1c398230bSWarner Losh /*- 2e79adb8eSGarrett Wollman * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 3df8bae1dSRodney W. Grimes * The Regents of the University of California. All rights reserved. 4dbc42409SLawrence Stewart * Copyright (c) 2007-2008,2010 5dbc42409SLawrence Stewart * Swinburne University of Technology, Melbourne, Australia. 6dbc42409SLawrence Stewart * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org> 7dbc42409SLawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation 8fa046d87SRobert Watson * Copyright (c) 2010-2011 Juniper Networks, Inc. 9dbc42409SLawrence Stewart * All rights reserved. 10dbc42409SLawrence Stewart * 11dbc42409SLawrence Stewart * Portions of this software were developed at the Centre for Advanced Internet 12891b8ed4SLawrence Stewart * Architectures, Swinburne University of Technology, by Lawrence Stewart, 13891b8ed4SLawrence Stewart * James Healy and David Hayes, made possible in part by a grant from the Cisco 14891b8ed4SLawrence Stewart * University Research Program Fund at Community Foundation Silicon Valley. 15dbc42409SLawrence Stewart * 16dbc42409SLawrence Stewart * Portions of this software were developed at the Centre for Advanced 17dbc42409SLawrence Stewart * Internet Architectures, Swinburne University of Technology, Melbourne, 18dbc42409SLawrence Stewart * Australia by David Hayes under sponsorship from the FreeBSD Foundation. 19df8bae1dSRodney W. Grimes * 20fa046d87SRobert Watson * Portions of this software were developed by Robert N. M. Watson under 21fa046d87SRobert Watson * contract to Juniper Networks, Inc. 22fa046d87SRobert Watson * 23df8bae1dSRodney W. Grimes * Redistribution and use in source and binary forms, with or without 24df8bae1dSRodney W. Grimes * modification, are permitted provided that the following conditions 25df8bae1dSRodney W. Grimes * are met: 26df8bae1dSRodney W. Grimes * 1. Redistributions of source code must retain the above copyright 27df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer. 28df8bae1dSRodney W. Grimes * 2. Redistributions in binary form must reproduce the above copyright 29df8bae1dSRodney W. Grimes * notice, this list of conditions and the following disclaimer in the 30df8bae1dSRodney W. Grimes * documentation and/or other materials provided with the distribution. 31df8bae1dSRodney W. Grimes * 4. Neither the name of the University nor the names of its contributors 32df8bae1dSRodney W. Grimes * may be used to endorse or promote products derived from this software 33df8bae1dSRodney W. Grimes * without specific prior written permission. 34df8bae1dSRodney W. Grimes * 35df8bae1dSRodney W. Grimes * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 36df8bae1dSRodney W. Grimes * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 37df8bae1dSRodney W. Grimes * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 38df8bae1dSRodney W. Grimes * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 39df8bae1dSRodney W. Grimes * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 40df8bae1dSRodney W. Grimes * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 41df8bae1dSRodney W. Grimes * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 42df8bae1dSRodney W. Grimes * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 43df8bae1dSRodney W. Grimes * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 44df8bae1dSRodney W. Grimes * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 45df8bae1dSRodney W. Grimes * SUCH DAMAGE. 46df8bae1dSRodney W. Grimes * 47e79adb8eSGarrett Wollman * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95 48df8bae1dSRodney W. Grimes */ 49df8bae1dSRodney W. Grimes 504b421e2dSMike Silbersack #include <sys/cdefs.h> 514b421e2dSMike Silbersack __FBSDID("$FreeBSD$"); 524b421e2dSMike Silbersack 53f9e354dfSJulian Elischer #include "opt_ipfw.h" /* for ipfw_fwd */ 541cfd4b53SBruce M Simpson #include "opt_inet.h" 55fb59c426SYoshinobu Inoue #include "opt_inet6.h" 563a2a9f79SYoshinobu Inoue #include "opt_ipsec.h" 570cc12cc5SJoerg Wunsch #include "opt_tcpdebug.h" 580cc12cc5SJoerg Wunsch 59df8bae1dSRodney W. Grimes #include <sys/param.h> 6098163b98SPoul-Henning Kamp #include <sys/kernel.h> 6139bc9de5SLawrence Stewart #include <sys/hhook.h> 62df8bae1dSRodney W. Grimes #include <sys/malloc.h> 63df8bae1dSRodney W. Grimes #include <sys/mbuf.h> 64a29f300eSGarrett Wollman #include <sys/proc.h> /* for proc0 declaration */ 65df8bae1dSRodney W. Grimes #include <sys/protosw.h> 6657f60867SMark Johnston #include <sys/sdt.h> 67960ed29cSSeigo Tanimura #include <sys/signalvar.h> 68df8bae1dSRodney W. Grimes #include <sys/socket.h> 69df8bae1dSRodney W. Grimes #include <sys/socketvar.h> 70960ed29cSSeigo Tanimura #include <sys/sysctl.h> 71816a3d83SPoul-Henning Kamp #include <sys/syslog.h> 72960ed29cSSeigo Tanimura #include <sys/systm.h> 73df8bae1dSRodney W. Grimes 74e79adb8eSGarrett Wollman #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */ 75e79adb8eSGarrett Wollman 7612e2e970SAndre Oppermann #include <vm/uma.h> 7712e2e970SAndre Oppermann 78df8bae1dSRodney W. Grimes #include <net/if.h> 7976039bc8SGleb Smirnoff #include <net/if_var.h> 80df8bae1dSRodney W. Grimes #include <net/route.h> 81530c0060SRobert Watson #include <net/vnet.h> 82df8bae1dSRodney W. Grimes 83773673c1SAndre Oppermann #define TCPSTATES /* for logging */ 84773673c1SAndre Oppermann 85dbc42409SLawrence Stewart #include <netinet/cc.h> 86df8bae1dSRodney W. Grimes #include <netinet/in.h> 8757f60867SMark Johnston #include <netinet/in_kdtrace.h> 88960ed29cSSeigo Tanimura #include <netinet/in_pcb.h> 89df8bae1dSRodney W. Grimes #include <netinet/in_systm.h> 90960ed29cSSeigo Tanimura #include <netinet/in_var.h> 91df8bae1dSRodney W. Grimes #include <netinet/ip.h> 928e8aab7aSAndre Oppermann #include <netinet/ip_icmp.h> /* required for icmp_var.h */ 93686cdd19SJun-ichiro itojun Hagino #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */ 94df8bae1dSRodney W. Grimes #include <netinet/ip_var.h> 95ef39adf0SAndre Oppermann #include <netinet/ip_options.h> 96686cdd19SJun-ichiro itojun Hagino #include <netinet/ip6.h> 97686cdd19SJun-ichiro itojun Hagino #include <netinet/icmp6.h> 98686cdd19SJun-ichiro itojun Hagino #include <netinet6/in6_pcb.h> 99960ed29cSSeigo Tanimura #include <netinet6/ip6_var.h> 100960ed29cSSeigo Tanimura #include <netinet6/nd6.h> 101df8bae1dSRodney W. Grimes #include <netinet/tcp_fsm.h> 102df8bae1dSRodney W. Grimes #include <netinet/tcp_seq.h> 103df8bae1dSRodney W. Grimes #include <netinet/tcp_timer.h> 104df8bae1dSRodney W. Grimes #include <netinet/tcp_var.h> 105fb59c426SYoshinobu Inoue #include <netinet6/tcp6_var.h> 106df8bae1dSRodney W. Grimes #include <netinet/tcpip.h> 107c325962bSMike Silbersack #include <netinet/tcp_syncache.h> 108610ee2f9SDavid Greenman #ifdef TCPDEBUG 109df8bae1dSRodney W. Grimes #include <netinet/tcp_debug.h> 110fb59c426SYoshinobu Inoue #endif /* TCPDEBUG */ 11109fe6320SNavdeep Parhar #ifdef TCP_OFFLOAD 11209fe6320SNavdeep Parhar #include <netinet/tcp_offload.h> 11309fe6320SNavdeep Parhar #endif 114fb59c426SYoshinobu Inoue 115b2630c29SGeorge V. Neville-Neil #ifdef IPSEC 116b9234fafSSam Leffler #include <netipsec/ipsec.h> 117b9234fafSSam Leffler #include <netipsec/ipsec6.h> 118b2630c29SGeorge V. Neville-Neil #endif /*IPSEC*/ 119b9234fafSSam Leffler 120db4f9cc7SJonathan Lemon #include <machine/in_cksum.h> 121db4f9cc7SJonathan Lemon 122aed55708SRobert Watson #include <security/mac/mac_framework.h> 123aed55708SRobert Watson 124dbc42409SLawrence Stewart const int tcprexmtthresh = 3; 1250312fbe9SPoul-Henning Kamp 126773673c1SAndre Oppermann int tcp_log_in_vain = 0; 127816a3d83SPoul-Henning Kamp SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW, 128eddfbb76SRobert Watson &tcp_log_in_vain, 0, 129eddfbb76SRobert Watson "Log all incoming TCP segments to closed ports"); 130816a3d83SPoul-Henning Kamp 13182cea7e6SBjoern A. Zeeb VNET_DEFINE(int, blackhole) = 0; 13282cea7e6SBjoern A. Zeeb #define V_blackhole VNET(blackhole) 133eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW, 134eddfbb76SRobert Watson &VNET_NAME(blackhole), 0, 135eddfbb76SRobert Watson "Do not send RST on segments to closed ports"); 13616f7f31fSGeoff Rehmet 13782cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_delack_enabled) = 1; 138eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW, 139eddfbb76SRobert Watson &VNET_NAME(tcp_delack_enabled), 0, 1403d177f46SBill Fumerola "Delay ACK to try and piggyback it onto a data packet"); 141f498eeeeSDavid Greenman 14282cea7e6SBjoern A. Zeeb VNET_DEFINE(int, drop_synfin) = 0; 14382cea7e6SBjoern A. Zeeb #define V_drop_synfin VNET(drop_synfin) 144eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW, 145eddfbb76SRobert Watson &VNET_NAME(drop_synfin), 0, 146eddfbb76SRobert Watson "Drop TCP packets with SYN+FIN set"); 147f8613305SDag-Erling Smørgrav 14882cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_do_rfc3042) = 1; 14982cea7e6SBjoern A. Zeeb #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042) 150eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW, 151eddfbb76SRobert Watson &VNET_NAME(tcp_do_rfc3042), 0, 152eddfbb76SRobert Watson "Enable RFC 3042 (Limited Transmit)"); 153582a954bSJeffrey Hsu 15482cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_do_rfc3390) = 1; 155eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW, 156eddfbb76SRobert Watson &VNET_NAME(tcp_do_rfc3390), 0, 157da3a8a1aSJeffrey Hsu "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); 158da3a8a1aSJeffrey Hsu 15909440655SAndre Oppermann SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0, 16009440655SAndre Oppermann "Experimental TCP extensions"); 16109440655SAndre Oppermann 16209440655SAndre Oppermann VNET_DEFINE(int, tcp_do_initcwnd10) = 1; 16309440655SAndre Oppermann SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW, 16409440655SAndre Oppermann &VNET_NAME(tcp_do_initcwnd10), 0, 165b8b4cfcdSSergey Kandaurov "Enable RFC 6928 (Increasing initial CWND to 10)"); 16609440655SAndre Oppermann 16782cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_do_rfc3465) = 1; 168eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW, 169eddfbb76SRobert Watson &VNET_NAME(tcp_do_rfc3465), 0, 17024cb0f22SLawrence Stewart "Enable RFC 3465 (Appropriate Byte Counting)"); 171eddfbb76SRobert Watson 17282cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_abc_l_var) = 2; 173eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW, 174eddfbb76SRobert Watson &VNET_NAME(tcp_abc_l_var), 2, 17524cb0f22SLawrence Stewart "Cap the max cwnd increment during slow-start to this number of segments"); 17624cb0f22SLawrence Stewart 1776472ac3dSEd Schouten static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN"); 178f2512ba1SRui Paulo 17982cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_do_ecn) = 0; 180eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW, 181eddfbb76SRobert Watson &VNET_NAME(tcp_do_ecn), 0, 182eddfbb76SRobert Watson "TCP ECN support"); 183eddfbb76SRobert Watson 18482cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_ecn_maxretries) = 1; 185eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW, 186eddfbb76SRobert Watson &VNET_NAME(tcp_ecn_maxretries), 0, 187eddfbb76SRobert Watson "Max retries before giving up on ECN"); 188eddfbb76SRobert Watson 18982cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_insecure_rst) = 0; 19082cea7e6SBjoern A. Zeeb #define V_tcp_insecure_rst VNET(tcp_insecure_rst) 191eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW, 192eddfbb76SRobert Watson &VNET_NAME(tcp_insecure_rst), 0, 1936489fe65SAndre Oppermann "Follow the old (insecure) criteria for accepting RST packets"); 194a69968eeSMike Silbersack 195fba0cea1SBjoern A. Zeeb VNET_DEFINE(int, tcp_recvspace) = 1024*64; 196873789cbSAndre Oppermann #define V_tcp_recvspace VNET(tcp_recvspace) 197ddd0c4a9SSergey Kandaurov SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW, 198873789cbSAndre Oppermann &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); 199873789cbSAndre Oppermann 20082cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; 20182cea7e6SBjoern A. Zeeb #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf) 202eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW, 203eddfbb76SRobert Watson &VNET_NAME(tcp_do_autorcvbuf), 0, 2048b615593SMarko Zec "Enable automatic receive buffer sizing"); 2056741ecf5SAndre Oppermann 20682cea7e6SBjoern A. Zeeb VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024; 20782cea7e6SBjoern A. Zeeb #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc) 208eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW, 209eddfbb76SRobert Watson &VNET_NAME(tcp_autorcvbuf_inc), 0, 2106489fe65SAndre Oppermann "Incrementor step size of automatic receive buffer"); 2116741ecf5SAndre Oppermann 212b233773bSBjoern A. Zeeb VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; 21382cea7e6SBjoern A. Zeeb #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max) 214eddfbb76SRobert Watson SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW, 215eddfbb76SRobert Watson &VNET_NAME(tcp_autorcvbuf_max), 0, 2168b615593SMarko Zec "Max size of automatic receive buffer"); 2176741ecf5SAndre Oppermann 218eddfbb76SRobert Watson VNET_DEFINE(struct inpcbhead, tcb); 21944e33a07SMarko Zec #define tcb6 tcb /* for KAME src sync over BSD*'s */ 22082cea7e6SBjoern A. Zeeb VNET_DEFINE(struct inpcbinfo, tcbinfo); 221df8bae1dSRodney W. Grimes 2225a53ca16SPaul Saab static void tcp_dooptions(struct tcpopt *, u_char *, int, int); 223679d9708SAndre Oppermann static void tcp_do_segment(struct mbuf *, struct tcphdr *, 224252ca428SRobert Watson struct socket *, struct tcpcb *, int, int, uint8_t, 225252ca428SRobert Watson int); 226302ce8d6SAndre Oppermann static void tcp_dropwithreset(struct mbuf *, struct tcphdr *, 227302ce8d6SAndre Oppermann struct tcpcb *, int, int); 2284d77a549SAlfred Perlstein static void tcp_pulloutofband(struct socket *, 2294d77a549SAlfred Perlstein struct tcphdr *, struct mbuf *, int); 2304d77a549SAlfred Perlstein static void tcp_xmit_timer(struct tcpcb *, int); 231c068736aSJeffrey Hsu static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *); 232dbc42409SLawrence Stewart static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th, 233dbc42409SLawrence Stewart uint16_t type); 234dbc42409SLawrence Stewart static void inline cc_conn_init(struct tcpcb *tp); 235dbc42409SLawrence Stewart static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th); 23639bc9de5SLawrence Stewart static void inline hhook_run_tcp_est_in(struct tcpcb *tp, 23739bc9de5SLawrence Stewart struct tcphdr *th, struct tcpopt *to); 238f2512ba1SRui Paulo 239315e3e38SRobert Watson /* 2405da0521fSAndrey V. Elsukov * TCP statistics are stored in an "array" of counter(9)s. 2415923c293SGleb Smirnoff */ 2425da0521fSAndrey V. Elsukov VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); 2435da0521fSAndrey V. Elsukov VNET_PCPUSTAT_SYSINIT(tcpstat); 2445da0521fSAndrey V. Elsukov SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, 2455da0521fSAndrey V. Elsukov tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); 2465923c293SGleb Smirnoff 2475923c293SGleb Smirnoff #ifdef VIMAGE 2485da0521fSAndrey V. Elsukov VNET_PCPUSTAT_SYSUNINIT(tcpstat); 2495923c293SGleb Smirnoff #endif /* VIMAGE */ 2505923c293SGleb Smirnoff /* 251315e3e38SRobert Watson * Kernel module interface for updating tcpstat. The argument is an index 2525923c293SGleb Smirnoff * into tcpstat treated as an array. 253315e3e38SRobert Watson */ 254315e3e38SRobert Watson void 255315e3e38SRobert Watson kmod_tcpstat_inc(int statnum) 256315e3e38SRobert Watson { 257315e3e38SRobert Watson 2585da0521fSAndrey V. Elsukov counter_u64_add(VNET(tcpstat)[statnum], 1); 259315e3e38SRobert Watson } 260315e3e38SRobert Watson 261dbc42409SLawrence Stewart /* 26239bc9de5SLawrence Stewart * Wrapper for the TCP established input helper hook. 26339bc9de5SLawrence Stewart */ 26439bc9de5SLawrence Stewart static void inline 26539bc9de5SLawrence Stewart hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) 26639bc9de5SLawrence Stewart { 26739bc9de5SLawrence Stewart struct tcp_hhook_data hhook_data; 26839bc9de5SLawrence Stewart 26939bc9de5SLawrence Stewart if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { 27039bc9de5SLawrence Stewart hhook_data.tp = tp; 27139bc9de5SLawrence Stewart hhook_data.th = th; 27239bc9de5SLawrence Stewart hhook_data.to = to; 27339bc9de5SLawrence Stewart 27439bc9de5SLawrence Stewart hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, 27539bc9de5SLawrence Stewart tp->osd); 27639bc9de5SLawrence Stewart } 27739bc9de5SLawrence Stewart } 27839bc9de5SLawrence Stewart 27939bc9de5SLawrence Stewart /* 280dbc42409SLawrence Stewart * CC wrapper hook functions 281dbc42409SLawrence Stewart */ 282f2512ba1SRui Paulo static void inline 283dbc42409SLawrence Stewart cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type) 284f2512ba1SRui Paulo { 285dbc42409SLawrence Stewart INP_WLOCK_ASSERT(tp->t_inpcb); 286f2512ba1SRui Paulo 287dbc42409SLawrence Stewart tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th); 2885b648e79SLawrence Stewart if (tp->snd_cwnd <= tp->snd_wnd) 289dbc42409SLawrence Stewart tp->ccv->flags |= CCF_CWND_LIMITED; 290dbc42409SLawrence Stewart else 291dbc42409SLawrence Stewart tp->ccv->flags &= ~CCF_CWND_LIMITED; 292dbc42409SLawrence Stewart 293dbc42409SLawrence Stewart if (type == CC_ACK) { 294dbc42409SLawrence Stewart if (tp->snd_cwnd > tp->snd_ssthresh) { 295dbc42409SLawrence Stewart tp->t_bytes_acked += min(tp->ccv->bytes_this_ack, 296dbc42409SLawrence Stewart V_tcp_abc_l_var * tp->t_maxseg); 297dbc42409SLawrence Stewart if (tp->t_bytes_acked >= tp->snd_cwnd) { 298dbc42409SLawrence Stewart tp->t_bytes_acked -= tp->snd_cwnd; 299dbc42409SLawrence Stewart tp->ccv->flags |= CCF_ABC_SENTAWND; 300dbc42409SLawrence Stewart } 301dbc42409SLawrence Stewart } else { 302dbc42409SLawrence Stewart tp->ccv->flags &= ~CCF_ABC_SENTAWND; 303dbc42409SLawrence Stewart tp->t_bytes_acked = 0; 304dbc42409SLawrence Stewart } 305dbc42409SLawrence Stewart } 306dbc42409SLawrence Stewart 307dbc42409SLawrence Stewart if (CC_ALGO(tp)->ack_received != NULL) { 308dbc42409SLawrence Stewart /* XXXLAS: Find a way to live without this */ 309dbc42409SLawrence Stewart tp->ccv->curack = th->th_ack; 310dbc42409SLawrence Stewart CC_ALGO(tp)->ack_received(tp->ccv, type); 311dbc42409SLawrence Stewart } 312dbc42409SLawrence Stewart } 313dbc42409SLawrence Stewart 314dbc42409SLawrence Stewart static void inline 315dbc42409SLawrence Stewart cc_conn_init(struct tcpcb *tp) 316dbc42409SLawrence Stewart { 317dbc42409SLawrence Stewart struct hc_metrics_lite metrics; 318dbc42409SLawrence Stewart struct inpcb *inp = tp->t_inpcb; 319dbc42409SLawrence Stewart int rtt; 320dbc42409SLawrence Stewart 321dbc42409SLawrence Stewart INP_WLOCK_ASSERT(tp->t_inpcb); 322dbc42409SLawrence Stewart 323dbc42409SLawrence Stewart tcp_hc_get(&inp->inp_inc, &metrics); 324dbc42409SLawrence Stewart 325dbc42409SLawrence Stewart if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { 326dbc42409SLawrence Stewart tp->t_srtt = rtt; 327dbc42409SLawrence Stewart tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE; 328dbc42409SLawrence Stewart TCPSTAT_INC(tcps_usedrtt); 329dbc42409SLawrence Stewart if (metrics.rmx_rttvar) { 330dbc42409SLawrence Stewart tp->t_rttvar = metrics.rmx_rttvar; 331dbc42409SLawrence Stewart TCPSTAT_INC(tcps_usedrttvar); 332dbc42409SLawrence Stewart } else { 333dbc42409SLawrence Stewart /* default variation is +- 1 rtt */ 334dbc42409SLawrence Stewart tp->t_rttvar = 335dbc42409SLawrence Stewart tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; 336dbc42409SLawrence Stewart } 337dbc42409SLawrence Stewart TCPT_RANGESET(tp->t_rxtcur, 338dbc42409SLawrence Stewart ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, 339dbc42409SLawrence Stewart tp->t_rttmin, TCPTV_REXMTMAX); 340dbc42409SLawrence Stewart } 341dbc42409SLawrence Stewart if (metrics.rmx_ssthresh) { 342dbc42409SLawrence Stewart /* 343dbc42409SLawrence Stewart * There's some sort of gateway or interface 344dbc42409SLawrence Stewart * buffer limit on the path. Use this to set 345dbc42409SLawrence Stewart * the slow start threshhold, but set the 346dbc42409SLawrence Stewart * threshold to no less than 2*mss. 347dbc42409SLawrence Stewart */ 348dbc42409SLawrence Stewart tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh); 349dbc42409SLawrence Stewart TCPSTAT_INC(tcps_usedssthresh); 350dbc42409SLawrence Stewart } 351dbc42409SLawrence Stewart 352dbc42409SLawrence Stewart /* 3539ec4a4ccSAndre Oppermann * Set the initial slow-start flight size. 354dbc42409SLawrence Stewart * 355cf8f04f4SAndre Oppermann * RFC5681 Section 3.1 specifies the default conservative values. 356cf8f04f4SAndre Oppermann * RFC3390 specifies slightly more aggressive values. 357b8b4cfcdSSergey Kandaurov * RFC6928 increases it to ten segments. 358cf8f04f4SAndre Oppermann * 359cf8f04f4SAndre Oppermann * If a SYN or SYN/ACK was lost and retransmitted, we have to 360cf8f04f4SAndre Oppermann * reduce the initial CWND to one segment as congestion is likely 361cf8f04f4SAndre Oppermann * requiring us to be cautious. 362dbc42409SLawrence Stewart */ 363cf8f04f4SAndre Oppermann if (tp->snd_cwnd == 1) 364cf8f04f4SAndre Oppermann tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */ 36509440655SAndre Oppermann else if (V_tcp_do_initcwnd10) 36609440655SAndre Oppermann tp->snd_cwnd = min(10 * tp->t_maxseg, 36709440655SAndre Oppermann max(2 * tp->t_maxseg, 14600)); 368cf8f04f4SAndre Oppermann else if (V_tcp_do_rfc3390) 369dbc42409SLawrence Stewart tp->snd_cwnd = min(4 * tp->t_maxseg, 370dbc42409SLawrence Stewart max(2 * tp->t_maxseg, 4380)); 37122efabd4SAndre Oppermann else { 37222efabd4SAndre Oppermann /* Per RFC5681 Section 3.1 */ 37322efabd4SAndre Oppermann if (tp->t_maxseg > 2190) 37422efabd4SAndre Oppermann tp->snd_cwnd = 2 * tp->t_maxseg; 37522efabd4SAndre Oppermann else if (tp->t_maxseg > 1095) 37622efabd4SAndre Oppermann tp->snd_cwnd = 3 * tp->t_maxseg; 377dbc42409SLawrence Stewart else 37822efabd4SAndre Oppermann tp->snd_cwnd = 4 * tp->t_maxseg; 37922efabd4SAndre Oppermann } 380dbc42409SLawrence Stewart 381dbc42409SLawrence Stewart if (CC_ALGO(tp)->conn_init != NULL) 382dbc42409SLawrence Stewart CC_ALGO(tp)->conn_init(tp->ccv); 383dbc42409SLawrence Stewart } 384dbc42409SLawrence Stewart 385dbc42409SLawrence Stewart void inline 386dbc42409SLawrence Stewart cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) 387dbc42409SLawrence Stewart { 388dbc42409SLawrence Stewart INP_WLOCK_ASSERT(tp->t_inpcb); 389dbc42409SLawrence Stewart 390dbc42409SLawrence Stewart switch(type) { 391dbc42409SLawrence Stewart case CC_NDUPACK: 392dbc42409SLawrence Stewart if (!IN_FASTRECOVERY(tp->t_flags)) { 393f2512ba1SRui Paulo tp->snd_recover = tp->snd_max; 394f2512ba1SRui Paulo if (tp->t_flags & TF_ECN_PERMIT) 395f2512ba1SRui Paulo tp->t_flags |= TF_ECN_SND_CWR; 396f2512ba1SRui Paulo } 397dbc42409SLawrence Stewart break; 398dbc42409SLawrence Stewart case CC_ECN: 399dbc42409SLawrence Stewart if (!IN_CONGRECOVERY(tp->t_flags)) { 400dbc42409SLawrence Stewart TCPSTAT_INC(tcps_ecn_rcwnd); 401dbc42409SLawrence Stewart tp->snd_recover = tp->snd_max; 402dbc42409SLawrence Stewart if (tp->t_flags & TF_ECN_PERMIT) 403dbc42409SLawrence Stewart tp->t_flags |= TF_ECN_SND_CWR; 404dbc42409SLawrence Stewart } 405dbc42409SLawrence Stewart break; 406dbc42409SLawrence Stewart case CC_RTO: 407dbc42409SLawrence Stewart tp->t_dupacks = 0; 408dbc42409SLawrence Stewart tp->t_bytes_acked = 0; 409dbc42409SLawrence Stewart EXIT_RECOVERY(tp->t_flags); 4106157935fSLawrence Stewart tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 / 4116157935fSLawrence Stewart tp->t_maxseg) * tp->t_maxseg; 412dbc42409SLawrence Stewart tp->snd_cwnd = tp->t_maxseg; 413dbc42409SLawrence Stewart break; 414dbc42409SLawrence Stewart case CC_RTO_ERR: 415dbc42409SLawrence Stewart TCPSTAT_INC(tcps_sndrexmitbad); 416dbc42409SLawrence Stewart /* RTO was unnecessary, so reset everything. */ 417dbc42409SLawrence Stewart tp->snd_cwnd = tp->snd_cwnd_prev; 418dbc42409SLawrence Stewart tp->snd_ssthresh = tp->snd_ssthresh_prev; 419dbc42409SLawrence Stewart tp->snd_recover = tp->snd_recover_prev; 420dbc42409SLawrence Stewart if (tp->t_flags & TF_WASFRECOVERY) 421dbc42409SLawrence Stewart ENTER_FASTRECOVERY(tp->t_flags); 422dbc42409SLawrence Stewart if (tp->t_flags & TF_WASCRECOVERY) 423dbc42409SLawrence Stewart ENTER_CONGRECOVERY(tp->t_flags); 424dbc42409SLawrence Stewart tp->snd_nxt = tp->snd_max; 425672dc4aeSJohn Baldwin tp->t_flags &= ~TF_PREVVALID; 426dbc42409SLawrence Stewart tp->t_badrxtwin = 0; 427dbc42409SLawrence Stewart break; 428dbc42409SLawrence Stewart } 429dbc42409SLawrence Stewart 430dbc42409SLawrence Stewart if (CC_ALGO(tp)->cong_signal != NULL) { 431dbc42409SLawrence Stewart if (th != NULL) 432dbc42409SLawrence Stewart tp->ccv->curack = th->th_ack; 433dbc42409SLawrence Stewart CC_ALGO(tp)->cong_signal(tp->ccv, type); 434dbc42409SLawrence Stewart } 435dbc42409SLawrence Stewart } 436dbc42409SLawrence Stewart 437dbc42409SLawrence Stewart static void inline 438dbc42409SLawrence Stewart cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) 439dbc42409SLawrence Stewart { 440dbc42409SLawrence Stewart INP_WLOCK_ASSERT(tp->t_inpcb); 441dbc42409SLawrence Stewart 442dbc42409SLawrence Stewart /* XXXLAS: KASSERT that we're in recovery? */ 443dbc42409SLawrence Stewart 444dbc42409SLawrence Stewart if (CC_ALGO(tp)->post_recovery != NULL) { 445dbc42409SLawrence Stewart tp->ccv->curack = th->th_ack; 446dbc42409SLawrence Stewart CC_ALGO(tp)->post_recovery(tp->ccv); 447dbc42409SLawrence Stewart } 448dbc42409SLawrence Stewart /* XXXLAS: EXIT_RECOVERY ? */ 449dbc42409SLawrence Stewart tp->t_bytes_acked = 0; 450dbc42409SLawrence Stewart } 4510312fbe9SPoul-Henning Kamp 4522903309aSAttilio Rao #ifdef TCP_SIGNATURE 4532903309aSAttilio Rao static inline int 4542903309aSAttilio Rao tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen, 4552903309aSAttilio Rao struct tcpopt *to, struct tcphdr *th, u_int tcpbflag) 4562903309aSAttilio Rao { 4572903309aSAttilio Rao int ret; 4582903309aSAttilio Rao 4592903309aSAttilio Rao tcp_fields_to_net(th); 4602903309aSAttilio Rao ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag); 4612903309aSAttilio Rao tcp_fields_to_host(th); 4622903309aSAttilio Rao return (ret); 4632903309aSAttilio Rao } 4642903309aSAttilio Rao #endif 4652903309aSAttilio Rao 466fb59c426SYoshinobu Inoue /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */ 467fb59c426SYoshinobu Inoue #ifdef INET6 468fb59c426SYoshinobu Inoue #define ND6_HINT(tp) \ 469fb59c426SYoshinobu Inoue do { \ 470fb59c426SYoshinobu Inoue if ((tp) && (tp)->t_inpcb && \ 47197d8d152SAndre Oppermann ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \ 47297d8d152SAndre Oppermann nd6_nud_hint(NULL, NULL, 0); \ 473fb59c426SYoshinobu Inoue } while (0) 474fb59c426SYoshinobu Inoue #else 475fb59c426SYoshinobu Inoue #define ND6_HINT(tp) 476fb59c426SYoshinobu Inoue #endif 477df8bae1dSRodney W. Grimes 478df8bae1dSRodney W. Grimes /* 479262c1c1aSMatthew Dillon * Indicate whether this ack should be delayed. We can delay the ack if 48085536381SHiren Panchasara * following conditions are met: 48185536381SHiren Panchasara * - There is no delayed ack timer in progress. 