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/signalvar.h> 67 #include <sys/socket.h> 68 #include <sys/socketvar.h> 69 #include <sys/sysctl.h> 70 #include <sys/syslog.h> 71 #include <sys/systm.h> 72 73 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 74 75 #include <vm/uma.h> 76 77 #include <net/if.h> 78 #include <net/route.h> 79 #include <net/vnet.h> 80 81 #define TCPSTATES /* for logging */ 82 83 #include <netinet/cc.h> 84 #include <netinet/in.h> 85 #include <netinet/in_pcb.h> 86 #include <netinet/in_systm.h> 87 #include <netinet/in_var.h> 88 #include <netinet/ip.h> 89 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 90 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 91 #include <netinet/ip_var.h> 92 #include <netinet/ip_options.h> 93 #include <netinet/ip6.h> 94 #include <netinet/icmp6.h> 95 #include <netinet6/in6_pcb.h> 96 #include <netinet6/ip6_var.h> 97 #include <netinet6/nd6.h> 98 #include <netinet/tcp_fsm.h> 99 #include <netinet/tcp_seq.h> 100 #include <netinet/tcp_timer.h> 101 #include <netinet/tcp_var.h> 102 #include <netinet6/tcp6_var.h> 103 #include <netinet/tcpip.h> 104 #include <netinet/tcp_syncache.h> 105 #ifdef TCPDEBUG 106 #include <netinet/tcp_debug.h> 107 #endif /* TCPDEBUG */ 108 #ifdef TCP_OFFLOAD 109 #include <netinet/tcp_offload.h> 110 #endif 111 112 #ifdef IPSEC 113 #include <netipsec/ipsec.h> 114 #include <netipsec/ipsec6.h> 115 #endif /*IPSEC*/ 116 117 #include <machine/in_cksum.h> 118 119 #include <security/mac/mac_framework.h> 120 121 const int tcprexmtthresh = 3; 122 123 VNET_DEFINE(struct tcpstat, tcpstat); 124 SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 125 &VNET_NAME(tcpstat), tcpstat, 126 "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 127 128 int tcp_log_in_vain = 0; 129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 130 &tcp_log_in_vain, 0, 131 "Log all incoming TCP segments to closed ports"); 132 133 VNET_DEFINE(int, blackhole) = 0; 134 #define V_blackhole VNET(blackhole) 135 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 136 &VNET_NAME(blackhole), 0, 137 "Do not send RST on segments to closed ports"); 138 139 VNET_DEFINE(int, tcp_delack_enabled) = 1; 140 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 141 &VNET_NAME(tcp_delack_enabled), 0, 142 "Delay ACK to try and piggyback it onto a data packet"); 143 144 VNET_DEFINE(int, drop_synfin) = 0; 145 #define V_drop_synfin VNET(drop_synfin) 146 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 147 &VNET_NAME(drop_synfin), 0, 148 "Drop TCP packets with SYN+FIN set"); 149 150 VNET_DEFINE(int, tcp_do_rfc3042) = 1; 151 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 152 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 153 &VNET_NAME(tcp_do_rfc3042), 0, 154 "Enable RFC 3042 (Limited Transmit)"); 155 156 VNET_DEFINE(int, tcp_do_rfc3390) = 1; 157 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 158 &VNET_NAME(tcp_do_rfc3390), 0, 159 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 160 161 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 162 "Experimental TCP extensions"); 163 164 VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 165 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 166 &VNET_NAME(tcp_do_initcwnd10), 0, 167 "Enable draft-ietf-tcpm-initcwnd-05 (Increasing initial CWND to 10)"); 168 169 VNET_DEFINE(int, tcp_do_rfc3465) = 1; 170 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 171 &VNET_NAME(tcp_do_rfc3465), 0, 172 "Enable RFC 3465 (Appropriate Byte Counting)"); 173 174 VNET_DEFINE(int, tcp_abc_l_var) = 2; 175 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 176 &VNET_NAME(tcp_abc_l_var), 2, 177 "Cap the max cwnd increment during slow-start to this number of segments"); 178 179 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 180 181 VNET_DEFINE(int, tcp_do_ecn) = 0; 182 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 183 &VNET_NAME(tcp_do_ecn), 0, 184 "TCP ECN support"); 185 186 VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 187 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 188 &VNET_NAME(tcp_ecn_maxretries), 0, 189 "Max retries before giving up on ECN"); 190 191 VNET_DEFINE(int, tcp_insecure_rst) = 0; 192 #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 193 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 194 &VNET_NAME(tcp_insecure_rst), 0, 195 "Follow the old (insecure) criteria for accepting RST packets"); 196 197 VNET_DEFINE(int, tcp_recvspace) = 1024*64; 198 #define V_tcp_recvspace VNET(tcp_recvspace) 199 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 200 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 201 202 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 203 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 204 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 205 &VNET_NAME(tcp_do_autorcvbuf), 0, 206 "Enable automatic receive buffer sizing"); 207 208 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 209 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 210 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 211 &VNET_NAME(tcp_autorcvbuf_inc), 0, 212 "Incrementor step size of automatic receive buffer"); 213 214 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 215 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 216 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 217 &VNET_NAME(tcp_autorcvbuf_max), 0, 218 "Max size of automatic receive buffer"); 219 220 VNET_DEFINE(struct inpcbhead, tcb); 221 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 222 VNET_DEFINE(struct inpcbinfo, tcbinfo); 223 224 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 225 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 226 struct socket *, struct tcpcb *, int, int, uint8_t, 227 int); 228 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 229 struct tcpcb *, int, int); 230 static void tcp_pulloutofband(struct socket *, 231 struct tcphdr *, struct mbuf *, int); 232 static void tcp_xmit_timer(struct tcpcb *, int); 233 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 234 static void inline tcp_fields_to_host(struct tcphdr *); 235 #ifdef TCP_SIGNATURE 236 static void inline tcp_fields_to_net(struct tcphdr *); 237 static int inline tcp_signature_verify_input(struct mbuf *, int, int, 238 int, struct tcpopt *, struct tcphdr *, u_int); 239 #endif 240 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 241 uint16_t type); 242 static void inline cc_conn_init(struct tcpcb *tp); 243 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 244 static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 245 struct tcphdr *th, struct tcpopt *to); 246 247 /* 248 * Kernel module interface for updating tcpstat. The argument is an index 249 * into tcpstat treated as an array of u_long. While this encodes the 250 * general layout of tcpstat into the caller, it doesn't encode its location, 251 * so that future changes to add, for example, per-CPU stats support won't 252 * cause binary compatibility problems for kernel modules. 253 */ 254 void 255 kmod_tcpstat_inc(int statnum) 256 { 257 258 (*((u_long *)&V_tcpstat + statnum))++; 259 } 260 261 /* 262 * Wrapper for the TCP established input helper hook. 263 */ 264 static void inline 265 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 266 { 267 struct tcp_hhook_data hhook_data; 268 269 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 270 hhook_data.tp = tp; 271 hhook_data.th = th; 272 hhook_data.to = to; 273 274 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 275 tp->osd); 276 } 277 } 278 279 /* 280 * CC wrapper hook functions 281 */ 282 static void inline 283 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 284 { 285 INP_WLOCK_ASSERT(tp->t_inpcb); 286 287 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 288 if (tp->snd_cwnd <= tp->snd_wnd) 289 tp->ccv->flags |= CCF_CWND_LIMITED; 290 else 291 tp->ccv->flags &= ~CCF_CWND_LIMITED; 292 293 if (type == CC_ACK) { 294 if (tp->snd_cwnd > tp->snd_ssthresh) { 295 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 296 V_tcp_abc_l_var * tp->t_maxseg); 297 if (tp->t_bytes_acked >= tp->snd_cwnd) { 298 tp->t_bytes_acked -= tp->snd_cwnd; 299 tp->ccv->flags |= CCF_ABC_SENTAWND; 300 } 301 } else { 302 tp->ccv->flags &= ~CCF_ABC_SENTAWND; 303 tp->t_bytes_acked = 0; 304 } 305 } 306 307 if (CC_ALGO(tp)->ack_received != NULL) { 308 /* XXXLAS: Find a way to live without this */ 309 tp->ccv->curack = th->th_ack; 310 CC_ALGO(tp)->ack_received(tp->ccv, type); 311 } 312 } 313 314 static void inline 315 cc_conn_init(struct tcpcb *tp) 316 { 317 struct hc_metrics_lite metrics; 318 struct inpcb *inp = tp->t_inpcb; 319 int rtt; 320 321 INP_WLOCK_ASSERT(tp->t_inpcb); 322 323 tcp_hc_get(&inp->inp_inc, &metrics); 324 325 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 326 tp->t_srtt = rtt; 327 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 328 TCPSTAT_INC(tcps_usedrtt); 329 if (metrics.rmx_rttvar) { 330 tp->t_rttvar = metrics.rmx_rttvar; 331 TCPSTAT_INC(tcps_usedrttvar); 332 } else { 333 /* default variation is +- 1 rtt */ 334 tp->t_rttvar = 335 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 336 } 337 TCPT_RANGESET(tp->t_rxtcur, 338 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 339 tp->t_rttmin, TCPTV_REXMTMAX); 340 } 341 if (metrics.rmx_ssthresh) { 342 /* 343 * There's some sort of gateway or interface 344 * buffer limit on the path. Use this to set 345 * the slow start threshhold, but set the 346 * threshold to no less than 2*mss. 347 */ 348 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 349 TCPSTAT_INC(tcps_usedssthresh); 350 } 351 352 /* 353 * Set the initial slow-start flight size. 354 * 355 * RFC5681 Section 3.1 specifies the default conservative values. 356 * RFC3390 specifies slightly more aggressive values. 357 * Draft-ietf-tcpm-initcwnd-05 increases it to ten segments. 358 * 359 * If a SYN or SYN/ACK was lost and retransmitted, we have to 360 * reduce the initial CWND to one segment as congestion is likely 361 * requiring us to be cautious. 362 */ 363 if (tp->snd_cwnd == 1) 364 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 365 else if (V_tcp_do_initcwnd10) 366 tp->snd_cwnd = min(10 * tp->t_maxseg, 367 max(2 * tp->t_maxseg, 14600)); 368 else if (V_tcp_do_rfc3390) 369 tp->snd_cwnd = min(4 * tp->t_maxseg, 370 max(2 * tp->t_maxseg, 4380)); 371 else { 372 /* Per RFC5681 Section 3.1 */ 373 if (tp->t_maxseg > 2190) 374 tp->snd_cwnd = 2 * tp->t_maxseg; 375 else if (tp->t_maxseg > 1095) 376 tp->snd_cwnd = 3 * tp->t_maxseg; 377 else 378 tp->snd_cwnd = 4 * tp->t_maxseg; 379 } 380 381 if (CC_ALGO(tp)->conn_init != NULL) 382 CC_ALGO(tp)->conn_init(tp->ccv); 383 } 384 385 void inline 386 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 387 { 388 INP_WLOCK_ASSERT(tp->t_inpcb); 389 390 switch(type) { 391 case CC_NDUPACK: 392 if (!IN_FASTRECOVERY(tp->t_flags)) { 393 tp->snd_recover = tp->snd_max; 394 if (tp->t_flags & TF_ECN_PERMIT) 395 tp->t_flags |= TF_ECN_SND_CWR; 396 } 397 break; 398 case CC_ECN: 399 if (!IN_CONGRECOVERY(tp->t_flags)) { 400 TCPSTAT_INC(tcps_ecn_rcwnd); 401 tp->snd_recover = tp->snd_max; 402 if (tp->t_flags & TF_ECN_PERMIT) 403 tp->t_flags |= TF_ECN_SND_CWR; 404 } 405 break; 406 case CC_RTO: 407 tp->t_dupacks = 0; 408 tp->t_bytes_acked = 0; 409 EXIT_RECOVERY(tp->t_flags); 410 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 411 tp->t_maxseg) * tp->t_maxseg; 412 tp->snd_cwnd = tp->t_maxseg; 413 break; 414 case CC_RTO_ERR: 415 TCPSTAT_INC(tcps_sndrexmitbad); 416 /* RTO was unnecessary, so reset everything. */ 417 tp->snd_cwnd = tp->snd_cwnd_prev; 418 tp->snd_ssthresh = tp->snd_ssthresh_prev; 419 tp->snd_recover = tp->snd_recover_prev; 420 if (tp->t_flags & TF_WASFRECOVERY) 421 ENTER_FASTRECOVERY(tp->t_flags); 422 if (tp->t_flags & TF_WASCRECOVERY) 423 ENTER_CONGRECOVERY(tp->t_flags); 424 tp->snd_nxt = tp->snd_max; 425 tp->t_flags &= ~TF_PREVVALID; 426 tp->t_badrxtwin = 0; 427 break; 428 } 429 430 if (CC_ALGO(tp)->cong_signal != NULL) { 431 if (th != NULL) 432 tp->ccv->curack = th->th_ack; 433 CC_ALGO(tp)->cong_signal(tp->ccv, type); 434 } 435 } 436 437 static void inline 438 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 439 { 440 INP_WLOCK_ASSERT(tp->t_inpcb); 441 442 /* XXXLAS: KASSERT that we're in recovery? */ 443 444 if (CC_ALGO(tp)->post_recovery != NULL) { 445 tp->ccv->curack = th->th_ack; 446 CC_ALGO(tp)->post_recovery(tp->ccv); 447 } 448 /* XXXLAS: EXIT_RECOVERY ? */ 449 tp->t_bytes_acked = 0; 450 } 451 452 static inline void 453 tcp_fields_to_host(struct tcphdr *th) 454 { 455 456 th->th_seq = ntohl(th->th_seq); 457 th->th_ack = ntohl(th->th_ack); 458 th->th_win = ntohs(th->th_win); 459 th->th_urp = ntohs(th->th_urp); 460 } 461 462 #ifdef TCP_SIGNATURE 463 static inline void 464 tcp_fields_to_net(struct tcphdr *th) 465 { 466 467 th->th_seq = htonl(th->th_seq); 468 th->th_ack = htonl(th->th_ack); 469 th->th_win = htons(th->th_win); 470 th->th_urp = htons(th->th_urp); 471 } 472 473 static inline int 474 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 475 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 476 { 477 int ret; 478 479 tcp_fields_to_net(th); 480 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 481 tcp_fields_to_host(th); 482 return (ret); 483 } 484 #endif 485 486 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 487 #ifdef INET6 488 #define ND6_HINT(tp) \ 489 do { \ 490 if ((tp) && (tp)->t_inpcb && \ 491 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 492 nd6_nud_hint(NULL, NULL, 0); \ 493 } while (0) 494 #else 495 #define ND6_HINT(tp) 496 #endif 497 498 /* 499 * Indicate whether this ack should be delayed. We can delay the ack if 500 * - there is no delayed ack timer in progress and 501 * - our last ack wasn't a 0-sized window. We never want to delay 502 * the ack that opens up a 0-sized window and 503 * - delayed acks are enabled or 504 * - this is a half-synchronized T/TCP connection. 