1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * Copyright (c) 2007-2008,2010 5 * Swinburne University of Technology, Melbourne, Australia. 6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7 * Copyright (c) 2010 The FreeBSD Foundation 8 * Copyright (c) 2010-2011 Juniper Networks, Inc. 9 * All rights reserved. 10 * 11 * Portions of this software were developed at the Centre for Advanced Internet 12 * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13 * James Healy and David Hayes, made possible in part by a grant from the Cisco 14 * University Research Program Fund at Community Foundation Silicon Valley. 15 * 16 * Portions of this software were developed at the Centre for Advanced 17 * Internet Architectures, Swinburne University of Technology, Melbourne, 18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19 * 20 * Portions of this software were developed by Robert N. M. Watson under 21 * contract to Juniper Networks, Inc. 22 * 23 * Redistribution and use in source and binary forms, with or without 24 * modification, are permitted provided that the following conditions 25 * are met: 26 * 1. Redistributions of source code must retain the above copyright 27 * notice, this list of conditions and the following disclaimer. 28 * 2. Redistributions in binary form must reproduce the above copyright 29 * notice, this list of conditions and the following disclaimer in the 30 * documentation and/or other materials provided with the distribution. 31 * 4. Neither the name of the University nor the names of its contributors 32 * may be used to endorse or promote products derived from this software 33 * without specific prior written permission. 34 * 35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45 * SUCH DAMAGE. 46 * 47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48 */ 49 50 #include <sys/cdefs.h> 51 __FBSDID("$FreeBSD$"); 52 53 #include "opt_ipfw.h" /* for ipfw_fwd */ 54 #include "opt_inet.h" 55 #include "opt_inet6.h" 56 #include "opt_ipsec.h" 57 #include "opt_tcpdebug.h" 58 59 #include <sys/param.h> 60 #include <sys/kernel.h> 61 #include <sys/hhook.h> 62 #include <sys/malloc.h> 63 #include <sys/mbuf.h> 64 #include <sys/proc.h> /* for proc0 declaration */ 65 #include <sys/protosw.h> 66 #include <sys/sdt.h> 67 #include <sys/signalvar.h> 68 #include <sys/socket.h> 69 #include <sys/socketvar.h> 70 #include <sys/sysctl.h> 71 #include <sys/syslog.h> 72 #include <sys/systm.h> 73 74 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 75 76 #include <vm/uma.h> 77 78 #include <net/if.h> 79 #include <net/if_var.h> 80 #include <net/route.h> 81 #include <net/vnet.h> 82 83 #define TCPSTATES /* for logging */ 84 85 #include <netinet/cc.h> 86 #include <netinet/in.h> 87 #include <netinet/in_kdtrace.h> 88 #include <netinet/in_pcb.h> 89 #include <netinet/in_systm.h> 90 #include <netinet/in_var.h> 91 #include <netinet/ip.h> 92 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 93 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 94 #include <netinet/ip_var.h> 95 #include <netinet/ip_options.h> 96 #include <netinet/ip6.h> 97 #include <netinet/icmp6.h> 98 #include <netinet6/in6_pcb.h> 99 #include <netinet6/ip6_var.h> 100 #include <netinet6/nd6.h> 101 #include <netinet/tcp_fsm.h> 102 #include <netinet/tcp_seq.h> 103 #include <netinet/tcp_timer.h> 104 #include <netinet/tcp_var.h> 105 #include <netinet6/tcp6_var.h> 106 #include <netinet/tcpip.h> 107 #include <netinet/tcp_syncache.h> 108 #ifdef TCPDEBUG 109 #include <netinet/tcp_debug.h> 110 #endif /* TCPDEBUG */ 111 #ifdef TCP_OFFLOAD 112 #include <netinet/tcp_offload.h> 113 #endif 114 115 #ifdef IPSEC 116 #include <netipsec/ipsec.h> 117 #include <netipsec/ipsec6.h> 118 #endif /*IPSEC*/ 119 120 #include <machine/in_cksum.h> 121 122 #include <security/mac/mac_framework.h> 123 124 const int tcprexmtthresh = 3; 125 126 int tcp_log_in_vain = 0; 127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 128 &tcp_log_in_vain, 0, 129 "Log all incoming TCP segments to closed ports"); 130 131 VNET_DEFINE(int, blackhole) = 0; 132 #define V_blackhole VNET(blackhole) 133 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, 134 &VNET_NAME(blackhole), 0, 135 "Do not send RST on segments to closed ports"); 136 137 VNET_DEFINE(int, tcp_delack_enabled) = 1; 138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, 139 &VNET_NAME(tcp_delack_enabled), 0, 140 "Delay ACK to try and piggyback it onto a data packet"); 141 142 VNET_DEFINE(int, drop_synfin) = 0; 143 #define V_drop_synfin VNET(drop_synfin) 144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, 145 &VNET_NAME(drop_synfin), 0, 146 "Drop TCP packets with SYN+FIN set"); 147 148 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 149 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, 151 &VNET_NAME(tcp_do_rfc3042), 0, 152 "Enable RFC 3042 (Limited Transmit)"); 153 154 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 155 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, 156 &VNET_NAME(tcp_do_rfc3390), 0, 157 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 158 159 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 160 "Experimental TCP extensions"); 161 162 VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 163 SYSCTL_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_VNET | CTLFLAG_RW, 164 &VNET_NAME(tcp_do_initcwnd10), 0, 165 "Enable RFC 6928 (Increasing initial CWND to 10)"); 166 167 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, 169 &VNET_NAME(tcp_do_rfc3465), 0, 170 "Enable RFC 3465 (Appropriate Byte Counting)"); 171 172 VNET_DEFINE(int, tcp_abc_l_var) = 2; 173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, 174 &VNET_NAME(tcp_abc_l_var), 2, 175 "Cap the max cwnd increment during slow-start to this number of segments"); 176 177 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 178 179 VNET_DEFINE(int, tcp_do_ecn) = 0; 180 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, 181 &VNET_NAME(tcp_do_ecn), 0, 182 "TCP ECN support"); 183 184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 185 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, 186 &VNET_NAME(tcp_ecn_maxretries), 0, 187 "Max retries before giving up on ECN"); 188 189 VNET_DEFINE(int, tcp_insecure_syn) = 0; 190 #define V_tcp_insecure_syn VNET(tcp_insecure_syn) 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, 192 &VNET_NAME(tcp_insecure_syn), 0, 193 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); 194 195 VNET_DEFINE(int, tcp_insecure_rst) = 0; 196 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, 198 &VNET_NAME(tcp_insecure_rst), 0, 199 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); 200 201 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 202 #define V_tcp_recvspace VNET(tcp_recvspace) 203 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, 204 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 205 206 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 207 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, 209 &VNET_NAME(tcp_do_autorcvbuf), 0, 210 "Enable automatic receive buffer sizing"); 211 212 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 213 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, 215 &VNET_NAME(tcp_autorcvbuf_inc), 0, 216 "Incrementor step size of automatic receive buffer"); 217 218 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 219 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, 221 &VNET_NAME(tcp_autorcvbuf_max), 0, 222 "Max size of automatic receive buffer"); 223 224 VNET_DEFINE(struct inpcbhead, tcb); 225 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 226 VNET_DEFINE(struct inpcbinfo, tcbinfo); 227 228 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 229 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 230 struct socket *, struct tcpcb *, int, int, uint8_t, 231 int); 232 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 233 struct tcpcb *, int, int); 234 static void tcp_pulloutofband(struct socket *, 235 struct tcphdr *, struct mbuf *, int); 236 static void tcp_xmit_timer(struct tcpcb *, int); 237 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 238 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 239 uint16_t type); 240 static void inline cc_conn_init(struct tcpcb *tp); 241 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 242 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 243 struct tcphdr *th, struct tcpopt *to); 244 245 /* 246 * TCP statistics are stored in an "array" of counter(9)s. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 VNET_PCPUSTAT_SYSINIT(tcpstat); 250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252 253 #ifdef VIMAGE 254 VNET_PCPUSTAT_SYSUNINIT(tcpstat); 255 #endif /* VIMAGE */ 256 /* 257 * Kernel module interface for updating tcpstat. The argument is an index 258 * into tcpstat treated as an array. 259 */ 260 void 261 kmod_tcpstat_inc(int statnum) 262 { 263 264 counter_u64_add(VNET(tcpstat)[statnum], 1); 265 } 266 267 /* 268 * Wrapper for the TCP established input helper hook. 269 */ 270 static void inline 271 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 272 { 273 struct tcp_hhook_data hhook_data; 274 275 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 276 hhook_data.tp = tp; 277 hhook_data.th = th; 278 hhook_data.to = to; 279 280 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 281 tp->osd); 282 } 283 } 284 285 /* 286 * CC wrapper hook functions 287 */ 288 static void inline 289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 290 { 291 INP_WLOCK_ASSERT(tp->t_inpcb); 292 293 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 294 if (tp->snd_cwnd <= tp->snd_wnd) 295 tp->ccv->flags |= CCF_CWND_LIMITED; 296 else 297 tp->ccv->flags &= ~CCF_CWND_LIMITED; 298 299 if (type == CC_ACK) { 300 if (tp->snd_cwnd > tp->snd_ssthresh) { 301 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 302 V_tcp_abc_l_var * tp->t_maxseg); 303 if (tp->t_bytes_acked >= tp->snd_cwnd) { 304 tp->t_bytes_acked -= tp->snd_cwnd; 305 tp->ccv->flags |= CCF_ABC_SENTAWND; 306 } 307 } else { 308 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 309 tp->t_bytes_acked = 0; 310 } 311 } 312 313 if (CC_ALGO(tp)->ack_received != NULL) { 314 /* XXXLAS: Find a way to live without this */ 315 tp->ccv->curack = th->th_ack; 316 CC_ALGO(tp)->ack_received(tp->ccv, type); 317 } 318 } 319 320 static void inline 321 cc_conn_init(struct tcpcb *tp) 322 { 323 struct hc_metrics_lite metrics; 324 struct inpcb *inp = tp->t_inpcb; 325 int rtt; 326 327 INP_WLOCK_ASSERT(tp->t_inpcb); 328 329 tcp_hc_get(&inp->inp_inc, &metrics); 330 331 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 332 tp->t_srtt = rtt; 333 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 334 TCPSTAT_INC(tcps_usedrtt); 335 if (metrics.rmx_rttvar) { 336 tp->t_rttvar = metrics.rmx_rttvar; 337 TCPSTAT_INC(tcps_usedrttvar); 338 } else { 339 /* default variation is +- 1 rtt */ 340 tp->t_rttvar = 341 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 342 } 343 TCPT_RANGESET(tp->t_rxtcur, 344 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 345 tp->t_rttmin, TCPTV_REXMTMAX); 346 } 347 if (metrics.rmx_ssthresh) { 348 /* 349 * There's some sort of gateway or interface 350 * buffer limit on the path. Use this to set 351 * the slow start threshhold, but set the 352 * threshold to no less than 2*mss. 353 */ 354 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 355 TCPSTAT_INC(tcps_usedssthresh); 356 } 357 358 /* 359 * Set the initial slow-start flight size. 360 * 361 * RFC5681 Section 3.1 specifies the default conservative values. 362 * RFC3390 specifies slightly more aggressive values. 363 * RFC6928 increases it to ten segments. 364 * 365 * If a SYN or SYN/ACK was lost and retransmitted, we have to 366 * reduce the initial CWND to one segment as congestion is likely 367 * requiring us to be cautious. 368 */ 369 if (tp->snd_cwnd == 1) 370 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 371 else if (V_tcp_do_initcwnd10) 372 tp->snd_cwnd = min(10 * tp->t_maxseg, 373 max(2 * tp->t_maxseg, 14600)); 374 else if (V_tcp_do_rfc3390) 375 tp->snd_cwnd = min(4 * tp->t_maxseg, 376 max(2 * tp->t_maxseg, 4380)); 377 else { 378 /* Per RFC5681 Section 3.1 */ 379 if (tp->t_maxseg > 2190) 380 tp->snd_cwnd = 2 * tp->t_maxseg; 381 else if (tp->t_maxseg > 1095) 382 tp->snd_cwnd = 3 * tp->t_maxseg; 383 else 384 tp->snd_cwnd = 4 * tp->t_maxseg; 385 } 386 387 if (CC_ALGO(tp)->conn_init != NULL) 388 CC_ALGO(tp)->conn_init(tp->ccv); 389 } 390 391 void inline 392 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 393 { 394 INP_WLOCK_ASSERT(tp->t_inpcb); 395 396 switch(type) { 397 case CC_NDUPACK: 398 if (!IN_FASTRECOVERY(tp->t_flags)) { 399 tp->snd_recover = tp->snd_max; 400 if (tp->t_flags & TF_ECN_PERMIT) 401 tp->t_flags |= TF_ECN_SND_CWR; 402 } 403 break; 404 case CC_ECN: 405 if (!IN_CONGRECOVERY(tp->t_flags)) { 406 TCPSTAT_INC(tcps_ecn_rcwnd); 407 tp->snd_recover = tp->snd_max; 408 if (tp->t_flags & TF_ECN_PERMIT) 409 tp->t_flags |= TF_ECN_SND_CWR; 410 } 411 break; 412 case CC_RTO: 413 tp->t_dupacks = 0; 414 tp->t_bytes_acked = 0; 415 EXIT_RECOVERY(tp->t_flags); 416 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 417 tp->t_maxseg) * tp->t_maxseg; 418 tp->snd_cwnd = tp->t_maxseg; 419 break; 420 case CC_RTO_ERR: 421 TCPSTAT_INC(tcps_sndrexmitbad); 422 /* RTO was unnecessary, so reset everything. */ 423 tp->snd_cwnd = tp->snd_cwnd_prev; 424 tp->snd_ssthresh = tp->snd_ssthresh_prev; 425 tp->snd_recover = tp->snd_recover_prev; 426 if (tp->t_flags & TF_WASFRECOVERY) 427 ENTER_FASTRECOVERY(tp->t_flags); 428 if (tp->t_flags & TF_WASCRECOVERY) 429 ENTER_CONGRECOVERY(tp->t_flags); 430 tp->snd_nxt = tp->snd_max; 431 tp->t_flags &= ~TF_PREVVALID; 432 tp->t_badrxtwin = 0; 433 break; 434 } 435 436 if (CC_ALGO(tp)->cong_signal != NULL) { 437 if (th != NULL) 438 tp->ccv->curack = th->th_ack; 439 CC_ALGO(tp)->cong_signal(tp->ccv, type); 440 } 441 } 442 443 static void inline 444 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 445 { 446 INP_WLOCK_ASSERT(tp->t_inpcb); 447 448 /* XXXLAS: KASSERT that we're in recovery? */ 449 450 if (CC_ALGO(tp)->post_recovery != NULL) { 451 tp->ccv->curack = th->th_ack; 452 CC_ALGO(tp)->post_recovery(tp->ccv); 453 } 454 /* XXXLAS: EXIT_RECOVERY ? */ 455 tp->t_bytes_acked = 0; 456 } 457 458 #ifdef TCP_SIGNATURE 459 static inline int 460 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 461 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 462 { 463 int ret; 464 465 tcp_fields_to_net(th); 466 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 467 tcp_fields_to_host(th); 468 return (ret); 469 } 470 #endif 471 472 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 473 #ifdef INET6 474 #define ND6_HINT(tp) \ 475 do { \ 476 if ((tp) && (tp)->t_inpcb && \ 477 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 478 nd6_nud_hint(NULL, NULL, 0); \ 479 } while (0) 480 #else 481 #define ND6_HINT(tp) 482 #endif 483 484 /* 485 * Indicate whether this ack should be delayed. We can delay the ack if 486 * following conditions are met: 487 * - There is no delayed ack timer in progress. 