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 * All rights reserved. 9 * 10 * Portions of this software were developed at the Centre for Advanced Internet 11 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 12 * James Healy and David Hayes, made possible in part by a grant from the Cisco 13 * University Research Program Fund at Community Foundation Silicon Valley. 14 * 15 * Portions of this software were developed at the Centre for Advanced 16 * Internet Architectures, Swinburne University of Technology, Melbourne, 17 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 18 * 19 * Redistribution and use in source and binary forms, with or without 20 * modification, are permitted provided that the following conditions 21 * are met: 22 * 1. Redistributions of source code must retain the above copyright 23 * notice, this list of conditions and the following disclaimer. 24 * 2. Redistributions in binary form must reproduce the above copyright 25 * notice, this list of conditions and the following disclaimer in the 26 * documentation and/or other materials provided with the distribution. 27 * 4. Neither the name of the University nor the names of its contributors 28 * may be used to endorse or promote products derived from this software 29 * without specific prior written permission. 30 * 31 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 32 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 33 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 34 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 35 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 36 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 37 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 38 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 39 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 40 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 41 * SUCH DAMAGE. 42 * 43 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 44 */ 45 46 #include <sys/cdefs.h> 47 __FBSDID("$FreeBSD$"); 48 49 #include "opt_ipfw.h" /* for ipfw_fwd */ 50 #include "opt_inet.h" 51 #include "opt_inet6.h" 52 #include "opt_ipsec.h" 53 #include "opt_tcpdebug.h" 54 55 #include <sys/param.h> 56 #include <sys/kernel.h> 57 #include <sys/hhook.h> 58 #include <sys/malloc.h> 59 #include <sys/mbuf.h> 60 #include <sys/proc.h> /* for proc0 declaration */ 61 #include <sys/protosw.h> 62 #include <sys/signalvar.h> 63 #include <sys/socket.h> 64 #include <sys/socketvar.h> 65 #include <sys/sysctl.h> 66 #include <sys/syslog.h> 67 #include <sys/systm.h> 68 69 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 70 71 #include <vm/uma.h> 72 73 #include <net/if.h> 74 #include <net/route.h> 75 #include <net/vnet.h> 76 77 #define TCPSTATES /* for logging */ 78 79 #include <netinet/cc.h> 80 #include <netinet/in.h> 81 #include <netinet/in_pcb.h> 82 #include <netinet/in_systm.h> 83 #include <netinet/in_var.h> 84 #include <netinet/ip.h> 85 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 86 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 87 #include <netinet/ip_var.h> 88 #include <netinet/ip_options.h> 89 #include <netinet/ip6.h> 90 #include <netinet/icmp6.h> 91 #include <netinet6/in6_pcb.h> 92 #include <netinet6/ip6_var.h> 93 #include <netinet6/nd6.h> 94 #include <netinet/tcp_fsm.h> 95 #include <netinet/tcp_seq.h> 96 #include <netinet/tcp_timer.h> 97 #include <netinet/tcp_var.h> 98 #include <netinet6/tcp6_var.h> 99 #include <netinet/tcpip.h> 100 #include <netinet/tcp_syncache.h> 101 #ifdef TCPDEBUG 102 #include <netinet/tcp_debug.h> 103 #endif /* TCPDEBUG */ 104 105 #ifdef IPSEC 106 #include <netipsec/ipsec.h> 107 #include <netipsec/ipsec6.h> 108 #endif /*IPSEC*/ 109 110 #include <machine/in_cksum.h> 111 112 #include <security/mac/mac_framework.h> 113 114 const int tcprexmtthresh = 3; 115 116 VNET_DEFINE(struct tcpstat, tcpstat); 117 SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 118 &VNET_NAME(tcpstat), tcpstat, 119 "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 120 121 int tcp_log_in_vain = 0; 122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 123 &tcp_log_in_vain, 0, 124 "Log all incoming TCP segments to closed ports"); 125 126 VNET_DEFINE(int, blackhole) = 0; 127 #define V_blackhole VNET(blackhole) 128 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 129 &VNET_NAME(blackhole), 0, 130 "Do not send RST on segments to closed ports"); 131 132 VNET_DEFINE(int, tcp_delack_enabled) = 1; 133 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 134 &VNET_NAME(tcp_delack_enabled), 0, 135 "Delay ACK to try and piggyback it onto a data packet"); 136 137 VNET_DEFINE(int, drop_synfin) = 0; 138 #define V_drop_synfin VNET(drop_synfin) 139 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 140 &VNET_NAME(drop_synfin), 0, 141 "Drop TCP packets with SYN+FIN set"); 142 143 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 144 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 145 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 146 &VNET_NAME(tcp_do_rfc3042), 0, 147 "Enable RFC 3042 (Limited Transmit)"); 148 149 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 151 &VNET_NAME(tcp_do_rfc3390), 0, 152 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 153 154 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 156 &VNET_NAME(tcp_do_rfc3465), 0, 157 "Enable RFC 3465 (Appropriate Byte Counting)"); 158 159 VNET_DEFINE(int, tcp_abc_l_var) = 2; 160 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 161 &VNET_NAME(tcp_abc_l_var), 2, 162 "Cap the max cwnd increment during slow-start to this number of segments"); 163 164 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 165 166 VNET_DEFINE(int, tcp_do_ecn) = 0; 167 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 168 &VNET_NAME(tcp_do_ecn), 0, 169 "TCP ECN support"); 170 171 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 172 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 173 &VNET_NAME(tcp_ecn_maxretries), 0, 174 "Max retries before giving up on ECN"); 175 176 VNET_DEFINE(int, tcp_insecure_rst) = 0; 177 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 178 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 179 &VNET_NAME(tcp_insecure_rst), 0, 180 "Follow the old (insecure) criteria for accepting RST packets"); 181 182 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 183 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 184 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 185 &VNET_NAME(tcp_do_autorcvbuf), 0, 186 "Enable automatic receive buffer sizing"); 187 188 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 189 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 190 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 191 &VNET_NAME(tcp_autorcvbuf_inc), 0, 192 "Incrementor step size of automatic receive buffer"); 193 194 VNET_DEFINE(int, tcp_autorcvbuf_max) = 256*1024; 195 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 196 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 197 &VNET_NAME(tcp_autorcvbuf_max), 0, 198 "Max size of automatic receive buffer"); 199 200 int tcp_read_locking = 1; 201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, read_locking, CTLFLAG_RW, 202 &tcp_read_locking, 0, "Enable read locking strategy"); 203 204 VNET_DEFINE(struct inpcbhead, tcb); 205 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 206 VNET_DEFINE(struct inpcbinfo, tcbinfo); 207 208 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 209 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 210 struct socket *, struct tcpcb *, int, int, uint8_t, 211 int); 212 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 213 struct tcpcb *, int, int); 214 static void tcp_pulloutofband(struct socket *, 215 struct tcphdr *, struct mbuf *, int); 216 static void tcp_xmit_timer(struct tcpcb *, int); 217 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 218 static void inline tcp_fields_to_host(struct tcphdr *); 219 #ifdef TCP_SIGNATURE 220 static void inline tcp_fields_to_net(struct tcphdr *); 221 static int inline tcp_signature_verify_input(struct mbuf *, int, int, 222 int, struct tcpopt *, struct tcphdr *, u_int); 223 #endif 224 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 225 uint16_t type); 226 static void inline cc_conn_init(struct tcpcb *tp); 227 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 228 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 229 struct tcphdr *th, struct tcpopt *to); 230 231 /* 232 * Kernel module interface for updating tcpstat. The argument is an index 233 * into tcpstat treated as an array of u_long. While this encodes the 234 * general layout of tcpstat into the caller, it doesn't encode its location, 235 * so that future changes to add, for example, per-CPU stats support won't 236 * cause binary compatibility problems for kernel modules. 237 */ 238 void 239 kmod_tcpstat_inc(int statnum) 240 { 241 242 (*((u_long *)&V_tcpstat + statnum))++; 243 } 244 245 /* 246 * Wrapper for the TCP established input helper hook. 247 */ 248 static void inline 249 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 250 { 251 struct tcp_hhook_data hhook_data; 252 253 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 254 hhook_data.tp = tp; 255 hhook_data.th = th; 256 hhook_data.to = to; 257 258 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 259 tp->osd); 260 } 261 } 262 263 /* 264 * CC wrapper hook functions 265 */ 266 static void inline 267 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 268 { 269 INP_WLOCK_ASSERT(tp->t_inpcb); 270 271 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 272 if (tp->snd_cwnd == min(tp->snd_cwnd, tp->snd_wnd)) 273 tp->ccv->flags |= CCF_CWND_LIMITED; 274 else 275 tp->ccv->flags &= ~CCF_CWND_LIMITED; 276 277 if (type == CC_ACK) { 278 if (tp->snd_cwnd > tp->snd_ssthresh) { 279 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 280 V_tcp_abc_l_var * tp->t_maxseg); 281 if (tp->t_bytes_acked >= tp->snd_cwnd) { 282 tp->t_bytes_acked -= tp->snd_cwnd; 283 tp->ccv->flags |= CCF_ABC_SENTAWND; 284 } 285 } else { 286 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 287 tp->t_bytes_acked = 0; 288 } 289 } 290 291 if (CC_ALGO(tp)->ack_received != NULL) { 292 /* XXXLAS: Find a way to live without this */ 293 tp->ccv->curack = th->th_ack; 294 CC_ALGO(tp)->ack_received(tp->ccv, type); 295 } 296 } 297 298 static void inline 299 cc_conn_init(struct tcpcb *tp) 300 { 301 struct hc_metrics_lite metrics; 302 struct inpcb *inp = tp->t_inpcb; 303 int rtt; 304 #ifdef INET6 305 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 306 #endif 307 308 INP_WLOCK_ASSERT(tp->t_inpcb); 309 310 tcp_hc_get(&inp->inp_inc, &metrics); 311 312 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 313 tp->t_srtt = rtt; 314 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 315 TCPSTAT_INC(tcps_usedrtt); 316 if (metrics.rmx_rttvar) { 317 tp->t_rttvar = metrics.rmx_rttvar; 318 TCPSTAT_INC(tcps_usedrttvar); 319 } else { 320 /* default variation is +- 1 rtt */ 321 tp->t_rttvar = 322 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 323 } 324 TCPT_RANGESET(tp->t_rxtcur, 325 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 326 tp->t_rttmin, TCPTV_REXMTMAX); 327 } 328 if (metrics.rmx_ssthresh) { 329 /* 330 * There's some sort of gateway or interface 331 * buffer limit on the path. Use this to set 332 * the slow start threshhold, but set the 333 * threshold to no less than 2*mss. 334 */ 335 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 336 TCPSTAT_INC(tcps_usedssthresh); 337 } 338 339 /* 340 * Set the slow-start flight size depending on whether this 341 * is a local network or not. 342 * 343 * Extend this so we cache the cwnd too and retrieve it here. 344 * Make cwnd even bigger than RFC3390 suggests but only if we 345 * have previous experience with the remote host. Be careful 346 * not make cwnd bigger than remote receive window or our own 347 * send socket buffer. Maybe put some additional upper bound 348 * on the retrieved cwnd. Should do incremental updates to 349 * hostcache when cwnd collapses so next connection doesn't 350 * overloads the path again. 351 * 352 * XXXAO: Initializing the CWND from the hostcache is broken 353 * and in its current form not RFC conformant. It is disabled 354 * until fixed or removed entirely. 