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_kdtrace.h" 58 #include "opt_tcpdebug.h" 59 60 #include <sys/param.h> 61 #include <sys/kernel.h> 62 #include <sys/hhook.h> 63 #include <sys/malloc.h> 64 #include <sys/mbuf.h> 65 #include <sys/proc.h> /* for proc0 declaration */ 66 #include <sys/protosw.h> 67 #include <sys/sdt.h> 68 #include <sys/signalvar.h> 69 #include <sys/socket.h> 70 #include <sys/socketvar.h> 71 #include <sys/sysctl.h> 72 #include <sys/syslog.h> 73 #include <sys/systm.h> 74 75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 76 77 #include <vm/uma.h> 78 79 #include <net/if.h> 80 #include <net/route.h> 81 #include <net/vnet.h> 82 83 #define TCPSTATES /* for logging */ 84 85 #include <netinet/cc.h> 86 #include <netinet/in.h> 87 #include <netinet/in_kdtrace.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/in_var.h> 91 #include <netinet/ip.h> 92 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 93 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 94 #include <netinet/ip_var.h> 95 #include <netinet/ip_options.h> 96 #include <netinet/ip6.h> 97 #include <netinet/icmp6.h> 98 #include <netinet6/in6_pcb.h> 99 #include <netinet6/ip6_var.h> 100 #include <netinet6/nd6.h> 101 #include <netinet/tcp_fsm.h> 102 #include <netinet/tcp_seq.h> 103 #include <netinet/tcp_timer.h> 104 #include <netinet/tcp_var.h> 105 #include <netinet6/tcp6_var.h> 106 #include <netinet/tcpip.h> 107 #include <netinet/tcp_syncache.h> 108 #ifdef TCPDEBUG 109 #include <netinet/tcp_debug.h> 110 #endif /* TCPDEBUG */ 111 #ifdef TCP_OFFLOAD 112 #include <netinet/tcp_offload.h> 113 #endif 114 115 #ifdef IPSEC 116 #include <netipsec/ipsec.h> 117 #include <netipsec/ipsec6.h> 118 #endif /*IPSEC*/ 119 120 #include <machine/in_cksum.h> 121 122 #include <security/mac/mac_framework.h> 123 124 const int tcprexmtthresh = 3; 125 126 int tcp_log_in_vain = 0; 127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 128 &tcp_log_in_vain, 0, 129 "Log all incoming TCP segments to closed ports"); 130 131 VNET_DEFINE(int, blackhole) = 0; 132 #define V_blackhole VNET(blackhole) 133 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 134 &VNET_NAME(blackhole), 0, 135 "Do not send RST on segments to closed ports"); 136 137 VNET_DEFINE(int, tcp_delack_enabled) = 1; 138 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 139 &VNET_NAME(tcp_delack_enabled), 0, 140 "Delay ACK to try and piggyback it onto a data packet"); 141 142 VNET_DEFINE(int, drop_synfin) = 0; 143 #define V_drop_synfin VNET(drop_synfin) 144 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 145 &VNET_NAME(drop_synfin), 0, 146 "Drop TCP packets with SYN+FIN set"); 147 148 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 149 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 151 &VNET_NAME(tcp_do_rfc3042), 0, 152 "Enable RFC 3042 (Limited Transmit)"); 153 154 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 156 &VNET_NAME(tcp_do_rfc3390), 0, 157 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 158 159 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 160 "Experimental TCP extensions"); 161 162 VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 163 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 164 &VNET_NAME(tcp_do_initcwnd10), 0, 165 "Enable draft-ietf-tcpm-initcwnd-05 (Increasing initial CWND to 10)"); 166 167 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 168 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 169 &VNET_NAME(tcp_do_rfc3465), 0, 170 "Enable RFC 3465 (Appropriate Byte Counting)"); 171 172 VNET_DEFINE(int, tcp_abc_l_var) = 2; 173 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 174 &VNET_NAME(tcp_abc_l_var), 2, 175 "Cap the max cwnd increment during slow-start to this number of segments"); 176 177 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 178 179 VNET_DEFINE(int, tcp_do_ecn) = 0; 180 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 181 &VNET_NAME(tcp_do_ecn), 0, 182 "TCP ECN support"); 183 184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 185 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 186 &VNET_NAME(tcp_ecn_maxretries), 0, 187 "Max retries before giving up on ECN"); 188 189 VNET_DEFINE(int, tcp_insecure_rst) = 0; 190 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 191 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 192 &VNET_NAME(tcp_insecure_rst), 0, 193 "Follow the old (insecure) criteria for accepting RST packets"); 194 195 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 196 #define V_tcp_recvspace VNET(tcp_recvspace) 197 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 198 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 199 200 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 201 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 202 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 203 &VNET_NAME(tcp_do_autorcvbuf), 0, 204 "Enable automatic receive buffer sizing"); 205 206 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 207 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 208 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 209 &VNET_NAME(tcp_autorcvbuf_inc), 0, 210 "Incrementor step size of automatic receive buffer"); 211 212 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 213 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 214 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 215 &VNET_NAME(tcp_autorcvbuf_max), 0, 216 "Max size of automatic receive buffer"); 217 218 VNET_DEFINE(struct inpcbhead, tcb); 219 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 220 VNET_DEFINE(struct inpcbinfo, tcbinfo); 221 222 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 223 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 224 struct socket *, struct tcpcb *, int, int, uint8_t, 225 int); 226 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 227 struct tcpcb *, int, int); 228 static void tcp_pulloutofband(struct socket *, 229 struct tcphdr *, struct mbuf *, int); 230 static void tcp_xmit_timer(struct tcpcb *, int); 231 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 232 static void inline tcp_fields_to_host(struct tcphdr *); 233 #ifdef TCP_SIGNATURE 234 static void inline tcp_fields_to_net(struct tcphdr *); 235 static int inline tcp_signature_verify_input(struct mbuf *, int, int, 236 int, struct tcpopt *, struct tcphdr *, u_int); 237 #endif 238 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 239 uint16_t type); 240 static void inline cc_conn_init(struct tcpcb *tp); 241 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 242 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 243 struct tcphdr *th, struct tcpopt *to); 244 245 /* 246 * TCP statistics are stored in an "array" of counter(9)s. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 VNET_PCPUSTAT_SYSINIT(tcpstat); 250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252 253 #ifdef VIMAGE 254 VNET_PCPUSTAT_SYSUNINIT(tcpstat); 255 #endif /* VIMAGE */ 256 /* 257 * Kernel module interface for updating tcpstat. The argument is an index 258 * into tcpstat treated as an array. 259 */ 260 void 261 kmod_tcpstat_inc(int statnum) 262 { 263 264 counter_u64_add(VNET(tcpstat)[statnum], 1); 265 } 266 267 /* 268 * Wrapper for the TCP established input helper hook. 269 */ 270 static void inline 271 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 272 { 273 struct tcp_hhook_data hhook_data; 274 275 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 276 hhook_data.tp = tp; 277 hhook_data.th = th; 278 hhook_data.to = to; 279 280 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 281 tp->osd); 282 } 283 } 284 285 /* 286 * CC wrapper hook functions 287 */ 288 static void inline 289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 290 { 291 INP_WLOCK_ASSERT(tp->t_inpcb); 292 293 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 294 if (tp->snd_cwnd <= tp->snd_wnd) 295 tp->ccv->flags |= CCF_CWND_LIMITED; 296 else 297 tp->ccv->flags &= ~CCF_CWND_LIMITED; 298 299 if (type == CC_ACK) { 300 if (tp->snd_cwnd > tp->snd_ssthresh) { 301 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 302 V_tcp_abc_l_var * tp->t_maxseg); 303 if (tp->t_bytes_acked >= tp->snd_cwnd) { 304 tp->t_bytes_acked -= tp->snd_cwnd; 305 tp->ccv->flags |= CCF_ABC_SENTAWND; 306 } 307 } else { 308 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 309 tp->t_bytes_acked = 0; 310 } 311 } 312 313 if (CC_ALGO(tp)->ack_received != NULL) { 314 /* XXXLAS: Find a way to live without this */ 315 tp->ccv->curack = th->th_ack; 316 CC_ALGO(tp)->ack_received(tp->ccv, type); 317 } 318 } 319 320 static void inline 321 cc_conn_init(struct tcpcb *tp) 322 { 323 struct hc_metrics_lite metrics; 324 struct inpcb *inp = tp->t_inpcb; 325 int rtt; 326 327 INP_WLOCK_ASSERT(tp->t_inpcb); 328 329 tcp_hc_get(&inp->inp_inc, &metrics); 330 331 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 332 tp->t_srtt = rtt; 333 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 334 TCPSTAT_INC(tcps_usedrtt); 335 if (metrics.rmx_rttvar) { 336 tp->t_rttvar = metrics.rmx_rttvar; 337 TCPSTAT_INC(tcps_usedrttvar); 338 } else { 339 /* default variation is +- 1 rtt */ 340 tp->t_rttvar = 341 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 342 } 343 TCPT_RANGESET(tp->t_rxtcur, 344 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 345 tp->t_rttmin, TCPTV_REXMTMAX); 346 } 347 if (metrics.rmx_ssthresh) { 348 /* 349 * There's some sort of gateway or interface 350 * buffer limit on the path. Use this to set 351 * the slow start threshhold, but set the 352 * threshold to no less than 2*mss. 353 */ 354 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 355 TCPSTAT_INC(tcps_usedssthresh); 356 } 357 358 /* 359 * Set the initial slow-start flight size. 360 * 361 * RFC5681 Section 3.1 specifies the default conservative values. 362 * RFC3390 specifies slightly more aggressive values. 363 * Draft-ietf-tcpm-initcwnd-05 increases it to ten segments. 364 * 365 * If a SYN or SYN/ACK was lost and retransmitted, we have to 366 * reduce the initial CWND to one segment as congestion is likely 367 * requiring us to be cautious. 368 */ 369 if (tp->snd_cwnd == 1) 370 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 371 else if (V_tcp_do_initcwnd10) 372 tp->snd_cwnd = min(10 * tp->t_maxseg, 373 max(2 * tp->t_maxseg, 14600)); 374 else if (V_tcp_do_rfc3390) 375 tp->snd_cwnd = min(4 * tp->t_maxseg, 376 max(2 * tp->t_maxseg, 4380)); 377 else { 378 /* Per RFC5681 Section 3.1 */ 379 if (tp->t_maxseg > 2190) 380 tp->snd_cwnd = 2 * tp->t_maxseg; 381 else if (tp->t_maxseg > 1095) 382 tp->snd_cwnd = 3 * tp->t_maxseg; 383 else 384 tp->snd_cwnd = 4 * tp->t_maxseg; 385 } 386 387 if (CC_ALGO(tp)->conn_init != NULL) 388 CC_ALGO(tp)->conn_init(tp->ccv); 389 } 390 391 void inline 392 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 393 { 394 INP_WLOCK_ASSERT(tp->t_inpcb); 395 396 switch(type) { 397 case CC_NDUPACK: 398 if (!IN_FASTRECOVERY(tp->t_flags)) { 399 tp->snd_recover = tp->snd_max; 400 if (tp->t_flags & TF_ECN_PERMIT) 401 tp->t_flags |= TF_ECN_SND_CWR; 402 } 403 break; 404 case CC_ECN: 405 if (!IN_CONGRECOVERY(tp->t_flags)) { 406 TCPSTAT_INC(tcps_ecn_rcwnd); 407 tp->snd_recover = tp->snd_max; 408 if (tp->t_flags & TF_ECN_PERMIT) 409 tp->t_flags |= TF_ECN_SND_CWR; 410 } 411 break; 412 case CC_RTO: 413 tp->t_dupacks = 0; 414 tp->t_bytes_acked = 0; 415 EXIT_RECOVERY(tp->t_flags); 416 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 417 tp->t_maxseg) * tp->t_maxseg; 418 tp->snd_cwnd = tp->t_maxseg; 419 break; 420 case CC_RTO_ERR: 421 TCPSTAT_INC(tcps_sndrexmitbad); 422 /* RTO was unnecessary, so reset everything. */ 423 tp->snd_cwnd = tp->snd_cwnd_prev; 424 tp->snd_ssthresh = tp->snd_ssthresh_prev; 425 tp->snd_recover = tp->snd_recover_prev; 426 if (tp->t_flags & TF_WASFRECOVERY) 427 ENTER_FASTRECOVERY(tp->t_flags); 428 if (tp->t_flags & TF_WASCRECOVERY) 429 ENTER_CONGRECOVERY(tp->t_flags); 430 tp->snd_nxt = tp->snd_max; 431 tp->t_flags &= ~TF_PREVVALID; 432 tp->t_badrxtwin = 0; 433 break; 434 } 435 436 if (CC_ALGO(tp)->cong_signal != NULL) { 437 if (th != NULL) 438 tp->ccv->curack = th->th_ack; 439 CC_ALGO(tp)->cong_signal(tp->ccv, type); 440 } 441 } 442 443 static void inline 444 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 445 { 446 INP_WLOCK_ASSERT(tp->t_inpcb); 447 448 /* XXXLAS: KASSERT that we're in recovery? */ 449 450 if (CC_ALGO(tp)->post_recovery != NULL) { 451 tp->ccv->curack = th->th_ack; 452 CC_ALGO(tp)->post_recovery(tp->ccv); 453 } 454 /* XXXLAS: EXIT_RECOVERY ? */ 455 tp->t_bytes_acked = 0; 456 } 457 458 static inline void 459 tcp_fields_to_host(struct tcphdr *th) 460 { 461 462 th->th_seq = ntohl(th->th_seq); 463 th->th_ack = ntohl(th->th_ack); 464 th->th_win = ntohs(th->th_win); 465 th->th_urp = ntohs(th->th_urp); 466 } 467 468 #ifdef TCP_SIGNATURE 469 static inline void 470 tcp_fields_to_net(struct tcphdr *th) 471 { 472 473 th->th_seq = htonl(th->th_seq); 474 th->th_ack = htonl(th->th_ack); 475 th->th_win = htons(th->th_win); 476 th->th_urp = htons(th->th_urp); 477 } 478 479 static inline int 480 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 481 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 482 { 483 int ret; 484 485 tcp_fields_to_net(th); 486 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 487 tcp_fields_to_host(th); 488 return (ret); 489 } 490 #endif 491 492 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 493 #ifdef INET6 494 #define ND6_HINT(tp) \ 495 do { \ 496 if ((tp) && (tp)->t_inpcb && \ 497 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 498 nd6_nud_hint(NULL, NULL, 0); \ 499 } while (0) 500 #else 501 #define ND6_HINT(tp) 502 #endif 503 504 /* 505 * Indicate whether this ack should be delayed. We can delay the ack if 506 * - there is no delayed ack timer in progress and 507 * - our last ack wasn't a 0-sized window. We never want to delay 508 * the ack that opens up a 0-sized window and 509 * - delayed acks are enabled or 510 * - this is a half-synchronized T/TCP connection. 