Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/netinet/tcp_input.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
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: releng/9.1/sys/netinet/tcp_input.c 271669 2014-09-16 09:50:19Z delphij $"); 52 53 #include "opt_ipfw.h" /* for ipfw_fwd */ 54 #include "opt_inet.h" 55 #include "opt_inet6.h" 56 #include "opt_ipsec.h" 57 #include "opt_tcpdebug.h" 58 59 #include <sys/param.h> 60 #include <sys/kernel.h> 61 #include <sys/hhook.h> 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/signalvar.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/sysctl.h> 70 #include <sys/syslog.h> 71 #include <sys/systm.h> 72 73 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 74 75 #include <vm/uma.h> 76 77 #include <net/if.h> 78 #include <net/route.h> 79 #include <net/vnet.h> 80 81 #define TCPSTATES /* for logging */ 82 83 #include <netinet/cc.h> 84 #include <netinet/in.h> 85 #include <netinet/in_pcb.h> 86 #include <netinet/in_systm.h> 87 #include <netinet/in_var.h> 88 #include <netinet/ip.h> 89 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 90 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 91 #include <netinet/ip_var.h> 92 #include <netinet/ip_options.h> 93 #include <netinet/ip6.h> 94 #include <netinet/icmp6.h> 95 #include <netinet6/in6_pcb.h> 96 #include <netinet6/ip6_var.h> 97 #include <netinet6/nd6.h> 98 #include <netinet/tcp_fsm.h> 99 #include <netinet/tcp_seq.h> 100 #include <netinet/tcp_timer.h> 101 #include <netinet/tcp_var.h> 102 #include <netinet6/tcp6_var.h> 103 #include <netinet/tcpip.h> 104 #include <netinet/tcp_syncache.h> 105 #ifdef TCPDEBUG 106 #include <netinet/tcp_debug.h> 107 #endif /* TCPDEBUG */ 108 109 #ifdef IPSEC 110 #include <netipsec/ipsec.h> 111 #include <netipsec/ipsec6.h> 112 #endif /*IPSEC*/ 113 114 #include <machine/in_cksum.h> 115 116 #include <security/mac/mac_framework.h> 117 118 const int tcprexmtthresh = 3; 119 120 VNET_DEFINE(struct tcpstat, tcpstat); 121 SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 122 &VNET_NAME(tcpstat), tcpstat, 123 "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 124 125 int tcp_log_in_vain = 0; 126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 127 &tcp_log_in_vain, 0, 128 "Log all incoming TCP segments to closed ports"); 129 130 VNET_DEFINE(int, blackhole) = 0; 131 #define V_blackhole VNET(blackhole) 132 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 133 &VNET_NAME(blackhole), 0, 134 "Do not send RST on segments to closed ports"); 135 136 VNET_DEFINE(int, tcp_delack_enabled) = 1; 137 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 138 &VNET_NAME(tcp_delack_enabled), 0, 139 "Delay ACK to try and piggyback it onto a data packet"); 140 141 VNET_DEFINE(int, drop_synfin) = 0; 142 #define V_drop_synfin VNET(drop_synfin) 143 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 144 &VNET_NAME(drop_synfin), 0, 145 "Drop TCP packets with SYN+FIN set"); 146 147 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 148 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 149 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 150 &VNET_NAME(tcp_do_rfc3042), 0, 151 "Enable RFC 3042 (Limited Transmit)"); 152 153 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 154 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 155 &VNET_NAME(tcp_do_rfc3390), 0, 156 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 157 158 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 159 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 160 &VNET_NAME(tcp_do_rfc3465), 0, 161 "Enable RFC 3465 (Appropriate Byte Counting)"); 162 163 VNET_DEFINE(int, tcp_abc_l_var) = 2; 164 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 165 &VNET_NAME(tcp_abc_l_var), 2, 166 "Cap the max cwnd increment during slow-start to this number of segments"); 167 168 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 169 170 VNET_DEFINE(int, tcp_do_ecn) = 0; 171 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 172 &VNET_NAME(tcp_do_ecn), 0, 173 "TCP ECN support"); 174 175 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 176 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 177 &VNET_NAME(tcp_ecn_maxretries), 0, 178 "Max retries before giving up on ECN"); 179 180 VNET_DEFINE(int, tcp_insecure_rst) = 0; 181 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 182 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 183 &VNET_NAME(tcp_insecure_rst), 0, 184 "Follow the old (insecure) criteria for accepting RST packets"); 185 186 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 187 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 188 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 189 &VNET_NAME(tcp_do_autorcvbuf), 0, 190 "Enable automatic receive buffer sizing"); 191 192 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 193 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 194 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 195 &VNET_NAME(tcp_autorcvbuf_inc), 0, 196 "Incrementor step size of automatic receive buffer"); 197 198 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 199 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 200 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 201 &VNET_NAME(tcp_autorcvbuf_max), 0, 202 "Max size of automatic receive buffer"); 203 204 VNET_DEFINE(struct inpcbhead, tcb); 205 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 206 VNET_DEFINE(struct inpcbinfo, tcbinfo); 207 208 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 209 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 210 struct socket *, struct tcpcb *, int, int, uint8_t, 211 int); 212 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 213 struct tcpcb *, int, int); 214 static void tcp_pulloutofband(struct socket *, 215 struct tcphdr *, struct mbuf *, int); 216 static void tcp_xmit_timer(struct tcpcb *, int); 217 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 218 static void inline tcp_fields_to_host(struct tcphdr *); 219 #ifdef TCP_SIGNATURE 220 static void inline tcp_fields_to_net(struct tcphdr *); 221 static int inline tcp_signature_verify_input(struct mbuf *, int, int, 222 int, struct tcpopt *, struct tcphdr *, u_int); 223 #endif 224 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 225 uint16_t type); 226 static void inline cc_conn_init(struct tcpcb *tp); 227 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 228 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 229 struct tcphdr *th, struct tcpopt *to); 230 231 /* 232 * Kernel module interface for updating tcpstat. The argument is an index 233 * into tcpstat treated as an array of u_long. While this encodes the 234 * general layout of tcpstat into the caller, it doesn't encode its location, 235 * so that future changes to add, for example, per-CPU stats support won't 236 * cause binary compatibility problems for kernel modules. 237 */ 238 void 239 kmod_tcpstat_inc(int statnum) 240 { 241 242 (*((u_long *)&V_tcpstat + statnum))++; 243 } 244 245 /* 246 * Wrapper for the TCP established input helper hook. 247 */ 248 static void inline 249 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 250 { 251 struct tcp_hhook_data hhook_data; 252 253 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 254 hhook_data.tp = tp; 255 hhook_data.th = th; 256 hhook_data.to = to; 257 258 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 259 tp->osd); 260 } 261 } 262 263 /* 264 * CC wrapper hook functions 265 */ 266 static void inline 267 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 268 { 269 INP_WLOCK_ASSERT(tp->t_inpcb); 270 271 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 272 if (tp->snd_cwnd == min(tp->snd_cwnd, tp->snd_wnd)) 273 tp->ccv->flags |= CCF_CWND_LIMITED; 274 else 275 tp->ccv->flags &= ~CCF_CWND_LIMITED; 276 277 if (type == CC_ACK) { 278 if (tp->snd_cwnd > tp->snd_ssthresh) { 279 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 280 V_tcp_abc_l_var * tp->t_maxseg); 281 if (tp->t_bytes_acked >= tp->snd_cwnd) { 282 tp->t_bytes_acked -= tp->snd_cwnd; 283 tp->ccv->flags |= CCF_ABC_SENTAWND; 284 } 285 } else { 286 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 287 tp->t_bytes_acked = 0; 288 } 289 } 290 291 if (CC_ALGO(tp)->ack_received != NULL) { 292 /* XXXLAS: Find a way to live without this */ 293 tp->ccv->curack = th->th_ack; 294 CC_ALGO(tp)->ack_received(tp->ccv, type); 295 } 296 } 297 298 static void inline 299 cc_conn_init(struct tcpcb *tp) 300 { 301 struct hc_metrics_lite metrics; 302 struct inpcb *inp = tp->t_inpcb; 303 int rtt; 304 #ifdef INET6 305 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 306 #endif 307 308 INP_WLOCK_ASSERT(tp->t_inpcb); 309 310 tcp_hc_get(&inp->inp_inc, &metrics); 311 312 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 313 tp->t_srtt = rtt; 314 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 315 TCPSTAT_INC(tcps_usedrtt); 316 if (metrics.rmx_rttvar) { 317 tp->t_rttvar = metrics.rmx_rttvar; 318 TCPSTAT_INC(tcps_usedrttvar); 319 } else { 320 /* default variation is +- 1 rtt */ 321 tp->t_rttvar = 322 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 323 } 324 TCPT_RANGESET(tp->t_rxtcur, 325 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 326 tp->t_rttmin, TCPTV_REXMTMAX); 327 } 328 if (metrics.rmx_ssthresh) { 329 /* 330 * There's some sort of gateway or interface 331 * buffer limit on the path. Use this to set 332 * the slow start threshhold, but set the 333 * threshold to no less than 2*mss. 334 */ 335 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 336 TCPSTAT_INC(tcps_usedssthresh); 337 } 338 339 /* 340 * Set the slow-start flight size depending on whether this 341 * is a local network or not. 342 * 343 * Extend this so we cache the cwnd too and retrieve it here. 344 * Make cwnd even bigger than RFC3390 suggests but only if we 345 * have previous experience with the remote host. Be careful 346 * not make cwnd bigger than remote receive window or our own 347 * send socket buffer. Maybe put some additional upper bound 348 * on the retrieved cwnd. Should do incremental updates to 349 * hostcache when cwnd collapses so next connection doesn't 350 * overloads the path again. 351 * 352 * XXXAO: Initializing the CWND from the hostcache is broken 353 * and in its current form not RFC conformant. It is disabled 354 * until fixed or removed entirely. 355 * 356 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 357 * We currently check only in syncache_socket for that. 358 */ 359 /* #define TCP_METRICS_CWND */ 360 #ifdef TCP_METRICS_CWND 361 if (metrics.rmx_cwnd) 362 tp->snd_cwnd = max(tp->t_maxseg, min(metrics.rmx_cwnd / 2, 363 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 364 else 365 #endif 366 if (V_tcp_do_rfc3390) 367 tp->snd_cwnd = min(4 * tp->t_maxseg, 368 max(2 * tp->t_maxseg, 4380)); 369 #ifdef INET6 370 else if (isipv6 && in6_localaddr(&inp->in6p_faddr)) 371 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local; 372 #endif 373 #if defined(INET) && defined(INET6) 374 else if (!isipv6 && in_localaddr(inp->inp_faddr)) 375 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local; 376 #endif 377 #ifdef INET 378 else if (in_localaddr(inp->inp_faddr)) 379 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local; 380 #endif 381 else 382 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz; 383 384 if (CC_ALGO(tp)->conn_init != NULL) 385 CC_ALGO(tp)->conn_init(tp->ccv); 386 } 387 388 void inline 389 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 390 { 391 INP_WLOCK_ASSERT(tp->t_inpcb); 392 393 switch(type) { 394 case CC_NDUPACK: 395 if (!IN_FASTRECOVERY(tp->t_flags)) { 396 tp->snd_recover = tp->snd_max; 397 if (tp->t_flags & TF_ECN_PERMIT) 398 tp->t_flags |= TF_ECN_SND_CWR; 399 } 400 break; 401 case CC_ECN: 402 if (!IN_CONGRECOVERY(tp->t_flags)) { 403 TCPSTAT_INC(tcps_ecn_rcwnd); 404 tp->snd_recover = tp->snd_max; 405 if (tp->t_flags & TF_ECN_PERMIT) 406 tp->t_flags |= TF_ECN_SND_CWR; 407 } 408 break; 409 case CC_RTO: 410 tp->t_dupacks = 0; 411 tp->t_bytes_acked = 0; 412 EXIT_RECOVERY(tp->t_flags); 413 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 414 tp->t_maxseg) * tp->t_maxseg; 415 tp->snd_cwnd = tp->t_maxseg; 416 break; 417 case CC_RTO_ERR: 418 TCPSTAT_INC(tcps_sndrexmitbad); 419 /* RTO was unnecessary, so reset everything. */ 420 tp->snd_cwnd = tp->snd_cwnd_prev; 421 tp->snd_ssthresh = tp->snd_ssthresh_prev; 422 tp->snd_recover = tp->snd_recover_prev; 423 if (tp->t_flags & TF_WASFRECOVERY) 424 ENTER_FASTRECOVERY(tp->t_flags); 425 if (tp->t_flags & TF_WASCRECOVERY) 426 ENTER_CONGRECOVERY(tp->t_flags); 427 tp->snd_nxt = tp->snd_max; 428 tp->t_flags &= ~TF_PREVVALID; 429 tp->t_badrxtwin = 0; 430 break; 431 } 432 433 if (CC_ALGO(tp)->cong_signal != NULL) { 434 if (th != NULL) 435 tp->ccv->curack = th->th_ack; 436 CC_ALGO(tp)->cong_signal(tp->ccv, type); 437 } 438 } 439 440 static void inline 441 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 442 { 443 INP_WLOCK_ASSERT(tp->t_inpcb); 444 445 /* XXXLAS: KASSERT that we're in recovery? */ 446 447 if (CC_ALGO(tp)->post_recovery != NULL) { 448 tp->ccv->curack = th->th_ack; 449 CC_ALGO(tp)->post_recovery(tp->ccv); 450 } 451 /* XXXLAS: EXIT_RECOVERY ? */ 452 tp->t_bytes_acked = 0; 453 } 454 455 static inline void 456 tcp_fields_to_host(struct tcphdr *th) 457 { 458 459 th->th_seq = ntohl(th->th_seq); 460 th->th_ack = ntohl(th->th_ack); 461 th->th_win = ntohs(th->th_win); 462 th->th_urp = ntohs(th->th_urp); 463 } 464 465 #ifdef TCP_SIGNATURE 466 static inline void 467 tcp_fields_to_net(struct tcphdr *th) 468 { 469 470 th->th_seq = htonl(th->th_seq); 471 th->th_ack = htonl(th->th_ack); 472 th->th_win = htons(th->th_win); 473 th->th_urp = htons(th->th_urp); 474 } 475 476 static inline int 477 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 478 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 479 { 480 int ret; 481 482 tcp_fields_to_net(th); 483 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 484 tcp_fields_to_host(th); 485 return (ret); 486 } 487 #endif 488 489 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 490 #ifdef INET6 491 #define ND6_HINT(tp) \ 492 do { \ 493 if ((tp) && (tp)->t_inpcb && \ 494 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 495 nd6_nud_hint(NULL, NULL, 0); \ 496 } while (0) 497 #else 498 #define ND6_HINT(tp) 499 #endif 500 501 /* 502 * Indicate whether this ack should be delayed. We can delay the ack if 503 * - there is no delayed ack timer in progress and 504 * - our last ack wasn't a 0-sized window. We never want to delay 505 * the ack that opens up a 0-sized window and 506 * - delayed acks are enabled or 507 * - this is a half-synchronized T/TCP connection. 