2123 if (thflags & TH_ACK) 2124 goto process_ACK; 2125 2126 goto step6; 2127 2128 /* 2129 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2130 * do normal processing. 2131 * 2132 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2133 */ 2134 case TCPS_LAST_ACK: 2135 case TCPS_CLOSING: 2136 break; /* continue normal processing */ 2137 } 2138 2139 /* 2140 * States other than LISTEN or SYN_SENT. 2141 * First check the RST flag and sequence number since reset segments 2142 * are exempt from the timestamp and connection count tests. This 2143 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2144 * below which allowed reset segments in half the sequence space 2145 * to fall though and be processed (which gives forged reset 2146 * segments with a random sequence number a 50 percent chance of 2147 * killing a connection). 2148 * Then check timestamp, if present. 2149 * Then check the connection count, if present. 2150 * Then check that at least some bytes of segment are within 2151 * receive window. If segment begins before rcv_nxt, 2152 * drop leading data (and SYN); if nothing left, just ack. 2153 */ 2154 if (thflags & TH_RST) { 2155 /* 2156 * RFC5961 Section 3.2 2157 * 2158 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2159 * - If RST is in window, we send challenge ACK. 2160 * 2161 * Note: to take into account delayed ACKs, we should 2162 * test against last_ack_sent instead of rcv_nxt. 2163 * Note 2: we handle special case of closed window, not 2164 * covered by the RFC. 2165 */ 2166 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2167 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2168 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2169 2170 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2171 KASSERT(ti_locked == TI_RLOCKED, 2172 ("%s: TH_RST ti_locked %d, th %p tp %p", 2173 __func__, ti_locked, th, tp)); 2174 KASSERT(tp->t_state != TCPS_SYN_SENT, 2175 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2176 __func__, th, tp)); 2177 2178 if (V_tcp_insecure_rst || 2179 tp->last_ack_sent == th->th_seq) { 2180 TCPSTAT_INC(tcps_drops); 2181 /* Drop the connection. */ 2182 switch (tp->t_state) { 2183 case TCPS_SYN_RECEIVED: 2184 so->so_error = ECONNREFUSED; 2185 goto close; 2186 case TCPS_ESTABLISHED: 2187 case TCPS_FIN_WAIT_1: 2188 case TCPS_FIN_WAIT_2: 2189 case TCPS_CLOSE_WAIT: 2190 case TCPS_CLOSING: 2191 case TCPS_LAST_ACK: 2192 so->so_error = ECONNRESET; 2193 close: 2194 /* FALLTHROUGH */ 2195 default: 2196 tp = tcp_close(tp); 2197 } 2198 } else { 2199 TCPSTAT_INC(tcps_badrst); 2200 /* Send challenge ACK. */ 2201 tcp_respond(tp, mtod(m, void *), th, m, 2202 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2203 tp->last_ack_sent = tp->rcv_nxt; 2204 m = NULL; 2205 } 2206 } 2207 goto drop; 2208 } 2209 2210 /* 2211 * RFC5961 Section 4.2 2212 * Send challenge ACK for any SYN in synchronized state. 2213 */ 2214 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && 2215 tp->t_state != TCPS_SYN_RECEIVED) { 2216 KASSERT(ti_locked == TI_RLOCKED, 2217 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2218 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2219 2220 TCPSTAT_INC(tcps_badsyn); 2221 if (V_tcp_insecure_syn && 2222 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2223 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2224 tp = tcp_drop(tp, ECONNRESET); 2225 rstreason = BANDLIM_UNLIMITED; 2226 } else { 2227 /* Send challenge ACK. */ 2228 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2229 tp->snd_nxt, TH_ACK); 2230 tp->last_ack_sent = tp->rcv_nxt; 2231 m = NULL; 2232 } 2233 goto drop; 2234 } 2235 2236 /* 2237 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2238 * and it's less than ts_recent, drop it. 2239 */ 2240 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2241 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2242 2243 /* Check to see if ts_recent is over 24 days old. */ 2244 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2245 /* 2246 * Invalidate ts_recent. If this segment updates 2247 * ts_recent, the age will be reset later and ts_recent 2248 * will get a valid value. If it does not, setting 2249 * ts_recent to zero will at least satisfy the 2250 * requirement that zero be placed in the timestamp 2251 * echo reply when ts_recent isn't valid. The 2252 * age isn't reset until we get a valid ts_recent 2253 * because we don't want out-of-order segments to be 2254 * dropped when ts_recent is old. 2255 */ 2256 tp->ts_recent = 0; 2257 } else { 2258 TCPSTAT_INC(tcps_rcvduppack); 2259 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2260 TCPSTAT_INC(tcps_pawsdrop); 2261 if (tlen) 2262 goto dropafterack; 2263 goto drop; 2264 } 2265 } 2266 2267 /* 2268 * In the SYN-RECEIVED state, validate that the packet belongs to 2269 * this connection before trimming the data to fit the receive 2270 * window. Check the sequence number versus IRS since we know 2271 * the sequence numbers haven't wrapped. This is a partial fix 2272 * for the "LAND" DoS attack. 2273 */ 2274 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2275 rstreason = BANDLIM_RST_OPENPORT; 2276 goto dropwithreset; 2277 } 2278 2279 todrop = tp->rcv_nxt - th->th_seq; 2280 if (todrop > 0) { 2281 if (thflags & TH_SYN) { 2282 thflags &= ~TH_SYN; 2283 th->th_seq++; 2284 if (th->th_urp > 1) 2285 th->th_urp--; 2286 else 2287 thflags &= ~TH_URG; 2288 todrop--; 2289 } 2290 /* 2291 * Following if statement from Stevens, vol. 2, p. 960. 2292 */ 2293 if (todrop > tlen 2294 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2295 /* 2296 * Any valid FIN must be to the left of the window. 2297 * At this point the FIN must be a duplicate or out 2298 * of sequence; drop it. 2299 */ 2300 thflags &= ~TH_FIN; 2301 2302 /* 2303 * Send an ACK to resynchronize and drop any data. 2304 * But keep on processing for RST or ACK. 2305 */ 2306 tp->t_flags |= TF_ACKNOW; 2307 todrop = tlen; 2308 TCPSTAT_INC(tcps_rcvduppack); 2309 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2310 } else { 2311 TCPSTAT_INC(tcps_rcvpartduppack); 2312 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2313 } 2314 drop_hdrlen += todrop; /* drop from the top afterwards */ 2315 th->th_seq += todrop; 2316 tlen -= todrop; 2317 if (th->th_urp > todrop) 2318 th->th_urp -= todrop; 2319 else { 2320 thflags &= ~TH_URG; 2321 th->th_urp = 0; 2322 } 2323 } 2324 2325 /* 2326 * If new data are received on a connection after the 2327 * user processes are gone, then RST the other end. 2328 */ 2329 if ((so->so_state & SS_NOFDREF) && 2330 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2331 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2332 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2333 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2334 2335 