diff --git a/sys/netinet/tcp_ecn.c b/sys/netinet/tcp_ecn.c index 34ecfe1c83c0..5332b3caf950 100644 --- a/sys/netinet/tcp_ecn.c +++ b/sys/netinet/tcp_ecn.c @@ -1,611 +1,612 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 * The Regents of the University of California. All rights reserved. * Copyright (c) 2007-2008,2010 * Swinburne University of Technology, Melbourne, Australia. * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * Copyright (c) 2010-2011 Juniper Networks, Inc. * Copyright (c) 2019 Richard Scheffenegger * All rights reserved. * * Portions of this software were developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by Lawrence Stewart, * James Healy and David Hayes, made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * Utility functions to deal with Explicit Congestion Notification in TCP * implementing the essential parts of the Accurate ECN extension * https://tools.ietf.org/html/draft-ietf-tcpm-accurate-ecn-09 */ #include #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static inline int tcp_ecn_get_ace(uint16_t); static inline void tcp_ecn_set_ace(uint16_t *, uint32_t); static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "TCP ECN"); VNET_DEFINE(int, tcp_do_ecn) = 2; SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_ecn), 0, "TCP ECN support"); VNET_DEFINE(int, tcp_ecn_maxretries) = 1; SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_ecn_maxretries), 0, "Max retries before giving up on ECN"); /* * Process incoming SYN,ACK packet */ void tcp_ecn_input_syn_sent(struct tcpcb *tp, uint16_t thflags, int iptos) { switch (V_tcp_do_ecn) { case 0: return; case 1: /* FALLTHROUGH */ case 2: /* RFC3168 ECN handling */ if ((thflags & (TH_CWR | TH_ECE)) == (0 | TH_ECE)) { tp->t_flags2 |= TF2_ECN_PERMIT; tp->t_flags2 &= ~TF2_ACE_PERMIT; TCPSTAT_INC(tcps_ecn_shs); } break; case 3: /* FALLTHROUGH */ case 4: - /* decoding Accurate ECN according to + /* + * Decoding Accurate ECN according to * table in section 3.1.1 - * on the SYN,ACK, process the AccECN + * + * On the SYN,ACK, process the AccECN * flags indicating the state the SYN * was delivered. * Reactions to Path ECN mangling can * come here. */ switch (thflags & (TH_AE | TH_CWR | TH_ECE)) { /* RFC3168 SYN */ case (0|0|TH_ECE): tp->t_flags2 |= TF2_ECN_PERMIT; tp->t_flags2 &= ~TF2_ACE_PERMIT; TCPSTAT_INC(tcps_ecn_shs); break; /* non-ECT SYN */ case (0|TH_CWR|0): tp->t_flags2 |= TF2_ACE_PERMIT; tp->t_flags2 &= ~TF2_ECN_PERMIT; tp->t_scep = 5; TCPSTAT_INC(tcps_ecn_shs); TCPSTAT_INC(tcps_ace_nect); break; /* ECT0 SYN */ case (TH_AE|0|0): tp->t_flags2 |= TF2_ACE_PERMIT; tp->t_flags2 &= ~TF2_ECN_PERMIT; tp->t_scep = 5; TCPSTAT_INC(tcps_ecn_shs); TCPSTAT_INC(tcps_ace_ect0); break; /* ECT1 SYN */ case (0|TH_CWR|TH_ECE): tp->t_flags2 |= TF2_ACE_PERMIT; tp->t_flags2 &= ~TF2_ECN_PERMIT; tp->t_scep = 5; TCPSTAT_INC(tcps_ecn_shs); TCPSTAT_INC(tcps_ace_ect1); break; /* CE SYN */ case (TH_AE|TH_CWR|0): tp->t_flags2 |= TF2_ACE_PERMIT; tp->t_flags2 &= ~TF2_ECN_PERMIT; tp->t_scep = 6; /* * reduce the IW to 2 MSS (to * account for delayed acks) if * the SYN,ACK was CE marked */ tp->snd_cwnd = 2 * tcp_maxseg(tp); TCPSTAT_INC(tcps_ecn_shs); TCPSTAT_INC(tcps_ace_nect); break; default: tp->t_flags2 &= ~(TF2_ECN_PERMIT | TF2_ACE_PERMIT); break; } /* * Set the AccECN Codepoints on * the outgoing to the ECN * state of the * according to table 3 in the * AccECN draft */ switch (iptos & IPTOS_ECN_MASK) { case (IPTOS_ECN_NOTECT): tp->t_rcep = 0b010; break; case (IPTOS_ECN_ECT0): tp->t_rcep = 0b100; break; case (IPTOS_ECN_ECT1): tp->t_rcep = 0b011; break; case (IPTOS_ECN_CE): tp->t_rcep = 0b110; break; } break; } } /* * Handle parallel SYN for ECN */ void tcp_ecn_input_parallel_syn(struct tcpcb *tp, uint16_t thflags, int iptos) { if (thflags & TH_ACK) return; switch (V_tcp_do_ecn) { case 0: return; case 1: /* FALLTHROUGH */ case 2: /* RFC3168 ECN handling */ if ((thflags & (TH_CWR | TH_ECE)) == (TH_CWR | TH_ECE)) { tp->t_flags2 |= TF2_ECN_PERMIT; tp->t_flags2 &= ~TF2_ACE_PERMIT; tp->t_flags2 |= TF2_ECN_SND_ECE; TCPSTAT_INC(tcps_ecn_shs); } break; case 3: /* FALLTHROUGH */ case 4: /* AccECN handling */ switch (thflags & (TH_AE | TH_CWR | TH_ECE)) { default: case (0|0|0): tp->t_flags2 &= ~(TF2_ECN_PERMIT | TF2_ACE_PERMIT); break; case (0|TH_CWR|TH_ECE): tp->t_flags2 |= TF2_ECN_PERMIT; tp->t_flags2 &= ~TF2_ACE_PERMIT; tp->t_flags2 |= TF2_ECN_SND_ECE; TCPSTAT_INC(tcps_ecn_shs); break; case (TH_AE|TH_CWR|TH_ECE): tp->t_flags2 |= TF2_ACE_PERMIT; tp->t_flags2 &= ~TF2_ECN_PERMIT; TCPSTAT_INC(tcps_ecn_shs); /* * Set the AccECN Codepoints on * the outgoing to the ECN * state of the * according to table 3 in the * AccECN draft */ switch (iptos & IPTOS_ECN_MASK) { case (IPTOS_ECN_NOTECT): tp->t_rcep = 0b010; break; case (IPTOS_ECN_ECT0): tp->t_rcep = 0b100; break; case (IPTOS_ECN_ECT1): tp->t_rcep = 0b011; break; case (IPTOS_ECN_CE): tp->t_rcep = 0b110; break; } break; } break; } } /* * TCP ECN processing. */ int tcp_ecn_input_segment(struct tcpcb *tp, uint16_t thflags, int tlen, int pkts, int iptos) { int delta_cep = 0; switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: TCPSTAT_INC(tcps_ecn_rcvce); break; case IPTOS_ECN_ECT0: TCPSTAT_INC(tcps_ecn_rcvect0); break; case IPTOS_ECN_ECT1: TCPSTAT_INC(tcps_ecn_rcvect1); break; } if (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT)) { if (tp->t_flags2 & TF2_ACE_PERMIT) { if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) tp->t_rcep += 1; if (tp->t_flags2 & TF2_ECN_PERMIT) { delta_cep = (tcp_ecn_get_ace(thflags) + 8 - (tp->t_scep & 7)) & 7; if (delta_cep < pkts) delta_cep = pkts - ((pkts - delta_cep) & 7); tp->t_scep += delta_cep; } else { /* * process the final ACK of the 3WHS * see table 3 in draft-ietf-tcpm-accurate-ecn */ switch (tcp_ecn_get_ace(thflags)) { case 0b010: /* nonECT SYN or SYN,ACK */ /* FALLTHROUGH */ case 0b011: /* ECT1 SYN or SYN,ACK */ /* FALLTHROUGH */ case 0b100: /* ECT0 SYN or SYN,ACK */ tp->t_scep = 5; break; case 0b110: /* CE SYN or SYN,ACK */ tp->t_scep = 6; tp->snd_cwnd = 2 * tcp_maxseg(tp); break; default: /* mangled AccECN handshake */ tp->t_scep = 5; break; } tp->t_flags2 |= TF2_ECN_PERMIT; } } else { /* RFC3168 ECN handling */ if ((thflags & (TH_SYN | TH_ECE)) == TH_ECE) { delta_cep = 1; tp->t_scep++; } if (thflags & TH_CWR) { tp->t_flags2 &= ~TF2_ECN_SND_ECE; tp->t_flags |= TF_ACKNOW; } if ((iptos & IPTOS_ECN_MASK) == IPTOS_ECN_CE) tp->t_flags2 |= TF2_ECN_SND_ECE; } /* Process a packet differently from RFC3168. */ cc_ecnpkt_handler_flags(tp, thflags, iptos); } return delta_cep; } /* * Send ECN setup packet header flags */ uint16_t tcp_ecn_output_syn_sent(struct tcpcb *tp) { uint16_t thflags = 0; if (V_tcp_do_ecn == 0) return thflags; if (V_tcp_do_ecn == 1) { /* Send a RFC3168 ECN setup packet */ if (tp->t_rxtshift >= 1) { if (tp->t_rxtshift <= V_tcp_ecn_maxretries) thflags = TH_ECE|TH_CWR; } else thflags = TH_ECE|TH_CWR; - } else - if (V_tcp_do_ecn == 3) { + } else if (V_tcp_do_ecn == 3) { /* Send an Accurate ECN setup packet */ if (tp->t_rxtshift >= 1) { if (tp->t_rxtshift <= V_tcp_ecn_maxretries) thflags = TH_ECE|TH_CWR|TH_AE; } else thflags = TH_ECE|TH_CWR|TH_AE; } return thflags; } /* * output processing of ECN feature * returning IP ECN header codepoint */ int tcp_ecn_output_established(struct tcpcb *tp, uint16_t *thflags, int len, bool rxmit) { int ipecn = IPTOS_ECN_NOTECT; bool newdata; /* * If the peer has ECN, mark data packets with * ECN capable transmission (ECT). * Ignore pure control packets, retransmissions * and window probes. */ newdata = (len > 0 && SEQ_GEQ(tp->snd_nxt, tp->snd_max) && !rxmit && !((tp->t_flags & TF_FORCEDATA) && len == 1)); /* RFC3168 ECN marking, only new data segments */ if (newdata) { if (tp->t_flags2 & TF2_ECN_USE_ECT1) { ipecn = IPTOS_ECN_ECT1; TCPSTAT_INC(tcps_ecn_sndect1); } else { ipecn = IPTOS_ECN_ECT0; TCPSTAT_INC(tcps_ecn_sndect0); } } /* * Reply with proper ECN notifications. */ if (tp->t_flags2 & TF2_ACE_PERMIT) { tcp_ecn_set_ace(thflags, tp->t_rcep); if (!