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_dooptions(&to, optp, optlen, TO_SYN); 1381 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1382 /* 1383 * Entry added to syncache and mbuf consumed. 1384 * Only the listen socket is unlocked by syncache_add(). 1385 */ 1386 if (ti_locked == TI_RLOCKED) { 1387 INP_INFO_RUNLOCK(&V_tcbinfo); 1388 ti_locked = TI_UNLOCKED; 1389 } 1390 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1391 return (IPPROTO_DONE); 1392 } else if (tp->t_state == TCPS_LISTEN) { 1393 /* 1394 * When a listen socket is torn down the SO_ACCEPTCONN 1395 * flag is removed first while connections are drained 1396 * from the accept queue in a unlock/lock cycle of the 1397 * ACCEPT_LOCK, opening a race condition allowing a SYN 1398 * attempt go through unhandled. 1399 */ 1400 goto dropunlock; 1401 } 1402 1403 #ifdef TCP_SIGNATURE 1404 if (sig_checked == 0) { 1405 tcp_dooptions(&to, optp, optlen, 1406 (thflags & TH_SYN) ? TO_SYN : 0); 1407 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1408 th, tp->t_flags)) { 1409 1410 /* 1411 * In SYN_SENT state if it receives an RST, it is 1412 * allowed for further processing. 1413 */ 1414 if ((thflags & TH_RST) == 0 || 1415 (tp->t_state == TCPS_SYN_SENT) == 0) 1416 goto dropunlock; 1417 } 1418 sig_checked = 1; 1419 } 1420 #endif 1421 1422 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1423 1424 /* 1425 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1426 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1427 * the inpcb, and unlocks pcbinfo. 1428 */ 1429 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1430 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1431 return (IPPROTO_DONE); 1432 1433 dropwithreset: 1434 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1435 1436 if (ti_locked == TI_RLOCKED) { 1437 INP_INFO_RUNLOCK(&V_tcbinfo); 1438 ti_locked = TI_UNLOCKED; 1439 } 1440 #ifdef INVARIANTS 1441 else { 1442 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1443 "ti_locked: %d", __func__, ti_locked)); 1444 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1445 } 1446 #endif 1447 1448 if (inp != NULL) { 1449 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1450 INP_WUNLOCK(inp); 1451 } else 1452 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1453 m = NULL; /* mbuf chain got consumed. */ 1454 goto drop; 1455 1456 dropunlock: 1457 if (m != NULL) 1458 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1459 1460 if (ti_locked == TI_RLOCKED) { 1461 INP_INFO_RUNLOCK(&V_tcbinfo); 1462 ti_locked = TI_UNLOCKED; 1463 } 1464 #ifdef INVARIANTS 1465 else { 1466 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1467 "ti_locked: %d", __func__, ti_locked)); 1468 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1469 } 1470 #endif 1471 1472 if (inp != NULL) 1473 INP_WUNLOCK(inp); 1474 1475 drop: 1476 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1477 if (s != NULL) 1478 free(s, M_TCPLOG); 1479 if (m != NULL) 1480 m_freem(m); 1481 return (IPPROTO_DONE); 1482 } 1483 1484 static void 1485 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1486 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1487 int ti_locked) 1488 { 1489 int thflags, acked, ourfinisacked, needoutput = 0; 1490 int rstreason, todrop, win; 1491 u_long tiwin; 1492 char *s; 1493 struct in_conninfo *inc; 1494 struct mbuf *mfree; 1495 struct tcpopt to; 1496 1497 #ifdef TCPDEBUG 1498 /* 1499 * The size of tcp_saveipgen must be the size of the max ip header, 1500 * now IPv6. 1501 */ 1502 u_char tcp_saveipgen[IP6_HDR_LEN]; 1503 struct tcphdr tcp_savetcp; 1504 short ostate = 0; 1505 #endif 1506 thflags = th->th_flags; 1507 inc = &tp->t_inpcb->inp_inc; 1508 tp->sackhint.last_sack_ack = 0; 1509 1510 /* 1511 * If this is either a state-changing packet or current state isn't 1512 * established, we require a write lock on tcbinfo. Otherwise, we 1513 * allow the tcbinfo to be in either alocked or unlocked, as the 1514 * caller may have unnecessarily acquired a write lock due to a race. 1515 */ 1516 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1517 tp->t_state != TCPS_ESTABLISHED) { 1518 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for " 1519 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1520 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1521 } else { 1522 #ifdef INVARIANTS 1523 if (ti_locked == TI_RLOCKED) 1524 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1525 else { 1526 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1527 "ti_locked: %d", __func__, ti_locked)); 1528 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1529 } 1530 #endif 1531 } 1532 INP_WLOCK_ASSERT(tp->t_inpcb); 1533 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1534 __func__)); 1535 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1536 __func__)); 1537 1538 /* 1539 * Segment received on connection. 1540 * Reset idle time and keep-alive timer. 1541 * XXX: This should be done after segment 1542 * validation to ignore broken/spoofed segs. 1543 */ 1544 tp->t_rcvtime = ticks; 1545 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1546 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1547 1548 /* 1549 * Scale up the window into a 32-bit value. 1550 * For the SYN_SENT state the scale is zero. 1551 */ 1552 tiwin = th->th_win << tp->snd_scale; 1553 1554 /* 1555 * TCP ECN processing. 1556 */ 1557 if (tp->t_flags & TF_ECN_PERMIT) { 1558 if (thflags & TH_CWR) 1559 tp->t_flags &= ~TF_ECN_SND_ECE; 1560 switch (iptos & IPTOS_ECN_MASK) { 1561 case IPTOS_ECN_CE: 1562 tp->t_flags |= TF_ECN_SND_ECE; 1563 TCPSTAT_INC(tcps_ecn_ce); 1564 break; 1565 case IPTOS_ECN_ECT0: 1566 TCPSTAT_INC(tcps_ecn_ect0); 1567 break; 1568 case IPTOS_ECN_ECT1: 1569 TCPSTAT_INC(tcps_ecn_ect1); 1570 break; 1571 } 1572 1573 /* Process a packet differently from RFC3168. */ 1574 cc_ecnpkt_handler(tp, th, iptos); 1575 1576 /* Congestion experienced. */ 1577 if (thflags & TH_ECE) { 1578 cc_cong_signal(tp, th, CC_ECN); 1579 } 1580 } 1581 1582 /* 1583 * Parse options on any incoming segment. 1584 */ 1585 tcp_dooptions(&to, (u_char *)(th + 1), 1586 (th->th_off << 2) - sizeof(struct tcphdr), 1587 (thflags & TH_SYN) ? TO_SYN : 0); 1588 1589 /* 1590 * If echoed timestamp is later than the current time, 1591 * fall back to non RFC1323 RTT calculation. Normalize 1592 * timestamp if syncookies were used when this connection 1593 * was established. 1594 */ 1595 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1596 to.to_tsecr -= tp->ts_offset; 1597 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1598 to.to_tsecr = 0; 1599 } 1600 /* 1601 * If timestamps were negotiated during SYN/ACK they should 1602 * appear on every segment during this session and vice versa. 1603 */ 1604 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1605 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1606 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1607 "no action\n", s, __func__); 1608 free(s, M_TCPLOG); 1609 } 1610 } 1611 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1612 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1613 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1614 "no action\n", s, __func__); 1615 free(s, M_TCPLOG); 1616 } 1617 } 1618 1619 /* 1620 * Process options only when we get SYN/ACK back. The SYN case 1621 * for incoming connections is handled in tcp_syncache. 1622 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1623 * or <SYN,ACK>) segment itself is never scaled. 1624 * XXX this is traditional behavior, may need to be cleaned up. 1625 */ 1626 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1627 if ((to.to_flags & TOF_SCALE) && 1628 (tp->t_flags & TF_REQ_SCALE)) { 1629 tp->t_flags |= TF_RCVD_SCALE; 1630 tp->snd_scale = to.to_wscale; 1631 } 1632 /* 1633 * Initial send window. It will be updated with 1634 * the next incoming segment to the scaled value. 1635 */ 1636 tp->snd_wnd = th->th_win; 1637 if (to.to_flags & TOF_TS) { 1638 tp->t_flags |= TF_RCVD_TSTMP; 1639 tp->ts_recent = to.to_tsval; 1640 tp->ts_recent_age = tcp_ts_getticks(); 1641 } 1642 if (to.to_flags & TOF_MSS) 1643 tcp_mss(tp, to.to_mss); 1644 if ((tp->t_flags & TF_SACK_PERMIT) && 1645 (to.to_flags & TOF_SACKPERM) == 0) 1646 tp->t_flags &= ~TF_SACK_PERMIT; 1647 } 1648 1649 /* 1650 * Header prediction: check for the two common cases 1651 * of a uni-directional data xfer. If the packet has 1652 * no control flags, is in-sequence, the window didn't 1653 * change and we're not retransmitting, it's a 1654 * candidate. If the length is zero and the ack moved 1655 * forward, we're the sender side of the xfer. Just 1656 * free the data acked & wake any higher level process 1657 * that was blocked waiting for space. If the length 1658 * is non-zero and the ack didn't move, we're the 1659 * receiver side. If we're getting packets in-order 1660 * (the reassembly queue is empty), add the data to 1661 * the socket buffer and note that we need a delayed ack. 1662 * Make sure that the hidden state-flags are also off. 1663 * Since we check for TCPS_ESTABLISHED first, it can only 1664 * be TH_NEEDSYN. 