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_syn) = 0; 190 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 191 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_RW, 192 &VNET_NAME(tcp_insecure_syn), 0, 193 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 194 195 VNET_DEFINE(int, tcp_insecure_rst) = 0; 196 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 197 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 198 &VNET_NAME(tcp_insecure_rst), 0, 199 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 200 201 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 202 #define V_tcp_recvspace VNET(tcp_recvspace) 203 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 204 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 205 206 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 207 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 208 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 209 &VNET_NAME(tcp_do_autorcvbuf), 0, 210 "Enable automatic receive buffer sizing"); 211 212 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 213 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 214 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 215 &VNET_NAME(tcp_autorcvbuf_inc), 0, 216 "Incrementor step size of automatic receive buffer"); 217 218 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 219 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 220 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 221 &VNET_NAME(tcp_autorcvbuf_max), 0, 222 "Max size of automatic receive buffer"); 223 224 VNET_DEFINE(struct inpcbhead, tcb); 225 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 226 VNET_DEFINE(struct inpcbinfo, tcbinfo); 227 228 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 229 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 230 struct socket *, struct tcpcb *, int, int, uint8_t, 231 int); 232 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 233 struct tcpcb *, int, int); 234 static void tcp_pulloutofband(struct socket *, 235 struct tcphdr *, struct mbuf *, int); 236 static void tcp_xmit_timer(struct tcpcb *, int); 237 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 238 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 239 uint16_t type); 240 static void inline cc_conn_init(struct tcpcb *tp); 241 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 242 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 243 struct tcphdr *th, struct tcpopt *to); 244 245 /* 246 * TCP statistics are stored in an "array" of counter(9)s. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 VNET_PCPUSTAT_SYSINIT(tcpstat); 250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252 253 #ifdef VIMAGE 254 VNET_PCPUSTAT_SYSUNINIT(tcpstat); 255 #endif /* VIMAGE */ 256 /* 257 * Kernel module interface for updating tcpstat. The argument is an index 258 * into tcpstat treated as an array. 259 */ 260 void 261 kmod_tcpstat_inc(int statnum) 262 { 263 264 counter_u64_add(VNET(tcpstat)[statnum], 1); 265 } 266 267 /* 268 * Wrapper for the TCP established input helper hook. 269 */ 270 static void inline 271 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 272 { 273 struct tcp_hhook_data hhook_data; 274 275 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 276 hhook_data.tp = tp; 277 hhook_data.th = th; 278 hhook_data.to = to; 279 280 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 281 tp->osd); 282 } 283 } 284 285 /* 286 * CC wrapper hook functions 287 */ 288 static void inline 289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 290 { 291 INP_WLOCK_ASSERT(tp->t_inpcb); 292 293 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 294 if (tp->snd_cwnd <= tp->snd_wnd) 295 tp->ccv->flags |= CCF_CWND_LIMITED; 296 else 297 tp->ccv->flags &= ~CCF_CWND_LIMITED; 298 299 if (type == CC_ACK) { 300 if (tp->snd_cwnd > tp->snd_ssthresh) { 301 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 302 V_tcp_abc_l_var * tp->t_maxseg); 303 if (tp->t_bytes_acked >= tp->snd_cwnd) { 304 tp->t_bytes_acked -= tp->snd_cwnd; 305 tp->ccv->flags |= CCF_ABC_SENTAWND; 306 } 307 } else { 308 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 309 tp->t_bytes_acked = 0; 310 } 311 } 312 313 if (CC_ALGO(tp)->ack_received != NULL) { 314 /* XXXLAS: Find a way to live without this */ 315 tp->ccv->curack = th->th_ack; 316 CC_ALGO(tp)->ack_received(tp->ccv, type); 317 } 318 } 319 320 static void inline 321 cc_conn_init(struct tcpcb *tp) 322 { 323 struct hc_metrics_lite metrics; 324 struct inpcb *inp = tp->t_inpcb; 325 int rtt; 326 327 INP_WLOCK_ASSERT(tp->t_inpcb); 328 329 tcp_hc_get(&inp->inp_inc, &metrics); 330 331 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 332 tp->t_srtt = rtt; 333 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 334 TCPSTAT_INC(tcps_usedrtt); 335 if (metrics.rmx_rttvar) { 336 tp->t_rttvar = metrics.rmx_rttvar; 337 TCPSTAT_INC(tcps_usedrttvar); 338 } else { 339 /* default variation is +- 1 rtt */ 340 tp->t_rttvar = 341 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 342 } 343 TCPT_RANGESET(tp->t_rxtcur, 344 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 345 tp->t_rttmin, TCPTV_REXMTMAX); 346 } 347 if (metrics.rmx_ssthresh) { 348 /* 349 * There's some sort of gateway or interface 350 * buffer limit on the path. Use this to set 351 * the slow start threshhold, but set the 352 * threshold to no less than 2*mss. 353 */ 354 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 355 TCPSTAT_INC(tcps_usedssthresh); 356 } 357 358 /* 359 * Set the initial slow-start flight size. 360 * 361 * RFC5681 Section 3.1 specifies the default conservative values. 362 * RFC3390 specifies slightly more aggressive values. 363 * RFC6928 increases it to ten segments. 364 * 365 * If a SYN or SYN/ACK was lost and retransmitted, we have to 366 * reduce the initial CWND to one segment as congestion is likely 367 * requiring us to be cautious. 368 */ 369 if (tp->snd_cwnd == 1) 370 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 371 else if (V_tcp_do_initcwnd10) 372 tp->snd_cwnd = min(10 * tp->t_maxseg, 373 max(2 * tp->t_maxseg, 14600)); 374 else if (V_tcp_do_rfc3390) 375 tp->snd_cwnd = min(4 * tp->t_maxseg, 376 max(2 * tp->t_maxseg, 4380)); 377 else { 378 /* Per RFC5681 Section 3.1 */ 379 if (tp->t_maxseg > 2190) 380 tp->snd_cwnd = 2 * tp->t_maxseg; 381 else if (tp->t_maxseg > 1095) 382 tp->snd_cwnd = 3 * tp->t_maxseg; 383 else 384 tp->snd_cwnd = 4 * tp->t_maxseg; 385 } 386 387 if (CC_ALGO(tp)->conn_init != NULL) 388 CC_ALGO(tp)->conn_init(tp->ccv); 389 } 390 391 void inline 392 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 393 { 394 INP_WLOCK_ASSERT(tp->t_inpcb); 395 396 switch(type) { 397 case CC_NDUPACK: 398 if (!IN_FASTRECOVERY(tp->t_flags)) { 399 tp->snd_recover = tp->snd_max; 400 if (tp->t_flags & TF_ECN_PERMIT) 401 tp->t_flags |= TF_ECN_SND_CWR; 402 } 403 break; 404 case CC_ECN: 405 if (!IN_CONGRECOVERY(tp->t_flags)) { 406 TCPSTAT_INC(tcps_ecn_rcwnd); 407 tp->snd_recover = tp->snd_max; 408 if (tp->t_flags & TF_ECN_PERMIT) 409 tp->t_flags |= TF_ECN_SND_CWR; 410 } 411 break; 412 case CC_RTO: 413 tp->t_dupacks = 0; 414 tp->t_bytes_acked = 0; 415 EXIT_RECOVERY(tp->t_flags); 416 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 417 tp->t_maxseg) * tp->t_maxseg; 418 tp->snd_cwnd = tp->t_maxseg; 419 break; 420 case CC_RTO_ERR: 421 TCPSTAT_INC(tcps_sndrexmitbad); 422 /* RTO was unnecessary, so reset everything. */ 423 tp->snd_cwnd = tp->snd_cwnd_prev; 424 tp->snd_ssthresh = tp->snd_ssthresh_prev; 425 tp->snd_recover = tp->snd_recover_prev; 426 if (tp->t_flags & TF_WASFRECOVERY) 427 ENTER_FASTRECOVERY(tp->t_flags); 428 if (tp->t_flags & TF_WASCRECOVERY) 429 ENTER_CONGRECOVERY(tp->t_flags); 430 tp->snd_nxt = tp->snd_max; 431 tp->t_flags &= ~TF_PREVVALID; 432 tp->t_badrxtwin = 0; 433 break; 434 } 435 436 if (CC_ALGO(tp)->cong_signal != NULL) { 437 if (th != NULL) 438 tp->ccv->curack = th->th_ack; 439 CC_ALGO(tp)->cong_signal(tp->ccv, type); 440 } 441 } 442 443 static void inline 444 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 445 { 446 INP_WLOCK_ASSERT(tp->t_inpcb); 447 448 /* XXXLAS: KASSERT that we're in recovery? */ 449 450 if (CC_ALGO(tp)->post_recovery != NULL) { 451 tp->ccv->curack = th->th_ack; 452 CC_ALGO(tp)->post_recovery(tp->ccv); 453 } 454 /* XXXLAS: EXIT_RECOVERY ? */ 455 tp->t_bytes_acked = 0; 456 } 457 458 #ifdef TCP_SIGNATURE 459 static inline int 460 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 461 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 462 { 463 int ret; 464 465 tcp_fields_to_net(th); 466 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 467 tcp_fields_to_host(th); 468 return (ret); 469 } 470 #endif 471 472 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 473 #ifdef INET6 474 #define ND6_HINT(tp) \ 475 do { \ 476 if ((tp) && (tp)->t_inpcb && \ 477 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 478 nd6_nud_hint(NULL, NULL, 0); \ 479 } while (0) 480 #else 481 #define ND6_HINT(tp) 482 #endif 483 484 /* 485 * Indicate whether this ack should be delayed. We can delay the ack if 486 * following conditions are met: 487 * - There is no delayed ack timer in progress. 488 * - Our last ack wasn't a 0-sized window. We never want to delay 489 * the ack that opens up a 0-sized window. 490 * - LRO wasn't used for this segment. We make sure by checking that the 491 * segment size is not larger than the MSS. 492 * - Delayed acks are enabled or this is a half-synchronized T/TCP 493 * connection. 494 */ 495 #define DELAY_ACK(tp, tlen) \ 496 ((!tcp_timer_active(tp, TT_DELACK) && \ 497 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 498 (tlen <= tp->t_maxopd) && \ 499 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 500 501 /* 502 * TCP input handling is split into multiple parts: 503 * tcp6_input is a thin wrapper around tcp_input for the extended 504 * ip6_protox[] call format in ip6_input 505 * tcp_input handles primary segment validation, inpcb lookup and 506 * SYN processing on listen sockets 507 * tcp_do_segment processes the ACK and text of the segment for 508 * establishing, established and closing connections 509 */ 510 #ifdef INET6 511 int 512 tcp6_input(struct mbuf **mp, int *offp, int proto) 513 { 514 struct mbuf *m = *mp; 515 struct in6_ifaddr *ia6; 516 517 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 518 519 /* 520 * draft-itojun-ipv6-tcp-to-anycast 521 * better place to put this in? 522 */ 523 ia6 = ip6_getdstifaddr(m); 524 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 525 struct ip6_hdr *ip6; 526 527 ifa_free(&ia6->ia_ifa); 528 ip6 = mtod(m, struct ip6_hdr *); 529 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 530 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 531 return (IPPROTO_DONE); 532 } 533 if (ia6) 534 ifa_free(&ia6->ia_ifa); 535 536 return (tcp_input(mp, offp, proto)); 537 } 538 #endif /* INET6 */ 539 540 int 541 tcp_input(struct mbuf **mp, int *offp, int proto) 542 { 543 struct mbuf *m = *mp; 544 struct tcphdr *th = NULL; 545 struct ip *ip = NULL; 546 struct inpcb *inp = NULL; 547 struct tcpcb *tp = NULL; 548 struct socket *so = NULL; 549 u_char *optp = NULL; 550 int off0; 551 int optlen = 0; 552 #ifdef INET 553 int len; 554 #endif 555 int tlen = 0, off; 556 int drop_hdrlen; 557 int thflags; 558 int rstreason = 0; /* For badport_bandlim accounting purposes */ 559 #ifdef TCP_SIGNATURE 560 uint8_t sig_checked = 0; 561 #endif 562 uint8_t iptos = 0; 563 struct m_tag *fwd_tag = NULL; 564 #ifdef INET6 565 struct ip6_hdr *ip6 = NULL; 566 int isipv6; 567 #else 568 const void *ip6 = NULL; 569 #endif /* INET6 */ 570 struct tcpopt to; /* options in this segment */ 571 char *s = NULL; /* address and port logging */ 572 int ti_locked; 573 #define TI_UNLOCKED 1 574 #define TI_WLOCKED 2 575 576 #ifdef TCPDEBUG 577 /* 578 * The size of tcp_saveipgen must be the size of the max ip header, 579 * now IPv6. 580 */ 581 u_char tcp_saveipgen[IP6_HDR_LEN]; 582 struct tcphdr tcp_savetcp; 583 short ostate = 0; 584 #endif 585 586 #ifdef INET6 587 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 588 #endif 589 590 off0 = *offp; 591 m = *mp; 592 *mp = NULL; 593 to.to_flags = 0; 594 TCPSTAT_INC(tcps_rcvtotal); 595 596 #ifdef INET6 597 if (isipv6) { 598 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 599 600 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 601 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 602 if (m == NULL) { 603 TCPSTAT_INC(tcps_rcvshort); 604 return (IPPROTO_DONE); 605 } 606 } 607 608 ip6 = mtod(m, struct ip6_hdr *); 609 th = (struct tcphdr *)((caddr_t)ip6 + off0); 610 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 611 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 612 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 613 th->th_sum = m->m_pkthdr.csum_data; 614 else 615 th->th_sum = in6_cksum_pseudo(ip6, tlen, 616 IPPROTO_TCP, m->m_pkthdr.csum_data); 617 th->th_sum ^= 0xffff; 618 } else 619 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 620 if (th->th_sum) { 621 TCPSTAT_INC(tcps_rcvbadsum); 622 goto drop; 623 } 624 625 /* 626 * Be proactive about unspecified IPv6 address in source. 627 * As we use all-zero to indicate unbounded/unconnected pcb, 628 * unspecified IPv6 address can be used to confuse us. 629 * 630 * Note that packets with unspecified IPv6 destination is 631 * already dropped in ip6_input. 