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_kdtrace.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/kernel.h> 62 #include <sys/hhook.h> 63 #include <sys/malloc.h> 64 #include <sys/mbuf.h> 65 #include <sys/proc.h> /* for proc0 declaration */ 66 #include <sys/protosw.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77 #include <vm/uma.h> 78 79 #include <net/if.h> 80 #include <net/if_var.h> 81 #include <net/route.h> 82 #include <net/vnet.h> 83 84 #define TCPSTATES /* for logging */ 85 86 #include <netinet/cc.h> 87 #include <netinet/in.h> 88 #include <netinet/in_kdtrace.h> 89 #include <netinet/in_pcb.h> 90 #include <netinet/in_systm.h> 91 #include <netinet/in_var.h> 92 #include <netinet/ip.h> 93 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 94 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 95 #include <netinet/ip_var.h> 96 #include <netinet/ip_options.h> 97 #include <netinet/ip6.h> 98 #include <netinet/icmp6.h> 99 #include <netinet6/in6_pcb.h> 100 #include <netinet6/ip6_var.h> 101 #include <netinet6/nd6.h> 102 #include <netinet/tcp_fsm.h> 103 #include <netinet/tcp_seq.h> 104 #include <netinet/tcp_timer.h> 105 #include <netinet/tcp_var.h> 106 #include <netinet6/tcp6_var.h> 107 #include <netinet/tcpip.h> 108 #include <netinet/tcp_syncache.h> 109 #ifdef TCPDEBUG 110 #include <netinet/tcp_debug.h> 111 #endif /* TCPDEBUG */ 112 #ifdef TCP_OFFLOAD 113 #include <netinet/tcp_offload.h> 114 #endif 115 116 #ifdef IPSEC 117 #include <netipsec/ipsec.h> 118 #include <netipsec/ipsec6.h> 119 #endif /*IPSEC*/ 120 121 #include <machine/in_cksum.h> 122 123 #include <security/mac/mac_framework.h> 124 125 const int tcprexmtthresh = 3; 126 127 int tcp_log_in_vain = 0; 128 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 129 &tcp_log_in_vain, 0, 130 "Log all incoming TCP segments to closed ports"); 131 132 VNET_DEFINE(int, blackhole) = 0; 133 #define V_blackhole VNET(blackhole) 134 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 135 &VNET_NAME(blackhole), 0, 136 "Do not send RST on segments to closed ports"); 137 138 VNET_DEFINE(int, tcp_delack_enabled) = 1; 139 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 140 &VNET_NAME(tcp_delack_enabled), 0, 141 "Delay ACK to try and piggyback it onto a data packet"); 142 143 VNET_DEFINE(int, drop_synfin) = 0; 144 #define V_drop_synfin VNET(drop_synfin) 145 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 146 &VNET_NAME(drop_synfin), 0, 147 "Drop TCP packets with SYN+FIN set"); 148 149 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 150 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 151 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 152 &VNET_NAME(tcp_do_rfc3042), 0, 153 "Enable RFC 3042 (Limited Transmit)"); 154 155 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 156 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 157 &VNET_NAME(tcp_do_rfc3390), 0, 158 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 159 160 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 161 "Experimental TCP extensions"); 162 163 VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 164 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 165 &VNET_NAME(tcp_do_initcwnd10), 0, 166 "Enable draft-ietf-tcpm-initcwnd-05 (Increasing initial CWND to 10)"); 167 168 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 169 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 170 &VNET_NAME(tcp_do_rfc3465), 0, 171 "Enable RFC 3465 (Appropriate Byte Counting)"); 172 173 VNET_DEFINE(int, tcp_abc_l_var) = 2; 174 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 175 &VNET_NAME(tcp_abc_l_var), 2, 176 "Cap the max cwnd increment during slow-start to this number of segments"); 177 178 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 179 180 VNET_DEFINE(int, tcp_do_ecn) = 0; 181 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 182 &VNET_NAME(tcp_do_ecn), 0, 183 "TCP ECN support"); 184 185 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 186 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 187 &VNET_NAME(tcp_ecn_maxretries), 0, 188 "Max retries before giving up on ECN"); 189 190 VNET_DEFINE(int, tcp_insecure_rst) = 0; 191 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 192 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 193 &VNET_NAME(tcp_insecure_rst), 0, 194 "Follow the old (insecure) criteria for accepting RST packets"); 195 196 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 197 #define V_tcp_recvspace VNET(tcp_recvspace) 198 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 199 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 200 201 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 202 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 203 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 204 &VNET_NAME(tcp_do_autorcvbuf), 0, 205 "Enable automatic receive buffer sizing"); 206 207 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 208 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 209 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 210 &VNET_NAME(tcp_autorcvbuf_inc), 0, 211 "Incrementor step size of automatic receive buffer"); 212 213 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 214 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 215 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 216 &VNET_NAME(tcp_autorcvbuf_max), 0, 217 "Max size of automatic receive buffer"); 218 219 VNET_DEFINE(struct inpcbhead, tcb); 220 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 221 VNET_DEFINE(struct inpcbinfo, tcbinfo); 222 223 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 224 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 225 struct socket *, struct tcpcb *, int, int, uint8_t, 226 int); 227 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 228 struct tcpcb *, int, int); 229 static void tcp_pulloutofband(struct socket *, 230 struct tcphdr *, struct mbuf *, int); 231 static void tcp_xmit_timer(struct tcpcb *, int); 232 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 233 static void inline tcp_fields_to_host(struct tcphdr *); 234 #ifdef TCP_SIGNATURE 235 static void inline tcp_fields_to_net(struct tcphdr *); 236 static int inline tcp_signature_verify_input(struct mbuf *, int, int, 237 int, struct tcpopt *, struct tcphdr *, u_int); 238 #endif 239 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 240 uint16_t type); 241 static void inline cc_conn_init(struct tcpcb *tp); 242 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 243 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 244 struct tcphdr *th, struct tcpopt *to); 245 246 /* 247 * TCP statistics are stored in an "array" of counter(9)s. 248 */ 249 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 250 VNET_PCPUSTAT_SYSINIT(tcpstat); 251 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 252 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 253 254 #ifdef VIMAGE 255 VNET_PCPUSTAT_SYSUNINIT(tcpstat); 256 #endif /* VIMAGE */ 257 /* 258 * Kernel module interface for updating tcpstat. The argument is an index 259 * into tcpstat treated as an array. 260 */ 261 void 262 kmod_tcpstat_inc(int statnum) 263 { 264 265 counter_u64_add(VNET(tcpstat)[statnum], 1); 266 } 267 268 /* 269 * Wrapper for the TCP established input helper hook. 270 */ 271 static void inline 272 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 273 { 274 struct tcp_hhook_data hhook_data; 275 276 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 277 hhook_data.tp = tp; 278 hhook_data.th = th; 279 hhook_data.to = to; 280 281 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 282 tp->osd); 283 } 284 } 285 286 /* 287 * CC wrapper hook functions 288 */ 289 static void inline 290 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 291 { 292 INP_WLOCK_ASSERT(tp->t_inpcb); 293 294 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 295 if (tp->snd_cwnd <= tp->snd_wnd) 296 tp->ccv->flags |= CCF_CWND_LIMITED; 297 else 298 tp->ccv->flags &= ~CCF_CWND_LIMITED; 299 300 if (type == CC_ACK) { 301 if (tp->snd_cwnd > tp->snd_ssthresh) { 302 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 303 V_tcp_abc_l_var * tp->t_maxseg); 304 if (tp->t_bytes_acked >= tp->snd_cwnd) { 305 tp->t_bytes_acked -= tp->snd_cwnd; 306 tp->ccv->flags |= CCF_ABC_SENTAWND; 307 } 308 } else { 309 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 310 tp->t_bytes_acked = 0; 311 } 312 } 313 314 if (CC_ALGO(tp)->ack_received != NULL) { 315 /* XXXLAS: Find a way to live without this */ 316 tp->ccv->curack = th->th_ack; 317 CC_ALGO(tp)->ack_received(tp->ccv, type); 318 } 319 } 320 321 static void inline 322 cc_conn_init(struct tcpcb *tp) 323 { 324 struct hc_metrics_lite metrics; 325 struct inpcb *inp = tp->t_inpcb; 326 int rtt; 327 328 INP_WLOCK_ASSERT(tp->t_inpcb); 329 330 tcp_hc_get(&inp->inp_inc, &metrics); 331 332 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 333 tp->t_srtt = rtt; 334 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 335 TCPSTAT_INC(tcps_usedrtt); 336 if (metrics.rmx_rttvar) { 337 tp->t_rttvar = metrics.rmx_rttvar; 338 TCPSTAT_INC(tcps_usedrttvar); 339 } else { 340 /* default variation is +- 1 rtt */ 341 tp->t_rttvar = 342 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 343 } 344 TCPT_RANGESET(tp->t_rxtcur, 345 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 346 tp->t_rttmin, TCPTV_REXMTMAX); 347 } 348 if (metrics.rmx_ssthresh) { 349 /* 350 * There's some sort of gateway or interface 351 * buffer limit on the path. Use this to set 352 * the slow start threshhold, but set the 353 * threshold to no less than 2*mss. 354 */ 355 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 356 TCPSTAT_INC(tcps_usedssthresh); 357 } 358 359 /* 360 * Set the initial slow-start flight size. 361 * 362 * RFC5681 Section 3.1 specifies the default conservative values. 363 * RFC3390 specifies slightly more aggressive values. 364 * Draft-ietf-tcpm-initcwnd-05 increases it to ten segments. 365 * 366 * If a SYN or SYN/ACK was lost and retransmitted, we have to 367 * reduce the initial CWND to one segment as congestion is likely 368 * requiring us to be cautious. 369 */ 370 if (tp->snd_cwnd == 1) 371 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 372 else if (V_tcp_do_initcwnd10) 373 tp->snd_cwnd = min(10 * tp->t_maxseg, 374 max(2 * tp->t_maxseg, 14600)); 375 else if (V_tcp_do_rfc3390) 376 tp->snd_cwnd = min(4 * tp->t_maxseg, 377 max(2 * tp->t_maxseg, 4380)); 378 else { 379 /* Per RFC5681 Section 3.1 */ 380 if (tp->t_maxseg > 2190) 381 tp->snd_cwnd = 2 * tp->t_maxseg; 382 else if (tp->t_maxseg > 1095) 383 tp->snd_cwnd = 3 * tp->t_maxseg; 384 else 385 tp->snd_cwnd = 4 * tp->t_maxseg; 386 } 387 388 if (CC_ALGO(tp)->conn_init != NULL) 389 CC_ALGO(tp)->conn_init(tp->ccv); 390 } 391 392 void inline 393 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 394 { 395 INP_WLOCK_ASSERT(tp->t_inpcb); 396 397 switch(type) { 398 case CC_NDUPACK: 399 if (!IN_FASTRECOVERY(tp->t_flags)) { 400 tp->snd_recover = tp->snd_max; 401 if (tp->t_flags & TF_ECN_PERMIT) 402 tp->t_flags |= TF_ECN_SND_CWR; 403 } 404 break; 405 case CC_ECN: 406 if (!IN_CONGRECOVERY(tp->t_flags)) { 407 TCPSTAT_INC(tcps_ecn_rcwnd); 408 tp->snd_recover = tp->snd_max; 409 if (tp->t_flags & TF_ECN_PERMIT) 410 tp->t_flags |= TF_ECN_SND_CWR; 411 } 412 break; 413 case CC_RTO: 414 tp->t_dupacks = 0; 415 tp->t_bytes_acked = 0; 416 EXIT_RECOVERY(tp->t_flags); 417 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 418 tp->t_maxseg) * tp->t_maxseg; 419 tp->snd_cwnd = tp->t_maxseg; 420 break; 421 case CC_RTO_ERR: 422 TCPSTAT_INC(tcps_sndrexmitbad); 423 /* RTO was unnecessary, so reset everything. */ 424 tp->snd_cwnd = tp->snd_cwnd_prev; 425 tp->snd_ssthresh = tp->snd_ssthresh_prev; 426 tp->snd_recover = tp->snd_recover_prev; 427 if (tp->t_flags & TF_WASFRECOVERY) 428 ENTER_FASTRECOVERY(tp->t_flags); 429 if (tp->t_flags & TF_WASCRECOVERY) 430 ENTER_CONGRECOVERY(tp->t_flags); 431 tp->snd_nxt = tp->snd_max; 432 tp->t_flags &= ~TF_PREVVALID; 433 tp->t_badrxtwin = 0; 434 break; 435 } 436 437 if (CC_ALGO(tp)->cong_signal != NULL) { 438 if (th != NULL) 439 tp->ccv->curack = th->th_ack; 440 CC_ALGO(tp)->cong_signal(tp->ccv, type); 441 } 442 } 443 444 static void inline 445 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 446 { 447 INP_WLOCK_ASSERT(tp->t_inpcb); 448 449 /* XXXLAS: KASSERT that we're in recovery? */ 450 451 if (CC_ALGO(tp)->post_recovery != NULL) { 452 tp->ccv->curack = th->th_ack; 453 CC_ALGO(tp)->post_recovery(tp->ccv); 454 } 455 /* XXXLAS: EXIT_RECOVERY ? */ 456 tp->t_bytes_acked = 0; 457 } 458 459 static inline void 460 tcp_fields_to_host(struct tcphdr *th) 461 { 462 463 th->th_seq = ntohl(th->th_seq); 464 th->th_ack = ntohl(th->th_ack); 465 th->th_win = ntohs(th->th_win); 466 th->th_urp = ntohs(th->th_urp); 467 } 468 469 #ifdef TCP_SIGNATURE 470 static inline void 471 tcp_fields_to_net(struct tcphdr *th) 472 { 473 474 th->th_seq = htonl(th->th_seq); 475 th->th_ack = htonl(th->th_ack); 476 th->th_win = htons(th->th_win); 477 th->th_urp = htons(th->th_urp); 478 } 479 480 static inline int 481 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 482 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 483 { 484 int ret; 485 486 tcp_fields_to_net(th); 487 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 488 tcp_fields_to_host(th); 489 return (ret); 490 } 491 #endif 492 493 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 494 #ifdef INET6 495 #define ND6_HINT(tp) \ 496 do { \ 497 if ((tp) && (tp)->t_inpcb && \ 498 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 499 nd6_nud_hint(NULL, NULL, 0); \ 500 } while (0) 501 #else 502 #define ND6_HINT(tp) 503 #endif 504 505 /* 506 * Indicate whether this ack should be delayed. We can delay the ack if 507 * - there is no delayed ack timer in progress and 508 * - our last ack wasn't a 0-sized window. We never want to delay 509 * the ack that opens up a 0-sized window and 510 * - delayed acks are enabled or 511 * - this is a half-synchronized T/TCP connection. 512 * - the segment size is not larger than the MSS and LRO wasn't used 513 * for this segment. 