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 <[email protected]> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 4. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_ipfw.h" /* for ipfw_fwd */ 54 #include "opt_inet.h" 55 #include "opt_inet6.h" 56 #include "opt_ipsec.h" 57 #include "opt_tcpdebug.h" 58 59 #include <sys/param.h> 60 #include <sys/kernel.h> 61 #include <sys/hhook.h> 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/signalvar.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/sysctl.h> 70 #include <sys/syslog.h> 71 #include <sys/systm.h> 72 73 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 74 75 #include <vm/uma.h> 76 77 #include <net/if.h> 78 #include <net/route.h> 79 #include <net/vnet.h> 80 81 #define TCPSTATES /* for logging */ 82 83 #include <netinet/cc.h> 84 #include <netinet/in.h> 85 #include <netinet/in_pcb.h> 86 #include <netinet/in_systm.h> 87 #include <netinet/in_var.h> 88 #include <netinet/ip.h> 89 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 90 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 91 #include <netinet/ip_var.h> 92 #include <netinet/ip_options.h> 93 #include <netinet/ip6.h> 94 #include <netinet/icmp6.h> 95 #include <netinet6/in6_pcb.h> 96 #include <netinet6/ip6_var.h> 97 #include <netinet6/nd6.h> 98 #include <netinet/tcp_fsm.h> 99 #include <netinet/tcp_seq.h> 100 #include <netinet/tcp_timer.h> 101 #include <netinet/tcp_var.h> 102 #include <netinet6/tcp6_var.h> 103 #include <netinet/tcpip.h> 104 #include <netinet/tcp_syncache.h> 105 #ifdef TCPDEBUG 106 #include <netinet/tcp_debug.h> 107 #endif /* TCPDEBUG */ 108 #ifdef TCP_OFFLOAD 109 #include <netinet/tcp_offload.h> 110 #endif 111 112 #ifdef IPSEC 113 #include <netipsec/ipsec.h> 114 #include <netipsec/ipsec6.h> 115 #endif /*IPSEC*/ 116 117 #include <machine/in_cksum.h> 118 119 #include <security/mac/mac_framework.h> 120 121 const int tcprexmtthresh = 3; 122 123 int tcp_log_in_vain = 0; 124 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 125 &tcp_log_in_vain, 0, 126 "Log all incoming TCP segments to closed ports"); 127 128 VNET_DEFINE(int, blackhole) = 0; 129 #define V_blackhole VNET(blackhole) 130 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 131 &VNET_NAME(blackhole), 0, 132 "Do not send RST on segments to closed ports"); 133 134 VNET_DEFINE(int, tcp_delack_enabled) = 1; 135 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 136 &VNET_NAME(tcp_delack_enabled), 0, 137 "Delay ACK to try and piggyback it onto a data packet"); 138 139 VNET_DEFINE(int, drop_synfin) = 0; 140 #define V_drop_synfin VNET(drop_synfin) 141 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 142 &VNET_NAME(drop_synfin), 0, 143 "Drop TCP packets with SYN+FIN set"); 144 145 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 146 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 147 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 148 &VNET_NAME(tcp_do_rfc3042), 0, 149 "Enable RFC 3042 (Limited Transmit)"); 150 151 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 152 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 153 &VNET_NAME(tcp_do_rfc3390), 0, 154 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 155 156 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 157 "Experimental TCP extensions"); 158 159 VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 160 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 161 &VNET_NAME(tcp_do_initcwnd10), 0, 162 "Enable draft-ietf-tcpm-initcwnd-05 (Increasing initial CWND to 10)"); 163 164 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 165 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 166 &VNET_NAME(tcp_do_rfc3465), 0, 167 "Enable RFC 3465 (Appropriate Byte Counting)"); 168 169 VNET_DEFINE(int, tcp_abc_l_var) = 2; 170 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 171 &VNET_NAME(tcp_abc_l_var), 2, 172 "Cap the max cwnd increment during slow-start to this number of segments"); 173 174 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 175 176 VNET_DEFINE(int, tcp_do_ecn) = 0; 177 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 178 &VNET_NAME(tcp_do_ecn), 0, 179 "TCP ECN support"); 180 181 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 182 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 183 &VNET_NAME(tcp_ecn_maxretries), 0, 184 "Max retries before giving up on ECN"); 185 186 VNET_DEFINE(int, tcp_insecure_rst) = 0; 187 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 188 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 189 &VNET_NAME(tcp_insecure_rst), 0, 190 "Follow the old (insecure) criteria for accepting RST packets"); 191 192 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 193 #define V_tcp_recvspace VNET(tcp_recvspace) 194 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 195 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 196 197 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 198 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 199 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 200 &VNET_NAME(tcp_do_autorcvbuf), 0, 201 "Enable automatic receive buffer sizing"); 202 203 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 204 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 205 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 206 &VNET_NAME(tcp_autorcvbuf_inc), 0, 207 "Incrementor step size of automatic receive buffer"); 208 209 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 210 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 211 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 212 &VNET_NAME(tcp_autorcvbuf_max), 0, 213 "Max size of automatic receive buffer"); 214 215 VNET_DEFINE(struct inpcbhead, tcb); 216 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 217 VNET_DEFINE(struct inpcbinfo, tcbinfo); 218 219 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 220 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 221 struct socket *, struct tcpcb *, int, int, uint8_t, 222 int); 223 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 224 struct tcpcb *, int, int); 225 static void tcp_pulloutofband(struct socket *, 226 struct tcphdr *, struct mbuf *, int); 227 static void tcp_xmit_timer(struct tcpcb *, int); 228 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 229 static void inline tcp_fields_to_host(struct tcphdr *); 230 #ifdef TCP_SIGNATURE 231 static void inline tcp_fields_to_net(struct tcphdr *); 232 static int inline tcp_signature_verify_input(struct mbuf *, int, int, 233 int, struct tcpopt *, struct tcphdr *, u_int); 234 #endif 235 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 236 uint16_t type); 237 static void inline cc_conn_init(struct tcpcb *tp); 238 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 239 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 240 struct tcphdr *th, struct tcpopt *to); 241 242 /* 243 * TCP statistics are stored in an "array" of counter(9)s. 244 */ 245 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 246 VNET_PCPUSTAT_SYSINIT(tcpstat); 247 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 248 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 249 250 #ifdef VIMAGE 251 VNET_PCPUSTAT_SYSUNINIT(tcpstat); 252 #endif /* VIMAGE */ 253 /* 254 * Kernel module interface for updating tcpstat. The argument is an index 255 * into tcpstat treated as an array. 256 */ 257 void 258 kmod_tcpstat_inc(int statnum) 259 { 260 261 counter_u64_add(VNET(tcpstat)[statnum], 1); 262 } 263 264 /* 265 * Wrapper for the TCP established input helper hook. 266 */ 267 static void inline 268 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 269 { 270 struct tcp_hhook_data hhook_data; 271 272 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 273 hhook_data.tp = tp; 274 hhook_data.th = th; 275 hhook_data.to = to; 276 277 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 278 tp->osd); 279 } 280 } 281 282 /* 283 * CC wrapper hook functions 284 */ 285 static void inline 286 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 287 { 288 INP_WLOCK_ASSERT(tp->t_inpcb); 289 290 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 291 if (tp->snd_cwnd <= tp->snd_wnd) 292 tp->ccv->flags |= CCF_CWND_LIMITED; 293 else 294 tp->ccv->flags &= ~CCF_CWND_LIMITED; 295 296 if (type == CC_ACK) { 297 if (tp->snd_cwnd > tp->snd_ssthresh) { 298 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 299 V_tcp_abc_l_var * tp->t_maxseg); 300 if (tp->t_bytes_acked >= tp->snd_cwnd) { 301 tp->t_bytes_acked -= tp->snd_cwnd; 302 tp->ccv->flags |= CCF_ABC_SENTAWND; 303 } 304 } else { 305 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 306 tp->t_bytes_acked = 0; 307 } 308 } 309 310 if (CC_ALGO(tp)->ack_received != NULL) { 311 /* XXXLAS: Find a way to live without this */ 312 tp->ccv->curack = th->th_ack; 313 CC_ALGO(tp)->ack_received(tp->ccv, type); 314 } 315 } 316 317 static void inline 318 cc_conn_init(struct tcpcb *tp) 319 { 320 struct hc_metrics_lite metrics; 321 struct inpcb *inp = tp->t_inpcb; 322 int rtt; 323 324 INP_WLOCK_ASSERT(tp->t_inpcb); 325 326 tcp_hc_get(&inp->inp_inc, &metrics); 327 328 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 329 tp->t_srtt = rtt; 330 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 331 TCPSTAT_INC(tcps_usedrtt); 332 if (metrics.rmx_rttvar) { 333 tp->t_rttvar = metrics.rmx_rttvar; 334 TCPSTAT_INC(tcps_usedrttvar); 335 } else { 336 /* default variation is +- 1 rtt */ 337 tp->t_rttvar = 338 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 339 } 340 TCPT_RANGESET(tp->t_rxtcur, 341 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 342 tp->t_rttmin, TCPTV_REXMTMAX); 343 } 344 if (metrics.rmx_ssthresh) { 345 /* 346 * There's some sort of gateway or interface 347 * buffer limit on the path. Use this to set 348 * the slow start threshhold, but set the 349 * threshold to no less than 2*mss. 350 */ 351 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 352 TCPSTAT_INC(tcps_usedssthresh); 353 } 354 355 /* 356 * Set the initial slow-start flight size. 357 * 358 * RFC5681 Section 3.1 specifies the default conservative values. 359 * RFC3390 specifies slightly more aggressive values. 360 * Draft-ietf-tcpm-initcwnd-05 increases it to ten segments. 361 * 362 * If a SYN or SYN/ACK was lost and retransmitted, we have to 363 * reduce the initial CWND to one segment as congestion is likely 364 * requiring us to be cautious. 365 */ 366 if (tp->snd_cwnd == 1) 367 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 368 else if (V_tcp_do_initcwnd10) 369 tp->snd_cwnd = min(10 * tp->t_maxseg, 370 max(2 * tp->t_maxseg, 14600)); 371 else if (V_tcp_do_rfc3390) 372 tp->snd_cwnd = min(4 * tp->t_maxseg, 373 max(2 * tp->t_maxseg, 4380)); 374 else { 375 /* Per RFC5681 Section 3.1 */ 376 if (tp->t_maxseg > 2190) 377 tp->snd_cwnd = 2 * tp->t_maxseg; 378 else if (tp->t_maxseg > 1095) 379 tp->snd_cwnd = 3 * tp->t_maxseg; 380 else 381 tp->snd_cwnd = 4 * tp->t_maxseg; 382 } 383 384 if (CC_ALGO(tp)->conn_init != NULL) 385 CC_ALGO(tp)->conn_init(tp->ccv); 386 } 387 388 void inline 389 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 390 { 391 INP_WLOCK_ASSERT(tp->t_inpcb); 392 393 switch(type) { 394 case CC_NDUPACK: 395 if (!IN_FASTRECOVERY(tp->t_flags)) { 396 tp->snd_recover = tp->snd_max; 397 if (tp->t_flags & TF_ECN_PERMIT) 398 tp->t_flags |= TF_ECN_SND_CWR; 399 } 400 break; 401 case CC_ECN: 402 if (!IN_CONGRECOVERY(tp->t_flags)) { 403 TCPSTAT_INC(tcps_ecn_rcwnd); 404 tp->snd_recover = tp->snd_max; 405 if (tp->t_flags & TF_ECN_PERMIT) 406 tp->t_flags |= TF_ECN_SND_CWR; 407 } 408 break; 409 case CC_RTO: 410 tp->t_dupacks = 0; 411 tp->t_bytes_acked = 0; 412 EXIT_RECOVERY(tp->t_flags); 413 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 414 tp->t_maxseg) * tp->t_maxseg; 415 tp->snd_cwnd = tp->t_maxseg; 416 break; 417 case CC_RTO_ERR: 418 TCPSTAT_INC(tcps_sndrexmitbad); 419 /* RTO was unnecessary, so reset everything. */ 420 tp->snd_cwnd = tp->snd_cwnd_prev; 421 tp->snd_ssthresh = tp->snd_ssthresh_prev; 422 tp->snd_recover = tp->snd_recover_prev; 423 if (tp->t_flags & TF_WASFRECOVERY) 424 ENTER_FASTRECOVERY(tp->t_flags); 425 if (tp->t_flags & TF_WASCRECOVERY) 426 ENTER_CONGRECOVERY(tp->t_flags); 427 tp->snd_nxt = tp->snd_max; 428 tp->t_flags &= ~TF_PREVVALID; 429 tp->t_badrxtwin = 0; 430 break; 431 } 432 433 if (CC_ALGO(tp)->cong_signal != NULL) { 434 if (th != NULL) 435 tp->ccv->curack = th->th_ack; 436 CC_ALGO(tp)->cong_signal(tp->ccv, type); 437 } 438 } 439 440 static void inline 441 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 442 { 443 INP_WLOCK_ASSERT(tp->t_inpcb); 444 445 /* XXXLAS: KASSERT that we're in recovery? */ 446 447 if (CC_ALGO(tp)->post_recovery != NULL) { 448 tp->ccv->curack = th->th_ack; 449 CC_ALGO(tp)->post_recovery(tp->ccv); 450 } 451 /* XXXLAS: EXIT_RECOVERY ? */ 452 tp->t_bytes_acked = 0; 453 } 454 455 static inline void 456 tcp_fields_to_host(struct tcphdr *th) 457 { 458 459 th->th_seq = ntohl(th->th_seq); 460 th->th_ack = ntohl(th->th_ack); 461 th->th_win = ntohs(th->th_win); 462 th->th_urp = ntohs(th->th_urp); 463 } 464 465 #ifdef TCP_SIGNATURE 466 static inline void 467 tcp_fields_to_net(struct tcphdr *th) 468 { 469 470 th->th_seq = htonl(th->th_seq); 471 th->th_ack = htonl(th->th_ack); 472 th->th_win = htons(th->th_win); 473 th->th_urp = htons(th->th_urp); 474 } 475 476 static inline int 477 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 478 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 479 { 480 int ret; 481 482 tcp_fields_to_net(th); 483 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 484 tcp_fields_to_host(th); 485 return (ret); 486 } 487 #endif 488 489 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 490 #ifdef INET6 491 #define ND6_HINT(tp) \ 492 do { \ 493 if ((tp) && (tp)->t_inpcb && \ 494 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 495 nd6_nud_hint(NULL, NULL, 0); \ 496 } while (0) 497 #else 498 #define ND6_HINT(tp) 499 #endif 500 501 /* 502 * Indicate whether this ack should be delayed. We can delay the ack if 503 * - there is no delayed ack timer in progress and 504 * - our last ack wasn't a 0-sized window. We never want to delay 505 * the ack that opens up a 0-sized window and 506 * - delayed acks are enabled or 507 * - this is a half-synchronized T/TCP connection. 