511 */ 512 #define DELAY_ACK(tp) \ 513 ((!tcp_timer_active(tp, TT_DELACK) && \ 514 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 515 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 516 517 /* 518 * TCP input handling is split into multiple parts: 519 * tcp6_input is a thin wrapper around tcp_input for the extended 520 * ip6_protox[] call format in ip6_input 521 * tcp_input handles primary segment validation, inpcb lookup and 522 * SYN processing on listen sockets 523 * tcp_do_segment processes the ACK and text of the segment for 524 * establishing, established and closing connections 525 */ 526 #ifdef INET6 527 int 528 tcp6_input(struct mbuf **mp, int *offp, int proto) 529 { 530 struct mbuf *m = *mp; 531 struct in6_ifaddr *ia6; 532 533 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 534 535 /* 536 * draft-itojun-ipv6-tcp-to-anycast 537 * better place to put this in? 538 */ 539 ia6 = ip6_getdstifaddr(m); 540 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 541 struct ip6_hdr *ip6; 542 543 ifa_free(&ia6->ia_ifa); 544 ip6 = mtod(m, struct ip6_hdr *); 545 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 546 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 547 return IPPROTO_DONE; 548 } 549 if (ia6) 550 ifa_free(&ia6->ia_ifa); 551 552 tcp_input(m, *offp); 553 return IPPROTO_DONE; 554 } 555 #endif /* INET6 */ 556 557 void 558 tcp_input(struct mbuf *m, int off0) 559 { 560 struct tcphdr *th = NULL; 561 struct ip *ip = NULL; 562 struct inpcb *inp = NULL; 563 struct tcpcb *tp = NULL; 564 struct socket *so = NULL; 565 u_char *optp = NULL; 566 int optlen = 0; 567 #ifdef INET 568 int len; 569 #endif 570 int tlen = 0, off; 571 int drop_hdrlen; 572 int thflags; 573 int rstreason = 0; /* For badport_bandlim accounting purposes */ 574 #ifdef TCP_SIGNATURE 575 uint8_t sig_checked = 0; 576 #endif 577 uint8_t iptos = 0; 578 struct m_tag *fwd_tag = NULL; 579 #ifdef INET6 580 struct ip6_hdr *ip6 = NULL; 581 int isipv6; 582 #else 583 const void *ip6 = NULL; 584 #endif /* INET6 */ 585 struct tcpopt to; /* options in this segment */ 586 char *s = NULL; /* address and port logging */ 587 int ti_locked; 588 #define TI_UNLOCKED 1 589 #define TI_WLOCKED 2 590 591 #ifdef TCPDEBUG 592 /* 593 * The size of tcp_saveipgen must be the size of the max ip header, 594 * now IPv6. 595 */ 596 u_char tcp_saveipgen[IP6_HDR_LEN]; 597 struct tcphdr tcp_savetcp; 598 short ostate = 0; 599 #endif 600 601 #ifdef INET6 602 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 603 #endif 604 605 to.to_flags = 0; 606 TCPSTAT_INC(tcps_rcvtotal); 607 608 #ifdef INET6 609 if (isipv6) { 610 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 611 612 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 613 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 614 if (m == NULL) { 615 TCPSTAT_INC(tcps_rcvshort); 616 return; 617 } 618 } 619 620 ip6 = mtod(m, struct ip6_hdr *); 621 th = (struct tcphdr *)((caddr_t)ip6 + off0); 622 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 623 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 624 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 625 th->th_sum = m->m_pkthdr.csum_data; 626 else 627 th->th_sum = in6_cksum_pseudo(ip6, tlen, 628 IPPROTO_TCP, m->m_pkthdr.csum_data); 629 th->th_sum ^= 0xffff; 630 } else 631 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 632 if (th->th_sum) { 633 TCPSTAT_INC(tcps_rcvbadsum); 634 goto drop; 635 } 636 637 /* 638 * Be proactive about unspecified IPv6 address in source. 639 * As we use all-zero to indicate unbounded/unconnected pcb, 640 * unspecified IPv6 address can be used to confuse us. 641 * 642 * Note that packets with unspecified IPv6 destination is 643 * already dropped in ip6_input. 644 */ 645 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 646 /* XXX stat */ 647 goto drop; 648 } 649 } 650 #endif 651 #if defined(INET) && defined(INET6) 652 else 653 #endif 654 #ifdef INET 655 { 656 /* 657 * Get IP and TCP header together in first mbuf. 658 * Note: IP leaves IP header in first mbuf. 659 */ 660 if (off0 > sizeof (struct ip)) { 661 ip_stripoptions(m); 662 off0 = sizeof(struct ip); 663 } 664 if (m->m_len < sizeof (struct tcpiphdr)) { 665 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 666 == NULL) { 667 TCPSTAT_INC(tcps_rcvshort); 668 return; 669 } 670 } 671 ip = mtod(m, struct ip *); 672 th = (struct tcphdr *)((caddr_t)ip + off0); 673 tlen = ntohs(ip->ip_len) - off0; 674 675 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 676 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 677 th->th_sum = m->m_pkthdr.csum_data; 678 else 679 th->th_sum = in_pseudo(ip->ip_src.s_addr, 680 ip->ip_dst.s_addr, 681 htonl(m->m_pkthdr.csum_data + tlen + 682 IPPROTO_TCP)); 683 th->th_sum ^= 0xffff; 684 } else { 685 struct ipovly *ipov = (struct ipovly *)ip; 686 687 /* 688 * Checksum extended TCP header and data. 689 */ 690 len = off0 + tlen; 691 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 692 ipov->ih_len = htons(tlen); 693 th->th_sum = in_cksum(m, len); 694 /* Reset length for SDT probes. */ 695 ip->ip_len = htons(tlen + off0); 696 } 697 698 if (th->th_sum) { 699 TCPSTAT_INC(tcps_rcvbadsum); 700 goto drop; 701 } 702 /* Re-initialization for later version check */ 703 ip->ip_v = IPVERSION; 704 } 705 #endif /* INET */ 706 707 #ifdef INET6 708 if (isipv6) 709 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 710 #endif 711 #if defined(INET) && defined(INET6) 712 else 713 #endif 714 #ifdef INET 715 iptos = ip->ip_tos; 716 #endif 717 718 /* 719 * Check that TCP offset makes sense, 720 * pull out TCP options and adjust length. XXX 721 */ 722 off = th->th_off << 2; 723 if (off < sizeof (struct tcphdr) || off > tlen) { 724 TCPSTAT_INC(tcps_rcvbadoff); 725 goto drop; 726 } 727 tlen -= off; /* tlen is used instead of ti->ti_len */ 728 if (off > sizeof (struct tcphdr)) { 729 #ifdef INET6 730 if (isipv6) { 731 IP6_EXTHDR_CHECK(m, off0, off, ); 732 ip6 = mtod(m, struct ip6_hdr *); 733 th = (struct tcphdr *)((caddr_t)ip6 + off0); 734 } 735 #endif 736 #if defined(INET) && defined(INET6) 737 else 738 #endif 739 #ifdef INET 740 { 741 if (m->m_len < sizeof(struct ip) + off) { 742 if ((m = m_pullup(m, sizeof (struct ip) + off)) 743 == NULL) { 744 TCPSTAT_INC(tcps_rcvshort); 745 return; 746 } 747 ip = mtod(m, struct ip *); 748 th = (struct tcphdr *)((caddr_t)ip + off0); 749 } 750 } 751 #endif 752 optlen = off - sizeof (struct tcphdr); 753 optp = (u_char *)(th + 1); 754 } 755 thflags = th->th_flags; 756 757 /* 758 * Convert TCP protocol specific fields to host format. 759 */ 760 tcp_fields_to_host(th); 761 762 /* 763 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 764 */ 765 drop_hdrlen = off0 + off; 766 767 /* 768 * Locate pcb for segment; if we're likely to add or remove a 769 * connection then first acquire pcbinfo lock. There are two cases 770 * where we might discover later we need a write lock despite the 771 * flags: ACKs moving a connection out of the syncache, and ACKs for 772 * a connection in TIMEWAIT. 773 */ 774 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 775 INP_INFO_WLOCK(&V_tcbinfo); 776 ti_locked = TI_WLOCKED; 777 } else 778 ti_locked = TI_UNLOCKED; 779 780 /* 781 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 782 */ 783 if ( 784 #ifdef INET6 785 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 786 #ifdef INET 787 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 788 #endif 789 #endif 790 #if defined(INET) && !defined(INET6) 791 (m->m_flags & M_IP_NEXTHOP) 792 #endif 793 ) 794 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 795 796 findpcb: 797 #ifdef INVARIANTS 798 if (ti_locked == TI_WLOCKED) { 799 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 800 } else { 801 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 802 } 803 #endif 804 #ifdef INET6 805 if (isipv6 && fwd_tag != NULL) { 806 struct sockaddr_in6 *next_hop6; 807 808 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 809 /* 810 * Transparently forwarded. Pretend to be the destination. 811 * Already got one like this? 812 */ 813 inp = in6_pcblookup_mbuf(&V_tcbinfo, 814 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 815 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 816 if (!inp) { 817 /* 818 * It's new. Try to find the ambushing socket. 819 * Because we've rewritten the destination address, 820 * any hardware-generated hash is ignored. 821 */ 822 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 823 th->th_sport, &next_hop6->sin6_addr, 824 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 825 th->th_dport, INPLOOKUP_WILDCARD | 826 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 827 } 828 } else if (isipv6) { 829 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 830 th->th_sport, &ip6->ip6_dst, th->th_dport, 831 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 832 m->m_pkthdr.rcvif, m); 833 } 834 #endif /* INET6 */ 835 #if defined(INET6) && defined(INET) 836 else 837 #endif 838 #ifdef INET 839 if (fwd_tag != NULL) { 840 struct sockaddr_in *next_hop; 841 842 next_hop = (struct sockaddr_in *)(fwd_tag+1); 843 /* 844 * Transparently forwarded. Pretend to be the destination. 845 * already got one like this? 846 */ 847 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 848 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 849 m->m_pkthdr.rcvif, m); 850 if (!inp) { 851 /* 852 * It's new. Try to find the ambushing socket. 853 * Because we've rewritten the destination address, 854 * any hardware-generated hash is ignored. 855 */ 856 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 857 th->th_sport, next_hop->sin_addr, 858 next_hop->sin_port ? ntohs(next_hop->sin_port) : 859 th->th_dport, INPLOOKUP_WILDCARD | 860 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 861 } 862 } else 863 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 864 th->th_sport, ip->ip_dst, th->th_dport, 865 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 866 m->m_pkthdr.rcvif, m); 867 #endif /* INET */ 868 869 /* 870 * If the INPCB does not exist then all data in the incoming 871 * segment is discarded and an appropriate RST is sent back. 872 * XXX MRT Send RST using which routing table? 873 */ 874 if (inp == NULL) { 875 /* 876 * Log communication attempts to ports that are not 877 * in use. 878 */ 879 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 880 tcp_log_in_vain == 2) { 881 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 882 log(LOG_INFO, "%s; %s: Connection attempt " 883 "to closed port\n", s, __func__); 884 } 885 /* 886 * When blackholing do not respond with a RST but 887 * completely ignore the segment and drop it. 888 */ 889 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 890 V_blackhole == 2) 891 goto dropunlock; 892 893 rstreason = BANDLIM_RST_CLOSEDPORT; 894 goto dropwithreset; 895 } 896 INP_WLOCK_ASSERT(inp); 897 if (!(inp->inp_flags & INP_HW_FLOWID) 898 && (m->m_flags & M_FLOWID) 899 && ((inp->inp_socket == NULL) 900 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 901 inp->inp_flags |= INP_HW_FLOWID; 902 inp->inp_flags &= ~INP_SW_FLOWID; 903 inp->inp_flowid = m->m_pkthdr.flowid; 904 } 905 #ifdef IPSEC 906 #ifdef INET6 907 if (isipv6 && ipsec6_in_reject(m, inp)) { 908 IPSEC6STAT_INC(ips_in_polvio); 909 goto dropunlock; 910 } else 911 #endif /* INET6 */ 912 if (ipsec4_in_reject(m, inp) != 0) { 913 IPSECSTAT_INC(ips_in_polvio); 914 goto dropunlock; 915 } 916 #endif /* IPSEC */ 917 918 /* 919 * Check the minimum TTL for socket. 920 */ 921 if (inp->inp_ip_minttl != 0) { 922 #ifdef INET6 923 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 924 goto dropunlock; 925 else 926 #endif 927 if (inp->inp_ip_minttl > ip->ip_ttl) 928 goto dropunlock; 929 } 930 931 /* 932 * A previous connection in TIMEWAIT state is supposed to catch stray 933 * or duplicate segments arriving late. If this segment was a 934 * legitimate new connection attempt, the old INPCB gets removed and 935 * we can try again to find a listening socket. 936 * 937 * At this point, due to earlier optimism, we may hold only an inpcb 938 * lock, and not the inpcbinfo write lock. If so, we need to try to 939 * acquire it, or if that fails, acquire a reference on the inpcb, 940 * drop all locks, acquire a global write lock, and then re-acquire 941 * the inpcb lock. We may at that point discover that another thread 942 * has tried to free the inpcb, in which case we need to loop back 943 * and try to find a new inpcb to deliver to. 944 * 945 * XXXRW: It may be time to rethink timewait locking. 946 */ 947 relocked: 948 if (inp->inp_flags & INP_TIMEWAIT) { 949 if (ti_locked == TI_UNLOCKED) { 950 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 951 in_pcbref(inp); 952 INP_WUNLOCK(inp); 953 INP_INFO_WLOCK(&V_tcbinfo); 954 ti_locked = TI_WLOCKED; 955 INP_WLOCK(inp); 956 if (in_pcbrele_wlocked(inp)) { 957 inp = NULL; 958 goto findpcb; 959 } 960 } else 961 ti_locked = TI_WLOCKED; 962 } 963 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 964 965 if (thflags & TH_SYN) 966 tcp_dooptions(&to, optp, optlen, TO_SYN); 967 /* 968 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 969 */ 970 if (tcp_twcheck(inp, &to, th, m, tlen)) 971 goto findpcb; 972 INP_INFO_WUNLOCK(&V_tcbinfo); 973 return; 974 } 975 /* 976 * The TCPCB may no longer exist if the connection is winding 977 * down or it is in the CLOSED state. Either way we drop the 978 * segment and send an appropriate response. 979 */ 980 tp = intotcpcb(inp); 981 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 982 rstreason = BANDLIM_RST_CLOSEDPORT; 983 goto dropwithreset; 984 } 985 986 #ifdef TCP_OFFLOAD 987 if (tp->t_flags & TF_TOE) { 988 tcp_offload_input(tp, m); 989 m = NULL; /* consumed by the TOE driver */ 990 goto dropunlock; 991 } 992 #endif 993 994 /* 995 * We've identified a valid inpcb, but it could be that we need an 996 * inpcbinfo write lock but don't hold it. In this case, attempt to 997 * acquire using the same strategy as the TIMEWAIT case above. If we 998 * relock, we have to jump back to 'relocked' as the connection might 999 * now be in TIMEWAIT. 