1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 4. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_ipfw.h" /* for ipfw_fwd */ 54 #include "opt_inet.h" 55 #include "opt_inet6.h" 56 #include "opt_ipsec.h" 57 #include "opt_tcpdebug.h" 58 59 #include <sys/param.h> 60 #include <sys/kernel.h> 61 #include <sys/hhook.h> 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/signalvar.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/sysctl.h> 70 #include <sys/syslog.h> 71 #include <sys/systm.h> 72 73 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 74 75 #include <vm/uma.h> 76 77 #include <net/if.h> 78 #include <net/route.h> 79 #include <net/vnet.h> 80 81 #define TCPSTATES /* for logging */ 82 83 #include <netinet/cc.h> 84 #include <netinet/in.h> 85 #include <netinet/in_pcb.h> 86 #include <netinet/in_systm.h> 87 #include <netinet/in_var.h> 88 #include <netinet/ip.h> 89 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 90 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 91 #include <netinet/ip_var.h> 92 #include <netinet/ip_options.h> 93 #include <netinet/ip6.h> 94 #include <netinet/icmp6.h> 95 #include <netinet6/in6_pcb.h> 96 #include <netinet6/ip6_var.h> 97 #include <netinet6/nd6.h> 98 #include <netinet/tcp_fsm.h> 99 #include <netinet/tcp_seq.h> 100 #include <netinet/tcp_timer.h> 101 #include <netinet/tcp_var.h> 102 #include <netinet6/tcp6_var.h> 103 #include <netinet/tcpip.h> 104 #include <netinet/tcp_syncache.h> 105 #ifdef TCPDEBUG 106 #include <netinet/tcp_debug.h> 107 #endif /* TCPDEBUG */ 108 #ifdef TCP_OFFLOAD 109 #include <netinet/tcp_offload.h> 110 #endif 111 112 #ifdef IPSEC 113 #include <netipsec/ipsec.h> 114 #include <netipsec/ipsec6.h> 115 #endif /*IPSEC*/ 116 117 #include <machine/in_cksum.h> 118 119 #include <security/mac/mac_framework.h> 120 121 const int tcprexmtthresh = 3; 122 123 VNET_DEFINE(struct tcpstat, tcpstat); 124 SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 125 &VNET_NAME(tcpstat), tcpstat, 126 "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 127 128 int tcp_log_in_vain = 0; 129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 130 &tcp_log_in_vain, 0, 131 "Log all incoming TCP segments to closed ports"); 132 133 VNET_DEFINE(int, blackhole) = 0; 134 #define V_blackhole VNET(blackhole) 135 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 136 &VNET_NAME(blackhole), 0, 137 "Do not send RST on segments to closed ports"); 138 139 VNET_DEFINE(int, tcp_delack_enabled) = 1; 140 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 141 &VNET_NAME(tcp_delack_enabled), 0, 142 "Delay ACK to try and piggyback it onto a data packet"); 143 144 VNET_DEFINE(int, drop_synfin) = 0; 145 #define V_drop_synfin VNET(drop_synfin) 146 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 147 &VNET_NAME(drop_synfin), 0, 148 "Drop TCP packets with SYN+FIN set"); 149 150 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 151 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 152 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 153 &VNET_NAME(tcp_do_rfc3042), 0, 154 "Enable RFC 3042 (Limited Transmit)"); 155 156 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 157 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 158 &VNET_NAME(tcp_do_rfc3390), 0, 159 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 160 161 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 162 "Experimental TCP extensions"); 163 164 VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 165 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 166 &VNET_NAME(tcp_do_initcwnd10), 0, 167 "Enable draft-ietf-tcpm-initcwnd-05 (Increasing initial CWND to 10)"); 168 169 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 170 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 171 &VNET_NAME(tcp_do_rfc3465), 0, 172 "Enable RFC 3465 (Appropriate Byte Counting)"); 173 174 VNET_DEFINE(int, tcp_abc_l_var) = 2; 175 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 176 &VNET_NAME(tcp_abc_l_var), 2, 177 "Cap the max cwnd increment during slow-start to this number of segments"); 178 179 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 180 181 VNET_DEFINE(int, tcp_do_ecn) = 0; 182 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 183 &VNET_NAME(tcp_do_ecn), 0, 184 "TCP ECN support"); 185 186 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 187 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 188 &VNET_NAME(tcp_ecn_maxretries), 0, 189 "Max retries before giving up on ECN"); 190 191 VNET_DEFINE(int, tcp_insecure_rst) = 0; 192 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 193 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 194 &VNET_NAME(tcp_insecure_rst), 0, 195 "Follow the old (insecure) criteria for accepting RST packets"); 196 197 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 198 #define V_tcp_recvspace VNET(tcp_recvspace) 199 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 200 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 201 202 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 203 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 204 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 205 &VNET_NAME(tcp_do_autorcvbuf), 0, 206 "Enable automatic receive buffer sizing"); 207 208 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 209 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 210 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 211 &VNET_NAME(tcp_autorcvbuf_inc), 0, 212 "Incrementor step size of automatic receive buffer"); 213 214 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 215 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 216 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 217 &VNET_NAME(tcp_autorcvbuf_max), 0, 218 "Max size of automatic receive buffer"); 219 220 VNET_DEFINE(struct inpcbhead, tcb); 221 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 222 VNET_DEFINE(struct inpcbinfo, tcbinfo); 223 224 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 225 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 226 struct socket *, struct tcpcb *, int, int, uint8_t, 227 int); 228 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 229 struct tcpcb *, int, int); 230 static void tcp_pulloutofband(struct socket *, 231 struct tcphdr *, struct mbuf *, int); 232 static void tcp_xmit_timer(struct tcpcb *, int); 233 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 234 static void inline tcp_fields_to_host(struct tcphdr *); 235 #ifdef TCP_SIGNATURE 236 static void inline tcp_fields_to_net(struct tcphdr *); 237 static int inline tcp_signature_verify_input(struct mbuf *, int, int, 238 int, struct tcpopt *, struct tcphdr *, u_int); 239 #endif 240 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 241 uint16_t type); 242 static void inline cc_conn_init(struct tcpcb *tp); 243 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 244 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 245 struct tcphdr *th, struct tcpopt *to); 246 247 /* 248 * Kernel module interface for updating tcpstat. The argument is an index 249 * into tcpstat treated as an array of u_long. While this encodes the 250 * general layout of tcpstat into the caller, it doesn't encode its location, 251 * so that future changes to add, for example, per-CPU stats support won't 252 * cause binary compatibility problems for kernel modules. 253 */ 254 void 255 kmod_tcpstat_inc(int statnum) 256 { 257 258 (*((u_long *)&V_tcpstat + statnum))++; 259 } 260 261 /* 262 * Wrapper for the TCP established input helper hook. 263 */ 264 static void inline 265 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 266 { 267 struct tcp_hhook_data hhook_data; 268 269 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 270 hhook_data.tp = tp; 271 hhook_data.th = th; 272 hhook_data.to = to; 273 274 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 275 tp->osd); 276 } 277 } 278 279 /* 280 * CC wrapper hook functions 281 */ 282 static void inline 283 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 284 { 285 INP_WLOCK_ASSERT(tp->t_inpcb); 286 287 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 288 if (tp->snd_cwnd == min(tp->snd_cwnd, tp->snd_wnd)) 289 tp->ccv->flags |= CCF_CWND_LIMITED; 290 else 291 tp->ccv->flags &= ~CCF_CWND_LIMITED; 292 293 if (type == CC_ACK) { 294 if (tp->snd_cwnd > tp->snd_ssthresh) { 295 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 296 V_tcp_abc_l_var * tp->t_maxseg); 297 if (tp->t_bytes_acked >= tp->snd_cwnd) { 298 tp->t_bytes_acked -= tp->snd_cwnd; 299 tp->ccv->flags |= CCF_ABC_SENTAWND; 300 } 301 } else { 302 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 303 tp->t_bytes_acked = 0; 304 } 305 } 306 307 if (CC_ALGO(tp)->ack_received != NULL) { 308 /* XXXLAS: Find a way to live without this */ 309 tp->ccv->curack = th->th_ack; 310 CC_ALGO(tp)->ack_received(tp->ccv, type); 311 } 312 } 313 314 static void inline 315 cc_conn_init(struct tcpcb *tp) 316 { 317 struct hc_metrics_lite metrics; 318 struct inpcb *inp = tp->t_inpcb; 319 int rtt; 320 321 INP_WLOCK_ASSERT(tp->t_inpcb); 322 323 tcp_hc_get(&inp->inp_inc, &metrics); 324 325 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 326 tp->t_srtt = rtt; 327 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 328 TCPSTAT_INC(tcps_usedrtt); 329 if (metrics.rmx_rttvar) { 330 tp->t_rttvar = metrics.rmx_rttvar; 331 TCPSTAT_INC(tcps_usedrttvar); 332 } else { 333 /* default variation is +- 1 rtt */ 334 tp->t_rttvar = 335 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 336 } 337 TCPT_RANGESET(tp->t_rxtcur, 338 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 339 tp->t_rttmin, TCPTV_REXMTMAX); 340 } 341 if (metrics.rmx_ssthresh) { 342 /* 343 * There's some sort of gateway or interface 344 * buffer limit on the path. Use this to set 345 * the slow start threshhold, but set the 346 * threshold to no less than 2*mss. 347 */ 348 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 349 TCPSTAT_INC(tcps_usedssthresh); 350 } 351 352 /* 353 * Set the initial slow-start flight size. 354 * 355 * RFC5681 Section 3.1 specifies the default conservative values. 356 * RFC3390 specifies slightly more aggressive values. 357 * Draft-ietf-tcpm-initcwnd-05 increases it to ten segments. 358 * 359 * If a SYN or SYN/ACK was lost and retransmitted, we have to 360 * reduce the initial CWND to one segment as congestion is likely 361 * requiring us to be cautious. 362 */ 363 if (tp->snd_cwnd == 1) 364 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 365 else if (V_tcp_do_initcwnd10) 366 tp->snd_cwnd = min(10 * tp->t_maxseg, 367 max(2 * tp->t_maxseg, 14600)); 368 else if (V_tcp_do_rfc3390) 369 tp->snd_cwnd = min(4 * tp->t_maxseg, 370 max(2 * tp->t_maxseg, 4380)); 371 else { 372 /* Per RFC5681 Section 3.1 */ 373 if (tp->t_maxseg > 2190) 374 tp->snd_cwnd = 2 * tp->t_maxseg; 375 else if (tp->t_maxseg > 1095) 376 tp->snd_cwnd = 3 * tp->t_maxseg; 377 else 378 tp->snd_cwnd = 4 * tp->t_maxseg; 379 } 380 381 if (CC_ALGO(tp)->conn_init != NULL) 382 CC_ALGO(tp)->conn_init(tp->ccv); 383 } 384 385 void inline 386 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 387 { 388 INP_WLOCK_ASSERT(tp->t_inpcb); 389 390 switch(type) { 391 case CC_NDUPACK: 392 if (!IN_FASTRECOVERY(tp->t_flags)) { 393 tp->snd_recover = tp->snd_max; 394 if (tp->t_flags & TF_ECN_PERMIT) 395 tp->t_flags |= TF_ECN_SND_CWR; 396 } 397 break; 398 case CC_ECN: 399 if (!IN_CONGRECOVERY(tp->t_flags)) { 400 TCPSTAT_INC(tcps_ecn_rcwnd); 401 tp->snd_recover = tp->snd_max; 402 if (tp->t_flags & TF_ECN_PERMIT) 403 tp->t_flags |= TF_ECN_SND_CWR; 404 } 405 break; 406 case CC_RTO: 407 tp->t_dupacks = 0; 408 tp->t_bytes_acked = 0; 409 EXIT_RECOVERY(tp->t_flags); 410 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 411 tp->t_maxseg) * tp->t_maxseg; 412 tp->snd_cwnd = tp->t_maxseg; 413 break; 414 case CC_RTO_ERR: 415 TCPSTAT_INC(tcps_sndrexmitbad); 416 /* RTO was unnecessary, so reset everything. */ 417 tp->snd_cwnd = tp->snd_cwnd_prev; 418 tp->snd_ssthresh = tp->snd_ssthresh_prev; 419 tp->snd_recover = tp->snd_recover_prev; 420 if (tp->t_flags & TF_WASFRECOVERY) 421 ENTER_FASTRECOVERY(tp->t_flags); 422 if (tp->t_flags & TF_WASCRECOVERY) 423 ENTER_CONGRECOVERY(tp->t_flags); 424 tp->snd_nxt = tp->snd_max; 425 tp->t_flags &= ~TF_PREVVALID; 426 tp->t_badrxtwin = 0; 427 break; 428 } 429 430 if (CC_ALGO(tp)->cong_signal != NULL) { 431 if (th != NULL) 432 tp->ccv->curack = th->th_ack; 433 CC_ALGO(tp)->cong_signal(tp->ccv, type); 434 } 435 } 436 437 static void inline 438 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 439 { 440 INP_WLOCK_ASSERT(tp->t_inpcb); 441 442 /* XXXLAS: KASSERT that we're in recovery? */ 443 444 if (CC_ALGO(tp)->post_recovery != NULL) { 445 tp->ccv->curack = th->th_ack; 446 CC_ALGO(tp)->post_recovery(tp->ccv); 447 } 448 /* XXXLAS: EXIT_RECOVERY ? */ 449 tp->t_bytes_acked = 0; 450 } 451 452 static inline void 453 tcp_fields_to_host(struct tcphdr *th) 454 { 455 456 th->th_seq = ntohl(th->th_seq); 457 th->th_ack = ntohl(th->th_ack); 458 th->th_win = ntohs(th->th_win); 459 th->th_urp = ntohs(th->th_urp); 460 } 461 462 #ifdef TCP_SIGNATURE 463 static inline void 464 tcp_fields_to_net(struct tcphdr *th) 465 { 466 467 th->th_seq = htonl(th->th_seq); 468 th->th_ack = htonl(th->th_ack); 469 th->th_win = htons(th->th_win); 470 th->th_urp = htons(th->th_urp); 471 } 472 473 static inline int 474 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 475 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 476 { 477 int ret; 478 479 tcp_fields_to_net(th); 480 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 481 tcp_fields_to_host(th); 482 return (ret); 483 } 484 #endif 485 486 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 487 #ifdef INET6 488 #define ND6_HINT(tp) \ 489 do { \ 490 if ((tp) && (tp)->t_inpcb && \ 491 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 492 nd6_nud_hint(NULL, NULL, 0); \ 493 } while (0) 494 #else 495 #define ND6_HINT(tp) 496 #endif 497 498 /* 499 * Indicate whether this ack should be delayed. We can delay the ack if 500 * - there is no delayed ack timer in progress and 501 * - our last ack wasn't a 0-sized window. We never want to delay 502 * the ack that opens up a 0-sized window and 503 * - delayed acks are enabled or 504 * - this is a half-synchronized T/TCP connection. 