48285536381SHiren Panchasara * - Our last ack wasn't a 0-sized window. We never want to delay 48385536381SHiren Panchasara * the ack that opens up a 0-sized window. 48485536381SHiren Panchasara * - LRO wasn't used for this segment. We make sure by checking that the 48585536381SHiren Panchasara * segment size is not larger than the MSS. 48685536381SHiren Panchasara * - Delayed acks are enabled or this is a half-synchronized T/TCP 48785536381SHiren Panchasara * connection. 488d8c85a26SJonathan Lemon */ 489c1e5a6e5SAndre Oppermann #define DELAY_ACK(tp, tlen) \ 490b8152ba7SAndre Oppermann ((!tcp_timer_active(tp, TT_DELACK) && \ 491a14c749fSJonathan Lemon (tp->t_flags & TF_RXWIN0SENT) == 0) && \ 492c1e5a6e5SAndre Oppermann (tlen <= tp->t_maxopd) && \ 493603724d3SBjoern A. Zeeb (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) 494d8c85a26SJonathan Lemon 495df8bae1dSRodney W. Grimes /* 4968d573cc1SAndre Oppermann * TCP input handling is split into multiple parts: 4978d573cc1SAndre Oppermann * tcp6_input is a thin wrapper around tcp_input for the extended 4988d573cc1SAndre Oppermann * ip6_protox[] call format in ip6_input 4998d573cc1SAndre Oppermann * tcp_input handles primary segment validation, inpcb lookup and 5008d573cc1SAndre Oppermann * SYN processing on listen sockets 5018d573cc1SAndre Oppermann * tcp_do_segment processes the ACK and text of the segment for 5028d573cc1SAndre Oppermann * establishing, established and closing connections 503df8bae1dSRodney W. Grimes */ 504fb59c426SYoshinobu Inoue #ifdef INET6 505fb59c426SYoshinobu Inoue int 506ad3f9ab3SAndre Oppermann tcp6_input(struct mbuf **mp, int *offp, int proto) 507fb59c426SYoshinobu Inoue { 508ad3f9ab3SAndre Oppermann struct mbuf *m = *mp; 50933841545SHajimu UMEMOTO struct in6_ifaddr *ia6; 510fb59c426SYoshinobu Inoue 511fb59c426SYoshinobu Inoue IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE); 512fb59c426SYoshinobu Inoue 513fb59c426SYoshinobu Inoue /* 514fb59c426SYoshinobu Inoue * draft-itojun-ipv6-tcp-to-anycast 515fb59c426SYoshinobu Inoue * better place to put this in? 516fb59c426SYoshinobu Inoue */ 51733841545SHajimu UMEMOTO ia6 = ip6_getdstifaddr(m); 51833841545SHajimu UMEMOTO if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { 519fb59c426SYoshinobu Inoue struct ip6_hdr *ip6; 520fb59c426SYoshinobu Inoue 5218c0fec80SRobert Watson ifa_free(&ia6->ia_ifa); 522fb59c426SYoshinobu Inoue ip6 = mtod(m, struct ip6_hdr *); 523fb59c426SYoshinobu Inoue icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, 524fb59c426SYoshinobu Inoue (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); 5258f5a8818SKevin Lo return (IPPROTO_DONE); 526fb59c426SYoshinobu Inoue } 52777d396fdSMaksim Yevmenkin if (ia6) 52877d396fdSMaksim Yevmenkin ifa_free(&ia6->ia_ifa); 529fb59c426SYoshinobu Inoue 5308f5a8818SKevin Lo return (tcp_input(mp, offp, proto)); 531fb59c426SYoshinobu Inoue } 532b287c6c7SBjoern A. Zeeb #endif /* INET6 */ 533fb59c426SYoshinobu Inoue 5348f5a8818SKevin Lo int 5358f5a8818SKevin Lo tcp_input(struct mbuf **mp, int *offp, int proto) 536df8bae1dSRodney W. Grimes { 5378f5a8818SKevin Lo struct mbuf *m = *mp; 538b287c6c7SBjoern A. Zeeb struct tcphdr *th = NULL; 539ad3f9ab3SAndre Oppermann struct ip *ip = NULL; 540ad3f9ab3SAndre Oppermann struct inpcb *inp = NULL; 541302ce8d6SAndre Oppermann struct tcpcb *tp = NULL; 542302ce8d6SAndre Oppermann struct socket *so = NULL; 543e79adb8eSGarrett Wollman u_char *optp = NULL; 5448f5a8818SKevin Lo int off0; 54526f9a767SRodney W. Grimes int optlen = 0; 546b287c6c7SBjoern A. Zeeb #ifdef INET 547b287c6c7SBjoern A. Zeeb int len; 548b287c6c7SBjoern A. Zeeb #endif 549b287c6c7SBjoern A. Zeeb int tlen = 0, off; 550fb59c426SYoshinobu Inoue int drop_hdrlen; 551ad3f9ab3SAndre Oppermann int thflags; 552302ce8d6SAndre Oppermann int rstreason = 0; /* For badport_bandlim accounting purposes */ 5532903309aSAttilio Rao #ifdef TCP_SIGNATURE 5542903309aSAttilio Rao uint8_t sig_checked = 0; 5552903309aSAttilio Rao #endif 556b287c6c7SBjoern A. Zeeb uint8_t iptos = 0; 557c1de64a4SAndrey V. Elsukov struct m_tag *fwd_tag = NULL; 558c068736aSJeffrey Hsu #ifdef INET6 559302ce8d6SAndre Oppermann struct ip6_hdr *ip6 = NULL; 560c068736aSJeffrey Hsu int isipv6; 561c068736aSJeffrey Hsu #else 562a160e630SAndre Oppermann const void *ip6 = NULL; 563b287c6c7SBjoern A. Zeeb #endif /* INET6 */ 564302ce8d6SAndre Oppermann struct tcpopt to; /* options in this segment */ 565a160e630SAndre Oppermann char *s = NULL; /* address and port logging */ 566252ca428SRobert Watson int ti_locked; 567252ca428SRobert Watson #define TI_UNLOCKED 1 568fa046d87SRobert Watson #define TI_WLOCKED 2 569f76fcf6dSJeffrey Hsu 570610ee2f9SDavid Greenman #ifdef TCPDEBUG 571c068736aSJeffrey Hsu /* 572c068736aSJeffrey Hsu * The size of tcp_saveipgen must be the size of the max ip header, 573c068736aSJeffrey Hsu * now IPv6. 574c068736aSJeffrey Hsu */ 57514739780SMaxim Konovalov u_char tcp_saveipgen[IP6_HDR_LEN]; 576410bb1bfSLuigi Rizzo struct tcphdr tcp_savetcp; 577610ee2f9SDavid Greenman short ostate = 0; 578610ee2f9SDavid Greenman #endif 579c068736aSJeffrey Hsu 580fb59c426SYoshinobu Inoue #ifdef INET6 581fb59c426SYoshinobu Inoue isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; 582fb59c426SYoshinobu Inoue #endif 583a0292f23SGarrett Wollman 5848f5a8818SKevin Lo off0 = *offp; 5858f5a8818SKevin Lo m = *mp; 5868f5a8818SKevin Lo *mp = NULL; 587302ce8d6SAndre Oppermann to.to_flags = 0; 58878b50714SRobert Watson TCPSTAT_INC(tcps_rcvtotal); 589fb59c426SYoshinobu Inoue 59004d3a452SHajimu UMEMOTO #ifdef INET6 591b287c6c7SBjoern A. Zeeb if (isipv6) { 5928d573cc1SAndre Oppermann /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */ 59345747ba5SBjoern A. Zeeb 59445747ba5SBjoern A. Zeeb if (m->m_len < (sizeof(*ip6) + sizeof(*th))) { 59545747ba5SBjoern A. Zeeb m = m_pullup(m, sizeof(*ip6) + sizeof(*th)); 59645747ba5SBjoern A. Zeeb if (m == NULL) { 59745747ba5SBjoern A. Zeeb TCPSTAT_INC(tcps_rcvshort); 5988f5a8818SKevin Lo return (IPPROTO_DONE); 59945747ba5SBjoern A. Zeeb } 60045747ba5SBjoern A. Zeeb } 60145747ba5SBjoern A. Zeeb 602fb59c426SYoshinobu Inoue ip6 = mtod(m, struct ip6_hdr *); 60345747ba5SBjoern A. Zeeb th = (struct tcphdr *)((caddr_t)ip6 + off0); 604fb59c426SYoshinobu Inoue tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; 605356ab07eSBjoern A. Zeeb if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { 60645747ba5SBjoern A. Zeeb if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 60745747ba5SBjoern A. Zeeb th->th_sum = m->m_pkthdr.csum_data; 60845747ba5SBjoern A. Zeeb else 60945747ba5SBjoern A. Zeeb th->th_sum = in6_cksum_pseudo(ip6, tlen, 61045747ba5SBjoern A. Zeeb IPPROTO_TCP, m->m_pkthdr.csum_data); 61145747ba5SBjoern A. Zeeb th->th_sum ^= 0xffff; 61245747ba5SBjoern A. Zeeb } else 61345747ba5SBjoern A. Zeeb th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); 61445747ba5SBjoern A. Zeeb if (th->th_sum) { 61578b50714SRobert Watson TCPSTAT_INC(tcps_rcvbadsum); 616fb59c426SYoshinobu Inoue goto drop; 617fb59c426SYoshinobu Inoue } 61833841545SHajimu UMEMOTO 61933841545SHajimu UMEMOTO /* 62033841545SHajimu UMEMOTO * Be proactive about unspecified IPv6 address in source. 62133841545SHajimu UMEMOTO * As we use all-zero to indicate unbounded/unconnected pcb, 62233841545SHajimu UMEMOTO * unspecified IPv6 address can be used to confuse us. 62333841545SHajimu UMEMOTO * 62433841545SHajimu UMEMOTO * Note that packets with unspecified IPv6 destination is 62533841545SHajimu UMEMOTO * already dropped in ip6_input. 62633841545SHajimu UMEMOTO */ 62733841545SHajimu UMEMOTO if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { 62833841545SHajimu UMEMOTO /* XXX stat */ 62933841545SHajimu UMEMOTO goto drop; 63033841545SHajimu UMEMOTO } 631b287c6c7SBjoern A. Zeeb } 63204d3a452SHajimu UMEMOTO #endif 633b287c6c7SBjoern A. Zeeb #if defined(INET) && defined(INET6) 634b287c6c7SBjoern A. Zeeb else 635b287c6c7SBjoern A. Zeeb #endif 636b287c6c7SBjoern A. Zeeb #ifdef INET 637b287c6c7SBjoern A. Zeeb { 638df8bae1dSRodney W. Grimes /* 639df8bae1dSRodney W. Grimes * Get IP and TCP header together in first mbuf. 640df8bae1dSRodney W. Grimes * Note: IP leaves IP header in first mbuf. 641df8bae1dSRodney W. Grimes */ 642fb59c426SYoshinobu Inoue if (off0 > sizeof (struct ip)) { 643105bd211SGleb Smirnoff ip_stripoptions(m); 644fb59c426SYoshinobu Inoue off0 = sizeof(struct ip); 645fb59c426SYoshinobu Inoue } 646df8bae1dSRodney W. Grimes if (m->m_len < sizeof (struct tcpiphdr)) { 6474dfdffe9SAndre Oppermann if ((m = m_pullup(m, sizeof (struct tcpiphdr))) 6484dfdffe9SAndre Oppermann == NULL) { 64978b50714SRobert Watson TCPSTAT_INC(tcps_rcvshort); 6508f5a8818SKevin Lo return (IPPROTO_DONE); 651df8bae1dSRodney W. Grimes } 652df8bae1dSRodney W. Grimes } 653fb59c426SYoshinobu Inoue ip = mtod(m, struct ip *); 654db4f9cc7SJonathan Lemon th = (struct tcphdr *)((caddr_t)ip + off0); 6558ad458a4SGleb Smirnoff tlen = ntohs(ip->ip_len) - off0; 656df8bae1dSRodney W. Grimes 657db4f9cc7SJonathan Lemon if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { 658db4f9cc7SJonathan Lemon if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) 659db4f9cc7SJonathan Lemon th->th_sum = m->m_pkthdr.csum_data; 660db4f9cc7SJonathan Lemon else 661db4f9cc7SJonathan Lemon th->th_sum = in_pseudo(ip->ip_src.s_addr, 662c068736aSJeffrey Hsu ip->ip_dst.s_addr, 6638f134647SGleb Smirnoff htonl(m->m_pkthdr.csum_data + tlen + 664c068736aSJeffrey Hsu IPPROTO_TCP)); 665db4f9cc7SJonathan Lemon th->th_sum ^= 0xffff; 666db4f9cc7SJonathan Lemon } else { 6678f134647SGleb Smirnoff struct ipovly *ipov = (struct ipovly *)ip; 6688f134647SGleb Smirnoff 669df8bae1dSRodney W. Grimes /* 670df8bae1dSRodney W. Grimes * Checksum extended TCP header and data. 671df8bae1dSRodney W. Grimes */ 6728f134647SGleb Smirnoff len = off0 + tlen; 673fb59c426SYoshinobu Inoue bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); 6748f134647SGleb Smirnoff ipov->ih_len = htons(tlen); 675fb59c426SYoshinobu Inoue th->th_sum = in_cksum(m, len); 67657f60867SMark Johnston /* Reset length for SDT probes. */ 67757f60867SMark Johnston ip->ip_len = htons(tlen + off0); 678db4f9cc7SJonathan Lemon } 67957f60867SMark Johnston 680fb59c426SYoshinobu Inoue if (th->th_sum) { 68178b50714SRobert Watson TCPSTAT_INC(tcps_rcvbadsum); 682df8bae1dSRodney W. Grimes goto drop; 683df8bae1dSRodney W. Grimes } 684fb59c426SYoshinobu Inoue /* Re-initialization for later version check */ 685fb59c426SYoshinobu Inoue ip->ip_v = IPVERSION; 686fb59c426SYoshinobu Inoue } 687b287c6c7SBjoern A. Zeeb #endif /* INET */ 688df8bae1dSRodney W. Grimes 689f2512ba1SRui Paulo #ifdef INET6 690f2512ba1SRui Paulo if (isipv6) 691f2512ba1SRui Paulo iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff; 692b287c6c7SBjoern A. Zeeb #endif 693b287c6c7SBjoern A. Zeeb #if defined(INET) && defined(INET6) 694f2512ba1SRui Paulo else 695f2512ba1SRui Paulo #endif 696b287c6c7SBjoern A. Zeeb #ifdef INET 697f2512ba1SRui Paulo iptos = ip->ip_tos; 698b287c6c7SBjoern A. Zeeb #endif 699f2512ba1SRui Paulo 700df8bae1dSRodney W. Grimes /* 701df8bae1dSRodney W. Grimes * Check that TCP offset makes sense, 702df8bae1dSRodney W. Grimes * pull out TCP options and adjust length. XXX 703df8bae1dSRodney W. Grimes */ 704fb59c426SYoshinobu Inoue off = th->th_off << 2; 705df8bae1dSRodney W. Grimes if (off < sizeof (struct tcphdr) || off > tlen) { 70678b50714SRobert Watson TCPSTAT_INC(tcps_rcvbadoff); 707df8bae1dSRodney W. Grimes goto drop; 708df8bae1dSRodney W. Grimes } 709fb59c426SYoshinobu Inoue tlen -= off; /* tlen is used instead of ti->ti_len */ 710df8bae1dSRodney W. Grimes if (off > sizeof (struct tcphdr)) { 71104d3a452SHajimu UMEMOTO #ifdef INET6 712b287c6c7SBjoern A. Zeeb if (isipv6) { 7138f5a8818SKevin Lo IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE); 714fb59c426SYoshinobu Inoue ip6 = mtod(m, struct ip6_hdr *); 715fb59c426SYoshinobu Inoue th = (struct tcphdr *)((caddr_t)ip6 + off0); 716b287c6c7SBjoern A. Zeeb } 71704d3a452SHajimu UMEMOTO #endif 718b287c6c7SBjoern A. Zeeb #if defined(INET) && defined(INET6) 719b287c6c7SBjoern A. Zeeb else 720b287c6c7SBjoern A. Zeeb #endif 721b287c6c7SBjoern A. Zeeb #ifdef INET 722b287c6c7SBjoern A. Zeeb { 723df8bae1dSRodney W. Grimes if (m->m_len < sizeof(struct ip) + off) { 724c068736aSJeffrey Hsu if ((m = m_pullup(m, sizeof (struct ip) + off)) 7254dfdffe9SAndre Oppermann == NULL) { 72678b50714SRobert Watson TCPSTAT_INC(tcps_rcvshort); 7278f5a8818SKevin Lo return (IPPROTO_DONE); 728df8bae1dSRodney W. Grimes } 729fb59c426SYoshinobu Inoue ip = mtod(m, struct ip *); 730fb59c426SYoshinobu Inoue th = (struct tcphdr *)((caddr_t)ip + off0); 731fb59c426SYoshinobu Inoue } 732df8bae1dSRodney W. Grimes } 733b287c6c7SBjoern A. Zeeb #endif 734df8bae1dSRodney W. Grimes optlen = off - sizeof (struct tcphdr); 735fb59c426SYoshinobu Inoue optp = (u_char *)(th + 1); 736df8bae1dSRodney W. Grimes } 737fb59c426SYoshinobu Inoue thflags = th->th_flags; 738df8bae1dSRodney W. Grimes 739e46cd3d4SDag-Erling Smørgrav /* 740df8bae1dSRodney W. Grimes * Convert TCP protocol specific fields to host format. 741df8bae1dSRodney W. Grimes */ 7422903309aSAttilio Rao tcp_fields_to_host(th); 743df8bae1dSRodney W. Grimes 744df8bae1dSRodney W. Grimes /* 745794235b7SAndre Oppermann * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. 746755c1f07SAndras Olah */ 747fb59c426SYoshinobu Inoue drop_hdrlen = off0 + off; 748755c1f07SAndras Olah 749755c1f07SAndras Olah /* 750fa046d87SRobert Watson * Locate pcb for segment; if we're likely to add or remove a 751a7c7f2a7SJohn Baldwin * connection then first acquire pcbinfo lock. There are three cases 752fa046d87SRobert Watson * where we might discover later we need a write lock despite the 753a7c7f2a7SJohn Baldwin * flags: ACKs moving a connection out of the syncache, ACKs for a 754a7c7f2a7SJohn Baldwin * connection in TIMEWAIT and SYNs not targeting a listening socket. 755df8bae1dSRodney W. Grimes */ 756a7c7f2a7SJohn Baldwin if ((thflags & (TH_FIN | TH_RST)) != 0) { 757603724d3SBjoern A. Zeeb INP_INFO_WLOCK(&V_tcbinfo); 758252ca428SRobert Watson ti_locked = TI_WLOCKED; 759fa046d87SRobert Watson } else 760fa046d87SRobert Watson ti_locked = TI_UNLOCKED; 761252ca428SRobert Watson 7628d573cc1SAndre Oppermann /* 7638d573cc1SAndre Oppermann * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. 7648d573cc1SAndre Oppermann */ 765b8056faeSGleb Smirnoff if ( 766b8056faeSGleb Smirnoff #ifdef INET6 767b8056faeSGleb Smirnoff (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) 768b8056faeSGleb Smirnoff #ifdef INET 769b8056faeSGleb Smirnoff || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) 770b8056faeSGleb Smirnoff #endif 771b8056faeSGleb Smirnoff #endif 772b8056faeSGleb Smirnoff #if defined(INET) && !defined(INET6) 773b8056faeSGleb Smirnoff (m->m_flags & M_IP_NEXTHOP) 774b8056faeSGleb Smirnoff #endif 775b8056faeSGleb Smirnoff ) 7769b932e9eSAndre Oppermann fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); 7779b932e9eSAndre Oppermann 778ce7ad664SEd Maste findpcb: 779ce7ad664SEd Maste #ifdef INVARIANTS 780ce7ad664SEd Maste if (ti_locked == TI_WLOCKED) { 781ce7ad664SEd Maste INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 782ce7ad664SEd Maste } else { 783ce7ad664SEd Maste INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 784ce7ad664SEd Maste } 785ce7ad664SEd Maste #endif 7868a006adbSBjoern A. Zeeb #ifdef INET6 7878a006adbSBjoern A. Zeeb if (isipv6 && fwd_tag != NULL) { 7888a006adbSBjoern A. Zeeb struct sockaddr_in6 *next_hop6; 7898a006adbSBjoern A. Zeeb 7908a006adbSBjoern A. Zeeb next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); 7918a006adbSBjoern A. Zeeb /* 7928a006adbSBjoern A. Zeeb * Transparently forwarded. Pretend to be the destination. 7938a006adbSBjoern A. Zeeb * Already got one like this? 7948a006adbSBjoern A. Zeeb */ 7958a006adbSBjoern A. Zeeb inp = in6_pcblookup_mbuf(&V_tcbinfo, 7968a006adbSBjoern A. Zeeb &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, 7978a006adbSBjoern A. Zeeb INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m); 7988a006adbSBjoern A. Zeeb if (!inp) { 7998a006adbSBjoern A. Zeeb /* 8008a006adbSBjoern A. Zeeb * It's new. Try to find the ambushing socket. 8018a006adbSBjoern A. Zeeb * Because we've rewritten the destination address, 8028a006adbSBjoern A. Zeeb * any hardware-generated hash is ignored. 8038a006adbSBjoern A. Zeeb */ 8048a006adbSBjoern A. Zeeb inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, 8058a006adbSBjoern A. Zeeb th->th_sport, &next_hop6->sin6_addr, 8068a006adbSBjoern A. Zeeb next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : 8078a006adbSBjoern A. Zeeb th->th_dport, INPLOOKUP_WILDCARD | 8088a006adbSBjoern A. Zeeb INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 8098a006adbSBjoern A. Zeeb } 810c1de64a4SAndrey V. Elsukov } else if (isipv6) { 8118a006adbSBjoern A. Zeeb inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, 8128a006adbSBjoern A. Zeeb th->th_sport, &ip6->ip6_dst, th->th_dport, 8138a006adbSBjoern A. Zeeb INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 8148a006adbSBjoern A. Zeeb m->m_pkthdr.rcvif, m); 8158a006adbSBjoern A. Zeeb } 8168a006adbSBjoern A. Zeeb #endif /* INET6 */ 8178a006adbSBjoern A. Zeeb #if defined(INET6) && defined(INET) 8188a006adbSBjoern A. Zeeb else 8198a006adbSBjoern A. Zeeb #endif 8208a006adbSBjoern A. Zeeb #ifdef INET 8218a006adbSBjoern A. Zeeb if (fwd_tag != NULL) { 8229b932e9eSAndre Oppermann struct sockaddr_in *next_hop; 8239b932e9eSAndre Oppermann 8249b932e9eSAndre Oppermann next_hop = (struct sockaddr_in *)(fwd_tag+1); 825f9e354dfSJulian Elischer /* 8262b25acc1SLuigi Rizzo * Transparently forwarded. Pretend to be the destination. 827f9e354dfSJulian Elischer * already got one like this? 828f9e354dfSJulian Elischer */ 829d3c1f003SRobert Watson inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, 830fa046d87SRobert Watson ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB, 831d3c1f003SRobert Watson m->m_pkthdr.rcvif, m); 832fa046d87SRobert Watson if (!inp) { 833fa046d87SRobert Watson /* 834fa046d87SRobert Watson * It's new. Try to find the ambushing socket. 835d3c1f003SRobert Watson * Because we've rewritten the destination address, 836d3c1f003SRobert Watson * any hardware-generated hash is ignored. 837fa046d87SRobert Watson */ 838fa046d87SRobert Watson inp = in_pcblookup(&V_tcbinfo, ip->ip_src, 839fa046d87SRobert Watson th->th_sport, next_hop->sin_addr, 840fa046d87SRobert Watson next_hop->sin_port ? ntohs(next_hop->sin_port) : 841fa046d87SRobert Watson th->th_dport, INPLOOKUP_WILDCARD | 842fa046d87SRobert Watson INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif); 843f9e354dfSJulian Elischer } 844b10fbdeaSAndre Oppermann } else 845d3c1f003SRobert Watson inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, 846fa046d87SRobert Watson th->th_sport, ip->ip_dst, th->th_dport, 847fa046d87SRobert Watson INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB, 848d3c1f003SRobert Watson m->m_pkthdr.rcvif, m); 8498a006adbSBjoern A. Zeeb #endif /* INET */ 850f9e354dfSJulian Elischer 851df8bae1dSRodney W. Grimes /* 852574b6964SAndre Oppermann * If the INPCB does not exist then all data in the incoming 853574b6964SAndre Oppermann * segment is discarded and an appropriate RST is sent back. 8548b07e49aSJulian Elischer * XXX MRT Send RST using which routing table? 855df8bae1dSRodney W. Grimes */ 856816a3d83SPoul-Henning Kamp if (inp == NULL) { 857574b6964SAndre Oppermann /* 858574b6964SAndre Oppermann * Log communication attempts to ports that are not 859574b6964SAndre Oppermann * in use. 860574b6964SAndre Oppermann */ 861574b6964SAndre Oppermann if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) || 862574b6964SAndre Oppermann tcp_log_in_vain == 2) { 863b7d747ecSAndre Oppermann if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) 864df541e5fSAndre Oppermann log(LOG_INFO, "%s; %s: Connection attempt " 865df541e5fSAndre Oppermann "to closed port\n", s, __func__); 8662e4e1b4cSGeoff Rehmet } 867574b6964SAndre Oppermann /* 868574b6964SAndre Oppermann * When blackholing do not respond with a RST but 869574b6964SAndre Oppermann * completely ignore the segment and drop it. 870574b6964SAndre Oppermann */ 871603724d3SBjoern A. Zeeb if ((V_blackhole == 1 && (thflags & TH_SYN)) || 872603724d3SBjoern A. Zeeb V_blackhole == 2) 8731929eae1SAndre Oppermann goto dropunlock; 874574b6964SAndre Oppermann 875a57815efSBosko Milekic rstreason = BANDLIM_RST_CLOSEDPORT; 876a57815efSBosko Milekic goto dropwithreset; 877816a3d83SPoul-Henning Kamp } 878fa046d87SRobert Watson INP_WLOCK_ASSERT(inp); 87980cb9f21SKip Macy if (!(inp->inp_flags & INP_HW_FLOWID) 88080cb9f21SKip Macy && (m->m_flags & M_FLOWID) 88180cb9f21SKip Macy && ((inp->inp_socket == NULL) 88280cb9f21SKip Macy || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) { 88380cb9f21SKip Macy inp->inp_flags |= INP_HW_FLOWID; 88480cb9f21SKip Macy inp->inp_flags &= ~INP_SW_FLOWID; 88580cb9f21SKip Macy inp->inp_flowid = m->m_pkthdr.flowid; 8862f719932SAdrian Chadd inp->inp_flowtype = M_HASHTYPE_GET(m); 88780cb9f21SKip Macy } 888a2589465SKen Smith #ifdef IPSEC 889a2589465SKen Smith #ifdef INET6 890a2589465SKen Smith if (isipv6 && ipsec6_in_reject(m, inp)) { 8916794f460SAndrey V. Elsukov IPSEC6STAT_INC(ips_in_polvio); 892a2589465SKen Smith goto dropunlock; 893a2589465SKen Smith } else 894a2589465SKen Smith #endif /* INET6 */ 895a2589465SKen Smith if (ipsec4_in_reject(m, inp) != 0) { 8966794f460SAndrey V. Elsukov IPSECSTAT_INC(ips_in_polvio); 897a2589465SKen Smith goto dropunlock; 898a2589465SKen Smith } 899a2589465SKen Smith #endif /* IPSEC */ 900a2589465SKen Smith 9018d573cc1SAndre Oppermann /* 9028d573cc1SAndre Oppermann * Check the minimum TTL for socket. 9038d573cc1SAndre Oppermann */ 90434f83c52SGeorge V. Neville-Neil if (inp->inp_ip_minttl != 0) { 90534f83c52SGeorge V. Neville-Neil #ifdef INET6 90634f83c52SGeorge V. Neville-Neil if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim) 907302ce8d6SAndre Oppermann goto dropunlock; 90802707462SAndre Oppermann else 90934f83c52SGeorge V. Neville-Neil #endif 91034f83c52SGeorge V. Neville-Neil if (inp->inp_ip_minttl > ip->ip_ttl) 911302ce8d6SAndre Oppermann goto dropunlock; 91234f83c52SGeorge V. Neville-Neil } 913936cd18dSAndre Oppermann 914340c35deSJonathan Lemon /* 915252ca428SRobert Watson * A previous connection in TIMEWAIT state is supposed to catch stray 916252ca428SRobert Watson * or duplicate segments arriving late. If this segment was a 9170989f56cSRobert Watson * legitimate new connection attempt, the old INPCB gets removed and 918252ca428SRobert Watson * we can try again to find a listening socket. 919252ca428SRobert Watson * 920fa046d87SRobert Watson * At this point, due to earlier optimism, we may hold only an inpcb 921fa046d87SRobert Watson * lock, and not the inpcbinfo write lock. If so, we need to try to 922fa046d87SRobert Watson * acquire it, or if that fails, acquire a reference on the inpcb, 923fa046d87SRobert Watson * drop all locks, acquire a global write lock, and then re-acquire 924fa046d87SRobert Watson * the inpcb lock. We may at that point discover that another thread 925fa046d87SRobert Watson * has tried to free the inpcb, in which case we need to loop back 926fa046d87SRobert Watson * and try to find a new inpcb to deliver to. 927fa046d87SRobert Watson * 928fa046d87SRobert Watson * XXXRW: It may be time to rethink timewait locking. 