tcp_input.c revision 171648
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 * $FreeBSD: head/sys/netinet/tcp_input.c 171648 2007-07-29 01:31:33Z mjacob $ 31 */ 32 33#include "opt_ipfw.h" /* for ipfw_fwd */ 34#include "opt_inet.h" 35#include "opt_inet6.h" 36#include "opt_ipsec.h" 37#include "opt_mac.h" 38#include "opt_tcpdebug.h" 39 40#include <sys/param.h> 41#include <sys/kernel.h> 42#include <sys/malloc.h> 43#include <sys/mbuf.h> 44#include <sys/proc.h> /* for proc0 declaration */ 45#include <sys/protosw.h> 46#include <sys/signalvar.h> 47#include <sys/socket.h> 48#include <sys/socketvar.h> 49#include <sys/sysctl.h> 50#include <sys/syslog.h> 51#include <sys/systm.h> 52 53#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 54 55#include <vm/uma.h> 56 57#include <net/if.h> 58#include <net/route.h> 59 60#define TCPSTATES /* for logging */ 61 62#include <netinet/in.h> 63#include <netinet/in_pcb.h> 64#include <netinet/in_systm.h> 65#include <netinet/in_var.h> 66#include <netinet/ip.h> 67#include <netinet/ip_icmp.h> /* required for icmp_var.h */ 68#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 69#include <netinet/ip_var.h> 70#include <netinet/ip_options.h> 71#include <netinet/ip6.h> 72#include <netinet/icmp6.h> 73#include <netinet6/in6_pcb.h> 74#include <netinet6/ip6_var.h> 75#include <netinet6/nd6.h> 76#include <netinet/tcp.h> 77#include <netinet/tcp_fsm.h> 78#include <netinet/tcp_seq.h> 79#include <netinet/tcp_timer.h> 80#include <netinet/tcp_var.h> 81#include <netinet6/tcp6_var.h> 82#include <netinet/tcpip.h> 83#include <netinet/tcp_syncache.h> 84#ifdef TCPDEBUG 85#include <netinet/tcp_debug.h> 86#endif /* TCPDEBUG */ 87 88#ifdef IPSEC 89#include <netipsec/ipsec.h> 90#include <netipsec/ipsec6.h> 91#endif /*IPSEC*/ 92 93#include <machine/in_cksum.h> 94 95#include <security/mac/mac_framework.h> 96 97static const int tcprexmtthresh = 3; 98 99struct tcpstat tcpstat; 100SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW, 101 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 102 103int tcp_log_in_vain = 0; 104SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 105 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports"); 106 107static int blackhole = 0; 108SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 109 &blackhole, 0, "Do not send RST on segments to closed ports"); 110 111int tcp_delack_enabled = 1; 112SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 113 &tcp_delack_enabled, 0, 114 "Delay ACK to try and piggyback it onto a data packet"); 115 116static int drop_synfin = 0; 117SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 118 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set"); 119 120static int tcp_do_rfc3042 = 1; 121SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 122 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)"); 123 124static int tcp_do_rfc3390 = 1; 125SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 126 &tcp_do_rfc3390, 0, 127 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 128 129static int tcp_insecure_rst = 0; 130SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 131 &tcp_insecure_rst, 0, 132 "Follow the old (insecure) criteria for accepting RST packets"); 133 134int tcp_do_autorcvbuf = 1; 135SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 136 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing"); 137 138int tcp_autorcvbuf_inc = 16*1024; 139SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 140 &tcp_autorcvbuf_inc, 0, 141 "Incrementor step size of automatic receive buffer"); 142 143int tcp_autorcvbuf_max = 256*1024; 144SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 145 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer"); 146 147struct inpcbhead tcb; 148#define tcb6 tcb /* for KAME src sync over BSD*'s */ 149struct inpcbinfo tcbinfo; 150 151static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 152static void tcp_do_segment(struct mbuf *, struct tcphdr *, 153 struct socket *, struct tcpcb *, int, int); 154static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 155 struct tcpcb *, int, int); 156static void tcp_pulloutofband(struct socket *, 157 struct tcphdr *, struct mbuf *, int); 158static void tcp_xmit_timer(struct tcpcb *, int); 159static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 160 161/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 162#ifdef INET6 163#define ND6_HINT(tp) \ 164do { \ 165 if ((tp) && (tp)->t_inpcb && \ 166 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 167 nd6_nud_hint(NULL, NULL, 0); \ 168} while (0) 169#else 170#define ND6_HINT(tp) 171#endif 172 173/* 174 * Indicate whether this ack should be delayed. We can delay the ack if 175 * - there is no delayed ack timer in progress and 176 * - our last ack wasn't a 0-sized window. We never want to delay 177 * the ack that opens up a 0-sized window and 178 * - delayed acks are enabled or 179 * - this is a half-synchronized T/TCP connection. 180 */ 181#define DELAY_ACK(tp) \ 182 ((!tcp_timer_active(tp, TT_DELACK) && \ 183 (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 184 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 185 186 187/* 188 * TCP input handling is split into multiple parts: 189 * tcp6_input is a thin wrapper around tcp_input for the extended 190 * ip6_protox[] call format in ip6_input 191 * tcp_input handles primary segment validation, inpcb lookup and 192 * SYN processing on listen sockets 193 * tcp_do_segment processes the ACK and text of the segment for 194 * establishing, established and closing connections 195 */ 196#ifdef INET6 197int 198tcp6_input(struct mbuf **mp, int *offp, int proto) 199{ 200 struct mbuf *m = *mp; 201 struct in6_ifaddr *ia6; 202 203 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 204 205 /* 206 * draft-itojun-ipv6-tcp-to-anycast 207 * better place to put this in? 208 */ 209 ia6 = ip6_getdstifaddr(m); 210 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 211 struct ip6_hdr *ip6; 212 213 ip6 = mtod(m, struct ip6_hdr *); 214 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 215 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 216 return IPPROTO_DONE; 217 } 218 219 tcp_input(m, *offp); 220 return IPPROTO_DONE; 221} 222#endif 223 224void 225tcp_input(struct mbuf *m, int off0) 226{ 227 struct tcphdr *th; 228 struct ip *ip = NULL; 229 struct ipovly *ipov; 230 struct inpcb *inp = NULL; 231 struct tcpcb *tp = NULL; 232 struct socket *so = NULL; 233 u_char *optp = NULL; 234 int optlen = 0; 235 int len, tlen, off; 236 int drop_hdrlen; 237 int thflags; 238 int rstreason = 0; /* For badport_bandlim accounting purposes */ 239#ifdef IPFIREWALL_FORWARD 240 struct m_tag *fwd_tag; 241#endif 242#ifdef INET6 243 struct ip6_hdr *ip6 = NULL; 244 int isipv6; 245#else 246 const void *ip6 = NULL; 247 const int isipv6 = 0; 248#endif 249 struct tcpopt to; /* options in this segment */ 250 char *s = NULL; /* address and port logging */ 251 252#ifdef TCPDEBUG 253 /* 254 * The size of tcp_saveipgen must be the size of the max ip header, 255 * now IPv6. 256 */ 257 u_char tcp_saveipgen[IP6_HDR_LEN]; 258 struct tcphdr tcp_savetcp; 259 short ostate = 0; 260#endif 261 262#ifdef INET6 263 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 264#endif 265 266 to.to_flags = 0; 267 tcpstat.tcps_rcvtotal++; 268 269 if (isipv6) { 270#ifdef INET6 271 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 272 ip6 = mtod(m, struct ip6_hdr *); 273 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 274 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) { 275 tcpstat.tcps_rcvbadsum++; 276 goto drop; 277 } 278 th = (struct tcphdr *)((caddr_t)ip6 + off0); 279 280 /* 281 * Be proactive about unspecified IPv6 address in source. 282 * As we use all-zero to indicate unbounded/unconnected pcb, 283 * unspecified IPv6 address can be used to confuse us. 284 * 285 * Note that packets with unspecified IPv6 destination is 286 * already dropped in ip6_input. 287 */ 288 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 289 /* XXX stat */ 290 goto drop; 291 } 292#else 293 th = NULL; /* XXX: Avoid compiler warning. */ 294#endif 295 } else { 296 /* 297 * Get IP and TCP header together in first mbuf. 298 * Note: IP leaves IP header in first mbuf. 299 */ 300 if (off0 > sizeof (struct ip)) { 301 ip_stripoptions(m, (struct mbuf *)0); 302 off0 = sizeof(struct ip); 303 } 304 if (m->m_len < sizeof (struct tcpiphdr)) { 305 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 306 == NULL) { 307 tcpstat.tcps_rcvshort++; 308 return; 309 } 310 } 311 ip = mtod(m, struct ip *); 312 ipov = (struct ipovly *)ip; 313 th = (struct tcphdr *)((caddr_t)ip + off0); 314 tlen = ip->ip_len; 315 316 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 317 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 318 th->th_sum = m->m_pkthdr.csum_data; 319 else 320 th->th_sum = in_pseudo(ip->ip_src.s_addr, 321 ip->ip_dst.s_addr, 322 htonl(m->m_pkthdr.csum_data + 323 ip->ip_len + 324 IPPROTO_TCP)); 325 th->th_sum ^= 0xffff; 326#ifdef TCPDEBUG 327 ipov->ih_len = (u_short)tlen; 328 ipov->ih_len = htons(ipov->ih_len); 329#endif 330 } else { 331 /* 332 * Checksum extended TCP header and data. 333 */ 334 len = sizeof (struct ip) + tlen; 335 bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 336 ipov->ih_len = (u_short)tlen; 337 ipov->ih_len = htons(ipov->ih_len); 338 th->th_sum = in_cksum(m, len); 339 } 340 if (th->th_sum) { 341 tcpstat.tcps_rcvbadsum++; 342 goto drop; 343 } 344 /* Re-initialization for later version check */ 345 ip->ip_v = IPVERSION; 346 } 347 348 /* 349 * Check that TCP offset makes sense, 350 * pull out TCP options and adjust length. XXX 351 */ 352 off = th->th_off << 2; 353 if (off < sizeof (struct tcphdr) || off > tlen) { 354 tcpstat.tcps_rcvbadoff++; 355 goto drop; 356 } 357 tlen -= off; /* tlen is used instead of ti->ti_len */ 358 if (off > sizeof (struct tcphdr)) { 359 if (isipv6) { 360#ifdef INET6 361 IP6_EXTHDR_CHECK(m, off0, off, ); 362 ip6 = mtod(m, struct ip6_hdr *); 363 th = (struct tcphdr *)((caddr_t)ip6 + off0); 364#endif 365 } else { 366 if (m->m_len < sizeof(struct ip) + off) { 367 if ((m = m_pullup(m, sizeof (struct ip) + off)) 368 == NULL) { 369 tcpstat.tcps_rcvshort++; 370 return; 371 } 372 ip = mtod(m, struct ip *); 373 ipov = (struct ipovly *)ip; 374 th = (struct tcphdr *)((caddr_t)ip + off0); 375 } 376 } 377 optlen = off - sizeof (struct tcphdr); 378 optp = (u_char *)(th + 1); 379 } 380 thflags = th->th_flags; 381 382 /* 383 * Convert TCP protocol specific fields to host format. 384 */ 385 th->th_seq = ntohl(th->th_seq); 386 th->th_ack = ntohl(th->th_ack); 387 th->th_win = ntohs(th->th_win); 388 th->th_urp = ntohs(th->th_urp); 389 390 /* 391 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 392 */ 393 drop_hdrlen = off0 + off; 394 395 /* 396 * Locate pcb for segment. 397 */ 398 INP_INFO_WLOCK(&tcbinfo); 399findpcb: 400 INP_INFO_WLOCK_ASSERT(&tcbinfo); 401#ifdef IPFIREWALL_FORWARD 402 /* 403 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 404 */ 405 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 406 407 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */ 408 struct sockaddr_in *next_hop; 409 410 next_hop = (struct sockaddr_in *)(fwd_tag+1); 411 /* 412 * Transparently forwarded. Pretend to be the destination. 413 * already got one like this? 414 */ 415 inp = in_pcblookup_hash(&tcbinfo, 416 ip->ip_src, th->th_sport, 417 ip->ip_dst, th->th_dport, 418 0, m->m_pkthdr.rcvif); 419 if (!inp) { 420 /* It's new. Try to find the ambushing socket. */ 421 inp = in_pcblookup_hash(&tcbinfo, 422 ip->ip_src, th->th_sport, 423 next_hop->sin_addr, 424 next_hop->sin_port ? 