508 */ 509 #define DELAY_ACK(tp) \ 510 ((!tcp_timer_active(tp, TT_DELACK) && \ 511 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 512 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 513 514 /* 515 * TCP input handling is split into multiple parts: 516 * tcp6_input is a thin wrapper around tcp_input for the extended 517 * ip6_protox[] call format in ip6_input 518 * tcp_input handles primary segment validation, inpcb lookup and 519 * SYN processing on listen sockets 520 * tcp_do_segment processes the ACK and text of the segment for 521 * establishing, established and closing connections 522 */ 523 #ifdef INET6 524 int 525 tcp6_input(struct mbuf **mp, int *offp, int proto) 526 { 527 struct mbuf *m = *mp; 528 struct in6_ifaddr *ia6; 529 530 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 531 532 /* 533 * draft-itojun-ipv6-tcp-to-anycast 534 * better place to put this in? 535 */ 536 ia6 = ip6_getdstifaddr(m); 537 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 538 struct ip6_hdr *ip6; 539 540 ifa_free(&ia6->ia_ifa); 541 ip6 = mtod(m, struct ip6_hdr *); 542 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 543 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 544 return IPPROTO_DONE; 545 } 546 if (ia6) 547 ifa_free(&ia6->ia_ifa); 548 549 tcp_input(m, *offp); 550 return IPPROTO_DONE; 551 } 552 #endif /* INET6 */ 553 554 void 555 tcp_input(struct mbuf *m, int off0) 556 { 557 struct tcphdr *th = NULL; 558 struct ip *ip = NULL; 559 #ifdef INET 560 struct ipovly *ipov; 561 #endif 562 struct inpcb *inp = NULL; 563 struct tcpcb *tp = NULL; 564 struct socket *so = NULL; 565 u_char *optp = NULL; 566 int optlen = 0; 567 #ifdef INET 568 int len; 569 #endif 570 int tlen = 0, off; 571 int drop_hdrlen; 572 int thflags; 573 int rstreason = 0; /* For badport_bandlim accounting purposes */ 574 #ifdef TCP_SIGNATURE 575 uint8_t sig_checked = 0; 576 #endif 577 uint8_t iptos = 0; 578 #ifdef IPFIREWALL_FORWARD 579 struct m_tag *fwd_tag; 580 #endif 581 #ifdef INET6 582 struct ip6_hdr *ip6 = NULL; 583 int isipv6; 584 #else 585 const void *ip6 = NULL; 586 #endif /* INET6 */ 587 struct tcpopt to; /* options in this segment */ 588 char *s = NULL; /* address and port logging */ 589 int ti_locked; 590 #define TI_UNLOCKED 1 591 #define TI_WLOCKED 2 592 593 #ifdef TCPDEBUG 594 /* 595 * The size of tcp_saveipgen must be the size of the max ip header, 596 * now IPv6. 597 */ 598 u_char tcp_saveipgen[IP6_HDR_LEN]; 599 struct tcphdr tcp_savetcp; 600 short ostate = 0; 601 #endif 602 603 #ifdef INET6 604 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 605 #endif 606 607 to.to_flags = 0; 608 TCPSTAT_INC(tcps_rcvtotal); 609 610 #ifdef INET6 611 if (isipv6) { 612 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 613 614 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 615 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 616 if (m == NULL) { 617 TCPSTAT_INC(tcps_rcvshort); 618 return; 619 } 620 } 621 622 ip6 = mtod(m, struct ip6_hdr *); 623 th = (struct tcphdr *)((caddr_t)ip6 + off0); 624 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 625 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 626 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 627 th->th_sum = m->m_pkthdr.csum_data; 628 else 629 th->th_sum = in6_cksum_pseudo(ip6, tlen, 630 IPPROTO_TCP, m->m_pkthdr.csum_data); 631 th->th_sum ^= 0xffff; 632 } else 633 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 634 if (th->th_sum) { 635 TCPSTAT_INC(tcps_rcvbadsum); 636 goto drop; 637 } 638 639 /* 640 * Be proactive about unspecified IPv6 address in source. 641 * As we use all-zero to indicate unbounded/unconnected pcb, 642 * unspecified IPv6 address can be used to confuse us. 643 * 644 * Note that packets with unspecified IPv6 destination is 645 * already dropped in ip6_input. 646 */ 647 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 648 /* XXX stat */ 649 goto drop; 650 } 651 } 652 #endif 653 #if defined(INET) && defined(INET6) 654 else 655 #endif 656 #ifdef INET 657 { 658 /* 659 * Get IP and TCP header together in first mbuf. 660 * Note: IP leaves IP header in first mbuf. 661 */ 662 if (off0 > sizeof (struct ip)) { 663 ip_stripoptions(m, (struct mbuf *)0); 664 off0 = sizeof(struct ip); 665 } 666 if (m->m_len < sizeof (struct tcpiphdr)) { 667 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 668 == NULL) { 669 TCPSTAT_INC(tcps_rcvshort); 670 return; 671 } 672 } 673 ip = mtod(m, struct ip *); 674 ipov = (struct ipovly *)ip; 675 th = (struct tcphdr *)((caddr_t)ip + off0); 676 tlen = ip->ip_len; 677 678 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 679 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 680 th->th_sum = m->m_pkthdr.csum_data; 681 else 682 th->th_sum = in_pseudo(ip->ip_src.s_addr, 683 ip->ip_dst.s_addr, 684 htonl(m->m_pkthdr.csum_data + 685 ip->ip_len + 686 IPPROTO_TCP)); 687 th->th_sum ^= 0xffff; 688 #ifdef TCPDEBUG 689 ipov->ih_len = (u_short)tlen; 690 ipov->ih_len = htons(ipov->ih_len); 691 #endif 692 } else { 693 /* 694 * Checksum extended TCP header and data. 695 */ 696 len = sizeof (struct ip) + tlen; 697 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 698 ipov->ih_len = (u_short)tlen; 699 ipov->ih_len = htons(ipov->ih_len); 700 th->th_sum = in_cksum(m, len); 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 ipov = (struct ipovly *)ip; 753 th = (struct tcphdr *)((caddr_t)ip + off0); 754 } 755 } 756 #endif 757 optlen = off - sizeof (struct tcphdr); 758 optp = (u_char *)(th + 1); 759 } 760 thflags = th->th_flags; 761 762 /* 763 * Convert TCP protocol specific fields to host format. 764 */ 765 tcp_fields_to_host(th); 766 767 /* 768 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 769 */ 770 drop_hdrlen = off0 + off; 771 772 /* 773 * Locate pcb for segment; if we're likely to add or remove a 774 * connection then first acquire pcbinfo lock. There are two cases 775 * where we might discover later we need a write lock despite the 776 * flags: ACKs moving a connection out of the syncache, and ACKs for 777 * a connection in TIMEWAIT. 778 */ 779 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 780 INP_INFO_WLOCK(&V_tcbinfo); 781 ti_locked = TI_WLOCKED; 782 } else 783 ti_locked = TI_UNLOCKED; 784 785 findpcb: 786 #ifdef INVARIANTS 787 if (ti_locked == TI_WLOCKED) { 788 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 789 } else { 790 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 791 } 792 #endif 793 794 #ifdef IPFIREWALL_FORWARD 795 /* 796 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 797 */ 798 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 799 #endif /* IPFIREWALL_FORWARD */ 800 801 #ifdef INET6 802 #ifdef IPFIREWALL_FORWARD 803 if (isipv6 && fwd_tag != NULL) { 804 struct sockaddr_in6 *next_hop6; 805 806 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 807 /* 808 * Transparently forwarded. Pretend to be the destination. 809 * Already got one like this? 810 */ 811 inp = in6_pcblookup_mbuf(&V_tcbinfo, 812 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 813 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 814 if (!inp) { 815 /* 816 * It's new. Try to find the ambushing socket. 817 * Because we've rewritten the destination address, 818 * any hardware-generated hash is ignored. 819 */ 820 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 821 th->th_sport, &next_hop6->sin6_addr, 822 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 823 th->th_dport, INPLOOKUP_WILDCARD | 824 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 825 } 826 /* Remove the tag from the packet. We don't need it anymore. */ 827 m_tag_delete(m, fwd_tag); 828 } else 829 #endif /* IPFIREWALL_FORWARD */ 830 if (isipv6) { 831 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 832 th->th_sport, &ip6->ip6_dst, th->th_dport, 833 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 834 m->m_pkthdr.rcvif, m); 835 } 836 #endif /* INET6 */ 837 #if defined(INET6) && defined(INET) 838 else 839 #endif 840 #ifdef INET 841 #ifdef IPFIREWALL_FORWARD 842 if (fwd_tag != NULL) { 843 struct sockaddr_in *next_hop; 844 845 next_hop = (struct sockaddr_in *)(fwd_tag+1); 846 /* 847 * Transparently forwarded. Pretend to be the destination. 848 * already got one like this? 849 */ 850 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 851 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 852 m->m_pkthdr.rcvif, m); 853 if (!inp) { 854 /* 855 * It's new. Try to find the ambushing socket. 856 * Because we've rewritten the destination address, 857 * any hardware-generated hash is ignored. 858 */ 859 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 860 th->th_sport, next_hop->sin_addr, 861 next_hop->sin_port ? ntohs(next_hop->sin_port) : 862 th->th_dport, INPLOOKUP_WILDCARD | 863 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 864 } 865 /* Remove the tag from the packet. We don't need it anymore. */ 866 m_tag_delete(m, fwd_tag); 867 } else 868 #endif /* IPFIREWALL_FORWARD */ 869 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 870 th->th_sport, ip->ip_dst, th->th_dport, 871 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 872 m->m_pkthdr.rcvif, m); 873 #endif /* INET */ 874 875 /* 876 * If the INPCB does not exist then all data in the incoming 877 * segment is discarded and an appropriate RST is sent back. 878 * XXX MRT Send RST using which routing table? 879 */ 880 if (inp == NULL) { 881 /* 882 * Log communication attempts to ports that are not 883 * in use. 884 */ 885 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 886 tcp_log_in_vain == 2) { 887 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 888 log(LOG_INFO, "%s; %s: Connection attempt " 889 "to closed port\n", s, __func__); 890 } 891 /* 892 * When blackholing do not respond with a RST but 893 * completely ignore the segment and drop it. 894 */ 895 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 896 V_blackhole == 2) 897 goto dropunlock; 898 899 rstreason = BANDLIM_RST_CLOSEDPORT; 900 goto dropwithreset; 901 } 902 INP_WLOCK_ASSERT(inp); 903 if (!(inp->inp_flags & INP_HW_FLOWID) 904 && (m->m_flags & M_FLOWID) 905 && ((inp->inp_socket == NULL) 906 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 907 inp->inp_flags |= INP_HW_FLOWID; 908 inp->inp_flags &= ~INP_SW_FLOWID; 909 inp->inp_flowid = m->m_pkthdr.flowid; 910 } 911 #ifdef IPSEC 912 #ifdef INET6 913 if (isipv6 && ipsec6_in_reject(m, inp)) { 914 V_ipsec6stat.in_polvio++; 915 goto dropunlock; 916 } else 917 #endif /* INET6 */ 918 if (ipsec4_in_reject(m, inp) != 0) { 919 V_ipsec4stat.in_polvio++; 920 goto dropunlock; 921 } 922 #endif /* IPSEC */ 923 924 /* 925 * Check the minimum TTL for socket. 926 */ 927 if (inp->inp_ip_minttl != 0) { 928 #ifdef INET6 929 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 930 goto dropunlock; 931 else 932 #endif 933 if (inp->inp_ip_minttl > ip->ip_ttl) 934 goto dropunlock; 935 } 936 937 /* 938 * A previous connection in TIMEWAIT state is supposed to catch stray 939 * or duplicate segments arriving late. If this segment was a 940 * legitimate new connection attempt the old INPCB gets removed and 941 * we can try again to find a listening socket. 942 * 943 * At this point, due to earlier optimism, we may hold only an inpcb 944 * lock, and not the inpcbinfo write lock. If so, we need to try to 945 * acquire it, or if that fails, acquire a reference on the inpcb, 946 * drop all locks, acquire a global write lock, and then re-acquire 947 * the inpcb lock. We may at that point discover that another thread 948 * has tried to free the inpcb, in which case we need to loop back 949 * and try to find a new inpcb to deliver to. 950 * 951 * XXXRW: It may be time to rethink timewait locking. 952 */ 953 relocked: 954 if (inp->inp_flags & INP_TIMEWAIT) { 955 if (ti_locked == TI_UNLOCKED) { 956 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 957 in_pcbref(inp); 958 INP_WUNLOCK(inp); 959 INP_INFO_WLOCK(&V_tcbinfo); 960 ti_locked = TI_WLOCKED; 961 INP_WLOCK(inp); 962 if (in_pcbrele_wlocked(inp)) { 963 inp = NULL; 964 goto findpcb; 965 } 966 } else 967 ti_locked = TI_WLOCKED; 968 } 969 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 970 971 if (thflags & TH_SYN) 972 tcp_dooptions(&to, optp, optlen, TO_SYN); 973 /* 974 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 975 */ 976 if (tcp_twcheck(inp, &to, th, m, tlen)) 977 goto findpcb; 978 INP_INFO_WUNLOCK(&V_tcbinfo); 979 return; 980 } 981 /* 982 * The TCPCB may no longer exist if the connection is winding 983 * down or it is in the CLOSED state. Either way we drop the 984 * segment and send an appropriate response. 985 */ 986 tp = intotcpcb(inp); 987 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 988 rstreason = BANDLIM_RST_CLOSEDPORT; 989 goto dropwithreset; 990 } 991 992 /* 993 * We've identified a valid inpcb, but it could be that we need an 994 * inpcbinfo write lock but don't hold it. In this case, attempt to 995 * acquire using the same strategy as the TIMEWAIT case above. If we 996 * relock, we have to jump back to 'relocked' as the connection might 997 * now be in TIMEWAIT. 