tcp_input.c revision 242249
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: head/sys/netinet/tcp_input.c 242249 2012-10-28 17:16:09Z andre $"); 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/pfil.h> 79#include <net/route.h> 80#include <net/vnet.h> 81 82#define TCPSTATES /* for logging */ 83 84#include <netinet/cc.h> 85#include <netinet/in.h> 86#include <netinet/in_pcb.h> 87#include <netinet/in_systm.h> 88#include <netinet/in_var.h> 89#include <netinet/ip.h> 90#include <netinet/ip_icmp.h> /* required for icmp_var.h */ 91#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 92#include <netinet/ip_var.h> 93#include <netinet/ip_options.h> 94#include <netinet/ip6.h> 95#include <netinet/icmp6.h> 96#include <netinet6/in6_pcb.h> 97#include <netinet6/ip6_var.h> 98#include <netinet6/nd6.h> 99#include <netinet/tcp_fsm.h> 100#include <netinet/tcp_seq.h> 101#include <netinet/tcp_timer.h> 102#include <netinet/tcp_var.h> 103#include <netinet6/tcp6_var.h> 104#include <netinet/tcpip.h> 105#include <netinet/tcp_syncache.h> 106#ifdef TCPDEBUG 107#include <netinet/tcp_debug.h> 108#endif /* TCPDEBUG */ 109#ifdef TCP_OFFLOAD 110#include <netinet/tcp_offload.h> 111#endif 112 113#ifdef IPSEC 114#include <netipsec/ipsec.h> 115#include <netipsec/ipsec6.h> 116#endif /*IPSEC*/ 117 118#include <machine/in_cksum.h> 119 120#include <security/mac/mac_framework.h> 121 122const int tcprexmtthresh = 3; 123 124VNET_DEFINE(struct tcpstat, tcpstat); 125SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 126 &VNET_NAME(tcpstat), tcpstat, 127 "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 128 129int tcp_log_in_vain = 0; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 131 &tcp_log_in_vain, 0, 132 "Log all incoming TCP segments to closed ports"); 133 134VNET_DEFINE(int, blackhole) = 0; 135#define V_blackhole VNET(blackhole) 136SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 137 &VNET_NAME(blackhole), 0, 138 "Do not send RST on segments to closed ports"); 139 140VNET_DEFINE(int, tcp_delack_enabled) = 1; 141SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 142 &VNET_NAME(tcp_delack_enabled), 0, 143 "Delay ACK to try and piggyback it onto a data packet"); 144 145VNET_DEFINE(int, drop_synfin) = 0; 146#define V_drop_synfin VNET(drop_synfin) 147SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 148 &VNET_NAME(drop_synfin), 0, 149 "Drop TCP packets with SYN+FIN set"); 150 151VNET_DEFINE(int, tcp_do_rfc3042) = 1; 152#define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 153SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 154 &VNET_NAME(tcp_do_rfc3042), 0, 155 "Enable RFC 3042 (Limited Transmit)"); 156 157VNET_DEFINE(int, tcp_do_rfc3390) = 1; 158SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 159 &VNET_NAME(tcp_do_rfc3390), 0, 160 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 161 162VNET_DEFINE(int, tcp_do_rfc3465) = 1; 163SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 164 &VNET_NAME(tcp_do_rfc3465), 0, 165 "Enable RFC 3465 (Appropriate Byte Counting)"); 166 167VNET_DEFINE(int, tcp_abc_l_var) = 2; 168SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 169 &VNET_NAME(tcp_abc_l_var), 2, 170 "Cap the max cwnd increment during slow-start to this number of segments"); 171 172static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 173 174VNET_DEFINE(int, tcp_do_ecn) = 0; 175SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 176 &VNET_NAME(tcp_do_ecn), 0, 177 "TCP ECN support"); 178 179VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 180SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 181 &VNET_NAME(tcp_ecn_maxretries), 0, 182 "Max retries before giving up on ECN"); 183 184VNET_DEFINE(int, tcp_insecure_rst) = 0; 185#define V_tcp_insecure_rst VNET(tcp_insecure_rst) 186SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 187 &VNET_NAME(tcp_insecure_rst), 0, 188 "Follow the old (insecure) criteria for accepting RST packets"); 189 190VNET_DEFINE(int, tcp_recvspace) = 1024*64; 191#define V_tcp_recvspace VNET(tcp_recvspace) 192SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 193 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 194 195VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 196#define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 197SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 198 &VNET_NAME(tcp_do_autorcvbuf), 0, 199 "Enable automatic receive buffer sizing"); 200 201VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 202#define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 203SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 204 &VNET_NAME(tcp_autorcvbuf_inc), 0, 205 "Incrementor step size of automatic receive buffer"); 206 207VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 208#define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 209SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 210 &VNET_NAME(tcp_autorcvbuf_max), 0, 211 "Max size of automatic receive buffer"); 212 213VNET_DEFINE(struct inpcbhead, tcb); 214#define tcb6 tcb /* for KAME src sync over BSD*'s */ 215VNET_DEFINE(struct inpcbinfo, tcbinfo); 216 217static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 218static void tcp_do_segment(struct mbuf *, struct tcphdr *, 219 struct socket *, struct tcpcb *, int, int, uint8_t, 220 int); 221static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 222 struct tcpcb *, int, int); 223static void tcp_pulloutofband(struct socket *, 224 struct tcphdr *, struct mbuf *, int); 225static void tcp_xmit_timer(struct tcpcb *, int); 226static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 227static void inline tcp_fields_to_host(struct tcphdr *); 228#ifdef TCP_SIGNATURE 229static void inline tcp_fields_to_net(struct tcphdr *); 230static int inline tcp_signature_verify_input(struct mbuf *, int, int, 231 int, struct tcpopt *, struct tcphdr *, u_int); 232#endif 233static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 234 uint16_t type); 235static void inline cc_conn_init(struct tcpcb *tp); 236static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 237static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 238 struct tcphdr *th, struct tcpopt *to); 239 240/* 241 * Kernel module interface for updating tcpstat. The argument is an index 242 * into tcpstat treated as an array of u_long. While this encodes the 243 * general layout of tcpstat into the caller, it doesn't encode its location, 244 * so that future changes to add, for example, per-CPU stats support won't 245 * cause binary compatibility problems for kernel modules. 246 */ 247void 248kmod_tcpstat_inc(int statnum) 249{ 250 251 (*((u_long *)&V_tcpstat + statnum))++; 252} 253 254/* 255 * Wrapper for the TCP established input helper hook. 256 */ 257static void inline 258hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 259{ 260 struct tcp_hhook_data hhook_data; 261 262 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 263 hhook_data.tp = tp; 264 hhook_data.th = th; 265 hhook_data.to = to; 266 267 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 268 tp->osd); 269 } 270} 271 272/* 273 * CC wrapper hook functions 274 */ 275static void inline 276cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 277{ 278 INP_WLOCK_ASSERT(tp->t_inpcb); 279 280 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 281 if (tp->snd_cwnd == min(tp->snd_cwnd, tp->snd_wnd)) 282 tp->ccv->flags |= CCF_CWND_LIMITED; 283 else 284 tp->ccv->flags &= ~CCF_CWND_LIMITED; 285 286 if (type == CC_ACK) { 287 if (tp->snd_cwnd > tp->snd_ssthresh) { 288 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 289 V_tcp_abc_l_var * tp->t_maxseg); 290 if (tp->t_bytes_acked >= tp->snd_cwnd) { 291 tp->t_bytes_acked -= tp->snd_cwnd; 292 tp->ccv->flags |= CCF_ABC_SENTAWND; 293 } 294 } else { 295 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 296 tp->t_bytes_acked = 0; 297 } 298 } 299 300 if (CC_ALGO(tp)->ack_received != NULL) { 301 /* XXXLAS: Find a way to live without this */ 302 tp->ccv->curack = th->th_ack; 303 CC_ALGO(tp)->ack_received(tp->ccv, type); 304 } 305} 306 307static void inline 308cc_conn_init(struct tcpcb *tp) 309{ 310 struct hc_metrics_lite metrics; 311 struct inpcb *inp = tp->t_inpcb; 312 int rtt; 313 314 INP_WLOCK_ASSERT(tp->t_inpcb); 315 316 tcp_hc_get(&inp->inp_inc, &metrics); 317 318 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 319 tp->t_srtt = rtt; 320 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 321 TCPSTAT_INC(tcps_usedrtt); 322 if (metrics.rmx_rttvar) { 323 tp->t_rttvar = metrics.rmx_rttvar; 324 TCPSTAT_INC(tcps_usedrttvar); 325 } else { 326 /* default variation is +- 1 rtt */ 327 tp->t_rttvar = 328 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 329 } 330 TCPT_RANGESET(tp->t_rxtcur, 331 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 332 tp->t_rttmin, TCPTV_REXMTMAX); 333 } 334 if (metrics.rmx_ssthresh) { 335 /* 336 * There's some sort of gateway or interface 337 * buffer limit on the path. Use this to set 338 * the slow start threshhold, but set the 339 * threshold to no less than 2*mss. 340 */ 341 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 342 TCPSTAT_INC(tcps_usedssthresh); 343 } 344 345 /* 346 * Set the initial slow-start flight size. 347 * 348 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 349 * XXX: We currently check only in syncache_socket for that. 350 */ 351 if (V_tcp_do_rfc3390) 352 tp->snd_cwnd = min(4 * tp->t_maxseg, 353 max(2 * tp->t_maxseg, 4380)); 354 else { 355 /* Per RFC5681 Section 3.1 */ 356 if (tp->t_maxseg > 2190) 357 tp->snd_cwnd = 2 * tp->t_maxseg; 358 else if (tp->t_maxseg > 1095) 359 tp->snd_cwnd = 3 * tp->t_maxseg; 360 else 361 tp->snd_cwnd = 4 * tp->t_maxseg; 362 } 363 364 if (CC_ALGO(tp)->conn_init != NULL) 365 CC_ALGO(tp)->conn_init(tp->ccv); 366} 367 368void inline 369cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 370{ 371 INP_WLOCK_ASSERT(tp->t_inpcb); 372 373 switch(type) { 374 case CC_NDUPACK: 375 if (!IN_FASTRECOVERY(tp->t_flags)) { 376 tp->snd_recover = tp->snd_max; 377 if (tp->t_flags & TF_ECN_PERMIT) 378 tp->t_flags |= TF_ECN_SND_CWR; 379 } 380 break; 381 case CC_ECN: 382 if (!IN_CONGRECOVERY(tp->t_flags)) { 383 TCPSTAT_INC(tcps_ecn_rcwnd); 384 tp->snd_recover = tp->snd_max; 385 if (tp->t_flags & TF_ECN_PERMIT) 386 tp->t_flags |= TF_ECN_SND_CWR; 387 } 388 break; 389 case CC_RTO: 390 tp->t_dupacks = 0; 391 tp->t_bytes_acked = 0; 392 EXIT_RECOVERY(tp->t_flags); 393 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 394 tp->t_maxseg) * tp->t_maxseg; 395 tp->snd_cwnd = tp->t_maxseg; 396 break; 397 case CC_RTO_ERR: 398 TCPSTAT_INC(tcps_sndrexmitbad); 399 /* RTO was unnecessary, so reset everything. */ 400 tp->snd_cwnd = tp->snd_cwnd_prev; 401 tp->snd_ssthresh = tp->snd_ssthresh_prev; 402 tp->snd_recover = tp->snd_recover_prev; 403 if (tp->t_flags & TF_WASFRECOVERY) 404 ENTER_FASTRECOVERY(tp->t_flags); 405 if (tp->t_flags & TF_WASCRECOVERY) 406 ENTER_CONGRECOVERY(tp->t_flags); 407 tp->snd_nxt = tp->snd_max; 408 tp->t_flags &= ~TF_PREVVALID; 409 tp->t_badrxtwin = 0; 410 break; 411 } 412 413 if (CC_ALGO(tp)->cong_signal != NULL) { 414 if (th != NULL) 415 tp->ccv->curack = th->th_ack; 416 CC_ALGO(tp)->cong_signal(tp->ccv, type); 417 } 418} 419 420static void inline 421cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 422{ 423 INP_WLOCK_ASSERT(tp->t_inpcb); 424 425 /* XXXLAS: KASSERT that we're in recovery? */ 426 427 if (CC_ALGO(tp)->post_recovery != NULL) { 428 tp->ccv->curack = th->th_ack; 429 CC_ALGO(tp)->post_recovery(tp->ccv); 430 } 431 /* XXXLAS: EXIT_RECOVERY ? */ 432 tp->t_bytes_acked = 0; 433} 434 435static inline void 436tcp_fields_to_host(struct tcphdr *th) 437{ 438 439 th->th_seq = ntohl(th->th_seq); 440 th->th_ack = ntohl(th->th_ack); 441 th->th_win = ntohs(th->th_win); 442 th->th_urp = ntohs(th->th_urp); 443} 444 445#ifdef TCP_SIGNATURE 446static inline void 447tcp_fields_to_net(struct tcphdr *th) 448{ 449 450 th->th_seq = htonl(th->th_seq); 451 th->th_ack = htonl(th->th_ack); 452 th->th_win = htons(th->th_win); 453 th->th_urp = htons(th->th_urp); 454} 455 456static inline int 457tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 458 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 459{ 460 int ret; 461 462 tcp_fields_to_net(th); 463 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 464 tcp_fields_to_host(th); 465 return (ret); 466} 467#endif 468 469/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 470#ifdef INET6 471#define ND6_HINT(tp) \ 472do { \ 473 if ((tp) && (tp)->t_inpcb && \ 474 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 475 nd6_nud_hint(NULL, NULL, 0); \ 476} while (0) 477#else 478#define ND6_HINT(tp) 479#endif 480 481/* 482 * Indicate whether this ack should be delayed. We can delay the ack if 483 * - there is no delayed ack timer in progress and 484 * - our last ack wasn't a 0-sized window. We never want to delay 485 * the ack that opens up a 0-sized window and 486 * - delayed acks are enabled or 487 * - this is a half-synchronized T/TCP connection. 