1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 4. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_ipfw.h" /* for ipfw_fwd */ 54 #include "opt_inet.h" 55 #include "opt_inet6.h" 56 #include "opt_ipsec.h" 57 #include "opt_tcpdebug.h" 58 59 #include <sys/param.h> 60 #include <sys/kernel.h> 61 #include <sys/hhook.h> 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/sdt.h> 67 #include <sys/signalvar.h> 68 #include <sys/socket.h> 69 #include <sys/socketvar.h> 70 #include <sys/sysctl.h> 71 #include <sys/syslog.h> 72 #include <sys/systm.h> 73 74 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 75 76 #include <vm/uma.h> 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/route.h> 81 #include <net/vnet.h> 82 83 #define TCPSTATES /* for logging */ 84 85 #include <netinet/cc.h> 86 #include <netinet/in.h> 87 #include <netinet/in_kdtrace.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/in_var.h> 91 #include <netinet/ip.h> 92 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 93 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 94 #include <netinet/ip_var.h> 95 #include <netinet/ip_options.h> 96 #include <netinet/ip6.h> 97 #include <netinet/icmp6.h> 98 #include <netinet6/in6_pcb.h> 99 #include <netinet6/ip6_var.h> 100 #include <netinet6/nd6.h> 101 #include <netinet/tcp_fsm.h> 102 #include <netinet/tcp_seq.h> 103 #include <netinet/tcp_timer.h> 104 #include <netinet/tcp_var.h> 105 #include <netinet6/tcp6_var.h> 106 #include <netinet/tcpip.h> 107 #include <netinet/tcp_syncache.h> 108 #ifdef TCPDEBUG 109 #include <netinet/tcp_debug.h> 110 #endif /* TCPDEBUG */ 111 #ifdef TCP_OFFLOAD 112 #include <netinet/tcp_offload.h> 113 #endif 114 115 #ifdef IPSEC 116 #include <netipsec/ipsec.h> 117 #include <netipsec/ipsec6.h> 118 #endif /*IPSEC*/ 119 120 #include <machine/in_cksum.h> 121 122 #include <security/mac/mac_framework.h> 123 124 const int tcprexmtthresh = 3; 125 126 int tcp_log_in_vain = 0; 127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 128 &tcp_log_in_vain, 0, 129 "Log all incoming TCP segments to closed ports"); 130 131 VNET_DEFINE(int, blackhole) = 0; 132 #define V_blackhole VNET(blackhole) 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 134 &VNET_NAME(blackhole), 0, 135 "Do not send RST on segments to closed ports"); 136 137 VNET_DEFINE(int, tcp_delack_enabled) = 1; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(tcp_delack_enabled), 0, 140 "Delay ACK to try and piggyback it onto a data packet"); 141 142 VNET_DEFINE(int, drop_synfin) = 0; 143 #define V_drop_synfin VNET(drop_synfin) 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 145 &VNET_NAME(drop_synfin), 0, 146 "Drop TCP packets with SYN+FIN set"); 147 148 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 149 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 151 &VNET_NAME(tcp_do_rfc3042), 0, 152 "Enable RFC 3042 (Limited Transmit)"); 153 154 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 156 &VNET_NAME(tcp_do_rfc3390), 0, 157 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 158 159 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 160 "Experimental TCP extensions"); 161 162 VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 163 SYSCTL_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_initcwnd10), 0, 165 "Enable RFC 6928 (Increasing initial CWND to 10)"); 166 167 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_rfc3465), 0, 170 "Enable RFC 3465 (Appropriate Byte Counting)"); 171 172 VNET_DEFINE(int, tcp_abc_l_var) = 2; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_abc_l_var), 2, 175 "Cap the max cwnd increment during slow-start to this number of segments"); 176 177 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 178 179 VNET_DEFINE(int, tcp_do_ecn) = 0; 180 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_do_ecn), 0, 182 "TCP ECN support"); 183 184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 185 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 186 &VNET_NAME(tcp_ecn_maxretries), 0, 187 "Max retries before giving up on ECN"); 188 189 VNET_DEFINE(int, tcp_insecure_syn) = 0; 190 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_insecure_syn), 0, 193 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 194 195 VNET_DEFINE(int, tcp_insecure_rst) = 0; 196 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_insecure_rst), 0, 199 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 200 201 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 202 #define V_tcp_recvspace VNET(tcp_recvspace) 203 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 204 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 205 206 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 207 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 209 &VNET_NAME(tcp_do_autorcvbuf), 0, 210 "Enable automatic receive buffer sizing"); 211 212 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 213 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 215 &VNET_NAME(tcp_autorcvbuf_inc), 0, 216 "Incrementor step size of automatic receive buffer"); 217 218 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 219 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 221 &VNET_NAME(tcp_autorcvbuf_max), 0, 222 "Max size of automatic receive buffer"); 223 224 VNET_DEFINE(struct inpcbhead, tcb); 225 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 226 VNET_DEFINE(struct inpcbinfo, tcbinfo); 227 228 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 229 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 230 struct socket *, struct tcpcb *, int, int, uint8_t, 231 int); 232 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 233 struct tcpcb *, int, int); 234 static void tcp_pulloutofband(struct socket *, 235 struct tcphdr *, struct mbuf *, int); 236 static void tcp_xmit_timer(struct tcpcb *, int); 237 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 238 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 239 uint16_t type); 240 static void inline cc_conn_init(struct tcpcb *tp); 241 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 242 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 243 struct tcphdr *th, struct tcpopt *to); 244 245 /* 246 * TCP statistics are stored in an "array" of counter(9)s. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 VNET_PCPUSTAT_SYSINIT(tcpstat); 250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252 253 #ifdef VIMAGE 254 VNET_PCPUSTAT_SYSUNINIT(tcpstat); 255 #endif /* VIMAGE */ 256 /* 257 * Kernel module interface for updating tcpstat. The argument is an index 258 * into tcpstat treated as an array. 259 */ 260 void 261 kmod_tcpstat_inc(int statnum) 262 { 263 264 counter_u64_add(VNET(tcpstat)[statnum], 1); 265 } 266 267 /* 268 * Wrapper for the TCP established input helper hook. 269 */ 270 static void inline 271 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 272 { 273 struct tcp_hhook_data hhook_data; 274 275 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 276 hhook_data.tp = tp; 277 hhook_data.th = th; 278 hhook_data.to = to; 279 280 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 281 tp->osd); 282 } 283 } 284 285 /* 286 * CC wrapper hook functions 287 */ 288 static void inline 289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 290 { 291 INP_WLOCK_ASSERT(tp->t_inpcb); 292 293 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 294 if (tp->snd_cwnd <= tp->snd_wnd) 295 tp->ccv->flags |= CCF_CWND_LIMITED; 296 else 297 tp->ccv->flags &= ~CCF_CWND_LIMITED; 298 299 if (type == CC_ACK) { 300 if (tp->snd_cwnd > tp->snd_ssthresh) { 301 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 302 V_tcp_abc_l_var * tp->t_maxseg); 303 if (tp->t_bytes_acked >= tp->snd_cwnd) { 304 tp->t_bytes_acked -= tp->snd_cwnd; 305 tp->ccv->flags |= CCF_ABC_SENTAWND; 306 } 307 } else { 308 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 309 tp->t_bytes_acked = 0; 310 } 311 } 312 313 if (CC_ALGO(tp)->ack_received != NULL) { 314 /* XXXLAS: Find a way to live without this */ 315 tp->ccv->curack = th->th_ack; 316 CC_ALGO(tp)->ack_received(tp->ccv, type); 317 } 318 } 319 320 static void inline 321 cc_conn_init(struct tcpcb *tp) 322 { 323 struct hc_metrics_lite metrics; 324 struct inpcb *inp = tp->t_inpcb; 325 int rtt; 326 327 INP_WLOCK_ASSERT(tp->t_inpcb); 328 329 tcp_hc_get(&inp->inp_inc, &metrics); 330 331 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 332 tp->t_srtt = rtt; 333 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 334 TCPSTAT_INC(tcps_usedrtt); 335 if (metrics.rmx_rttvar) { 336 tp->t_rttvar = metrics.rmx_rttvar; 337 TCPSTAT_INC(tcps_usedrttvar); 338 } else { 339 /* default variation is +- 1 rtt */ 340 tp->t_rttvar = 341 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 342 } 343 TCPT_RANGESET(tp->t_rxtcur, 344 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 345 tp->t_rttmin, TCPTV_REXMTMAX); 346 } 347 if (metrics.rmx_ssthresh) { 348 /* 349 * There's some sort of gateway or interface 350 * buffer limit on the path. Use this to set 351 * the slow start threshhold, but set the 352 * threshold to no less than 2*mss. 353 */ 354 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 355 TCPSTAT_INC(tcps_usedssthresh); 356 } 357 358 /* 359 * Set the initial slow-start flight size. 360 * 361 * RFC5681 Section 3.1 specifies the default conservative values. 362 * RFC3390 specifies slightly more aggressive values. 363 * RFC6928 increases it to ten segments. 364 * 365 * If a SYN or SYN/ACK was lost and retransmitted, we have to 366 * reduce the initial CWND to one segment as congestion is likely 367 * requiring us to be cautious. 368 */ 369 if (tp->snd_cwnd == 1) 370 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 371 else if (V_tcp_do_initcwnd10) 372 tp->snd_cwnd = min(10 * tp->t_maxseg, 373 max(2 * tp->t_maxseg, 14600)); 374 else if (V_tcp_do_rfc3390) 375 tp->snd_cwnd = min(4 * tp->t_maxseg, 376 max(2 * tp->t_maxseg, 4380)); 377 else { 378 /* Per RFC5681 Section 3.1 */ 379 if (tp->t_maxseg > 2190) 380 tp->snd_cwnd = 2 * tp->t_maxseg; 381 else if (tp->t_maxseg > 1095) 382 tp->snd_cwnd = 3 * tp->t_maxseg; 383 else 384 tp->snd_cwnd = 4 * tp->t_maxseg; 385 } 386 387 if (CC_ALGO(tp)->conn_init != NULL) 388 CC_ALGO(tp)->conn_init(tp->ccv); 389 } 390 391 void inline 392 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 393 { 394 INP_WLOCK_ASSERT(tp->t_inpcb); 395 396 switch(type) { 397 case CC_NDUPACK: 398 if (!IN_FASTRECOVERY(tp->t_flags)) { 399 tp->snd_recover = tp->snd_max; 400 if (tp->t_flags & TF_ECN_PERMIT) 401 tp->t_flags |= TF_ECN_SND_CWR; 402 } 403 break; 404 case CC_ECN: 405 if (!IN_CONGRECOVERY(tp->t_flags)) { 406 TCPSTAT_INC(tcps_ecn_rcwnd); 407 tp->snd_recover = tp->snd_max; 408 if (tp->t_flags & TF_ECN_PERMIT) 409 tp->t_flags |= TF_ECN_SND_CWR; 410 } 411 break; 412 case CC_RTO: 413 tp->t_dupacks = 0; 414 tp->t_bytes_acked = 0; 415 EXIT_RECOVERY(tp->t_flags); 416 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 417 tp->t_maxseg) * tp->t_maxseg; 418 tp->snd_cwnd = tp->t_maxseg; 419 break; 420 case CC_RTO_ERR: 421 TCPSTAT_INC(tcps_sndrexmitbad); 422 /* RTO was unnecessary, so reset everything. */ 423 tp->snd_cwnd = tp->snd_cwnd_prev; 424 tp->snd_ssthresh = tp->snd_ssthresh_prev; 425 tp->snd_recover = tp->snd_recover_prev; 426 if (tp->t_flags & TF_WASFRECOVERY) 427 ENTER_FASTRECOVERY(tp->t_flags); 428 if (tp->t_flags & TF_WASCRECOVERY) 429 ENTER_CONGRECOVERY(tp->t_flags); 430 tp->snd_nxt = tp->snd_max; 431 tp->t_flags &= ~TF_PREVVALID; 432 tp->t_badrxtwin = 0; 433 break; 434 } 435 436 if (CC_ALGO(tp)->cong_signal != NULL) { 437 if (th != NULL) 438 tp->ccv->curack = th->th_ack; 439 CC_ALGO(tp)->cong_signal(tp->ccv, type); 440 } 441 } 442 443 static void inline 444 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 445 { 446 INP_WLOCK_ASSERT(tp->t_inpcb); 447 448 /* XXXLAS: KASSERT that we're in recovery? */ 449 450 if (CC_ALGO(tp)->post_recovery != NULL) { 451 tp->ccv->curack = th->th_ack; 452 CC_ALGO(tp)->post_recovery(tp->ccv); 453 } 454 /* XXXLAS: EXIT_RECOVERY ? */ 455 tp->t_bytes_acked = 0; 456 } 457 458 #ifdef TCP_SIGNATURE 459 static inline int 460 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 461 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 462 { 463 int ret; 464 465 tcp_fields_to_net(th); 466 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 467 tcp_fields_to_host(th); 468 return (ret); 469 } 470 #endif 471 472 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 473 #ifdef INET6 474 #define ND6_HINT(tp) \ 475 do { \ 476 if ((tp) && (tp)->t_inpcb && \ 477 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 478 nd6_nud_hint(NULL, NULL, 0); \ 479 } while (0) 480 #else 481 #define ND6_HINT(tp) 482 #endif 483 484 /* 485 * Indicate whether this ack should be delayed. We can delay the ack if 486 * following conditions are met: 487 * - There is no delayed ack timer in progress. 488 * - Our last ack wasn't a 0-sized window. We never want to delay 489 * the ack that opens up a 0-sized window. 490 * - LRO wasn't used for this segment. We make sure by checking that the 491 * segment size is not larger than the MSS. 492 * - Delayed acks are enabled or this is a half-synchronized T/TCP 493 * connection. 