1 /*- 2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 30 */ 31 32 #include <sys/cdefs.h> 33 __FBSDID("$FreeBSD$"); 34 35 #include "opt_ipfw.h" /* for ipfw_fwd */ 36 #include "opt_inet.h" 37 #include "opt_inet6.h" 38 #include "opt_ipsec.h" 39 #include "opt_mac.h" 40 #include "opt_tcpdebug.h" 41 42 #include <sys/param.h> 43 #include <sys/kernel.h> 44 #include <sys/malloc.h> 45 #include <sys/mbuf.h> 46 #include <sys/proc.h> /* for proc0 declaration */ 47 #include <sys/protosw.h> 48 #include <sys/signalvar.h> 49 #include <sys/socket.h> 50 #include <sys/socketvar.h> 51 #include <sys/sysctl.h> 52 #include <sys/syslog.h> 53 #include <sys/systm.h> 54 #include <sys/vimage.h> 55 56 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 57 58 #include <vm/uma.h> 59 60 #include <net/if.h> 61 #include <net/route.h> 62 63 #define TCPSTATES /* for logging */ 64 65 #include <netinet/in.h> 66 #include <netinet/in_pcb.h> 67 #include <netinet/in_systm.h> 68 #include <netinet/in_var.h> 69 #include <netinet/ip.h> 70 #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 71 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 72 #include <netinet/ip_var.h> 73 #include <netinet/ip_options.h> 74 #include <netinet/ip6.h> 75 #include <netinet/icmp6.h> 76 #include <netinet6/in6_pcb.h> 77 #include <netinet6/ip6_var.h> 78 #include <netinet6/nd6.h> 79 #include <netinet/tcp.h> 80 #include <netinet/tcp_fsm.h> 81 #include <netinet/tcp_seq.h> 82 #include <netinet/tcp_timer.h> 83 #include <netinet/tcp_var.h> 84 #include <netinet6/tcp6_var.h> 85 #include <netinet/tcpip.h> 86 #include <netinet/tcp_syncache.h> 87 #ifdef TCPDEBUG 88 #include <netinet/tcp_debug.h> 89 #endif /* TCPDEBUG */ 90 #include <netinet/vinet.h> 91 92 #ifdef INET6 93 #include <netinet6/vinet6.h> 94 #endif 95 96 #ifdef IPSEC 97 #include <netipsec/ipsec.h> 98 #include <netipsec/ipsec6.h> 99 #endif /*IPSEC*/ 100 101 #include <machine/in_cksum.h> 102 103 #include <security/mac/mac_framework.h> 104 105 static const int tcprexmtthresh = 3; 106 107 #ifdef VIMAGE_GLOBALS 108 struct tcpstat tcpstat; 109 int blackhole; 110 int tcp_delack_enabled; 111 int drop_synfin; 112 int tcp_do_rfc3042; 113 int tcp_do_rfc3390; 114 int tcp_do_ecn; 115 int tcp_ecn_maxretries; 116 int tcp_insecure_rst; 117 int tcp_do_autorcvbuf; 118 int tcp_autorcvbuf_inc; 119 int tcp_autorcvbuf_max; 120 int tcp_do_rfc3465; 121 int tcp_abc_l_var; 122 #endif 123 124 SYSCTL_V_STRUCT(V_NET, vnet_inet, _net_inet_tcp, TCPCTL_STATS, stats, 125 CTLFLAG_RW, tcpstat , tcpstat, 126 "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 127 128 int tcp_log_in_vain = 0; 129 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 130 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports"); 131 132 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 133 blackhole, 0, "Do not send RST on segments to closed ports"); 134 135 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, delayed_ack, 136 CTLFLAG_RW, tcp_delack_enabled, 0, 137 "Delay ACK to try and piggyback it onto a data packet"); 138 139 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, drop_synfin, 140 CTLFLAG_RW, drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 141 142 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 143 tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 144 145 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 146 tcp_do_rfc3390, 0, 147 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 148 149 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 150 tcp_do_rfc3465, 0, 151 "Enable RFC 3465 (Appropriate Byte Counting)"); 152 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 153 tcp_abc_l_var, 2, 154 "Cap the max cwnd increment during slow-start to this number of segments"); 155 156 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 157 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_ecn, OID_AUTO, enable, 158 CTLFLAG_RW, tcp_do_ecn, 0, "TCP ECN support"); 159 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp_ecn, OID_AUTO, maxretries, 160 CTLFLAG_RW, tcp_ecn_maxretries, 0, "Max retries before giving up on ECN"); 161 162 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, insecure_rst, 163 CTLFLAG_RW, tcp_insecure_rst, 0, 164 "Follow the old (insecure) criteria for accepting RST packets"); 165 166 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, recvbuf_auto, 167 CTLFLAG_RW, tcp_do_autorcvbuf, 0, 168 "Enable automatic receive buffer sizing"); 169 170 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, recvbuf_inc, 171 CTLFLAG_RW, tcp_autorcvbuf_inc, 0, 172 "Incrementor step size of automatic receive buffer"); 173 174 SYSCTL_V_INT(V_NET, vnet_inet, _net_inet_tcp, OID_AUTO, recvbuf_max, 175 CTLFLAG_RW, tcp_autorcvbuf_max, 0, 176 "Max size of automatic receive buffer"); 177 178 int tcp_read_locking = 1; 179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, read_locking, CTLFLAG_RW, 180 &tcp_read_locking, 0, "Enable read locking strategy"); 181 182 int tcp_rlock_atfirst; 183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rlock_atfirst, CTLFLAG_RD, 184 &tcp_rlock_atfirst, 0, ""); 185 186 int tcp_wlock_atfirst; 187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_wlock_atfirst, CTLFLAG_RD, 188 &tcp_wlock_atfirst, 0, ""); 189 190 int tcp_wlock_upgraded; 191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, wlock_upgraded, CTLFLAG_RD, 192 &tcp_wlock_upgraded, 0, ""); 193 194 int tcp_wlock_relocked; 195 SYSCTL_INT(_net_inet_tcp, OID_AUTO, wlock_relocked, CTLFLAG_RD, 196 &tcp_wlock_relocked, 0, ""); 197 198 int tcp_wlock_looped; 199 SYSCTL_INT(_net_inet_tcp, OID_AUTO, wlock_looped, CTLFLAG_RD, 200 &tcp_wlock_looped, 0, ""); 201 202 #ifdef VIMAGE_GLOBALS 203 struct inpcbhead tcb; 204 struct inpcbinfo tcbinfo; 205 #endif 206 #define tcb6 tcb /* for KAME src sync over BSD*'s */ 207 208 static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 209 static void tcp_do_segment(struct mbuf *, struct tcphdr *, 210 struct socket *, struct tcpcb *, int, int, uint8_t, 211 int); 212 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 213 struct tcpcb *, int, int); 214 static void tcp_pulloutofband(struct socket *, 215 struct tcphdr *, struct mbuf *, int); 216 static void tcp_xmit_timer(struct tcpcb *, int); 217 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 218 static void inline 219 tcp_congestion_exp(struct tcpcb *); 220 221 static void inline 222 tcp_congestion_exp(struct tcpcb *tp) 223 { 224 u_int win; 225 226 win = min(tp->snd_wnd, tp->snd_cwnd) / 227 2 / tp->t_maxseg; 228 if (win < 2) 229 win = 2; 230 tp->snd_ssthresh = win * tp->t_maxseg; 231 ENTER_FASTRECOVERY(tp); 232 tp->snd_recover = tp->snd_max; 233 if (tp->t_flags & TF_ECN_PERMIT) 234 tp->t_flags |= TF_ECN_SND_CWR; 235 } 236 237 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 238 #ifdef INET6 239 #define ND6_HINT(tp) \ 240 do { \ 241 if ((tp) && (tp)->t_inpcb && \ 242 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 243 nd6_nud_hint(NULL, NULL, 0); \ 244 } while (0) 245 #else 246 #define ND6_HINT(tp) 247 #endif 248 249 /* 250 * Indicate whether this ack should be delayed. We can delay the ack if 251 * - there is no delayed ack timer in progress and 252 * - our last ack wasn't a 0-sized window. We never want to delay 253 * the ack that opens up a 0-sized window and 254 * - delayed acks are enabled or 255 * - this is a half-synchronized T/TCP connection. 256 */ 257 #define DELAY_ACK(tp) \ 258 ((!tcp_timer_active(tp, TT_DELACK) && \ 259 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 260 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 261 262 /* 263 * TCP input handling is split into multiple parts: 264 * tcp6_input is a thin wrapper around tcp_input for the extended 265 * ip6_protox[] call format in ip6_input 266 * tcp_input handles primary segment validation, inpcb lookup and 267 * SYN processing on listen sockets 268 * tcp_do_segment processes the ACK and text of the segment for 269 * establishing, established and closing connections 270 */ 271 #ifdef INET6 272 int 273 tcp6_input(struct mbuf **mp, int *offp, int proto) 274 { 275 INIT_VNET_INET6(curvnet); 276 struct mbuf *m = *mp; 277 struct in6_ifaddr *ia6; 278 279 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 280 281 /* 282 * draft-itojun-ipv6-tcp-to-anycast 283 * better place to put this in? 284 */ 285 ia6 = ip6_getdstifaddr(m); 286 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 287 struct ip6_hdr *ip6; 288 289 ip6 = mtod(m, struct ip6_hdr *); 290 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 291 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 292 return IPPROTO_DONE; 293 } 294 295 tcp_input(m, *offp); 296 return IPPROTO_DONE; 297 } 298 #endif 299 300 void 301 tcp_input(struct mbuf *m, int off0) 302 { 303 INIT_VNET_INET(curvnet); 304 #ifdef INET6 305 INIT_VNET_INET6(curvnet); 306 #endif 307 #ifdef IPSEC 308 INIT_VNET_IPSEC(curvnet); 309 #endif 310 struct tcphdr *th; 311 struct ip *ip = NULL; 312 struct ipovly *ipov; 313 struct inpcb *inp = NULL; 314 struct tcpcb *tp = NULL; 315 struct socket *so = NULL; 316 u_char *optp = NULL; 317 int optlen = 0; 318 int len, tlen, off; 319 int drop_hdrlen; 320 int thflags; 321 int rstreason = 0; /* For badport_bandlim accounting purposes */ 322 uint8_t iptos; 323 #ifdef IPFIREWALL_FORWARD 324 struct m_tag *fwd_tag; 325 #endif 326 #ifdef INET6 327 struct ip6_hdr *ip6 = NULL; 328 int isipv6; 329 #else 330 const void *ip6 = NULL; 331 const int isipv6 = 0; 332 #endif 333 struct tcpopt to; /* options in this segment */ 334 char *s = NULL; /* address and port logging */ 335 int ti_locked; 336 #define TI_UNLOCKED 1 337 #define TI_RLOCKED 2 338 #define TI_WLOCKED 3 339 340 #ifdef TCPDEBUG 341 /* 342 * The size of tcp_saveipgen must be the size of the max ip header, 343 * now IPv6. 344 */ 345 u_char tcp_saveipgen[IP6_HDR_LEN]; 346 struct tcphdr tcp_savetcp; 347 short ostate = 0; 348 #endif 349 350 #ifdef INET6 351 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 352 #endif 353 354 to.to_flags = 0; 355 V_tcpstat.tcps_rcvtotal++; 356 357 if (isipv6) { 358 #ifdef INET6 359 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 360 ip6 = mtod(m, struct ip6_hdr *); 361 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 362 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 363 V_tcpstat.tcps_rcvbadsum++; 364 goto drop; 365 } 366 th = (struct tcphdr *)((caddr_t)ip6 + off0); 367 368 /* 369 * Be proactive about unspecified IPv6 address in source. 370 * As we use all-zero to indicate unbounded/unconnected pcb, 371 * unspecified IPv6 address can be used to confuse us. 372 * 373 * Note that packets with unspecified IPv6 destination is 374 * already dropped in ip6_input. 375 */ 376 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 377 /* XXX stat */ 378 goto drop; 379 } 380 #else 381 th = NULL; /* XXX: Avoid compiler warning. */ 382 #endif 383 } else { 384 /* 385 * Get IP and TCP header together in first mbuf. 386 * Note: IP leaves IP header in first mbuf. 387 */ 388 if (off0 > sizeof (struct ip)) { 389 ip_stripoptions(m, (struct mbuf *)0); 390 off0 = sizeof(struct ip); 391 } 392 if (m->m_len < sizeof (struct tcpiphdr)) { 393 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 394 == NULL) { 395 V_tcpstat.tcps_rcvshort++; 396 return; 397 } 398 } 399 ip = mtod(m, struct ip *); 400 ipov = (struct ipovly *)ip; 401 th = (struct tcphdr *)((caddr_t)ip + off0); 402 tlen = ip->ip_len; 403 404 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 405 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 406 th->th_sum = m->m_pkthdr.csum_data; 407 else 408 th->th_sum = in_pseudo(ip->ip_src.s_addr, 409 ip->ip_dst.s_addr, 410 htonl(m->m_pkthdr.csum_data + 411 ip->ip_len + 412 IPPROTO_TCP)); 413 th->th_sum ^= 0xffff; 414 #ifdef TCPDEBUG 415 ipov->ih_len = (u_short)tlen; 416 ipov->ih_len = htons(ipov->ih_len); 417 #endif 418 } else { 419 /* 420 * Checksum extended TCP header and data. 421 */ 422 len = sizeof (struct ip) + tlen; 423 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 424 ipov->ih_len = (u_short)tlen; 425 ipov->ih_len = htons(ipov->ih_len); 426 th->th_sum = in_cksum(m, len); 427 } 428 if (th->th_sum) { 429 V_tcpstat.tcps_rcvbadsum++; 430 goto drop; 431 } 432 /* Re-initialization for later version check */ 433 ip->ip_v = IPVERSION; 434 } 435 436 #ifdef INET6 437 if (isipv6) 438 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 439 else 440 #endif 441 iptos = ip->ip_tos; 442 443 /* 444 * Check that TCP offset makes sense, 445 * pull out TCP options and adjust length. XXX 446 */ 447 off = th->th_off << 2; 448 if (off < sizeof (struct tcphdr) || off > tlen) { 449 V_tcpstat.tcps_rcvbadoff++; 450 goto drop; 451 } 452 tlen -= off; /* tlen is used instead of ti->ti_len */ 453 if (off > sizeof (struct tcphdr)) { 454 if (isipv6) { 455 #ifdef INET6 456 IP6_EXTHDR_CHECK(m, off0, off, ); 457 ip6 = mtod(m, struct ip6_hdr *); 458 th = (struct tcphdr *)((caddr_t)ip6 + off0); 459 #endif 460 } else { 461 if (m->m_len < sizeof(struct ip) + off) { 462 if ((m = m_pullup(m, sizeof (struct ip) + off)) 463 == NULL) { 464 V_tcpstat.tcps_rcvshort++; 465 return; 466 } 467 ip = mtod(m, struct ip *); 468 ipov = (struct ipovly *)ip; 469 th = (struct tcphdr *)((caddr_t)ip + off0); 470 } 471 } 472 optlen = off - sizeof (struct tcphdr); 473 optp = (u_char *)(th + 1); 474 } 475 thflags = th->th_flags; 476 477 /* 478 * Convert TCP protocol specific fields to host format. 479 */ 480 th->th_seq = ntohl(th->th_seq); 481 th->th_ack = ntohl(th->th_ack); 482 th->th_win = ntohs(th->th_win); 483 th->th_urp = ntohs(th->th_urp); 484 485 /* 486 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 487 */ 488 drop_hdrlen = off0 + off; 489 490 /* 491 * Locate pcb for segment, which requires a lock on tcbinfo. 