505 */ 506 #define DELAY_ACK(tp) \ 507 ((!tcp_timer_active(tp, TT_DELACK) && \ 508 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 509 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 510 511 /* 512 * TCP input handling is split into multiple parts: 513 * tcp6_input is a thin wrapper around tcp_input for the extended 514 * ip6_protox[] call format in ip6_input 515 * tcp_input handles primary segment validation, inpcb lookup and 516 * SYN processing on listen sockets 517 * tcp_do_segment processes the ACK and text of the segment for 518 * establishing, established and closing connections 519 */ 520 #ifdef INET6 521 int 522 tcp6_input(struct mbuf **mp, int *offp, int proto) 523 { 524 struct mbuf *m = *mp; 525 struct in6_ifaddr *ia6; 526 527 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 528 529 /* 530 * draft-itojun-ipv6-tcp-to-anycast 531 * better place to put this in? 532 */ 533 ia6 = ip6_getdstifaddr(m); 534 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 535 struct ip6_hdr *ip6; 536 537 ifa_free(&ia6->ia_ifa); 538 ip6 = mtod(m, struct ip6_hdr *); 539 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 540 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 541 return IPPROTO_DONE; 542 } 543 if (ia6) 544 ifa_free(&ia6->ia_ifa); 545 546 tcp_input(m, *offp); 547 return IPPROTO_DONE; 548 } 549 #endif /* INET6 */ 550 551 void 552 tcp_input(struct mbuf *m, int off0) 553 { 554 struct tcphdr *th = NULL; 555 struct ip *ip = NULL; 556 struct inpcb *inp = NULL; 557 struct tcpcb *tp = NULL; 558 struct socket *so = NULL; 559 u_char *optp = NULL; 560 int optlen = 0; 561 #ifdef INET 562 int len; 563 #endif 564 int tlen = 0, off; 565 int drop_hdrlen; 566 int thflags; 567 int rstreason = 0; /* For badport_bandlim accounting purposes */ 568 #ifdef TCP_SIGNATURE 569 uint8_t sig_checked = 0; 570 #endif 571 uint8_t iptos = 0; 572 struct m_tag *fwd_tag = NULL; 573 #ifdef INET6 574 struct ip6_hdr *ip6 = NULL; 575 int isipv6; 576 #else 577 const void *ip6 = NULL; 578 #endif /* INET6 */ 579 struct tcpopt to; /* options in this segment */ 580 char *s = NULL; /* address and port logging */ 581 int ti_locked; 582 #define TI_UNLOCKED 1 583 #define TI_WLOCKED 2 584 585 #ifdef TCPDEBUG 586 /* 587 * The size of tcp_saveipgen must be the size of the max ip header, 588 * now IPv6. 589 */ 590 u_char tcp_saveipgen[IP6_HDR_LEN]; 591 struct tcphdr tcp_savetcp; 592 short ostate = 0; 593 #endif 594 595 #ifdef INET6 596 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 597 #endif 598 599 to.to_flags = 0; 600 TCPSTAT_INC(tcps_rcvtotal); 601 602 #ifdef INET6 603 if (isipv6) { 604 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 605 606 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 607 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 608 if (m == NULL) { 609 TCPSTAT_INC(tcps_rcvshort); 610 return; 611 } 612 } 613 614 ip6 = mtod(m, struct ip6_hdr *); 615 th = (struct tcphdr *)((caddr_t)ip6 + off0); 616 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 617 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 618 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 619 th->th_sum = m->m_pkthdr.csum_data; 620 else 621 th->th_sum = in6_cksum_pseudo(ip6, tlen, 622 IPPROTO_TCP, m->m_pkthdr.csum_data); 623 th->th_sum ^= 0xffff; 624 } else 625 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 626 if (th->th_sum) { 627 TCPSTAT_INC(tcps_rcvbadsum); 628 goto drop; 629 } 630 631 /* 632 * Be proactive about unspecified IPv6 address in source. 633 * As we use all-zero to indicate unbounded/unconnected pcb, 634 * unspecified IPv6 address can be used to confuse us. 635 * 636 * Note that packets with unspecified IPv6 destination is 637 * already dropped in ip6_input. 638 */ 639 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 640 /* XXX stat */ 641 goto drop; 642 } 643 } 644 #endif 645 #if defined(INET) && defined(INET6) 646 else 647 #endif 648 #ifdef INET 649 { 650 /* 651 * Get IP and TCP header together in first mbuf. 652 * Note: IP leaves IP header in first mbuf. 653 */ 654 if (off0 > sizeof (struct ip)) { 655 ip_stripoptions(m); 656 off0 = sizeof(struct ip); 657 } 658 if (m->m_len < sizeof (struct tcpiphdr)) { 659 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 660 == NULL) { 661 TCPSTAT_INC(tcps_rcvshort); 662 return; 663 } 664 } 665 ip = mtod(m, struct ip *); 666 th = (struct tcphdr *)((caddr_t)ip + off0); 667 tlen = ntohs(ip->ip_len) - off0; 668 669 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 670 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 671 th->th_sum = m->m_pkthdr.csum_data; 672 else 673 th->th_sum = in_pseudo(ip->ip_src.s_addr, 674 ip->ip_dst.s_addr, 675 htonl(m->m_pkthdr.csum_data + tlen + 676 IPPROTO_TCP)); 677 th->th_sum ^= 0xffff; 678 } else { 679 struct ipovly *ipov = (struct ipovly *)ip; 680 681 /* 682 * Checksum extended TCP header and data. 683 */ 684 len = off0 + tlen; 685 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 686 ipov->ih_len = htons(tlen); 687 th->th_sum = in_cksum(m, len); 688 } 689 if (th->th_sum) { 690 TCPSTAT_INC(tcps_rcvbadsum); 691 goto drop; 692 } 693 /* Re-initialization for later version check */ 694 ip->ip_v = IPVERSION; 695 } 696 #endif /* INET */ 697 698 #ifdef INET6 699 if (isipv6) 700 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 701 #endif 702 #if defined(INET) && defined(INET6) 703 else 704 #endif 705 #ifdef INET 706 iptos = ip->ip_tos; 707 #endif 708 709 /* 710 * Check that TCP offset makes sense, 711 * pull out TCP options and adjust length. XXX 712 */ 713 off = th->th_off << 2; 714 if (off < sizeof (struct tcphdr) || off > tlen) { 715 TCPSTAT_INC(tcps_rcvbadoff); 716 goto drop; 717 } 718 tlen -= off; /* tlen is used instead of ti->ti_len */ 719 if (off > sizeof (struct tcphdr)) { 720 #ifdef INET6 721 if (isipv6) { 722 IP6_EXTHDR_CHECK(m, off0, off, ); 723 ip6 = mtod(m, struct ip6_hdr *); 724 th = (struct tcphdr *)((caddr_t)ip6 + off0); 725 } 726 #endif 727 #if defined(INET) && defined(INET6) 728 else 729 #endif 730 #ifdef INET 731 { 732 if (m->m_len < sizeof(struct ip) + off) { 733 if ((m = m_pullup(m, sizeof (struct ip) + off)) 734 == NULL) { 735 TCPSTAT_INC(tcps_rcvshort); 736 return; 737 } 738 ip = mtod(m, struct 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; if we're likely to add or remove a 760 * connection then first acquire pcbinfo lock. There are two cases 761 * where we might discover later we need a write lock despite the 762 * flags: ACKs moving a connection out of the syncache, and ACKs for 763 * a connection in TIMEWAIT. 764 */ 765 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 766 INP_INFO_WLOCK(&V_tcbinfo); 767 ti_locked = TI_WLOCKED; 768 } else 769 ti_locked = TI_UNLOCKED; 770 771 findpcb: 772 #ifdef INVARIANTS 773 if (ti_locked == TI_WLOCKED) { 774 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 775 } else { 776 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 777 } 778 #endif 779 780 /* 781 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 782 */ 783 if (m->m_flags & M_IP_NEXTHOP) 784 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 785 786 #ifdef INET6 787 if (isipv6 && fwd_tag != NULL) { 788 struct sockaddr_in6 *next_hop6; 789 790 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 791 /* 792 * Transparently forwarded. Pretend to be the destination. 793 * Already got one like this? 794 */ 795 inp = in6_pcblookup_mbuf(&V_tcbinfo, 796 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 797 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 798 if (!inp) { 799 /* 800 * It's new. Try to find the ambushing socket. 801 * Because we've rewritten the destination address, 802 * any hardware-generated hash is ignored. 803 */ 804 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 805 th->th_sport, &next_hop6->sin6_addr, 806 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 807 th->th_dport, INPLOOKUP_WILDCARD | 808 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 809 } 810 /* Remove the tag from the packet. We don't need it anymore. */ 811 m_tag_delete(m, fwd_tag); 812 m->m_flags &= ~M_IP_NEXTHOP; 813 } else if (isipv6) { 814 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 815 th->th_sport, &ip6->ip6_dst, th->th_dport, 816 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 817 m->m_pkthdr.rcvif, m); 818 } 819 #endif /* INET6 */ 820 #if defined(INET6) && defined(INET) 821 else 822 #endif 823 #ifdef INET 824 if (fwd_tag != NULL) { 825 struct sockaddr_in *next_hop; 826 827 next_hop = (struct sockaddr_in *)(fwd_tag+1); 828 /* 829 * Transparently forwarded. Pretend to be the destination. 830 * already got one like this? 831 */ 832 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 833 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 834 m->m_pkthdr.rcvif, m); 835 if (!inp) { 836 /* 837 * It's new. Try to find the ambushing socket. 838 * Because we've rewritten the destination address, 839 * any hardware-generated hash is ignored. 840 */ 841 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 842 th->th_sport, next_hop->sin_addr, 843 next_hop->sin_port ? ntohs(next_hop->sin_port) : 844 th->th_dport, INPLOOKUP_WILDCARD | 845 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 846 } 847 /* Remove the tag from the packet. We don't need it anymore. */ 848 m_tag_delete(m, fwd_tag); 849 m->m_flags &= ~M_IP_NEXTHOP; 850 } else 851 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 852 th->th_sport, ip->ip_dst, th->th_dport, 853 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 854 m->m_pkthdr.rcvif, m); 855 #endif /* INET */ 856 857 /* 858 * If the INPCB does not exist then all data in the incoming 859 * segment is discarded and an appropriate RST is sent back. 860 * XXX MRT Send RST using which routing table? 861 */ 862 if (inp == NULL) { 863 /* 864 * Log communication attempts to ports that are not 865 * in use. 866 */ 867 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 868 tcp_log_in_vain == 2) { 869 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 870 log(LOG_INFO, "%s; %s: Connection attempt " 871 "to closed port\n", s, __func__); 872 } 873 /* 874 * When blackholing do not respond with a RST but 875 * completely ignore the segment and drop it. 876 */ 877 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 878 V_blackhole == 2) 879 goto dropunlock; 880 881 rstreason = BANDLIM_RST_CLOSEDPORT; 882 goto dropwithreset; 883 } 884 INP_WLOCK_ASSERT(inp); 885 if (!(inp->inp_flags & INP_HW_FLOWID) 886 && (m->m_flags & M_FLOWID) 887 && ((inp->inp_socket == NULL) 888 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 889 inp->inp_flags |= INP_HW_FLOWID; 890 inp->inp_flags &= ~INP_SW_FLOWID; 891 inp->inp_flowid = m->m_pkthdr.flowid; 892 } 893 #ifdef IPSEC 894 #ifdef INET6 895 if (isipv6 && ipsec6_in_reject(m, inp)) { 896 V_ipsec6stat.in_polvio++; 897 goto dropunlock; 898 } else 899 #endif /* INET6 */ 900 if (ipsec4_in_reject(m, inp) != 0) { 901 V_ipsec4stat.in_polvio++; 902 goto dropunlock; 903 } 904 #endif /* IPSEC */ 905 906 /* 907 * Check the minimum TTL for socket. 908 */ 909 if (inp->inp_ip_minttl != 0) { 910 #ifdef INET6 911 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 912 goto dropunlock; 913 else 914 #endif 915 if (inp->inp_ip_minttl > ip->ip_ttl) 916 goto dropunlock; 917 } 918 919 /* 920 * A previous connection in TIMEWAIT state is supposed to catch stray 921 * or duplicate segments arriving late. If this segment was a 922 * legitimate new connection attempt, the old INPCB gets removed and 923 * we can try again to find a listening socket. 924 * 925 * At this point, due to earlier optimism, we may hold only an inpcb 926 * lock, and not the inpcbinfo write lock. If so, we need to try to 927 * acquire it, or if that fails, acquire a reference on the inpcb, 928 * drop all locks, acquire a global write lock, and then re-acquire 929 * the inpcb lock. We may at that point discover that another thread 930 * has tried to free the inpcb, in which case we need to loop back 931 * and try to find a new inpcb to deliver to. 932 * 933 * XXXRW: It may be time to rethink timewait locking. 934 */ 935 relocked: 936 if (inp->inp_flags & INP_TIMEWAIT) { 937 if (ti_locked == TI_UNLOCKED) { 938 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 939 in_pcbref(inp); 940 INP_WUNLOCK(inp); 941 INP_INFO_WLOCK(&V_tcbinfo); 942 ti_locked = TI_WLOCKED; 943 INP_WLOCK(inp); 944 if (in_pcbrele_wlocked(inp)) { 945 inp = NULL; 946 goto findpcb; 947 } 948 } else 949 ti_locked = TI_WLOCKED; 950 } 951 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 952 953 if (thflags & TH_SYN) 954 tcp_dooptions(&to, optp, optlen, TO_SYN); 955 /* 956 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 957 */ 958 if (tcp_twcheck(inp, &to, th, m, tlen)) 959 goto findpcb; 960 INP_INFO_WUNLOCK(&V_tcbinfo); 961 return; 962 } 963 /* 964 * The TCPCB may no longer exist if the connection is winding 965 * down or it is in the CLOSED state. Either way we drop the 966 * segment and send an appropriate response. 967 */ 968 tp = intotcpcb(inp); 969 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 970 rstreason = BANDLIM_RST_CLOSEDPORT; 971 goto dropwithreset; 972 } 973 974 #ifdef TCP_OFFLOAD 975 if (tp->t_flags & TF_TOE) { 976 tcp_offload_input(tp, m); 977 m = NULL; /* consumed by the TOE driver */ 978 goto dropunlock; 979 } 980 #endif 981 982 /* 983 * We've identified a valid inpcb, but it could be that we need an 984 * inpcbinfo write lock but don't hold it. In this case, attempt to 985 * acquire using the same strategy as the TIMEWAIT case above. If we 986 * relock, we have to jump back to 'relocked' as the connection might 987 * now be in TIMEWAIT. 