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2336 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2337 "after socket was closed, " 2338 "sending RST and removing tcpcb\n", 2339 s, __func__, tcpstates[tp->t_state], tlen); 2340 free(s, M_TCPLOG); 2341 } 2342 tp = tcp_close(tp); 2343 TCPSTAT_INC(tcps_rcvafterclose); 2344 rstreason = BANDLIM_UNLIMITED; 2345 goto dropwithreset; 2346 } 2347 2348 /* 2349 * If segment ends after window, drop trailing data 2350 * (and PUSH and FIN); if nothing left, just ACK. 2351 */ 2352 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2353 if (todrop > 0) { 2354 TCPSTAT_INC(tcps_rcvpackafterwin); 2355 if (todrop >= tlen) { 2356 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2357 /* 2358 * If window is closed can only take segments at 2359 * window edge, and have to drop data and PUSH from 2360 * incoming segments. Continue processing, but 2361 * remember to ack. Otherwise, drop segment 2362 * and ack. 2363 */ 2364 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2365 tp->t_flags |= TF_ACKNOW; 2366 TCPSTAT_INC(tcps_rcvwinprobe); 2367 } else 2368 goto dropafterack; 2369 } else 2370 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2371 m_adj(m, -todrop); 2372 tlen -= todrop; 2373 thflags &= ~(TH_PUSH|TH_FIN); 2374 } 2375 2376 /* 2377 * If last ACK falls within this segment's sequence numbers, 2378 * record its timestamp. 2379 * NOTE: 2380 * 1) That the test incorporates suggestions from the latest 2381 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2382 * 2) That updating only on newer timestamps interferes with 2383 * our earlier PAWS tests, so this check should be solely 2384 * predicated on the sequence space of this segment. 2385 * 3) That we modify the segment boundary check to be 2386 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2387 * instead of RFC1323's 2388 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2389 * This modified check allows us to overcome RFC1323's 2390 * limitations as described in Stevens TCP/IP Illustrated 2391 * Vol. 2 p.869. In such cases, we can still calculate the 2392 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2393 */ 2394 if ((to.to_flags & TOF_TS) != 0 && 2395 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2396 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2397 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2398 tp->ts_recent_age = tcp_ts_getticks(); 2399 tp->ts_recent = to.to_tsval; 2400 } 2401 2402 /* 2403 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2404 * flag is on (half-synchronized state), then queue data for 2405 * later processing; else drop segment and return. 2406 */ 2407 if ((thflags & TH_ACK) == 0) { 2408 if (tp->t_state == TCPS_SYN_RECEIVED || 2409 (tp->t_flags & TF_NEEDSYN)) { 2410 #ifdef TCP_RFC7413 2411 if (tp->t_state == TCPS_SYN_RECEIVED && 2412 IS_FASTOPEN(tp->t_flags)) { 2413 tp->snd_wnd = tiwin; 2414 cc_conn_init(tp); 2415 } 2416 #endif 2417 goto step6; 2418 } else if (tp->t_flags & TF_ACKNOW) 2419 goto dropafterack; 2420 else 2421 goto drop; 2422 } 2423 2424 /* 2425 * Ack processing. 2426 */ 2427 switch (tp->t_state) { 2428 2429 /* 2430 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2431 * ESTABLISHED state and continue processing. 2432 * The ACK was checked above. 2433 */ 2434 case TCPS_SYN_RECEIVED: 2435 2436 TCPSTAT_INC(tcps_connects); 2437 soisconnected(so); 2438 /* Do window scaling? */ 2439 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2440 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2441 tp->rcv_scale = tp->request_r_scale; 2442 tp->snd_wnd = tiwin; 2443 } 2444 /* 2445 * Make transitions: 2446 * SYN-RECEIVED -> ESTABLISHED 2447 * SYN-RECEIVED* -> FIN-WAIT-1 2448 */ 2449 tp->t_starttime = ticks; 2450 if (tp->t_flags & TF_NEEDFIN) { 2451 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2452 tp->t_flags &= ~TF_NEEDFIN; 2453 } else { 2454 tcp_state_change(tp, TCPS_ESTABLISHED); 2455 TCP_PROBE5(accept__established, NULL, tp, 2456 m, tp, th); 2457 #ifdef TCP_RFC7413 2458 if (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 #endif 2477 cc_conn_init(tp); 2478 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2479 } 2480 /* 2481 * If segment contains data or ACK, will call tcp_reass() 2482 * later; if not, do so now to pass queued data to user. 2483 */ 2484 if (tlen == 0 && (thflags & TH_FIN) == 0) 2485 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2486 (struct mbuf *)0); 2487 tp->snd_wl1 = th->th_seq - 1; 2488 /* FALLTHROUGH */ 2489 2490 /* 2491 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2492 * ACKs. If the ack is in the range 2493 * tp->snd_una < th->th_ack <= tp->snd_max 2494 * then advance tp->snd_una to th->th_ack and drop 2495 * data from the retransmission queue. If this ACK reflects 2496 * more up to date window information we update our window information. 2497 */ 2498 case TCPS_ESTABLISHED: 2499 case TCPS_FIN_WAIT_1: 2500 case TCPS_FIN_WAIT_2: 2501 case TCPS_CLOSE_WAIT: 2502 case TCPS_CLOSING: 2503 case TCPS_LAST_ACK: 2504 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2505 TCPSTAT_INC(tcps_rcvacktoomuch); 2506 goto dropafterack; 2507 } 2508 if ((tp->t_flags & TF_SACK_PERMIT) && 2509 ((to.to_flags & TOF_SACK) || 2510 !TAILQ_EMPTY(&tp->snd_holes))) 2511 sack_changed = tcp_sack_doack(tp, &to, th->th_ack); 2512 else 2513 /* 2514 * Reset the value so that previous (valid) value 2515 * from the last ack with SACK doesn't get used. 2516 */ 2517 tp->sackhint.sacked_bytes = 0; 2518 2519 #ifdef TCP_HHOOK 2520 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2521 hhook_run_tcp_est_in(tp, th, &to); 2522 #endif 2523 2524 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2525 u_int maxseg; 2526 2527 maxseg = tcp_maxseg(tp); 2528 if (tlen == 0 && 2529 (tiwin == tp->snd_wnd || 2530 (tp->t_flags & TF_SACK_PERMIT))) { 2531 /* 2532 * If this is the first time we've seen a 2533 * FIN from the remote, this is not a 2534 * duplicate and it needs to be processed 2535 * normally. This happens during a 2536 * simultaneous close. 2537 */ 2538 if ((thflags & TH_FIN) && 2539 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2540 tp->t_dupacks = 0; 2541 break; 2542 } 2543 TCPSTAT_INC(tcps_rcvdupack); 2544 /* 2545 * If we have outstanding data (other than 2546 * a window probe), this is a completely 2547 * duplicate ack (ie, window info didn't 2548 * change and FIN isn't set), 2549 * the ack is the biggest we've 2550 * seen and we've seen exactly our rexmt 2551 * threshold of them, assume a packet 2552 * has been dropped and retransmit it. 2553 * Kludge snd_nxt & the congestion 2554 * window so we send only this one 2555 * packet. 2556 * 2557 * We know we're losing at the current 2558 * window size so do congestion avoidance 2559 * (set ssthresh to half the current window 2560 * and pull our congestion window back to 2561 * the new ssthresh). 