492 * Optimisticaly acquire a global read lock rather than a write lock 493 * unless header flags necessarily imply a state change. There are 494 * two cases where we might discover later we need a write lock 495 * despite the flags: ACKs moving a connection out of the syncache, 496 * and ACKs for a connection in TIMEWAIT. 497 */ 498 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 499 tcp_read_locking == 0) { 500 INP_INFO_WLOCK(&V_tcbinfo); 501 ti_locked = TI_WLOCKED; 502 tcp_wlock_atfirst++; 503 } else { 504 INP_INFO_RLOCK(&V_tcbinfo); 505 ti_locked = TI_RLOCKED; 506 tcp_rlock_atfirst++; 507 } 508 509 findpcb: 510 #ifdef INVARIANTS 511 if (ti_locked == TI_RLOCKED) 512 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 513 else if (ti_locked == TI_WLOCKED) 514 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 515 else 516 panic("%s: findpcb ti_locked %d\n", __func__, ti_locked); 517 #endif 518 519 #ifdef IPFIREWALL_FORWARD 520 /* 521 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 522 */ 523 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 524 525 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 526 struct sockaddr_in *next_hop; 527 528 next_hop = (struct sockaddr_in *)(fwd_tag+1); 529 /* 530 * Transparently forwarded. Pretend to be the destination. 531 * already got one like this? 532 */ 533 inp = in_pcblookup_hash(&V_tcbinfo, 534 ip->ip_src, th->th_sport, 535 ip->ip_dst, th->th_dport, 536 0, m->m_pkthdr.rcvif); 537 if (!inp) { 538 /* It's new. Try to find the ambushing socket. */ 539 inp = in_pcblookup_hash(&V_tcbinfo, 540 ip->ip_src, th->th_sport, 541 next_hop->sin_addr, 542 next_hop->sin_port ? 543 ntohs(next_hop->sin_port) : 544 th->th_dport, 545 INPLOOKUP_WILDCARD, 546 m->m_pkthdr.rcvif); 547 } 548 /* Remove the tag from the packet. We don't need it anymore. */ 549 m_tag_delete(m, fwd_tag); 550 } else 551 #endif /* IPFIREWALL_FORWARD */ 552 { 553 if (isipv6) { 554 #ifdef INET6 555 inp = in6_pcblookup_hash(&V_tcbinfo, 556 &ip6->ip6_src, th->th_sport, 557 &ip6->ip6_dst, th->th_dport, 558 INPLOOKUP_WILDCARD, 559 m->m_pkthdr.rcvif); 560 #endif 561 } else 562 inp = in_pcblookup_hash(&V_tcbinfo, 563 ip->ip_src, th->th_sport, 564 ip->ip_dst, th->th_dport, 565 INPLOOKUP_WILDCARD, 566 m->m_pkthdr.rcvif); 567 } 568 569 /* 570 * If the INPCB does not exist then all data in the incoming 571 * segment is discarded and an appropriate RST is sent back. 572 * XXX MRT Send RST using which routing table? 573 */ 574 if (inp == NULL) { 575 /* 576 * Log communication attempts to ports that are not 577 * in use. 578 */ 579 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 580 tcp_log_in_vain == 2) { 581 if ((s = tcp_log_addrs(NULL, th, (void *)ip, ip6))) 582 log(LOG_INFO, "%s; %s: Connection attempt " 583 "to closed port\n", s, __func__); 584 } 585 /* 586 * When blackholing do not respond with a RST but 587 * completely ignore the segment and drop it. 588 */ 589 if ((V_blackhole == 1 && (thflags & TH_SYN)) || 590 V_blackhole == 2) 591 goto dropunlock; 592 593 rstreason = BANDLIM_RST_CLOSEDPORT; 594 goto dropwithreset; 595 } 596 INP_WLOCK(inp); 597 598 #ifdef IPSEC 599 #ifdef INET6 600 if (isipv6 && ipsec6_in_reject(m, inp)) { 601 V_ipsec6stat.in_polvio++; 602 goto dropunlock; 603 } else 604 #endif /* INET6 */ 605 if (ipsec4_in_reject(m, inp) != 0) { 606 V_ipsec4stat.in_polvio++; 607 goto dropunlock; 608 } 609 #endif /* IPSEC */ 610 611 /* 612 * Check the minimum TTL for socket. 613 */ 614 if (inp->inp_ip_minttl != 0) { 615 #ifdef INET6 616 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 617 goto dropunlock; 618 else 619 #endif 620 if (inp->inp_ip_minttl > ip->ip_ttl) 621 goto dropunlock; 622 } 623 624 /* 625 * A previous connection in TIMEWAIT state is supposed to catch stray 626 * or duplicate segments arriving late. If this segment was a 627 * legitimate new connection attempt the old INPCB gets removed and 628 * we can try again to find a listening socket. 629 * 630 * At this point, due to earlier optimism, we may hold a read lock on 631 * the inpcbinfo, rather than a write lock. If so, we need to 632 * upgrade, or if that fails, acquire a reference on the inpcb, drop 633 * all locks, acquire a global write lock, and then re-acquire the 634 * inpcb lock. We may at that point discover that another thread has 635 * tried to free the inpcb, in which case we need to loop back and 636 * try to find a new inpcb to deliver to. 637 */ 638 if (inp->inp_vflag & INP_TIMEWAIT) { 639 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 640 ("%s: INP_TIMEWAIT ti_locked %d", __func__, ti_locked)); 641 642 if (ti_locked == TI_RLOCKED) { 643 if (rw_try_upgrade(&V_tcbinfo.ipi_lock) == 0) { 644 in_pcbref(inp); 645 INP_WUNLOCK(inp); 646 INP_INFO_RUNLOCK(&V_tcbinfo); 647 INP_INFO_WLOCK(&V_tcbinfo); 648 ti_locked = TI_WLOCKED; 649 INP_WLOCK(inp); 650 if (in_pcbrele(inp)) { 651 tcp_wlock_looped++; 652 inp = NULL; 653 goto findpcb; 654 } 655 tcp_wlock_relocked++; 656 } else { 657 ti_locked = TI_WLOCKED; 658 tcp_wlock_upgraded++; 659 } 660 } 661 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 662 663 if (thflags & TH_SYN) 664 tcp_dooptions(&to, optp, optlen, TO_SYN); 665 /* 666 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 667 */ 668 if (tcp_twcheck(inp, &to, th, m, tlen)) 669 goto findpcb; 670 INP_INFO_WUNLOCK(&V_tcbinfo); 671 return; 672 } 673 /* 674 * The TCPCB may no longer exist if the connection is winding 675 * down or it is in the CLOSED state. Either way we drop the 676 * segment and send an appropriate response. 677 */ 678 tp = intotcpcb(inp); 679 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 680 rstreason = BANDLIM_RST_CLOSEDPORT; 681 goto dropwithreset; 682 } 683 684 /* 685 * We've identified a valid inpcb, but it could be that we need an 686 * inpcbinfo write lock and have only a read lock. In this case, 687 * attempt to upgrade/relock using the same strategy as the TIMEWAIT 688 * case above. 689 */ 690 if (tp->t_state != TCPS_ESTABLISHED || 691 (thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 692 tcp_read_locking == 0) { 693 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 694 ("%s: upgrade check ti_locked %d", __func__, ti_locked)); 695 696 if (ti_locked == TI_RLOCKED) { 697 if (rw_try_upgrade(&V_tcbinfo.ipi_lock) == 0) { 698 in_pcbref(inp); 699 INP_WUNLOCK(inp); 700 INP_INFO_RUNLOCK(&V_tcbinfo); 701 INP_INFO_WLOCK(&V_tcbinfo); 702 ti_locked = TI_WLOCKED; 703 INP_WLOCK(inp); 704 if (in_pcbrele(inp)) { 705 tcp_wlock_looped++; 706 inp = NULL; 707 goto findpcb; 708 } 709 tcp_wlock_relocked++; 710 } else { 711 ti_locked = TI_WLOCKED; 712 tcp_wlock_upgraded++; 713 } 714 } 715 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 716 } 717 718 #ifdef MAC 719 INP_WLOCK_ASSERT(inp); 720 if (mac_inpcb_check_deliver(inp, m)) 721 goto dropunlock; 722 #endif 723 so = inp->inp_socket; 724 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 725 #ifdef TCPDEBUG 726 if (so->so_options & SO_DEBUG) { 727 ostate = tp->t_state; 728 if (isipv6) { 729 #ifdef INET6 730 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 731 #endif 732 } else 733 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 734 tcp_savetcp = *th; 735 } 736 #endif 737 /* 738 * When the socket is accepting connections (the INPCB is in LISTEN 739 * state) we look into the SYN cache if this is a new connection 740 * attempt or the completion of a previous one. 741 */ 742 if (so->so_options & SO_ACCEPTCONN) { 743 struct in_conninfo inc; 744 745 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 746 "tp not listening", __func__)); 747 748 bzero(&inc, sizeof(inc)); 749 #ifdef INET6 750 if (isipv6) { 751 inc.inc_flags |= INC_ISIPV6; 752 inc.inc6_faddr = ip6->ip6_src; 753 inc.inc6_laddr = ip6->ip6_dst; 754 } else 755 #endif 756 { 757 inc.inc_faddr = ip->ip_src; 758 inc.inc_laddr = ip->ip_dst; 759 } 760 inc.inc_fport = th->th_sport; 761 inc.inc_lport = th->th_dport; 762 763 /* 764 * Check for an existing connection attempt in syncache if 765 * the flag is only ACK. A successful lookup creates a new 766 * socket appended to the listen queue in SYN_RECEIVED state. 767 */ 768 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 769 /* 770 * Parse the TCP options here because 771 * syncookies need access to the reflected 772 * timestamp. 773 */ 774 tcp_dooptions(&to, optp, optlen, 0); 775 /* 776 * NB: syncache_expand() doesn't unlock 777 * inp and tcpinfo locks. 778 */ 779 if (!syncache_expand(&inc, &to, th, &so, m)) { 780 /* 781 * No syncache entry or ACK was not 782 * for our SYN/ACK. Send a RST. 783 * NB: syncache did its own logging 784 * of the failure cause. 785 */ 786 rstreason = BANDLIM_RST_OPENPORT; 787 goto dropwithreset; 788 } 789 if (so == NULL) { 790 /* 791 * We completed the 3-way handshake 792 * but could not allocate a socket 793 * either due to memory shortage, 794 * listen queue length limits or 795 * global socket limits. Send RST 796 * or wait and have the remote end 797 * retransmit the ACK for another 798 * try. 799 */ 800 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 801 log(LOG_DEBUG, "%s; %s: Listen socket: " 802 "Socket allocation failed due to " 803 "limits or memory shortage, %s\n", 804 s, __func__, 805 V_tcp_sc_rst_sock_fail ? 806 "sending RST" : "try again"); 807 if (V_tcp_sc_rst_sock_fail) { 808 rstreason = BANDLIM_UNLIMITED; 809 goto dropwithreset; 810 } else 811 goto dropunlock; 812 } 813 /* 814 * Socket is created in state SYN_RECEIVED. 815 * Unlock the listen socket, lock the newly 816 * created socket and update the tp variable. 817 */ 818 INP_WUNLOCK(inp); /* listen socket */ 819 inp = sotoinpcb(so); 820 INP_WLOCK(inp); /* new connection */ 821 tp = intotcpcb(inp); 822 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 823 ("%s: ", __func__)); 824 /* 825 * Process the segment and the data it 826 * contains. tcp_do_segment() consumes 827 * the mbuf chain and unlocks the inpcb. 828 */ 829 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 830 iptos, ti_locked); 831 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 832 return; 833 } 834 /* 835 * Segment flag validation for new connection attempts: 836 * 837 * Our (SYN|ACK) response was rejected. 838 * Check with syncache and remove entry to prevent 839 * retransmits. 840 * 841 * NB: syncache_chkrst does its own logging of failure 842 * causes. 843 */ 844 if (thflags & TH_RST) { 845 syncache_chkrst(&inc, th); 846 goto dropunlock; 847 } 848 /* 849 * We can't do anything without SYN. 850 */ 851 if ((thflags & TH_SYN) == 0) { 852 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 853 log(LOG_DEBUG, "%s; %s: Listen socket: " 854 "SYN is missing, segment ignored\n", 855 s, __func__); 856 V_tcpstat.tcps_badsyn++; 857 goto dropunlock; 858 } 859 /* 860 * (SYN|ACK) is bogus on a listen socket. 861 */ 862 if (thflags & TH_ACK) { 863 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 864 log(LOG_DEBUG, "%s; %s: Listen socket: " 865 "SYN|ACK invalid, segment rejected\n", 866 s, __func__); 867 syncache_badack(&inc); /* XXX: Not needed! */ 868 V_tcpstat.tcps_badsyn++; 869 rstreason = BANDLIM_RST_OPENPORT; 870 goto dropwithreset; 871 } 872 /* 873 * If the drop_synfin option is enabled, drop all 874 * segments with both the SYN and FIN bits set. 875 * This prevents e.g. nmap from identifying the 876 * TCP/IP stack. 877 * XXX: Poor reasoning. nmap has other methods 878 * and is constantly refining its stack detection 879 * strategies. 880 * XXX: This is a violation of the TCP specification 881 * and was used by RFC1644. 882 */ 883 if ((thflags & TH_FIN) && V_drop_synfin) { 884 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 885 log(LOG_DEBUG, "%s; %s: Listen socket: " 886 "SYN|FIN segment ignored (based on " 887 "sysctl setting)\n", s, __func__); 888 V_tcpstat.tcps_badsyn++; 889 goto dropunlock; 890 } 891 /* 892 * Segment's flags are (SYN) or (SYN|FIN). 893 * 894 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 895 * as they do not affect the state of the TCP FSM. 896 * The data pointed to by TH_URG and th_urp is ignored. 897 */ 898 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 899 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 900 KASSERT(thflags & (TH_SYN), 901 ("%s: Listen socket: TH_SYN not set", __func__)); 902 #ifdef INET6 903 /* 904 * If deprecated address is forbidden, 905 * we do not accept SYN to deprecated interface 906 * address to prevent any new inbound connection from 907 * getting established. 908 * When we do not accept SYN, we send a TCP RST, 909 * with deprecated source address (instead of dropping 910 * it). We compromise it as it is much better for peer 911 * to send a RST, and RST will be the final packet 912 * for the exchange. 913 * 914 * If we do not forbid deprecated addresses, we accept 915 * the SYN packet. RFC2462 does not suggest dropping 916 * SYN in this case. 917 * If we decipher RFC2462 5.5.4, it says like this: 918 * 1. use of deprecated addr with existing 919 * communication is okay - "SHOULD continue to be 920 * used" 921 * 2. use of it with new communication: 922 * (2a) "SHOULD NOT be used if alternate address 923 * with sufficient scope is available" 924 * (2b) nothing mentioned otherwise. 925 * Here we fall into (2b) case as we have no choice in 926 * our source address selection - we must obey the peer. 927 * 928 * The wording in RFC2462 is confusing, and there are 929 * multiple description text for deprecated address 930 * handling - worse, they are not exactly the same. 931 * I believe 5.5.4 is the best one, so we follow 5.5.4. 