1000 */ 1001 #ifdef INVARIANTS 1002 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 1003 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1004 #endif 1005 if (tp->t_state != TCPS_ESTABLISHED) { 1006 if (ti_locked == TI_UNLOCKED) { 1007 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 1008 in_pcbref(inp); 1009 INP_WUNLOCK(inp); 1010 INP_INFO_WLOCK(&V_tcbinfo); 1011 ti_locked = TI_WLOCKED; 1012 INP_WLOCK(inp); 1013 if (in_pcbrele_wlocked(inp)) { 1014 inp = NULL; 1015 goto findpcb; 1016 } 1017 goto relocked; 1018 } else 1019 ti_locked = TI_WLOCKED; 1020 } 1021 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1022 } 1023 1024 #ifdef MAC 1025 INP_WLOCK_ASSERT(inp); 1026 if (mac_inpcb_check_deliver(inp, m)) 1027 goto dropunlock; 1028 #endif 1029 so = inp->inp_socket; 1030 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1031 #ifdef TCPDEBUG 1032 if (so->so_options & SO_DEBUG) { 1033 ostate = tp->t_state; 1034 #ifdef INET6 1035 if (isipv6) { 1036 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1037 } else 1038 #endif 1039 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1040 tcp_savetcp = *th; 1041 } 1042 #endif /* TCPDEBUG */ 1043 /* 1044 * When the socket is accepting connections (the INPCB is in LISTEN 1045 * state) we look into the SYN cache if this is a new connection 1046 * attempt or the completion of a previous one. Because listen 1047 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1048 * held in this case. 1049 */ 1050 if (so->so_options & SO_ACCEPTCONN) { 1051 struct in_conninfo inc; 1052 1053 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1054 "tp not listening", __func__)); 1055 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1056 1057 bzero(&inc, sizeof(inc)); 1058 #ifdef INET6 1059 if (isipv6) { 1060 inc.inc_flags |= INC_ISIPV6; 1061 inc.inc6_faddr = ip6->ip6_src; 1062 inc.inc6_laddr = ip6->ip6_dst; 1063 } else 1064 #endif 1065 { 1066 inc.inc_faddr = ip->ip_src; 1067 inc.inc_laddr = ip->ip_dst; 1068 } 1069 inc.inc_fport = th->th_sport; 1070 inc.inc_lport = th->th_dport; 1071 inc.inc_fibnum = so->so_fibnum; 1072 1073 /* 1074 * Check for an existing connection attempt in syncache if 1075 * the flag is only ACK. A successful lookup creates a new 1076 * socket appended to the listen queue in SYN_RECEIVED state. 1077 */ 1078 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1079 /* 1080 * Parse the TCP options here because 1081 * syncookies need access to the reflected 1082 * timestamp. 1083 */ 1084 tcp_dooptions(&to, optp, optlen, 0); 1085 /* 1086 * NB: syncache_expand() doesn't unlock 1087 * inp and tcpinfo locks. 1088 */ 1089 if (!syncache_expand(&inc, &to, th, &so, m)) { 1090 /* 1091 * No syncache entry or ACK was not 1092 * for our SYN/ACK. Send a RST. 1093 * NB: syncache did its own logging 1094 * of the failure cause. 1095 */ 1096 rstreason = BANDLIM_RST_OPENPORT; 1097 goto dropwithreset; 1098 } 1099 if (so == NULL) { 1100 /* 1101 * We completed the 3-way handshake 1102 * but could not allocate a socket 1103 * either due to memory shortage, 1104 * listen queue length limits or 1105 * global socket limits. Send RST 1106 * or wait and have the remote end 1107 * retransmit the ACK for another 1108 * try. 1109 */ 1110 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1111 log(LOG_DEBUG, "%s; %s: Listen socket: " 1112 "Socket allocation failed due to " 1113 "limits or memory shortage, %s\n", 1114 s, __func__, 1115 V_tcp_sc_rst_sock_fail ? 1116 "sending RST" : "try again"); 1117 if (V_tcp_sc_rst_sock_fail) { 1118 rstreason = BANDLIM_UNLIMITED; 1119 goto dropwithreset; 1120 } else 1121 goto dropunlock; 1122 } 1123 /* 1124 * Socket is created in state SYN_RECEIVED. 1125 * Unlock the listen socket, lock the newly 1126 * created socket and update the tp variable. 1127 */ 1128 INP_WUNLOCK(inp); /* listen socket */ 1129 inp = sotoinpcb(so); 1130 INP_WLOCK(inp); /* new connection */ 1131 tp = intotcpcb(inp); 1132 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1133 ("%s: ", __func__)); 1134 #ifdef TCP_SIGNATURE 1135 if (sig_checked == 0) { 1136 tcp_dooptions(&to, optp, optlen, 1137 (thflags & TH_SYN) ? TO_SYN : 0); 1138 if (!tcp_signature_verify_input(m, off0, tlen, 1139 optlen, &to, th, tp->t_flags)) { 1140 1141 /* 1142 * In SYN_SENT state if it receives an 1143 * RST, it is allowed for further 1144 * processing. 1145 */ 1146 if ((thflags & TH_RST) == 0 || 1147 (tp->t_state == TCPS_SYN_SENT) == 0) 1148 goto dropunlock; 1149 } 1150 sig_checked = 1; 1151 } 1152 #endif 1153 1154 /* 1155 * Process the segment and the data it 1156 * contains. tcp_do_segment() consumes 1157 * the mbuf chain and unlocks the inpcb. 1158 */ 1159 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1160 iptos, ti_locked); 1161 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1162 return; 1163 } 1164 /* 1165 * Segment flag validation for new connection attempts: 1166 * 1167 * Our (SYN|ACK) response was rejected. 1168 * Check with syncache and remove entry to prevent 1169 * retransmits. 1170 * 1171 * NB: syncache_chkrst does its own logging of failure 1172 * causes. 1173 */ 1174 if (thflags & TH_RST) { 1175 syncache_chkrst(&inc, th); 1176 goto dropunlock; 1177 } 1178 /* 1179 * We can't do anything without SYN. 1180 */ 1181 if ((thflags & TH_SYN) == 0) { 1182 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1183 log(LOG_DEBUG, "%s; %s: Listen socket: " 1184 "SYN is missing, segment ignored\n", 1185 s, __func__); 1186 TCPSTAT_INC(tcps_badsyn); 1187 goto dropunlock; 1188 } 1189 /* 1190 * (SYN|ACK) is bogus on a listen socket. 1191 */ 1192 if (thflags & TH_ACK) { 1193 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1194 log(LOG_DEBUG, "%s; %s: Listen socket: " 1195 "SYN|ACK invalid, segment rejected\n", 1196 s, __func__); 1197 syncache_badack(&inc); /* XXX: Not needed! */ 1198 TCPSTAT_INC(tcps_badsyn); 1199 rstreason = BANDLIM_RST_OPENPORT; 1200 goto dropwithreset; 1201 } 1202 /* 1203 * If the drop_synfin option is enabled, drop all 1204 * segments with both the SYN and FIN bits set. 1205 * This prevents e.g. nmap from identifying the 1206 * TCP/IP stack. 1207 * XXX: Poor reasoning. nmap has other methods 1208 * and is constantly refining its stack detection 1209 * strategies. 1210 * XXX: This is a violation of the TCP specification 1211 * and was used by RFC1644. 1212 */ 1213 if ((thflags & TH_FIN) && V_drop_synfin) { 1214 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1215 log(LOG_DEBUG, "%s; %s: Listen socket: " 1216 "SYN|FIN segment ignored (based on " 1217 "sysctl setting)\n", s, __func__); 1218 TCPSTAT_INC(tcps_badsyn); 1219 goto dropunlock; 1220 } 1221 /* 1222 * Segment's flags are (SYN) or (SYN|FIN). 1223 * 1224 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1225 * as they do not affect the state of the TCP FSM. 1226 * The data pointed to by TH_URG and th_urp is ignored. 1227 */ 1228 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1229 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1230 KASSERT(thflags & (TH_SYN), 1231 ("%s: Listen socket: TH_SYN not set", __func__)); 1232 #ifdef INET6 1233 /* 1234 * If deprecated address is forbidden, 1235 * we do not accept SYN to deprecated interface 1236 * address to prevent any new inbound connection from 1237 * getting established. 1238 * When we do not accept SYN, we send a TCP RST, 1239 * with deprecated source address (instead of dropping 1240 * it). We compromise it as it is much better for peer 1241 * to send a RST, and RST will be the final packet 1242 * for the exchange. 1243 * 1244 * If we do not forbid deprecated addresses, we accept 1245 * the SYN packet. RFC2462 does not suggest dropping 1246 * SYN in this case. 1247 * If we decipher RFC2462 5.5.4, it says like this: 1248 * 1. use of deprecated addr with existing 1249 * communication is okay - "SHOULD continue to be 1250 * used" 1251 * 2. use of it with new communication: 1252 * (2a) "SHOULD NOT be used if alternate address 1253 * with sufficient scope is available" 1254 * (2b) nothing mentioned otherwise. 1255 * Here we fall into (2b) case as we have no choice in 1256 * our source address selection - we must obey the peer. 1257 * 1258 * The wording in RFC2462 is confusing, and there are 1259 * multiple description text for deprecated address 1260 * handling - worse, they are not exactly the same. 1261 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1262 */ 1263 if (isipv6 && !V_ip6_use_deprecated) { 1264 struct in6_ifaddr *ia6; 1265 1266 ia6 = ip6_getdstifaddr(m); 1267 if (ia6 != NULL && 1268 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1269 ifa_free(&ia6->ia_ifa); 1270 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1271 log(LOG_DEBUG, "%s; %s: Listen socket: " 1272 "Connection attempt to deprecated " 1273 "IPv6 address rejected\n", 1274 s, __func__); 1275 rstreason = BANDLIM_RST_OPENPORT; 1276 goto dropwithreset; 1277 } 1278 if (ia6) 1279 ifa_free(&ia6->ia_ifa); 1280 } 1281 #endif /* INET6 */ 1282 /* 1283 * Basic sanity checks on incoming SYN requests: 1284 * Don't respond if the destination is a link layer 1285 * broadcast according to RFC1122 4.2.3.10, p. 104. 1286 * If it is from this socket it must be forged. 1287 * Don't respond if the source or destination is a 1288 * global or subnet broad- or multicast address. 1289 * Note that it is quite possible to receive unicast 1290 * link-layer packets with a broadcast IP address. Use 1291 * in_broadcast() to find them. 1292 */ 1293 if (m->m_flags & (M_BCAST|M_MCAST)) { 1294 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1295 log(LOG_DEBUG, "%s; %s: Listen socket: " 1296 "Connection attempt from broad- or multicast " 1297 "link layer address ignored\n", s, __func__); 1298 goto dropunlock; 1299 } 1300 #ifdef INET6 1301 if (isipv6) { 1302 if (th->th_dport == th->th_sport && 1303 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1304 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1305 log(LOG_DEBUG, "%s; %s: Listen socket: " 1306 "Connection attempt to/from self " 1307 "ignored\n", s, __func__); 1308 goto dropunlock; 1309 } 1310 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1311 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1312 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1313 log(LOG_DEBUG, "%s; %s: Listen socket: " 1314 "Connection attempt from/to multicast " 1315 "address ignored\n", s, __func__); 1316 goto dropunlock; 1317 } 1318 } 1319 #endif 1320 #if defined(INET) && defined(INET6) 1321 else 1322 #endif 1323 #ifdef INET 1324 { 1325 if (th->th_dport == th->th_sport && 1326 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1327 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1328 log(LOG_DEBUG, "%s; %s: Listen socket: " 1329 "Connection attempt from/to self " 1330 "ignored\n", s, __func__); 1331 goto dropunlock; 1332 } 1333 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1334 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1335 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1336 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1337 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1338 log(LOG_DEBUG, "%s; %s: Listen socket: " 1339 "Connection attempt from/to broad- " 1340 "or multicast address ignored\n", 1341 s, __func__); 1342 goto dropunlock; 1343 } 1344 } 1345 #endif 1346 /* 1347 * SYN appears to be valid. Create compressed TCP state 1348 * for syncache. 1349 */ 1350 #ifdef TCPDEBUG 1351 if (so->so_options & SO_DEBUG) 1352 tcp_trace(TA_INPUT, ostate, tp, 1353 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1354 #endif 1355 tcp_dooptions(&to, optp, optlen, TO_SYN); 1356 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1357 /* 1358 * Entry added to syncache and mbuf consumed. 1359 * Everything already unlocked by syncache_add(). 1360 */ 1361 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1362 return; 1363 } else if (tp->t_state == TCPS_LISTEN) { 1364 /* 1365 * When a listen socket is torn down the SO_ACCEPTCONN 1366 * flag is removed first while connections are drained 1367 * from the accept queue in a unlock/lock cycle of the 1368 * ACCEPT_LOCK, opening a race condition allowing a SYN 1369 * attempt go through unhandled. 1370 */ 1371 goto dropunlock; 1372 } 1373 1374 #ifdef TCP_SIGNATURE 1375 if (sig_checked == 0) { 1376 tcp_dooptions(&to, optp, optlen, 1377 (thflags & TH_SYN) ? TO_SYN : 0); 1378 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1379 th, tp->t_flags)) { 1380 1381 /* 1382 * In SYN_SENT state if it receives an RST, it is 1383 * allowed for further processing. 1384 */ 1385 if ((thflags & TH_RST) == 0 || 1386 (tp->t_state == TCPS_SYN_SENT) == 0) 1387 goto dropunlock; 1388 } 1389 sig_checked = 1; 1390 } 1391 #endif 1392 1393 TCP_PROBE5(receive, NULL, tp, m->m_data, tp, th); 1394 1395 /* 1396 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1397 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1398 * the inpcb, and unlocks pcbinfo. 1399 */ 1400 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1401 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1402 return; 1403 1404 dropwithreset: 1405 TCP_PROBE5(receive, NULL, tp, m->m_data, tp, th); 1406 1407 if (ti_locked == TI_WLOCKED) { 1408 INP_INFO_WUNLOCK(&V_tcbinfo); 1409 ti_locked = TI_UNLOCKED; 1410 } 1411 #ifdef INVARIANTS 1412 else { 1413 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1414 "ti_locked: %d", __func__, ti_locked)); 1415 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1416 } 1417 #endif 1418 1419 if (inp != NULL) { 1420 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1421 INP_WUNLOCK(inp); 1422 } else 1423 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1424 m = NULL; /* mbuf chain got consumed. */ 1425 goto drop; 1426 1427 dropunlock: 1428 if (m != NULL) 1429 TCP_PROBE5(receive, NULL, tp, m->m_data, tp, th); 1430 1431 if (ti_locked == TI_WLOCKED) { 1432 INP_INFO_WUNLOCK(&V_tcbinfo); 1433 ti_locked = TI_UNLOCKED; 1434 } 1435 #ifdef INVARIANTS 1436 else { 1437 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1438 "ti_locked: %d", __func__, ti_locked)); 1439 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1440 } 1441 #endif 1442 1443 if (inp != NULL) 1444 INP_WUNLOCK(inp); 1445 1446 drop: 1447 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1448 if (s != NULL) 1449 free(s, M_TCPLOG); 1450 if (m != NULL) 1451 m_freem(m); 1452 } 1453 1454 static void 1455 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1456 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1457 int ti_locked) 1458 { 1459 int thflags, acked, ourfinisacked, needoutput = 0; 1460 int rstreason, todrop, win; 1461 u_long tiwin; 1462 char *s; 1463 struct in_conninfo *inc; 1464 struct tcpopt to; 1465 1466 #ifdef TCPDEBUG 1467 /* 1468 * The size of tcp_saveipgen must be the size of the max ip header, 1469 * now IPv6. 