488 */ 489#define DELAY_ACK(tp) \ 490 ((!tcp_timer_active(tp, TT_DELACK) && \ 491 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 492 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 493 494/* 495 * TCP input handling is split into multiple parts: 496 * tcp6_input is a thin wrapper around tcp_input for the extended 497 * ip6_protox[] call format in ip6_input 498 * tcp_input handles primary segment validation, inpcb lookup and 499 * SYN processing on listen sockets 500 * tcp_do_segment processes the ACK and text of the segment for 501 * establishing, established and closing connections 502 */ 503#ifdef INET6 504int 505tcp6_input(struct mbuf **mp, int *offp, int proto) 506{ 507 struct mbuf *m = *mp; 508 struct in6_ifaddr *ia6; 509 510 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 511 512 /* 513 * draft-itojun-ipv6-tcp-to-anycast 514 * better place to put this in? 515 */ 516 ia6 = ip6_getdstifaddr(m); 517 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 518 struct ip6_hdr *ip6; 519 520 ifa_free(&ia6->ia_ifa); 521 ip6 = mtod(m, struct ip6_hdr *); 522 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 523 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 524 return IPPROTO_DONE; 525 } 526 if (ia6) 527 ifa_free(&ia6->ia_ifa); 528 529 tcp_input(m, *offp); 530 return IPPROTO_DONE; 531} 532#endif /* INET6 */ 533 534void 535tcp_input(struct mbuf *m, int off0) 536{ 537 struct tcphdr *th = NULL; 538 struct ip *ip = NULL; 539 struct inpcb *inp = NULL; 540 struct tcpcb *tp = NULL; 541 struct socket *so = NULL; 542 u_char *optp = NULL; 543 int optlen = 0; 544#ifdef INET 545 int len; 546#endif 547 int tlen = 0, off; 548 int drop_hdrlen; 549 int thflags; 550 int rstreason = 0; /* For badport_bandlim accounting purposes */ 551#ifdef TCP_SIGNATURE 552 uint8_t sig_checked = 0; 553#endif 554 uint8_t iptos = 0; 555 struct m_tag *fwd_tag = NULL; 556#ifdef INET6 557 struct ip6_hdr *ip6 = NULL; 558 int isipv6; 559#else 560 const void *ip6 = NULL; 561#endif /* INET6 */ 562 struct tcpopt to; /* options in this segment */ 563 char *s = NULL; /* address and port logging */ 564 int ti_locked; 565#define TI_UNLOCKED 1 566#define TI_WLOCKED 2 567 568#ifdef TCPDEBUG 569 /* 570 * The size of tcp_saveipgen must be the size of the max ip header, 571 * now IPv6. 572 */ 573 u_char tcp_saveipgen[IP6_HDR_LEN]; 574 struct tcphdr tcp_savetcp; 575 short ostate = 0; 576#endif 577 578#ifdef INET6 579 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 580#endif 581 582 to.to_flags = 0; 583 TCPSTAT_INC(tcps_rcvtotal); 584 585#ifdef INET6 586 if (isipv6) { 587 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 588 589 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 590 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 591 if (m == NULL) { 592 TCPSTAT_INC(tcps_rcvshort); 593 return; 594 } 595 } 596 597 ip6 = mtod(m, struct ip6_hdr *); 598 th = (struct tcphdr *)((caddr_t)ip6 + off0); 599 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 600 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 601 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 602 th->th_sum = m->m_pkthdr.csum_data; 603 else 604 th->th_sum = in6_cksum_pseudo(ip6, tlen, 605 IPPROTO_TCP, m->m_pkthdr.csum_data); 606 th->th_sum ^= 0xffff; 607 } else 608 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 609 if (th->th_sum) { 610 TCPSTAT_INC(tcps_rcvbadsum); 611 goto drop; 612 } 613 614 /* 615 * Be proactive about unspecified IPv6 address in source. 616 * As we use all-zero to indicate unbounded/unconnected pcb, 617 * unspecified IPv6 address can be used to confuse us. 618 * 619 * Note that packets with unspecified IPv6 destination is 620 * already dropped in ip6_input. 621 */ 622 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 623 /* XXX stat */ 624 goto drop; 625 } 626 } 627#endif 628#if defined(INET) && defined(INET6) 629 else 630#endif 631#ifdef INET 632 { 633 /* 634 * Get IP and TCP header together in first mbuf. 635 * Note: IP leaves IP header in first mbuf. 636 */ 637 if (off0 > sizeof (struct ip)) { 638 ip_stripoptions(m); 639 off0 = sizeof(struct ip); 640 } 641 if (m->m_len < sizeof (struct tcpiphdr)) { 642 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 643 == NULL) { 644 TCPSTAT_INC(tcps_rcvshort); 645 return; 646 } 647 } 648 ip = mtod(m, struct ip *); 649 th = (struct tcphdr *)((caddr_t)ip + off0); 650 tlen = ntohs(ip->ip_len) - off0; 651 652 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 653 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 654 th->th_sum = m->m_pkthdr.csum_data; 655 else 656 th->th_sum = in_pseudo(ip->ip_src.s_addr, 657 ip->ip_dst.s_addr, 658 htonl(m->m_pkthdr.csum_data + tlen + 659 IPPROTO_TCP)); 660 th->th_sum ^= 0xffff; 661 } else { 662 struct ipovly *ipov = (struct ipovly *)ip; 663 664 /* 665 * Checksum extended TCP header and data. 666 */ 667 len = off0 + tlen; 668 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 669 ipov->ih_len = htons(tlen); 670 th->th_sum = in_cksum(m, len); 671 } 672 if (th->th_sum) { 673 TCPSTAT_INC(tcps_rcvbadsum); 674 goto drop; 675 } 676 /* Re-initialization for later version check */ 677 ip->ip_v = IPVERSION; 678 } 679#endif /* INET */ 680 681#ifdef INET6 682 if (isipv6) 683 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 684#endif 685#if defined(INET) && defined(INET6) 686 else 687#endif 688#ifdef INET 689 iptos = ip->ip_tos; 690#endif 691 692 /* 693 * Check that TCP offset makes sense, 694 * pull out TCP options and adjust length. XXX 695 */ 696 off = th->th_off << 2; 697 if (off < sizeof (struct tcphdr) || off > tlen) { 698 TCPSTAT_INC(tcps_rcvbadoff); 699 goto drop; 700 } 701 tlen -= off; /* tlen is used instead of ti->ti_len */ 702 if (off > sizeof (struct tcphdr)) { 703#ifdef INET6 704 if (isipv6) { 705 IP6_EXTHDR_CHECK(m, off0, off, ); 706 ip6 = mtod(m, struct ip6_hdr *); 707 th = (struct tcphdr *)((caddr_t)ip6 + off0); 708 } 709#endif 710#if defined(INET) && defined(INET6) 711 else 712#endif 713#ifdef INET 714 { 715 if (m->m_len < sizeof(struct ip) + off) { 716 if ((m = m_pullup(m, sizeof (struct ip) + off)) 717 == NULL) { 718 TCPSTAT_INC(tcps_rcvshort); 719 return; 720 } 721 ip = mtod(m, struct ip *); 722 th = (struct tcphdr *)((caddr_t)ip + off0); 723 } 724 } 725#endif 726 optlen = off - sizeof (struct tcphdr); 727 optp = (u_char *)(th + 1); 728 } 729 thflags = th->th_flags; 730 731 /* 732 * Convert TCP protocol specific fields to host format. 733 */ 734 tcp_fields_to_host(th); 735 736 /* 737 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 738 */ 739 drop_hdrlen = off0 + off; 740 741 /* 742 * Locate pcb for segment; if we're likely to add or remove a 743 * connection then first acquire pcbinfo lock. There are two cases 744 * where we might discover later we need a write lock despite the 745 * flags: ACKs moving a connection out of the syncache, and ACKs for 746 * a connection in TIMEWAIT. 747 */ 748 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 749 INP_INFO_WLOCK(&V_tcbinfo); 750 ti_locked = TI_WLOCKED; 751 } else 752 ti_locked = TI_UNLOCKED; 753 754findpcb: 755#ifdef INVARIANTS 756 if (ti_locked == TI_WLOCKED) { 757 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 758 } else { 759 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 760 } 761#endif 762 763 /* 764 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 765 */ 766 if (V_pfilforward != 0) 767 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 768 769#ifdef INET6 770 if (isipv6 && fwd_tag != NULL) { 771 struct sockaddr_in6 *next_hop6; 772 773 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 774 /* 775 * Transparently forwarded. Pretend to be the destination. 776 * Already got one like this? 777 */ 778 inp = in6_pcblookup_mbuf(&V_tcbinfo, 779 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 780 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 781 if (!inp) { 782 /* 783 * It's new. Try to find the ambushing socket. 784 * Because we've rewritten the destination address, 785 * any hardware-generated hash is ignored. 786 */ 787 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 788 th->th_sport, &next_hop6->sin6_addr, 789 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 790 th->th_dport, INPLOOKUP_WILDCARD | 791 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 792 } 793 /* Remove the tag from the packet. We don't need it anymore. */ 794 m_tag_delete(m, fwd_tag); 795 } else if (isipv6) { 796 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 797 th->th_sport, &ip6->ip6_dst, th->th_dport, 798 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 799 m->m_pkthdr.rcvif, m); 800 } 801#endif /* INET6 */ 802#if defined(INET6) && defined(INET) 803 else 804#endif 805#ifdef INET 806 if (fwd_tag != NULL) { 807 struct sockaddr_in *next_hop; 808 809 next_hop = (struct sockaddr_in *)(fwd_tag+1); 810 /* 811 * Transparently forwarded. Pretend to be the destination. 812 * already got one like this? 813 */ 814 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 815 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 816 m->m_pkthdr.rcvif, m); 817 if (!inp) { 818 /* 819 * It's new. Try to find the ambushing socket. 820 * Because we've rewritten the destination address, 821 * any hardware-generated hash is ignored. 822 */ 823 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 824 th->th_sport, next_hop->sin_addr, 825 next_hop->sin_port ? ntohs(next_hop->sin_port) : 826 th->th_dport, INPLOOKUP_WILDCARD | 827 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 828 } 829 /* Remove the tag from the packet. We don't need it anymore. */ 830 m_tag_delete(m, fwd_tag); 831 } else 832 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 833 th->th_sport, ip->ip_dst, th->th_dport, 834 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 835 m->m_pkthdr.rcvif, m); 836#endif /* INET */ 837 838 /* 839 * If the INPCB does not exist then all data in the incoming 840 * segment is discarded and an appropriate RST is sent back. 841 * XXX MRT Send RST using which routing table? 842 */ 843 if (inp == NULL) { 844 /* 845 * Log communication attempts to ports that are not 846 * in use. 847 */ 848 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 849 tcp_log_in_vain == 2) { 850 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 851 log(LOG_INFO, "%s; %s: Connection attempt " 852 "to closed port\n", s, __func__); 853 } 854 /* 855 * When blackholing do not respond with a RST but 856 * completely ignore the segment and drop it. 857 */ 858 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 859 V_blackhole == 2) 860 goto dropunlock; 861 862 rstreason = BANDLIM_RST_CLOSEDPORT; 863 goto dropwithreset; 864 } 865 INP_WLOCK_ASSERT(inp); 866 if (!(inp->inp_flags & INP_HW_FLOWID) 867 && (m->m_flags & M_FLOWID) 868 && ((inp->inp_socket == NULL) 869 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 870 inp->inp_flags |= INP_HW_FLOWID; 871 inp->inp_flags &= ~INP_SW_FLOWID; 872 inp->inp_flowid = m->m_pkthdr.flowid; 873 } 874#ifdef IPSEC 875#ifdef INET6 876 if (isipv6 && ipsec6_in_reject(m, inp)) { 877 V_ipsec6stat.in_polvio++; 878 goto dropunlock; 879 } else 880#endif /* INET6 */ 881 if (ipsec4_in_reject(m, inp) != 0) { 882 V_ipsec4stat.in_polvio++; 883 goto dropunlock; 884 } 885#endif /* IPSEC */ 886 887 /* 888 * Check the minimum TTL for socket. 889 */ 890 if (inp->inp_ip_minttl != 0) { 891#ifdef INET6 892 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 893 goto dropunlock; 894 else 895#endif 896 if (inp->inp_ip_minttl > ip->ip_ttl) 897 goto dropunlock; 898 } 899 900 /* 901 * A previous connection in TIMEWAIT state is supposed to catch stray 902 * or duplicate segments arriving late. If this segment was a 903 * legitimate new connection attempt, the old INPCB gets removed and 904 * we can try again to find a listening socket. 905 * 906 * At this point, due to earlier optimism, we may hold only an inpcb 907 * lock, and not the inpcbinfo write lock. If so, we need to try to 908 * acquire it, or if that fails, acquire a reference on the inpcb, 909 * drop all locks, acquire a global write lock, and then re-acquire 910 * the inpcb lock. We may at that point discover that another thread 911 * has tried to free the inpcb, in which case we need to loop back 912 * and try to find a new inpcb to deliver to. 913 * 914 * XXXRW: It may be time to rethink timewait locking. 915 */ 916relocked: 917 if (inp->inp_flags & INP_TIMEWAIT) { 918 if (ti_locked == TI_UNLOCKED) { 919 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 920 in_pcbref(inp); 921 INP_WUNLOCK(inp); 922 INP_INFO_WLOCK(&V_tcbinfo); 923 ti_locked = TI_WLOCKED; 924 INP_WLOCK(inp); 925 if (in_pcbrele_wlocked(inp)) { 926 inp = NULL; 927 goto findpcb; 928 } 929 } else 930 ti_locked = TI_WLOCKED; 931 } 932 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 933 934 if (thflags & TH_SYN) 935 tcp_dooptions(&to, optp, optlen, TO_SYN); 936 /* 937 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 938 */ 939 if (tcp_twcheck(inp, &to, th, m, tlen)) 940 goto findpcb; 941 INP_INFO_WUNLOCK(&V_tcbinfo); 942 return; 943 } 944 /* 945 * The TCPCB may no longer exist if the connection is winding 946 * down or it is in the CLOSED state. Either way we drop the 947 * segment and send an appropriate response. 948 */ 949 tp = intotcpcb(inp); 950 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 951 rstreason = BANDLIM_RST_CLOSEDPORT; 952 goto dropwithreset; 953 } 954 955#ifdef TCP_OFFLOAD 956 if (tp->t_flags & TF_TOE) { 957 tcp_offload_input(tp, m); 958 m = NULL; /* consumed by the TOE driver */ 959 goto dropunlock; 960 } 961#endif 962 963 /* 964 * We've identified a valid inpcb, but it could be that we need an 965 * inpcbinfo write lock but don't hold it. In this case, attempt to 966 * acquire using the same strategy as the TIMEWAIT case above. If we 967 * relock, we have to jump back to 'relocked' as the connection might 968 * now be in TIMEWAIT. 