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_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, 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_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, 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_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, 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_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, 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_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, 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_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, 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_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, 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_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, 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_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 181 &VNET_NAME(tcp_do_ecn), 0, 182 "TCP ECN support"); 183 184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 185 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 186 &VNET_NAME(tcp_ecn_maxretries), 0, 187 "Max retries before giving up on ECN"); 188 189 VNET_DEFINE(int, tcp_insecure_rst) = 0; 190 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 191 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 192 &VNET_NAME(tcp_insecure_rst), 0, 193 "Follow the old (insecure) criteria for accepting RST packets"); 194 195 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 196 #define V_tcp_recvspace VNET(tcp_recvspace) 197 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 198 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 199 200 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 201 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 202 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 203 &VNET_NAME(tcp_do_autorcvbuf), 0, 204 "Enable automatic receive buffer sizing"); 205 206 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 207 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 208 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 209 &VNET_NAME(tcp_autorcvbuf_inc), 0, 210 "Incrementor step size of automatic receive buffer"); 211 212 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 213 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 214 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 215 &VNET_NAME(tcp_autorcvbuf_max), 0, 216 "Max size of automatic receive buffer"); 217 218 VNET_DEFINE(struct inpcbhead, tcb); 219 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 220 VNET_DEFINE(struct inpcbinfo, tcbinfo); 221 222 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 223 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 224 struct socket *, struct tcpcb *, int, int, uint8_t, 225 int); 226 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 227 struct tcpcb *, int, int); 228 static void tcp_pulloutofband(struct socket *, 229 struct tcphdr *, struct mbuf *, int); 230 static void tcp_xmit_timer(struct tcpcb *, int); 231 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 232 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 233 uint16_t type); 234 static void inline cc_conn_init(struct tcpcb *tp); 235 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 236 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 237 struct tcphdr *th, struct tcpopt *to); 238 239 /* 240 * TCP statistics are stored in an "array" of counter(9)s. 241 */ 242 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 243 VNET_PCPUSTAT_SYSINIT(tcpstat); 244 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 245 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 246 247 #ifdef VIMAGE 248 VNET_PCPUSTAT_SYSUNINIT(tcpstat); 249 #endif /* VIMAGE */ 250 /* 251 * Kernel module interface for updating tcpstat. The argument is an index 252 * into tcpstat treated as an array. 253 */ 254 void 255 kmod_tcpstat_inc(int statnum) 256 { 257 258 counter_u64_add(VNET(tcpstat)[statnum], 1); 259 } 260 261 /* 262 * Wrapper for the TCP established input helper hook. 263 */ 264 static void inline 265 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 266 { 267 struct tcp_hhook_data hhook_data; 268 269 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 270 hhook_data.tp = tp; 271 hhook_data.th = th; 272 hhook_data.to = to; 273 274 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 275 tp->osd); 276 } 277 } 278 279 /* 280 * CC wrapper hook functions 281 */ 282 static void inline 283 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 284 { 285 INP_WLOCK_ASSERT(tp->t_inpcb); 286 287 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 288 if (tp->snd_cwnd <= tp->snd_wnd) 289 tp->ccv->flags |= CCF_CWND_LIMITED; 290 else 291 tp->ccv->flags &= ~CCF_CWND_LIMITED; 292 293 if (type == CC_ACK) { 294 if (tp->snd_cwnd > tp->snd_ssthresh) { 295 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 296 V_tcp_abc_l_var * tp->t_maxseg); 297 if (tp->t_bytes_acked >= tp->snd_cwnd) { 298 tp->t_bytes_acked -= tp->snd_cwnd; 299 tp->ccv->flags |= CCF_ABC_SENTAWND; 300 } 301 } else { 302 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 303 tp->t_bytes_acked = 0; 304 } 305 } 306 307 if (CC_ALGO(tp)->ack_received != NULL) { 308 /* XXXLAS: Find a way to live without this */ 309 tp->ccv->curack = th->th_ack; 310 CC_ALGO(tp)->ack_received(tp->ccv, type); 311 } 312 } 313 314 static void inline 315 cc_conn_init(struct tcpcb *tp) 316 { 317 struct hc_metrics_lite metrics; 318 struct inpcb *inp = tp->t_inpcb; 319 int rtt; 320 321 INP_WLOCK_ASSERT(tp->t_inpcb); 322 323 tcp_hc_get(&inp->inp_inc, &metrics); 324 325 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 326 tp->t_srtt = rtt; 327 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 328 TCPSTAT_INC(tcps_usedrtt); 329 if (metrics.rmx_rttvar) { 330 tp->t_rttvar = metrics.rmx_rttvar; 331 TCPSTAT_INC(tcps_usedrttvar); 332 } else { 333 /* default variation is +- 1 rtt */ 334 tp->t_rttvar = 335 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 336 } 337 TCPT_RANGESET(tp->t_rxtcur, 338 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 339 tp->t_rttmin, TCPTV_REXMTMAX); 340 } 341 if (metrics.rmx_ssthresh) { 342 /* 343 * There's some sort of gateway or interface 344 * buffer limit on the path. Use this to set 345 * the slow start threshhold, but set the 346 * threshold to no less than 2*mss. 347 */ 348 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 349 TCPSTAT_INC(tcps_usedssthresh); 350 } 351 352 /* 353 * Set the initial slow-start flight size. 354 * 355 * RFC5681 Section 3.1 specifies the default conservative values. 356 * RFC3390 specifies slightly more aggressive values. 357 * RFC6928 increases it to ten segments. 358 * 359 * If a SYN or SYN/ACK was lost and retransmitted, we have to 360 * reduce the initial CWND to one segment as congestion is likely 361 * requiring us to be cautious. 362 */ 363 if (tp->snd_cwnd == 1) 364 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 365 else if (V_tcp_do_initcwnd10) 366 tp->snd_cwnd = min(10 * tp->t_maxseg, 367 max(2 * tp->t_maxseg, 14600)); 368 else if (V_tcp_do_rfc3390) 369 tp->snd_cwnd = min(4 * tp->t_maxseg, 370 max(2 * tp->t_maxseg, 4380)); 371 else { 372 /* Per RFC5681 Section 3.1 */ 373 if (tp->t_maxseg > 2190) 374 tp->snd_cwnd = 2 * tp->t_maxseg; 375 else if (tp->t_maxseg > 1095) 376 tp->snd_cwnd = 3 * tp->t_maxseg; 377 else 378 tp->snd_cwnd = 4 * tp->t_maxseg; 379 } 380 381 if (CC_ALGO(tp)->conn_init != NULL) 382 CC_ALGO(tp)->conn_init(tp->ccv); 383 } 384 385 void inline 386 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 387 { 388 INP_WLOCK_ASSERT(tp->t_inpcb); 389 390 switch(type) { 391 case CC_NDUPACK: 392 if (!IN_FASTRECOVERY(tp->t_flags)) { 393 tp->snd_recover = tp->snd_max; 394 if (tp->t_flags & TF_ECN_PERMIT) 395 tp->t_flags |= TF_ECN_SND_CWR; 396 } 397 break; 398 case CC_ECN: 399 if (!IN_CONGRECOVERY(tp->t_flags)) { 400 TCPSTAT_INC(tcps_ecn_rcwnd); 401 tp->snd_recover = tp->snd_max; 402 if (tp->t_flags & TF_ECN_PERMIT) 403 tp->t_flags |= TF_ECN_SND_CWR; 404 } 405 break; 406 case CC_RTO: 407 tp->t_dupacks = 0; 408 tp->t_bytes_acked = 0; 409 EXIT_RECOVERY(tp->t_flags); 410 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 411 tp->t_maxseg) * tp->t_maxseg; 412 tp->snd_cwnd = tp->t_maxseg; 413 break; 414 case CC_RTO_ERR: 415 TCPSTAT_INC(tcps_sndrexmitbad); 416 /* RTO was unnecessary, so reset everything. */ 417 tp->snd_cwnd = tp->snd_cwnd_prev; 418 tp->snd_ssthresh = tp->snd_ssthresh_prev; 419 tp->snd_recover = tp->snd_recover_prev; 420 if (tp->t_flags & TF_WASFRECOVERY) 421 ENTER_FASTRECOVERY(tp->t_flags); 422 if (tp->t_flags & TF_WASCRECOVERY) 423 ENTER_CONGRECOVERY(tp->t_flags); 424 tp->snd_nxt = tp->snd_max; 425 tp->t_flags &= ~TF_PREVVALID; 426 tp->t_badrxtwin = 0; 427 break; 428 } 429 430 if (CC_ALGO(tp)->cong_signal != NULL) { 431 if (th != NULL) 432 tp->ccv->curack = th->th_ack; 433 CC_ALGO(tp)->cong_signal(tp->ccv, type); 434 } 435 } 436 437 static void inline 438 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 439 { 440 INP_WLOCK_ASSERT(tp->t_inpcb); 441 442 /* XXXLAS: KASSERT that we're in recovery? */ 443 444 if (CC_ALGO(tp)->post_recovery != NULL) { 445 tp->ccv->curack = th->th_ack; 446 CC_ALGO(tp)->post_recovery(tp->ccv); 447 } 448 /* XXXLAS: EXIT_RECOVERY ? */ 449 tp->t_bytes_acked = 0; 450 } 451 452 #ifdef TCP_SIGNATURE 453 static inline int 454 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 455 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 456 { 457 int ret; 458 459 tcp_fields_to_net(th); 460 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 461 tcp_fields_to_host(th); 462 return (ret); 463 } 464 #endif 465 466 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 467 #ifdef INET6 468 #define ND6_HINT(tp) \ 469 do { \ 470 if ((tp) && (tp)->t_inpcb && \ 471 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 472 nd6_nud_hint(NULL, NULL, 0); \ 473 } while (0) 474 #else 475 #define ND6_HINT(tp) 476 #endif 477 478 /* 479 * Indicate whether this ack should be delayed. We can delay the ack if 480 * following conditions are met: 481 * - There is no delayed ack timer in progress. 482 * - Our last ack wasn't a 0-sized window. We never want to delay 483 * the ack that opens up a 0-sized window. 484 * - LRO wasn't used for this segment. We make sure by checking that the 485 * segment size is not larger than the MSS. 486 * - Delayed acks are enabled or this is a half-synchronized T/TCP 487 * connection. 488 */ 489 #define DELAY_ACK(tp, tlen) \ 490 ((!tcp_timer_active(tp, TT_DELACK) && \ 491 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 492 (tlen <= tp->t_maxopd) && \ 493 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 494 495 /* 496 * TCP input handling is split into multiple parts: 497 * tcp6_input is a thin wrapper around tcp_input for the extended 498 * ip6_protox[] call format in ip6_input 499 * tcp_input handles primary segment validation, inpcb lookup and 500 * SYN processing on listen sockets 501 * tcp_do_segment processes the ACK and text of the segment for 502 * establishing, established and closing connections 503 */ 504 #ifdef INET6 505 int 506 tcp6_input(struct mbuf **mp, int *offp, int proto) 507 { 508 struct mbuf *m = *mp; 509 struct in6_ifaddr *ia6; 510 511 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 512 513 /* 514 * draft-itojun-ipv6-tcp-to-anycast 515 * better place to put this in? 516 */ 517 ia6 = ip6_getdstifaddr(m); 518 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 519 struct ip6_hdr *ip6; 520 521 ifa_free(&ia6->ia_ifa); 522 ip6 = mtod(m, struct ip6_hdr *); 523 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 524 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 525 return IPPROTO_DONE; 526 } 527 if (ia6) 528 ifa_free(&ia6->ia_ifa); 529 530 tcp_input(m, *offp); 531 return IPPROTO_DONE; 532 } 533 #endif /* INET6 */ 534 535 void 536 tcp_input(struct mbuf *m, int off0) 537 { 538 struct tcphdr *th = NULL; 539 struct ip *ip = NULL; 540 struct inpcb *inp = NULL; 541 struct tcpcb *tp = NULL; 542 struct socket *so = NULL; 543 u_char *optp = NULL; 544 int optlen = 0; 545 #ifdef INET 546 int len; 547 #endif 548 int tlen = 0, off; 549 int drop_hdrlen; 550 int thflags; 551 int rstreason = 0; /* For badport_bandlim accounting purposes */ 552 #ifdef TCP_SIGNATURE 553 uint8_t sig_checked = 0; 554 #endif 555 uint8_t iptos = 0; 556 struct m_tag *fwd_tag = NULL; 557 #ifdef INET6 558 struct ip6_hdr *ip6 = NULL; 559 int isipv6; 560 #else 561 const void *ip6 = NULL; 562 #endif /* INET6 */ 563 struct tcpopt to; /* options in this segment */ 564 char *s = NULL; /* address and port logging */ 565 int ti_locked; 566 #define TI_UNLOCKED 1 567 #define TI_WLOCKED 2 568 569 #ifdef TCPDEBUG 570 /* 571 * The size of tcp_saveipgen must be the size of the max ip header, 572 * now IPv6. 573 */ 574 u_char tcp_saveipgen[IP6_HDR_LEN]; 575 struct tcphdr tcp_savetcp; 576 short ostate = 0; 577 #endif 578 579 #ifdef INET6 580 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 581 #endif 582 583 to.to_flags = 0; 584 TCPSTAT_INC(tcps_rcvtotal); 585 586 #ifdef INET6 587 if (isipv6) { 588 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 589 590 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 591 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 592 if (m == NULL) { 593 TCPSTAT_INC(tcps_rcvshort); 594 return; 595 } 596 } 597 598 ip6 = mtod(m, struct ip6_hdr *); 599 th = (struct tcphdr *)((caddr_t)ip6 + off0); 600 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 601 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 602 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 603 th->th_sum = m->m_pkthdr.csum_data; 604 else 605 th->th_sum = in6_cksum_pseudo(ip6, tlen, 606 IPPROTO_TCP, m->m_pkthdr.csum_data); 607 th->th_sum ^= 0xffff; 608 } else 609 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 610 if (th->th_sum) { 611 TCPSTAT_INC(tcps_rcvbadsum); 612 goto drop; 613 } 614 615 /* 616 * Be proactive about unspecified IPv6 address in source. 