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_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 134 &VNET_NAME(blackhole), 0, 135 "Do not send RST on segments to closed ports"); 136 137 VNET_DEFINE(int, tcp_delack_enabled) = 1; 138 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 139 &VNET_NAME(tcp_delack_enabled), 0, 140 "Delay ACK to try and piggyback it onto a data packet"); 141 142 VNET_DEFINE(int, drop_synfin) = 0; 143 #define V_drop_synfin VNET(drop_synfin) 144 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 145 &VNET_NAME(drop_synfin), 0, 146 "Drop TCP packets with SYN+FIN set"); 147 148 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 149 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 151 &VNET_NAME(tcp_do_rfc3042), 0, 152 "Enable RFC 3042 (Limited Transmit)"); 153 154 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 156 &VNET_NAME(tcp_do_rfc3390), 0, 157 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 158 159 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 160 "Experimental TCP extensions"); 161 162 VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 163 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 164 &VNET_NAME(tcp_do_initcwnd10), 0, 165 "Enable RFC 6928 (Increasing initial CWND to 10)"); 166 167 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 168 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 169 &VNET_NAME(tcp_do_rfc3465), 0, 170 "Enable RFC 3465 (Appropriate Byte Counting)"); 171 172 VNET_DEFINE(int, tcp_abc_l_var) = 2; 173 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 174 &VNET_NAME(tcp_abc_l_var), 2, 175 "Cap the max cwnd increment during slow-start to this number of segments"); 176 177 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 178 179 VNET_DEFINE(int, tcp_do_ecn) = 0; 180 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 181 &VNET_NAME(tcp_do_ecn), 0, 182 "TCP ECN support"); 183 184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 185 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 186 &VNET_NAME(tcp_ecn_maxretries), 0, 187 "Max retries before giving up on ECN"); 188 189 VNET_DEFINE(int, tcp_insecure_rst) = 0; 190 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 191 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 192 &VNET_NAME(tcp_insecure_rst), 0, 193 "Follow the old (insecure) criteria for accepting RST packets"); 194 195 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 196 #define V_tcp_recvspace VNET(tcp_recvspace) 197 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 198 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 199 200 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 201 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 202 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 203 &VNET_NAME(tcp_do_autorcvbuf), 0, 204 "Enable automatic receive buffer sizing"); 205 206 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 207 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 208 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 209 &VNET_NAME(tcp_autorcvbuf_inc), 0, 210 "Incrementor step size of automatic receive buffer"); 211 212 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 213 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 214 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 215 &VNET_NAME(tcp_autorcvbuf_max), 0, 216 "Max size of automatic receive buffer"); 217 218 VNET_DEFINE(struct inpcbhead, tcb); 219 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 220 VNET_DEFINE(struct inpcbinfo, tcbinfo); 221 222 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 223 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 224 struct socket *, struct tcpcb *, int, int, uint8_t, 225 int); 226 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 227 struct tcpcb *, int, int); 228 static void tcp_pulloutofband(struct socket *, 229 struct tcphdr *, struct mbuf *, int); 230 static void tcp_xmit_timer(struct tcpcb *, int); 231 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 232 static void inline tcp_fields_to_host(struct tcphdr *); 233 #ifdef TCP_SIGNATURE 234 static void inline tcp_fields_to_net(struct tcphdr *); 235 static int inline tcp_signature_verify_input(struct mbuf *, int, int, 236 int, struct tcpopt *, struct tcphdr *, u_int); 237 #endif 238 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 239 uint16_t type); 240 static void inline cc_conn_init(struct tcpcb *tp); 241 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 242 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 243 struct tcphdr *th, struct tcpopt *to); 244 245 /* 246 * TCP statistics are stored in an "array" of counter(9)s. 247 */ 248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 249 VNET_PCPUSTAT_SYSINIT(tcpstat); 250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 252 253 #ifdef VIMAGE 254 VNET_PCPUSTAT_SYSUNINIT(tcpstat); 255 #endif /* VIMAGE */ 256 /* 257 * Kernel module interface for updating tcpstat. The argument is an index 258 * into tcpstat treated as an array. 259 */ 260 void 261 kmod_tcpstat_inc(int statnum) 262 { 263 264 counter_u64_add(VNET(tcpstat)[statnum], 1); 265 } 266 267 /* 268 * Wrapper for the TCP established input helper hook. 269 */ 270 static void inline 271 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 272 { 273 struct tcp_hhook_data hhook_data; 274 275 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 276 hhook_data.tp = tp; 277 hhook_data.th = th; 278 hhook_data.to = to; 279 280 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 281 tp->osd); 282 } 283 } 284 285 /* 286 * CC wrapper hook functions 287 */ 288 static void inline 289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 290 { 291 INP_WLOCK_ASSERT(tp->t_inpcb); 292 293 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 294 if (tp->snd_cwnd <= tp->snd_wnd) 295 tp->ccv->flags |= CCF_CWND_LIMITED; 296 else 297 tp->ccv->flags &= ~CCF_CWND_LIMITED; 298 299 if (type == CC_ACK) { 300 if (tp->snd_cwnd > tp->snd_ssthresh) { 301 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 302 V_tcp_abc_l_var * tp->t_maxseg); 303 if (tp->t_bytes_acked >= tp->snd_cwnd) { 304 tp->t_bytes_acked -= tp->snd_cwnd; 305 tp->ccv->flags |= CCF_ABC_SENTAWND; 306 } 307 } else { 308 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 309 tp->t_bytes_acked = 0; 310 } 311 } 312 313 if (CC_ALGO(tp)->ack_received != NULL) { 314 /* XXXLAS: Find a way to live without this */ 315 tp->ccv->curack = th->th_ack; 316 CC_ALGO(tp)->ack_received(tp->ccv, type); 317 } 318 } 319 320 static void inline 321 cc_conn_init(struct tcpcb *tp) 322 { 323 struct hc_metrics_lite metrics; 324 struct inpcb *inp = tp->t_inpcb; 325 int rtt; 326 327 INP_WLOCK_ASSERT(tp->t_inpcb); 328 329 tcp_hc_get(&inp->inp_inc, &metrics); 330 331 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 332 tp->t_srtt = rtt; 333 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 334 TCPSTAT_INC(tcps_usedrtt); 335 if (metrics.rmx_rttvar) { 336 tp->t_rttvar = metrics.rmx_rttvar; 337 TCPSTAT_INC(tcps_usedrttvar); 338 } else { 339 /* default variation is +- 1 rtt */ 340 tp->t_rttvar = 341 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 342 } 343 TCPT_RANGESET(tp->t_rxtcur, 344 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 345 tp->t_rttmin, TCPTV_REXMTMAX); 346 } 347 if (metrics.rmx_ssthresh) { 348 /* 349 * There's some sort of gateway or interface 350 * buffer limit on the path. Use this to set 351 * the slow start threshhold, but set the 352 * threshold to no less than 2*mss. 353 */ 354 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 355 TCPSTAT_INC(tcps_usedssthresh); 356 } 357 358 /* 359 * Set the initial slow-start flight size. 360 * 361 * RFC5681 Section 3.1 specifies the default conservative values. 362 * RFC3390 specifies slightly more aggressive values. 363 * 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 static inline void 459 tcp_fields_to_host(struct tcphdr *th) 460 { 461 462 th->th_seq = ntohl(th->th_seq); 463 th->th_ack = ntohl(th->th_ack); 464 th->th_win = ntohs(th->th_win); 465 th->th_urp = ntohs(th->th_urp); 466 } 467 468 #ifdef TCP_SIGNATURE 469 static inline void 470 tcp_fields_to_net(struct tcphdr *th) 471 { 472 473 th->th_seq = htonl(th->th_seq); 474 th->th_ack = htonl(th->th_ack); 475 th->th_win = htons(th->th_win); 476 th->th_urp = htons(th->th_urp); 477 } 478 479 static inline int 480 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 481 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 482 { 483 int ret; 484 485 tcp_fields_to_net(th); 486 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 487 tcp_fields_to_host(th); 488 return (ret); 489 } 490 #endif 491 492 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 493 #ifdef INET6 494 #define ND6_HINT(tp) \ 495 do { \ 496 if ((tp) && (tp)->t_inpcb && \ 497 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 498 nd6_nud_hint(NULL, NULL, 0); \ 499 } while (0) 500 #else 501 #define ND6_HINT(tp) 502 #endif 503 504 /* 505 * Indicate whether this ack should be delayed. We can delay the ack if 506 * following conditions are met: 507 * - There is no delayed ack timer in progress. 508 * - Our last ack wasn't a 0-sized window. We never want to delay 509 * the ack that opens up a 0-sized window. 510 * - LRO wasn't used for this segment. We make sure by checking that the 511 * segment size is not larger than the MSS. 512 * - Delayed acks are enabled or this is a half-synchronized T/TCP 513 * connection. 514 */ 515 #define DELAY_ACK(tp, tlen) \ 516 ((!tcp_timer_active(tp, TT_DELACK) && \ 517 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 518 (tlen <= tp->t_maxopd) && \ 519 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 520 521 /* 522 * TCP input handling is split into multiple parts: 523 * tcp6_input is a thin wrapper around tcp_input for the extended 524 * ip6_protox[] call format in ip6_input 525 * tcp_input handles primary segment validation, inpcb lookup and 526 * SYN processing on listen sockets 527 * tcp_do_segment processes the ACK and text of the segment for 528 * establishing, established and closing connections 529 */ 530 #ifdef INET6 531 int 532 tcp6_input(struct mbuf **mp, int *offp, int proto) 533 { 534 struct mbuf *m = *mp; 535 struct in6_ifaddr *ia6; 536 537 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 538 539 /* 540 * draft-itojun-ipv6-tcp-to-anycast 541 * better place to put this in? 542 */ 543 ia6 = ip6_getdstifaddr(m); 544 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 545 struct ip6_hdr *ip6; 546 547 ifa_free(&ia6->ia_ifa); 548 ip6 = mtod(m, struct ip6_hdr *); 549 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 550 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 551 return IPPROTO_DONE; 552 } 553 if (ia6) 554 ifa_free(&ia6->ia_ifa); 555 556 tcp_input(m, *offp); 557 return IPPROTO_DONE; 558 } 559 #endif /* INET6 */ 560 561 void 562 tcp_input(struct mbuf *m, int off0) 563 { 564 struct tcphdr *th = NULL; 565 struct ip *ip = NULL; 566 struct inpcb *inp = NULL; 567 struct tcpcb *tp = NULL; 568 struct socket *so = NULL; 569 u_char *optp = NULL; 570 int optlen = 0; 571 #ifdef INET 572 int len; 573 #endif 574 int tlen = 0, off; 575 int drop_hdrlen; 576 int thflags; 577 int rstreason = 0; /* For badport_bandlim accounting purposes */ 578 #ifdef TCP_SIGNATURE 579 uint8_t sig_checked = 0; 580 #endif 581 uint8_t iptos = 0; 582 struct m_tag *fwd_tag = NULL; 583 #ifdef INET6 584 struct ip6_hdr *ip6 = NULL; 585 int isipv6; 586 #else 587 const void *ip6 = NULL; 588 #endif /* INET6 */ 589 struct tcpopt to; /* options in this segment */ 590 char *s = NULL; /* address and port logging */ 591 int ti_locked; 592 #define TI_UNLOCKED 1 593 #define TI_WLOCKED 2 594 595 #ifdef TCPDEBUG 596 /* 597 * The size of tcp_saveipgen must be the size of the max ip header, 598 * now IPv6. 599 */ 600 u_char tcp_saveipgen[IP6_HDR_LEN]; 601 struct tcphdr tcp_savetcp; 602 short ostate = 0; 603 #endif 604 605 #ifdef INET6 606 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 607 #endif 608 609 to.to_flags = 0; 610 TCPSTAT_INC(tcps_rcvtotal); 611 612 #ifdef INET6 613 if (isipv6) { 614 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 615 616 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 617 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 618 if (m == NULL) { 619 TCPSTAT_INC(tcps_rcvshort); 620 return; 621 } 622 } 623 624 ip6 = mtod(m, struct ip6_hdr *); 625 th = (struct tcphdr *)((caddr_t)ip6 + off0); 626 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 627 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 628 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 629 th->th_sum = m->m_pkthdr.csum_data; 630 else 631 th->th_sum = in6_cksum_pseudo(ip6, tlen, 632 IPPROTO_TCP, m->m_pkthdr.csum_data); 633 th->th_sum ^= 0xffff; 634 } else 635 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 636 if (th->th_sum) { 637 TCPSTAT_INC(tcps_rcvbadsum); 638 goto drop; 639 } 640 641 /* 642 * Be proactive about unspecified IPv6 address in source. 643 * As we use all-zero to indicate unbounded/unconnected pcb, 644 * unspecified IPv6 address can be used to confuse us. 645 * 646 * Note that packets with unspecified IPv6 destination is 647 * already dropped in ip6_input. 