494 */ 495 #define DELAY_ACK(tp, tlen) \ 496 ((!tcp_timer_active(tp, TT_DELACK) && \ 497 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 498 (tlen <= tp->t_maxopd) && \ 499 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 500 501 /* 502 * TCP input handling is split into multiple parts: 503 * tcp6_input is a thin wrapper around tcp_input for the extended 504 * ip6_protox[] call format in ip6_input 505 * tcp_input handles primary segment validation, inpcb lookup and 506 * SYN processing on listen sockets 507 * tcp_do_segment processes the ACK and text of the segment for 508 * establishing, established and closing connections 509 */ 510 #ifdef INET6 511 int 512 tcp6_input(struct mbuf **mp, int *offp, int proto) 513 { 514 struct mbuf *m = *mp; 515 struct in6_ifaddr *ia6; 516 struct ip6_hdr *ip6; 517 518 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 519 520 /* 521 * draft-itojun-ipv6-tcp-to-anycast 522 * better place to put this in? 523 */ 524 ip6 = mtod(m, struct ip6_hdr *); 525 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 526 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 527 struct ip6_hdr *ip6; 528 529 ifa_free(&ia6->ia_ifa); 530 ip6 = mtod(m, struct ip6_hdr *); 531 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 532 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 533 return (IPPROTO_DONE); 534 } 535 if (ia6) 536 ifa_free(&ia6->ia_ifa); 537 538 return (tcp_input(mp, offp, proto)); 539 } 540 #endif /* INET6 */ 541 542 int 543 tcp_input(struct mbuf **mp, int *offp, int proto) 544 { 545 struct mbuf *m = *mp; 546 struct tcphdr *th = NULL; 547 struct ip *ip = NULL; 548 struct inpcb *inp = NULL; 549 struct tcpcb *tp = NULL; 550 struct socket *so = NULL; 551 u_char *optp = NULL; 552 int off0; 553 int optlen = 0; 554 #ifdef INET 555 int len; 556 #endif 557 int tlen = 0, off; 558 int drop_hdrlen; 559 int thflags; 560 int rstreason = 0; /* For badport_bandlim accounting purposes */ 561 #ifdef TCP_SIGNATURE 562 uint8_t sig_checked = 0; 563 #endif 564 uint8_t iptos = 0; 565 struct m_tag *fwd_tag = NULL; 566 #ifdef INET6 567 struct ip6_hdr *ip6 = NULL; 568 int isipv6; 569 #else 570 const void *ip6 = NULL; 571 #endif /* INET6 */ 572 struct tcpopt to; /* options in this segment */ 573 char *s = NULL; /* address and port logging */ 574 int ti_locked; 575 #define TI_UNLOCKED 1 576 #define TI_WLOCKED 2 577 578 #ifdef TCPDEBUG 579 /* 580 * The size of tcp_saveipgen must be the size of the max ip header, 581 * now IPv6. 582 */ 583 u_char tcp_saveipgen[IP6_HDR_LEN]; 584 struct tcphdr tcp_savetcp; 585 short ostate = 0; 586 #endif 587 588 #ifdef INET6 589 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 590 #endif 591 592 off0 = *offp; 593 m = *mp; 594 *mp = NULL; 595 to.to_flags = 0; 596 TCPSTAT_INC(tcps_rcvtotal); 597 598 #ifdef INET6 599 if (isipv6) { 600 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 601 602 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 603 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 604 if (m == NULL) { 605 TCPSTAT_INC(tcps_rcvshort); 606 return (IPPROTO_DONE); 607 } 608 } 609 610 ip6 = mtod(m, struct ip6_hdr *); 611 th = (struct tcphdr *)((caddr_t)ip6 + off0); 612 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 613 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 614 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 615 th->th_sum = m->m_pkthdr.csum_data; 616 else 617 th->th_sum = in6_cksum_pseudo(ip6, tlen, 618 IPPROTO_TCP, m->m_pkthdr.csum_data); 619 th->th_sum ^= 0xffff; 620 } else 621 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 622 if (th->th_sum) { 623 TCPSTAT_INC(tcps_rcvbadsum); 624 goto drop; 625 } 626 627 /* 628 * Be proactive about unspecified IPv6 address in source. 629 * As we use all-zero to indicate unbounded/unconnected pcb, 630 * unspecified IPv6 address can be used to confuse us. 631 * 632 * Note that packets with unspecified IPv6 destination is 633 * already dropped in ip6_input. 634 */ 635 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 636 /* XXX stat */ 637 goto drop; 638 } 639 } 640 #endif 641 #if defined(INET) && defined(INET6) 642 else 643 #endif 644 #ifdef INET 645 { 646 /* 647 * Get IP and TCP header together in first mbuf. 648 * Note: IP leaves IP header in first mbuf. 649 */ 650 if (off0 > sizeof (struct ip)) { 651 ip_stripoptions(m); 652 off0 = sizeof(struct ip); 653 } 654 if (m->m_len < sizeof (struct tcpiphdr)) { 655 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 656 == NULL) { 657 TCPSTAT_INC(tcps_rcvshort); 658 return (IPPROTO_DONE); 659 } 660 } 661 ip = mtod(m, struct ip *); 662 th = (struct tcphdr *)((caddr_t)ip + off0); 663 tlen = ntohs(ip->ip_len) - off0; 664 665 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 666 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 667 th->th_sum = m->m_pkthdr.csum_data; 668 else 669 th->th_sum = in_pseudo(ip->ip_src.s_addr, 670 ip->ip_dst.s_addr, 671 htonl(m->m_pkthdr.csum_data + tlen + 672 IPPROTO_TCP)); 673 th->th_sum ^= 0xffff; 674 } else { 675 struct ipovly *ipov = (struct ipovly *)ip; 676 677 /* 678 * Checksum extended TCP header and data. 679 */ 680 len = off0 + tlen; 681 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 682 ipov->ih_len = htons(tlen); 683 th->th_sum = in_cksum(m, len); 684 /* Reset length for SDT probes. */ 685 ip->ip_len = htons(tlen + off0); 686 } 687 688 if (th->th_sum) { 689 TCPSTAT_INC(tcps_rcvbadsum); 690 goto drop; 691 } 692 /* Re-initialization for later version check */ 693 ip->ip_v = IPVERSION; 694 } 695 #endif /* INET */ 696 697 #ifdef INET6 698 if (isipv6) 699 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 700 #endif 701 #if defined(INET) && defined(INET6) 702 else 703 #endif 704 #ifdef INET 705 iptos = ip->ip_tos; 706 #endif 707 708 /* 709 * Check that TCP offset makes sense, 710 * pull out TCP options and adjust length. XXX 711 */ 712 off = th->th_off << 2; 713 if (off < sizeof (struct tcphdr) || off > tlen) { 714 TCPSTAT_INC(tcps_rcvbadoff); 715 goto drop; 716 } 717 tlen -= off; /* tlen is used instead of ti->ti_len */ 718 if (off > sizeof (struct tcphdr)) { 719 #ifdef INET6 720 if (isipv6) { 721 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 722 ip6 = mtod(m, struct ip6_hdr *); 723 th = (struct tcphdr *)((caddr_t)ip6 + off0); 724 } 725 #endif 726 #if defined(INET) && defined(INET6) 727 else 728 #endif 729 #ifdef INET 730 { 731 if (m->m_len < sizeof(struct ip) + off) { 732 if ((m = m_pullup(m, sizeof (struct ip) + off)) 733 == NULL) { 734 TCPSTAT_INC(tcps_rcvshort); 735 return (IPPROTO_DONE); 736 } 737 ip = mtod(m, struct ip *); 738 th = (struct tcphdr *)((caddr_t)ip + off0); 739 } 740 } 741 #endif 742 optlen = off - sizeof (struct tcphdr); 743 optp = (u_char *)(th + 1); 744 } 745 thflags = th->th_flags; 746 747 /* 748 * Convert TCP protocol specific fields to host format. 749 */ 750 tcp_fields_to_host(th); 751 752 /* 753 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 754 */ 755 drop_hdrlen = off0 + off; 756 757 /* 758 * Locate pcb for segment; if we're likely to add or remove a 759 * connection then first acquire pcbinfo lock. There are three cases 760 * where we might discover later we need a write lock despite the 761 * flags: ACKs moving a connection out of the syncache, ACKs for a 762 * connection in TIMEWAIT and SYNs not targeting a listening socket. 763 */ 764 if ((thflags & (TH_FIN | TH_RST)) != 0) { 765 INP_INFO_WLOCK(&V_tcbinfo); 766 ti_locked = TI_WLOCKED; 767 } else 768 ti_locked = TI_UNLOCKED; 769 770 /* 771 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 772 */ 773 if ( 774 #ifdef INET6 775 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 776 #ifdef INET 777 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 778 #endif 779 #endif 780 #if defined(INET) && !defined(INET6) 781 (m->m_flags & M_IP_NEXTHOP) 782 #endif 783 ) 784 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 785 786 findpcb: 787 #ifdef INVARIANTS 788 if (ti_locked == TI_WLOCKED) { 789 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 790 } else { 791 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 792 } 793 #endif 794 #ifdef INET6 795 if (isipv6 && fwd_tag != NULL) { 796 struct sockaddr_in6 *next_hop6; 797 798 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 799 /* 800 * Transparently forwarded. Pretend to be the destination. 801 * Already got one like this? 802 */ 803 inp = in6_pcblookup_mbuf(&V_tcbinfo, 804 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 805 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 806 if (!inp) { 807 /* 808 * It's new. Try to find the ambushing socket. 809 * Because we've rewritten the destination address, 810 * any hardware-generated hash is ignored. 811 */ 812 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 813 th->th_sport, &next_hop6->sin6_addr, 814 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 815 th->th_dport, INPLOOKUP_WILDCARD | 816 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 817 } 818 } else if (isipv6) { 819 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 820 th->th_sport, &ip6->ip6_dst, th->th_dport, 821 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 822 m->m_pkthdr.rcvif, m); 823 } 824 #endif /* INET6 */ 825 #if defined(INET6) && defined(INET) 826 else 827 #endif 828 #ifdef INET 829 if (fwd_tag != NULL) { 830 struct sockaddr_in *next_hop; 831 832 next_hop = (struct sockaddr_in *)(fwd_tag+1); 833 /* 834 * Transparently forwarded. Pretend to be the destination. 835 * already got one like this? 836 */ 837 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 838 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 839 m->m_pkthdr.rcvif, m); 840 if (!inp) { 841 /* 842 * It's new. Try to find the ambushing socket. 843 * Because we've rewritten the destination address, 844 * any hardware-generated hash is ignored. 845 */ 846 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 847 th->th_sport, next_hop->sin_addr, 848 next_hop->sin_port ? ntohs(next_hop->sin_port) : 849 th->th_dport, INPLOOKUP_WILDCARD | 850 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 851 } 852 } else 853 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 854 th->th_sport, ip->ip_dst, th->th_dport, 855 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 856 m->m_pkthdr.rcvif, m); 857 #endif /* INET */ 858 859 /* 860 * If the INPCB does not exist then all data in the incoming 861 * segment is discarded and an appropriate RST is sent back. 862 * XXX MRT Send RST using which routing table? 863 */ 864 if (inp == NULL) { 865 /* 866 * Log communication attempts to ports that are not 867 * in use. 868 */ 869 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 870 tcp_log_in_vain == 2) { 871 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 872 log(LOG_INFO, "%s; %s: Connection attempt " 873 "to closed port\n", s, __func__); 874 } 875 /* 876 * When blackholing do not respond with a RST but 877 * completely ignore the segment and drop it. 878 */ 879 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 880 V_blackhole == 2) 881 goto dropunlock; 882 883 rstreason = BANDLIM_RST_CLOSEDPORT; 884 goto dropwithreset; 885 } 886 INP_WLOCK_ASSERT(inp); 887 if (!(inp->inp_flags & INP_HW_FLOWID) 888 && (m->m_flags & M_FLOWID) 889 && ((inp->inp_socket == NULL) 890 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 891 inp->inp_flags |= INP_HW_FLOWID; 892 inp->inp_flags &= ~INP_SW_FLOWID; 893 inp->inp_flowid = m->m_pkthdr.flowid; 894 inp->inp_flowtype = M_HASHTYPE_GET(m); 895 } 896 #ifdef IPSEC 897 #ifdef INET6 898 if (isipv6 && ipsec6_in_reject(m, inp)) { 899 IPSEC6STAT_INC(ips_in_polvio); 900 goto dropunlock; 901 } else 902 #endif /* INET6 */ 903 if (ipsec4_in_reject(m, inp) != 0) { 904 IPSECSTAT_INC(ips_in_polvio); 905 goto dropunlock; 906 } 907 #endif /* IPSEC */ 908 909 /* 910 * Check the minimum TTL for socket. 911 */ 912 if (inp->inp_ip_minttl != 0) { 913 #ifdef INET6 914 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 915 goto dropunlock; 916 else 917 #endif 918 if (inp->inp_ip_minttl > ip->ip_ttl) 919 goto dropunlock; 920 } 921 922 /* 923 * A previous connection in TIMEWAIT state is supposed to catch stray 924 * or duplicate segments arriving late. If this segment was a 925 * legitimate new connection attempt, the old INPCB gets removed and 926 * we can try again to find a listening socket. 927 * 928 * At this point, due to earlier optimism, we may hold only an inpcb 929 * lock, and not the inpcbinfo write lock. If so, we need to try to 930 * acquire it, or if that fails, acquire a reference on the inpcb, 931 * drop all locks, acquire a global write lock, and then re-acquire 932 * the inpcb lock. We may at that point discover that another thread 933 * has tried to free the inpcb, in which case we need to loop back 934 * and try to find a new inpcb to deliver to. 935 * 936 * XXXRW: It may be time to rethink timewait locking. 937 */ 938 relocked: 939 if (inp->inp_flags & INP_TIMEWAIT) { 940 if (ti_locked == TI_UNLOCKED) { 941 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 942 in_pcbref(inp); 943 INP_WUNLOCK(inp); 944 INP_INFO_WLOCK(&V_tcbinfo); 945 ti_locked = TI_WLOCKED; 946 INP_WLOCK(inp); 947 if (in_pcbrele_wlocked(inp)) { 948 inp = NULL; 949 goto findpcb; 950 } 951 } else 952 ti_locked = TI_WLOCKED; 953 } 954 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 955 956 if (thflags & TH_SYN) 957 tcp_dooptions(&to, optp, optlen, TO_SYN); 958 /* 959 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 960 */ 961 if (tcp_twcheck(inp, &to, th, m, tlen)) 962 goto findpcb; 963 INP_INFO_WUNLOCK(&V_tcbinfo); 964 return (IPPROTO_DONE); 965 } 966 /* 967 * The TCPCB may no longer exist if the connection is winding 968 * down or it is in the CLOSED state. Either way we drop the 969 * segment and send an appropriate response. 970 */ 971 tp = intotcpcb(inp); 972 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 973 rstreason = BANDLIM_RST_CLOSEDPORT; 974 goto dropwithreset; 975 } 976 977 #ifdef TCP_OFFLOAD 978 if (tp->t_flags & TF_TOE) { 979 tcp_offload_input(tp, m); 980 m = NULL; /* consumed by the TOE driver */ 981 goto dropunlock; 982 } 983 #endif 984 985 /* 986 * We've identified a valid inpcb, but it could be that we need an 987 * inpcbinfo write lock but don't hold it. In this case, attempt to 988 * acquire using the same strategy as the TIMEWAIT case above. If we 989 * relock, we have to jump back to 'relocked' as the connection might 990 * now be in TIMEWAIT. 