1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 4. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_ipfw.h" /* for ipfw_fwd */ 54 #include "opt_inet.h" 55 #include "opt_inet6.h" 56 #include "opt_ipsec.h" 57 #include "opt_tcpdebug.h" 58 59 #include <sys/param.h> 60 #include <sys/kernel.h> 61 #include <sys/hhook.h> 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/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) = 256*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 } 546 547 tcp_input(m, *offp); 548 return IPPROTO_DONE; 549 } 550 #endif /* INET6 */ 551 552 void 553 tcp_input(struct mbuf *m, int off0) 554 { 555 struct tcphdr *th = NULL; 556 struct ip *ip = NULL; 557 #ifdef INET 558 struct ipovly *ipov; 559 #endif 560 struct inpcb *inp = NULL; 561 struct tcpcb *tp = NULL; 562 struct socket *so = NULL; 563 u_char *optp = NULL; 564 int optlen = 0; 565 #ifdef INET 566 int len; 567 #endif 568 int tlen = 0, off; 569 int drop_hdrlen; 570 int thflags; 571 int rstreason = 0; /* For badport_bandlim accounting purposes */ 572 #ifdef TCP_SIGNATURE 573 uint8_t sig_checked = 0; 574 #endif 575 uint8_t iptos = 0; 576 #ifdef INET 577 #ifdef IPFIREWALL_FORWARD 578 struct m_tag *fwd_tag; 579 #endif 580 #endif /* INET */ 581 #ifdef INET6 582 struct ip6_hdr *ip6 = NULL; 583 int isipv6; 584 #else 585 const void *ip6 = NULL; 586 #if (defined(INET) && defined(IPFIREWALL_FORWARD)) || defined(TCPDEBUG) 587 const int isipv6 = 0; 588 #endif 589 #endif /* INET6 */ 590 struct tcpopt to; /* options in this segment */ 591 char *s = NULL; /* address and port logging */ 592 int ti_locked; 593 #define TI_UNLOCKED 1 594 #define TI_WLOCKED 2 595 596 #ifdef TCPDEBUG 597 /* 598 * The size of tcp_saveipgen must be the size of the max ip header, 599 * now IPv6. 600 */ 601 u_char tcp_saveipgen[IP6_HDR_LEN]; 602 struct tcphdr tcp_savetcp; 603 short ostate = 0; 604 #endif 605 606 #ifdef INET6 607 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 608 #endif 609 610 to.to_flags = 0; 611 TCPSTAT_INC(tcps_rcvtotal); 612 613 #ifdef INET6 614 if (isipv6) { 615 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 616 ip6 = mtod(m, struct ip6_hdr *); 617 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 618 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 619 TCPSTAT_INC(tcps_rcvbadsum); 620 goto drop; 621 } 622 th = (struct tcphdr *)((caddr_t)ip6 + off0); 623 624 /* 625 * Be proactive about unspecified IPv6 address in source. 626 * As we use all-zero to indicate unbounded/unconnected pcb, 627 * unspecified IPv6 address can be used to confuse us. 628 * 629 * Note that packets with unspecified IPv6 destination is 630 * already dropped in ip6_input. 631 */ 632 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 633 /* XXX stat */ 634 goto drop; 635 } 636 } 637 #endif 638 #if defined(INET) && defined(INET6) 639 else 640 #endif 641 #ifdef INET 642 { 643 /* 644 * Get IP and TCP header together in first mbuf. 645 * Note: IP leaves IP header in first mbuf. 646 */ 647 if (off0 > sizeof (struct ip)) { 648 ip_stripoptions(m, (struct mbuf *)0); 649 off0 = sizeof(struct ip); 650 } 651 if (m->m_len < sizeof (struct tcpiphdr)) { 652 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 653 == NULL) { 654 TCPSTAT_INC(tcps_rcvshort); 655 return; 656 } 657 } 658 ip = mtod(m, struct ip *); 659 ipov = (struct ipovly *)ip; 660 th = (struct tcphdr *)((caddr_t)ip + off0); 661 tlen = ip->ip_len; 662 663 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 664 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 665 th->th_sum = m->m_pkthdr.csum_data; 666 else 667 th->th_sum = in_pseudo(ip->ip_src.s_addr, 668 ip->ip_dst.s_addr, 669 htonl(m->m_pkthdr.csum_data + 670 ip->ip_len + 671 IPPROTO_TCP)); 672 th->th_sum ^= 0xffff; 673 #ifdef TCPDEBUG 674 ipov->ih_len = (u_short)tlen; 675 ipov->ih_len = htons(ipov->ih_len); 676 #endif 677 } else { 678 /* 679 * Checksum extended TCP header and data. 680 */ 681 len = sizeof (struct ip) + tlen; 682 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 683 ipov->ih_len = (u_short)tlen; 684 ipov->ih_len = htons(ipov->ih_len); 685 th->th_sum = in_cksum(m, len); 686 } 687 if (th->th_sum) { 688 TCPSTAT_INC(tcps_rcvbadsum); 689 goto drop; 690 } 691 /* Re-initialization for later version check */ 692 ip->ip_v = IPVERSION; 693 } 694 #endif /* INET */ 695 696 #ifdef INET6 697 if (isipv6) 698 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 699 #endif 700 #if defined(INET) && defined(INET6) 701 else 702 #endif 703 #ifdef INET 704 iptos = ip->ip_tos; 705 #endif 706 707 /* 708 * Check that TCP offset makes sense, 709 * pull out TCP options and adjust length. XXX 710 */ 711 off = th->th_off << 2; 712 if (off < sizeof (struct tcphdr) || off > tlen) { 713 TCPSTAT_INC(tcps_rcvbadoff); 714 goto drop; 715 } 716 tlen -= off; /* tlen is used instead of ti->ti_len */ 717 if (off > sizeof (struct tcphdr)) { 718 #ifdef INET6 719 if (isipv6) { 720 IP6_EXTHDR_CHECK(m, off0, off, ); 721 ip6 = mtod(m, struct ip6_hdr *); 722 th = (struct tcphdr *)((caddr_t)ip6 + off0); 723 } 724 #endif 725 #if defined(INET) && defined(INET6) 726 else 727 #endif 728 #ifdef INET 729 { 730 if (m->m_len < sizeof(struct ip) + off) { 731 if ((m = m_pullup(m, sizeof (struct ip) + off)) 732 == NULL) { 733 TCPSTAT_INC(tcps_rcvshort); 734 return; 735 } 736 ip = mtod(m, struct ip *); 737 ipov = (struct ipovly *)ip; 738 th = (struct tcphdr *)((caddr_t)ip + off0); 739 } 740 } 741 #endif 742 optlen = off - sizeof (struct tcphdr); 743 optp = (u_char *)(th + 1); 744 } 745 thflags = th->th_flags; 746 747 /* 748 * Convert TCP protocol specific fields to host format. 749 */ 750 tcp_fields_to_host(th); 751 752 /* 753 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 754 */ 755 drop_hdrlen = off0 + off; 756 757 /* 758 * Locate pcb for segment; if we're likely to add or remove a 759 * connection then first acquire pcbinfo lock. There are two cases 760 * where we might discover later we need a write lock despite the 761 * flags: ACKs moving a connection out of the syncache, and ACKs for 762 * a connection in TIMEWAIT. 763 */ 764 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 765 INP_INFO_WLOCK(&V_tcbinfo); 766 ti_locked = TI_WLOCKED; 767 } else 768 ti_locked = TI_UNLOCKED; 769 770 findpcb: 771 #ifdef INVARIANTS 772 if (ti_locked == TI_WLOCKED) { 773 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 774 } else { 775 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 776 } 777 #endif 778 779 #ifdef INET 780 #ifdef IPFIREWALL_FORWARD 781 /* 782 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 783 */ 784 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 785 786 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 787 struct sockaddr_in *next_hop; 788 789 next_hop = (struct sockaddr_in *)(fwd_tag+1); 790 /* 791 * Transparently forwarded. Pretend to be the destination. 792 * already got one like this? 793 */ 794 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 795 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 796 m->m_pkthdr.rcvif, m); 797 if (!inp) { 798 /* 799 * It's new. Try to find the ambushing socket. 800 * Because we've rewritten the destination address, 801 * any hardware-generated hash is ignored. 802 */ 803 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 804 th->th_sport, next_hop->sin_addr, 805 next_hop->sin_port ? ntohs(next_hop->sin_port) : 806 th->th_dport, INPLOOKUP_WILDCARD | 807 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 808 } 809 /* Remove the tag from the packet. We don't need it anymore. */ 810 m_tag_delete(m, fwd_tag); 811 } else 812 #endif /* IPFIREWALL_FORWARD */ 813 #endif /* INET */ 814 { 815 #ifdef INET6 816 if (isipv6) 817 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 818 th->th_sport, &ip6->ip6_dst, th->th_dport, 819 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 820 m->m_pkthdr.rcvif, m); 821 #endif 822 #if defined(INET) && defined(INET6) 823 else 824 #endif 825 #ifdef INET 826 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 827 th->th_sport, ip->ip_dst, th->th_dport, 828 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 829 m->m_pkthdr.rcvif, m); 830 #endif 831 } 832 833 /* 834 * If the INPCB does not exist then all data in the incoming 835 * segment is discarded and an appropriate RST is sent back. 836 * XXX MRT Send RST using which routing table? 837 */ 838 if (inp == NULL) { 839 /* 840 * Log communication attempts to ports that are not 841 * in use. 842 */ 843 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 844 tcp_log_in_vain == 2) { 845 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 846 log(LOG_INFO, "%s; %s: Connection attempt " 847 "to closed port\n", s, __func__); 848 } 849 /* 850 * When blackholing do not respond with a RST but 851 * completely ignore the segment and drop it. 852 */ 853 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 854 V_blackhole == 2) 855 goto dropunlock; 856 857 rstreason = BANDLIM_RST_CLOSEDPORT; 858 goto dropwithreset; 859 } 860 INP_WLOCK_ASSERT(inp); 861 if (!(inp->inp_flags & INP_HW_FLOWID) 862 && (m->m_flags & M_FLOWID) 863 && ((inp->inp_socket == NULL) 864 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 865 inp->inp_flags |= INP_HW_FLOWID; 866 inp->inp_flags &= ~INP_SW_FLOWID; 867 inp->inp_flowid = m->m_pkthdr.flowid; 868 } 869 #ifdef IPSEC 870 #ifdef INET6 871 if (isipv6 && ipsec6_in_reject(m, inp)) { 872 V_ipsec6stat.in_polvio++; 873 goto dropunlock; 874 } else 875 #endif /* INET6 */ 876 if (ipsec4_in_reject(m, inp) != 0) { 877 V_ipsec4stat.in_polvio++; 878 goto dropunlock; 879 } 880 #endif /* IPSEC */ 881 882 /* 883 * Check the minimum TTL for socket. 884 */ 885 if (inp->inp_ip_minttl != 0) { 886 #ifdef INET6 887 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 888 goto dropunlock; 889 else 890 #endif 891 if (inp->inp_ip_minttl > ip->ip_ttl) 892 goto dropunlock; 893 } 894 895 /* 896 * A previous connection in TIMEWAIT state is supposed to catch stray 897 * or duplicate segments arriving late. If this segment was a 898 * legitimate new connection attempt the old INPCB gets removed and 899 * we can try again to find a listening socket. 900 * 901 * At this point, due to earlier optimism, we may hold only an inpcb 902 * lock, and not the inpcbinfo write lock. If so, we need to try to 903 * acquire it, or if that fails, acquire a reference on the inpcb, 904 * drop all locks, acquire a global write lock, and then re-acquire 905 * the inpcb lock. We may at that point discover that another thread 906 * has tried to free the inpcb, in which case we need to loop back 907 * and try to find a new inpcb to deliver to. 908 * 909 * XXXRW: It may be time to rethink timewait locking. 910 */ 911 relocked: 912 if (inp->inp_flags & INP_TIMEWAIT) { 913 if (ti_locked == TI_UNLOCKED) { 914 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 915 in_pcbref(inp); 916 INP_WUNLOCK(inp); 917 INP_INFO_WLOCK(&V_tcbinfo); 918 ti_locked = TI_WLOCKED; 919 INP_WLOCK(inp); 920 if (in_pcbrele_wlocked(inp)) { 921 inp = NULL; 922 goto findpcb; 923 } 924 } else 925 ti_locked = TI_WLOCKED; 926 } 927 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 928 929 #ifdef TCP_SIGNATURE 930 tcp_dooptions(&to, optp, optlen, 931 (thflags & TH_SYN) ? TO_SYN : 0); 932 if (sig_checked == 0) { 933 tp = intotcpcb(inp); 934 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 935 rstreason = BANDLIM_RST_CLOSEDPORT; 936 goto dropwithreset; 937 } 938 if (!tcp_signature_verify_input(m, off0, tlen, optlen, 939 &to, th, tp->t_flags)) 940 goto dropunlock; 941 sig_checked = 1; 942 } 943 #else 944 if (thflags & TH_SYN) 945 tcp_dooptions(&to, optp, optlen, TO_SYN); 946 #endif 947 /* 948 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 949 */ 950 if (tcp_twcheck(inp, &to, th, m, tlen)) 951 goto findpcb; 952 INP_INFO_WUNLOCK(&V_tcbinfo); 953 return; 954 } 955 /* 956 * The TCPCB may no longer exist if the connection is winding 957 * down or it is in the CLOSED state. Either way we drop the 958 * segment and send an appropriate response. 