648 */ 649 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 650 /* XXX stat */ 651 goto drop; 652 } 653 } 654 #endif 655 #if defined(INET) && defined(INET6) 656 else 657 #endif 658 #ifdef INET 659 { 660 /* 661 * Get IP and TCP header together in first mbuf. 662 * Note: IP leaves IP header in first mbuf. 663 */ 664 if (off0 > sizeof (struct ip)) { 665 ip_stripoptions(m); 666 off0 = sizeof(struct ip); 667 } 668 if (m->m_len < sizeof (struct tcpiphdr)) { 669 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 670 == NULL) { 671 TCPSTAT_INC(tcps_rcvshort); 672 return; 673 } 674 } 675 ip = mtod(m, struct ip *); 676 th = (struct tcphdr *)((caddr_t)ip + off0); 677 tlen = ntohs(ip->ip_len) - off0; 678 679 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 680 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 681 th->th_sum = m->m_pkthdr.csum_data; 682 else 683 th->th_sum = in_pseudo(ip->ip_src.s_addr, 684 ip->ip_dst.s_addr, 685 htonl(m->m_pkthdr.csum_data + tlen + 686 IPPROTO_TCP)); 687 th->th_sum ^= 0xffff; 688 } else { 689 struct ipovly *ipov = (struct ipovly *)ip; 690 691 /* 692 * Checksum extended TCP header and data. 693 */ 694 len = off0 + tlen; 695 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 696 ipov->ih_len = htons(tlen); 697 th->th_sum = in_cksum(m, len); 698 /* Reset length for SDT probes. */ 699 ip->ip_len = htons(tlen + off0); 700 } 701 702 if (th->th_sum) { 703 TCPSTAT_INC(tcps_rcvbadsum); 704 goto drop; 705 } 706 /* Re-initialization for later version check */ 707 ip->ip_v = IPVERSION; 708 } 709 #endif /* INET */ 710 711 #ifdef INET6 712 if (isipv6) 713 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 714 #endif 715 #if defined(INET) && defined(INET6) 716 else 717 #endif 718 #ifdef INET 719 iptos = ip->ip_tos; 720 #endif 721 722 /* 723 * Check that TCP offset makes sense, 724 * pull out TCP options and adjust length. XXX 725 */ 726 off = th->th_off << 2; 727 if (off < sizeof (struct tcphdr) || off > tlen) { 728 TCPSTAT_INC(tcps_rcvbadoff); 729 goto drop; 730 } 731 tlen -= off; /* tlen is used instead of ti->ti_len */ 732 if (off > sizeof (struct tcphdr)) { 733 #ifdef INET6 734 if (isipv6) { 735 IP6_EXTHDR_CHECK(m, off0, off, ); 736 ip6 = mtod(m, struct ip6_hdr *); 737 th = (struct tcphdr *)((caddr_t)ip6 + off0); 738 } 739 #endif 740 #if defined(INET) && defined(INET6) 741 else 742 #endif 743 #ifdef INET 744 { 745 if (m->m_len < sizeof(struct ip) + off) { 746 if ((m = m_pullup(m, sizeof (struct ip) + off)) 747 == NULL) { 748 TCPSTAT_INC(tcps_rcvshort); 749 return; 750 } 751 ip = mtod(m, struct ip *); 752 th = (struct tcphdr *)((caddr_t)ip + off0); 753 } 754 } 755 #endif 756 optlen = off - sizeof (struct tcphdr); 757 optp = (u_char *)(th + 1); 758 } 759 thflags = th->th_flags; 760 761 /* 762 * Convert TCP protocol specific fields to host format. 763 */ 764 tcp_fields_to_host(th); 765 766 /* 767 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 768 */ 769 drop_hdrlen = off0 + off; 770 771 /* 772 * Locate pcb for segment; if we're likely to add or remove a 773 * connection then first acquire pcbinfo lock. There are two cases 774 * where we might discover later we need a write lock despite the 775 * flags: ACKs moving a connection out of the syncache, and ACKs for 776 * a connection in TIMEWAIT. 777 */ 778 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 779 INP_INFO_WLOCK(&V_tcbinfo); 780 ti_locked = TI_WLOCKED; 781 } else 782 ti_locked = TI_UNLOCKED; 783 784 /* 785 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 786 */ 787 if ( 788 #ifdef INET6 789 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 790 #ifdef INET 791 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 792 #endif 793 #endif 794 #if defined(INET) && !defined(INET6) 795 (m->m_flags & M_IP_NEXTHOP) 796 #endif 797 ) 798 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 799 800 findpcb: 801 #ifdef INVARIANTS 802 if (ti_locked == TI_WLOCKED) { 803 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 804 } else { 805 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 806 } 807 #endif 808 #ifdef INET6 809 if (isipv6 && fwd_tag != NULL) { 810 struct sockaddr_in6 *next_hop6; 811 812 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 813 /* 814 * Transparently forwarded. Pretend to be the destination. 815 * Already got one like this? 816 */ 817 inp = in6_pcblookup_mbuf(&V_tcbinfo, 818 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 819 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 820 if (!inp) { 821 /* 822 * It's new. Try to find the ambushing socket. 823 * Because we've rewritten the destination address, 824 * any hardware-generated hash is ignored. 825 */ 826 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 827 th->th_sport, &next_hop6->sin6_addr, 828 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 829 th->th_dport, INPLOOKUP_WILDCARD | 830 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 831 } 832 } else if (isipv6) { 833 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 834 th->th_sport, &ip6->ip6_dst, th->th_dport, 835 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 836 m->m_pkthdr.rcvif, m); 837 } 838 #endif /* INET6 */ 839 #if defined(INET6) && defined(INET) 840 else 841 #endif 842 #ifdef INET 843 if (fwd_tag != NULL) { 844 struct sockaddr_in *next_hop; 845 846 next_hop = (struct sockaddr_in *)(fwd_tag+1); 847 /* 848 * Transparently forwarded. Pretend to be the destination. 849 * already got one like this? 850 */ 851 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 852 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 853 m->m_pkthdr.rcvif, m); 854 if (!inp) { 855 /* 856 * It's new. Try to find the ambushing socket. 857 * Because we've rewritten the destination address, 858 * any hardware-generated hash is ignored. 859 */ 860 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 861 th->th_sport, next_hop->sin_addr, 862 next_hop->sin_port ? ntohs(next_hop->sin_port) : 863 th->th_dport, INPLOOKUP_WILDCARD | 864 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 865 } 866 } else 867 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 868 th->th_sport, ip->ip_dst, th->th_dport, 869 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 870 m->m_pkthdr.rcvif, m); 871 #endif /* INET */ 872 873 /* 874 * If the INPCB does not exist then all data in the incoming 875 * segment is discarded and an appropriate RST is sent back. 876 * XXX MRT Send RST using which routing table? 877 */ 878 if (inp == NULL) { 879 /* 880 * Log communication attempts to ports that are not 881 * in use. 882 */ 883 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 884 tcp_log_in_vain == 2) { 885 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 886 log(LOG_INFO, "%s; %s: Connection attempt " 887 "to closed port\n", s, __func__); 888 } 889 /* 890 * When blackholing do not respond with a RST but 891 * completely ignore the segment and drop it. 892 */ 893 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 894 V_blackhole == 2) 895 goto dropunlock; 896 897 rstreason = BANDLIM_RST_CLOSEDPORT; 898 goto dropwithreset; 899 } 900 INP_WLOCK_ASSERT(inp); 901 if (!(inp->inp_flags & INP_HW_FLOWID) 902 && (m->m_flags & M_FLOWID) 903 && ((inp->inp_socket == NULL) 904 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 905 inp->inp_flags |= INP_HW_FLOWID; 906 inp->inp_flags &= ~INP_SW_FLOWID; 907 inp->inp_flowid = m->m_pkthdr.flowid; 908 inp->inp_flowtype = M_HASHTYPE_GET(m); 909 } 910 #ifdef IPSEC 911 #ifdef INET6 912 if (isipv6 && ipsec6_in_reject(m, inp)) { 913 IPSEC6STAT_INC(ips_in_polvio); 914 goto dropunlock; 915 } else 916 #endif /* INET6 */ 917 if (ipsec4_in_reject(m, inp) != 0) { 918 IPSECSTAT_INC(ips_in_polvio); 919 goto dropunlock; 920 } 921 #endif /* IPSEC */ 922 923 /* 924 * Check the minimum TTL for socket. 925 */ 926 if (inp->inp_ip_minttl != 0) { 927 #ifdef INET6 928 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 929 goto dropunlock; 930 else 931 #endif 932 if (inp->inp_ip_minttl > ip->ip_ttl) 933 goto dropunlock; 934 } 935 936 /* 937 * A previous connection in TIMEWAIT state is supposed to catch stray 938 * or duplicate segments arriving late. If this segment was a 939 * legitimate new connection attempt, the old INPCB gets removed and 940 * we can try again to find a listening socket. 941 * 942 * At this point, due to earlier optimism, we may hold only an inpcb 943 * lock, and not the inpcbinfo write lock. If so, we need to try to 944 * acquire it, or if that fails, acquire a reference on the inpcb, 945 * drop all locks, acquire a global write lock, and then re-acquire 946 * the inpcb lock. We may at that point discover that another thread 947 * has tried to free the inpcb, in which case we need to loop back 948 * and try to find a new inpcb to deliver to. 949 * 950 * XXXRW: It may be time to rethink timewait locking. 951 */ 952 relocked: 953 if (inp->inp_flags & INP_TIMEWAIT) { 954 if (ti_locked == TI_UNLOCKED) { 955 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 956 in_pcbref(inp); 957 INP_WUNLOCK(inp); 958 INP_INFO_WLOCK(&V_tcbinfo); 959 ti_locked = TI_WLOCKED; 960 INP_WLOCK(inp); 961 if (in_pcbrele_wlocked(inp)) { 962 inp = NULL; 963 goto findpcb; 964 } 965 } else 966 ti_locked = TI_WLOCKED; 967 } 968 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 969 970 if (thflags & TH_SYN) 971 tcp_dooptions(&to, optp, optlen, TO_SYN); 972 /* 973 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 974 */ 975 if (tcp_twcheck(inp, &to, th, m, tlen)) 976 goto findpcb; 977 INP_INFO_WUNLOCK(&V_tcbinfo); 978 return; 979 } 980 /* 981 * The TCPCB may no longer exist if the connection is winding 982 * down or it is in the CLOSED state. Either way we drop the 983 * segment and send an appropriate response. 984 */ 985 tp = intotcpcb(inp); 986 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 987 rstreason = BANDLIM_RST_CLOSEDPORT; 988 goto dropwithreset; 989 } 990 991 #ifdef TCP_OFFLOAD 992 if (tp->t_flags & TF_TOE) { 993 tcp_offload_input(tp, m); 994 m = NULL; /* consumed by the TOE driver */ 995 goto dropunlock; 996 } 997 #endif 998 999 /* 1000 * We've identified a valid inpcb, but it could be that we need an 1001 * inpcbinfo write lock but don't hold it. In this case, attempt to 1002 * acquire using the same strategy as the TIMEWAIT case above. If we 1003 * relock, we have to jump back to 'relocked' as the connection might 1004 * now be in TIMEWAIT. 1005 */ 1006 #ifdef INVARIANTS 1007 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 1008 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1009 #endif 1010 if (tp->t_state != TCPS_ESTABLISHED) { 1011 if (ti_locked == TI_UNLOCKED) { 1012 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 1013 in_pcbref(inp); 1014 INP_WUNLOCK(inp); 1015 INP_INFO_WLOCK(&V_tcbinfo); 1016 ti_locked = TI_WLOCKED; 1017 INP_WLOCK(inp); 1018 if (in_pcbrele_wlocked(inp)) { 1019 inp = NULL; 1020 goto findpcb; 1021 } 1022 goto relocked; 1023 } else 1024 ti_locked = TI_WLOCKED; 1025 } 1026 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1027 } 1028 1029 #ifdef MAC 1030 INP_WLOCK_ASSERT(inp); 1031 if (mac_inpcb_check_deliver(inp, m)) 1032 goto dropunlock; 1033 #endif 1034 so = inp->inp_socket; 1035 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1036 #ifdef TCPDEBUG 1037 if (so->so_options & SO_DEBUG) { 1038 ostate = tp->t_state; 1039 #ifdef INET6 1040 if (isipv6) { 1041 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1042 } else 1043 #endif 1044 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1045 tcp_savetcp = *th; 1046 } 1047 #endif /* TCPDEBUG */ 1048 /* 1049 * When the socket is accepting connections (the INPCB is in LISTEN 1050 * state) we look into the SYN cache if this is a new connection 1051 * attempt or the completion of a previous one. Because listen 1052 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1053 * held in this case. 1054 */ 1055 if (so->so_options & SO_ACCEPTCONN) { 1056 struct in_conninfo inc; 1057 1058 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1059 "tp not listening", __func__)); 1060 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1061 1062 bzero(&inc, sizeof(inc)); 1063 #ifdef INET6 1064 if (isipv6) { 1065 inc.inc_flags |= INC_ISIPV6; 1066 inc.inc6_faddr = ip6->ip6_src; 1067 inc.inc6_laddr = ip6->ip6_dst; 1068 } else 1069 #endif 1070 { 1071 inc.inc_faddr = ip->ip_src; 1072 inc.inc_laddr = ip->ip_dst; 1073 } 1074 inc.inc_fport = th->th_sport; 1075 inc.inc_lport = th->th_dport; 1076 inc.inc_fibnum = so->so_fibnum; 1077 1078 /* 1079 * Check for an existing connection attempt in syncache if 1080 * the flag is only ACK. A successful lookup creates a new 1081 * socket appended to the listen queue in SYN_RECEIVED state. 1082 */ 1083 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1084 /* 1085 * Parse the TCP options here because 1086 * syncookies need access to the reflected 1087 * timestamp. 1088 */ 1089 tcp_dooptions(&to, optp, optlen, 0); 1090 /* 1091 * NB: syncache_expand() doesn't unlock 1092 * inp and tcpinfo locks. 1093 */ 1094 if (!syncache_expand(&inc, &to, th, &so, m)) { 1095 /* 1096 * No syncache entry or ACK was not 1097 * for our SYN/ACK. Send a RST. 1098 * NB: syncache did its own logging 1099 * of the failure cause. 1100 */ 1101 rstreason = BANDLIM_RST_OPENPORT; 1102 goto dropwithreset; 1103 } 1104 if (so == NULL) { 1105 /* 1106 * We completed the 3-way handshake 1107 * but could not allocate a socket 1108 * either due to memory shortage, 1109 * listen queue length limits or 1110 * global socket limits. Send RST 1111 * or wait and have the remote end 1112 * retransmit the ACK for another 1113 * try. 1114 */ 1115 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1116 log(LOG_DEBUG, "%s; %s: Listen socket: " 1117 "Socket allocation failed due to " 1118 "limits or memory shortage, %s\n", 1119 s, __func__, 1120 V_tcp_sc_rst_sock_fail ? 1121 "sending RST" : "try again"); 1122 if (V_tcp_sc_rst_sock_fail) { 1123 rstreason = BANDLIM_UNLIMITED; 1124 goto dropwithreset; 1125 } else 1126 goto dropunlock; 1127 } 1128 /* 1129 * Socket is created in state SYN_RECEIVED. 1130 * Unlock the listen socket, lock the newly 1131 * created socket and update the tp variable. 