969 */ 970#ifdef INVARIANTS 971 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 972 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 973#endif 974 if (tp->t_state != TCPS_ESTABLISHED) { 975 if (ti_locked == TI_UNLOCKED) { 976 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 977 in_pcbref(inp); 978 INP_WUNLOCK(inp); 979 INP_INFO_WLOCK(&V_tcbinfo); 980 ti_locked = TI_WLOCKED; 981 INP_WLOCK(inp); 982 if (in_pcbrele_wlocked(inp)) { 983 inp = NULL; 984 goto findpcb; 985 } 986 goto relocked; 987 } else 988 ti_locked = TI_WLOCKED; 989 } 990 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 991 } 992 993#ifdef MAC 994 INP_WLOCK_ASSERT(inp); 995 if (mac_inpcb_check_deliver(inp, m)) 996 goto dropunlock; 997#endif 998 so = inp->inp_socket; 999 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1000#ifdef TCPDEBUG 1001 if (so->so_options & SO_DEBUG) { 1002 ostate = tp->t_state; 1003#ifdef INET6 1004 if (isipv6) { 1005 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1006 } else 1007#endif 1008 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1009 tcp_savetcp = *th; 1010 } 1011#endif /* TCPDEBUG */ 1012 /* 1013 * When the socket is accepting connections (the INPCB is in LISTEN 1014 * state) we look into the SYN cache if this is a new connection 1015 * attempt or the completion of a previous one. Because listen 1016 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1017 * held in this case. 1018 */ 1019 if (so->so_options & SO_ACCEPTCONN) { 1020 struct in_conninfo inc; 1021 1022 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1023 "tp not listening", __func__)); 1024 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1025 1026 bzero(&inc, sizeof(inc)); 1027#ifdef INET6 1028 if (isipv6) { 1029 inc.inc_flags |= INC_ISIPV6; 1030 inc.inc6_faddr = ip6->ip6_src; 1031 inc.inc6_laddr = ip6->ip6_dst; 1032 } else 1033#endif 1034 { 1035 inc.inc_faddr = ip->ip_src; 1036 inc.inc_laddr = ip->ip_dst; 1037 } 1038 inc.inc_fport = th->th_sport; 1039 inc.inc_lport = th->th_dport; 1040 inc.inc_fibnum = so->so_fibnum; 1041 1042 /* 1043 * Check for an existing connection attempt in syncache if 1044 * the flag is only ACK. A successful lookup creates a new 1045 * socket appended to the listen queue in SYN_RECEIVED state. 1046 */ 1047 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1048 /* 1049 * Parse the TCP options here because 1050 * syncookies need access to the reflected 1051 * timestamp. 1052 */ 1053 tcp_dooptions(&to, optp, optlen, 0); 1054 /* 1055 * NB: syncache_expand() doesn't unlock 1056 * inp and tcpinfo locks. 1057 */ 1058 if (!syncache_expand(&inc, &to, th, &so, m)) { 1059 /* 1060 * No syncache entry or ACK was not 1061 * for our SYN/ACK. Send a RST. 1062 * NB: syncache did its own logging 1063 * of the failure cause. 1064 */ 1065 rstreason = BANDLIM_RST_OPENPORT; 1066 goto dropwithreset; 1067 } 1068 if (so == NULL) { 1069 /* 1070 * We completed the 3-way handshake 1071 * but could not allocate a socket 1072 * either due to memory shortage, 1073 * listen queue length limits or 1074 * global socket limits. Send RST 1075 * or wait and have the remote end 1076 * retransmit the ACK for another 1077 * try. 1078 */ 1079 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1080 log(LOG_DEBUG, "%s; %s: Listen socket: " 1081 "Socket allocation failed due to " 1082 "limits or memory shortage, %s\n", 1083 s, __func__, 1084 V_tcp_sc_rst_sock_fail ? 1085 "sending RST" : "try again"); 1086 if (V_tcp_sc_rst_sock_fail) { 1087 rstreason = BANDLIM_UNLIMITED; 1088 goto dropwithreset; 1089 } else 1090 goto dropunlock; 1091 } 1092 /* 1093 * Socket is created in state SYN_RECEIVED. 1094 * Unlock the listen socket, lock the newly 1095 * created socket and update the tp variable. 1096 */ 1097 INP_WUNLOCK(inp); /* listen socket */ 1098 inp = sotoinpcb(so); 1099 INP_WLOCK(inp); /* new connection */ 1100 tp = intotcpcb(inp); 1101 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1102 ("%s: ", __func__)); 1103#ifdef TCP_SIGNATURE 1104 if (sig_checked == 0) { 1105 tcp_dooptions(&to, optp, optlen, 1106 (thflags & TH_SYN) ? TO_SYN : 0); 1107 if (!tcp_signature_verify_input(m, off0, tlen, 1108 optlen, &to, th, tp->t_flags)) { 1109 1110 /* 1111 * In SYN_SENT state if it receives an 1112 * RST, it is allowed for further 1113 * processing. 1114 */ 1115 if ((thflags & TH_RST) == 0 || 1116 (tp->t_state == TCPS_SYN_SENT) == 0) 1117 goto dropunlock; 1118 } 1119 sig_checked = 1; 1120 } 1121#endif 1122 1123 /* 1124 * Process the segment and the data it 1125 * contains. tcp_do_segment() consumes 1126 * the mbuf chain and unlocks the inpcb. 1127 */ 1128 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1129 iptos, ti_locked); 1130 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1131 return; 1132 } 1133 /* 1134 * Segment flag validation for new connection attempts: 1135 * 1136 * Our (SYN|ACK) response was rejected. 1137 * Check with syncache and remove entry to prevent 1138 * retransmits. 1139 * 1140 * NB: syncache_chkrst does its own logging of failure 1141 * causes. 1142 */ 1143 if (thflags & TH_RST) { 1144 syncache_chkrst(&inc, th); 1145 goto dropunlock; 1146 } 1147 /* 1148 * We can't do anything without SYN. 1149 */ 1150 if ((thflags & TH_SYN) == 0) { 1151 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1152 log(LOG_DEBUG, "%s; %s: Listen socket: " 1153 "SYN is missing, segment ignored\n", 1154 s, __func__); 1155 TCPSTAT_INC(tcps_badsyn); 1156 goto dropunlock; 1157 } 1158 /* 1159 * (SYN|ACK) is bogus on a listen socket. 1160 */ 1161 if (thflags & TH_ACK) { 1162 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1163 log(LOG_DEBUG, "%s; %s: Listen socket: " 1164 "SYN|ACK invalid, segment rejected\n", 1165 s, __func__); 1166 syncache_badack(&inc); /* XXX: Not needed! */ 1167 TCPSTAT_INC(tcps_badsyn); 1168 rstreason = BANDLIM_RST_OPENPORT; 1169 goto dropwithreset; 1170 } 1171 /* 1172 * If the drop_synfin option is enabled, drop all 1173 * segments with both the SYN and FIN bits set. 1174 * This prevents e.g. nmap from identifying the 1175 * TCP/IP stack. 1176 * XXX: Poor reasoning. nmap has other methods 1177 * and is constantly refining its stack detection 1178 * strategies. 1179 * XXX: This is a violation of the TCP specification 1180 * and was used by RFC1644. 1181 */ 1182 if ((thflags & TH_FIN) && V_drop_synfin) { 1183 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1184 log(LOG_DEBUG, "%s; %s: Listen socket: " 1185 "SYN|FIN segment ignored (based on " 1186 "sysctl setting)\n", s, __func__); 1187 TCPSTAT_INC(tcps_badsyn); 1188 goto dropunlock; 1189 } 1190 /* 1191 * Segment's flags are (SYN) or (SYN|FIN). 1192 * 1193 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1194 * as they do not affect the state of the TCP FSM. 1195 * The data pointed to by TH_URG and th_urp is ignored. 1196 */ 1197 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1198 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1199 KASSERT(thflags & (TH_SYN), 1200 ("%s: Listen socket: TH_SYN not set", __func__)); 1201#ifdef INET6 1202 /* 1203 * If deprecated address is forbidden, 1204 * we do not accept SYN to deprecated interface 1205 * address to prevent any new inbound connection from 1206 * getting established. 1207 * When we do not accept SYN, we send a TCP RST, 1208 * with deprecated source address (instead of dropping 1209 * it). We compromise it as it is much better for peer 1210 * to send a RST, and RST will be the final packet 1211 * for the exchange. 1212 * 1213 * If we do not forbid deprecated addresses, we accept 1214 * the SYN packet. RFC2462 does not suggest dropping 1215 * SYN in this case. 1216 * If we decipher RFC2462 5.5.4, it says like this: 1217 * 1. use of deprecated addr with existing 1218 * communication is okay - "SHOULD continue to be 1219 * used" 1220 * 2. use of it with new communication: 1221 * (2a) "SHOULD NOT be used if alternate address 1222 * with sufficient scope is available" 1223 * (2b) nothing mentioned otherwise. 1224 * Here we fall into (2b) case as we have no choice in 1225 * our source address selection - we must obey the peer. 1226 * 1227 * The wording in RFC2462 is confusing, and there are 1228 * multiple description text for deprecated address 1229 * handling - worse, they are not exactly the same. 1230 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1231 */ 1232 if (isipv6 && !V_ip6_use_deprecated) { 1233 struct in6_ifaddr *ia6; 1234 1235 ia6 = ip6_getdstifaddr(m); 1236 if (ia6 != NULL && 1237 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1238 ifa_free(&ia6->ia_ifa); 1239 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1240 log(LOG_DEBUG, "%s; %s: Listen socket: " 1241 "Connection attempt to deprecated " 1242 "IPv6 address rejected\n", 1243 s, __func__); 1244 rstreason = BANDLIM_RST_OPENPORT; 1245 goto dropwithreset; 1246 } 1247 if (ia6) 1248 ifa_free(&ia6->ia_ifa); 1249 } 1250#endif /* INET6 */ 1251 /* 1252 * Basic sanity checks on incoming SYN requests: 1253 * Don't respond if the destination is a link layer 1254 * broadcast according to RFC1122 4.2.3.10, p. 104. 1255 * If it is from this socket it must be forged. 1256 * Don't respond if the source or destination is a 1257 * global or subnet broad- or multicast address. 1258 * Note that it is quite possible to receive unicast 1259 * link-layer packets with a broadcast IP address. Use 1260 * in_broadcast() to find them. 1261 */ 1262 if (m->m_flags & (M_BCAST|M_MCAST)) { 1263 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1264 log(LOG_DEBUG, "%s; %s: Listen socket: " 1265 "Connection attempt from broad- or multicast " 1266 "link layer address ignored\n", s, __func__); 1267 goto dropunlock; 1268 } 1269#ifdef INET6 1270 if (isipv6) { 1271 if (th->th_dport == th->th_sport && 1272 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1273 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1274 log(LOG_DEBUG, "%s; %s: Listen socket: " 1275 "Connection attempt to/from self " 1276 "ignored\n", s, __func__); 1277 goto dropunlock; 1278 } 1279 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1280 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1281 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1282 log(LOG_DEBUG, "%s; %s: Listen socket: " 1283 "Connection attempt from/to multicast " 1284 "address ignored\n", s, __func__); 1285 goto dropunlock; 1286 } 1287 } 1288#endif 1289#if defined(INET) && defined(INET6) 1290 else 1291#endif 1292#ifdef INET 1293 { 1294 if (th->th_dport == th->th_sport && 1295 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1296 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1297 log(LOG_DEBUG, "%s; %s: Listen socket: " 1298 "Connection attempt from/to self " 1299 "ignored\n", s, __func__); 1300 goto dropunlock; 1301 } 1302 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1303 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1304 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1305 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1306 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1307 log(LOG_DEBUG, "%s; %s: Listen socket: " 1308 "Connection attempt from/to broad- " 1309 "or multicast address ignored\n", 1310 s, __func__); 1311 goto dropunlock; 1312 } 1313 } 1314#endif 1315 /* 1316 * SYN appears to be valid. Create compressed TCP state 1317 * for syncache. 1318 */ 1319#ifdef TCPDEBUG 1320 if (so->so_options & SO_DEBUG) 1321 tcp_trace(TA_INPUT, ostate, tp, 1322 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1323#endif 1324 tcp_dooptions(&to, optp, optlen, TO_SYN); 1325 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1326 /* 1327 * Entry added to syncache and mbuf consumed. 1328 * Everything already unlocked by syncache_add(). 1329 */ 1330 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1331 return; 1332 } 1333 1334#ifdef TCP_SIGNATURE 1335 if (sig_checked == 0) { 1336 tcp_dooptions(&to, optp, optlen, 1337 (thflags & TH_SYN) ? TO_SYN : 0); 1338 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1339 th, tp->t_flags)) { 1340 1341 /* 1342 * In SYN_SENT state if it receives an RST, it is 1343 * allowed for further processing. 1344 */ 1345 if ((thflags & TH_RST) == 0 || 1346 (tp->t_state == TCPS_SYN_SENT) == 0) 1347 goto dropunlock; 1348 } 1349 sig_checked = 1; 1350 } 1351#endif 1352 1353 /* 1354 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1355 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1356 * the inpcb, and unlocks pcbinfo. 1357 */ 1358 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1359 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1360 return; 1361 1362dropwithreset: 1363 if (ti_locked == TI_WLOCKED) { 1364 INP_INFO_WUNLOCK(&V_tcbinfo); 1365 ti_locked = TI_UNLOCKED; 1366 } 1367#ifdef INVARIANTS 1368 else { 1369 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1370 "ti_locked: %d", __func__, ti_locked)); 1371 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1372 } 1373#endif 1374 1375 if (inp != NULL) { 1376 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1377 INP_WUNLOCK(inp); 1378 } else 1379 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1380 m = NULL; /* mbuf chain got consumed. */ 1381 goto drop; 1382 1383dropunlock: 1384 if (ti_locked == TI_WLOCKED) { 1385 INP_INFO_WUNLOCK(&V_tcbinfo); 1386 ti_locked = TI_UNLOCKED; 1387 } 1388#ifdef INVARIANTS 1389 else { 1390 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1391 "ti_locked: %d", __func__, ti_locked)); 1392 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1393 } 1394#endif 1395 1396 if (inp != NULL) 1397 INP_WUNLOCK(inp); 1398 1399drop: 1400 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1401 if (s != NULL) 1402 free(s, M_TCPLOG); 1403 if (m != NULL) 1404 m_freem(m); 1405} 1406 1407static void 1408tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1409 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1410 int ti_locked) 1411{ 1412 int thflags, acked, ourfinisacked, needoutput = 0; 1413 int rstreason, todrop, win; 1414 u_long tiwin; 1415 struct tcpopt to; 1416 1417#ifdef TCPDEBUG 1418 /* 1419 * The size of tcp_saveipgen must be the size of the max ip header, 1420 * now IPv6. 1421 */ 1422 u_char tcp_saveipgen[IP6_HDR_LEN]; 1423 struct tcphdr tcp_savetcp; 1424 short ostate = 0; 1425#endif 1426 thflags = th->th_flags; 1427 tp->sackhint.last_sack_ack = 0; 1428 1429 /* 1430 * If this is either a state-changing packet or current state isn't 1431 * established, we require a write lock on tcbinfo. Otherwise, we 1432 * allow the tcbinfo to be in either alocked or unlocked, as the 1433 * caller may have unnecessarily acquired a write lock due to a race. 1434 */ 1435 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1436 tp->t_state != TCPS_ESTABLISHED) { 1437 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1438 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1439 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1440 } else { 1441#ifdef INVARIANTS 1442 if (ti_locked == TI_WLOCKED) 1443 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1444 else { 1445 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1446 "ti_locked: %d", __func__, ti_locked)); 1447 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1448 } 1449#endif 1450 } 1451 INP_WLOCK_ASSERT(tp->t_inpcb); 1452 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1453 __func__)); 1454 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1455 __func__)); 1456 1457 /* 1458 * Segment received on connection. 