355 * 356 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 357 * We currently check only in syncache_socket for that. 358 */ 359 /* #define TCP_METRICS_CWND */ 360 #ifdef TCP_METRICS_CWND 361 if (metrics.rmx_cwnd) 362 tp->snd_cwnd = max(tp->t_maxseg, min(metrics.rmx_cwnd / 2, 363 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 364 else 365 #endif 366 if (V_tcp_do_rfc3390) 367 tp->snd_cwnd = min(4 * tp->t_maxseg, 368 max(2 * tp->t_maxseg, 4380)); 369 #ifdef INET6 370 else if (isipv6 && in6_localaddr(&inp->in6p_faddr)) 371 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local; 372 #endif 373 #if defined(INET) && defined(INET6) 374 else if (!isipv6 && in_localaddr(inp->inp_faddr)) 375 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local; 376 #endif 377 #ifdef INET 378 else if (in_localaddr(inp->inp_faddr)) 379 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local; 380 #endif 381 else 382 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz; 383 384 if (CC_ALGO(tp)->conn_init != NULL) 385 CC_ALGO(tp)->conn_init(tp->ccv); 386 } 387 388 void inline 389 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 390 { 391 INP_WLOCK_ASSERT(tp->t_inpcb); 392 393 switch(type) { 394 case CC_NDUPACK: 395 if (!IN_FASTRECOVERY(tp->t_flags)) { 396 tp->snd_recover = tp->snd_max; 397 if (tp->t_flags & TF_ECN_PERMIT) 398 tp->t_flags |= TF_ECN_SND_CWR; 399 } 400 break; 401 case CC_ECN: 402 if (!IN_CONGRECOVERY(tp->t_flags)) { 403 TCPSTAT_INC(tcps_ecn_rcwnd); 404 tp->snd_recover = tp->snd_max; 405 if (tp->t_flags & TF_ECN_PERMIT) 406 tp->t_flags |= TF_ECN_SND_CWR; 407 } 408 break; 409 case CC_RTO: 410 tp->t_dupacks = 0; 411 tp->t_bytes_acked = 0; 412 EXIT_RECOVERY(tp->t_flags); 413 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 414 tp->t_maxseg) * tp->t_maxseg; 415 tp->snd_cwnd = tp->t_maxseg; 416 break; 417 case CC_RTO_ERR: 418 TCPSTAT_INC(tcps_sndrexmitbad); 419 /* RTO was unnecessary, so reset everything. */ 420 tp->snd_cwnd = tp->snd_cwnd_prev; 421 tp->snd_ssthresh = tp->snd_ssthresh_prev; 422 tp->snd_recover = tp->snd_recover_prev; 423 if (tp->t_flags & TF_WASFRECOVERY) 424 ENTER_FASTRECOVERY(tp->t_flags); 425 if (tp->t_flags & TF_WASCRECOVERY) 426 ENTER_CONGRECOVERY(tp->t_flags); 427 tp->snd_nxt = tp->snd_max; 428 tp->t_flags &= ~TF_PREVVALID; 429 tp->t_badrxtwin = 0; 430 break; 431 } 432 433 if (CC_ALGO(tp)->cong_signal != NULL) { 434 if (th != NULL) 435 tp->ccv->curack = th->th_ack; 436 CC_ALGO(tp)->cong_signal(tp->ccv, type); 437 } 438 } 439 440 static void inline 441 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 442 { 443 INP_WLOCK_ASSERT(tp->t_inpcb); 444 445 /* XXXLAS: KASSERT that we're in recovery? */ 446 447 if (CC_ALGO(tp)->post_recovery != NULL) { 448 tp->ccv->curack = th->th_ack; 449 CC_ALGO(tp)->post_recovery(tp->ccv); 450 } 451 /* XXXLAS: EXIT_RECOVERY ? */ 452 tp->t_bytes_acked = 0; 453 } 454 455 static inline void 456 tcp_fields_to_host(struct tcphdr *th) 457 { 458 459 th->th_seq = ntohl(th->th_seq); 460 th->th_ack = ntohl(th->th_ack); 461 th->th_win = ntohs(th->th_win); 462 th->th_urp = ntohs(th->th_urp); 463 } 464 465 #ifdef TCP_SIGNATURE 466 static inline void 467 tcp_fields_to_net(struct tcphdr *th) 468 { 469 470 th->th_seq = htonl(th->th_seq); 471 th->th_ack = htonl(th->th_ack); 472 th->th_win = htons(th->th_win); 473 th->th_urp = htons(th->th_urp); 474 } 475 476 static inline int 477 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 478 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 479 { 480 int ret; 481 482 tcp_fields_to_net(th); 483 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 484 tcp_fields_to_host(th); 485 return (ret); 486 } 487 #endif 488 489 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 490 #ifdef INET6 491 #define ND6_HINT(tp) \ 492 do { \ 493 if ((tp) && (tp)->t_inpcb && \ 494 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 495 nd6_nud_hint(NULL, NULL, 0); \ 496 } while (0) 497 #else 498 #define ND6_HINT(tp) 499 #endif 500 501 /* 502 * Indicate whether this ack should be delayed. We can delay the ack if 503 * - there is no delayed ack timer in progress and 504 * - our last ack wasn't a 0-sized window. We never want to delay 505 * the ack that opens up a 0-sized window and 506 * - delayed acks are enabled or 507 * - this is a half-synchronized T/TCP connection. 508 */ 509 #define DELAY_ACK(tp) \ 510 ((!tcp_timer_active(tp, TT_DELACK) && \ 511 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 512 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 513 514 /* 515 * TCP input handling is split into multiple parts: 516 * tcp6_input is a thin wrapper around tcp_input for the extended 517 * ip6_protox[] call format in ip6_input 518 * tcp_input handles primary segment validation, inpcb lookup and 519 * SYN processing on listen sockets 520 * tcp_do_segment processes the ACK and text of the segment for 521 * establishing, established and closing connections 522 */ 523 #ifdef INET6 524 int 525 tcp6_input(struct mbuf **mp, int *offp, int proto) 526 { 527 struct mbuf *m = *mp; 528 struct in6_ifaddr *ia6; 529 530 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 531 532 /* 533 * draft-itojun-ipv6-tcp-to-anycast 534 * better place to put this in? 535 */ 536 ia6 = ip6_getdstifaddr(m); 537 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 538 struct ip6_hdr *ip6; 539 540 ifa_free(&ia6->ia_ifa); 541 ip6 = mtod(m, struct ip6_hdr *); 542 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 543 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 544 return IPPROTO_DONE; 545 } 546 547 tcp_input(m, *offp); 548 return IPPROTO_DONE; 549 } 550 #endif /* INET6 */ 551 552 void 553 tcp_input(struct mbuf *m, int off0) 554 { 555 struct tcphdr *th = NULL; 556 struct ip *ip = NULL; 557 #ifdef INET 558 struct ipovly *ipov; 559 #endif 560 struct inpcb *inp = NULL; 561 struct tcpcb *tp = NULL; 562 struct socket *so = NULL; 563 u_char *optp = NULL; 564 int optlen = 0; 565 #ifdef INET 566 int len; 567 #endif 568 int tlen = 0, off; 569 int drop_hdrlen; 570 int thflags; 571 int rstreason = 0; /* For badport_bandlim accounting purposes */ 572 #ifdef TCP_SIGNATURE 573 uint8_t sig_checked = 0; 574 #endif 575 uint8_t iptos = 0; 576 #ifdef INET 577 #ifdef IPFIREWALL_FORWARD 578 struct m_tag *fwd_tag; 579 #endif 580 #endif /* INET */ 581 #ifdef INET6 582 struct ip6_hdr *ip6 = NULL; 583 int isipv6; 584 #else 585 const void *ip6 = NULL; 586 #if (defined(INET) && defined(IPFIREWALL_FORWARD)) || defined(TCPDEBUG) 587 const int isipv6 = 0; 588 #endif 589 #endif /* INET6 */ 590 struct tcpopt to; /* options in this segment */ 591 char *s = NULL; /* address and port logging */ 592 int ti_locked; 593 #define TI_UNLOCKED 1 594 #define TI_RLOCKED 2 595 #define TI_WLOCKED 3 596 597 #ifdef TCPDEBUG 598 /* 599 * The size of tcp_saveipgen must be the size of the max ip header, 600 * now IPv6. 601 */ 602 u_char tcp_saveipgen[IP6_HDR_LEN]; 603 struct tcphdr tcp_savetcp; 604 short ostate = 0; 605 #endif 606 607 #ifdef INET6 608 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 609 #endif 610 611 to.to_flags = 0; 612 TCPSTAT_INC(tcps_rcvtotal); 613 614 #ifdef INET6 615 if (isipv6) { 616 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 617 ip6 = mtod(m, struct ip6_hdr *); 618 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 619 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 620 TCPSTAT_INC(tcps_rcvbadsum); 621 goto drop; 622 } 623 th = (struct tcphdr *)((caddr_t)ip6 + off0); 624 625 /* 626 * Be proactive about unspecified IPv6 address in source. 627 * As we use all-zero to indicate unbounded/unconnected pcb, 628 * unspecified IPv6 address can be used to confuse us. 629 * 630 * Note that packets with unspecified IPv6 destination is 631 * already dropped in ip6_input. 632 */ 633 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 634 /* XXX stat */ 635 goto drop; 636 } 637 } 638 #endif 639 #if defined(INET) && defined(INET6) 640 else 641 #endif 642 #ifdef INET 643 { 644 /* 645 * Get IP and TCP header together in first mbuf. 646 * Note: IP leaves IP header in first mbuf. 647 */ 648 if (off0 > sizeof (struct ip)) { 649 ip_stripoptions(m, (struct mbuf *)0); 650 off0 = sizeof(struct ip); 651 } 652 if (m->m_len < sizeof (struct tcpiphdr)) { 653 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 654 == NULL) { 655 TCPSTAT_INC(tcps_rcvshort); 656 return; 657 } 658 } 659 ip = mtod(m, struct ip *); 660 ipov = (struct ipovly *)ip; 661 th = (struct tcphdr *)((caddr_t)ip + off0); 662 tlen = ip->ip_len; 663 664 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 665 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 666 th->th_sum = m->m_pkthdr.csum_data; 667 else 668 th->th_sum = in_pseudo(ip->ip_src.s_addr, 669 ip->ip_dst.s_addr, 670 htonl(m->m_pkthdr.csum_data + 671 ip->ip_len + 672 IPPROTO_TCP)); 673 th->th_sum ^= 0xffff; 674 #ifdef TCPDEBUG 675 ipov->ih_len = (u_short)tlen; 676 ipov->ih_len = htons(ipov->ih_len); 677 #endif 678 } else { 679 /* 680 * Checksum extended TCP header and data. 681 */ 682 len = sizeof (struct ip) + tlen; 683 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 684 ipov->ih_len = (u_short)tlen; 685 ipov->ih_len = htons(ipov->ih_len); 686 th->th_sum = in_cksum(m, len); 687 } 688 if (th->th_sum) { 689 TCPSTAT_INC(tcps_rcvbadsum); 690 goto drop; 691 } 692 /* Re-initialization for later version check */ 693 ip->ip_v = IPVERSION; 694 } 695 #endif /* INET */ 696 697 #ifdef INET6 698 if (isipv6) 699 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 700 #endif 701 #if defined(INET) && defined(INET6) 702 else 703 #endif 704 #ifdef INET 705 iptos = ip->ip_tos; 706 #endif 707 708 /* 709 * Check that TCP offset makes sense, 710 * pull out TCP options and adjust length. XXX 711 */ 712 off = th->th_off << 2; 713 if (off < sizeof (struct tcphdr) || off > tlen) { 714 TCPSTAT_INC(tcps_rcvbadoff); 715 goto drop; 716 } 717 tlen -= off; /* tlen is used instead of ti->ti_len */ 718 if (off > sizeof (struct tcphdr)) { 719 #ifdef INET6 720 if (isipv6) { 721 IP6_EXTHDR_CHECK(m, off0, off, ); 722 ip6 = mtod(m, struct ip6_hdr *); 723 th = (struct tcphdr *)((caddr_t)ip6 + off0); 724 } 725 #endif 726 #if defined(INET) && defined(INET6) 727 else 728 #endif 729 #ifdef INET 730 { 731 if (m->m_len < sizeof(struct ip) + off) { 732 if ((m = m_pullup(m, sizeof (struct ip) + off)) 733 == NULL) { 734 TCPSTAT_INC(tcps_rcvshort); 735 return; 736 } 737 ip = mtod(m, struct ip *); 738 ipov = (struct ipovly *)ip; 739 th = (struct tcphdr *)((caddr_t)ip + off0); 740 } 741 } 742 #endif 743 optlen = off - sizeof (struct tcphdr); 744 optp = (u_char *)(th + 1); 745 } 746 thflags = th->th_flags; 747 748 /* 749 * Convert TCP protocol specific fields to host format. 750 */ 751 tcp_fields_to_host(th); 752 753 /* 754 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 755 */ 756 drop_hdrlen = off0 + off; 757 758 /* 759 * Locate pcb for segment, which requires a lock on tcbinfo. 760 * Optimisticaly acquire a global read lock rather than a write lock 761 * unless header flags necessarily imply a state change. There are 762 * two cases where we might discover later we need a write lock 763 * despite the flags: ACKs moving a connection out of the syncache, 764 * and ACKs for a connection in TIMEWAIT. 765 */ 766 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 767 tcp_read_locking == 0) { 768 INP_INFO_WLOCK(&V_tcbinfo); 769 ti_locked = TI_WLOCKED; 770 } else { 771 INP_INFO_RLOCK(&V_tcbinfo); 772 ti_locked = TI_RLOCKED; 773 } 774 775 findpcb: 776 #ifdef INVARIANTS 777 if (ti_locked == TI_RLOCKED) 778 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 779 else if (ti_locked == TI_WLOCKED) 780 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 781 else 782 panic("%s: findpcb ti_locked %d\n", __func__, ti_locked); 783 #endif 784 785 #ifdef INET 786 #ifdef IPFIREWALL_FORWARD 787 /* 788 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 789 */ 790 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 791 792 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 793 struct sockaddr_in *next_hop; 794 795 next_hop = (struct sockaddr_in *)(fwd_tag+1); 796 /* 797 * Transparently forwarded. Pretend to be the destination. 798 * already got one like this? 799 */ 800 inp = in_pcblookup_hash(&V_tcbinfo, 801 ip->ip_src, th->th_sport, 802 ip->ip_dst, th->th_dport, 803 0, m->m_pkthdr.rcvif); 804 if (!inp) { 805 /* It's new. Try to find the ambushing socket. */ 806 inp = in_pcblookup_hash(&V_tcbinfo, 807 ip->ip_src, th->th_sport, 808 next_hop->sin_addr, 809 next_hop->sin_port ? 810 ntohs(next_hop->sin_port) : 811 th->th_dport, 812 INPLOOKUP_WILDCARD, 813 m->m_pkthdr.rcvif); 814 } 815 /* Remove the tag from the packet. We don't need it anymore. */ 816 m_tag_delete(m, fwd_tag); 817 } else 818 #endif /* IPFIREWALL_FORWARD */ 819 #endif /* INET */ 820 { 821 #ifdef INET6 822 if (isipv6) 823 inp = in6_pcblookup_hash(&V_tcbinfo, 824 &ip6->ip6_src, th->th_sport, 825 &ip6->ip6_dst, th->th_dport, 826 INPLOOKUP_WILDCARD, 827 m->m_pkthdr.rcvif); 828 #endif 829 #if defined(INET) && defined(INET6) 830 else 831 #endif 832 #ifdef INET 833 inp = in_pcblookup_hash(&V_tcbinfo, 834 ip->ip_src, th->th_sport, 835 ip->ip_dst, th->th_dport, 836 INPLOOKUP_WILDCARD, 837 m->m_pkthdr.rcvif); 838 #endif 839 } 840 841 /* 842 * If the INPCB does not exist then all data in the incoming 843 * segment is discarded and an appropriate RST is sent back. 844 * XXX MRT Send RST using which routing table? 845 */ 846 if (inp == NULL) { 847 /* 848 * Log communication attempts to ports that are not 849 * in use. 850 */ 851 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 852 tcp_log_in_vain == 2) { 853 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 854 log(LOG_INFO, "%s; %s: Connection attempt " 855 "to closed port\n", s, __func__); 856 } 857 /* 858 * When blackholing do not respond with a RST but 859 * completely ignore the segment and drop it. 860 */ 861 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 862 V_blackhole == 2) 863 goto dropunlock; 864 865 rstreason = BANDLIM_RST_CLOSEDPORT; 866 goto dropwithreset; 867 } 868 INP_WLOCK(inp); 869 if (!(inp->inp_flags & INP_HW_FLOWID) 870 && (m->m_flags & M_FLOWID) 871 && ((inp->inp_socket == NULL) 872 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 873 inp->inp_flags |= INP_HW_FLOWID; 874 inp->inp_flags &= ~INP_SW_FLOWID; 875 inp->inp_flowid = m->m_pkthdr.flowid; 876 } 877 #ifdef IPSEC 878 #ifdef INET6 879 if (isipv6 && ipsec6_in_reject(m, inp)) { 880 V_ipsec6stat.in_polvio++; 881 goto dropunlock; 882 } else 883 #endif /* INET6 */ 884 if (ipsec4_in_reject(m, inp) != 0) { 885 V_ipsec4stat.in_polvio++; 886 goto dropunlock; 887 } 888 #endif /* IPSEC */ 889 890 /* 891 * Check the minimum TTL for socket. 892 */ 893 if (inp->inp_ip_minttl != 0) { 894 #ifdef INET6 895 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 896 goto dropunlock; 897 else 898 #endif 899 if (inp->inp_ip_minttl > ip->ip_ttl) 900 goto dropunlock; 901 } 902 903 /* 904 * A previous connection in TIMEWAIT state is supposed to catch stray 905 * or duplicate segments arriving late. If this segment was a 906 * legitimate new connection attempt the old INPCB gets removed and 907 * we can try again to find a listening socket. 908 * 909 * At this point, due to earlier optimism, we may hold a read lock on 910 * the inpcbinfo, rather than a write lock. If so, we need to 911 * upgrade, or if that fails, acquire a reference on the inpcb, drop 912 * all locks, acquire a global write lock, and then re-acquire the 913 * inpcb lock. We may at that point discover that another thread has 914 * tried to free the inpcb, in which case we need to loop back and 915 * try to find a new inpcb to deliver to. 916 */ 917 relocked: 918 if (inp->inp_flags & INP_TIMEWAIT) { 919 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 920 ("%s: INP_TIMEWAIT ti_locked %d", __func__, ti_locked)); 921 922 if (ti_locked == TI_RLOCKED) { 923 if (INP_INFO_TRY_UPGRADE(&V_tcbinfo) == 0) { 924 in_pcbref(inp); 925 INP_WUNLOCK(inp); 926 INP_INFO_RUNLOCK(&V_tcbinfo); 927 INP_INFO_WLOCK(&V_tcbinfo); 928 ti_locked = TI_WLOCKED; 929 INP_WLOCK(inp); 930 if (in_pcbrele(inp)) { 931 inp = NULL; 932 goto findpcb; 933 } 934 } else 935 ti_locked = TI_WLOCKED; 936 } 937 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 938 939 #ifdef TCP_SIGNATURE 940 tcp_dooptions(&to, optp, optlen, 941 (thflags & TH_SYN) ? TO_SYN : 0); 942 if (sig_checked == 0) { 943 tp = intotcpcb(inp); 944 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 945 rstreason = BANDLIM_RST_CLOSEDPORT; 946 goto dropwithreset; 947 } 948 if (!tcp_signature_verify_input(m, off0, tlen, optlen, 949 &to, th, tp->t_flags)) 950 goto dropunlock; 951 sig_checked = 1; 952 } 953 #else 954 if (thflags & TH_SYN) 955 tcp_dooptions(&to, optp, optlen, TO_SYN); 956 #endif 957 /* 958 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 959 */ 960 if (tcp_twcheck(inp, &to, th, m, tlen)) 961 goto findpcb; 962 INP_INFO_WUNLOCK(&V_tcbinfo); 963 return; 964 } 965 /* 966 * The TCPCB may no longer exist if the connection is winding 967 * down or it is in the CLOSED state. Either way we drop the 968 * segment and send an appropriate response. 