632 */ 633 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 634 /* XXX stat */ 635 goto drop; 636 } 637 } 638 #endif 639 #if defined(INET) && defined(INET6) 640 else 641 #endif 642 #ifdef INET 643 { 644 /* 645 * Get IP and TCP header together in first mbuf. 646 * Note: IP leaves IP header in first mbuf. 647 */ 648 if (off0 > sizeof (struct ip)) { 649 ip_stripoptions(m); 650 off0 = sizeof(struct ip); 651 } 652 if (m->m_len < sizeof (struct tcpiphdr)) { 653 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 654 == NULL) { 655 TCPSTAT_INC(tcps_rcvshort); 656 return (IPPROTO_DONE); 657 } 658 } 659 ip = mtod(m, struct ip *); 660 th = (struct tcphdr *)((caddr_t)ip + off0); 661 tlen = ntohs(ip->ip_len) - off0; 662 663 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 664 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 665 th->th_sum = m->m_pkthdr.csum_data; 666 else 667 th->th_sum = in_pseudo(ip->ip_src.s_addr, 668 ip->ip_dst.s_addr, 669 htonl(m->m_pkthdr.csum_data + tlen + 670 IPPROTO_TCP)); 671 th->th_sum ^= 0xffff; 672 } else { 673 struct ipovly *ipov = (struct ipovly *)ip; 674 675 /* 676 * Checksum extended TCP header and data. 677 */ 678 len = off0 + tlen; 679 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 680 ipov->ih_len = htons(tlen); 681 th->th_sum = in_cksum(m, len); 682 /* Reset length for SDT probes. */ 683 ip->ip_len = htons(tlen + off0); 684 } 685 686 if (th->th_sum) { 687 TCPSTAT_INC(tcps_rcvbadsum); 688 goto drop; 689 } 690 /* Re-initialization for later version check */ 691 ip->ip_v = IPVERSION; 692 } 693 #endif /* INET */ 694 695 #ifdef INET6 696 if (isipv6) 697 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 698 #endif 699 #if defined(INET) && defined(INET6) 700 else 701 #endif 702 #ifdef INET 703 iptos = ip->ip_tos; 704 #endif 705 706 /* 707 * Check that TCP offset makes sense, 708 * pull out TCP options and adjust length. XXX 709 */ 710 off = th->th_off << 2; 711 if (off < sizeof (struct tcphdr) || off > tlen) { 712 TCPSTAT_INC(tcps_rcvbadoff); 713 goto drop; 714 } 715 tlen -= off; /* tlen is used instead of ti->ti_len */ 716 if (off > sizeof (struct tcphdr)) { 717 #ifdef INET6 718 if (isipv6) { 719 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 720 ip6 = mtod(m, struct ip6_hdr *); 721 th = (struct tcphdr *)((caddr_t)ip6 + off0); 722 } 723 #endif 724 #if defined(INET) && defined(INET6) 725 else 726 #endif 727 #ifdef INET 728 { 729 if (m->m_len < sizeof(struct ip) + off) { 730 if ((m = m_pullup(m, sizeof (struct ip) + off)) 731 == NULL) { 732 TCPSTAT_INC(tcps_rcvshort); 733 return (IPPROTO_DONE); 734 } 735 ip = mtod(m, struct ip *); 736 th = (struct tcphdr *)((caddr_t)ip + off0); 737 } 738 } 739 #endif 740 optlen = off - sizeof (struct tcphdr); 741 optp = (u_char *)(th + 1); 742 } 743 thflags = th->th_flags; 744 745 /* 746 * Convert TCP protocol specific fields to host format. 747 */ 748 tcp_fields_to_host(th); 749 750 /* 751 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 752 */ 753 drop_hdrlen = off0 + off; 754 755 /* 756 * Locate pcb for segment; if we're likely to add or remove a 757 * connection then first acquire pcbinfo lock. There are three cases 758 * where we might discover later we need a write lock despite the 759 * flags: ACKs moving a connection out of the syncache, ACKs for a 760 * connection in TIMEWAIT and SYNs not targeting a listening socket. 761 */ 762 if ((thflags & (TH_FIN | TH_RST)) != 0) { 763 INP_INFO_WLOCK(&V_tcbinfo); 764 ti_locked = TI_WLOCKED; 765 } else 766 ti_locked = TI_UNLOCKED; 767 768 /* 769 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 770 */ 771 if ( 772 #ifdef INET6 773 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 774 #ifdef INET 775 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 776 #endif 777 #endif 778 #if defined(INET) && !defined(INET6) 779 (m->m_flags & M_IP_NEXTHOP) 780 #endif 781 ) 782 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 783 784 findpcb: 785 #ifdef INVARIANTS 786 if (ti_locked == TI_WLOCKED) { 787 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 788 } else { 789 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 790 } 791 #endif 792 #ifdef INET6 793 if (isipv6 && fwd_tag != NULL) { 794 struct sockaddr_in6 *next_hop6; 795 796 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 797 /* 798 * Transparently forwarded. Pretend to be the destination. 799 * Already got one like this? 800 */ 801 inp = in6_pcblookup_mbuf(&V_tcbinfo, 802 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 803 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 804 if (!inp) { 805 /* 806 * It's new. Try to find the ambushing socket. 807 * Because we've rewritten the destination address, 808 * any hardware-generated hash is ignored. 809 */ 810 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 811 th->th_sport, &next_hop6->sin6_addr, 812 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 813 th->th_dport, INPLOOKUP_WILDCARD | 814 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 815 } 816 } else if (isipv6) { 817 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 818 th->th_sport, &ip6->ip6_dst, th->th_dport, 819 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 820 m->m_pkthdr.rcvif, m); 821 } 822 #endif /* INET6 */ 823 #if defined(INET6) && defined(INET) 824 else 825 #endif 826 #ifdef INET 827 if (fwd_tag != NULL) { 828 struct sockaddr_in *next_hop; 829 830 next_hop = (struct sockaddr_in *)(fwd_tag+1); 831 /* 832 * Transparently forwarded. Pretend to be the destination. 833 * already got one like this? 834 */ 835 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 836 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 837 m->m_pkthdr.rcvif, m); 838 if (!inp) { 839 /* 840 * It's new. Try to find the ambushing socket. 841 * Because we've rewritten the destination address, 842 * any hardware-generated hash is ignored. 843 */ 844 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 845 th->th_sport, next_hop->sin_addr, 846 next_hop->sin_port ? ntohs(next_hop->sin_port) : 847 th->th_dport, INPLOOKUP_WILDCARD | 848 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 849 } 850 } else 851 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 852 th->th_sport, ip->ip_dst, th->th_dport, 853 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 854 m->m_pkthdr.rcvif, m); 855 #endif /* INET */ 856 857 /* 858 * If the INPCB does not exist then all data in the incoming 859 * segment is discarded and an appropriate RST is sent back. 860 * XXX MRT Send RST using which routing table? 861 */ 862 if (inp == NULL) { 863 /* 864 * Log communication attempts to ports that are not 865 * in use. 866 */ 867 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 868 tcp_log_in_vain == 2) { 869 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 870 log(LOG_INFO, "%s; %s: Connection attempt " 871 "to closed port\n", s, __func__); 872 } 873 /* 874 * When blackholing do not respond with a RST but 875 * completely ignore the segment and drop it. 876 */ 877 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 878 V_blackhole == 2) 879 goto dropunlock; 880 881 rstreason = BANDLIM_RST_CLOSEDPORT; 882 goto dropwithreset; 883 } 884 INP_WLOCK_ASSERT(inp); 885 if (!(inp->inp_flags & INP_HW_FLOWID) 886 && (m->m_flags & M_FLOWID) 887 && ((inp->inp_socket == NULL) 888 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 889 inp->inp_flags |= INP_HW_FLOWID; 890 inp->inp_flags &= ~INP_SW_FLOWID; 891 inp->inp_flowid = m->m_pkthdr.flowid; 892 inp->inp_flowtype = M_HASHTYPE_GET(m); 893 } 894 #ifdef IPSEC 895 #ifdef INET6 896 if (isipv6 && ipsec6_in_reject(m, inp)) { 897 IPSEC6STAT_INC(ips_in_polvio); 898 goto dropunlock; 899 } else 900 #endif /* INET6 */ 901 if (ipsec4_in_reject(m, inp) != 0) { 902 IPSECSTAT_INC(ips_in_polvio); 903 goto dropunlock; 904 } 905 #endif /* IPSEC */ 906 907 /* 908 * Check the minimum TTL for socket. 909 */ 910 if (inp->inp_ip_minttl != 0) { 911 #ifdef INET6 912 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 913 goto dropunlock; 914 else 915 #endif 916 if (inp->inp_ip_minttl > ip->ip_ttl) 917 goto dropunlock; 918 } 919 920 /* 921 * A previous connection in TIMEWAIT state is supposed to catch stray 922 * or duplicate segments arriving late. If this segment was a 923 * legitimate new connection attempt, the old INPCB gets removed and 924 * we can try again to find a listening socket. 925 * 926 * At this point, due to earlier optimism, we may hold only an inpcb 927 * lock, and not the inpcbinfo write lock. If so, we need to try to 928 * acquire it, or if that fails, acquire a reference on the inpcb, 929 * drop all locks, acquire a global write lock, and then re-acquire 930 * the inpcb lock. We may at that point discover that another thread 931 * has tried to free the inpcb, in which case we need to loop back 932 * and try to find a new inpcb to deliver to. 933 * 934 * XXXRW: It may be time to rethink timewait locking. 935 */ 936 relocked: 937 if (inp->inp_flags & INP_TIMEWAIT) { 938 if (ti_locked == TI_UNLOCKED) { 939 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 940 in_pcbref(inp); 941 INP_WUNLOCK(inp); 942 INP_INFO_WLOCK(&V_tcbinfo); 943 ti_locked = TI_WLOCKED; 944 INP_WLOCK(inp); 945 if (in_pcbrele_wlocked(inp)) { 946 inp = NULL; 947 goto findpcb; 948 } 949 } else 950 ti_locked = TI_WLOCKED; 951 } 952 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 953 954 if (thflags & TH_SYN) 955 tcp_dooptions(&to, optp, optlen, TO_SYN); 956 /* 957 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 958 */ 959 if (tcp_twcheck(inp, &to, th, m, tlen)) 960 goto findpcb; 961 INP_INFO_WUNLOCK(&V_tcbinfo); 962 return (IPPROTO_DONE); 963 } 964 /* 965 * The TCPCB may no longer exist if the connection is winding 966 * down or it is in the CLOSED state. Either way we drop the 967 * segment and send an appropriate response. 968 */ 969 tp = intotcpcb(inp); 970 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 971 rstreason = BANDLIM_RST_CLOSEDPORT; 972 goto dropwithreset; 973 } 974 975 #ifdef TCP_OFFLOAD 976 if (tp->t_flags & TF_TOE) { 977 tcp_offload_input(tp, m); 978 m = NULL; /* consumed by the TOE driver */ 979 goto dropunlock; 980 } 981 #endif 982 983 /* 984 * We've identified a valid inpcb, but it could be that we need an 985 * inpcbinfo write lock but don't hold it. In this case, attempt to 986 * acquire using the same strategy as the TIMEWAIT case above. If we 987 * relock, we have to jump back to 'relocked' as the connection might 988 * now be in TIMEWAIT. 989 */ 990 #ifdef INVARIANTS 991 if ((thflags & (TH_FIN | TH_RST)) != 0) 992 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 993 #endif 994 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 995 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN)))) { 996 if (ti_locked == TI_UNLOCKED) { 997 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 998 in_pcbref(inp); 999 INP_WUNLOCK(inp); 1000 INP_INFO_WLOCK(&V_tcbinfo); 1001 ti_locked = TI_WLOCKED; 1002 INP_WLOCK(inp); 1003 if (in_pcbrele_wlocked(inp)) { 1004 inp = NULL; 1005 goto findpcb; 1006 } 1007 goto relocked; 1008 } else 1009 ti_locked = TI_WLOCKED; 1010 } 1011 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1012 } 1013 1014 #ifdef MAC 1015 INP_WLOCK_ASSERT(inp); 1016 if (mac_inpcb_check_deliver(inp, m)) 1017 goto dropunlock; 1018 #endif 1019 so = inp->inp_socket; 1020 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1021 #ifdef TCPDEBUG 1022 if (so->so_options & SO_DEBUG) { 1023 ostate = tp->t_state; 1024 #ifdef INET6 1025 if (isipv6) { 1026 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1027 } else 1028 #endif 1029 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1030 tcp_savetcp = *th; 1031 } 1032 #endif /* TCPDEBUG */ 1033 /* 1034 * When the socket is accepting connections (the INPCB is in LISTEN 1035 * state) we look into the SYN cache if this is a new connection 1036 * attempt or the completion of a previous one. 1037 */ 1038 if (so->so_options & SO_ACCEPTCONN) { 1039 struct in_conninfo inc; 1040 1041 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1042 "tp not listening", __func__)); 1043 bzero(&inc, sizeof(inc)); 1044 #ifdef INET6 1045 if (isipv6) { 1046 inc.inc_flags |= INC_ISIPV6; 1047 inc.inc6_faddr = ip6->ip6_src; 1048 inc.inc6_laddr = ip6->ip6_dst; 1049 } else 1050 #endif 1051 { 1052 inc.inc_faddr = ip->ip_src; 1053 inc.inc_laddr = ip->ip_dst; 1054 } 1055 inc.inc_fport = th->th_sport; 1056 inc.inc_lport = th->th_dport; 1057 inc.inc_fibnum = so->so_fibnum; 1058 1059 /* 1060 * Check for an existing connection attempt in syncache if 1061 * the flag is only ACK. A successful lookup creates a new 1062 * socket appended to the listen queue in SYN_RECEIVED state. 1063 */ 1064 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1065 1066 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1067 /* 1068 * Parse the TCP options here because 1069 * syncookies need access to the reflected 1070 * timestamp. 1071 */ 1072 tcp_dooptions(&to, optp, optlen, 0); 1073 /* 1074 * NB: syncache_expand() doesn't unlock 1075 * inp and tcpinfo locks. 1076 */ 1077 if (!syncache_expand(&inc, &to, th, &so, m)) { 1078 /* 1079 * No syncache entry or ACK was not 1080 * for our SYN/ACK. Send a RST. 1081 * NB: syncache did its own logging 1082 * of the failure cause. 1083 */ 1084 rstreason = BANDLIM_RST_OPENPORT; 1085 goto dropwithreset; 1086 } 1087 if (so == NULL) { 1088 /* 1089 * We completed the 3-way handshake 1090 * but could not allocate a socket 1091 * either due to memory shortage, 1092 * listen queue length limits or 1093 * global socket limits. Send RST 1094 * or wait and have the remote end 1095 * retransmit the ACK for another 1096 * try. 1097 */ 1098 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1099 log(LOG_DEBUG, "%s; %s: Listen socket: " 1100 "Socket allocation failed due to " 1101 "limits or memory shortage, %s\n", 1102 s, __func__, 1103 V_tcp_sc_rst_sock_fail ? 