929340c35deSJonathan Lemon */ 930883e9bc4SRobert Watson relocked: 931ad71fe3cSRobert Watson if (inp->inp_flags & INP_TIMEWAIT) { 932fa046d87SRobert Watson if (ti_locked == TI_UNLOCKED) { 933fa046d87SRobert Watson if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 934252ca428SRobert Watson in_pcbref(inp); 935252ca428SRobert Watson INP_WUNLOCK(inp); 936252ca428SRobert Watson INP_INFO_WLOCK(&V_tcbinfo); 937252ca428SRobert Watson ti_locked = TI_WLOCKED; 938252ca428SRobert Watson INP_WLOCK(inp); 939fa046d87SRobert Watson if (in_pcbrele_wlocked(inp)) { 940252ca428SRobert Watson inp = NULL; 941252ca428SRobert Watson goto findpcb; 942252ca428SRobert Watson } 943f681a5fdSRobert Watson } else 944252ca428SRobert Watson ti_locked = TI_WLOCKED; 945252ca428SRobert Watson } 946252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 947252ca428SRobert Watson 948340c35deSJonathan Lemon if (thflags & TH_SYN) 949f72167f4SAndre Oppermann tcp_dooptions(&to, optp, optlen, TO_SYN); 9508d573cc1SAndre Oppermann /* 9518d573cc1SAndre Oppermann * NB: tcp_twcheck unlocks the INP and frees the mbuf. 9528d573cc1SAndre Oppermann */ 9532104448fSAndre Oppermann if (tcp_twcheck(inp, &to, th, m, tlen)) 954340c35deSJonathan Lemon goto findpcb; 955603724d3SBjoern A. Zeeb INP_INFO_WUNLOCK(&V_tcbinfo); 9568f5a8818SKevin Lo return (IPPROTO_DONE); 957340c35deSJonathan Lemon } 958794235b7SAndre Oppermann /* 959794235b7SAndre Oppermann * The TCPCB may no longer exist if the connection is winding 960794235b7SAndre Oppermann * down or it is in the CLOSED state. Either way we drop the 961794235b7SAndre Oppermann * segment and send an appropriate response. 962794235b7SAndre Oppermann */ 963df8bae1dSRodney W. Grimes tp = intotcpcb(inp); 964a160e630SAndre Oppermann if (tp == NULL || tp->t_state == TCPS_CLOSED) { 965a57815efSBosko Milekic rstreason = BANDLIM_RST_CLOSEDPORT; 966a57815efSBosko Milekic goto dropwithreset; 96709f81a46SBosko Milekic } 968df8bae1dSRodney W. Grimes 96909fe6320SNavdeep Parhar #ifdef TCP_OFFLOAD 97009fe6320SNavdeep Parhar if (tp->t_flags & TF_TOE) { 97109fe6320SNavdeep Parhar tcp_offload_input(tp, m); 97209fe6320SNavdeep Parhar m = NULL; /* consumed by the TOE driver */ 97309fe6320SNavdeep Parhar goto dropunlock; 97409fe6320SNavdeep Parhar } 97509fe6320SNavdeep Parhar #endif 97609fe6320SNavdeep Parhar 977252ca428SRobert Watson /* 978252ca428SRobert Watson * We've identified a valid inpcb, but it could be that we need an 979fa046d87SRobert Watson * inpcbinfo write lock but don't hold it. In this case, attempt to 980fa046d87SRobert Watson * acquire using the same strategy as the TIMEWAIT case above. If we 981fa046d87SRobert Watson * relock, we have to jump back to 'relocked' as the connection might 982fa046d87SRobert Watson * now be in TIMEWAIT. 983252ca428SRobert Watson */ 984fa046d87SRobert Watson #ifdef INVARIANTS 985a7c7f2a7SJohn Baldwin if ((thflags & (TH_FIN | TH_RST)) != 0) 986fa046d87SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 987fa046d87SRobert Watson #endif 988a7c7f2a7SJohn Baldwin if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) || 989a7c7f2a7SJohn Baldwin (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN)))) { 990fa046d87SRobert Watson if (ti_locked == TI_UNLOCKED) { 991fa046d87SRobert Watson if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) { 992252ca428SRobert Watson in_pcbref(inp); 993252ca428SRobert Watson INP_WUNLOCK(inp); 994252ca428SRobert Watson INP_INFO_WLOCK(&V_tcbinfo); 995252ca428SRobert Watson ti_locked = TI_WLOCKED; 996252ca428SRobert Watson INP_WLOCK(inp); 997fa046d87SRobert Watson if (in_pcbrele_wlocked(inp)) { 998252ca428SRobert Watson inp = NULL; 999252ca428SRobert Watson goto findpcb; 1000252ca428SRobert Watson } 1001883e9bc4SRobert Watson goto relocked; 1002f681a5fdSRobert Watson } else 1003252ca428SRobert Watson ti_locked = TI_WLOCKED; 1004252ca428SRobert Watson } 1005252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1006252ca428SRobert Watson } 1007252ca428SRobert Watson 1008c488362eSRobert Watson #ifdef MAC 10098501a69cSRobert Watson INP_WLOCK_ASSERT(inp); 101030d239bcSRobert Watson if (mac_inpcb_check_deliver(inp, m)) 1011302ce8d6SAndre Oppermann goto dropunlock; 1012c488362eSRobert Watson #endif 1013a557af22SRobert Watson so = inp->inp_socket; 1014302ce8d6SAndre Oppermann KASSERT(so != NULL, ("%s: so == NULL", __func__)); 1015610ee2f9SDavid Greenman #ifdef TCPDEBUG 1016df8bae1dSRodney W. Grimes if (so->so_options & SO_DEBUG) { 1017df8bae1dSRodney W. Grimes ostate = tp->t_state; 101810fe523eSMaxim Konovalov #ifdef INET6 10196f697424SBjoern A. Zeeb if (isipv6) { 1020540e8b7eSJeffrey Hsu bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6)); 1021d30d90dcSMaxim Konovalov } else 10226f697424SBjoern A. Zeeb #endif 1023540e8b7eSJeffrey Hsu bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip)); 1024fb59c426SYoshinobu Inoue tcp_savetcp = *th; 1025df8bae1dSRodney W. Grimes } 1026b287c6c7SBjoern A. Zeeb #endif /* TCPDEBUG */ 1027db33b3e6SAndre Oppermann /* 1028db33b3e6SAndre Oppermann * When the socket is accepting connections (the INPCB is in LISTEN 1029db33b3e6SAndre Oppermann * state) we look into the SYN cache if this is a new connection 1030a7c7f2a7SJohn Baldwin * attempt or the completion of a previous one. 1031db33b3e6SAndre Oppermann */ 1032540e8b7eSJeffrey Hsu if (so->so_options & SO_ACCEPTCONN) { 1033540e8b7eSJeffrey Hsu struct in_conninfo inc; 1034540e8b7eSJeffrey Hsu 10357824d002SAndre Oppermann KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but " 10367824d002SAndre Oppermann "tp not listening", __func__)); 10378bfb1918SAndre Oppermann bzero(&inc, sizeof(inc)); 1038302ce8d6SAndre Oppermann #ifdef INET6 1039be2ac88cSJonathan Lemon if (isipv6) { 1040dcdb4371SBjoern A. Zeeb inc.inc_flags |= INC_ISIPV6; 1041be2ac88cSJonathan Lemon inc.inc6_faddr = ip6->ip6_src; 1042be2ac88cSJonathan Lemon inc.inc6_laddr = ip6->ip6_dst; 1043302ce8d6SAndre Oppermann } else 1044302ce8d6SAndre Oppermann #endif 1045302ce8d6SAndre Oppermann { 1046be2ac88cSJonathan Lemon inc.inc_faddr = ip->ip_src; 1047be2ac88cSJonathan Lemon inc.inc_laddr = ip->ip_dst; 1048be2ac88cSJonathan Lemon } 1049be2ac88cSJonathan Lemon inc.inc_fport = th->th_sport; 1050be2ac88cSJonathan Lemon inc.inc_lport = th->th_dport; 10517973fba3SJulian Elischer inc.inc_fibnum = so->so_fibnum; 1052be2ac88cSJonathan Lemon 1053fb59c426SYoshinobu Inoue /* 10548d573cc1SAndre Oppermann * Check for an existing connection attempt in syncache if 10558d573cc1SAndre Oppermann * the flag is only ACK. A successful lookup creates a new 10568d573cc1SAndre Oppermann * socket appended to the listen queue in SYN_RECEIVED state. 1057fb59c426SYoshinobu Inoue */ 1058be2ac88cSJonathan Lemon if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { 1059a7c7f2a7SJohn Baldwin 1060a7c7f2a7SJohn Baldwin INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1061bf6d304aSAndre Oppermann /* 1062bf6d304aSAndre Oppermann * Parse the TCP options here because 1063bf6d304aSAndre Oppermann * syncookies need access to the reflected 1064bf6d304aSAndre Oppermann * timestamp. 1065bf6d304aSAndre Oppermann */ 1066bf6d304aSAndre Oppermann tcp_dooptions(&to, optp, optlen, 0); 10679fa198beSAndre Oppermann /* 10689fa198beSAndre Oppermann * NB: syncache_expand() doesn't unlock 10699fa198beSAndre Oppermann * inp and tcpinfo locks. 10709fa198beSAndre Oppermann */ 1071bf6d304aSAndre Oppermann if (!syncache_expand(&inc, &to, th, &so, m)) { 10724195b4afSPaul Traina /* 1073e207f800SAndre Oppermann * No syncache entry or ACK was not 1074be2ac88cSJonathan Lemon * for our SYN/ACK. Send a RST. 10758d573cc1SAndre Oppermann * NB: syncache did its own logging 10768d573cc1SAndre Oppermann * of the failure cause. 10774195b4afSPaul Traina */ 1078be2ac88cSJonathan Lemon rstreason = BANDLIM_RST_OPENPORT; 1079be2ac88cSJonathan Lemon goto dropwithreset; 1080be2ac88cSJonathan Lemon } 1081f76fcf6dSJeffrey Hsu if (so == NULL) { 1082be2ac88cSJonathan Lemon /* 1083e207f800SAndre Oppermann * We completed the 3-way handshake 1084e207f800SAndre Oppermann * but could not allocate a socket 1085e207f800SAndre Oppermann * either due to memory shortage, 1086e207f800SAndre Oppermann * listen queue length limits or 1087a160e630SAndre Oppermann * global socket limits. Send RST 1088a160e630SAndre Oppermann * or wait and have the remote end 1089a160e630SAndre Oppermann * retransmit the ACK for another 1090a160e630SAndre Oppermann * try. 1091be2ac88cSJonathan Lemon */ 1092a160e630SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1093a160e630SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 10948d573cc1SAndre Oppermann "Socket allocation failed due to " 10958d573cc1SAndre Oppermann "limits or memory shortage, %s\n", 1096ac957cd2SJulian Elischer s, __func__, 1097ac957cd2SJulian Elischer V_tcp_sc_rst_sock_fail ? 1098ac957cd2SJulian Elischer "sending RST" : "try again"); 1099603724d3SBjoern A. Zeeb if (V_tcp_sc_rst_sock_fail) { 1100e207f800SAndre Oppermann rstreason = BANDLIM_UNLIMITED; 1101e207f800SAndre Oppermann goto dropwithreset; 1102a160e630SAndre Oppermann } else 1103a160e630SAndre Oppermann goto dropunlock; 1104f76fcf6dSJeffrey Hsu } 1105be2ac88cSJonathan Lemon /* 1106be2ac88cSJonathan Lemon * Socket is created in state SYN_RECEIVED. 11078d573cc1SAndre Oppermann * Unlock the listen socket, lock the newly 11088d573cc1SAndre Oppermann * created socket and update the tp variable. 1109be2ac88cSJonathan Lemon */ 11108501a69cSRobert Watson INP_WUNLOCK(inp); /* listen socket */ 1111be2ac88cSJonathan Lemon inp = sotoinpcb(so); 11128501a69cSRobert Watson INP_WLOCK(inp); /* new connection */ 1113be2ac88cSJonathan Lemon tp = intotcpcb(inp); 11148d573cc1SAndre Oppermann KASSERT(tp->t_state == TCPS_SYN_RECEIVED, 11158d573cc1SAndre Oppermann ("%s: ", __func__)); 11162903309aSAttilio Rao #ifdef TCP_SIGNATURE 11172903309aSAttilio Rao if (sig_checked == 0) { 11182903309aSAttilio Rao tcp_dooptions(&to, optp, optlen, 11192903309aSAttilio Rao (thflags & TH_SYN) ? TO_SYN : 0); 11202903309aSAttilio Rao if (!tcp_signature_verify_input(m, off0, tlen, 11212903309aSAttilio Rao optlen, &to, th, tp->t_flags)) { 11222903309aSAttilio Rao 11232903309aSAttilio Rao /* 11242903309aSAttilio Rao * In SYN_SENT state if it receives an 11252903309aSAttilio Rao * RST, it is allowed for further 11262903309aSAttilio Rao * processing. 11272903309aSAttilio Rao */ 11282903309aSAttilio Rao if ((thflags & TH_RST) == 0 || 11292903309aSAttilio Rao (tp->t_state == TCPS_SYN_SENT) == 0) 11302903309aSAttilio Rao goto dropunlock; 11312903309aSAttilio Rao } 11322903309aSAttilio Rao sig_checked = 1; 11332903309aSAttilio Rao } 11342903309aSAttilio Rao #endif 11352903309aSAttilio Rao 1136be2ac88cSJonathan Lemon /* 1137302ce8d6SAndre Oppermann * Process the segment and the data it 1138302ce8d6SAndre Oppermann * contains. tcp_do_segment() consumes 1139302ce8d6SAndre Oppermann * the mbuf chain and unlocks the inpcb. 1140302ce8d6SAndre Oppermann */ 1141f2512ba1SRui Paulo tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, 1142252ca428SRobert Watson iptos, ti_locked); 1143603724d3SBjoern A. Zeeb INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 11448f5a8818SKevin Lo return (IPPROTO_DONE); 1145be2ac88cSJonathan Lemon } 1146a160e630SAndre Oppermann /* 11478d573cc1SAndre Oppermann * Segment flag validation for new connection attempts: 11488d573cc1SAndre Oppermann * 1149a160e630SAndre Oppermann * Our (SYN|ACK) response was rejected. 1150a160e630SAndre Oppermann * Check with syncache and remove entry to prevent 1151a160e630SAndre Oppermann * retransmits. 115219bc77c5SAndre Oppermann * 115319bc77c5SAndre Oppermann * NB: syncache_chkrst does its own logging of failure 115419bc77c5SAndre Oppermann * causes. 1155a160e630SAndre Oppermann */ 1156a160e630SAndre Oppermann if (thflags & TH_RST) { 115719bc77c5SAndre Oppermann syncache_chkrst(&inc, th); 1158a160e630SAndre Oppermann goto dropunlock; 1159a160e630SAndre Oppermann } 1160a160e630SAndre Oppermann /* 1161a160e630SAndre Oppermann * We can't do anything without SYN. 1162a160e630SAndre Oppermann */ 1163a160e630SAndre Oppermann if ((thflags & TH_SYN) == 0) { 1164a160e630SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1165a160e630SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 1166773673c1SAndre Oppermann "SYN is missing, segment ignored\n", 11678d573cc1SAndre Oppermann s, __func__); 116878b50714SRobert Watson TCPSTAT_INC(tcps_badsyn); 1169a160e630SAndre Oppermann goto dropunlock; 1170a160e630SAndre Oppermann } 1171a160e630SAndre Oppermann /* 1172a160e630SAndre Oppermann * (SYN|ACK) is bogus on a listen socket. 1173a160e630SAndre Oppermann */ 1174fb59c426SYoshinobu Inoue if (thflags & TH_ACK) { 1175a160e630SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1176a160e630SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 11778d573cc1SAndre Oppermann "SYN|ACK invalid, segment rejected\n", 11788d573cc1SAndre Oppermann s, __func__); 1179a160e630SAndre Oppermann syncache_badack(&inc); /* XXX: Not needed! */ 118078b50714SRobert Watson TCPSTAT_INC(tcps_badsyn); 1181a57815efSBosko Milekic rstreason = BANDLIM_RST_OPENPORT; 1182a57815efSBosko Milekic goto dropwithreset; 11834195b4afSPaul Traina } 1184a160e630SAndre Oppermann /* 1185a160e630SAndre Oppermann * If the drop_synfin option is enabled, drop all 1186a160e630SAndre Oppermann * segments with both the SYN and FIN bits set. 1187a160e630SAndre Oppermann * This prevents e.g. nmap from identifying the 1188a160e630SAndre Oppermann * TCP/IP stack. 1189a160e630SAndre Oppermann * XXX: Poor reasoning. nmap has other methods 1190a160e630SAndre Oppermann * and is constantly refining its stack detection 1191a160e630SAndre Oppermann * strategies. 11928d573cc1SAndre Oppermann * XXX: This is a violation of the TCP specification 1193a160e630SAndre Oppermann * and was used by RFC1644. 1194a160e630SAndre Oppermann */ 1195603724d3SBjoern A. Zeeb if ((thflags & TH_FIN) && V_drop_synfin) { 1196a160e630SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1197a160e630SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 1198773673c1SAndre Oppermann "SYN|FIN segment ignored (based on " 11998d573cc1SAndre Oppermann "sysctl setting)\n", s, __func__); 120078b50714SRobert Watson TCPSTAT_INC(tcps_badsyn); 1201302ce8d6SAndre Oppermann goto dropunlock; 1202ebb0cbeaSPaul Traina } 1203be2ac88cSJonathan Lemon /* 1204be2ac88cSJonathan Lemon * Segment's flags are (SYN) or (SYN|FIN). 1205a160e630SAndre Oppermann * 1206a160e630SAndre Oppermann * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored 1207a160e630SAndre Oppermann * as they do not affect the state of the TCP FSM. 1208a160e630SAndre Oppermann * The data pointed to by TH_URG and th_urp is ignored. 1209be2ac88cSJonathan Lemon */ 1210a160e630SAndre Oppermann KASSERT((thflags & (TH_RST|TH_ACK)) == 0, 1211a160e630SAndre Oppermann ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); 1212a160e630SAndre Oppermann KASSERT(thflags & (TH_SYN), 1213a160e630SAndre Oppermann ("%s: Listen socket: TH_SYN not set", __func__)); 121433841545SHajimu UMEMOTO #ifdef INET6 121533841545SHajimu UMEMOTO /* 121633841545SHajimu UMEMOTO * If deprecated address is forbidden, 121733841545SHajimu UMEMOTO * we do not accept SYN to deprecated interface 121833841545SHajimu UMEMOTO * address to prevent any new inbound connection from 121933841545SHajimu UMEMOTO * getting established. 122033841545SHajimu UMEMOTO * When we do not accept SYN, we send a TCP RST, 122133841545SHajimu UMEMOTO * with deprecated source address (instead of dropping 122233841545SHajimu UMEMOTO * it). We compromise it as it is much better for peer 122333841545SHajimu UMEMOTO * to send a RST, and RST will be the final packet 122433841545SHajimu UMEMOTO * for the exchange. 122533841545SHajimu UMEMOTO * 122633841545SHajimu UMEMOTO * If we do not forbid deprecated addresses, we accept 122733841545SHajimu UMEMOTO * the SYN packet. RFC2462 does not suggest dropping 122833841545SHajimu UMEMOTO * SYN in this case. 122933841545SHajimu UMEMOTO * If we decipher RFC2462 5.5.4, it says like this: 123033841545SHajimu UMEMOTO * 1. use of deprecated addr with existing 123133841545SHajimu UMEMOTO * communication is okay - "SHOULD continue to be 123233841545SHajimu UMEMOTO * used" 123333841545SHajimu UMEMOTO * 2. use of it with new communication: 123433841545SHajimu UMEMOTO * (2a) "SHOULD NOT be used if alternate address 123533841545SHajimu UMEMOTO * with sufficient scope is available" 123633841545SHajimu UMEMOTO * (2b) nothing mentioned otherwise. 123733841545SHajimu UMEMOTO * Here we fall into (2b) case as we have no choice in 123833841545SHajimu UMEMOTO * our source address selection - we must obey the peer. 123933841545SHajimu UMEMOTO * 124033841545SHajimu UMEMOTO * The wording in RFC2462 is confusing, and there are 124133841545SHajimu UMEMOTO * multiple description text for deprecated address 124233841545SHajimu UMEMOTO * handling - worse, they are not exactly the same. 124333841545SHajimu UMEMOTO * I believe 5.5.4 is the best one, so we follow 5.5.4. 124433841545SHajimu UMEMOTO */ 1245603724d3SBjoern A. Zeeb if (isipv6 && !V_ip6_use_deprecated) { 124633841545SHajimu UMEMOTO struct in6_ifaddr *ia6; 124733841545SHajimu UMEMOTO 12488c0fec80SRobert Watson ia6 = ip6_getdstifaddr(m); 12498c0fec80SRobert Watson if (ia6 != NULL && 125033841545SHajimu UMEMOTO (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { 12518c0fec80SRobert Watson ifa_free(&ia6->ia_ifa); 1252a160e630SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1253a160e630SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 12548d573cc1SAndre Oppermann "Connection attempt to deprecated " 12558d573cc1SAndre Oppermann "IPv6 address rejected\n", 1256a160e630SAndre Oppermann s, __func__); 125733841545SHajimu UMEMOTO rstreason = BANDLIM_RST_OPENPORT; 125833841545SHajimu UMEMOTO goto dropwithreset; 125933841545SHajimu UMEMOTO } 126077d396fdSMaksim Yevmenkin if (ia6) 12618c0fec80SRobert Watson ifa_free(&ia6->ia_ifa); 126233841545SHajimu UMEMOTO } 1263b287c6c7SBjoern A. Zeeb #endif /* INET6 */ 126465f28919SJesper Skriver /* 1265db33b3e6SAndre Oppermann * Basic sanity checks on incoming SYN requests: 12668d573cc1SAndre Oppermann * Don't respond if the destination is a link layer 1267db33b3e6SAndre Oppermann * broadcast according to RFC1122 4.2.3.10, p. 104. 12688d573cc1SAndre Oppermann * If it is from this socket it must be forged. 12698d573cc1SAndre Oppermann * Don't respond if the source or destination is a 12708d573cc1SAndre Oppermann * global or subnet broad- or multicast address. 127193ec91baSCrist J. Clark * Note that it is quite possible to receive unicast 127293ec91baSCrist J. Clark * link-layer packets with a broadcast IP address. Use 127393ec91baSCrist J. Clark * in_broadcast() to find them. 1274be2ac88cSJonathan Lemon */ 12758d573cc1SAndre Oppermann if (m->m_flags & (M_BCAST|M_MCAST)) { 12768d573cc1SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 12778d573cc1SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 12788d573cc1SAndre Oppermann "Connection attempt from broad- or multicast " 1279773673c1SAndre Oppermann "link layer address ignored\n", s, __func__); 1280302ce8d6SAndre Oppermann goto dropunlock; 12818d573cc1SAndre Oppermann } 1282db33b3e6SAndre Oppermann #ifdef INET6 1283b287c6c7SBjoern A. Zeeb if (isipv6) { 1284db33b3e6SAndre Oppermann if (th->th_dport == th->th_sport && 1285a160e630SAndre Oppermann IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { 1286a160e630SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1287a160e630SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 12888d573cc1SAndre Oppermann "Connection attempt to/from self " 1289773673c1SAndre Oppermann "ignored\n", s, __func__); 1290302ce8d6SAndre Oppermann goto dropunlock; 1291a160e630SAndre Oppermann } 1292be2ac88cSJonathan Lemon if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 1293a160e630SAndre Oppermann IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { 1294a160e630SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1295a160e630SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 12968d573cc1SAndre Oppermann "Connection attempt from/to multicast " 1297773673c1SAndre Oppermann "address ignored\n", s, __func__); 1298302ce8d6SAndre Oppermann goto dropunlock; 1299a160e630SAndre Oppermann } 1300b287c6c7SBjoern A. Zeeb } 1301db33b3e6SAndre Oppermann #endif 1302b287c6c7SBjoern A. Zeeb #if defined(INET) && defined(INET6) 1303b287c6c7SBjoern A. Zeeb else 1304b287c6c7SBjoern A. Zeeb #endif 1305b287c6c7SBjoern A. Zeeb #ifdef INET 1306b287c6c7SBjoern A. Zeeb { 1307db33b3e6SAndre Oppermann if (th->th_dport == th->th_sport && 1308a160e630SAndre Oppermann ip->ip_dst.s_addr == ip->ip_src.s_addr) { 1309a160e630SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1310a160e630SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 13118d573cc1SAndre Oppermann "Connection attempt from/to self " 1312773673c1SAndre Oppermann "ignored\n", s, __func__); 1313302ce8d6SAndre Oppermann goto dropunlock; 1314a160e630SAndre Oppermann } 1315be2ac88cSJonathan Lemon if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 1316be2ac88cSJonathan Lemon IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 13172ca2159fSCrist J. Clark ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 1318a160e630SAndre Oppermann in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { 1319a160e630SAndre Oppermann if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) 1320a160e630SAndre Oppermann log(LOG_DEBUG, "%s; %s: Listen socket: " 13218d573cc1SAndre Oppermann "Connection attempt from/to broad- " 1322773673c1SAndre Oppermann "or multicast address ignored\n", 1323a160e630SAndre Oppermann s, __func__); 1324302ce8d6SAndre Oppermann goto dropunlock; 1325c068736aSJeffrey Hsu } 1326a160e630SAndre Oppermann } 1327b287c6c7SBjoern A. Zeeb #endif 1328be2ac88cSJonathan Lemon /* 1329db33b3e6SAndre Oppermann * SYN appears to be valid. Create compressed TCP state 1330db33b3e6SAndre Oppermann * for syncache. 1331be2ac88cSJonathan Lemon */ 13323c653157SHartmut Brandt #ifdef TCPDEBUG 13333c653157SHartmut Brandt if (so->so_options & SO_DEBUG) 13343c653157SHartmut Brandt tcp_trace(TA_INPUT, ostate, tp, 13353c653157SHartmut Brandt (void *)tcp_saveipgen, &tcp_savetcp, 0); 13363c653157SHartmut Brandt #endif 1337f72167f4SAndre Oppermann tcp_dooptions(&to, optp, optlen, TO_SYN); 133809fe6320SNavdeep Parhar syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL); 1339be2ac88cSJonathan Lemon /* 13404d6e7130SAndre Oppermann * Entry added to syncache and mbuf consumed. 1341a7c7f2a7SJohn Baldwin * Only the listen socket is unlocked by syncache_add(). 1342be2ac88cSJonathan Lemon */ 1343a7c7f2a7SJohn Baldwin if (ti_locked == TI_WLOCKED) { 1344a7c7f2a7SJohn Baldwin INP_INFO_WUNLOCK(&V_tcbinfo); 1345a7c7f2a7SJohn Baldwin ti_locked = TI_UNLOCKED; 1346a7c7f2a7SJohn Baldwin } 1347603724d3SBjoern A. Zeeb INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 13488f5a8818SKevin Lo return (IPPROTO_DONE); 1349982c1675SAndre Oppermann } else if (tp->t_state == TCPS_LISTEN) { 1350982c1675SAndre Oppermann /* 1351982c1675SAndre Oppermann * When a listen socket is torn down the SO_ACCEPTCONN 1352982c1675SAndre Oppermann * flag is removed first while connections are drained 1353982c1675SAndre Oppermann * from the accept queue in a unlock/lock cycle of the 1354982c1675SAndre Oppermann * ACCEPT_LOCK, opening a race condition allowing a SYN 1355982c1675SAndre Oppermann * attempt go through unhandled. 1356982c1675SAndre Oppermann */ 1357982c1675SAndre Oppermann goto dropunlock; 1358f76fcf6dSJeffrey Hsu } 1359db33b3e6SAndre Oppermann 13602903309aSAttilio Rao #ifdef TCP_SIGNATURE 13612903309aSAttilio Rao if (sig_checked == 0) { 13622903309aSAttilio Rao tcp_dooptions(&to, optp, optlen, 13632903309aSAttilio Rao (thflags & TH_SYN) ? TO_SYN : 0); 13642903309aSAttilio Rao if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to, 13652903309aSAttilio Rao th, tp->t_flags)) { 13662903309aSAttilio Rao 13672903309aSAttilio Rao /* 13682903309aSAttilio Rao * In SYN_SENT state if it receives an RST, it is 13692903309aSAttilio Rao * allowed for further processing. 13702903309aSAttilio Rao */ 13712903309aSAttilio Rao if ((thflags & TH_RST) == 0 || 13722903309aSAttilio Rao (tp->t_state == TCPS_SYN_SENT) == 0) 13732903309aSAttilio Rao goto dropunlock; 13742903309aSAttilio Rao } 13752903309aSAttilio Rao sig_checked = 1; 13762903309aSAttilio Rao } 13772903309aSAttilio Rao #endif 13782903309aSAttilio Rao 1379fa22ce15SAdrian Chadd TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 138057f60867SMark Johnston 1381302ce8d6SAndre Oppermann /* 13828d573cc1SAndre Oppermann * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later 13838d573cc1SAndre Oppermann * state. tcp_do_segment() always consumes the mbuf chain, unlocks 13848d573cc1SAndre Oppermann * the inpcb, and unlocks pcbinfo. 