1132 */ 1133 INP_WUNLOCK(inp); /* listen socket */ 1134 inp = sotoinpcb(so); 1135 INP_WLOCK(inp); /* new connection */ 1136 tp = intotcpcb(inp); 1137 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1138 ("%s: ", __func__)); 1139 #ifdef TCP_SIGNATURE 1140 if (sig_checked == 0) { 1141 tcp_dooptions(&to, optp, optlen, 1142 (thflags & TH_SYN) ? TO_SYN : 0); 1143 if (!tcp_signature_verify_input(m, off0, tlen, 1144 optlen, &to, th, tp->t_flags)) { 1145 1146 /* 1147 * In SYN_SENT state if it receives an 1148 * RST, it is allowed for further 1149 * processing. 1150 */ 1151 if ((thflags & TH_RST) == 0 || 1152 (tp->t_state == TCPS_SYN_SENT) == 0) 1153 goto dropunlock; 1154 } 1155 sig_checked = 1; 1156 } 1157 #endif 1158 1159 /* 1160 * Process the segment and the data it 1161 * contains. tcp_do_segment() consumes 1162 * the mbuf chain and unlocks the inpcb. 1163 */ 1164 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1165 iptos, ti_locked); 1166 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1167 return; 1168 } 1169 /* 1170 * Segment flag validation for new connection attempts: 1171 * 1172 * Our (SYN|ACK) response was rejected. 1173 * Check with syncache and remove entry to prevent 1174 * retransmits. 1175 * 1176 * NB: syncache_chkrst does its own logging of failure 1177 * causes. 1178 */ 1179 if (thflags & TH_RST) { 1180 syncache_chkrst(&inc, th); 1181 goto dropunlock; 1182 } 1183 /* 1184 * We can't do anything without SYN. 1185 */ 1186 if ((thflags & TH_SYN) == 0) { 1187 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1188 log(LOG_DEBUG, "%s; %s: Listen socket: " 1189 "SYN is missing, segment ignored\n", 1190 s, __func__); 1191 TCPSTAT_INC(tcps_badsyn); 1192 goto dropunlock; 1193 } 1194 /* 1195 * (SYN|ACK) is bogus on a listen socket. 1196 */ 1197 if (thflags & TH_ACK) { 1198 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1199 log(LOG_DEBUG, "%s; %s: Listen socket: " 1200 "SYN|ACK invalid, segment rejected\n", 1201 s, __func__); 1202 syncache_badack(&inc); /* XXX: Not needed! */ 1203 TCPSTAT_INC(tcps_badsyn); 1204 rstreason = BANDLIM_RST_OPENPORT; 1205 goto dropwithreset; 1206 } 1207 /* 1208 * If the drop_synfin option is enabled, drop all 1209 * segments with both the SYN and FIN bits set. 1210 * This prevents e.g. nmap from identifying the 1211 * TCP/IP stack. 1212 * XXX: Poor reasoning. nmap has other methods 1213 * and is constantly refining its stack detection 1214 * strategies. 1215 * XXX: This is a violation of the TCP specification 1216 * and was used by RFC1644. 1217 */ 1218 if ((thflags & TH_FIN) && V_drop_synfin) { 1219 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1220 log(LOG_DEBUG, "%s; %s: Listen socket: " 1221 "SYN|FIN segment ignored (based on " 1222 "sysctl setting)\n", s, __func__); 1223 TCPSTAT_INC(tcps_badsyn); 1224 goto dropunlock; 1225 } 1226 /* 1227 * Segment's flags are (SYN) or (SYN|FIN). 1228 * 1229 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1230 * as they do not affect the state of the TCP FSM. 1231 * The data pointed to by TH_URG and th_urp is ignored. 1232 */ 1233 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1234 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1235 KASSERT(thflags & (TH_SYN), 1236 ("%s: Listen socket: TH_SYN not set", __func__)); 1237 #ifdef INET6 1238 /* 1239 * If deprecated address is forbidden, 1240 * we do not accept SYN to deprecated interface 1241 * address to prevent any new inbound connection from 1242 * getting established. 1243 * When we do not accept SYN, we send a TCP RST, 1244 * with deprecated source address (instead of dropping 1245 * it). We compromise it as it is much better for peer 1246 * to send a RST, and RST will be the final packet 1247 * for the exchange. 1248 * 1249 * If we do not forbid deprecated addresses, we accept 1250 * the SYN packet. RFC2462 does not suggest dropping 1251 * SYN in this case. 1252 * If we decipher RFC2462 5.5.4, it says like this: 1253 * 1. use of deprecated addr with existing 1254 * communication is okay - "SHOULD continue to be 1255 * used" 1256 * 2. use of it with new communication: 1257 * (2a) "SHOULD NOT be used if alternate address 1258 * with sufficient scope is available" 1259 * (2b) nothing mentioned otherwise. 1260 * Here we fall into (2b) case as we have no choice in 1261 * our source address selection - we must obey the peer. 1262 * 1263 * The wording in RFC2462 is confusing, and there are 1264 * multiple description text for deprecated address 1265 * handling - worse, they are not exactly the same. 1266 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1267 */ 1268 if (isipv6 && !V_ip6_use_deprecated) { 1269 struct in6_ifaddr *ia6; 1270 1271 ia6 = ip6_getdstifaddr(m); 1272 if (ia6 != NULL && 1273 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1274 ifa_free(&ia6->ia_ifa); 1275 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1276 log(LOG_DEBUG, "%s; %s: Listen socket: " 1277 "Connection attempt to deprecated " 1278 "IPv6 address rejected\n", 1279 s, __func__); 1280 rstreason = BANDLIM_RST_OPENPORT; 1281 goto dropwithreset; 1282 } 1283 if (ia6) 1284 ifa_free(&ia6->ia_ifa); 1285 } 1286 #endif /* INET6 */ 1287 /* 1288 * Basic sanity checks on incoming SYN requests: 1289 * Don't respond if the destination is a link layer 1290 * broadcast according to RFC1122 4.2.3.10, p. 104. 1291 * If it is from this socket it must be forged. 1292 * Don't respond if the source or destination is a 1293 * global or subnet broad- or multicast address. 1294 * Note that it is quite possible to receive unicast 1295 * link-layer packets with a broadcast IP address. Use 1296 * in_broadcast() to find them. 1297 */ 1298 if (m->m_flags & (M_BCAST|M_MCAST)) { 1299 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1300 log(LOG_DEBUG, "%s; %s: Listen socket: " 1301 "Connection attempt from broad- or multicast " 1302 "link layer address ignored\n", s, __func__); 1303 goto dropunlock; 1304 } 1305 #ifdef INET6 1306 if (isipv6) { 1307 if (th->th_dport == th->th_sport && 1308 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1309 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1310 log(LOG_DEBUG, "%s; %s: Listen socket: " 1311 "Connection attempt to/from self " 1312 "ignored\n", s, __func__); 1313 goto dropunlock; 1314 } 1315 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1316 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1318 log(LOG_DEBUG, "%s; %s: Listen socket: " 1319 "Connection attempt from/to multicast " 1320 "address ignored\n", s, __func__); 1321 goto dropunlock; 1322 } 1323 } 1324 #endif 1325 #if defined(INET) && defined(INET6) 1326 else 1327 #endif 1328 #ifdef INET 1329 { 1330 if (th->th_dport == th->th_sport && 1331 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1332 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1333 log(LOG_DEBUG, "%s; %s: Listen socket: " 1334 "Connection attempt from/to self " 1335 "ignored\n", s, __func__); 1336 goto dropunlock; 1337 } 1338 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1339 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1340 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1341 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1342 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1343 log(LOG_DEBUG, "%s; %s: Listen socket: " 1344 "Connection attempt from/to broad- " 1345 "or multicast address ignored\n", 1346 s, __func__); 1347 goto dropunlock; 1348 } 1349 } 1350 #endif 1351 /* 1352 * SYN appears to be valid. Create compressed TCP state 1353 * for syncache. 1354 */ 1355 #ifdef TCPDEBUG 1356 if (so->so_options & SO_DEBUG) 1357 tcp_trace(TA_INPUT, ostate, tp, 1358 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1359 #endif 1360 tcp_dooptions(&to, optp, optlen, TO_SYN); 1361 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1362 /* 1363 * Entry added to syncache and mbuf consumed. 1364 * Everything already unlocked by syncache_add(). 1365 */ 1366 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1367 return; 1368 } else if (tp->t_state == TCPS_LISTEN) { 1369 /* 1370 * When a listen socket is torn down the SO_ACCEPTCONN 1371 * flag is removed first while connections are drained 1372 * from the accept queue in a unlock/lock cycle of the 1373 * ACCEPT_LOCK, opening a race condition allowing a SYN 1374 * attempt go through unhandled. 1375 */ 1376 goto dropunlock; 1377 } 1378 1379 #ifdef TCP_SIGNATURE 1380 if (sig_checked == 0) { 1381 tcp_dooptions(&to, optp, optlen, 1382 (thflags & TH_SYN) ? TO_SYN : 0); 1383 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1384 th, tp->t_flags)) { 1385 1386 /* 1387 * In SYN_SENT state if it receives an RST, it is 1388 * allowed for further processing. 1389 */ 1390 if ((thflags & TH_RST) == 0 || 1391 (tp->t_state == TCPS_SYN_SENT) == 0) 1392 goto dropunlock; 1393 } 1394 sig_checked = 1; 1395 } 1396 #endif 1397 1398 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1399 1400 /* 1401 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1402 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1403 * the inpcb, and unlocks pcbinfo. 1404 */ 1405 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1406 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1407 return; 1408 1409 dropwithreset: 1410 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1411 1412 if (ti_locked == TI_WLOCKED) { 1413 INP_INFO_WUNLOCK(&V_tcbinfo); 1414 ti_locked = TI_UNLOCKED; 1415 } 1416 #ifdef INVARIANTS 1417 else { 1418 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1419 "ti_locked: %d", __func__, ti_locked)); 1420 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1421 } 1422 #endif 1423 1424 if (inp != NULL) { 1425 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1426 INP_WUNLOCK(inp); 1427 } else 1428 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1429 m = NULL; /* mbuf chain got consumed. */ 1430 goto drop; 1431 1432 dropunlock: 1433 if (m != NULL) 1434 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1435 1436 if (ti_locked == TI_WLOCKED) { 1437 INP_INFO_WUNLOCK(&V_tcbinfo); 1438 ti_locked = TI_UNLOCKED; 1439 } 1440 #ifdef INVARIANTS 1441 else { 1442 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1443 "ti_locked: %d", __func__, ti_locked)); 1444 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1445 } 1446 #endif 1447 1448 if (inp != NULL) 1449 INP_WUNLOCK(inp); 1450 1451 drop: 1452 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1453 if (s != NULL) 1454 free(s, M_TCPLOG); 1455 if (m != NULL) 1456 m_freem(m); 1457 } 1458 1459 static void 1460 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1461 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1462 int ti_locked) 1463 { 1464 int thflags, acked, ourfinisacked, needoutput = 0; 1465 int rstreason, todrop, win; 1466 u_long tiwin; 1467 char *s; 1468 struct in_conninfo *inc; 1469 struct mbuf *mfree; 1470 struct tcpopt to; 1471 1472 #ifdef TCPDEBUG 1473 /* 1474 * The size of tcp_saveipgen must be the size of the max ip header, 1475 * now IPv6. 1476 */ 1477 u_char tcp_saveipgen[IP6_HDR_LEN]; 1478 struct tcphdr tcp_savetcp; 1479 short ostate = 0; 1480 #endif 1481 thflags = th->th_flags; 1482 inc = &tp->t_inpcb->inp_inc; 1483 tp->sackhint.last_sack_ack = 0; 1484 1485 /* 1486 * If this is either a state-changing packet or current state isn't 1487 * established, we require a write lock on tcbinfo. Otherwise, we 1488 * allow the tcbinfo to be in either alocked or unlocked, as the 1489 * caller may have unnecessarily acquired a write lock due to a race. 1490 */ 1491 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1492 tp->t_state != TCPS_ESTABLISHED) { 1493 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1494 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1495 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1496 } else { 1497 #ifdef INVARIANTS 1498 if (ti_locked == TI_WLOCKED) 1499 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1500 else { 1501 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1502 "ti_locked: %d", __func__, ti_locked)); 1503 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1504 } 1505 #endif 1506 } 1507 INP_WLOCK_ASSERT(tp->t_inpcb); 1508 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1509 __func__)); 1510 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1511 __func__)); 1512 1513 /* 1514 * Segment received on connection. 1515 * Reset idle time and keep-alive timer. 1516 * XXX: This should be done after segment 1517 * validation to ignore broken/spoofed segs. 1518 */ 1519 tp->t_rcvtime = ticks; 1520 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1521 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1522 1523 /* 1524 * Unscale the window into a 32-bit value. 1525 * For the SYN_SENT state the scale is zero. 1526 */ 1527 tiwin = th->th_win << tp->snd_scale; 1528 1529 /* 1530 * TCP ECN processing. 1531 */ 1532 if (tp->t_flags & TF_ECN_PERMIT) { 1533 if (thflags & TH_CWR) 1534 tp->t_flags &= ~TF_ECN_SND_ECE; 1535 switch (iptos & IPTOS_ECN_MASK) { 1536 case IPTOS_ECN_CE: 1537 tp->t_flags |= TF_ECN_SND_ECE; 1538 TCPSTAT_INC(tcps_ecn_ce); 1539 break; 1540 case IPTOS_ECN_ECT0: 1541 TCPSTAT_INC(tcps_ecn_ect0); 1542 break; 1543 case IPTOS_ECN_ECT1: 1544 TCPSTAT_INC(tcps_ecn_ect1); 1545 break; 1546 } 1547 /* Congestion experienced. */ 1548 if (thflags & TH_ECE) { 1549 cc_cong_signal(tp, th, CC_ECN); 1550 } 1551 } 1552 1553 /* 1554 * Parse options on any incoming segment. 1555 */ 1556 tcp_dooptions(&to, (u_char *)(th + 1), 1557 (th->th_off << 2) - sizeof(struct tcphdr), 1558 (thflags & TH_SYN) ? TO_SYN : 0); 1559 1560 /* 1561 * If echoed timestamp is later than the current time, 1562 * fall back to non RFC1323 RTT calculation. Normalize 1563 * timestamp if syncookies were used when this connection 1564 * was established. 