505 */ 506 #define DELAY_ACK(tp) \ 507 ((!tcp_timer_active(tp, TT_DELACK) && \ 508 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 509 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 510 511 /* 512 * TCP input handling is split into multiple parts: 513 * tcp6_input is a thin wrapper around tcp_input for the extended 514 * ip6_protox[] call format in ip6_input 515 * tcp_input handles primary segment validation, inpcb lookup and 516 * SYN processing on listen sockets 517 * tcp_do_segment processes the ACK and text of the segment for 518 * establishing, established and closing connections 519 */ 520 #ifdef INET6 521 int 522 tcp6_input(struct mbuf **mp, int *offp, int proto) 523 { 524 struct mbuf *m = *mp; 525 struct in6_ifaddr *ia6; 526 527 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 528 529 /* 530 * draft-itojun-ipv6-tcp-to-anycast 531 * better place to put this in? 532 */ 533 ia6 = ip6_getdstifaddr(m); 534 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 535 struct ip6_hdr *ip6; 536 537 ifa_free(&ia6->ia_ifa); 538 ip6 = mtod(m, struct ip6_hdr *); 539 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 540 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 541 return IPPROTO_DONE; 542 } 543 if (ia6) 544 ifa_free(&ia6->ia_ifa); 545 546 tcp_input(m, *offp); 547 return IPPROTO_DONE; 548 } 549 #endif /* INET6 */ 550 551 void 552 tcp_input(struct mbuf *m, int off0) 553 { 554 struct tcphdr *th = NULL; 555 struct ip *ip = NULL; 556 struct inpcb *inp = NULL; 557 struct tcpcb *tp = NULL; 558 struct socket *so = NULL; 559 u_char *optp = NULL; 560 int optlen = 0; 561 #ifdef INET 562 int len; 563 #endif 564 int tlen = 0, off; 565 int drop_hdrlen; 566 int thflags; 567 int rstreason = 0; /* For badport_bandlim accounting purposes */ 568 #ifdef TCP_SIGNATURE 569 uint8_t sig_checked = 0; 570 #endif 571 uint8_t iptos = 0; 572 struct m_tag *fwd_tag = NULL; 573 #ifdef INET6 574 struct ip6_hdr *ip6 = NULL; 575 int isipv6; 576 #else 577 const void *ip6 = NULL; 578 #endif /* INET6 */ 579 struct tcpopt to; /* options in this segment */ 580 char *s = NULL; /* address and port logging */ 581 int ti_locked; 582 #define TI_UNLOCKED 1 583 #define TI_WLOCKED 2 584 585 #ifdef TCPDEBUG 586 /* 587 * The size of tcp_saveipgen must be the size of the max ip header, 588 * now IPv6. 589 */ 590 u_char tcp_saveipgen[IP6_HDR_LEN]; 591 struct tcphdr tcp_savetcp; 592 short ostate = 0; 593 #endif 594 595 #ifdef INET6 596 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 597 #endif 598 599 to.to_flags = 0; 600 TCPSTAT_INC(tcps_rcvtotal); 601 602 #ifdef INET6 603 if (isipv6) { 604 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 605 606 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 607 m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 608 if (m == NULL) { 609 TCPSTAT_INC(tcps_rcvshort); 610 return; 611 } 612 } 613 614 ip6 = mtod(m, struct ip6_hdr *); 615 th = (struct tcphdr *)((caddr_t)ip6 + off0); 616 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 617 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 618 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 619 th->th_sum = m->m_pkthdr.csum_data; 620 else 621 th->th_sum = in6_cksum_pseudo(ip6, tlen, 622 IPPROTO_TCP, m->m_pkthdr.csum_data); 623 th->th_sum ^= 0xffff; 624 } else 625 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 626 if (th->th_sum) { 627 TCPSTAT_INC(tcps_rcvbadsum); 628 goto drop; 629 } 630 631 /* 632 * Be proactive about unspecified IPv6 address in source. 633 * As we use all-zero to indicate unbounded/unconnected pcb, 634 * unspecified IPv6 address can be used to confuse us. 635 * 636 * Note that packets with unspecified IPv6 destination is 637 * already dropped in ip6_input. 638 */ 639 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 640 /* XXX stat */ 641 goto drop; 642 } 643 } 644 #endif 645 #if defined(INET) && defined(INET6) 646 else 647 #endif 648 #ifdef INET 649 { 650 /* 651 * Get IP and TCP header together in first mbuf. 652 * Note: IP leaves IP header in first mbuf. 653 */ 654 if (off0 > sizeof (struct ip)) { 655 ip_stripoptions(m); 656 off0 = sizeof(struct ip); 657 } 658 if (m->m_len < sizeof (struct tcpiphdr)) { 659 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 660 == NULL) { 661 TCPSTAT_INC(tcps_rcvshort); 662 return; 663 } 664 } 665 ip = mtod(m, struct ip *); 666 th = (struct tcphdr *)((caddr_t)ip + off0); 667 tlen = ntohs(ip->ip_len) - off0; 668 669 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 670 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 671 th->th_sum = m->m_pkthdr.csum_data; 672 else 673 th->th_sum = in_pseudo(ip->ip_src.s_addr, 674 ip->ip_dst.s_addr, 675 htonl(m->m_pkthdr.csum_data + tlen + 676 IPPROTO_TCP)); 677 th->th_sum ^= 0xffff; 678 } else { 679 struct ipovly *ipov = (struct ipovly *)ip; 680 681 /* 682 * Checksum extended TCP header and data. 683 */ 684 len = off0 + tlen; 685 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 686 ipov->ih_len = htons(tlen); 687 th->th_sum = in_cksum(m, len); 688 } 689 if (th->th_sum) { 690 TCPSTAT_INC(tcps_rcvbadsum); 691 goto drop; 692 } 693 /* Re-initialization for later version check */ 694 ip->ip_v = IPVERSION; 695 } 696 #endif /* INET */ 697 698 #ifdef INET6 699 if (isipv6) 700 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 701 #endif 702 #if defined(INET) && defined(INET6) 703 else 704 #endif 705 #ifdef INET 706 iptos = ip->ip_tos; 707 #endif 708 709 /* 710 * Check that TCP offset makes sense, 711 * pull out TCP options and adjust length. XXX 712 */ 713 off = th->th_off << 2; 714 if (off < sizeof (struct tcphdr) || off > tlen) { 715 TCPSTAT_INC(tcps_rcvbadoff); 716 goto drop; 717 } 718 tlen -= off; /* tlen is used instead of ti->ti_len */ 719 if (off > sizeof (struct tcphdr)) { 720 #ifdef INET6 721 if (isipv6) { 722 IP6_EXTHDR_CHECK(m, off0, off, ); 723 ip6 = mtod(m, struct ip6_hdr *); 724 th = (struct tcphdr *)((caddr_t)ip6 + off0); 725 } 726 #endif 727 #if defined(INET) && defined(INET6) 728 else 729 #endif 730 #ifdef INET 731 { 732 if (m->m_len < sizeof(struct ip) + off) { 733 if ((m = m_pullup(m, sizeof (struct ip) + off)) 734 == NULL) { 735 TCPSTAT_INC(tcps_rcvshort); 736 return; 737 } 738 ip = mtod(m, struct ip *); 739 th = (struct tcphdr *)((caddr_t)ip + off0); 740 } 741 } 742 #endif 743 optlen = off - sizeof (struct tcphdr); 744 optp = (u_char *)(th + 1); 745 } 746 thflags = th->th_flags; 747 748 /* 749 * Convert TCP protocol specific fields to host format. 750 */ 751 tcp_fields_to_host(th); 752 753 /* 754 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 755 */ 756 drop_hdrlen = off0 + off; 757 758 /* 759 * Locate pcb for segment; if we're likely to add or remove a 760 * connection then first acquire pcbinfo lock. There are two cases 761 * where we might discover later we need a write lock despite the 762 * flags: ACKs moving a connection out of the syncache, and ACKs for 763 * a connection in TIMEWAIT. 764 */ 765 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) { 766 INP_INFO_WLOCK(&V_tcbinfo); 767 ti_locked = TI_WLOCKED; 768 } else 769 ti_locked = TI_UNLOCKED; 770 771 /* 772 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 773 */ 774 if ( 775 #ifdef INET6 776 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 777 #ifdef INET 778 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 779 #endif 780 #endif 781 #if defined(INET) && !defined(INET6) 782 (m->m_flags & M_IP_NEXTHOP) 783 #endif 784 ) 785 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 786 787 findpcb: 788 #ifdef INVARIANTS 789 if (ti_locked == TI_WLOCKED) { 790 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 791 } else { 792 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 793 } 794 #endif 795 #ifdef INET6 796 if (isipv6 && fwd_tag != NULL) { 797 struct sockaddr_in6 *next_hop6; 798 799 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 800 /* 801 * Transparently forwarded. Pretend to be the destination. 802 * Already got one like this? 803 */ 804 inp = in6_pcblookup_mbuf(&V_tcbinfo, 805 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 806 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 807 if (!inp) { 808 /* 809 * It's new. Try to find the ambushing socket. 810 * Because we've rewritten the destination address, 811 * any hardware-generated hash is ignored. 812 */ 813 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 814 th->th_sport, &next_hop6->sin6_addr, 815 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 816 th->th_dport, INPLOOKUP_WILDCARD | 817 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 818 } 819 } else if (isipv6) { 820 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 821 th->th_sport, &ip6->ip6_dst, th->th_dport, 822 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 823 m->m_pkthdr.rcvif, m); 824 } 825 #endif /* INET6 */ 826 #if defined(INET6) && defined(INET) 827 else 828 #endif 829 #ifdef INET 830 if (fwd_tag != NULL) { 831 struct sockaddr_in *next_hop; 832 833 next_hop = (struct sockaddr_in *)(fwd_tag+1); 834 /* 835 * Transparently forwarded. Pretend to be the destination. 836 * already got one like this? 837 */ 838 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 839 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 840 m->m_pkthdr.rcvif, m); 841 if (!inp) { 842 /* 843 * It's new. Try to find the ambushing socket. 844 * Because we've rewritten the destination address, 845 * any hardware-generated hash is ignored. 846 */ 847 inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 848 th->th_sport, next_hop->sin_addr, 849 next_hop->sin_port ? ntohs(next_hop->sin_port) : 850 th->th_dport, INPLOOKUP_WILDCARD | 851 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 852 } 853 } else 854 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 855 th->th_sport, ip->ip_dst, th->th_dport, 856 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 857 m->m_pkthdr.rcvif, m); 858 #endif /* INET */ 859 860 /* 861 * If the INPCB does not exist then all data in the incoming 862 * segment is discarded and an appropriate RST is sent back. 863 * XXX MRT Send RST using which routing table? 864 */ 865 if (inp == NULL) { 866 /* 867 * Log communication attempts to ports that are not 868 * in use. 869 */ 870 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 871 tcp_log_in_vain == 2) { 872 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 873 log(LOG_INFO, "%s; %s: Connection attempt " 874 "to closed port\n", s, __func__); 875 } 876 /* 877 * When blackholing do not respond with a RST but 878 * completely ignore the segment and drop it. 879 */ 880 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 881 V_blackhole == 2) 882 goto dropunlock; 883 884 rstreason = BANDLIM_RST_CLOSEDPORT; 885 goto dropwithreset; 886 } 887 INP_WLOCK_ASSERT(inp); 888 if (!(inp->inp_flags & INP_HW_FLOWID) 889 && (m->m_flags & M_FLOWID) 890 && ((inp->inp_socket == NULL) 891 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 892 inp->inp_flags |= INP_HW_FLOWID; 893 inp->inp_flags &= ~INP_SW_FLOWID; 894 inp->inp_flowid = m->m_pkthdr.flowid; 895 } 896 #ifdef IPSEC 897 #ifdef INET6 898 if (isipv6 && ipsec6_in_reject(m, inp)) { 899 V_ipsec6stat.in_polvio++; 900 goto dropunlock; 901 } else 902 #endif /* INET6 */ 903 if (ipsec4_in_reject(m, inp) != 0) { 904 V_ipsec4stat.in_polvio++; 905 goto dropunlock; 906 } 907 #endif /* IPSEC */ 908 909 /* 910 * Check the minimum TTL for socket. 911 */ 912 if (inp->inp_ip_minttl != 0) { 913 #ifdef INET6 914 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 915 goto dropunlock; 916 else 917 #endif 918 if (inp->inp_ip_minttl > ip->ip_ttl) 919 goto dropunlock; 920 } 921 922 /* 923 * A previous connection in TIMEWAIT state is supposed to catch stray 924 * or duplicate segments arriving late. If this segment was a 925 * legitimate new connection attempt, the old INPCB gets removed and 926 * we can try again to find a listening socket. 927 * 928 * At this point, due to earlier optimism, we may hold only an inpcb 929 * lock, and not the inpcbinfo write lock. If so, we need to try to 930 * acquire it, or if that fails, acquire a reference on the inpcb, 931 * drop all locks, acquire a global write lock, and then re-acquire 932 * the inpcb lock. We may at that point discover that another thread 933 * has tried to free the inpcb, in which case we need to loop back 934 * and try to find a new inpcb to deliver to. 935 * 936 * XXXRW: It may be time to rethink timewait locking. 937 */ 938 relocked: 939 if (inp->inp_flags & INP_TIMEWAIT) { 940 if (ti_locked == TI_UNLOCKED) { 941 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 942 in_pcbref(inp); 943 INP_WUNLOCK(inp); 944 INP_INFO_WLOCK(&V_tcbinfo); 945 ti_locked = TI_WLOCKED; 946 INP_WLOCK(inp); 947 if (in_pcbrele_wlocked(inp)) { 948 inp = NULL; 949 goto findpcb; 950 } 951 } else 952 ti_locked = TI_WLOCKED; 953 } 954 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 955 956 if (thflags & TH_SYN) 957 tcp_dooptions(&to, optp, optlen, TO_SYN); 958 /* 959 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 960 */ 961 if (tcp_twcheck(inp, &to, th, m, tlen)) 962 goto findpcb; 963 INP_INFO_WUNLOCK(&V_tcbinfo); 964 return; 965 } 966 /* 967 * The TCPCB may no longer exist if the connection is winding 968 * down or it is in the CLOSED state. Either way we drop the 969 * segment and send an appropriate response. 970 */ 971 tp = intotcpcb(inp); 972 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 973 rstreason = BANDLIM_RST_CLOSEDPORT; 974 goto dropwithreset; 975 } 976 977 #ifdef TCP_OFFLOAD 978 if (tp->t_flags & TF_TOE) { 979 tcp_offload_input(tp, m); 980 m = NULL; /* consumed by the TOE driver */ 981 goto dropunlock; 982 } 983 #endif 984 985 /* 986 * We've identified a valid inpcb, but it could be that we need an 987 * inpcbinfo write lock but don't hold it. In this case, attempt to 988 * acquire using the same strategy as the TIMEWAIT case above. If we 989 * relock, we have to jump back to 'relocked' as the connection might 990 * now be in TIMEWAIT. 