1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 4. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_ipfw.h" /* for ipfw_fwd */ 54 #include "opt_inet.h" 55 #include "opt_inet6.h" 56 #include "opt_ipsec.h" 57 #include "opt_tcpdebug.h" 58 59 #include <sys/param.h> 60 #include <sys/kernel.h> 61 #include <sys/hhook.h> 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/sdt.h> 67 #include <sys/signalvar.h> 68 #include <sys/socket.h> 69 #include <sys/socketvar.h> 70 #include <sys/sysctl.h> 71 #include <sys/syslog.h> 72 #include <sys/systm.h> 73 74 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 75 76 #include <vm/uma.h> 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/route.h> 81 #include <net/vnet.h> 82 83 #define TCPSTATES /* for logging */ 84 85 #include <netinet/cc.h> 86 #include <netinet/in.h> 87 #include <netinet/in_kdtrace.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/in_var.h> 91 #include <netinet/ip.h> 92 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 93 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 94 #include <netinet/ip_var.h> 95 #include <netinet/ip_options.h> 96 #include <netinet/ip6.h> 97 #include <netinet/icmp6.h> 98 #include <netinet6/in6_pcb.h> 99 #include <netinet6/ip6_var.h> 100 #include <netinet6/nd6.h> 101 #include <netinet/tcp_fsm.h> 102 #include <netinet/tcp_seq.h> 103 #include <netinet/tcp_timer.h> 104 #include <netinet/tcp_var.h> 105 #include <netinet6/tcp6_var.h> 106 #include <netinet/tcpip.h> 107 #include <netinet/tcp_syncache.h> 108 #ifdef TCPDEBUG 109 #include <netinet/tcp_debug.h> 110 #endif /* TCPDEBUG */ 111 #ifdef TCP_OFFLOAD 112 #include <netinet/tcp_offload.h> 113 #endif 114 115 #ifdef IPSEC 116 #include <netipsec/ipsec.h> 117 #include <netipsec/ipsec6.h> 118 #endif /*IPSEC*/ 119 120 #include <machine/in_cksum.h> 121 122 #include <security/mac/mac_framework.h> 123 124 const int tcprexmtthresh = 3; 125 126 int tcp_log_in_vain = 0; 127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 128 &tcp_log_in_vain, 0, 129 "Log all incoming TCP segments to closed ports"); 130 131 VNET_DEFINE(int, blackhole) = 0; 132 #define V_blackhole VNET(blackhole) 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 134 &VNET_NAME(blackhole), 0, 135 "Do not send RST on segments to closed ports"); 136 137 VNET_DEFINE(int, tcp_delack_enabled) = 1; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(tcp_delack_enabled), 0, 140 "Delay ACK to try and piggyback it onto a data packet"); 141 142 VNET_DEFINE(int, drop_synfin) = 0; 143 #define V_drop_synfin VNET(drop_synfin) 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 145 &VNET_NAME(drop_synfin), 0, 146 "Drop TCP packets with SYN+FIN set"); 147 148 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 149 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 151 &VNET_NAME(tcp_do_rfc3042), 0, 152 "Enable RFC 3042 (Limited Transmit)"); 153 154 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 156 &VNET_NAME(tcp_do_rfc3390), 0, 157 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 158 159 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 160 "Experimental TCP extensions"); 161 162 VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 163 SYSCTL_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_initcwnd10), 0, 165 "Enable RFC 6928 (Increasing initial CWND to 10)"); 166 167 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_rfc3465), 0, 170 "Enable RFC 3465 (Appropriate Byte Counting)"); 171 172 VNET_DEFINE(int, tcp_abc_l_var) = 2; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_abc_l_var), 2, 175 "Cap the max cwnd increment during slow-start to this number of segments"); 176 177 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 178 179 VNET_DEFINE(int, tcp_do_ecn) = 0; 180 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_do_ecn), 0, 182 "TCP ECN support"); 183 184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 185 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 186 &VNET_NAME(tcp_ecn_maxretries), 0, 187 "Max retries before giving up on ECN"); 188 189 VNET_DEFINE(int, tcp_insecure_syn) = 0; 190 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_insecure_syn), 0, 193 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 194 195 VNET_DEFINE(int, tcp_insecure_rst) = 0; 196 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_insecure_rst), 0, 199 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 200 201 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 202 #define V_tcp_recvspace VNET(tcp_recvspace) 203 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 204 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 205 206 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 207 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 209 &VNET_NAME(tcp_do_autorcvbuf), 0, 210 "Enable automatic receive buffer sizing"); 211 212 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 213 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 215 &VNET_NAME(tcp_autorcvbuf_inc), 0, 216 "Incrementor step size of automatic receive buffer"); 217 218 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 219 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 221 &VNET_NAME(tcp_autorcvbuf_max), 0, 222 "Max size of automatic receive buffer"); 223 224 VNET_DEFINE(struct inpcbhead, tcb); 225 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 226 VNET_DEFINE(struct inpcbinfo, tcbinfo); 227 228 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 229 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 230 struct socket *, struct tcpcb *, int, int, uint8_t, 231 int); 232 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 233 struct tcpcb *, int, int); 234 static void tcp_pulloutofband(struct socket *, 235 struct tcphdr *, struct mbuf *, int); 236 static void tcp_xmit_timer(struct tcpcb *, int); 237 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 238 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 239 uint16_t type); 240 static void inline cc_conn_init(struct tcpcb *tp); 241 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 242 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 243 struct tcphdr *th, struct tcpopt *to); 244 245 /* 246 * TCP statistics are stored in an "array" of counter(9)s. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 VNET_PCPUSTAT_SYSINIT(tcpstat); 250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252 253 #ifdef VIMAGE 254 VNET_PCPUSTAT_SYSUNINIT(tcpstat); 255 #endif /* VIMAGE */ 256 /* 257 * Kernel module interface for updating tcpstat. The argument is an index 258 * into tcpstat treated as an array. 259 */ 260 void 261 kmod_tcpstat_inc(int statnum) 262 { 263 264 counter_u64_add(VNET(tcpstat)[statnum], 1); 265 } 266 267 /* 268 * Wrapper for the TCP established input helper hook. 269 */ 270 static void inline 271 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 272 { 273 struct tcp_hhook_data hhook_data; 274 275 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 276 hhook_data.tp = tp; 277 hhook_data.th = th; 278 hhook_data.to = to; 279 280 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 281 tp->osd); 282 } 283 } 284 285 /* 286 * CC wrapper hook functions 287 */ 288 static void inline 289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 290 { 291 INP_WLOCK_ASSERT(tp->t_inpcb); 292 293 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 294 if (tp->snd_cwnd <= tp->snd_wnd) 295 tp->ccv->flags |= CCF_CWND_LIMITED; 296 else 297 tp->ccv->flags &= ~CCF_CWND_LIMITED; 298 299 if (type == CC_ACK) { 300 if (tp->snd_cwnd > tp->snd_ssthresh) { 301 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 302 V_tcp_abc_l_var * tp->t_maxseg); 303 if (tp->t_bytes_acked >= tp->snd_cwnd) { 304 tp->t_bytes_acked -= tp->snd_cwnd; 305 tp->ccv->flags |= CCF_ABC_SENTAWND; 306 } 307 } else { 308 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 309 tp->t_bytes_acked = 0; 310 } 311 } 312 313 if (CC_ALGO(tp)->ack_received != NULL) { 314 /* XXXLAS: Find a way to live without this */ 315 tp->ccv->curack = th->th_ack; 316 CC_ALGO(tp)->ack_received(tp->ccv, type); 317 } 318 } 319 320 static void inline 321 cc_conn_init(struct tcpcb *tp) 322 { 323 struct hc_metrics_lite metrics; 324 struct inpcb *inp = tp->t_inpcb; 325 int rtt; 326 327 INP_WLOCK_ASSERT(tp->t_inpcb); 328 329 tcp_hc_get(&inp->inp_inc, &metrics); 330 331 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 332 tp->t_srtt = rtt; 333 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 334 TCPSTAT_INC(tcps_usedrtt); 335 if (metrics.rmx_rttvar) { 336 tp->t_rttvar = metrics.rmx_rttvar; 337 TCPSTAT_INC(tcps_usedrttvar); 338 } else { 339 /* default variation is +- 1 rtt */ 340 tp->t_rttvar = 341 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 342 } 343 TCPT_RANGESET(tp->t_rxtcur, 344 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 345 tp->t_rttmin, TCPTV_REXMTMAX); 346 } 347 if (metrics.rmx_ssthresh) { 348 /* 349 * There's some sort of gateway or interface 350 * buffer limit on the path. Use this to set 351 * the slow start threshhold, but set the 352 * threshold to no less than 2*mss. 353 */ 354 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 355 TCPSTAT_INC(tcps_usedssthresh); 356 } 357 358 /* 359 * Set the initial slow-start flight size. 360 * 361 * RFC5681 Section 3.1 specifies the default conservative values. 362 * RFC3390 specifies slightly more aggressive values. 363 * RFC6928 increases it to ten segments. 364 * 365 * If a SYN or SYN/ACK was lost and retransmitted, we have to 366 * reduce the initial CWND to one segment as congestion is likely 367 * requiring us to be cautious. 368 */ 369 if (tp->snd_cwnd == 1) 370 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 371 else if (V_tcp_do_initcwnd10) 372 tp->snd_cwnd = min(10 * tp->t_maxseg, 373 max(2 * tp->t_maxseg, 14600)); 374 else if (V_tcp_do_rfc3390) 375 tp->snd_cwnd = min(4 * tp->t_maxseg, 376 max(2 * tp->t_maxseg, 4380)); 377 else { 378 /* Per RFC5681 Section 3.1 */ 379 if (tp->t_maxseg > 2190) 380 tp->snd_cwnd = 2 * tp->t_maxseg; 381 else if (tp->t_maxseg > 1095) 382 tp->snd_cwnd = 3 * tp->t_maxseg; 383 else 384 tp->snd_cwnd = 4 * tp->t_maxseg; 385 } 386 387 if (CC_ALGO(tp)->conn_init != NULL) 388 CC_ALGO(tp)->conn_init(tp->ccv); 389 } 390 391 void inline 392 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 393 { 394 INP_WLOCK_ASSERT(tp->t_inpcb); 395 396 switch(type) { 397 case CC_NDUPACK: 398 if (!IN_FASTRECOVERY(tp->t_flags)) { 399 tp->snd_recover = tp->snd_max; 400 if (tp->t_flags & TF_ECN_PERMIT) 401 tp->t_flags |= TF_ECN_SND_CWR; 402 } 403 break; 404 case CC_ECN: 405 if (!IN_CONGRECOVERY(tp->t_flags)) { 406 TCPSTAT_INC(tcps_ecn_rcwnd); 407 tp->snd_recover = tp->snd_max; 408 if (tp->t_flags & TF_ECN_PERMIT) 409 tp->t_flags |= TF_ECN_SND_CWR; 410 } 411 break; 412 case CC_RTO: 413 tp->t_dupacks = 0; 414 tp->t_bytes_acked = 0; 415 EXIT_RECOVERY(tp->t_flags); 416 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 417 tp->t_maxseg) * tp->t_maxseg; 418 tp->snd_cwnd = tp->t_maxseg; 419 break; 420 case CC_RTO_ERR: 421 TCPSTAT_INC(tcps_sndrexmitbad); 422 /* RTO was unnecessary, so reset everything. */ 423 tp->snd_cwnd = tp->snd_cwnd_prev; 424 tp->snd_ssthresh = tp->snd_ssthresh_prev; 425 tp->snd_recover = tp->snd_recover_prev; 426 if (tp->t_flags & TF_WASFRECOVERY) 427 ENTER_FASTRECOVERY(tp->t_flags); 428 if (tp->t_flags & TF_WASCRECOVERY) 429 ENTER_CONGRECOVERY(tp->t_flags); 430 tp->snd_nxt = tp->snd_max; 431 tp->t_flags &= ~TF_PREVVALID; 432 tp->t_badrxtwin = 0; 433 break; 434 } 435 436 if (CC_ALGO(tp)->cong_signal != NULL) { 437 if (th != NULL) 438 tp->ccv->curack = th->th_ack; 439 CC_ALGO(tp)->cong_signal(tp->ccv, type); 440 } 441 } 442 443 static void inline 444 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 445 { 446 INP_WLOCK_ASSERT(tp->t_inpcb); 447 448 /* XXXLAS: KASSERT that we're in recovery? */ 449 450 if (CC_ALGO(tp)->post_recovery != NULL) { 451 tp->ccv->curack = th->th_ack; 452 CC_ALGO(tp)->post_recovery(tp->ccv); 453 } 454 /* XXXLAS: EXIT_RECOVERY ? */ 455 tp->t_bytes_acked = 0; 456 } 457 458 #ifdef TCP_SIGNATURE 459 static inline int 460 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 461 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 462 { 463 int ret; 464 465 tcp_fields_to_net(th); 466 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 467 tcp_fields_to_host(th); 468 return (ret); 469 } 470 #endif 471 472 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 473 #ifdef INET6 474 #define ND6_HINT(tp) \ 475 do { \ 476 if ((tp) && (tp)->t_inpcb && \ 477 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 478 nd6_nud_hint(NULL, NULL, 0); \ 479 } while (0) 480 #else 481 #define ND6_HINT(tp) 482 #endif 483 484 /* 485 * Indicate whether this ack should be delayed. We can delay the ack if 486 * following conditions are met: 487 * - There is no delayed ack timer in progress. 488 * - Our last ack wasn't a 0-sized window. We never want to delay 489 * the ack that opens up a 0-sized window. 490 * - LRO wasn't used for this segment. We make sure by checking that the 491 * segment size is not larger than the MSS. 492 * - Delayed acks are enabled or this is a half-synchronized T/TCP 493 * connection. 