1459 * Reset idle time and keep-alive timer. 1460 * XXX: This should be done after segment 1461 * validation to ignore broken/spoofed segs. 1462 */ 1463 tp->t_rcvtime = ticks; 1464 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1465 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1466 1467 /* 1468 * Unscale the window into a 32-bit value. 1469 * For the SYN_SENT state the scale is zero. 1470 */ 1471 tiwin = th->th_win << tp->snd_scale; 1472 1473 /* 1474 * TCP ECN processing. 1475 */ 1476 if (tp->t_flags & TF_ECN_PERMIT) { 1477 if (thflags & TH_CWR) 1478 tp->t_flags &= ~TF_ECN_SND_ECE; 1479 switch (iptos & IPTOS_ECN_MASK) { 1480 case IPTOS_ECN_CE: 1481 tp->t_flags |= TF_ECN_SND_ECE; 1482 TCPSTAT_INC(tcps_ecn_ce); 1483 break; 1484 case IPTOS_ECN_ECT0: 1485 TCPSTAT_INC(tcps_ecn_ect0); 1486 break; 1487 case IPTOS_ECN_ECT1: 1488 TCPSTAT_INC(tcps_ecn_ect1); 1489 break; 1490 } 1491 /* Congestion experienced. */ 1492 if (thflags & TH_ECE) { 1493 cc_cong_signal(tp, th, CC_ECN); 1494 } 1495 } 1496 1497 /* 1498 * Parse options on any incoming segment. 1499 */ 1500 tcp_dooptions(&to, (u_char *)(th + 1), 1501 (th->th_off << 2) - sizeof(struct tcphdr), 1502 (thflags & TH_SYN) ? TO_SYN : 0); 1503 1504 /* 1505 * If echoed timestamp is later than the current time, 1506 * fall back to non RFC1323 RTT calculation. Normalize 1507 * timestamp if syncookies were used when this connection 1508 * was established. 1509 */ 1510 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1511 to.to_tsecr -= tp->ts_offset; 1512 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1513 to.to_tsecr = 0; 1514 } 1515 1516 /* 1517 * Process options only when we get SYN/ACK back. The SYN case 1518 * for incoming connections is handled in tcp_syncache. 1519 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1520 * or <SYN,ACK>) segment itself is never scaled. 1521 * XXX this is traditional behavior, may need to be cleaned up. 1522 */ 1523 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1524 if ((to.to_flags & TOF_SCALE) && 1525 (tp->t_flags & TF_REQ_SCALE)) { 1526 tp->t_flags |= TF_RCVD_SCALE; 1527 tp->snd_scale = to.to_wscale; 1528 } 1529 /* 1530 * Initial send window. It will be updated with 1531 * the next incoming segment to the scaled value. 1532 */ 1533 tp->snd_wnd = th->th_win; 1534 if (to.to_flags & TOF_TS) { 1535 tp->t_flags |= TF_RCVD_TSTMP; 1536 tp->ts_recent = to.to_tsval; 1537 tp->ts_recent_age = tcp_ts_getticks(); 1538 } 1539 if (to.to_flags & TOF_MSS) 1540 tcp_mss(tp, to.to_mss); 1541 if ((tp->t_flags & TF_SACK_PERMIT) && 1542 (to.to_flags & TOF_SACKPERM) == 0) 1543 tp->t_flags &= ~TF_SACK_PERMIT; 1544 } 1545 1546 /* 1547 * Header prediction: check for the two common cases 1548 * of a uni-directional data xfer. If the packet has 1549 * no control flags, is in-sequence, the window didn't 1550 * change and we're not retransmitting, it's a 1551 * candidate. If the length is zero and the ack moved 1552 * forward, we're the sender side of the xfer. Just 1553 * free the data acked & wake any higher level process 1554 * that was blocked waiting for space. If the length 1555 * is non-zero and the ack didn't move, we're the 1556 * receiver side. If we're getting packets in-order 1557 * (the reassembly queue is empty), add the data to 1558 * the socket buffer and note that we need a delayed ack. 1559 * Make sure that the hidden state-flags are also off. 1560 * Since we check for TCPS_ESTABLISHED first, it can only 1561 * be TH_NEEDSYN. 1562 */ 1563 if (tp->t_state == TCPS_ESTABLISHED && 1564 th->th_seq == tp->rcv_nxt && 1565 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1566 tp->snd_nxt == tp->snd_max && 1567 tiwin && tiwin == tp->snd_wnd && 1568 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1569 LIST_EMPTY(&tp->t_segq) && 1570 ((to.to_flags & TOF_TS) == 0 || 1571 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1572 1573 /* 1574 * If last ACK falls within this segment's sequence numbers, 1575 * record the timestamp. 1576 * NOTE that the test is modified according to the latest 1577 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1578 */ 1579 if ((to.to_flags & TOF_TS) != 0 && 1580 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1581 tp->ts_recent_age = tcp_ts_getticks(); 1582 tp->ts_recent = to.to_tsval; 1583 } 1584 1585 if (tlen == 0) { 1586 if (SEQ_GT(th->th_ack, tp->snd_una) && 1587 SEQ_LEQ(th->th_ack, tp->snd_max) && 1588 !IN_RECOVERY(tp->t_flags) && 1589 (to.to_flags & TOF_SACK) == 0 && 1590 TAILQ_EMPTY(&tp->snd_holes)) { 1591 /* 1592 * This is a pure ack for outstanding data. 1593 */ 1594 if (ti_locked == TI_WLOCKED) 1595 INP_INFO_WUNLOCK(&V_tcbinfo); 1596 ti_locked = TI_UNLOCKED; 1597 1598 TCPSTAT_INC(tcps_predack); 1599 1600 /* 1601 * "bad retransmit" recovery. 1602 */ 1603 if (tp->t_rxtshift == 1 && 1604 tp->t_flags & TF_PREVVALID && 1605 (int)(ticks - tp->t_badrxtwin) < 0) { 1606 cc_cong_signal(tp, th, CC_RTO_ERR); 1607 } 1608 1609 /* 1610 * Recalculate the transmit timer / rtt. 1611 * 1612 * Some boxes send broken timestamp replies 1613 * during the SYN+ACK phase, ignore 1614 * timestamps of 0 or we could calculate a 1615 * huge RTT and blow up the retransmit timer. 1616 */ 1617 if ((to.to_flags & TOF_TS) != 0 && 1618 to.to_tsecr) { 1619 u_int t; 1620 1621 t = tcp_ts_getticks() - to.to_tsecr; 1622 if (!tp->t_rttlow || tp->t_rttlow > t) 1623 tp->t_rttlow = t; 1624 tcp_xmit_timer(tp, 1625 TCP_TS_TO_TICKS(t) + 1); 1626 } else if (tp->t_rtttime && 1627 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1628 if (!tp->t_rttlow || 1629 tp->t_rttlow > ticks - tp->t_rtttime) 1630 tp->t_rttlow = ticks - tp->t_rtttime; 1631 tcp_xmit_timer(tp, 1632 ticks - tp->t_rtttime); 1633 } 1634 acked = BYTES_THIS_ACK(tp, th); 1635 1636 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1637 hhook_run_tcp_est_in(tp, th, &to); 1638 1639 TCPSTAT_INC(tcps_rcvackpack); 1640 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1641 sbdrop(&so->so_snd, acked); 1642 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1643 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1644 tp->snd_recover = th->th_ack - 1; 1645 1646 /* 1647 * Let the congestion control algorithm update 1648 * congestion control related information. This 1649 * typically means increasing the congestion 1650 * window. 1651 */ 1652 cc_ack_received(tp, th, CC_ACK); 1653 1654 tp->snd_una = th->th_ack; 1655 /* 1656 * Pull snd_wl2 up to prevent seq wrap relative 1657 * to th_ack. 1658 */ 1659 tp->snd_wl2 = th->th_ack; 1660 tp->t_dupacks = 0; 1661 m_freem(m); 1662 ND6_HINT(tp); /* Some progress has been made. */ 1663 1664 /* 1665 * If all outstanding data are acked, stop 1666 * retransmit timer, otherwise restart timer 1667 * using current (possibly backed-off) value. 1668 * If process is waiting for space, 1669 * wakeup/selwakeup/signal. If data 1670 * are ready to send, let tcp_output 1671 * decide between more output or persist. 1672 */ 1673#ifdef TCPDEBUG 1674 if (so->so_options & SO_DEBUG) 1675 tcp_trace(TA_INPUT, ostate, tp, 1676 (void *)tcp_saveipgen, 1677 &tcp_savetcp, 0); 1678#endif 1679 if (tp->snd_una == tp->snd_max) 1680 tcp_timer_activate(tp, TT_REXMT, 0); 1681 else if (!tcp_timer_active(tp, TT_PERSIST)) 1682 tcp_timer_activate(tp, TT_REXMT, 1683 tp->t_rxtcur); 1684 sowwakeup(so); 1685 if (so->so_snd.sb_cc) 1686 (void) tcp_output(tp); 1687 goto check_delack; 1688 } 1689 } else if (th->th_ack == tp->snd_una && 1690 tlen <= sbspace(&so->so_rcv)) { 1691 int newsize = 0; /* automatic sockbuf scaling */ 1692 1693 /* 1694 * This is a pure, in-sequence data packet with 1695 * nothing on the reassembly queue and we have enough 1696 * buffer space to take it. 1697 */ 1698 if (ti_locked == TI_WLOCKED) 1699 INP_INFO_WUNLOCK(&V_tcbinfo); 1700 ti_locked = TI_UNLOCKED; 1701 1702 /* Clean receiver SACK report if present */ 1703 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1704 tcp_clean_sackreport(tp); 1705 TCPSTAT_INC(tcps_preddat); 1706 tp->rcv_nxt += tlen; 1707 /* 1708 * Pull snd_wl1 up to prevent seq wrap relative to 1709 * th_seq. 1710 */ 1711 tp->snd_wl1 = th->th_seq; 1712 /* 1713 * Pull rcv_up up to prevent seq wrap relative to 1714 * rcv_nxt. 1715 */ 1716 tp->rcv_up = tp->rcv_nxt; 1717 TCPSTAT_INC(tcps_rcvpack); 1718 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1719 ND6_HINT(tp); /* Some progress has been made */ 1720#ifdef TCPDEBUG 1721 if (so->so_options & SO_DEBUG) 1722 tcp_trace(TA_INPUT, ostate, tp, 1723 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1724#endif 1725 /* 1726 * Automatic sizing of receive socket buffer. Often the send 1727 * buffer size is not optimally adjusted to the actual network 1728 * conditions at hand (delay bandwidth product). Setting the 1729 * buffer size too small limits throughput on links with high 1730 * bandwidth and high delay (eg. trans-continental/oceanic links). 1731 * 1732 * On the receive side the socket buffer memory is only rarely 1733 * used to any significant extent. This allows us to be much 1734 * more aggressive in scaling the receive socket buffer. For 1735 * the case that the buffer space is actually used to a large 1736 * extent and we run out of kernel memory we can simply drop 1737 * the new segments; TCP on the sender will just retransmit it 1738 * later. Setting the buffer size too big may only consume too 1739 * much kernel memory if the application doesn't read() from 1740 * the socket or packet loss or reordering makes use of the 1741 * reassembly queue. 1742 * 1743 * The criteria to step up the receive buffer one notch are: 1744 * 1. the number of bytes received during the time it takes 1745 * one timestamp to be reflected back to us (the RTT); 1746 * 2. received bytes per RTT is within seven eighth of the 1747 * current socket buffer size; 1748 * 3. receive buffer size has not hit maximal automatic size; 1749 * 1750 * This algorithm does one step per RTT at most and only if 1751 * we receive a bulk stream w/o packet losses or reorderings. 1752 * Shrinking the buffer during idle times is not necessary as 1753 * it doesn't consume any memory when idle. 1754 * 1755 * TODO: Only step up if the application is actually serving 1756 * the buffer to better manage the socket buffer resources. 1757 */ 1758 if (V_tcp_do_autorcvbuf && 1759 to.to_tsecr && 1760 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1761 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1762 to.to_tsecr - tp->rfbuf_ts < hz) { 1763 if (tp->rfbuf_cnt > 1764 (so->so_rcv.sb_hiwat / 8 * 7) && 1765 so->so_rcv.sb_hiwat < 1766 V_tcp_autorcvbuf_max) { 1767 newsize = 1768 min(so->so_rcv.sb_hiwat + 1769 V_tcp_autorcvbuf_inc, 1770 V_tcp_autorcvbuf_max); 1771 } 1772 /* Start over with next RTT. */ 1773 tp->rfbuf_ts = 0; 1774 tp->rfbuf_cnt = 0; 1775 } else 1776 tp->rfbuf_cnt += tlen; /* add up */ 1777 } 1778 1779 /* Add data to socket buffer. */ 1780 SOCKBUF_LOCK(&so->so_rcv); 1781 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1782 m_freem(m); 1783 } else { 1784 /* 1785 * Set new socket buffer size. 1786 * Give up when limit is reached. 1787 */ 1788 if (newsize) 1789 if (!sbreserve_locked(&so->so_rcv, 1790 newsize, so, NULL)) 1791 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1792 m_adj(m, drop_hdrlen); /* delayed header drop */ 1793 sbappendstream_locked(&so->so_rcv, m); 1794 } 1795 /* NB: sorwakeup_locked() does an implicit unlock. */ 1796 sorwakeup_locked(so); 1797 if (DELAY_ACK(tp)) { 1798 tp->t_flags |= TF_DELACK; 1799 } else { 1800 tp->t_flags |= TF_ACKNOW; 1801 tcp_output(tp); 1802 } 1803 goto check_delack; 1804 } 1805 } 1806 1807 /* 1808 * Calculate amount of space in receive window, 1809 * and then do TCP input processing. 1810 * Receive window is amount of space in rcv queue, 1811 * but not less than advertised window. 1812 */ 1813 win = sbspace(&so->so_rcv); 1814 if (win < 0) 1815 win = 0; 1816 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1817 1818 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1819 tp->rfbuf_ts = 0; 1820 tp->rfbuf_cnt = 0; 1821 1822 switch (tp->t_state) { 1823 1824 /* 1825 * If the state is SYN_RECEIVED: 1826 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1827 */ 1828 case TCPS_SYN_RECEIVED: 1829 if ((thflags & TH_ACK) && 1830 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1831 SEQ_GT(th->th_ack, tp->snd_max))) { 1832 rstreason = BANDLIM_RST_OPENPORT; 1833 goto dropwithreset; 1834 } 1835 break; 1836 1837 /* 1838 * If the state is SYN_SENT: 1839 * if seg contains an ACK, but not for our SYN, drop the input. 1840 * if seg contains a RST, then drop the connection. 1841 * if seg does not contain SYN, then drop it. 1842 * Otherwise this is an acceptable SYN segment 1843 * initialize tp->rcv_nxt and tp->irs 1844 * if seg contains ack then advance tp->snd_una 1845 * if seg contains an ECE and ECN support is enabled, the stream 1846 * is ECN capable. 1847 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1848 * arrange for segment to be acked (eventually) 1849 * continue processing rest of data/controls, beginning with URG 1850 */ 1851 case TCPS_SYN_SENT: 1852 if ((thflags & TH_ACK) && 1853 (SEQ_LEQ(th->th_ack, tp->iss) || 1854 SEQ_GT(th->th_ack, tp->snd_max))) { 1855 rstreason = BANDLIM_UNLIMITED; 1856 goto dropwithreset; 1857 } 1858 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1859 tp = tcp_drop(tp, ECONNREFUSED); 1860 if (thflags & TH_RST) 1861 goto drop; 1862 if (!