959 */ 960 tp = intotcpcb(inp); 961 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 962 rstreason = BANDLIM_RST_CLOSEDPORT; 963 goto dropwithreset; 964 } 965 966 /* 967 * We've identified a valid inpcb, but it could be that we need an 968 * inpcbinfo write lock but don't hold it. In this case, attempt to 969 * acquire using the same strategy as the TIMEWAIT case above. If we 970 * relock, we have to jump back to 'relocked' as the connection might 971 * now be in TIMEWAIT. 972 */ 973 #ifdef INVARIANTS 974 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 975 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 976 #endif 977 if (tp->t_state != TCPS_ESTABLISHED) { 978 if (ti_locked == TI_UNLOCKED) { 979 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 980 in_pcbref(inp); 981 INP_WUNLOCK(inp); 982 INP_INFO_WLOCK(&V_tcbinfo); 983 ti_locked = TI_WLOCKED; 984 INP_WLOCK(inp); 985 if (in_pcbrele_wlocked(inp)) { 986 inp = NULL; 987 goto findpcb; 988 } 989 goto relocked; 990 } else 991 ti_locked = TI_WLOCKED; 992 } 993 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 994 } 995 996 #ifdef MAC 997 INP_WLOCK_ASSERT(inp); 998 if (mac_inpcb_check_deliver(inp, m)) 999 goto dropunlock; 1000 #endif 1001 so = inp->inp_socket; 1002 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1003 #ifdef TCPDEBUG 1004 if (so->so_options & SO_DEBUG) { 1005 ostate = tp->t_state; 1006 if (isipv6) { 1007 #ifdef INET6 1008 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1009 #endif 1010 } else 1011 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1012 tcp_savetcp = *th; 1013 } 1014 #endif /* TCPDEBUG */ 1015 /* 1016 * When the socket is accepting connections (the INPCB is in LISTEN 1017 * state) we look into the SYN cache if this is a new connection 1018 * attempt or the completion of a previous one. Because listen 1019 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1020 * held in this case. 1021 */ 1022 if (so->so_options & SO_ACCEPTCONN) { 1023 struct in_conninfo inc; 1024 1025 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1026 "tp not listening", __func__)); 1027 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1028 1029 bzero(&inc, sizeof(inc)); 1030 #ifdef INET6 1031 if (isipv6) { 1032 inc.inc_flags |= INC_ISIPV6; 1033 inc.inc6_faddr = ip6->ip6_src; 1034 inc.inc6_laddr = ip6->ip6_dst; 1035 } else 1036 #endif 1037 { 1038 inc.inc_faddr = ip->ip_src; 1039 inc.inc_laddr = ip->ip_dst; 1040 } 1041 inc.inc_fport = th->th_sport; 1042 inc.inc_lport = th->th_dport; 1043 inc.inc_fibnum = so->so_fibnum; 1044 1045 /* 1046 * Check for an existing connection attempt in syncache if 1047 * the flag is only ACK. A successful lookup creates a new 1048 * socket appended to the listen queue in SYN_RECEIVED state. 1049 */ 1050 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1051 /* 1052 * Parse the TCP options here because 1053 * syncookies need access to the reflected 1054 * timestamp. 1055 */ 1056 tcp_dooptions(&to, optp, optlen, 0); 1057 /* 1058 * NB: syncache_expand() doesn't unlock 1059 * inp and tcpinfo locks. 1060 */ 1061 if (!syncache_expand(&inc, &to, th, &so, m)) { 1062 /* 1063 * No syncache entry or ACK was not 1064 * for our SYN/ACK. Send a RST. 1065 * NB: syncache did its own logging 1066 * of the failure cause. 1067 */ 1068 rstreason = BANDLIM_RST_OPENPORT; 1069 goto dropwithreset; 1070 } 1071 if (so == NULL) { 1072 /* 1073 * We completed the 3-way handshake 1074 * but could not allocate a socket 1075 * either due to memory shortage, 1076 * listen queue length limits or 1077 * global socket limits. Send RST 1078 * or wait and have the remote end 1079 * retransmit the ACK for another 1080 * try. 1081 */ 1082 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1083 log(LOG_DEBUG, "%s; %s: Listen socket: " 1084 "Socket allocation failed due to " 1085 "limits or memory shortage, %s\n", 1086 s, __func__, 1087 V_tcp_sc_rst_sock_fail ? 1088 "sending RST" : "try again"); 1089 if (V_tcp_sc_rst_sock_fail) { 1090 rstreason = BANDLIM_UNLIMITED; 1091 goto dropwithreset; 1092 } else 1093 goto dropunlock; 1094 } 1095 /* 1096 * Socket is created in state SYN_RECEIVED. 1097 * Unlock the listen socket, lock the newly 1098 * created socket and update the tp variable. 1099 */ 1100 INP_WUNLOCK(inp); /* listen socket */ 1101 inp = sotoinpcb(so); 1102 INP_WLOCK(inp); /* new connection */ 1103 tp = intotcpcb(inp); 1104 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1105 ("%s: ", __func__)); 1106 #ifdef TCP_SIGNATURE 1107 if (sig_checked == 0) { 1108 tcp_dooptions(&to, optp, optlen, 1109 (thflags & TH_SYN) ? TO_SYN : 0); 1110 if (!tcp_signature_verify_input(m, off0, tlen, 1111 optlen, &to, th, tp->t_flags)) { 1112 1113 /* 1114 * In SYN_SENT state if it receives an 1115 * RST, it is allowed for further 1116 * processing. 1117 */ 1118 if ((thflags & TH_RST) == 0 || 1119 (tp->t_state == TCPS_SYN_SENT) == 0) 1120 goto dropunlock; 1121 } 1122 sig_checked = 1; 1123 } 1124 #endif 1125 1126 /* 1127 * Process the segment and the data it 1128 * contains. tcp_do_segment() consumes 1129 * the mbuf chain and unlocks the inpcb. 1130 */ 1131 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1132 iptos, ti_locked); 1133 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1134 return; 1135 } 1136 /* 1137 * Segment flag validation for new connection attempts: 1138 * 1139 * Our (SYN|ACK) response was rejected. 1140 * Check with syncache and remove entry to prevent 1141 * retransmits. 1142 * 1143 * NB: syncache_chkrst does its own logging of failure 1144 * causes. 1145 */ 1146 if (thflags & TH_RST) { 1147 syncache_chkrst(&inc, th); 1148 goto dropunlock; 1149 } 1150 /* 1151 * We can't do anything without SYN. 1152 */ 1153 if ((thflags & TH_SYN) == 0) { 1154 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1155 log(LOG_DEBUG, "%s; %s: Listen socket: " 1156 "SYN is missing, segment ignored\n", 1157 s, __func__); 1158 TCPSTAT_INC(tcps_badsyn); 1159 goto dropunlock; 1160 } 1161 /* 1162 * (SYN|ACK) is bogus on a listen socket. 1163 */ 1164 if (thflags & TH_ACK) { 1165 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1166 log(LOG_DEBUG, "%s; %s: Listen socket: " 1167 "SYN|ACK invalid, segment rejected\n", 1168 s, __func__); 1169 syncache_badack(&inc); /* XXX: Not needed! */ 1170 TCPSTAT_INC(tcps_badsyn); 1171 rstreason = BANDLIM_RST_OPENPORT; 1172 goto dropwithreset; 1173 } 1174 /* 1175 * If the drop_synfin option is enabled, drop all 1176 * segments with both the SYN and FIN bits set. 1177 * This prevents e.g. nmap from identifying the 1178 * TCP/IP stack. 1179 * XXX: Poor reasoning. nmap has other methods 1180 * and is constantly refining its stack detection 1181 * strategies. 1182 * XXX: This is a violation of the TCP specification 1183 * and was used by RFC1644. 1184 */ 1185 if ((thflags & TH_FIN) && V_drop_synfin) { 1186 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1187 log(LOG_DEBUG, "%s; %s: Listen socket: " 1188 "SYN|FIN segment ignored (based on " 1189 "sysctl setting)\n", s, __func__); 1190 TCPSTAT_INC(tcps_badsyn); 1191 goto dropunlock; 1192 } 1193 /* 1194 * Segment's flags are (SYN) or (SYN|FIN). 1195 * 1196 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1197 * as they do not affect the state of the TCP FSM. 1198 * The data pointed to by TH_URG and th_urp is ignored. 1199 */ 1200 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1201 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1202 KASSERT(thflags & (TH_SYN), 1203 ("%s: Listen socket: TH_SYN not set", __func__)); 1204 #ifdef INET6 1205 /* 1206 * If deprecated address is forbidden, 1207 * we do not accept SYN to deprecated interface 1208 * address to prevent any new inbound connection from 1209 * getting established. 1210 * When we do not accept SYN, we send a TCP RST, 1211 * with deprecated source address (instead of dropping 1212 * it). We compromise it as it is much better for peer 1213 * to send a RST, and RST will be the final packet 1214 * for the exchange. 1215 * 1216 * If we do not forbid deprecated addresses, we accept 1217 * the SYN packet. RFC2462 does not suggest dropping 1218 * SYN in this case. 1219 * If we decipher RFC2462 5.5.4, it says like this: 1220 * 1. use of deprecated addr with existing 1221 * communication is okay - "SHOULD continue to be 1222 * used" 1223 * 2. use of it with new communication: 1224 * (2a) "SHOULD NOT be used if alternate address 1225 * with sufficient scope is available" 1226 * (2b) nothing mentioned otherwise. 1227 * Here we fall into (2b) case as we have no choice in 1228 * our source address selection - we must obey the peer. 1229 * 1230 * The wording in RFC2462 is confusing, and there are 1231 * multiple description text for deprecated address 1232 * handling - worse, they are not exactly the same. 1233 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1234 */ 1235 if (isipv6 && !V_ip6_use_deprecated) { 1236 struct in6_ifaddr *ia6; 1237 1238 ia6 = ip6_getdstifaddr(m); 1239 if (ia6 != NULL && 1240 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1241 ifa_free(&ia6->ia_ifa); 1242 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1243 log(LOG_DEBUG, "%s; %s: Listen socket: " 1244 "Connection attempt to deprecated " 1245 "IPv6 address rejected\n", 1246 s, __func__); 1247 rstreason = BANDLIM_RST_OPENPORT; 1248 goto dropwithreset; 1249 } 1250 ifa_free(&ia6->ia_ifa); 1251 } 1252 #endif /* INET6 */ 1253 /* 1254 * Basic sanity checks on incoming SYN requests: 1255 * Don't respond if the destination is a link layer 1256 * broadcast according to RFC1122 4.2.3.10, p. 104. 1257 * If it is from this socket it must be forged. 1258 * Don't respond if the source or destination is a 1259 * global or subnet broad- or multicast address. 1260 * Note that it is quite possible to receive unicast 1261 * link-layer packets with a broadcast IP address. Use 1262 * in_broadcast() to find them. 1263 */ 1264 if (m->m_flags & (M_BCAST|M_MCAST)) { 1265 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1266 log(LOG_DEBUG, "%s; %s: Listen socket: " 1267 "Connection attempt from broad- or multicast " 1268 "link layer address ignored\n", s, __func__); 1269 goto dropunlock; 1270 } 1271 #ifdef INET6 1272 if (isipv6) { 1273 if (th->th_dport == th->th_sport && 1274 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1275 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1276 log(LOG_DEBUG, "%s; %s: Listen socket: " 1277 "Connection attempt to/from self " 1278 "ignored\n", s, __func__); 1279 goto dropunlock; 1280 } 1281 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1282 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1283 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1284 log(LOG_DEBUG, "%s; %s: Listen socket: " 1285 "Connection attempt from/to multicast " 1286 "address ignored\n", s, __func__); 1287 goto dropunlock; 1288 } 1289 } 1290 #endif 1291 #if defined(INET) && defined(INET6) 1292 else 1293 #endif 1294 #ifdef INET 1295 { 1296 if (th->th_dport == th->th_sport && 1297 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1299 log(LOG_DEBUG, "%s; %s: Listen socket: " 1300 "Connection attempt from/to self " 1301 "ignored\n", s, __func__); 1302 goto dropunlock; 1303 } 1304 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1305 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1306 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1307 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1309 log(LOG_DEBUG, "%s; %s: Listen socket: " 1310 "Connection attempt from/to broad- " 1311 "or multicast address ignored\n", 1312 s, __func__); 1313 goto dropunlock; 1314 } 1315 } 1316 #endif 1317 /* 1318 * SYN appears to be valid. Create compressed TCP state 1319 * for syncache. 1320 */ 1321 #ifdef TCPDEBUG 1322 if (so->so_options & SO_DEBUG) 1323 tcp_trace(TA_INPUT, ostate, tp, 1324 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1325 #endif 1326 tcp_dooptions(&to, optp, optlen, TO_SYN); 1327 syncache_add(&inc, &to, th, inp, &so, m); 1328 /* 1329 * Entry added to syncache and mbuf consumed. 1330 * Everything already unlocked by syncache_add(). 1331 */ 1332 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1333 return; 1334 } 1335 1336 #ifdef TCP_SIGNATURE 1337 if (sig_checked == 0) { 1338 tcp_dooptions(&to, optp, optlen, 1339 (thflags & TH_SYN) ? TO_SYN : 0); 1340 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1341 th, tp->t_flags)) { 1342 1343 /* 1344 * In SYN_SENT state if it receives an RST, it is 1345 * allowed for further processing. 