494 */ 495 #define DELAY_ACK(tp, tlen) \ 496 ((!tcp_timer_active(tp, TT_DELACK) && \ 497 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 498 (tlen <= tp->t_maxopd) && \ 499 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 500 501 /* 502 * TCP input handling is split into multiple parts: 503 * tcp6_input is a thin wrapper around tcp_input for the extended 504 * ip6_protox[] call format in ip6_input 505 * tcp_input handles primary segment validation, inpcb lookup and 506 * SYN processing on listen sockets 507 * tcp_do_segment processes the ACK and text of the segment for 508 * establishing, established and closing connections 509 */ 510 #ifdef INET6 511 int 512 tcp6_input(struct mbuf **mp, int *offp, int proto) 513 { 514 struct mbuf *m = *mp; 515 struct in6_ifaddr *ia6; 516 struct ip6_hdr *ip6; 517 518 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 519 520 /* 521 * draft-itojun-ipv6-tcp-to-anycast 522 * better place to put this in? 523 */ 524 ip6 = mtod(m, struct ip6_hdr *); 525 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 526 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 527 struct ip6_hdr *ip6; 528 529 ifa_free(&ia6->ia_ifa); 530 ip6 = mtod(m, struct ip6_hdr *); 531 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 532 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 533 return (IPPROTO_DONE); 534 } 535 if (ia6) 536 ifa_free(&ia6->ia_ifa); 537 538 return (tcp_input(mp, offp, proto)); 539 } 540 #endif /* INET6 */ 541 542 int 543 tcp_input(struct mbuf **mp, int *offp, int proto) 544 { 545 struct mbuf *m = *mp; 546 struct tcphdr *th = NULL; 547 struct ip *ip = NULL; 548 struct inpcb *inp = NULL; 549 struct tcpcb *tp = NULL; 550 struct socket *so = NULL; 551 u_char *optp = NULL; 552 int off0; 553 int optlen = 0; 554 #ifdef INET 555 int len; 556 #endif 557 int tlen = 0, off; 558 int drop_hdrlen; 559 int thflags; 560 int rstreason = 0; /* For badport_bandlim accounting purposes */ 561 #ifdef TCP_SIGNATURE 562 uint8_t sig_checked = 0; 563 #endif 564 uint8_t iptos = 0; 565 struct m_tag *fwd_tag = NULL; 566 #ifdef INET6 567 struct ip6_hdr *ip6 = NULL; 568 int isipv6; 569 #else 570 const void *ip6 = NULL; 571 #endif /* INET6 */ 572 struct tcpopt to; /* options in this segment */ 573 char *s = NULL; /* address and port logging */ 574 int ti_locked; 575 #define TI_UNLOCKED 1 576 #define TI_WLOCKED 2 577 578 #ifdef TCPDEBUG 579 /* 580 * The size of tcp_saveipgen must be the size of the max ip header, 581 * now IPv6. 582 */ 583 u_char tcp_saveipgen[IP6_HDR_LEN]; 584 struct tcphdr tcp_savetcp; 585 short ostate = 0; 586 #endif 587 588 #ifdef INET6 589 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 590 #endif 591 592 off0 = *offp; 593 m = *mp; 594 *mp = NULL; 595 to.to_flags = 0; 596 TCPSTAT_INC(tcps_rcvtotal); 597 598 #ifdef INET6 599 if (isipv6) { 600 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 601 602 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 603 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 604 if (m == NULL) { 605 TCPSTAT_INC(tcps_rcvshort); 606 return (IPPROTO_DONE); 607 } 608 } 609 610 ip6 = mtod(m, struct ip6_hdr *); 611 th = (struct tcphdr *)((caddr_t)ip6 + off0); 612 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 613 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 614 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 615 th->th_sum = m->m_pkthdr.csum_data; 616 else 617 th->th_sum = in6_cksum_pseudo(ip6, tlen, 618 IPPROTO_TCP, m->m_pkthdr.csum_data); 619 th->th_sum ^= 0xffff; 620 } else 621 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 622 if (th->th_sum) { 623 TCPSTAT_INC(tcps_rcvbadsum); 624 goto drop; 625 } 626 627 /* 628 * Be proactive about unspecified IPv6 address in source. 629 * As we use all-zero to indicate unbounded/unconnected pcb, 630 * unspecified IPv6 address can be used to confuse us. 631 * 632 * Note that packets with unspecified IPv6 destination is 633 * already dropped in ip6_input. 634 */ 635 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 636 /* XXX stat */ 637 goto drop; 638 } 639 } 640 #endif 641 #if defined(INET) && defined(INET6) 642 else 643 #endif 644 #ifdef INET 645 { 646 /* 647 * Get IP and TCP header together in first mbuf. 648 * Note: IP leaves IP header in first mbuf. 649 */ 650 if (off0 > sizeof (struct ip)) { 651 ip_stripoptions(m); 652 off0 = sizeof(struct ip); 653 } 654 if (m->m_len < sizeof (struct tcpiphdr)) { 655 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 656 == NULL) { 657 TCPSTAT_INC(tcps_rcvshort); 658 return (IPPROTO_DONE); 659 } 660 } 661 ip = mtod(m, struct ip *); 662 th = (struct tcphdr *)((caddr_t)ip + off0); 663 tlen = ntohs(ip->ip_len) - off0; 664 665 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 666 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 667 th->th_sum = m->m_pkthdr.csum_data; 668 else 669 th->th_sum = in_pseudo(ip->ip_src.s_addr, 670 ip->ip_dst.s_addr, 671 htonl(m->m_pkthdr.csum_data + tlen + 672 IPPROTO_TCP)); 673 th->th_sum ^= 0xffff; 674 } else { 675 struct ipovly *ipov = (struct ipovly *)ip; 676 677 /* 678 * Checksum extended TCP header and data. 679 */ 680 len = off0 + tlen; 681 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 682 ipov->ih_len = htons(tlen); 683 th->th_sum = in_cksum(m, len); 684 /* Reset length for SDT probes. */ 685 ip->ip_len = htons(tlen + off0); 686 } 687 688 if (th->th_sum) { 689 TCPSTAT_INC(tcps_rcvbadsum); 690 goto drop; 691 } 692 /* Re-initialization for later version check */ 693 ip->ip_v = IPVERSION; 694 } 695 #endif /* INET */ 696 697 #ifdef INET6 698 if (isipv6) 699 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 700 #endif 701 #if defined(INET) && defined(INET6) 702 else 703 #endif 704 #ifdef INET 705 iptos = ip->ip_tos; 706 #endif 707 708 /* 709 * Check that TCP offset makes sense, 710 * pull out TCP options and adjust length. XXX 711 */ 712 off = th->th_off << 2; 713 if (off < sizeof (struct tcphdr) || off > tlen) { 714 TCPSTAT_INC(tcps_rcvbadoff); 715 goto drop; 716 } 717 tlen -= off; /* tlen is used instead of ti->ti_len */ 718 if (off > sizeof (struct tcphdr)) { 719 #ifdef INET6 720 if (isipv6) { 721 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 722 ip6 = mtod(m, struct ip6_hdr *); 723 th = (struct tcphdr *)((caddr_t)ip6 + off0); 724 } 725 #endif 726 #if defined(INET) && defined(INET6) 727 else 728 #endif 729 #ifdef INET 730 { 731 if (m->m_len < sizeof(struct ip) + off) { 732 if ((m = m_pullup(m, sizeof (struct ip) + off)) 733 == NULL) { 734 TCPSTAT_INC(tcps_rcvshort); 735 return (IPPROTO_DONE); 736 } 737 ip = mtod(m, struct ip *); 738 th = (struct tcphdr *)((caddr_t)ip + off0); 739 } 740 } 741 #endif 742 optlen = off - sizeof (struct tcphdr); 743 optp = (u_char *)(th + 1); 744 } 745 thflags = th->th_flags; 746 747 /* 748 * Convert TCP protocol specific fields to host format. 749 */ 750 tcp_fields_to_host(th); 751 752 /* 753 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 754 */ 755 drop_hdrlen = off0 + off; 756 757 /* 758 * Locate pcb for segment; if we're likely to add or remove a 759 * connection then first acquire pcbinfo lock. There are three cases 760 * where we might discover later we need a write lock despite the 761 * flags: ACKs moving a connection out of the syncache, ACKs for a 762 * connection in TIMEWAIT and SYNs not targeting a listening socket. 763 */ 764 if ((thflags & (TH_FIN | TH_RST)) != 0) { 765 INP_INFO_WLOCK(&V_tcbinfo); 766 ti_locked = TI_WLOCKED; 767 } else 768 ti_locked = TI_UNLOCKED; 769 770 /* 771 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 772 */ 773 if ( 774 #ifdef INET6 775 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 776 #ifdef INET 777 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 778 #endif 779 #endif 780 #if defined(INET) && !defined(INET6) 781 (m->m_flags & M_IP_NEXTHOP) 782 #endif 783 ) 784 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 785 786 findpcb: 787 #ifdef INVARIANTS 788 if (ti_locked == TI_WLOCKED) { 789 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 790 } else { 791 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 792 } 793 #endif 794 #ifdef INET6 795 if (isipv6 && fwd_tag != NULL) { 796 struct sockaddr_in6 *next_hop6; 797 798 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 799 /* 800 * Transparently forwarded. Pretend to be the destination. 801 * Already got one like this? 802 */ 803 inp = in6_pcblookup_mbuf(&V_tcbinfo, 804 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 805 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 806 if (!inp) { 807 /* 808 * It's new. Try to find the ambushing socket. 809 * Because we've rewritten the destination address, 810 * any hardware-generated hash is ignored. 811 */ 812 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 813 th->th_sport, &next_hop6->sin6_addr, 814 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 815 th->th_dport, INPLOOKUP_WILDCARD | 816 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 817 } 818 } else if (isipv6) { 819 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 820 th->th_sport, &ip6->ip6_dst, th->th_dport, 821 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 822 m->m_pkthdr.rcvif, m); 823 } 824 #endif /* INET6 */ 825 #if defined(INET6) && defined(INET) 826 else 827 #endif 828 #ifdef INET 829 if (fwd_tag != NULL) { 830 struct sockaddr_in *next_hop; 831 832 next_hop = (struct sockaddr_in *)(fwd_tag+1); 833 /* 834 * Transparently forwarded. Pretend to be the destination. 835 * already got one like this? 836 */ 837 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 838 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 839 m->m_pkthdr.rcvif, m); 840 if (!inp) { 841 /* 842 * It's new. Try to find the ambushing socket. 843 * Because we've rewritten the destination address, 844 * any hardware-generated hash is ignored. 845 */ 846 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 847 th->th_sport, next_hop->sin_addr, 848 next_hop->sin_port ? ntohs(next_hop->sin_port) : 849 th->th_dport, INPLOOKUP_WILDCARD | 850 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 851 } 852 } else 853 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 854 th->th_sport, ip->ip_dst, th->th_dport, 855 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 856 m->m_pkthdr.rcvif, m); 857 #endif /* INET */ 858 859 /* 860 * If the INPCB does not exist then all data in the incoming 861 * segment is discarded and an appropriate RST is sent back. 862 * XXX MRT Send RST using which routing table? 863 */ 864 if (inp == NULL) { 865 /* 866 * Log communication attempts to ports that are not 867 * in use. 868 */ 869 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 870 tcp_log_in_vain == 2) { 871 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 872 log(LOG_INFO, "%s; %s: Connection attempt " 873 "to closed port\n", s, __func__); 874 } 875 /* 876 * When blackholing do not respond with a RST but 877 * completely ignore the segment and drop it. 878 */ 879 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 880 V_blackhole == 2) 881 goto dropunlock; 882 883 rstreason = BANDLIM_RST_CLOSEDPORT; 884 goto dropwithreset; 885 } 886 INP_WLOCK_ASSERT(inp); 887 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 888 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 889 ((inp->inp_socket == NULL) || 890 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 891 inp->inp_flowid = m->m_pkthdr.flowid; 892 inp->inp_flowtype = M_HASHTYPE_GET(m); 893 } 894 #ifdef IPSEC 895 #ifdef INET6 896 if (isipv6 && ipsec6_in_reject(m, inp)) { 897 goto dropunlock; 898 } else 899 #endif /* INET6 */ 900 if (ipsec4_in_reject(m, inp) != 0) { 901 goto dropunlock; 902 } 903 #endif /* IPSEC */ 904 905 /* 906 * Check the minimum TTL for socket. 907 */ 908 if (inp->inp_ip_minttl != 0) { 909 #ifdef INET6 910 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 911 goto dropunlock; 912 else 913 #endif 914 if (inp->inp_ip_minttl > ip->ip_ttl) 915 goto dropunlock; 916 } 917 918 /* 919 * A previous connection in TIMEWAIT state is supposed to catch stray 920 * or duplicate segments arriving late. If this segment was a 921 * legitimate new connection attempt, the old INPCB gets removed and 922 * we can try again to find a listening socket. 923 * 924 * At this point, due to earlier optimism, we may hold only an inpcb 925 * lock, and not the inpcbinfo write lock. If so, we need to try to 926 * acquire it, or if that fails, acquire a reference on the inpcb, 927 * drop all locks, acquire a global write lock, and then re-acquire 928 * the inpcb lock. We may at that point discover that another thread 929 * has tried to free the inpcb, in which case we need to loop back 930 * and try to find a new inpcb to deliver to. 931 * 932 * XXXRW: It may be time to rethink timewait locking. 933 */ 934 relocked: 935 if (inp->inp_flags & INP_TIMEWAIT) { 936 if (ti_locked == TI_UNLOCKED) { 937 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 938 in_pcbref(inp); 939 INP_WUNLOCK(inp); 940 INP_INFO_WLOCK(&V_tcbinfo); 941 ti_locked = TI_WLOCKED; 942 INP_WLOCK(inp); 943 if (in_pcbrele_wlocked(inp)) { 944 inp = NULL; 945 goto findpcb; 946 } 947 } else 948 ti_locked = TI_WLOCKED; 949 } 950 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 951 952 if (thflags & TH_SYN) 953 tcp_dooptions(&to, optp, optlen, TO_SYN); 954 /* 955 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 956 */ 957 if (tcp_twcheck(inp, &to, th, m, tlen)) 958 goto findpcb; 959 INP_INFO_WUNLOCK(&V_tcbinfo); 960 return (IPPROTO_DONE); 961 } 962 /* 963 * The TCPCB may no longer exist if the connection is winding 964 * down or it is in the CLOSED state. Either way we drop the 965 * segment and send an appropriate response. 966 */ 967 tp = intotcpcb(inp); 968 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 969 rstreason = BANDLIM_RST_CLOSEDPORT; 970 goto dropwithreset; 971 } 972 973 #ifdef TCP_OFFLOAD 974 if (tp->t_flags & TF_TOE) { 975 tcp_offload_input(tp, m); 976 m = NULL; /* consumed by the TOE driver */ 977 goto dropunlock; 978 } 979 #endif 980 981 /* 982 * We've identified a valid inpcb, but it could be that we need an 983 * inpcbinfo write lock but don't hold it. In this case, attempt to 984 * acquire using the same strategy as the TIMEWAIT case above. If we 985 * relock, we have to jump back to 'relocked' as the connection might 986 * now be in TIMEWAIT. 