1470 */ 1471 u_char tcp_saveipgen[IP6_HDR_LEN]; 1472 struct tcphdr tcp_savetcp; 1473 short ostate = 0; 1474 #endif 1475 thflags = th->th_flags; 1476 inc = &tp->t_inpcb->inp_inc; 1477 tp->sackhint.last_sack_ack = 0; 1478 1479 /* 1480 * If this is either a state-changing packet or current state isn't 1481 * established, we require a write lock on tcbinfo. Otherwise, we 1482 * allow the tcbinfo to be in either alocked or unlocked, as the 1483 * caller may have unnecessarily acquired a write lock due to a race. 1484 */ 1485 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1486 tp->t_state != TCPS_ESTABLISHED) { 1487 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1488 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1489 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1490 } else { 1491 #ifdef INVARIANTS 1492 if (ti_locked == TI_WLOCKED) 1493 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1494 else { 1495 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1496 "ti_locked: %d", __func__, ti_locked)); 1497 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1498 } 1499 #endif 1500 } 1501 INP_WLOCK_ASSERT(tp->t_inpcb); 1502 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1503 __func__)); 1504 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1505 __func__)); 1506 1507 /* 1508 * Segment received on connection. 1509 * Reset idle time and keep-alive timer. 1510 * XXX: This should be done after segment 1511 * validation to ignore broken/spoofed segs. 1512 */ 1513 tp->t_rcvtime = ticks; 1514 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1515 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1516 1517 /* 1518 * Unscale the window into a 32-bit value. 1519 * For the SYN_SENT state the scale is zero. 1520 */ 1521 tiwin = th->th_win << tp->snd_scale; 1522 1523 /* 1524 * TCP ECN processing. 1525 */ 1526 if (tp->t_flags & TF_ECN_PERMIT) { 1527 if (thflags & TH_CWR) 1528 tp->t_flags &= ~TF_ECN_SND_ECE; 1529 switch (iptos & IPTOS_ECN_MASK) { 1530 case IPTOS_ECN_CE: 1531 tp->t_flags |= TF_ECN_SND_ECE; 1532 TCPSTAT_INC(tcps_ecn_ce); 1533 break; 1534 case IPTOS_ECN_ECT0: 1535 TCPSTAT_INC(tcps_ecn_ect0); 1536 break; 1537 case IPTOS_ECN_ECT1: 1538 TCPSTAT_INC(tcps_ecn_ect1); 1539 break; 1540 } 1541 /* Congestion experienced. */ 1542 if (thflags & TH_ECE) { 1543 cc_cong_signal(tp, th, CC_ECN); 1544 } 1545 } 1546 1547 /* 1548 * Parse options on any incoming segment. 1549 */ 1550 tcp_dooptions(&to, (u_char *)(th + 1), 1551 (th->th_off << 2) - sizeof(struct tcphdr), 1552 (thflags & TH_SYN) ? TO_SYN : 0); 1553 1554 /* 1555 * If echoed timestamp is later than the current time, 1556 * fall back to non RFC1323 RTT calculation. Normalize 1557 * timestamp if syncookies were used when this connection 1558 * was established. 1559 */ 1560 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1561 to.to_tsecr -= tp->ts_offset; 1562 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1563 to.to_tsecr = 0; 1564 } 1565 /* 1566 * If timestamps were negotiated during SYN/ACK they should 1567 * appear on every segment during this session and vice versa. 1568 */ 1569 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1570 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1571 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1572 "no action\n", s, __func__); 1573 free(s, M_TCPLOG); 1574 } 1575 } 1576 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1577 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1578 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1579 "no action\n", s, __func__); 1580 free(s, M_TCPLOG); 1581 } 1582 } 1583 1584 /* 1585 * Process options only when we get SYN/ACK back. The SYN case 1586 * for incoming connections is handled in tcp_syncache. 1587 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1588 * or <SYN,ACK>) segment itself is never scaled. 1589 * XXX this is traditional behavior, may need to be cleaned up. 1590 */ 1591 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1592 if ((to.to_flags & TOF_SCALE) && 1593 (tp->t_flags & TF_REQ_SCALE)) { 1594 tp->t_flags |= TF_RCVD_SCALE; 1595 tp->snd_scale = to.to_wscale; 1596 } 1597 /* 1598 * Initial send window. It will be updated with 1599 * the next incoming segment to the scaled value. 1600 */ 1601 tp->snd_wnd = th->th_win; 1602 if (to.to_flags & TOF_TS) { 1603 tp->t_flags |= TF_RCVD_TSTMP; 1604 tp->ts_recent = to.to_tsval; 1605 tp->ts_recent_age = tcp_ts_getticks(); 1606 } 1607 if (to.to_flags & TOF_MSS) 1608 tcp_mss(tp, to.to_mss); 1609 if ((tp->t_flags & TF_SACK_PERMIT) && 1610 (to.to_flags & TOF_SACKPERM) == 0) 1611 tp->t_flags &= ~TF_SACK_PERMIT; 1612 } 1613 1614 /* 1615 * Header prediction: check for the two common cases 1616 * of a uni-directional data xfer. If the packet has 1617 * no control flags, is in-sequence, the window didn't 1618 * change and we're not retransmitting, it's a 1619 * candidate. If the length is zero and the ack moved 1620 * forward, we're the sender side of the xfer. Just 1621 * free the data acked & wake any higher level process 1622 * that was blocked waiting for space. If the length 1623 * is non-zero and the ack didn't move, we're the 1624 * receiver side. If we're getting packets in-order 1625 * (the reassembly queue is empty), add the data to 1626 * the socket buffer and note that we need a delayed ack. 1627 * Make sure that the hidden state-flags are also off. 1628 * Since we check for TCPS_ESTABLISHED first, it can only 1629 * be TH_NEEDSYN. 1630 */ 1631 if (tp->t_state == TCPS_ESTABLISHED && 1632 th->th_seq == tp->rcv_nxt && 1633 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1634 tp->snd_nxt == tp->snd_max && 1635 tiwin && tiwin == tp->snd_wnd && 1636 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1637 LIST_EMPTY(&tp->t_segq) && 1638 ((to.to_flags & TOF_TS) == 0 || 1639 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1640 1641 /* 1642 * If last ACK falls within this segment's sequence numbers, 1643 * record the timestamp. 1644 * NOTE that the test is modified according to the latest 1645 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1646 */ 1647 if ((to.to_flags & TOF_TS) != 0 && 1648 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1649 tp->ts_recent_age = tcp_ts_getticks(); 1650 tp->ts_recent = to.to_tsval; 1651 } 1652 1653 if (tlen == 0) { 1654 if (SEQ_GT(th->th_ack, tp->snd_una) && 1655 SEQ_LEQ(th->th_ack, tp->snd_max) && 1656 !IN_RECOVERY(tp->t_flags) && 1657 (to.to_flags & TOF_SACK) == 0 && 1658 TAILQ_EMPTY(&tp->snd_holes)) { 1659 /* 1660 * This is a pure ack for outstanding data. 1661 */ 1662 if (ti_locked == TI_WLOCKED) 1663 INP_INFO_WUNLOCK(&V_tcbinfo); 1664 ti_locked = TI_UNLOCKED; 1665 1666 TCPSTAT_INC(tcps_predack); 1667 1668 /* 1669 * "bad retransmit" recovery. 1670 */ 1671 if (tp->t_rxtshift == 1 && 1672 tp->t_flags & TF_PREVVALID && 1673 (int)(ticks - tp->t_badrxtwin) < 0) { 1674 cc_cong_signal(tp, th, CC_RTO_ERR); 1675 } 1676 1677 /* 1678 * Recalculate the transmit timer / rtt. 1679 * 1680 * Some boxes send broken timestamp replies 1681 * during the SYN+ACK phase, ignore 1682 * timestamps of 0 or we could calculate a 1683 * huge RTT and blow up the retransmit timer. 1684 */ 1685 if ((to.to_flags & TOF_TS) != 0 && 1686 to.to_tsecr) { 1687 u_int t; 1688 1689 t = tcp_ts_getticks() - to.to_tsecr; 1690 if (!tp->t_rttlow || tp->t_rttlow > t) 1691 tp->t_rttlow = t; 1692 tcp_xmit_timer(tp, 1693 TCP_TS_TO_TICKS(t) + 1); 1694 } else if (tp->t_rtttime && 1695 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1696 if (!tp->t_rttlow || 1697 tp->t_rttlow > ticks - tp->t_rtttime) 1698 tp->t_rttlow = ticks - tp->t_rtttime; 1699 tcp_xmit_timer(tp, 1700 ticks - tp->t_rtttime); 1701 } 1702 acked = BYTES_THIS_ACK(tp, th); 1703 1704 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1705 hhook_run_tcp_est_in(tp, th, &to); 1706 1707 TCPSTAT_INC(tcps_rcvackpack); 1708 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1709 sbdrop(&so->so_snd, acked); 1710 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1711 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1712 tp->snd_recover = th->th_ack - 1; 1713 1714 /* 1715 * Let the congestion control algorithm update 1716 * congestion control related information. This 1717 * typically means increasing the congestion 1718 * window. 1719 */ 1720 cc_ack_received(tp, th, CC_ACK); 1721 1722 tp->snd_una = th->th_ack; 1723 /* 1724 * Pull snd_wl2 up to prevent seq wrap relative 1725 * to th_ack. 1726 */ 1727 tp->snd_wl2 = th->th_ack; 1728 tp->t_dupacks = 0; 1729 m_freem(m); 1730 ND6_HINT(tp); /* Some progress has been made. */ 1731 1732 /* 1733 * If all outstanding data are acked, stop 1734 * retransmit timer, otherwise restart timer 1735 * using current (possibly backed-off) value. 1736 * If process is waiting for space, 1737 * wakeup/selwakeup/signal. If data 1738 * are ready to send, let tcp_output 1739 * decide between more output or persist. 1740 */ 1741 #ifdef TCPDEBUG 1742 if (so->so_options & SO_DEBUG) 1743 tcp_trace(TA_INPUT, ostate, tp, 1744 (void *)tcp_saveipgen, 1745 &tcp_savetcp, 0); 1746 #endif 1747 if (tp->snd_una == tp->snd_max) 1748 tcp_timer_activate(tp, TT_REXMT, 0); 1749 else if (!tcp_timer_active(tp, TT_PERSIST)) 1750 tcp_timer_activate(tp, TT_REXMT, 1751 tp->t_rxtcur); 1752 sowwakeup(so); 1753 if (so->so_snd.sb_cc) 1754 (void) tcp_output(tp); 1755 goto check_delack; 1756 } 1757 } else if (th->th_ack == tp->snd_una && 1758 tlen <= sbspace(&so->so_rcv)) { 1759 int newsize = 0; /* automatic sockbuf scaling */ 1760 1761 /* 1762 * This is a pure, in-sequence data packet with 1763 * nothing on the reassembly queue and we have enough 1764 * buffer space to take it. 1765 */ 1766 if (ti_locked == TI_WLOCKED) 1767 INP_INFO_WUNLOCK(&V_tcbinfo); 1768 ti_locked = TI_UNLOCKED; 1769 1770 /* Clean receiver SACK report if present */ 1771 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1772 tcp_clean_sackreport(tp); 1773 TCPSTAT_INC(tcps_preddat); 1774 tp->rcv_nxt += tlen; 1775 /* 1776 * Pull snd_wl1 up to prevent seq wrap relative to 1777 * th_seq. 1778 */ 1779 tp->snd_wl1 = th->th_seq; 1780 /* 1781 * Pull rcv_up up to prevent seq wrap relative to 1782 * rcv_nxt. 1783 */ 1784 tp->rcv_up = tp->rcv_nxt; 1785 TCPSTAT_INC(tcps_rcvpack); 1786 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1787 ND6_HINT(tp); /* Some progress has been made */ 1788 #ifdef TCPDEBUG 1789 if (so->so_options & SO_DEBUG) 1790 tcp_trace(TA_INPUT, ostate, tp, 1791 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1792 #endif 1793 /* 1794 * Automatic sizing of receive socket buffer. Often the send 1795 * buffer size is not optimally adjusted to the actual network 1796 * conditions at hand (delay bandwidth product). Setting the 1797 * buffer size too small limits throughput on links with high 1798 * bandwidth and high delay (eg. trans-continental/oceanic links). 1799 * 1800 * On the receive side the socket buffer memory is only rarely 1801 * used to any significant extent. This allows us to be much 1802 * more aggressive in scaling the receive socket buffer. For 1803 * the case that the buffer space is actually used to a large 1804 * extent and we run out of kernel memory we can simply drop 1805 * the new segments; TCP on the sender will just retransmit it 1806 * later. Setting the buffer size too big may only consume too 1807 * much kernel memory if the application doesn't read() from 1808 * the socket or packet loss or reordering makes use of the 1809 * reassembly queue. 1810 * 1811 * The criteria to step up the receive buffer one notch are: 1812 * 1. the number of bytes received during the time it takes 1813 * one timestamp to be reflected back to us (the RTT); 1814 * 2. received bytes per RTT is within seven eighth of the 1815 * current socket buffer size; 1816 * 3. receive buffer size has not hit maximal automatic size; 1817 * 1818 * This algorithm does one step per RTT at most and only if 1819 * we receive a bulk stream w/o packet losses or reorderings. 1820 * Shrinking the buffer during idle times is not necessary as 1821 * it doesn't consume any memory when idle. 1822 * 1823 * TODO: Only step up if the application is actually serving 1824 * the buffer to better manage the socket buffer resources. 1825 */ 1826 if (V_tcp_do_autorcvbuf && 1827 to.to_tsecr && 1828 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1829 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1830 to.to_tsecr - tp->rfbuf_ts < hz) { 1831 if (tp->rfbuf_cnt > 1832 (so->so_rcv.sb_hiwat / 8 * 7) && 1833 so->so_rcv.sb_hiwat < 1834 V_tcp_autorcvbuf_max) { 1835 newsize = 1836 min(so->so_rcv.sb_hiwat + 1837 V_tcp_autorcvbuf_inc, 1838 V_tcp_autorcvbuf_max); 1839 } 1840 /* Start over with next RTT. */ 1841 tp->rfbuf_ts = 0; 1842 tp->rfbuf_cnt = 0; 1843 } else 1844 tp->rfbuf_cnt += tlen; /* add up */ 1845 } 1846 1847 /* Add data to socket buffer. */ 1848 SOCKBUF_LOCK(&so->so_rcv); 1849 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1850 m_freem(m); 1851 } else { 1852 /* 1853 * Set new socket buffer size. 1854 * Give up when limit is reached. 1855 */ 1856 if (newsize) 1857 if (!sbreserve_locked(&so->so_rcv, 1858 newsize, so, NULL)) 1859 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1860 m_adj(m, drop_hdrlen); /* delayed header drop */ 1861 sbappendstream_locked(&so->so_rcv, m); 1862 } 1863 /* NB: sorwakeup_locked() does an implicit unlock. */ 1864 sorwakeup_locked(so); 1865 if (DELAY_ACK(tp)) { 1866 tp->t_flags |= TF_DELACK; 1867 } else { 1868 tp->t_flags |= TF_ACKNOW; 1869 tcp_output(tp); 1870 } 1871 goto check_delack; 1872 } 1873 } 1874 1875 /* 1876 * Calculate amount of space in receive window, 1877 * and then do TCP input processing. 1878 * Receive window is amount of space in rcv queue, 1879 * but not less than advertised window. 1880 */ 1881 win = sbspace(&so->so_rcv); 1882 if (win < 0) 1883 win = 0; 1884 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1885 1886 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1887 tp->rfbuf_ts = 0; 1888 tp->rfbuf_cnt = 0; 1889 1890 switch (tp->t_state) { 1891 1892 /* 1893 * If the state is SYN_RECEIVED: 1894 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1895 */ 1896 case TCPS_SYN_RECEIVED: 1897 if ((thflags & TH_ACK) && 1898 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1899 SEQ_GT(th->th_ack, tp->snd_max))) { 1900 rstreason = BANDLIM_RST_OPENPORT; 1901 goto dropwithreset; 1902 } 1903 break; 1904 1905 /* 1906 * If the state is SYN_SENT: 1907 * if seg contains an ACK, but not for our SYN, drop the input. 1908 * if seg contains a RST, then drop the connection. 