508 */ 509 #define DELAY_ACK(tp) \ 510 ((!tcp_timer_active(tp, TT_DELACK) && \ 511 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 512 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 513 514 /* 515 * TCP input handling is split into multiple parts: 516 * tcp6_input is a thin wrapper around tcp_input for the extended 517 * ip6_protox[] call format in ip6_input 518 * tcp_input handles primary segment validation, inpcb lookup and 519 * SYN processing on listen sockets 520 * tcp_do_segment processes the ACK and text of the segment for 521 * establishing, established and closing connections 522 */ 523 #ifdef INET6 524 int 525 tcp6_input(struct mbuf **mp, int *offp, int proto) 526 { 527 struct mbuf *m = *mp; 528 struct in6_ifaddr *ia6; 529 530 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 531 532 /* 533 * draft-itojun-ipv6-tcp-to-anycast 534 * better place to put this in? 535 */ 536 ia6 = ip6_getdstifaddr(m); 537 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 538 struct ip6_hdr *ip6; 539 540 ifa_free(&ia6->ia_ifa); 541 ip6 = mtod(m, struct ip6_hdr *); 542 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 543 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 544 return IPPROTO_DONE; 545 } 546 if (ia6) 547 ifa_free(&ia6->ia_ifa); 548 549 tcp_input(m, *offp); 550 return IPPROTO_DONE; 551 } 552 #endif /* INET6 */ 553 554 void 555 tcp_input(struct mbuf *m, int off0) 556 { 557 struct tcphdr *th = NULL; 558 struct ip *ip = NULL; 559 struct inpcb *inp = NULL; 560 struct tcpcb *tp = NULL; 561 struct socket *so = NULL; 562 u_char *optp = NULL; 563 int optlen = 0; 564 #ifdef INET 565 int len; 566 #endif 567 int tlen = 0, off; 568 int drop_hdrlen; 569 int thflags; 570 int rstreason = 0; /* For badport_bandlim accounting purposes */ 571 #ifdef TCP_SIGNATURE 572 uint8_t sig_checked = 0; 573 #endif 574 uint8_t iptos = 0; 575 struct m_tag *fwd_tag = NULL; 576 #ifdef INET6 577 struct ip6_hdr *ip6 = NULL; 578 int isipv6; 579 #else 580 const void *ip6 = NULL; 581 #endif /* INET6 */ 582 struct tcpopt to; /* options in this segment */ 583 char *s = NULL; /* address and port logging */ 584 int ti_locked; 585 #define TI_UNLOCKED 1 586 #define TI_WLOCKED 2 587 588 #ifdef TCPDEBUG 589 /* 590 * The size of tcp_saveipgen must be the size of the max ip header, 591 * now IPv6. 592 */ 593 u_char tcp_saveipgen[IP6_HDR_LEN]; 594 struct tcphdr tcp_savetcp; 595 short ostate = 0; 596 #endif 597 598 #ifdef INET6 599 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 600 #endif 601 602 to.to_flags = 0; 603 TCPSTAT_INC(tcps_rcvtotal); 604 605 #ifdef INET6 606 if (isipv6) { 607 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 608 609 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 610 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 611 if (m == NULL) { 612 TCPSTAT_INC(tcps_rcvshort); 613 return; 614 } 615 } 616 617 ip6 = mtod(m, struct ip6_hdr *); 618 th = (struct tcphdr *)((caddr_t)ip6 + off0); 619 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 620 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 621 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 622 th->th_sum = m->m_pkthdr.csum_data; 623 else 624 th->th_sum = in6_cksum_pseudo(ip6, tlen, 625 IPPROTO_TCP, m->m_pkthdr.csum_data); 626 th->th_sum ^= 0xffff; 627 } else 628 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 629 if (th->th_sum) { 630 TCPSTAT_INC(tcps_rcvbadsum); 631 goto drop; 632 } 633 634 /* 635 * Be proactive about unspecified IPv6 address in source. 636 * As we use all-zero to indicate unbounded/unconnected pcb, 637 * unspecified IPv6 address can be used to confuse us. 638 * 639 * Note that packets with unspecified IPv6 destination is 640 * already dropped in ip6_input. 641 */ 642 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 643 /* XXX stat */ 644 goto drop; 645 } 646 } 647 #endif 648 #if defined(INET) && defined(INET6) 649 else 650 #endif 651 #ifdef INET 652 { 653 /* 654 * Get IP and TCP header together in first mbuf. 655 * Note: IP leaves IP header in first mbuf. 656 */ 657 if (off0 > sizeof (struct ip)) { 658 ip_stripoptions(m); 659 off0 = sizeof(struct ip); 660 } 661 if (m->m_len < sizeof (struct tcpiphdr)) { 662 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 663 == NULL) { 664 TCPSTAT_INC(tcps_rcvshort); 665 return; 666 } 667 } 668 ip = mtod(m, struct ip *); 669 th = (struct tcphdr *)((caddr_t)ip + off0); 670 tlen = ntohs(ip->ip_len) - off0; 671 672 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 673 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 674 th->th_sum = m->m_pkthdr.csum_data; 675 else 676 th->th_sum = in_pseudo(ip->ip_src.s_addr, 677 ip->ip_dst.s_addr, 678 htonl(m->m_pkthdr.csum_data + tlen + 679 IPPROTO_TCP)); 680 th->th_sum ^= 0xffff; 681 } else { 682 struct ipovly *ipov = (struct ipovly *)ip; 683 684 /* 685 * Checksum extended TCP header and data. 686 */ 687 len = off0 + tlen; 688 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 689 ipov->ih_len = htons(tlen); 690 th->th_sum = in_cksum(m, len); 691 } 692 if (th->th_sum) { 693 TCPSTAT_INC(tcps_rcvbadsum); 694 goto drop; 695 } 696 /* Re-initialization for later version check */ 697 ip->ip_v = IPVERSION; 698 } 699 #endif /* INET */ 700 701 #ifdef INET6 702 if (isipv6) 703 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 704 #endif 705 #if defined(INET) && defined(INET6) 706 else 707 #endif 708 #ifdef INET 709 iptos = ip->ip_tos; 710 #endif 711 712 /* 713 * Check that TCP offset makes sense, 714 * pull out TCP options and adjust length. XXX 715 */ 716 off = th->th_off << 2; 717 if (off < sizeof (struct tcphdr) || off > tlen) { 718 TCPSTAT_INC(tcps_rcvbadoff); 719 goto drop; 720 } 721 tlen -= off; /* tlen is used instead of ti->ti_len */ 722 if (off > sizeof (struct tcphdr)) { 723 #ifdef INET6 724 if (isipv6) { 725 IP6_EXTHDR_CHECK(m, off0, off, ); 726 ip6 = mtod(m, struct ip6_hdr *); 727 th = (struct tcphdr *)((caddr_t)ip6 + off0); 728 } 729 #endif 730 #if defined(INET) && defined(INET6) 731 else 732 #endif 733 #ifdef INET 734 { 735 if (m->m_len < sizeof(struct ip) + off) { 736 if ((m = m_pullup(m, sizeof (struct ip) + off)) 737 == NULL) { 738 TCPSTAT_INC(tcps_rcvshort); 739 return; 740 } 741 ip = mtod(m, struct ip *); 742 th = (struct tcphdr *)((caddr_t)ip + off0); 743 } 744 } 745 #endif 746 optlen = off - sizeof (struct tcphdr); 747 optp = (u_char *)(th + 1); 748 } 749 thflags = th->th_flags; 750 751 /* 752 * Convert TCP protocol specific fields to host format. 753 */ 754 tcp_fields_to_host(th); 755 756 /* 757 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 758 */ 759 drop_hdrlen = off0 + off; 760 761 /* 762 * Locate pcb for segment; if we're likely to add or remove a 763 * connection then first acquire pcbinfo lock. There are two cases 764 * where we might discover later we need a write lock despite the 765 * flags: ACKs moving a connection out of the syncache, and ACKs for 766 * a connection in TIMEWAIT. 767 */ 768 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 769 INP_INFO_WLOCK(&V_tcbinfo); 770 ti_locked = TI_WLOCKED; 771 } else 772 ti_locked = TI_UNLOCKED; 773 774 /* 775 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 776 */ 777 if ( 778 #ifdef INET6 779 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 780 #ifdef INET 781 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 782 #endif 783 #endif 784 #if defined(INET) && !defined(INET6) 785 (m->m_flags & M_IP_NEXTHOP) 786 #endif 787 ) 788 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 789 790 findpcb: 791 #ifdef INVARIANTS 792 if (ti_locked == TI_WLOCKED) { 793 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 794 } else { 795 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 796 } 797 #endif 798 #ifdef INET6 799 if (isipv6 && fwd_tag != NULL) { 800 struct sockaddr_in6 *next_hop6; 801 802 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 803 /* 804 * Transparently forwarded. Pretend to be the destination. 805 * Already got one like this? 806 */ 807 inp = in6_pcblookup_mbuf(&V_tcbinfo, 808 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 809 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 810 if (!inp) { 811 /* 812 * It's new. Try to find the ambushing socket. 813 * Because we've rewritten the destination address, 814 * any hardware-generated hash is ignored. 815 */ 816 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 817 th->th_sport, &next_hop6->sin6_addr, 818 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 819 th->th_dport, INPLOOKUP_WILDCARD | 820 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 821 } 822 } else if (isipv6) { 823 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 824 th->th_sport, &ip6->ip6_dst, th->th_dport, 825 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 826 m->m_pkthdr.rcvif, m); 827 } 828 #endif /* INET6 */ 829 #if defined(INET6) && defined(INET) 830 else 831 #endif 832 #ifdef INET 833 if (fwd_tag != NULL) { 834 struct sockaddr_in *next_hop; 835 836 next_hop = (struct sockaddr_in *)(fwd_tag+1); 837 /* 838 * Transparently forwarded. Pretend to be the destination. 839 * already got one like this? 840 */ 841 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 842 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 843 m->m_pkthdr.rcvif, m); 844 if (!inp) { 845 /* 846 * It's new. Try to find the ambushing socket. 847 * Because we've rewritten the destination address, 848 * any hardware-generated hash is ignored. 849 */ 850 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 851 th->th_sport, next_hop->sin_addr, 852 next_hop->sin_port ? ntohs(next_hop->sin_port) : 853 th->th_dport, INPLOOKUP_WILDCARD | 854 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 855 } 856 } else 857 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 858 th->th_sport, ip->ip_dst, th->th_dport, 859 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 860 m->m_pkthdr.rcvif, m); 861 #endif /* INET */ 862 863 /* 864 * If the INPCB does not exist then all data in the incoming 865 * segment is discarded and an appropriate RST is sent back. 866 * XXX MRT Send RST using which routing table? 867 */ 868 if (inp == NULL) { 869 /* 870 * Log communication attempts to ports that are not 871 * in use. 872 */ 873 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 874 tcp_log_in_vain == 2) { 875 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 876 log(LOG_INFO, "%s; %s: Connection attempt " 877 "to closed port\n", s, __func__); 878 } 879 /* 880 * When blackholing do not respond with a RST but 881 * completely ignore the segment and drop it. 882 */ 883 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 884 V_blackhole == 2) 885 goto dropunlock; 886 887 rstreason = BANDLIM_RST_CLOSEDPORT; 888 goto dropwithreset; 889 } 890 INP_WLOCK_ASSERT(inp); 891 if (!(inp->inp_flags & INP_HW_FLOWID) 892 && (m->m_flags & M_FLOWID) 893 && ((inp->inp_socket == NULL) 894 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 895 inp->inp_flags |= INP_HW_FLOWID; 896 inp->inp_flags &= ~INP_SW_FLOWID; 897 inp->inp_flowid = m->m_pkthdr.flowid; 898 } 899 #ifdef IPSEC 900 #ifdef INET6 901 if (isipv6 && ipsec6_in_reject(m, inp)) { 902 IPSEC6STAT_INC(in_polvio); 903 goto dropunlock; 904 } else 905 #endif /* INET6 */ 906 if (ipsec4_in_reject(m, inp) != 0) { 907 IPSECSTAT_INC(in_polvio); 908 goto dropunlock; 909 } 910 #endif /* IPSEC */ 911 912 /* 913 * Check the minimum TTL for socket. 914 */ 915 if (inp->inp_ip_minttl != 0) { 916 #ifdef INET6 917 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 918 goto dropunlock; 919 else 920 #endif 921 if (inp->inp_ip_minttl > ip->ip_ttl) 922 goto dropunlock; 923 } 924 925 /* 926 * A previous connection in TIMEWAIT state is supposed to catch stray 927 * or duplicate segments arriving late. If this segment was a 928 * legitimate new connection attempt, the old INPCB gets removed and 929 * we can try again to find a listening socket. 930 * 931 * At this point, due to earlier optimism, we may hold only an inpcb 932 * lock, and not the inpcbinfo write lock. If so, we need to try to 933 * acquire it, or if that fails, acquire a reference on the inpcb, 934 * drop all locks, acquire a global write lock, and then re-acquire 935 * the inpcb lock. We may at that point discover that another thread 936 * has tried to free the inpcb, in which case we need to loop back 937 * and try to find a new inpcb to deliver to. 938 * 939 * XXXRW: It may be time to rethink timewait locking. 940 */ 941 relocked: 942 if (inp->inp_flags & INP_TIMEWAIT) { 943 if (ti_locked == TI_UNLOCKED) { 944 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 945 in_pcbref(inp); 946 INP_WUNLOCK(inp); 947 INP_INFO_WLOCK(&V_tcbinfo); 948 ti_locked = TI_WLOCKED; 949 INP_WLOCK(inp); 950 if (in_pcbrele_wlocked(inp)) { 951 inp = NULL; 952 goto findpcb; 953 } 954 } else 955 ti_locked = TI_WLOCKED; 956 } 957 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 958 959 if (thflags & TH_SYN) 960 tcp_dooptions(&to, optp, optlen, TO_SYN); 961 /* 962 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 963 */ 964 if (tcp_twcheck(inp, &to, th, m, tlen)) 965 goto findpcb; 966 INP_INFO_WUNLOCK(&V_tcbinfo); 967 return; 968 } 969 /* 970 * The TCPCB may no longer exist if the connection is winding 971 * down or it is in the CLOSED state. Either way we drop the 972 * segment and send an appropriate response. 973 */ 974 tp = intotcpcb(inp); 975 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 976 rstreason = BANDLIM_RST_CLOSEDPORT; 977 goto dropwithreset; 978 } 979 980 #ifdef TCP_OFFLOAD 981 if (tp->t_flags & TF_TOE) { 982 tcp_offload_input(tp, m); 983 m = NULL; /* consumed by the TOE driver */ 984 goto dropunlock; 985 } 986 #endif 987 988 /* 989 * We've identified a valid inpcb, but it could be that we need an 990 * inpcbinfo write lock but don't hold it. In this case, attempt to 991 * acquire using the same strategy as the TIMEWAIT case above. If we 992 * relock, we have to jump back to 'relocked' as the connection might 993 * now be in TIMEWAIT. 