987 */ 988 #ifdef INVARIANTS 989 if ((thflags & (TH_FIN | TH_RST)) != 0) 990 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 991 #endif 992 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 993 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN)))) { 994 if (ti_locked == TI_UNLOCKED) { 995 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 996 in_pcbref(inp); 997 INP_WUNLOCK(inp); 998 INP_INFO_WLOCK(&V_tcbinfo); 999 ti_locked = TI_WLOCKED; 1000 INP_WLOCK(inp); 1001 if (in_pcbrele_wlocked(inp)) { 1002 inp = NULL; 1003 goto findpcb; 1004 } 1005 goto relocked; 1006 } else 1007 ti_locked = TI_WLOCKED; 1008 } 1009 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1010 } 1011 1012 #ifdef MAC 1013 INP_WLOCK_ASSERT(inp); 1014 if (mac_inpcb_check_deliver(inp, m)) 1015 goto dropunlock; 1016 #endif 1017 so = inp->inp_socket; 1018 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1019 #ifdef TCPDEBUG 1020 if (so->so_options & SO_DEBUG) { 1021 ostate = tp->t_state; 1022 #ifdef INET6 1023 if (isipv6) { 1024 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1025 } else 1026 #endif 1027 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1028 tcp_savetcp = *th; 1029 } 1030 #endif /* TCPDEBUG */ 1031 /* 1032 * When the socket is accepting connections (the INPCB is in LISTEN 1033 * state) we look into the SYN cache if this is a new connection 1034 * attempt or the completion of a previous one. 1035 */ 1036 if (so->so_options & SO_ACCEPTCONN) { 1037 struct in_conninfo inc; 1038 1039 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1040 "tp not listening", __func__)); 1041 bzero(&inc, sizeof(inc)); 1042 #ifdef INET6 1043 if (isipv6) { 1044 inc.inc_flags |= INC_ISIPV6; 1045 inc.inc6_faddr = ip6->ip6_src; 1046 inc.inc6_laddr = ip6->ip6_dst; 1047 } else 1048 #endif 1049 { 1050 inc.inc_faddr = ip->ip_src; 1051 inc.inc_laddr = ip->ip_dst; 1052 } 1053 inc.inc_fport = th->th_sport; 1054 inc.inc_lport = th->th_dport; 1055 inc.inc_fibnum = so->so_fibnum; 1056 1057 /* 1058 * Check for an existing connection attempt in syncache if 1059 * the flag is only ACK. A successful lookup creates a new 1060 * socket appended to the listen queue in SYN_RECEIVED state. 1061 */ 1062 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1063 1064 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1065 /* 1066 * Parse the TCP options here because 1067 * syncookies need access to the reflected 1068 * timestamp. 1069 */ 1070 tcp_dooptions(&to, optp, optlen, 0); 1071 /* 1072 * NB: syncache_expand() doesn't unlock 1073 * inp and tcpinfo locks. 1074 */ 1075 if (!syncache_expand(&inc, &to, th, &so, m)) { 1076 /* 1077 * No syncache entry or ACK was not 1078 * for our SYN/ACK. Send a RST. 1079 * NB: syncache did its own logging 1080 * of the failure cause. 1081 */ 1082 rstreason = BANDLIM_RST_OPENPORT; 1083 goto dropwithreset; 1084 } 1085 if (so == NULL) { 1086 /* 1087 * We completed the 3-way handshake 1088 * but could not allocate a socket 1089 * either due to memory shortage, 1090 * listen queue length limits or 1091 * global socket limits. Send RST 1092 * or wait and have the remote end 1093 * retransmit the ACK for another 1094 * try. 1095 */ 1096 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1097 log(LOG_DEBUG, "%s; %s: Listen socket: " 1098 "Socket allocation failed due to " 1099 "limits or memory shortage, %s\n", 1100 s, __func__, 1101 V_tcp_sc_rst_sock_fail ? 1102 "sending RST" : "try again"); 1103 if (V_tcp_sc_rst_sock_fail) { 1104 rstreason = BANDLIM_UNLIMITED; 1105 goto dropwithreset; 1106 } else 1107 goto dropunlock; 1108 } 1109 /* 1110 * Socket is created in state SYN_RECEIVED. 1111 * Unlock the listen socket, lock the newly 1112 * created socket and update the tp variable. 1113 */ 1114 INP_WUNLOCK(inp); /* listen socket */ 1115 inp = sotoinpcb(so); 1116 INP_WLOCK(inp); /* new connection */ 1117 tp = intotcpcb(inp); 1118 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1119 ("%s: ", __func__)); 1120 #ifdef TCP_SIGNATURE 1121 if (sig_checked == 0) { 1122 tcp_dooptions(&to, optp, optlen, 1123 (thflags & TH_SYN) ? TO_SYN : 0); 1124 if (!tcp_signature_verify_input(m, off0, tlen, 1125 optlen, &to, th, tp->t_flags)) { 1126 1127 /* 1128 * In SYN_SENT state if it receives an 1129 * RST, it is allowed for further 1130 * processing. 1131 */ 1132 if ((thflags & TH_RST) == 0 || 1133 (tp->t_state == TCPS_SYN_SENT) == 0) 1134 goto dropunlock; 1135 } 1136 sig_checked = 1; 1137 } 1138 #endif 1139 1140 /* 1141 * Process the segment and the data it 1142 * contains. tcp_do_segment() consumes 1143 * the mbuf chain and unlocks the inpcb. 1144 */ 1145 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1146 iptos, ti_locked); 1147 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1148 return (IPPROTO_DONE); 1149 } 1150 /* 1151 * Segment flag validation for new connection attempts: 1152 * 1153 * Our (SYN|ACK) response was rejected. 1154 * Check with syncache and remove entry to prevent 1155 * retransmits. 1156 * 1157 * NB: syncache_chkrst does its own logging of failure 1158 * causes. 1159 */ 1160 if (thflags & TH_RST) { 1161 syncache_chkrst(&inc, th); 1162 goto dropunlock; 1163 } 1164 /* 1165 * We can't do anything without SYN. 1166 */ 1167 if ((thflags & TH_SYN) == 0) { 1168 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1169 log(LOG_DEBUG, "%s; %s: Listen socket: " 1170 "SYN is missing, segment ignored\n", 1171 s, __func__); 1172 TCPSTAT_INC(tcps_badsyn); 1173 goto dropunlock; 1174 } 1175 /* 1176 * (SYN|ACK) is bogus on a listen socket. 1177 */ 1178 if (thflags & TH_ACK) { 1179 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1180 log(LOG_DEBUG, "%s; %s: Listen socket: " 1181 "SYN|ACK invalid, segment rejected\n", 1182 s, __func__); 1183 syncache_badack(&inc); /* XXX: Not needed! */ 1184 TCPSTAT_INC(tcps_badsyn); 1185 rstreason = BANDLIM_RST_OPENPORT; 1186 goto dropwithreset; 1187 } 1188 /* 1189 * If the drop_synfin option is enabled, drop all 1190 * segments with both the SYN and FIN bits set. 1191 * This prevents e.g. nmap from identifying the 1192 * TCP/IP stack. 1193 * XXX: Poor reasoning. nmap has other methods 1194 * and is constantly refining its stack detection 1195 * strategies. 1196 * XXX: This is a violation of the TCP specification 1197 * and was used by RFC1644. 1198 */ 1199 if ((thflags & TH_FIN) && V_drop_synfin) { 1200 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1201 log(LOG_DEBUG, "%s; %s: Listen socket: " 1202 "SYN|FIN segment ignored (based on " 1203 "sysctl setting)\n", s, __func__); 1204 TCPSTAT_INC(tcps_badsyn); 1205 goto dropunlock; 1206 } 1207 /* 1208 * Segment's flags are (SYN) or (SYN|FIN). 1209 * 1210 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1211 * as they do not affect the state of the TCP FSM. 1212 * The data pointed to by TH_URG and th_urp is ignored. 1213 */ 1214 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1215 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1216 KASSERT(thflags & (TH_SYN), 1217 ("%s: Listen socket: TH_SYN not set", __func__)); 1218 #ifdef INET6 1219 /* 1220 * If deprecated address is forbidden, 1221 * we do not accept SYN to deprecated interface 1222 * address to prevent any new inbound connection from 1223 * getting established. 1224 * When we do not accept SYN, we send a TCP RST, 1225 * with deprecated source address (instead of dropping 1226 * it). We compromise it as it is much better for peer 1227 * to send a RST, and RST will be the final packet 1228 * for the exchange. 1229 * 1230 * If we do not forbid deprecated addresses, we accept 1231 * the SYN packet. RFC2462 does not suggest dropping 1232 * SYN in this case. 1233 * If we decipher RFC2462 5.5.4, it says like this: 1234 * 1. use of deprecated addr with existing 1235 * communication is okay - "SHOULD continue to be 1236 * used" 1237 * 2. use of it with new communication: 1238 * (2a) "SHOULD NOT be used if alternate address 1239 * with sufficient scope is available" 1240 * (2b) nothing mentioned otherwise. 1241 * Here we fall into (2b) case as we have no choice in 1242 * our source address selection - we must obey the peer. 1243 * 1244 * The wording in RFC2462 is confusing, and there are 1245 * multiple description text for deprecated address 1246 * handling - worse, they are not exactly the same. 1247 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1248 */ 1249 if (isipv6 && !V_ip6_use_deprecated) { 1250 struct in6_ifaddr *ia6; 1251 1252 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1253 if (ia6 != NULL && 1254 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1255 ifa_free(&ia6->ia_ifa); 1256 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1257 log(LOG_DEBUG, "%s; %s: Listen socket: " 1258 "Connection attempt to deprecated " 1259 "IPv6 address rejected\n", 1260 s, __func__); 1261 rstreason = BANDLIM_RST_OPENPORT; 1262 goto dropwithreset; 1263 } 1264 if (ia6) 1265 ifa_free(&ia6->ia_ifa); 1266 } 1267 #endif /* INET6 */ 1268 /* 1269 * Basic sanity checks on incoming SYN requests: 1270 * Don't respond if the destination is a link layer 1271 * broadcast according to RFC1122 4.2.3.10, p. 104. 1272 * If it is from this socket it must be forged. 1273 * Don't respond if the source or destination is a 1274 * global or subnet broad- or multicast address. 1275 * Note that it is quite possible to receive unicast 1276 * link-layer packets with a broadcast IP address. Use 1277 * in_broadcast() to find them. 1278 */ 1279 if (m->m_flags & (M_BCAST|M_MCAST)) { 1280 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1281 log(LOG_DEBUG, "%s; %s: Listen socket: " 1282 "Connection attempt from broad- or multicast " 1283 "link layer address ignored\n", s, __func__); 1284 goto dropunlock; 1285 } 1286 #ifdef INET6 1287 if (isipv6) { 1288 if (th->th_dport == th->th_sport && 1289 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1290 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1291 log(LOG_DEBUG, "%s; %s: Listen socket: " 1292 "Connection attempt to/from self " 1293 "ignored\n", s, __func__); 1294 goto dropunlock; 1295 } 1296 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1297 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1299 log(LOG_DEBUG, "%s; %s: Listen socket: " 1300 "Connection attempt from/to multicast " 1301 "address ignored\n", s, __func__); 1302 goto dropunlock; 1303 } 1304 } 1305 #endif 1306 #if defined(INET) && defined(INET6) 1307 else 1308 #endif 1309 #ifdef INET 1310 { 1311 if (th->th_dport == th->th_sport && 1312 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1313 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1314 log(LOG_DEBUG, "%s; %s: Listen socket: " 1315 "Connection attempt from/to self " 1316 "ignored\n", s, __func__); 1317 goto dropunlock; 1318 } 1319 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1320 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1321 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1322 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1323 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1324 log(LOG_DEBUG, "%s; %s: Listen socket: " 1325 "Connection attempt from/to broad- " 1326 "or multicast address ignored\n", 1327 s, __func__); 1328 goto dropunlock; 1329 } 1330 } 1331 #endif 1332 /* 1333 * SYN appears to be valid. Create compressed TCP state 1334 * for syncache. 1335 */ 1336 #ifdef TCPDEBUG 1337 if (so->so_options & SO_DEBUG) 1338 tcp_trace(TA_INPUT, ostate, tp, 1339 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1340 #endif 1341 tcp_dooptions(&to, optp, optlen, TO_SYN); 1342 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1343 /* 1344 * Entry added to syncache and mbuf consumed. 1345 * Only the listen socket is unlocked by syncache_add(). 1346 */ 1347 if (ti_locked == TI_WLOCKED) { 1348 INP_INFO_WUNLOCK(&V_tcbinfo); 1349 ti_locked = TI_UNLOCKED; 1350 } 1351 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1352 return (IPPROTO_DONE); 1353 } else if (tp->t_state == TCPS_LISTEN) { 1354 /* 1355 * When a listen socket is torn down the SO_ACCEPTCONN 1356 * flag is removed first while connections are drained 1357 * from the accept queue in a unlock/lock cycle of the 1358 * ACCEPT_LOCK, opening a race condition allowing a SYN 1359 * attempt go through unhandled. 1360 */ 1361 goto dropunlock; 1362 } 1363 1364 #ifdef TCP_SIGNATURE 1365 if (sig_checked == 0) { 1366 tcp_dooptions(&to, optp, optlen, 1367 (thflags & TH_SYN) ? TO_SYN : 0); 1368 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1369 th, tp->t_flags)) { 1370 1371 /* 1372 * In SYN_SENT state if it receives an RST, it is 1373 * allowed for further processing. 1374 */ 1375 if ((thflags & TH_RST) == 0 || 1376 (tp->t_state == TCPS_SYN_SENT) == 0) 1377 goto dropunlock; 1378 } 1379 sig_checked = 1; 1380 } 1381 #endif 1382 1383 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1384 1385 /* 1386 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1387 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1388 * the inpcb, and unlocks pcbinfo. 1389 */ 1390 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1391 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1392 return (IPPROTO_DONE); 1393 1394 dropwithreset: 1395 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1396 1397 if (ti_locked == TI_WLOCKED) { 1398 INP_INFO_WUNLOCK(&V_tcbinfo); 1399 ti_locked = TI_UNLOCKED; 1400 } 1401 #ifdef INVARIANTS 1402 else { 1403 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1404 "ti_locked: %d", __func__, ti_locked)); 1405 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1406 } 1407 #endif 1408 1409 if (inp != NULL) { 1410 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1411 INP_WUNLOCK(inp); 1412 } else 1413 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1414 m = NULL; /* mbuf chain got consumed. */ 1415 goto drop; 1416 1417 dropunlock: 1418 if (m != NULL) 1419 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1420 1421 if (ti_locked == TI_WLOCKED) { 1422 INP_INFO_WUNLOCK(&V_tcbinfo); 1423 ti_locked = TI_UNLOCKED; 1424 } 1425 #ifdef INVARIANTS 1426 else { 1427 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1428 "ti_locked: %d", __func__, ti_locked)); 1429 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1430 } 1431 #endif 1432 1433 if (inp != NULL) 1434 INP_WUNLOCK(inp); 1435 1436 drop: 1437 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1438 if (s != NULL) 1439 free(s, M_TCPLOG); 1440 if (m != NULL) 1441 m_freem(m); 1442 return (IPPROTO_DONE); 1443 } 1444 1445 static void 1446 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1447 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1448 int ti_locked) 1449 { 1450 int thflags, acked, ourfinisacked, needoutput = 0; 1451 int rstreason, todrop, win; 1452 u_long tiwin; 1453 char *s; 1454 struct in_conninfo *inc; 1455 struct mbuf *mfree; 1456 struct tcpopt to; 1457 1458 #ifdef TCPDEBUG 1459 /* 1460 * The size of tcp_saveipgen must be the size of the max ip header, 1461 * now IPv6. 