969 */ 970 tp = intotcpcb(inp); 971 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 972 rstreason = BANDLIM_RST_CLOSEDPORT; 973 goto dropwithreset; 974 } 975 976 /* 977 * We've identified a valid inpcb, but it could be that we need an 978 * inpcbinfo write lock and have only a read lock. In this case, 979 * attempt to upgrade/relock using the same strategy as the TIMEWAIT 980 * case above. If we relock, we have to jump back to 'relocked' as 981 * the connection might now be in TIMEWAIT. 982 */ 983 if (tp->t_state != TCPS_ESTABLISHED || 984 (thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 985 tcp_read_locking == 0) { 986 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 987 ("%s: upgrade check ti_locked %d", __func__, ti_locked)); 988 989 if (ti_locked == TI_RLOCKED) { 990 if (INP_INFO_TRY_UPGRADE(&V_tcbinfo) == 0) { 991 in_pcbref(inp); 992 INP_WUNLOCK(inp); 993 INP_INFO_RUNLOCK(&V_tcbinfo); 994 INP_INFO_WLOCK(&V_tcbinfo); 995 ti_locked = TI_WLOCKED; 996 INP_WLOCK(inp); 997 if (in_pcbrele(inp)) { 998 inp = NULL; 999 goto findpcb; 1000 } 1001 goto relocked; 1002 } else 1003 ti_locked = TI_WLOCKED; 1004 } 1005 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1006 } 1007 1008 #ifdef MAC 1009 INP_WLOCK_ASSERT(inp); 1010 if (mac_inpcb_check_deliver(inp, m)) 1011 goto dropunlock; 1012 #endif 1013 so = inp->inp_socket; 1014 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1015 #ifdef TCPDEBUG 1016 if (so->so_options & SO_DEBUG) { 1017 ostate = tp->t_state; 1018 if (isipv6) { 1019 #ifdef INET6 1020 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1021 #endif 1022 } else 1023 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1024 tcp_savetcp = *th; 1025 } 1026 #endif /* TCPDEBUG */ 1027 /* 1028 * When the socket is accepting connections (the INPCB is in LISTEN 1029 * state) we look into the SYN cache if this is a new connection 1030 * attempt or the completion of a previous one. 1031 */ 1032 if (so->so_options & SO_ACCEPTCONN) { 1033 struct in_conninfo inc; 1034 1035 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1036 "tp not listening", __func__)); 1037 1038 bzero(&inc, sizeof(inc)); 1039 #ifdef INET6 1040 if (isipv6) { 1041 inc.inc_flags |= INC_ISIPV6; 1042 inc.inc6_faddr = ip6->ip6_src; 1043 inc.inc6_laddr = ip6->ip6_dst; 1044 } else 1045 #endif 1046 { 1047 inc.inc_faddr = ip->ip_src; 1048 inc.inc_laddr = ip->ip_dst; 1049 } 1050 inc.inc_fport = th->th_sport; 1051 inc.inc_lport = th->th_dport; 1052 inc.inc_fibnum = so->so_fibnum; 1053 1054 /* 1055 * Check for an existing connection attempt in syncache if 1056 * the flag is only ACK. A successful lookup creates a new 1057 * socket appended to the listen queue in SYN_RECEIVED state. 1058 */ 1059 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1060 /* 1061 * Parse the TCP options here because 1062 * syncookies need access to the reflected 1063 * timestamp. 1064 */ 1065 tcp_dooptions(&to, optp, optlen, 0); 1066 /* 1067 * NB: syncache_expand() doesn't unlock 1068 * inp and tcpinfo locks. 1069 */ 1070 if (!syncache_expand(&inc, &to, th, &so, m)) { 1071 /* 1072 * No syncache entry or ACK was not 1073 * for our SYN/ACK. Send a RST. 1074 * NB: syncache did its own logging 1075 * of the failure cause. 1076 */ 1077 rstreason = BANDLIM_RST_OPENPORT; 1078 goto dropwithreset; 1079 } 1080 if (so == NULL) { 1081 /* 1082 * We completed the 3-way handshake 1083 * but could not allocate a socket 1084 * either due to memory shortage, 1085 * listen queue length limits or 1086 * global socket limits. Send RST 1087 * or wait and have the remote end 1088 * retransmit the ACK for another 1089 * try. 1090 */ 1091 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1092 log(LOG_DEBUG, "%s; %s: Listen socket: " 1093 "Socket allocation failed due to " 1094 "limits or memory shortage, %s\n", 1095 s, __func__, 1096 V_tcp_sc_rst_sock_fail ? 1097 "sending RST" : "try again"); 1098 if (V_tcp_sc_rst_sock_fail) { 1099 rstreason = BANDLIM_UNLIMITED; 1100 goto dropwithreset; 1101 } else 1102 goto dropunlock; 1103 } 1104 /* 1105 * Socket is created in state SYN_RECEIVED. 1106 * Unlock the listen socket, lock the newly 1107 * created socket and update the tp variable. 1108 */ 1109 INP_WUNLOCK(inp); /* listen socket */ 1110 inp = sotoinpcb(so); 1111 INP_WLOCK(inp); /* new connection */ 1112 tp = intotcpcb(inp); 1113 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1114 ("%s: ", __func__)); 1115 #ifdef TCP_SIGNATURE 1116 if (sig_checked == 0) { 1117 tcp_dooptions(&to, optp, optlen, 1118 (thflags & TH_SYN) ? TO_SYN : 0); 1119 if (!tcp_signature_verify_input(m, off0, tlen, 1120 optlen, &to, th, tp->t_flags)) { 1121 1122 /* 1123 * In SYN_SENT state if it receives an 1124 * RST, it is allowed for further 1125 * processing. 1126 */ 1127 if ((thflags & TH_RST) == 0 || 1128 (tp->t_state == TCPS_SYN_SENT) == 0) 1129 goto dropunlock; 1130 } 1131 sig_checked = 1; 1132 } 1133 #endif 1134 1135 /* 1136 * Process the segment and the data it 1137 * contains. tcp_do_segment() consumes 1138 * the mbuf chain and unlocks the inpcb. 1139 */ 1140 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1141 iptos, ti_locked); 1142 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1143 return; 1144 } 1145 /* 1146 * Segment flag validation for new connection attempts: 1147 * 1148 * Our (SYN|ACK) response was rejected. 1149 * Check with syncache and remove entry to prevent 1150 * retransmits. 1151 * 1152 * NB: syncache_chkrst does its own logging of failure 1153 * causes. 1154 */ 1155 if (thflags & TH_RST) { 1156 syncache_chkrst(&inc, th); 1157 goto dropunlock; 1158 } 1159 /* 1160 * We can't do anything without SYN. 1161 */ 1162 if ((thflags & TH_SYN) == 0) { 1163 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1164 log(LOG_DEBUG, "%s; %s: Listen socket: " 1165 "SYN is missing, segment ignored\n", 1166 s, __func__); 1167 TCPSTAT_INC(tcps_badsyn); 1168 goto dropunlock; 1169 } 1170 /* 1171 * (SYN|ACK) is bogus on a listen socket. 1172 */ 1173 if (thflags & TH_ACK) { 1174 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1175 log(LOG_DEBUG, "%s; %s: Listen socket: " 1176 "SYN|ACK invalid, segment rejected\n", 1177 s, __func__); 1178 syncache_badack(&inc); /* XXX: Not needed! */ 1179 TCPSTAT_INC(tcps_badsyn); 1180 rstreason = BANDLIM_RST_OPENPORT; 1181 goto dropwithreset; 1182 } 1183 /* 1184 * If the drop_synfin option is enabled, drop all 1185 * segments with both the SYN and FIN bits set. 1186 * This prevents e.g. nmap from identifying the 1187 * TCP/IP stack. 1188 * XXX: Poor reasoning. nmap has other methods 1189 * and is constantly refining its stack detection 1190 * strategies. 1191 * XXX: This is a violation of the TCP specification 1192 * and was used by RFC1644. 1193 */ 1194 if ((thflags & TH_FIN) && V_drop_synfin) { 1195 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1196 log(LOG_DEBUG, "%s; %s: Listen socket: " 1197 "SYN|FIN segment ignored (based on " 1198 "sysctl setting)\n", s, __func__); 1199 TCPSTAT_INC(tcps_badsyn); 1200 goto dropunlock; 1201 } 1202 /* 1203 * Segment's flags are (SYN) or (SYN|FIN). 1204 * 1205 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1206 * as they do not affect the state of the TCP FSM. 1207 * The data pointed to by TH_URG and th_urp is ignored. 1208 */ 1209 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1210 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1211 KASSERT(thflags & (TH_SYN), 1212 ("%s: Listen socket: TH_SYN not set", __func__)); 1213 #ifdef INET6 1214 /* 1215 * If deprecated address is forbidden, 1216 * we do not accept SYN to deprecated interface 1217 * address to prevent any new inbound connection from 1218 * getting established. 1219 * When we do not accept SYN, we send a TCP RST, 1220 * with deprecated source address (instead of dropping 1221 * it). We compromise it as it is much better for peer 1222 * to send a RST, and RST will be the final packet 1223 * for the exchange. 1224 * 1225 * If we do not forbid deprecated addresses, we accept 1226 * the SYN packet. RFC2462 does not suggest dropping 1227 * SYN in this case. 1228 * If we decipher RFC2462 5.5.4, it says like this: 1229 * 1. use of deprecated addr with existing 1230 * communication is okay - "SHOULD continue to be 1231 * used" 1232 * 2. use of it with new communication: 1233 * (2a) "SHOULD NOT be used if alternate address 1234 * with sufficient scope is available" 1235 * (2b) nothing mentioned otherwise. 1236 * Here we fall into (2b) case as we have no choice in 1237 * our source address selection - we must obey the peer. 1238 * 1239 * The wording in RFC2462 is confusing, and there are 1240 * multiple description text for deprecated address 1241 * handling - worse, they are not exactly the same. 1242 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1243 */ 1244 if (isipv6 && !V_ip6_use_deprecated) { 1245 struct in6_ifaddr *ia6; 1246 1247 ia6 = ip6_getdstifaddr(m); 1248 if (ia6 != NULL && 1249 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1250 ifa_free(&ia6->ia_ifa); 1251 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1252 log(LOG_DEBUG, "%s; %s: Listen socket: " 1253 "Connection attempt to deprecated " 1254 "IPv6 address rejected\n", 1255 s, __func__); 1256 rstreason = BANDLIM_RST_OPENPORT; 1257 goto dropwithreset; 1258 } 1259 ifa_free(&ia6->ia_ifa); 1260 } 1261 #endif /* INET6 */ 1262 /* 1263 * Basic sanity checks on incoming SYN requests: 1264 * Don't respond if the destination is a link layer 1265 * broadcast according to RFC1122 4.2.3.10, p. 104. 1266 * If it is from this socket it must be forged. 1267 * Don't respond if the source or destination is a 1268 * global or subnet broad- or multicast address. 1269 * Note that it is quite possible to receive unicast 1270 * link-layer packets with a broadcast IP address. Use 1271 * in_broadcast() to find them. 1272 */ 1273 if (m->m_flags & (M_BCAST|M_MCAST)) { 1274 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1275 log(LOG_DEBUG, "%s; %s: Listen socket: " 1276 "Connection attempt from broad- or multicast " 1277 "link layer address ignored\n", s, __func__); 1278 goto dropunlock; 1279 } 1280 #ifdef INET6 1281 if (isipv6) { 1282 if (th->th_dport == th->th_sport && 1283 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1284 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1285 log(LOG_DEBUG, "%s; %s: Listen socket: " 1286 "Connection attempt to/from self " 1287 "ignored\n", s, __func__); 1288 goto dropunlock; 1289 } 1290 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1291 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1292 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1293 log(LOG_DEBUG, "%s; %s: Listen socket: " 1294 "Connection attempt from/to multicast " 1295 "address ignored\n", s, __func__); 1296 goto dropunlock; 1297 } 1298 } 1299 #endif 1300 #if defined(INET) && defined(INET6) 1301 else 1302 #endif 1303 #ifdef INET 1304 { 1305 if (th->th_dport == th->th_sport && 1306 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1307 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1308 log(LOG_DEBUG, "%s; %s: Listen socket: " 1309 "Connection attempt from/to self " 1310 "ignored\n", s, __func__); 1311 goto dropunlock; 1312 } 1313 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1314 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1315 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1316 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1318 log(LOG_DEBUG, "%s; %s: Listen socket: " 1319 "Connection attempt from/to broad- " 1320 "or multicast address ignored\n", 1321 s, __func__); 1322 goto dropunlock; 1323 } 1324 } 1325 #endif 1326 /* 1327 * SYN appears to be valid. Create compressed TCP state 1328 * for syncache. 1329 */ 1330 #ifdef TCPDEBUG 1331 if (so->so_options & SO_DEBUG) 1332 tcp_trace(TA_INPUT, ostate, tp, 1333 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1334 #endif 1335 tcp_dooptions(&to, optp, optlen, TO_SYN); 1336 syncache_add(&inc, &to, th, inp, &so, m); 1337 /* 1338 * Entry added to syncache and mbuf consumed. 1339 * Everything already unlocked by syncache_add(). 1340 */ 1341 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1342 return; 1343 } 1344 1345 #ifdef TCP_SIGNATURE 1346 if (sig_checked == 0) { 1347 tcp_dooptions(&to, optp, optlen, 1348 (thflags & TH_SYN) ? TO_SYN : 0); 1349 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1350 th, tp->t_flags)) { 1351 1352 /* 1353 * In SYN_SENT state if it receives an RST, it is 1354 * allowed for further processing. 1355 */ 1356 if ((thflags & TH_RST) == 0 || 1357 (tp->t_state == TCPS_SYN_SENT) == 0) 1358 goto dropunlock; 1359 } 1360 sig_checked = 1; 1361 } 1362 #endif 1363 1364 /* 1365 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1366 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1367 * the inpcb, and unlocks pcbinfo. 1368 */ 1369 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1370 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1371 return; 1372 1373 dropwithreset: 1374 if (ti_locked == TI_RLOCKED) 1375 INP_INFO_RUNLOCK(&V_tcbinfo); 1376 else if (ti_locked == TI_WLOCKED) 1377 INP_INFO_WUNLOCK(&V_tcbinfo); 1378 else 1379 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked); 1380 ti_locked = TI_UNLOCKED; 1381 1382 if (inp != NULL) { 1383 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1384 INP_WUNLOCK(inp); 1385 } else 1386 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1387 m = NULL; /* mbuf chain got consumed. */ 1388 goto drop; 1389 1390 dropunlock: 1391 if (ti_locked == TI_RLOCKED) 1392 INP_INFO_RUNLOCK(&V_tcbinfo); 1393 else if (ti_locked == TI_WLOCKED) 1394 INP_INFO_WUNLOCK(&V_tcbinfo); 1395 else 1396 panic("%s: dropunlock ti_locked %d", __func__, ti_locked); 1397 ti_locked = TI_UNLOCKED; 1398 1399 if (inp != NULL) 1400 INP_WUNLOCK(inp); 1401 1402 drop: 1403 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1404 if (s != NULL) 1405 free(s, M_TCPLOG); 1406 if (m != NULL) 1407 m_freem(m); 1408 } 1409 1410 static void 1411 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1412 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1413 int ti_locked) 1414 { 1415 int thflags, acked, ourfinisacked, needoutput = 0; 1416 int rstreason, todrop, win; 1417 u_long tiwin; 1418 struct tcpopt to; 1419 1420 #ifdef TCPDEBUG 1421 /* 1422 * The size of tcp_saveipgen must be the size of the max ip header, 1423 * now IPv6. 