1385302ce8d6SAndre Oppermann */ 1386252ca428SRobert Watson tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked); 1387603724d3SBjoern A. Zeeb INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 13888f5a8818SKevin Lo return (IPPROTO_DONE); 1389be2ac88cSJonathan Lemon 1390302ce8d6SAndre Oppermann dropwithreset: 1391fa22ce15SAdrian Chadd TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 139257f60867SMark Johnston 1393fa046d87SRobert Watson if (ti_locked == TI_WLOCKED) { 1394dd8ac7f9SRobert Watson INP_INFO_WUNLOCK(&V_tcbinfo); 1395252ca428SRobert Watson ti_locked = TI_UNLOCKED; 1396fa046d87SRobert Watson } 1397fa046d87SRobert Watson #ifdef INVARIANTS 1398fa046d87SRobert Watson else { 1399fa046d87SRobert Watson KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset " 1400fa046d87SRobert Watson "ti_locked: %d", __func__, ti_locked)); 1401fa046d87SRobert Watson INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1402fa046d87SRobert Watson } 1403fa046d87SRobert Watson #endif 1404014ea782SRobert Watson 1405014ea782SRobert Watson if (inp != NULL) { 1406302ce8d6SAndre Oppermann tcp_dropwithreset(m, th, tp, tlen, rstreason); 1407014ea782SRobert Watson INP_WUNLOCK(inp); 1408dd8ac7f9SRobert Watson } else 1409014ea782SRobert Watson tcp_dropwithreset(m, th, NULL, tlen, rstreason); 1410302ce8d6SAndre Oppermann m = NULL; /* mbuf chain got consumed. */ 1411014ea782SRobert Watson goto drop; 1412014ea782SRobert Watson 1413302ce8d6SAndre Oppermann dropunlock: 141457f60867SMark Johnston if (m != NULL) 1415fa22ce15SAdrian Chadd TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th); 141657f60867SMark Johnston 1417fa046d87SRobert Watson if (ti_locked == TI_WLOCKED) { 1418252ca428SRobert Watson INP_INFO_WUNLOCK(&V_tcbinfo); 1419252ca428SRobert Watson ti_locked = TI_UNLOCKED; 1420fa046d87SRobert Watson } 1421fa046d87SRobert Watson #ifdef INVARIANTS 1422fa046d87SRobert Watson else { 1423fa046d87SRobert Watson KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock " 1424fa046d87SRobert Watson "ti_locked: %d", __func__, ti_locked)); 1425fa046d87SRobert Watson INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1426fa046d87SRobert Watson } 1427fa046d87SRobert Watson #endif 1428252ca428SRobert Watson 14299fa198beSAndre Oppermann if (inp != NULL) 14308501a69cSRobert Watson INP_WUNLOCK(inp); 1431014ea782SRobert Watson 1432302ce8d6SAndre Oppermann drop: 1433603724d3SBjoern A. Zeeb INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1434a160e630SAndre Oppermann if (s != NULL) 1435a160e630SAndre Oppermann free(s, M_TCPLOG); 1436302ce8d6SAndre Oppermann if (m != NULL) 1437302ce8d6SAndre Oppermann m_freem(m); 14388f5a8818SKevin Lo return (IPPROTO_DONE); 1439302ce8d6SAndre Oppermann } 1440302ce8d6SAndre Oppermann 1441679d9708SAndre Oppermann static void 1442302ce8d6SAndre Oppermann tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so, 1443252ca428SRobert Watson struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos, 1444252ca428SRobert Watson int ti_locked) 1445302ce8d6SAndre Oppermann { 1446302ce8d6SAndre Oppermann int thflags, acked, ourfinisacked, needoutput = 0; 1447302ce8d6SAndre Oppermann int rstreason, todrop, win; 1448302ce8d6SAndre Oppermann u_long tiwin; 144907dacf03SAndre Oppermann char *s; 145007dacf03SAndre Oppermann struct in_conninfo *inc; 1451c11a15bfSGleb Smirnoff struct mbuf *mfree; 1452302ce8d6SAndre Oppermann struct tcpopt to; 1453302ce8d6SAndre Oppermann 145414739780SMaxim Konovalov #ifdef TCPDEBUG 145514739780SMaxim Konovalov /* 145614739780SMaxim Konovalov * The size of tcp_saveipgen must be the size of the max ip header, 145714739780SMaxim Konovalov * now IPv6. 145814739780SMaxim Konovalov */ 145914739780SMaxim Konovalov u_char tcp_saveipgen[IP6_HDR_LEN]; 146014739780SMaxim Konovalov struct tcphdr tcp_savetcp; 146114739780SMaxim Konovalov short ostate = 0; 146214739780SMaxim Konovalov #endif 1463302ce8d6SAndre Oppermann thflags = th->th_flags; 146407dacf03SAndre Oppermann inc = &tp->t_inpcb->inp_inc; 1465d64a46eaSLawrence Stewart tp->sackhint.last_sack_ack = 0; 1466302ce8d6SAndre Oppermann 1467252ca428SRobert Watson /* 1468252ca428SRobert Watson * If this is either a state-changing packet or current state isn't 1469252ca428SRobert Watson * established, we require a write lock on tcbinfo. Otherwise, we 14700989f56cSRobert Watson * allow the tcbinfo to be in either alocked or unlocked, as the 14710989f56cSRobert Watson * caller may have unnecessarily acquired a write lock due to a race. 1472252ca428SRobert Watson */ 1473252ca428SRobert Watson if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 || 1474252ca428SRobert Watson tp->t_state != TCPS_ESTABLISHED) { 1475252ca428SRobert Watson KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for " 1476252ca428SRobert Watson "SYN/FIN/RST/!EST", __func__, ti_locked)); 1477603724d3SBjoern A. Zeeb INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1478252ca428SRobert Watson } else { 1479252ca428SRobert Watson #ifdef INVARIANTS 1480fa046d87SRobert Watson if (ti_locked == TI_WLOCKED) 1481252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 1482fa046d87SRobert Watson else { 1483fa046d87SRobert Watson KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST " 1484fa046d87SRobert Watson "ti_locked: %d", __func__, ti_locked)); 1485fa046d87SRobert Watson INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 1486fa046d87SRobert Watson } 1487252ca428SRobert Watson #endif 1488252ca428SRobert Watson } 14898501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 1490679d9708SAndre Oppermann KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", 1491679d9708SAndre Oppermann __func__)); 1492679d9708SAndre Oppermann KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", 1493679d9708SAndre Oppermann __func__)); 1494df8bae1dSRodney W. Grimes 1495df8bae1dSRodney W. Grimes /* 1496df8bae1dSRodney W. Grimes * Segment received on connection. 1497df8bae1dSRodney W. Grimes * Reset idle time and keep-alive timer. 1498e8949f74SAndre Oppermann * XXX: This should be done after segment 1499e8949f74SAndre Oppermann * validation to ignore broken/spoofed segs. 1500df8bae1dSRodney W. Grimes */ 15019b8b58e0SJonathan Lemon tp->t_rcvtime = ticks; 15027ff19458SPaul Traina if (TCPS_HAVEESTABLISHED(tp->t_state)) 15039077f387SGleb Smirnoff tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 1504df8bae1dSRodney W. Grimes 1505df8bae1dSRodney W. Grimes /* 1506464fcfbcSAndre Oppermann * Unscale the window into a 32-bit value. 1507e8949f74SAndre Oppermann * For the SYN_SENT state the scale is zero. 1508464fcfbcSAndre Oppermann */ 1509464fcfbcSAndre Oppermann tiwin = th->th_win << tp->snd_scale; 1510464fcfbcSAndre Oppermann 1511464fcfbcSAndre Oppermann /* 1512f2512ba1SRui Paulo * TCP ECN processing. 1513f2512ba1SRui Paulo */ 1514f2512ba1SRui Paulo if (tp->t_flags & TF_ECN_PERMIT) { 15159c251892SRui Paulo if (thflags & TH_CWR) 15169c251892SRui Paulo tp->t_flags &= ~TF_ECN_SND_ECE; 1517f2512ba1SRui Paulo switch (iptos & IPTOS_ECN_MASK) { 1518f2512ba1SRui Paulo case IPTOS_ECN_CE: 1519f2512ba1SRui Paulo tp->t_flags |= TF_ECN_SND_ECE; 152078b50714SRobert Watson TCPSTAT_INC(tcps_ecn_ce); 1521f2512ba1SRui Paulo break; 1522f2512ba1SRui Paulo case IPTOS_ECN_ECT0: 152378b50714SRobert Watson TCPSTAT_INC(tcps_ecn_ect0); 1524f2512ba1SRui Paulo break; 1525f2512ba1SRui Paulo case IPTOS_ECN_ECT1: 152678b50714SRobert Watson TCPSTAT_INC(tcps_ecn_ect1); 1527f2512ba1SRui Paulo break; 1528f2512ba1SRui Paulo } 1529dbc42409SLawrence Stewart /* Congestion experienced. */ 1530dbc42409SLawrence Stewart if (thflags & TH_ECE) { 1531dbc42409SLawrence Stewart cc_cong_signal(tp, th, CC_ECN); 1532f2512ba1SRui Paulo } 1533f2512ba1SRui Paulo } 1534f2512ba1SRui Paulo 1535f2512ba1SRui Paulo /* 1536f72167f4SAndre Oppermann * Parse options on any incoming segment. 1537f72167f4SAndre Oppermann */ 1538302ce8d6SAndre Oppermann tcp_dooptions(&to, (u_char *)(th + 1), 1539302ce8d6SAndre Oppermann (th->th_off << 2) - sizeof(struct tcphdr), 1540302ce8d6SAndre Oppermann (thflags & TH_SYN) ? TO_SYN : 0); 1541f72167f4SAndre Oppermann 1542f72167f4SAndre Oppermann /* 1543f72167f4SAndre Oppermann * If echoed timestamp is later than the current time, 1544bf6d304aSAndre Oppermann * fall back to non RFC1323 RTT calculation. Normalize 1545bf6d304aSAndre Oppermann * timestamp if syncookies were used when this connection 1546bf6d304aSAndre Oppermann * was established. 1547f72167f4SAndre Oppermann */ 1548bf6d304aSAndre Oppermann if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { 1549e16fa5caSJohn-Mark Gurney to.to_tsecr -= tp->ts_offset; 1550d8951c8aSBjoern A. Zeeb if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) 1551f72167f4SAndre Oppermann to.to_tsecr = 0; 1552bf6d304aSAndre Oppermann } 155307dacf03SAndre Oppermann /* 155407dacf03SAndre Oppermann * If timestamps were negotiated during SYN/ACK they should 155507dacf03SAndre Oppermann * appear on every segment during this session and vice versa. 155607dacf03SAndre Oppermann */ 155707dacf03SAndre Oppermann if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { 155807dacf03SAndre Oppermann if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 155907dacf03SAndre Oppermann log(LOG_DEBUG, "%s; %s: Timestamp missing, " 156007dacf03SAndre Oppermann "no action\n", s, __func__); 156107dacf03SAndre Oppermann free(s, M_TCPLOG); 156207dacf03SAndre Oppermann } 156307dacf03SAndre Oppermann } 156407dacf03SAndre Oppermann if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { 156507dacf03SAndre Oppermann if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 156607dacf03SAndre Oppermann log(LOG_DEBUG, "%s; %s: Timestamp not expected, " 156707dacf03SAndre Oppermann "no action\n", s, __func__); 156807dacf03SAndre Oppermann free(s, M_TCPLOG); 156907dacf03SAndre Oppermann } 157007dacf03SAndre Oppermann } 1571f72167f4SAndre Oppermann 1572f72167f4SAndre Oppermann /* 157397d8d152SAndre Oppermann * Process options only when we get SYN/ACK back. The SYN case 157497d8d152SAndre Oppermann * for incoming connections is handled in tcp_syncache. 15755396d0f8SAndre Oppermann * According to RFC1323 the window field in a SYN (i.e., a <SYN> 15765396d0f8SAndre Oppermann * or <SYN,ACK>) segment itself is never scaled. 157797d8d152SAndre Oppermann * XXX this is traditional behavior, may need to be cleaned up. 1578df8bae1dSRodney W. Grimes */ 15790a389eabSSimon L. B. Nielsen if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { 1580464fcfbcSAndre Oppermann if ((to.to_flags & TOF_SCALE) && 1581464fcfbcSAndre Oppermann (tp->t_flags & TF_REQ_SCALE)) { 1582be2ac88cSJonathan Lemon tp->t_flags |= TF_RCVD_SCALE; 158302a1a643SAndre Oppermann tp->snd_scale = to.to_wscale; 1584be2ac88cSJonathan Lemon } 15855396d0f8SAndre Oppermann /* 15865396d0f8SAndre Oppermann * Initial send window. It will be updated with 15875396d0f8SAndre Oppermann * the next incoming segment to the scaled value. 15885396d0f8SAndre Oppermann */ 15895396d0f8SAndre Oppermann tp->snd_wnd = th->th_win; 1590be2ac88cSJonathan Lemon if (to.to_flags & TOF_TS) { 1591be2ac88cSJonathan Lemon tp->t_flags |= TF_RCVD_TSTMP; 1592be2ac88cSJonathan Lemon tp->ts_recent = to.to_tsval; 1593d8951c8aSBjoern A. Zeeb tp->ts_recent_age = tcp_ts_getticks(); 1594be2ac88cSJonathan Lemon } 1595be2ac88cSJonathan Lemon if (to.to_flags & TOF_MSS) 1596be2ac88cSJonathan Lemon tcp_mss(tp, to.to_mss); 15973529149eSAndre Oppermann if ((tp->t_flags & TF_SACK_PERMIT) && 15983529149eSAndre Oppermann (to.to_flags & TOF_SACKPERM) == 0) 15993529149eSAndre Oppermann tp->t_flags &= ~TF_SACK_PERMIT; 16006d90faf3SPaul Saab } 16016d90faf3SPaul Saab 1602df8bae1dSRodney W. Grimes /* 1603df8bae1dSRodney W. Grimes * Header prediction: check for the two common cases 1604df8bae1dSRodney W. Grimes * of a uni-directional data xfer. If the packet has 1605df8bae1dSRodney W. Grimes * no control flags, is in-sequence, the window didn't 1606df8bae1dSRodney W. Grimes * change and we're not retransmitting, it's a 1607df8bae1dSRodney W. Grimes * candidate. If the length is zero and the ack moved 1608df8bae1dSRodney W. Grimes * forward, we're the sender side of the xfer. Just 1609df8bae1dSRodney W. Grimes * free the data acked & wake any higher level process 1610df8bae1dSRodney W. Grimes * that was blocked waiting for space. If the length 1611df8bae1dSRodney W. Grimes * is non-zero and the ack didn't move, we're the 1612df8bae1dSRodney W. Grimes * receiver side. If we're getting packets in-order 1613df8bae1dSRodney W. Grimes * (the reassembly queue is empty), add the data to 1614df8bae1dSRodney W. Grimes * the socket buffer and note that we need a delayed ack. 1615a0292f23SGarrett Wollman * Make sure that the hidden state-flags are also off. 1616c5ad39b9SAndre Oppermann * Since we check for TCPS_ESTABLISHED first, it can only 1617a0292f23SGarrett Wollman * be TH_NEEDSYN. 1618df8bae1dSRodney W. Grimes */ 1619df8bae1dSRodney W. Grimes if (tp->t_state == TCPS_ESTABLISHED && 1620c5ad39b9SAndre Oppermann th->th_seq == tp->rcv_nxt && 1621fb59c426SYoshinobu Inoue (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && 1622c5ad39b9SAndre Oppermann tp->snd_nxt == tp->snd_max && 1623c5ad39b9SAndre Oppermann tiwin && tiwin == tp->snd_wnd && 1624a0292f23SGarrett Wollman ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && 1625e407b67bSGleb Smirnoff tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 || 1626c5ad39b9SAndre Oppermann TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { 1627df8bae1dSRodney W. Grimes 1628df8bae1dSRodney W. Grimes /* 1629df8bae1dSRodney W. Grimes * If last ACK falls within this segment's sequence numbers, 1630df8bae1dSRodney W. Grimes * record the timestamp. 1631a0292f23SGarrett Wollman * NOTE that the test is modified according to the latest 1632a0292f23SGarrett Wollman * proposal of the tcplw@cray.com list (Braden 1993/04/26). 1633df8bae1dSRodney W. Grimes */ 1634be2ac88cSJonathan Lemon if ((to.to_flags & TOF_TS) != 0 && 1635fb59c426SYoshinobu Inoue SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { 1636d8951c8aSBjoern A. Zeeb tp->ts_recent_age = tcp_ts_getticks(); 1637a0292f23SGarrett Wollman tp->ts_recent = to.to_tsval; 1638df8bae1dSRodney W. Grimes } 1639df8bae1dSRodney W. Grimes 1640fb59c426SYoshinobu Inoue if (tlen == 0) { 1641fb59c426SYoshinobu Inoue if (SEQ_GT(th->th_ack, tp->snd_una) && 1642fb59c426SYoshinobu Inoue SEQ_LEQ(th->th_ack, tp->snd_max) && 1643dbc42409SLawrence Stewart !IN_RECOVERY(tp->t_flags) && 1644fc30a251SAndre Oppermann (to.to_flags & TOF_SACK) == 0 && 1645dbc42409SLawrence Stewart TAILQ_EMPTY(&tp->snd_holes)) { 1646df8bae1dSRodney W. Grimes /* 1647e8949f74SAndre Oppermann * This is a pure ack for outstanding data. 1648df8bae1dSRodney W. Grimes */ 1649fa046d87SRobert Watson if (ti_locked == TI_WLOCKED) 1650252ca428SRobert Watson INP_INFO_WUNLOCK(&V_tcbinfo); 1651252ca428SRobert Watson ti_locked = TI_UNLOCKED; 1652252ca428SRobert Watson 165378b50714SRobert Watson TCPSTAT_INC(tcps_predack); 1654252ca428SRobert Watson 16559b8b58e0SJonathan Lemon /* 1656e8949f74SAndre Oppermann * "bad retransmit" recovery. 16579b8b58e0SJonathan Lemon */ 16589b8b58e0SJonathan Lemon if (tp->t_rxtshift == 1 && 1659672dc4aeSJohn Baldwin tp->t_flags & TF_PREVVALID && 16606dfb8b31SJohn Baldwin (int)(ticks - tp->t_badrxtwin) < 0) { 1661dbc42409SLawrence Stewart cc_cong_signal(tp, th, CC_RTO_ERR); 16629b8b58e0SJonathan Lemon } 1663fa55172bSMatthew Dillon 1664fa55172bSMatthew Dillon /* 1665fa55172bSMatthew Dillon * Recalculate the transmit timer / rtt. 1666fa55172bSMatthew Dillon * 1667fa55172bSMatthew Dillon * Some boxes send broken timestamp replies 1668fa55172bSMatthew Dillon * during the SYN+ACK phase, ignore 1669fa55172bSMatthew Dillon * timestamps of 0 or we could calculate a 1670fa55172bSMatthew Dillon * huge RTT and blow up the retransmit timer. 1671fa55172bSMatthew Dillon */ 1672fa55172bSMatthew Dillon if ((to.to_flags & TOF_TS) != 0 && 1673fa55172bSMatthew Dillon to.to_tsecr) { 1674d8951c8aSBjoern A. Zeeb u_int t; 1675d8951c8aSBjoern A. Zeeb 1676d8951c8aSBjoern A. Zeeb t = tcp_ts_getticks() - to.to_tsecr; 1677d8951c8aSBjoern A. Zeeb if (!tp->t_rttlow || tp->t_rttlow > t) 1678d8951c8aSBjoern A. Zeeb tp->t_rttlow = t; 1679a0292f23SGarrett Wollman tcp_xmit_timer(tp, 1680d8951c8aSBjoern A. Zeeb TCP_TS_TO_TICKS(t) + 1); 1681fa55172bSMatthew Dillon } else if (tp->t_rtttime && 1682fa55172bSMatthew Dillon SEQ_GT(th->th_ack, tp->t_rtseq)) { 1683eaf80179SAndre Oppermann if (!tp->t_rttlow || 1684eaf80179SAndre Oppermann tp->t_rttlow > ticks - tp->t_rtttime) 1685eaf80179SAndre Oppermann tp->t_rttlow = ticks - tp->t_rtttime; 1686c068736aSJeffrey Hsu tcp_xmit_timer(tp, 1687c068736aSJeffrey Hsu ticks - tp->t_rtttime); 1688fa55172bSMatthew Dillon } 1689dbc42409SLawrence Stewart acked = BYTES_THIS_ACK(tp, th); 169039bc9de5SLawrence Stewart 169139bc9de5SLawrence Stewart /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 169239bc9de5SLawrence Stewart hhook_run_tcp_est_in(tp, th, &to); 169339bc9de5SLawrence Stewart 169478b50714SRobert Watson TCPSTAT_INC(tcps_rcvackpack); 169578b50714SRobert Watson TCPSTAT_ADD(tcps_rcvackbyte, acked); 1696df8bae1dSRodney W. Grimes sbdrop(&so->so_snd, acked); 16979d11646dSJeffrey Hsu if (SEQ_GT(tp->snd_una, tp->snd_recover) && 16989d11646dSJeffrey Hsu SEQ_LEQ(th->th_ack, tp->snd_recover)) 16999d11646dSJeffrey Hsu tp->snd_recover = th->th_ack - 1; 1700dbc42409SLawrence Stewart 1701dbc42409SLawrence Stewart /* 1702dbc42409SLawrence Stewart * Let the congestion control algorithm update 1703dbc42409SLawrence Stewart * congestion control related information. This 1704dbc42409SLawrence Stewart * typically means increasing the congestion 1705dbc42409SLawrence Stewart * window. 1706dbc42409SLawrence Stewart */ 1707dbc42409SLawrence Stewart cc_ack_received(tp, th, CC_ACK); 1708dbc42409SLawrence Stewart 17099d11646dSJeffrey Hsu tp->snd_una = th->th_ack; 17101645d090SJeff Roberson /* 1711e8949f74SAndre Oppermann * Pull snd_wl2 up to prevent seq wrap relative 17121645d090SJeff Roberson * to th_ack. 17131645d090SJeff Roberson */ 1714cb942153SJeffrey Hsu tp->snd_wl2 = th->th_ack; 1715b5addd85SJeffrey Hsu tp->t_dupacks = 0; 1716df8bae1dSRodney W. Grimes m_freem(m); 1717e8949f74SAndre Oppermann ND6_HINT(tp); /* Some progress has been made. */ 1718df8bae1dSRodney W. Grimes 1719df8bae1dSRodney W. Grimes /* 1720df8bae1dSRodney W. Grimes * If all outstanding data are acked, stop 1721df8bae1dSRodney W. Grimes * retransmit timer, otherwise restart timer 1722df8bae1dSRodney W. Grimes * using current (possibly backed-off) value. 1723df8bae1dSRodney W. Grimes * If process is waiting for space, 1724df8bae1dSRodney W. Grimes * wakeup/selwakeup/signal. If data 1725df8bae1dSRodney W. Grimes * are ready to send, let tcp_output 1726df8bae1dSRodney W. Grimes * decide between more output or persist. 1727e8949f74SAndre Oppermann */ 17283c653157SHartmut Brandt #ifdef TCPDEBUG 17293c653157SHartmut Brandt if (so->so_options & SO_DEBUG) 17303c653157SHartmut Brandt tcp_trace(TA_INPUT, ostate, tp, 17313c653157SHartmut Brandt (void *)tcp_saveipgen, 17323c653157SHartmut Brandt &tcp_savetcp, 0); 17333c653157SHartmut Brandt #endif 1734df8bae1dSRodney W. Grimes if (tp->snd_una == tp->snd_max) 1735b8152ba7SAndre Oppermann tcp_timer_activate(tp, TT_REXMT, 0); 1736b8152ba7SAndre Oppermann else if (!tcp_timer_active(tp, TT_PERSIST)) 1737b8152ba7SAndre Oppermann tcp_timer_activate(tp, TT_REXMT, 1738b8152ba7SAndre Oppermann tp->t_rxtcur); 1739df8bae1dSRodney W. Grimes sowwakeup(so); 1740df8bae1dSRodney W. Grimes if (so->so_snd.sb_cc) 1741df8bae1dSRodney W. Grimes (void) tcp_output(tp); 1742a14c749fSJonathan Lemon goto check_delack; 1743df8bae1dSRodney W. Grimes } 1744fb59c426SYoshinobu Inoue } else if (th->th_ack == tp->snd_una && 1745fb59c426SYoshinobu Inoue tlen <= sbspace(&so->so_rcv)) { 17466741ecf5SAndre Oppermann int newsize = 0; /* automatic sockbuf scaling */ 17476741ecf5SAndre Oppermann 1748df8bae1dSRodney W. Grimes /* 1749252ca428SRobert Watson * This is a pure, in-sequence data packet with 1750252ca428SRobert Watson * nothing on the reassembly queue and we have enough 1751252ca428SRobert Watson * buffer space to take it. 1752df8bae1dSRodney W. Grimes */ 1753fa046d87SRobert Watson if (ti_locked == TI_WLOCKED) 1754252ca428SRobert Watson INP_INFO_WUNLOCK(&V_tcbinfo); 1755252ca428SRobert Watson ti_locked = TI_UNLOCKED; 1756252ca428SRobert Watson 17576d90faf3SPaul Saab /* Clean receiver SACK report if present */ 17583529149eSAndre Oppermann if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) 17596d90faf3SPaul Saab tcp_clean_sackreport(tp); 176078b50714SRobert Watson TCPSTAT_INC(tcps_preddat); 1761fb59c426SYoshinobu Inoue tp->rcv_nxt += tlen; 17621645d090SJeff Roberson /* 17631645d090SJeff Roberson * Pull snd_wl1 up to prevent seq wrap relative to 17641645d090SJeff Roberson * th_seq. 17651645d090SJeff Roberson */ 176660ee3bb2SAndre Oppermann tp->snd_wl1 = th->th_seq; 17671645d090SJeff Roberson /* 17681645d090SJeff Roberson * Pull rcv_up up to prevent seq wrap relative to 17691645d090SJeff Roberson * rcv_nxt. 17701645d090SJeff Roberson */ 17711645d090SJeff Roberson tp->rcv_up = tp->rcv_nxt; 177278b50714SRobert Watson TCPSTAT_INC(tcps_rcvpack); 177378b50714SRobert Watson TCPSTAT_ADD(tcps_rcvbyte, tlen); 1774e8949f74SAndre Oppermann ND6_HINT(tp); /* Some progress has been made */ 17753c653157SHartmut Brandt #ifdef TCPDEBUG 17763c653157SHartmut Brandt if (so->so_options & SO_DEBUG) 17773c653157SHartmut Brandt tcp_trace(TA_INPUT, ostate, tp, 17783c653157SHartmut Brandt (void *)tcp_saveipgen, &tcp_savetcp, 0); 17793c653157SHartmut Brandt #endif 17806741ecf5SAndre Oppermann /* 17816741ecf5SAndre Oppermann * Automatic sizing of receive socket buffer. Often the send 17826741ecf5SAndre Oppermann * buffer size is not optimally adjusted to the actual network 17836741ecf5SAndre Oppermann * conditions at hand (delay bandwidth product). Setting the 17846741ecf5SAndre Oppermann * buffer size too small limits throughput on links with high 17856741ecf5SAndre Oppermann * bandwidth and high delay (eg. trans-continental/oceanic links). 17866741ecf5SAndre Oppermann * 17876741ecf5SAndre Oppermann * On the receive side the socket buffer memory is only rarely 17886741ecf5SAndre Oppermann * used to any significant extent. This allows us to be much 17896741ecf5SAndre Oppermann * more aggressive in scaling the receive socket buffer. For 17906741ecf5SAndre Oppermann * the case that the buffer space is actually used to a large 17916741ecf5SAndre Oppermann * extent and we run out of kernel memory we can simply drop 17926741ecf5SAndre Oppermann * the new segments; TCP on the sender will just retransmit it 17936741ecf5SAndre Oppermann * later. Setting the buffer size too big may only consume too 17946741ecf5SAndre Oppermann * much kernel memory if the application doesn't read() from 17956741ecf5SAndre Oppermann * the socket or packet loss or reordering makes use of the 17966741ecf5SAndre Oppermann * reassembly queue. 17976741ecf5SAndre Oppermann * 17986741ecf5SAndre Oppermann * The criteria to step up the receive buffer one notch are: 1799f7469d3eSHiren Panchasara * 1. Application has not set receive buffer size with 1800f7469d3eSHiren Panchasara * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. 1801f7469d3eSHiren Panchasara * 2. the number of bytes received during the time it takes 18026741ecf5SAndre Oppermann * one timestamp to be reflected back to us (the RTT); 1803f7469d3eSHiren Panchasara * 3. received bytes per RTT is within seven eighth of the 18046741ecf5SAndre Oppermann * current socket buffer size; 1805f7469d3eSHiren Panchasara * 4. receive buffer size has not hit maximal automatic size; 18066741ecf5SAndre Oppermann * 18076741ecf5SAndre Oppermann * This algorithm does one step per RTT at most and only if 18086741ecf5SAndre Oppermann * we receive a bulk stream w/o packet losses or reorderings. 18096741ecf5SAndre Oppermann * Shrinking the buffer during idle times is not necessary as 18106741ecf5SAndre Oppermann * it doesn't consume any memory when idle. 18116741ecf5SAndre Oppermann * 18126741ecf5SAndre Oppermann * TODO: Only step up if the application is actually serving 18136741ecf5SAndre Oppermann * the buffer to better manage the socket buffer resources. 1814df8bae1dSRodney W. Grimes */ 1815603724d3SBjoern A. Zeeb if (V_tcp_do_autorcvbuf && 18166741ecf5SAndre Oppermann to.to_tsecr && 18176741ecf5SAndre Oppermann (so->so_rcv.sb_flags & SB_AUTOSIZE)) { 18188502ec25SAndre Oppermann if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) && 18196741ecf5SAndre Oppermann to.to_tsecr - tp->rfbuf_ts < hz) { 18206741ecf5SAndre Oppermann if (tp->rfbuf_cnt > 18216741ecf5SAndre Oppermann (so->so_rcv.sb_hiwat / 8 * 7) && 18226741ecf5SAndre Oppermann so->so_rcv.sb_hiwat < 1823603724d3SBjoern A. Zeeb V_tcp_autorcvbuf_max) { 18246741ecf5SAndre Oppermann newsize = 18256741ecf5SAndre Oppermann min(so->so_rcv.sb_hiwat + 1826603724d3SBjoern A. Zeeb V_tcp_autorcvbuf_inc, 1827603724d3SBjoern A. Zeeb V_tcp_autorcvbuf_max); 18286741ecf5SAndre Oppermann } 18296741ecf5SAndre Oppermann /* Start over with next RTT. */ 18306741ecf5SAndre Oppermann tp->rfbuf_ts = 0; 18316741ecf5SAndre Oppermann tp->rfbuf_cnt = 0; 18326741ecf5SAndre Oppermann } else 18336741ecf5SAndre Oppermann tp->rfbuf_cnt += tlen; /* add up */ 18346741ecf5SAndre Oppermann } 18356741ecf5SAndre Oppermann 18366741ecf5SAndre Oppermann /* Add data to socket buffer. */ 18371e4d7da7SRobert Watson SOCKBUF_LOCK(&so->so_rcv); 1838c0b99ffaSRobert Watson if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 1839c1c36a2cSMike Silbersack m_freem(m); 1840c1c36a2cSMike Silbersack } else { 18416741ecf5SAndre Oppermann /* 18426741ecf5SAndre Oppermann * Set new socket buffer size. 18436741ecf5SAndre Oppermann * Give up when limit is reached. 