2562 * 2563 * Dup acks mean that packets have left the 2564 * network (they're now cached at the receiver) 2565 * so bump cwnd by the amount in the receiver 2566 * to keep a constant cwnd packets in the 2567 * network. 2568 * 2569 * When using TCP ECN, notify the peer that 2570 * we reduced the cwnd. 2571 */ 2572 /* 2573 * Following 2 kinds of acks should not affect 2574 * dupack counting: 2575 * 1) Old acks 2576 * 2) Acks with SACK but without any new SACK 2577 * information in them. These could result from 2578 * any anomaly in the network like a switch 2579 * duplicating packets or a possible DoS attack. 2580 */ 2581 if (th->th_ack != tp->snd_una || 2582 ((tp->t_flags & TF_SACK_PERMIT) && 2583 !sack_changed)) 2584 break; 2585 else if (!tcp_timer_active(tp, TT_REXMT)) 2586 tp->t_dupacks = 0; 2587 else if (++tp->t_dupacks > tcprexmtthresh || 2588 IN_FASTRECOVERY(tp->t_flags)) { 2589 cc_ack_received(tp, th, nsegs, 2590 CC_DUPACK); 2591 if ((tp->t_flags & TF_SACK_PERMIT) && 2592 IN_FASTRECOVERY(tp->t_flags)) { 2593 int awnd; 2594 2595 /* 2596 * Compute the amount of data in flight first. 2597 * We can inject new data into the pipe iff 2598 * we have less than 1/2 the original window's 2599 * worth of data in flight. 2600 */ 2601 if (V_tcp_do_rfc6675_pipe) 2602 awnd = tcp_compute_pipe(tp); 2603 else 2604 awnd = (tp->snd_nxt - tp->snd_fack) + 2605 tp->sackhint.sack_bytes_rexmit; 2606 2607 if (awnd < tp->snd_ssthresh) { 2608 tp->snd_cwnd += maxseg; 2609 if (tp->snd_cwnd > tp->snd_ssthresh) 2610 tp->snd_cwnd = tp->snd_ssthresh; 2611 } 2612 } else 2613 tp->snd_cwnd += maxseg; 2614 (void) tp->t_fb->tfb_tcp_output(tp); 2615 goto drop; 2616 } else if (tp->t_dupacks == tcprexmtthresh) { 2617 tcp_seq onxt = tp->snd_nxt; 2618 2619 /* 2620 * If we're doing sack, check to 2621 * see if we're already in sack 2622 * recovery. If we're not doing sack, 2623 * check to see if we're in newreno 2624 * recovery. 2625 */ 2626 if (tp->t_flags & TF_SACK_PERMIT) { 2627 if (IN_FASTRECOVERY(tp->t_flags)) { 2628 tp->t_dupacks = 0; 2629 break; 2630 } 2631 } else { 2632 if (SEQ_LEQ(th->th_ack, 2633 tp->snd_recover)) { 2634 tp->t_dupacks = 0; 2635 break; 2636 } 2637 } 2638 /* Congestion signal before ack. */ 2639 cc_cong_signal(tp, th, CC_NDUPACK); 2640 cc_ack_received(tp, th, nsegs, 2641 CC_DUPACK); 2642 tcp_timer_activate(tp, TT_REXMT, 0); 2643 tp->t_rtttime = 0; 2644 if (tp->t_flags & TF_SACK_PERMIT) { 2645 TCPSTAT_INC( 2646 tcps_sack_recovery_episode); 2647 tp->sack_newdata = tp->snd_nxt; 2648 tp->snd_cwnd = maxseg; 2649 (void) tp->t_fb->tfb_tcp_output(tp); 2650 goto drop; 2651 } 2652 tp->snd_nxt = th->th_ack; 2653 tp->snd_cwnd = maxseg; 2654 (void) tp->t_fb->tfb_tcp_output(tp); 2655 KASSERT(tp->snd_limited <= 2, 2656 ("%s: tp->snd_limited too big", 2657 __func__)); 2658 tp->snd_cwnd = tp->snd_ssthresh + 2659 maxseg * 2660 (tp->t_dupacks - tp->snd_limited); 2661 if (SEQ_GT(onxt, tp->snd_nxt)) 2662 tp->snd_nxt = onxt; 2663 goto drop; 2664 } else if (V_tcp_do_rfc3042) { 2665 /* 2666 * Process first and second duplicate 2667 * ACKs. Each indicates a segment 2668 * leaving the network, creating room 2669 * for more. Make sure we can send a 2670 * packet on reception of each duplicate 2671 * ACK by increasing snd_cwnd by one 2672 * segment. Restore the original 2673 * snd_cwnd after packet transmission. 2674 */ 2675 cc_ack_received(tp, th, nsegs, 2676 CC_DUPACK); 2677 uint32_t oldcwnd = tp->snd_cwnd; 2678 tcp_seq oldsndmax = tp->snd_max; 2679 u_int sent; 2680 int avail; 2681 2682 KASSERT(tp->t_dupacks == 1 || 2683 tp->t_dupacks == 2, 2684 ("%s: dupacks not 1 or 2", 2685 __func__)); 2686 if (tp->t_dupacks == 1) 2687 tp->snd_limited = 0; 2688 tp->snd_cwnd = 2689 (tp->snd_nxt - tp->snd_una) + 2690 (tp->t_dupacks - tp->snd_limited) * 2691 maxseg; 2692 /* 2693 * Only call tcp_output when there 2694 * is new data available to be sent. 2695 * Otherwise we would send pure ACKs. 2696 */ 2697 SOCKBUF_LOCK(&so->so_snd); 2698 avail = sbavail(&so->so_snd) - 2699 (tp->snd_nxt - tp->snd_una); 2700 SOCKBUF_UNLOCK(&so->so_snd); 2701 if (avail > 0) 2702 (void) tp->t_fb->tfb_tcp_output(tp); 2703 sent = tp->snd_max - oldsndmax; 2704 if (sent > maxseg) { 2705 KASSERT((tp->t_dupacks == 2 && 2706 tp->snd_limited == 0) || 2707 (sent == maxseg + 1 && 2708 tp->t_flags & TF_SENTFIN), 2709 ("%s: sent too much", 2710 __func__)); 2711 tp->snd_limited = 2; 2712 } else if (sent > 0) 2713 ++tp->snd_limited; 2714 tp->snd_cwnd = oldcwnd; 2715 goto drop; 2716 } 2717 } 2718 break; 2719 } else { 2720 /* 2721 * This ack is advancing the left edge, reset the 2722 * counter. 2723 */ 2724 tp->t_dupacks = 0; 2725 /* 2726 * If this ack also has new SACK info, increment the 2727 * counter as per rfc6675. 2728 */ 2729 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed) 2730 tp->t_dupacks++; 2731 } 2732 2733 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2734 ("%s: th_ack <= snd_una", __func__)); 2735 2736 /* 2737 * If the congestion window was inflated to account 2738 * for the other side's cached packets, retract it. 2739 */ 2740 if (IN_FASTRECOVERY(tp->t_flags)) { 2741 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2742 if (tp->t_flags & TF_SACK_PERMIT) 2743 tcp_sack_partialack(tp, th); 2744 else 2745 tcp_newreno_partial_ack(tp, th); 2746 } else 2747 cc_post_recovery(tp, th); 2748 } 2749 /* 2750 * If we reach this point, ACK is not a duplicate, 2751 * i.e., it ACKs something we sent. 2752 */ 2753 if (tp->t_flags & TF_NEEDSYN) { 2754 /* 2755 * T/TCP: Connection was half-synchronized, and our 2756 * SYN has been ACK'd (so connection is now fully 2757 * synchronized). Go to non-starred state, 2758 * increment snd_una for ACK of SYN, and check if 2759 * we can do window scaling. 2760 */ 2761 tp->t_flags &= ~TF_NEEDSYN; 2762 tp->snd_una++; 2763 /* Do window scaling? */ 2764 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2765 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2766 tp->rcv_scale = tp->request_r_scale; 2767 /* Send window already scaled. */ 2768 } 2769 } 2770 2771 process_ACK: 2772 INP_WLOCK_ASSERT(tp->t_inpcb); 2773 2774 acked = BYTES_THIS_ACK(tp, th); 2775 KASSERT(acked >= 0, ("%s: acked unexepectedly negative " 2776 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, 2777 tp->snd_una, th->th_ack, tp, m)); 2778 TCPSTAT_ADD(tcps_rcvackpack, nsegs); 2779 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2780 2781 /* 2782 * If we just performed our first retransmit, and the ACK 2783 * arrives within our recovery window, then it was a mistake 2784 * to do the retransmit in the first place. Recover our 2785 * original cwnd and ssthresh, and proceed to transmit where 2786 * we left off. 2787 */ 2788 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2789 (int)(ticks - tp->t_badrxtwin) < 0) 2790 cc_cong_signal(tp, th, CC_RTO_ERR); 2791 2792 /* 2793 * If we have a timestamp reply, update smoothed 2794 * round trip time. If no timestamp is present but 2795 * transmit timer is running and timed sequence 2796 * number was acked, update smoothed round trip time. 2797 * Since we now have an rtt measurement, cancel the 2798 * timer backoff (cf., Phil Karn's retransmit alg.). 