1665 */ 1666 if (tp->t_state == TCPS_ESTABLISHED && 1667 th->th_seq == tp->rcv_nxt && 1668 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1669 tp->snd_nxt == tp->snd_max && 1670 tiwin && tiwin == tp->snd_wnd && 1671 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1672 LIST_EMPTY(&tp->t_segq) && 1673 ((to.to_flags & TOF_TS) == 0 || 1674 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1675 1676 /* 1677 * If last ACK falls within this segment's sequence numbers, 1678 * record the timestamp. 1679 * NOTE that the test is modified according to the latest 1680 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1681 */ 1682 if ((to.to_flags & TOF_TS) != 0 && 1683 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1684 tp->ts_recent_age = tcp_ts_getticks(); 1685 tp->ts_recent = to.to_tsval; 1686 } 1687 1688 if (tlen == 0) { 1689 if (SEQ_GT(th->th_ack, tp->snd_una) && 1690 SEQ_LEQ(th->th_ack, tp->snd_max) && 1691 !IN_RECOVERY(tp->t_flags) && 1692 (to.to_flags & TOF_SACK) == 0 && 1693 TAILQ_EMPTY(&tp->snd_holes)) { 1694 /* 1695 * This is a pure ack for outstanding data. 1696 */ 1697 if (ti_locked == TI_RLOCKED) 1698 INP_INFO_RUNLOCK(&V_tcbinfo); 1699 ti_locked = TI_UNLOCKED; 1700 1701 TCPSTAT_INC(tcps_predack); 1702 1703 /* 1704 * "bad retransmit" recovery. 1705 */ 1706 if (tp->t_rxtshift == 1 && 1707 tp->t_flags & TF_PREVVALID && 1708 (int)(ticks - tp->t_badrxtwin) < 0) { 1709 cc_cong_signal(tp, th, CC_RTO_ERR); 1710 } 1711 1712 /* 1713 * Recalculate the transmit timer / rtt. 1714 * 1715 * Some boxes send broken timestamp replies 1716 * during the SYN+ACK phase, ignore 1717 * timestamps of 0 or we could calculate a 1718 * huge RTT and blow up the retransmit timer. 1719 */ 1720 if ((to.to_flags & TOF_TS) != 0 && 1721 to.to_tsecr) { 1722 u_int t; 1723 1724 t = tcp_ts_getticks() - to.to_tsecr; 1725 if (!tp->t_rttlow || tp->t_rttlow > t) 1726 tp->t_rttlow = t; 1727 tcp_xmit_timer(tp, 1728 TCP_TS_TO_TICKS(t) + 1); 1729 } else if (tp->t_rtttime && 1730 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1731 if (!tp->t_rttlow || 1732 tp->t_rttlow > ticks - tp->t_rtttime) 1733 tp->t_rttlow = ticks - tp->t_rtttime; 1734 tcp_xmit_timer(tp, 1735 ticks - tp->t_rtttime); 1736 } 1737 acked = BYTES_THIS_ACK(tp, th); 1738 1739 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1740 hhook_run_tcp_est_in(tp, th, &to); 1741 1742 TCPSTAT_INC(tcps_rcvackpack); 1743 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1744 sbdrop(&so->so_snd, acked); 1745 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1746 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1747 tp->snd_recover = th->th_ack - 1; 1748 1749 /* 1750 * Let the congestion control algorithm update 1751 * congestion control related information. This 1752 * typically means increasing the congestion 1753 * window. 1754 */ 1755 cc_ack_received(tp, th, CC_ACK); 1756 1757 tp->snd_una = th->th_ack; 1758 /* 1759 * Pull snd_wl2 up to prevent seq wrap relative 1760 * to th_ack. 1761 */ 1762 tp->snd_wl2 = th->th_ack; 1763 tp->t_dupacks = 0; 1764 m_freem(m); 1765 ND6_HINT(tp); /* Some progress has been made. */ 1766 1767 /* 1768 * If all outstanding data are acked, stop 1769 * retransmit timer, otherwise restart timer 1770 * using current (possibly backed-off) value. 1771 * If process is waiting for space, 1772 * wakeup/selwakeup/signal. If data 1773 * are ready to send, let tcp_output 1774 * decide between more output or persist. 1775 */ 1776 #ifdef TCPDEBUG 1777 if (so->so_options & SO_DEBUG) 1778 tcp_trace(TA_INPUT, ostate, tp, 1779 (void *)tcp_saveipgen, 1780 &tcp_savetcp, 0); 1781 #endif 1782 if (tp->snd_una == tp->snd_max) 1783 tcp_timer_activate(tp, TT_REXMT, 0); 1784 else if (!tcp_timer_active(tp, TT_PERSIST)) 1785 tcp_timer_activate(tp, TT_REXMT, 1786 tp->t_rxtcur); 1787 sowwakeup(so); 1788 if (sbavail(&so->so_snd)) 1789 (void) tcp_output(tp); 1790 goto check_delack; 1791 } 1792 } else if (th->th_ack == tp->snd_una && 1793 tlen <= sbspace(&so->so_rcv)) { 1794 int newsize = 0; /* automatic sockbuf scaling */ 1795 1796 /* 1797 * This is a pure, in-sequence data packet with 1798 * nothing on the reassembly queue and we have enough 1799 * buffer space to take it. 1800 */ 1801 if (ti_locked == TI_RLOCKED) 1802 INP_INFO_RUNLOCK(&V_tcbinfo); 1803 ti_locked = TI_UNLOCKED; 1804 1805 /* Clean receiver SACK report if present */ 1806 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1807 tcp_clean_sackreport(tp); 1808 TCPSTAT_INC(tcps_preddat); 1809 tp->rcv_nxt += tlen; 1810 /* 1811 * Pull snd_wl1 up to prevent seq wrap relative to 1812 * th_seq. 1813 */ 1814 tp->snd_wl1 = th->th_seq; 1815 /* 1816 * Pull rcv_up up to prevent seq wrap relative to 1817 * rcv_nxt. 1818 */ 1819 tp->rcv_up = tp->rcv_nxt; 1820 TCPSTAT_INC(tcps_rcvpack); 1821 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1822 ND6_HINT(tp); /* Some progress has been made */ 1823 #ifdef TCPDEBUG 1824 if (so->so_options & SO_DEBUG) 1825 tcp_trace(TA_INPUT, ostate, tp, 1826 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1827 #endif 1828 /* 1829 * Automatic sizing of receive socket buffer. Often the send 1830 * buffer size is not optimally adjusted to the actual network 1831 * conditions at hand (delay bandwidth product). Setting the 1832 * buffer size too small limits throughput on links with high 1833 * bandwidth and high delay (eg. trans-continental/oceanic links). 1834 * 1835 * On the receive side the socket buffer memory is only rarely 1836 * used to any significant extent. This allows us to be much 1837 * more aggressive in scaling the receive socket buffer. For 1838 * the case that the buffer space is actually used to a large 1839 * extent and we run out of kernel memory we can simply drop 1840 * the new segments; TCP on the sender will just retransmit it 1841 * later. Setting the buffer size too big may only consume too 1842 * much kernel memory if the application doesn't read() from 1843 * the socket or packet loss or reordering makes use of the 1844 * reassembly queue. 1845 * 1846 * The criteria to step up the receive buffer one notch are: 1847 * 1. Application has not set receive buffer size with 1848 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1849 * 2. the number of bytes received during the time it takes 1850 * one timestamp to be reflected back to us (the RTT); 1851 * 3. received bytes per RTT is within seven eighth of the 1852 * current socket buffer size; 1853 * 4. receive buffer size has not hit maximal automatic size; 1854 * 1855 * This algorithm does one step per RTT at most and only if 1856 * we receive a bulk stream w/o packet losses or reorderings. 1857 * Shrinking the buffer during idle times is not necessary as 1858 * it doesn't consume any memory when idle. 1859 * 1860 * TODO: Only step up if the application is actually serving 1861 * the buffer to better manage the socket buffer resources. 1862 */ 1863 if (V_tcp_do_autorcvbuf && 1864 (to.to_flags & TOF_TS) && 1865 to.to_tsecr && 1866 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1867 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1868 to.to_tsecr - tp->rfbuf_ts < hz) { 1869 if (tp->rfbuf_cnt > 1870 (so->so_rcv.sb_hiwat / 8 * 7) && 1871 so->so_rcv.sb_hiwat < 1872 V_tcp_autorcvbuf_max) { 1873 newsize = 1874 min(so->so_rcv.sb_hiwat + 1875 V_tcp_autorcvbuf_inc, 1876 V_tcp_autorcvbuf_max); 1877 } 1878 /* Start over with next RTT. */ 1879 tp->rfbuf_ts = 0; 1880 tp->rfbuf_cnt = 0; 1881 } else 1882 tp->rfbuf_cnt += tlen; /* add up */ 1883 } 1884 1885 /* Add data to socket buffer. */ 1886 SOCKBUF_LOCK(&so->so_rcv); 1887 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1888 m_freem(m); 1889 } else { 1890 /* 1891 * Set new socket buffer size. 1892 * Give up when limit is reached. 1893 */ 1894 if (newsize) 1895 if (!sbreserve_locked(&so->so_rcv, 1896 newsize, so, NULL)) 1897 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1898 m_adj(m, drop_hdrlen); /* delayed header drop */ 1899 sbappendstream_locked(&so->so_rcv, m, 0); 1900 } 1901 /* NB: sorwakeup_locked() does an implicit unlock. */ 1902 sorwakeup_locked(so); 1903 if (DELAY_ACK(tp, tlen)) { 1904 tp->t_flags |= TF_DELACK; 1905 } else { 1906 tp->t_flags |= TF_ACKNOW; 1907 tcp_output(tp); 1908 } 1909 goto check_delack; 1910 } 1911 } 1912 1913 /* 1914 * Calculate amount of space in receive window, 1915 * and then do TCP input processing. 1916 * Receive window is amount of space in rcv queue, 1917 * but not less than advertised window. 1918 */ 1919 win = sbspace(&so->so_rcv); 1920 if (win < 0) 1921 win = 0; 1922 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1923 1924 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1925 tp->rfbuf_ts = 0; 1926 tp->rfbuf_cnt = 0; 1927 1928 switch (tp->t_state) { 1929 1930 /* 1931 * If the state is SYN_RECEIVED: 1932 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1933 */ 1934 case TCPS_SYN_RECEIVED: 1935 if ((thflags & TH_ACK) && 1936 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1937 SEQ_GT(th->th_ack, tp->snd_max))) { 1938 rstreason = BANDLIM_RST_OPENPORT; 1939 goto dropwithreset; 1940 } 1941 break; 1942 1943 /* 1944 * If the state is SYN_SENT: 1945 * if seg contains an ACK, but not for our SYN, drop the input. 1946 * if seg contains a RST, then drop the connection. 