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_WLOCKED 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_WLOCK(&V_tcbinfo); 801 ti_locked = TI_WLOCKED; 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_WLOCKED) { 824 INP_INFO_WLOCK_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_WLOCK(&V_tcbinfo) == 0) { 973 in_pcbref(inp); 974 INP_WUNLOCK(inp); 975 INP_INFO_WLOCK(&V_tcbinfo); 976 ti_locked = TI_WLOCKED; 977 INP_WLOCK(inp); 978 if (in_pcbrele_wlocked(inp)) { 979 inp = NULL; 980 goto findpcb; 981 } 982 } else 983 ti_locked = TI_WLOCKED; 984 } 985 INP_INFO_WLOCK_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_WUNLOCK(&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_WLOCK_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_WLOCK(&V_tcbinfo) == 0) { 1031 in_pcbref(inp); 1032 INP_WUNLOCK(inp); 1033 INP_INFO_WLOCK(&V_tcbinfo); 1034 ti_locked = TI_WLOCKED; 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_WLOCKED; 1043 } 1044 INP_INFO_WLOCK_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_WLOCK_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 INP_WLOCK(inp); /* new connection */ 1152 tp = intotcpcb(inp); 1153 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1154 ("%s: ", __func__)); 1155 #ifdef TCP_SIGNATURE 1156 if (sig_checked == 0) { 1157 tcp_dooptions(&to, optp, optlen, 1158 (thflags & TH_SYN) ? TO_SYN : 0); 1159 if (!tcp_signature_verify_input(m, off0, tlen, 1160 optlen, &to, th, tp->t_flags)) { 1161 1162 /* 1163 * In SYN_SENT state if it receives an 1164 * RST, it is allowed for further 1165 * processing. 1166 */ 1167 if ((thflags & TH_RST) == 0 || 1168 (tp->t_state == TCPS_SYN_SENT) == 0) 1169 goto dropunlock; 1170 } 1171 sig_checked = 1; 1172 } 1173 #endif 1174 1175 /* 1176 * Process the segment and the data it 1177 * contains. tcp_do_segment() consumes 1178 * the mbuf chain and unlocks the inpcb. 1179 */ 1180 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1181 iptos, ti_locked); 1182 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1183 return (IPPROTO_DONE); 1184 } 1185 /* 1186 * Segment flag validation for new connection attempts: 1187 * 1188 * Our (SYN|ACK) response was rejected. 1189 * Check with syncache and remove entry to prevent 1190 * retransmits. 1191 * 1192 * NB: syncache_chkrst does its own logging of failure 1193 * causes. 1194 */ 1195 if (thflags & TH_RST) { 1196 syncache_chkrst(&inc, th); 1197 goto dropunlock; 1198 } 1199 /* 1200 * We can't do anything without SYN. 1201 */ 1202 if ((thflags & TH_SYN) == 0) { 1203 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1204 log(LOG_DEBUG, "%s; %s: Listen socket: " 1205 "SYN is missing, segment ignored\n", 1206 s, __func__); 1207 TCPSTAT_INC(tcps_badsyn); 1208 goto dropunlock; 1209 } 1210 /* 1211 * (SYN|ACK) is bogus on a listen socket. 1212 */ 1213 if (thflags & TH_ACK) { 1214 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1215 log(LOG_DEBUG, "%s; %s: Listen socket: " 1216 "SYN|ACK invalid, segment rejected\n", 1217 s, __func__); 1218 syncache_badack(&inc); /* XXX: Not needed! */ 1219 TCPSTAT_INC(tcps_badsyn); 1220 rstreason = BANDLIM_RST_OPENPORT; 1221 goto dropwithreset; 1222 } 1223 /* 1224 * If the drop_synfin option is enabled, drop all 1225 * segments with both the SYN and FIN bits set. 1226 * This prevents e.g. nmap from identifying the 1227 * TCP/IP stack. 1228 * XXX: Poor reasoning. nmap has other methods 1229 * and is constantly refining its stack detection 1230 * strategies. 1231 * XXX: This is a violation of the TCP specification 1232 * and was used by RFC1644. 1233 */ 1234 if ((thflags & TH_FIN) && V_drop_synfin) { 1235 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1236 log(LOG_DEBUG, "%s; %s: Listen socket: " 1237 "SYN|FIN segment ignored (based on " 1238 "sysctl setting)\n", s, __func__); 1239 TCPSTAT_INC(tcps_badsyn); 1240 goto dropunlock; 1241 } 1242 /* 1243 * Segment's flags are (SYN) or (SYN|FIN). 1244 * 1245 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1246 * as they do not affect the state of the TCP FSM. 1247 * The data pointed to by TH_URG and th_urp is ignored. 1248 */ 1249 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1250 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1251 KASSERT(thflags & (TH_SYN), 1252 ("%s: Listen socket: TH_SYN not set", __func__)); 1253 #ifdef INET6 1254 /* 1255 * If deprecated address is forbidden, 1256 * we do not accept SYN to deprecated interface 1257 * address to prevent any new inbound connection from 1258 * getting established. 1259 * When we do not accept SYN, we send a TCP RST, 1260 * with deprecated source address (instead of dropping 1261 * it). We compromise it as it is much better for peer 1262 * to send a RST, and RST will be the final packet 1263 * for the exchange. 1264 * 1265 * If we do not forbid deprecated addresses, we accept 1266 * the SYN packet. RFC2462 does not suggest dropping 1267 * SYN in this case. 1268 * If we decipher RFC2462 5.5.4, it says like this: 1269 * 1. use of deprecated addr with existing 1270 * communication is okay - "SHOULD continue to be 1271 * used" 1272 * 2. use of it with new communication: 1273 * (2a) "SHOULD NOT be used if alternate address 1274 * with sufficient scope is available" 1275 * (2b) nothing mentioned otherwise. 1276 * Here we fall into (2b) case as we have no choice in 1277 * our source address selection - we must obey the peer. 1278 * 1279 * The wording in RFC2462 is confusing, and there are 1280 * multiple description text for deprecated address 1281 * handling - worse, they are not exactly the same. 1282 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1283 */ 1284 if (isipv6 && !V_ip6_use_deprecated) { 1285 struct in6_ifaddr *ia6; 1286 1287 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */); 1288 if (ia6 != NULL && 1289 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1290 ifa_free(&ia6->ia_ifa); 1291 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1292 log(LOG_DEBUG, "%s; %s: Listen socket: " 1293 "Connection attempt to deprecated " 1294 "IPv6 address rejected\n", 1295 s, __func__); 1296 rstreason = BANDLIM_RST_OPENPORT; 1297 goto dropwithreset; 1298 } 1299 if (ia6) 1300 ifa_free(&ia6->ia_ifa); 1301 } 1302 #endif /* INET6 */ 1303 /* 1304 * Basic sanity checks on incoming SYN requests: 1305 * Don't respond if the destination is a link layer 1306 * broadcast according to RFC1122 4.2.3.10, p. 104. 1307 * If it is from this socket it must be forged. 1308 * Don't respond if the source or destination is a 1309 * global or subnet broad- or multicast address. 1310 * Note that it is quite possible to receive unicast 1311 * link-layer packets with a broadcast IP address. Use 1312 * in_broadcast() to find them. 1313 */ 1314 if (m->m_flags & (M_BCAST|M_MCAST)) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from broad- or multicast " 1318 "link layer address ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 #ifdef INET6 1322 if (isipv6) { 1323 if (th->th_dport == th->th_sport && 1324 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1325 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1326 log(LOG_DEBUG, "%s; %s: Listen socket: " 1327 "Connection attempt to/from self " 1328 "ignored\n", s, __func__); 1329 goto dropunlock; 1330 } 1331 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1332 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1333 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1334 log(LOG_DEBUG, "%s; %s: Listen socket: " 1335 "Connection attempt from/to multicast " 1336 "address ignored\n", s, __func__); 1337 goto dropunlock; 1338 } 1339 } 1340 #endif 1341 #if defined(INET) && defined(INET6) 1342 else 1343 #endif 1344 #ifdef INET 1345 { 1346 if (th->th_dport == th->th_sport && 1347 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1348 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1349 log(LOG_DEBUG, "%s; %s: Listen socket: " 1350 "Connection attempt from/to self " 1351 "ignored\n", s, __func__); 1352 goto dropunlock; 1353 } 1354 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1355 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1356 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1357 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1358 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1359 log(LOG_DEBUG, "%s; %s: Listen socket: " 1360 "Connection attempt from/to broad- " 1361 "or multicast address ignored\n", 1362 s, __func__); 1363 goto dropunlock; 1364 } 1365 } 1366 #endif 1367 /* 1368 * SYN appears to be valid. Create compressed TCP state 1369 * for syncache. 1370 */ 1371 #ifdef TCPDEBUG 1372 if (so->so_options & SO_DEBUG) 1373 tcp_trace(TA_INPUT, ostate, tp, 1374 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1375 #endif 1376 tcp_dooptions(&to, optp, optlen, TO_SYN); 1377 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1378 /* 1379 * Entry added to syncache and mbuf consumed. 1380 * Only the listen socket is unlocked by syncache_add(). 1381 */ 1382 if (ti_locked == TI_WLOCKED) { 1383 INP_INFO_WUNLOCK(&V_tcbinfo); 1384 ti_locked = TI_UNLOCKED; 1385 } 1386 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1387 return (IPPROTO_DONE); 1388 } else if (tp->t_state == TCPS_LISTEN) { 1389 /* 1390 * When a listen socket is torn down the SO_ACCEPTCONN 1391 * flag is removed first while connections are drained 1392 * from the accept queue in a unlock/lock cycle of the 1393 * ACCEPT_LOCK, opening a race condition allowing a SYN 1394 * attempt go through unhandled. 1395 */ 1396 goto dropunlock; 1397 } 1398 1399 #ifdef TCP_SIGNATURE 1400 if (sig_checked == 0) { 1401 tcp_dooptions(&to, optp, optlen, 1402 (thflags & TH_SYN) ? TO_SYN : 0); 1403 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1404 th, tp->t_flags)) { 1405 1406 /* 1407 * In SYN_SENT state if it receives an RST, it is 1408 * allowed for further processing. 1409 */ 1410 if ((thflags & TH_RST) == 0 || 1411 (tp->t_state == TCPS_SYN_SENT) == 0) 1412 goto dropunlock; 1413 } 1414 sig_checked = 1; 1415 } 1416 #endif 1417 1418 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1419 1420 /* 1421 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1422 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1423 * the inpcb, and unlocks pcbinfo. 