1424 */ 1425 u_char tcp_saveipgen[IP6_HDR_LEN]; 1426 struct tcphdr tcp_savetcp; 1427 short ostate = 0; 1428 #endif 1429 thflags = th->th_flags; 1430 tp->sackhint.last_sack_ack = 0; 1431 1432 /* 1433 * If this is either a state-changing packet or current state isn't 1434 * established, we require a write lock on tcbinfo. Otherwise, we 1435 * allow either a read lock or a write lock, as we may have acquired 1436 * a write lock due to a race. 1437 * 1438 * Require a global write lock for SYN/FIN/RST segments or 1439 * non-established connections; otherwise accept either a read or 1440 * write lock, as we may have conservatively acquired a write lock in 1441 * certain cases in tcp_input() (is this still true?). Currently we 1442 * will never enter with no lock, so we try to drop it quickly in the 1443 * common pure ack/pure data cases. 1444 */ 1445 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1446 tp->t_state != TCPS_ESTABLISHED) { 1447 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1448 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1449 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1450 } else { 1451 #ifdef INVARIANTS 1452 if (ti_locked == TI_RLOCKED) 1453 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1454 else if (ti_locked == TI_WLOCKED) 1455 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1456 else 1457 panic("%s: ti_locked %d for EST", __func__, 1458 ti_locked); 1459 #endif 1460 } 1461 INP_WLOCK_ASSERT(tp->t_inpcb); 1462 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1463 __func__)); 1464 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1465 __func__)); 1466 1467 /* 1468 * Segment received on connection. 1469 * Reset idle time and keep-alive timer. 1470 * XXX: This should be done after segment 1471 * validation to ignore broken/spoofed segs. 1472 */ 1473 tp->t_rcvtime = ticks; 1474 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1475 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1476 1477 /* 1478 * Unscale the window into a 32-bit value. 1479 * For the SYN_SENT state the scale is zero. 1480 */ 1481 tiwin = th->th_win << tp->snd_scale; 1482 1483 /* 1484 * TCP ECN processing. 1485 */ 1486 if (tp->t_flags & TF_ECN_PERMIT) { 1487 if (thflags & TH_CWR) 1488 tp->t_flags &= ~TF_ECN_SND_ECE; 1489 switch (iptos & IPTOS_ECN_MASK) { 1490 case IPTOS_ECN_CE: 1491 tp->t_flags |= TF_ECN_SND_ECE; 1492 TCPSTAT_INC(tcps_ecn_ce); 1493 break; 1494 case IPTOS_ECN_ECT0: 1495 TCPSTAT_INC(tcps_ecn_ect0); 1496 break; 1497 case IPTOS_ECN_ECT1: 1498 TCPSTAT_INC(tcps_ecn_ect1); 1499 break; 1500 } 1501 /* Congestion experienced. */ 1502 if (thflags & TH_ECE) { 1503 cc_cong_signal(tp, th, CC_ECN); 1504 } 1505 } 1506 1507 /* 1508 * Parse options on any incoming segment. 1509 */ 1510 tcp_dooptions(&to, (u_char *)(th + 1), 1511 (th->th_off << 2) - sizeof(struct tcphdr), 1512 (thflags & TH_SYN) ? TO_SYN : 0); 1513 1514 /* 1515 * If echoed timestamp is later than the current time, 1516 * fall back to non RFC1323 RTT calculation. Normalize 1517 * timestamp if syncookies were used when this connection 1518 * was established. 1519 */ 1520 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1521 to.to_tsecr -= tp->ts_offset; 1522 if (TSTMP_GT(to.to_tsecr, ticks)) 1523 to.to_tsecr = 0; 1524 } 1525 1526 /* 1527 * Process options only when we get SYN/ACK back. The SYN case 1528 * for incoming connections is handled in tcp_syncache. 1529 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1530 * or <SYN,ACK>) segment itself is never scaled. 1531 * XXX this is traditional behavior, may need to be cleaned up. 1532 */ 1533 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1534 if ((to.to_flags & TOF_SCALE) && 1535 (tp->t_flags & TF_REQ_SCALE)) { 1536 tp->t_flags |= TF_RCVD_SCALE; 1537 tp->snd_scale = to.to_wscale; 1538 } 1539 /* 1540 * Initial send window. It will be updated with 1541 * the next incoming segment to the scaled value. 1542 */ 1543 tp->snd_wnd = th->th_win; 1544 if (to.to_flags & TOF_TS) { 1545 tp->t_flags |= TF_RCVD_TSTMP; 1546 tp->ts_recent = to.to_tsval; 1547 tp->ts_recent_age = ticks; 1548 } 1549 if (to.to_flags & TOF_MSS) 1550 tcp_mss(tp, to.to_mss); 1551 if ((tp->t_flags & TF_SACK_PERMIT) && 1552 (to.to_flags & TOF_SACKPERM) == 0) 1553 tp->t_flags &= ~TF_SACK_PERMIT; 1554 } 1555 1556 /* 1557 * Header prediction: check for the two common cases 1558 * of a uni-directional data xfer. If the packet has 1559 * no control flags, is in-sequence, the window didn't 1560 * change and we're not retransmitting, it's a 1561 * candidate. If the length is zero and the ack moved 1562 * forward, we're the sender side of the xfer. Just 1563 * free the data acked & wake any higher level process 1564 * that was blocked waiting for space. If the length 1565 * is non-zero and the ack didn't move, we're the 1566 * receiver side. If we're getting packets in-order 1567 * (the reassembly queue is empty), add the data to 1568 * the socket buffer and note that we need a delayed ack. 1569 * Make sure that the hidden state-flags are also off. 1570 * Since we check for TCPS_ESTABLISHED first, it can only 1571 * be TH_NEEDSYN. 1572 */ 1573 if (tp->t_state == TCPS_ESTABLISHED && 1574 th->th_seq == tp->rcv_nxt && 1575 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1576 tp->snd_nxt == tp->snd_max && 1577 tiwin && tiwin == tp->snd_wnd && 1578 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1579 LIST_EMPTY(&tp->t_segq) && 1580 ((to.to_flags & TOF_TS) == 0 || 1581 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1582 1583 /* 1584 * If last ACK falls within this segment's sequence numbers, 1585 * record the timestamp. 1586 * NOTE that the test is modified according to the latest 1587 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1588 */ 1589 if ((to.to_flags & TOF_TS) != 0 && 1590 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1591 tp->ts_recent_age = ticks; 1592 tp->ts_recent = to.to_tsval; 1593 } 1594 1595 if (tlen == 0) { 1596 if (SEQ_GT(th->th_ack, tp->snd_una) && 1597 SEQ_LEQ(th->th_ack, tp->snd_max) && 1598 !IN_RECOVERY(tp->t_flags) && 1599 (to.to_flags & TOF_SACK) == 0 && 1600 TAILQ_EMPTY(&tp->snd_holes)) { 1601 /* 1602 * This is a pure ack for outstanding data. 1603 */ 1604 if (ti_locked == TI_RLOCKED) 1605 INP_INFO_RUNLOCK(&V_tcbinfo); 1606 else if (ti_locked == TI_WLOCKED) 1607 INP_INFO_WUNLOCK(&V_tcbinfo); 1608 else 1609 panic("%s: ti_locked %d on pure ACK", 1610 __func__, ti_locked); 1611 ti_locked = TI_UNLOCKED; 1612 1613 TCPSTAT_INC(tcps_predack); 1614 1615 /* 1616 * "bad retransmit" recovery. 1617 */ 1618 if (tp->t_rxtshift == 1 && 1619 tp->t_flags & TF_PREVVALID && 1620 (int)(ticks - tp->t_badrxtwin) < 0) { 1621 cc_cong_signal(tp, th, CC_RTO_ERR); 1622 } 1623 1624 /* 1625 * Recalculate the transmit timer / rtt. 1626 * 1627 * Some boxes send broken timestamp replies 1628 * during the SYN+ACK phase, ignore 1629 * timestamps of 0 or we could calculate a 1630 * huge RTT and blow up the retransmit timer. 1631 */ 1632 if ((to.to_flags & TOF_TS) != 0 && 1633 to.to_tsecr) { 1634 if (!tp->t_rttlow || 1635 tp->t_rttlow > ticks - to.to_tsecr) 1636 tp->t_rttlow = ticks - to.to_tsecr; 1637 tcp_xmit_timer(tp, 1638 ticks - to.to_tsecr + 1); 1639 } else if (tp->t_rtttime && 1640 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1641 if (!tp->t_rttlow || 1642 tp->t_rttlow > ticks - tp->t_rtttime) 1643 tp->t_rttlow = ticks - tp->t_rtttime; 1644 tcp_xmit_timer(tp, 1645 ticks - tp->t_rtttime); 1646 } 1647 acked = BYTES_THIS_ACK(tp, th); 1648 1649 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1650 hhook_run_tcp_est_in(tp, th, &to); 1651 1652 TCPSTAT_INC(tcps_rcvackpack); 1653 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1654 sbdrop(&so->so_snd, acked); 1655 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1656 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1657 tp->snd_recover = th->th_ack - 1; 1658 1659 /* 1660 * Let the congestion control algorithm update 1661 * congestion control related information. This 1662 * typically means increasing the congestion 1663 * window. 1664 */ 1665 cc_ack_received(tp, th, CC_ACK); 1666 1667 tp->snd_una = th->th_ack; 1668 /* 1669 * Pull snd_wl2 up to prevent seq wrap relative 1670 * to th_ack. 1671 */ 1672 tp->snd_wl2 = th->th_ack; 1673 tp->t_dupacks = 0; 1674 m_freem(m); 1675 ND6_HINT(tp); /* Some progress has been made. */ 1676 1677 /* 1678 * If all outstanding data are acked, stop 1679 * retransmit timer, otherwise restart timer 1680 * using current (possibly backed-off) value. 1681 * If process is waiting for space, 1682 * wakeup/selwakeup/signal. If data 1683 * are ready to send, let tcp_output 1684 * decide between more output or persist. 1685 */ 1686 #ifdef TCPDEBUG 1687 if (so->so_options & SO_DEBUG) 1688 tcp_trace(TA_INPUT, ostate, tp, 1689 (void *)tcp_saveipgen, 1690 &tcp_savetcp, 0); 1691 #endif 1692 if (tp->snd_una == tp->snd_max) 1693 tcp_timer_activate(tp, TT_REXMT, 0); 1694 else if (!tcp_timer_active(tp, TT_PERSIST)) 1695 tcp_timer_activate(tp, TT_REXMT, 1696 tp->t_rxtcur); 1697 sowwakeup(so); 1698 if (so->so_snd.sb_cc) 1699 (void) tcp_output(tp); 1700 goto check_delack; 1701 } 1702 } else if (th->th_ack == tp->snd_una && 1703 tlen <= sbspace(&so->so_rcv)) { 1704 int newsize = 0; /* automatic sockbuf scaling */ 1705 1706 /* 1707 * This is a pure, in-sequence data packet with 1708 * nothing on the reassembly queue and we have enough 1709 * buffer space to take it. 1710 */ 1711 if (ti_locked == TI_RLOCKED) 1712 INP_INFO_RUNLOCK(&V_tcbinfo); 1713 else if (ti_locked == TI_WLOCKED) 1714 INP_INFO_WUNLOCK(&V_tcbinfo); 1715 else 1716 panic("%s: ti_locked %d on pure data " 1717 "segment", __func__, ti_locked); 1718 ti_locked = TI_UNLOCKED; 1719 1720 /* Clean receiver SACK report if present */ 1721 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1722 tcp_clean_sackreport(tp); 1723 TCPSTAT_INC(tcps_preddat); 1724 tp->rcv_nxt += tlen; 1725 /* 1726 * Pull snd_wl1 up to prevent seq wrap relative to 1727 * th_seq. 1728 */ 1729 tp->snd_wl1 = th->th_seq; 1730 /* 1731 * Pull rcv_up up to prevent seq wrap relative to 1732 * rcv_nxt. 1733 */ 1734 tp->rcv_up = tp->rcv_nxt; 1735 TCPSTAT_INC(tcps_rcvpack); 1736 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1737 ND6_HINT(tp); /* Some progress has been made */ 1738 #ifdef TCPDEBUG 1739 if (so->so_options & SO_DEBUG) 1740 tcp_trace(TA_INPUT, ostate, tp, 1741 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1742 #endif 1743 /* 1744 * Automatic sizing of receive socket buffer. Often the send 1745 * buffer size is not optimally adjusted to the actual network 1746 * conditions at hand (delay bandwidth product). Setting the 1747 * buffer size too small limits throughput on links with high 1748 * bandwidth and high delay (eg. trans-continental/oceanic links). 1749 * 1750 * On the receive side the socket buffer memory is only rarely 1751 * used to any significant extent. This allows us to be much 1752 * more aggressive in scaling the receive socket buffer. For 1753 * the case that the buffer space is actually used to a large 1754 * extent and we run out of kernel memory we can simply drop 1755 * the new segments; TCP on the sender will just retransmit it 1756 * later. Setting the buffer size too big may only consume too 1757 * much kernel memory if the application doesn't read() from 1758 * the socket or packet loss or reordering makes use of the 1759 * reassembly queue. 1760 * 1761 * The criteria to step up the receive buffer one notch are: 1762 * 1. the number of bytes received during the time it takes 1763 * one timestamp to be reflected back to us (the RTT); 1764 * 2. received bytes per RTT is within seven eighth of the 1765 * current socket buffer size; 1766 * 3. receive buffer size has not hit maximal automatic size; 1767 * 1768 * This algorithm does one step per RTT at most and only if 1769 * we receive a bulk stream w/o packet losses or reorderings. 1770 * Shrinking the buffer during idle times is not necessary as 1771 * it doesn't consume any memory when idle. 1772 * 1773 * TODO: Only step up if the application is actually serving 1774 * the buffer to better manage the socket buffer resources. 1775 */ 1776 if (V_tcp_do_autorcvbuf && 1777 to.to_tsecr && 1778 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1779 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1780 to.to_tsecr - tp->rfbuf_ts < hz) { 1781 if (tp->rfbuf_cnt > 1782 (so->so_rcv.sb_hiwat / 8 * 7) && 1783 so->so_rcv.sb_hiwat < 1784 V_tcp_autorcvbuf_max) { 1785 newsize = 1786 min(so->so_rcv.sb_hiwat + 1787 V_tcp_autorcvbuf_inc, 1788 V_tcp_autorcvbuf_max); 1789 } 1790 /* Start over with next RTT. */ 1791 tp->rfbuf_ts = 0; 1792 tp->rfbuf_cnt = 0; 1793 } else 1794 tp->rfbuf_cnt += tlen; /* add up */ 1795 } 1796 1797 /* Add data to socket buffer. */ 1798 SOCKBUF_LOCK(&so->so_rcv); 1799 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1800 m_freem(m); 1801 } else { 1802 /* 1803 * Set new socket buffer size. 1804 * Give up when limit is reached. 1805 */ 1806 if (newsize) 1807 if (!sbreserve_locked(&so->so_rcv, 1808 newsize, so, NULL)) 1809 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1810 m_adj(m, drop_hdrlen); /* delayed header drop */ 1811 sbappendstream_locked(&so->so_rcv, m); 1812 } 1813 /* NB: sorwakeup_locked() does an implicit unlock. */ 1814 sorwakeup_locked(so); 1815 if (DELAY_ACK(tp)) { 1816 tp->t_flags |= TF_DELACK; 1817 } else { 1818 tp->t_flags |= TF_ACKNOW; 1819 tcp_output(tp); 1820 } 1821 goto check_delack; 1822 } 1823 } 1824 1825 /* 1826 * Calculate amount of space in receive window, 1827 * and then do TCP input processing. 1828 * Receive window is amount of space in rcv queue, 1829 * but not less than advertised window. 