998 */ 999 #ifdef INVARIANTS 1000 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 1001 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1002 #endif 1003 if (tp->t_state != TCPS_ESTABLISHED) { 1004 if (ti_locked == TI_UNLOCKED) { 1005 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 1006 in_pcbref(inp); 1007 INP_WUNLOCK(inp); 1008 INP_INFO_WLOCK(&V_tcbinfo); 1009 ti_locked = TI_WLOCKED; 1010 INP_WLOCK(inp); 1011 if (in_pcbrele_wlocked(inp)) { 1012 inp = NULL; 1013 goto findpcb; 1014 } 1015 goto relocked; 1016 } else 1017 ti_locked = TI_WLOCKED; 1018 } 1019 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1020 } 1021 1022 #ifdef MAC 1023 INP_WLOCK_ASSERT(inp); 1024 if (mac_inpcb_check_deliver(inp, m)) 1025 goto dropunlock; 1026 #endif 1027 so = inp->inp_socket; 1028 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1029 #ifdef TCPDEBUG 1030 if (so->so_options & SO_DEBUG) { 1031 ostate = tp->t_state; 1032 #ifdef INET6 1033 if (isipv6) { 1034 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1035 } else 1036 #endif 1037 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1038 tcp_savetcp = *th; 1039 } 1040 #endif /* TCPDEBUG */ 1041 /* 1042 * When the socket is accepting connections (the INPCB is in LISTEN 1043 * state) we look into the SYN cache if this is a new connection 1044 * attempt or the completion of a previous one. Because listen 1045 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1046 * held in this case. 1047 */ 1048 if (so->so_options & SO_ACCEPTCONN) { 1049 struct in_conninfo inc; 1050 1051 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1052 "tp not listening", __func__)); 1053 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1054 1055 bzero(&inc, sizeof(inc)); 1056 #ifdef INET6 1057 if (isipv6) { 1058 inc.inc_flags |= INC_ISIPV6; 1059 inc.inc6_faddr = ip6->ip6_src; 1060 inc.inc6_laddr = ip6->ip6_dst; 1061 } else 1062 #endif 1063 { 1064 inc.inc_faddr = ip->ip_src; 1065 inc.inc_laddr = ip->ip_dst; 1066 } 1067 inc.inc_fport = th->th_sport; 1068 inc.inc_lport = th->th_dport; 1069 inc.inc_fibnum = so->so_fibnum; 1070 1071 /* 1072 * Check for an existing connection attempt in syncache if 1073 * the flag is only ACK. A successful lookup creates a new 1074 * socket appended to the listen queue in SYN_RECEIVED state. 1075 */ 1076 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1077 /* 1078 * Parse the TCP options here because 1079 * syncookies need access to the reflected 1080 * timestamp. 1081 */ 1082 tcp_dooptions(&to, optp, optlen, 0); 1083 /* 1084 * NB: syncache_expand() doesn't unlock 1085 * inp and tcpinfo locks. 1086 */ 1087 if (!syncache_expand(&inc, &to, th, &so, m)) { 1088 /* 1089 * No syncache entry or ACK was not 1090 * for our SYN/ACK. Send a RST. 1091 * NB: syncache did its own logging 1092 * of the failure cause. 1093 */ 1094 rstreason = BANDLIM_RST_OPENPORT; 1095 goto dropwithreset; 1096 } 1097 if (so == NULL) { 1098 /* 1099 * We completed the 3-way handshake 1100 * but could not allocate a socket 1101 * either due to memory shortage, 1102 * listen queue length limits or 1103 * global socket limits. Send RST 1104 * or wait and have the remote end 1105 * retransmit the ACK for another 1106 * try. 1107 */ 1108 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1109 log(LOG_DEBUG, "%s; %s: Listen socket: " 1110 "Socket allocation failed due to " 1111 "limits or memory shortage, %s\n", 1112 s, __func__, 1113 V_tcp_sc_rst_sock_fail ? 1114 "sending RST" : "try again"); 1115 if (V_tcp_sc_rst_sock_fail) { 1116 rstreason = BANDLIM_UNLIMITED; 1117 goto dropwithreset; 1118 } else 1119 goto dropunlock; 1120 } 1121 /* 1122 * Socket is created in state SYN_RECEIVED. 1123 * Unlock the listen socket, lock the newly 1124 * created socket and update the tp variable. 1125 */ 1126 INP_WUNLOCK(inp); /* listen socket */ 1127 inp = sotoinpcb(so); 1128 INP_WLOCK(inp); /* new connection */ 1129 tp = intotcpcb(inp); 1130 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1131 ("%s: ", __func__)); 1132 #ifdef TCP_SIGNATURE 1133 if (sig_checked == 0) { 1134 tcp_dooptions(&to, optp, optlen, 1135 (thflags & TH_SYN) ? TO_SYN : 0); 1136 if (!tcp_signature_verify_input(m, off0, tlen, 1137 optlen, &to, th, tp->t_flags)) { 1138 1139 /* 1140 * In SYN_SENT state if it receives an 1141 * RST, it is allowed for further 1142 * processing. 1143 */ 1144 if ((thflags & TH_RST) == 0 || 1145 (tp->t_state == TCPS_SYN_SENT) == 0) 1146 goto dropunlock; 1147 } 1148 sig_checked = 1; 1149 } 1150 #endif 1151 1152 /* 1153 * Process the segment and the data it 1154 * contains. tcp_do_segment() consumes 1155 * the mbuf chain and unlocks the inpcb. 1156 */ 1157 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1158 iptos, ti_locked); 1159 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1160 return; 1161 } 1162 /* 1163 * Segment flag validation for new connection attempts: 1164 * 1165 * Our (SYN|ACK) response was rejected. 1166 * Check with syncache and remove entry to prevent 1167 * retransmits. 1168 * 1169 * NB: syncache_chkrst does its own logging of failure 1170 * causes. 1171 */ 1172 if (thflags & TH_RST) { 1173 syncache_chkrst(&inc, th); 1174 goto dropunlock; 1175 } 1176 /* 1177 * We can't do anything without SYN. 1178 */ 1179 if ((thflags & TH_SYN) == 0) { 1180 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1181 log(LOG_DEBUG, "%s; %s: Listen socket: " 1182 "SYN is missing, segment ignored\n", 1183 s, __func__); 1184 TCPSTAT_INC(tcps_badsyn); 1185 goto dropunlock; 1186 } 1187 /* 1188 * (SYN|ACK) is bogus on a listen socket. 1189 */ 1190 if (thflags & TH_ACK) { 1191 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1192 log(LOG_DEBUG, "%s; %s: Listen socket: " 1193 "SYN|ACK invalid, segment rejected\n", 1194 s, __func__); 1195 syncache_badack(&inc); /* XXX: Not needed! */ 1196 TCPSTAT_INC(tcps_badsyn); 1197 rstreason = BANDLIM_RST_OPENPORT; 1198 goto dropwithreset; 1199 } 1200 /* 1201 * If the drop_synfin option is enabled, drop all 1202 * segments with both the SYN and FIN bits set. 1203 * This prevents e.g. nmap from identifying the 1204 * TCP/IP stack. 1205 * XXX: Poor reasoning. nmap has other methods 1206 * and is constantly refining its stack detection 1207 * strategies. 1208 * XXX: This is a violation of the TCP specification 1209 * and was used by RFC1644. 1210 */ 1211 if ((thflags & TH_FIN) && V_drop_synfin) { 1212 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1213 log(LOG_DEBUG, "%s; %s: Listen socket: " 1214 "SYN|FIN segment ignored (based on " 1215 "sysctl setting)\n", s, __func__); 1216 TCPSTAT_INC(tcps_badsyn); 1217 goto dropunlock; 1218 } 1219 /* 1220 * Segment's flags are (SYN) or (SYN|FIN). 1221 * 1222 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1223 * as they do not affect the state of the TCP FSM. 1224 * The data pointed to by TH_URG and th_urp is ignored. 1225 */ 1226 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1227 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1228 KASSERT(thflags & (TH_SYN), 1229 ("%s: Listen socket: TH_SYN not set", __func__)); 1230 #ifdef INET6 1231 /* 1232 * If deprecated address is forbidden, 1233 * we do not accept SYN to deprecated interface 1234 * address to prevent any new inbound connection from 1235 * getting established. 1236 * When we do not accept SYN, we send a TCP RST, 1237 * with deprecated source address (instead of dropping 1238 * it). We compromise it as it is much better for peer 1239 * to send a RST, and RST will be the final packet 1240 * for the exchange. 1241 * 1242 * If we do not forbid deprecated addresses, we accept 1243 * the SYN packet. RFC2462 does not suggest dropping 1244 * SYN in this case. 1245 * If we decipher RFC2462 5.5.4, it says like this: 1246 * 1. use of deprecated addr with existing 1247 * communication is okay - "SHOULD continue to be 1248 * used" 1249 * 2. use of it with new communication: 1250 * (2a) "SHOULD NOT be used if alternate address 1251 * with sufficient scope is available" 1252 * (2b) nothing mentioned otherwise. 1253 * Here we fall into (2b) case as we have no choice in 1254 * our source address selection - we must obey the peer. 1255 * 1256 * The wording in RFC2462 is confusing, and there are 1257 * multiple description text for deprecated address 1258 * handling - worse, they are not exactly the same. 1259 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1260 */ 1261 if (isipv6 && !V_ip6_use_deprecated) { 1262 struct in6_ifaddr *ia6; 1263 1264 ia6 = ip6_getdstifaddr(m); 1265 if (ia6 != NULL && 1266 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1267 ifa_free(&ia6->ia_ifa); 1268 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1269 log(LOG_DEBUG, "%s; %s: Listen socket: " 1270 "Connection attempt to deprecated " 1271 "IPv6 address rejected\n", 1272 s, __func__); 1273 rstreason = BANDLIM_RST_OPENPORT; 1274 goto dropwithreset; 1275 } 1276 if (ia6) 1277 ifa_free(&ia6->ia_ifa); 1278 } 1279 #endif /* INET6 */ 1280 /* 1281 * Basic sanity checks on incoming SYN requests: 1282 * Don't respond if the destination is a link layer 1283 * broadcast according to RFC1122 4.2.3.10, p. 104. 1284 * If it is from this socket it must be forged. 1285 * Don't respond if the source or destination is a 1286 * global or subnet broad- or multicast address. 1287 * Note that it is quite possible to receive unicast 1288 * link-layer packets with a broadcast IP address. Use 1289 * in_broadcast() to find them. 1290 */ 1291 if (m->m_flags & (M_BCAST|M_MCAST)) { 1292 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1293 log(LOG_DEBUG, "%s; %s: Listen socket: " 1294 "Connection attempt from broad- or multicast " 1295 "link layer address ignored\n", s, __func__); 1296 goto dropunlock; 1297 } 1298 #ifdef INET6 1299 if (isipv6) { 1300 if (th->th_dport == th->th_sport && 1301 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1302 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1303 log(LOG_DEBUG, "%s; %s: Listen socket: " 1304 "Connection attempt to/from self " 1305 "ignored\n", s, __func__); 1306 goto dropunlock; 1307 } 1308 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1309 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1310 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1311 log(LOG_DEBUG, "%s; %s: Listen socket: " 1312 "Connection attempt from/to multicast " 1313 "address ignored\n", s, __func__); 1314 goto dropunlock; 1315 } 1316 } 1317 #endif 1318 #if defined(INET) && defined(INET6) 1319 else 1320 #endif 1321 #ifdef INET 1322 { 1323 if (th->th_dport == th->th_sport && 1324 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1326 log(LOG_DEBUG, "%s; %s: Listen socket: " 1327 "Connection attempt from/to self " 1328 "ignored\n", s, __func__); 1329 goto dropunlock; 1330 } 1331 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1332 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1333 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1334 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1335 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1336 log(LOG_DEBUG, "%s; %s: Listen socket: " 1337 "Connection attempt from/to broad- " 1338 "or multicast address ignored\n", 1339 s, __func__); 1340 goto dropunlock; 1341 } 1342 } 1343 #endif 1344 /* 1345 * SYN appears to be valid. Create compressed TCP state 1346 * for syncache. 1347 */ 1348 #ifdef TCPDEBUG 1349 if (so->so_options & SO_DEBUG) 1350 tcp_trace(TA_INPUT, ostate, tp, 1351 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1352 #endif 1353 tcp_dooptions(&to, optp, optlen, TO_SYN); 1354 syncache_add(&inc, &to, th, inp, &so, m); 1355 /* 1356 * Entry added to syncache and mbuf consumed. 1357 * Everything already unlocked by syncache_add(). 1358 */ 1359 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1360 return; 1361 } 1362 1363 #ifdef TCP_SIGNATURE 1364 if (sig_checked == 0) { 1365 tcp_dooptions(&to, optp, optlen, 1366 (thflags & TH_SYN) ? TO_SYN : 0); 1367 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1368 th, tp->t_flags)) { 1369 1370 /* 1371 * In SYN_SENT state if it receives an RST, it is 1372 * allowed for further processing. 1373 */ 1374 if ((thflags & TH_RST) == 0 || 1375 (tp->t_state == TCPS_SYN_SENT) == 0) 1376 goto dropunlock; 1377 } 1378 sig_checked = 1; 1379 } 1380 #endif 1381 1382 /* 1383 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1384 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1385 * the inpcb, and unlocks pcbinfo. 1386 */ 1387 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1388 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1389 return; 1390 1391 dropwithreset: 1392 if (ti_locked == TI_WLOCKED) { 1393 INP_INFO_WUNLOCK(&V_tcbinfo); 1394 ti_locked = TI_UNLOCKED; 1395 } 1396 #ifdef INVARIANTS 1397 else { 1398 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1399 "ti_locked: %d", __func__, ti_locked)); 1400 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1401 } 1402 #endif 1403 1404 if (inp != NULL) { 1405 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1406 INP_WUNLOCK(inp); 1407 } else 1408 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1409 m = NULL; /* mbuf chain got consumed. */ 1410 goto drop; 1411 1412 dropunlock: 1413 if (ti_locked == TI_WLOCKED) { 1414 INP_INFO_WUNLOCK(&V_tcbinfo); 1415 ti_locked = TI_UNLOCKED; 1416 } 1417 #ifdef INVARIANTS 1418 else { 1419 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1420 "ti_locked: %d", __func__, ti_locked)); 1421 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1422 } 1423 #endif 1424 1425 if (inp != NULL) 1426 INP_WUNLOCK(inp); 1427 1428 drop: 1429 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1430 if (s != NULL) 1431 free(s, M_TCPLOG); 1432 if (m != NULL) 1433 m_freem(m); 1434 } 1435 1436 static void 1437 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1438 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1439 int ti_locked) 1440 { 1441 int thflags, acked, ourfinisacked, needoutput = 0; 1442 int rstreason, todrop, win; 1443 u_long tiwin; 1444 struct tcpopt to; 1445 1446 #ifdef TCPDEBUG 1447 /* 1448 * The size of tcp_saveipgen must be the size of the max ip header, 1449 * now IPv6. 