514 */ 515 #define DELAY_ACK(tp, tlen) \ 516 ((!tcp_timer_active(tp, TT_DELACK) && \ 517 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 518 (tlen <= tp->t_maxopd) && \ 519 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 520 521 /* 522 * TCP input handling is split into multiple parts: 523 * tcp6_input is a thin wrapper around tcp_input for the extended 524 * ip6_protox[] call format in ip6_input 525 * tcp_input handles primary segment validation, inpcb lookup and 526 * SYN processing on listen sockets 527 * tcp_do_segment processes the ACK and text of the segment for 528 * establishing, established and closing connections 529 */ 530 #ifdef INET6 531 int 532 tcp6_input(struct mbuf **mp, int *offp, int proto) 533 { 534 struct mbuf *m = *mp; 535 struct in6_ifaddr *ia6; 536 537 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 538 539 /* 540 * draft-itojun-ipv6-tcp-to-anycast 541 * better place to put this in? 542 */ 543 ia6 = ip6_getdstifaddr(m); 544 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 545 struct ip6_hdr *ip6; 546 547 ifa_free(&ia6->ia_ifa); 548 ip6 = mtod(m, struct ip6_hdr *); 549 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 550 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 551 return IPPROTO_DONE; 552 } 553 if (ia6) 554 ifa_free(&ia6->ia_ifa); 555 556 tcp_input(m, *offp); 557 return IPPROTO_DONE; 558 } 559 #endif /* INET6 */ 560 561 void 562 tcp_input(struct mbuf *m, int off0) 563 { 564 struct tcphdr *th = NULL; 565 struct ip *ip = NULL; 566 struct inpcb *inp = NULL; 567 struct tcpcb *tp = NULL; 568 struct socket *so = NULL; 569 u_char *optp = NULL; 570 int optlen = 0; 571 #ifdef INET 572 int len; 573 #endif 574 int tlen = 0, off; 575 int drop_hdrlen; 576 int thflags; 577 int rstreason = 0; /* For badport_bandlim accounting purposes */ 578 #ifdef TCP_SIGNATURE 579 uint8_t sig_checked = 0; 580 #endif 581 uint8_t iptos = 0; 582 struct m_tag *fwd_tag = NULL; 583 #ifdef INET6 584 struct ip6_hdr *ip6 = NULL; 585 int isipv6; 586 #else 587 const void *ip6 = NULL; 588 #endif /* INET6 */ 589 struct tcpopt to; /* options in this segment */ 590 char *s = NULL; /* address and port logging */ 591 int ti_locked; 592 #define TI_UNLOCKED 1 593 #define TI_WLOCKED 2 594 595 #ifdef TCPDEBUG 596 /* 597 * The size of tcp_saveipgen must be the size of the max ip header, 598 * now IPv6. 599 */ 600 u_char tcp_saveipgen[IP6_HDR_LEN]; 601 struct tcphdr tcp_savetcp; 602 short ostate = 0; 603 #endif 604 605 #ifdef INET6 606 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 607 #endif 608 609 to.to_flags = 0; 610 TCPSTAT_INC(tcps_rcvtotal); 611 612 #ifdef INET6 613 if (isipv6) { 614 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 615 616 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 617 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 618 if (m == NULL) { 619 TCPSTAT_INC(tcps_rcvshort); 620 return; 621 } 622 } 623 624 ip6 = mtod(m, struct ip6_hdr *); 625 th = (struct tcphdr *)((caddr_t)ip6 + off0); 626 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 627 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 628 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 629 th->th_sum = m->m_pkthdr.csum_data; 630 else 631 th->th_sum = in6_cksum_pseudo(ip6, tlen, 632 IPPROTO_TCP, m->m_pkthdr.csum_data); 633 th->th_sum ^= 0xffff; 634 } else 635 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 636 if (th->th_sum) { 637 TCPSTAT_INC(tcps_rcvbadsum); 638 goto drop; 639 } 640 641 /* 642 * Be proactive about unspecified IPv6 address in source. 643 * As we use all-zero to indicate unbounded/unconnected pcb, 644 * unspecified IPv6 address can be used to confuse us. 645 * 646 * Note that packets with unspecified IPv6 destination is 647 * already dropped in ip6_input. 648 */ 649 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 650 /* XXX stat */ 651 goto drop; 652 } 653 } 654 #endif 655 #if defined(INET) && defined(INET6) 656 else 657 #endif 658 #ifdef INET 659 { 660 /* 661 * Get IP and TCP header together in first mbuf. 662 * Note: IP leaves IP header in first mbuf. 663 */ 664 if (off0 > sizeof (struct ip)) { 665 ip_stripoptions(m); 666 off0 = sizeof(struct ip); 667 } 668 if (m->m_len < sizeof (struct tcpiphdr)) { 669 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 670 == NULL) { 671 TCPSTAT_INC(tcps_rcvshort); 672 return; 673 } 674 } 675 ip = mtod(m, struct ip *); 676 th = (struct tcphdr *)((caddr_t)ip + off0); 677 tlen = ntohs(ip->ip_len) - off0; 678 679 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 680 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 681 th->th_sum = m->m_pkthdr.csum_data; 682 else 683 th->th_sum = in_pseudo(ip->ip_src.s_addr, 684 ip->ip_dst.s_addr, 685 htonl(m->m_pkthdr.csum_data + tlen + 686 IPPROTO_TCP)); 687 th->th_sum ^= 0xffff; 688 } else { 689 struct ipovly *ipov = (struct ipovly *)ip; 690 691 /* 692 * Checksum extended TCP header and data. 693 */ 694 len = off0 + tlen; 695 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 696 ipov->ih_len = htons(tlen); 697 th->th_sum = in_cksum(m, len); 698 /* Reset length for SDT probes. */ 699 ip->ip_len = htons(tlen + off0); 700 } 701 702 if (th->th_sum) { 703 TCPSTAT_INC(tcps_rcvbadsum); 704 goto drop; 705 } 706 /* Re-initialization for later version check */ 707 ip->ip_v = IPVERSION; 708 } 709 #endif /* INET */ 710 711 #ifdef INET6 712 if (isipv6) 713 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 714 #endif 715 #if defined(INET) && defined(INET6) 716 else 717 #endif 718 #ifdef INET 719 iptos = ip->ip_tos; 720 #endif 721 722 /* 723 * Check that TCP offset makes sense, 724 * pull out TCP options and adjust length. XXX 725 */ 726 off = th->th_off << 2; 727 if (off < sizeof (struct tcphdr) || off > tlen) { 728 TCPSTAT_INC(tcps_rcvbadoff); 729 goto drop; 730 } 731 tlen -= off; /* tlen is used instead of ti->ti_len */ 732 if (off > sizeof (struct tcphdr)) { 733 #ifdef INET6 734 if (isipv6) { 735 IP6_EXTHDR_CHECK(m, off0, off, ); 736 ip6 = mtod(m, struct ip6_hdr *); 737 th = (struct tcphdr *)((caddr_t)ip6 + off0); 738 } 739 #endif 740 #if defined(INET) && defined(INET6) 741 else 742 #endif 743 #ifdef INET 744 { 745 if (m->m_len < sizeof(struct ip) + off) { 746 if ((m = m_pullup(m, sizeof (struct ip) + off)) 747 == NULL) { 748 TCPSTAT_INC(tcps_rcvshort); 749 return; 750 } 751 ip = mtod(m, struct ip *); 752 th = (struct tcphdr *)((caddr_t)ip + off0); 753 } 754 } 755 #endif 756 optlen = off - sizeof (struct tcphdr); 757 optp = (u_char *)(th + 1); 758 } 759 thflags = th->th_flags; 760 761 /* 762 * Convert TCP protocol specific fields to host format. 763 */ 764 tcp_fields_to_host(th); 765 766 /* 767 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 768 */ 769 drop_hdrlen = off0 + off; 770 771 /* 772 * Locate pcb for segment; if we're likely to add or remove a 773 * connection then first acquire pcbinfo lock. There are two cases 774 * where we might discover later we need a write lock despite the 775 * flags: ACKs moving a connection out of the syncache, and ACKs for 776 * a connection in TIMEWAIT. 777 */ 778 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 779 INP_INFO_WLOCK(&V_tcbinfo); 780 ti_locked = TI_WLOCKED; 781 } else 782 ti_locked = TI_UNLOCKED; 783 784 /* 785 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 786 */ 787 if ( 788 #ifdef INET6 789 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 790 #ifdef INET 791 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 792 #endif 793 #endif 794 #if defined(INET) && !defined(INET6) 795 (m->m_flags & M_IP_NEXTHOP) 796 #endif 797 ) 798 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 799 800 findpcb: 801 #ifdef INVARIANTS 802 if (ti_locked == TI_WLOCKED) { 803 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 804 } else { 805 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 806 } 807 #endif 808 #ifdef INET6 809 if (isipv6 && fwd_tag != NULL) { 810 struct sockaddr_in6 *next_hop6; 811 812 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 813 /* 814 * Transparently forwarded. Pretend to be the destination. 815 * Already got one like this? 816 */ 817 inp = in6_pcblookup_mbuf(&V_tcbinfo, 818 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 819 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 820 if (!inp) { 821 /* 822 * It's new. Try to find the ambushing socket. 823 * Because we've rewritten the destination address, 824 * any hardware-generated hash is ignored. 825 */ 826 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 827 th->th_sport, &next_hop6->sin6_addr, 828 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 829 th->th_dport, INPLOOKUP_WILDCARD | 830 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 831 } 832 } else if (isipv6) { 833 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 834 th->th_sport, &ip6->ip6_dst, th->th_dport, 835 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 836 m->m_pkthdr.rcvif, m); 837 } 838 #endif /* INET6 */ 839 #if defined(INET6) && defined(INET) 840 else 841 #endif 842 #ifdef INET 843 if (fwd_tag != NULL) { 844 struct sockaddr_in *next_hop; 845 846 next_hop = (struct sockaddr_in *)(fwd_tag+1); 847 /* 848 * Transparently forwarded. Pretend to be the destination. 849 * already got one like this? 850 */ 851 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 852 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 853 m->m_pkthdr.rcvif, m); 854 if (!inp) { 855 /* 856 * It's new. Try to find the ambushing socket. 857 * Because we've rewritten the destination address, 858 * any hardware-generated hash is ignored. 859 */ 860 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 861 th->th_sport, next_hop->sin_addr, 862 next_hop->sin_port ? ntohs(next_hop->sin_port) : 863 th->th_dport, INPLOOKUP_WILDCARD | 864 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 865 } 866 } else 867 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 868 th->th_sport, ip->ip_dst, th->th_dport, 869 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 870 m->m_pkthdr.rcvif, m); 871 #endif /* INET */ 872 873 /* 874 * If the INPCB does not exist then all data in the incoming 875 * segment is discarded and an appropriate RST is sent back. 876 * XXX MRT Send RST using which routing table? 877 */ 878 if (inp == NULL) { 879 /* 880 * Log communication attempts to ports that are not 881 * in use. 882 */ 883 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 884 tcp_log_in_vain == 2) { 885 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 886 log(LOG_INFO, "%s; %s: Connection attempt " 887 "to closed port\n", s, __func__); 888 } 889 /* 890 * When blackholing do not respond with a RST but 891 * completely ignore the segment and drop it. 892 */ 893 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 894 V_blackhole == 2) 895 goto dropunlock; 896 897 rstreason = BANDLIM_RST_CLOSEDPORT; 898 goto dropwithreset; 899 } 900 INP_WLOCK_ASSERT(inp); 901 if (!(inp->inp_flags & INP_HW_FLOWID) 902 && (m->m_flags & M_FLOWID) 903 && ((inp->inp_socket == NULL) 904 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 905 inp->inp_flags |= INP_HW_FLOWID; 906 inp->inp_flags &= ~INP_SW_FLOWID; 907 inp->inp_flowid = m->m_pkthdr.flowid; 908 } 909 #ifdef IPSEC 910 #ifdef INET6 911 if (isipv6 && ipsec6_in_reject(m, inp)) { 912 IPSEC6STAT_INC(ips_in_polvio); 913 goto dropunlock; 914 } else 915 #endif /* INET6 */ 916 if (ipsec4_in_reject(m, inp) != 0) { 917 IPSECSTAT_INC(ips_in_polvio); 918 goto dropunlock; 919 } 920 #endif /* IPSEC */ 921 922 /* 923 * Check the minimum TTL for socket. 924 */ 925 if (inp->inp_ip_minttl != 0) { 926 #ifdef INET6 927 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 928 goto dropunlock; 929 else 930 #endif 931 if (inp->inp_ip_minttl > ip->ip_ttl) 932 goto dropunlock; 933 } 934 935 /* 936 * A previous connection in TIMEWAIT state is supposed to catch stray 937 * or duplicate segments arriving late. If this segment was a 938 * legitimate new connection attempt, the old INPCB gets removed and 939 * we can try again to find a listening socket. 940 * 941 * At this point, due to earlier optimism, we may hold only an inpcb 942 * lock, and not the inpcbinfo write lock. If so, we need to try to 943 * acquire it, or if that fails, acquire a reference on the inpcb, 944 * drop all locks, acquire a global write lock, and then re-acquire 945 * the inpcb lock. We may at that point discover that another thread 946 * has tried to free the inpcb, in which case we need to loop back 947 * and try to find a new inpcb to deliver to. 948 * 949 * XXXRW: It may be time to rethink timewait locking. 950 */ 951 relocked: 952 if (inp->inp_flags & INP_TIMEWAIT) { 953 if (ti_locked == TI_UNLOCKED) { 954 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 955 in_pcbref(inp); 956 INP_WUNLOCK(inp); 957 INP_INFO_WLOCK(&V_tcbinfo); 958 ti_locked = TI_WLOCKED; 959 INP_WLOCK(inp); 960 if (in_pcbrele_wlocked(inp)) { 961 inp = NULL; 962 goto findpcb; 963 } 964 } else 965 ti_locked = TI_WLOCKED; 966 } 967 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 968 969 if (thflags & TH_SYN) 970 tcp_dooptions(&to, optp, optlen, TO_SYN); 971 /* 972 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 973 */ 974 if (tcp_twcheck(inp, &to, th, m, tlen)) 975 goto findpcb; 976 INP_INFO_WUNLOCK(&V_tcbinfo); 977 return; 978 } 979 /* 980 * The TCPCB may no longer exist if the connection is winding 981 * down or it is in the CLOSED state. Either way we drop the 982 * segment and send an appropriate response. 983 */ 984 tp = intotcpcb(inp); 985 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 986 rstreason = BANDLIM_RST_CLOSEDPORT; 987 goto dropwithreset; 988 } 989 990 #ifdef TCP_OFFLOAD 991 if (tp->t_flags & TF_TOE) { 992 tcp_offload_input(tp, m); 993 m = NULL; /* consumed by the TOE driver */ 994 goto dropunlock; 995 } 996 #endif 997 998 /* 999 * We've identified a valid inpcb, but it could be that we need an 1000 * inpcbinfo write lock but don't hold it. In this case, attempt to 1001 * acquire using the same strategy as the TIMEWAIT case above. If we 1002 * relock, we have to jump back to 'relocked' as the connection might 1003 * now be in TIMEWAIT. 