994 */ 995 #ifdef INVARIANTS 996 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 997 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 998 #endif 999 if (tp->t_state != TCPS_ESTABLISHED) { 1000 if (ti_locked == TI_UNLOCKED) { 1001 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 1002 in_pcbref(inp); 1003 INP_WUNLOCK(inp); 1004 INP_INFO_WLOCK(&V_tcbinfo); 1005 ti_locked = TI_WLOCKED; 1006 INP_WLOCK(inp); 1007 if (in_pcbrele_wlocked(inp)) { 1008 inp = NULL; 1009 goto findpcb; 1010 } 1011 goto relocked; 1012 } else 1013 ti_locked = TI_WLOCKED; 1014 } 1015 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1016 } 1017 1018 #ifdef MAC 1019 INP_WLOCK_ASSERT(inp); 1020 if (mac_inpcb_check_deliver(inp, m)) 1021 goto dropunlock; 1022 #endif 1023 so = inp->inp_socket; 1024 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1025 #ifdef TCPDEBUG 1026 if (so->so_options & SO_DEBUG) { 1027 ostate = tp->t_state; 1028 #ifdef INET6 1029 if (isipv6) { 1030 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1031 } else 1032 #endif 1033 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1034 tcp_savetcp = *th; 1035 } 1036 #endif /* TCPDEBUG */ 1037 /* 1038 * When the socket is accepting connections (the INPCB is in LISTEN 1039 * state) we look into the SYN cache if this is a new connection 1040 * attempt or the completion of a previous one. Because listen 1041 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1042 * held in this case. 1043 */ 1044 if (so->so_options & SO_ACCEPTCONN) { 1045 struct in_conninfo inc; 1046 1047 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1048 "tp not listening", __func__)); 1049 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1050 1051 bzero(&inc, sizeof(inc)); 1052 #ifdef INET6 1053 if (isipv6) { 1054 inc.inc_flags |= INC_ISIPV6; 1055 inc.inc6_faddr = ip6->ip6_src; 1056 inc.inc6_laddr = ip6->ip6_dst; 1057 } else 1058 #endif 1059 { 1060 inc.inc_faddr = ip->ip_src; 1061 inc.inc_laddr = ip->ip_dst; 1062 } 1063 inc.inc_fport = th->th_sport; 1064 inc.inc_lport = th->th_dport; 1065 inc.inc_fibnum = so->so_fibnum; 1066 1067 /* 1068 * Check for an existing connection attempt in syncache if 1069 * the flag is only ACK. A successful lookup creates a new 1070 * socket appended to the listen queue in SYN_RECEIVED state. 1071 */ 1072 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1073 /* 1074 * Parse the TCP options here because 1075 * syncookies need access to the reflected 1076 * timestamp. 1077 */ 1078 tcp_dooptions(&to, optp, optlen, 0); 1079 /* 1080 * NB: syncache_expand() doesn't unlock 1081 * inp and tcpinfo locks. 1082 */ 1083 if (!syncache_expand(&inc, &to, th, &so, m)) { 1084 /* 1085 * No syncache entry or ACK was not 1086 * for our SYN/ACK. Send a RST. 1087 * NB: syncache did its own logging 1088 * of the failure cause. 1089 */ 1090 rstreason = BANDLIM_RST_OPENPORT; 1091 goto dropwithreset; 1092 } 1093 if (so == NULL) { 1094 /* 1095 * We completed the 3-way handshake 1096 * but could not allocate a socket 1097 * either due to memory shortage, 1098 * listen queue length limits or 1099 * global socket limits. Send RST 1100 * or wait and have the remote end 1101 * retransmit the ACK for another 1102 * try. 1103 */ 1104 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1105 log(LOG_DEBUG, "%s; %s: Listen socket: " 1106 "Socket allocation failed due to " 1107 "limits or memory shortage, %s\n", 1108 s, __func__, 1109 V_tcp_sc_rst_sock_fail ? 1110 "sending RST" : "try again"); 1111 if (V_tcp_sc_rst_sock_fail) { 1112 rstreason = BANDLIM_UNLIMITED; 1113 goto dropwithreset; 1114 } else 1115 goto dropunlock; 1116 } 1117 /* 1118 * Socket is created in state SYN_RECEIVED. 1119 * Unlock the listen socket, lock the newly 1120 * created socket and update the tp variable. 1121 */ 1122 INP_WUNLOCK(inp); /* listen socket */ 1123 inp = sotoinpcb(so); 1124 INP_WLOCK(inp); /* new connection */ 1125 tp = intotcpcb(inp); 1126 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1127 ("%s: ", __func__)); 1128 #ifdef TCP_SIGNATURE 1129 if (sig_checked == 0) { 1130 tcp_dooptions(&to, optp, optlen, 1131 (thflags & TH_SYN) ? TO_SYN : 0); 1132 if (!tcp_signature_verify_input(m, off0, tlen, 1133 optlen, &to, th, tp->t_flags)) { 1134 1135 /* 1136 * In SYN_SENT state if it receives an 1137 * RST, it is allowed for further 1138 * processing. 1139 */ 1140 if ((thflags & TH_RST) == 0 || 1141 (tp->t_state == TCPS_SYN_SENT) == 0) 1142 goto dropunlock; 1143 } 1144 sig_checked = 1; 1145 } 1146 #endif 1147 1148 /* 1149 * Process the segment and the data it 1150 * contains. tcp_do_segment() consumes 1151 * the mbuf chain and unlocks the inpcb. 1152 */ 1153 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1154 iptos, ti_locked); 1155 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1156 return; 1157 } 1158 /* 1159 * Segment flag validation for new connection attempts: 1160 * 1161 * Our (SYN|ACK) response was rejected. 1162 * Check with syncache and remove entry to prevent 1163 * retransmits. 1164 * 1165 * NB: syncache_chkrst does its own logging of failure 1166 * causes. 1167 */ 1168 if (thflags & TH_RST) { 1169 syncache_chkrst(&inc, th); 1170 goto dropunlock; 1171 } 1172 /* 1173 * We can't do anything without SYN. 1174 */ 1175 if ((thflags & TH_SYN) == 0) { 1176 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1177 log(LOG_DEBUG, "%s; %s: Listen socket: " 1178 "SYN is missing, segment ignored\n", 1179 s, __func__); 1180 TCPSTAT_INC(tcps_badsyn); 1181 goto dropunlock; 1182 } 1183 /* 1184 * (SYN|ACK) is bogus on a listen socket. 1185 */ 1186 if (thflags & TH_ACK) { 1187 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1188 log(LOG_DEBUG, "%s; %s: Listen socket: " 1189 "SYN|ACK invalid, segment rejected\n", 1190 s, __func__); 1191 syncache_badack(&inc); /* XXX: Not needed! */ 1192 TCPSTAT_INC(tcps_badsyn); 1193 rstreason = BANDLIM_RST_OPENPORT; 1194 goto dropwithreset; 1195 } 1196 /* 1197 * If the drop_synfin option is enabled, drop all 1198 * segments with both the SYN and FIN bits set. 1199 * This prevents e.g. nmap from identifying the 1200 * TCP/IP stack. 1201 * XXX: Poor reasoning. nmap has other methods 1202 * and is constantly refining its stack detection 1203 * strategies. 1204 * XXX: This is a violation of the TCP specification 1205 * and was used by RFC1644. 1206 */ 1207 if ((thflags & TH_FIN) && V_drop_synfin) { 1208 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1209 log(LOG_DEBUG, "%s; %s: Listen socket: " 1210 "SYN|FIN segment ignored (based on " 1211 "sysctl setting)\n", s, __func__); 1212 TCPSTAT_INC(tcps_badsyn); 1213 goto dropunlock; 1214 } 1215 /* 1216 * Segment's flags are (SYN) or (SYN|FIN). 1217 * 1218 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1219 * as they do not affect the state of the TCP FSM. 1220 * The data pointed to by TH_URG and th_urp is ignored. 1221 */ 1222 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1223 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1224 KASSERT(thflags & (TH_SYN), 1225 ("%s: Listen socket: TH_SYN not set", __func__)); 1226 #ifdef INET6 1227 /* 1228 * If deprecated address is forbidden, 1229 * we do not accept SYN to deprecated interface 1230 * address to prevent any new inbound connection from 1231 * getting established. 1232 * When we do not accept SYN, we send a TCP RST, 1233 * with deprecated source address (instead of dropping 1234 * it). We compromise it as it is much better for peer 1235 * to send a RST, and RST will be the final packet 1236 * for the exchange. 1237 * 1238 * If we do not forbid deprecated addresses, we accept 1239 * the SYN packet. RFC2462 does not suggest dropping 1240 * SYN in this case. 1241 * If we decipher RFC2462 5.5.4, it says like this: 1242 * 1. use of deprecated addr with existing 1243 * communication is okay - "SHOULD continue to be 1244 * used" 1245 * 2. use of it with new communication: 1246 * (2a) "SHOULD NOT be used if alternate address 1247 * with sufficient scope is available" 1248 * (2b) nothing mentioned otherwise. 1249 * Here we fall into (2b) case as we have no choice in 1250 * our source address selection - we must obey the peer. 1251 * 1252 * The wording in RFC2462 is confusing, and there are 1253 * multiple description text for deprecated address 1254 * handling - worse, they are not exactly the same. 1255 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1256 */ 1257 if (isipv6 && !V_ip6_use_deprecated) { 1258 struct in6_ifaddr *ia6; 1259 1260 ia6 = ip6_getdstifaddr(m); 1261 if (ia6 != NULL && 1262 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1263 ifa_free(&ia6->ia_ifa); 1264 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1265 log(LOG_DEBUG, "%s; %s: Listen socket: " 1266 "Connection attempt to deprecated " 1267 "IPv6 address rejected\n", 1268 s, __func__); 1269 rstreason = BANDLIM_RST_OPENPORT; 1270 goto dropwithreset; 1271 } 1272 if (ia6) 1273 ifa_free(&ia6->ia_ifa); 1274 } 1275 #endif /* INET6 */ 1276 /* 1277 * Basic sanity checks on incoming SYN requests: 1278 * Don't respond if the destination is a link layer 1279 * broadcast according to RFC1122 4.2.3.10, p. 104. 1280 * If it is from this socket it must be forged. 1281 * Don't respond if the source or destination is a 1282 * global or subnet broad- or multicast address. 1283 * Note that it is quite possible to receive unicast 1284 * link-layer packets with a broadcast IP address. Use 1285 * in_broadcast() to find them. 1286 */ 1287 if (m->m_flags & (M_BCAST|M_MCAST)) { 1288 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1289 log(LOG_DEBUG, "%s; %s: Listen socket: " 1290 "Connection attempt from broad- or multicast " 1291 "link layer address ignored\n", s, __func__); 1292 goto dropunlock; 1293 } 1294 #ifdef INET6 1295 if (isipv6) { 1296 if (th->th_dport == th->th_sport && 1297 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1299 log(LOG_DEBUG, "%s; %s: Listen socket: " 1300 "Connection attempt to/from self " 1301 "ignored\n", s, __func__); 1302 goto dropunlock; 1303 } 1304 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1305 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1306 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1307 log(LOG_DEBUG, "%s; %s: Listen socket: " 1308 "Connection attempt from/to multicast " 1309 "address ignored\n", s, __func__); 1310 goto dropunlock; 1311 } 1312 } 1313 #endif 1314 #if defined(INET) && defined(INET6) 1315 else 1316 #endif 1317 #ifdef INET 1318 { 1319 if (th->th_dport == th->th_sport && 1320 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1321 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1322 log(LOG_DEBUG, "%s; %s: Listen socket: " 1323 "Connection attempt from/to self " 1324 "ignored\n", s, __func__); 1325 goto dropunlock; 1326 } 1327 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1328 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1329 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1330 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1331 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1332 log(LOG_DEBUG, "%s; %s: Listen socket: " 1333 "Connection attempt from/to broad- " 1334 "or multicast address ignored\n", 1335 s, __func__); 1336 goto dropunlock; 1337 } 1338 } 1339 #endif 1340 /* 1341 * SYN appears to be valid. Create compressed TCP state 1342 * for syncache. 1343 */ 1344 #ifdef TCPDEBUG 1345 if (so->so_options & SO_DEBUG) 1346 tcp_trace(TA_INPUT, ostate, tp, 1347 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1348 #endif 1349 tcp_dooptions(&to, optp, optlen, TO_SYN); 1350 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1351 /* 1352 * Entry added to syncache and mbuf consumed. 1353 * Everything already unlocked by syncache_add(). 1354 */ 1355 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1356 return; 1357 } else if (tp->t_state == TCPS_LISTEN) { 1358 /* 1359 * When a listen socket is torn down the SO_ACCEPTCONN 1360 * flag is removed first while connections are drained 1361 * from the accept queue in a unlock/lock cycle of the 1362 * ACCEPT_LOCK, opening a race condition allowing a SYN 1363 * attempt go through unhandled. 1364 */ 1365 goto dropunlock; 1366 } 1367 1368 #ifdef TCP_SIGNATURE 1369 if (sig_checked == 0) { 1370 tcp_dooptions(&to, optp, optlen, 1371 (thflags & TH_SYN) ? TO_SYN : 0); 1372 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1373 th, tp->t_flags)) { 1374 1375 /* 1376 * In SYN_SENT state if it receives an RST, it is 1377 * allowed for further processing. 1378 */ 1379 if ((thflags & TH_RST) == 0 || 1380 (tp->t_state == TCPS_SYN_SENT) == 0) 1381 goto dropunlock; 1382 } 1383 sig_checked = 1; 1384 } 1385 #endif 1386 1387 /* 1388 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1389 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1390 * the inpcb, and unlocks pcbinfo. 1391 */ 1392 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1393 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1394 return; 1395 1396 dropwithreset: 1397 if (ti_locked == TI_WLOCKED) { 1398 INP_INFO_WUNLOCK(&V_tcbinfo); 1399 ti_locked = TI_UNLOCKED; 1400 } 1401 #ifdef INVARIANTS 1402 else { 1403 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1404 "ti_locked: %d", __func__, ti_locked)); 1405 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1406 } 1407 #endif 1408 1409 if (inp != NULL) { 1410 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1411 INP_WUNLOCK(inp); 1412 } else 1413 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1414 m = NULL; /* mbuf chain got consumed. */ 1415 goto drop; 1416 1417 dropunlock: 1418 if (ti_locked == TI_WLOCKED) { 1419 INP_INFO_WUNLOCK(&V_tcbinfo); 1420 ti_locked = TI_UNLOCKED; 1421 } 1422 #ifdef INVARIANTS 1423 else { 1424 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1425 "ti_locked: %d", __func__, ti_locked)); 1426 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1427 } 1428 #endif 1429 1430 if (inp != NULL) 1431 INP_WUNLOCK(inp); 1432 1433 drop: 1434 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1435 if (s != NULL) 1436 free(s, M_TCPLOG); 1437 if (m != NULL) 1438 m_freem(m); 1439 } 1440 1441 static void 1442 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1443 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1444 int ti_locked) 1445 { 1446 int thflags, acked, ourfinisacked, needoutput = 0; 1447 int rstreason, todrop, win; 1448 u_long tiwin; 1449 struct tcpopt to; 1450 1451 #ifdef TCPDEBUG 1452 /* 1453 * The size of tcp_saveipgen must be the size of the max ip header, 1454 * now IPv6. 