1830 */ 1831 win = sbspace(&so->so_rcv); 1832 if (win < 0) 1833 win = 0; 1834 KASSERT(SEQ_GEQ(tp->rcv_adv, tp->rcv_nxt), 1835 ("tcp_input negative window: tp %p rcv_nxt %u rcv_adv %u", tp, 1836 tp->rcv_nxt, tp->rcv_adv)); 1837 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1838 1839 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1840 tp->rfbuf_ts = 0; 1841 tp->rfbuf_cnt = 0; 1842 1843 switch (tp->t_state) { 1844 1845 /* 1846 * If the state is SYN_RECEIVED: 1847 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1848 */ 1849 case TCPS_SYN_RECEIVED: 1850 if ((thflags & TH_ACK) && 1851 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1852 SEQ_GT(th->th_ack, tp->snd_max))) { 1853 rstreason = BANDLIM_RST_OPENPORT; 1854 goto dropwithreset; 1855 } 1856 break; 1857 1858 /* 1859 * If the state is SYN_SENT: 1860 * if seg contains an ACK, but not for our SYN, drop the input. 1861 * if seg contains a RST, then drop the connection. 1862 * if seg does not contain SYN, then drop it. 1863 * Otherwise this is an acceptable SYN segment 1864 * initialize tp->rcv_nxt and tp->irs 1865 * if seg contains ack then advance tp->snd_una 1866 * if seg contains an ECE and ECN support is enabled, the stream 1867 * is ECN capable. 1868 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1869 * arrange for segment to be acked (eventually) 1870 * continue processing rest of data/controls, beginning with URG 1871 */ 1872 case TCPS_SYN_SENT: 1873 if ((thflags & TH_ACK) && 1874 (SEQ_LEQ(th->th_ack, tp->iss) || 1875 SEQ_GT(th->th_ack, tp->snd_max))) { 1876 rstreason = BANDLIM_UNLIMITED; 1877 goto dropwithreset; 1878 } 1879 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1880 tp = tcp_drop(tp, ECONNREFUSED); 1881 if (thflags & TH_RST) 1882 goto drop; 1883 if (!(thflags & TH_SYN)) 1884 goto drop; 1885 1886 tp->irs = th->th_seq; 1887 tcp_rcvseqinit(tp); 1888 if (thflags & TH_ACK) { 1889 TCPSTAT_INC(tcps_connects); 1890 soisconnected(so); 1891 #ifdef MAC 1892 mac_socketpeer_set_from_mbuf(m, so); 1893 #endif 1894 /* Do window scaling on this connection? */ 1895 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1896 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1897 tp->rcv_scale = tp->request_r_scale; 1898 } 1899 tp->rcv_adv += imin(tp->rcv_wnd, 1900 TCP_MAXWIN << tp->rcv_scale); 1901 tp->snd_una++; /* SYN is acked */ 1902 /* 1903 * If there's data, delay ACK; if there's also a FIN 1904 * ACKNOW will be turned on later. 1905 */ 1906 if (DELAY_ACK(tp) && tlen != 0) 1907 tcp_timer_activate(tp, TT_DELACK, 1908 tcp_delacktime); 1909 else 1910 tp->t_flags |= TF_ACKNOW; 1911 1912 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1913 tp->t_flags |= TF_ECN_PERMIT; 1914 TCPSTAT_INC(tcps_ecn_shs); 1915 } 1916 1917 /* 1918 * Received <SYN,ACK> in SYN_SENT[*] state. 1919 * Transitions: 1920 * SYN_SENT --> ESTABLISHED 1921 * SYN_SENT* --> FIN_WAIT_1 1922 */ 1923 tp->t_starttime = ticks; 1924 if (tp->t_flags & TF_NEEDFIN) { 1925 tp->t_state = TCPS_FIN_WAIT_1; 1926 tp->t_flags &= ~TF_NEEDFIN; 1927 thflags &= ~TH_SYN; 1928 } else { 1929 tp->t_state = TCPS_ESTABLISHED; 1930 cc_conn_init(tp); 1931 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1932 } 1933 } else { 1934 /* 1935 * Received initial SYN in SYN-SENT[*] state => 1936 * simultaneous open. If segment contains CC option 1937 * and there is a cached CC, apply TAO test. 1938 * If it succeeds, connection is * half-synchronized. 1939 * Otherwise, do 3-way handshake: 1940 * SYN-SENT -> SYN-RECEIVED 1941 * SYN-SENT* -> SYN-RECEIVED* 1942 * If there was no CC option, clear cached CC value. 1943 */ 1944 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1945 tcp_timer_activate(tp, TT_REXMT, 0); 1946 tp->t_state = TCPS_SYN_RECEIVED; 1947 } 1948 1949 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1950 "ti_locked %d", __func__, ti_locked)); 1951 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1952 INP_WLOCK_ASSERT(tp->t_inpcb); 1953 1954 /* 1955 * Advance th->th_seq to correspond to first data byte. 1956 * If data, trim to stay within window, 1957 * dropping FIN if necessary. 1958 */ 1959 th->th_seq++; 1960 if (tlen > tp->rcv_wnd) { 1961 todrop = tlen - tp->rcv_wnd; 1962 m_adj(m, -todrop); 1963 tlen = tp->rcv_wnd; 1964 thflags &= ~TH_FIN; 1965 TCPSTAT_INC(tcps_rcvpackafterwin); 1966 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1967 } 1968 tp->snd_wl1 = th->th_seq - 1; 1969 tp->rcv_up = th->th_seq; 1970 /* 1971 * Client side of transaction: already sent SYN and data. 1972 * If the remote host used T/TCP to validate the SYN, 1973 * our data will be ACK'd; if so, enter normal data segment 1974 * processing in the middle of step 5, ack processing. 1975 * Otherwise, goto step 6. 1976 */ 1977 if (thflags & TH_ACK) 1978 goto process_ACK; 1979 1980 goto step6; 1981 1982 /* 1983 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1984 * do normal processing. 1985 * 1986 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1987 */ 1988 case TCPS_LAST_ACK: 1989 case TCPS_CLOSING: 1990 break; /* continue normal processing */ 1991 } 1992 1993 /* 1994 * States other than LISTEN or SYN_SENT. 1995 * First check the RST flag and sequence number since reset segments 1996 * are exempt from the timestamp and connection count tests. This 1997 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1998 * below which allowed reset segments in half the sequence space 1999 * to fall though and be processed (which gives forged reset 2000 * segments with a random sequence number a 50 percent chance of 2001 * killing a connection). 2002 * Then check timestamp, if present. 2003 * Then check the connection count, if present. 2004 * Then check that at least some bytes of segment are within 2005 * receive window. If segment begins before rcv_nxt, 2006 * drop leading data (and SYN); if nothing left, just ack. 2007 * 2008 * 2009 * If the RST bit is set, check the sequence number to see 2010 * if this is a valid reset segment. 2011 * RFC 793 page 37: 2012 * In all states except SYN-SENT, all reset (RST) segments 2013 * are validated by checking their SEQ-fields. A reset is 2014 * valid if its sequence number is in the window. 2015 * Note: this does not take into account delayed ACKs, so 2016 * we should test against last_ack_sent instead of rcv_nxt. 2017 * The sequence number in the reset segment is normally an 2018 * echo of our outgoing acknowlegement numbers, but some hosts 2019 * send a reset with the sequence number at the rightmost edge 2020 * of our receive window, and we have to handle this case. 2021 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2022 * that brute force RST attacks are possible. To combat this, 2023 * we use a much stricter check while in the ESTABLISHED state, 2024 * only accepting RSTs where the sequence number is equal to 2025 * last_ack_sent. In all other states (the states in which a 2026 * RST is more likely), the more permissive check is used. 2027 * If we have multiple segments in flight, the initial reset 2028 * segment sequence numbers will be to the left of last_ack_sent, 2029 * but they will eventually catch up. 2030 * In any case, it never made sense to trim reset segments to 2031 * fit the receive window since RFC 1122 says: 2032 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2033 * 2034 * A TCP SHOULD allow a received RST segment to include data. 2035 * 2036 * DISCUSSION 2037 * It has been suggested that a RST segment could contain 2038 * ASCII text that encoded and explained the cause of the 2039 * RST. No standard has yet been established for such 2040 * data. 2041 * 2042 * If the reset segment passes the sequence number test examine 2043 * the state: 2044 * SYN_RECEIVED STATE: 2045 * If passive open, return to LISTEN state. 2046 * If active open, inform user that connection was refused. 2047 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2048 * Inform user that connection was reset, and close tcb. 2049 * CLOSING, LAST_ACK STATES: 2050 * Close the tcb. 2051 * TIME_WAIT STATE: 2052 * Drop the segment - see Stevens, vol. 2, p. 964 and 2053 * RFC 1337. 2054 */ 2055 if (thflags & TH_RST) { 2056 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2057 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2058 switch (tp->t_state) { 2059 2060 case TCPS_SYN_RECEIVED: 2061 so->so_error = ECONNREFUSED; 2062 goto close; 2063 2064 case TCPS_ESTABLISHED: 2065 if (V_tcp_insecure_rst == 0 && 2066 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2067 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2068 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2069 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2070 TCPSTAT_INC(tcps_badrst); 2071 goto drop; 2072 } 2073 /* FALLTHROUGH */ 2074 case TCPS_FIN_WAIT_1: 2075 case TCPS_FIN_WAIT_2: 2076 case TCPS_CLOSE_WAIT: 2077 so->so_error = ECONNRESET; 2078 close: 2079 KASSERT(ti_locked == TI_WLOCKED, 2080 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2081 ti_locked)); 2082 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2083 2084 tp->t_state = TCPS_CLOSED; 2085 TCPSTAT_INC(tcps_drops); 2086 tp = tcp_close(tp); 2087 break; 2088 2089 case TCPS_CLOSING: 2090 case TCPS_LAST_ACK: 2091 KASSERT(ti_locked == TI_WLOCKED, 2092 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2093 ti_locked)); 2094 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2095 2096 tp = tcp_close(tp); 2097 break; 2098 } 2099 } 2100 goto drop; 2101 } 2102 2103 /* 2104 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2105 * and it's less than ts_recent, drop it. 2106 */ 2107 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2108 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2109 2110 /* Check to see if ts_recent is over 24 days old. */ 2111 if (ticks - tp->ts_recent_age > TCP_PAWS_IDLE) { 2112 /* 2113 * Invalidate ts_recent. If this segment updates 2114 * ts_recent, the age will be reset later and ts_recent 2115 * will get a valid value. If it does not, setting 2116 * ts_recent to zero will at least satisfy the 2117 * requirement that zero be placed in the timestamp 2118 * echo reply when ts_recent isn't valid. The 2119 * age isn't reset until we get a valid ts_recent 2120 * because we don't want out-of-order segments to be 2121 * dropped when ts_recent is old. 2122 */ 2123 tp->ts_recent = 0; 2124 } else { 2125 TCPSTAT_INC(tcps_rcvduppack); 2126 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2127 TCPSTAT_INC(tcps_pawsdrop); 2128 if (tlen) 2129 goto dropafterack; 2130 goto drop; 2131 } 2132 } 2133 2134 /* 2135 * In the SYN-RECEIVED state, validate that the packet belongs to 2136 * this connection before trimming the data to fit the receive 2137 * window. Check the sequence number versus IRS since we know 2138 * the sequence numbers haven't wrapped. This is a partial fix 2139 * for the "LAND" DoS attack. 2140 */ 2141 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2142 rstreason = BANDLIM_RST_OPENPORT; 2143 goto dropwithreset; 2144 } 2145 2146 todrop = tp->rcv_nxt - th->th_seq; 2147 if (todrop > 0) { 2148 /* 2149 * If this is a duplicate SYN for our current connection, 2150 * advance over it and pretend and it's not a SYN. 2151 */ 2152 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2153 thflags &= ~TH_SYN; 2154 th->th_seq++; 2155 if (th->th_urp > 1) 2156 th->th_urp--; 2157 else 2158 thflags &= ~TH_URG; 2159 todrop--; 2160 } 2161 /* 2162 * Following if statement from Stevens, vol. 2, p. 960. 2163 */ 2164 if (todrop > tlen 2165 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2166 /* 2167 * Any valid FIN must be to the left of the window. 2168 * At this point the FIN must be a duplicate or out 2169 * of sequence; drop it. 2170 */ 2171 thflags &= ~TH_FIN; 2172 2173 /* 2174 * Send an ACK to resynchronize and drop any data. 2175 * But keep on processing for RST or ACK. 2176 */ 2177 tp->t_flags |= TF_ACKNOW; 2178 todrop = tlen; 2179 TCPSTAT_INC(tcps_rcvduppack); 2180 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2181 } else { 2182 TCPSTAT_INC(tcps_rcvpartduppack); 2183 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2184 } 2185 drop_hdrlen += todrop; /* drop from the top afterwards */ 2186 th->th_seq += todrop; 2187 tlen -= todrop; 2188 if (th->th_urp > todrop) 2189 th->th_urp -= todrop; 2190 else { 2191 thflags &= ~TH_URG; 2192 th->th_urp = 0; 2193 } 2194 } 2195 2196 /* 2197 * If new data are received on a connection after the 2198 * user processes are gone, then RST the other end. 2199 */ 2200 if ((so->so_state & SS_NOFDREF) && 2201 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2202 char *s; 2203 2204 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2205 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2206 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2207 2208 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 2209 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 2210 "was closed, sending RST and removing tcpcb\n", 2211 s, __func__, tcpstates[tp->t_state], tlen); 2212 free(s, M_TCPLOG); 2213 } 2214 tp = tcp_close(tp); 2215 TCPSTAT_INC(tcps_rcvafterclose); 2216 rstreason = BANDLIM_UNLIMITED; 2217 goto dropwithreset; 2218 } 2219 2220 /* 2221 * If segment ends after window, drop trailing data 2222 * (and PUSH and FIN); if nothing left, just ACK. 2223 */ 2224 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2225 if (todrop > 0) { 2226 TCPSTAT_INC(tcps_rcvpackafterwin); 2227 if (todrop >= tlen) { 2228 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2229 /* 2230 * If window is closed can only take segments at 2231 * window edge, and have to drop data and PUSH from 2232 * incoming segments. Continue processing, but 2233 * remember to ack. Otherwise, drop segment 2234 * and ack. 2235 */ 2236 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2237 tp->t_flags |= TF_ACKNOW; 2238 TCPSTAT_INC(tcps_rcvwinprobe); 2239 } else 2240 goto dropafterack; 2241 } else 2242 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2243 m_adj(m, -todrop); 2244 tlen -= todrop; 2245 thflags &= ~(TH_PUSH|TH_FIN); 2246 } 2247 2248 /* 2249 * If last ACK falls within this segment's sequence numbers, 2250 * record its timestamp. 2251 * NOTE: 2252 * 1) That the test incorporates suggestions from the latest 2253 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2254 * 2) That updating only on newer timestamps interferes with 2255 * our earlier PAWS tests, so this check should be solely 2256 * predicated on the sequence space of this segment. 