617 * As we use all-zero to indicate unbounded/unconnected pcb, 618 * unspecified IPv6 address can be used to confuse us. 619 * 620 * Note that packets with unspecified IPv6 destination is 621 * already dropped in ip6_input. 622 */ 623 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 624 /* XXX stat */ 625 goto drop; 626 } 627 } 628 #endif 629 #if defined(INET) && defined(INET6) 630 else 631 #endif 632 #ifdef INET 633 { 634 /* 635 * Get IP and TCP header together in first mbuf. 636 * Note: IP leaves IP header in first mbuf. 637 */ 638 if (off0 > sizeof (struct ip)) { 639 ip_stripoptions(m); 640 off0 = sizeof(struct ip); 641 } 642 if (m->m_len < sizeof (struct tcpiphdr)) { 643 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 644 == NULL) { 645 TCPSTAT_INC(tcps_rcvshort); 646 return; 647 } 648 } 649 ip = mtod(m, struct ip *); 650 th = (struct tcphdr *)((caddr_t)ip + off0); 651 tlen = ntohs(ip->ip_len) - off0; 652 653 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 654 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 655 th->th_sum = m->m_pkthdr.csum_data; 656 else 657 th->th_sum = in_pseudo(ip->ip_src.s_addr, 658 ip->ip_dst.s_addr, 659 htonl(m->m_pkthdr.csum_data + tlen + 660 IPPROTO_TCP)); 661 th->th_sum ^= 0xffff; 662 } else { 663 struct ipovly *ipov = (struct ipovly *)ip; 664 665 /* 666 * Checksum extended TCP header and data. 667 */ 668 len = off0 + tlen; 669 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 670 ipov->ih_len = htons(tlen); 671 th->th_sum = in_cksum(m, len); 672 /* Reset length for SDT probes. */ 673 ip->ip_len = htons(tlen + off0); 674 } 675 676 if (th->th_sum) { 677 TCPSTAT_INC(tcps_rcvbadsum); 678 goto drop; 679 } 680 /* Re-initialization for later version check */ 681 ip->ip_v = IPVERSION; 682 } 683 #endif /* INET */ 684 685 #ifdef INET6 686 if (isipv6) 687 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 688 #endif 689 #if defined(INET) && defined(INET6) 690 else 691 #endif 692 #ifdef INET 693 iptos = ip->ip_tos; 694 #endif 695 696 /* 697 * Check that TCP offset makes sense, 698 * pull out TCP options and adjust length. XXX 699 */ 700 off = th->th_off << 2; 701 if (off < sizeof (struct tcphdr) || off > tlen) { 702 TCPSTAT_INC(tcps_rcvbadoff); 703 goto drop; 704 } 705 tlen -= off; /* tlen is used instead of ti->ti_len */ 706 if (off > sizeof (struct tcphdr)) { 707 #ifdef INET6 708 if (isipv6) { 709 IP6_EXTHDR_CHECK(m, off0, off, ); 710 ip6 = mtod(m, struct ip6_hdr *); 711 th = (struct tcphdr *)((caddr_t)ip6 + off0); 712 } 713 #endif 714 #if defined(INET) && defined(INET6) 715 else 716 #endif 717 #ifdef INET 718 { 719 if (m->m_len < sizeof(struct ip) + off) { 720 if ((m = m_pullup(m, sizeof (struct ip) + off)) 721 == NULL) { 722 TCPSTAT_INC(tcps_rcvshort); 723 return; 724 } 725 ip = mtod(m, struct ip *); 726 th = (struct tcphdr *)((caddr_t)ip + off0); 727 } 728 } 729 #endif 730 optlen = off - sizeof (struct tcphdr); 731 optp = (u_char *)(th + 1); 732 } 733 thflags = th->th_flags; 734 735 /* 736 * Convert TCP protocol specific fields to host format. 737 */ 738 tcp_fields_to_host(th); 739 740 /* 741 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 742 */ 743 drop_hdrlen = off0 + off; 744 745 /* 746 * Locate pcb for segment; if we're likely to add or remove a 747 * connection then first acquire pcbinfo lock. There are two cases 748 * where we might discover later we need a write lock despite the 749 * flags: ACKs moving a connection out of the syncache, and ACKs for 750 * a connection in TIMEWAIT. 751 */ 752 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 753 INP_INFO_WLOCK(&V_tcbinfo); 754 ti_locked = TI_WLOCKED; 755 } else 756 ti_locked = TI_UNLOCKED; 757 758 /* 759 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 760 */ 761 if ( 762 #ifdef INET6 763 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 764 #ifdef INET 765 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 766 #endif 767 #endif 768 #if defined(INET) && !defined(INET6) 769 (m->m_flags & M_IP_NEXTHOP) 770 #endif 771 ) 772 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 773 774 findpcb: 775 #ifdef INVARIANTS 776 if (ti_locked == TI_WLOCKED) { 777 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 778 } else { 779 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 780 } 781 #endif 782 #ifdef INET6 783 if (isipv6 && fwd_tag != NULL) { 784 struct sockaddr_in6 *next_hop6; 785 786 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 787 /* 788 * Transparently forwarded. Pretend to be the destination. 789 * Already got one like this? 790 */ 791 inp = in6_pcblookup_mbuf(&V_tcbinfo, 792 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 793 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 794 if (!inp) { 795 /* 796 * It's new. Try to find the ambushing socket. 797 * Because we've rewritten the destination address, 798 * any hardware-generated hash is ignored. 799 */ 800 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 801 th->th_sport, &next_hop6->sin6_addr, 802 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 803 th->th_dport, INPLOOKUP_WILDCARD | 804 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 805 } 806 } else if (isipv6) { 807 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 808 th->th_sport, &ip6->ip6_dst, th->th_dport, 809 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 810 m->m_pkthdr.rcvif, m); 811 } 812 #endif /* INET6 */ 813 #if defined(INET6) && defined(INET) 814 else 815 #endif 816 #ifdef INET 817 if (fwd_tag != NULL) { 818 struct sockaddr_in *next_hop; 819 820 next_hop = (struct sockaddr_in *)(fwd_tag+1); 821 /* 822 * Transparently forwarded. Pretend to be the destination. 823 * already got one like this? 824 */ 825 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 826 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 827 m->m_pkthdr.rcvif, m); 828 if (!inp) { 829 /* 830 * It's new. Try to find the ambushing socket. 831 * Because we've rewritten the destination address, 832 * any hardware-generated hash is ignored. 833 */ 834 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 835 th->th_sport, next_hop->sin_addr, 836 next_hop->sin_port ? ntohs(next_hop->sin_port) : 837 th->th_dport, INPLOOKUP_WILDCARD | 838 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 839 } 840 } else 841 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 842 th->th_sport, ip->ip_dst, th->th_dport, 843 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 844 m->m_pkthdr.rcvif, m); 845 #endif /* INET */ 846 847 /* 848 * If the INPCB does not exist then all data in the incoming 849 * segment is discarded and an appropriate RST is sent back. 850 * XXX MRT Send RST using which routing table? 851 */ 852 if (inp == NULL) { 853 /* 854 * Log communication attempts to ports that are not 855 * in use. 856 */ 857 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 858 tcp_log_in_vain == 2) { 859 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 860 log(LOG_INFO, "%s; %s: Connection attempt " 861 "to closed port\n", s, __func__); 862 } 863 /* 864 * When blackholing do not respond with a RST but 865 * completely ignore the segment and drop it. 866 */ 867 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 868 V_blackhole == 2) 869 goto dropunlock; 870 871 rstreason = BANDLIM_RST_CLOSEDPORT; 872 goto dropwithreset; 873 } 874 INP_WLOCK_ASSERT(inp); 875 if (!(inp->inp_flags & INP_HW_FLOWID) 876 && (m->m_flags & M_FLOWID) 877 && ((inp->inp_socket == NULL) 878 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 879 inp->inp_flags |= INP_HW_FLOWID; 880 inp->inp_flags &= ~INP_SW_FLOWID; 881 inp->inp_flowid = m->m_pkthdr.flowid; 882 inp->inp_flowtype = M_HASHTYPE_GET(m); 883 } 884 #ifdef IPSEC 885 #ifdef INET6 886 if (isipv6 && ipsec6_in_reject(m, inp)) { 887 IPSEC6STAT_INC(ips_in_polvio); 888 goto dropunlock; 889 } else 890 #endif /* INET6 */ 891 if (ipsec4_in_reject(m, inp) != 0) { 892 IPSECSTAT_INC(ips_in_polvio); 893 goto dropunlock; 894 } 895 #endif /* IPSEC */ 896 897 /* 898 * Check the minimum TTL for socket. 899 */ 900 if (inp->inp_ip_minttl != 0) { 901 #ifdef INET6 902 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 903 goto dropunlock; 904 else 905 #endif 906 if (inp->inp_ip_minttl > ip->ip_ttl) 907 goto dropunlock; 908 } 909 910 /* 911 * A previous connection in TIMEWAIT state is supposed to catch stray 912 * or duplicate segments arriving late. If this segment was a 913 * legitimate new connection attempt, the old INPCB gets removed and 914 * we can try again to find a listening socket. 915 * 916 * At this point, due to earlier optimism, we may hold only an inpcb 917 * lock, and not the inpcbinfo write lock. If so, we need to try to 918 * acquire it, or if that fails, acquire a reference on the inpcb, 919 * drop all locks, acquire a global write lock, and then re-acquire 920 * the inpcb lock. We may at that point discover that another thread 921 * has tried to free the inpcb, in which case we need to loop back 922 * and try to find a new inpcb to deliver to. 923 * 924 * XXXRW: It may be time to rethink timewait locking. 925 */ 926 relocked: 927 if (inp->inp_flags & INP_TIMEWAIT) { 928 if (ti_locked == TI_UNLOCKED) { 929 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 930 in_pcbref(inp); 931 INP_WUNLOCK(inp); 932 INP_INFO_WLOCK(&V_tcbinfo); 933 ti_locked = TI_WLOCKED; 934 INP_WLOCK(inp); 935 if (in_pcbrele_wlocked(inp)) { 936 inp = NULL; 937 goto findpcb; 938 } 939 } else 940 ti_locked = TI_WLOCKED; 941 } 942 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 943 944 if (thflags & TH_SYN) 945 tcp_dooptions(&to, optp, optlen, TO_SYN); 946 /* 947 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 948 */ 949 if (tcp_twcheck(inp, &to, th, m, tlen)) 950 goto findpcb; 951 INP_INFO_WUNLOCK(&V_tcbinfo); 952 return; 953 } 954 /* 955 * The TCPCB may no longer exist if the connection is winding 956 * down or it is in the CLOSED state. Either way we drop the 957 * segment and send an appropriate response. 958 */ 959 tp = intotcpcb(inp); 960 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 961 rstreason = BANDLIM_RST_CLOSEDPORT; 962 goto dropwithreset; 963 } 964 965 #ifdef TCP_OFFLOAD 966 if (tp->t_flags & TF_TOE) { 967 tcp_offload_input(tp, m); 968 m = NULL; /* consumed by the TOE driver */ 969 goto dropunlock; 970 } 971 #endif 972 973 /* 974 * We've identified a valid inpcb, but it could be that we need an 975 * inpcbinfo write lock but don't hold it. In this case, attempt to 976 * acquire using the same strategy as the TIMEWAIT case above. If we 977 * relock, we have to jump back to 'relocked' as the connection might 978 * now be in TIMEWAIT. 979 */ 980 #ifdef INVARIANTS 981 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 982 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 983 #endif 984 if (tp->t_state != TCPS_ESTABLISHED) { 985 if (ti_locked == TI_UNLOCKED) { 986 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 987 in_pcbref(inp); 988 INP_WUNLOCK(inp); 989 INP_INFO_WLOCK(&V_tcbinfo); 990 ti_locked = TI_WLOCKED; 991 INP_WLOCK(inp); 992 if (in_pcbrele_wlocked(inp)) { 993 inp = NULL; 994 goto findpcb; 995 } 996 goto relocked; 997 } else 998 ti_locked = TI_WLOCKED; 999 } 1000 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1001 } 1002 1003 #ifdef MAC 1004 INP_WLOCK_ASSERT(inp); 1005 if (mac_inpcb_check_deliver(inp, m)) 1006 goto dropunlock; 1007 #endif 1008 so = inp->inp_socket; 1009 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1010 #ifdef TCPDEBUG 1011 if (so->so_options & SO_DEBUG) { 1012 ostate = tp->t_state; 1013 #ifdef INET6 1014 if (isipv6) { 1015 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1016 } else 1017 #endif 1018 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1019 tcp_savetcp = *th; 1020 } 1021 #endif /* TCPDEBUG */ 1022 /* 1023 * When the socket is accepting connections (the INPCB is in LISTEN 1024 * state) we look into the SYN cache if this is a new connection 1025 * attempt or the completion of a previous one. Because listen 1026 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1027 * held in this case. 1028 */ 1029 if (so->so_options & SO_ACCEPTCONN) { 1030 struct in_conninfo inc; 1031 1032 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1033 "tp not listening", __func__)); 1034 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1035 1036 bzero(&inc, sizeof(inc)); 1037 #ifdef INET6 1038 if (isipv6) { 1039 inc.inc_flags |= INC_ISIPV6; 1040 inc.inc6_faddr = ip6->ip6_src; 1041 inc.inc6_laddr = ip6->ip6_dst; 1042 } else 1043 #endif 1044 { 1045 inc.inc_faddr = ip->ip_src; 1046 inc.inc_laddr = ip->ip_dst; 1047 } 1048 inc.inc_fport = th->th_sport; 1049 inc.inc_lport = th->th_dport; 1050 inc.inc_fibnum = so->so_fibnum; 1051 1052 /* 1053 * Check for an existing connection attempt in syncache if 1054 * the flag is only ACK. A successful lookup creates a new 1055 * socket appended to the listen queue in SYN_RECEIVED state. 1056 */ 1057 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1058 /* 1059 * Parse the TCP options here because 1060 * syncookies need access to the reflected 1061 * timestamp. 1062 */ 1063 tcp_dooptions(&to, optp, optlen, 0); 1064 /* 1065 * NB: syncache_expand() doesn't unlock 1066 * inp and tcpinfo locks. 1067 */ 1068 if (!syncache_expand(&inc, &to, th, &so, m)) { 1069 /* 1070 * No syncache entry or ACK was not 1071 * for our SYN/ACK. Send a RST. 1072 * NB: syncache did its own logging 1073 * of the failure cause. 