(tp->t_flags2 & TF2_ECN_PERMIT)) { /* * here we process the final * ACK of the 3WHS */ if (tp->t_rcep == 0b110) { tp->t_rcep = 6; } else { tp->t_rcep = 5; } tp->t_flags2 |= TF2_ECN_PERMIT; } } else { if (newdata && (tp->t_flags2 & TF2_ECN_SND_CWR)) { *thflags |= TH_CWR; tp->t_flags2 &= ~TF2_ECN_SND_CWR; } if (tp->t_flags2 & TF2_ECN_SND_ECE) *thflags |= TH_ECE; } return ipecn; } /* * Set up the ECN related tcpcb fields from * a syncache entry */ void tcp_ecn_syncache_socket(struct tcpcb *tp, struct syncache *sc) { if (sc->sc_flags & SCF_ECN_MASK) { switch (sc->sc_flags & SCF_ECN_MASK) { case SCF_ECN: tp->t_flags2 |= TF2_ECN_PERMIT; break; case SCF_ACE_N: /* FALLTHROUGH */ case SCF_ACE_0: /* FALLTHROUGH */ case SCF_ACE_1: tp->t_flags2 |= TF2_ACE_PERMIT; tp->t_scep = 5; tp->t_rcep = 5; break; case SCF_ACE_CE: tp->t_flags2 |= TF2_ACE_PERMIT; tp->t_scep = 6; tp->t_rcep = 6; break; } } } /* * Process a packets ECN information, and provide the * syncache with the relevant information. */ int tcp_ecn_syncache_add(uint16_t thflags, int iptos) { int scflags = 0; switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: TCPSTAT_INC(tcps_ecn_rcvce); break; case IPTOS_ECN_ECT0: TCPSTAT_INC(tcps_ecn_rcvect0); break; case IPTOS_ECN_ECT1: TCPSTAT_INC(tcps_ecn_rcvect1); break; } switch (thflags & (TH_AE|TH_CWR|TH_ECE)) { /* no ECN */ case (0|0|0): break; /* legacy ECN */ case (0|TH_CWR|TH_ECE): scflags = SCF_ECN; break; /* Accurate ECN */ case (TH_AE|TH_CWR|TH_ECE): if ((V_tcp_do_ecn == 3) || (V_tcp_do_ecn == 4)) { switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: scflags = SCF_ACE_CE; break; case IPTOS_ECN_ECT0: scflags = SCF_ACE_0; break; case IPTOS_ECN_ECT1: scflags = SCF_ACE_1; break; case IPTOS_ECN_NOTECT: scflags = SCF_ACE_N; break; } } else scflags = SCF_ECN; break; /* Default Case (section 3.1.2) */ default: if ((V_tcp_do_ecn == 3) || (V_tcp_do_ecn == 4)) { switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: scflags = SCF_ACE_CE; break; case IPTOS_ECN_ECT0: scflags = SCF_ACE_0; break; case IPTOS_ECN_ECT1: scflags = SCF_ACE_1; break; case IPTOS_ECN_NOTECT: scflags = SCF_ACE_N; break; } } break; } return scflags; } /* * Set up the ECN information for the from * syncache information. */ uint16_t tcp_ecn_syncache_respond(uint16_t thflags, struct syncache *sc) { if ((thflags & TH_SYN) && (sc->sc_flags & SCF_ECN_MASK)) { switch (sc->sc_flags & SCF_ECN_MASK) { case SCF_ECN: thflags |= (0 | 0 | TH_ECE); TCPSTAT_INC(tcps_ecn_shs); break; case SCF_ACE_N: thflags |= (0 | TH_CWR | 0); TCPSTAT_INC(tcps_ecn_shs); TCPSTAT_INC(tcps_ace_nect); break; case SCF_ACE_0: thflags |= (TH_AE | 0 | 0); TCPSTAT_INC(tcps_ecn_shs); TCPSTAT_INC(tcps_ace_ect0); break; case SCF_ACE_1: thflags |= (0 | TH_ECE | TH_CWR); TCPSTAT_INC(tcps_ecn_shs); TCPSTAT_INC(tcps_ace_ect1); break; case SCF_ACE_CE: thflags |= (TH_AE | TH_CWR | 0); TCPSTAT_INC(tcps_ecn_shs); TCPSTAT_INC(tcps_ace_ce); break; } } return thflags; } static inline int tcp_ecn_get_ace(uint16_t thflags) { return ((thflags & (TH_AE|TH_CWR|TH_ECE)) >> TH_ACE_SHIFT); } static inline void tcp_ecn_set_ace(uint16_t *thflags, uint32_t t_rcep) { *thflags &= ~(TH_AE|TH_CWR|TH_ECE); *thflags |= ((t_rcep << TH_ACE_SHIFT) & (TH_AE|TH_CWR|TH_ECE)); } diff --git a/sys/netinet/tcp_input.c b/sys/netinet/tcp_input.c index 02b042fd3273..98564ff67f22 100644 --- a/sys/netinet/tcp_input.c +++ b/sys/netinet/tcp_input.c @@ -1,4118 +1,4134 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 * The Regents of the University of California. All rights reserved. * Copyright (c) 2007-2008,2010 * Swinburne University of Technology, Melbourne, Australia. * Copyright (c) 2009-2010 Lawrence Stewart * Copyright (c) 2010 The FreeBSD Foundation * Copyright (c) 2010-2011 Juniper Networks, Inc. * All rights reserved. * * Portions of this software were developed at the Centre for Advanced Internet * Architectures, Swinburne University of Technology, by Lawrence Stewart, * James Healy and David Hayes, made possible in part by a grant from the Cisco * University Research Program Fund at Community Foundation Silicon Valley. * * Portions of this software were developed at the Centre for Advanced * Internet Architectures, Swinburne University of Technology, Melbourne, * Australia by David Hayes under sponsorship from the FreeBSD Foundation. * * Portions of this software were developed by Robert N. M. Watson under * contract to Juniper Networks, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_rss.h" #include #include #include #ifdef TCP_HHOOK #include #endif #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include #include #include #include #include #include /* before tcp_seq.h, for tcp_random18() */ #include #include #include #include #include #include #define TCPSTATES /* for logging */ #include #include #include #include #include #include #include /* required for icmp_var.h */ #include /* for ICMP_BANDLIM */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef TCPPCAP #include #endif #include #ifdef TCP_OFFLOAD #include #endif #include #include #include #include #include const int tcprexmtthresh = 3; VNET_DEFINE(int, tcp_log_in_vain) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_log_in_vain), 0, "Log all incoming TCP segments to closed ports"); VNET_DEFINE(int, blackhole) = 0; #define V_blackhole VNET(blackhole) SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(blackhole), 0, "Do not send RST on segments to closed ports"); VNET_DEFINE(bool, blackhole_local) = false; #define V_blackhole_local VNET(blackhole_local) SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(blackhole_local), false, "Enforce net.inet.tcp.blackhole for locally originated packets"); VNET_DEFINE(int, tcp_delack_enabled) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_delack_enabled), 0, "Delay ACK to try and piggyback it onto a data packet"); VNET_DEFINE(int, drop_synfin) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(drop_synfin), 0, "Drop TCP packets with SYN+FIN set"); VNET_DEFINE(int, tcp_do_prr) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_prr), 1, "Enable Proportional Rate Reduction per RFC 6937"); VNET_DEFINE(int, tcp_do_newcwv) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_newcwv), 0, "Enable New Congestion Window Validation per RFC7661"); VNET_DEFINE(int, tcp_do_rfc3042) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3042), 0, "Enable RFC 3042 (Limited Transmit)"); VNET_DEFINE(int, tcp_do_rfc3390) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3390), 0, "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)"); VNET_DEFINE(int, tcp_initcwnd_segments) = 10; SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0, "Slow-start flight size (initial congestion window) in number of segments"); VNET_DEFINE(int, tcp_do_rfc3465) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_rfc3465), 0, "Enable RFC 3465 (Appropriate Byte Counting)"); VNET_DEFINE(int, tcp_abc_l_var) = 2; SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_abc_l_var), 2, "Cap the max cwnd increment during slow-start to this number of segments"); VNET_DEFINE(int, tcp_insecure_syn) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_insecure_syn), 0, "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets"); VNET_DEFINE(int, tcp_insecure_rst) = 0; SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_insecure_rst), 0, "Follow RFC793 instead of RFC5961 criteria for accepting RST packets"); VNET_DEFINE(int, tcp_recvspace) = 1024*64; #define V_tcp_recvspace VNET(tcp_recvspace) SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size"); VNET_DEFINE(int, tcp_do_autorcvbuf) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_autorcvbuf), 0, "Enable automatic receive buffer sizing"); VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024; SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autorcvbuf_max), 0, "Max size of automatic receive buffer"); VNET_DEFINE(struct inpcbinfo, tcbinfo); /* * TCP statistics are stored in an array of counter(9)s, which size matches * size of struct tcpstat. TCP running connection count is a regular array. */ VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat); SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat, tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)"); VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]); SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD | CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES, "TCP connection counts by TCP state"); /* * Kernel module interface for updating tcpstat. The first argument is an index * into tcpstat treated as an array. */ void kmod_tcpstat_add(int statnum, int val) { counter_u64_add(VNET(tcpstat)[statnum], val); } /* * Make sure that we only start a SACK loss recovery when * receiving a duplicate ACK with a SACK block, and also * complete SACK loss recovery in case the other end * reneges. */ static bool inline tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to) { return ((tp->t_flags & TF_SACK_PERMIT) && ((to->to_flags & TOF_SACK) || (!TAILQ_EMPTY(&tp->snd_holes)))); } #ifdef