488 * - Our last ack wasn't a 0-sized window. We never want to delay 489 * the ack that opens up a 0-sized window. 490 * - LRO wasn't used for this segment. We make sure by checking that the 491 * segment size is not larger than the MSS. 492 * - Delayed acks are enabled or this is a half-synchronized T/TCP 493 * connection. 494 */ 495 #define DELAY_ACK(tp, tlen) \ 496 ((!tcp_timer_active(tp, TT_DELACK) && \ 497 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 498 (tlen <= tp->t_maxopd) && \ 499 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 500 501 static void inline 502 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) 503 { 504 INP_WLOCK_ASSERT(tp->t_inpcb); 505 506 if (CC_ALGO(tp)->ecnpkt_handler != NULL) { 507 switch (iptos & IPTOS_ECN_MASK) { 508 case IPTOS_ECN_CE: 509 tp->ccv->flags |= CCF_IPHDR_CE; 510 break; 511 case IPTOS_ECN_ECT0: 512 tp->ccv->flags &= ~CCF_IPHDR_CE; 513 break; 514 case IPTOS_ECN_ECT1: 515 tp->ccv->flags &= ~CCF_IPHDR_CE; 516 break; 517 } 518 519 if (th->th_flags & TH_CWR) 520 tp->ccv->flags |= CCF_TCPHDR_CWR; 521 else 522 tp->ccv->flags &= ~CCF_TCPHDR_CWR; 523 524 if (tp->t_flags & TF_DELACK) 525 tp->ccv->flags |= CCF_DELACK; 526 else 527 tp->ccv->flags &= ~CCF_DELACK; 528 529 CC_ALGO(tp)->ecnpkt_handler(tp->ccv); 530 531 if (tp->ccv->flags & CCF_ACKNOW) 532 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 533 } 534 } 535 536 /* 537 * TCP input handling is split into multiple parts: 538 * tcp6_input is a thin wrapper around tcp_input for the extended 539 * ip6_protox[] call format in ip6_input 540 * tcp_input handles primary segment validation, inpcb lookup and 541 * SYN processing on listen sockets 542 * tcp_do_segment processes the ACK and text of the segment for 543 * establishing, established and closing connections 544 */ 545 #ifdef INET6 546 int 547 tcp6_input(struct mbuf **mp, int *offp, int proto) 548 { 549 struct mbuf *m = *mp; 550 struct in6_ifaddr *ia6; 551 struct ip6_hdr *ip6; 552 553 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 554 555 /* 556 * draft-itojun-ipv6-tcp-to-anycast 557 * better place to put this in? 558 */ 559 ip6 = mtod(m, struct ip6_hdr *); 560 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 561 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 562 struct ip6_hdr *ip6; 563 564 ifa_free(&ia6->ia_ifa); 565 ip6 = mtod(m, struct ip6_hdr *); 566 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 567 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 568 return (IPPROTO_DONE); 569 } 570 if (ia6) 571 ifa_free(&ia6->ia_ifa); 572 573 return (tcp_input(mp, offp, proto)); 574 } 575 #endif /* INET6 */ 576 577 int 578 tcp_input(struct mbuf **mp, int *offp, int proto) 579 { 580 struct mbuf *m = *mp; 581 struct tcphdr *th = NULL; 582 struct ip *ip = NULL; 583 struct inpcb *inp = NULL; 584 struct tcpcb *tp = NULL; 585 struct socket *so = NULL; 586 u_char *optp = NULL; 587 int off0; 588 int optlen = 0; 589 #ifdef INET 590 int len; 591 #endif 592 int tlen = 0, off; 593 int drop_hdrlen; 594 int thflags; 595 int rstreason = 0; /* For badport_bandlim accounting purposes */ 596 #ifdef TCP_SIGNATURE 597 uint8_t sig_checked = 0; 598 #endif 599 uint8_t iptos = 0; 600 struct m_tag *fwd_tag = NULL; 601 #ifdef INET6 602 struct ip6_hdr *ip6 = NULL; 603 int isipv6; 604 #else 605 const void *ip6 = NULL; 606 #endif /* INET6 */ 607 struct tcpopt to; /* options in this segment */ 608 char *s = NULL; /* address and port logging */ 609 int ti_locked; 610 #define TI_UNLOCKED 1 611 #define TI_RLOCKED 2 612 613 #ifdef TCPDEBUG 614 /* 615 * The size of tcp_saveipgen must be the size of the max ip header, 616 * now IPv6. 617 */ 618 u_char tcp_saveipgen[IP6_HDR_LEN]; 619 struct tcphdr tcp_savetcp; 620 short ostate = 0; 621 #endif 622 623 #ifdef INET6 624 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 625 #endif 626 627 off0 = *offp; 628 m = *mp; 629 *mp = NULL; 630 to.to_flags = 0; 631 TCPSTAT_INC(tcps_rcvtotal); 632 633 #ifdef INET6 634 if (isipv6) { 635 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 636 637 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 638 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 639 if (m == NULL) { 640 TCPSTAT_INC(tcps_rcvshort); 641 return (IPPROTO_DONE); 642 } 643 } 644 645 ip6 = mtod(m, struct ip6_hdr *); 646 th = (struct tcphdr *)((caddr_t)ip6 + off0); 647 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 648 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 649 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 650 th->th_sum = m->m_pkthdr.csum_data; 651 else 652 th->th_sum = in6_cksum_pseudo(ip6, tlen, 653 IPPROTO_TCP, m->m_pkthdr.csum_data); 654 th->th_sum ^= 0xffff; 655 } else 656 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 657 if (th->th_sum) { 658 TCPSTAT_INC(tcps_rcvbadsum); 659 goto drop; 660 } 661 662 /* 663 * Be proactive about unspecified IPv6 address in source. 664 * As we use all-zero to indicate unbounded/unconnected pcb, 665 * unspecified IPv6 address can be used to confuse us. 666 * 667 * Note that packets with unspecified IPv6 destination is 668 * already dropped in ip6_input. 669 */ 670 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 671 /* XXX stat */ 672 goto drop; 673 } 674 } 675 #endif 676 #if defined(INET) && defined(INET6) 677 else 678 #endif 679 #ifdef INET 680 { 681 /* 682 * Get IP and TCP header together in first mbuf. 683 * Note: IP leaves IP header in first mbuf. 684 */ 685 if (off0 > sizeof (struct ip)) { 686 ip_stripoptions(m); 687 off0 = sizeof(struct ip); 688 } 689 if (m->m_len < sizeof (struct tcpiphdr)) { 690 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 691 == NULL) { 692 TCPSTAT_INC(tcps_rcvshort); 693 return (IPPROTO_DONE); 694 } 695 } 696 ip = mtod(m, struct ip *); 697 th = (struct tcphdr *)((caddr_t)ip + off0); 698 tlen = ntohs(ip->ip_len) - off0; 699 700 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 701 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 702 th->th_sum = m->m_pkthdr.csum_data; 703 else 704 th->th_sum = in_pseudo(ip->ip_src.s_addr, 705 ip->ip_dst.s_addr, 706 htonl(m->m_pkthdr.csum_data + tlen + 707 IPPROTO_TCP)); 708 th->th_sum ^= 0xffff; 709 } else { 710 struct ipovly *ipov = (struct ipovly *)ip; 711 712 /* 713 * Checksum extended TCP header and data. 714 */ 715 len = off0 + tlen; 716 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 717 ipov->ih_len = htons(tlen); 718 th->th_sum = in_cksum(m, len); 719 /* Reset length for SDT probes. */ 720 ip->ip_len = htons(tlen + off0); 721 } 722 723 if (th->th_sum) { 724 TCPSTAT_INC(tcps_rcvbadsum); 725 goto drop; 726 } 727 /* Re-initialization for later version check */ 728 ip->ip_v = IPVERSION; 729 } 730 #endif /* INET */ 731 732 #ifdef INET6 733 if (isipv6) 734 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 735 #endif 736 #if defined(INET) && defined(INET6) 737 else 738 #endif 739 #ifdef INET 740 iptos = ip->ip_tos; 741 #endif 742 743 /* 744 * Check that TCP offset makes sense, 745 * pull out TCP options and adjust length. XXX 746 */ 747 off = th->th_off << 2; 748 if (off < sizeof (struct tcphdr) || off > tlen) { 749 TCPSTAT_INC(tcps_rcvbadoff); 750 goto drop; 751 } 752 tlen -= off; /* tlen is used instead of ti->ti_len */ 753 if (off > sizeof (struct tcphdr)) { 754 #ifdef INET6 755 if (isipv6) { 756 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 757 ip6 = mtod(m, struct ip6_hdr *); 758 th = (struct tcphdr *)((caddr_t)ip6 + off0); 759 } 760 #endif 761 #if defined(INET) && defined(INET6) 762 else 763 #endif 764 #ifdef INET 765 { 766 if (m->m_len < sizeof(struct ip) + off) { 767 if ((m = m_pullup(m, sizeof (struct ip) + off)) 768 == NULL) { 769 TCPSTAT_INC(tcps_rcvshort); 770 return (IPPROTO_DONE); 771 } 772 ip = mtod(m, struct ip *); 773 th = (struct tcphdr *)((caddr_t)ip + off0); 774 } 775 } 776 #endif 777 optlen = off - sizeof (struct tcphdr); 778 optp = (u_char *)(th + 1); 779 } 780 thflags = th->th_flags; 781 782 /* 783 * Convert TCP protocol specific fields to host format. 784 */ 785 tcp_fields_to_host(th); 786 787 /* 788 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 789 */ 790 drop_hdrlen = off0 + off; 791 792 /* 793 * Locate pcb for segment; if we're likely to add or remove a 794 * connection then first acquire pcbinfo lock. There are three cases 795 * where we might discover later we need a write lock despite the 796 * flags: ACKs moving a connection out of the syncache, ACKs for a 797 * connection in TIMEWAIT and SYNs not targeting a listening socket. 798 */ 799 if ((thflags & (TH_FIN | TH_RST)) != 0) { 800 INP_INFO_RLOCK(&V_tcbinfo); 801 ti_locked = TI_RLOCKED; 802 } else 803 ti_locked = TI_UNLOCKED; 804 805 /* 806 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 807 */ 808 if ( 809 #ifdef INET6 810 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 811 #ifdef INET 812 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 813 #endif 814 #endif 815 #if defined(INET) && !defined(INET6) 816 (m->m_flags & M_IP_NEXTHOP) 817 #endif 818 ) 819 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 820 821 findpcb: 822 #ifdef INVARIANTS 823 if (ti_locked == TI_RLOCKED) { 824 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 825 } else { 826 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 827 } 828 #endif 829 #ifdef INET6 830 if (isipv6 && fwd_tag != NULL) { 831 struct sockaddr_in6 *next_hop6; 832 833 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 834 /* 835 * Transparently forwarded. Pretend to be the destination. 836 * Already got one like this? 837 */ 838 inp = in6_pcblookup_mbuf(&V_tcbinfo, 839 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 840 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 841 if (!inp) { 842 /* 843 * It's new. Try to find the ambushing socket. 844 * Because we've rewritten the destination address, 845 * any hardware-generated hash is ignored. 846 */ 847 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 848 th->th_sport, &next_hop6->sin6_addr, 849 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 850 th->th_dport, INPLOOKUP_WILDCARD | 851 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 852 } 853 } else if (isipv6) { 854 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 855 th->th_sport, &ip6->ip6_dst, th->th_dport, 856 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 857 m->m_pkthdr.rcvif, m); 858 } 859 #endif /* INET6 */ 860 #if defined(INET6) && defined(INET) 861 else 862 #endif 863 #ifdef INET 864 if (fwd_tag != NULL) { 865 struct sockaddr_in *next_hop; 866 867 next_hop = (struct sockaddr_in *)(fwd_tag+1); 868 /* 869 * Transparently forwarded. Pretend to be the destination. 870 * already got one like this? 871 */ 872 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 873 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 874 m->m_pkthdr.rcvif, m); 875 if (!inp) { 876 /* 877 * It's new. Try to find the ambushing socket. 878 * Because we've rewritten the destination address, 879 * any hardware-generated hash is ignored. 880 */ 881 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 882 th->th_sport, next_hop->sin_addr, 883 next_hop->sin_port ? ntohs(next_hop->sin_port) : 884 th->th_dport, INPLOOKUP_WILDCARD | 885 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 886 } 887 } else 888 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 889 th->th_sport, ip->ip_dst, th->th_dport, 890 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 891 m->m_pkthdr.rcvif, m); 892 #endif /* INET */ 893 894 /* 895 * If the INPCB does not exist then all data in the incoming 896 * segment is discarded and an appropriate RST is sent back. 897 * XXX MRT Send RST using which routing table? 898 */ 899 if (inp == NULL) { 900 /* 901 * Log communication attempts to ports that are not 902 * in use. 903 */ 904 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 905 tcp_log_in_vain == 2) { 906 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 907 log(LOG_INFO, "%s; %s: Connection attempt " 908 "to closed port\n", s, __func__); 909 } 910 /* 911 * When blackholing do not respond with a RST but 912 * completely ignore the segment and drop it. 913 */ 914 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 915 V_blackhole == 2) 916 goto dropunlock; 917 918 rstreason = BANDLIM_RST_CLOSEDPORT; 919 goto dropwithreset; 920 } 921 INP_WLOCK_ASSERT(inp); 922 if ((inp->inp_flowtype == M_HASHTYPE_NONE) && 923 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) && 924 ((inp->inp_socket == NULL) || 925 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) { 926 inp->inp_flowid = m->m_pkthdr.flowid; 927 inp->inp_flowtype = M_HASHTYPE_GET(m); 928 } 929 #ifdef IPSEC 930 #ifdef INET6 931 if (isipv6 && ipsec6_in_reject(m, inp)) { 932 goto dropunlock; 933 } else 934 #endif /* INET6 */ 935 if (ipsec4_in_reject(m, inp) != 0) { 936 goto dropunlock; 937 } 938 #endif /* IPSEC */ 939 940 /* 941 * Check the minimum TTL for socket. 942 */ 943 if (inp->inp_ip_minttl != 0) { 944 #ifdef INET6 945 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 946 goto dropunlock; 947 else 948 #endif 949 if (inp->inp_ip_minttl > ip->ip_ttl) 950 goto dropunlock; 951 } 952 953 /* 954 * A previous connection in TIMEWAIT state is supposed to catch stray 955 * or duplicate segments arriving late. If this segment was a 956 * legitimate new connection attempt, the old INPCB gets removed and 957 * we can try again to find a listening socket. 958 * 959 * At this point, due to earlier optimism, we may hold only an inpcb 960 * lock, and not the inpcbinfo write lock. If so, we need to try to 961 * acquire it, or if that fails, acquire a reference on the inpcb, 962 * drop all locks, acquire a global write lock, and then re-acquire 963 * the inpcb lock. We may at that point discover that another thread 964 * has tried to free the inpcb, in which case we need to loop back 965 * and try to find a new inpcb to deliver to. 966 * 967 * XXXRW: It may be time to rethink timewait locking. 