1074 */ 1075 rstreason = BANDLIM_RST_OPENPORT; 1076 goto dropwithreset; 1077 } 1078 if (so == NULL) { 1079 /* 1080 * We completed the 3-way handshake 1081 * but could not allocate a socket 1082 * either due to memory shortage, 1083 * listen queue length limits or 1084 * global socket limits. Send RST 1085 * or wait and have the remote end 1086 * retransmit the ACK for another 1087 * try. 1088 */ 1089 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1090 log(LOG_DEBUG, "%s; %s: Listen socket: " 1091 "Socket allocation failed due to " 1092 "limits or memory shortage, %s\n", 1093 s, __func__, 1094 V_tcp_sc_rst_sock_fail ? 1095 "sending RST" : "try again"); 1096 if (V_tcp_sc_rst_sock_fail) { 1097 rstreason = BANDLIM_UNLIMITED; 1098 goto dropwithreset; 1099 } else 1100 goto dropunlock; 1101 } 1102 /* 1103 * Socket is created in state SYN_RECEIVED. 1104 * Unlock the listen socket, lock the newly 1105 * created socket and update the tp variable. 1106 */ 1107 INP_WUNLOCK(inp); /* listen socket */ 1108 inp = sotoinpcb(so); 1109 INP_WLOCK(inp); /* new connection */ 1110 tp = intotcpcb(inp); 1111 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1112 ("%s: ", __func__)); 1113 #ifdef TCP_SIGNATURE 1114 if (sig_checked == 0) { 1115 tcp_dooptions(&to, optp, optlen, 1116 (thflags & TH_SYN) ? TO_SYN : 0); 1117 if (!tcp_signature_verify_input(m, off0, tlen, 1118 optlen, &to, th, tp->t_flags)) { 1119 1120 /* 1121 * In SYN_SENT state if it receives an 1122 * RST, it is allowed for further 1123 * processing. 1124 */ 1125 if ((thflags & TH_RST) == 0 || 1126 (tp->t_state == TCPS_SYN_SENT) == 0) 1127 goto dropunlock; 1128 } 1129 sig_checked = 1; 1130 } 1131 #endif 1132 1133 /* 1134 * Process the segment and the data it 1135 * contains. tcp_do_segment() consumes 1136 * the mbuf chain and unlocks the inpcb. 1137 */ 1138 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1139 iptos, ti_locked); 1140 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1141 return; 1142 } 1143 /* 1144 * Segment flag validation for new connection attempts: 1145 * 1146 * Our (SYN|ACK) response was rejected. 1147 * Check with syncache and remove entry to prevent 1148 * retransmits. 1149 * 1150 * NB: syncache_chkrst does its own logging of failure 1151 * causes. 1152 */ 1153 if (thflags & TH_RST) { 1154 syncache_chkrst(&inc, th); 1155 goto dropunlock; 1156 } 1157 /* 1158 * We can't do anything without SYN. 1159 */ 1160 if ((thflags & TH_SYN) == 0) { 1161 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1162 log(LOG_DEBUG, "%s; %s: Listen socket: " 1163 "SYN is missing, segment ignored\n", 1164 s, __func__); 1165 TCPSTAT_INC(tcps_badsyn); 1166 goto dropunlock; 1167 } 1168 /* 1169 * (SYN|ACK) is bogus on a listen socket. 1170 */ 1171 if (thflags & TH_ACK) { 1172 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1173 log(LOG_DEBUG, "%s; %s: Listen socket: " 1174 "SYN|ACK invalid, segment rejected\n", 1175 s, __func__); 1176 syncache_badack(&inc); /* XXX: Not needed! */ 1177 TCPSTAT_INC(tcps_badsyn); 1178 rstreason = BANDLIM_RST_OPENPORT; 1179 goto dropwithreset; 1180 } 1181 /* 1182 * If the drop_synfin option is enabled, drop all 1183 * segments with both the SYN and FIN bits set. 1184 * This prevents e.g. nmap from identifying the 1185 * TCP/IP stack. 1186 * XXX: Poor reasoning. nmap has other methods 1187 * and is constantly refining its stack detection 1188 * strategies. 1189 * XXX: This is a violation of the TCP specification 1190 * and was used by RFC1644. 1191 */ 1192 if ((thflags & TH_FIN) && V_drop_synfin) { 1193 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1194 log(LOG_DEBUG, "%s; %s: Listen socket: " 1195 "SYN|FIN segment ignored (based on " 1196 "sysctl setting)\n", s, __func__); 1197 TCPSTAT_INC(tcps_badsyn); 1198 goto dropunlock; 1199 } 1200 /* 1201 * Segment's flags are (SYN) or (SYN|FIN). 1202 * 1203 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1204 * as they do not affect the state of the TCP FSM. 1205 * The data pointed to by TH_URG and th_urp is ignored. 1206 */ 1207 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1208 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1209 KASSERT(thflags & (TH_SYN), 1210 ("%s: Listen socket: TH_SYN not set", __func__)); 1211 #ifdef INET6 1212 /* 1213 * If deprecated address is forbidden, 1214 * we do not accept SYN to deprecated interface 1215 * address to prevent any new inbound connection from 1216 * getting established. 1217 * When we do not accept SYN, we send a TCP RST, 1218 * with deprecated source address (instead of dropping 1219 * it). We compromise it as it is much better for peer 1220 * to send a RST, and RST will be the final packet 1221 * for the exchange. 1222 * 1223 * If we do not forbid deprecated addresses, we accept 1224 * the SYN packet. RFC2462 does not suggest dropping 1225 * SYN in this case. 1226 * If we decipher RFC2462 5.5.4, it says like this: 1227 * 1. use of deprecated addr with existing 1228 * communication is okay - "SHOULD continue to be 1229 * used" 1230 * 2. use of it with new communication: 1231 * (2a) "SHOULD NOT be used if alternate address 1232 * with sufficient scope is available" 1233 * (2b) nothing mentioned otherwise. 1234 * Here we fall into (2b) case as we have no choice in 1235 * our source address selection - we must obey the peer. 1236 * 1237 * The wording in RFC2462 is confusing, and there are 1238 * multiple description text for deprecated address 1239 * handling - worse, they are not exactly the same. 1240 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1241 */ 1242 if (isipv6 && !V_ip6_use_deprecated) { 1243 struct in6_ifaddr *ia6; 1244 1245 ia6 = ip6_getdstifaddr(m); 1246 if (ia6 != NULL && 1247 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1248 ifa_free(&ia6->ia_ifa); 1249 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1250 log(LOG_DEBUG, "%s; %s: Listen socket: " 1251 "Connection attempt to deprecated " 1252 "IPv6 address rejected\n", 1253 s, __func__); 1254 rstreason = BANDLIM_RST_OPENPORT; 1255 goto dropwithreset; 1256 } 1257 if (ia6) 1258 ifa_free(&ia6->ia_ifa); 1259 } 1260 #endif /* INET6 */ 1261 /* 1262 * Basic sanity checks on incoming SYN requests: 1263 * Don't respond if the destination is a link layer 1264 * broadcast according to RFC1122 4.2.3.10, p. 104. 1265 * If it is from this socket it must be forged. 1266 * Don't respond if the source or destination is a 1267 * global or subnet broad- or multicast address. 1268 * Note that it is quite possible to receive unicast 1269 * link-layer packets with a broadcast IP address. Use 1270 * in_broadcast() to find them. 1271 */ 1272 if (m->m_flags & (M_BCAST|M_MCAST)) { 1273 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1274 log(LOG_DEBUG, "%s; %s: Listen socket: " 1275 "Connection attempt from broad- or multicast " 1276 "link layer address ignored\n", s, __func__); 1277 goto dropunlock; 1278 } 1279 #ifdef INET6 1280 if (isipv6) { 1281 if (th->th_dport == th->th_sport && 1282 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1283 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1284 log(LOG_DEBUG, "%s; %s: Listen socket: " 1285 "Connection attempt to/from self " 1286 "ignored\n", s, __func__); 1287 goto dropunlock; 1288 } 1289 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1290 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1291 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1292 log(LOG_DEBUG, "%s; %s: Listen socket: " 1293 "Connection attempt from/to multicast " 1294 "address ignored\n", s, __func__); 1295 goto dropunlock; 1296 } 1297 } 1298 #endif 1299 #if defined(INET) && defined(INET6) 1300 else 1301 #endif 1302 #ifdef INET 1303 { 1304 if (th->th_dport == th->th_sport && 1305 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1306 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1307 log(LOG_DEBUG, "%s; %s: Listen socket: " 1308 "Connection attempt from/to self " 1309 "ignored\n", s, __func__); 1310 goto dropunlock; 1311 } 1312 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1313 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1314 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1315 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1316 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1317 log(LOG_DEBUG, "%s; %s: Listen socket: " 1318 "Connection attempt from/to broad- " 1319 "or multicast address ignored\n", 1320 s, __func__); 1321 goto dropunlock; 1322 } 1323 } 1324 #endif 1325 /* 1326 * SYN appears to be valid. Create compressed TCP state 1327 * for syncache. 1328 */ 1329 #ifdef TCPDEBUG 1330 if (so->so_options & SO_DEBUG) 1331 tcp_trace(TA_INPUT, ostate, tp, 1332 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1333 #endif 1334 tcp_dooptions(&to, optp, optlen, TO_SYN); 1335 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1336 /* 1337 * Entry added to syncache and mbuf consumed. 1338 * Everything already unlocked by syncache_add(). 1339 */ 1340 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1341 return; 1342 } else if (tp->t_state == TCPS_LISTEN) { 1343 /* 1344 * When a listen socket is torn down the SO_ACCEPTCONN 1345 * flag is removed first while connections are drained 1346 * from the accept queue in a unlock/lock cycle of the 1347 * ACCEPT_LOCK, opening a race condition allowing a SYN 1348 * attempt go through unhandled. 1349 */ 1350 goto dropunlock; 1351 } 1352 1353 #ifdef TCP_SIGNATURE 1354 if (sig_checked == 0) { 1355 tcp_dooptions(&to, optp, optlen, 1356 (thflags & TH_SYN) ? TO_SYN : 0); 1357 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1358 th, tp->t_flags)) { 1359 1360 /* 1361 * In SYN_SENT state if it receives an RST, it is 1362 * allowed for further processing. 1363 */ 1364 if ((thflags & TH_RST) == 0 || 1365 (tp->t_state == TCPS_SYN_SENT) == 0) 1366 goto dropunlock; 1367 } 1368 sig_checked = 1; 1369 } 1370 #endif 1371 1372 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1373 1374 /* 1375 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1376 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1377 * the inpcb, and unlocks pcbinfo. 1378 */ 1379 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1380 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1381 return; 1382 1383 dropwithreset: 1384 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1385 1386 if (ti_locked == TI_WLOCKED) { 1387 INP_INFO_WUNLOCK(&V_tcbinfo); 1388 ti_locked = TI_UNLOCKED; 1389 } 1390 #ifdef INVARIANTS 1391 else { 1392 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1393 "ti_locked: %d", __func__, ti_locked)); 1394 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1395 } 1396 #endif 1397 1398 if (inp != NULL) { 1399 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1400 INP_WUNLOCK(inp); 1401 } else 1402 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1403 m = NULL; /* mbuf chain got consumed. */ 1404 goto drop; 1405 1406 dropunlock: 1407 if (m != NULL) 1408 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1409 1410 if (ti_locked == TI_WLOCKED) { 1411 INP_INFO_WUNLOCK(&V_tcbinfo); 1412 ti_locked = TI_UNLOCKED; 1413 } 1414 #ifdef INVARIANTS 1415 else { 1416 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1417 "ti_locked: %d", __func__, ti_locked)); 1418 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1419 } 1420 #endif 1421 1422 if (inp != NULL) 1423 INP_WUNLOCK(inp); 1424 1425 drop: 1426 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1427 if (s != NULL) 1428 free(s, M_TCPLOG); 1429 if (m != NULL) 1430 m_freem(m); 1431 } 1432 1433 static void 1434 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1435 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1436 int ti_locked) 1437 { 1438 int thflags, acked, ourfinisacked, needoutput = 0; 1439 int rstreason, todrop, win; 1440 u_long tiwin; 1441 char *s; 1442 struct in_conninfo *inc; 1443 struct mbuf *mfree; 1444 struct tcpopt to; 1445 1446 #ifdef TCPDEBUG 1447 /* 1448 * The size of tcp_saveipgen must be the size of the max ip header, 1449 * now IPv6. 1450 */ 1451 u_char tcp_saveipgen[IP6_HDR_LEN]; 1452 struct tcphdr tcp_savetcp; 1453 short ostate = 0; 1454 #endif 1455 thflags = th->th_flags; 1456 inc = &tp->t_inpcb->inp_inc; 1457 tp->sackhint.last_sack_ack = 0; 1458 1459 /* 1460 * If this is either a state-changing packet or current state isn't 1461 * established, we require a write lock on tcbinfo. Otherwise, we 1462 * allow the tcbinfo to be in either alocked or unlocked, as the 1463 * caller may have unnecessarily acquired a write lock due to a race. 1464 */ 1465 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1466 tp->t_state != TCPS_ESTABLISHED) { 1467 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1468 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1469 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1470 } else { 1471 #ifdef INVARIANTS 1472 if (ti_locked == TI_WLOCKED) 1473 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1474 else { 1475 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1476 "ti_locked: %d", __func__, ti_locked)); 1477 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1478 } 1479 #endif 1480 } 1481 INP_WLOCK_ASSERT(tp->t_inpcb); 1482 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1483 __func__)); 1484 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1485 __func__)); 1486 1487 /* 1488 * Segment received on connection. 1489 * Reset idle time and keep-alive timer. 1490 * XXX: This should be done after segment 1491 * validation to ignore broken/spoofed segs. 1492 */ 1493 tp->t_rcvtime = ticks; 1494 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1495 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1496 1497 /* 1498 * Unscale the window into a 32-bit value. 1499 * For the SYN_SENT state the scale is zero. 1500 */ 1501 tiwin = th->th_win << tp->snd_scale; 1502 1503 /* 1504 * TCP ECN processing. 1505 */ 1506 if (tp->t_flags & TF_ECN_PERMIT) { 1507 if (thflags & TH_CWR) 1508 tp->t_flags &= ~TF_ECN_SND_ECE; 1509 switch (iptos & IPTOS_ECN_MASK) { 1510 case IPTOS_ECN_CE: 1511 tp->t_flags |= TF_ECN_SND_ECE; 1512 TCPSTAT_INC(tcps_ecn_ce); 1513 break; 1514 case IPTOS_ECN_ECT0: 1515 TCPSTAT_INC(tcps_ecn_ect0); 1516 break; 1517 case IPTOS_ECN_ECT1: 1518 TCPSTAT_INC(tcps_ecn_ect1); 1519 break; 1520 } 1521 /* Congestion experienced. */ 1522 if (thflags & TH_ECE) { 1523 cc_cong_signal(tp, th, CC_ECN); 1524 } 1525 } 1526 1527 /* 1528 * Parse options on any incoming segment. 