1462 */ 1463 u_char tcp_saveipgen[IP6_HDR_LEN]; 1464 struct tcphdr tcp_savetcp; 1465 short ostate = 0; 1466 #endif 1467 thflags = th->th_flags; 1468 inc = &tp->t_inpcb->inp_inc; 1469 tp->sackhint.last_sack_ack = 0; 1470 1471 /* 1472 * If this is either a state-changing packet or current state isn't 1473 * established, we require a write lock on tcbinfo. Otherwise, we 1474 * allow the tcbinfo to be in either alocked or unlocked, as the 1475 * caller may have unnecessarily acquired a write lock due to a race. 1476 */ 1477 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1478 tp->t_state != TCPS_ESTABLISHED) { 1479 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1480 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1481 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1482 } else { 1483 #ifdef INVARIANTS 1484 if (ti_locked == TI_WLOCKED) 1485 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1486 else { 1487 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1488 "ti_locked: %d", __func__, ti_locked)); 1489 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1490 } 1491 #endif 1492 } 1493 INP_WLOCK_ASSERT(tp->t_inpcb); 1494 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1495 __func__)); 1496 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1497 __func__)); 1498 1499 /* 1500 * Segment received on connection. 1501 * Reset idle time and keep-alive timer. 1502 * XXX: This should be done after segment 1503 * validation to ignore broken/spoofed segs. 1504 */ 1505 tp->t_rcvtime = ticks; 1506 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1507 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1508 1509 /* 1510 * Unscale the window into a 32-bit value. 1511 * For the SYN_SENT state the scale is zero. 1512 */ 1513 tiwin = th->th_win << tp->snd_scale; 1514 1515 /* 1516 * TCP ECN processing. 1517 */ 1518 if (tp->t_flags & TF_ECN_PERMIT) { 1519 if (thflags & TH_CWR) 1520 tp->t_flags &= ~TF_ECN_SND_ECE; 1521 switch (iptos & IPTOS_ECN_MASK) { 1522 case IPTOS_ECN_CE: 1523 tp->t_flags |= TF_ECN_SND_ECE; 1524 TCPSTAT_INC(tcps_ecn_ce); 1525 break; 1526 case IPTOS_ECN_ECT0: 1527 TCPSTAT_INC(tcps_ecn_ect0); 1528 break; 1529 case IPTOS_ECN_ECT1: 1530 TCPSTAT_INC(tcps_ecn_ect1); 1531 break; 1532 } 1533 /* Congestion experienced. */ 1534 if (thflags & TH_ECE) { 1535 cc_cong_signal(tp, th, CC_ECN); 1536 } 1537 } 1538 1539 /* 1540 * Parse options on any incoming segment. 1541 */ 1542 tcp_dooptions(&to, (u_char *)(th + 1), 1543 (th->th_off << 2) - sizeof(struct tcphdr), 1544 (thflags & TH_SYN) ? TO_SYN : 0); 1545 1546 /* 1547 * If echoed timestamp is later than the current time, 1548 * fall back to non RFC1323 RTT calculation. Normalize 1549 * timestamp if syncookies were used when this connection 1550 * was established. 1551 */ 1552 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1553 to.to_tsecr -= tp->ts_offset; 1554 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1555 to.to_tsecr = 0; 1556 } 1557 /* 1558 * If timestamps were negotiated during SYN/ACK they should 1559 * appear on every segment during this session and vice versa. 1560 */ 1561 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1562 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1563 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1564 "no action\n", s, __func__); 1565 free(s, M_TCPLOG); 1566 } 1567 } 1568 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1569 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1570 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1571 "no action\n", s, __func__); 1572 free(s, M_TCPLOG); 1573 } 1574 } 1575 1576 /* 1577 * Process options only when we get SYN/ACK back. The SYN case 1578 * for incoming connections is handled in tcp_syncache. 1579 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1580 * or <SYN,ACK>) segment itself is never scaled. 1581 * XXX this is traditional behavior, may need to be cleaned up. 1582 */ 1583 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1584 if ((to.to_flags & TOF_SCALE) && 1585 (tp->t_flags & TF_REQ_SCALE)) { 1586 tp->t_flags |= TF_RCVD_SCALE; 1587 tp->snd_scale = to.to_wscale; 1588 } 1589 /* 1590 * Initial send window. It will be updated with 1591 * the next incoming segment to the scaled value. 1592 */ 1593 tp->snd_wnd = th->th_win; 1594 if (to.to_flags & TOF_TS) { 1595 tp->t_flags |= TF_RCVD_TSTMP; 1596 tp->ts_recent = to.to_tsval; 1597 tp->ts_recent_age = tcp_ts_getticks(); 1598 } 1599 if (to.to_flags & TOF_MSS) 1600 tcp_mss(tp, to.to_mss); 1601 if ((tp->t_flags & TF_SACK_PERMIT) && 1602 (to.to_flags & TOF_SACKPERM) == 0) 1603 tp->t_flags &= ~TF_SACK_PERMIT; 1604 } 1605 1606 /* 1607 * Header prediction: check for the two common cases 1608 * of a uni-directional data xfer. If the packet has 1609 * no control flags, is in-sequence, the window didn't 1610 * change and we're not retransmitting, it's a 1611 * candidate. If the length is zero and the ack moved 1612 * forward, we're the sender side of the xfer. Just 1613 * free the data acked & wake any higher level process 1614 * that was blocked waiting for space. If the length 1615 * is non-zero and the ack didn't move, we're the 1616 * receiver side. If we're getting packets in-order 1617 * (the reassembly queue is empty), add the data to 1618 * the socket buffer and note that we need a delayed ack. 1619 * Make sure that the hidden state-flags are also off. 1620 * Since we check for TCPS_ESTABLISHED first, it can only 1621 * be TH_NEEDSYN. 1622 */ 1623 if (tp->t_state == TCPS_ESTABLISHED && 1624 th->th_seq == tp->rcv_nxt && 1625 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1626 tp->snd_nxt == tp->snd_max && 1627 tiwin && tiwin == tp->snd_wnd && 1628 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1629 tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 || 1630 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1631 1632 /* 1633 * If last ACK falls within this segment's sequence numbers, 1634 * record the timestamp. 1635 * NOTE that the test is modified according to the latest 1636 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1637 */ 1638 if ((to.to_flags & TOF_TS) != 0 && 1639 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1640 tp->ts_recent_age = tcp_ts_getticks(); 1641 tp->ts_recent = to.to_tsval; 1642 } 1643 1644 if (tlen == 0) { 1645 if (SEQ_GT(th->th_ack, tp->snd_una) && 1646 SEQ_LEQ(th->th_ack, tp->snd_max) && 1647 !IN_RECOVERY(tp->t_flags) && 1648 (to.to_flags & TOF_SACK) == 0 && 1649 TAILQ_EMPTY(&tp->snd_holes)) { 1650 /* 1651 * This is a pure ack for outstanding data. 1652 */ 1653 if (ti_locked == TI_WLOCKED) 1654 INP_INFO_WUNLOCK(&V_tcbinfo); 1655 ti_locked = TI_UNLOCKED; 1656 1657 TCPSTAT_INC(tcps_predack); 1658 1659 /* 1660 * "bad retransmit" recovery. 1661 */ 1662 if (tp->t_rxtshift == 1 && 1663 tp->t_flags & TF_PREVVALID && 1664 (int)(ticks - tp->t_badrxtwin) < 0) { 1665 cc_cong_signal(tp, th, CC_RTO_ERR); 1666 } 1667 1668 /* 1669 * Recalculate the transmit timer / rtt. 1670 * 1671 * Some boxes send broken timestamp replies 1672 * during the SYN+ACK phase, ignore 1673 * timestamps of 0 or we could calculate a 1674 * huge RTT and blow up the retransmit timer. 1675 */ 1676 if ((to.to_flags & TOF_TS) != 0 && 1677 to.to_tsecr) { 1678 u_int t; 1679 1680 t = tcp_ts_getticks() - to.to_tsecr; 1681 if (!tp->t_rttlow || tp->t_rttlow > t) 1682 tp->t_rttlow = t; 1683 tcp_xmit_timer(tp, 1684 TCP_TS_TO_TICKS(t) + 1); 1685 } else if (tp->t_rtttime && 1686 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1687 if (!tp->t_rttlow || 1688 tp->t_rttlow > ticks - tp->t_rtttime) 1689 tp->t_rttlow = ticks - tp->t_rtttime; 1690 tcp_xmit_timer(tp, 1691 ticks - tp->t_rtttime); 1692 } 1693 acked = BYTES_THIS_ACK(tp, th); 1694 1695 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1696 hhook_run_tcp_est_in(tp, th, &to); 1697 1698 TCPSTAT_INC(tcps_rcvackpack); 1699 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1700 sbdrop(&so->so_snd, acked); 1701 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1702 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1703 tp->snd_recover = th->th_ack - 1; 1704 1705 /* 1706 * Let the congestion control algorithm update 1707 * congestion control related information. This 1708 * typically means increasing the congestion 1709 * window. 1710 */ 1711 cc_ack_received(tp, th, CC_ACK); 1712 1713 tp->snd_una = th->th_ack; 1714 /* 1715 * Pull snd_wl2 up to prevent seq wrap relative 1716 * to th_ack. 1717 */ 1718 tp->snd_wl2 = th->th_ack; 1719 tp->t_dupacks = 0; 1720 m_freem(m); 1721 ND6_HINT(tp); /* Some progress has been made. */ 1722 1723 /* 1724 * If all outstanding data are acked, stop 1725 * retransmit timer, otherwise restart timer 1726 * using current (possibly backed-off) value. 1727 * If process is waiting for space, 1728 * wakeup/selwakeup/signal. If data 1729 * are ready to send, let tcp_output 1730 * decide between more output or persist. 1731 */ 1732 #ifdef TCPDEBUG 1733 if (so->so_options & SO_DEBUG) 1734 tcp_trace(TA_INPUT, ostate, tp, 1735 (void *)tcp_saveipgen, 1736 &tcp_savetcp, 0); 1737 #endif 1738 if (tp->snd_una == tp->snd_max) 1739 tcp_timer_activate(tp, TT_REXMT, 0); 1740 else if (!tcp_timer_active(tp, TT_PERSIST)) 1741 tcp_timer_activate(tp, TT_REXMT, 1742 tp->t_rxtcur); 1743 sowwakeup(so); 1744 if (sbavail(&so->so_snd)) 1745 (void) tcp_output(tp); 1746 goto check_delack; 1747 } 1748 } else if (th->th_ack == tp->snd_una && 1749 tlen <= sbspace(&so->so_rcv)) { 1750 int newsize = 0; /* automatic sockbuf scaling */ 1751 1752 /* 1753 * This is a pure, in-sequence data packet with 1754 * nothing on the reassembly queue and we have enough 1755 * buffer space to take it. 1756 */ 1757 if (ti_locked == TI_WLOCKED) 1758 INP_INFO_WUNLOCK(&V_tcbinfo); 1759 ti_locked = TI_UNLOCKED; 1760 1761 /* Clean receiver SACK report if present */ 1762 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1763 tcp_clean_sackreport(tp); 1764 TCPSTAT_INC(tcps_preddat); 1765 tp->rcv_nxt += tlen; 1766 /* 1767 * Pull snd_wl1 up to prevent seq wrap relative to 1768 * th_seq. 1769 */ 1770 tp->snd_wl1 = th->th_seq; 1771 /* 1772 * Pull rcv_up up to prevent seq wrap relative to 1773 * rcv_nxt. 1774 */ 1775 tp->rcv_up = tp->rcv_nxt; 1776 TCPSTAT_INC(tcps_rcvpack); 1777 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1778 ND6_HINT(tp); /* Some progress has been made */ 1779 #ifdef TCPDEBUG 1780 if (so->so_options & SO_DEBUG) 1781 tcp_trace(TA_INPUT, ostate, tp, 1782 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1783 #endif 1784 /* 1785 * Automatic sizing of receive socket buffer. Often the send 1786 * buffer size is not optimally adjusted to the actual network 1787 * conditions at hand (delay bandwidth product). Setting the 1788 * buffer size too small limits throughput on links with high 1789 * bandwidth and high delay (eg. trans-continental/oceanic links). 1790 * 1791 * On the receive side the socket buffer memory is only rarely 1792 * used to any significant extent. This allows us to be much 1793 * more aggressive in scaling the receive socket buffer. For 1794 * the case that the buffer space is actually used to a large 1795 * extent and we run out of kernel memory we can simply drop 1796 * the new segments; TCP on the sender will just retransmit it 1797 * later. Setting the buffer size too big may only consume too 1798 * much kernel memory if the application doesn't read() from 1799 * the socket or packet loss or reordering makes use of the 1800 * reassembly queue. 1801 * 1802 * The criteria to step up the receive buffer one notch are: 1803 * 1. Application has not set receive buffer size with 1804 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1805 * 2. the number of bytes received during the time it takes 1806 * one timestamp to be reflected back to us (the RTT); 1807 * 3. received bytes per RTT is within seven eighth of the 1808 * current socket buffer size; 1809 * 4. receive buffer size has not hit maximal automatic size; 1810 * 1811 * This algorithm does one step per RTT at most and only if 1812 * we receive a bulk stream w/o packet losses or reorderings. 1813 * Shrinking the buffer during idle times is not necessary as 1814 * it doesn't consume any memory when idle. 1815 * 1816 * TODO: Only step up if the application is actually serving 1817 * the buffer to better manage the socket buffer resources. 1818 */ 1819 if (V_tcp_do_autorcvbuf && 1820 to.to_tsecr && 1821 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1822 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1823 to.to_tsecr - tp->rfbuf_ts < hz) { 1824 if (tp->rfbuf_cnt > 1825 (so->so_rcv.sb_hiwat / 8 * 7) && 1826 so->so_rcv.sb_hiwat < 1827 V_tcp_autorcvbuf_max) { 1828 newsize = 1829 min(so->so_rcv.sb_hiwat + 1830 V_tcp_autorcvbuf_inc, 1831 V_tcp_autorcvbuf_max); 1832 } 1833 /* Start over with next RTT. */ 1834 tp->rfbuf_ts = 0; 1835 tp->rfbuf_cnt = 0; 1836 } else 1837 tp->rfbuf_cnt += tlen; /* add up */ 1838 } 1839 1840 /* Add data to socket buffer. */ 1841 SOCKBUF_LOCK(&so->so_rcv); 1842 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1843 m_freem(m); 1844 } else { 1845 /* 1846 * Set new socket buffer size. 1847 * Give up when limit is reached. 1848 */ 1849 if (newsize) 1850 if (!sbreserve_locked(&so->so_rcv, 1851 newsize, so, NULL)) 1852 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1853 m_adj(m, drop_hdrlen); /* delayed header drop */ 1854 sbappendstream_locked(&so->so_rcv, m, 0); 1855 } 1856 /* NB: sorwakeup_locked() does an implicit unlock. */ 1857 sorwakeup_locked(so); 1858 if (DELAY_ACK(tp, tlen)) { 1859 tp->t_flags |= TF_DELACK; 1860 } else { 1861 tp->t_flags |= TF_ACKNOW; 1862 tcp_output(tp); 1863 } 1864 goto check_delack; 1865 } 1866 } 1867 1868 /* 1869 * Calculate amount of space in receive window, 1870 * and then do TCP input processing. 1871 * Receive window is amount of space in rcv queue, 1872 * but not less than advertised window. 