1004 */ 1005 #ifdef INVARIANTS 1006 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 1007 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1008 #endif 1009 if (tp->t_state != TCPS_ESTABLISHED) { 1010 if (ti_locked == TI_UNLOCKED) { 1011 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 1012 in_pcbref(inp); 1013 INP_WUNLOCK(inp); 1014 INP_INFO_WLOCK(&V_tcbinfo); 1015 ti_locked = TI_WLOCKED; 1016 INP_WLOCK(inp); 1017 if (in_pcbrele_wlocked(inp)) { 1018 inp = NULL; 1019 goto findpcb; 1020 } 1021 goto relocked; 1022 } else 1023 ti_locked = TI_WLOCKED; 1024 } 1025 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1026 } 1027 1028 #ifdef MAC 1029 INP_WLOCK_ASSERT(inp); 1030 if (mac_inpcb_check_deliver(inp, m)) 1031 goto dropunlock; 1032 #endif 1033 so = inp->inp_socket; 1034 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1035 #ifdef TCPDEBUG 1036 if (so->so_options & SO_DEBUG) { 1037 ostate = tp->t_state; 1038 #ifdef INET6 1039 if (isipv6) { 1040 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1041 } else 1042 #endif 1043 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1044 tcp_savetcp = *th; 1045 } 1046 #endif /* TCPDEBUG */ 1047 /* 1048 * When the socket is accepting connections (the INPCB is in LISTEN 1049 * state) we look into the SYN cache if this is a new connection 1050 * attempt or the completion of a previous one. Because listen 1051 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1052 * held in this case. 1053 */ 1054 if (so->so_options & SO_ACCEPTCONN) { 1055 struct in_conninfo inc; 1056 1057 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1058 "tp not listening", __func__)); 1059 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1060 1061 bzero(&inc, sizeof(inc)); 1062 #ifdef INET6 1063 if (isipv6) { 1064 inc.inc_flags |= INC_ISIPV6; 1065 inc.inc6_faddr = ip6->ip6_src; 1066 inc.inc6_laddr = ip6->ip6_dst; 1067 } else 1068 #endif 1069 { 1070 inc.inc_faddr = ip->ip_src; 1071 inc.inc_laddr = ip->ip_dst; 1072 } 1073 inc.inc_fport = th->th_sport; 1074 inc.inc_lport = th->th_dport; 1075 inc.inc_fibnum = so->so_fibnum; 1076 1077 /* 1078 * Check for an existing connection attempt in syncache if 1079 * the flag is only ACK. A successful lookup creates a new 1080 * socket appended to the listen queue in SYN_RECEIVED state. 1081 */ 1082 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1083 /* 1084 * Parse the TCP options here because 1085 * syncookies need access to the reflected 1086 * timestamp. 1087 */ 1088 tcp_dooptions(&to, optp, optlen, 0); 1089 /* 1090 * NB: syncache_expand() doesn't unlock 1091 * inp and tcpinfo locks. 1092 */ 1093 if (!syncache_expand(&inc, &to, th, &so, m)) { 1094 /* 1095 * No syncache entry or ACK was not 1096 * for our SYN/ACK. Send a RST. 1097 * NB: syncache did its own logging 1098 * of the failure cause. 1099 */ 1100 rstreason = BANDLIM_RST_OPENPORT; 1101 goto dropwithreset; 1102 } 1103 if (so == NULL) { 1104 /* 1105 * We completed the 3-way handshake 1106 * but could not allocate a socket 1107 * either due to memory shortage, 1108 * listen queue length limits or 1109 * global socket limits. Send RST 1110 * or wait and have the remote end 1111 * retransmit the ACK for another 1112 * try. 1113 */ 1114 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1115 log(LOG_DEBUG, "%s; %s: Listen socket: " 1116 "Socket allocation failed due to " 1117 "limits or memory shortage, %s\n", 1118 s, __func__, 1119 V_tcp_sc_rst_sock_fail ? 1120 "sending RST" : "try again"); 1121 if (V_tcp_sc_rst_sock_fail) { 1122 rstreason = BANDLIM_UNLIMITED; 1123 goto dropwithreset; 1124 } else 1125 goto dropunlock; 1126 } 1127 /* 1128 * Socket is created in state SYN_RECEIVED. 1129 * Unlock the listen socket, lock the newly 1130 * created socket and update the tp variable. 1131 */ 1132 INP_WUNLOCK(inp); /* listen socket */ 1133 inp = sotoinpcb(so); 1134 INP_WLOCK(inp); /* new connection */ 1135 tp = intotcpcb(inp); 1136 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1137 ("%s: ", __func__)); 1138 #ifdef TCP_SIGNATURE 1139 if (sig_checked == 0) { 1140 tcp_dooptions(&to, optp, optlen, 1141 (thflags & TH_SYN) ? TO_SYN : 0); 1142 if (!tcp_signature_verify_input(m, off0, tlen, 1143 optlen, &to, th, tp->t_flags)) { 1144 1145 /* 1146 * In SYN_SENT state if it receives an 1147 * RST, it is allowed for further 1148 * processing. 1149 */ 1150 if ((thflags & TH_RST) == 0 || 1151 (tp->t_state == TCPS_SYN_SENT) == 0) 1152 goto dropunlock; 1153 } 1154 sig_checked = 1; 1155 } 1156 #endif 1157 1158 /* 1159 * Process the segment and the data it 1160 * contains. tcp_do_segment() consumes 1161 * the mbuf chain and unlocks the inpcb. 1162 */ 1163 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1164 iptos, ti_locked); 1165 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1166 return; 1167 } 1168 /* 1169 * Segment flag validation for new connection attempts: 1170 * 1171 * Our (SYN|ACK) response was rejected. 1172 * Check with syncache and remove entry to prevent 1173 * retransmits. 1174 * 1175 * NB: syncache_chkrst does its own logging of failure 1176 * causes. 1177 */ 1178 if (thflags & TH_RST) { 1179 syncache_chkrst(&inc, th); 1180 goto dropunlock; 1181 } 1182 /* 1183 * We can't do anything without SYN. 1184 */ 1185 if ((thflags & TH_SYN) == 0) { 1186 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1187 log(LOG_DEBUG, "%s; %s: Listen socket: " 1188 "SYN is missing, segment ignored\n", 1189 s, __func__); 1190 TCPSTAT_INC(tcps_badsyn); 1191 goto dropunlock; 1192 } 1193 /* 1194 * (SYN|ACK) is bogus on a listen socket. 1195 */ 1196 if (thflags & TH_ACK) { 1197 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1198 log(LOG_DEBUG, "%s; %s: Listen socket: " 1199 "SYN|ACK invalid, segment rejected\n", 1200 s, __func__); 1201 syncache_badack(&inc); /* XXX: Not needed! */ 1202 TCPSTAT_INC(tcps_badsyn); 1203 rstreason = BANDLIM_RST_OPENPORT; 1204 goto dropwithreset; 1205 } 1206 /* 1207 * If the drop_synfin option is enabled, drop all 1208 * segments with both the SYN and FIN bits set. 1209 * This prevents e.g. nmap from identifying the 1210 * TCP/IP stack. 1211 * XXX: Poor reasoning. nmap has other methods 1212 * and is constantly refining its stack detection 1213 * strategies. 1214 * XXX: This is a violation of the TCP specification 1215 * and was used by RFC1644. 1216 */ 1217 if ((thflags & TH_FIN) && V_drop_synfin) { 1218 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1219 log(LOG_DEBUG, "%s; %s: Listen socket: " 1220 "SYN|FIN segment ignored (based on " 1221 "sysctl setting)\n", s, __func__); 1222 TCPSTAT_INC(tcps_badsyn); 1223 goto dropunlock; 1224 } 1225 /* 1226 * Segment's flags are (SYN) or (SYN|FIN). 1227 * 1228 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1229 * as they do not affect the state of the TCP FSM. 1230 * The data pointed to by TH_URG and th_urp is ignored. 1231 */ 1232 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1233 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1234 KASSERT(thflags & (TH_SYN), 1235 ("%s: Listen socket: TH_SYN not set", __func__)); 1236 #ifdef INET6 1237 /* 1238 * If deprecated address is forbidden, 1239 * we do not accept SYN to deprecated interface 1240 * address to prevent any new inbound connection from 1241 * getting established. 1242 * When we do not accept SYN, we send a TCP RST, 1243 * with deprecated source address (instead of dropping 1244 * it). We compromise it as it is much better for peer 1245 * to send a RST, and RST will be the final packet 1246 * for the exchange. 1247 * 1248 * If we do not forbid deprecated addresses, we accept 1249 * the SYN packet. RFC2462 does not suggest dropping 1250 * SYN in this case. 1251 * If we decipher RFC2462 5.5.4, it says like this: 1252 * 1. use of deprecated addr with existing 1253 * communication is okay - "SHOULD continue to be 1254 * used" 1255 * 2. use of it with new communication: 1256 * (2a) "SHOULD NOT be used if alternate address 1257 * with sufficient scope is available" 1258 * (2b) nothing mentioned otherwise. 1259 * Here we fall into (2b) case as we have no choice in 1260 * our source address selection - we must obey the peer. 1261 * 1262 * The wording in RFC2462 is confusing, and there are 1263 * multiple description text for deprecated address 1264 * handling - worse, they are not exactly the same. 1265 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1266 */ 1267 if (isipv6 && !V_ip6_use_deprecated) { 1268 struct in6_ifaddr *ia6; 1269 1270 ia6 = ip6_getdstifaddr(m); 1271 if (ia6 != NULL && 1272 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1273 ifa_free(&ia6->ia_ifa); 1274 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1275 log(LOG_DEBUG, "%s; %s: Listen socket: " 1276 "Connection attempt to deprecated " 1277 "IPv6 address rejected\n", 1278 s, __func__); 1279 rstreason = BANDLIM_RST_OPENPORT; 1280 goto dropwithreset; 1281 } 1282 if (ia6) 1283 ifa_free(&ia6->ia_ifa); 1284 } 1285 #endif /* INET6 */ 1286 /* 1287 * Basic sanity checks on incoming SYN requests: 1288 * Don't respond if the destination is a link layer 1289 * broadcast according to RFC1122 4.2.3.10, p. 104. 1290 * If it is from this socket it must be forged. 1291 * Don't respond if the source or destination is a 1292 * global or subnet broad- or multicast address. 1293 * Note that it is quite possible to receive unicast 1294 * link-layer packets with a broadcast IP address. Use 1295 * in_broadcast() to find them. 1296 */ 1297 if (m->m_flags & (M_BCAST|M_MCAST)) { 1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1299 log(LOG_DEBUG, "%s; %s: Listen socket: " 1300 "Connection attempt from broad- or multicast " 1301 "link layer address ignored\n", s, __func__); 1302 goto dropunlock; 1303 } 1304 #ifdef INET6 1305 if (isipv6) { 1306 if (th->th_dport == th->th_sport && 1307 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1309 log(LOG_DEBUG, "%s; %s: Listen socket: " 1310 "Connection attempt to/from self " 1311 "ignored\n", s, __func__); 1312 goto dropunlock; 1313 } 1314 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1315 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1316 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1317 log(LOG_DEBUG, "%s; %s: Listen socket: " 1318 "Connection attempt from/to multicast " 1319 "address ignored\n", s, __func__); 1320 goto dropunlock; 1321 } 1322 } 1323 #endif 1324 #if defined(INET) && defined(INET6) 1325 else 1326 #endif 1327 #ifdef INET 1328 { 1329 if (th->th_dport == th->th_sport && 1330 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1331 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1332 log(LOG_DEBUG, "%s; %s: Listen socket: " 1333 "Connection attempt from/to self " 1334 "ignored\n", s, __func__); 1335 goto dropunlock; 1336 } 1337 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1338 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1339 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1340 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1341 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1342 log(LOG_DEBUG, "%s; %s: Listen socket: " 1343 "Connection attempt from/to broad- " 1344 "or multicast address ignored\n", 1345 s, __func__); 1346 goto dropunlock; 1347 } 1348 } 1349 #endif 1350 /* 1351 * SYN appears to be valid. Create compressed TCP state 1352 * for syncache. 1353 */ 1354 #ifdef TCPDEBUG 1355 if (so->so_options & SO_DEBUG) 1356 tcp_trace(TA_INPUT, ostate, tp, 1357 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1358 #endif 1359 tcp_dooptions(&to, optp, optlen, TO_SYN); 1360 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1361 /* 1362 * Entry added to syncache and mbuf consumed. 1363 * Everything already unlocked by syncache_add(). 1364 */ 1365 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1366 return; 1367 } else if (tp->t_state == TCPS_LISTEN) { 1368 /* 1369 * When a listen socket is torn down the SO_ACCEPTCONN 1370 * flag is removed first while connections are drained 1371 * from the accept queue in a unlock/lock cycle of the 1372 * ACCEPT_LOCK, opening a race condition allowing a SYN 1373 * attempt go through unhandled. 1374 */ 1375 goto dropunlock; 1376 } 1377 1378 #ifdef TCP_SIGNATURE 1379 if (sig_checked == 0) { 1380 tcp_dooptions(&to, optp, optlen, 1381 (thflags & TH_SYN) ? TO_SYN : 0); 1382 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1383 th, tp->t_flags)) { 1384 1385 /* 1386 * In SYN_SENT state if it receives an RST, it is 1387 * allowed for further processing. 1388 */ 1389 if ((thflags & TH_RST) == 0 || 1390 (tp->t_state == TCPS_SYN_SENT) == 0) 1391 goto dropunlock; 1392 } 1393 sig_checked = 1; 1394 } 1395 #endif 1396 1397 TCP_PROBE5(receive, NULL, tp, m->m_data, tp, th); 1398 1399 /* 1400 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1401 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1402 * the inpcb, and unlocks pcbinfo. 1403 */ 1404 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1405 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1406 return; 1407 1408 dropwithreset: 1409 TCP_PROBE5(receive, NULL, tp, m->m_data, tp, th); 1410 1411 if (ti_locked == TI_WLOCKED) { 1412 INP_INFO_WUNLOCK(&V_tcbinfo); 1413 ti_locked = TI_UNLOCKED; 1414 } 1415 #ifdef INVARIANTS 1416 else { 1417 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1418 "ti_locked: %d", __func__, ti_locked)); 1419 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1420 } 1421 #endif 1422 1423 if (inp != NULL) { 1424 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1425 INP_WUNLOCK(inp); 1426 } else 1427 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1428 m = NULL; /* mbuf chain got consumed. */ 1429 goto drop; 1430 1431 dropunlock: 1432 if (m != NULL) 1433 TCP_PROBE5(receive, NULL, tp, m->m_data, tp, th); 1434 1435 if (ti_locked == TI_WLOCKED) { 1436 INP_INFO_WUNLOCK(&V_tcbinfo); 1437 ti_locked = TI_UNLOCKED; 1438 } 1439 #ifdef INVARIANTS 1440 else { 1441 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1442 "ti_locked: %d", __func__, ti_locked)); 1443 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1444 } 1445 #endif 1446 1447 if (inp != NULL) 1448 INP_WUNLOCK(inp); 1449 1450 drop: 1451 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1452 if (s != NULL) 1453 free(s, M_TCPLOG); 1454 if (m != NULL) 1455 m_freem(m); 1456 } 1457 1458 static void 1459 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1460 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1461 int ti_locked) 1462 { 1463 int thflags, acked, ourfinisacked, needoutput = 0; 1464 int rstreason, todrop, win; 1465 u_long tiwin; 1466 char *s; 1467 struct in_conninfo *inc; 1468 struct mbuf *mfree; 1469 struct tcpopt to; 1470 1471 #ifdef TCPDEBUG 1472 /* 1473 * The size of tcp_saveipgen must be the size of the max ip header, 1474 * now IPv6. 