1909 * if seg does not contain SYN, then drop it. 1910 * Otherwise this is an acceptable SYN segment 1911 * initialize tp->rcv_nxt and tp->irs 1912 * if seg contains ack then advance tp->snd_una 1913 * if seg contains an ECE and ECN support is enabled, the stream 1914 * is ECN capable. 1915 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1916 * arrange for segment to be acked (eventually) 1917 * continue processing rest of data/controls, beginning with URG 1918 */ 1919 case TCPS_SYN_SENT: 1920 if ((thflags & TH_ACK) && 1921 (SEQ_LEQ(th->th_ack, tp->iss) || 1922 SEQ_GT(th->th_ack, tp->snd_max))) { 1923 rstreason = BANDLIM_UNLIMITED; 1924 goto dropwithreset; 1925 } 1926 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1927 TCP_PROBE5(connect_refused, NULL, tp, m->m_data, tp, 1928 th); 1929 tp = tcp_drop(tp, ECONNREFUSED); 1930 } 1931 if (thflags & TH_RST) 1932 goto drop; 1933 if (!(thflags & TH_SYN)) 1934 goto drop; 1935 1936 tp->irs = th->th_seq; 1937 tcp_rcvseqinit(tp); 1938 if (thflags & TH_ACK) { 1939 TCPSTAT_INC(tcps_connects); 1940 soisconnected(so); 1941 #ifdef MAC 1942 mac_socketpeer_set_from_mbuf(m, so); 1943 #endif 1944 /* Do window scaling on this connection? */ 1945 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1946 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1947 tp->rcv_scale = tp->request_r_scale; 1948 } 1949 tp->rcv_adv += imin(tp->rcv_wnd, 1950 TCP_MAXWIN << tp->rcv_scale); 1951 tp->snd_una++; /* SYN is acked */ 1952 /* 1953 * If there's data, delay ACK; if there's also a FIN 1954 * ACKNOW will be turned on later. 1955 */ 1956 if (DELAY_ACK(tp) && tlen != 0) 1957 tcp_timer_activate(tp, TT_DELACK, 1958 tcp_delacktime); 1959 else 1960 tp->t_flags |= TF_ACKNOW; 1961 1962 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1963 tp->t_flags |= TF_ECN_PERMIT; 1964 TCPSTAT_INC(tcps_ecn_shs); 1965 } 1966 1967 /* 1968 * Received <SYN,ACK> in SYN_SENT[*] state. 1969 * Transitions: 1970 * SYN_SENT --> ESTABLISHED 1971 * SYN_SENT* --> FIN_WAIT_1 1972 */ 1973 tp->t_starttime = ticks; 1974 if (tp->t_flags & TF_NEEDFIN) { 1975 tcp_state_change(tp, TCPS_FIN_WAIT_1); 1976 tp->t_flags &= ~TF_NEEDFIN; 1977 thflags &= ~TH_SYN; 1978 } else { 1979 tcp_state_change(tp, TCPS_ESTABLISHED); 1980 TCP_PROBE5(connect_established, NULL, tp, 1981 m->m_data, tp, th); 1982 cc_conn_init(tp); 1983 tcp_timer_activate(tp, TT_KEEP, 1984 TP_KEEPIDLE(tp)); 1985 } 1986 } else { 1987 /* 1988 * Received initial SYN in SYN-SENT[*] state => 1989 * simultaneous open. If segment contains CC option 1990 * and there is a cached CC, apply TAO test. 1991 * If it succeeds, connection is * half-synchronized. 1992 * Otherwise, do 3-way handshake: 1993 * SYN-SENT -> SYN-RECEIVED 1994 * SYN-SENT* -> SYN-RECEIVED* 1995 * If there was no CC option, clear cached CC value. 1996 */ 1997 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1998 tcp_timer_activate(tp, TT_REXMT, 0); 1999 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2000 } 2001 2002 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 2003 "ti_locked %d", __func__, ti_locked)); 2004 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2005 INP_WLOCK_ASSERT(tp->t_inpcb); 2006 2007 /* 2008 * Advance th->th_seq to correspond to first data byte. 2009 * If data, trim to stay within window, 2010 * dropping FIN if necessary. 2011 */ 2012 th->th_seq++; 2013 if (tlen > tp->rcv_wnd) { 2014 todrop = tlen - tp->rcv_wnd; 2015 m_adj(m, -todrop); 2016 tlen = tp->rcv_wnd; 2017 thflags &= ~TH_FIN; 2018 TCPSTAT_INC(tcps_rcvpackafterwin); 2019 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2020 } 2021 tp->snd_wl1 = th->th_seq - 1; 2022 tp->rcv_up = th->th_seq; 2023 /* 2024 * Client side of transaction: already sent SYN and data. 2025 * If the remote host used T/TCP to validate the SYN, 2026 * our data will be ACK'd; if so, enter normal data segment 2027 * processing in the middle of step 5, ack processing. 2028 * Otherwise, goto step 6. 2029 */ 2030 if (thflags & TH_ACK) 2031 goto process_ACK; 2032 2033 goto step6; 2034 2035 /* 2036 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2037 * do normal processing. 2038 * 2039 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2040 */ 2041 case TCPS_LAST_ACK: 2042 case TCPS_CLOSING: 2043 break; /* continue normal processing */ 2044 } 2045 2046 /* 2047 * States other than LISTEN or SYN_SENT. 2048 * First check the RST flag and sequence number since reset segments 2049 * are exempt from the timestamp and connection count tests. This 2050 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2051 * below which allowed reset segments in half the sequence space 2052 * to fall though and be processed (which gives forged reset 2053 * segments with a random sequence number a 50 percent chance of 2054 * killing a connection). 2055 * Then check timestamp, if present. 2056 * Then check the connection count, if present. 2057 * Then check that at least some bytes of segment are within 2058 * receive window. If segment begins before rcv_nxt, 2059 * drop leading data (and SYN); if nothing left, just ack. 2060 * 2061 * 2062 * If the RST bit is set, check the sequence number to see 2063 * if this is a valid reset segment. 2064 * RFC 793 page 37: 2065 * In all states except SYN-SENT, all reset (RST) segments 2066 * are validated by checking their SEQ-fields. A reset is 2067 * valid if its sequence number is in the window. 2068 * Note: this does not take into account delayed ACKs, so 2069 * we should test against last_ack_sent instead of rcv_nxt. 2070 * The sequence number in the reset segment is normally an 2071 * echo of our outgoing acknowlegement numbers, but some hosts 2072 * send a reset with the sequence number at the rightmost edge 2073 * of our receive window, and we have to handle this case. 2074 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2075 * that brute force RST attacks are possible. To combat this, 2076 * we use a much stricter check while in the ESTABLISHED state, 2077 * only accepting RSTs where the sequence number is equal to 2078 * last_ack_sent. In all other states (the states in which a 2079 * RST is more likely), the more permissive check is used. 2080 * If we have multiple segments in flight, the initial reset 2081 * segment sequence numbers will be to the left of last_ack_sent, 2082 * but they will eventually catch up. 2083 * In any case, it never made sense to trim reset segments to 2084 * fit the receive window since RFC 1122 says: 2085 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2086 * 2087 * A TCP SHOULD allow a received RST segment to include data. 2088 * 2089 * DISCUSSION 2090 * It has been suggested that a RST segment could contain 2091 * ASCII text that encoded and explained the cause of the 2092 * RST. No standard has yet been established for such 2093 * data. 2094 * 2095 * If the reset segment passes the sequence number test examine 2096 * the state: 2097 * SYN_RECEIVED STATE: 2098 * If passive open, return to LISTEN state. 2099 * If active open, inform user that connection was refused. 2100 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2101 * Inform user that connection was reset, and close tcb. 2102 * CLOSING, LAST_ACK STATES: 2103 * Close the tcb. 2104 * TIME_WAIT STATE: 2105 * Drop the segment - see Stevens, vol. 2, p. 964 and 2106 * RFC 1337. 2107 */ 2108 if (thflags & TH_RST) { 2109 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2110 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2111 switch (tp->t_state) { 2112 2113 case TCPS_SYN_RECEIVED: 2114 so->so_error = ECONNREFUSED; 2115 goto close; 2116 2117 case TCPS_ESTABLISHED: 2118 if (V_tcp_insecure_rst == 0 && 2119 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2120 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2121 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2122 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2123 TCPSTAT_INC(tcps_badrst); 2124 goto drop; 2125 } 2126 /* FALLTHROUGH */ 2127 case TCPS_FIN_WAIT_1: 2128 case TCPS_FIN_WAIT_2: 2129 case TCPS_CLOSE_WAIT: 2130 so->so_error = ECONNRESET; 2131 close: 2132 KASSERT(ti_locked == TI_WLOCKED, 2133 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2134 ti_locked)); 2135 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2136 2137 tcp_state_change(tp, TCPS_CLOSED); 2138 TCPSTAT_INC(tcps_drops); 2139 tp = tcp_close(tp); 2140 break; 2141 2142 case TCPS_CLOSING: 2143 case TCPS_LAST_ACK: 2144 KASSERT(ti_locked == TI_WLOCKED, 2145 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2146 ti_locked)); 2147 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2148 2149 tp = tcp_close(tp); 2150 break; 2151 } 2152 } 2153 goto drop; 2154 } 2155 2156 /* 2157 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2158 * and it's less than ts_recent, drop it. 2159 */ 2160 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2161 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2162 2163 /* Check to see if ts_recent is over 24 days old. */ 2164 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2165 /* 2166 * Invalidate ts_recent. If this segment updates 2167 * ts_recent, the age will be reset later and ts_recent 2168 * will get a valid value. If it does not, setting 2169 * ts_recent to zero will at least satisfy the 2170 * requirement that zero be placed in the timestamp 2171 * echo reply when ts_recent isn't valid. The 2172 * age isn't reset until we get a valid ts_recent 2173 * because we don't want out-of-order segments to be 2174 * dropped when ts_recent is old. 2175 */ 2176 tp->ts_recent = 0; 2177 } else { 2178 TCPSTAT_INC(tcps_rcvduppack); 2179 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2180 TCPSTAT_INC(tcps_pawsdrop); 2181 if (tlen) 2182 goto dropafterack; 2183 goto drop; 2184 } 2185 } 2186 2187 /* 2188 * In the SYN-RECEIVED state, validate that the packet belongs to 2189 * this connection before trimming the data to fit the receive 2190 * window. Check the sequence number versus IRS since we know 2191 * the sequence numbers haven't wrapped. This is a partial fix 2192 * for the "LAND" DoS attack. 2193 */ 2194 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2195 rstreason = BANDLIM_RST_OPENPORT; 2196 goto dropwithreset; 2197 } 2198 2199 todrop = tp->rcv_nxt - th->th_seq; 2200 if (todrop > 0) { 2201 /* 2202 * If this is a duplicate SYN for our current connection, 2203 * advance over it and pretend and it's not a SYN. 2204 */ 2205 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2206 thflags &= ~TH_SYN; 2207 th->th_seq++; 2208 if (th->th_urp > 1) 2209 th->th_urp--; 2210 else 2211 thflags &= ~TH_URG; 2212 todrop--; 2213 } 2214 /* 2215 * Following if statement from Stevens, vol. 2, p. 960. 2216 */ 2217 if (todrop > tlen 2218 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2219 /* 2220 * Any valid FIN must be to the left of the window. 2221 * At this point the FIN must be a duplicate or out 2222 * of sequence; drop it. 2223 */ 2224 thflags &= ~TH_FIN; 2225 2226 /* 2227 * Send an ACK to resynchronize and drop any data. 2228 * But keep on processing for RST or ACK. 2229 */ 2230 tp->t_flags |= TF_ACKNOW; 2231 todrop = tlen; 2232 TCPSTAT_INC(tcps_rcvduppack); 2233 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2234 } else { 2235 TCPSTAT_INC(tcps_rcvpartduppack); 2236 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2237 } 2238 drop_hdrlen += todrop; /* drop from the top afterwards */ 2239 th->th_seq += todrop; 2240 tlen -= todrop; 2241 if (th->th_urp > todrop) 2242 th->th_urp -= todrop; 2243 else { 2244 thflags &= ~TH_URG; 2245 th->th_urp = 0; 2246 } 2247 } 2248 2249 /* 2250 * If new data are received on a connection after the 2251 * user processes are gone, then RST the other end. 2252 */ 2253 if ((so->so_state & SS_NOFDREF) && 2254 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2255 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2256 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2257 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2258 2259 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2260 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2261 "after socket was closed, " 2262 "sending RST and removing tcpcb\n", 2263 s, __func__, tcpstates[tp->t_state], tlen); 2264 free(s, M_TCPLOG); 2265 } 2266 tp = tcp_close(tp); 2267 TCPSTAT_INC(tcps_rcvafterclose); 2268 rstreason = BANDLIM_UNLIMITED; 2269 goto dropwithreset; 2270 } 2271 2272 /* 2273 * If segment ends after window, drop trailing data 2274 * (and PUSH and FIN); if nothing left, just ACK. 2275 */ 2276 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2277 if (todrop > 0) { 2278 TCPSTAT_INC(tcps_rcvpackafterwin); 2279 if (todrop >= tlen) { 2280 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2281 /* 2282 * If window is closed can only take segments at 2283 * window edge, and have to drop data and PUSH from 2284 * incoming segments. Continue processing, but 2285 * remember to ack. Otherwise, drop segment 2286 * and ack. 2287 */ 2288 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2289 tp->t_flags |= TF_ACKNOW; 2290 TCPSTAT_INC(tcps_rcvwinprobe); 2291 } else 2292 goto dropafterack; 2293 } else 2294 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2295 m_adj(m, -todrop); 2296 tlen -= todrop; 2297 thflags &= ~(TH_PUSH|TH_FIN); 2298 } 2299 2300 /* 2301 * If last ACK falls within this segment's sequence numbers, 2302 * record its timestamp. 2303 * NOTE: 2304 * 1) That the test incorporates suggestions from the latest 2305 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2306 * 2) That updating only on newer timestamps interferes with 2307 * our earlier PAWS tests, so this check should be solely 2308 * predicated on the sequence space of this segment. 2309 * 3) That we modify the segment boundary check to be 2310 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2311 * instead of RFC1323's 2312 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2313 * This modified check allows us to overcome RFC1323's 2314 * limitations as described in Stevens TCP/IP Illustrated 2315 * Vol. 2 p.869. In such cases, we can still calculate the 2316 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2317 */ 2318 if ((to.to_flags & TOF_TS) != 0 && 2319 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2320 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2321 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2322 tp->ts_recent_age = tcp_ts_getticks(); 2323 tp->ts_recent = to.to_tsval; 2324 } 2325 2326 /* 2327 * If a SYN is in the window, then this is an 2328 * error and we send an RST and drop the connection. 