425 ntohs(next_hop->sin_port) : 426 th->th_dport, 427 INPLOOKUP_WILDCARD, 428 m->m_pkthdr.rcvif); 429 } 430 /* Remove the tag from the packet. We don't need it anymore. */ 431 m_tag_delete(m, fwd_tag); 432 } else 433#endif /* IPFIREWALL_FORWARD */ 434 { 435 if (isipv6) { 436#ifdef INET6 437 inp = in6_pcblookup_hash(&tcbinfo, 438 &ip6->ip6_src, th->th_sport, 439 &ip6->ip6_dst, th->th_dport, 440 INPLOOKUP_WILDCARD, 441 m->m_pkthdr.rcvif); 442#endif 443 } else 444 inp = in_pcblookup_hash(&tcbinfo, 445 ip->ip_src, th->th_sport, 446 ip->ip_dst, th->th_dport, 447 INPLOOKUP_WILDCARD, 448 m->m_pkthdr.rcvif); 449 } 450 451#ifdef IPSEC 452#ifdef INET6 453 if (isipv6 && inp != NULL && ipsec6_in_reject(m, inp)) { 454 ipsec6stat.in_polvio++; 455 goto dropunlock; 456 } else 457#endif /* INET6 */ 458 if (inp != NULL && ipsec4_in_reject(m, inp)) { 459 ipsec4stat.in_polvio++; 460 goto dropunlock; 461 } 462#endif /* IPSEC */ 463 464 /* 465 * If the INPCB does not exist then all data in the incoming 466 * segment is discarded and an appropriate RST is sent back. 467 */ 468 if (inp == NULL) { 469 /* 470 * Log communication attempts to ports that are not 471 * in use. 472 */ 473 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 474 tcp_log_in_vain == 2) { 475 if ((s = tcp_log_addrs(NULL, th, (void *)ip, ip6))) 476 log(LOG_INFO, "%s; %s: Connection attempt " 477 "to closed port\n", s, __func__); 478 } 479 /* 480 * When blackholing do not respond with a RST but 481 * completely ignore the segment and drop it. 482 */ 483 if ((blackhole == 1 && (thflags & TH_SYN)) || 484 blackhole == 2) 485 goto dropunlock; 486 487 rstreason = BANDLIM_RST_CLOSEDPORT; 488 goto dropwithreset; 489 } 490 INP_LOCK(inp); 491 492 /* 493 * Check the minimum TTL for socket. 494 */ 495 if (inp->inp_ip_minttl != 0) { 496#ifdef INET6 497 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 498 goto dropunlock; 499 else 500#endif 501 if (inp->inp_ip_minttl > ip->ip_ttl) 502 goto dropunlock; 503 } 504 505 /* 506 * A previous connection in TIMEWAIT state is supposed to catch 507 * stray or duplicate segments arriving late. If this segment 508 * was a legitimate new connection attempt the old INPCB gets 509 * removed and we can try again to find a listening socket. 510 */ 511 if (inp->inp_vflag & INP_TIMEWAIT) { 512 if (thflags & TH_SYN) 513 tcp_dooptions(&to, optp, optlen, TO_SYN); 514 /* 515 * NB: tcp_twcheck unlocks the INP and frees the mbuf. 516 */ 517 if (tcp_twcheck(inp, &to, th, m, tlen)) 518 goto findpcb; 519 INP_INFO_WUNLOCK(&tcbinfo); 520 return; 521 } 522 /* 523 * The TCPCB may no longer exist if the connection is winding 524 * down or it is in the CLOSED state. Either way we drop the 525 * segment and send an appropriate response. 526 */ 527 tp = intotcpcb(inp); 528 if (tp == NULL || tp->t_state == TCPS_CLOSED) { 529 rstreason = BANDLIM_RST_CLOSEDPORT; 530 goto dropwithreset; 531 } 532 533#ifdef MAC 534 INP_LOCK_ASSERT(inp); 535 if (mac_check_inpcb_deliver(inp, m)) 536 goto dropunlock; 537#endif 538 so = inp->inp_socket; 539 KASSERT(so != NULL, ("%s: so == NULL", __func__)); 540#ifdef TCPDEBUG 541 if (so->so_options & SO_DEBUG) { 542 ostate = tp->t_state; 543 if (isipv6) { 544#ifdef INET6 545 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 546#endif 547 } else 548 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 549 tcp_savetcp = *th; 550 } 551#endif 552 /* 553 * When the socket is accepting connections (the INPCB is in LISTEN 554 * state) we look into the SYN cache if this is a new connection 555 * attempt or the completion of a previous one. 556 */ 557 if (so->so_options & SO_ACCEPTCONN) { 558 struct in_conninfo inc; 559 560 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 561 "tp not listening", __func__)); 562 563 bzero(&inc, sizeof(inc)); 564 inc.inc_isipv6 = isipv6; 565#ifdef INET6 566 if (isipv6) { 567 inc.inc6_faddr = ip6->ip6_src; 568 inc.inc6_laddr = ip6->ip6_dst; 569 } else 570#endif 571 { 572 inc.inc_faddr = ip->ip_src; 573 inc.inc_laddr = ip->ip_dst; 574 } 575 inc.inc_fport = th->th_sport; 576 inc.inc_lport = th->th_dport; 577 578 /* 579 * Check for an existing connection attempt in syncache if 580 * the flag is only ACK. A successful lookup creates a new 581 * socket appended to the listen queue in SYN_RECEIVED state. 582 */ 583 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 584 /* 585 * Parse the TCP options here because 586 * syncookies need access to the reflected 587 * timestamp. 588 */ 589 tcp_dooptions(&to, optp, optlen, 0); 590 /* 591 * NB: syncache_expand() doesn't unlock 592 * inp and tcpinfo locks. 593 */ 594 if (!syncache_expand(&inc, &to, th, &so, m)) { 595 /* 596 * No syncache entry or ACK was not 597 * for our SYN/ACK. Send a RST. 598 * NB: syncache did its own logging 599 * of the failure cause. 600 */ 601 rstreason = BANDLIM_RST_OPENPORT; 602 goto dropwithreset; 603 } 604 if (so == NULL) { 605 /* 606 * We completed the 3-way handshake 607 * but could not allocate a socket 608 * either due to memory shortage, 609 * listen queue length limits or 610 * global socket limits. Send RST 611 * or wait and have the remote end 612 * retransmit the ACK for another 613 * try. 614 */ 615 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 616 log(LOG_DEBUG, "%s; %s: Listen socket: " 617 "Socket allocation failed due to " 618 "limits or memory shortage, %s\n", 619 s, __func__, (tcp_sc_rst_sock_fail ? 620 "sending RST" : "try again")); 621 if (tcp_sc_rst_sock_fail) { 622 rstreason = BANDLIM_UNLIMITED; 623 goto dropwithreset; 624 } else 625 goto dropunlock; 626 } 627 /* 628 * Socket is created in state SYN_RECEIVED. 629 * Unlock the listen socket, lock the newly 630 * created socket and update the tp variable. 631 */ 632 INP_UNLOCK(inp); /* listen socket */ 633 inp = sotoinpcb(so); 634 INP_LOCK(inp); /* new connection */ 635 tp = intotcpcb(inp); 636 KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 637 ("%s: ", __func__)); 638 /* 639 * Process the segment and the data it 640 * contains. tcp_do_segment() consumes 641 * the mbuf chain and unlocks the inpcb. 642 */ 643 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen); 644 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 645 return; 646 } 647 /* 648 * Segment flag validation for new connection attempts: 649 * 650 * Our (SYN|ACK) response was rejected. 651 * Check with syncache and remove entry to prevent 652 * retransmits. 653 * 654 * NB: syncache_chkrst does its own logging of failure 655 * causes. 656 */ 657 if (thflags & TH_RST) { 658 syncache_chkrst(&inc, th); 659 goto dropunlock; 660 } 661 /* 662 * We can't do anything without SYN. 663 */ 664 if ((thflags & TH_SYN) == 0) { 665 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 666 log(LOG_DEBUG, "%s; %s: Listen socket: " 667 "SYN is missing, segment ignored\n", 668 s, __func__); 669 tcpstat.tcps_badsyn++; 670 goto dropunlock; 671 } 672 /* 673 * (SYN|ACK) is bogus on a listen socket. 674 */ 675 if (thflags & TH_ACK) { 676 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 677 log(LOG_DEBUG, "%s; %s: Listen socket: " 678 "SYN|ACK invalid, segment rejected\n", 679 s, __func__); 680 syncache_badack(&inc); /* XXX: Not needed! */ 681 tcpstat.tcps_badsyn++; 682 rstreason = BANDLIM_RST_OPENPORT; 683 goto dropwithreset; 684 } 685 /* 686 * If the drop_synfin option is enabled, drop all 687 * segments with both the SYN and FIN bits set. 688 * This prevents e.g. nmap from identifying the 689 * TCP/IP stack. 690 * XXX: Poor reasoning. nmap has other methods 691 * and is constantly refining its stack detection 692 * strategies. 693 * XXX: This is a violation of the TCP specification 694 * and was used by RFC1644. 695 */ 696 if ((thflags & TH_FIN) && drop_synfin) { 697 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 698 log(LOG_DEBUG, "%s; %s: Listen socket: " 699 "SYN|FIN segment ignored (based on " 700 "sysctl setting)\n", s, __func__); 701 tcpstat.tcps_badsyn++; 702 goto dropunlock; 703 } 704 /* 705 * Segment's flags are (SYN) or (SYN|FIN). 706 * 707 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 708 * as they do not affect the state of the TCP FSM. 709 * The data pointed to by TH_URG and th_urp is ignored. 710 */ 711 KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 712 ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 713 KASSERT(thflags & (TH_SYN), 714 ("%s: Listen socket: TH_SYN not set", __func__)); 715#ifdef INET6 716 /* 717 * If deprecated address is forbidden, 718 * we do not accept SYN to deprecated interface 719 * address to prevent any new inbound connection from 720 * getting established. 721 * When we do not accept SYN, we send a TCP RST, 722 * with deprecated source address (instead of dropping 723 * it). We compromise it as it is much better for peer 724 * to send a RST, and RST will be the final packet 725 * for the exchange. 726 * 727 * If we do not forbid deprecated addresses, we accept 728 * the SYN packet. RFC2462 does not suggest dropping 729 * SYN in this case. 730 * If we decipher RFC2462 5.5.4, it says like this: 731 * 1. use of deprecated addr with existing 732 * communication is okay - "SHOULD continue to be 733 * used" 734 * 2. use of it with new communication: 735 * (2a) "SHOULD NOT be used if alternate address 736 * with sufficient scope is available" 737 * (2b) nothing mentioned otherwise. 738 * Here we fall into (2b) case as we have no choice in 739 * our source address selection - we must obey the peer. 740 * 741 * The wording in RFC2462 is confusing, and there are 742 * multiple description text for deprecated address 743 * handling - worse, they are not exactly the same. 744 * I believe 5.5.4 is the best one, so we follow 5.5.4. 745 */ 746 if (isipv6 && !ip6_use_deprecated) { 747 struct in6_ifaddr *ia6; 748 749 if ((ia6 = ip6_getdstifaddr(m)) && 750 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 751 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 752 log(LOG_DEBUG, "%s; %s: Listen socket: " 753 "Connection attempt to deprecated " 754 "IPv6 address rejected\n", 755 s, __func__); 756 rstreason = BANDLIM_RST_OPENPORT; 757 goto dropwithreset; 758 } 759 } 760#endif 761 /* 762 * Basic sanity checks on incoming SYN requests: 763 * Don't respond if the destination is a link layer 764 * broadcast according to RFC1122 4.2.3.10, p. 104. 765 * If it is from this socket it must be forged. 766 * Don't respond if the source or destination is a 767 * global or subnet broad- or multicast address. 768 * Note that it is quite possible to receive unicast 769 * link-layer packets with a broadcast IP address. Use 770 * in_broadcast() to find them. 771 */ 772 if (m->m_flags & (M_BCAST|M_MCAST)) { 773 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 774 log(LOG_DEBUG, "%s; %s: Listen socket: " 775 "Connection attempt from broad- or multicast " 776 "link layer address ignored\n", s, __func__); 777 goto dropunlock; 778 } 779 if (isipv6) { 780#ifdef INET6 781 if (th->th_dport == th->th_sport && 782 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 783 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 784 log(LOG_DEBUG, "%s; %s: Listen socket: " 785 "Connection attempt to/from self " 786 "ignored\n", s, __func__); 787 goto dropunlock; 788 } 789 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 790 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 791 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 792 log(LOG_DEBUG, "%s; %s: Listen socket: " 793 "Connection attempt from/to multicast " 794 "address ignored\n", s, __func__); 795 goto dropunlock; 796 } 797#endif 798 } else { 799 if (th->th_dport == th->th_sport && 800 ip->ip_dst.s_addr == ip->ip_src.s_addr) { 801 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 802 log(LOG_DEBUG, "%s; %s: Listen socket: " 803 "Connection attempt from/to self " 804 "ignored\n", s, __func__); 805 goto dropunlock; 806 } 807 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 808 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 809 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 810 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 811 if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 812 log(LOG_DEBUG, "%s; %s: Listen socket: " 813 "Connection attempt from/to broad- " 814 "or multicast address ignored\n", 815 s, __func__); 816 goto dropunlock; 817 } 818 } 819 /* 820 * SYN appears to be valid. Create compressed TCP state 821 * for syncache. 