1450 */ 1451 u_char tcp_saveipgen[IP6_HDR_LEN]; 1452 struct tcphdr tcp_savetcp; 1453 short ostate = 0; 1454 #endif 1455 thflags = th->th_flags; 1456 tp->sackhint.last_sack_ack = 0; 1457 1458 /* 1459 * If this is either a state-changing packet or current state isn't 1460 * established, we require a write lock on tcbinfo. Otherwise, we 1461 * allow either a read lock or a write lock, as we may have acquired 1462 * a write lock due to a race. 1463 * 1464 * Require a global write lock for SYN/FIN/RST segments or 1465 * non-established connections; otherwise accept either a read or 1466 * write lock, as we may have conservatively acquired a write lock in 1467 * certain cases in tcp_input() (is this still true?). Currently we 1468 * will never enter with no lock, so we try to drop it quickly in the 1469 * common pure ack/pure data cases. 1470 */ 1471 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1472 tp->t_state != TCPS_ESTABLISHED) { 1473 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1474 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1475 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1476 } else { 1477 #ifdef INVARIANTS 1478 if (ti_locked == TI_WLOCKED) 1479 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1480 else { 1481 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1482 "ti_locked: %d", __func__, ti_locked)); 1483 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1484 } 1485 #endif 1486 } 1487 INP_WLOCK_ASSERT(tp->t_inpcb); 1488 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1489 __func__)); 1490 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1491 __func__)); 1492 1493 /* 1494 * Segment received on connection. 1495 * Reset idle time and keep-alive timer. 1496 * XXX: This should be done after segment 1497 * validation to ignore broken/spoofed segs. 1498 */ 1499 tp->t_rcvtime = ticks; 1500 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1501 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1502 1503 /* 1504 * Unscale the window into a 32-bit value. 1505 * For the SYN_SENT state the scale is zero. 1506 */ 1507 tiwin = th->th_win << tp->snd_scale; 1508 1509 /* 1510 * TCP ECN processing. 1511 */ 1512 if (tp->t_flags & TF_ECN_PERMIT) { 1513 if (thflags & TH_CWR) 1514 tp->t_flags &= ~TF_ECN_SND_ECE; 1515 switch (iptos & IPTOS_ECN_MASK) { 1516 case IPTOS_ECN_CE: 1517 tp->t_flags |= TF_ECN_SND_ECE; 1518 TCPSTAT_INC(tcps_ecn_ce); 1519 break; 1520 case IPTOS_ECN_ECT0: 1521 TCPSTAT_INC(tcps_ecn_ect0); 1522 break; 1523 case IPTOS_ECN_ECT1: 1524 TCPSTAT_INC(tcps_ecn_ect1); 1525 break; 1526 } 1527 /* Congestion experienced. */ 1528 if (thflags & TH_ECE) { 1529 cc_cong_signal(tp, th, CC_ECN); 1530 } 1531 } 1532 1533 /* 1534 * Parse options on any incoming segment. 1535 */ 1536 tcp_dooptions(&to, (u_char *)(th + 1), 1537 (th->th_off << 2) - sizeof(struct tcphdr), 1538 (thflags & TH_SYN) ? TO_SYN : 0); 1539 1540 /* 1541 * If echoed timestamp is later than the current time, 1542 * fall back to non RFC1323 RTT calculation. Normalize 1543 * timestamp if syncookies were used when this connection 1544 * was established. 1545 */ 1546 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1547 to.to_tsecr -= tp->ts_offset; 1548 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1549 to.to_tsecr = 0; 1550 } 1551 1552 /* 1553 * Process options only when we get SYN/ACK back. The SYN case 1554 * for incoming connections is handled in tcp_syncache. 1555 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1556 * or <SYN,ACK>) segment itself is never scaled. 1557 * XXX this is traditional behavior, may need to be cleaned up. 1558 */ 1559 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1560 if ((to.to_flags & TOF_SCALE) && 1561 (tp->t_flags & TF_REQ_SCALE)) { 1562 tp->t_flags |= TF_RCVD_SCALE; 1563 tp->snd_scale = to.to_wscale; 1564 } 1565 /* 1566 * Initial send window. It will be updated with 1567 * the next incoming segment to the scaled value. 1568 */ 1569 tp->snd_wnd = th->th_win; 1570 if (to.to_flags & TOF_TS) { 1571 tp->t_flags |= TF_RCVD_TSTMP; 1572 tp->ts_recent = to.to_tsval; 1573 tp->ts_recent_age = tcp_ts_getticks(); 1574 } 1575 if (to.to_flags & TOF_MSS) 1576 tcp_mss(tp, to.to_mss); 1577 if ((tp->t_flags & TF_SACK_PERMIT) && 1578 (to.to_flags & TOF_SACKPERM) == 0) 1579 tp->t_flags &= ~TF_SACK_PERMIT; 1580 } 1581 1582 /* 1583 * Header prediction: check for the two common cases 1584 * of a uni-directional data xfer. If the packet has 1585 * no control flags, is in-sequence, the window didn't 1586 * change and we're not retransmitting, it's a 1587 * candidate. If the length is zero and the ack moved 1588 * forward, we're the sender side of the xfer. Just 1589 * free the data acked & wake any higher level process 1590 * that was blocked waiting for space. If the length 1591 * is non-zero and the ack didn't move, we're the 1592 * receiver side. If we're getting packets in-order 1593 * (the reassembly queue is empty), add the data to 1594 * the socket buffer and note that we need a delayed ack. 1595 * Make sure that the hidden state-flags are also off. 1596 * Since we check for TCPS_ESTABLISHED first, it can only 1597 * be TH_NEEDSYN. 1598 */ 1599 if (tp->t_state == TCPS_ESTABLISHED && 1600 th->th_seq == tp->rcv_nxt && 1601 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1602 tp->snd_nxt == tp->snd_max && 1603 tiwin && tiwin == tp->snd_wnd && 1604 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1605 LIST_EMPTY(&tp->t_segq) && 1606 ((to.to_flags & TOF_TS) == 0 || 1607 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1608 1609 /* 1610 * If last ACK falls within this segment's sequence numbers, 1611 * record the timestamp. 1612 * NOTE that the test is modified according to the latest 1613 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1614 */ 1615 if ((to.to_flags & TOF_TS) != 0 && 1616 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1617 tp->ts_recent_age = tcp_ts_getticks(); 1618 tp->ts_recent = to.to_tsval; 1619 } 1620 1621 if (tlen == 0) { 1622 if (SEQ_GT(th->th_ack, tp->snd_una) && 1623 SEQ_LEQ(th->th_ack, tp->snd_max) && 1624 !IN_RECOVERY(tp->t_flags) && 1625 (to.to_flags & TOF_SACK) == 0 && 1626 TAILQ_EMPTY(&tp->snd_holes)) { 1627 /* 1628 * This is a pure ack for outstanding data. 1629 */ 1630 if (ti_locked == TI_WLOCKED) 1631 INP_INFO_WUNLOCK(&V_tcbinfo); 1632 ti_locked = TI_UNLOCKED; 1633 1634 TCPSTAT_INC(tcps_predack); 1635 1636 /* 1637 * "bad retransmit" recovery. 1638 */ 1639 if (tp->t_rxtshift == 1 && 1640 tp->t_flags & TF_PREVVALID && 1641 (int)(ticks - tp->t_badrxtwin) < 0) { 1642 cc_cong_signal(tp, th, CC_RTO_ERR); 1643 } 1644 1645 /* 1646 * Recalculate the transmit timer / rtt. 1647 * 1648 * Some boxes send broken timestamp replies 1649 * during the SYN+ACK phase, ignore 1650 * timestamps of 0 or we could calculate a 1651 * huge RTT and blow up the retransmit timer. 1652 */ 1653 if ((to.to_flags & TOF_TS) != 0 && 1654 to.to_tsecr) { 1655 u_int t; 1656 1657 t = tcp_ts_getticks() - to.to_tsecr; 1658 if (!tp->t_rttlow || tp->t_rttlow > t) 1659 tp->t_rttlow = t; 1660 tcp_xmit_timer(tp, 1661 TCP_TS_TO_TICKS(t) + 1); 1662 } else if (tp->t_rtttime && 1663 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1664 if (!tp->t_rttlow || 1665 tp->t_rttlow > ticks - tp->t_rtttime) 1666 tp->t_rttlow = ticks - tp->t_rtttime; 1667 tcp_xmit_timer(tp, 1668 ticks - tp->t_rtttime); 1669 } 1670 acked = BYTES_THIS_ACK(tp, th); 1671 1672 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1673 hhook_run_tcp_est_in(tp, th, &to); 1674 1675 TCPSTAT_INC(tcps_rcvackpack); 1676 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1677 sbdrop(&so->so_snd, acked); 1678 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1679 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1680 tp->snd_recover = th->th_ack - 1; 1681 1682 /* 1683 * Let the congestion control algorithm update 1684 * congestion control related information. This 1685 * typically means increasing the congestion 1686 * window. 1687 */ 1688 cc_ack_received(tp, th, CC_ACK); 1689 1690 tp->snd_una = th->th_ack; 1691 /* 1692 * Pull snd_wl2 up to prevent seq wrap relative 1693 * to th_ack. 1694 */ 1695 tp->snd_wl2 = th->th_ack; 1696 tp->t_dupacks = 0; 1697 m_freem(m); 1698 ND6_HINT(tp); /* Some progress has been made. */ 1699 1700 /* 1701 * If all outstanding data are acked, stop 1702 * retransmit timer, otherwise restart timer 1703 * using current (possibly backed-off) value. 1704 * If process is waiting for space, 1705 * wakeup/selwakeup/signal. If data 1706 * are ready to send, let tcp_output 1707 * decide between more output or persist. 1708 */ 1709 #ifdef TCPDEBUG 1710 if (so->so_options & SO_DEBUG) 1711 tcp_trace(TA_INPUT, ostate, tp, 1712 (void *)tcp_saveipgen, 1713 &tcp_savetcp, 0); 1714 #endif 1715 if (tp->snd_una == tp->snd_max) 1716 tcp_timer_activate(tp, TT_REXMT, 0); 1717 else if (!tcp_timer_active(tp, TT_PERSIST)) 1718 tcp_timer_activate(tp, TT_REXMT, 1719 tp->t_rxtcur); 1720 sowwakeup(so); 1721 if (so->so_snd.sb_cc) 1722 (void) tcp_output(tp); 1723 goto check_delack; 1724 } 1725 } else if (th->th_ack == tp->snd_una && 1726 tlen <= sbspace(&so->so_rcv)) { 1727 int newsize = 0; /* automatic sockbuf scaling */ 1728 1729 /* 1730 * This is a pure, in-sequence data packet with 1731 * nothing on the reassembly queue and we have enough 1732 * buffer space to take it. 1733 */ 1734 if (ti_locked == TI_WLOCKED) 1735 INP_INFO_WUNLOCK(&V_tcbinfo); 1736 ti_locked = TI_UNLOCKED; 1737 1738 /* Clean receiver SACK report if present */ 1739 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1740 tcp_clean_sackreport(tp); 1741 TCPSTAT_INC(tcps_preddat); 1742 tp->rcv_nxt += tlen; 1743 /* 1744 * Pull snd_wl1 up to prevent seq wrap relative to 1745 * th_seq. 1746 */ 1747 tp->snd_wl1 = th->th_seq; 1748 /* 1749 * Pull rcv_up up to prevent seq wrap relative to 1750 * rcv_nxt. 1751 */ 1752 tp->rcv_up = tp->rcv_nxt; 1753 TCPSTAT_INC(tcps_rcvpack); 1754 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1755 ND6_HINT(tp); /* Some progress has been made */ 1756 #ifdef TCPDEBUG 1757 if (so->so_options & SO_DEBUG) 1758 tcp_trace(TA_INPUT, ostate, tp, 1759 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1760 #endif 1761 /* 1762 * Automatic sizing of receive socket buffer. Often the send 1763 * buffer size is not optimally adjusted to the actual network 1764 * conditions at hand (delay bandwidth product). Setting the 1765 * buffer size too small limits throughput on links with high 1766 * bandwidth and high delay (eg. trans-continental/oceanic links). 1767 * 1768 * On the receive side the socket buffer memory is only rarely 1769 * used to any significant extent. This allows us to be much 1770 * more aggressive in scaling the receive socket buffer. For 1771 * the case that the buffer space is actually used to a large 1772 * extent and we run out of kernel memory we can simply drop 1773 * the new segments; TCP on the sender will just retransmit it 1774 * later. Setting the buffer size too big may only consume too 1775 * much kernel memory if the application doesn't read() from 1776 * the socket or packet loss or reordering makes use of the 1777 * reassembly queue. 1778 * 1779 * The criteria to step up the receive buffer one notch are: 1780 * 1. the number of bytes received during the time it takes 1781 * one timestamp to be reflected back to us (the RTT); 1782 * 2. received bytes per RTT is within seven eighth of the 1783 * current socket buffer size; 1784 * 3. receive buffer size has not hit maximal automatic size; 1785 * 1786 * This algorithm does one step per RTT at most and only if 1787 * we receive a bulk stream w/o packet losses or reorderings. 1788 * Shrinking the buffer during idle times is not necessary as 1789 * it doesn't consume any memory when idle. 1790 * 1791 * TODO: Only step up if the application is actually serving 1792 * the buffer to better manage the socket buffer resources. 1793 */ 1794 if (V_tcp_do_autorcvbuf && 1795 to.to_tsecr && 1796 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1797 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1798 to.to_tsecr - tp->rfbuf_ts < hz) { 1799 if (tp->rfbuf_cnt > 1800 (so->so_rcv.sb_hiwat / 8 * 7) && 1801 so->so_rcv.sb_hiwat < 1802 V_tcp_autorcvbuf_max) { 1803 newsize = 1804 min(so->so_rcv.sb_hiwat + 1805 V_tcp_autorcvbuf_inc, 1806 V_tcp_autorcvbuf_max); 1807 } 1808 /* Start over with next RTT. */ 1809 tp->rfbuf_ts = 0; 1810 tp->rfbuf_cnt = 0; 1811 } else 1812 tp->rfbuf_cnt += tlen; /* add up */ 1813 } 1814 1815 /* Add data to socket buffer. */ 1816 SOCKBUF_LOCK(&so->so_rcv); 1817 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1818 m_freem(m); 1819 } else { 1820 /* 1821 * Set new socket buffer size. 1822 * Give up when limit is reached. 1823 */ 1824 if (newsize) 1825 if (!sbreserve_locked(&so->so_rcv, 1826 newsize, so, NULL)) 1827 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1828 m_adj(m, drop_hdrlen); /* delayed header drop */ 1829 sbappendstream_locked(&so->so_rcv, m); 1830 } 1831 /* NB: sorwakeup_locked() does an implicit unlock. */ 1832 sorwakeup_locked(so); 1833 if (DELAY_ACK(tp)) { 1834 tp->t_flags |= TF_DELACK; 1835 } else { 1836 tp->t_flags |= TF_ACKNOW; 1837 tcp_output(tp); 1838 } 1839 goto check_delack; 1840 } 1841 } 1842 1843 /* 1844 * Calculate amount of space in receive window, 1845 * and then do TCP input processing. 1846 * Receive window is amount of space in rcv queue, 1847 * but not less than advertised window. 1848 */ 1849 win = sbspace(&so->so_rcv); 1850 if (win < 0) 1851 win = 0; 1852 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1853 1854 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1855 tp->rfbuf_ts = 0; 1856 tp->rfbuf_cnt = 0; 1857 1858 switch (tp->t_state) { 1859 1860 /* 1861 * If the state is SYN_RECEIVED: 1862 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1863 */ 1864 case TCPS_SYN_RECEIVED: 1865 if ((thflags & TH_ACK) && 1866 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1867 SEQ_GT(th->th_ack, tp->snd_max))) { 1868 rstreason = BANDLIM_RST_OPENPORT; 1869 goto dropwithreset; 1870 } 1871 break; 1872 1873 /* 1874 * If the state is SYN_SENT: 1875 * if seg contains an ACK, but not for our SYN, drop the input. 1876 * if seg contains a RST, then drop the connection. 