932 */ 933 if (isipv6 && !V_ip6_use_deprecated) { 934 struct in6_ifaddr *ia6; 935 936 if ((ia6 = ip6_getdstifaddr(m)) && 937 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 938 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 939 log(LOG_DEBUG, "%s; %s: Listen socket: " 940 "Connection attempt to deprecated " 941 "IPv6 address rejected\n", 942 s, __func__); 943 rstreason = BANDLIM_RST_OPENPORT; 944 goto dropwithreset; 945 } 946 } 947 #endif 948 /* 949 * Basic sanity checks on incoming SYN requests: 950 * Don't respond if the destination is a link layer 951 * broadcast according to RFC1122 4.2.3.10, p. 104. 952 * If it is from this socket it must be forged. 953 * Don't respond if the source or destination is a 954 * global or subnet broad- or multicast address. 955 * Note that it is quite possible to receive unicast 956 * link-layer packets with a broadcast IP address. Use 957 * in_broadcast() to find them. 958 */ 959 if (m->m_flags & (M_BCAST|M_MCAST)) { 960 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 961 log(LOG_DEBUG, "%s; %s: Listen socket: " 962 "Connection attempt from broad- or multicast " 963 "link layer address ignored\n", s, __func__); 964 goto dropunlock; 965 } 966 if (isipv6) { 967 #ifdef INET6 968 if (th->th_dport == th->th_sport && 969 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 970 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 971 log(LOG_DEBUG, "%s; %s: Listen socket: " 972 "Connection attempt to/from self " 973 "ignored\n", s, __func__); 974 goto dropunlock; 975 } 976 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 977 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 978 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 979 log(LOG_DEBUG, "%s; %s: Listen socket: " 980 "Connection attempt from/to multicast " 981 "address ignored\n", s, __func__); 982 goto dropunlock; 983 } 984 #endif 985 } else { 986 if (th->th_dport == th->th_sport && 987 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 988 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 989 log(LOG_DEBUG, "%s; %s: Listen socket: " 990 "Connection attempt from/to self " 991 "ignored\n", s, __func__); 992 goto dropunlock; 993 } 994 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 995 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 996 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 997 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 998 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 999 log(LOG_DEBUG, "%s; %s: Listen socket: " 1000 "Connection attempt from/to broad- " 1001 "or multicast address ignored\n", 1002 s, __func__); 1003 goto dropunlock; 1004 } 1005 } 1006 /* 1007 * SYN appears to be valid. Create compressed TCP state 1008 * for syncache. 1009 */ 1010 #ifdef TCPDEBUG 1011 if (so->so_options & SO_DEBUG) 1012 tcp_trace(TA_INPUT, ostate, tp, 1013 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1014 #endif 1015 tcp_dooptions(&to, optp, optlen, TO_SYN); 1016 syncache_add(&inc, &to, th, inp, &so, m); 1017 /* 1018 * Entry added to syncache and mbuf consumed. 1019 * Everything already unlocked by syncache_add(). 1020 */ 1021 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1022 return; 1023 } 1024 1025 /* 1026 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 1027 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 1028 * the inpcb, and unlocks pcbinfo. 1029 */ 1030 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1031 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1032 return; 1033 1034 dropwithreset: 1035 if (ti_locked == TI_RLOCKED) 1036 INP_INFO_RUNLOCK(&V_tcbinfo); 1037 else if (ti_locked == TI_WLOCKED) 1038 INP_INFO_WUNLOCK(&V_tcbinfo); 1039 else 1040 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked); 1041 ti_locked = TI_UNLOCKED; 1042 1043 if (inp != NULL) { 1044 tcp_dropwithreset(m, th, tp, tlen, rstreason); 1045 INP_WUNLOCK(inp); 1046 } else 1047 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1048 m = NULL; /* mbuf chain got consumed. */ 1049 goto drop; 1050 1051 dropunlock: 1052 if (ti_locked == TI_RLOCKED) 1053 INP_INFO_RUNLOCK(&V_tcbinfo); 1054 else if (ti_locked == TI_WLOCKED) 1055 INP_INFO_WUNLOCK(&V_tcbinfo); 1056 else 1057 panic("%s: dropunlock ti_locked %d", __func__, ti_locked); 1058 ti_locked = TI_UNLOCKED; 1059 1060 if (inp != NULL) 1061 INP_WUNLOCK(inp); 1062 1063 drop: 1064 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1065 if (s != NULL) 1066 free(s, M_TCPLOG); 1067 if (m != NULL) 1068 m_freem(m); 1069 } 1070 1071 static void 1072 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1073 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1074 int ti_locked) 1075 { 1076 INIT_VNET_INET(tp->t_vnet); 1077 int thflags, acked, ourfinisacked, needoutput = 0; 1078 int rstreason, todrop, win; 1079 u_long tiwin; 1080 struct tcpopt to; 1081 1082 #ifdef TCPDEBUG 1083 /* 1084 * The size of tcp_saveipgen must be the size of the max ip header, 1085 * now IPv6. 1086 */ 1087 u_char tcp_saveipgen[IP6_HDR_LEN]; 1088 struct tcphdr tcp_savetcp; 1089 short ostate = 0; 1090 #endif 1091 thflags = th->th_flags; 1092 1093 /* 1094 * If this is either a state-changing packet or current state isn't 1095 * established, we require a write lock on tcbinfo. Otherwise, we 1096 * allow either a read lock or a write lock, as we may have acquired 1097 * a write lock due to a race. 1098 * 1099 * Require a global write lock for SYN/FIN/RST segments or 1100 * non-established connections; otherwise accept either a read or 1101 * write lock, as we may have conservatively acquired a write lock in 1102 * certain cases in tcp_input() (is this still true?). Currently we 1103 * will never enter with no lock, so we try to drop it quickly in the 1104 * common pure ack/pure data cases. 1105 */ 1106 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1107 tp->t_state != TCPS_ESTABLISHED) { 1108 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1109 "SYN/FIN/RST/!EST", __func__, ti_locked)); 1110 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1111 } else { 1112 #ifdef INVARIANTS 1113 if (ti_locked == TI_RLOCKED) 1114 INP_INFO_RLOCK_ASSERT(&V_tcbinfo); 1115 else if (ti_locked == TI_WLOCKED) 1116 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1117 else 1118 panic("%s: ti_locked %d for EST", __func__, 1119 ti_locked); 1120 #endif 1121 } 1122 INP_WLOCK_ASSERT(tp->t_inpcb); 1123 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1124 __func__)); 1125 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1126 __func__)); 1127 1128 /* 1129 * Segment received on connection. 1130 * Reset idle time and keep-alive timer. 1131 * XXX: This should be done after segment 1132 * validation to ignore broken/spoofed segs. 1133 */ 1134 tp->t_rcvtime = ticks; 1135 if (TCPS_HAVEESTABLISHED(tp->t_state)) 1136 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1137 1138 /* 1139 * Unscale the window into a 32-bit value. 1140 * For the SYN_SENT state the scale is zero. 1141 */ 1142 tiwin = th->th_win << tp->snd_scale; 1143 1144 /* 1145 * TCP ECN processing. 1146 */ 1147 if (tp->t_flags & TF_ECN_PERMIT) { 1148 switch (iptos & IPTOS_ECN_MASK) { 1149 case IPTOS_ECN_CE: 1150 tp->t_flags |= TF_ECN_SND_ECE; 1151 V_tcpstat.tcps_ecn_ce++; 1152 break; 1153 case IPTOS_ECN_ECT0: 1154 V_tcpstat.tcps_ecn_ect0++; 1155 break; 1156 case IPTOS_ECN_ECT1: 1157 V_tcpstat.tcps_ecn_ect1++; 1158 break; 1159 } 1160 1161 if (thflags & TH_CWR) 1162 tp->t_flags &= ~TF_ECN_SND_ECE; 1163 1164 /* 1165 * Congestion experienced. 1166 * Ignore if we are already trying to recover. 1167 */ 1168 if ((thflags & TH_ECE) && 1169 SEQ_LEQ(th->th_ack, tp->snd_recover)) { 1170 V_tcpstat.tcps_ecn_rcwnd++; 1171 tcp_congestion_exp(tp); 1172 } 1173 } 1174 1175 /* 1176 * Parse options on any incoming segment. 1177 */ 1178 tcp_dooptions(&to, (u_char *)(th + 1), 1179 (th->th_off << 2) - sizeof(struct tcphdr), 1180 (thflags & TH_SYN) ? TO_SYN : 0); 1181 1182 /* 1183 * If echoed timestamp is later than the current time, 1184 * fall back to non RFC1323 RTT calculation. Normalize 1185 * timestamp if syncookies were used when this connection 1186 * was established. 1187 */ 1188 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1189 to.to_tsecr -= tp->ts_offset; 1190 if (TSTMP_GT(to.to_tsecr, ticks)) 1191 to.to_tsecr = 0; 1192 } 1193 1194 /* 1195 * Process options only when we get SYN/ACK back. The SYN case 1196 * for incoming connections is handled in tcp_syncache. 1197 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 1198 * or <SYN,ACK>) segment itself is never scaled. 1199 * XXX this is traditional behavior, may need to be cleaned up. 1200 */ 1201 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1202 if ((to.to_flags & TOF_SCALE) && 1203 (tp->t_flags & TF_REQ_SCALE)) { 1204 tp->t_flags |= TF_RCVD_SCALE; 1205 tp->snd_scale = to.to_wscale; 1206 } 1207 /* 1208 * Initial send window. It will be updated with 1209 * the next incoming segment to the scaled value. 1210 */ 1211 tp->snd_wnd = th->th_win; 1212 if (to.to_flags & TOF_TS) { 1213 tp->t_flags |= TF_RCVD_TSTMP; 1214 tp->ts_recent = to.to_tsval; 1215 tp->ts_recent_age = ticks; 1216 } 1217 if (to.to_flags & TOF_MSS) 1218 tcp_mss(tp, to.to_mss); 1219 if ((tp->t_flags & TF_SACK_PERMIT) && 1220 (to.to_flags & TOF_SACKPERM) == 0) 1221 tp->t_flags &= ~TF_SACK_PERMIT; 1222 } 1223 1224 /* 1225 * Header prediction: check for the two common cases 1226 * of a uni-directional data xfer. If the packet has 1227 * no control flags, is in-sequence, the window didn't 1228 * change and we're not retransmitting, it's a 1229 * candidate. If the length is zero and the ack moved 1230 * forward, we're the sender side of the xfer. Just 1231 * free the data acked & wake any higher level process 1232 * that was blocked waiting for space. If the length 1233 * is non-zero and the ack didn't move, we're the 1234 * receiver side. If we're getting packets in-order 1235 * (the reassembly queue is empty), add the data to 1236 * the socket buffer and note that we need a delayed ack. 1237 * Make sure that the hidden state-flags are also off. 1238 * Since we check for TCPS_ESTABLISHED first, it can only 1239 * be TH_NEEDSYN. 1240 */ 1241 if (tp->t_state == TCPS_ESTABLISHED && 1242 th->th_seq == tp->rcv_nxt && 1243 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1244 tp->snd_nxt == tp->snd_max && 1245 tiwin && tiwin == tp->snd_wnd && 1246 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1247 LIST_EMPTY(&tp->t_segq) && 1248 ((to.to_flags & TOF_TS) == 0 || 1249 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1250 1251 /* 1252 * If last ACK falls within this segment's sequence numbers, 1253 * record the timestamp. 1254 * NOTE that the test is modified according to the latest 1255 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1256 */ 1257 if ((to.to_flags & TOF_TS) != 0 && 1258 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1259 tp->ts_recent_age = ticks; 1260 tp->ts_recent = to.to_tsval; 1261 } 1262 1263 if (tlen == 0) { 1264 if (SEQ_GT(th->th_ack, tp->snd_una) && 1265 SEQ_LEQ(th->th_ack, tp->snd_max) && 1266 tp->snd_cwnd >= tp->snd_wnd && 1267 ((!V_tcp_do_newreno && 1268 !(tp->t_flags & TF_SACK_PERMIT) && 1269 tp->t_dupacks < tcprexmtthresh) || 1270 ((V_tcp_do_newreno || 1271 (tp->t_flags & TF_SACK_PERMIT)) && 1272 !IN_FASTRECOVERY(tp) && 1273 (to.to_flags & TOF_SACK) == 0 && 1274 TAILQ_EMPTY(&tp->snd_holes)))) { 1275 /* 1276 * This is a pure ack for outstanding data. 1277 */ 1278 if (ti_locked == TI_RLOCKED) 1279 INP_INFO_RUNLOCK(&V_tcbinfo); 1280 else if (ti_locked == TI_WLOCKED) 1281 INP_INFO_WUNLOCK(&V_tcbinfo); 1282 else 1283 panic("%s: ti_locked %d on pure ACK", 1284 __func__, ti_locked); 1285 ti_locked = TI_UNLOCKED; 1286 1287 ++V_tcpstat.tcps_predack; 1288 1289 /* 1290 * "bad retransmit" recovery. 1291 */ 1292 if (tp->t_rxtshift == 1 && 1293 ticks < tp->t_badrxtwin) { 1294 ++V_tcpstat.tcps_sndrexmitbad; 1295 tp->snd_cwnd = tp->snd_cwnd_prev; 1296 tp->snd_ssthresh = 1297 tp->snd_ssthresh_prev; 1298 tp->snd_recover = tp->snd_recover_prev; 1299 if (tp->t_flags & TF_WASFRECOVERY) 1300 ENTER_FASTRECOVERY(tp); 1301 tp->snd_nxt = tp->snd_max; 1302 tp->t_badrxtwin = 0; 1303 } 1304 1305 /* 1306 * Recalculate the transmit timer / rtt. 1307 * 1308 * Some boxes send broken timestamp replies 1309 * during the SYN+ACK phase, ignore 1310 * timestamps of 0 or we could calculate a 1311 * huge RTT and blow up the retransmit timer. 1312 */ 1313 if ((to.to_flags & TOF_TS) != 0 && 1314 to.to_tsecr) { 1315 if (!tp->t_rttlow || 1316 tp->t_rttlow > ticks - to.to_tsecr) 1317 tp->t_rttlow = ticks - to.to_tsecr; 1318 tcp_xmit_timer(tp, 1319 ticks - to.to_tsecr + 1); 1320 } else if (tp->t_rtttime && 1321 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1322 if (!tp->t_rttlow || 1323 tp->t_rttlow > ticks - tp->t_rtttime) 1324 tp->t_rttlow = ticks - tp->t_rtttime; 1325 tcp_xmit_timer(tp, 1326 ticks - tp->t_rtttime); 1327 } 1328 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1329 acked = th->th_ack - tp->snd_una; 1330 V_tcpstat.tcps_rcvackpack++; 1331 V_tcpstat.tcps_rcvackbyte += acked; 1332 sbdrop(&so->so_snd, acked); 1333 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1334 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1335 tp->snd_recover = th->th_ack - 1; 1336 tp->snd_una = th->th_ack; 1337 /* 1338 * Pull snd_wl2 up to prevent seq wrap relative 1339 * to th_ack. 1340 */ 1341 tp->snd_wl2 = th->th_ack; 1342 tp->t_dupacks = 0; 1343 m_freem(m); 1344 ND6_HINT(tp); /* Some progress has been made. */ 1345 1346 /* 1347 * If all outstanding data are acked, stop 1348 * retransmit timer, otherwise restart timer 1349 * using current (possibly backed-off) value. 