1104 "sending RST" : "try again"); 1105 if (V_tcp_sc_rst_sock_fail) { 1106 rstreason = BANDLIM_UNLIMITED; 1107 goto dropwithreset; 1108 } else 1109 goto dropunlock; 1110 } 1111 /* 1112 * Socket is created in state SYN_RECEIVED. 1113 * Unlock the listen socket, lock the newly 1114 * created socket and update the tp variable. 1115 */ 1116 INP_WUNLOCK(inp); /* listen socket */ 1117 inp = sotoinpcb(so); 1118 INP_WLOCK(inp); /* new connection */ 1119 tp = intotcpcb(inp); 1120 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1121 ("%s: ", __func__)); 1122 #ifdef TCP_SIGNATURE 1123 if (sig_checked == 0) { 1124 tcp_dooptions(&to, optp, optlen, 1125 (thflags & TH_SYN) ? TO_SYN : 0); 1126 if (!tcp_signature_verify_input(m, off0, tlen, 1127 optlen, &to, th, tp->t_flags)) { 1128 1129 /* 1130 * In SYN_SENT state if it receives an 1131 * RST, it is allowed for further 1132 * processing. 1133 */ 1134 if ((thflags & TH_RST) == 0 || 1135 (tp->t_state == TCPS_SYN_SENT) == 0) 1136 goto dropunlock; 1137 } 1138 sig_checked = 1; 1139 } 1140 #endif 1141 1142 /* 1143 * Process the segment and the data it 1144 * contains. tcp_do_segment() consumes 1145 * the mbuf chain and unlocks the inpcb. 1146 */ 1147 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1148 iptos, ti_locked); 1149 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1150 return (IPPROTO_DONE); 1151 } 1152 /* 1153 * Segment flag validation for new connection attempts: 1154 * 1155 * Our (SYN|ACK) response was rejected. 1156 * Check with syncache and remove entry to prevent 1157 * retransmits. 1158 * 1159 * NB: syncache_chkrst does its own logging of failure 1160 * causes. 1161 */ 1162 if (thflags & TH_RST) { 1163 syncache_chkrst(&inc, th); 1164 goto dropunlock; 1165 } 1166 /* 1167 * We can't do anything without SYN. 1168 */ 1169 if ((thflags & TH_SYN) == 0) { 1170 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1171 log(LOG_DEBUG, "%s; %s: Listen socket: " 1172 "SYN is missing, segment ignored\n", 1173 s, __func__); 1174 TCPSTAT_INC(tcps_badsyn); 1175 goto dropunlock; 1176 } 1177 /* 1178 * (SYN|ACK) is bogus on a listen socket. 1179 */ 1180 if (thflags & TH_ACK) { 1181 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1182 log(LOG_DEBUG, "%s; %s: Listen socket: " 1183 "SYN|ACK invalid, segment rejected\n", 1184 s, __func__); 1185 syncache_badack(&inc); /* XXX: Not needed! */ 1186 TCPSTAT_INC(tcps_badsyn); 1187 rstreason = BANDLIM_RST_OPENPORT; 1188 goto dropwithreset; 1189 } 1190 /* 1191 * If the drop_synfin option is enabled, drop all 1192 * segments with both the SYN and FIN bits set. 1193 * This prevents e.g. nmap from identifying the 1194 * TCP/IP stack. 1195 * XXX: Poor reasoning. nmap has other methods 1196 * and is constantly refining its stack detection 1197 * strategies. 1198 * XXX: This is a violation of the TCP specification 1199 * and was used by RFC1644. 1200 */ 1201 if ((thflags & TH_FIN) && V_drop_synfin) { 1202 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1203 log(LOG_DEBUG, "%s; %s: Listen socket: " 1204 "SYN|FIN segment ignored (based on " 1205 "sysctl setting)\n", s, __func__); 1206 TCPSTAT_INC(tcps_badsyn); 1207 goto dropunlock; 1208 } 1209 /* 1210 * Segment's flags are (SYN) or (SYN|FIN). 1211 * 1212 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1213 * as they do not affect the state of the TCP FSM. 1214 * The data pointed to by TH_URG and th_urp is ignored. 1215 */ 1216 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1217 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1218 KASSERT(thflags & (TH_SYN), 1219 ("%s: Listen socket: TH_SYN not set", __func__)); 1220 #ifdef INET6 1221 /* 1222 * If deprecated address is forbidden, 1223 * we do not accept SYN to deprecated interface 1224 * address to prevent any new inbound connection from 1225 * getting established. 1226 * When we do not accept SYN, we send a TCP RST, 1227 * with deprecated source address (instead of dropping 1228 * it). We compromise it as it is much better for peer 1229 * to send a RST, and RST will be the final packet 1230 * for the exchange. 1231 * 1232 * If we do not forbid deprecated addresses, we accept 1233 * the SYN packet. RFC2462 does not suggest dropping 1234 * SYN in this case. 1235 * If we decipher RFC2462 5.5.4, it says like this: 1236 * 1. use of deprecated addr with existing 1237 * communication is okay - "SHOULD continue to be 1238 * used" 1239 * 2. use of it with new communication: 1240 * (2a) "SHOULD NOT be used if alternate address 1241 * with sufficient scope is available" 1242 * (2b) nothing mentioned otherwise. 1243 * Here we fall into (2b) case as we have no choice in 1244 * our source address selection - we must obey the peer. 1245 * 1246 * The wording in RFC2462 is confusing, and there are 1247 * multiple description text for deprecated address 1248 * handling - worse, they are not exactly the same. 1249 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1250 */ 1251 if (isipv6 && !V_ip6_use_deprecated) { 1252 struct in6_ifaddr *ia6; 1253 1254 ia6 = ip6_getdstifaddr(m); 1255 if (ia6 != NULL && 1256 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1257 ifa_free(&ia6->ia_ifa); 1258 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1259 log(LOG_DEBUG, "%s; %s: Listen socket: " 1260 "Connection attempt to deprecated " 1261 "IPv6 address rejected\n", 1262 s, __func__); 1263 rstreason = BANDLIM_RST_OPENPORT; 1264 goto dropwithreset; 1265 } 1266 if (ia6) 1267 ifa_free(&ia6->ia_ifa); 1268 } 1269 #endif /* INET6 */ 1270 /* 1271 * Basic sanity checks on incoming SYN requests: 1272 * Don't respond if the destination is a link layer 1273 * broadcast according to RFC1122 4.2.3.10, p. 104. 1274 * If it is from this socket it must be forged. 1275 * Don't respond if the source or destination is a 1276 * global or subnet broad- or multicast address. 1277 * Note that it is quite possible to receive unicast 1278 * link-layer packets with a broadcast IP address. Use 1279 * in_broadcast() to find them. 1280 */ 1281 if (m->m_flags & (M_BCAST|M_MCAST)) { 1282 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1283 log(LOG_DEBUG, "%s; %s: Listen socket: " 1284 "Connection attempt from broad- or multicast " 1285 "link layer address ignored\n", s, __func__); 1286 goto dropunlock; 1287 } 1288 #ifdef INET6 1289 if (isipv6) { 1290 if (th->th_dport == th->th_sport && 1291 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1292 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1293 log(LOG_DEBUG, "%s; %s: Listen socket: " 1294 "Connection attempt to/from self " 1295 "ignored\n", s, __func__); 1296 goto dropunlock; 1297 } 1298 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1299 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1300 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1301 log(LOG_DEBUG, "%s; %s: Listen socket: " 1302 "Connection attempt from/to multicast " 1303 "address ignored\n", s, __func__); 1304 goto dropunlock; 1305 } 1306 } 1307 #endif 1308 #if defined(INET) && defined(INET6) 1309 else 1310 #endif 1311 #ifdef INET 1312 { 1313 if (th->th_dport == th->th_sport && 1314 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from/to self " 1318 "ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1322 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1323 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1324 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1326 log(LOG_DEBUG, "%s; %s: Listen socket: " 1327 "Connection attempt from/to broad- " 1328 "or multicast address ignored\n", 1329 s, __func__); 1330 goto dropunlock; 1331 } 1332 } 1333 #endif 1334 /* 1335 * SYN appears to be valid. Create compressed TCP state 1336 * for syncache. 1337 */ 1338 #ifdef TCPDEBUG 1339 if (so->so_options & SO_DEBUG) 1340 tcp_trace(TA_INPUT, ostate, tp, 1341 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1342 #endif 1343 tcp_dooptions(&to, optp, optlen, TO_SYN); 1344 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1345 /* 1346 * Entry added to syncache and mbuf consumed. 1347 * Only the listen socket is unlocked by syncache_add(). 1348 */ 1349 if (ti_locked == TI_WLOCKED) { 1350 INP_INFO_WUNLOCK(&V_tcbinfo); 1351 ti_locked = TI_UNLOCKED; 1352 } 1353 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1354 return (IPPROTO_DONE); 1355 } else if (tp->t_state == TCPS_LISTEN) { 1356 /* 1357 * When a listen socket is torn down the SO_ACCEPTCONN 1358 * flag is removed first while connections are drained 1359 * from the accept queue in a unlock/lock cycle of the 1360 * ACCEPT_LOCK, opening a race condition allowing a SYN 1361 * attempt go through unhandled. 1362 */ 1363 goto dropunlock; 1364 } 1365 1366 #ifdef TCP_SIGNATURE 1367 if (sig_checked == 0) { 1368 tcp_dooptions(&to, optp, optlen, 1369 (thflags & TH_SYN) ? TO_SYN : 0); 1370 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1371 th, tp->t_flags)) { 1372 1373 /* 1374 * In SYN_SENT state if it receives an RST, it is 1375 * allowed for further processing. 1376 */ 1377 if ((thflags & TH_RST) == 0 || 1378 (tp->t_state == TCPS_SYN_SENT) == 0) 1379 goto dropunlock; 1380 } 1381 sig_checked = 1; 1382 } 1383 #endif 1384 1385 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1386 1387 /* 1388 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1389 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1390 * the inpcb, and unlocks pcbinfo. 1391 */ 1392 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1393 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1394 return (IPPROTO_DONE); 1395 1396 dropwithreset: 1397 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1398 1399 if (ti_locked == TI_WLOCKED) { 1400 INP_INFO_WUNLOCK(&V_tcbinfo); 1401 ti_locked = TI_UNLOCKED; 1402 } 1403 #ifdef INVARIANTS 1404 else { 1405 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1406 "ti_locked: %d", __func__, ti_locked)); 1407 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1408 } 1409 #endif 1410 1411 if (inp != NULL) { 1412 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1413 INP_WUNLOCK(inp); 1414 } else 1415 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1416 m = NULL; /* mbuf chain got consumed. */ 1417 goto drop; 1418 1419 dropunlock: 1420 if (m != NULL) 1421 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1422 1423 if (ti_locked == TI_WLOCKED) { 1424 INP_INFO_WUNLOCK(&V_tcbinfo); 1425 ti_locked = TI_UNLOCKED; 1426 } 1427 #ifdef INVARIANTS 1428 else { 1429 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1430 "ti_locked: %d", __func__, ti_locked)); 1431 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1432 } 1433 #endif 1434 1435 if (inp != NULL) 1436 INP_WUNLOCK(inp); 1437 1438 drop: 1439 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1440 if (s != NULL) 1441 free(s, M_TCPLOG); 1442 if (m != NULL) 1443 m_freem(m); 1444 return (IPPROTO_DONE); 1445 } 1446 1447 static void 1448 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1449 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1450 int ti_locked) 1451 { 1452 int thflags, acked, ourfinisacked, needoutput = 0; 1453 int rstreason, todrop, win; 1454 u_long tiwin; 1455 char *s; 1456 struct in_conninfo *inc; 1457 struct mbuf *mfree; 1458 struct tcpopt to; 1459 1460 #ifdef TCPDEBUG 1461 /* 1462 * The size of tcp_saveipgen must be the size of the max ip header, 1463 * now IPv6. 1464 */ 1465 u_char tcp_saveipgen[IP6_HDR_LEN]; 1466 struct tcphdr tcp_savetcp; 1467 short ostate = 0; 1468 #endif 1469 thflags = th->th_flags; 1470 inc = &tp->t_inpcb->inp_inc; 1471 tp->sackhint.last_sack_ack = 0; 1472 1473 /* 1474 * If this is either a state-changing packet or current state isn't 1475 * established, we require a write lock on tcbinfo. Otherwise, we 1476 * allow the tcbinfo to be in either alocked or unlocked, as the 1477 * caller may have unnecessarily acquired a write lock due to a race. 1478 */ 1479 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1480 tp->t_state != TCPS_ESTABLISHED) { 1481 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1482 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1483 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1484 } else { 1485 #ifdef INVARIANTS 1486 if (ti_locked == TI_WLOCKED) 1487 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1488 else { 1489 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1490 "ti_locked: %d", __func__, ti_locked)); 1491 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1492 } 1493 #endif 1494 } 1495 INP_WLOCK_ASSERT(tp->t_inpcb); 1496 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1497 __func__)); 1498 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1499 __func__)); 1500 1501 /* 1502 * Segment received on connection. 1503 * Reset idle time and keep-alive timer. 1504 * XXX: This should be done after segment 1505 * validation to ignore broken/spoofed segs. 1506 */ 1507 tp->t_rcvtime = ticks; 1508 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1509 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1510 1511 /* 1512 * Unscale the window into a 32-bit value. 1513 * For the SYN_SENT state the scale is zero. 1514 */ 1515 tiwin = th->th_win << tp->snd_scale; 1516 1517 /* 1518 * TCP ECN processing. 1519 */ 1520 if (tp->t_flags & TF_ECN_PERMIT) { 1521 if (thflags & TH_CWR) 1522 tp->t_flags &= ~TF_ECN_SND_ECE; 1523 switch (iptos & IPTOS_ECN_MASK) { 1524 case IPTOS_ECN_CE: 1525 tp->t_flags |= TF_ECN_SND_ECE; 1526 TCPSTAT_INC(tcps_ecn_ce); 1527 break; 1528 case IPTOS_ECN_ECT0: 1529 TCPSTAT_INC(tcps_ecn_ect0); 1530 break; 1531 case IPTOS_ECN_ECT1: 1532 TCPSTAT_INC(tcps_ecn_ect1); 1533 break; 1534 } 1535 /* Congestion experienced. */ 1536 if (thflags & TH_ECE) { 1537 cc_cong_signal(tp, th, CC_ECN); 1538 } 1539 } 1540 1541 /* 1542 * Parse options on any incoming segment. 