1455 */ 1456 u_char tcp_saveipgen[IP6_HDR_LEN]; 1457 struct tcphdr tcp_savetcp; 1458 short ostate = 0; 1459 #endif 1460 thflags = th->th_flags; 1461 tp->sackhint.last_sack_ack = 0; 1462 1463 /* 1464 * If this is either a state-changing packet or current state isn't 1465 * established, we require a write lock on tcbinfo. Otherwise, we 1466 * allow the tcbinfo to be in either alocked or unlocked, as the 1467 * caller may have unnecessarily acquired a write lock due to a race. 1468 */ 1469 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1470 tp->t_state != TCPS_ESTABLISHED) { 1471 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1472 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1473 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1474 } else { 1475 #ifdef INVARIANTS 1476 if (ti_locked == TI_WLOCKED) 1477 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1478 else { 1479 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1480 "ti_locked: %d", __func__, ti_locked)); 1481 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1482 } 1483 #endif 1484 } 1485 INP_WLOCK_ASSERT(tp->t_inpcb); 1486 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1487 __func__)); 1488 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1489 __func__)); 1490 1491 /* 1492 * Segment received on connection. 1493 * Reset idle time and keep-alive timer. 1494 * XXX: This should be done after segment 1495 * validation to ignore broken/spoofed segs. 1496 */ 1497 tp->t_rcvtime = ticks; 1498 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1499 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1500 1501 /* 1502 * Unscale the window into a 32-bit value. 1503 * For the SYN_SENT state the scale is zero. 1504 */ 1505 tiwin = th->th_win << tp->snd_scale; 1506 1507 /* 1508 * TCP ECN processing. 1509 */ 1510 if (tp->t_flags & TF_ECN_PERMIT) { 1511 if (thflags & TH_CWR) 1512 tp->t_flags &= ~TF_ECN_SND_ECE; 1513 switch (iptos & IPTOS_ECN_MASK) { 1514 case IPTOS_ECN_CE: 1515 tp->t_flags |= TF_ECN_SND_ECE; 1516 TCPSTAT_INC(tcps_ecn_ce); 1517 break; 1518 case IPTOS_ECN_ECT0: 1519 TCPSTAT_INC(tcps_ecn_ect0); 1520 break; 1521 case IPTOS_ECN_ECT1: 1522 TCPSTAT_INC(tcps_ecn_ect1); 1523 break; 1524 } 1525 /* Congestion experienced. */ 1526 if (thflags & TH_ECE) { 1527 cc_cong_signal(tp, th, CC_ECN); 1528 } 1529 } 1530 1531 /* 1532 * Parse options on any incoming segment. 1533 */ 1534 tcp_dooptions(&to, (u_char *)(th + 1), 1535 (th->th_off << 2) - sizeof(struct tcphdr), 1536 (thflags & TH_SYN) ? TO_SYN : 0); 1537 1538 /* 1539 * If echoed timestamp is later than the current time, 1540 * fall back to non RFC1323 RTT calculation. Normalize 1541 * timestamp if syncookies were used when this connection 1542 * was established. 1543 */ 1544 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1545 to.to_tsecr -= tp->ts_offset; 1546 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1547 to.to_tsecr = 0; 1548 } 1549 1550 /* 1551 * Process options only when we get SYN/ACK back. The SYN case 1552 * for incoming connections is handled in tcp_syncache. 1553 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1554 * or <SYN,ACK>) segment itself is never scaled. 1555 * XXX this is traditional behavior, may need to be cleaned up. 1556 */ 1557 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1558 if ((to.to_flags & TOF_SCALE) && 1559 (tp->t_flags & TF_REQ_SCALE)) { 1560 tp->t_flags |= TF_RCVD_SCALE; 1561 tp->snd_scale = to.to_wscale; 1562 } 1563 /* 1564 * Initial send window. It will be updated with 1565 * the next incoming segment to the scaled value. 1566 */ 1567 tp->snd_wnd = th->th_win; 1568 if (to.to_flags & TOF_TS) { 1569 tp->t_flags |= TF_RCVD_TSTMP; 1570 tp->ts_recent = to.to_tsval; 1571 tp->ts_recent_age = tcp_ts_getticks(); 1572 } 1573 if (to.to_flags & TOF_MSS) 1574 tcp_mss(tp, to.to_mss); 1575 if ((tp->t_flags & TF_SACK_PERMIT) && 1576 (to.to_flags & TOF_SACKPERM) == 0) 1577 tp->t_flags &= ~TF_SACK_PERMIT; 1578 } 1579 1580 /* 1581 * Header prediction: check for the two common cases 1582 * of a uni-directional data xfer. If the packet has 1583 * no control flags, is in-sequence, the window didn't 1584 * change and we're not retransmitting, it's a 1585 * candidate. If the length is zero and the ack moved 1586 * forward, we're the sender side of the xfer. Just 1587 * free the data acked & wake any higher level process 1588 * that was blocked waiting for space. If the length 1589 * is non-zero and the ack didn't move, we're the 1590 * receiver side. If we're getting packets in-order 1591 * (the reassembly queue is empty), add the data to 1592 * the socket buffer and note that we need a delayed ack. 1593 * Make sure that the hidden state-flags are also off. 1594 * Since we check for TCPS_ESTABLISHED first, it can only 1595 * be TH_NEEDSYN. 1596 */ 1597 if (tp->t_state == TCPS_ESTABLISHED && 1598 th->th_seq == tp->rcv_nxt && 1599 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1600 tp->snd_nxt == tp->snd_max && 1601 tiwin && tiwin == tp->snd_wnd && 1602 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1603 LIST_EMPTY(&tp->t_segq) && 1604 ((to.to_flags & TOF_TS) == 0 || 1605 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1606 1607 /* 1608 * If last ACK falls within this segment's sequence numbers, 1609 * record the timestamp. 1610 * NOTE that the test is modified according to the latest 1611 * proposal of the [email protected] list (Braden 1993/04/26). 1612 */ 1613 if ((to.to_flags & TOF_TS) != 0 && 1614 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1615 tp->ts_recent_age = tcp_ts_getticks(); 1616 tp->ts_recent = to.to_tsval; 1617 } 1618 1619 if (tlen == 0) { 1620 if (SEQ_GT(th->th_ack, tp->snd_una) && 1621 SEQ_LEQ(th->th_ack, tp->snd_max) && 1622 !IN_RECOVERY(tp->t_flags) && 1623 (to.to_flags & TOF_SACK) == 0 && 1624 TAILQ_EMPTY(&tp->snd_holes)) { 1625 /* 1626 * This is a pure ack for outstanding data. 1627 */ 1628 if (ti_locked == TI_WLOCKED) 1629 INP_INFO_WUNLOCK(&V_tcbinfo); 1630 ti_locked = TI_UNLOCKED; 1631 1632 TCPSTAT_INC(tcps_predack); 1633 1634 /* 1635 * "bad retransmit" recovery. 1636 */ 1637 if (tp->t_rxtshift == 1 && 1638 tp->t_flags & TF_PREVVALID && 1639 (int)(ticks - tp->t_badrxtwin) < 0) { 1640 cc_cong_signal(tp, th, CC_RTO_ERR); 1641 } 1642 1643 /* 1644 * Recalculate the transmit timer / rtt. 1645 * 1646 * Some boxes send broken timestamp replies 1647 * during the SYN+ACK phase, ignore 1648 * timestamps of 0 or we could calculate a 1649 * huge RTT and blow up the retransmit timer. 1650 */ 1651 if ((to.to_flags & TOF_TS) != 0 && 1652 to.to_tsecr) { 1653 u_int t; 1654 1655 t = tcp_ts_getticks() - to.to_tsecr; 1656 if (!tp->t_rttlow || tp->t_rttlow > t) 1657 tp->t_rttlow = t; 1658 tcp_xmit_timer(tp, 1659 TCP_TS_TO_TICKS(t) + 1); 1660 } else if (tp->t_rtttime && 1661 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1662 if (!tp->t_rttlow || 1663 tp->t_rttlow > ticks - tp->t_rtttime) 1664 tp->t_rttlow = ticks - tp->t_rtttime; 1665 tcp_xmit_timer(tp, 1666 ticks - tp->t_rtttime); 1667 } 1668 acked = BYTES_THIS_ACK(tp, th); 1669 1670 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1671 hhook_run_tcp_est_in(tp, th, &to); 1672 1673 TCPSTAT_INC(tcps_rcvackpack); 1674 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1675 sbdrop(&so->so_snd, acked); 1676 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1677 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1678 tp->snd_recover = th->th_ack - 1; 1679 1680 /* 1681 * Let the congestion control algorithm update 1682 * congestion control related information. This 1683 * typically means increasing the congestion 1684 * window. 1685 */ 1686 cc_ack_received(tp, th, CC_ACK); 1687 1688 tp->snd_una = th->th_ack; 1689 /* 1690 * Pull snd_wl2 up to prevent seq wrap relative 1691 * to th_ack. 1692 */ 1693 tp->snd_wl2 = th->th_ack; 1694 tp->t_dupacks = 0; 1695 m_freem(m); 1696 ND6_HINT(tp); /* Some progress has been made. */ 1697 1698 /* 1699 * If all outstanding data are acked, stop 1700 * retransmit timer, otherwise restart timer 1701 * using current (possibly backed-off) value. 1702 * If process is waiting for space, 1703 * wakeup/selwakeup/signal. If data 1704 * are ready to send, let tcp_output 1705 * decide between more output or persist. 1706 */ 1707 #ifdef TCPDEBUG 1708 if (so->so_options & SO_DEBUG) 1709 tcp_trace(TA_INPUT, ostate, tp, 1710 (void *)tcp_saveipgen, 1711 &tcp_savetcp, 0); 1712 #endif 1713 if (tp->snd_una == tp->snd_max) 1714 tcp_timer_activate(tp, TT_REXMT, 0); 1715 else if (!tcp_timer_active(tp, TT_PERSIST)) 1716 tcp_timer_activate(tp, TT_REXMT, 1717 tp->t_rxtcur); 1718 sowwakeup(so); 1719 if (so->so_snd.sb_cc) 1720 (void) tcp_output(tp); 1721 goto check_delack; 1722 } 1723 } else if (th->th_ack == tp->snd_una && 1724 tlen <= sbspace(&so->so_rcv)) { 1725 int newsize = 0; /* automatic sockbuf scaling */ 1726 1727 /* 1728 * This is a pure, in-sequence data packet with 1729 * nothing on the reassembly queue and we have enough 1730 * buffer space to take it. 1731 */ 1732 if (ti_locked == TI_WLOCKED) 1733 INP_INFO_WUNLOCK(&V_tcbinfo); 1734 ti_locked = TI_UNLOCKED; 1735 1736 /* Clean receiver SACK report if present */ 1737 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1738 tcp_clean_sackreport(tp); 1739 TCPSTAT_INC(tcps_preddat); 1740 tp->rcv_nxt += tlen; 1741 /* 1742 * Pull snd_wl1 up to prevent seq wrap relative to 1743 * th_seq. 1744 */ 1745 tp->snd_wl1 = th->th_seq; 1746 /* 1747 * Pull rcv_up up to prevent seq wrap relative to 1748 * rcv_nxt. 1749 */ 1750 tp->rcv_up = tp->rcv_nxt; 1751 TCPSTAT_INC(tcps_rcvpack); 1752 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1753 ND6_HINT(tp); /* Some progress has been made */ 1754 #ifdef TCPDEBUG 1755 if (so->so_options & SO_DEBUG) 1756 tcp_trace(TA_INPUT, ostate, tp, 1757 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1758 #endif 1759 /* 1760 * Automatic sizing of receive socket buffer. Often the send 1761 * buffer size is not optimally adjusted to the actual network 1762 * conditions at hand (delay bandwidth product). Setting the 1763 * buffer size too small limits throughput on links with high 1764 * bandwidth and high delay (eg. trans-continental/oceanic links). 1765 * 1766 * On the receive side the socket buffer memory is only rarely 1767 * used to any significant extent. This allows us to be much 1768 * more aggressive in scaling the receive socket buffer. For 1769 * the case that the buffer space is actually used to a large 1770 * extent and we run out of kernel memory we can simply drop 1771 * the new segments; TCP on the sender will just retransmit it 1772 * later. Setting the buffer size too big may only consume too 1773 * much kernel memory if the application doesn't read() from 1774 * the socket or packet loss or reordering makes use of the 1775 * reassembly queue. 1776 * 1777 * The criteria to step up the receive buffer one notch are: 1778 * 1. the number of bytes received during the time it takes 1779 * one timestamp to be reflected back to us (the RTT); 1780 * 2. received bytes per RTT is within seven eighth of the 1781 * current socket buffer size; 1782 * 3. receive buffer size has not hit maximal automatic size; 1783 * 1784 * This algorithm does one step per RTT at most and only if 1785 * we receive a bulk stream w/o packet losses or reorderings. 1786 * Shrinking the buffer during idle times is not necessary as 1787 * it doesn't consume any memory when idle. 1788 * 1789 * TODO: Only step up if the application is actually serving 1790 * the buffer to better manage the socket buffer resources. 1791 */ 1792 if (V_tcp_do_autorcvbuf && 1793 to.to_tsecr && 1794 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1795 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1796 to.to_tsecr - tp->rfbuf_ts < hz) { 1797 if (tp->rfbuf_cnt > 1798 (so->so_rcv.sb_hiwat / 8 * 7) && 1799 so->so_rcv.sb_hiwat < 1800 V_tcp_autorcvbuf_max) { 1801 newsize = 1802 min(so->so_rcv.sb_hiwat + 1803 V_tcp_autorcvbuf_inc, 1804 V_tcp_autorcvbuf_max); 1805 } 1806 /* Start over with next RTT. */ 1807 tp->rfbuf_ts = 0; 1808 tp->rfbuf_cnt = 0; 1809 } else 1810 tp->rfbuf_cnt += tlen; /* add up */ 1811 } 1812 1813 /* Add data to socket buffer. */ 1814 SOCKBUF_LOCK(&so->so_rcv); 1815 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1816 m_freem(m); 1817 } else { 1818 /* 1819 * Set new socket buffer size. 1820 * Give up when limit is reached. 1821 */ 1822 if (newsize) 1823 if (!sbreserve_locked(&so->so_rcv, 1824 newsize, so, NULL)) 1825 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1826 m_adj(m, drop_hdrlen); /* delayed header drop */ 1827 sbappendstream_locked(&so->so_rcv, m); 1828 } 1829 /* NB: sorwakeup_locked() does an implicit unlock. */ 1830 sorwakeup_locked(so); 1831 if (DELAY_ACK(tp)) { 1832 tp->t_flags |= TF_DELACK; 1833 } else { 1834 tp->t_flags |= TF_ACKNOW; 1835 tcp_output(tp); 1836 } 1837 goto check_delack; 1838 } 1839 } 1840 1841 /* 1842 * Calculate amount of space in receive window, 1843 * and then do TCP input processing. 1844 * Receive window is amount of space in rcv queue, 1845 * but not less than advertised window. 1846 */ 1847 win = sbspace(&so->so_rcv); 1848 if (win < 0) 1849 win = 0; 1850 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1851 1852 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1853 tp->rfbuf_ts = 0; 1854 tp->rfbuf_cnt = 0; 1855 1856 switch (tp->t_state) { 1857 1858 /* 1859 * If the state is SYN_RECEIVED: 1860 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1861 */ 1862 case TCPS_SYN_RECEIVED: 1863 if ((thflags & TH_ACK) && 1864 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1865 SEQ_GT(th->th_ack, tp->snd_max))) { 1866 rstreason = BANDLIM_RST_OPENPORT; 1867 goto dropwithreset; 1868 } 1869 break; 1870 1871 /* 1872 * If the state is SYN_SENT: 1873 * if seg contains an ACK, but not for our SYN, drop the input. 1874 * if seg contains a RST, then drop the connection. 1875 * if seg does not contain SYN, then drop it. 1876 * Otherwise this is an acceptable SYN segment 1877 * initialize tp->rcv_nxt and tp->irs 1878 * if seg contains ack then advance tp->snd_una 1879 * if seg contains an ECE and ECN support is enabled, the stream 1880 * is ECN capable. 