TCP_HHOOK /* * Wrapper for the TCP established input helper hook. */ void hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to) { struct tcp_hhook_data hhook_data; if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) { hhook_data.tp = tp; hhook_data.th = th; hhook_data.to = to; hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data, &tp->t_osd); } } #endif /* * CC wrapper hook functions */ void cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs, uint16_t type) { #ifdef STATS int32_t gput; #endif INP_WLOCK_ASSERT(tptoinpcb(tp)); tp->t_ccv.nsegs = nsegs; tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th); if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) || (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) && (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2)))) tp->t_ccv.flags |= CCF_CWND_LIMITED; else tp->t_ccv.flags &= ~CCF_CWND_LIMITED; if (type == CC_ACK) { #ifdef STATS stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF, ((int32_t)tp->snd_cwnd) - tp->snd_wnd); if (!IN_RECOVERY(tp->t_flags)) stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN, tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs)); if ((tp->t_flags & TF_GPUTINPROG) && SEQ_GEQ(th->th_ack, tp->gput_ack)) { /* * Compute goodput in bits per millisecond. */ gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) / max(1, tcp_ts_getticks() - tp->gput_ts); stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT, gput); /* * XXXLAS: This is a temporary hack, and should be * chained off VOI_TCP_GPUT when stats(9) grows an API * to deal with chained VOIs. */ if (tp->t_stats_gput_prev > 0) stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_GPUT_ND, ((gput - tp->t_stats_gput_prev) * 100) / tp->t_stats_gput_prev); tp->t_flags &= ~TF_GPUTINPROG; tp->t_stats_gput_prev = gput; } #endif /* STATS */ if (tp->snd_cwnd > tp->snd_ssthresh) { tp->t_bytes_acked += tp->t_ccv.bytes_this_ack; if (tp->t_bytes_acked >= tp->snd_cwnd) { tp->t_bytes_acked -= tp->snd_cwnd; tp->t_ccv.flags |= CCF_ABC_SENTAWND; } } else { tp->t_ccv.flags &= ~CCF_ABC_SENTAWND; tp->t_bytes_acked = 0; } } if (CC_ALGO(tp)->ack_received != NULL) { /* XXXLAS: Find a way to live without this */ tp->t_ccv.curack = th->th_ack; CC_ALGO(tp)->ack_received(&tp->t_ccv, type); } #ifdef STATS stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd); #endif } void cc_conn_init(struct tcpcb *tp) { struct hc_metrics_lite metrics; struct inpcb *inp = tptoinpcb(tp); u_int maxseg; int rtt; INP_WLOCK_ASSERT(inp); tcp_hc_get(&inp->inp_inc, &metrics); maxseg = tcp_maxseg(tp); if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) { tp->t_srtt = rtt; TCPSTAT_INC(tcps_usedrtt); if (metrics.rmx_rttvar) { tp->t_rttvar = metrics.rmx_rttvar; TCPSTAT_INC(tcps_usedrttvar); } else { /* default variation is +- 1 rtt */ tp->t_rttvar = tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE; } TCPT_RANGESET(tp->t_rxtcur, ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1, tp->t_rttmin, TCPTV_REXMTMAX); } if (metrics.rmx_ssthresh) { /* * There's some sort of gateway or interface * buffer limit on the path. Use this to set * the slow start threshold, but set the * threshold to no less than 2*mss. */ tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh); TCPSTAT_INC(tcps_usedssthresh); } /* * Set the initial slow-start flight size. * * If a SYN or SYN/ACK was lost and retransmitted, we have to * reduce the initial CWND to one segment as congestion is likely * requiring us to be cautious. */ if (tp->snd_cwnd == 1) tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */ else tp->snd_cwnd = tcp_compute_initwnd(maxseg); if (CC_ALGO(tp)->conn_init != NULL) CC_ALGO(tp)->conn_init(&tp->t_ccv); } void inline cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type) { INP_WLOCK_ASSERT(tptoinpcb(tp)); #ifdef STATS stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type); #endif switch(type) { case CC_NDUPACK: if (!IN_FASTRECOVERY(tp->t_flags)) { tp->snd_recover = tp->snd_max; if (tp->t_flags2 & TF2_ECN_PERMIT) tp->t_flags2 |= TF2_ECN_SND_CWR; } break; case CC_ECN: if (!IN_CONGRECOVERY(tp->t_flags) || /* * Allow ECN reaction on ACK to CWR, if * that data segment was also CE marked. */ SEQ_GEQ(th->th_ack, tp->snd_recover)) { EXIT_CONGRECOVERY(tp->t_flags); TCPSTAT_INC(tcps_ecn_rcwnd); tp->snd_recover = tp->snd_max + 1; if (tp->t_flags2 & TF2_ECN_PERMIT) tp->t_flags2 |= TF2_ECN_SND_CWR; } break; case CC_RTO: tp->t_dupacks = 0; tp->t_bytes_acked = 0; if ((tp->t_rxtshift > 1) || !((tp->t_flags & TF_SACK_PERMIT) && (!TAILQ_EMPTY(&tp->snd_holes)))) EXIT_RECOVERY(tp->t_flags); if (tp->t_flags2 & TF2_ECN_PERMIT) tp->t_flags2 |= TF2_ECN_SND_CWR; break; case CC_RTO_ERR: TCPSTAT_INC(tcps_sndrexmitbad); /* RTO was unnecessary, so reset everything. */ tp->snd_cwnd = tp->snd_cwnd_prev; tp->snd_ssthresh = tp->snd_ssthresh_prev; tp->snd_recover = tp->snd_recover_prev; if (tp->t_flags & TF_WASFRECOVERY) ENTER_FASTRECOVERY(tp->t_flags); if (tp->t_flags & TF_WASCRECOVERY) ENTER_CONGRECOVERY(tp->t_flags); tp->snd_nxt = tp->snd_max; tp->t_flags &= ~TF_PREVVALID; tp->t_badrxtwin = 0; break; } if (CC_ALGO(tp)->cong_signal != NULL) { if (th != NULL) tp->t_ccv.curack = th->th_ack; CC_ALGO(tp)->cong_signal(&tp->t_ccv, type); } } void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th) { INP_WLOCK_ASSERT(tptoinpcb(tp)); /* XXXLAS: KASSERT that we're in recovery? */ if (CC_ALGO(tp)->post_recovery != NULL) { tp->t_ccv.curack = th->th_ack; CC_ALGO(tp)->post_recovery(&tp->t_ccv); } /* XXXLAS: EXIT_RECOVERY ? */ tp->t_bytes_acked = 0; tp->sackhint.delivered_data = 0; tp->sackhint.prr_delivered = 0; tp->sackhint.prr_out = 0; } /* * Indicate whether this ack should be delayed. We can delay the ack if * following conditions are met: * - There is no delayed ack timer in progress. * - Our last ack wasn't a 0-sized window. We never want to delay * the ack that opens up a 0-sized window. * - LRO wasn't used for this segment. We make sure by checking that the * segment size is not larger than the MSS. */ #define DELAY_ACK(tp, tlen) \ ((!tcp_timer_active(tp, TT_DELACK) && \ (tp->t_flags & TF_RXWIN0SENT) == 0) && \ (tlen <= tp->t_maxseg) && \ (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN))) void inline cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos) { INP_WLOCK_ASSERT(tptoinpcb(tp)); if (CC_ALGO(tp)->ecnpkt_handler != NULL) { switch (iptos & IPTOS_ECN_MASK) { case IPTOS_ECN_CE: tp->t_ccv.flags |= CCF_IPHDR_CE; break; case IPTOS_ECN_ECT0: /* FALLTHROUGH */ case IPTOS_ECN_ECT1: /* FALLTHROUGH */ case IPTOS_ECN_NOTECT: tp->t_ccv.flags &= ~CCF_IPHDR_CE; break; } if (flags & TH_CWR) tp->t_ccv.flags |= CCF_TCPHDR_CWR; else tp->t_ccv.flags &= ~CCF_TCPHDR_CWR; CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv); if (tp->t_ccv.flags & CCF_ACKNOW) { tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); tp->t_flags |= TF_ACKNOW; } } } void inline cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos) { cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos); } /* * TCP input handling is split into multiple parts: * tcp6_input is a thin wrapper around tcp_input for the extended * ip6_protox[] call format in ip6_input * tcp_input handles primary segment validation, inpcb lookup and * SYN processing on listen sockets * tcp_do_segment processes the ACK and text of the segment for * establishing, established and closing connections */ #ifdef INET6 int tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) { struct mbuf *m; struct in6_ifaddr *ia6; struct ip6_hdr *ip6; m = *mp; if (m->m_len < *offp + sizeof(struct tcphdr)) { m = m_pullup(m, *offp + sizeof(struct tcphdr)); if (m == NULL) { *mp = m; TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } /* * draft-itojun-ipv6-tcp-to-anycast * better place to put this in? */ ip6 = mtod(m, struct ip6_hdr *); ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) { icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR, (caddr_t)&ip6->ip6_dst - (caddr_t)ip6); *mp = NULL; return (IPPROTO_DONE); } *mp = m; return (tcp_input_with_port(mp, offp, proto, port)); } int tcp6_input(struct mbuf **mp, int *offp, int proto) { return(tcp6_input_with_port(mp, offp, proto, 0)); } #endif /* INET6 */ int tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port) { struct mbuf *m = *mp; struct tcphdr *th = NULL; struct ip *ip = NULL; struct inpcb *inp = NULL; struct tcpcb *tp = NULL; struct socket *so = NULL; u_char *optp = NULL; int off0; int optlen = 0; #ifdef INET int len; uint8_t ipttl; #endif int tlen = 