1424 */ 1425 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1426 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1427 return (IPPROTO_DONE); 1428 1429 dropwithreset: 1430 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1431 1432 if (ti_locked == TI_WLOCKED) { 1433 INP_INFO_WUNLOCK(&V_tcbinfo); 1434 ti_locked = TI_UNLOCKED; 1435 } 1436 #ifdef INVARIANTS 1437 else { 1438 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1439 "ti_locked: %d", __func__, ti_locked)); 1440 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1441 } 1442 #endif 1443 1444 if (inp != NULL) { 1445 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1446 INP_WUNLOCK(inp); 1447 } else 1448 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1449 m = NULL; /* mbuf chain got consumed. */ 1450 goto drop; 1451 1452 dropunlock: 1453 if (m != NULL) 1454 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1455 1456 if (ti_locked == TI_WLOCKED) { 1457 INP_INFO_WUNLOCK(&V_tcbinfo); 1458 ti_locked = TI_UNLOCKED; 1459 } 1460 #ifdef INVARIANTS 1461 else { 1462 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1463 "ti_locked: %d", __func__, ti_locked)); 1464 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1465 } 1466 #endif 1467 1468 if (inp != NULL) 1469 INP_WUNLOCK(inp); 1470 1471 drop: 1472 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1473 if (s != NULL) 1474 free(s, M_TCPLOG); 1475 if (m != NULL) 1476 m_freem(m); 1477 return (IPPROTO_DONE); 1478 } 1479 1480 static void 1481 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1482 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1483 int ti_locked) 1484 { 1485 int thflags, acked, ourfinisacked, needoutput = 0; 1486 int rstreason, todrop, win; 1487 u_long tiwin; 1488 char *s; 1489 struct in_conninfo *inc; 1490 struct mbuf *mfree; 1491 struct tcpopt to; 1492 1493 #ifdef TCPDEBUG 1494 /* 1495 * The size of tcp_saveipgen must be the size of the max ip header, 1496 * now IPv6. 1497 */ 1498 u_char tcp_saveipgen[IP6_HDR_LEN]; 1499 struct tcphdr tcp_savetcp; 1500 short ostate = 0; 1501 #endif 1502 thflags = th->th_flags; 1503 inc = &tp->t_inpcb->inp_inc; 1504 tp->sackhint.last_sack_ack = 0; 1505 1506 /* 1507 * If this is either a state-changing packet or current state isn't 1508 * established, we require a write lock on tcbinfo. Otherwise, we 1509 * allow the tcbinfo to be in either alocked or unlocked, as the 1510 * caller may have unnecessarily acquired a write lock due to a race. 1511 */ 1512 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1513 tp->t_state != TCPS_ESTABLISHED) { 1514 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1515 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1516 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1517 } else { 1518 #ifdef INVARIANTS 1519 if (ti_locked == TI_WLOCKED) 1520 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1521 else { 1522 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1523 "ti_locked: %d", __func__, ti_locked)); 1524 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1525 } 1526 #endif 1527 } 1528 INP_WLOCK_ASSERT(tp->t_inpcb); 1529 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1530 __func__)); 1531 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1532 __func__)); 1533 1534 /* 1535 * Segment received on connection. 1536 * Reset idle time and keep-alive timer. 1537 * XXX: This should be done after segment 1538 * validation to ignore broken/spoofed segs. 1539 */ 1540 tp->t_rcvtime = ticks; 1541 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1542 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1543 1544 /* 1545 * Scale up the window into a 32-bit value. 1546 * For the SYN_SENT state the scale is zero. 1547 */ 1548 tiwin = th->th_win << tp->snd_scale; 1549 1550 /* 1551 * TCP ECN processing. 1552 */ 1553 if (tp->t_flags & TF_ECN_PERMIT) { 1554 if (thflags & TH_CWR) 1555 tp->t_flags &= ~TF_ECN_SND_ECE; 1556 switch (iptos & IPTOS_ECN_MASK) { 1557 case IPTOS_ECN_CE: 1558 tp->t_flags |= TF_ECN_SND_ECE; 1559 TCPSTAT_INC(tcps_ecn_ce); 1560 break; 1561 case IPTOS_ECN_ECT0: 1562 TCPSTAT_INC(tcps_ecn_ect0); 1563 break; 1564 case IPTOS_ECN_ECT1: 1565 TCPSTAT_INC(tcps_ecn_ect1); 1566 break; 1567 } 1568 1569 /* Process a packet differently from RFC3168. */ 1570 cc_ecnpkt_handler(tp, th, iptos); 1571 1572 /* Congestion experienced. */ 1573 if (thflags & TH_ECE) { 1574 cc_cong_signal(tp, th, CC_ECN); 1575 } 1576 } 1577 1578 /* 1579 * Parse options on any incoming segment. 1580 */ 1581 tcp_dooptions(&to, (u_char *)(th + 1), 1582 (th->th_off << 2) - sizeof(struct tcphdr), 1583 (thflags & TH_SYN) ? TO_SYN : 0); 1584 1585 /* 1586 * If echoed timestamp is later than the current time, 1587 * fall back to non RFC1323 RTT calculation. Normalize 1588 * timestamp if syncookies were used when this connection 1589 * was established. 1590 */ 1591 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1592 to.to_tsecr -= tp->ts_offset; 1593 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1594 to.to_tsecr = 0; 1595 } 1596 /* 1597 * If timestamps were negotiated during SYN/ACK they should 1598 * appear on every segment during this session and vice versa. 1599 */ 1600 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1601 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1602 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1603 "no action\n", s, __func__); 1604 free(s, M_TCPLOG); 1605 } 1606 } 1607 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1608 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1609 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1610 "no action\n", s, __func__); 1611 free(s, M_TCPLOG); 1612 } 1613 } 1614 1615 /* 1616 * Process options only when we get SYN/ACK back. The SYN case 1617 * for incoming connections is handled in tcp_syncache. 1618 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1619 * or <SYN,ACK>) segment itself is never scaled. 1620 * XXX this is traditional behavior, may need to be cleaned up. 1621 */ 1622 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1623 if ((to.to_flags & TOF_SCALE) && 1624 (tp->t_flags & TF_REQ_SCALE)) { 1625 tp->t_flags |= TF_RCVD_SCALE; 1626 tp->snd_scale = to.to_wscale; 1627 } 1628 /* 1629 * Initial send window. It will be updated with 1630 * the next incoming segment to the scaled value. 1631 */ 1632 tp->snd_wnd = th->th_win; 1633 if (to.to_flags & TOF_TS) { 1634 tp->t_flags |= TF_RCVD_TSTMP; 1635 tp->ts_recent = to.to_tsval; 1636 tp->ts_recent_age = tcp_ts_getticks(); 1637 } 1638 if (to.to_flags & TOF_MSS) 1639 tcp_mss(tp, to.to_mss); 1640 if ((tp->t_flags & TF_SACK_PERMIT) && 1641 (to.to_flags & TOF_SACKPERM) == 0) 1642 tp->t_flags &= ~TF_SACK_PERMIT; 1643 } 1644 1645 /* 1646 * Header prediction: check for the two common cases 1647 * of a uni-directional data xfer. If the packet has 1648 * no control flags, is in-sequence, the window didn't 1649 * change and we're not retransmitting, it's a 1650 * candidate. If the length is zero and the ack moved 1651 * forward, we're the sender side of the xfer. Just 1652 * free the data acked & wake any higher level process 1653 * that was blocked waiting for space. If the length 1654 * is non-zero and the ack didn't move, we're the 1655 * receiver side. If we're getting packets in-order 1656 * (the reassembly queue is empty), add the data to 1657 * the socket buffer and note that we need a delayed ack. 1658 * Make sure that the hidden state-flags are also off. 1659 * Since we check for TCPS_ESTABLISHED first, it can only 1660 * be TH_NEEDSYN. 1661 */ 1662 if (tp->t_state == TCPS_ESTABLISHED && 1663 th->th_seq == tp->rcv_nxt && 1664 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1665 tp->snd_nxt == tp->snd_max && 1666 tiwin && tiwin == tp->snd_wnd && 1667 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1668 tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 || 1669 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1670 1671 /* 1672 * If last ACK falls within this segment's sequence numbers, 1673 * record the timestamp. 1674 * NOTE that the test is modified according to the latest 1675 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1676 */ 1677 if ((to.to_flags & TOF_TS) != 0 && 1678 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1679 tp->ts_recent_age = tcp_ts_getticks(); 1680 tp->ts_recent = to.to_tsval; 1681 } 1682 1683 if (tlen == 0) { 1684 if (SEQ_GT(th->th_ack, tp->snd_una) && 1685 SEQ_LEQ(th->th_ack, tp->snd_max) && 1686 !IN_RECOVERY(tp->t_flags) && 1687 (to.to_flags & TOF_SACK) == 0 && 1688 TAILQ_EMPTY(&tp->snd_holes)) { 1689 /* 1690 * This is a pure ack for outstanding data. 1691 */ 1692 if (ti_locked == TI_WLOCKED) 1693 INP_INFO_WUNLOCK(&V_tcbinfo); 1694 ti_locked = TI_UNLOCKED; 1695 1696 TCPSTAT_INC(tcps_predack); 1697 1698 /* 1699 * "bad retransmit" recovery. 1700 */ 1701 if (tp->t_rxtshift == 1 && 1702 tp->t_flags & TF_PREVVALID && 1703 (int)(ticks - tp->t_badrxtwin) < 0) { 1704 cc_cong_signal(tp, th, CC_RTO_ERR); 1705 } 1706 1707 /* 1708 * Recalculate the transmit timer / rtt. 1709 * 1710 * Some boxes send broken timestamp replies 1711 * during the SYN+ACK phase, ignore 1712 * timestamps of 0 or we could calculate a 1713 * huge RTT and blow up the retransmit timer. 1714 */ 1715 if ((to.to_flags & TOF_TS) != 0 && 1716 to.to_tsecr) { 1717 u_int t; 1718 1719 t = tcp_ts_getticks() - to.to_tsecr; 1720 if (!tp->t_rttlow || tp->t_rttlow > t) 1721 tp->t_rttlow = t; 1722 tcp_xmit_timer(tp, 1723 TCP_TS_TO_TICKS(t) + 1); 1724 } else if (tp->t_rtttime && 1725 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1726 if (!tp->t_rttlow || 1727 tp->t_rttlow > ticks - tp->t_rtttime) 1728 tp->t_rttlow = ticks - tp->t_rtttime; 1729 tcp_xmit_timer(tp, 1730 ticks - tp->t_rtttime); 1731 } 1732 acked = BYTES_THIS_ACK(tp, th); 1733 1734 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1735 hhook_run_tcp_est_in(tp, th, &to); 1736 1737 TCPSTAT_INC(tcps_rcvackpack); 1738 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1739 sbdrop(&so->so_snd, acked); 1740 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1741 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1742 tp->snd_recover = th->th_ack - 1; 1743 1744 /* 1745 * Let the congestion control algorithm update 1746 * congestion control related information. This 1747 * typically means increasing the congestion 1748 * window. 