1346 */ 1347 if ((thflags & TH_RST) == 0 || 1348 (tp->t_state == TCPS_SYN_SENT) == 0) 1349 goto dropunlock; 1350 } 1351 sig_checked = 1; 1352 } 1353 #endif 1354 1355 /* 1356 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1357 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1358 * the inpcb, and unlocks pcbinfo. 1359 */ 1360 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1361 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1362 return; 1363 1364 dropwithreset: 1365 if (ti_locked == TI_WLOCKED) { 1366 INP_INFO_WUNLOCK(&V_tcbinfo); 1367 ti_locked = TI_UNLOCKED; 1368 } 1369 #ifdef INVARIANTS 1370 else { 1371 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1372 "ti_locked: %d", __func__, ti_locked)); 1373 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1374 } 1375 #endif 1376 1377 if (inp != NULL) { 1378 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1379 INP_WUNLOCK(inp); 1380 } else 1381 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1382 m = NULL; /* mbuf chain got consumed. */ 1383 goto drop; 1384 1385 dropunlock: 1386 if (ti_locked == TI_WLOCKED) { 1387 INP_INFO_WUNLOCK(&V_tcbinfo); 1388 ti_locked = TI_UNLOCKED; 1389 } 1390 #ifdef INVARIANTS 1391 else { 1392 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1393 "ti_locked: %d", __func__, ti_locked)); 1394 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1395 } 1396 #endif 1397 1398 if (inp != NULL) 1399 INP_WUNLOCK(inp); 1400 1401 drop: 1402 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1403 if (s != NULL) 1404 free(s, M_TCPLOG); 1405 if (m != NULL) 1406 m_freem(m); 1407 } 1408 1409 static void 1410 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1411 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1412 int ti_locked) 1413 { 1414 int thflags, acked, ourfinisacked, needoutput = 0; 1415 int rstreason, todrop, win; 1416 u_long tiwin; 1417 struct tcpopt to; 1418 1419 #ifdef TCPDEBUG 1420 /* 1421 * The size of tcp_saveipgen must be the size of the max ip header, 1422 * now IPv6. 1423 */ 1424 u_char tcp_saveipgen[IP6_HDR_LEN]; 1425 struct tcphdr tcp_savetcp; 1426 short ostate = 0; 1427 #endif 1428 thflags = th->th_flags; 1429 tp->sackhint.last_sack_ack = 0; 1430 1431 /* 1432 * If this is either a state-changing packet or current state isn't 1433 * established, we require a write lock on tcbinfo. Otherwise, we 1434 * allow either a read lock or a write lock, as we may have acquired 1435 * a write lock due to a race. 1436 * 1437 * Require a global write lock for SYN/FIN/RST segments or 1438 * non-established connections; otherwise accept either a read or 1439 * write lock, as we may have conservatively acquired a write lock in 1440 * certain cases in tcp_input() (is this still true?). Currently we 1441 * will never enter with no lock, so we try to drop it quickly in the 1442 * common pure ack/pure data cases. 1443 */ 1444 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1445 tp->t_state != TCPS_ESTABLISHED) { 1446 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1447 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1448 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1449 } else { 1450 #ifdef INVARIANTS 1451 if (ti_locked == TI_WLOCKED) 1452 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1453 else { 1454 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1455 "ti_locked: %d", __func__, ti_locked)); 1456 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1457 } 1458 #endif 1459 } 1460 INP_WLOCK_ASSERT(tp->t_inpcb); 1461 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1462 __func__)); 1463 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1464 __func__)); 1465 1466 /* 1467 * Segment received on connection. 1468 * Reset idle time and keep-alive timer. 1469 * XXX: This should be done after segment 1470 * validation to ignore broken/spoofed segs. 1471 */ 1472 tp->t_rcvtime = ticks; 1473 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1474 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1475 1476 /* 1477 * Unscale the window into a 32-bit value. 1478 * For the SYN_SENT state the scale is zero. 1479 */ 1480 tiwin = th->th_win << tp->snd_scale; 1481 1482 /* 1483 * TCP ECN processing. 1484 */ 1485 if (tp->t_flags & TF_ECN_PERMIT) { 1486 if (thflags & TH_CWR) 1487 tp->t_flags &= ~TF_ECN_SND_ECE; 1488 switch (iptos & IPTOS_ECN_MASK) { 1489 case IPTOS_ECN_CE: 1490 tp->t_flags |= TF_ECN_SND_ECE; 1491 TCPSTAT_INC(tcps_ecn_ce); 1492 break; 1493 case IPTOS_ECN_ECT0: 1494 TCPSTAT_INC(tcps_ecn_ect0); 1495 break; 1496 case IPTOS_ECN_ECT1: 1497 TCPSTAT_INC(tcps_ecn_ect1); 1498 break; 1499 } 1500 /* Congestion experienced. */ 1501 if (thflags & TH_ECE) { 1502 cc_cong_signal(tp, th, CC_ECN); 1503 } 1504 } 1505 1506 /* 1507 * Parse options on any incoming segment. 1508 */ 1509 tcp_dooptions(&to, (u_char *)(th + 1), 1510 (th->th_off << 2) - sizeof(struct tcphdr), 1511 (thflags & TH_SYN) ? TO_SYN : 0); 1512 1513 /* 1514 * If echoed timestamp is later than the current time, 1515 * fall back to non RFC1323 RTT calculation. Normalize 1516 * timestamp if syncookies were used when this connection 1517 * was established. 1518 */ 1519 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1520 to.to_tsecr -= tp->ts_offset; 1521 if (TSTMP_GT(to.to_tsecr, ticks)) 1522 to.to_tsecr = 0; 1523 } 1524 1525 /* 1526 * Process options only when we get SYN/ACK back. The SYN case 1527 * for incoming connections is handled in tcp_syncache. 1528 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1529 * or <SYN,ACK>) segment itself is never scaled. 1530 * XXX this is traditional behavior, may need to be cleaned up. 1531 */ 1532 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1533 if ((to.to_flags & TOF_SCALE) && 1534 (tp->t_flags & TF_REQ_SCALE)) { 1535 tp->t_flags |= TF_RCVD_SCALE; 1536 tp->snd_scale = to.to_wscale; 1537 } 1538 /* 1539 * Initial send window. It will be updated with 1540 * the next incoming segment to the scaled value. 1541 */ 1542 tp->snd_wnd = th->th_win; 1543 if (to.to_flags & TOF_TS) { 1544 tp->t_flags |= TF_RCVD_TSTMP; 1545 tp->ts_recent = to.to_tsval; 1546 tp->ts_recent_age = ticks; 1547 } 1548 if (to.to_flags & TOF_MSS) 1549 tcp_mss(tp, to.to_mss); 1550 if ((tp->t_flags & TF_SACK_PERMIT) && 1551 (to.to_flags & TOF_SACKPERM) == 0) 1552 tp->t_flags &= ~TF_SACK_PERMIT; 1553 } 1554 1555 /* 1556 * Header prediction: check for the two common cases 1557 * of a uni-directional data xfer. If the packet has 1558 * no control flags, is in-sequence, the window didn't 1559 * change and we're not retransmitting, it's a 1560 * candidate. If the length is zero and the ack moved 1561 * forward, we're the sender side of the xfer. Just 1562 * free the data acked & wake any higher level process 1563 * that was blocked waiting for space. If the length 1564 * is non-zero and the ack didn't move, we're the 1565 * receiver side. If we're getting packets in-order 1566 * (the reassembly queue is empty), add the data to 1567 * the socket buffer and note that we need a delayed ack. 1568 * Make sure that the hidden state-flags are also off. 1569 * Since we check for TCPS_ESTABLISHED first, it can only 1570 * be TH_NEEDSYN. 1571 */ 1572 if (tp->t_state == TCPS_ESTABLISHED && 1573 th->th_seq == tp->rcv_nxt && 1574 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1575 tp->snd_nxt == tp->snd_max && 1576 tiwin && tiwin == tp->snd_wnd && 1577 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1578 LIST_EMPTY(&tp->t_segq) && 1579 ((to.to_flags & TOF_TS) == 0 || 1580 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1581 1582 /* 1583 * If last ACK falls within this segment's sequence numbers, 1584 * record the timestamp. 1585 * NOTE that the test is modified according to the latest 1586 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1587 */ 1588 if ((to.to_flags & TOF_TS) != 0 && 1589 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1590 tp->ts_recent_age = ticks; 1591 tp->ts_recent = to.to_tsval; 1592 } 1593 1594 if (tlen == 0) { 1595 if (SEQ_GT(th->th_ack, tp->snd_una) && 1596 SEQ_LEQ(th->th_ack, tp->snd_max) && 1597 !IN_RECOVERY(tp->t_flags) && 1598 (to.to_flags & TOF_SACK) == 0 && 1599 TAILQ_EMPTY(&tp->snd_holes)) { 1600 /* 1601 * This is a pure ack for outstanding data. 1602 */ 1603 if (ti_locked == TI_WLOCKED) 1604 INP_INFO_WUNLOCK(&V_tcbinfo); 1605 ti_locked = TI_UNLOCKED; 1606 1607 TCPSTAT_INC(tcps_predack); 1608 1609 /* 1610 * "bad retransmit" recovery. 1611 */ 1612 if (tp->t_rxtshift == 1 && 1613 tp->t_flags & TF_PREVVALID && 1614 (int)(ticks - tp->t_badrxtwin) < 0) { 1615 cc_cong_signal(tp, th, CC_RTO_ERR); 1616 } 1617 1618 /* 1619 * Recalculate the transmit timer / rtt. 1620 * 1621 * Some boxes send broken timestamp replies 1622 * during the SYN+ACK phase, ignore 1623 * timestamps of 0 or we could calculate a 1624 * huge RTT and blow up the retransmit timer. 1625 */ 1626 if ((to.to_flags & TOF_TS) != 0 && 1627 to.to_tsecr) { 1628 if (!tp->t_rttlow || 1629 tp->t_rttlow > ticks - to.to_tsecr) 1630 tp->t_rttlow = ticks - to.to_tsecr; 1631 tcp_xmit_timer(tp, 1632 ticks - to.to_tsecr + 1); 1633 } else if (tp->t_rtttime && 1634 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1635 if (!tp->t_rttlow || 1636 tp->t_rttlow > ticks - tp->t_rtttime) 1637 tp->t_rttlow = ticks - tp->t_rtttime; 1638 tcp_xmit_timer(tp, 1639 ticks - tp->t_rtttime); 1640 } 1641 acked = BYTES_THIS_ACK(tp, th); 1642 1643 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1644 hhook_run_tcp_est_in(tp, th, &to); 1645 1646 TCPSTAT_INC(tcps_rcvackpack); 1647 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1648 sbdrop(&so->so_snd, acked); 1649 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1650 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1651 tp->snd_recover = th->th_ack - 1; 1652 1653 /* 1654 * Let the congestion control algorithm update 1655 * congestion control related information. This 1656 * typically means increasing the congestion 1657 * window. 1658 */ 1659 cc_ack_received(tp, th, CC_ACK); 1660 1661 tp->snd_una = th->th_ack; 1662 /* 1663 * Pull snd_wl2 up to prevent seq wrap relative 1664 * to th_ack. 1665 */ 1666 tp->snd_wl2 = th->th_ack; 1667 tp->t_dupacks = 0; 1668 m_freem(m); 1669 ND6_HINT(tp); /* Some progress has been made. */ 1670 1671 /* 1672 * If all outstanding data are acked, stop 1673 * retransmit timer, otherwise restart timer 1674 * using current (possibly backed-off) value. 1675 * If process is waiting for space, 1676 * wakeup/selwakeup/signal. If data 1677 * are ready to send, let tcp_output 1678 * decide between more output or persist. 1679 */ 1680 #ifdef TCPDEBUG 1681 if (so->so_options & SO_DEBUG) 1682 tcp_trace(TA_INPUT, ostate, tp, 1683 (void *)tcp_saveipgen, 1684 &tcp_savetcp, 0); 1685 #endif 1686 if (tp->snd_una == tp->snd_max) 1687 tcp_timer_activate(tp, TT_REXMT, 0); 1688 else if (!tcp_timer_active(tp, TT_PERSIST)) 1689 tcp_timer_activate(tp, TT_REXMT, 1690 tp->t_rxtcur); 1691 sowwakeup(so); 1692 if (so->so_snd.sb_cc) 1693 (void) tcp_output(tp); 1694 goto check_delack; 1695 } 1696 } else if (th->th_ack == tp->snd_una && 1697 tlen <= sbspace(&so->so_rcv)) { 1698 int newsize = 0; /* automatic sockbuf scaling */ 1699 1700 /* 1701 * This is a pure, in-sequence data packet with 1702 * nothing on the reassembly queue and we have enough 1703 * buffer space to take it. 1704 */ 1705 if (ti_locked == TI_WLOCKED) 1706 INP_INFO_WUNLOCK(&V_tcbinfo); 1707 ti_locked = TI_UNLOCKED; 1708 1709 /* Clean receiver SACK report if present */ 1710 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1711 tcp_clean_sackreport(tp); 1712 TCPSTAT_INC(tcps_preddat); 1713 tp->rcv_nxt += tlen; 1714 /* 1715 * Pull snd_wl1 up to prevent seq wrap relative to 1716 * th_seq. 1717 */ 1718 tp->snd_wl1 = th->th_seq; 1719 /* 1720 * Pull rcv_up up to prevent seq wrap relative to 1721 * rcv_nxt. 1722 */ 1723 tp->rcv_up = tp->rcv_nxt; 1724 TCPSTAT_INC(tcps_rcvpack); 1725 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1726 ND6_HINT(tp); /* Some progress has been made */ 1727 #ifdef TCPDEBUG 1728 if (so->so_options & SO_DEBUG) 1729 tcp_trace(TA_INPUT, ostate, tp, 1730 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1731 #endif 1732 /* 1733 * Automatic sizing of receive socket buffer. Often the send 1734 * buffer size is not optimally adjusted to the actual network 1735 * conditions at hand (delay bandwidth product). Setting the 1736 * buffer size too small limits throughput on links with high 1737 * bandwidth and high delay (eg. trans-continental/oceanic links). 