18446741ecf5SAndre Oppermann */ 18456741ecf5SAndre Oppermann if (newsize) 18466741ecf5SAndre Oppermann if (!sbreserve_locked(&so->so_rcv, 18476c8286e4SRobert Watson newsize, so, NULL)) 18486741ecf5SAndre Oppermann so->so_rcv.sb_flags &= ~SB_AUTOSIZE; 1849fb59c426SYoshinobu Inoue m_adj(m, drop_hdrlen); /* delayed header drop */ 18501e4d7da7SRobert Watson sbappendstream_locked(&so->so_rcv, m); 1851c1c36a2cSMike Silbersack } 1852e8949f74SAndre Oppermann /* NB: sorwakeup_locked() does an implicit unlock. */ 18531e4d7da7SRobert Watson sorwakeup_locked(so); 1854c1e5a6e5SAndre Oppermann if (DELAY_ACK(tp, tlen)) { 18553bfd6421SJonathan Lemon tp->t_flags |= TF_DELACK; 1856f498eeeeSDavid Greenman } else { 1857e612a582SDavid Greenman tp->t_flags |= TF_ACKNOW; 1858e612a582SDavid Greenman tcp_output(tp); 1859e612a582SDavid Greenman } 1860a14c749fSJonathan Lemon goto check_delack; 1861df8bae1dSRodney W. Grimes } 1862df8bae1dSRodney W. Grimes } 1863df8bae1dSRodney W. Grimes 1864df8bae1dSRodney W. Grimes /* 1865df8bae1dSRodney W. Grimes * Calculate amount of space in receive window, 1866df8bae1dSRodney W. Grimes * and then do TCP input processing. 1867df8bae1dSRodney W. Grimes * Receive window is amount of space in rcv queue, 1868df8bae1dSRodney W. Grimes * but not less than advertised window. 1869df8bae1dSRodney W. Grimes */ 1870df8bae1dSRodney W. Grimes win = sbspace(&so->so_rcv); 1871df8bae1dSRodney W. Grimes if (win < 0) 1872df8bae1dSRodney W. Grimes win = 0; 187366e39adcSJohn Polstra tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); 1874df8bae1dSRodney W. Grimes 18756741ecf5SAndre Oppermann /* Reset receive buffer auto scaling when not in bulk receive mode. */ 18766741ecf5SAndre Oppermann tp->rfbuf_ts = 0; 18776741ecf5SAndre Oppermann tp->rfbuf_cnt = 0; 18786741ecf5SAndre Oppermann 1879df8bae1dSRodney W. Grimes switch (tp->t_state) { 1880df8bae1dSRodney W. Grimes 1881df8bae1dSRodney W. Grimes /* 1882764d8cefSBill Fenner * If the state is SYN_RECEIVED: 1883764d8cefSBill Fenner * if seg contains an ACK, but not for our SYN/ACK, send a RST. 1884764d8cefSBill Fenner */ 1885764d8cefSBill Fenner case TCPS_SYN_RECEIVED: 1886fb59c426SYoshinobu Inoue if ((thflags & TH_ACK) && 1887fb59c426SYoshinobu Inoue (SEQ_LEQ(th->th_ack, tp->snd_una) || 1888a57815efSBosko Milekic SEQ_GT(th->th_ack, tp->snd_max))) { 1889a57815efSBosko Milekic rstreason = BANDLIM_RST_OPENPORT; 1890a57815efSBosko Milekic goto dropwithreset; 1891a57815efSBosko Milekic } 1892764d8cefSBill Fenner break; 1893764d8cefSBill Fenner 1894764d8cefSBill Fenner /* 1895df8bae1dSRodney W. Grimes * If the state is SYN_SENT: 1896df8bae1dSRodney W. Grimes * if seg contains an ACK, but not for our SYN, drop the input. 1897df8bae1dSRodney W. Grimes * if seg contains a RST, then drop the connection. 1898df8bae1dSRodney W. Grimes * if seg does not contain SYN, then drop it. 1899df8bae1dSRodney W. Grimes * Otherwise this is an acceptable SYN segment 1900df8bae1dSRodney W. Grimes * initialize tp->rcv_nxt and tp->irs 1901df8bae1dSRodney W. Grimes * if seg contains ack then advance tp->snd_una 1902f2512ba1SRui Paulo * if seg contains an ECE and ECN support is enabled, the stream 1903f2512ba1SRui Paulo * is ECN capable. 1904df8bae1dSRodney W. Grimes * if SYN has been acked change to ESTABLISHED else SYN_RCVD state 1905df8bae1dSRodney W. Grimes * arrange for segment to be acked (eventually) 1906df8bae1dSRodney W. Grimes * continue processing rest of data/controls, beginning with URG 1907df8bae1dSRodney W. Grimes */ 1908df8bae1dSRodney W. Grimes case TCPS_SYN_SENT: 1909fb59c426SYoshinobu Inoue if ((thflags & TH_ACK) && 1910fb59c426SYoshinobu Inoue (SEQ_LEQ(th->th_ack, tp->iss) || 1911fb59c426SYoshinobu Inoue SEQ_GT(th->th_ack, tp->snd_max))) { 1912a57815efSBosko Milekic rstreason = BANDLIM_UNLIMITED; 1913a0292f23SGarrett Wollman goto dropwithreset; 1914a0292f23SGarrett Wollman } 191557f60867SMark Johnston if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { 1916d9fae5abSAndriy Gapon TCP_PROBE5(connect__refused, NULL, tp, 1917fa22ce15SAdrian Chadd mtod(m, const char *), tp, th); 1918df8bae1dSRodney W. Grimes tp = tcp_drop(tp, ECONNREFUSED); 191957f60867SMark Johnston } 19200ca3f933SAndre Oppermann if (thflags & TH_RST) 1921df8bae1dSRodney W. Grimes goto drop; 19220ca3f933SAndre Oppermann if (!(thflags & TH_SYN)) 1923df8bae1dSRodney W. Grimes goto drop; 1924464fcfbcSAndre Oppermann 1925fb59c426SYoshinobu Inoue tp->irs = th->th_seq; 1926df8bae1dSRodney W. Grimes tcp_rcvseqinit(tp); 1927fb59c426SYoshinobu Inoue if (thflags & TH_ACK) { 192878b50714SRobert Watson TCPSTAT_INC(tcps_connects); 1929845799c1SAndras Olah soisconnected(so); 1930c488362eSRobert Watson #ifdef MAC 193130d239bcSRobert Watson mac_socketpeer_set_from_mbuf(m, so); 1932c488362eSRobert Watson #endif 1933845799c1SAndras Olah /* Do window scaling on this connection? */ 1934845799c1SAndras Olah if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 1935845799c1SAndras Olah (TF_RCVD_SCALE|TF_REQ_SCALE)) { 1936845799c1SAndras Olah tp->rcv_scale = tp->request_r_scale; 1937845799c1SAndras Olah } 1938766282cbSJohn Baldwin tp->rcv_adv += imin(tp->rcv_wnd, 1939766282cbSJohn Baldwin TCP_MAXWIN << tp->rcv_scale); 1940a0292f23SGarrett Wollman tp->snd_una++; /* SYN is acked */ 1941a0292f23SGarrett Wollman /* 1942a0292f23SGarrett Wollman * If there's data, delay ACK; if there's also a FIN 1943a0292f23SGarrett Wollman * ACKNOW will be turned on later. 1944a0292f23SGarrett Wollman */ 1945c1e5a6e5SAndre Oppermann if (DELAY_ACK(tp, tlen) && tlen != 0) 1946b8152ba7SAndre Oppermann tcp_timer_activate(tp, TT_DELACK, 1947b8152ba7SAndre Oppermann tcp_delacktime); 1948a0292f23SGarrett Wollman else 1949a0292f23SGarrett Wollman tp->t_flags |= TF_ACKNOW; 1950f2512ba1SRui Paulo 1951603724d3SBjoern A. Zeeb if ((thflags & TH_ECE) && V_tcp_do_ecn) { 1952f2512ba1SRui Paulo tp->t_flags |= TF_ECN_PERMIT; 195378b50714SRobert Watson TCPSTAT_INC(tcps_ecn_shs); 1954f2512ba1SRui Paulo } 1955f2512ba1SRui Paulo 1956a0292f23SGarrett Wollman /* 1957a0292f23SGarrett Wollman * Received <SYN,ACK> in SYN_SENT[*] state. 1958a0292f23SGarrett Wollman * Transitions: 1959a0292f23SGarrett Wollman * SYN_SENT --> ESTABLISHED 1960a0292f23SGarrett Wollman * SYN_SENT* --> FIN_WAIT_1 1961a0292f23SGarrett Wollman */ 19629b8b58e0SJonathan Lemon tp->t_starttime = ticks; 1963a0292f23SGarrett Wollman if (tp->t_flags & TF_NEEDFIN) { 196457f60867SMark Johnston tcp_state_change(tp, TCPS_FIN_WAIT_1); 1965a0292f23SGarrett Wollman tp->t_flags &= ~TF_NEEDFIN; 1966fb59c426SYoshinobu Inoue thflags &= ~TH_SYN; 19677ff19458SPaul Traina } else { 196857f60867SMark Johnston tcp_state_change(tp, TCPS_ESTABLISHED); 1969d9fae5abSAndriy Gapon TCP_PROBE5(connect__established, NULL, tp, 1970fa22ce15SAdrian Chadd mtod(m, const char *), tp, th); 1971dbc42409SLawrence Stewart cc_conn_init(tp); 19729077f387SGleb Smirnoff tcp_timer_activate(tp, TT_KEEP, 19739077f387SGleb Smirnoff TP_KEEPIDLE(tp)); 19747ff19458SPaul Traina } 1975a0292f23SGarrett Wollman } else { 1976a0292f23SGarrett Wollman /* 1977c068736aSJeffrey Hsu * Received initial SYN in SYN-SENT[*] state => 1978153edc50SHiren Panchasara * simultaneous open. 1979c068736aSJeffrey Hsu * If it succeeds, connection is * half-synchronized. 1980c068736aSJeffrey Hsu * Otherwise, do 3-way handshake: 1981a0292f23SGarrett Wollman * SYN-SENT -> SYN-RECEIVED 1982a0292f23SGarrett Wollman * SYN-SENT* -> SYN-RECEIVED* 1983a0292f23SGarrett Wollman */ 19844b8e98d6SQing Li tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN); 1985b8152ba7SAndre Oppermann tcp_timer_activate(tp, TT_REXMT, 0); 198657f60867SMark Johnston tcp_state_change(tp, TCPS_SYN_RECEIVED); 1987a0292f23SGarrett Wollman } 1988df8bae1dSRodney W. Grimes 1989252ca428SRobert Watson KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: " 1990252ca428SRobert Watson "ti_locked %d", __func__, ti_locked)); 1991252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 19928501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 19937cfc6904SRobert Watson 1994df8bae1dSRodney W. Grimes /* 1995fb59c426SYoshinobu Inoue * Advance th->th_seq to correspond to first data byte. 1996df8bae1dSRodney W. Grimes * If data, trim to stay within window, 1997df8bae1dSRodney W. Grimes * dropping FIN if necessary. 1998df8bae1dSRodney W. Grimes */ 1999fb59c426SYoshinobu Inoue th->th_seq++; 2000fb59c426SYoshinobu Inoue if (tlen > tp->rcv_wnd) { 2001fb59c426SYoshinobu Inoue todrop = tlen - tp->rcv_wnd; 2002df8bae1dSRodney W. Grimes m_adj(m, -todrop); 2003fb59c426SYoshinobu Inoue tlen = tp->rcv_wnd; 2004fb59c426SYoshinobu Inoue thflags &= ~TH_FIN; 200578b50714SRobert Watson TCPSTAT_INC(tcps_rcvpackafterwin); 200678b50714SRobert Watson TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2007df8bae1dSRodney W. Grimes } 2008fb59c426SYoshinobu Inoue tp->snd_wl1 = th->th_seq - 1; 2009fb59c426SYoshinobu Inoue tp->rcv_up = th->th_seq; 2010a0292f23SGarrett Wollman /* 2011a0292f23SGarrett Wollman * Client side of transaction: already sent SYN and data. 2012a0292f23SGarrett Wollman * If the remote host used T/TCP to validate the SYN, 2013a0292f23SGarrett Wollman * our data will be ACK'd; if so, enter normal data segment 2014a0292f23SGarrett Wollman * processing in the middle of step 5, ack processing. 2015a0292f23SGarrett Wollman * Otherwise, goto step 6. 2016a0292f23SGarrett Wollman */ 2017fb59c426SYoshinobu Inoue if (thflags & TH_ACK) 2018a0292f23SGarrett Wollman goto process_ACK; 2019c068736aSJeffrey Hsu 2020df8bae1dSRodney W. Grimes goto step6; 2021c068736aSJeffrey Hsu 2022a0292f23SGarrett Wollman /* 2023a0292f23SGarrett Wollman * If the state is LAST_ACK or CLOSING or TIME_WAIT: 2024c94c54e4SAndre Oppermann * do normal processing. 2025a0292f23SGarrett Wollman * 2026c94c54e4SAndre Oppermann * NB: Leftover from RFC1644 T/TCP. Cases to be reused later. 2027a0292f23SGarrett Wollman */ 2028a0292f23SGarrett Wollman case TCPS_LAST_ACK: 2029a0292f23SGarrett Wollman case TCPS_CLOSING: 2030a0292f23SGarrett Wollman break; /* continue normal processing */ 2031df8bae1dSRodney W. Grimes } 2032df8bae1dSRodney W. Grimes 2033df8bae1dSRodney W. Grimes /* 2034df8bae1dSRodney W. Grimes * States other than LISTEN or SYN_SENT. 203580ab7c0eSGarrett Wollman * First check the RST flag and sequence number since reset segments 203680ab7c0eSGarrett Wollman * are exempt from the timestamp and connection count tests. This 203780ab7c0eSGarrett Wollman * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix 203880ab7c0eSGarrett Wollman * below which allowed reset segments in half the sequence space 203980ab7c0eSGarrett Wollman * to fall though and be processed (which gives forged reset 204080ab7c0eSGarrett Wollman * segments with a random sequence number a 50 percent chance of 204180ab7c0eSGarrett Wollman * killing a connection). 204280ab7c0eSGarrett Wollman * Then check timestamp, if present. 2043a0292f23SGarrett Wollman * Then check the connection count, if present. 2044df8bae1dSRodney W. Grimes * Then check that at least some bytes of segment are within 2045df8bae1dSRodney W. Grimes * receive window. If segment begins before rcv_nxt, 2046df8bae1dSRodney W. Grimes * drop leading data (and SYN); if nothing left, just ack. 2047df8bae1dSRodney W. Grimes * 204880ab7c0eSGarrett Wollman * 204980ab7c0eSGarrett Wollman * If the RST bit is set, check the sequence number to see 205080ab7c0eSGarrett Wollman * if this is a valid reset segment. 205180ab7c0eSGarrett Wollman * RFC 793 page 37: 205280ab7c0eSGarrett Wollman * In all states except SYN-SENT, all reset (RST) segments 205380ab7c0eSGarrett Wollman * are validated by checking their SEQ-fields. A reset is 205480ab7c0eSGarrett Wollman * valid if its sequence number is in the window. 205580ab7c0eSGarrett Wollman * Note: this does not take into account delayed ACKs, so 205680ab7c0eSGarrett Wollman * we should test against last_ack_sent instead of rcv_nxt. 20571a244a61SJonathan Lemon * The sequence number in the reset segment is normally an 20581a244a61SJonathan Lemon * echo of our outgoing acknowlegement numbers, but some hosts 20591a244a61SJonathan Lemon * send a reset with the sequence number at the rightmost edge 20601a244a61SJonathan Lemon * of our receive window, and we have to handle this case. 206180dd2a81SMike Silbersack * Note 2: Paul Watson's paper "Slipping in the Window" has shown 206280dd2a81SMike Silbersack * that brute force RST attacks are possible. To combat this, 206380dd2a81SMike Silbersack * we use a much stricter check while in the ESTABLISHED state, 206480dd2a81SMike Silbersack * only accepting RSTs where the sequence number is equal to 206580dd2a81SMike Silbersack * last_ack_sent. In all other states (the states in which a 206680dd2a81SMike Silbersack * RST is more likely), the more permissive check is used. 20678e5c87f4SBjoern A. Zeeb * If we have multiple segments in flight, the initial reset 206880ab7c0eSGarrett Wollman * segment sequence numbers will be to the left of last_ack_sent, 206980ab7c0eSGarrett Wollman * but they will eventually catch up. 207080ab7c0eSGarrett Wollman * In any case, it never made sense to trim reset segments to 207180ab7c0eSGarrett Wollman * fit the receive window since RFC 1122 says: 207280ab7c0eSGarrett Wollman * 4.2.2.12 RST Segment: RFC-793 Section 3.4 207380ab7c0eSGarrett Wollman * 207480ab7c0eSGarrett Wollman * A TCP SHOULD allow a received RST segment to include data. 207580ab7c0eSGarrett Wollman * 207680ab7c0eSGarrett Wollman * DISCUSSION 207780ab7c0eSGarrett Wollman * It has been suggested that a RST segment could contain 207880ab7c0eSGarrett Wollman * ASCII text that encoded and explained the cause of the 207980ab7c0eSGarrett Wollman * RST. No standard has yet been established for such 208080ab7c0eSGarrett Wollman * data. 208180ab7c0eSGarrett Wollman * 208280ab7c0eSGarrett Wollman * If the reset segment passes the sequence number test examine 208380ab7c0eSGarrett Wollman * the state: 208480ab7c0eSGarrett Wollman * SYN_RECEIVED STATE: 208580ab7c0eSGarrett Wollman * If passive open, return to LISTEN state. 208680ab7c0eSGarrett Wollman * If active open, inform user that connection was refused. 2087745bab7fSDima Dorfman * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES: 208880ab7c0eSGarrett Wollman * Inform user that connection was reset, and close tcb. 2089e9bd3a37SJonathan M. Bresler * CLOSING, LAST_ACK STATES: 209080ab7c0eSGarrett Wollman * Close the tcb. 2091e9bd3a37SJonathan M. Bresler * TIME_WAIT STATE: 209280ab7c0eSGarrett Wollman * Drop the segment - see Stevens, vol. 2, p. 964 and 209380ab7c0eSGarrett Wollman * RFC 1337. 209480ab7c0eSGarrett Wollman */ 2095fb59c426SYoshinobu Inoue if (thflags & TH_RST) { 209695ad8418SQing Li if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 209795ad8418SQing Li SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { 209880ab7c0eSGarrett Wollman switch (tp->t_state) { 209980ab7c0eSGarrett Wollman 210080ab7c0eSGarrett Wollman case TCPS_SYN_RECEIVED: 210180ab7c0eSGarrett Wollman so->so_error = ECONNREFUSED; 210280ab7c0eSGarrett Wollman goto close; 210380ab7c0eSGarrett Wollman 210480ab7c0eSGarrett Wollman case TCPS_ESTABLISHED: 2105603724d3SBjoern A. Zeeb if (V_tcp_insecure_rst == 0 && 210695ad8418SQing Li !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) && 210795ad8418SQing Li SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) && 210895ad8418SQing Li !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) && 210995ad8418SQing Li SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) { 211078b50714SRobert Watson TCPSTAT_INC(tcps_badrst); 211180dd2a81SMike Silbersack goto drop; 211280dd2a81SMike Silbersack } 2113564aab1fSAndre Oppermann /* FALLTHROUGH */ 211480ab7c0eSGarrett Wollman case TCPS_FIN_WAIT_1: 211580ab7c0eSGarrett Wollman case TCPS_FIN_WAIT_2: 211680ab7c0eSGarrett Wollman case TCPS_CLOSE_WAIT: 211780ab7c0eSGarrett Wollman so->so_error = ECONNRESET; 211880ab7c0eSGarrett Wollman close: 2119252ca428SRobert Watson KASSERT(ti_locked == TI_WLOCKED, 2120252ca428SRobert Watson ("tcp_do_segment: TH_RST 1 ti_locked %d", 2121252ca428SRobert Watson ti_locked)); 2122252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2123252ca428SRobert Watson 212457f60867SMark Johnston tcp_state_change(tp, TCPS_CLOSED); 212578b50714SRobert Watson TCPSTAT_INC(tcps_drops); 212680ab7c0eSGarrett Wollman tp = tcp_close(tp); 212780ab7c0eSGarrett Wollman break; 212880ab7c0eSGarrett Wollman 212980ab7c0eSGarrett Wollman case TCPS_CLOSING: 213080ab7c0eSGarrett Wollman case TCPS_LAST_ACK: 2131252ca428SRobert Watson KASSERT(ti_locked == TI_WLOCKED, 2132252ca428SRobert Watson ("tcp_do_segment: TH_RST 2 ti_locked %d", 2133252ca428SRobert Watson ti_locked)); 2134252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2135252ca428SRobert Watson 213680ab7c0eSGarrett Wollman tp = tcp_close(tp); 213780ab7c0eSGarrett Wollman break; 213880ab7c0eSGarrett Wollman } 213980ab7c0eSGarrett Wollman } 214080ab7c0eSGarrett Wollman goto drop; 214180ab7c0eSGarrett Wollman } 214280ab7c0eSGarrett Wollman 214380ab7c0eSGarrett Wollman /* 2144df8bae1dSRodney W. Grimes * RFC 1323 PAWS: If we have a timestamp reply on this segment 2145df8bae1dSRodney W. Grimes * and it's less than ts_recent, drop it. 2146df8bae1dSRodney W. Grimes */ 2147be2ac88cSJonathan Lemon if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 214880ab7c0eSGarrett Wollman TSTMP_LT(to.to_tsval, tp->ts_recent)) { 2149df8bae1dSRodney W. Grimes 2150df8bae1dSRodney W. Grimes /* Check to see if ts_recent is over 24 days old. */ 2151d8951c8aSBjoern A. Zeeb if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { 2152df8bae1dSRodney W. Grimes /* 2153df8bae1dSRodney W. Grimes * Invalidate ts_recent. If this segment updates 2154df8bae1dSRodney W. Grimes * ts_recent, the age will be reset later and ts_recent 2155df8bae1dSRodney W. Grimes * will get a valid value. If it does not, setting 2156df8bae1dSRodney W. Grimes * ts_recent to zero will at least satisfy the 2157df8bae1dSRodney W. Grimes * requirement that zero be placed in the timestamp 2158df8bae1dSRodney W. Grimes * echo reply when ts_recent isn't valid. The 2159df8bae1dSRodney W. Grimes * age isn't reset until we get a valid ts_recent 2160df8bae1dSRodney W. Grimes * because we don't want out-of-order segments to be 2161df8bae1dSRodney W. Grimes * dropped when ts_recent is old. 2162df8bae1dSRodney W. Grimes */ 2163df8bae1dSRodney W. Grimes tp->ts_recent = 0; 2164df8bae1dSRodney W. Grimes } else { 216578b50714SRobert Watson TCPSTAT_INC(tcps_rcvduppack); 216678b50714SRobert Watson TCPSTAT_ADD(tcps_rcvdupbyte, tlen); 216778b50714SRobert Watson TCPSTAT_INC(tcps_pawsdrop); 216807fd333dSMatthew Dillon if (tlen) 2169df8bae1dSRodney W. Grimes goto dropafterack; 21701ab4789dSMatthew Dillon goto drop; 21711ab4789dSMatthew Dillon } 2172df8bae1dSRodney W. Grimes } 2173df8bae1dSRodney W. Grimes 2174a0292f23SGarrett Wollman /* 217580ab7c0eSGarrett Wollman * In the SYN-RECEIVED state, validate that the packet belongs to 217680ab7c0eSGarrett Wollman * this connection before trimming the data to fit the receive 217780ab7c0eSGarrett Wollman * window. Check the sequence number versus IRS since we know 217880ab7c0eSGarrett Wollman * the sequence numbers haven't wrapped. This is a partial fix 217980ab7c0eSGarrett Wollman * for the "LAND" DoS attack. 218080ab7c0eSGarrett Wollman */ 2181a57815efSBosko Milekic if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { 2182a57815efSBosko Milekic rstreason = BANDLIM_RST_OPENPORT; 2183a57815efSBosko Milekic goto dropwithreset; 2184a57815efSBosko Milekic } 218580ab7c0eSGarrett Wollman 2186fb59c426SYoshinobu Inoue todrop = tp->rcv_nxt - th->th_seq; 2187df8bae1dSRodney W. Grimes if (todrop > 0) { 2188*831ad37eSXin LI if (thflags & TH_SYN) { 2189fb59c426SYoshinobu Inoue thflags &= ~TH_SYN; 2190fb59c426SYoshinobu Inoue th->th_seq++; 2191fb59c426SYoshinobu Inoue if (th->th_urp > 1) 2192fb59c426SYoshinobu Inoue th->th_urp--; 2193df8bae1dSRodney W. Grimes else 2194fb59c426SYoshinobu Inoue thflags &= ~TH_URG; 2195df8bae1dSRodney W. Grimes todrop--; 2196df8bae1dSRodney W. Grimes } 2197df8bae1dSRodney W. Grimes /* 2198dac20301SGarrett Wollman * Following if statement from Stevens, vol. 2, p. 960. 2199df8bae1dSRodney W. Grimes */ 2200fb59c426SYoshinobu Inoue if (todrop > tlen 2201fb59c426SYoshinobu Inoue || (todrop == tlen && (thflags & TH_FIN) == 0)) { 2202dac20301SGarrett Wollman /* 2203dac20301SGarrett Wollman * Any valid FIN must be to the left of the window. 2204dac20301SGarrett Wollman * At this point the FIN must be a duplicate or out 2205dac20301SGarrett Wollman * of sequence; drop it. 2206dac20301SGarrett Wollman */ 2207fb59c426SYoshinobu Inoue thflags &= ~TH_FIN; 2208dac20301SGarrett Wollman 2209df8bae1dSRodney W. Grimes /* 2210dac20301SGarrett Wollman * Send an ACK to resynchronize and drop any data. 2211dac20301SGarrett Wollman * But keep on processing for RST or ACK. 2212df8bae1dSRodney W. Grimes */ 2213dac20301SGarrett Wollman tp->t_flags |= TF_ACKNOW; 2214fb59c426SYoshinobu Inoue todrop = tlen; 221578b50714SRobert Watson TCPSTAT_INC(tcps_rcvduppack); 221678b50714SRobert Watson TCPSTAT_ADD(tcps_rcvdupbyte, todrop); 2217df8bae1dSRodney W. Grimes } else { 221878b50714SRobert Watson TCPSTAT_INC(tcps_rcvpartduppack); 221978b50714SRobert Watson TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); 2220df8bae1dSRodney W. Grimes } 2221fb59c426SYoshinobu Inoue drop_hdrlen += todrop; /* drop from the top afterwards */ 2222fb59c426SYoshinobu Inoue th->th_seq += todrop; 2223fb59c426SYoshinobu Inoue tlen -= todrop; 2224fb59c426SYoshinobu Inoue if (th->th_urp > todrop) 2225fb59c426SYoshinobu Inoue th->th_urp -= todrop; 2226df8bae1dSRodney W. Grimes else { 2227fb59c426SYoshinobu Inoue thflags &= ~TH_URG; 2228fb59c426SYoshinobu Inoue th->th_urp = 0; 2229df8bae1dSRodney W. Grimes } 2230df8bae1dSRodney W. Grimes } 2231df8bae1dSRodney W. Grimes 2232df8bae1dSRodney W. Grimes /* 2233df8bae1dSRodney W. Grimes * If new data are received on a connection after the 2234df8bae1dSRodney W. Grimes * user processes are gone, then RST the other end. 2235df8bae1dSRodney W. Grimes */ 2236df8bae1dSRodney W. Grimes if ((so->so_state & SS_NOFDREF) && 2237fb59c426SYoshinobu Inoue tp->t_state > TCPS_CLOSE_WAIT && tlen) { 2238252ca428SRobert Watson KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && " 2239252ca428SRobert Watson "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked)); 2240252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2241252ca428SRobert Watson 224207dacf03SAndre Oppermann if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { 224307dacf03SAndre Oppermann log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " 224407dacf03SAndre Oppermann "after socket was closed, " 224507dacf03SAndre Oppermann "sending RST and removing tcpcb\n", 2246e31d8aa3SMike Silbersack s, __func__, tcpstates[tp->t_state], tlen); 2247773673c1SAndre Oppermann free(s, M_TCPLOG); 2248773673c1SAndre Oppermann } 2249df8bae1dSRodney W. Grimes tp = tcp_close(tp); 225078b50714SRobert Watson TCPSTAT_INC(tcps_rcvafterclose); 2251a57815efSBosko Milekic rstreason = BANDLIM_UNLIMITED; 2252df8bae1dSRodney W. Grimes goto dropwithreset; 2253df8bae1dSRodney W. Grimes } 2254df8bae1dSRodney W. Grimes 2255df8bae1dSRodney W. Grimes /* 2256df8bae1dSRodney W. Grimes * If segment ends after window, drop trailing data 2257df8bae1dSRodney W. Grimes * (and PUSH and FIN); if nothing left, just ACK. 2258df8bae1dSRodney W. Grimes */ 2259fb59c426SYoshinobu Inoue todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); 2260df8bae1dSRodney W. Grimes if (todrop > 0) { 226178b50714SRobert Watson TCPSTAT_INC(tcps_rcvpackafterwin); 2262fb59c426SYoshinobu Inoue if (todrop >= tlen) { 226378b50714SRobert Watson TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); 2264df8bae1dSRodney W. Grimes /* 2265df8bae1dSRodney W. Grimes * If window is closed can only take segments at 2266df8bae1dSRodney W. Grimes * window edge, and have to drop data and PUSH from 2267df8bae1dSRodney W. Grimes * incoming segments. Continue processing, but 2268df8bae1dSRodney W. Grimes * remember to ack. Otherwise, drop segment 2269df8bae1dSRodney W. Grimes * and ack. 2270df8bae1dSRodney W. Grimes */ 2271fb59c426SYoshinobu Inoue if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { 2272df8bae1dSRodney W. Grimes tp->t_flags |= TF_ACKNOW; 227378b50714SRobert Watson TCPSTAT_INC(tcps_rcvwinprobe); 2274df8bae1dSRodney W. Grimes } else 2275df8bae1dSRodney W. Grimes goto dropafterack; 2276df8bae1dSRodney W. Grimes } else 227778b50714SRobert Watson TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); 2278df8bae1dSRodney W. Grimes m_adj(m, -todrop); 2279fb59c426SYoshinobu Inoue tlen -= todrop; 2280fb59c426SYoshinobu Inoue thflags &= ~(TH_PUSH|TH_FIN); 2281df8bae1dSRodney W. Grimes } 2282df8bae1dSRodney W. Grimes 2283df8bae1dSRodney W. Grimes /* 2284df8bae1dSRodney W. Grimes * If last ACK falls within this segment's sequence numbers, 2285df8bae1dSRodney W. Grimes * record its timestamp. 2286cf09195bSPaul Saab * NOTE: 2287cf09195bSPaul Saab * 1) That the test incorporates suggestions from the latest 2288a0292f23SGarrett Wollman * proposal of the tcplw@cray.com list (Braden 1993/04/26). 2289cf09195bSPaul Saab * 2) That updating only on newer timestamps interferes with 2290cf09195bSPaul Saab * our earlier PAWS tests, so this check should be solely 2291cf09195bSPaul Saab * predicated on the sequence space of this segment. 