1475 */ 1476 u_char tcp_saveipgen[IP6_HDR_LEN]; 1477 struct tcphdr tcp_savetcp; 1478 short ostate = 0; 1479 #endif 1480 thflags = th->th_flags; 1481 inc = &tp->t_inpcb->inp_inc; 1482 tp->sackhint.last_sack_ack = 0; 1483 1484 /* 1485 * If this is either a state-changing packet or current state isn't 1486 * established, we require a write lock on tcbinfo. Otherwise, we 1487 * allow the tcbinfo to be in either alocked or unlocked, as the 1488 * caller may have unnecessarily acquired a write lock due to a race. 1489 */ 1490 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1491 tp->t_state != TCPS_ESTABLISHED) { 1492 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1493 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1494 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1495 } else { 1496 #ifdef INVARIANTS 1497 if (ti_locked == TI_WLOCKED) 1498 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1499 else { 1500 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1501 "ti_locked: %d", __func__, ti_locked)); 1502 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1503 } 1504 #endif 1505 } 1506 INP_WLOCK_ASSERT(tp->t_inpcb); 1507 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1508 __func__)); 1509 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1510 __func__)); 1511 1512 /* 1513 * Segment received on connection. 1514 * Reset idle time and keep-alive timer. 1515 * XXX: This should be done after segment 1516 * validation to ignore broken/spoofed segs. 1517 */ 1518 tp->t_rcvtime = ticks; 1519 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1520 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1521 1522 /* 1523 * Unscale the window into a 32-bit value. 1524 * For the SYN_SENT state the scale is zero. 1525 */ 1526 tiwin = th->th_win << tp->snd_scale; 1527 1528 /* 1529 * TCP ECN processing. 1530 */ 1531 if (tp->t_flags & TF_ECN_PERMIT) { 1532 if (thflags & TH_CWR) 1533 tp->t_flags &= ~TF_ECN_SND_ECE; 1534 switch (iptos & IPTOS_ECN_MASK) { 1535 case IPTOS_ECN_CE: 1536 tp->t_flags |= TF_ECN_SND_ECE; 1537 TCPSTAT_INC(tcps_ecn_ce); 1538 break; 1539 case IPTOS_ECN_ECT0: 1540 TCPSTAT_INC(tcps_ecn_ect0); 1541 break; 1542 case IPTOS_ECN_ECT1: 1543 TCPSTAT_INC(tcps_ecn_ect1); 1544 break; 1545 } 1546 /* Congestion experienced. */ 1547 if (thflags & TH_ECE) { 1548 cc_cong_signal(tp, th, CC_ECN); 1549 } 1550 } 1551 1552 /* 1553 * Parse options on any incoming segment. 1554 */ 1555 tcp_dooptions(&to, (u_char *)(th + 1), 1556 (th->th_off << 2) - sizeof(struct tcphdr), 1557 (thflags & TH_SYN) ? TO_SYN : 0); 1558 1559 /* 1560 * If echoed timestamp is later than the current time, 1561 * fall back to non RFC1323 RTT calculation. Normalize 1562 * timestamp if syncookies were used when this connection 1563 * was established. 1564 */ 1565 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1566 to.to_tsecr -= tp->ts_offset; 1567 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1568 to.to_tsecr = 0; 1569 } 1570 /* 1571 * If timestamps were negotiated during SYN/ACK they should 1572 * appear on every segment during this session and vice versa. 1573 */ 1574 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1575 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1576 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1577 "no action\n", s, __func__); 1578 free(s, M_TCPLOG); 1579 } 1580 } 1581 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1582 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1583 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1584 "no action\n", s, __func__); 1585 free(s, M_TCPLOG); 1586 } 1587 } 1588 1589 /* 1590 * Process options only when we get SYN/ACK back. The SYN case 1591 * for incoming connections is handled in tcp_syncache. 1592 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1593 * or <SYN,ACK>) segment itself is never scaled. 1594 * XXX this is traditional behavior, may need to be cleaned up. 1595 */ 1596 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1597 if ((to.to_flags & TOF_SCALE) && 1598 (tp->t_flags & TF_REQ_SCALE)) { 1599 tp->t_flags |= TF_RCVD_SCALE; 1600 tp->snd_scale = to.to_wscale; 1601 } 1602 /* 1603 * Initial send window. It will be updated with 1604 * the next incoming segment to the scaled value. 1605 */ 1606 tp->snd_wnd = th->th_win; 1607 if (to.to_flags & TOF_TS) { 1608 tp->t_flags |= TF_RCVD_TSTMP; 1609 tp->ts_recent = to.to_tsval; 1610 tp->ts_recent_age = tcp_ts_getticks(); 1611 } 1612 if (to.to_flags & TOF_MSS) 1613 tcp_mss(tp, to.to_mss); 1614 if ((tp->t_flags & TF_SACK_PERMIT) && 1615 (to.to_flags & TOF_SACKPERM) == 0) 1616 tp->t_flags &= ~TF_SACK_PERMIT; 1617 } 1618 1619 /* 1620 * Header prediction: check for the two common cases 1621 * of a uni-directional data xfer. If the packet has 1622 * no control flags, is in-sequence, the window didn't 1623 * change and we're not retransmitting, it's a 1624 * candidate. If the length is zero and the ack moved 1625 * forward, we're the sender side of the xfer. Just 1626 * free the data acked & wake any higher level process 1627 * that was blocked waiting for space. If the length 1628 * is non-zero and the ack didn't move, we're the 1629 * receiver side. If we're getting packets in-order 1630 * (the reassembly queue is empty), add the data to 1631 * the socket buffer and note that we need a delayed ack. 1632 * Make sure that the hidden state-flags are also off. 1633 * Since we check for TCPS_ESTABLISHED first, it can only 1634 * be TH_NEEDSYN. 1635 */ 1636 if (tp->t_state == TCPS_ESTABLISHED && 1637 th->th_seq == tp->rcv_nxt && 1638 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1639 tp->snd_nxt == tp->snd_max && 1640 tiwin && tiwin == tp->snd_wnd && 1641 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1642 LIST_EMPTY(&tp->t_segq) && 1643 ((to.to_flags & TOF_TS) == 0 || 1644 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1645 1646 /* 1647 * If last ACK falls within this segment's sequence numbers, 1648 * record the timestamp. 1649 * NOTE that the test is modified according to the latest 1650 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1651 */ 1652 if ((to.to_flags & TOF_TS) != 0 && 1653 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1654 tp->ts_recent_age = tcp_ts_getticks(); 1655 tp->ts_recent = to.to_tsval; 1656 } 1657 1658 if (tlen == 0) { 1659 if (SEQ_GT(th->th_ack, tp->snd_una) && 1660 SEQ_LEQ(th->th_ack, tp->snd_max) && 1661 !IN_RECOVERY(tp->t_flags) && 1662 (to.to_flags & TOF_SACK) == 0 && 1663 TAILQ_EMPTY(&tp->snd_holes)) { 1664 /* 1665 * This is a pure ack for outstanding data. 1666 */ 1667 if (ti_locked == TI_WLOCKED) 1668 INP_INFO_WUNLOCK(&V_tcbinfo); 1669 ti_locked = TI_UNLOCKED; 1670 1671 TCPSTAT_INC(tcps_predack); 1672 1673 /* 1674 * "bad retransmit" recovery. 1675 */ 1676 if (tp->t_rxtshift == 1 && 1677 tp->t_flags & TF_PREVVALID && 1678 (int)(ticks - tp->t_badrxtwin) < 0) { 1679 cc_cong_signal(tp, th, CC_RTO_ERR); 1680 } 1681 1682 /* 1683 * Recalculate the transmit timer / rtt. 1684 * 1685 * Some boxes send broken timestamp replies 1686 * during the SYN+ACK phase, ignore 1687 * timestamps of 0 or we could calculate a 1688 * huge RTT and blow up the retransmit timer. 1689 */ 1690 if ((to.to_flags & TOF_TS) != 0 && 1691 to.to_tsecr) { 1692 u_int t; 1693 1694 t = tcp_ts_getticks() - to.to_tsecr; 1695 if (!tp->t_rttlow || tp->t_rttlow > t) 1696 tp->t_rttlow = t; 1697 tcp_xmit_timer(tp, 1698 TCP_TS_TO_TICKS(t) + 1); 1699 } else if (tp->t_rtttime && 1700 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1701 if (!tp->t_rttlow || 1702 tp->t_rttlow > ticks - tp->t_rtttime) 1703 tp->t_rttlow = ticks - tp->t_rtttime; 1704 tcp_xmit_timer(tp, 1705 ticks - tp->t_rtttime); 1706 } 1707 acked = BYTES_THIS_ACK(tp, th); 1708 1709 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1710 hhook_run_tcp_est_in(tp, th, &to); 1711 1712 TCPSTAT_INC(tcps_rcvackpack); 1713 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1714 sbdrop(&so->so_snd, acked); 1715 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1716 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1717 tp->snd_recover = th->th_ack - 1; 1718 1719 /* 1720 * Let the congestion control algorithm update 1721 * congestion control related information. This 1722 * typically means increasing the congestion 1723 * window. 1724 */ 1725 cc_ack_received(tp, th, CC_ACK); 1726 1727 tp->snd_una = th->th_ack; 1728 /* 1729 * Pull snd_wl2 up to prevent seq wrap relative 1730 * to th_ack. 1731 */ 1732 tp->snd_wl2 = th->th_ack; 1733 tp->t_dupacks = 0; 1734 m_freem(m); 1735 ND6_HINT(tp); /* Some progress has been made. */ 1736 1737 /* 1738 * If all outstanding data are acked, stop 1739 * retransmit timer, otherwise restart timer 1740 * using current (possibly backed-off) value. 1741 * If process is waiting for space, 1742 * wakeup/selwakeup/signal. If data 1743 * are ready to send, let tcp_output 1744 * decide between more output or persist. 1745 */ 1746 #ifdef TCPDEBUG 1747 if (so->so_options & SO_DEBUG) 1748 tcp_trace(TA_INPUT, ostate, tp, 1749 (void *)tcp_saveipgen, 1750 &tcp_savetcp, 0); 1751 #endif 1752 if (tp->snd_una == tp->snd_max) 1753 tcp_timer_activate(tp, TT_REXMT, 0); 1754 else if (!tcp_timer_active(tp, TT_PERSIST)) 1755 tcp_timer_activate(tp, TT_REXMT, 1756 tp->t_rxtcur); 1757 sowwakeup(so); 1758 if (so->so_snd.sb_cc) 1759 (void) tcp_output(tp); 1760 goto check_delack; 1761 } 1762 } else if (th->th_ack == tp->snd_una && 1763 tlen <= sbspace(&so->so_rcv)) { 1764 int newsize = 0; /* automatic sockbuf scaling */ 1765 1766 /* 1767 * This is a pure, in-sequence data packet with 1768 * nothing on the reassembly queue and we have enough 1769 * buffer space to take it. 1770 */ 1771 if (ti_locked == TI_WLOCKED) 1772 INP_INFO_WUNLOCK(&V_tcbinfo); 1773 ti_locked = TI_UNLOCKED; 1774 1775 /* Clean receiver SACK report if present */ 1776 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1777 tcp_clean_sackreport(tp); 1778 TCPSTAT_INC(tcps_preddat); 1779 tp->rcv_nxt += tlen; 1780 /* 1781 * Pull snd_wl1 up to prevent seq wrap relative to 1782 * th_seq. 1783 */ 1784 tp->snd_wl1 = th->th_seq; 1785 /* 1786 * Pull rcv_up up to prevent seq wrap relative to 1787 * rcv_nxt. 1788 */ 1789 tp->rcv_up = tp->rcv_nxt; 1790 TCPSTAT_INC(tcps_rcvpack); 1791 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1792 ND6_HINT(tp); /* Some progress has been made */ 1793 #ifdef TCPDEBUG 1794 if (so->so_options & SO_DEBUG) 1795 tcp_trace(TA_INPUT, ostate, tp, 1796 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1797 #endif 1798 /* 1799 * Automatic sizing of receive socket buffer. Often the send 1800 * buffer size is not optimally adjusted to the actual network 1801 * conditions at hand (delay bandwidth product). Setting the 1802 * buffer size too small limits throughput on links with high 1803 * bandwidth and high delay (eg. trans-continental/oceanic links). 1804 * 1805 * On the receive side the socket buffer memory is only rarely 1806 * used to any significant extent. This allows us to be much 1807 * more aggressive in scaling the receive socket buffer. For 1808 * the case that the buffer space is actually used to a large 1809 * extent and we run out of kernel memory we can simply drop 1810 * the new segments; TCP on the sender will just retransmit it 1811 * later. Setting the buffer size too big may only consume too 1812 * much kernel memory if the application doesn't read() from 1813 * the socket or packet loss or reordering makes use of the 1814 * reassembly queue. 1815 * 1816 * The criteria to step up the receive buffer one notch are: 1817 * 1. the number of bytes received during the time it takes 1818 * one timestamp to be reflected back to us (the RTT); 1819 * 2. received bytes per RTT is within seven eighth of the 1820 * current socket buffer size; 1821 * 3. receive buffer size has not hit maximal automatic size; 1822 * 1823 * This algorithm does one step per RTT at most and only if 1824 * we receive a bulk stream w/o packet losses or reorderings. 1825 * Shrinking the buffer during idle times is not necessary as 1826 * it doesn't consume any memory when idle. 1827 * 1828 * TODO: Only step up if the application is actually serving 1829 * the buffer to better manage the socket buffer resources. 1830 */ 1831 if (V_tcp_do_autorcvbuf && 1832 to.to_tsecr && 1833 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1834 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1835 to.to_tsecr - tp->rfbuf_ts < hz) { 1836 if (tp->rfbuf_cnt > 1837 (so->so_rcv.sb_hiwat / 8 * 7) && 1838 so->so_rcv.sb_hiwat < 1839 V_tcp_autorcvbuf_max) { 1840 newsize = 1841 min(so->so_rcv.sb_hiwat + 1842 V_tcp_autorcvbuf_inc, 1843 V_tcp_autorcvbuf_max); 1844 } 1845 /* Start over with next RTT. */ 1846 tp->rfbuf_ts = 0; 1847 tp->rfbuf_cnt = 0; 1848 } else 1849 tp->rfbuf_cnt += tlen; /* add up */ 1850 } 1851 1852 /* Add data to socket buffer. */ 1853 SOCKBUF_LOCK(&so->so_rcv); 1854 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1855 m_freem(m); 1856 } else { 1857 /* 1858 * Set new socket buffer size. 1859 * Give up when limit is reached. 1860 */ 1861 if (newsize) 1862 if (!sbreserve_locked(&so->so_rcv, 1863 newsize, so, NULL)) 1864 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1865 m_adj(m, drop_hdrlen); /* delayed header drop */ 1866 sbappendstream_locked(&so->so_rcv, m); 1867 } 1868 /* NB: sorwakeup_locked() does an implicit unlock. */ 1869 sorwakeup_locked(so); 1870 if (DELAY_ACK(tp, tlen)) { 1871 tp->t_flags |= TF_DELACK; 1872 } else { 1873 tp->t_flags |= TF_ACKNOW; 1874 tcp_output(tp); 1875 } 1876 goto check_delack; 1877 } 1878 } 1879 1880 /* 1881 * Calculate amount of space in receive window, 1882 * and then do TCP input processing. 1883 * Receive window is amount of space in rcv queue, 1884 * but not less than advertised window. 1885 */ 1886 win = sbspace(&so->so_rcv); 1887 if (win < 0) 1888 win = 0; 1889 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1890 1891 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1892 tp->rfbuf_ts = 0; 1893 tp->rfbuf_cnt = 0; 1894 1895 switch (tp->t_state) { 1896 1897 /* 1898 * If the state is SYN_RECEIVED: 1899 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1900 */ 1901 case TCPS_SYN_RECEIVED: 1902 if ((thflags & TH_ACK) && 1903 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1904 SEQ_GT(th->th_ack, tp->snd_max))) { 1905 rstreason = BANDLIM_RST_OPENPORT; 1906 goto dropwithreset; 1907 } 1908 break; 1909 1910 /* 1911 * If the state is SYN_SENT: 1912 * if seg contains an ACK, but not for our SYN, drop the input. 1913 * if seg contains a RST, then drop the connection. 