1565 */ 1566 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1567 to.to_tsecr -= tp->ts_offset; 1568 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1569 to.to_tsecr = 0; 1570 } 1571 /* 1572 * If timestamps were negotiated during SYN/ACK they should 1573 * appear on every segment during this session and vice versa. 1574 */ 1575 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1576 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1577 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1578 "no action\n", s, __func__); 1579 free(s, M_TCPLOG); 1580 } 1581 } 1582 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1583 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1584 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1585 "no action\n", s, __func__); 1586 free(s, M_TCPLOG); 1587 } 1588 } 1589 1590 /* 1591 * Process options only when we get SYN/ACK back. The SYN case 1592 * for incoming connections is handled in tcp_syncache. 1593 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1594 * or <SYN,ACK>) segment itself is never scaled. 1595 * XXX this is traditional behavior, may need to be cleaned up. 1596 */ 1597 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1598 if ((to.to_flags & TOF_SCALE) && 1599 (tp->t_flags & TF_REQ_SCALE)) { 1600 tp->t_flags |= TF_RCVD_SCALE; 1601 tp->snd_scale = to.to_wscale; 1602 } 1603 /* 1604 * Initial send window. It will be updated with 1605 * the next incoming segment to the scaled value. 1606 */ 1607 tp->snd_wnd = th->th_win; 1608 if (to.to_flags & TOF_TS) { 1609 tp->t_flags |= TF_RCVD_TSTMP; 1610 tp->ts_recent = to.to_tsval; 1611 tp->ts_recent_age = tcp_ts_getticks(); 1612 } 1613 if (to.to_flags & TOF_MSS) 1614 tcp_mss(tp, to.to_mss); 1615 if ((tp->t_flags & TF_SACK_PERMIT) && 1616 (to.to_flags & TOF_SACKPERM) == 0) 1617 tp->t_flags &= ~TF_SACK_PERMIT; 1618 } 1619 1620 /* 1621 * Header prediction: check for the two common cases 1622 * of a uni-directional data xfer. If the packet has 1623 * no control flags, is in-sequence, the window didn't 1624 * change and we're not retransmitting, it's a 1625 * candidate. If the length is zero and the ack moved 1626 * forward, we're the sender side of the xfer. Just 1627 * free the data acked & wake any higher level process 1628 * that was blocked waiting for space. If the length 1629 * is non-zero and the ack didn't move, we're the 1630 * receiver side. If we're getting packets in-order 1631 * (the reassembly queue is empty), add the data to 1632 * the socket buffer and note that we need a delayed ack. 1633 * Make sure that the hidden state-flags are also off. 1634 * Since we check for TCPS_ESTABLISHED first, it can only 1635 * be TH_NEEDSYN. 1636 */ 1637 if (tp->t_state == TCPS_ESTABLISHED && 1638 th->th_seq == tp->rcv_nxt && 1639 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1640 tp->snd_nxt == tp->snd_max && 1641 tiwin && tiwin == tp->snd_wnd && 1642 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1643 tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 || 1644 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1645 1646 /* 1647 * If last ACK falls within this segment's sequence numbers, 1648 * record the timestamp. 1649 * NOTE that the test is modified according to the latest 1650 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1651 */ 1652 if ((to.to_flags & TOF_TS) != 0 && 1653 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1654 tp->ts_recent_age = tcp_ts_getticks(); 1655 tp->ts_recent = to.to_tsval; 1656 } 1657 1658 if (tlen == 0) { 1659 if (SEQ_GT(th->th_ack, tp->snd_una) && 1660 SEQ_LEQ(th->th_ack, tp->snd_max) && 1661 !IN_RECOVERY(tp->t_flags) && 1662 (to.to_flags & TOF_SACK) == 0 && 1663 TAILQ_EMPTY(&tp->snd_holes)) { 1664 /* 1665 * This is a pure ack for outstanding data. 1666 */ 1667 if (ti_locked == TI_WLOCKED) 1668 INP_INFO_WUNLOCK(&V_tcbinfo); 1669 ti_locked = TI_UNLOCKED; 1670 1671 TCPSTAT_INC(tcps_predack); 1672 1673 /* 1674 * "bad retransmit" recovery. 1675 */ 1676 if (tp->t_rxtshift == 1 && 1677 tp->t_flags & TF_PREVVALID && 1678 (int)(ticks - tp->t_badrxtwin) < 0) { 1679 cc_cong_signal(tp, th, CC_RTO_ERR); 1680 } 1681 1682 /* 1683 * Recalculate the transmit timer / rtt. 1684 * 1685 * Some boxes send broken timestamp replies 1686 * during the SYN+ACK phase, ignore 1687 * timestamps of 0 or we could calculate a 1688 * huge RTT and blow up the retransmit timer. 1689 */ 1690 if ((to.to_flags & TOF_TS) != 0 && 1691 to.to_tsecr) { 1692 u_int t; 1693 1694 t = tcp_ts_getticks() - to.to_tsecr; 1695 if (!tp->t_rttlow || tp->t_rttlow > t) 1696 tp->t_rttlow = t; 1697 tcp_xmit_timer(tp, 1698 TCP_TS_TO_TICKS(t) + 1); 1699 } else if (tp->t_rtttime && 1700 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1701 if (!tp->t_rttlow || 1702 tp->t_rttlow > ticks - tp->t_rtttime) 1703 tp->t_rttlow = ticks - tp->t_rtttime; 1704 tcp_xmit_timer(tp, 1705 ticks - tp->t_rtttime); 1706 } 1707 acked = BYTES_THIS_ACK(tp, th); 1708 1709 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1710 hhook_run_tcp_est_in(tp, th, &to); 1711 1712 TCPSTAT_INC(tcps_rcvackpack); 1713 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1714 sbdrop(&so->so_snd, acked); 1715 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1716 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1717 tp->snd_recover = th->th_ack - 1; 1718 1719 /* 1720 * Let the congestion control algorithm update 1721 * congestion control related information. This 1722 * typically means increasing the congestion 1723 * window. 1724 */ 1725 cc_ack_received(tp, th, CC_ACK); 1726 1727 tp->snd_una = th->th_ack; 1728 /* 1729 * Pull snd_wl2 up to prevent seq wrap relative 1730 * to th_ack. 1731 */ 1732 tp->snd_wl2 = th->th_ack; 1733 tp->t_dupacks = 0; 1734 m_freem(m); 1735 ND6_HINT(tp); /* Some progress has been made. */ 1736 1737 /* 1738 * If all outstanding data are acked, stop 1739 * retransmit timer, otherwise restart timer 1740 * using current (possibly backed-off) value. 1741 * If process is waiting for space, 1742 * wakeup/selwakeup/signal. If data 1743 * are ready to send, let tcp_output 1744 * decide between more output or persist. 1745 */ 1746 #ifdef TCPDEBUG 1747 if (so->so_options & SO_DEBUG) 1748 tcp_trace(TA_INPUT, ostate, tp, 1749 (void *)tcp_saveipgen, 1750 &tcp_savetcp, 0); 1751 #endif 1752 if (tp->snd_una == tp->snd_max) 1753 tcp_timer_activate(tp, TT_REXMT, 0); 1754 else if (!tcp_timer_active(tp, TT_PERSIST)) 1755 tcp_timer_activate(tp, TT_REXMT, 1756 tp->t_rxtcur); 1757 sowwakeup(so); 1758 if (so->so_snd.sb_cc) 1759 (void) tcp_output(tp); 1760 goto check_delack; 1761 } 1762 } else if (th->th_ack == tp->snd_una && 1763 tlen <= sbspace(&so->so_rcv)) { 1764 int newsize = 0; /* automatic sockbuf scaling */ 1765 1766 /* 1767 * This is a pure, in-sequence data packet with 1768 * nothing on the reassembly queue and we have enough 1769 * buffer space to take it. 1770 */ 1771 if (ti_locked == TI_WLOCKED) 1772 INP_INFO_WUNLOCK(&V_tcbinfo); 1773 ti_locked = TI_UNLOCKED; 1774 1775 /* Clean receiver SACK report if present */ 1776 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1777 tcp_clean_sackreport(tp); 1778 TCPSTAT_INC(tcps_preddat); 1779 tp->rcv_nxt += tlen; 1780 /* 1781 * Pull snd_wl1 up to prevent seq wrap relative to 1782 * th_seq. 1783 */ 1784 tp->snd_wl1 = th->th_seq; 1785 /* 1786 * Pull rcv_up up to prevent seq wrap relative to 1787 * rcv_nxt. 1788 */ 1789 tp->rcv_up = tp->rcv_nxt; 1790 TCPSTAT_INC(tcps_rcvpack); 1791 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1792 ND6_HINT(tp); /* Some progress has been made */ 1793 #ifdef TCPDEBUG 1794 if (so->so_options & SO_DEBUG) 1795 tcp_trace(TA_INPUT, ostate, tp, 1796 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1797 #endif 1798 /* 1799 * Automatic sizing of receive socket buffer. Often the send 1800 * buffer size is not optimally adjusted to the actual network 1801 * conditions at hand (delay bandwidth product). Setting the 1802 * buffer size too small limits throughput on links with high 1803 * bandwidth and high delay (eg. trans-continental/oceanic links). 1804 * 1805 * On the receive side the socket buffer memory is only rarely 1806 * used to any significant extent. This allows us to be much 1807 * more aggressive in scaling the receive socket buffer. For 1808 * the case that the buffer space is actually used to a large 1809 * extent and we run out of kernel memory we can simply drop 1810 * the new segments; TCP on the sender will just retransmit it 1811 * later. Setting the buffer size too big may only consume too 1812 * much kernel memory if the application doesn't read() from 1813 * the socket or packet loss or reordering makes use of the 1814 * reassembly queue. 1815 * 1816 * The criteria to step up the receive buffer one notch are: 1817 * 1. the number of bytes received during the time it takes 1818 * one timestamp to be reflected back to us (the RTT); 1819 * 2. received bytes per RTT is within seven eighth of the 1820 * current socket buffer size; 1821 * 3. receive buffer size has not hit maximal automatic size; 1822 * 1823 * This algorithm does one step per RTT at most and only if 1824 * we receive a bulk stream w/o packet losses or reorderings. 1825 * Shrinking the buffer during idle times is not necessary as 1826 * it doesn't consume any memory when idle. 1827 * 1828 * TODO: Only step up if the application is actually serving 1829 * the buffer to better manage the socket buffer resources. 1830 */ 1831 if (V_tcp_do_autorcvbuf && 1832 to.to_tsecr && 1833 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1834 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1835 to.to_tsecr - tp->rfbuf_ts < hz) { 1836 if (tp->rfbuf_cnt > 1837 (so->so_rcv.sb_hiwat / 8 * 7) && 1838 so->so_rcv.sb_hiwat < 1839 V_tcp_autorcvbuf_max) { 1840 newsize = 1841 min(so->so_rcv.sb_hiwat + 1842 V_tcp_autorcvbuf_inc, 1843 V_tcp_autorcvbuf_max); 1844 } 1845 /* Start over with next RTT. */ 1846 tp->rfbuf_ts = 0; 1847 tp->rfbuf_cnt = 0; 1848 } else 1849 tp->rfbuf_cnt += tlen; /* add up */ 1850 } 1851 1852 /* Add data to socket buffer. */ 1853 SOCKBUF_LOCK(&so->so_rcv); 1854 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1855 m_freem(m); 1856 } else { 1857 /* 1858 * Set new socket buffer size. 1859 * Give up when limit is reached. 1860 */ 1861 if (newsize) 1862 if (!sbreserve_locked(&so->so_rcv, 1863 newsize, so, NULL)) 1864 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1865 m_adj(m, drop_hdrlen); /* delayed header drop */ 1866 sbappendstream_locked(&so->so_rcv, m); 1867 } 1868 /* NB: sorwakeup_locked() does an implicit unlock. */ 1869 sorwakeup_locked(so); 1870 if (DELAY_ACK(tp, tlen)) { 1871 tp->t_flags |= TF_DELACK; 1872 } else { 1873 tp->t_flags |= TF_ACKNOW; 1874 tcp_output(tp); 1875 } 1876 goto check_delack; 1877 } 1878 } 1879 1880 /* 1881 * Calculate amount of space in receive window, 1882 * and then do TCP input processing. 1883 * Receive window is amount of space in rcv queue, 1884 * but not less than advertised window. 1885 */ 1886 win = sbspace(&so->so_rcv); 1887 if (win < 0) 1888 win = 0; 1889 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1890 1891 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1892 tp->rfbuf_ts = 0; 1893 tp->rfbuf_cnt = 0; 1894 1895 switch (tp->t_state) { 1896 1897 /* 1898 * If the state is SYN_RECEIVED: 1899 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1900 */ 1901 case TCPS_SYN_RECEIVED: 1902 if ((thflags & TH_ACK) && 1903 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1904 SEQ_GT(th->th_ack, tp->snd_max))) { 1905 rstreason = BANDLIM_RST_OPENPORT; 1906 goto dropwithreset; 1907 } 1908 break; 1909 1910 /* 1911 * If the state is SYN_SENT: 1912 * if seg contains an ACK, but not for our SYN, drop the input. 1913 * if seg contains a RST, then drop the connection. 1914 * if seg does not contain SYN, then drop it. 1915 * Otherwise this is an acceptable SYN segment 1916 * initialize tp->rcv_nxt and tp->irs 1917 * if seg contains ack then advance tp->snd_una 1918 * if seg contains an ECE and ECN support is enabled, the stream 1919 * is ECN capable. 1920 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1921 * arrange for segment to be acked (eventually) 1922 * continue processing rest of data/controls, beginning with URG 1923 */ 1924 case TCPS_SYN_SENT: 1925 if ((thflags & TH_ACK) && 1926 (SEQ_LEQ(th->th_ack, tp->iss) || 1927 SEQ_GT(th->th_ack, tp->snd_max))) { 1928 rstreason = BANDLIM_UNLIMITED; 1929 goto dropwithreset; 1930 } 1931 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1932 TCP_PROBE5(connect__refused, NULL, tp, 1933 mtod(m, const char *), tp, th); 1934 tp = tcp_drop(tp, ECONNREFUSED); 1935 } 1936 if (thflags & TH_RST) 1937 goto drop; 1938 if (!(thflags & TH_SYN)) 1939 goto drop; 1940 1941 tp->irs = th->th_seq; 1942 tcp_rcvseqinit(tp); 1943 if (thflags & TH_ACK) { 1944 TCPSTAT_INC(tcps_connects); 1945 soisconnected(so); 1946 #ifdef MAC 1947 mac_socketpeer_set_from_mbuf(m, so); 1948 #endif 1949 /* Do window scaling on this connection? */ 1950 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1951 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1952 tp->rcv_scale = tp->request_r_scale; 1953 } 1954 tp->rcv_adv += imin(tp->rcv_wnd, 1955 TCP_MAXWIN << tp->rcv_scale); 1956 tp->snd_una++; /* SYN is acked */ 1957 /* 1958 * If there's data, delay ACK; if there's also a FIN 1959 * ACKNOW will be turned on later. 1960 */ 1961 if (DELAY_ACK(tp, tlen) && tlen != 0) 1962 tcp_timer_activate(tp, TT_DELACK, 1963 tcp_delacktime); 1964 else 1965 tp->t_flags |= TF_ACKNOW; 1966 1967 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1968 tp->t_flags |= TF_ECN_PERMIT; 1969 TCPSTAT_INC(tcps_ecn_shs); 1970 } 1971 1972 /* 1973 * Received <SYN,ACK> in SYN_SENT[*] state. 