991 */ 992 #ifdef INVARIANTS 993 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) 994 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 995 #endif 996 if (tp->t_state != TCPS_ESTABLISHED) { 997 if (ti_locked == TI_UNLOCKED) { 998 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 999 in_pcbref(inp); 1000 INP_WUNLOCK(inp); 1001 INP_INFO_WLOCK(&V_tcbinfo); 1002 ti_locked = TI_WLOCKED; 1003 INP_WLOCK(inp); 1004 if (in_pcbrele_wlocked(inp)) { 1005 inp = NULL; 1006 goto findpcb; 1007 } 1008 goto relocked; 1009 } else 1010 ti_locked = TI_WLOCKED; 1011 } 1012 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1013 } 1014 1015 #ifdef MAC 1016 INP_WLOCK_ASSERT(inp); 1017 if (mac_inpcb_check_deliver(inp, m)) 1018 goto dropunlock; 1019 #endif 1020 so = inp->inp_socket; 1021 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1022 #ifdef TCPDEBUG 1023 if (so->so_options & SO_DEBUG) { 1024 ostate = tp->t_state; 1025 #ifdef INET6 1026 if (isipv6) { 1027 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1028 } else 1029 #endif 1030 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1031 tcp_savetcp = *th; 1032 } 1033 #endif /* TCPDEBUG */ 1034 /* 1035 * When the socket is accepting connections (the INPCB is in LISTEN 1036 * state) we look into the SYN cache if this is a new connection 1037 * attempt or the completion of a previous one. Because listen 1038 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be 1039 * held in this case. 1040 */ 1041 if (so->so_options & SO_ACCEPTCONN) { 1042 struct in_conninfo inc; 1043 1044 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 1045 "tp not listening", __func__)); 1046 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1047 1048 bzero(&inc, sizeof(inc)); 1049 #ifdef INET6 1050 if (isipv6) { 1051 inc.inc_flags |= INC_ISIPV6; 1052 inc.inc6_faddr = ip6->ip6_src; 1053 inc.inc6_laddr = ip6->ip6_dst; 1054 } else 1055 #endif 1056 { 1057 inc.inc_faddr = ip->ip_src; 1058 inc.inc_laddr = ip->ip_dst; 1059 } 1060 inc.inc_fport = th->th_sport; 1061 inc.inc_lport = th->th_dport; 1062 inc.inc_fibnum = so->so_fibnum; 1063 1064 /* 1065 * Check for an existing connection attempt in syncache if 1066 * the flag is only ACK. A successful lookup creates a new 1067 * socket appended to the listen queue in SYN_RECEIVED state. 1068 */ 1069 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1070 /* 1071 * Parse the TCP options here because 1072 * syncookies need access to the reflected 1073 * timestamp. 1074 */ 1075 tcp_dooptions(&to, optp, optlen, 0); 1076 /* 1077 * NB: syncache_expand() doesn't unlock 1078 * inp and tcpinfo locks. 1079 */ 1080 if (!syncache_expand(&inc, &to, th, &so, m)) { 1081 /* 1082 * No syncache entry or ACK was not 1083 * for our SYN/ACK. Send a RST. 1084 * NB: syncache did its own logging 1085 * of the failure cause. 1086 */ 1087 rstreason = BANDLIM_RST_OPENPORT; 1088 goto dropwithreset; 1089 } 1090 if (so == NULL) { 1091 /* 1092 * We completed the 3-way handshake 1093 * but could not allocate a socket 1094 * either due to memory shortage, 1095 * listen queue length limits or 1096 * global socket limits. Send RST 1097 * or wait and have the remote end 1098 * retransmit the ACK for another 1099 * try. 1100 */ 1101 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1102 log(LOG_DEBUG, "%s; %s: Listen socket: " 1103 "Socket allocation failed due to " 1104 "limits or memory shortage, %s\n", 1105 s, __func__, 1106 V_tcp_sc_rst_sock_fail ? 1107 "sending RST" : "try again"); 1108 if (V_tcp_sc_rst_sock_fail) { 1109 rstreason = BANDLIM_UNLIMITED; 1110 goto dropwithreset; 1111 } else 1112 goto dropunlock; 1113 } 1114 /* 1115 * Socket is created in state SYN_RECEIVED. 1116 * Unlock the listen socket, lock the newly 1117 * created socket and update the tp variable. 1118 */ 1119 INP_WUNLOCK(inp); /* listen socket */ 1120 inp = sotoinpcb(so); 1121 INP_WLOCK(inp); /* new connection */ 1122 tp = intotcpcb(inp); 1123 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 1124 ("%s: ", __func__)); 1125 #ifdef TCP_SIGNATURE 1126 if (sig_checked == 0) { 1127 tcp_dooptions(&to, optp, optlen, 1128 (thflags & TH_SYN) ? TO_SYN : 0); 1129 if (!tcp_signature_verify_input(m, off0, tlen, 1130 optlen, &to, th, tp->t_flags)) { 1131 1132 /* 1133 * In SYN_SENT state if it receives an 1134 * RST, it is allowed for further 1135 * processing. 1136 */ 1137 if ((thflags & TH_RST) == 0 || 1138 (tp->t_state == TCPS_SYN_SENT) == 0) 1139 goto dropunlock; 1140 } 1141 sig_checked = 1; 1142 } 1143 #endif 1144 1145 /* 1146 * Process the segment and the data it 1147 * contains. tcp_do_segment() consumes 1148 * the mbuf chain and unlocks the inpcb. 1149 */ 1150 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1151 iptos, ti_locked); 1152 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1153 return; 1154 } 1155 /* 1156 * Segment flag validation for new connection attempts: 1157 * 1158 * Our (SYN|ACK) response was rejected. 1159 * Check with syncache and remove entry to prevent 1160 * retransmits. 1161 * 1162 * NB: syncache_chkrst does its own logging of failure 1163 * causes. 1164 */ 1165 if (thflags & TH_RST) { 1166 syncache_chkrst(&inc, th); 1167 goto dropunlock; 1168 } 1169 /* 1170 * We can't do anything without SYN. 1171 */ 1172 if ((thflags & TH_SYN) == 0) { 1173 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1174 log(LOG_DEBUG, "%s; %s: Listen socket: " 1175 "SYN is missing, segment ignored\n", 1176 s, __func__); 1177 TCPSTAT_INC(tcps_badsyn); 1178 goto dropunlock; 1179 } 1180 /* 1181 * (SYN|ACK) is bogus on a listen socket. 1182 */ 1183 if (thflags & TH_ACK) { 1184 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1185 log(LOG_DEBUG, "%s; %s: Listen socket: " 1186 "SYN|ACK invalid, segment rejected\n", 1187 s, __func__); 1188 syncache_badack(&inc); /* XXX: Not needed! */ 1189 TCPSTAT_INC(tcps_badsyn); 1190 rstreason = BANDLIM_RST_OPENPORT; 1191 goto dropwithreset; 1192 } 1193 /* 1194 * If the drop_synfin option is enabled, drop all 1195 * segments with both the SYN and FIN bits set. 1196 * This prevents e.g. nmap from identifying the 1197 * TCP/IP stack. 1198 * XXX: Poor reasoning. nmap has other methods 1199 * and is constantly refining its stack detection 1200 * strategies. 1201 * XXX: This is a violation of the TCP specification 1202 * and was used by RFC1644. 1203 */ 1204 if ((thflags & TH_FIN) && V_drop_synfin) { 1205 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1206 log(LOG_DEBUG, "%s; %s: Listen socket: " 1207 "SYN|FIN segment ignored (based on " 1208 "sysctl setting)\n", s, __func__); 1209 TCPSTAT_INC(tcps_badsyn); 1210 goto dropunlock; 1211 } 1212 /* 1213 * Segment's flags are (SYN) or (SYN|FIN). 1214 * 1215 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1216 * as they do not affect the state of the TCP FSM. 1217 * The data pointed to by TH_URG and th_urp is ignored. 1218 */ 1219 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1220 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1221 KASSERT(thflags & (TH_SYN), 1222 ("%s: Listen socket: TH_SYN not set", __func__)); 1223 #ifdef INET6 1224 /* 1225 * If deprecated address is forbidden, 1226 * we do not accept SYN to deprecated interface 1227 * address to prevent any new inbound connection from 1228 * getting established. 1229 * When we do not accept SYN, we send a TCP RST, 1230 * with deprecated source address (instead of dropping 1231 * it). We compromise it as it is much better for peer 1232 * to send a RST, and RST will be the final packet 1233 * for the exchange. 1234 * 1235 * If we do not forbid deprecated addresses, we accept 1236 * the SYN packet. RFC2462 does not suggest dropping 1237 * SYN in this case. 1238 * If we decipher RFC2462 5.5.4, it says like this: 1239 * 1. use of deprecated addr with existing 1240 * communication is okay - "SHOULD continue to be 1241 * used" 1242 * 2. use of it with new communication: 1243 * (2a) "SHOULD NOT be used if alternate address 1244 * with sufficient scope is available" 1245 * (2b) nothing mentioned otherwise. 1246 * Here we fall into (2b) case as we have no choice in 1247 * our source address selection - we must obey the peer. 1248 * 1249 * The wording in RFC2462 is confusing, and there are 1250 * multiple description text for deprecated address 1251 * handling - worse, they are not exactly the same. 1252 * I believe 5.5.4 is the best one, so we follow 5.5.4. 1253 */ 1254 if (isipv6 && !V_ip6_use_deprecated) { 1255 struct in6_ifaddr *ia6; 1256 1257 ia6 = ip6_getdstifaddr(m); 1258 if (ia6 != NULL && 1259 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 1260 ifa_free(&ia6->ia_ifa); 1261 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1262 log(LOG_DEBUG, "%s; %s: Listen socket: " 1263 "Connection attempt to deprecated " 1264 "IPv6 address rejected\n", 1265 s, __func__); 1266 rstreason = BANDLIM_RST_OPENPORT; 1267 goto dropwithreset; 1268 } 1269 if (ia6) 1270 ifa_free(&ia6->ia_ifa); 1271 } 1272 #endif /* INET6 */ 1273 /* 1274 * Basic sanity checks on incoming SYN requests: 1275 * Don't respond if the destination is a link layer 1276 * broadcast according to RFC1122 4.2.3.10, p. 104. 1277 * If it is from this socket it must be forged. 1278 * Don't respond if the source or destination is a 1279 * global or subnet broad- or multicast address. 1280 * Note that it is quite possible to receive unicast 1281 * link-layer packets with a broadcast IP address. Use 1282 * in_broadcast() to find them. 1283 */ 1284 if (m->m_flags & (M_BCAST|M_MCAST)) { 1285 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1286 log(LOG_DEBUG, "%s; %s: Listen socket: " 1287 "Connection attempt from broad- or multicast " 1288 "link layer address ignored\n", s, __func__); 1289 goto dropunlock; 1290 } 1291 #ifdef INET6 1292 if (isipv6) { 1293 if (th->th_dport == th->th_sport && 1294 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1295 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1296 log(LOG_DEBUG, "%s; %s: Listen socket: " 1297 "Connection attempt to/from self " 1298 "ignored\n", s, __func__); 1299 goto dropunlock; 1300 } 1301 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1302 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1304 log(LOG_DEBUG, "%s; %s: Listen socket: " 1305 "Connection attempt from/to multicast " 1306 "address ignored\n", s, __func__); 1307 goto dropunlock; 1308 } 1309 } 1310 #endif 1311 #if defined(INET) && defined(INET6) 1312 else 1313 #endif 1314 #ifdef INET 1315 { 1316 if (th->th_dport == th->th_sport && 1317 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1318 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1319 log(LOG_DEBUG, "%s; %s: Listen socket: " 1320 "Connection attempt from/to self " 1321 "ignored\n", s, __func__); 1322 goto dropunlock; 1323 } 1324 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1325 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 1326 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1327 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1328 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1329 log(LOG_DEBUG, "%s; %s: Listen socket: " 1330 "Connection attempt from/to broad- " 1331 "or multicast address ignored\n", 1332 s, __func__); 1333 goto dropunlock; 1334 } 1335 } 1336 #endif 1337 /* 1338 * SYN appears to be valid. Create compressed TCP state 1339 * for syncache. 1340 */ 1341 #ifdef TCPDEBUG 1342 if (so->so_options & SO_DEBUG) 1343 tcp_trace(TA_INPUT, ostate, tp, 1344 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1345 #endif 1346 tcp_dooptions(&to, optp, optlen, TO_SYN); 1347 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1348 /* 1349 * Entry added to syncache and mbuf consumed. 1350 * Everything already unlocked by syncache_add(). 1351 */ 1352 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1353 return; 1354 } 1355 1356 #ifdef TCP_SIGNATURE 1357 if (sig_checked == 0) { 1358 tcp_dooptions(&to, optp, optlen, 1359 (thflags & TH_SYN) ? TO_SYN : 0); 1360 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 1361 th, tp->t_flags)) { 1362 1363 /* 1364 * In SYN_SENT state if it receives an RST, it is 1365 * allowed for further processing. 1366 */ 1367 if ((thflags & TH_RST) == 0 || 1368 (tp->t_state == TCPS_SYN_SENT) == 0) 1369 goto dropunlock; 1370 } 1371 sig_checked = 1; 1372 } 1373 #endif 1374 1375 /* 1376 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1377 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1378 * the inpcb, and unlocks pcbinfo. 1379 */ 1380 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1381 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1382 return; 1383 1384 dropwithreset: 1385 if (ti_locked == TI_WLOCKED) { 1386 INP_INFO_WUNLOCK(&V_tcbinfo); 1387 ti_locked = TI_UNLOCKED; 1388 } 1389 #ifdef INVARIANTS 1390 else { 1391 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1392 "ti_locked: %d", __func__, ti_locked)); 1393 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1394 } 1395 #endif 1396 1397 if (inp != NULL) { 1398 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1399 INP_WUNLOCK(inp); 1400 } else 1401 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1402 m = NULL; /* mbuf chain got consumed. */ 1403 goto drop; 1404 1405 dropunlock: 1406 if (ti_locked == TI_WLOCKED) { 1407 INP_INFO_WUNLOCK(&V_tcbinfo); 1408 ti_locked = TI_UNLOCKED; 1409 } 1410 #ifdef INVARIANTS 1411 else { 1412 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1413 "ti_locked: %d", __func__, ti_locked)); 1414 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1415 } 1416 #endif 1417 1418 if (inp != NULL) 1419 INP_WUNLOCK(inp); 1420 1421 drop: 1422 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1423 if (s != NULL) 1424 free(s, M_TCPLOG); 1425 if (m != NULL) 1426 m_freem(m); 1427 } 1428 1429 static void 1430 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1431 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1432 int ti_locked) 1433 { 1434 int thflags, acked, ourfinisacked, needoutput = 0; 1435 int rstreason, todrop, win; 1436 u_long tiwin; 1437 struct tcpopt to; 1438 1439 #ifdef TCPDEBUG 1440 /* 1441 * The size of tcp_saveipgen must be the size of the max ip header, 1442 * now IPv6. 