822 */ 823#ifdef TCPDEBUG 824 if (so->so_options & SO_DEBUG) 825 tcp_trace(TA_INPUT, ostate, tp, 826 (void *)tcp_saveipgen, &tcp_savetcp, 0); 827#endif 828 tcp_dooptions(&to, optp, optlen, TO_SYN); 829 syncache_add(&inc, &to, th, inp, &so, m); 830 /* 831 * Entry added to syncache and mbuf consumed. 832 * Everything already unlocked by syncache_add(). 833 */ 834 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 835 return; 836 } 837 838 /* 839 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 840 * state. tcp_do_segment() always consumes the mbuf chain, unlocks 841 * the inpcb, and unlocks pcbinfo. 842 */ 843 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen); 844 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 845 return; 846 847dropwithreset: 848 INP_INFO_WLOCK_ASSERT(&tcbinfo); 849 tcp_dropwithreset(m, th, tp, tlen, rstreason); 850 m = NULL; /* mbuf chain got consumed. */ 851dropunlock: 852 INP_INFO_WLOCK_ASSERT(&tcbinfo); 853 if (inp != NULL) 854 INP_UNLOCK(inp); 855 INP_INFO_WUNLOCK(&tcbinfo); 856drop: 857 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 858 if (s != NULL) 859 free(s, M_TCPLOG); 860 if (m != NULL) 861 m_freem(m); 862 return; 863} 864 865static void 866tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 867 struct tcpcb *tp, int drop_hdrlen, int tlen) 868{ 869 int thflags, acked, ourfinisacked, needoutput = 0; 870 int headlocked = 1; 871 int rstreason, todrop, win; 872 u_long tiwin; 873 struct tcpopt to; 874 875#ifdef TCPDEBUG 876 /* 877 * The size of tcp_saveipgen must be the size of the max ip header, 878 * now IPv6. 879 */ 880 u_char tcp_saveipgen[IP6_HDR_LEN]; 881 struct tcphdr tcp_savetcp; 882 short ostate = 0; 883#endif 884 thflags = th->th_flags; 885 886 INP_INFO_WLOCK_ASSERT(&tcbinfo); 887 INP_LOCK_ASSERT(tp->t_inpcb); 888 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 889 __func__)); 890 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 891 __func__)); 892 893 /* 894 * Segment received on connection. 895 * Reset idle time and keep-alive timer. 896 * XXX: This should be done after segment 897 * validation to ignore broken/spoofed segs. 898 */ 899 tp->t_rcvtime = ticks; 900 if (TCPS_HAVEESTABLISHED(tp->t_state)) 901 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 902 903 /* 904 * Unscale the window into a 32-bit value. 905 * For the SYN_SENT state the scale is zero. 906 */ 907 tiwin = th->th_win << tp->snd_scale; 908 909 /* 910 * Parse options on any incoming segment. 911 */ 912 tcp_dooptions(&to, (u_char *)(th + 1), 913 (th->th_off << 2) - sizeof(struct tcphdr), 914 (thflags & TH_SYN) ? TO_SYN : 0); 915 916 /* 917 * If echoed timestamp is later than the current time, 918 * fall back to non RFC1323 RTT calculation. Normalize 919 * timestamp if syncookies were used when this connection 920 * was established. 921 */ 922 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 923 to.to_tsecr -= tp->ts_offset; 924 if (TSTMP_GT(to.to_tsecr, ticks)) 925 to.to_tsecr = 0; 926 } 927 928 /* 929 * Process options only when we get SYN/ACK back. The SYN case 930 * for incoming connections is handled in tcp_syncache. 931 * According to RFC1323 the window field in a SYN (i.e., a <SYN> 932 * or <SYN,ACK>) segment itself is never scaled. 933 * XXX this is traditional behavior, may need to be cleaned up. 934 */ 935 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 936 if ((to.to_flags & TOF_SCALE) && 937 (tp->t_flags & TF_REQ_SCALE)) { 938 tp->t_flags |= TF_RCVD_SCALE; 939 tp->snd_scale = to.to_wscale; 940 } 941 /* 942 * Initial send window. It will be updated with 943 * the next incoming segment to the scaled value. 944 */ 945 tp->snd_wnd = th->th_win; 946 if (to.to_flags & TOF_TS) { 947 tp->t_flags |= TF_RCVD_TSTMP; 948 tp->ts_recent = to.to_tsval; 949 tp->ts_recent_age = ticks; 950 } 951 if (to.to_flags & TOF_MSS) 952 tcp_mss(tp, to.to_mss); 953 if ((tp->t_flags & TF_SACK_PERMIT) && 954 (to.to_flags & TOF_SACKPERM) == 0) 955 tp->t_flags &= ~TF_SACK_PERMIT; 956 } 957 958 /* 959 * Header prediction: check for the two common cases 960 * of a uni-directional data xfer. If the packet has 961 * no control flags, is in-sequence, the window didn't 962 * change and we're not retransmitting, it's a 963 * candidate. If the length is zero and the ack moved 964 * forward, we're the sender side of the xfer. Just 965 * free the data acked & wake any higher level process 966 * that was blocked waiting for space. If the length 967 * is non-zero and the ack didn't move, we're the 968 * receiver side. If we're getting packets in-order 969 * (the reassembly queue is empty), add the data to 970 * the socket buffer and note that we need a delayed ack. 971 * Make sure that the hidden state-flags are also off. 972 * Since we check for TCPS_ESTABLISHED first, it can only 973 * be TH_NEEDSYN. 974 */ 975 if (tp->t_state == TCPS_ESTABLISHED && 976 th->th_seq == tp->rcv_nxt && 977 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 978 tp->snd_nxt == tp->snd_max && 979 tiwin && tiwin == tp->snd_wnd && 980 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 981 LIST_EMPTY(&tp->t_segq) && 982 ((to.to_flags & TOF_TS) == 0 || 983 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 984 985 /* 986 * If last ACK falls within this segment's sequence numbers, 987 * record the timestamp. 988 * NOTE that the test is modified according to the latest 989 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 990 */ 991 if ((to.to_flags & TOF_TS) != 0 && 992 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 993 tp->ts_recent_age = ticks; 994 tp->ts_recent = to.to_tsval; 995 } 996 997 if (tlen == 0) { 998 if (SEQ_GT(th->th_ack, tp->snd_una) && 999 SEQ_LEQ(th->th_ack, tp->snd_max) && 1000 tp->snd_cwnd >= tp->snd_wnd && 1001 ((!tcp_do_newreno && 1002 !(tp->t_flags & TF_SACK_PERMIT) && 1003 tp->t_dupacks < tcprexmtthresh) || 1004 ((tcp_do_newreno || 1005 (tp->t_flags & TF_SACK_PERMIT)) && 1006 !IN_FASTRECOVERY(tp) && 1007 (to.to_flags & TOF_SACK) == 0 && 1008 TAILQ_EMPTY(&tp->snd_holes)))) { 1009 KASSERT(headlocked, 1010 ("%s: headlocked", __func__)); 1011 INP_INFO_WUNLOCK(&tcbinfo); 1012 headlocked = 0; 1013 /* 1014 * This is a pure ack for outstanding data. 1015 */ 1016 ++tcpstat.tcps_predack; 1017 /* 1018 * "bad retransmit" recovery. 1019 */ 1020 if (tp->t_rxtshift == 1 && 1021 ticks < tp->t_badrxtwin) { 1022 ++tcpstat.tcps_sndrexmitbad; 1023 tp->snd_cwnd = tp->snd_cwnd_prev; 1024 tp->snd_ssthresh = 1025 tp->snd_ssthresh_prev; 1026 tp->snd_recover = tp->snd_recover_prev; 1027 if (tp->t_flags & TF_WASFRECOVERY) 1028 ENTER_FASTRECOVERY(tp); 1029 tp->snd_nxt = tp->snd_max; 1030 tp->t_badrxtwin = 0; 1031 } 1032 1033 /* 1034 * Recalculate the transmit timer / rtt. 1035 * 1036 * Some boxes send broken timestamp replies 1037 * during the SYN+ACK phase, ignore 1038 * timestamps of 0 or we could calculate a 1039 * huge RTT and blow up the retransmit timer. 1040 */ 1041 if ((to.to_flags & TOF_TS) != 0 && 1042 to.to_tsecr) { 1043 if (!tp->t_rttlow || 1044 tp->t_rttlow > ticks - to.to_tsecr) 1045 tp->t_rttlow = ticks - to.to_tsecr; 1046 tcp_xmit_timer(tp, 1047 ticks - to.to_tsecr + 1); 1048 } else if (tp->t_rtttime && 1049 SEQ_GT(th->th_ack, tp->t_rtseq)) { 1050 if (!tp->t_rttlow || 1051 tp->t_rttlow > ticks - tp->t_rtttime) 1052 tp->t_rttlow = ticks - tp->t_rtttime; 1053 tcp_xmit_timer(tp, 1054 ticks - tp->t_rtttime); 1055 } 1056 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1057 acked = th->th_ack - tp->snd_una; 1058 tcpstat.tcps_rcvackpack++; 1059 tcpstat.tcps_rcvackbyte += acked; 1060 sbdrop(&so->so_snd, acked); 1061 if (SEQ_GT(tp->snd_una, tp->snd_recover) && 1062 SEQ_LEQ(th->th_ack, tp->snd_recover)) 1063 tp->snd_recover = th->th_ack - 1; 1064 tp->snd_una = th->th_ack; 1065 /* 1066 * Pull snd_wl2 up to prevent seq wrap relative 1067 * to th_ack. 1068 */ 1069 tp->snd_wl2 = th->th_ack; 1070 tp->t_dupacks = 0; 1071 m_freem(m); 1072 ND6_HINT(tp); /* Some progress has been made. */ 1073 1074 /* 1075 * If all outstanding data are acked, stop 1076 * retransmit timer, otherwise restart timer 1077 * using current (possibly backed-off) value. 1078 * If process is waiting for space, 1079 * wakeup/selwakeup/signal. If data 1080 * are ready to send, let tcp_output 1081 * decide between more output or persist. 1082 */ 1083#ifdef TCPDEBUG 1084 if (so->so_options & SO_DEBUG) 1085 tcp_trace(TA_INPUT, ostate, tp, 1086 (void *)tcp_saveipgen, 1087 &tcp_savetcp, 0); 1088#endif 1089 if (tp->snd_una == tp->snd_max) 1090 tcp_timer_activate(tp, TT_REXMT, 0); 1091 else if (!tcp_timer_active(tp, TT_PERSIST)) 1092 tcp_timer_activate(tp, TT_REXMT, 1093 tp->t_rxtcur); 1094 sowwakeup(so); 1095 if (so->so_snd.sb_cc) 1096 (void) tcp_output(tp); 1097 goto check_delack; 1098 } 1099 } else if (th->th_ack == tp->snd_una && 1100 tlen <= sbspace(&so->so_rcv)) { 1101 int newsize = 0; /* automatic sockbuf scaling */ 1102 1103 KASSERT(headlocked, ("%s: headlocked", __func__)); 1104 INP_INFO_WUNLOCK(&tcbinfo); 1105 headlocked = 0; 1106 /* 1107 * This is a pure, in-sequence data packet 1108 * with nothing on the reassembly queue and 1109 * we have enough buffer space to take it. 1110 */ 1111 /* Clean receiver SACK report if present */ 1112 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 1113 tcp_clean_sackreport(tp); 1114 ++tcpstat.tcps_preddat; 1115 tp->rcv_nxt += tlen; 1116 /* 1117 * Pull snd_wl1 up to prevent seq wrap relative to 1118 * th_seq. 1119 */ 1120 tp->snd_wl1 = th->th_seq; 1121 /* 1122 * Pull rcv_up up to prevent seq wrap relative to 1123 * rcv_nxt. 1124 */ 1125 tp->rcv_up = tp->rcv_nxt; 1126 tcpstat.tcps_rcvpack++; 1127 tcpstat.tcps_rcvbyte += tlen; 1128 ND6_HINT(tp); /* Some progress has been made */ 1129#ifdef TCPDEBUG 1130 if (so->so_options & SO_DEBUG) 1131 tcp_trace(TA_INPUT, ostate, tp, 1132 (void *)tcp_saveipgen, &tcp_savetcp, 0); 1133#endif 1134 /* 1135 * Automatic sizing of receive socket buffer. Often the send 1136 * buffer size is not optimally adjusted to the actual network 1137 * conditions at hand (delay bandwidth product). Setting the 1138 * buffer size too small limits throughput on links with high 1139 * bandwidth and high delay (eg. trans-continental/oceanic links). 1140 * 1141 * On the receive side the socket buffer memory is only rarely 1142 * used to any significant extent. This allows us to be much 1143 * more aggressive in scaling the receive socket buffer. For 1144 * the case that the buffer space is actually used to a large 1145 * extent and we run out of kernel memory we can simply drop 1146 * the new segments; TCP on the sender will just retransmit it 1147 * later. Setting the buffer size too big may only consume too 1148 * much kernel memory if the application doesn't read() from 1149 * the socket or packet loss or reordering makes use of the 1150 * reassembly queue. 1151 * 1152 * The criteria to step up the receive buffer one notch are: 1153 * 1. the number of bytes received during the time it takes 1154 * one timestamp to be reflected back to us (the RTT); 1155 * 2. received bytes per RTT is within seven eighth of the 1156 * current socket buffer size; 1157 * 3. receive buffer size has not hit maximal automatic size; 1158 * 1159 * This algorithm does one step per RTT at most and only if 1160 * we receive a bulk stream w/o packet losses or reorderings. 1161 * Shrinking the buffer during idle times is not necessary as 1162 * it doesn't consume any memory when idle. 1163 * 1164 * TODO: Only step up if the application is actually serving 1165 * the buffer to better manage the socket buffer resources. 