1873 */ 1874 win = sbspace(&so->so_rcv); 1875 if (win < 0) 1876 win = 0; 1877 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1878 1879 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1880 tp->rfbuf_ts = 0; 1881 tp->rfbuf_cnt = 0; 1882 1883 switch (tp->t_state) { 1884 1885 /* 1886 * If the state is SYN_RECEIVED: 1887 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1888 */ 1889 case TCPS_SYN_RECEIVED: 1890 if ((thflags & TH_ACK) && 1891 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1892 SEQ_GT(th->th_ack, tp->snd_max))) { 1893 rstreason = BANDLIM_RST_OPENPORT; 1894 goto dropwithreset; 1895 } 1896 break; 1897 1898 /* 1899 * If the state is SYN_SENT: 1900 * if seg contains an ACK, but not for our SYN, drop the input. 1901 * if seg contains a RST, then drop the connection. 1902 * if seg does not contain SYN, then drop it. 1903 * Otherwise this is an acceptable SYN segment 1904 * initialize tp->rcv_nxt and tp->irs 1905 * if seg contains ack then advance tp->snd_una 1906 * if seg contains an ECE and ECN support is enabled, the stream 1907 * is ECN capable. 1908 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1909 * arrange for segment to be acked (eventually) 1910 * continue processing rest of data/controls, beginning with URG 1911 */ 1912 case TCPS_SYN_SENT: 1913 if ((thflags & TH_ACK) && 1914 (SEQ_LEQ(th->th_ack, tp->iss) || 1915 SEQ_GT(th->th_ack, tp->snd_max))) { 1916 rstreason = BANDLIM_UNLIMITED; 1917 goto dropwithreset; 1918 } 1919 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1920 TCP_PROBE5(connect__refused, NULL, tp, 1921 mtod(m, const char *), tp, th); 1922 tp = tcp_drop(tp, ECONNREFUSED); 1923 } 1924 if (thflags & TH_RST) 1925 goto drop; 1926 if (!(thflags & TH_SYN)) 1927 goto drop; 1928 1929 tp->irs = th->th_seq; 1930 tcp_rcvseqinit(tp); 1931 if (thflags & TH_ACK) { 1932 TCPSTAT_INC(tcps_connects); 1933 soisconnected(so); 1934 #ifdef MAC 1935 mac_socketpeer_set_from_mbuf(m, so); 1936 #endif 1937 /* Do window scaling on this connection? */ 1938 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1939 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1940 tp->rcv_scale = tp->request_r_scale; 1941 } 1942 tp->rcv_adv += imin(tp->rcv_wnd, 1943 TCP_MAXWIN << tp->rcv_scale); 1944 tp->snd_una++; /* SYN is acked */ 1945 /* 1946 * If there's data, delay ACK; if there's also a FIN 1947 * ACKNOW will be turned on later. 1948 */ 1949 if (DELAY_ACK(tp, tlen) && tlen != 0) 1950 tcp_timer_activate(tp, TT_DELACK, 1951 tcp_delacktime); 1952 else 1953 tp->t_flags |= TF_ACKNOW; 1954 1955 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1956 tp->t_flags |= TF_ECN_PERMIT; 1957 TCPSTAT_INC(tcps_ecn_shs); 1958 } 1959 1960 /* 1961 * Received <SYN,ACK> in SYN_SENT[*] state. 1962 * Transitions: 1963 * SYN_SENT --> ESTABLISHED 1964 * SYN_SENT* --> FIN_WAIT_1 1965 */ 1966 tp->t_starttime = ticks; 1967 if (tp->t_flags & TF_NEEDFIN) { 1968 tcp_state_change(tp, TCPS_FIN_WAIT_1); 1969 tp->t_flags &= ~TF_NEEDFIN; 1970 thflags &= ~TH_SYN; 1971 } else { 1972 tcp_state_change(tp, TCPS_ESTABLISHED); 1973 TCP_PROBE5(connect__established, NULL, tp, 1974 mtod(m, const char *), tp, th); 1975 cc_conn_init(tp); 1976 tcp_timer_activate(tp, TT_KEEP, 1977 TP_KEEPIDLE(tp)); 1978 } 1979 } else { 1980 /* 1981 * Received initial SYN in SYN-SENT[*] state => 1982 * simultaneous open. 1983 * If it succeeds, connection is * half-synchronized. 1984 * Otherwise, do 3-way handshake: 1985 * SYN-SENT -> SYN-RECEIVED 1986 * SYN-SENT* -> SYN-RECEIVED* 1987 */ 1988 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1989 tcp_timer_activate(tp, TT_REXMT, 0); 1990 tcp_state_change(tp, TCPS_SYN_RECEIVED); 1991 } 1992 1993 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1994 "ti_locked %d", __func__, ti_locked)); 1995 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1996 INP_WLOCK_ASSERT(tp->t_inpcb); 1997 1998 /* 1999 * Advance th->th_seq to correspond to first data byte. 2000 * If data, trim to stay within window, 2001 * dropping FIN if necessary. 2002 */ 2003 th->th_seq++; 2004 if (tlen > tp->rcv_wnd) { 2005 todrop = tlen - tp->rcv_wnd; 2006 m_adj(m, -todrop); 2007 tlen = tp->rcv_wnd; 2008 thflags &= ~TH_FIN; 2009 TCPSTAT_INC(tcps_rcvpackafterwin); 2010 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2011 } 2012 tp->snd_wl1 = th->th_seq - 1; 2013 tp->rcv_up = th->th_seq; 2014 /* 2015 * Client side of transaction: already sent SYN and data. 2016 * If the remote host used T/TCP to validate the SYN, 2017 * our data will be ACK'd; if so, enter normal data segment 2018 * processing in the middle of step 5, ack processing. 2019 * Otherwise, goto step 6. 2020 */ 2021 if (thflags & TH_ACK) 2022 goto process_ACK; 2023 2024 goto step6; 2025 2026 /* 2027 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2028 * do normal processing. 2029 * 2030 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2031 */ 2032 case TCPS_LAST_ACK: 2033 case TCPS_CLOSING: 2034 break; /* continue normal processing */ 2035 } 2036 2037 /* 2038 * States other than LISTEN or SYN_SENT. 2039 * First check the RST flag and sequence number since reset segments 2040 * are exempt from the timestamp and connection count tests. This 2041 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2042 * below which allowed reset segments in half the sequence space 2043 * to fall though and be processed (which gives forged reset 2044 * segments with a random sequence number a 50 percent chance of 2045 * killing a connection). 2046 * Then check timestamp, if present. 2047 * Then check the connection count, if present. 2048 * Then check that at least some bytes of segment are within 2049 * receive window. If segment begins before rcv_nxt, 2050 * drop leading data (and SYN); if nothing left, just ack. 2051 */ 2052 if (thflags & TH_RST) { 2053 /* 2054 * RFC5961 Section 3.2 2055 * 2056 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2057 * - If RST is in window, we send challenge ACK. 2058 * 2059 * Note: to take into account delayed ACKs, we should 2060 * test against last_ack_sent instead of rcv_nxt. 2061 * Note 2: we handle special case of closed window, not 2062 * covered by the RFC. 2063 */ 2064 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2065 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2066 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2067 2068 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2069 KASSERT(ti_locked == TI_WLOCKED, 2070 ("%s: TH_RST ti_locked %d, th %p tp %p", 2071 __func__, ti_locked, th, tp)); 2072 KASSERT(tp->t_state != TCPS_SYN_SENT, 2073 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2074 __func__, th, tp)); 2075 2076 if (V_tcp_insecure_rst || 2077 tp->last_ack_sent == th->th_seq) { 2078 TCPSTAT_INC(tcps_drops); 2079 /* Drop the connection. */ 2080 switch (tp->t_state) { 2081 case TCPS_SYN_RECEIVED: 2082 so->so_error = ECONNREFUSED; 2083 goto close; 2084 case TCPS_ESTABLISHED: 2085 case TCPS_FIN_WAIT_1: 2086 case TCPS_FIN_WAIT_2: 2087 case TCPS_CLOSE_WAIT: 2088 so->so_error = ECONNRESET; 2089 close: 2090 tcp_state_change(tp, TCPS_CLOSED); 2091 /* FALLTHROUGH */ 2092 default: 2093 tp = tcp_close(tp); 2094 } 2095 } else { 2096 TCPSTAT_INC(tcps_badrst); 2097 /* Send challenge ACK. */ 2098 tcp_respond(tp, mtod(m, void *), th, m, 2099 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2100 tp->last_ack_sent = tp->rcv_nxt; 2101 m = NULL; 2102 } 2103 } 2104 goto drop; 2105 } 2106 2107 /* 2108 * RFC5961 Section 4.2 2109 * Send challenge ACK for any SYN in synchronized state. 2110 */ 2111 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT) { 2112 KASSERT(ti_locked == TI_WLOCKED, 2113 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2114 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2115 2116 TCPSTAT_INC(tcps_badsyn); 2117 if (V_tcp_insecure_syn && 2118 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2119 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2120 tp = tcp_drop(tp, ECONNRESET); 2121 rstreason = BANDLIM_UNLIMITED; 2122 } else { 2123 /* Send challenge ACK. */ 2124 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2125 tp->snd_nxt, TH_ACK); 2126 tp->last_ack_sent = tp->rcv_nxt; 2127 m = NULL; 2128 } 2129 goto drop; 2130 } 2131 2132 /* 2133 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2134 * and it's less than ts_recent, drop it. 2135 */ 2136 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2137 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2138 2139 /* Check to see if ts_recent is over 24 days old. */ 2140 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2141 /* 2142 * Invalidate ts_recent. If this segment updates 2143 * ts_recent, the age will be reset later and ts_recent 2144 * will get a valid value. If it does not, setting 2145 * ts_recent to zero will at least satisfy the 2146 * requirement that zero be placed in the timestamp 2147 * echo reply when ts_recent isn't valid. The 2148 * age isn't reset until we get a valid ts_recent 2149 * because we don't want out-of-order segments to be 2150 * dropped when ts_recent is old. 2151 */ 2152 tp->ts_recent = 0; 2153 } else { 2154 TCPSTAT_INC(tcps_rcvduppack); 2155 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2156 TCPSTAT_INC(tcps_pawsdrop); 2157 if (tlen) 2158 goto dropafterack; 2159 goto drop; 2160 } 2161 } 2162 2163 /* 2164 * In the SYN-RECEIVED state, validate that the packet belongs to 2165 * this connection before trimming the data to fit the receive 2166 * window. Check the sequence number versus IRS since we know 2167 * the sequence numbers haven't wrapped. This is a partial fix 2168 * for the "LAND" DoS attack. 2169 */ 2170 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2171 rstreason = BANDLIM_RST_OPENPORT; 2172 goto dropwithreset; 2173 } 2174 2175 todrop = tp->rcv_nxt - th->th_seq; 2176 if (todrop > 0) { 2177 if (thflags & TH_SYN) { 2178 thflags &= ~TH_SYN; 2179 th->th_seq++; 2180 if (th->th_urp > 1) 2181 th->th_urp--; 2182 else 2183 thflags &= ~TH_URG; 2184 todrop--; 2185 } 2186 /* 2187 * Following if statement from Stevens, vol. 2, p. 960. 2188 */ 2189 if (todrop > tlen 2190 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2191 /* 2192 * Any valid FIN must be to the left of the window. 2193 * At this point the FIN must be a duplicate or out 2194 * of sequence; drop it. 2195 */ 2196 thflags &= ~TH_FIN; 2197 2198 /* 2199 * Send an ACK to resynchronize and drop any data. 2200 * But keep on processing for RST or ACK. 2201 */ 2202 tp->t_flags |= TF_ACKNOW; 2203 todrop = tlen; 2204 TCPSTAT_INC(tcps_rcvduppack); 2205 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2206 } else { 2207 TCPSTAT_INC(tcps_rcvpartduppack); 2208 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2209 } 2210 drop_hdrlen += todrop; /* drop from the top afterwards */ 2211 th->th_seq += todrop; 2212 tlen -= todrop; 2213 if (th->th_urp > todrop) 2214 th->th_urp -= todrop; 2215 else { 2216 thflags &= ~TH_URG; 2217 th->th_urp = 0; 2218 } 2219 } 2220 2221 /* 2222 * If new data are received on a connection after the 2223 * user processes are gone, then RST the other end. 2224 */ 2225 if ((so->so_state & SS_NOFDREF) && 2226 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2227 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2228 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2229 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2230 2231 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2232 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2233 "after socket was closed, " 2234 "sending RST and removing tcpcb\n", 2235 s, __func__, tcpstates[tp->t_state], tlen); 2236 free(s, M_TCPLOG); 2237 } 2238 tp = tcp_close(tp); 2239 TCPSTAT_INC(tcps_rcvafterclose); 2240 rstreason = BANDLIM_UNLIMITED; 2241 goto dropwithreset; 2242 } 2243 2244 /* 2245 * If segment ends after window, drop trailing data 2246 * (and PUSH and FIN); if nothing left, just ACK. 2247 */ 2248 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2249 if (todrop > 0) { 2250 TCPSTAT_INC(tcps_rcvpackafterwin); 2251 if (todrop >= tlen) { 2252 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2253 /* 2254 * If window is closed can only take segments at 2255 * window edge, and have to drop data and PUSH from 2256 * incoming segments. Continue processing, but 2257 * remember to ack. Otherwise, drop segment 2258 * and ack. 2259 */ 2260 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2261 tp->t_flags |= TF_ACKNOW; 2262 TCPSTAT_INC(tcps_rcvwinprobe); 2263 } else 2264 goto dropafterack; 2265 } else 2266 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2267 m_adj(m, -todrop); 2268 tlen -= todrop; 2269 thflags &= ~(TH_PUSH|TH_FIN); 2270 } 2271 2272 /* 2273 * If last ACK falls within this segment's sequence numbers, 2274 * record its timestamp. 2275 * NOTE: 2276 * 1) That the test incorporates suggestions from the latest 2277 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2278 * 2) That updating only on newer timestamps interferes with 2279 * our earlier PAWS tests, so this check should be solely 2280 * predicated on the sequence space of this segment. 2281 * 3) That we modify the segment boundary check to be 2282 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2283 * instead of RFC1323's 2284 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2285 * This modified check allows us to overcome RFC1323's 2286 * limitations as described in Stevens TCP/IP Illustrated 2287 * Vol. 2 p.869. In such cases, we can still calculate the 2288 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2289 */ 2290 if ((to.to_flags & TOF_TS) != 0 && 2291 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2292 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2293 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2294 tp->ts_recent_age = tcp_ts_getticks(); 2295 tp->ts_recent = to.to_tsval; 2296 } 2297 2298 /* 2299 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2300 * flag is on (half-synchronized state), then queue data for 2301 * later processing; else drop segment and return. 