1350 * If process is waiting for space, 1351 * wakeup/selwakeup/signal. If data 1352 * are ready to send, let tcp_output 1353 * decide between more output or persist. 1354 */ 1355 #ifdef TCPDEBUG 1356 if (so->so_options & SO_DEBUG) 1357 tcp_trace(TA_INPUT, ostate, tp, 1358 (void *)tcp_saveipgen, 1359 &tcp_savetcp, 0); 1360 #endif 1361 if (tp->snd_una == tp->snd_max) 1362 tcp_timer_activate(tp, TT_REXMT, 0); 1363 else if (!tcp_timer_active(tp, TT_PERSIST)) 1364 tcp_timer_activate(tp, TT_REXMT, 1365 tp->t_rxtcur); 1366 sowwakeup(so); 1367 if (so->so_snd.sb_cc) 1368 (void) tcp_output(tp); 1369 goto check_delack; 1370 } 1371 } else if (th->th_ack == tp->snd_una && 1372 tlen <= sbspace(&so->so_rcv)) { 1373 int newsize = 0; /* automatic sockbuf scaling */ 1374 1375 /* 1376 * This is a pure, in-sequence data packet with 1377 * nothing on the reassembly queue and we have enough 1378 * buffer space to take it. 1379 */ 1380 if (ti_locked == TI_RLOCKED) 1381 INP_INFO_RUNLOCK(&V_tcbinfo); 1382 else if (ti_locked == TI_WLOCKED) 1383 INP_INFO_WUNLOCK(&V_tcbinfo); 1384 else 1385 panic("%s: ti_locked %d on pure data " 1386 "segment", __func__, ti_locked); 1387 ti_locked = TI_UNLOCKED; 1388 1389 /* Clean receiver SACK report if present */ 1390 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1391 tcp_clean_sackreport(tp); 1392 ++V_tcpstat.tcps_preddat; 1393 tp->rcv_nxt += tlen; 1394 /* 1395 * Pull snd_wl1 up to prevent seq wrap relative to 1396 * th_seq. 1397 */ 1398 tp->snd_wl1 = th->th_seq; 1399 /* 1400 * Pull rcv_up up to prevent seq wrap relative to 1401 * rcv_nxt. 1402 */ 1403 tp->rcv_up = tp->rcv_nxt; 1404 V_tcpstat.tcps_rcvpack++; 1405 V_tcpstat.tcps_rcvbyte += tlen; 1406 ND6_HINT(tp); /* Some progress has been made */ 1407 #ifdef TCPDEBUG 1408 if (so->so_options & SO_DEBUG) 1409 tcp_trace(TA_INPUT, ostate, tp, 1410 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1411 #endif 1412 /* 1413 * Automatic sizing of receive socket buffer. Often the send 1414 * buffer size is not optimally adjusted to the actual network 1415 * conditions at hand (delay bandwidth product). Setting the 1416 * buffer size too small limits throughput on links with high 1417 * bandwidth and high delay (eg. trans-continental/oceanic links). 1418 * 1419 * On the receive side the socket buffer memory is only rarely 1420 * used to any significant extent. This allows us to be much 1421 * more aggressive in scaling the receive socket buffer. For 1422 * the case that the buffer space is actually used to a large 1423 * extent and we run out of kernel memory we can simply drop 1424 * the new segments; TCP on the sender will just retransmit it 1425 * later. Setting the buffer size too big may only consume too 1426 * much kernel memory if the application doesn't read() from 1427 * the socket or packet loss or reordering makes use of the 1428 * reassembly queue. 1429 * 1430 * The criteria to step up the receive buffer one notch are: 1431 * 1. the number of bytes received during the time it takes 1432 * one timestamp to be reflected back to us (the RTT); 1433 * 2. received bytes per RTT is within seven eighth of the 1434 * current socket buffer size; 1435 * 3. receive buffer size has not hit maximal automatic size; 1436 * 1437 * This algorithm does one step per RTT at most and only if 1438 * we receive a bulk stream w/o packet losses or reorderings. 1439 * Shrinking the buffer during idle times is not necessary as 1440 * it doesn't consume any memory when idle. 1441 * 1442 * TODO: Only step up if the application is actually serving 1443 * the buffer to better manage the socket buffer resources. 1444 */ 1445 if (V_tcp_do_autorcvbuf && 1446 to.to_tsecr && 1447 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1448 if (to.to_tsecr > tp->rfbuf_ts && 1449 to.to_tsecr - tp->rfbuf_ts < hz) { 1450 if (tp->rfbuf_cnt > 1451 (so->so_rcv.sb_hiwat / 8 * 7) && 1452 so->so_rcv.sb_hiwat < 1453 V_tcp_autorcvbuf_max) { 1454 newsize = 1455 min(so->so_rcv.sb_hiwat + 1456 V_tcp_autorcvbuf_inc, 1457 V_tcp_autorcvbuf_max); 1458 } 1459 /* Start over with next RTT. */ 1460 tp->rfbuf_ts = 0; 1461 tp->rfbuf_cnt = 0; 1462 } else 1463 tp->rfbuf_cnt += tlen; /* add up */ 1464 } 1465 1466 /* Add data to socket buffer. */ 1467 SOCKBUF_LOCK(&so->so_rcv); 1468 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1469 m_freem(m); 1470 } else { 1471 /* 1472 * Set new socket buffer size. 1473 * Give up when limit is reached. 1474 */ 1475 if (newsize) 1476 if (!sbreserve_locked(&so->so_rcv, 1477 newsize, so, NULL)) 1478 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1479 m_adj(m, drop_hdrlen); /* delayed header drop */ 1480 sbappendstream_locked(&so->so_rcv, m); 1481 } 1482 /* NB: sorwakeup_locked() does an implicit unlock. */ 1483 sorwakeup_locked(so); 1484 if (DELAY_ACK(tp)) { 1485 tp->t_flags |= TF_DELACK; 1486 } else { 1487 tp->t_flags |= TF_ACKNOW; 1488 tcp_output(tp); 1489 } 1490 goto check_delack; 1491 } 1492 } 1493 1494 /* 1495 * Calculate amount of space in receive window, 1496 * and then do TCP input processing. 1497 * Receive window is amount of space in rcv queue, 1498 * but not less than advertised window. 1499 */ 1500 win = sbspace(&so->so_rcv); 1501 if (win < 0) 1502 win = 0; 1503 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1504 1505 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1506 tp->rfbuf_ts = 0; 1507 tp->rfbuf_cnt = 0; 1508 1509 switch (tp->t_state) { 1510 1511 /* 1512 * If the state is SYN_RECEIVED: 1513 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1514 */ 1515 case TCPS_SYN_RECEIVED: 1516 if ((thflags & TH_ACK) && 1517 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1518 SEQ_GT(th->th_ack, tp->snd_max))) { 1519 rstreason = BANDLIM_RST_OPENPORT; 1520 goto dropwithreset; 1521 } 1522 break; 1523 1524 /* 1525 * If the state is SYN_SENT: 1526 * if seg contains an ACK, but not for our SYN, drop the input. 1527 * if seg contains a RST, then drop the connection. 1528 * if seg does not contain SYN, then drop it. 1529 * Otherwise this is an acceptable SYN segment 1530 * initialize tp->rcv_nxt and tp->irs 1531 * if seg contains ack then advance tp->snd_una 1532 * if seg contains an ECE and ECN support is enabled, the stream 1533 * is ECN capable. 1534 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1535 * arrange for segment to be acked (eventually) 1536 * continue processing rest of data/controls, beginning with URG 1537 */ 1538 case TCPS_SYN_SENT: 1539 if ((thflags & TH_ACK) && 1540 (SEQ_LEQ(th->th_ack, tp->iss) || 1541 SEQ_GT(th->th_ack, tp->snd_max))) { 1542 rstreason = BANDLIM_UNLIMITED; 1543 goto dropwithreset; 1544 } 1545 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1546 tp = tcp_drop(tp, ECONNREFUSED); 1547 if (thflags & TH_RST) 1548 goto drop; 1549 if (!(thflags & TH_SYN)) 1550 goto drop; 1551 1552 tp->irs = th->th_seq; 1553 tcp_rcvseqinit(tp); 1554 if (thflags & TH_ACK) { 1555 V_tcpstat.tcps_connects++; 1556 soisconnected(so); 1557 #ifdef MAC 1558 SOCK_LOCK(so); 1559 mac_socketpeer_set_from_mbuf(m, so); 1560 SOCK_UNLOCK(so); 1561 #endif 1562 /* Do window scaling on this connection? */ 1563 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1564 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1565 tp->rcv_scale = tp->request_r_scale; 1566 } 1567 tp->rcv_adv += tp->rcv_wnd; 1568 tp->snd_una++; /* SYN is acked */ 1569 /* 1570 * If there's data, delay ACK; if there's also a FIN 1571 * ACKNOW will be turned on later. 1572 */ 1573 if (DELAY_ACK(tp) && tlen != 0) 1574 tcp_timer_activate(tp, TT_DELACK, 1575 tcp_delacktime); 1576 else 1577 tp->t_flags |= TF_ACKNOW; 1578 1579 if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1580 tp->t_flags |= TF_ECN_PERMIT; 1581 V_tcpstat.tcps_ecn_shs++; 1582 } 1583 1584 /* 1585 * Received <SYN,ACK> in SYN_SENT[*] state. 1586 * Transitions: 1587 * SYN_SENT --> ESTABLISHED 1588 * SYN_SENT* --> FIN_WAIT_1 1589 */ 1590 tp->t_starttime = ticks; 1591 if (tp->t_flags & TF_NEEDFIN) { 1592 tp->t_state = TCPS_FIN_WAIT_1; 1593 tp->t_flags &= ~TF_NEEDFIN; 1594 thflags &= ~TH_SYN; 1595 } else { 1596 tp->t_state = TCPS_ESTABLISHED; 1597 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1598 } 1599 } else { 1600 /* 1601 * Received initial SYN in SYN-SENT[*] state => 1602 * simultaneous open. If segment contains CC option 1603 * and there is a cached CC, apply TAO test. 1604 * If it succeeds, connection is * half-synchronized. 1605 * Otherwise, do 3-way handshake: 1606 * SYN-SENT -> SYN-RECEIVED 1607 * SYN-SENT* -> SYN-RECEIVED* 1608 * If there was no CC option, clear cached CC value. 1609 */ 1610 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1611 tcp_timer_activate(tp, TT_REXMT, 0); 1612 tp->t_state = TCPS_SYN_RECEIVED; 1613 } 1614 1615 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1616 "ti_locked %d", __func__, ti_locked)); 1617 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1618 INP_WLOCK_ASSERT(tp->t_inpcb); 1619 1620 /* 1621 * Advance th->th_seq to correspond to first data byte. 1622 * If data, trim to stay within window, 1623 * dropping FIN if necessary. 1624 */ 1625 th->th_seq++; 1626 if (tlen > tp->rcv_wnd) { 1627 todrop = tlen - tp->rcv_wnd; 1628 m_adj(m, -todrop); 1629 tlen = tp->rcv_wnd; 1630 thflags &= ~TH_FIN; 1631 V_tcpstat.tcps_rcvpackafterwin++; 1632 V_tcpstat.tcps_rcvbyteafterwin += todrop; 1633 } 1634 tp->snd_wl1 = th->th_seq - 1; 1635 tp->rcv_up = th->th_seq; 1636 /* 1637 * Client side of transaction: already sent SYN and data. 1638 * If the remote host used T/TCP to validate the SYN, 1639 * our data will be ACK'd; if so, enter normal data segment 1640 * processing in the middle of step 5, ack processing. 1641 * Otherwise, goto step 6. 1642 */ 1643 if (thflags & TH_ACK) 1644 goto process_ACK; 1645 1646 goto step6; 1647 1648 /* 1649 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1650 * do normal processing. 1651 * 1652 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1653 */ 1654 case TCPS_LAST_ACK: 1655 case TCPS_CLOSING: 1656 break; /* continue normal processing */ 1657 } 1658 1659 /* 1660 * States other than LISTEN or SYN_SENT. 1661 * First check the RST flag and sequence number since reset segments 1662 * are exempt from the timestamp and connection count tests. This 1663 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1664 * below which allowed reset segments in half the sequence space 1665 * to fall though and be processed (which gives forged reset 1666 * segments with a random sequence number a 50 percent chance of 1667 * killing a connection). 1668 * Then check timestamp, if present. 1669 * Then check the connection count, if present. 1670 * Then check that at least some bytes of segment are within 1671 * receive window. If segment begins before rcv_nxt, 1672 * drop leading data (and SYN); if nothing left, just ack. 1673 * 1674 * 1675 * If the RST bit is set, check the sequence number to see 1676 * if this is a valid reset segment. 1677 * RFC 793 page 37: 1678 * In all states except SYN-SENT, all reset (RST) segments 1679 * are validated by checking their SEQ-fields. A reset is 1680 * valid if its sequence number is in the window. 1681 * Note: this does not take into account delayed ACKs, so 1682 * we should test against last_ack_sent instead of rcv_nxt. 1683 * The sequence number in the reset segment is normally an 1684 * echo of our outgoing acknowlegement numbers, but some hosts 1685 * send a reset with the sequence number at the rightmost edge 1686 * of our receive window, and we have to handle this case. 1687 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1688 * that brute force RST attacks are possible. To combat this, 1689 * we use a much stricter check while in the ESTABLISHED state, 1690 * only accepting RSTs where the sequence number is equal to 1691 * last_ack_sent. In all other states (the states in which a 1692 * RST is more likely), the more permissive check is used. 1693 * If we have multiple segments in flight, the initial reset 1694 * segment sequence numbers will be to the left of last_ack_sent, 1695 * but they will eventually catch up. 1696 * In any case, it never made sense to trim reset segments to 1697 * fit the receive window since RFC 1122 says: 1698 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1699 * 1700 * A TCP SHOULD allow a received RST segment to include data. 1701 * 1702 * DISCUSSION 1703 * It has been suggested that a RST segment could contain 1704 * ASCII text that encoded and explained the cause of the 1705 * RST. No standard has yet been established for such 1706 * data. 1707 * 1708 * If the reset segment passes the sequence number test examine 1709 * the state: 1710 * SYN_RECEIVED STATE: 1711 * If passive open, return to LISTEN state. 1712 * If active open, inform user that connection was refused. 1713 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1714 * Inform user that connection was reset, and close tcb. 1715 * CLOSING, LAST_ACK STATES: 1716 * Close the tcb. 1717 * TIME_WAIT STATE: 1718 * Drop the segment - see Stevens, vol. 2, p. 964 and 1719 * RFC 1337. 