988 */ 989 #ifdef INVARIANTS 990 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 991 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 992 #endif 993 if (tp->t_state != TCPS_ESTABLISHED) { 994 if (ti_locked == TI_UNLOCKED) { 995 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 996 in_pcbref(inp); 997 INP_WUNLOCK(inp); 998 INP_INFO_WLOCK(&V_tcbinfo); 999 ti_locked = TI_WLOCKED; 1000 INP_WLOCK(inp); 1001 if (in_pcbrele_wlocked(inp)) { 1002 inp = NULL; 1003 goto findpcb; 1004 } 1005 goto relocked; 1006 } else 1007 ti_locked = TI_WLOCKED; 1008 } 1009 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1010 } 1011 1012 #ifdef MAC 1013 INP_WLOCK_ASSERT(inp); 1014 if (mac_inpcb_check_deliver(inp, m)) 1015 goto dropunlock; 1016 #endif 1017 so = inp->inp_socket; 1018 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1019 #ifdef TCPDEBUG 1020 if (so->so_options & SO_DEBUG) { 1021 ostate = tp->t_state; 1022 #ifdef INET6 1023 if (isipv6) { 1024 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1025 } else 1026 #endif 1027 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1028 tcp_savetcp = *th; 1029 } 1030 #endif /* TCPDEBUG */ 1031 /* 1032 * When the socket is accepting connections (the INPCB is in LISTEN 1033 * state) we look into the SYN cache if this is a new connection 1034 * attempt or the completion of a previous one. Because listen 1035 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1036 * held in this case. 1037 */ 1038 if (so->so_options & SO_ACCEPTCONN) { 1039 struct in_conninfo inc; 1040 1041 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1042 "tp not listening", __func__)); 1043 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1044 1045 bzero(&inc, sizeof(inc)); 1046 #ifdef INET6 1047 if (isipv6) { 1048 inc.inc_flags |= INC_ISIPV6; 1049 inc.inc6_faddr = ip6->ip6_src; 1050 inc.inc6_laddr = ip6->ip6_dst; 1051 } else 1052 #endif 1053 { 1054 inc.inc_faddr = ip->ip_src; 1055 inc.inc_laddr = ip->ip_dst; 1056 } 1057 inc.inc_fport = th->th_sport; 1058 inc.inc_lport = th->th_dport; 1059 inc.inc_fibnum = so->so_fibnum; 1060 1061 /* 1062 * Check for an existing connection attempt in syncache if 1063 * the flag is only ACK. A successful lookup creates a new 1064 * socket appended to the listen queue in SYN_RECEIVED state. 1065 */ 1066 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1067 /* 1068 * Parse the TCP options here because 1069 * syncookies need access to the reflected 1070 * timestamp. 1071 */ 1072 tcp_dooptions(&to, optp, optlen, 0); 1073 /* 1074 * NB: syncache_expand() doesn't unlock 1075 * inp and tcpinfo locks. 1076 */ 1077 if (!syncache_expand(&inc, &to, th, &so, m)) { 1078 /* 1079 * No syncache entry or ACK was not 1080 * for our SYN/ACK. Send a RST. 1081 * NB: syncache did its own logging 1082 * of the failure cause. 1083 */ 1084 rstreason = BANDLIM_RST_OPENPORT; 1085 goto dropwithreset; 1086 } 1087 if (so == NULL) { 1088 /* 1089 * We completed the 3-way handshake 1090 * but could not allocate a socket 1091 * either due to memory shortage, 1092 * listen queue length limits or 1093 * global socket limits. Send RST 1094 * or wait and have the remote end 1095 * retransmit the ACK for another 1096 * try. 1097 */ 1098 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1099 log(LOG_DEBUG, "%s; %s: Listen socket: " 1100 "Socket allocation failed due to " 1101 "limits or memory shortage, %s\n", 1102 s, __func__, 1103 V_tcp_sc_rst_sock_fail ? 1104 "sending RST" : "try again"); 1105 if (V_tcp_sc_rst_sock_fail) { 1106 rstreason = BANDLIM_UNLIMITED; 1107 goto dropwithreset; 1108 } else 1109 goto dropunlock; 1110 } 1111 /* 1112 * Socket is created in state SYN_RECEIVED. 1113 * Unlock the listen socket, lock the newly 1114 * created socket and update the tp variable. 1115 */ 1116 INP_WUNLOCK(inp); /* listen socket */ 1117 inp = sotoinpcb(so); 1118 INP_WLOCK(inp); /* new connection */ 1119 tp = intotcpcb(inp); 1120 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1121 ("%s: ", __func__)); 1122 #ifdef TCP_SIGNATURE 1123 if (sig_checked == 0) { 1124 tcp_dooptions(&to, optp, optlen, 1125 (thflags & TH_SYN) ? TO_SYN : 0); 1126 if (!tcp_signature_verify_input(m, off0, tlen, 1127 optlen, &to, th, tp->t_flags)) { 1128 1129 /* 1130 * In SYN_SENT state if it receives an 1131 * RST, it is allowed for further 1132 * processing. 1133 */ 1134 if ((thflags & TH_RST) == 0 || 1135 (tp->t_state == TCPS_SYN_SENT) == 0) 1136 goto dropunlock; 1137 } 1138 sig_checked = 1; 1139 } 1140 #endif 1141 1142 /* 1143 * Process the segment and the data it 1144 * contains. tcp_do_segment() consumes 1145 * the mbuf chain and unlocks the inpcb. 1146 */ 1147 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1148 iptos, ti_locked); 1149 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1150 return; 1151 } 1152 /* 1153 * Segment flag validation for new connection attempts: 1154 * 1155 * Our (SYN|ACK) response was rejected. 1156 * Check with syncache and remove entry to prevent 1157 * retransmits. 1158 * 1159 * NB: syncache_chkrst does its own logging of failure 1160 * causes. 1161 */ 1162 if (thflags & TH_RST) { 1163 syncache_chkrst(&inc, th); 1164 goto dropunlock; 1165 } 1166 /* 1167 * We can't do anything without SYN. 1168 */ 1169 if ((thflags & TH_SYN) == 0) { 1170 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1171 log(LOG_DEBUG, "%s; %s: Listen socket: " 1172 "SYN is missing, segment ignored\n", 1173 s, __func__); 1174 TCPSTAT_INC(tcps_badsyn); 1175 goto dropunlock; 1176 } 1177 /* 1178 * (SYN|ACK) is bogus on a listen socket. 1179 */ 1180 if (thflags & TH_ACK) { 1181 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1182 log(LOG_DEBUG, "%s; %s: Listen socket: " 1183 "SYN|ACK invalid, segment rejected\n", 1184 s, __func__); 1185 syncache_badack(&inc); /* XXX: Not needed! */ 1186 TCPSTAT_INC(tcps_badsyn); 1187 rstreason = BANDLIM_RST_OPENPORT; 1188 goto dropwithreset; 1189 } 1190 /* 1191 * If the drop_synfin option is enabled, drop all 1192 * segments with both the SYN and FIN bits set. 1193 * This prevents e.g. nmap from identifying the 1194 * TCP/IP stack. 1195 * XXX: Poor reasoning. nmap has other methods 1196 * and is constantly refining its stack detection 1197 * strategies. 1198 * XXX: This is a violation of the TCP specification 1199 * and was used by RFC1644. 1200 */ 1201 if ((thflags & TH_FIN) && V_drop_synfin) { 1202 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1203 log(LOG_DEBUG, "%s; %s: Listen socket: " 1204 "SYN|FIN segment ignored (based on " 1205 "sysctl setting)\n", s, __func__); 1206 TCPSTAT_INC(tcps_badsyn); 1207 goto dropunlock; 1208 } 1209 /* 1210 * Segment's flags are (SYN) or (SYN|FIN). 1211 * 1212 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1213 * as they do not affect the state of the TCP FSM. 1214 * The data pointed to by TH_URG and th_urp is ignored. 1215 */ 1216 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1217 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1218 KASSERT(thflags & (TH_SYN), 1219 ("%s: Listen socket: TH_SYN not set", __func__)); 1220 #ifdef INET6 1221 /* 1222 * If deprecated address is forbidden, 1223 * we do not accept SYN to deprecated interface 1224 * address to prevent any new inbound connection from 1225 * getting established. 1226 * When we do not accept SYN, we send a TCP RST, 1227 * with deprecated source address (instead of dropping 1228 * it). We compromise it as it is much better for peer 1229 * to send a RST, and RST will be the final packet 1230 * for the exchange. 1231 * 1232 * If we do not forbid deprecated addresses, we accept 1233 * the SYN packet. RFC2462 does not suggest dropping 1234 * SYN in this case. 1235 * If we decipher RFC2462 5.5.4, it says like this: 1236 * 1. use of deprecated addr with existing 1237 * communication is okay - "SHOULD continue to be 1238 * used" 1239 * 2. use of it with new communication: 1240 * (2a) "SHOULD NOT be used if alternate address 1241 * with sufficient scope is available" 1242 * (2b) nothing mentioned otherwise. 1243 * Here we fall into (2b) case as we have no choice in 1244 * our source address selection - we must obey the peer. 1245 * 1246 * The wording in RFC2462 is confusing, and there are 1247 * multiple description text for deprecated address 1248 * handling - worse, they are not exactly the same. 1249 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1250 */ 1251 if (isipv6 && !V_ip6_use_deprecated) { 1252 struct in6_ifaddr *ia6; 1253 1254 ia6 = ip6_getdstifaddr(m); 1255 if (ia6 != NULL && 1256 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1257 ifa_free(&ia6->ia_ifa); 1258 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1259 log(LOG_DEBUG, "%s; %s: Listen socket: " 1260 "Connection attempt to deprecated " 1261 "IPv6 address rejected\n", 1262 s, __func__); 1263 rstreason = BANDLIM_RST_OPENPORT; 1264 goto dropwithreset; 1265 } 1266 if (ia6) 1267 ifa_free(&ia6->ia_ifa); 1268 } 1269 #endif /* INET6 */ 1270 /* 1271 * Basic sanity checks on incoming SYN requests: 1272 * Don't respond if the destination is a link layer 1273 * broadcast according to RFC1122 4.2.3.10, p. 104. 1274 * If it is from this socket it must be forged. 1275 * Don't respond if the source or destination is a 1276 * global or subnet broad- or multicast address. 1277 * Note that it is quite possible to receive unicast 1278 * link-layer packets with a broadcast IP address. Use 1279 * in_broadcast() to find them. 1280 */ 1281 if (m->m_flags & (M_BCAST|M_MCAST)) { 1282 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1283 log(LOG_DEBUG, "%s; %s: Listen socket: " 1284 "Connection attempt from broad- or multicast " 1285 "link layer address ignored\n", s, __func__); 1286 goto dropunlock; 1287 } 1288 #ifdef INET6 1289 if (isipv6) { 1290 if (th->th_dport == th->th_sport && 1291 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1292 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1293 log(LOG_DEBUG, "%s; %s: Listen socket: " 1294 "Connection attempt to/from self " 1295 "ignored\n", s, __func__); 1296 goto dropunlock; 1297 } 1298 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1299 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1300 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1301 log(LOG_DEBUG, "%s; %s: Listen socket: " 1302 "Connection attempt from/to multicast " 1303 "address ignored\n", s, __func__); 1304 goto dropunlock; 1305 } 1306 } 1307 #endif 1308 #if defined(INET) && defined(INET6) 1309 else 1310 #endif 1311 #ifdef INET 1312 { 1313 if (th->th_dport == th->th_sport && 1314 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from/to self " 1318 "ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1322 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1323 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1324 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1326 log(LOG_DEBUG, "%s; %s: Listen socket: " 1327 "Connection attempt from/to broad- " 1328 "or multicast address ignored\n", 1329 s, __func__); 1330 goto dropunlock; 1331 } 1332 } 1333 #endif 1334 /* 1335 * SYN appears to be valid. Create compressed TCP state 1336 * for syncache. 1337 */ 1338 #ifdef TCPDEBUG 1339 if (so->so_options & SO_DEBUG) 1340 tcp_trace(TA_INPUT, ostate, tp, 1341 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1342 #endif 1343 tcp_dooptions(&to, optp, optlen, TO_SYN); 1344 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1345 /* 1346 * Entry added to syncache and mbuf consumed. 1347 * Everything already unlocked by syncache_add(). 1348 */ 1349 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1350 return; 1351 } 1352 1353 #ifdef TCP_SIGNATURE 1354 if (sig_checked == 0) { 1355 tcp_dooptions(&to, optp, optlen, 1356 (thflags & TH_SYN) ? TO_SYN : 0); 1357 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1358 th, tp->t_flags)) { 1359 1360 /* 1361 * In SYN_SENT state if it receives an RST, it is 1362 * allowed for further processing. 1363 */ 1364 if ((thflags & TH_RST) == 0 || 1365 (tp->t_state == TCPS_SYN_SENT) == 0) 1366 goto dropunlock; 1367 } 1368 sig_checked = 1; 1369 } 1370 #endif 1371 1372 /* 1373 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1374 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1375 * the inpcb, and unlocks pcbinfo. 1376 */ 1377 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1378 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1379 return; 1380 1381 dropwithreset: 1382 if (ti_locked == TI_WLOCKED) { 1383 INP_INFO_WUNLOCK(&V_tcbinfo); 1384 ti_locked = TI_UNLOCKED; 1385 } 1386 #ifdef INVARIANTS 1387 else { 1388 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1389 "ti_locked: %d", __func__, ti_locked)); 1390 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1391 } 1392 #endif 1393 1394 if (inp != NULL) { 1395 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1396 INP_WUNLOCK(inp); 1397 } else 1398 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1399 m = NULL; /* mbuf chain got consumed. */ 1400 goto drop; 1401 1402 dropunlock: 1403 if (ti_locked == TI_WLOCKED) { 1404 INP_INFO_WUNLOCK(&V_tcbinfo); 1405 ti_locked = TI_UNLOCKED; 1406 } 1407 #ifdef INVARIANTS 1408 else { 1409 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1410 "ti_locked: %d", __func__, ti_locked)); 1411 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1412 } 1413 #endif 1414 1415 if (inp != NULL) 1416 INP_WUNLOCK(inp); 1417 1418 drop: 1419 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1420 if (s != NULL) 1421 free(s, M_TCPLOG); 1422 if (m != NULL) 1423 m_freem(m); 1424 } 1425 1426 static void 1427 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1428 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1429 int ti_locked) 1430 { 1431 int thflags, acked, ourfinisacked, needoutput = 0; 1432 int rstreason, todrop, win; 1433 u_long tiwin; 1434 struct tcpopt to; 1435 1436 #ifdef TCPDEBUG 1437 /* 1438 * The size of tcp_saveipgen must be the size of the max ip header, 1439 * now IPv6. 1440 */ 1441 u_char tcp_saveipgen[IP6_HDR_LEN]; 1442 struct tcphdr tcp_savetcp; 1443 short ostate = 0; 1444 #endif 1445 thflags = th->th_flags; 1446 tp->sackhint.last_sack_ack = 0; 1447 1448 /* 1449 * If this is either a state-changing packet or current state isn't 1450 * established, we require a write lock on tcbinfo. Otherwise, we 1451 * allow the tcbinfo to be in either alocked or unlocked, as the 1452 * caller may have unnecessarily acquired a write lock due to a race. 1453 */ 1454 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1455 tp->t_state != TCPS_ESTABLISHED) { 1456 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1457 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1458 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1459 } else { 1460 #ifdef INVARIANTS 1461 if (ti_locked == TI_WLOCKED) 1462 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1463 else { 1464 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1465 "ti_locked: %d", __func__, ti_locked)); 1466 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1467 } 1468 #endif 1469 } 1470 INP_WLOCK_ASSERT(tp->t_inpcb); 1471 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1472 __func__)); 1473 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1474 __func__)); 1475 1476 /* 1477 * Segment received on connection. 