1529 */ 1530 tcp_dooptions(&to, (u_char *)(th + 1), 1531 (th->th_off << 2) - sizeof(struct tcphdr), 1532 (thflags & TH_SYN) ? TO_SYN : 0); 1533 1534 /* 1535 * If echoed timestamp is later than the current time, 1536 * fall back to non RFC1323 RTT calculation. Normalize 1537 * timestamp if syncookies were used when this connection 1538 * was established. 1539 */ 1540 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1541 to.to_tsecr -= tp->ts_offset; 1542 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1543 to.to_tsecr = 0; 1544 } 1545 /* 1546 * If timestamps were negotiated during SYN/ACK they should 1547 * appear on every segment during this session and vice versa. 1548 */ 1549 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1550 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1551 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1552 "no action\n", s, __func__); 1553 free(s, M_TCPLOG); 1554 } 1555 } 1556 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1557 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1558 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1559 "no action\n", s, __func__); 1560 free(s, M_TCPLOG); 1561 } 1562 } 1563 1564 /* 1565 * Process options only when we get SYN/ACK back. The SYN case 1566 * for incoming connections is handled in tcp_syncache. 1567 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1568 * or <SYN,ACK>) segment itself is never scaled. 1569 * XXX this is traditional behavior, may need to be cleaned up. 1570 */ 1571 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1572 if ((to.to_flags & TOF_SCALE) && 1573 (tp->t_flags & TF_REQ_SCALE)) { 1574 tp->t_flags |= TF_RCVD_SCALE; 1575 tp->snd_scale = to.to_wscale; 1576 } 1577 /* 1578 * Initial send window. It will be updated with 1579 * the next incoming segment to the scaled value. 1580 */ 1581 tp->snd_wnd = th->th_win; 1582 if (to.to_flags & TOF_TS) { 1583 tp->t_flags |= TF_RCVD_TSTMP; 1584 tp->ts_recent = to.to_tsval; 1585 tp->ts_recent_age = tcp_ts_getticks(); 1586 } 1587 if (to.to_flags & TOF_MSS) 1588 tcp_mss(tp, to.to_mss); 1589 if ((tp->t_flags & TF_SACK_PERMIT) && 1590 (to.to_flags & TOF_SACKPERM) == 0) 1591 tp->t_flags &= ~TF_SACK_PERMIT; 1592 } 1593 1594 /* 1595 * Header prediction: check for the two common cases 1596 * of a uni-directional data xfer. If the packet has 1597 * no control flags, is in-sequence, the window didn't 1598 * change and we're not retransmitting, it's a 1599 * candidate. If the length is zero and the ack moved 1600 * forward, we're the sender side of the xfer. Just 1601 * free the data acked & wake any higher level process 1602 * that was blocked waiting for space. If the length 1603 * is non-zero and the ack didn't move, we're the 1604 * receiver side. If we're getting packets in-order 1605 * (the reassembly queue is empty), add the data to 1606 * the socket buffer and note that we need a delayed ack. 1607 * Make sure that the hidden state-flags are also off. 1608 * Since we check for TCPS_ESTABLISHED first, it can only 1609 * be TH_NEEDSYN. 1610 */ 1611 if (tp->t_state == TCPS_ESTABLISHED && 1612 th->th_seq == tp->rcv_nxt && 1613 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1614 tp->snd_nxt == tp->snd_max && 1615 tiwin && tiwin == tp->snd_wnd && 1616 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1617 tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 || 1618 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1619 1620 /* 1621 * If last ACK falls within this segment's sequence numbers, 1622 * record the timestamp. 1623 * NOTE that the test is modified according to the latest 1624 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1625 */ 1626 if ((to.to_flags & TOF_TS) != 0 && 1627 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1628 tp->ts_recent_age = tcp_ts_getticks(); 1629 tp->ts_recent = to.to_tsval; 1630 } 1631 1632 if (tlen == 0) { 1633 if (SEQ_GT(th->th_ack, tp->snd_una) && 1634 SEQ_LEQ(th->th_ack, tp->snd_max) && 1635 !IN_RECOVERY(tp->t_flags) && 1636 (to.to_flags & TOF_SACK) == 0 && 1637 TAILQ_EMPTY(&tp->snd_holes)) { 1638 /* 1639 * This is a pure ack for outstanding data. 1640 */ 1641 if (ti_locked == TI_WLOCKED) 1642 INP_INFO_WUNLOCK(&V_tcbinfo); 1643 ti_locked = TI_UNLOCKED; 1644 1645 TCPSTAT_INC(tcps_predack); 1646 1647 /* 1648 * "bad retransmit" recovery. 1649 */ 1650 if (tp->t_rxtshift == 1 && 1651 tp->t_flags & TF_PREVVALID && 1652 (int)(ticks - tp->t_badrxtwin) < 0) { 1653 cc_cong_signal(tp, th, CC_RTO_ERR); 1654 } 1655 1656 /* 1657 * Recalculate the transmit timer / rtt. 1658 * 1659 * Some boxes send broken timestamp replies 1660 * during the SYN+ACK phase, ignore 1661 * timestamps of 0 or we could calculate a 1662 * huge RTT and blow up the retransmit timer. 1663 */ 1664 if ((to.to_flags & TOF_TS) != 0 && 1665 to.to_tsecr) { 1666 u_int t; 1667 1668 t = tcp_ts_getticks() - to.to_tsecr; 1669 if (!tp->t_rttlow || tp->t_rttlow > t) 1670 tp->t_rttlow = t; 1671 tcp_xmit_timer(tp, 1672 TCP_TS_TO_TICKS(t) + 1); 1673 } else if (tp->t_rtttime && 1674 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1675 if (!tp->t_rttlow || 1676 tp->t_rttlow > ticks - tp->t_rtttime) 1677 tp->t_rttlow = ticks - tp->t_rtttime; 1678 tcp_xmit_timer(tp, 1679 ticks - tp->t_rtttime); 1680 } 1681 acked = BYTES_THIS_ACK(tp, th); 1682 1683 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1684 hhook_run_tcp_est_in(tp, th, &to); 1685 1686 TCPSTAT_INC(tcps_rcvackpack); 1687 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1688 sbdrop(&so->so_snd, acked); 1689 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1690 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1691 tp->snd_recover = th->th_ack - 1; 1692 1693 /* 1694 * Let the congestion control algorithm update 1695 * congestion control related information. This 1696 * typically means increasing the congestion 1697 * window. 1698 */ 1699 cc_ack_received(tp, th, CC_ACK); 1700 1701 tp->snd_una = th->th_ack; 1702 /* 1703 * Pull snd_wl2 up to prevent seq wrap relative 1704 * to th_ack. 1705 */ 1706 tp->snd_wl2 = th->th_ack; 1707 tp->t_dupacks = 0; 1708 m_freem(m); 1709 ND6_HINT(tp); /* Some progress has been made. */ 1710 1711 /* 1712 * If all outstanding data are acked, stop 1713 * retransmit timer, otherwise restart timer 1714 * using current (possibly backed-off) value. 1715 * If process is waiting for space, 1716 * wakeup/selwakeup/signal. If data 1717 * are ready to send, let tcp_output 1718 * decide between more output or persist. 1719 */ 1720 #ifdef TCPDEBUG 1721 if (so->so_options & SO_DEBUG) 1722 tcp_trace(TA_INPUT, ostate, tp, 1723 (void *)tcp_saveipgen, 1724 &tcp_savetcp, 0); 1725 #endif 1726 if (tp->snd_una == tp->snd_max) 1727 tcp_timer_activate(tp, TT_REXMT, 0); 1728 else if (!tcp_timer_active(tp, TT_PERSIST)) 1729 tcp_timer_activate(tp, TT_REXMT, 1730 tp->t_rxtcur); 1731 sowwakeup(so); 1732 if (so->so_snd.sb_cc) 1733 (void) tcp_output(tp); 1734 goto check_delack; 1735 } 1736 } else if (th->th_ack == tp->snd_una && 1737 tlen <= sbspace(&so->so_rcv)) { 1738 int newsize = 0; /* automatic sockbuf scaling */ 1739 1740 /* 1741 * This is a pure, in-sequence data packet with 1742 * nothing on the reassembly queue and we have enough 1743 * buffer space to take it. 1744 */ 1745 if (ti_locked == TI_WLOCKED) 1746 INP_INFO_WUNLOCK(&V_tcbinfo); 1747 ti_locked = TI_UNLOCKED; 1748 1749 /* Clean receiver SACK report if present */ 1750 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1751 tcp_clean_sackreport(tp); 1752 TCPSTAT_INC(tcps_preddat); 1753 tp->rcv_nxt += tlen; 1754 /* 1755 * Pull snd_wl1 up to prevent seq wrap relative to 1756 * th_seq. 1757 */ 1758 tp->snd_wl1 = th->th_seq; 1759 /* 1760 * Pull rcv_up up to prevent seq wrap relative to 1761 * rcv_nxt. 1762 */ 1763 tp->rcv_up = tp->rcv_nxt; 1764 TCPSTAT_INC(tcps_rcvpack); 1765 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1766 ND6_HINT(tp); /* Some progress has been made */ 1767 #ifdef TCPDEBUG 1768 if (so->so_options & SO_DEBUG) 1769 tcp_trace(TA_INPUT, ostate, tp, 1770 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1771 #endif 1772 /* 1773 * Automatic sizing of receive socket buffer. Often the send 1774 * buffer size is not optimally adjusted to the actual network 1775 * conditions at hand (delay bandwidth product). Setting the 1776 * buffer size too small limits throughput on links with high 1777 * bandwidth and high delay (eg. trans-continental/oceanic links). 1778 * 1779 * On the receive side the socket buffer memory is only rarely 1780 * used to any significant extent. This allows us to be much 1781 * more aggressive in scaling the receive socket buffer. For 1782 * the case that the buffer space is actually used to a large 1783 * extent and we run out of kernel memory we can simply drop 1784 * the new segments; TCP on the sender will just retransmit it 1785 * later. Setting the buffer size too big may only consume too 1786 * much kernel memory if the application doesn't read() from 1787 * the socket or packet loss or reordering makes use of the 1788 * reassembly queue. 1789 * 1790 * The criteria to step up the receive buffer one notch are: 1791 * 1. the number of bytes received during the time it takes 1792 * one timestamp to be reflected back to us (the RTT); 1793 * 2. received bytes per RTT is within seven eighth of the 1794 * current socket buffer size; 1795 * 3. receive buffer size has not hit maximal automatic size; 1796 * 1797 * This algorithm does one step per RTT at most and only if 1798 * we receive a bulk stream w/o packet losses or reorderings. 1799 * Shrinking the buffer during idle times is not necessary as 1800 * it doesn't consume any memory when idle. 1801 * 1802 * TODO: Only step up if the application is actually serving 1803 * the buffer to better manage the socket buffer resources. 1804 */ 1805 if (V_tcp_do_autorcvbuf && 1806 to.to_tsecr && 1807 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1808 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1809 to.to_tsecr - tp->rfbuf_ts < hz) { 1810 if (tp->rfbuf_cnt > 1811 (so->so_rcv.sb_hiwat / 8 * 7) && 1812 so->so_rcv.sb_hiwat < 1813 V_tcp_autorcvbuf_max) { 1814 newsize = 1815 min(so->so_rcv.sb_hiwat + 1816 V_tcp_autorcvbuf_inc, 1817 V_tcp_autorcvbuf_max); 1818 } 1819 /* Start over with next RTT. */ 1820 tp->rfbuf_ts = 0; 1821 tp->rfbuf_cnt = 0; 1822 } else 1823 tp->rfbuf_cnt += tlen; /* add up */ 1824 } 1825 1826 /* Add data to socket buffer. */ 1827 SOCKBUF_LOCK(&so->so_rcv); 1828 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1829 m_freem(m); 1830 } else { 1831 /* 1832 * Set new socket buffer size. 1833 * Give up when limit is reached. 1834 */ 1835 if (newsize) 1836 if (!sbreserve_locked(&so->so_rcv, 1837 newsize, so, NULL)) 1838 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1839 m_adj(m, drop_hdrlen); /* delayed header drop */ 1840 sbappendstream_locked(&so->so_rcv, m); 1841 } 1842 /* NB: sorwakeup_locked() does an implicit unlock. */ 1843 sorwakeup_locked(so); 1844 if (DELAY_ACK(tp, tlen)) { 1845 tp->t_flags |= TF_DELACK; 1846 } else { 1847 tp->t_flags |= TF_ACKNOW; 1848 tcp_output(tp); 1849 } 1850 goto check_delack; 1851 } 1852 } 1853 1854 /* 1855 * Calculate amount of space in receive window, 1856 * and then do TCP input processing. 1857 * Receive window is amount of space in rcv queue, 1858 * but not less than advertised window. 1859 */ 1860 win = sbspace(&so->so_rcv); 1861 if (win < 0) 1862 win = 0; 1863 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1864 1865 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1866 tp->rfbuf_ts = 0; 1867 tp->rfbuf_cnt = 0; 1868 1869 switch (tp->t_state) { 1870 1871 /* 1872 * If the state is SYN_RECEIVED: 1873 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1874 */ 1875 case TCPS_SYN_RECEIVED: 1876 if ((thflags & TH_ACK) && 1877 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1878 SEQ_GT(th->th_ack, tp->snd_max))) { 1879 rstreason = BANDLIM_RST_OPENPORT; 1880 goto dropwithreset; 1881 } 1882 break; 1883 1884 /* 1885 * If the state is SYN_SENT: 1886 * if seg contains an ACK, but not for our SYN, drop the input. 1887 * if seg contains a RST, then drop the connection. 1888 * if seg does not contain SYN, then drop it. 1889 * Otherwise this is an acceptable SYN segment 1890 * initialize tp->rcv_nxt and tp->irs 1891 * if seg contains ack then advance tp->snd_una 1892 * if seg contains an ECE and ECN support is enabled, the stream 1893 * is ECN capable. 1894 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1895 * arrange for segment to be acked (eventually) 1896 * continue processing rest of data/controls, beginning with URG 1897 */ 1898 case TCPS_SYN_SENT: 1899 if ((thflags & TH_ACK) && 1900 (SEQ_LEQ(th->th_ack, tp->iss) || 1901 SEQ_GT(th->th_ack, tp->snd_max))) { 1902 rstreason = BANDLIM_UNLIMITED; 1903 goto dropwithreset; 1904 } 1905 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1906 TCP_PROBE5(connect__refused, NULL, tp, 1907 mtod(m, const char *), tp, th); 1908 tp = tcp_drop(tp, ECONNREFUSED); 1909 } 1910 if (thflags & TH_RST) 1911 goto drop; 1912 if (!(thflags & TH_SYN)) 1913 goto drop; 1914 1915 tp->irs = th->th_seq; 1916 tcp_rcvseqinit(tp); 1917 if (thflags & TH_ACK) { 1918 TCPSTAT_INC(tcps_connects); 1919 soisconnected(so); 1920 #ifdef MAC 1921 mac_socketpeer_set_from_mbuf(m, so); 1922 #endif 1923 /* Do window scaling on this connection? */ 1924 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1925 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1926 tp->rcv_scale = tp->request_r_scale; 1927 } 1928 tp->rcv_adv += imin(tp->rcv_wnd, 1929 TCP_MAXWIN << tp->rcv_scale); 1930 tp->snd_una++; /* SYN is acked */ 1931 /* 1932 * If there's data, delay ACK; if there's also a FIN 1933 * ACKNOW will be turned on later. 