968 */ 969 relocked: 970 if (inp->inp_flags & INP_TIMEWAIT) { 971 if (ti_locked == TI_UNLOCKED) { 972 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 973 in_pcbref(inp); 974 INP_WUNLOCK(inp); 975 INP_INFO_RLOCK(&V_tcbinfo); 976 ti_locked = TI_RLOCKED; 977 INP_WLOCK(inp); 978 if (in_pcbrele_wlocked(inp)) { 979 inp = NULL; 980 goto findpcb; 981 } 982 } else 983 ti_locked = TI_RLOCKED; 984 } 985 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 986 987 if (thflags & TH_SYN) 988 tcp_dooptions(&to, optp, optlen, TO_SYN); 989 /* 990 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 991 */ 992 if (tcp_twcheck(inp, &to, th, m, tlen)) 993 goto findpcb; 994 INP_INFO_RUNLOCK(&V_tcbinfo); 995 return (IPPROTO_DONE); 996 } 997 /* 998 * The TCPCB may no longer exist if the connection is winding 999 * down or it is in the CLOSED state. Either way we drop the 1000 * segment and send an appropriate response. 1001 */ 1002 tp = intotcpcb(inp); 1003 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 1004 rstreason = BANDLIM_RST_CLOSEDPORT; 1005 goto dropwithreset; 1006 } 1007 1008 #ifdef TCP_OFFLOAD 1009 if (tp->t_flags & TF_TOE) { 1010 tcp_offload_input(tp, m); 1011 m = NULL; /* consumed by the TOE driver */ 1012 goto dropunlock; 1013 } 1014 #endif 1015 1016 /* 1017 * We've identified a valid inpcb, but it could be that we need an 1018 * inpcbinfo write lock but don't hold it. In this case, attempt to 1019 * acquire using the same strategy as the TIMEWAIT case above. If we 1020 * relock, we have to jump back to 'relocked' as the connection might 1021 * now be in TIMEWAIT. 1022 */ 1023 #ifdef INVARIANTS 1024 if ((thflags & (TH_FIN | TH_RST)) != 0) 1025 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1026 #endif 1027 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 1028 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN)))) { 1029 if (ti_locked == TI_UNLOCKED) { 1030 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) { 1031 in_pcbref(inp); 1032 INP_WUNLOCK(inp); 1033 INP_INFO_RLOCK(&V_tcbinfo); 1034 ti_locked = TI_RLOCKED; 1035 INP_WLOCK(inp); 1036 if (in_pcbrele_wlocked(inp)) { 1037 inp = NULL; 1038 goto findpcb; 1039 } 1040 goto relocked; 1041 } else 1042 ti_locked = TI_RLOCKED; 1043 } 1044 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1045 } 1046 1047 #ifdef MAC 1048 INP_WLOCK_ASSERT(inp); 1049 if (mac_inpcb_check_deliver(inp, m)) 1050 goto dropunlock; 1051 #endif 1052 so = inp->inp_socket; 1053 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1054 #ifdef TCPDEBUG 1055 if (so->so_options & SO_DEBUG) { 1056 ostate = tp->t_state; 1057 #ifdef INET6 1058 if (isipv6) { 1059 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1060 } else 1061 #endif 1062 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1063 tcp_savetcp = *th; 1064 } 1065 #endif /* TCPDEBUG */ 1066 /* 1067 * When the socket is accepting connections (the INPCB is in LISTEN 1068 * state) we look into the SYN cache if this is a new connection 1069 * attempt or the completion of a previous one. 1070 */ 1071 if (so->so_options & SO_ACCEPTCONN) { 1072 struct in_conninfo inc; 1073 1074 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1075 "tp not listening", __func__)); 1076 bzero(&inc, sizeof(inc)); 1077 #ifdef INET6 1078 if (isipv6) { 1079 inc.inc_flags |= INC_ISIPV6; 1080 inc.inc6_faddr = ip6->ip6_src; 1081 inc.inc6_laddr = ip6->ip6_dst; 1082 } else 1083 #endif 1084 { 1085 inc.inc_faddr = ip->ip_src; 1086 inc.inc_laddr = ip->ip_dst; 1087 } 1088 inc.inc_fport = th->th_sport; 1089 inc.inc_lport = th->th_dport; 1090 inc.inc_fibnum = so->so_fibnum; 1091 1092 /* 1093 * Check for an existing connection attempt in syncache if 1094 * the flag is only ACK. A successful lookup creates a new 1095 * socket appended to the listen queue in SYN_RECEIVED state. 1096 */ 1097 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1098 1099 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1100 /* 1101 * Parse the TCP options here because 1102 * syncookies need access to the reflected 1103 * timestamp. 1104 */ 1105 tcp_dooptions(&to, optp, optlen, 0); 1106 /* 1107 * NB: syncache_expand() doesn't unlock 1108 * inp and tcpinfo locks. 1109 */ 1110 if (!syncache_expand(&inc, &to, th, &so, m)) { 1111 /* 1112 * No syncache entry or ACK was not 1113 * for our SYN/ACK. Send a RST. 1114 * NB: syncache did its own logging 1115 * of the failure cause. 1116 */ 1117 rstreason = BANDLIM_RST_OPENPORT; 1118 goto dropwithreset; 1119 } 1120 if (so == NULL) { 1121 /* 1122 * We completed the 3-way handshake 1123 * but could not allocate a socket 1124 * either due to memory shortage, 1125 * listen queue length limits or 1126 * global socket limits. Send RST 1127 * or wait and have the remote end 1128 * retransmit the ACK for another 1129 * try. 1130 */ 1131 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1132 log(LOG_DEBUG, "%s; %s: Listen socket: " 1133 "Socket allocation failed due to " 1134 "limits or memory shortage, %s\n", 1135 s, __func__, 1136 V_tcp_sc_rst_sock_fail ? 1137 "sending RST" : "try again"); 1138 if (V_tcp_sc_rst_sock_fail) { 1139 rstreason = BANDLIM_UNLIMITED; 1140 goto dropwithreset; 1141 } else 1142 goto dropunlock; 1143 } 1144 /* 1145 * Socket is created in state SYN_RECEIVED. 1146 * Unlock the listen socket, lock the newly 1147 * created socket and update the tp variable. 1148 */ 1149 INP_WUNLOCK(inp); /* listen socket */ 1150 inp = sotoinpcb(so); 1151 /* 1152 * New connection inpcb is already locked by 1153 * syncache_expand(). 1154 */ 1155 INP_WLOCK_ASSERT(inp); 1156 tp = intotcpcb(inp); 1157 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1158 ("%s: ", __func__)); 1159 #ifdef TCP_SIGNATURE 1160 if (sig_checked == 0) { 1161 tcp_dooptions(&to, optp, optlen, 1162 (thflags & TH_SYN) ? TO_SYN : 0); 1163 if (!tcp_signature_verify_input(m, off0, tlen, 1164 optlen, &to, th, tp->t_flags)) { 1165 1166 /* 1167 * In SYN_SENT state if it receives an 1168 * RST, it is allowed for further 1169 * processing. 1170 */ 1171 if ((thflags & TH_RST) == 0 || 1172 (tp->t_state == TCPS_SYN_SENT) == 0) 1173 goto dropunlock; 1174 } 1175 sig_checked = 1; 1176 } 1177 #endif 1178 1179 /* 1180 * Process the segment and the data it 1181 * contains. tcp_do_segment() consumes 1182 * the mbuf chain and unlocks the inpcb. 1183 */ 1184 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1185 iptos, ti_locked); 1186 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1187 return (IPPROTO_DONE); 1188 } 1189 /* 1190 * Segment flag validation for new connection attempts: 1191 * 1192 * Our (SYN|ACK) response was rejected. 1193 * Check with syncache and remove entry to prevent 1194 * retransmits. 1195 * 1196 * NB: syncache_chkrst does its own logging of failure 1197 * causes. 1198 */ 1199 if (thflags & TH_RST) { 1200 syncache_chkrst(&inc, th); 1201 goto dropunlock; 1202 } 1203 /* 1204 * We can't do anything without SYN. 1205 */ 1206 if ((thflags & TH_SYN) == 0) { 1207 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1208 log(LOG_DEBUG, "%s; %s: Listen socket: " 1209 "SYN is missing, segment ignored\n", 1210 s, __func__); 1211 TCPSTAT_INC(tcps_badsyn); 1212 goto dropunlock; 1213 } 1214 /* 1215 * (SYN|ACK) is bogus on a listen socket. 1216 */ 1217 if (thflags & TH_ACK) { 1218 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1219 log(LOG_DEBUG, "%s; %s: Listen socket: " 1220 "SYN|ACK invalid, segment rejected\n", 1221 s, __func__); 1222 syncache_badack(&inc); /* XXX: Not needed! */ 1223 TCPSTAT_INC(tcps_badsyn); 1224 rstreason = BANDLIM_RST_OPENPORT; 1225 goto dropwithreset; 1226 } 1227 /* 1228 * If the drop_synfin option is enabled, drop all 1229 * segments with both the SYN and FIN bits set. 1230 * This prevents e.g. nmap from identifying the 1231 * TCP/IP stack. 1232 * XXX: Poor reasoning. nmap has other methods 1233 * and is constantly refining its stack detection 1234 * strategies. 1235 * XXX: This is a violation of the TCP specification 1236 * and was used by RFC1644. 1237 */ 1238 if ((thflags & TH_FIN) && V_drop_synfin) { 1239 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1240 log(LOG_DEBUG, "%s; %s: Listen socket: " 1241 "SYN|FIN segment ignored (based on " 1242 "sysctl setting)\n", s, __func__); 1243 TCPSTAT_INC(tcps_badsyn); 1244 goto dropunlock; 1245 } 1246 /* 1247 * Segment's flags are (SYN) or (SYN|FIN). 1248 * 1249 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1250 * as they do not affect the state of the TCP FSM. 1251 * The data pointed to by TH_URG and th_urp is ignored. 1252 */ 1253 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1254 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1255 KASSERT(thflags & (TH_SYN), 1256 ("%s: Listen socket: TH_SYN not set", __func__)); 1257 #ifdef INET6 1258 /* 1259 * If deprecated address is forbidden, 1260 * we do not accept SYN to deprecated interface 1261 * address to prevent any new inbound connection from 1262 * getting established. 1263 * When we do not accept SYN, we send a TCP RST, 1264 * with deprecated source address (instead of dropping 1265 * it). We compromise it as it is much better for peer 1266 * to send a RST, and RST will be the final packet 1267 * for the exchange. 1268 * 1269 * If we do not forbid deprecated addresses, we accept 1270 * the SYN packet. RFC2462 does not suggest dropping 1271 * SYN in this case. 1272 * If we decipher RFC2462 5.5.4, it says like this: 1273 * 1. use of deprecated addr with existing 1274 * communication is okay - "SHOULD continue to be 1275 * used" 1276 * 2. use of it with new communication: 1277 * (2a) "SHOULD NOT be used if alternate address 1278 * with sufficient scope is available" 1279 * (2b) nothing mentioned otherwise. 1280 * Here we fall into (2b) case as we have no choice in 1281 * our source address selection - we must obey the peer. 1282 * 1283 * The wording in RFC2462 is confusing, and there are 1284 * multiple description text for deprecated address 1285 * handling - worse, they are not exactly the same. 1286 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1287 */ 1288 if (isipv6 && !V_ip6_use_deprecated) { 1289 struct in6_ifaddr *ia6; 1290 1291 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1292 if (ia6 != NULL && 1293 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1294 ifa_free(&ia6->ia_ifa); 1295 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1296 log(LOG_DEBUG, "%s; %s: Listen socket: " 1297 "Connection attempt to deprecated " 1298 "IPv6 address rejected\n", 1299 s, __func__); 1300 rstreason = BANDLIM_RST_OPENPORT; 1301 goto dropwithreset; 1302 } 1303 if (ia6) 1304 ifa_free(&ia6->ia_ifa); 1305 } 1306 #endif /* INET6 */ 1307 /* 1308 * Basic sanity checks on incoming SYN requests: 1309 * Don't respond if the destination is a link layer 1310 * broadcast according to RFC1122 4.2.3.10, p. 104. 1311 * If it is from this socket it must be forged. 1312 * Don't respond if the source or destination is a 1313 * global or subnet broad- or multicast address. 1314 * Note that it is quite possible to receive unicast 1315 * link-layer packets with a broadcast IP address. Use 1316 * in_broadcast() to find them. 1317 */ 1318 if (m->m_flags & (M_BCAST|M_MCAST)) { 1319 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1320 log(LOG_DEBUG, "%s; %s: Listen socket: " 1321 "Connection attempt from broad- or multicast " 1322 "link layer address ignored\n", s, __func__); 1323 goto dropunlock; 1324 } 1325 #ifdef INET6 1326 if (isipv6) { 1327 if (th->th_dport == th->th_sport && 1328 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1329 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1330 log(LOG_DEBUG, "%s; %s: Listen socket: " 1331 "Connection attempt to/from self " 1332 "ignored\n", s, __func__); 1333 goto dropunlock; 1334 } 1335 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1336 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1337 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1338 log(LOG_DEBUG, "%s; %s: Listen socket: " 1339 "Connection attempt from/to multicast " 1340 "address ignored\n", s, __func__); 1341 goto dropunlock; 1342 } 1343 } 1344 #endif 1345 #if defined(INET) && defined(INET6) 1346 else 1347 #endif 1348 #ifdef INET 1349 { 1350 if (th->th_dport == th->th_sport && 1351 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1352 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1353 log(LOG_DEBUG, "%s; %s: Listen socket: " 1354 "Connection attempt from/to self " 1355 "ignored\n", s, __func__); 1356 goto dropunlock; 1357 } 1358 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1359 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1360 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1361 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1362 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1363 log(LOG_DEBUG, "%s; %s: Listen socket: " 1364 "Connection attempt from/to broad- " 1365 "or multicast address ignored\n", 1366 s, __func__); 1367 goto dropunlock; 1368 } 1369 } 1370 #endif 1371 /* 1372 * SYN appears to be valid. Create compressed TCP state 1373 * for syncache. 1374 */ 1375 #ifdef TCPDEBUG 1376 if (so->so_options & SO_DEBUG) 1377 tcp_trace(TA_INPUT, ostate, tp, 1378 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1379 #endif 1380 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 1381 tcp_dooptions(&to, optp, optlen, TO_SYN); 1382 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1383 /* 1384 * Entry added to syncache and mbuf consumed. 1385 * Only the listen socket is unlocked by syncache_add(). 1386 */ 1387 if (ti_locked == TI_RLOCKED) { 1388 INP_INFO_RUNLOCK(&V_tcbinfo); 1389 ti_locked = TI_UNLOCKED; 1390 } 1391 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1392 return (IPPROTO_DONE); 1393 } else if (tp->t_state == TCPS_LISTEN) { 1394 /* 1395 * When a listen socket is torn down the SO_ACCEPTCONN 1396 * flag is removed first while connections are drained 1397 * from the accept queue in a unlock/lock cycle of the 1398 * ACCEPT_LOCK, opening a race condition allowing a SYN 1399 * attempt go through unhandled. 1400 */ 1401 goto dropunlock; 1402 } 1403 1404 #ifdef TCP_SIGNATURE 1405 if (sig_checked == 0) { 1406 tcp_dooptions(&to, optp, optlen, 1407 (thflags & TH_SYN) ? TO_SYN : 0); 1408 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1409 th, tp->t_flags)) { 1410 1411 /* 1412 * In SYN_SENT state if it receives an RST, it is 1413 * allowed for further processing. 1414 */ 1415 if ((thflags & TH_RST) == 0 || 1416 (tp->t_state == TCPS_SYN_SENT) == 0) 1417 goto dropunlock; 1418 } 1419 sig_checked = 1; 1420 } 1421 #endif 1422 1423 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1424 1425 /* 1426 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1427 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1428 * the inpcb, and unlocks pcbinfo. 