2329 */ 2330 if (thflags & TH_SYN) { 2331 KASSERT(ti_locked == TI_WLOCKED, 2332 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2333 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2334 2335 tp = tcp_drop(tp, ECONNRESET); 2336 rstreason = BANDLIM_UNLIMITED; 2337 goto drop; 2338 } 2339 2340 /* 2341 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2342 * flag is on (half-synchronized state), then queue data for 2343 * later processing; else drop segment and return. 2344 */ 2345 if ((thflags & TH_ACK) == 0) { 2346 if (tp->t_state == TCPS_SYN_RECEIVED || 2347 (tp->t_flags & TF_NEEDSYN)) 2348 goto step6; 2349 else if (tp->t_flags & TF_ACKNOW) 2350 goto dropafterack; 2351 else 2352 goto drop; 2353 } 2354 2355 /* 2356 * Ack processing. 2357 */ 2358 switch (tp->t_state) { 2359 2360 /* 2361 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2362 * ESTABLISHED state and continue processing. 2363 * The ACK was checked above. 2364 */ 2365 case TCPS_SYN_RECEIVED: 2366 2367 TCPSTAT_INC(tcps_connects); 2368 soisconnected(so); 2369 /* Do window scaling? */ 2370 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2371 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2372 tp->rcv_scale = tp->request_r_scale; 2373 tp->snd_wnd = tiwin; 2374 } 2375 /* 2376 * Make transitions: 2377 * SYN-RECEIVED -> ESTABLISHED 2378 * SYN-RECEIVED* -> FIN-WAIT-1 2379 */ 2380 tp->t_starttime = ticks; 2381 if (tp->t_flags & TF_NEEDFIN) { 2382 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2383 tp->t_flags &= ~TF_NEEDFIN; 2384 } else { 2385 tcp_state_change(tp, TCPS_ESTABLISHED); 2386 TCP_PROBE5(accept_established, NULL, tp, m->m_data, tp, 2387 th); 2388 cc_conn_init(tp); 2389 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2390 } 2391 /* 2392 * If segment contains data or ACK, will call tcp_reass() 2393 * later; if not, do so now to pass queued data to user. 2394 */ 2395 if (tlen == 0 && (thflags & TH_FIN) == 0) 2396 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2397 (struct mbuf *)0); 2398 tp->snd_wl1 = th->th_seq - 1; 2399 /* FALLTHROUGH */ 2400 2401 /* 2402 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2403 * ACKs. If the ack is in the range 2404 * tp->snd_una < th->th_ack <= tp->snd_max 2405 * then advance tp->snd_una to th->th_ack and drop 2406 * data from the retransmission queue. If this ACK reflects 2407 * more up to date window information we update our window information. 2408 */ 2409 case TCPS_ESTABLISHED: 2410 case TCPS_FIN_WAIT_1: 2411 case TCPS_FIN_WAIT_2: 2412 case TCPS_CLOSE_WAIT: 2413 case TCPS_CLOSING: 2414 case TCPS_LAST_ACK: 2415 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2416 TCPSTAT_INC(tcps_rcvacktoomuch); 2417 goto dropafterack; 2418 } 2419 if ((tp->t_flags & TF_SACK_PERMIT) && 2420 ((to.to_flags & TOF_SACK) || 2421 !TAILQ_EMPTY(&tp->snd_holes))) 2422 tcp_sack_doack(tp, &to, th->th_ack); 2423 2424 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2425 hhook_run_tcp_est_in(tp, th, &to); 2426 2427 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2428 if (tlen == 0 && tiwin == tp->snd_wnd) { 2429 TCPSTAT_INC(tcps_rcvdupack); 2430 /* 2431 * If we have outstanding data (other than 2432 * a window probe), this is a completely 2433 * duplicate ack (ie, window info didn't 2434 * change), the ack is the biggest we've 2435 * seen and we've seen exactly our rexmt 2436 * threshhold of them, assume a packet 2437 * has been dropped and retransmit it. 2438 * Kludge snd_nxt & the congestion 2439 * window so we send only this one 2440 * packet. 2441 * 2442 * We know we're losing at the current 2443 * window size so do congestion avoidance 2444 * (set ssthresh to half the current window 2445 * and pull our congestion window back to 2446 * the new ssthresh). 2447 * 2448 * Dup acks mean that packets have left the 2449 * network (they're now cached at the receiver) 2450 * so bump cwnd by the amount in the receiver 2451 * to keep a constant cwnd packets in the 2452 * network. 2453 * 2454 * When using TCP ECN, notify the peer that 2455 * we reduced the cwnd. 2456 */ 2457 if (!tcp_timer_active(tp, TT_REXMT) || 2458 th->th_ack != tp->snd_una) 2459 tp->t_dupacks = 0; 2460 else if (++tp->t_dupacks > tcprexmtthresh || 2461 IN_FASTRECOVERY(tp->t_flags)) { 2462 cc_ack_received(tp, th, CC_DUPACK); 2463 if ((tp->t_flags & TF_SACK_PERMIT) && 2464 IN_FASTRECOVERY(tp->t_flags)) { 2465 int awnd; 2466 2467 /* 2468 * Compute the amount of data in flight first. 2469 * We can inject new data into the pipe iff 2470 * we have less than 1/2 the original window's 2471 * worth of data in flight. 2472 */ 2473 awnd = (tp->snd_nxt - tp->snd_fack) + 2474 tp->sackhint.sack_bytes_rexmit; 2475 if (awnd < tp->snd_ssthresh) { 2476 tp->snd_cwnd += tp->t_maxseg; 2477 if (tp->snd_cwnd > tp->snd_ssthresh) 2478 tp->snd_cwnd = tp->snd_ssthresh; 2479 } 2480 } else 2481 tp->snd_cwnd += tp->t_maxseg; 2482 if ((thflags & TH_FIN) && 2483 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2484 /* 2485 * If its a fin we need to process 2486 * it to avoid a race where both 2487 * sides enter FIN-WAIT and send FIN|ACK 2488 * at the same time. 2489 */ 2490 break; 2491 } 2492 (void) tcp_output(tp); 2493 goto drop; 2494 } else if (tp->t_dupacks == tcprexmtthresh) { 2495 tcp_seq onxt = tp->snd_nxt; 2496 2497 /* 2498 * If we're doing sack, check to 2499 * see if we're already in sack 2500 * recovery. If we're not doing sack, 2501 * check to see if we're in newreno 2502 * recovery. 2503 */ 2504 if (tp->t_flags & TF_SACK_PERMIT) { 2505 if (IN_FASTRECOVERY(tp->t_flags)) { 2506 tp->t_dupacks = 0; 2507 break; 2508 } 2509 } else { 2510 if (SEQ_LEQ(th->th_ack, 2511 tp->snd_recover)) { 2512 tp->t_dupacks = 0; 2513 break; 2514 } 2515 } 2516 /* Congestion signal before ack. */ 2517 cc_cong_signal(tp, th, CC_NDUPACK); 2518 cc_ack_received(tp, th, CC_DUPACK); 2519 tcp_timer_activate(tp, TT_REXMT, 0); 2520 tp->t_rtttime = 0; 2521 if (tp->t_flags & TF_SACK_PERMIT) { 2522 TCPSTAT_INC( 2523 tcps_sack_recovery_episode); 2524 tp->sack_newdata = tp->snd_nxt; 2525 tp->snd_cwnd = tp->t_maxseg; 2526 (void) tcp_output(tp); 2527 goto drop; 2528 } 2529 tp->snd_nxt = th->th_ack; 2530 tp->snd_cwnd = tp->t_maxseg; 2531 if ((thflags & TH_FIN) && 2532 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2533 /* 2534 * If its a fin we need to process 2535 * it to avoid a race where both 2536 * sides enter FIN-WAIT and send FIN|ACK 2537 * at the same time. 2538 */ 2539 break; 2540 } 2541 (void) tcp_output(tp); 2542 KASSERT(tp->snd_limited <= 2, 2543 ("%s: tp->snd_limited too big", 2544 __func__)); 2545 tp->snd_cwnd = tp->snd_ssthresh + 2546 tp->t_maxseg * 2547 (tp->t_dupacks - tp->snd_limited); 2548 if (SEQ_GT(onxt, tp->snd_nxt)) 2549 tp->snd_nxt = onxt; 2550 goto drop; 2551 } else if (V_tcp_do_rfc3042) { 2552 cc_ack_received(tp, th, CC_DUPACK); 2553 u_long oldcwnd = tp->snd_cwnd; 2554 tcp_seq oldsndmax = tp->snd_max; 2555 u_int sent; 2556 int avail; 2557 2558 KASSERT(tp->t_dupacks == 1 || 2559 tp->t_dupacks == 2, 2560 ("%s: dupacks not 1 or 2", 2561 __func__)); 2562 if (tp->t_dupacks == 1) 2563 tp->snd_limited = 0; 2564 tp->snd_cwnd = 2565 (tp->snd_nxt - tp->snd_una) + 2566 (tp->t_dupacks - tp->snd_limited) * 2567 tp->t_maxseg; 2568 if ((thflags & TH_FIN) && 2569 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2570 /* 2571 * If its a fin we need to process 2572 * it to avoid a race where both 2573 * sides enter FIN-WAIT and send FIN|ACK 2574 * at the same time. 2575 */ 2576 break; 2577 } 2578 /* 2579 * Only call tcp_output when there 2580 * is new data available to be sent. 2581 * Otherwise we would send pure ACKs. 2582 */ 2583 SOCKBUF_LOCK(&so->so_snd); 2584 avail = so->so_snd.sb_cc - 2585 (tp->snd_nxt - tp->snd_una); 2586 SOCKBUF_UNLOCK(&so->so_snd); 2587 if (avail > 0) 2588 (void) tcp_output(tp); 2589 sent = tp->snd_max - oldsndmax; 2590 if (sent > tp->t_maxseg) { 2591 KASSERT((tp->t_dupacks == 2 && 2592 tp->snd_limited == 0) || 2593 (sent == tp->t_maxseg + 1 && 2594 tp->t_flags & TF_SENTFIN), 2595 ("%s: sent too much", 2596 __func__)); 2597 tp->snd_limited = 2; 2598 } else if (sent > 0) 2599 ++tp->snd_limited; 2600 tp->snd_cwnd = oldcwnd; 2601 goto drop; 2602 } 2603 } else 2604 tp->t_dupacks = 0; 2605 break; 2606 } 2607 2608 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2609 ("%s: th_ack <= snd_una", __func__)); 2610 2611 /* 2612 * If the congestion window was inflated to account 2613 * for the other side's cached packets, retract it. 2614 */ 2615 if (IN_FASTRECOVERY(tp->t_flags)) { 2616 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2617 if (tp->t_flags & TF_SACK_PERMIT) 2618 tcp_sack_partialack(tp, th); 2619 else 2620 tcp_newreno_partial_ack(tp, th); 2621 } else 2622 cc_post_recovery(tp, th); 2623 } 2624 tp->t_dupacks = 0; 2625 /* 2626 * If we reach this point, ACK is not a duplicate, 2627 * i.e., it ACKs something we sent. 2628 */ 2629 if (tp->t_flags & TF_NEEDSYN) { 2630 /* 2631 * T/TCP: Connection was half-synchronized, and our 2632 * SYN has been ACK'd (so connection is now fully 2633 * synchronized). Go to non-starred state, 2634 * increment snd_una for ACK of SYN, and check if 2635 * we can do window scaling. 2636 */ 2637 tp->t_flags &= ~TF_NEEDSYN; 2638 tp->snd_una++; 2639 /* Do window scaling? */ 2640 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2641 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2642 tp->rcv_scale = tp->request_r_scale; 2643 /* Send window already scaled. */ 2644 } 2645 } 2646 2647 process_ACK: 2648 INP_WLOCK_ASSERT(tp->t_inpcb); 2649 2650 acked = BYTES_THIS_ACK(tp, th); 2651 TCPSTAT_INC(tcps_rcvackpack); 2652 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2653 2654 /* 2655 * If we just performed our first retransmit, and the ACK 2656 * arrives within our recovery window, then it was a mistake 2657 * to do the retransmit in the first place. Recover our 2658 * original cwnd and ssthresh, and proceed to transmit where 2659 * we left off. 2660 */ 2661 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2662 (int)(ticks - tp->t_badrxtwin) < 0) 2663 cc_cong_signal(tp, th, CC_RTO_ERR); 2664 2665 /* 2666 * If we have a timestamp reply, update smoothed 2667 * round trip time. If no timestamp is present but 2668 * transmit timer is running and timed sequence 2669 * number was acked, update smoothed round trip time. 2670 * Since we now have an rtt measurement, cancel the 2671 * timer backoff (cf., Phil Karn's retransmit alg.). 2672 * Recompute the initial retransmit timer. 2673 * 2674 * Some boxes send broken timestamp replies 2675 * during the SYN+ACK phase, ignore 2676 * timestamps of 0 or we could calculate a 2677 * huge RTT and blow up the retransmit timer. 2678 */ 2679 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2680 u_int t; 2681 2682 t = tcp_ts_getticks() - to.to_tsecr; 2683 if (!tp->t_rttlow || tp->t_rttlow > t) 2684 tp->t_rttlow = t; 2685 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2686 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2687 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2688 tp->t_rttlow = ticks - tp->t_rtttime; 2689 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2690 } 2691 2692 /* 2693 * If all outstanding data is acked, stop retransmit 2694 * timer and remember to restart (more output or persist). 2695 * If there is more data to be acked, restart retransmit 2696 * timer, using current (possibly backed-off) value. 2697 */ 2698 if (th->th_ack == tp->snd_max) { 2699 tcp_timer_activate(tp, TT_REXMT, 0); 2700 needoutput = 1; 2701 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2702 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2703 2704 /* 2705 * If no data (only SYN) was ACK'd, 2706 * skip rest of ACK processing. 2707 */ 2708 if (acked == 0) 2709 goto step6; 2710 2711 /* 2712 * Let the congestion control algorithm update congestion 2713 * control related information. This typically means increasing 2714 * the congestion window. 2715 */ 2716 cc_ack_received(tp, th, CC_ACK); 2717 2718 SOCKBUF_LOCK(&so->so_snd); 2719 if (acked > so->so_snd.sb_cc) { 2720 tp->snd_wnd -= so->so_snd.sb_cc; 2721 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2722 ourfinisacked = 1; 2723 } else { 2724 sbdrop_locked(&so->so_snd, acked); 2725 tp->snd_wnd -= acked; 2726 ourfinisacked = 0; 2727 } 2728 /* NB: sowwakeup_locked() does an implicit unlock. */ 2729 sowwakeup_locked(so); 2730 /* Detect una wraparound. */ 2731 if (!IN_RECOVERY(tp->t_flags) && 2732 SEQ_GT(tp->snd_una, tp->snd_recover) && 2733 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2734 tp->snd_recover = th->th_ack - 1; 2735 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2736 if (IN_RECOVERY(tp->t_flags) && 2737 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2738 EXIT_RECOVERY(tp->t_flags); 2739 } 2740 tp->snd_una = th->th_ack; 2741 if (tp->t_flags & TF_SACK_PERMIT) { 2742 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2743 tp->snd_recover = tp->snd_una; 2744 } 2745 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2746 tp->snd_nxt = tp->snd_una; 2747 2748 switch (tp->t_state) { 2749 2750 /* 2751 * In FIN_WAIT_1 STATE in addition to the processing 2752 * for the ESTABLISHED state if our FIN is now acknowledged 2753 * then enter FIN_WAIT_2. 2754 */ 2755 case TCPS_FIN_WAIT_1: 2756 if (ourfinisacked) { 2757 /* 2758 * If we can't receive any more 2759 * data, then closing user can proceed. 2760 * Starting the timer is contrary to the 2761 * specification, but if we don't get a FIN 2762 * we'll hang forever. 2763 * 2764 * XXXjl: 2765 * we should release the tp also, and use a 2766 * compressed state. 2767 */ 2768 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2769 soisdisconnected(so); 2770 tcp_timer_activate(tp, TT_2MSL, 2771 (tcp_fast_finwait2_recycle ? 2772 tcp_finwait2_timeout : 2773 TP_MAXIDLE(tp))); 2774 } 2775 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2776 } 2777 break; 2778 2779 /* 2780 * In CLOSING STATE in addition to the processing for 2781 * the ESTABLISHED state if the ACK acknowledges our FIN 2782 * then enter the TIME-WAIT state, otherwise ignore 2783 * the segment. 