2799 * Recompute the initial retransmit timer. 2800 * 2801 * Some boxes send broken timestamp replies 2802 * during the SYN+ACK phase, ignore 2803 * timestamps of 0 or we could calculate a 2804 * huge RTT and blow up the retransmit timer. 2805 */ 2806 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2807 uint32_t t; 2808 2809 t = tcp_ts_getticks() - to.to_tsecr; 2810 if (!tp->t_rttlow || tp->t_rttlow > t) 2811 tp->t_rttlow = t; 2812 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2813 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2814 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2815 tp->t_rttlow = ticks - tp->t_rtttime; 2816 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2817 } 2818 2819 /* 2820 * If all outstanding data is acked, stop retransmit 2821 * timer and remember to restart (more output or persist). 2822 * If there is more data to be acked, restart retransmit 2823 * timer, using current (possibly backed-off) value. 2824 */ 2825 if (th->th_ack == tp->snd_max) { 2826 tcp_timer_activate(tp, TT_REXMT, 0); 2827 needoutput = 1; 2828 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2829 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2830 2831 /* 2832 * If no data (only SYN) was ACK'd, 2833 * skip rest of ACK processing. 2834 */ 2835 if (acked == 0) 2836 goto step6; 2837 2838 /* 2839 * Let the congestion control algorithm update congestion 2840 * control related information. This typically means increasing 2841 * the congestion window. 2842 */ 2843 cc_ack_received(tp, th, nsegs, CC_ACK); 2844 2845 SOCKBUF_LOCK(&so->so_snd); 2846 if (acked > sbavail(&so->so_snd)) { 2847 if (tp->snd_wnd >= sbavail(&so->so_snd)) 2848 tp->snd_wnd -= sbavail(&so->so_snd); 2849 else 2850 tp->snd_wnd = 0; 2851 mfree = sbcut_locked(&so->so_snd, 2852 (int)sbavail(&so->so_snd)); 2853 ourfinisacked = 1; 2854 } else { 2855 mfree = sbcut_locked(&so->so_snd, acked); 2856 if (tp->snd_wnd >= (uint32_t) acked) 2857 tp->snd_wnd -= acked; 2858 else 2859 tp->snd_wnd = 0; 2860 ourfinisacked = 0; 2861 } 2862 /* NB: sowwakeup_locked() does an implicit unlock. */ 2863 sowwakeup_locked(so); 2864 m_freem(mfree); 2865 /* Detect una wraparound. */ 2866 if (!IN_RECOVERY(tp->t_flags) && 2867 SEQ_GT(tp->snd_una, tp->snd_recover) && 2868 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2869 tp->snd_recover = th->th_ack - 1; 2870 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2871 if (IN_RECOVERY(tp->t_flags) && 2872 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2873 EXIT_RECOVERY(tp->t_flags); 2874 } 2875 tp->snd_una = th->th_ack; 2876 if (tp->t_flags & TF_SACK_PERMIT) { 2877 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2878 tp->snd_recover = tp->snd_una; 2879 } 2880 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2881 tp->snd_nxt = tp->snd_una; 2882 2883 switch (tp->t_state) { 2884 2885 /* 2886 * In FIN_WAIT_1 STATE in addition to the processing 2887 * for the ESTABLISHED state if our FIN is now acknowledged 2888 * then enter FIN_WAIT_2. 2889 */ 2890 case TCPS_FIN_WAIT_1: 2891 if (ourfinisacked) { 2892 /* 2893 * If we can't receive any more 2894 * data, then closing user can proceed. 2895 * Starting the timer is contrary to the 2896 * specification, but if we don't get a FIN 2897 * we'll hang forever. 2898 * 2899 * XXXjl: 2900 * we should release the tp also, and use a 2901 * compressed state. 2902 */ 2903 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2904 soisdisconnected(so); 2905 tcp_timer_activate(tp, TT_2MSL, 2906 (tcp_fast_finwait2_recycle ? 2907 tcp_finwait2_timeout : 2908 TP_MAXIDLE(tp))); 2909 } 2910 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2911 } 2912 break; 2913 2914 /* 2915 * In CLOSING STATE in addition to the processing for 2916 * the ESTABLISHED state if the ACK acknowledges our FIN 2917 * then enter the TIME-WAIT state, otherwise ignore 2918 * the segment. 2919 */ 2920 case TCPS_CLOSING: 2921 if (ourfinisacked) { 2922 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2923 tcp_twstart(tp); 2924 INP_INFO_RUNLOCK(&V_tcbinfo); 2925 m_freem(m); 2926 return; 2927 } 2928 break; 2929 2930 /* 2931 * In LAST_ACK, we may still be waiting for data to drain 2932 * and/or to be acked, as well as for the ack of our FIN. 2933 * If our FIN is now acknowledged, delete the TCB, 2934 * enter the closed state and return. 2935 */ 2936 case TCPS_LAST_ACK: 2937 if (ourfinisacked) { 2938 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2939 tp = tcp_close(tp); 2940 goto drop; 2941 } 2942 break; 2943 } 2944 } 2945 2946 step6: 2947 INP_WLOCK_ASSERT(tp->t_inpcb); 2948 2949 /* 2950 * Update window information. 2951 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2952 */ 2953 if ((thflags & TH_ACK) && 2954 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2955 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2956 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2957 /* keep track of pure window updates */ 2958 if (tlen == 0 && 2959 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2960 TCPSTAT_INC(tcps_rcvwinupd); 2961 tp->snd_wnd = tiwin; 2962 tp->snd_wl1 = th->th_seq; 2963 tp->snd_wl2 = th->th_ack; 2964 if (tp->snd_wnd > tp->max_sndwnd) 2965 tp->max_sndwnd = tp->snd_wnd; 2966 needoutput = 1; 2967 } 2968 2969 /* 2970 * Process segments with URG. 2971 */ 2972 if ((thflags & TH_URG) && th->th_urp && 2973 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2974 /* 2975 * This is a kludge, but if we receive and accept 2976 * random urgent pointers, we'll crash in 2977 * soreceive. It's hard to imagine someone 2978 * actually wanting to send this much urgent data. 2979 */ 2980 SOCKBUF_LOCK(&so->so_rcv); 2981 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2982 th->th_urp = 0; /* XXX */ 2983 thflags &= ~TH_URG; /* XXX */ 2984 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2985 goto dodata; /* XXX */ 2986 } 2987 /* 2988 * If this segment advances the known urgent pointer, 2989 * then mark the data stream. This should not happen 2990 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2991 * a FIN has been received from the remote side. 2992 * In these states we ignore the URG. 2993 * 2994 * According to RFC961 (Assigned Protocols), 2995 * the urgent pointer points to the last octet 2996 * of urgent data. We continue, however, 2997 * to consider it to indicate the first octet 2998 * of data past the urgent section as the original 2999 * spec states (in one of two places). 