1881 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1882 * arrange for segment to be acked (eventually) 1883 * continue processing rest of data/controls, beginning with URG 1884 */ 1885 case TCPS_SYN_SENT: 1886 if ((thflags & TH_ACK) && 1887 (SEQ_LEQ(th->th_ack, tp->iss) || 1888 SEQ_GT(th->th_ack, tp->snd_max))) { 1889 rstreason = BANDLIM_UNLIMITED; 1890 goto dropwithreset; 1891 } 1892 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1893 tp = tcp_drop(tp, ECONNREFUSED); 1894 if (thflags & TH_RST) 1895 goto drop; 1896 if (!(thflags & TH_SYN)) 1897 goto drop; 1898 1899 tp->irs = th->th_seq; 1900 tcp_rcvseqinit(tp); 1901 if (thflags & TH_ACK) { 1902 TCPSTAT_INC(tcps_connects); 1903 soisconnected(so); 1904 #ifdef MAC 1905 mac_socketpeer_set_from_mbuf(m, so); 1906 #endif 1907 /* Do window scaling on this connection? */ 1908 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1909 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1910 tp->rcv_scale = tp->request_r_scale; 1911 } 1912 tp->rcv_adv += imin(tp->rcv_wnd, 1913 TCP_MAXWIN << tp->rcv_scale); 1914 tp->snd_una++; /* SYN is acked */ 1915 /* 1916 * If there's data, delay ACK; if there's also a FIN 1917 * ACKNOW will be turned on later. 1918 */ 1919 if (DELAY_ACK(tp) && tlen != 0) 1920 tcp_timer_activate(tp, TT_DELACK, 1921 tcp_delacktime); 1922 else 1923 tp->t_flags |= TF_ACKNOW; 1924 1925 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1926 tp->t_flags |= TF_ECN_PERMIT; 1927 TCPSTAT_INC(tcps_ecn_shs); 1928 } 1929 1930 /* 1931 * Received <SYN,ACK> in SYN_SENT[*] state. 1932 * Transitions: 1933 * SYN_SENT --> ESTABLISHED 1934 * SYN_SENT* --> FIN_WAIT_1 1935 */ 1936 tp->t_starttime = ticks; 1937 if (tp->t_flags & TF_NEEDFIN) { 1938 tp->t_state = TCPS_FIN_WAIT_1; 1939 tp->t_flags &= ~TF_NEEDFIN; 1940 thflags &= ~TH_SYN; 1941 } else { 1942 tp->t_state = TCPS_ESTABLISHED; 1943 cc_conn_init(tp); 1944 tcp_timer_activate(tp, TT_KEEP, 1945 TP_KEEPIDLE(tp)); 1946 } 1947 } else { 1948 /* 1949 * Received initial SYN in SYN-SENT[*] state => 1950 * simultaneous open. If segment contains CC option 1951 * and there is a cached CC, apply TAO test. 1952 * If it succeeds, connection is * half-synchronized. 1953 * Otherwise, do 3-way handshake: 1954 * SYN-SENT -> SYN-RECEIVED 1955 * SYN-SENT* -> SYN-RECEIVED* 1956 * If there was no CC option, clear cached CC value. 1957 */ 1958 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1959 tcp_timer_activate(tp, TT_REXMT, 0); 1960 tp->t_state = TCPS_SYN_RECEIVED; 1961 } 1962 1963 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1964 "ti_locked %d", __func__, ti_locked)); 1965 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1966 INP_WLOCK_ASSERT(tp->t_inpcb); 1967 1968 /* 1969 * Advance th->th_seq to correspond to first data byte. 1970 * If data, trim to stay within window, 1971 * dropping FIN if necessary. 1972 */ 1973 th->th_seq++; 1974 if (tlen > tp->rcv_wnd) { 1975 todrop = tlen - tp->rcv_wnd; 1976 m_adj(m, -todrop); 1977 tlen = tp->rcv_wnd; 1978 thflags &= ~TH_FIN; 1979 TCPSTAT_INC(tcps_rcvpackafterwin); 1980 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1981 } 1982 tp->snd_wl1 = th->th_seq - 1; 1983 tp->rcv_up = th->th_seq; 1984 /* 1985 * Client side of transaction: already sent SYN and data. 1986 * If the remote host used T/TCP to validate the SYN, 1987 * our data will be ACK'd; if so, enter normal data segment 1988 * processing in the middle of step 5, ack processing. 1989 * Otherwise, goto step 6. 1990 */ 1991 if (thflags & TH_ACK) 1992 goto process_ACK; 1993 1994 goto step6; 1995 1996 /* 1997 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1998 * do normal processing. 1999 * 2000 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2001 */ 2002 case TCPS_LAST_ACK: 2003 case TCPS_CLOSING: 2004 break; /* continue normal processing */ 2005 } 2006 2007 /* 2008 * States other than LISTEN or SYN_SENT. 2009 * First check the RST flag and sequence number since reset segments 2010 * are exempt from the timestamp and connection count tests. This 2011 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2012 * below which allowed reset segments in half the sequence space 2013 * to fall though and be processed (which gives forged reset 2014 * segments with a random sequence number a 50 percent chance of 2015 * killing a connection). 2016 * Then check timestamp, if present. 2017 * Then check the connection count, if present. 2018 * Then check that at least some bytes of segment are within 2019 * receive window. If segment begins before rcv_nxt, 2020 * drop leading data (and SYN); if nothing left, just ack. 2021 * 2022 * 2023 * If the RST bit is set, check the sequence number to see 2024 * if this is a valid reset segment. 2025 * RFC 793 page 37: 2026 * In all states except SYN-SENT, all reset (RST) segments 2027 * are validated by checking their SEQ-fields. A reset is 2028 * valid if its sequence number is in the window. 2029 * Note: this does not take into account delayed ACKs, so 2030 * we should test against last_ack_sent instead of rcv_nxt. 2031 * The sequence number in the reset segment is normally an 2032 * echo of our outgoing acknowlegement numbers, but some hosts 2033 * send a reset with the sequence number at the rightmost edge 2034 * of our receive window, and we have to handle this case. 2035 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2036 * that brute force RST attacks are possible. To combat this, 2037 * we use a much stricter check while in the ESTABLISHED state, 2038 * only accepting RSTs where the sequence number is equal to 2039 * last_ack_sent. In all other states (the states in which a 2040 * RST is more likely), the more permissive check is used. 2041 * If we have multiple segments in flight, the initial reset 2042 * segment sequence numbers will be to the left of last_ack_sent, 2043 * but they will eventually catch up. 2044 * In any case, it never made sense to trim reset segments to 2045 * fit the receive window since RFC 1122 says: 2046 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2047 * 2048 * A TCP SHOULD allow a received RST segment to include data. 2049 * 2050 * DISCUSSION 2051 * It has been suggested that a RST segment could contain 2052 * ASCII text that encoded and explained the cause of the 2053 * RST. No standard has yet been established for such 2054 * data. 2055 * 2056 * If the reset segment passes the sequence number test examine 2057 * the state: 2058 * SYN_RECEIVED STATE: 2059 * If passive open, return to LISTEN state. 2060 * If active open, inform user that connection was refused. 2061 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2062 * Inform user that connection was reset, and close tcb. 2063 * CLOSING, LAST_ACK STATES: 2064 * Close the tcb. 2065 * TIME_WAIT STATE: 2066 * Drop the segment - see Stevens, vol. 2, p. 964 and 2067 * RFC 1337. 2068 */ 2069 if (thflags & TH_RST) { 2070 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2071 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2072 switch (tp->t_state) { 2073 2074 case TCPS_SYN_RECEIVED: 2075 so->so_error = ECONNREFUSED; 2076 goto close; 2077 2078 case TCPS_ESTABLISHED: 2079 if (V_tcp_insecure_rst == 0 && 2080 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2081 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2082 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2083 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2084 TCPSTAT_INC(tcps_badrst); 2085 goto drop; 2086 } 2087 /* FALLTHROUGH */ 2088 case TCPS_FIN_WAIT_1: 2089 case TCPS_FIN_WAIT_2: 2090 case TCPS_CLOSE_WAIT: 2091 so->so_error = ECONNRESET; 2092 close: 2093 KASSERT(ti_locked == TI_WLOCKED, 2094 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2095 ti_locked)); 2096 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2097 2098 tp->t_state = TCPS_CLOSED; 2099 TCPSTAT_INC(tcps_drops); 2100 tp = tcp_close(tp); 2101 break; 2102 2103 case TCPS_CLOSING: 2104 case TCPS_LAST_ACK: 2105 KASSERT(ti_locked == TI_WLOCKED, 2106 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2107 ti_locked)); 2108 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2109 2110 tp = tcp_close(tp); 2111 break; 2112 } 2113 } 2114 goto drop; 2115 } 2116 2117 /* 2118 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2119 * and it's less than ts_recent, drop it. 2120 */ 2121 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2122 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2123 2124 /* Check to see if ts_recent is over 24 days old. */ 2125 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2126 /* 2127 * Invalidate ts_recent. If this segment updates 2128 * ts_recent, the age will be reset later and ts_recent 2129 * will get a valid value. If it does not, setting 2130 * ts_recent to zero will at least satisfy the 2131 * requirement that zero be placed in the timestamp 2132 * echo reply when ts_recent isn't valid. The 2133 * age isn't reset until we get a valid ts_recent 2134 * because we don't want out-of-order segments to be 2135 * dropped when ts_recent is old. 2136 */ 2137 tp->ts_recent = 0; 2138 } else { 2139 TCPSTAT_INC(tcps_rcvduppack); 2140 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2141 TCPSTAT_INC(tcps_pawsdrop); 2142 if (tlen) 2143 goto dropafterack; 2144 goto drop; 2145 } 2146 } 2147 2148 /* 2149 * In the SYN-RECEIVED state, validate that the packet belongs to 2150 * this connection before trimming the data to fit the receive 2151 * window. Check the sequence number versus IRS since we know 2152 * the sequence numbers haven't wrapped. This is a partial fix 2153 * for the "LAND" DoS attack. 2154 */ 2155 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2156 rstreason = BANDLIM_RST_OPENPORT; 2157 goto dropwithreset; 2158 } 2159 2160 todrop = tp->rcv_nxt - th->th_seq; 2161 if (todrop > 0) { 2162 /* 2163 * If this is a duplicate SYN for our current connection, 2164 * advance over it and pretend and it's not a SYN. 2165 */ 2166 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2167 thflags &= ~TH_SYN; 2168 th->th_seq++; 2169 if (th->th_urp > 1) 2170 th->th_urp--; 2171 else 2172 thflags &= ~TH_URG; 2173 todrop--; 2174 } 2175 /* 2176 * Following if statement from Stevens, vol. 2, p. 960. 2177 */ 2178 if (todrop > tlen 2179 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2180 /* 2181 * Any valid FIN must be to the left of the window. 2182 * At this point the FIN must be a duplicate or out 2183 * of sequence; drop it. 2184 */ 2185 thflags &= ~TH_FIN; 2186 2187 /* 2188 * Send an ACK to resynchronize and drop any data. 2189 * But keep on processing for RST or ACK. 2190 */ 2191 tp->t_flags |= TF_ACKNOW; 2192 todrop = tlen; 2193 TCPSTAT_INC(tcps_rcvduppack); 2194 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2195 } else { 2196 TCPSTAT_INC(tcps_rcvpartduppack); 2197 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2198 } 2199 drop_hdrlen += todrop; /* drop from the top afterwards */ 2200 th->th_seq += todrop; 2201 tlen -= todrop; 2202 if (th->th_urp > todrop) 2203 th->th_urp -= todrop; 2204 else { 2205 thflags &= ~TH_URG; 2206 th->th_urp = 0; 2207 } 2208 } 2209 2210 /* 2211 * If new data are received on a connection after the 2212 * user processes are gone, then RST the other end. 2213 */ 2214 if ((so->so_state & SS_NOFDREF) && 2215 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2216 char *s; 2217 2218 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2219 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2220 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2221 2222 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 2223 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 2224 "was closed, sending RST and removing tcpcb\n", 2225 s, __func__, tcpstates[tp->t_state], tlen); 2226 free(s, M_TCPLOG); 2227 } 2228 tp = tcp_close(tp); 2229 TCPSTAT_INC(tcps_rcvafterclose); 2230 rstreason = BANDLIM_UNLIMITED; 2231 goto dropwithreset; 2232 } 2233 2234 /* 2235 * If segment ends after window, drop trailing data 2236 * (and PUSH and FIN); if nothing left, just ACK. 2237 */ 2238 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2239 if (todrop > 0) { 2240 TCPSTAT_INC(tcps_rcvpackafterwin); 2241 if (todrop >= tlen) { 2242 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2243 /* 2244 * If window is closed can only take segments at 2245 * window edge, and have to drop data and PUSH from 2246 * incoming segments. Continue processing, but 2247 * remember to ack. Otherwise, drop segment 2248 * and ack. 2249 */ 2250 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2251 tp->t_flags |= TF_ACKNOW; 2252 TCPSTAT_INC(tcps_rcvwinprobe); 2253 } else 2254 goto dropafterack; 2255 } else 2256 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2257 m_adj(m, -todrop); 2258 tlen -= todrop; 2259 thflags &= ~(TH_PUSH|TH_FIN); 2260 } 2261 2262 /* 2263 * If last ACK falls within this segment's sequence numbers, 2264 * record its timestamp. 2265 * NOTE: 2266 * 1) That the test incorporates suggestions from the latest 2267 * proposal of the [email protected] list (Braden 1993/04/26). 2268 * 2) That updating only on newer timestamps interferes with 2269 * our earlier PAWS tests, so this check should be solely 2270 * predicated on the sequence space of this segment. 2271 * 3) That we modify the segment boundary check to be 2272 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2273 * instead of RFC1323's 2274 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2275 * This modified check allows us to overcome RFC1323's 2276 * limitations as described in Stevens TCP/IP Illustrated 2277 * Vol. 2 p.869. In such cases, we can still calculate the 2278 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2279 */ 2280 if ((to.to_flags & TOF_TS) != 0 && 2281 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2282 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2283 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2284 tp->ts_recent_age = tcp_ts_getticks(); 2285 tp->ts_recent = to.to_tsval; 2286 } 2287 2288 /* 2289 * If a SYN is in the window, then this is an 2290 * error and we send an RST and drop the connection. 