1974 * Transitions: 1975 * SYN_SENT --> ESTABLISHED 1976 * SYN_SENT* --> FIN_WAIT_1 1977 */ 1978 tp->t_starttime = ticks; 1979 if (tp->t_flags & TF_NEEDFIN) { 1980 tcp_state_change(tp, TCPS_FIN_WAIT_1); 1981 tp->t_flags &= ~TF_NEEDFIN; 1982 thflags &= ~TH_SYN; 1983 } else { 1984 tcp_state_change(tp, TCPS_ESTABLISHED); 1985 TCP_PROBE5(connect__established, NULL, tp, 1986 mtod(m, const char *), tp, th); 1987 cc_conn_init(tp); 1988 tcp_timer_activate(tp, TT_KEEP, 1989 TP_KEEPIDLE(tp)); 1990 } 1991 } else { 1992 /* 1993 * Received initial SYN in SYN-SENT[*] state => 1994 * simultaneous open. 1995 * If it succeeds, connection is * half-synchronized. 1996 * Otherwise, do 3-way handshake: 1997 * SYN-SENT -> SYN-RECEIVED 1998 * SYN-SENT* -> SYN-RECEIVED* 1999 */ 2000 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2001 tcp_timer_activate(tp, TT_REXMT, 0); 2002 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2003 } 2004 2005 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 2006 "ti_locked %d", __func__, ti_locked)); 2007 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2008 INP_WLOCK_ASSERT(tp->t_inpcb); 2009 2010 /* 2011 * Advance th->th_seq to correspond to first data byte. 2012 * If data, trim to stay within window, 2013 * dropping FIN if necessary. 2014 */ 2015 th->th_seq++; 2016 if (tlen > tp->rcv_wnd) { 2017 todrop = tlen - tp->rcv_wnd; 2018 m_adj(m, -todrop); 2019 tlen = tp->rcv_wnd; 2020 thflags &= ~TH_FIN; 2021 TCPSTAT_INC(tcps_rcvpackafterwin); 2022 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2023 } 2024 tp->snd_wl1 = th->th_seq - 1; 2025 tp->rcv_up = th->th_seq; 2026 /* 2027 * Client side of transaction: already sent SYN and data. 2028 * If the remote host used T/TCP to validate the SYN, 2029 * our data will be ACK'd; if so, enter normal data segment 2030 * processing in the middle of step 5, ack processing. 2031 * Otherwise, goto step 6. 2032 */ 2033 if (thflags & TH_ACK) 2034 goto process_ACK; 2035 2036 goto step6; 2037 2038 /* 2039 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2040 * do normal processing. 2041 * 2042 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2043 */ 2044 case TCPS_LAST_ACK: 2045 case TCPS_CLOSING: 2046 break; /* continue normal processing */ 2047 } 2048 2049 /* 2050 * States other than LISTEN or SYN_SENT. 2051 * First check the RST flag and sequence number since reset segments 2052 * are exempt from the timestamp and connection count tests. This 2053 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2054 * below which allowed reset segments in half the sequence space 2055 * to fall though and be processed (which gives forged reset 2056 * segments with a random sequence number a 50 percent chance of 2057 * killing a connection). 2058 * Then check timestamp, if present. 2059 * Then check the connection count, if present. 2060 * Then check that at least some bytes of segment are within 2061 * receive window. If segment begins before rcv_nxt, 2062 * drop leading data (and SYN); if nothing left, just ack. 2063 * 2064 * 2065 * If the RST bit is set, check the sequence number to see 2066 * if this is a valid reset segment. 2067 * RFC 793 page 37: 2068 * In all states except SYN-SENT, all reset (RST) segments 2069 * are validated by checking their SEQ-fields. A reset is 2070 * valid if its sequence number is in the window. 2071 * Note: this does not take into account delayed ACKs, so 2072 * we should test against last_ack_sent instead of rcv_nxt. 2073 * The sequence number in the reset segment is normally an 2074 * echo of our outgoing acknowlegement numbers, but some hosts 2075 * send a reset with the sequence number at the rightmost edge 2076 * of our receive window, and we have to handle this case. 2077 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2078 * that brute force RST attacks are possible. To combat this, 2079 * we use a much stricter check while in the ESTABLISHED state, 2080 * only accepting RSTs where the sequence number is equal to 2081 * last_ack_sent. In all other states (the states in which a 2082 * RST is more likely), the more permissive check is used. 2083 * If we have multiple segments in flight, the initial reset 2084 * segment sequence numbers will be to the left of last_ack_sent, 2085 * but they will eventually catch up. 2086 * In any case, it never made sense to trim reset segments to 2087 * fit the receive window since RFC 1122 says: 2088 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2089 * 2090 * A TCP SHOULD allow a received RST segment to include data. 2091 * 2092 * DISCUSSION 2093 * It has been suggested that a RST segment could contain 2094 * ASCII text that encoded and explained the cause of the 2095 * RST. No standard has yet been established for such 2096 * data. 2097 * 2098 * If the reset segment passes the sequence number test examine 2099 * the state: 2100 * SYN_RECEIVED STATE: 2101 * If passive open, return to LISTEN state. 2102 * If active open, inform user that connection was refused. 2103 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2104 * Inform user that connection was reset, and close tcb. 2105 * CLOSING, LAST_ACK STATES: 2106 * Close the tcb. 2107 * TIME_WAIT STATE: 2108 * Drop the segment - see Stevens, vol. 2, p. 964 and 2109 * RFC 1337. 2110 */ 2111 if (thflags & TH_RST) { 2112 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2113 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2114 switch (tp->t_state) { 2115 2116 case TCPS_SYN_RECEIVED: 2117 so->so_error = ECONNREFUSED; 2118 goto close; 2119 2120 case TCPS_ESTABLISHED: 2121 if (V_tcp_insecure_rst == 0 && 2122 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2123 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2124 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2125 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2126 TCPSTAT_INC(tcps_badrst); 2127 goto drop; 2128 } 2129 /* FALLTHROUGH */ 2130 case TCPS_FIN_WAIT_1: 2131 case TCPS_FIN_WAIT_2: 2132 case TCPS_CLOSE_WAIT: 2133 so->so_error = ECONNRESET; 2134 close: 2135 KASSERT(ti_locked == TI_WLOCKED, 2136 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2137 ti_locked)); 2138 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2139 2140 tcp_state_change(tp, TCPS_CLOSED); 2141 TCPSTAT_INC(tcps_drops); 2142 tp = tcp_close(tp); 2143 break; 2144 2145 case TCPS_CLOSING: 2146 case TCPS_LAST_ACK: 2147 KASSERT(ti_locked == TI_WLOCKED, 2148 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2149 ti_locked)); 2150 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2151 2152 tp = tcp_close(tp); 2153 break; 2154 } 2155 } 2156 goto drop; 2157 } 2158 2159 /* 2160 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2161 * and it's less than ts_recent, drop it. 2162 */ 2163 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2164 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2165 2166 /* Check to see if ts_recent is over 24 days old. */ 2167 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2168 /* 2169 * Invalidate ts_recent. If this segment updates 2170 * ts_recent, the age will be reset later and ts_recent 2171 * will get a valid value. If it does not, setting 2172 * ts_recent to zero will at least satisfy the 2173 * requirement that zero be placed in the timestamp 2174 * echo reply when ts_recent isn't valid. The 2175 * age isn't reset until we get a valid ts_recent 2176 * because we don't want out-of-order segments to be 2177 * dropped when ts_recent is old. 2178 */ 2179 tp->ts_recent = 0; 2180 } else { 2181 TCPSTAT_INC(tcps_rcvduppack); 2182 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2183 TCPSTAT_INC(tcps_pawsdrop); 2184 if (tlen) 2185 goto dropafterack; 2186 goto drop; 2187 } 2188 } 2189 2190 /* 2191 * In the SYN-RECEIVED state, validate that the packet belongs to 2192 * this connection before trimming the data to fit the receive 2193 * window. Check the sequence number versus IRS since we know 2194 * the sequence numbers haven't wrapped. This is a partial fix 2195 * for the "LAND" DoS attack. 2196 */ 2197 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2198 rstreason = BANDLIM_RST_OPENPORT; 2199 goto dropwithreset; 2200 } 2201 2202 todrop = tp->rcv_nxt - th->th_seq; 2203 if (todrop > 0) { 2204 /* 2205 * If this is a duplicate SYN for our current connection, 2206 * advance over it and pretend and it's not a SYN. 2207 */ 2208 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2209 thflags &= ~TH_SYN; 2210 th->th_seq++; 2211 if (th->th_urp > 1) 2212 th->th_urp--; 2213 else 2214 thflags &= ~TH_URG; 2215 todrop--; 2216 } 2217 /* 2218 * Following if statement from Stevens, vol. 2, p. 960. 2219 */ 2220 if (todrop > tlen 2221 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2222 /* 2223 * Any valid FIN must be to the left of the window. 2224 * At this point the FIN must be a duplicate or out 2225 * of sequence; drop it. 2226 */ 2227 thflags &= ~TH_FIN; 2228 2229 /* 2230 * Send an ACK to resynchronize and drop any data. 2231 * But keep on processing for RST or ACK. 2232 */ 2233 tp->t_flags |= TF_ACKNOW; 2234 todrop = tlen; 2235 TCPSTAT_INC(tcps_rcvduppack); 2236 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2237 } else { 2238 TCPSTAT_INC(tcps_rcvpartduppack); 2239 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2240 } 2241 drop_hdrlen += todrop; /* drop from the top afterwards */ 2242 th->th_seq += todrop; 2243 tlen -= todrop; 2244 if (th->th_urp > todrop) 2245 th->th_urp -= todrop; 2246 else { 2247 thflags &= ~TH_URG; 2248 th->th_urp = 0; 2249 } 2250 } 2251 2252 /* 2253 * If new data are received on a connection after the 2254 * user processes are gone, then RST the other end. 2255 */ 2256 if ((so->so_state & SS_NOFDREF) && 2257 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2258 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2259 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2260 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2261 2262 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2263 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2264 "after socket was closed, " 2265 "sending RST and removing tcpcb\n", 2266 s, __func__, tcpstates[tp->t_state], tlen); 2267 free(s, M_TCPLOG); 2268 } 2269 tp = tcp_close(tp); 2270 TCPSTAT_INC(tcps_rcvafterclose); 2271 rstreason = BANDLIM_UNLIMITED; 2272 goto dropwithreset; 2273 } 2274 2275 /* 2276 * If segment ends after window, drop trailing data 2277 * (and PUSH and FIN); if nothing left, just ACK. 2278 */ 2279 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2280 if (todrop > 0) { 2281 TCPSTAT_INC(tcps_rcvpackafterwin); 2282 if (todrop >= tlen) { 2283 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2284 /* 2285 * If window is closed can only take segments at 2286 * window edge, and have to drop data and PUSH from 2287 * incoming segments. Continue processing, but 2288 * remember to ack. Otherwise, drop segment 2289 * and ack. 2290 */ 2291 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2292 tp->t_flags |= TF_ACKNOW; 2293 TCPSTAT_INC(tcps_rcvwinprobe); 2294 } else 2295 goto dropafterack; 2296 } else 2297 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2298 m_adj(m, -todrop); 2299 tlen -= todrop; 2300 thflags &= ~(TH_PUSH|TH_FIN); 2301 } 2302 2303 /* 2304 * If last ACK falls within this segment's sequence numbers, 2305 * record its timestamp. 2306 * NOTE: 2307 * 1) That the test incorporates suggestions from the latest 2308 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2309 * 2) That updating only on newer timestamps interferes with 2310 * our earlier PAWS tests, so this check should be solely 2311 * predicated on the sequence space of this segment. 2312 * 3) That we modify the segment boundary check to be 2313 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2314 * instead of RFC1323's 2315 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2316 * This modified check allows us to overcome RFC1323's 2317 * limitations as described in Stevens TCP/IP Illustrated 2318 * Vol. 2 p.869. In such cases, we can still calculate the 2319 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2320 */ 2321 if ((to.to_flags & TOF_TS) != 0 && 2322 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2323 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2324 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2325 tp->ts_recent_age = tcp_ts_getticks(); 2326 tp->ts_recent = to.to_tsval; 2327 } 2328 2329 /* 2330 * If a SYN is in the window, then this is an 2331 * error and we send an RST and drop the connection. 2332 */ 2333 if (thflags & TH_SYN) { 2334 KASSERT(ti_locked == TI_WLOCKED, 2335 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2336 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2337 2338 tp = tcp_drop(tp, ECONNRESET); 2339 rstreason = BANDLIM_UNLIMITED; 2340 goto drop; 2341 } 2342 2343 /* 2344 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2345 * flag is on (half-synchronized state), then queue data for 2346 * later processing; else drop segment and return. 2347 */ 2348 if ((thflags & TH_ACK) == 0) { 2349 if (tp->t_state == TCPS_SYN_RECEIVED || 2350 (tp->t_flags & TF_NEEDSYN)) 2351 goto step6; 2352 else if (tp->t_flags & TF_ACKNOW) 2353 goto dropafterack; 2354 else 2355 goto drop; 2356 } 2357 2358 /* 2359 * Ack processing. 2360 */ 2361 switch (tp->t_state) { 2362 2363 /* 2364 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2365 * ESTABLISHED state and continue processing. 2366 * The ACK was checked above. 