1720 */ 1721 if (thflags & TH_RST) { 1722 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1723 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1724 switch (tp->t_state) { 1725 1726 case TCPS_SYN_RECEIVED: 1727 so->so_error = ECONNREFUSED; 1728 goto close; 1729 1730 case TCPS_ESTABLISHED: 1731 if (V_tcp_insecure_rst == 0 && 1732 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 1733 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 1734 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1735 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 1736 V_tcpstat.tcps_badrst++; 1737 goto drop; 1738 } 1739 /* FALLTHROUGH */ 1740 case TCPS_FIN_WAIT_1: 1741 case TCPS_FIN_WAIT_2: 1742 case TCPS_CLOSE_WAIT: 1743 so->so_error = ECONNRESET; 1744 close: 1745 KASSERT(ti_locked == TI_WLOCKED, 1746 ("tcp_do_segment: TH_RST 1 ti_locked %d", 1747 ti_locked)); 1748 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1749 1750 tp->t_state = TCPS_CLOSED; 1751 V_tcpstat.tcps_drops++; 1752 tp = tcp_close(tp); 1753 break; 1754 1755 case TCPS_CLOSING: 1756 case TCPS_LAST_ACK: 1757 KASSERT(ti_locked == TI_WLOCKED, 1758 ("tcp_do_segment: TH_RST 2 ti_locked %d", 1759 ti_locked)); 1760 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1761 1762 tp = tcp_close(tp); 1763 break; 1764 } 1765 } 1766 goto drop; 1767 } 1768 1769 /* 1770 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1771 * and it's less than ts_recent, drop it. 1772 */ 1773 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1774 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1775 1776 /* Check to see if ts_recent is over 24 days old. */ 1777 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1778 /* 1779 * Invalidate ts_recent. If this segment updates 1780 * ts_recent, the age will be reset later and ts_recent 1781 * will get a valid value. If it does not, setting 1782 * ts_recent to zero will at least satisfy the 1783 * requirement that zero be placed in the timestamp 1784 * echo reply when ts_recent isn't valid. The 1785 * age isn't reset until we get a valid ts_recent 1786 * because we don't want out-of-order segments to be 1787 * dropped when ts_recent is old. 1788 */ 1789 tp->ts_recent = 0; 1790 } else { 1791 V_tcpstat.tcps_rcvduppack++; 1792 V_tcpstat.tcps_rcvdupbyte += tlen; 1793 V_tcpstat.tcps_pawsdrop++; 1794 if (tlen) 1795 goto dropafterack; 1796 goto drop; 1797 } 1798 } 1799 1800 /* 1801 * In the SYN-RECEIVED state, validate that the packet belongs to 1802 * this connection before trimming the data to fit the receive 1803 * window. Check the sequence number versus IRS since we know 1804 * the sequence numbers haven't wrapped. This is a partial fix 1805 * for the "LAND" DoS attack. 1806 */ 1807 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1808 rstreason = BANDLIM_RST_OPENPORT; 1809 goto dropwithreset; 1810 } 1811 1812 todrop = tp->rcv_nxt - th->th_seq; 1813 if (todrop > 0) { 1814 if (thflags & TH_SYN) { 1815 thflags &= ~TH_SYN; 1816 th->th_seq++; 1817 if (th->th_urp > 1) 1818 th->th_urp--; 1819 else 1820 thflags &= ~TH_URG; 1821 todrop--; 1822 } 1823 /* 1824 * Following if statement from Stevens, vol. 2, p. 960. 1825 */ 1826 if (todrop > tlen 1827 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1828 /* 1829 * Any valid FIN must be to the left of the window. 1830 * At this point the FIN must be a duplicate or out 1831 * of sequence; drop it. 1832 */ 1833 thflags &= ~TH_FIN; 1834 1835 /* 1836 * Send an ACK to resynchronize and drop any data. 1837 * But keep on processing for RST or ACK. 1838 */ 1839 tp->t_flags |= TF_ACKNOW; 1840 todrop = tlen; 1841 V_tcpstat.tcps_rcvduppack++; 1842 V_tcpstat.tcps_rcvdupbyte += todrop; 1843 } else { 1844 V_tcpstat.tcps_rcvpartduppack++; 1845 V_tcpstat.tcps_rcvpartdupbyte += todrop; 1846 } 1847 drop_hdrlen += todrop; /* drop from the top afterwards */ 1848 th->th_seq += todrop; 1849 tlen -= todrop; 1850 if (th->th_urp > todrop) 1851 th->th_urp -= todrop; 1852 else { 1853 thflags &= ~TH_URG; 1854 th->th_urp = 0; 1855 } 1856 } 1857 1858 /* 1859 * If new data are received on a connection after the 1860 * user processes are gone, then RST the other end. 1861 */ 1862 if ((so->so_state & SS_NOFDREF) && 1863 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1864 char *s; 1865 1866 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 1867 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 1868 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1869 1870 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 1871 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket " 1872 "was closed, sending RST and removing tcpcb\n", 1873 s, __func__, tcpstates[tp->t_state], tlen); 1874 free(s, M_TCPLOG); 1875 } 1876 tp = tcp_close(tp); 1877 V_tcpstat.tcps_rcvafterclose++; 1878 rstreason = BANDLIM_UNLIMITED; 1879 goto dropwithreset; 1880 } 1881 1882 /* 1883 * If segment ends after window, drop trailing data 1884 * (and PUSH and FIN); if nothing left, just ACK. 1885 */ 1886 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1887 if (todrop > 0) { 1888 V_tcpstat.tcps_rcvpackafterwin++; 1889 if (todrop >= tlen) { 1890 V_tcpstat.tcps_rcvbyteafterwin += tlen; 1891 /* 1892 * If window is closed can only take segments at 1893 * window edge, and have to drop data and PUSH from 1894 * incoming segments. Continue processing, but 1895 * remember to ack. Otherwise, drop segment 1896 * and ack. 1897 */ 1898 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1899 tp->t_flags |= TF_ACKNOW; 1900 V_tcpstat.tcps_rcvwinprobe++; 1901 } else 1902 goto dropafterack; 1903 } else 1904 V_tcpstat.tcps_rcvbyteafterwin += todrop; 1905 m_adj(m, -todrop); 1906 tlen -= todrop; 1907 thflags &= ~(TH_PUSH|TH_FIN); 1908 } 1909 1910 /* 1911 * If last ACK falls within this segment's sequence numbers, 1912 * record its timestamp. 1913 * NOTE: 1914 * 1) That the test incorporates suggestions from the latest 1915 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1916 * 2) That updating only on newer timestamps interferes with 1917 * our earlier PAWS tests, so this check should be solely 1918 * predicated on the sequence space of this segment. 1919 * 3) That we modify the segment boundary check to be 1920 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1921 * instead of RFC1323's 1922 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1923 * This modified check allows us to overcome RFC1323's 1924 * limitations as described in Stevens TCP/IP Illustrated 1925 * Vol. 2 p.869. In such cases, we can still calculate the 1926 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1927 */ 1928 if ((to.to_flags & TOF_TS) != 0 && 1929 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1930 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1931 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1932 tp->ts_recent_age = ticks; 1933 tp->ts_recent = to.to_tsval; 1934 } 1935 1936 /* 1937 * If a SYN is in the window, then this is an 1938 * error and we send an RST and drop the connection. 1939 */ 1940 if (thflags & TH_SYN) { 1941 KASSERT(ti_locked == TI_WLOCKED, 1942 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 1943 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1944 1945 tp = tcp_drop(tp, ECONNRESET); 1946 rstreason = BANDLIM_UNLIMITED; 1947 goto drop; 1948 } 1949 1950 /* 1951 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1952 * flag is on (half-synchronized state), then queue data for 1953 * later processing; else drop segment and return. 1954 */ 1955 if ((thflags & TH_ACK) == 0) { 1956 if (tp->t_state == TCPS_SYN_RECEIVED || 1957 (tp->t_flags & TF_NEEDSYN)) 1958 goto step6; 1959 else if (tp->t_flags & TF_ACKNOW) 1960 goto dropafterack; 1961 else 1962 goto drop; 1963 } 1964 1965 /* 1966 * Ack processing. 1967 */ 1968 switch (tp->t_state) { 1969 1970 /* 1971 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1972 * ESTABLISHED state and continue processing. 1973 * The ACK was checked above. 1974 */ 1975 case TCPS_SYN_RECEIVED: 1976 1977 V_tcpstat.tcps_connects++; 1978 soisconnected(so); 1979 /* Do window scaling? */ 1980 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1981 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1982 tp->rcv_scale = tp->request_r_scale; 1983 tp->snd_wnd = tiwin; 1984 } 1985 /* 1986 * Make transitions: 1987 * SYN-RECEIVED -> ESTABLISHED 1988 * SYN-RECEIVED* -> FIN-WAIT-1 1989 */ 1990 tp->t_starttime = ticks; 1991 if (tp->t_flags & TF_NEEDFIN) { 1992 tp->t_state = TCPS_FIN_WAIT_1; 1993 tp->t_flags &= ~TF_NEEDFIN; 1994 } else { 1995 tp->t_state = TCPS_ESTABLISHED; 1996 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1997 } 1998 /* 1999 * If segment contains data or ACK, will call tcp_reass() 2000 * later; if not, do so now to pass queued data to user. 2001 */ 2002 if (tlen == 0 && (thflags & TH_FIN) == 0) 2003 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2004 (struct mbuf *)0); 2005 tp->snd_wl1 = th->th_seq - 1; 2006 /* FALLTHROUGH */ 2007 2008 /* 2009 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2010 * ACKs. If the ack is in the range 2011 * tp->snd_una < th->th_ack <= tp->snd_max 2012 * then advance tp->snd_una to th->th_ack and drop 2013 * data from the retransmission queue. If this ACK reflects 2014 * more up to date window information we update our window information. 2015 */ 2016 case TCPS_ESTABLISHED: 2017 case TCPS_FIN_WAIT_1: 2018 case TCPS_FIN_WAIT_2: 2019 case TCPS_CLOSE_WAIT: 2020 case TCPS_CLOSING: 2021 case TCPS_LAST_ACK: 2022 if (SEQ_GT(th->th_ack, tp->snd_max)) { 2023 V_tcpstat.tcps_rcvacktoomuch++; 2024 goto dropafterack; 2025 } 2026 if ((tp->t_flags & TF_SACK_PERMIT) && 2027 ((to.to_flags & TOF_SACK) || 2028 !TAILQ_EMPTY(&tp->snd_holes))) 2029 tcp_sack_doack(tp, &to, th->th_ack); 2030 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 2031 if (tlen == 0 && tiwin == tp->snd_wnd) { 2032 V_tcpstat.tcps_rcvdupack++; 2033 /* 2034 * If we have outstanding data (other than 2035 * a window probe), this is a completely 2036 * duplicate ack (ie, window info didn't 2037 * change), the ack is the biggest we've 2038 * seen and we've seen exactly our rexmt 2039 * threshhold of them, assume a packet 2040 * has been dropped and retransmit it. 2041 * Kludge snd_nxt & the congestion 2042 * window so we send only this one 2043 * packet. 2044 * 2045 * We know we're losing at the current 2046 * window size so do congestion avoidance 2047 * (set ssthresh to half the current window 2048 * and pull our congestion window back to 2049 * the new ssthresh). 2050 * 2051 * Dup acks mean that packets have left the 2052 * network (they're now cached at the receiver) 2053 * so bump cwnd by the amount in the receiver 2054 * to keep a constant cwnd packets in the 2055 * network. 2056 * 2057 * When using TCP ECN, notify the peer that 2058 * we reduced the cwnd. 2059 */ 2060 if (!tcp_timer_active(tp, TT_REXMT) || 2061 th->th_ack != tp->snd_una) 2062 tp->t_dupacks = 0; 2063 else if (++tp->t_dupacks > tcprexmtthresh || 2064 ((V_tcp_do_newreno || 2065 (tp->t_flags & TF_SACK_PERMIT)) && 2066 IN_FASTRECOVERY(tp))) { 2067 if ((tp->t_flags & TF_SACK_PERMIT) && 2068 IN_FASTRECOVERY(tp)) { 2069 int awnd; 2070 2071 /* 2072 * Compute the amount of data in flight first. 2073 * We can inject new data into the pipe iff 2074 * we have less than 1/2 the original window's 2075 * worth of data in flight. 2076 */ 2077 awnd = (tp->snd_nxt - tp->snd_fack) + 2078 tp->sackhint.sack_bytes_rexmit; 2079 if (awnd < tp->snd_ssthresh) { 2080 tp->snd_cwnd += tp->t_maxseg; 2081 if (tp->snd_cwnd > tp->snd_ssthresh) 2082 tp->snd_cwnd = tp->snd_ssthresh; 2083 } 2084 } else 2085 tp->snd_cwnd += tp->t_maxseg; 2086 (void) tcp_output(tp); 2087 goto drop; 2088 } else if (tp->t_dupacks == tcprexmtthresh) { 2089 tcp_seq onxt = tp->snd_nxt; 2090 2091 /* 2092 * If we're doing sack, check to 2093 * see if we're already in sack 2094 * recovery. If we're not doing sack, 2095 * check to see if we're in newreno 2096 * recovery. 2097 */ 2098 if (tp->t_flags & TF_SACK_PERMIT) { 2099 if (IN_FASTRECOVERY(tp)) { 2100 tp->t_dupacks = 0; 2101 break; 2102 } 2103 } else if (V_tcp_do_newreno || 2104 V_tcp_do_ecn) { 2105 if (SEQ_LEQ(th->th_ack, 2106 tp->snd_recover)) { 2107 tp->t_dupacks = 0; 2108 break; 2109 } 2110 } 2111 tcp_congestion_exp(tp); 2112 tcp_timer_activate(tp, TT_REXMT, 0); 2113 tp->t_rtttime = 0; 2114 if (tp->t_flags & TF_SACK_PERMIT) { 2115 V_tcpstat.tcps_sack_recovery_episode++; 2116 tp->sack_newdata = tp->snd_nxt; 2117 tp->snd_cwnd = tp->t_maxseg; 2118 (void) tcp_output(tp); 2119 goto drop; 2120 } 2121 tp->snd_nxt = th->th_ack; 2122 tp->snd_cwnd = tp->t_maxseg; 2123 (void) tcp_output(tp); 2124 KASSERT(tp->snd_limited <= 2, 2125 ("%s: tp->snd_limited too big", 2126 __func__)); 2127 tp->snd_cwnd = tp->snd_ssthresh + 2128 tp->t_maxseg * 2129 (tp->t_dupacks - tp->snd_limited); 2130 if (SEQ_GT(onxt, tp->snd_nxt)) 2131 tp->snd_nxt = onxt; 2132 goto drop; 2133 } else if (V_tcp_do_rfc3042) { 2134 u_long oldcwnd = tp->snd_cwnd; 2135 tcp_seq oldsndmax = tp->snd_max; 2136 u_int sent; 2137 2138 KASSERT(tp->t_dupacks == 1 || 2139 tp->t_dupacks == 2, 2140 ("%s: dupacks not 1 or 2", 2141 __func__)); 2142 if (tp->t_dupacks == 1) 2143 tp->snd_limited = 0; 2144 tp->snd_cwnd = 2145 (tp->snd_nxt - tp->snd_una) + 2146 (tp->t_dupacks - tp->snd_limited) * 2147 tp->t_maxseg; 2148 (void) tcp_output(tp); 2149 sent = tp->snd_max - oldsndmax; 2150 if (sent > tp->t_maxseg) { 2151 KASSERT((tp->t_dupacks == 2 && 2152 tp->snd_limited == 0) || 2153 (sent == tp->t_maxseg + 1 && 2154 tp->t_flags & TF_SENTFIN), 2155 ("%s: sent too much", 2156 __func__)); 2157 tp->snd_limited = 2; 2158 } else if (sent > 0) 2159 ++tp->snd_limited; 2160 tp->snd_cwnd = oldcwnd; 2161 goto drop; 2162 } 2163 } else 2164 tp->t_dupacks = 0; 2165 break; 2166 } 2167 2168 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2169 ("%s: th_ack <= snd_una", __func__)); 2170 2171 /* 2172 * If the congestion window was inflated to account 2173 * for the other side's cached packets, retract it. 2174 */ 2175 if (V_tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) { 2176 if (IN_FASTRECOVERY(tp)) { 2177 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 2178 if (tp->t_flags & TF_SACK_PERMIT) 2179 tcp_sack_partialack(tp, th); 2180 else 2181 tcp_newreno_partial_ack(tp, th); 2182 } else { 2183 /* 2184 * Out of fast recovery. 