1478 * Reset idle time and keep-alive timer. 1479 * XXX: This should be done after segment 1480 * validation to ignore broken/spoofed segs. 1481 */ 1482 tp->t_rcvtime = ticks; 1483 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1484 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1485 1486 /* 1487 * Unscale the window into a 32-bit value. 1488 * For the SYN_SENT state the scale is zero. 1489 */ 1490 tiwin = th->th_win << tp->snd_scale; 1491 1492 /* 1493 * TCP ECN processing. 1494 */ 1495 if (tp->t_flags & TF_ECN_PERMIT) { 1496 if (thflags & TH_CWR) 1497 tp->t_flags &= ~TF_ECN_SND_ECE; 1498 switch (iptos & IPTOS_ECN_MASK) { 1499 case IPTOS_ECN_CE: 1500 tp->t_flags |= TF_ECN_SND_ECE; 1501 TCPSTAT_INC(tcps_ecn_ce); 1502 break; 1503 case IPTOS_ECN_ECT0: 1504 TCPSTAT_INC(tcps_ecn_ect0); 1505 break; 1506 case IPTOS_ECN_ECT1: 1507 TCPSTAT_INC(tcps_ecn_ect1); 1508 break; 1509 } 1510 /* Congestion experienced. */ 1511 if (thflags & TH_ECE) { 1512 cc_cong_signal(tp, th, CC_ECN); 1513 } 1514 } 1515 1516 /* 1517 * Parse options on any incoming segment. 1518 */ 1519 tcp_dooptions(&to, (u_char *)(th + 1), 1520 (th->th_off << 2) - sizeof(struct tcphdr), 1521 (thflags & TH_SYN) ? TO_SYN : 0); 1522 1523 /* 1524 * If echoed timestamp is later than the current time, 1525 * fall back to non RFC1323 RTT calculation. Normalize 1526 * timestamp if syncookies were used when this connection 1527 * was established. 1528 */ 1529 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1530 to.to_tsecr -= tp->ts_offset; 1531 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1532 to.to_tsecr = 0; 1533 } 1534 1535 /* 1536 * Process options only when we get SYN/ACK back. The SYN case 1537 * for incoming connections is handled in tcp_syncache. 1538 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1539 * or <SYN,ACK>) segment itself is never scaled. 1540 * XXX this is traditional behavior, may need to be cleaned up. 1541 */ 1542 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1543 if ((to.to_flags & TOF_SCALE) && 1544 (tp->t_flags & TF_REQ_SCALE)) { 1545 tp->t_flags |= TF_RCVD_SCALE; 1546 tp->snd_scale = to.to_wscale; 1547 } 1548 /* 1549 * Initial send window. It will be updated with 1550 * the next incoming segment to the scaled value. 1551 */ 1552 tp->snd_wnd = th->th_win; 1553 if (to.to_flags & TOF_TS) { 1554 tp->t_flags |= TF_RCVD_TSTMP; 1555 tp->ts_recent = to.to_tsval; 1556 tp->ts_recent_age = tcp_ts_getticks(); 1557 } 1558 if (to.to_flags & TOF_MSS) 1559 tcp_mss(tp, to.to_mss); 1560 if ((tp->t_flags & TF_SACK_PERMIT) && 1561 (to.to_flags & TOF_SACKPERM) == 0) 1562 tp->t_flags &= ~TF_SACK_PERMIT; 1563 } 1564 1565 /* 1566 * Header prediction: check for the two common cases 1567 * of a uni-directional data xfer. If the packet has 1568 * no control flags, is in-sequence, the window didn't 1569 * change and we're not retransmitting, it's a 1570 * candidate. If the length is zero and the ack moved 1571 * forward, we're the sender side of the xfer. Just 1572 * free the data acked & wake any higher level process 1573 * that was blocked waiting for space. If the length 1574 * is non-zero and the ack didn't move, we're the 1575 * receiver side. If we're getting packets in-order 1576 * (the reassembly queue is empty), add the data to 1577 * the socket buffer and note that we need a delayed ack. 1578 * Make sure that the hidden state-flags are also off. 1579 * Since we check for TCPS_ESTABLISHED first, it can only 1580 * be TH_NEEDSYN. 1581 */ 1582 if (tp->t_state == TCPS_ESTABLISHED && 1583 th->th_seq == tp->rcv_nxt && 1584 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1585 tp->snd_nxt == tp->snd_max && 1586 tiwin && tiwin == tp->snd_wnd && 1587 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1588 LIST_EMPTY(&tp->t_segq) && 1589 ((to.to_flags & TOF_TS) == 0 || 1590 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1591 1592 /* 1593 * If last ACK falls within this segment's sequence numbers, 1594 * record the timestamp. 1595 * NOTE that the test is modified according to the latest 1596 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1597 */ 1598 if ((to.to_flags & TOF_TS) != 0 && 1599 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1600 tp->ts_recent_age = tcp_ts_getticks(); 1601 tp->ts_recent = to.to_tsval; 1602 } 1603 1604 if (tlen == 0) { 1605 if (SEQ_GT(th->th_ack, tp->snd_una) && 1606 SEQ_LEQ(th->th_ack, tp->snd_max) && 1607 !IN_RECOVERY(tp->t_flags) && 1608 (to.to_flags & TOF_SACK) == 0 && 1609 TAILQ_EMPTY(&tp->snd_holes)) { 1610 /* 1611 * This is a pure ack for outstanding data. 1612 */ 1613 if (ti_locked == TI_WLOCKED) 1614 INP_INFO_WUNLOCK(&V_tcbinfo); 1615 ti_locked = TI_UNLOCKED; 1616 1617 TCPSTAT_INC(tcps_predack); 1618 1619 /* 1620 * "bad retransmit" recovery. 1621 */ 1622 if (tp->t_rxtshift == 1 && 1623 tp->t_flags & TF_PREVVALID && 1624 (int)(ticks - tp->t_badrxtwin) < 0) { 1625 cc_cong_signal(tp, th, CC_RTO_ERR); 1626 } 1627 1628 /* 1629 * Recalculate the transmit timer / rtt. 1630 * 1631 * Some boxes send broken timestamp replies 1632 * during the SYN+ACK phase, ignore 1633 * timestamps of 0 or we could calculate a 1634 * huge RTT and blow up the retransmit timer. 1635 */ 1636 if ((to.to_flags & TOF_TS) != 0 && 1637 to.to_tsecr) { 1638 u_int t; 1639 1640 t = tcp_ts_getticks() - to.to_tsecr; 1641 if (!tp->t_rttlow || tp->t_rttlow > t) 1642 tp->t_rttlow = t; 1643 tcp_xmit_timer(tp, 1644 TCP_TS_TO_TICKS(t) + 1); 1645 } else if (tp->t_rtttime && 1646 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1647 if (!tp->t_rttlow || 1648 tp->t_rttlow > ticks - tp->t_rtttime) 1649 tp->t_rttlow = ticks - tp->t_rtttime; 1650 tcp_xmit_timer(tp, 1651 ticks - tp->t_rtttime); 1652 } 1653 acked = BYTES_THIS_ACK(tp, th); 1654 1655 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1656 hhook_run_tcp_est_in(tp, th, &to); 1657 1658 TCPSTAT_INC(tcps_rcvackpack); 1659 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1660 sbdrop(&so->so_snd, acked); 1661 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1662 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1663 tp->snd_recover = th->th_ack - 1; 1664 1665 /* 1666 * Let the congestion control algorithm update 1667 * congestion control related information. This 1668 * typically means increasing the congestion 1669 * window. 1670 */ 1671 cc_ack_received(tp, th, CC_ACK); 1672 1673 tp->snd_una = th->th_ack; 1674 /* 1675 * Pull snd_wl2 up to prevent seq wrap relative 1676 * to th_ack. 1677 */ 1678 tp->snd_wl2 = th->th_ack; 1679 tp->t_dupacks = 0; 1680 m_freem(m); 1681 ND6_HINT(tp); /* Some progress has been made. */ 1682 1683 /* 1684 * If all outstanding data are acked, stop 1685 * retransmit timer, otherwise restart timer 1686 * using current (possibly backed-off) value. 1687 * If process is waiting for space, 1688 * wakeup/selwakeup/signal. If data 1689 * are ready to send, let tcp_output 1690 * decide between more output or persist. 1691 */ 1692 #ifdef TCPDEBUG 1693 if (so->so_options & SO_DEBUG) 1694 tcp_trace(TA_INPUT, ostate, tp, 1695 (void *)tcp_saveipgen, 1696 &tcp_savetcp, 0); 1697 #endif 1698 if (tp->snd_una == tp->snd_max) 1699 tcp_timer_activate(tp, TT_REXMT, 0); 1700 else if (!tcp_timer_active(tp, TT_PERSIST)) 1701 tcp_timer_activate(tp, TT_REXMT, 1702 tp->t_rxtcur); 1703 sowwakeup(so); 1704 if (so->so_snd.sb_cc) 1705 (void) tcp_output(tp); 1706 goto check_delack; 1707 } 1708 } else if (th->th_ack == tp->snd_una && 1709 tlen <= sbspace(&so->so_rcv)) { 1710 int newsize = 0; /* automatic sockbuf scaling */ 1711 1712 /* 1713 * This is a pure, in-sequence data packet with 1714 * nothing on the reassembly queue and we have enough 1715 * buffer space to take it. 1716 */ 1717 if (ti_locked == TI_WLOCKED) 1718 INP_INFO_WUNLOCK(&V_tcbinfo); 1719 ti_locked = TI_UNLOCKED; 1720 1721 /* Clean receiver SACK report if present */ 1722 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1723 tcp_clean_sackreport(tp); 1724 TCPSTAT_INC(tcps_preddat); 1725 tp->rcv_nxt += tlen; 1726 /* 1727 * Pull snd_wl1 up to prevent seq wrap relative to 1728 * th_seq. 1729 */ 1730 tp->snd_wl1 = th->th_seq + tlen; 1731 /* 1732 * Pull rcv_up up to prevent seq wrap relative to 1733 * rcv_nxt. 1734 */ 1735 tp->rcv_up = tp->rcv_nxt; 1736 TCPSTAT_INC(tcps_rcvpack); 1737 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1738 ND6_HINT(tp); /* Some progress has been made */ 1739 #ifdef TCPDEBUG 1740 if (so->so_options & SO_DEBUG) 1741 tcp_trace(TA_INPUT, ostate, tp, 1742 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1743 #endif 1744 /* 1745 * Automatic sizing of receive socket buffer. Often the send 1746 * buffer size is not optimally adjusted to the actual network 1747 * conditions at hand (delay bandwidth product). Setting the 1748 * buffer size too small limits throughput on links with high 1749 * bandwidth and high delay (eg. trans-continental/oceanic links). 1750 * 1751 * On the receive side the socket buffer memory is only rarely 1752 * used to any significant extent. This allows us to be much 1753 * more aggressive in scaling the receive socket buffer. For 1754 * the case that the buffer space is actually used to a large 1755 * extent and we run out of kernel memory we can simply drop 1756 * the new segments; TCP on the sender will just retransmit it 1757 * later. Setting the buffer size too big may only consume too 1758 * much kernel memory if the application doesn't read() from 1759 * the socket or packet loss or reordering makes use of the 1760 * reassembly queue. 1761 * 1762 * The criteria to step up the receive buffer one notch are: 1763 * 1. the number of bytes received during the time it takes 1764 * one timestamp to be reflected back to us (the RTT); 1765 * 2. received bytes per RTT is within seven eighth of the 1766 * current socket buffer size; 1767 * 3. receive buffer size has not hit maximal automatic size; 1768 * 1769 * This algorithm does one step per RTT at most and only if 1770 * we receive a bulk stream w/o packet losses or reorderings. 1771 * Shrinking the buffer during idle times is not necessary as 1772 * it doesn't consume any memory when idle. 1773 * 1774 * TODO: Only step up if the application is actually serving 1775 * the buffer to better manage the socket buffer resources. 1776 */ 1777 if (V_tcp_do_autorcvbuf && 1778 to.to_tsecr && 1779 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1780 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1781 to.to_tsecr - tp->rfbuf_ts < hz) { 1782 if (tp->rfbuf_cnt > 1783 (so->so_rcv.sb_hiwat / 8 * 7) && 1784 so->so_rcv.sb_hiwat < 1785 V_tcp_autorcvbuf_max) { 1786 newsize = 1787 min(so->so_rcv.sb_hiwat + 1788 V_tcp_autorcvbuf_inc, 1789 V_tcp_autorcvbuf_max); 1790 } 1791 /* Start over with next RTT. */ 1792 tp->rfbuf_ts = 0; 1793 tp->rfbuf_cnt = 0; 1794 } else 1795 tp->rfbuf_cnt += tlen; /* add up */ 1796 } 1797 1798 /* Add data to socket buffer. */ 1799 SOCKBUF_LOCK(&so->so_rcv); 1800 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1801 m_freem(m); 1802 } else { 1803 /* 1804 * Set new socket buffer size. 1805 * Give up when limit is reached. 1806 */ 1807 if (newsize) 1808 if (!sbreserve_locked(&so->so_rcv, 1809 newsize, so, NULL)) 1810 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1811 m_adj(m, drop_hdrlen); /* delayed header drop */ 1812 sbappendstream_locked(&so->so_rcv, m); 1813 } 1814 /* NB: sorwakeup_locked() does an implicit unlock. */ 1815 sorwakeup_locked(so); 1816 if (DELAY_ACK(tp)) { 1817 tp->t_flags |= TF_DELACK; 1818 } else { 1819 tp->t_flags |= TF_ACKNOW; 1820 tcp_output(tp); 1821 } 1822 goto check_delack; 1823 } 1824 } 1825 1826 /* 1827 * Calculate amount of space in receive window, 1828 * and then do TCP input processing. 1829 * Receive window is amount of space in rcv queue, 1830 * but not less than advertised window. 1831 */ 1832 win = sbspace(&so->so_rcv); 1833 if (win < 0) 1834 win = 0; 1835 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1836 1837 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1838 tp->rfbuf_ts = 0; 1839 tp->rfbuf_cnt = 0; 1840 1841 switch (tp->t_state) { 1842 1843 /* 1844 * If the state is SYN_RECEIVED: 1845 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1846 */ 1847 case TCPS_SYN_RECEIVED: 1848 if ((thflags & TH_ACK) && 1849 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1850 SEQ_GT(th->th_ack, tp->snd_max))) { 1851 rstreason = BANDLIM_RST_OPENPORT; 1852 goto dropwithreset; 1853 } 1854 break; 1855 1856 /* 1857 * If the state is SYN_SENT: 1858 * if seg contains an ACK, but not for our SYN, drop the input. 1859 * if seg contains a RST, then drop the connection. 1860 * if seg does not contain SYN, then drop it. 1861 * Otherwise this is an acceptable SYN segment 1862 * initialize tp->rcv_nxt and tp->irs 1863 * if seg contains ack then advance tp->snd_una 1864 * if seg contains an ECE and ECN support is enabled, the stream 1865 * is ECN capable. 1866 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1867 * arrange for segment to be acked (eventually) 1868 * continue processing rest of data/controls, beginning with URG 1869 */ 1870 case TCPS_SYN_SENT: 1871 if ((thflags & TH_ACK) && 1872 (SEQ_LEQ(th->th_ack, tp->iss) || 1873 SEQ_GT(th->th_ack, tp->snd_max))) { 1874 rstreason = BANDLIM_UNLIMITED; 1875 goto dropwithreset; 1876 } 1877 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1878 tp = tcp_drop(tp, ECONNREFUSED); 1879 if (thflags & TH_RST) 1880 goto drop; 1881 if (!(thflags & TH_SYN)) 1882 goto drop; 1883 1884 tp->irs = th->th_seq; 1885 tcp_rcvseqinit(tp); 1886 if (thflags & TH_ACK) { 1887 TCPSTAT_INC(tcps_connects); 1888 soisconnected(so); 1889 #ifdef MAC 1890 mac_socketpeer_set_from_mbuf(m, so); 1891 #endif 1892 /* Do window scaling on this connection? */ 1893 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1894 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1895 tp->rcv_scale = tp->request_r_scale; 1896 } 1897 tp->rcv_adv += imin(tp->rcv_wnd, 1898 TCP_MAXWIN << tp->rcv_scale); 1899 tp->snd_una++; /* SYN is acked */ 1900 /* 1901 * If there's data, delay ACK; if there's also a FIN 1902 * ACKNOW will be turned on later. 