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 * - there is no delayed ack timer in progress and 507 * - our last ack wasn't a 0-sized window. We never want to delay 508 * the ack that opens up a 0-sized window and 509 * - delayed acks are enabled or 510 * - this is a half-synchronized T/TCP connection. 511 * - the segment size is not larger than the MSS and LRO wasn't used 512 * for this segment. 513 */ 514 #define DELAY_ACK(tp, tlen) \ 515 ((!tcp_timer_active(tp, TT_DELACK) && \ 516 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 517 (tlen <= tp->t_maxopd) && \ 518 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 519 520 /* 521 * TCP input handling is split into multiple parts: 522 * tcp6_input is a thin wrapper around tcp_input for the extended 523 * ip6_protox[] call format in ip6_input 524 * tcp_input handles primary segment validation, inpcb lookup and 525 * SYN processing on listen sockets 526 * tcp_do_segment processes the ACK and text of the segment for 527 * establishing, established and closing connections 528 */ 529 #ifdef INET6 530 int 531 tcp6_input(struct mbuf **mp, int *offp, int proto) 532 { 533 struct mbuf *m = *mp; 534 struct in6_ifaddr *ia6; 535 536 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 537 538 /* 539 * draft-itojun-ipv6-tcp-to-anycast 540 * better place to put this in? 541 */ 542 ia6 = ip6_getdstifaddr(m); 543 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 544 struct ip6_hdr *ip6; 545 546 ifa_free(&ia6->ia_ifa); 547 ip6 = mtod(m, struct ip6_hdr *); 548 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 549 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 550 return IPPROTO_DONE; 551 } 552 if (ia6) 553 ifa_free(&ia6->ia_ifa); 554 555 tcp_input(m, *offp); 556 return IPPROTO_DONE; 557 } 558 #endif /* INET6 */ 559 560 void 561 tcp_input(struct mbuf *m, int off0) 562 { 563 struct tcphdr *th = NULL; 564 struct ip *ip = NULL; 565 struct inpcb *inp = NULL; 566 struct tcpcb *tp = NULL; 567 struct socket *so = NULL; 568 u_char *optp = NULL; 569 int optlen = 0; 570 #ifdef INET 571 int len; 572 #endif 573 int tlen = 0, off; 574 int drop_hdrlen; 575 int thflags; 576 int rstreason = 0; /* For badport_bandlim accounting purposes */ 577 #ifdef TCP_SIGNATURE 578 uint8_t sig_checked = 0; 579 #endif 580 uint8_t iptos = 0; 581 struct m_tag *fwd_tag = NULL; 582 #ifdef INET6 583 struct ip6_hdr *ip6 = NULL; 584 int isipv6; 585 #else 586 const void *ip6 = NULL; 587 #endif /* INET6 */ 588 struct tcpopt to; /* options in this segment */ 589 char *s = NULL; /* address and port logging */ 590 int ti_locked; 591 #define TI_UNLOCKED 1 592 #define TI_WLOCKED 2 593 594 #ifdef TCPDEBUG 595 /* 596 * The size of tcp_saveipgen must be the size of the max ip header, 597 * now IPv6. 598 */ 599 u_char tcp_saveipgen[IP6_HDR_LEN]; 600 struct tcphdr tcp_savetcp; 601 short ostate = 0; 602 #endif 603 604 #ifdef INET6 605 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 606 #endif 607 608 to.to_flags = 0; 609 TCPSTAT_INC(tcps_rcvtotal); 610 611 #ifdef INET6 612 if (isipv6) { 613 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 614 615 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 616 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 617 if (m == NULL) { 618 TCPSTAT_INC(tcps_rcvshort); 619 return; 620 } 621 } 622 623 ip6 = mtod(m, struct ip6_hdr *); 624 th = (struct tcphdr *)((caddr_t)ip6 + off0); 625 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 626 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 627 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 628 th->th_sum = m->m_pkthdr.csum_data; 629 else 630 th->th_sum = in6_cksum_pseudo(ip6, tlen, 631 IPPROTO_TCP, m->m_pkthdr.csum_data); 632 th->th_sum ^= 0xffff; 633 } else 634 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 635 if (th->th_sum) { 636 TCPSTAT_INC(tcps_rcvbadsum); 637 goto drop; 638 } 639 640 /* 641 * Be proactive about unspecified IPv6 address in source. 642 * As we use all-zero to indicate unbounded/unconnected pcb, 643 * unspecified IPv6 address can be used to confuse us. 644 * 645 * Note that packets with unspecified IPv6 destination is 646 * already dropped in ip6_input. 647 */ 648 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 649 /* XXX stat */ 650 goto drop; 651 } 652 } 653 #endif 654 #if defined(INET) && defined(INET6) 655 else 656 #endif 657 #ifdef INET 658 { 659 /* 660 * Get IP and TCP header together in first mbuf. 661 * Note: IP leaves IP header in first mbuf. 662 */ 663 if (off0 > sizeof (struct ip)) { 664 ip_stripoptions(m); 665 off0 = sizeof(struct ip); 666 } 667 if (m->m_len < sizeof (struct tcpiphdr)) { 668 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 669 == NULL) { 670 TCPSTAT_INC(tcps_rcvshort); 671 return; 672 } 673 } 674 ip = mtod(m, struct ip *); 675 th = (struct tcphdr *)((caddr_t)ip + off0); 676 tlen = ntohs(ip->ip_len) - off0; 677 678 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 679 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 680 th->th_sum = m->m_pkthdr.csum_data; 681 else 682 th->th_sum = in_pseudo(ip->ip_src.s_addr, 683 ip->ip_dst.s_addr, 684 htonl(m->m_pkthdr.csum_data + tlen + 685 IPPROTO_TCP)); 686 th->th_sum ^= 0xffff; 687 } else { 688 struct ipovly *ipov = (struct ipovly *)ip; 689 690 /* 691 * Checksum extended TCP header and data. 692 */ 693 len = off0 + tlen; 694 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 695 ipov->ih_len = htons(tlen); 696 th->th_sum = in_cksum(m, len); 697 /* Reset length for SDT probes. */ 698 ip->ip_len = htons(tlen + off0); 699 } 700 701 if (th->th_sum) { 702 TCPSTAT_INC(tcps_rcvbadsum); 703 goto drop; 704 } 705 /* Re-initialization for later version check */ 706 ip->ip_v = IPVERSION; 707 } 708 #endif /* INET */ 709 710 #ifdef INET6 711 if (isipv6) 712 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 713 #endif 714 #if defined(INET) && defined(INET6) 715 else 716 #endif 717 #ifdef INET 718 iptos = ip->ip_tos; 719 #endif 720 721 /* 722 * Check that TCP offset makes sense, 723 * pull out TCP options and adjust length. XXX 724 */ 725 off = th->th_off << 2; 726 if (off < sizeof (struct tcphdr) || off > tlen) { 727 TCPSTAT_INC(tcps_rcvbadoff); 728 goto drop; 729 } 730 tlen -= off; /* tlen is used instead of ti->ti_len */ 731 if (off > sizeof (struct tcphdr)) { 732 #ifdef INET6 733 if (isipv6) { 734 IP6_EXTHDR_CHECK(m, off0, off, ); 735 ip6 = mtod(m, struct ip6_hdr *); 736 th = (struct tcphdr *)((caddr_t)ip6 + off0); 737 } 738 #endif 739 #if defined(INET) && defined(INET6) 740 else 741 #endif 742 #ifdef INET 743 { 744 if (m->m_len < sizeof(struct ip) + off) { 745 if ((m = m_pullup(m, sizeof (struct ip) + off)) 746 == NULL) { 747 TCPSTAT_INC(tcps_rcvshort); 748 return; 749 } 750 ip = mtod(m, struct ip *); 751 th = (struct tcphdr *)((caddr_t)ip + off0); 752 } 753 } 754 #endif 755 optlen = off - sizeof (struct tcphdr); 756 optp = (u_char *)(th + 1); 757 } 758 thflags = th->th_flags; 759 760 /* 761 * Convert TCP protocol specific fields to host format. 762 */ 763 tcp_fields_to_host(th); 764 765 /* 766 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 767 */ 768 drop_hdrlen = off0 + off; 769 770 /* 771 * Locate pcb for segment; if we're likely to add or remove a 772 * connection then first acquire pcbinfo lock. There are two cases 773 * where we might discover later we need a write lock despite the 774 * flags: ACKs moving a connection out of the syncache, and ACKs for 775 * a connection in TIMEWAIT. 776 */ 777 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 778 INP_INFO_WLOCK(&V_tcbinfo); 779 ti_locked = TI_WLOCKED; 780 } else 781 ti_locked = TI_UNLOCKED; 782 783 /* 784 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 785 */ 786 if ( 787 #ifdef INET6 788 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 789 #ifdef INET 790 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 791 #endif 792 #endif 793 #if defined(INET) && !defined(INET6) 794 (m->m_flags & M_IP_NEXTHOP) 795 #endif 796 ) 797 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 798 799 findpcb: 800 #ifdef INVARIANTS 801 if (ti_locked == TI_WLOCKED) { 802 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 803 } else { 804 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 805 } 806 #endif 807 #ifdef INET6 808 if (isipv6 && fwd_tag != NULL) { 809 struct sockaddr_in6 *next_hop6; 810 811 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 812 /* 813 * Transparently forwarded. Pretend to be the destination. 814 * Already got one like this? 815 */ 816 inp = in6_pcblookup_mbuf(&V_tcbinfo, 817 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 818 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 819 if (!inp) { 820 /* 821 * It's new. Try to find the ambushing socket. 822 * Because we've rewritten the destination address, 823 * any hardware-generated hash is ignored. 824 */ 825 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 826 th->th_sport, &next_hop6->sin6_addr, 827 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 828 th->th_dport, INPLOOKUP_WILDCARD | 829 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 830 } 831 } else if (isipv6) { 832 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 833 th->th_sport, &ip6->ip6_dst, th->th_dport, 834 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 835 m->m_pkthdr.rcvif, m); 836 } 837 #endif /* INET6 */ 838 #if defined(INET6) && defined(INET) 839 else 840 #endif 841 #ifdef INET 842 if (fwd_tag != NULL) { 843 struct sockaddr_in *next_hop; 844 845 next_hop = (struct sockaddr_in *)(fwd_tag+1); 846 /* 847 * Transparently forwarded. Pretend to be the destination. 848 * already got one like this? 849 */ 850 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 851 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 852 m->m_pkthdr.rcvif, m); 853 if (!inp) { 854 /* 855 * It's new. Try to find the ambushing socket. 856 * Because we've rewritten the destination address, 857 * any hardware-generated hash is ignored. 858 */ 859 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 860 th->th_sport, next_hop->sin_addr, 861 next_hop->sin_port ? ntohs(next_hop->sin_port) : 862 th->th_dport, INPLOOKUP_WILDCARD | 863 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 864 } 865 } else 866 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 867 th->th_sport, ip->ip_dst, th->th_dport, 868 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 869 m->m_pkthdr.rcvif, m); 870 #endif /* INET */ 871 872 /* 873 * If the INPCB does not exist then all data in the incoming 874 * segment is discarded and an appropriate RST is sent back. 875 * XXX MRT Send RST using which routing table? 876 */ 877 if (inp == NULL) { 878 /* 879 * Log communication attempts to ports that are not 880 * in use. 881 */ 882 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 883 tcp_log_in_vain == 2) { 884 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 885 log(LOG_INFO, "%s; %s: Connection attempt " 886 "to closed port\n", s, __func__); 887 } 888 /* 889 * When blackholing do not respond with a RST but 890 * completely ignore the segment and drop it. 891 */ 892 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 893 V_blackhole == 2) 894 goto dropunlock; 895 896 rstreason = BANDLIM_RST_CLOSEDPORT; 897 goto dropwithreset; 898 } 899 INP_WLOCK_ASSERT(inp); 900 if (!(inp->inp_flags & INP_HW_FLOWID) 901 && (m->m_flags & M_FLOWID) 902 && ((inp->inp_socket == NULL) 903 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 904 inp->inp_flags |= INP_HW_FLOWID; 905 inp->inp_flags &= ~INP_SW_FLOWID; 906 inp->inp_flowid = m->m_pkthdr.flowid; 907 } 908 #ifdef IPSEC 909 #ifdef INET6 910 if (isipv6 && ipsec6_in_reject(m, inp)) { 911 IPSEC6STAT_INC(ips_in_polvio); 912 goto dropunlock; 913 } else 914 #endif /* INET6 */ 915 if (ipsec4_in_reject(m, inp) != 0) { 916 IPSECSTAT_INC(ips_in_polvio); 917 goto dropunlock; 918 } 919 #endif /* IPSEC */ 920 921 /* 922 * Check the minimum TTL for socket. 923 */ 924 if (inp->inp_ip_minttl != 0) { 925 #ifdef INET6 926 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 927 goto dropunlock; 928 else 929 #endif 930 if (inp->inp_ip_minttl > ip->ip_ttl) 931 goto dropunlock; 932 } 933 934 /* 935 * A previous connection in TIMEWAIT state is supposed to catch stray 936 * or duplicate segments arriving late. If this segment was a 937 * legitimate new connection attempt, the old INPCB gets removed and 938 * we can try again to find a listening socket. 