1934 */ 1935 if (DELAY_ACK(tp, tlen) && tlen != 0) 1936 tcp_timer_activate(tp, TT_DELACK, 1937 tcp_delacktime); 1938 else 1939 tp->t_flags |= TF_ACKNOW; 1940 1941 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1942 tp->t_flags |= TF_ECN_PERMIT; 1943 TCPSTAT_INC(tcps_ecn_shs); 1944 } 1945 1946 /* 1947 * Received <SYN,ACK> in SYN_SENT[*] state. 1948 * Transitions: 1949 * SYN_SENT --> ESTABLISHED 1950 * SYN_SENT* --> FIN_WAIT_1 1951 */ 1952 tp->t_starttime = ticks; 1953 if (tp->t_flags & TF_NEEDFIN) { 1954 tcp_state_change(tp, TCPS_FIN_WAIT_1); 1955 tp->t_flags &= ~TF_NEEDFIN; 1956 thflags &= ~TH_SYN; 1957 } else { 1958 tcp_state_change(tp, TCPS_ESTABLISHED); 1959 TCP_PROBE5(connect__established, NULL, tp, 1960 mtod(m, const char *), tp, th); 1961 cc_conn_init(tp); 1962 tcp_timer_activate(tp, TT_KEEP, 1963 TP_KEEPIDLE(tp)); 1964 } 1965 } else { 1966 /* 1967 * Received initial SYN in SYN-SENT[*] state => 1968 * simultaneous open. 1969 * If it succeeds, connection is * half-synchronized. 1970 * Otherwise, do 3-way handshake: 1971 * SYN-SENT -> SYN-RECEIVED 1972 * SYN-SENT* -> SYN-RECEIVED* 1973 */ 1974 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1975 tcp_timer_activate(tp, TT_REXMT, 0); 1976 tcp_state_change(tp, TCPS_SYN_RECEIVED); 1977 } 1978 1979 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1980 "ti_locked %d", __func__, ti_locked)); 1981 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1982 INP_WLOCK_ASSERT(tp->t_inpcb); 1983 1984 /* 1985 * Advance th->th_seq to correspond to first data byte. 1986 * If data, trim to stay within window, 1987 * dropping FIN if necessary. 1988 */ 1989 th->th_seq++; 1990 if (tlen > tp->rcv_wnd) { 1991 todrop = tlen - tp->rcv_wnd; 1992 m_adj(m, -todrop); 1993 tlen = tp->rcv_wnd; 1994 thflags &= ~TH_FIN; 1995 TCPSTAT_INC(tcps_rcvpackafterwin); 1996 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1997 } 1998 tp->snd_wl1 = th->th_seq - 1; 1999 tp->rcv_up = th->th_seq; 2000 /* 2001 * Client side of transaction: already sent SYN and data. 2002 * If the remote host used T/TCP to validate the SYN, 2003 * our data will be ACK'd; if so, enter normal data segment 2004 * processing in the middle of step 5, ack processing. 2005 * Otherwise, goto step 6. 2006 */ 2007 if (thflags & TH_ACK) 2008 goto process_ACK; 2009 2010 goto step6; 2011 2012 /* 2013 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2014 * do normal processing. 2015 * 2016 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2017 */ 2018 case TCPS_LAST_ACK: 2019 case TCPS_CLOSING: 2020 break; /* continue normal processing */ 2021 } 2022 2023 /* 2024 * States other than LISTEN or SYN_SENT. 2025 * First check the RST flag and sequence number since reset segments 2026 * are exempt from the timestamp and connection count tests. This 2027 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2028 * below which allowed reset segments in half the sequence space 2029 * to fall though and be processed (which gives forged reset 2030 * segments with a random sequence number a 50 percent chance of 2031 * killing a connection). 2032 * Then check timestamp, if present. 2033 * Then check the connection count, if present. 2034 * Then check that at least some bytes of segment are within 2035 * receive window. If segment begins before rcv_nxt, 2036 * drop leading data (and SYN); if nothing left, just ack. 2037 * 2038 * 2039 * If the RST bit is set, check the sequence number to see 2040 * if this is a valid reset segment. 2041 * RFC 793 page 37: 2042 * In all states except SYN-SENT, all reset (RST) segments 2043 * are validated by checking their SEQ-fields. A reset is 2044 * valid if its sequence number is in the window. 2045 * Note: this does not take into account delayed ACKs, so 2046 * we should test against last_ack_sent instead of rcv_nxt. 2047 * The sequence number in the reset segment is normally an 2048 * echo of our outgoing acknowlegement numbers, but some hosts 2049 * send a reset with the sequence number at the rightmost edge 2050 * of our receive window, and we have to handle this case. 2051 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2052 * that brute force RST attacks are possible. To combat this, 2053 * we use a much stricter check while in the ESTABLISHED state, 2054 * only accepting RSTs where the sequence number is equal to 2055 * last_ack_sent. In all other states (the states in which a 2056 * RST is more likely), the more permissive check is used. 2057 * If we have multiple segments in flight, the initial reset 2058 * segment sequence numbers will be to the left of last_ack_sent, 2059 * but they will eventually catch up. 2060 * In any case, it never made sense to trim reset segments to 2061 * fit the receive window since RFC 1122 says: 2062 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2063 * 2064 * A TCP SHOULD allow a received RST segment to include data. 2065 * 2066 * DISCUSSION 2067 * It has been suggested that a RST segment could contain 2068 * ASCII text that encoded and explained the cause of the 2069 * RST. No standard has yet been established for such 2070 * data. 2071 * 2072 * If the reset segment passes the sequence number test examine 2073 * the state: 2074 * SYN_RECEIVED STATE: 2075 * If passive open, return to LISTEN state. 2076 * If active open, inform user that connection was refused. 2077 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2078 * Inform user that connection was reset, and close tcb. 2079 * CLOSING, LAST_ACK STATES: 2080 * Close the tcb. 2081 * TIME_WAIT STATE: 2082 * Drop the segment - see Stevens, vol. 2, p. 964 and 2083 * RFC 1337. 2084 */ 2085 if (thflags & TH_RST) { 2086 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2087 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2088 switch (tp->t_state) { 2089 2090 case TCPS_SYN_RECEIVED: 2091 so->so_error = ECONNREFUSED; 2092 goto close; 2093 2094 case TCPS_ESTABLISHED: 2095 if (V_tcp_insecure_rst == 0 && 2096 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2097 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2098 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2099 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2100 TCPSTAT_INC(tcps_badrst); 2101 goto drop; 2102 } 2103 /* FALLTHROUGH */ 2104 case TCPS_FIN_WAIT_1: 2105 case TCPS_FIN_WAIT_2: 2106 case TCPS_CLOSE_WAIT: 2107 so->so_error = ECONNRESET; 2108 close: 2109 KASSERT(ti_locked == TI_WLOCKED, 2110 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2111 ti_locked)); 2112 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2113 2114 tcp_state_change(tp, TCPS_CLOSED); 2115 TCPSTAT_INC(tcps_drops); 2116 tp = tcp_close(tp); 2117 break; 2118 2119 case TCPS_CLOSING: 2120 case TCPS_LAST_ACK: 2121 KASSERT(ti_locked == TI_WLOCKED, 2122 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2123 ti_locked)); 2124 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2125 2126 tp = tcp_close(tp); 2127 break; 2128 } 2129 } 2130 goto drop; 2131 } 2132 2133 /* 2134 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2135 * and it's less than ts_recent, drop it. 2136 */ 2137 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2138 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2139 2140 /* Check to see if ts_recent is over 24 days old. */ 2141 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2142 /* 2143 * Invalidate ts_recent. If this segment updates 2144 * ts_recent, the age will be reset later and ts_recent 2145 * will get a valid value. If it does not, setting 2146 * ts_recent to zero will at least satisfy the 2147 * requirement that zero be placed in the timestamp 2148 * echo reply when ts_recent isn't valid. The 2149 * age isn't reset until we get a valid ts_recent 2150 * because we don't want out-of-order segments to be 2151 * dropped when ts_recent is old. 2152 */ 2153 tp->ts_recent = 0; 2154 } else { 2155 TCPSTAT_INC(tcps_rcvduppack); 2156 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2157 TCPSTAT_INC(tcps_pawsdrop); 2158 if (tlen) 2159 goto dropafterack; 2160 goto drop; 2161 } 2162 } 2163 2164 /* 2165 * In the SYN-RECEIVED state, validate that the packet belongs to 2166 * this connection before trimming the data to fit the receive 2167 * window. Check the sequence number versus IRS since we know 2168 * the sequence numbers haven't wrapped. This is a partial fix 2169 * for the "LAND" DoS attack. 2170 */ 2171 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2172 rstreason = BANDLIM_RST_OPENPORT; 2173 goto dropwithreset; 2174 } 2175 2176 todrop = tp->rcv_nxt - th->th_seq; 2177 if (todrop > 0) { 2178 /* 2179 * If this is a duplicate SYN for our current connection, 2180 * advance over it and pretend and it's not a SYN. 2181 */ 2182 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2183 thflags &= ~TH_SYN; 2184 th->th_seq++; 2185 if (th->th_urp > 1) 2186 th->th_urp--; 2187 else 2188 thflags &= ~TH_URG; 2189 todrop--; 2190 } 2191 /* 2192 * Following if statement from Stevens, vol. 2, p. 960. 2193 */ 2194 if (todrop > tlen 2195 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2196 /* 2197 * Any valid FIN must be to the left of the window. 2198 * At this point the FIN must be a duplicate or out 2199 * of sequence; drop it. 2200 */ 2201 thflags &= ~TH_FIN; 2202 2203 /* 2204 * Send an ACK to resynchronize and drop any data. 2205 * But keep on processing for RST or ACK. 2206 */ 2207 tp->t_flags |= TF_ACKNOW; 2208 todrop = tlen; 2209 TCPSTAT_INC(tcps_rcvduppack); 2210 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2211 } else { 2212 TCPSTAT_INC(tcps_rcvpartduppack); 2213 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2214 } 2215 drop_hdrlen += todrop; /* drop from the top afterwards */ 2216 th->th_seq += todrop; 2217 tlen -= todrop; 2218 if (th->th_urp > todrop) 2219 th->th_urp -= todrop; 2220 else { 2221 thflags &= ~TH_URG; 2222 th->th_urp = 0; 2223 } 2224 } 2225 2226 /* 2227 * If new data are received on a connection after the 2228 * user processes are gone, then RST the other end. 2229 */ 2230 if ((so->so_state & SS_NOFDREF) && 2231 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2232 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2233 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2234 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2235 2236 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2237 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2238 "after socket was closed, " 2239 "sending RST and removing tcpcb\n", 2240 s, __func__, tcpstates[tp->t_state], tlen); 2241 free(s, M_TCPLOG); 2242 } 2243 tp = tcp_close(tp); 2244 TCPSTAT_INC(tcps_rcvafterclose); 2245 rstreason = BANDLIM_UNLIMITED; 2246 goto dropwithreset; 2247 } 2248 2249 /* 2250 * If segment ends after window, drop trailing data 2251 * (and PUSH and FIN); if nothing left, just ACK. 2252 */ 2253 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2254 if (todrop > 0) { 2255 TCPSTAT_INC(tcps_rcvpackafterwin); 2256 if (todrop >= tlen) { 2257 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2258 /* 2259 * If window is closed can only take segments at 2260 * window edge, and have to drop data and PUSH from 2261 * incoming segments. Continue processing, but 2262 * remember to ack. Otherwise, drop segment 2263 * and ack. 2264 */ 2265 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2266 tp->t_flags |= TF_ACKNOW; 2267 TCPSTAT_INC(tcps_rcvwinprobe); 2268 } else 2269 goto dropafterack; 2270 } else 2271 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2272 m_adj(m, -todrop); 2273 tlen -= todrop; 2274 thflags &= ~(TH_PUSH|TH_FIN); 2275 } 2276 2277 /* 2278 * If last ACK falls within this segment's sequence numbers, 2279 * record its timestamp. 2280 * NOTE: 2281 * 1) That the test incorporates suggestions from the latest 2282 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2283 * 2) That updating only on newer timestamps interferes with 2284 * our earlier PAWS tests, so this check should be solely 2285 * predicated on the sequence space of this segment. 2286 * 3) That we modify the segment boundary check to be 2287 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2288 * instead of RFC1323's 2289 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2290 * This modified check allows us to overcome RFC1323's 2291 * limitations as described in Stevens TCP/IP Illustrated 2292 * Vol. 2 p.869. In such cases, we can still calculate the 2293 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2294 */ 2295 if ((to.to_flags & TOF_TS) != 0 && 2296 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2297 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2298 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2299 tp->ts_recent_age = tcp_ts_getticks(); 2300 tp->ts_recent = to.to_tsval; 2301 } 2302 2303 /* 2304 * If a SYN is in the window, then this is an 2305 * error and we send an RST and drop the connection. 2306 */ 2307 if (thflags & TH_SYN) { 2308 KASSERT(ti_locked == TI_WLOCKED, 2309 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2310 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2311 2312 tp = tcp_drop(tp, ECONNRESET); 2313 rstreason = BANDLIM_UNLIMITED; 2314 goto drop; 2315 } 2316 2317 /* 2318 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2319 * flag is on (half-synchronized state), then queue data for 2320 * later processing; else drop segment and return. 2321 */ 2322 if ((thflags & TH_ACK) == 0) { 2323 if (tp->t_state == TCPS_SYN_RECEIVED || 2324 (tp->t_flags & TF_NEEDSYN)) 2325 goto step6; 2326 else if (tp->t_flags & TF_ACKNOW) 2327 goto dropafterack; 2328 else 2329 goto drop; 2330 } 2331 2332 /* 2333 * Ack processing. 2334 */ 2335 switch (tp->t_state) { 2336 2337 /* 2338 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2339 * ESTABLISHED state and continue processing. 2340 * The ACK was checked above. 