2784 */ 2785 case TCPS_CLOSING: 2786 if (ourfinisacked) { 2787 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2788 tcp_twstart(tp); 2789 INP_INFO_WUNLOCK(&V_tcbinfo); 2790 m_freem(m); 2791 return; 2792 } 2793 break; 2794 2795 /* 2796 * In LAST_ACK, we may still be waiting for data to drain 2797 * and/or to be acked, as well as for the ack of our FIN. 2798 * If our FIN is now acknowledged, delete the TCB, 2799 * enter the closed state and return. 2800 */ 2801 case TCPS_LAST_ACK: 2802 if (ourfinisacked) { 2803 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2804 tp = tcp_close(tp); 2805 goto drop; 2806 } 2807 break; 2808 } 2809 } 2810 2811 step6: 2812 INP_WLOCK_ASSERT(tp->t_inpcb); 2813 2814 /* 2815 * Update window information. 2816 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2817 */ 2818 if ((thflags & TH_ACK) && 2819 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2820 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2821 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2822 /* keep track of pure window updates */ 2823 if (tlen == 0 && 2824 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2825 TCPSTAT_INC(tcps_rcvwinupd); 2826 tp->snd_wnd = tiwin; 2827 tp->snd_wl1 = th->th_seq; 2828 tp->snd_wl2 = th->th_ack; 2829 if (tp->snd_wnd > tp->max_sndwnd) 2830 tp->max_sndwnd = tp->snd_wnd; 2831 needoutput = 1; 2832 } 2833 2834 /* 2835 * Process segments with URG. 2836 */ 2837 if ((thflags & TH_URG) && th->th_urp && 2838 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2839 /* 2840 * This is a kludge, but if we receive and accept 2841 * random urgent pointers, we'll crash in 2842 * soreceive. It's hard to imagine someone 2843 * actually wanting to send this much urgent data. 2844 */ 2845 SOCKBUF_LOCK(&so->so_rcv); 2846 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2847 th->th_urp = 0; /* XXX */ 2848 thflags &= ~TH_URG; /* XXX */ 2849 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2850 goto dodata; /* XXX */ 2851 } 2852 /* 2853 * If this segment advances the known urgent pointer, 2854 * then mark the data stream. This should not happen 2855 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2856 * a FIN has been received from the remote side. 2857 * In these states we ignore the URG. 2858 * 2859 * According to RFC961 (Assigned Protocols), 2860 * the urgent pointer points to the last octet 2861 * of urgent data. We continue, however, 2862 * to consider it to indicate the first octet 2863 * of data past the urgent section as the original 2864 * spec states (in one of two places). 2865 */ 2866 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2867 tp->rcv_up = th->th_seq + th->th_urp; 2868 so->so_oobmark = so->so_rcv.sb_cc + 2869 (tp->rcv_up - tp->rcv_nxt) - 1; 2870 if (so->so_oobmark == 0) 2871 so->so_rcv.sb_state |= SBS_RCVATMARK; 2872 sohasoutofband(so); 2873 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2874 } 2875 SOCKBUF_UNLOCK(&so->so_rcv); 2876 /* 2877 * Remove out of band data so doesn't get presented to user. 2878 * This can happen independent of advancing the URG pointer, 2879 * but if two URG's are pending at once, some out-of-band 2880 * data may creep in... ick. 2881 */ 2882 if (th->th_urp <= (u_long)tlen && 2883 !(so->so_options & SO_OOBINLINE)) { 2884 /* hdr drop is delayed */ 2885 tcp_pulloutofband(so, th, m, drop_hdrlen); 2886 } 2887 } else { 2888 /* 2889 * If no out of band data is expected, 2890 * pull receive urgent pointer along 2891 * with the receive window. 2892 */ 2893 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2894 tp->rcv_up = tp->rcv_nxt; 2895 } 2896 dodata: /* XXX */ 2897 INP_WLOCK_ASSERT(tp->t_inpcb); 2898 2899 /* 2900 * Process the segment text, merging it into the TCP sequencing queue, 2901 * and arranging for acknowledgment of receipt if necessary. 2902 * This process logically involves adjusting tp->rcv_wnd as data 2903 * is presented to the user (this happens in tcp_usrreq.c, 2904 * case PRU_RCVD). If a FIN has already been received on this 2905 * connection then we just ignore the text. 2906 */ 2907 if ((tlen || (thflags & TH_FIN)) && 2908 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2909 tcp_seq save_start = th->th_seq; 2910 m_adj(m, drop_hdrlen); /* delayed header drop */ 2911 /* 2912 * Insert segment which includes th into TCP reassembly queue 2913 * with control block tp. Set thflags to whether reassembly now 2914 * includes a segment with FIN. This handles the common case 2915 * inline (segment is the next to be received on an established 2916 * connection, and the queue is empty), avoiding linkage into 2917 * and removal from the queue and repetition of various 2918 * conversions. 2919 * Set DELACK for segments received in order, but ack 2920 * immediately when segments are out of order (so 2921 * fast retransmit can work). 2922 */ 2923 if (th->th_seq == tp->rcv_nxt && 2924 LIST_EMPTY(&tp->t_segq) && 2925 TCPS_HAVEESTABLISHED(tp->t_state)) { 2926 if (DELAY_ACK(tp)) 2927 tp->t_flags |= TF_DELACK; 2928 else 2929 tp->t_flags |= TF_ACKNOW; 2930 tp->rcv_nxt += tlen; 2931 thflags = th->th_flags & TH_FIN; 2932 TCPSTAT_INC(tcps_rcvpack); 2933 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2934 ND6_HINT(tp); 2935 SOCKBUF_LOCK(&so->so_rcv); 2936 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2937 m_freem(m); 2938 else 2939 sbappendstream_locked(&so->so_rcv, m); 2940 /* NB: sorwakeup_locked() does an implicit unlock. */ 2941 sorwakeup_locked(so); 2942 } else { 2943 /* 2944 * XXX: Due to the header drop above "th" is 2945 * theoretically invalid by now. Fortunately 2946 * m_adj() doesn't actually frees any mbufs 2947 * when trimming from the head. 2948 */ 2949 thflags = tcp_reass(tp, th, &tlen, m); 2950 tp->t_flags |= TF_ACKNOW; 2951 } 2952 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2953 tcp_update_sack_list(tp, save_start, save_start + tlen); 2954 #if 0 2955 /* 2956 * Note the amount of data that peer has sent into 2957 * our window, in order to estimate the sender's 2958 * buffer size. 2959 * XXX: Unused. 2960 */ 2961 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2962 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2963 else 2964 len = so->so_rcv.sb_hiwat; 2965 #endif 2966 } else { 2967 m_freem(m); 2968 thflags &= ~TH_FIN; 2969 } 2970 2971 /* 2972 * If FIN is received ACK the FIN and let the user know 2973 * that the connection is closing. 2974 */ 2975 if (thflags & TH_FIN) { 2976 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2977 socantrcvmore(so); 2978 /* 2979 * If connection is half-synchronized 2980 * (ie NEEDSYN flag on) then delay ACK, 2981 * so it may be piggybacked when SYN is sent. 2982 * Otherwise, since we received a FIN then no 2983 * more input can be expected, send ACK now. 2984 */ 2985 if (tp->t_flags & TF_NEEDSYN) 2986 tp->t_flags |= TF_DELACK; 2987 else 2988 tp->t_flags |= TF_ACKNOW; 2989 tp->rcv_nxt++; 2990 } 2991 switch (tp->t_state) { 2992 2993 /* 2994 * In SYN_RECEIVED and ESTABLISHED STATES 2995 * enter the CLOSE_WAIT state. 2996 */ 2997 case TCPS_SYN_RECEIVED: 2998 tp->t_starttime = ticks; 2999 /* FALLTHROUGH */ 3000 case TCPS_ESTABLISHED: 3001 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3002 break; 3003 3004 /* 3005 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3006 * enter the CLOSING state. 3007 */ 3008 case TCPS_FIN_WAIT_1: 3009 tcp_state_change(tp, TCPS_CLOSING); 3010 break; 3011 3012 /* 3013 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3014 * starting the time-wait timer, turning off the other 3015 * standard timers. 3016 */ 3017 case TCPS_FIN_WAIT_2: 3018 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 3019 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 3020 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3021 ti_locked)); 3022 3023 tcp_twstart(tp); 3024 INP_INFO_WUNLOCK(&V_tcbinfo); 3025 return; 3026 } 3027 } 3028 if (ti_locked == TI_WLOCKED) 3029 INP_INFO_WUNLOCK(&V_tcbinfo); 3030 ti_locked = TI_UNLOCKED; 3031 3032 #ifdef TCPDEBUG 3033 if (so->so_options & SO_DEBUG) 3034 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3035 &tcp_savetcp, 0); 3036 #endif 3037 3038 /* 3039 * Return any desired output. 3040 */ 3041 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3042 (void) tcp_output(tp); 3043 3044 check_delack: 3045 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3046 __func__, ti_locked)); 3047 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3048 INP_WLOCK_ASSERT(tp->t_inpcb); 3049 3050 if (tp->t_flags & TF_DELACK) { 3051 tp->t_flags &= ~TF_DELACK; 3052 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3053 } 3054 INP_WUNLOCK(tp->t_inpcb); 3055 return; 3056 3057 dropafterack: 3058 /* 3059 * Generate an ACK dropping incoming segment if it occupies 3060 * sequence space, where the ACK reflects our state. 3061 * 3062 * We can now skip the test for the RST flag since all 3063 * paths to this code happen after packets containing 3064 * RST have been dropped. 3065 * 3066 * In the SYN-RECEIVED state, don't send an ACK unless the 3067 * segment we received passes the SYN-RECEIVED ACK test. 3068 * If it fails send a RST. This breaks the loop in the 3069 * "LAND" DoS attack, and also prevents an ACK storm 3070 * between two listening ports that have been sent forged 3071 * SYN segments, each with the source address of the other. 3072 */ 3073 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3074 (SEQ_GT(tp->snd_una, th->th_ack) || 3075 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3076 rstreason = BANDLIM_RST_OPENPORT; 3077 goto dropwithreset; 3078 } 3079 #ifdef TCPDEBUG 3080 if (so->so_options & SO_DEBUG) 3081 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3082 &tcp_savetcp, 0); 3083 #endif 3084 if (ti_locked == TI_WLOCKED) 3085 INP_INFO_WUNLOCK(&V_tcbinfo); 3086 ti_locked = TI_UNLOCKED; 3087 3088 tp->t_flags |= TF_ACKNOW; 3089 (void) tcp_output(tp); 3090 INP_WUNLOCK(tp->t_inpcb); 3091 m_freem(m); 3092 return; 3093 3094 dropwithreset: 3095 if (ti_locked == TI_WLOCKED) 3096 INP_INFO_WUNLOCK(&V_tcbinfo); 3097 ti_locked = TI_UNLOCKED; 3098 3099 if (tp != NULL) { 3100 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3101 INP_WUNLOCK(tp->t_inpcb); 3102 } else 3103 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3104 return; 3105 3106 drop: 3107 if (ti_locked == TI_WLOCKED) { 3108 INP_INFO_WUNLOCK(&V_tcbinfo); 3109 ti_locked = TI_UNLOCKED; 3110 } 3111 #ifdef INVARIANTS 3112 else 3113 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3114 #endif 3115 3116 /* 3117 * Drop space held by incoming segment and return. 3118 */ 3119 #ifdef TCPDEBUG 3120 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3121 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3122 &tcp_savetcp, 0); 3123 #endif 3124 if (tp != NULL) 3125 INP_WUNLOCK(tp->t_inpcb); 3126 m_freem(m); 3127 } 3128 3129 /* 3130 * Issue RST and make ACK acceptable to originator of segment. 3131 * The mbuf must still include the original packet header. 3132 * tp may be NULL. 3133 */ 3134 static void 3135 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3136 int tlen, int rstreason) 3137 { 3138 #ifdef INET 3139 struct ip *ip; 3140 #endif 3141 #ifdef INET6 3142 struct ip6_hdr *ip6; 3143 #endif 3144 3145 if (tp != NULL) { 3146 INP_WLOCK_ASSERT(tp->t_inpcb); 3147 } 3148 3149 /* Don't bother if destination was broadcast/multicast. */ 3150 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3151 goto drop; 3152 #ifdef INET6 3153 if (mtod(m, struct ip *)->ip_v == 6) { 3154 ip6 = mtod(m, struct ip6_hdr *); 3155 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3156 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3157 goto drop; 3158 /* IPv6 anycast check is done at tcp6_input() */ 3159 } 3160 #endif 3161 #if defined(INET) && defined(INET6) 3162 else 3163 #endif 3164 #ifdef INET 3165 { 3166 ip = mtod(m, struct ip *); 3167 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3168 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3169 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3170 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3171 goto drop; 3172 } 3173 #endif 3174 3175 /* Perform bandwidth limiting. */ 3176 if (badport_bandlim(rstreason) < 0) 3177 goto drop; 3178 3179 /* tcp_respond consumes the mbuf chain. */ 3180 if (th->th_flags & TH_ACK) { 3181 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3182 th->th_ack, TH_RST); 3183 } else { 3184 if (th->th_flags & TH_SYN) 3185 tlen++; 3186 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3187 (tcp_seq)0, TH_RST|TH_ACK); 3188 } 3189 return; 3190 drop: 3191 m_freem(m); 3192 } 3193 3194 /* 3195 * Parse TCP options and place in tcpopt. 3196 */ 3197 static void 3198 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3199 { 3200 int opt, optlen; 3201 3202 to->to_flags = 0; 3203 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3204 opt = cp[0]; 3205 if (opt == TCPOPT_EOL) 3206 break; 3207 if (opt == TCPOPT_NOP) 3208 optlen = 1; 3209 else { 3210 if (cnt < 2) 3211 break; 3212 optlen = cp[1]; 3213 if (optlen < 2 || optlen > cnt) 3214 break; 3215 } 3216 switch (opt) { 3217 case TCPOPT_MAXSEG: 3218 if (optlen != TCPOLEN_MAXSEG) 3219 continue; 3220 if (!(flags & TO_SYN)) 3221 continue; 3222 to->to_flags |= TOF_MSS; 3223 bcopy((char *)cp + 2, 3224 (char *)&to->to_mss, sizeof(to->to_mss)); 3225 to->to_mss = ntohs(to->to_mss); 3226 break; 3227 case TCPOPT_WINDOW: 3228 if (optlen != TCPOLEN_WINDOW) 3229 continue; 3230 if (!(flags & TO_SYN)) 3231 continue; 3232 to->to_flags |= TOF_SCALE; 3233 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3234 break; 3235 case TCPOPT_TIMESTAMP: 3236 if (optlen != TCPOLEN_TIMESTAMP) 3237 continue; 3238 to->to_flags |= TOF_TS; 3239 bcopy((char *)cp + 2, 3240 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3241 to->to_tsval = ntohl(to->to_tsval); 3242 bcopy((char *)cp + 6, 3243 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3244 to->to_tsecr = ntohl(to->to_tsecr); 3245 break; 3246 #ifdef TCP_SIGNATURE 3247 /* 3248 * XXX In order to reply to a host which has set the 3249 * TCP_SIGNATURE option in its initial SYN, we have to 3250 * record the fact that the option was observed here 3251 * for the syncache code to perform the correct response. 3252 */ 3253 case TCPOPT_SIGNATURE: 3254 if (optlen != TCPOLEN_SIGNATURE) 3255 continue; 3256 to->to_flags |= TOF_SIGNATURE; 3257 to->to_signature = cp + 2; 3258 break; 3259 #endif 3260 case TCPOPT_SACK_PERMITTED: 3261 if (optlen != TCPOLEN_SACK_PERMITTED) 3262 continue; 3263 if (!