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