2341 */ 2342 case TCPS_SYN_RECEIVED: 2343 2344 TCPSTAT_INC(tcps_connects); 2345 soisconnected(so); 2346 /* Do window scaling? */ 2347 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2348 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2349 tp->rcv_scale = tp->request_r_scale; 2350 tp->snd_wnd = tiwin; 2351 } 2352 /* 2353 * Make transitions: 2354 * SYN-RECEIVED -> ESTABLISHED 2355 * SYN-RECEIVED* -> FIN-WAIT-1 2356 */ 2357 tp->t_starttime = ticks; 2358 if (tp->t_flags & TF_NEEDFIN) { 2359 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2360 tp->t_flags &= ~TF_NEEDFIN; 2361 } else { 2362 tcp_state_change(tp, TCPS_ESTABLISHED); 2363 TCP_PROBE5(accept__established, NULL, tp, 2364 mtod(m, const char *), tp, th); 2365 cc_conn_init(tp); 2366 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2367 } 2368 /* 2369 * If segment contains data or ACK, will call tcp_reass() 2370 * later; if not, do so now to pass queued data to user. 2371 */ 2372 if (tlen == 0 && (thflags & TH_FIN) == 0) 2373 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2374 (struct mbuf *)0); 2375 tp->snd_wl1 = th->th_seq - 1; 2376 /* FALLTHROUGH */ 2377 2378 /* 2379 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2380 * ACKs. If the ack is in the range 2381 * tp->snd_una < th->th_ack <= tp->snd_max 2382 * then advance tp->snd_una to th->th_ack and drop 2383 * data from the retransmission queue. If this ACK reflects 2384 * more up to date window information we update our window information. 2385 */ 2386 case TCPS_ESTABLISHED: 2387 case TCPS_FIN_WAIT_1: 2388 case TCPS_FIN_WAIT_2: 2389 case TCPS_CLOSE_WAIT: 2390 case TCPS_CLOSING: 2391 case TCPS_LAST_ACK: 2392 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2393 TCPSTAT_INC(tcps_rcvacktoomuch); 2394 goto dropafterack; 2395 } 2396 if ((tp->t_flags & TF_SACK_PERMIT) && 2397 ((to.to_flags & TOF_SACK) || 2398 !TAILQ_EMPTY(&tp->snd_holes))) 2399 tcp_sack_doack(tp, &to, th->th_ack); 2400 2401 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2402 hhook_run_tcp_est_in(tp, th, &to); 2403 2404 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2405 if (tlen == 0 && tiwin == tp->snd_wnd) { 2406 /* 2407 * If this is the first time we've seen a 2408 * FIN from the remote, this is not a 2409 * duplicate and it needs to be processed 2410 * normally. This happens during a 2411 * simultaneous close. 2412 */ 2413 if ((thflags & TH_FIN) && 2414 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2415 tp->t_dupacks = 0; 2416 break; 2417 } 2418 TCPSTAT_INC(tcps_rcvdupack); 2419 /* 2420 * If we have outstanding data (other than 2421 * a window probe), this is a completely 2422 * duplicate ack (ie, window info didn't 2423 * change and FIN isn't set), 2424 * the ack is the biggest we've 2425 * seen and we've seen exactly our rexmt 2426 * threshhold of them, assume a packet 2427 * has been dropped and retransmit it. 2428 * Kludge snd_nxt & the congestion 2429 * window so we send only this one 2430 * packet. 2431 * 2432 * We know we're losing at the current 2433 * window size so do congestion avoidance 2434 * (set ssthresh to half the current window 2435 * and pull our congestion window back to 2436 * the new ssthresh). 2437 * 2438 * Dup acks mean that packets have left the 2439 * network (they're now cached at the receiver) 2440 * so bump cwnd by the amount in the receiver 2441 * to keep a constant cwnd packets in the 2442 * network. 2443 * 2444 * When using TCP ECN, notify the peer that 2445 * we reduced the cwnd. 2446 */ 2447 if (!tcp_timer_active(tp, TT_REXMT) || 2448 th->th_ack != tp->snd_una) 2449 tp->t_dupacks = 0; 2450 else if (++tp->t_dupacks > tcprexmtthresh || 2451 IN_FASTRECOVERY(tp->t_flags)) { 2452 cc_ack_received(tp, th, CC_DUPACK); 2453 if ((tp->t_flags & TF_SACK_PERMIT) && 2454 IN_FASTRECOVERY(tp->t_flags)) { 2455 int awnd; 2456 2457 /* 2458 * Compute the amount of data in flight first. 2459 * We can inject new data into the pipe iff 2460 * we have less than 1/2 the original window's 2461 * worth of data in flight. 2462 */ 2463 awnd = (tp->snd_nxt - tp->snd_fack) + 2464 tp->sackhint.sack_bytes_rexmit; 2465 if (awnd < tp->snd_ssthresh) { 2466 tp->snd_cwnd += tp->t_maxseg; 2467 if (tp->snd_cwnd > tp->snd_ssthresh) 2468 tp->snd_cwnd = tp->snd_ssthresh; 2469 } 2470 } else 2471 tp->snd_cwnd += tp->t_maxseg; 2472 (void) tcp_output(tp); 2473 goto drop; 2474 } else if (tp->t_dupacks == tcprexmtthresh) { 2475 tcp_seq onxt = tp->snd_nxt; 2476 2477 /* 2478 * If we're doing sack, check to 2479 * see if we're already in sack 2480 * recovery. If we're not doing sack, 2481 * check to see if we're in newreno 2482 * recovery. 2483 */ 2484 if (tp->t_flags & TF_SACK_PERMIT) { 2485 if (IN_FASTRECOVERY(tp->t_flags)) { 2486 tp->t_dupacks = 0; 2487 break; 2488 } 2489 } else { 2490 if (SEQ_LEQ(th->th_ack, 2491 tp->snd_recover)) { 2492 tp->t_dupacks = 0; 2493 break; 2494 } 2495 } 2496 /* Congestion signal before ack. */ 2497 cc_cong_signal(tp, th, CC_NDUPACK); 2498 cc_ack_received(tp, th, CC_DUPACK); 2499 tcp_timer_activate(tp, TT_REXMT, 0); 2500 tp->t_rtttime = 0; 2501 if (tp->t_flags & TF_SACK_PERMIT) { 2502 TCPSTAT_INC( 2503 tcps_sack_recovery_episode); 2504 tp->sack_newdata = tp->snd_nxt; 2505 tp->snd_cwnd = tp->t_maxseg; 2506 (void) tcp_output(tp); 2507 goto drop; 2508 } 2509 tp->snd_nxt = th->th_ack; 2510 tp->snd_cwnd = tp->t_maxseg; 2511 (void) tcp_output(tp); 2512 KASSERT(tp->snd_limited <= 2, 2513 ("%s: tp->snd_limited too big", 2514 __func__)); 2515 tp->snd_cwnd = tp->snd_ssthresh + 2516 tp->t_maxseg * 2517 (tp->t_dupacks - tp->snd_limited); 2518 if (SEQ_GT(onxt, tp->snd_nxt)) 2519 tp->snd_nxt = onxt; 2520 goto drop; 2521 } else if (V_tcp_do_rfc3042) { 2522 cc_ack_received(tp, th, CC_DUPACK); 2523 u_long oldcwnd = tp->snd_cwnd; 2524 tcp_seq oldsndmax = tp->snd_max; 2525 u_int sent; 2526 int avail; 2527 2528 KASSERT(tp->t_dupacks == 1 || 2529 tp->t_dupacks == 2, 2530 ("%s: dupacks not 1 or 2", 2531 __func__)); 2532 if (tp->t_dupacks == 1) 2533 tp->snd_limited = 0; 2534 tp->snd_cwnd = 2535 (tp->snd_nxt - tp->snd_una) + 2536 (tp->t_dupacks - tp->snd_limited) * 2537 tp->t_maxseg; 2538 /* 2539 * Only call tcp_output when there 2540 * is new data available to be sent. 2541 * Otherwise we would send pure ACKs. 2542 */ 2543 SOCKBUF_LOCK(&so->so_snd); 2544 avail = so->so_snd.sb_cc - 2545 (tp->snd_nxt - tp->snd_una); 2546 SOCKBUF_UNLOCK(&so->so_snd); 2547 if (avail > 0) 2548 (void) tcp_output(tp); 2549 sent = tp->snd_max - oldsndmax; 2550 if (sent > tp->t_maxseg) { 2551 KASSERT((tp->t_dupacks == 2 && 2552 tp->snd_limited == 0) || 2553 (sent == tp->t_maxseg + 1 && 2554 tp->t_flags & TF_SENTFIN), 2555 ("%s: sent too much", 2556 __func__)); 2557 tp->snd_limited = 2; 2558 } else if (sent > 0) 2559 ++tp->snd_limited; 2560 tp->snd_cwnd = oldcwnd; 2561 goto drop; 2562 } 2563 } else 2564 tp->t_dupacks = 0; 2565 break; 2566 } 2567 2568 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2569 ("%s: th_ack <= snd_una", __func__)); 2570 2571 /* 2572 * If the congestion window was inflated to account 2573 * for the other side's cached packets, retract it. 2574 */ 2575 if (IN_FASTRECOVERY(tp->t_flags)) { 2576 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2577 if (tp->t_flags & TF_SACK_PERMIT) 2578 tcp_sack_partialack(tp, th); 2579 else 2580 tcp_newreno_partial_ack(tp, th); 2581 } else 2582 cc_post_recovery(tp, th); 2583 } 2584 tp->t_dupacks = 0; 2585 /* 2586 * If we reach this point, ACK is not a duplicate, 2587 * i.e., it ACKs something we sent. 2588 */ 2589 if (tp->t_flags & TF_NEEDSYN) { 2590 /* 2591 * T/TCP: Connection was half-synchronized, and our 2592 * SYN has been ACK'd (so connection is now fully 2593 * synchronized). Go to non-starred state, 2594 * increment snd_una for ACK of SYN, and check if 2595 * we can do window scaling. 2596 */ 2597 tp->t_flags &= ~TF_NEEDSYN; 2598 tp->snd_una++; 2599 /* Do window scaling? */ 2600 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2601 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2602 tp->rcv_scale = tp->request_r_scale; 2603 /* Send window already scaled. */ 2604 } 2605 } 2606 2607 process_ACK: 2608 INP_WLOCK_ASSERT(tp->t_inpcb); 2609 2610 acked = BYTES_THIS_ACK(tp, th); 2611 TCPSTAT_INC(tcps_rcvackpack); 2612 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2613 2614 /* 2615 * If we just performed our first retransmit, and the ACK 2616 * arrives within our recovery window, then it was a mistake 2617 * to do the retransmit in the first place. Recover our 2618 * original cwnd and ssthresh, and proceed to transmit where 2619 * we left off. 2620 */ 2621 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2622 (int)(ticks - tp->t_badrxtwin) < 0) 2623 cc_cong_signal(tp, th, CC_RTO_ERR); 2624 2625 /* 2626 * If we have a timestamp reply, update smoothed 2627 * round trip time. If no timestamp is present but 2628 * transmit timer is running and timed sequence 2629 * number was acked, update smoothed round trip time. 2630 * Since we now have an rtt measurement, cancel the 2631 * timer backoff (cf., Phil Karn's retransmit alg.). 2632 * Recompute the initial retransmit timer. 2633 * 2634 * Some boxes send broken timestamp replies 2635 * during the SYN+ACK phase, ignore 2636 * timestamps of 0 or we could calculate a 2637 * huge RTT and blow up the retransmit timer. 2638 */ 2639 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2640 u_int t; 2641 2642 t = tcp_ts_getticks() - to.to_tsecr; 2643 if (!tp->t_rttlow || tp->t_rttlow > t) 2644 tp->t_rttlow = t; 2645 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2646 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2647 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2648 tp->t_rttlow = ticks - tp->t_rtttime; 2649 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2650 } 2651 2652 /* 2653 * If all outstanding data is acked, stop retransmit 2654 * timer and remember to restart (more output or persist). 2655 * If there is more data to be acked, restart retransmit 2656 * timer, using current (possibly backed-off) value. 2657 */ 2658 if (th->th_ack == tp->snd_max) { 2659 tcp_timer_activate(tp, TT_REXMT, 0); 2660 needoutput = 1; 2661 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2662 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2663 2664 /* 2665 * If no data (only SYN) was ACK'd, 2666 * skip rest of ACK processing. 2667 */ 2668 if (acked == 0) 2669 goto step6; 2670 2671 /* 2672 * Let the congestion control algorithm update congestion 2673 * control related information. This typically means increasing 2674 * the congestion window. 2675 */ 2676 cc_ack_received(tp, th, CC_ACK); 2677 2678 SOCKBUF_LOCK(&so->so_snd); 2679 if (acked > so->so_snd.sb_cc) { 2680 tp->snd_wnd -= so->so_snd.sb_cc; 2681 mfree = sbcut_locked(&so->so_snd, 2682 (int)so->so_snd.sb_cc); 2683 ourfinisacked = 1; 2684 } else { 2685 mfree = sbcut_locked(&so->so_snd, acked); 2686 tp->snd_wnd -= acked; 2687 ourfinisacked = 0; 2688 } 2689 /* NB: sowwakeup_locked() does an implicit unlock. */ 2690 sowwakeup_locked(so); 2691 m_freem(mfree); 2692 /* Detect una wraparound. */ 2693 if (!IN_RECOVERY(tp->t_flags) && 2694 SEQ_GT(tp->snd_una, tp->snd_recover) && 2695 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2696 tp->snd_recover = th->th_ack - 1; 2697 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2698 if (IN_RECOVERY(tp->t_flags) && 2699 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2700 EXIT_RECOVERY(tp->t_flags); 2701 } 2702 tp->snd_una = th->th_ack; 2703 if (tp->t_flags & TF_SACK_PERMIT) { 2704 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2705 tp->snd_recover = tp->snd_una; 2706 } 2707 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2708 tp->snd_nxt = tp->snd_una; 2709 2710 switch (tp->t_state) { 2711 2712 /* 2713 * In FIN_WAIT_1 STATE in addition to the processing 2714 * for the ESTABLISHED state if our FIN is now acknowledged 2715 * then enter FIN_WAIT_2. 2716 */ 2717 case TCPS_FIN_WAIT_1: 2718 if (ourfinisacked) { 2719 /* 2720 * If we can't receive any more 2721 * data, then closing user can proceed. 2722 * Starting the timer is contrary to the 2723 * specification, but if we don't get a FIN 2724 * we'll hang forever. 2725 * 2726 * XXXjl: 2727 * we should release the tp also, and use a 2728 * compressed state. 2729 */ 2730 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2731 soisdisconnected(so); 2732 tcp_timer_activate(tp, TT_2MSL, 2733 (tcp_fast_finwait2_recycle ? 2734 tcp_finwait2_timeout : 2735 TP_MAXIDLE(tp))); 2736 } 2737 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2738 } 2739 break; 2740 2741 /* 2742 * In CLOSING STATE in addition to the processing for 2743 * the ESTABLISHED state if the ACK acknowledges our FIN 2744 * then enter the TIME-WAIT state, otherwise ignore 2745 * the segment. 2746 */ 2747 case TCPS_CLOSING: 2748 if (ourfinisacked) { 2749 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2750 tcp_twstart(tp); 2751 INP_INFO_WUNLOCK(&V_tcbinfo); 2752 m_freem(m); 2753 return; 2754 } 2755 break; 2756 2757 /* 2758 * In LAST_ACK, we may still be waiting for data to drain 2759 * and/or to be acked, as well as for the ack of our FIN. 2760 * If our FIN is now acknowledged, delete the TCB, 2761 * enter the closed state and return. 2762 */ 2763 case TCPS_LAST_ACK: 2764 if (ourfinisacked) { 2765 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2766 tp = tcp_close(tp); 2767 goto drop; 2768 } 2769 break; 2770 } 2771 } 2772 2773 step6: 2774 INP_WLOCK_ASSERT(tp->t_inpcb); 2775 2776 /* 2777 * Update window information. 