939 * 940 * At this point, due to earlier optimism, we may hold only an inpcb 941 * lock, and not the inpcbinfo write lock. If so, we need to try to 942 * acquire it, or if that fails, acquire a reference on the inpcb, 943 * drop all locks, acquire a global write lock, and then re-acquire 944 * the inpcb lock. We may at that point discover that another thread 945 * has tried to free the inpcb, in which case we need to loop back 946 * and try to find a new inpcb to deliver to. 947 * 948 * XXXRW: It may be time to rethink timewait locking. 949 */ 950 relocked: 951 if (inp->inp_flags & INP_TIMEWAIT) { 952 if (ti_locked == TI_UNLOCKED) { 953 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 954 in_pcbref(inp); 955 INP_WUNLOCK(inp); 956 INP_INFO_WLOCK(&V_tcbinfo); 957 ti_locked = TI_WLOCKED; 958 INP_WLOCK(inp); 959 if (in_pcbrele_wlocked(inp)) { 960 inp = NULL; 961 goto findpcb; 962 } 963 } else 964 ti_locked = TI_WLOCKED; 965 } 966 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 967 968 if (thflags & TH_SYN) 969 tcp_dooptions(&to, optp, optlen, TO_SYN); 970 /* 971 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 972 */ 973 if (tcp_twcheck(inp, &to, th, m, tlen)) 974 goto findpcb; 975 INP_INFO_WUNLOCK(&V_tcbinfo); 976 return; 977 } 978 /* 979 * The TCPCB may no longer exist if the connection is winding 980 * down or it is in the CLOSED state. Either way we drop the 981 * segment and send an appropriate response. 982 */ 983 tp = intotcpcb(inp); 984 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 985 rstreason = BANDLIM_RST_CLOSEDPORT; 986 goto dropwithreset; 987 } 988 989 #ifdef TCP_OFFLOAD 990 if (tp->t_flags & TF_TOE) { 991 tcp_offload_input(tp, m); 992 m = NULL; /* consumed by the TOE driver */ 993 goto dropunlock; 994 } 995 #endif 996 997 /* 998 * We've identified a valid inpcb, but it could be that we need an 999 * inpcbinfo write lock but don't hold it. In this case, attempt to 1000 * acquire using the same strategy as the TIMEWAIT case above. If we 1001 * relock, we have to jump back to 'relocked' as the connection might 1002 * now be in TIMEWAIT. 1003 */ 1004 #ifdef INVARIANTS 1005 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 1006 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1007 #endif 1008 if (tp->t_state != TCPS_ESTABLISHED) { 1009 if (ti_locked == TI_UNLOCKED) { 1010 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 1011 in_pcbref(inp); 1012 INP_WUNLOCK(inp); 1013 INP_INFO_WLOCK(&V_tcbinfo); 1014 ti_locked = TI_WLOCKED; 1015 INP_WLOCK(inp); 1016 if (in_pcbrele_wlocked(inp)) { 1017 inp = NULL; 1018 goto findpcb; 1019 } 1020 goto relocked; 1021 } else 1022 ti_locked = TI_WLOCKED; 1023 } 1024 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1025 } 1026 1027 #ifdef MAC 1028 INP_WLOCK_ASSERT(inp); 1029 if (mac_inpcb_check_deliver(inp, m)) 1030 goto dropunlock; 1031 #endif 1032 so = inp->inp_socket; 1033 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1034 #ifdef TCPDEBUG 1035 if (so->so_options & SO_DEBUG) { 1036 ostate = tp->t_state; 1037 #ifdef INET6 1038 if (isipv6) { 1039 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1040 } else 1041 #endif 1042 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1043 tcp_savetcp = *th; 1044 } 1045 #endif /* TCPDEBUG */ 1046 /* 1047 * When the socket is accepting connections (the INPCB is in LISTEN 1048 * state) we look into the SYN cache if this is a new connection 1049 * attempt or the completion of a previous one. Because listen 1050 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1051 * held in this case. 1052 */ 1053 if (so->so_options & SO_ACCEPTCONN) { 1054 struct in_conninfo inc; 1055 1056 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1057 "tp not listening", __func__)); 1058 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1059 1060 bzero(&inc, sizeof(inc)); 1061 #ifdef INET6 1062 if (isipv6) { 1063 inc.inc_flags |= INC_ISIPV6; 1064 inc.inc6_faddr = ip6->ip6_src; 1065 inc.inc6_laddr = ip6->ip6_dst; 1066 } else 1067 #endif 1068 { 1069 inc.inc_faddr = ip->ip_src; 1070 inc.inc_laddr = ip->ip_dst; 1071 } 1072 inc.inc_fport = th->th_sport; 1073 inc.inc_lport = th->th_dport; 1074 inc.inc_fibnum = so->so_fibnum; 1075 1076 /* 1077 * Check for an existing connection attempt in syncache if 1078 * the flag is only ACK. A successful lookup creates a new 1079 * socket appended to the listen queue in SYN_RECEIVED state. 1080 */ 1081 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1082 /* 1083 * Parse the TCP options here because 1084 * syncookies need access to the reflected 1085 * timestamp. 1086 */ 1087 tcp_dooptions(&to, optp, optlen, 0); 1088 /* 1089 * NB: syncache_expand() doesn't unlock 1090 * inp and tcpinfo locks. 1091 */ 1092 if (!syncache_expand(&inc, &to, th, &so, m)) { 1093 /* 1094 * No syncache entry or ACK was not 1095 * for our SYN/ACK. Send a RST. 1096 * NB: syncache did its own logging 1097 * of the failure cause. 1098 */ 1099 rstreason = BANDLIM_RST_OPENPORT; 1100 goto dropwithreset; 1101 } 1102 if (so == NULL) { 1103 /* 1104 * We completed the 3-way handshake 1105 * but could not allocate a socket 1106 * either due to memory shortage, 1107 * listen queue length limits or 1108 * global socket limits. Send RST 1109 * or wait and have the remote end 1110 * retransmit the ACK for another 1111 * try. 1112 */ 1113 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1114 log(LOG_DEBUG, "%s; %s: Listen socket: " 1115 "Socket allocation failed due to " 1116 "limits or memory shortage, %s\n", 1117 s, __func__, 1118 V_tcp_sc_rst_sock_fail ? 1119 "sending RST" : "try again"); 1120 if (V_tcp_sc_rst_sock_fail) { 1121 rstreason = BANDLIM_UNLIMITED; 1122 goto dropwithreset; 1123 } else 1124 goto dropunlock; 1125 } 1126 /* 1127 * Socket is created in state SYN_RECEIVED. 1128 * Unlock the listen socket, lock the newly 1129 * created socket and update the tp variable. 1130 */ 1131 INP_WUNLOCK(inp); /* listen socket */ 1132 inp = sotoinpcb(so); 1133 INP_WLOCK(inp); /* new connection */ 1134 tp = intotcpcb(inp); 1135 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1136 ("%s: ", __func__)); 1137 #ifdef TCP_SIGNATURE 1138 if (sig_checked == 0) { 1139 tcp_dooptions(&to, optp, optlen, 1140 (thflags & TH_SYN) ? TO_SYN : 0); 1141 if (!tcp_signature_verify_input(m, off0, tlen, 1142 optlen, &to, th, tp->t_flags)) { 1143 1144 /* 1145 * In SYN_SENT state if it receives an 1146 * RST, it is allowed for further 1147 * processing. 1148 */ 1149 if ((thflags & TH_RST) == 0 || 1150 (tp->t_state == TCPS_SYN_SENT) == 0) 1151 goto dropunlock; 1152 } 1153 sig_checked = 1; 1154 } 1155 #endif 1156 1157 /* 1158 * Process the segment and the data it 1159 * contains. tcp_do_segment() consumes 1160 * the mbuf chain and unlocks the inpcb. 1161 */ 1162 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1163 iptos, ti_locked); 1164 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1165 return; 1166 } 1167 /* 1168 * Segment flag validation for new connection attempts: 1169 * 1170 * Our (SYN|ACK) response was rejected. 1171 * Check with syncache and remove entry to prevent 1172 * retransmits. 1173 * 1174 * NB: syncache_chkrst does its own logging of failure 1175 * causes. 1176 */ 1177 if (thflags & TH_RST) { 1178 syncache_chkrst(&inc, th); 1179 goto dropunlock; 1180 } 1181 /* 1182 * We can't do anything without SYN. 1183 */ 1184 if ((thflags & TH_SYN) == 0) { 1185 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1186 log(LOG_DEBUG, "%s; %s: Listen socket: " 1187 "SYN is missing, segment ignored\n", 1188 s, __func__); 1189 TCPSTAT_INC(tcps_badsyn); 1190 goto dropunlock; 1191 } 1192 /* 1193 * (SYN|ACK) is bogus on a listen socket. 1194 */ 1195 if (thflags & TH_ACK) { 1196 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1197 log(LOG_DEBUG, "%s; %s: Listen socket: " 1198 "SYN|ACK invalid, segment rejected\n", 1199 s, __func__); 1200 syncache_badack(&inc); /* XXX: Not needed! */ 1201 TCPSTAT_INC(tcps_badsyn); 1202 rstreason = BANDLIM_RST_OPENPORT; 1203 goto dropwithreset; 1204 } 1205 /* 1206 * If the drop_synfin option is enabled, drop all 1207 * segments with both the SYN and FIN bits set. 1208 * This prevents e.g. nmap from identifying the 1209 * TCP/IP stack. 1210 * XXX: Poor reasoning. nmap has other methods 1211 * and is constantly refining its stack detection 1212 * strategies. 1213 * XXX: This is a violation of the TCP specification 1214 * and was used by RFC1644. 1215 */ 1216 if ((thflags & TH_FIN) && V_drop_synfin) { 1217 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1218 log(LOG_DEBUG, "%s; %s: Listen socket: " 1219 "SYN|FIN segment ignored (based on " 1220 "sysctl setting)\n", s, __func__); 1221 TCPSTAT_INC(tcps_badsyn); 1222 goto dropunlock; 1223 } 1224 /* 1225 * Segment's flags are (SYN) or (SYN|FIN). 1226 * 1227 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1228 * as they do not affect the state of the TCP FSM. 1229 * The data pointed to by TH_URG and th_urp is ignored. 1230 */ 1231 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1232 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1233 KASSERT(thflags & (TH_SYN), 1234 ("%s: Listen socket: TH_SYN not set", __func__)); 1235 #ifdef INET6 1236 /* 1237 * If deprecated address is forbidden, 1238 * we do not accept SYN to deprecated interface 1239 * address to prevent any new inbound connection from 1240 * getting established. 1241 * When we do not accept SYN, we send a TCP RST, 1242 * with deprecated source address (instead of dropping 1243 * it). We compromise it as it is much better for peer 1244 * to send a RST, and RST will be the final packet 1245 * for the exchange. 1246 * 1247 * If we do not forbid deprecated addresses, we accept 1248 * the SYN packet. RFC2462 does not suggest dropping 1249 * SYN in this case. 1250 * If we decipher RFC2462 5.5.4, it says like this: 1251 * 1. use of deprecated addr with existing 1252 * communication is okay - "SHOULD continue to be 1253 * used" 1254 * 2. use of it with new communication: 1255 * (2a) "SHOULD NOT be used if alternate address 1256 * with sufficient scope is available" 1257 * (2b) nothing mentioned otherwise. 1258 * Here we fall into (2b) case as we have no choice in 1259 * our source address selection - we must obey the peer. 1260 * 1261 * The wording in RFC2462 is confusing, and there are 1262 * multiple description text for deprecated address 1263 * handling - worse, they are not exactly the same. 1264 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1265 */ 1266 if (isipv6 && !V_ip6_use_deprecated) { 1267 struct in6_ifaddr *ia6; 1268 1269 ia6 = ip6_getdstifaddr(m); 1270 if (ia6 != NULL && 1271 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1272 ifa_free(&ia6->ia_ifa); 1273 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1274 log(LOG_DEBUG, "%s; %s: Listen socket: " 1275 "Connection attempt to deprecated " 1276 "IPv6 address rejected\n", 1277 s, __func__); 1278 rstreason = BANDLIM_RST_OPENPORT; 1279 goto dropwithreset; 1280 } 1281 if (ia6) 1282 ifa_free(&ia6->ia_ifa); 1283 } 1284 #endif /* INET6 */ 1285 /* 1286 * Basic sanity checks on incoming SYN requests: 1287 * Don't respond if the destination is a link layer 1288 * broadcast according to RFC1122 4.2.3.10, p. 104. 1289 * If it is from this socket it must be forged. 1290 * Don't respond if the source or destination is a 1291 * global or subnet broad- or multicast address. 1292 * Note that it is quite possible to receive unicast 1293 * link-layer packets with a broadcast IP address. Use 1294 * in_broadcast() to find them. 1295 */ 1296 if (m->m_flags & (M_BCAST|M_MCAST)) { 1297 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1298 log(LOG_DEBUG, "%s; %s: Listen socket: " 1299 "Connection attempt from broad- or multicast " 1300 "link layer address ignored\n", s, __func__); 1301 goto dropunlock; 1302 } 1303 #ifdef INET6 1304 if (isipv6) { 1305 if (th->th_dport == th->th_sport && 1306 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1307 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1308 log(LOG_DEBUG, "%s; %s: Listen socket: " 1309 "Connection attempt to/from self " 1310 "ignored\n", s, __func__); 1311 goto dropunlock; 1312 } 1313 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1314 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1315 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1316 log(LOG_DEBUG, "%s; %s: Listen socket: " 1317 "Connection attempt from/to multicast " 1318 "address ignored\n", s, __func__); 1319 goto dropunlock; 1320 } 1321 } 1322 #endif 1323 #if defined(INET) && defined(INET6) 1324 else 1325 #endif 1326 #ifdef INET 1327 { 1328 if (th->th_dport == th->th_sport && 1329 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1330 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1331 log(LOG_DEBUG, "%s; %s: Listen socket: " 1332 "Connection attempt from/to self " 1333 "ignored\n", s, __func__); 1334 goto dropunlock; 1335 } 1336 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1337 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1338 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1339 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1340 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1341 log(LOG_DEBUG, "%s; %s: Listen socket: " 1342 "Connection attempt from/to broad- " 1343 "or multicast address ignored\n", 1344 s, __func__); 1345 goto dropunlock; 1346 } 1347 } 1348 #endif 1349 /* 1350 * SYN appears to be valid. Create compressed TCP state 1351 * for syncache. 