2367 */ 2368 case TCPS_SYN_RECEIVED: 2369 2370 TCPSTAT_INC(tcps_connects); 2371 soisconnected(so); 2372 /* Do window scaling? */ 2373 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2374 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2375 tp->rcv_scale = tp->request_r_scale; 2376 tp->snd_wnd = tiwin; 2377 } 2378 /* 2379 * Make transitions: 2380 * SYN-RECEIVED -> ESTABLISHED 2381 * SYN-RECEIVED* -> FIN-WAIT-1 2382 */ 2383 tp->t_starttime = ticks; 2384 if (tp->t_flags & TF_NEEDFIN) { 2385 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2386 tp->t_flags &= ~TF_NEEDFIN; 2387 } else { 2388 tcp_state_change(tp, TCPS_ESTABLISHED); 2389 TCP_PROBE5(accept__established, NULL, tp, 2390 mtod(m, const char *), tp, th); 2391 cc_conn_init(tp); 2392 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2393 } 2394 /* 2395 * If segment contains data or ACK, will call tcp_reass() 2396 * later; if not, do so now to pass queued data to user. 2397 */ 2398 if (tlen == 0 && (thflags & TH_FIN) == 0) 2399 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2400 (struct mbuf *)0); 2401 tp->snd_wl1 = th->th_seq - 1; 2402 /* FALLTHROUGH */ 2403 2404 /* 2405 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2406 * ACKs. If the ack is in the range 2407 * tp->snd_una < th->th_ack <= tp->snd_max 2408 * then advance tp->snd_una to th->th_ack and drop 2409 * data from the retransmission queue. If this ACK reflects 2410 * more up to date window information we update our window information. 2411 */ 2412 case TCPS_ESTABLISHED: 2413 case TCPS_FIN_WAIT_1: 2414 case TCPS_FIN_WAIT_2: 2415 case TCPS_CLOSE_WAIT: 2416 case TCPS_CLOSING: 2417 case TCPS_LAST_ACK: 2418 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2419 TCPSTAT_INC(tcps_rcvacktoomuch); 2420 goto dropafterack; 2421 } 2422 if ((tp->t_flags & TF_SACK_PERMIT) && 2423 ((to.to_flags & TOF_SACK) || 2424 !TAILQ_EMPTY(&tp->snd_holes))) 2425 tcp_sack_doack(tp, &to, th->th_ack); 2426 2427 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2428 hhook_run_tcp_est_in(tp, th, &to); 2429 2430 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2431 if (tlen == 0 && tiwin == tp->snd_wnd) { 2432 /* 2433 * If this is the first time we've seen a 2434 * FIN from the remote, this is not a 2435 * duplicate and it needs to be processed 2436 * normally. This happens during a 2437 * simultaneous close. 2438 */ 2439 if ((thflags & TH_FIN) && 2440 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2441 tp->t_dupacks = 0; 2442 break; 2443 } 2444 TCPSTAT_INC(tcps_rcvdupack); 2445 /* 2446 * If we have outstanding data (other than 2447 * a window probe), this is a completely 2448 * duplicate ack (ie, window info didn't 2449 * change and FIN isn't set), 2450 * the ack is the biggest we've 2451 * seen and we've seen exactly our rexmt 2452 * threshhold of them, assume a packet 2453 * has been dropped and retransmit it. 2454 * Kludge snd_nxt & the congestion 2455 * window so we send only this one 2456 * packet. 2457 * 2458 * We know we're losing at the current 2459 * window size so do congestion avoidance 2460 * (set ssthresh to half the current window 2461 * and pull our congestion window back to 2462 * the new ssthresh). 2463 * 2464 * Dup acks mean that packets have left the 2465 * network (they're now cached at the receiver) 2466 * so bump cwnd by the amount in the receiver 2467 * to keep a constant cwnd packets in the 2468 * network. 2469 * 2470 * When using TCP ECN, notify the peer that 2471 * we reduced the cwnd. 2472 */ 2473 if (!tcp_timer_active(tp, TT_REXMT) || 2474 th->th_ack != tp->snd_una) 2475 tp->t_dupacks = 0; 2476 else if (++tp->t_dupacks > tcprexmtthresh || 2477 IN_FASTRECOVERY(tp->t_flags)) { 2478 cc_ack_received(tp, th, CC_DUPACK); 2479 if ((tp->t_flags & TF_SACK_PERMIT) && 2480 IN_FASTRECOVERY(tp->t_flags)) { 2481 int awnd; 2482 2483 /* 2484 * Compute the amount of data in flight first. 2485 * We can inject new data into the pipe iff 2486 * we have less than 1/2 the original window's 2487 * worth of data in flight. 2488 */ 2489 awnd = (tp->snd_nxt - tp->snd_fack) + 2490 tp->sackhint.sack_bytes_rexmit; 2491 if (awnd < tp->snd_ssthresh) { 2492 tp->snd_cwnd += tp->t_maxseg; 2493 if (tp->snd_cwnd > tp->snd_ssthresh) 2494 tp->snd_cwnd = tp->snd_ssthresh; 2495 } 2496 } else 2497 tp->snd_cwnd += tp->t_maxseg; 2498 (void) tcp_output(tp); 2499 goto drop; 2500 } else if (tp->t_dupacks == tcprexmtthresh) { 2501 tcp_seq onxt = tp->snd_nxt; 2502 2503 /* 2504 * If we're doing sack, check to 2505 * see if we're already in sack 2506 * recovery. If we're not doing sack, 2507 * check to see if we're in newreno 2508 * recovery. 2509 */ 2510 if (tp->t_flags & TF_SACK_PERMIT) { 2511 if (IN_FASTRECOVERY(tp->t_flags)) { 2512 tp->t_dupacks = 0; 2513 break; 2514 } 2515 } else { 2516 if (SEQ_LEQ(th->th_ack, 2517 tp->snd_recover)) { 2518 tp->t_dupacks = 0; 2519 break; 2520 } 2521 } 2522 /* Congestion signal before ack. */ 2523 cc_cong_signal(tp, th, CC_NDUPACK); 2524 cc_ack_received(tp, th, CC_DUPACK); 2525 tcp_timer_activate(tp, TT_REXMT, 0); 2526 tp->t_rtttime = 0; 2527 if (tp->t_flags & TF_SACK_PERMIT) { 2528 TCPSTAT_INC( 2529 tcps_sack_recovery_episode); 2530 tp->sack_newdata = tp->snd_nxt; 2531 tp->snd_cwnd = tp->t_maxseg; 2532 (void) tcp_output(tp); 2533 goto drop; 2534 } 2535 tp->snd_nxt = th->th_ack; 2536 tp->snd_cwnd = tp->t_maxseg; 2537 (void) tcp_output(tp); 2538 KASSERT(tp->snd_limited <= 2, 2539 ("%s: tp->snd_limited too big", 2540 __func__)); 2541 tp->snd_cwnd = tp->snd_ssthresh + 2542 tp->t_maxseg * 2543 (tp->t_dupacks - tp->snd_limited); 2544 if (SEQ_GT(onxt, tp->snd_nxt)) 2545 tp->snd_nxt = onxt; 2546 goto drop; 2547 } else if (V_tcp_do_rfc3042) { 2548 cc_ack_received(tp, th, CC_DUPACK); 2549 u_long oldcwnd = tp->snd_cwnd; 2550 tcp_seq oldsndmax = tp->snd_max; 2551 u_int sent; 2552 int avail; 2553 2554 KASSERT(tp->t_dupacks == 1 || 2555 tp->t_dupacks == 2, 2556 ("%s: dupacks not 1 or 2", 2557 __func__)); 2558 if (tp->t_dupacks == 1) 2559 tp->snd_limited = 0; 2560 tp->snd_cwnd = 2561 (tp->snd_nxt - tp->snd_una) + 2562 (tp->t_dupacks - tp->snd_limited) * 2563 tp->t_maxseg; 2564 /* 2565 * Only call tcp_output when there 2566 * is new data available to be sent. 2567 * Otherwise we would send pure ACKs. 2568 */ 2569 SOCKBUF_LOCK(&so->so_snd); 2570 avail = so->so_snd.sb_cc - 2571 (tp->snd_nxt - tp->snd_una); 2572 SOCKBUF_UNLOCK(&so->so_snd); 2573 if (avail > 0) 2574 (void) tcp_output(tp); 2575 sent = tp->snd_max - oldsndmax; 2576 if (sent > tp->t_maxseg) { 2577 KASSERT((tp->t_dupacks == 2 && 2578 tp->snd_limited == 0) || 2579 (sent == tp->t_maxseg + 1 && 2580 tp->t_flags & TF_SENTFIN), 2581 ("%s: sent too much", 2582 __func__)); 2583 tp->snd_limited = 2; 2584 } else if (sent > 0) 2585 ++tp->snd_limited; 2586 tp->snd_cwnd = oldcwnd; 2587 goto drop; 2588 } 2589 } else 2590 tp->t_dupacks = 0; 2591 break; 2592 } 2593 2594 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2595 ("%s: th_ack <= snd_una", __func__)); 2596 2597 /* 2598 * If the congestion window was inflated to account 2599 * for the other side's cached packets, retract it. 2600 */ 2601 if (IN_FASTRECOVERY(tp->t_flags)) { 2602 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2603 if (tp->t_flags & TF_SACK_PERMIT) 2604 tcp_sack_partialack(tp, th); 2605 else 2606 tcp_newreno_partial_ack(tp, th); 2607 } else 2608 cc_post_recovery(tp, th); 2609 } 2610 tp->t_dupacks = 0; 2611 /* 2612 * If we reach this point, ACK is not a duplicate, 2613 * i.e., it ACKs something we sent. 2614 */ 2615 if (tp->t_flags & TF_NEEDSYN) { 2616 /* 2617 * T/TCP: Connection was half-synchronized, and our 2618 * SYN has been ACK'd (so connection is now fully 2619 * synchronized). Go to non-starred state, 2620 * increment snd_una for ACK of SYN, and check if 2621 * we can do window scaling. 2622 */ 2623 tp->t_flags &= ~TF_NEEDSYN; 2624 tp->snd_una++; 2625 /* Do window scaling? */ 2626 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2627 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2628 tp->rcv_scale = tp->request_r_scale; 2629 /* Send window already scaled. */ 2630 } 2631 } 2632 2633 process_ACK: 2634 INP_WLOCK_ASSERT(tp->t_inpcb); 2635 2636 acked = BYTES_THIS_ACK(tp, th); 2637 TCPSTAT_INC(tcps_rcvackpack); 2638 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2639 2640 /* 2641 * If we just performed our first retransmit, and the ACK 2642 * arrives within our recovery window, then it was a mistake 2643 * to do the retransmit in the first place. Recover our 2644 * original cwnd and ssthresh, and proceed to transmit where 2645 * we left off. 2646 */ 2647 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2648 (int)(ticks - tp->t_badrxtwin) < 0) 2649 cc_cong_signal(tp, th, CC_RTO_ERR); 2650 2651 /* 2652 * If we have a timestamp reply, update smoothed 2653 * round trip time. If no timestamp is present but 2654 * transmit timer is running and timed sequence 2655 * number was acked, update smoothed round trip time. 2656 * Since we now have an rtt measurement, cancel the 2657 * timer backoff (cf., Phil Karn's retransmit alg.). 2658 * Recompute the initial retransmit timer. 2659 * 2660 * Some boxes send broken timestamp replies 2661 * during the SYN+ACK phase, ignore 2662 * timestamps of 0 or we could calculate a 2663 * huge RTT and blow up the retransmit timer. 2664 */ 2665 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2666 u_int t; 2667 2668 t = tcp_ts_getticks() - to.to_tsecr; 2669 if (!tp->t_rttlow || tp->t_rttlow > t) 2670 tp->t_rttlow = t; 2671 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2672 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2673 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2674 tp->t_rttlow = ticks - tp->t_rtttime; 2675 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2676 } 2677 2678 /* 2679 * If all outstanding data is acked, stop retransmit 2680 * timer and remember to restart (more output or persist). 2681 * If there is more data to be acked, restart retransmit 2682 * timer, using current (possibly backed-off) value. 2683 */ 2684 if (th->th_ack == tp->snd_max) { 2685 tcp_timer_activate(tp, TT_REXMT, 0); 2686 needoutput = 1; 2687 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2688 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2689 2690 /* 2691 * If no data (only SYN) was ACK'd, 2692 * skip rest of ACK processing. 2693 */ 2694 if (acked == 0) 2695 goto step6; 2696 2697 /* 2698 * Let the congestion control algorithm update congestion 2699 * control related information. This typically means increasing 2700 * the congestion window. 2701 */ 2702 cc_ack_received(tp, th, CC_ACK); 2703 2704 SOCKBUF_LOCK(&so->so_snd); 2705 if (acked > so->so_snd.sb_cc) { 2706 tp->snd_wnd -= so->so_snd.sb_cc; 2707 mfree = sbcut_locked(&so->so_snd, 2708 (int)so->so_snd.sb_cc); 2709 ourfinisacked = 1; 2710 } else { 2711 mfree = sbcut_locked(&so->so_snd, acked); 2712 tp->snd_wnd -= acked; 2713 ourfinisacked = 0; 2714 } 2715 /* NB: sowwakeup_locked() does an implicit unlock. */ 2716 sowwakeup_locked(so); 2717 m_freem(mfree); 2718 /* Detect una wraparound. */ 2719 if (!IN_RECOVERY(tp->t_flags) && 2720 SEQ_GT(tp->snd_una, tp->snd_recover) && 2721 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2722 tp->snd_recover = th->th_ack - 1; 2723 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2724 if (IN_RECOVERY(tp->t_flags) && 2725 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2726 EXIT_RECOVERY(tp->t_flags); 2727 } 2728 tp->snd_una = th->th_ack; 2729 if (tp->t_flags & TF_SACK_PERMIT) { 2730 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2731 tp->snd_recover = tp->snd_una; 2732 } 2733 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2734 tp->snd_nxt = tp->snd_una; 2735 2736 switch (tp->t_state) { 2737 2738 /* 2739 * In FIN_WAIT_1 STATE in addition to the processing 2740 * for the ESTABLISHED state if our FIN is now acknowledged 2741 * then enter FIN_WAIT_2. 2742 */ 2743 case TCPS_FIN_WAIT_1: 2744 if (ourfinisacked) { 2745 /* 2746 * If we can't receive any more 2747 * data, then closing user can proceed. 2748 * Starting the timer is contrary to the 2749 * specification, but if we don't get a FIN 2750 * we'll hang forever. 2751 * 2752 * XXXjl: 2753 * we should release the tp also, and use a 2754 * compressed state. 2755 */ 2756 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2757 soisdisconnected(so); 2758 tcp_timer_activate(tp, TT_2MSL, 2759 (tcp_fast_finwait2_recycle ? 2760 tcp_finwait2_timeout : 2761 TP_MAXIDLE(tp))); 2762 } 2763 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2764 } 2765 break; 2766 2767 /* 2768 * In CLOSING STATE in addition to the processing for 2769 * the ESTABLISHED state if the ACK acknowledges our FIN 2770 * then enter the TIME-WAIT state, otherwise ignore 2771 * the segment. 2772 */ 2773 case TCPS_CLOSING: 2774 if (ourfinisacked) { 2775 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2776 tcp_twstart(tp); 2777 INP_INFO_WUNLOCK(&V_tcbinfo); 2778 m_freem(m); 2779 return; 2780 } 2781 break; 2782 2783 /* 2784 * In LAST_ACK, we may still be waiting for data to drain 2785 * and/or to be acked, as well as for the ack of our FIN. 2786 * If our FIN is now acknowledged, delete the TCB, 2787 * enter the closed state and return. 2788 */ 2789 case TCPS_LAST_ACK: 2790 if (ourfinisacked) { 2791 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2792 tp = tcp_close(tp); 2793 goto drop; 2794 } 2795 break; 2796 } 2797 } 2798 2799 step6: 2800 INP_WLOCK_ASSERT(tp->t_inpcb); 2801 2802 /* 2803 * Update window information. 