2778 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2779 */ 2780 if ((thflags & TH_ACK) && 2781 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2782 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2783 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2784 /* keep track of pure window updates */ 2785 if (tlen == 0 && 2786 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2787 TCPSTAT_INC(tcps_rcvwinupd); 2788 tp->snd_wnd = tiwin; 2789 tp->snd_wl1 = th->th_seq; 2790 tp->snd_wl2 = th->th_ack; 2791 if (tp->snd_wnd > tp->max_sndwnd) 2792 tp->max_sndwnd = tp->snd_wnd; 2793 needoutput = 1; 2794 } 2795 2796 /* 2797 * Process segments with URG. 2798 */ 2799 if ((thflags & TH_URG) && th->th_urp && 2800 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2801 /* 2802 * This is a kludge, but if we receive and accept 2803 * random urgent pointers, we'll crash in 2804 * soreceive. It's hard to imagine someone 2805 * actually wanting to send this much urgent data. 2806 */ 2807 SOCKBUF_LOCK(&so->so_rcv); 2808 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2809 th->th_urp = 0; /* XXX */ 2810 thflags &= ~TH_URG; /* XXX */ 2811 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2812 goto dodata; /* XXX */ 2813 } 2814 /* 2815 * If this segment advances the known urgent pointer, 2816 * then mark the data stream. This should not happen 2817 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2818 * a FIN has been received from the remote side. 2819 * In these states we ignore the URG. 2820 * 2821 * According to RFC961 (Assigned Protocols), 2822 * the urgent pointer points to the last octet 2823 * of urgent data. We continue, however, 2824 * to consider it to indicate the first octet 2825 * of data past the urgent section as the original 2826 * spec states (in one of two places). 2827 */ 2828 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2829 tp->rcv_up = th->th_seq + th->th_urp; 2830 so->so_oobmark = so->so_rcv.sb_cc + 2831 (tp->rcv_up - tp->rcv_nxt) - 1; 2832 if (so->so_oobmark == 0) 2833 so->so_rcv.sb_state |= SBS_RCVATMARK; 2834 sohasoutofband(so); 2835 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2836 } 2837 SOCKBUF_UNLOCK(&so->so_rcv); 2838 /* 2839 * Remove out of band data so doesn't get presented to user. 2840 * This can happen independent of advancing the URG pointer, 2841 * but if two URG's are pending at once, some out-of-band 2842 * data may creep in... ick. 2843 */ 2844 if (th->th_urp <= (u_long)tlen && 2845 !(so->so_options & SO_OOBINLINE)) { 2846 /* hdr drop is delayed */ 2847 tcp_pulloutofband(so, th, m, drop_hdrlen); 2848 } 2849 } else { 2850 /* 2851 * If no out of band data is expected, 2852 * pull receive urgent pointer along 2853 * with the receive window. 2854 */ 2855 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2856 tp->rcv_up = tp->rcv_nxt; 2857 } 2858 dodata: /* XXX */ 2859 INP_WLOCK_ASSERT(tp->t_inpcb); 2860 2861 /* 2862 * Process the segment text, merging it into the TCP sequencing queue, 2863 * and arranging for acknowledgment of receipt if necessary. 2864 * This process logically involves adjusting tp->rcv_wnd as data 2865 * is presented to the user (this happens in tcp_usrreq.c, 2866 * case PRU_RCVD). If a FIN has already been received on this 2867 * connection then we just ignore the text. 2868 */ 2869 if ((tlen || (thflags & TH_FIN)) && 2870 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2871 tcp_seq save_start = th->th_seq; 2872 m_adj(m, drop_hdrlen); /* delayed header drop */ 2873 /* 2874 * Insert segment which includes th into TCP reassembly queue 2875 * with control block tp. Set thflags to whether reassembly now 2876 * includes a segment with FIN. This handles the common case 2877 * inline (segment is the next to be received on an established 2878 * connection, and the queue is empty), avoiding linkage into 2879 * and removal from the queue and repetition of various 2880 * conversions. 2881 * Set DELACK for segments received in order, but ack 2882 * immediately when segments are out of order (so 2883 * fast retransmit can work). 2884 */ 2885 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL && 2886 TCPS_HAVEESTABLISHED(tp->t_state)) { 2887 if (DELAY_ACK(tp, tlen)) 2888 tp->t_flags |= TF_DELACK; 2889 else 2890 tp->t_flags |= TF_ACKNOW; 2891 tp->rcv_nxt += tlen; 2892 thflags = th->th_flags & TH_FIN; 2893 TCPSTAT_INC(tcps_rcvpack); 2894 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2895 ND6_HINT(tp); 2896 SOCKBUF_LOCK(&so->so_rcv); 2897 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2898 m_freem(m); 2899 else 2900 sbappendstream_locked(&so->so_rcv, m); 2901 /* NB: sorwakeup_locked() does an implicit unlock. */ 2902 sorwakeup_locked(so); 2903 } else { 2904 /* 2905 * XXX: Due to the header drop above "th" is 2906 * theoretically invalid by now. Fortunately 2907 * m_adj() doesn't actually frees any mbufs 2908 * when trimming from the head. 2909 */ 2910 thflags = tcp_reass(tp, th, &tlen, m); 2911 tp->t_flags |= TF_ACKNOW; 2912 } 2913 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2914 tcp_update_sack_list(tp, save_start, save_start + tlen); 2915 #if 0 2916 /* 2917 * Note the amount of data that peer has sent into 2918 * our window, in order to estimate the sender's 2919 * buffer size. 2920 * XXX: Unused. 2921 */ 2922 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2923 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2924 else 2925 len = so->so_rcv.sb_hiwat; 2926 #endif 2927 } else { 2928 m_freem(m); 2929 thflags &= ~TH_FIN; 2930 } 2931 2932 /* 2933 * If FIN is received ACK the FIN and let the user know 2934 * that the connection is closing. 2935 */ 2936 if (thflags & TH_FIN) { 2937 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2938 socantrcvmore(so); 2939 /* 2940 * If connection is half-synchronized 2941 * (ie NEEDSYN flag on) then delay ACK, 2942 * so it may be piggybacked when SYN is sent. 2943 * Otherwise, since we received a FIN then no 2944 * more input can be expected, send ACK now. 2945 */ 2946 if (tp->t_flags & TF_NEEDSYN) 2947 tp->t_flags |= TF_DELACK; 2948 else 2949 tp->t_flags |= TF_ACKNOW; 2950 tp->rcv_nxt++; 2951 } 2952 switch (tp->t_state) { 2953 2954 /* 2955 * In SYN_RECEIVED and ESTABLISHED STATES 2956 * enter the CLOSE_WAIT state. 2957 */ 2958 case TCPS_SYN_RECEIVED: 2959 tp->t_starttime = ticks; 2960 /* FALLTHROUGH */ 2961 case TCPS_ESTABLISHED: 2962 tcp_state_change(tp, TCPS_CLOSE_WAIT); 2963 break; 2964 2965 /* 2966 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2967 * enter the CLOSING state. 2968 */ 2969 case TCPS_FIN_WAIT_1: 2970 tcp_state_change(tp, TCPS_CLOSING); 2971 break; 2972 2973 /* 2974 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2975 * starting the time-wait timer, turning off the other 2976 * standard timers. 2977 */ 2978 case TCPS_FIN_WAIT_2: 2979 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2980 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2981 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2982 ti_locked)); 2983 2984 tcp_twstart(tp); 2985 INP_INFO_WUNLOCK(&V_tcbinfo); 2986 return; 2987 } 2988 } 2989 if (ti_locked == TI_WLOCKED) 2990 INP_INFO_WUNLOCK(&V_tcbinfo); 2991 ti_locked = TI_UNLOCKED; 2992 2993 #ifdef TCPDEBUG 2994 if (so->so_options & SO_DEBUG) 2995 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2996 &tcp_savetcp, 0); 2997 #endif 2998 2999 /* 3000 * Return any desired output. 3001 */ 3002 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3003 (void) tcp_output(tp); 3004 3005 check_delack: 3006 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3007 __func__, ti_locked)); 3008 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3009 INP_WLOCK_ASSERT(tp->t_inpcb); 3010 3011 if (tp->t_flags & TF_DELACK) { 3012 tp->t_flags &= ~TF_DELACK; 3013 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3014 } 3015 INP_WUNLOCK(tp->t_inpcb); 3016 return; 3017 3018 dropafterack: 3019 /* 3020 * Generate an ACK dropping incoming segment if it occupies 3021 * sequence space, where the ACK reflects our state. 3022 * 3023 * We can now skip the test for the RST flag since all 3024 * paths to this code happen after packets containing 3025 * RST have been dropped. 3026 * 3027 * In the SYN-RECEIVED state, don't send an ACK unless the 3028 * segment we received passes the SYN-RECEIVED ACK test. 3029 * If it fails send a RST. This breaks the loop in the 3030 * "LAND" DoS attack, and also prevents an ACK storm 3031 * between two listening ports that have been sent forged 3032 * SYN segments, each with the source address of the other. 3033 */ 3034 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3035 (SEQ_GT(tp->snd_una, th->th_ack) || 3036 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3037 rstreason = BANDLIM_RST_OPENPORT; 3038 goto dropwithreset; 3039 } 3040 #ifdef TCPDEBUG 3041 if (so->so_options & SO_DEBUG) 3042 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3043 &tcp_savetcp, 0); 3044 #endif 3045 if (ti_locked == TI_WLOCKED) 3046 INP_INFO_WUNLOCK(&V_tcbinfo); 3047 ti_locked = TI_UNLOCKED; 3048 3049 tp->t_flags |= TF_ACKNOW; 3050 (void) tcp_output(tp); 3051 INP_WUNLOCK(tp->t_inpcb); 3052 m_freem(m); 3053 return; 3054 3055 dropwithreset: 3056 if (ti_locked == TI_WLOCKED) 3057 INP_INFO_WUNLOCK(&V_tcbinfo); 3058 ti_locked = TI_UNLOCKED; 3059 3060 if (tp != NULL) { 3061 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3062 INP_WUNLOCK(tp->t_inpcb); 3063 } else 3064 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3065 return; 3066 3067 drop: 3068 if (ti_locked == TI_WLOCKED) { 3069 INP_INFO_WUNLOCK(&V_tcbinfo); 3070 ti_locked = TI_UNLOCKED; 3071 } 3072 #ifdef INVARIANTS 3073 else 3074 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3075 #endif 3076 3077 /* 3078 * Drop space held by incoming segment and return. 3079 */ 3080 #ifdef TCPDEBUG 3081 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3082 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3083 &tcp_savetcp, 0); 3084 #endif 3085 if (tp != NULL) 3086 INP_WUNLOCK(tp->t_inpcb); 3087 m_freem(m); 3088 } 3089 3090 /* 3091 * Issue RST and make ACK acceptable to originator of segment. 3092 * The mbuf must still include the original packet header. 3093 * tp may be NULL. 3094 */ 3095 static void 3096 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3097 int tlen, int rstreason) 3098 { 3099 #ifdef INET 3100 struct ip *ip; 3101 #endif 3102 #ifdef INET6 3103 struct ip6_hdr *ip6; 3104 #endif 3105 3106 if (tp != NULL) { 3107 INP_WLOCK_ASSERT(tp->t_inpcb); 3108 } 3109 3110 /* Don't bother if destination was broadcast/multicast. */ 3111 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3112 goto drop; 3113 #ifdef INET6 3114 if (mtod(m, struct ip *)->ip_v == 6) { 3115 ip6 = mtod(m, struct ip6_hdr *); 3116 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3117 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3118 goto drop; 3119 /* IPv6 anycast check is done at tcp6_input() */ 3120 } 3121 #endif 3122 #if defined(INET) && defined(INET6) 3123 else 3124 #endif 3125 #ifdef INET 3126 { 3127 ip = mtod(m, struct ip *); 3128 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3129 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3130 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3131 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3132 goto drop; 3133 } 3134 #endif 3135 3136 /* Perform bandwidth limiting. */ 3137 if (badport_bandlim(rstreason) < 0) 3138 goto drop; 3139 3140 /* tcp_respond consumes the mbuf chain. */ 3141 if (th->th_flags & TH_ACK) { 3142 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3143 th->th_ack, TH_RST); 3144 } else { 3145 if (th->th_flags & TH_SYN) 3146 tlen++; 3147 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3148 (tcp_seq)0, TH_RST|TH_ACK); 3149 } 3150 return; 3151 drop: 3152 m_freem(m); 3153 } 3154 3155 /* 3156 * Parse TCP options and place in tcpopt. 3157 */ 3158 static void 3159 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3160 { 3161 int opt, optlen; 3162 3163 to->to_flags = 0; 3164 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3165 opt = cp[0]; 3166 if (opt == TCPOPT_EOL) 3167 break; 3168 if (opt == TCPOPT_NOP) 3169 optlen = 1; 3170 else { 3171 if (cnt < 2) 3172 break; 3173 optlen = cp[1]; 3174 if (optlen < 2 || optlen > cnt) 3175 break; 3176 } 3177 switch (opt) { 3178 case TCPOPT_MAXSEG: 3179 if (optlen != TCPOLEN_MAXSEG) 3180 continue; 3181 if (!(flags & TO_SYN)) 3182 continue; 3183 to->to_flags |= TOF_MSS; 3184 bcopy((char *)cp + 2, 3185 (char *)&to->to_mss, sizeof(to->to_mss)); 3186 to->to_mss = ntohs(to->to_mss); 3187 break; 3188 case TCPOPT_WINDOW: 3189 if (optlen != TCPOLEN_WINDOW) 3190 continue; 3191 if (!(flags & TO_SYN)) 3192 continue; 3193 to->to_flags |= TOF_SCALE; 3194 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3195 break; 3196 case TCPOPT_TIMESTAMP: 3197 if (optlen != TCPOLEN_TIMESTAMP) 3198 continue; 3199 to->to_flags |= TOF_TS; 3200 bcopy((char *)cp + 2, 3201 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3202 to->to_tsval = ntohl(to->to_tsval); 3203 bcopy((char *)cp + 6, 3204 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3205 to->to_tsecr = ntohl(to->to_tsecr); 3206 break; 3207 #ifdef TCP_SIGNATURE 3208 /* 3209 * XXX In order to reply to a host which has set the 3210 * TCP_SIGNATURE option in its initial SYN, we have to 3211 * record the fact that the option was observed here 3212 * for the syncache code to perform the correct response. 3213 */ 3214 case TCPOPT_SIGNATURE: 3215 if (optlen != TCPOLEN_SIGNATURE) 3216 continue; 3217 to->to_flags |= TOF_SIGNATURE; 3218 to->to_signature = cp + 2; 3219 break; 3220 #endif 3221 case TCPOPT_SACK_PERMITTED: 3222 if (optlen != TCPOLEN_SACK_PERMITTED) 3223 continue; 3224 if (!(flags & TO_SYN)) 3225 continue; 3226 if (!V_tcp_do_sack) 3227 continue; 3228 to->to_flags |= TOF_SACKPERM; 3229 break; 3230 case TCPOPT_SACK: 3231 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3232 continue; 3233 if (flags & TO_SYN) 3234 continue; 3235 to->to_flags |= TOF_SACK; 3236 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3237 to->to_sacks = cp + 2; 3238 TCPSTAT_INC(tcps_sack_rcv_blocks); 3239 break; 3240 default: 3241 continue; 3242 } 3243 } 3244 } 3245 3246 /* 3247 * Pull out of band byte out of a segment so 3248 * it doesn't appear in the user's data queue. 