2302 */ 2303 if ((thflags & TH_ACK) == 0) { 2304 if (tp->t_state == TCPS_SYN_RECEIVED || 2305 (tp->t_flags & TF_NEEDSYN)) 2306 goto step6; 2307 else if (tp->t_flags & TF_ACKNOW) 2308 goto dropafterack; 2309 else 2310 goto drop; 2311 } 2312 2313 /* 2314 * Ack processing. 2315 */ 2316 switch (tp->t_state) { 2317 2318 /* 2319 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2320 * ESTABLISHED state and continue processing. 2321 * The ACK was checked above. 2322 */ 2323 case TCPS_SYN_RECEIVED: 2324 2325 TCPSTAT_INC(tcps_connects); 2326 soisconnected(so); 2327 /* Do window scaling? */ 2328 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2329 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2330 tp->rcv_scale = tp->request_r_scale; 2331 tp->snd_wnd = tiwin; 2332 } 2333 /* 2334 * Make transitions: 2335 * SYN-RECEIVED -> ESTABLISHED 2336 * SYN-RECEIVED* -> FIN-WAIT-1 2337 */ 2338 tp->t_starttime = ticks; 2339 if (tp->t_flags & TF_NEEDFIN) { 2340 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2341 tp->t_flags &= ~TF_NEEDFIN; 2342 } else { 2343 tcp_state_change(tp, TCPS_ESTABLISHED); 2344 TCP_PROBE5(accept__established, NULL, tp, 2345 mtod(m, const char *), tp, th); 2346 cc_conn_init(tp); 2347 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2348 } 2349 /* 2350 * If segment contains data or ACK, will call tcp_reass() 2351 * later; if not, do so now to pass queued data to user. 2352 */ 2353 if (tlen == 0 && (thflags & TH_FIN) == 0) 2354 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2355 (struct mbuf *)0); 2356 tp->snd_wl1 = th->th_seq - 1; 2357 /* FALLTHROUGH */ 2358 2359 /* 2360 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2361 * ACKs. If the ack is in the range 2362 * tp->snd_una < th->th_ack <= tp->snd_max 2363 * then advance tp->snd_una to th->th_ack and drop 2364 * data from the retransmission queue. If this ACK reflects 2365 * more up to date window information we update our window information. 2366 */ 2367 case TCPS_ESTABLISHED: 2368 case TCPS_FIN_WAIT_1: 2369 case TCPS_FIN_WAIT_2: 2370 case TCPS_CLOSE_WAIT: 2371 case TCPS_CLOSING: 2372 case TCPS_LAST_ACK: 2373 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2374 TCPSTAT_INC(tcps_rcvacktoomuch); 2375 goto dropafterack; 2376 } 2377 if ((tp->t_flags & TF_SACK_PERMIT) && 2378 ((to.to_flags & TOF_SACK) || 2379 !TAILQ_EMPTY(&tp->snd_holes))) 2380 tcp_sack_doack(tp, &to, th->th_ack); 2381 2382 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2383 hhook_run_tcp_est_in(tp, th, &to); 2384 2385 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2386 if (tlen == 0 && tiwin == tp->snd_wnd) { 2387 /* 2388 * If this is the first time we've seen a 2389 * FIN from the remote, this is not a 2390 * duplicate and it needs to be processed 2391 * normally. This happens during a 2392 * simultaneous close. 2393 */ 2394 if ((thflags & TH_FIN) && 2395 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2396 tp->t_dupacks = 0; 2397 break; 2398 } 2399 TCPSTAT_INC(tcps_rcvdupack); 2400 /* 2401 * If we have outstanding data (other than 2402 * a window probe), this is a completely 2403 * duplicate ack (ie, window info didn't 2404 * change and FIN isn't set), 2405 * the ack is the biggest we've 2406 * seen and we've seen exactly our rexmt 2407 * threshhold of them, assume a packet 2408 * has been dropped and retransmit it. 2409 * Kludge snd_nxt & the congestion 2410 * window so we send only this one 2411 * packet. 2412 * 2413 * We know we're losing at the current 2414 * window size so do congestion avoidance 2415 * (set ssthresh to half the current window 2416 * and pull our congestion window back to 2417 * the new ssthresh). 2418 * 2419 * Dup acks mean that packets have left the 2420 * network (they're now cached at the receiver) 2421 * so bump cwnd by the amount in the receiver 2422 * to keep a constant cwnd packets in the 2423 * network. 2424 * 2425 * When using TCP ECN, notify the peer that 2426 * we reduced the cwnd. 2427 */ 2428 if (!tcp_timer_active(tp, TT_REXMT) || 2429 th->th_ack != tp->snd_una) 2430 tp->t_dupacks = 0; 2431 else if (++tp->t_dupacks > tcprexmtthresh || 2432 IN_FASTRECOVERY(tp->t_flags)) { 2433 cc_ack_received(tp, th, CC_DUPACK); 2434 if ((tp->t_flags & TF_SACK_PERMIT) && 2435 IN_FASTRECOVERY(tp->t_flags)) { 2436 int awnd; 2437 2438 /* 2439 * Compute the amount of data in flight first. 2440 * We can inject new data into the pipe iff 2441 * we have less than 1/2 the original window's 2442 * worth of data in flight. 2443 */ 2444 awnd = (tp->snd_nxt - tp->snd_fack) + 2445 tp->sackhint.sack_bytes_rexmit; 2446 if (awnd < tp->snd_ssthresh) { 2447 tp->snd_cwnd += tp->t_maxseg; 2448 if (tp->snd_cwnd > tp->snd_ssthresh) 2449 tp->snd_cwnd = tp->snd_ssthresh; 2450 } 2451 } else 2452 tp->snd_cwnd += tp->t_maxseg; 2453 (void) tcp_output(tp); 2454 goto drop; 2455 } else if (tp->t_dupacks == tcprexmtthresh) { 2456 tcp_seq onxt = tp->snd_nxt; 2457 2458 /* 2459 * If we're doing sack, check to 2460 * see if we're already in sack 2461 * recovery. If we're not doing sack, 2462 * check to see if we're in newreno 2463 * recovery. 2464 */ 2465 if (tp->t_flags & TF_SACK_PERMIT) { 2466 if (IN_FASTRECOVERY(tp->t_flags)) { 2467 tp->t_dupacks = 0; 2468 break; 2469 } 2470 } else { 2471 if (SEQ_LEQ(th->th_ack, 2472 tp->snd_recover)) { 2473 tp->t_dupacks = 0; 2474 break; 2475 } 2476 } 2477 /* Congestion signal before ack. */ 2478 cc_cong_signal(tp, th, CC_NDUPACK); 2479 cc_ack_received(tp, th, CC_DUPACK); 2480 tcp_timer_activate(tp, TT_REXMT, 0); 2481 tp->t_rtttime = 0; 2482 if (tp->t_flags & TF_SACK_PERMIT) { 2483 TCPSTAT_INC( 2484 tcps_sack_recovery_episode); 2485 tp->sack_newdata = tp->snd_nxt; 2486 tp->snd_cwnd = tp->t_maxseg; 2487 (void) tcp_output(tp); 2488 goto drop; 2489 } 2490 tp->snd_nxt = th->th_ack; 2491 tp->snd_cwnd = tp->t_maxseg; 2492 (void) tcp_output(tp); 2493 KASSERT(tp->snd_limited <= 2, 2494 ("%s: tp->snd_limited too big", 2495 __func__)); 2496 tp->snd_cwnd = tp->snd_ssthresh + 2497 tp->t_maxseg * 2498 (tp->t_dupacks - tp->snd_limited); 2499 if (SEQ_GT(onxt, tp->snd_nxt)) 2500 tp->snd_nxt = onxt; 2501 goto drop; 2502 } else if (V_tcp_do_rfc3042) { 2503 cc_ack_received(tp, th, CC_DUPACK); 2504 u_long oldcwnd = tp->snd_cwnd; 2505 tcp_seq oldsndmax = tp->snd_max; 2506 u_int sent; 2507 int avail; 2508 2509 KASSERT(tp->t_dupacks == 1 || 2510 tp->t_dupacks == 2, 2511 ("%s: dupacks not 1 or 2", 2512 __func__)); 2513 if (tp->t_dupacks == 1) 2514 tp->snd_limited = 0; 2515 tp->snd_cwnd = 2516 (tp->snd_nxt - tp->snd_una) + 2517 (tp->t_dupacks - tp->snd_limited) * 2518 tp->t_maxseg; 2519 /* 2520 * Only call tcp_output when there 2521 * is new data available to be sent. 2522 * Otherwise we would send pure ACKs. 2523 */ 2524 SOCKBUF_LOCK(&so->so_snd); 2525 avail = sbavail(&so->so_snd) - 2526 (tp->snd_nxt - tp->snd_una); 2527 SOCKBUF_UNLOCK(&so->so_snd); 2528 if (avail > 0) 2529 (void) tcp_output(tp); 2530 sent = tp->snd_max - oldsndmax; 2531 if (sent > tp->t_maxseg) { 2532 KASSERT((tp->t_dupacks == 2 && 2533 tp->snd_limited == 0) || 2534 (sent == tp->t_maxseg + 1 && 2535 tp->t_flags & TF_SENTFIN), 2536 ("%s: sent too much", 2537 __func__)); 2538 tp->snd_limited = 2; 2539 } else if (sent > 0) 2540 ++tp->snd_limited; 2541 tp->snd_cwnd = oldcwnd; 2542 goto drop; 2543 } 2544 } else 2545 tp->t_dupacks = 0; 2546 break; 2547 } 2548 2549 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2550 ("%s: th_ack <= snd_una", __func__)); 2551 2552 /* 2553 * If the congestion window was inflated to account 2554 * for the other side's cached packets, retract it. 2555 */ 2556 if (IN_FASTRECOVERY(tp->t_flags)) { 2557 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2558 if (tp->t_flags & TF_SACK_PERMIT) 2559 tcp_sack_partialack(tp, th); 2560 else 2561 tcp_newreno_partial_ack(tp, th); 2562 } else 2563 cc_post_recovery(tp, th); 2564 } 2565 tp->t_dupacks = 0; 2566 /* 2567 * If we reach this point, ACK is not a duplicate, 2568 * i.e., it ACKs something we sent. 2569 */ 2570 if (tp->t_flags & TF_NEEDSYN) { 2571 /* 2572 * T/TCP: Connection was half-synchronized, and our 2573 * SYN has been ACK'd (so connection is now fully 2574 * synchronized). Go to non-starred state, 2575 * increment snd_una for ACK of SYN, and check if 2576 * we can do window scaling. 2577 */ 2578 tp->t_flags &= ~TF_NEEDSYN; 2579 tp->snd_una++; 2580 /* Do window scaling? */ 2581 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2582 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2583 tp->rcv_scale = tp->request_r_scale; 2584 /* Send window already scaled. */ 2585 } 2586 } 2587 2588 process_ACK: 2589 INP_WLOCK_ASSERT(tp->t_inpcb); 2590 2591 acked = BYTES_THIS_ACK(tp, th); 2592 TCPSTAT_INC(tcps_rcvackpack); 2593 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2594 2595 /* 2596 * If we just performed our first retransmit, and the ACK 2597 * arrives within our recovery window, then it was a mistake 2598 * to do the retransmit in the first place. Recover our 2599 * original cwnd and ssthresh, and proceed to transmit where 2600 * we left off. 2601 */ 2602 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2603 (int)(ticks - tp->t_badrxtwin) < 0) 2604 cc_cong_signal(tp, th, CC_RTO_ERR); 2605 2606 /* 2607 * If we have a timestamp reply, update smoothed 2608 * round trip time. If no timestamp is present but 2609 * transmit timer is running and timed sequence 2610 * number was acked, update smoothed round trip time. 2611 * Since we now have an rtt measurement, cancel the 2612 * timer backoff (cf., Phil Karn's retransmit alg.). 2613 * Recompute the initial retransmit timer. 2614 * 2615 * Some boxes send broken timestamp replies 2616 * during the SYN+ACK phase, ignore 2617 * timestamps of 0 or we could calculate a 2618 * huge RTT and blow up the retransmit timer. 2619 */ 2620 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2621 u_int t; 2622 2623 t = tcp_ts_getticks() - to.to_tsecr; 2624 if (!tp->t_rttlow || tp->t_rttlow > t) 2625 tp->t_rttlow = t; 2626 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2627 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2628 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2629 tp->t_rttlow = ticks - tp->t_rtttime; 2630 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2631 } 2632 2633 /* 2634 * If all outstanding data is acked, stop retransmit 2635 * timer and remember to restart (more output or persist). 2636 * If there is more data to be acked, restart retransmit 2637 * timer, using current (possibly backed-off) value. 2638 */ 2639 if (th->th_ack == tp->snd_max) { 2640 tcp_timer_activate(tp, TT_REXMT, 0); 2641 needoutput = 1; 2642 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2643 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2644 2645 /* 2646 * If no data (only SYN) was ACK'd, 2647 * skip rest of ACK processing. 2648 */ 2649 if (acked == 0) 2650 goto step6; 2651 2652 /* 2653 * Let the congestion control algorithm update congestion 2654 * control related information. This typically means increasing 2655 * the congestion window. 2656 */ 2657 cc_ack_received(tp, th, CC_ACK); 2658 2659 SOCKBUF_LOCK(&so->so_snd); 2660 if (acked > sbavail(&so->so_snd)) { 2661 tp->snd_wnd -= sbavail(&so->so_snd); 2662 mfree = sbcut_locked(&so->so_snd, 2663 (int)sbavail(&so->so_snd)); 2664 ourfinisacked = 1; 2665 } else { 2666 mfree = sbcut_locked(&so->so_snd, acked); 2667 tp->snd_wnd -= acked; 2668 ourfinisacked = 0; 2669 } 2670 /* NB: sowwakeup_locked() does an implicit unlock. */ 2671 sowwakeup_locked(so); 2672 m_freem(mfree); 2673 /* Detect una wraparound. */ 2674 if (!IN_RECOVERY(tp->t_flags) && 2675 SEQ_GT(tp->snd_una, tp->snd_recover) && 2676 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2677 tp->snd_recover = th->th_ack - 1; 2678 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2679 if (IN_RECOVERY(tp->t_flags) && 2680 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2681 EXIT_RECOVERY(tp->t_flags); 2682 } 2683 tp->snd_una = th->th_ack; 2684 if (tp->t_flags & TF_SACK_PERMIT) { 2685 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2686 tp->snd_recover = tp->snd_una; 2687 } 2688 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2689 tp->snd_nxt = tp->snd_una; 2690 2691 switch (tp->t_state) { 2692 2693 /* 2694 * In FIN_WAIT_1 STATE in addition to the processing 2695 * for the ESTABLISHED state if our FIN is now acknowledged 2696 * then enter FIN_WAIT_2. 2697 */ 2698 case TCPS_FIN_WAIT_1: 2699 if (ourfinisacked) { 2700 /* 2701 * If we can't receive any more 2702 * data, then closing user can proceed. 2703 * Starting the timer is contrary to the 2704 * specification, but if we don't get a FIN 2705 * we'll hang forever. 2706 * 2707 * XXXjl: 2708 * we should release the tp also, and use a 2709 * compressed state. 2710 */ 2711 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2712 soisdisconnected(so); 2713 tcp_timer_activate(tp, TT_2MSL, 2714 (tcp_fast_finwait2_recycle ? 2715 tcp_finwait2_timeout : 2716 TP_MAXIDLE(tp))); 2717 } 2718 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2719 } 2720 break; 2721 2722 /* 2723 * In CLOSING STATE in addition to the processing for 2724 * the ESTABLISHED state if the ACK acknowledges our FIN 2725 * then enter the TIME-WAIT state, otherwise ignore 2726 * the segment. 2727 */ 2728 case TCPS_CLOSING: 2729 if (ourfinisacked) { 2730 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2731 tcp_twstart(tp); 2732 INP_INFO_WUNLOCK(&V_tcbinfo); 2733 m_freem(m); 2734 return; 2735 } 2736 break; 2737 2738 /* 2739 * In LAST_ACK, we may still be waiting for data to drain 2740 * and/or to be acked, as well as for the ack of our FIN. 2741 * If our FIN is now acknowledged, delete the TCB, 2742 * enter the closed state and return. 2743 */ 2744 case TCPS_LAST_ACK: 2745 if (ourfinisacked) { 2746 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2747 tp = tcp_close(tp); 2748 goto drop; 2749 } 2750 break; 2751 } 2752 } 2753 2754 step6: 2755 INP_WLOCK_ASSERT(tp->t_inpcb); 2756 2757 /* 2758 * Update window information. 