2804 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2805 */ 2806 if ((thflags & TH_ACK) && 2807 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2808 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2809 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2810 /* keep track of pure window updates */ 2811 if (tlen == 0 && 2812 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2813 TCPSTAT_INC(tcps_rcvwinupd); 2814 tp->snd_wnd = tiwin; 2815 tp->snd_wl1 = th->th_seq; 2816 tp->snd_wl2 = th->th_ack; 2817 if (tp->snd_wnd > tp->max_sndwnd) 2818 tp->max_sndwnd = tp->snd_wnd; 2819 needoutput = 1; 2820 } 2821 2822 /* 2823 * Process segments with URG. 2824 */ 2825 if ((thflags & TH_URG) && th->th_urp && 2826 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2827 /* 2828 * This is a kludge, but if we receive and accept 2829 * random urgent pointers, we'll crash in 2830 * soreceive. It's hard to imagine someone 2831 * actually wanting to send this much urgent data. 2832 */ 2833 SOCKBUF_LOCK(&so->so_rcv); 2834 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2835 th->th_urp = 0; /* XXX */ 2836 thflags &= ~TH_URG; /* XXX */ 2837 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2838 goto dodata; /* XXX */ 2839 } 2840 /* 2841 * If this segment advances the known urgent pointer, 2842 * then mark the data stream. This should not happen 2843 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2844 * a FIN has been received from the remote side. 2845 * In these states we ignore the URG. 2846 * 2847 * According to RFC961 (Assigned Protocols), 2848 * the urgent pointer points to the last octet 2849 * of urgent data. We continue, however, 2850 * to consider it to indicate the first octet 2851 * of data past the urgent section as the original 2852 * spec states (in one of two places). 2853 */ 2854 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2855 tp->rcv_up = th->th_seq + th->th_urp; 2856 so->so_oobmark = so->so_rcv.sb_cc + 2857 (tp->rcv_up - tp->rcv_nxt) - 1; 2858 if (so->so_oobmark == 0) 2859 so->so_rcv.sb_state |= SBS_RCVATMARK; 2860 sohasoutofband(so); 2861 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2862 } 2863 SOCKBUF_UNLOCK(&so->so_rcv); 2864 /* 2865 * Remove out of band data so doesn't get presented to user. 2866 * This can happen independent of advancing the URG pointer, 2867 * but if two URG's are pending at once, some out-of-band 2868 * data may creep in... ick. 2869 */ 2870 if (th->th_urp <= (u_long)tlen && 2871 !(so->so_options & SO_OOBINLINE)) { 2872 /* hdr drop is delayed */ 2873 tcp_pulloutofband(so, th, m, drop_hdrlen); 2874 } 2875 } else { 2876 /* 2877 * If no out of band data is expected, 2878 * pull receive urgent pointer along 2879 * with the receive window. 2880 */ 2881 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2882 tp->rcv_up = tp->rcv_nxt; 2883 } 2884 dodata: /* XXX */ 2885 INP_WLOCK_ASSERT(tp->t_inpcb); 2886 2887 /* 2888 * Process the segment text, merging it into the TCP sequencing queue, 2889 * and arranging for acknowledgment of receipt if necessary. 2890 * This process logically involves adjusting tp->rcv_wnd as data 2891 * is presented to the user (this happens in tcp_usrreq.c, 2892 * case PRU_RCVD). If a FIN has already been received on this 2893 * connection then we just ignore the text. 2894 */ 2895 if ((tlen || (thflags & TH_FIN)) && 2896 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2897 tcp_seq save_start = th->th_seq; 2898 m_adj(m, drop_hdrlen); /* delayed header drop */ 2899 /* 2900 * Insert segment which includes th into TCP reassembly queue 2901 * with control block tp. Set thflags to whether reassembly now 2902 * includes a segment with FIN. This handles the common case 2903 * inline (segment is the next to be received on an established 2904 * connection, and the queue is empty), avoiding linkage into 2905 * and removal from the queue and repetition of various 2906 * conversions. 2907 * Set DELACK for segments received in order, but ack 2908 * immediately when segments are out of order (so 2909 * fast retransmit can work). 2910 */ 2911 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL && 2912 TCPS_HAVEESTABLISHED(tp->t_state)) { 2913 if (DELAY_ACK(tp, tlen)) 2914 tp->t_flags |= TF_DELACK; 2915 else 2916 tp->t_flags |= TF_ACKNOW; 2917 tp->rcv_nxt += tlen; 2918 thflags = th->th_flags & TH_FIN; 2919 TCPSTAT_INC(tcps_rcvpack); 2920 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2921 ND6_HINT(tp); 2922 SOCKBUF_LOCK(&so->so_rcv); 2923 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2924 m_freem(m); 2925 else 2926 sbappendstream_locked(&so->so_rcv, m); 2927 /* NB: sorwakeup_locked() does an implicit unlock. */ 2928 sorwakeup_locked(so); 2929 } else { 2930 /* 2931 * XXX: Due to the header drop above "th" is 2932 * theoretically invalid by now. Fortunately 2933 * m_adj() doesn't actually frees any mbufs 2934 * when trimming from the head. 2935 */ 2936 thflags = tcp_reass(tp, th, &tlen, m); 2937 tp->t_flags |= TF_ACKNOW; 2938 } 2939 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2940 tcp_update_sack_list(tp, save_start, save_start + tlen); 2941 #if 0 2942 /* 2943 * Note the amount of data that peer has sent into 2944 * our window, in order to estimate the sender's 2945 * buffer size. 2946 * XXX: Unused. 2947 */ 2948 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2949 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2950 else 2951 len = so->so_rcv.sb_hiwat; 2952 #endif 2953 } else { 2954 m_freem(m); 2955 thflags &= ~TH_FIN; 2956 } 2957 2958 /* 2959 * If FIN is received ACK the FIN and let the user know 2960 * that the connection is closing. 2961 */ 2962 if (thflags & TH_FIN) { 2963 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2964 socantrcvmore(so); 2965 /* 2966 * If connection is half-synchronized 2967 * (ie NEEDSYN flag on) then delay ACK, 2968 * so it may be piggybacked when SYN is sent. 2969 * Otherwise, since we received a FIN then no 2970 * more input can be expected, send ACK now. 2971 */ 2972 if (tp->t_flags & TF_NEEDSYN) 2973 tp->t_flags |= TF_DELACK; 2974 else 2975 tp->t_flags |= TF_ACKNOW; 2976 tp->rcv_nxt++; 2977 } 2978 switch (tp->t_state) { 2979 2980 /* 2981 * In SYN_RECEIVED and ESTABLISHED STATES 2982 * enter the CLOSE_WAIT state. 2983 */ 2984 case TCPS_SYN_RECEIVED: 2985 tp->t_starttime = ticks; 2986 /* FALLTHROUGH */ 2987 case TCPS_ESTABLISHED: 2988 tcp_state_change(tp, TCPS_CLOSE_WAIT); 2989 break; 2990 2991 /* 2992 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2993 * enter the CLOSING state. 2994 */ 2995 case TCPS_FIN_WAIT_1: 2996 tcp_state_change(tp, TCPS_CLOSING); 2997 break; 2998 2999 /* 3000 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3001 * starting the time-wait timer, turning off the other 3002 * standard timers. 3003 */ 3004 case TCPS_FIN_WAIT_2: 3005 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 3006 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 3007 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3008 ti_locked)); 3009 3010 tcp_twstart(tp); 3011 INP_INFO_WUNLOCK(&V_tcbinfo); 3012 return; 3013 } 3014 } 3015 if (ti_locked == TI_WLOCKED) 3016 INP_INFO_WUNLOCK(&V_tcbinfo); 3017 ti_locked = TI_UNLOCKED; 3018 3019 #ifdef TCPDEBUG 3020 if (so->so_options & SO_DEBUG) 3021 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3022 &tcp_savetcp, 0); 3023 #endif 3024 3025 /* 3026 * Return any desired output. 3027 */ 3028 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3029 (void) tcp_output(tp); 3030 3031 check_delack: 3032 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3033 __func__, ti_locked)); 3034 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3035 INP_WLOCK_ASSERT(tp->t_inpcb); 3036 3037 if (tp->t_flags & TF_DELACK) { 3038 tp->t_flags &= ~TF_DELACK; 3039 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3040 } 3041 INP_WUNLOCK(tp->t_inpcb); 3042 return; 3043 3044 dropafterack: 3045 /* 3046 * Generate an ACK dropping incoming segment if it occupies 3047 * sequence space, where the ACK reflects our state. 3048 * 3049 * We can now skip the test for the RST flag since all 3050 * paths to this code happen after packets containing 3051 * RST have been dropped. 3052 * 3053 * In the SYN-RECEIVED state, don't send an ACK unless the 3054 * segment we received passes the SYN-RECEIVED ACK test. 3055 * If it fails send a RST. This breaks the loop in the 3056 * "LAND" DoS attack, and also prevents an ACK storm 3057 * between two listening ports that have been sent forged 3058 * SYN segments, each with the source address of the other. 3059 */ 3060 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3061 (SEQ_GT(tp->snd_una, th->th_ack) || 3062 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3063 rstreason = BANDLIM_RST_OPENPORT; 3064 goto dropwithreset; 3065 } 3066 #ifdef TCPDEBUG 3067 if (so->so_options & SO_DEBUG) 3068 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3069 &tcp_savetcp, 0); 3070 #endif 3071 if (ti_locked == TI_WLOCKED) 3072 INP_INFO_WUNLOCK(&V_tcbinfo); 3073 ti_locked = TI_UNLOCKED; 3074 3075 tp->t_flags |= TF_ACKNOW; 3076 (void) tcp_output(tp); 3077 INP_WUNLOCK(tp->t_inpcb); 3078 m_freem(m); 3079 return; 3080 3081 dropwithreset: 3082 if (ti_locked == TI_WLOCKED) 3083 INP_INFO_WUNLOCK(&V_tcbinfo); 3084 ti_locked = TI_UNLOCKED; 3085 3086 if (tp != NULL) { 3087 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3088 INP_WUNLOCK(tp->t_inpcb); 3089 } else 3090 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3091 return; 3092 3093 drop: 3094 if (ti_locked == TI_WLOCKED) { 3095 INP_INFO_WUNLOCK(&V_tcbinfo); 3096 ti_locked = TI_UNLOCKED; 3097 } 3098 #ifdef INVARIANTS 3099 else 3100 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3101 #endif 3102 3103 /* 3104 * Drop space held by incoming segment and return. 3105 */ 3106 #ifdef TCPDEBUG 3107 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3108 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3109 &tcp_savetcp, 0); 3110 #endif 3111 if (tp != NULL) 3112 INP_WUNLOCK(tp->t_inpcb); 3113 m_freem(m); 3114 } 3115 3116 /* 3117 * Issue RST and make ACK acceptable to originator of segment. 3118 * The mbuf must still include the original packet header. 3119 * tp may be NULL. 3120 */ 3121 static void 3122 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3123 int tlen, int rstreason) 3124 { 3125 #ifdef INET 3126 struct ip *ip; 3127 #endif 3128 #ifdef INET6 3129 struct ip6_hdr *ip6; 3130 #endif 3131 3132 if (tp != NULL) { 3133 INP_WLOCK_ASSERT(tp->t_inpcb); 3134 } 3135 3136 /* Don't bother if destination was broadcast/multicast. */ 3137 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3138 goto drop; 3139 #ifdef INET6 3140 if (mtod(m, struct ip *)->ip_v == 6) { 3141 ip6 = mtod(m, struct ip6_hdr *); 3142 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3143 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3144 goto drop; 3145 /* IPv6 anycast check is done at tcp6_input() */ 3146 } 3147 #endif 3148 #if defined(INET) && defined(INET6) 3149 else 3150 #endif 3151 #ifdef INET 3152 { 3153 ip = mtod(m, struct ip *); 3154 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3155 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3156 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3157 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3158 goto drop; 3159 } 3160 #endif 3161 3162 /* Perform bandwidth limiting. */ 3163 if (badport_bandlim(rstreason) < 0) 3164 goto drop; 3165 3166 /* tcp_respond consumes the mbuf chain. */ 3167 if (th->th_flags & TH_ACK) { 3168 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3169 th->th_ack, TH_RST); 3170 } else { 3171 if (th->th_flags & TH_SYN) 3172 tlen++; 3173 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3174 (tcp_seq)0, TH_RST|TH_ACK); 3175 } 3176 return; 3177 drop: 3178 m_freem(m); 3179 } 3180 3181 /* 3182 * Parse TCP options and place in tcpopt. 3183 */ 3184 static void 3185 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3186 { 3187 int opt, optlen; 3188 3189 to->to_flags = 0; 3190 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3191 opt = cp[0]; 3192 if (opt == TCPOPT_EOL) 3193 break; 3194 if (opt == TCPOPT_NOP) 3195 optlen = 1; 3196 else { 3197 if (cnt < 2) 3198 break; 3199 optlen = cp[1]; 3200 if (optlen < 2 || optlen > cnt) 3201 break; 3202 } 3203 switch (opt) { 3204 case TCPOPT_MAXSEG: 3205 if (optlen != TCPOLEN_MAXSEG) 3206 continue; 3207 if (!(flags & TO_SYN)) 3208 continue; 3209 to->to_flags |= TOF_MSS; 3210 bcopy((char *)cp + 2, 3211 (char *)&to->to_mss, sizeof(to->to_mss)); 3212 to->to_mss = ntohs(to->to_mss); 3213 break; 3214 case TCPOPT_WINDOW: 3215 if (optlen != TCPOLEN_WINDOW) 3216 continue; 3217 if (!(flags & TO_SYN)) 3218 continue; 3219 to->to_flags |= TOF_SCALE; 3220 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3221 break; 3222 case TCPOPT_TIMESTAMP: 3223 if (optlen != TCPOLEN_TIMESTAMP) 3224 continue; 3225 to->to_flags |= TOF_TS; 3226 bcopy((char *)cp + 2, 3227 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3228 to->to_tsval = ntohl(to->to_tsval); 3229 bcopy((char *)cp + 6, 3230 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3231 to->to_tsecr = ntohl(to->to_tsecr); 3232 break; 3233 #ifdef TCP_SIGNATURE 3234 /* 3235 * XXX In order to reply to a host which has set the 3236 * TCP_SIGNATURE option in its initial SYN, we have to 3237 * record the fact that the option was observed here 3238 * for the syncache code to perform the correct response. 3239 */ 3240 case TCPOPT_SIGNATURE: 3241 if (optlen != TCPOLEN_SIGNATURE) 3242 continue; 3243 to->to_flags |= TOF_SIGNATURE; 3244 to->to_signature = cp + 2; 3245 break; 3246 #endif 3247 case TCPOPT_SACK_PERMITTED: 3248 if (optlen != TCPOLEN_SACK_PERMITTED) 3249 continue; 3250 if (!(flags & TO_SYN)) 3251 continue; 3252 if (!V_tcp_do_sack) 3253 continue; 3254 to->to_flags |= TOF_SACKPERM; 3255 break; 3256 case TCPOPT_SACK: 3257 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3258 continue; 3259 if (flags & TO_SYN) 3260 continue; 3261 to->to_flags |= TOF_SACK; 3262 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3263 to->to_sacks = cp + 2; 3264 TCPSTAT_INC(tcps_sack_rcv_blocks); 3265 break; 3266 default: 3267 continue; 3268 } 3269 } 3270 } 3271 3272 /* 3273 * Pull out of band byte out of a segment so 3274 * it doesn't appear in the user's data queue. 3275 * It is still reflected in the segment length for 3276 * sequencing purposes. 