1429 */ 1430 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1431 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1432 return (IPPROTO_DONE); 1433 1434 dropwithreset: 1435 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1436 1437 if (ti_locked == TI_RLOCKED) { 1438 INP_INFO_RUNLOCK(&V_tcbinfo); 1439 ti_locked = TI_UNLOCKED; 1440 } 1441 #ifdef INVARIANTS 1442 else { 1443 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1444 "ti_locked: %d", __func__, ti_locked)); 1445 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1446 } 1447 #endif 1448 1449 if (inp != NULL) { 1450 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1451 INP_WUNLOCK(inp); 1452 } else 1453 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1454 m = NULL; /* mbuf chain got consumed. */ 1455 goto drop; 1456 1457 dropunlock: 1458 if (m != NULL) 1459 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1460 1461 if (ti_locked == TI_RLOCKED) { 1462 INP_INFO_RUNLOCK(&V_tcbinfo); 1463 ti_locked = TI_UNLOCKED; 1464 } 1465 #ifdef INVARIANTS 1466 else { 1467 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1468 "ti_locked: %d", __func__, ti_locked)); 1469 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1470 } 1471 #endif 1472 1473 if (inp != NULL) 1474 INP_WUNLOCK(inp); 1475 1476 drop: 1477 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1478 if (s != NULL) 1479 free(s, M_TCPLOG); 1480 if (m != NULL) 1481 m_freem(m); 1482 return (IPPROTO_DONE); 1483 } 1484 1485 static void 1486 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1487 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1488 int ti_locked) 1489 { 1490 int thflags, acked, ourfinisacked, needoutput = 0; 1491 int rstreason, todrop, win; 1492 u_long tiwin; 1493 char *s; 1494 struct in_conninfo *inc; 1495 struct mbuf *mfree; 1496 struct tcpopt to; 1497 1498 #ifdef TCPDEBUG 1499 /* 1500 * The size of tcp_saveipgen must be the size of the max ip header, 1501 * now IPv6. 1502 */ 1503 u_char tcp_saveipgen[IP6_HDR_LEN]; 1504 struct tcphdr tcp_savetcp; 1505 short ostate = 0; 1506 #endif 1507 thflags = th->th_flags; 1508 inc = &tp->t_inpcb->inp_inc; 1509 tp->sackhint.last_sack_ack = 0; 1510 1511 /* 1512 * If this is either a state-changing packet or current state isn't 1513 * established, we require a write lock on tcbinfo. Otherwise, we 1514 * allow the tcbinfo to be in either alocked or unlocked, as the 1515 * caller may have unnecessarily acquired a write lock due to a race. 1516 */ 1517 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1518 tp->t_state != TCPS_ESTABLISHED) { 1519 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1520 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1521 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1522 } else { 1523 #ifdef INVARIANTS 1524 if (ti_locked == TI_RLOCKED) 1525 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1526 else { 1527 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1528 "ti_locked: %d", __func__, ti_locked)); 1529 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1530 } 1531 #endif 1532 } 1533 INP_WLOCK_ASSERT(tp->t_inpcb); 1534 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1535 __func__)); 1536 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1537 __func__)); 1538 1539 /* 1540 * Segment received on connection. 1541 * Reset idle time and keep-alive timer. 1542 * XXX: This should be done after segment 1543 * validation to ignore broken/spoofed segs. 1544 */ 1545 tp->t_rcvtime = ticks; 1546 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1547 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1548 1549 /* 1550 * Scale up the window into a 32-bit value. 1551 * For the SYN_SENT state the scale is zero. 1552 */ 1553 tiwin = th->th_win << tp->snd_scale; 1554 1555 /* 1556 * TCP ECN processing. 1557 */ 1558 if (tp->t_flags & TF_ECN_PERMIT) { 1559 if (thflags & TH_CWR) 1560 tp->t_flags &= ~TF_ECN_SND_ECE; 1561 switch (iptos & IPTOS_ECN_MASK) { 1562 case IPTOS_ECN_CE: 1563 tp->t_flags |= TF_ECN_SND_ECE; 1564 TCPSTAT_INC(tcps_ecn_ce); 1565 break; 1566 case IPTOS_ECN_ECT0: 1567 TCPSTAT_INC(tcps_ecn_ect0); 1568 break; 1569 case IPTOS_ECN_ECT1: 1570 TCPSTAT_INC(tcps_ecn_ect1); 1571 break; 1572 } 1573 1574 /* Process a packet differently from RFC3168. */ 1575 cc_ecnpkt_handler(tp, th, iptos); 1576 1577 /* Congestion experienced. */ 1578 if (thflags & TH_ECE) { 1579 cc_cong_signal(tp, th, CC_ECN); 1580 } 1581 } 1582 1583 /* 1584 * Parse options on any incoming segment. 1585 */ 1586 tcp_dooptions(&to, (u_char *)(th + 1), 1587 (th->th_off << 2) - sizeof(struct tcphdr), 1588 (thflags & TH_SYN) ? TO_SYN : 0); 1589 1590 /* 1591 * If echoed timestamp is later than the current time, 1592 * fall back to non RFC1323 RTT calculation. Normalize 1593 * timestamp if syncookies were used when this connection 1594 * was established. 1595 */ 1596 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1597 to.to_tsecr -= tp->ts_offset; 1598 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1599 to.to_tsecr = 0; 1600 } 1601 /* 1602 * If timestamps were negotiated during SYN/ACK they should 1603 * appear on every segment during this session and vice versa. 1604 */ 1605 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1606 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1607 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1608 "no action\n", s, __func__); 1609 free(s, M_TCPLOG); 1610 } 1611 } 1612 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1613 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1614 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1615 "no action\n", s, __func__); 1616 free(s, M_TCPLOG); 1617 } 1618 } 1619 1620 /* 1621 * Process options only when we get SYN/ACK back. The SYN case 1622 * for incoming connections is handled in tcp_syncache. 1623 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1624 * or <SYN,ACK>) segment itself is never scaled. 1625 * XXX this is traditional behavior, may need to be cleaned up. 1626 */ 1627 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1628 if ((to.to_flags & TOF_SCALE) && 1629 (tp->t_flags & TF_REQ_SCALE)) { 1630 tp->t_flags |= TF_RCVD_SCALE; 1631 tp->snd_scale = to.to_wscale; 1632 } 1633 /* 1634 * Initial send window. It will be updated with 1635 * the next incoming segment to the scaled value. 1636 */ 1637 tp->snd_wnd = th->th_win; 1638 if (to.to_flags & TOF_TS) { 1639 tp->t_flags |= TF_RCVD_TSTMP; 1640 tp->ts_recent = to.to_tsval; 1641 tp->ts_recent_age = tcp_ts_getticks(); 1642 } 1643 if (to.to_flags & TOF_MSS) 1644 tcp_mss(tp, to.to_mss); 1645 if ((tp->t_flags & TF_SACK_PERMIT) && 1646 (to.to_flags & TOF_SACKPERM) == 0) 1647 tp->t_flags &= ~TF_SACK_PERMIT; 1648 } 1649 1650 /* 1651 * Header prediction: check for the two common cases 1652 * of a uni-directional data xfer. If the packet has 1653 * no control flags, is in-sequence, the window didn't 1654 * change and we're not retransmitting, it's a 1655 * candidate. If the length is zero and the ack moved 1656 * forward, we're the sender side of the xfer. Just 1657 * free the data acked & wake any higher level process 1658 * that was blocked waiting for space. If the length 1659 * is non-zero and the ack didn't move, we're the 1660 * receiver side. If we're getting packets in-order 1661 * (the reassembly queue is empty), add the data to 1662 * the socket buffer and note that we need a delayed ack. 1663 * Make sure that the hidden state-flags are also off. 1664 * Since we check for TCPS_ESTABLISHED first, it can only 1665 * be TH_NEEDSYN. 1666 */ 1667 if (tp->t_state == TCPS_ESTABLISHED && 1668 th->th_seq == tp->rcv_nxt && 1669 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1670 tp->snd_nxt == tp->snd_max && 1671 tiwin && tiwin == tp->snd_wnd && 1672 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1673 LIST_EMPTY(&tp->t_segq) && 1674 ((to.to_flags & TOF_TS) == 0 || 1675 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1676 1677 /* 1678 * If last ACK falls within this segment's sequence numbers, 1679 * record the timestamp. 1680 * NOTE that the test is modified according to the latest 1681 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1682 */ 1683 if ((to.to_flags & TOF_TS) != 0 && 1684 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1685 tp->ts_recent_age = tcp_ts_getticks(); 1686 tp->ts_recent = to.to_tsval; 1687 } 1688 1689 if (tlen == 0) { 1690 if (SEQ_GT(th->th_ack, tp->snd_una) && 1691 SEQ_LEQ(th->th_ack, tp->snd_max) && 1692 !IN_RECOVERY(tp->t_flags) && 1693 (to.to_flags & TOF_SACK) == 0 && 1694 TAILQ_EMPTY(&tp->snd_holes)) { 1695 /* 1696 * This is a pure ack for outstanding data. 1697 */ 1698 if (ti_locked == TI_RLOCKED) 1699 INP_INFO_RUNLOCK(&V_tcbinfo); 1700 ti_locked = TI_UNLOCKED; 1701 1702 TCPSTAT_INC(tcps_predack); 1703 1704 /* 1705 * "bad retransmit" recovery. 1706 */ 1707 if (tp->t_rxtshift == 1 && 1708 tp->t_flags & TF_PREVVALID && 1709 (int)(ticks - tp->t_badrxtwin) < 0) { 1710 cc_cong_signal(tp, th, CC_RTO_ERR); 1711 } 1712 1713 /* 1714 * Recalculate the transmit timer / rtt. 1715 * 1716 * Some boxes send broken timestamp replies 1717 * during the SYN+ACK phase, ignore 1718 * timestamps of 0 or we could calculate a 1719 * huge RTT and blow up the retransmit timer. 1720 */ 1721 if ((to.to_flags & TOF_TS) != 0 && 1722 to.to_tsecr) { 1723 u_int t; 1724 1725 t = tcp_ts_getticks() - to.to_tsecr; 1726 if (!tp->t_rttlow || tp->t_rttlow > t) 1727 tp->t_rttlow = t; 1728 tcp_xmit_timer(tp, 1729 TCP_TS_TO_TICKS(t) + 1); 1730 } else if (tp->t_rtttime && 1731 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1732 if (!tp->t_rttlow || 1733 tp->t_rttlow > ticks - tp->t_rtttime) 1734 tp->t_rttlow = ticks - tp->t_rtttime; 1735 tcp_xmit_timer(tp, 1736 ticks - tp->t_rtttime); 1737 } 1738 acked = BYTES_THIS_ACK(tp, th); 1739 1740 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1741 hhook_run_tcp_est_in(tp, th, &to); 1742 1743 TCPSTAT_INC(tcps_rcvackpack); 1744 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1745 sbdrop(&so->so_snd, acked); 1746 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1747 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1748 tp->snd_recover = th->th_ack - 1; 1749 1750 /* 1751 * Let the congestion control algorithm update 1752 * congestion control related information. This 1753 * typically means increasing the congestion 1754 * window. 1755 */ 1756 cc_ack_received(tp, th, CC_ACK); 1757 1758 tp->snd_una = th->th_ack; 1759 /* 1760 * Pull snd_wl2 up to prevent seq wrap relative 1761 * to th_ack. 1762 */ 1763 tp->snd_wl2 = th->th_ack; 1764 tp->t_dupacks = 0; 1765 m_freem(m); 1766 ND6_HINT(tp); /* Some progress has been made. */ 1767 1768 /* 1769 * If all outstanding data are acked, stop 1770 * retransmit timer, otherwise restart timer 1771 * using current (possibly backed-off) value. 1772 * If process is waiting for space, 1773 * wakeup/selwakeup/signal. If data 1774 * are ready to send, let tcp_output 1775 * decide between more output or persist. 1776 */ 1777 #ifdef TCPDEBUG 1778 if (so->so_options & SO_DEBUG) 1779 tcp_trace(TA_INPUT, ostate, tp, 1780 (void *)tcp_saveipgen, 1781 &tcp_savetcp, 0); 1782 #endif 1783 TCP_PROBE3(debug__input, tp, th, 1784 mtod(m, const char *)); 1785 if (tp->snd_una == tp->snd_max) 1786 tcp_timer_activate(tp, TT_REXMT, 0); 1787 else if (!tcp_timer_active(tp, TT_PERSIST)) 1788 tcp_timer_activate(tp, TT_REXMT, 1789 tp->t_rxtcur); 1790 sowwakeup(so); 1791 if (sbavail(&so->so_snd)) 1792 (void) tcp_output(tp); 1793 goto check_delack; 1794 } 1795 } else if (th->th_ack == tp->snd_una && 1796 tlen <= sbspace(&so->so_rcv)) { 1797 int newsize = 0; /* automatic sockbuf scaling */ 1798 1799 /* 1800 * This is a pure, in-sequence data packet with 1801 * nothing on the reassembly queue and we have enough 1802 * buffer space to take it. 1803 */ 1804 if (ti_locked == TI_RLOCKED) 1805 INP_INFO_RUNLOCK(&V_tcbinfo); 1806 ti_locked = TI_UNLOCKED; 1807 1808 /* Clean receiver SACK report if present */ 1809 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1810 tcp_clean_sackreport(tp); 1811 TCPSTAT_INC(tcps_preddat); 1812 tp->rcv_nxt += tlen; 1813 /* 1814 * Pull snd_wl1 up to prevent seq wrap relative to 1815 * th_seq. 1816 */ 1817 tp->snd_wl1 = th->th_seq; 1818 /* 1819 * Pull rcv_up up to prevent seq wrap relative to 1820 * rcv_nxt. 1821 */ 1822 tp->rcv_up = tp->rcv_nxt; 1823 TCPSTAT_INC(tcps_rcvpack); 1824 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1825 ND6_HINT(tp); /* Some progress has been made */ 1826 #ifdef TCPDEBUG 1827 if (so->so_options & SO_DEBUG) 1828 tcp_trace(TA_INPUT, ostate, tp, 1829 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1830 #endif 1831 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 1832 1833 /* 1834 * Automatic sizing of receive socket buffer. Often the send 1835 * buffer size is not optimally adjusted to the actual network 1836 * conditions at hand (delay bandwidth product). Setting the 1837 * buffer size too small limits throughput on links with high 1838 * bandwidth and high delay (eg. trans-continental/oceanic links). 1839 * 1840 * On the receive side the socket buffer memory is only rarely 1841 * used to any significant extent. This allows us to be much 1842 * more aggressive in scaling the receive socket buffer. For 1843 * the case that the buffer space is actually used to a large 1844 * extent and we run out of kernel memory we can simply drop 1845 * the new segments; TCP on the sender will just retransmit it 1846 * later. Setting the buffer size too big may only consume too 1847 * much kernel memory if the application doesn't read() from 1848 * the socket or packet loss or reordering makes use of the 1849 * reassembly queue. 1850 * 1851 * The criteria to step up the receive buffer one notch are: 1852 * 1. Application has not set receive buffer size with 1853 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1854 * 2. the number of bytes received during the time it takes 1855 * one timestamp to be reflected back to us (the RTT); 1856 * 3. received bytes per RTT is within seven eighth of the 1857 * current socket buffer size; 1858 * 4. receive buffer size has not hit maximal automatic size; 1859 * 1860 * This algorithm does one step per RTT at most and only if 1861 * we receive a bulk stream w/o packet losses or reorderings. 1862 * Shrinking the buffer during idle times is not necessary as 1863 * it doesn't consume any memory when idle. 1864 * 1865 * TODO: Only step up if the application is actually serving 1866 * the buffer to better manage the socket buffer resources. 