1947 * if seg does not contain SYN, then drop it. 1948 * Otherwise this is an acceptable SYN segment 1949 * initialize tp->rcv_nxt and tp->irs 1950 * if seg contains ack then advance tp->snd_una 1951 * if seg contains an ECE and ECN support is enabled, the stream 1952 * is ECN capable. 1953 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1954 * arrange for segment to be acked (eventually) 1955 * continue processing rest of data/controls, beginning with URG 1956 */ 1957 case TCPS_SYN_SENT: 1958 if ((thflags & TH_ACK) && 1959 (SEQ_LEQ(th->th_ack, tp->iss) || 1960 SEQ_GT(th->th_ack, tp->snd_max))) { 1961 rstreason = BANDLIM_UNLIMITED; 1962 goto dropwithreset; 1963 } 1964 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1965 TCP_PROBE5(connect__refused, NULL, tp, 1966 mtod(m, const char *), tp, th); 1967 tp = tcp_drop(tp, ECONNREFUSED); 1968 } 1969 if (thflags & TH_RST) 1970 goto drop; 1971 if (!(thflags & TH_SYN)) 1972 goto drop; 1973 1974 tp->irs = th->th_seq; 1975 tcp_rcvseqinit(tp); 1976 if (thflags & TH_ACK) { 1977 TCPSTAT_INC(tcps_connects); 1978 soisconnected(so); 1979 #ifdef MAC 1980 mac_socketpeer_set_from_mbuf(m, so); 1981 #endif 1982 /* Do window scaling on this connection? */ 1983 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1984 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1985 tp->rcv_scale = tp->request_r_scale; 1986 } 1987 tp->rcv_adv += imin(tp->rcv_wnd, 1988 TCP_MAXWIN << tp->rcv_scale); 1989 tp->snd_una++; /* SYN is acked */ 1990 /* 1991 * If there's data, delay ACK; if there's also a FIN 1992 * ACKNOW will be turned on later. 1993 */ 1994 if (DELAY_ACK(tp, tlen) && tlen != 0) 1995 tcp_timer_activate(tp, TT_DELACK, 1996 tcp_delacktime); 1997 else 1998 tp->t_flags |= TF_ACKNOW; 1999 2000 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 2001 tp->t_flags |= TF_ECN_PERMIT; 2002 TCPSTAT_INC(tcps_ecn_shs); 2003 } 2004 2005 /* 2006 * Received <SYN,ACK> in SYN_SENT[*] state. 2007 * Transitions: 2008 * SYN_SENT --> ESTABLISHED 2009 * SYN_SENT* --> FIN_WAIT_1 2010 */ 2011 tp->t_starttime = ticks; 2012 if (tp->t_flags & TF_NEEDFIN) { 2013 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2014 tp->t_flags &= ~TF_NEEDFIN; 2015 thflags &= ~TH_SYN; 2016 } else { 2017 tcp_state_change(tp, TCPS_ESTABLISHED); 2018 TCP_PROBE5(connect__established, NULL, tp, 2019 mtod(m, const char *), tp, th); 2020 cc_conn_init(tp); 2021 tcp_timer_activate(tp, TT_KEEP, 2022 TP_KEEPIDLE(tp)); 2023 } 2024 } else { 2025 /* 2026 * Received initial SYN in SYN-SENT[*] state => 2027 * simultaneous open. 2028 * If it succeeds, connection is * half-synchronized. 2029 * Otherwise, do 3-way handshake: 2030 * SYN-SENT -> SYN-RECEIVED 2031 * SYN-SENT* -> SYN-RECEIVED* 2032 */ 2033 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2034 tcp_timer_activate(tp, TT_REXMT, 0); 2035 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2036 } 2037 2038 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: " 2039 "ti_locked %d", __func__, ti_locked)); 2040 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2041 INP_WLOCK_ASSERT(tp->t_inpcb); 2042 2043 /* 2044 * Advance th->th_seq to correspond to first data byte. 2045 * If data, trim to stay within window, 2046 * dropping FIN if necessary. 2047 */ 2048 th->th_seq++; 2049 if (tlen > tp->rcv_wnd) { 2050 todrop = tlen - tp->rcv_wnd; 2051 m_adj(m, -todrop); 2052 tlen = tp->rcv_wnd; 2053 thflags &= ~TH_FIN; 2054 TCPSTAT_INC(tcps_rcvpackafterwin); 2055 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2056 } 2057 tp->snd_wl1 = th->th_seq - 1; 2058 tp->rcv_up = th->th_seq; 2059 /* 2060 * Client side of transaction: already sent SYN and data. 2061 * If the remote host used T/TCP to validate the SYN, 2062 * our data will be ACK'd; if so, enter normal data segment 2063 * processing in the middle of step 5, ack processing. 2064 * Otherwise, goto step 6. 2065 */ 2066 if (thflags & TH_ACK) 2067 goto process_ACK; 2068 2069 goto step6; 2070 2071 /* 2072 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2073 * do normal processing. 2074 * 2075 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2076 */ 2077 case TCPS_LAST_ACK: 2078 case TCPS_CLOSING: 2079 break; /* continue normal processing */ 2080 } 2081 2082 /* 2083 * States other than LISTEN or SYN_SENT. 2084 * First check the RST flag and sequence number since reset segments 2085 * are exempt from the timestamp and connection count tests. This 2086 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2087 * below which allowed reset segments in half the sequence space 2088 * to fall though and be processed (which gives forged reset 2089 * segments with a random sequence number a 50 percent chance of 2090 * killing a connection). 2091 * Then check timestamp, if present. 2092 * Then check the connection count, if present. 2093 * Then check that at least some bytes of segment are within 2094 * receive window. If segment begins before rcv_nxt, 2095 * drop leading data (and SYN); if nothing left, just ack. 2096 */ 2097 if (thflags & TH_RST) { 2098 /* 2099 * RFC5961 Section 3.2 2100 * 2101 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2102 * - If RST is in window, we send challenge ACK. 2103 * 2104 * Note: to take into account delayed ACKs, we should 2105 * test against last_ack_sent instead of rcv_nxt. 2106 * Note 2: we handle special case of closed window, not 2107 * covered by the RFC. 2108 */ 2109 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2110 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2111 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2112 2113 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2114 KASSERT(ti_locked == TI_RLOCKED, 2115 ("%s: TH_RST ti_locked %d, th %p tp %p", 2116 __func__, ti_locked, th, tp)); 2117 KASSERT(tp->t_state != TCPS_SYN_SENT, 2118 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2119 __func__, th, tp)); 2120 2121 if (V_tcp_insecure_rst || 2122 tp->last_ack_sent == th->th_seq) { 2123 TCPSTAT_INC(tcps_drops); 2124 /* Drop the connection. */ 2125 switch (tp->t_state) { 2126 case TCPS_SYN_RECEIVED: 2127 so->so_error = ECONNREFUSED; 2128 goto close; 2129 case TCPS_ESTABLISHED: 2130 case TCPS_FIN_WAIT_1: 2131 case TCPS_FIN_WAIT_2: 2132 case TCPS_CLOSE_WAIT: 2133 so->so_error = ECONNRESET; 2134 close: 2135 tcp_state_change(tp, TCPS_CLOSED); 2136 /* FALLTHROUGH */ 2137 default: 2138 tp = tcp_close(tp); 2139 } 2140 } else { 2141 TCPSTAT_INC(tcps_badrst); 2142 /* Send challenge ACK. */ 2143 tcp_respond(tp, mtod(m, void *), th, m, 2144 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2145 tp->last_ack_sent = tp->rcv_nxt; 2146 m = NULL; 2147 } 2148 } 2149 goto drop; 2150 } 2151 2152 /* 2153 * RFC5961 Section 4.2 2154 * Send challenge ACK for any SYN in synchronized state. 2155 */ 2156 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT) { 2157 KASSERT(ti_locked == TI_RLOCKED, 2158 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2159 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2160 2161 TCPSTAT_INC(tcps_badsyn); 2162 if (V_tcp_insecure_syn && 2163 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2164 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2165 tp = tcp_drop(tp, ECONNRESET); 2166 rstreason = BANDLIM_UNLIMITED; 2167 } else { 2168 /* Send challenge ACK. */ 2169 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2170 tp->snd_nxt, TH_ACK); 2171 tp->last_ack_sent = tp->rcv_nxt; 2172 m = NULL; 2173 } 2174 goto drop; 2175 } 2176 2177 /* 2178 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2179 * and it's less than ts_recent, drop it. 2180 */ 2181 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2182 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2183 2184 /* Check to see if ts_recent is over 24 days old. */ 2185 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2186 /* 2187 * Invalidate ts_recent. If this segment updates 2188 * ts_recent, the age will be reset later and ts_recent 2189 * will get a valid value. If it does not, setting 2190 * ts_recent to zero will at least satisfy the 2191 * requirement that zero be placed in the timestamp 2192 * echo reply when ts_recent isn't valid. The 2193 * age isn't reset until we get a valid ts_recent 2194 * because we don't want out-of-order segments to be 2195 * dropped when ts_recent is old. 2196 */ 2197 tp->ts_recent = 0; 2198 } else { 2199 TCPSTAT_INC(tcps_rcvduppack); 2200 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2201 TCPSTAT_INC(tcps_pawsdrop); 2202 if (tlen) 2203 goto dropafterack; 2204 goto drop; 2205 } 2206 } 2207 2208 /* 2209 * In the SYN-RECEIVED state, validate that the packet belongs to 2210 * this connection before trimming the data to fit the receive 2211 * window. Check the sequence number versus IRS since we know 2212 * the sequence numbers haven't wrapped. This is a partial fix 2213 * for the "LAND" DoS attack. 2214 */ 2215 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2216 rstreason = BANDLIM_RST_OPENPORT; 2217 goto dropwithreset; 2218 } 2219 2220 todrop = tp->rcv_nxt - th->th_seq; 2221 if (todrop > 0) { 2222 if (thflags & TH_SYN) { 2223 thflags &= ~TH_SYN; 2224 th->th_seq++; 2225 if (th->th_urp > 1) 2226 th->th_urp--; 2227 else 2228 thflags &= ~TH_URG; 2229 todrop--; 2230 } 2231 /* 2232 * Following if statement from Stevens, vol. 2, p. 960. 2233 */ 2234 if (todrop > tlen 2235 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2236 /* 2237 * Any valid FIN must be to the left of the window. 