3000 */ 3001 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 3002 tp->rcv_up = th->th_seq + th->th_urp; 3003 so->so_oobmark = sbavail(&so->so_rcv) + 3004 (tp->rcv_up - tp->rcv_nxt) - 1; 3005 if (so->so_oobmark == 0) 3006 so->so_rcv.sb_state |= SBS_RCVATMARK; 3007 sohasoutofband(so); 3008 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 3009 } 3010 SOCKBUF_UNLOCK(&so->so_rcv); 3011 /* 3012 * Remove out of band data so doesn't get presented to user. 3013 * This can happen independent of advancing the URG pointer, 3014 * but if two URG's are pending at once, some out-of-band 3015 * data may creep in... ick. 3016 */ 3017 if (th->th_urp <= (uint32_t)tlen && 3018 !(so->so_options & SO_OOBINLINE)) { 3019 /* hdr drop is delayed */ 3020 tcp_pulloutofband(so, th, m, drop_hdrlen); 3021 } 3022 } else { 3023 /* 3024 * If no out of band data is expected, 3025 * pull receive urgent pointer along 3026 * with the receive window. 3027 */ 3028 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 3029 tp->rcv_up = tp->rcv_nxt; 3030 } 3031 dodata: /* XXX */ 3032 INP_WLOCK_ASSERT(tp->t_inpcb); 3033 3034 /* 3035 * Process the segment text, merging it into the TCP sequencing queue, 3036 * and arranging for acknowledgment of receipt if necessary. 3037 * This process logically involves adjusting tp->rcv_wnd as data 3038 * is presented to the user (this happens in tcp_usrreq.c, 3039 * case PRU_RCVD). If a FIN has already been received on this 3040 * connection then we just ignore the text. 3041 */ 3042 #ifdef TCP_RFC7413 3043 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && 3044 IS_FASTOPEN(tp->t_flags)); 3045 #else 3046 #define tfo_syn (false) 3047 #endif 3048 if ((tlen || (thflags & TH_FIN) || tfo_syn) && 3049 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3050 tcp_seq save_start = th->th_seq; 3051 m_adj(m, drop_hdrlen); /* delayed header drop */ 3052 /* 3053 * Insert segment which includes th into TCP reassembly queue 3054 * with control block tp. Set thflags to whether reassembly now 3055 * includes a segment with FIN. This handles the common case 3056 * inline (segment is the next to be received on an established 3057 * connection, and the queue is empty), avoiding linkage into 3058 * and removal from the queue and repetition of various 3059 * conversions. 3060 * Set DELACK for segments received in order, but ack 3061 * immediately when segments are out of order (so 3062 * fast retransmit can work). 3063 */ 3064 if (th->th_seq == tp->rcv_nxt && 3065 LIST_EMPTY(&tp->t_segq) && 3066 (TCPS_HAVEESTABLISHED(tp->t_state) || 3067 tfo_syn)) { 3068 if (DELAY_ACK(tp, tlen) || tfo_syn) 3069 tp->t_flags |= TF_DELACK; 3070 else 3071 tp->t_flags |= TF_ACKNOW; 3072 tp->rcv_nxt += tlen; 3073 thflags = th->th_flags & TH_FIN; 3074 TCPSTAT_INC(tcps_rcvpack); 3075 TCPSTAT_ADD(tcps_rcvbyte, tlen); 3076 SOCKBUF_LOCK(&so->so_rcv); 3077 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 3078 m_freem(m); 3079 else 3080 sbappendstream_locked(&so->so_rcv, m, 0); 3081 /* NB: sorwakeup_locked() does an implicit unlock. */ 3082 sorwakeup_locked(so); 3083 } else { 3084 /* 3085 * XXX: Due to the header drop above "th" is 3086 * theoretically invalid by now. Fortunately 3087 * m_adj() doesn't actually frees any mbufs 3088 * when trimming from the head. 3089 */ 3090 thflags = tcp_reass(tp, th, &tlen, m); 3091 tp->t_flags |= TF_ACKNOW; 3092 } 3093 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 3094 tcp_update_sack_list(tp, save_start, save_start + tlen); 3095 #if 0 3096 /* 3097 * Note the amount of data that peer has sent into 3098 * our window, in order to estimate the sender's 3099 * buffer size. 3100 * XXX: Unused. 3101 */ 3102 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 3103 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 3104 else 3105 len = so->so_rcv.sb_hiwat; 3106 #endif 3107 } else { 3108 m_freem(m); 3109 thflags &= ~TH_FIN; 3110 } 3111 3112 /* 3113 * If FIN is received ACK the FIN and let the user know 3114 * that the connection is closing. 3115 */ 3116 if (thflags & TH_FIN) { 3117 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 3118 socantrcvmore(so); 3119 /* 3120 * If connection is half-synchronized 3121 * (ie NEEDSYN flag on) then delay ACK, 3122 * so it may be piggybacked when SYN is sent. 3123 * Otherwise, since we received a FIN then no 3124 * more input can be expected, send ACK now. 3125 */ 3126 if (tp->t_flags & TF_NEEDSYN) 3127 tp->t_flags |= TF_DELACK; 3128 else 3129 tp->t_flags |= TF_ACKNOW; 3130 tp->rcv_nxt++; 3131 } 3132 switch (tp->t_state) { 3133 3134 /* 3135 * In SYN_RECEIVED and ESTABLISHED STATES 3136 * enter the CLOSE_WAIT state. 3137 */ 3138 case TCPS_SYN_RECEIVED: 3139 tp->t_starttime = ticks; 3140 /* FALLTHROUGH */ 3141 case TCPS_ESTABLISHED: 3142 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3143 break; 3144 3145 /* 3146 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3147 * enter the CLOSING state. 3148 */ 3149 case TCPS_FIN_WAIT_1: 3150 tcp_state_change(tp, TCPS_CLOSING); 3151 break; 3152 3153 /* 3154 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3155 * starting the time-wait timer, turning off the other 3156 * standard timers. 3157 */ 3158 case TCPS_FIN_WAIT_2: 3159 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3160 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3161 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3162 ti_locked)); 3163 3164 tcp_twstart(tp); 3165 INP_INFO_RUNLOCK(&V_tcbinfo); 3166 return; 3167 } 3168 } 3169 if (ti_locked == TI_RLOCKED) 3170 INP_INFO_RUNLOCK(&V_tcbinfo); 3171 ti_locked = TI_UNLOCKED; 3172 3173 #ifdef TCPDEBUG 3174 if (so->so_options & SO_DEBUG) 3175 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3176 &tcp_savetcp, 0); 3177 #endif 3178 TCP_PROBE3(debug__input, tp, th, m); 3179 3180 /* 3181 * Return any desired output. 3182 */ 3183 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3184 (void) tp->t_fb->tfb_tcp_output(tp); 3185 3186 check_delack: 3187 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3188 __func__, ti_locked)); 3189 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3190 INP_WLOCK_ASSERT(tp->t_inpcb); 3191 3192 if (tp->t_flags & TF_DELACK) { 3193 tp->t_flags &= ~TF_DELACK; 3194 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3195 } 3196 INP_WUNLOCK(tp->t_inpcb); 3197 return; 3198 3199 dropafterack: 3200 /* 3201 * Generate an ACK dropping incoming segment if it occupies 3202 * sequence space, where the ACK reflects our state. 3203 * 3204 * We can now skip the test for the RST flag since all 3205 * paths to this code happen after packets containing 3206 * RST have been dropped. 