1352 */ 1353 #ifdef TCPDEBUG 1354 if (so->so_options & SO_DEBUG) 1355 tcp_trace(TA_INPUT, ostate, tp, 1356 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1357 #endif 1358 tcp_dooptions(&to, optp, optlen, TO_SYN); 1359 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1360 /* 1361 * Entry added to syncache and mbuf consumed. 1362 * Everything already unlocked by syncache_add(). 1363 */ 1364 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1365 return; 1366 } else if (tp->t_state == TCPS_LISTEN) { 1367 /* 1368 * When a listen socket is torn down the SO_ACCEPTCONN 1369 * flag is removed first while connections are drained 1370 * from the accept queue in a unlock/lock cycle of the 1371 * ACCEPT_LOCK, opening a race condition allowing a SYN 1372 * attempt go through unhandled. 1373 */ 1374 goto dropunlock; 1375 } 1376 1377 #ifdef TCP_SIGNATURE 1378 if (sig_checked == 0) { 1379 tcp_dooptions(&to, optp, optlen, 1380 (thflags & TH_SYN) ? TO_SYN : 0); 1381 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1382 th, tp->t_flags)) { 1383 1384 /* 1385 * In SYN_SENT state if it receives an RST, it is 1386 * allowed for further processing. 1387 */ 1388 if ((thflags & TH_RST) == 0 || 1389 (tp->t_state == TCPS_SYN_SENT) == 0) 1390 goto dropunlock; 1391 } 1392 sig_checked = 1; 1393 } 1394 #endif 1395 1396 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1397 1398 /* 1399 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1400 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1401 * the inpcb, and unlocks pcbinfo. 1402 */ 1403 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1404 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1405 return; 1406 1407 dropwithreset: 1408 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1409 1410 if (ti_locked == TI_WLOCKED) { 1411 INP_INFO_WUNLOCK(&V_tcbinfo); 1412 ti_locked = TI_UNLOCKED; 1413 } 1414 #ifdef INVARIANTS 1415 else { 1416 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1417 "ti_locked: %d", __func__, ti_locked)); 1418 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1419 } 1420 #endif 1421 1422 if (inp != NULL) { 1423 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1424 INP_WUNLOCK(inp); 1425 } else 1426 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1427 m = NULL; /* mbuf chain got consumed. */ 1428 goto drop; 1429 1430 dropunlock: 1431 if (m != NULL) 1432 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 1433 1434 if (ti_locked == TI_WLOCKED) { 1435 INP_INFO_WUNLOCK(&V_tcbinfo); 1436 ti_locked = TI_UNLOCKED; 1437 } 1438 #ifdef INVARIANTS 1439 else { 1440 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1441 "ti_locked: %d", __func__, ti_locked)); 1442 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1443 } 1444 #endif 1445 1446 if (inp != NULL) 1447 INP_WUNLOCK(inp); 1448 1449 drop: 1450 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1451 if (s != NULL) 1452 free(s, M_TCPLOG); 1453 if (m != NULL) 1454 m_freem(m); 1455 } 1456 1457 static void 1458 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1459 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1460 int ti_locked) 1461 { 1462 int thflags, acked, ourfinisacked, needoutput = 0; 1463 int rstreason, todrop, win; 1464 u_long tiwin; 1465 char *s; 1466 struct in_conninfo *inc; 1467 struct mbuf *mfree; 1468 struct tcpopt to; 1469 1470 #ifdef TCPDEBUG 1471 /* 1472 * The size of tcp_saveipgen must be the size of the max ip header, 1473 * now IPv6. 1474 */ 1475 u_char tcp_saveipgen[IP6_HDR_LEN]; 1476 struct tcphdr tcp_savetcp; 1477 short ostate = 0; 1478 #endif 1479 thflags = th->th_flags; 1480 inc = &tp->t_inpcb->inp_inc; 1481 tp->sackhint.last_sack_ack = 0; 1482 1483 /* 1484 * If this is either a state-changing packet or current state isn't 1485 * established, we require a write lock on tcbinfo. Otherwise, we 1486 * allow the tcbinfo to be in either alocked or unlocked, as the 1487 * caller may have unnecessarily acquired a write lock due to a race. 1488 */ 1489 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1490 tp->t_state != TCPS_ESTABLISHED) { 1491 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1492 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1493 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1494 } else { 1495 #ifdef INVARIANTS 1496 if (ti_locked == TI_WLOCKED) 1497 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1498 else { 1499 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1500 "ti_locked: %d", __func__, ti_locked)); 1501 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1502 } 1503 #endif 1504 } 1505 INP_WLOCK_ASSERT(tp->t_inpcb); 1506 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1507 __func__)); 1508 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1509 __func__)); 1510 1511 /* 1512 * Segment received on connection. 1513 * Reset idle time and keep-alive timer. 1514 * XXX: This should be done after segment 1515 * validation to ignore broken/spoofed segs. 1516 */ 1517 tp->t_rcvtime = ticks; 1518 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1519 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1520 1521 /* 1522 * Unscale the window into a 32-bit value. 1523 * For the SYN_SENT state the scale is zero. 1524 */ 1525 tiwin = th->th_win << tp->snd_scale; 1526 1527 /* 1528 * TCP ECN processing. 1529 */ 1530 if (tp->t_flags & TF_ECN_PERMIT) { 1531 if (thflags & TH_CWR) 1532 tp->t_flags &= ~TF_ECN_SND_ECE; 1533 switch (iptos & IPTOS_ECN_MASK) { 1534 case IPTOS_ECN_CE: 1535 tp->t_flags |= TF_ECN_SND_ECE; 1536 TCPSTAT_INC(tcps_ecn_ce); 1537 break; 1538 case IPTOS_ECN_ECT0: 1539 TCPSTAT_INC(tcps_ecn_ect0); 1540 break; 1541 case IPTOS_ECN_ECT1: 1542 TCPSTAT_INC(tcps_ecn_ect1); 1543 break; 1544 } 1545 /* Congestion experienced. */ 1546 if (thflags & TH_ECE) { 1547 cc_cong_signal(tp, th, CC_ECN); 1548 } 1549 } 1550 1551 /* 1552 * Parse options on any incoming segment. 1553 */ 1554 tcp_dooptions(&to, (u_char *)(th + 1), 1555 (th->th_off << 2) - sizeof(struct tcphdr), 1556 (thflags & TH_SYN) ? TO_SYN : 0); 1557 1558 /* 1559 * If echoed timestamp is later than the current time, 1560 * fall back to non RFC1323 RTT calculation. Normalize 1561 * timestamp if syncookies were used when this connection 1562 * was established. 1563 */ 1564 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1565 to.to_tsecr -= tp->ts_offset; 1566 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1567 to.to_tsecr = 0; 1568 } 1569 /* 1570 * If timestamps were negotiated during SYN/ACK they should 1571 * appear on every segment during this session and vice versa. 1572 */ 1573 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 1574 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1575 log(LOG_DEBUG, "%s; %s: Timestamp missing, " 1576 "no action\n", s, __func__); 1577 free(s, M_TCPLOG); 1578 } 1579 } 1580 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 1581 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 1582 log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 1583 "no action\n", s, __func__); 1584 free(s, M_TCPLOG); 1585 } 1586 } 1587 1588 /* 1589 * Process options only when we get SYN/ACK back. The SYN case 1590 * for incoming connections is handled in tcp_syncache. 1591 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1592 * or <SYN,ACK>) segment itself is never scaled. 1593 * XXX this is traditional behavior, may need to be cleaned up. 1594 */ 1595 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1596 if ((to.to_flags & TOF_SCALE) && 1597 (tp->t_flags & TF_REQ_SCALE)) { 1598 tp->t_flags |= TF_RCVD_SCALE; 1599 tp->snd_scale = to.to_wscale; 1600 } 1601 /* 1602 * Initial send window. It will be updated with 1603 * the next incoming segment to the scaled value. 1604 */ 1605 tp->snd_wnd = th->th_win; 1606 if (to.to_flags & TOF_TS) { 1607 tp->t_flags |= TF_RCVD_TSTMP; 1608 tp->ts_recent = to.to_tsval; 1609 tp->ts_recent_age = tcp_ts_getticks(); 1610 } 1611 if (to.to_flags & TOF_MSS) 1612 tcp_mss(tp, to.to_mss); 1613 if ((tp->t_flags & TF_SACK_PERMIT) && 1614 (to.to_flags & TOF_SACKPERM) == 0) 1615 tp->t_flags &= ~TF_SACK_PERMIT; 1616 } 1617 1618 /* 1619 * Header prediction: check for the two common cases 1620 * of a uni-directional data xfer. If the packet has 1621 * no control flags, is in-sequence, the window didn't 1622 * change and we're not retransmitting, it's a 1623 * candidate. If the length is zero and the ack moved 1624 * forward, we're the sender side of the xfer. Just 1625 * free the data acked & wake any higher level process 1626 * that was blocked waiting for space. If the length 1627 * is non-zero and the ack didn't move, we're the 1628 * receiver side. If we're getting packets in-order 1629 * (the reassembly queue is empty), add the data to 1630 * the socket buffer and note that we need a delayed ack. 1631 * Make sure that the hidden state-flags are also off. 1632 * Since we check for TCPS_ESTABLISHED first, it can only 1633 * be TH_NEEDSYN. 1634 */ 1635 if (tp->t_state == TCPS_ESTABLISHED && 1636 th->th_seq == tp->rcv_nxt && 1637 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1638 tp->snd_nxt == tp->snd_max && 1639 tiwin && tiwin == tp->snd_wnd && 1640 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1641 LIST_EMPTY(&tp->t_segq) && 1642 ((to.to_flags & TOF_TS) == 0 || 1643 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1644 1645 /* 1646 * If last ACK falls within this segment's sequence numbers, 1647 * record the timestamp. 1648 * NOTE that the test is modified according to the latest 1649 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1650 */ 1651 if ((to.to_flags & TOF_TS) != 0 && 1652 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1653 tp->ts_recent_age = tcp_ts_getticks(); 1654 tp->ts_recent = to.to_tsval; 1655 } 1656 1657 if (tlen == 0) { 1658 if (SEQ_GT(th->th_ack, tp->snd_una) && 1659 SEQ_LEQ(th->th_ack, tp->snd_max) && 1660 !IN_RECOVERY(tp->t_flags) && 1661 (to.to_flags & TOF_SACK) == 0 && 1662 TAILQ_EMPTY(&tp->snd_holes)) { 1663 /* 1664 * This is a pure ack for outstanding data. 1665 */ 1666 if (ti_locked == TI_WLOCKED) 1667 INP_INFO_WUNLOCK(&V_tcbinfo); 1668 ti_locked = TI_UNLOCKED; 1669 1670 TCPSTAT_INC(tcps_predack); 1671 1672 /* 1673 * "bad retransmit" recovery. 1674 */ 1675 if (tp->t_rxtshift == 1 && 1676 tp->t_flags & TF_PREVVALID && 1677 (int)(ticks - tp->t_badrxtwin) < 0) { 1678 cc_cong_signal(tp, th, CC_RTO_ERR); 1679 } 1680 1681 /* 1682 * Recalculate the transmit timer / rtt. 1683 * 1684 * Some boxes send broken timestamp replies 1685 * during the SYN+ACK phase, ignore 1686 * timestamps of 0 or we could calculate a 1687 * huge RTT and blow up the retransmit timer. 1688 */ 1689 if ((to.to_flags & TOF_TS) != 0 && 1690 to.to_tsecr) { 1691 u_int t; 1692 1693 t = tcp_ts_getticks() - to.to_tsecr; 1694 if (!tp->t_rttlow || tp->t_rttlow > t) 1695 tp->t_rttlow = t; 1696 tcp_xmit_timer(tp, 1697 TCP_TS_TO_TICKS(t) + 1); 1698 } else if (tp->t_rtttime && 1699 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1700 if (!tp->t_rttlow || 1701 tp->t_rttlow > ticks - tp->t_rtttime) 1702 tp->t_rttlow = ticks - tp->t_rtttime; 1703 tcp_xmit_timer(tp, 1704 ticks - tp->t_rtttime); 1705 } 1706 acked = BYTES_THIS_ACK(tp, th); 1707 1708 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1709 hhook_run_tcp_est_in(tp, th, &to); 1710 1711 TCPSTAT_INC(tcps_rcvackpack); 1712 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1713 sbdrop(&so->so_snd, acked); 1714 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1715 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1716 tp->snd_recover = th->th_ack - 1; 1717 1718 /* 1719 * Let the congestion control algorithm update 1720 * congestion control related information. This 1721 * typically means increasing the congestion 1722 * window. 