1749 */ 1750 cc_ack_received(tp, th, CC_ACK); 1751 1752 tp->snd_una = th->th_ack; 1753 /* 1754 * Pull snd_wl2 up to prevent seq wrap relative 1755 * to th_ack. 1756 */ 1757 tp->snd_wl2 = th->th_ack; 1758 tp->t_dupacks = 0; 1759 m_freem(m); 1760 ND6_HINT(tp); /* Some progress has been made. */ 1761 1762 /* 1763 * If all outstanding data are acked, stop 1764 * retransmit timer, otherwise restart timer 1765 * using current (possibly backed-off) value. 1766 * If process is waiting for space, 1767 * wakeup/selwakeup/signal. If data 1768 * are ready to send, let tcp_output 1769 * decide between more output or persist. 1770 */ 1771 #ifdef TCPDEBUG 1772 if (so->so_options & SO_DEBUG) 1773 tcp_trace(TA_INPUT, ostate, tp, 1774 (void *)tcp_saveipgen, 1775 &tcp_savetcp, 0); 1776 #endif 1777 if (tp->snd_una == tp->snd_max) 1778 tcp_timer_activate(tp, TT_REXMT, 0); 1779 else if (!tcp_timer_active(tp, TT_PERSIST)) 1780 tcp_timer_activate(tp, TT_REXMT, 1781 tp->t_rxtcur); 1782 sowwakeup(so); 1783 if (sbavail(&so->so_snd)) 1784 (void) tcp_output(tp); 1785 goto check_delack; 1786 } 1787 } else if (th->th_ack == tp->snd_una && 1788 tlen <= sbspace(&so->so_rcv)) { 1789 int newsize = 0; /* automatic sockbuf scaling */ 1790 1791 /* 1792 * This is a pure, in-sequence data packet with 1793 * nothing on the reassembly queue and we have enough 1794 * buffer space to take it. 1795 */ 1796 if (ti_locked == TI_WLOCKED) 1797 INP_INFO_WUNLOCK(&V_tcbinfo); 1798 ti_locked = TI_UNLOCKED; 1799 1800 /* Clean receiver SACK report if present */ 1801 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1802 tcp_clean_sackreport(tp); 1803 TCPSTAT_INC(tcps_preddat); 1804 tp->rcv_nxt += tlen; 1805 /* 1806 * Pull snd_wl1 up to prevent seq wrap relative to 1807 * th_seq. 1808 */ 1809 tp->snd_wl1 = th->th_seq; 1810 /* 1811 * Pull rcv_up up to prevent seq wrap relative to 1812 * rcv_nxt. 1813 */ 1814 tp->rcv_up = tp->rcv_nxt; 1815 TCPSTAT_INC(tcps_rcvpack); 1816 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1817 ND6_HINT(tp); /* Some progress has been made */ 1818 #ifdef TCPDEBUG 1819 if (so->so_options & SO_DEBUG) 1820 tcp_trace(TA_INPUT, ostate, tp, 1821 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1822 #endif 1823 /* 1824 * Automatic sizing of receive socket buffer. Often the send 1825 * buffer size is not optimally adjusted to the actual network 1826 * conditions at hand (delay bandwidth product). Setting the 1827 * buffer size too small limits throughput on links with high 1828 * bandwidth and high delay (eg. trans-continental/oceanic links). 1829 * 1830 * On the receive side the socket buffer memory is only rarely 1831 * used to any significant extent. This allows us to be much 1832 * more aggressive in scaling the receive socket buffer. For 1833 * the case that the buffer space is actually used to a large 1834 * extent and we run out of kernel memory we can simply drop 1835 * the new segments; TCP on the sender will just retransmit it 1836 * later. Setting the buffer size too big may only consume too 1837 * much kernel memory if the application doesn't read() from 1838 * the socket or packet loss or reordering makes use of the 1839 * reassembly queue. 1840 * 1841 * The criteria to step up the receive buffer one notch are: 1842 * 1. Application has not set receive buffer size with 1843 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1844 * 2. the number of bytes received during the time it takes 1845 * one timestamp to be reflected back to us (the RTT); 1846 * 3. received bytes per RTT is within seven eighth of the 1847 * current socket buffer size; 1848 * 4. receive buffer size has not hit maximal automatic size; 1849 * 1850 * This algorithm does one step per RTT at most and only if 1851 * we receive a bulk stream w/o packet losses or reorderings. 1852 * Shrinking the buffer during idle times is not necessary as 1853 * it doesn't consume any memory when idle. 1854 * 1855 * TODO: Only step up if the application is actually serving 1856 * the buffer to better manage the socket buffer resources. 1857 */ 1858 if (V_tcp_do_autorcvbuf && 1859 (to.to_flags & TOF_TS) && 1860 to.to_tsecr && 1861 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1862 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1863 to.to_tsecr - tp->rfbuf_ts < hz) { 1864 if (tp->rfbuf_cnt > 1865 (so->so_rcv.sb_hiwat / 8 * 7) && 1866 so->so_rcv.sb_hiwat < 1867 V_tcp_autorcvbuf_max) { 1868 newsize = 1869 min(so->so_rcv.sb_hiwat + 1870 V_tcp_autorcvbuf_inc, 1871 V_tcp_autorcvbuf_max); 1872 } 1873 /* Start over with next RTT. */ 1874 tp->rfbuf_ts = 0; 1875 tp->rfbuf_cnt = 0; 1876 } else 1877 tp->rfbuf_cnt += tlen; /* add up */ 1878 } 1879 1880 /* Add data to socket buffer. */ 1881 SOCKBUF_LOCK(&so->so_rcv); 1882 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1883 m_freem(m); 1884 } else { 1885 /* 1886 * Set new socket buffer size. 1887 * Give up when limit is reached. 1888 */ 1889 if (newsize) 1890 if (!sbreserve_locked(&so->so_rcv, 1891 newsize, so, NULL)) 1892 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1893 m_adj(m, drop_hdrlen); /* delayed header drop */ 1894 sbappendstream_locked(&so->so_rcv, m, 0); 1895 } 1896 /* NB: sorwakeup_locked() does an implicit unlock. */ 1897 sorwakeup_locked(so); 1898 if (DELAY_ACK(tp, tlen)) { 1899 tp->t_flags |= TF_DELACK; 1900 } else { 1901 tp->t_flags |= TF_ACKNOW; 1902 tcp_output(tp); 1903 } 1904 goto check_delack; 1905 } 1906 } 1907 1908 /* 1909 * Calculate amount of space in receive window, 1910 * and then do TCP input processing. 1911 * Receive window is amount of space in rcv queue, 1912 * but not less than advertised window. 1913 */ 1914 win = sbspace(&so->so_rcv); 1915 if (win < 0) 1916 win = 0; 1917 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1918 1919 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1920 tp->rfbuf_ts = 0; 1921 tp->rfbuf_cnt = 0; 1922 1923 switch (tp->t_state) { 1924 1925 /* 1926 * If the state is SYN_RECEIVED: 1927 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1928 */ 1929 case TCPS_SYN_RECEIVED: 1930 if ((thflags & TH_ACK) && 1931 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1932 SEQ_GT(th->th_ack, tp->snd_max))) { 1933 rstreason = BANDLIM_RST_OPENPORT; 1934 goto dropwithreset; 1935 } 1936 break; 1937 1938 /* 1939 * If the state is SYN_SENT: 1940 * if seg contains an ACK, but not for our SYN, drop the input. 1941 * if seg contains a RST, then drop the connection. 1942 * if seg does not contain SYN, then drop it. 1943 * Otherwise this is an acceptable SYN segment 1944 * initialize tp->rcv_nxt and tp->irs 1945 * if seg contains ack then advance tp->snd_una 1946 * if seg contains an ECE and ECN support is enabled, the stream 1947 * is ECN capable. 1948 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1949 * arrange for segment to be acked (eventually) 1950 * continue processing rest of data/controls, beginning with URG 1951 */ 1952 case TCPS_SYN_SENT: 1953 if ((thflags & TH_ACK) && 1954 (SEQ_LEQ(th->th_ack, tp->iss) || 1955 SEQ_GT(th->th_ack, tp->snd_max))) { 1956 rstreason = BANDLIM_UNLIMITED; 1957 goto dropwithreset; 1958 } 1959 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1960 TCP_PROBE5(connect__refused, NULL, tp, 1961 mtod(m, const char *), tp, th); 1962 tp = tcp_drop(tp, ECONNREFUSED); 1963 } 1964 if (thflags & TH_RST) 1965 goto drop; 1966 if (!(thflags & TH_SYN)) 1967 goto drop; 1968 1969 tp->irs = th->th_seq; 1970 tcp_rcvseqinit(tp); 1971 if (thflags & TH_ACK) { 1972 TCPSTAT_INC(tcps_connects); 1973 soisconnected(so); 1974 #ifdef MAC 1975 mac_socketpeer_set_from_mbuf(m, so); 1976 #endif 1977 /* Do window scaling on this connection? */ 1978 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1979 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1980 tp->rcv_scale = tp->request_r_scale; 1981 } 1982 tp->rcv_adv += imin(tp->rcv_wnd, 1983 TCP_MAXWIN << tp->rcv_scale); 1984 tp->snd_una++; /* SYN is acked */ 1985 /* 1986 * If there's data, delay ACK; if there's also a FIN 1987 * ACKNOW will be turned on later. 1988 */ 1989 if (DELAY_ACK(tp, tlen) && tlen != 0) 1990 tcp_timer_activate(tp, TT_DELACK, 1991 tcp_delacktime); 1992 else 1993 tp->t_flags |= TF_ACKNOW; 1994 1995 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1996 tp->t_flags |= TF_ECN_PERMIT; 1997 TCPSTAT_INC(tcps_ecn_shs); 1998 } 1999 2000 /* 2001 * Received <SYN,ACK> in SYN_SENT[*] state. 2002 * Transitions: 2003 * SYN_SENT --> ESTABLISHED 2004 * SYN_SENT* --> FIN_WAIT_1 2005 */ 2006 tp->t_starttime = ticks; 2007 if (tp->t_flags & TF_NEEDFIN) { 2008 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2009 tp->t_flags &= ~TF_NEEDFIN; 2010 thflags &= ~TH_SYN; 2011 } else { 2012 tcp_state_change(tp, TCPS_ESTABLISHED); 2013 TCP_PROBE5(connect__established, NULL, tp, 2014 mtod(m, const char *), tp, th); 2015 cc_conn_init(tp); 2016 tcp_timer_activate(tp, TT_KEEP, 2017 TP_KEEPIDLE(tp)); 2018 } 2019 } else { 2020 /* 2021 * Received initial SYN in SYN-SENT[*] state => 2022 * simultaneous open. 2023 * If it succeeds, connection is * half-synchronized. 2024 * Otherwise, do 3-way handshake: 2025 * SYN-SENT -> SYN-RECEIVED 2026 * SYN-SENT* -> SYN-RECEIVED* 2027 */ 2028 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2029 tcp_timer_activate(tp, TT_REXMT, 0); 2030 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2031 } 2032 2033 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 2034 "ti_locked %d", __func__, ti_locked)); 2035 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2036 INP_WLOCK_ASSERT(tp->t_inpcb); 2037 2038 /* 2039 * Advance th->th_seq to correspond to first data byte. 2040 * If data, trim to stay within window, 2041 * dropping FIN if necessary. 2042 */ 2043 th->th_seq++; 2044 if (tlen > tp->rcv_wnd) { 2045 todrop = tlen - tp->rcv_wnd; 2046 m_adj(m, -todrop); 2047 tlen = tp->rcv_wnd; 2048 thflags &= ~TH_FIN; 2049 TCPSTAT_INC(tcps_rcvpackafterwin); 2050 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2051 } 2052 tp->snd_wl1 = th->th_seq - 1; 2053 tp->rcv_up = th->th_seq; 2054 /* 2055 * Client side of transaction: already sent SYN and data. 2056 * If the remote host used T/TCP to validate the SYN, 2057 * our data will be ACK'd; if so, enter normal data segment 2058 * processing in the middle of step 5, ack processing. 2059 * Otherwise, goto step 6. 