1903 */ 1904 if (DELAY_ACK(tp) && tlen != 0) 1905 tcp_timer_activate(tp, TT_DELACK, 1906 tcp_delacktime); 1907 else 1908 tp->t_flags |= TF_ACKNOW; 1909 1910 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1911 tp->t_flags |= TF_ECN_PERMIT; 1912 TCPSTAT_INC(tcps_ecn_shs); 1913 } 1914 1915 /* 1916 * Received <SYN,ACK> in SYN_SENT[*] state. 1917 * Transitions: 1918 * SYN_SENT --> ESTABLISHED 1919 * SYN_SENT* --> FIN_WAIT_1 1920 */ 1921 tp->t_starttime = ticks; 1922 if (tp->t_flags & TF_NEEDFIN) { 1923 tp->t_state = TCPS_FIN_WAIT_1; 1924 tp->t_flags &= ~TF_NEEDFIN; 1925 thflags &= ~TH_SYN; 1926 } else { 1927 tp->t_state = TCPS_ESTABLISHED; 1928 cc_conn_init(tp); 1929 tcp_timer_activate(tp, TT_KEEP, 1930 TP_KEEPIDLE(tp)); 1931 } 1932 } else { 1933 /* 1934 * Received initial SYN in SYN-SENT[*] state => 1935 * simultaneous open. If segment contains CC option 1936 * and there is a cached CC, apply TAO test. 1937 * If it succeeds, connection is * half-synchronized. 1938 * Otherwise, do 3-way handshake: 1939 * SYN-SENT -> SYN-RECEIVED 1940 * SYN-SENT* -> SYN-RECEIVED* 1941 * If there was no CC option, clear cached CC value. 1942 */ 1943 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1944 tcp_timer_activate(tp, TT_REXMT, 0); 1945 tp->t_state = TCPS_SYN_RECEIVED; 1946 } 1947 1948 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1949 "ti_locked %d", __func__, ti_locked)); 1950 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1951 INP_WLOCK_ASSERT(tp->t_inpcb); 1952 1953 /* 1954 * Advance th->th_seq to correspond to first data byte. 1955 * If data, trim to stay within window, 1956 * dropping FIN if necessary. 1957 */ 1958 th->th_seq++; 1959 if (tlen > tp->rcv_wnd) { 1960 todrop = tlen - tp->rcv_wnd; 1961 m_adj(m, -todrop); 1962 tlen = tp->rcv_wnd; 1963 thflags &= ~TH_FIN; 1964 TCPSTAT_INC(tcps_rcvpackafterwin); 1965 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1966 } 1967 tp->snd_wl1 = th->th_seq - 1; 1968 tp->rcv_up = th->th_seq; 1969 /* 1970 * Client side of transaction: already sent SYN and data. 1971 * If the remote host used T/TCP to validate the SYN, 1972 * our data will be ACK'd; if so, enter normal data segment 1973 * processing in the middle of step 5, ack processing. 1974 * Otherwise, goto step 6. 1975 */ 1976 if (thflags & TH_ACK) 1977 goto process_ACK; 1978 1979 goto step6; 1980 1981 /* 1982 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1983 * do normal processing. 1984 * 1985 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1986 */ 1987 case TCPS_LAST_ACK: 1988 case TCPS_CLOSING: 1989 break; /* continue normal processing */ 1990 } 1991 1992 /* 1993 * States other than LISTEN or SYN_SENT. 1994 * First check the RST flag and sequence number since reset segments 1995 * are exempt from the timestamp and connection count tests. This 1996 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1997 * below which allowed reset segments in half the sequence space 1998 * to fall though and be processed (which gives forged reset 1999 * segments with a random sequence number a 50 percent chance of 2000 * killing a connection). 2001 * Then check timestamp, if present. 2002 * Then check the connection count, if present. 2003 * Then check that at least some bytes of segment are within 2004 * receive window. If segment begins before rcv_nxt, 2005 * drop leading data (and SYN); if nothing left, just ack. 2006 * 2007 * 2008 * If the RST bit is set, check the sequence number to see 2009 * if this is a valid reset segment. 2010 * RFC 793 page 37: 2011 * In all states except SYN-SENT, all reset (RST) segments 2012 * are validated by checking their SEQ-fields. A reset is 2013 * valid if its sequence number is in the window. 2014 * Note: this does not take into account delayed ACKs, so 2015 * we should test against last_ack_sent instead of rcv_nxt. 2016 * The sequence number in the reset segment is normally an 2017 * echo of our outgoing acknowlegement numbers, but some hosts 2018 * send a reset with the sequence number at the rightmost edge 2019 * of our receive window, and we have to handle this case. 2020 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2021 * that brute force RST attacks are possible. To combat this, 2022 * we use a much stricter check while in the ESTABLISHED state, 2023 * only accepting RSTs where the sequence number is equal to 2024 * last_ack_sent. In all other states (the states in which a 2025 * RST is more likely), the more permissive check is used. 2026 * If we have multiple segments in flight, the initial reset 2027 * segment sequence numbers will be to the left of last_ack_sent, 2028 * but they will eventually catch up. 2029 * In any case, it never made sense to trim reset segments to 2030 * fit the receive window since RFC 1122 says: 2031 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2032 * 2033 * A TCP SHOULD allow a received RST segment to include data. 2034 * 2035 * DISCUSSION 2036 * It has been suggested that a RST segment could contain 2037 * ASCII text that encoded and explained the cause of the 2038 * RST. No standard has yet been established for such 2039 * data. 2040 * 2041 * If the reset segment passes the sequence number test examine 2042 * the state: 2043 * SYN_RECEIVED STATE: 2044 * If passive open, return to LISTEN state. 2045 * If active open, inform user that connection was refused. 2046 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2047 * Inform user that connection was reset, and close tcb. 2048 * CLOSING, LAST_ACK STATES: 2049 * Close the tcb. 2050 * TIME_WAIT STATE: 2051 * Drop the segment - see Stevens, vol. 2, p. 964 and 2052 * RFC 1337. 2053 */ 2054 if (thflags & TH_RST) { 2055 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2056 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2057 switch (tp->t_state) { 2058 2059 case TCPS_SYN_RECEIVED: 2060 so->so_error = ECONNREFUSED; 2061 goto close; 2062 2063 case TCPS_ESTABLISHED: 2064 if (V_tcp_insecure_rst == 0 && 2065 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2066 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2067 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2068 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2069 TCPSTAT_INC(tcps_badrst); 2070 goto drop; 2071 } 2072 /* FALLTHROUGH */ 2073 case TCPS_FIN_WAIT_1: 2074 case TCPS_FIN_WAIT_2: 2075 case TCPS_CLOSE_WAIT: 2076 so->so_error = ECONNRESET; 2077 close: 2078 KASSERT(ti_locked == TI_WLOCKED, 2079 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2080 ti_locked)); 2081 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2082 2083 tp->t_state = TCPS_CLOSED; 2084 TCPSTAT_INC(tcps_drops); 2085 tp = tcp_close(tp); 2086 break; 2087 2088 case TCPS_CLOSING: 2089 case TCPS_LAST_ACK: 2090 KASSERT(ti_locked == TI_WLOCKED, 2091 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2092 ti_locked)); 2093 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2094 2095 tp = tcp_close(tp); 2096 break; 2097 } 2098 } 2099 goto drop; 2100 } 2101 2102 /* 2103 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2104 * and it's less than ts_recent, drop it. 2105 */ 2106 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2107 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2108 2109 /* Check to see if ts_recent is over 24 days old. */ 2110 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2111 /* 2112 * Invalidate ts_recent. If this segment updates 2113 * ts_recent, the age will be reset later and ts_recent 2114 * will get a valid value. If it does not, setting 2115 * ts_recent to zero will at least satisfy the 2116 * requirement that zero be placed in the timestamp 2117 * echo reply when ts_recent isn't valid. The 2118 * age isn't reset until we get a valid ts_recent 2119 * because we don't want out-of-order segments to be 2120 * dropped when ts_recent is old. 2121 */ 2122 tp->ts_recent = 0; 2123 } else { 2124 TCPSTAT_INC(tcps_rcvduppack); 2125 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2126 TCPSTAT_INC(tcps_pawsdrop); 2127 if (tlen) 2128 goto dropafterack; 2129 goto drop; 2130 } 2131 } 2132 2133 /* 2134 * In the SYN-RECEIVED state, validate that the packet belongs to 2135 * this connection before trimming the data to fit the receive 2136 * window. Check the sequence number versus IRS since we know 2137 * the sequence numbers haven't wrapped. This is a partial fix 2138 * for the "LAND" DoS attack. 2139 */ 2140 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2141 rstreason = BANDLIM_RST_OPENPORT; 2142 goto dropwithreset; 2143 } 2144 2145 todrop = tp->rcv_nxt - th->th_seq; 2146 if (todrop > 0) { 2147 /* 2148 * If this is a duplicate SYN for our current connection, 2149 * advance over it and pretend and it's not a SYN. 2150 */ 2151 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2152 thflags &= ~TH_SYN; 2153 th->th_seq++; 2154 if (th->th_urp > 1) 2155 th->th_urp--; 2156 else 2157 thflags &= ~TH_URG; 2158 todrop--; 2159 } 2160 /* 2161 * Following if statement from Stevens, vol. 2, p. 960. 2162 */ 2163 if (todrop > tlen 2164 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2165 /* 2166 * Any valid FIN must be to the left of the window. 2167 * At this point the FIN must be a duplicate or out 2168 * of sequence; drop it. 2169 */ 2170 thflags &= ~TH_FIN; 2171 2172 /* 2173 * Send an ACK to resynchronize and drop any data. 2174 * But keep on processing for RST or ACK. 2175 */ 2176 tp->t_flags |= TF_ACKNOW; 2177 todrop = tlen; 2178 TCPSTAT_INC(tcps_rcvduppack); 2179 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2180 } else { 2181 TCPSTAT_INC(tcps_rcvpartduppack); 2182 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2183 } 2184 drop_hdrlen += todrop; /* drop from the top afterwards */ 2185 th->th_seq += todrop; 2186 tlen -= todrop; 2187 if (th->th_urp > todrop) 2188 th->th_urp -= todrop; 2189 else { 2190 thflags &= ~TH_URG; 2191 th->th_urp = 0; 2192 } 2193 } 2194 2195 /* 2196 * If new data are received on a connection after the 2197 * user processes are gone, then RST the other end. 2198 */ 2199 if ((so->so_state & SS_NOFDREF) && 2200 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2201 char *s; 2202 2203 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2204 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2205 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2206 2207 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 2208 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 2209 "was closed, sending RST and removing tcpcb\n", 2210 s, __func__, tcpstates[tp->t_state], tlen); 2211 free(s, M_TCPLOG); 2212 } 2213 tp = tcp_close(tp); 2214 TCPSTAT_INC(tcps_rcvafterclose); 2215 rstreason = BANDLIM_UNLIMITED; 2216 goto dropwithreset; 2217 } 2218 2219 /* 2220 * If segment ends after window, drop trailing data 2221 * (and PUSH and FIN); if nothing left, just ACK. 2222 */ 2223 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2224 if (todrop > 0) { 2225 TCPSTAT_INC(tcps_rcvpackafterwin); 2226 if (todrop >= tlen) { 2227 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2228 /* 2229 * If window is closed can only take segments at 2230 * window edge, and have to drop data and PUSH from 2231 * incoming segments. Continue processing, but 2232 * remember to ack. Otherwise, drop segment 2233 * and ack. 2234 */ 2235 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2236 tp->t_flags |= TF_ACKNOW; 2237 TCPSTAT_INC(tcps_rcvwinprobe); 2238 } else 2239 goto dropafterack; 2240 } else 2241 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2242 m_adj(m, -todrop); 2243 tlen -= todrop; 2244 thflags &= ~(TH_PUSH|TH_FIN); 2245 } 2246 2247 /* 2248 * If last ACK falls within this segment's sequence numbers, 2249 * record its timestamp. 2250 * NOTE: 2251 * 1) That the test incorporates suggestions from the latest 2252 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2253 * 2) That updating only on newer timestamps interferes with 2254 * our earlier PAWS tests, so this check should be solely 2255 * predicated on the sequence space of this segment. 2256 * 3) That we modify the segment boundary check to be 2257 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2258 * instead of RFC1323's 2259 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2260 * This modified check allows us to overcome RFC1323's 2261 * limitations as described in Stevens TCP/IP Illustrated 2262 * Vol. 2 p.869. In such cases, we can still calculate the 2263 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2264 */ 2265 if ((to.to_flags & TOF_TS) != 0 && 2266 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2267 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2268 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2269 tp->ts_recent_age = tcp_ts_getticks(); 2270 tp->ts_recent = to.to_tsval; 2271 } 2272 2273 /* 2274 * If a SYN is in the window, then this is an 2275 * error and we send an RST and drop the connection. 2276 */ 2277 if (thflags & TH_SYN) { 2278 KASSERT(ti_locked == TI_WLOCKED, 2279 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2280 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2281 2282 tp = tcp_drop(tp, ECONNRESET); 2283 rstreason = BANDLIM_UNLIMITED; 2284 goto drop; 2285 } 2286 2287 /* 2288 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2289 * flag is on (half-synchronized state), then queue data for 2290 * later processing; else drop segment and return. 2291 */ 2292 if ((thflags & TH_ACK) == 0) { 2293 if (tp->t_state == TCPS_SYN_RECEIVED || 2294 (tp->t_flags & TF_NEEDSYN)) 2295 goto step6; 2296 else if (tp->t_flags & TF_ACKNOW) 2297 goto dropafterack; 2298 else 2299 goto drop; 2300 } 2301 2302 /* 2303 * Ack processing. 2304 */ 2305 switch (tp->t_state) { 2306 2307 /* 2308 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2309 * ESTABLISHED state and continue processing. 2310 * The ACK was checked above. 