(thflags & TH_SYN)) 1863 goto drop; 1864 1865 tp->irs = th->th_seq; 1866 tcp_rcvseqinit(tp); 1867 if (thflags & TH_ACK) { 1868 TCPSTAT_INC(tcps_connects); 1869 soisconnected(so); 1870#ifdef MAC 1871 mac_socketpeer_set_from_mbuf(m, so); 1872#endif 1873 /* Do window scaling on this connection? */ 1874 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1875 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1876 tp->rcv_scale = tp->request_r_scale; 1877 } 1878 tp->rcv_adv += imin(tp->rcv_wnd, 1879 TCP_MAXWIN << tp->rcv_scale); 1880 tp->snd_una++; /* SYN is acked */ 1881 /* 1882 * If there's data, delay ACK; if there's also a FIN 1883 * ACKNOW will be turned on later. 1884 */ 1885 if (DELAY_ACK(tp) && tlen != 0) 1886 tcp_timer_activate(tp, TT_DELACK, 1887 tcp_delacktime); 1888 else 1889 tp->t_flags |= TF_ACKNOW; 1890 1891 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1892 tp->t_flags |= TF_ECN_PERMIT; 1893 TCPSTAT_INC(tcps_ecn_shs); 1894 } 1895 1896 /* 1897 * Received <SYN,ACK> in SYN_SENT[*] state. 1898 * Transitions: 1899 * SYN_SENT --> ESTABLISHED 1900 * SYN_SENT* --> FIN_WAIT_1 1901 */ 1902 tp->t_starttime = ticks; 1903 if (tp->t_flags & TF_NEEDFIN) { 1904 tp->t_state = TCPS_FIN_WAIT_1; 1905 tp->t_flags &= ~TF_NEEDFIN; 1906 thflags &= ~TH_SYN; 1907 } else { 1908 tp->t_state = TCPS_ESTABLISHED; 1909 cc_conn_init(tp); 1910 tcp_timer_activate(tp, TT_KEEP, 1911 TP_KEEPIDLE(tp)); 1912 } 1913 } else { 1914 /* 1915 * Received initial SYN in SYN-SENT[*] state => 1916 * simultaneous open. If segment contains CC option 1917 * and there is a cached CC, apply TAO test. 1918 * If it succeeds, connection is * half-synchronized. 1919 * Otherwise, do 3-way handshake: 1920 * SYN-SENT -> SYN-RECEIVED 1921 * SYN-SENT* -> SYN-RECEIVED* 1922 * If there was no CC option, clear cached CC value. 1923 */ 1924 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1925 tcp_timer_activate(tp, TT_REXMT, 0); 1926 tp->t_state = TCPS_SYN_RECEIVED; 1927 } 1928 1929 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1930 "ti_locked %d", __func__, ti_locked)); 1931 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1932 INP_WLOCK_ASSERT(tp->t_inpcb); 1933 1934 /* 1935 * Advance th->th_seq to correspond to first data byte. 1936 * If data, trim to stay within window, 1937 * dropping FIN if necessary. 1938 */ 1939 th->th_seq++; 1940 if (tlen > tp->rcv_wnd) { 1941 todrop = tlen - tp->rcv_wnd; 1942 m_adj(m, -todrop); 1943 tlen = tp->rcv_wnd; 1944 thflags &= ~TH_FIN; 1945 TCPSTAT_INC(tcps_rcvpackafterwin); 1946 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1947 } 1948 tp->snd_wl1 = th->th_seq - 1; 1949 tp->rcv_up = th->th_seq; 1950 /* 1951 * Client side of transaction: already sent SYN and data. 1952 * If the remote host used T/TCP to validate the SYN, 1953 * our data will be ACK'd; if so, enter normal data segment 1954 * processing in the middle of step 5, ack processing. 1955 * Otherwise, goto step 6. 1956 */ 1957 if (thflags & TH_ACK) 1958 goto process_ACK; 1959 1960 goto step6; 1961 1962 /* 1963 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1964 * do normal processing. 1965 * 1966 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1967 */ 1968 case TCPS_LAST_ACK: 1969 case TCPS_CLOSING: 1970 break; /* continue normal processing */ 1971 } 1972 1973 /* 1974 * States other than LISTEN or SYN_SENT. 1975 * First check the RST flag and sequence number since reset segments 1976 * are exempt from the timestamp and connection count tests. This 1977 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1978 * below which allowed reset segments in half the sequence space 1979 * to fall though and be processed (which gives forged reset 1980 * segments with a random sequence number a 50 percent chance of 1981 * killing a connection). 1982 * Then check timestamp, if present. 1983 * Then check the connection count, if present. 1984 * Then check that at least some bytes of segment are within 1985 * receive window. If segment begins before rcv_nxt, 1986 * drop leading data (and SYN); if nothing left, just ack. 1987 * 1988 * 1989 * If the RST bit is set, check the sequence number to see 1990 * if this is a valid reset segment. 1991 * RFC 793 page 37: 1992 * In all states except SYN-SENT, all reset (RST) segments 1993 * are validated by checking their SEQ-fields. A reset is 1994 * valid if its sequence number is in the window. 1995 * Note: this does not take into account delayed ACKs, so 1996 * we should test against last_ack_sent instead of rcv_nxt. 1997 * The sequence number in the reset segment is normally an 1998 * echo of our outgoing acknowlegement numbers, but some hosts 1999 * send a reset with the sequence number at the rightmost edge 2000 * of our receive window, and we have to handle this case. 2001 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2002 * that brute force RST attacks are possible. To combat this, 2003 * we use a much stricter check while in the ESTABLISHED state, 2004 * only accepting RSTs where the sequence number is equal to 2005 * last_ack_sent. In all other states (the states in which a 2006 * RST is more likely), the more permissive check is used. 2007 * If we have multiple segments in flight, the initial reset 2008 * segment sequence numbers will be to the left of last_ack_sent, 2009 * but they will eventually catch up. 2010 * In any case, it never made sense to trim reset segments to 2011 * fit the receive window since RFC 1122 says: 2012 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2013 * 2014 * A TCP SHOULD allow a received RST segment to include data. 2015 * 2016 * DISCUSSION 2017 * It has been suggested that a RST segment could contain 2018 * ASCII text that encoded and explained the cause of the 2019 * RST. No standard has yet been established for such 2020 * data. 2021 * 2022 * If the reset segment passes the sequence number test examine 2023 * the state: 2024 * SYN_RECEIVED STATE: 2025 * If passive open, return to LISTEN state. 2026 * If active open, inform user that connection was refused. 2027 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2028 * Inform user that connection was reset, and close tcb. 2029 * CLOSING, LAST_ACK STATES: 2030 * Close the tcb. 2031 * TIME_WAIT STATE: 2032 * Drop the segment - see Stevens, vol. 2, p. 964 and 2033 * RFC 1337. 2034 */ 2035 if (thflags & TH_RST) { 2036 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2037 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2038 switch (tp->t_state) { 2039 2040 case TCPS_SYN_RECEIVED: 2041 so->so_error = ECONNREFUSED; 2042 goto close; 2043 2044 case TCPS_ESTABLISHED: 2045 if (V_tcp_insecure_rst == 0 && 2046 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2047 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2048 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2049 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2050 TCPSTAT_INC(tcps_badrst); 2051 goto drop; 2052 } 2053 /* FALLTHROUGH */ 2054 case TCPS_FIN_WAIT_1: 2055 case TCPS_FIN_WAIT_2: 2056 case TCPS_CLOSE_WAIT: 2057 so->so_error = ECONNRESET; 2058 close: 2059 KASSERT(ti_locked == TI_WLOCKED, 2060 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2061 ti_locked)); 2062 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2063 2064 tp->t_state = TCPS_CLOSED; 2065 TCPSTAT_INC(tcps_drops); 2066 tp = tcp_close(tp); 2067 break; 2068 2069 case TCPS_CLOSING: 2070 case TCPS_LAST_ACK: 2071 KASSERT(ti_locked == TI_WLOCKED, 2072 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2073 ti_locked)); 2074 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2075 2076 tp = tcp_close(tp); 2077 break; 2078 } 2079 } 2080 goto drop; 2081 } 2082 2083 /* 2084 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2085 * and it's less than ts_recent, drop it. 2086 */ 2087 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2088 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2089 2090 /* Check to see if ts_recent is over 24 days old. */ 2091 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2092 /* 2093 * Invalidate ts_recent. If this segment updates 2094 * ts_recent, the age will be reset later and ts_recent 2095 * will get a valid value. If it does not, setting 2096 * ts_recent to zero will at least satisfy the 2097 * requirement that zero be placed in the timestamp 2098 * echo reply when ts_recent isn't valid. The 2099 * age isn't reset until we get a valid ts_recent 2100 * because we don't want out-of-order segments to be 2101 * dropped when ts_recent is old. 2102 */ 2103 tp->ts_recent = 0; 2104 } else { 2105 TCPSTAT_INC(tcps_rcvduppack); 2106 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2107 TCPSTAT_INC(tcps_pawsdrop); 2108 if (tlen) 2109 goto dropafterack; 2110 goto drop; 2111 } 2112 } 2113 2114 /* 2115 * In the SYN-RECEIVED state, validate that the packet belongs to 2116 * this connection before trimming the data to fit the receive 2117 * window. Check the sequence number versus IRS since we know 2118 * the sequence numbers haven't wrapped. This is a partial fix 2119 * for the "LAND" DoS attack. 2120 */ 2121 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2122 rstreason = BANDLIM_RST_OPENPORT; 2123 goto dropwithreset; 2124 } 2125 2126 todrop = tp->rcv_nxt - th->th_seq; 2127 if (todrop > 0) { 2128 /* 2129 * If this is a duplicate SYN for our current connection, 2130 * advance over it and pretend and it's not a SYN. 2131 */ 2132 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2133 thflags &= ~TH_SYN; 2134 th->th_seq++; 2135 if (th->th_urp > 1) 2136 th->th_urp--; 2137 else 2138 thflags &= ~TH_URG; 2139 todrop--; 2140 } 2141 /* 2142 * Following if statement from Stevens, vol. 2, p. 960. 2143 */ 2144 if (todrop > tlen 2145 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2146 /* 2147 * Any valid FIN must be to the left of the window. 2148 * At this point the FIN must be a duplicate or out 2149 * of sequence; drop it. 2150 */ 2151 thflags &= ~TH_FIN; 2152 2153 /* 2154 * Send an ACK to resynchronize and drop any data. 2155 * But keep on processing for RST or ACK. 2156 */ 2157 tp->t_flags |= TF_ACKNOW; 2158 todrop = tlen; 2159 TCPSTAT_INC(tcps_rcvduppack); 2160 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2161 } else { 2162 TCPSTAT_INC(tcps_rcvpartduppack); 2163 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2164 } 2165 drop_hdrlen += todrop; /* drop from the top afterwards */ 2166 th->th_seq += todrop; 2167 tlen -= todrop; 2168 if (th->th_urp > todrop) 2169 th->th_urp -= todrop; 2170 else { 2171 thflags &= ~TH_URG; 2172 th->th_urp = 0; 2173 } 2174 } 2175 2176 /* 2177 * If new data are received on a connection after the 2178 * user processes are gone, then RST the other end. 2179 */ 2180 if ((so->so_state & SS_NOFDREF) && 2181 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2182 char *s; 2183 2184 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2185 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2186 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2187 2188 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 2189 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 2190 "was closed, sending RST and removing tcpcb\n", 2191 s, __func__, tcpstates[tp->t_state], tlen); 2192 free(s, M_TCPLOG); 2193 } 2194 tp = tcp_close(tp); 2195 TCPSTAT_INC(tcps_rcvafterclose); 2196 rstreason = BANDLIM_UNLIMITED; 2197 goto dropwithreset; 2198 } 2199 2200 /* 2201 * If segment ends after window, drop trailing data 2202 * (and PUSH and FIN); if nothing left, just ACK. 2203 */ 2204 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2205 if (todrop > 0) { 2206 TCPSTAT_INC(tcps_rcvpackafterwin); 2207 if (todrop >= tlen) { 2208 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2209 /* 2210 * If window is closed can only take segments at 2211 * window edge, and have to drop data and PUSH from 2212 * incoming segments. Continue processing, but 2213 * remember to ack. Otherwise, drop segment 2214 * and ack. 2215 */ 2216 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2217 tp->t_flags |= TF_ACKNOW; 2218 TCPSTAT_INC(tcps_rcvwinprobe); 2219 } else 2220 goto dropafterack; 2221 } else 2222 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2223 m_adj(m, -todrop); 2224 tlen -= todrop; 2225 thflags &= ~(TH_PUSH|TH_FIN); 2226 } 2227 2228 /* 2229 * If last ACK falls within this segment's sequence numbers, 2230 * record its timestamp. 2231 * NOTE: 2232 * 1) That the test incorporates suggestions from the latest 2233 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2234 * 2) That updating only on newer timestamps interferes with 2235 * our earlier PAWS tests, so this check should be solely 2236 * predicated on the sequence space of this segment. 2237 * 3) That we modify the segment boundary check to be 2238 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2239 * instead of RFC1323's 2240 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2241 * This modified check allows us to overcome RFC1323's 2242 * limitations as described in Stevens TCP/IP Illustrated 2243 * Vol. 2 p.869. In such cases, we can still calculate the 2244 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2245 */ 2246 if ((to.to_flags & TOF_TS) != 0 && 2247 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2248 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2249 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2250 tp->ts_recent_age = tcp_ts_getticks(); 2251 tp->ts_recent = to.to_tsval; 2252 } 2253 2254 /* 2255 * If a SYN is in the window, then this is an 2256 * error and we send an RST and drop the connection. 2257 */ 2258 if (thflags & TH_SYN) { 2259 KASSERT(ti_locked == TI_WLOCKED, 2260 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2261 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2262 2263 tp = tcp_drop(tp, ECONNRESET); 2264 rstreason = BANDLIM_UNLIMITED; 2265 goto drop; 2266 } 2267 2268 /* 2269 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2270 * flag is on (half-synchronized state), then queue data for 2271 * later processing; else drop segment and return. 2272 */ 2273 if ((thflags & TH_ACK) == 0) { 2274 if (tp->t_state == TCPS_SYN_RECEIVED || 2275 (tp->t_flags & TF_NEEDSYN)) 2276 goto step6; 2277 else if (tp->t_flags & TF_ACKNOW) 2278 goto dropafterack; 2279 else 2280 goto drop; 2281 } 2282 2283 /* 2284 * Ack processing. 