1877 * if seg does not contain SYN, then drop it. 1878 * Otherwise this is an acceptable SYN segment 1879 * initialize tp->rcv_nxt and tp->irs 1880 * if seg contains ack then advance tp->snd_una 1881 * if seg contains an ECE and ECN support is enabled, the stream 1882 * is ECN capable. 1883 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1884 * arrange for segment to be acked (eventually) 1885 * continue processing rest of data/controls, beginning with URG 1886 */ 1887 case TCPS_SYN_SENT: 1888 if ((thflags & TH_ACK) && 1889 (SEQ_LEQ(th->th_ack, tp->iss) || 1890 SEQ_GT(th->th_ack, tp->snd_max))) { 1891 rstreason = BANDLIM_UNLIMITED; 1892 goto dropwithreset; 1893 } 1894 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1895 tp = tcp_drop(tp, ECONNREFUSED); 1896 if (thflags & TH_RST) 1897 goto drop; 1898 if (!(thflags & TH_SYN)) 1899 goto drop; 1900 1901 tp->irs = th->th_seq; 1902 tcp_rcvseqinit(tp); 1903 if (thflags & TH_ACK) { 1904 TCPSTAT_INC(tcps_connects); 1905 soisconnected(so); 1906 #ifdef MAC 1907 mac_socketpeer_set_from_mbuf(m, so); 1908 #endif 1909 /* Do window scaling on this connection? */ 1910 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1911 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1912 tp->rcv_scale = tp->request_r_scale; 1913 } 1914 tp->rcv_adv += imin(tp->rcv_wnd, 1915 TCP_MAXWIN << tp->rcv_scale); 1916 tp->snd_una++; /* SYN is acked */ 1917 /* 1918 * If there's data, delay ACK; if there's also a FIN 1919 * ACKNOW will be turned on later. 1920 */ 1921 if (DELAY_ACK(tp) && tlen != 0) 1922 tcp_timer_activate(tp, TT_DELACK, 1923 tcp_delacktime); 1924 else 1925 tp->t_flags |= TF_ACKNOW; 1926 1927 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1928 tp->t_flags |= TF_ECN_PERMIT; 1929 TCPSTAT_INC(tcps_ecn_shs); 1930 } 1931 1932 /* 1933 * Received <SYN,ACK> in SYN_SENT[*] state. 1934 * Transitions: 1935 * SYN_SENT --> ESTABLISHED 1936 * SYN_SENT* --> FIN_WAIT_1 1937 */ 1938 tp->t_starttime = ticks; 1939 if (tp->t_flags & TF_NEEDFIN) { 1940 tp->t_state = TCPS_FIN_WAIT_1; 1941 tp->t_flags &= ~TF_NEEDFIN; 1942 thflags &= ~TH_SYN; 1943 } else { 1944 tp->t_state = TCPS_ESTABLISHED; 1945 cc_conn_init(tp); 1946 tcp_timer_activate(tp, TT_KEEP, 1947 TP_KEEPIDLE(tp)); 1948 } 1949 } else { 1950 /* 1951 * Received initial SYN in SYN-SENT[*] state => 1952 * simultaneous open. If segment contains CC option 1953 * and there is a cached CC, apply TAO test. 1954 * If it succeeds, connection is * half-synchronized. 1955 * Otherwise, do 3-way handshake: 1956 * SYN-SENT -> SYN-RECEIVED 1957 * SYN-SENT* -> SYN-RECEIVED* 1958 * If there was no CC option, clear cached CC value. 1959 */ 1960 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1961 tcp_timer_activate(tp, TT_REXMT, 0); 1962 tp->t_state = TCPS_SYN_RECEIVED; 1963 } 1964 1965 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1966 "ti_locked %d", __func__, ti_locked)); 1967 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1968 INP_WLOCK_ASSERT(tp->t_inpcb); 1969 1970 /* 1971 * Advance th->th_seq to correspond to first data byte. 1972 * If data, trim to stay within window, 1973 * dropping FIN if necessary. 1974 */ 1975 th->th_seq++; 1976 if (tlen > tp->rcv_wnd) { 1977 todrop = tlen - tp->rcv_wnd; 1978 m_adj(m, -todrop); 1979 tlen = tp->rcv_wnd; 1980 thflags &= ~TH_FIN; 1981 TCPSTAT_INC(tcps_rcvpackafterwin); 1982 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1983 } 1984 tp->snd_wl1 = th->th_seq - 1; 1985 tp->rcv_up = th->th_seq; 1986 /* 1987 * Client side of transaction: already sent SYN and data. 1988 * If the remote host used T/TCP to validate the SYN, 1989 * our data will be ACK'd; if so, enter normal data segment 1990 * processing in the middle of step 5, ack processing. 1991 * Otherwise, goto step 6. 1992 */ 1993 if (thflags & TH_ACK) 1994 goto process_ACK; 1995 1996 goto step6; 1997 1998 /* 1999 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2000 * do normal processing. 2001 * 2002 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2003 */ 2004 case TCPS_LAST_ACK: 2005 case TCPS_CLOSING: 2006 break; /* continue normal processing */ 2007 } 2008 2009 /* 2010 * States other than LISTEN or SYN_SENT. 2011 * First check the RST flag and sequence number since reset segments 2012 * are exempt from the timestamp and connection count tests. This 2013 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2014 * below which allowed reset segments in half the sequence space 2015 * to fall though and be processed (which gives forged reset 2016 * segments with a random sequence number a 50 percent chance of 2017 * killing a connection). 2018 * Then check timestamp, if present. 2019 * Then check the connection count, if present. 2020 * Then check that at least some bytes of segment are within 2021 * receive window. If segment begins before rcv_nxt, 2022 * drop leading data (and SYN); if nothing left, just ack. 2023 * 2024 * 2025 * If the RST bit is set, check the sequence number to see 2026 * if this is a valid reset segment. 2027 * RFC 793 page 37: 2028 * In all states except SYN-SENT, all reset (RST) segments 2029 * are validated by checking their SEQ-fields. A reset is 2030 * valid if its sequence number is in the window. 2031 * Note: this does not take into account delayed ACKs, so 2032 * we should test against last_ack_sent instead of rcv_nxt. 2033 * The sequence number in the reset segment is normally an 2034 * echo of our outgoing acknowlegement numbers, but some hosts 2035 * send a reset with the sequence number at the rightmost edge 2036 * of our receive window, and we have to handle this case. 2037 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2038 * that brute force RST attacks are possible. To combat this, 2039 * we use a much stricter check while in the ESTABLISHED state, 2040 * only accepting RSTs where the sequence number is equal to 2041 * last_ack_sent. In all other states (the states in which a 2042 * RST is more likely), the more permissive check is used. 2043 * If we have multiple segments in flight, the initial reset 2044 * segment sequence numbers will be to the left of last_ack_sent, 2045 * but they will eventually catch up. 2046 * In any case, it never made sense to trim reset segments to 2047 * fit the receive window since RFC 1122 says: 2048 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2049 * 2050 * A TCP SHOULD allow a received RST segment to include data. 2051 * 2052 * DISCUSSION 2053 * It has been suggested that a RST segment could contain 2054 * ASCII text that encoded and explained the cause of the 2055 * RST. No standard has yet been established for such 2056 * data. 2057 * 2058 * If the reset segment passes the sequence number test examine 2059 * the state: 2060 * SYN_RECEIVED STATE: 2061 * If passive open, return to LISTEN state. 2062 * If active open, inform user that connection was refused. 2063 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2064 * Inform user that connection was reset, and close tcb. 2065 * CLOSING, LAST_ACK STATES: 2066 * Close the tcb. 2067 * TIME_WAIT STATE: 2068 * Drop the segment - see Stevens, vol. 2, p. 964 and 2069 * RFC 1337. 2070 */ 2071 if (thflags & TH_RST) { 2072 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2073 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2074 switch (tp->t_state) { 2075 2076 case TCPS_SYN_RECEIVED: 2077 so->so_error = ECONNREFUSED; 2078 goto close; 2079 2080 case TCPS_ESTABLISHED: 2081 if (V_tcp_insecure_rst == 0 && 2082 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2083 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2084 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2085 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2086 TCPSTAT_INC(tcps_badrst); 2087 goto drop; 2088 } 2089 /* FALLTHROUGH */ 2090 case TCPS_FIN_WAIT_1: 2091 case TCPS_FIN_WAIT_2: 2092 case TCPS_CLOSE_WAIT: 2093 so->so_error = ECONNRESET; 2094 close: 2095 KASSERT(ti_locked == TI_WLOCKED, 2096 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2097 ti_locked)); 2098 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2099 2100 tp->t_state = TCPS_CLOSED; 2101 TCPSTAT_INC(tcps_drops); 2102 tp = tcp_close(tp); 2103 break; 2104 2105 case TCPS_CLOSING: 2106 case TCPS_LAST_ACK: 2107 KASSERT(ti_locked == TI_WLOCKED, 2108 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2109 ti_locked)); 2110 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2111 2112 tp = tcp_close(tp); 2113 break; 2114 } 2115 } 2116 goto drop; 2117 } 2118 2119 /* 2120 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2121 * and it's less than ts_recent, drop it. 2122 */ 2123 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2124 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2125 2126 /* Check to see if ts_recent is over 24 days old. */ 2127 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2128 /* 2129 * Invalidate ts_recent. If this segment updates 2130 * ts_recent, the age will be reset later and ts_recent 2131 * will get a valid value. If it does not, setting 2132 * ts_recent to zero will at least satisfy the 2133 * requirement that zero be placed in the timestamp 2134 * echo reply when ts_recent isn't valid. The 2135 * age isn't reset until we get a valid ts_recent 2136 * because we don't want out-of-order segments to be 2137 * dropped when ts_recent is old. 2138 */ 2139 tp->ts_recent = 0; 2140 } else { 2141 TCPSTAT_INC(tcps_rcvduppack); 2142 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2143 TCPSTAT_INC(tcps_pawsdrop); 2144 if (tlen) 2145 goto dropafterack; 2146 goto drop; 2147 } 2148 } 2149 2150 /* 2151 * In the SYN-RECEIVED state, validate that the packet belongs to 2152 * this connection before trimming the data to fit the receive 2153 * window. Check the sequence number versus IRS since we know 2154 * the sequence numbers haven't wrapped. This is a partial fix 2155 * for the "LAND" DoS attack. 2156 */ 2157 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2158 rstreason = BANDLIM_RST_OPENPORT; 2159 goto dropwithreset; 2160 } 2161 2162 todrop = tp->rcv_nxt - th->th_seq; 2163 if (todrop > 0) { 2164 if (thflags & TH_SYN) { 2165 thflags &= ~TH_SYN; 2166 th->th_seq++; 2167 if (th->th_urp > 1) 2168 th->th_urp--; 2169 else 2170 thflags &= ~TH_URG; 2171 todrop--; 2172 } 2173 /* 2174 * Following if statement from Stevens, vol. 2, p. 960. 2175 */ 2176 if (todrop > tlen 2177 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2178 /* 2179 * Any valid FIN must be to the left of the window. 2180 * At this point the FIN must be a duplicate or out 2181 * of sequence; drop it. 2182 */ 2183 thflags &= ~TH_FIN; 2184 2185 /* 2186 * Send an ACK to resynchronize and drop any data. 2187 * But keep on processing for RST or ACK. 2188 */ 2189 tp->t_flags |= TF_ACKNOW; 2190 todrop = tlen; 2191 TCPSTAT_INC(tcps_rcvduppack); 2192 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2193 } else { 2194 TCPSTAT_INC(tcps_rcvpartduppack); 2195 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2196 } 2197 drop_hdrlen += todrop; /* drop from the top afterwards */ 2198 th->th_seq += todrop; 2199 tlen -= todrop; 2200 if (th->th_urp > todrop) 2201 th->th_urp -= todrop; 2202 else { 2203 thflags &= ~TH_URG; 2204 th->th_urp = 0; 2205 } 2206 } 2207 2208 /* 2209 * If new data are received on a connection after the 2210 * user processes are gone, then RST the other end. 2211 */ 2212 if ((so->so_state & SS_NOFDREF) && 2213 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2214 char *s; 2215 2216 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2217 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2218 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2219 2220 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 2221 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 2222 "was closed, sending RST and removing tcpcb\n", 2223 s, __func__, tcpstates[tp->t_state], tlen); 2224 free(s, M_TCPLOG); 2225 } 2226 tp = tcp_close(tp); 2227 TCPSTAT_INC(tcps_rcvafterclose); 2228 rstreason = BANDLIM_UNLIMITED; 2229 goto dropwithreset; 2230 } 2231 2232 /* 2233 * If segment ends after window, drop trailing data 2234 * (and PUSH and FIN); if nothing left, just ACK. 2235 */ 2236 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2237 if (todrop > 0) { 2238 TCPSTAT_INC(tcps_rcvpackafterwin); 2239 if (todrop >= tlen) { 2240 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2241 /* 2242 * If window is closed can only take segments at 2243 * window edge, and have to drop data and PUSH from 2244 * incoming segments. Continue processing, but 2245 * remember to ack. Otherwise, drop segment 2246 * and ack. 2247 */ 2248 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2249 tp->t_flags |= TF_ACKNOW; 2250 TCPSTAT_INC(tcps_rcvwinprobe); 2251 } else 2252 goto dropafterack; 2253 } else 2254 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2255 m_adj(m, -todrop); 2256 tlen -= todrop; 2257 thflags &= ~(TH_PUSH|TH_FIN); 2258 } 2259 2260 /* 2261 * If last ACK falls within this segment's sequence numbers, 2262 * record its timestamp. 2263 * NOTE: 2264 * 1) That the test incorporates suggestions from the latest 2265 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2266 * 2) That updating only on newer timestamps interferes with 2267 * our earlier PAWS tests, so this check should be solely 2268 * predicated on the sequence space of this segment. 2269 * 3) That we modify the segment boundary check to be 2270 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2271 * instead of RFC1323's 2272 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2273 * This modified check allows us to overcome RFC1323's 2274 * limitations as described in Stevens TCP/IP Illustrated 2275 * Vol. 2 p.869. In such cases, we can still calculate the 2276 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2277 */ 2278 if ((to.to_flags & TOF_TS) != 0 && 2279 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2280 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2281 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2282 tp->ts_recent_age = tcp_ts_getticks(); 2283 tp->ts_recent = to.to_tsval; 2284 } 2285 2286 /* 2287 * If a SYN is in the window, then this is an 2288 * error and we send an RST and drop the connection. 