1543 */ 1544 tcp_dooptions(&to, (u_char *)(th + 1), 1545 (th->th_off << 2) - sizeof(struct tcphdr), 1546 (thflags & TH_SYN) ? TO_SYN : 0); 1547 1548 /* 1549 * If echoed timestamp is later than the current time, 1550 * fall back to non RFC1323 RTT calculation. Normalize 1551 * timestamp if syncookies were used when this connection 1552 * was established. 1553 */ 1554 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1555 to.to_tsecr -= tp->ts_offset; 1556 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1557 to.to_tsecr = 0; 1558 } 1559 /* 1560 * If timestamps were negotiated during SYN/ACK they should 1561 * appear on every segment during this session and vice versa. 1562 */ 1563 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1564 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1565 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1566 "no action\n", s, __func__); 1567 free(s, M_TCPLOG); 1568 } 1569 } 1570 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1571 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1572 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1573 "no action\n", s, __func__); 1574 free(s, M_TCPLOG); 1575 } 1576 } 1577 1578 /* 1579 * Process options only when we get SYN/ACK back. The SYN case 1580 * for incoming connections is handled in tcp_syncache. 1581 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1582 * or <SYN,ACK>) segment itself is never scaled. 1583 * XXX this is traditional behavior, may need to be cleaned up. 1584 */ 1585 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1586 if ((to.to_flags & TOF_SCALE) && 1587 (tp->t_flags & TF_REQ_SCALE)) { 1588 tp->t_flags |= TF_RCVD_SCALE; 1589 tp->snd_scale = to.to_wscale; 1590 } 1591 /* 1592 * Initial send window. It will be updated with 1593 * the next incoming segment to the scaled value. 1594 */ 1595 tp->snd_wnd = th->th_win; 1596 if (to.to_flags & TOF_TS) { 1597 tp->t_flags |= TF_RCVD_TSTMP; 1598 tp->ts_recent = to.to_tsval; 1599 tp->ts_recent_age = tcp_ts_getticks(); 1600 } 1601 if (to.to_flags & TOF_MSS) 1602 tcp_mss(tp, to.to_mss); 1603 if ((tp->t_flags & TF_SACK_PERMIT) && 1604 (to.to_flags & TOF_SACKPERM) == 0) 1605 tp->t_flags &= ~TF_SACK_PERMIT; 1606 } 1607 1608 /* 1609 * Header prediction: check for the two common cases 1610 * of a uni-directional data xfer. If the packet has 1611 * no control flags, is in-sequence, the window didn't 1612 * change and we're not retransmitting, it's a 1613 * candidate. If the length is zero and the ack moved 1614 * forward, we're the sender side of the xfer. Just 1615 * free the data acked & wake any higher level process 1616 * that was blocked waiting for space. If the length 1617 * is non-zero and the ack didn't move, we're the 1618 * receiver side. If we're getting packets in-order 1619 * (the reassembly queue is empty), add the data to 1620 * the socket buffer and note that we need a delayed ack. 1621 * Make sure that the hidden state-flags are also off. 1622 * Since we check for TCPS_ESTABLISHED first, it can only 1623 * be TH_NEEDSYN. 1624 */ 1625 if (tp->t_state == TCPS_ESTABLISHED && 1626 th->th_seq == tp->rcv_nxt && 1627 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1628 tp->snd_nxt == tp->snd_max && 1629 tiwin && tiwin == tp->snd_wnd && 1630 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1631 tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 || 1632 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1633 1634 /* 1635 * If last ACK falls within this segment's sequence numbers, 1636 * record the timestamp. 1637 * NOTE that the test is modified according to the latest 1638 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1639 */ 1640 if ((to.to_flags & TOF_TS) != 0 && 1641 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1642 tp->ts_recent_age = tcp_ts_getticks(); 1643 tp->ts_recent = to.to_tsval; 1644 } 1645 1646 if (tlen == 0) { 1647 if (SEQ_GT(th->th_ack, tp->snd_una) && 1648 SEQ_LEQ(th->th_ack, tp->snd_max) && 1649 !IN_RECOVERY(tp->t_flags) && 1650 (to.to_flags & TOF_SACK) == 0 && 1651 TAILQ_EMPTY(&tp->snd_holes)) { 1652 /* 1653 * This is a pure ack for outstanding data. 1654 */ 1655 if (ti_locked == TI_WLOCKED) 1656 INP_INFO_WUNLOCK(&V_tcbinfo); 1657 ti_locked = TI_UNLOCKED; 1658 1659 TCPSTAT_INC(tcps_predack); 1660 1661 /* 1662 * "bad retransmit" recovery. 1663 */ 1664 if (tp->t_rxtshift == 1 && 1665 tp->t_flags & TF_PREVVALID && 1666 (int)(ticks - tp->t_badrxtwin) < 0) { 1667 cc_cong_signal(tp, th, CC_RTO_ERR); 1668 } 1669 1670 /* 1671 * Recalculate the transmit timer / rtt. 1672 * 1673 * Some boxes send broken timestamp replies 1674 * during the SYN+ACK phase, ignore 1675 * timestamps of 0 or we could calculate a 1676 * huge RTT and blow up the retransmit timer. 1677 */ 1678 if ((to.to_flags & TOF_TS) != 0 && 1679 to.to_tsecr) { 1680 u_int t; 1681 1682 t = tcp_ts_getticks() - to.to_tsecr; 1683 if (!tp->t_rttlow || tp->t_rttlow > t) 1684 tp->t_rttlow = t; 1685 tcp_xmit_timer(tp, 1686 TCP_TS_TO_TICKS(t) + 1); 1687 } else if (tp->t_rtttime && 1688 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1689 if (!tp->t_rttlow || 1690 tp->t_rttlow > ticks - tp->t_rtttime) 1691 tp->t_rttlow = ticks - tp->t_rtttime; 1692 tcp_xmit_timer(tp, 1693 ticks - tp->t_rtttime); 1694 } 1695 acked = BYTES_THIS_ACK(tp, th); 1696 1697 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1698 hhook_run_tcp_est_in(tp, th, &to); 1699 1700 TCPSTAT_INC(tcps_rcvackpack); 1701 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1702 sbdrop(&so->so_snd, acked); 1703 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1704 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1705 tp->snd_recover = th->th_ack - 1; 1706 1707 /* 1708 * Let the congestion control algorithm update 1709 * congestion control related information. This 1710 * typically means increasing the congestion 1711 * window. 1712 */ 1713 cc_ack_received(tp, th, CC_ACK); 1714 1715 tp->snd_una = th->th_ack; 1716 /* 1717 * Pull snd_wl2 up to prevent seq wrap relative 1718 * to th_ack. 1719 */ 1720 tp->snd_wl2 = th->th_ack; 1721 tp->t_dupacks = 0; 1722 m_freem(m); 1723 ND6_HINT(tp); /* Some progress has been made. */ 1724 1725 /* 1726 * If all outstanding data are acked, stop 1727 * retransmit timer, otherwise restart timer 1728 * using current (possibly backed-off) value. 1729 * If process is waiting for space, 1730 * wakeup/selwakeup/signal. If data 1731 * are ready to send, let tcp_output 1732 * decide between more output or persist. 1733 */ 1734 #ifdef TCPDEBUG 1735 if (so->so_options & SO_DEBUG) 1736 tcp_trace(TA_INPUT, ostate, tp, 1737 (void *)tcp_saveipgen, 1738 &tcp_savetcp, 0); 1739 #endif 1740 if (tp->snd_una == tp->snd_max) 1741 tcp_timer_activate(tp, TT_REXMT, 0); 1742 else if (!tcp_timer_active(tp, TT_PERSIST)) 1743 tcp_timer_activate(tp, TT_REXMT, 1744 tp->t_rxtcur); 1745 sowwakeup(so); 1746 if (so->so_snd.sb_cc) 1747 (void) tcp_output(tp); 1748 goto check_delack; 1749 } 1750 } else if (th->th_ack == tp->snd_una && 1751 tlen <= sbspace(&so->so_rcv)) { 1752 int newsize = 0; /* automatic sockbuf scaling */ 1753 1754 /* 1755 * This is a pure, in-sequence data packet with 1756 * nothing on the reassembly queue and we have enough 1757 * buffer space to take it. 1758 */ 1759 if (ti_locked == TI_WLOCKED) 1760 INP_INFO_WUNLOCK(&V_tcbinfo); 1761 ti_locked = TI_UNLOCKED; 1762 1763 /* Clean receiver SACK report if present */ 1764 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1765 tcp_clean_sackreport(tp); 1766 TCPSTAT_INC(tcps_preddat); 1767 tp->rcv_nxt += tlen; 1768 /* 1769 * Pull snd_wl1 up to prevent seq wrap relative to 1770 * th_seq. 1771 */ 1772 tp->snd_wl1 = th->th_seq; 1773 /* 1774 * Pull rcv_up up to prevent seq wrap relative to 1775 * rcv_nxt. 1776 */ 1777 tp->rcv_up = tp->rcv_nxt; 1778 TCPSTAT_INC(tcps_rcvpack); 1779 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1780 ND6_HINT(tp); /* Some progress has been made */ 1781 #ifdef TCPDEBUG 1782 if (so->so_options & SO_DEBUG) 1783 tcp_trace(TA_INPUT, ostate, tp, 1784 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1785 #endif 1786 /* 1787 * Automatic sizing of receive socket buffer. Often the send 1788 * buffer size is not optimally adjusted to the actual network 1789 * conditions at hand (delay bandwidth product). Setting the 1790 * buffer size too small limits throughput on links with high 1791 * bandwidth and high delay (eg. trans-continental/oceanic links). 1792 * 1793 * On the receive side the socket buffer memory is only rarely 1794 * used to any significant extent. This allows us to be much 1795 * more aggressive in scaling the receive socket buffer. For 1796 * the case that the buffer space is actually used to a large 1797 * extent and we run out of kernel memory we can simply drop 1798 * the new segments; TCP on the sender will just retransmit it 1799 * later. Setting the buffer size too big may only consume too 1800 * much kernel memory if the application doesn't read() from 1801 * the socket or packet loss or reordering makes use of the 1802 * reassembly queue. 1803 * 1804 * The criteria to step up the receive buffer one notch are: 1805 * 1. Application has not set receive buffer size with 1806 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1807 * 2. the number of bytes received during the time it takes 1808 * one timestamp to be reflected back to us (the RTT); 1809 * 3. received bytes per RTT is within seven eighth of the 1810 * current socket buffer size; 1811 * 4. receive buffer size has not hit maximal automatic size; 1812 * 1813 * This algorithm does one step per RTT at most and only if 1814 * we receive a bulk stream w/o packet losses or reorderings. 1815 * Shrinking the buffer during idle times is not necessary as 1816 * it doesn't consume any memory when idle. 1817 * 1818 * TODO: Only step up if the application is actually serving 1819 * the buffer to better manage the socket buffer resources. 1820 */ 1821 if (V_tcp_do_autorcvbuf && 1822 to.to_tsecr && 1823 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1824 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1825 to.to_tsecr - tp->rfbuf_ts < hz) { 1826 if (tp->rfbuf_cnt > 1827 (so->so_rcv.sb_hiwat / 8 * 7) && 1828 so->so_rcv.sb_hiwat < 1829 V_tcp_autorcvbuf_max) { 1830 newsize = 1831 min(so->so_rcv.sb_hiwat + 1832 V_tcp_autorcvbuf_inc, 1833 V_tcp_autorcvbuf_max); 1834 } 1835 /* Start over with next RTT. */ 1836 tp->rfbuf_ts = 0; 1837 tp->rfbuf_cnt = 0; 1838 } else 1839 tp->rfbuf_cnt += tlen; /* add up */ 1840 } 1841 1842 /* Add data to socket buffer. */ 1843 SOCKBUF_LOCK(&so->so_rcv); 1844 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1845 m_freem(m); 1846 } else { 1847 /* 1848 * Set new socket buffer size. 1849 * Give up when limit is reached. 1850 */ 1851 if (newsize) 1852 if (!sbreserve_locked(&so->so_rcv, 1853 newsize, so, NULL)) 1854 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1855 m_adj(m, drop_hdrlen); /* delayed header drop */ 1856 sbappendstream_locked(&so->so_rcv, m); 1857 } 1858 /* NB: sorwakeup_locked() does an implicit unlock. */ 1859 sorwakeup_locked(so); 1860 if (DELAY_ACK(tp, tlen)) { 1861 tp->t_flags |= TF_DELACK; 1862 } else { 1863 tp->t_flags |= TF_ACKNOW; 1864 tcp_output(tp); 1865 } 1866 goto check_delack; 1867 } 1868 } 1869 1870 /* 1871 * Calculate amount of space in receive window, 1872 * and then do TCP input processing. 1873 * Receive window is amount of space in rcv queue, 1874 * but not less than advertised window. 1875 */ 1876 win = sbspace(&so->so_rcv); 1877 if (win < 0) 1878 win = 0; 1879 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1880 1881 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1882 tp->rfbuf_ts = 0; 1883 tp->rfbuf_cnt = 0; 1884 1885 switch (tp->t_state) { 1886 1887 /* 1888 * If the state is SYN_RECEIVED: 1889 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1890 */ 1891 case TCPS_SYN_RECEIVED: 1892 if ((thflags & TH_ACK) && 1893 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1894 SEQ_GT(th->th_ack, tp->snd_max))) { 1895 rstreason = BANDLIM_RST_OPENPORT; 1896 goto dropwithreset; 1897 } 1898 break; 1899 1900 /* 1901 * If the state is SYN_SENT: 1902 * if seg contains an ACK, but not for our SYN, drop the input. 1903 * if seg contains a RST, then drop the connection. 1904 * if seg does not contain SYN, then drop it. 1905 * Otherwise this is an acceptable SYN segment 1906 * initialize tp->rcv_nxt and tp->irs 1907 * if seg contains ack then advance tp->snd_una 1908 * if seg contains an ECE and ECN support is enabled, the stream 1909 * is ECN capable. 1910 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1911 * arrange for segment to be acked (eventually) 1912 * continue processing rest of data/controls, beginning with URG 1913 */ 1914 case TCPS_SYN_SENT: 1915 if ((thflags & TH_ACK) && 1916 (SEQ_LEQ(th->th_ack, tp->iss) || 1917 SEQ_GT(th->th_ack, tp->snd_max))) { 1918 rstreason = BANDLIM_UNLIMITED; 1919 goto dropwithreset; 1920 } 1921 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1922 TCP_PROBE5(connect__refused, NULL, tp, 1923 mtod(m, const char *), tp, th); 1924 tp = tcp_drop(tp, ECONNREFUSED); 1925 } 1926 if (thflags & TH_RST) 1927 goto drop; 1928 if (!