(flags & TO_SYN)) 3264 continue; 3265 if (!V_tcp_do_sack) 3266 continue; 3267 to->to_flags |= TOF_SACKPERM; 3268 break; 3269 case TCPOPT_SACK: 3270 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3271 continue; 3272 if (flags & TO_SYN) 3273 continue; 3274 to->to_flags |= TOF_SACK; 3275 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3276 to->to_sacks = cp + 2; 3277 TCPSTAT_INC(tcps_sack_rcv_blocks); 3278 break; 3279 default: 3280 continue; 3281 } 3282 } 3283 } 3284 3285 /* 3286 * Pull out of band byte out of a segment so 3287 * it doesn't appear in the user's data queue. 3288 * It is still reflected in the segment length for 3289 * sequencing purposes. 3290 */ 3291 static void 3292 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3293 int off) 3294 { 3295 int cnt = off + th->th_urp - 1; 3296 3297 while (cnt >= 0) { 3298 if (m->m_len > cnt) { 3299 char *cp = mtod(m, caddr_t) + cnt; 3300 struct tcpcb *tp = sototcpcb(so); 3301 3302 INP_WLOCK_ASSERT(tp->t_inpcb); 3303 3304 tp->t_iobc = *cp; 3305 tp->t_oobflags |= TCPOOB_HAVEDATA; 3306 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3307 m->m_len--; 3308 if (m->m_flags & M_PKTHDR) 3309 m->m_pkthdr.len--; 3310 return; 3311 } 3312 cnt -= m->m_len; 3313 m = m->m_next; 3314 if (m == NULL) 3315 break; 3316 } 3317 panic("tcp_pulloutofband"); 3318 } 3319 3320 /* 3321 * Collect new round-trip time estimate 3322 * and update averages and current timeout. 3323 */ 3324 static void 3325 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3326 { 3327 int delta; 3328 3329 INP_WLOCK_ASSERT(tp->t_inpcb); 3330 3331 TCPSTAT_INC(tcps_rttupdated); 3332 tp->t_rttupdated++; 3333 if (tp->t_srtt != 0) { 3334 /* 3335 * srtt is stored as fixed point with 5 bits after the 3336 * binary point (i.e., scaled by 8). The following magic 3337 * is equivalent to the smoothing algorithm in rfc793 with 3338 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3339 * point). Adjust rtt to origin 0. 3340 */ 3341 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3342 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3343 3344 if ((tp->t_srtt += delta) <= 0) 3345 tp->t_srtt = 1; 3346 3347 /* 3348 * We accumulate a smoothed rtt variance (actually, a 3349 * smoothed mean difference), then set the retransmit 3350 * timer to smoothed rtt + 4 times the smoothed variance. 3351 * rttvar is stored as fixed point with 4 bits after the 3352 * binary point (scaled by 16). The following is 3353 * equivalent to rfc793 smoothing with an alpha of .75 3354 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3355 * rfc793's wired-in beta. 3356 */ 3357 if (delta < 0) 3358 delta = -delta; 3359 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3360 if ((tp->t_rttvar += delta) <= 0) 3361 tp->t_rttvar = 1; 3362 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3363 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3364 } else { 3365 /* 3366 * No rtt measurement yet - use the unsmoothed rtt. 3367 * Set the variance to half the rtt (so our first 3368 * retransmit happens at 3*rtt). 3369 */ 3370 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3371 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3372 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3373 } 3374 tp->t_rtttime = 0; 3375 tp->t_rxtshift = 0; 3376 3377 /* 3378 * the retransmit should happen at rtt + 4 * rttvar. 3379 * Because of the way we do the smoothing, srtt and rttvar 3380 * will each average +1/2 tick of bias. When we compute 3381 * the retransmit timer, we want 1/2 tick of rounding and 3382 * 1 extra tick because of +-1/2 tick uncertainty in the 3383 * firing of the timer. The bias will give us exactly the 3384 * 1.5 tick we need. But, because the bias is 3385 * statistical, we have to test that we don't drop below 3386 * the minimum feasible timer (which is 2 ticks). 3387 */ 3388 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3389 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3390 3391 /* 3392 * We received an ack for a packet that wasn't retransmitted; 3393 * it is probably safe to discard any error indications we've 3394 * received recently. This isn't quite right, but close enough 3395 * for now (a route might have failed after we sent a segment, 3396 * and the return path might not be symmetrical). 3397 */ 3398 tp->t_softerror = 0; 3399 } 3400 3401 /* 3402 * Determine a reasonable value for maxseg size. 3403 * If the route is known, check route for mtu. 3404 * If none, use an mss that can be handled on the outgoing interface 3405 * without forcing IP to fragment. If no route is found, route has no mtu, 3406 * or the destination isn't local, use a default, hopefully conservative 3407 * size (usually 512 or the default IP max size, but no more than the mtu 3408 * of the interface), as we can't discover anything about intervening 3409 * gateways or networks. We also initialize the congestion/slow start 3410 * window to be a single segment if the destination isn't local. 3411 * While looking at the routing entry, we also initialize other path-dependent 3412 * parameters from pre-set or cached values in the routing entry. 3413 * 3414 * Also take into account the space needed for options that we 3415 * send regularly. Make maxseg shorter by that amount to assure 3416 * that we can send maxseg amount of data even when the options 3417 * are present. Store the upper limit of the length of options plus 3418 * data in maxopd. 3419 * 3420 * NOTE that this routine is only called when we process an incoming 3421 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3422 * settings are handled in tcp_mssopt(). 3423 */ 3424 void 3425 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3426 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3427 { 3428 int mss = 0; 3429 u_long maxmtu = 0; 3430 struct inpcb *inp = tp->t_inpcb; 3431 struct hc_metrics_lite metrics; 3432 int origoffer; 3433 #ifdef INET6 3434 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3435 size_t min_protoh = isipv6 ? 3436 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3437 sizeof (struct tcpiphdr); 3438 #else 3439 const size_t min_protoh = sizeof(struct tcpiphdr); 3440 #endif 3441 3442 INP_WLOCK_ASSERT(tp->t_inpcb); 3443 3444 if (mtuoffer != -1) { 3445 KASSERT(offer == -1, ("%s: conflict", __func__)); 3446 offer = mtuoffer - min_protoh; 3447 } 3448 origoffer = offer; 3449 3450 /* Initialize. */ 3451 #ifdef INET6 3452 if (isipv6) { 3453 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3454 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3455 } 3456 #endif 3457 #if defined(INET) && defined(INET6) 3458 else 3459 #endif 3460 #ifdef INET 3461 { 3462 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3463 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3464 } 3465 #endif 3466 3467 /* 3468 * No route to sender, stay with default mss and return. 3469 */ 3470 if (maxmtu == 0) { 3471 /* 3472 * In case we return early we need to initialize metrics 3473 * to a defined state as tcp_hc_get() would do for us 3474 * if there was no cache hit. 3475 */ 3476 if (metricptr != NULL) 3477 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3478 return; 3479 } 3480 3481 /* What have we got? */ 3482 switch (offer) { 3483 case 0: 3484 /* 3485 * Offer == 0 means that there was no MSS on the SYN 3486 * segment, in this case we use tcp_mssdflt as 3487 * already assigned to t_maxopd above. 3488 */ 3489 offer = tp->t_maxopd; 3490 break; 3491 3492 case -1: 3493 /* 3494 * Offer == -1 means that we didn't receive SYN yet. 3495 */ 3496 /* FALLTHROUGH */ 3497 3498 default: 3499 /* 3500 * Prevent DoS attack with too small MSS. Round up 3501 * to at least minmss. 3502 */ 3503 offer = max(offer, V_tcp_minmss); 3504 } 3505 3506 /* 3507 * rmx information is now retrieved from tcp_hostcache. 3508 */ 3509 tcp_hc_get(&inp->inp_inc, &metrics); 3510 if (metricptr != NULL) 3511 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3512 3513 /* 3514 * If there's a discovered mtu int tcp hostcache, use it 3515 * else, use the link mtu. 3516 */ 3517 if (metrics.rmx_mtu) 3518 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3519 else { 3520 #ifdef INET6 3521 if (isipv6) { 3522 mss = maxmtu - min_protoh; 3523 if (!V_path_mtu_discovery && 3524 !in6_localaddr(&inp->in6p_faddr)) 3525 mss = min(mss, V_tcp_v6mssdflt); 3526 } 3527 #endif 3528 #if defined(INET) && defined(INET6) 3529 else 3530 #endif 3531 #ifdef INET 3532 { 3533 mss = maxmtu - min_protoh; 3534 if (!V_path_mtu_discovery && 3535 !in_localaddr(inp->inp_faddr)) 3536 mss = min(mss, V_tcp_mssdflt); 3537 } 3538 #endif 3539 /* 3540 * XXX - The above conditional (mss = maxmtu - min_protoh) 3541 * probably violates the TCP spec. 3542 * The problem is that, since we don't know the 3543 * other end's MSS, we are supposed to use a conservative 3544 * default. But, if we do that, then MTU discovery will 3545 * never actually take place, because the conservative 3546 * default is much less than the MTUs typically seen 3547 * on the Internet today. For the moment, we'll sweep 3548 * this under the carpet. 3549 * 3550 * The conservative default might not actually be a problem 3551 * if the only case this occurs is when sending an initial 3552 * SYN with options and data to a host we've never talked 3553 * to before. Then, they will reply with an MSS value which 3554 * will get recorded and the new parameters should get 3555 * recomputed. For Further Study. 3556 */ 3557 } 3558 mss = min(mss, offer); 3559 3560 /* 3561 * Sanity check: make sure that maxopd will be large 3562 * enough to allow some data on segments even if the 3563 * all the option space is used (40bytes). Otherwise 3564 * funny things may happen in tcp_output. 3565 */ 3566 mss = max(mss, 64); 3567 3568 /* 3569 * maxopd stores the maximum length of data AND options 3570 * in a segment; maxseg is the amount of data in a normal 3571 * segment. We need to store this value (maxopd) apart 3572 * from maxseg, because now every segment carries options 3573 * and thus we normally have somewhat less data in segments. 3574 */ 3575 tp->t_maxopd = mss; 3576 3577 /* 3578 * origoffer==-1 indicates that no segments were received yet. 3579 * In this case we just guess. 3580 */ 3581 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3582 (origoffer == -1 || 3583 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3584 mss -= TCPOLEN_TSTAMP_APPA; 3585 3586 tp->t_maxseg = mss; 3587 } 3588 3589 void 3590 tcp_mss(struct tcpcb *tp, int offer) 3591 { 3592 int mss; 3593 u_long bufsize; 3594 struct inpcb *inp; 3595 struct socket *so; 3596 struct hc_metrics_lite metrics; 3597 struct tcp_ifcap cap; 3598 3599 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3600 3601 bzero(&cap, sizeof(cap)); 3602 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3603 3604 mss = tp->t_maxseg; 3605 inp = tp->t_inpcb; 3606 3607 /* 3608 * If there's a pipesize, change the socket buffer to that size, 3609 * don't change if sb_hiwat is different than default (then it 3610 * has been changed on purpose with setsockopt). 3611 * Make the socket buffers an integral number of mss units; 3612 * if the mss is larger than the socket buffer, decrease the mss. 3613 */ 3614 so = inp->inp_socket; 3615 SOCKBUF_LOCK(&so->so_snd); 3616 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3617 bufsize = metrics.rmx_sendpipe; 3618 else 3619 bufsize = so->so_snd.sb_hiwat; 3620 if (bufsize < mss) 3621 mss = bufsize; 3622 else { 3623 bufsize = roundup(bufsize, mss); 3624 if (bufsize > sb_max) 3625 bufsize = sb_max; 3626 if (bufsize > so->so_snd.sb_hiwat) 3627 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3628 } 3629 SOCKBUF_UNLOCK(&so->so_snd); 3630 tp->t_maxseg = mss; 3631 3632 SOCKBUF_LOCK(&so->so_rcv); 3633 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3634 bufsize = metrics.rmx_recvpipe; 3635 else 3636 bufsize = so->so_rcv.sb_hiwat; 3637 if (bufsize > mss) { 3638 bufsize = roundup(bufsize, mss); 3639 if (bufsize > sb_max) 3640 bufsize = sb_max; 3641 if (bufsize > so->so_rcv.sb_hiwat) 3642 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3643 } 3644 SOCKBUF_UNLOCK(&so->so_rcv); 3645 3646 /* Check the interface for TSO capabilities. */ 3647 if (cap.ifcap & CSUM_TSO) { 3648 tp->t_flags |= TF_TSO; 3649 tp->t_tsomax = cap.tsomax; 3650 } 3651 } 3652 3653 /* 3654 * Determine the MSS option to send on an outgoing SYN. 3655 */ 3656 int 3657 tcp_mssopt(struct in_conninfo *inc) 3658 { 3659 int mss = 0; 3660 u_long maxmtu = 0; 3661 u_long thcmtu = 0; 3662 size_t min_protoh; 3663 3664 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3665 3666 #ifdef INET6 3667 if (inc->inc_flags & INC_ISIPV6) { 3668 mss = V_tcp_v6mssdflt; 3669 maxmtu = tcp_maxmtu6(inc, NULL); 3670 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3671 } 3672 #endif 3673 #if defined(INET) && defined(INET6) 3674 else 3675 #endif 3676 #ifdef INET 3677 { 3678 mss = V_tcp_mssdflt; 3679 maxmtu = tcp_maxmtu(inc, NULL); 3680 min_protoh = sizeof(struct tcpiphdr); 3681 } 3682 #endif 3683 #if defined(INET6) || defined(INET) 3684 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3685 #endif 3686 3687 if (maxmtu && thcmtu) 3688 mss = min(maxmtu, thcmtu) - min_protoh; 3689 else if (maxmtu || thcmtu) 3690 mss = max(maxmtu, thcmtu) - min_protoh; 3691 3692 return (mss); 3693 } 3694 3695 3696 /* 3697 * On a partial ack arrives, force the retransmission of the 3698 * next unacknowledged segment. Do not clear tp->t_dupacks. 3699 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3700 * be started again. 3701 */ 3702 static void 3703 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3704 { 3705 tcp_seq onxt = tp->snd_nxt; 3706 u_long ocwnd = tp->snd_cwnd; 3707 3708 INP_WLOCK_ASSERT(tp->t_inpcb); 3709 3710 tcp_timer_activate(tp, TT_REXMT, 0); 3711 tp->t_rtttime = 0; 3712 tp->snd_nxt = th->th_ack; 3713 /* 3714 * Set snd_cwnd to one segment beyond acknowledged offset. 3715 * (tp->snd_una has not yet been updated when this function is called.) 3716 */ 3717 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3718 tp->t_flags |= TF_ACKNOW; 3719 (void) tcp_output(tp); 3720 tp->snd_cwnd = ocwnd; 3721 if (SEQ_GT(onxt, tp->snd_nxt)) 3722 tp->snd_nxt = onxt; 3723 /* 3724 * Partial window deflation. Relies on fact that tp->snd_una 3725 * not updated yet. 3726 */ 3727 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3728 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3729 else 3730 tp->snd_cwnd = 0; 3731 tp->snd_cwnd += tp->t_maxseg; 3732 } 3733