3277 */ 3278 static void 3279 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3280 int off) 3281 { 3282 int cnt = off + th->th_urp - 1; 3283 3284 while (cnt >= 0) { 3285 if (m->m_len > cnt) { 3286 char *cp = mtod(m, caddr_t) + cnt; 3287 struct tcpcb *tp = sototcpcb(so); 3288 3289 INP_WLOCK_ASSERT(tp->t_inpcb); 3290 3291 tp->t_iobc = *cp; 3292 tp->t_oobflags |= TCPOOB_HAVEDATA; 3293 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3294 m->m_len--; 3295 if (m->m_flags & M_PKTHDR) 3296 m->m_pkthdr.len--; 3297 return; 3298 } 3299 cnt -= m->m_len; 3300 m = m->m_next; 3301 if (m == NULL) 3302 break; 3303 } 3304 panic("tcp_pulloutofband"); 3305 } 3306 3307 /* 3308 * Collect new round-trip time estimate 3309 * and update averages and current timeout. 3310 */ 3311 static void 3312 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3313 { 3314 int delta; 3315 3316 INP_WLOCK_ASSERT(tp->t_inpcb); 3317 3318 TCPSTAT_INC(tcps_rttupdated); 3319 tp->t_rttupdated++; 3320 if (tp->t_srtt != 0) { 3321 /* 3322 * srtt is stored as fixed point with 5 bits after the 3323 * binary point (i.e., scaled by 8). The following magic 3324 * is equivalent to the smoothing algorithm in rfc793 with 3325 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3326 * point). Adjust rtt to origin 0. 3327 */ 3328 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3329 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3330 3331 if ((tp->t_srtt += delta) <= 0) 3332 tp->t_srtt = 1; 3333 3334 /* 3335 * We accumulate a smoothed rtt variance (actually, a 3336 * smoothed mean difference), then set the retransmit 3337 * timer to smoothed rtt + 4 times the smoothed variance. 3338 * rttvar is stored as fixed point with 4 bits after the 3339 * binary point (scaled by 16). The following is 3340 * equivalent to rfc793 smoothing with an alpha of .75 3341 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3342 * rfc793's wired-in beta. 3343 */ 3344 if (delta < 0) 3345 delta = -delta; 3346 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3347 if ((tp->t_rttvar += delta) <= 0) 3348 tp->t_rttvar = 1; 3349 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3350 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3351 } else { 3352 /* 3353 * No rtt measurement yet - use the unsmoothed rtt. 3354 * Set the variance to half the rtt (so our first 3355 * retransmit happens at 3*rtt). 3356 */ 3357 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3358 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3359 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3360 } 3361 tp->t_rtttime = 0; 3362 tp->t_rxtshift = 0; 3363 3364 /* 3365 * the retransmit should happen at rtt + 4 * rttvar. 3366 * Because of the way we do the smoothing, srtt and rttvar 3367 * will each average +1/2 tick of bias. When we compute 3368 * the retransmit timer, we want 1/2 tick of rounding and 3369 * 1 extra tick because of +-1/2 tick uncertainty in the 3370 * firing of the timer. The bias will give us exactly the 3371 * 1.5 tick we need. But, because the bias is 3372 * statistical, we have to test that we don't drop below 3373 * the minimum feasible timer (which is 2 ticks). 3374 */ 3375 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3376 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3377 3378 /* 3379 * We received an ack for a packet that wasn't retransmitted; 3380 * it is probably safe to discard any error indications we've 3381 * received recently. This isn't quite right, but close enough 3382 * for now (a route might have failed after we sent a segment, 3383 * and the return path might not be symmetrical). 3384 */ 3385 tp->t_softerror = 0; 3386 } 3387 3388 /* 3389 * Determine a reasonable value for maxseg size. 3390 * If the route is known, check route for mtu. 3391 * If none, use an mss that can be handled on the outgoing interface 3392 * without forcing IP to fragment. If no route is found, route has no mtu, 3393 * or the destination isn't local, use a default, hopefully conservative 3394 * size (usually 512 or the default IP max size, but no more than the mtu 3395 * of the interface), as we can't discover anything about intervening 3396 * gateways or networks. We also initialize the congestion/slow start 3397 * window to be a single segment if the destination isn't local. 3398 * While looking at the routing entry, we also initialize other path-dependent 3399 * parameters from pre-set or cached values in the routing entry. 3400 * 3401 * Also take into account the space needed for options that we 3402 * send regularly. Make maxseg shorter by that amount to assure 3403 * that we can send maxseg amount of data even when the options 3404 * are present. Store the upper limit of the length of options plus 3405 * data in maxopd. 3406 * 3407 * NOTE that this routine is only called when we process an incoming 3408 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3409 * settings are handled in tcp_mssopt(). 3410 */ 3411 void 3412 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3413 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3414 { 3415 int mss = 0; 3416 u_long maxmtu = 0; 3417 struct inpcb *inp = tp->t_inpcb; 3418 struct hc_metrics_lite metrics; 3419 int origoffer; 3420 #ifdef INET6 3421 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3422 size_t min_protoh = isipv6 ? 3423 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3424 sizeof (struct tcpiphdr); 3425 #else 3426 const size_t min_protoh = sizeof(struct tcpiphdr); 3427 #endif 3428 3429 INP_WLOCK_ASSERT(tp->t_inpcb); 3430 3431 if (mtuoffer != -1) { 3432 KASSERT(offer == -1, ("%s: conflict", __func__)); 3433 offer = mtuoffer - min_protoh; 3434 } 3435 origoffer = offer; 3436 3437 /* Initialize. */ 3438 #ifdef INET6 3439 if (isipv6) { 3440 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3441 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3442 } 3443 #endif 3444 #if defined(INET) && defined(INET6) 3445 else 3446 #endif 3447 #ifdef INET 3448 { 3449 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3450 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3451 } 3452 #endif 3453 3454 /* 3455 * No route to sender, stay with default mss and return. 3456 */ 3457 if (maxmtu == 0) { 3458 /* 3459 * In case we return early we need to initialize metrics 3460 * to a defined state as tcp_hc_get() would do for us 3461 * if there was no cache hit. 3462 */ 3463 if (metricptr != NULL) 3464 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3465 return; 3466 } 3467 3468 /* What have we got? */ 3469 switch (offer) { 3470 case 0: 3471 /* 3472 * Offer == 0 means that there was no MSS on the SYN 3473 * segment, in this case we use tcp_mssdflt as 3474 * already assigned to t_maxopd above. 3475 */ 3476 offer = tp->t_maxopd; 3477 break; 3478 3479 case -1: 3480 /* 3481 * Offer == -1 means that we didn't receive SYN yet. 3482 */ 3483 /* FALLTHROUGH */ 3484 3485 default: 3486 /* 3487 * Prevent DoS attack with too small MSS. Round up 3488 * to at least minmss. 3489 */ 3490 offer = max(offer, V_tcp_minmss); 3491 } 3492 3493 /* 3494 * rmx information is now retrieved from tcp_hostcache. 3495 */ 3496 tcp_hc_get(&inp->inp_inc, &metrics); 3497 if (metricptr != NULL) 3498 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3499 3500 /* 3501 * If there's a discovered mtu int tcp hostcache, use it 3502 * else, use the link mtu. 3503 */ 3504 if (metrics.rmx_mtu) 3505 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3506 else { 3507 #ifdef INET6 3508 if (isipv6) { 3509 mss = maxmtu - min_protoh; 3510 if (!V_path_mtu_discovery && 3511 !in6_localaddr(&inp->in6p_faddr)) 3512 mss = min(mss, V_tcp_v6mssdflt); 3513 } 3514 #endif 3515 #if defined(INET) && defined(INET6) 3516 else 3517 #endif 3518 #ifdef INET 3519 { 3520 mss = maxmtu - min_protoh; 3521 if (!V_path_mtu_discovery && 3522 !in_localaddr(inp->inp_faddr)) 3523 mss = min(mss, V_tcp_mssdflt); 3524 } 3525 #endif 3526 /* 3527 * XXX - The above conditional (mss = maxmtu - min_protoh) 3528 * probably violates the TCP spec. 3529 * The problem is that, since we don't know the 3530 * other end's MSS, we are supposed to use a conservative 3531 * default. But, if we do that, then MTU discovery will 3532 * never actually take place, because the conservative 3533 * default is much less than the MTUs typically seen 3534 * on the Internet today. For the moment, we'll sweep 3535 * this under the carpet. 3536 * 3537 * The conservative default might not actually be a problem 3538 * if the only case this occurs is when sending an initial 3539 * SYN with options and data to a host we've never talked 3540 * to before. Then, they will reply with an MSS value which 3541 * will get recorded and the new parameters should get 3542 * recomputed. For Further Study. 3543 */ 3544 } 3545 mss = min(mss, offer); 3546 3547 /* 3548 * Sanity check: make sure that maxopd will be large 3549 * enough to allow some data on segments even if the 3550 * all the option space is used (40bytes). Otherwise 3551 * funny things may happen in tcp_output. 3552 */ 3553 mss = max(mss, 64); 3554 3555 /* 3556 * maxopd stores the maximum length of data AND options 3557 * in a segment; maxseg is the amount of data in a normal 3558 * segment. We need to store this value (maxopd) apart 3559 * from maxseg, because now every segment carries options 3560 * and thus we normally have somewhat less data in segments. 3561 */ 3562 tp->t_maxopd = mss; 3563 3564 /* 3565 * origoffer==-1 indicates that no segments were received yet. 3566 * In this case we just guess. 3567 */ 3568 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3569 (origoffer == -1 || 3570 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3571 mss -= TCPOLEN_TSTAMP_APPA; 3572 3573 tp->t_maxseg = mss; 3574 } 3575 3576 void 3577 tcp_mss(struct tcpcb *tp, int offer) 3578 { 3579 int mss; 3580 u_long bufsize; 3581 struct inpcb *inp; 3582 struct socket *so; 3583 struct hc_metrics_lite metrics; 3584 struct tcp_ifcap cap; 3585 3586 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3587 3588 bzero(&cap, sizeof(cap)); 3589 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3590 3591 mss = tp->t_maxseg; 3592 inp = tp->t_inpcb; 3593 3594 /* 3595 * If there's a pipesize, change the socket buffer to that size, 3596 * don't change if sb_hiwat is different than default (then it 3597 * has been changed on purpose with setsockopt). 3598 * Make the socket buffers an integral number of mss units; 3599 * if the mss is larger than the socket buffer, decrease the mss. 3600 */ 3601 so = inp->inp_socket; 3602 SOCKBUF_LOCK(&so->so_snd); 3603 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3604 bufsize = metrics.rmx_sendpipe; 3605 else 3606 bufsize = so->so_snd.sb_hiwat; 3607 if (bufsize < mss) 3608 mss = bufsize; 3609 else { 3610 bufsize = roundup(bufsize, mss); 3611 if (bufsize > sb_max) 3612 bufsize = sb_max; 3613 if (bufsize > so->so_snd.sb_hiwat) 3614 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3615 } 3616 SOCKBUF_UNLOCK(&so->so_snd); 3617 tp->t_maxseg = mss; 3618 3619 SOCKBUF_LOCK(&so->so_rcv); 3620 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3621 bufsize = metrics.rmx_recvpipe; 3622 else 3623 bufsize = so->so_rcv.sb_hiwat; 3624 if (bufsize > mss) { 3625 bufsize = roundup(bufsize, mss); 3626 if (bufsize > sb_max) 3627 bufsize = sb_max; 3628 if (bufsize > so->so_rcv.sb_hiwat) 3629 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3630 } 3631 SOCKBUF_UNLOCK(&so->so_rcv); 3632 3633 /* Check the interface for TSO capabilities. */ 3634 if (cap.ifcap & CSUM_TSO) { 3635 tp->t_flags |= TF_TSO; 3636 tp->t_tsomax = cap.tsomax; 3637 } 3638 } 3639 3640 /* 3641 * Determine the MSS option to send on an outgoing SYN. 3642 */ 3643 int 3644 tcp_mssopt(struct in_conninfo *inc) 3645 { 3646 int mss = 0; 3647 u_long maxmtu = 0; 3648 u_long thcmtu = 0; 3649 size_t min_protoh; 3650 3651 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3652 3653 #ifdef INET6 3654 if (inc->inc_flags & INC_ISIPV6) { 3655 mss = V_tcp_v6mssdflt; 3656 maxmtu = tcp_maxmtu6(inc, NULL); 3657 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3658 } 3659 #endif 3660 #if defined(INET) && defined(INET6) 3661 else 3662 #endif 3663 #ifdef INET 3664 { 3665 mss = V_tcp_mssdflt; 3666 maxmtu = tcp_maxmtu(inc, NULL); 3667 min_protoh = sizeof(struct tcpiphdr); 3668 } 3669 #endif 3670 #if defined(INET6) || defined(INET) 3671 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3672 #endif 3673 3674 if (maxmtu && thcmtu) 3675 mss = min(maxmtu, thcmtu) - min_protoh; 3676 else if (maxmtu || thcmtu) 3677 mss = max(maxmtu, thcmtu) - min_protoh; 3678 3679 return (mss); 3680 } 3681 3682 3683 /* 3684 * On a partial ack arrives, force the retransmission of the 3685 * next unacknowledged segment. Do not clear tp->t_dupacks. 3686 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3687 * be started again. 3688 */ 3689 static void 3690 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3691 { 3692 tcp_seq onxt = tp->snd_nxt; 3693 u_long ocwnd = tp->snd_cwnd; 3694 3695 INP_WLOCK_ASSERT(tp->t_inpcb); 3696 3697 tcp_timer_activate(tp, TT_REXMT, 0); 3698 tp->t_rtttime = 0; 3699 tp->snd_nxt = th->th_ack; 3700 /* 3701 * Set snd_cwnd to one segment beyond acknowledged offset. 3702 * (tp->snd_una has not yet been updated when this function is called.) 3703 */ 3704 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3705 tp->t_flags |= TF_ACKNOW; 3706 (void) tcp_output(tp); 3707 tp->snd_cwnd = ocwnd; 3708 if (SEQ_GT(onxt, tp->snd_nxt)) 3709 tp->snd_nxt = onxt; 3710 /* 3711 * Partial window deflation. Relies on fact that tp->snd_una 3712 * not updated yet. 3713 */ 3714 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3715 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3716 else 3717 tp->snd_cwnd = 0; 3718 tp->snd_cwnd += tp->t_maxseg; 3719 } 3720