0, off; int drop_hdrlen; int thflags; int rstreason = 0; /* For badport_bandlim accounting purposes */ int lookupflag; uint8_t iptos; struct m_tag *fwd_tag = NULL; #ifdef INET6 struct ip6_hdr *ip6 = NULL; int isipv6; #else const void *ip6 = NULL; #endif /* INET6 */ struct tcpopt to; /* options in this segment */ char *s = NULL; /* address and port logging */ NET_EPOCH_ASSERT(); #ifdef INET6 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0; #endif off0 = *offp; m = *mp; *mp = NULL; to.to_flags = 0; TCPSTAT_INC(tcps_rcvtotal); m->m_pkthdr.tcp_tun_port = port; #ifdef INET6 if (isipv6) { ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)((caddr_t)ip6 + off0); tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0; if (port) goto skip6_csum; if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) th->th_sum = m->m_pkthdr.csum_data; else th->th_sum = in6_cksum_pseudo(ip6, tlen, IPPROTO_TCP, m->m_pkthdr.csum_data); th->th_sum ^= 0xffff; } else th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen); if (th->th_sum) { TCPSTAT_INC(tcps_rcvbadsum); goto drop; } skip6_csum: /* * Be proactive about unspecified IPv6 address in source. * As we use all-zero to indicate unbounded/unconnected pcb, * unspecified IPv6 address can be used to confuse us. * * Note that packets with unspecified IPv6 destination is * already dropped in ip6_input. */ KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst), ("%s: unspecified destination v6 address", __func__)); if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) { IP6STAT_INC(ip6s_badscope); /* XXX */ goto drop; } iptos = IPV6_TRAFFIC_CLASS(ip6); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { /* * Get IP and TCP header together in first mbuf. * Note: IP leaves IP header in first mbuf. */ if (off0 > sizeof (struct ip)) { ip_stripoptions(m); off0 = sizeof(struct ip); } if (m->m_len < sizeof (struct tcpiphdr)) { if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } ip = mtod(m, struct ip *); th = (struct tcphdr *)((caddr_t)ip + off0); tlen = ntohs(ip->ip_len) - off0; iptos = ip->ip_tos; if (port) goto skip_csum; if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) { if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR) th->th_sum = m->m_pkthdr.csum_data; else th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htonl(m->m_pkthdr.csum_data + tlen + IPPROTO_TCP)); th->th_sum ^= 0xffff; } else { struct ipovly *ipov = (struct ipovly *)ip; /* * Checksum extended TCP header and data. */ len = off0 + tlen; ipttl = ip->ip_ttl; bzero(ipov->ih_x1, sizeof(ipov->ih_x1)); ipov->ih_len = htons(tlen); th->th_sum = in_cksum(m, len); /* Reset length for SDT probes. */ ip->ip_len = htons(len); /* Reset TOS bits */ ip->ip_tos = iptos; /* Re-initialization for later version check */ ip->ip_ttl = ipttl; ip->ip_v = IPVERSION; ip->ip_hl = off0 >> 2; } skip_csum: if (th->th_sum && (port == 0)) { TCPSTAT_INC(tcps_rcvbadsum); goto drop; } KASSERT(ip->ip_dst.s_addr != INADDR_ANY, ("%s: unspecified destination v4 address", __func__)); if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) { IPSTAT_INC(ips_badaddr); goto drop; } } #endif /* INET */ /* * Check that TCP offset makes sense, * pull out TCP options and adjust length. XXX */ off = th->th_off << 2; if (off < sizeof (struct tcphdr) || off > tlen) { TCPSTAT_INC(tcps_rcvbadoff); goto drop; } tlen -= off; /* tlen is used instead of ti->ti_len */ if (off > sizeof (struct tcphdr)) { #ifdef INET6 if (isipv6) { if (m->m_len < off0 + off) { m = m_pullup(m, off0 + off); if (m == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } } ip6 = mtod(m, struct ip6_hdr *); th = (struct tcphdr *)((caddr_t)ip6 + off0); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { if (m->m_len < sizeof(struct ip) + off) { if ((m = m_pullup(m, sizeof (struct ip) + off)) == NULL) { TCPSTAT_INC(tcps_rcvshort); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); th = (struct tcphdr *)((caddr_t)ip + off0); } } #endif optlen = off - sizeof (struct tcphdr); optp = (u_char *)(th + 1); } thflags = tcp_get_flags(th); /* * Convert TCP protocol specific fields to host format. */ tcp_fields_to_host(th); /* * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options. */ drop_hdrlen = off0 + off; /* * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain. */ if ( #ifdef INET6 (isipv6 && (m->m_flags & M_IP6_NEXTHOP)) #ifdef INET || (!isipv6 && (m->m_flags & M_IP_NEXTHOP)) #endif #endif #if defined(INET) && !defined(INET6) (m->m_flags & M_IP_NEXTHOP) #endif ) fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL); /* * For initial SYN packets we don't need write lock on matching * PCB, be it a listening one or a synchronized one. The packet * shall not modify its state. */ lookupflag = INPLOOKUP_WILDCARD | ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ? INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB); findpcb: #ifdef INET6 if (isipv6 && fwd_tag != NULL) { struct sockaddr_in6 *next_hop6; next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1); /* * Transparently forwarded. Pretend to be the destination. * Already got one like this? */ inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &next_hop6->sin6_addr, next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) : th->th_dport, lookupflag, m->m_pkthdr.rcvif); } } else if (isipv6) { inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); } #endif /* INET6 */ #if defined(INET6) && defined(INET) else #endif #ifdef INET if (fwd_tag != NULL) { struct sockaddr_in *next_hop; next_hop = (struct sockaddr_in *)(fwd_tag+1); /* * Transparently forwarded. Pretend to be the destination. * already got one like this? */ inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m); if (!inp) { /* * It's new. Try to find the ambushing socket. * Because we've rewritten the destination address, * any hardware-generated hash is ignored. */ inp = in_pcblookup(&V_tcbinfo, ip->ip_src, th->th_sport, next_hop->sin_addr, next_hop->sin_port ? ntohs(next_hop->sin_port) : th->th_dport, lookupflag, m->m_pkthdr.rcvif); } } else inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport, ip->ip_dst, th->th_dport, lookupflag, m->m_pkthdr.rcvif, m); #endif /* INET */ /* * If the INPCB does not exist then all data in the incoming * segment is discarded and an appropriate RST is sent back. * XXX MRT Send RST using which routing table? */ if (inp == NULL) { if (rstreason != 0) { /* We came here after second (safety) lookup. */ MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0); goto dropwithreset; } /* * Log communication attempts to ports that are not * in use. */ if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) || V_tcp_log_in_vain == 2) { if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6))) log(LOG_INFO, "%s; %s: Connection attempt " "to closed port\n", s, __func__); } /* * When blackholing do not respond with a RST but * completely ignore the segment and drop it. */ if (((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole == 2) && (V_blackhole_local || ( #ifdef INET6 isipv6 ? !in6_localaddr(&ip6->ip6_src) : #endif #ifdef INET !in_localip(ip->ip_src) #else true #endif ))) goto dropunlock; rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } INP_LOCK_ASSERT(inp); if ((inp->inp_flowtype == M_HASHTYPE_NONE) && !SOLISTENING(inp->inp_socket)) { if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) { inp->inp_flowid = m->m_pkthdr.flowid; inp->inp_flowtype = M_HASHTYPE_GET(m); #ifdef RSS } else { /* assign flowid by software RSS hash */ #ifdef INET6 if (isipv6) { rss_proto_software_hash_v6(&inp->in6p_faddr, &inp->in6p_laddr, inp->inp_fport, inp->inp_lport, IPPROTO_TCP, &inp->inp_flowid, &inp->inp_flowtype); } else #endif /* INET6 */ { rss_proto_software_hash_v4(inp->inp_faddr, inp->inp_laddr, inp->inp_fport, inp->inp_lport, IPPROTO_TCP, &inp->inp_flowid, &inp->inp_flowtype); } #endif /* RSS */ } } #if defined(IPSEC) || defined(IPSEC_SUPPORT) #ifdef INET6 if (isipv6 && IPSEC_ENABLED(ipv6) && IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) { goto dropunlock; } #ifdef INET else #endif #endif /* INET6 */ #ifdef INET if (IPSEC_ENABLED(ipv4) && IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) { goto dropunlock; } #endif /* INET */ #endif /* IPSEC */ /* * Check the minimum TTL for socket. */ if (inp->inp_ip_minttl != 0) { #ifdef INET6 if (isipv6) { if (inp->inp_ip_minttl > ip6->ip6_hlim) goto dropunlock; } else #endif if (inp->inp_ip_minttl > ip->ip_ttl) goto dropunlock; } tp = intotcpcb(inp); switch (tp->t_state) { case TCPS_TIME_WAIT: /* * A