1867 */ 1868 if (V_tcp_do_autorcvbuf && 1869 (to.to_flags & TOF_TS) && 1870 to.to_tsecr && 1871 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1872 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1873 to.to_tsecr - tp->rfbuf_ts < hz) { 1874 if (tp->rfbuf_cnt > 1875 (so->so_rcv.sb_hiwat / 8 * 7) && 1876 so->so_rcv.sb_hiwat < 1877 V_tcp_autorcvbuf_max) { 1878 newsize = 1879 min(so->so_rcv.sb_hiwat + 1880 V_tcp_autorcvbuf_inc, 1881 V_tcp_autorcvbuf_max); 1882 } 1883 /* Start over with next RTT. */ 1884 tp->rfbuf_ts = 0; 1885 tp->rfbuf_cnt = 0; 1886 } else 1887 tp->rfbuf_cnt += tlen; /* add up */ 1888 } 1889 1890 /* Add data to socket buffer. */ 1891 SOCKBUF_LOCK(&so->so_rcv); 1892 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1893 m_freem(m); 1894 } else { 1895 /* 1896 * Set new socket buffer size. 1897 * Give up when limit is reached. 1898 */ 1899 if (newsize) 1900 if (!sbreserve_locked(&so->so_rcv, 1901 newsize, so, NULL)) 1902 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1903 m_adj(m, drop_hdrlen); /* delayed header drop */ 1904 sbappendstream_locked(&so->so_rcv, m, 0); 1905 } 1906 /* NB: sorwakeup_locked() does an implicit unlock. */ 1907 sorwakeup_locked(so); 1908 if (DELAY_ACK(tp, tlen)) { 1909 tp->t_flags |= TF_DELACK; 1910 } else { 1911 tp->t_flags |= TF_ACKNOW; 1912 tcp_output(tp); 1913 } 1914 goto check_delack; 1915 } 1916 } 1917 1918 /* 1919 * Calculate amount of space in receive window, 1920 * and then do TCP input processing. 1921 * Receive window is amount of space in rcv queue, 1922 * but not less than advertised window. 1923 */ 1924 win = sbspace(&so->so_rcv); 1925 if (win < 0) 1926 win = 0; 1927 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1928 1929 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1930 tp->rfbuf_ts = 0; 1931 tp->rfbuf_cnt = 0; 1932 1933 switch (tp->t_state) { 1934 1935 /* 1936 * If the state is SYN_RECEIVED: 1937 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1938 */ 1939 case TCPS_SYN_RECEIVED: 1940 if ((thflags & TH_ACK) && 1941 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1942 SEQ_GT(th->th_ack, tp->snd_max))) { 1943 rstreason = BANDLIM_RST_OPENPORT; 1944 goto dropwithreset; 1945 } 1946 break; 1947 1948 /* 1949 * If the state is SYN_SENT: 1950 * if seg contains an ACK, but not for our SYN, drop the input. 1951 * if seg contains a RST, then drop the connection. 1952 * if seg does not contain SYN, then drop it. 1953 * Otherwise this is an acceptable SYN segment 1954 * initialize tp->rcv_nxt and tp->irs 1955 * if seg contains ack then advance tp->snd_una 1956 * if seg contains an ECE and ECN support is enabled, the stream 1957 * is ECN capable. 1958 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1959 * arrange for segment to be acked (eventually) 1960 * continue processing rest of data/controls, beginning with URG 1961 */ 1962 case TCPS_SYN_SENT: 1963 if ((thflags & TH_ACK) && 1964 (SEQ_LEQ(th->th_ack, tp->iss) || 1965 SEQ_GT(th->th_ack, tp->snd_max))) { 1966 rstreason = BANDLIM_UNLIMITED; 1967 goto dropwithreset; 1968 } 1969 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1970 TCP_PROBE5(connect__refused, NULL, tp, 1971 mtod(m, const char *), tp, th); 1972 tp = tcp_drop(tp, ECONNREFUSED); 1973 } 1974 if (thflags & TH_RST) 1975 goto drop; 1976 if (!(thflags & TH_SYN)) 1977 goto drop; 1978 1979 tp->irs = th->th_seq; 1980 tcp_rcvseqinit(tp); 1981 if (thflags & TH_ACK) { 1982 TCPSTAT_INC(tcps_connects); 1983 soisconnected(so); 1984 #ifdef MAC 1985 mac_socketpeer_set_from_mbuf(m, so); 1986 #endif 1987 /* Do window scaling on this connection? */ 1988 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1989 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1990 tp->rcv_scale = tp->request_r_scale; 1991 } 1992 tp->rcv_adv += imin(tp->rcv_wnd, 1993 TCP_MAXWIN << tp->rcv_scale); 1994 tp->snd_una++; /* SYN is acked */ 1995 /* 1996 * If there's data, delay ACK; if there's also a FIN 1997 * ACKNOW will be turned on later. 1998 */ 1999 if (DELAY_ACK(tp, tlen) && tlen != 0) 2000 tcp_timer_activate(tp, TT_DELACK, 2001 tcp_delacktime); 2002 else 2003 tp->t_flags |= TF_ACKNOW; 2004 2005 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2006 tp->t_flags |= TF_ECN_PERMIT; 2007 TCPSTAT_INC(tcps_ecn_shs); 2008 } 2009 2010 /* 2011 * Received <SYN,ACK> in SYN_SENT[*] state. 2012 * Transitions: 2013 * SYN_SENT --> ESTABLISHED 2014 * SYN_SENT* --> FIN_WAIT_1 2015 */ 2016 tp->t_starttime = ticks; 2017 if (tp->t_flags & TF_NEEDFIN) { 2018 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2019 tp->t_flags &= ~TF_NEEDFIN; 2020 thflags &= ~TH_SYN; 2021 } else { 2022 tcp_state_change(tp, TCPS_ESTABLISHED); 2023 TCP_PROBE5(connect__established, NULL, tp, 2024 mtod(m, const char *), tp, th); 2025 cc_conn_init(tp); 2026 tcp_timer_activate(tp, TT_KEEP, 2027 TP_KEEPIDLE(tp)); 2028 } 2029 } else { 2030 /* 2031 * Received initial SYN in SYN-SENT[*] state => 2032 * simultaneous open. 2033 * If it succeeds, connection is * half-synchronized. 2034 * Otherwise, do 3-way handshake: 2035 * SYN-SENT -> SYN-RECEIVED 2036 * SYN-SENT* -> SYN-RECEIVED* 2037 */ 2038 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2039 tcp_timer_activate(tp, TT_REXMT, 0); 2040 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2041 } 2042 2043 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2044 "ti_locked %d", __func__, ti_locked)); 2045 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2046 INP_WLOCK_ASSERT(tp->t_inpcb); 2047 2048 /* 2049 * Advance th->th_seq to correspond to first data byte. 2050 * If data, trim to stay within window, 2051 * dropping FIN if necessary. 2052 */ 2053 th->th_seq++; 2054 if (tlen > tp->rcv_wnd) { 2055 todrop = tlen - tp->rcv_wnd; 2056 m_adj(m, -todrop); 2057 tlen = tp->rcv_wnd; 2058 thflags &= ~TH_FIN; 2059 TCPSTAT_INC(tcps_rcvpackafterwin); 2060 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2061 } 2062 tp->snd_wl1 = th->th_seq - 1; 2063 tp->rcv_up = th->th_seq; 2064 /* 2065 * Client side of transaction: already sent SYN and data. 2066 * If the remote host used T/TCP to validate the SYN, 2067 * our data will be ACK'd; if so, enter normal data segment 2068 * processing in the middle of step 5, ack processing. 2069 * Otherwise, goto step 6. 2070 */ 2071 if (thflags & TH_ACK) 2072 goto process_ACK; 2073 2074 goto step6; 2075 2076 /* 2077 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2078 * do normal processing. 2079 * 2080 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2081 */ 2082 case TCPS_LAST_ACK: 2083 case TCPS_CLOSING: 2084 break; /* continue normal processing */ 2085 } 2086 2087 /* 2088 * States other than LISTEN or SYN_SENT. 2089 * First check the RST flag and sequence number since reset segments 2090 * are exempt from the timestamp and connection count tests. This 2091 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2092 * below which allowed reset segments in half the sequence space 2093 * to fall though and be processed (which gives forged reset 2094 * segments with a random sequence number a 50 percent chance of 2095 * killing a connection). 2096 * Then check timestamp, if present. 2097 * Then check the connection count, if present. 2098 * Then check that at least some bytes of segment are within 2099 * receive window. If segment begins before rcv_nxt, 2100 * drop leading data (and SYN); if nothing left, just ack. 2101 */ 2102 if (thflags & TH_RST) { 2103 /* 2104 * RFC5961 Section 3.2 2105 * 2106 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2107 * - If RST is in window, we send challenge ACK. 2108 * 2109 * Note: to take into account delayed ACKs, we should 2110 * test against last_ack_sent instead of rcv_nxt. 2111 * Note 2: we handle special case of closed window, not 2112 * covered by the RFC. 2113 */ 2114 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2115 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2116 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2117 2118 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2119 KASSERT(ti_locked == TI_RLOCKED, 2120 ("%s: TH_RST ti_locked %d, th %p tp %p", 2121 __func__, ti_locked, th, tp)); 2122 KASSERT(tp->t_state != TCPS_SYN_SENT, 2123 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2124 __func__, th, tp)); 2125 2126 if (V_tcp_insecure_rst || 2127 tp->last_ack_sent == th->th_seq) { 2128 TCPSTAT_INC(tcps_drops); 2129 /* Drop the connection. */ 2130 switch (tp->t_state) { 2131 case TCPS_SYN_RECEIVED: 2132 so->so_error = ECONNREFUSED; 2133 goto close; 2134 case TCPS_ESTABLISHED: 2135 case TCPS_FIN_WAIT_1: 2136 case TCPS_FIN_WAIT_2: 2137 case TCPS_CLOSE_WAIT: 2138 so->so_error = ECONNRESET; 2139 close: 2140 tcp_state_change(tp, TCPS_CLOSED); 2141 /* FALLTHROUGH */ 2142 default: 2143 tp = tcp_close(tp); 2144 } 2145 } else { 2146 TCPSTAT_INC(tcps_badrst); 2147 /* Send challenge ACK. */ 2148 tcp_respond(tp, mtod(m, void *), th, m, 2149 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2150 tp->last_ack_sent = tp->rcv_nxt; 2151 m = NULL; 2152 } 2153 } 2154 goto drop; 2155 } 2156 2157 /* 2158 * RFC5961 Section 4.2 2159 * Send challenge ACK for any SYN in synchronized state. 2160 */ 2161 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT) { 2162 KASSERT(ti_locked == TI_RLOCKED, 2163 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2164 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2165 2166 TCPSTAT_INC(tcps_badsyn); 2167 if (V_tcp_insecure_syn && 2168 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2169 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2170 tp = tcp_drop(tp, ECONNRESET); 2171 rstreason = BANDLIM_UNLIMITED; 2172 } else { 2173 /* Send challenge ACK. */ 2174 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2175 tp->snd_nxt, TH_ACK); 2176 tp->last_ack_sent = tp->rcv_nxt; 2177 m = NULL; 2178 } 2179 goto drop; 2180 } 2181 2182 /* 2183 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2184 * and it's less than ts_recent, drop it. 2185 */ 2186 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2187 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2188 2189 /* Check to see if ts_recent is over 24 days old. */ 2190 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2191 /* 2192 * Invalidate ts_recent. If this segment updates 2193 * ts_recent, the age will be reset later and ts_recent 2194 * will get a valid value. If it does not, setting 2195 * ts_recent to zero will at least satisfy the 2196 * requirement that zero be placed in the timestamp 2197 * echo reply when ts_recent isn't valid. The 2198 * age isn't reset until we get a valid ts_recent 2199 * because we don't want out-of-order segments to be 2200 * dropped when ts_recent is old. 2201 */ 2202 tp->ts_recent = 0; 2203 } else { 2204 TCPSTAT_INC(tcps_rcvduppack); 2205 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2206 TCPSTAT_INC(tcps_pawsdrop); 2207 if (tlen) 2208 goto dropafterack; 2209 goto drop; 2210 } 2211 } 2212 2213 /* 2214 * In the SYN-RECEIVED state, validate that the packet belongs to 2215 * this connection before trimming the data to fit the receive 2216 * window. Check the sequence number versus IRS since we know 2217 * the sequence numbers haven't wrapped. This is a partial fix 2218 * for the "LAND" DoS attack. 2219 */ 2220 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2221 rstreason = BANDLIM_RST_OPENPORT; 2222 goto dropwithreset; 2223 } 2224 2225 todrop = tp->rcv_nxt - th->th_seq; 2226 if (todrop > 0) { 2227 if (thflags & TH_SYN) { 2228 thflags &= ~TH_SYN; 2229 th->th_seq++; 2230 if (th->th_urp > 1) 2231 th->th_urp--; 2232 else 2233 thflags &= ~TH_URG; 2234 todrop--; 2235 } 2236 /* 2237 * Following if statement from Stevens, vol. 2, p. 960. 2238 */ 2239 if (todrop > tlen 2240 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2241 /* 2242 * Any valid FIN must be to the left of the window. 2243 * At this point the FIN must be a duplicate or out 2244 * of sequence; drop it. 2245 */ 2246 thflags &= ~TH_FIN; 2247 2248 /* 2249 * Send an ACK to resynchronize and drop any data. 2250 * But keep on processing for RST or ACK. 2251 */ 2252 tp->t_flags |= TF_ACKNOW; 2253 todrop = tlen; 2254 TCPSTAT_INC(tcps_rcvduppack); 2255 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2256 } else { 2257 TCPSTAT_INC(tcps_rcvpartduppack); 2258 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2259 } 2260 drop_hdrlen += todrop; /* drop from the top afterwards */ 2261 th->th_seq += todrop; 2262 tlen -= todrop; 2263 if (th->th_urp > todrop) 2264 th->th_urp -= todrop; 2265 else { 2266 thflags &= ~TH_URG; 2267 th->th_urp = 0; 2268 } 2269 } 2270 2271 /* 2272 * If new data are received on a connection after the 2273 * user processes are gone, then RST the other end. 2274 */ 2275 if ((so->so_state & SS_NOFDREF) && 2276 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2277 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2278 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2279 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2280 2281 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2282 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2283 "after socket was closed, " 2284 "sending RST and removing tcpcb\n", 2285 s, __func__, tcpstates[tp->t_state], tlen); 2286 free(s, M_TCPLOG); 2287 } 2288 tp = tcp_close(tp); 2289 TCPSTAT_INC(tcps_rcvafterclose); 2290 rstreason = BANDLIM_UNLIMITED; 2291 goto dropwithreset; 2292 } 2293 2294 /* 2295 * If segment ends after window, drop trailing data 2296 * (and PUSH and FIN); if nothing left, just ACK. 2297 */ 2298 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2299 if (todrop > 0) { 2300 TCPSTAT_INC(tcps_rcvpackafterwin); 2301 if (todrop >= tlen) { 2302 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2303 /* 2304 * If window is closed can only take segments at 2305 * window edge, and have to drop data and PUSH from 2306 * incoming segments. Continue processing, but 2307 * remember to ack. Otherwise, drop segment 2308 * and ack. 2309 */ 2310 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2311 tp->t_flags |= TF_ACKNOW; 2312 TCPSTAT_INC(tcps_rcvwinprobe); 2313 } else 2314 goto dropafterack; 2315 } else 2316 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2317 m_adj(m, -todrop); 2318 tlen -= todrop; 2319 thflags &= ~(TH_PUSH|TH_FIN); 2320 } 2321 2322 /* 2323 * If last ACK falls within this segment's sequence numbers, 2324 * record its timestamp. 