2238 * At this point the FIN must be a duplicate or out 2239 * of sequence; drop it. 2240 */ 2241 thflags &= ~TH_FIN; 2242 2243 /* 2244 * Send an ACK to resynchronize and drop any data. 2245 * But keep on processing for RST or ACK. 2246 */ 2247 tp->t_flags |= TF_ACKNOW; 2248 todrop = tlen; 2249 TCPSTAT_INC(tcps_rcvduppack); 2250 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2251 } else { 2252 TCPSTAT_INC(tcps_rcvpartduppack); 2253 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2254 } 2255 drop_hdrlen += todrop; /* drop from the top afterwards */ 2256 th->th_seq += todrop; 2257 tlen -= todrop; 2258 if (th->th_urp > todrop) 2259 th->th_urp -= todrop; 2260 else { 2261 thflags &= ~TH_URG; 2262 th->th_urp = 0; 2263 } 2264 } 2265 2266 /* 2267 * If new data are received on a connection after the 2268 * user processes are gone, then RST the other end. 2269 */ 2270 if ((so->so_state & SS_NOFDREF) && 2271 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2272 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && " 2273 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2274 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2275 2276 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2277 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2278 "after socket was closed, " 2279 "sending RST and removing tcpcb\n", 2280 s, __func__, tcpstates[tp->t_state], tlen); 2281 free(s, M_TCPLOG); 2282 } 2283 tp = tcp_close(tp); 2284 TCPSTAT_INC(tcps_rcvafterclose); 2285 rstreason = BANDLIM_UNLIMITED; 2286 goto dropwithreset; 2287 } 2288 2289 /* 2290 * If segment ends after window, drop trailing data 2291 * (and PUSH and FIN); if nothing left, just ACK. 2292 */ 2293 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2294 if (todrop > 0) { 2295 TCPSTAT_INC(tcps_rcvpackafterwin); 2296 if (todrop >= tlen) { 2297 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2298 /* 2299 * If window is closed can only take segments at 2300 * window edge, and have to drop data and PUSH from 2301 * incoming segments. Continue processing, but 2302 * remember to ack. Otherwise, drop segment 2303 * and ack. 2304 */ 2305 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2306 tp->t_flags |= TF_ACKNOW; 2307 TCPSTAT_INC(tcps_rcvwinprobe); 2308 } else 2309 goto dropafterack; 2310 } else 2311 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2312 m_adj(m, -todrop); 2313 tlen -= todrop; 2314 thflags &= ~(TH_PUSH|TH_FIN); 2315 } 2316 2317 /* 2318 * If last ACK falls within this segment's sequence numbers, 2319 * record its timestamp. 2320 * NOTE: 2321 * 1) That the test incorporates suggestions from the latest 2322 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2323 * 2) That updating only on newer timestamps interferes with 2324 * our earlier PAWS tests, so this check should be solely 2325 * predicated on the sequence space of this segment. 2326 * 3) That we modify the segment boundary check to be 2327 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2328 * instead of RFC1323's 2329 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2330 * This modified check allows us to overcome RFC1323's 2331 * limitations as described in Stevens TCP/IP Illustrated 2332 * Vol. 2 p.869. In such cases, we can still calculate the 2333 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2334 */ 2335 if ((to.to_flags & TOF_TS) != 0 && 2336 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2337 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2338 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2339 tp->ts_recent_age = tcp_ts_getticks(); 2340 tp->ts_recent = to.to_tsval; 2341 } 2342 2343 /* 2344 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2345 * flag is on (half-synchronized state), then queue data for 2346 * later processing; else drop segment and return. 2347 */ 2348 if ((thflags & TH_ACK) == 0) { 2349 if (tp->t_state == TCPS_SYN_RECEIVED || 2350 (tp->t_flags & TF_NEEDSYN)) 2351 goto step6; 2352 else if (tp->t_flags & TF_ACKNOW) 2353 goto dropafterack; 2354 else 2355 goto drop; 2356 } 2357 2358 /* 2359 * Ack processing. 2360 */ 2361 switch (tp->t_state) { 2362 2363 /* 2364 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2365 * ESTABLISHED state and continue processing. 2366 * The ACK was checked above. 2367 */ 2368 case TCPS_SYN_RECEIVED: 2369 2370 TCPSTAT_INC(tcps_connects); 2371 soisconnected(so); 2372 /* Do window scaling? */ 2373 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2374 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2375 tp->rcv_scale = tp->request_r_scale; 2376 tp->snd_wnd = tiwin; 2377 } 2378 /* 2379 * Make transitions: 2380 * SYN-RECEIVED -> ESTABLISHED 2381 * SYN-RECEIVED* -> FIN-WAIT-1 2382 */ 2383 tp->t_starttime = ticks; 2384 if (tp->t_flags & TF_NEEDFIN) { 2385 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2386 tp->t_flags &= ~TF_NEEDFIN; 2387 } else { 2388 tcp_state_change(tp, TCPS_ESTABLISHED); 2389 TCP_PROBE5(accept__established, NULL, tp, 2390 mtod(m, const char *), tp, th); 2391 cc_conn_init(tp); 2392 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2393 } 2394 /* 2395 * If segment contains data or ACK, will call tcp_reass() 2396 * later; if not, do so now to pass queued data to user. 2397 */ 2398 if (tlen == 0 && (thflags & TH_FIN) == 0) 2399 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2400 (struct mbuf *)0); 2401 tp->snd_wl1 = th->th_seq - 1; 2402 /* FALLTHROUGH */ 2403 2404 /* 2405 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2406 * ACKs. If the ack is in the range 2407 * tp->snd_una < th->th_ack <= tp->snd_max 2408 * then advance tp->snd_una to th->th_ack and drop 2409 * data from the retransmission queue. If this ACK reflects 2410 * more up to date window information we update our window information. 2411 */ 2412 case TCPS_ESTABLISHED: 2413 case TCPS_FIN_WAIT_1: 2414 case TCPS_FIN_WAIT_2: 2415 case TCPS_CLOSE_WAIT: 2416 case TCPS_CLOSING: 2417 case TCPS_LAST_ACK: 2418 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2419 TCPSTAT_INC(tcps_rcvacktoomuch); 2420 goto dropafterack; 2421 } 2422 if ((tp->t_flags & TF_SACK_PERMIT) && 2423 ((to.to_flags & TOF_SACK) || 2424 !TAILQ_EMPTY(&tp->snd_holes))) 2425 tcp_sack_doack(tp, &to, th->th_ack); 2426 2427 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2428 hhook_run_tcp_est_in(tp, th, &to); 2429 2430 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2431 if (tlen == 0 && tiwin == tp->snd_wnd) { 2432 /* 2433 * If this is the first time we've seen a 2434 * FIN from the remote, this is not a 2435 * duplicate and it needs to be processed 2436 * normally. This happens during a 2437 * simultaneous close. 2438 */ 2439 if ((thflags & TH_FIN) && 2440 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2441 tp->t_dupacks = 0; 2442 break; 2443 } 2444 TCPSTAT_INC(tcps_rcvdupack); 2445 /* 2446 * If we have outstanding data (other than 2447 * a window probe), this is a completely 2448 * duplicate ack (ie, window info didn't 2449 * change and FIN isn't set), 2450 * the ack is the biggest we've 2451 * seen and we've seen exactly our rexmt 2452 * threshhold of them, assume a packet 2453 * has been dropped and retransmit it. 2454 * Kludge snd_nxt & the congestion 2455 * window so we send only this one 2456 * packet. 2457 * 2458 * We know we're losing at the current 2459 * window size so do congestion avoidance 2460 * (set ssthresh to half the current window 2461 * and pull our congestion window back to 2462 * the new ssthresh). 2463 * 2464 * Dup acks mean that packets have left the 2465 * network (they're now cached at the receiver) 2466 * so bump cwnd by the amount in the receiver 2467 * to keep a constant cwnd packets in the 2468 * network. 2469 * 2470 * When using TCP ECN, notify the peer that 2471 * we reduced the cwnd. 2472 */ 2473 if (!tcp_timer_active(tp, TT_REXMT) || 2474 th->th_ack != tp->snd_una) 2475 tp->t_dupacks = 0; 2476 else if (++tp->t_dupacks > tcprexmtthresh || 2477 IN_FASTRECOVERY(tp->t_flags)) { 2478 cc_ack_received(tp, th, CC_DUPACK); 2479 if ((tp->t_flags & TF_SACK_PERMIT) && 2480 IN_FASTRECOVERY(tp->t_flags)) { 2481 int awnd; 2482 2483 /* 2484 * Compute the amount of data in flight first. 2485 * We can inject new data into the pipe iff 2486 * we have less than 1/2 the original window's 2487 * worth of data in flight. 2488 */ 2489 awnd = (tp->snd_nxt - tp->snd_fack) + 2490 tp->sackhint.sack_bytes_rexmit; 2491 if (awnd < tp->snd_ssthresh) { 2492 tp->snd_cwnd += tp->t_maxseg; 2493 if (tp->snd_cwnd > tp->snd_ssthresh) 2494 tp->snd_cwnd = tp->snd_ssthresh; 2495 } 2496 } else 2497 tp->snd_cwnd += tp->t_maxseg; 2498 (void) tcp_output(tp); 2499 goto drop; 2500 } else if (tp->t_dupacks == tcprexmtthresh) { 2501 tcp_seq onxt = tp->snd_nxt; 2502 2503 /* 2504 * If we're doing sack, check to 2505 * see if we're already in sack 2506 * recovery. If we're not doing sack, 2507 * check to see if we're in newreno 2508 * recovery. 