1738 * 1739 * On the receive side the socket buffer memory is only rarely 1740 * used to any significant extent. This allows us to be much 1741 * more aggressive in scaling the receive socket buffer. For 1742 * the case that the buffer space is actually used to a large 1743 * extent and we run out of kernel memory we can simply drop 1744 * the new segments; TCP on the sender will just retransmit it 1745 * later. Setting the buffer size too big may only consume too 1746 * much kernel memory if the application doesn't read() from 1747 * the socket or packet loss or reordering makes use of the 1748 * reassembly queue. 1749 * 1750 * The criteria to step up the receive buffer one notch are: 1751 * 1. the number of bytes received during the time it takes 1752 * one timestamp to be reflected back to us (the RTT); 1753 * 2. received bytes per RTT is within seven eighth of the 1754 * current socket buffer size; 1755 * 3. receive buffer size has not hit maximal automatic size; 1756 * 1757 * This algorithm does one step per RTT at most and only if 1758 * we receive a bulk stream w/o packet losses or reorderings. 1759 * Shrinking the buffer during idle times is not necessary as 1760 * it doesn't consume any memory when idle. 1761 * 1762 * TODO: Only step up if the application is actually serving 1763 * the buffer to better manage the socket buffer resources. 1764 */ 1765 if (V_tcp_do_autorcvbuf && 1766 to.to_tsecr && 1767 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1768 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1769 to.to_tsecr - tp->rfbuf_ts < hz) { 1770 if (tp->rfbuf_cnt > 1771 (so->so_rcv.sb_hiwat / 8 * 7) && 1772 so->so_rcv.sb_hiwat < 1773 V_tcp_autorcvbuf_max) { 1774 newsize = 1775 min(so->so_rcv.sb_hiwat + 1776 V_tcp_autorcvbuf_inc, 1777 V_tcp_autorcvbuf_max); 1778 } 1779 /* Start over with next RTT. */ 1780 tp->rfbuf_ts = 0; 1781 tp->rfbuf_cnt = 0; 1782 } else 1783 tp->rfbuf_cnt += tlen; /* add up */ 1784 } 1785 1786 /* Add data to socket buffer. */ 1787 SOCKBUF_LOCK(&so->so_rcv); 1788 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1789 m_freem(m); 1790 } else { 1791 /* 1792 * Set new socket buffer size. 1793 * Give up when limit is reached. 1794 */ 1795 if (newsize) 1796 if (!sbreserve_locked(&so->so_rcv, 1797 newsize, so, NULL)) 1798 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1799 m_adj(m, drop_hdrlen); /* delayed header drop */ 1800 sbappendstream_locked(&so->so_rcv, m); 1801 } 1802 /* NB: sorwakeup_locked() does an implicit unlock. */ 1803 sorwakeup_locked(so); 1804 if (DELAY_ACK(tp)) { 1805 tp->t_flags |= TF_DELACK; 1806 } else { 1807 tp->t_flags |= TF_ACKNOW; 1808 tcp_output(tp); 1809 } 1810 goto check_delack; 1811 } 1812 } 1813 1814 /* 1815 * Calculate amount of space in receive window, 1816 * and then do TCP input processing. 1817 * Receive window is amount of space in rcv queue, 1818 * but not less than advertised window. 1819 */ 1820 win = sbspace(&so->so_rcv); 1821 if (win < 0) 1822 win = 0; 1823 KASSERT(SEQ_GEQ(tp->rcv_adv, tp->rcv_nxt), 1824 ("tcp_input negative window: tp %p rcv_nxt %u rcv_adv %u", tp, 1825 tp->rcv_nxt, tp->rcv_adv)); 1826 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1827 1828 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1829 tp->rfbuf_ts = 0; 1830 tp->rfbuf_cnt = 0; 1831 1832 switch (tp->t_state) { 1833 1834 /* 1835 * If the state is SYN_RECEIVED: 1836 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1837 */ 1838 case TCPS_SYN_RECEIVED: 1839 if ((thflags & TH_ACK) && 1840 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1841 SEQ_GT(th->th_ack, tp->snd_max))) { 1842 rstreason = BANDLIM_RST_OPENPORT; 1843 goto dropwithreset; 1844 } 1845 break; 1846 1847 /* 1848 * If the state is SYN_SENT: 1849 * if seg contains an ACK, but not for our SYN, drop the input. 1850 * if seg contains a RST, then drop the connection. 1851 * if seg does not contain SYN, then drop it. 1852 * Otherwise this is an acceptable SYN segment 1853 * initialize tp->rcv_nxt and tp->irs 1854 * if seg contains ack then advance tp->snd_una 1855 * if seg contains an ECE and ECN support is enabled, the stream 1856 * is ECN capable. 1857 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1858 * arrange for segment to be acked (eventually) 1859 * continue processing rest of data/controls, beginning with URG 1860 */ 1861 case TCPS_SYN_SENT: 1862 if ((thflags & TH_ACK) && 1863 (SEQ_LEQ(th->th_ack, tp->iss) || 1864 SEQ_GT(th->th_ack, tp->snd_max))) { 1865 rstreason = BANDLIM_UNLIMITED; 1866 goto dropwithreset; 1867 } 1868 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1869 tp = tcp_drop(tp, ECONNREFUSED); 1870 if (thflags & TH_RST) 1871 goto drop; 1872 if (!(thflags & TH_SYN)) 1873 goto drop; 1874 1875 tp->irs = th->th_seq; 1876 tcp_rcvseqinit(tp); 1877 if (thflags & TH_ACK) { 1878 TCPSTAT_INC(tcps_connects); 1879 soisconnected(so); 1880 #ifdef MAC 1881 mac_socketpeer_set_from_mbuf(m, so); 1882 #endif 1883 /* Do window scaling on this connection? */ 1884 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1885 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1886 tp->rcv_scale = tp->request_r_scale; 1887 } 1888 tp->rcv_adv += imin(tp->rcv_wnd, 1889 TCP_MAXWIN << tp->rcv_scale); 1890 tp->snd_una++; /* SYN is acked */ 1891 /* 1892 * If there's data, delay ACK; if there's also a FIN 1893 * ACKNOW will be turned on later. 1894 */ 1895 if (DELAY_ACK(tp) && tlen != 0) 1896 tcp_timer_activate(tp, TT_DELACK, 1897 tcp_delacktime); 1898 else 1899 tp->t_flags |= TF_ACKNOW; 1900 1901 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1902 tp->t_flags |= TF_ECN_PERMIT; 1903 TCPSTAT_INC(tcps_ecn_shs); 1904 } 1905 1906 /* 1907 * Received <SYN,ACK> in SYN_SENT[*] state. 1908 * Transitions: 1909 * SYN_SENT --> ESTABLISHED 1910 * SYN_SENT* --> FIN_WAIT_1 1911 */ 1912 tp->t_starttime = ticks; 1913 if (tp->t_flags & TF_NEEDFIN) { 1914 tp->t_state = TCPS_FIN_WAIT_1; 1915 tp->t_flags &= ~TF_NEEDFIN; 1916 thflags &= ~TH_SYN; 1917 } else { 1918 tp->t_state = TCPS_ESTABLISHED; 1919 cc_conn_init(tp); 1920 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1921 } 1922 } else { 1923 /* 1924 * Received initial SYN in SYN-SENT[*] state => 1925 * simultaneous open. If segment contains CC option 1926 * and there is a cached CC, apply TAO test. 1927 * If it succeeds, connection is * half-synchronized. 1928 * Otherwise, do 3-way handshake: 1929 * SYN-SENT -> SYN-RECEIVED 1930 * SYN-SENT* -> SYN-RECEIVED* 1931 * If there was no CC option, clear cached CC value. 1932 */ 1933 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1934 tcp_timer_activate(tp, TT_REXMT, 0); 1935 tp->t_state = TCPS_SYN_RECEIVED; 1936 } 1937 1938 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1939 "ti_locked %d", __func__, ti_locked)); 1940 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1941 INP_WLOCK_ASSERT(tp->t_inpcb); 1942 1943 /* 1944 * Advance th->th_seq to correspond to first data byte. 1945 * If data, trim to stay within window, 1946 * dropping FIN if necessary. 1947 */ 1948 th->th_seq++; 1949 if (tlen > tp->rcv_wnd) { 1950 todrop = tlen - tp->rcv_wnd; 1951 m_adj(m, -todrop); 1952 tlen = tp->rcv_wnd; 1953 thflags &= ~TH_FIN; 1954 TCPSTAT_INC(tcps_rcvpackafterwin); 1955 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1956 } 1957 tp->snd_wl1 = th->th_seq - 1; 1958 tp->rcv_up = th->th_seq; 1959 /* 1960 * Client side of transaction: already sent SYN and data. 1961 * If the remote host used T/TCP to validate the SYN, 1962 * our data will be ACK'd; if so, enter normal data segment 1963 * processing in the middle of step 5, ack processing. 1964 * Otherwise, goto step 6. 1965 */ 1966 if (thflags & TH_ACK) 1967 goto process_ACK; 1968 1969 goto step6; 1970 1971 /* 1972 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1973 * do normal processing. 1974 * 1975 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1976 */ 1977 case TCPS_LAST_ACK: 1978 case TCPS_CLOSING: 1979 break; /* continue normal processing */ 1980 } 1981 1982 /* 1983 * States other than LISTEN or SYN_SENT. 1984 * First check the RST flag and sequence number since reset segments 1985 * are exempt from the timestamp and connection count tests. This 1986 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1987 * below which allowed reset segments in half the sequence space 1988 * to fall though and be processed (which gives forged reset 1989 * segments with a random sequence number a 50 percent chance of 1990 * killing a connection). 1991 * Then check timestamp, if present. 1992 * Then check the connection count, if present. 1993 * Then check that at least some bytes of segment are within 1994 * receive window. If segment begins before rcv_nxt, 1995 * drop leading data (and SYN); if nothing left, just ack. 1996 * 1997 * 1998 * If the RST bit is set, check the sequence number to see 1999 * if this is a valid reset segment. 2000 * RFC 793 page 37: 2001 * In all states except SYN-SENT, all reset (RST) segments 2002 * are validated by checking their SEQ-fields. A reset is 2003 * valid if its sequence number is in the window. 2004 * Note: this does not take into account delayed ACKs, so 2005 * we should test against last_ack_sent instead of rcv_nxt. 2006 * The sequence number in the reset segment is normally an 2007 * echo of our outgoing acknowlegement numbers, but some hosts 2008 * send a reset with the sequence number at the rightmost edge 2009 * of our receive window, and we have to handle this case. 2010 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2011 * that brute force RST attacks are possible. To combat this, 2012 * we use a much stricter check while in the ESTABLISHED state, 2013 * only accepting RSTs where the sequence number is equal to 2014 * last_ack_sent. In all other states (the states in which a 2015 * RST is more likely), the more permissive check is used. 2016 * If we have multiple segments in flight, the initial reset 2017 * segment sequence numbers will be to the left of last_ack_sent, 2018 * but they will eventually catch up. 2019 * In any case, it never made sense to trim reset segments to 2020 * fit the receive window since RFC 1122 says: 2021 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2022 * 2023 * A TCP SHOULD allow a received RST segment to include data. 2024 * 2025 * DISCUSSION 2026 * It has been suggested that a RST segment could contain 2027 * ASCII text that encoded and explained the cause of the 2028 * RST. No standard has yet been established for such 2029 * data. 2030 * 2031 * If the reset segment passes the sequence number test examine 2032 * the state: 2033 * SYN_RECEIVED STATE: 2034 * If passive open, return to LISTEN state. 2035 * If active open, inform user that connection was refused. 2036 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2037 * Inform user that connection was reset, and close tcb. 2038 * CLOSING, LAST_ACK STATES: 2039 * Close the tcb. 2040 * TIME_WAIT STATE: 2041 * Drop the segment - see Stevens, vol. 2, p. 964 and 2042 * RFC 1337. 