3207 * 3208 * In the SYN-RECEIVED state, don't send an ACK unless the 3209 * segment we received passes the SYN-RECEIVED ACK test. 3210 * If it fails send a RST. This breaks the loop in the 3211 * "LAND" DoS attack, and also prevents an ACK storm 3212 * between two listening ports that have been sent forged 3213 * SYN segments, each with the source address of the other. 3214 */ 3215 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3216 (SEQ_GT(tp->snd_una, th->th_ack) || 3217 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3218 rstreason = BANDLIM_RST_OPENPORT; 3219 goto dropwithreset; 3220 } 3221 #ifdef TCPDEBUG 3222 if (so->so_options & SO_DEBUG) 3223 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3224 &tcp_savetcp, 0); 3225 #endif 3226 TCP_PROBE3(debug__input, tp, th, m); 3227 if (ti_locked == TI_RLOCKED) 3228 INP_INFO_RUNLOCK(&V_tcbinfo); 3229 ti_locked = TI_UNLOCKED; 3230 3231 tp->t_flags |= TF_ACKNOW; 3232 (void) tp->t_fb->tfb_tcp_output(tp); 3233 INP_WUNLOCK(tp->t_inpcb); 3234 m_freem(m); 3235 return; 3236 3237 dropwithreset: 3238 if (ti_locked == TI_RLOCKED) 3239 INP_INFO_RUNLOCK(&V_tcbinfo); 3240 ti_locked = TI_UNLOCKED; 3241 3242 if (tp != NULL) { 3243 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3244 INP_WUNLOCK(tp->t_inpcb); 3245 } else 3246 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3247 return; 3248 3249 drop: 3250 if (ti_locked == TI_RLOCKED) { 3251 INP_INFO_RUNLOCK(&V_tcbinfo); 3252 ti_locked = TI_UNLOCKED; 3253 } 3254 #ifdef INVARIANTS 3255 else 3256 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3257 #endif 3258 3259 /* 3260 * Drop space held by incoming segment and return. 3261 */ 3262 #ifdef TCPDEBUG 3263 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3264 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3265 &tcp_savetcp, 0); 3266 #endif 3267 TCP_PROBE3(debug__input, tp, th, m); 3268 if (tp != NULL) 3269 INP_WUNLOCK(tp->t_inpcb); 3270 m_freem(m); 3271 #ifndef TCP_RFC7413 3272 #undef tfo_syn 3273 #endif 3274 } 3275 3276 /* 3277 * Issue RST and make ACK acceptable to originator of segment. 3278 * The mbuf must still include the original packet header. 3279 * tp may be NULL. 3280 */ 3281 void 3282 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3283 int tlen, int rstreason) 3284 { 3285 #ifdef INET 3286 struct ip *ip; 3287 #endif 3288 #ifdef INET6 3289 struct ip6_hdr *ip6; 3290 #endif 3291 3292 if (tp != NULL) { 3293 INP_WLOCK_ASSERT(tp->t_inpcb); 3294 } 3295 3296 /* Don't bother if destination was broadcast/multicast. */ 3297 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3298 goto drop; 3299 #ifdef INET6 3300 if (mtod(m, struct ip *)->ip_v == 6) { 3301 ip6 = mtod(m, struct ip6_hdr *); 3302 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3303 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3304 goto drop; 3305 /* IPv6 anycast check is done at tcp6_input() */ 3306 } 3307 #endif 3308 #if defined(INET) && defined(INET6) 3309 else 3310 #endif 3311 #ifdef INET 3312 { 3313 ip = mtod(m, struct ip *); 3314 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3315 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3316 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3317 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3318 goto drop; 3319 } 3320 #endif 3321 3322 /* Perform bandwidth limiting. */ 3323 if (badport_bandlim(rstreason) < 0) 3324 goto drop; 3325 3326 /* tcp_respond consumes the mbuf chain. */ 3327 if (th->th_flags & TH_ACK) { 3328 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3329 th->th_ack, TH_RST); 3330 } else { 3331 if (th->th_flags & TH_SYN) 3332 tlen++; 3333 if (th->th_flags & TH_FIN) 3334 tlen++; 3335 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3336 (tcp_seq)0, TH_RST|TH_ACK); 3337 } 3338 return; 3339 drop: 3340 m_freem(m); 3341 } 3342 3343 /* 3344 * Parse TCP options and place in tcpopt. 3345 */ 3346 void 3347 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3348 { 3349 int opt, optlen; 3350 3351 to->to_flags = 0; 3352 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3353 opt = cp[0]; 3354 if (opt == TCPOPT_EOL) 3355 break; 3356 if (opt == TCPOPT_NOP) 3357 optlen = 1; 3358 else { 3359 if (cnt < 2) 3360 break; 3361 optlen = cp[1]; 3362 if (optlen < 2 || optlen > cnt) 3363 break; 3364 } 3365 switch (opt) { 3366 case TCPOPT_MAXSEG: 3367 if (optlen != TCPOLEN_MAXSEG) 3368 continue; 3369 if (!(flags & TO_SYN)) 3370 continue; 3371 to->to_flags |= TOF_MSS; 3372 bcopy((char *)cp + 2, 3373 (char *)&to->to_mss, sizeof(to->to_mss)); 3374 to->to_mss = ntohs(to->to_mss); 3375 break; 3376 case TCPOPT_WINDOW: 3377 if (optlen != TCPOLEN_WINDOW) 3378 continue; 3379 if (!(flags & TO_SYN)) 3380 continue; 3381 to->to_flags |= TOF_SCALE; 3382 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3383 break; 3384 case TCPOPT_TIMESTAMP: 3385 if (optlen != TCPOLEN_TIMESTAMP) 3386 continue; 3387 to->to_flags |= TOF_TS; 3388 bcopy((char *)cp + 2, 3389 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3390 to->to_tsval = ntohl(to->to_tsval); 3391 bcopy((char *)cp + 6, 3392 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3393 to->to_tsecr = ntohl(to->to_tsecr); 3394 break; 3395 case TCPOPT_SIGNATURE: 3396 /* 3397 * In order to reply to a host which has set the 3398 * TCP_SIGNATURE option in its initial SYN, we have 3399 * to record the fact that the option was observed 3400 * here for the syncache code to perform the correct 3401 * response. 3402 */ 3403 if (optlen != TCPOLEN_SIGNATURE) 3404 continue; 3405 to->to_flags |= TOF_SIGNATURE; 3406 to->to_signature = cp + 2; 3407 break; 3408 case TCPOPT_SACK_PERMITTED: 3409 if (optlen != TCPOLEN_SACK_PERMITTED) 3410 continue; 3411 if (!(flags & TO_SYN)) 3412 continue; 3413 if (!V_tcp_do_sack) 3414 continue; 3415 to->to_flags |= TOF_SACKPERM; 3416 break; 3417 case TCPOPT_SACK: 3418 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3419 continue; 3420 if (flags & TO_SYN) 3421 continue; 3422 to->to_flags |= TOF_SACK; 3423 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3424 to->to_sacks = cp + 2; 3425 TCPSTAT_INC(tcps_sack_rcv_blocks); 3426 break; 3427 #ifdef TCP_RFC7413 3428 case TCPOPT_FAST_OPEN: 3429 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) && 3430 (optlen < TCPOLEN_FAST_OPEN_MIN) && 3431 (optlen > TCPOLEN_FAST_OPEN_MAX)) 3432 continue; 3433 if (!(flags & TO_SYN)) 3434 continue; 3435 if (!V_tcp_fastopen_enabled) 3436 continue; 3437 to->to_flags |= TOF_FASTOPEN; 3438 to->to_tfo_len = optlen - 2; 3439 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; 3440 break; 3441 #endif 3442 default: 3443 continue; 3444 } 3445 } 3446 } 3447 3448 /* 3449 * Pull out of band byte out of a segment so 3450 * it doesn't appear in the user's data queue. 3451 * It is still reflected in the segment length for 3452 * sequencing purposes. 