1443 */ 1444 u_char tcp_saveipgen[IP6_HDR_LEN]; 1445 struct tcphdr tcp_savetcp; 1446 short ostate = 0; 1447 #endif 1448 thflags = th->th_flags; 1449 tp->sackhint.last_sack_ack = 0; 1450 1451 /* 1452 * If this is either a state-changing packet or current state isn't 1453 * established, we require a write lock on tcbinfo. Otherwise, we 1454 * allow the tcbinfo to be in either alocked or unlocked, as the 1455 * caller may have unnecessarily acquired a write lock due to a race. 1456 */ 1457 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1458 tp->t_state != TCPS_ESTABLISHED) { 1459 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1460 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1461 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1462 } else { 1463 #ifdef INVARIANTS 1464 if (ti_locked == TI_WLOCKED) 1465 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1466 else { 1467 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1468 "ti_locked: %d", __func__, ti_locked)); 1469 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1470 } 1471 #endif 1472 } 1473 INP_WLOCK_ASSERT(tp->t_inpcb); 1474 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1475 __func__)); 1476 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1477 __func__)); 1478 1479 /* 1480 * Segment received on connection. 1481 * Reset idle time and keep-alive timer. 1482 * XXX: This should be done after segment 1483 * validation to ignore broken/spoofed segs. 1484 */ 1485 tp->t_rcvtime = ticks; 1486 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1487 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1488 1489 /* 1490 * Unscale the window into a 32-bit value. 1491 * For the SYN_SENT state the scale is zero. 1492 */ 1493 tiwin = th->th_win << tp->snd_scale; 1494 1495 /* 1496 * TCP ECN processing. 1497 */ 1498 if (tp->t_flags & TF_ECN_PERMIT) { 1499 if (thflags & TH_CWR) 1500 tp->t_flags &= ~TF_ECN_SND_ECE; 1501 switch (iptos & IPTOS_ECN_MASK) { 1502 case IPTOS_ECN_CE: 1503 tp->t_flags |= TF_ECN_SND_ECE; 1504 TCPSTAT_INC(tcps_ecn_ce); 1505 break; 1506 case IPTOS_ECN_ECT0: 1507 TCPSTAT_INC(tcps_ecn_ect0); 1508 break; 1509 case IPTOS_ECN_ECT1: 1510 TCPSTAT_INC(tcps_ecn_ect1); 1511 break; 1512 } 1513 /* Congestion experienced. */ 1514 if (thflags & TH_ECE) { 1515 cc_cong_signal(tp, th, CC_ECN); 1516 } 1517 } 1518 1519 /* 1520 * Parse options on any incoming segment. 1521 */ 1522 tcp_dooptions(&to, (u_char *)(th + 1), 1523 (th->th_off << 2) - sizeof(struct tcphdr), 1524 (thflags & TH_SYN) ? TO_SYN : 0); 1525 1526 /* 1527 * If echoed timestamp is later than the current time, 1528 * fall back to non RFC1323 RTT calculation. Normalize 1529 * timestamp if syncookies were used when this connection 1530 * was established. 1531 */ 1532 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1533 to.to_tsecr -= tp->ts_offset; 1534 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1535 to.to_tsecr = 0; 1536 } 1537 1538 /* 1539 * Process options only when we get SYN/ACK back. The SYN case 1540 * for incoming connections is handled in tcp_syncache. 1541 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1542 * or <SYN,ACK>) segment itself is never scaled. 1543 * XXX this is traditional behavior, may need to be cleaned up. 1544 */ 1545 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1546 if ((to.to_flags & TOF_SCALE) && 1547 (tp->t_flags & TF_REQ_SCALE)) { 1548 tp->t_flags |= TF_RCVD_SCALE; 1549 tp->snd_scale = to.to_wscale; 1550 } 1551 /* 1552 * Initial send window. It will be updated with 1553 * the next incoming segment to the scaled value. 1554 */ 1555 tp->snd_wnd = th->th_win; 1556 if (to.to_flags & TOF_TS) { 1557 tp->t_flags |= TF_RCVD_TSTMP; 1558 tp->ts_recent = to.to_tsval; 1559 tp->ts_recent_age = tcp_ts_getticks(); 1560 } 1561 if (to.to_flags & TOF_MSS) 1562 tcp_mss(tp, to.to_mss); 1563 if ((tp->t_flags & TF_SACK_PERMIT) && 1564 (to.to_flags & TOF_SACKPERM) == 0) 1565 tp->t_flags &= ~TF_SACK_PERMIT; 1566 } 1567 1568 /* 1569 * Header prediction: check for the two common cases 1570 * of a uni-directional data xfer. If the packet has 1571 * no control flags, is in-sequence, the window didn't 1572 * change and we're not retransmitting, it's a 1573 * candidate. If the length is zero and the ack moved 1574 * forward, we're the sender side of the xfer. Just 1575 * free the data acked & wake any higher level process 1576 * that was blocked waiting for space. If the length 1577 * is non-zero and the ack didn't move, we're the 1578 * receiver side. If we're getting packets in-order 1579 * (the reassembly queue is empty), add the data to 1580 * the socket buffer and note that we need a delayed ack. 1581 * Make sure that the hidden state-flags are also off. 1582 * Since we check for TCPS_ESTABLISHED first, it can only 1583 * be TH_NEEDSYN. 1584 */ 1585 if (tp->t_state == TCPS_ESTABLISHED && 1586 th->th_seq == tp->rcv_nxt && 1587 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1588 tp->snd_nxt == tp->snd_max && 1589 tiwin && tiwin == tp->snd_wnd && 1590 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1591 LIST_EMPTY(&tp->t_segq) && 1592 ((to.to_flags & TOF_TS) == 0 || 1593 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1594 1595 /* 1596 * If last ACK falls within this segment's sequence numbers, 1597 * record the timestamp. 1598 * NOTE that the test is modified according to the latest 1599 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1600 */ 1601 if ((to.to_flags & TOF_TS) != 0 && 1602 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1603 tp->ts_recent_age = tcp_ts_getticks(); 1604 tp->ts_recent = to.to_tsval; 1605 } 1606 1607 if (tlen == 0) { 1608 if (SEQ_GT(th->th_ack, tp->snd_una) && 1609 SEQ_LEQ(th->th_ack, tp->snd_max) && 1610 !IN_RECOVERY(tp->t_flags) && 1611 (to.to_flags & TOF_SACK) == 0 && 1612 TAILQ_EMPTY(&tp->snd_holes)) { 1613 /* 1614 * This is a pure ack for outstanding data. 1615 */ 1616 if (ti_locked == TI_WLOCKED) 1617 INP_INFO_WUNLOCK(&V_tcbinfo); 1618 ti_locked = TI_UNLOCKED; 1619 1620 TCPSTAT_INC(tcps_predack); 1621 1622 /* 1623 * "bad retransmit" recovery. 1624 */ 1625 if (tp->t_rxtshift == 1 && 1626 tp->t_flags & TF_PREVVALID && 1627 (int)(ticks - tp->t_badrxtwin) < 0) { 1628 cc_cong_signal(tp, th, CC_RTO_ERR); 1629 } 1630 1631 /* 1632 * Recalculate the transmit timer / rtt. 1633 * 1634 * Some boxes send broken timestamp replies 1635 * during the SYN+ACK phase, ignore 1636 * timestamps of 0 or we could calculate a 1637 * huge RTT and blow up the retransmit timer. 1638 */ 1639 if ((to.to_flags & TOF_TS) != 0 && 1640 to.to_tsecr) { 1641 u_int t; 1642 1643 t = tcp_ts_getticks() - to.to_tsecr; 1644 if (!tp->t_rttlow || tp->t_rttlow > t) 1645 tp->t_rttlow = t; 1646 tcp_xmit_timer(tp, 1647 TCP_TS_TO_TICKS(t) + 1); 1648 } else if (tp->t_rtttime && 1649 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1650 if (!tp->t_rttlow || 1651 tp->t_rttlow > ticks - tp->t_rtttime) 1652 tp->t_rttlow = ticks - tp->t_rtttime; 1653 tcp_xmit_timer(tp, 1654 ticks - tp->t_rtttime); 1655 } 1656 acked = BYTES_THIS_ACK(tp, th); 1657 1658 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 1659 hhook_run_tcp_est_in(tp, th, &to); 1660 1661 TCPSTAT_INC(tcps_rcvackpack); 1662 TCPSTAT_ADD(tcps_rcvackbyte, acked); 1663 sbdrop(&so->so_snd, acked); 1664 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1665 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1666 tp->snd_recover = th->th_ack - 1; 1667 1668 /* 1669 * Let the congestion control algorithm update 1670 * congestion control related information. This 1671 * typically means increasing the congestion 1672 * window. 1673 */ 1674 cc_ack_received(tp, th, CC_ACK); 1675 1676 tp->snd_una = th->th_ack; 1677 /* 1678 * Pull snd_wl2 up to prevent seq wrap relative 1679 * to th_ack. 1680 */ 1681 tp->snd_wl2 = th->th_ack; 1682 tp->t_dupacks = 0; 1683 m_freem(m); 1684 ND6_HINT(tp); /* Some progress has been made. */ 1685 1686 /* 1687 * If all outstanding data are acked, stop 1688 * retransmit timer, otherwise restart timer 1689 * using current (possibly backed-off) value. 1690 * If process is waiting for space, 1691 * wakeup/selwakeup/signal. If data 1692 * are ready to send, let tcp_output 1693 * decide between more output or persist. 1694 */ 1695 #ifdef TCPDEBUG 1696 if (so->so_options & SO_DEBUG) 1697 tcp_trace(TA_INPUT, ostate, tp, 1698 (void *)tcp_saveipgen, 1699 &tcp_savetcp, 0); 1700 #endif 1701 if (tp->snd_una == tp->snd_max) 1702 tcp_timer_activate(tp, TT_REXMT, 0); 1703 else if (!tcp_timer_active(tp, TT_PERSIST)) 1704 tcp_timer_activate(tp, TT_REXMT, 1705 tp->t_rxtcur); 1706 sowwakeup(so); 1707 if (so->so_snd.sb_cc) 1708 (void) tcp_output(tp); 1709 goto check_delack; 1710 } 1711 } else if (th->th_ack == tp->snd_una && 1712 tlen <= sbspace(&so->so_rcv)) { 1713 int newsize = 0; /* automatic sockbuf scaling */ 1714 1715 /* 1716 * This is a pure, in-sequence data packet with 1717 * nothing on the reassembly queue and we have enough 1718 * buffer space to take it. 1719 */ 1720 if (ti_locked == TI_WLOCKED) 1721 INP_INFO_WUNLOCK(&V_tcbinfo); 1722 ti_locked = TI_UNLOCKED; 1723 1724 /* Clean receiver SACK report if present */ 1725 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1726 tcp_clean_sackreport(tp); 1727 TCPSTAT_INC(tcps_preddat); 1728 tp->rcv_nxt += tlen; 1729 /* 1730 * Pull snd_wl1 up to prevent seq wrap relative to 1731 * th_seq. 1732 */ 1733 tp->snd_wl1 = th->th_seq; 1734 /* 1735 * Pull rcv_up up to prevent seq wrap relative to 1736 * rcv_nxt. 1737 */ 1738 tp->rcv_up = tp->rcv_nxt; 1739 TCPSTAT_INC(tcps_rcvpack); 1740 TCPSTAT_ADD(tcps_rcvbyte, tlen); 1741 ND6_HINT(tp); /* Some progress has been made */ 1742 #ifdef TCPDEBUG 1743 if (so->so_options & SO_DEBUG) 1744 tcp_trace(TA_INPUT, ostate, tp, 1745 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1746 #endif 1747 /* 1748 * Automatic sizing of receive socket buffer. Often the send 1749 * buffer size is not optimally adjusted to the actual network 1750 * conditions at hand (delay bandwidth product). Setting the 1751 * buffer size too small limits throughput on links with high 1752 * bandwidth and high delay (eg. trans-continental/oceanic links). 1753 * 1754 * On the receive side the socket buffer memory is only rarely 1755 * used to any significant extent. This allows us to be much 1756 * more aggressive in scaling the receive socket buffer. For 1757 * the case that the buffer space is actually used to a large 1758 * extent and we run out of kernel memory we can simply drop 1759 * the new segments; TCP on the sender will just retransmit it 1760 * later. Setting the buffer size too big may only consume too 1761 * much kernel memory if the application doesn't read() from 1762 * the socket or packet loss or reordering makes use of the 1763 * reassembly queue. 1764 * 1765 * The criteria to step up the receive buffer one notch are: 1766 * 1. the number of bytes received during the time it takes 1767 * one timestamp to be reflected back to us (the RTT); 1768 * 2. received bytes per RTT is within seven eighth of the 1769 * current socket buffer size; 1770 * 3. receive buffer size has not hit maximal automatic size; 1771 * 1772 * This algorithm does one step per RTT at most and only if 1773 * we receive a bulk stream w/o packet losses or reorderings. 1774 * Shrinking the buffer during idle times is not necessary as 1775 * it doesn't consume any memory when idle. 1776 * 1777 * TODO: Only step up if the application is actually serving 1778 * the buffer to better manage the socket buffer resources. 1779 */ 1780 if (V_tcp_do_autorcvbuf && 1781 to.to_tsecr && 1782 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1783 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 1784 to.to_tsecr - tp->rfbuf_ts < hz) { 1785 if (tp->rfbuf_cnt > 1786 (so->so_rcv.sb_hiwat / 8 * 7) && 1787 so->so_rcv.sb_hiwat < 1788 V_tcp_autorcvbuf_max) { 1789 newsize = 1790 min(so->so_rcv.sb_hiwat + 1791 V_tcp_autorcvbuf_inc, 1792 V_tcp_autorcvbuf_max); 1793 } 1794 /* Start over with next RTT. */ 1795 tp->rfbuf_ts = 0; 1796 tp->rfbuf_cnt = 0; 1797 } else 1798 tp->rfbuf_cnt += tlen; /* add up */ 1799 } 1800 1801 /* Add data to socket buffer. */ 1802 SOCKBUF_LOCK(&so->so_rcv); 1803 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1804 m_freem(m); 1805 } else { 1806 /* 1807 * Set new socket buffer size. 1808 * Give up when limit is reached. 1809 */ 1810 if (newsize) 1811 if (!sbreserve_locked(&so->so_rcv, 1812 newsize, so, NULL)) 1813 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1814 m_adj(m, drop_hdrlen); /* delayed header drop */ 1815 sbappendstream_locked(&so->so_rcv, m); 1816 } 1817 /* NB: sorwakeup_locked() does an implicit unlock. */ 1818 sorwakeup_locked(so); 1819 if (DELAY_ACK(tp)) { 1820 tp->t_flags |= TF_DELACK; 1821 } else { 1822 tp->t_flags |= TF_ACKNOW; 1823 tcp_output(tp); 1824 } 1825 goto check_delack; 1826 } 1827 } 1828 1829 /* 1830 * Calculate amount of space in receive window, 1831 * and then do TCP input processing. 1832 * Receive window is amount of space in rcv queue, 1833 * but not less than advertised window. 1834 */ 1835 win = sbspace(&so->so_rcv); 1836 if (win < 0) 1837 win = 0; 1838 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1839 1840 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1841 tp->rfbuf_ts = 0; 1842 tp->rfbuf_cnt = 0; 1843 1844 switch (tp->t_state) { 1845 1846 /* 1847 * If the state is SYN_RECEIVED: 1848 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1849 */ 1850 case TCPS_SYN_RECEIVED: 1851 if ((thflags & TH_ACK) && 1852 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1853 SEQ_GT(th->th_ack, tp->snd_max))) { 1854 rstreason = BANDLIM_RST_OPENPORT; 1855 goto dropwithreset; 1856 } 1857 break; 1858 1859 /* 1860 * If the state is SYN_SENT: 1861 * if seg contains an ACK, but not for our SYN, drop the input. 1862 * if seg contains a RST, then drop the connection. 1863 * if seg does not contain SYN, then drop it. 1864 * Otherwise this is an acceptable SYN segment 1865 * initialize tp->rcv_nxt and tp->irs 1866 * if seg contains ack then advance tp->snd_una 1867 * if seg contains an ECE and ECN support is enabled, the stream 1868 * is ECN capable. 