2285 */ 2286 switch (tp->t_state) { 2287 2288 /* 2289 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2290 * ESTABLISHED state and continue processing. 2291 * The ACK was checked above. 2292 */ 2293 case TCPS_SYN_RECEIVED: 2294 2295 TCPSTAT_INC(tcps_connects); 2296 soisconnected(so); 2297 /* Do window scaling? */ 2298 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2299 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2300 tp->rcv_scale = tp->request_r_scale; 2301 tp->snd_wnd = tiwin; 2302 } 2303 /* 2304 * Make transitions: 2305 * SYN-RECEIVED -> ESTABLISHED 2306 * SYN-RECEIVED* -> FIN-WAIT-1 2307 */ 2308 tp->t_starttime = ticks; 2309 if (tp->t_flags & TF_NEEDFIN) { 2310 tp->t_state = TCPS_FIN_WAIT_1; 2311 tp->t_flags &= ~TF_NEEDFIN; 2312 } else { 2313 tp->t_state = TCPS_ESTABLISHED; 2314 cc_conn_init(tp); 2315 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2316 } 2317 /* 2318 * If segment contains data or ACK, will call tcp_reass() 2319 * later; if not, do so now to pass queued data to user. 2320 */ 2321 if (tlen == 0 && (thflags & TH_FIN) == 0) 2322 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2323 (struct mbuf *)0); 2324 tp->snd_wl1 = th->th_seq - 1; 2325 /* FALLTHROUGH */ 2326 2327 /* 2328 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2329 * ACKs. If the ack is in the range 2330 * tp->snd_una < th->th_ack <= tp->snd_max 2331 * then advance tp->snd_una to th->th_ack and drop 2332 * data from the retransmission queue. If this ACK reflects 2333 * more up to date window information we update our window information. 2334 */ 2335 case TCPS_ESTABLISHED: 2336 case TCPS_FIN_WAIT_1: 2337 case TCPS_FIN_WAIT_2: 2338 case TCPS_CLOSE_WAIT: 2339 case TCPS_CLOSING: 2340 case TCPS_LAST_ACK: 2341 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2342 TCPSTAT_INC(tcps_rcvacktoomuch); 2343 goto dropafterack; 2344 } 2345 if ((tp->t_flags & TF_SACK_PERMIT) && 2346 ((to.to_flags & TOF_SACK) || 2347 !TAILQ_EMPTY(&tp->snd_holes))) 2348 tcp_sack_doack(tp, &to, th->th_ack); 2349 2350 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2351 hhook_run_tcp_est_in(tp, th, &to); 2352 2353 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2354 if (tlen == 0 && tiwin == tp->snd_wnd) { 2355 TCPSTAT_INC(tcps_rcvdupack); 2356 /* 2357 * If we have outstanding data (other than 2358 * a window probe), this is a completely 2359 * duplicate ack (ie, window info didn't 2360 * change), the ack is the biggest we've 2361 * seen and we've seen exactly our rexmt 2362 * threshhold of them, assume a packet 2363 * has been dropped and retransmit it. 2364 * Kludge snd_nxt & the congestion 2365 * window so we send only this one 2366 * packet. 2367 * 2368 * We know we're losing at the current 2369 * window size so do congestion avoidance 2370 * (set ssthresh to half the current window 2371 * and pull our congestion window back to 2372 * the new ssthresh). 2373 * 2374 * Dup acks mean that packets have left the 2375 * network (they're now cached at the receiver) 2376 * so bump cwnd by the amount in the receiver 2377 * to keep a constant cwnd packets in the 2378 * network. 2379 * 2380 * When using TCP ECN, notify the peer that 2381 * we reduced the cwnd. 2382 */ 2383 if (!tcp_timer_active(tp, TT_REXMT) || 2384 th->th_ack != tp->snd_una) 2385 tp->t_dupacks = 0; 2386 else if (++tp->t_dupacks > tcprexmtthresh || 2387 IN_FASTRECOVERY(tp->t_flags)) { 2388 cc_ack_received(tp, th, CC_DUPACK); 2389 if ((tp->t_flags & TF_SACK_PERMIT) && 2390 IN_FASTRECOVERY(tp->t_flags)) { 2391 int awnd; 2392 2393 /* 2394 * Compute the amount of data in flight first. 2395 * We can inject new data into the pipe iff 2396 * we have less than 1/2 the original window's 2397 * worth of data in flight. 2398 */ 2399 awnd = (tp->snd_nxt - tp->snd_fack) + 2400 tp->sackhint.sack_bytes_rexmit; 2401 if (awnd < tp->snd_ssthresh) { 2402 tp->snd_cwnd += tp->t_maxseg; 2403 if (tp->snd_cwnd > tp->snd_ssthresh) 2404 tp->snd_cwnd = tp->snd_ssthresh; 2405 } 2406 } else 2407 tp->snd_cwnd += tp->t_maxseg; 2408 if ((thflags & TH_FIN) && 2409 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2410 /* 2411 * If its a fin we need to process 2412 * it to avoid a race where both 2413 * sides enter FIN-WAIT and send FIN|ACK 2414 * at the same time. 2415 */ 2416 break; 2417 } 2418 (void) tcp_output(tp); 2419 goto drop; 2420 } else if (tp->t_dupacks == tcprexmtthresh) { 2421 tcp_seq onxt = tp->snd_nxt; 2422 2423 /* 2424 * If we're doing sack, check to 2425 * see if we're already in sack 2426 * recovery. If we're not doing sack, 2427 * check to see if we're in newreno 2428 * recovery. 2429 */ 2430 if (tp->t_flags & TF_SACK_PERMIT) { 2431 if (IN_FASTRECOVERY(tp->t_flags)) { 2432 tp->t_dupacks = 0; 2433 break; 2434 } 2435 } else { 2436 if (SEQ_LEQ(th->th_ack, 2437 tp->snd_recover)) { 2438 tp->t_dupacks = 0; 2439 break; 2440 } 2441 } 2442 /* Congestion signal before ack. */ 2443 cc_cong_signal(tp, th, CC_NDUPACK); 2444 cc_ack_received(tp, th, CC_DUPACK); 2445 tcp_timer_activate(tp, TT_REXMT, 0); 2446 tp->t_rtttime = 0; 2447 if (tp->t_flags & TF_SACK_PERMIT) { 2448 TCPSTAT_INC( 2449 tcps_sack_recovery_episode); 2450 tp->sack_newdata = tp->snd_nxt; 2451 tp->snd_cwnd = tp->t_maxseg; 2452 (void) tcp_output(tp); 2453 goto drop; 2454 } 2455 tp->snd_nxt = th->th_ack; 2456 tp->snd_cwnd = tp->t_maxseg; 2457 if ((thflags & TH_FIN) && 2458 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2459 /* 2460 * If its a fin we need to process 2461 * it to avoid a race where both 2462 * sides enter FIN-WAIT and send FIN|ACK 2463 * at the same time. 2464 */ 2465 break; 2466 } 2467 (void) tcp_output(tp); 2468 KASSERT(tp->snd_limited <= 2, 2469 ("%s: tp->snd_limited too big", 2470 __func__)); 2471 tp->snd_cwnd = tp->snd_ssthresh + 2472 tp->t_maxseg * 2473 (tp->t_dupacks - tp->snd_limited); 2474 if (SEQ_GT(onxt, tp->snd_nxt)) 2475 tp->snd_nxt = onxt; 2476 goto drop; 2477 } else if (V_tcp_do_rfc3042) { 2478 cc_ack_received(tp, th, CC_DUPACK); 2479 u_long oldcwnd = tp->snd_cwnd; 2480 tcp_seq oldsndmax = tp->snd_max; 2481 u_int sent; 2482 2483 KASSERT(tp->t_dupacks == 1 || 2484 tp->t_dupacks == 2, 2485 ("%s: dupacks not 1 or 2", 2486 __func__)); 2487 if (tp->t_dupacks == 1) 2488 tp->snd_limited = 0; 2489 tp->snd_cwnd = 2490 (tp->snd_nxt - tp->snd_una) + 2491 (tp->t_dupacks - tp->snd_limited) * 2492 tp->t_maxseg; 2493 if ((thflags & TH_FIN) && 2494 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2495 /* 2496 * If its a fin we need to process 2497 * it to avoid a race where both 2498 * sides enter FIN-WAIT and send FIN|ACK 2499 * at the same time. 2500 */ 2501 break; 2502 } 2503 (void) tcp_output(tp); 2504 sent = tp->snd_max - oldsndmax; 2505 if (sent > tp->t_maxseg) { 2506 KASSERT((tp->t_dupacks == 2 && 2507 tp->snd_limited == 0) || 2508 (sent == tp->t_maxseg + 1 && 2509 tp->t_flags & TF_SENTFIN), 2510 ("%s: sent too much", 2511 __func__)); 2512 tp->snd_limited = 2; 2513 } else if (sent > 0) 2514 ++tp->snd_limited; 2515 tp->snd_cwnd = oldcwnd; 2516 goto drop; 2517 } 2518 } else 2519 tp->t_dupacks = 0; 2520 break; 2521 } 2522 2523 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2524 ("%s: th_ack <= snd_una", __func__)); 2525 2526 /* 2527 * If the congestion window was inflated to account 2528 * for the other side's cached packets, retract it. 2529 */ 2530 if (IN_FASTRECOVERY(tp->t_flags)) { 2531 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2532 if (tp->t_flags & TF_SACK_PERMIT) 2533 tcp_sack_partialack(tp, th); 2534 else 2535 tcp_newreno_partial_ack(tp, th); 2536 } else 2537 cc_post_recovery(tp, th); 2538 } 2539 tp->t_dupacks = 0; 2540 /* 2541 * If we reach this point, ACK is not a duplicate, 2542 * i.e., it ACKs something we sent. 2543 */ 2544 if (tp->t_flags & TF_NEEDSYN) { 2545 /* 2546 * T/TCP: Connection was half-synchronized, and our 2547 * SYN has been ACK'd (so connection is now fully 2548 * synchronized). Go to non-starred state, 2549 * increment snd_una for ACK of SYN, and check if 2550 * we can do window scaling. 2551 */ 2552 tp->t_flags &= ~TF_NEEDSYN; 2553 tp->snd_una++; 2554 /* Do window scaling? */ 2555 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2556 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2557 tp->rcv_scale = tp->request_r_scale; 2558 /* Send window already scaled. */ 2559 } 2560 } 2561 2562process_ACK: 2563 INP_WLOCK_ASSERT(tp->t_inpcb); 2564 2565 acked = BYTES_THIS_ACK(tp, th); 2566 TCPSTAT_INC(tcps_rcvackpack); 2567 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2568 2569 /* 2570 * If we just performed our first retransmit, and the ACK 2571 * arrives within our recovery window, then it was a mistake 2572 * to do the retransmit in the first place. Recover our 2573 * original cwnd and ssthresh, and proceed to transmit where 2574 * we left off. 2575 */ 2576 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2577 (int)(ticks - tp->t_badrxtwin) < 0) 2578 cc_cong_signal(tp, th, CC_RTO_ERR); 2579 2580 /* 2581 * If we have a timestamp reply, update smoothed 2582 * round trip time. If no timestamp is present but 2583 * transmit timer is running and timed sequence 2584 * number was acked, update smoothed round trip time. 2585 * Since we now have an rtt measurement, cancel the 2586 * timer backoff (cf., Phil Karn's retransmit alg.). 2587 * Recompute the initial retransmit timer. 2588 * 2589 * Some boxes send broken timestamp replies 2590 * during the SYN+ACK phase, ignore 2591 * timestamps of 0 or we could calculate a 2592 * huge RTT and blow up the retransmit timer. 2593 */ 2594 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2595 u_int t; 2596 2597 t = tcp_ts_getticks() - to.to_tsecr; 2598 if (!tp->t_rttlow || tp->t_rttlow > t) 2599 tp->t_rttlow = t; 2600 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2601 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2602 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2603 tp->t_rttlow = ticks - tp->t_rtttime; 2604 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2605 } 2606 2607 /* 2608 * If all outstanding data is acked, stop retransmit 2609 * timer and remember to restart (more output or persist). 2610 * If there is more data to be acked, restart retransmit 2611 * timer, using current (possibly backed-off) value. 2612 */ 2613 if (th->th_ack == tp->snd_max) { 2614 tcp_timer_activate(tp, TT_REXMT, 0); 2615 needoutput = 1; 2616 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2617 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2618 2619 /* 2620 * If no data (only SYN) was ACK'd, 2621 * skip rest of ACK processing. 2622 */ 2623 if (acked == 0) 2624 goto step6; 2625 2626 /* 2627 * Let the congestion control algorithm update congestion 2628 * control related information. This typically means increasing 2629 * the congestion window. 2630 */ 2631 cc_ack_received(tp, th, CC_ACK); 2632 2633 SOCKBUF_LOCK(&so->so_snd); 2634 if (acked > so->so_snd.sb_cc) { 2635 tp->snd_wnd -= so->so_snd.sb_cc; 2636 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2637 ourfinisacked = 1; 2638 } else { 2639 sbdrop_locked(&so->so_snd, acked); 2640 tp->snd_wnd -= acked; 2641 ourfinisacked = 0; 2642 } 2643 /* NB: sowwakeup_locked() does an implicit unlock. */ 2644 sowwakeup_locked(so); 2645 /* Detect una wraparound. */ 2646 if (!IN_RECOVERY(tp->t_flags) && 2647 SEQ_GT(tp->snd_una, tp->snd_recover) && 2648 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2649 tp->snd_recover = th->th_ack - 1; 2650 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2651 if (IN_RECOVERY(tp->t_flags) && 2652 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2653 EXIT_RECOVERY(tp->t_flags); 2654 } 2655 tp->snd_una = th->th_ack; 2656 if (tp->t_flags & TF_SACK_PERMIT) { 2657 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2658 tp->snd_recover = tp->snd_una; 2659 } 2660 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2661 tp->snd_nxt = tp->snd_una; 2662 2663 switch (tp->t_state) { 2664 2665 /* 2666 * In FIN_WAIT_1 STATE in addition to the processing 2667 * for the ESTABLISHED state if our FIN is now acknowledged 2668 * then enter FIN_WAIT_2. 2669 */ 2670 case TCPS_FIN_WAIT_1: 2671 if (ourfinisacked) { 2672 /* 2673 * If we can't receive any more 2674 * data, then closing user can proceed. 2675 * Starting the timer is contrary to the 2676 * specification, but if we don't get a FIN 2677 * we'll hang forever. 2678 * 2679 * XXXjl: 2680 * we should release the tp also, and use a 2681 * compressed state. 2682 */ 2683 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2684 soisdisconnected(so); 2685 tcp_timer_activate(tp, TT_2MSL, 2686 (tcp_fast_finwait2_recycle ? 2687 tcp_finwait2_timeout : 2688 TP_MAXIDLE(tp))); 2689 } 2690 tp->t_state = TCPS_FIN_WAIT_2; 2691 } 2692 break; 2693 2694 /* 2695 * In CLOSING STATE in addition to the processing for 2696 * the ESTABLISHED state if the ACK acknowledges our FIN 2697 * then enter the TIME-WAIT state, otherwise ignore 2698 * the segment. 2699 */ 2700 case TCPS_CLOSING: 2701 if (ourfinisacked) { 2702 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2703 tcp_twstart(tp); 2704 INP_INFO_WUNLOCK(&V_tcbinfo); 2705 m_freem(m); 2706 return; 2707 } 2708 break; 2709 2710 /* 2711 * In LAST_ACK, we may still be waiting for data to drain 2712 * and/or to be acked, as well as for the ack of our FIN. 2713 * If our FIN is now acknowledged, delete the TCB, 2714 * enter the closed state and return. 2715 */ 2716 case TCPS_LAST_ACK: 2717 if (ourfinisacked) { 2718 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2719 tp = tcp_close(tp); 2720 goto drop; 2721 } 2722 break; 2723 } 2724 } 2725 2726step6: 2727 INP_WLOCK_ASSERT(tp->t_inpcb); 2728 2729 /* 2730 * Update window information. 