previous connection in TIMEWAIT state is supposed to catch * stray or duplicate segments arriving late. If this segment * was a legitimate new connection attempt, the old INPCB gets * removed and we can try again to find a listening socket. */ tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) ? TO_SYN : 0); /* * tcp_twcheck unlocks the inp always, and frees the m if fails. */ if (tcp_twcheck(inp, &to, th, m, tlen)) goto findpcb; return (IPPROTO_DONE); case TCPS_CLOSED: /* * The TCPCB may no longer exist if the connection is winding * down or it is in the CLOSED state. Either way we drop the * segment and send an appropriate response. */ rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) { rstreason = BANDLIM_RST_CLOSEDPORT; goto dropwithreset; } #ifdef TCP_OFFLOAD if (tp->t_flags & TF_TOE) { tcp_offload_input(tp, m); m = NULL; /* consumed by the TOE driver */ goto dropunlock; } #endif #ifdef MAC if (mac_inpcb_check_deliver(inp, m)) goto dropunlock; #endif so = inp->inp_socket; KASSERT(so != NULL, ("%s: so == NULL", __func__)); /* * When the socket is accepting connections (the INPCB is in LISTEN * state) we look into the SYN cache if this is a new connection * attempt or the completion of a previous one. */ KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so), ("%s: so accepting but tp %p not listening", __func__, tp)); if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) { struct in_conninfo inc; bzero(&inc, sizeof(inc)); #ifdef INET6 if (isipv6) { inc.inc_flags |= INC_ISIPV6; if (inp->inp_inc.inc_flags & INC_IPV6MINMTU) inc.inc_flags |= INC_IPV6MINMTU; inc.inc6_faddr = ip6->ip6_src; inc.inc6_laddr = ip6->ip6_dst; } else #endif { inc.inc_faddr = ip->ip_src; inc.inc_laddr = ip->ip_dst; } inc.inc_fport = th->th_sport; inc.inc_lport = th->th_dport; inc.inc_fibnum = so->so_fibnum; /* * Check for an existing connection attempt in syncache if * the flag is only ACK. A successful lookup creates a new * socket appended to the listen queue in SYN_RECEIVED state. */ if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) { /* * Parse the TCP options here because * syncookies need access to the reflected * timestamp. */ tcp_dooptions(&to, optp, optlen, 0); /* * NB: syncache_expand() doesn't unlock inp. */ rstreason = syncache_expand(&inc, &to, th, &so, m, port); if (rstreason < 0) { /* * A failing TCP MD5 signature comparison * must result in the segment being dropped * and must not produce any response back * to the sender. */ goto dropunlock; } else if (rstreason == 0) { /* * No syncache entry, or ACK was not for our * SYN/ACK. Do our protection against double * ACK. If peer sent us 2 ACKs, then for the * first one syncache_expand() successfully * converted syncache entry into a socket, * while we were waiting on the inpcb lock. We * don't want to sent RST for the second ACK, * so we perform second lookup without wildcard * match, hoping to find the new socket. If * the ACK is stray indeed, rstreason would * hint the above code that the lookup was a * second attempt. * * NB: syncache did its own logging * of the failure cause. */ INP_WUNLOCK(inp); rstreason = BANDLIM_RST_OPENPORT; lookupflag &= ~INPLOOKUP_WILDCARD; goto findpcb; } tfo_socket_result: if (so == NULL) { /* * We completed the 3-way handshake * but could not allocate a socket * either due to memory shortage, * listen queue length limits or * global socket limits. Send RST * or wait and have the remote end * retransmit the ACK for another * try. */ if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Socket allocation failed due to " "limits or memory shortage, %s\n", s, __func__, V_tcp_sc_rst_sock_fail ? "sending RST" : "try again"); if (V_tcp_sc_rst_sock_fail) { rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } else goto dropunlock; } /* * Socket is created in state SYN_RECEIVED. * Unlock the listen socket, lock the newly * created socket and update the tp variable. * If we came here via jump to tfo_socket_result, * then listening socket is read-locked. */ INP_UNLOCK(inp); /* listen socket */ inp = sotoinpcb(so); /* * New connection inpcb is already locked by * syncache_expand(). */ INP_WLOCK_ASSERT(inp); tp = intotcpcb(inp); KASSERT(tp->t_state == TCPS_SYN_RECEIVED, ("%s: ", __func__)); /* * Process the segment and the data it * contains. tcp_do_segment() consumes * the mbuf chain and unlocks the inpcb. */ TCP_PROBE5(receive, NULL, tp, m, tp, th); tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); return (IPPROTO_DONE); } /* * Segment flag validation for new connection attempts: * * Our (SYN|ACK) response was rejected. * Check with syncache and remove entry to prevent * retransmits. * * NB: syncache_chkrst does its own logging of failure * causes. */ if (thflags & TH_RST) { syncache_chkrst(&inc, th, m, port); goto dropunlock; } /* * We can't do anything without SYN. */ if ((thflags & TH_SYN) == 0) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN is missing, segment ignored\n", s, __func__); TCPSTAT_INC(tcps_badsyn); goto dropunlock; } /* * (SYN|ACK) is bogus on a listen socket. */ if (thflags & TH_ACK) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN|ACK invalid, segment rejected\n", s, __func__); syncache_badack(&inc, port); /* XXX: Not needed! */ TCPSTAT_INC(tcps_badsyn); rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } /* * If the drop_synfin option is enabled, drop all * segments with both the SYN and FIN bits set. * This prevents e.g. nmap from identifying the * TCP/IP stack. * XXX: Poor reasoning. nmap has other methods * and is constantly refining its stack detection * strategies. * XXX: This is a violation of the TCP specification * and was used by RFC1644. */ if ((thflags & TH_FIN) && V_drop_synfin) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "SYN|FIN segment ignored (based on " "sysctl setting)\n", s, __func__); TCPSTAT_INC(tcps_badsyn); goto dropunlock; } /* * Segment's flags are (SYN) or (SYN|FIN). * * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored * as they do not affect the state of the TCP FSM. * The data pointed to by TH_URG and th_urp is ignored. */ KASSERT((thflags & (TH_RST|TH_ACK)) == 0, ("%s: Listen socket: TH_RST or TH_ACK set", __func__)); KASSERT(thflags & (TH_SYN), ("%s: Listen socket: TH_SYN not set", __func__)); INP_RLOCK_ASSERT(inp); #ifdef INET6 /* * If deprecated address is forbidden, * we do not accept SYN to deprecated interface * address to prevent any new inbound connection from * getting established. * When we do not accept SYN, we send a TCP RST, * with deprecated source address (instead of dropping * it). We compromise it as it is much better for peer * to send a RST, and RST will be the final packet * for the exchange. * * If we do not forbid deprecated addresses, we accept * the SYN packet. RFC2462 does not suggest dropping * SYN in this case. * If we decipher RFC2462 5.5.4, it says like this: * 1. use of deprecated addr with existing * communication is okay - "SHOULD continue to be * used" * 2. use of it with new communication: * (2a) "SHOULD NOT be used if alternate address * with sufficient scope is available" * (2b) nothing mentioned otherwise. * Here we fall into (2b) case as we have no choice in * our source address selection - we must obey the peer. * * The wording in RFC2462 is confusing, and there are * multiple description text for deprecated address * handling - worse, they are not exactly the same. * I believe 5.5.4 is the best one, so we follow 5.5.4. */ if (isipv6 && !V_ip6_use_deprecated) { struct in6_ifaddr *ia6; ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false); if (ia6 != NULL && (ia6->ia6_flags & IN6_IFF_DEPRECATED)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt to deprecated " "IPv6 address rejected\n", s, __func__); rstreason = BANDLIM_RST_OPENPORT; goto dropwithreset; } } #endif /* INET6 */ /* * Basic sanity checks on incoming SYN requests: * Don't respond if the destination is a link layer * broadcast according to RFC1122 4.2.3.10, p. 104. * If it is from this socket it must be forged. * Don't respond if the source or destination is a * global or subnet broad- or multicast address. * Note that it is quite possible to receive unicast * link-layer packets with a broadcast IP address. Use * in_broadcast() to find them. */ if (m->m_flags & (M_BCAST|M_MCAST)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from broad- or multicast " "link layer address ignored\n", s, __func__); goto dropunlock; } #ifdef INET6 if (isipv6) { if (th->th_dport == th->th_sport && IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt to/from self " "ignored\n", s, __func__); goto dropunlock; } if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to multicast " "address ignored\n", s, __func__); goto dropunlock; } } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { if (th->th_dport == th->th_sport && ip->ip_dst.s_addr == ip->ip_src.s_addr) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to self " "ignored\n", s, __func__); goto dropunlock; } if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) { if ((s = tcp_log_addrs(&inc, th, NULL, NULL))) log(LOG_DEBUG, "%s; %s: Listen socket: " "Connection attempt from/to broad- " "or multicast address ignored\n", s, __func__); goto dropunlock; } } #endif /* * SYN appears to be valid. Create compressed TCP state * for syncache. */ TCP_PROBE3(debug__input, tp, th, m); tcp_dooptions(&to, optp, optlen, TO_SYN); if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL, iptos, port)) != NULL) goto tfo_socket_result; /* * Entry added to syncache and mbuf consumed. * Only the listen socket is unlocked by syncache_add(). */ return (IPPROTO_DONE); } else if (tp->t_state == TCPS_LISTEN) { /* * When a listen socket is torn down the SO_ACCEPTCONN * flag is removed first while connections are drained * from the accept queue in a unlock/lock cycle of the * ACCEPT_LOCK, opening a race condition allowing a SYN * attempt go through unhandled. */ goto dropunlock; } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (tp->t_flags & TF_SIGNATURE) { tcp_dooptions(&to, optp, optlen, thflags); if ((to.to_flags & TOF_SIGNATURE) == 0) { TCPSTAT_INC(tcps_sig_err_nosigopt); goto dropunlock; } if (!TCPMD5_ENABLED() || TCPMD5_INPUT(m, th, to.to_signature) != 0) goto dropunlock; } #endif TCP_PROBE5(receive, NULL, tp, m, tp, th); /* * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later * state. tcp_do_segment() always consumes the mbuf chain, unlocks * the inpcb, and unlocks pcbinfo. * * XXXGL: in case of a pure SYN arriving on existing connection * TCP stacks won't need to modify the PCB, they would either drop * the segment silently, or send a challenge ACK. However, we try * to upgrade the lock, because calling convention for stacks is * write-lock on PCB. If upgrade fails, drop the SYN. */ if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0) goto dropunlock; tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos); return (IPPROTO_DONE); dropwithreset: TCP_PROBE5(receive, NULL, tp, m, tp, th); if (inp != NULL) { tcp_dropwithreset(m, th, tp, tlen, rstreason); INP_UNLOCK(inp); } else tcp_dropwithreset(m, th, NULL, tlen, rstreason); m = NULL; /* mbuf chain got consumed. */ goto drop; dropunlock: if (m != NULL) TCP_PROBE5(receive, NULL, tp, m, tp, th); if (inp != NULL) INP_UNLOCK(inp); drop: if (s != NULL) free(s, M_TCPLOG); if (m != NULL) m_freem(m); return (IPPROTO_DONE); } /* * Automatic sizing of receive socket buffer. Often the send * buffer size is not optimally adjusted to the actual network * conditions at hand (delay bandwidth product). Setting the * buffer size too small limits throughput on links with high * bandwidth and high delay (eg. trans-continental/oceanic links). * * On the receive side the socket buffer memory is only rarely * used to any significant extent. This allows us to be much * more aggressive in scaling the receive socket buffer. For * the case that the buffer space is actually used to a large * extent and we run out of kernel memory we can simply drop * the new segments; TCP on the sender will just retransmit it * later. Setting the buffer size too big may only consume too * much kernel memory if the application doesn't read() from * the socket or packet loss or reordering makes use of the * reassembly queue. * * The criteria to step up the receive buffer one notch are: * 1. Application has not set receive buffer size with * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE. * 2. the number of bytes received during 1/2 of an sRTT * is at least 3/8 of the current socket buffer size. * 3. receive buffer size has not hit maximal automatic size; * * If all of the criteria are met we increaset the socket buffer * by a 1/2 (bounded by the max). This allows us to keep ahead * of slow-start but also makes it so our peer never gets limited * by our rwnd which we then open up causing a burst. * * This algorithm does two steps per RTT at most and only if * we receive a bulk stream w/o packet losses or reorderings. * Shrinking the buffer during idle times is not necessary as * it doesn't consume any memory when idle. * * TODO: Only step up if the application is actually serving * the buffer to better manage the socket buffer resources. */ int tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so, struct tcpcb *tp, int tlen) { int newsize = 0; if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) && tp->t_srtt != 0 && tp->rfbuf_ts != 0 && TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) > ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) { if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) && so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) { newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max); } TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize); /* Start over with next RTT. */ tp->rfbuf_ts = 0; tp->rfbuf_cnt = 0; } else { tp->rfbuf_cnt += tlen; /* add up */ } return (newsize); } int tcp_input(struct mbuf **mp, int *offp, int proto) { return(tcp_input_with_port(mp, offp, proto, 0)); } static void tcp_handle_wakeup(struct tcpcb *tp) { INP_WLOCK_ASSERT(tptoinpcb(tp)); if (tp->t_flags & TF_WAKESOR) { struct socket *so = tptosocket(tp); tp->t_flags &= ~TF_WAKESOR; SOCKBUF_LOCK_ASSERT(&so->so_rcv); sorwakeup_locked(so); } } void tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th, int drop_hdrlen, int tlen, uint8_t iptos) { uint16_t thflags; int acked, ourfinisacked, needoutput = 0; sackstatus_t sack_changed; int rstreason, todrop, win, incforsyn = 0; uint32_t tiwin; uint16_t nsegs; char *s; struct inpcb *inp = tptoinpcb(tp); struct socket *so = tptosocket(tp); struct in_conninfo *inc = &inp->inp_inc; struct mbuf *mfree; struct tcpopt to; int tfo_syn; u_int maxseg = 0; thflags = tcp_get_flags(th); tp->sackhint.last_sack_ack = 0; sack_changed = SACK_NOCHANGE; nsegs = max(1, m->m_pkthdr.lro_nsegs); NET_EPOCH_ASSERT(); INP_WLOCK_ASSERT(inp); KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN", __func__)); KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT", __func__)); #ifdef TCPPCAP /* Save segment, if requested. */ tcp_pcap_add(th, m, &(tp->t_inpkts)); #endif TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0, tlen, NULL, true); if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: " "SYN|FIN segment ignored (based on " "sysctl setting)\n", s, __func__); free(s, M_TCPLOG); } goto drop; } /* * If a segment with the ACK-bit set arrives in the SYN-SENT state * check SEQ.ACK first. */ if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) { rstreason = BANDLIM_UNLIMITED; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } /* * Segment received on connection. * Reset idle time and keep-alive timer. * XXX: This should be done after segment * validation to ignore broken/spoofed segs. */ if (tp->t_idle_reduce && (tp->snd_max == tp->snd_una) && ((ticks - tp->t_rcvtime) >= tp->t_rxtcur)) cc_after_idle(tp); tp->t_rcvtime = ticks; if (thflags & TH_FIN) tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN); /* * Scale up the window into a 32-bit value. * For the SYN_SENT state the scale is zero. */ tiwin = th->th_win << tp->snd_scale; #ifdef STATS stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin); #endif /* * TCP ECN processing. */ if (tcp_ecn_input_segment(tp, thflags, tlen, tcp_packets_this_ack(tp, th->th_ack), iptos)) cc_cong_signal(tp, th, CC_ECN); /* * Parse options on any incoming segment. */ tcp_dooptions(&to, (u_char *)(th + 1), (th->th_off << 2) - sizeof(struct tcphdr), (thflags & TH_SYN) ? TO_SYN : 0); #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if ((tp->t_flags & TF_SIGNATURE) != 0 && (to.to_flags & TOF_SIGNATURE) == 0) { TCPSTAT_INC(tcps_sig_err_sigopt); /* XXX: should drop? */ } #endif /* * If echoed timestamp is later than the current time, * fall back to non RFC1323 RTT calculation. Normalize * timestamp if syncookies were used when this connection * was established. */ if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) { to.to_tsecr -= tp->ts_offset; - if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) + if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) { to.to_tsecr = 0; - else if (tp->t_rxtshift == 1 && + } else if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && tp->t_badrxtwin != 0 && - TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) + TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) { cc_cong_signal(tp, th, CC_RTO_ERR); + } } /* * Process options only when we get SYN/ACK back. The SYN case * for incoming connections is handled in tcp_syncache. * According to RFC1323 the window field in a SYN (i.e., a * or ) segment itself is never scaled. * XXX this is traditional behavior, may need to be cleaned up. */ if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) { /* Handle parallel SYN for ECN */ tcp_ecn_input_parallel_syn(tp, thflags, iptos); if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE) && !