1869 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1870 * arrange for segment to be acked (eventually) 1871 * continue processing rest of data/controls, beginning with URG 1872 */ 1873 case TCPS_SYN_SENT: 1874 if ((thflags & TH_ACK) && 1875 (SEQ_LEQ(th->th_ack, tp->iss) || 1876 SEQ_GT(th->th_ack, tp->snd_max))) { 1877 rstreason = BANDLIM_UNLIMITED; 1878 goto dropwithreset; 1879 } 1880 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1881 tp = tcp_drop(tp, ECONNREFUSED); 1882 if (thflags & TH_RST) 1883 goto drop; 1884 if (!(thflags & TH_SYN)) 1885 goto drop; 1886 1887 tp->irs = th->th_seq; 1888 tcp_rcvseqinit(tp); 1889 if (thflags & TH_ACK) { 1890 TCPSTAT_INC(tcps_connects); 1891 soisconnected(so); 1892 #ifdef MAC 1893 mac_socketpeer_set_from_mbuf(m, so); 1894 #endif 1895 /* Do window scaling on this connection? */ 1896 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1897 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1898 tp->rcv_scale = tp->request_r_scale; 1899 } 1900 tp->rcv_adv += imin(tp->rcv_wnd, 1901 TCP_MAXWIN << tp->rcv_scale); 1902 tp->snd_una++; /* SYN is acked */ 1903 /* 1904 * If there's data, delay ACK; if there's also a FIN 1905 * ACKNOW will be turned on later. 1906 */ 1907 if (DELAY_ACK(tp) && tlen != 0) 1908 tcp_timer_activate(tp, TT_DELACK, 1909 tcp_delacktime); 1910 else 1911 tp->t_flags |= TF_ACKNOW; 1912 1913 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1914 tp->t_flags |= TF_ECN_PERMIT; 1915 TCPSTAT_INC(tcps_ecn_shs); 1916 } 1917 1918 /* 1919 * Received <SYN,ACK> in SYN_SENT[*] state. 1920 * Transitions: 1921 * SYN_SENT --> ESTABLISHED 1922 * SYN_SENT* --> FIN_WAIT_1 1923 */ 1924 tp->t_starttime = ticks; 1925 if (tp->t_flags & TF_NEEDFIN) { 1926 tp->t_state = TCPS_FIN_WAIT_1; 1927 tp->t_flags &= ~TF_NEEDFIN; 1928 thflags &= ~TH_SYN; 1929 } else { 1930 tp->t_state = TCPS_ESTABLISHED; 1931 cc_conn_init(tp); 1932 tcp_timer_activate(tp, TT_KEEP, 1933 TP_KEEPIDLE(tp)); 1934 } 1935 } else { 1936 /* 1937 * Received initial SYN in SYN-SENT[*] state => 1938 * simultaneous open. If segment contains CC option 1939 * and there is a cached CC, apply TAO test. 1940 * If it succeeds, connection is * half-synchronized. 1941 * Otherwise, do 3-way handshake: 1942 * SYN-SENT -> SYN-RECEIVED 1943 * SYN-SENT* -> SYN-RECEIVED* 1944 * If there was no CC option, clear cached CC value. 1945 */ 1946 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1947 tcp_timer_activate(tp, TT_REXMT, 0); 1948 tp->t_state = TCPS_SYN_RECEIVED; 1949 } 1950 1951 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1952 "ti_locked %d", __func__, ti_locked)); 1953 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1954 INP_WLOCK_ASSERT(tp->t_inpcb); 1955 1956 /* 1957 * Advance th->th_seq to correspond to first data byte. 1958 * If data, trim to stay within window, 1959 * dropping FIN if necessary. 1960 */ 1961 th->th_seq++; 1962 if (tlen > tp->rcv_wnd) { 1963 todrop = tlen - tp->rcv_wnd; 1964 m_adj(m, -todrop); 1965 tlen = tp->rcv_wnd; 1966 thflags &= ~TH_FIN; 1967 TCPSTAT_INC(tcps_rcvpackafterwin); 1968 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 1969 } 1970 tp->snd_wl1 = th->th_seq - 1; 1971 tp->rcv_up = th->th_seq; 1972 /* 1973 * Client side of transaction: already sent SYN and data. 1974 * If the remote host used T/TCP to validate the SYN, 1975 * our data will be ACK'd; if so, enter normal data segment 1976 * processing in the middle of step 5, ack processing. 1977 * Otherwise, goto step 6. 1978 */ 1979 if (thflags & TH_ACK) 1980 goto process_ACK; 1981 1982 goto step6; 1983 1984 /* 1985 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1986 * do normal processing. 1987 * 1988 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1989 */ 1990 case TCPS_LAST_ACK: 1991 case TCPS_CLOSING: 1992 break; /* continue normal processing */ 1993 } 1994 1995 /* 1996 * States other than LISTEN or SYN_SENT. 1997 * First check the RST flag and sequence number since reset segments 1998 * are exempt from the timestamp and connection count tests. This 1999 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 2000 * below which allowed reset segments in half the sequence space 2001 * to fall though and be processed (which gives forged reset 2002 * segments with a random sequence number a 50 percent chance of 2003 * killing a connection). 2004 * Then check timestamp, if present. 2005 * Then check the connection count, if present. 2006 * Then check that at least some bytes of segment are within 2007 * receive window. If segment begins before rcv_nxt, 2008 * drop leading data (and SYN); if nothing left, just ack. 2009 * 2010 * 2011 * If the RST bit is set, check the sequence number to see 2012 * if this is a valid reset segment. 2013 * RFC 793 page 37: 2014 * In all states except SYN-SENT, all reset (RST) segments 2015 * are validated by checking their SEQ-fields. A reset is 2016 * valid if its sequence number is in the window. 2017 * Note: this does not take into account delayed ACKs, so 2018 * we should test against last_ack_sent instead of rcv_nxt. 2019 * The sequence number in the reset segment is normally an 2020 * echo of our outgoing acknowlegement numbers, but some hosts 2021 * send a reset with the sequence number at the rightmost edge 2022 * of our receive window, and we have to handle this case. 2023 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 2024 * that brute force RST attacks are possible. To combat this, 2025 * we use a much stricter check while in the ESTABLISHED state, 2026 * only accepting RSTs where the sequence number is equal to 2027 * last_ack_sent. In all other states (the states in which a 2028 * RST is more likely), the more permissive check is used. 2029 * If we have multiple segments in flight, the initial reset 2030 * segment sequence numbers will be to the left of last_ack_sent, 2031 * but they will eventually catch up. 2032 * In any case, it never made sense to trim reset segments to 2033 * fit the receive window since RFC 1122 says: 2034 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 2035 * 2036 * A TCP SHOULD allow a received RST segment to include data. 2037 * 2038 * DISCUSSION 2039 * It has been suggested that a RST segment could contain 2040 * ASCII text that encoded and explained the cause of the 2041 * RST. No standard has yet been established for such 2042 * data. 2043 * 2044 * If the reset segment passes the sequence number test examine 2045 * the state: 2046 * SYN_RECEIVED STATE: 2047 * If passive open, return to LISTEN state. 2048 * If active open, inform user that connection was refused. 2049 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 2050 * Inform user that connection was reset, and close tcb. 2051 * CLOSING, LAST_ACK STATES: 2052 * Close the tcb. 2053 * TIME_WAIT STATE: 2054 * Drop the segment - see Stevens, vol. 2, p. 964 and 2055 * RFC 1337. 2056 */ 2057 if (thflags & TH_RST) { 2058 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2059 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 2060 switch (tp->t_state) { 2061 2062 case TCPS_SYN_RECEIVED: 2063 so->so_error = ECONNREFUSED; 2064 goto close; 2065 2066 case TCPS_ESTABLISHED: 2067 if (V_tcp_insecure_rst == 0 && 2068 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 2069 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 2070 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 2071 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 2072 TCPSTAT_INC(tcps_badrst); 2073 goto drop; 2074 } 2075 /* FALLTHROUGH */ 2076 case TCPS_FIN_WAIT_1: 2077 case TCPS_FIN_WAIT_2: 2078 case TCPS_CLOSE_WAIT: 2079 so->so_error = ECONNRESET; 2080 close: 2081 KASSERT(ti_locked == TI_WLOCKED, 2082 ("tcp_do_segment: TH_RST 1 ti_locked %d", 2083 ti_locked)); 2084 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2085 2086 tp->t_state = TCPS_CLOSED; 2087 TCPSTAT_INC(tcps_drops); 2088 tp = tcp_close(tp); 2089 break; 2090 2091 case TCPS_CLOSING: 2092 case TCPS_LAST_ACK: 2093 KASSERT(ti_locked == TI_WLOCKED, 2094 ("tcp_do_segment: TH_RST 2 ti_locked %d", 2095 ti_locked)); 2096 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2097 2098 tp = tcp_close(tp); 2099 break; 2100 } 2101 } 2102 goto drop; 2103 } 2104 2105 /* 2106 * RFC 1323 PAWS: If we have a timestamp reply on this segment 2107 * and it's less than ts_recent, drop it. 2108 */ 2109 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 2110 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2111 2112 /* Check to see if ts_recent is over 24 days old. */ 2113 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2114 /* 2115 * Invalidate ts_recent. If this segment updates 2116 * ts_recent, the age will be reset later and ts_recent 2117 * will get a valid value. If it does not, setting 2118 * ts_recent to zero will at least satisfy the 2119 * requirement that zero be placed in the timestamp 2120 * echo reply when ts_recent isn't valid. The 2121 * age isn't reset until we get a valid ts_recent 2122 * because we don't want out-of-order segments to be 2123 * dropped when ts_recent is old. 2124 */ 2125 tp->ts_recent = 0; 2126 } else { 2127 TCPSTAT_INC(tcps_rcvduppack); 2128 TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 2129 TCPSTAT_INC(tcps_pawsdrop); 2130 if (tlen) 2131 goto dropafterack; 2132 goto drop; 2133 } 2134 } 2135 2136 /* 2137 * In the SYN-RECEIVED state, validate that the packet belongs to 2138 * this connection before trimming the data to fit the receive 2139 * window. Check the sequence number versus IRS since we know 2140 * the sequence numbers haven't wrapped. This is a partial fix 2141 * for the "LAND" DoS attack. 2142 */ 2143 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2144 rstreason = BANDLIM_RST_OPENPORT; 2145 goto dropwithreset; 2146 } 2147 2148 todrop = tp->rcv_nxt - th->th_seq; 2149 if (todrop > 0) { 2150 /* 2151 * If this is a duplicate SYN for our current connection, 2152 * advance over it and pretend and it's not a SYN. 2153 */ 2154 if (thflags & TH_SYN && th->th_seq == tp->irs) { 2155 thflags &= ~TH_SYN; 2156 th->th_seq++; 2157 if (th->th_urp > 1) 2158 th->th_urp--; 2159 else 2160 thflags &= ~TH_URG; 2161 todrop--; 2162 } 2163 /* 2164 * Following if statement from Stevens, vol. 2, p. 960. 2165 */ 2166 if (todrop > tlen 2167 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2168 /* 2169 * Any valid FIN must be to the left of the window. 2170 * At this point the FIN must be a duplicate or out 2171 * of sequence; drop it. 2172 */ 2173 thflags &= ~TH_FIN; 2174 2175 /* 2176 * Send an ACK to resynchronize and drop any data. 2177 * But keep on processing for RST or ACK. 2178 */ 2179 tp->t_flags |= TF_ACKNOW; 2180 todrop = tlen; 2181 TCPSTAT_INC(tcps_rcvduppack); 2182 TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2183 } else { 2184 TCPSTAT_INC(tcps_rcvpartduppack); 2185 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2186 } 2187 drop_hdrlen += todrop; /* drop from the top afterwards */ 2188 th->th_seq += todrop; 2189 tlen -= todrop; 2190 if (th->th_urp > todrop) 2191 th->th_urp -= todrop; 2192 else { 2193 thflags &= ~TH_URG; 2194 th->th_urp = 0; 2195 } 2196 } 2197 2198 /* 2199 * If new data are received on a connection after the 2200 * user processes are gone, then RST the other end. 2201 */ 2202 if ((so->so_state & SS_NOFDREF) && 2203 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2204 char *s; 2205 2206 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2207 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2208 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2209 2210 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 2211 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 2212 "was closed, sending RST and removing tcpcb\n", 2213 s, __func__, tcpstates[tp->t_state], tlen); 2214 free(s, M_TCPLOG); 2215 } 2216 tp = tcp_close(tp); 2217 TCPSTAT_INC(tcps_rcvafterclose); 2218 rstreason = BANDLIM_UNLIMITED; 2219 goto dropwithreset; 2220 } 2221 2222 /* 2223 * If segment ends after window, drop trailing data 2224 * (and PUSH and FIN); if nothing left, just ACK. 2225 */ 2226 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2227 if (todrop > 0) { 2228 TCPSTAT_INC(tcps_rcvpackafterwin); 2229 if (todrop >= tlen) { 2230 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2231 /* 2232 * If window is closed can only take segments at 2233 * window edge, and have to drop data and PUSH from 2234 * incoming segments. Continue processing, but 2235 * remember to ack. Otherwise, drop segment 2236 * and ack. 2237 */ 2238 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2239 tp->t_flags |= TF_ACKNOW; 2240 TCPSTAT_INC(tcps_rcvwinprobe); 2241 } else 2242 goto dropafterack; 2243 } else 2244 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2245 m_adj(m, -todrop); 2246 tlen -= todrop; 2247 thflags &= ~(TH_PUSH|TH_FIN); 2248 } 2249 2250 /* 2251 * If last ACK falls within this segment's sequence numbers, 2252 * record its timestamp. 2253 * NOTE: 2254 * 1) That the test incorporates suggestions from the latest 2255 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2256 * 2) That updating only on newer timestamps interferes with 2257 * our earlier PAWS tests, so this check should be solely 2258 * predicated on the sequence space of this segment. 2259 * 3) That we modify the segment boundary check to be 2260 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2261 * instead of RFC1323's 2262 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2263 * This modified check allows us to overcome RFC1323's 2264 * limitations as described in Stevens TCP/IP Illustrated 2265 * Vol. 2 p.869. In such cases, we can still calculate the 2266 * RTT correctly when RCV.NXT == Last.ACK.Sent. 2267 */ 2268 if ((to.to_flags & TOF_TS) != 0 && 2269 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2270 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2271 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2272 tp->ts_recent_age = tcp_ts_getticks(); 2273 tp->ts_recent = to.to_tsval; 2274 } 2275 2276 /* 2277 * If a SYN is in the window, then this is an 2278 * error and we send an RST and drop the connection. 