1914 * if seg does not contain SYN, then drop it. 1915 * Otherwise this is an acceptable SYN segment 1916 * initialize tp->rcv_nxt and tp->irs 1917 * if seg contains ack then advance tp->snd_una 1918 * if seg contains an ECE and ECN support is enabled, the stream 1919 * is ECN capable. 1920 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1921 * arrange for segment to be acked (eventually) 1922 * continue processing rest of data/controls, beginning with URG 1923 */ 1924 case TCPS_SYN_SENT: 1925 if ((thflags & TH_ACK) && 1926 (SEQ_LEQ(th->th_ack, tp->iss) || 1927 SEQ_GT(th->th_ack, tp->snd_max))) { 1928 rstreason = BANDLIM_UNLIMITED; 1929 goto dropwithreset; 1930 } 1931 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1932 TCP_PROBE5(connect_refused, NULL, tp, m->m_data, tp, 1933 th); 1934 tp = tcp_drop(tp, ECONNREFUSED); 1935 } 1936 if (thflags & TH_RST) 1937 goto drop; 1938 if (!(thflags & TH_SYN)) 1939 goto drop; 1940 1941 tp->irs = th->th_seq; 1942 tcp_rcvseqinit(tp); 1943 if (thflags & TH_ACK) { 1944 TCPSTAT_INC(tcps_connects); 1945 soisconnected(so); 1946 #ifdef MAC 1947 mac_socketpeer_set_from_mbuf(m, so); 1948 #endif 1949 /* Do window scaling on this connection? */ 1950 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1951 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1952 tp->rcv_scale = tp->request_r_scale; 1953 } 1954 tp->rcv_adv += imin(tp->rcv_wnd, 1955 TCP_MAXWIN << tp->rcv_scale); 1956 tp->snd_una++; /* SYN is acked */ 1957 /* 1958 * If there's data, delay ACK; if there's also a FIN 1959 * ACKNOW will be turned on later. 1960 */ 1961 if (DELAY_ACK(tp, tlen) && tlen != 0) 1962 tcp_timer_activate(tp, TT_DELACK, 1963 tcp_delacktime); 1964 else 1965 tp->t_flags |= TF_ACKNOW; 1966 1967 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1968 tp->t_flags |= TF_ECN_PERMIT; 1969 TCPSTAT_INC(tcps_ecn_shs); 1970 } 1971 1972 /* 1973 * Received <SYN,ACK> in SYN_SENT[*] state. 1974 * Transitions: 1975 * SYN_SENT --> ESTABLISHED 1976 * SYN_SENT* --> FIN_WAIT_1 1977 */ 1978 tp->t_starttime = ticks; 1979 if (tp->t_flags & TF_NEEDFIN) { 1980 tcp_state_change(tp, TCPS_FIN_WAIT_1); 1981 tp->t_flags &= ~TF_NEEDFIN; 1982 thflags &= ~TH_SYN; 1983 } else { 1984 tcp_state_change(tp, TCPS_ESTABLISHED); 1985 TCP_PROBE5(connect_established, NULL, tp, 1986 m->m_data, tp, th); 1987 cc_conn_init(tp); 1988 tcp_timer_activate(tp, TT_KEEP, 1989 TP_KEEPIDLE(tp)); 1990 } 1991 } else { 1992 /* 1993 * Received initial SYN in SYN-SENT[*] state => 1994 * simultaneous open. If segment contains CC option 1995 * and there is a cached CC, apply TAO test. 1996 * If it succeeds, connection is * half-synchronized. 1997 * Otherwise, do 3-way handshake: 1998 * SYN-SENT -> SYN-RECEIVED 1999 * SYN-SENT* -> SYN-RECEIVED* 2000 * If there was no CC option, clear cached CC value. 2001 */ 2002 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2003 tcp_timer_activate(tp, TT_REXMT, 0); 2004 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2005 } 2006 2007 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 2008 "ti_locked %d", __func__, ti_locked)); 2009 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2010 INP_WLOCK_ASSERT(tp->t_inpcb); 2011 2012 /* 2013 * Advance th->th_seq to correspond to first data byte. 2014 * If data, trim to stay within window, 2015 * dropping FIN if necessary. 2016 */ 2017 th->th_seq++; 2018 if (tlen > tp->rcv_wnd) { 2019 todrop = tlen - tp->rcv_wnd; 2020 m_adj(m, -todrop); 2021 tlen = tp->rcv_wnd; 2022 thflags &= ~TH_FIN; 2023 TCPSTAT_INC(tcps_rcvpackafterwin); 2024 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2025 } 2026 tp->snd_wl1 = th->th_seq - 1; 2027 tp->rcv_up = th->th_seq; 2028 /* 2029 * Client side of transaction: already sent SYN and data. 2030 * If the remote host used T/TCP to validate the SYN, 2031 * our data will be ACK'd; if so, enter normal data segment 2032 * processing in the middle of step 5, ack processing. 2033 * Otherwise, goto step 6. 2034 */ 2035 if (thflags & TH_ACK) 2036 goto process_ACK; 2037 2038 goto step6; 2039 2040 /* 2041 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2042 * do normal processing. 2043 * 2044 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2045 */ 2046 case TCPS_LAST_ACK: 2047 case TCPS_CLOSING: 2048 break; /* continue normal processing */ 2049 } 2050 2051 /* 2052 * States other than LISTEN or SYN_SENT. 2053 * First check the RST flag and sequence number since reset segments 2054 * are exempt from the timestamp and connection count tests. This 2055 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2056 * below which allowed reset segments in half the sequence space 2057 * to fall though and be processed (which gives forged reset 2058 * segments with a random sequence number a 50 percent chance of 2059 * killing a connection). 2060 * Then check timestamp, if present. 2061 * Then check the connection count, if present. 2062 * Then check that at least some bytes of segment are within 2063 * receive window. If segment begins before rcv_nxt, 2064 * drop leading data (and SYN); if nothing left, just ack. 2065 * 2066 * 2067 * If the RST bit is set, check the sequence number to see 2068 * if this is a valid reset segment. 2069 * RFC 793 page 37: 2070 * In all states except SYN-SENT, all reset (RST) segments 2071 * are validated by checking their SEQ-fields. A reset is 2072 * valid if its sequence number is in the window. 2073 * Note: this does not take into account delayed ACKs, so 2074 * we should test against last_ack_sent instead of rcv_nxt. 2075 * The sequence number in the reset segment is normally an 2076 * echo of our outgoing acknowlegement numbers, but some hosts 2077 * send a reset with the sequence number at the rightmost edge 2078 * of our receive window, and we have to handle this case. 2079 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2080 * that brute force RST attacks are possible. To combat this, 2081 * we use a much stricter check while in the ESTABLISHED state, 2082 * only accepting RSTs where the sequence number is equal to 2083 * last_ack_sent. In all other states (the states in which a 2084 * RST is more likely), the more permissive check is used. 2085 * If we have multiple segments in flight, the initial reset 2086 * segment sequence numbers will be to the left of last_ack_sent, 2087 * but they will eventually catch up. 2088 * In any case, it never made sense to trim reset segments to 2089 * fit the receive window since RFC 1122 says: 2090 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2091 * 2092 * A TCP SHOULD allow a received RST segment to include data. 2093 * 2094 * DISCUSSION 2095 * It has been suggested that a RST segment could contain 2096 * ASCII text that encoded and explained the cause of the 2097 * RST. No standard has yet been established for such 2098 * data. 2099 * 2100 * If the reset segment passes the sequence number test examine 2101 * the state: 2102 * SYN_RECEIVED STATE: 2103 * If passive open, return to LISTEN state. 2104 * If active open, inform user that connection was refused. 2105 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2106 * Inform user that connection was reset, and close tcb. 2107 * CLOSING, LAST_ACK STATES: 2108 * Close the tcb. 2109 * TIME_WAIT STATE: 2110 * Drop the segment - see Stevens, vol. 2, p. 964 and 2111 * RFC 1337. 2112 */ 2113 if (thflags & TH_RST) { 2114 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2115 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2116 switch (tp->t_state) { 2117 2118 case TCPS_SYN_RECEIVED: 2119 so->so_error = ECONNREFUSED; 2120 goto close; 2121 2122 case TCPS_ESTABLISHED: 2123 if (V_tcp_insecure_rst == 0 && 2124 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2125 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2126 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2127 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2128 TCPSTAT_INC(tcps_badrst); 2129 goto drop; 2130 } 2131 /* FALLTHROUGH */ 2132 case TCPS_FIN_WAIT_1: 2133 case TCPS_FIN_WAIT_2: 2134 case TCPS_CLOSE_WAIT: 2135 so->so_error = ECONNRESET; 2136 close: 2137 KASSERT(ti_locked == TI_WLOCKED, 2138 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2139 ti_locked)); 2140 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2141 2142 tcp_state_change(tp, TCPS_CLOSED); 2143 TCPSTAT_INC(tcps_drops); 2144 tp = tcp_close(tp); 2145 break; 2146 2147 case TCPS_CLOSING: 2148 case TCPS_LAST_ACK: 2149 KASSERT(ti_locked == TI_WLOCKED, 2150 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2151 ti_locked)); 2152 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2153 2154 tp = tcp_close(tp); 2155 break; 2156 } 2157 } 2158 goto drop; 2159 } 2160 2161 /* 2162 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2163 * and it's less than ts_recent, drop it. 2164 */ 2165 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2166 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2167 2168 /* Check to see if ts_recent is over 24 days old. */ 2169 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2170 /* 2171 * Invalidate ts_recent. If this segment updates 2172 * ts_recent, the age will be reset later and ts_recent 2173 * will get a valid value. If it does not, setting 2174 * ts_recent to zero will at least satisfy the 2175 * requirement that zero be placed in the timestamp 2176 * echo reply when ts_recent isn't valid. The 2177 * age isn't reset until we get a valid ts_recent 2178 * because we don't want out-of-order segments to be 2179 * dropped when ts_recent is old. 2180 */ 2181 tp->ts_recent = 0; 2182 } else { 2183 TCPSTAT_INC(tcps_rcvduppack); 2184 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2185 TCPSTAT_INC(tcps_pawsdrop); 2186 if (tlen) 2187 goto dropafterack; 2188 goto drop; 2189 } 2190 } 2191 2192 /* 2193 * In the SYN-RECEIVED state, validate that the packet belongs to 2194 * this connection before trimming the data to fit the receive 2195 * window. Check the sequence number versus IRS since we know 2196 * the sequence numbers haven't wrapped. This is a partial fix 2197 * for the "LAND" DoS attack. 2198 */ 2199 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2200 rstreason = BANDLIM_RST_OPENPORT; 2201 goto dropwithreset; 2202 } 2203 2204 todrop = tp->rcv_nxt - th->th_seq; 2205 if (todrop > 0) { 2206 /* 2207 * If this is a duplicate SYN for our current connection, 2208 * advance over it and pretend and it's not a SYN. 2209 */ 2210 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2211 thflags &= ~TH_SYN; 2212 th->th_seq++; 2213 if (th->th_urp > 1) 2214 th->th_urp--; 2215 else 2216 thflags &= ~TH_URG; 2217 todrop--; 2218 } 2219 /* 2220 * Following if statement from Stevens, vol. 2, p. 960. 2221 */ 2222 if (todrop > tlen 2223 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2224 /* 2225 * Any valid FIN must be to the left of the window. 2226 * At this point the FIN must be a duplicate or out 2227 * of sequence; drop it. 2228 */ 2229 thflags &= ~TH_FIN; 2230 2231 /* 2232 * Send an ACK to resynchronize and drop any data. 2233 * But keep on processing for RST or ACK. 2234 */ 2235 tp->t_flags |= TF_ACKNOW; 2236 todrop = tlen; 2237 TCPSTAT_INC(tcps_rcvduppack); 2238 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2239 } else { 2240 TCPSTAT_INC(tcps_rcvpartduppack); 2241 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2242 } 2243 drop_hdrlen += todrop; /* drop from the top afterwards */ 2244 th->th_seq += todrop; 2245 tlen -= todrop; 2246 if (th->th_urp > todrop) 2247 th->th_urp -= todrop; 2248 else { 2249 thflags &= ~TH_URG; 2250 th->th_urp = 0; 2251 } 2252 } 2253 2254 /* 2255 * If new data are received on a connection after the 2256 * user processes are gone, then RST the other end. 2257 */ 2258 if ((so->so_state & SS_NOFDREF) && 2259 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2260 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2261 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2262 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2263 2264 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2265 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2266 "after socket was closed, " 2267 "sending RST and removing tcpcb\n", 2268 s, __func__, tcpstates[tp->t_state], tlen); 2269 free(s, M_TCPLOG); 2270 } 2271 tp = tcp_close(tp); 2272 TCPSTAT_INC(tcps_rcvafterclose); 2273 rstreason = BANDLIM_UNLIMITED; 2274 goto dropwithreset; 2275 } 2276 2277 /* 2278 * If segment ends after window, drop trailing data 2279 * (and PUSH and FIN); if nothing left, just ACK. 2280 */ 2281 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2282 if (todrop > 0) { 2283 TCPSTAT_INC(tcps_rcvpackafterwin); 2284 if (todrop >= tlen) { 2285 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2286 /* 2287 * If window is closed can only take segments at 2288 * window edge, and have to drop data and PUSH from 2289 * incoming segments. Continue processing, but 2290 * remember to ack. Otherwise, drop segment 2291 * and ack. 2292 */ 2293 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2294 tp->t_flags |= TF_ACKNOW; 2295 TCPSTAT_INC(tcps_rcvwinprobe); 2296 } else 2297 goto dropafterack; 2298 } else 2299 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2300 m_adj(m, -todrop); 2301 tlen -= todrop; 2302 thflags &= ~(TH_PUSH|TH_FIN); 2303 } 2304 2305 /* 2306 * If last ACK falls within this segment's sequence numbers, 2307 * record its timestamp. 2308 * NOTE: 2309 * 1) That the test incorporates suggestions from the latest 2310 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2311 * 2) That updating only on newer timestamps interferes with 2312 * our earlier PAWS tests, so this check should be solely 2313 * predicated on the sequence space of this segment. 2314 * 3) That we modify the segment boundary check to be 2315 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2316 * instead of RFC1323's 2317 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2318 * This modified check allows us to overcome RFC1323's 2319 * limitations as described in Stevens TCP/IP Illustrated 2320 * Vol. 2 p.869. In such cases, we can still calculate the 2321 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2322 */ 2323 if ((to.to_flags & TOF_TS) != 0 && 2324 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2325 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2326 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2327 tp->ts_recent_age = tcp_ts_getticks(); 2328 tp->ts_recent = to.to_tsval; 2329 } 2330 2331 /* 2332 * If a SYN is in the window, then this is an 2333 * error and we send an RST and drop the connection. 