2311 */ 2312 case TCPS_SYN_RECEIVED: 2313 2314 TCPSTAT_INC(tcps_connects); 2315 soisconnected(so); 2316 /* Do window scaling? */ 2317 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2318 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2319 tp->rcv_scale = tp->request_r_scale; 2320 tp->snd_wnd = tiwin; 2321 } 2322 /* 2323 * Make transitions: 2324 * SYN-RECEIVED -> ESTABLISHED 2325 * SYN-RECEIVED* -> FIN-WAIT-1 2326 */ 2327 tp->t_starttime = ticks; 2328 if (tp->t_flags & TF_NEEDFIN) { 2329 tp->t_state = TCPS_FIN_WAIT_1; 2330 tp->t_flags &= ~TF_NEEDFIN; 2331 } else { 2332 tp->t_state = TCPS_ESTABLISHED; 2333 cc_conn_init(tp); 2334 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2335 } 2336 /* 2337 * If segment contains data or ACK, will call tcp_reass() 2338 * later; if not, do so now to pass queued data to user. 2339 */ 2340 if (tlen == 0 && (thflags & TH_FIN) == 0) 2341 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2342 (struct mbuf *)0); 2343 /* FALLTHROUGH */ 2344 2345 /* 2346 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2347 * ACKs. If the ack is in the range 2348 * tp->snd_una < th->th_ack <= tp->snd_max 2349 * then advance tp->snd_una to th->th_ack and drop 2350 * data from the retransmission queue. If this ACK reflects 2351 * more up to date window information we update our window information. 2352 */ 2353 case TCPS_ESTABLISHED: 2354 case TCPS_FIN_WAIT_1: 2355 case TCPS_FIN_WAIT_2: 2356 case TCPS_CLOSE_WAIT: 2357 case TCPS_CLOSING: 2358 case TCPS_LAST_ACK: 2359 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2360 TCPSTAT_INC(tcps_rcvacktoomuch); 2361 goto dropafterack; 2362 } 2363 if ((tp->t_flags & TF_SACK_PERMIT) && 2364 ((to.to_flags & TOF_SACK) || 2365 !TAILQ_EMPTY(&tp->snd_holes))) 2366 tcp_sack_doack(tp, &to, th->th_ack); 2367 2368 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2369 hhook_run_tcp_est_in(tp, th, &to); 2370 2371 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2372 if (tlen == 0 && tiwin == tp->snd_wnd) { 2373 TCPSTAT_INC(tcps_rcvdupack); 2374 /* 2375 * If we have outstanding data (other than 2376 * a window probe), this is a completely 2377 * duplicate ack (ie, window info didn't 2378 * change), the ack is the biggest we've 2379 * seen and we've seen exactly our rexmt 2380 * threshhold of them, assume a packet 2381 * has been dropped and retransmit it. 2382 * Kludge snd_nxt & the congestion 2383 * window so we send only this one 2384 * packet. 2385 * 2386 * We know we're losing at the current 2387 * window size so do congestion avoidance 2388 * (set ssthresh to half the current window 2389 * and pull our congestion window back to 2390 * the new ssthresh). 2391 * 2392 * Dup acks mean that packets have left the 2393 * network (they're now cached at the receiver) 2394 * so bump cwnd by the amount in the receiver 2395 * to keep a constant cwnd packets in the 2396 * network. 2397 * 2398 * When using TCP ECN, notify the peer that 2399 * we reduced the cwnd. 2400 */ 2401 if (!tcp_timer_active(tp, TT_REXMT) || 2402 th->th_ack != tp->snd_una) 2403 tp->t_dupacks = 0; 2404 else if (++tp->t_dupacks > tcprexmtthresh || 2405 IN_FASTRECOVERY(tp->t_flags)) { 2406 cc_ack_received(tp, th, CC_DUPACK); 2407 if ((tp->t_flags & TF_SACK_PERMIT) && 2408 IN_FASTRECOVERY(tp->t_flags)) { 2409 int awnd; 2410 2411 /* 2412 * Compute the amount of data in flight first. 2413 * We can inject new data into the pipe iff 2414 * we have less than 1/2 the original window's 2415 * worth of data in flight. 2416 */ 2417 awnd = (tp->snd_nxt - tp->snd_fack) + 2418 tp->sackhint.sack_bytes_rexmit; 2419 if (awnd < tp->snd_ssthresh) { 2420 tp->snd_cwnd += tp->t_maxseg; 2421 if (tp->snd_cwnd > tp->snd_ssthresh) 2422 tp->snd_cwnd = tp->snd_ssthresh; 2423 } 2424 } else 2425 tp->snd_cwnd += tp->t_maxseg; 2426 if ((thflags & TH_FIN) && 2427 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2428 /* 2429 * If its a fin we need to process 2430 * it to avoid a race where both 2431 * sides enter FIN-WAIT and send FIN|ACK 2432 * at the same time. 2433 */ 2434 break; 2435 } 2436 (void) tcp_output(tp); 2437 goto drop; 2438 } else if (tp->t_dupacks == tcprexmtthresh) { 2439 tcp_seq onxt = tp->snd_nxt; 2440 2441 /* 2442 * If we're doing sack, check to 2443 * see if we're already in sack 2444 * recovery. If we're not doing sack, 2445 * check to see if we're in newreno 2446 * recovery. 2447 */ 2448 if (tp->t_flags & TF_SACK_PERMIT) { 2449 if (IN_FASTRECOVERY(tp->t_flags)) { 2450 tp->t_dupacks = 0; 2451 break; 2452 } 2453 } else { 2454 if (SEQ_LEQ(th->th_ack, 2455 tp->snd_recover)) { 2456 tp->t_dupacks = 0; 2457 break; 2458 } 2459 } 2460 /* Congestion signal before ack. */ 2461 cc_cong_signal(tp, th, CC_NDUPACK); 2462 cc_ack_received(tp, th, CC_DUPACK); 2463 tcp_timer_activate(tp, TT_REXMT, 0); 2464 tp->t_rtttime = 0; 2465 if (tp->t_flags & TF_SACK_PERMIT) { 2466 TCPSTAT_INC( 2467 tcps_sack_recovery_episode); 2468 tp->sack_newdata = tp->snd_nxt; 2469 tp->snd_cwnd = tp->t_maxseg; 2470 (void) tcp_output(tp); 2471 goto drop; 2472 } 2473 tp->snd_nxt = th->th_ack; 2474 tp->snd_cwnd = tp->t_maxseg; 2475 if ((thflags & TH_FIN) && 2476 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2477 /* 2478 * If its a fin we need to process 2479 * it to avoid a race where both 2480 * sides enter FIN-WAIT and send FIN|ACK 2481 * at the same time. 2482 */ 2483 break; 2484 } 2485 (void) tcp_output(tp); 2486 KASSERT(tp->snd_limited <= 2, 2487 ("%s: tp->snd_limited too big", 2488 __func__)); 2489 tp->snd_cwnd = tp->snd_ssthresh + 2490 tp->t_maxseg * 2491 (tp->t_dupacks - tp->snd_limited); 2492 if (SEQ_GT(onxt, tp->snd_nxt)) 2493 tp->snd_nxt = onxt; 2494 goto drop; 2495 } else if (V_tcp_do_rfc3042) { 2496 cc_ack_received(tp, th, CC_DUPACK); 2497 u_long oldcwnd = tp->snd_cwnd; 2498 tcp_seq oldsndmax = tp->snd_max; 2499 u_int sent; 2500 2501 KASSERT(tp->t_dupacks == 1 || 2502 tp->t_dupacks == 2, 2503 ("%s: dupacks not 1 or 2", 2504 __func__)); 2505 if (tp->t_dupacks == 1) 2506 tp->snd_limited = 0; 2507 tp->snd_cwnd = 2508 (tp->snd_nxt - tp->snd_una) + 2509 (tp->t_dupacks - tp->snd_limited) * 2510 tp->t_maxseg; 2511 if ((thflags & TH_FIN) && 2512 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2513 /* 2514 * If its a fin we need to process 2515 * it to avoid a race where both 2516 * sides enter FIN-WAIT and send FIN|ACK 2517 * at the same time. 2518 */ 2519 break; 2520 } 2521 (void) tcp_output(tp); 2522 sent = tp->snd_max - oldsndmax; 2523 if (sent > tp->t_maxseg) { 2524 KASSERT((tp->t_dupacks == 2 && 2525 tp->snd_limited == 0) || 2526 (sent == tp->t_maxseg + 1 && 2527 tp->t_flags & TF_SENTFIN), 2528 ("%s: sent too much", 2529 __func__)); 2530 tp->snd_limited = 2; 2531 } else if (sent > 0) 2532 ++tp->snd_limited; 2533 tp->snd_cwnd = oldcwnd; 2534 goto drop; 2535 } 2536 } else 2537 tp->t_dupacks = 0; 2538 break; 2539 } 2540 2541 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2542 ("%s: th_ack <= snd_una", __func__)); 2543 2544 /* 2545 * If the congestion window was inflated to account 2546 * for the other side's cached packets, retract it. 2547 */ 2548 if (IN_FASTRECOVERY(tp->t_flags)) { 2549 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2550 if (tp->t_flags & TF_SACK_PERMIT) 2551 tcp_sack_partialack(tp, th); 2552 else 2553 tcp_newreno_partial_ack(tp, th); 2554 } else 2555 cc_post_recovery(tp, th); 2556 } 2557 tp->t_dupacks = 0; 2558 /* 2559 * If we reach this point, ACK is not a duplicate, 2560 * i.e., it ACKs something we sent. 2561 */ 2562 if (tp->t_flags & TF_NEEDSYN) { 2563 /* 2564 * T/TCP: Connection was half-synchronized, and our 2565 * SYN has been ACK'd (so connection is now fully 2566 * synchronized). Go to non-starred state, 2567 * increment snd_una for ACK of SYN, and check if 2568 * we can do window scaling. 2569 */ 2570 tp->t_flags &= ~TF_NEEDSYN; 2571 tp->snd_una++; 2572 /* Do window scaling? */ 2573 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2574 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2575 tp->rcv_scale = tp->request_r_scale; 2576 /* Send window already scaled. */ 2577 } 2578 } 2579 2580 process_ACK: 2581 INP_WLOCK_ASSERT(tp->t_inpcb); 2582 2583 acked = BYTES_THIS_ACK(tp, th); 2584 TCPSTAT_INC(tcps_rcvackpack); 2585 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2586 2587 /* 2588 * If we just performed our first retransmit, and the ACK 2589 * arrives within our recovery window, then it was a mistake 2590 * to do the retransmit in the first place. Recover our 2591 * original cwnd and ssthresh, and proceed to transmit where 2592 * we left off. 2593 */ 2594 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2595 (int)(ticks - tp->t_badrxtwin) < 0) 2596 cc_cong_signal(tp, th, CC_RTO_ERR); 2597 2598 /* 2599 * If we have a timestamp reply, update smoothed 2600 * round trip time. If no timestamp is present but 2601 * transmit timer is running and timed sequence 2602 * number was acked, update smoothed round trip time. 2603 * Since we now have an rtt measurement, cancel the 2604 * timer backoff (cf., Phil Karn's retransmit alg.). 2605 * Recompute the initial retransmit timer. 2606 * 2607 * Some boxes send broken timestamp replies 2608 * during the SYN+ACK phase, ignore 2609 * timestamps of 0 or we could calculate a 2610 * huge RTT and blow up the retransmit timer. 2611 */ 2612 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2613 u_int t; 2614 2615 t = tcp_ts_getticks() - to.to_tsecr; 2616 if (!tp->t_rttlow || tp->t_rttlow > t) 2617 tp->t_rttlow = t; 2618 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2619 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2620 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2621 tp->t_rttlow = ticks - tp->t_rtttime; 2622 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2623 } 2624 2625 /* 2626 * If all outstanding data is acked, stop retransmit 2627 * timer and remember to restart (more output or persist). 2628 * If there is more data to be acked, restart retransmit 2629 * timer, using current (possibly backed-off) value. 2630 */ 2631 if (th->th_ack == tp->snd_max) { 2632 tcp_timer_activate(tp, TT_REXMT, 0); 2633 needoutput = 1; 2634 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2635 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2636 2637 /* 2638 * If no data (only SYN) was ACK'd, 2639 * skip rest of ACK processing. 2640 */ 2641 if (acked == 0) 2642 goto step6; 2643 2644 /* 2645 * Let the congestion control algorithm update congestion 2646 * control related information. This typically means increasing 2647 * the congestion window. 2648 */ 2649 cc_ack_received(tp, th, CC_ACK); 2650 2651 SOCKBUF_LOCK(&so->so_snd); 2652 if (acked > so->so_snd.sb_cc) { 2653 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2654 ourfinisacked = 1; 2655 } else { 2656 sbdrop_locked(&so->so_snd, acked); 2657 ourfinisacked = 0; 2658 } 2659 /* NB: sowwakeup_locked() does an implicit unlock. */ 2660 sowwakeup_locked(so); 2661 /* Detect una wraparound. */ 2662 if (!IN_RECOVERY(tp->t_flags) && 2663 SEQ_GT(tp->snd_una, tp->snd_recover) && 2664 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2665 tp->snd_recover = th->th_ack - 1; 2666 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2667 if (IN_RECOVERY(tp->t_flags) && 2668 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2669 EXIT_RECOVERY(tp->t_flags); 2670 } 2671 tp->snd_una = th->th_ack; 2672 if (tp->t_flags & TF_SACK_PERMIT) { 2673 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2674 tp->snd_recover = tp->snd_una; 2675 } 2676 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2677 tp->snd_nxt = tp->snd_una; 2678 2679 switch (tp->t_state) { 2680 2681 /* 2682 * In FIN_WAIT_1 STATE in addition to the processing 2683 * for the ESTABLISHED state if our FIN is now acknowledged 2684 * then enter FIN_WAIT_2. 2685 */ 2686 case TCPS_FIN_WAIT_1: 2687 if (ourfinisacked) { 2688 /* 2689 * If we can't receive any more 2690 * data, then closing user can proceed. 2691 * Starting the timer is contrary to the 2692 * specification, but if we don't get a FIN 2693 * we'll hang forever. 2694 * 2695 * XXXjl: 2696 * we should release the tp also, and use a 2697 * compressed state. 2698 */ 2699 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2700 soisdisconnected(so); 2701 tcp_timer_activate(tp, TT_2MSL, 2702 (tcp_fast_finwait2_recycle ? 2703 tcp_finwait2_timeout : 2704 TP_MAXIDLE(tp))); 2705 } 2706 tp->t_state = TCPS_FIN_WAIT_2; 2707 } 2708 break; 2709 2710 /* 2711 * In CLOSING STATE in addition to the processing for 2712 * the ESTABLISHED state if the ACK acknowledges our FIN 2713 * then enter the TIME-WAIT state, otherwise ignore 2714 * the segment. 2715 */ 2716 case TCPS_CLOSING: 2717 if (ourfinisacked) { 2718 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2719 tcp_twstart(tp); 2720 INP_INFO_WUNLOCK(&V_tcbinfo); 2721 m_freem(m); 2722 return; 2723 } 2724 break; 2725 2726 /* 2727 * In LAST_ACK, we may still be waiting for data to drain 2728 * and/or to be acked, as well as for the ack of our FIN. 2729 * If our FIN is now acknowledged, delete the TCB, 2730 * enter the closed state and return. 2731 */ 2732 case TCPS_LAST_ACK: 2733 if (ourfinisacked) { 2734 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2735 tp = tcp_close(tp); 2736 goto drop; 2737 } 2738 break; 2739 } 2740 } 2741 2742 step6: 2743 INP_WLOCK_ASSERT(tp->t_inpcb); 2744 2745 /* 2746 * Window update acceptance logic. We have to be careful not 2747 * to accept window updates from old segments in the presence 2748 * of reordering or duplication. 2749 * 2750 * A window update is valid when: 2751 * - the segment ACK's new data. 2752 * - the segment carries new data and its ACK is current. 2753 * - the segment matches the current SEQ and ACK but increases 2754 * the window. This is the escape from persist mode, if there 2755 * data to be sent. 2756 * 2757 * XXXAO: The presence of new SACK information would allow to 2758 * accept window updates during retransmits. We don't have an 2759 * easy way to test for that the moment. 2760 * 2761 * NB: The other side isn't allowed to shrink the window when 2762 * not sending or acking new data. This behavior is strongly 2763 * discouraged by RFC793, section 3.7, page 42 anyways. 2764 * 2765 * XXXAO: tiwin >= minmss to avoid jitter? 