2292cf09195bSPaul Saab * 3) That we modify the segment boundary check to be 2293cf09195bSPaul Saab * Last.ACK.Sent <= SEG.SEQ + SEG.Len 2294cf09195bSPaul Saab * instead of RFC1323's 2295cf09195bSPaul Saab * Last.ACK.Sent < SEG.SEQ + SEG.Len, 2296cf09195bSPaul Saab * This modified check allows us to overcome RFC1323's 2297cf09195bSPaul Saab * limitations as described in Stevens TCP/IP Illustrated 2298cf09195bSPaul Saab * Vol. 2 p.869. In such cases, we can still calculate the 2299cf09195bSPaul Saab * RTT correctly when RCV.NXT == Last.ACK.Sent. 2300df8bae1dSRodney W. Grimes */ 2301be2ac88cSJonathan Lemon if ((to.to_flags & TOF_TS) != 0 && 2302cf09195bSPaul Saab SEQ_LEQ(th->th_seq, tp->last_ack_sent) && 2303cf09195bSPaul Saab SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + 2304cf09195bSPaul Saab ((thflags & (TH_SYN|TH_FIN)) != 0))) { 2305d8951c8aSBjoern A. Zeeb tp->ts_recent_age = tcp_ts_getticks(); 2306a0292f23SGarrett Wollman tp->ts_recent = to.to_tsval; 2307df8bae1dSRodney W. Grimes } 2308df8bae1dSRodney W. Grimes 2309df8bae1dSRodney W. Grimes /* 2310df8bae1dSRodney W. Grimes * If a SYN is in the window, then this is an 2311df8bae1dSRodney W. Grimes * error and we send an RST and drop the connection. 2312df8bae1dSRodney W. Grimes */ 2313fb59c426SYoshinobu Inoue if (thflags & TH_SYN) { 2314252ca428SRobert Watson KASSERT(ti_locked == TI_WLOCKED, 2315252ca428SRobert Watson ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked)); 2316252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2317252ca428SRobert Watson 2318df8bae1dSRodney W. Grimes tp = tcp_drop(tp, ECONNRESET); 2319a57815efSBosko Milekic rstreason = BANDLIM_UNLIMITED; 2320122aad88SAndre Oppermann goto drop; 2321df8bae1dSRodney W. Grimes } 2322df8bae1dSRodney W. Grimes 2323a0292f23SGarrett Wollman /* 2324a0292f23SGarrett Wollman * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN 2325a0292f23SGarrett Wollman * flag is on (half-synchronized state), then queue data for 2326a0292f23SGarrett Wollman * later processing; else drop segment and return. 2327a0292f23SGarrett Wollman */ 2328fb59c426SYoshinobu Inoue if ((thflags & TH_ACK) == 0) { 2329a0292f23SGarrett Wollman if (tp->t_state == TCPS_SYN_RECEIVED || 2330a0292f23SGarrett Wollman (tp->t_flags & TF_NEEDSYN)) 2331a0292f23SGarrett Wollman goto step6; 23325e1aa279SDavid Malone else if (tp->t_flags & TF_ACKNOW) 23335e1aa279SDavid Malone goto dropafterack; 2334a0292f23SGarrett Wollman else 2335a0292f23SGarrett Wollman goto drop; 2336a0292f23SGarrett Wollman } 2337df8bae1dSRodney W. Grimes 2338df8bae1dSRodney W. Grimes /* 2339df8bae1dSRodney W. Grimes * Ack processing. 2340df8bae1dSRodney W. Grimes */ 2341df8bae1dSRodney W. Grimes switch (tp->t_state) { 2342df8bae1dSRodney W. Grimes 2343df8bae1dSRodney W. Grimes /* 2344764d8cefSBill Fenner * In SYN_RECEIVED state, the ack ACKs our SYN, so enter 2345764d8cefSBill Fenner * ESTABLISHED state and continue processing. 2346764d8cefSBill Fenner * The ACK was checked above. 2347df8bae1dSRodney W. Grimes */ 2348df8bae1dSRodney W. Grimes case TCPS_SYN_RECEIVED: 2349a0292f23SGarrett Wollman 235078b50714SRobert Watson TCPSTAT_INC(tcps_connects); 2351df8bae1dSRodney W. Grimes soisconnected(so); 2352df8bae1dSRodney W. Grimes /* Do window scaling? */ 2353df8bae1dSRodney W. Grimes if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 2354df8bae1dSRodney W. Grimes (TF_RCVD_SCALE|TF_REQ_SCALE)) { 2355df8bae1dSRodney W. Grimes tp->rcv_scale = tp->request_r_scale; 2356464fcfbcSAndre Oppermann tp->snd_wnd = tiwin; 2357df8bae1dSRodney W. Grimes } 2358a0292f23SGarrett Wollman /* 2359a0292f23SGarrett Wollman * Make transitions: 2360a0292f23SGarrett Wollman * SYN-RECEIVED -> ESTABLISHED 2361a0292f23SGarrett Wollman * SYN-RECEIVED* -> FIN-WAIT-1 2362a0292f23SGarrett Wollman */ 23639b8b58e0SJonathan Lemon tp->t_starttime = ticks; 2364a0292f23SGarrett Wollman if (tp->t_flags & TF_NEEDFIN) { 236557f60867SMark Johnston tcp_state_change(tp, TCPS_FIN_WAIT_1); 2366a0292f23SGarrett Wollman tp->t_flags &= ~TF_NEEDFIN; 23677ff19458SPaul Traina } else { 236857f60867SMark Johnston tcp_state_change(tp, TCPS_ESTABLISHED); 2369d9fae5abSAndriy Gapon TCP_PROBE5(accept__established, NULL, tp, 2370fa22ce15SAdrian Chadd mtod(m, const char *), tp, th); 2371dbc42409SLawrence Stewart cc_conn_init(tp); 23729077f387SGleb Smirnoff tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); 23737ff19458SPaul Traina } 2374a0292f23SGarrett Wollman /* 2375a0292f23SGarrett Wollman * If segment contains data or ACK, will call tcp_reass() 2376a0292f23SGarrett Wollman * later; if not, do so now to pass queued data to user. 2377a0292f23SGarrett Wollman */ 2378fb59c426SYoshinobu Inoue if (tlen == 0 && (thflags & TH_FIN) == 0) 2379fb59c426SYoshinobu Inoue (void) tcp_reass(tp, (struct tcphdr *)0, 0, 2380a0292f23SGarrett Wollman (struct mbuf *)0); 238160ee3bb2SAndre Oppermann tp->snd_wl1 = th->th_seq - 1; 238293b0017fSPhilippe Charnier /* FALLTHROUGH */ 2383df8bae1dSRodney W. Grimes 2384df8bae1dSRodney W. Grimes /* 2385df8bae1dSRodney W. Grimes * In ESTABLISHED state: drop duplicate ACKs; ACK out of range 2386df8bae1dSRodney W. Grimes * ACKs. If the ack is in the range 2387fb59c426SYoshinobu Inoue * tp->snd_una < th->th_ack <= tp->snd_max 2388fb59c426SYoshinobu Inoue * then advance tp->snd_una to th->th_ack and drop 2389df8bae1dSRodney W. Grimes * data from the retransmission queue. If this ACK reflects 2390df8bae1dSRodney W. Grimes * more up to date window information we update our window information. 2391df8bae1dSRodney W. Grimes */ 2392df8bae1dSRodney W. Grimes case TCPS_ESTABLISHED: 2393df8bae1dSRodney W. Grimes case TCPS_FIN_WAIT_1: 2394df8bae1dSRodney W. Grimes case TCPS_FIN_WAIT_2: 2395df8bae1dSRodney W. Grimes case TCPS_CLOSE_WAIT: 2396df8bae1dSRodney W. Grimes case TCPS_CLOSING: 2397df8bae1dSRodney W. Grimes case TCPS_LAST_ACK: 23985a53ca16SPaul Saab if (SEQ_GT(th->th_ack, tp->snd_max)) { 239978b50714SRobert Watson TCPSTAT_INC(tcps_rcvacktoomuch); 24005a53ca16SPaul Saab goto dropafterack; 24015a53ca16SPaul Saab } 24023529149eSAndre Oppermann if ((tp->t_flags & TF_SACK_PERMIT) && 2403fc30a251SAndre Oppermann ((to.to_flags & TOF_SACK) || 2404fc30a251SAndre Oppermann !TAILQ_EMPTY(&tp->snd_holes))) 24055a53ca16SPaul Saab tcp_sack_doack(tp, &to, th->th_ack); 240639bc9de5SLawrence Stewart 240739bc9de5SLawrence Stewart /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ 240839bc9de5SLawrence Stewart hhook_run_tcp_est_in(tp, th, &to); 240939bc9de5SLawrence Stewart 2410fb59c426SYoshinobu Inoue if (SEQ_LEQ(th->th_ack, tp->snd_una)) { 241162b90589SPeter Wemm if (tlen == 0 && tiwin == tp->snd_wnd) { 241262b90589SPeter Wemm /* 241362b90589SPeter Wemm * If this is the first time we've seen a 241462b90589SPeter Wemm * FIN from the remote, this is not a 241562b90589SPeter Wemm * duplicate and it needs to be processed 241662b90589SPeter Wemm * normally. This happens during a 241762b90589SPeter Wemm * simultaneous close. 241862b90589SPeter Wemm */ 241962b90589SPeter Wemm if ((thflags & TH_FIN) && 242062b90589SPeter Wemm (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { 242162b90589SPeter Wemm tp->t_dupacks = 0; 242262b90589SPeter Wemm break; 242362b90589SPeter Wemm } 242478b50714SRobert Watson TCPSTAT_INC(tcps_rcvdupack); 2425df8bae1dSRodney W. Grimes /* 2426df8bae1dSRodney W. Grimes * If we have outstanding data (other than 2427df8bae1dSRodney W. Grimes * a window probe), this is a completely 2428df8bae1dSRodney W. Grimes * duplicate ack (ie, window info didn't 24295f30ec9bSEitan Adler * change and FIN isn't set), 24305f30ec9bSEitan Adler * the ack is the biggest we've 2431df8bae1dSRodney W. Grimes * seen and we've seen exactly our rexmt 2432df8bae1dSRodney W. Grimes * threshhold of them, assume a packet 2433df8bae1dSRodney W. Grimes * has been dropped and retransmit it. 2434df8bae1dSRodney W. Grimes * Kludge snd_nxt & the congestion 2435df8bae1dSRodney W. Grimes * window so we send only this one 2436df8bae1dSRodney W. Grimes * packet. 2437df8bae1dSRodney W. Grimes * 2438df8bae1dSRodney W. Grimes * We know we're losing at the current 2439df8bae1dSRodney W. Grimes * window size so do congestion avoidance 2440df8bae1dSRodney W. Grimes * (set ssthresh to half the current window 2441df8bae1dSRodney W. Grimes * and pull our congestion window back to 2442df8bae1dSRodney W. Grimes * the new ssthresh). 2443df8bae1dSRodney W. Grimes * 2444df8bae1dSRodney W. Grimes * Dup acks mean that packets have left the 2445df8bae1dSRodney W. Grimes * network (they're now cached at the receiver) 2446df8bae1dSRodney W. Grimes * so bump cwnd by the amount in the receiver 2447df8bae1dSRodney W. Grimes * to keep a constant cwnd packets in the 2448df8bae1dSRodney W. Grimes * network. 2449f2512ba1SRui Paulo * 2450f2512ba1SRui Paulo * When using TCP ECN, notify the peer that 2451f2512ba1SRui Paulo * we reduced the cwnd. 2452df8bae1dSRodney W. Grimes */ 2453b8152ba7SAndre Oppermann if (!tcp_timer_active(tp, TT_REXMT) || 2454fb59c426SYoshinobu Inoue th->th_ack != tp->snd_una) 2455df8bae1dSRodney W. Grimes tp->t_dupacks = 0; 2456cb942153SJeffrey Hsu else if (++tp->t_dupacks > tcprexmtthresh || 2457dbc42409SLawrence Stewart IN_FASTRECOVERY(tp->t_flags)) { 2458dbc42409SLawrence Stewart cc_ack_received(tp, th, CC_DUPACK); 24593529149eSAndre Oppermann if ((tp->t_flags & TF_SACK_PERMIT) && 2460dbc42409SLawrence Stewart IN_FASTRECOVERY(tp->t_flags)) { 2461808f11b7SPaul Saab int awnd; 246225e6f9edSPaul Saab 246325e6f9edSPaul Saab /* 246425e6f9edSPaul Saab * Compute the amount of data in flight first. 246525e6f9edSPaul Saab * We can inject new data into the pipe iff 246625e6f9edSPaul Saab * we have less than 1/2 the original window's 246725e6f9edSPaul Saab * worth of data in flight. 246825e6f9edSPaul Saab */ 2469808f11b7SPaul Saab awnd = (tp->snd_nxt - tp->snd_fack) + 24700077b016SPaul Saab tp->sackhint.sack_bytes_rexmit; 2471808f11b7SPaul Saab if (awnd < tp->snd_ssthresh) { 247225e6f9edSPaul Saab tp->snd_cwnd += tp->t_maxseg; 247325e6f9edSPaul Saab if (tp->snd_cwnd > tp->snd_ssthresh) 247425e6f9edSPaul Saab tp->snd_cwnd = tp->snd_ssthresh; 247525e6f9edSPaul Saab } 247625e6f9edSPaul Saab } else 247746f58482SJonathan Lemon tp->snd_cwnd += tp->t_maxseg; 247846f58482SJonathan Lemon (void) tcp_output(tp); 247946f58482SJonathan Lemon goto drop; 2480cb942153SJeffrey Hsu } else if (tp->t_dupacks == tcprexmtthresh) { 2481cb942153SJeffrey Hsu tcp_seq onxt = tp->snd_nxt; 2482a0445c2eSJayanth Vijayaraghavan 2483a0445c2eSJayanth Vijayaraghavan /* 2484a0445c2eSJayanth Vijayaraghavan * If we're doing sack, check to 2485a0445c2eSJayanth Vijayaraghavan * see if we're already in sack 2486a0445c2eSJayanth Vijayaraghavan * recovery. If we're not doing sack, 2487a0445c2eSJayanth Vijayaraghavan * check to see if we're in newreno 2488a0445c2eSJayanth Vijayaraghavan * recovery. 2489a0445c2eSJayanth Vijayaraghavan */ 24903529149eSAndre Oppermann if (tp->t_flags & TF_SACK_PERMIT) { 2491dbc42409SLawrence Stewart if (IN_FASTRECOVERY(tp->t_flags)) { 2492a0445c2eSJayanth Vijayaraghavan tp->t_dupacks = 0; 2493a0445c2eSJayanth Vijayaraghavan break; 2494a0445c2eSJayanth Vijayaraghavan } 2495dbc42409SLawrence Stewart } else { 2496a0445c2eSJayanth Vijayaraghavan if (SEQ_LEQ(th->th_ack, 24979d11646dSJeffrey Hsu tp->snd_recover)) { 2498cb942153SJeffrey Hsu tp->t_dupacks = 0; 2499cb942153SJeffrey Hsu break; 250046f58482SJonathan Lemon } 2501a0445c2eSJayanth Vijayaraghavan } 2502dbc42409SLawrence Stewart /* Congestion signal before ack. */ 2503dbc42409SLawrence Stewart cc_cong_signal(tp, th, CC_NDUPACK); 2504dbc42409SLawrence Stewart cc_ack_received(tp, th, CC_DUPACK); 2505b8152ba7SAndre Oppermann tcp_timer_activate(tp, TT_REXMT, 0); 25069b8b58e0SJonathan Lemon tp->t_rtttime = 0; 25073529149eSAndre Oppermann if (tp->t_flags & TF_SACK_PERMIT) { 250878b50714SRobert Watson TCPSTAT_INC( 250978b50714SRobert Watson tcps_sack_recovery_episode); 2510a55db2b6SPaul Saab tp->sack_newdata = tp->snd_nxt; 25118db456bfSPaul Saab tp->snd_cwnd = tp->t_maxseg; 25126d90faf3SPaul Saab (void) tcp_output(tp); 25136d90faf3SPaul Saab goto drop; 25146d90faf3SPaul Saab } 2515fb59c426SYoshinobu Inoue tp->snd_nxt = th->th_ack; 2516df8bae1dSRodney W. Grimes tp->snd_cwnd = tp->t_maxseg; 2517df8bae1dSRodney W. Grimes (void) tcp_output(tp); 251848d2549cSJeffrey Hsu KASSERT(tp->snd_limited <= 2, 2519302ce8d6SAndre Oppermann ("%s: tp->snd_limited too big", 2520302ce8d6SAndre Oppermann __func__)); 2521df8bae1dSRodney W. Grimes tp->snd_cwnd = tp->snd_ssthresh + 252248d2549cSJeffrey Hsu tp->t_maxseg * 252348d2549cSJeffrey Hsu (tp->t_dupacks - tp->snd_limited); 2524df8bae1dSRodney W. Grimes if (SEQ_GT(onxt, tp->snd_nxt)) 2525df8bae1dSRodney W. Grimes tp->snd_nxt = onxt; 2526df8bae1dSRodney W. Grimes goto drop; 2527603724d3SBjoern A. Zeeb } else if (V_tcp_do_rfc3042) { 2528dbc42409SLawrence Stewart cc_ack_received(tp, th, CC_DUPACK); 2529582a954bSJeffrey Hsu u_long oldcwnd = tp->snd_cwnd; 253048d2549cSJeffrey Hsu tcp_seq oldsndmax = tp->snd_max; 253148d2549cSJeffrey Hsu u_int sent; 25325628dd08SAndre Oppermann int avail; 253389c02376SJeffrey Hsu 2534582a954bSJeffrey Hsu KASSERT(tp->t_dupacks == 1 || 2535582a954bSJeffrey Hsu tp->t_dupacks == 2, 2536302ce8d6SAndre Oppermann ("%s: dupacks not 1 or 2", 2537302ce8d6SAndre Oppermann __func__)); 253861a36e3dSJeffrey Hsu if (tp->t_dupacks == 1) 253948d2549cSJeffrey Hsu tp->snd_limited = 0; 254061a36e3dSJeffrey Hsu tp->snd_cwnd = 254161a36e3dSJeffrey Hsu (tp->snd_nxt - tp->snd_una) + 254261a36e3dSJeffrey Hsu (tp->t_dupacks - tp->snd_limited) * 254361a36e3dSJeffrey Hsu tp->t_maxseg; 25445628dd08SAndre Oppermann /* 25455628dd08SAndre Oppermann * Only call tcp_output when there 25465628dd08SAndre Oppermann * is new data available to be sent. 25475628dd08SAndre Oppermann * Otherwise we would send pure ACKs. 25485628dd08SAndre Oppermann */ 25495628dd08SAndre Oppermann SOCKBUF_LOCK(&so->so_snd); 25505628dd08SAndre Oppermann avail = so->so_snd.sb_cc - 25515628dd08SAndre Oppermann (tp->snd_nxt - tp->snd_una); 25525628dd08SAndre Oppermann SOCKBUF_UNLOCK(&so->so_snd); 25535628dd08SAndre Oppermann if (avail > 0) 2554582a954bSJeffrey Hsu (void) tcp_output(tp); 255548d2549cSJeffrey Hsu sent = tp->snd_max - oldsndmax; 255648d2549cSJeffrey Hsu if (sent > tp->t_maxseg) { 255789c02376SJeffrey Hsu KASSERT((tp->t_dupacks == 2 && 255889c02376SJeffrey Hsu tp->snd_limited == 0) || 255989c02376SJeffrey Hsu (sent == tp->t_maxseg + 1 && 256089c02376SJeffrey Hsu tp->t_flags & TF_SENTFIN), 2561302ce8d6SAndre Oppermann ("%s: sent too much", 2562302ce8d6SAndre Oppermann __func__)); 256348d2549cSJeffrey Hsu tp->snd_limited = 2; 256448d2549cSJeffrey Hsu } else if (sent > 0) 256548d2549cSJeffrey Hsu ++tp->snd_limited; 2566582a954bSJeffrey Hsu tp->snd_cwnd = oldcwnd; 2567582a954bSJeffrey Hsu goto drop; 2568df8bae1dSRodney W. Grimes } 2569df8bae1dSRodney W. Grimes } else 2570df8bae1dSRodney W. Grimes tp->t_dupacks = 0; 2571df8bae1dSRodney W. Grimes break; 2572df8bae1dSRodney W. Grimes } 2573cb942153SJeffrey Hsu 2574302ce8d6SAndre Oppermann KASSERT(SEQ_GT(th->th_ack, tp->snd_una), 2575302ce8d6SAndre Oppermann ("%s: th_ack <= snd_una", __func__)); 2576cb942153SJeffrey Hsu 2577df8bae1dSRodney W. Grimes /* 2578df8bae1dSRodney W. Grimes * If the congestion window was inflated to account 2579df8bae1dSRodney W. Grimes * for the other side's cached packets, retract it. 2580df8bae1dSRodney W. Grimes */ 2581dbc42409SLawrence Stewart if (IN_FASTRECOVERY(tp->t_flags)) { 2582cb942153SJeffrey Hsu if (SEQ_LT(th->th_ack, tp->snd_recover)) { 25833529149eSAndre Oppermann if (tp->t_flags & TF_SACK_PERMIT) 25846d90faf3SPaul Saab tcp_sack_partialack(tp, th); 25856d90faf3SPaul Saab else 2586c068736aSJeffrey Hsu tcp_newreno_partial_ack(tp, th); 2587dbc42409SLawrence Stewart } else 2588dbc42409SLawrence Stewart cc_post_recovery(tp, th); 258946f58482SJonathan Lemon } 2590cb942153SJeffrey Hsu tp->t_dupacks = 0; 2591a0292f23SGarrett Wollman /* 2592a0292f23SGarrett Wollman * If we reach this point, ACK is not a duplicate, 2593a0292f23SGarrett Wollman * i.e., it ACKs something we sent. 2594a0292f23SGarrett Wollman */ 2595a0292f23SGarrett Wollman if (tp->t_flags & TF_NEEDSYN) { 2596a0292f23SGarrett Wollman /* 2597a0292f23SGarrett Wollman * T/TCP: Connection was half-synchronized, and our 2598a0292f23SGarrett Wollman * SYN has been ACK'd (so connection is now fully 259907e43e10SAndras Olah * synchronized). Go to non-starred state, 260007e43e10SAndras Olah * increment snd_una for ACK of SYN, and check if 260107e43e10SAndras Olah * we can do window scaling. 2602a0292f23SGarrett Wollman */ 2603a0292f23SGarrett Wollman tp->t_flags &= ~TF_NEEDSYN; 2604a0292f23SGarrett Wollman tp->snd_una++; 260507e43e10SAndras Olah /* Do window scaling? */ 260607e43e10SAndras Olah if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == 260707e43e10SAndras Olah (TF_RCVD_SCALE|TF_REQ_SCALE)) { 260807e43e10SAndras Olah tp->rcv_scale = tp->request_r_scale; 2609464fcfbcSAndre Oppermann /* Send window already scaled. */ 261007e43e10SAndras Olah } 2611a0292f23SGarrett Wollman } 2612a0292f23SGarrett Wollman 2613a0292f23SGarrett Wollman process_ACK: 26148501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 26157cfc6904SRobert Watson 2616dbc42409SLawrence Stewart acked = BYTES_THIS_ACK(tp, th); 261778b50714SRobert Watson TCPSTAT_INC(tcps_rcvackpack); 261878b50714SRobert Watson TCPSTAT_ADD(tcps_rcvackbyte, acked); 2619df8bae1dSRodney W. Grimes 2620df8bae1dSRodney W. Grimes /* 26219b8b58e0SJonathan Lemon * If we just performed our first retransmit, and the ACK 26229b8b58e0SJonathan Lemon * arrives within our recovery window, then it was a mistake 26239b8b58e0SJonathan Lemon * to do the retransmit in the first place. Recover our 26249b8b58e0SJonathan Lemon * original cwnd and ssthresh, and proceed to transmit where 26259b8b58e0SJonathan Lemon * we left off. 26269b8b58e0SJonathan Lemon */ 2627672dc4aeSJohn Baldwin if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && 2628672dc4aeSJohn Baldwin (int)(ticks - tp->t_badrxtwin) < 0) 2629dbc42409SLawrence Stewart cc_cong_signal(tp, th, CC_RTO_ERR); 26309b8b58e0SJonathan Lemon 26319b8b58e0SJonathan Lemon /* 2632df8bae1dSRodney W. Grimes * If we have a timestamp reply, update smoothed 2633df8bae1dSRodney W. Grimes * round trip time. If no timestamp is present but 2634df8bae1dSRodney W. Grimes * transmit timer is running and timed sequence 2635df8bae1dSRodney W. Grimes * number was acked, update smoothed round trip time. 2636df8bae1dSRodney W. Grimes * Since we now have an rtt measurement, cancel the 2637df8bae1dSRodney W. Grimes * timer backoff (cf., Phil Karn's retransmit alg.). 2638df8bae1dSRodney W. Grimes * Recompute the initial retransmit timer. 2639fa55172bSMatthew Dillon * 2640fa55172bSMatthew Dillon * Some boxes send broken timestamp replies 2641fa55172bSMatthew Dillon * during the SYN+ACK phase, ignore 2642fa55172bSMatthew Dillon * timestamps of 0 or we could calculate a 2643fa55172bSMatthew Dillon * huge RTT and blow up the retransmit timer. 2644df8bae1dSRodney W. Grimes */ 2645d8951c8aSBjoern A. Zeeb if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { 2646d8951c8aSBjoern A. Zeeb u_int t; 2647d8951c8aSBjoern A. Zeeb 2648d8951c8aSBjoern A. Zeeb t = tcp_ts_getticks() - to.to_tsecr; 2649d8951c8aSBjoern A. Zeeb if (!tp->t_rttlow || tp->t_rttlow > t) 2650d8951c8aSBjoern A. Zeeb tp->t_rttlow = t; 2651d8951c8aSBjoern A. Zeeb tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); 2652fa55172bSMatthew Dillon } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { 2653eaf80179SAndre Oppermann if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) 2654eaf80179SAndre Oppermann tp->t_rttlow = ticks - tp->t_rtttime; 26559b8b58e0SJonathan Lemon tcp_xmit_timer(tp, ticks - tp->t_rtttime); 2656fa55172bSMatthew Dillon } 2657df8bae1dSRodney W. Grimes 2658df8bae1dSRodney W. Grimes /* 2659df8bae1dSRodney W. Grimes * If all outstanding data is acked, stop retransmit 2660df8bae1dSRodney W. Grimes * timer and remember to restart (more output or persist). 2661df8bae1dSRodney W. Grimes * If there is more data to be acked, restart retransmit 2662df8bae1dSRodney W. Grimes * timer, using current (possibly backed-off) value. 2663df8bae1dSRodney W. Grimes */ 2664fb59c426SYoshinobu Inoue if (th->th_ack == tp->snd_max) { 2665b8152ba7SAndre Oppermann tcp_timer_activate(tp, TT_REXMT, 0); 2666df8bae1dSRodney W. Grimes needoutput = 1; 2667b8152ba7SAndre Oppermann } else if (!tcp_timer_active(tp, TT_PERSIST)) 2668b8152ba7SAndre Oppermann tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur); 2669a0292f23SGarrett Wollman 2670a0292f23SGarrett Wollman /* 2671a0292f23SGarrett Wollman * If no data (only SYN) was ACK'd, 2672a0292f23SGarrett Wollman * skip rest of ACK processing. 2673a0292f23SGarrett Wollman */ 2674a0292f23SGarrett Wollman if (acked == 0) 2675a0292f23SGarrett Wollman goto step6; 2676a0292f23SGarrett Wollman 2677df8bae1dSRodney W. Grimes /* 2678dbc42409SLawrence Stewart * Let the congestion control algorithm update congestion 2679dbc42409SLawrence Stewart * control related information. This typically means increasing 2680dbc42409SLawrence Stewart * the congestion window. 2681df8bae1dSRodney W. Grimes */ 2682dbc42409SLawrence Stewart cc_ack_received(tp, th, CC_ACK); 2683dbc42409SLawrence Stewart 26845905999bSRobert Watson SOCKBUF_LOCK(&so->so_snd); 2685df8bae1dSRodney W. Grimes if (acked > so->so_snd.sb_cc) { 268660ee3bb2SAndre Oppermann tp->snd_wnd -= so->so_snd.sb_cc; 2687c11a15bfSGleb Smirnoff mfree = sbcut_locked(&so->so_snd, 2688c11a15bfSGleb Smirnoff (int)so->so_snd.sb_cc); 2689df8bae1dSRodney W. Grimes ourfinisacked = 1; 2690df8bae1dSRodney W. Grimes } else { 2691c11a15bfSGleb Smirnoff mfree = sbcut_locked(&so->so_snd, acked); 269260ee3bb2SAndre Oppermann tp->snd_wnd -= acked; 2693df8bae1dSRodney W. Grimes ourfinisacked = 0; 2694df8bae1dSRodney W. Grimes } 2695564aab1fSAndre Oppermann /* NB: sowwakeup_locked() does an implicit unlock. */ 26961e4d7da7SRobert Watson sowwakeup_locked(so); 2697c11a15bfSGleb Smirnoff m_freem(mfree); 2698e8949f74SAndre Oppermann /* Detect una wraparound. */ 2699dbc42409SLawrence Stewart if (!IN_RECOVERY(tp->t_flags) && 27009d11646dSJeffrey Hsu SEQ_GT(tp->snd_una, tp->snd_recover) && 27019d11646dSJeffrey Hsu SEQ_LEQ(th->th_ack, tp->snd_recover)) 27029d11646dSJeffrey Hsu tp->snd_recover = th->th_ack - 1; 2703dbc42409SLawrence Stewart /* XXXLAS: Can this be moved up into cc_post_recovery? */ 2704dbc42409SLawrence Stewart if (IN_RECOVERY(tp->t_flags) && 270524cb0f22SLawrence Stewart SEQ_GEQ(th->th_ack, tp->snd_recover)) { 2706dbc42409SLawrence Stewart EXIT_RECOVERY(tp->t_flags); 270724cb0f22SLawrence Stewart } 2708fb59c426SYoshinobu Inoue tp->snd_una = th->th_ack; 27093529149eSAndre Oppermann if (tp->t_flags & TF_SACK_PERMIT) { 27106d90faf3SPaul Saab if (SEQ_GT(tp->snd_una, tp->snd_recover)) 27116d90faf3SPaul Saab tp->snd_recover = tp->snd_una; 27126d90faf3SPaul Saab } 2713df8bae1dSRodney W. Grimes if (SEQ_LT(tp->snd_nxt, tp->snd_una)) 2714df8bae1dSRodney W. Grimes tp->snd_nxt = tp->snd_una; 2715df8bae1dSRodney W. Grimes 2716df8bae1dSRodney W. Grimes switch (tp->t_state) { 2717df8bae1dSRodney W. Grimes 2718df8bae1dSRodney W. Grimes /* 2719df8bae1dSRodney W. Grimes * In FIN_WAIT_1 STATE in addition to the processing 2720df8bae1dSRodney W. Grimes * for the ESTABLISHED state if our FIN is now acknowledged 2721df8bae1dSRodney W. Grimes * then enter FIN_WAIT_2. 2722df8bae1dSRodney W. Grimes */ 2723df8bae1dSRodney W. Grimes case TCPS_FIN_WAIT_1: 2724df8bae1dSRodney W. Grimes if (ourfinisacked) { 2725df8bae1dSRodney W. Grimes /* 2726df8bae1dSRodney W. Grimes * If we can't receive any more 2727df8bae1dSRodney W. Grimes * data, then closing user can proceed. 2728df8bae1dSRodney W. Grimes * Starting the timer is contrary to the 2729df8bae1dSRodney W. Grimes * specification, but if we don't get a FIN 2730df8bae1dSRodney W. Grimes * we'll hang forever. 2731e8949f74SAndre Oppermann * 2732e8949f74SAndre Oppermann * XXXjl: 2733340c35deSJonathan Lemon * we should release the tp also, and use a 2734340c35deSJonathan Lemon * compressed state. 2735340c35deSJonathan Lemon */ 2736c0b99ffaSRobert Watson if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { 273703e49181SSeigo Tanimura soisdisconnected(so); 27389077f387SGleb Smirnoff tcp_timer_activate(tp, TT_2MSL, 27399077f387SGleb Smirnoff (tcp_fast_finwait2_recycle ? 27409077f387SGleb Smirnoff tcp_finwait2_timeout : 27419077f387SGleb Smirnoff TP_MAXIDLE(tp))); 27424cc20ab1SSeigo Tanimura } 274357f60867SMark Johnston tcp_state_change(tp, TCPS_FIN_WAIT_2); 2744df8bae1dSRodney W. Grimes } 2745df8bae1dSRodney W. Grimes break; 2746df8bae1dSRodney W. Grimes 2747df8bae1dSRodney W. Grimes /* 2748df8bae1dSRodney W. Grimes * In CLOSING STATE in addition to the processing for 2749df8bae1dSRodney W. Grimes * the ESTABLISHED state if the ACK acknowledges our FIN 2750df8bae1dSRodney W. Grimes * then enter the TIME-WAIT state, otherwise ignore 2751df8bae1dSRodney W. Grimes * the segment. 2752df8bae1dSRodney W. Grimes */ 2753df8bae1dSRodney W. Grimes case TCPS_CLOSING: 2754df8bae1dSRodney W. Grimes if (ourfinisacked) { 2755252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 275611a20fb8SJeffrey Hsu tcp_twstart(tp); 2757603724d3SBjoern A. Zeeb INP_INFO_WUNLOCK(&V_tcbinfo); 2758340c35deSJonathan Lemon m_freem(m); 2759679d9708SAndre Oppermann return; 2760df8bae1dSRodney W. Grimes } 2761df8bae1dSRodney W. Grimes break; 2762df8bae1dSRodney W. Grimes 2763df8bae1dSRodney W. Grimes /* 2764df8bae1dSRodney W. Grimes * In LAST_ACK, we may still be waiting for data to drain 2765df8bae1dSRodney W. Grimes * and/or to be acked, as well as for the ack of our FIN. 2766df8bae1dSRodney W. Grimes * If our FIN is now acknowledged, delete the TCB, 2767df8bae1dSRodney W. Grimes * enter the closed state and return. 