2060 */ 2061 if (thflags & TH_ACK) 2062 goto process_ACK; 2063 2064 goto step6; 2065 2066 /* 2067 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2068 * do normal processing. 2069 * 2070 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2071 */ 2072 case TCPS_LAST_ACK: 2073 case TCPS_CLOSING: 2074 break; /* continue normal processing */ 2075 } 2076 2077 /* 2078 * States other than LISTEN or SYN_SENT. 2079 * First check the RST flag and sequence number since reset segments 2080 * are exempt from the timestamp and connection count tests. This 2081 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2082 * below which allowed reset segments in half the sequence space 2083 * to fall though and be processed (which gives forged reset 2084 * segments with a random sequence number a 50 percent chance of 2085 * killing a connection). 2086 * Then check timestamp, if present. 2087 * Then check the connection count, if present. 2088 * Then check that at least some bytes of segment are within 2089 * receive window. If segment begins before rcv_nxt, 2090 * drop leading data (and SYN); if nothing left, just ack. 2091 */ 2092 if (thflags & TH_RST) { 2093 /* 2094 * RFC5961 Section 3.2 2095 * 2096 * - RST drops connection only if SEG.SEQ == RCV.NXT. 2097 * - If RST is in window, we send challenge ACK. 2098 * 2099 * Note: to take into account delayed ACKs, we should 2100 * test against last_ack_sent instead of rcv_nxt. 2101 * Note 2: we handle special case of closed window, not 2102 * covered by the RFC. 2103 */ 2104 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2105 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || 2106 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { 2107 2108 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2109 KASSERT(ti_locked == TI_WLOCKED, 2110 ("%s: TH_RST ti_locked %d, th %p tp %p", 2111 __func__, ti_locked, th, tp)); 2112 KASSERT(tp->t_state != TCPS_SYN_SENT, 2113 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", 2114 __func__, th, tp)); 2115 2116 if (V_tcp_insecure_rst || 2117 tp->last_ack_sent == th->th_seq) { 2118 TCPSTAT_INC(tcps_drops); 2119 /* Drop the connection. */ 2120 switch (tp->t_state) { 2121 case TCPS_SYN_RECEIVED: 2122 so->so_error = ECONNREFUSED; 2123 goto close; 2124 case TCPS_ESTABLISHED: 2125 case TCPS_FIN_WAIT_1: 2126 case TCPS_FIN_WAIT_2: 2127 case TCPS_CLOSE_WAIT: 2128 so->so_error = ECONNRESET; 2129 close: 2130 tcp_state_change(tp, TCPS_CLOSED); 2131 /* FALLTHROUGH */ 2132 default: 2133 tp = tcp_close(tp); 2134 } 2135 } else { 2136 TCPSTAT_INC(tcps_badrst); 2137 /* Send challenge ACK. */ 2138 tcp_respond(tp, mtod(m, void *), th, m, 2139 tp->rcv_nxt, tp->snd_nxt, TH_ACK); 2140 tp->last_ack_sent = tp->rcv_nxt; 2141 m = NULL; 2142 } 2143 } 2144 goto drop; 2145 } 2146 2147 /* 2148 * RFC5961 Section 4.2 2149 * Send challenge ACK for any SYN in synchronized state. 2150 */ 2151 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT) { 2152 KASSERT(ti_locked == TI_WLOCKED, 2153 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2154 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2155 2156 TCPSTAT_INC(tcps_badsyn); 2157 if (V_tcp_insecure_syn && 2158 SEQ_GEQ(th->th_seq, tp->last_ack_sent) && 2159 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2160 tp = tcp_drop(tp, ECONNRESET); 2161 rstreason = BANDLIM_UNLIMITED; 2162 } else { 2163 /* Send challenge ACK. */ 2164 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 2165 tp->snd_nxt, TH_ACK); 2166 tp->last_ack_sent = tp->rcv_nxt; 2167 m = NULL; 2168 } 2169 goto drop; 2170 } 2171 2172 /* 2173 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2174 * and it's less than ts_recent, drop it. 2175 */ 2176 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2177 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2178 2179 /* Check to see if ts_recent is over 24 days old. */ 2180 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2181 /* 2182 * Invalidate ts_recent. If this segment updates 2183 * ts_recent, the age will be reset later and ts_recent 2184 * will get a valid value. If it does not, setting 2185 * ts_recent to zero will at least satisfy the 2186 * requirement that zero be placed in the timestamp 2187 * echo reply when ts_recent isn't valid. The 2188 * age isn't reset until we get a valid ts_recent 2189 * because we don't want out-of-order segments to be 2190 * dropped when ts_recent is old. 2191 */ 2192 tp->ts_recent = 0; 2193 } else { 2194 TCPSTAT_INC(tcps_rcvduppack); 2195 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2196 TCPSTAT_INC(tcps_pawsdrop); 2197 if (tlen) 2198 goto dropafterack; 2199 goto drop; 2200 } 2201 } 2202 2203 /* 2204 * In the SYN-RECEIVED state, validate that the packet belongs to 2205 * this connection before trimming the data to fit the receive 2206 * window. Check the sequence number versus IRS since we know 2207 * the sequence numbers haven't wrapped. This is a partial fix 2208 * for the "LAND" DoS attack. 2209 */ 2210 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2211 rstreason = BANDLIM_RST_OPENPORT; 2212 goto dropwithreset; 2213 } 2214 2215 todrop = tp->rcv_nxt - th->th_seq; 2216 if (todrop > 0) { 2217 if (thflags & TH_SYN) { 2218 thflags &= ~TH_SYN; 2219 th->th_seq++; 2220 if (th->th_urp > 1) 2221 th->th_urp--; 2222 else 2223 thflags &= ~TH_URG; 2224 todrop--; 2225 } 2226 /* 2227 * Following if statement from Stevens, vol. 2, p. 960. 2228 */ 2229 if (todrop > tlen 2230 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2231 /* 2232 * Any valid FIN must be to the left of the window. 2233 * At this point the FIN must be a duplicate or out 2234 * of sequence; drop it. 2235 */ 2236 thflags &= ~TH_FIN; 2237 2238 /* 2239 * Send an ACK to resynchronize and drop any data. 2240 * But keep on processing for RST or ACK. 2241 */ 2242 tp->t_flags |= TF_ACKNOW; 2243 todrop = tlen; 2244 TCPSTAT_INC(tcps_rcvduppack); 2245 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2246 } else { 2247 TCPSTAT_INC(tcps_rcvpartduppack); 2248 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2249 } 2250 drop_hdrlen += todrop; /* drop from the top afterwards */ 2251 th->th_seq += todrop; 2252 tlen -= todrop; 2253 if (th->th_urp > todrop) 2254 th->th_urp -= todrop; 2255 else { 2256 thflags &= ~TH_URG; 2257 th->th_urp = 0; 2258 } 2259 } 2260 2261 /* 2262 * If new data are received on a connection after the 2263 * user processes are gone, then RST the other end. 2264 */ 2265 if ((so->so_state & SS_NOFDREF) && 2266 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2267 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2268 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2269 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2270 2271 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2272 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2273 "after socket was closed, " 2274 "sending RST and removing tcpcb\n", 2275 s, __func__, tcpstates[tp->t_state], tlen); 2276 free(s, M_TCPLOG); 2277 } 2278 tp = tcp_close(tp); 2279 TCPSTAT_INC(tcps_rcvafterclose); 2280 rstreason = BANDLIM_UNLIMITED; 2281 goto dropwithreset; 2282 } 2283 2284 /* 2285 * If segment ends after window, drop trailing data 2286 * (and PUSH and FIN); if nothing left, just ACK. 2287 */ 2288 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2289 if (todrop > 0) { 2290 TCPSTAT_INC(tcps_rcvpackafterwin); 2291 if (todrop >= tlen) { 2292 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2293 /* 2294 * If window is closed can only take segments at 2295 * window edge, and have to drop data and PUSH from 2296 * incoming segments. Continue processing, but 2297 * remember to ack. Otherwise, drop segment 2298 * and ack. 2299 */ 2300 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2301 tp->t_flags |= TF_ACKNOW; 2302 TCPSTAT_INC(tcps_rcvwinprobe); 2303 } else 2304 goto dropafterack; 2305 } else 2306 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2307 m_adj(m, -todrop); 2308 tlen -= todrop; 2309 thflags &= ~(TH_PUSH|TH_FIN); 2310 } 2311 2312 /* 2313 * If last ACK falls within this segment's sequence numbers, 2314 * record its timestamp. 2315 * NOTE: 2316 * 1) That the test incorporates suggestions from the latest 2317 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2318 * 2) That updating only on newer timestamps interferes with 2319 * our earlier PAWS tests, so this check should be solely 2320 * predicated on the sequence space of this segment. 2321 * 3) That we modify the segment boundary check to be 2322 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2323 * instead of RFC1323's 2324 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2325 * This modified check allows us to overcome RFC1323's 2326 * limitations as described in Stevens TCP/IP Illustrated 2327 * Vol. 2 p.869. In such cases, we can still calculate the 2328 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2329 */ 2330 if ((to.to_flags & TOF_TS) != 0 && 2331 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2332 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2333 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2334 tp->ts_recent_age = tcp_ts_getticks(); 2335 tp->ts_recent = to.to_tsval; 2336 } 2337 2338 /* 2339 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2340 * flag is on (half-synchronized state), then queue data for 2341 * later processing; else drop segment and return. 2342 */ 2343 if ((thflags & TH_ACK) == 0) { 2344 if (tp->t_state == TCPS_SYN_RECEIVED || 2345 (tp->t_flags & TF_NEEDSYN)) 2346 goto step6; 2347 else if (tp->t_flags & TF_ACKNOW) 2348 goto dropafterack; 2349 else 2350 goto drop; 2351 } 2352 2353 /* 2354 * Ack processing. 2355 */ 2356 switch (tp->t_state) { 2357 2358 /* 2359 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2360 * ESTABLISHED state and continue processing. 2361 * The ACK was checked above. 