(thflags & TH_SYN)) 1929 goto drop; 1930 1931 tp->irs = th->th_seq; 1932 tcp_rcvseqinit(tp); 1933 if (thflags & TH_ACK) { 1934 TCPSTAT_INC(tcps_connects); 1935 soisconnected(so); 1936 #ifdef MAC 1937 mac_socketpeer_set_from_mbuf(m, so); 1938 #endif 1939 /* Do window scaling on this connection? */ 1940 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1941 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1942 tp->rcv_scale = tp->request_r_scale; 1943 } 1944 tp->rcv_adv += imin(tp->rcv_wnd, 1945 TCP_MAXWIN << tp->rcv_scale); 1946 tp->snd_una++; /* SYN is acked */ 1947 /* 1948 * If there's data, delay ACK; if there's also a FIN 1949 * ACKNOW will be turned on later. 1950 */ 1951 if (DELAY_ACK(tp, tlen) && tlen != 0) 1952 tcp_timer_activate(tp, TT_DELACK, 1953 tcp_delacktime); 1954 else 1955 tp->t_flags |= TF_ACKNOW; 1956 1957 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1958 tp->t_flags |= TF_ECN_PERMIT; 1959 TCPSTAT_INC(tcps_ecn_shs); 1960 } 1961 1962 /* 1963 * Received <SYN,ACK> in SYN_SENT[*] state. 1964 * Transitions: 1965 * SYN_SENT --> ESTABLISHED 1966 * SYN_SENT* --> FIN_WAIT_1 1967 */ 1968 tp->t_starttime = ticks; 1969 if (tp->t_flags & TF_NEEDFIN) { 1970 tcp_state_change(tp, TCPS_FIN_WAIT_1); 1971 tp->t_flags &= ~TF_NEEDFIN; 1972 thflags &= ~TH_SYN; 1973 } else { 1974 tcp_state_change(tp, TCPS_ESTABLISHED); 1975 TCP_PROBE5(connect__established, NULL, tp, 1976 mtod(m, const char *), tp, th); 1977 cc_conn_init(tp); 1978 tcp_timer_activate(tp, TT_KEEP, 1979 TP_KEEPIDLE(tp)); 1980 } 1981 } else { 1982 /* 1983 * Received initial SYN in SYN-SENT[*] state => 1984 * simultaneous open. 1985 * If it succeeds, connection is * half-synchronized. 1986 * Otherwise, do 3-way handshake: 1987 * SYN-SENT -> SYN-RECEIVED 1988 * SYN-SENT* -> SYN-RECEIVED* 1989 */ 1990 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1991 tcp_timer_activate(tp, TT_REXMT, 0); 1992 tcp_state_change(tp, TCPS_SYN_RECEIVED); 1993 } 1994 1995 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1996 "ti_locked %d", __func__, ti_locked)); 1997 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1998 INP_WLOCK_ASSERT(tp->t_inpcb); 1999 2000 /* 2001 * Advance th->th_seq to correspond to first data byte. 2002 * If data, trim to stay within window, 2003 * dropping FIN if necessary. 2004 */ 2005 th->th_seq++; 2006 if (tlen > tp->rcv_wnd) { 2007 todrop = tlen - tp->rcv_wnd; 2008 m_adj(m, -todrop); 2009 tlen = tp->rcv_wnd; 2010 thflags &= ~TH_FIN; 2011 TCPSTAT_INC(tcps_rcvpackafterwin); 2012 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2013 } 2014 tp->snd_wl1 = th->th_seq - 1; 2015 tp->rcv_up = th->th_seq; 2016 /* 2017 * Client side of transaction: already sent SYN and data. 2018 * If the remote host used T/TCP to validate the SYN, 2019 * our data will be ACK'd; if so, enter normal data segment 2020 * processing in the middle of step 5, ack processing. 2021 * Otherwise, goto step 6. 2022 */ 2023 if (thflags & TH_ACK) 2024 goto process_ACK; 2025 2026 goto step6; 2027 2028 /* 2029 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2030 * do normal processing. 2031 * 2032 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2033 */ 2034 case TCPS_LAST_ACK: 2035 case TCPS_CLOSING: 2036 break; /* continue normal processing */ 2037 } 2038 2039 /* 2040 * States other than LISTEN or SYN_SENT. 2041 * First check the RST flag and sequence number since reset segments 2042 * are exempt from the timestamp and connection count tests. This 2043 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2044 * below which allowed reset segments in half the sequence space 2045 * to fall though and be processed (which gives forged reset 2046 * segments with a random sequence number a 50 percent chance of 2047 * killing a connection). 2048 * Then check timestamp, if present. 2049 * Then check the connection count, if present. 2050 * Then check that at least some bytes of segment are within 2051 * receive window. If segment begins before rcv_nxt, 2052 * drop leading data (and SYN); if nothing left, just ack. 2053 */ 2054 if (thflags & TH_RST) { 2055 /* 2056 * RFC5961 Section 3.2 2057 * 2058 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2059 * - If RST is in window, we send challenge ACK. 2060 * 2061 * Note: to take into account delayed ACKs, we should 2062 * test against last_ack_sent instead of rcv_nxt. 2063 * Note 2: we handle special case of closed window, not 2064 * covered by the RFC. 2065 */ 2066 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2067 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2068 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2069 2070 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2071 KASSERT(ti_locked == TI_WLOCKED, 2072 ("%s: TH_RST ti_locked %d, th %p tp %p", 2073 __func__, ti_locked, th, tp)); 2074 KASSERT(tp->t_state != TCPS_SYN_SENT, 2075 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2076 __func__, th, tp)); 2077 2078 if (V_tcp_insecure_rst || 2079 tp->last_ack_sent == th->th_seq) { 2080 TCPSTAT_INC(tcps_drops); 2081 /* Drop the connection. */ 2082 switch (tp->t_state) { 2083 case TCPS_SYN_RECEIVED: 2084 so->so_error = ECONNREFUSED; 2085 goto close; 2086 case TCPS_ESTABLISHED: 2087 case TCPS_FIN_WAIT_1: 2088 case TCPS_FIN_WAIT_2: 2089 case TCPS_CLOSE_WAIT: 2090 so->so_error = ECONNRESET; 2091 close: 2092 tcp_state_change(tp, TCPS_CLOSED); 2093 /* FALLTHROUGH */ 2094 default: 2095 tp = tcp_close(tp); 2096 } 2097 } else { 2098 TCPSTAT_INC(tcps_badrst); 2099 /* Send challenge ACK. */ 2100 tcp_respond(tp, mtod(m, void *), th, m, 2101 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2102 tp->last_ack_sent = tp->rcv_nxt; 2103 m = NULL; 2104 } 2105 } 2106 goto drop; 2107 } 2108 2109 /* 2110 * RFC5961 Section 4.2 2111 * Send challenge ACK for any SYN in synchronized state. 2112 */ 2113 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT) { 2114 KASSERT(ti_locked == TI_WLOCKED, 2115 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2116 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2117 2118 TCPSTAT_INC(tcps_badsyn); 2119 if (V_tcp_insecure_syn && 2120 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2121 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2122 tp = tcp_drop(tp, ECONNRESET); 2123 rstreason = BANDLIM_UNLIMITED; 2124 } else { 2125 /* Send challenge ACK. */ 2126 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2127 tp->snd_nxt, TH_ACK); 2128 tp->last_ack_sent = tp->rcv_nxt; 2129 m = NULL; 2130 } 2131 goto drop; 2132 } 2133 2134 /* 2135 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2136 * and it's less than ts_recent, drop it. 2137 */ 2138 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2139 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2140 2141 /* Check to see if ts_recent is over 24 days old. */ 2142 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2143 /* 2144 * Invalidate ts_recent. If this segment updates 2145 * ts_recent, the age will be reset later and ts_recent 2146 * will get a valid value. If it does not, setting 2147 * ts_recent to zero will at least satisfy the 2148 * requirement that zero be placed in the timestamp 2149 * echo reply when ts_recent isn't valid. The 2150 * age isn't reset until we get a valid ts_recent 2151 * because we don't want out-of-order segments to be 2152 * dropped when ts_recent is old. 2153 */ 2154 tp->ts_recent = 0; 2155 } else { 2156 TCPSTAT_INC(tcps_rcvduppack); 2157 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2158 TCPSTAT_INC(tcps_pawsdrop); 2159 if (tlen) 2160 goto dropafterack; 2161 goto drop; 2162 } 2163 } 2164 2165 /* 2166 * In the SYN-RECEIVED state, validate that the packet belongs to 2167 * this connection before trimming the data to fit the receive 2168 * window. Check the sequence number versus IRS since we know 2169 * the sequence numbers haven't wrapped. This is a partial fix 2170 * for the "LAND" DoS attack. 2171 */ 2172 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2173 rstreason = BANDLIM_RST_OPENPORT; 2174 goto dropwithreset; 2175 } 2176 2177 todrop = tp->rcv_nxt - th->th_seq; 2178 if (todrop > 0) { 2179 if (thflags & TH_SYN) { 2180 thflags &= ~TH_SYN; 2181 th->th_seq++; 2182 if (th->th_urp > 1) 2183 th->th_urp--; 2184 else 2185 thflags &= ~TH_URG; 2186 todrop--; 2187 } 2188 /* 2189 * Following if statement from Stevens, vol. 2, p. 960. 2190 */ 2191 if (todrop > tlen 2192 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2193 /* 2194 * Any valid FIN must be to the left of the window. 2195 * At this point the FIN must be a duplicate or out 2196 * of sequence; drop it. 2197 */ 2198 thflags &= ~TH_FIN; 2199 2200 /* 2201 * Send an ACK to resynchronize and drop any data. 2202 * But keep on processing for RST or ACK. 2203 */ 2204 tp->t_flags |= TF_ACKNOW; 2205 todrop = tlen; 2206 TCPSTAT_INC(tcps_rcvduppack); 2207 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2208 } else { 2209 TCPSTAT_INC(tcps_rcvpartduppack); 2210 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2211 } 2212 drop_hdrlen += todrop; /* drop from the top afterwards */ 2213 th->th_seq += todrop; 2214 tlen -= todrop; 2215 if (th->th_urp > todrop) 2216 th->th_urp -= todrop; 2217 else { 2218 thflags &= ~TH_URG; 2219 th->th_urp = 0; 2220 } 2221 } 2222 2223 /* 2224 * If new data are received on a connection after the 2225 * user processes are gone, then RST the other end. 2226 */ 2227 if ((so->so_state & SS_NOFDREF) && 2228 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2229 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2230 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2231 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2232 2233 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2234 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2235 "after socket was closed, " 2236 "sending RST and removing tcpcb\n", 2237 s, __func__, tcpstates[tp->t_state], tlen); 2238 free(s, M_TCPLOG); 2239 } 2240 tp = tcp_close(tp); 2241 TCPSTAT_INC(tcps_rcvafterclose); 2242 rstreason = BANDLIM_UNLIMITED; 2243 goto dropwithreset; 2244 } 2245 2246 /* 2247 * If segment ends after window, drop trailing data 2248 * (and PUSH and FIN); if nothing left, just ACK. 2249 */ 2250 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2251 if (todrop > 0) { 2252 TCPSTAT_INC(tcps_rcvpackafterwin); 2253 if (todrop >= tlen) { 2254 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2255 /* 2256 * If window is closed can only take segments at 2257 * window edge, and have to drop data and PUSH from 2258 * incoming segments. Continue processing, but 2259 * remember to ack. Otherwise, drop segment 2260 * and ack. 2261 */ 2262 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2263 tp->t_flags |= TF_ACKNOW; 2264 TCPSTAT_INC(tcps_rcvwinprobe); 2265 } else 2266 goto dropafterack; 2267 } else 2268 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2269 m_adj(m, -todrop); 2270 tlen -= todrop; 2271 thflags &= ~(TH_PUSH|TH_FIN); 2272 } 2273 2274 /* 2275 * If last ACK falls within this segment's sequence numbers, 2276 * record its timestamp. 2277 * NOTE: 2278 * 1) That the test incorporates suggestions from the latest 2279 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2280 * 2) That updating only on newer timestamps interferes with 2281 * our earlier PAWS tests, so this check should be solely 2282 * predicated on the sequence space of this segment. 2283 * 3) That we modify the segment boundary check to be 2284 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2285 * instead of RFC1323's 2286 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2287 * This modified check allows us to overcome RFC1323's 2288 * limitations as described in Stevens TCP/IP Illustrated 2289 * Vol. 2 p.869. In such cases, we can still calculate the 2290 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2291 */ 2292 if ((to.to_flags & TOF_TS) != 0 && 2293 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2294 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2295 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2296 tp->ts_recent_age = tcp_ts_getticks(); 2297 tp->ts_recent = to.to_tsval; 2298 } 2299 2300 /* 2301 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2302 * flag is on (half-synchronized state), then queue data for 2303 * later processing; else drop segment and return. 2304 */ 2305 if ((thflags & TH_ACK) == 0) { 2306 if (tp->t_state == TCPS_SYN_RECEIVED || 2307 (tp->t_flags & TF_NEEDSYN)) 2308 goto step6; 2309 else if (tp->t_flags & TF_ACKNOW) 2310 goto dropafterack; 2311 else 2312 goto drop; 2313 } 2314 2315 /* 2316 * Ack processing. 2317 */ 2318 switch (tp->t_state) { 2319 2320 /* 2321 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2322 * ESTABLISHED state and continue processing. 2323 * The ACK was checked above. 2324 */ 2325 case TCPS_SYN_RECEIVED: 2326 2327 TCPSTAT_INC(tcps_connects); 2328 soisconnected(so); 2329 /* Do window scaling? */ 2330 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2331 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2332 tp->rcv_scale = tp->request_r_scale; 2333 tp->snd_wnd = tiwin; 2334 } 2335 /* 2336 * Make transitions: 2337 * SYN-RECEIVED -> ESTABLISHED 2338 * SYN-RECEIVED* -> FIN-WAIT-1 2339 */ 2340 tp->t_starttime = ticks; 2341 if (tp->t_flags & TF_NEEDFIN) { 2342 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2343 tp->t_flags &= ~TF_NEEDFIN; 2344 } else { 2345 tcp_state_change(tp, TCPS_ESTABLISHED); 2346 TCP_PROBE5(accept__established, NULL, tp, 2347 mtod(m, const char *), tp, th); 2348 cc_conn_init(tp); 2349 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2350 } 2351 /* 2352 * If segment contains data or ACK, will call tcp_reass() 2353 * later; if not, do so now to pass queued data to user. 2354 */ 2355 if (tlen == 0 && (thflags & TH_FIN) == 0) 2356 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2357 (struct mbuf *)0); 2358 tp->snd_wl1 = th->th_seq - 1; 2359 /* FALLTHROUGH */ 2360 2361 /* 2362 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2363 * ACKs. If the ack is in the range 2364 * tp->snd_una < th->th_ack <= tp->snd_max 2365 * then advance tp->snd_una to th->th_ack and drop 2366 * data from the retransmission queue. If this ACK reflects 2367 * more up to date window information we update our window information. 