2289 */ 2290 if (thflags & TH_SYN) { 2291 KASSERT(ti_locked == TI_WLOCKED, 2292 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2293 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2294 2295 tp = tcp_drop(tp, ECONNRESET); 2296 rstreason = BANDLIM_UNLIMITED; 2297 goto drop; 2298 } 2299 2300 /* 2301 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2302 * flag is on (half-synchronized state), then queue data for 2303 * later processing; else drop segment and return. 2304 */ 2305 if ((thflags & TH_ACK) == 0) { 2306 if (tp->t_state == TCPS_SYN_RECEIVED || 2307 (tp->t_flags & TF_NEEDSYN)) 2308 goto step6; 2309 else if (tp->t_flags & TF_ACKNOW) 2310 goto dropafterack; 2311 else 2312 goto drop; 2313 } 2314 2315 /* 2316 * Ack processing. 2317 */ 2318 switch (tp->t_state) { 2319 2320 /* 2321 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2322 * ESTABLISHED state and continue processing. 2323 * The ACK was checked above. 2324 */ 2325 case TCPS_SYN_RECEIVED: 2326 2327 TCPSTAT_INC(tcps_connects); 2328 soisconnected(so); 2329 /* Do window scaling? */ 2330 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2331 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2332 tp->rcv_scale = tp->request_r_scale; 2333 tp->snd_wnd = tiwin; 2334 } 2335 /* 2336 * Make transitions: 2337 * SYN-RECEIVED -> ESTABLISHED 2338 * SYN-RECEIVED* -> FIN-WAIT-1 2339 */ 2340 tp->t_starttime = ticks; 2341 if (tp->t_flags & TF_NEEDFIN) { 2342 tp->t_state = TCPS_FIN_WAIT_1; 2343 tp->t_flags &= ~TF_NEEDFIN; 2344 } else { 2345 tp->t_state = TCPS_ESTABLISHED; 2346 cc_conn_init(tp); 2347 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2348 } 2349 /* 2350 * If segment contains data or ACK, will call tcp_reass() 2351 * later; if not, do so now to pass queued data to user. 2352 */ 2353 if (tlen == 0 && (thflags & TH_FIN) == 0) 2354 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2355 (struct mbuf *)0); 2356 tp->snd_wl1 = th->th_seq - 1; 2357 /* FALLTHROUGH */ 2358 2359 /* 2360 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2361 * ACKs. If the ack is in the range 2362 * tp->snd_una < th->th_ack <= tp->snd_max 2363 * then advance tp->snd_una to th->th_ack and drop 2364 * data from the retransmission queue. If this ACK reflects 2365 * more up to date window information we update our window information. 2366 */ 2367 case TCPS_ESTABLISHED: 2368 case TCPS_FIN_WAIT_1: 2369 case TCPS_FIN_WAIT_2: 2370 case TCPS_CLOSE_WAIT: 2371 case TCPS_CLOSING: 2372 case TCPS_LAST_ACK: 2373 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2374 TCPSTAT_INC(tcps_rcvacktoomuch); 2375 goto dropafterack; 2376 } 2377 if ((tp->t_flags & TF_SACK_PERMIT) && 2378 ((to.to_flags & TOF_SACK) || 2379 !TAILQ_EMPTY(&tp->snd_holes))) 2380 tcp_sack_doack(tp, &to, th->th_ack); 2381 2382 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2383 hhook_run_tcp_est_in(tp, th, &to); 2384 2385 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2386 if (tlen == 0 && tiwin == tp->snd_wnd) { 2387 TCPSTAT_INC(tcps_rcvdupack); 2388 /* 2389 * If we have outstanding data (other than 2390 * a window probe), this is a completely 2391 * duplicate ack (ie, window info didn't 2392 * change), the ack is the biggest we've 2393 * seen and we've seen exactly our rexmt 2394 * threshhold of them, assume a packet 2395 * has been dropped and retransmit it. 2396 * Kludge snd_nxt & the congestion 2397 * window so we send only this one 2398 * packet. 2399 * 2400 * We know we're losing at the current 2401 * window size so do congestion avoidance 2402 * (set ssthresh to half the current window 2403 * and pull our congestion window back to 2404 * the new ssthresh). 2405 * 2406 * Dup acks mean that packets have left the 2407 * network (they're now cached at the receiver) 2408 * so bump cwnd by the amount in the receiver 2409 * to keep a constant cwnd packets in the 2410 * network. 2411 * 2412 * When using TCP ECN, notify the peer that 2413 * we reduced the cwnd. 2414 */ 2415 if (!tcp_timer_active(tp, TT_REXMT) || 2416 th->th_ack != tp->snd_una) 2417 tp->t_dupacks = 0; 2418 else if (++tp->t_dupacks > tcprexmtthresh || 2419 IN_FASTRECOVERY(tp->t_flags)) { 2420 cc_ack_received(tp, th, CC_DUPACK); 2421 if ((tp->t_flags & TF_SACK_PERMIT) && 2422 IN_FASTRECOVERY(tp->t_flags)) { 2423 int awnd; 2424 2425 /* 2426 * Compute the amount of data in flight first. 2427 * We can inject new data into the pipe iff 2428 * we have less than 1/2 the original window's 2429 * worth of data in flight. 2430 */ 2431 awnd = (tp->snd_nxt - tp->snd_fack) + 2432 tp->sackhint.sack_bytes_rexmit; 2433 if (awnd < tp->snd_ssthresh) { 2434 tp->snd_cwnd += tp->t_maxseg; 2435 if (tp->snd_cwnd > tp->snd_ssthresh) 2436 tp->snd_cwnd = tp->snd_ssthresh; 2437 } 2438 } else 2439 tp->snd_cwnd += tp->t_maxseg; 2440 (void) tcp_output(tp); 2441 goto drop; 2442 } else if (tp->t_dupacks == tcprexmtthresh) { 2443 tcp_seq onxt = tp->snd_nxt; 2444 2445 /* 2446 * If we're doing sack, check to 2447 * see if we're already in sack 2448 * recovery. If we're not doing sack, 2449 * check to see if we're in newreno 2450 * recovery. 2451 */ 2452 if (tp->t_flags & TF_SACK_PERMIT) { 2453 if (IN_FASTRECOVERY(tp->t_flags)) { 2454 tp->t_dupacks = 0; 2455 break; 2456 } 2457 } else { 2458 if (SEQ_LEQ(th->th_ack, 2459 tp->snd_recover)) { 2460 tp->t_dupacks = 0; 2461 break; 2462 } 2463 } 2464 /* Congestion signal before ack. */ 2465 cc_cong_signal(tp, th, CC_NDUPACK); 2466 cc_ack_received(tp, th, CC_DUPACK); 2467 tcp_timer_activate(tp, TT_REXMT, 0); 2468 tp->t_rtttime = 0; 2469 if (tp->t_flags & TF_SACK_PERMIT) { 2470 TCPSTAT_INC( 2471 tcps_sack_recovery_episode); 2472 tp->sack_newdata = tp->snd_nxt; 2473 tp->snd_cwnd = tp->t_maxseg; 2474 (void) tcp_output(tp); 2475 goto drop; 2476 } 2477 tp->snd_nxt = th->th_ack; 2478 tp->snd_cwnd = tp->t_maxseg; 2479 (void) tcp_output(tp); 2480 KASSERT(tp->snd_limited <= 2, 2481 ("%s: tp->snd_limited too big", 2482 __func__)); 2483 tp->snd_cwnd = tp->snd_ssthresh + 2484 tp->t_maxseg * 2485 (tp->t_dupacks - tp->snd_limited); 2486 if (SEQ_GT(onxt, tp->snd_nxt)) 2487 tp->snd_nxt = onxt; 2488 goto drop; 2489 } else if (V_tcp_do_rfc3042) { 2490 cc_ack_received(tp, th, CC_DUPACK); 2491 u_long oldcwnd = tp->snd_cwnd; 2492 tcp_seq oldsndmax = tp->snd_max; 2493 u_int sent; 2494 2495 KASSERT(tp->t_dupacks == 1 || 2496 tp->t_dupacks == 2, 2497 ("%s: dupacks not 1 or 2", 2498 __func__)); 2499 if (tp->t_dupacks == 1) 2500 tp->snd_limited = 0; 2501 tp->snd_cwnd = 2502 (tp->snd_nxt - tp->snd_una) + 2503 (tp->t_dupacks - tp->snd_limited) * 2504 tp->t_maxseg; 2505 (void) tcp_output(tp); 2506 sent = tp->snd_max - oldsndmax; 2507 if (sent > tp->t_maxseg) { 2508 KASSERT((tp->t_dupacks == 2 && 2509 tp->snd_limited == 0) || 2510 (sent == tp->t_maxseg + 1 && 2511 tp->t_flags & TF_SENTFIN), 2512 ("%s: sent too much", 2513 __func__)); 2514 tp->snd_limited = 2; 2515 } else if (sent > 0) 2516 ++tp->snd_limited; 2517 tp->snd_cwnd = oldcwnd; 2518 goto drop; 2519 } 2520 } else 2521 tp->t_dupacks = 0; 2522 break; 2523 } 2524 2525 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2526 ("%s: th_ack <= snd_una", __func__)); 2527 2528 /* 2529 * If the congestion window was inflated to account 2530 * for the other side's cached packets, retract it. 2531 */ 2532 if (IN_FASTRECOVERY(tp->t_flags)) { 2533 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2534 if (tp->t_flags & TF_SACK_PERMIT) 2535 tcp_sack_partialack(tp, th); 2536 else 2537 tcp_newreno_partial_ack(tp, th); 2538 } else 2539 cc_post_recovery(tp, th); 2540 } 2541 tp->t_dupacks = 0; 2542 /* 2543 * If we reach this point, ACK is not a duplicate, 2544 * i.e., it ACKs something we sent. 2545 */ 2546 if (tp->t_flags & TF_NEEDSYN) { 2547 /* 2548 * T/TCP: Connection was half-synchronized, and our 2549 * SYN has been ACK'd (so connection is now fully 2550 * synchronized). Go to non-starred state, 2551 * increment snd_una for ACK of SYN, and check if 2552 * we can do window scaling. 2553 */ 2554 tp->t_flags &= ~TF_NEEDSYN; 2555 tp->snd_una++; 2556 /* Do window scaling? */ 2557 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2558 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2559 tp->rcv_scale = tp->request_r_scale; 2560 /* Send window already scaled. */ 2561 } 2562 } 2563 2564 process_ACK: 2565 INP_WLOCK_ASSERT(tp->t_inpcb); 2566 2567 acked = BYTES_THIS_ACK(tp, th); 2568 TCPSTAT_INC(tcps_rcvackpack); 2569 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2570 2571 /* 2572 * If we just performed our first retransmit, and the ACK 2573 * arrives within our recovery window, then it was a mistake 2574 * to do the retransmit in the first place. Recover our 2575 * original cwnd and ssthresh, and proceed to transmit where 2576 * we left off. 2577 */ 2578 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2579 (int)(ticks - tp->t_badrxtwin) < 0) 2580 cc_cong_signal(tp, th, CC_RTO_ERR); 2581 2582 /* 2583 * If we have a timestamp reply, update smoothed 2584 * round trip time. If no timestamp is present but 2585 * transmit timer is running and timed sequence 2586 * number was acked, update smoothed round trip time. 2587 * Since we now have an rtt measurement, cancel the 2588 * timer backoff (cf., Phil Karn's retransmit alg.). 2589 * Recompute the initial retransmit timer. 2590 * 2591 * Some boxes send broken timestamp replies 2592 * during the SYN+ACK phase, ignore 2593 * timestamps of 0 or we could calculate a 2594 * huge RTT and blow up the retransmit timer. 2595 */ 2596 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2597 u_int t; 2598 2599 t = tcp_ts_getticks() - to.to_tsecr; 2600 if (!tp->t_rttlow || tp->t_rttlow > t) 2601 tp->t_rttlow = t; 2602 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2603 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2604 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2605 tp->t_rttlow = ticks - tp->t_rtttime; 2606 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2607 } 2608 2609 /* 2610 * If all outstanding data is acked, stop retransmit 2611 * timer and remember to restart (more output or persist). 2612 * If there is more data to be acked, restart retransmit 2613 * timer, using current (possibly backed-off) value. 2614 */ 2615 if (th->th_ack == tp->snd_max) { 2616 tcp_timer_activate(tp, TT_REXMT, 0); 2617 needoutput = 1; 2618 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2619 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2620 2621 /* 2622 * If no data (only SYN) was ACK'd, 2623 * skip rest of ACK processing. 2624 */ 2625 if (acked == 0) 2626 goto step6; 2627 2628 /* 2629 * Let the congestion control algorithm update congestion 2630 * control related information. This typically means increasing 2631 * the congestion window. 2632 */ 2633 cc_ack_received(tp, th, CC_ACK); 2634 2635 SOCKBUF_LOCK(&so->so_snd); 2636 if (acked > so->so_snd.sb_cc) { 2637 tp->snd_wnd -= so->so_snd.sb_cc; 2638 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2639 ourfinisacked = 1; 2640 } else { 2641 sbdrop_locked(&so->so_snd, acked); 2642 tp->snd_wnd -= acked; 2643 ourfinisacked = 0; 2644 } 2645 /* NB: sowwakeup_locked() does an implicit unlock. */ 2646 sowwakeup_locked(so); 2647 /* Detect una wraparound. */ 2648 if (!IN_RECOVERY(tp->t_flags) && 2649 SEQ_GT(tp->snd_una, tp->snd_recover) && 2650 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2651 tp->snd_recover = th->th_ack - 1; 2652 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2653 if (IN_RECOVERY(tp->t_flags) && 2654 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2655 EXIT_RECOVERY(tp->t_flags); 2656 } 2657 tp->snd_una = th->th_ack; 2658 if (tp->t_flags & TF_SACK_PERMIT) { 2659 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2660 tp->snd_recover = tp->snd_una; 2661 } 2662 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2663 tp->snd_nxt = tp->snd_una; 2664 2665 switch (tp->t_state) { 2666 2667 /* 2668 * In FIN_WAIT_1 STATE in addition to the processing 2669 * for the ESTABLISHED state if our FIN is now acknowledged 2670 * then enter FIN_WAIT_2. 2671 */ 2672 case TCPS_FIN_WAIT_1: 2673 if (ourfinisacked) { 2674 /* 2675 * If we can't receive any more 2676 * data, then closing user can proceed. 2677 * Starting the timer is contrary to the 2678 * specification, but if we don't get a FIN 2679 * we'll hang forever. 2680 * 2681 * XXXjl: 2682 * we should release the tp also, and use a 2683 * compressed state. 2684 */ 2685 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2686 soisdisconnected(so); 2687 tcp_timer_activate(tp, TT_2MSL, 2688 (tcp_fast_finwait2_recycle ? 2689 tcp_finwait2_timeout : 2690 TP_MAXIDLE(tp))); 2691 } 2692 tp->t_state = TCPS_FIN_WAIT_2; 2693 } 2694 break; 2695 2696 /* 2697 * In CLOSING STATE in addition to the processing for 2698 * the ESTABLISHED state if the ACK acknowledges our FIN 2699 * then enter the TIME-WAIT state, otherwise ignore 2700 * the segment. 2701 */ 2702 case TCPS_CLOSING: 2703 if (ourfinisacked) { 2704 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2705 tcp_twstart(tp); 2706 INP_INFO_WUNLOCK(&V_tcbinfo); 2707 m_freem(m); 2708 return; 2709 } 2710 break; 2711 2712 /* 2713 * In LAST_ACK, we may still be waiting for data to drain 2714 * and/or to be acked, as well as for the ack of our FIN. 2715 * If our FIN is now acknowledged, delete the TCB, 2716 * enter the closed state and return. 2717 */ 2718 case TCPS_LAST_ACK: 2719 if (ourfinisacked) { 2720 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2721 tp = tcp_close(tp); 2722 goto drop; 2723 } 2724 break; 2725 } 2726 } 2727 2728 step6: 2729 INP_WLOCK_ASSERT(tp->t_inpcb); 2730 2731 /* 2732 * Update window information. 