2759 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2760 */ 2761 if ((thflags & TH_ACK) && 2762 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2763 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2764 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2765 /* keep track of pure window updates */ 2766 if (tlen == 0 && 2767 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2768 TCPSTAT_INC(tcps_rcvwinupd); 2769 tp->snd_wnd = tiwin; 2770 tp->snd_wl1 = th->th_seq; 2771 tp->snd_wl2 = th->th_ack; 2772 if (tp->snd_wnd > tp->max_sndwnd) 2773 tp->max_sndwnd = tp->snd_wnd; 2774 needoutput = 1; 2775 } 2776 2777 /* 2778 * Process segments with URG. 2779 */ 2780 if ((thflags & TH_URG) && th->th_urp && 2781 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2782 /* 2783 * This is a kludge, but if we receive and accept 2784 * random urgent pointers, we'll crash in 2785 * soreceive. It's hard to imagine someone 2786 * actually wanting to send this much urgent data. 2787 */ 2788 SOCKBUF_LOCK(&so->so_rcv); 2789 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2790 th->th_urp = 0; /* XXX */ 2791 thflags &= ~TH_URG; /* XXX */ 2792 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2793 goto dodata; /* XXX */ 2794 } 2795 /* 2796 * If this segment advances the known urgent pointer, 2797 * then mark the data stream. This should not happen 2798 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2799 * a FIN has been received from the remote side. 2800 * In these states we ignore the URG. 2801 * 2802 * According to RFC961 (Assigned Protocols), 2803 * the urgent pointer points to the last octet 2804 * of urgent data. We continue, however, 2805 * to consider it to indicate the first octet 2806 * of data past the urgent section as the original 2807 * spec states (in one of two places). 2808 */ 2809 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2810 tp->rcv_up = th->th_seq + th->th_urp; 2811 so->so_oobmark = sbavail(&so->so_rcv) + 2812 (tp->rcv_up - tp->rcv_nxt) - 1; 2813 if (so->so_oobmark == 0) 2814 so->so_rcv.sb_state |= SBS_RCVATMARK; 2815 sohasoutofband(so); 2816 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2817 } 2818 SOCKBUF_UNLOCK(&so->so_rcv); 2819 /* 2820 * Remove out of band data so doesn't get presented to user. 2821 * This can happen independent of advancing the URG pointer, 2822 * but if two URG's are pending at once, some out-of-band 2823 * data may creep in... ick. 2824 */ 2825 if (th->th_urp <= (u_long)tlen && 2826 !(so->so_options & SO_OOBINLINE)) { 2827 /* hdr drop is delayed */ 2828 tcp_pulloutofband(so, th, m, drop_hdrlen); 2829 } 2830 } else { 2831 /* 2832 * If no out of band data is expected, 2833 * pull receive urgent pointer along 2834 * with the receive window. 2835 */ 2836 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2837 tp->rcv_up = tp->rcv_nxt; 2838 } 2839 dodata: /* XXX */ 2840 INP_WLOCK_ASSERT(tp->t_inpcb); 2841 2842 /* 2843 * Process the segment text, merging it into the TCP sequencing queue, 2844 * and arranging for acknowledgment of receipt if necessary. 2845 * This process logically involves adjusting tp->rcv_wnd as data 2846 * is presented to the user (this happens in tcp_usrreq.c, 2847 * case PRU_RCVD). If a FIN has already been received on this 2848 * connection then we just ignore the text. 2849 */ 2850 if ((tlen || (thflags & TH_FIN)) && 2851 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2852 tcp_seq save_start = th->th_seq; 2853 m_adj(m, drop_hdrlen); /* delayed header drop */ 2854 /* 2855 * Insert segment which includes th into TCP reassembly queue 2856 * with control block tp. Set thflags to whether reassembly now 2857 * includes a segment with FIN. This handles the common case 2858 * inline (segment is the next to be received on an established 2859 * connection, and the queue is empty), avoiding linkage into 2860 * and removal from the queue and repetition of various 2861 * conversions. 2862 * Set DELACK for segments received in order, but ack 2863 * immediately when segments are out of order (so 2864 * fast retransmit can work). 2865 */ 2866 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL && 2867 TCPS_HAVEESTABLISHED(tp->t_state)) { 2868 if (DELAY_ACK(tp, tlen)) 2869 tp->t_flags |= TF_DELACK; 2870 else 2871 tp->t_flags |= TF_ACKNOW; 2872 tp->rcv_nxt += tlen; 2873 thflags = th->th_flags & TH_FIN; 2874 TCPSTAT_INC(tcps_rcvpack); 2875 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2876 ND6_HINT(tp); 2877 SOCKBUF_LOCK(&so->so_rcv); 2878 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2879 m_freem(m); 2880 else 2881 sbappendstream_locked(&so->so_rcv, m, 0); 2882 /* NB: sorwakeup_locked() does an implicit unlock. */ 2883 sorwakeup_locked(so); 2884 } else { 2885 /* 2886 * XXX: Due to the header drop above "th" is 2887 * theoretically invalid by now. Fortunately 2888 * m_adj() doesn't actually frees any mbufs 2889 * when trimming from the head. 2890 */ 2891 thflags = tcp_reass(tp, th, &tlen, m); 2892 tp->t_flags |= TF_ACKNOW; 2893 } 2894 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2895 tcp_update_sack_list(tp, save_start, save_start + tlen); 2896 #if 0 2897 /* 2898 * Note the amount of data that peer has sent into 2899 * our window, in order to estimate the sender's 2900 * buffer size. 2901 * XXX: Unused. 2902 */ 2903 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2904 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2905 else 2906 len = so->so_rcv.sb_hiwat; 2907 #endif 2908 } else { 2909 m_freem(m); 2910 thflags &= ~TH_FIN; 2911 } 2912 2913 /* 2914 * If FIN is received ACK the FIN and let the user know 2915 * that the connection is closing. 2916 */ 2917 if (thflags & TH_FIN) { 2918 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2919 socantrcvmore(so); 2920 /* 2921 * If connection is half-synchronized 2922 * (ie NEEDSYN flag on) then delay ACK, 2923 * so it may be piggybacked when SYN is sent. 2924 * Otherwise, since we received a FIN then no 2925 * more input can be expected, send ACK now. 2926 */ 2927 if (tp->t_flags & TF_NEEDSYN) 2928 tp->t_flags |= TF_DELACK; 2929 else 2930 tp->t_flags |= TF_ACKNOW; 2931 tp->rcv_nxt++; 2932 } 2933 switch (tp->t_state) { 2934 2935 /* 2936 * In SYN_RECEIVED and ESTABLISHED STATES 2937 * enter the CLOSE_WAIT state. 2938 */ 2939 case TCPS_SYN_RECEIVED: 2940 tp->t_starttime = ticks; 2941 /* FALLTHROUGH */ 2942 case TCPS_ESTABLISHED: 2943 tcp_state_change(tp, TCPS_CLOSE_WAIT); 2944 break; 2945 2946 /* 2947 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2948 * enter the CLOSING state. 2949 */ 2950 case TCPS_FIN_WAIT_1: 2951 tcp_state_change(tp, TCPS_CLOSING); 2952 break; 2953 2954 /* 2955 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2956 * starting the time-wait timer, turning off the other 2957 * standard timers. 2958 */ 2959 case TCPS_FIN_WAIT_2: 2960 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2961 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2962 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2963 ti_locked)); 2964 2965 tcp_twstart(tp); 2966 INP_INFO_WUNLOCK(&V_tcbinfo); 2967 return; 2968 } 2969 } 2970 if (ti_locked == TI_WLOCKED) 2971 INP_INFO_WUNLOCK(&V_tcbinfo); 2972 ti_locked = TI_UNLOCKED; 2973 2974 #ifdef TCPDEBUG 2975 if (so->so_options & SO_DEBUG) 2976 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2977 &tcp_savetcp, 0); 2978 #endif 2979 2980 /* 2981 * Return any desired output. 2982 */ 2983 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2984 (void) tcp_output(tp); 2985 2986 check_delack: 2987 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2988 __func__, ti_locked)); 2989 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2990 INP_WLOCK_ASSERT(tp->t_inpcb); 2991 2992 if (tp->t_flags & TF_DELACK) { 2993 tp->t_flags &= ~TF_DELACK; 2994 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2995 } 2996 INP_WUNLOCK(tp->t_inpcb); 2997 return; 2998 2999 dropafterack: 3000 /* 3001 * Generate an ACK dropping incoming segment if it occupies 3002 * sequence space, where the ACK reflects our state. 3003 * 3004 * We can now skip the test for the RST flag since all 3005 * paths to this code happen after packets containing 3006 * RST have been dropped. 3007 * 3008 * In the SYN-RECEIVED state, don't send an ACK unless the 3009 * segment we received passes the SYN-RECEIVED ACK test. 3010 * If it fails send a RST. This breaks the loop in the 3011 * "LAND" DoS attack, and also prevents an ACK storm 3012 * between two listening ports that have been sent forged 3013 * SYN segments, each with the source address of the other. 3014 */ 3015 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3016 (SEQ_GT(tp->snd_una, th->th_ack) || 3017 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3018 rstreason = BANDLIM_RST_OPENPORT; 3019 goto dropwithreset; 3020 } 3021 #ifdef TCPDEBUG 3022 if (so->so_options & SO_DEBUG) 3023 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3024 &tcp_savetcp, 0); 3025 #endif 3026 if (ti_locked == TI_WLOCKED) 3027 INP_INFO_WUNLOCK(&V_tcbinfo); 3028 ti_locked = TI_UNLOCKED; 3029 3030 tp->t_flags |= TF_ACKNOW; 3031 (void) tcp_output(tp); 3032 INP_WUNLOCK(tp->t_inpcb); 3033 m_freem(m); 3034 return; 3035 3036 dropwithreset: 3037 if (ti_locked == TI_WLOCKED) 3038 INP_INFO_WUNLOCK(&V_tcbinfo); 3039 ti_locked = TI_UNLOCKED; 3040 3041 if (tp != NULL) { 3042 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3043 INP_WUNLOCK(tp->t_inpcb); 3044 } else 3045 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3046 return; 3047 3048 drop: 3049 if (ti_locked == TI_WLOCKED) { 3050 INP_INFO_WUNLOCK(&V_tcbinfo); 3051 ti_locked = TI_UNLOCKED; 3052 } 3053 #ifdef INVARIANTS 3054 else 3055 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3056 #endif 3057 3058 /* 3059 * Drop space held by incoming segment and return. 3060 */ 3061 #ifdef TCPDEBUG 3062 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3063 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3064 &tcp_savetcp, 0); 3065 #endif 3066 if (tp != NULL) 3067 INP_WUNLOCK(tp->t_inpcb); 3068 m_freem(m); 3069 } 3070 3071 /* 3072 * Issue RST and make ACK acceptable to originator of segment. 3073 * The mbuf must still include the original packet header. 3074 * tp may be NULL. 3075 */ 3076 static void 3077 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3078 int tlen, int rstreason) 3079 { 3080 #ifdef INET 3081 struct ip *ip; 3082 #endif 3083 #ifdef INET6 3084 struct ip6_hdr *ip6; 3085 #endif 3086 3087 if (tp != NULL) { 3088 INP_WLOCK_ASSERT(tp->t_inpcb); 3089 } 3090 3091 /* Don't bother if destination was broadcast/multicast. */ 3092 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3093 goto drop; 3094 #ifdef INET6 3095 if (mtod(m, struct ip *)->ip_v == 6) { 3096 ip6 = mtod(m, struct ip6_hdr *); 3097 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3098 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3099 goto drop; 3100 /* IPv6 anycast check is done at tcp6_input() */ 3101 } 3102 #endif 3103 #if defined(INET) && defined(INET6) 3104 else 3105 #endif 3106 #ifdef INET 3107 { 3108 ip = mtod(m, struct ip *); 3109 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3110 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3111 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3112 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3113 goto drop; 3114 } 3115 #endif 3116 3117 /* Perform bandwidth limiting. */ 3118 if (badport_bandlim(rstreason) < 0) 3119 goto drop; 3120 3121 /* tcp_respond consumes the mbuf chain. */ 3122 if (th->th_flags & TH_ACK) { 3123 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3124 th->th_ack, TH_RST); 3125 } else { 3126 if (th->th_flags & TH_SYN) 3127 tlen++; 3128 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3129 (tcp_seq)0, TH_RST|TH_ACK); 3130 } 3131 return; 3132 drop: 3133 m_freem(m); 3134 } 3135 3136 /* 3137 * Parse TCP options and place in tcpopt. 3138 */ 3139 static void 3140 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3141 { 3142 int opt, optlen; 3143 3144 to->to_flags = 0; 3145 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3146 opt = cp[0]; 3147 if (opt == TCPOPT_EOL) 3148 break; 3149 if (opt == TCPOPT_NOP) 3150 optlen = 1; 3151 else { 3152 if (cnt < 2) 3153 break; 3154 optlen = cp[1]; 3155 if (optlen < 2 || optlen > cnt) 3156 break; 3157 } 3158 switch (opt) { 3159 case TCPOPT_MAXSEG: 3160 if (optlen != TCPOLEN_MAXSEG) 3161 continue; 3162 if (!(flags & TO_SYN)) 3163 continue; 3164 to->to_flags |= TOF_MSS; 3165 bcopy((char *)cp + 2, 3166 (char *)&to->to_mss, sizeof(to->to_mss)); 3167 to->to_mss = ntohs(to->to_mss); 3168 break; 3169 case TCPOPT_WINDOW: 3170 if (optlen != TCPOLEN_WINDOW) 3171 continue; 3172 if (!(flags & TO_SYN)) 3173 continue; 3174 to->to_flags |= TOF_SCALE; 3175 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3176 break; 3177 case TCPOPT_TIMESTAMP: 3178 if (optlen != TCPOLEN_TIMESTAMP) 3179 continue; 3180 to->to_flags |= TOF_TS; 3181 bcopy((char *)cp + 2, 3182 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3183 to->to_tsval = ntohl(to->to_tsval); 3184 bcopy((char *)cp + 6, 3185 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3186 to->to_tsecr = ntohl(to->to_tsecr); 3187 break; 3188 #ifdef TCP_SIGNATURE 3189 /* 3190 * XXX In order to reply to a host which has set the 3191 * TCP_SIGNATURE option in its initial SYN, we have to 3192 * record the fact that the option was observed here 3193 * for the syncache code to perform the correct response. 3194 */ 3195 case TCPOPT_SIGNATURE: 3196 if (optlen != TCPOLEN_SIGNATURE) 3197 continue; 3198 to->to_flags |= TOF_SIGNATURE; 3199 to->to_signature = cp + 2; 3200 break; 3201 #endif 3202 case TCPOPT_SACK_PERMITTED: 3203 if (optlen != TCPOLEN_SACK_PERMITTED) 3204 continue; 3205 if (!(flags & TO_SYN)) 3206 continue; 3207 if (!V_tcp_do_sack) 3208 continue; 3209 to->to_flags |= TOF_SACKPERM; 3210 break; 3211 case TCPOPT_SACK: 3212 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3213 continue; 3214 if (flags & TO_SYN) 3215 continue; 3216 to->to_flags |= TOF_SACK; 3217 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3218 to->to_sacks = cp + 2; 3219 TCPSTAT_INC(tcps_sack_rcv_blocks); 3220 break; 3221 default: 3222 continue; 3223 } 3224 } 3225 } 3226 3227 /* 3228 * Pull out of band byte out of a segment so 3229 * it doesn't appear in the user's data queue. 