2509 */ 2510 if (tp->t_flags & TF_SACK_PERMIT) { 2511 if (IN_FASTRECOVERY(tp->t_flags)) { 2512 tp->t_dupacks = 0; 2513 break; 2514 } 2515 } else { 2516 if (SEQ_LEQ(th->th_ack, 2517 tp->snd_recover)) { 2518 tp->t_dupacks = 0; 2519 break; 2520 } 2521 } 2522 /* Congestion signal before ack. */ 2523 cc_cong_signal(tp, th, CC_NDUPACK); 2524 cc_ack_received(tp, th, CC_DUPACK); 2525 tcp_timer_activate(tp, TT_REXMT, 0); 2526 tp->t_rtttime = 0; 2527 if (tp->t_flags & TF_SACK_PERMIT) { 2528 TCPSTAT_INC( 2529 tcps_sack_recovery_episode); 2530 tp->sack_newdata = tp->snd_nxt; 2531 tp->snd_cwnd = tp->t_maxseg; 2532 (void) tcp_output(tp); 2533 goto drop; 2534 } 2535 tp->snd_nxt = th->th_ack; 2536 tp->snd_cwnd = tp->t_maxseg; 2537 (void) tcp_output(tp); 2538 KASSERT(tp->snd_limited <= 2, 2539 ("%s: tp->snd_limited too big", 2540 __func__)); 2541 tp->snd_cwnd = tp->snd_ssthresh + 2542 tp->t_maxseg * 2543 (tp->t_dupacks - tp->snd_limited); 2544 if (SEQ_GT(onxt, tp->snd_nxt)) 2545 tp->snd_nxt = onxt; 2546 goto drop; 2547 } else if (V_tcp_do_rfc3042) { 2548 cc_ack_received(tp, th, CC_DUPACK); 2549 u_long oldcwnd = tp->snd_cwnd; 2550 tcp_seq oldsndmax = tp->snd_max; 2551 u_int sent; 2552 int avail; 2553 2554 KASSERT(tp->t_dupacks == 1 || 2555 tp->t_dupacks == 2, 2556 ("%s: dupacks not 1 or 2", 2557 __func__)); 2558 if (tp->t_dupacks == 1) 2559 tp->snd_limited = 0; 2560 tp->snd_cwnd = 2561 (tp->snd_nxt - tp->snd_una) + 2562 (tp->t_dupacks - tp->snd_limited) * 2563 tp->t_maxseg; 2564 /* 2565 * Only call tcp_output when there 2566 * is new data available to be sent. 2567 * Otherwise we would send pure ACKs. 2568 */ 2569 SOCKBUF_LOCK(&so->so_snd); 2570 avail = sbavail(&so->so_snd) - 2571 (tp->snd_nxt - tp->snd_una); 2572 SOCKBUF_UNLOCK(&so->so_snd); 2573 if (avail > 0) 2574 (void) tcp_output(tp); 2575 sent = tp->snd_max - oldsndmax; 2576 if (sent > tp->t_maxseg) { 2577 KASSERT((tp->t_dupacks == 2 && 2578 tp->snd_limited == 0) || 2579 (sent == tp->t_maxseg + 1 && 2580 tp->t_flags & TF_SENTFIN), 2581 ("%s: sent too much", 2582 __func__)); 2583 tp->snd_limited = 2; 2584 } else if (sent > 0) 2585 ++tp->snd_limited; 2586 tp->snd_cwnd = oldcwnd; 2587 goto drop; 2588 } 2589 } else 2590 tp->t_dupacks = 0; 2591 break; 2592 } 2593 2594 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2595 ("%s: th_ack <= snd_una", __func__)); 2596 2597 /* 2598 * If the congestion window was inflated to account 2599 * for the other side's cached packets, retract it. 2600 */ 2601 if (IN_FASTRECOVERY(tp->t_flags)) { 2602 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2603 if (tp->t_flags & TF_SACK_PERMIT) 2604 tcp_sack_partialack(tp, th); 2605 else 2606 tcp_newreno_partial_ack(tp, th); 2607 } else 2608 cc_post_recovery(tp, th); 2609 } 2610 tp->t_dupacks = 0; 2611 /* 2612 * If we reach this point, ACK is not a duplicate, 2613 * i.e., it ACKs something we sent. 2614 */ 2615 if (tp->t_flags & TF_NEEDSYN) { 2616 /* 2617 * T/TCP: Connection was half-synchronized, and our 2618 * SYN has been ACK'd (so connection is now fully 2619 * synchronized). Go to non-starred state, 2620 * increment snd_una for ACK of SYN, and check if 2621 * we can do window scaling. 2622 */ 2623 tp->t_flags &= ~TF_NEEDSYN; 2624 tp->snd_una++; 2625 /* Do window scaling? */ 2626 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2627 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2628 tp->rcv_scale = tp->request_r_scale; 2629 /* Send window already scaled. */ 2630 } 2631 } 2632 2633 process_ACK: 2634 INP_WLOCK_ASSERT(tp->t_inpcb); 2635 2636 acked = BYTES_THIS_ACK(tp, th); 2637 TCPSTAT_INC(tcps_rcvackpack); 2638 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2639 2640 /* 2641 * If we just performed our first retransmit, and the ACK 2642 * arrives within our recovery window, then it was a mistake 2643 * to do the retransmit in the first place. Recover our 2644 * original cwnd and ssthresh, and proceed to transmit where 2645 * we left off. 2646 */ 2647 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2648 (int)(ticks - tp->t_badrxtwin) < 0) 2649 cc_cong_signal(tp, th, CC_RTO_ERR); 2650 2651 /* 2652 * If we have a timestamp reply, update smoothed 2653 * round trip time. If no timestamp is present but 2654 * transmit timer is running and timed sequence 2655 * number was acked, update smoothed round trip time. 2656 * Since we now have an rtt measurement, cancel the 2657 * timer backoff (cf., Phil Karn's retransmit alg.). 2658 * Recompute the initial retransmit timer. 2659 * 2660 * Some boxes send broken timestamp replies 2661 * during the SYN+ACK phase, ignore 2662 * timestamps of 0 or we could calculate a 2663 * huge RTT and blow up the retransmit timer. 2664 */ 2665 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2666 u_int t; 2667 2668 t = tcp_ts_getticks() - to.to_tsecr; 2669 if (!tp->t_rttlow || tp->t_rttlow > t) 2670 tp->t_rttlow = t; 2671 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2672 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2673 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2674 tp->t_rttlow = ticks - tp->t_rtttime; 2675 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2676 } 2677 2678 /* 2679 * If all outstanding data is acked, stop retransmit 2680 * timer and remember to restart (more output or persist). 2681 * If there is more data to be acked, restart retransmit 2682 * timer, using current (possibly backed-off) value. 2683 */ 2684 if (th->th_ack == tp->snd_max) { 2685 tcp_timer_activate(tp, TT_REXMT, 0); 2686 needoutput = 1; 2687 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2688 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2689 2690 /* 2691 * If no data (only SYN) was ACK'd, 2692 * skip rest of ACK processing. 2693 */ 2694 if (acked == 0) 2695 goto step6; 2696 2697 /* 2698 * Let the congestion control algorithm update congestion 2699 * control related information. This typically means increasing 2700 * the congestion window. 2701 */ 2702 cc_ack_received(tp, th, CC_ACK); 2703 2704 SOCKBUF_LOCK(&so->so_snd); 2705 if (acked > sbavail(&so->so_snd)) { 2706 tp->snd_wnd -= sbavail(&so->so_snd); 2707 mfree = sbcut_locked(&so->so_snd, 2708 (int)sbavail(&so->so_snd)); 2709 ourfinisacked = 1; 2710 } else { 2711 mfree = sbcut_locked(&so->so_snd, acked); 2712 tp->snd_wnd -= acked; 2713 ourfinisacked = 0; 2714 } 2715 /* NB: sowwakeup_locked() does an implicit unlock. */ 2716 sowwakeup_locked(so); 2717 m_freem(mfree); 2718 /* Detect una wraparound. */ 2719 if (!IN_RECOVERY(tp->t_flags) && 2720 SEQ_GT(tp->snd_una, tp->snd_recover) && 2721 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2722 tp->snd_recover = th->th_ack - 1; 2723 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2724 if (IN_RECOVERY(tp->t_flags) && 2725 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2726 EXIT_RECOVERY(tp->t_flags); 2727 } 2728 tp->snd_una = th->th_ack; 2729 if (tp->t_flags & TF_SACK_PERMIT) { 2730 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2731 tp->snd_recover = tp->snd_una; 2732 } 2733 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2734 tp->snd_nxt = tp->snd_una; 2735 2736 switch (tp->t_state) { 2737 2738 /* 2739 * In FIN_WAIT_1 STATE in addition to the processing 2740 * for the ESTABLISHED state if our FIN is now acknowledged 2741 * then enter FIN_WAIT_2. 2742 */ 2743 case TCPS_FIN_WAIT_1: 2744 if (ourfinisacked) { 2745 /* 2746 * If we can't receive any more 2747 * data, then closing user can proceed. 2748 * Starting the timer is contrary to the 2749 * specification, but if we don't get a FIN 2750 * we'll hang forever. 2751 * 2752 * XXXjl: 2753 * we should release the tp also, and use a 2754 * compressed state. 2755 */ 2756 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2757 soisdisconnected(so); 2758 tcp_timer_activate(tp, TT_2MSL, 2759 (tcp_fast_finwait2_recycle ? 2760 tcp_finwait2_timeout : 2761 TP_MAXIDLE(tp))); 2762 } 2763 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2764 } 2765 break; 2766 2767 /* 2768 * In CLOSING STATE in addition to the processing for 2769 * the ESTABLISHED state if the ACK acknowledges our FIN 2770 * then enter the TIME-WAIT state, otherwise ignore 2771 * the segment. 2772 */ 2773 case TCPS_CLOSING: 2774 if (ourfinisacked) { 2775 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2776 tcp_twstart(tp); 2777 INP_INFO_RUNLOCK(&V_tcbinfo); 2778 m_freem(m); 2779 return; 2780 } 2781 break; 2782 2783 /* 2784 * In LAST_ACK, we may still be waiting for data to drain 2785 * and/or to be acked, as well as for the ack of our FIN. 2786 * If our FIN is now acknowledged, delete the TCB, 2787 * enter the closed state and return. 2788 */ 2789 case TCPS_LAST_ACK: 2790 if (ourfinisacked) { 2791 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 2792 tp = tcp_close(tp); 2793 goto drop; 2794 } 2795 break; 2796 } 2797 } 2798 2799 step6: 2800 INP_WLOCK_ASSERT(tp->t_inpcb); 2801 2802 /* 2803 * Update window information. 