2766 */ 2767 if ((thflags & TH_ACK) && tiwin != tp->snd_wnd && 2768 (SEQ_GT(th->th_ack, tp->snd_wl2) || 2769 (th->th_ack == tp->snd_wl2 && 2770 (SEQ_GT(th->th_seq + tlen, tp->snd_wl1) || 2771 (th->th_seq == tp->snd_wl1 && tlen == 0 && tiwin > tp->snd_wnd))))) { 2772 #if 0 2773 char *s; 2774 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 2775 log(LOG_DEBUG, "%s; %s: window update %lu -> %lu\n", 2776 s, __func__, tp->snd_wnd, tiwin); 2777 free(s, M_TCPLOG); 2778 } 2779 #endif 2780 /* Keep track of pure window updates. */ 2781 if (th->th_seq == tp->snd_wl1 && tlen == 0 && 2782 tiwin > tp->snd_wnd) 2783 TCPSTAT_INC(tcps_rcvwinupd); 2784 /* 2785 * When the new window is larger, nudge output 2786 * as we may be able to send more data. 2787 */ 2788 if (tiwin > tp->snd_wnd) 2789 needoutput = 1; 2790 tp->snd_wnd = tiwin; 2791 if (tp->snd_wnd > tp->max_sndwnd) 2792 tp->max_sndwnd = tp->snd_wnd; 2793 } 2794 if (SEQ_GT(th->th_ack, tp->snd_wl2)) 2795 tp->snd_wl2 = th->th_ack; 2796 2797 /* 2798 * Process segments with URG. 2799 */ 2800 if ((thflags & TH_URG) && th->th_urp && 2801 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2802 /* 2803 * This is a kludge, but if we receive and accept 2804 * random urgent pointers, we'll crash in 2805 * soreceive. It's hard to imagine someone 2806 * actually wanting to send this much urgent data. 2807 */ 2808 SOCKBUF_LOCK(&so->so_rcv); 2809 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2810 th->th_urp = 0; /* XXX */ 2811 thflags &= ~TH_URG; /* XXX */ 2812 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2813 goto dodata; /* XXX */ 2814 } 2815 /* 2816 * If this segment advances the known urgent pointer, 2817 * then mark the data stream. This should not happen 2818 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2819 * a FIN has been received from the remote side. 2820 * In these states we ignore the URG. 2821 * 2822 * According to RFC961 (Assigned Protocols), 2823 * the urgent pointer points to the last octet 2824 * of urgent data. We continue, however, 2825 * to consider it to indicate the first octet 2826 * of data past the urgent section as the original 2827 * spec states (in one of two places). 2828 */ 2829 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2830 tp->rcv_up = th->th_seq + th->th_urp; 2831 so->so_oobmark = so->so_rcv.sb_cc + 2832 (tp->rcv_up - tp->rcv_nxt) - 1; 2833 if (so->so_oobmark == 0) 2834 so->so_rcv.sb_state |= SBS_RCVATMARK; 2835 sohasoutofband(so); 2836 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2837 } 2838 SOCKBUF_UNLOCK(&so->so_rcv); 2839 /* 2840 * Remove out of band data so doesn't get presented to user. 2841 * This can happen independent of advancing the URG pointer, 2842 * but if two URG's are pending at once, some out-of-band 2843 * data may creep in... ick. 2844 */ 2845 if (th->th_urp <= (u_long)tlen && 2846 !(so->so_options & SO_OOBINLINE)) { 2847 /* hdr drop is delayed */ 2848 tcp_pulloutofband(so, th, m, drop_hdrlen); 2849 } 2850 } else { 2851 /* 2852 * If no out of band data is expected, 2853 * pull receive urgent pointer along 2854 * with the receive window. 2855 */ 2856 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2857 tp->rcv_up = tp->rcv_nxt; 2858 } 2859 dodata: /* XXX */ 2860 INP_WLOCK_ASSERT(tp->t_inpcb); 2861 2862 /* 2863 * Process the segment text, merging it into the TCP sequencing queue, 2864 * and arranging for acknowledgment of receipt if necessary. 2865 * This process logically involves adjusting tp->rcv_wnd as data 2866 * is presented to the user (this happens in tcp_usrreq.c, 2867 * case PRU_RCVD). If a FIN has already been received on this 2868 * connection then we just ignore the text. 2869 */ 2870 if ((tlen || (thflags & TH_FIN)) && 2871 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2872 tcp_seq save_start = th->th_seq; 2873 m_adj(m, drop_hdrlen); /* delayed header drop */ 2874 /* 2875 * Insert segment which includes th into TCP reassembly queue 2876 * with control block tp. Set thflags to whether reassembly now 2877 * includes a segment with FIN. This handles the common case 2878 * inline (segment is the next to be received on an established 2879 * connection, and the queue is empty), avoiding linkage into 2880 * and removal from the queue and repetition of various 2881 * conversions. 2882 * Set DELACK for segments received in order, but ack 2883 * immediately when segments are out of order (so 2884 * fast retransmit can work). 2885 */ 2886 if (th->th_seq == tp->rcv_nxt && 2887 LIST_EMPTY(&tp->t_segq) && 2888 TCPS_HAVEESTABLISHED(tp->t_state)) { 2889 if (DELAY_ACK(tp)) 2890 tp->t_flags |= TF_DELACK; 2891 else 2892 tp->t_flags |= TF_ACKNOW; 2893 tp->rcv_nxt += tlen; 2894 thflags = th->th_flags & TH_FIN; 2895 TCPSTAT_INC(tcps_rcvpack); 2896 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2897 ND6_HINT(tp); 2898 SOCKBUF_LOCK(&so->so_rcv); 2899 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2900 m_freem(m); 2901 else 2902 sbappendstream_locked(&so->so_rcv, m); 2903 /* NB: sorwakeup_locked() does an implicit unlock. */ 2904 sorwakeup_locked(so); 2905 } else { 2906 /* 2907 * XXX: Due to the header drop above "th" is 2908 * theoretically invalid by now. Fortunately 2909 * m_adj() doesn't actually frees any mbufs 2910 * when trimming from the head. 2911 */ 2912 thflags = tcp_reass(tp, th, &tlen, m); 2913 tp->t_flags |= TF_ACKNOW; 2914 } 2915 if (SEQ_GT(th->th_seq, tp->snd_wl1)) 2916 tp->snd_wl1 = th->th_seq + tlen; 2917 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2918 tcp_update_sack_list(tp, save_start, save_start + tlen); 2919 #if 0 2920 /* 2921 * Note the amount of data that peer has sent into 2922 * our window, in order to estimate the sender's 2923 * buffer size. 2924 * XXX: Unused. 2925 */ 2926 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2927 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2928 else 2929 len = so->so_rcv.sb_hiwat; 2930 #endif 2931 } else { 2932 m_freem(m); 2933 thflags &= ~TH_FIN; 2934 } 2935 2936 /* 2937 * If FIN is received ACK the FIN and let the user know 2938 * that the connection is closing. 2939 */ 2940 if (thflags & TH_FIN) { 2941 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2942 socantrcvmore(so); 2943 /* 2944 * If connection is half-synchronized 2945 * (ie NEEDSYN flag on) then delay ACK, 2946 * so it may be piggybacked when SYN is sent. 2947 * Otherwise, since we received a FIN then no 2948 * more input can be expected, send ACK now. 2949 */ 2950 if (tp->t_flags & TF_NEEDSYN) 2951 tp->t_flags |= TF_DELACK; 2952 else 2953 tp->t_flags |= TF_ACKNOW; 2954 tp->rcv_nxt++; 2955 } 2956 switch (tp->t_state) { 2957 2958 /* 2959 * In SYN_RECEIVED and ESTABLISHED STATES 2960 * enter the CLOSE_WAIT state. 2961 */ 2962 case TCPS_SYN_RECEIVED: 2963 tp->t_starttime = ticks; 2964 /* FALLTHROUGH */ 2965 case TCPS_ESTABLISHED: 2966 tp->t_state = TCPS_CLOSE_WAIT; 2967 break; 2968 2969 /* 2970 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2971 * enter the CLOSING state. 2972 */ 2973 case TCPS_FIN_WAIT_1: 2974 tp->t_state = TCPS_CLOSING; 2975 break; 2976 2977 /* 2978 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2979 * starting the time-wait timer, turning off the other 2980 * standard timers. 2981 */ 2982 case TCPS_FIN_WAIT_2: 2983 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2984 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2985 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2986 ti_locked)); 2987 2988 tcp_twstart(tp); 2989 INP_INFO_WUNLOCK(&V_tcbinfo); 2990 return; 2991 } 2992 } 2993 if (ti_locked == TI_WLOCKED) 2994 INP_INFO_WUNLOCK(&V_tcbinfo); 2995 ti_locked = TI_UNLOCKED; 2996 2997 #ifdef TCPDEBUG 2998 if (so->so_options & SO_DEBUG) 2999 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3000 &tcp_savetcp, 0); 3001 #endif 3002 3003 /* 3004 * Return any desired output. 3005 */ 3006 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3007 (void) tcp_output(tp); 3008 3009 check_delack: 3010 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3011 __func__, ti_locked)); 3012 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3013 INP_WLOCK_ASSERT(tp->t_inpcb); 3014 3015 if (tp->t_flags & TF_DELACK) { 3016 tp->t_flags &= ~TF_DELACK; 3017 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3018 } 3019 INP_WUNLOCK(tp->t_inpcb); 3020 return; 3021 3022 dropafterack: 3023 /* 3024 * Generate an ACK dropping incoming segment if it occupies 3025 * sequence space, where the ACK reflects our state. 3026 * 3027 * We can now skip the test for the RST flag since all 3028 * paths to this code happen after packets containing 3029 * RST have been dropped. 3030 * 3031 * In the SYN-RECEIVED state, don't send an ACK unless the 3032 * segment we received passes the SYN-RECEIVED ACK test. 3033 * If it fails send a RST. This breaks the loop in the 3034 * "LAND" DoS attack, and also prevents an ACK storm 3035 * between two listening ports that have been sent forged 3036 * SYN segments, each with the source address of the other. 3037 */ 3038 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3039 (SEQ_GT(tp->snd_una, th->th_ack) || 3040 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3041 rstreason = BANDLIM_RST_OPENPORT; 3042 goto dropwithreset; 3043 } 3044 #ifdef TCPDEBUG 3045 if (so->so_options & SO_DEBUG) 3046 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3047 &tcp_savetcp, 0); 3048 #endif 3049 if (ti_locked == TI_WLOCKED) 3050 INP_INFO_WUNLOCK(&V_tcbinfo); 3051 ti_locked = TI_UNLOCKED; 3052 3053 tp->t_flags |= TF_ACKNOW; 3054 (void) tcp_output(tp); 3055 INP_WUNLOCK(tp->t_inpcb); 3056 m_freem(m); 3057 return; 3058 3059 dropwithreset: 3060 if (ti_locked == TI_WLOCKED) 3061 INP_INFO_WUNLOCK(&V_tcbinfo); 3062 ti_locked = TI_UNLOCKED; 3063 3064 if (tp != NULL) { 3065 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3066 INP_WUNLOCK(tp->t_inpcb); 3067 } else 3068 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3069 return; 3070 3071 drop: 3072 if (ti_locked == TI_WLOCKED) { 3073 INP_INFO_WUNLOCK(&V_tcbinfo); 3074 ti_locked = TI_UNLOCKED; 3075 } 3076 #ifdef INVARIANTS 3077 else 3078 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3079 #endif 3080 3081 /* 3082 * Drop space held by incoming segment and return. 3083 */ 3084 #ifdef TCPDEBUG 3085 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3086 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3087 &tcp_savetcp, 0); 3088 #endif 3089 if (tp != NULL) 3090 INP_WUNLOCK(tp->t_inpcb); 3091 m_freem(m); 3092 } 3093 3094 /* 3095 * Issue RST and make ACK acceptable to originator of segment. 3096 * The mbuf must still include the original packet header. 3097 * tp may be NULL. 3098 */ 3099 static void 3100 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3101 int tlen, int rstreason) 3102 { 3103 #ifdef INET 3104 struct ip *ip; 3105 #endif 3106 #ifdef INET6 3107 struct ip6_hdr *ip6; 3108 #endif 3109 3110 if (tp != NULL) { 3111 INP_WLOCK_ASSERT(tp->t_inpcb); 3112 } 3113 3114 /* Don't bother if destination was broadcast/multicast. */ 3115 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3116 goto drop; 3117 #ifdef INET6 3118 if (mtod(m, struct ip *)->ip_v == 6) { 3119 ip6 = mtod(m, struct ip6_hdr *); 3120 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3121 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3122 goto drop; 3123 /* IPv6 anycast check is done at tcp6_input() */ 3124 } 3125 #endif 3126 #if defined(INET) && defined(INET6) 3127 else 3128 #endif 3129 #ifdef INET 3130 { 3131 ip = mtod(m, struct ip *); 3132 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3133 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3134 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3135 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3136 goto drop; 3137 } 3138 #endif 3139 3140 /* Perform bandwidth limiting. */ 3141 if (badport_bandlim(rstreason) < 0) 3142 goto drop; 3143 3144 /* tcp_respond consumes the mbuf chain. */ 3145 if (th->th_flags & TH_ACK) { 3146 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3147 th->th_ack, TH_RST); 3148 } else { 3149 if (th->th_flags & TH_SYN) 3150 tlen++; 3151 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3152 (tcp_seq)0, TH_RST|TH_ACK); 3153 } 3154 return; 3155 drop: 3156 m_freem(m); 3157 } 3158 3159 /* 3160 * Parse TCP options and place in tcpopt. 3161 */ 3162 static void 3163 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3164 { 3165 int opt, optlen; 3166 3167 to->to_flags = 0; 3168 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3169 opt = cp[0]; 3170 if (opt == TCPOPT_EOL) 3171 break; 3172 if (opt == TCPOPT_NOP) 3173 optlen = 1; 3174 else { 3175 if (cnt < 2) 3176 break; 3177 optlen = cp[1]; 3178 if (optlen < 2 || optlen > cnt) 3179 break; 3180 } 3181 switch (opt) { 3182 case TCPOPT_MAXSEG: 3183 if (optlen != TCPOLEN_MAXSEG) 3184 continue; 3185 if (!(flags & TO_SYN)) 3186 continue; 3187 to->to_flags |= TOF_MSS; 3188 bcopy((char *)cp + 2, 3189 (char *)&to->to_mss, sizeof(to->to_mss)); 3190 to->to_mss = ntohs(to->to_mss); 3191 break; 3192 case TCPOPT_WINDOW: 3193 if (optlen != TCPOLEN_WINDOW) 3194 continue; 3195 if (!(flags & TO_SYN)) 3196 continue; 3197 to->to_flags |= TOF_SCALE; 3198 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3199 break; 3200 case TCPOPT_TIMESTAMP: 3201 if (optlen != TCPOLEN_TIMESTAMP) 3202 continue; 3203 to->to_flags |= TOF_TS; 3204 bcopy((char *)cp + 2, 3205 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3206 to->to_tsval = ntohl(to->to_tsval); 3207 bcopy((char *)cp + 6, 3208 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3209 to->to_tsecr = ntohl(to->to_tsecr); 3210 break; 3211 #ifdef TCP_SIGNATURE 3212 /* 3213 * XXX In order to reply to a host which has set the 3214 * TCP_SIGNATURE option in its initial SYN, we have to 3215 * record the fact that the option was observed here 3216 * for the syncache code to perform the correct response. 3217 */ 3218 case TCPOPT_SIGNATURE: 3219 if (optlen != TCPOLEN_SIGNATURE) 3220 continue; 3221 to->to_flags |= TOF_SIGNATURE; 3222 to->to_signature = cp + 2; 3223 break; 3224 #endif 3225 case TCPOPT_SACK_PERMITTED: 3226 if (optlen != TCPOLEN_SACK_PERMITTED) 3227 continue; 3228 if (!