2257 * 3) That we modify the segment boundary check to be 2258 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2259 * instead of RFC1323's 2260 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2261 * This modified check allows us to overcome RFC1323's 2262 * limitations as described in Stevens TCP/IP Illustrated 2263 * Vol. 2 p.869. In such cases, we can still calculate the 2264 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2265 */ 2266 if ((to.to_flags & TOF_TS) != 0 && 2267 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2268 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2269 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2270 tp->ts_recent_age = ticks; 2271 tp->ts_recent = to.to_tsval; 2272 } 2273 2274 /* 2275 * If a SYN is in the window, then this is an 2276 * error and we send an RST and drop the connection. 2277 */ 2278 if (thflags & TH_SYN) { 2279 KASSERT(ti_locked == TI_WLOCKED, 2280 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2281 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2282 2283 tp = tcp_drop(tp, ECONNRESET); 2284 rstreason = BANDLIM_UNLIMITED; 2285 goto drop; 2286 } 2287 2288 /* 2289 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2290 * flag is on (half-synchronized state), then queue data for 2291 * later processing; else drop segment and return. 2292 */ 2293 if ((thflags & TH_ACK) == 0) { 2294 if (tp->t_state == TCPS_SYN_RECEIVED || 2295 (tp->t_flags & TF_NEEDSYN)) 2296 goto step6; 2297 else if (tp->t_flags & TF_ACKNOW) 2298 goto dropafterack; 2299 else 2300 goto drop; 2301 } 2302 2303 /* 2304 * Ack processing. 2305 */ 2306 switch (tp->t_state) { 2307 2308 /* 2309 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2310 * ESTABLISHED state and continue processing. 2311 * The ACK was checked above. 2312 */ 2313 case TCPS_SYN_RECEIVED: 2314 2315 TCPSTAT_INC(tcps_connects); 2316 soisconnected(so); 2317 /* Do window scaling? */ 2318 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2319 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2320 tp->rcv_scale = tp->request_r_scale; 2321 tp->snd_wnd = tiwin; 2322 } 2323 /* 2324 * Make transitions: 2325 * SYN-RECEIVED -> ESTABLISHED 2326 * SYN-RECEIVED* -> FIN-WAIT-1 2327 */ 2328 tp->t_starttime = ticks; 2329 if (tp->t_flags & TF_NEEDFIN) { 2330 tp->t_state = TCPS_FIN_WAIT_1; 2331 tp->t_flags &= ~TF_NEEDFIN; 2332 } else { 2333 tp->t_state = TCPS_ESTABLISHED; 2334 cc_conn_init(tp); 2335 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 2336 } 2337 /* 2338 * If segment contains data or ACK, will call tcp_reass() 2339 * later; if not, do so now to pass queued data to user. 2340 */ 2341 if (tlen == 0 && (thflags & TH_FIN) == 0) 2342 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2343 (struct mbuf *)0); 2344 tp->snd_wl1 = th->th_seq - 1; 2345 /* FALLTHROUGH */ 2346 2347 /* 2348 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2349 * ACKs. If the ack is in the range 2350 * tp->snd_una < th->th_ack <= tp->snd_max 2351 * then advance tp->snd_una to th->th_ack and drop 2352 * data from the retransmission queue. If this ACK reflects 2353 * more up to date window information we update our window information. 2354 */ 2355 case TCPS_ESTABLISHED: 2356 case TCPS_FIN_WAIT_1: 2357 case TCPS_FIN_WAIT_2: 2358 case TCPS_CLOSE_WAIT: 2359 case TCPS_CLOSING: 2360 case TCPS_LAST_ACK: 2361 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2362 TCPSTAT_INC(tcps_rcvacktoomuch); 2363 goto dropafterack; 2364 } 2365 if ((tp->t_flags & TF_SACK_PERMIT) && 2366 ((to.to_flags & TOF_SACK) || 2367 !TAILQ_EMPTY(&tp->snd_holes))) 2368 tcp_sack_doack(tp, &to, th->th_ack); 2369 2370 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2371 hhook_run_tcp_est_in(tp, th, &to); 2372 2373 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2374 if (tlen == 0 && tiwin == tp->snd_wnd) { 2375 TCPSTAT_INC(tcps_rcvdupack); 2376 /* 2377 * If we have outstanding data (other than 2378 * a window probe), this is a completely 2379 * duplicate ack (ie, window info didn't 2380 * change), the ack is the biggest we've 2381 * seen and we've seen exactly our rexmt 2382 * threshhold of them, assume a packet 2383 * has been dropped and retransmit it. 2384 * Kludge snd_nxt & the congestion 2385 * window so we send only this one 2386 * packet. 2387 * 2388 * We know we're losing at the current 2389 * window size so do congestion avoidance 2390 * (set ssthresh to half the current window 2391 * and pull our congestion window back to 2392 * the new ssthresh). 2393 * 2394 * Dup acks mean that packets have left the 2395 * network (they're now cached at the receiver) 2396 * so bump cwnd by the amount in the receiver 2397 * to keep a constant cwnd packets in the 2398 * network. 2399 * 2400 * When using TCP ECN, notify the peer that 2401 * we reduced the cwnd. 2402 */ 2403 if (!tcp_timer_active(tp, TT_REXMT) || 2404 th->th_ack != tp->snd_una) 2405 tp->t_dupacks = 0; 2406 else if (++tp->t_dupacks > tcprexmtthresh || 2407 IN_FASTRECOVERY(tp->t_flags)) { 2408 cc_ack_received(tp, th, CC_DUPACK); 2409 if ((tp->t_flags & TF_SACK_PERMIT) && 2410 IN_FASTRECOVERY(tp->t_flags)) { 2411 int awnd; 2412 2413 /* 2414 * Compute the amount of data in flight first. 2415 * We can inject new data into the pipe iff 2416 * we have less than 1/2 the original window's 2417 * worth of data in flight. 2418 */ 2419 awnd = (tp->snd_nxt - tp->snd_fack) + 2420 tp->sackhint.sack_bytes_rexmit; 2421 if (awnd < tp->snd_ssthresh) { 2422 tp->snd_cwnd += tp->t_maxseg; 2423 if (tp->snd_cwnd > tp->snd_ssthresh) 2424 tp->snd_cwnd = tp->snd_ssthresh; 2425 } 2426 } else 2427 tp->snd_cwnd += tp->t_maxseg; 2428 (void) tcp_output(tp); 2429 goto drop; 2430 } else if (tp->t_dupacks == tcprexmtthresh) { 2431 tcp_seq onxt = tp->snd_nxt; 2432 2433 /* 2434 * If we're doing sack, check to 2435 * see if we're already in sack 2436 * recovery. If we're not doing sack, 2437 * check to see if we're in newreno 2438 * recovery. 2439 */ 2440 if (tp->t_flags & TF_SACK_PERMIT) { 2441 if (IN_FASTRECOVERY(tp->t_flags)) { 2442 tp->t_dupacks = 0; 2443 break; 2444 } 2445 } else { 2446 if (SEQ_LEQ(th->th_ack, 2447 tp->snd_recover)) { 2448 tp->t_dupacks = 0; 2449 break; 2450 } 2451 } 2452 /* Congestion signal before ack. */ 2453 cc_cong_signal(tp, th, CC_NDUPACK); 2454 cc_ack_received(tp, th, CC_DUPACK); 2455 tcp_timer_activate(tp, TT_REXMT, 0); 2456 tp->t_rtttime = 0; 2457 if (tp->t_flags & TF_SACK_PERMIT) { 2458 TCPSTAT_INC( 2459 tcps_sack_recovery_episode); 2460 tp->sack_newdata = tp->snd_nxt; 2461 tp->snd_cwnd = tp->t_maxseg; 2462 (void) tcp_output(tp); 2463 goto drop; 2464 } 2465 tp->snd_nxt = th->th_ack; 2466 tp->snd_cwnd = tp->t_maxseg; 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 (void) tcp_output(tp); 2494 sent = tp->snd_max - oldsndmax; 2495 if (sent > tp->t_maxseg) { 2496 KASSERT((tp->t_dupacks == 2 && 2497 tp->snd_limited == 0) || 2498 (sent == tp->t_maxseg + 1 && 2499 tp->t_flags & TF_SENTFIN), 2500 ("%s: sent too much", 2501 __func__)); 2502 tp->snd_limited = 2; 2503 } else if (sent > 0) 2504 ++tp->snd_limited; 2505 tp->snd_cwnd = oldcwnd; 2506 goto drop; 2507 } 2508 } else 2509 tp->t_dupacks = 0; 2510 break; 2511 } 2512 2513 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2514 ("%s: th_ack <= snd_una", __func__)); 2515 2516 /* 2517 * If the congestion window was inflated to account 2518 * for the other side's cached packets, retract it. 2519 */ 2520 if (IN_FASTRECOVERY(tp->t_flags)) { 2521 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2522 if (tp->t_flags & TF_SACK_PERMIT) 2523 tcp_sack_partialack(tp, th); 2524 else 2525 tcp_newreno_partial_ack(tp, th); 2526 } else 2527 cc_post_recovery(tp, th); 2528 } 2529 tp->t_dupacks = 0; 2530 /* 2531 * If we reach this point, ACK is not a duplicate, 2532 * i.e., it ACKs something we sent. 2533 */ 2534 if (tp->t_flags & TF_NEEDSYN) { 2535 /* 2536 * T/TCP: Connection was half-synchronized, and our 2537 * SYN has been ACK'd (so connection is now fully 2538 * synchronized). Go to non-starred state, 2539 * increment snd_una for ACK of SYN, and check if 2540 * we can do window scaling. 2541 */ 2542 tp->t_flags &= ~TF_NEEDSYN; 2543 tp->snd_una++; 2544 /* Do window scaling? */ 2545 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2546 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2547 tp->rcv_scale = tp->request_r_scale; 2548 /* Send window already scaled. */ 2549 } 2550 } 2551 2552 process_ACK: 2553 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2554 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2555 ("tcp_input: process_ACK ti_locked %d", ti_locked)); 2556 INP_WLOCK_ASSERT(tp->t_inpcb); 2557 2558 acked = BYTES_THIS_ACK(tp, th); 2559 TCPSTAT_INC(tcps_rcvackpack); 2560 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2561 2562 /* 2563 * If we just performed our first retransmit, and the ACK 2564 * arrives within our recovery window, then it was a mistake 2565 * to do the retransmit in the first place. Recover our 2566 * original cwnd and ssthresh, and proceed to transmit where 2567 * we left off. 2568 */ 2569 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2570 (int)(ticks - tp->t_badrxtwin) < 0) 2571 cc_cong_signal(tp, th, CC_RTO_ERR); 2572 2573 /* 2574 * If we have a timestamp reply, update smoothed 2575 * round trip time. If no timestamp is present but 2576 * transmit timer is running and timed sequence 2577 * number was acked, update smoothed round trip time. 2578 * Since we now have an rtt measurement, cancel the 2579 * timer backoff (cf., Phil Karn's retransmit alg.). 2580 * Recompute the initial retransmit timer. 2581 * 2582 * Some boxes send broken timestamp replies 2583 * during the SYN+ACK phase, ignore 2584 * timestamps of 0 or we could calculate a 2585 * huge RTT and blow up the retransmit timer. 2586 */ 2587 if ((to.to_flags & TOF_TS) != 0 && 2588 to.to_tsecr) { 2589 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2590 tp->t_rttlow = ticks - to.to_tsecr; 2591 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2592 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2593 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2594 tp->t_rttlow = ticks - tp->t_rtttime; 2595 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2596 } 2597 2598 /* 2599 * If all outstanding data is acked, stop retransmit 2600 * timer and remember to restart (more output or persist). 2601 * If there is more data to be acked, restart retransmit 2602 * timer, using current (possibly backed-off) value. 2603 */ 2604 if (th->th_ack == tp->snd_max) { 2605 tcp_timer_activate(tp, TT_REXMT, 0); 2606 needoutput = 1; 2607 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2608 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2609 2610 /* 2611 * If no data (only SYN) was ACK'd, 2612 * skip rest of ACK processing. 2613 */ 2614 if (acked == 0) 2615 goto step6; 2616 2617 /* 2618 * Let the congestion control algorithm update congestion 2619 * control related information. This typically means increasing 2620 * the congestion window. 2621 */ 2622 cc_ack_received(tp, th, CC_ACK); 2623 2624 SOCKBUF_LOCK(&so->so_snd); 2625 if (acked > so->so_snd.sb_cc) { 2626 tp->snd_wnd -= so->so_snd.sb_cc; 2627 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2628 ourfinisacked = 1; 2629 } else { 2630 sbdrop_locked(&so->so_snd, acked); 2631 tp->snd_wnd -= acked; 2632 ourfinisacked = 0; 2633 } 2634 /* NB: sowwakeup_locked() does an implicit unlock. */ 2635 sowwakeup_locked(so); 2636 /* Detect una wraparound. */ 2637 if (!IN_RECOVERY(tp->t_flags) && 2638 SEQ_GT(tp->snd_una, tp->snd_recover) && 2639 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2640 tp->snd_recover = th->th_ack - 1; 2641 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2642 if (IN_RECOVERY(tp->t_flags) && 2643 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2644 EXIT_RECOVERY(tp->t_flags); 2645 } 2646 tp->snd_una = th->th_ack; 2647 if (tp->t_flags & TF_SACK_PERMIT) { 2648 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2649 tp->snd_recover = tp->snd_una; 2650 } 2651 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2652 tp->snd_nxt = tp->snd_una; 2653 2654 switch (tp->t_state) { 2655 2656 /* 2657 * In FIN_WAIT_1 STATE in addition to the processing 2658 * for the ESTABLISHED state if our FIN is now acknowledged 2659 * then enter FIN_WAIT_2. 2660 */ 2661 case TCPS_FIN_WAIT_1: 2662 if (ourfinisacked) { 2663 /* 2664 * If we can't receive any more 2665 * data, then closing user can proceed. 2666 * Starting the timer is contrary to the 2667 * specification, but if we don't get a FIN 2668 * we'll hang forever. 2669 * 2670 * XXXjl: 2671 * we should release the tp also, and use a 2672 * compressed state. 2673 */ 2674 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2675 int timeout; 2676 2677 soisdisconnected(so); 2678 timeout = (tcp_fast_finwait2_recycle) ? 2679 tcp_finwait2_timeout : tcp_maxidle; 2680 tcp_timer_activate(tp, TT_2MSL, timeout); 2681 } 2682 tp->t_state = TCPS_FIN_WAIT_2; 2683 } 2684 break; 2685 2686 /* 2687 * In CLOSING STATE in addition to the processing for 2688 * the ESTABLISHED state if the ACK acknowledges our FIN 2689 * then enter the TIME-WAIT state, otherwise ignore 2690 * the segment. 2691 */ 2692 case TCPS_CLOSING: 2693 if (ourfinisacked) { 2694 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2695 tcp_twstart(tp); 2696 INP_INFO_WUNLOCK(&V_tcbinfo); 2697 m_freem(m); 2698 return; 2699 } 2700 break; 2701 2702 /* 2703 * In LAST_ACK, we may still be waiting for data to drain 2704 * and/or to be acked, as well as for the ack of our FIN. 2705 * If our FIN is now acknowledged, delete the TCB, 2706 * enter the closed state and return. 2707 */ 2708 case TCPS_LAST_ACK: 2709 if (ourfinisacked) { 2710 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2711 tp = tcp_close(tp); 2712 goto drop; 2713 } 2714 break; 2715 } 2716 } 2717 2718 step6: 2719 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2720 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2721 ("tcp_do_segment: step6 ti_locked %d", ti_locked)); 2722 INP_WLOCK_ASSERT(tp->t_inpcb); 2723 2724 /* 2725 * Update window information. 