2325 * NOTE: 2326 * 1) That the test incorporates suggestions from the latest 2327 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2328 * 2) That updating only on newer timestamps interferes with 2329 * our earlier PAWS tests, so this check should be solely 2330 * predicated on the sequence space of this segment. 2331 * 3) That we modify the segment boundary check to be 2332 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2333 * instead of RFC1323's 2334 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2335 * This modified check allows us to overcome RFC1323's 2336 * limitations as described in Stevens TCP/IP Illustrated 2337 * Vol. 2 p.869. In such cases, we can still calculate the 2338 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2339 */ 2340 if ((to.to_flags & TOF_TS) != 0 && 2341 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2342 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2343 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2344 tp->ts_recent_age = tcp_ts_getticks(); 2345 tp->ts_recent = to.to_tsval; 2346 } 2347 2348 /* 2349 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2350 * flag is on (half-synchronized state), then queue data for 2351 * later processing; else drop segment and return. 2352 */ 2353 if ((thflags & TH_ACK) == 0) { 2354 if (tp->t_state == TCPS_SYN_RECEIVED || 2355 (tp->t_flags & TF_NEEDSYN)) 2356 goto step6; 2357 else if (tp->t_flags & TF_ACKNOW) 2358 goto dropafterack; 2359 else 2360 goto drop; 2361 } 2362 2363 /* 2364 * Ack processing. 2365 */ 2366 switch (tp->t_state) { 2367 2368 /* 2369 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2370 * ESTABLISHED state and continue processing. 2371 * The ACK was checked above. 2372 */ 2373 case TCPS_SYN_RECEIVED: 2374 2375 TCPSTAT_INC(tcps_connects); 2376 soisconnected(so); 2377 /* Do window scaling? */ 2378 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2379 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2380 tp->rcv_scale = tp->request_r_scale; 2381 tp->snd_wnd = tiwin; 2382 } 2383 /* 2384 * Make transitions: 2385 * SYN-RECEIVED -> ESTABLISHED 2386 * SYN-RECEIVED* -> FIN-WAIT-1 2387 */ 2388 tp->t_starttime = ticks; 2389 if (tp->t_flags & TF_NEEDFIN) { 2390 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2391 tp->t_flags &= ~TF_NEEDFIN; 2392 } else { 2393 tcp_state_change(tp, TCPS_ESTABLISHED); 2394 TCP_PROBE5(accept__established, NULL, tp, 2395 mtod(m, const char *), tp, th); 2396 cc_conn_init(tp); 2397 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2398 } 2399 /* 2400 * If segment contains data or ACK, will call tcp_reass() 2401 * later; if not, do so now to pass queued data to user. 2402 */ 2403 if (tlen == 0 && (thflags & TH_FIN) == 0) 2404 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2405 (struct mbuf *)0); 2406 tp->snd_wl1 = th->th_seq - 1; 2407 /* FALLTHROUGH */ 2408 2409 /* 2410 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2411 * ACKs. If the ack is in the range 2412 * tp->snd_una < th->th_ack <= tp->snd_max 2413 * then advance tp->snd_una to th->th_ack and drop 2414 * data from the retransmission queue. If this ACK reflects 2415 * more up to date window information we update our window information. 2416 */ 2417 case TCPS_ESTABLISHED: 2418 case TCPS_FIN_WAIT_1: 2419 case TCPS_FIN_WAIT_2: 2420 case TCPS_CLOSE_WAIT: 2421 case TCPS_CLOSING: 2422 case TCPS_LAST_ACK: 2423 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2424 TCPSTAT_INC(tcps_rcvacktoomuch); 2425 goto dropafterack; 2426 } 2427 if ((tp->t_flags & TF_SACK_PERMIT) && 2428 ((to.to_flags & TOF_SACK) || 2429 !TAILQ_EMPTY(&tp->snd_holes))) 2430 tcp_sack_doack(tp, &to, th->th_ack); 2431 2432 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2433 hhook_run_tcp_est_in(tp, th, &to); 2434 2435 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2436 if (tlen == 0 && tiwin == tp->snd_wnd) { 2437 /* 2438 * If this is the first time we've seen a 2439 * FIN from the remote, this is not a 2440 * duplicate and it needs to be processed 2441 * normally. This happens during a 2442 * simultaneous close. 2443 */ 2444 if ((thflags & TH_FIN) && 2445 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2446 tp->t_dupacks = 0; 2447 break; 2448 } 2449 TCPSTAT_INC(tcps_rcvdupack); 2450 /* 2451 * If we have outstanding data (other than 2452 * a window probe), this is a completely 2453 * duplicate ack (ie, window info didn't 2454 * change and FIN isn't set), 2455 * the ack is the biggest we've 2456 * seen and we've seen exactly our rexmt 2457 * threshhold of them, assume a packet 2458 * has been dropped and retransmit it. 2459 * Kludge snd_nxt & the congestion 2460 * window so we send only this one 2461 * packet. 2462 * 2463 * We know we're losing at the current 2464 * window size so do congestion avoidance 2465 * (set ssthresh to half the current window 2466 * and pull our congestion window back to 2467 * the new ssthresh). 2468 * 2469 * Dup acks mean that packets have left the 2470 * network (they're now cached at the receiver) 2471 * so bump cwnd by the amount in the receiver 2472 * to keep a constant cwnd packets in the 2473 * network. 2474 * 2475 * When using TCP ECN, notify the peer that 2476 * we reduced the cwnd. 2477 */ 2478 if (!tcp_timer_active(tp, TT_REXMT) || 2479 th->th_ack != tp->snd_una) 2480 tp->t_dupacks = 0; 2481 else if (++tp->t_dupacks > tcprexmtthresh || 2482 IN_FASTRECOVERY(tp->t_flags)) { 2483 cc_ack_received(tp, th, CC_DUPACK); 2484 if ((tp->t_flags & TF_SACK_PERMIT) && 2485 IN_FASTRECOVERY(tp->t_flags)) { 2486 int awnd; 2487 2488 /* 2489 * Compute the amount of data in flight first. 2490 * We can inject new data into the pipe iff 2491 * we have less than 1/2 the original window's 2492 * worth of data in flight. 2493 */ 2494 awnd = (tp->snd_nxt - tp->snd_fack) + 2495 tp->sackhint.sack_bytes_rexmit; 2496 if (awnd < tp->snd_ssthresh) { 2497 tp->snd_cwnd += tp->t_maxseg; 2498 if (tp->snd_cwnd > tp->snd_ssthresh) 2499 tp->snd_cwnd = tp->snd_ssthresh; 2500 } 2501 } else 2502 tp->snd_cwnd += tp->t_maxseg; 2503 (void) tcp_output(tp); 2504 goto drop; 2505 } else if (tp->t_dupacks == tcprexmtthresh) { 2506 tcp_seq onxt = tp->snd_nxt; 2507 2508 /* 2509 * If we're doing sack, check to 2510 * see if we're already in sack 2511 * recovery. If we're not doing sack, 2512 * check to see if we're in newreno 2513 * recovery. 2514 */ 2515 if (tp->t_flags & TF_SACK_PERMIT) { 2516 if (IN_FASTRECOVERY(tp->t_flags)) { 2517 tp->t_dupacks = 0; 2518 break; 2519 } 2520 } else { 2521 if (SEQ_LEQ(th->th_ack, 2522 tp->snd_recover)) { 2523 tp->t_dupacks = 0; 2524 break; 2525 } 2526 } 2527 /* Congestion signal before ack. */ 2528 cc_cong_signal(tp, th, CC_NDUPACK); 2529 cc_ack_received(tp, th, CC_DUPACK); 2530 tcp_timer_activate(tp, TT_REXMT, 0); 2531 tp->t_rtttime = 0; 2532 if (tp->t_flags & TF_SACK_PERMIT) { 2533 TCPSTAT_INC( 2534 tcps_sack_recovery_episode); 2535 tp->sack_newdata = tp->snd_nxt; 2536 tp->snd_cwnd = tp->t_maxseg; 2537 (void) tcp_output(tp); 2538 goto drop; 2539 } 2540 tp->snd_nxt = th->th_ack; 2541 tp->snd_cwnd = tp->t_maxseg; 2542 (void) tcp_output(tp); 2543 KASSERT(tp->snd_limited <= 2, 2544 ("%s: tp->snd_limited too big", 2545 __func__)); 2546 tp->snd_cwnd = tp->snd_ssthresh + 2547 tp->t_maxseg * 2548 (tp->t_dupacks - tp->snd_limited); 2549 if (SEQ_GT(onxt, tp->snd_nxt)) 2550 tp->snd_nxt = onxt; 2551 goto drop; 2552 } else if (V_tcp_do_rfc3042) { 2553 cc_ack_received(tp, th, CC_DUPACK); 2554 u_long oldcwnd = tp->snd_cwnd; 2555 tcp_seq oldsndmax = tp->snd_max; 2556 u_int sent; 2557 int avail; 2558 2559 KASSERT(tp->t_dupacks == 1 || 2560 tp->t_dupacks == 2, 2561 ("%s: dupacks not 1 or 2", 2562 __func__)); 2563 if (tp->t_dupacks == 1) 2564 tp->snd_limited = 0; 2565 tp->snd_cwnd = 2566 (tp->snd_nxt - tp->snd_una) + 2567 (tp->t_dupacks - tp->snd_limited) * 2568 tp->t_maxseg; 2569 /* 2570 * Only call tcp_output when there 2571 * is new data available to be sent. 2572 * Otherwise we would send pure ACKs. 2573 */ 2574 SOCKBUF_LOCK(&so->so_snd); 2575 avail = sbavail(&so->so_snd) - 2576 (tp->snd_nxt - tp->snd_una); 2577 SOCKBUF_UNLOCK(&so->so_snd); 2578 if (avail > 0) 2579 (void) tcp_output(tp); 2580 sent = tp->snd_max - oldsndmax; 2581 if (sent > tp->t_maxseg) { 2582 KASSERT((tp->t_dupacks == 2 && 2583 tp->snd_limited == 0) || 2584 (sent == tp->t_maxseg + 1 && 2585 tp->t_flags & TF_SENTFIN), 2586 ("%s: sent too much", 2587 __func__)); 2588 tp->snd_limited = 2; 2589 } else if (sent > 0) 2590 ++tp->snd_limited; 2591 tp->snd_cwnd = oldcwnd; 2592 goto drop; 2593 } 2594 } else 2595 tp->t_dupacks = 0; 2596 break; 2597 } 2598 2599 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2600 ("%s: th_ack <= snd_una", __func__)); 2601 2602 /* 2603 * If the congestion window was inflated to account 2604 * for the other side's cached packets, retract it. 2605 */ 2606 if (IN_FASTRECOVERY(tp->t_flags)) { 2607 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2608 if (tp->t_flags & TF_SACK_PERMIT) 2609 tcp_sack_partialack(tp, th); 2610 else 2611 tcp_newreno_partial_ack(tp, th); 2612 } else 2613 cc_post_recovery(tp, th); 2614 } 2615 tp->t_dupacks = 0; 2616 /* 2617 * If we reach this point, ACK is not a duplicate, 2618 * i.e., it ACKs something we sent. 2619 */ 2620 if (tp->t_flags & TF_NEEDSYN) { 2621 /* 2622 * T/TCP: Connection was half-synchronized, and our 2623 * SYN has been ACK'd (so connection is now fully 2624 * synchronized). Go to non-starred state, 2625 * increment snd_una for ACK of SYN, and check if 2626 * we can do window scaling. 2627 */ 2628 tp->t_flags &= ~TF_NEEDSYN; 2629 tp->snd_una++; 2630 /* Do window scaling? */ 2631 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2632 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2633 tp->rcv_scale = tp->request_r_scale; 2634 /* Send window already scaled. */ 2635 } 2636 } 2637 2638 process_ACK: 2639 INP_WLOCK_ASSERT(tp->t_inpcb); 2640 2641 acked = BYTES_THIS_ACK(tp, th); 2642 TCPSTAT_INC(tcps_rcvackpack); 2643 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2644 2645 /* 2646 * If we just performed our first retransmit, and the ACK 2647 * arrives within our recovery window, then it was a mistake 2648 * to do the retransmit in the first place. Recover our 2649 * original cwnd and ssthresh, and proceed to transmit where 2650 * we left off. 2651 */ 2652 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2653 (int)(ticks - tp->t_badrxtwin) < 0) 2654 cc_cong_signal(tp, th, CC_RTO_ERR); 2655 2656 /* 2657 * If we have a timestamp reply, update smoothed 2658 * round trip time. If no timestamp is present but 2659 * transmit timer is running and timed sequence 2660 * number was acked, update smoothed round trip time. 2661 * Since we now have an rtt measurement, cancel the 2662 * timer backoff (cf., Phil Karn's retransmit alg.). 2663 * Recompute the initial retransmit timer. 2664 * 2665 * Some boxes send broken timestamp replies 2666 * during the SYN+ACK phase, ignore 2667 * timestamps of 0 or we could calculate a 2668 * huge RTT and blow up the retransmit timer. 2669 */ 2670 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2671 u_int t; 2672 2673 t = tcp_ts_getticks() - to.to_tsecr; 2674 if (!tp->t_rttlow || tp->t_rttlow > t) 2675 tp->t_rttlow = t; 2676 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2677 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2678 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2679 tp->t_rttlow = ticks - tp->t_rtttime; 2680 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2681 } 2682 2683 /* 2684 * If all outstanding data is acked, stop retransmit 2685 * timer and remember to restart (more output or persist). 2686 * If there is more data to be acked, restart retransmit 2687 * timer, using current (possibly backed-off) value. 2688 */ 2689 if (th->th_ack == tp->snd_max) { 2690 tcp_timer_activate(tp, TT_REXMT, 0); 2691 needoutput = 1; 2692 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2693 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2694 2695 /* 2696 * If no data (only SYN) was ACK'd, 2697 * skip rest of ACK processing. 2698 */ 2699 if (acked == 0) 2700 goto step6; 2701 2702 /* 2703 * Let the congestion control algorithm update congestion 2704 * control related information. This typically means increasing 2705 * the congestion window. 2706 */ 2707 cc_ack_received(tp, th, CC_ACK); 2708 2709 SOCKBUF_LOCK(&so->so_snd); 2710 if (acked > sbavail(&so->so_snd)) { 2711 tp->snd_wnd -= sbavail(&so->so_snd); 2712 mfree = sbcut_locked(&so->so_snd, 2713 (int)sbavail(&so->so_snd)); 2714 ourfinisacked = 1; 2715 } else { 2716 mfree = sbcut_locked(&so->so_snd, acked); 2717 tp->snd_wnd -= acked; 2718 ourfinisacked = 0; 2719 } 2720 /* NB: sowwakeup_locked() does an implicit unlock. */ 2721 sowwakeup_locked(so); 2722 m_freem(mfree); 2723 /* Detect una wraparound. */ 2724 if (!IN_RECOVERY(tp->t_flags) && 2725 SEQ_GT(tp->snd_una, tp->snd_recover) && 2726 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2727 tp->snd_recover = th->th_ack - 1; 2728 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2729 if (IN_RECOVERY(tp->t_flags) && 2730 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2731 EXIT_RECOVERY(tp->t_flags); 2732 } 2733 tp->snd_una = th->th_ack; 2734 if (tp->t_flags & TF_SACK_PERMIT) { 2735 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2736 tp->snd_recover = tp->snd_una; 2737 } 2738 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2739 tp->snd_nxt = tp->snd_una; 2740 2741 switch (tp->t_state) { 2742 2743 /* 2744 * In FIN_WAIT_1 STATE in addition to the processing 2745 * for the ESTABLISHED state if our FIN is now acknowledged 2746 * then enter FIN_WAIT_2. 2747 */ 2748 case TCPS_FIN_WAIT_1: 2749 if (ourfinisacked) { 2750 /* 2751 * If we can't receive any more 2752 * data, then closing user can proceed. 2753 * Starting the timer is contrary to the 2754 * specification, but if we don't get a FIN 2755 * we'll hang forever. 2756 * 2757 * XXXjl: 2758 * we should release the tp also, and use a 2759 * compressed state. 2760 */ 2761 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2762 soisdisconnected(so); 2763 tcp_timer_activate(tp, TT_2MSL, 2764 (tcp_fast_finwait2_recycle ? 2765 tcp_finwait2_timeout : 2766 TP_MAXIDLE(tp))); 2767 } 2768 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2769 } 2770 break; 2771 2772 /* 2773 * In CLOSING STATE in addition to the processing for 2774 * the ESTABLISHED state if the ACK acknowledges our FIN 2775 * then enter the TIME-WAIT state, otherwise ignore 2776 * the segment. 