2279 */ 2280 if (thflags & TH_SYN) { 2281 KASSERT(ti_locked == TI_WLOCKED, 2282 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2283 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2284 2285 tp = tcp_drop(tp, ECONNRESET); 2286 rstreason = BANDLIM_UNLIMITED; 2287 goto drop; 2288 } 2289 2290 /* 2291 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2292 * flag is on (half-synchronized state), then queue data for 2293 * later processing; else drop segment and return. 2294 */ 2295 if ((thflags & TH_ACK) == 0) { 2296 if (tp->t_state == TCPS_SYN_RECEIVED || 2297 (tp->t_flags & TF_NEEDSYN)) 2298 goto step6; 2299 else if (tp->t_flags & TF_ACKNOW) 2300 goto dropafterack; 2301 else 2302 goto drop; 2303 } 2304 2305 /* 2306 * Ack processing. 2307 */ 2308 switch (tp->t_state) { 2309 2310 /* 2311 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2312 * ESTABLISHED state and continue processing. 2313 * The ACK was checked above. 2314 */ 2315 case TCPS_SYN_RECEIVED: 2316 2317 TCPSTAT_INC(tcps_connects); 2318 soisconnected(so); 2319 /* Do window scaling? */ 2320 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2321 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2322 tp->rcv_scale = tp->request_r_scale; 2323 tp->snd_wnd = tiwin; 2324 } 2325 /* 2326 * Make transitions: 2327 * SYN-RECEIVED -> ESTABLISHED 2328 * SYN-RECEIVED* -> FIN-WAIT-1 2329 */ 2330 tp->t_starttime = ticks; 2331 if (tp->t_flags & TF_NEEDFIN) { 2332 tp->t_state = TCPS_FIN_WAIT_1; 2333 tp->t_flags &= ~TF_NEEDFIN; 2334 } else { 2335 tp->t_state = TCPS_ESTABLISHED; 2336 cc_conn_init(tp); 2337 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 2338 } 2339 /* 2340 * If segment contains data or ACK, will call tcp_reass() 2341 * later; if not, do so now to pass queued data to user. 2342 */ 2343 if (tlen == 0 && (thflags & TH_FIN) == 0) 2344 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2345 (struct mbuf *)0); 2346 tp->snd_wl1 = th->th_seq - 1; 2347 /* FALLTHROUGH */ 2348 2349 /* 2350 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2351 * ACKs. If the ack is in the range 2352 * tp->snd_una < th->th_ack <= tp->snd_max 2353 * then advance tp->snd_una to th->th_ack and drop 2354 * data from the retransmission queue. If this ACK reflects 2355 * more up to date window information we update our window information. 2356 */ 2357 case TCPS_ESTABLISHED: 2358 case TCPS_FIN_WAIT_1: 2359 case TCPS_FIN_WAIT_2: 2360 case TCPS_CLOSE_WAIT: 2361 case TCPS_CLOSING: 2362 case TCPS_LAST_ACK: 2363 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2364 TCPSTAT_INC(tcps_rcvacktoomuch); 2365 goto dropafterack; 2366 } 2367 if ((tp->t_flags & TF_SACK_PERMIT) && 2368 ((to.to_flags & TOF_SACK) || 2369 !TAILQ_EMPTY(&tp->snd_holes))) 2370 tcp_sack_doack(tp, &to, th->th_ack); 2371 2372 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 2373 hhook_run_tcp_est_in(tp, th, &to); 2374 2375 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2376 if (tlen == 0 && tiwin == tp->snd_wnd) { 2377 TCPSTAT_INC(tcps_rcvdupack); 2378 /* 2379 * If we have outstanding data (other than 2380 * a window probe), this is a completely 2381 * duplicate ack (ie, window info didn't 2382 * change), the ack is the biggest we've 2383 * seen and we've seen exactly our rexmt 2384 * threshhold of them, assume a packet 2385 * has been dropped and retransmit it. 2386 * Kludge snd_nxt & the congestion 2387 * window so we send only this one 2388 * packet. 2389 * 2390 * We know we're losing at the current 2391 * window size so do congestion avoidance 2392 * (set ssthresh to half the current window 2393 * and pull our congestion window back to 2394 * the new ssthresh). 2395 * 2396 * Dup acks mean that packets have left the 2397 * network (they're now cached at the receiver) 2398 * so bump cwnd by the amount in the receiver 2399 * to keep a constant cwnd packets in the 2400 * network. 2401 * 2402 * When using TCP ECN, notify the peer that 2403 * we reduced the cwnd. 2404 */ 2405 if (!tcp_timer_active(tp, TT_REXMT) || 2406 th->th_ack != tp->snd_una) 2407 tp->t_dupacks = 0; 2408 else if (++tp->t_dupacks > tcprexmtthresh || 2409 IN_FASTRECOVERY(tp->t_flags)) { 2410 cc_ack_received(tp, th, CC_DUPACK); 2411 if ((tp->t_flags & TF_SACK_PERMIT) && 2412 IN_FASTRECOVERY(tp->t_flags)) { 2413 int awnd; 2414 2415 /* 2416 * Compute the amount of data in flight first. 2417 * We can inject new data into the pipe iff 2418 * we have less than 1/2 the original window's 2419 * worth of data in flight. 2420 */ 2421 awnd = (tp->snd_nxt - tp->snd_fack) + 2422 tp->sackhint.sack_bytes_rexmit; 2423 if (awnd < tp->snd_ssthresh) { 2424 tp->snd_cwnd += tp->t_maxseg; 2425 if (tp->snd_cwnd > tp->snd_ssthresh) 2426 tp->snd_cwnd = tp->snd_ssthresh; 2427 } 2428 } else 2429 tp->snd_cwnd += tp->t_maxseg; 2430 if ((thflags & TH_FIN) && 2431 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2432 /* 2433 * If its a fin we need to process 2434 * it to avoid a race where both 2435 * sides enter FIN-WAIT and send FIN|ACK 2436 * at the same time. 2437 */ 2438 break; 2439 } 2440 (void) tcp_output(tp); 2441 goto drop; 2442 } else if (tp->t_dupacks == tcprexmtthresh) { 2443 tcp_seq onxt = tp->snd_nxt; 2444 2445 /* 2446 * If we're doing sack, check to 2447 * see if we're already in sack 2448 * recovery. If we're not doing sack, 2449 * check to see if we're in newreno 2450 * recovery. 2451 */ 2452 if (tp->t_flags & TF_SACK_PERMIT) { 2453 if (IN_FASTRECOVERY(tp->t_flags)) { 2454 tp->t_dupacks = 0; 2455 break; 2456 } 2457 } else { 2458 if (SEQ_LEQ(th->th_ack, 2459 tp->snd_recover)) { 2460 tp->t_dupacks = 0; 2461 break; 2462 } 2463 } 2464 /* Congestion signal before ack. */ 2465 cc_cong_signal(tp, th, CC_NDUPACK); 2466 cc_ack_received(tp, th, CC_DUPACK); 2467 tcp_timer_activate(tp, TT_REXMT, 0); 2468 tp->t_rtttime = 0; 2469 if (tp->t_flags & TF_SACK_PERMIT) { 2470 TCPSTAT_INC( 2471 tcps_sack_recovery_episode); 2472 tp->sack_newdata = tp->snd_nxt; 2473 tp->snd_cwnd = tp->t_maxseg; 2474 (void) tcp_output(tp); 2475 goto drop; 2476 } 2477 tp->snd_nxt = th->th_ack; 2478 tp->snd_cwnd = tp->t_maxseg; 2479 if ((thflags & TH_FIN) && 2480 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2481 /* 2482 * If its a fin we need to process 2483 * it to avoid a race where both 2484 * sides enter FIN-WAIT and send FIN|ACK 2485 * at the same time. 2486 */ 2487 break; 2488 } 2489 (void) tcp_output(tp); 2490 KASSERT(tp->snd_limited <= 2, 2491 ("%s: tp->snd_limited too big", 2492 __func__)); 2493 tp->snd_cwnd = tp->snd_ssthresh + 2494 tp->t_maxseg * 2495 (tp->t_dupacks - tp->snd_limited); 2496 if (SEQ_GT(onxt, tp->snd_nxt)) 2497 tp->snd_nxt = onxt; 2498 goto drop; 2499 } else if (V_tcp_do_rfc3042) { 2500 cc_ack_received(tp, th, CC_DUPACK); 2501 u_long oldcwnd = tp->snd_cwnd; 2502 tcp_seq oldsndmax = tp->snd_max; 2503 u_int sent; 2504 2505 KASSERT(tp->t_dupacks == 1 || 2506 tp->t_dupacks == 2, 2507 ("%s: dupacks not 1 or 2", 2508 __func__)); 2509 if (tp->t_dupacks == 1) 2510 tp->snd_limited = 0; 2511 tp->snd_cwnd = 2512 (tp->snd_nxt - tp->snd_una) + 2513 (tp->t_dupacks - tp->snd_limited) * 2514 tp->t_maxseg; 2515 if ((thflags & TH_FIN) && 2516 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 2517 /* 2518 * If its a fin we need to process 2519 * it to avoid a race where both 2520 * sides enter FIN-WAIT and send FIN|ACK 2521 * at the same time. 2522 */ 2523 break; 2524 } 2525 (void) tcp_output(tp); 2526 sent = tp->snd_max - oldsndmax; 2527 if (sent > tp->t_maxseg) { 2528 KASSERT((tp->t_dupacks == 2 && 2529 tp->snd_limited == 0) || 2530 (sent == tp->t_maxseg + 1 && 2531 tp->t_flags & TF_SENTFIN), 2532 ("%s: sent too much", 2533 __func__)); 2534 tp->snd_limited = 2; 2535 } else if (sent > 0) 2536 ++tp->snd_limited; 2537 tp->snd_cwnd = oldcwnd; 2538 goto drop; 2539 } 2540 } else 2541 tp->t_dupacks = 0; 2542 break; 2543 } 2544 2545 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2546 ("%s: th_ack <= snd_una", __func__)); 2547 2548 /* 2549 * If the congestion window was inflated to account 2550 * for the other side's cached packets, retract it. 2551 */ 2552 if (IN_FASTRECOVERY(tp->t_flags)) { 2553 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2554 if (tp->t_flags & TF_SACK_PERMIT) 2555 tcp_sack_partialack(tp, th); 2556 else 2557 tcp_newreno_partial_ack(tp, th); 2558 } else 2559 cc_post_recovery(tp, th); 2560 } 2561 tp->t_dupacks = 0; 2562 /* 2563 * If we reach this point, ACK is not a duplicate, 2564 * i.e., it ACKs something we sent. 2565 */ 2566 if (tp->t_flags & TF_NEEDSYN) { 2567 /* 2568 * T/TCP: Connection was half-synchronized, and our 2569 * SYN has been ACK'd (so connection is now fully 2570 * synchronized). Go to non-starred state, 2571 * increment snd_una for ACK of SYN, and check if 2572 * we can do window scaling. 2573 */ 2574 tp->t_flags &= ~TF_NEEDSYN; 2575 tp->snd_una++; 2576 /* Do window scaling? */ 2577 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2578 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2579 tp->rcv_scale = tp->request_r_scale; 2580 /* Send window already scaled. */ 2581 } 2582 } 2583 2584 process_ACK: 2585 INP_WLOCK_ASSERT(tp->t_inpcb); 2586 2587 acked = BYTES_THIS_ACK(tp, th); 2588 TCPSTAT_INC(tcps_rcvackpack); 2589 TCPSTAT_ADD(tcps_rcvackbyte, acked); 2590 2591 /* 2592 * If we just performed our first retransmit, and the ACK 2593 * arrives within our recovery window, then it was a mistake 2594 * to do the retransmit in the first place. Recover our 2595 * original cwnd and ssthresh, and proceed to transmit where 2596 * we left off. 2597 */ 2598 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2599 (int)(ticks - tp->t_badrxtwin) < 0) 2600 cc_cong_signal(tp, th, CC_RTO_ERR); 2601 2602 /* 2603 * If we have a timestamp reply, update smoothed 2604 * round trip time. If no timestamp is present but 2605 * transmit timer is running and timed sequence 2606 * number was acked, update smoothed round trip time. 2607 * Since we now have an rtt measurement, cancel the 2608 * timer backoff (cf., Phil Karn's retransmit alg.). 2609 * Recompute the initial retransmit timer. 2610 * 2611 * Some boxes send broken timestamp replies 2612 * during the SYN+ACK phase, ignore 2613 * timestamps of 0 or we could calculate a 2614 * huge RTT and blow up the retransmit timer. 2615 */ 2616 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2617 u_int t; 2618 2619 t = tcp_ts_getticks() - to.to_tsecr; 2620 if (!tp->t_rttlow || tp->t_rttlow > t) 2621 tp->t_rttlow = t; 2622 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2623 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2624 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2625 tp->t_rttlow = ticks - tp->t_rtttime; 2626 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2627 } 2628 2629 /* 2630 * If all outstanding data is acked, stop retransmit 2631 * timer and remember to restart (more output or persist). 2632 * If there is more data to be acked, restart retransmit 2633 * timer, using current (possibly backed-off) value. 2634 */ 2635 if (th->th_ack == tp->snd_max) { 2636 tcp_timer_activate(tp, TT_REXMT, 0); 2637 needoutput = 1; 2638 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2639 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2640 2641 /* 2642 * If no data (only SYN) was ACK'd, 2643 * skip rest of ACK processing. 2644 */ 2645 if (acked == 0) 2646 goto step6; 2647 2648 /* 2649 * Let the congestion control algorithm update congestion 2650 * control related information. This typically means increasing 2651 * the congestion window. 2652 */ 2653 cc_ack_received(tp, th, CC_ACK); 2654 2655 SOCKBUF_LOCK(&so->so_snd); 2656 if (acked > so->so_snd.sb_cc) { 2657 tp->snd_wnd -= so->so_snd.sb_cc; 2658 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2659 ourfinisacked = 1; 2660 } else { 2661 sbdrop_locked(&so->so_snd, acked); 2662 tp->snd_wnd -= acked; 2663 ourfinisacked = 0; 2664 } 2665 /* NB: sowwakeup_locked() does an implicit unlock. */ 2666 sowwakeup_locked(so); 2667 /* Detect una wraparound. */ 2668 if (!IN_RECOVERY(tp->t_flags) && 2669 SEQ_GT(tp->snd_una, tp->snd_recover) && 2670 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2671 tp->snd_recover = th->th_ack - 1; 2672 /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2673 if (IN_RECOVERY(tp->t_flags) && 2674 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2675 EXIT_RECOVERY(tp->t_flags); 2676 } 2677 tp->snd_una = th->th_ack; 2678 if (tp->t_flags & TF_SACK_PERMIT) { 2679 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2680 tp->snd_recover = tp->snd_una; 2681 } 2682 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2683 tp->snd_nxt = tp->snd_una; 2684 2685 switch (tp->t_state) { 2686 2687 /* 2688 * In FIN_WAIT_1 STATE in addition to the processing 2689 * for the ESTABLISHED state if our FIN is now acknowledged 2690 * then enter FIN_WAIT_2. 2691 */ 2692 case TCPS_FIN_WAIT_1: 2693 if (ourfinisacked) { 2694 /* 2695 * If we can't receive any more 2696 * data, then closing user can proceed. 2697 * Starting the timer is contrary to the 2698 * specification, but if we don't get a FIN 2699 * we'll hang forever. 2700 * 2701 * XXXjl: 2702 * we should release the tp also, and use a 2703 * compressed state. 2704 */ 2705 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2706 soisdisconnected(so); 2707 tcp_timer_activate(tp, TT_2MSL, 2708 (tcp_fast_finwait2_recycle ? 2709 tcp_finwait2_timeout : 2710 TP_MAXIDLE(tp))); 2711 } 2712 tp->t_state = TCPS_FIN_WAIT_2; 2713 } 2714 break; 2715 2716 /* 2717 * In CLOSING STATE in addition to the processing for 2718 * the ESTABLISHED state if the ACK acknowledges our FIN 2719 * then enter the TIME-WAIT state, otherwise ignore 2720 * the segment. 2721 */ 2722 case TCPS_CLOSING: 2723 if (ourfinisacked) { 2724 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2725 tcp_twstart(tp); 2726 INP_INFO_WUNLOCK(&V_tcbinfo); 2727 m_freem(m); 2728 return; 2729 } 2730 break; 2731 2732 /* 2733 * In LAST_ACK, we may still be waiting for data to drain 2734 * and/or to be acked, as well as for the ack of our FIN. 2735 * If our FIN is now acknowledged, delete the TCB, 2736 * enter the closed state and return. 2737 */ 2738 case TCPS_LAST_ACK: 2739 if (ourfinisacked) { 2740 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2741 tp = tcp_close(tp); 2742 goto drop; 2743 } 2744 break; 2745 } 2746 } 2747 2748 step6: 2749 INP_WLOCK_ASSERT(tp->t_inpcb); 2750 2751 /* 2752 * Update window information. 