991 */ 992 #ifdef INVARIANTS 993 if ((thflags & (TH_FIN | TH_RST)) != 0) 994 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 995 #endif 996 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 997 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN)))) { 998 if (ti_locked == TI_UNLOCKED) { 999 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 1000 in_pcbref(inp); 1001 INP_WUNLOCK(inp); 1002 INP_INFO_WLOCK(&V_tcbinfo); 1003 ti_locked = TI_WLOCKED; 1004 INP_WLOCK(inp); 1005 if (in_pcbrele_wlocked(inp)) { 1006 inp = NULL; 1007 goto findpcb; 1008 } 1009 goto relocked; 1010 } else 1011 ti_locked = TI_WLOCKED; 1012 } 1013 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1014 } 1015 1016 #ifdef MAC 1017 INP_WLOCK_ASSERT(inp); 1018 if (mac_inpcb_check_deliver(inp, m)) 1019 goto dropunlock; 1020 #endif 1021 so = inp->inp_socket; 1022 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1023 #ifdef TCPDEBUG 1024 if (so->so_options & SO_DEBUG) { 1025 ostate = tp->t_state; 1026 #ifdef INET6 1027 if (isipv6) { 1028 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1029 } else 1030 #endif 1031 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1032 tcp_savetcp = *th; 1033 } 1034 #endif /* TCPDEBUG */ 1035 /* 1036 * When the socket is accepting connections (the INPCB is in LISTEN 1037 * state) we look into the SYN cache if this is a new connection 1038 * attempt or the completion of a previous one. 1039 */ 1040 if (so->so_options & SO_ACCEPTCONN) { 1041 struct in_conninfo inc; 1042 1043 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1044 "tp not listening", __func__)); 1045 bzero(&inc, sizeof(inc)); 1046 #ifdef INET6 1047 if (isipv6) { 1048 inc.inc_flags |= INC_ISIPV6; 1049 inc.inc6_faddr = ip6->ip6_src; 1050 inc.inc6_laddr = ip6->ip6_dst; 1051 } else 1052 #endif 1053 { 1054 inc.inc_faddr = ip->ip_src; 1055 inc.inc_laddr = ip->ip_dst; 1056 } 1057 inc.inc_fport = th->th_sport; 1058 inc.inc_lport = th->th_dport; 1059 inc.inc_fibnum = so->so_fibnum; 1060 1061 /* 1062 * Check for an existing connection attempt in syncache if 1063 * the flag is only ACK. A successful lookup creates a new 1064 * socket appended to the listen queue in SYN_RECEIVED state. 1065 */ 1066 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1067 1068 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1069 /* 1070 * Parse the TCP options here because 1071 * syncookies need access to the reflected 1072 * timestamp. 1073 */ 1074 tcp_dooptions(&to, optp, optlen, 0); 1075 /* 1076 * NB: syncache_expand() doesn't unlock 1077 * inp and tcpinfo locks. 1078 */ 1079 if (!syncache_expand(&inc, &to, th, &so, m)) { 1080 /* 1081 * No syncache entry or ACK was not 1082 * for our SYN/ACK. Send a RST. 1083 * NB: syncache did its own logging 1084 * of the failure cause. 1085 */ 1086 rstreason = BANDLIM_RST_OPENPORT; 1087 goto dropwithreset; 1088 } 1089 if (so == NULL) { 1090 /* 1091 * We completed the 3-way handshake 1092 * but could not allocate a socket 1093 * either due to memory shortage, 1094 * listen queue length limits or 1095 * global socket limits. Send RST 1096 * or wait and have the remote end 1097 * retransmit the ACK for another 1098 * try. 1099 */ 1100 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1101 log(LOG_DEBUG, "%s; %s: Listen socket: " 1102 "Socket allocation failed due to " 1103 "limits or memory shortage, %s\n", 1104 s, __func__, 1105 V_tcp_sc_rst_sock_fail ? 1106 "sending RST" : "try again"); 1107 if (V_tcp_sc_rst_sock_fail) { 1108 rstreason = BANDLIM_UNLIMITED; 1109 goto dropwithreset; 1110 } else 1111 goto dropunlock; 1112 } 1113 /* 1114 * Socket is created in state SYN_RECEIVED. 1115 * Unlock the listen socket, lock the newly 1116 * created socket and update the tp variable. 1117 */ 1118 INP_WUNLOCK(inp); /* listen socket */ 1119 inp = sotoinpcb(so); 1120 INP_WLOCK(inp); /* new connection */ 1121 tp = intotcpcb(inp); 1122 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1123 ("%s: ", __func__)); 1124 #ifdef TCP_SIGNATURE 1125 if (sig_checked == 0) { 1126 tcp_dooptions(&to, optp, optlen, 1127 (thflags & TH_SYN) ? TO_SYN : 0); 1128 if (!tcp_signature_verify_input(m, off0, tlen, 1129 optlen, &to, th, tp->t_flags)) { 1130 1131 /* 1132 * In SYN_SENT state if it receives an 1133 * RST, it is allowed for further 1134 * processing. 1135 */ 1136 if ((thflags & TH_RST) == 0 || 1137 (tp->t_state == TCPS_SYN_SENT) == 0) 1138 goto dropunlock; 1139 } 1140 sig_checked = 1; 1141 } 1142 #endif 1143 1144 /* 1145 * Process the segment and the data it 1146 * contains. tcp_do_segment() consumes 1147 * the mbuf chain and unlocks the inpcb. 1148 */ 1149 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1150 iptos, ti_locked); 1151 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1152 return (IPPROTO_DONE); 1153 } 1154 /* 1155 * Segment flag validation for new connection attempts: 1156 * 1157 * Our (SYN|ACK) response was rejected. 1158 * Check with syncache and remove entry to prevent 1159 * retransmits. 1160 * 1161 * NB: syncache_chkrst does its own logging of failure 1162 * causes. 1163 */ 1164 if (thflags & TH_RST) { 1165 syncache_chkrst(&inc, th); 1166 goto dropunlock; 1167 } 1168 /* 1169 * We can't do anything without SYN. 1170 */ 1171 if ((thflags & TH_SYN) == 0) { 1172 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1173 log(LOG_DEBUG, "%s; %s: Listen socket: " 1174 "SYN is missing, segment ignored\n", 1175 s, __func__); 1176 TCPSTAT_INC(tcps_badsyn); 1177 goto dropunlock; 1178 } 1179 /* 1180 * (SYN|ACK) is bogus on a listen socket. 1181 */ 1182 if (thflags & TH_ACK) { 1183 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1184 log(LOG_DEBUG, "%s; %s: Listen socket: " 1185 "SYN|ACK invalid, segment rejected\n", 1186 s, __func__); 1187 syncache_badack(&inc); /* XXX: Not needed! */ 1188 TCPSTAT_INC(tcps_badsyn); 1189 rstreason = BANDLIM_RST_OPENPORT; 1190 goto dropwithreset; 1191 } 1192 /* 1193 * If the drop_synfin option is enabled, drop all 1194 * segments with both the SYN and FIN bits set. 1195 * This prevents e.g. nmap from identifying the 1196 * TCP/IP stack. 1197 * XXX: Poor reasoning. nmap has other methods 1198 * and is constantly refining its stack detection 1199 * strategies. 1200 * XXX: This is a violation of the TCP specification 1201 * and was used by RFC1644. 1202 */ 1203 if ((thflags & TH_FIN) && V_drop_synfin) { 1204 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1205 log(LOG_DEBUG, "%s; %s: Listen socket: " 1206 "SYN|FIN segment ignored (based on " 1207 "sysctl setting)\n", s, __func__); 1208 TCPSTAT_INC(tcps_badsyn); 1209 goto dropunlock; 1210 } 1211 /* 1212 * Segment's flags are (SYN) or (SYN|FIN). 1213 * 1214 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1215 * as they do not affect the state of the TCP FSM. 1216 * The data pointed to by TH_URG and th_urp is ignored. 1217 */ 1218 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1219 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1220 KASSERT(thflags & (TH_SYN), 1221 ("%s: Listen socket: TH_SYN not set", __func__)); 1222 #ifdef INET6 1223 /* 1224 * If deprecated address is forbidden, 1225 * we do not accept SYN to deprecated interface 1226 * address to prevent any new inbound connection from 1227 * getting established. 1228 * When we do not accept SYN, we send a TCP RST, 1229 * with deprecated source address (instead of dropping 1230 * it). We compromise it as it is much better for peer 1231 * to send a RST, and RST will be the final packet 1232 * for the exchange. 1233 * 1234 * If we do not forbid deprecated addresses, we accept 1235 * the SYN packet. RFC2462 does not suggest dropping 1236 * SYN in this case. 1237 * If we decipher RFC2462 5.5.4, it says like this: 1238 * 1. use of deprecated addr with existing 1239 * communication is okay - "SHOULD continue to be 1240 * used" 1241 * 2. use of it with new communication: 1242 * (2a) "SHOULD NOT be used if alternate address 1243 * with sufficient scope is available" 1244 * (2b) nothing mentioned otherwise. 1245 * Here we fall into (2b) case as we have no choice in 1246 * our source address selection - we must obey the peer. 1247 * 1248 * The wording in RFC2462 is confusing, and there are 1249 * multiple description text for deprecated address 1250 * handling - worse, they are not exactly the same. 1251 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1252 */ 1253 if (isipv6 && !V_ip6_use_deprecated) { 1254 struct in6_ifaddr *ia6; 1255 1256 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1257 if (ia6 != NULL && 1258 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1259 ifa_free(&ia6->ia_ifa); 1260 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1261 log(LOG_DEBUG, "%s; %s: Listen socket: " 1262 "Connection attempt to deprecated " 1263 "IPv6 address rejected\n", 1264 s, __func__); 1265 rstreason = BANDLIM_RST_OPENPORT; 1266 goto dropwithreset; 1267 } 1268 if (ia6) 1269 ifa_free(&ia6->ia_ifa); 1270 } 1271 #endif /* INET6 */ 1272 /* 1273 * Basic sanity checks on incoming SYN requests: 1274 * Don't respond if the destination is a link layer 1275 * broadcast according to RFC1122 4.2.3.10, p. 104. 1276 * If it is from this socket it must be forged. 1277 * Don't respond if the source or destination is a 1278 * global or subnet broad- or multicast address. 1279 * Note that it is quite possible to receive unicast 1280 * link-layer packets with a broadcast IP address. Use 1281 * in_broadcast() to find them. 1282 */ 1283 if (m->m_flags & (M_BCAST|M_MCAST)) { 1284 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1285 log(LOG_DEBUG, "%s; %s: Listen socket: " 1286 "Connection attempt from broad- or multicast " 1287 "link layer address ignored\n", s, __func__); 1288 goto dropunlock; 1289 } 1290 #ifdef INET6 1291 if (isipv6) { 1292 if (th->th_dport == th->th_sport && 1293 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1294 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1295 log(LOG_DEBUG, "%s; %s: Listen socket: " 1296 "Connection attempt to/from self " 1297 "ignored\n", s, __func__); 1298 goto dropunlock; 1299 } 1300 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1301 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1302 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1303 log(LOG_DEBUG, "%s; %s: Listen socket: " 1304 "Connection attempt from/to multicast " 1305 "address ignored\n", s, __func__); 1306 goto dropunlock; 1307 } 1308 } 1309 #endif 1310 #if defined(INET) && defined(INET6) 1311 else 1312 #endif 1313 #ifdef INET 1314 { 1315 if (th->th_dport == th->th_sport && 1316 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1318 log(LOG_DEBUG, "%s; %s: Listen socket: " 1319 "Connection attempt from/to self " 1320 "ignored\n", s, __func__); 1321 goto dropunlock; 1322 } 1323 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1324 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1325 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1326 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1327 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1328 log(LOG_DEBUG, "%s; %s: Listen socket: " 1329 "Connection attempt from/to broad- " 1330 "or multicast address ignored\n", 1331 s, __func__); 1332 goto dropunlock; 1333 } 1334 } 1335 #endif 1336 /* 1337 * SYN appears to be valid. Create compressed TCP state 1338 * for syncache. 1339 */ 1340 #ifdef TCPDEBUG 1341 if (so->so_options & SO_DEBUG) 1342 tcp_trace(TA_INPUT, ostate, tp, 1343 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1344 #endif 1345 tcp_dooptions(&to, optp, optlen, TO_SYN); 1346 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1347 /* 1348 * Entry added to syncache and mbuf consumed. 1349 * Only the listen socket is unlocked by syncache_add(). 1350 */ 1351 if (ti_locked == TI_WLOCKED) { 1352 INP_INFO_WUNLOCK(&V_tcbinfo); 1353 ti_locked = TI_UNLOCKED; 1354 } 1355 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1356 return (IPPROTO_DONE); 1357 } else if (tp->t_state == TCPS_LISTEN) { 1358 /* 1359 * When a listen socket is torn down the SO_ACCEPTCONN 1360 * flag is removed first while connections are drained 1361 * from the accept queue in a unlock/lock cycle of the 1362 * ACCEPT_LOCK, opening a race condition allowing a SYN 1363 * attempt go through unhandled. 1364 */ 1365 goto dropunlock; 1366 } 1367 1368 #ifdef TCP_SIGNATURE 1369 if (sig_checked == 0) { 1370 tcp_dooptions(&to, optp, optlen, 1371 (thflags & TH_SYN) ? TO_SYN : 0); 1372 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1373 th, tp->t_flags)) { 1374 1375 /* 1376 * In SYN_SENT state if it receives an RST, it is 1377 * allowed for further processing. 1378 */ 1379 if ((thflags & TH_RST) == 0 || 1380 (tp->t_state == TCPS_SYN_SENT) == 0) 1381 goto dropunlock; 1382 } 1383 sig_checked = 1; 1384 } 1385 #endif 1386 1387 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1388 1389 /* 1390 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1391 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1392 * the inpcb, and unlocks pcbinfo. 1393 */ 1394 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1395 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1396 return (IPPROTO_DONE); 1397 1398 dropwithreset: 1399 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1400 1401 if (ti_locked == TI_WLOCKED) { 1402 INP_INFO_WUNLOCK(&V_tcbinfo); 1403 ti_locked = TI_UNLOCKED; 1404 } 1405 #ifdef INVARIANTS 1406 else { 1407 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1408 "ti_locked: %d", __func__, ti_locked)); 1409 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1410 } 1411 #endif 1412 1413 if (inp != NULL) { 1414 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1415 INP_WUNLOCK(inp); 1416 } else 1417 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1418 m = NULL; /* mbuf chain got consumed. */ 1419 goto drop; 1420 1421 dropunlock: 1422 if (m != NULL) 1423 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1424 1425 if (ti_locked == TI_WLOCKED) { 1426 INP_INFO_WUNLOCK(&V_tcbinfo); 1427 ti_locked = TI_UNLOCKED; 1428 } 1429 #ifdef INVARIANTS 1430 else { 1431 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1432 "ti_locked: %d", __func__, ti_locked)); 1433 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1434 } 1435 #endif 1436 1437 if (inp != NULL) 1438 INP_WUNLOCK(inp); 1439 1440 drop: 1441 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1442 if (s != NULL) 1443 free(s, M_TCPLOG); 1444 if (m != NULL) 1445 m_freem(m); 1446 return (IPPROTO_DONE); 1447 } 1448 1449 static void 1450 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1451 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1452 int ti_locked) 1453 { 1454 int thflags, acked, ourfinisacked, needoutput = 0; 1455 int rstreason, todrop, win; 1456 u_long tiwin; 1457 char *s; 1458 struct in_conninfo *inc; 1459 struct mbuf *mfree; 1460 struct tcpopt to; 1461 1462 #ifdef TCPDEBUG 1463 /* 1464 * The size of tcp_saveipgen must be the size of the max ip header, 1465 * now IPv6. 