2777 */ 2778 case TCPS_CLOSING: 2779 if (ourfinisacked) { 2780 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2781 tcp_twstart(tp); 2782 INP_INFO_RUNLOCK(&V_tcbinfo); 2783 m_freem(m); 2784 return; 2785 } 2786 break; 2787 2788 /* 2789 * In LAST_ACK, we may still be waiting for data to drain 2790 * and/or to be acked, as well as for the ack of our FIN. 2791 * If our FIN is now acknowledged, delete the TCB, 2792 * enter the closed state and return. 2793 */ 2794 case TCPS_LAST_ACK: 2795 if (ourfinisacked) { 2796 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2797 tp = tcp_close(tp); 2798 goto drop; 2799 } 2800 break; 2801 } 2802 } 2803 2804 step6: 2805 INP_WLOCK_ASSERT(tp->t_inpcb); 2806 2807 /* 2808 * Update window information. 2809 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2810 */ 2811 if ((thflags & TH_ACK) && 2812 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2813 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2814 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2815 /* keep track of pure window updates */ 2816 if (tlen == 0 && 2817 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2818 TCPSTAT_INC(tcps_rcvwinupd); 2819 tp->snd_wnd = tiwin; 2820 tp->snd_wl1 = th->th_seq; 2821 tp->snd_wl2 = th->th_ack; 2822 if (tp->snd_wnd > tp->max_sndwnd) 2823 tp->max_sndwnd = tp->snd_wnd; 2824 needoutput = 1; 2825 } 2826 2827 /* 2828 * Process segments with URG. 2829 */ 2830 if ((thflags & TH_URG) && th->th_urp && 2831 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2832 /* 2833 * This is a kludge, but if we receive and accept 2834 * random urgent pointers, we'll crash in 2835 * soreceive. It's hard to imagine someone 2836 * actually wanting to send this much urgent data. 2837 */ 2838 SOCKBUF_LOCK(&so->so_rcv); 2839 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2840 th->th_urp = 0; /* XXX */ 2841 thflags &= ~TH_URG; /* XXX */ 2842 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2843 goto dodata; /* XXX */ 2844 } 2845 /* 2846 * If this segment advances the known urgent pointer, 2847 * then mark the data stream. This should not happen 2848 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2849 * a FIN has been received from the remote side. 2850 * In these states we ignore the URG. 2851 * 2852 * According to RFC961 (Assigned Protocols), 2853 * the urgent pointer points to the last octet 2854 * of urgent data. We continue, however, 2855 * to consider it to indicate the first octet 2856 * of data past the urgent section as the original 2857 * spec states (in one of two places). 2858 */ 2859 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2860 tp->rcv_up = th->th_seq + th->th_urp; 2861 so->so_oobmark = sbavail(&so->so_rcv) + 2862 (tp->rcv_up - tp->rcv_nxt) - 1; 2863 if (so->so_oobmark == 0) 2864 so->so_rcv.sb_state |= SBS_RCVATMARK; 2865 sohasoutofband(so); 2866 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2867 } 2868 SOCKBUF_UNLOCK(&so->so_rcv); 2869 /* 2870 * Remove out of band data so doesn't get presented to user. 2871 * This can happen independent of advancing the URG pointer, 2872 * but if two URG's are pending at once, some out-of-band 2873 * data may creep in... ick. 2874 */ 2875 if (th->th_urp <= (u_long)tlen && 2876 !(so->so_options & SO_OOBINLINE)) { 2877 /* hdr drop is delayed */ 2878 tcp_pulloutofband(so, th, m, drop_hdrlen); 2879 } 2880 } else { 2881 /* 2882 * If no out of band data is expected, 2883 * pull receive urgent pointer along 2884 * with the receive window. 2885 */ 2886 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2887 tp->rcv_up = tp->rcv_nxt; 2888 } 2889 dodata: /* XXX */ 2890 INP_WLOCK_ASSERT(tp->t_inpcb); 2891 2892 /* 2893 * Process the segment text, merging it into the TCP sequencing queue, 2894 * and arranging for acknowledgment of receipt if necessary. 2895 * This process logically involves adjusting tp->rcv_wnd as data 2896 * is presented to the user (this happens in tcp_usrreq.c, 2897 * case PRU_RCVD). If a FIN has already been received on this 2898 * connection then we just ignore the text. 2899 */ 2900 if ((tlen || (thflags & TH_FIN)) && 2901 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2902 tcp_seq save_start = th->th_seq; 2903 m_adj(m, drop_hdrlen); /* delayed header drop */ 2904 /* 2905 * Insert segment which includes th into TCP reassembly queue 2906 * with control block tp. Set thflags to whether reassembly now 2907 * includes a segment with FIN. This handles the common case 2908 * inline (segment is the next to be received on an established 2909 * connection, and the queue is empty), avoiding linkage into 2910 * and removal from the queue and repetition of various 2911 * conversions. 2912 * Set DELACK for segments received in order, but ack 2913 * immediately when segments are out of order (so 2914 * fast retransmit can work). 2915 */ 2916 if (th->th_seq == tp->rcv_nxt && 2917 LIST_EMPTY(&tp->t_segq) && 2918 TCPS_HAVEESTABLISHED(tp->t_state)) { 2919 if (DELAY_ACK(tp, tlen)) 2920 tp->t_flags |= TF_DELACK; 2921 else 2922 tp->t_flags |= TF_ACKNOW; 2923 tp->rcv_nxt += tlen; 2924 thflags = th->th_flags & TH_FIN; 2925 TCPSTAT_INC(tcps_rcvpack); 2926 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2927 ND6_HINT(tp); 2928 SOCKBUF_LOCK(&so->so_rcv); 2929 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2930 m_freem(m); 2931 else 2932 sbappendstream_locked(&so->so_rcv, m, 0); 2933 /* NB: sorwakeup_locked() does an implicit unlock. */ 2934 sorwakeup_locked(so); 2935 } else { 2936 /* 2937 * XXX: Due to the header drop above "th" is 2938 * theoretically invalid by now. Fortunately 2939 * m_adj() doesn't actually frees any mbufs 2940 * when trimming from the head. 2941 */ 2942 thflags = tcp_reass(tp, th, &tlen, m); 2943 tp->t_flags |= TF_ACKNOW; 2944 } 2945 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2946 tcp_update_sack_list(tp, save_start, save_start + tlen); 2947 #if 0 2948 /* 2949 * Note the amount of data that peer has sent into 2950 * our window, in order to estimate the sender's 2951 * buffer size. 2952 * XXX: Unused. 2953 */ 2954 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2955 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2956 else 2957 len = so->so_rcv.sb_hiwat; 2958 #endif 2959 } else { 2960 m_freem(m); 2961 thflags &= ~TH_FIN; 2962 } 2963 2964 /* 2965 * If FIN is received ACK the FIN and let the user know 2966 * that the connection is closing. 2967 */ 2968 if (thflags & TH_FIN) { 2969 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2970 socantrcvmore(so); 2971 /* 2972 * If connection is half-synchronized 2973 * (ie NEEDSYN flag on) then delay ACK, 2974 * so it may be piggybacked when SYN is sent. 2975 * Otherwise, since we received a FIN then no 2976 * more input can be expected, send ACK now. 2977 */ 2978 if (tp->t_flags & TF_NEEDSYN) 2979 tp->t_flags |= TF_DELACK; 2980 else 2981 tp->t_flags |= TF_ACKNOW; 2982 tp->rcv_nxt++; 2983 } 2984 switch (tp->t_state) { 2985 2986 /* 2987 * In SYN_RECEIVED and ESTABLISHED STATES 2988 * enter the CLOSE_WAIT state. 2989 */ 2990 case TCPS_SYN_RECEIVED: 2991 tp->t_starttime = ticks; 2992 /* FALLTHROUGH */ 2993 case TCPS_ESTABLISHED: 2994 tcp_state_change(tp, TCPS_CLOSE_WAIT); 2995 break; 2996 2997 /* 2998 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2999 * enter the CLOSING state. 3000 */ 3001 case TCPS_FIN_WAIT_1: 3002 tcp_state_change(tp, TCPS_CLOSING); 3003 break; 3004 3005 /* 3006 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3007 * starting the time-wait timer, turning off the other 3008 * standard timers. 3009 */ 3010 case TCPS_FIN_WAIT_2: 3011 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3012 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3013 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3014 ti_locked)); 3015 3016 tcp_twstart(tp); 3017 INP_INFO_RUNLOCK(&V_tcbinfo); 3018 return; 3019 } 3020 } 3021 if (ti_locked == TI_RLOCKED) 3022 INP_INFO_RUNLOCK(&V_tcbinfo); 3023 ti_locked = TI_UNLOCKED; 3024 3025 #ifdef TCPDEBUG 3026 if (so->so_options & SO_DEBUG) 3027 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3028 &tcp_savetcp, 0); 3029 #endif 3030 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 3031 3032 /* 3033 * Return any desired output. 3034 */ 3035 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3036 (void) tcp_output(tp); 3037 3038 check_delack: 3039 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3040 __func__, ti_locked)); 3041 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3042 INP_WLOCK_ASSERT(tp->t_inpcb); 3043 3044 if (tp->t_flags & TF_DELACK) { 3045 tp->t_flags &= ~TF_DELACK; 3046 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3047 } 3048 INP_WUNLOCK(tp->t_inpcb); 3049 return; 3050 3051 dropafterack: 3052 /* 3053 * Generate an ACK dropping incoming segment if it occupies 3054 * sequence space, where the ACK reflects our state. 3055 * 3056 * We can now skip the test for the RST flag since all 3057 * paths to this code happen after packets containing 3058 * RST have been dropped. 3059 * 3060 * In the SYN-RECEIVED state, don't send an ACK unless the 3061 * segment we received passes the SYN-RECEIVED ACK test. 3062 * If it fails send a RST. This breaks the loop in the 3063 * "LAND" DoS attack, and also prevents an ACK storm 3064 * between two listening ports that have been sent forged 3065 * SYN segments, each with the source address of the other. 3066 */ 3067 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3068 (SEQ_GT(tp->snd_una, th->th_ack) || 3069 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3070 rstreason = BANDLIM_RST_OPENPORT; 3071 goto dropwithreset; 3072 } 3073 #ifdef TCPDEBUG 3074 if (so->so_options & SO_DEBUG) 3075 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3076 &tcp_savetcp, 0); 3077 #endif 3078 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 3079 if (ti_locked == TI_RLOCKED) 3080 INP_INFO_RUNLOCK(&V_tcbinfo); 3081 ti_locked = TI_UNLOCKED; 3082 3083 tp->t_flags |= TF_ACKNOW; 3084 (void) tcp_output(tp); 3085 INP_WUNLOCK(tp->t_inpcb); 3086 m_freem(m); 3087 return; 3088 3089 dropwithreset: 3090 if (ti_locked == TI_RLOCKED) 3091 INP_INFO_RUNLOCK(&V_tcbinfo); 3092 ti_locked = TI_UNLOCKED; 3093 3094 if (tp != NULL) { 3095 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3096 INP_WUNLOCK(tp->t_inpcb); 3097 } else 3098 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3099 return; 3100 3101 drop: 3102 if (ti_locked == TI_RLOCKED) { 3103 INP_INFO_RUNLOCK(&V_tcbinfo); 3104 ti_locked = TI_UNLOCKED; 3105 } 3106 #ifdef INVARIANTS 3107 else 3108 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3109 #endif 3110 3111 /* 3112 * Drop space held by incoming segment and return. 3113 */ 3114 #ifdef TCPDEBUG 3115 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3116 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3117 &tcp_savetcp, 0); 3118 #endif 3119 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *)); 3120 if (tp != NULL) 3121 INP_WUNLOCK(tp->t_inpcb); 3122 m_freem(m); 3123 } 3124 3125 /* 3126 * Issue RST and make ACK acceptable to originator of segment. 3127 * The mbuf must still include the original packet header. 3128 * tp may be NULL. 3129 */ 3130 static void 3131 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3132 int tlen, int rstreason) 3133 { 3134 #ifdef INET 3135 struct ip *ip; 3136 #endif 3137 #ifdef INET6 3138 struct ip6_hdr *ip6; 3139 #endif 3140 3141 if (tp != NULL) { 3142 INP_WLOCK_ASSERT(tp->t_inpcb); 3143 } 3144 3145 /* Don't bother if destination was broadcast/multicast. */ 3146 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3147 goto drop; 3148 #ifdef INET6 3149 if (mtod(m, struct ip *)->ip_v == 6) { 3150 ip6 = mtod(m, struct ip6_hdr *); 3151 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3152 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3153 goto drop; 3154 /* IPv6 anycast check is done at tcp6_input() */ 3155 } 3156 #endif 3157 #if defined(INET) && defined(INET6) 3158 else 3159 #endif 3160 #ifdef INET 3161 { 3162 ip = mtod(m, struct ip *); 3163 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3164 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3165 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3166 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3167 goto drop; 3168 } 3169 #endif 3170 3171 /* Perform bandwidth limiting. */ 3172 if (badport_bandlim(rstreason) < 0) 3173 goto drop; 3174 3175 /* tcp_respond consumes the mbuf chain. */ 3176 if (th->th_flags & TH_ACK) { 3177 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3178 th->th_ack, TH_RST); 3179 } else { 3180 if (th->th_flags & TH_SYN) 3181 tlen++; 3182 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3183 (tcp_seq)0, TH_RST|TH_ACK); 3184 } 3185 return; 3186 drop: 3187 m_freem(m); 3188 } 3189 3190 /* 3191 * Parse TCP options and place in tcpopt. 3192 */ 3193 static void 3194 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3195 { 3196 int opt, optlen; 3197 3198 to->to_flags = 0; 3199 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3200 opt = cp[0]; 3201 if (opt == TCPOPT_EOL) 3202 break; 3203 if (opt == TCPOPT_NOP) 3204 optlen = 1; 3205 else { 3206 if (cnt < 2) 3207 break; 3208 optlen = cp[1]; 3209 if (optlen < 2 || optlen > cnt) 3210 break; 3211 } 3212 switch (opt) { 3213 case TCPOPT_MAXSEG: 3214 if (optlen != TCPOLEN_MAXSEG) 3215 continue; 3216 if (!(flags & TO_SYN)) 3217 continue; 3218 to->to_flags |= TOF_MSS; 3219 bcopy((char *)cp + 2, 3220 (char *)&to->to_mss, sizeof(to->to_mss)); 3221 to->to_mss = ntohs(to->to_mss); 3222 break; 3223 case TCPOPT_WINDOW: 3224 if (optlen != TCPOLEN_WINDOW) 3225 continue; 3226 if (!(flags & TO_SYN)) 3227 continue; 3228 to->to_flags |= TOF_SCALE; 3229 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3230 break; 3231 case TCPOPT_TIMESTAMP: 3232 if (optlen != TCPOLEN_TIMESTAMP) 3233 continue; 3234 to->to_flags |= TOF_TS; 3235 bcopy((char *)cp + 2, 3236 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3237 to->to_tsval = ntohl(to->to_tsval); 3238 bcopy((char *)cp + 6, 3239 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3240 to->to_tsecr = ntohl(to->to_tsecr); 3241 break; 3242 #ifdef TCP_SIGNATURE 3243 /* 3244 * XXX In order to reply to a host which has set the 3245 * TCP_SIGNATURE option in its initial SYN, we have to 3246 * record the fact that the option was observed here 3247 * for the syncache code to perform the correct response. 3248 */ 3249 case TCPOPT_SIGNATURE: 3250 if (optlen != TCPOLEN_SIGNATURE) 3251 continue; 3252 to->to_flags |= TOF_SIGNATURE; 3253 to->to_signature = cp + 2; 3254 break; 3255 #endif 3256 case TCPOPT_SACK_PERMITTED: 3257 if (optlen != TCPOLEN_SACK_PERMITTED) 3258 continue; 3259 if (!