3453 */ 3454 void 3455 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3456 int off) 3457 { 3458 int cnt = off + th->th_urp - 1; 3459 3460 while (cnt >= 0) { 3461 if (m->m_len > cnt) { 3462 char *cp = mtod(m, caddr_t) + cnt; 3463 struct tcpcb *tp = sototcpcb(so); 3464 3465 INP_WLOCK_ASSERT(tp->t_inpcb); 3466 3467 tp->t_iobc = *cp; 3468 tp->t_oobflags |= TCPOOB_HAVEDATA; 3469 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3470 m->m_len--; 3471 if (m->m_flags & M_PKTHDR) 3472 m->m_pkthdr.len--; 3473 return; 3474 } 3475 cnt -= m->m_len; 3476 m = m->m_next; 3477 if (m == NULL) 3478 break; 3479 } 3480 panic("tcp_pulloutofband"); 3481 } 3482 3483 /* 3484 * Collect new round-trip time estimate 3485 * and update averages and current timeout. 3486 */ 3487 void 3488 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3489 { 3490 int delta; 3491 3492 INP_WLOCK_ASSERT(tp->t_inpcb); 3493 3494 TCPSTAT_INC(tcps_rttupdated); 3495 tp->t_rttupdated++; 3496 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { 3497 /* 3498 * srtt is stored as fixed point with 5 bits after the 3499 * binary point (i.e., scaled by 8). The following magic 3500 * is equivalent to the smoothing algorithm in rfc793 with 3501 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3502 * point). Adjust rtt to origin 0. 3503 */ 3504 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3505 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3506 3507 if ((tp->t_srtt += delta) <= 0) 3508 tp->t_srtt = 1; 3509 3510 /* 3511 * We accumulate a smoothed rtt variance (actually, a 3512 * smoothed mean difference), then set the retransmit 3513 * timer to smoothed rtt + 4 times the smoothed variance. 3514 * rttvar is stored as fixed point with 4 bits after the 3515 * binary point (scaled by 16). The following is 3516 * equivalent to rfc793 smoothing with an alpha of .75 3517 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3518 * rfc793's wired-in beta. 3519 */ 3520 if (delta < 0) 3521 delta = -delta; 3522 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3523 if ((tp->t_rttvar += delta) <= 0) 3524 tp->t_rttvar = 1; 3525 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3526 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3527 } else { 3528 /* 3529 * No rtt measurement yet - use the unsmoothed rtt. 3530 * Set the variance to half the rtt (so our first 3531 * retransmit happens at 3*rtt). 3532 */ 3533 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3534 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3535 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3536 } 3537 tp->t_rtttime = 0; 3538 tp->t_rxtshift = 0; 3539 3540 /* 3541 * the retransmit should happen at rtt + 4 * rttvar. 3542 * Because of the way we do the smoothing, srtt and rttvar 3543 * will each average +1/2 tick of bias. When we compute 3544 * the retransmit timer, we want 1/2 tick of rounding and 3545 * 1 extra tick because of +-1/2 tick uncertainty in the 3546 * firing of the timer. The bias will give us exactly the 3547 * 1.5 tick we need. But, because the bias is 3548 * statistical, we have to test that we don't drop below 3549 * the minimum feasible timer (which is 2 ticks). 3550 */ 3551 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3552 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3553 3554 /* 3555 * We received an ack for a packet that wasn't retransmitted; 3556 * it is probably safe to discard any error indications we've 3557 * received recently. This isn't quite right, but close enough 3558 * for now (a route might have failed after we sent a segment, 3559 * and the return path might not be symmetrical). 3560 */ 3561 tp->t_softerror = 0; 3562 } 3563 3564 /* 3565 * Determine a reasonable value for maxseg size. 3566 * If the route is known, check route for mtu. 3567 * If none, use an mss that can be handled on the outgoing interface 3568 * without forcing IP to fragment. If no route is found, route has no mtu, 3569 * or the destination isn't local, use a default, hopefully conservative 3570 * size (usually 512 or the default IP max size, but no more than the mtu 3571 * of the interface), as we can't discover anything about intervening 3572 * gateways or networks. We also initialize the congestion/slow start 3573 * window to be a single segment if the destination isn't local. 3574 * While looking at the routing entry, we also initialize other path-dependent 3575 * parameters from pre-set or cached values in the routing entry. 3576 * 3577 * NOTE that resulting t_maxseg doesn't include space for TCP options or 3578 * IP options, e.g. IPSEC data, since length of this data may vary, and 3579 * thus it is calculated for every segment separately in tcp_output(). 3580 * 3581 * NOTE that this routine is only called when we process an incoming 3582 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3583 * settings are handled in tcp_mssopt(). 3584 */ 3585 void 3586 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3587 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3588 { 3589 int mss = 0; 3590 uint32_t maxmtu = 0; 3591 struct inpcb *inp = tp->t_inpcb; 3592 struct hc_metrics_lite metrics; 3593 #ifdef INET6 3594 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3595 size_t min_protoh = isipv6 ? 3596 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3597 sizeof (struct tcpiphdr); 3598 #else 3599 const size_t min_protoh = sizeof(struct tcpiphdr); 3600 #endif 3601 3602 INP_WLOCK_ASSERT(tp->t_inpcb); 3603 3604 if (mtuoffer != -1) { 3605 KASSERT(offer == -1, ("%s: conflict", __func__)); 3606 offer = mtuoffer - min_protoh; 3607 } 3608 3609 /* Initialize. */ 3610 #ifdef INET6 3611 if (isipv6) { 3612 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3613 tp->t_maxseg = V_tcp_v6mssdflt; 3614 } 3615 #endif 3616 #if defined(INET) && defined(INET6) 3617 else 3618 #endif 3619 #ifdef INET 3620 { 3621 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3622 tp->t_maxseg = V_tcp_mssdflt; 3623 } 3624 #endif 3625 3626 /* 3627 * No route to sender, stay with default mss and return. 3628 */ 3629 if (maxmtu == 0) { 3630 /* 3631 * In case we return early we need to initialize metrics 3632 * to a defined state as tcp_hc_get() would do for us 3633 * if there was no cache hit. 3634 */ 3635 if (metricptr != NULL) 3636 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3637 return; 3638 } 3639 3640 /* What have we got? */ 3641 switch (offer) { 3642 case 0: 3643 /* 3644 * Offer == 0 means that there was no MSS on the SYN 3645 * segment, in this case we use tcp_mssdflt as 3646 * already assigned to t_maxseg above. 3647 */ 3648 offer = tp->t_maxseg; 3649 break; 3650 3651 case -1: 3652 /* 3653 * Offer == -1 means that we didn't receive SYN yet. 3654 */ 3655 /* FALLTHROUGH */ 3656 3657 default: 3658 /* 3659 * Prevent DoS attack with too small MSS. Round up 3660 * to at least minmss. 3661 */ 3662 offer = max(offer, V_tcp_minmss); 3663 } 3664 3665 /* 3666 * rmx information is now retrieved from tcp_hostcache. 