2731 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2732 */ 2733 if ((thflags & TH_ACK) && 2734 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2735 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2736 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2737 /* keep track of pure window updates */ 2738 if (tlen == 0 && 2739 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2740 TCPSTAT_INC(tcps_rcvwinupd); 2741 tp->snd_wnd = tiwin; 2742 tp->snd_wl1 = th->th_seq; 2743 tp->snd_wl2 = th->th_ack; 2744 if (tp->snd_wnd > tp->max_sndwnd) 2745 tp->max_sndwnd = tp->snd_wnd; 2746 needoutput = 1; 2747 } 2748 2749 /* 2750 * Process segments with URG. 2751 */ 2752 if ((thflags & TH_URG) && th->th_urp && 2753 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2754 /* 2755 * This is a kludge, but if we receive and accept 2756 * random urgent pointers, we'll crash in 2757 * soreceive. It's hard to imagine someone 2758 * actually wanting to send this much urgent data. 2759 */ 2760 SOCKBUF_LOCK(&so->so_rcv); 2761 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2762 th->th_urp = 0; /* XXX */ 2763 thflags &= ~TH_URG; /* XXX */ 2764 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2765 goto dodata; /* XXX */ 2766 } 2767 /* 2768 * If this segment advances the known urgent pointer, 2769 * then mark the data stream. This should not happen 2770 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2771 * a FIN has been received from the remote side. 2772 * In these states we ignore the URG. 2773 * 2774 * According to RFC961 (Assigned Protocols), 2775 * the urgent pointer points to the last octet 2776 * of urgent data. We continue, however, 2777 * to consider it to indicate the first octet 2778 * of data past the urgent section as the original 2779 * spec states (in one of two places). 2780 */ 2781 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2782 tp->rcv_up = th->th_seq + th->th_urp; 2783 so->so_oobmark = so->so_rcv.sb_cc + 2784 (tp->rcv_up - tp->rcv_nxt) - 1; 2785 if (so->so_oobmark == 0) 2786 so->so_rcv.sb_state |= SBS_RCVATMARK; 2787 sohasoutofband(so); 2788 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2789 } 2790 SOCKBUF_UNLOCK(&so->so_rcv); 2791 /* 2792 * Remove out of band data so doesn't get presented to user. 2793 * This can happen independent of advancing the URG pointer, 2794 * but if two URG's are pending at once, some out-of-band 2795 * data may creep in... ick. 2796 */ 2797 if (th->th_urp <= (u_long)tlen && 2798 !(so->so_options & SO_OOBINLINE)) { 2799 /* hdr drop is delayed */ 2800 tcp_pulloutofband(so, th, m, drop_hdrlen); 2801 } 2802 } else { 2803 /* 2804 * If no out of band data is expected, 2805 * pull receive urgent pointer along 2806 * with the receive window. 2807 */ 2808 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2809 tp->rcv_up = tp->rcv_nxt; 2810 } 2811dodata: /* XXX */ 2812 INP_WLOCK_ASSERT(tp->t_inpcb); 2813 2814 /* 2815 * Process the segment text, merging it into the TCP sequencing queue, 2816 * and arranging for acknowledgment of receipt if necessary. 2817 * This process logically involves adjusting tp->rcv_wnd as data 2818 * is presented to the user (this happens in tcp_usrreq.c, 2819 * case PRU_RCVD). If a FIN has already been received on this 2820 * connection then we just ignore the text. 2821 */ 2822 if ((tlen || (thflags & TH_FIN)) && 2823 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2824 tcp_seq save_start = th->th_seq; 2825 m_adj(m, drop_hdrlen); /* delayed header drop */ 2826 /* 2827 * Insert segment which includes th into TCP reassembly queue 2828 * with control block tp. Set thflags to whether reassembly now 2829 * includes a segment with FIN. This handles the common case 2830 * inline (segment is the next to be received on an established 2831 * connection, and the queue is empty), avoiding linkage into 2832 * and removal from the queue and repetition of various 2833 * conversions. 2834 * Set DELACK for segments received in order, but ack 2835 * immediately when segments are out of order (so 2836 * fast retransmit can work). 2837 */ 2838 if (th->th_seq == tp->rcv_nxt && 2839 LIST_EMPTY(&tp->t_segq) && 2840 TCPS_HAVEESTABLISHED(tp->t_state)) { 2841 if (DELAY_ACK(tp)) 2842 tp->t_flags |= TF_DELACK; 2843 else 2844 tp->t_flags |= TF_ACKNOW; 2845 tp->rcv_nxt += tlen; 2846 thflags = th->th_flags & TH_FIN; 2847 TCPSTAT_INC(tcps_rcvpack); 2848 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2849 ND6_HINT(tp); 2850 SOCKBUF_LOCK(&so->so_rcv); 2851 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2852 m_freem(m); 2853 else 2854 sbappendstream_locked(&so->so_rcv, m); 2855 /* NB: sorwakeup_locked() does an implicit unlock. */ 2856 sorwakeup_locked(so); 2857 } else { 2858 /* 2859 * XXX: Due to the header drop above "th" is 2860 * theoretically invalid by now. Fortunately 2861 * m_adj() doesn't actually frees any mbufs 2862 * when trimming from the head. 2863 */ 2864 thflags = tcp_reass(tp, th, &tlen, m); 2865 tp->t_flags |= TF_ACKNOW; 2866 } 2867 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2868 tcp_update_sack_list(tp, save_start, save_start + tlen); 2869#if 0 2870 /* 2871 * Note the amount of data that peer has sent into 2872 * our window, in order to estimate the sender's 2873 * buffer size. 2874 * XXX: Unused. 2875 */ 2876 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2877 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2878 else 2879 len = so->so_rcv.sb_hiwat; 2880#endif 2881 } else { 2882 m_freem(m); 2883 thflags &= ~TH_FIN; 2884 } 2885 2886 /* 2887 * If FIN is received ACK the FIN and let the user know 2888 * that the connection is closing. 2889 */ 2890 if (thflags & TH_FIN) { 2891 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2892 socantrcvmore(so); 2893 /* 2894 * If connection is half-synchronized 2895 * (ie NEEDSYN flag on) then delay ACK, 2896 * so it may be piggybacked when SYN is sent. 2897 * Otherwise, since we received a FIN then no 2898 * more input can be expected, send ACK now. 2899 */ 2900 if (tp->t_flags & TF_NEEDSYN) 2901 tp->t_flags |= TF_DELACK; 2902 else 2903 tp->t_flags |= TF_ACKNOW; 2904 tp->rcv_nxt++; 2905 } 2906 switch (tp->t_state) { 2907 2908 /* 2909 * In SYN_RECEIVED and ESTABLISHED STATES 2910 * enter the CLOSE_WAIT state. 2911 */ 2912 case TCPS_SYN_RECEIVED: 2913 tp->t_starttime = ticks; 2914 /* FALLTHROUGH */ 2915 case TCPS_ESTABLISHED: 2916 tp->t_state = TCPS_CLOSE_WAIT; 2917 break; 2918 2919 /* 2920 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2921 * enter the CLOSING state. 2922 */ 2923 case TCPS_FIN_WAIT_1: 2924 tp->t_state = TCPS_CLOSING; 2925 break; 2926 2927 /* 2928 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2929 * starting the time-wait timer, turning off the other 2930 * standard timers. 2931 */ 2932 case TCPS_FIN_WAIT_2: 2933 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2934 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2935 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2936 ti_locked)); 2937 2938 tcp_twstart(tp); 2939 INP_INFO_WUNLOCK(&V_tcbinfo); 2940 return; 2941 } 2942 } 2943 if (ti_locked == TI_WLOCKED) 2944 INP_INFO_WUNLOCK(&V_tcbinfo); 2945 ti_locked = TI_UNLOCKED; 2946 2947#ifdef TCPDEBUG 2948 if (so->so_options & SO_DEBUG) 2949 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2950 &tcp_savetcp, 0); 2951#endif 2952 2953 /* 2954 * Return any desired output. 2955 */ 2956 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2957 (void) tcp_output(tp); 2958 2959check_delack: 2960 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2961 __func__, ti_locked)); 2962 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2963 INP_WLOCK_ASSERT(tp->t_inpcb); 2964 2965 if (tp->t_flags & TF_DELACK) { 2966 tp->t_flags &= ~TF_DELACK; 2967 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2968 } 2969 INP_WUNLOCK(tp->t_inpcb); 2970 return; 2971 2972dropafterack: 2973 /* 2974 * Generate an ACK dropping incoming segment if it occupies 2975 * sequence space, where the ACK reflects our state. 2976 * 2977 * We can now skip the test for the RST flag since all 2978 * paths to this code happen after packets containing 2979 * RST have been dropped. 2980 * 2981 * In the SYN-RECEIVED state, don't send an ACK unless the 2982 * segment we received passes the SYN-RECEIVED ACK test. 2983 * If it fails send a RST. This breaks the loop in the 2984 * "LAND" DoS attack, and also prevents an ACK storm 2985 * between two listening ports that have been sent forged 2986 * SYN segments, each with the source address of the other. 2987 */ 2988 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2989 (SEQ_GT(tp->snd_una, th->th_ack) || 2990 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2991 rstreason = BANDLIM_RST_OPENPORT; 2992 goto dropwithreset; 2993 } 2994#ifdef TCPDEBUG 2995 if (so->so_options & SO_DEBUG) 2996 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2997 &tcp_savetcp, 0); 2998#endif 2999 if (ti_locked == TI_WLOCKED) 3000 INP_INFO_WUNLOCK(&V_tcbinfo); 3001 ti_locked = TI_UNLOCKED; 3002 3003 tp->t_flags |= TF_ACKNOW; 3004 (void) tcp_output(tp); 3005 INP_WUNLOCK(tp->t_inpcb); 3006 m_freem(m); 3007 return; 3008 3009dropwithreset: 3010 if (ti_locked == TI_WLOCKED) 3011 INP_INFO_WUNLOCK(&V_tcbinfo); 3012 ti_locked = TI_UNLOCKED; 3013 3014 if (tp != NULL) { 3015 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3016 INP_WUNLOCK(tp->t_inpcb); 3017 } else 3018 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3019 return; 3020 3021drop: 3022 if (ti_locked == TI_WLOCKED) { 3023 INP_INFO_WUNLOCK(&V_tcbinfo); 3024 ti_locked = TI_UNLOCKED; 3025 } 3026#ifdef INVARIANTS 3027 else 3028 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3029#endif 3030 3031 /* 3032 * Drop space held by incoming segment and return. 3033 */ 3034#ifdef TCPDEBUG 3035 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3036 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3037 &tcp_savetcp, 0); 3038#endif 3039 if (tp != NULL) 3040 INP_WUNLOCK(tp->t_inpcb); 3041 m_freem(m); 3042} 3043 3044/* 3045 * Issue RST and make ACK acceptable to originator of segment. 3046 * The mbuf must still include the original packet header. 3047 * tp may be NULL. 3048 */ 3049static void 3050tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3051 int tlen, int rstreason) 3052{ 3053#ifdef INET 3054 struct ip *ip; 3055#endif 3056#ifdef INET6 3057 struct ip6_hdr *ip6; 3058#endif 3059 3060 if (tp != NULL) { 3061 INP_WLOCK_ASSERT(tp->t_inpcb); 3062 } 3063 3064 /* Don't bother if destination was broadcast/multicast. */ 3065 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3066 goto drop; 3067#ifdef INET6 3068 if (mtod(m, struct ip *)->ip_v == 6) { 3069 ip6 = mtod(m, struct ip6_hdr *); 3070 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3071 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3072 goto drop; 3073 /* IPv6 anycast check is done at tcp6_input() */ 3074 } 3075#endif 3076#if defined(INET) && defined(INET6) 3077 else 3078#endif 3079#ifdef INET 3080 { 3081 ip = mtod(m, struct ip *); 3082 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3083 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3084 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3085 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3086 goto drop; 3087 } 3088#endif 3089 3090 /* Perform bandwidth limiting. */ 3091 if (badport_bandlim(rstreason) < 0) 3092 goto drop; 3093 3094 /* tcp_respond consumes the mbuf chain. */ 3095 if (th->th_flags & TH_ACK) { 3096 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3097 th->th_ack, TH_RST); 3098 } else { 3099 if (th->th_flags & TH_SYN) 3100 tlen++; 3101 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3102 (tcp_seq)0, TH_RST|TH_ACK); 3103 } 3104 return; 3105drop: 3106 m_freem(m); 3107} 3108 3109/* 3110 * Parse TCP options and place in tcpopt. 3111 */ 3112static void 3113tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3114{ 3115 int opt, optlen; 3116 3117 to->to_flags = 0; 3118 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3119 opt = cp[0]; 3120 if (opt == TCPOPT_EOL) 3121 break; 3122 if (opt == TCPOPT_NOP) 3123 optlen = 1; 3124 else { 3125 if (cnt < 2) 3126 break; 3127 optlen = cp[1]; 3128 if (optlen < 2 || optlen > cnt) 3129 break; 3130 } 3131 switch (opt) { 3132 case TCPOPT_MAXSEG: 3133 if (optlen != TCPOLEN_MAXSEG) 3134 continue; 3135 if (!(flags & TO_SYN)) 3136 continue; 3137 to->to_flags |= TOF_MSS; 3138 bcopy((char *)cp + 2, 3139 (char *)&to->to_mss, sizeof(to->to_mss)); 3140 to->to_mss = ntohs(to->to_mss); 3141 break; 3142 case TCPOPT_WINDOW: 3143 if (optlen != TCPOLEN_WINDOW) 3144 continue; 3145 if (!(flags & TO_SYN)) 3146 continue; 3147 to->to_flags |= TOF_SCALE; 3148 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3149 break; 3150 case TCPOPT_TIMESTAMP: 3151 if (optlen != TCPOLEN_TIMESTAMP) 3152 continue; 3153 to->to_flags |= TOF_TS; 3154 bcopy((char *)cp + 2, 3155 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3156 to->to_tsval = ntohl(to->to_tsval); 3157 bcopy((char *)cp + 6, 3158 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3159 to->to_tsecr = ntohl(to->to_tsecr); 3160 break; 3161#ifdef TCP_SIGNATURE 3162 /* 3163 * XXX In order to reply to a host which has set the 3164 * TCP_SIGNATURE option in its initial SYN, we have to 3165 * record the fact that the option was observed here 3166 * for the syncache code to perform the correct response. 3167 */ 3168 case TCPOPT_SIGNATURE: 3169 if (optlen != TCPOLEN_SIGNATURE) 3170 continue; 3171 to->to_flags |= TOF_SIGNATURE; 3172 to->to_signature = cp + 2; 3173 break; 3174#endif 3175 case TCPOPT_SACK_PERMITTED: 3176 if (optlen != TCPOLEN_SACK_PERMITTED) 3177 continue; 3178 if (!(flags & TO_SYN)) 3179 continue; 3180 if (!V_tcp_do_sack) 3181 continue; 3182 to->to_flags |= TOF_SACKPERM; 3183 break; 3184 case TCPOPT_SACK: 3185 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3186 continue; 3187 if (flags & TO_SYN) 3188 continue; 3189 to->to_flags |= TOF_SACK; 3190 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3191 to->to_sacks = cp + 2; 3192 TCPSTAT_INC(tcps_sack_rcv_blocks); 3193 break; 3194 default: 3195 continue; 3196 } 3197 } 3198} 3199 3200/* 3201 * Pull out of band byte out of a segment so 3202 * it doesn't appear in the user's data queue. 