(flags & TO_SYN)) 3229 continue; 3230 if (!V_tcp_do_sack) 3231 continue; 3232 to->to_flags |= TOF_SACKPERM; 3233 break; 3234 case TCPOPT_SACK: 3235 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3236 continue; 3237 if (flags & TO_SYN) 3238 continue; 3239 to->to_flags |= TOF_SACK; 3240 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3241 to->to_sacks = cp + 2; 3242 TCPSTAT_INC(tcps_sack_rcv_blocks); 3243 break; 3244 default: 3245 continue; 3246 } 3247 } 3248 } 3249 3250 /* 3251 * Pull out of band byte out of a segment so 3252 * it doesn't appear in the user's data queue. 3253 * It is still reflected in the segment length for 3254 * sequencing purposes. 3255 */ 3256 static void 3257 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3258 int off) 3259 { 3260 int cnt = off + th->th_urp - 1; 3261 3262 while (cnt >= 0) { 3263 if (m->m_len > cnt) { 3264 char *cp = mtod(m, caddr_t) + cnt; 3265 struct tcpcb *tp = sototcpcb(so); 3266 3267 INP_WLOCK_ASSERT(tp->t_inpcb); 3268 3269 tp->t_iobc = *cp; 3270 tp->t_oobflags |= TCPOOB_HAVEDATA; 3271 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3272 m->m_len--; 3273 if (m->m_flags & M_PKTHDR) 3274 m->m_pkthdr.len--; 3275 return; 3276 } 3277 cnt -= m->m_len; 3278 m = m->m_next; 3279 if (m == NULL) 3280 break; 3281 } 3282 panic("tcp_pulloutofband"); 3283 } 3284 3285 /* 3286 * Collect new round-trip time estimate 3287 * and update averages and current timeout. 3288 */ 3289 static void 3290 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3291 { 3292 int delta; 3293 3294 INP_WLOCK_ASSERT(tp->t_inpcb); 3295 3296 TCPSTAT_INC(tcps_rttupdated); 3297 tp->t_rttupdated++; 3298 if (tp->t_srtt != 0) { 3299 /* 3300 * srtt is stored as fixed point with 5 bits after the 3301 * binary point (i.e., scaled by 8). The following magic 3302 * is equivalent to the smoothing algorithm in rfc793 with 3303 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3304 * point). Adjust rtt to origin 0. 3305 */ 3306 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3307 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3308 3309 if ((tp->t_srtt += delta) <= 0) 3310 tp->t_srtt = 1; 3311 3312 /* 3313 * We accumulate a smoothed rtt variance (actually, a 3314 * smoothed mean difference), then set the retransmit 3315 * timer to smoothed rtt + 4 times the smoothed variance. 3316 * rttvar is stored as fixed point with 4 bits after the 3317 * binary point (scaled by 16). The following is 3318 * equivalent to rfc793 smoothing with an alpha of .75 3319 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3320 * rfc793's wired-in beta. 3321 */ 3322 if (delta < 0) 3323 delta = -delta; 3324 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3325 if ((tp->t_rttvar += delta) <= 0) 3326 tp->t_rttvar = 1; 3327 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3328 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3329 } else { 3330 /* 3331 * No rtt measurement yet - use the unsmoothed rtt. 3332 * Set the variance to half the rtt (so our first 3333 * retransmit happens at 3*rtt). 3334 */ 3335 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3336 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3337 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3338 } 3339 tp->t_rtttime = 0; 3340 tp->t_rxtshift = 0; 3341 3342 /* 3343 * the retransmit should happen at rtt + 4 * rttvar. 3344 * Because of the way we do the smoothing, srtt and rttvar 3345 * will each average +1/2 tick of bias. When we compute 3346 * the retransmit timer, we want 1/2 tick of rounding and 3347 * 1 extra tick because of +-1/2 tick uncertainty in the 3348 * firing of the timer. The bias will give us exactly the 3349 * 1.5 tick we need. But, because the bias is 3350 * statistical, we have to test that we don't drop below 3351 * the minimum feasible timer (which is 2 ticks). 3352 */ 3353 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3354 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3355 3356 /* 3357 * We received an ack for a packet that wasn't retransmitted; 3358 * it is probably safe to discard any error indications we've 3359 * received recently. This isn't quite right, but close enough 3360 * for now (a route might have failed after we sent a segment, 3361 * and the return path might not be symmetrical). 3362 */ 3363 tp->t_softerror = 0; 3364 } 3365 3366 /* 3367 * Determine a reasonable value for maxseg size. 3368 * If the route is known, check route for mtu. 3369 * If none, use an mss that can be handled on the outgoing interface 3370 * without forcing IP to fragment. If no route is found, route has no mtu, 3371 * or the destination isn't local, use a default, hopefully conservative 3372 * size (usually 512 or the default IP max size, but no more than the mtu 3373 * of the interface), as we can't discover anything about intervening 3374 * gateways or networks. We also initialize the congestion/slow start 3375 * window to be a single segment if the destination isn't local. 3376 * While looking at the routing entry, we also initialize other path-dependent 3377 * parameters from pre-set or cached values in the routing entry. 3378 * 3379 * Also take into account the space needed for options that we 3380 * send regularly. Make maxseg shorter by that amount to assure 3381 * that we can send maxseg amount of data even when the options 3382 * are present. Store the upper limit of the length of options plus 3383 * data in maxopd. 3384 * 3385 * NOTE that this routine is only called when we process an incoming 3386 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3387 * settings are handled in tcp_mssopt(). 3388 */ 3389 void 3390 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3391 struct hc_metrics_lite *metricptr, int *mtuflags) 3392 { 3393 int mss = 0; 3394 u_long maxmtu = 0; 3395 struct inpcb *inp = tp->t_inpcb; 3396 struct hc_metrics_lite metrics; 3397 int origoffer; 3398 #ifdef INET6 3399 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3400 size_t min_protoh = isipv6 ? 3401 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3402 sizeof (struct tcpiphdr); 3403 #else 3404 const size_t min_protoh = sizeof(struct tcpiphdr); 3405 #endif 3406 3407 INP_WLOCK_ASSERT(tp->t_inpcb); 3408 3409 if (mtuoffer != -1) { 3410 KASSERT(offer == -1, ("%s: conflict", __func__)); 3411 offer = mtuoffer - min_protoh; 3412 } 3413 origoffer = offer; 3414 3415 /* Initialize. */ 3416 #ifdef INET6 3417 if (isipv6) { 3418 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags); 3419 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3420 } 3421 #endif 3422 #if defined(INET) && defined(INET6) 3423 else 3424 #endif 3425 #ifdef INET 3426 { 3427 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags); 3428 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3429 } 3430 #endif 3431 3432 /* 3433 * No route to sender, stay with default mss and return. 3434 */ 3435 if (maxmtu == 0) { 3436 /* 3437 * In case we return early we need to initialize metrics 3438 * to a defined state as tcp_hc_get() would do for us 3439 * if there was no cache hit. 3440 */ 3441 if (metricptr != NULL) 3442 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3443 return; 3444 } 3445 3446 /* What have we got? */ 3447 switch (offer) { 3448 case 0: 3449 /* 3450 * Offer == 0 means that there was no MSS on the SYN 3451 * segment, in this case we use tcp_mssdflt as 3452 * already assigned to t_maxopd above. 3453 */ 3454 offer = tp->t_maxopd; 3455 break; 3456 3457 case -1: 3458 /* 3459 * Offer == -1 means that we didn't receive SYN yet. 3460 */ 3461 /* FALLTHROUGH */ 3462 3463 default: 3464 /* 3465 * Prevent DoS attack with too small MSS. Round up 3466 * to at least minmss. 3467 */ 3468 offer = max(offer, V_tcp_minmss); 3469 } 3470 3471 /* 3472 * rmx information is now retrieved from tcp_hostcache. 3473 */ 3474 tcp_hc_get(&inp->inp_inc, &metrics); 3475 if (metricptr != NULL) 3476 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3477 3478 /* 3479 * If there's a discovered mtu int tcp hostcache, use it 3480 * else, use the link mtu. 3481 */ 3482 if (metrics.rmx_mtu) 3483 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3484 else { 3485 #ifdef INET6 3486 if (isipv6) { 3487 mss = maxmtu - min_protoh; 3488 if (!V_path_mtu_discovery && 3489 !in6_localaddr(&inp->in6p_faddr)) 3490 mss = min(mss, V_tcp_v6mssdflt); 3491 } 3492 #endif 3493 #if defined(INET) && defined(INET6) 3494 else 3495 #endif 3496 #ifdef INET 3497 { 3498 mss = maxmtu - min_protoh; 3499 if (!V_path_mtu_discovery && 3500 !in_localaddr(inp->inp_faddr)) 3501 mss = min(mss, V_tcp_mssdflt); 3502 } 3503 #endif 3504 /* 3505 * XXX - The above conditional (mss = maxmtu - min_protoh) 3506 * probably violates the TCP spec. 3507 * The problem is that, since we don't know the 3508 * other end's MSS, we are supposed to use a conservative 3509 * default. But, if we do that, then MTU discovery will 3510 * never actually take place, because the conservative 3511 * default is much less than the MTUs typically seen 3512 * on the Internet today. For the moment, we'll sweep 3513 * this under the carpet. 3514 * 3515 * The conservative default might not actually be a problem 3516 * if the only case this occurs is when sending an initial 3517 * SYN with options and data to a host we've never talked 3518 * to before. Then, they will reply with an MSS value which 3519 * will get recorded and the new parameters should get 3520 * recomputed. For Further Study. 3521 */ 3522 } 3523 mss = min(mss, offer); 3524 3525 /* 3526 * Sanity check: make sure that maxopd will be large 3527 * enough to allow some data on segments even if the 3528 * all the option space is used (40bytes). Otherwise 3529 * funny things may happen in tcp_output. 3530 */ 3531 mss = max(mss, 64); 3532 3533 /* 3534 * maxopd stores the maximum length of data AND options 3535 * in a segment; maxseg is the amount of data in a normal 3536 * segment. We need to store this value (maxopd) apart 3537 * from maxseg, because now every segment carries options 3538 * and thus we normally have somewhat less data in segments. 3539 */ 3540 tp->t_maxopd = mss; 3541 3542 /* 3543 * origoffer==-1 indicates that no segments were received yet. 3544 * In this case we just guess. 3545 */ 3546 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3547 (origoffer == -1 || 3548 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3549 mss -= TCPOLEN_TSTAMP_APPA; 3550 3551 tp->t_maxseg = mss; 3552 } 3553 3554 void 3555 tcp_mss(struct tcpcb *tp, int offer) 3556 { 3557 int mss; 3558 u_long bufsize; 3559 struct inpcb *inp; 3560 struct socket *so; 3561 struct hc_metrics_lite metrics; 3562 int mtuflags = 0; 3563 3564 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3565 3566 tcp_mss_update(tp, offer, -1, &metrics, &mtuflags); 3567 3568 mss = tp->t_maxseg; 3569 inp = tp->t_inpcb; 3570 3571 /* 3572 * If there's a pipesize, change the socket buffer to that size, 3573 * don't change if sb_hiwat is different than default (then it 3574 * has been changed on purpose with setsockopt). 3575 * Make the socket buffers an integral number of mss units; 3576 * if the mss is larger than the socket buffer, decrease the mss. 3577 */ 3578 so = inp->inp_socket; 3579 SOCKBUF_LOCK(&so->so_snd); 3580 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3581 bufsize = metrics.rmx_sendpipe; 3582 else 3583 bufsize = so->so_snd.sb_hiwat; 3584 if (bufsize < mss) 3585 mss = bufsize; 3586 else { 3587 bufsize = roundup(bufsize, mss); 3588 if (bufsize > sb_max) 3589 bufsize = sb_max; 3590 if (bufsize > so->so_snd.sb_hiwat) 3591 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3592 } 3593 SOCKBUF_UNLOCK(&so->so_snd); 3594 tp->t_maxseg = mss; 3595 3596 SOCKBUF_LOCK(&so->so_rcv); 3597 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3598 bufsize = metrics.rmx_recvpipe; 3599 else 3600 bufsize = so->so_rcv.sb_hiwat; 3601 if (bufsize > mss) { 3602 bufsize = roundup(bufsize, mss); 3603 if (bufsize > sb_max) 3604 bufsize = sb_max; 3605 if (bufsize > so->so_rcv.sb_hiwat) 3606 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3607 } 3608 SOCKBUF_UNLOCK(&so->so_rcv); 3609 3610 /* Check the interface for TSO capabilities. */ 3611 if (mtuflags & CSUM_TSO) 3612 tp->t_flags |= TF_TSO; 3613 } 3614 3615 /* 3616 * Determine the MSS option to send on an outgoing SYN. 3617 */ 3618 int 3619 tcp_mssopt(struct in_conninfo *inc) 3620 { 3621 int mss = 0; 3622 u_long maxmtu = 0; 3623 u_long thcmtu = 0; 3624 size_t min_protoh; 3625 3626 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3627 3628 #ifdef INET6 3629 if (inc->inc_flags & INC_ISIPV6) { 3630 mss = V_tcp_v6mssdflt; 3631 maxmtu = tcp_maxmtu6(inc, NULL); 3632 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3633 } 3634 #endif 3635 #if defined(INET) && defined(INET6) 3636 else 3637 #endif 3638 #ifdef INET 3639 { 3640 mss = V_tcp_mssdflt; 3641 maxmtu = tcp_maxmtu(inc, NULL); 3642 min_protoh = sizeof(struct tcpiphdr); 3643 } 3644 #endif 3645 #if defined(INET6) || defined(INET) 3646 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3647 #endif 3648 3649 if (maxmtu && thcmtu) 3650 mss = min(maxmtu, thcmtu) - min_protoh; 3651 else if (maxmtu || thcmtu) 3652 mss = max(maxmtu, thcmtu) - min_protoh; 3653 3654 return (mss); 3655 } 3656 3657 3658 /* 3659 * On a partial ack arrives, force the retransmission of the 3660 * next unacknowledged segment. Do not clear tp->t_dupacks. 3661 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3662 * be started again. 3663 */ 3664 static void 3665 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3666 { 3667 tcp_seq onxt = tp->snd_nxt; 3668 u_long ocwnd = tp->snd_cwnd; 3669 3670 INP_WLOCK_ASSERT(tp->t_inpcb); 3671 3672 tcp_timer_activate(tp, TT_REXMT, 0); 3673 tp->t_rtttime = 0; 3674 tp->snd_nxt = th->th_ack; 3675 /* 3676 * Set snd_cwnd to one segment beyond acknowledged offset. 3677 * (tp->snd_una has not yet been updated when this function is called.) 3678 */ 3679 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3680 tp->t_flags |= TF_ACKNOW; 3681 (void) tcp_output(tp); 3682 tp->snd_cwnd = ocwnd; 3683 if (SEQ_GT(onxt, tp->snd_nxt)) 3684 tp->snd_nxt = onxt; 3685 /* 3686 * Partial window deflation. Relies on fact that tp->snd_una 3687 * not updated yet. 3688 */ 3689 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3690 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3691 else 3692 tp->snd_cwnd = 0; 3693 tp->snd_cwnd += tp->t_maxseg; 3694 } 3695