2185 * Window inflation should have left us 2186 * with approximately snd_ssthresh 2187 * outstanding data. 2188 * But in case we would be inclined to 2189 * send a burst, better to do it via 2190 * the slow start mechanism. 2191 */ 2192 if (SEQ_GT(th->th_ack + 2193 tp->snd_ssthresh, 2194 tp->snd_max)) 2195 tp->snd_cwnd = tp->snd_max - 2196 th->th_ack + 2197 tp->t_maxseg; 2198 else 2199 tp->snd_cwnd = tp->snd_ssthresh; 2200 } 2201 } 2202 } else { 2203 if (tp->t_dupacks >= tcprexmtthresh && 2204 tp->snd_cwnd > tp->snd_ssthresh) 2205 tp->snd_cwnd = tp->snd_ssthresh; 2206 } 2207 tp->t_dupacks = 0; 2208 /* 2209 * If we reach this point, ACK is not a duplicate, 2210 * i.e., it ACKs something we sent. 2211 */ 2212 if (tp->t_flags & TF_NEEDSYN) { 2213 /* 2214 * T/TCP: Connection was half-synchronized, and our 2215 * SYN has been ACK'd (so connection is now fully 2216 * synchronized). Go to non-starred state, 2217 * increment snd_una for ACK of SYN, and check if 2218 * we can do window scaling. 2219 */ 2220 tp->t_flags &= ~TF_NEEDSYN; 2221 tp->snd_una++; 2222 /* Do window scaling? */ 2223 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2224 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2225 tp->rcv_scale = tp->request_r_scale; 2226 /* Send window already scaled. */ 2227 } 2228 } 2229 2230 process_ACK: 2231 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2232 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2233 ("tcp_input: process_ACK ti_locked %d", ti_locked)); 2234 INP_WLOCK_ASSERT(tp->t_inpcb); 2235 2236 acked = th->th_ack - tp->snd_una; 2237 V_tcpstat.tcps_rcvackpack++; 2238 V_tcpstat.tcps_rcvackbyte += acked; 2239 2240 /* 2241 * If we just performed our first retransmit, and the ACK 2242 * arrives within our recovery window, then it was a mistake 2243 * to do the retransmit in the first place. Recover our 2244 * original cwnd and ssthresh, and proceed to transmit where 2245 * we left off. 2246 */ 2247 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 2248 ++V_tcpstat.tcps_sndrexmitbad; 2249 tp->snd_cwnd = tp->snd_cwnd_prev; 2250 tp->snd_ssthresh = tp->snd_ssthresh_prev; 2251 tp->snd_recover = tp->snd_recover_prev; 2252 if (tp->t_flags & TF_WASFRECOVERY) 2253 ENTER_FASTRECOVERY(tp); 2254 tp->snd_nxt = tp->snd_max; 2255 tp->t_badrxtwin = 0; /* XXX probably not required */ 2256 } 2257 2258 /* 2259 * If we have a timestamp reply, update smoothed 2260 * round trip time. If no timestamp is present but 2261 * transmit timer is running and timed sequence 2262 * number was acked, update smoothed round trip time. 2263 * Since we now have an rtt measurement, cancel the 2264 * timer backoff (cf., Phil Karn's retransmit alg.). 2265 * Recompute the initial retransmit timer. 2266 * 2267 * Some boxes send broken timestamp replies 2268 * during the SYN+ACK phase, ignore 2269 * timestamps of 0 or we could calculate a 2270 * huge RTT and blow up the retransmit timer. 2271 */ 2272 if ((to.to_flags & TOF_TS) != 0 && 2273 to.to_tsecr) { 2274 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 2275 tp->t_rttlow = ticks - to.to_tsecr; 2276 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 2277 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2278 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2279 tp->t_rttlow = ticks - tp->t_rtttime; 2280 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2281 } 2282 tcp_xmit_bandwidth_limit(tp, th->th_ack); 2283 2284 /* 2285 * If all outstanding data is acked, stop retransmit 2286 * timer and remember to restart (more output or persist). 2287 * If there is more data to be acked, restart retransmit 2288 * timer, using current (possibly backed-off) value. 2289 */ 2290 if (th->th_ack == tp->snd_max) { 2291 tcp_timer_activate(tp, TT_REXMT, 0); 2292 needoutput = 1; 2293 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2294 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2295 2296 /* 2297 * If no data (only SYN) was ACK'd, 2298 * skip rest of ACK processing. 2299 */ 2300 if (acked == 0) 2301 goto step6; 2302 2303 /* 2304 * When new data is acked, open the congestion window. 2305 * Method depends on which congestion control state we're 2306 * in (slow start or cong avoid) and if ABC (RFC 3465) is 2307 * enabled. 2308 * 2309 * slow start: cwnd <= ssthresh 2310 * cong avoid: cwnd > ssthresh 2311 * 2312 * slow start and ABC (RFC 3465): 2313 * Grow cwnd exponentially by the amount of data 2314 * ACKed capping the max increment per ACK to 2315 * (abc_l_var * maxseg) bytes. 2316 * 2317 * slow start without ABC (RFC 2581): 2318 * Grow cwnd exponentially by maxseg per ACK. 2319 * 2320 * cong avoid and ABC (RFC 3465): 2321 * Grow cwnd linearly by maxseg per RTT for each 2322 * cwnd worth of ACKed data. 2323 * 2324 * cong avoid without ABC (RFC 2581): 2325 * Grow cwnd linearly by approximately maxseg per RTT using 2326 * maxseg^2 / cwnd per ACK as the increment. 2327 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to 2328 * avoid capping cwnd. 2329 */ 2330 if ((!V_tcp_do_newreno && !(tp->t_flags & TF_SACK_PERMIT)) || 2331 !IN_FASTRECOVERY(tp)) { 2332 u_int cw = tp->snd_cwnd; 2333 u_int incr = tp->t_maxseg; 2334 /* In congestion avoidance? */ 2335 if (cw > tp->snd_ssthresh) { 2336 if (V_tcp_do_rfc3465) { 2337 tp->t_bytes_acked += acked; 2338 if (tp->t_bytes_acked >= tp->snd_cwnd) 2339 tp->t_bytes_acked -= cw; 2340 else 2341 incr = 0; 2342 } 2343 else 2344 incr = max((incr * incr / cw), 1); 2345 /* 2346 * In slow-start with ABC enabled and no RTO in sight? 2347 * (Must not use abc_l_var > 1 if slow starting after an 2348 * RTO. On RTO, snd_nxt = snd_una, so the snd_nxt == 2349 * snd_max check is sufficient to handle this). 2350 */ 2351 } else if (V_tcp_do_rfc3465 && 2352 tp->snd_nxt == tp->snd_max) 2353 incr = min(acked, 2354 V_tcp_abc_l_var * tp->t_maxseg); 2355 /* ABC is on by default, so (incr == 0) frequently. */ 2356 if (incr > 0) 2357 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2358 } 2359 SOCKBUF_LOCK(&so->so_snd); 2360 if (acked > so->so_snd.sb_cc) { 2361 tp->snd_wnd -= so->so_snd.sb_cc; 2362 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2363 ourfinisacked = 1; 2364 } else { 2365 sbdrop_locked(&so->so_snd, acked); 2366 tp->snd_wnd -= acked; 2367 ourfinisacked = 0; 2368 } 2369 /* NB: sowwakeup_locked() does an implicit unlock. */ 2370 sowwakeup_locked(so); 2371 /* Detect una wraparound. */ 2372 if ((V_tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2373 !IN_FASTRECOVERY(tp) && 2374 SEQ_GT(tp->snd_una, tp->snd_recover) && 2375 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2376 tp->snd_recover = th->th_ack - 1; 2377 if ((V_tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2378 IN_FASTRECOVERY(tp) && 2379 SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2380 EXIT_FASTRECOVERY(tp); 2381 tp->t_bytes_acked = 0; 2382 } 2383 tp->snd_una = th->th_ack; 2384 if (tp->t_flags & TF_SACK_PERMIT) { 2385 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2386 tp->snd_recover = tp->snd_una; 2387 } 2388 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2389 tp->snd_nxt = tp->snd_una; 2390 2391 switch (tp->t_state) { 2392 2393 /* 2394 * In FIN_WAIT_1 STATE in addition to the processing 2395 * for the ESTABLISHED state if our FIN is now acknowledged 2396 * then enter FIN_WAIT_2. 2397 */ 2398 case TCPS_FIN_WAIT_1: 2399 if (ourfinisacked) { 2400 /* 2401 * If we can't receive any more 2402 * data, then closing user can proceed. 2403 * Starting the timer is contrary to the 2404 * specification, but if we don't get a FIN 2405 * we'll hang forever. 2406 * 2407 * XXXjl: 2408 * we should release the tp also, and use a 2409 * compressed state. 2410 */ 2411 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2412 int timeout; 2413 2414 soisdisconnected(so); 2415 timeout = (tcp_fast_finwait2_recycle) ? 2416 tcp_finwait2_timeout : tcp_maxidle; 2417 tcp_timer_activate(tp, TT_2MSL, timeout); 2418 } 2419 tp->t_state = TCPS_FIN_WAIT_2; 2420 } 2421 break; 2422 2423 /* 2424 * In CLOSING STATE in addition to the processing for 2425 * the ESTABLISHED state if the ACK acknowledges our FIN 2426 * then enter the TIME-WAIT state, otherwise ignore 2427 * the segment. 2428 */ 2429 case TCPS_CLOSING: 2430 if (ourfinisacked) { 2431 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2432 tcp_twstart(tp); 2433 INP_INFO_WUNLOCK(&V_tcbinfo); 2434 m_freem(m); 2435 return; 2436 } 2437 break; 2438 2439 /* 2440 * In LAST_ACK, we may still be waiting for data to drain 2441 * and/or to be acked, as well as for the ack of our FIN. 2442 * If our FIN is now acknowledged, delete the TCB, 2443 * enter the closed state and return. 2444 */ 2445 case TCPS_LAST_ACK: 2446 if (ourfinisacked) { 2447 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2448 tp = tcp_close(tp); 2449 goto drop; 2450 } 2451 break; 2452 } 2453 } 2454 2455 step6: 2456 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2457 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2458 ("tcp_do_segment: step6 ti_locked %d", ti_locked)); 2459 INP_WLOCK_ASSERT(tp->t_inpcb); 2460 2461 /* 2462 * Update window information. 2463 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2464 */ 2465 if ((thflags & TH_ACK) && 2466 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2467 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2468 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2469 /* keep track of pure window updates */ 2470 if (tlen == 0 && 2471 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2472 V_tcpstat.tcps_rcvwinupd++; 2473 tp->snd_wnd = tiwin; 2474 tp->snd_wl1 = th->th_seq; 2475 tp->snd_wl2 = th->th_ack; 2476 if (tp->snd_wnd > tp->max_sndwnd) 2477 tp->max_sndwnd = tp->snd_wnd; 2478 needoutput = 1; 2479 } 2480 2481 /* 2482 * Process segments with URG. 2483 */ 2484 if ((thflags & TH_URG) && th->th_urp && 2485 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2486 /* 2487 * This is a kludge, but if we receive and accept 2488 * random urgent pointers, we'll crash in 2489 * soreceive. It's hard to imagine someone 2490 * actually wanting to send this much urgent data. 2491 */ 2492 SOCKBUF_LOCK(&so->so_rcv); 2493 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2494 th->th_urp = 0; /* XXX */ 2495 thflags &= ~TH_URG; /* XXX */ 2496 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2497 goto dodata; /* XXX */ 2498 } 2499 /* 2500 * If this segment advances the known urgent pointer, 2501 * then mark the data stream. This should not happen 2502 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2503 * a FIN has been received from the remote side. 2504 * In these states we ignore the URG. 2505 * 2506 * According to RFC961 (Assigned Protocols), 2507 * the urgent pointer points to the last octet 2508 * of urgent data. We continue, however, 2509 * to consider it to indicate the first octet 2510 * of data past the urgent section as the original 2511 * spec states (in one of two places). 2512 */ 2513 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2514 tp->rcv_up = th->th_seq + th->th_urp; 2515 so->so_oobmark = so->so_rcv.sb_cc + 2516 (tp->rcv_up - tp->rcv_nxt) - 1; 2517 if (so->so_oobmark == 0) 2518 so->so_rcv.sb_state |= SBS_RCVATMARK; 2519 sohasoutofband(so); 2520 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2521 } 2522 SOCKBUF_UNLOCK(&so->so_rcv); 2523 /* 2524 * Remove out of band data so doesn't get presented to user. 2525 * This can happen independent of advancing the URG pointer, 2526 * but if two URG's are pending at once, some out-of-band 2527 * data may creep in... ick. 2528 */ 2529 if (th->th_urp <= (u_long)tlen && 2530 !(so->so_options & SO_OOBINLINE)) { 2531 /* hdr drop is delayed */ 2532 tcp_pulloutofband(so, th, m, drop_hdrlen); 2533 } 2534 } else { 2535 /* 2536 * If no out of band data is expected, 2537 * pull receive urgent pointer along 2538 * with the receive window. 2539 */ 2540 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2541 tp->rcv_up = tp->rcv_nxt; 2542 } 2543 dodata: /* XXX */ 2544 INP_INFO_LOCK_ASSERT(&V_tcbinfo); 2545 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2546 ("tcp_do_segment: dodata ti_locked %d", ti_locked)); 2547 INP_WLOCK_ASSERT(tp->t_inpcb); 2548 2549 /* 2550 * Process the segment text, merging it into the TCP sequencing queue, 2551 * and arranging for acknowledgment of receipt if necessary. 2552 * This process logically involves adjusting tp->rcv_wnd as data 2553 * is presented to the user (this happens in tcp_usrreq.c, 2554 * case PRU_RCVD). If a FIN has already been received on this 2555 * connection then we just ignore the text. 2556 */ 2557 if ((tlen || (thflags & TH_FIN)) && 2558 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2559 tcp_seq save_start = th->th_seq; 2560 m_adj(m, drop_hdrlen); /* delayed header drop */ 2561 /* 2562 * Insert segment which includes th into TCP reassembly queue 2563 * with control block tp. Set thflags to whether reassembly now 2564 * includes a segment with FIN. This handles the common case 2565 * inline (segment is the next to be received on an established 2566 * connection, and the queue is empty), avoiding linkage into 2567 * and removal from the queue and repetition of various 2568 * conversions. 2569 * Set DELACK for segments received in order, but ack 2570 * immediately when segments are out of order (so 2571 * fast retransmit can work). 