1466 */ 1467 u_char tcp_saveipgen[IP6_HDR_LEN]; 1468 struct tcphdr tcp_savetcp; 1469 short ostate = 0; 1470 #endif 1471 thflags = th->th_flags; 1472 inc = &tp->t_inpcb->inp_inc; 1473 tp->sackhint.last_sack_ack = 0; 1474 1475 /* 1476 * If this is either a state-changing packet or current state isn't 1477 * established, we require a write lock on tcbinfo. Otherwise, we 1478 * allow the tcbinfo to be in either alocked or unlocked, as the 1479 * caller may have unnecessarily acquired a write lock due to a race. 1480 */ 1481 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1482 tp->t_state != TCPS_ESTABLISHED) { 1483 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1484 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1485 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1486 } else { 1487 #ifdef INVARIANTS 1488 if (ti_locked == TI_WLOCKED) 1489 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1490 else { 1491 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1492 "ti_locked: %d", __func__, ti_locked)); 1493 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1494 } 1495 #endif 1496 } 1497 INP_WLOCK_ASSERT(tp->t_inpcb); 1498 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1499 __func__)); 1500 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1501 __func__)); 1502 1503 /* 1504 * Segment received on connection. 1505 * Reset idle time and keep-alive timer. 1506 * XXX: This should be done after segment 1507 * validation to ignore broken/spoofed segs. 1508 */ 1509 tp->t_rcvtime = ticks; 1510 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1511 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1512 1513 /* 1514 * Unscale the window into a 32-bit value. 1515 * For the SYN_SENT state the scale is zero. 1516 */ 1517 tiwin = th->th_win << tp->snd_scale; 1518 1519 /* 1520 * TCP ECN processing. 1521 */ 1522 if (tp->t_flags & TF_ECN_PERMIT) { 1523 if (thflags & TH_CWR) 1524 tp->t_flags &= ~TF_ECN_SND_ECE; 1525 switch (iptos & IPTOS_ECN_MASK) { 1526 case IPTOS_ECN_CE: 1527 tp->t_flags |= TF_ECN_SND_ECE; 1528 TCPSTAT_INC(tcps_ecn_ce); 1529 break; 1530 case IPTOS_ECN_ECT0: 1531 TCPSTAT_INC(tcps_ecn_ect0); 1532 break; 1533 case IPTOS_ECN_ECT1: 1534 TCPSTAT_INC(tcps_ecn_ect1); 1535 break; 1536 } 1537 /* Congestion experienced. */ 1538 if (thflags & TH_ECE) { 1539 cc_cong_signal(tp, th, CC_ECN); 1540 } 1541 } 1542 1543 /* 1544 * Parse options on any incoming segment. 1545 */ 1546 tcp_dooptions(&to, (u_char *)(th + 1), 1547 (th->th_off << 2) - sizeof(struct tcphdr), 1548 (thflags & TH_SYN) ? TO_SYN : 0); 1549 1550 /* 1551 * If echoed timestamp is later than the current time, 1552 * fall back to non RFC1323 RTT calculation. Normalize 1553 * timestamp if syncookies were used when this connection 1554 * was established. 1555 */ 1556 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1557 to.to_tsecr -= tp->ts_offset; 1558 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1559 to.to_tsecr = 0; 1560 } 1561 /* 1562 * If timestamps were negotiated during SYN/ACK they should 1563 * appear on every segment during this session and vice versa. 1564 */ 1565 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1566 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1567 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1568 "no action\n", s, __func__); 1569 free(s, M_TCPLOG); 1570 } 1571 } 1572 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1573 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1574 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1575 "no action\n", s, __func__); 1576 free(s, M_TCPLOG); 1577 } 1578 } 1579 1580 /* 1581 * Process options only when we get SYN/ACK back. The SYN case 1582 * for incoming connections is handled in tcp_syncache. 1583 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1584 * or <SYN,ACK>) segment itself is never scaled. 1585 * XXX this is traditional behavior, may need to be cleaned up. 1586 */ 1587 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1588 if ((to.to_flags & TOF_SCALE) && 1589 (tp->t_flags & TF_REQ_SCALE)) { 1590 tp->t_flags |= TF_RCVD_SCALE; 1591 tp->snd_scale = to.to_wscale; 1592 } 1593 /* 1594 * Initial send window. It will be updated with 1595 * the next incoming segment to the scaled value. 1596 */ 1597 tp->snd_wnd = th->th_win; 1598 if (to.to_flags & TOF_TS) { 1599 tp->t_flags |= TF_RCVD_TSTMP; 1600 tp->ts_recent = to.to_tsval; 1601 tp->ts_recent_age = tcp_ts_getticks(); 1602 } 1603 if (to.to_flags & TOF_MSS) 1604 tcp_mss(tp, to.to_mss); 1605 if ((tp->t_flags & TF_SACK_PERMIT) && 1606 (to.to_flags & TOF_SACKPERM) == 0) 1607 tp->t_flags &= ~TF_SACK_PERMIT; 1608 } 1609 1610 /* 1611 * Header prediction: check for the two common cases 1612 * of a uni-directional data xfer. If the packet has 1613 * no control flags, is in-sequence, the window didn't 1614 * change and we're not retransmitting, it's a 1615 * candidate. If the length is zero and the ack moved 1616 * forward, we're the sender side of the xfer. Just 1617 * free the data acked & wake any higher level process 1618 * that was blocked waiting for space. If the length 1619 * is non-zero and the ack didn't move, we're the 1620 * receiver side. If we're getting packets in-order 1621 * (the reassembly queue is empty), add the data to 1622 * the socket buffer and note that we need a delayed ack. 1623 * Make sure that the hidden state-flags are also off. 1624 * Since we check for TCPS_ESTABLISHED first, it can only 1625 * be TH_NEEDSYN. 1626 */ 1627 if (tp->t_state == TCPS_ESTABLISHED && 1628 th->th_seq == tp->rcv_nxt && 1629 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1630 tp->snd_nxt == tp->snd_max && 1631 tiwin && tiwin == tp->snd_wnd && 1632 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1633 tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 || 1634 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1635 1636 /* 1637 * If last ACK falls within this segment's sequence numbers, 1638 * record the timestamp. 1639 * NOTE that the test is modified according to the latest 1640 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1641 */ 1642 if ((to.to_flags & TOF_TS) != 0 && 1643 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1644 tp->ts_recent_age = tcp_ts_getticks(); 1645 tp->ts_recent = to.to_tsval; 1646 } 1647 1648 if (tlen == 0) { 1649 if (SEQ_GT(th->th_ack, tp->snd_una) && 1650 SEQ_LEQ(th->th_ack, tp->snd_max) && 1651 !IN_RECOVERY(tp->t_flags) && 1652 (to.to_flags & TOF_SACK) == 0 && 1653 TAILQ_EMPTY(&tp->snd_holes)) { 1654 /* 1655 * This is a pure ack for outstanding data. 1656 */ 1657 if (ti_locked == TI_WLOCKED) 1658 INP_INFO_WUNLOCK(&V_tcbinfo); 1659 ti_locked = TI_UNLOCKED; 1660 1661 TCPSTAT_INC(tcps_predack); 1662 1663 /* 1664 * "bad retransmit" recovery. 1665 */ 1666 if (tp->t_rxtshift == 1 && 1667 tp->t_flags & TF_PREVVALID && 1668 (int)(ticks - tp->t_badrxtwin) < 0) { 1669 cc_cong_signal(tp, th, CC_RTO_ERR); 1670 } 1671 1672 /* 1673 * Recalculate the transmit timer / rtt. 1674 * 1675 * Some boxes send broken timestamp replies 1676 * during the SYN+ACK phase, ignore 1677 * timestamps of 0 or we could calculate a 1678 * huge RTT and blow up the retransmit timer. 1679 */ 1680 if ((to.to_flags & TOF_TS) != 0 && 1681 to.to_tsecr) { 1682 u_int t; 1683 1684 t = tcp_ts_getticks() - to.to_tsecr; 1685 if (!tp->t_rttlow || tp->t_rttlow > t) 1686 tp->t_rttlow = t; 1687 tcp_xmit_timer(tp, 1688 TCP_TS_TO_TICKS(t) + 1); 1689 } else if (tp->t_rtttime && 1690 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1691 if (!tp->t_rttlow || 1692 tp->t_rttlow > ticks - tp->t_rtttime) 1693 tp->t_rttlow = ticks - tp->t_rtttime; 1694 tcp_xmit_timer(tp, 1695 ticks - tp->t_rtttime); 1696 } 1697 acked = BYTES_THIS_ACK(tp, th); 1698 1699 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1700 hhook_run_tcp_est_in(tp, th, &to); 1701 1702 TCPSTAT_INC(tcps_rcvackpack); 1703 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1704 sbdrop(&so->so_snd, acked); 1705 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1706 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1707 tp->snd_recover = th->th_ack - 1; 1708 1709 /* 1710 * Let the congestion control algorithm update 1711 * congestion control related information. This 1712 * typically means increasing the congestion 1713 * window. 1714 */ 1715 cc_ack_received(tp, th, CC_ACK); 1716 1717 tp->snd_una = th->th_ack; 1718 /* 1719 * Pull snd_wl2 up to prevent seq wrap relative 1720 * to th_ack. 1721 */ 1722 tp->snd_wl2 = th->th_ack; 1723 tp->t_dupacks = 0; 1724 m_freem(m); 1725 ND6_HINT(tp); /* Some progress has been made. */ 1726 1727 /* 1728 * If all outstanding data are acked, stop 1729 * retransmit timer, otherwise restart timer 1730 * using current (possibly backed-off) value. 1731 * If process is waiting for space, 1732 * wakeup/selwakeup/signal. If data 1733 * are ready to send, let tcp_output 1734 * decide between more output or persist. 1735 */ 1736 #ifdef TCPDEBUG 1737 if (so->so_options & SO_DEBUG) 1738 tcp_trace(TA_INPUT, ostate, tp, 1739 (void *)tcp_saveipgen, 1740 &tcp_savetcp, 0); 1741 #endif 1742 if (tp->snd_una == tp->snd_max) 1743 tcp_timer_activate(tp, TT_REXMT, 0); 1744 else if (!tcp_timer_active(tp, TT_PERSIST)) 1745 tcp_timer_activate(tp, TT_REXMT, 1746 tp->t_rxtcur); 1747 sowwakeup(so); 1748 if (sbavail(&so->so_snd)) 1749 (void) tcp_output(tp); 1750 goto check_delack; 1751 } 1752 } else if (th->th_ack == tp->snd_una && 1753 tlen <= sbspace(&so->so_rcv)) { 1754 int newsize = 0; /* automatic sockbuf scaling */ 1755 1756 /* 1757 * This is a pure, in-sequence data packet with 1758 * nothing on the reassembly queue and we have enough 1759 * buffer space to take it. 1760 */ 1761 if (ti_locked == TI_WLOCKED) 1762 INP_INFO_WUNLOCK(&V_tcbinfo); 1763 ti_locked = TI_UNLOCKED; 1764 1765 /* Clean receiver SACK report if present */ 1766 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1767 tcp_clean_sackreport(tp); 1768 TCPSTAT_INC(tcps_preddat); 1769 tp->rcv_nxt += tlen; 1770 /* 1771 * Pull snd_wl1 up to prevent seq wrap relative to 1772 * th_seq. 1773 */ 1774 tp->snd_wl1 = th->th_seq; 1775 /* 1776 * Pull rcv_up up to prevent seq wrap relative to 1777 * rcv_nxt. 1778 */ 1779 tp->rcv_up = tp->rcv_nxt; 1780 TCPSTAT_INC(tcps_rcvpack); 1781 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1782 ND6_HINT(tp); /* Some progress has been made */ 1783 #ifdef TCPDEBUG 1784 if (so->so_options & SO_DEBUG) 1785 tcp_trace(TA_INPUT, ostate, tp, 1786 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1787 #endif 1788 /* 1789 * Automatic sizing of receive socket buffer. Often the send 1790 * buffer size is not optimally adjusted to the actual network 1791 * conditions at hand (delay bandwidth product). Setting the 1792 * buffer size too small limits throughput on links with high 1793 * bandwidth and high delay (eg. trans-continental/oceanic links). 1794 * 1795 * On the receive side the socket buffer memory is only rarely 1796 * used to any significant extent. This allows us to be much 1797 * more aggressive in scaling the receive socket buffer. For 1798 * the case that the buffer space is actually used to a large 1799 * extent and we run out of kernel memory we can simply drop 1800 * the new segments; TCP on the sender will just retransmit it 1801 * later. Setting the buffer size too big may only consume too 1802 * much kernel memory if the application doesn't read() from 1803 * the socket or packet loss or reordering makes use of the 1804 * reassembly queue. 1805 * 1806 * The criteria to step up the receive buffer one notch are: 1807 * 1. Application has not set receive buffer size with 1808 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1809 * 2. the number of bytes received during the time it takes 1810 * one timestamp to be reflected back to us (the RTT); 1811 * 3. received bytes per RTT is within seven eighth of the 1812 * current socket buffer size; 1813 * 4. receive buffer size has not hit maximal automatic size; 1814 * 1815 * This algorithm does one step per RTT at most and only if 1816 * we receive a bulk stream w/o packet losses or reorderings. 1817 * Shrinking the buffer during idle times is not necessary as 1818 * it doesn't consume any memory when idle. 1819 * 1820 * TODO: Only step up if the application is actually serving 1821 * the buffer to better manage the socket buffer resources. 1822 */ 1823 if (V_tcp_do_autorcvbuf && 1824 to.to_tsecr && 1825 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1826 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1827 to.to_tsecr - tp->rfbuf_ts < hz) { 1828 if (tp->rfbuf_cnt > 1829 (so->so_rcv.sb_hiwat / 8 * 7) && 1830 so->so_rcv.sb_hiwat < 1831 V_tcp_autorcvbuf_max) { 1832 newsize = 1833 min(so->so_rcv.sb_hiwat + 1834 V_tcp_autorcvbuf_inc, 1835 V_tcp_autorcvbuf_max); 1836 } 1837 /* Start over with next RTT. */ 1838 tp->rfbuf_ts = 0; 1839 tp->rfbuf_cnt = 0; 1840 } else 1841 tp->rfbuf_cnt += tlen; /* add up */ 1842 } 1843 1844 /* Add data to socket buffer. */ 1845 SOCKBUF_LOCK(&so->so_rcv); 1846 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1847 m_freem(m); 1848 } else { 1849 /* 1850 * Set new socket buffer size. 1851 * Give up when limit is reached. 1852 */ 1853 if (newsize) 1854 if (!sbreserve_locked(&so->so_rcv, 1855 newsize, so, NULL)) 1856 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1857 m_adj(m, drop_hdrlen); /* delayed header drop */ 1858 sbappendstream_locked(&so->so_rcv, m); 1859 } 1860 /* NB: sorwakeup_locked() does an implicit unlock. */ 1861 sorwakeup_locked(so); 1862 if (DELAY_ACK(tp, tlen)) { 1863 tp->t_flags |= TF_DELACK; 1864 } else { 1865 tp->t_flags |= TF_ACKNOW; 1866 tcp_output(tp); 1867 } 1868 goto check_delack; 1869 } 1870 } 1871 1872 /* 1873 * Calculate amount of space in receive window, 1874 * and then do TCP input processing. 