1166 */ 1167 if (tcp_do_autorcvbuf && 1168 to.to_tsecr && 1169 (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 1170 if (to.to_tsecr > tp->rfbuf_ts && 1171 to.to_tsecr - tp->rfbuf_ts < hz) { 1172 if (tp->rfbuf_cnt > 1173 (so->so_rcv.sb_hiwat / 8 * 7) && 1174 so->so_rcv.sb_hiwat < 1175 tcp_autorcvbuf_max) { 1176 newsize = 1177 min(so->so_rcv.sb_hiwat + 1178 tcp_autorcvbuf_inc, 1179 tcp_autorcvbuf_max); 1180 } 1181 /* Start over with next RTT. */ 1182 tp->rfbuf_ts = 0; 1183 tp->rfbuf_cnt = 0; 1184 } else 1185 tp->rfbuf_cnt += tlen; /* add up */ 1186 } 1187 1188 /* Add data to socket buffer. */ 1189 SOCKBUF_LOCK(&so->so_rcv); 1190 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1191 m_freem(m); 1192 } else { 1193 /* 1194 * Set new socket buffer size. 1195 * Give up when limit is reached. 1196 */ 1197 if (newsize) 1198 if (!sbreserve_locked(&so->so_rcv, 1199 newsize, so, curthread)) 1200 so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1201 m_adj(m, drop_hdrlen); /* delayed header drop */ 1202 sbappendstream_locked(&so->so_rcv, m); 1203 } 1204 /* NB: sorwakeup_locked() does an implicit unlock. */ 1205 sorwakeup_locked(so); 1206 if (DELAY_ACK(tp)) { 1207 tp->t_flags |= TF_DELACK; 1208 } else { 1209 tp->t_flags |= TF_ACKNOW; 1210 tcp_output(tp); 1211 } 1212 goto check_delack; 1213 } 1214 } 1215 1216 /* 1217 * Calculate amount of space in receive window, 1218 * and then do TCP input processing. 1219 * Receive window is amount of space in rcv queue, 1220 * but not less than advertised window. 1221 */ 1222 win = sbspace(&so->so_rcv); 1223 if (win < 0) 1224 win = 0; 1225 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1226 1227 /* Reset receive buffer auto scaling when not in bulk receive mode. */ 1228 tp->rfbuf_ts = 0; 1229 tp->rfbuf_cnt = 0; 1230 1231 switch (tp->t_state) { 1232 1233 /* 1234 * If the state is SYN_RECEIVED: 1235 * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1236 */ 1237 case TCPS_SYN_RECEIVED: 1238 if ((thflags & TH_ACK) && 1239 (SEQ_LEQ(th->th_ack, tp->snd_una) || 1240 SEQ_GT(th->th_ack, tp->snd_max))) { 1241 rstreason = BANDLIM_RST_OPENPORT; 1242 goto dropwithreset; 1243 } 1244 break; 1245 1246 /* 1247 * If the state is SYN_SENT: 1248 * if seg contains an ACK, but not for our SYN, drop the input. 1249 * if seg contains a RST, then drop the connection. 1250 * if seg does not contain SYN, then drop it. 1251 * Otherwise this is an acceptable SYN segment 1252 * initialize tp->rcv_nxt and tp->irs 1253 * if seg contains ack then advance tp->snd_una 1254 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1255 * arrange for segment to be acked (eventually) 1256 * continue processing rest of data/controls, beginning with URG 1257 */ 1258 case TCPS_SYN_SENT: 1259 if ((thflags & TH_ACK) && 1260 (SEQ_LEQ(th->th_ack, tp->iss) || 1261 SEQ_GT(th->th_ack, tp->snd_max))) { 1262 rstreason = BANDLIM_UNLIMITED; 1263 goto dropwithreset; 1264 } 1265 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) 1266 tp = tcp_drop(tp, ECONNREFUSED); 1267 if (thflags & TH_RST) 1268 goto drop; 1269 if (!(thflags & TH_SYN)) 1270 goto drop; 1271 1272 tp->irs = th->th_seq; 1273 tcp_rcvseqinit(tp); 1274 if (thflags & TH_ACK) { 1275 tcpstat.tcps_connects++; 1276 soisconnected(so); 1277#ifdef MAC 1278 SOCK_LOCK(so); 1279 mac_set_socket_peer_from_mbuf(m, so); 1280 SOCK_UNLOCK(so); 1281#endif 1282 /* Do window scaling on this connection? */ 1283 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1284 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1285 tp->rcv_scale = tp->request_r_scale; 1286 } 1287 tp->rcv_adv += tp->rcv_wnd; 1288 tp->snd_una++; /* SYN is acked */ 1289 /* 1290 * If there's data, delay ACK; if there's also a FIN 1291 * ACKNOW will be turned on later. 1292 */ 1293 if (DELAY_ACK(tp) && tlen != 0) 1294 tcp_timer_activate(tp, TT_DELACK, 1295 tcp_delacktime); 1296 else 1297 tp->t_flags |= TF_ACKNOW; 1298 /* 1299 * Received <SYN,ACK> in SYN_SENT[*] state. 1300 * Transitions: 1301 * SYN_SENT --> ESTABLISHED 1302 * SYN_SENT* --> FIN_WAIT_1 1303 */ 1304 tp->t_starttime = ticks; 1305 if (tp->t_flags & TF_NEEDFIN) { 1306 tp->t_state = TCPS_FIN_WAIT_1; 1307 tp->t_flags &= ~TF_NEEDFIN; 1308 thflags &= ~TH_SYN; 1309 } else { 1310 tp->t_state = TCPS_ESTABLISHED; 1311 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1312 } 1313 } else { 1314 /* 1315 * Received initial SYN in SYN-SENT[*] state => 1316 * simultaneous open. If segment contains CC option 1317 * and there is a cached CC, apply TAO test. 1318 * If it succeeds, connection is * half-synchronized. 1319 * Otherwise, do 3-way handshake: 1320 * SYN-SENT -> SYN-RECEIVED 1321 * SYN-SENT* -> SYN-RECEIVED* 1322 * If there was no CC option, clear cached CC value. 1323 */ 1324 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1325 tcp_timer_activate(tp, TT_REXMT, 0); 1326 tp->t_state = TCPS_SYN_RECEIVED; 1327 } 1328 1329 KASSERT(headlocked, ("%s: trimthenstep6: head not locked", 1330 __func__)); 1331 INP_LOCK_ASSERT(tp->t_inpcb); 1332 1333 /* 1334 * Advance th->th_seq to correspond to first data byte. 1335 * If data, trim to stay within window, 1336 * dropping FIN if necessary. 1337 */ 1338 th->th_seq++; 1339 if (tlen > tp->rcv_wnd) { 1340 todrop = tlen - tp->rcv_wnd; 1341 m_adj(m, -todrop); 1342 tlen = tp->rcv_wnd; 1343 thflags &= ~TH_FIN; 1344 tcpstat.tcps_rcvpackafterwin++; 1345 tcpstat.tcps_rcvbyteafterwin += todrop; 1346 } 1347 tp->snd_wl1 = th->th_seq - 1; 1348 tp->rcv_up = th->th_seq; 1349 /* 1350 * Client side of transaction: already sent SYN and data. 1351 * If the remote host used T/TCP to validate the SYN, 1352 * our data will be ACK'd; if so, enter normal data segment 1353 * processing in the middle of step 5, ack processing. 1354 * Otherwise, goto step 6. 1355 */ 1356 if (thflags & TH_ACK) 1357 goto process_ACK; 1358 1359 goto step6; 1360 1361 /* 1362 * If the state is LAST_ACK or CLOSING or TIME_WAIT: 1363 * do normal processing. 1364 * 1365 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 1366 */ 1367 case TCPS_LAST_ACK: 1368 case TCPS_CLOSING: 1369 break; /* continue normal processing */ 1370 } 1371 1372 /* 1373 * States other than LISTEN or SYN_SENT. 1374 * First check the RST flag and sequence number since reset segments 1375 * are exempt from the timestamp and connection count tests. This 1376 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 1377 * below which allowed reset segments in half the sequence space 1378 * to fall though and be processed (which gives forged reset 1379 * segments with a random sequence number a 50 percent chance of 1380 * killing a connection). 1381 * Then check timestamp, if present. 1382 * Then check the connection count, if present. 1383 * Then check that at least some bytes of segment are within 1384 * receive window. If segment begins before rcv_nxt, 1385 * drop leading data (and SYN); if nothing left, just ack. 1386 * 1387 * 1388 * If the RST bit is set, check the sequence number to see 1389 * if this is a valid reset segment. 1390 * RFC 793 page 37: 1391 * In all states except SYN-SENT, all reset (RST) segments 1392 * are validated by checking their SEQ-fields. A reset is 1393 * valid if its sequence number is in the window. 1394 * Note: this does not take into account delayed ACKs, so 1395 * we should test against last_ack_sent instead of rcv_nxt. 1396 * The sequence number in the reset segment is normally an 1397 * echo of our outgoing acknowlegement numbers, but some hosts 1398 * send a reset with the sequence number at the rightmost edge 1399 * of our receive window, and we have to handle this case. 1400 * Note 2: Paul Watson's paper "Slipping in the Window" has shown 1401 * that brute force RST attacks are possible. To combat this, 1402 * we use a much stricter check while in the ESTABLISHED state, 1403 * only accepting RSTs where the sequence number is equal to 1404 * last_ack_sent. In all other states (the states in which a 1405 * RST is more likely), the more permissive check is used. 1406 * If we have multiple segments in flight, the intial reset 1407 * segment sequence numbers will be to the left of last_ack_sent, 1408 * but they will eventually catch up. 1409 * In any case, it never made sense to trim reset segments to 1410 * fit the receive window since RFC 1122 says: 1411 * 4.2.2.12 RST Segment: RFC-793 Section 3.4 1412 * 1413 * A TCP SHOULD allow a received RST segment to include data. 1414 * 1415 * DISCUSSION 1416 * It has been suggested that a RST segment could contain 1417 * ASCII text that encoded and explained the cause of the 1418 * RST. No standard has yet been established for such 1419 * data. 1420 * 1421 * If the reset segment passes the sequence number test examine 1422 * the state: 1423 * SYN_RECEIVED STATE: 1424 * If passive open, return to LISTEN state. 1425 * If active open, inform user that connection was refused. 1426 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 1427 * Inform user that connection was reset, and close tcb. 1428 * CLOSING, LAST_ACK STATES: 1429 * Close the tcb. 1430 * TIME_WAIT STATE: 1431 * Drop the segment - see Stevens, vol. 2, p. 964 and 1432 * RFC 1337. 1433 */ 1434 if (thflags & TH_RST) { 1435 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1436 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 1437 switch (tp->t_state) { 1438 1439 case TCPS_SYN_RECEIVED: 1440 so->so_error = ECONNREFUSED; 1441 goto close; 1442 1443 case TCPS_ESTABLISHED: 1444 if (tcp_insecure_rst == 0 && 1445 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 1446 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 1447 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 1448 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 1449 tcpstat.tcps_badrst++; 1450 goto drop; 1451 } 1452 /* FALLTHROUGH */ 1453 case TCPS_FIN_WAIT_1: 1454 case TCPS_FIN_WAIT_2: 1455 case TCPS_CLOSE_WAIT: 1456 so->so_error = ECONNRESET; 1457 close: 1458 tp->t_state = TCPS_CLOSED; 1459 tcpstat.tcps_drops++; 1460 KASSERT(headlocked, ("%s: trimthenstep6: " 1461 "tcp_close: head not locked", __func__)); 1462 tp = tcp_close(tp); 1463 break; 1464 1465 case TCPS_CLOSING: 1466 case TCPS_LAST_ACK: 1467 KASSERT(headlocked, ("%s: trimthenstep6: " 1468 "tcp_close.2: head not locked", __func__)); 1469 tp = tcp_close(tp); 1470 break; 1471 } 1472 } 1473 goto drop; 1474 } 1475 1476 /* 1477 * RFC 1323 PAWS: If we have a timestamp reply on this segment 1478 * and it's less than ts_recent, drop it. 1479 */ 1480 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 1481 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 1482 1483 /* Check to see if ts_recent is over 24 days old. */ 1484 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) { 1485 /* 1486 * Invalidate ts_recent. If this segment updates 1487 * ts_recent, the age will be reset later and ts_recent 1488 * will get a valid value. If it does not, setting 1489 * ts_recent to zero will at least satisfy the 1490 * requirement that zero be placed in the timestamp 1491 * echo reply when ts_recent isn't valid. The 1492 * age isn't reset until we get a valid ts_recent 1493 * because we don't want out-of-order segments to be 1494 * dropped when ts_recent is old. 1495 */ 1496 tp->ts_recent = 0; 1497 } else { 1498 tcpstat.tcps_rcvduppack++; 1499 tcpstat.tcps_rcvdupbyte += tlen; 1500 tcpstat.tcps_pawsdrop++; 1501 if (tlen) 1502 goto dropafterack; 1503 goto drop; 1504 } 1505 } 1506 1507 /* 1508 * In the SYN-RECEIVED state, validate that the packet belongs to 1509 * this connection before trimming the data to fit the receive 1510 * window. Check the sequence number versus IRS since we know 1511 * the sequence numbers haven't wrapped. This is a partial fix 1512 * for the "LAND" DoS attack. 