2572 */ 2573 if (th->th_seq == tp->rcv_nxt && 2574 LIST_EMPTY(&tp->t_segq) && 2575 TCPS_HAVEESTABLISHED(tp->t_state)) { 2576 if (DELAY_ACK(tp)) 2577 tp->t_flags |= TF_DELACK; 2578 else 2579 tp->t_flags |= TF_ACKNOW; 2580 tp->rcv_nxt += tlen; 2581 thflags = th->th_flags & TH_FIN; 2582 V_tcpstat.tcps_rcvpack++; 2583 V_tcpstat.tcps_rcvbyte += tlen; 2584 ND6_HINT(tp); 2585 SOCKBUF_LOCK(&so->so_rcv); 2586 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2587 m_freem(m); 2588 else 2589 sbappendstream_locked(&so->so_rcv, m); 2590 /* NB: sorwakeup_locked() does an implicit unlock. */ 2591 sorwakeup_locked(so); 2592 } else { 2593 /* 2594 * XXX: Due to the header drop above "th" is 2595 * theoretically invalid by now. Fortunately 2596 * m_adj() doesn't actually frees any mbufs 2597 * when trimming from the head. 2598 */ 2599 thflags = tcp_reass(tp, th, &tlen, m); 2600 tp->t_flags |= TF_ACKNOW; 2601 } 2602 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2603 tcp_update_sack_list(tp, save_start, save_start + tlen); 2604 #if 0 2605 /* 2606 * Note the amount of data that peer has sent into 2607 * our window, in order to estimate the sender's 2608 * buffer size. 2609 * XXX: Unused. 2610 */ 2611 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2612 #endif 2613 } else { 2614 m_freem(m); 2615 thflags &= ~TH_FIN; 2616 } 2617 2618 /* 2619 * If FIN is received ACK the FIN and let the user know 2620 * that the connection is closing. 2621 */ 2622 if (thflags & TH_FIN) { 2623 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2624 socantrcvmore(so); 2625 /* 2626 * If connection is half-synchronized 2627 * (ie NEEDSYN flag on) then delay ACK, 2628 * so it may be piggybacked when SYN is sent. 2629 * Otherwise, since we received a FIN then no 2630 * more input can be expected, send ACK now. 2631 */ 2632 if (tp->t_flags & TF_NEEDSYN) 2633 tp->t_flags |= TF_DELACK; 2634 else 2635 tp->t_flags |= TF_ACKNOW; 2636 tp->rcv_nxt++; 2637 } 2638 switch (tp->t_state) { 2639 2640 /* 2641 * In SYN_RECEIVED and ESTABLISHED STATES 2642 * enter the CLOSE_WAIT state. 2643 */ 2644 case TCPS_SYN_RECEIVED: 2645 tp->t_starttime = ticks; 2646 /* FALLTHROUGH */ 2647 case TCPS_ESTABLISHED: 2648 tp->t_state = TCPS_CLOSE_WAIT; 2649 break; 2650 2651 /* 2652 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2653 * enter the CLOSING state. 2654 */ 2655 case TCPS_FIN_WAIT_1: 2656 tp->t_state = TCPS_CLOSING; 2657 break; 2658 2659 /* 2660 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2661 * starting the time-wait timer, turning off the other 2662 * standard timers. 2663 */ 2664 case TCPS_FIN_WAIT_2: 2665 INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2666 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2667 "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2668 ti_locked)); 2669 2670 tcp_twstart(tp); 2671 INP_INFO_WUNLOCK(&V_tcbinfo); 2672 return; 2673 } 2674 } 2675 if (ti_locked == TI_RLOCKED) 2676 INP_INFO_RUNLOCK(&V_tcbinfo); 2677 else if (ti_locked == TI_WLOCKED) 2678 INP_INFO_WUNLOCK(&V_tcbinfo); 2679 else 2680 panic("%s: dodata epilogue ti_locked %d", __func__, 2681 ti_locked); 2682 ti_locked = TI_UNLOCKED; 2683 2684 #ifdef TCPDEBUG 2685 if (so->so_options & SO_DEBUG) 2686 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2687 &tcp_savetcp, 0); 2688 #endif 2689 2690 /* 2691 * Return any desired output. 2692 */ 2693 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2694 (void) tcp_output(tp); 2695 2696 check_delack: 2697 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 2698 __func__, ti_locked)); 2699 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2700 INP_WLOCK_ASSERT(tp->t_inpcb); 2701 2702 if (tp->t_flags & TF_DELACK) { 2703 tp->t_flags &= ~TF_DELACK; 2704 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2705 } 2706 INP_WUNLOCK(tp->t_inpcb); 2707 return; 2708 2709 dropafterack: 2710 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED, 2711 ("tcp_do_segment: dropafterack ti_locked %d", ti_locked)); 2712 2713 /* 2714 * Generate an ACK dropping incoming segment if it occupies 2715 * sequence space, where the ACK reflects our state. 2716 * 2717 * We can now skip the test for the RST flag since all 2718 * paths to this code happen after packets containing 2719 * RST have been dropped. 2720 * 2721 * In the SYN-RECEIVED state, don't send an ACK unless the 2722 * segment we received passes the SYN-RECEIVED ACK test. 2723 * If it fails send a RST. This breaks the loop in the 2724 * "LAND" DoS attack, and also prevents an ACK storm 2725 * between two listening ports that have been sent forged 2726 * SYN segments, each with the source address of the other. 2727 */ 2728 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2729 (SEQ_GT(tp->snd_una, th->th_ack) || 2730 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2731 rstreason = BANDLIM_RST_OPENPORT; 2732 goto dropwithreset; 2733 } 2734 #ifdef TCPDEBUG 2735 if (so->so_options & SO_DEBUG) 2736 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2737 &tcp_savetcp, 0); 2738 #endif 2739 if (ti_locked == TI_RLOCKED) 2740 INP_INFO_RUNLOCK(&V_tcbinfo); 2741 else if (ti_locked == TI_WLOCKED) 2742 INP_INFO_WUNLOCK(&V_tcbinfo); 2743 else 2744 panic("%s: dropafterack epilogue ti_locked %d", __func__, 2745 ti_locked); 2746 ti_locked = TI_UNLOCKED; 2747 2748 tp->t_flags |= TF_ACKNOW; 2749 (void) tcp_output(tp); 2750 INP_WUNLOCK(tp->t_inpcb); 2751 m_freem(m); 2752 return; 2753 2754 dropwithreset: 2755 if (ti_locked == TI_RLOCKED) 2756 INP_INFO_RUNLOCK(&V_tcbinfo); 2757 else if (ti_locked == TI_WLOCKED) 2758 INP_INFO_WUNLOCK(&V_tcbinfo); 2759 else 2760 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked); 2761 ti_locked = TI_UNLOCKED; 2762 2763 if (tp != NULL) { 2764 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2765 INP_WUNLOCK(tp->t_inpcb); 2766 } else 2767 tcp_dropwithreset(m, th, NULL, tlen, rstreason); 2768 return; 2769 2770 drop: 2771 if (ti_locked == TI_RLOCKED) 2772 INP_INFO_RUNLOCK(&V_tcbinfo); 2773 else if (ti_locked == TI_WLOCKED) 2774 INP_INFO_WUNLOCK(&V_tcbinfo); 2775 #ifdef INVARIANTS 2776 else 2777 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 2778 #endif 2779 ti_locked = TI_UNLOCKED; 2780 2781 /* 2782 * Drop space held by incoming segment and return. 2783 */ 2784 #ifdef TCPDEBUG 2785 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2786 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2787 &tcp_savetcp, 0); 2788 #endif 2789 if (tp != NULL) 2790 INP_WUNLOCK(tp->t_inpcb); 2791 m_freem(m); 2792 } 2793 2794 /* 2795 * Issue RST and make ACK acceptable to originator of segment. 2796 * The mbuf must still include the original packet header. 2797 * tp may be NULL. 2798 */ 2799 static void 2800 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 2801 int tlen, int rstreason) 2802 { 2803 struct ip *ip; 2804 #ifdef INET6 2805 struct ip6_hdr *ip6; 2806 #endif 2807 2808 if (tp != NULL) { 2809 INP_WLOCK_ASSERT(tp->t_inpcb); 2810 } 2811 2812 /* Don't bother if destination was broadcast/multicast. */ 2813 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2814 goto drop; 2815 #ifdef INET6 2816 if (mtod(m, struct ip *)->ip_v == 6) { 2817 ip6 = mtod(m, struct ip6_hdr *); 2818 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2819 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2820 goto drop; 2821 /* IPv6 anycast check is done at tcp6_input() */ 2822 } else 2823 #endif 2824 { 2825 ip = mtod(m, struct ip *); 2826 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2827 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2828 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2829 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2830 goto drop; 2831 } 2832 2833 /* Perform bandwidth limiting. */ 2834 if (badport_bandlim(rstreason) < 0) 2835 goto drop; 2836 2837 /* tcp_respond consumes the mbuf chain. */ 2838 if (th->th_flags & TH_ACK) { 2839 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 2840 th->th_ack, TH_RST); 2841 } else { 2842 if (th->th_flags & TH_SYN) 2843 tlen++; 2844 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2845 (tcp_seq)0, TH_RST|TH_ACK); 2846 } 2847 return; 2848 drop: 2849 m_freem(m); 2850 } 2851 2852 /* 2853 * Parse TCP options and place in tcpopt. 2854 */ 2855 static void 2856 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 2857 { 2858 INIT_VNET_INET(curvnet); 2859 int opt, optlen; 2860 2861 to->to_flags = 0; 2862 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2863 opt = cp[0]; 2864 if (opt == TCPOPT_EOL) 2865 break; 2866 if (opt == TCPOPT_NOP) 2867 optlen = 1; 2868 else { 2869 if (cnt < 2) 2870 break; 2871 optlen = cp[1]; 2872 if (optlen < 2 || optlen > cnt) 2873 break; 2874 } 2875 switch (opt) { 2876 case TCPOPT_MAXSEG: 2877 if (optlen != TCPOLEN_MAXSEG) 2878 continue; 2879 if (!(flags & TO_SYN)) 2880 continue; 2881 to->to_flags |= TOF_MSS; 2882 bcopy((char *)cp + 2, 2883 (char *)&to->to_mss, sizeof(to->to_mss)); 2884 to->to_mss = ntohs(to->to_mss); 2885 break; 2886 case TCPOPT_WINDOW: 2887 if (optlen != TCPOLEN_WINDOW) 2888 continue; 2889 if (!(flags & TO_SYN)) 2890 continue; 2891 to->to_flags |= TOF_SCALE; 2892 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 2893 break; 2894 case TCPOPT_TIMESTAMP: 2895 if (optlen != TCPOLEN_TIMESTAMP) 2896 continue; 2897 to->to_flags |= TOF_TS; 2898 bcopy((char *)cp + 2, 2899 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2900 to->to_tsval = ntohl(to->to_tsval); 2901 bcopy((char *)cp + 6, 2902 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2903 to->to_tsecr = ntohl(to->to_tsecr); 2904 break; 2905 #ifdef TCP_SIGNATURE 2906 /* 2907 * XXX In order to reply to a host which has set the 2908 * TCP_SIGNATURE option in its initial SYN, we have to 2909 * record the fact that the option was observed here 2910 * for the syncache code to perform the correct response. 2911 */ 2912 case TCPOPT_SIGNATURE: 2913 if (optlen != TCPOLEN_SIGNATURE) 2914 continue; 2915 to->to_flags |= TOF_SIGNATURE; 2916 to->to_signature = cp + 2; 2917 break; 2918 #endif 2919 case TCPOPT_SACK_PERMITTED: 2920 if (optlen != TCPOLEN_SACK_PERMITTED) 2921 continue; 2922 if (!(flags & TO_SYN)) 2923 continue; 2924 if (!V_tcp_do_sack) 2925 continue; 2926 to->to_flags |= TOF_SACKPERM; 2927 break; 2928 case TCPOPT_SACK: 2929 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2930 continue; 2931 if (flags & TO_SYN) 2932 continue; 2933 to->to_flags |= TOF_SACK; 2934 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2935 to->to_sacks = cp + 2; 2936 V_tcpstat.tcps_sack_rcv_blocks++; 2937 break; 2938 default: 2939 continue; 2940 } 2941 } 2942 } 2943 2944 /* 2945 * Pull out of band byte out of a segment so 2946 * it doesn't appear in the user's data queue. 2947 * It is still reflected in the segment length for 2948 * sequencing purposes. 2949 */ 2950 static void 2951 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 2952 int off) 2953 { 2954 int cnt = off + th->th_urp - 1; 2955 2956 while (cnt >= 0) { 2957 if (m->m_len > cnt) { 2958 char *cp = mtod(m, caddr_t) + cnt; 2959 struct tcpcb *tp = sototcpcb(so); 2960 2961 INP_WLOCK_ASSERT(tp->t_inpcb); 2962 2963 tp->t_iobc = *cp; 2964 tp->t_oobflags |= TCPOOB_HAVEDATA; 2965 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2966 m->m_len--; 2967 if (m->m_flags & M_PKTHDR) 2968 m->m_pkthdr.len--; 2969 return; 2970 } 2971 cnt -= m->m_len; 2972 m = m->m_next; 2973 if (m == NULL) 2974 break; 2975 } 2976 panic("tcp_pulloutofband"); 2977 } 2978 2979 /* 2980 * Collect new round-trip time estimate 2981 * and update averages and current timeout. 2982 */ 2983 static void 2984 tcp_xmit_timer(struct tcpcb *tp, int rtt) 2985 { 2986 INIT_VNET_INET(tp->t_inpcb->inp_vnet); 2987 int delta; 2988 2989 INP_WLOCK_ASSERT(tp->t_inpcb); 2990 2991 V_tcpstat.tcps_rttupdated++; 2992 tp->t_rttupdated++; 2993 if (tp->t_srtt != 0) { 2994 /* 2995 * srtt is stored as fixed point with 5 bits after the 2996 * binary point (i.e., scaled by 8). The following magic 2997 * is equivalent to the smoothing algorithm in rfc793 with 2998 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2999 * point). Adjust rtt to origin 0. 3000 */ 3001 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3002 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3003 3004 if ((tp->t_srtt += delta) <= 0) 3005 tp->t_srtt = 1; 3006 3007 /* 3008 * We accumulate a smoothed rtt variance (actually, a 3009 * smoothed mean difference), then set the retransmit 3010 * timer to smoothed rtt + 4 times the smoothed variance. 3011 * rttvar is stored as fixed point with 4 bits after the 3012 * binary point (scaled by 16). The following is 3013 * equivalent to rfc793 smoothing with an alpha of .75 3014 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3015 * rfc793's wired-in beta. 3016 */ 3017 if (delta < 0) 3018 delta = -delta; 3019 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3020 if ((tp->t_rttvar += delta) <= 0) 3021 tp->t_rttvar = 1; 3022 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 3023 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3024 } else { 3025 /* 3026 * No rtt measurement yet - use the unsmoothed rtt. 3027 * Set the variance to half the rtt (so our first 3028 * retransmit happens at 3*rtt). 3029 */ 3030 tp->t_srtt = rtt << TCP_RTT_SHIFT; 3031 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 3032 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3033 } 3034 tp->t_rtttime = 0; 3035 tp->t_rxtshift = 0; 3036 3037 /* 3038 * the retransmit should happen at rtt + 4 * rttvar. 