3230 * It is still reflected in the segment length for 3231 * sequencing purposes. 3232 */ 3233 static void 3234 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3235 int off) 3236 { 3237 int cnt = off + th->th_urp - 1; 3238 3239 while (cnt >= 0) { 3240 if (m->m_len > cnt) { 3241 char *cp = mtod(m, caddr_t) + cnt; 3242 struct tcpcb *tp = sototcpcb(so); 3243 3244 INP_WLOCK_ASSERT(tp->t_inpcb); 3245 3246 tp->t_iobc = *cp; 3247 tp->t_oobflags |= TCPOOB_HAVEDATA; 3248 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3249 m->m_len--; 3250 if (m->m_flags & M_PKTHDR) 3251 m->m_pkthdr.len--; 3252 return; 3253 } 3254 cnt -= m->m_len; 3255 m = m->m_next; 3256 if (m == NULL) 3257 break; 3258 } 3259 panic("tcp_pulloutofband"); 3260 } 3261 3262 /* 3263 * Collect new round-trip time estimate 3264 * and update averages and current timeout. 3265 */ 3266 static void 3267 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3268 { 3269 int delta; 3270 3271 INP_WLOCK_ASSERT(tp->t_inpcb); 3272 3273 TCPSTAT_INC(tcps_rttupdated); 3274 tp->t_rttupdated++; 3275 if (tp->t_srtt != 0) { 3276 /* 3277 * srtt is stored as fixed point with 5 bits after the 3278 * binary point (i.e., scaled by 8). The following magic 3279 * is equivalent to the smoothing algorithm in rfc793 with 3280 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3281 * point). Adjust rtt to origin 0. 3282 */ 3283 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3284 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3285 3286 if ((tp->t_srtt += delta) <= 0) 3287 tp->t_srtt = 1; 3288 3289 /* 3290 * We accumulate a smoothed rtt variance (actually, a 3291 * smoothed mean difference), then set the retransmit 3292 * timer to smoothed rtt + 4 times the smoothed variance. 3293 * rttvar is stored as fixed point with 4 bits after the 3294 * binary point (scaled by 16). The following is 3295 * equivalent to rfc793 smoothing with an alpha of .75 3296 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3297 * rfc793's wired-in beta. 3298 */ 3299 if (delta < 0) 3300 delta = -delta; 3301 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3302 if ((tp->t_rttvar += delta) <= 0) 3303 tp->t_rttvar = 1; 3304 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3305 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3306 } else { 3307 /* 3308 * No rtt measurement yet - use the unsmoothed rtt. 3309 * Set the variance to half the rtt (so our first 3310 * retransmit happens at 3*rtt). 3311 */ 3312 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3313 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3314 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3315 } 3316 tp->t_rtttime = 0; 3317 tp->t_rxtshift = 0; 3318 3319 /* 3320 * the retransmit should happen at rtt + 4 * rttvar. 3321 * Because of the way we do the smoothing, srtt and rttvar 3322 * will each average +1/2 tick of bias. When we compute 3323 * the retransmit timer, we want 1/2 tick of rounding and 3324 * 1 extra tick because of +-1/2 tick uncertainty in the 3325 * firing of the timer. The bias will give us exactly the 3326 * 1.5 tick we need. But, because the bias is 3327 * statistical, we have to test that we don't drop below 3328 * the minimum feasible timer (which is 2 ticks). 3329 */ 3330 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3331 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3332 3333 /* 3334 * We received an ack for a packet that wasn't retransmitted; 3335 * it is probably safe to discard any error indications we've 3336 * received recently. This isn't quite right, but close enough 3337 * for now (a route might have failed after we sent a segment, 3338 * and the return path might not be symmetrical). 3339 */ 3340 tp->t_softerror = 0; 3341 } 3342 3343 /* 3344 * Determine a reasonable value for maxseg size. 3345 * If the route is known, check route for mtu. 3346 * If none, use an mss that can be handled on the outgoing interface 3347 * without forcing IP to fragment. If no route is found, route has no mtu, 3348 * or the destination isn't local, use a default, hopefully conservative 3349 * size (usually 512 or the default IP max size, but no more than the mtu 3350 * of the interface), as we can't discover anything about intervening 3351 * gateways or networks. We also initialize the congestion/slow start 3352 * window to be a single segment if the destination isn't local. 3353 * While looking at the routing entry, we also initialize other path-dependent 3354 * parameters from pre-set or cached values in the routing entry. 3355 * 3356 * Also take into account the space needed for options that we 3357 * send regularly. Make maxseg shorter by that amount to assure 3358 * that we can send maxseg amount of data even when the options 3359 * are present. Store the upper limit of the length of options plus 3360 * data in maxopd. 3361 * 3362 * NOTE that this routine is only called when we process an incoming 3363 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3364 * settings are handled in tcp_mssopt(). 3365 */ 3366 void 3367 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3368 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3369 { 3370 int mss = 0; 3371 u_long maxmtu = 0; 3372 struct inpcb *inp = tp->t_inpcb; 3373 struct hc_metrics_lite metrics; 3374 int origoffer; 3375 #ifdef INET6 3376 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3377 size_t min_protoh = isipv6 ? 3378 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3379 sizeof (struct tcpiphdr); 3380 #else 3381 const size_t min_protoh = sizeof(struct tcpiphdr); 3382 #endif 3383 3384 INP_WLOCK_ASSERT(tp->t_inpcb); 3385 3386 if (mtuoffer != -1) { 3387 KASSERT(offer == -1, ("%s: conflict", __func__)); 3388 offer = mtuoffer - min_protoh; 3389 } 3390 origoffer = offer; 3391 3392 /* Initialize. */ 3393 #ifdef INET6 3394 if (isipv6) { 3395 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3396 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3397 } 3398 #endif 3399 #if defined(INET) && defined(INET6) 3400 else 3401 #endif 3402 #ifdef INET 3403 { 3404 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3405 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3406 } 3407 #endif 3408 3409 /* 3410 * No route to sender, stay with default mss and return. 3411 */ 3412 if (maxmtu == 0) { 3413 /* 3414 * In case we return early we need to initialize metrics 3415 * to a defined state as tcp_hc_get() would do for us 3416 * if there was no cache hit. 3417 */ 3418 if (metricptr != NULL) 3419 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3420 return; 3421 } 3422 3423 /* What have we got? */ 3424 switch (offer) { 3425 case 0: 3426 /* 3427 * Offer == 0 means that there was no MSS on the SYN 3428 * segment, in this case we use tcp_mssdflt as 3429 * already assigned to t_maxopd above. 3430 */ 3431 offer = tp->t_maxopd; 3432 break; 3433 3434 case -1: 3435 /* 3436 * Offer == -1 means that we didn't receive SYN yet. 3437 */ 3438 /* FALLTHROUGH */ 3439 3440 default: 3441 /* 3442 * Prevent DoS attack with too small MSS. Round up 3443 * to at least minmss. 3444 */ 3445 offer = max(offer, V_tcp_minmss); 3446 } 3447 3448 /* 3449 * rmx information is now retrieved from tcp_hostcache. 3450 */ 3451 tcp_hc_get(&inp->inp_inc, &metrics); 3452 if (metricptr != NULL) 3453 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3454 3455 /* 3456 * If there's a discovered mtu in tcp hostcache, use it. 3457 * Else, use the link mtu. 3458 */ 3459 if (metrics.rmx_mtu) 3460 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3461 else { 3462 #ifdef INET6 3463 if (isipv6) { 3464 mss = maxmtu - min_protoh; 3465 if (!V_path_mtu_discovery && 3466 !in6_localaddr(&inp->in6p_faddr)) 3467 mss = min(mss, V_tcp_v6mssdflt); 3468 } 3469 #endif 3470 #if defined(INET) && defined(INET6) 3471 else 3472 #endif 3473 #ifdef INET 3474 { 3475 mss = maxmtu - min_protoh; 3476 if (!V_path_mtu_discovery && 3477 !in_localaddr(inp->inp_faddr)) 3478 mss = min(mss, V_tcp_mssdflt); 3479 } 3480 #endif 3481 /* 3482 * XXX - The above conditional (mss = maxmtu - min_protoh) 3483 * probably violates the TCP spec. 3484 * The problem is that, since we don't know the 3485 * other end's MSS, we are supposed to use a conservative 3486 * default. But, if we do that, then MTU discovery will 3487 * never actually take place, because the conservative 3488 * default is much less than the MTUs typically seen 3489 * on the Internet today. For the moment, we'll sweep 3490 * this under the carpet. 3491 * 3492 * The conservative default might not actually be a problem 3493 * if the only case this occurs is when sending an initial 3494 * SYN with options and data to a host we've never talked 3495 * to before. Then, they will reply with an MSS value which 3496 * will get recorded and the new parameters should get 3497 * recomputed. For Further Study. 3498 */ 3499 } 3500 mss = min(mss, offer); 3501 3502 /* 3503 * Sanity check: make sure that maxopd will be large 3504 * enough to allow some data on segments even if the 3505 * all the option space is used (40bytes). Otherwise 3506 * funny things may happen in tcp_output. 3507 */ 3508 mss = max(mss, 64); 3509 3510 /* 3511 * maxopd stores the maximum length of data AND options 3512 * in a segment; maxseg is the amount of data in a normal 3513 * segment. We need to store this value (maxopd) apart 3514 * from maxseg, because now every segment carries options 3515 * and thus we normally have somewhat less data in segments. 3516 */ 3517 tp->t_maxopd = mss; 3518 3519 /* 3520 * origoffer==-1 indicates that no segments were received yet. 3521 * In this case we just guess. 3522 */ 3523 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3524 (origoffer == -1 || 3525 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3526 mss -= TCPOLEN_TSTAMP_APPA; 3527 3528 tp->t_maxseg = mss; 3529 } 3530 3531 void 3532 tcp_mss(struct tcpcb *tp, int offer) 3533 { 3534 int mss; 3535 u_long bufsize; 3536 struct inpcb *inp; 3537 struct socket *so; 3538 struct hc_metrics_lite metrics; 3539 struct tcp_ifcap cap; 3540 3541 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3542 3543 bzero(&cap, sizeof(cap)); 3544 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3545 3546 mss = tp->t_maxseg; 3547 inp = tp->t_inpcb; 3548 3549 /* 3550 * If there's a pipesize, change the socket buffer to that size, 3551 * don't change if sb_hiwat is different than default (then it 3552 * has been changed on purpose with setsockopt). 3553 * Make the socket buffers an integral number of mss units; 3554 * if the mss is larger than the socket buffer, decrease the mss. 3555 */ 3556 so = inp->inp_socket; 3557 SOCKBUF_LOCK(&so->so_snd); 3558 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3559 bufsize = metrics.rmx_sendpipe; 3560 else 3561 bufsize = so->so_snd.sb_hiwat; 3562 if (bufsize < mss) 3563 mss = bufsize; 3564 else { 3565 bufsize = roundup(bufsize, mss); 3566 if (bufsize > sb_max) 3567 bufsize = sb_max; 3568 if (bufsize > so->so_snd.sb_hiwat) 3569 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3570 } 3571 SOCKBUF_UNLOCK(&so->so_snd); 3572 tp->t_maxseg = mss; 3573 3574 SOCKBUF_LOCK(&so->so_rcv); 3575 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3576 bufsize = metrics.rmx_recvpipe; 3577 else 3578 bufsize = so->so_rcv.sb_hiwat; 3579 if (bufsize > mss) { 3580 bufsize = roundup(bufsize, mss); 3581 if (bufsize > sb_max) 3582 bufsize = sb_max; 3583 if (bufsize > so->so_rcv.sb_hiwat) 3584 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3585 } 3586 SOCKBUF_UNLOCK(&so->so_rcv); 3587 3588 /* Check the interface for TSO capabilities. */ 3589 if (cap.ifcap & CSUM_TSO) { 3590 tp->t_flags |= TF_TSO; 3591 tp->t_tsomax = cap.tsomax; 3592 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3593 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3594 } 3595 } 3596 3597 /* 3598 * Determine the MSS option to send on an outgoing SYN. 3599 */ 3600 int 3601 tcp_mssopt(struct in_conninfo *inc) 3602 { 3603 int mss = 0; 3604 u_long maxmtu = 0; 3605 u_long thcmtu = 0; 3606 size_t min_protoh; 3607 3608 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3609 3610 #ifdef INET6 3611 if (inc->inc_flags & INC_ISIPV6) { 3612 mss = V_tcp_v6mssdflt; 3613 maxmtu = tcp_maxmtu6(inc, NULL); 3614 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3615 } 3616 #endif 3617 #if defined(INET) && defined(INET6) 3618 else 3619 #endif 3620 #ifdef INET 3621 { 3622 mss = V_tcp_mssdflt; 3623 maxmtu = tcp_maxmtu(inc, NULL); 3624 min_protoh = sizeof(struct tcpiphdr); 3625 } 3626 #endif 3627 #if defined(INET6) || defined(INET) 3628 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3629 #endif 3630 3631 if (maxmtu && thcmtu) 3632 mss = min(maxmtu, thcmtu) - min_protoh; 3633 else if (maxmtu || thcmtu) 3634 mss = max(maxmtu, thcmtu) - min_protoh; 3635 3636 return (mss); 3637 } 3638 3639 3640 /* 3641 * On a partial ack arrives, force the retransmission of the 3642 * next unacknowledged segment. Do not clear tp->t_dupacks. 3643 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3644 * be started again. 3645 */ 3646 static void 3647 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3648 { 3649 tcp_seq onxt = tp->snd_nxt; 3650 u_long ocwnd = tp->snd_cwnd; 3651 3652 INP_WLOCK_ASSERT(tp->t_inpcb); 3653 3654 tcp_timer_activate(tp, TT_REXMT, 0); 3655 tp->t_rtttime = 0; 3656 tp->snd_nxt = th->th_ack; 3657 /* 3658 * Set snd_cwnd to one segment beyond acknowledged offset. 3659 * (tp->snd_una has not yet been updated when this function is called.) 3660 */ 3661 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3662 tp->t_flags |= TF_ACKNOW; 3663 (void) tcp_output(tp); 3664 tp->snd_cwnd = ocwnd; 3665 if (SEQ_GT(onxt, tp->snd_nxt)) 3666 tp->snd_nxt = onxt; 3667 /* 3668 * Partial window deflation. Relies on fact that tp->snd_una 3669 * not updated yet. 3670 */ 3671 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3672 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3673 else 3674 tp->snd_cwnd = 0; 3675 tp->snd_cwnd += tp->t_maxseg; 3676 } 3677