2043 */ 2044 if (thflags & TH_RST) { 2045 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2046 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2047 switch (tp->t_state) { 2048 2049 case TCPS_SYN_RECEIVED: 2050 so->so_error = ECONNREFUSED; 2051 goto close; 2052 2053 case TCPS_ESTABLISHED: 2054 if (V_tcp_insecure_rst == 0 && 2055 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2056 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2057 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2058 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2059 TCPSTAT_INC(tcps_badrst); 2060 goto drop; 2061 } 2062 /* FALLTHROUGH */ 2063 case TCPS_FIN_WAIT_1: 2064 case TCPS_FIN_WAIT_2: 2065 case TCPS_CLOSE_WAIT: 2066 so->so_error = ECONNRESET; 2067 close: 2068 KASSERT(ti_locked == TI_WLOCKED, 2069 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2070 ti_locked)); 2071 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2072 2073 tp->t_state = TCPS_CLOSED; 2074 TCPSTAT_INC(tcps_drops); 2075 tp = tcp_close(tp); 2076 break; 2077 2078 case TCPS_CLOSING: 2079 case TCPS_LAST_ACK: 2080 KASSERT(ti_locked == TI_WLOCKED, 2081 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2082 ti_locked)); 2083 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2084 2085 tp = tcp_close(tp); 2086 break; 2087 } 2088 } 2089 goto drop; 2090 } 2091 2092 /* 2093 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2094 * and it's less than ts_recent, drop it. 2095 */ 2096 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2097 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2098 2099 /* Check to see if ts_recent is over 24 days old. */ 2100 if (ticks - tp->ts_recent_age > TCP_PAWS_IDLE) { 2101 /* 2102 * Invalidate ts_recent. If this segment updates 2103 * ts_recent, the age will be reset later and ts_recent 2104 * will get a valid value. If it does not, setting 2105 * ts_recent to zero will at least satisfy the 2106 * requirement that zero be placed in the timestamp 2107 * echo reply when ts_recent isn't valid. The 2108 * age isn't reset until we get a valid ts_recent 2109 * because we don't want out-of-order segments to be 2110 * dropped when ts_recent is old. 2111 */ 2112 tp->ts_recent = 0; 2113 } else { 2114 TCPSTAT_INC(tcps_rcvduppack); 2115 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2116 TCPSTAT_INC(tcps_pawsdrop); 2117 if (tlen) 2118 goto dropafterack; 2119 goto drop; 2120 } 2121 } 2122 2123 /* 2124 * In the SYN-RECEIVED state, validate that the packet belongs to 2125 * this connection before trimming the data to fit the receive 2126 * window. Check the sequence number versus IRS since we know 2127 * the sequence numbers haven't wrapped. This is a partial fix 2128 * for the "LAND" DoS attack. 2129 */ 2130 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2131 rstreason = BANDLIM_RST_OPENPORT; 2132 goto dropwithreset; 2133 } 2134 2135 todrop = tp->rcv_nxt - th->th_seq; 2136 if (todrop > 0) { 2137 /* 2138 * If this is a duplicate SYN for our current connection, 2139 * advance over it and pretend and it's not a SYN. 2140 */ 2141 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2142 thflags &= ~TH_SYN; 2143 th->th_seq++; 2144 if (th->th_urp > 1) 2145 th->th_urp--; 2146 else 2147 thflags &= ~TH_URG; 2148 todrop--; 2149 } 2150 /* 2151 * Following if statement from Stevens, vol. 2, p. 960. 2152 */ 2153 if (todrop > tlen 2154 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2155 /* 2156 * Any valid FIN must be to the left of the window. 2157 * At this point the FIN must be a duplicate or out 2158 * of sequence; drop it. 2159 */ 2160 thflags &= ~TH_FIN; 2161 2162 /* 2163 * Send an ACK to resynchronize and drop any data. 2164 * But keep on processing for RST or ACK. 2165 */ 2166 tp->t_flags |= TF_ACKNOW; 2167 todrop = tlen; 2168 TCPSTAT_INC(tcps_rcvduppack); 2169 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2170 } else { 2171 TCPSTAT_INC(tcps_rcvpartduppack); 2172 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2173 } 2174 drop_hdrlen += todrop; /* drop from the top afterwards */ 2175 th->th_seq += todrop; 2176 tlen -= todrop; 2177 if (th->th_urp > todrop) 2178 th->th_urp -= todrop; 2179 else { 2180 thflags &= ~TH_URG; 2181 th->th_urp = 0; 2182 } 2183 } 2184 2185 /* 2186 * If new data are received on a connection after the 2187 * user processes are gone, then RST the other end. 2188 */ 2189 if ((so->so_state & SS_NOFDREF) && 2190 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2191 char *s; 2192 2193 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2194 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2195 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2196 2197 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 2198 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 2199 "was closed, sending RST and removing tcpcb\n", 2200 s, __func__, tcpstates[tp->t_state], tlen); 2201 free(s, M_TCPLOG); 2202 } 2203 tp = tcp_close(tp); 2204 TCPSTAT_INC(tcps_rcvafterclose); 2205 rstreason = BANDLIM_UNLIMITED; 2206 goto dropwithreset; 2207 } 2208 2209 /* 2210 * If segment ends after window, drop trailing data 2211 * (and PUSH and FIN); if nothing left, just ACK. 2212 */ 2213 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2214 if (todrop > 0) { 2215 TCPSTAT_INC(tcps_rcvpackafterwin); 2216 if (todrop >= tlen) { 2217 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2218 /* 2219 * If window is closed can only take segments at 2220 * window edge, and have to drop data and PUSH from 2221 * incoming segments. Continue processing, but 2222 * remember to ack. Otherwise, drop segment 2223 * and ack. 2224 */ 2225 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2226 tp->t_flags |= TF_ACKNOW; 2227 TCPSTAT_INC(tcps_rcvwinprobe); 2228 } else 2229 goto dropafterack; 2230 } else 2231 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2232 m_adj(m, -todrop); 2233 tlen -= todrop; 2234 thflags &= ~(TH_PUSH|TH_FIN); 2235 } 2236 2237 /* 2238 * If last ACK falls within this segment's sequence numbers, 2239 * record its timestamp. 2240 * NOTE: 2241 * 1) That the test incorporates suggestions from the latest 2242 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2243 * 2) That updating only on newer timestamps interferes with 2244 * our earlier PAWS tests, so this check should be solely 2245 * predicated on the sequence space of this segment. 2246 * 3) That we modify the segment boundary check to be 2247 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2248 * instead of RFC1323's 2249 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2250 * This modified check allows us to overcome RFC1323's 2251 * limitations as described in Stevens TCP/IP Illustrated 2252 * Vol. 2 p.869. In such cases, we can still calculate the 2253 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2254 */ 2255 if ((to.to_flags & TOF_TS) != 0 && 2256 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2257 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2258 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2259 tp->ts_recent_age = ticks; 2260 tp->ts_recent = to.to_tsval; 2261 } 2262 2263 /* 2264 * If a SYN is in the window, then this is an 2265 * error and we send an RST and drop the connection. 2266 */ 2267 if (thflags & TH_SYN) { 2268 KASSERT(ti_locked == TI_WLOCKED, 2269 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2270 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2271 2272 tp = tcp_drop(tp, ECONNRESET); 2273 rstreason = BANDLIM_UNLIMITED; 2274 goto drop; 2275 } 2276 2277 /* 2278 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2279 * flag is on (half-synchronized state), then queue data for 2280 * later processing; else drop segment and return. 2281 */ 2282 if ((thflags & TH_ACK) == 0) { 2283 if (tp->t_state == TCPS_SYN_RECEIVED || 2284 (tp->t_flags & TF_NEEDSYN)) 2285 goto step6; 2286 else if (tp->t_flags & TF_ACKNOW) 2287 goto dropafterack; 2288 else 2289 goto drop; 2290 } 2291 2292 /* 2293 * Ack processing. 2294 */ 2295 switch (tp->t_state) { 2296 2297 /* 2298 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2299 * ESTABLISHED state and continue processing. 2300 * The ACK was checked above. 2301 */ 2302 case TCPS_SYN_RECEIVED: 2303 2304 TCPSTAT_INC(tcps_connects); 2305 soisconnected(so); 2306 /* Do window scaling? */ 2307 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2308 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2309 tp->rcv_scale = tp->request_r_scale; 2310 tp->snd_wnd = tiwin; 2311 } 2312 /* 2313 * Make transitions: 2314 * SYN-RECEIVED -> ESTABLISHED 2315 * SYN-RECEIVED* -> FIN-WAIT-1 2316 */ 2317 tp->t_starttime = ticks; 2318 if (tp->t_flags & TF_NEEDFIN) { 2319 tp->t_state = TCPS_FIN_WAIT_1; 2320 tp->t_flags &= ~TF_NEEDFIN; 2321 } else { 2322 tp->t_state = TCPS_ESTABLISHED; 2323 cc_conn_init(tp); 2324 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 2325 } 2326 /* 2327 * If segment contains data or ACK, will call tcp_reass() 2328 * later; if not, do so now to pass queued data to user. 2329 */ 2330 if (tlen == 0 && (thflags & TH_FIN) == 0) 2331 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2332 (struct mbuf *)0); 2333 tp->snd_wl1 = th->th_seq - 1; 2334 /* FALLTHROUGH */ 2335 2336 /* 2337 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2338 * ACKs. If the ack is in the range 2339 * tp->snd_una < th->th_ack <= tp->snd_max 2340 * then advance tp->snd_una to th->th_ack and drop 2341 * data from the retransmission queue. If this ACK reflects 2342 * more up to date window information we update our window information. 2343 */ 2344 case TCPS_ESTABLISHED: 2345 case TCPS_FIN_WAIT_1: 2346 case TCPS_FIN_WAIT_2: 2347 case TCPS_CLOSE_WAIT: 2348 case TCPS_CLOSING: 2349 case TCPS_LAST_ACK: 2350 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2351 TCPSTAT_INC(tcps_rcvacktoomuch); 2352 goto dropafterack; 2353 } 2354 if ((tp->t_flags & TF_SACK_PERMIT) && 2355 ((to.to_flags & TOF_SACK) || 2356 !TAILQ_EMPTY(&tp->snd_holes))) 2357 tcp_sack_doack(tp, &to, th->th_ack); 2358 2359 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2360 hhook_run_tcp_est_in(tp, th, &to); 2361 2362 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2363 if (tlen == 0 && tiwin == tp->snd_wnd) { 2364 TCPSTAT_INC(tcps_rcvdupack); 2365 /* 2366 * If we have outstanding data (other than 2367 * a window probe), this is a completely 2368 * duplicate ack (ie, window info didn't 2369 * change), the ack is the biggest we've 2370 * seen and we've seen exactly our rexmt 2371 * threshhold of them, assume a packet 2372 * has been dropped and retransmit it. 2373 * Kludge snd_nxt & the congestion 2374 * window so we send only this one 2375 * packet. 2376 * 2377 * We know we're losing at the current 2378 * window size so do congestion avoidance 2379 * (set ssthresh to half the current window 2380 * and pull our congestion window back to 2381 * the new ssthresh). 2382 * 2383 * Dup acks mean that packets have left the 2384 * network (they're now cached at the receiver) 2385 * so bump cwnd by the amount in the receiver 2386 * to keep a constant cwnd packets in the 2387 * network. 2388 * 2389 * When using TCP ECN, notify the peer that 2390 * we reduced the cwnd. 2391 */ 2392 if (!tcp_timer_active(tp, TT_REXMT) || 2393 th->th_ack != tp->snd_una) 2394 tp->t_dupacks = 0; 2395 else if (++tp->t_dupacks > tcprexmtthresh || 2396 IN_FASTRECOVERY(tp->t_flags)) { 2397 cc_ack_received(tp, th, CC_DUPACK); 2398 if ((tp->t_flags & TF_SACK_PERMIT) && 2399 IN_FASTRECOVERY(tp->t_flags)) { 2400 int awnd; 2401 2402 /* 2403 * Compute the amount of data in flight first. 2404 * We can inject new data into the pipe iff 2405 * we have less than 1/2 the original window's 2406 * worth of data in flight. 2407 */ 2408 awnd = (tp->snd_nxt - tp->snd_fack) + 2409 tp->sackhint.sack_bytes_rexmit; 2410 if (awnd < tp->snd_ssthresh) { 2411 tp->snd_cwnd += tp->t_maxseg; 2412 if (tp->snd_cwnd > tp->snd_ssthresh) 2413 tp->snd_cwnd = tp->snd_ssthresh; 2414 } 2415 } else 2416 tp->snd_cwnd += tp->t_maxseg; 2417 (void) tcp_output(tp); 2418 goto drop; 2419 } else if (tp->t_dupacks == tcprexmtthresh) { 2420 tcp_seq onxt = tp->snd_nxt; 2421 2422 /* 2423 * If we're doing sack, check to 2424 * see if we're already in sack 2425 * recovery. If we're not doing sack, 2426 * check to see if we're in newreno 2427 * recovery. 