1875 * Receive window is amount of space in rcv queue, 1876 * but not less than advertised window. 1877 */ 1878 win = sbspace(&so->so_rcv); 1879 if (win < 0) 1880 win = 0; 1881 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1882 1883 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1884 tp->rfbuf_ts = 0; 1885 tp->rfbuf_cnt = 0; 1886 1887 switch (tp->t_state) { 1888 1889 /* 1890 * If the state is SYN_RECEIVED: 1891 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1892 */ 1893 case TCPS_SYN_RECEIVED: 1894 if ((thflags & TH_ACK) && 1895 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1896 SEQ_GT(th->th_ack, tp->snd_max))) { 1897 rstreason = BANDLIM_RST_OPENPORT; 1898 goto dropwithreset; 1899 } 1900 break; 1901 1902 /* 1903 * If the state is SYN_SENT: 1904 * if seg contains an ACK, but not for our SYN, drop the input. 1905 * if seg contains a RST, then drop the connection. 1906 * if seg does not contain SYN, then drop it. 1907 * Otherwise this is an acceptable SYN segment 1908 * initialize tp->rcv_nxt and tp->irs 1909 * if seg contains ack then advance tp->snd_una 1910 * if seg contains an ECE and ECN support is enabled, the stream 1911 * is ECN capable. 1912 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1913 * arrange for segment to be acked (eventually) 1914 * continue processing rest of data/controls, beginning with URG 1915 */ 1916 case TCPS_SYN_SENT: 1917 if ((thflags & TH_ACK) && 1918 (SEQ_LEQ(th->th_ack, tp->iss) || 1919 SEQ_GT(th->th_ack, tp->snd_max))) { 1920 rstreason = BANDLIM_UNLIMITED; 1921 goto dropwithreset; 1922 } 1923 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1924 TCP_PROBE5(connect__refused, NULL, tp, 1925 mtod(m, const char *), tp, th); 1926 tp = tcp_drop(tp, ECONNREFUSED); 1927 } 1928 if (thflags & TH_RST) 1929 goto drop; 1930 if (!(thflags & TH_SYN)) 1931 goto drop; 1932 1933 tp->irs = th->th_seq; 1934 tcp_rcvseqinit(tp); 1935 if (thflags & TH_ACK) { 1936 TCPSTAT_INC(tcps_connects); 1937 soisconnected(so); 1938 #ifdef MAC 1939 mac_socketpeer_set_from_mbuf(m, so); 1940 #endif 1941 /* Do window scaling on this connection? */ 1942 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1943 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1944 tp->rcv_scale = tp->request_r_scale; 1945 } 1946 tp->rcv_adv += imin(tp->rcv_wnd, 1947 TCP_MAXWIN << tp->rcv_scale); 1948 tp->snd_una++; /* SYN is acked */ 1949 /* 1950 * If there's data, delay ACK; if there's also a FIN 1951 * ACKNOW will be turned on later. 1952 */ 1953 if (DELAY_ACK(tp, tlen) && tlen != 0) 1954 tcp_timer_activate(tp, TT_DELACK, 1955 tcp_delacktime); 1956 else 1957 tp->t_flags |= TF_ACKNOW; 1958 1959 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1960 tp->t_flags |= TF_ECN_PERMIT; 1961 TCPSTAT_INC(tcps_ecn_shs); 1962 } 1963 1964 /* 1965 * Received <SYN,ACK> in SYN_SENT[*] state. 1966 * Transitions: 1967 * SYN_SENT --> ESTABLISHED 1968 * SYN_SENT* --> FIN_WAIT_1 1969 */ 1970 tp->t_starttime = ticks; 1971 if (tp->t_flags & TF_NEEDFIN) { 1972 tcp_state_change(tp, TCPS_FIN_WAIT_1); 1973 tp->t_flags &= ~TF_NEEDFIN; 1974 thflags &= ~TH_SYN; 1975 } else { 1976 tcp_state_change(tp, TCPS_ESTABLISHED); 1977 TCP_PROBE5(connect__established, NULL, tp, 1978 mtod(m, const char *), tp, th); 1979 cc_conn_init(tp); 1980 tcp_timer_activate(tp, TT_KEEP, 1981 TP_KEEPIDLE(tp)); 1982 } 1983 } else { 1984 /* 1985 * Received initial SYN in SYN-SENT[*] state => 1986 * simultaneous open. 1987 * If it succeeds, connection is * half-synchronized. 1988 * Otherwise, do 3-way handshake: 1989 * SYN-SENT -> SYN-RECEIVED 1990 * SYN-SENT* -> SYN-RECEIVED* 1991 */ 1992 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1993 tcp_timer_activate(tp, TT_REXMT, 0); 1994 tcp_state_change(tp, TCPS_SYN_RECEIVED); 1995 } 1996 1997 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1998 "ti_locked %d", __func__, ti_locked)); 1999 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2000 INP_WLOCK_ASSERT(tp->t_inpcb); 2001 2002 /* 2003 * Advance th->th_seq to correspond to first data byte. 2004 * If data, trim to stay within window, 2005 * dropping FIN if necessary. 2006 */ 2007 th->th_seq++; 2008 if (tlen > tp->rcv_wnd) { 2009 todrop = tlen - tp->rcv_wnd; 2010 m_adj(m, -todrop); 2011 tlen = tp->rcv_wnd; 2012 thflags &= ~TH_FIN; 2013 TCPSTAT_INC(tcps_rcvpackafterwin); 2014 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2015 } 2016 tp->snd_wl1 = th->th_seq - 1; 2017 tp->rcv_up = th->th_seq; 2018 /* 2019 * Client side of transaction: already sent SYN and data. 2020 * If the remote host used T/TCP to validate the SYN, 2021 * our data will be ACK'd; if so, enter normal data segment 2022 * processing in the middle of step 5, ack processing. 2023 * Otherwise, goto step 6. 2024 */ 2025 if (thflags & TH_ACK) 2026 goto process_ACK; 2027 2028 goto step6; 2029 2030 /* 2031 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2032 * do normal processing. 2033 * 2034 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2035 */ 2036 case TCPS_LAST_ACK: 2037 case TCPS_CLOSING: 2038 break; /* continue normal processing */ 2039 } 2040 2041 /* 2042 * States other than LISTEN or SYN_SENT. 2043 * First check the RST flag and sequence number since reset segments 2044 * are exempt from the timestamp and connection count tests. This 2045 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2046 * below which allowed reset segments in half the sequence space 2047 * to fall though and be processed (which gives forged reset 2048 * segments with a random sequence number a 50 percent chance of 2049 * killing a connection). 2050 * Then check timestamp, if present. 2051 * Then check the connection count, if present. 2052 * Then check that at least some bytes of segment are within 2053 * receive window. If segment begins before rcv_nxt, 2054 * drop leading data (and SYN); if nothing left, just ack. 2055 */ 2056 if (thflags & TH_RST) { 2057 /* 2058 * RFC5961 Section 3.2 2059 * 2060 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2061 * - If RST is in window, we send challenge ACK. 2062 * 2063 * Note: to take into account delayed ACKs, we should 2064 * test against last_ack_sent instead of rcv_nxt. 2065 * Note 2: we handle special case of closed window, not 2066 * covered by the RFC. 2067 */ 2068 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2069 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2070 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2071 2072 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2073 KASSERT(ti_locked == TI_WLOCKED, 2074 ("%s: TH_RST ti_locked %d, th %p tp %p", 2075 __func__, ti_locked, th, tp)); 2076 KASSERT(tp->t_state != TCPS_SYN_SENT, 2077 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2078 __func__, th, tp)); 2079 2080 if (V_tcp_insecure_rst || 2081 tp->last_ack_sent == th->th_seq) { 2082 TCPSTAT_INC(tcps_drops); 2083 /* Drop the connection. */ 2084 switch (tp->t_state) { 2085 case TCPS_SYN_RECEIVED: 2086 so->so_error = ECONNREFUSED; 2087 goto close; 2088 case TCPS_ESTABLISHED: 2089 case TCPS_FIN_WAIT_1: 2090 case TCPS_FIN_WAIT_2: 2091 case TCPS_CLOSE_WAIT: 2092 so->so_error = ECONNRESET; 2093 close: 2094 tcp_state_change(tp, TCPS_CLOSED); 2095 /* FALLTHROUGH */ 2096 default: 2097 tp = tcp_close(tp); 2098 } 2099 } else { 2100 TCPSTAT_INC(tcps_badrst); 2101 /* Send challenge ACK. */ 2102 tcp_respond(tp, mtod(m, void *), th, m, 2103 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2104 tp->last_ack_sent = tp->rcv_nxt; 2105 m = NULL; 2106 } 2107 } 2108 goto drop; 2109 } 2110 2111 /* 2112 * RFC5961 Section 4.2 2113 * Send challenge ACK for any SYN in synchronized state. 2114 */ 2115 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT) { 2116 KASSERT(ti_locked == TI_WLOCKED, 2117 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2118 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2119 2120 TCPSTAT_INC(tcps_badsyn); 2121 if (V_tcp_insecure_syn && 2122 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2123 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2124 tp = tcp_drop(tp, ECONNRESET); 2125 rstreason = BANDLIM_UNLIMITED; 2126 } else { 2127 /* Send challenge ACK. */ 2128 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2129 tp->snd_nxt, TH_ACK); 2130 tp->last_ack_sent = tp->rcv_nxt; 2131 m = NULL; 2132 } 2133 goto drop; 2134 } 2135 2136 /* 2137 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2138 * and it's less than ts_recent, drop it. 2139 */ 2140 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2141 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2142 2143 /* Check to see if ts_recent is over 24 days old. */ 2144 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2145 /* 2146 * Invalidate ts_recent. If this segment updates 2147 * ts_recent, the age will be reset later and ts_recent 2148 * will get a valid value. If it does not, setting 2149 * ts_recent to zero will at least satisfy the 2150 * requirement that zero be placed in the timestamp 2151 * echo reply when ts_recent isn't valid. The 2152 * age isn't reset until we get a valid ts_recent 2153 * because we don't want out-of-order segments to be 2154 * dropped when ts_recent is old. 2155 */ 2156 tp->ts_recent = 0; 2157 } else { 2158 TCPSTAT_INC(tcps_rcvduppack); 2159 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2160 TCPSTAT_INC(tcps_pawsdrop); 2161 if (tlen) 2162 goto dropafterack; 2163 goto drop; 2164 } 2165 } 2166 2167 /* 2168 * In the SYN-RECEIVED state, validate that the packet belongs to 2169 * this connection before trimming the data to fit the receive 2170 * window. Check the sequence number versus IRS since we know 2171 * the sequence numbers haven't wrapped. This is a partial fix 2172 * for the "LAND" DoS attack. 2173 */ 2174 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2175 rstreason = BANDLIM_RST_OPENPORT; 2176 goto dropwithreset; 2177 } 2178 2179 todrop = tp->rcv_nxt - th->th_seq; 2180 if (todrop > 0) { 2181 if (thflags & TH_SYN) { 2182 thflags &= ~TH_SYN; 2183 th->th_seq++; 2184 if (th->th_urp > 1) 2185 th->th_urp--; 2186 else 2187 thflags &= ~TH_URG; 2188 todrop--; 2189 } 2190 /* 2191 * Following if statement from Stevens, vol. 2, p. 960. 2192 */ 2193 if (todrop > tlen 2194 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2195 /* 2196 * Any valid FIN must be to the left of the window. 2197 * At this point the FIN must be a duplicate or out 2198 * of sequence; drop it. 2199 */ 2200 thflags &= ~TH_FIN; 2201 2202 /* 2203 * Send an ACK to resynchronize and drop any data. 2204 * But keep on processing for RST or ACK. 2205 */ 2206 tp->t_flags |= TF_ACKNOW; 2207 todrop = tlen; 2208 TCPSTAT_INC(tcps_rcvduppack); 2209 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2210 } else { 2211 TCPSTAT_INC(tcps_rcvpartduppack); 2212 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2213 } 2214 drop_hdrlen += todrop; /* drop from the top afterwards */ 2215 th->th_seq += todrop; 2216 tlen -= todrop; 2217 if (th->th_urp > todrop) 2218 th->th_urp -= todrop; 2219 else { 2220 thflags &= ~TH_URG; 2221 th->th_urp = 0; 2222 } 2223 } 2224 2225 /* 2226 * If new data are received on a connection after the 2227 * user processes are gone, then RST the other end. 2228 */ 2229 if ((so->so_state & SS_NOFDREF) && 2230 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2231 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2232 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2233 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2234 2235 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2236 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2237 "after socket was closed, " 2238 "sending RST and removing tcpcb\n", 2239 s, __func__, tcpstates[tp->t_state], tlen); 2240 free(s, M_TCPLOG); 2241 } 2242 tp = tcp_close(tp); 2243 TCPSTAT_INC(tcps_rcvafterclose); 2244 rstreason = BANDLIM_UNLIMITED; 2245 goto dropwithreset; 2246 } 2247 2248 /* 2249 * If segment ends after window, drop trailing data 2250 * (and PUSH and FIN); if nothing left, just ACK. 2251 */ 2252 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2253 if (todrop > 0) { 2254 TCPSTAT_INC(tcps_rcvpackafterwin); 2255 if (todrop >= tlen) { 2256 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2257 /* 2258 * If window is closed can only take segments at 2259 * window edge, and have to drop data and PUSH from 2260 * incoming segments. Continue processing, but 2261 * remember to ack. Otherwise, drop segment 2262 * and ack. 2263 */ 2264 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2265 tp->t_flags |= TF_ACKNOW; 2266 TCPSTAT_INC(tcps_rcvwinprobe); 2267 } else 2268 goto dropafterack; 2269 } else 2270 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2271 m_adj(m, -todrop); 2272 tlen -= todrop; 2273 thflags &= ~(TH_PUSH|TH_FIN); 2274 } 2275 2276 /* 2277 * If last ACK falls within this segment's sequence numbers, 2278 * record its timestamp. 2279 * NOTE: 2280 * 1) That the test incorporates suggestions from the latest 2281 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2282 * 2) That updating only on newer timestamps interferes with 2283 * our earlier PAWS tests, so this check should be solely 2284 * predicated on the sequence space of this segment. 2285 * 3) That we modify the segment boundary check to be 2286 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2287 * instead of RFC1323's 2288 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2289 * This modified check allows us to overcome RFC1323's 2290 * limitations as described in Stevens TCP/IP Illustrated 2291 * Vol. 2 p.869. In such cases, we can still calculate the 2292 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2293 */ 2294 if ((to.to_flags & TOF_TS) != 0 && 2295 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2296 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2297 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2298 tp->ts_recent_age = tcp_ts_getticks(); 2299 tp->ts_recent = to.to_tsval; 2300 } 2301 2302 /* 2303 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2304 * flag is on (half-synchronized state), then queue data for 2305 * later processing; else drop segment and return. 