2733 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2734 */ 2735 if ((thflags & TH_ACK) && 2736 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2737 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2738 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2739 /* keep track of pure window updates */ 2740 if (tlen == 0 && 2741 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2742 TCPSTAT_INC(tcps_rcvwinupd); 2743 tp->snd_wnd = tiwin; 2744 tp->snd_wl1 = th->th_seq; 2745 tp->snd_wl2 = th->th_ack; 2746 if (tp->snd_wnd > tp->max_sndwnd) 2747 tp->max_sndwnd = tp->snd_wnd; 2748 needoutput = 1; 2749 } 2750 2751 /* 2752 * Process segments with URG. 2753 */ 2754 if ((thflags & TH_URG) && th->th_urp && 2755 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2756 /* 2757 * This is a kludge, but if we receive and accept 2758 * random urgent pointers, we'll crash in 2759 * soreceive. It's hard to imagine someone 2760 * actually wanting to send this much urgent data. 2761 */ 2762 SOCKBUF_LOCK(&so->so_rcv); 2763 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2764 th->th_urp = 0; /* XXX */ 2765 thflags &= ~TH_URG; /* XXX */ 2766 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2767 goto dodata; /* XXX */ 2768 } 2769 /* 2770 * If this segment advances the known urgent pointer, 2771 * then mark the data stream. This should not happen 2772 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2773 * a FIN has been received from the remote side. 2774 * In these states we ignore the URG. 2775 * 2776 * According to RFC961 (Assigned Protocols), 2777 * the urgent pointer points to the last octet 2778 * of urgent data. We continue, however, 2779 * to consider it to indicate the first octet 2780 * of data past the urgent section as the original 2781 * spec states (in one of two places). 2782 */ 2783 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2784 tp->rcv_up = th->th_seq + th->th_urp; 2785 so->so_oobmark = so->so_rcv.sb_cc + 2786 (tp->rcv_up - tp->rcv_nxt) - 1; 2787 if (so->so_oobmark == 0) 2788 so->so_rcv.sb_state |= SBS_RCVATMARK; 2789 sohasoutofband(so); 2790 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2791 } 2792 SOCKBUF_UNLOCK(&so->so_rcv); 2793 /* 2794 * Remove out of band data so doesn't get presented to user. 2795 * This can happen independent of advancing the URG pointer, 2796 * but if two URG's are pending at once, some out-of-band 2797 * data may creep in... ick. 2798 */ 2799 if (th->th_urp <= (u_long)tlen && 2800 !(so->so_options & SO_OOBINLINE)) { 2801 /* hdr drop is delayed */ 2802 tcp_pulloutofband(so, th, m, drop_hdrlen); 2803 } 2804 } else { 2805 /* 2806 * If no out of band data is expected, 2807 * pull receive urgent pointer along 2808 * with the receive window. 2809 */ 2810 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2811 tp->rcv_up = tp->rcv_nxt; 2812 } 2813 dodata: /* XXX */ 2814 INP_WLOCK_ASSERT(tp->t_inpcb); 2815 2816 /* 2817 * Process the segment text, merging it into the TCP sequencing queue, 2818 * and arranging for acknowledgment of receipt if necessary. 2819 * This process logically involves adjusting tp->rcv_wnd as data 2820 * is presented to the user (this happens in tcp_usrreq.c, 2821 * case PRU_RCVD). If a FIN has already been received on this 2822 * connection then we just ignore the text. 2823 */ 2824 if ((tlen || (thflags & TH_FIN)) && 2825 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2826 tcp_seq save_start = th->th_seq; 2827 m_adj(m, drop_hdrlen); /* delayed header drop */ 2828 /* 2829 * Insert segment which includes th into TCP reassembly queue 2830 * with control block tp. Set thflags to whether reassembly now 2831 * includes a segment with FIN. This handles the common case 2832 * inline (segment is the next to be received on an established 2833 * connection, and the queue is empty), avoiding linkage into 2834 * and removal from the queue and repetition of various 2835 * conversions. 2836 * Set DELACK for segments received in order, but ack 2837 * immediately when segments are out of order (so 2838 * fast retransmit can work). 2839 */ 2840 if (th->th_seq == tp->rcv_nxt && 2841 LIST_EMPTY(&tp->t_segq) && 2842 TCPS_HAVEESTABLISHED(tp->t_state)) { 2843 if (DELAY_ACK(tp)) 2844 tp->t_flags |= TF_DELACK; 2845 else 2846 tp->t_flags |= TF_ACKNOW; 2847 tp->rcv_nxt += tlen; 2848 thflags = th->th_flags & TH_FIN; 2849 TCPSTAT_INC(tcps_rcvpack); 2850 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2851 ND6_HINT(tp); 2852 SOCKBUF_LOCK(&so->so_rcv); 2853 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2854 m_freem(m); 2855 else 2856 sbappendstream_locked(&so->so_rcv, m); 2857 /* NB: sorwakeup_locked() does an implicit unlock. */ 2858 sorwakeup_locked(so); 2859 } else { 2860 /* 2861 * XXX: Due to the header drop above "th" is 2862 * theoretically invalid by now. Fortunately 2863 * m_adj() doesn't actually frees any mbufs 2864 * when trimming from the head. 2865 */ 2866 thflags = tcp_reass(tp, th, &tlen, m); 2867 tp->t_flags |= TF_ACKNOW; 2868 } 2869 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2870 tcp_update_sack_list(tp, save_start, save_start + tlen); 2871 #if 0 2872 /* 2873 * Note the amount of data that peer has sent into 2874 * our window, in order to estimate the sender's 2875 * buffer size. 2876 * XXX: Unused. 2877 */ 2878 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2879 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2880 else 2881 len = so->so_rcv.sb_hiwat; 2882 #endif 2883 } else { 2884 m_freem(m); 2885 thflags &= ~TH_FIN; 2886 } 2887 2888 /* 2889 * If FIN is received ACK the FIN and let the user know 2890 * that the connection is closing. 2891 */ 2892 if (thflags & TH_FIN) { 2893 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2894 socantrcvmore(so); 2895 /* 2896 * If connection is half-synchronized 2897 * (ie NEEDSYN flag on) then delay ACK, 2898 * so it may be piggybacked when SYN is sent. 2899 * Otherwise, since we received a FIN then no 2900 * more input can be expected, send ACK now. 2901 */ 2902 if (tp->t_flags & TF_NEEDSYN) 2903 tp->t_flags |= TF_DELACK; 2904 else 2905 tp->t_flags |= TF_ACKNOW; 2906 tp->rcv_nxt++; 2907 } 2908 switch (tp->t_state) { 2909 2910 /* 2911 * In SYN_RECEIVED and ESTABLISHED STATES 2912 * enter the CLOSE_WAIT state. 2913 */ 2914 case TCPS_SYN_RECEIVED: 2915 tp->t_starttime = ticks; 2916 /* FALLTHROUGH */ 2917 case TCPS_ESTABLISHED: 2918 tp->t_state = TCPS_CLOSE_WAIT; 2919 break; 2920 2921 /* 2922 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2923 * enter the CLOSING state. 2924 */ 2925 case TCPS_FIN_WAIT_1: 2926 tp->t_state = TCPS_CLOSING; 2927 break; 2928 2929 /* 2930 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2931 * starting the time-wait timer, turning off the other 2932 * standard timers. 2933 */ 2934 case TCPS_FIN_WAIT_2: 2935 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2936 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2937 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2938 ti_locked)); 2939 2940 tcp_twstart(tp); 2941 INP_INFO_WUNLOCK(&V_tcbinfo); 2942 return; 2943 } 2944 } 2945 if (ti_locked == TI_WLOCKED) 2946 INP_INFO_WUNLOCK(&V_tcbinfo); 2947 ti_locked = TI_UNLOCKED; 2948 2949 #ifdef TCPDEBUG 2950 if (so->so_options & SO_DEBUG) 2951 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2952 &tcp_savetcp, 0); 2953 #endif 2954 2955 /* 2956 * Return any desired output. 2957 */ 2958 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2959 (void) tcp_output(tp); 2960 2961 check_delack: 2962 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2963 __func__, ti_locked)); 2964 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2965 INP_WLOCK_ASSERT(tp->t_inpcb); 2966 2967 if (tp->t_flags & TF_DELACK) { 2968 tp->t_flags &= ~TF_DELACK; 2969 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2970 } 2971 INP_WUNLOCK(tp->t_inpcb); 2972 return; 2973 2974 dropafterack: 2975 /* 2976 * Generate an ACK dropping incoming segment if it occupies 2977 * sequence space, where the ACK reflects our state. 2978 * 2979 * We can now skip the test for the RST flag since all 2980 * paths to this code happen after packets containing 2981 * RST have been dropped. 2982 * 2983 * In the SYN-RECEIVED state, don't send an ACK unless the 2984 * segment we received passes the SYN-RECEIVED ACK test. 2985 * If it fails send a RST. This breaks the loop in the 2986 * "LAND" DoS attack, and also prevents an ACK storm 2987 * between two listening ports that have been sent forged 2988 * SYN segments, each with the source address of the other. 2989 */ 2990 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2991 (SEQ_GT(tp->snd_una, th->th_ack) || 2992 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2993 rstreason = BANDLIM_RST_OPENPORT; 2994 goto dropwithreset; 2995 } 2996 #ifdef TCPDEBUG 2997 if (so->so_options & SO_DEBUG) 2998 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2999 &tcp_savetcp, 0); 3000 #endif 3001 if (ti_locked == TI_WLOCKED) 3002 INP_INFO_WUNLOCK(&V_tcbinfo); 3003 ti_locked = TI_UNLOCKED; 3004 3005 tp->t_flags |= TF_ACKNOW; 3006 (void) tcp_output(tp); 3007 INP_WUNLOCK(tp->t_inpcb); 3008 m_freem(m); 3009 return; 3010 3011 dropwithreset: 3012 if (ti_locked == TI_WLOCKED) 3013 INP_INFO_WUNLOCK(&V_tcbinfo); 3014 ti_locked = TI_UNLOCKED; 3015 3016 if (tp != NULL) { 3017 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3018 INP_WUNLOCK(tp->t_inpcb); 3019 } else 3020 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3021 return; 3022 3023 drop: 3024 if (ti_locked == TI_WLOCKED) { 3025 INP_INFO_WUNLOCK(&V_tcbinfo); 3026 ti_locked = TI_UNLOCKED; 3027 } 3028 #ifdef INVARIANTS 3029 else 3030 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3031 #endif 3032 3033 /* 3034 * Drop space held by incoming segment and return. 3035 */ 3036 #ifdef TCPDEBUG 3037 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3038 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3039 &tcp_savetcp, 0); 3040 #endif 3041 if (tp != NULL) 3042 INP_WUNLOCK(tp->t_inpcb); 3043 m_freem(m); 3044 } 3045 3046 /* 3047 * Issue RST and make ACK acceptable to originator of segment. 3048 * The mbuf must still include the original packet header. 3049 * tp may be NULL. 3050 */ 3051 static void 3052 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3053 int tlen, int rstreason) 3054 { 3055 #ifdef INET 3056 struct ip *ip; 3057 #endif 3058 #ifdef INET6 3059 struct ip6_hdr *ip6; 3060 #endif 3061 3062 if (tp != NULL) { 3063 INP_WLOCK_ASSERT(tp->t_inpcb); 3064 } 3065 3066 /* Don't bother if destination was broadcast/multicast. */ 3067 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3068 goto drop; 3069 #ifdef INET6 3070 if (mtod(m, struct ip *)->ip_v == 6) { 3071 ip6 = mtod(m, struct ip6_hdr *); 3072 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3073 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3074 goto drop; 3075 /* IPv6 anycast check is done at tcp6_input() */ 3076 } 3077 #endif 3078 #if defined(INET) && defined(INET6) 3079 else 3080 #endif 3081 #ifdef INET 3082 { 3083 ip = mtod(m, struct ip *); 3084 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3085 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3086 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3087 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3088 goto drop; 3089 } 3090 #endif 3091 3092 /* Perform bandwidth limiting. */ 3093 if (badport_bandlim(rstreason) < 0) 3094 goto drop; 3095 3096 /* tcp_respond consumes the mbuf chain. */ 3097 if (th->th_flags & TH_ACK) { 3098 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3099 th->th_ack, TH_RST); 3100 } else { 3101 if (th->th_flags & TH_SYN) 3102 tlen++; 3103 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3104 (tcp_seq)0, TH_RST|TH_ACK); 3105 } 3106 return; 3107 drop: 3108 m_freem(m); 3109 } 3110 3111 /* 3112 * Parse TCP options and place in tcpopt. 3113 */ 3114 static void 3115 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3116 { 3117 int opt, optlen; 3118 3119 to->to_flags = 0; 3120 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3121 opt = cp[0]; 3122 if (opt == TCPOPT_EOL) 3123 break; 3124 if (opt == TCPOPT_NOP) 3125 optlen = 1; 3126 else { 3127 if (cnt < 2) 3128 break; 3129 optlen = cp[1]; 3130 if (optlen < 2 || optlen > cnt) 3131 break; 3132 } 3133 switch (opt) { 3134 case TCPOPT_MAXSEG: 3135 if (optlen != TCPOLEN_MAXSEG) 3136 continue; 3137 if (!(flags & TO_SYN)) 3138 continue; 3139 to->to_flags |= TOF_MSS; 3140 bcopy((char *)cp + 2, 3141 (char *)&to->to_mss, sizeof(to->to_mss)); 3142 to->to_mss = ntohs(to->to_mss); 3143 break; 3144 case TCPOPT_WINDOW: 3145 if (optlen != TCPOLEN_WINDOW) 3146 continue; 3147 if (!(flags & TO_SYN)) 3148 continue; 3149 to->to_flags |= TOF_SCALE; 3150 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3151 break; 3152 case TCPOPT_TIMESTAMP: 3153 if (optlen != TCPOLEN_TIMESTAMP) 3154 continue; 3155 to->to_flags |= TOF_TS; 3156 bcopy((char *)cp + 2, 3157 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3158 to->to_tsval = ntohl(to->to_tsval); 3159 bcopy((char *)cp + 6, 3160 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3161 to->to_tsecr = ntohl(to->to_tsecr); 3162 break; 3163 #ifdef TCP_SIGNATURE 3164 /* 3165 * XXX In order to reply to a host which has set the 3166 * TCP_SIGNATURE option in its initial SYN, we have to 3167 * record the fact that the option was observed here 3168 * for the syncache code to perform the correct response. 3169 */ 3170 case TCPOPT_SIGNATURE: 3171 if (optlen != TCPOLEN_SIGNATURE) 3172 continue; 3173 to->to_flags |= TOF_SIGNATURE; 3174 to->to_signature = cp + 2; 3175 break; 3176 #endif 3177 case TCPOPT_SACK_PERMITTED: 3178 if (optlen != TCPOLEN_SACK_PERMITTED) 3179 continue; 3180 if (!(flags & TO_SYN)) 3181 continue; 3182 if (!V_tcp_do_sack) 3183 continue; 3184 to->to_flags |= TOF_SACKPERM; 3185 break; 3186 case TCPOPT_SACK: 3187 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3188 continue; 3189 if (flags & TO_SYN) 3190 continue; 3191 to->to_flags |= TOF_SACK; 3192 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3193 to->to_sacks = cp + 2; 3194 TCPSTAT_INC(tcps_sack_rcv_blocks); 3195 break; 3196 default: 3197 continue; 3198 } 3199 } 3200 } 3201 3202 /* 3203 * Pull out of band byte out of a segment so 3204 * it doesn't appear in the user's data queue. 3205 * It is still reflected in the segment length for 3206 * sequencing purposes. 