1723 */ 1724 cc_ack_received(tp, th, CC_ACK); 1725 1726 tp->snd_una = th->th_ack; 1727 /* 1728 * Pull snd_wl2 up to prevent seq wrap relative 1729 * to th_ack. 1730 */ 1731 tp->snd_wl2 = th->th_ack; 1732 tp->t_dupacks = 0; 1733 m_freem(m); 1734 ND6_HINT(tp); /* Some progress has been made. */ 1735 1736 /* 1737 * If all outstanding data are acked, stop 1738 * retransmit timer, otherwise restart timer 1739 * using current (possibly backed-off) value. 1740 * If process is waiting for space, 1741 * wakeup/selwakeup/signal. If data 1742 * are ready to send, let tcp_output 1743 * decide between more output or persist. 1744 */ 1745 #ifdef TCPDEBUG 1746 if (so->so_options & SO_DEBUG) 1747 tcp_trace(TA_INPUT, ostate, tp, 1748 (void *)tcp_saveipgen, 1749 &tcp_savetcp, 0); 1750 #endif 1751 if (tp->snd_una == tp->snd_max) 1752 tcp_timer_activate(tp, TT_REXMT, 0); 1753 else if (!tcp_timer_active(tp, TT_PERSIST)) 1754 tcp_timer_activate(tp, TT_REXMT, 1755 tp->t_rxtcur); 1756 sowwakeup(so); 1757 if (so->so_snd.sb_cc) 1758 (void) tcp_output(tp); 1759 goto check_delack; 1760 } 1761 } else if (th->th_ack == tp->snd_una && 1762 tlen <= sbspace(&so->so_rcv)) { 1763 int newsize = 0; /* automatic sockbuf scaling */ 1764 1765 /* 1766 * This is a pure, in-sequence data packet with 1767 * nothing on the reassembly queue and we have enough 1768 * buffer space to take it. 1769 */ 1770 if (ti_locked == TI_WLOCKED) 1771 INP_INFO_WUNLOCK(&V_tcbinfo); 1772 ti_locked = TI_UNLOCKED; 1773 1774 /* Clean receiver SACK report if present */ 1775 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1776 tcp_clean_sackreport(tp); 1777 TCPSTAT_INC(tcps_preddat); 1778 tp->rcv_nxt += tlen; 1779 /* 1780 * Pull snd_wl1 up to prevent seq wrap relative to 1781 * th_seq. 1782 */ 1783 tp->snd_wl1 = th->th_seq; 1784 /* 1785 * Pull rcv_up up to prevent seq wrap relative to 1786 * rcv_nxt. 1787 */ 1788 tp->rcv_up = tp->rcv_nxt; 1789 TCPSTAT_INC(tcps_rcvpack); 1790 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1791 ND6_HINT(tp); /* Some progress has been made */ 1792 #ifdef TCPDEBUG 1793 if (so->so_options & SO_DEBUG) 1794 tcp_trace(TA_INPUT, ostate, tp, 1795 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1796 #endif 1797 /* 1798 * Automatic sizing of receive socket buffer. Often the send 1799 * buffer size is not optimally adjusted to the actual network 1800 * conditions at hand (delay bandwidth product). Setting the 1801 * buffer size too small limits throughput on links with high 1802 * bandwidth and high delay (eg. trans-continental/oceanic links). 1803 * 1804 * On the receive side the socket buffer memory is only rarely 1805 * used to any significant extent. This allows us to be much 1806 * more aggressive in scaling the receive socket buffer. For 1807 * the case that the buffer space is actually used to a large 1808 * extent and we run out of kernel memory we can simply drop 1809 * the new segments; TCP on the sender will just retransmit it 1810 * later. Setting the buffer size too big may only consume too 1811 * much kernel memory if the application doesn't read() from 1812 * the socket or packet loss or reordering makes use of the 1813 * reassembly queue. 1814 * 1815 * The criteria to step up the receive buffer one notch are: 1816 * 1. the number of bytes received during the time it takes 1817 * one timestamp to be reflected back to us (the RTT); 1818 * 2. received bytes per RTT is within seven eighth of the 1819 * current socket buffer size; 1820 * 3. receive buffer size has not hit maximal automatic size; 1821 * 1822 * This algorithm does one step per RTT at most and only if 1823 * we receive a bulk stream w/o packet losses or reorderings. 1824 * Shrinking the buffer during idle times is not necessary as 1825 * it doesn't consume any memory when idle. 1826 * 1827 * TODO: Only step up if the application is actually serving 1828 * the buffer to better manage the socket buffer resources. 1829 */ 1830 if (V_tcp_do_autorcvbuf && 1831 to.to_tsecr && 1832 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1833 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1834 to.to_tsecr - tp->rfbuf_ts < hz) { 1835 if (tp->rfbuf_cnt > 1836 (so->so_rcv.sb_hiwat / 8 * 7) && 1837 so->so_rcv.sb_hiwat < 1838 V_tcp_autorcvbuf_max) { 1839 newsize = 1840 min(so->so_rcv.sb_hiwat + 1841 V_tcp_autorcvbuf_inc, 1842 V_tcp_autorcvbuf_max); 1843 } 1844 /* Start over with next RTT. */ 1845 tp->rfbuf_ts = 0; 1846 tp->rfbuf_cnt = 0; 1847 } else 1848 tp->rfbuf_cnt += tlen; /* add up */ 1849 } 1850 1851 /* Add data to socket buffer. */ 1852 SOCKBUF_LOCK(&so->so_rcv); 1853 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1854 m_freem(m); 1855 } else { 1856 /* 1857 * Set new socket buffer size. 1858 * Give up when limit is reached. 1859 */ 1860 if (newsize) 1861 if (!sbreserve_locked(&so->so_rcv, 1862 newsize, so, NULL)) 1863 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1864 m_adj(m, drop_hdrlen); /* delayed header drop */ 1865 sbappendstream_locked(&so->so_rcv, m); 1866 } 1867 /* NB: sorwakeup_locked() does an implicit unlock. */ 1868 sorwakeup_locked(so); 1869 if (DELAY_ACK(tp, tlen)) { 1870 tp->t_flags |= TF_DELACK; 1871 } else { 1872 tp->t_flags |= TF_ACKNOW; 1873 tcp_output(tp); 1874 } 1875 goto check_delack; 1876 } 1877 } 1878 1879 /* 1880 * Calculate amount of space in receive window, 1881 * and then do TCP input processing. 1882 * Receive window is amount of space in rcv queue, 1883 * but not less than advertised window. 1884 */ 1885 win = sbspace(&so->so_rcv); 1886 if (win < 0) 1887 win = 0; 1888 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1889 1890 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1891 tp->rfbuf_ts = 0; 1892 tp->rfbuf_cnt = 0; 1893 1894 switch (tp->t_state) { 1895 1896 /* 1897 * If the state is SYN_RECEIVED: 1898 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1899 */ 1900 case TCPS_SYN_RECEIVED: 1901 if ((thflags & TH_ACK) && 1902 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1903 SEQ_GT(th->th_ack, tp->snd_max))) { 1904 rstreason = BANDLIM_RST_OPENPORT; 1905 goto dropwithreset; 1906 } 1907 break; 1908 1909 /* 1910 * If the state is SYN_SENT: 1911 * if seg contains an ACK, but not for our SYN, drop the input. 1912 * if seg contains a RST, then drop the connection. 1913 * if seg does not contain SYN, then drop it. 1914 * Otherwise this is an acceptable SYN segment 1915 * initialize tp->rcv_nxt and tp->irs 1916 * if seg contains ack then advance tp->snd_una 1917 * if seg contains an ECE and ECN support is enabled, the stream 1918 * is ECN capable. 1919 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1920 * arrange for segment to be acked (eventually) 1921 * continue processing rest of data/controls, beginning with URG 1922 */ 1923 case TCPS_SYN_SENT: 1924 if ((thflags & TH_ACK) && 1925 (SEQ_LEQ(th->th_ack, tp->iss) || 1926 SEQ_GT(th->th_ack, tp->snd_max))) { 1927 rstreason = BANDLIM_UNLIMITED; 1928 goto dropwithreset; 1929 } 1930 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1931 TCP_PROBE5(connect__refused, NULL, tp, 1932 mtod(m, const char *), tp, th); 1933 tp = tcp_drop(tp, ECONNREFUSED); 1934 } 1935 if (thflags & TH_RST) 1936 goto drop; 1937 if (!(thflags & TH_SYN)) 1938 goto drop; 1939 1940 tp->irs = th->th_seq; 1941 tcp_rcvseqinit(tp); 1942 if (thflags & TH_ACK) { 1943 TCPSTAT_INC(tcps_connects); 1944 soisconnected(so); 1945 #ifdef MAC 1946 mac_socketpeer_set_from_mbuf(m, so); 1947 #endif 1948 /* Do window scaling on this connection? */ 1949 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1950 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1951 tp->rcv_scale = tp->request_r_scale; 1952 } 1953 tp->rcv_adv += imin(tp->rcv_wnd, 1954 TCP_MAXWIN << tp->rcv_scale); 1955 tp->snd_una++; /* SYN is acked */ 1956 /* 1957 * If there's data, delay ACK; if there's also a FIN 1958 * ACKNOW will be turned on later. 1959 */ 1960 if (DELAY_ACK(tp, tlen) && tlen != 0) 1961 tcp_timer_activate(tp, TT_DELACK, 1962 tcp_delacktime); 1963 else 1964 tp->t_flags |= TF_ACKNOW; 1965 1966 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1967 tp->t_flags |= TF_ECN_PERMIT; 1968 TCPSTAT_INC(tcps_ecn_shs); 1969 } 1970 1971 /* 1972 * Received <SYN,ACK> in SYN_SENT[*] state. 1973 * Transitions: 1974 * SYN_SENT --> ESTABLISHED 1975 * SYN_SENT* --> FIN_WAIT_1 1976 */ 1977 tp->t_starttime = ticks; 1978 if (tp->t_flags & TF_NEEDFIN) { 1979 tcp_state_change(tp, TCPS_FIN_WAIT_1); 1980 tp->t_flags &= ~TF_NEEDFIN; 1981 thflags &= ~TH_SYN; 1982 } else { 1983 tcp_state_change(tp, TCPS_ESTABLISHED); 1984 TCP_PROBE5(connect__established, NULL, tp, 1985 mtod(m, const char *), tp, th); 1986 cc_conn_init(tp); 1987 tcp_timer_activate(tp, TT_KEEP, 1988 TP_KEEPIDLE(tp)); 1989 } 1990 } else { 1991 /* 1992 * Received initial SYN in SYN-SENT[*] state => 1993 * simultaneous open. If segment contains CC option 1994 * and there is a cached CC, apply TAO test. 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 * If there was no CC option, clear cached CC value. 2000 */ 2001 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 2002 tcp_timer_activate(tp, TT_REXMT, 0); 2003 tcp_state_change(tp, TCPS_SYN_RECEIVED); 2004 } 2005 2006 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 2007 "ti_locked %d", __func__, ti_locked)); 2008 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2009 INP_WLOCK_ASSERT(tp->t_inpcb); 2010 2011 /* 2012 * Advance th->th_seq to correspond to first data byte. 2013 * If data, trim to stay within window, 2014 * dropping FIN if necessary. 2015 */ 2016 th->th_seq++; 2017 if (tlen > tp->rcv_wnd) { 2018 todrop = tlen - tp->rcv_wnd; 2019 m_adj(m, -todrop); 2020 tlen = tp->rcv_wnd; 2021 thflags &= ~TH_FIN; 2022 TCPSTAT_INC(tcps_rcvpackafterwin); 2023 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2024 } 2025 tp->snd_wl1 = th->th_seq - 1; 2026 tp->rcv_up = th->th_seq; 2027 /* 2028 * Client side of transaction: already sent SYN and data. 2029 * If the remote host used T/TCP to validate the SYN, 2030 * our data will be ACK'd; if so, enter normal data segment 2031 * processing in the middle of step 5, ack processing. 2032 * Otherwise, goto step 6. 2033 */ 2034 if (thflags & TH_ACK) 2035 goto process_ACK; 2036 2037 goto step6; 2038 2039 /* 2040 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2041 * do normal processing. 2042 * 2043 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2044 */ 2045 case TCPS_LAST_ACK: 2046 case TCPS_CLOSING: 2047 break; /* continue normal processing */ 2048 } 2049 2050 /* 2051 * States other than LISTEN or SYN_SENT. 2052 * First check the RST flag and sequence number since reset segments 2053 * are exempt from the timestamp and connection count tests. This 2054 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2055 * below which allowed reset segments in half the sequence space 2056 * to fall though and be processed (which gives forged reset 2057 * segments with a random sequence number a 50 percent chance of 2058 * killing a connection). 2059 * Then check timestamp, if present. 2060 * Then check the connection count, if present. 2061 * Then check that at least some bytes of segment are within 2062 * receive window. If segment begins before rcv_nxt, 2063 * drop leading data (and SYN); if nothing left, just ack. 2064 * 2065 * 2066 * If the RST bit is set, check the sequence number to see 2067 * if this is a valid reset segment. 2068 * RFC 793 page 37: 2069 * In all states except SYN-SENT, all reset (RST) segments 2070 * are validated by checking their SEQ-fields. A reset is 2071 * valid if its sequence number is in the window. 2072 * Note: this does not take into account delayed ACKs, so 2073 * we should test against last_ack_sent instead of rcv_nxt. 2074 * The sequence number in the reset segment is normally an 2075 * echo of our outgoing acknowlegement numbers, but some hosts 2076 * send a reset with the sequence number at the rightmost edge 2077 * of our receive window, and we have to handle this case. 2078 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2079 * that brute force RST attacks are possible. To combat this, 2080 * we use a much stricter check while in the ESTABLISHED state, 2081 * only accepting RSTs where the sequence number is equal to 2082 * last_ack_sent. In all other states (the states in which a 2083 * RST is more likely), the more permissive check is used. 2084 * If we have multiple segments in flight, the initial reset 2085 * segment sequence numbers will be to the left of last_ack_sent, 2086 * but they will eventually catch up. 2087 * In any case, it never made sense to trim reset segments to 2088 * fit the receive window since RFC 1122 says: 2089 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2090 * 2091 * A TCP SHOULD allow a received RST segment to include data. 2092 * 2093 * DISCUSSION 2094 * It has been suggested that a RST segment could contain 2095 * ASCII text that encoded and explained the cause of the 2096 * RST. No standard has yet been established for such 2097 * data. 2098 * 2099 * If the reset segment passes the sequence number test examine 2100 * the state: 2101 * SYN_RECEIVED STATE: 2102 * If passive open, return to LISTEN state. 