2306 */ 2307 if ((thflags & TH_ACK) == 0) { 2308 if (tp->t_state == TCPS_SYN_RECEIVED || 2309 (tp->t_flags & TF_NEEDSYN)) 2310 goto step6; 2311 else if (tp->t_flags & TF_ACKNOW) 2312 goto dropafterack; 2313 else 2314 goto drop; 2315 } 2316 2317 /* 2318 * Ack processing. 2319 */ 2320 switch (tp->t_state) { 2321 2322 /* 2323 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2324 * ESTABLISHED state and continue processing. 2325 * The ACK was checked above. 2326 */ 2327 case TCPS_SYN_RECEIVED: 2328 2329 TCPSTAT_INC(tcps_connects); 2330 soisconnected(so); 2331 /* Do window scaling? */ 2332 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2333 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2334 tp->rcv_scale = tp->request_r_scale; 2335 tp->snd_wnd = tiwin; 2336 } 2337 /* 2338 * Make transitions: 2339 * SYN-RECEIVED -> ESTABLISHED 2340 * SYN-RECEIVED* -> FIN-WAIT-1 2341 */ 2342 tp->t_starttime = ticks; 2343 if (tp->t_flags & TF_NEEDFIN) { 2344 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2345 tp->t_flags &= ~TF_NEEDFIN; 2346 } else { 2347 tcp_state_change(tp, TCPS_ESTABLISHED); 2348 TCP_PROBE5(accept__established, NULL, tp, 2349 mtod(m, const char *), tp, th); 2350 cc_conn_init(tp); 2351 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2352 } 2353 /* 2354 * If segment contains data or ACK, will call tcp_reass() 2355 * later; if not, do so now to pass queued data to user. 2356 */ 2357 if (tlen == 0 && (thflags & TH_FIN) == 0) 2358 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2359 (struct mbuf *)0); 2360 tp->snd_wl1 = th->th_seq - 1; 2361 /* FALLTHROUGH */ 2362 2363 /* 2364 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2365 * ACKs. If the ack is in the range 2366 * tp->snd_una < th->th_ack <= tp->snd_max 2367 * then advance tp->snd_una to th->th_ack and drop 2368 * data from the retransmission queue. If this ACK reflects 2369 * more up to date window information we update our window information. 2370 */ 2371 case TCPS_ESTABLISHED: 2372 case TCPS_FIN_WAIT_1: 2373 case TCPS_FIN_WAIT_2: 2374 case TCPS_CLOSE_WAIT: 2375 case TCPS_CLOSING: 2376 case TCPS_LAST_ACK: 2377 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2378 TCPSTAT_INC(tcps_rcvacktoomuch); 2379 goto dropafterack; 2380 } 2381 if ((tp->t_flags & TF_SACK_PERMIT) && 2382 ((to.to_flags & TOF_SACK) || 2383 !TAILQ_EMPTY(&tp->snd_holes))) 2384 tcp_sack_doack(tp, &to, th->th_ack); 2385 2386 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2387 hhook_run_tcp_est_in(tp, th, &to); 2388 2389 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2390 if (tlen == 0 && tiwin == tp->snd_wnd) { 2391 /* 2392 * If this is the first time we've seen a 2393 * FIN from the remote, this is not a 2394 * duplicate and it needs to be processed 2395 * normally. This happens during a 2396 * simultaneous close. 2397 */ 2398 if ((thflags & TH_FIN) && 2399 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2400 tp->t_dupacks = 0; 2401 break; 2402 } 2403 TCPSTAT_INC(tcps_rcvdupack); 2404 /* 2405 * If we have outstanding data (other than 2406 * a window probe), this is a completely 2407 * duplicate ack (ie, window info didn't 2408 * change and FIN isn't set), 2409 * the ack is the biggest we've 2410 * seen and we've seen exactly our rexmt 2411 * threshhold of them, assume a packet 2412 * has been dropped and retransmit it. 2413 * Kludge snd_nxt & the congestion 2414 * window so we send only this one 2415 * packet. 2416 * 2417 * We know we're losing at the current 2418 * window size so do congestion avoidance 2419 * (set ssthresh to half the current window 2420 * and pull our congestion window back to 2421 * the new ssthresh). 2422 * 2423 * Dup acks mean that packets have left the 2424 * network (they're now cached at the receiver) 2425 * so bump cwnd by the amount in the receiver 2426 * to keep a constant cwnd packets in the 2427 * network. 2428 * 2429 * When using TCP ECN, notify the peer that 2430 * we reduced the cwnd. 2431 */ 2432 if (!tcp_timer_active(tp, TT_REXMT) || 2433 th->th_ack != tp->snd_una) 2434 tp->t_dupacks = 0; 2435 else if (++tp->t_dupacks > tcprexmtthresh || 2436 IN_FASTRECOVERY(tp->t_flags)) { 2437 cc_ack_received(tp, th, CC_DUPACK); 2438 if ((tp->t_flags & TF_SACK_PERMIT) && 2439 IN_FASTRECOVERY(tp->t_flags)) { 2440 int awnd; 2441 2442 /* 2443 * Compute the amount of data in flight first. 2444 * We can inject new data into the pipe iff 2445 * we have less than 1/2 the original window's 2446 * worth of data in flight. 2447 */ 2448 awnd = (tp->snd_nxt - tp->snd_fack) + 2449 tp->sackhint.sack_bytes_rexmit; 2450 if (awnd < tp->snd_ssthresh) { 2451 tp->snd_cwnd += tp->t_maxseg; 2452 if (tp->snd_cwnd > tp->snd_ssthresh) 2453 tp->snd_cwnd = tp->snd_ssthresh; 2454 } 2455 } else 2456 tp->snd_cwnd += tp->t_maxseg; 2457 (void) tcp_output(tp); 2458 goto drop; 2459 } else if (tp->t_dupacks == tcprexmtthresh) { 2460 tcp_seq onxt = tp->snd_nxt; 2461 2462 /* 2463 * If we're doing sack, check to 2464 * see if we're already in sack 2465 * recovery. If we're not doing sack, 2466 * check to see if we're in newreno 2467 * recovery. 2468 */ 2469 if (tp->t_flags & TF_SACK_PERMIT) { 2470 if (IN_FASTRECOVERY(tp->t_flags)) { 2471 tp->t_dupacks = 0; 2472 break; 2473 } 2474 } else { 2475 if (SEQ_LEQ(th->th_ack, 2476 tp->snd_recover)) { 2477 tp->t_dupacks = 0; 2478 break; 2479 } 2480 } 2481 /* Congestion signal before ack. */ 2482 cc_cong_signal(tp, th, CC_NDUPACK); 2483 cc_ack_received(tp, th, CC_DUPACK); 2484 tcp_timer_activate(tp, TT_REXMT, 0); 2485 tp->t_rtttime = 0; 2486 if (tp->t_flags & TF_SACK_PERMIT) { 2487 TCPSTAT_INC( 2488 tcps_sack_recovery_episode); 2489 tp->sack_newdata = tp->snd_nxt; 2490 tp->snd_cwnd = tp->t_maxseg; 2491 (void) tcp_output(tp); 2492 goto drop; 2493 } 2494 tp->snd_nxt = th->th_ack; 2495 tp->snd_cwnd = tp->t_maxseg; 2496 (void) tcp_output(tp); 2497 KASSERT(tp->snd_limited <= 2, 2498 ("%s: tp->snd_limited too big", 2499 __func__)); 2500 tp->snd_cwnd = tp->snd_ssthresh + 2501 tp->t_maxseg * 2502 (tp->t_dupacks - tp->snd_limited); 2503 if (SEQ_GT(onxt, tp->snd_nxt)) 2504 tp->snd_nxt = onxt; 2505 goto drop; 2506 } else if (V_tcp_do_rfc3042) { 2507 cc_ack_received(tp, th, CC_DUPACK); 2508 u_long oldcwnd = tp->snd_cwnd; 2509 tcp_seq oldsndmax = tp->snd_max; 2510 u_int sent; 2511 int avail; 2512 2513 KASSERT(tp->t_dupacks == 1 || 2514 tp->t_dupacks == 2, 2515 ("%s: dupacks not 1 or 2", 2516 __func__)); 2517 if (tp->t_dupacks == 1) 2518 tp->snd_limited = 0; 2519 tp->snd_cwnd = 2520 (tp->snd_nxt - tp->snd_una) + 2521 (tp->t_dupacks - tp->snd_limited) * 2522 tp->t_maxseg; 2523 /* 2524 * Only call tcp_output when there 2525 * is new data available to be sent. 2526 * Otherwise we would send pure ACKs. 2527 */ 2528 SOCKBUF_LOCK(&so->so_snd); 2529 avail = sbavail(&so->so_snd) - 2530 (tp->snd_nxt - tp->snd_una); 2531 SOCKBUF_UNLOCK(&so->so_snd); 2532 if (avail > 0) 2533 (void) tcp_output(tp); 2534 sent = tp->snd_max - oldsndmax; 2535 if (sent > tp->t_maxseg) { 2536 KASSERT((tp->t_dupacks == 2 && 2537 tp->snd_limited == 0) || 2538 (sent == tp->t_maxseg + 1 && 2539 tp->t_flags & TF_SENTFIN), 2540 ("%s: sent too much", 2541 __func__)); 2542 tp->snd_limited = 2; 2543 } else if (sent > 0) 2544 ++tp->snd_limited; 2545 tp->snd_cwnd = oldcwnd; 2546 goto drop; 2547 } 2548 } else 2549 tp->t_dupacks = 0; 2550 break; 2551 } 2552 2553 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2554 ("%s: th_ack <= snd_una", __func__)); 2555 2556 /* 2557 * If the congestion window was inflated to account 2558 * for the other side's cached packets, retract it. 2559 */ 2560 if (IN_FASTRECOVERY(tp->t_flags)) { 2561 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2562 if (tp->t_flags & TF_SACK_PERMIT) 2563 tcp_sack_partialack(tp, th); 2564 else 2565 tcp_newreno_partial_ack(tp, th); 2566 } else 2567 cc_post_recovery(tp, th); 2568 } 2569 tp->t_dupacks = 0; 2570 /* 2571 * If we reach this point, ACK is not a duplicate, 2572 * i.e., it ACKs something we sent. 2573 */ 2574 if (tp->t_flags & TF_NEEDSYN) { 2575 /* 2576 * T/TCP: Connection was half-synchronized, and our 2577 * SYN has been ACK'd (so connection is now fully 2578 * synchronized). Go to non-starred state, 2579 * increment snd_una for ACK of SYN, and check if 2580 * we can do window scaling. 2581 */ 2582 tp->t_flags &= ~TF_NEEDSYN; 2583 tp->snd_una++; 2584 /* Do window scaling? */ 2585 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2586 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2587 tp->rcv_scale = tp->request_r_scale; 2588 /* Send window already scaled. */ 2589 } 2590 } 2591 2592 process_ACK: 2593 INP_WLOCK_ASSERT(tp->t_inpcb); 2594 2595 acked = BYTES_THIS_ACK(tp, th); 2596 TCPSTAT_INC(tcps_rcvackpack); 2597 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2598 2599 /* 2600 * If we just performed our first retransmit, and the ACK 2601 * arrives within our recovery window, then it was a mistake 2602 * to do the retransmit in the first place. Recover our 2603 * original cwnd and ssthresh, and proceed to transmit where 2604 * we left off. 2605 */ 2606 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2607 (int)(ticks - tp->t_badrxtwin) < 0) 2608 cc_cong_signal(tp, th, CC_RTO_ERR); 2609 2610 /* 2611 * If we have a timestamp reply, update smoothed 2612 * round trip time. If no timestamp is present but 2613 * transmit timer is running and timed sequence 2614 * number was acked, update smoothed round trip time. 2615 * Since we now have an rtt measurement, cancel the 2616 * timer backoff (cf., Phil Karn's retransmit alg.). 2617 * Recompute the initial retransmit timer. 2618 * 2619 * Some boxes send broken timestamp replies 2620 * during the SYN+ACK phase, ignore 2621 * timestamps of 0 or we could calculate a 2622 * huge RTT and blow up the retransmit timer. 2623 */ 2624 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2625 u_int t; 2626 2627 t = tcp_ts_getticks() - to.to_tsecr; 2628 if (!tp->t_rttlow || tp->t_rttlow > t) 2629 tp->t_rttlow = t; 2630 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2631 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2632 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2633 tp->t_rttlow = ticks - tp->t_rtttime; 2634 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2635 } 2636 2637 /* 2638 * If all outstanding data is acked, stop retransmit 2639 * timer and remember to restart (more output or persist). 2640 * If there is more data to be acked, restart retransmit 2641 * timer, using current (possibly backed-off) value. 2642 */ 2643 if (th->th_ack == tp->snd_max) { 2644 tcp_timer_activate(tp, TT_REXMT, 0); 2645 needoutput = 1; 2646 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2647 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2648 2649 /* 2650 * If no data (only SYN) was ACK'd, 2651 * skip rest of ACK processing. 2652 */ 2653 if (acked == 0) 2654 goto step6; 2655 2656 /* 2657 * Let the congestion control algorithm update congestion 2658 * control related information. This typically means increasing 2659 * the congestion window. 2660 */ 2661 cc_ack_received(tp, th, CC_ACK); 2662 2663 SOCKBUF_LOCK(&so->so_snd); 2664 if (acked > sbavail(&so->so_snd)) { 2665 tp->snd_wnd -= sbavail(&so->so_snd); 2666 mfree = sbcut_locked(&so->so_snd, 2667 (int)sbavail(&so->so_snd)); 2668 ourfinisacked = 1; 2669 } else { 2670 mfree = sbcut_locked(&so->so_snd, acked); 2671 tp->snd_wnd -= acked; 2672 ourfinisacked = 0; 2673 } 2674 /* NB: sowwakeup_locked() does an implicit unlock. */ 2675 sowwakeup_locked(so); 2676 m_freem(mfree); 2677 /* Detect una wraparound. */ 2678 if (!IN_RECOVERY(tp->t_flags) && 2679 SEQ_GT(tp->snd_una, tp->snd_recover) && 2680 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2681 tp->snd_recover = th->th_ack - 1; 2682 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2683 if (IN_RECOVERY(tp->t_flags) && 2684 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2685 EXIT_RECOVERY(tp->t_flags); 2686 } 2687 tp->snd_una = th->th_ack; 2688 if (tp->t_flags & TF_SACK_PERMIT) { 2689 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2690 tp->snd_recover = tp->snd_una; 2691 } 2692 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2693 tp->snd_nxt = tp->snd_una; 2694 2695 switch (tp->t_state) { 2696 2697 /* 2698 * In FIN_WAIT_1 STATE in addition to the processing 2699 * for the ESTABLISHED state if our FIN is now acknowledged 2700 * then enter FIN_WAIT_2. 2701 */ 2702 case TCPS_FIN_WAIT_1: 2703 if (ourfinisacked) { 2704 /* 2705 * If we can't receive any more 2706 * data, then closing user can proceed. 2707 * Starting the timer is contrary to the 2708 * specification, but if we don't get a FIN 2709 * we'll hang forever. 2710 * 2711 * XXXjl: 2712 * we should release the tp also, and use a 2713 * compressed state. 2714 */ 2715 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2716 soisdisconnected(so); 2717 tcp_timer_activate(tp, TT_2MSL, 2718 (tcp_fast_finwait2_recycle ? 2719 tcp_finwait2_timeout : 2720 TP_MAXIDLE(tp))); 2721 } 2722 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2723 } 2724 break; 2725 2726 /* 2727 * In CLOSING STATE in addition to the processing for 2728 * the ESTABLISHED state if the ACK acknowledges our FIN 2729 * then enter the TIME-WAIT state, otherwise ignore 2730 * the segment. 2731 */ 2732 case TCPS_CLOSING: 2733 if (ourfinisacked) { 2734 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2735 tcp_twstart(tp); 2736 INP_INFO_WUNLOCK(&V_tcbinfo); 2737 m_freem(m); 2738 return; 2739 } 2740 break; 2741 2742 /* 2743 * In LAST_ACK, we may still be waiting for data to drain 2744 * and/or to be acked, as well as for the ack of our FIN. 2745 * If our FIN is now acknowledged, delete the TCB, 2746 * enter the closed state and return. 2747 */ 2748 case TCPS_LAST_ACK: 2749 if (ourfinisacked) { 2750 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2751 tp = tcp_close(tp); 2752 goto drop; 2753 } 2754 break; 2755 } 2756 } 2757 2758 step6: 2759 INP_WLOCK_ASSERT(tp->t_inpcb); 2760 2761 /* 2762 * Update window information. 