2334 */ 2335 if (thflags & TH_SYN) { 2336 KASSERT(ti_locked == TI_WLOCKED, 2337 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2338 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2339 2340 tp = tcp_drop(tp, ECONNRESET); 2341 rstreason = BANDLIM_UNLIMITED; 2342 goto drop; 2343 } 2344 2345 /* 2346 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2347 * flag is on (half-synchronized state), then queue data for 2348 * later processing; else drop segment and return. 2349 */ 2350 if ((thflags & TH_ACK) == 0) { 2351 if (tp->t_state == TCPS_SYN_RECEIVED || 2352 (tp->t_flags & TF_NEEDSYN)) 2353 goto step6; 2354 else if (tp->t_flags & TF_ACKNOW) 2355 goto dropafterack; 2356 else 2357 goto drop; 2358 } 2359 2360 /* 2361 * Ack processing. 2362 */ 2363 switch (tp->t_state) { 2364 2365 /* 2366 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2367 * ESTABLISHED state and continue processing. 2368 * The ACK was checked above. 2369 */ 2370 case TCPS_SYN_RECEIVED: 2371 2372 TCPSTAT_INC(tcps_connects); 2373 soisconnected(so); 2374 /* Do window scaling? */ 2375 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2376 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2377 tp->rcv_scale = tp->request_r_scale; 2378 tp->snd_wnd = tiwin; 2379 } 2380 /* 2381 * Make transitions: 2382 * SYN-RECEIVED -> ESTABLISHED 2383 * SYN-RECEIVED* -> FIN-WAIT-1 2384 */ 2385 tp->t_starttime = ticks; 2386 if (tp->t_flags & TF_NEEDFIN) { 2387 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2388 tp->t_flags &= ~TF_NEEDFIN; 2389 } else { 2390 tcp_state_change(tp, TCPS_ESTABLISHED); 2391 TCP_PROBE5(accept_established, NULL, tp, m->m_data, tp, 2392 th); 2393 cc_conn_init(tp); 2394 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2395 } 2396 /* 2397 * If segment contains data or ACK, will call tcp_reass() 2398 * later; if not, do so now to pass queued data to user. 2399 */ 2400 if (tlen == 0 && (thflags & TH_FIN) == 0) 2401 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2402 (struct mbuf *)0); 2403 tp->snd_wl1 = th->th_seq - 1; 2404 /* FALLTHROUGH */ 2405 2406 /* 2407 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2408 * ACKs. If the ack is in the range 2409 * tp->snd_una < th->th_ack <= tp->snd_max 2410 * then advance tp->snd_una to th->th_ack and drop 2411 * data from the retransmission queue. If this ACK reflects 2412 * more up to date window information we update our window information. 2413 */ 2414 case TCPS_ESTABLISHED: 2415 case TCPS_FIN_WAIT_1: 2416 case TCPS_FIN_WAIT_2: 2417 case TCPS_CLOSE_WAIT: 2418 case TCPS_CLOSING: 2419 case TCPS_LAST_ACK: 2420 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2421 TCPSTAT_INC(tcps_rcvacktoomuch); 2422 goto dropafterack; 2423 } 2424 if ((tp->t_flags & TF_SACK_PERMIT) && 2425 ((to.to_flags & TOF_SACK) || 2426 !TAILQ_EMPTY(&tp->snd_holes))) 2427 tcp_sack_doack(tp, &to, th->th_ack); 2428 2429 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2430 hhook_run_tcp_est_in(tp, th, &to); 2431 2432 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2433 if (tlen == 0 && tiwin == tp->snd_wnd) { 2434 TCPSTAT_INC(tcps_rcvdupack); 2435 /* 2436 * If we have outstanding data (other than 2437 * a window probe), this is a completely 2438 * duplicate ack (ie, window info didn't 2439 * change), the ack is the biggest we've 2440 * seen and we've seen exactly our rexmt 2441 * threshhold of them, assume a packet 2442 * has been dropped and retransmit it. 2443 * Kludge snd_nxt & the congestion 2444 * window so we send only this one 2445 * packet. 2446 * 2447 * We know we're losing at the current 2448 * window size so do congestion avoidance 2449 * (set ssthresh to half the current window 2450 * and pull our congestion window back to 2451 * the new ssthresh). 2452 * 2453 * Dup acks mean that packets have left the 2454 * network (they're now cached at the receiver) 2455 * so bump cwnd by the amount in the receiver 2456 * to keep a constant cwnd packets in the 2457 * network. 2458 * 2459 * When using TCP ECN, notify the peer that 2460 * we reduced the cwnd. 2461 */ 2462 if (!tcp_timer_active(tp, TT_REXMT) || 2463 th->th_ack != tp->snd_una) 2464 tp->t_dupacks = 0; 2465 else if (++tp->t_dupacks > tcprexmtthresh || 2466 IN_FASTRECOVERY(tp->t_flags)) { 2467 cc_ack_received(tp, th, CC_DUPACK); 2468 if ((tp->t_flags & TF_SACK_PERMIT) && 2469 IN_FASTRECOVERY(tp->t_flags)) { 2470 int awnd; 2471 2472 /* 2473 * Compute the amount of data in flight first. 2474 * We can inject new data into the pipe iff 2475 * we have less than 1/2 the original window's 2476 * worth of data in flight. 2477 */ 2478 awnd = (tp->snd_nxt - tp->snd_fack) + 2479 tp->sackhint.sack_bytes_rexmit; 2480 if (awnd < tp->snd_ssthresh) { 2481 tp->snd_cwnd += tp->t_maxseg; 2482 if (tp->snd_cwnd > tp->snd_ssthresh) 2483 tp->snd_cwnd = tp->snd_ssthresh; 2484 } 2485 } else 2486 tp->snd_cwnd += tp->t_maxseg; 2487 if ((thflags & TH_FIN) && 2488 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2489 /* 2490 * If its a fin we need to process 2491 * it to avoid a race where both 2492 * sides enter FIN-WAIT and send FIN|ACK 2493 * at the same time. 2494 */ 2495 break; 2496 } 2497 (void) tcp_output(tp); 2498 goto drop; 2499 } else if (tp->t_dupacks == tcprexmtthresh) { 2500 tcp_seq onxt = tp->snd_nxt; 2501 2502 /* 2503 * If we're doing sack, check to 2504 * see if we're already in sack 2505 * recovery. If we're not doing sack, 2506 * check to see if we're in newreno 2507 * recovery. 2508 */ 2509 if (tp->t_flags & TF_SACK_PERMIT) { 2510 if (IN_FASTRECOVERY(tp->t_flags)) { 2511 tp->t_dupacks = 0; 2512 break; 2513 } 2514 } else { 2515 if (SEQ_LEQ(th->th_ack, 2516 tp->snd_recover)) { 2517 tp->t_dupacks = 0; 2518 break; 2519 } 2520 } 2521 /* Congestion signal before ack. */ 2522 cc_cong_signal(tp, th, CC_NDUPACK); 2523 cc_ack_received(tp, th, CC_DUPACK); 2524 tcp_timer_activate(tp, TT_REXMT, 0); 2525 tp->t_rtttime = 0; 2526 if (tp->t_flags & TF_SACK_PERMIT) { 2527 TCPSTAT_INC( 2528 tcps_sack_recovery_episode); 2529 tp->sack_newdata = tp->snd_nxt; 2530 tp->snd_cwnd = tp->t_maxseg; 2531 (void) tcp_output(tp); 2532 goto drop; 2533 } 2534 tp->snd_nxt = th->th_ack; 2535 tp->snd_cwnd = tp->t_maxseg; 2536 if ((thflags & TH_FIN) && 2537 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2538 /* 2539 * If its a fin we need to process 2540 * it to avoid a race where both 2541 * sides enter FIN-WAIT and send FIN|ACK 2542 * at the same time. 2543 */ 2544 break; 2545 } 2546 (void) tcp_output(tp); 2547 KASSERT(tp->snd_limited <= 2, 2548 ("%s: tp->snd_limited too big", 2549 __func__)); 2550 tp->snd_cwnd = tp->snd_ssthresh + 2551 tp->t_maxseg * 2552 (tp->t_dupacks - tp->snd_limited); 2553 if (SEQ_GT(onxt, tp->snd_nxt)) 2554 tp->snd_nxt = onxt; 2555 goto drop; 2556 } else if (V_tcp_do_rfc3042) { 2557 cc_ack_received(tp, th, CC_DUPACK); 2558 u_long oldcwnd = tp->snd_cwnd; 2559 tcp_seq oldsndmax = tp->snd_max; 2560 u_int sent; 2561 int avail; 2562 2563 KASSERT(tp->t_dupacks == 1 || 2564 tp->t_dupacks == 2, 2565 ("%s: dupacks not 1 or 2", 2566 __func__)); 2567 if (tp->t_dupacks == 1) 2568 tp->snd_limited = 0; 2569 tp->snd_cwnd = 2570 (tp->snd_nxt - tp->snd_una) + 2571 (tp->t_dupacks - tp->snd_limited) * 2572 tp->t_maxseg; 2573 if ((thflags & TH_FIN) && 2574 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2575 /* 2576 * If its a fin we need to process 2577 * it to avoid a race where both 2578 * sides enter FIN-WAIT and send FIN|ACK 2579 * at the same time. 2580 */ 2581 break; 2582 } 2583 /* 2584 * Only call tcp_output when there 2585 * is new data available to be sent. 2586 * Otherwise we would send pure ACKs. 2587 */ 2588 SOCKBUF_LOCK(&so->so_snd); 2589 avail = so->so_snd.sb_cc - 2590 (tp->snd_nxt - tp->snd_una); 2591 SOCKBUF_UNLOCK(&so->so_snd); 2592 if (avail > 0) 2593 (void) tcp_output(tp); 2594 sent = tp->snd_max - oldsndmax; 2595 if (sent > tp->t_maxseg) { 2596 KASSERT((tp->t_dupacks == 2 && 2597 tp->snd_limited == 0) || 2598 (sent == tp->t_maxseg + 1 && 2599 tp->t_flags & TF_SENTFIN), 2600 ("%s: sent too much", 2601 __func__)); 2602 tp->snd_limited = 2; 2603 } else if (sent > 0) 2604 ++tp->snd_limited; 2605 tp->snd_cwnd = oldcwnd; 2606 goto drop; 2607 } 2608 } else 2609 tp->t_dupacks = 0; 2610 break; 2611 } 2612 2613 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2614 ("%s: th_ack <= snd_una", __func__)); 2615 2616 /* 2617 * If the congestion window was inflated to account 2618 * for the other side's cached packets, retract it. 2619 */ 2620 if (IN_FASTRECOVERY(tp->t_flags)) { 2621 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2622 if (tp->t_flags & TF_SACK_PERMIT) 2623 tcp_sack_partialack(tp, th); 2624 else 2625 tcp_newreno_partial_ack(tp, th); 2626 } else 2627 cc_post_recovery(tp, th); 2628 } 2629 tp->t_dupacks = 0; 2630 /* 2631 * If we reach this point, ACK is not a duplicate, 2632 * i.e., it ACKs something we sent. 2633 */ 2634 if (tp->t_flags & TF_NEEDSYN) { 2635 /* 2636 * T/TCP: Connection was half-synchronized, and our 2637 * SYN has been ACK'd (so connection is now fully 2638 * synchronized). Go to non-starred state, 2639 * increment snd_una for ACK of SYN, and check if 2640 * we can do window scaling. 2641 */ 2642 tp->t_flags &= ~TF_NEEDSYN; 2643 tp->snd_una++; 2644 /* Do window scaling? */ 2645 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2646 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2647 tp->rcv_scale = tp->request_r_scale; 2648 /* Send window already scaled. */ 2649 } 2650 } 2651 2652 process_ACK: 2653 INP_WLOCK_ASSERT(tp->t_inpcb); 2654 2655 acked = BYTES_THIS_ACK(tp, th); 2656 TCPSTAT_INC(tcps_rcvackpack); 2657 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2658 2659 /* 2660 * If we just performed our first retransmit, and the ACK 2661 * arrives within our recovery window, then it was a mistake 2662 * to do the retransmit in the first place. Recover our 2663 * original cwnd and ssthresh, and proceed to transmit where 2664 * we left off. 2665 */ 2666 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2667 (int)(ticks - tp->t_badrxtwin) < 0) 2668 cc_cong_signal(tp, th, CC_RTO_ERR); 2669 2670 /* 2671 * If we have a timestamp reply, update smoothed 2672 * round trip time. If no timestamp is present but 2673 * transmit timer is running and timed sequence 2674 * number was acked, update smoothed round trip time. 2675 * Since we now have an rtt measurement, cancel the 2676 * timer backoff (cf., Phil Karn's retransmit alg.). 2677 * Recompute the initial retransmit timer. 2678 * 2679 * Some boxes send broken timestamp replies 2680 * during the SYN+ACK phase, ignore 2681 * timestamps of 0 or we could calculate a 2682 * huge RTT and blow up the retransmit timer. 2683 */ 2684 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2685 u_int t; 2686 2687 t = tcp_ts_getticks() - to.to_tsecr; 2688 if (!tp->t_rttlow || tp->t_rttlow > t) 2689 tp->t_rttlow = t; 2690 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2691 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2692 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2693 tp->t_rttlow = ticks - tp->t_rtttime; 2694 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2695 } 2696 2697 /* 2698 * If all outstanding data is acked, stop retransmit 2699 * timer and remember to restart (more output or persist). 2700 * If there is more data to be acked, restart retransmit 2701 * timer, using current (possibly backed-off) value. 2702 */ 2703 if (th->th_ack == tp->snd_max) { 2704 tcp_timer_activate(tp, TT_REXMT, 0); 2705 needoutput = 1; 2706 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2707 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2708 2709 /* 2710 * If no data (only SYN) was ACK'd, 2711 * skip rest of ACK processing. 2712 */ 2713 if (acked == 0) 2714 goto step6; 2715 2716 /* 2717 * Let the congestion control algorithm update congestion 2718 * control related information. This typically means increasing 2719 * the congestion window. 2720 */ 2721 cc_ack_received(tp, th, CC_ACK); 2722 2723 SOCKBUF_LOCK(&so->so_snd); 2724 if (acked > so->so_snd.sb_cc) { 2725 tp->snd_wnd -= so->so_snd.sb_cc; 2726 mfree = sbcut_locked(&so->so_snd, 2727 (int)so->so_snd.sb_cc); 2728 ourfinisacked = 1; 2729 } else { 2730 mfree = sbcut_locked(&so->so_snd, acked); 2731 tp->snd_wnd -= acked; 2732 ourfinisacked = 0; 2733 } 2734 /* NB: sowwakeup_locked() does an implicit unlock. */ 2735 sowwakeup_locked(so); 2736 m_freem(mfree); 2737 /* Detect una wraparound. */ 2738 if (!IN_RECOVERY(tp->t_flags) && 2739 SEQ_GT(tp->snd_una, tp->snd_recover) && 2740 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2741 tp->snd_recover = th->th_ack - 1; 2742 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2743 if (IN_RECOVERY(tp->t_flags) && 2744 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2745 EXIT_RECOVERY(tp->t_flags); 2746 } 2747 tp->snd_una = th->th_ack; 2748 if (tp->t_flags & TF_SACK_PERMIT) { 2749 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2750 tp->snd_recover = tp->snd_una; 2751 } 2752 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2753 tp->snd_nxt = tp->snd_una; 2754 2755 switch (tp->t_state) { 2756 2757 /* 2758 * In FIN_WAIT_1 STATE in addition to the processing 2759 * for the ESTABLISHED state if our FIN is now acknowledged 2760 * then enter FIN_WAIT_2. 2761 */ 2762 case TCPS_FIN_WAIT_1: 2763 if (ourfinisacked) { 2764 /* 2765 * If we can't receive any more 2766 * data, then closing user can proceed. 2767 * Starting the timer is contrary to the 2768 * specification, but if we don't get a FIN 2769 * we'll hang forever. 2770 * 2771 * XXXjl: 2772 * we should release the tp also, and use a 2773 * compressed state. 2774 */ 2775 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2776 soisdisconnected(so); 2777 tcp_timer_activate(tp, TT_2MSL, 2778 (tcp_fast_finwait2_recycle ? 2779 tcp_finwait2_timeout : 2780 TP_MAXIDLE(tp))); 2781 } 2782 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2783 } 2784 break; 2785 2786 /* 2787 * In CLOSING STATE in addition to the processing for 2788 * the ESTABLISHED state if the ACK acknowledges our FIN 2789 * then enter the TIME-WAIT state, otherwise ignore 2790 * the segment. 