2291 */ 2292 if (thflags & TH_SYN) { 2293 KASSERT(ti_locked == TI_WLOCKED, 2294 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2295 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2296 2297 tp = tcp_drop(tp, ECONNRESET); 2298 rstreason = BANDLIM_UNLIMITED; 2299 goto drop; 2300 } 2301 2302 /* 2303 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2304 * flag is on (half-synchronized state), then queue data for 2305 * later processing; else drop segment and return. 2306 */ 2307 if ((thflags & TH_ACK) == 0) { 2308 if (tp->t_state == TCPS_SYN_RECEIVED || 2309 (tp->t_flags & TF_NEEDSYN)) 2310 goto step6; 2311 else if (tp->t_flags & TF_ACKNOW) 2312 goto dropafterack; 2313 else 2314 goto drop; 2315 } 2316 2317 /* 2318 * Ack processing. 2319 */ 2320 switch (tp->t_state) { 2321 2322 /* 2323 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2324 * ESTABLISHED state and continue processing. 2325 * The ACK was checked above. 2326 */ 2327 case TCPS_SYN_RECEIVED: 2328 2329 TCPSTAT_INC(tcps_connects); 2330 soisconnected(so); 2331 /* Do window scaling? */ 2332 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2333 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2334 tp->rcv_scale = tp->request_r_scale; 2335 tp->snd_wnd = tiwin; 2336 } 2337 /* 2338 * Make transitions: 2339 * SYN-RECEIVED -> ESTABLISHED 2340 * SYN-RECEIVED* -> FIN-WAIT-1 2341 */ 2342 tp->t_starttime = ticks; 2343 if (tp->t_flags & TF_NEEDFIN) { 2344 tp->t_state = TCPS_FIN_WAIT_1; 2345 tp->t_flags &= ~TF_NEEDFIN; 2346 } else { 2347 tp->t_state = TCPS_ESTABLISHED; 2348 cc_conn_init(tp); 2349 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2350 } 2351 /* 2352 * If segment contains data or ACK, will call tcp_reass() 2353 * later; if not, do so now to pass queued data to user. 2354 */ 2355 if (tlen == 0 && (thflags & TH_FIN) == 0) 2356 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2357 (struct mbuf *)0); 2358 tp->snd_wl1 = th->th_seq - 1; 2359 /* FALLTHROUGH */ 2360 2361 /* 2362 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2363 * ACKs. If the ack is in the range 2364 * tp->snd_una < th->th_ack <= tp->snd_max 2365 * then advance tp->snd_una to th->th_ack and drop 2366 * data from the retransmission queue. If this ACK reflects 2367 * more up to date window information we update our window information. 2368 */ 2369 case TCPS_ESTABLISHED: 2370 case TCPS_FIN_WAIT_1: 2371 case TCPS_FIN_WAIT_2: 2372 case TCPS_CLOSE_WAIT: 2373 case TCPS_CLOSING: 2374 case TCPS_LAST_ACK: 2375 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2376 TCPSTAT_INC(tcps_rcvacktoomuch); 2377 goto dropafterack; 2378 } 2379 if ((tp->t_flags & TF_SACK_PERMIT) && 2380 ((to.to_flags & TOF_SACK) || 2381 !TAILQ_EMPTY(&tp->snd_holes))) 2382 tcp_sack_doack(tp, &to, th->th_ack); 2383 2384 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2385 hhook_run_tcp_est_in(tp, th, &to); 2386 2387 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2388 if (tlen == 0 && tiwin == tp->snd_wnd) { 2389 TCPSTAT_INC(tcps_rcvdupack); 2390 /* 2391 * If we have outstanding data (other than 2392 * a window probe), this is a completely 2393 * duplicate ack (ie, window info didn't 2394 * change), the ack is the biggest we've 2395 * seen and we've seen exactly our rexmt 2396 * threshhold of them, assume a packet 2397 * has been dropped and retransmit it. 2398 * Kludge snd_nxt & the congestion 2399 * window so we send only this one 2400 * packet. 2401 * 2402 * We know we're losing at the current 2403 * window size so do congestion avoidance 2404 * (set ssthresh to half the current window 2405 * and pull our congestion window back to 2406 * the new ssthresh). 2407 * 2408 * Dup acks mean that packets have left the 2409 * network (they're now cached at the receiver) 2410 * so bump cwnd by the amount in the receiver 2411 * to keep a constant cwnd packets in the 2412 * network. 2413 * 2414 * When using TCP ECN, notify the peer that 2415 * we reduced the cwnd. 2416 */ 2417 if (!tcp_timer_active(tp, TT_REXMT) || 2418 th->th_ack != tp->snd_una) 2419 tp->t_dupacks = 0; 2420 else if (++tp->t_dupacks > tcprexmtthresh || 2421 IN_FASTRECOVERY(tp->t_flags)) { 2422 cc_ack_received(tp, th, CC_DUPACK); 2423 if ((tp->t_flags & TF_SACK_PERMIT) && 2424 IN_FASTRECOVERY(tp->t_flags)) { 2425 int awnd; 2426 2427 /* 2428 * Compute the amount of data in flight first. 2429 * We can inject new data into the pipe iff 2430 * we have less than 1/2 the original window's 2431 * worth of data in flight. 2432 */ 2433 awnd = (tp->snd_nxt - tp->snd_fack) + 2434 tp->sackhint.sack_bytes_rexmit; 2435 if (awnd < tp->snd_ssthresh) { 2436 tp->snd_cwnd += tp->t_maxseg; 2437 if (tp->snd_cwnd > tp->snd_ssthresh) 2438 tp->snd_cwnd = tp->snd_ssthresh; 2439 } 2440 } else 2441 tp->snd_cwnd += tp->t_maxseg; 2442 if ((thflags & TH_FIN) && 2443 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2444 /* 2445 * If its a fin we need to process 2446 * it to avoid a race where both 2447 * sides enter FIN-WAIT and send FIN|ACK 2448 * at the same time. 2449 */ 2450 break; 2451 } 2452 (void) tcp_output(tp); 2453 goto drop; 2454 } else if (tp->t_dupacks == tcprexmtthresh) { 2455 tcp_seq onxt = tp->snd_nxt; 2456 2457 /* 2458 * If we're doing sack, check to 2459 * see if we're already in sack 2460 * recovery. If we're not doing sack, 2461 * check to see if we're in newreno 2462 * recovery. 2463 */ 2464 if (tp->t_flags & TF_SACK_PERMIT) { 2465 if (IN_FASTRECOVERY(tp->t_flags)) { 2466 tp->t_dupacks = 0; 2467 break; 2468 } 2469 } else { 2470 if (SEQ_LEQ(th->th_ack, 2471 tp->snd_recover)) { 2472 tp->t_dupacks = 0; 2473 break; 2474 } 2475 } 2476 /* Congestion signal before ack. */ 2477 cc_cong_signal(tp, th, CC_NDUPACK); 2478 cc_ack_received(tp, th, CC_DUPACK); 2479 tcp_timer_activate(tp, TT_REXMT, 0); 2480 tp->t_rtttime = 0; 2481 if (tp->t_flags & TF_SACK_PERMIT) { 2482 TCPSTAT_INC( 2483 tcps_sack_recovery_episode); 2484 tp->sack_newdata = tp->snd_nxt; 2485 tp->snd_cwnd = tp->t_maxseg; 2486 (void) tcp_output(tp); 2487 goto drop; 2488 } 2489 tp->snd_nxt = th->th_ack; 2490 tp->snd_cwnd = tp->t_maxseg; 2491 if ((thflags & TH_FIN) && 2492 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2493 /* 2494 * If its a fin we need to process 2495 * it to avoid a race where both 2496 * sides enter FIN-WAIT and send FIN|ACK 2497 * at the same time. 2498 */ 2499 break; 2500 } 2501 (void) tcp_output(tp); 2502 KASSERT(tp->snd_limited <= 2, 2503 ("%s: tp->snd_limited too big", 2504 __func__)); 2505 tp->snd_cwnd = tp->snd_ssthresh + 2506 tp->t_maxseg * 2507 (tp->t_dupacks - tp->snd_limited); 2508 if (SEQ_GT(onxt, tp->snd_nxt)) 2509 tp->snd_nxt = onxt; 2510 goto drop; 2511 } else if (V_tcp_do_rfc3042) { 2512 cc_ack_received(tp, th, CC_DUPACK); 2513 u_long oldcwnd = tp->snd_cwnd; 2514 tcp_seq oldsndmax = tp->snd_max; 2515 u_int sent; 2516 int avail; 2517 2518 KASSERT(tp->t_dupacks == 1 || 2519 tp->t_dupacks == 2, 2520 ("%s: dupacks not 1 or 2", 2521 __func__)); 2522 if (tp->t_dupacks == 1) 2523 tp->snd_limited = 0; 2524 tp->snd_cwnd = 2525 (tp->snd_nxt - tp->snd_una) + 2526 (tp->t_dupacks - tp->snd_limited) * 2527 tp->t_maxseg; 2528 if ((thflags & TH_FIN) && 2529 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2530 /* 2531 * If its a fin we need to process 2532 * it to avoid a race where both 2533 * sides enter FIN-WAIT and send FIN|ACK 2534 * at the same time. 2535 */ 2536 break; 2537 } 2538 /* 2539 * Only call tcp_output when there 2540 * is new data available to be sent. 2541 * Otherwise we would send pure ACKs. 2542 */ 2543 SOCKBUF_LOCK(&so->so_snd); 2544 avail = so->so_snd.sb_cc - 2545 (tp->snd_nxt - tp->snd_una); 2546 SOCKBUF_UNLOCK(&so->so_snd); 2547 if (avail > 0) 2548 (void) tcp_output(tp); 2549 sent = tp->snd_max - oldsndmax; 2550 if (sent > tp->t_maxseg) { 2551 KASSERT((tp->t_dupacks == 2 && 2552 tp->snd_limited == 0) || 2553 (sent == tp->t_maxseg + 1 && 2554 tp->t_flags & TF_SENTFIN), 2555 ("%s: sent too much", 2556 __func__)); 2557 tp->snd_limited = 2; 2558 } else if (sent > 0) 2559 ++tp->snd_limited; 2560 tp->snd_cwnd = oldcwnd; 2561 goto drop; 2562 } 2563 } else 2564 tp->t_dupacks = 0; 2565 break; 2566 } 2567 2568 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2569 ("%s: th_ack <= snd_una", __func__)); 2570 2571 /* 2572 * If the congestion window was inflated to account 2573 * for the other side's cached packets, retract it. 2574 */ 2575 if (IN_FASTRECOVERY(tp->t_flags)) { 2576 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2577 if (tp->t_flags & TF_SACK_PERMIT) 2578 tcp_sack_partialack(tp, th); 2579 else 2580 tcp_newreno_partial_ack(tp, th); 2581 } else 2582 cc_post_recovery(tp, th); 2583 } 2584 tp->t_dupacks = 0; 2585 /* 2586 * If we reach this point, ACK is not a duplicate, 2587 * i.e., it ACKs something we sent. 2588 */ 2589 if (tp->t_flags & TF_NEEDSYN) { 2590 /* 2591 * T/TCP: Connection was half-synchronized, and our 2592 * SYN has been ACK'd (so connection is now fully 2593 * synchronized). Go to non-starred state, 2594 * increment snd_una for ACK of SYN, and check if 2595 * we can do window scaling. 2596 */ 2597 tp->t_flags &= ~TF_NEEDSYN; 2598 tp->snd_una++; 2599 /* Do window scaling? */ 2600 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2601 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2602 tp->rcv_scale = tp->request_r_scale; 2603 /* Send window already scaled. */ 2604 } 2605 } 2606 2607 process_ACK: 2608 INP_WLOCK_ASSERT(tp->t_inpcb); 2609 2610 acked = BYTES_THIS_ACK(tp, th); 2611 TCPSTAT_INC(tcps_rcvackpack); 2612 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2613 2614 /* 2615 * If we just performed our first retransmit, and the ACK 2616 * arrives within our recovery window, then it was a mistake 2617 * to do the retransmit in the first place. Recover our 2618 * original cwnd and ssthresh, and proceed to transmit where 2619 * we left off. 2620 */ 2621 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2622 (int)(ticks - tp->t_badrxtwin) < 0) 2623 cc_cong_signal(tp, th, CC_RTO_ERR); 2624 2625 /* 2626 * If we have a timestamp reply, update smoothed 2627 * round trip time. If no timestamp is present but 2628 * transmit timer is running and timed sequence 2629 * number was acked, update smoothed round trip time. 2630 * Since we now have an rtt measurement, cancel the 2631 * timer backoff (cf., Phil Karn's retransmit alg.). 2632 * Recompute the initial retransmit timer. 2633 * 2634 * Some boxes send broken timestamp replies 2635 * during the SYN+ACK phase, ignore 2636 * timestamps of 0 or we could calculate a 2637 * huge RTT and blow up the retransmit timer. 2638 */ 2639 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2640 u_int t; 2641 2642 t = tcp_ts_getticks() - to.to_tsecr; 2643 if (!tp->t_rttlow || tp->t_rttlow > t) 2644 tp->t_rttlow = t; 2645 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2646 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2647 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2648 tp->t_rttlow = ticks - tp->t_rtttime; 2649 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2650 } 2651 2652 /* 2653 * If all outstanding data is acked, stop retransmit 2654 * timer and remember to restart (more output or persist). 2655 * If there is more data to be acked, restart retransmit 2656 * timer, using current (possibly backed-off) value. 2657 */ 2658 if (th->th_ack == tp->snd_max) { 2659 tcp_timer_activate(tp, TT_REXMT, 0); 2660 needoutput = 1; 2661 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2662 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2663 2664 /* 2665 * If no data (only SYN) was ACK'd, 2666 * skip rest of ACK processing. 2667 */ 2668 if (acked == 0) 2669 goto step6; 2670 2671 /* 2672 * Let the congestion control algorithm update congestion 2673 * control related information. This typically means increasing 2674 * the congestion window. 2675 */ 2676 cc_ack_received(tp, th, CC_ACK); 2677 2678 SOCKBUF_LOCK(&so->so_snd); 2679 if (acked > so->so_snd.sb_cc) { 2680 tp->snd_wnd -= so->so_snd.sb_cc; 2681 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2682 ourfinisacked = 1; 2683 } else { 2684 sbdrop_locked(&so->so_snd, acked); 2685 tp->snd_wnd -= acked; 2686 ourfinisacked = 0; 2687 } 2688 /* NB: sowwakeup_locked() does an implicit unlock. */ 2689 sowwakeup_locked(so); 2690 /* Detect una wraparound. */ 2691 if (!IN_RECOVERY(tp->t_flags) && 2692 SEQ_GT(tp->snd_una, tp->snd_recover) && 2693 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2694 tp->snd_recover = th->th_ack - 1; 2695 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2696 if (IN_RECOVERY(tp->t_flags) && 2697 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2698 EXIT_RECOVERY(tp->t_flags); 2699 } 2700 tp->snd_una = th->th_ack; 2701 if (tp->t_flags & TF_SACK_PERMIT) { 2702 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2703 tp->snd_recover = tp->snd_una; 2704 } 2705 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2706 tp->snd_nxt = tp->snd_una; 2707 2708 switch (tp->t_state) { 2709 2710 /* 2711 * In FIN_WAIT_1 STATE in addition to the processing 2712 * for the ESTABLISHED state if our FIN is now acknowledged 2713 * then enter FIN_WAIT_2. 2714 */ 2715 case TCPS_FIN_WAIT_1: 2716 if (ourfinisacked) { 2717 /* 2718 * If we can't receive any more 2719 * data, then closing user can proceed. 2720 * Starting the timer is contrary to the 2721 * specification, but if we don't get a FIN 2722 * we'll hang forever. 2723 * 2724 * XXXjl: 2725 * we should release the tp also, and use a 2726 * compressed state. 2727 */ 2728 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2729 soisdisconnected(so); 2730 tcp_timer_activate(tp, TT_2MSL, 2731 (tcp_fast_finwait2_recycle ? 2732 tcp_finwait2_timeout : 2733 TP_MAXIDLE(tp))); 2734 } 2735 tp->t_state = TCPS_FIN_WAIT_2; 2736 } 2737 break; 2738 2739 /* 2740 * In CLOSING STATE in addition to the processing for 2741 * the ESTABLISHED state if the ACK acknowledges our FIN 2742 * then enter the TIME-WAIT state, otherwise ignore 2743 * the segment. 2744 */ 2745 case TCPS_CLOSING: 2746 if (ourfinisacked) { 2747 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2748 tcp_twstart(tp); 2749 INP_INFO_WUNLOCK(&V_tcbinfo); 2750 m_freem(m); 2751 return; 2752 } 2753 break; 2754 2755 /* 2756 * In LAST_ACK, we may still be waiting for data to drain 2757 * and/or to be acked, as well as for the ack of our FIN. 2758 * If our FIN is now acknowledged, delete the TCB, 2759 * enter the closed state and return. 