2362 */ 2363 case TCPS_SYN_RECEIVED: 2364 2365 TCPSTAT_INC(tcps_connects); 2366 soisconnected(so); 2367 /* Do window scaling? */ 2368 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2369 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2370 tp->rcv_scale = tp->request_r_scale; 2371 tp->snd_wnd = tiwin; 2372 } 2373 /* 2374 * Make transitions: 2375 * SYN-RECEIVED -> ESTABLISHED 2376 * SYN-RECEIVED* -> FIN-WAIT-1 2377 */ 2378 tp->t_starttime = ticks; 2379 if (tp->t_flags & TF_NEEDFIN) { 2380 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2381 tp->t_flags &= ~TF_NEEDFIN; 2382 } else { 2383 tcp_state_change(tp, TCPS_ESTABLISHED); 2384 TCP_PROBE5(accept__established, NULL, tp, 2385 mtod(m, const char *), tp, th); 2386 cc_conn_init(tp); 2387 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2388 } 2389 /* 2390 * If segment contains data or ACK, will call tcp_reass() 2391 * later; if not, do so now to pass queued data to user. 2392 */ 2393 if (tlen == 0 && (thflags & TH_FIN) == 0) 2394 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2395 (struct mbuf *)0); 2396 tp->snd_wl1 = th->th_seq - 1; 2397 /* FALLTHROUGH */ 2398 2399 /* 2400 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2401 * ACKs. If the ack is in the range 2402 * tp->snd_una < th->th_ack <= tp->snd_max 2403 * then advance tp->snd_una to th->th_ack and drop 2404 * data from the retransmission queue. If this ACK reflects 2405 * more up to date window information we update our window information. 2406 */ 2407 case TCPS_ESTABLISHED: 2408 case TCPS_FIN_WAIT_1: 2409 case TCPS_FIN_WAIT_2: 2410 case TCPS_CLOSE_WAIT: 2411 case TCPS_CLOSING: 2412 case TCPS_LAST_ACK: 2413 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2414 TCPSTAT_INC(tcps_rcvacktoomuch); 2415 goto dropafterack; 2416 } 2417 if ((tp->t_flags & TF_SACK_PERMIT) && 2418 ((to.to_flags & TOF_SACK) || 2419 !TAILQ_EMPTY(&tp->snd_holes))) 2420 tcp_sack_doack(tp, &to, th->th_ack); 2421 2422 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2423 hhook_run_tcp_est_in(tp, th, &to); 2424 2425 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2426 if (tlen == 0 && tiwin == tp->snd_wnd) { 2427 /* 2428 * If this is the first time we've seen a 2429 * FIN from the remote, this is not a 2430 * duplicate and it needs to be processed 2431 * normally. This happens during a 2432 * simultaneous close. 2433 */ 2434 if ((thflags & TH_FIN) && 2435 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2436 tp->t_dupacks = 0; 2437 break; 2438 } 2439 TCPSTAT_INC(tcps_rcvdupack); 2440 /* 2441 * If we have outstanding data (other than 2442 * a window probe), this is a completely 2443 * duplicate ack (ie, window info didn't 2444 * change and FIN isn't set), 2445 * the ack is the biggest we've 2446 * seen and we've seen exactly our rexmt 2447 * threshhold of them, assume a packet 2448 * has been dropped and retransmit it. 2449 * Kludge snd_nxt & the congestion 2450 * window so we send only this one 2451 * packet. 2452 * 2453 * We know we're losing at the current 2454 * window size so do congestion avoidance 2455 * (set ssthresh to half the current window 2456 * and pull our congestion window back to 2457 * the new ssthresh). 2458 * 2459 * Dup acks mean that packets have left the 2460 * network (they're now cached at the receiver) 2461 * so bump cwnd by the amount in the receiver 2462 * to keep a constant cwnd packets in the 2463 * network. 2464 * 2465 * When using TCP ECN, notify the peer that 2466 * we reduced the cwnd. 2467 */ 2468 if (!tcp_timer_active(tp, TT_REXMT) || 2469 th->th_ack != tp->snd_una) 2470 tp->t_dupacks = 0; 2471 else if (++tp->t_dupacks > tcprexmtthresh || 2472 IN_FASTRECOVERY(tp->t_flags)) { 2473 cc_ack_received(tp, th, CC_DUPACK); 2474 if ((tp->t_flags & TF_SACK_PERMIT) && 2475 IN_FASTRECOVERY(tp->t_flags)) { 2476 int awnd; 2477 2478 /* 2479 * Compute the amount of data in flight first. 2480 * We can inject new data into the pipe iff 2481 * we have less than 1/2 the original window's 2482 * worth of data in flight. 2483 */ 2484 awnd = (tp->snd_nxt - tp->snd_fack) + 2485 tp->sackhint.sack_bytes_rexmit; 2486 if (awnd < tp->snd_ssthresh) { 2487 tp->snd_cwnd += tp->t_maxseg; 2488 if (tp->snd_cwnd > tp->snd_ssthresh) 2489 tp->snd_cwnd = tp->snd_ssthresh; 2490 } 2491 } else 2492 tp->snd_cwnd += tp->t_maxseg; 2493 (void) tcp_output(tp); 2494 goto drop; 2495 } else if (tp->t_dupacks == tcprexmtthresh) { 2496 tcp_seq onxt = tp->snd_nxt; 2497 2498 /* 2499 * If we're doing sack, check to 2500 * see if we're already in sack 2501 * recovery. If we're not doing sack, 2502 * check to see if we're in newreno 2503 * recovery. 2504 */ 2505 if (tp->t_flags & TF_SACK_PERMIT) { 2506 if (IN_FASTRECOVERY(tp->t_flags)) { 2507 tp->t_dupacks = 0; 2508 break; 2509 } 2510 } else { 2511 if (SEQ_LEQ(th->th_ack, 2512 tp->snd_recover)) { 2513 tp->t_dupacks = 0; 2514 break; 2515 } 2516 } 2517 /* Congestion signal before ack. */ 2518 cc_cong_signal(tp, th, CC_NDUPACK); 2519 cc_ack_received(tp, th, CC_DUPACK); 2520 tcp_timer_activate(tp, TT_REXMT, 0); 2521 tp->t_rtttime = 0; 2522 if (tp->t_flags & TF_SACK_PERMIT) { 2523 TCPSTAT_INC( 2524 tcps_sack_recovery_episode); 2525 tp->sack_newdata = tp->snd_nxt; 2526 tp->snd_cwnd = tp->t_maxseg; 2527 (void) tcp_output(tp); 2528 goto drop; 2529 } 2530 tp->snd_nxt = th->th_ack; 2531 tp->snd_cwnd = tp->t_maxseg; 2532 (void) tcp_output(tp); 2533 KASSERT(tp->snd_limited <= 2, 2534 ("%s: tp->snd_limited too big", 2535 __func__)); 2536 tp->snd_cwnd = tp->snd_ssthresh + 2537 tp->t_maxseg * 2538 (tp->t_dupacks - tp->snd_limited); 2539 if (SEQ_GT(onxt, tp->snd_nxt)) 2540 tp->snd_nxt = onxt; 2541 goto drop; 2542 } else if (V_tcp_do_rfc3042) { 2543 cc_ack_received(tp, th, CC_DUPACK); 2544 u_long oldcwnd = tp->snd_cwnd; 2545 tcp_seq oldsndmax = tp->snd_max; 2546 u_int sent; 2547 int avail; 2548 2549 KASSERT(tp->t_dupacks == 1 || 2550 tp->t_dupacks == 2, 2551 ("%s: dupacks not 1 or 2", 2552 __func__)); 2553 if (tp->t_dupacks == 1) 2554 tp->snd_limited = 0; 2555 tp->snd_cwnd = 2556 (tp->snd_nxt - tp->snd_una) + 2557 (tp->t_dupacks - tp->snd_limited) * 2558 tp->t_maxseg; 2559 /* 2560 * Only call tcp_output when there 2561 * is new data available to be sent. 2562 * Otherwise we would send pure ACKs. 2563 */ 2564 SOCKBUF_LOCK(&so->so_snd); 2565 avail = sbavail(&so->so_snd) - 2566 (tp->snd_nxt - tp->snd_una); 2567 SOCKBUF_UNLOCK(&so->so_snd); 2568 if (avail > 0) 2569 (void) tcp_output(tp); 2570 sent = tp->snd_max - oldsndmax; 2571 if (sent > tp->t_maxseg) { 2572 KASSERT((tp->t_dupacks == 2 && 2573 tp->snd_limited == 0) || 2574 (sent == tp->t_maxseg + 1 && 2575 tp->t_flags & TF_SENTFIN), 2576 ("%s: sent too much", 2577 __func__)); 2578 tp->snd_limited = 2; 2579 } else if (sent > 0) 2580 ++tp->snd_limited; 2581 tp->snd_cwnd = oldcwnd; 2582 goto drop; 2583 } 2584 } else 2585 tp->t_dupacks = 0; 2586 break; 2587 } 2588 2589 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2590 ("%s: th_ack <= snd_una", __func__)); 2591 2592 /* 2593 * If the congestion window was inflated to account 2594 * for the other side's cached packets, retract it. 2595 */ 2596 if (IN_FASTRECOVERY(tp->t_flags)) { 2597 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2598 if (tp->t_flags & TF_SACK_PERMIT) 2599 tcp_sack_partialack(tp, th); 2600 else 2601 tcp_newreno_partial_ack(tp, th); 2602 } else 2603 cc_post_recovery(tp, th); 2604 } 2605 tp->t_dupacks = 0; 2606 /* 2607 * If we reach this point, ACK is not a duplicate, 2608 * i.e., it ACKs something we sent. 2609 */ 2610 if (tp->t_flags & TF_NEEDSYN) { 2611 /* 2612 * T/TCP: Connection was half-synchronized, and our 2613 * SYN has been ACK'd (so connection is now fully 2614 * synchronized). Go to non-starred state, 2615 * increment snd_una for ACK of SYN, and check if 2616 * we can do window scaling. 2617 */ 2618 tp->t_flags &= ~TF_NEEDSYN; 2619 tp->snd_una++; 2620 /* Do window scaling? */ 2621 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2622 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2623 tp->rcv_scale = tp->request_r_scale; 2624 /* Send window already scaled. */ 2625 } 2626 } 2627 2628 process_ACK: 2629 INP_WLOCK_ASSERT(tp->t_inpcb); 2630 2631 acked = BYTES_THIS_ACK(tp, th); 2632 TCPSTAT_INC(tcps_rcvackpack); 2633 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2634 2635 /* 2636 * If we just performed our first retransmit, and the ACK 2637 * arrives within our recovery window, then it was a mistake 2638 * to do the retransmit in the first place. Recover our 2639 * original cwnd and ssthresh, and proceed to transmit where 2640 * we left off. 2641 */ 2642 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2643 (int)(ticks - tp->t_badrxtwin) < 0) 2644 cc_cong_signal(tp, th, CC_RTO_ERR); 2645 2646 /* 2647 * If we have a timestamp reply, update smoothed 2648 * round trip time. If no timestamp is present but 2649 * transmit timer is running and timed sequence 2650 * number was acked, update smoothed round trip time. 2651 * Since we now have an rtt measurement, cancel the 2652 * timer backoff (cf., Phil Karn's retransmit alg.). 2653 * Recompute the initial retransmit timer. 2654 * 2655 * Some boxes send broken timestamp replies 2656 * during the SYN+ACK phase, ignore 2657 * timestamps of 0 or we could calculate a 2658 * huge RTT and blow up the retransmit timer. 2659 */ 2660 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2661 u_int t; 2662 2663 t = tcp_ts_getticks() - to.to_tsecr; 2664 if (!tp->t_rttlow || tp->t_rttlow > t) 2665 tp->t_rttlow = t; 2666 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2667 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2668 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2669 tp->t_rttlow = ticks - tp->t_rtttime; 2670 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2671 } 2672 2673 /* 2674 * If all outstanding data is acked, stop retransmit 2675 * timer and remember to restart (more output or persist). 2676 * If there is more data to be acked, restart retransmit 2677 * timer, using current (possibly backed-off) value. 2678 */ 2679 if (th->th_ack == tp->snd_max) { 2680 tcp_timer_activate(tp, TT_REXMT, 0); 2681 needoutput = 1; 2682 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2683 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2684 2685 /* 2686 * If no data (only SYN) was ACK'd, 2687 * skip rest of ACK processing. 2688 */ 2689 if (acked == 0) 2690 goto step6; 2691 2692 /* 2693 * Let the congestion control algorithm update congestion 2694 * control related information. This typically means increasing 2695 * the congestion window. 