2103 * If active open, inform user that connection was refused. 2104 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2105 * Inform user that connection was reset, and close tcb. 2106 * CLOSING, LAST_ACK STATES: 2107 * Close the tcb. 2108 * TIME_WAIT STATE: 2109 * Drop the segment - see Stevens, vol. 2, p. 964 and 2110 * RFC 1337. 2111 */ 2112 if (thflags & TH_RST) { 2113 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2114 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2115 switch (tp->t_state) { 2116 2117 case TCPS_SYN_RECEIVED: 2118 so->so_error = ECONNREFUSED; 2119 goto close; 2120 2121 case TCPS_ESTABLISHED: 2122 if (V_tcp_insecure_rst == 0 && 2123 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2124 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2125 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2126 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2127 TCPSTAT_INC(tcps_badrst); 2128 goto drop; 2129 } 2130 /* FALLTHROUGH */ 2131 case TCPS_FIN_WAIT_1: 2132 case TCPS_FIN_WAIT_2: 2133 case TCPS_CLOSE_WAIT: 2134 so->so_error = ECONNRESET; 2135 close: 2136 KASSERT(ti_locked == TI_WLOCKED, 2137 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2138 ti_locked)); 2139 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2140 2141 tcp_state_change(tp, TCPS_CLOSED); 2142 TCPSTAT_INC(tcps_drops); 2143 tp = tcp_close(tp); 2144 break; 2145 2146 case TCPS_CLOSING: 2147 case TCPS_LAST_ACK: 2148 KASSERT(ti_locked == TI_WLOCKED, 2149 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2150 ti_locked)); 2151 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2152 2153 tp = tcp_close(tp); 2154 break; 2155 } 2156 } 2157 goto drop; 2158 } 2159 2160 /* 2161 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2162 * and it's less than ts_recent, drop it. 2163 */ 2164 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2165 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2166 2167 /* Check to see if ts_recent is over 24 days old. */ 2168 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2169 /* 2170 * Invalidate ts_recent. If this segment updates 2171 * ts_recent, the age will be reset later and ts_recent 2172 * will get a valid value. If it does not, setting 2173 * ts_recent to zero will at least satisfy the 2174 * requirement that zero be placed in the timestamp 2175 * echo reply when ts_recent isn't valid. The 2176 * age isn't reset until we get a valid ts_recent 2177 * because we don't want out-of-order segments to be 2178 * dropped when ts_recent is old. 2179 */ 2180 tp->ts_recent = 0; 2181 } else { 2182 TCPSTAT_INC(tcps_rcvduppack); 2183 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2184 TCPSTAT_INC(tcps_pawsdrop); 2185 if (tlen) 2186 goto dropafterack; 2187 goto drop; 2188 } 2189 } 2190 2191 /* 2192 * In the SYN-RECEIVED state, validate that the packet belongs to 2193 * this connection before trimming the data to fit the receive 2194 * window. Check the sequence number versus IRS since we know 2195 * the sequence numbers haven't wrapped. This is a partial fix 2196 * for the "LAND" DoS attack. 2197 */ 2198 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2199 rstreason = BANDLIM_RST_OPENPORT; 2200 goto dropwithreset; 2201 } 2202 2203 todrop = tp->rcv_nxt - th->th_seq; 2204 if (todrop > 0) { 2205 /* 2206 * If this is a duplicate SYN for our current connection, 2207 * advance over it and pretend and it's not a SYN. 2208 */ 2209 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2210 thflags &= ~TH_SYN; 2211 th->th_seq++; 2212 if (th->th_urp > 1) 2213 th->th_urp--; 2214 else 2215 thflags &= ~TH_URG; 2216 todrop--; 2217 } 2218 /* 2219 * Following if statement from Stevens, vol. 2, p. 960. 2220 */ 2221 if (todrop > tlen 2222 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2223 /* 2224 * Any valid FIN must be to the left of the window. 2225 * At this point the FIN must be a duplicate or out 2226 * of sequence; drop it. 2227 */ 2228 thflags &= ~TH_FIN; 2229 2230 /* 2231 * Send an ACK to resynchronize and drop any data. 2232 * But keep on processing for RST or ACK. 2233 */ 2234 tp->t_flags |= TF_ACKNOW; 2235 todrop = tlen; 2236 TCPSTAT_INC(tcps_rcvduppack); 2237 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2238 } else { 2239 TCPSTAT_INC(tcps_rcvpartduppack); 2240 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2241 } 2242 drop_hdrlen += todrop; /* drop from the top afterwards */ 2243 th->th_seq += todrop; 2244 tlen -= todrop; 2245 if (th->th_urp > todrop) 2246 th->th_urp -= todrop; 2247 else { 2248 thflags &= ~TH_URG; 2249 th->th_urp = 0; 2250 } 2251 } 2252 2253 /* 2254 * If new data are received on a connection after the 2255 * user processes are gone, then RST the other end. 2256 */ 2257 if ((so->so_state & SS_NOFDREF) && 2258 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2259 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2260 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2261 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2262 2263 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 2264 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 2265 "after socket was closed, " 2266 "sending RST and removing tcpcb\n", 2267 s, __func__, tcpstates[tp->t_state], tlen); 2268 free(s, M_TCPLOG); 2269 } 2270 tp = tcp_close(tp); 2271 TCPSTAT_INC(tcps_rcvafterclose); 2272 rstreason = BANDLIM_UNLIMITED; 2273 goto dropwithreset; 2274 } 2275 2276 /* 2277 * If segment ends after window, drop trailing data 2278 * (and PUSH and FIN); if nothing left, just ACK. 2279 */ 2280 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2281 if (todrop > 0) { 2282 TCPSTAT_INC(tcps_rcvpackafterwin); 2283 if (todrop >= tlen) { 2284 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2285 /* 2286 * If window is closed can only take segments at 2287 * window edge, and have to drop data and PUSH from 2288 * incoming segments. Continue processing, but 2289 * remember to ack. Otherwise, drop segment 2290 * and ack. 2291 */ 2292 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2293 tp->t_flags |= TF_ACKNOW; 2294 TCPSTAT_INC(tcps_rcvwinprobe); 2295 } else 2296 goto dropafterack; 2297 } else 2298 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2299 m_adj(m, -todrop); 2300 tlen -= todrop; 2301 thflags &= ~(TH_PUSH|TH_FIN); 2302 } 2303 2304 /* 2305 * If last ACK falls within this segment's sequence numbers, 2306 * record its timestamp. 2307 * NOTE: 2308 * 1) That the test incorporates suggestions from the latest 2309 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2310 * 2) That updating only on newer timestamps interferes with 2311 * our earlier PAWS tests, so this check should be solely 2312 * predicated on the sequence space of this segment. 2313 * 3) That we modify the segment boundary check to be 2314 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2315 * instead of RFC1323's 2316 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2317 * This modified check allows us to overcome RFC1323's 2318 * limitations as described in Stevens TCP/IP Illustrated 2319 * Vol. 2 p.869. In such cases, we can still calculate the 2320 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2321 */ 2322 if ((to.to_flags & TOF_TS) != 0 && 2323 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2324 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2325 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2326 tp->ts_recent_age = tcp_ts_getticks(); 2327 tp->ts_recent = to.to_tsval; 2328 } 2329 2330 /* 2331 * If a SYN is in the window, then this is an 2332 * error and we send an RST and drop the connection. 2333 */ 2334 if (thflags & TH_SYN) { 2335 KASSERT(ti_locked == TI_WLOCKED, 2336 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2337 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2338 2339 tp = tcp_drop(tp, ECONNRESET); 2340 rstreason = BANDLIM_UNLIMITED; 2341 goto drop; 2342 } 2343 2344 /* 2345 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2346 * flag is on (half-synchronized state), then queue data for 2347 * later processing; else drop segment and return. 2348 */ 2349 if ((thflags & TH_ACK) == 0) { 2350 if (tp->t_state == TCPS_SYN_RECEIVED || 2351 (tp->t_flags & TF_NEEDSYN)) 2352 goto step6; 2353 else if (tp->t_flags & TF_ACKNOW) 2354 goto dropafterack; 2355 else 2356 goto drop; 2357 } 2358 2359 /* 2360 * Ack processing. 2361 */ 2362 switch (tp->t_state) { 2363 2364 /* 2365 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2366 * ESTABLISHED state and continue processing. 2367 * The ACK was checked above. 2368 */ 2369 case TCPS_SYN_RECEIVED: 2370 2371 TCPSTAT_INC(tcps_connects); 2372 soisconnected(so); 2373 /* Do window scaling? */ 2374 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2375 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2376 tp->rcv_scale = tp->request_r_scale; 2377 tp->snd_wnd = tiwin; 2378 } 2379 /* 2380 * Make transitions: 2381 * SYN-RECEIVED -> ESTABLISHED 2382 * SYN-RECEIVED* -> FIN-WAIT-1 2383 */ 2384 tp->t_starttime = ticks; 2385 if (tp->t_flags & TF_NEEDFIN) { 2386 tcp_state_change(tp, TCPS_FIN_WAIT_1); 2387 tp->t_flags &= ~TF_NEEDFIN; 2388 } else { 2389 tcp_state_change(tp, TCPS_ESTABLISHED); 2390 TCP_PROBE5(accept__established, NULL, tp, 2391 mtod(m, const char *), tp, th); 2392 cc_conn_init(tp); 2393 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2394 } 2395 /* 2396 * If segment contains data or ACK, will call tcp_reass() 2397 * later; if not, do so now to pass queued data to user. 2398 */ 2399 if (tlen == 0 && (thflags & TH_FIN) == 0) 2400 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2401 (struct mbuf *)0); 2402 tp->snd_wl1 = th->th_seq - 1; 2403 /* FALLTHROUGH */ 2404 2405 /* 2406 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2407 * ACKs. If the ack is in the range 2408 * tp->snd_una < th->th_ack <= tp->snd_max 2409 * then advance tp->snd_una to th->th_ack and drop 2410 * data from the retransmission queue. If this ACK reflects 2411 * more up to date window information we update our window information. 2412 */ 2413 case TCPS_ESTABLISHED: 2414 case TCPS_FIN_WAIT_1: 2415 case TCPS_FIN_WAIT_2: 2416 case TCPS_CLOSE_WAIT: 2417 case TCPS_CLOSING: 2418 case TCPS_LAST_ACK: 2419 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2420 TCPSTAT_INC(tcps_rcvacktoomuch); 2421 goto dropafterack; 2422 } 2423 if ((tp->t_flags & TF_SACK_PERMIT) && 2424 ((to.to_flags & TOF_SACK) || 2425 !TAILQ_EMPTY(&tp->snd_holes))) 2426 tcp_sack_doack(tp, &to, th->th_ack); 2427 2428 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2429 hhook_run_tcp_est_in(tp, th, &to); 2430 2431 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2432 if (tlen == 0 && tiwin == tp->snd_wnd && 2433 !(thflags & TH_FIN)) { 2434 TCPSTAT_INC(tcps_rcvdupack); 2435 /* 2436 * If we have outstanding data (other than 2437 * a window probe), this is a completely 2438 * duplicate ack (ie, window info didn't 2439 * change and FIN isn't set), 2440 * the ack is the biggest we've 2441 * seen and we've seen exactly our rexmt 2442 * threshhold of them, assume a packet 2443 * has been dropped and retransmit it. 2444 * Kludge snd_nxt & the congestion 2445 * window so we send only this one 2446 * packet. 2447 * 2448 * We know we're losing at the current 2449 * window size so do congestion avoidance 2450 * (set ssthresh to half the current window 2451 * and pull our congestion window back to 2452 * the new ssthresh). 2453 * 2454 * Dup acks mean that packets have left the 2455 * network (they're now cached at the receiver) 2456 * so bump cwnd by the amount in the receiver 2457 * to keep a constant cwnd packets in the 2458 * network. 2459 * 2460 * When using TCP ECN, notify the peer that 2461 * we reduced the cwnd. 2462 */ 2463 if (!tcp_timer_active(tp, TT_REXMT) || 2464 th->th_ack != tp->snd_una) 2465 tp->t_dupacks = 0; 2466 else if (++tp->t_dupacks > tcprexmtthresh || 2467 IN_FASTRECOVERY(tp->t_flags)) { 2468 cc_ack_received(tp, th, CC_DUPACK); 2469 if ((tp->t_flags & TF_SACK_PERMIT) && 2470 IN_FASTRECOVERY(tp->t_flags)) { 2471 int awnd; 2472 2473 /* 2474 * Compute the amount of data in flight first. 