2368 */ 2369 case TCPS_ESTABLISHED: 2370 case TCPS_FIN_WAIT_1: 2371 case TCPS_FIN_WAIT_2: 2372 case TCPS_CLOSE_WAIT: 2373 case TCPS_CLOSING: 2374 case TCPS_LAST_ACK: 2375 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2376 TCPSTAT_INC(tcps_rcvacktoomuch); 2377 goto dropafterack; 2378 } 2379 if ((tp->t_flags & TF_SACK_PERMIT) && 2380 ((to.to_flags & TOF_SACK) || 2381 !TAILQ_EMPTY(&tp->snd_holes))) 2382 tcp_sack_doack(tp, &to, th->th_ack); 2383 2384 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2385 hhook_run_tcp_est_in(tp, th, &to); 2386 2387 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2388 if (tlen == 0 && tiwin == tp->snd_wnd) { 2389 /* 2390 * If this is the first time we've seen a 2391 * FIN from the remote, this is not a 2392 * duplicate and it needs to be processed 2393 * normally. This happens during a 2394 * simultaneous close. 2395 */ 2396 if ((thflags & TH_FIN) && 2397 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2398 tp->t_dupacks = 0; 2399 break; 2400 } 2401 TCPSTAT_INC(tcps_rcvdupack); 2402 /* 2403 * If we have outstanding data (other than 2404 * a window probe), this is a completely 2405 * duplicate ack (ie, window info didn't 2406 * change and FIN isn't set), 2407 * the ack is the biggest we've 2408 * seen and we've seen exactly our rexmt 2409 * threshhold of them, assume a packet 2410 * has been dropped and retransmit it. 2411 * Kludge snd_nxt & the congestion 2412 * window so we send only this one 2413 * packet. 2414 * 2415 * We know we're losing at the current 2416 * window size so do congestion avoidance 2417 * (set ssthresh to half the current window 2418 * and pull our congestion window back to 2419 * the new ssthresh). 2420 * 2421 * Dup acks mean that packets have left the 2422 * network (they're now cached at the receiver) 2423 * so bump cwnd by the amount in the receiver 2424 * to keep a constant cwnd packets in the 2425 * network. 2426 * 2427 * When using TCP ECN, notify the peer that 2428 * we reduced the cwnd. 2429 */ 2430 if (!tcp_timer_active(tp, TT_REXMT) || 2431 th->th_ack != tp->snd_una) 2432 tp->t_dupacks = 0; 2433 else if (++tp->t_dupacks > tcprexmtthresh || 2434 IN_FASTRECOVERY(tp->t_flags)) { 2435 cc_ack_received(tp, th, CC_DUPACK); 2436 if ((tp->t_flags & TF_SACK_PERMIT) && 2437 IN_FASTRECOVERY(tp->t_flags)) { 2438 int awnd; 2439 2440 /* 2441 * Compute the amount of data in flight first. 2442 * We can inject new data into the pipe iff 2443 * we have less than 1/2 the original window's 2444 * worth of data in flight. 2445 */ 2446 awnd = (tp->snd_nxt - tp->snd_fack) + 2447 tp->sackhint.sack_bytes_rexmit; 2448 if (awnd < tp->snd_ssthresh) { 2449 tp->snd_cwnd += tp->t_maxseg; 2450 if (tp->snd_cwnd > tp->snd_ssthresh) 2451 tp->snd_cwnd = tp->snd_ssthresh; 2452 } 2453 } else 2454 tp->snd_cwnd += tp->t_maxseg; 2455 (void) tcp_output(tp); 2456 goto drop; 2457 } else if (tp->t_dupacks == tcprexmtthresh) { 2458 tcp_seq onxt = tp->snd_nxt; 2459 2460 /* 2461 * If we're doing sack, check to 2462 * see if we're already in sack 2463 * recovery. If we're not doing sack, 2464 * check to see if we're in newreno 2465 * recovery. 2466 */ 2467 if (tp->t_flags & TF_SACK_PERMIT) { 2468 if (IN_FASTRECOVERY(tp->t_flags)) { 2469 tp->t_dupacks = 0; 2470 break; 2471 } 2472 } else { 2473 if (SEQ_LEQ(th->th_ack, 2474 tp->snd_recover)) { 2475 tp->t_dupacks = 0; 2476 break; 2477 } 2478 } 2479 /* Congestion signal before ack. */ 2480 cc_cong_signal(tp, th, CC_NDUPACK); 2481 cc_ack_received(tp, th, CC_DUPACK); 2482 tcp_timer_activate(tp, TT_REXMT, 0); 2483 tp->t_rtttime = 0; 2484 if (tp->t_flags & TF_SACK_PERMIT) { 2485 TCPSTAT_INC( 2486 tcps_sack_recovery_episode); 2487 tp->sack_newdata = tp->snd_nxt; 2488 tp->snd_cwnd = tp->t_maxseg; 2489 (void) tcp_output(tp); 2490 goto drop; 2491 } 2492 tp->snd_nxt = th->th_ack; 2493 tp->snd_cwnd = tp->t_maxseg; 2494 (void) tcp_output(tp); 2495 KASSERT(tp->snd_limited <= 2, 2496 ("%s: tp->snd_limited too big", 2497 __func__)); 2498 tp->snd_cwnd = tp->snd_ssthresh + 2499 tp->t_maxseg * 2500 (tp->t_dupacks - tp->snd_limited); 2501 if (SEQ_GT(onxt, tp->snd_nxt)) 2502 tp->snd_nxt = onxt; 2503 goto drop; 2504 } else if (V_tcp_do_rfc3042) { 2505 cc_ack_received(tp, th, CC_DUPACK); 2506 u_long oldcwnd = tp->snd_cwnd; 2507 tcp_seq oldsndmax = tp->snd_max; 2508 u_int sent; 2509 int avail; 2510 2511 KASSERT(tp->t_dupacks == 1 || 2512 tp->t_dupacks == 2, 2513 ("%s: dupacks not 1 or 2", 2514 __func__)); 2515 if (tp->t_dupacks == 1) 2516 tp->snd_limited = 0; 2517 tp->snd_cwnd = 2518 (tp->snd_nxt - tp->snd_una) + 2519 (tp->t_dupacks - tp->snd_limited) * 2520 tp->t_maxseg; 2521 /* 2522 * Only call tcp_output when there 2523 * is new data available to be sent. 2524 * Otherwise we would send pure ACKs. 2525 */ 2526 SOCKBUF_LOCK(&so->so_snd); 2527 avail = so->so_snd.sb_cc - 2528 (tp->snd_nxt - tp->snd_una); 2529 SOCKBUF_UNLOCK(&so->so_snd); 2530 if (avail > 0) 2531 (void) tcp_output(tp); 2532 sent = tp->snd_max - oldsndmax; 2533 if (sent > tp->t_maxseg) { 2534 KASSERT((tp->t_dupacks == 2 && 2535 tp->snd_limited == 0) || 2536 (sent == tp->t_maxseg + 1 && 2537 tp->t_flags & TF_SENTFIN), 2538 ("%s: sent too much", 2539 __func__)); 2540 tp->snd_limited = 2; 2541 } else if (sent > 0) 2542 ++tp->snd_limited; 2543 tp->snd_cwnd = oldcwnd; 2544 goto drop; 2545 } 2546 } else 2547 tp->t_dupacks = 0; 2548 break; 2549 } 2550 2551 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2552 ("%s: th_ack <= snd_una", __func__)); 2553 2554 /* 2555 * If the congestion window was inflated to account 2556 * for the other side's cached packets, retract it. 2557 */ 2558 if (IN_FASTRECOVERY(tp->t_flags)) { 2559 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2560 if (tp->t_flags & TF_SACK_PERMIT) 2561 tcp_sack_partialack(tp, th); 2562 else 2563 tcp_newreno_partial_ack(tp, th); 2564 } else 2565 cc_post_recovery(tp, th); 2566 } 2567 tp->t_dupacks = 0; 2568 /* 2569 * If we reach this point, ACK is not a duplicate, 2570 * i.e., it ACKs something we sent. 2571 */ 2572 if (tp->t_flags & TF_NEEDSYN) { 2573 /* 2574 * T/TCP: Connection was half-synchronized, and our 2575 * SYN has been ACK'd (so connection is now fully 2576 * synchronized). Go to non-starred state, 2577 * increment snd_una for ACK of SYN, and check if 2578 * we can do window scaling. 2579 */ 2580 tp->t_flags &= ~TF_NEEDSYN; 2581 tp->snd_una++; 2582 /* Do window scaling? */ 2583 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2584 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2585 tp->rcv_scale = tp->request_r_scale; 2586 /* Send window already scaled. */ 2587 } 2588 } 2589 2590 process_ACK: 2591 INP_WLOCK_ASSERT(tp->t_inpcb); 2592 2593 acked = BYTES_THIS_ACK(tp, th); 2594 TCPSTAT_INC(tcps_rcvackpack); 2595 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2596 2597 /* 2598 * If we just performed our first retransmit, and the ACK 2599 * arrives within our recovery window, then it was a mistake 2600 * to do the retransmit in the first place. Recover our 2601 * original cwnd and ssthresh, and proceed to transmit where 2602 * we left off. 2603 */ 2604 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2605 (int)(ticks - tp->t_badrxtwin) < 0) 2606 cc_cong_signal(tp, th, CC_RTO_ERR); 2607 2608 /* 2609 * If we have a timestamp reply, update smoothed 2610 * round trip time. If no timestamp is present but 2611 * transmit timer is running and timed sequence 2612 * number was acked, update smoothed round trip time. 2613 * Since we now have an rtt measurement, cancel the 2614 * timer backoff (cf., Phil Karn's retransmit alg.). 2615 * Recompute the initial retransmit timer. 2616 * 2617 * Some boxes send broken timestamp replies 2618 * during the SYN+ACK phase, ignore 2619 * timestamps of 0 or we could calculate a 2620 * huge RTT and blow up the retransmit timer. 2621 */ 2622 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2623 u_int t; 2624 2625 t = tcp_ts_getticks() - to.to_tsecr; 2626 if (!tp->t_rttlow || tp->t_rttlow > t) 2627 tp->t_rttlow = t; 2628 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2629 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2630 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2631 tp->t_rttlow = ticks - tp->t_rtttime; 2632 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2633 } 2634 2635 /* 2636 * If all outstanding data is acked, stop retransmit 2637 * timer and remember to restart (more output or persist). 2638 * If there is more data to be acked, restart retransmit 2639 * timer, using current (possibly backed-off) value. 2640 */ 2641 if (th->th_ack == tp->snd_max) { 2642 tcp_timer_activate(tp, TT_REXMT, 0); 2643 needoutput = 1; 2644 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2645 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2646 2647 /* 2648 * If no data (only SYN) was ACK'd, 2649 * skip rest of ACK processing. 2650 */ 2651 if (acked == 0) 2652 goto step6; 2653 2654 /* 2655 * Let the congestion control algorithm update congestion 2656 * control related information. This typically means increasing 2657 * the congestion window. 2658 */ 2659 cc_ack_received(tp, th, CC_ACK); 2660 2661 SOCKBUF_LOCK(&so->so_snd); 2662 if (acked > so->so_snd.sb_cc) { 2663 tp->snd_wnd -= so->so_snd.sb_cc; 2664 mfree = sbcut_locked(&so->so_snd, 2665 (int)so->so_snd.sb_cc); 2666 ourfinisacked = 1; 2667 } else { 2668 mfree = sbcut_locked(&so->so_snd, acked); 2669 tp->snd_wnd -= acked; 2670 ourfinisacked = 0; 2671 } 2672 /* NB: sowwakeup_locked() does an implicit unlock. */ 2673 sowwakeup_locked(so); 2674 m_freem(mfree); 2675 /* Detect una wraparound. */ 2676 if (!IN_RECOVERY(tp->t_flags) && 2677 SEQ_GT(tp->snd_una, tp->snd_recover) && 2678 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2679 tp->snd_recover = th->th_ack - 1; 2680 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2681 if (IN_RECOVERY(tp->t_flags) && 2682 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2683 EXIT_RECOVERY(tp->t_flags); 2684 } 2685 tp->snd_una = th->th_ack; 2686 if (tp->t_flags & TF_SACK_PERMIT) { 2687 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2688 tp->snd_recover = tp->snd_una; 2689 } 2690 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2691 tp->snd_nxt = tp->snd_una; 2692 2693 switch (tp->t_state) { 2694 2695 /* 2696 * In FIN_WAIT_1 STATE in addition to the processing 2697 * for the ESTABLISHED state if our FIN is now acknowledged 2698 * then enter FIN_WAIT_2. 2699 */ 2700 case TCPS_FIN_WAIT_1: 2701 if (ourfinisacked) { 2702 /* 2703 * If we can't receive any more 2704 * data, then closing user can proceed. 2705 * Starting the timer is contrary to the 2706 * specification, but if we don't get a FIN 2707 * we'll hang forever. 2708 * 2709 * XXXjl: 2710 * we should release the tp also, and use a 2711 * compressed state. 2712 */ 2713 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2714 soisdisconnected(so); 2715 tcp_timer_activate(tp, TT_2MSL, 2716 (tcp_fast_finwait2_recycle ? 2717 tcp_finwait2_timeout : 2718 TP_MAXIDLE(tp))); 2719 } 2720 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2721 } 2722 break; 2723 2724 /* 2725 * In CLOSING STATE in addition to the processing for 2726 * the ESTABLISHED state if the ACK acknowledges our FIN 2727 * then enter the TIME-WAIT state, otherwise ignore 2728 * the segment. 2729 */ 2730 case TCPS_CLOSING: 2731 if (ourfinisacked) { 2732 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2733 tcp_twstart(tp); 2734 INP_INFO_WUNLOCK(&V_tcbinfo); 2735 m_freem(m); 2736 return; 2737 } 2738 break; 2739 2740 /* 2741 * In LAST_ACK, we may still be waiting for data to drain 2742 * and/or to be acked, as well as for the ack of our FIN. 2743 * If our FIN is now acknowledged, delete the TCB, 2744 * enter the closed state and return. 2745 */ 2746 case TCPS_LAST_ACK: 2747 if (ourfinisacked) { 2748 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2749 tp = tcp_close(tp); 2750 goto drop; 2751 } 2752 break; 2753 } 2754 } 2755 2756 step6: 2757 INP_WLOCK_ASSERT(tp->t_inpcb); 2758 2759 /* 2760 * Update window information. 2761 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2762 */ 2763 if ((thflags & TH_ACK) && 2764 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2765 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2766 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2767 /* keep track of pure window updates */ 2768 if (tlen == 0 && 2769 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2770 TCPSTAT_INC(tcps_rcvwinupd); 2771 tp->snd_wnd = tiwin; 2772 tp->snd_wl1 = th->th_seq; 2773 tp->snd_wl2 = th->th_ack; 2774 if (tp->snd_wnd > tp->max_sndwnd) 2775 tp->max_sndwnd = tp->snd_wnd; 2776 needoutput = 1; 2777 } 2778 2779 /* 2780 * Process segments with URG. 2781 */ 2782 if ((thflags & TH_URG) && th->th_urp && 2783 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2784 /* 2785 * This is a kludge, but if we receive and accept 2786 * random urgent pointers, we'll crash in 2787 * soreceive. It's hard to imagine someone 2788 * actually wanting to send this much urgent data. 2789 */ 2790 SOCKBUF_LOCK(&so->so_rcv); 2791 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2792 th->th_urp = 0; /* XXX */ 2793 thflags &= ~TH_URG; /* XXX */ 2794 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2795 goto dodata; /* XXX */ 2796 } 2797 /* 2798 * If this segment advances the known urgent pointer, 2799 * then mark the data stream. This should not happen 2800 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2801 * a FIN has been received from the remote side. 2802 * In these states we ignore the URG. 2803 * 2804 * According to RFC961 (Assigned Protocols), 2805 * the urgent pointer points to the last octet 2806 * of urgent data. We continue, however, 2807 * to consider it to indicate the first octet 2808 * of data past the urgent section as the original 2809 * spec states (in one of two places). 