2791 */ 2792 case TCPS_CLOSING: 2793 if (ourfinisacked) { 2794 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2795 tcp_twstart(tp); 2796 INP_INFO_WUNLOCK(&V_tcbinfo); 2797 m_freem(m); 2798 return; 2799 } 2800 break; 2801 2802 /* 2803 * In LAST_ACK, we may still be waiting for data to drain 2804 * and/or to be acked, as well as for the ack of our FIN. 2805 * If our FIN is now acknowledged, delete the TCB, 2806 * enter the closed state and return. 2807 */ 2808 case TCPS_LAST_ACK: 2809 if (ourfinisacked) { 2810 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2811 tp = tcp_close(tp); 2812 goto drop; 2813 } 2814 break; 2815 } 2816 } 2817 2818 step6: 2819 INP_WLOCK_ASSERT(tp->t_inpcb); 2820 2821 /* 2822 * Update window information. 2823 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2824 */ 2825 if ((thflags & TH_ACK) && 2826 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2827 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2828 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2829 /* keep track of pure window updates */ 2830 if (tlen == 0 && 2831 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2832 TCPSTAT_INC(tcps_rcvwinupd); 2833 tp->snd_wnd = tiwin; 2834 tp->snd_wl1 = th->th_seq; 2835 tp->snd_wl2 = th->th_ack; 2836 if (tp->snd_wnd > tp->max_sndwnd) 2837 tp->max_sndwnd = tp->snd_wnd; 2838 needoutput = 1; 2839 } 2840 2841 /* 2842 * Process segments with URG. 2843 */ 2844 if ((thflags & TH_URG) && th->th_urp && 2845 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2846 /* 2847 * This is a kludge, but if we receive and accept 2848 * random urgent pointers, we'll crash in 2849 * soreceive. It's hard to imagine someone 2850 * actually wanting to send this much urgent data. 2851 */ 2852 SOCKBUF_LOCK(&so->so_rcv); 2853 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2854 th->th_urp = 0; /* XXX */ 2855 thflags &= ~TH_URG; /* XXX */ 2856 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2857 goto dodata; /* XXX */ 2858 } 2859 /* 2860 * If this segment advances the known urgent pointer, 2861 * then mark the data stream. This should not happen 2862 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2863 * a FIN has been received from the remote side. 2864 * In these states we ignore the URG. 2865 * 2866 * According to RFC961 (Assigned Protocols), 2867 * the urgent pointer points to the last octet 2868 * of urgent data. We continue, however, 2869 * to consider it to indicate the first octet 2870 * of data past the urgent section as the original 2871 * spec states (in one of two places). 2872 */ 2873 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2874 tp->rcv_up = th->th_seq + th->th_urp; 2875 so->so_oobmark = so->so_rcv.sb_cc + 2876 (tp->rcv_up - tp->rcv_nxt) - 1; 2877 if (so->so_oobmark == 0) 2878 so->so_rcv.sb_state |= SBS_RCVATMARK; 2879 sohasoutofband(so); 2880 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2881 } 2882 SOCKBUF_UNLOCK(&so->so_rcv); 2883 /* 2884 * Remove out of band data so doesn't get presented to user. 2885 * This can happen independent of advancing the URG pointer, 2886 * but if two URG's are pending at once, some out-of-band 2887 * data may creep in... ick. 2888 */ 2889 if (th->th_urp <= (u_long)tlen && 2890 !(so->so_options & SO_OOBINLINE)) { 2891 /* hdr drop is delayed */ 2892 tcp_pulloutofband(so, th, m, drop_hdrlen); 2893 } 2894 } else { 2895 /* 2896 * If no out of band data is expected, 2897 * pull receive urgent pointer along 2898 * with the receive window. 2899 */ 2900 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2901 tp->rcv_up = tp->rcv_nxt; 2902 } 2903 dodata: /* XXX */ 2904 INP_WLOCK_ASSERT(tp->t_inpcb); 2905 2906 /* 2907 * Process the segment text, merging it into the TCP sequencing queue, 2908 * and arranging for acknowledgment of receipt if necessary. 2909 * This process logically involves adjusting tp->rcv_wnd as data 2910 * is presented to the user (this happens in tcp_usrreq.c, 2911 * case PRU_RCVD). If a FIN has already been received on this 2912 * connection then we just ignore the text. 2913 */ 2914 if ((tlen || (thflags & TH_FIN)) && 2915 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2916 tcp_seq save_start = th->th_seq; 2917 m_adj(m, drop_hdrlen); /* delayed header drop */ 2918 /* 2919 * Insert segment which includes th into TCP reassembly queue 2920 * with control block tp. Set thflags to whether reassembly now 2921 * includes a segment with FIN. This handles the common case 2922 * inline (segment is the next to be received on an established 2923 * connection, and the queue is empty), avoiding linkage into 2924 * and removal from the queue and repetition of various 2925 * conversions. 2926 * Set DELACK for segments received in order, but ack 2927 * immediately when segments are out of order (so 2928 * fast retransmit can work). 2929 */ 2930 if (th->th_seq == tp->rcv_nxt && 2931 LIST_EMPTY(&tp->t_segq) && 2932 TCPS_HAVEESTABLISHED(tp->t_state)) { 2933 if (DELAY_ACK(tp, tlen)) 2934 tp->t_flags |= TF_DELACK; 2935 else 2936 tp->t_flags |= TF_ACKNOW; 2937 tp->rcv_nxt += tlen; 2938 thflags = th->th_flags & TH_FIN; 2939 TCPSTAT_INC(tcps_rcvpack); 2940 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2941 ND6_HINT(tp); 2942 SOCKBUF_LOCK(&so->so_rcv); 2943 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2944 m_freem(m); 2945 else 2946 sbappendstream_locked(&so->so_rcv, m); 2947 /* NB: sorwakeup_locked() does an implicit unlock. */ 2948 sorwakeup_locked(so); 2949 } else { 2950 /* 2951 * XXX: Due to the header drop above "th" is 2952 * theoretically invalid by now. Fortunately 2953 * m_adj() doesn't actually frees any mbufs 2954 * when trimming from the head. 2955 */ 2956 thflags = tcp_reass(tp, th, &tlen, m); 2957 tp->t_flags |= TF_ACKNOW; 2958 } 2959 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2960 tcp_update_sack_list(tp, save_start, save_start + tlen); 2961 #if 0 2962 /* 2963 * Note the amount of data that peer has sent into 2964 * our window, in order to estimate the sender's 2965 * buffer size. 2966 * XXX: Unused. 2967 */ 2968 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2969 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2970 else 2971 len = so->so_rcv.sb_hiwat; 2972 #endif 2973 } else { 2974 m_freem(m); 2975 thflags &= ~TH_FIN; 2976 } 2977 2978 /* 2979 * If FIN is received ACK the FIN and let the user know 2980 * that the connection is closing. 2981 */ 2982 if (thflags & TH_FIN) { 2983 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2984 socantrcvmore(so); 2985 /* 2986 * If connection is half-synchronized 2987 * (ie NEEDSYN flag on) then delay ACK, 2988 * so it may be piggybacked when SYN is sent. 2989 * Otherwise, since we received a FIN then no 2990 * more input can be expected, send ACK now. 2991 */ 2992 if (tp->t_flags & TF_NEEDSYN) 2993 tp->t_flags |= TF_DELACK; 2994 else 2995 tp->t_flags |= TF_ACKNOW; 2996 tp->rcv_nxt++; 2997 } 2998 switch (tp->t_state) { 2999 3000 /* 3001 * In SYN_RECEIVED and ESTABLISHED STATES 3002 * enter the CLOSE_WAIT state. 3003 */ 3004 case TCPS_SYN_RECEIVED: 3005 tp->t_starttime = ticks; 3006 /* FALLTHROUGH */ 3007 case TCPS_ESTABLISHED: 3008 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3009 break; 3010 3011 /* 3012 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3013 * enter the CLOSING state. 3014 */ 3015 case TCPS_FIN_WAIT_1: 3016 tcp_state_change(tp, TCPS_CLOSING); 3017 break; 3018 3019 /* 3020 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3021 * starting the time-wait timer, turning off the other 3022 * standard timers. 3023 */ 3024 case TCPS_FIN_WAIT_2: 3025 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 3026 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 3027 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3028 ti_locked)); 3029 3030 tcp_twstart(tp); 3031 INP_INFO_WUNLOCK(&V_tcbinfo); 3032 return; 3033 } 3034 } 3035 if (ti_locked == TI_WLOCKED) 3036 INP_INFO_WUNLOCK(&V_tcbinfo); 3037 ti_locked = TI_UNLOCKED; 3038 3039 #ifdef TCPDEBUG 3040 if (so->so_options & SO_DEBUG) 3041 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3042 &tcp_savetcp, 0); 3043 #endif 3044 3045 /* 3046 * Return any desired output. 3047 */ 3048 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3049 (void) tcp_output(tp); 3050 3051 check_delack: 3052 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3053 __func__, ti_locked)); 3054 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3055 INP_WLOCK_ASSERT(tp->t_inpcb); 3056 3057 if (tp->t_flags & TF_DELACK) { 3058 tp->t_flags &= ~TF_DELACK; 3059 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3060 } 3061 INP_WUNLOCK(tp->t_inpcb); 3062 return; 3063 3064 dropafterack: 3065 /* 3066 * Generate an ACK dropping incoming segment if it occupies 3067 * sequence space, where the ACK reflects our state. 3068 * 3069 * We can now skip the test for the RST flag since all 3070 * paths to this code happen after packets containing 3071 * RST have been dropped. 3072 * 3073 * In the SYN-RECEIVED state, don't send an ACK unless the 3074 * segment we received passes the SYN-RECEIVED ACK test. 3075 * If it fails send a RST. This breaks the loop in the 3076 * "LAND" DoS attack, and also prevents an ACK storm 3077 * between two listening ports that have been sent forged 3078 * SYN segments, each with the source address of the other. 3079 */ 3080 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3081 (SEQ_GT(tp->snd_una, th->th_ack) || 3082 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3083 rstreason = BANDLIM_RST_OPENPORT; 3084 goto dropwithreset; 3085 } 3086 #ifdef TCPDEBUG 3087 if (so->so_options & SO_DEBUG) 3088 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3089 &tcp_savetcp, 0); 3090 #endif 3091 if (ti_locked == TI_WLOCKED) 3092 INP_INFO_WUNLOCK(&V_tcbinfo); 3093 ti_locked = TI_UNLOCKED; 3094 3095 tp->t_flags |= TF_ACKNOW; 3096 (void) tcp_output(tp); 3097 INP_WUNLOCK(tp->t_inpcb); 3098 m_freem(m); 3099 return; 3100 3101 dropwithreset: 3102 if (ti_locked == TI_WLOCKED) 3103 INP_INFO_WUNLOCK(&V_tcbinfo); 3104 ti_locked = TI_UNLOCKED; 3105 3106 if (tp != NULL) { 3107 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3108 INP_WUNLOCK(tp->t_inpcb); 3109 } else 3110 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3111 return; 3112 3113 drop: 3114 if (ti_locked == TI_WLOCKED) { 3115 INP_INFO_WUNLOCK(&V_tcbinfo); 3116 ti_locked = TI_UNLOCKED; 3117 } 3118 #ifdef INVARIANTS 3119 else 3120 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3121 #endif 3122 3123 /* 3124 * Drop space held by incoming segment and return. 3125 */ 3126 #ifdef TCPDEBUG 3127 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3128 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3129 &tcp_savetcp, 0); 3130 #endif 3131 if (tp != NULL) 3132 INP_WUNLOCK(tp->t_inpcb); 3133 m_freem(m); 3134 } 3135 3136 /* 3137 * Issue RST and make ACK acceptable to originator of segment. 3138 * The mbuf must still include the original packet header. 3139 * tp may be NULL. 3140 */ 3141 static void 3142 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3143 int tlen, int rstreason) 3144 { 3145 #ifdef INET 3146 struct ip *ip; 3147 #endif 3148 #ifdef INET6 3149 struct ip6_hdr *ip6; 3150 #endif 3151 3152 if (tp != NULL) { 3153 INP_WLOCK_ASSERT(tp->t_inpcb); 3154 } 3155 3156 /* Don't bother if destination was broadcast/multicast. */ 3157 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3158 goto drop; 3159 #ifdef INET6 3160 if (mtod(m, struct ip *)->ip_v == 6) { 3161 ip6 = mtod(m, struct ip6_hdr *); 3162 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3163 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3164 goto drop; 3165 /* IPv6 anycast check is done at tcp6_input() */ 3166 } 3167 #endif 3168 #if defined(INET) && defined(INET6) 3169 else 3170 #endif 3171 #ifdef INET 3172 { 3173 ip = mtod(m, struct ip *); 3174 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3175 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3176 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3177 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3178 goto drop; 3179 } 3180 #endif 3181 3182 /* Perform bandwidth limiting. */ 3183 if (badport_bandlim(rstreason) < 0) 3184 goto drop; 3185 3186 /* tcp_respond consumes the mbuf chain. */ 3187 if (th->th_flags & TH_ACK) { 3188 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3189 th->th_ack, TH_RST); 3190 } else { 3191 if (th->th_flags & TH_SYN) 3192 tlen++; 3193 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3194 (tcp_seq)0, TH_RST|TH_ACK); 3195 } 3196 return; 3197 drop: 3198 m_freem(m); 3199 } 3200 3201 /* 3202 * Parse TCP options and place in tcpopt. 3203 */ 3204 static void 3205 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3206 { 3207 int opt, optlen; 3208 3209 to->to_flags = 0; 3210 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3211 opt = cp[0]; 3212 if (opt == TCPOPT_EOL) 3213 break; 3214 if (opt == TCPOPT_NOP) 3215 optlen = 1; 3216 else { 3217 if (cnt < 2) 3218 break; 3219 optlen = cp[1]; 3220 if (optlen < 2 || optlen > cnt) 3221 break; 3222 } 3223 switch (opt) { 3224 case TCPOPT_MAXSEG: 3225 if (optlen != TCPOLEN_MAXSEG) 3226 continue; 3227 if (!(flags & TO_SYN)) 3228 continue; 3229 to->to_flags |= TOF_MSS; 3230 bcopy((char *)cp + 2, 3231 (char *)&to->to_mss, sizeof(to->to_mss)); 3232 to->to_mss = ntohs(to->to_mss); 3233 break; 3234 case TCPOPT_WINDOW: 3235 if (optlen != TCPOLEN_WINDOW) 3236 continue; 3237 if (!(flags & TO_SYN)) 3238 continue; 3239 to->to_flags |= TOF_SCALE; 3240 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3241 break; 3242 case TCPOPT_TIMESTAMP: 3243 if (optlen != TCPOLEN_TIMESTAMP) 3244 continue; 3245 to->to_flags |= TOF_TS; 3246 bcopy((char *)cp + 2, 3247 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3248 to->to_tsval = ntohl(to->to_tsval); 3249 bcopy((char *)cp + 6, 3250 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3251 to->to_tsecr = ntohl(to->to_tsecr); 3252 break; 3253 #ifdef TCP_SIGNATURE 3254 /* 3255 * XXX In order to reply to a host which has set the 3256 * TCP_SIGNATURE option in its initial SYN, we have to 3257 * record the fact that the option was observed here 3258 * for the syncache code to perform the correct response. 3259 */ 3260 case TCPOPT_SIGNATURE: 3261 if (optlen != TCPOLEN_SIGNATURE) 3262 continue; 3263 to->to_flags |= TOF_SIGNATURE; 3264 to->to_signature = cp + 2; 3265 break; 3266 #endif 3267 case TCPOPT_SACK_PERMITTED: 3268 if (optlen != TCPOLEN_SACK_PERMITTED) 3269 continue; 3270 if (!