2475 * We can inject new data into the pipe iff 2476 * we have less than 1/2 the original window's 2477 * worth of data in flight. 2478 */ 2479 awnd = (tp->snd_nxt - tp->snd_fack) + 2480 tp->sackhint.sack_bytes_rexmit; 2481 if (awnd < tp->snd_ssthresh) { 2482 tp->snd_cwnd += tp->t_maxseg; 2483 if (tp->snd_cwnd > tp->snd_ssthresh) 2484 tp->snd_cwnd = tp->snd_ssthresh; 2485 } 2486 } else 2487 tp->snd_cwnd += tp->t_maxseg; 2488 if ((thflags & TH_FIN) && 2489 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2490 /* 2491 * If its a fin we need to process 2492 * it to avoid a race where both 2493 * sides enter FIN-WAIT and send FIN|ACK 2494 * at the same time. 2495 */ 2496 break; 2497 } 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 if ((thflags & TH_FIN) && 2538 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2539 /* 2540 * If its a fin we need to process 2541 * it to avoid a race where both 2542 * sides enter FIN-WAIT and send FIN|ACK 2543 * at the same time. 2544 */ 2545 break; 2546 } 2547 (void) tcp_output(tp); 2548 KASSERT(tp->snd_limited <= 2, 2549 ("%s: tp->snd_limited too big", 2550 __func__)); 2551 tp->snd_cwnd = tp->snd_ssthresh + 2552 tp->t_maxseg * 2553 (tp->t_dupacks - tp->snd_limited); 2554 if (SEQ_GT(onxt, tp->snd_nxt)) 2555 tp->snd_nxt = onxt; 2556 goto drop; 2557 } else if (V_tcp_do_rfc3042) { 2558 cc_ack_received(tp, th, CC_DUPACK); 2559 u_long oldcwnd = tp->snd_cwnd; 2560 tcp_seq oldsndmax = tp->snd_max; 2561 u_int sent; 2562 int avail; 2563 2564 KASSERT(tp->t_dupacks == 1 || 2565 tp->t_dupacks == 2, 2566 ("%s: dupacks not 1 or 2", 2567 __func__)); 2568 if (tp->t_dupacks == 1) 2569 tp->snd_limited = 0; 2570 tp->snd_cwnd = 2571 (tp->snd_nxt - tp->snd_una) + 2572 (tp->t_dupacks - tp->snd_limited) * 2573 tp->t_maxseg; 2574 if ((thflags & TH_FIN) && 2575 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2576 /* 2577 * If its a fin we need to process 2578 * it to avoid a race where both 2579 * sides enter FIN-WAIT and send FIN|ACK 2580 * at the same time. 2581 */ 2582 break; 2583 } 2584 /* 2585 * Only call tcp_output when there 2586 * is new data available to be sent. 2587 * Otherwise we would send pure ACKs. 2588 */ 2589 SOCKBUF_LOCK(&so->so_snd); 2590 avail = so->so_snd.sb_cc - 2591 (tp->snd_nxt - tp->snd_una); 2592 SOCKBUF_UNLOCK(&so->so_snd); 2593 if (avail > 0) 2594 (void) tcp_output(tp); 2595 sent = tp->snd_max - oldsndmax; 2596 if (sent > tp->t_maxseg) { 2597 KASSERT((tp->t_dupacks == 2 && 2598 tp->snd_limited == 0) || 2599 (sent == tp->t_maxseg + 1 && 2600 tp->t_flags & TF_SENTFIN), 2601 ("%s: sent too much", 2602 __func__)); 2603 tp->snd_limited = 2; 2604 } else if (sent > 0) 2605 ++tp->snd_limited; 2606 tp->snd_cwnd = oldcwnd; 2607 goto drop; 2608 } 2609 } else 2610 tp->t_dupacks = 0; 2611 break; 2612 } 2613 2614 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2615 ("%s: th_ack <= snd_una", __func__)); 2616 2617 /* 2618 * If the congestion window was inflated to account 2619 * for the other side's cached packets, retract it. 2620 */ 2621 if (IN_FASTRECOVERY(tp->t_flags)) { 2622 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2623 if (tp->t_flags & TF_SACK_PERMIT) 2624 tcp_sack_partialack(tp, th); 2625 else 2626 tcp_newreno_partial_ack(tp, th); 2627 } else 2628 cc_post_recovery(tp, th); 2629 } 2630 tp->t_dupacks = 0; 2631 /* 2632 * If we reach this point, ACK is not a duplicate, 2633 * i.e., it ACKs something we sent. 2634 */ 2635 if (tp->t_flags & TF_NEEDSYN) { 2636 /* 2637 * T/TCP: Connection was half-synchronized, and our 2638 * SYN has been ACK'd (so connection is now fully 2639 * synchronized). Go to non-starred state, 2640 * increment snd_una for ACK of SYN, and check if 2641 * we can do window scaling. 2642 */ 2643 tp->t_flags &= ~TF_NEEDSYN; 2644 tp->snd_una++; 2645 /* Do window scaling? */ 2646 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2647 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2648 tp->rcv_scale = tp->request_r_scale; 2649 /* Send window already scaled. */ 2650 } 2651 } 2652 2653 process_ACK: 2654 INP_WLOCK_ASSERT(tp->t_inpcb); 2655 2656 acked = BYTES_THIS_ACK(tp, th); 2657 TCPSTAT_INC(tcps_rcvackpack); 2658 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2659 2660 /* 2661 * If we just performed our first retransmit, and the ACK 2662 * arrives within our recovery window, then it was a mistake 2663 * to do the retransmit in the first place. Recover our 2664 * original cwnd and ssthresh, and proceed to transmit where 2665 * we left off. 2666 */ 2667 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2668 (int)(ticks - tp->t_badrxtwin) < 0) 2669 cc_cong_signal(tp, th, CC_RTO_ERR); 2670 2671 /* 2672 * If we have a timestamp reply, update smoothed 2673 * round trip time. If no timestamp is present but 2674 * transmit timer is running and timed sequence 2675 * number was acked, update smoothed round trip time. 2676 * Since we now have an rtt measurement, cancel the 2677 * timer backoff (cf., Phil Karn's retransmit alg.). 2678 * Recompute the initial retransmit timer. 2679 * 2680 * Some boxes send broken timestamp replies 2681 * during the SYN+ACK phase, ignore 2682 * timestamps of 0 or we could calculate a 2683 * huge RTT and blow up the retransmit timer. 2684 */ 2685 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2686 u_int t; 2687 2688 t = tcp_ts_getticks() - to.to_tsecr; 2689 if (!tp->t_rttlow || tp->t_rttlow > t) 2690 tp->t_rttlow = t; 2691 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2692 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2693 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2694 tp->t_rttlow = ticks - tp->t_rtttime; 2695 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2696 } 2697 2698 /* 2699 * If all outstanding data is acked, stop retransmit 2700 * timer and remember to restart (more output or persist). 2701 * If there is more data to be acked, restart retransmit 2702 * timer, using current (possibly backed-off) value. 2703 */ 2704 if (th->th_ack == tp->snd_max) { 2705 tcp_timer_activate(tp, TT_REXMT, 0); 2706 needoutput = 1; 2707 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2708 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2709 2710 /* 2711 * If no data (only SYN) was ACK'd, 2712 * skip rest of ACK processing. 2713 */ 2714 if (acked == 0) 2715 goto step6; 2716 2717 /* 2718 * Let the congestion control algorithm update congestion 2719 * control related information. This typically means increasing 2720 * the congestion window. 2721 */ 2722 cc_ack_received(tp, th, CC_ACK); 2723 2724 SOCKBUF_LOCK(&so->so_snd); 2725 if (acked > so->so_snd.sb_cc) { 2726 tp->snd_wnd -= so->so_snd.sb_cc; 2727 mfree = sbcut_locked(&so->so_snd, 2728 (int)so->so_snd.sb_cc); 2729 ourfinisacked = 1; 2730 } else { 2731 mfree = sbcut_locked(&so->so_snd, acked); 2732 tp->snd_wnd -= acked; 2733 ourfinisacked = 0; 2734 } 2735 /* NB: sowwakeup_locked() does an implicit unlock. */ 2736 sowwakeup_locked(so); 2737 m_freem(mfree); 2738 /* Detect una wraparound. */ 2739 if (!IN_RECOVERY(tp->t_flags) && 2740 SEQ_GT(tp->snd_una, tp->snd_recover) && 2741 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2742 tp->snd_recover = th->th_ack - 1; 2743 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2744 if (IN_RECOVERY(tp->t_flags) && 2745 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2746 EXIT_RECOVERY(tp->t_flags); 2747 } 2748 tp->snd_una = th->th_ack; 2749 if (tp->t_flags & TF_SACK_PERMIT) { 2750 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2751 tp->snd_recover = tp->snd_una; 2752 } 2753 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2754 tp->snd_nxt = tp->snd_una; 2755 2756 switch (tp->t_state) { 2757 2758 /* 2759 * In FIN_WAIT_1 STATE in addition to the processing 2760 * for the ESTABLISHED state if our FIN is now acknowledged 2761 * then enter FIN_WAIT_2. 2762 */ 2763 case TCPS_FIN_WAIT_1: 2764 if (ourfinisacked) { 2765 /* 2766 * If we can't receive any more 2767 * data, then closing user can proceed. 2768 * Starting the timer is contrary to the 2769 * specification, but if we don't get a FIN 2770 * we'll hang forever. 2771 * 2772 * XXXjl: 2773 * we should release the tp also, and use a 2774 * compressed state. 2775 */ 2776 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2777 soisdisconnected(so); 2778 tcp_timer_activate(tp, TT_2MSL, 2779 (tcp_fast_finwait2_recycle ? 2780 tcp_finwait2_timeout : 2781 TP_MAXIDLE(tp))); 2782 } 2783 tcp_state_change(tp, TCPS_FIN_WAIT_2); 2784 } 2785 break; 2786 2787 /* 2788 * In CLOSING STATE in addition to the processing for 2789 * the ESTABLISHED state if the ACK acknowledges our FIN 2790 * then enter the TIME-WAIT state, otherwise ignore 2791 * the segment. 2792 */ 2793 case TCPS_CLOSING: 2794 if (ourfinisacked) { 2795 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2796 tcp_twstart(tp); 2797 INP_INFO_WUNLOCK(&V_tcbinfo); 2798 m_freem(m); 2799 return; 2800 } 2801 break; 2802 2803 /* 2804 * In LAST_ACK, we may still be waiting for data to drain 2805 * and/or to be acked, as well as for the ack of our FIN. 2806 * If our FIN is now acknowledged, delete the TCB, 2807 * enter the closed state and return. 2808 */ 2809 case TCPS_LAST_ACK: 2810 if (ourfinisacked) { 2811 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2812 tp = tcp_close(tp); 2813 goto drop; 2814 } 2815 break; 2816 } 2817 } 2818 2819 step6: 2820 INP_WLOCK_ASSERT(tp->t_inpcb); 2821 2822 /* 2823 * Update window information. 2824 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2825 */ 2826 if ((thflags & TH_ACK) && 2827 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2828 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2829 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2830 /* keep track of pure window updates */ 2831 if (tlen == 0 && 2832 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2833 TCPSTAT_INC(tcps_rcvwinupd); 2834 tp->snd_wnd = tiwin; 2835 tp->snd_wl1 = th->th_seq; 2836 tp->snd_wl2 = th->th_ack; 2837 if (tp->snd_wnd > tp->max_sndwnd) 2838 tp->max_sndwnd = tp->snd_wnd; 2839 needoutput = 1; 2840 } 2841 2842 /* 2843 * Process segments with URG. 2844 */ 2845 if ((thflags & TH_URG) && th->th_urp && 2846 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2847 /* 2848 * This is a kludge, but if we receive and accept 2849 * random urgent pointers, we'll crash in 2850 * soreceive. It's hard to imagine someone 2851 * actually wanting to send this much urgent data. 2852 */ 2853 SOCKBUF_LOCK(&so->so_rcv); 2854 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2855 th->th_urp = 0; /* XXX */ 2856 thflags &= ~TH_URG; /* XXX */ 2857 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2858 goto dodata; /* XXX */ 2859 } 2860 /* 2861 * If this segment advances the known urgent pointer, 2862 * then mark the data stream. This should not happen 2863 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2864 * a FIN has been received from the remote side. 2865 * In these states we ignore the URG. 2866 * 2867 * According to RFC961 (Assigned Protocols), 2868 * the urgent pointer points to the last octet 2869 * of urgent data. We continue, however, 2870 * to consider it to indicate the first octet 2871 * of data past the urgent section as the original 2872 * spec states (in one of two places). 2873 */ 2874 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2875 tp->rcv_up = th->th_seq + th->th_urp; 2876 so->so_oobmark = so->so_rcv.sb_cc + 2877 (tp->rcv_up - tp->rcv_nxt) - 1; 2878 if (so->so_oobmark == 0) 2879 so->so_rcv.sb_state |= SBS_RCVATMARK; 2880 sohasoutofband(so); 2881 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2882 } 2883 SOCKBUF_UNLOCK(&so->so_rcv); 2884 /* 2885 * Remove out of band data so doesn't get presented to user. 2886 * This can happen independent of advancing the URG pointer, 2887 * but if two URG's are pending at once, some out-of-band 2888 * data may creep in... ick. 2889 */ 2890 if (th->th_urp <= (u_long)tlen && 2891 !(so->so_options & SO_OOBINLINE)) { 2892 /* hdr drop is delayed */ 2893 tcp_pulloutofband(so, th, m, drop_hdrlen); 2894 } 2895 } else { 2896 /* 2897 * If no out of band data is expected, 2898 * pull receive urgent pointer along 2899 * with the receive window. 2900 */ 2901 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2902 tp->rcv_up = tp->rcv_nxt; 2903 } 2904 dodata: /* XXX */ 2905 INP_WLOCK_ASSERT(tp->t_inpcb); 2906 2907 /* 2908 * Process the segment text, merging it into the TCP sequencing queue, 2909 * and arranging for acknowledgment of receipt if necessary. 