2810 */ 2811 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2812 tp->rcv_up = th->th_seq + th->th_urp; 2813 so->so_oobmark = so->so_rcv.sb_cc + 2814 (tp->rcv_up - tp->rcv_nxt) - 1; 2815 if (so->so_oobmark == 0) 2816 so->so_rcv.sb_state |= SBS_RCVATMARK; 2817 sohasoutofband(so); 2818 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2819 } 2820 SOCKBUF_UNLOCK(&so->so_rcv); 2821 /* 2822 * Remove out of band data so doesn't get presented to user. 2823 * This can happen independent of advancing the URG pointer, 2824 * but if two URG's are pending at once, some out-of-band 2825 * data may creep in... ick. 2826 */ 2827 if (th->th_urp <= (u_long)tlen && 2828 !(so->so_options & SO_OOBINLINE)) { 2829 /* hdr drop is delayed */ 2830 tcp_pulloutofband(so, th, m, drop_hdrlen); 2831 } 2832 } else { 2833 /* 2834 * If no out of band data is expected, 2835 * pull receive urgent pointer along 2836 * with the receive window. 2837 */ 2838 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2839 tp->rcv_up = tp->rcv_nxt; 2840 } 2841 dodata: /* XXX */ 2842 INP_WLOCK_ASSERT(tp->t_inpcb); 2843 2844 /* 2845 * Process the segment text, merging it into the TCP sequencing queue, 2846 * and arranging for acknowledgment of receipt if necessary. 2847 * This process logically involves adjusting tp->rcv_wnd as data 2848 * is presented to the user (this happens in tcp_usrreq.c, 2849 * case PRU_RCVD). If a FIN has already been received on this 2850 * connection then we just ignore the text. 2851 */ 2852 if ((tlen || (thflags & TH_FIN)) && 2853 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2854 tcp_seq save_start = th->th_seq; 2855 m_adj(m, drop_hdrlen); /* delayed header drop */ 2856 /* 2857 * Insert segment which includes th into TCP reassembly queue 2858 * with control block tp. Set thflags to whether reassembly now 2859 * includes a segment with FIN. This handles the common case 2860 * inline (segment is the next to be received on an established 2861 * connection, and the queue is empty), avoiding linkage into 2862 * and removal from the queue and repetition of various 2863 * conversions. 2864 * Set DELACK for segments received in order, but ack 2865 * immediately when segments are out of order (so 2866 * fast retransmit can work). 2867 */ 2868 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL && 2869 TCPS_HAVEESTABLISHED(tp->t_state)) { 2870 if (DELAY_ACK(tp, tlen)) 2871 tp->t_flags |= TF_DELACK; 2872 else 2873 tp->t_flags |= TF_ACKNOW; 2874 tp->rcv_nxt += tlen; 2875 thflags = th->th_flags & TH_FIN; 2876 TCPSTAT_INC(tcps_rcvpack); 2877 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2878 ND6_HINT(tp); 2879 SOCKBUF_LOCK(&so->so_rcv); 2880 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2881 m_freem(m); 2882 else 2883 sbappendstream_locked(&so->so_rcv, m); 2884 /* NB: sorwakeup_locked() does an implicit unlock. */ 2885 sorwakeup_locked(so); 2886 } else { 2887 /* 2888 * XXX: Due to the header drop above "th" is 2889 * theoretically invalid by now. Fortunately 2890 * m_adj() doesn't actually frees any mbufs 2891 * when trimming from the head. 2892 */ 2893 thflags = tcp_reass(tp, th, &tlen, m); 2894 tp->t_flags |= TF_ACKNOW; 2895 } 2896 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2897 tcp_update_sack_list(tp, save_start, save_start + tlen); 2898 #if 0 2899 /* 2900 * Note the amount of data that peer has sent into 2901 * our window, in order to estimate the sender's 2902 * buffer size. 2903 * XXX: Unused. 2904 */ 2905 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2906 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2907 else 2908 len = so->so_rcv.sb_hiwat; 2909 #endif 2910 } else { 2911 m_freem(m); 2912 thflags &= ~TH_FIN; 2913 } 2914 2915 /* 2916 * If FIN is received ACK the FIN and let the user know 2917 * that the connection is closing. 2918 */ 2919 if (thflags & TH_FIN) { 2920 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2921 socantrcvmore(so); 2922 /* 2923 * If connection is half-synchronized 2924 * (ie NEEDSYN flag on) then delay ACK, 2925 * so it may be piggybacked when SYN is sent. 2926 * Otherwise, since we received a FIN then no 2927 * more input can be expected, send ACK now. 2928 */ 2929 if (tp->t_flags & TF_NEEDSYN) 2930 tp->t_flags |= TF_DELACK; 2931 else 2932 tp->t_flags |= TF_ACKNOW; 2933 tp->rcv_nxt++; 2934 } 2935 switch (tp->t_state) { 2936 2937 /* 2938 * In SYN_RECEIVED and ESTABLISHED STATES 2939 * enter the CLOSE_WAIT state. 2940 */ 2941 case TCPS_SYN_RECEIVED: 2942 tp->t_starttime = ticks; 2943 /* FALLTHROUGH */ 2944 case TCPS_ESTABLISHED: 2945 tcp_state_change(tp, TCPS_CLOSE_WAIT); 2946 break; 2947 2948 /* 2949 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2950 * enter the CLOSING state. 2951 */ 2952 case TCPS_FIN_WAIT_1: 2953 tcp_state_change(tp, TCPS_CLOSING); 2954 break; 2955 2956 /* 2957 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2958 * starting the time-wait timer, turning off the other 2959 * standard timers. 2960 */ 2961 case TCPS_FIN_WAIT_2: 2962 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2963 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2964 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2965 ti_locked)); 2966 2967 tcp_twstart(tp); 2968 INP_INFO_WUNLOCK(&V_tcbinfo); 2969 return; 2970 } 2971 } 2972 if (ti_locked == TI_WLOCKED) 2973 INP_INFO_WUNLOCK(&V_tcbinfo); 2974 ti_locked = TI_UNLOCKED; 2975 2976 #ifdef TCPDEBUG 2977 if (so->so_options & SO_DEBUG) 2978 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2979 &tcp_savetcp, 0); 2980 #endif 2981 2982 /* 2983 * Return any desired output. 2984 */ 2985 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2986 (void) tcp_output(tp); 2987 2988 check_delack: 2989 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2990 __func__, ti_locked)); 2991 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2992 INP_WLOCK_ASSERT(tp->t_inpcb); 2993 2994 if (tp->t_flags & TF_DELACK) { 2995 tp->t_flags &= ~TF_DELACK; 2996 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2997 } 2998 INP_WUNLOCK(tp->t_inpcb); 2999 return; 3000 3001 dropafterack: 3002 /* 3003 * Generate an ACK dropping incoming segment if it occupies 3004 * sequence space, where the ACK reflects our state. 3005 * 3006 * We can now skip the test for the RST flag since all 3007 * paths to this code happen after packets containing 3008 * RST have been dropped. 3009 * 3010 * In the SYN-RECEIVED state, don't send an ACK unless the 3011 * segment we received passes the SYN-RECEIVED ACK test. 3012 * If it fails send a RST. This breaks the loop in the 3013 * "LAND" DoS attack, and also prevents an ACK storm 3014 * between two listening ports that have been sent forged 3015 * SYN segments, each with the source address of the other. 3016 */ 3017 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3018 (SEQ_GT(tp->snd_una, th->th_ack) || 3019 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3020 rstreason = BANDLIM_RST_OPENPORT; 3021 goto dropwithreset; 3022 } 3023 #ifdef TCPDEBUG 3024 if (so->so_options & SO_DEBUG) 3025 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3026 &tcp_savetcp, 0); 3027 #endif 3028 if (ti_locked == TI_WLOCKED) 3029 INP_INFO_WUNLOCK(&V_tcbinfo); 3030 ti_locked = TI_UNLOCKED; 3031 3032 tp->t_flags |= TF_ACKNOW; 3033 (void) tcp_output(tp); 3034 INP_WUNLOCK(tp->t_inpcb); 3035 m_freem(m); 3036 return; 3037 3038 dropwithreset: 3039 if (ti_locked == TI_WLOCKED) 3040 INP_INFO_WUNLOCK(&V_tcbinfo); 3041 ti_locked = TI_UNLOCKED; 3042 3043 if (tp != NULL) { 3044 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3045 INP_WUNLOCK(tp->t_inpcb); 3046 } else 3047 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3048 return; 3049 3050 drop: 3051 if (ti_locked == TI_WLOCKED) { 3052 INP_INFO_WUNLOCK(&V_tcbinfo); 3053 ti_locked = TI_UNLOCKED; 3054 } 3055 #ifdef INVARIANTS 3056 else 3057 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3058 #endif 3059 3060 /* 3061 * Drop space held by incoming segment and return. 3062 */ 3063 #ifdef TCPDEBUG 3064 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3065 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3066 &tcp_savetcp, 0); 3067 #endif 3068 if (tp != NULL) 3069 INP_WUNLOCK(tp->t_inpcb); 3070 m_freem(m); 3071 } 3072 3073 /* 3074 * Issue RST and make ACK acceptable to originator of segment. 3075 * The mbuf must still include the original packet header. 3076 * tp may be NULL. 3077 */ 3078 static void 3079 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3080 int tlen, int rstreason) 3081 { 3082 #ifdef INET 3083 struct ip *ip; 3084 #endif 3085 #ifdef INET6 3086 struct ip6_hdr *ip6; 3087 #endif 3088 3089 if (tp != NULL) { 3090 INP_WLOCK_ASSERT(tp->t_inpcb); 3091 } 3092 3093 /* Don't bother if destination was broadcast/multicast. */ 3094 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3095 goto drop; 3096 #ifdef INET6 3097 if (mtod(m, struct ip *)->ip_v == 6) { 3098 ip6 = mtod(m, struct ip6_hdr *); 3099 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3100 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3101 goto drop; 3102 /* IPv6 anycast check is done at tcp6_input() */ 3103 } 3104 #endif 3105 #if defined(INET) && defined(INET6) 3106 else 3107 #endif 3108 #ifdef INET 3109 { 3110 ip = mtod(m, struct ip *); 3111 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3112 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3113 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3114 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3115 goto drop; 3116 } 3117 #endif 3118 3119 /* Perform bandwidth limiting. */ 3120 if (badport_bandlim(rstreason) < 0) 3121 goto drop; 3122 3123 /* tcp_respond consumes the mbuf chain. */ 3124 if (th->th_flags & TH_ACK) { 3125 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3126 th->th_ack, TH_RST); 3127 } else { 3128 if (th->th_flags & TH_SYN) 3129 tlen++; 3130 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3131 (tcp_seq)0, TH_RST|TH_ACK); 3132 } 3133 return; 3134 drop: 3135 m_freem(m); 3136 } 3137 3138 /* 3139 * Parse TCP options and place in tcpopt. 3140 */ 3141 static void 3142 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3143 { 3144 int opt, optlen; 3145 3146 to->to_flags = 0; 3147 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3148 opt = cp[0]; 3149 if (opt == TCPOPT_EOL) 3150 break; 3151 if (opt == TCPOPT_NOP) 3152 optlen = 1; 3153 else { 3154 if (cnt < 2) 3155 break; 3156 optlen = cp[1]; 3157 if (optlen < 2 || optlen > cnt) 3158 break; 3159 } 3160 switch (opt) { 3161 case TCPOPT_MAXSEG: 3162 if (optlen != TCPOLEN_MAXSEG) 3163 continue; 3164 if (!(flags & TO_SYN)) 3165 continue; 3166 to->to_flags |= TOF_MSS; 3167 bcopy((char *)cp + 2, 3168 (char *)&to->to_mss, sizeof(to->to_mss)); 3169 to->to_mss = ntohs(to->to_mss); 3170 break; 3171 case TCPOPT_WINDOW: 3172 if (optlen != TCPOLEN_WINDOW) 3173 continue; 3174 if (!(flags & TO_SYN)) 3175 continue; 3176 to->to_flags |= TOF_SCALE; 3177 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3178 break; 3179 case TCPOPT_TIMESTAMP: 3180 if (optlen != TCPOLEN_TIMESTAMP) 3181 continue; 3182 to->to_flags |= TOF_TS; 3183 bcopy((char *)cp + 2, 3184 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3185 to->to_tsval = ntohl(to->to_tsval); 3186 bcopy((char *)cp + 6, 3187 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3188 to->to_tsecr = ntohl(to->to_tsecr); 3189 break; 3190 #ifdef TCP_SIGNATURE 3191 /* 3192 * XXX In order to reply to a host which has set the 3193 * TCP_SIGNATURE option in its initial SYN, we have to 3194 * record the fact that the option was observed here 3195 * for the syncache code to perform the correct response. 3196 */ 3197 case TCPOPT_SIGNATURE: 3198 if (optlen != TCPOLEN_SIGNATURE) 3199 continue; 3200 to->to_flags |= TOF_SIGNATURE; 3201 to->to_signature = cp + 2; 3202 break; 3203 #endif 3204 case TCPOPT_SACK_PERMITTED: 3205 if (optlen != TCPOLEN_SACK_PERMITTED) 3206 continue; 3207 if (!(flags & TO_SYN)) 3208 continue; 3209 if (!V_tcp_do_sack) 3210 continue; 3211 to->to_flags |= TOF_SACKPERM; 3212 break; 3213 case TCPOPT_SACK: 3214 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3215 continue; 3216 if (flags & TO_SYN) 3217 continue; 3218 to->to_flags |= TOF_SACK; 3219 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3220 to->to_sacks = cp + 2; 3221 TCPSTAT_INC(tcps_sack_rcv_blocks); 3222 break; 3223 default: 3224 continue; 3225 } 3226 } 3227 } 3228 3229 /* 3230 * Pull out of band byte out of a segment so 3231 * it doesn't appear in the user's data queue. 3232 * It is still reflected in the segment length for 3233 * sequencing purposes. 