2760 */ 2761 case TCPS_LAST_ACK: 2762 if (ourfinisacked) { 2763 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2764 tp = tcp_close(tp); 2765 goto drop; 2766 } 2767 break; 2768 } 2769 } 2770 2771 step6: 2772 INP_WLOCK_ASSERT(tp->t_inpcb); 2773 2774 /* 2775 * Update window information. 2776 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2777 */ 2778 if ((thflags & TH_ACK) && 2779 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2780 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2781 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2782 /* keep track of pure window updates */ 2783 if (tlen == 0 && 2784 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2785 TCPSTAT_INC(tcps_rcvwinupd); 2786 tp->snd_wnd = tiwin; 2787 tp->snd_wl1 = th->th_seq; 2788 tp->snd_wl2 = th->th_ack; 2789 if (tp->snd_wnd > tp->max_sndwnd) 2790 tp->max_sndwnd = tp->snd_wnd; 2791 needoutput = 1; 2792 } 2793 2794 /* 2795 * Process segments with URG. 2796 */ 2797 if ((thflags & TH_URG) && th->th_urp && 2798 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2799 /* 2800 * This is a kludge, but if we receive and accept 2801 * random urgent pointers, we'll crash in 2802 * soreceive. It's hard to imagine someone 2803 * actually wanting to send this much urgent data. 2804 */ 2805 SOCKBUF_LOCK(&so->so_rcv); 2806 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2807 th->th_urp = 0; /* XXX */ 2808 thflags &= ~TH_URG; /* XXX */ 2809 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2810 goto dodata; /* XXX */ 2811 } 2812 /* 2813 * If this segment advances the known urgent pointer, 2814 * then mark the data stream. This should not happen 2815 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2816 * a FIN has been received from the remote side. 2817 * In these states we ignore the URG. 2818 * 2819 * According to RFC961 (Assigned Protocols), 2820 * the urgent pointer points to the last octet 2821 * of urgent data. We continue, however, 2822 * to consider it to indicate the first octet 2823 * of data past the urgent section as the original 2824 * spec states (in one of two places). 2825 */ 2826 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2827 tp->rcv_up = th->th_seq + th->th_urp; 2828 so->so_oobmark = so->so_rcv.sb_cc + 2829 (tp->rcv_up - tp->rcv_nxt) - 1; 2830 if (so->so_oobmark == 0) 2831 so->so_rcv.sb_state |= SBS_RCVATMARK; 2832 sohasoutofband(so); 2833 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2834 } 2835 SOCKBUF_UNLOCK(&so->so_rcv); 2836 /* 2837 * Remove out of band data so doesn't get presented to user. 2838 * This can happen independent of advancing the URG pointer, 2839 * but if two URG's are pending at once, some out-of-band 2840 * data may creep in... ick. 2841 */ 2842 if (th->th_urp <= (u_long)tlen && 2843 !(so->so_options & SO_OOBINLINE)) { 2844 /* hdr drop is delayed */ 2845 tcp_pulloutofband(so, th, m, drop_hdrlen); 2846 } 2847 } else { 2848 /* 2849 * If no out of band data is expected, 2850 * pull receive urgent pointer along 2851 * with the receive window. 2852 */ 2853 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2854 tp->rcv_up = tp->rcv_nxt; 2855 } 2856 dodata: /* XXX */ 2857 INP_WLOCK_ASSERT(tp->t_inpcb); 2858 2859 /* 2860 * Process the segment text, merging it into the TCP sequencing queue, 2861 * and arranging for acknowledgment of receipt if necessary. 2862 * This process logically involves adjusting tp->rcv_wnd as data 2863 * is presented to the user (this happens in tcp_usrreq.c, 2864 * case PRU_RCVD). If a FIN has already been received on this 2865 * connection then we just ignore the text. 2866 */ 2867 if ((tlen || (thflags & TH_FIN)) && 2868 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2869 tcp_seq save_start = th->th_seq; 2870 m_adj(m, drop_hdrlen); /* delayed header drop */ 2871 /* 2872 * Insert segment which includes th into TCP reassembly queue 2873 * with control block tp. Set thflags to whether reassembly now 2874 * includes a segment with FIN. This handles the common case 2875 * inline (segment is the next to be received on an established 2876 * connection, and the queue is empty), avoiding linkage into 2877 * and removal from the queue and repetition of various 2878 * conversions. 2879 * Set DELACK for segments received in order, but ack 2880 * immediately when segments are out of order (so 2881 * fast retransmit can work). 2882 */ 2883 if (th->th_seq == tp->rcv_nxt && 2884 LIST_EMPTY(&tp->t_segq) && 2885 TCPS_HAVEESTABLISHED(tp->t_state)) { 2886 if (DELAY_ACK(tp)) 2887 tp->t_flags |= TF_DELACK; 2888 else 2889 tp->t_flags |= TF_ACKNOW; 2890 tp->rcv_nxt += tlen; 2891 thflags = th->th_flags & TH_FIN; 2892 TCPSTAT_INC(tcps_rcvpack); 2893 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2894 ND6_HINT(tp); 2895 SOCKBUF_LOCK(&so->so_rcv); 2896 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2897 m_freem(m); 2898 else 2899 sbappendstream_locked(&so->so_rcv, m); 2900 /* NB: sorwakeup_locked() does an implicit unlock. */ 2901 sorwakeup_locked(so); 2902 } else { 2903 /* 2904 * XXX: Due to the header drop above "th" is 2905 * theoretically invalid by now. Fortunately 2906 * m_adj() doesn't actually frees any mbufs 2907 * when trimming from the head. 2908 */ 2909 thflags = tcp_reass(tp, th, &tlen, m); 2910 tp->t_flags |= TF_ACKNOW; 2911 } 2912 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2913 tcp_update_sack_list(tp, save_start, save_start + tlen); 2914 #if 0 2915 /* 2916 * Note the amount of data that peer has sent into 2917 * our window, in order to estimate the sender's 2918 * buffer size. 2919 * XXX: Unused. 2920 */ 2921 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2922 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2923 else 2924 len = so->so_rcv.sb_hiwat; 2925 #endif 2926 } else { 2927 m_freem(m); 2928 thflags &= ~TH_FIN; 2929 } 2930 2931 /* 2932 * If FIN is received ACK the FIN and let the user know 2933 * that the connection is closing. 2934 */ 2935 if (thflags & TH_FIN) { 2936 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2937 socantrcvmore(so); 2938 /* 2939 * If connection is half-synchronized 2940 * (ie NEEDSYN flag on) then delay ACK, 2941 * so it may be piggybacked when SYN is sent. 2942 * Otherwise, since we received a FIN then no 2943 * more input can be expected, send ACK now. 2944 */ 2945 if (tp->t_flags & TF_NEEDSYN) 2946 tp->t_flags |= TF_DELACK; 2947 else 2948 tp->t_flags |= TF_ACKNOW; 2949 tp->rcv_nxt++; 2950 } 2951 switch (tp->t_state) { 2952 2953 /* 2954 * In SYN_RECEIVED and ESTABLISHED STATES 2955 * enter the CLOSE_WAIT state. 2956 */ 2957 case TCPS_SYN_RECEIVED: 2958 tp->t_starttime = ticks; 2959 /* FALLTHROUGH */ 2960 case TCPS_ESTABLISHED: 2961 tp->t_state = TCPS_CLOSE_WAIT; 2962 break; 2963 2964 /* 2965 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2966 * enter the CLOSING state. 2967 */ 2968 case TCPS_FIN_WAIT_1: 2969 tp->t_state = TCPS_CLOSING; 2970 break; 2971 2972 /* 2973 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2974 * starting the time-wait timer, turning off the other 2975 * standard timers. 2976 */ 2977 case TCPS_FIN_WAIT_2: 2978 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2979 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2980 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2981 ti_locked)); 2982 2983 tcp_twstart(tp); 2984 INP_INFO_WUNLOCK(&V_tcbinfo); 2985 return; 2986 } 2987 } 2988 if (ti_locked == TI_WLOCKED) 2989 INP_INFO_WUNLOCK(&V_tcbinfo); 2990 ti_locked = TI_UNLOCKED; 2991 2992 #ifdef TCPDEBUG 2993 if (so->so_options & SO_DEBUG) 2994 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2995 &tcp_savetcp, 0); 2996 #endif 2997 2998 /* 2999 * Return any desired output. 3000 */ 3001 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3002 (void) tcp_output(tp); 3003 3004 check_delack: 3005 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3006 __func__, ti_locked)); 3007 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3008 INP_WLOCK_ASSERT(tp->t_inpcb); 3009 3010 if (tp->t_flags & TF_DELACK) { 3011 tp->t_flags &= ~TF_DELACK; 3012 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3013 } 3014 INP_WUNLOCK(tp->t_inpcb); 3015 return; 3016 3017 dropafterack: 3018 /* 3019 * Generate an ACK dropping incoming segment if it occupies 3020 * sequence space, where the ACK reflects our state. 3021 * 3022 * We can now skip the test for the RST flag since all 3023 * paths to this code happen after packets containing 3024 * RST have been dropped. 3025 * 3026 * In the SYN-RECEIVED state, don't send an ACK unless the 3027 * segment we received passes the SYN-RECEIVED ACK test. 3028 * If it fails send a RST. This breaks the loop in the 3029 * "LAND" DoS attack, and also prevents an ACK storm 3030 * between two listening ports that have been sent forged 3031 * SYN segments, each with the source address of the other. 3032 */ 3033 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3034 (SEQ_GT(tp->snd_una, th->th_ack) || 3035 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3036 rstreason = BANDLIM_RST_OPENPORT; 3037 goto dropwithreset; 3038 } 3039 #ifdef TCPDEBUG 3040 if (so->so_options & SO_DEBUG) 3041 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3042 &tcp_savetcp, 0); 3043 #endif 3044 if (ti_locked == TI_WLOCKED) 3045 INP_INFO_WUNLOCK(&V_tcbinfo); 3046 ti_locked = TI_UNLOCKED; 3047 3048 tp->t_flags |= TF_ACKNOW; 3049 (void) tcp_output(tp); 3050 INP_WUNLOCK(tp->t_inpcb); 3051 m_freem(m); 3052 return; 3053 3054 dropwithreset: 3055 if (ti_locked == TI_WLOCKED) 3056 INP_INFO_WUNLOCK(&V_tcbinfo); 3057 ti_locked = TI_UNLOCKED; 3058 3059 if (tp != NULL) { 3060 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3061 INP_WUNLOCK(tp->t_inpcb); 3062 } else 3063 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3064 return; 3065 3066 drop: 3067 if (ti_locked == TI_WLOCKED) { 3068 INP_INFO_WUNLOCK(&V_tcbinfo); 3069 ti_locked = TI_UNLOCKED; 3070 } 3071 #ifdef INVARIANTS 3072 else 3073 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3074 #endif 3075 3076 /* 3077 * Drop space held by incoming segment and return. 3078 */ 3079 #ifdef TCPDEBUG 3080 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3081 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3082 &tcp_savetcp, 0); 3083 #endif 3084 if (tp != NULL) 3085 INP_WUNLOCK(tp->t_inpcb); 3086 m_freem(m); 3087 } 3088 3089 /* 3090 * Issue RST and make ACK acceptable to originator of segment. 3091 * The mbuf must still include the original packet header. 3092 * tp may be NULL. 3093 */ 3094 static void 3095 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3096 int tlen, int rstreason) 3097 { 3098 #ifdef INET 3099 struct ip *ip; 3100 #endif 3101 #ifdef INET6 3102 struct ip6_hdr *ip6; 3103 #endif 3104 3105 if (tp != NULL) { 3106 INP_WLOCK_ASSERT(tp->t_inpcb); 3107 } 3108 3109 /* Don't bother if destination was broadcast/multicast. */ 3110 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3111 goto drop; 3112 #ifdef INET6 3113 if (mtod(m, struct ip *)->ip_v == 6) { 3114 ip6 = mtod(m, struct ip6_hdr *); 3115 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3116 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3117 goto drop; 3118 /* IPv6 anycast check is done at tcp6_input() */ 3119 } 3120 #endif 3121 #if defined(INET) && defined(INET6) 3122 else 3123 #endif 3124 #ifdef INET 3125 { 3126 ip = mtod(m, struct ip *); 3127 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3128 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3129 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3130 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3131 goto drop; 3132 } 3133 #endif 3134 3135 /* Perform bandwidth limiting. */ 3136 if (badport_bandlim(rstreason) < 0) 3137 goto drop; 3138 3139 /* tcp_respond consumes the mbuf chain. */ 3140 if (th->th_flags & TH_ACK) { 3141 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3142 th->th_ack, TH_RST); 3143 } else { 3144 if (th->th_flags & TH_SYN) 3145 tlen++; 3146 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3147 (tcp_seq)0, TH_RST|TH_ACK); 3148 } 3149 return; 3150 drop: 3151 m_freem(m); 3152 } 3153 3154 /* 3155 * Parse TCP options and place in tcpopt. 3156 */ 3157 static void 3158 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3159 { 3160 int opt, optlen; 3161 3162 to->to_flags = 0; 3163 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3164 opt = cp[0]; 3165 if (opt == TCPOPT_EOL) 3166 break; 3167 if (opt == TCPOPT_NOP) 3168 optlen = 1; 3169 else { 3170 if (cnt < 2) 3171 break; 3172 optlen = cp[1]; 3173 if (optlen < 2 || optlen > cnt) 3174 break; 3175 } 3176 switch (opt) { 3177 case TCPOPT_MAXSEG: 3178 if (optlen != TCPOLEN_MAXSEG) 3179 continue; 3180 if (!(flags & TO_SYN)) 3181 continue; 3182 to->to_flags |= TOF_MSS; 3183 bcopy((char *)cp + 2, 3184 (char *)&to->to_mss, sizeof(to->to_mss)); 3185 to->to_mss = ntohs(to->to_mss); 3186 break; 3187 case TCPOPT_WINDOW: 3188 if (optlen != TCPOLEN_WINDOW) 3189 continue; 3190 if (!(flags & TO_SYN)) 3191 continue; 3192 to->to_flags |= TOF_SCALE; 3193 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3194 break; 3195 case TCPOPT_TIMESTAMP: 3196 if (optlen != TCPOLEN_TIMESTAMP) 3197 continue; 3198 to->to_flags |= TOF_TS; 3199 bcopy((char *)cp + 2, 3200 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3201 to->to_tsval = ntohl(to->to_tsval); 3202 bcopy((char *)cp + 6, 3203 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3204 to->to_tsecr = ntohl(to->to_tsecr); 3205 break; 3206 #ifdef TCP_SIGNATURE 3207 /* 3208 * XXX In order to reply to a host which has set the 3209 * TCP_SIGNATURE option in its initial SYN, we have to 3210 * record the fact that the option was observed here 3211 * for the syncache code to perform the correct response. 3212 */ 3213 case TCPOPT_SIGNATURE: 3214 if (optlen != TCPOLEN_SIGNATURE) 3215 continue; 3216 to->to_flags |= TOF_SIGNATURE; 3217 to->to_signature = cp + 2; 3218 break; 3219 #endif 3220 case TCPOPT_SACK_PERMITTED: 3221 if (optlen != TCPOLEN_SACK_PERMITTED) 3222 continue; 3223 if (!