(flags & TO_SYN)) 3271 continue; 3272 if (!V_tcp_do_sack) 3273 continue; 3274 to->to_flags |= TOF_SACKPERM; 3275 break; 3276 case TCPOPT_SACK: 3277 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3278 continue; 3279 if (flags & TO_SYN) 3280 continue; 3281 to->to_flags |= TOF_SACK; 3282 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3283 to->to_sacks = cp + 2; 3284 TCPSTAT_INC(tcps_sack_rcv_blocks); 3285 break; 3286 default: 3287 continue; 3288 } 3289 } 3290 } 3291 3292 /* 3293 * Pull out of band byte out of a segment so 3294 * it doesn't appear in the user's data queue. 3295 * It is still reflected in the segment length for 3296 * sequencing purposes. 3297 */ 3298 static void 3299 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3300 int off) 3301 { 3302 int cnt = off + th->th_urp - 1; 3303 3304 while (cnt >= 0) { 3305 if (m->m_len > cnt) { 3306 char *cp = mtod(m, caddr_t) + cnt; 3307 struct tcpcb *tp = sototcpcb(so); 3308 3309 INP_WLOCK_ASSERT(tp->t_inpcb); 3310 3311 tp->t_iobc = *cp; 3312 tp->t_oobflags |= TCPOOB_HAVEDATA; 3313 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3314 m->m_len--; 3315 if (m->m_flags & M_PKTHDR) 3316 m->m_pkthdr.len--; 3317 return; 3318 } 3319 cnt -= m->m_len; 3320 m = m->m_next; 3321 if (m == NULL) 3322 break; 3323 } 3324 panic("tcp_pulloutofband"); 3325 } 3326 3327 /* 3328 * Collect new round-trip time estimate 3329 * and update averages and current timeout. 3330 */ 3331 static void 3332 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3333 { 3334 int delta; 3335 3336 INP_WLOCK_ASSERT(tp->t_inpcb); 3337 3338 TCPSTAT_INC(tcps_rttupdated); 3339 tp->t_rttupdated++; 3340 if (tp->t_srtt != 0) { 3341 /* 3342 * srtt is stored as fixed point with 5 bits after the 3343 * binary point (i.e., scaled by 8). The following magic 3344 * is equivalent to the smoothing algorithm in rfc793 with 3345 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3346 * point). Adjust rtt to origin 0. 3347 */ 3348 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3349 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3350 3351 if ((tp->t_srtt += delta) <= 0) 3352 tp->t_srtt = 1; 3353 3354 /* 3355 * We accumulate a smoothed rtt variance (actually, a 3356 * smoothed mean difference), then set the retransmit 3357 * timer to smoothed rtt + 4 times the smoothed variance. 3358 * rttvar is stored as fixed point with 4 bits after the 3359 * binary point (scaled by 16). The following is 3360 * equivalent to rfc793 smoothing with an alpha of .75 3361 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3362 * rfc793's wired-in beta. 3363 */ 3364 if (delta < 0) 3365 delta = -delta; 3366 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3367 if ((tp->t_rttvar += delta) <= 0) 3368 tp->t_rttvar = 1; 3369 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3370 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3371 } else { 3372 /* 3373 * No rtt measurement yet - use the unsmoothed rtt. 3374 * Set the variance to half the rtt (so our first 3375 * retransmit happens at 3*rtt). 3376 */ 3377 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3378 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3379 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3380 } 3381 tp->t_rtttime = 0; 3382 tp->t_rxtshift = 0; 3383 3384 /* 3385 * the retransmit should happen at rtt + 4 * rttvar. 3386 * Because of the way we do the smoothing, srtt and rttvar 3387 * will each average +1/2 tick of bias. When we compute 3388 * the retransmit timer, we want 1/2 tick of rounding and 3389 * 1 extra tick because of +-1/2 tick uncertainty in the 3390 * firing of the timer. The bias will give us exactly the 3391 * 1.5 tick we need. But, because the bias is 3392 * statistical, we have to test that we don't drop below 3393 * the minimum feasible timer (which is 2 ticks). 3394 */ 3395 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3396 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3397 3398 /* 3399 * We received an ack for a packet that wasn't retransmitted; 3400 * it is probably safe to discard any error indications we've 3401 * received recently. This isn't quite right, but close enough 3402 * for now (a route might have failed after we sent a segment, 3403 * and the return path might not be symmetrical). 3404 */ 3405 tp->t_softerror = 0; 3406 } 3407 3408 /* 3409 * Determine a reasonable value for maxseg size. 3410 * If the route is known, check route for mtu. 3411 * If none, use an mss that can be handled on the outgoing interface 3412 * without forcing IP to fragment. If no route is found, route has no mtu, 3413 * or the destination isn't local, use a default, hopefully conservative 3414 * size (usually 512 or the default IP max size, but no more than the mtu 3415 * of the interface), as we can't discover anything about intervening 3416 * gateways or networks. We also initialize the congestion/slow start 3417 * window to be a single segment if the destination isn't local. 3418 * While looking at the routing entry, we also initialize other path-dependent 3419 * parameters from pre-set or cached values in the routing entry. 3420 * 3421 * Also take into account the space needed for options that we 3422 * send regularly. Make maxseg shorter by that amount to assure 3423 * that we can send maxseg amount of data even when the options 3424 * are present. Store the upper limit of the length of options plus 3425 * data in maxopd. 3426 * 3427 * NOTE that this routine is only called when we process an incoming 3428 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3429 * settings are handled in tcp_mssopt(). 3430 */ 3431 void 3432 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3433 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3434 { 3435 int mss = 0; 3436 u_long maxmtu = 0; 3437 struct inpcb *inp = tp->t_inpcb; 3438 struct hc_metrics_lite metrics; 3439 int origoffer; 3440 #ifdef INET6 3441 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3442 size_t min_protoh = isipv6 ? 3443 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3444 sizeof (struct tcpiphdr); 3445 #else 3446 const size_t min_protoh = sizeof(struct tcpiphdr); 3447 #endif 3448 3449 INP_WLOCK_ASSERT(tp->t_inpcb); 3450 3451 if (mtuoffer != -1) { 3452 KASSERT(offer == -1, ("%s: conflict", __func__)); 3453 offer = mtuoffer - min_protoh; 3454 } 3455 origoffer = offer; 3456 3457 /* Initialize. */ 3458 #ifdef INET6 3459 if (isipv6) { 3460 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3461 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3462 } 3463 #endif 3464 #if defined(INET) && defined(INET6) 3465 else 3466 #endif 3467 #ifdef INET 3468 { 3469 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3470 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3471 } 3472 #endif 3473 3474 /* 3475 * No route to sender, stay with default mss and return. 3476 */ 3477 if (maxmtu == 0) { 3478 /* 3479 * In case we return early we need to initialize metrics 3480 * to a defined state as tcp_hc_get() would do for us 3481 * if there was no cache hit. 3482 */ 3483 if (metricptr != NULL) 3484 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3485 return; 3486 } 3487 3488 /* What have we got? */ 3489 switch (offer) { 3490 case 0: 3491 /* 3492 * Offer == 0 means that there was no MSS on the SYN 3493 * segment, in this case we use tcp_mssdflt as 3494 * already assigned to t_maxopd above. 3495 */ 3496 offer = tp->t_maxopd; 3497 break; 3498 3499 case -1: 3500 /* 3501 * Offer == -1 means that we didn't receive SYN yet. 3502 */ 3503 /* FALLTHROUGH */ 3504 3505 default: 3506 /* 3507 * Prevent DoS attack with too small MSS. Round up 3508 * to at least minmss. 3509 */ 3510 offer = max(offer, V_tcp_minmss); 3511 } 3512 3513 /* 3514 * rmx information is now retrieved from tcp_hostcache. 3515 */ 3516 tcp_hc_get(&inp->inp_inc, &metrics); 3517 if (metricptr != NULL) 3518 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3519 3520 /* 3521 * If there's a discovered mtu int tcp hostcache, use it 3522 * else, use the link mtu. 3523 */ 3524 if (metrics.rmx_mtu) 3525 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3526 else { 3527 #ifdef INET6 3528 if (isipv6) { 3529 mss = maxmtu - min_protoh; 3530 if (!V_path_mtu_discovery && 3531 !in6_localaddr(&inp->in6p_faddr)) 3532 mss = min(mss, V_tcp_v6mssdflt); 3533 } 3534 #endif 3535 #if defined(INET) && defined(INET6) 3536 else 3537 #endif 3538 #ifdef INET 3539 { 3540 mss = maxmtu - min_protoh; 3541 if (!V_path_mtu_discovery && 3542 !in_localaddr(inp->inp_faddr)) 3543 mss = min(mss, V_tcp_mssdflt); 3544 } 3545 #endif 3546 /* 3547 * XXX - The above conditional (mss = maxmtu - min_protoh) 3548 * probably violates the TCP spec. 3549 * The problem is that, since we don't know the 3550 * other end's MSS, we are supposed to use a conservative 3551 * default. But, if we do that, then MTU discovery will 3552 * never actually take place, because the conservative 3553 * default is much less than the MTUs typically seen 3554 * on the Internet today. For the moment, we'll sweep 3555 * this under the carpet. 3556 * 3557 * The conservative default might not actually be a problem 3558 * if the only case this occurs is when sending an initial 3559 * SYN with options and data to a host we've never talked 3560 * to before. Then, they will reply with an MSS value which 3561 * will get recorded and the new parameters should get 3562 * recomputed. For Further Study. 3563 */ 3564 } 3565 mss = min(mss, offer); 3566 3567 /* 3568 * Sanity check: make sure that maxopd will be large 3569 * enough to allow some data on segments even if the 3570 * all the option space is used (40bytes). Otherwise 3571 * funny things may happen in tcp_output. 3572 */ 3573 mss = max(mss, 64); 3574 3575 /* 3576 * maxopd stores the maximum length of data AND options 3577 * in a segment; maxseg is the amount of data in a normal 3578 * segment. We need to store this value (maxopd) apart 3579 * from maxseg, because now every segment carries options 3580 * and thus we normally have somewhat less data in segments. 3581 */ 3582 tp->t_maxopd = mss; 3583 3584 /* 3585 * origoffer==-1 indicates that no segments were received yet. 3586 * In this case we just guess. 3587 */ 3588 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3589 (origoffer == -1 || 3590 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3591 mss -= TCPOLEN_TSTAMP_APPA; 3592 3593 tp->t_maxseg = mss; 3594 } 3595 3596 void 3597 tcp_mss(struct tcpcb *tp, int offer) 3598 { 3599 int mss; 3600 u_long bufsize; 3601 struct inpcb *inp; 3602 struct socket *so; 3603 struct hc_metrics_lite metrics; 3604 struct tcp_ifcap cap; 3605 3606 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3607 3608 bzero(&cap, sizeof(cap)); 3609 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3610 3611 mss = tp->t_maxseg; 3612 inp = tp->t_inpcb; 3613 3614 /* 3615 * If there's a pipesize, change the socket buffer to that size, 3616 * don't change if sb_hiwat is different than default (then it 3617 * has been changed on purpose with setsockopt). 3618 * Make the socket buffers an integral number of mss units; 3619 * if the mss is larger than the socket buffer, decrease the mss. 3620 */ 3621 so = inp->inp_socket; 3622 SOCKBUF_LOCK(&so->so_snd); 3623 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3624 bufsize = metrics.rmx_sendpipe; 3625 else 3626 bufsize = so->so_snd.sb_hiwat; 3627 if (bufsize < mss) 3628 mss = bufsize; 3629 else { 3630 bufsize = roundup(bufsize, mss); 3631 if (bufsize > sb_max) 3632 bufsize = sb_max; 3633 if (bufsize > so->so_snd.sb_hiwat) 3634 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3635 } 3636 SOCKBUF_UNLOCK(&so->so_snd); 3637 tp->t_maxseg = mss; 3638 3639 SOCKBUF_LOCK(&so->so_rcv); 3640 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3641 bufsize = metrics.rmx_recvpipe; 3642 else 3643 bufsize = so->so_rcv.sb_hiwat; 3644 if (bufsize > mss) { 3645 bufsize = roundup(bufsize, mss); 3646 if (bufsize > sb_max) 3647 bufsize = sb_max; 3648 if (bufsize > so->so_rcv.sb_hiwat) 3649 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3650 } 3651 SOCKBUF_UNLOCK(&so->so_rcv); 3652 3653 /* Check the interface for TSO capabilities. */ 3654 if (cap.ifcap & CSUM_TSO) { 3655 tp->t_flags |= TF_TSO; 3656 tp->t_tsomax = cap.tsomax; 3657 } 3658 } 3659 3660 /* 3661 * Determine the MSS option to send on an outgoing SYN. 3662 */ 3663 int 3664 tcp_mssopt(struct in_conninfo *inc) 3665 { 3666 int mss = 0; 3667 u_long maxmtu = 0; 3668 u_long thcmtu = 0; 3669 size_t min_protoh; 3670 3671 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3672 3673 #ifdef INET6 3674 if (inc->inc_flags & INC_ISIPV6) { 3675 mss = V_tcp_v6mssdflt; 3676 maxmtu = tcp_maxmtu6(inc, NULL); 3677 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3678 } 3679 #endif 3680 #if defined(INET) && defined(INET6) 3681 else 3682 #endif 3683 #ifdef INET 3684 { 3685 mss = V_tcp_mssdflt; 3686 maxmtu = tcp_maxmtu(inc, NULL); 3687 min_protoh = sizeof(struct tcpiphdr); 3688 } 3689 #endif 3690 #if defined(INET6) || defined(INET) 3691 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3692 #endif 3693 3694 if (maxmtu && thcmtu) 3695 mss = min(maxmtu, thcmtu) - min_protoh; 3696 else if (maxmtu || thcmtu) 3697 mss = max(maxmtu, thcmtu) - min_protoh; 3698 3699 return (mss); 3700 } 3701 3702 3703 /* 3704 * On a partial ack arrives, force the retransmission of the 3705 * next unacknowledged segment. Do not clear tp->t_dupacks. 3706 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3707 * be started again. 3708 */ 3709 static void 3710 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3711 { 3712 tcp_seq onxt = tp->snd_nxt; 3713 u_long ocwnd = tp->snd_cwnd; 3714 3715 INP_WLOCK_ASSERT(tp->t_inpcb); 3716 3717 tcp_timer_activate(tp, TT_REXMT, 0); 3718 tp->t_rtttime = 0; 3719 tp->snd_nxt = th->th_ack; 3720 /* 3721 * Set snd_cwnd to one segment beyond acknowledged offset. 3722 * (tp->snd_una has not yet been updated when this function is called.) 3723 */ 3724 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3725 tp->t_flags |= TF_ACKNOW; 3726 (void) tcp_output(tp); 3727 tp->snd_cwnd = ocwnd; 3728 if (SEQ_GT(onxt, tp->snd_nxt)) 3729 tp->snd_nxt = onxt; 3730 /* 3731 * Partial window deflation. Relies on fact that tp->snd_una 3732 * not updated yet. 3733 */ 3734 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3735 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3736 else 3737 tp->snd_cwnd = 0; 3738 tp->snd_cwnd += tp->t_maxseg; 3739 } 3740