1513 */ 1514 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 1515 rstreason = BANDLIM_RST_OPENPORT; 1516 goto dropwithreset; 1517 } 1518 1519 todrop = tp->rcv_nxt - th->th_seq; 1520 if (todrop > 0) { 1521 if (thflags & TH_SYN) { 1522 thflags &= ~TH_SYN; 1523 th->th_seq++; 1524 if (th->th_urp > 1) 1525 th->th_urp--; 1526 else 1527 thflags &= ~TH_URG; 1528 todrop--; 1529 } 1530 /* 1531 * Following if statement from Stevens, vol. 2, p. 960. 1532 */ 1533 if (todrop > tlen 1534 || (todrop == tlen && (thflags & TH_FIN) == 0)) { 1535 /* 1536 * Any valid FIN must be to the left of the window. 1537 * At this point the FIN must be a duplicate or out 1538 * of sequence; drop it. 1539 */ 1540 thflags &= ~TH_FIN; 1541 1542 /* 1543 * Send an ACK to resynchronize and drop any data. 1544 * But keep on processing for RST or ACK. 1545 */ 1546 tp->t_flags |= TF_ACKNOW; 1547 todrop = tlen; 1548 tcpstat.tcps_rcvduppack++; 1549 tcpstat.tcps_rcvdupbyte += todrop; 1550 } else { 1551 tcpstat.tcps_rcvpartduppack++; 1552 tcpstat.tcps_rcvpartdupbyte += todrop; 1553 } 1554 drop_hdrlen += todrop; /* drop from the top afterwards */ 1555 th->th_seq += todrop; 1556 tlen -= todrop; 1557 if (th->th_urp > todrop) 1558 th->th_urp -= todrop; 1559 else { 1560 thflags &= ~TH_URG; 1561 th->th_urp = 0; 1562 } 1563 } 1564 1565 /* 1566 * If new data are received on a connection after the 1567 * user processes are gone, then RST the other end. 1568 */ 1569 if ((so->so_state & SS_NOFDREF) && 1570 tp->t_state > TCPS_CLOSE_WAIT && tlen) { 1571 char *s; 1572 1573 KASSERT(headlocked, ("%s: trimthenstep6: tcp_close.3: head " 1574 "not locked", __func__)); 1575 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) { 1576#ifdef TCPDEBUG 1577 log(LOG_DEBUG, "%s; %s: %s: Received data after socket " 1578 "was closed, sending RST and removing tcpcb\n", 1579 s, __func__, tcpstates[tp->t_state]); 1580#else 1581 log(LOG_DEBUG, "%s; %s: Received data after socket " 1582 "was closed, sending RST and removing tcpcb\n", 1583 s, __func__); 1584#endif 1585 free(s, M_TCPLOG); 1586 } 1587 tp = tcp_close(tp); 1588 tcpstat.tcps_rcvafterclose++; 1589 rstreason = BANDLIM_UNLIMITED; 1590 goto dropwithreset; 1591 } 1592 1593 /* 1594 * If segment ends after window, drop trailing data 1595 * (and PUSH and FIN); if nothing left, just ACK. 1596 */ 1597 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 1598 if (todrop > 0) { 1599 tcpstat.tcps_rcvpackafterwin++; 1600 if (todrop >= tlen) { 1601 tcpstat.tcps_rcvbyteafterwin += tlen; 1602 /* 1603 * If window is closed can only take segments at 1604 * window edge, and have to drop data and PUSH from 1605 * incoming segments. Continue processing, but 1606 * remember to ack. Otherwise, drop segment 1607 * and ack. 1608 */ 1609 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 1610 tp->t_flags |= TF_ACKNOW; 1611 tcpstat.tcps_rcvwinprobe++; 1612 } else 1613 goto dropafterack; 1614 } else 1615 tcpstat.tcps_rcvbyteafterwin += todrop; 1616 m_adj(m, -todrop); 1617 tlen -= todrop; 1618 thflags &= ~(TH_PUSH|TH_FIN); 1619 } 1620 1621 /* 1622 * If last ACK falls within this segment's sequence numbers, 1623 * record its timestamp. 1624 * NOTE: 1625 * 1) That the test incorporates suggestions from the latest 1626 * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1627 * 2) That updating only on newer timestamps interferes with 1628 * our earlier PAWS tests, so this check should be solely 1629 * predicated on the sequence space of this segment. 1630 * 3) That we modify the segment boundary check to be 1631 * Last.ACK.Sent <= SEG.SEQ + SEG.Len 1632 * instead of RFC1323's 1633 * Last.ACK.Sent < SEG.SEQ + SEG.Len, 1634 * This modified check allows us to overcome RFC1323's 1635 * limitations as described in Stevens TCP/IP Illustrated 1636 * Vol. 2 p.869. In such cases, we can still calculate the 1637 * RTT correctly when RCV.NXT == Last.ACK.Sent. 1638 */ 1639 if ((to.to_flags & TOF_TS) != 0 && 1640 SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 1641 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 1642 ((thflags & (TH_SYN|TH_FIN)) != 0))) { 1643 tp->ts_recent_age = ticks; 1644 tp->ts_recent = to.to_tsval; 1645 } 1646 1647 /* 1648 * If a SYN is in the window, then this is an 1649 * error and we send an RST and drop the connection. 1650 */ 1651 if (thflags & TH_SYN) { 1652 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: " 1653 "head not locked", __func__)); 1654 tp = tcp_drop(tp, ECONNRESET); 1655 rstreason = BANDLIM_UNLIMITED; 1656 goto drop; 1657 } 1658 1659 /* 1660 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 1661 * flag is on (half-synchronized state), then queue data for 1662 * later processing; else drop segment and return. 1663 */ 1664 if ((thflags & TH_ACK) == 0) { 1665 if (tp->t_state == TCPS_SYN_RECEIVED || 1666 (tp->t_flags & TF_NEEDSYN)) 1667 goto step6; 1668 else if (tp->t_flags & TF_ACKNOW) 1669 goto dropafterack; 1670 else 1671 goto drop; 1672 } 1673 1674 /* 1675 * Ack processing. 1676 */ 1677 switch (tp->t_state) { 1678 1679 /* 1680 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 1681 * ESTABLISHED state and continue processing. 1682 * The ACK was checked above. 1683 */ 1684 case TCPS_SYN_RECEIVED: 1685 1686 tcpstat.tcps_connects++; 1687 soisconnected(so); 1688 /* Do window scaling? */ 1689 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1690 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1691 tp->rcv_scale = tp->request_r_scale; 1692 tp->snd_wnd = tiwin; 1693 } 1694 /* 1695 * Make transitions: 1696 * SYN-RECEIVED -> ESTABLISHED 1697 * SYN-RECEIVED* -> FIN-WAIT-1 1698 */ 1699 tp->t_starttime = ticks; 1700 if (tp->t_flags & TF_NEEDFIN) { 1701 tp->t_state = TCPS_FIN_WAIT_1; 1702 tp->t_flags &= ~TF_NEEDFIN; 1703 } else { 1704 tp->t_state = TCPS_ESTABLISHED; 1705 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle); 1706 } 1707 /* 1708 * If segment contains data or ACK, will call tcp_reass() 1709 * later; if not, do so now to pass queued data to user. 1710 */ 1711 if (tlen == 0 && (thflags & TH_FIN) == 0) 1712 (void) tcp_reass(tp, (struct tcphdr *)0, 0, 1713 (struct mbuf *)0); 1714 tp->snd_wl1 = th->th_seq - 1; 1715 /* FALLTHROUGH */ 1716 1717 /* 1718 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 1719 * ACKs. If the ack is in the range 1720 * tp->snd_una < th->th_ack <= tp->snd_max 1721 * then advance tp->snd_una to th->th_ack and drop 1722 * data from the retransmission queue. If this ACK reflects 1723 * more up to date window information we update our window information. 1724 */ 1725 case TCPS_ESTABLISHED: 1726 case TCPS_FIN_WAIT_1: 1727 case TCPS_FIN_WAIT_2: 1728 case TCPS_CLOSE_WAIT: 1729 case TCPS_CLOSING: 1730 case TCPS_LAST_ACK: 1731 if (SEQ_GT(th->th_ack, tp->snd_max)) { 1732 tcpstat.tcps_rcvacktoomuch++; 1733 goto dropafterack; 1734 } 1735 if ((tp->t_flags & TF_SACK_PERMIT) && 1736 ((to.to_flags & TOF_SACK) || 1737 !TAILQ_EMPTY(&tp->snd_holes))) 1738 tcp_sack_doack(tp, &to, th->th_ack); 1739 if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 1740 if (tlen == 0 && tiwin == tp->snd_wnd) { 1741 tcpstat.tcps_rcvdupack++; 1742 /* 1743 * If we have outstanding data (other than 1744 * a window probe), this is a completely 1745 * duplicate ack (ie, window info didn't 1746 * change), the ack is the biggest we've 1747 * seen and we've seen exactly our rexmt 1748 * threshhold of them, assume a packet 1749 * has been dropped and retransmit it. 1750 * Kludge snd_nxt & the congestion 1751 * window so we send only this one 1752 * packet. 1753 * 1754 * We know we're losing at the current 1755 * window size so do congestion avoidance 1756 * (set ssthresh to half the current window 1757 * and pull our congestion window back to 1758 * the new ssthresh). 1759 * 1760 * Dup acks mean that packets have left the 1761 * network (they're now cached at the receiver) 1762 * so bump cwnd by the amount in the receiver 1763 * to keep a constant cwnd packets in the 1764 * network. 1765 */ 1766 if (!tcp_timer_active(tp, TT_REXMT) || 1767 th->th_ack != tp->snd_una) 1768 tp->t_dupacks = 0; 1769 else if (++tp->t_dupacks > tcprexmtthresh || 1770 ((tcp_do_newreno || 1771 (tp->t_flags & TF_SACK_PERMIT)) && 1772 IN_FASTRECOVERY(tp))) { 1773 if ((tp->t_flags & TF_SACK_PERMIT) && 1774 IN_FASTRECOVERY(tp)) { 1775 int awnd; 1776 1777 /* 1778 * Compute the amount of data in flight first. 1779 * We can inject new data into the pipe iff 1780 * we have less than 1/2 the original window's 1781 * worth of data in flight. 1782 */ 1783 awnd = (tp->snd_nxt - tp->snd_fack) + 1784 tp->sackhint.sack_bytes_rexmit; 1785 if (awnd < tp->snd_ssthresh) { 1786 tp->snd_cwnd += tp->t_maxseg; 1787 if (tp->snd_cwnd > tp->snd_ssthresh) 1788 tp->snd_cwnd = tp->snd_ssthresh; 1789 } 1790 } else 1791 tp->snd_cwnd += tp->t_maxseg; 1792 (void) tcp_output(tp); 1793 goto drop; 1794 } else if (tp->t_dupacks == tcprexmtthresh) { 1795 tcp_seq onxt = tp->snd_nxt; 1796 u_int win; 1797 1798 /* 1799 * If we're doing sack, check to 1800 * see if we're already in sack 1801 * recovery. If we're not doing sack, 1802 * check to see if we're in newreno 1803 * recovery. 1804 */ 1805 if (tp->t_flags & TF_SACK_PERMIT) { 1806 if (IN_FASTRECOVERY(tp)) { 1807 tp->t_dupacks = 0; 1808 break; 1809 } 1810 } else if (tcp_do_newreno) { 1811 if (SEQ_LEQ(th->th_ack, 1812 tp->snd_recover)) { 1813 tp->t_dupacks = 0; 1814 break; 1815 } 1816 } 1817 win = min(tp->snd_wnd, tp->snd_cwnd) / 1818 2 / tp->t_maxseg; 1819 if (win < 2) 1820 win = 2; 1821 tp->snd_ssthresh = win * tp->t_maxseg; 1822 ENTER_FASTRECOVERY(tp); 1823 tp->snd_recover = tp->snd_max; 1824 tcp_timer_activate(tp, TT_REXMT, 0); 1825 tp->t_rtttime = 0; 1826 if (tp->t_flags & TF_SACK_PERMIT) { 1827 tcpstat.tcps_sack_recovery_episode++; 1828 tp->sack_newdata = tp->snd_nxt; 1829 tp->snd_cwnd = tp->t_maxseg; 1830 (void) tcp_output(tp); 1831 goto drop; 1832 } 1833 tp->snd_nxt = th->th_ack; 1834 tp->snd_cwnd = tp->t_maxseg; 1835 (void) tcp_output(tp); 1836 KASSERT(tp->snd_limited <= 2, 1837 ("%s: tp->snd_limited too big", 1838 __func__)); 1839 tp->snd_cwnd = tp->snd_ssthresh + 1840 tp->t_maxseg * 1841 (tp->t_dupacks - tp->snd_limited); 1842 if (SEQ_GT(onxt, tp->snd_nxt)) 1843 tp->snd_nxt = onxt; 1844 goto drop; 1845 } else if (tcp_do_rfc3042) { 1846 u_long oldcwnd = tp->snd_cwnd; 1847 tcp_seq oldsndmax = tp->snd_max; 1848 u_int sent; 1849 1850 KASSERT(tp->t_dupacks == 1 || 1851 tp->t_dupacks == 2, 1852 ("%s: dupacks not 1 or 2", 1853 __func__)); 1854 if (tp->t_dupacks == 1) 1855 tp->snd_limited = 0; 1856 tp->snd_cwnd = 1857 (tp->snd_nxt - tp->snd_una) + 1858 (tp->t_dupacks - tp->snd_limited) * 1859 tp->t_maxseg; 1860 (void) tcp_output(tp); 1861 sent = tp->snd_max - oldsndmax; 1862 if (sent > tp->t_maxseg) { 1863 KASSERT((tp->t_dupacks == 2 && 1864 tp->snd_limited == 0) || 1865 (sent == tp->t_maxseg + 1 && 1866 tp->t_flags & TF_SENTFIN), 1867 ("%s: sent too much", 1868 __func__)); 1869 tp->snd_limited = 2; 1870 } else if (sent > 0) 1871 ++tp->snd_limited; 1872 tp->snd_cwnd = oldcwnd; 1873 goto drop; 1874 } 1875 } else 1876 tp->t_dupacks = 0; 1877 break; 1878 } 1879 1880 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 1881 ("%s: th_ack <= snd_una", __func__)); 1882 1883 /* 1884 * If the congestion window was inflated to account 1885 * for the other side's cached packets, retract it. 1886 */ 1887 if (tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) { 1888 if (IN_FASTRECOVERY(tp)) { 1889 if (SEQ_LT(th->th_ack, tp->snd_recover)) { 1890 if (tp->t_flags & TF_SACK_PERMIT) 1891 tcp_sack_partialack(tp, th); 1892 else 1893 tcp_newreno_partial_ack(tp, th); 1894 } else { 1895 /* 1896 * Out of fast recovery. 1897 * Window inflation should have left us 1898 * with approximately snd_ssthresh 1899 * outstanding data. 1900 * But in case we would be inclined to 1901 * send a burst, better to do it via 1902 * the slow start mechanism. 