2696 */ 2697 cc_ack_received(tp, th, CC_ACK); 2698 2699 SOCKBUF_LOCK(&so->so_snd); 2700 if (acked > sbavail(&so->so_snd)) { 2701 tp->snd_wnd -= sbavail(&so->so_snd); 2702 mfree = sbcut_locked(&so->so_snd, 2703 (int)sbavail(&so->so_snd)); 2704 ourfinisacked = 1; 2705 } else { 2706 mfree = sbcut_locked(&so->so_snd, acked); 2707 tp->snd_wnd -= acked; 2708 ourfinisacked = 0; 2709 } 2710 /* NB: sowwakeup_locked() does an implicit unlock. */ 2711 sowwakeup_locked(so); 2712 m_freem(mfree); 2713 /* Detect una wraparound. */ 2714 if (!IN_RECOVERY(tp->t_flags) && 2715 SEQ_GT(tp->snd_una, tp->snd_recover) && 2716 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2717 tp->snd_recover = th->th_ack - 1; 2718 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2719 if (IN_RECOVERY(tp->t_flags) && 2720 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2721 EXIT_RECOVERY(tp->t_flags); 2722 } 2723 tp->snd_una = th->th_ack; 2724 if (tp->t_flags & TF_SACK_PERMIT) { 2725 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2726 tp->snd_recover = tp->snd_una; 2727 } 2728 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2729 tp->snd_nxt = tp->snd_una; 2730 2731 switch (tp->t_state) { 2732 2733 /* 2734 * In FIN_WAIT_1 STATE in addition to the processing 2735 * for the ESTABLISHED state if our FIN is now acknowledged 2736 * then enter FIN_WAIT_2. 2737 */ 2738 case TCPS_FIN_WAIT_1: 2739 if (ourfinisacked) { 2740 /* 2741 * If we can't receive any more 2742 * data, then closing user can proceed. 2743 * Starting the timer is contrary to the 2744 * specification, but if we don't get a FIN 2745 * we'll hang forever. 2746 * 2747 * XXXjl: 2748 * we should release the tp also, and use a 2749 * compressed state. 2750 */ 2751 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2752 soisdisconnected(so); 2753 tcp_timer_activate(tp, TT_2MSL, 2754 (tcp_fast_finwait2_recycle ? 2755 tcp_finwait2_timeout : 2756 TP_MAXIDLE(tp))); 2757 } 2758 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2759 } 2760 break; 2761 2762 /* 2763 * In CLOSING STATE in addition to the processing for 2764 * the ESTABLISHED state if the ACK acknowledges our FIN 2765 * then enter the TIME-WAIT state, otherwise ignore 2766 * the segment. 2767 */ 2768 case TCPS_CLOSING: 2769 if (ourfinisacked) { 2770 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2771 tcp_twstart(tp); 2772 INP_INFO_WUNLOCK(&V_tcbinfo); 2773 m_freem(m); 2774 return; 2775 } 2776 break; 2777 2778 /* 2779 * In LAST_ACK, we may still be waiting for data to drain 2780 * and/or to be acked, as well as for the ack of our FIN. 2781 * If our FIN is now acknowledged, delete the TCB, 2782 * enter the closed state and return. 2783 */ 2784 case TCPS_LAST_ACK: 2785 if (ourfinisacked) { 2786 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2787 tp = tcp_close(tp); 2788 goto drop; 2789 } 2790 break; 2791 } 2792 } 2793 2794 step6: 2795 INP_WLOCK_ASSERT(tp->t_inpcb); 2796 2797 /* 2798 * Update window information. 2799 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2800 */ 2801 if ((thflags & TH_ACK) && 2802 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2803 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2804 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2805 /* keep track of pure window updates */ 2806 if (tlen == 0 && 2807 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2808 TCPSTAT_INC(tcps_rcvwinupd); 2809 tp->snd_wnd = tiwin; 2810 tp->snd_wl1 = th->th_seq; 2811 tp->snd_wl2 = th->th_ack; 2812 if (tp->snd_wnd > tp->max_sndwnd) 2813 tp->max_sndwnd = tp->snd_wnd; 2814 needoutput = 1; 2815 } 2816 2817 /* 2818 * Process segments with URG. 2819 */ 2820 if ((thflags & TH_URG) && th->th_urp && 2821 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2822 /* 2823 * This is a kludge, but if we receive and accept 2824 * random urgent pointers, we'll crash in 2825 * soreceive. It's hard to imagine someone 2826 * actually wanting to send this much urgent data. 2827 */ 2828 SOCKBUF_LOCK(&so->so_rcv); 2829 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { 2830 th->th_urp = 0; /* XXX */ 2831 thflags &= ~TH_URG; /* XXX */ 2832 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2833 goto dodata; /* XXX */ 2834 } 2835 /* 2836 * If this segment advances the known urgent pointer, 2837 * then mark the data stream. This should not happen 2838 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2839 * a FIN has been received from the remote side. 2840 * In these states we ignore the URG. 2841 * 2842 * According to RFC961 (Assigned Protocols), 2843 * the urgent pointer points to the last octet 2844 * of urgent data. We continue, however, 2845 * to consider it to indicate the first octet 2846 * of data past the urgent section as the original 2847 * spec states (in one of two places). 2848 */ 2849 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2850 tp->rcv_up = th->th_seq + th->th_urp; 2851 so->so_oobmark = sbavail(&so->so_rcv) + 2852 (tp->rcv_up - tp->rcv_nxt) - 1; 2853 if (so->so_oobmark == 0) 2854 so->so_rcv.sb_state |= SBS_RCVATMARK; 2855 sohasoutofband(so); 2856 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2857 } 2858 SOCKBUF_UNLOCK(&so->so_rcv); 2859 /* 2860 * Remove out of band data so doesn't get presented to user. 2861 * This can happen independent of advancing the URG pointer, 2862 * but if two URG's are pending at once, some out-of-band 2863 * data may creep in... ick. 2864 */ 2865 if (th->th_urp <= (u_long)tlen && 2866 !(so->so_options & SO_OOBINLINE)) { 2867 /* hdr drop is delayed */ 2868 tcp_pulloutofband(so, th, m, drop_hdrlen); 2869 } 2870 } else { 2871 /* 2872 * If no out of band data is expected, 2873 * pull receive urgent pointer along 2874 * with the receive window. 2875 */ 2876 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2877 tp->rcv_up = tp->rcv_nxt; 2878 } 2879 dodata: /* XXX */ 2880 INP_WLOCK_ASSERT(tp->t_inpcb); 2881 2882 /* 2883 * Process the segment text, merging it into the TCP sequencing queue, 2884 * and arranging for acknowledgment of receipt if necessary. 2885 * This process logically involves adjusting tp->rcv_wnd as data 2886 * is presented to the user (this happens in tcp_usrreq.c, 2887 * case PRU_RCVD). If a FIN has already been received on this 2888 * connection then we just ignore the text. 2889 */ 2890 if ((tlen || (thflags & TH_FIN)) && 2891 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2892 tcp_seq save_start = th->th_seq; 2893 m_adj(m, drop_hdrlen); /* delayed header drop */ 2894 /* 2895 * Insert segment which includes th into TCP reassembly queue 2896 * with control block tp. Set thflags to whether reassembly now 2897 * includes a segment with FIN. This handles the common case 2898 * inline (segment is the next to be received on an established 2899 * connection, and the queue is empty), avoiding linkage into 2900 * and removal from the queue and repetition of various 2901 * conversions. 2902 * Set DELACK for segments received in order, but ack 2903 * immediately when segments are out of order (so 2904 * fast retransmit can work). 2905 */ 2906 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL && 2907 TCPS_HAVEESTABLISHED(tp->t_state)) { 2908 if (DELAY_ACK(tp, tlen)) 2909 tp->t_flags |= TF_DELACK; 2910 else 2911 tp->t_flags |= TF_ACKNOW; 2912 tp->rcv_nxt += tlen; 2913 thflags = th->th_flags & TH_FIN; 2914 TCPSTAT_INC(tcps_rcvpack); 2915 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2916 ND6_HINT(tp); 2917 SOCKBUF_LOCK(&so->so_rcv); 2918 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2919 m_freem(m); 2920 else 2921 sbappendstream_locked(&so->so_rcv, m, 0); 2922 /* NB: sorwakeup_locked() does an implicit unlock. */ 2923 sorwakeup_locked(so); 2924 } else { 2925 /* 2926 * XXX: Due to the header drop above "th" is 2927 * theoretically invalid by now. Fortunately 2928 * m_adj() doesn't actually frees any mbufs 2929 * when trimming from the head. 2930 */ 2931 thflags = tcp_reass(tp, th, &tlen, m); 2932 tp->t_flags |= TF_ACKNOW; 2933 } 2934 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2935 tcp_update_sack_list(tp, save_start, save_start + tlen); 2936 #if 0 2937 /* 2938 * Note the amount of data that peer has sent into 2939 * our window, in order to estimate the sender's 2940 * buffer size. 2941 * XXX: Unused. 2942 */ 2943 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2944 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2945 else 2946 len = so->so_rcv.sb_hiwat; 2947 #endif 2948 } else { 2949 m_freem(m); 2950 thflags &= ~TH_FIN; 2951 } 2952 2953 /* 2954 * If FIN is received ACK the FIN and let the user know 2955 * that the connection is closing. 2956 */ 2957 if (thflags & TH_FIN) { 2958 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2959 socantrcvmore(so); 2960 /* 2961 * If connection is half-synchronized 2962 * (ie NEEDSYN flag on) then delay ACK, 2963 * so it may be piggybacked when SYN is sent. 2964 * Otherwise, since we received a FIN then no 2965 * more input can be expected, send ACK now. 2966 */ 2967 if (tp->t_flags & TF_NEEDSYN) 2968 tp->t_flags |= TF_DELACK; 2969 else 2970 tp->t_flags |= TF_ACKNOW; 2971 tp->rcv_nxt++; 2972 } 2973 switch (tp->t_state) { 2974 2975 /* 2976 * In SYN_RECEIVED and ESTABLISHED STATES 2977 * enter the CLOSE_WAIT state. 2978 */ 2979 case TCPS_SYN_RECEIVED: 2980 tp->t_starttime = ticks; 2981 /* FALLTHROUGH */ 2982 case TCPS_ESTABLISHED: 2983 tcp_state_change(tp, TCPS_CLOSE_WAIT); 2984 break; 2985 2986 /* 2987 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2988 * enter the CLOSING state. 2989 */ 2990 case TCPS_FIN_WAIT_1: 2991 tcp_state_change(tp, TCPS_CLOSING); 2992 break; 2993 2994 /* 2995 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2996 * starting the time-wait timer, turning off the other 2997 * standard timers. 2998 */ 2999 case TCPS_FIN_WAIT_2: 3000 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 3001 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 3002 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3003 ti_locked)); 3004 3005 tcp_twstart(tp); 3006 INP_INFO_WUNLOCK(&V_tcbinfo); 3007 return; 3008 } 3009 } 3010 if (ti_locked == TI_WLOCKED) 3011 INP_INFO_WUNLOCK(&V_tcbinfo); 3012 ti_locked = TI_UNLOCKED; 3013 3014 #ifdef TCPDEBUG 3015 if (so->so_options & SO_DEBUG) 3016 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3017 &tcp_savetcp, 0); 3018 #endif 3019 3020 /* 3021 * Return any desired output. 