2763 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2764 */ 2765 if ((thflags & TH_ACK) && 2766 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2767 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2768 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2769 /* keep track of pure window updates */ 2770 if (tlen == 0 && 2771 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2772 TCPSTAT_INC(tcps_rcvwinupd); 2773 tp->snd_wnd = tiwin; 2774 tp->snd_wl1 = th->th_seq; 2775 tp->snd_wl2 = th->th_ack; 2776 if (tp->snd_wnd > tp->max_sndwnd) 2777 tp->max_sndwnd = tp->snd_wnd; 2778 needoutput = 1; 2779 } 2780 2781 /* 2782 * Process segments with URG. 2783 */ 2784 if ((thflags & TH_URG) && th->th_urp && 2785 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2786 /* 2787 * This is a kludge, but if we receive and accept 2788 * random urgent pointers, we'll crash in 2789 * soreceive. It's hard to imagine someone 2790 * actually wanting to send this much urgent data. 2791 */ 2792 SOCKBUF_LOCK(&so->so_rcv); 2793 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2794 th->th_urp = 0; /* XXX */ 2795 thflags &= ~TH_URG; /* XXX */ 2796 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2797 goto dodata; /* XXX */ 2798 } 2799 /* 2800 * If this segment advances the known urgent pointer, 2801 * then mark the data stream. This should not happen 2802 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2803 * a FIN has been received from the remote side. 2804 * In these states we ignore the URG. 2805 * 2806 * According to RFC961 (Assigned Protocols), 2807 * the urgent pointer points to the last octet 2808 * of urgent data. We continue, however, 2809 * to consider it to indicate the first octet 2810 * of data past the urgent section as the original 2811 * spec states (in one of two places). 2812 */ 2813 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2814 tp->rcv_up = th->th_seq + th->th_urp; 2815 so->so_oobmark = sbavail(&so->so_rcv) + 2816 (tp->rcv_up - tp->rcv_nxt) - 1; 2817 if (so->so_oobmark == 0) 2818 so->so_rcv.sb_state |= SBS_RCVATMARK; 2819 sohasoutofband(so); 2820 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2821 } 2822 SOCKBUF_UNLOCK(&so->so_rcv); 2823 /* 2824 * Remove out of band data so doesn't get presented to user. 2825 * This can happen independent of advancing the URG pointer, 2826 * but if two URG's are pending at once, some out-of-band 2827 * data may creep in... ick. 2828 */ 2829 if (th->th_urp <= (u_long)tlen && 2830 !(so->so_options & SO_OOBINLINE)) { 2831 /* hdr drop is delayed */ 2832 tcp_pulloutofband(so, th, m, drop_hdrlen); 2833 } 2834 } else { 2835 /* 2836 * If no out of band data is expected, 2837 * pull receive urgent pointer along 2838 * with the receive window. 2839 */ 2840 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2841 tp->rcv_up = tp->rcv_nxt; 2842 } 2843 dodata: /* XXX */ 2844 INP_WLOCK_ASSERT(tp->t_inpcb); 2845 2846 /* 2847 * Process the segment text, merging it into the TCP sequencing queue, 2848 * and arranging for acknowledgment of receipt if necessary. 2849 * This process logically involves adjusting tp->rcv_wnd as data 2850 * is presented to the user (this happens in tcp_usrreq.c, 2851 * case PRU_RCVD). If a FIN has already been received on this 2852 * connection then we just ignore the text. 2853 */ 2854 if ((tlen || (thflags & TH_FIN)) && 2855 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2856 tcp_seq save_start = th->th_seq; 2857 m_adj(m, drop_hdrlen); /* delayed header drop */ 2858 /* 2859 * Insert segment which includes th into TCP reassembly queue 2860 * with control block tp. Set thflags to whether reassembly now 2861 * includes a segment with FIN. This handles the common case 2862 * inline (segment is the next to be received on an established 2863 * connection, and the queue is empty), avoiding linkage into 2864 * and removal from the queue and repetition of various 2865 * conversions. 2866 * Set DELACK for segments received in order, but ack 2867 * immediately when segments are out of order (so 2868 * fast retransmit can work). 2869 */ 2870 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL && 2871 TCPS_HAVEESTABLISHED(tp->t_state)) { 2872 if (DELAY_ACK(tp, tlen)) 2873 tp->t_flags |= TF_DELACK; 2874 else 2875 tp->t_flags |= TF_ACKNOW; 2876 tp->rcv_nxt += tlen; 2877 thflags = th->th_flags & TH_FIN; 2878 TCPSTAT_INC(tcps_rcvpack); 2879 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2880 ND6_HINT(tp); 2881 SOCKBUF_LOCK(&so->so_rcv); 2882 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2883 m_freem(m); 2884 else 2885 sbappendstream_locked(&so->so_rcv, m); 2886 /* NB: sorwakeup_locked() does an implicit unlock. */ 2887 sorwakeup_locked(so); 2888 } else { 2889 /* 2890 * XXX: Due to the header drop above "th" is 2891 * theoretically invalid by now. Fortunately 2892 * m_adj() doesn't actually frees any mbufs 2893 * when trimming from the head. 2894 */ 2895 thflags = tcp_reass(tp, th, &tlen, m); 2896 tp->t_flags |= TF_ACKNOW; 2897 } 2898 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2899 tcp_update_sack_list(tp, save_start, save_start + tlen); 2900 #if 0 2901 /* 2902 * Note the amount of data that peer has sent into 2903 * our window, in order to estimate the sender's 2904 * buffer size. 2905 * XXX: Unused. 2906 */ 2907 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2908 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2909 else 2910 len = so->so_rcv.sb_hiwat; 2911 #endif 2912 } else { 2913 m_freem(m); 2914 thflags &= ~TH_FIN; 2915 } 2916 2917 /* 2918 * If FIN is received ACK the FIN and let the user know 2919 * that the connection is closing. 2920 */ 2921 if (thflags & TH_FIN) { 2922 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2923 socantrcvmore(so); 2924 /* 2925 * If connection is half-synchronized 2926 * (ie NEEDSYN flag on) then delay ACK, 2927 * so it may be piggybacked when SYN is sent. 2928 * Otherwise, since we received a FIN then no 2929 * more input can be expected, send ACK now. 2930 */ 2931 if (tp->t_flags & TF_NEEDSYN) 2932 tp->t_flags |= TF_DELACK; 2933 else 2934 tp->t_flags |= TF_ACKNOW; 2935 tp->rcv_nxt++; 2936 } 2937 switch (tp->t_state) { 2938 2939 /* 2940 * In SYN_RECEIVED and ESTABLISHED STATES 2941 * enter the CLOSE_WAIT state. 2942 */ 2943 case TCPS_SYN_RECEIVED: 2944 tp->t_starttime = ticks; 2945 /* FALLTHROUGH */ 2946 case TCPS_ESTABLISHED: 2947 tcp_state_change(tp, TCPS_CLOSE_WAIT); 2948 break; 2949 2950 /* 2951 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2952 * enter the CLOSING state. 2953 */ 2954 case TCPS_FIN_WAIT_1: 2955 tcp_state_change(tp, TCPS_CLOSING); 2956 break; 2957 2958 /* 2959 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2960 * starting the time-wait timer, turning off the other 2961 * standard timers. 2962 */ 2963 case TCPS_FIN_WAIT_2: 2964 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2965 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2966 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2967 ti_locked)); 2968 2969 tcp_twstart(tp); 2970 INP_INFO_WUNLOCK(&V_tcbinfo); 2971 return; 2972 } 2973 } 2974 if (ti_locked == TI_WLOCKED) 2975 INP_INFO_WUNLOCK(&V_tcbinfo); 2976 ti_locked = TI_UNLOCKED; 2977 2978 #ifdef TCPDEBUG 2979 if (so->so_options & SO_DEBUG) 2980 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2981 &tcp_savetcp, 0); 2982 #endif 2983 2984 /* 2985 * Return any desired output. 2986 */ 2987 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2988 (void) tcp_output(tp); 2989 2990 check_delack: 2991 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2992 __func__, ti_locked)); 2993 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2994 INP_WLOCK_ASSERT(tp->t_inpcb); 2995 2996 if (tp->t_flags & TF_DELACK) { 2997 tp->t_flags &= ~TF_DELACK; 2998 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2999 } 3000 INP_WUNLOCK(tp->t_inpcb); 3001 return; 3002 3003 dropafterack: 3004 /* 3005 * Generate an ACK dropping incoming segment if it occupies 3006 * sequence space, where the ACK reflects our state. 3007 * 3008 * We can now skip the test for the RST flag since all 3009 * paths to this code happen after packets containing 3010 * RST have been dropped. 3011 * 3012 * In the SYN-RECEIVED state, don't send an ACK unless the 3013 * segment we received passes the SYN-RECEIVED ACK test. 3014 * If it fails send a RST. This breaks the loop in the 3015 * "LAND" DoS attack, and also prevents an ACK storm 3016 * between two listening ports that have been sent forged 3017 * SYN segments, each with the source address of the other. 3018 */ 3019 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3020 (SEQ_GT(tp->snd_una, th->th_ack) || 3021 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3022 rstreason = BANDLIM_RST_OPENPORT; 3023 goto dropwithreset; 3024 } 3025 #ifdef TCPDEBUG 3026 if (so->so_options & SO_DEBUG) 3027 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3028 &tcp_savetcp, 0); 3029 #endif 3030 if (ti_locked == TI_WLOCKED) 3031 INP_INFO_WUNLOCK(&V_tcbinfo); 3032 ti_locked = TI_UNLOCKED; 3033 3034 tp->t_flags |= TF_ACKNOW; 3035 (void) tcp_output(tp); 3036 INP_WUNLOCK(tp->t_inpcb); 3037 m_freem(m); 3038 return; 3039 3040 dropwithreset: 3041 if (ti_locked == TI_WLOCKED) 3042 INP_INFO_WUNLOCK(&V_tcbinfo); 3043 ti_locked = TI_UNLOCKED; 3044 3045 if (tp != NULL) { 3046 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3047 INP_WUNLOCK(tp->t_inpcb); 3048 } else 3049 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3050 return; 3051 3052 drop: 3053 if (ti_locked == TI_WLOCKED) { 3054 INP_INFO_WUNLOCK(&V_tcbinfo); 3055 ti_locked = TI_UNLOCKED; 3056 } 3057 #ifdef INVARIANTS 3058 else 3059 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3060 #endif 3061 3062 /* 3063 * Drop space held by incoming segment and return. 3064 */ 3065 #ifdef TCPDEBUG 3066 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3067 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3068 &tcp_savetcp, 0); 3069 #endif 3070 if (tp != NULL) 3071 INP_WUNLOCK(tp->t_inpcb); 3072 m_freem(m); 3073 } 3074 3075 /* 3076 * Issue RST and make ACK acceptable to originator of segment. 3077 * The mbuf must still include the original packet header. 3078 * tp may be NULL. 3079 */ 3080 static void 3081 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3082 int tlen, int rstreason) 3083 { 3084 #ifdef INET 3085 struct ip *ip; 3086 #endif 3087 #ifdef INET6 3088 struct ip6_hdr *ip6; 3089 #endif 3090 3091 if (tp != NULL) { 3092 INP_WLOCK_ASSERT(tp->t_inpcb); 3093 } 3094 3095 /* Don't bother if destination was broadcast/multicast. */ 3096 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3097 goto drop; 3098 #ifdef INET6 3099 if (mtod(m, struct ip *)->ip_v == 6) { 3100 ip6 = mtod(m, struct ip6_hdr *); 3101 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3102 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3103 goto drop; 3104 /* IPv6 anycast check is done at tcp6_input() */ 3105 } 3106 #endif 3107 #if defined(INET) && defined(INET6) 3108 else 3109 #endif 3110 #ifdef INET 3111 { 3112 ip = mtod(m, struct ip *); 3113 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3114 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3115 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3116 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3117 goto drop; 3118 } 3119 #endif 3120 3121 /* Perform bandwidth limiting. */ 3122 if (badport_bandlim(rstreason) < 0) 3123 goto drop; 3124 3125 /* tcp_respond consumes the mbuf chain. */ 3126 if (th->th_flags & TH_ACK) { 3127 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3128 th->th_ack, TH_RST); 3129 } else { 3130 if (th->th_flags & TH_SYN) 3131 tlen++; 3132 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3133 (tcp_seq)0, TH_RST|TH_ACK); 3134 } 3135 return; 3136 drop: 3137 m_freem(m); 3138 } 3139 3140 /* 3141 * Parse TCP options and place in tcpopt. 3142 */ 3143 static void 3144 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3145 { 3146 int opt, optlen; 3147 3148 to->to_flags = 0; 3149 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3150 opt = cp[0]; 3151 if (opt == TCPOPT_EOL) 3152 break; 3153 if (opt == TCPOPT_NOP) 3154 optlen = 1; 3155 else { 3156 if (cnt < 2) 3157 break; 3158 optlen = cp[1]; 3159 if (optlen < 2 || optlen > cnt) 3160 break; 3161 } 3162 switch (opt) { 3163 case TCPOPT_MAXSEG: 3164 if (optlen != TCPOLEN_MAXSEG) 3165 continue; 3166 if (!(flags & TO_SYN)) 3167 continue; 3168 to->to_flags |= TOF_MSS; 3169 bcopy((char *)cp + 2, 3170 (char *)&to->to_mss, sizeof(to->to_mss)); 3171 to->to_mss = ntohs(to->to_mss); 3172 break; 3173 case TCPOPT_WINDOW: 3174 if (optlen != TCPOLEN_WINDOW) 3175 continue; 3176 if (!(flags & TO_SYN)) 3177 continue; 3178 to->to_flags |= TOF_SCALE; 3179 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3180 break; 3181 case TCPOPT_TIMESTAMP: 3182 if (optlen != TCPOLEN_TIMESTAMP) 3183 continue; 3184 to->to_flags |= TOF_TS; 3185 bcopy((char *)cp + 2, 3186 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3187 to->to_tsval = ntohl(to->to_tsval); 3188 bcopy((char *)cp + 6, 3189 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3190 to->to_tsecr = ntohl(to->to_tsecr); 3191 break; 3192 #ifdef TCP_SIGNATURE 3193 /* 3194 * XXX In order to reply to a host which has set the 3195 * TCP_SIGNATURE option in its initial SYN, we have to 3196 * record the fact that the option was observed here 3197 * for the syncache code to perform the correct response. 3198 */ 3199 case TCPOPT_SIGNATURE: 3200 if (optlen != TCPOLEN_SIGNATURE) 3201 continue; 3202 to->to_flags |= TOF_SIGNATURE; 3203 to->to_signature = cp + 2; 3204 break; 3205 #endif 3206 case TCPOPT_SACK_PERMITTED: 3207 if (optlen != TCPOLEN_SACK_PERMITTED) 3208 continue; 3209 if (!(flags & TO_SYN)) 3210 continue; 3211 if (!V_tcp_do_sack) 3212 continue; 3213 to->to_flags |= TOF_SACKPERM; 3214 break; 3215 case TCPOPT_SACK: 3216 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3217 continue; 3218 if (flags & TO_SYN) 3219 continue; 3220 to->to_flags |= TOF_SACK; 3221 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3222 to->to_sacks = cp + 2; 3223 TCPSTAT_INC(tcps_sack_rcv_blocks); 3224 break; 3225 default: 3226 continue; 3227 } 3228 } 3229 } 3230 3231 /* 3232 * Pull out of band byte out of a segment so 3233 * it doesn't appear in the user's data queue. 