(flags & TO_SYN)) 3260 continue; 3261 if (!V_tcp_do_sack) 3262 continue; 3263 to->to_flags |= TOF_SACKPERM; 3264 break; 3265 case TCPOPT_SACK: 3266 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3267 continue; 3268 if (flags & TO_SYN) 3269 continue; 3270 to->to_flags |= TOF_SACK; 3271 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3272 to->to_sacks = cp + 2; 3273 TCPSTAT_INC(tcps_sack_rcv_blocks); 3274 break; 3275 default: 3276 continue; 3277 } 3278 } 3279 } 3280 3281 /* 3282 * Pull out of band byte out of a segment so 3283 * it doesn't appear in the user's data queue. 3284 * It is still reflected in the segment length for 3285 * sequencing purposes. 3286 */ 3287 static void 3288 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3289 int off) 3290 { 3291 int cnt = off + th->th_urp - 1; 3292 3293 while (cnt >= 0) { 3294 if (m->m_len > cnt) { 3295 char *cp = mtod(m, caddr_t) + cnt; 3296 struct tcpcb *tp = sototcpcb(so); 3297 3298 INP_WLOCK_ASSERT(tp->t_inpcb); 3299 3300 tp->t_iobc = *cp; 3301 tp->t_oobflags |= TCPOOB_HAVEDATA; 3302 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3303 m->m_len--; 3304 if (m->m_flags & M_PKTHDR) 3305 m->m_pkthdr.len--; 3306 return; 3307 } 3308 cnt -= m->m_len; 3309 m = m->m_next; 3310 if (m == NULL) 3311 break; 3312 } 3313 panic("tcp_pulloutofband"); 3314 } 3315 3316 /* 3317 * Collect new round-trip time estimate 3318 * and update averages and current timeout. 3319 */ 3320 static void 3321 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3322 { 3323 int delta; 3324 3325 INP_WLOCK_ASSERT(tp->t_inpcb); 3326 3327 TCPSTAT_INC(tcps_rttupdated); 3328 tp->t_rttupdated++; 3329 if (tp->t_srtt != 0) { 3330 /* 3331 * srtt is stored as fixed point with 5 bits after the 3332 * binary point (i.e., scaled by 8). The following magic 3333 * is equivalent to the smoothing algorithm in rfc793 with 3334 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3335 * point). Adjust rtt to origin 0. 3336 */ 3337 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3338 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3339 3340 if ((tp->t_srtt += delta) <= 0) 3341 tp->t_srtt = 1; 3342 3343 /* 3344 * We accumulate a smoothed rtt variance (actually, a 3345 * smoothed mean difference), then set the retransmit 3346 * timer to smoothed rtt + 4 times the smoothed variance. 3347 * rttvar is stored as fixed point with 4 bits after the 3348 * binary point (scaled by 16). The following is 3349 * equivalent to rfc793 smoothing with an alpha of .75 3350 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3351 * rfc793's wired-in beta. 3352 */ 3353 if (delta < 0) 3354 delta = -delta; 3355 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3356 if ((tp->t_rttvar += delta) <= 0) 3357 tp->t_rttvar = 1; 3358 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3359 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3360 } else { 3361 /* 3362 * No rtt measurement yet - use the unsmoothed rtt. 3363 * Set the variance to half the rtt (so our first 3364 * retransmit happens at 3*rtt). 3365 */ 3366 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3367 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3368 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3369 } 3370 tp->t_rtttime = 0; 3371 tp->t_rxtshift = 0; 3372 3373 /* 3374 * the retransmit should happen at rtt + 4 * rttvar. 3375 * Because of the way we do the smoothing, srtt and rttvar 3376 * will each average +1/2 tick of bias. When we compute 3377 * the retransmit timer, we want 1/2 tick of rounding and 3378 * 1 extra tick because of +-1/2 tick uncertainty in the 3379 * firing of the timer. The bias will give us exactly the 3380 * 1.5 tick we need. But, because the bias is 3381 * statistical, we have to test that we don't drop below 3382 * the minimum feasible timer (which is 2 ticks). 3383 */ 3384 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3385 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3386 3387 /* 3388 * We received an ack for a packet that wasn't retransmitted; 3389 * it is probably safe to discard any error indications we've 3390 * received recently. This isn't quite right, but close enough 3391 * for now (a route might have failed after we sent a segment, 3392 * and the return path might not be symmetrical). 3393 */ 3394 tp->t_softerror = 0; 3395 } 3396 3397 /* 3398 * Determine a reasonable value for maxseg size. 3399 * If the route is known, check route for mtu. 3400 * If none, use an mss that can be handled on the outgoing interface 3401 * without forcing IP to fragment. If no route is found, route has no mtu, 3402 * or the destination isn't local, use a default, hopefully conservative 3403 * size (usually 512 or the default IP max size, but no more than the mtu 3404 * of the interface), as we can't discover anything about intervening 3405 * gateways or networks. We also initialize the congestion/slow start 3406 * window to be a single segment if the destination isn't local. 3407 * While looking at the routing entry, we also initialize other path-dependent 3408 * parameters from pre-set or cached values in the routing entry. 3409 * 3410 * Also take into account the space needed for options that we 3411 * send regularly. Make maxseg shorter by that amount to assure 3412 * that we can send maxseg amount of data even when the options 3413 * are present. Store the upper limit of the length of options plus 3414 * data in maxopd. 3415 * 3416 * NOTE that this routine is only called when we process an incoming 3417 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3418 * settings are handled in tcp_mssopt(). 3419 */ 3420 void 3421 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3422 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3423 { 3424 int mss = 0; 3425 u_long maxmtu = 0; 3426 struct inpcb *inp = tp->t_inpcb; 3427 struct hc_metrics_lite metrics; 3428 int origoffer; 3429 #ifdef INET6 3430 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3431 size_t min_protoh = isipv6 ? 3432 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3433 sizeof (struct tcpiphdr); 3434 #else 3435 const size_t min_protoh = sizeof(struct tcpiphdr); 3436 #endif 3437 3438 INP_WLOCK_ASSERT(tp->t_inpcb); 3439 3440 if (mtuoffer != -1) { 3441 KASSERT(offer == -1, ("%s: conflict", __func__)); 3442 offer = mtuoffer - min_protoh; 3443 } 3444 origoffer = offer; 3445 3446 /* Initialize. */ 3447 #ifdef INET6 3448 if (isipv6) { 3449 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3450 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3451 } 3452 #endif 3453 #if defined(INET) && defined(INET6) 3454 else 3455 #endif 3456 #ifdef INET 3457 { 3458 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3459 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3460 } 3461 #endif 3462 3463 /* 3464 * No route to sender, stay with default mss and return. 3465 */ 3466 if (maxmtu == 0) { 3467 /* 3468 * In case we return early we need to initialize metrics 3469 * to a defined state as tcp_hc_get() would do for us 3470 * if there was no cache hit. 3471 */ 3472 if (metricptr != NULL) 3473 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3474 return; 3475 } 3476 3477 /* What have we got? */ 3478 switch (offer) { 3479 case 0: 3480 /* 3481 * Offer == 0 means that there was no MSS on the SYN 3482 * segment, in this case we use tcp_mssdflt as 3483 * already assigned to t_maxopd above. 3484 */ 3485 offer = tp->t_maxopd; 3486 break; 3487 3488 case -1: 3489 /* 3490 * Offer == -1 means that we didn't receive SYN yet. 3491 */ 3492 /* FALLTHROUGH */ 3493 3494 default: 3495 /* 3496 * Prevent DoS attack with too small MSS. Round up 3497 * to at least minmss. 3498 */ 3499 offer = max(offer, V_tcp_minmss); 3500 } 3501 3502 /* 3503 * rmx information is now retrieved from tcp_hostcache. 3504 */ 3505 tcp_hc_get(&inp->inp_inc, &metrics); 3506 if (metricptr != NULL) 3507 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3508 3509 /* 3510 * If there's a discovered mtu in tcp hostcache, use it. 3511 * Else, use the link mtu. 3512 */ 3513 if (metrics.rmx_mtu) 3514 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3515 else { 3516 #ifdef INET6 3517 if (isipv6) { 3518 mss = maxmtu - min_protoh; 3519 if (!V_path_mtu_discovery && 3520 !in6_localaddr(&inp->in6p_faddr)) 3521 mss = min(mss, V_tcp_v6mssdflt); 3522 } 3523 #endif 3524 #if defined(INET) && defined(INET6) 3525 else 3526 #endif 3527 #ifdef INET 3528 { 3529 mss = maxmtu - min_protoh; 3530 if (!V_path_mtu_discovery && 3531 !in_localaddr(inp->inp_faddr)) 3532 mss = min(mss, V_tcp_mssdflt); 3533 } 3534 #endif 3535 /* 3536 * XXX - The above conditional (mss = maxmtu - min_protoh) 3537 * probably violates the TCP spec. 3538 * The problem is that, since we don't know the 3539 * other end's MSS, we are supposed to use a conservative 3540 * default. But, if we do that, then MTU discovery will 3541 * never actually take place, because the conservative 3542 * default is much less than the MTUs typically seen 3543 * on the Internet today. For the moment, we'll sweep 3544 * this under the carpet. 3545 * 3546 * The conservative default might not actually be a problem 3547 * if the only case this occurs is when sending an initial 3548 * SYN with options and data to a host we've never talked 3549 * to before. Then, they will reply with an MSS value which 3550 * will get recorded and the new parameters should get 3551 * recomputed. For Further Study. 3552 */ 3553 } 3554 mss = min(mss, offer); 3555 3556 /* 3557 * Sanity check: make sure that maxopd will be large 3558 * enough to allow some data on segments even if the 3559 * all the option space is used (40bytes). Otherwise 3560 * funny things may happen in tcp_output. 3561 */ 3562 mss = max(mss, 64); 3563 3564 /* 3565 * maxopd stores the maximum length of data AND options 3566 * in a segment; maxseg is the amount of data in a normal 3567 * segment. We need to store this value (maxopd) apart 3568 * from maxseg, because now every segment carries options 3569 * and thus we normally have somewhat less data in segments. 3570 */ 3571 tp->t_maxopd = mss; 3572 3573 /* 3574 * origoffer==-1 indicates that no segments were received yet. 3575 * In this case we just guess. 3576 */ 3577 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3578 (origoffer == -1 || 3579 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3580 mss -= TCPOLEN_TSTAMP_APPA; 3581 3582 tp->t_maxseg = mss; 3583 } 3584 3585 void 3586 tcp_mss(struct tcpcb *tp, int offer) 3587 { 3588 int mss; 3589 u_long bufsize; 3590 struct inpcb *inp; 3591 struct socket *so; 3592 struct hc_metrics_lite metrics; 3593 struct tcp_ifcap cap; 3594 3595 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3596 3597 bzero(&cap, sizeof(cap)); 3598 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3599 3600 mss = tp->t_maxseg; 3601 inp = tp->t_inpcb; 3602 3603 /* 3604 * If there's a pipesize, change the socket buffer to that size, 3605 * don't change if sb_hiwat is different than default (then it 3606 * has been changed on purpose with setsockopt). 3607 * Make the socket buffers an integral number of mss units; 3608 * if the mss is larger than the socket buffer, decrease the mss. 3609 */ 3610 so = inp->inp_socket; 3611 SOCKBUF_LOCK(&so->so_snd); 3612 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3613 bufsize = metrics.rmx_sendpipe; 3614 else 3615 bufsize = so->so_snd.sb_hiwat; 3616 if (bufsize < mss) 3617 mss = bufsize; 3618 else { 3619 bufsize = roundup(bufsize, mss); 3620 if (bufsize > sb_max) 3621 bufsize = sb_max; 3622 if (bufsize > so->so_snd.sb_hiwat) 3623 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3624 } 3625 SOCKBUF_UNLOCK(&so->so_snd); 3626 tp->t_maxseg = mss; 3627 3628 SOCKBUF_LOCK(&so->so_rcv); 3629 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3630 bufsize = metrics.rmx_recvpipe; 3631 else 3632 bufsize = so->so_rcv.sb_hiwat; 3633 if (bufsize > mss) { 3634 bufsize = roundup(bufsize, mss); 3635 if (bufsize > sb_max) 3636 bufsize = sb_max; 3637 if (bufsize > so->so_rcv.sb_hiwat) 3638 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3639 } 3640 SOCKBUF_UNLOCK(&so->so_rcv); 3641 3642 /* Check the interface for TSO capabilities. */ 3643 if (cap.ifcap & CSUM_TSO) { 3644 tp->t_flags |= TF_TSO; 3645 tp->t_tsomax = cap.tsomax; 3646 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3647 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3648 } 3649 } 3650 3651 /* 3652 * Determine the MSS option to send on an outgoing SYN. 3653 */ 3654 int 3655 tcp_mssopt(struct in_conninfo *inc) 3656 { 3657 int mss = 0; 3658 u_long maxmtu = 0; 3659 u_long thcmtu = 0; 3660 size_t min_protoh; 3661 3662 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3663 3664 #ifdef INET6 3665 if (inc->inc_flags & INC_ISIPV6) { 3666 mss = V_tcp_v6mssdflt; 3667 maxmtu = tcp_maxmtu6(inc, NULL); 3668 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3669 } 3670 #endif 3671 #if defined(INET) && defined(INET6) 3672 else 3673 #endif 3674 #ifdef INET 3675 { 3676 mss = V_tcp_mssdflt; 3677 maxmtu = tcp_maxmtu(inc, NULL); 3678 min_protoh = sizeof(struct tcpiphdr); 3679 } 3680 #endif 3681 #if defined(INET6) || defined(INET) 3682 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3683 #endif 3684 3685 if (maxmtu && thcmtu) 3686 mss = min(maxmtu, thcmtu) - min_protoh; 3687 else if (maxmtu || thcmtu) 3688 mss = max(maxmtu, thcmtu) - min_protoh; 3689 3690 return (mss); 3691 } 3692 3693 3694 /* 3695 * On a partial ack arrives, force the retransmission of the 3696 * next unacknowledged segment. Do not clear tp->t_dupacks. 3697 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3698 * be started again. 3699 */ 3700 static void 3701 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3702 { 3703 tcp_seq onxt = tp->snd_nxt; 3704 u_long ocwnd = tp->snd_cwnd; 3705 3706 INP_WLOCK_ASSERT(tp->t_inpcb); 3707 3708 tcp_timer_activate(tp, TT_REXMT, 0); 3709 tp->t_rtttime = 0; 3710 tp->snd_nxt = th->th_ack; 3711 /* 3712 * Set snd_cwnd to one segment beyond acknowledged offset. 3713 * (tp->snd_una has not yet been updated when this function is called.) 3714 */ 3715 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3716 tp->t_flags |= TF_ACKNOW; 3717 (void) tcp_output(tp); 3718 tp->snd_cwnd = ocwnd; 3719 if (SEQ_GT(onxt, tp->snd_nxt)) 3720 tp->snd_nxt = onxt; 3721 /* 3722 * Partial window deflation. Relies on fact that tp->snd_una 3723 * not updated yet. 3724 */ 3725 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3726 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3727 else 3728 tp->snd_cwnd = 0; 3729 tp->snd_cwnd += tp->t_maxseg; 3730 } 3731