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