2910 * This process logically involves adjusting tp->rcv_wnd as data 2911 * is presented to the user (this happens in tcp_usrreq.c, 2912 * case PRU_RCVD). If a FIN has already been received on this 2913 * connection then we just ignore the text. 2914 */ 2915 if ((tlen || (thflags & TH_FIN)) && 2916 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2917 tcp_seq save_start = th->th_seq; 2918 m_adj(m, drop_hdrlen); /* delayed header drop */ 2919 /* 2920 * Insert segment which includes th into TCP reassembly queue 2921 * with control block tp. Set thflags to whether reassembly now 2922 * includes a segment with FIN. This handles the common case 2923 * inline (segment is the next to be received on an established 2924 * connection, and the queue is empty), avoiding linkage into 2925 * and removal from the queue and repetition of various 2926 * conversions. 2927 * Set DELACK for segments received in order, but ack 2928 * immediately when segments are out of order (so 2929 * fast retransmit can work). 2930 */ 2931 if (th->th_seq == tp->rcv_nxt && 2932 LIST_EMPTY(&tp->t_segq) && 2933 TCPS_HAVEESTABLISHED(tp->t_state)) { 2934 if (DELAY_ACK(tp, tlen)) 2935 tp->t_flags |= TF_DELACK; 2936 else 2937 tp->t_flags |= TF_ACKNOW; 2938 tp->rcv_nxt += tlen; 2939 thflags = th->th_flags & TH_FIN; 2940 TCPSTAT_INC(tcps_rcvpack); 2941 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2942 ND6_HINT(tp); 2943 SOCKBUF_LOCK(&so->so_rcv); 2944 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2945 m_freem(m); 2946 else 2947 sbappendstream_locked(&so->so_rcv, m); 2948 /* NB: sorwakeup_locked() does an implicit unlock. */ 2949 sorwakeup_locked(so); 2950 } else { 2951 /* 2952 * XXX: Due to the header drop above "th" is 2953 * theoretically invalid by now. Fortunately 2954 * m_adj() doesn't actually frees any mbufs 2955 * when trimming from the head. 2956 */ 2957 thflags = tcp_reass(tp, th, &tlen, m); 2958 tp->t_flags |= TF_ACKNOW; 2959 } 2960 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2961 tcp_update_sack_list(tp, save_start, save_start + tlen); 2962 #if 0 2963 /* 2964 * Note the amount of data that peer has sent into 2965 * our window, in order to estimate the sender's 2966 * buffer size. 2967 * XXX: Unused. 2968 */ 2969 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2970 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2971 else 2972 len = so->so_rcv.sb_hiwat; 2973 #endif 2974 } else { 2975 m_freem(m); 2976 thflags &= ~TH_FIN; 2977 } 2978 2979 /* 2980 * If FIN is received ACK the FIN and let the user know 2981 * that the connection is closing. 2982 */ 2983 if (thflags & TH_FIN) { 2984 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2985 socantrcvmore(so); 2986 /* 2987 * If connection is half-synchronized 2988 * (ie NEEDSYN flag on) then delay ACK, 2989 * so it may be piggybacked when SYN is sent. 2990 * Otherwise, since we received a FIN then no 2991 * more input can be expected, send ACK now. 2992 */ 2993 if (tp->t_flags & TF_NEEDSYN) 2994 tp->t_flags |= TF_DELACK; 2995 else 2996 tp->t_flags |= TF_ACKNOW; 2997 tp->rcv_nxt++; 2998 } 2999 switch (tp->t_state) { 3000 3001 /* 3002 * In SYN_RECEIVED and ESTABLISHED STATES 3003 * enter the CLOSE_WAIT state. 3004 */ 3005 case TCPS_SYN_RECEIVED: 3006 tp->t_starttime = ticks; 3007 /* FALLTHROUGH */ 3008 case TCPS_ESTABLISHED: 3009 tcp_state_change(tp, TCPS_CLOSE_WAIT); 3010 break; 3011 3012 /* 3013 * If still in FIN_WAIT_1 STATE FIN has not been acked so 3014 * enter the CLOSING state. 3015 */ 3016 case TCPS_FIN_WAIT_1: 3017 tcp_state_change(tp, TCPS_CLOSING); 3018 break; 3019 3020 /* 3021 * In FIN_WAIT_2 state enter the TIME_WAIT state, 3022 * starting the time-wait timer, turning off the other 3023 * standard timers. 3024 */ 3025 case TCPS_FIN_WAIT_2: 3026 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 3027 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 3028 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 3029 ti_locked)); 3030 3031 tcp_twstart(tp); 3032 INP_INFO_WUNLOCK(&V_tcbinfo); 3033 return; 3034 } 3035 } 3036 if (ti_locked == TI_WLOCKED) 3037 INP_INFO_WUNLOCK(&V_tcbinfo); 3038 ti_locked = TI_UNLOCKED; 3039 3040 #ifdef TCPDEBUG 3041 if (so->so_options & SO_DEBUG) 3042 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3043 &tcp_savetcp, 0); 3044 #endif 3045 3046 /* 3047 * Return any desired output. 3048 */ 3049 if (needoutput || (tp->t_flags & TF_ACKNOW)) 3050 (void) tcp_output(tp); 3051 3052 check_delack: 3053 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3054 __func__, ti_locked)); 3055 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3056 INP_WLOCK_ASSERT(tp->t_inpcb); 3057 3058 if (tp->t_flags & TF_DELACK) { 3059 tp->t_flags &= ~TF_DELACK; 3060 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 3061 } 3062 INP_WUNLOCK(tp->t_inpcb); 3063 return; 3064 3065 dropafterack: 3066 /* 3067 * Generate an ACK dropping incoming segment if it occupies 3068 * sequence space, where the ACK reflects our state. 3069 * 3070 * We can now skip the test for the RST flag since all 3071 * paths to this code happen after packets containing 3072 * RST have been dropped. 3073 * 3074 * In the SYN-RECEIVED state, don't send an ACK unless the 3075 * segment we received passes the SYN-RECEIVED ACK test. 3076 * If it fails send a RST. This breaks the loop in the 3077 * "LAND" DoS attack, and also prevents an ACK storm 3078 * between two listening ports that have been sent forged 3079 * SYN segments, each with the source address of the other. 3080 */ 3081 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3082 (SEQ_GT(tp->snd_una, th->th_ack) || 3083 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3084 rstreason = BANDLIM_RST_OPENPORT; 3085 goto dropwithreset; 3086 } 3087 #ifdef TCPDEBUG 3088 if (so->so_options & SO_DEBUG) 3089 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3090 &tcp_savetcp, 0); 3091 #endif 3092 if (ti_locked == TI_WLOCKED) 3093 INP_INFO_WUNLOCK(&V_tcbinfo); 3094 ti_locked = TI_UNLOCKED; 3095 3096 tp->t_flags |= TF_ACKNOW; 3097 (void) tcp_output(tp); 3098 INP_WUNLOCK(tp->t_inpcb); 3099 m_freem(m); 3100 return; 3101 3102 dropwithreset: 3103 if (ti_locked == TI_WLOCKED) 3104 INP_INFO_WUNLOCK(&V_tcbinfo); 3105 ti_locked = TI_UNLOCKED; 3106 3107 if (tp != NULL) { 3108 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3109 INP_WUNLOCK(tp->t_inpcb); 3110 } else 3111 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3112 return; 3113 3114 drop: 3115 if (ti_locked == TI_WLOCKED) { 3116 INP_INFO_WUNLOCK(&V_tcbinfo); 3117 ti_locked = TI_UNLOCKED; 3118 } 3119 #ifdef INVARIANTS 3120 else 3121 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3122 #endif 3123 3124 /* 3125 * Drop space held by incoming segment and return. 3126 */ 3127 #ifdef TCPDEBUG 3128 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3129 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3130 &tcp_savetcp, 0); 3131 #endif 3132 if (tp != NULL) 3133 INP_WUNLOCK(tp->t_inpcb); 3134 m_freem(m); 3135 } 3136 3137 /* 3138 * Issue RST and make ACK acceptable to originator of segment. 3139 * The mbuf must still include the original packet header. 3140 * tp may be NULL. 3141 */ 3142 static void 3143 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3144 int tlen, int rstreason) 3145 { 3146 #ifdef INET 3147 struct ip *ip; 3148 #endif 3149 #ifdef INET6 3150 struct ip6_hdr *ip6; 3151 #endif 3152 3153 if (tp != NULL) { 3154 INP_WLOCK_ASSERT(tp->t_inpcb); 3155 } 3156 3157 /* Don't bother if destination was broadcast/multicast. */ 3158 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3159 goto drop; 3160 #ifdef INET6 3161 if (mtod(m, struct ip *)->ip_v == 6) { 3162 ip6 = mtod(m, struct ip6_hdr *); 3163 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3164 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3165 goto drop; 3166 /* IPv6 anycast check is done at tcp6_input() */ 3167 } 3168 #endif 3169 #if defined(INET) && defined(INET6) 3170 else 3171 #endif 3172 #ifdef INET 3173 { 3174 ip = mtod(m, struct ip *); 3175 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3176 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3177 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3178 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3179 goto drop; 3180 } 3181 #endif 3182 3183 /* Perform bandwidth limiting. */ 3184 if (badport_bandlim(rstreason) < 0) 3185 goto drop; 3186 3187 /* tcp_respond consumes the mbuf chain. */ 3188 if (th->th_flags & TH_ACK) { 3189 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3190 th->th_ack, TH_RST); 3191 } else { 3192 if (th->th_flags & TH_SYN) 3193 tlen++; 3194 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3195 (tcp_seq)0, TH_RST|TH_ACK); 3196 } 3197 return; 3198 drop: 3199 m_freem(m); 3200 } 3201 3202 /* 3203 * Parse TCP options and place in tcpopt. 3204 */ 3205 static void 3206 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3207 { 3208 int opt, optlen; 3209 3210 to->to_flags = 0; 3211 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3212 opt = cp[0]; 3213 if (opt == TCPOPT_EOL) 3214 break; 3215 if (opt == TCPOPT_NOP) 3216 optlen = 1; 3217 else { 3218 if (cnt < 2) 3219 break; 3220 optlen = cp[1]; 3221 if (optlen < 2 || optlen > cnt) 3222 break; 3223 } 3224 switch (opt) { 3225 case TCPOPT_MAXSEG: 3226 if (optlen != TCPOLEN_MAXSEG) 3227 continue; 3228 if (!(flags & TO_SYN)) 3229 continue; 3230 to->to_flags |= TOF_MSS; 3231 bcopy((char *)cp + 2, 3232 (char *)&to->to_mss, sizeof(to->to_mss)); 3233 to->to_mss = ntohs(to->to_mss); 3234 break; 3235 case TCPOPT_WINDOW: 3236 if (optlen != TCPOLEN_WINDOW) 3237 continue; 3238 if (!(flags & TO_SYN)) 3239 continue; 3240 to->to_flags |= TOF_SCALE; 3241 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3242 break; 3243 case TCPOPT_TIMESTAMP: 3244 if (optlen != TCPOLEN_TIMESTAMP) 3245 continue; 3246 to->to_flags |= TOF_TS; 3247 bcopy((char *)cp + 2, 3248 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3249 to->to_tsval = ntohl(to->to_tsval); 3250 bcopy((char *)cp + 6, 3251 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3252 to->to_tsecr = ntohl(to->to_tsecr); 3253 break; 3254 #ifdef TCP_SIGNATURE 3255 /* 3256 * XXX In order to reply to a host which has set the 3257 * TCP_SIGNATURE option in its initial SYN, we have to 3258 * record the fact that the option was observed here 3259 * for the syncache code to perform the correct response. 3260 */ 3261 case TCPOPT_SIGNATURE: 3262 if (optlen != TCPOLEN_SIGNATURE) 3263 continue; 3264 to->to_flags |= TOF_SIGNATURE; 3265 to->to_signature = cp + 2; 3266 break; 3267 #endif 3268 case TCPOPT_SACK_PERMITTED: 3269 if (optlen != TCPOLEN_SACK_PERMITTED) 3270 continue; 3271 if (!(flags & TO_SYN)) 3272 continue; 3273 if (!V_tcp_do_sack) 3274 continue; 3275 to->to_flags |= TOF_SACKPERM; 3276 break; 3277 case TCPOPT_SACK: 3278 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3279 continue; 3280 if (flags & TO_SYN) 3281 continue; 3282 to->to_flags |= TOF_SACK; 3283 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3284 to->to_sacks = cp + 2; 3285 TCPSTAT_INC(tcps_sack_rcv_blocks); 3286 break; 3287 default: 3288 continue; 3289 } 3290 } 3291 } 3292 3293 /* 3294 * Pull out of band byte out of a segment so 3295 * it doesn't appear in the user's data queue. 3296 * It is still reflected in the segment length for 3297 * sequencing purposes. 3298 */ 3299 static void 3300 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3301 int off) 3302 { 3303 int cnt = off + th->th_urp - 1; 3304 3305 while (cnt >= 0) { 3306 if (m->m_len > cnt) { 3307 char *cp = mtod(m, caddr_t) + cnt; 3308 struct tcpcb *tp = sototcpcb(so); 3309 3310 INP_WLOCK_ASSERT(tp->t_inpcb); 3311 3312 tp->t_iobc = *cp; 3313 tp->t_oobflags |= TCPOOB_HAVEDATA; 3314 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3315 m->m_len--; 3316 if (m->m_flags & M_PKTHDR) 3317 m->m_pkthdr.len--; 3318 return; 3319 } 3320 cnt -= m->m_len; 3321 m = m->m_next; 3322 if (m == NULL) 3323 break; 3324 } 3325 panic("tcp_pulloutofband"); 3326 } 3327 3328 /* 3329 * Collect new round-trip time estimate 3330 * and update averages and current timeout. 3331 */ 3332 static void 3333 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3334 { 3335 int delta; 3336 3337 INP_WLOCK_ASSERT(tp->t_inpcb); 3338 3339 TCPSTAT_INC(tcps_rttupdated); 3340 tp->t_rttupdated++; 3341 if (tp->t_srtt != 0) { 3342 /* 3343 * srtt is stored as fixed point with 5 bits after the 3344 * binary point (i.e., scaled by 8). The following magic 3345 * is equivalent to the smoothing algorithm in rfc793 with 3346 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3347 * point). Adjust rtt to origin 0. 