2753 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2754 */ 2755 if ((thflags & TH_ACK) && 2756 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2757 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2758 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2759 /* keep track of pure window updates */ 2760 if (tlen == 0 && 2761 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2762 TCPSTAT_INC(tcps_rcvwinupd); 2763 tp->snd_wnd = tiwin; 2764 tp->snd_wl1 = th->th_seq; 2765 tp->snd_wl2 = th->th_ack; 2766 if (tp->snd_wnd > tp->max_sndwnd) 2767 tp->max_sndwnd = tp->snd_wnd; 2768 needoutput = 1; 2769 } 2770 2771 /* 2772 * Process segments with URG. 2773 */ 2774 if ((thflags & TH_URG) && th->th_urp && 2775 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2776 /* 2777 * This is a kludge, but if we receive and accept 2778 * random urgent pointers, we'll crash in 2779 * soreceive. It's hard to imagine someone 2780 * actually wanting to send this much urgent data. 2781 */ 2782 SOCKBUF_LOCK(&so->so_rcv); 2783 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2784 th->th_urp = 0; /* XXX */ 2785 thflags &= ~TH_URG; /* XXX */ 2786 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2787 goto dodata; /* XXX */ 2788 } 2789 /* 2790 * If this segment advances the known urgent pointer, 2791 * then mark the data stream. This should not happen 2792 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2793 * a FIN has been received from the remote side. 2794 * In these states we ignore the URG. 2795 * 2796 * According to RFC961 (Assigned Protocols), 2797 * the urgent pointer points to the last octet 2798 * of urgent data. We continue, however, 2799 * to consider it to indicate the first octet 2800 * of data past the urgent section as the original 2801 * spec states (in one of two places). 2802 */ 2803 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2804 tp->rcv_up = th->th_seq + th->th_urp; 2805 so->so_oobmark = so->so_rcv.sb_cc + 2806 (tp->rcv_up - tp->rcv_nxt) - 1; 2807 if (so->so_oobmark == 0) 2808 so->so_rcv.sb_state |= SBS_RCVATMARK; 2809 sohasoutofband(so); 2810 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2811 } 2812 SOCKBUF_UNLOCK(&so->so_rcv); 2813 /* 2814 * Remove out of band data so doesn't get presented to user. 2815 * This can happen independent of advancing the URG pointer, 2816 * but if two URG's are pending at once, some out-of-band 2817 * data may creep in... ick. 2818 */ 2819 if (th->th_urp <= (u_long)tlen && 2820 !(so->so_options & SO_OOBINLINE)) { 2821 /* hdr drop is delayed */ 2822 tcp_pulloutofband(so, th, m, drop_hdrlen); 2823 } 2824 } else { 2825 /* 2826 * If no out of band data is expected, 2827 * pull receive urgent pointer along 2828 * with the receive window. 2829 */ 2830 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2831 tp->rcv_up = tp->rcv_nxt; 2832 } 2833 dodata: /* XXX */ 2834 INP_WLOCK_ASSERT(tp->t_inpcb); 2835 2836 /* 2837 * Process the segment text, merging it into the TCP sequencing queue, 2838 * and arranging for acknowledgment of receipt if necessary. 2839 * This process logically involves adjusting tp->rcv_wnd as data 2840 * is presented to the user (this happens in tcp_usrreq.c, 2841 * case PRU_RCVD). If a FIN has already been received on this 2842 * connection then we just ignore the text. 2843 */ 2844 if ((tlen || (thflags & TH_FIN)) && 2845 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2846 tcp_seq save_start = th->th_seq; 2847 m_adj(m, drop_hdrlen); /* delayed header drop */ 2848 /* 2849 * Insert segment which includes th into TCP reassembly queue 2850 * with control block tp. Set thflags to whether reassembly now 2851 * includes a segment with FIN. This handles the common case 2852 * inline (segment is the next to be received on an established 2853 * connection, and the queue is empty), avoiding linkage into 2854 * and removal from the queue and repetition of various 2855 * conversions. 2856 * Set DELACK for segments received in order, but ack 2857 * immediately when segments are out of order (so 2858 * fast retransmit can work). 2859 */ 2860 if (th->th_seq == tp->rcv_nxt && 2861 LIST_EMPTY(&tp->t_segq) && 2862 TCPS_HAVEESTABLISHED(tp->t_state)) { 2863 if (DELAY_ACK(tp)) 2864 tp->t_flags |= TF_DELACK; 2865 else 2866 tp->t_flags |= TF_ACKNOW; 2867 tp->rcv_nxt += tlen; 2868 thflags = th->th_flags & TH_FIN; 2869 TCPSTAT_INC(tcps_rcvpack); 2870 TCPSTAT_ADD(tcps_rcvbyte, tlen); 2871 ND6_HINT(tp); 2872 SOCKBUF_LOCK(&so->so_rcv); 2873 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2874 m_freem(m); 2875 else 2876 sbappendstream_locked(&so->so_rcv, m); 2877 /* NB: sorwakeup_locked() does an implicit unlock. */ 2878 sorwakeup_locked(so); 2879 } else { 2880 /* 2881 * XXX: Due to the header drop above "th" is 2882 * theoretically invalid by now. Fortunately 2883 * m_adj() doesn't actually frees any mbufs 2884 * when trimming from the head. 2885 */ 2886 thflags = tcp_reass(tp, th, &tlen, m); 2887 tp->t_flags |= TF_ACKNOW; 2888 } 2889 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2890 tcp_update_sack_list(tp, save_start, save_start + tlen); 2891 #if 0 2892 /* 2893 * Note the amount of data that peer has sent into 2894 * our window, in order to estimate the sender's 2895 * buffer size. 2896 * XXX: Unused. 2897 */ 2898 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2899 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2900 else 2901 len = so->so_rcv.sb_hiwat; 2902 #endif 2903 } else { 2904 m_freem(m); 2905 thflags &= ~TH_FIN; 2906 } 2907 2908 /* 2909 * If FIN is received ACK the FIN and let the user know 2910 * that the connection is closing. 2911 */ 2912 if (thflags & TH_FIN) { 2913 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2914 socantrcvmore(so); 2915 /* 2916 * If connection is half-synchronized 2917 * (ie NEEDSYN flag on) then delay ACK, 2918 * so it may be piggybacked when SYN is sent. 2919 * Otherwise, since we received a FIN then no 2920 * more input can be expected, send ACK now. 2921 */ 2922 if (tp->t_flags & TF_NEEDSYN) 2923 tp->t_flags |= TF_DELACK; 2924 else 2925 tp->t_flags |= TF_ACKNOW; 2926 tp->rcv_nxt++; 2927 } 2928 switch (tp->t_state) { 2929 2930 /* 2931 * In SYN_RECEIVED and ESTABLISHED STATES 2932 * enter the CLOSE_WAIT state. 2933 */ 2934 case TCPS_SYN_RECEIVED: 2935 tp->t_starttime = ticks; 2936 /* FALLTHROUGH */ 2937 case TCPS_ESTABLISHED: 2938 tp->t_state = TCPS_CLOSE_WAIT; 2939 break; 2940 2941 /* 2942 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2943 * enter the CLOSING state. 2944 */ 2945 case TCPS_FIN_WAIT_1: 2946 tp->t_state = TCPS_CLOSING; 2947 break; 2948 2949 /* 2950 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2951 * starting the time-wait timer, turning off the other 2952 * standard timers. 2953 */ 2954 case TCPS_FIN_WAIT_2: 2955 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2956 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2957 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2958 ti_locked)); 2959 2960 tcp_twstart(tp); 2961 INP_INFO_WUNLOCK(&V_tcbinfo); 2962 return; 2963 } 2964 } 2965 if (ti_locked == TI_WLOCKED) 2966 INP_INFO_WUNLOCK(&V_tcbinfo); 2967 ti_locked = TI_UNLOCKED; 2968 2969 #ifdef TCPDEBUG 2970 if (so->so_options & SO_DEBUG) 2971 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2972 &tcp_savetcp, 0); 2973 #endif 2974 2975 /* 2976 * Return any desired output. 2977 */ 2978 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2979 (void) tcp_output(tp); 2980 2981 check_delack: 2982 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2983 __func__, ti_locked)); 2984 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2985 INP_WLOCK_ASSERT(tp->t_inpcb); 2986 2987 if (tp->t_flags & TF_DELACK) { 2988 tp->t_flags &= ~TF_DELACK; 2989 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2990 } 2991 INP_WUNLOCK(tp->t_inpcb); 2992 return; 2993 2994 dropafterack: 2995 /* 2996 * Generate an ACK dropping incoming segment if it occupies 2997 * sequence space, where the ACK reflects our state. 2998 * 2999 * We can now skip the test for the RST flag since all 3000 * paths to this code happen after packets containing 3001 * RST have been dropped. 3002 * 3003 * In the SYN-RECEIVED state, don't send an ACK unless the 3004 * segment we received passes the SYN-RECEIVED ACK test. 3005 * If it fails send a RST. This breaks the loop in the 3006 * "LAND" DoS attack, and also prevents an ACK storm 3007 * between two listening ports that have been sent forged 3008 * SYN segments, each with the source address of the other. 3009 */ 3010 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3011 (SEQ_GT(tp->snd_una, th->th_ack) || 3012 SEQ_GT(th->th_ack, tp->snd_max)) ) { 3013 rstreason = BANDLIM_RST_OPENPORT; 3014 goto dropwithreset; 3015 } 3016 #ifdef TCPDEBUG 3017 if (so->so_options & SO_DEBUG) 3018 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3019 &tcp_savetcp, 0); 3020 #endif 3021 if (ti_locked == TI_WLOCKED) 3022 INP_INFO_WUNLOCK(&V_tcbinfo); 3023 ti_locked = TI_UNLOCKED; 3024 3025 tp->t_flags |= TF_ACKNOW; 3026 (void) tcp_output(tp); 3027 INP_WUNLOCK(tp->t_inpcb); 3028 m_freem(m); 3029 return; 3030 3031 dropwithreset: 3032 if (ti_locked == TI_WLOCKED) 3033 INP_INFO_WUNLOCK(&V_tcbinfo); 3034 ti_locked = TI_UNLOCKED; 3035 3036 if (tp != NULL) { 3037 tcp_dropwithreset(m, th, tp, tlen, rstreason); 3038 INP_WUNLOCK(tp->t_inpcb); 3039 } else 3040 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3041 return; 3042 3043 drop: 3044 if (ti_locked == TI_WLOCKED) { 3045 INP_INFO_WUNLOCK(&V_tcbinfo); 3046 ti_locked = TI_UNLOCKED; 3047 } 3048 #ifdef INVARIANTS 3049 else 3050 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3051 #endif 3052 3053 /* 3054 * Drop space held by incoming segment and return. 3055 */ 3056 #ifdef TCPDEBUG 3057 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3058 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3059 &tcp_savetcp, 0); 3060 #endif 3061 if (tp != NULL) 3062 INP_WUNLOCK(tp->t_inpcb); 3063 m_freem(m); 3064 } 3065 3066 /* 3067 * Issue RST and make ACK acceptable to originator of segment. 3068 * The mbuf must still include the original packet header. 3069 * tp may be NULL. 3070 */ 3071 static void 3072 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3073 int tlen, int rstreason) 3074 { 3075 #ifdef INET 3076 struct ip *ip; 3077 #endif 3078 #ifdef INET6 3079 struct ip6_hdr *ip6; 3080 #endif 3081 3082 if (tp != NULL) { 3083 INP_WLOCK_ASSERT(tp->t_inpcb); 3084 } 3085 3086 /* Don't bother if destination was broadcast/multicast. */ 3087 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3088 goto drop; 3089 #ifdef INET6 3090 if (mtod(m, struct ip *)->ip_v == 6) { 3091 ip6 = mtod(m, struct ip6_hdr *); 3092 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3093 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3094 goto drop; 3095 /* IPv6 anycast check is done at tcp6_input() */ 3096 } 3097 #endif 3098 #if defined(INET) && defined(INET6) 3099 else 3100 #endif 3101 #ifdef INET 3102 { 3103 ip = mtod(m, struct ip *); 3104 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3105 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3106 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3107 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3108 goto drop; 3109 } 3110 #endif 3111 3112 /* Perform bandwidth limiting. */ 3113 if (badport_bandlim(rstreason) < 0) 3114 goto drop; 3115 3116 /* tcp_respond consumes the mbuf chain. */ 3117 if (th->th_flags & TH_ACK) { 3118 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3119 th->th_ack, TH_RST); 3120 } else { 3121 if (th->th_flags & TH_SYN) 3122 tlen++; 3123 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3124 (tcp_seq)0, TH_RST|TH_ACK); 3125 } 3126 return; 3127 drop: 3128 m_freem(m); 3129 } 3130 3131 /* 3132 * Parse TCP options and place in tcpopt. 3133 */ 3134 static void 3135 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3136 { 3137 int opt, optlen; 3138 3139 to->to_flags = 0; 3140 for (; cnt > 0; cnt -= optlen, cp += optlen) { 3141 opt = cp[0]; 3142 if (opt == TCPOPT_EOL) 3143 break; 3144 if (opt == TCPOPT_NOP) 3145 optlen = 1; 3146 else { 3147 if (cnt < 2) 3148 break; 3149 optlen = cp[1]; 3150 if (optlen < 2 || optlen > cnt) 3151 break; 3152 } 3153 switch (opt) { 3154 case TCPOPT_MAXSEG: 3155 if (optlen != TCPOLEN_MAXSEG) 3156 continue; 3157 if (!(flags & TO_SYN)) 3158 continue; 3159 to->to_flags |= TOF_MSS; 3160 bcopy((char *)cp + 2, 3161 (char *)&to->to_mss, sizeof(to->to_mss)); 3162 to->to_mss = ntohs(to->to_mss); 3163 break; 3164 case TCPOPT_WINDOW: 3165 if (optlen != TCPOLEN_WINDOW) 3166 continue; 3167 if (!(flags & TO_SYN)) 3168 continue; 3169 to->to_flags |= TOF_SCALE; 3170 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3171 break; 3172 case TCPOPT_TIMESTAMP: 3173 if (optlen != TCPOLEN_TIMESTAMP) 3174 continue; 3175 to->to_flags |= TOF_TS; 3176 bcopy((char *)cp + 2, 3177 (char *)&to->to_tsval, sizeof(to->to_tsval)); 3178 to->to_tsval = ntohl(to->to_tsval); 3179 bcopy((char *)cp + 6, 3180 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3181 to->to_tsecr = ntohl(to->to_tsecr); 3182 break; 3183 #ifdef TCP_SIGNATURE 3184 /* 3185 * XXX In order to reply to a host which has set the 3186 * TCP_SIGNATURE option in its initial SYN, we have to 3187 * record the fact that the option was observed here 3188 * for the syncache code to perform the correct response. 3189 */ 3190 case TCPOPT_SIGNATURE: 3191 if (optlen != TCPOLEN_SIGNATURE) 3192 continue; 3193 to->to_flags |= TOF_SIGNATURE; 3194 to->to_signature = cp + 2; 3195 break; 3196 #endif 3197 case TCPOPT_SACK_PERMITTED: 3198 if (optlen != TCPOLEN_SACK_PERMITTED) 3199 continue; 3200 if (!(flags & TO_SYN)) 3201 continue; 3202 if (!V_tcp_do_sack) 3203 continue; 3204 to->to_flags |= TOF_SACKPERM; 3205 break; 3206 case TCPOPT_SACK: 3207 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 3208 continue; 3209 if (flags & TO_SYN) 3210 continue; 3211 to->to_flags |= TOF_SACK; 3212 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 3213 to->to_sacks = cp + 2; 3214 TCPSTAT_INC(tcps_sack_rcv_blocks); 3215 break; 3216 default: 3217 continue; 3218 } 3219 } 3220 } 3221 3222 /* 3223 * Pull out of band byte out of a segment so 3224 * it doesn't appear in the user's data queue. 