3249 * It is still reflected in the segment length for 3250 * sequencing purposes. 3251 */ 3252 static void 3253 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3254 int off) 3255 { 3256 int cnt = off + th->th_urp - 1; 3257 3258 while (cnt >= 0) { 3259 if (m->m_len > cnt) { 3260 char *cp = mtod(m, caddr_t) + cnt; 3261 struct tcpcb *tp = sototcpcb(so); 3262 3263 INP_WLOCK_ASSERT(tp->t_inpcb); 3264 3265 tp->t_iobc = *cp; 3266 tp->t_oobflags |= TCPOOB_HAVEDATA; 3267 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3268 m->m_len--; 3269 if (m->m_flags & M_PKTHDR) 3270 m->m_pkthdr.len--; 3271 return; 3272 } 3273 cnt -= m->m_len; 3274 m = m->m_next; 3275 if (m == NULL) 3276 break; 3277 } 3278 panic("tcp_pulloutofband"); 3279 } 3280 3281 /* 3282 * Collect new round-trip time estimate 3283 * and update averages and current timeout. 3284 */ 3285 static void 3286 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3287 { 3288 int delta; 3289 3290 INP_WLOCK_ASSERT(tp->t_inpcb); 3291 3292 TCPSTAT_INC(tcps_rttupdated); 3293 tp->t_rttupdated++; 3294 if (tp->t_srtt != 0) { 3295 /* 3296 * srtt is stored as fixed point with 5 bits after the 3297 * binary point (i.e., scaled by 8). The following magic 3298 * is equivalent to the smoothing algorithm in rfc793 with 3299 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3300 * point). Adjust rtt to origin 0. 3301 */ 3302 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3303 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3304 3305 if ((tp->t_srtt += delta) <= 0) 3306 tp->t_srtt = 1; 3307 3308 /* 3309 * We accumulate a smoothed rtt variance (actually, a 3310 * smoothed mean difference), then set the retransmit 3311 * timer to smoothed rtt + 4 times the smoothed variance. 3312 * rttvar is stored as fixed point with 4 bits after the 3313 * binary point (scaled by 16). The following is 3314 * equivalent to rfc793 smoothing with an alpha of .75 3315 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3316 * rfc793's wired-in beta. 3317 */ 3318 if (delta < 0) 3319 delta = -delta; 3320 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3321 if ((tp->t_rttvar += delta) <= 0) 3322 tp->t_rttvar = 1; 3323 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3324 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3325 } else { 3326 /* 3327 * No rtt measurement yet - use the unsmoothed rtt. 3328 * Set the variance to half the rtt (so our first 3329 * retransmit happens at 3*rtt). 3330 */ 3331 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3332 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3333 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3334 } 3335 tp->t_rtttime = 0; 3336 tp->t_rxtshift = 0; 3337 3338 /* 3339 * the retransmit should happen at rtt + 4 * rttvar. 3340 * Because of the way we do the smoothing, srtt and rttvar 3341 * will each average +1/2 tick of bias. When we compute 3342 * the retransmit timer, we want 1/2 tick of rounding and 3343 * 1 extra tick because of +-1/2 tick uncertainty in the 3344 * firing of the timer. The bias will give us exactly the 3345 * 1.5 tick we need. But, because the bias is 3346 * statistical, we have to test that we don't drop below 3347 * the minimum feasible timer (which is 2 ticks). 3348 */ 3349 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3350 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3351 3352 /* 3353 * We received an ack for a packet that wasn't retransmitted; 3354 * it is probably safe to discard any error indications we've 3355 * received recently. This isn't quite right, but close enough 3356 * for now (a route might have failed after we sent a segment, 3357 * and the return path might not be symmetrical). 3358 */ 3359 tp->t_softerror = 0; 3360 } 3361 3362 /* 3363 * Determine a reasonable value for maxseg size. 3364 * If the route is known, check route for mtu. 3365 * If none, use an mss that can be handled on the outgoing interface 3366 * without forcing IP to fragment. If no route is found, route has no mtu, 3367 * or the destination isn't local, use a default, hopefully conservative 3368 * size (usually 512 or the default IP max size, but no more than the mtu 3369 * of the interface), as we can't discover anything about intervening 3370 * gateways or networks. We also initialize the congestion/slow start 3371 * window to be a single segment if the destination isn't local. 3372 * While looking at the routing entry, we also initialize other path-dependent 3373 * parameters from pre-set or cached values in the routing entry. 3374 * 3375 * Also take into account the space needed for options that we 3376 * send regularly. Make maxseg shorter by that amount to assure 3377 * that we can send maxseg amount of data even when the options 3378 * are present. Store the upper limit of the length of options plus 3379 * data in maxopd. 3380 * 3381 * NOTE that this routine is only called when we process an incoming 3382 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3383 * settings are handled in tcp_mssopt(). 3384 */ 3385 void 3386 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3387 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3388 { 3389 int mss = 0; 3390 u_long maxmtu = 0; 3391 struct inpcb *inp = tp->t_inpcb; 3392 struct hc_metrics_lite metrics; 3393 int origoffer; 3394 #ifdef INET6 3395 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3396 size_t min_protoh = isipv6 ? 3397 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3398 sizeof (struct tcpiphdr); 3399 #else 3400 const size_t min_protoh = sizeof(struct tcpiphdr); 3401 #endif 3402 3403 INP_WLOCK_ASSERT(tp->t_inpcb); 3404 3405 if (mtuoffer != -1) { 3406 KASSERT(offer == -1, ("%s: conflict", __func__)); 3407 offer = mtuoffer - min_protoh; 3408 } 3409 origoffer = offer; 3410 3411 /* Initialize. */ 3412 #ifdef INET6 3413 if (isipv6) { 3414 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3415 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3416 } 3417 #endif 3418 #if defined(INET) && defined(INET6) 3419 else 3420 #endif 3421 #ifdef INET 3422 { 3423 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3424 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3425 } 3426 #endif 3427 3428 /* 3429 * No route to sender, stay with default mss and return. 3430 */ 3431 if (maxmtu == 0) { 3432 /* 3433 * In case we return early we need to initialize metrics 3434 * to a defined state as tcp_hc_get() would do for us 3435 * if there was no cache hit. 3436 */ 3437 if (metricptr != NULL) 3438 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3439 return; 3440 } 3441 3442 /* What have we got? */ 3443 switch (offer) { 3444 case 0: 3445 /* 3446 * Offer == 0 means that there was no MSS on the SYN 3447 * segment, in this case we use tcp_mssdflt as 3448 * already assigned to t_maxopd above. 3449 */ 3450 offer = tp->t_maxopd; 3451 break; 3452 3453 case -1: 3454 /* 3455 * Offer == -1 means that we didn't receive SYN yet. 3456 */ 3457 /* FALLTHROUGH */ 3458 3459 default: 3460 /* 3461 * Prevent DoS attack with too small MSS. Round up 3462 * to at least minmss. 3463 */ 3464 offer = max(offer, V_tcp_minmss); 3465 } 3466 3467 /* 3468 * rmx information is now retrieved from tcp_hostcache. 3469 */ 3470 tcp_hc_get(&inp->inp_inc, &metrics); 3471 if (metricptr != NULL) 3472 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3473 3474 /* 3475 * If there's a discovered mtu in tcp hostcache, use it. 3476 * Else, use the link mtu. 3477 */ 3478 if (metrics.rmx_mtu) 3479 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3480 else { 3481 #ifdef INET6 3482 if (isipv6) { 3483 mss = maxmtu - min_protoh; 3484 if (!V_path_mtu_discovery && 3485 !in6_localaddr(&inp->in6p_faddr)) 3486 mss = min(mss, V_tcp_v6mssdflt); 3487 } 3488 #endif 3489 #if defined(INET) && defined(INET6) 3490 else 3491 #endif 3492 #ifdef INET 3493 { 3494 mss = maxmtu - min_protoh; 3495 if (!V_path_mtu_discovery && 3496 !in_localaddr(inp->inp_faddr)) 3497 mss = min(mss, V_tcp_mssdflt); 3498 } 3499 #endif 3500 /* 3501 * XXX - The above conditional (mss = maxmtu - min_protoh) 3502 * probably violates the TCP spec. 3503 * The problem is that, since we don't know the 3504 * other end's MSS, we are supposed to use a conservative 3505 * default. But, if we do that, then MTU discovery will 3506 * never actually take place, because the conservative 3507 * default is much less than the MTUs typically seen 3508 * on the Internet today. For the moment, we'll sweep 3509 * this under the carpet. 3510 * 3511 * The conservative default might not actually be a problem 3512 * if the only case this occurs is when sending an initial 3513 * SYN with options and data to a host we've never talked 3514 * to before. Then, they will reply with an MSS value which 3515 * will get recorded and the new parameters should get 3516 * recomputed. For Further Study. 3517 */ 3518 } 3519 mss = min(mss, offer); 3520 3521 /* 3522 * Sanity check: make sure that maxopd will be large 3523 * enough to allow some data on segments even if the 3524 * all the option space is used (40bytes). Otherwise 3525 * funny things may happen in tcp_output. 3526 */ 3527 mss = max(mss, 64); 3528 3529 /* 3530 * maxopd stores the maximum length of data AND options 3531 * in a segment; maxseg is the amount of data in a normal 3532 * segment. We need to store this value (maxopd) apart 3533 * from maxseg, because now every segment carries options 3534 * and thus we normally have somewhat less data in segments. 3535 */ 3536 tp->t_maxopd = mss; 3537 3538 /* 3539 * origoffer==-1 indicates that no segments were received yet. 3540 * In this case we just guess. 3541 */ 3542 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3543 (origoffer == -1 || 3544 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3545 mss -= TCPOLEN_TSTAMP_APPA; 3546 3547 tp->t_maxseg = mss; 3548 } 3549 3550 void 3551 tcp_mss(struct tcpcb *tp, int offer) 3552 { 3553 int mss; 3554 u_long bufsize; 3555 struct inpcb *inp; 3556 struct socket *so; 3557 struct hc_metrics_lite metrics; 3558 struct tcp_ifcap cap; 3559 3560 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3561 3562 bzero(&cap, sizeof(cap)); 3563 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3564 3565 mss = tp->t_maxseg; 3566 inp = tp->t_inpcb; 3567 3568 /* 3569 * If there's a pipesize, change the socket buffer to that size, 3570 * don't change if sb_hiwat is different than default (then it 3571 * has been changed on purpose with setsockopt). 3572 * Make the socket buffers an integral number of mss units; 3573 * if the mss is larger than the socket buffer, decrease the mss. 3574 */ 3575 so = inp->inp_socket; 3576 SOCKBUF_LOCK(&so->so_snd); 3577 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3578 bufsize = metrics.rmx_sendpipe; 3579 else 3580 bufsize = so->so_snd.sb_hiwat; 3581 if (bufsize < mss) 3582 mss = bufsize; 3583 else { 3584 bufsize = roundup(bufsize, mss); 3585 if (bufsize > sb_max) 3586 bufsize = sb_max; 3587 if (bufsize > so->so_snd.sb_hiwat) 3588 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3589 } 3590 SOCKBUF_UNLOCK(&so->so_snd); 3591 tp->t_maxseg = mss; 3592 3593 SOCKBUF_LOCK(&so->so_rcv); 3594 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3595 bufsize = metrics.rmx_recvpipe; 3596 else 3597 bufsize = so->so_rcv.sb_hiwat; 3598 if (bufsize > mss) { 3599 bufsize = roundup(bufsize, mss); 3600 if (bufsize > sb_max) 3601 bufsize = sb_max; 3602 if (bufsize > so->so_rcv.sb_hiwat) 3603 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3604 } 3605 SOCKBUF_UNLOCK(&so->so_rcv); 3606 3607 /* Check the interface for TSO capabilities. */ 3608 if (cap.ifcap & CSUM_TSO) { 3609 tp->t_flags |= TF_TSO; 3610 tp->t_tsomax = cap.tsomax; 3611 } 3612 } 3613 3614 /* 3615 * Determine the MSS option to send on an outgoing SYN. 3616 */ 3617 int 3618 tcp_mssopt(struct in_conninfo *inc) 3619 { 3620 int mss = 0; 3621 u_long maxmtu = 0; 3622 u_long thcmtu = 0; 3623 size_t min_protoh; 3624 3625 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3626 3627 #ifdef INET6 3628 if (inc->inc_flags & INC_ISIPV6) { 3629 mss = V_tcp_v6mssdflt; 3630 maxmtu = tcp_maxmtu6(inc, NULL); 3631 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3632 } 3633 #endif 3634 #if defined(INET) && defined(INET6) 3635 else 3636 #endif 3637 #ifdef INET 3638 { 3639 mss = V_tcp_mssdflt; 3640 maxmtu = tcp_maxmtu(inc, NULL); 3641 min_protoh = sizeof(struct tcpiphdr); 3642 } 3643 #endif 3644 #if defined(INET6) || defined(INET) 3645 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3646 #endif 3647 3648 if (maxmtu && thcmtu) 3649 mss = min(maxmtu, thcmtu) - min_protoh; 3650 else if (maxmtu || thcmtu) 3651 mss = max(maxmtu, thcmtu) - min_protoh; 3652 3653 return (mss); 3654 } 3655 3656 3657 /* 3658 * On a partial ack arrives, force the retransmission of the 3659 * next unacknowledged segment. Do not clear tp->t_dupacks. 3660 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3661 * be started again. 3662 */ 3663 static void 3664 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3665 { 3666 tcp_seq onxt = tp->snd_nxt; 3667 u_long ocwnd = tp->snd_cwnd; 3668 3669 INP_WLOCK_ASSERT(tp->t_inpcb); 3670 3671 tcp_timer_activate(tp, TT_REXMT, 0); 3672 tp->t_rtttime = 0; 3673 tp->snd_nxt = th->th_ack; 3674 /* 3675 * Set snd_cwnd to one segment beyond acknowledged offset. 3676 * (tp->snd_una has not yet been updated when this function is called.) 3677 */ 3678 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3679 tp->t_flags |= TF_ACKNOW; 3680 (void) tcp_output(tp); 3681 tp->snd_cwnd = ocwnd; 3682 if (SEQ_GT(onxt, tp->snd_nxt)) 3683 tp->snd_nxt = onxt; 3684 /* 3685 * Partial window deflation. Relies on fact that tp->snd_una 3686 * not updated yet. 3687 */ 3688 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3689 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3690 else 3691 tp->snd_cwnd = 0; 3692 tp->snd_cwnd += tp->t_maxseg; 3693 } 3694