2804 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2805 */ 2806 if ((thflags & TH_ACK) && 2807 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2808 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2809 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2810 /* keep track of pure window updates */ 2811 if (tlen == 0 && 2812 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2813 TCPSTAT_INC(tcps_rcvwinupd); 2814 tp->snd_wnd = tiwin; 2815 tp->snd_wl1 = th->th_seq; 2816 tp->snd_wl2 = th->th_ack; 2817 if (tp->snd_wnd > tp->max_sndwnd) 2818 tp->max_sndwnd = tp->snd_wnd; 2819 needoutput = 1; 2820 } 2821 2822 /* 2823 * Process segments with URG. 2824 */ 2825 if ((thflags & TH_URG) && th->th_urp && 2826 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2827 /* 2828 * This is a kludge, but if we receive and accept 2829 * random urgent pointers, we'll crash in 2830 * soreceive. It's hard to imagine someone 2831 * actually wanting to send this much urgent data. 2832 */ 2833 SOCKBUF_LOCK(&so->so_rcv); 2834 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2835 th->th_urp = 0; /* XXX */ 2836 thflags &= ~TH_URG; /* XXX */ 2837 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2838 goto dodata; /* XXX */ 2839 } 2840 /* 2841 * If this segment advances the known urgent pointer, 2842 * then mark the data stream. This should not happen 2843 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2844 * a FIN has been received from the remote side. 2845 * In these states we ignore the URG. 2846 * 2847 * According to RFC961 (Assigned Protocols), 2848 * the urgent pointer points to the last octet 2849 * of urgent data. We continue, however, 2850 * to consider it to indicate the first octet 2851 * of data past the urgent section as the original 2852 * spec states (in one of two places). 2853 */ 2854 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2855 tp->rcv_up = th->th_seq + th->th_urp; 2856 so->so_oobmark = sbavail(&so->so_rcv) + 2857 (tp->rcv_up - tp->rcv_nxt) - 1; 2858 if (so->so_oobmark == 0) 2859 so->so_rcv.sb_state |= SBS_RCVATMARK; 2860 sohasoutofband(so); 2861 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2862 } 2863 SOCKBUF_UNLOCK(&so->so_rcv); 2864 /* 2865 * Remove out of band data so doesn't get presented to user. 2866 * This can happen independent of advancing the URG pointer, 2867 * but if two URG's are pending at once, some out-of-band 2868 * data may creep in... ick. 2869 */ 2870 if (th->th_urp <= (u_long)tlen && 2871 !(so->so_options & SO_OOBINLINE)) { 2872 /* hdr drop is delayed */ 2873 tcp_pulloutofband(so, th, m, drop_hdrlen); 2874 } 2875 } else { 2876 /* 2877 * If no out of band data is expected, 2878 * pull receive urgent pointer along 2879 * with the receive window. 2880 */ 2881 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2882 tp->rcv_up = tp->rcv_nxt; 2883 } 2884 dodata: /* XXX */ 2885 INP_WLOCK_ASSERT(tp->t_inpcb); 2886 2887 /* 2888 * Process the segment text, merging it into the TCP sequencing queue, 2889 * and arranging for acknowledgment of receipt if necessary. 2890 * This process logically involves adjusting tp->rcv_wnd as data 2891 * is presented to the user (this happens in tcp_usrreq.c, 2892 * case PRU_RCVD). If a FIN has already been received on this 2893 * connection then we just ignore the text. 2894 */ 2895 if ((tlen || (thflags & TH_FIN)) && 2896 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2897 tcp_seq save_start = th->th_seq; 2898 m_adj(m, drop_hdrlen); /* delayed header drop */ 2899 /* 2900 * Insert segment which includes th into TCP reassembly queue 2901 * with control block tp. Set thflags to whether reassembly now 2902 * includes a segment with FIN. This handles the common case 2903 * inline (segment is the next to be received on an established 2904 * connection, and the queue is empty), avoiding linkage into 2905 * and removal from the queue and repetition of various 2906 * conversions. 2907 * Set DELACK for segments received in order, but ack 2908 * immediately when segments are out of order (so 2909 * fast retransmit can work). 2910 */ 2911 if (th->th_seq == tp->rcv_nxt && 2912 LIST_EMPTY(&tp->t_segq) && 2913 TCPS_HAVEESTABLISHED(tp->t_state)) { 2914 if (DELAY_ACK(tp, tlen)) 2915 tp->t_flags |= TF_DELACK; 2916 else 2917 tp->t_flags |= TF_ACKNOW; 2918 tp->rcv_nxt += tlen; 2919 thflags = th->th_flags & TH_FIN; 2920 TCPSTAT_INC(tcps_rcvpack); 2921 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2922 ND6_HINT(tp); 2923 SOCKBUF_LOCK(&so->so_rcv); 2924 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2925 m_freem(m); 2926 else 2927 sbappendstream_locked(&so->so_rcv, m, 0); 2928 /* NB: sorwakeup_locked() does an implicit unlock. */ 2929 sorwakeup_locked(so); 2930 } else { 2931 /* 2932 * XXX: Due to the header drop above "th" is 2933 * theoretically invalid by now. Fortunately 2934 * m_adj() doesn't actually frees any mbufs 2935 * when trimming from the head. 2936 */ 2937 thflags = tcp_reass(tp, th, &tlen, m); 2938 tp->t_flags |= TF_ACKNOW; 2939 } 2940 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2941 tcp_update_sack_list(tp, save_start, save_start + tlen); 2942 #if 0 2943 /* 2944 * Note the amount of data that peer has sent into 2945 * our window, in order to estimate the sender's 2946 * buffer size. 2947 * XXX: Unused. 2948 */ 2949 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2950 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2951 else 2952 len = so->so_rcv.sb_hiwat; 2953 #endif 2954 } else { 2955 m_freem(m); 2956 thflags &= ~TH_FIN; 2957 } 2958 2959 /* 2960 * If FIN is received ACK the FIN and let the user know 2961 * that the connection is closing. 2962 */ 2963 if (thflags & TH_FIN) { 2964 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2965 socantrcvmore(so); 2966 /* 2967 * If connection is half-synchronized 2968 * (ie NEEDSYN flag on) then delay ACK, 2969 * so it may be piggybacked when SYN is sent. 2970 * Otherwise, since we received a FIN then no 2971 * more input can be expected, send ACK now. 2972 */ 2973 if (tp->t_flags & TF_NEEDSYN) 2974 tp->t_flags |= TF_DELACK; 2975 else 2976 tp->t_flags |= TF_ACKNOW; 2977 tp->rcv_nxt++; 2978 } 2979 switch (tp->t_state) { 2980 2981 /* 2982 * In SYN_RECEIVED and ESTABLISHED STATES 2983 * enter the CLOSE_WAIT state. 2984 */ 2985 case TCPS_SYN_RECEIVED: 2986 tp->t_starttime = ticks; 2987 /* FALLTHROUGH */ 2988 case TCPS_ESTABLISHED: 2989 tcp_state_change(tp, TCPS_CLOSE_WAIT); 2990 break; 2991 2992 /* 2993 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2994 * enter the CLOSING state. 2995 */ 2996 case TCPS_FIN_WAIT_1: 2997 tcp_state_change(tp, TCPS_CLOSING); 2998 break; 2999 3000 /* 3001 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3002 * starting the time-wait timer, turning off the other 3003 * standard timers. 3004 */ 3005 case TCPS_FIN_WAIT_2: 3006 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 3007 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata " 3008 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3009 ti_locked)); 3010 3011 tcp_twstart(tp); 3012 INP_INFO_RUNLOCK(&V_tcbinfo); 3013 return; 3014 } 3015 } 3016 if (ti_locked == TI_RLOCKED) 3017 INP_INFO_RUNLOCK(&V_tcbinfo); 3018 ti_locked = TI_UNLOCKED; 3019 3020 #ifdef TCPDEBUG 3021 if (so->so_options & SO_DEBUG) 3022 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3023 &tcp_savetcp, 0); 3024 #endif 3025 3026 /* 3027 * Return any desired output. 3028 */ 3029 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3030 (void) tcp_output(tp); 3031 3032 check_delack: 3033 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3034 __func__, ti_locked)); 3035 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3036 INP_WLOCK_ASSERT(tp->t_inpcb); 3037 3038 if (tp->t_flags & TF_DELACK) { 3039 tp->t_flags &= ~TF_DELACK; 3040 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3041 } 3042 INP_WUNLOCK(tp->t_inpcb); 3043 return; 3044 3045 dropafterack: 3046 /* 3047 * Generate an ACK dropping incoming segment if it occupies 3048 * sequence space, where the ACK reflects our state. 3049 * 3050 * We can now skip the test for the RST flag since all 3051 * paths to this code happen after packets containing 3052 * RST have been dropped. 3053 * 3054 * In the SYN-RECEIVED state, don't send an ACK unless the 3055 * segment we received passes the SYN-RECEIVED ACK test. 3056 * If it fails send a RST. This breaks the loop in the 3057 * "LAND" DoS attack, and also prevents an ACK storm 3058 * between two listening ports that have been sent forged 3059 * SYN segments, each with the source address of the other. 3060 */ 3061 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3062 (SEQ_GT(tp->snd_una, th->th_ack) || 3063 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3064 rstreason = BANDLIM_RST_OPENPORT; 3065 goto dropwithreset; 3066 } 3067 #ifdef TCPDEBUG 3068 if (so->so_options & SO_DEBUG) 3069 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3070 &tcp_savetcp, 0); 3071 #endif 3072 if (ti_locked == TI_RLOCKED) 3073 INP_INFO_RUNLOCK(&V_tcbinfo); 3074 ti_locked = TI_UNLOCKED; 3075 3076 tp->t_flags |= TF_ACKNOW; 3077 (void) tcp_output(tp); 3078 INP_WUNLOCK(tp->t_inpcb); 3079 m_freem(m); 3080 return; 3081 3082 dropwithreset: 3083 if (ti_locked == TI_RLOCKED) 3084 INP_INFO_RUNLOCK(&V_tcbinfo); 3085 ti_locked = TI_UNLOCKED; 3086 3087 if (tp != NULL) { 3088 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3089 INP_WUNLOCK(tp->t_inpcb); 3090 } else 3091 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3092 return; 3093 3094 drop: 3095 if (ti_locked == TI_RLOCKED) { 3096 INP_INFO_RUNLOCK(&V_tcbinfo); 3097 ti_locked = TI_UNLOCKED; 3098 } 3099 #ifdef INVARIANTS 3100 else 3101 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3102 #endif 3103 3104 /* 3105 * Drop space held by incoming segment and return. 