2428 */ 2429 if (tp->t_flags & TF_SACK_PERMIT) { 2430 if (IN_FASTRECOVERY(tp->t_flags)) { 2431 tp->t_dupacks = 0; 2432 break; 2433 } 2434 } else { 2435 if (SEQ_LEQ(th->th_ack, 2436 tp->snd_recover)) { 2437 tp->t_dupacks = 0; 2438 break; 2439 } 2440 } 2441 /* Congestion signal before ack. */ 2442 cc_cong_signal(tp, th, CC_NDUPACK); 2443 cc_ack_received(tp, th, CC_DUPACK); 2444 tcp_timer_activate(tp, TT_REXMT, 0); 2445 tp->t_rtttime = 0; 2446 if (tp->t_flags & TF_SACK_PERMIT) { 2447 TCPSTAT_INC( 2448 tcps_sack_recovery_episode); 2449 tp->sack_newdata = tp->snd_nxt; 2450 tp->snd_cwnd = tp->t_maxseg; 2451 (void) tcp_output(tp); 2452 goto drop; 2453 } 2454 tp->snd_nxt = th->th_ack; 2455 tp->snd_cwnd = tp->t_maxseg; 2456 (void) tcp_output(tp); 2457 KASSERT(tp->snd_limited <= 2, 2458 ("%s: tp->snd_limited too big", 2459 __func__)); 2460 tp->snd_cwnd = tp->snd_ssthresh + 2461 tp->t_maxseg * 2462 (tp->t_dupacks - tp->snd_limited); 2463 if (SEQ_GT(onxt, tp->snd_nxt)) 2464 tp->snd_nxt = onxt; 2465 goto drop; 2466 } else if (V_tcp_do_rfc3042) { 2467 cc_ack_received(tp, th, CC_DUPACK); 2468 u_long oldcwnd = tp->snd_cwnd; 2469 tcp_seq oldsndmax = tp->snd_max; 2470 u_int sent; 2471 2472 KASSERT(tp->t_dupacks == 1 || 2473 tp->t_dupacks == 2, 2474 ("%s: dupacks not 1 or 2", 2475 __func__)); 2476 if (tp->t_dupacks == 1) 2477 tp->snd_limited = 0; 2478 tp->snd_cwnd = 2479 (tp->snd_nxt - tp->snd_una) + 2480 (tp->t_dupacks - tp->snd_limited) * 2481 tp->t_maxseg; 2482 (void) tcp_output(tp); 2483 sent = tp->snd_max - oldsndmax; 2484 if (sent > tp->t_maxseg) { 2485 KASSERT((tp->t_dupacks == 2 && 2486 tp->snd_limited == 0) || 2487 (sent == tp->t_maxseg + 1 && 2488 tp->t_flags & TF_SENTFIN), 2489 ("%s: sent too much", 2490 __func__)); 2491 tp->snd_limited = 2; 2492 } else if (sent > 0) 2493 ++tp->snd_limited; 2494 tp->snd_cwnd = oldcwnd; 2495 goto drop; 2496 } 2497 } else 2498 tp->t_dupacks = 0; 2499 break; 2500 } 2501 2502 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2503 ("%s: th_ack <= snd_una", __func__)); 2504 2505 /* 2506 * If the congestion window was inflated to account 2507 * for the other side's cached packets, retract it. 2508 */ 2509 if (IN_FASTRECOVERY(tp->t_flags)) { 2510 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2511 if (tp->t_flags & TF_SACK_PERMIT) 2512 tcp_sack_partialack(tp, th); 2513 else 2514 tcp_newreno_partial_ack(tp, th); 2515 } else 2516 cc_post_recovery(tp, th); 2517 } 2518 tp->t_dupacks = 0; 2519 /* 2520 * If we reach this point, ACK is not a duplicate, 2521 * i.e., it ACKs something we sent. 2522 */ 2523 if (tp->t_flags & TF_NEEDSYN) { 2524 /* 2525 * T/TCP: Connection was half-synchronized, and our 2526 * SYN has been ACK'd (so connection is now fully 2527 * synchronized). Go to non-starred state, 2528 * increment snd_una for ACK of SYN, and check if 2529 * we can do window scaling. 2530 */ 2531 tp->t_flags &= ~TF_NEEDSYN; 2532 tp->snd_una++; 2533 /* Do window scaling? */ 2534 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2535 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2536 tp->rcv_scale = tp->request_r_scale; 2537 /* Send window already scaled. */ 2538 } 2539 } 2540 2541 process_ACK: 2542 INP_WLOCK_ASSERT(tp->t_inpcb); 2543 2544 acked = BYTES_THIS_ACK(tp, th); 2545 TCPSTAT_INC(tcps_rcvackpack); 2546 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2547 2548 /* 2549 * If we just performed our first retransmit, and the ACK 2550 * arrives within our recovery window, then it was a mistake 2551 * to do the retransmit in the first place. Recover our 2552 * original cwnd and ssthresh, and proceed to transmit where 2553 * we left off. 2554 */ 2555 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2556 (int)(ticks - tp->t_badrxtwin) < 0) 2557 cc_cong_signal(tp, th, CC_RTO_ERR); 2558 2559 /* 2560 * If we have a timestamp reply, update smoothed 2561 * round trip time. If no timestamp is present but 2562 * transmit timer is running and timed sequence 2563 * number was acked, update smoothed round trip time. 2564 * Since we now have an rtt measurement, cancel the 2565 * timer backoff (cf., Phil Karn's retransmit alg.). 2566 * Recompute the initial retransmit timer. 2567 * 2568 * Some boxes send broken timestamp replies 2569 * during the SYN+ACK phase, ignore 2570 * timestamps of 0 or we could calculate a 2571 * huge RTT and blow up the retransmit timer. 2572 */ 2573 if ((to.to_flags & TOF_TS) != 0 && 2574 to.to_tsecr) { 2575 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2576 tp->t_rttlow = ticks - to.to_tsecr; 2577 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2578 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2579 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2580 tp->t_rttlow = ticks - tp->t_rtttime; 2581 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2582 } 2583 2584 /* 2585 * If all outstanding data is acked, stop retransmit 2586 * timer and remember to restart (more output or persist). 2587 * If there is more data to be acked, restart retransmit 2588 * timer, using current (possibly backed-off) value. 2589 */ 2590 if (th->th_ack == tp->snd_max) { 2591 tcp_timer_activate(tp, TT_REXMT, 0); 2592 needoutput = 1; 2593 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2594 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2595 2596 /* 2597 * If no data (only SYN) was ACK'd, 2598 * skip rest of ACK processing. 2599 */ 2600 if (acked == 0) 2601 goto step6; 2602 2603 /* 2604 * Let the congestion control algorithm update congestion 2605 * control related information. This typically means increasing 2606 * the congestion window. 2607 */ 2608 cc_ack_received(tp, th, CC_ACK); 2609 2610 SOCKBUF_LOCK(&so->so_snd); 2611 if (acked > so->so_snd.sb_cc) { 2612 tp->snd_wnd -= so->so_snd.sb_cc; 2613 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2614 ourfinisacked = 1; 2615 } else { 2616 sbdrop_locked(&so->so_snd, acked); 2617 tp->snd_wnd -= acked; 2618 ourfinisacked = 0; 2619 } 2620 /* NB: sowwakeup_locked() does an implicit unlock. */ 2621 sowwakeup_locked(so); 2622 /* Detect una wraparound. */ 2623 if (!IN_RECOVERY(tp->t_flags) && 2624 SEQ_GT(tp->snd_una, tp->snd_recover) && 2625 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2626 tp->snd_recover = th->th_ack - 1; 2627 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2628 if (IN_RECOVERY(tp->t_flags) && 2629 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2630 EXIT_RECOVERY(tp->t_flags); 2631 } 2632 tp->snd_una = th->th_ack; 2633 if (tp->t_flags & TF_SACK_PERMIT) { 2634 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2635 tp->snd_recover = tp->snd_una; 2636 } 2637 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2638 tp->snd_nxt = tp->snd_una; 2639 2640 switch (tp->t_state) { 2641 2642 /* 2643 * In FIN_WAIT_1 STATE in addition to the processing 2644 * for the ESTABLISHED state if our FIN is now acknowledged 2645 * then enter FIN_WAIT_2. 2646 */ 2647 case TCPS_FIN_WAIT_1: 2648 if (ourfinisacked) { 2649 /* 2650 * If we can't receive any more 2651 * data, then closing user can proceed. 2652 * Starting the timer is contrary to the 2653 * specification, but if we don't get a FIN 2654 * we'll hang forever. 2655 * 2656 * XXXjl: 2657 * we should release the tp also, and use a 2658 * compressed state. 2659 */ 2660 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2661 int timeout; 2662 2663 soisdisconnected(so); 2664 timeout = (tcp_fast_finwait2_recycle) ? 2665 tcp_finwait2_timeout : tcp_maxidle; 2666 tcp_timer_activate(tp, TT_2MSL, timeout); 2667 } 2668 tp->t_state = TCPS_FIN_WAIT_2; 2669 } 2670 break; 2671 2672 /* 2673 * In CLOSING STATE in addition to the processing for 2674 * the ESTABLISHED state if the ACK acknowledges our FIN 2675 * then enter the TIME-WAIT state, otherwise ignore 2676 * the segment. 2677 */ 2678 case TCPS_CLOSING: 2679 if (ourfinisacked) { 2680 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2681 tcp_twstart(tp); 2682 INP_INFO_WUNLOCK(&V_tcbinfo); 2683 m_freem(m); 2684 return; 2685 } 2686 break; 2687 2688 /* 2689 * In LAST_ACK, we may still be waiting for data to drain 2690 * and/or to be acked, as well as for the ack of our FIN. 2691 * If our FIN is now acknowledged, delete the TCB, 2692 * enter the closed state and return. 2693 */ 2694 case TCPS_LAST_ACK: 2695 if (ourfinisacked) { 2696 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2697 tp = tcp_close(tp); 2698 goto drop; 2699 } 2700 break; 2701 } 2702 } 2703 2704 step6: 2705 INP_WLOCK_ASSERT(tp->t_inpcb); 2706 2707 /* 2708 * Update window information. 2709 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2710 */ 2711 if ((thflags & TH_ACK) && 2712 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2713 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2714 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2715 /* keep track of pure window updates */ 2716 if (tlen == 0 && 2717 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2718 TCPSTAT_INC(tcps_rcvwinupd); 2719 tp->snd_wnd = tiwin; 2720 tp->snd_wl1 = th->th_seq; 2721 tp->snd_wl2 = th->th_ack; 2722 if (tp->snd_wnd > tp->max_sndwnd) 2723 tp->max_sndwnd = tp->snd_wnd; 2724 needoutput = 1; 2725 } 2726 2727 /* 2728 * Process segments with URG. 2729 */ 2730 if ((thflags & TH_URG) && th->th_urp && 2731 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2732 /* 2733 * This is a kludge, but if we receive and accept 2734 * random urgent pointers, we'll crash in 2735 * soreceive. It's hard to imagine someone 2736 * actually wanting to send this much urgent data. 2737 */ 2738 SOCKBUF_LOCK(&so->so_rcv); 2739 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2740 th->th_urp = 0; /* XXX */ 2741 thflags &= ~TH_URG; /* XXX */ 2742 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2743 goto dodata; /* XXX */ 2744 } 2745 /* 2746 * If this segment advances the known urgent pointer, 2747 * then mark the data stream. This should not happen 2748 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2749 * a FIN has been received from the remote side. 2750 * In these states we ignore the URG. 2751 * 2752 * According to RFC961 (Assigned Protocols), 2753 * the urgent pointer points to the last octet 2754 * of urgent data. We continue, however, 2755 * to consider it to indicate the first octet 2756 * of data past the urgent section as the original 2757 * spec states (in one of two places). 2758 */ 2759 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2760 tp->rcv_up = th->th_seq + th->th_urp; 2761 so->so_oobmark = so->so_rcv.sb_cc + 2762 (tp->rcv_up - tp->rcv_nxt) - 1; 2763 if (so->so_oobmark == 0) 2764 so->so_rcv.sb_state |= SBS_RCVATMARK; 2765 sohasoutofband(so); 2766 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2767 } 2768 SOCKBUF_UNLOCK(&so->so_rcv); 2769 /* 2770 * Remove out of band data so doesn't get presented to user. 2771 * This can happen independent of advancing the URG pointer, 2772 * but if two URG's are pending at once, some out-of-band 2773 * data may creep in... ick. 2774 */ 2775 if (th->th_urp <= (u_long)tlen && 2776 !(so->so_options & SO_OOBINLINE)) { 2777 /* hdr drop is delayed */ 2778 tcp_pulloutofband(so, th, m, drop_hdrlen); 2779 } 2780 } else { 2781 /* 2782 * If no out of band data is expected, 2783 * pull receive urgent pointer along 2784 * with the receive window. 2785 */ 2786 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2787 tp->rcv_up = tp->rcv_nxt; 2788 } 2789 dodata: /* XXX */ 2790 INP_WLOCK_ASSERT(tp->t_inpcb); 2791 2792 /* 2793 * Process the segment text, merging it into the TCP sequencing queue, 2794 * and arranging for acknowledgment of receipt if necessary. 2795 * This process logically involves adjusting tp->rcv_wnd as data 2796 * is presented to the user (this happens in tcp_usrreq.c, 2797 * case PRU_RCVD). If a FIN has already been received on this 2798 * connection then we just ignore the text. 2799 */ 2800 if ((tlen || (thflags & TH_FIN)) && 2801 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2802 tcp_seq save_start = th->th_seq; 2803 m_adj(m, drop_hdrlen); /* delayed header drop */ 2804 /* 2805 * Insert segment which includes th into TCP reassembly queue 2806 * with control block tp. Set thflags to whether reassembly now 2807 * includes a segment with FIN. This handles the common case 2808 * inline (segment is the next to be received on an established 2809 * connection, and the queue is empty), avoiding linkage into 2810 * and removal from the queue and repetition of various 2811 * conversions. 2812 * Set DELACK for segments received in order, but ack 2813 * immediately when segments are out of order (so 2814 * fast retransmit can work). 