3234 */ 3235 static void 3236 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3237 int off) 3238 { 3239 int cnt = off + th->th_urp - 1; 3240 3241 while (cnt >= 0) { 3242 if (m->m_len > cnt) { 3243 char *cp = mtod(m, caddr_t) + cnt; 3244 struct tcpcb *tp = sototcpcb(so); 3245 3246 INP_WLOCK_ASSERT(tp->t_inpcb); 3247 3248 tp->t_iobc = *cp; 3249 tp->t_oobflags |= TCPOOB_HAVEDATA; 3250 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3251 m->m_len--; 3252 if (m->m_flags & M_PKTHDR) 3253 m->m_pkthdr.len--; 3254 return; 3255 } 3256 cnt -= m->m_len; 3257 m = m->m_next; 3258 if (m == NULL) 3259 break; 3260 } 3261 panic("tcp_pulloutofband"); 3262 } 3263 3264 /* 3265 * Collect new round-trip time estimate 3266 * and update averages and current timeout. 3267 */ 3268 static void 3269 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3270 { 3271 int delta; 3272 3273 INP_WLOCK_ASSERT(tp->t_inpcb); 3274 3275 TCPSTAT_INC(tcps_rttupdated); 3276 tp->t_rttupdated++; 3277 if (tp->t_srtt != 0) { 3278 /* 3279 * srtt is stored as fixed point with 5 bits after the 3280 * binary point (i.e., scaled by 8). The following magic 3281 * is equivalent to the smoothing algorithm in rfc793 with 3282 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3283 * point). Adjust rtt to origin 0. 3284 */ 3285 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3286 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3287 3288 if ((tp->t_srtt += delta) <= 0) 3289 tp->t_srtt = 1; 3290 3291 /* 3292 * We accumulate a smoothed rtt variance (actually, a 3293 * smoothed mean difference), then set the retransmit 3294 * timer to smoothed rtt + 4 times the smoothed variance. 3295 * rttvar is stored as fixed point with 4 bits after the 3296 * binary point (scaled by 16). The following is 3297 * equivalent to rfc793 smoothing with an alpha of .75 3298 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3299 * rfc793's wired-in beta. 3300 */ 3301 if (delta < 0) 3302 delta = -delta; 3303 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3304 if ((tp->t_rttvar += delta) <= 0) 3305 tp->t_rttvar = 1; 3306 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3307 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3308 } else { 3309 /* 3310 * No rtt measurement yet - use the unsmoothed rtt. 3311 * Set the variance to half the rtt (so our first 3312 * retransmit happens at 3*rtt). 3313 */ 3314 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3315 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3316 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3317 } 3318 tp->t_rtttime = 0; 3319 tp->t_rxtshift = 0; 3320 3321 /* 3322 * the retransmit should happen at rtt + 4 * rttvar. 3323 * Because of the way we do the smoothing, srtt and rttvar 3324 * will each average +1/2 tick of bias. When we compute 3325 * the retransmit timer, we want 1/2 tick of rounding and 3326 * 1 extra tick because of +-1/2 tick uncertainty in the 3327 * firing of the timer. The bias will give us exactly the 3328 * 1.5 tick we need. But, because the bias is 3329 * statistical, we have to test that we don't drop below 3330 * the minimum feasible timer (which is 2 ticks). 3331 */ 3332 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3333 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3334 3335 /* 3336 * We received an ack for a packet that wasn't retransmitted; 3337 * it is probably safe to discard any error indications we've 3338 * received recently. This isn't quite right, but close enough 3339 * for now (a route might have failed after we sent a segment, 3340 * and the return path might not be symmetrical). 3341 */ 3342 tp->t_softerror = 0; 3343 } 3344 3345 /* 3346 * Determine a reasonable value for maxseg size. 3347 * If the route is known, check route for mtu. 3348 * If none, use an mss that can be handled on the outgoing interface 3349 * without forcing IP to fragment. If no route is found, route has no mtu, 3350 * or the destination isn't local, use a default, hopefully conservative 3351 * size (usually 512 or the default IP max size, but no more than the mtu 3352 * of the interface), as we can't discover anything about intervening 3353 * gateways or networks. We also initialize the congestion/slow start 3354 * window to be a single segment if the destination isn't local. 3355 * While looking at the routing entry, we also initialize other path-dependent 3356 * parameters from pre-set or cached values in the routing entry. 3357 * 3358 * Also take into account the space needed for options that we 3359 * send regularly. Make maxseg shorter by that amount to assure 3360 * that we can send maxseg amount of data even when the options 3361 * are present. Store the upper limit of the length of options plus 3362 * data in maxopd. 3363 * 3364 * NOTE that this routine is only called when we process an incoming 3365 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3366 * settings are handled in tcp_mssopt(). 3367 */ 3368 void 3369 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3370 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3371 { 3372 int mss = 0; 3373 u_long maxmtu = 0; 3374 struct inpcb *inp = tp->t_inpcb; 3375 struct hc_metrics_lite metrics; 3376 int origoffer; 3377 #ifdef INET6 3378 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3379 size_t min_protoh = isipv6 ? 3380 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3381 sizeof (struct tcpiphdr); 3382 #else 3383 const size_t min_protoh = sizeof(struct tcpiphdr); 3384 #endif 3385 3386 INP_WLOCK_ASSERT(tp->t_inpcb); 3387 3388 if (mtuoffer != -1) { 3389 KASSERT(offer == -1, ("%s: conflict", __func__)); 3390 offer = mtuoffer - min_protoh; 3391 } 3392 origoffer = offer; 3393 3394 /* Initialize. */ 3395 #ifdef INET6 3396 if (isipv6) { 3397 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3398 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3399 } 3400 #endif 3401 #if defined(INET) && defined(INET6) 3402 else 3403 #endif 3404 #ifdef INET 3405 { 3406 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3407 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3408 } 3409 #endif 3410 3411 /* 3412 * No route to sender, stay with default mss and return. 3413 */ 3414 if (maxmtu == 0) { 3415 /* 3416 * In case we return early we need to initialize metrics 3417 * to a defined state as tcp_hc_get() would do for us 3418 * if there was no cache hit. 3419 */ 3420 if (metricptr != NULL) 3421 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3422 return; 3423 } 3424 3425 /* What have we got? */ 3426 switch (offer) { 3427 case 0: 3428 /* 3429 * Offer == 0 means that there was no MSS on the SYN 3430 * segment, in this case we use tcp_mssdflt as 3431 * already assigned to t_maxopd above. 3432 */ 3433 offer = tp->t_maxopd; 3434 break; 3435 3436 case -1: 3437 /* 3438 * Offer == -1 means that we didn't receive SYN yet. 3439 */ 3440 /* FALLTHROUGH */ 3441 3442 default: 3443 /* 3444 * Prevent DoS attack with too small MSS. Round up 3445 * to at least minmss. 3446 */ 3447 offer = max(offer, V_tcp_minmss); 3448 } 3449 3450 /* 3451 * rmx information is now retrieved from tcp_hostcache. 3452 */ 3453 tcp_hc_get(&inp->inp_inc, &metrics); 3454 if (metricptr != NULL) 3455 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3456 3457 /* 3458 * If there's a discovered mtu in tcp hostcache, use it. 3459 * Else, use the link mtu. 3460 */ 3461 if (metrics.rmx_mtu) 3462 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3463 else { 3464 #ifdef INET6 3465 if (isipv6) { 3466 mss = maxmtu - min_protoh; 3467 if (!V_path_mtu_discovery && 3468 !in6_localaddr(&inp->in6p_faddr)) 3469 mss = min(mss, V_tcp_v6mssdflt); 3470 } 3471 #endif 3472 #if defined(INET) && defined(INET6) 3473 else 3474 #endif 3475 #ifdef INET 3476 { 3477 mss = maxmtu - min_protoh; 3478 if (!V_path_mtu_discovery && 3479 !in_localaddr(inp->inp_faddr)) 3480 mss = min(mss, V_tcp_mssdflt); 3481 } 3482 #endif 3483 /* 3484 * XXX - The above conditional (mss = maxmtu - min_protoh) 3485 * probably violates the TCP spec. 3486 * The problem is that, since we don't know the 3487 * other end's MSS, we are supposed to use a conservative 3488 * default. But, if we do that, then MTU discovery will 3489 * never actually take place, because the conservative 3490 * default is much less than the MTUs typically seen 3491 * on the Internet today. For the moment, we'll sweep 3492 * this under the carpet. 3493 * 3494 * The conservative default might not actually be a problem 3495 * if the only case this occurs is when sending an initial 3496 * SYN with options and data to a host we've never talked 3497 * to before. Then, they will reply with an MSS value which 3498 * will get recorded and the new parameters should get 3499 * recomputed. For Further Study. 3500 */ 3501 } 3502 mss = min(mss, offer); 3503 3504 /* 3505 * Sanity check: make sure that maxopd will be large 3506 * enough to allow some data on segments even if the 3507 * all the option space is used (40bytes). Otherwise 3508 * funny things may happen in tcp_output. 3509 */ 3510 mss = max(mss, 64); 3511 3512 /* 3513 * maxopd stores the maximum length of data AND options 3514 * in a segment; maxseg is the amount of data in a normal 3515 * segment. We need to store this value (maxopd) apart 3516 * from maxseg, because now every segment carries options 3517 * and thus we normally have somewhat less data in segments. 3518 */ 3519 tp->t_maxopd = mss; 3520 3521 /* 3522 * origoffer==-1 indicates that no segments were received yet. 3523 * In this case we just guess. 3524 */ 3525 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3526 (origoffer == -1 || 3527 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3528 mss -= TCPOLEN_TSTAMP_APPA; 3529 3530 tp->t_maxseg = mss; 3531 } 3532 3533 void 3534 tcp_mss(struct tcpcb *tp, int offer) 3535 { 3536 int mss; 3537 u_long bufsize; 3538 struct inpcb *inp; 3539 struct socket *so; 3540 struct hc_metrics_lite metrics; 3541 struct tcp_ifcap cap; 3542 3543 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3544 3545 bzero(&cap, sizeof(cap)); 3546 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3547 3548 mss = tp->t_maxseg; 3549 inp = tp->t_inpcb; 3550 3551 /* 3552 * If there's a pipesize, change the socket buffer to that size, 3553 * don't change if sb_hiwat is different than default (then it 3554 * has been changed on purpose with setsockopt). 3555 * Make the socket buffers an integral number of mss units; 3556 * if the mss is larger than the socket buffer, decrease the mss. 3557 */ 3558 so = inp->inp_socket; 3559 SOCKBUF_LOCK(&so->so_snd); 3560 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3561 bufsize = metrics.rmx_sendpipe; 3562 else 3563 bufsize = so->so_snd.sb_hiwat; 3564 if (bufsize < mss) 3565 mss = bufsize; 3566 else { 3567 bufsize = roundup(bufsize, mss); 3568 if (bufsize > sb_max) 3569 bufsize = sb_max; 3570 if (bufsize > so->so_snd.sb_hiwat) 3571 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3572 } 3573 SOCKBUF_UNLOCK(&so->so_snd); 3574 tp->t_maxseg = mss; 3575 3576 SOCKBUF_LOCK(&so->so_rcv); 3577 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3578 bufsize = metrics.rmx_recvpipe; 3579 else 3580 bufsize = so->so_rcv.sb_hiwat; 3581 if (bufsize > mss) { 3582 bufsize = roundup(bufsize, mss); 3583 if (bufsize > sb_max) 3584 bufsize = sb_max; 3585 if (bufsize > so->so_rcv.sb_hiwat) 3586 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3587 } 3588 SOCKBUF_UNLOCK(&so->so_rcv); 3589 3590 /* Check the interface for TSO capabilities. */ 3591 if (cap.ifcap & CSUM_TSO) { 3592 tp->t_flags |= TF_TSO; 3593 tp->t_tsomax = cap.tsomax; 3594 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3595 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3596 } 3597 } 3598 3599 /* 3600 * Determine the MSS option to send on an outgoing SYN. 3601 */ 3602 int 3603 tcp_mssopt(struct in_conninfo *inc) 3604 { 3605 int mss = 0; 3606 u_long maxmtu = 0; 3607 u_long thcmtu = 0; 3608 size_t min_protoh; 3609 3610 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3611 3612 #ifdef INET6 3613 if (inc->inc_flags & INC_ISIPV6) { 3614 mss = V_tcp_v6mssdflt; 3615 maxmtu = tcp_maxmtu6(inc, NULL); 3616 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3617 } 3618 #endif 3619 #if defined(INET) && defined(INET6) 3620 else 3621 #endif 3622 #ifdef INET 3623 { 3624 mss = V_tcp_mssdflt; 3625 maxmtu = tcp_maxmtu(inc, NULL); 3626 min_protoh = sizeof(struct tcpiphdr); 3627 } 3628 #endif 3629 #if defined(INET6) || defined(INET) 3630 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3631 #endif 3632 3633 if (maxmtu && thcmtu) 3634 mss = min(maxmtu, thcmtu) - min_protoh; 3635 else if (maxmtu || thcmtu) 3636 mss = max(maxmtu, thcmtu) - min_protoh; 3637 3638 return (mss); 3639 } 3640 3641 3642 /* 3643 * On a partial ack arrives, force the retransmission of the 3644 * next unacknowledged segment. Do not clear tp->t_dupacks. 3645 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3646 * be started again. 3647 */ 3648 static void 3649 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3650 { 3651 tcp_seq onxt = tp->snd_nxt; 3652 u_long ocwnd = tp->snd_cwnd; 3653 3654 INP_WLOCK_ASSERT(tp->t_inpcb); 3655 3656 tcp_timer_activate(tp, TT_REXMT, 0); 3657 tp->t_rtttime = 0; 3658 tp->snd_nxt = th->th_ack; 3659 /* 3660 * Set snd_cwnd to one segment beyond acknowledged offset. 3661 * (tp->snd_una has not yet been updated when this function is called.) 3662 */ 3663 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3664 tp->t_flags |= TF_ACKNOW; 3665 (void) tcp_output(tp); 3666 tp->snd_cwnd = ocwnd; 3667 if (SEQ_GT(onxt, tp->snd_nxt)) 3668 tp->snd_nxt = onxt; 3669 /* 3670 * Partial window deflation. Relies on fact that tp->snd_una 3671 * not updated yet. 3672 */ 3673 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3674 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3675 else 3676 tp->snd_cwnd = 0; 3677 tp->snd_cwnd += tp->t_maxseg; 3678 } 3679