1903 */ 1904 if (SEQ_GT(th->th_ack + 1905 tp->snd_ssthresh, 1906 tp->snd_max)) 1907 tp->snd_cwnd = tp->snd_max - 1908 th->th_ack + 1909 tp->t_maxseg; 1910 else 1911 tp->snd_cwnd = tp->snd_ssthresh; 1912 } 1913 } 1914 } else { 1915 if (tp->t_dupacks >= tcprexmtthresh && 1916 tp->snd_cwnd > tp->snd_ssthresh) 1917 tp->snd_cwnd = tp->snd_ssthresh; 1918 } 1919 tp->t_dupacks = 0; 1920 /* 1921 * If we reach this point, ACK is not a duplicate, 1922 * i.e., it ACKs something we sent. 1923 */ 1924 if (tp->t_flags & TF_NEEDSYN) { 1925 /* 1926 * T/TCP: Connection was half-synchronized, and our 1927 * SYN has been ACK'd (so connection is now fully 1928 * synchronized). Go to non-starred state, 1929 * increment snd_una for ACK of SYN, and check if 1930 * we can do window scaling. 1931 */ 1932 tp->t_flags &= ~TF_NEEDSYN; 1933 tp->snd_una++; 1934 /* Do window scaling? */ 1935 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1936 (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1937 tp->rcv_scale = tp->request_r_scale; 1938 /* Send window already scaled. */ 1939 } 1940 } 1941 1942process_ACK: 1943 KASSERT(headlocked, ("%s: process_ACK: head not locked", 1944 __func__)); 1945 INP_LOCK_ASSERT(tp->t_inpcb); 1946 1947 acked = th->th_ack - tp->snd_una; 1948 tcpstat.tcps_rcvackpack++; 1949 tcpstat.tcps_rcvackbyte += acked; 1950 1951 /* 1952 * If we just performed our first retransmit, and the ACK 1953 * arrives within our recovery window, then it was a mistake 1954 * to do the retransmit in the first place. Recover our 1955 * original cwnd and ssthresh, and proceed to transmit where 1956 * we left off. 1957 */ 1958 if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) { 1959 ++tcpstat.tcps_sndrexmitbad; 1960 tp->snd_cwnd = tp->snd_cwnd_prev; 1961 tp->snd_ssthresh = tp->snd_ssthresh_prev; 1962 tp->snd_recover = tp->snd_recover_prev; 1963 if (tp->t_flags & TF_WASFRECOVERY) 1964 ENTER_FASTRECOVERY(tp); 1965 tp->snd_nxt = tp->snd_max; 1966 tp->t_badrxtwin = 0; /* XXX probably not required */ 1967 } 1968 1969 /* 1970 * If we have a timestamp reply, update smoothed 1971 * round trip time. If no timestamp is present but 1972 * transmit timer is running and timed sequence 1973 * number was acked, update smoothed round trip time. 1974 * Since we now have an rtt measurement, cancel the 1975 * timer backoff (cf., Phil Karn's retransmit alg.). 1976 * Recompute the initial retransmit timer. 1977 * 1978 * Some boxes send broken timestamp replies 1979 * during the SYN+ACK phase, ignore 1980 * timestamps of 0 or we could calculate a 1981 * huge RTT and blow up the retransmit timer. 1982 */ 1983 if ((to.to_flags & TOF_TS) != 0 && 1984 to.to_tsecr) { 1985 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr) 1986 tp->t_rttlow = ticks - to.to_tsecr; 1987 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1); 1988 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 1989 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 1990 tp->t_rttlow = ticks - tp->t_rtttime; 1991 tcp_xmit_timer(tp, ticks - tp->t_rtttime); 1992 } 1993 tcp_xmit_bandwidth_limit(tp, th->th_ack); 1994 1995 /* 1996 * If all outstanding data is acked, stop retransmit 1997 * timer and remember to restart (more output or persist). 1998 * If there is more data to be acked, restart retransmit 1999 * timer, using current (possibly backed-off) value. 2000 */ 2001 if (th->th_ack == tp->snd_max) { 2002 tcp_timer_activate(tp, TT_REXMT, 0); 2003 needoutput = 1; 2004 } else if (!tcp_timer_active(tp, TT_PERSIST)) 2005 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2006 2007 /* 2008 * If no data (only SYN) was ACK'd, 2009 * skip rest of ACK processing. 2010 */ 2011 if (acked == 0) 2012 goto step6; 2013 2014 /* 2015 * When new data is acked, open the congestion window. 2016 * If the window gives us less than ssthresh packets 2017 * in flight, open exponentially (maxseg per packet). 2018 * Otherwise open linearly: maxseg per window 2019 * (maxseg^2 / cwnd per packet). 2020 */ 2021 if ((!tcp_do_newreno && !(tp->t_flags & TF_SACK_PERMIT)) || 2022 !IN_FASTRECOVERY(tp)) { 2023 u_int cw = tp->snd_cwnd; 2024 u_int incr = tp->t_maxseg; 2025 if (cw > tp->snd_ssthresh) 2026 incr = incr * incr / cw; 2027 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale); 2028 } 2029 SOCKBUF_LOCK(&so->so_snd); 2030 if (acked > so->so_snd.sb_cc) { 2031 tp->snd_wnd -= so->so_snd.sb_cc; 2032 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc); 2033 ourfinisacked = 1; 2034 } else { 2035 sbdrop_locked(&so->so_snd, acked); 2036 tp->snd_wnd -= acked; 2037 ourfinisacked = 0; 2038 } 2039 /* NB: sowwakeup_locked() does an implicit unlock. */ 2040 sowwakeup_locked(so); 2041 /* Detect una wraparound. */ 2042 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2043 !IN_FASTRECOVERY(tp) && 2044 SEQ_GT(tp->snd_una, tp->snd_recover) && 2045 SEQ_LEQ(th->th_ack, tp->snd_recover)) 2046 tp->snd_recover = th->th_ack - 1; 2047 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) && 2048 IN_FASTRECOVERY(tp) && 2049 SEQ_GEQ(th->th_ack, tp->snd_recover)) 2050 EXIT_FASTRECOVERY(tp); 2051 tp->snd_una = th->th_ack; 2052 if (tp->t_flags & TF_SACK_PERMIT) { 2053 if (SEQ_GT(tp->snd_una, tp->snd_recover)) 2054 tp->snd_recover = tp->snd_una; 2055 } 2056 if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2057 tp->snd_nxt = tp->snd_una; 2058 2059 switch (tp->t_state) { 2060 2061 /* 2062 * In FIN_WAIT_1 STATE in addition to the processing 2063 * for the ESTABLISHED state if our FIN is now acknowledged 2064 * then enter FIN_WAIT_2. 2065 */ 2066 case TCPS_FIN_WAIT_1: 2067 if (ourfinisacked) { 2068 /* 2069 * If we can't receive any more 2070 * data, then closing user can proceed. 2071 * Starting the timer is contrary to the 2072 * specification, but if we don't get a FIN 2073 * we'll hang forever. 2074 * 2075 * XXXjl: 2076 * we should release the tp also, and use a 2077 * compressed state. 2078 */ 2079 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 2080 int timeout; 2081 2082 soisdisconnected(so); 2083 timeout = (tcp_fast_finwait2_recycle) ? 2084 tcp_finwait2_timeout : tcp_maxidle; 2085 tcp_timer_activate(tp, TT_2MSL, timeout); 2086 } 2087 tp->t_state = TCPS_FIN_WAIT_2; 2088 } 2089 break; 2090 2091 /* 2092 * In CLOSING STATE in addition to the processing for 2093 * the ESTABLISHED state if the ACK acknowledges our FIN 2094 * then enter the TIME-WAIT state, otherwise ignore 2095 * the segment. 2096 */ 2097 case TCPS_CLOSING: 2098 if (ourfinisacked) { 2099 KASSERT(headlocked, ("%s: process_ACK: " 2100 "head not locked", __func__)); 2101 tcp_twstart(tp); 2102 INP_INFO_WUNLOCK(&tcbinfo); 2103 headlocked = 0; 2104 m_freem(m); 2105 return; 2106 } 2107 break; 2108 2109 /* 2110 * In LAST_ACK, we may still be waiting for data to drain 2111 * and/or to be acked, as well as for the ack of our FIN. 2112 * If our FIN is now acknowledged, delete the TCB, 2113 * enter the closed state and return. 2114 */ 2115 case TCPS_LAST_ACK: 2116 if (ourfinisacked) { 2117 KASSERT(headlocked, ("%s: process_ACK: " 2118 "tcp_close: head not locked", __func__)); 2119 tp = tcp_close(tp); 2120 goto drop; 2121 } 2122 break; 2123 } 2124 } 2125 2126step6: 2127 KASSERT(headlocked, ("%s: step6: head not locked", __func__)); 2128 INP_LOCK_ASSERT(tp->t_inpcb); 2129 2130 /* 2131 * Update window information. 2132 * Don't look at window if no ACK: TAC's send garbage on first SYN. 2133 */ 2134 if ((thflags & TH_ACK) && 2135 (SEQ_LT(tp->snd_wl1, th->th_seq) || 2136 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 2137 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 2138 /* keep track of pure window updates */ 2139 if (tlen == 0 && 2140 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 2141 tcpstat.tcps_rcvwinupd++; 2142 tp->snd_wnd = tiwin; 2143 tp->snd_wl1 = th->th_seq; 2144 tp->snd_wl2 = th->th_ack; 2145 if (tp->snd_wnd > tp->max_sndwnd) 2146 tp->max_sndwnd = tp->snd_wnd; 2147 needoutput = 1; 2148 } 2149 2150 /* 2151 * Process segments with URG. 2152 */ 2153 if ((thflags & TH_URG) && th->th_urp && 2154 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2155 /* 2156 * This is a kludge, but if we receive and accept 2157 * random urgent pointers, we'll crash in 2158 * soreceive. It's hard to imagine someone 2159 * actually wanting to send this much urgent data. 2160 */ 2161 SOCKBUF_LOCK(&so->so_rcv); 2162 if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2163 th->th_urp = 0; /* XXX */ 2164 thflags &= ~TH_URG; /* XXX */ 2165 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2166 goto dodata; /* XXX */ 2167 } 2168 /* 2169 * If this segment advances the known urgent pointer, 2170 * then mark the data stream. This should not happen 2171 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2172 * a FIN has been received from the remote side. 2173 * In these states we ignore the URG. 2174 * 2175 * According to RFC961 (Assigned Protocols), 2176 * the urgent pointer points to the last octet 2177 * of urgent data. We continue, however, 2178 * to consider it to indicate the first octet 2179 * of data past the urgent section as the original 2180 * spec states (in one of two places). 2181 */ 2182 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2183 tp->rcv_up = th->th_seq + th->th_urp; 2184 so->so_oobmark = so->so_rcv.sb_cc + 2185 (tp->rcv_up - tp->rcv_nxt) - 1; 2186 if (so->so_oobmark == 0) 2187 so->so_rcv.sb_state |= SBS_RCVATMARK; 2188 sohasoutofband(so); 2189 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2190 } 2191 SOCKBUF_UNLOCK(&so->so_rcv); 2192 /* 2193 * Remove out of band data so doesn't get presented to user. 2194 * This can happen independent of advancing the URG pointer, 2195 * but if two URG's are pending at once, some out-of-band 2196 * data may creep in... ick. 2197 */ 2198 if (th->th_urp <= (u_long)tlen && 2199 !(so->so_options & SO_OOBINLINE)) { 2200 /* hdr drop is delayed */ 2201 tcp_pulloutofband(so, th, m, drop_hdrlen); 2202 } 2203 } else { 2204 /* 2205 * If no out of band data is expected, 2206 * pull receive urgent pointer along 2207 * with the receive window. 2208 */ 2209 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2210 tp->rcv_up = tp->rcv_nxt; 2211 } 2212dodata: /* XXX */ 2213 KASSERT(headlocked, ("%s: dodata: head not locked", __func__)); 2214 INP_LOCK_ASSERT(tp->t_inpcb); 2215 2216 /* 2217 * Process the segment text, merging it into the TCP sequencing queue, 2218 * and arranging for acknowledgment of receipt if necessary. 2219 * This process logically involves adjusting tp->rcv_wnd as data 2220 * is presented to the user (this happens in tcp_usrreq.c, 2221 * case PRU_RCVD). If a FIN has already been received on this 2222 * connection then we just ignore the text. 2223 */ 2224 if ((tlen || (thflags & TH_FIN)) && 2225 TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2226 tcp_seq save_start = th->th_seq; 2227 m_adj(m, drop_hdrlen); /* delayed header drop */ 2228 /* 2229 * Insert segment which includes th into TCP reassembly queue 2230 * with control block tp. Set thflags to whether reassembly now 2231 * includes a segment with FIN. This handles the common case 2232 * inline (segment is the next to be received on an established 2233 * connection, and the queue is empty), avoiding linkage into 2234 * and removal from the queue and repetition of various 2235 * conversions. 2236 * Set DELACK for segments received in order, but ack 2237 * immediately when segments are out of order (so 2238 * fast retransmit can work). 