3207 */ 3208 static void 3209 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3210 int off) 3211 { 3212 int cnt = off + th->th_urp - 1; 3213 3214 while (cnt >= 0) { 3215 if (m->m_len > cnt) { 3216 char *cp = mtod(m, caddr_t) + cnt; 3217 struct tcpcb *tp = sototcpcb(so); 3218 3219 INP_WLOCK_ASSERT(tp->t_inpcb); 3220 3221 tp->t_iobc = *cp; 3222 tp->t_oobflags |= TCPOOB_HAVEDATA; 3223 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3224 m->m_len--; 3225 if (m->m_flags & M_PKTHDR) 3226 m->m_pkthdr.len--; 3227 return; 3228 } 3229 cnt -= m->m_len; 3230 m = m->m_next; 3231 if (m == NULL) 3232 break; 3233 } 3234 panic("tcp_pulloutofband"); 3235 } 3236 3237 /* 3238 * Collect new round-trip time estimate 3239 * and update averages and current timeout. 3240 */ 3241 static void 3242 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3243 { 3244 int delta; 3245 3246 INP_WLOCK_ASSERT(tp->t_inpcb); 3247 3248 TCPSTAT_INC(tcps_rttupdated); 3249 tp->t_rttupdated++; 3250 if (tp->t_srtt != 0) { 3251 /* 3252 * srtt is stored as fixed point with 5 bits after the 3253 * binary point (i.e., scaled by 8). The following magic 3254 * is equivalent to the smoothing algorithm in rfc793 with 3255 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3256 * point). Adjust rtt to origin 0. 3257 */ 3258 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3259 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3260 3261 if ((tp->t_srtt += delta) <= 0) 3262 tp->t_srtt = 1; 3263 3264 /* 3265 * We accumulate a smoothed rtt variance (actually, a 3266 * smoothed mean difference), then set the retransmit 3267 * timer to smoothed rtt + 4 times the smoothed variance. 3268 * rttvar is stored as fixed point with 4 bits after the 3269 * binary point (scaled by 16). The following is 3270 * equivalent to rfc793 smoothing with an alpha of .75 3271 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3272 * rfc793's wired-in beta. 3273 */ 3274 if (delta < 0) 3275 delta = -delta; 3276 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3277 if ((tp->t_rttvar += delta) <= 0) 3278 tp->t_rttvar = 1; 3279 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3280 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3281 } else { 3282 /* 3283 * No rtt measurement yet - use the unsmoothed rtt. 3284 * Set the variance to half the rtt (so our first 3285 * retransmit happens at 3*rtt). 3286 */ 3287 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3288 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3289 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3290 } 3291 tp->t_rtttime = 0; 3292 tp->t_rxtshift = 0; 3293 3294 /* 3295 * the retransmit should happen at rtt + 4 * rttvar. 3296 * Because of the way we do the smoothing, srtt and rttvar 3297 * will each average +1/2 tick of bias. When we compute 3298 * the retransmit timer, we want 1/2 tick of rounding and 3299 * 1 extra tick because of +-1/2 tick uncertainty in the 3300 * firing of the timer. The bias will give us exactly the 3301 * 1.5 tick we need. But, because the bias is 3302 * statistical, we have to test that we don't drop below 3303 * the minimum feasible timer (which is 2 ticks). 3304 */ 3305 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3306 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3307 3308 /* 3309 * We received an ack for a packet that wasn't retransmitted; 3310 * it is probably safe to discard any error indications we've 3311 * received recently. This isn't quite right, but close enough 3312 * for now (a route might have failed after we sent a segment, 3313 * and the return path might not be symmetrical). 3314 */ 3315 tp->t_softerror = 0; 3316 } 3317 3318 /* 3319 * Determine a reasonable value for maxseg size. 3320 * If the route is known, check route for mtu. 3321 * If none, use an mss that can be handled on the outgoing 3322 * interface without forcing IP to fragment; if bigger than 3323 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 3324 * to utilize large mbufs. If no route is found, route has no mtu, 3325 * or the destination isn't local, use a default, hopefully conservative 3326 * size (usually 512 or the default IP max size, but no more than the mtu 3327 * of the interface), as we can't discover anything about intervening 3328 * gateways or networks. We also initialize the congestion/slow start 3329 * window to be a single segment if the destination isn't local. 3330 * While looking at the routing entry, we also initialize other path-dependent 3331 * parameters from pre-set or cached values in the routing entry. 3332 * 3333 * Also take into account the space needed for options that we 3334 * send regularly. Make maxseg shorter by that amount to assure 3335 * that we can send maxseg amount of data even when the options 3336 * are present. Store the upper limit of the length of options plus 3337 * data in maxopd. 3338 * 3339 * NOTE that this routine is only called when we process an incoming 3340 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3341 * settings are handled in tcp_mssopt(). 3342 */ 3343 void 3344 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3345 struct hc_metrics_lite *metricptr, int *mtuflags) 3346 { 3347 int mss = 0; 3348 u_long maxmtu = 0; 3349 struct inpcb *inp = tp->t_inpcb; 3350 struct hc_metrics_lite metrics; 3351 int origoffer; 3352 #ifdef INET6 3353 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3354 size_t min_protoh = isipv6 ? 3355 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3356 sizeof (struct tcpiphdr); 3357 #else 3358 const size_t min_protoh = sizeof(struct tcpiphdr); 3359 #endif 3360 3361 INP_WLOCK_ASSERT(tp->t_inpcb); 3362 3363 if (mtuoffer != -1) { 3364 KASSERT(offer == -1, ("%s: conflict", __func__)); 3365 offer = mtuoffer - min_protoh; 3366 } 3367 origoffer = offer; 3368 3369 /* Initialize. */ 3370 #ifdef INET6 3371 if (isipv6) { 3372 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags); 3373 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3374 } 3375 #endif 3376 #if defined(INET) && defined(INET6) 3377 else 3378 #endif 3379 #ifdef INET 3380 { 3381 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags); 3382 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3383 } 3384 #endif 3385 3386 /* 3387 * No route to sender, stay with default mss and return. 3388 */ 3389 if (maxmtu == 0) { 3390 /* 3391 * In case we return early we need to initialize metrics 3392 * to a defined state as tcp_hc_get() would do for us 3393 * if there was no cache hit. 3394 */ 3395 if (metricptr != NULL) 3396 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3397 return; 3398 } 3399 3400 /* What have we got? */ 3401 switch (offer) { 3402 case 0: 3403 /* 3404 * Offer == 0 means that there was no MSS on the SYN 3405 * segment, in this case we use tcp_mssdflt as 3406 * already assigned to t_maxopd above. 3407 */ 3408 offer = tp->t_maxopd; 3409 break; 3410 3411 case -1: 3412 /* 3413 * Offer == -1 means that we didn't receive SYN yet. 3414 */ 3415 /* FALLTHROUGH */ 3416 3417 default: 3418 /* 3419 * Prevent DoS attack with too small MSS. Round up 3420 * to at least minmss. 3421 */ 3422 offer = max(offer, V_tcp_minmss); 3423 } 3424 3425 /* 3426 * rmx information is now retrieved from tcp_hostcache. 3427 */ 3428 tcp_hc_get(&inp->inp_inc, &metrics); 3429 if (metricptr != NULL) 3430 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3431 3432 /* 3433 * If there's a discovered mtu int tcp hostcache, use it 3434 * else, use the link mtu. 3435 */ 3436 if (metrics.rmx_mtu) 3437 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3438 else { 3439 #ifdef INET6 3440 if (isipv6) { 3441 mss = maxmtu - min_protoh; 3442 if (!V_path_mtu_discovery && 3443 !in6_localaddr(&inp->in6p_faddr)) 3444 mss = min(mss, V_tcp_v6mssdflt); 3445 } 3446 #endif 3447 #if defined(INET) && defined(INET6) 3448 else 3449 #endif 3450 #ifdef INET 3451 { 3452 mss = maxmtu - min_protoh; 3453 if (!V_path_mtu_discovery && 3454 !in_localaddr(inp->inp_faddr)) 3455 mss = min(mss, V_tcp_mssdflt); 3456 } 3457 #endif 3458 /* 3459 * XXX - The above conditional (mss = maxmtu - min_protoh) 3460 * probably violates the TCP spec. 3461 * The problem is that, since we don't know the 3462 * other end's MSS, we are supposed to use a conservative 3463 * default. But, if we do that, then MTU discovery will 3464 * never actually take place, because the conservative 3465 * default is much less than the MTUs typically seen 3466 * on the Internet today. For the moment, we'll sweep 3467 * this under the carpet. 3468 * 3469 * The conservative default might not actually be a problem 3470 * if the only case this occurs is when sending an initial 3471 * SYN with options and data to a host we've never talked 3472 * to before. Then, they will reply with an MSS value which 3473 * will get recorded and the new parameters should get 3474 * recomputed. For Further Study. 3475 */ 3476 } 3477 mss = min(mss, offer); 3478 3479 /* 3480 * Sanity check: make sure that maxopd will be large 3481 * enough to allow some data on segments even if the 3482 * all the option space is used (40bytes). Otherwise 3483 * funny things may happen in tcp_output. 3484 */ 3485 mss = max(mss, 64); 3486 3487 /* 3488 * maxopd stores the maximum length of data AND options 3489 * in a segment; maxseg is the amount of data in a normal 3490 * segment. We need to store this value (maxopd) apart 3491 * from maxseg, because now every segment carries options 3492 * and thus we normally have somewhat less data in segments. 3493 */ 3494 tp->t_maxopd = mss; 3495 3496 /* 3497 * origoffer==-1 indicates that no segments were received yet. 3498 * In this case we just guess. 3499 */ 3500 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3501 (origoffer == -1 || 3502 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3503 mss -= TCPOLEN_TSTAMP_APPA; 3504 3505 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3506 if (mss > MCLBYTES) 3507 mss &= ~(MCLBYTES-1); 3508 #else 3509 if (mss > MCLBYTES) 3510 mss = mss / MCLBYTES * MCLBYTES; 3511 #endif 3512 tp->t_maxseg = mss; 3513 } 3514 3515 void 3516 tcp_mss(struct tcpcb *tp, int offer) 3517 { 3518 int mss; 3519 u_long bufsize; 3520 struct inpcb *inp; 3521 struct socket *so; 3522 struct hc_metrics_lite metrics; 3523 int mtuflags = 0; 3524 3525 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3526 3527 tcp_mss_update(tp, offer, -1, &metrics, &mtuflags); 3528 3529 mss = tp->t_maxseg; 3530 inp = tp->t_inpcb; 3531 3532 /* 3533 * If there's a pipesize, change the socket buffer to that size, 3534 * don't change if sb_hiwat is different than default (then it 3535 * has been changed on purpose with setsockopt). 3536 * Make the socket buffers an integral number of mss units; 3537 * if the mss is larger than the socket buffer, decrease the mss. 3538 */ 3539 so = inp->inp_socket; 3540 SOCKBUF_LOCK(&so->so_snd); 3541 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3542 bufsize = metrics.rmx_sendpipe; 3543 else 3544 bufsize = so->so_snd.sb_hiwat; 3545 if (bufsize < mss) 3546 mss = bufsize; 3547 else { 3548 bufsize = roundup(bufsize, mss); 3549 if (bufsize > sb_max) 3550 bufsize = sb_max; 3551 if (bufsize > so->so_snd.sb_hiwat) 3552 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3553 } 3554 SOCKBUF_UNLOCK(&so->so_snd); 3555 tp->t_maxseg = mss; 3556 3557 SOCKBUF_LOCK(&so->so_rcv); 3558 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3559 bufsize = metrics.rmx_recvpipe; 3560 else 3561 bufsize = so->so_rcv.sb_hiwat; 3562 if (bufsize > mss) { 3563 bufsize = roundup(bufsize, mss); 3564 if (bufsize > sb_max) 3565 bufsize = sb_max; 3566 if (bufsize > so->so_rcv.sb_hiwat) 3567 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3568 } 3569 SOCKBUF_UNLOCK(&so->so_rcv); 3570 3571 /* Check the interface for TSO capabilities. */ 3572 if (mtuflags & CSUM_TSO) 3573 tp->t_flags |= TF_TSO; 3574 } 3575 3576 /* 3577 * Determine the MSS option to send on an outgoing SYN. 3578 */ 3579 int 3580 tcp_mssopt(struct in_conninfo *inc) 3581 { 3582 int mss = 0; 3583 u_long maxmtu = 0; 3584 u_long thcmtu = 0; 3585 size_t min_protoh; 3586 3587 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3588 3589 #ifdef INET6 3590 if (inc->inc_flags & INC_ISIPV6) { 3591 mss = V_tcp_v6mssdflt; 3592 maxmtu = tcp_maxmtu6(inc, NULL); 3593 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3594 } 3595 #endif 3596 #if defined(INET) && defined(INET6) 3597 else 3598 #endif 3599 #ifdef INET 3600 { 3601 mss = V_tcp_mssdflt; 3602 maxmtu = tcp_maxmtu(inc, NULL); 3603 min_protoh = sizeof(struct tcpiphdr); 3604 } 3605 #endif 3606 #if defined(INET6) || defined(INET) 3607 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3608 #endif 3609 3610 if (maxmtu && thcmtu) 3611 mss = min(maxmtu, thcmtu) - min_protoh; 3612 else if (maxmtu || thcmtu) 3613 mss = max(maxmtu, thcmtu) - min_protoh; 3614 3615 return (mss); 3616 } 3617 3618 3619 /* 3620 * On a partial ack arrives, force the retransmission of the 3621 * next unacknowledged segment. Do not clear tp->t_dupacks. 3622 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3623 * be started again. 3624 */ 3625 static void 3626 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3627 { 3628 tcp_seq onxt = tp->snd_nxt; 3629 u_long ocwnd = tp->snd_cwnd; 3630 3631 INP_WLOCK_ASSERT(tp->t_inpcb); 3632 3633 tcp_timer_activate(tp, TT_REXMT, 0); 3634 tp->t_rtttime = 0; 3635 tp->snd_nxt = th->th_ack; 3636 /* 3637 * Set snd_cwnd to one segment beyond acknowledged offset. 3638 * (tp->snd_una has not yet been updated when this function is called.) 3639 */ 3640 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3641 tp->t_flags |= TF_ACKNOW; 3642 (void) tcp_output(tp); 3643 tp->snd_cwnd = ocwnd; 3644 if (SEQ_GT(onxt, tp->snd_nxt)) 3645 tp->snd_nxt = onxt; 3646 /* 3647 * Partial window deflation. Relies on fact that tp->snd_una 3648 * not updated yet. 3649 */ 3650 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3651 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3652 else 3653 tp->snd_cwnd = 0; 3654 tp->snd_cwnd += tp->t_maxseg; 3655 }
Cache object: de629295524abde6c678926519f28116
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]