2768df8bae1dSRodney W. Grimes */ 2769df8bae1dSRodney W. Grimes case TCPS_LAST_ACK: 2770df8bae1dSRodney W. Grimes if (ourfinisacked) { 2771252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2772df8bae1dSRodney W. Grimes tp = tcp_close(tp); 2773df8bae1dSRodney W. Grimes goto drop; 2774df8bae1dSRodney W. Grimes } 2775df8bae1dSRodney W. Grimes break; 2776df8bae1dSRodney W. Grimes } 2777df8bae1dSRodney W. Grimes } 2778df8bae1dSRodney W. Grimes 2779df8bae1dSRodney W. Grimes step6: 27808501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 27817cfc6904SRobert Watson 2782df8bae1dSRodney W. Grimes /* 278360ee3bb2SAndre Oppermann * Update window information. 278460ee3bb2SAndre Oppermann * Don't look at window if no ACK: TAC's send garbage on first SYN. 2785df8bae1dSRodney W. Grimes */ 278660ee3bb2SAndre Oppermann if ((thflags & TH_ACK) && 278760ee3bb2SAndre Oppermann (SEQ_LT(tp->snd_wl1, th->th_seq) || 278860ee3bb2SAndre Oppermann (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || 278960ee3bb2SAndre Oppermann (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { 279060ee3bb2SAndre Oppermann /* keep track of pure window updates */ 279160ee3bb2SAndre Oppermann if (tlen == 0 && 279260ee3bb2SAndre Oppermann tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) 279378b50714SRobert Watson TCPSTAT_INC(tcps_rcvwinupd); 2794df8bae1dSRodney W. Grimes tp->snd_wnd = tiwin; 279560ee3bb2SAndre Oppermann tp->snd_wl1 = th->th_seq; 279660ee3bb2SAndre Oppermann tp->snd_wl2 = th->th_ack; 2797df8bae1dSRodney W. Grimes if (tp->snd_wnd > tp->max_sndwnd) 2798df8bae1dSRodney W. Grimes tp->max_sndwnd = tp->snd_wnd; 279960ee3bb2SAndre Oppermann needoutput = 1; 2800df8bae1dSRodney W. Grimes } 2801df8bae1dSRodney W. Grimes 2802df8bae1dSRodney W. Grimes /* 2803df8bae1dSRodney W. Grimes * Process segments with URG. 2804df8bae1dSRodney W. Grimes */ 2805fb59c426SYoshinobu Inoue if ((thflags & TH_URG) && th->th_urp && 2806df8bae1dSRodney W. Grimes TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2807df8bae1dSRodney W. Grimes /* 2808df8bae1dSRodney W. Grimes * This is a kludge, but if we receive and accept 2809df8bae1dSRodney W. Grimes * random urgent pointers, we'll crash in 2810df8bae1dSRodney W. Grimes * soreceive. It's hard to imagine someone 2811df8bae1dSRodney W. Grimes * actually wanting to send this much urgent data. 2812df8bae1dSRodney W. Grimes */ 281318ad5842SRobert Watson SOCKBUF_LOCK(&so->so_rcv); 2814fb59c426SYoshinobu Inoue if (th->th_urp + so->so_rcv.sb_cc > sb_max) { 2815fb59c426SYoshinobu Inoue th->th_urp = 0; /* XXX */ 2816fb59c426SYoshinobu Inoue thflags &= ~TH_URG; /* XXX */ 281718ad5842SRobert Watson SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ 2818df8bae1dSRodney W. Grimes goto dodata; /* XXX */ 2819df8bae1dSRodney W. Grimes } 2820df8bae1dSRodney W. Grimes /* 2821df8bae1dSRodney W. Grimes * If this segment advances the known urgent pointer, 2822df8bae1dSRodney W. Grimes * then mark the data stream. This should not happen 2823df8bae1dSRodney W. Grimes * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since 2824df8bae1dSRodney W. Grimes * a FIN has been received from the remote side. 2825df8bae1dSRodney W. Grimes * In these states we ignore the URG. 2826df8bae1dSRodney W. Grimes * 2827df8bae1dSRodney W. Grimes * According to RFC961 (Assigned Protocols), 2828df8bae1dSRodney W. Grimes * the urgent pointer points to the last octet 2829df8bae1dSRodney W. Grimes * of urgent data. We continue, however, 2830df8bae1dSRodney W. Grimes * to consider it to indicate the first octet 2831df8bae1dSRodney W. Grimes * of data past the urgent section as the original 2832df8bae1dSRodney W. Grimes * spec states (in one of two places). 2833df8bae1dSRodney W. Grimes */ 2834fb59c426SYoshinobu Inoue if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { 2835fb59c426SYoshinobu Inoue tp->rcv_up = th->th_seq + th->th_urp; 2836df8bae1dSRodney W. Grimes so->so_oobmark = so->so_rcv.sb_cc + 2837df8bae1dSRodney W. Grimes (tp->rcv_up - tp->rcv_nxt) - 1; 2838927c5ceaSRobert Watson if (so->so_oobmark == 0) 2839c0b99ffaSRobert Watson so->so_rcv.sb_state |= SBS_RCVATMARK; 2840df8bae1dSRodney W. Grimes sohasoutofband(so); 2841df8bae1dSRodney W. Grimes tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); 2842df8bae1dSRodney W. Grimes } 284318ad5842SRobert Watson SOCKBUF_UNLOCK(&so->so_rcv); 2844df8bae1dSRodney W. Grimes /* 2845df8bae1dSRodney W. Grimes * Remove out of band data so doesn't get presented to user. 2846df8bae1dSRodney W. Grimes * This can happen independent of advancing the URG pointer, 2847df8bae1dSRodney W. Grimes * but if two URG's are pending at once, some out-of-band 2848df8bae1dSRodney W. Grimes * data may creep in... ick. 2849df8bae1dSRodney W. Grimes */ 285030613f56SJeffrey Hsu if (th->th_urp <= (u_long)tlen && 285130613f56SJeffrey Hsu !(so->so_options & SO_OOBINLINE)) { 285230613f56SJeffrey Hsu /* hdr drop is delayed */ 285330613f56SJeffrey Hsu tcp_pulloutofband(so, th, m, drop_hdrlen); 285430613f56SJeffrey Hsu } 2855c068736aSJeffrey Hsu } else { 2856df8bae1dSRodney W. Grimes /* 2857df8bae1dSRodney W. Grimes * If no out of band data is expected, 2858df8bae1dSRodney W. Grimes * pull receive urgent pointer along 2859df8bae1dSRodney W. Grimes * with the receive window. 2860df8bae1dSRodney W. Grimes */ 2861df8bae1dSRodney W. Grimes if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) 2862df8bae1dSRodney W. Grimes tp->rcv_up = tp->rcv_nxt; 2863c068736aSJeffrey Hsu } 2864df8bae1dSRodney W. Grimes dodata: /* XXX */ 28658501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 28667cfc6904SRobert Watson 2867df8bae1dSRodney W. Grimes /* 2868df8bae1dSRodney W. Grimes * Process the segment text, merging it into the TCP sequencing queue, 2869df8bae1dSRodney W. Grimes * and arranging for acknowledgment of receipt if necessary. 2870df8bae1dSRodney W. Grimes * This process logically involves adjusting tp->rcv_wnd as data 2871df8bae1dSRodney W. Grimes * is presented to the user (this happens in tcp_usrreq.c, 2872df8bae1dSRodney W. Grimes * case PRU_RCVD). If a FIN has already been received on this 2873df8bae1dSRodney W. Grimes * connection then we just ignore the text. 2874df8bae1dSRodney W. Grimes */ 2875fb59c426SYoshinobu Inoue if ((tlen || (thflags & TH_FIN)) && 2876df8bae1dSRodney W. Grimes TCPS_HAVERCVDFIN(tp->t_state) == 0) { 287769e03620SPaul Saab tcp_seq save_start = th->th_seq; 2878fb59c426SYoshinobu Inoue m_adj(m, drop_hdrlen); /* delayed header drop */ 2879e4b64281SJesper Skriver /* 2880c068736aSJeffrey Hsu * Insert segment which includes th into TCP reassembly queue 2881c068736aSJeffrey Hsu * with control block tp. Set thflags to whether reassembly now 2882c068736aSJeffrey Hsu * includes a segment with FIN. This handles the common case 2883c068736aSJeffrey Hsu * inline (segment is the next to be received on an established 2884c068736aSJeffrey Hsu * connection, and the queue is empty), avoiding linkage into 2885c068736aSJeffrey Hsu * and removal from the queue and repetition of various 2886c068736aSJeffrey Hsu * conversions. 2887c068736aSJeffrey Hsu * Set DELACK for segments received in order, but ack 2888c068736aSJeffrey Hsu * immediately when segments are out of order (so 2889c068736aSJeffrey Hsu * fast retransmit can work). 2890e4b64281SJesper Skriver */ 2891e407b67bSGleb Smirnoff if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL && 2892e4b64281SJesper Skriver TCPS_HAVEESTABLISHED(tp->t_state)) { 2893c1e5a6e5SAndre Oppermann if (DELAY_ACK(tp, tlen)) 28943bfd6421SJonathan Lemon tp->t_flags |= TF_DELACK; 2895e4b64281SJesper Skriver else 2896e4b64281SJesper Skriver tp->t_flags |= TF_ACKNOW; 2897e4b64281SJesper Skriver tp->rcv_nxt += tlen; 2898e4b64281SJesper Skriver thflags = th->th_flags & TH_FIN; 289978b50714SRobert Watson TCPSTAT_INC(tcps_rcvpack); 290078b50714SRobert Watson TCPSTAT_ADD(tcps_rcvbyte, tlen); 2901e4b64281SJesper Skriver ND6_HINT(tp); 29021e4d7da7SRobert Watson SOCKBUF_LOCK(&so->so_rcv); 2903c0b99ffaSRobert Watson if (so->so_rcv.sb_state & SBS_CANTRCVMORE) 2904c1c36a2cSMike Silbersack m_freem(m); 2905c1c36a2cSMike Silbersack else 29061e4d7da7SRobert Watson sbappendstream_locked(&so->so_rcv, m); 2907e8949f74SAndre Oppermann /* NB: sorwakeup_locked() does an implicit unlock. */ 29081e4d7da7SRobert Watson sorwakeup_locked(so); 2909e4b64281SJesper Skriver } else { 2910e8949f74SAndre Oppermann /* 2911e8949f74SAndre Oppermann * XXX: Due to the header drop above "th" is 2912e8949f74SAndre Oppermann * theoretically invalid by now. Fortunately 2913e8949f74SAndre Oppermann * m_adj() doesn't actually frees any mbufs 2914e8949f74SAndre Oppermann * when trimming from the head. 2915e8949f74SAndre Oppermann */ 2916e4b64281SJesper Skriver thflags = tcp_reass(tp, th, &tlen, m); 2917e4b64281SJesper Skriver tp->t_flags |= TF_ACKNOW; 2918e4b64281SJesper Skriver } 29193529149eSAndre Oppermann if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT)) 2920f194524fSAndre Oppermann tcp_update_sack_list(tp, save_start, save_start + tlen); 2921302ce8d6SAndre Oppermann #if 0 2922df8bae1dSRodney W. Grimes /* 2923df8bae1dSRodney W. Grimes * Note the amount of data that peer has sent into 2924df8bae1dSRodney W. Grimes * our window, in order to estimate the sender's 2925df8bae1dSRodney W. Grimes * buffer size. 2926302ce8d6SAndre Oppermann * XXX: Unused. 2927df8bae1dSRodney W. Grimes */ 2928f701e30dSJohn Baldwin if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) 2929df8bae1dSRodney W. Grimes len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); 2930f701e30dSJohn Baldwin else 2931f701e30dSJohn Baldwin len = so->so_rcv.sb_hiwat; 2932302ce8d6SAndre Oppermann #endif 2933df8bae1dSRodney W. Grimes } else { 2934df8bae1dSRodney W. Grimes m_freem(m); 2935fb59c426SYoshinobu Inoue thflags &= ~TH_FIN; 2936df8bae1dSRodney W. Grimes } 2937df8bae1dSRodney W. Grimes 2938df8bae1dSRodney W. Grimes /* 2939df8bae1dSRodney W. Grimes * If FIN is received ACK the FIN and let the user know 2940df8bae1dSRodney W. Grimes * that the connection is closing. 2941df8bae1dSRodney W. Grimes */ 2942fb59c426SYoshinobu Inoue if (thflags & TH_FIN) { 2943df8bae1dSRodney W. Grimes if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { 2944df8bae1dSRodney W. Grimes socantrcvmore(so); 2945a0292f23SGarrett Wollman /* 2946a0292f23SGarrett Wollman * If connection is half-synchronized 2947d3eede9dSAndras Olah * (ie NEEDSYN flag on) then delay ACK, 2948a0292f23SGarrett Wollman * so it may be piggybacked when SYN is sent. 2949a0292f23SGarrett Wollman * Otherwise, since we received a FIN then no 2950a0292f23SGarrett Wollman * more input can be expected, send ACK now. 2951a0292f23SGarrett Wollman */ 29523bfd6421SJonathan Lemon if (tp->t_flags & TF_NEEDSYN) 29533bfd6421SJonathan Lemon tp->t_flags |= TF_DELACK; 2954a0292f23SGarrett Wollman else 2955df8bae1dSRodney W. Grimes tp->t_flags |= TF_ACKNOW; 2956df8bae1dSRodney W. Grimes tp->rcv_nxt++; 2957df8bae1dSRodney W. Grimes } 2958df8bae1dSRodney W. Grimes switch (tp->t_state) { 2959df8bae1dSRodney W. Grimes 2960df8bae1dSRodney W. Grimes /* 2961df8bae1dSRodney W. Grimes * In SYN_RECEIVED and ESTABLISHED STATES 2962df8bae1dSRodney W. Grimes * enter the CLOSE_WAIT state. 2963df8bae1dSRodney W. Grimes */ 2964df8bae1dSRodney W. Grimes case TCPS_SYN_RECEIVED: 29659b8b58e0SJonathan Lemon tp->t_starttime = ticks; 29669b8b58e0SJonathan Lemon /* FALLTHROUGH */ 2967df8bae1dSRodney W. Grimes case TCPS_ESTABLISHED: 296857f60867SMark Johnston tcp_state_change(tp, TCPS_CLOSE_WAIT); 2969df8bae1dSRodney W. Grimes break; 2970df8bae1dSRodney W. Grimes 2971df8bae1dSRodney W. Grimes /* 2972df8bae1dSRodney W. Grimes * If still in FIN_WAIT_1 STATE FIN has not been acked so 2973df8bae1dSRodney W. Grimes * enter the CLOSING state. 2974df8bae1dSRodney W. Grimes */ 2975df8bae1dSRodney W. Grimes case TCPS_FIN_WAIT_1: 297657f60867SMark Johnston tcp_state_change(tp, TCPS_CLOSING); 2977df8bae1dSRodney W. Grimes break; 2978df8bae1dSRodney W. Grimes 2979df8bae1dSRodney W. Grimes /* 2980df8bae1dSRodney W. Grimes * In FIN_WAIT_2 state enter the TIME_WAIT state, 2981df8bae1dSRodney W. Grimes * starting the time-wait timer, turning off the other 2982df8bae1dSRodney W. Grimes * standard timers. 2983df8bae1dSRodney W. Grimes */ 2984df8bae1dSRodney W. Grimes case TCPS_FIN_WAIT_2: 2985252ca428SRobert Watson INP_INFO_WLOCK_ASSERT(&V_tcbinfo); 2986252ca428SRobert Watson KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata " 2987252ca428SRobert Watson "TCP_FIN_WAIT_2 ti_locked: %d", __func__, 2988252ca428SRobert Watson ti_locked)); 2989252ca428SRobert Watson 2990340c35deSJonathan Lemon tcp_twstart(tp); 2991603724d3SBjoern A. Zeeb INP_INFO_WUNLOCK(&V_tcbinfo); 2992679d9708SAndre Oppermann return; 2993df8bae1dSRodney W. Grimes } 2994df8bae1dSRodney W. Grimes } 2995fa046d87SRobert Watson if (ti_locked == TI_WLOCKED) 2996603724d3SBjoern A. Zeeb INP_INFO_WUNLOCK(&V_tcbinfo); 2997252ca428SRobert Watson ti_locked = TI_UNLOCKED; 2998252ca428SRobert Watson 2999610ee2f9SDavid Greenman #ifdef TCPDEBUG 30004cc20ab1SSeigo Tanimura if (so->so_options & SO_DEBUG) 3001fb59c426SYoshinobu Inoue tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen, 3002fb59c426SYoshinobu Inoue &tcp_savetcp, 0); 3003610ee2f9SDavid Greenman #endif 3004df8bae1dSRodney W. Grimes 3005df8bae1dSRodney W. Grimes /* 3006df8bae1dSRodney W. Grimes * Return any desired output. 3007df8bae1dSRodney W. Grimes */ 3008df8bae1dSRodney W. Grimes if (needoutput || (tp->t_flags & TF_ACKNOW)) 3009df8bae1dSRodney W. Grimes (void) tcp_output(tp); 30107792ea27SJeffrey Hsu 3011a14c749fSJonathan Lemon check_delack: 3012252ca428SRobert Watson KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d", 3013252ca428SRobert Watson __func__, ti_locked)); 3014603724d3SBjoern A. Zeeb INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 30158501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 3016252ca428SRobert Watson 3017a14c749fSJonathan Lemon if (tp->t_flags & TF_DELACK) { 30183bfd6421SJonathan Lemon tp->t_flags &= ~TF_DELACK; 3019b8152ba7SAndre Oppermann tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); 30203bfd6421SJonathan Lemon } 30218501a69cSRobert Watson INP_WUNLOCK(tp->t_inpcb); 3022679d9708SAndre Oppermann return; 3023df8bae1dSRodney W. Grimes 3024df8bae1dSRodney W. Grimes dropafterack: 3025df8bae1dSRodney W. Grimes /* 3026df8bae1dSRodney W. Grimes * Generate an ACK dropping incoming segment if it occupies 3027df8bae1dSRodney W. Grimes * sequence space, where the ACK reflects our state. 302880ab7c0eSGarrett Wollman * 302980ab7c0eSGarrett Wollman * We can now skip the test for the RST flag since all 303080ab7c0eSGarrett Wollman * paths to this code happen after packets containing 303180ab7c0eSGarrett Wollman * RST have been dropped. 303280ab7c0eSGarrett Wollman * 303380ab7c0eSGarrett Wollman * In the SYN-RECEIVED state, don't send an ACK unless the 303480ab7c0eSGarrett Wollman * segment we received passes the SYN-RECEIVED ACK test. 303580ab7c0eSGarrett Wollman * If it fails send a RST. This breaks the loop in the 303680ab7c0eSGarrett Wollman * "LAND" DoS attack, and also prevents an ACK storm 303780ab7c0eSGarrett Wollman * between two listening ports that have been sent forged 303880ab7c0eSGarrett Wollman * SYN segments, each with the source address of the other. 3039df8bae1dSRodney W. Grimes */ 3040fb59c426SYoshinobu Inoue if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && 3041fb59c426SYoshinobu Inoue (SEQ_GT(tp->snd_una, th->th_ack) || 3042a57815efSBosko Milekic SEQ_GT(th->th_ack, tp->snd_max)) ) { 3043a57815efSBosko Milekic rstreason = BANDLIM_RST_OPENPORT; 3044a57815efSBosko Milekic goto dropwithreset; 3045a57815efSBosko Milekic } 3046a0292f23SGarrett Wollman #ifdef TCPDEBUG 30474cc20ab1SSeigo Tanimura if (so->so_options & SO_DEBUG) 3048fb59c426SYoshinobu Inoue tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3049fb59c426SYoshinobu Inoue &tcp_savetcp, 0); 3050a0292f23SGarrett Wollman #endif 3051fa046d87SRobert Watson if (ti_locked == TI_WLOCKED) 3052603724d3SBjoern A. Zeeb INP_INFO_WUNLOCK(&V_tcbinfo); 3053252ca428SRobert Watson ti_locked = TI_UNLOCKED; 3054252ca428SRobert Watson 3055df8bae1dSRodney W. Grimes tp->t_flags |= TF_ACKNOW; 3056df8bae1dSRodney W. Grimes (void) tcp_output(tp); 30578501a69cSRobert Watson INP_WUNLOCK(tp->t_inpcb); 3058d6915262SRobert Watson m_freem(m); 3059679d9708SAndre Oppermann return; 3060df8bae1dSRodney W. Grimes 3061df8bae1dSRodney W. Grimes dropwithreset: 3062fa046d87SRobert Watson if (ti_locked == TI_WLOCKED) 3063dd8ac7f9SRobert Watson INP_INFO_WUNLOCK(&V_tcbinfo); 3064252ca428SRobert Watson ti_locked = TI_UNLOCKED; 3065a57815efSBosko Milekic 3066a0ca0871SRobert Watson if (tp != NULL) { 3067302ce8d6SAndre Oppermann tcp_dropwithreset(m, th, tp, tlen, rstreason); 30688501a69cSRobert Watson INP_WUNLOCK(tp->t_inpcb); 3069dd8ac7f9SRobert Watson } else 3070a0ca0871SRobert Watson tcp_dropwithreset(m, th, NULL, tlen, rstreason); 3071679d9708SAndre Oppermann return; 3072df8bae1dSRodney W. Grimes 3073df8bae1dSRodney W. Grimes drop: 3074fa046d87SRobert Watson if (ti_locked == TI_WLOCKED) { 3075252ca428SRobert Watson INP_INFO_WUNLOCK(&V_tcbinfo); 3076fa046d87SRobert Watson ti_locked = TI_UNLOCKED; 3077fa046d87SRobert Watson } 3078252ca428SRobert Watson #ifdef INVARIANTS 3079252ca428SRobert Watson else 3080252ca428SRobert Watson INP_INFO_UNLOCK_ASSERT(&V_tcbinfo); 3081252ca428SRobert Watson #endif 3082252ca428SRobert Watson 3083df8bae1dSRodney W. Grimes /* 3084df8bae1dSRodney W. Grimes * Drop space held by incoming segment and return. 3085df8bae1dSRodney W. Grimes */ 3086610ee2f9SDavid Greenman #ifdef TCPDEBUG 30871c53f806SRobert Watson if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG)) 3088fb59c426SYoshinobu Inoue tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen, 3089fb59c426SYoshinobu Inoue &tcp_savetcp, 0); 3090a0292f23SGarrett Wollman #endif 30911c53f806SRobert Watson if (tp != NULL) 30928501a69cSRobert Watson INP_WUNLOCK(tp->t_inpcb); 3093d6915262SRobert Watson m_freem(m); 3094302ce8d6SAndre Oppermann } 3095302ce8d6SAndre Oppermann 3096302ce8d6SAndre Oppermann /* 30970ca3f933SAndre Oppermann * Issue RST and make ACK acceptable to originator of segment. 30980ca3f933SAndre Oppermann * The mbuf must still include the original packet header. 30990ca3f933SAndre Oppermann * tp may be NULL. 3100302ce8d6SAndre Oppermann */ 3101302ce8d6SAndre Oppermann static void 3102302ce8d6SAndre Oppermann tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, 3103302ce8d6SAndre Oppermann int tlen, int rstreason) 3104302ce8d6SAndre Oppermann { 3105b287c6c7SBjoern A. Zeeb #ifdef INET 3106302ce8d6SAndre Oppermann struct ip *ip; 3107b287c6c7SBjoern A. Zeeb #endif 3108302ce8d6SAndre Oppermann #ifdef INET6 3109302ce8d6SAndre Oppermann struct ip6_hdr *ip6; 3110302ce8d6SAndre Oppermann #endif 31117a3244ccSRobert Watson 31127a3244ccSRobert Watson if (tp != NULL) { 31138501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 31147a3244ccSRobert Watson } 31157a3244ccSRobert Watson 31160ca3f933SAndre Oppermann /* Don't bother if destination was broadcast/multicast. */ 3117302ce8d6SAndre Oppermann if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) 3118302ce8d6SAndre Oppermann goto drop; 3119302ce8d6SAndre Oppermann #ifdef INET6 3120302ce8d6SAndre Oppermann if (mtod(m, struct ip *)->ip_v == 6) { 3121302ce8d6SAndre Oppermann ip6 = mtod(m, struct ip6_hdr *); 3122302ce8d6SAndre Oppermann if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || 3123302ce8d6SAndre Oppermann IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) 3124302ce8d6SAndre Oppermann goto drop; 3125302ce8d6SAndre Oppermann /* IPv6 anycast check is done at tcp6_input() */ 3126b287c6c7SBjoern A. Zeeb } 3127302ce8d6SAndre Oppermann #endif 3128b287c6c7SBjoern A. Zeeb #if defined(INET) && defined(INET6) 3129b287c6c7SBjoern A. Zeeb else 3130b287c6c7SBjoern A. Zeeb #endif 3131b287c6c7SBjoern A. Zeeb #ifdef INET 3132302ce8d6SAndre Oppermann { 3133302ce8d6SAndre Oppermann ip = mtod(m, struct ip *); 3134302ce8d6SAndre Oppermann if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || 3135302ce8d6SAndre Oppermann IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || 3136302ce8d6SAndre Oppermann ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || 3137302ce8d6SAndre Oppermann in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) 3138302ce8d6SAndre Oppermann goto drop; 3139302ce8d6SAndre Oppermann } 3140b287c6c7SBjoern A. Zeeb #endif 3141302ce8d6SAndre Oppermann 3142302ce8d6SAndre Oppermann /* Perform bandwidth limiting. */ 3143302ce8d6SAndre Oppermann if (badport_bandlim(rstreason) < 0) 3144302ce8d6SAndre Oppermann goto drop; 3145302ce8d6SAndre Oppermann 3146302ce8d6SAndre Oppermann /* tcp_respond consumes the mbuf chain. */ 3147302ce8d6SAndre Oppermann if (th->th_flags & TH_ACK) { 3148302ce8d6SAndre Oppermann tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, 3149302ce8d6SAndre Oppermann th->th_ack, TH_RST); 3150302ce8d6SAndre Oppermann } else { 3151302ce8d6SAndre Oppermann if (th->th_flags & TH_SYN) 3152302ce8d6SAndre Oppermann tlen++; 3153302ce8d6SAndre Oppermann tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, 3154302ce8d6SAndre Oppermann (tcp_seq)0, TH_RST|TH_ACK); 3155302ce8d6SAndre Oppermann } 3156302ce8d6SAndre Oppermann return; 3157302ce8d6SAndre Oppermann drop: 3158302ce8d6SAndre Oppermann m_freem(m); 3159df8bae1dSRodney W. Grimes } 3160df8bae1dSRodney W. Grimes 3161be2ac88cSJonathan Lemon /* 3162be2ac88cSJonathan Lemon * Parse TCP options and place in tcpopt. 3163be2ac88cSJonathan Lemon */ 31640312fbe9SPoul-Henning Kamp static void 3165ad3f9ab3SAndre Oppermann tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) 3166df8bae1dSRodney W. Grimes { 3167df8bae1dSRodney W. Grimes int opt, optlen; 3168df8bae1dSRodney W. Grimes 3169be2ac88cSJonathan Lemon to->to_flags = 0; 3170df8bae1dSRodney W. Grimes for (; cnt > 0; cnt -= optlen, cp += optlen) { 3171df8bae1dSRodney W. Grimes opt = cp[0]; 3172df8bae1dSRodney W. Grimes if (opt == TCPOPT_EOL) 3173df8bae1dSRodney W. Grimes break; 3174df8bae1dSRodney W. Grimes if (opt == TCPOPT_NOP) 3175df8bae1dSRodney W. Grimes optlen = 1; 3176df8bae1dSRodney W. Grimes else { 3177b474779fSJun-ichiro itojun Hagino if (cnt < 2) 3178b474779fSJun-ichiro itojun Hagino break; 3179df8bae1dSRodney W. Grimes optlen = cp[1]; 3180b474779fSJun-ichiro itojun Hagino if (optlen < 2 || optlen > cnt) 3181df8bae1dSRodney W. Grimes break; 3182df8bae1dSRodney W. Grimes } 3183df8bae1dSRodney W. Grimes switch (opt) { 3184df8bae1dSRodney W. Grimes case TCPOPT_MAXSEG: 3185df8bae1dSRodney W. Grimes if (optlen != TCPOLEN_MAXSEG) 3186df8bae1dSRodney W. Grimes continue; 3187f72167f4SAndre Oppermann if (!(flags & TO_SYN)) 3188df8bae1dSRodney W. Grimes continue; 3189be2ac88cSJonathan Lemon to->to_flags |= TOF_MSS; 3190be2ac88cSJonathan Lemon bcopy((char *)cp + 2, 3191be2ac88cSJonathan Lemon (char *)&to->to_mss, sizeof(to->to_mss)); 3192fd8e4ebcSMike Barcroft to->to_mss = ntohs(to->to_mss); 3193df8bae1dSRodney W. Grimes break; 3194df8bae1dSRodney W. Grimes case TCPOPT_WINDOW: 3195df8bae1dSRodney W. Grimes if (optlen != TCPOLEN_WINDOW) 3196df8bae1dSRodney W. Grimes continue; 3197f72167f4SAndre Oppermann if (!(flags & TO_SYN)) 3198df8bae1dSRodney W. Grimes continue; 3199be2ac88cSJonathan Lemon to->to_flags |= TOF_SCALE; 320002a1a643SAndre Oppermann to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); 3201df8bae1dSRodney W. Grimes break; 3202df8bae1dSRodney W. Grimes case TCPOPT_TIMESTAMP: 3203df8bae1dSRodney W. Grimes if (optlen != TCPOLEN_TIMESTAMP) 3204df8bae1dSRodney W. Grimes continue; 3205be2ac88cSJonathan Lemon to->to_flags |= TOF_TS; 3206a0292f23SGarrett Wollman bcopy((char *)cp + 2, 3207a0292f23SGarrett Wollman (char *)&to->to_tsval, sizeof(to->to_tsval)); 3208fd8e4ebcSMike Barcroft to->to_tsval = ntohl(to->to_tsval); 3209a0292f23SGarrett Wollman bcopy((char *)cp + 6, 3210a0292f23SGarrett Wollman (char *)&to->to_tsecr, sizeof(to->to_tsecr)); 3211fd8e4ebcSMike Barcroft to->to_tsecr = ntohl(to->to_tsecr); 3212df8bae1dSRodney W. Grimes break; 32131cfd4b53SBruce M Simpson #ifdef TCP_SIGNATURE 32141cfd4b53SBruce M Simpson /* 32151cfd4b53SBruce M Simpson * XXX In order to reply to a host which has set the 32161cfd4b53SBruce M Simpson * TCP_SIGNATURE option in its initial SYN, we have to 32171cfd4b53SBruce M Simpson * record the fact that the option was observed here 32181cfd4b53SBruce M Simpson * for the syncache code to perform the correct response. 32191cfd4b53SBruce M Simpson */ 32201cfd4b53SBruce M Simpson case TCPOPT_SIGNATURE: 32211cfd4b53SBruce M Simpson if (optlen != TCPOLEN_SIGNATURE) 32221cfd4b53SBruce M Simpson continue; 3223df47e437SAndre Oppermann to->to_flags |= TOF_SIGNATURE; 3224df47e437SAndre Oppermann to->to_signature = cp + 2; 32251cfd4b53SBruce M Simpson break; 3226265ed012SBruce M Simpson #endif 32276d90faf3SPaul Saab case TCPOPT_SACK_PERMITTED: 3228f72167f4SAndre Oppermann if (optlen != TCPOLEN_SACK_PERMITTED) 32296d90faf3SPaul Saab continue; 3230f72167f4SAndre Oppermann if (!