3106 */ 3107 #ifdef TCPDEBUG 3108 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3109 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3110 &tcp_savetcp, 0); 3111 #endif 3112 if (tp != NULL) 3113 INP_WUNLOCK(tp->t_inpcb); 3114 m_freem(m); 3115 } 3116 3117 /* 3118 * Issue RST and make ACK acceptable to originator of segment. 3119 * The mbuf must still include the original packet header. 3120 * tp may be NULL. 3121 */ 3122 static void 3123 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3124 int tlen, int rstreason) 3125 { 3126 #ifdef INET 3127 struct ip *ip; 3128 #endif 3129 #ifdef INET6 3130 struct ip6_hdr *ip6; 3131 #endif 3132 3133 if (tp != NULL) { 3134 INP_WLOCK_ASSERT(tp->t_inpcb); 3135 } 3136 3137 /* Don't bother if destination was broadcast/multicast. */ 3138 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3139 goto drop; 3140 #ifdef INET6 3141 if (mtod(m, struct ip *)->ip_v == 6) { 3142 ip6 = mtod(m, struct ip6_hdr *); 3143 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3144 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3145 goto drop; 3146 /* IPv6 anycast check is done at tcp6_input() */ 3147 } 3148 #endif 3149 #if defined(INET) && defined(INET6) 3150 else 3151 #endif 3152 #ifdef INET 3153 { 3154 ip = mtod(m, struct ip *); 3155 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3156 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3157 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3158 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3159 goto drop; 3160 } 3161 #endif 3162 3163 /* Perform bandwidth limiting. */ 3164 if (badport_bandlim(rstreason) < 0) 3165 goto drop; 3166 3167 /* tcp_respond consumes the mbuf chain. */ 3168 if (th->th_flags & TH_ACK) { 3169 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3170 th->th_ack, TH_RST); 3171 } else { 3172 if (th->th_flags & TH_SYN) 3173 tlen++; 3174 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3175 (tcp_seq)0, TH_RST|TH_ACK); 3176 } 3177 return; 3178 drop: 3179 m_freem(m); 3180 } 3181 3182 /* 3183 * Parse TCP options and place in tcpopt. 3184 */ 3185 static void 3186 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3187 { 3188 int opt, optlen; 3189 3190 to->to_flags = 0; 3191 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3192 opt = cp[0]; 3193 if (opt == TCPOPT_EOL) 3194 break; 3195 if (opt == TCPOPT_NOP) 3196 optlen = 1; 3197 else { 3198 if (cnt < 2) 3199 break; 3200 optlen = cp[1]; 3201 if (optlen < 2 || optlen > cnt) 3202 break; 3203 } 3204 switch (opt) { 3205 case TCPOPT_MAXSEG: 3206 if (optlen != TCPOLEN_MAXSEG) 3207 continue; 3208 if (!(flags & TO_SYN)) 3209 continue; 3210 to->to_flags |= TOF_MSS; 3211 bcopy((char *)cp + 2, 3212 (char *)&to->to_mss, sizeof(to->to_mss)); 3213 to->to_mss = ntohs(to->to_mss); 3214 break; 3215 case TCPOPT_WINDOW: 3216 if (optlen != TCPOLEN_WINDOW) 3217 continue; 3218 if (!(flags & TO_SYN)) 3219 continue; 3220 to->to_flags |= TOF_SCALE; 3221 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3222 break; 3223 case TCPOPT_TIMESTAMP: 3224 if (optlen != TCPOLEN_TIMESTAMP) 3225 continue; 3226 to->to_flags |= TOF_TS; 3227 bcopy((char *)cp + 2, 3228 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3229 to->to_tsval = ntohl(to->to_tsval); 3230 bcopy((char *)cp + 6, 3231 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3232 to->to_tsecr = ntohl(to->to_tsecr); 3233 break; 3234 #ifdef TCP_SIGNATURE 3235 /* 3236 * XXX In order to reply to a host which has set the 3237 * TCP_SIGNATURE option in its initial SYN, we have to 3238 * record the fact that the option was observed here 3239 * for the syncache code to perform the correct response. 3240 */ 3241 case TCPOPT_SIGNATURE: 3242 if (optlen != TCPOLEN_SIGNATURE) 3243 continue; 3244 to->to_flags |= TOF_SIGNATURE; 3245 to->to_signature = cp + 2; 3246 break; 3247 #endif 3248 case TCPOPT_SACK_PERMITTED: 3249 if (optlen != TCPOLEN_SACK_PERMITTED) 3250 continue; 3251 if (!(flags & TO_SYN)) 3252 continue; 3253 if (!V_tcp_do_sack) 3254 continue; 3255 to->to_flags |= TOF_SACKPERM; 3256 break; 3257 case TCPOPT_SACK: 3258 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3259 continue; 3260 if (flags & TO_SYN) 3261 continue; 3262 to->to_flags |= TOF_SACK; 3263 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3264 to->to_sacks = cp + 2; 3265 TCPSTAT_INC(tcps_sack_rcv_blocks); 3266 break; 3267 default: 3268 continue; 3269 } 3270 } 3271 } 3272 3273 /* 3274 * Pull out of band byte out of a segment so 3275 * it doesn't appear in the user's data queue. 3276 * It is still reflected in the segment length for 3277 * sequencing purposes. 3278 */ 3279 static void 3280 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3281 int off) 3282 { 3283 int cnt = off + th->th_urp - 1; 3284 3285 while (cnt >= 0) { 3286 if (m->m_len > cnt) { 3287 char *cp = mtod(m, caddr_t) + cnt; 3288 struct tcpcb *tp = sototcpcb(so); 3289 3290 INP_WLOCK_ASSERT(tp->t_inpcb); 3291 3292 tp->t_iobc = *cp; 3293 tp->t_oobflags |= TCPOOB_HAVEDATA; 3294 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3295 m->m_len--; 3296 if (m->m_flags & M_PKTHDR) 3297 m->m_pkthdr.len--; 3298 return; 3299 } 3300 cnt -= m->m_len; 3301 m = m->m_next; 3302 if (m == NULL) 3303 break; 3304 } 3305 panic("tcp_pulloutofband"); 3306 } 3307 3308 /* 3309 * Collect new round-trip time estimate 3310 * and update averages and current timeout. 3311 */ 3312 static void 3313 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3314 { 3315 int delta; 3316 3317 INP_WLOCK_ASSERT(tp->t_inpcb); 3318 3319 TCPSTAT_INC(tcps_rttupdated); 3320 tp->t_rttupdated++; 3321 if (tp->t_srtt != 0) { 3322 /* 3323 * srtt is stored as fixed point with 5 bits after the 3324 * binary point (i.e., scaled by 8). The following magic 3325 * is equivalent to the smoothing algorithm in rfc793 with 3326 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3327 * point). Adjust rtt to origin 0. 3328 */ 3329 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3330 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3331 3332 if ((tp->t_srtt += delta) <= 0) 3333 tp->t_srtt = 1; 3334 3335 /* 3336 * We accumulate a smoothed rtt variance (actually, a 3337 * smoothed mean difference), then set the retransmit 3338 * timer to smoothed rtt + 4 times the smoothed variance. 3339 * rttvar is stored as fixed point with 4 bits after the 3340 * binary point (scaled by 16). The following is 3341 * equivalent to rfc793 smoothing with an alpha of .75 3342 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3343 * rfc793's wired-in beta. 3344 */ 3345 if (delta < 0) 3346 delta = -delta; 3347 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3348 if ((tp->t_rttvar += delta) <= 0) 3349 tp->t_rttvar = 1; 3350 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3351 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3352 } else { 3353 /* 3354 * No rtt measurement yet - use the unsmoothed rtt. 3355 * Set the variance to half the rtt (so our first 3356 * retransmit happens at 3*rtt). 3357 */ 3358 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3359 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3360 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3361 } 3362 tp->t_rtttime = 0; 3363 tp->t_rxtshift = 0; 3364 3365 /* 3366 * the retransmit should happen at rtt + 4 * rttvar. 3367 * Because of the way we do the smoothing, srtt and rttvar 3368 * will each average +1/2 tick of bias. When we compute 3369 * the retransmit timer, we want 1/2 tick of rounding and 3370 * 1 extra tick because of +-1/2 tick uncertainty in the 3371 * firing of the timer. The bias will give us exactly the 3372 * 1.5 tick we need. But, because the bias is 3373 * statistical, we have to test that we don't drop below 3374 * the minimum feasible timer (which is 2 ticks). 3375 */ 3376 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3377 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3378 3379 /* 3380 * We received an ack for a packet that wasn't retransmitted; 3381 * it is probably safe to discard any error indications we've 3382 * received recently. This isn't quite right, but close enough 3383 * for now (a route might have failed after we sent a segment, 3384 * and the return path might not be symmetrical). 3385 */ 3386 tp->t_softerror = 0; 3387 } 3388 3389 /* 3390 * Determine a reasonable value for maxseg size. 3391 * If the route is known, check route for mtu. 3392 * If none, use an mss that can be handled on the outgoing interface 3393 * without forcing IP to fragment. If no route is found, route has no mtu, 3394 * or the destination isn't local, use a default, hopefully conservative 3395 * size (usually 512 or the default IP max size, but no more than the mtu 3396 * of the interface), as we can't discover anything about intervening 3397 * gateways or networks. We also initialize the congestion/slow start 3398 * window to be a single segment if the destination isn't local. 3399 * While looking at the routing entry, we also initialize other path-dependent 3400 * parameters from pre-set or cached values in the routing entry. 3401 * 3402 * Also take into account the space needed for options that we 3403 * send regularly. Make maxseg shorter by that amount to assure 3404 * that we can send maxseg amount of data even when the options 3405 * are present. Store the upper limit of the length of options plus 3406 * data in maxopd. 