3667 */ 3668 tcp_hc_get(&inp->inp_inc, &metrics); 3669 if (metricptr != NULL) 3670 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3671 3672 /* 3673 * If there's a discovered mtu in tcp hostcache, use it. 3674 * Else, use the link mtu. 3675 */ 3676 if (metrics.rmx_mtu) 3677 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3678 else { 3679 #ifdef INET6 3680 if (isipv6) { 3681 mss = maxmtu - min_protoh; 3682 if (!V_path_mtu_discovery && 3683 !in6_localaddr(&inp->in6p_faddr)) 3684 mss = min(mss, V_tcp_v6mssdflt); 3685 } 3686 #endif 3687 #if defined(INET) && defined(INET6) 3688 else 3689 #endif 3690 #ifdef INET 3691 { 3692 mss = maxmtu - min_protoh; 3693 if (!V_path_mtu_discovery && 3694 !in_localaddr(inp->inp_faddr)) 3695 mss = min(mss, V_tcp_mssdflt); 3696 } 3697 #endif 3698 /* 3699 * XXX - The above conditional (mss = maxmtu - min_protoh) 3700 * probably violates the TCP spec. 3701 * The problem is that, since we don't know the 3702 * other end's MSS, we are supposed to use a conservative 3703 * default. But, if we do that, then MTU discovery will 3704 * never actually take place, because the conservative 3705 * default is much less than the MTUs typically seen 3706 * on the Internet today. For the moment, we'll sweep 3707 * this under the carpet. 3708 * 3709 * The conservative default might not actually be a problem 3710 * if the only case this occurs is when sending an initial 3711 * SYN with options and data to a host we've never talked 3712 * to before. Then, they will reply with an MSS value which 3713 * will get recorded and the new parameters should get 3714 * recomputed. For Further Study. 3715 */ 3716 } 3717 mss = min(mss, offer); 3718 3719 /* 3720 * Sanity check: make sure that maxseg will be large 3721 * enough to allow some data on segments even if the 3722 * all the option space is used (40bytes). Otherwise 3723 * funny things may happen in tcp_output. 3724 * 3725 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3726 */ 3727 mss = max(mss, 64); 3728 3729 tp->t_maxseg = mss; 3730 } 3731 3732 void 3733 tcp_mss(struct tcpcb *tp, int offer) 3734 { 3735 int mss; 3736 uint32_t bufsize; 3737 struct inpcb *inp; 3738 struct socket *so; 3739 struct hc_metrics_lite metrics; 3740 struct tcp_ifcap cap; 3741 3742 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3743 3744 bzero(&cap, sizeof(cap)); 3745 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3746 3747 mss = tp->t_maxseg; 3748 inp = tp->t_inpcb; 3749 3750 /* 3751 * If there's a pipesize, change the socket buffer to that size, 3752 * don't change if sb_hiwat is different than default (then it 3753 * has been changed on purpose with setsockopt). 3754 * Make the socket buffers an integral number of mss units; 3755 * if the mss is larger than the socket buffer, decrease the mss. 3756 */ 3757 so = inp->inp_socket; 3758 SOCKBUF_LOCK(&so->so_snd); 3759 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3760 bufsize = metrics.rmx_sendpipe; 3761 else 3762 bufsize = so->so_snd.sb_hiwat; 3763 if (bufsize < mss) 3764 mss = bufsize; 3765 else { 3766 bufsize = roundup(bufsize, mss); 3767 if (bufsize > sb_max) 3768 bufsize = sb_max; 3769 if (bufsize > so->so_snd.sb_hiwat) 3770 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3771 } 3772 SOCKBUF_UNLOCK(&so->so_snd); 3773 /* 3774 * Sanity check: make sure that maxseg will be large 3775 * enough to allow some data on segments even if the 3776 * all the option space is used (40bytes). Otherwise 3777 * funny things may happen in tcp_output. 3778 * 3779 * XXXGL: shouldn't we reserve space for IP/IPv6 options? 3780 */ 3781 tp->t_maxseg = max(mss, 64); 3782 3783 SOCKBUF_LOCK(&so->so_rcv); 3784 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3785 bufsize = metrics.rmx_recvpipe; 3786 else 3787 bufsize = so->so_rcv.sb_hiwat; 3788 if (bufsize > mss) { 3789 bufsize = roundup(bufsize, mss); 3790 if (bufsize > sb_max) 3791 bufsize = sb_max; 3792 if (bufsize > so->so_rcv.sb_hiwat) 3793 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3794 } 3795 SOCKBUF_UNLOCK(&so->so_rcv); 3796 3797 /* Check the interface for TSO capabilities. */ 3798 if (cap.ifcap & CSUM_TSO) { 3799 tp->t_flags |= TF_TSO; 3800 tp->t_tsomax = cap.tsomax; 3801 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3802 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3803 } 3804 } 3805 3806 /* 3807 * Determine the MSS option to send on an outgoing SYN. 3808 */ 3809 int 3810 tcp_mssopt(struct in_conninfo *inc) 3811 { 3812 int mss = 0; 3813 uint32_t thcmtu = 0; 3814 uint32_t maxmtu = 0; 3815 size_t min_protoh; 3816 3817 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3818 3819 #ifdef INET6 3820 if (inc->inc_flags & INC_ISIPV6) { 3821 mss = V_tcp_v6mssdflt; 3822 maxmtu = tcp_maxmtu6(inc, NULL); 3823 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3824 } 3825 #endif 3826 #if defined(INET) && defined(INET6) 3827 else 3828 #endif 3829 #ifdef INET 3830 { 3831 mss = V_tcp_mssdflt; 3832 maxmtu = tcp_maxmtu(inc, NULL); 3833 min_protoh = sizeof(struct tcpiphdr); 3834 } 3835 #endif 3836 #if defined(INET6) || defined(INET) 3837 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3838 #endif 3839 3840 if (maxmtu && thcmtu) 3841 mss = min(maxmtu, thcmtu) - min_protoh; 3842 else if (maxmtu || thcmtu) 3843 mss = max(maxmtu, thcmtu) - min_protoh; 3844 3845 return (mss); 3846 } 3847 3848 3849 /* 3850 * On a partial ack arrives, force the retransmission of the 3851 * next unacknowledged segment. Do not clear tp->t_dupacks. 3852 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3853 * be started again. 3854 */ 3855 void 3856 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3857 { 3858 tcp_seq onxt = tp->snd_nxt; 3859 uint32_t ocwnd = tp->snd_cwnd; 3860 u_int maxseg = tcp_maxseg(tp); 3861 3862 INP_WLOCK_ASSERT(tp->t_inpcb); 3863 3864 tcp_timer_activate(tp, TT_REXMT, 0); 3865 tp->t_rtttime = 0; 3866 tp->snd_nxt = th->th_ack; 3867 /* 3868 * Set snd_cwnd to one segment beyond acknowledged offset. 3869 * (tp->snd_una has not yet been updated when this function is called.) 3870 */ 3871 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); 3872 tp->t_flags |= TF_ACKNOW; 3873 (void) tp->t_fb->tfb_tcp_output(tp); 3874 tp->snd_cwnd = ocwnd; 3875 if (SEQ_GT(onxt, tp->snd_nxt)) 3876 tp->snd_nxt = onxt; 3877 /* 3878 * Partial window deflation. Relies on fact that tp->snd_una 3879 * not updated yet. 3880 */ 3881 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3882 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3883 else 3884 tp->snd_cwnd = 0; 3885 tp->snd_cwnd += maxseg; 3886 } 3887 3888 int 3889 tcp_compute_pipe(struct tcpcb *tp) 3890 { 3891 return (tp->snd_max - tp->snd_una + 3892 tp->sackhint.sack_bytes_rexmit - 3893 tp->sackhint.sacked_bytes); 3894 } 3895