3022 */ 3023 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3024 (void) tcp_output(tp); 3025 3026 check_delack: 3027 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3028 __func__, ti_locked)); 3029 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3030 INP_WLOCK_ASSERT(tp->t_inpcb); 3031 3032 if (tp->t_flags & TF_DELACK) { 3033 tp->t_flags &= ~TF_DELACK; 3034 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3035 } 3036 INP_WUNLOCK(tp->t_inpcb); 3037 return; 3038 3039 dropafterack: 3040 /* 3041 * Generate an ACK dropping incoming segment if it occupies 3042 * sequence space, where the ACK reflects our state. 3043 * 3044 * We can now skip the test for the RST flag since all 3045 * paths to this code happen after packets containing 3046 * RST have been dropped. 3047 * 3048 * In the SYN-RECEIVED state, don't send an ACK unless the 3049 * segment we received passes the SYN-RECEIVED ACK test. 3050 * If it fails send a RST. This breaks the loop in the 3051 * "LAND" DoS attack, and also prevents an ACK storm 3052 * between two listening ports that have been sent forged 3053 * SYN segments, each with the source address of the other. 3054 */ 3055 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3056 (SEQ_GT(tp->snd_una, th->th_ack) || 3057 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3058 rstreason = BANDLIM_RST_OPENPORT; 3059 goto dropwithreset; 3060 } 3061 #ifdef TCPDEBUG 3062 if (so->so_options & SO_DEBUG) 3063 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3064 &tcp_savetcp, 0); 3065 #endif 3066 if (ti_locked == TI_WLOCKED) 3067 INP_INFO_WUNLOCK(&V_tcbinfo); 3068 ti_locked = TI_UNLOCKED; 3069 3070 tp->t_flags |= TF_ACKNOW; 3071 (void) tcp_output(tp); 3072 INP_WUNLOCK(tp->t_inpcb); 3073 m_freem(m); 3074 return; 3075 3076 dropwithreset: 3077 if (ti_locked == TI_WLOCKED) 3078 INP_INFO_WUNLOCK(&V_tcbinfo); 3079 ti_locked = TI_UNLOCKED; 3080 3081 if (tp != NULL) { 3082 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3083 INP_WUNLOCK(tp->t_inpcb); 3084 } else 3085 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3086 return; 3087 3088 drop: 3089 if (ti_locked == TI_WLOCKED) { 3090 INP_INFO_WUNLOCK(&V_tcbinfo); 3091 ti_locked = TI_UNLOCKED; 3092 } 3093 #ifdef INVARIANTS 3094 else 3095 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3096 #endif 3097 3098 /* 3099 * Drop space held by incoming segment and return. 3100 */ 3101 #ifdef TCPDEBUG 3102 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3103 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3104 &tcp_savetcp, 0); 3105 #endif 3106 if (tp != NULL) 3107 INP_WUNLOCK(tp->t_inpcb); 3108 m_freem(m); 3109 } 3110 3111 /* 3112 * Issue RST and make ACK acceptable to originator of segment. 3113 * The mbuf must still include the original packet header. 3114 * tp may be NULL. 3115 */ 3116 static void 3117 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3118 int tlen, int rstreason) 3119 { 3120 #ifdef INET 3121 struct ip *ip; 3122 #endif 3123 #ifdef INET6 3124 struct ip6_hdr *ip6; 3125 #endif 3126 3127 if (tp != NULL) { 3128 INP_WLOCK_ASSERT(tp->t_inpcb); 3129 } 3130 3131 /* Don't bother if destination was broadcast/multicast. */ 3132 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3133 goto drop; 3134 #ifdef INET6 3135 if (mtod(m, struct ip *)->ip_v == 6) { 3136 ip6 = mtod(m, struct ip6_hdr *); 3137 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3138 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3139 goto drop; 3140 /* IPv6 anycast check is done at tcp6_input() */ 3141 } 3142 #endif 3143 #if defined(INET) && defined(INET6) 3144 else 3145 #endif 3146 #ifdef INET 3147 { 3148 ip = mtod(m, struct ip *); 3149 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3150 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3151 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3152 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3153 goto drop; 3154 } 3155 #endif 3156 3157 /* Perform bandwidth limiting. */ 3158 if (badport_bandlim(rstreason) < 0) 3159 goto drop; 3160 3161 /* tcp_respond consumes the mbuf chain. */ 3162 if (th->th_flags & TH_ACK) { 3163 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3164 th->th_ack, TH_RST); 3165 } else { 3166 if (th->th_flags & TH_SYN) 3167 tlen++; 3168 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3169 (tcp_seq)0, TH_RST|TH_ACK); 3170 } 3171 return; 3172 drop: 3173 m_freem(m); 3174 } 3175 3176 /* 3177 * Parse TCP options and place in tcpopt. 3178 */ 3179 static void 3180 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3181 { 3182 int opt, optlen; 3183 3184 to->to_flags = 0; 3185 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3186 opt = cp[0]; 3187 if (opt == TCPOPT_EOL) 3188 break; 3189 if (opt == TCPOPT_NOP) 3190 optlen = 1; 3191 else { 3192 if (cnt < 2) 3193 break; 3194 optlen = cp[1]; 3195 if (optlen < 2 || optlen > cnt) 3196 break; 3197 } 3198 switch (opt) { 3199 case TCPOPT_MAXSEG: 3200 if (optlen != TCPOLEN_MAXSEG) 3201 continue; 3202 if (!(flags & TO_SYN)) 3203 continue; 3204 to->to_flags |= TOF_MSS; 3205 bcopy((char *)cp + 2, 3206 (char *)&to->to_mss, sizeof(to->to_mss)); 3207 to->to_mss = ntohs(to->to_mss); 3208 break; 3209 case TCPOPT_WINDOW: 3210 if (optlen != TCPOLEN_WINDOW) 3211 continue; 3212 if (!(flags & TO_SYN)) 3213 continue; 3214 to->to_flags |= TOF_SCALE; 3215 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3216 break; 3217 case TCPOPT_TIMESTAMP: 3218 if (optlen != TCPOLEN_TIMESTAMP) 3219 continue; 3220 to->to_flags |= TOF_TS; 3221 bcopy((char *)cp + 2, 3222 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3223 to->to_tsval = ntohl(to->to_tsval); 3224 bcopy((char *)cp + 6, 3225 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3226 to->to_tsecr = ntohl(to->to_tsecr); 3227 break; 3228 #ifdef TCP_SIGNATURE 3229 /* 3230 * XXX In order to reply to a host which has set the 3231 * TCP_SIGNATURE option in its initial SYN, we have to 3232 * record the fact that the option was observed here 3233 * for the syncache code to perform the correct response. 3234 */ 3235 case TCPOPT_SIGNATURE: 3236 if (optlen != TCPOLEN_SIGNATURE) 3237 continue; 3238 to->to_flags |= TOF_SIGNATURE; 3239 to->to_signature = cp + 2; 3240 break; 3241 #endif 3242 case TCPOPT_SACK_PERMITTED: 3243 if (optlen != TCPOLEN_SACK_PERMITTED) 3244 continue; 3245 if (!(flags & TO_SYN)) 3246 continue; 3247 if (!V_tcp_do_sack) 3248 continue; 3249 to->to_flags |= TOF_SACKPERM; 3250 break; 3251 case TCPOPT_SACK: 3252 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3253 continue; 3254 if (flags & TO_SYN) 3255 continue; 3256 to->to_flags |= TOF_SACK; 3257 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3258 to->to_sacks = cp + 2; 3259 TCPSTAT_INC(tcps_sack_rcv_blocks); 3260 break; 3261 default: 3262 continue; 3263 } 3264 } 3265 } 3266 3267 /* 3268 * Pull out of band byte out of a segment so 3269 * it doesn't appear in the user's data queue. 3270 * It is still reflected in the segment length for 3271 * sequencing purposes. 3272 */ 3273 static void 3274 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3275 int off) 3276 { 3277 int cnt = off + th->th_urp - 1; 3278 3279 while (cnt >= 0) { 3280 if (m->m_len > cnt) { 3281 char *cp = mtod(m, caddr_t) + cnt; 3282 struct tcpcb *tp = sototcpcb(so); 3283 3284 INP_WLOCK_ASSERT(tp->t_inpcb); 3285 3286 tp->t_iobc = *cp; 3287 tp->t_oobflags |= TCPOOB_HAVEDATA; 3288 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3289 m->m_len--; 3290 if (m->m_flags & M_PKTHDR) 3291 m->m_pkthdr.len--; 3292 return; 3293 } 3294 cnt -= m->m_len; 3295 m = m->m_next; 3296 if (m == NULL) 3297 break; 3298 } 3299 panic("tcp_pulloutofband"); 3300 } 3301 3302 /* 3303 * Collect new round-trip time estimate 3304 * and update averages and current timeout. 3305 */ 3306 static void 3307 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3308 { 3309 int delta; 3310 3311 INP_WLOCK_ASSERT(tp->t_inpcb); 3312 3313 TCPSTAT_INC(tcps_rttupdated); 3314 tp->t_rttupdated++; 3315 if (tp->t_srtt != 0) { 3316 /* 3317 * srtt is stored as fixed point with 5 bits after the 3318 * binary point (i.e., scaled by 8). The following magic 3319 * is equivalent to the smoothing algorithm in rfc793 with 3320 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3321 * point). Adjust rtt to origin 0. 3322 */ 3323 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3324 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3325 3326 if ((tp->t_srtt += delta) <= 0) 3327 tp->t_srtt = 1; 3328 3329 /* 3330 * We accumulate a smoothed rtt variance (actually, a 3331 * smoothed mean difference), then set the retransmit 3332 * timer to smoothed rtt + 4 times the smoothed variance. 3333 * rttvar is stored as fixed point with 4 bits after the 3334 * binary point (scaled by 16). The following is 3335 * equivalent to rfc793 smoothing with an alpha of .75 3336 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3337 * rfc793's wired-in beta. 3338 */ 3339 if (delta < 0) 3340 delta = -delta; 3341 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3342 if ((tp->t_rttvar += delta) <= 0) 3343 tp->t_rttvar = 1; 3344 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3345 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3346 } else { 3347 /* 3348 * No rtt measurement yet - use the unsmoothed rtt. 3349 * Set the variance to half the rtt (so our first 3350 * retransmit happens at 3*rtt). 3351 */ 3352 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3353 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3354 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3355 } 3356 tp->t_rtttime = 0; 3357 tp->t_rxtshift = 0; 3358 3359 /* 3360 * the retransmit should happen at rtt + 4 * rttvar. 3361 * Because of the way we do the smoothing, srtt and rttvar 3362 * will each average +1/2 tick of bias. When we compute 3363 * the retransmit timer, we want 1/2 tick of rounding and 3364 * 1 extra tick because of +-1/2 tick uncertainty in the 3365 * firing of the timer. The bias will give us exactly the 3366 * 1.5 tick we need. But, because the bias is 3367 * statistical, we have to test that we don't drop below 3368 * the minimum feasible timer (which is 2 ticks). 3369 */ 3370 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3371 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3372 3373 /* 3374 * We received an ack for a packet that wasn't retransmitted; 3375 * it is probably safe to discard any error indications we've 3376 * received recently. This isn't quite right, but close enough 3377 * for now (a route might have failed after we sent a segment, 3378 * and the return path might not be symmetrical). 