3203 * It is still reflected in the segment length for 3204 * sequencing purposes. 3205 */ 3206static void 3207tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3208 int off) 3209{ 3210 int cnt = off + th->th_urp - 1; 3211 3212 while (cnt >= 0) { 3213 if (m->m_len > cnt) { 3214 char *cp = mtod(m, caddr_t) + cnt; 3215 struct tcpcb *tp = sototcpcb(so); 3216 3217 INP_WLOCK_ASSERT(tp->t_inpcb); 3218 3219 tp->t_iobc = *cp; 3220 tp->t_oobflags |= TCPOOB_HAVEDATA; 3221 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3222 m->m_len--; 3223 if (m->m_flags & M_PKTHDR) 3224 m->m_pkthdr.len--; 3225 return; 3226 } 3227 cnt -= m->m_len; 3228 m = m->m_next; 3229 if (m == NULL) 3230 break; 3231 } 3232 panic("tcp_pulloutofband"); 3233} 3234 3235/* 3236 * Collect new round-trip time estimate 3237 * and update averages and current timeout. 3238 */ 3239static void 3240tcp_xmit_timer(struct tcpcb *tp, int rtt) 3241{ 3242 int delta; 3243 3244 INP_WLOCK_ASSERT(tp->t_inpcb); 3245 3246 TCPSTAT_INC(tcps_rttupdated); 3247 tp->t_rttupdated++; 3248 if (tp->t_srtt != 0) { 3249 /* 3250 * srtt is stored as fixed point with 5 bits after the 3251 * binary point (i.e., scaled by 8). The following magic 3252 * is equivalent to the smoothing algorithm in rfc793 with 3253 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3254 * point). Adjust rtt to origin 0. 3255 */ 3256 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3257 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3258 3259 if ((tp->t_srtt += delta) <= 0) 3260 tp->t_srtt = 1; 3261 3262 /* 3263 * We accumulate a smoothed rtt variance (actually, a 3264 * smoothed mean difference), then set the retransmit 3265 * timer to smoothed rtt + 4 times the smoothed variance. 3266 * rttvar is stored as fixed point with 4 bits after the 3267 * binary point (scaled by 16). The following is 3268 * equivalent to rfc793 smoothing with an alpha of .75 3269 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3270 * rfc793's wired-in beta. 3271 */ 3272 if (delta < 0) 3273 delta = -delta; 3274 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3275 if ((tp->t_rttvar += delta) <= 0) 3276 tp->t_rttvar = 1; 3277 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3278 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3279 } else { 3280 /* 3281 * No rtt measurement yet - use the unsmoothed rtt. 3282 * Set the variance to half the rtt (so our first 3283 * retransmit happens at 3*rtt). 3284 */ 3285 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3286 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3287 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3288 } 3289 tp->t_rtttime = 0; 3290 tp->t_rxtshift = 0; 3291 3292 /* 3293 * the retransmit should happen at rtt + 4 * rttvar. 3294 * Because of the way we do the smoothing, srtt and rttvar 3295 * will each average +1/2 tick of bias. When we compute 3296 * the retransmit timer, we want 1/2 tick of rounding and 3297 * 1 extra tick because of +-1/2 tick uncertainty in the 3298 * firing of the timer. The bias will give us exactly the 3299 * 1.5 tick we need. But, because the bias is 3300 * statistical, we have to test that we don't drop below 3301 * the minimum feasible timer (which is 2 ticks). 3302 */ 3303 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3304 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3305 3306 /* 3307 * We received an ack for a packet that wasn't retransmitted; 3308 * it is probably safe to discard any error indications we've 3309 * received recently. This isn't quite right, but close enough 3310 * for now (a route might have failed after we sent a segment, 3311 * and the return path might not be symmetrical). 3312 */ 3313 tp->t_softerror = 0; 3314} 3315 3316/* 3317 * Determine a reasonable value for maxseg size. 3318 * If the route is known, check route for mtu. 3319 * If none, use an mss that can be handled on the outgoing 3320 * interface without forcing IP to fragment; if bigger than 3321 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 3322 * to utilize large mbufs. If no route is found, route has no mtu, 3323 * or the destination isn't local, use a default, hopefully conservative 3324 * size (usually 512 or the default IP max size, but no more than the mtu 3325 * of the interface), as we can't discover anything about intervening 3326 * gateways or networks. We also initialize the congestion/slow start 3327 * window to be a single segment if the destination isn't local. 3328 * While looking at the routing entry, we also initialize other path-dependent 3329 * parameters from pre-set or cached values in the routing entry. 3330 * 3331 * Also take into account the space needed for options that we 3332 * send regularly. Make maxseg shorter by that amount to assure 3333 * that we can send maxseg amount of data even when the options 3334 * are present. Store the upper limit of the length of options plus 3335 * data in maxopd. 3336 * 3337 * NOTE that this routine is only called when we process an incoming 3338 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3339 * settings are handled in tcp_mssopt(). 3340 */ 3341void 3342tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3343 struct hc_metrics_lite *metricptr, int *mtuflags) 3344{ 3345 int mss = 0; 3346 u_long maxmtu = 0; 3347 struct inpcb *inp = tp->t_inpcb; 3348 struct hc_metrics_lite metrics; 3349 int origoffer; 3350#ifdef INET6 3351 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3352 size_t min_protoh = isipv6 ? 3353 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3354 sizeof (struct tcpiphdr); 3355#else 3356 const size_t min_protoh = sizeof(struct tcpiphdr); 3357#endif 3358 3359 INP_WLOCK_ASSERT(tp->t_inpcb); 3360 3361 if (mtuoffer != -1) { 3362 KASSERT(offer == -1, ("%s: conflict", __func__)); 3363 offer = mtuoffer - min_protoh; 3364 } 3365 origoffer = offer; 3366 3367 /* Initialize. */ 3368#ifdef INET6 3369 if (isipv6) { 3370 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags); 3371 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3372 } 3373#endif 3374#if defined(INET) && defined(INET6) 3375 else 3376#endif 3377#ifdef INET 3378 { 3379 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags); 3380 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3381 } 3382#endif 3383 3384 /* 3385 * No route to sender, stay with default mss and return. 3386 */ 3387 if (maxmtu == 0) { 3388 /* 3389 * In case we return early we need to initialize metrics 3390 * to a defined state as tcp_hc_get() would do for us 3391 * if there was no cache hit. 3392 */ 3393 if (metricptr != NULL) 3394 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3395 return; 3396 } 3397 3398 /* What have we got? */ 3399 switch (offer) { 3400 case 0: 3401 /* 3402 * Offer == 0 means that there was no MSS on the SYN 3403 * segment, in this case we use tcp_mssdflt as 3404 * already assigned to t_maxopd above. 3405 */ 3406 offer = tp->t_maxopd; 3407 break; 3408 3409 case -1: 3410 /* 3411 * Offer == -1 means that we didn't receive SYN yet. 3412 */ 3413 /* FALLTHROUGH */ 3414 3415 default: 3416 /* 3417 * Prevent DoS attack with too small MSS. Round up 3418 * to at least minmss. 3419 */ 3420 offer = max(offer, V_tcp_minmss); 3421 } 3422 3423 /* 3424 * rmx information is now retrieved from tcp_hostcache. 3425 */ 3426 tcp_hc_get(&inp->inp_inc, &metrics); 3427 if (metricptr != NULL) 3428 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3429 3430 /* 3431 * If there's a discovered mtu int tcp hostcache, use it 3432 * else, use the link mtu. 3433 */ 3434 if (metrics.rmx_mtu) 3435 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3436 else { 3437#ifdef INET6 3438 if (isipv6) { 3439 mss = maxmtu - min_protoh; 3440 if (!V_path_mtu_discovery && 3441 !in6_localaddr(&inp->in6p_faddr)) 3442 mss = min(mss, V_tcp_v6mssdflt); 3443 } 3444#endif 3445#if defined(INET) && defined(INET6) 3446 else 3447#endif 3448#ifdef INET 3449 { 3450 mss = maxmtu - min_protoh; 3451 if (!V_path_mtu_discovery && 3452 !in_localaddr(inp->inp_faddr)) 3453 mss = min(mss, V_tcp_mssdflt); 3454 } 3455#endif 3456 /* 3457 * XXX - The above conditional (mss = maxmtu - min_protoh) 3458 * probably violates the TCP spec. 3459 * The problem is that, since we don't know the 3460 * other end's MSS, we are supposed to use a conservative 3461 * default. But, if we do that, then MTU discovery will 3462 * never actually take place, because the conservative 3463 * default is much less than the MTUs typically seen 3464 * on the Internet today. For the moment, we'll sweep 3465 * this under the carpet. 3466 * 3467 * The conservative default might not actually be a problem 3468 * if the only case this occurs is when sending an initial 3469 * SYN with options and data to a host we've never talked 3470 * to before. Then, they will reply with an MSS value which 3471 * will get recorded and the new parameters should get 3472 * recomputed. For Further Study. 3473 */ 3474 } 3475 mss = min(mss, offer); 3476 3477 /* 3478 * Sanity check: make sure that maxopd will be large 3479 * enough to allow some data on segments even if the 3480 * all the option space is used (40bytes). Otherwise 3481 * funny things may happen in tcp_output. 3482 */ 3483 mss = max(mss, 64); 3484 3485 /* 3486 * maxopd stores the maximum length of data AND options 3487 * in a segment; maxseg is the amount of data in a normal 3488 * segment. We need to store this value (maxopd) apart 3489 * from maxseg, because now every segment carries options 3490 * and thus we normally have somewhat less data in segments. 3491 */ 3492 tp->t_maxopd = mss; 3493 3494 /* 3495 * origoffer==-1 indicates that no segments were received yet. 3496 * In this case we just guess. 3497 */ 3498 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3499 (origoffer == -1 || 3500 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3501 mss -= TCPOLEN_TSTAMP_APPA; 3502 3503#if (MCLBYTES & (MCLBYTES - 1)) == 0 3504 if (mss > MCLBYTES) 3505 mss &= ~(MCLBYTES-1); 3506#else 3507 if (mss > MCLBYTES) 3508 mss = mss / MCLBYTES * MCLBYTES; 3509#endif 3510 tp->t_maxseg = mss; 3511} 3512 3513void 3514tcp_mss(struct tcpcb *tp, int offer) 3515{ 3516 int mss; 3517 u_long bufsize; 3518 struct inpcb *inp; 3519 struct socket *so; 3520 struct hc_metrics_lite metrics; 3521 int mtuflags = 0; 3522 3523 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3524 3525 tcp_mss_update(tp, offer, -1, &metrics, &mtuflags); 3526 3527 mss = tp->t_maxseg; 3528 inp = tp->t_inpcb; 3529 3530 /* 3531 * If there's a pipesize, change the socket buffer to that size, 3532 * don't change if sb_hiwat is different than default (then it 3533 * has been changed on purpose with setsockopt). 3534 * Make the socket buffers an integral number of mss units; 3535 * if the mss is larger than the socket buffer, decrease the mss. 3536 */ 3537 so = inp->inp_socket; 3538 SOCKBUF_LOCK(&so->so_snd); 3539 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3540 bufsize = metrics.rmx_sendpipe; 3541 else 3542 bufsize = so->so_snd.sb_hiwat; 3543 if (bufsize < mss) 3544 mss = bufsize; 3545 else { 3546 bufsize = roundup(bufsize, mss); 3547 if (bufsize > sb_max) 3548 bufsize = sb_max; 3549 if (bufsize > so->so_snd.sb_hiwat) 3550 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3551 } 3552 SOCKBUF_UNLOCK(&so->so_snd); 3553 tp->t_maxseg = mss; 3554 3555 SOCKBUF_LOCK(&so->so_rcv); 3556 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3557 bufsize = metrics.rmx_recvpipe; 3558 else 3559 bufsize = so->so_rcv.sb_hiwat; 3560 if (bufsize > mss) { 3561 bufsize = roundup(bufsize, mss); 3562 if (bufsize > sb_max) 3563 bufsize = sb_max; 3564 if (bufsize > so->so_rcv.sb_hiwat) 3565 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3566 } 3567 SOCKBUF_UNLOCK(&so->so_rcv); 3568 3569 /* Check the interface for TSO capabilities. */ 3570 if (mtuflags & CSUM_TSO) 3571 tp->t_flags |= TF_TSO; 3572} 3573 3574/* 3575 * Determine the MSS option to send on an outgoing SYN. 3576 */ 3577int 3578tcp_mssopt(struct in_conninfo *inc) 3579{ 3580 int mss = 0; 3581 u_long maxmtu = 0; 3582 u_long thcmtu = 0; 3583 size_t min_protoh; 3584 3585 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3586 3587#ifdef INET6 3588 if (inc->inc_flags & INC_ISIPV6) { 3589 mss = V_tcp_v6mssdflt; 3590 maxmtu = tcp_maxmtu6(inc, NULL); 3591 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3592 } 3593#endif 3594#if defined(INET) && defined(INET6) 3595 else 3596#endif 3597#ifdef INET 3598 { 3599 mss = V_tcp_mssdflt; 3600 maxmtu = tcp_maxmtu(inc, NULL); 3601 min_protoh = sizeof(struct tcpiphdr); 3602 } 3603#endif 3604#if defined(INET6) || defined(INET) 3605 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3606#endif 3607 3608 if (maxmtu && thcmtu) 3609 mss = min(maxmtu, thcmtu) - min_protoh; 3610 else if (maxmtu || thcmtu) 3611 mss = max(maxmtu, thcmtu) - min_protoh; 3612 3613 return (mss); 3614} 3615 3616 3617/* 3618 * On a partial ack arrives, force the retransmission of the 3619 * next unacknowledged segment. Do not clear tp->t_dupacks. 3620 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3621 * be started again. 3622 */ 3623static void 3624tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3625{ 3626 tcp_seq onxt = tp->snd_nxt; 3627 u_long ocwnd = tp->snd_cwnd; 3628 3629 INP_WLOCK_ASSERT(tp->t_inpcb); 3630 3631 tcp_timer_activate(tp, TT_REXMT, 0); 3632 tp->t_rtttime = 0; 3633 tp->snd_nxt = th->th_ack; 3634 /* 3635 * Set snd_cwnd to one segment beyond acknowledged offset. 3636 * (tp->snd_una has not yet been updated when this function is called.) 3637 */ 3638 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3639 tp->t_flags |= TF_ACKNOW; 3640 (void) tcp_output(tp); 3641 tp->snd_cwnd = ocwnd; 3642 if (SEQ_GT(onxt, tp->snd_nxt)) 3643 tp->snd_nxt = onxt; 3644 /* 3645 * Partial window deflation. Relies on fact that tp->snd_una 3646 * not updated yet. 3647 */ 3648 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3649 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3650 else 3651 tp->snd_cwnd = 0; 3652 tp->snd_cwnd += tp->t_maxseg; 3653} 3654