3234 * It is still reflected in the segment length for 3235 * sequencing purposes. 3236 */ 3237 static void 3238 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3239 int off) 3240 { 3241 int cnt = off + th->th_urp - 1; 3242 3243 while (cnt >= 0) { 3244 if (m->m_len > cnt) { 3245 char *cp = mtod(m, caddr_t) + cnt; 3246 struct tcpcb *tp = sototcpcb(so); 3247 3248 INP_WLOCK_ASSERT(tp->t_inpcb); 3249 3250 tp->t_iobc = *cp; 3251 tp->t_oobflags |= TCPOOB_HAVEDATA; 3252 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3253 m->m_len--; 3254 if (m->m_flags & M_PKTHDR) 3255 m->m_pkthdr.len--; 3256 return; 3257 } 3258 cnt -= m->m_len; 3259 m = m->m_next; 3260 if (m == NULL) 3261 break; 3262 } 3263 panic("tcp_pulloutofband"); 3264 } 3265 3266 /* 3267 * Collect new round-trip time estimate 3268 * and update averages and current timeout. 3269 */ 3270 static void 3271 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3272 { 3273 int delta; 3274 3275 INP_WLOCK_ASSERT(tp->t_inpcb); 3276 3277 TCPSTAT_INC(tcps_rttupdated); 3278 tp->t_rttupdated++; 3279 if (tp->t_srtt != 0) { 3280 /* 3281 * srtt is stored as fixed point with 5 bits after the 3282 * binary point (i.e., scaled by 8). The following magic 3283 * is equivalent to the smoothing algorithm in rfc793 with 3284 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3285 * point). Adjust rtt to origin 0. 3286 */ 3287 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3288 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3289 3290 if ((tp->t_srtt += delta) <= 0) 3291 tp->t_srtt = 1; 3292 3293 /* 3294 * We accumulate a smoothed rtt variance (actually, a 3295 * smoothed mean difference), then set the retransmit 3296 * timer to smoothed rtt + 4 times the smoothed variance. 3297 * rttvar is stored as fixed point with 4 bits after the 3298 * binary point (scaled by 16). The following is 3299 * equivalent to rfc793 smoothing with an alpha of .75 3300 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3301 * rfc793's wired-in beta. 3302 */ 3303 if (delta < 0) 3304 delta = -delta; 3305 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3306 if ((tp->t_rttvar += delta) <= 0) 3307 tp->t_rttvar = 1; 3308 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3309 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3310 } else { 3311 /* 3312 * No rtt measurement yet - use the unsmoothed rtt. 3313 * Set the variance to half the rtt (so our first 3314 * retransmit happens at 3*rtt). 3315 */ 3316 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3317 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3318 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3319 } 3320 tp->t_rtttime = 0; 3321 tp->t_rxtshift = 0; 3322 3323 /* 3324 * the retransmit should happen at rtt + 4 * rttvar. 3325 * Because of the way we do the smoothing, srtt and rttvar 3326 * will each average +1/2 tick of bias. When we compute 3327 * the retransmit timer, we want 1/2 tick of rounding and 3328 * 1 extra tick because of +-1/2 tick uncertainty in the 3329 * firing of the timer. The bias will give us exactly the 3330 * 1.5 tick we need. But, because the bias is 3331 * statistical, we have to test that we don't drop below 3332 * the minimum feasible timer (which is 2 ticks). 3333 */ 3334 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3335 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3336 3337 /* 3338 * We received an ack for a packet that wasn't retransmitted; 3339 * it is probably safe to discard any error indications we've 3340 * received recently. This isn't quite right, but close enough 3341 * for now (a route might have failed after we sent a segment, 3342 * and the return path might not be symmetrical). 3343 */ 3344 tp->t_softerror = 0; 3345 } 3346 3347 /* 3348 * Determine a reasonable value for maxseg size. 3349 * If the route is known, check route for mtu. 3350 * If none, use an mss that can be handled on the outgoing interface 3351 * without forcing IP to fragment. If no route is found, route has no mtu, 3352 * or the destination isn't local, use a default, hopefully conservative 3353 * size (usually 512 or the default IP max size, but no more than the mtu 3354 * of the interface), as we can't discover anything about intervening 3355 * gateways or networks. We also initialize the congestion/slow start 3356 * window to be a single segment if the destination isn't local. 3357 * While looking at the routing entry, we also initialize other path-dependent 3358 * parameters from pre-set or cached values in the routing entry. 3359 * 3360 * Also take into account the space needed for options that we 3361 * send regularly. Make maxseg shorter by that amount to assure 3362 * that we can send maxseg amount of data even when the options 3363 * are present. Store the upper limit of the length of options plus 3364 * data in maxopd. 3365 * 3366 * NOTE that this routine is only called when we process an incoming 3367 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3368 * settings are handled in tcp_mssopt(). 3369 */ 3370 void 3371 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3372 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3373 { 3374 int mss = 0; 3375 u_long maxmtu = 0; 3376 struct inpcb *inp = tp->t_inpcb; 3377 struct hc_metrics_lite metrics; 3378 int origoffer; 3379 #ifdef INET6 3380 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3381 size_t min_protoh = isipv6 ? 3382 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3383 sizeof (struct tcpiphdr); 3384 #else 3385 const size_t min_protoh = sizeof(struct tcpiphdr); 3386 #endif 3387 3388 INP_WLOCK_ASSERT(tp->t_inpcb); 3389 3390 if (mtuoffer != -1) { 3391 KASSERT(offer == -1, ("%s: conflict", __func__)); 3392 offer = mtuoffer - min_protoh; 3393 } 3394 origoffer = offer; 3395 3396 /* Initialize. */ 3397 #ifdef INET6 3398 if (isipv6) { 3399 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3400 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3401 } 3402 #endif 3403 #if defined(INET) && defined(INET6) 3404 else 3405 #endif 3406 #ifdef INET 3407 { 3408 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3409 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3410 } 3411 #endif 3412 3413 /* 3414 * No route to sender, stay with default mss and return. 3415 */ 3416 if (maxmtu == 0) { 3417 /* 3418 * In case we return early we need to initialize metrics 3419 * to a defined state as tcp_hc_get() would do for us 3420 * if there was no cache hit. 3421 */ 3422 if (metricptr != NULL) 3423 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3424 return; 3425 } 3426 3427 /* What have we got? */ 3428 switch (offer) { 3429 case 0: 3430 /* 3431 * Offer == 0 means that there was no MSS on the SYN 3432 * segment, in this case we use tcp_mssdflt as 3433 * already assigned to t_maxopd above. 3434 */ 3435 offer = tp->t_maxopd; 3436 break; 3437 3438 case -1: 3439 /* 3440 * Offer == -1 means that we didn't receive SYN yet. 3441 */ 3442 /* FALLTHROUGH */ 3443 3444 default: 3445 /* 3446 * Prevent DoS attack with too small MSS. Round up 3447 * to at least minmss. 3448 */ 3449 offer = max(offer, V_tcp_minmss); 3450 } 3451 3452 /* 3453 * rmx information is now retrieved from tcp_hostcache. 3454 */ 3455 tcp_hc_get(&inp->inp_inc, &metrics); 3456 if (metricptr != NULL) 3457 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3458 3459 /* 3460 * If there's a discovered mtu in tcp hostcache, use it. 3461 * Else, use the link mtu. 3462 */ 3463 if (metrics.rmx_mtu) 3464 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3465 else { 3466 #ifdef INET6 3467 if (isipv6) { 3468 mss = maxmtu - min_protoh; 3469 if (!V_path_mtu_discovery && 3470 !in6_localaddr(&inp->in6p_faddr)) 3471 mss = min(mss, V_tcp_v6mssdflt); 3472 } 3473 #endif 3474 #if defined(INET) && defined(INET6) 3475 else 3476 #endif 3477 #ifdef INET 3478 { 3479 mss = maxmtu - min_protoh; 3480 if (!V_path_mtu_discovery && 3481 !in_localaddr(inp->inp_faddr)) 3482 mss = min(mss, V_tcp_mssdflt); 3483 } 3484 #endif 3485 /* 3486 * XXX - The above conditional (mss = maxmtu - min_protoh) 3487 * probably violates the TCP spec. 3488 * The problem is that, since we don't know the 3489 * other end's MSS, we are supposed to use a conservative 3490 * default. But, if we do that, then MTU discovery will 3491 * never actually take place, because the conservative 3492 * default is much less than the MTUs typically seen 3493 * on the Internet today. For the moment, we'll sweep 3494 * this under the carpet. 3495 * 3496 * The conservative default might not actually be a problem 3497 * if the only case this occurs is when sending an initial 3498 * SYN with options and data to a host we've never talked 3499 * to before. Then, they will reply with an MSS value which 3500 * will get recorded and the new parameters should get 3501 * recomputed. For Further Study. 3502 */ 3503 } 3504 mss = min(mss, offer); 3505 3506 /* 3507 * Sanity check: make sure that maxopd will be large 3508 * enough to allow some data on segments even if the 3509 * all the option space is used (40bytes). Otherwise 3510 * funny things may happen in tcp_output. 3511 */ 3512 mss = max(mss, 64); 3513 3514 /* 3515 * maxopd stores the maximum length of data AND options 3516 * in a segment; maxseg is the amount of data in a normal 3517 * segment. We need to store this value (maxopd) apart 3518 * from maxseg, because now every segment carries options 3519 * and thus we normally have somewhat less data in segments. 3520 */ 3521 tp->t_maxopd = mss; 3522 3523 /* 3524 * origoffer==-1 indicates that no segments were received yet. 3525 * In this case we just guess. 3526 */ 3527 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3528 (origoffer == -1 || 3529 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3530 mss -= TCPOLEN_TSTAMP_APPA; 3531 3532 tp->t_maxseg = mss; 3533 } 3534 3535 void 3536 tcp_mss(struct tcpcb *tp, int offer) 3537 { 3538 int mss; 3539 u_long bufsize; 3540 struct inpcb *inp; 3541 struct socket *so; 3542 struct hc_metrics_lite metrics; 3543 struct tcp_ifcap cap; 3544 3545 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3546 3547 bzero(&cap, sizeof(cap)); 3548 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3549 3550 mss = tp->t_maxseg; 3551 inp = tp->t_inpcb; 3552 3553 /* 3554 * If there's a pipesize, change the socket buffer to that size, 3555 * don't change if sb_hiwat is different than default (then it 3556 * has been changed on purpose with setsockopt). 3557 * Make the socket buffers an integral number of mss units; 3558 * if the mss is larger than the socket buffer, decrease the mss. 3559 */ 3560 so = inp->inp_socket; 3561 SOCKBUF_LOCK(&so->so_snd); 3562 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3563 bufsize = metrics.rmx_sendpipe; 3564 else 3565 bufsize = so->so_snd.sb_hiwat; 3566 if (bufsize < mss) 3567 mss = bufsize; 3568 else { 3569 bufsize = roundup(bufsize, mss); 3570 if (bufsize > sb_max) 3571 bufsize = sb_max; 3572 if (bufsize > so->so_snd.sb_hiwat) 3573 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3574 } 3575 SOCKBUF_UNLOCK(&so->so_snd); 3576 tp->t_maxseg = mss; 3577 3578 SOCKBUF_LOCK(&so->so_rcv); 3579 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3580 bufsize = metrics.rmx_recvpipe; 3581 else 3582 bufsize = so->so_rcv.sb_hiwat; 3583 if (bufsize > mss) { 3584 bufsize = roundup(bufsize, mss); 3585 if (bufsize > sb_max) 3586 bufsize = sb_max; 3587 if (bufsize > so->so_rcv.sb_hiwat) 3588 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3589 } 3590 SOCKBUF_UNLOCK(&so->so_rcv); 3591 3592 /* Check the interface for TSO capabilities. */ 3593 if (cap.ifcap & CSUM_TSO) { 3594 tp->t_flags |= TF_TSO; 3595 tp->t_tsomax = cap.tsomax; 3596 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3597 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3598 } 3599 } 3600 3601 /* 3602 * Determine the MSS option to send on an outgoing SYN. 3603 */ 3604 int 3605 tcp_mssopt(struct in_conninfo *inc) 3606 { 3607 int mss = 0; 3608 u_long maxmtu = 0; 3609 u_long thcmtu = 0; 3610 size_t min_protoh; 3611 3612 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3613 3614 #ifdef INET6 3615 if (inc->inc_flags & INC_ISIPV6) { 3616 mss = V_tcp_v6mssdflt; 3617 maxmtu = tcp_maxmtu6(inc, NULL); 3618 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3619 } 3620 #endif 3621 #if defined(INET) && defined(INET6) 3622 else 3623 #endif 3624 #ifdef INET 3625 { 3626 mss = V_tcp_mssdflt; 3627 maxmtu = tcp_maxmtu(inc, NULL); 3628 min_protoh = sizeof(struct tcpiphdr); 3629 } 3630 #endif 3631 #if defined(INET6) || defined(INET) 3632 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3633 #endif 3634 3635 if (maxmtu && thcmtu) 3636 mss = min(maxmtu, thcmtu) - min_protoh; 3637 else if (maxmtu || thcmtu) 3638 mss = max(maxmtu, thcmtu) - min_protoh; 3639 3640 return (mss); 3641 } 3642 3643 3644 /* 3645 * On a partial ack arrives, force the retransmission of the 3646 * next unacknowledged segment. Do not clear tp->t_dupacks. 3647 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3648 * be started again. 3649 */ 3650 static void 3651 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3652 { 3653 tcp_seq onxt = tp->snd_nxt; 3654 u_long ocwnd = tp->snd_cwnd; 3655 3656 INP_WLOCK_ASSERT(tp->t_inpcb); 3657 3658 tcp_timer_activate(tp, TT_REXMT, 0); 3659 tp->t_rtttime = 0; 3660 tp->snd_nxt = th->th_ack; 3661 /* 3662 * Set snd_cwnd to one segment beyond acknowledged offset. 3663 * (tp->snd_una has not yet been updated when this function is called.) 3664 */ 3665 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3666 tp->t_flags |= TF_ACKNOW; 3667 (void) tcp_output(tp); 3668 tp->snd_cwnd = ocwnd; 3669 if (SEQ_GT(onxt, tp->snd_nxt)) 3670 tp->snd_nxt = onxt; 3671 /* 3672 * Partial window deflation. Relies on fact that tp->snd_una 3673 * not updated yet. 3674 */ 3675 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3676 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3677 else 3678 tp->snd_cwnd = 0; 3679 tp->snd_cwnd += tp->t_maxseg; 3680 } 3681