(flags & TO_SYN)) 3231f72167f4SAndre Oppermann continue; 3232603724d3SBjoern A. Zeeb if (!V_tcp_do_sack) 3233f72167f4SAndre Oppermann continue; 3234fc30a251SAndre Oppermann to->to_flags |= TOF_SACKPERM; 32356d90faf3SPaul Saab break; 32366d90faf3SPaul Saab case TCPOPT_SACK: 32375a53ca16SPaul Saab if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) 32386d90faf3SPaul Saab continue; 3239b728e902SAndre Oppermann if (flags & TO_SYN) 3240b728e902SAndre Oppermann continue; 3241fc30a251SAndre Oppermann to->to_flags |= TOF_SACK; 32425a53ca16SPaul Saab to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; 32435a53ca16SPaul Saab to->to_sacks = cp + 2; 324478b50714SRobert Watson TCPSTAT_INC(tcps_sack_rcv_blocks); 32456d90faf3SPaul Saab break; 3246be2ac88cSJonathan Lemon default: 3247be2ac88cSJonathan Lemon continue; 3248df8bae1dSRodney W. Grimes } 3249df8bae1dSRodney W. Grimes } 3250df8bae1dSRodney W. Grimes } 3251df8bae1dSRodney W. Grimes 3252df8bae1dSRodney W. Grimes /* 3253df8bae1dSRodney W. Grimes * Pull out of band byte out of a segment so 3254df8bae1dSRodney W. Grimes * it doesn't appear in the user's data queue. 3255df8bae1dSRodney W. Grimes * It is still reflected in the segment length for 3256df8bae1dSRodney W. Grimes * sequencing purposes. 3257df8bae1dSRodney W. Grimes */ 32580312fbe9SPoul-Henning Kamp static void 3259ad3f9ab3SAndre Oppermann tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, 3260ad3f9ab3SAndre Oppermann int off) 3261df8bae1dSRodney W. Grimes { 3262fb59c426SYoshinobu Inoue int cnt = off + th->th_urp - 1; 3263df8bae1dSRodney W. Grimes 3264df8bae1dSRodney W. Grimes while (cnt >= 0) { 3265df8bae1dSRodney W. Grimes if (m->m_len > cnt) { 3266df8bae1dSRodney W. Grimes char *cp = mtod(m, caddr_t) + cnt; 3267df8bae1dSRodney W. Grimes struct tcpcb *tp = sototcpcb(so); 3268df8bae1dSRodney W. Grimes 32698501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 32707a3244ccSRobert Watson 3271df8bae1dSRodney W. Grimes tp->t_iobc = *cp; 3272df8bae1dSRodney W. Grimes tp->t_oobflags |= TCPOOB_HAVEDATA; 3273df8bae1dSRodney W. Grimes bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); 3274df8bae1dSRodney W. Grimes m->m_len--; 327569a34685SYoshinobu Inoue if (m->m_flags & M_PKTHDR) 327669a34685SYoshinobu Inoue m->m_pkthdr.len--; 3277df8bae1dSRodney W. Grimes return; 3278df8bae1dSRodney W. Grimes } 3279df8bae1dSRodney W. Grimes cnt -= m->m_len; 3280df8bae1dSRodney W. Grimes m = m->m_next; 32814dfdffe9SAndre Oppermann if (m == NULL) 3282df8bae1dSRodney W. Grimes break; 3283df8bae1dSRodney W. Grimes } 3284df8bae1dSRodney W. Grimes panic("tcp_pulloutofband"); 3285df8bae1dSRodney W. Grimes } 3286df8bae1dSRodney W. Grimes 3287df8bae1dSRodney W. Grimes /* 3288df8bae1dSRodney W. Grimes * Collect new round-trip time estimate 3289df8bae1dSRodney W. Grimes * and update averages and current timeout. 3290df8bae1dSRodney W. Grimes */ 32910312fbe9SPoul-Henning Kamp static void 3292ad3f9ab3SAndre Oppermann tcp_xmit_timer(struct tcpcb *tp, int rtt) 3293df8bae1dSRodney W. Grimes { 3294ad3f9ab3SAndre Oppermann int delta; 3295233e8c18SGarrett Wollman 32968501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 32972be3bf22SRobert Watson 329878b50714SRobert Watson TCPSTAT_INC(tcps_rttupdated); 3299233e8c18SGarrett Wollman tp->t_rttupdated++; 3300233e8c18SGarrett Wollman if (tp->t_srtt != 0) { 3301233e8c18SGarrett Wollman /* 3302233e8c18SGarrett Wollman * srtt is stored as fixed point with 5 bits after the 3303233e8c18SGarrett Wollman * binary point (i.e., scaled by 8). The following magic 3304233e8c18SGarrett Wollman * is equivalent to the smoothing algorithm in rfc793 with 3305233e8c18SGarrett Wollman * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed 3306233e8c18SGarrett Wollman * point). Adjust rtt to origin 0. 3307233e8c18SGarrett Wollman */ 3308233e8c18SGarrett Wollman delta = ((rtt - 1) << TCP_DELTA_SHIFT) 3309233e8c18SGarrett Wollman - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); 3310233e8c18SGarrett Wollman 3311233e8c18SGarrett Wollman if ((tp->t_srtt += delta) <= 0) 3312233e8c18SGarrett Wollman tp->t_srtt = 1; 3313233e8c18SGarrett Wollman 3314233e8c18SGarrett Wollman /* 3315233e8c18SGarrett Wollman * We accumulate a smoothed rtt variance (actually, a 3316233e8c18SGarrett Wollman * smoothed mean difference), then set the retransmit 3317233e8c18SGarrett Wollman * timer to smoothed rtt + 4 times the smoothed variance. 3318233e8c18SGarrett Wollman * rttvar is stored as fixed point with 4 bits after the 3319233e8c18SGarrett Wollman * binary point (scaled by 16). The following is 3320233e8c18SGarrett Wollman * equivalent to rfc793 smoothing with an alpha of .75 3321233e8c18SGarrett Wollman * (rttvar = rttvar*3/4 + |delta| / 4). This replaces 3322233e8c18SGarrett Wollman * rfc793's wired-in beta. 3323233e8c18SGarrett Wollman */ 3324233e8c18SGarrett Wollman if (delta < 0) 3325233e8c18SGarrett Wollman delta = -delta; 3326233e8c18SGarrett Wollman delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); 3327233e8c18SGarrett Wollman if ((tp->t_rttvar += delta) <= 0) 3328233e8c18SGarrett Wollman tp->t_rttvar = 1; 33291fcc99b5SMatthew Dillon if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar) 33301fcc99b5SMatthew Dillon tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3331233e8c18SGarrett Wollman } else { 3332233e8c18SGarrett Wollman /* 3333233e8c18SGarrett Wollman * No rtt measurement yet - use the unsmoothed rtt. 3334233e8c18SGarrett Wollman * Set the variance to half the rtt (so our first 3335233e8c18SGarrett Wollman * retransmit happens at 3*rtt). 3336233e8c18SGarrett Wollman */ 3337233e8c18SGarrett Wollman tp->t_srtt = rtt << TCP_RTT_SHIFT; 3338233e8c18SGarrett Wollman tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); 33391fcc99b5SMatthew Dillon tp->t_rttbest = tp->t_srtt + tp->t_rttvar; 3340233e8c18SGarrett Wollman } 33419b8b58e0SJonathan Lemon tp->t_rtttime = 0; 3342df8bae1dSRodney W. Grimes tp->t_rxtshift = 0; 3343df8bae1dSRodney W. Grimes 3344df8bae1dSRodney W. Grimes /* 3345df8bae1dSRodney W. Grimes * the retransmit should happen at rtt + 4 * rttvar. 3346df8bae1dSRodney W. Grimes * Because of the way we do the smoothing, srtt and rttvar 3347df8bae1dSRodney W. Grimes * will each average +1/2 tick of bias. When we compute 3348df8bae1dSRodney W. Grimes * the retransmit timer, we want 1/2 tick of rounding and 3349df8bae1dSRodney W. Grimes * 1 extra tick because of +-1/2 tick uncertainty in the 3350df8bae1dSRodney W. Grimes * firing of the timer. The bias will give us exactly the 3351df8bae1dSRodney W. Grimes * 1.5 tick we need. But, because the bias is 3352df8bae1dSRodney W. Grimes * statistical, we have to test that we don't drop below 3353df8bae1dSRodney W. Grimes * the minimum feasible timer (which is 2 ticks). 3354df8bae1dSRodney W. Grimes */ 3355233e8c18SGarrett Wollman TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), 33569e2874b0SGarrett Wollman max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); 3357df8bae1dSRodney W. Grimes 3358df8bae1dSRodney W. Grimes /* 3359df8bae1dSRodney W. Grimes * We received an ack for a packet that wasn't retransmitted; 3360df8bae1dSRodney W. Grimes * it is probably safe to discard any error indications we've 3361df8bae1dSRodney W. Grimes * received recently. This isn't quite right, but close enough 3362df8bae1dSRodney W. Grimes * for now (a route might have failed after we sent a segment, 3363df8bae1dSRodney W. Grimes * and the return path might not be symmetrical). 3364df8bae1dSRodney W. Grimes */ 3365df8bae1dSRodney W. Grimes tp->t_softerror = 0; 3366df8bae1dSRodney W. Grimes } 3367df8bae1dSRodney W. Grimes 3368df8bae1dSRodney W. Grimes /* 3369df8bae1dSRodney W. Grimes * Determine a reasonable value for maxseg size. 3370df8bae1dSRodney W. Grimes * If the route is known, check route for mtu. 33714faaea55SAndre Oppermann * If none, use an mss that can be handled on the outgoing interface 33724faaea55SAndre Oppermann * without forcing IP to fragment. If no route is found, route has no mtu, 3373df8bae1dSRodney W. Grimes * or the destination isn't local, use a default, hopefully conservative 3374df8bae1dSRodney W. Grimes * size (usually 512 or the default IP max size, but no more than the mtu 3375df8bae1dSRodney W. Grimes * of the interface), as we can't discover anything about intervening 3376df8bae1dSRodney W. Grimes * gateways or networks. We also initialize the congestion/slow start 3377df8bae1dSRodney W. Grimes * window to be a single segment if the destination isn't local. 3378df8bae1dSRodney W. Grimes * While looking at the routing entry, we also initialize other path-dependent 3379df8bae1dSRodney W. Grimes * parameters from pre-set or cached values in the routing entry. 3380a0292f23SGarrett Wollman * 3381a0292f23SGarrett Wollman * Also take into account the space needed for options that we 3382a0292f23SGarrett Wollman * send regularly. Make maxseg shorter by that amount to assure 3383a0292f23SGarrett Wollman * that we can send maxseg amount of data even when the options 3384a0292f23SGarrett Wollman * are present. Store the upper limit of the length of options plus 3385a0292f23SGarrett Wollman * data in maxopd. 3386a0292f23SGarrett Wollman * 338797d8d152SAndre Oppermann * NOTE that this routine is only called when we process an incoming 3388ef341ee1SGleb Smirnoff * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS 3389ef341ee1SGleb Smirnoff * settings are handled in tcp_mssopt(). 3390df8bae1dSRodney W. Grimes */ 3391a0292f23SGarrett Wollman void 3392ef341ee1SGleb Smirnoff tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, 33933c914c54SAndre Oppermann struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) 3394df8bae1dSRodney W. Grimes { 3395b287c6c7SBjoern A. Zeeb int mss = 0; 3396b287c6c7SBjoern A. Zeeb u_long maxmtu = 0; 3397c068736aSJeffrey Hsu struct inpcb *inp = tp->t_inpcb; 339897d8d152SAndre Oppermann struct hc_metrics_lite metrics; 3399ef341ee1SGleb Smirnoff int origoffer; 3400fb59c426SYoshinobu Inoue #ifdef INET6 3401c068736aSJeffrey Hsu int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; 3402c068736aSJeffrey Hsu size_t min_protoh = isipv6 ? 3403c068736aSJeffrey Hsu sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : 3404c068736aSJeffrey Hsu sizeof (struct tcpiphdr); 3405c068736aSJeffrey Hsu #else 3406c068736aSJeffrey Hsu const size_t min_protoh = sizeof(struct tcpiphdr); 3407fb59c426SYoshinobu Inoue #endif 3408df8bae1dSRodney W. Grimes 34098501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 34107a3244ccSRobert Watson 3411ef341ee1SGleb Smirnoff if (mtuoffer != -1) { 3412ef341ee1SGleb Smirnoff KASSERT(offer == -1, ("%s: conflict", __func__)); 3413ef341ee1SGleb Smirnoff offer = mtuoffer - min_protoh; 3414ef341ee1SGleb Smirnoff } 3415ef341ee1SGleb Smirnoff origoffer = offer; 3416ef341ee1SGleb Smirnoff 3417e8949f74SAndre Oppermann /* Initialize. */ 341897d8d152SAndre Oppermann #ifdef INET6 341997d8d152SAndre Oppermann if (isipv6) { 34203c914c54SAndre Oppermann maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); 3421603724d3SBjoern A. Zeeb tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt; 3422b287c6c7SBjoern A. Zeeb } 342397d8d152SAndre Oppermann #endif 3424b287c6c7SBjoern A. Zeeb #if defined(INET) && defined(INET6) 3425b287c6c7SBjoern A. Zeeb else 3426b287c6c7SBjoern A. Zeeb #endif 3427b287c6c7SBjoern A. Zeeb #ifdef INET 342897d8d152SAndre Oppermann { 34293c914c54SAndre Oppermann maxmtu = tcp_maxmtu(&inp->inp_inc, cap); 3430603724d3SBjoern A. Zeeb tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt; 3431df8bae1dSRodney W. Grimes } 3432b287c6c7SBjoern A. Zeeb #endif 3433df8bae1dSRodney W. Grimes 343497d8d152SAndre Oppermann /* 3435e8949f74SAndre Oppermann * No route to sender, stay with default mss and return. 343697d8d152SAndre Oppermann */ 34376f01cac6SBjoern A. Zeeb if (maxmtu == 0) { 34386f01cac6SBjoern A. Zeeb /* 34398e5c87f4SBjoern A. Zeeb * In case we return early we need to initialize metrics 34406f01cac6SBjoern A. Zeeb * to a defined state as tcp_hc_get() would do for us 34416f01cac6SBjoern A. Zeeb * if there was no cache hit. 34426f01cac6SBjoern A. Zeeb */ 34436f01cac6SBjoern A. Zeeb if (metricptr != NULL) 34446f01cac6SBjoern A. Zeeb bzero(metricptr, sizeof(struct hc_metrics_lite)); 344597d8d152SAndre Oppermann return; 34466f01cac6SBjoern A. Zeeb } 344797d8d152SAndre Oppermann 3448e8949f74SAndre Oppermann /* What have we got? */ 344997d8d152SAndre Oppermann switch (offer) { 345097d8d152SAndre Oppermann case 0: 345197d8d152SAndre Oppermann /* 345297d8d152SAndre Oppermann * Offer == 0 means that there was no MSS on the SYN 3453c3b02504SBjoern A. Zeeb * segment, in this case we use tcp_mssdflt as 3454c3b02504SBjoern A. Zeeb * already assigned to t_maxopd above. 345597d8d152SAndre Oppermann */ 3456c3b02504SBjoern A. Zeeb offer = tp->t_maxopd; 345797d8d152SAndre Oppermann break; 345897d8d152SAndre Oppermann 345997d8d152SAndre Oppermann case -1: 3460a0292f23SGarrett Wollman /* 3461c94c54e4SAndre Oppermann * Offer == -1 means that we didn't receive SYN yet. 3462a0292f23SGarrett Wollman */ 346397d8d152SAndre Oppermann /* FALLTHROUGH */ 346497d8d152SAndre Oppermann 346597d8d152SAndre Oppermann default: 3466a0292f23SGarrett Wollman /* 346753369ac9SAndre Oppermann * Prevent DoS attack with too small MSS. Round up 346853369ac9SAndre Oppermann * to at least minmss. 346953369ac9SAndre Oppermann */ 3470603724d3SBjoern A. Zeeb offer = max(offer, V_tcp_minmss); 347197d8d152SAndre Oppermann } 3472a0292f23SGarrett Wollman 3473df8bae1dSRodney W. Grimes /* 3474e8949f74SAndre Oppermann * rmx information is now retrieved from tcp_hostcache. 3475df8bae1dSRodney W. Grimes */ 347697d8d152SAndre Oppermann tcp_hc_get(&inp->inp_inc, &metrics); 34773cee92e0SBjoern A. Zeeb if (metricptr != NULL) 34783cee92e0SBjoern A. Zeeb bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); 347997d8d152SAndre Oppermann 3480df8bae1dSRodney W. Grimes /* 3481cc412412SHiren Panchasara * If there's a discovered mtu in tcp hostcache, use it. 3482cc412412SHiren Panchasara * Else, use the link mtu. 3483df8bae1dSRodney W. Grimes */ 348497d8d152SAndre Oppermann if (metrics.rmx_mtu) 34852d166c02SAndre Oppermann mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; 3486c068736aSJeffrey Hsu else { 348731b3783cSHajimu UMEMOTO #ifdef INET6 3488fb59c426SYoshinobu Inoue if (isipv6) { 348997d8d152SAndre Oppermann mss = maxmtu - min_protoh; 3490603724d3SBjoern A. Zeeb if (!V_path_mtu_discovery && 349197d8d152SAndre Oppermann !in6_localaddr(&inp->in6p_faddr)) 3492603724d3SBjoern A. Zeeb mss = min(mss, V_tcp_v6mssdflt); 3493b287c6c7SBjoern A. Zeeb } 3494b3399803SHajimu UMEMOTO #endif 3495b287c6c7SBjoern A. Zeeb #if defined(INET) && defined(INET6) 3496b287c6c7SBjoern A. Zeeb else 3497b287c6c7SBjoern A. Zeeb #endif 3498b287c6c7SBjoern A. Zeeb #ifdef INET 349997d8d152SAndre Oppermann { 350097d8d152SAndre Oppermann mss = maxmtu - min_protoh; 3501603724d3SBjoern A. Zeeb if (!V_path_mtu_discovery && 350297d8d152SAndre Oppermann !in_localaddr(inp->inp_faddr)) 3503603724d3SBjoern A. Zeeb mss = min(mss, V_tcp_mssdflt); 3504a0292f23SGarrett Wollman } 3505b287c6c7SBjoern A. Zeeb #endif 35063cee92e0SBjoern A. Zeeb /* 35073cee92e0SBjoern A. Zeeb * XXX - The above conditional (mss = maxmtu - min_protoh) 35083cee92e0SBjoern A. Zeeb * probably violates the TCP spec. 35093cee92e0SBjoern A. Zeeb * The problem is that, since we don't know the 35103cee92e0SBjoern A. Zeeb * other end's MSS, we are supposed to use a conservative 35113cee92e0SBjoern A. Zeeb * default. But, if we do that, then MTU discovery will 35123cee92e0SBjoern A. Zeeb * never actually take place, because the conservative 35133cee92e0SBjoern A. Zeeb * default is much less than the MTUs typically seen 35143cee92e0SBjoern A. Zeeb * on the Internet today. For the moment, we'll sweep 35153cee92e0SBjoern A. Zeeb * this under the carpet. 35163cee92e0SBjoern A. Zeeb * 35173cee92e0SBjoern A. Zeeb * The conservative default might not actually be a problem 35183cee92e0SBjoern A. Zeeb * if the only case this occurs is when sending an initial 35193cee92e0SBjoern A. Zeeb * SYN with options and data to a host we've never talked 35203cee92e0SBjoern A. Zeeb * to before. Then, they will reply with an MSS value which 35213cee92e0SBjoern A. Zeeb * will get recorded and the new parameters should get 35223cee92e0SBjoern A. Zeeb * recomputed. For Further Study. 35233cee92e0SBjoern A. Zeeb */ 352497d8d152SAndre Oppermann } 3525a0292f23SGarrett Wollman mss = min(mss, offer); 352697d8d152SAndre Oppermann 3527a0292f23SGarrett Wollman /* 35283cee92e0SBjoern A. Zeeb * Sanity check: make sure that maxopd will be large 35293cee92e0SBjoern A. Zeeb * enough to allow some data on segments even if the 35303cee92e0SBjoern A. Zeeb * all the option space is used (40bytes). Otherwise 35313cee92e0SBjoern A. Zeeb * funny things may happen in tcp_output. 35323cee92e0SBjoern A. Zeeb */ 35333cee92e0SBjoern A. Zeeb mss = max(mss, 64); 35343cee92e0SBjoern A. Zeeb 35353cee92e0SBjoern A. Zeeb /* 3536a0292f23SGarrett Wollman * maxopd stores the maximum length of data AND options 3537a0292f23SGarrett Wollman * in a segment; maxseg is the amount of data in a normal 3538a0292f23SGarrett Wollman * segment. We need to store this value (maxopd) apart 3539a0292f23SGarrett Wollman * from maxseg, because now every segment carries options 3540a0292f23SGarrett Wollman * and thus we normally have somewhat less data in segments. 3541a0292f23SGarrett Wollman */ 3542a0292f23SGarrett Wollman tp->t_maxopd = mss; 3543a0292f23SGarrett Wollman 3544a0292f23SGarrett Wollman /* 3545e8949f74SAndre Oppermann * origoffer==-1 indicates that no segments were received yet. 3546c94c54e4SAndre Oppermann * In this case we just guess. 3547a0292f23SGarrett Wollman */ 3548a0292f23SGarrett Wollman if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP && 3549a0292f23SGarrett Wollman (origoffer == -1 || 3550a0292f23SGarrett Wollman (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)) 3551a0292f23SGarrett Wollman mss -= TCPOLEN_TSTAMP_APPA; 3552a0292f23SGarrett Wollman 355397d8d152SAndre Oppermann tp->t_maxseg = mss; 35543cee92e0SBjoern A. Zeeb } 35553cee92e0SBjoern A. Zeeb 35563cee92e0SBjoern A. Zeeb void 35573cee92e0SBjoern A. Zeeb tcp_mss(struct tcpcb *tp, int offer) 35583cee92e0SBjoern A. Zeeb { 3559dbc42409SLawrence Stewart int mss; 35603cee92e0SBjoern A. Zeeb u_long bufsize; 35613cee92e0SBjoern A. Zeeb struct inpcb *inp; 35623cee92e0SBjoern A. Zeeb struct socket *so; 35633cee92e0SBjoern A. Zeeb struct hc_metrics_lite metrics; 35643c914c54SAndre Oppermann struct tcp_ifcap cap; 3565dbc42409SLawrence Stewart 35663cee92e0SBjoern A. Zeeb KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); 35673cee92e0SBjoern A. Zeeb 35683c914c54SAndre Oppermann bzero(&cap, sizeof(cap)); 35693c914c54SAndre Oppermann tcp_mss_update(tp, offer, -1, &metrics, &cap); 35703cee92e0SBjoern A. Zeeb 35713cee92e0SBjoern A. Zeeb mss = tp->t_maxseg; 35723cee92e0SBjoern A. Zeeb inp = tp->t_inpcb; 357397d8d152SAndre Oppermann 3574df8bae1dSRodney W. Grimes /* 357597d8d152SAndre Oppermann * If there's a pipesize, change the socket buffer to that size, 357697d8d152SAndre Oppermann * don't change if sb_hiwat is different than default (then it 357797d8d152SAndre Oppermann * has been changed on purpose with setsockopt). 357897d8d152SAndre Oppermann * Make the socket buffers an integral number of mss units; 357997d8d152SAndre Oppermann * if the mss is larger than the socket buffer, decrease the mss. 3580df8bae1dSRodney W. Grimes */ 3581c3b02504SBjoern A. Zeeb so = inp->inp_socket; 35823f11a2f3SRobert Watson SOCKBUF_LOCK(&so->so_snd); 3583e233e2acSAndre Oppermann if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) 358497d8d152SAndre Oppermann bufsize = metrics.rmx_sendpipe; 358597d8d152SAndre Oppermann else 3586df8bae1dSRodney W. Grimes bufsize = so->so_snd.sb_hiwat; 3587df8bae1dSRodney W. Grimes if (bufsize < mss) 3588df8bae1dSRodney W. Grimes mss = bufsize; 3589df8bae1dSRodney W. Grimes else { 3590df8bae1dSRodney W. Grimes bufsize = roundup(bufsize, mss); 3591df8bae1dSRodney W. Grimes if (bufsize > sb_max) 3592df8bae1dSRodney W. Grimes bufsize = sb_max; 359388c39af3SRuslan Ermilov if (bufsize > so->so_snd.sb_hiwat) 35943f11a2f3SRobert Watson (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL); 3595df8bae1dSRodney W. Grimes } 35963f11a2f3SRobert Watson SOCKBUF_UNLOCK(&so->so_snd); 3597df8bae1dSRodney W. Grimes tp->t_maxseg = mss; 3598df8bae1dSRodney W. Grimes 35993f11a2f3SRobert Watson SOCKBUF_LOCK(&so->so_rcv); 3600e233e2acSAndre Oppermann if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) 360197d8d152SAndre Oppermann bufsize = metrics.rmx_recvpipe; 360297d8d152SAndre Oppermann else 3603df8bae1dSRodney W. Grimes bufsize = so->so_rcv.sb_hiwat; 3604df8bae1dSRodney W. Grimes if (bufsize > mss) { 3605df8bae1dSRodney W. Grimes bufsize = roundup(bufsize, mss); 3606df8bae1dSRodney W. Grimes if (bufsize > sb_max) 3607df8bae1dSRodney W. Grimes bufsize = sb_max; 360888c39af3SRuslan Ermilov if (bufsize > so->so_rcv.sb_hiwat) 36093f11a2f3SRobert Watson (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL); 3610df8bae1dSRodney W. Grimes } 36113f11a2f3SRobert Watson SOCKBUF_UNLOCK(&so->so_rcv); 361291d6cfa6SBjoern A. Zeeb 361391d6cfa6SBjoern A. Zeeb /* Check the interface for TSO capabilities. */ 36143c914c54SAndre Oppermann if (cap.ifcap & CSUM_TSO) { 361591d6cfa6SBjoern A. Zeeb tp->t_flags |= TF_TSO; 36163c914c54SAndre Oppermann tp->t_tsomax = cap.tsomax; 36173c914c54SAndre Oppermann } 3618a0292f23SGarrett Wollman } 3619a0292f23SGarrett Wollman 3620a0292f23SGarrett Wollman /* 3621a0292f23SGarrett Wollman * Determine the MSS option to send on an outgoing SYN. 3622a0292f23SGarrett Wollman */ 3623a0292f23SGarrett Wollman int 3624ad3f9ab3SAndre Oppermann tcp_mssopt(struct in_conninfo *inc) 3625a0292f23SGarrett Wollman { 362697d8d152SAndre Oppermann int mss = 0; 362797d8d152SAndre Oppermann u_long maxmtu = 0; 362897d8d152SAndre Oppermann u_long thcmtu = 0; 362997d8d152SAndre Oppermann size_t min_protoh; 3630a0292f23SGarrett Wollman 363197d8d152SAndre Oppermann KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); 3632a0292f23SGarrett Wollman 363331b3783cSHajimu UMEMOTO #ifdef INET6 3634dcdb4371SBjoern A. Zeeb if (inc->inc_flags & INC_ISIPV6) { 3635603724d3SBjoern A. Zeeb mss = V_tcp_v6mssdflt; 3636233dcce1SAndre Oppermann maxmtu = tcp_maxmtu6(inc, NULL); 363797d8d152SAndre Oppermann min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); 3638b287c6c7SBjoern A. Zeeb } 363931b3783cSHajimu UMEMOTO #endif 3640b287c6c7SBjoern A. Zeeb #if defined(INET) && defined(INET6) 3641b287c6c7SBjoern A. Zeeb else 3642b287c6c7SBjoern A. Zeeb #endif 3643b287c6c7SBjoern A. Zeeb #ifdef INET 364497d8d152SAndre Oppermann { 3645603724d3SBjoern A. Zeeb mss = V_tcp_mssdflt; 3646233dcce1SAndre Oppermann maxmtu = tcp_maxmtu(inc, NULL); 364797d8d152SAndre Oppermann min_protoh = sizeof(struct tcpiphdr); 364897d8d152SAndre Oppermann } 3649b287c6c7SBjoern A. Zeeb #endif 36503a9391deSBjoern A. Zeeb #if defined(INET6) || defined(INET) 36513a9391deSBjoern A. Zeeb thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ 36523a9391deSBjoern A. Zeeb #endif 36533a9391deSBjoern A. Zeeb 365497d8d152SAndre Oppermann if (maxmtu && thcmtu) 365597d8d152SAndre Oppermann mss = min(maxmtu, thcmtu) - min_protoh; 365697d8d152SAndre Oppermann else if (maxmtu || thcmtu) 365797d8d152SAndre Oppermann mss = max(maxmtu, thcmtu) - min_protoh; 365897d8d152SAndre Oppermann 365997d8d152SAndre Oppermann return (mss); 3660df8bae1dSRodney W. Grimes } 366146f58482SJonathan Lemon 366246f58482SJonathan Lemon 366346f58482SJonathan Lemon /* 3664c068736aSJeffrey Hsu * On a partial ack arrives, force the retransmission of the 3665c068736aSJeffrey Hsu * next unacknowledged segment. Do not clear tp->t_dupacks. 3666c068736aSJeffrey Hsu * By setting snd_nxt to ti_ack, this forces retransmission timer to 3667c068736aSJeffrey Hsu * be started again. 366846f58482SJonathan Lemon */ 3669c068736aSJeffrey Hsu static void 3670ad3f9ab3SAndre Oppermann tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) 367146f58482SJonathan Lemon { 367246f58482SJonathan Lemon tcp_seq onxt = tp->snd_nxt; 367346f58482SJonathan Lemon u_long ocwnd = tp->snd_cwnd; 367446f58482SJonathan Lemon 36758501a69cSRobert Watson INP_WLOCK_ASSERT(tp->t_inpcb); 36767a3244ccSRobert Watson 3677b8152ba7SAndre Oppermann tcp_timer_activate(tp, TT_REXMT, 0); 367846f58482SJonathan Lemon tp->t_rtttime = 0; 367946f58482SJonathan Lemon tp->snd_nxt = th->th_ack; 36806b2a5f92SJayanth Vijayaraghavan /* 3681c068736aSJeffrey Hsu * Set snd_cwnd to one segment beyond acknowledged offset. 3682c068736aSJeffrey Hsu * (tp->snd_una has not yet been updated when this function is called.) 36836b2a5f92SJayanth Vijayaraghavan */ 3684dbc42409SLawrence Stewart tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th); 3685a84db8f4SMatthew Dillon tp->t_flags |= TF_ACKNOW; 36866b2a5f92SJayanth Vijayaraghavan (void) tcp_output(tp); 368746f58482SJonathan Lemon tp->snd_cwnd = ocwnd; 368846f58482SJonathan Lemon if (SEQ_GT(onxt, tp->snd_nxt)) 368946f58482SJonathan Lemon tp->snd_nxt = onxt; 369046f58482SJonathan Lemon /* 369146f58482SJonathan Lemon * Partial window deflation. Relies on fact that tp->snd_una 369246f58482SJonathan Lemon * not updated yet. 369346f58482SJonathan Lemon */ 3694dbc42409SLawrence Stewart if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) 3695dbc42409SLawrence Stewart tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); 3696d7587117SPaul Saab else 3697d7587117SPaul Saab tp->snd_cwnd = 0; 3698d7587117SPaul Saab tp->snd_cwnd += tp->t_maxseg; 369946f58482SJonathan Lemon } 3700