3348 */ 3349 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3350 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3351 3352 if ((tp->t_srtt += delta) <= 0) 3353 tp->t_srtt = 1; 3354 3355 /* 3356 * We accumulate a smoothed rtt variance (actually, a 3357 * smoothed mean difference), then set the retransmit 3358 * timer to smoothed rtt + 4 times the smoothed variance. 3359 * rttvar is stored as fixed point with 4 bits after the 3360 * binary point (scaled by 16). The following is 3361 * equivalent to rfc793 smoothing with an alpha of .75 3362 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3363 * rfc793's wired-in beta. 3364 */ 3365 if (delta < 0) 3366 delta = -delta; 3367 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3368 if ((tp->t_rttvar += delta) <= 0) 3369 tp->t_rttvar = 1; 3370 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3371 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3372 } else { 3373 /* 3374 * No rtt measurement yet - use the unsmoothed rtt. 3375 * Set the variance to half the rtt (so our first 3376 * retransmit happens at 3*rtt). 3377 */ 3378 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3379 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3380 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3381 } 3382 tp->t_rtttime = 0; 3383 tp->t_rxtshift = 0; 3384 3385 /* 3386 * the retransmit should happen at rtt + 4 * rttvar. 3387 * Because of the way we do the smoothing, srtt and rttvar 3388 * will each average +1/2 tick of bias. When we compute 3389 * the retransmit timer, we want 1/2 tick of rounding and 3390 * 1 extra tick because of +-1/2 tick uncertainty in the 3391 * firing of the timer. The bias will give us exactly the 3392 * 1.5 tick we need. But, because the bias is 3393 * statistical, we have to test that we don't drop below 3394 * the minimum feasible timer (which is 2 ticks). 3395 */ 3396 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3397 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3398 3399 /* 3400 * We received an ack for a packet that wasn't retransmitted; 3401 * it is probably safe to discard any error indications we've 3402 * received recently. This isn't quite right, but close enough 3403 * for now (a route might have failed after we sent a segment, 3404 * and the return path might not be symmetrical). 3405 */ 3406 tp->t_softerror = 0; 3407 } 3408 3409 /* 3410 * Determine a reasonable value for maxseg size. 3411 * If the route is known, check route for mtu. 3412 * If none, use an mss that can be handled on the outgoing interface 3413 * without forcing IP to fragment. If no route is found, route has no mtu, 3414 * or the destination isn't local, use a default, hopefully conservative 3415 * size (usually 512 or the default IP max size, but no more than the mtu 3416 * of the interface), as we can't discover anything about intervening 3417 * gateways or networks. We also initialize the congestion/slow start 3418 * window to be a single segment if the destination isn't local. 3419 * While looking at the routing entry, we also initialize other path-dependent 3420 * parameters from pre-set or cached values in the routing entry. 3421 * 3422 * Also take into account the space needed for options that we 3423 * send regularly. Make maxseg shorter by that amount to assure 3424 * that we can send maxseg amount of data even when the options 3425 * are present. Store the upper limit of the length of options plus 3426 * data in maxopd. 3427 * 3428 * NOTE that this routine is only called when we process an incoming 3429 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3430 * settings are handled in tcp_mssopt(). 3431 */ 3432 void 3433 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3434 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3435 { 3436 int mss = 0; 3437 u_long maxmtu = 0; 3438 struct inpcb *inp = tp->t_inpcb; 3439 struct hc_metrics_lite metrics; 3440 int origoffer; 3441 #ifdef INET6 3442 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3443 size_t min_protoh = isipv6 ? 3444 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3445 sizeof (struct tcpiphdr); 3446 #else 3447 const size_t min_protoh = sizeof(struct tcpiphdr); 3448 #endif 3449 3450 INP_WLOCK_ASSERT(tp->t_inpcb); 3451 3452 if (mtuoffer != -1) { 3453 KASSERT(offer == -1, ("%s: conflict", __func__)); 3454 offer = mtuoffer - min_protoh; 3455 } 3456 origoffer = offer; 3457 3458 /* Initialize. */ 3459 #ifdef INET6 3460 if (isipv6) { 3461 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3462 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3463 } 3464 #endif 3465 #if defined(INET) && defined(INET6) 3466 else 3467 #endif 3468 #ifdef INET 3469 { 3470 maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3471 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3472 } 3473 #endif 3474 3475 /* 3476 * No route to sender, stay with default mss and return. 3477 */ 3478 if (maxmtu == 0) { 3479 /* 3480 * In case we return early we need to initialize metrics 3481 * to a defined state as tcp_hc_get() would do for us 3482 * if there was no cache hit. 3483 */ 3484 if (metricptr != NULL) 3485 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3486 return; 3487 } 3488 3489 /* What have we got? */ 3490 switch (offer) { 3491 case 0: 3492 /* 3493 * Offer == 0 means that there was no MSS on the SYN 3494 * segment, in this case we use tcp_mssdflt as 3495 * already assigned to t_maxopd above. 3496 */ 3497 offer = tp->t_maxopd; 3498 break; 3499 3500 case -1: 3501 /* 3502 * Offer == -1 means that we didn't receive SYN yet. 3503 */ 3504 /* FALLTHROUGH */ 3505 3506 default: 3507 /* 3508 * Prevent DoS attack with too small MSS. Round up 3509 * to at least minmss. 3510 */ 3511 offer = max(offer, V_tcp_minmss); 3512 } 3513 3514 /* 3515 * rmx information is now retrieved from tcp_hostcache. 3516 */ 3517 tcp_hc_get(&inp->inp_inc, &metrics); 3518 if (metricptr != NULL) 3519 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3520 3521 /* 3522 * If there's a discovered mtu int tcp hostcache, use it 3523 * else, use the link mtu. 3524 */ 3525 if (metrics.rmx_mtu) 3526 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3527 else { 3528 #ifdef INET6 3529 if (isipv6) { 3530 mss = maxmtu - min_protoh; 3531 if (!V_path_mtu_discovery && 3532 !in6_localaddr(&inp->in6p_faddr)) 3533 mss = min(mss, V_tcp_v6mssdflt); 3534 } 3535 #endif 3536 #if defined(INET) && defined(INET6) 3537 else 3538 #endif 3539 #ifdef INET 3540 { 3541 mss = maxmtu - min_protoh; 3542 if (!V_path_mtu_discovery && 3543 !in_localaddr(inp->inp_faddr)) 3544 mss = min(mss, V_tcp_mssdflt); 3545 } 3546 #endif 3547 /* 3548 * XXX - The above conditional (mss = maxmtu - min_protoh) 3549 * probably violates the TCP spec. 3550 * The problem is that, since we don't know the 3551 * other end's MSS, we are supposed to use a conservative 3552 * default. But, if we do that, then MTU discovery will 3553 * never actually take place, because the conservative 3554 * default is much less than the MTUs typically seen 3555 * on the Internet today. For the moment, we'll sweep 3556 * this under the carpet. 3557 * 3558 * The conservative default might not actually be a problem 3559 * if the only case this occurs is when sending an initial 3560 * SYN with options and data to a host we've never talked 3561 * to before. Then, they will reply with an MSS value which 3562 * will get recorded and the new parameters should get 3563 * recomputed. For Further Study. 3564 */ 3565 } 3566 mss = min(mss, offer); 3567 3568 /* 3569 * Sanity check: make sure that maxopd will be large 3570 * enough to allow some data on segments even if the 3571 * all the option space is used (40bytes). Otherwise 3572 * funny things may happen in tcp_output. 3573 */ 3574 mss = max(mss, 64); 3575 3576 /* 3577 * maxopd stores the maximum length of data AND options 3578 * in a segment; maxseg is the amount of data in a normal 3579 * segment. We need to store this value (maxopd) apart 3580 * from maxseg, because now every segment carries options 3581 * and thus we normally have somewhat less data in segments. 3582 */ 3583 tp->t_maxopd = mss; 3584 3585 /* 3586 * origoffer==-1 indicates that no segments were received yet. 3587 * In this case we just guess. 3588 */ 3589 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3590 (origoffer == -1 || 3591 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3592 mss -= TCPOLEN_TSTAMP_APPA; 3593 3594 tp->t_maxseg = mss; 3595 } 3596 3597 void 3598 tcp_mss(struct tcpcb *tp, int offer) 3599 { 3600 int mss; 3601 u_long bufsize; 3602 struct inpcb *inp; 3603 struct socket *so; 3604 struct hc_metrics_lite metrics; 3605 struct tcp_ifcap cap; 3606 3607 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3608 3609 bzero(&cap, sizeof(cap)); 3610 tcp_mss_update(tp, offer, -1, &metrics, &cap); 3611 3612 mss = tp->t_maxseg; 3613 inp = tp->t_inpcb; 3614 3615 /* 3616 * If there's a pipesize, change the socket buffer to that size, 3617 * don't change if sb_hiwat is different than default (then it 3618 * has been changed on purpose with setsockopt). 3619 * Make the socket buffers an integral number of mss units; 3620 * if the mss is larger than the socket buffer, decrease the mss. 3621 */ 3622 so = inp->inp_socket; 3623 SOCKBUF_LOCK(&so->so_snd); 3624 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3625 bufsize = metrics.rmx_sendpipe; 3626 else 3627 bufsize = so->so_snd.sb_hiwat; 3628 if (bufsize < mss) 3629 mss = bufsize; 3630 else { 3631 bufsize = roundup(bufsize, mss); 3632 if (bufsize > sb_max) 3633 bufsize = sb_max; 3634 if (bufsize > so->so_snd.sb_hiwat) 3635 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3636 } 3637 SOCKBUF_UNLOCK(&so->so_snd); 3638 tp->t_maxseg = mss; 3639 3640 SOCKBUF_LOCK(&so->so_rcv); 3641 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3642 bufsize = metrics.rmx_recvpipe; 3643 else 3644 bufsize = so->so_rcv.sb_hiwat; 3645 if (bufsize > mss) { 3646 bufsize = roundup(bufsize, mss); 3647 if (bufsize > sb_max) 3648 bufsize = sb_max; 3649 if (bufsize > so->so_rcv.sb_hiwat) 3650 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3651 } 3652 SOCKBUF_UNLOCK(&so->so_rcv); 3653 3654 /* Check the interface for TSO capabilities. */ 3655 if (cap.ifcap & CSUM_TSO) { 3656 tp->t_flags |= TF_TSO; 3657 tp->t_tsomax = cap.tsomax; 3658 } 3659 } 3660 3661 /* 3662 * Determine the MSS option to send on an outgoing SYN. 3663 */ 3664 int 3665 tcp_mssopt(struct in_conninfo *inc) 3666 { 3667 int mss = 0; 3668 u_long maxmtu = 0; 3669 u_long thcmtu = 0; 3670 size_t min_protoh; 3671 3672 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3673 3674 #ifdef INET6 3675 if (inc->inc_flags & INC_ISIPV6) { 3676 mss = V_tcp_v6mssdflt; 3677 maxmtu = tcp_maxmtu6(inc, NULL); 3678 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3679 } 3680 #endif 3681 #if defined(INET) && defined(INET6) 3682 else 3683 #endif 3684 #ifdef INET 3685 { 3686 mss = V_tcp_mssdflt; 3687 maxmtu = tcp_maxmtu(inc, NULL); 3688 min_protoh = sizeof(struct tcpiphdr); 3689 } 3690 #endif 3691 #if defined(INET6) || defined(INET) 3692 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3693 #endif 3694 3695 if (maxmtu && thcmtu) 3696 mss = min(maxmtu, thcmtu) - min_protoh; 3697 else if (maxmtu || thcmtu) 3698 mss = max(maxmtu, thcmtu) - min_protoh; 3699 3700 return (mss); 3701 } 3702 3703 3704 /* 3705 * On a partial ack arrives, force the retransmission of the 3706 * next unacknowledged segment. Do not clear tp->t_dupacks. 3707 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3708 * be started again. 3709 */ 3710 static void 3711 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3712 { 3713 tcp_seq onxt = tp->snd_nxt; 3714 u_long ocwnd = tp->snd_cwnd; 3715 3716 INP_WLOCK_ASSERT(tp->t_inpcb); 3717 3718 tcp_timer_activate(tp, TT_REXMT, 0); 3719 tp->t_rtttime = 0; 3720 tp->snd_nxt = th->th_ack; 3721 /* 3722 * Set snd_cwnd to one segment beyond acknowledged offset. 3723 * (tp->snd_una has not yet been updated when this function is called.) 3724 */ 3725 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3726 tp->t_flags |= TF_ACKNOW; 3727 (void) tcp_output(tp); 3728 tp->snd_cwnd = ocwnd; 3729 if (SEQ_GT(onxt, tp->snd_nxt)) 3730 tp->snd_nxt = onxt; 3731 /* 3732 * Partial window deflation. Relies on fact that tp->snd_una 3733 * not updated yet. 3734 */ 3735 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3736 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3737 else 3738 tp->snd_cwnd = 0; 3739 tp->snd_cwnd += tp->t_maxseg; 3740 } 3741