(flags & TO_SYN)) 3224 continue; 3225 if (!V_tcp_do_sack) 3226 continue; 3227 to->to_flags |= TOF_SACKPERM; 3228 break; 3229 case TCPOPT_SACK: 3230 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3231 continue; 3232 if (flags & TO_SYN) 3233 continue; 3234 to->to_flags |= TOF_SACK; 3235 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3236 to->to_sacks = cp + 2; 3237 TCPSTAT_INC(tcps_sack_rcv_blocks); 3238 break; 3239 default: 3240 continue; 3241 } 3242 } 3243 } 3244 3245 /* 3246 * Pull out of band byte out of a segment so 3247 * it doesn't appear in the user's data queue. 3248 * It is still reflected in the segment length for 3249 * sequencing purposes. 3250 */ 3251 static void 3252 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3253 int off) 3254 { 3255 int cnt = off + th->th_urp - 1; 3256 3257 while (cnt >= 0) { 3258 if (m->m_len > cnt) { 3259 char *cp = mtod(m, caddr_t) + cnt; 3260 struct tcpcb *tp = sototcpcb(so); 3261 3262 INP_WLOCK_ASSERT(tp->t_inpcb); 3263 3264 tp->t_iobc = *cp; 3265 tp->t_oobflags |= TCPOOB_HAVEDATA; 3266 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3267 m->m_len--; 3268 if (m->m_flags & M_PKTHDR) 3269 m->m_pkthdr.len--; 3270 return; 3271 } 3272 cnt -= m->m_len; 3273 m = m->m_next; 3274 if (m == NULL) 3275 break; 3276 } 3277 panic("tcp_pulloutofband"); 3278 } 3279 3280 /* 3281 * Collect new round-trip time estimate 3282 * and update averages and current timeout. 3283 */ 3284 static void 3285 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3286 { 3287 int delta; 3288 3289 INP_WLOCK_ASSERT(tp->t_inpcb); 3290 3291 TCPSTAT_INC(tcps_rttupdated); 3292 tp->t_rttupdated++; 3293 if (tp->t_srtt != 0) { 3294 /* 3295 * srtt is stored as fixed point with 5 bits after the 3296 * binary point (i.e., scaled by 8). The following magic 3297 * is equivalent to the smoothing algorithm in rfc793 with 3298 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3299 * point). Adjust rtt to origin 0. 3300 */ 3301 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3302 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3303 3304 if ((tp->t_srtt += delta) <= 0) 3305 tp->t_srtt = 1; 3306 3307 /* 3308 * We accumulate a smoothed rtt variance (actually, a 3309 * smoothed mean difference), then set the retransmit 3310 * timer to smoothed rtt + 4 times the smoothed variance. 3311 * rttvar is stored as fixed point with 4 bits after the 3312 * binary point (scaled by 16). The following is 3313 * equivalent to rfc793 smoothing with an alpha of .75 3314 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3315 * rfc793's wired-in beta. 3316 */ 3317 if (delta < 0) 3318 delta = -delta; 3319 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3320 if ((tp->t_rttvar += delta) <= 0) 3321 tp->t_rttvar = 1; 3322 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3323 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3324 } else { 3325 /* 3326 * No rtt measurement yet - use the unsmoothed rtt. 3327 * Set the variance to half the rtt (so our first 3328 * retransmit happens at 3*rtt). 3329 */ 3330 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3331 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3332 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3333 } 3334 tp->t_rtttime = 0; 3335 tp->t_rxtshift = 0; 3336 3337 /* 3338 * the retransmit should happen at rtt + 4 * rttvar. 3339 * Because of the way we do the smoothing, srtt and rttvar 3340 * will each average +1/2 tick of bias. When we compute 3341 * the retransmit timer, we want 1/2 tick of rounding and 3342 * 1 extra tick because of +-1/2 tick uncertainty in the 3343 * firing of the timer. The bias will give us exactly the 3344 * 1.5 tick we need. But, because the bias is 3345 * statistical, we have to test that we don't drop below 3346 * the minimum feasible timer (which is 2 ticks). 3347 */ 3348 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3349 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3350 3351 /* 3352 * We received an ack for a packet that wasn't retransmitted; 3353 * it is probably safe to discard any error indications we've 3354 * received recently. This isn't quite right, but close enough 3355 * for now (a route might have failed after we sent a segment, 3356 * and the return path might not be symmetrical). 3357 */ 3358 tp->t_softerror = 0; 3359 } 3360 3361 /* 3362 * Determine a reasonable value for maxseg size. 3363 * If the route is known, check route for mtu. 3364 * If none, use an mss that can be handled on the outgoing interface 3365 * without forcing IP to fragment. If no route is found, route has no mtu, 3366 * or the destination isn't local, use a default, hopefully conservative 3367 * size (usually 512 or the default IP max size, but no more than the mtu 3368 * of the interface), as we can't discover anything about intervening 3369 * gateways or networks. We also initialize the congestion/slow start 3370 * window to be a single segment if the destination isn't local. 3371 * While looking at the routing entry, we also initialize other path-dependent 3372 * parameters from pre-set or cached values in the routing entry. 3373 * 3374 * Also take into account the space needed for options that we 3375 * send regularly. Make maxseg shorter by that amount to assure 3376 * that we can send maxseg amount of data even when the options 3377 * are present. Store the upper limit of the length of options plus 3378 * data in maxopd. 3379 * 3380 * NOTE that this routine is only called when we process an incoming 3381 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3382 * settings are handled in tcp_mssopt(). 3383 */ 3384 void 3385 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3386 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3387 { 3388 int mss = 0; 3389 u_long maxmtu = 0; 3390 struct inpcb *inp = tp->t_inpcb; 3391 struct hc_metrics_lite metrics; 3392 int origoffer; 3393 #ifdef INET6 3394 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3395 size_t min_protoh = isipv6 ? 3396 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3397 sizeof (struct tcpiphdr); 3398 #else 3399 const size_t min_protoh = sizeof(struct tcpiphdr); 3400 #endif 3401 3402 INP_WLOCK_ASSERT(tp->t_inpcb); 3403 3404 if (mtuoffer != -1) { 3405 KASSERT(offer == -1, ("%s: conflict", __func__)); 3406 offer = mtuoffer - min_protoh; 3407 } 3408 origoffer = offer; 3409 3410 /* Initialize. */ 3411 #ifdef INET6 3412 if (isipv6) { 3413 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3414 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3415 } 3416 #endif 3417 #if defined(INET) && defined(INET6) 3418 else 3419 #endif 3420 #ifdef INET 3421 { 3422 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3423 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3424 } 3425 #endif 3426 3427 /* 3428 * No route to sender, stay with default mss and return. 3429 */ 3430 if (maxmtu == 0) { 3431 /* 3432 * In case we return early we need to initialize metrics 3433 * to a defined state as tcp_hc_get() would do for us 3434 * if there was no cache hit. 3435 */ 3436 if (metricptr != NULL) 3437 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3438 return; 3439 } 3440 3441 /* What have we got? */ 3442 switch (offer) { 3443 case 0: 3444 /* 3445 * Offer == 0 means that there was no MSS on the SYN 3446 * segment, in this case we use tcp_mssdflt as 3447 * already assigned to t_maxopd above. 3448 */ 3449 offer = tp->t_maxopd; 3450 break; 3451 3452 case -1: 3453 /* 3454 * Offer == -1 means that we didn't receive SYN yet. 3455 */ 3456 /* FALLTHROUGH */ 3457 3458 default: 3459 /* 3460 * Prevent DoS attack with too small MSS. Round up 3461 * to at least minmss. 3462 */ 3463 offer = max(offer, V_tcp_minmss); 3464 } 3465 3466 /* 3467 * rmx information is now retrieved from tcp_hostcache. 3468 */ 3469 tcp_hc_get(&inp->inp_inc, &metrics); 3470 if (metricptr != NULL) 3471 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3472 3473 /* 3474 * If there's a discovered mtu int tcp hostcache, use it 3475 * else, use the link mtu. 3476 */ 3477 if (metrics.rmx_mtu) 3478 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3479 else { 3480 #ifdef INET6 3481 if (isipv6) { 3482 mss = maxmtu - min_protoh; 3483 if (!V_path_mtu_discovery && 3484 !in6_localaddr(&inp->in6p_faddr)) 3485 mss = min(mss, V_tcp_v6mssdflt); 3486 } 3487 #endif 3488 #if defined(INET) && defined(INET6) 3489 else 3490 #endif 3491 #ifdef INET 3492 { 3493 mss = maxmtu - min_protoh; 3494 if (!V_path_mtu_discovery && 3495 !in_localaddr(inp->inp_faddr)) 3496 mss = min(mss, V_tcp_mssdflt); 3497 } 3498 #endif 3499 /* 3500 * XXX - The above conditional (mss = maxmtu - min_protoh) 3501 * probably violates the TCP spec. 3502 * The problem is that, since we don't know the 3503 * other end's MSS, we are supposed to use a conservative 3504 * default. But, if we do that, then MTU discovery will 3505 * never actually take place, because the conservative 3506 * default is much less than the MTUs typically seen 3507 * on the Internet today. For the moment, we'll sweep 3508 * this under the carpet. 3509 * 3510 * The conservative default might not actually be a problem 3511 * if the only case this occurs is when sending an initial 3512 * SYN with options and data to a host we've never talked 3513 * to before. Then, they will reply with an MSS value which 3514 * will get recorded and the new parameters should get 3515 * recomputed. For Further Study. 3516 */ 3517 } 3518 mss = min(mss, offer); 3519 3520 /* 3521 * Sanity check: make sure that maxopd will be large 3522 * enough to allow some data on segments even if the 3523 * all the option space is used (40bytes). Otherwise 3524 * funny things may happen in tcp_output. 3525 */ 3526 mss = max(mss, 64); 3527 3528 /* 3529 * maxopd stores the maximum length of data AND options 3530 * in a segment; maxseg is the amount of data in a normal 3531 * segment. We need to store this value (maxopd) apart 3532 * from maxseg, because now every segment carries options 3533 * and thus we normally have somewhat less data in segments. 3534 */ 3535 tp->t_maxopd = mss; 3536 3537 /* 3538 * origoffer==-1 indicates that no segments were received yet. 3539 * In this case we just guess. 3540 */ 3541 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3542 (origoffer == -1 || 3543 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3544 mss -= TCPOLEN_TSTAMP_APPA; 3545 3546 tp->t_maxseg = mss; 3547 } 3548 3549 void 3550 tcp_mss(struct tcpcb *tp, int offer) 3551 { 3552 int mss; 3553 u_long bufsize; 3554 struct inpcb *inp; 3555 struct socket *so; 3556 struct hc_metrics_lite metrics; 3557 struct tcp_ifcap cap; 3558 3559 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3560 3561 bzero(&cap, sizeof(cap)); 3562 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3563 3564 mss = tp->t_maxseg; 3565 inp = tp->t_inpcb; 3566 3567 /* 3568 * If there's a pipesize, change the socket buffer to that size, 3569 * don't change if sb_hiwat is different than default (then it 3570 * has been changed on purpose with setsockopt). 3571 * Make the socket buffers an integral number of mss units; 3572 * if the mss is larger than the socket buffer, decrease the mss. 3573 */ 3574 so = inp->inp_socket; 3575 SOCKBUF_LOCK(&so->so_snd); 3576 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3577 bufsize = metrics.rmx_sendpipe; 3578 else 3579 bufsize = so->so_snd.sb_hiwat; 3580 if (bufsize < mss) 3581 mss = bufsize; 3582 else { 3583 bufsize = roundup(bufsize, mss); 3584 if (bufsize > sb_max) 3585 bufsize = sb_max; 3586 if (bufsize > so->so_snd.sb_hiwat) 3587 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3588 } 3589 SOCKBUF_UNLOCK(&so->so_snd); 3590 tp->t_maxseg = mss; 3591 3592 SOCKBUF_LOCK(&so->so_rcv); 3593 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3594 bufsize = metrics.rmx_recvpipe; 3595 else 3596 bufsize = so->so_rcv.sb_hiwat; 3597 if (bufsize > mss) { 3598 bufsize = roundup(bufsize, mss); 3599 if (bufsize > sb_max) 3600 bufsize = sb_max; 3601 if (bufsize > so->so_rcv.sb_hiwat) 3602 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3603 } 3604 SOCKBUF_UNLOCK(&so->so_rcv); 3605 3606 /* Check the interface for TSO capabilities. */ 3607 if (cap.ifcap & CSUM_TSO) { 3608 tp->t_flags |= TF_TSO; 3609 tp->t_tsomax = cap.tsomax; 3610 } 3611 } 3612 3613 /* 3614 * Determine the MSS option to send on an outgoing SYN. 3615 */ 3616 int 3617 tcp_mssopt(struct in_conninfo *inc) 3618 { 3619 int mss = 0; 3620 u_long maxmtu = 0; 3621 u_long thcmtu = 0; 3622 size_t min_protoh; 3623 3624 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3625 3626 #ifdef INET6 3627 if (inc->inc_flags & INC_ISIPV6) { 3628 mss = V_tcp_v6mssdflt; 3629 maxmtu = tcp_maxmtu6(inc, NULL); 3630 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3631 } 3632 #endif 3633 #if defined(INET) && defined(INET6) 3634 else 3635 #endif 3636 #ifdef INET 3637 { 3638 mss = V_tcp_mssdflt; 3639 maxmtu = tcp_maxmtu(inc, NULL); 3640 min_protoh = sizeof(struct tcpiphdr); 3641 } 3642 #endif 3643 #if defined(INET6) || defined(INET) 3644 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3645 #endif 3646 3647 if (maxmtu && thcmtu) 3648 mss = min(maxmtu, thcmtu) - min_protoh; 3649 else if (maxmtu || thcmtu) 3650 mss = max(maxmtu, thcmtu) - min_protoh; 3651 3652 return (mss); 3653 } 3654 3655 3656 /* 3657 * On a partial ack arrives, force the retransmission of the 3658 * next unacknowledged segment. Do not clear tp->t_dupacks. 3659 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3660 * be started again. 3661 */ 3662 static void 3663 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3664 { 3665 tcp_seq onxt = tp->snd_nxt; 3666 u_long ocwnd = tp->snd_cwnd; 3667 3668 INP_WLOCK_ASSERT(tp->t_inpcb); 3669 3670 tcp_timer_activate(tp, TT_REXMT, 0); 3671 tp->t_rtttime = 0; 3672 tp->snd_nxt = th->th_ack; 3673 /* 3674 * Set snd_cwnd to one segment beyond acknowledged offset. 3675 * (tp->snd_una has not yet been updated when this function is called.) 3676 */ 3677 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3678 tp->t_flags |= TF_ACKNOW; 3679 (void) tcp_output(tp); 3680 tp->snd_cwnd = ocwnd; 3681 if (SEQ_GT(onxt, tp->snd_nxt)) 3682 tp->snd_nxt = onxt; 3683 /* 3684 * Partial window deflation. Relies on fact that tp->snd_una 3685 * not updated yet. 3686 */ 3687 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3688 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3689 else 3690 tp->snd_cwnd = 0; 3691 tp->snd_cwnd += tp->t_maxseg; 3692 } 3693