3039 * Because of the way we do the smoothing, srtt and rttvar 3040 * will each average +1/2 tick of bias. When we compute 3041 * the retransmit timer, we want 1/2 tick of rounding and 3042 * 1 extra tick because of +-1/2 tick uncertainty in the 3043 * firing of the timer. The bias will give us exactly the 3044 * 1.5 tick we need. But, because the bias is 3045 * statistical, we have to test that we don't drop below 3046 * the minimum feasible timer (which is 2 ticks). 3047 */ 3048 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 3049 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3050 3051 /* 3052 * We received an ack for a packet that wasn't retransmitted; 3053 * it is probably safe to discard any error indications we've 3054 * received recently. This isn't quite right, but close enough 3055 * for now (a route might have failed after we sent a segment, 3056 * and the return path might not be symmetrical). 3057 */ 3058 tp->t_softerror = 0; 3059 } 3060 3061 /* 3062 * Determine a reasonable value for maxseg size. 3063 * If the route is known, check route for mtu. 3064 * If none, use an mss that can be handled on the outgoing 3065 * interface without forcing IP to fragment; if bigger than 3066 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 3067 * to utilize large mbufs. If no route is found, route has no mtu, 3068 * or the destination isn't local, use a default, hopefully conservative 3069 * size (usually 512 or the default IP max size, but no more than the mtu 3070 * of the interface), as we can't discover anything about intervening 3071 * gateways or networks. We also initialize the congestion/slow start 3072 * window to be a single segment if the destination isn't local. 3073 * While looking at the routing entry, we also initialize other path-dependent 3074 * parameters from pre-set or cached values in the routing entry. 3075 * 3076 * Also take into account the space needed for options that we 3077 * send regularly. Make maxseg shorter by that amount to assure 3078 * that we can send maxseg amount of data even when the options 3079 * are present. Store the upper limit of the length of options plus 3080 * data in maxopd. 3081 * 3082 * In case of T/TCP, we call this routine during implicit connection 3083 * setup as well (offer = -1), to initialize maxseg from the cached 3084 * MSS of our peer. 3085 * 3086 * NOTE that this routine is only called when we process an incoming 3087 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 3088 */ 3089 void 3090 tcp_mss_update(struct tcpcb *tp, int offer, 3091 struct hc_metrics_lite *metricptr, int *mtuflags) 3092 { 3093 INIT_VNET_INET(tp->t_inpcb->inp_vnet); 3094 int mss; 3095 u_long maxmtu; 3096 struct inpcb *inp = tp->t_inpcb; 3097 struct hc_metrics_lite metrics; 3098 int origoffer = offer; 3099 #ifdef INET6 3100 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3101 size_t min_protoh = isipv6 ? 3102 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3103 sizeof (struct tcpiphdr); 3104 #else 3105 const size_t min_protoh = sizeof(struct tcpiphdr); 3106 #endif 3107 3108 INP_WLOCK_ASSERT(tp->t_inpcb); 3109 3110 /* Initialize. */ 3111 #ifdef INET6 3112 if (isipv6) { 3113 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags); 3114 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3115 } else 3116 #endif 3117 { 3118 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags); 3119 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3120 } 3121 3122 /* 3123 * No route to sender, stay with default mss and return. 3124 */ 3125 if (maxmtu == 0) { 3126 /* 3127 * In case we return early we need to initialize metrics 3128 * to a defined state as tcp_hc_get() would do for us 3129 * if there was no cache hit. 3130 */ 3131 if (metricptr != NULL) 3132 bzero(metricptr, sizeof(struct hc_metrics_lite)); 3133 return; 3134 } 3135 3136 /* What have we got? */ 3137 switch (offer) { 3138 case 0: 3139 /* 3140 * Offer == 0 means that there was no MSS on the SYN 3141 * segment, in this case we use tcp_mssdflt as 3142 * already assigned to t_maxopd above. 3143 */ 3144 offer = tp->t_maxopd; 3145 break; 3146 3147 case -1: 3148 /* 3149 * Offer == -1 means that we didn't receive SYN yet. 3150 */ 3151 /* FALLTHROUGH */ 3152 3153 default: 3154 /* 3155 * Prevent DoS attack with too small MSS. Round up 3156 * to at least minmss. 3157 */ 3158 offer = max(offer, V_tcp_minmss); 3159 } 3160 3161 /* 3162 * rmx information is now retrieved from tcp_hostcache. 3163 */ 3164 tcp_hc_get(&inp->inp_inc, &metrics); 3165 if (metricptr != NULL) 3166 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 3167 3168 /* 3169 * If there's a discovered mtu int tcp hostcache, use it 3170 * else, use the link mtu. 3171 */ 3172 if (metrics.rmx_mtu) 3173 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3174 else { 3175 #ifdef INET6 3176 if (isipv6) { 3177 mss = maxmtu - min_protoh; 3178 if (!V_path_mtu_discovery && 3179 !in6_localaddr(&inp->in6p_faddr)) 3180 mss = min(mss, V_tcp_v6mssdflt); 3181 } else 3182 #endif 3183 { 3184 mss = maxmtu - min_protoh; 3185 if (!V_path_mtu_discovery && 3186 !in_localaddr(inp->inp_faddr)) 3187 mss = min(mss, V_tcp_mssdflt); 3188 } 3189 /* 3190 * XXX - The above conditional (mss = maxmtu - min_protoh) 3191 * probably violates the TCP spec. 3192 * The problem is that, since we don't know the 3193 * other end's MSS, we are supposed to use a conservative 3194 * default. But, if we do that, then MTU discovery will 3195 * never actually take place, because the conservative 3196 * default is much less than the MTUs typically seen 3197 * on the Internet today. For the moment, we'll sweep 3198 * this under the carpet. 3199 * 3200 * The conservative default might not actually be a problem 3201 * if the only case this occurs is when sending an initial 3202 * SYN with options and data to a host we've never talked 3203 * to before. Then, they will reply with an MSS value which 3204 * will get recorded and the new parameters should get 3205 * recomputed. For Further Study. 3206 */ 3207 } 3208 mss = min(mss, offer); 3209 3210 /* 3211 * Sanity check: make sure that maxopd will be large 3212 * enough to allow some data on segments even if the 3213 * all the option space is used (40bytes). Otherwise 3214 * funny things may happen in tcp_output. 3215 */ 3216 mss = max(mss, 64); 3217 3218 /* 3219 * maxopd stores the maximum length of data AND options 3220 * in a segment; maxseg is the amount of data in a normal 3221 * segment. We need to store this value (maxopd) apart 3222 * from maxseg, because now every segment carries options 3223 * and thus we normally have somewhat less data in segments. 3224 */ 3225 tp->t_maxopd = mss; 3226 3227 /* 3228 * origoffer==-1 indicates that no segments were received yet. 3229 * In this case we just guess. 3230 */ 3231 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3232 (origoffer == -1 || 3233 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3234 mss -= TCPOLEN_TSTAMP_APPA; 3235 3236 #if (MCLBYTES & (MCLBYTES - 1)) == 0 3237 if (mss > MCLBYTES) 3238 mss &= ~(MCLBYTES-1); 3239 #else 3240 if (mss > MCLBYTES) 3241 mss = mss / MCLBYTES * MCLBYTES; 3242 #endif 3243 tp->t_maxseg = mss; 3244 } 3245 3246 void 3247 tcp_mss(struct tcpcb *tp, int offer) 3248 { 3249 int rtt, mss; 3250 u_long bufsize; 3251 struct inpcb *inp; 3252 struct socket *so; 3253 struct hc_metrics_lite metrics; 3254 int mtuflags = 0; 3255 #ifdef INET6 3256 int isipv6; 3257 #endif 3258 KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 3259 INIT_VNET_INET(tp->t_vnet); 3260 3261 tcp_mss_update(tp, offer, &metrics, &mtuflags); 3262 3263 mss = tp->t_maxseg; 3264 inp = tp->t_inpcb; 3265 #ifdef INET6 3266 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3267 #endif 3268 3269 /* 3270 * If there's a pipesize, change the socket buffer to that size, 3271 * don't change if sb_hiwat is different than default (then it 3272 * has been changed on purpose with setsockopt). 3273 * Make the socket buffers an integral number of mss units; 3274 * if the mss is larger than the socket buffer, decrease the mss. 3275 */ 3276 so = inp->inp_socket; 3277 SOCKBUF_LOCK(&so->so_snd); 3278 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 3279 bufsize = metrics.rmx_sendpipe; 3280 else 3281 bufsize = so->so_snd.sb_hiwat; 3282 if (bufsize < mss) 3283 mss = bufsize; 3284 else { 3285 bufsize = roundup(bufsize, mss); 3286 if (bufsize > sb_max) 3287 bufsize = sb_max; 3288 if (bufsize > so->so_snd.sb_hiwat) 3289 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3290 } 3291 SOCKBUF_UNLOCK(&so->so_snd); 3292 tp->t_maxseg = mss; 3293 3294 SOCKBUF_LOCK(&so->so_rcv); 3295 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 3296 bufsize = metrics.rmx_recvpipe; 3297 else 3298 bufsize = so->so_rcv.sb_hiwat; 3299 if (bufsize > mss) { 3300 bufsize = roundup(bufsize, mss); 3301 if (bufsize > sb_max) 3302 bufsize = sb_max; 3303 if (bufsize > so->so_rcv.sb_hiwat) 3304 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3305 } 3306 SOCKBUF_UNLOCK(&so->so_rcv); 3307 /* 3308 * While we're here, check the others too. 3309 */ 3310 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 3311 tp->t_srtt = rtt; 3312 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 3313 V_tcpstat.tcps_usedrtt++; 3314 if (metrics.rmx_rttvar) { 3315 tp->t_rttvar = metrics.rmx_rttvar; 3316 V_tcpstat.tcps_usedrttvar++; 3317 } else { 3318 /* default variation is +- 1 rtt */ 3319 tp->t_rttvar = 3320 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 3321 } 3322 TCPT_RANGESET(tp->t_rxtcur, 3323 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 3324 tp->t_rttmin, TCPTV_REXMTMAX); 3325 } 3326 if (metrics.rmx_ssthresh) { 3327 /* 3328 * There's some sort of gateway or interface 3329 * buffer limit on the path. Use this to set 3330 * the slow start threshhold, but set the 3331 * threshold to no less than 2*mss. 3332 */ 3333 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 3334 V_tcpstat.tcps_usedssthresh++; 3335 } 3336 if (metrics.rmx_bandwidth) 3337 tp->snd_bandwidth = metrics.rmx_bandwidth; 3338 3339 /* 3340 * Set the slow-start flight size depending on whether this 3341 * is a local network or not. 3342 * 3343 * Extend this so we cache the cwnd too and retrieve it here. 3344 * Make cwnd even bigger than RFC3390 suggests but only if we 3345 * have previous experience with the remote host. Be careful 3346 * not make cwnd bigger than remote receive window or our own 3347 * send socket buffer. Maybe put some additional upper bound 3348 * on the retrieved cwnd. Should do incremental updates to 3349 * hostcache when cwnd collapses so next connection doesn't 3350 * overloads the path again. 3351 * 3352 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 3353 * We currently check only in syncache_socket for that. 3354 */ 3355 #define TCP_METRICS_CWND 3356 #ifdef TCP_METRICS_CWND 3357 if (metrics.rmx_cwnd) 3358 tp->snd_cwnd = max(mss, 3359 min(metrics.rmx_cwnd / 2, 3360 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 3361 else 3362 #endif 3363 if (V_tcp_do_rfc3390) 3364 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 3365 #ifdef INET6 3366 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 3367 (!isipv6 && in_localaddr(inp->inp_faddr))) 3368 #else 3369 else if (in_localaddr(inp->inp_faddr)) 3370 #endif 3371 tp->snd_cwnd = mss * V_ss_fltsz_local; 3372 else 3373 tp->snd_cwnd = mss * V_ss_fltsz; 3374 3375 /* Check the interface for TSO capabilities. */ 3376 if (mtuflags & CSUM_TSO) 3377 tp->t_flags |= TF_TSO; 3378 } 3379 3380 /* 3381 * Determine the MSS option to send on an outgoing SYN. 3382 */ 3383 int 3384 tcp_mssopt(struct in_conninfo *inc) 3385 { 3386 INIT_VNET_INET(curvnet); 3387 int mss = 0; 3388 u_long maxmtu = 0; 3389 u_long thcmtu = 0; 3390 size_t min_protoh; 3391 3392 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3393 3394 #ifdef INET6 3395 if (inc->inc_flags & INC_ISIPV6) { 3396 mss = V_tcp_v6mssdflt; 3397 maxmtu = tcp_maxmtu6(inc, NULL); 3398 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3399 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3400 } else 3401 #endif 3402 { 3403 mss = V_tcp_mssdflt; 3404 maxmtu = tcp_maxmtu(inc, NULL); 3405 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 3406 min_protoh = sizeof(struct tcpiphdr); 3407 } 3408 if (maxmtu && thcmtu) 3409 mss = min(maxmtu, thcmtu) - min_protoh; 3410 else if (maxmtu || thcmtu) 3411 mss = max(maxmtu, thcmtu) - min_protoh; 3412 3413 return (mss); 3414 } 3415 3416 3417 /* 3418 * On a partial ack arrives, force the retransmission of the 3419 * next unacknowledged segment. Do not clear tp->t_dupacks. 3420 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3421 * be started again. 3422 */ 3423 static void 3424 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3425 { 3426 tcp_seq onxt = tp->snd_nxt; 3427 u_long ocwnd = tp->snd_cwnd; 3428 3429 INP_WLOCK_ASSERT(tp->t_inpcb); 3430 3431 tcp_timer_activate(tp, TT_REXMT, 0); 3432 tp->t_rtttime = 0; 3433 tp->snd_nxt = th->th_ack; 3434 /* 3435 * Set snd_cwnd to one segment beyond acknowledged offset. 3436 * (tp->snd_una has not yet been updated when this function is called.) 3437 */ 3438 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3439 tp->t_flags |= TF_ACKNOW; 3440 (void) tcp_output(tp); 3441 tp->snd_cwnd = ocwnd; 3442 if (SEQ_GT(onxt, tp->snd_nxt)) 3443 tp->snd_nxt = onxt; 3444 /* 3445 * Partial window deflation. Relies on fact that tp->snd_una 3446 * not updated yet. 3447 */ 3448 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3449 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3450 else 3451 tp->snd_cwnd = 0; 3452 tp->snd_cwnd += tp->t_maxseg; 3453 } 3454