2726 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2727 */ 2728 if ((thflags & TH_ACK) && 2729 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2730 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2731 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2732 /* keep track of pure window updates */ 2733 if (tlen == 0 && 2734 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2735 TCPSTAT_INC(tcps_rcvwinupd); 2736 tp->snd_wnd = tiwin; 2737 tp->snd_wl1 = th->th_seq; 2738 tp->snd_wl2 = th->th_ack; 2739 if (tp->snd_wnd > tp->max_sndwnd) 2740 tp->max_sndwnd = tp->snd_wnd; 2741 needoutput = 1; 2742 } 2743 2744 /* 2745 * Process segments with URG. 2746 */ 2747 if ((thflags & TH_URG) && th->th_urp && 2748 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2749 /* 2750 * This is a kludge, but if we receive and accept 2751 * random urgent pointers, we'll crash in 2752 * soreceive. It's hard to imagine someone 2753 * actually wanting to send this much urgent data. 2754 */ 2755 SOCKBUF_LOCK(&so->so_rcv); 2756 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2757 th->th_urp = 0; /* XXX */ 2758 thflags &= ~TH_URG; /* XXX */ 2759 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2760 goto dodata; /* XXX */ 2761 } 2762 /* 2763 * If this segment advances the known urgent pointer, 2764 * then mark the data stream. This should not happen 2765 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2766 * a FIN has been received from the remote side. 2767 * In these states we ignore the URG. 2768 * 2769 * According to RFC961 (Assigned Protocols), 2770 * the urgent pointer points to the last octet 2771 * of urgent data. We continue, however, 2772 * to consider it to indicate the first octet 2773 * of data past the urgent section as the original 2774 * spec states (in one of two places). 2775 */ 2776 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2777 tp->rcv_up = th->th_seq + th->th_urp; 2778 so->so_oobmark = so->so_rcv.sb_cc + 2779 (tp->rcv_up - tp->rcv_nxt) - 1; 2780 if (so->so_oobmark == 0) 2781 so->so_rcv.sb_state |= SBS_RCVATMARK; 2782 sohasoutofband(so); 2783 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2784 } 2785 SOCKBUF_UNLOCK(&so->so_rcv); 2786 /* 2787 * Remove out of band data so doesn't get presented to user. 2788 * This can happen independent of advancing the URG pointer, 2789 * but if two URG's are pending at once, some out-of-band 2790 * data may creep in... ick. 2791 */ 2792 if (th->th_urp <= (u_long)tlen && 2793 !(so->so_options & SO_OOBINLINE)) { 2794 /* hdr drop is delayed */ 2795 tcp_pulloutofband(so, th, m, drop_hdrlen); 2796 } 2797 } else { 2798 /* 2799 * If no out of band data is expected, 2800 * pull receive urgent pointer along 2801 * with the receive window. 2802 */ 2803 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2804 tp->rcv_up = tp->rcv_nxt; 2805 } 2806 dodata: /* XXX */ 2807 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2808 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2809 ("tcp_do_segment: dodata ti_locked %d", ti_locked)); 2810 INP_WLOCK_ASSERT(tp->t_inpcb); 2811 2812 /* 2813 * Process the segment text, merging it into the TCP sequencing queue, 2814 * and arranging for acknowledgment of receipt if necessary. 2815 * This process logically involves adjusting tp->rcv_wnd as data 2816 * is presented to the user (this happens in tcp_usrreq.c, 2817 * case PRU_RCVD). If a FIN has already been received on this 2818 * connection then we just ignore the text. 2819 */ 2820 if ((tlen || (thflags & TH_FIN)) && 2821 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2822 tcp_seq save_start = th->th_seq; 2823 m_adj(m, drop_hdrlen); /* delayed header drop */ 2824 /* 2825 * Insert segment which includes th into TCP reassembly queue 2826 * with control block tp. Set thflags to whether reassembly now 2827 * includes a segment with FIN. This handles the common case 2828 * inline (segment is the next to be received on an established 2829 * connection, and the queue is empty), avoiding linkage into 2830 * and removal from the queue and repetition of various 2831 * conversions. 2832 * Set DELACK for segments received in order, but ack 2833 * immediately when segments are out of order (so 2834 * fast retransmit can work). 2835 */ 2836 if (th->th_seq == tp->rcv_nxt && 2837 LIST_EMPTY(&tp->t_segq) && 2838 TCPS_HAVEESTABLISHED(tp->t_state)) { 2839 if (DELAY_ACK(tp)) 2840 tp->t_flags |= TF_DELACK; 2841 else 2842 tp->t_flags |= TF_ACKNOW; 2843 tp->rcv_nxt += tlen; 2844 thflags = th->th_flags & TH_FIN; 2845 TCPSTAT_INC(tcps_rcvpack); 2846 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2847 ND6_HINT(tp); 2848 SOCKBUF_LOCK(&so->so_rcv); 2849 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2850 m_freem(m); 2851 else 2852 sbappendstream_locked(&so->so_rcv, m); 2853 /* NB: sorwakeup_locked() does an implicit unlock. */ 2854 sorwakeup_locked(so); 2855 } else { 2856 /* 2857 * XXX: Due to the header drop above "th" is 2858 * theoretically invalid by now. Fortunately 2859 * m_adj() doesn't actually frees any mbufs 2860 * when trimming from the head. 2861 */ 2862 thflags = tcp_reass(tp, th, &tlen, m); 2863 tp->t_flags |= TF_ACKNOW; 2864 } 2865 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2866 tcp_update_sack_list(tp, save_start, save_start + tlen); 2867 #if 0 2868 /* 2869 * Note the amount of data that peer has sent into 2870 * our window, in order to estimate the sender's 2871 * buffer size. 2872 * XXX: Unused. 2873 */ 2874 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2875 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2876 else 2877 len = so->so_rcv.sb_hiwat; 2878 #endif 2879 } else { 2880 m_freem(m); 2881 thflags &= ~TH_FIN; 2882 } 2883 2884 /* 2885 * If FIN is received ACK the FIN and let the user know 2886 * that the connection is closing. 2887 */ 2888 if (thflags & TH_FIN) { 2889 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2890 socantrcvmore(so); 2891 /* 2892 * If connection is half-synchronized 2893 * (ie NEEDSYN flag on) then delay ACK, 2894 * so it may be piggybacked when SYN is sent. 2895 * Otherwise, since we received a FIN then no 2896 * more input can be expected, send ACK now. 2897 */ 2898 if (tp->t_flags & TF_NEEDSYN) 2899 tp->t_flags |= TF_DELACK; 2900 else 2901 tp->t_flags |= TF_ACKNOW; 2902 tp->rcv_nxt++; 2903 } 2904 switch (tp->t_state) { 2905 2906 /* 2907 * In SYN_RECEIVED and ESTABLISHED STATES 2908 * enter the CLOSE_WAIT state. 2909 */ 2910 case TCPS_SYN_RECEIVED: 2911 tp->t_starttime = ticks; 2912 /* FALLTHROUGH */ 2913 case TCPS_ESTABLISHED: 2914 tp->t_state = TCPS_CLOSE_WAIT; 2915 break; 2916 2917 /* 2918 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2919 * enter the CLOSING state. 2920 */ 2921 case TCPS_FIN_WAIT_1: 2922 tp->t_state = TCPS_CLOSING; 2923 break; 2924 2925 /* 2926 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2927 * starting the time-wait timer, turning off the other 2928 * standard timers. 2929 */ 2930 case TCPS_FIN_WAIT_2: 2931 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2932 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2933 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2934 ti_locked)); 2935 2936 tcp_twstart(tp); 2937 INP_INFO_WUNLOCK(&V_tcbinfo); 2938 return; 2939 } 2940 } 2941 if (ti_locked == TI_RLOCKED) 2942 INP_INFO_RUNLOCK(&V_tcbinfo); 2943 else if (ti_locked == TI_WLOCKED) 2944 INP_INFO_WUNLOCK(&V_tcbinfo); 2945 else 2946 panic("%s: dodata epilogue ti_locked %d", __func__, 2947 ti_locked); 2948 ti_locked = TI_UNLOCKED; 2949 2950 #ifdef TCPDEBUG 2951 if (so->so_options & SO_DEBUG) 2952 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2953 &tcp_savetcp, 0); 2954 #endif 2955 2956 /* 2957 * Return any desired output. 2958 */ 2959 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2960 (void) tcp_output(tp); 2961 2962 check_delack: 2963 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2964 __func__, ti_locked)); 2965 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2966 INP_WLOCK_ASSERT(tp->t_inpcb); 2967 2968 if (tp->t_flags & TF_DELACK) { 2969 tp->t_flags &= ~TF_DELACK; 2970 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2971 } 2972 INP_WUNLOCK(tp->t_inpcb); 2973 return; 2974 2975 dropafterack: 2976 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2977 ("tcp_do_segment: dropafterack ti_locked %d", ti_locked)); 2978 2979 /* 2980 * Generate an ACK dropping incoming segment if it occupies 2981 * sequence space, where the ACK reflects our state. 2982 * 2983 * We can now skip the test for the RST flag since all 2984 * paths to this code happen after packets containing 2985 * RST have been dropped. 2986 * 2987 * In the SYN-RECEIVED state, don't send an ACK unless the 2988 * segment we received passes the SYN-RECEIVED ACK test. 2989 * If it fails send a RST. This breaks the loop in the 2990 * "LAND" DoS attack, and also prevents an ACK storm 2991 * between two listening ports that have been sent forged 2992 * SYN segments, each with the source address of the other. 2993 */ 2994 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2995 (SEQ_GT(tp->snd_una, th->th_ack) || 2996 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2997 rstreason = BANDLIM_RST_OPENPORT; 2998 goto dropwithreset; 2999 } 3000 #ifdef TCPDEBUG 3001 if (so->so_options & SO_DEBUG) 3002 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3003 &tcp_savetcp, 0); 3004 #endif 3005 if (ti_locked == TI_RLOCKED) 3006 INP_INFO_RUNLOCK(&V_tcbinfo); 3007 else if (ti_locked == TI_WLOCKED) 3008 INP_INFO_WUNLOCK(&V_tcbinfo); 3009 else 3010 panic("%s: dropafterack epilogue ti_locked %d", __func__, 3011 ti_locked); 3012 ti_locked = TI_UNLOCKED; 3013 3014 tp->t_flags |= TF_ACKNOW; 3015 (void) tcp_output(tp); 3016 INP_WUNLOCK(tp->t_inpcb); 3017 m_freem(m); 3018 return; 3019 3020 dropwithreset: 3021 if (ti_locked == TI_RLOCKED) 3022 INP_INFO_RUNLOCK(&V_tcbinfo); 3023 else if (ti_locked == TI_WLOCKED) 3024 INP_INFO_WUNLOCK(&V_tcbinfo); 3025 else 3026 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked); 3027 ti_locked = TI_UNLOCKED; 3028 3029 if (tp != NULL) { 3030 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3031 INP_WUNLOCK(tp->t_inpcb); 3032 } else 3033 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3034 return; 3035 3036 drop: 3037 if (ti_locked == TI_RLOCKED) 3038 INP_INFO_RUNLOCK(&V_tcbinfo); 3039 else if (ti_locked == TI_WLOCKED) 3040 INP_INFO_WUNLOCK(&V_tcbinfo); 3041 #ifdef INVARIANTS 3042 else 3043 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3044 #endif 3045 ti_locked = TI_UNLOCKED; 3046 3047 /* 3048 * Drop space held by incoming segment and return. 3049 */ 3050 #ifdef TCPDEBUG 3051 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3052 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3053 &tcp_savetcp, 0); 3054 #endif 3055 if (tp != NULL) 3056 INP_WUNLOCK(tp->t_inpcb); 3057 m_freem(m); 3058 } 3059 3060 /* 3061 * Issue RST and make ACK acceptable to originator of segment. 3062 * The mbuf must still include the original packet header. 3063 * tp may be NULL. 3064 */ 3065 static void 3066 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3067 int tlen, int rstreason) 3068 { 3069 #ifdef INET 3070 struct ip *ip; 3071 #endif 3072 #ifdef INET6 3073 struct ip6_hdr *ip6; 3074 #endif 3075 3076 if (tp != NULL) { 3077 INP_WLOCK_ASSERT(tp->t_inpcb); 3078 } 3079 3080 /* Don't bother if destination was broadcast/multicast. */ 3081 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3082 goto drop; 3083 #ifdef INET6 3084 if (mtod(m, struct ip *)->ip_v == 6) { 3085 ip6 = mtod(m, struct ip6_hdr *); 3086 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3087 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3088 goto drop; 3089 /* IPv6 anycast check is done at tcp6_input() */ 3090 } 3091 #endif 3092 #if defined(INET) && defined(INET6) 3093 else 3094 #endif 3095 #ifdef INET 3096 { 3097 ip = mtod(m, struct ip *); 3098 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3099 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3100 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3101 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3102 goto drop; 3103 } 3104 #endif 3105 3106 /* Perform bandwidth limiting. */ 3107 if (badport_bandlim(rstreason) < 0) 3108 goto drop; 3109 3110 /* tcp_respond consumes the mbuf chain. */ 3111 if (th->th_flags & TH_ACK) { 3112 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3113 th->th_ack, TH_RST); 3114 } else { 3115 if (th->th_flags & TH_SYN) 3116 tlen++; 3117 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3118 (tcp_seq)0, TH_RST|TH_ACK); 3119 } 3120 return; 3121 drop: 3122 m_freem(m); 3123 } 3124 3125 /* 3126 * Parse TCP options and place in tcpopt. 3127 */ 3128 static void 3129 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3130 { 3131 int opt, optlen; 3132 3133 to->to_flags = 0; 3134 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3135 opt = cp[0]; 3136 if (opt == TCPOPT_EOL) 3137 break; 3138 if (opt == TCPOPT_NOP) 3139 optlen = 1; 3140 else { 3141 if (cnt < 2) 3142 break; 3143 optlen = cp[1]; 3144 if (optlen < 2 || optlen > cnt) 3145 break; 3146 } 3147 switch (opt) { 3148 case TCPOPT_MAXSEG: 3149 if (optlen != TCPOLEN_MAXSEG) 3150 continue; 3151 if (!(flags & TO_SYN)) 3152 continue; 3153 to->to_flags |= TOF_MSS; 3154 bcopy((char *)cp + 2, 3155 (char *)&to->to_mss, sizeof(to->to_mss)); 3156 to->to_mss = ntohs(to->to_mss); 3157 break; 3158 case TCPOPT_WINDOW: 3159 if (optlen != TCPOLEN_WINDOW) 3160 continue; 3161 if (!(flags & TO_SYN)) 3162 continue; 3163 to->to_flags |= TOF_SCALE; 3164 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3165 break; 3166 case TCPOPT_TIMESTAMP: 3167 if (optlen != TCPOLEN_TIMESTAMP) 3168 continue; 3169 to->to_flags |= TOF_TS; 3170 bcopy((char *)cp + 2, 3171 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3172 to->to_tsval = ntohl(to->to_tsval); 3173 bcopy((char *)cp + 6, 3174 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3175 to->to_tsecr = ntohl(to->to_tsecr); 3176 break; 3177 #ifdef TCP_SIGNATURE 3178 /* 3179 * XXX In order to reply to a host which has set the 3180 * TCP_SIGNATURE option in its initial SYN, we have to 3181 * record the fact that the option was observed here 3182 * for the syncache code to perform the correct response. 3183 */ 3184 case TCPOPT_SIGNATURE: 3185 if (optlen != TCPOLEN_SIGNATURE) 3186 continue; 3187 to->to_flags |= TOF_SIGNATURE; 3188 to->to_signature = cp + 2; 3189 break; 3190 #endif 3191 case TCPOPT_SACK_PERMITTED: 3192 if (optlen != TCPOLEN_SACK_PERMITTED) 3193 continue; 3194 if (!