3407 * 3408 * NOTE that this routine is only called when we process an incoming 3409 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3410 * settings are handled in tcp_mssopt(). 3411 */ 3412 void 3413 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3414 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3415 { 3416 int mss = 0; 3417 u_long maxmtu = 0; 3418 struct inpcb *inp = tp->t_inpcb; 3419 struct hc_metrics_lite metrics; 3420 int origoffer; 3421 #ifdef INET6 3422 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3423 size_t min_protoh = isipv6 ? 3424 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3425 sizeof (struct tcpiphdr); 3426 #else 3427 const size_t min_protoh = sizeof(struct tcpiphdr); 3428 #endif 3429 3430 INP_WLOCK_ASSERT(tp->t_inpcb); 3431 3432 if (mtuoffer != -1) { 3433 KASSERT(offer == -1, ("%s: conflict", __func__)); 3434 offer = mtuoffer - min_protoh; 3435 } 3436 origoffer = offer; 3437 3438 /* Initialize. */ 3439 #ifdef INET6 3440 if (isipv6) { 3441 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3442 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3443 } 3444 #endif 3445 #if defined(INET) && defined(INET6) 3446 else 3447 #endif 3448 #ifdef INET 3449 { 3450 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3451 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3452 } 3453 #endif 3454 3455 /* 3456 * No route to sender, stay with default mss and return. 3457 */ 3458 if (maxmtu == 0) { 3459 /* 3460 * In case we return early we need to initialize metrics 3461 * to a defined state as tcp_hc_get() would do for us 3462 * if there was no cache hit. 3463 */ 3464 if (metricptr != NULL) 3465 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3466 return; 3467 } 3468 3469 /* What have we got? */ 3470 switch (offer) { 3471 case 0: 3472 /* 3473 * Offer == 0 means that there was no MSS on the SYN 3474 * segment, in this case we use tcp_mssdflt as 3475 * already assigned to t_maxopd above. 3476 */ 3477 offer = tp->t_maxopd; 3478 break; 3479 3480 case -1: 3481 /* 3482 * Offer == -1 means that we didn't receive SYN yet. 3483 */ 3484 /* FALLTHROUGH */ 3485 3486 default: 3487 /* 3488 * Prevent DoS attack with too small MSS. Round up 3489 * to at least minmss. 3490 */ 3491 offer = max(offer, V_tcp_minmss); 3492 } 3493 3494 /* 3495 * rmx information is now retrieved from tcp_hostcache. 3496 */ 3497 tcp_hc_get(&inp->inp_inc, &metrics); 3498 if (metricptr != NULL) 3499 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3500 3501 /* 3502 * If there's a discovered mtu in tcp hostcache, use it. 3503 * Else, use the link mtu. 3504 */ 3505 if (metrics.rmx_mtu) 3506 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3507 else { 3508 #ifdef INET6 3509 if (isipv6) { 3510 mss = maxmtu - min_protoh; 3511 if (!V_path_mtu_discovery && 3512 !in6_localaddr(&inp->in6p_faddr)) 3513 mss = min(mss, V_tcp_v6mssdflt); 3514 } 3515 #endif 3516 #if defined(INET) && defined(INET6) 3517 else 3518 #endif 3519 #ifdef INET 3520 { 3521 mss = maxmtu - min_protoh; 3522 if (!V_path_mtu_discovery && 3523 !in_localaddr(inp->inp_faddr)) 3524 mss = min(mss, V_tcp_mssdflt); 3525 } 3526 #endif 3527 /* 3528 * XXX - The above conditional (mss = maxmtu - min_protoh) 3529 * probably violates the TCP spec. 3530 * The problem is that, since we don't know the 3531 * other end's MSS, we are supposed to use a conservative 3532 * default. But, if we do that, then MTU discovery will 3533 * never actually take place, because the conservative 3534 * default is much less than the MTUs typically seen 3535 * on the Internet today. For the moment, we'll sweep 3536 * this under the carpet. 3537 * 3538 * The conservative default might not actually be a problem 3539 * if the only case this occurs is when sending an initial 3540 * SYN with options and data to a host we've never talked 3541 * to before. Then, they will reply with an MSS value which 3542 * will get recorded and the new parameters should get 3543 * recomputed. For Further Study. 3544 */ 3545 } 3546 mss = min(mss, offer); 3547 3548 /* 3549 * Sanity check: make sure that maxopd will be large 3550 * enough to allow some data on segments even if the 3551 * all the option space is used (40bytes). Otherwise 3552 * funny things may happen in tcp_output. 3553 */ 3554 mss = max(mss, 64); 3555 3556 /* 3557 * maxopd stores the maximum length of data AND options 3558 * in a segment; maxseg is the amount of data in a normal 3559 * segment. We need to store this value (maxopd) apart 3560 * from maxseg, because now every segment carries options 3561 * and thus we normally have somewhat less data in segments. 3562 */ 3563 tp->t_maxopd = mss; 3564 3565 /* 3566 * origoffer==-1 indicates that no segments were received yet. 3567 * In this case we just guess. 3568 */ 3569 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3570 (origoffer == -1 || 3571 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3572 mss -= TCPOLEN_TSTAMP_APPA; 3573 3574 tp->t_maxseg = mss; 3575 } 3576 3577 void 3578 tcp_mss(struct tcpcb *tp, int offer) 3579 { 3580 int mss; 3581 u_long bufsize; 3582 struct inpcb *inp; 3583 struct socket *so; 3584 struct hc_metrics_lite metrics; 3585 struct tcp_ifcap cap; 3586 3587 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3588 3589 bzero(&cap, sizeof(cap)); 3590 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3591 3592 mss = tp->t_maxseg; 3593 inp = tp->t_inpcb; 3594 3595 /* 3596 * If there's a pipesize, change the socket buffer to that size, 3597 * don't change if sb_hiwat is different than default (then it 3598 * has been changed on purpose with setsockopt). 3599 * Make the socket buffers an integral number of mss units; 3600 * if the mss is larger than the socket buffer, decrease the mss. 3601 */ 3602 so = inp->inp_socket; 3603 SOCKBUF_LOCK(&so->so_snd); 3604 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3605 bufsize = metrics.rmx_sendpipe; 3606 else 3607 bufsize = so->so_snd.sb_hiwat; 3608 if (bufsize < mss) 3609 mss = bufsize; 3610 else { 3611 bufsize = roundup(bufsize, mss); 3612 if (bufsize > sb_max) 3613 bufsize = sb_max; 3614 if (bufsize > so->so_snd.sb_hiwat) 3615 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3616 } 3617 SOCKBUF_UNLOCK(&so->so_snd); 3618 tp->t_maxseg = mss; 3619 3620 SOCKBUF_LOCK(&so->so_rcv); 3621 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3622 bufsize = metrics.rmx_recvpipe; 3623 else 3624 bufsize = so->so_rcv.sb_hiwat; 3625 if (bufsize > mss) { 3626 bufsize = roundup(bufsize, mss); 3627 if (bufsize > sb_max) 3628 bufsize = sb_max; 3629 if (bufsize > so->so_rcv.sb_hiwat) 3630 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3631 } 3632 SOCKBUF_UNLOCK(&so->so_rcv); 3633 3634 /* Check the interface for TSO capabilities. */ 3635 if (cap.ifcap & CSUM_TSO) { 3636 tp->t_flags |= TF_TSO; 3637 tp->t_tsomax = cap.tsomax; 3638 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3639 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3640 } 3641 } 3642 3643 /* 3644 * Determine the MSS option to send on an outgoing SYN. 3645 */ 3646 int 3647 tcp_mssopt(struct in_conninfo *inc) 3648 { 3649 int mss = 0; 3650 u_long maxmtu = 0; 3651 u_long thcmtu = 0; 3652 size_t min_protoh; 3653 3654 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3655 3656 #ifdef INET6 3657 if (inc->inc_flags & INC_ISIPV6) { 3658 mss = V_tcp_v6mssdflt; 3659 maxmtu = tcp_maxmtu6(inc, NULL); 3660 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3661 } 3662 #endif 3663 #if defined(INET) && defined(INET6) 3664 else 3665 #endif 3666 #ifdef INET 3667 { 3668 mss = V_tcp_mssdflt; 3669 maxmtu = tcp_maxmtu(inc, NULL); 3670 min_protoh = sizeof(struct tcpiphdr); 3671 } 3672 #endif 3673 #if defined(INET6) || defined(INET) 3674 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3675 #endif 3676 3677 if (maxmtu && thcmtu) 3678 mss = min(maxmtu, thcmtu) - min_protoh; 3679 else if (maxmtu || thcmtu) 3680 mss = max(maxmtu, thcmtu) - min_protoh; 3681 3682 return (mss); 3683 } 3684 3685 3686 /* 3687 * On a partial ack arrives, force the retransmission of the 3688 * next unacknowledged segment. Do not clear tp->t_dupacks. 3689 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3690 * be started again. 3691 */ 3692 static void 3693 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3694 { 3695 tcp_seq onxt = tp->snd_nxt; 3696 u_long ocwnd = tp->snd_cwnd; 3697 3698 INP_WLOCK_ASSERT(tp->t_inpcb); 3699 3700 tcp_timer_activate(tp, TT_REXMT, 0); 3701 tp->t_rtttime = 0; 3702 tp->snd_nxt = th->th_ack; 3703 /* 3704 * Set snd_cwnd to one segment beyond acknowledged offset. 3705 * (tp->snd_una has not yet been updated when this function is called.) 3706 */ 3707 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3708 tp->t_flags |= TF_ACKNOW; 3709 (void) tcp_output(tp); 3710 tp->snd_cwnd = ocwnd; 3711 if (SEQ_GT(onxt, tp->snd_nxt)) 3712 tp->snd_nxt = onxt; 3713 /* 3714 * Partial window deflation. Relies on fact that tp->snd_una 3715 * not updated yet. 3716 */ 3717 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3718 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3719 else 3720 tp->snd_cwnd = 0; 3721 tp->snd_cwnd += tp->t_maxseg; 3722 } 3723