2239 */ 2240 if (th->th_seq == tp->rcv_nxt && 2241 LIST_EMPTY(&tp->t_segq) && 2242 TCPS_HAVEESTABLISHED(tp->t_state)) { 2243 if (DELAY_ACK(tp)) 2244 tp->t_flags |= TF_DELACK; 2245 else 2246 tp->t_flags |= TF_ACKNOW; 2247 tp->rcv_nxt += tlen; 2248 thflags = th->th_flags & TH_FIN; 2249 tcpstat.tcps_rcvpack++; 2250 tcpstat.tcps_rcvbyte += tlen; 2251 ND6_HINT(tp); 2252 SOCKBUF_LOCK(&so->so_rcv); 2253 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2254 m_freem(m); 2255 else 2256 sbappendstream_locked(&so->so_rcv, m); 2257 /* NB: sorwakeup_locked() does an implicit unlock. */ 2258 sorwakeup_locked(so); 2259 } else { 2260 /* 2261 * XXX: Due to the header drop above "th" is 2262 * theoretically invalid by now. Fortunately 2263 * m_adj() doesn't actually frees any mbufs 2264 * when trimming from the head. 2265 */ 2266 thflags = tcp_reass(tp, th, &tlen, m); 2267 tp->t_flags |= TF_ACKNOW; 2268 } 2269 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2270 tcp_update_sack_list(tp, save_start, save_start + tlen); 2271#if 0 2272 /* 2273 * Note the amount of data that peer has sent into 2274 * our window, in order to estimate the sender's 2275 * buffer size. 2276 * XXX: Unused. 2277 */ 2278 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2279#endif 2280 } else { 2281 m_freem(m); 2282 thflags &= ~TH_FIN; 2283 } 2284 2285 /* 2286 * If FIN is received ACK the FIN and let the user know 2287 * that the connection is closing. 2288 */ 2289 if (thflags & TH_FIN) { 2290 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2291 socantrcvmore(so); 2292 /* 2293 * If connection is half-synchronized 2294 * (ie NEEDSYN flag on) then delay ACK, 2295 * so it may be piggybacked when SYN is sent. 2296 * Otherwise, since we received a FIN then no 2297 * more input can be expected, send ACK now. 2298 */ 2299 if (tp->t_flags & TF_NEEDSYN) 2300 tp->t_flags |= TF_DELACK; 2301 else 2302 tp->t_flags |= TF_ACKNOW; 2303 tp->rcv_nxt++; 2304 } 2305 switch (tp->t_state) { 2306 2307 /* 2308 * In SYN_RECEIVED and ESTABLISHED STATES 2309 * enter the CLOSE_WAIT state. 2310 */ 2311 case TCPS_SYN_RECEIVED: 2312 tp->t_starttime = ticks; 2313 /* FALLTHROUGH */ 2314 case TCPS_ESTABLISHED: 2315 tp->t_state = TCPS_CLOSE_WAIT; 2316 break; 2317 2318 /* 2319 * If still in FIN_WAIT_1 STATE FIN has not been acked so 2320 * enter the CLOSING state. 2321 */ 2322 case TCPS_FIN_WAIT_1: 2323 tp->t_state = TCPS_CLOSING; 2324 break; 2325 2326 /* 2327 * In FIN_WAIT_2 state enter the TIME_WAIT state, 2328 * starting the time-wait timer, turning off the other 2329 * standard timers. 2330 */ 2331 case TCPS_FIN_WAIT_2: 2332 KASSERT(headlocked == 1, ("%s: dodata: " 2333 "TCP_FIN_WAIT_2: head not locked", __func__)); 2334 tcp_twstart(tp); 2335 INP_INFO_WUNLOCK(&tcbinfo); 2336 return; 2337 } 2338 } 2339 INP_INFO_WUNLOCK(&tcbinfo); 2340 headlocked = 0; 2341#ifdef TCPDEBUG 2342 if (so->so_options & SO_DEBUG) 2343 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 2344 &tcp_savetcp, 0); 2345#endif 2346 2347 /* 2348 * Return any desired output. 2349 */ 2350 if (needoutput || (tp->t_flags & TF_ACKNOW)) 2351 (void) tcp_output(tp); 2352 2353check_delack: 2354 KASSERT(headlocked == 0, ("%s: check_delack: head locked", 2355 __func__)); 2356 INP_INFO_UNLOCK_ASSERT(&tcbinfo); 2357 INP_LOCK_ASSERT(tp->t_inpcb); 2358 if (tp->t_flags & TF_DELACK) { 2359 tp->t_flags &= ~TF_DELACK; 2360 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 2361 } 2362 INP_UNLOCK(tp->t_inpcb); 2363 return; 2364 2365dropafterack: 2366 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__)); 2367 /* 2368 * Generate an ACK dropping incoming segment if it occupies 2369 * sequence space, where the ACK reflects our state. 2370 * 2371 * We can now skip the test for the RST flag since all 2372 * paths to this code happen after packets containing 2373 * RST have been dropped. 2374 * 2375 * In the SYN-RECEIVED state, don't send an ACK unless the 2376 * segment we received passes the SYN-RECEIVED ACK test. 2377 * If it fails send a RST. This breaks the loop in the 2378 * "LAND" DoS attack, and also prevents an ACK storm 2379 * between two listening ports that have been sent forged 2380 * SYN segments, each with the source address of the other. 2381 */ 2382 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 2383 (SEQ_GT(tp->snd_una, th->th_ack) || 2384 SEQ_GT(th->th_ack, tp->snd_max)) ) { 2385 rstreason = BANDLIM_RST_OPENPORT; 2386 goto dropwithreset; 2387 } 2388#ifdef TCPDEBUG 2389 if (so->so_options & SO_DEBUG) 2390 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2391 &tcp_savetcp, 0); 2392#endif 2393 KASSERT(headlocked, ("%s: headlocked should be 1", __func__)); 2394 INP_INFO_WUNLOCK(&tcbinfo); 2395 tp->t_flags |= TF_ACKNOW; 2396 (void) tcp_output(tp); 2397 INP_UNLOCK(tp->t_inpcb); 2398 m_freem(m); 2399 return; 2400 2401dropwithreset: 2402 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__)); 2403 2404 tcp_dropwithreset(m, th, tp, tlen, rstreason); 2405 2406 if (tp != NULL) 2407 INP_UNLOCK(tp->t_inpcb); 2408 if (headlocked) 2409 INP_INFO_WUNLOCK(&tcbinfo); 2410 return; 2411 2412drop: 2413 /* 2414 * Drop space held by incoming segment and return. 2415 */ 2416#ifdef TCPDEBUG 2417 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 2418 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 2419 &tcp_savetcp, 0); 2420#endif 2421 if (tp != NULL) 2422 INP_UNLOCK(tp->t_inpcb); 2423 if (headlocked) 2424 INP_INFO_WUNLOCK(&tcbinfo); 2425 m_freem(m); 2426 return; 2427} 2428 2429/* 2430 * Issue RST and make ACK acceptable to originator of segment. 2431 * The mbuf must still include the original packet header. 2432 * tp may be NULL. 2433 */ 2434static void 2435tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 2436 int tlen, int rstreason) 2437{ 2438 struct ip *ip; 2439#ifdef INET6 2440 struct ip6_hdr *ip6; 2441#endif 2442 /* Don't bother if destination was broadcast/multicast. */ 2443 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 2444 goto drop; 2445#ifdef INET6 2446 if (mtod(m, struct ip *)->ip_v == 6) { 2447 ip6 = mtod(m, struct ip6_hdr *); 2448 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 2449 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 2450 goto drop; 2451 /* IPv6 anycast check is done at tcp6_input() */ 2452 } else 2453#endif 2454 { 2455 ip = mtod(m, struct ip *); 2456 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 2457 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 2458 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 2459 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 2460 goto drop; 2461 } 2462 2463 /* Perform bandwidth limiting. */ 2464 if (badport_bandlim(rstreason) < 0) 2465 goto drop; 2466 2467 /* tcp_respond consumes the mbuf chain. */ 2468 if (th->th_flags & TH_ACK) { 2469 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 2470 th->th_ack, TH_RST); 2471 } else { 2472 if (th->th_flags & TH_SYN) 2473 tlen++; 2474 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 2475 (tcp_seq)0, TH_RST|TH_ACK); 2476 } 2477 return; 2478drop: 2479 m_freem(m); 2480 return; 2481} 2482 2483/* 2484 * Parse TCP options and place in tcpopt. 2485 */ 2486static void 2487tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 2488{ 2489 int opt, optlen; 2490 2491 to->to_flags = 0; 2492 for (; cnt > 0; cnt -= optlen, cp += optlen) { 2493 opt = cp[0]; 2494 if (opt == TCPOPT_EOL) 2495 break; 2496 if (opt == TCPOPT_NOP) 2497 optlen = 1; 2498 else { 2499 if (cnt < 2) 2500 break; 2501 optlen = cp[1]; 2502 if (optlen < 2 || optlen > cnt) 2503 break; 2504 } 2505 switch (opt) { 2506 case TCPOPT_MAXSEG: 2507 if (optlen != TCPOLEN_MAXSEG) 2508 continue; 2509 if (!(flags & TO_SYN)) 2510 continue; 2511 to->to_flags |= TOF_MSS; 2512 bcopy((char *)cp + 2, 2513 (char *)&to->to_mss, sizeof(to->to_mss)); 2514 to->to_mss = ntohs(to->to_mss); 2515 break; 2516 case TCPOPT_WINDOW: 2517 if (optlen != TCPOLEN_WINDOW) 2518 continue; 2519 if (!(flags & TO_SYN)) 2520 continue; 2521 to->to_flags |= TOF_SCALE; 2522 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 2523 break; 2524 case TCPOPT_TIMESTAMP: 2525 if (optlen != TCPOLEN_TIMESTAMP) 2526 continue; 2527 to->to_flags |= TOF_TS; 2528 bcopy((char *)cp + 2, 2529 (char *)&to->to_tsval, sizeof(to->to_tsval)); 2530 to->to_tsval = ntohl(to->to_tsval); 2531 bcopy((char *)cp + 6, 2532 (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 2533 to->to_tsecr = ntohl(to->to_tsecr); 2534 break; 2535#ifdef TCP_SIGNATURE 2536 /* 2537 * XXX In order to reply to a host which has set the 2538 * TCP_SIGNATURE option in its initial SYN, we have to 2539 * record the fact that the option was observed here 2540 * for the syncache code to perform the correct response. 2541 */ 2542 case TCPOPT_SIGNATURE: 2543 if (optlen != TCPOLEN_SIGNATURE) 2544 continue; 2545 to->to_flags |= TOF_SIGNATURE; 2546 to->to_signature = cp + 2; 2547 break; 2548#endif 2549 case TCPOPT_SACK_PERMITTED: 2550 if (optlen != TCPOLEN_SACK_PERMITTED) 2551 continue; 2552 if (!(flags & TO_SYN)) 2553 continue; 2554 if (!tcp_do_sack) 2555 continue; 2556 to->to_flags |= TOF_SACKPERM; 2557 break; 2558 case TCPOPT_SACK: 2559 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 2560 continue; 2561 if (flags & TO_SYN) 2562 continue; 2563 to->to_flags |= TOF_SACK; 2564 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 2565 to->to_sacks = cp + 2; 2566 tcpstat.tcps_sack_rcv_blocks++; 2567 break; 2568 default: 2569 continue; 2570 } 2571 } 2572} 2573 2574/* 2575 * Pull out of band byte out of a segment so 2576 * it doesn't appear in the user's data queue. 2577 * It is still reflected in the segment length for 2578 * sequencing purposes. 2579 */ 2580static void 2581tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 2582 int off) 2583{ 2584 int cnt = off + th->th_urp - 1; 2585 2586 while (cnt >= 0) { 2587 if (m->m_len > cnt) { 2588 char *cp = mtod(m, caddr_t) + cnt; 2589 struct tcpcb *tp = sototcpcb(so); 2590 2591 tp->t_iobc = *cp; 2592 tp->t_oobflags |= TCPOOB_HAVEDATA; 2593 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 2594 m->m_len--; 2595 if (m->m_flags & M_PKTHDR) 2596 m->m_pkthdr.len--; 2597 return; 2598 } 2599 cnt -= m->m_len; 2600 m = m->m_next; 2601 if (m == NULL) 2602 break; 2603 } 2604 panic("tcp_pulloutofband"); 2605} 2606 2607/* 2608 * Collect new round-trip time estimate 2609 * and update averages and current timeout. 2610 */ 2611static void 2612tcp_xmit_timer(struct tcpcb *tp, int rtt) 2613{ 2614 int delta; 2615 2616 INP_LOCK_ASSERT(tp->t_inpcb); 2617 2618 tcpstat.tcps_rttupdated++; 2619 tp->t_rttupdated++; 2620 if (tp->t_srtt != 0) { 2621 /* 2622 * srtt is stored as fixed point with 5 bits after the 2623 * binary point (i.e., scaled by 8). The following magic 2624 * is equivalent to the smoothing algorithm in rfc793 with 2625 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 2626 * point). Adjust rtt to origin 0. 2627 */ 2628 delta = ((rtt - 1) << TCP_DELTA_SHIFT) 2629 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 2630 2631 if ((tp->t_srtt += delta) <= 0) 2632 tp->t_srtt = 1; 2633 2634 /* 2635 * We accumulate a smoothed rtt variance (actually, a 2636 * smoothed mean difference), then set the retransmit 2637 * timer to smoothed rtt + 4 times the smoothed variance. 2638 * rttvar is stored as fixed point with 4 bits after the 2639 * binary point (scaled by 16). The following is 2640 * equivalent to rfc793 smoothing with an alpha of .75 2641 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 2642 * rfc793's wired-in beta. 2643 */ 2644 if (delta < 0) 2645 delta = -delta; 2646 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 2647 if ((tp->t_rttvar += delta) <= 0) 2648 tp->t_rttvar = 1; 2649 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 2650 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2651 } else { 2652 /* 2653 * No rtt measurement yet - use the unsmoothed rtt. 2654 * Set the variance to half the rtt (so our first 2655 * retransmit happens at 3*rtt). 2656 */ 2657 tp->t_srtt = rtt << TCP_RTT_SHIFT; 2658 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 2659 tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 2660 } 2661 tp->t_rtttime = 0; 2662 tp->t_rxtshift = 0; 2663 2664 /* 2665 * the retransmit should happen at rtt + 4 * rttvar. 