(flags & TO_SYN)) 3195 continue; 3196 if (!V_tcp_do_sack) 3197 continue; 3198 to->to_flags |= TOF_SACKPERM; 3199 break; 3200 case TCPOPT_SACK: 3201 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3202 continue; 3203 if (flags & TO_SYN) 3204 continue; 3205 to->to_flags |= TOF_SACK; 3206 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3207 to->to_sacks = cp + 2; 3208 TCPSTAT_INC(tcps_sack_rcv_blocks); 3209 break; 3210 default: 3211 continue; 3212 } 3213 } 3214 } 3215 3216 /* 3217 * Pull out of band byte out of a segment so 3218 * it doesn't appear in the user's data queue. 3219 * It is still reflected in the segment length for 3220 * sequencing purposes. 3221 */ 3222 static void 3223 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3224 int off) 3225 { 3226 int cnt = off + th->th_urp - 1; 3227 3228 while (cnt >= 0) { 3229 if (m->m_len > cnt) { 3230 char *cp = mtod(m, caddr_t) + cnt; 3231 struct tcpcb *tp = sototcpcb(so); 3232 3233 INP_WLOCK_ASSERT(tp->t_inpcb); 3234 3235 tp->t_iobc = *cp; 3236 tp->t_oobflags |= TCPOOB_HAVEDATA; 3237 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3238 m->m_len--; 3239 if (m->m_flags & M_PKTHDR) 3240 m->m_pkthdr.len--; 3241 return; 3242 } 3243 cnt -= m->m_len; 3244 m = m->m_next; 3245 if (m == NULL) 3246 break; 3247 } 3248 panic("tcp_pulloutofband"); 3249 } 3250 3251 /* 3252 * Collect new round-trip time estimate 3253 * and update averages and current timeout. 3254 */ 3255 static void 3256 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3257 { 3258 int delta; 3259 3260 INP_WLOCK_ASSERT(tp->t_inpcb); 3261 3262 TCPSTAT_INC(tcps_rttupdated); 3263 tp->t_rttupdated++; 3264 if (tp->t_srtt != 0) { 3265 /* 3266 * srtt is stored as fixed point with 5 bits after the 3267 * binary point (i.e., scaled by 8). The following magic 3268 * is equivalent to the smoothing algorithm in rfc793 with 3269 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3270 * point). Adjust rtt to origin 0. 3271 */ 3272 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3273 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3274 3275 if ((tp->t_srtt += delta) <= 0) 3276 tp->t_srtt = 1; 3277 3278 /* 3279 * We accumulate a smoothed rtt variance (actually, a 3280 * smoothed mean difference), then set the retransmit 3281 * timer to smoothed rtt + 4 times the smoothed variance. 3282 * rttvar is stored as fixed point with 4 bits after the 3283 * binary point (scaled by 16). The following is 3284 * equivalent to rfc793 smoothing with an alpha of .75 3285 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3286 * rfc793's wired-in beta. 3287 */ 3288 if (delta < 0) 3289 delta = -delta; 3290 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3291 if ((tp->t_rttvar += delta) <= 0) 3292 tp->t_rttvar = 1; 3293 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3294 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3295 } else { 3296 /* 3297 * No rtt measurement yet - use the unsmoothed rtt. 3298 * Set the variance to half the rtt (so our first 3299 * retransmit happens at 3*rtt). 3300 */ 3301 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3302 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3303 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3304 } 3305 tp->t_rtttime = 0; 3306 tp->t_rxtshift = 0; 3307 3308 /* 3309 * the retransmit should happen at rtt + 4 * rttvar. 3310 * Because of the way we do the smoothing, srtt and rttvar 3311 * will each average +1/2 tick of bias. When we compute 3312 * the retransmit timer, we want 1/2 tick of rounding and 3313 * 1 extra tick because of +-1/2 tick uncertainty in the 3314 * firing of the timer. The bias will give us exactly the 3315 * 1.5 tick we need. But, because the bias is 3316 * statistical, we have to test that we don't drop below 3317 * the minimum feasible timer (which is 2 ticks). 3318 */ 3319 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3320 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3321 3322 /* 3323 * We received an ack for a packet that wasn't retransmitted; 3324 * it is probably safe to discard any error indications we've 3325 * received recently. This isn't quite right, but close enough 3326 * for now (a route might have failed after we sent a segment, 3327 * and the return path might not be symmetrical). 3328 */ 3329 tp->t_softerror = 0; 3330 } 3331 3332 /* 3333 * Determine a reasonable value for maxseg size. 3334 * If the route is known, check route for mtu. 3335 * If none, use an mss that can be handled on the outgoing 3336 * interface without forcing IP to fragment; if bigger than 3337 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 3338 * to utilize large mbufs. If no route is found, route has no mtu, 3339 * or the destination isn't local, use a default, hopefully conservative 3340 * size (usually 512 or the default IP max size, but no more than the mtu 3341 * of the interface), as we can't discover anything about intervening 3342 * gateways or networks. We also initialize the congestion/slow start 3343 * window to be a single segment if the destination isn't local. 3344 * While looking at the routing entry, we also initialize other path-dependent 3345 * parameters from pre-set or cached values in the routing entry. 3346 * 3347 * Also take into account the space needed for options that we 3348 * send regularly. Make maxseg shorter by that amount to assure 3349 * that we can send maxseg amount of data even when the options 3350 * are present. Store the upper limit of the length of options plus 3351 * data in maxopd. 3352 * 3353 * In case of T/TCP, we call this routine during implicit connection 3354 * setup as well (offer = -1), to initialize maxseg from the cached 3355 * MSS of our peer. 3356 * 3357 * NOTE that this routine is only called when we process an incoming 3358 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 3359 */ 3360 void 3361 tcp_mss_update(struct tcpcb *tp, int offer, 3362 struct hc_metrics_lite *metricptr, int *mtuflags) 3363 { 3364 int mss = 0; 3365 u_long maxmtu = 0; 3366 struct inpcb *inp = tp->t_inpcb; 3367 struct hc_metrics_lite metrics; 3368 int origoffer = offer; 3369 #ifdef INET6 3370 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3371 size_t min_protoh = isipv6 ? 3372 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3373 sizeof (struct tcpiphdr); 3374 #else 3375 const size_t min_protoh = sizeof(struct tcpiphdr); 3376 #endif 3377 3378 INP_WLOCK_ASSERT(tp->t_inpcb); 3379 3380 /* Initialize. */ 3381 #ifdef INET6 3382 if (isipv6) { 3383 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags); 3384 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3385 } 3386 #endif 3387 #if defined(INET) && defined(INET6) 3388 else 3389 #endif 3390 #ifdef INET 3391 { 3392 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags); 3393 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3394 } 3395 #endif 3396 3397 /* 3398 * No route to sender, stay with default mss and return. 3399 */ 3400 if (maxmtu == 0) { 3401 /* 3402 * In case we return early we need to initialize metrics 3403 * to a defined state as tcp_hc_get() would do for us 3404 * if there was no cache hit. 3405 */ 3406 if (metricptr != NULL) 3407 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3408 return; 3409 } 3410 3411 /* What have we got? */ 3412 switch (offer) { 3413 case 0: 3414 /* 3415 * Offer == 0 means that there was no MSS on the SYN 3416 * segment, in this case we use tcp_mssdflt as 3417 * already assigned to t_maxopd above. 3418 */ 3419 offer = tp->t_maxopd; 3420 break; 3421 3422 case -1: 3423 /* 3424 * Offer == -1 means that we didn't receive SYN yet. 3425 */ 3426 /* FALLTHROUGH */ 3427 3428 default: 3429 /* 3430 * Prevent DoS attack with too small MSS. Round up 3431 * to at least minmss. 3432 */ 3433 offer = max(offer, V_tcp_minmss); 3434 } 3435 3436 /* 3437 * rmx information is now retrieved from tcp_hostcache. 3438 */ 3439 tcp_hc_get(&inp->inp_inc, &metrics); 3440 if (metricptr != NULL) 3441 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3442 3443 /* 3444 * If there's a discovered mtu int tcp hostcache, use it 3445 * else, use the link mtu. 3446 */ 3447 if (metrics.rmx_mtu) 3448 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3449 else { 3450 #ifdef INET6 3451 if (isipv6) { 3452 mss = maxmtu - min_protoh; 3453 if (!V_path_mtu_discovery && 3454 !in6_localaddr(&inp->in6p_faddr)) 3455 mss = min(mss, V_tcp_v6mssdflt); 3456 } 3457 #endif 3458 #if defined(INET) && defined(INET6) 3459 else 3460 #endif 3461 #ifdef INET 3462 { 3463 mss = maxmtu - min_protoh; 3464 if (!V_path_mtu_discovery && 3465 !in_localaddr(inp->inp_faddr)) 3466 mss = min(mss, V_tcp_mssdflt); 3467 } 3468 #endif 3469 /* 3470 * XXX - The above conditional (mss = maxmtu - min_protoh) 3471 * probably violates the TCP spec. 3472 * The problem is that, since we don't know the 3473 * other end's MSS, we are supposed to use a conservative 3474 * default. But, if we do that, then MTU discovery will 3475 * never actually take place, because the conservative 3476 * default is much less than the MTUs typically seen 3477 * on the Internet today. For the moment, we'll sweep 3478 * this under the carpet. 3479 * 3480 * The conservative default might not actually be a problem 3481 * if the only case this occurs is when sending an initial 3482 * SYN with options and data to a host we've never talked 3483 * to before. Then, they will reply with an MSS value which 3484 * will get recorded and the new parameters should get 3485 * recomputed. For Further Study. 3486 */ 3487 } 3488 mss = min(mss, offer); 3489 3490 /* 3491 * Sanity check: make sure that maxopd will be large 3492 * enough to allow some data on segments even if the 3493 * all the option space is used (40bytes). Otherwise 3494 * funny things may happen in tcp_output. 3495 */ 3496 mss = max(mss, 64); 3497 3498 /* 3499 * maxopd stores the maximum length of data AND options 3500 * in a segment; maxseg is the amount of data in a normal 3501 * segment. We need to store this value (maxopd) apart 3502 * from maxseg, because now every segment carries options 3503 * and thus we normally have somewhat less data in segments. 3504 */ 3505 tp->t_maxopd = mss; 3506 3507 /* 3508 * origoffer==-1 indicates that no segments were received yet. 3509 * In this case we just guess. 3510 */ 3511 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3512 (origoffer == -1 || 3513 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3514 mss -= TCPOLEN_TSTAMP_APPA; 3515 3516 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3517 if (mss > MCLBYTES) 3518 mss &= ~(MCLBYTES-1); 3519 #else 3520 if (mss > MCLBYTES) 3521 mss = mss / MCLBYTES * MCLBYTES; 3522 #endif 3523 tp->t_maxseg = mss; 3524 } 3525 3526 void 3527 tcp_mss(struct tcpcb *tp, int offer) 3528 { 3529 int mss; 3530 u_long bufsize; 3531 struct inpcb *inp; 3532 struct socket *so; 3533 struct hc_metrics_lite metrics; 3534 int mtuflags = 0; 3535 3536 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3537 3538 tcp_mss_update(tp, offer, &metrics, &mtuflags); 3539 3540 mss = tp->t_maxseg; 3541 inp = tp->t_inpcb; 3542 3543 /* 3544 * If there's a pipesize, change the socket buffer to that size, 3545 * don't change if sb_hiwat is different than default (then it 3546 * has been changed on purpose with setsockopt). 3547 * Make the socket buffers an integral number of mss units; 3548 * if the mss is larger than the socket buffer, decrease the mss. 3549 */ 3550 so = inp->inp_socket; 3551 SOCKBUF_LOCK(&so->so_snd); 3552 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3553 bufsize = metrics.rmx_sendpipe; 3554 else 3555 bufsize = so->so_snd.sb_hiwat; 3556 if (bufsize < mss) 3557 mss = bufsize; 3558 else { 3559 bufsize = roundup(bufsize, mss); 3560 if (bufsize > sb_max) 3561 bufsize = sb_max; 3562 if (bufsize > so->so_snd.sb_hiwat) 3563 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3564 } 3565 SOCKBUF_UNLOCK(&so->so_snd); 3566 tp->t_maxseg = mss; 3567 3568 SOCKBUF_LOCK(&so->so_rcv); 3569 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3570 bufsize = metrics.rmx_recvpipe; 3571 else 3572 bufsize = so->so_rcv.sb_hiwat; 3573 if (bufsize > mss) { 3574 bufsize = roundup(bufsize, mss); 3575 if (bufsize > sb_max) 3576 bufsize = sb_max; 3577 if (bufsize > so->so_rcv.sb_hiwat) 3578 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3579 } 3580 SOCKBUF_UNLOCK(&so->so_rcv); 3581 3582 /* Check the interface for TSO capabilities. */ 3583 if (mtuflags & CSUM_TSO) 3584 tp->t_flags |= TF_TSO; 3585 } 3586 3587 /* 3588 * Determine the MSS option to send on an outgoing SYN. 3589 */ 3590 int 3591 tcp_mssopt(struct in_conninfo *inc) 3592 { 3593 int mss = 0; 3594 u_long maxmtu = 0; 3595 u_long thcmtu = 0; 3596 size_t min_protoh; 3597 3598 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3599 3600 #ifdef INET6 3601 if (inc->inc_flags & INC_ISIPV6) { 3602 mss = V_tcp_v6mssdflt; 3603 maxmtu = tcp_maxmtu6(inc, NULL); 3604 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3605 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3606 } 3607 #endif 3608 #if defined(INET) && defined(INET6) 3609 else 3610 #endif 3611 #ifdef INET 3612 { 3613 mss = V_tcp_mssdflt; 3614 maxmtu = tcp_maxmtu(inc, NULL); 3615 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3616 min_protoh = sizeof(struct tcpiphdr); 3617 } 3618 #endif 3619 if (maxmtu && thcmtu) 3620 mss = min(maxmtu, thcmtu) - min_protoh; 3621 else if (maxmtu || thcmtu) 3622 mss = max(maxmtu, thcmtu) - min_protoh; 3623 3624 return (mss); 3625 } 3626 3627 3628 /* 3629 * On a partial ack arrives, force the retransmission of the 3630 * next unacknowledged segment. Do not clear tp->t_dupacks. 3631 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3632 * be started again. 3633 */ 3634 static void 3635 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3636 { 3637 tcp_seq onxt = tp->snd_nxt; 3638 u_long ocwnd = tp->snd_cwnd; 3639 3640 INP_WLOCK_ASSERT(tp->t_inpcb); 3641 3642 tcp_timer_activate(tp, TT_REXMT, 0); 3643 tp->t_rtttime = 0; 3644 tp->snd_nxt = th->th_ack; 3645 /* 3646 * Set snd_cwnd to one segment beyond acknowledged offset. 3647 * (tp->snd_una has not yet been updated when this function is called.) 3648 */ 3649 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3650 tp->t_flags |= TF_ACKNOW; 3651 (void) tcp_output(tp); 3652 tp->snd_cwnd = ocwnd; 3653 if (SEQ_GT(onxt, tp->snd_nxt)) 3654 tp->snd_nxt = onxt; 3655 /* 3656 * Partial window deflation. Relies on fact that tp->snd_una 3657 * not updated yet. 3658 */ 3659 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3660 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3661 else 3662 tp->snd_cwnd = 0; 3663 tp->snd_cwnd += tp->t_maxseg; 3664 } 3665