2815 */ 2816 if (th->th_seq == tp->rcv_nxt && 2817 LIST_EMPTY(&tp->t_segq) && 2818 TCPS_HAVEESTABLISHED(tp->t_state)) { 2819 if (DELAY_ACK(tp)) 2820 tp->t_flags |= TF_DELACK; 2821 else 2822 tp->t_flags |= TF_ACKNOW; 2823 tp->rcv_nxt += tlen; 2824 thflags = th->th_flags & TH_FIN; 2825 TCPSTAT_INC(tcps_rcvpack); 2826 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2827 ND6_HINT(tp); 2828 SOCKBUF_LOCK(&so->so_rcv); 2829 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2830 m_freem(m); 2831 else 2832 sbappendstream_locked(&so->so_rcv, m); 2833 /* NB: sorwakeup_locked() does an implicit unlock. */ 2834 sorwakeup_locked(so); 2835 } else { 2836 /* 2837 * XXX: Due to the header drop above "th" is 2838 * theoretically invalid by now. Fortunately 2839 * m_adj() doesn't actually frees any mbufs 2840 * when trimming from the head. 2841 */ 2842 thflags = tcp_reass(tp, th, &tlen, m); 2843 tp->t_flags |= TF_ACKNOW; 2844 } 2845 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2846 tcp_update_sack_list(tp, save_start, save_start + tlen); 2847 #if 0 2848 /* 2849 * Note the amount of data that peer has sent into 2850 * our window, in order to estimate the sender's 2851 * buffer size. 2852 * XXX: Unused. 2853 */ 2854 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2855 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2856 else 2857 len = so->so_rcv.sb_hiwat; 2858 #endif 2859 } else { 2860 m_freem(m); 2861 thflags &= ~TH_FIN; 2862 } 2863 2864 /* 2865 * If FIN is received ACK the FIN and let the user know 2866 * that the connection is closing. 2867 */ 2868 if (thflags & TH_FIN) { 2869 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2870 socantrcvmore(so); 2871 /* 2872 * If connection is half-synchronized 2873 * (ie NEEDSYN flag on) then delay ACK, 2874 * so it may be piggybacked when SYN is sent. 2875 * Otherwise, since we received a FIN then no 2876 * more input can be expected, send ACK now. 2877 */ 2878 if (tp->t_flags & TF_NEEDSYN) 2879 tp->t_flags |= TF_DELACK; 2880 else 2881 tp->t_flags |= TF_ACKNOW; 2882 tp->rcv_nxt++; 2883 } 2884 switch (tp->t_state) { 2885 2886 /* 2887 * In SYN_RECEIVED and ESTABLISHED STATES 2888 * enter the CLOSE_WAIT state. 2889 */ 2890 case TCPS_SYN_RECEIVED: 2891 tp->t_starttime = ticks; 2892 /* FALLTHROUGH */ 2893 case TCPS_ESTABLISHED: 2894 tp->t_state = TCPS_CLOSE_WAIT; 2895 break; 2896 2897 /* 2898 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2899 * enter the CLOSING state. 2900 */ 2901 case TCPS_FIN_WAIT_1: 2902 tp->t_state = TCPS_CLOSING; 2903 break; 2904 2905 /* 2906 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2907 * starting the time-wait timer, turning off the other 2908 * standard timers. 2909 */ 2910 case TCPS_FIN_WAIT_2: 2911 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2912 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2913 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2914 ti_locked)); 2915 2916 tcp_twstart(tp); 2917 INP_INFO_WUNLOCK(&V_tcbinfo); 2918 return; 2919 } 2920 } 2921 if (ti_locked == TI_WLOCKED) 2922 INP_INFO_WUNLOCK(&V_tcbinfo); 2923 ti_locked = TI_UNLOCKED; 2924 2925 #ifdef TCPDEBUG 2926 if (so->so_options & SO_DEBUG) 2927 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2928 &tcp_savetcp, 0); 2929 #endif 2930 2931 /* 2932 * Return any desired output. 2933 */ 2934 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2935 (void) tcp_output(tp); 2936 2937 check_delack: 2938 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2939 __func__, ti_locked)); 2940 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2941 INP_WLOCK_ASSERT(tp->t_inpcb); 2942 2943 if (tp->t_flags & TF_DELACK) { 2944 tp->t_flags &= ~TF_DELACK; 2945 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2946 } 2947 INP_WUNLOCK(tp->t_inpcb); 2948 return; 2949 2950 dropafterack: 2951 /* 2952 * Generate an ACK dropping incoming segment if it occupies 2953 * sequence space, where the ACK reflects our state. 2954 * 2955 * We can now skip the test for the RST flag since all 2956 * paths to this code happen after packets containing 2957 * RST have been dropped. 2958 * 2959 * In the SYN-RECEIVED state, don't send an ACK unless the 2960 * segment we received passes the SYN-RECEIVED ACK test. 2961 * If it fails send a RST. This breaks the loop in the 2962 * "LAND" DoS attack, and also prevents an ACK storm 2963 * between two listening ports that have been sent forged 2964 * SYN segments, each with the source address of the other. 2965 */ 2966 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2967 (SEQ_GT(tp->snd_una, th->th_ack) || 2968 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2969 rstreason = BANDLIM_RST_OPENPORT; 2970 goto dropwithreset; 2971 } 2972 #ifdef TCPDEBUG 2973 if (so->so_options & SO_DEBUG) 2974 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2975 &tcp_savetcp, 0); 2976 #endif 2977 if (ti_locked == TI_WLOCKED) 2978 INP_INFO_WUNLOCK(&V_tcbinfo); 2979 ti_locked = TI_UNLOCKED; 2980 2981 tp->t_flags |= TF_ACKNOW; 2982 (void) tcp_output(tp); 2983 INP_WUNLOCK(tp->t_inpcb); 2984 m_freem(m); 2985 return; 2986 2987 dropwithreset: 2988 if (ti_locked == TI_WLOCKED) 2989 INP_INFO_WUNLOCK(&V_tcbinfo); 2990 ti_locked = TI_UNLOCKED; 2991 2992 if (tp != NULL) { 2993 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2994 INP_WUNLOCK(tp->t_inpcb); 2995 } else 2996 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 2997 return; 2998 2999 drop: 3000 if (ti_locked == TI_WLOCKED) { 3001 INP_INFO_WUNLOCK(&V_tcbinfo); 3002 ti_locked = TI_UNLOCKED; 3003 } 3004 #ifdef INVARIANTS 3005 else 3006 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3007 #endif 3008 3009 /* 3010 * Drop space held by incoming segment and return. 3011 */ 3012 #ifdef TCPDEBUG 3013 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3014 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3015 &tcp_savetcp, 0); 3016 #endif 3017 if (tp != NULL) 3018 INP_WUNLOCK(tp->t_inpcb); 3019 m_freem(m); 3020 } 3021 3022 /* 3023 * Issue RST and make ACK acceptable to originator of segment. 3024 * The mbuf must still include the original packet header. 3025 * tp may be NULL. 3026 */ 3027 static void 3028 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3029 int tlen, int rstreason) 3030 { 3031 #ifdef INET 3032 struct ip *ip; 3033 #endif 3034 #ifdef INET6 3035 struct ip6_hdr *ip6; 3036 #endif 3037 3038 if (tp != NULL) { 3039 INP_WLOCK_ASSERT(tp->t_inpcb); 3040 } 3041 3042 /* Don't bother if destination was broadcast/multicast. */ 3043 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3044 goto drop; 3045 #ifdef INET6 3046 if (mtod(m, struct ip *)->ip_v == 6) { 3047 ip6 = mtod(m, struct ip6_hdr *); 3048 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3049 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3050 goto drop; 3051 /* IPv6 anycast check is done at tcp6_input() */ 3052 } 3053 #endif 3054 #if defined(INET) && defined(INET6) 3055 else 3056 #endif 3057 #ifdef INET 3058 { 3059 ip = mtod(m, struct ip *); 3060 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3061 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3062 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3063 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3064 goto drop; 3065 } 3066 #endif 3067 3068 /* Perform bandwidth limiting. */ 3069 if (badport_bandlim(rstreason) < 0) 3070 goto drop; 3071 3072 /* tcp_respond consumes the mbuf chain. */ 3073 if (th->th_flags & TH_ACK) { 3074 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3075 th->th_ack, TH_RST); 3076 } else { 3077 if (th->th_flags & TH_SYN) 3078 tlen++; 3079 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3080 (tcp_seq)0, TH_RST|TH_ACK); 3081 } 3082 return; 3083 drop: 3084 m_freem(m); 3085 } 3086 3087 /* 3088 * Parse TCP options and place in tcpopt. 3089 */ 3090 static void 3091 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3092 { 3093 int opt, optlen; 3094 3095 to->to_flags = 0; 3096 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3097 opt = cp[0]; 3098 if (opt == TCPOPT_EOL) 3099 break; 3100 if (opt == TCPOPT_NOP) 3101 optlen = 1; 3102 else { 3103 if (cnt < 2) 3104 break; 3105 optlen = cp[1]; 3106 if (optlen < 2 || optlen > cnt) 3107 break; 3108 } 3109 switch (opt) { 3110 case TCPOPT_MAXSEG: 3111 if (optlen != TCPOLEN_MAXSEG) 3112 continue; 3113 if (!(flags & TO_SYN)) 3114 continue; 3115 to->to_flags |= TOF_MSS; 3116 bcopy((char *)cp + 2, 3117 (char *)&to->to_mss, sizeof(to->to_mss)); 3118 to->to_mss = ntohs(to->to_mss); 3119 break; 3120 case TCPOPT_WINDOW: 3121 if (optlen != TCPOLEN_WINDOW) 3122 continue; 3123 if (!(flags & TO_SYN)) 3124 continue; 3125 to->to_flags |= TOF_SCALE; 3126 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3127 break; 3128 case TCPOPT_TIMESTAMP: 3129 if (optlen != TCPOLEN_TIMESTAMP) 3130 continue; 3131 to->to_flags |= TOF_TS; 3132 bcopy((char *)cp + 2, 3133 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3134 to->to_tsval = ntohl(to->to_tsval); 3135 bcopy((char *)cp + 6, 3136 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3137 to->to_tsecr = ntohl(to->to_tsecr); 3138 break; 3139 #ifdef TCP_SIGNATURE 3140 /* 3141 * XXX In order to reply to a host which has set the 3142 * TCP_SIGNATURE option in its initial SYN, we have to 3143 * record the fact that the option was observed here 3144 * for the syncache code to perform the correct response. 3145 */ 3146 case TCPOPT_SIGNATURE: 3147 if (optlen != TCPOLEN_SIGNATURE) 3148 continue; 3149 to->to_flags |= TOF_SIGNATURE; 3150 to->to_signature = cp + 2; 3151 break; 3152 #endif 3153 case TCPOPT_SACK_PERMITTED: 3154 if (optlen != TCPOLEN_SACK_PERMITTED) 3155 continue; 3156 if (!(flags & TO_SYN)) 3157 continue; 3158 if (!V_tcp_do_sack) 3159 continue; 3160 to->to_flags |= TOF_SACKPERM; 3161 break; 3162 case TCPOPT_SACK: 3163 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3164 continue; 3165 if (flags & TO_SYN) 3166 continue; 3167 to->to_flags |= TOF_SACK; 3168 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3169 to->to_sacks = cp + 2; 3170 TCPSTAT_INC(tcps_sack_rcv_blocks); 3171 break; 3172 default: 3173 continue; 3174 } 3175 } 3176 } 3177 3178 /* 3179 * Pull out of band byte out of a segment so 3180 * it doesn't appear in the user's data queue. 3181 * It is still reflected in the segment length for 3182 * sequencing purposes. 3183 */ 3184 static void 3185 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3186 int off) 3187 { 3188 int cnt = off + th->th_urp - 1; 3189 3190 while (cnt >= 0) { 3191 if (m->m_len > cnt) { 3192 char *cp = mtod(m, caddr_t) + cnt; 3193 struct tcpcb *tp = sototcpcb(so); 3194 3195 INP_WLOCK_ASSERT(tp->t_inpcb); 3196 3197 tp->t_iobc = *cp; 3198 tp->t_oobflags |= TCPOOB_HAVEDATA; 3199 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3200 m->m_len--; 3201 if (m->m_flags & M_PKTHDR) 3202 m->m_pkthdr.len--; 3203 return; 3204 } 3205 cnt -= m->m_len; 3206 m = m->m_next; 3207 if (m == NULL) 3208 break; 3209 } 3210 panic("tcp_pulloutofband"); 3211 } 3212 3213 /* 3214 * Collect new round-trip time estimate 3215 * and update averages and current timeout. 3216 */ 3217 static void 3218 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3219 { 3220 int delta; 3221 3222 INP_WLOCK_ASSERT(tp->t_inpcb); 3223 3224 TCPSTAT_INC(tcps_rttupdated); 3225 tp->t_rttupdated++; 3226 if (tp->t_srtt != 0) { 3227 /* 3228 * srtt is stored as fixed point with 5 bits after the 3229 * binary point (i.e., scaled by 8). The following magic 3230 * is equivalent to the smoothing algorithm in rfc793 with 3231 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3232 * point). Adjust rtt to origin 0. 3233 */ 3234 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3235 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3236 3237 if ((tp->t_srtt += delta) <= 0) 3238 tp->t_srtt = 1; 3239 3240 /* 3241 * We accumulate a smoothed rtt variance (actually, a 3242 * smoothed mean difference), then set the retransmit 3243 * timer to smoothed rtt + 4 times the smoothed variance. 