3379 */ 3380 tp->t_softerror = 0; 3381 } 3382 3383 /* 3384 * Determine a reasonable value for maxseg size. 3385 * If the route is known, check route for mtu. 3386 * If none, use an mss that can be handled on the outgoing interface 3387 * without forcing IP to fragment. If no route is found, route has no mtu, 3388 * or the destination isn't local, use a default, hopefully conservative 3389 * size (usually 512 or the default IP max size, but no more than the mtu 3390 * of the interface), as we can't discover anything about intervening 3391 * gateways or networks. We also initialize the congestion/slow start 3392 * window to be a single segment if the destination isn't local. 3393 * While looking at the routing entry, we also initialize other path-dependent 3394 * parameters from pre-set or cached values in the routing entry. 3395 * 3396 * Also take into account the space needed for options that we 3397 * send regularly. Make maxseg shorter by that amount to assure 3398 * that we can send maxseg amount of data even when the options 3399 * are present. Store the upper limit of the length of options plus 3400 * data in maxopd. 3401 * 3402 * NOTE that this routine is only called when we process an incoming 3403 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3404 * settings are handled in tcp_mssopt(). 3405 */ 3406 void 3407 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3408 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3409 { 3410 int mss = 0; 3411 u_long maxmtu = 0; 3412 struct inpcb *inp = tp->t_inpcb; 3413 struct hc_metrics_lite metrics; 3414 int origoffer; 3415 #ifdef INET6 3416 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3417 size_t min_protoh = isipv6 ? 3418 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3419 sizeof (struct tcpiphdr); 3420 #else 3421 const size_t min_protoh = sizeof(struct tcpiphdr); 3422 #endif 3423 3424 INP_WLOCK_ASSERT(tp->t_inpcb); 3425 3426 if (mtuoffer != -1) { 3427 KASSERT(offer == -1, ("%s: conflict", __func__)); 3428 offer = mtuoffer - min_protoh; 3429 } 3430 origoffer = offer; 3431 3432 /* Initialize. */ 3433 #ifdef INET6 3434 if (isipv6) { 3435 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3436 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3437 } 3438 #endif 3439 #if defined(INET) && defined(INET6) 3440 else 3441 #endif 3442 #ifdef INET 3443 { 3444 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3445 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3446 } 3447 #endif 3448 3449 /* 3450 * No route to sender, stay with default mss and return. 3451 */ 3452 if (maxmtu == 0) { 3453 /* 3454 * In case we return early we need to initialize metrics 3455 * to a defined state as tcp_hc_get() would do for us 3456 * if there was no cache hit. 3457 */ 3458 if (metricptr != NULL) 3459 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3460 return; 3461 } 3462 3463 /* What have we got? */ 3464 switch (offer) { 3465 case 0: 3466 /* 3467 * Offer == 0 means that there was no MSS on the SYN 3468 * segment, in this case we use tcp_mssdflt as 3469 * already assigned to t_maxopd above. 3470 */ 3471 offer = tp->t_maxopd; 3472 break; 3473 3474 case -1: 3475 /* 3476 * Offer == -1 means that we didn't receive SYN yet. 3477 */ 3478 /* FALLTHROUGH */ 3479 3480 default: 3481 /* 3482 * Prevent DoS attack with too small MSS. Round up 3483 * to at least minmss. 3484 */ 3485 offer = max(offer, V_tcp_minmss); 3486 } 3487 3488 /* 3489 * rmx information is now retrieved from tcp_hostcache. 3490 */ 3491 tcp_hc_get(&inp->inp_inc, &metrics); 3492 if (metricptr != NULL) 3493 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3494 3495 /* 3496 * If there's a discovered mtu in tcp hostcache, use it. 3497 * Else, use the link mtu. 3498 */ 3499 if (metrics.rmx_mtu) 3500 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3501 else { 3502 #ifdef INET6 3503 if (isipv6) { 3504 mss = maxmtu - min_protoh; 3505 if (!V_path_mtu_discovery && 3506 !in6_localaddr(&inp->in6p_faddr)) 3507 mss = min(mss, V_tcp_v6mssdflt); 3508 } 3509 #endif 3510 #if defined(INET) && defined(INET6) 3511 else 3512 #endif 3513 #ifdef INET 3514 { 3515 mss = maxmtu - min_protoh; 3516 if (!V_path_mtu_discovery && 3517 !in_localaddr(inp->inp_faddr)) 3518 mss = min(mss, V_tcp_mssdflt); 3519 } 3520 #endif 3521 /* 3522 * XXX - The above conditional (mss = maxmtu - min_protoh) 3523 * probably violates the TCP spec. 3524 * The problem is that, since we don't know the 3525 * other end's MSS, we are supposed to use a conservative 3526 * default. But, if we do that, then MTU discovery will 3527 * never actually take place, because the conservative 3528 * default is much less than the MTUs typically seen 3529 * on the Internet today. For the moment, we'll sweep 3530 * this under the carpet. 3531 * 3532 * The conservative default might not actually be a problem 3533 * if the only case this occurs is when sending an initial 3534 * SYN with options and data to a host we've never talked 3535 * to before. Then, they will reply with an MSS value which 3536 * will get recorded and the new parameters should get 3537 * recomputed. For Further Study. 3538 */ 3539 } 3540 mss = min(mss, offer); 3541 3542 /* 3543 * Sanity check: make sure that maxopd will be large 3544 * enough to allow some data on segments even if the 3545 * all the option space is used (40bytes). Otherwise 3546 * funny things may happen in tcp_output. 3547 */ 3548 mss = max(mss, 64); 3549 3550 /* 3551 * maxopd stores the maximum length of data AND options 3552 * in a segment; maxseg is the amount of data in a normal 3553 * segment. We need to store this value (maxopd) apart 3554 * from maxseg, because now every segment carries options 3555 * and thus we normally have somewhat less data in segments. 3556 */ 3557 tp->t_maxopd = mss; 3558 3559 /* 3560 * origoffer==-1 indicates that no segments were received yet. 3561 * In this case we just guess. 3562 */ 3563 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3564 (origoffer == -1 || 3565 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3566 mss -= TCPOLEN_TSTAMP_APPA; 3567 3568 tp->t_maxseg = mss; 3569 } 3570 3571 void 3572 tcp_mss(struct tcpcb *tp, int offer) 3573 { 3574 int mss; 3575 u_long bufsize; 3576 struct inpcb *inp; 3577 struct socket *so; 3578 struct hc_metrics_lite metrics; 3579 struct tcp_ifcap cap; 3580 3581 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3582 3583 bzero(&cap, sizeof(cap)); 3584 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3585 3586 mss = tp->t_maxseg; 3587 inp = tp->t_inpcb; 3588 3589 /* 3590 * If there's a pipesize, change the socket buffer to that size, 3591 * don't change if sb_hiwat is different than default (then it 3592 * has been changed on purpose with setsockopt). 3593 * Make the socket buffers an integral number of mss units; 3594 * if the mss is larger than the socket buffer, decrease the mss. 3595 */ 3596 so = inp->inp_socket; 3597 SOCKBUF_LOCK(&so->so_snd); 3598 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3599 bufsize = metrics.rmx_sendpipe; 3600 else 3601 bufsize = so->so_snd.sb_hiwat; 3602 if (bufsize < mss) 3603 mss = bufsize; 3604 else { 3605 bufsize = roundup(bufsize, mss); 3606 if (bufsize > sb_max) 3607 bufsize = sb_max; 3608 if (bufsize > so->so_snd.sb_hiwat) 3609 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3610 } 3611 SOCKBUF_UNLOCK(&so->so_snd); 3612 tp->t_maxseg = mss; 3613 3614 SOCKBUF_LOCK(&so->so_rcv); 3615 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3616 bufsize = metrics.rmx_recvpipe; 3617 else 3618 bufsize = so->so_rcv.sb_hiwat; 3619 if (bufsize > mss) { 3620 bufsize = roundup(bufsize, mss); 3621 if (bufsize > sb_max) 3622 bufsize = sb_max; 3623 if (bufsize > so->so_rcv.sb_hiwat) 3624 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3625 } 3626 SOCKBUF_UNLOCK(&so->so_rcv); 3627 3628 /* Check the interface for TSO capabilities. */ 3629 if (cap.ifcap & CSUM_TSO) { 3630 tp->t_flags |= TF_TSO; 3631 tp->t_tsomax = cap.tsomax; 3632 tp->t_tsomaxsegcount = cap.tsomaxsegcount; 3633 tp->t_tsomaxsegsize = cap.tsomaxsegsize; 3634 } 3635 } 3636 3637 /* 3638 * Determine the MSS option to send on an outgoing SYN. 3639 */ 3640 int 3641 tcp_mssopt(struct in_conninfo *inc) 3642 { 3643 int mss = 0; 3644 u_long maxmtu = 0; 3645 u_long thcmtu = 0; 3646 size_t min_protoh; 3647 3648 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3649 3650 #ifdef INET6 3651 if (inc->inc_flags & INC_ISIPV6) { 3652 mss = V_tcp_v6mssdflt; 3653 maxmtu = tcp_maxmtu6(inc, NULL); 3654 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3655 } 3656 #endif 3657 #if defined(INET) && defined(INET6) 3658 else 3659 #endif 3660 #ifdef INET 3661 { 3662 mss = V_tcp_mssdflt; 3663 maxmtu = tcp_maxmtu(inc, NULL); 3664 min_protoh = sizeof(struct tcpiphdr); 3665 } 3666 #endif 3667 #if defined(INET6) || defined(INET) 3668 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3669 #endif 3670 3671 if (maxmtu && thcmtu) 3672 mss = min(maxmtu, thcmtu) - min_protoh; 3673 else if (maxmtu || thcmtu) 3674 mss = max(maxmtu, thcmtu) - min_protoh; 3675 3676 return (mss); 3677 } 3678 3679 3680 /* 3681 * On a partial ack arrives, force the retransmission of the 3682 * next unacknowledged segment. Do not clear tp->t_dupacks. 3683 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3684 * be started again. 3685 */ 3686 static void 3687 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3688 { 3689 tcp_seq onxt = tp->snd_nxt; 3690 u_long ocwnd = tp->snd_cwnd; 3691 3692 INP_WLOCK_ASSERT(tp->t_inpcb); 3693 3694 tcp_timer_activate(tp, TT_REXMT, 0); 3695 tp->t_rtttime = 0; 3696 tp->snd_nxt = th->th_ack; 3697 /* 3698 * Set snd_cwnd to one segment beyond acknowledged offset. 3699 * (tp->snd_una has not yet been updated when this function is called.) 3700 */ 3701 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3702 tp->t_flags |= TF_ACKNOW; 3703 (void) tcp_output(tp); 3704 tp->snd_cwnd = ocwnd; 3705 if (SEQ_GT(onxt, tp->snd_nxt)) 3706 tp->snd_nxt = onxt; 3707 /* 3708 * Partial window deflation. Relies on fact that tp->snd_una 3709 * not updated yet. 3710 */ 3711 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3712 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3713 else 3714 tp->snd_cwnd = 0; 3715 tp->snd_cwnd += tp->t_maxseg; 3716 } 3717