2666 * Because of the way we do the smoothing, srtt and rttvar 2667 * will each average +1/2 tick of bias. When we compute 2668 * the retransmit timer, we want 1/2 tick of rounding and 2669 * 1 extra tick because of +-1/2 tick uncertainty in the 2670 * firing of the timer. The bias will give us exactly the 2671 * 1.5 tick we need. But, because the bias is 2672 * statistical, we have to test that we don't drop below 2673 * the minimum feasible timer (which is 2 ticks). 2674 */ 2675 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 2676 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 2677 2678 /* 2679 * We received an ack for a packet that wasn't retransmitted; 2680 * it is probably safe to discard any error indications we've 2681 * received recently. This isn't quite right, but close enough 2682 * for now (a route might have failed after we sent a segment, 2683 * and the return path might not be symmetrical). 2684 */ 2685 tp->t_softerror = 0; 2686} 2687 2688/* 2689 * Determine a reasonable value for maxseg size. 2690 * If the route is known, check route for mtu. 2691 * If none, use an mss that can be handled on the outgoing 2692 * interface without forcing IP to fragment; if bigger than 2693 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES 2694 * to utilize large mbufs. If no route is found, route has no mtu, 2695 * or the destination isn't local, use a default, hopefully conservative 2696 * size (usually 512 or the default IP max size, but no more than the mtu 2697 * of the interface), as we can't discover anything about intervening 2698 * gateways or networks. We also initialize the congestion/slow start 2699 * window to be a single segment if the destination isn't local. 2700 * While looking at the routing entry, we also initialize other path-dependent 2701 * parameters from pre-set or cached values in the routing entry. 2702 * 2703 * Also take into account the space needed for options that we 2704 * send regularly. Make maxseg shorter by that amount to assure 2705 * that we can send maxseg amount of data even when the options 2706 * are present. Store the upper limit of the length of options plus 2707 * data in maxopd. 2708 * 2709 * In case of T/TCP, we call this routine during implicit connection 2710 * setup as well (offer = -1), to initialize maxseg from the cached 2711 * MSS of our peer. 2712 * 2713 * NOTE that this routine is only called when we process an incoming 2714 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt(). 2715 */ 2716void 2717tcp_mss(struct tcpcb *tp, int offer) 2718{ 2719 int rtt, mss; 2720 u_long bufsize; 2721 u_long maxmtu; 2722 struct inpcb *inp = tp->t_inpcb; 2723 struct socket *so; 2724 struct hc_metrics_lite metrics; 2725 int origoffer = offer; 2726 int mtuflags = 0; 2727#ifdef INET6 2728 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 2729 size_t min_protoh = isipv6 ? 2730 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 2731 sizeof (struct tcpiphdr); 2732#else 2733 const size_t min_protoh = sizeof(struct tcpiphdr); 2734#endif 2735 2736 /* Initialize. */ 2737#ifdef INET6 2738 if (isipv6) { 2739 maxmtu = tcp_maxmtu6(&inp->inp_inc, &mtuflags); 2740 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt; 2741 } else 2742#endif 2743 { 2744 maxmtu = tcp_maxmtu(&inp->inp_inc, &mtuflags); 2745 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt; 2746 } 2747 so = inp->inp_socket; 2748 2749 /* 2750 * No route to sender, stay with default mss and return. 2751 */ 2752 if (maxmtu == 0) 2753 return; 2754 2755 /* What have we got? */ 2756 switch (offer) { 2757 case 0: 2758 /* 2759 * Offer == 0 means that there was no MSS on the SYN 2760 * segment, in this case we use tcp_mssdflt. 2761 */ 2762 offer = 2763#ifdef INET6 2764 isipv6 ? tcp_v6mssdflt : 2765#endif 2766 tcp_mssdflt; 2767 break; 2768 2769 case -1: 2770 /* 2771 * Offer == -1 means that we didn't receive SYN yet. 2772 */ 2773 /* FALLTHROUGH */ 2774 2775 default: 2776 /* 2777 * Prevent DoS attack with too small MSS. Round up 2778 * to at least minmss. 2779 */ 2780 offer = max(offer, tcp_minmss); 2781 /* 2782 * Sanity check: make sure that maxopd will be large 2783 * enough to allow some data on segments even if the 2784 * all the option space is used (40bytes). Otherwise 2785 * funny things may happen in tcp_output. 2786 */ 2787 offer = max(offer, 64); 2788 } 2789 2790 /* 2791 * rmx information is now retrieved from tcp_hostcache. 2792 */ 2793 tcp_hc_get(&inp->inp_inc, &metrics); 2794 2795 /* 2796 * If there's a discovered mtu int tcp hostcache, use it 2797 * else, use the link mtu. 2798 */ 2799 if (metrics.rmx_mtu) 2800 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 2801 else { 2802#ifdef INET6 2803 if (isipv6) { 2804 mss = maxmtu - min_protoh; 2805 if (!path_mtu_discovery && 2806 !in6_localaddr(&inp->in6p_faddr)) 2807 mss = min(mss, tcp_v6mssdflt); 2808 } else 2809#endif 2810 { 2811 mss = maxmtu - min_protoh; 2812 if (!path_mtu_discovery && 2813 !in_localaddr(inp->inp_faddr)) 2814 mss = min(mss, tcp_mssdflt); 2815 } 2816 } 2817 mss = min(mss, offer); 2818 2819 /* 2820 * maxopd stores the maximum length of data AND options 2821 * in a segment; maxseg is the amount of data in a normal 2822 * segment. We need to store this value (maxopd) apart 2823 * from maxseg, because now every segment carries options 2824 * and thus we normally have somewhat less data in segments. 2825 */ 2826 tp->t_maxopd = mss; 2827 2828 /* 2829 * origoffer==-1 indicates that no segments were received yet. 2830 * In this case we just guess. 2831 */ 2832 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 2833 (origoffer == -1 || 2834 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 2835 mss -= TCPOLEN_TSTAMP_APPA; 2836 tp->t_maxseg = mss; 2837 2838#if (MCLBYTES & (MCLBYTES - 1)) == 0 2839 if (mss > MCLBYTES) 2840 mss &= ~(MCLBYTES-1); 2841#else 2842 if (mss > MCLBYTES) 2843 mss = mss / MCLBYTES * MCLBYTES; 2844#endif 2845 tp->t_maxseg = mss; 2846 2847 /* 2848 * If there's a pipesize, change the socket buffer to that size, 2849 * don't change if sb_hiwat is different than default (then it 2850 * has been changed on purpose with setsockopt). 2851 * Make the socket buffers an integral number of mss units; 2852 * if the mss is larger than the socket buffer, decrease the mss. 2853 */ 2854 SOCKBUF_LOCK(&so->so_snd); 2855 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe) 2856 bufsize = metrics.rmx_sendpipe; 2857 else 2858 bufsize = so->so_snd.sb_hiwat; 2859 if (bufsize < mss) 2860 mss = bufsize; 2861 else { 2862 bufsize = roundup(bufsize, mss); 2863 if (bufsize > sb_max) 2864 bufsize = sb_max; 2865 if (bufsize > so->so_snd.sb_hiwat) 2866 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 2867 } 2868 SOCKBUF_UNLOCK(&so->so_snd); 2869 tp->t_maxseg = mss; 2870 2871 SOCKBUF_LOCK(&so->so_rcv); 2872 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe) 2873 bufsize = metrics.rmx_recvpipe; 2874 else 2875 bufsize = so->so_rcv.sb_hiwat; 2876 if (bufsize > mss) { 2877 bufsize = roundup(bufsize, mss); 2878 if (bufsize > sb_max) 2879 bufsize = sb_max; 2880 if (bufsize > so->so_rcv.sb_hiwat) 2881 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 2882 } 2883 SOCKBUF_UNLOCK(&so->so_rcv); 2884 /* 2885 * While we're here, check the others too. 2886 */ 2887 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 2888 tp->t_srtt = rtt; 2889 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 2890 tcpstat.tcps_usedrtt++; 2891 if (metrics.rmx_rttvar) { 2892 tp->t_rttvar = metrics.rmx_rttvar; 2893 tcpstat.tcps_usedrttvar++; 2894 } else { 2895 /* default variation is +- 1 rtt */ 2896 tp->t_rttvar = 2897 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 2898 } 2899 TCPT_RANGESET(tp->t_rxtcur, 2900 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 2901 tp->t_rttmin, TCPTV_REXMTMAX); 2902 } 2903 if (metrics.rmx_ssthresh) { 2904 /* 2905 * There's some sort of gateway or interface 2906 * buffer limit on the path. Use this to set 2907 * the slow start threshhold, but set the 2908 * threshold to no less than 2*mss. 2909 */ 2910 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh); 2911 tcpstat.tcps_usedssthresh++; 2912 } 2913 if (metrics.rmx_bandwidth) 2914 tp->snd_bandwidth = metrics.rmx_bandwidth; 2915 2916 /* 2917 * Set the slow-start flight size depending on whether this 2918 * is a local network or not. 2919 * 2920 * Extend this so we cache the cwnd too and retrieve it here. 2921 * Make cwnd even bigger than RFC3390 suggests but only if we 2922 * have previous experience with the remote host. Be careful 2923 * not make cwnd bigger than remote receive window or our own 2924 * send socket buffer. Maybe put some additional upper bound 2925 * on the retrieved cwnd. Should do incremental updates to 2926 * hostcache when cwnd collapses so next connection doesn't 2927 * overloads the path again. 2928 * 2929 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost. 2930 * We currently check only in syncache_socket for that. 2931 */ 2932#define TCP_METRICS_CWND 2933#ifdef TCP_METRICS_CWND 2934 if (metrics.rmx_cwnd) 2935 tp->snd_cwnd = max(mss, 2936 min(metrics.rmx_cwnd / 2, 2937 min(tp->snd_wnd, so->so_snd.sb_hiwat))); 2938 else 2939#endif 2940 if (tcp_do_rfc3390) 2941 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380)); 2942#ifdef INET6 2943 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) || 2944 (!isipv6 && in_localaddr(inp->inp_faddr))) 2945#else 2946 else if (in_localaddr(inp->inp_faddr)) 2947#endif 2948 tp->snd_cwnd = mss * ss_fltsz_local; 2949 else 2950 tp->snd_cwnd = mss * ss_fltsz; 2951 2952 /* Check the interface for TSO capabilities. */ 2953 if (mtuflags & CSUM_TSO) 2954 tp->t_flags |= TF_TSO; 2955} 2956 2957/* 2958 * Determine the MSS option to send on an outgoing SYN. 2959 */ 2960int 2961tcp_mssopt(struct in_conninfo *inc) 2962{ 2963 int mss = 0; 2964 u_long maxmtu = 0; 2965 u_long thcmtu = 0; 2966 size_t min_protoh; 2967#ifdef INET6 2968 int isipv6 = inc->inc_isipv6 ? 1 : 0; 2969#endif 2970 2971 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 2972 2973#ifdef INET6 2974 if (isipv6) { 2975 mss = tcp_v6mssdflt; 2976 maxmtu = tcp_maxmtu6(inc, NULL); 2977 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2978 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 2979 } else 2980#endif 2981 { 2982 mss = tcp_mssdflt; 2983 maxmtu = tcp_maxmtu(inc, NULL); 2984 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 2985 min_protoh = sizeof(struct tcpiphdr); 2986 } 2987 if (maxmtu && thcmtu) 2988 mss = min(maxmtu, thcmtu) - min_protoh; 2989 else if (maxmtu || thcmtu) 2990 mss = max(maxmtu, thcmtu) - min_protoh; 2991 2992 return (mss); 2993} 2994 2995 2996/* 2997 * On a partial ack arrives, force the retransmission of the 2998 * next unacknowledged segment. Do not clear tp->t_dupacks. 2999 * By setting snd_nxt to ti_ack, this forces retransmission timer to 3000 * be started again. 3001 */ 3002static void 3003tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 3004{ 3005 tcp_seq onxt = tp->snd_nxt; 3006 u_long ocwnd = tp->snd_cwnd; 3007 3008 tcp_timer_activate(tp, TT_REXMT, 0); 3009 tp->t_rtttime = 0; 3010 tp->snd_nxt = th->th_ack; 3011 /* 3012 * Set snd_cwnd to one segment beyond acknowledged offset. 3013 * (tp->snd_una has not yet been updated when this function is called.) 3014 */ 3015 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una); 3016 tp->t_flags |= TF_ACKNOW; 3017 (void) tcp_output(tp); 3018 tp->snd_cwnd = ocwnd; 3019 if (SEQ_GT(onxt, tp->snd_nxt)) 3020 tp->snd_nxt = onxt; 3021 /* 3022 * Partial window deflation. Relies on fact that tp->snd_una 3023 * not updated yet. 3024 */ 3025 if (tp->snd_cwnd > th->th_ack - tp->snd_una) 3026 tp->snd_cwnd -= th->th_ack - tp->snd_una; 3027 else 3028 tp->snd_cwnd = 0; 3029 tp->snd_cwnd += tp->t_maxseg; 3030} 3031