3244 * rttvar is stored as fixed point with 4 bits after the 3245 * binary point (scaled by 16). The following is 3246 * equivalent to rfc793 smoothing with an alpha of .75 3247 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3248 * rfc793's wired-in beta. 3249 */ 3250 if (delta < 0) 3251 delta = -delta; 3252 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3253 if ((tp->t_rttvar += delta) <= 0) 3254 tp->t_rttvar = 1; 3255 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3256 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3257 } else { 3258 /* 3259 * No rtt measurement yet - use the unsmoothed rtt. 3260 * Set the variance to half the rtt (so our first 3261 * retransmit happens at 3*rtt). 3262 */ 3263 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3264 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3265 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3266 } 3267 tp->t_rtttime = 0; 3268 tp->t_rxtshift = 0; 3269 3270 /* 3271 * the retransmit should happen at rtt + 4 * rttvar. 3272 * Because of the way we do the smoothing, srtt and rttvar 3273 * will each average +1/2 tick of bias. When we compute 3274 * the retransmit timer, we want 1/2 tick of rounding and 3275 * 1 extra tick because of +-1/2 tick uncertainty in the 3276 * firing of the timer. The bias will give us exactly the 3277 * 1.5 tick we need. But, because the bias is 3278 * statistical, we have to test that we don't drop below 3279 * the minimum feasible timer (which is 2 ticks). 3280 */ 3281 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3282 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3283 3284 /* 3285 * We received an ack for a packet that wasn't retransmitted; 3286 * it is probably safe to discard any error indications we've 3287 * received recently. This isn't quite right, but close enough 3288 * for now (a route might have failed after we sent a segment, 3289 * and the return path might not be symmetrical). 3290 */ 3291 tp->t_softerror = 0; 3292 } 3293 3294 /* 3295 * Determine a reasonable value for maxseg size. 3296 * If the route is known, check route for mtu. 3297 * If none, use an mss that can be handled on the outgoing 3298 * interface without forcing IP to fragment; if bigger than 3299 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 3300 * to utilize large mbufs. If no route is found, route has no mtu, 3301 * or the destination isn't local, use a default, hopefully conservative 3302 * size (usually 512 or the default IP max size, but no more than the mtu 3303 * of the interface), as we can't discover anything about intervening 3304 * gateways or networks. We also initialize the congestion/slow start 3305 * window to be a single segment if the destination isn't local. 3306 * While looking at the routing entry, we also initialize other path-dependent 3307 * parameters from pre-set or cached values in the routing entry. 3308 * 3309 * Also take into account the space needed for options that we 3310 * send regularly. Make maxseg shorter by that amount to assure 3311 * that we can send maxseg amount of data even when the options 3312 * are present. Store the upper limit of the length of options plus 3313 * data in maxopd. 3314 * 3315 * In case of T/TCP, we call this routine during implicit connection 3316 * setup as well (offer = -1), to initialize maxseg from the cached 3317 * MSS of our peer. 3318 * 3319 * NOTE that this routine is only called when we process an incoming 3320 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 3321 */ 3322 void 3323 tcp_mss_update(struct tcpcb *tp, int offer, 3324 struct hc_metrics_lite *metricptr, int *mtuflags) 3325 { 3326 int mss = 0; 3327 u_long maxmtu = 0; 3328 struct inpcb *inp = tp->t_inpcb; 3329 struct hc_metrics_lite metrics; 3330 int origoffer = offer; 3331 #ifdef INET6 3332 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3333 size_t min_protoh = isipv6 ? 3334 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3335 sizeof (struct tcpiphdr); 3336 #else 3337 const size_t min_protoh = sizeof(struct tcpiphdr); 3338 #endif 3339 3340 INP_WLOCK_ASSERT(tp->t_inpcb); 3341 3342 /* Initialize. */ 3343 #ifdef INET6 3344 if (isipv6) { 3345 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags); 3346 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3347 } 3348 #endif 3349 #if defined(INET) && defined(INET6) 3350 else 3351 #endif 3352 #ifdef INET 3353 { 3354 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags); 3355 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3356 } 3357 #endif 3358 3359 /* 3360 * No route to sender, stay with default mss and return. 3361 */ 3362 if (maxmtu == 0) { 3363 /* 3364 * In case we return early we need to initialize metrics 3365 * to a defined state as tcp_hc_get() would do for us 3366 * if there was no cache hit. 3367 */ 3368 if (metricptr != NULL) 3369 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3370 return; 3371 } 3372 3373 /* What have we got? */ 3374 switch (offer) { 3375 case 0: 3376 /* 3377 * Offer == 0 means that there was no MSS on the SYN 3378 * segment, in this case we use tcp_mssdflt as 3379 * already assigned to t_maxopd above. 3380 */ 3381 offer = tp->t_maxopd; 3382 break; 3383 3384 case -1: 3385 /* 3386 * Offer == -1 means that we didn't receive SYN yet. 3387 */ 3388 /* FALLTHROUGH */ 3389 3390 default: 3391 /* 3392 * Prevent DoS attack with too small MSS. Round up 3393 * to at least minmss. 3394 */ 3395 offer = max(offer, V_tcp_minmss); 3396 } 3397 3398 /* 3399 * rmx information is now retrieved from tcp_hostcache. 3400 */ 3401 tcp_hc_get(&inp->inp_inc, &metrics); 3402 if (metricptr != NULL) 3403 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3404 3405 /* 3406 * If there's a discovered mtu int tcp hostcache, use it 3407 * else, use the link mtu. 3408 */ 3409 if (metrics.rmx_mtu) 3410 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3411 else { 3412 #ifdef INET6 3413 if (isipv6) { 3414 mss = maxmtu - min_protoh; 3415 if (!V_path_mtu_discovery && 3416 !in6_localaddr(&inp->in6p_faddr)) 3417 mss = min(mss, V_tcp_v6mssdflt); 3418 } 3419 #endif 3420 #if defined(INET) && defined(INET6) 3421 else 3422 #endif 3423 #ifdef INET 3424 { 3425 mss = maxmtu - min_protoh; 3426 if (!V_path_mtu_discovery && 3427 !in_localaddr(inp->inp_faddr)) 3428 mss = min(mss, V_tcp_mssdflt); 3429 } 3430 #endif 3431 /* 3432 * XXX - The above conditional (mss = maxmtu - min_protoh) 3433 * probably violates the TCP spec. 3434 * The problem is that, since we don't know the 3435 * other end's MSS, we are supposed to use a conservative 3436 * default. But, if we do that, then MTU discovery will 3437 * never actually take place, because the conservative 3438 * default is much less than the MTUs typically seen 3439 * on the Internet today. For the moment, we'll sweep 3440 * this under the carpet. 3441 * 3442 * The conservative default might not actually be a problem 3443 * if the only case this occurs is when sending an initial 3444 * SYN with options and data to a host we've never talked 3445 * to before. Then, they will reply with an MSS value which 3446 * will get recorded and the new parameters should get 3447 * recomputed. For Further Study. 3448 */ 3449 } 3450 mss = min(mss, offer); 3451 3452 /* 3453 * Sanity check: make sure that maxopd will be large 3454 * enough to allow some data on segments even if the 3455 * all the option space is used (40bytes). Otherwise 3456 * funny things may happen in tcp_output. 3457 */ 3458 mss = max(mss, 64); 3459 3460 /* 3461 * maxopd stores the maximum length of data AND options 3462 * in a segment; maxseg is the amount of data in a normal 3463 * segment. We need to store this value (maxopd) apart 3464 * from maxseg, because now every segment carries options 3465 * and thus we normally have somewhat less data in segments. 3466 */ 3467 tp->t_maxopd = mss; 3468 3469 /* 3470 * origoffer==-1 indicates that no segments were received yet. 3471 * In this case we just guess. 3472 */ 3473 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3474 (origoffer == -1 || 3475 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3476 mss -= TCPOLEN_TSTAMP_APPA; 3477 3478 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3479 if (mss > MCLBYTES) 3480 mss &= ~(MCLBYTES-1); 3481 #else 3482 if (mss > MCLBYTES) 3483 mss = mss / MCLBYTES * MCLBYTES; 3484 #endif 3485 tp->t_maxseg = mss; 3486 } 3487 3488 void 3489 tcp_mss(struct tcpcb *tp, int offer) 3490 { 3491 int mss; 3492 u_long bufsize; 3493 struct inpcb *inp; 3494 struct socket *so; 3495 struct hc_metrics_lite metrics; 3496 int mtuflags = 0; 3497 3498 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3499 3500 tcp_mss_update(tp, offer, &metrics, &mtuflags); 3501 3502 mss = tp->t_maxseg; 3503 inp = tp->t_inpcb; 3504 3505 /* 3506 * If there's a pipesize, change the socket buffer to that size, 3507 * don't change if sb_hiwat is different than default (then it 3508 * has been changed on purpose with setsockopt). 3509 * Make the socket buffers an integral number of mss units; 3510 * if the mss is larger than the socket buffer, decrease the mss. 3511 */ 3512 so = inp->inp_socket; 3513 SOCKBUF_LOCK(&so->so_snd); 3514 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3515 bufsize = metrics.rmx_sendpipe; 3516 else 3517 bufsize = so->so_snd.sb_hiwat; 3518 if (bufsize < mss) 3519 mss = bufsize; 3520 else { 3521 bufsize = roundup(bufsize, mss); 3522 if (bufsize > sb_max) 3523 bufsize = sb_max; 3524 if (bufsize > so->so_snd.sb_hiwat) 3525 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3526 } 3527 SOCKBUF_UNLOCK(&so->so_snd); 3528 tp->t_maxseg = mss; 3529 3530 SOCKBUF_LOCK(&so->so_rcv); 3531 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3532 bufsize = metrics.rmx_recvpipe; 3533 else 3534 bufsize = so->so_rcv.sb_hiwat; 3535 if (bufsize > mss) { 3536 bufsize = roundup(bufsize, mss); 3537 if (bufsize > sb_max) 3538 bufsize = sb_max; 3539 if (bufsize > so->so_rcv.sb_hiwat) 3540 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3541 } 3542 SOCKBUF_UNLOCK(&so->so_rcv); 3543 3544 /* Check the interface for TSO capabilities. */ 3545 if (mtuflags & CSUM_TSO) 3546 tp->t_flags |= TF_TSO; 3547 } 3548 3549 /* 3550 * Determine the MSS option to send on an outgoing SYN. 3551 */ 3552 int 3553 tcp_mssopt(struct in_conninfo *inc) 3554 { 3555 int mss = 0; 3556 u_long maxmtu = 0; 3557 u_long thcmtu = 0; 3558 size_t min_protoh; 3559 3560 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3561 3562 #ifdef INET6 3563 if (inc->inc_flags & INC_ISIPV6) { 3564 mss = V_tcp_v6mssdflt; 3565 maxmtu = tcp_maxmtu6(inc, NULL); 3566 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3567 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3568 } 3569 #endif 3570 #if defined(INET) && defined(INET6) 3571 else 3572 #endif 3573 #ifdef INET 3574 { 3575 mss = V_tcp_mssdflt; 3576 maxmtu = tcp_maxmtu(inc, NULL); 3577 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3578 min_protoh = sizeof(struct tcpiphdr); 3579 } 3580 #endif 3581 if (maxmtu && thcmtu) 3582 mss = min(maxmtu, thcmtu) - min_protoh; 3583 else if (maxmtu || thcmtu) 3584 mss = max(maxmtu, thcmtu) - min_protoh; 3585 3586 return (mss); 3587 } 3588 3589 3590 /* 3591 * On a partial ack arrives, force the retransmission of the 3592 * next unacknowledged segment. Do not clear tp->t_dupacks. 3593 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3594 * be started again. 3595 */ 3596 static void 3597 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3598 { 3599 tcp_seq onxt = tp->snd_nxt; 3600 u_long ocwnd = tp->snd_cwnd; 3601 3602 INP_WLOCK_ASSERT(tp->t_inpcb); 3603 3604 tcp_timer_activate(tp, TT_REXMT, 0); 3605 tp->t_rtttime = 0; 3606 tp->snd_nxt = th->th_ack; 3607 /* 3608 * Set snd_cwnd to one segment beyond acknowledged offset. 3609 * (tp->snd_una has not yet been updated when this function is called.) 3610 */ 3611 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3612 tp->t_flags |= TF_ACKNOW; 3613 (void) tcp_output(tp); 3614 tp->snd_cwnd = ocwnd; 3615 if (SEQ_GT(onxt, tp->snd_nxt)) 3616 tp->snd_nxt = onxt; 3617 /* 3618 * Partial window deflation. Relies on fact that tp->snd_una 3619 * not updated yet. 3620 */ 3621 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3622 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3623 else 3624 tp->snd_cwnd = 0; 3625 tp->snd_cwnd += tp->t_maxseg; 3626 } 3627