3225 * It is still reflected in the segment length for 3226 * sequencing purposes. 3227 */ 3228 static void 3229 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3230 int off) 3231 { 3232 int cnt = off + th->th_urp - 1; 3233 3234 while (cnt >= 0) { 3235 if (m->m_len > cnt) { 3236 char *cp = mtod(m, caddr_t) + cnt; 3237 struct tcpcb *tp = sototcpcb(so); 3238 3239 INP_WLOCK_ASSERT(tp->t_inpcb); 3240 3241 tp->t_iobc = *cp; 3242 tp->t_oobflags |= TCPOOB_HAVEDATA; 3243 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3244 m->m_len--; 3245 if (m->m_flags & M_PKTHDR) 3246 m->m_pkthdr.len--; 3247 return; 3248 } 3249 cnt -= m->m_len; 3250 m = m->m_next; 3251 if (m == NULL) 3252 break; 3253 } 3254 panic("tcp_pulloutofband"); 3255 } 3256 3257 /* 3258 * Collect new round-trip time estimate 3259 * and update averages and current timeout. 3260 */ 3261 static void 3262 tcp_xmit_timer(struct tcpcb *tp, int rtt) 3263 { 3264 int delta; 3265 3266 INP_WLOCK_ASSERT(tp->t_inpcb); 3267 3268 TCPSTAT_INC(tcps_rttupdated); 3269 tp->t_rttupdated++; 3270 if (tp->t_srtt != 0) { 3271 /* 3272 * srtt is stored as fixed point with 5 bits after the 3273 * binary point (i.e., scaled by 8). The following magic 3274 * is equivalent to the smoothing algorithm in rfc793 with 3275 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3276 * point). Adjust rtt to origin 0. 3277 */ 3278 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3279 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3280 3281 if ((tp->t_srtt += delta) <= 0) 3282 tp->t_srtt = 1; 3283 3284 /* 3285 * We accumulate a smoothed rtt variance (actually, a 3286 * smoothed mean difference), then set the retransmit 3287 * timer to smoothed rtt + 4 times the smoothed variance. 3288 * rttvar is stored as fixed point with 4 bits after the 3289 * binary point (scaled by 16). The following is 3290 * equivalent to rfc793 smoothing with an alpha of .75 3291 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3292 * rfc793's wired-in beta. 3293 */ 3294 if (delta < 0) 3295 delta = -delta; 3296 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3297 if ((tp->t_rttvar += delta) <= 0) 3298 tp->t_rttvar = 1; 3299 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3300 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3301 } else { 3302 /* 3303 * No rtt measurement yet - use the unsmoothed rtt. 3304 * Set the variance to half the rtt (so our first 3305 * retransmit happens at 3*rtt). 3306 */ 3307 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3308 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3309 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3310 } 3311 tp->t_rtttime = 0; 3312 tp->t_rxtshift = 0; 3313 3314 /* 3315 * the retransmit should happen at rtt + 4 * rttvar. 3316 * Because of the way we do the smoothing, srtt and rttvar 3317 * will each average +1/2 tick of bias. When we compute 3318 * the retransmit timer, we want 1/2 tick of rounding and 3319 * 1 extra tick because of +-1/2 tick uncertainty in the 3320 * firing of the timer. The bias will give us exactly the 3321 * 1.5 tick we need. But, because the bias is 3322 * statistical, we have to test that we don't drop below 3323 * the minimum feasible timer (which is 2 ticks). 3324 */ 3325 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3326 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3327 3328 /* 3329 * We received an ack for a packet that wasn't retransmitted; 3330 * it is probably safe to discard any error indications we've 3331 * received recently. This isn't quite right, but close enough 3332 * for now (a route might have failed after we sent a segment, 3333 * and the return path might not be symmetrical). 3334 */ 3335 tp->t_softerror = 0; 3336 } 3337 3338 /* 3339 * Determine a reasonable value for maxseg size. 3340 * If the route is known, check route for mtu. 3341 * If none, use an mss that can be handled on the outgoing interface 3342 * without forcing IP to fragment. If no route is found, route has no mtu, 3343 * or the destination isn't local, use a default, hopefully conservative 3344 * size (usually 512 or the default IP max size, but no more than the mtu 3345 * of the interface), as we can't discover anything about intervening 3346 * gateways or networks. We also initialize the congestion/slow start 3347 * window to be a single segment if the destination isn't local. 3348 * While looking at the routing entry, we also initialize other path-dependent 3349 * parameters from pre-set or cached values in the routing entry. 3350 * 3351 * Also take into account the space needed for options that we 3352 * send regularly. Make maxseg shorter by that amount to assure 3353 * that we can send maxseg amount of data even when the options 3354 * are present. Store the upper limit of the length of options plus 3355 * data in maxopd. 3356 * 3357 * NOTE that this routine is only called when we process an incoming 3358 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3359 * settings are handled in tcp_mssopt(). 3360 */ 3361 void 3362 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 3363 struct hc_metrics_lite *metricptr, int *mtuflags) 3364 { 3365 int mss = 0; 3366 u_long maxmtu = 0; 3367 struct inpcb *inp = tp->t_inpcb; 3368 struct hc_metrics_lite metrics; 3369 int origoffer; 3370 #ifdef INET6 3371 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3372 size_t min_protoh = isipv6 ? 3373 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3374 sizeof (struct tcpiphdr); 3375 #else 3376 const size_t min_protoh = sizeof(struct tcpiphdr); 3377 #endif 3378 3379 INP_WLOCK_ASSERT(tp->t_inpcb); 3380 3381 if (mtuoffer != -1) { 3382 KASSERT(offer == -1, ("%s: conflict", __func__)); 3383 offer = mtuoffer - min_protoh; 3384 } 3385 origoffer = offer; 3386 3387 /* Initialize. */ 3388 #ifdef INET6 3389 if (isipv6) { 3390 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags); 3391 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3392 } 3393 #endif 3394 #if defined(INET) && defined(INET6) 3395 else 3396 #endif 3397 #ifdef INET 3398 { 3399 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags); 3400 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3401 } 3402 #endif 3403 3404 /* 3405 * No route to sender, stay with default mss and return. 3406 */ 3407 if (maxmtu == 0) { 3408 /* 3409 * In case we return early we need to initialize metrics 3410 * to a defined state as tcp_hc_get() would do for us 3411 * if there was no cache hit. 3412 */ 3413 if (metricptr != NULL) 3414 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3415 return; 3416 } 3417 3418 /* What have we got? */ 3419 switch (offer) { 3420 case 0: 3421 /* 3422 * Offer == 0 means that there was no MSS on the SYN 3423 * segment, in this case we use tcp_mssdflt as 3424 * already assigned to t_maxopd above. 3425 */ 3426 offer = tp->t_maxopd; 3427 break; 3428 3429 case -1: 3430 /* 3431 * Offer == -1 means that we didn't receive SYN yet. 3432 */ 3433 /* FALLTHROUGH */ 3434 3435 default: 3436 /* 3437 * Prevent DoS attack with too small MSS. Round up 3438 * to at least minmss. 3439 */ 3440 offer = max(offer, V_tcp_minmss); 3441 } 3442 3443 /* 3444 * rmx information is now retrieved from tcp_hostcache. 3445 */ 3446 tcp_hc_get(&inp->inp_inc, &metrics); 3447 if (metricptr != NULL) 3448 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3449 3450 /* 3451 * If there's a discovered mtu int tcp hostcache, use it 3452 * else, use the link mtu. 3453 */ 3454 if (metrics.rmx_mtu) 3455 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3456 else { 3457 #ifdef INET6 3458 if (isipv6) { 3459 mss = maxmtu - min_protoh; 3460 if (!V_path_mtu_discovery && 3461 !in6_localaddr(&inp->in6p_faddr)) 3462 mss = min(mss, V_tcp_v6mssdflt); 3463 } 3464 #endif 3465 #if defined(INET) && defined(INET6) 3466 else 3467 #endif 3468 #ifdef INET 3469 { 3470 mss = maxmtu - min_protoh; 3471 if (!V_path_mtu_discovery && 3472 !in_localaddr(inp->inp_faddr)) 3473 mss = min(mss, V_tcp_mssdflt); 3474 } 3475 #endif 3476 /* 3477 * XXX - The above conditional (mss = maxmtu - min_protoh) 3478 * probably violates the TCP spec. 3479 * The problem is that, since we don't know the 3480 * other end's MSS, we are supposed to use a conservative 3481 * default. But, if we do that, then MTU discovery will 3482 * never actually take place, because the conservative 3483 * default is much less than the MTUs typically seen 3484 * on the Internet today. For the moment, we'll sweep 3485 * this under the carpet. 3486 * 3487 * The conservative default might not actually be a problem 3488 * if the only case this occurs is when sending an initial 3489 * SYN with options and data to a host we've never talked 3490 * to before. Then, they will reply with an MSS value which 3491 * will get recorded and the new parameters should get 3492 * recomputed. For Further Study. 3493 */ 3494 } 3495 mss = min(mss, offer); 3496 3497 /* 3498 * Sanity check: make sure that maxopd will be large 3499 * enough to allow some data on segments even if the 3500 * all the option space is used (40bytes). Otherwise 3501 * funny things may happen in tcp_output. 3502 */ 3503 mss = max(mss, 64); 3504 3505 /* 3506 * maxopd stores the maximum length of data AND options 3507 * in a segment; maxseg is the amount of data in a normal 3508 * segment. We need to store this value (maxopd) apart 3509 * from maxseg, because now every segment carries options 3510 * and thus we normally have somewhat less data in segments. 3511 */ 3512 tp->t_maxopd = mss; 3513 3514 /* 3515 * origoffer==-1 indicates that no segments were received yet. 3516 * In this case we just guess. 3517 */ 3518 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3519 (origoffer == -1 || 3520 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3521 mss -= TCPOLEN_TSTAMP_APPA; 3522 3523 tp->t_maxseg = mss; 3524 } 3525 3526 void 3527 tcp_mss(struct tcpcb *tp, int offer) 3528 { 3529 int mss; 3530 u_long bufsize; 3531 struct inpcb *inp; 3532 struct socket *so; 3533 struct hc_metrics_lite metrics; 3534 int mtuflags = 0; 3535 3536 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3537 3538 tcp_mss_update(tp, offer, -1, &metrics, &mtuflags); 3539 3540 mss = tp->t_maxseg; 3541 inp = tp->t_inpcb; 3542 3543 /* 3544 * If there's a pipesize, change the socket buffer to that size, 3545 * don't change if sb_hiwat is different than default (then it 3546 * has been changed on purpose with setsockopt). 3547 * Make the socket buffers an integral number of mss units; 3548 * if the mss is larger than the socket buffer, decrease the mss. 3549 */ 3550 so = inp->inp_socket; 3551 SOCKBUF_LOCK(&so->so_snd); 3552 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 3553 bufsize = metrics.rmx_sendpipe; 3554 else 3555 bufsize = so->so_snd.sb_hiwat; 3556 if (bufsize < mss) 3557 mss = bufsize; 3558 else { 3559 bufsize = roundup(bufsize, mss); 3560 if (bufsize > sb_max) 3561 bufsize = sb_max; 3562 if (bufsize > so->so_snd.sb_hiwat) 3563 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3564 } 3565 SOCKBUF_UNLOCK(&so->so_snd); 3566 tp->t_maxseg = mss; 3567 3568 SOCKBUF_LOCK(&so->so_rcv); 3569 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 3570 bufsize = metrics.rmx_recvpipe; 3571 else 3572 bufsize = so->so_rcv.sb_hiwat; 3573 if (bufsize > mss) { 3574 bufsize = roundup(bufsize, mss); 3575 if (bufsize > sb_max) 3576 bufsize = sb_max; 3577 if (bufsize > so->so_rcv.sb_hiwat) 3578 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3579 } 3580 SOCKBUF_UNLOCK(&so->so_rcv); 3581 3582 /* Check the interface for TSO capabilities. */ 3583 if (mtuflags & CSUM_TSO) 3584 tp->t_flags |= TF_TSO; 3585 } 3586 3587 /* 3588 * Determine the MSS option to send on an outgoing SYN. 3589 */ 3590 int 3591 tcp_mssopt(struct in_conninfo *inc) 3592 { 3593 int mss = 0; 3594 u_long maxmtu = 0; 3595 u_long thcmtu = 0; 3596 size_t min_protoh; 3597 3598 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3599 3600 #ifdef INET6 3601 if (inc->inc_flags & INC_ISIPV6) { 3602 mss = V_tcp_v6mssdflt; 3603 maxmtu = tcp_maxmtu6(inc, NULL); 3604 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3605 } 3606 #endif 3607 #if defined(INET) && defined(INET6) 3608 else 3609 #endif 3610 #ifdef INET 3611 { 3612 mss = V_tcp_mssdflt; 3613 maxmtu = tcp_maxmtu(inc, NULL); 3614 min_protoh = sizeof(struct tcpiphdr); 3615 } 3616 #endif 3617 #if defined(INET6) || defined(INET) 3618 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3619 #endif 3620 3621 if (maxmtu && thcmtu) 3622 mss = min(maxmtu, thcmtu) - min_protoh; 3623 else if (maxmtu || thcmtu) 3624 mss = max(maxmtu, thcmtu) - min_protoh; 3625 3626 return (mss); 3627 } 3628 3629 3630 /* 3631 * On a partial ack arrives, force the retransmission of the 3632 * next unacknowledged segment. Do not clear tp->t_dupacks. 3633 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3634 * be started again. 3635 */ 3636 static void 3637 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3638 { 3639 tcp_seq onxt = tp->snd_nxt; 3640 u_long ocwnd = tp->snd_cwnd; 3641 3642 INP_WLOCK_ASSERT(tp->t_inpcb); 3643 3644 tcp_timer_activate(tp, TT_REXMT, 0); 3645 tp->t_rtttime = 0; 3646 tp->snd_nxt = th->th_ack; 3647 /* 3648 * Set snd_cwnd to one segment beyond acknowledged offset. 3649 * (tp->snd_una has not yet been updated when this function is called.) 3650 */ 3651 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3652 tp->t_flags |= TF_ACKNOW; 3653 (void) tcp_output(tp); 3654 tp->snd_cwnd = ocwnd; 3655 if (SEQ_GT(onxt, tp->snd_nxt)) 3656 tp->snd_nxt = onxt; 3657 /* 3658 * Partial window deflation. Relies on fact that tp->snd_una 3659 * not updated yet. 3660 */ 3661 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3662 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3663 else 3664 tp->snd_cwnd = 0; 3665 tp->snd_cwnd += tp->t_maxseg; 3666 } 3667