(tp->t_flags & TF_NOOPT)) { tp->t_flags |= TF_RCVD_SCALE; tp->snd_scale = to.to_wscale; - } else + } else { tp->t_flags &= ~TF_REQ_SCALE; + } /* * Initial send window. It will be updated with * the next incoming segment to the scaled value. */ tp->snd_wnd = th->th_win; if ((to.to_flags & TOF_TS) && (tp->t_flags & TF_REQ_TSTMP) && !(tp->t_flags & TF_NOOPT)) { tp->t_flags |= TF_RCVD_TSTMP; tp->ts_recent = to.to_tsval; tp->ts_recent_age = tcp_ts_getticks(); - } else + } else { tp->t_flags &= ~TF_REQ_TSTMP; - if (to.to_flags & TOF_MSS) + } + if (to.to_flags & TOF_MSS) { tcp_mss(tp, to.to_mss); + } if ((tp->t_flags & TF_SACK_PERMIT) && (!(to.to_flags & TOF_SACKPERM) || - (tp->t_flags & TF_NOOPT))) + (tp->t_flags & TF_NOOPT))) { tp->t_flags &= ~TF_SACK_PERMIT; + } if (IS_FASTOPEN(tp->t_flags)) { if ((to.to_flags & TOF_FASTOPEN) && !(tp->t_flags & TF_NOOPT)) { uint16_t mss; - if (to.to_flags & TOF_MSS) + if (to.to_flags & TOF_MSS) { mss = to.to_mss; - else - if ((inp->inp_vflag & INP_IPV6) != 0) + } else { + if ((inp->inp_vflag & INP_IPV6) != 0) { mss = TCP6_MSS; - else + } else { mss = TCP_MSS; + } + } tcp_fastopen_update_cache(tp, mss, to.to_tfo_len, to.to_tfo_cookie); - } else + } else { tcp_fastopen_disable_path(tp); + } } } /* * If timestamps were negotiated during SYN/ACK and a * segment without a timestamp is received, silently drop * the segment, unless it is a RST segment or missing timestamps are * tolerated. * See section 3.2 of RFC 7323. */ if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) { if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp missing, " "segment processed normally\n", s, __func__); free(s, M_TCPLOG); } } else { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp missing, " "segment silently dropped\n", s, __func__); free(s, M_TCPLOG); } goto drop; } } /* * If timestamps were not negotiated during SYN/ACK and a * segment with a timestamp is received, ignore the * timestamp and process the packet normally. * See section 3.2 of RFC 7323. */ if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: Timestamp not expected, " "segment processed normally\n", s, __func__); free(s, M_TCPLOG); } } /* * Header prediction: check for the two common cases * of a uni-directional data xfer. If the packet has * no control flags, is in-sequence, the window didn't * change and we're not retransmitting, it's a * candidate. If the length is zero and the ack moved * forward, we're the sender side of the xfer. Just * free the data acked & wake any higher level process * that was blocked waiting for space. If the length * is non-zero and the ack didn't move, we're the * receiver side. If we're getting packets in-order * (the reassembly queue is empty), add the data to * the socket buffer and note that we need a delayed ack. * Make sure that the hidden state-flags are also off. * Since we check for TCPS_ESTABLISHED first, it can only * be TH_NEEDSYN. */ if (tp->t_state == TCPS_ESTABLISHED && th->th_seq == tp->rcv_nxt && (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK && tp->snd_nxt == tp->snd_max && tiwin && tiwin == tp->snd_wnd && ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) && SEGQ_EMPTY(tp) && ((to.to_flags & TOF_TS) == 0 || TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) { /* * If last ACK falls within this segment's sequence numbers, * record the timestamp. * NOTE that the test is modified according to the latest * proposal of the tcplw@cray.com list (Braden 1993/04/26). */ if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) { tp->ts_recent_age = tcp_ts_getticks(); tp->ts_recent = to.to_tsval; } if (tlen == 0) { if (SEQ_GT(th->th_ack, tp->snd_una) && SEQ_LEQ(th->th_ack, tp->snd_max) && !IN_RECOVERY(tp->t_flags) && (to.to_flags & TOF_SACK) == 0 && TAILQ_EMPTY(&tp->snd_holes)) { /* * This is a pure ack for outstanding data. */ TCPSTAT_INC(tcps_predack); /* * "bad retransmit" recovery without timestamps. */ if ((to.to_flags & TOF_TS) == 0 && tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && tp->t_badrxtwin != 0 && TSTMP_LT(ticks, tp->t_badrxtwin)) { cc_cong_signal(tp, th, CC_RTO_ERR); } /* * Recalculate the transmit timer / rtt. * * Some boxes send broken timestamp replies * during the SYN+ACK phase, ignore * timestamps of 0 or we could calculate a * huge RTT and blow up the retransmit timer. */ if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { uint32_t t; t = tcp_ts_getticks() - to.to_tsecr; if (!tp->t_rttlow || tp->t_rttlow > t) tp->t_rttlow = t; tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) tp->t_rttlow = ticks - tp->t_rtttime; tcp_xmit_timer(tp, ticks - tp->t_rtttime); } acked = BYTES_THIS_ACK(tp, th); #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ hhook_run_tcp_est_in(tp, th, &to); #endif TCPSTAT_ADD(tcps_rcvackpack, nsegs); TCPSTAT_ADD(tcps_rcvackbyte, acked); sbdrop(&so->so_snd, acked); if (SEQ_GT(tp->snd_una, tp->snd_recover) && SEQ_LEQ(th->th_ack, tp->snd_recover)) tp->snd_recover = th->th_ack - 1; /* * Let the congestion control algorithm update * congestion control related information. This * typically means increasing the congestion * window. */ cc_ack_received(tp, th, nsegs, CC_ACK); tp->snd_una = th->th_ack; /* * Pull snd_wl2 up to prevent seq wrap relative * to th_ack. */ tp->snd_wl2 = th->th_ack; tp->t_dupacks = 0; m_freem(m); /* * If all outstanding data are acked, stop * retransmit timer, otherwise restart timer * using current (possibly backed-off) value. * If process is waiting for space, * wakeup/selwakeup/signal. If data * are ready to send, let tcp_output * decide between more output or persist. */ TCP_PROBE3(debug__input, tp, th, m); /* * Clear t_acktime if remote side has ACKd * all data in the socket buffer. * Otherwise, update t_acktime if we received * a sufficiently large ACK. */ if (sbavail(&so->so_snd) == 0) tp->t_acktime = 0; else if (acked > 1) tp->t_acktime = ticks; if (tp->snd_una == tp->snd_max) tcp_timer_activate(tp, TT_REXMT, 0); else if (!tcp_timer_active(tp, TT_PERSIST)) tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); sowwakeup(so); /* * Only call tcp_output when there * is new data available to be sent * or we need to send an ACK. */ - if (SEQ_GT(tp->snd_una + sbavail(&so->so_snd), - tp->snd_max) || tp->t_flags & TF_ACKNOW) + if ((tp->t_flags & TF_ACKNOW) || + (sbavail(&so->so_snd) >= + SEQ_SUB(tp->snd_max, tp->snd_una))) { (void) tcp_output(tp); + } goto check_delack; } } else if (th->th_ack == tp->snd_una && tlen <= sbspace(&so->so_rcv)) { int newsize = 0; /* automatic sockbuf scaling */ /* * This is a pure, in-sequence data packet with * nothing on the reassembly queue and we have enough * buffer space to take it. */ /* Clean receiver SACK report if present */ if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks) tcp_clean_sackreport(tp); TCPSTAT_INC(tcps_preddat); tp->rcv_nxt += tlen; if (tlen && ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && (tp->t_fbyte_in == 0)) { tp->t_fbyte_in = ticks; if (tp->t_fbyte_in == 0) tp->t_fbyte_in = 1; if (tp->t_fbyte_out && tp->t_fbyte_in) tp->t_flags2 |= TF2_FBYTES_COMPLETE; } /* * Pull snd_wl1 up to prevent seq wrap relative to * th_seq. */ tp->snd_wl1 = th->th_seq; /* * Pull rcv_up up to prevent seq wrap relative to * rcv_nxt. */ tp->rcv_up = tp->rcv_nxt; TCPSTAT_ADD(tcps_rcvpack, nsegs); TCPSTAT_ADD(tcps_rcvbyte, tlen); TCP_PROBE3(debug__input, tp, th, m); newsize = tcp_autorcvbuf(m, th, so, tp, tlen); /* Add data to socket buffer. */ SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { m_freem(m); } else { /* * Set new socket buffer size. * Give up when limit is reached. */ if (newsize) if (!sbreserve_locked(so, SO_RCV, newsize, NULL)) so->so_rcv.sb_flags &= ~SB_AUTOSIZE; m_adj(m, drop_hdrlen); /* delayed header drop */ sbappendstream_locked(&so->so_rcv, m, 0); } /* NB: sorwakeup_locked() does an implicit unlock. */ sorwakeup_locked(so); if (DELAY_ACK(tp, tlen)) { tp->t_flags |= TF_DELACK; } else { tp->t_flags |= TF_ACKNOW; tcp_output(tp); } goto check_delack; } } /* * Calculate amount of space in receive window, * and then do TCP input processing. * Receive window is amount of space in rcv queue, * but not less than advertised window. */ win = sbspace(&so->so_rcv); if (win < 0) win = 0; tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt)); switch (tp->t_state) { /* * If the state is SYN_RECEIVED: * if seg contains an ACK, but not for our SYN/ACK, send a RST. */ case TCPS_SYN_RECEIVED: if (thflags & TH_RST) { /* Handle RST segments later. */ break; } if ((thflags & TH_ACK) && (SEQ_LEQ(th->th_ack, tp->snd_una) || SEQ_GT(th->th_ack, tp->snd_max))) { rstreason = BANDLIM_RST_OPENPORT; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } if (IS_FASTOPEN(tp->t_flags)) { /* * When a TFO connection is in SYN_RECEIVED, the * only valid packets are the initial SYN, a * retransmit/copy of the initial SYN (possibly with * a subset of the original data), a valid ACK, a * FIN, or a RST. */ if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) { rstreason = BANDLIM_RST_OPENPORT; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } else if (thflags & TH_SYN) { /* non-initial SYN is ignored */ if ((tcp_timer_active(tp, TT_DELACK) || tcp_timer_active(tp, TT_REXMT))) goto drop; } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) { goto drop; } } break; /* * If the state is SYN_SENT: * if seg contains a RST with valid ACK (SEQ.ACK has already * been verified), then drop the connection. * if seg contains a RST without an ACK, drop the seg. * if seg does not contain SYN, then drop the seg. * Otherwise this is an acceptable SYN segment * initialize tp->rcv_nxt and tp->irs * if seg contains ack then advance tp->snd_una * if seg contains an ECE and ECN support is enabled, the stream * is ECN capable. * if SYN has been acked change to ESTABLISHED else SYN_RCVD state * arrange for segment to be acked (eventually) * continue processing rest of data/controls, beginning with URG */ case TCPS_SYN_SENT: if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) { TCP_PROBE5(connect__refused, NULL, tp, m, tp, th); tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); tp = tcp_drop(tp, ECONNREFUSED); } if (thflags & TH_RST) goto drop; if (!(thflags & TH_SYN)) goto drop; tp->irs = th->th_seq; tcp_rcvseqinit(tp); if (thflags & TH_ACK) { int tfo_partial_ack = 0; TCPSTAT_INC(tcps_connects); soisconnected(so); #ifdef MAC mac_socketpeer_set_from_mbuf(m, so); #endif /* Do window scaling on this connection? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; } tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN << tp->rcv_scale); tp->snd_una++; /* SYN is acked */ if (SEQ_LT(tp->snd_nxt, tp->snd_una)) tp->snd_nxt = tp->snd_una; /* * If not all the data that was sent in the TFO SYN * has been acked, resend the remainder right away. */ if (IS_FASTOPEN(tp->t_flags) && (tp->snd_una != tp->snd_max)) { tp->snd_nxt = th->th_ack; tfo_partial_ack = 1; } /* * If there's data, delay ACK; if there's also a FIN * ACKNOW will be turned on later. */ if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack) tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); else tp->t_flags |= TF_ACKNOW; tcp_ecn_input_syn_sent(tp, thflags, iptos); /* * Received in SYN_SENT[*] state. * Transitions: * SYN_SENT --> ESTABLISHED * SYN_SENT* --> FIN_WAIT_1 */ tp->t_starttime = ticks; if (tp->t_flags & TF_NEEDFIN) { tp->t_acktime = ticks; tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; thflags &= ~TH_SYN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(connect__established, NULL, tp, m, tp, th); cc_conn_init(tp); tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); } } else { /* * Received initial SYN in SYN-SENT[*] state => * simultaneous open. * If it succeeds, connection is * half-synchronized. * Otherwise, do 3-way handshake: * SYN-SENT -> SYN-RECEIVED * SYN-SENT* -> SYN-RECEIVED* */ tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN); tcp_timer_activate(tp, TT_REXMT, 0); tcp_state_change(tp, TCPS_SYN_RECEIVED); } /* * Advance th->th_seq to correspond to first data byte. * If data, trim to stay within window, * dropping FIN if necessary. */ th->th_seq++; if (tlen > tp->rcv_wnd) { todrop = tlen - tp->rcv_wnd; m_adj(m, -todrop); tlen = tp->rcv_wnd; thflags &= ~TH_FIN; TCPSTAT_INC(tcps_rcvpackafterwin); TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); } tp->snd_wl1 = th->th_seq - 1; tp->rcv_up = th->th_seq; /* * Client side of transaction: already sent SYN and data. * If the remote host used T/TCP to validate the SYN, * our data will be ACK'd; if so, enter normal data segment * processing in the middle of step 5, ack processing. * Otherwise, goto step 6. */ if (thflags & TH_ACK) goto process_ACK; goto step6; } /* * States other than LISTEN or SYN_SENT. * First check the RST flag and sequence number since reset segments * are exempt from the timestamp and connection count tests. This * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix * below which allowed reset segments in half the sequence space * to fall though and be processed (which gives forged reset * segments with a random sequence number a 50 percent chance of * killing a connection). * Then check timestamp, if present. * Then check the connection count, if present. * Then check that at least some bytes of segment are within * receive window. If segment begins before rcv_nxt, * drop leading data (and SYN); if nothing left, just ack. */ if (thflags & TH_RST) { /* * RFC5961 Section 3.2 * * - RST drops connection only if SEG.SEQ == RCV.NXT. * - If RST is in window, we send challenge ACK. * * Note: to take into account delayed ACKs, we should * test against last_ack_sent instead of rcv_nxt. * Note 2: we handle special case of closed window, not * covered by the RFC. */ if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) && SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) || (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) { KASSERT(tp->t_state != TCPS_SYN_SENT, ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p", __func__, th, tp)); if (V_tcp_insecure_rst || tp->last_ack_sent == th->th_seq) { TCPSTAT_INC(tcps_drops); /* Drop the connection. */ switch (tp->t_state) { case TCPS_SYN_RECEIVED: so->so_error = ECONNREFUSED; goto close; case TCPS_ESTABLISHED: case TCPS_FIN_WAIT_1: case TCPS_FIN_WAIT_2: case TCPS_CLOSE_WAIT: case TCPS_CLOSING: case TCPS_LAST_ACK: so->so_error = ECONNRESET; close: /* FALLTHROUGH */ default: tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST); tp = tcp_close(tp); } } else { TCPSTAT_INC(tcps_badrst); /* Send challenge ACK. */ tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, tp->snd_nxt, TH_ACK); tp->last_ack_sent = tp->rcv_nxt; m = NULL; } } goto drop; } /* * RFC5961 Section 4.2 * Send challenge ACK for any SYN in synchronized state. */ if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT && tp->t_state != TCPS_SYN_RECEIVED) { TCPSTAT_INC(tcps_badsyn); if (V_tcp_insecure_syn && SEQ_GEQ(th->th_seq, tp->last_ack_sent) && SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) { tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); tp = tcp_drop(tp, ECONNRESET); rstreason = BANDLIM_UNLIMITED; } else { tcp_ecn_input_syn_sent(tp, thflags, iptos); /* Send challenge ACK. */ tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, tp->snd_nxt, TH_ACK); tp->last_ack_sent = tp->rcv_nxt; m = NULL; } goto drop; } /* * RFC 1323 PAWS: If we have a timestamp reply on this segment * and it's less than ts_recent, drop it. */ if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && TSTMP_LT(to.to_tsval, tp->ts_recent)) { /* Check to see if ts_recent is over 24 days old. */ if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) { /* * Invalidate ts_recent. If this segment updates * ts_recent, the age will be reset later and ts_recent * will get a valid value. If it does not, setting * ts_recent to zero will at least satisfy the * requirement that zero be placed in the timestamp * echo reply when ts_recent isn't valid. The * age isn't reset until we get a valid ts_recent * because we don't want out-of-order segments to be * dropped when ts_recent is old. */ tp->ts_recent = 0; } else { TCPSTAT_INC(tcps_rcvduppack); TCPSTAT_ADD(tcps_rcvdupbyte, tlen); TCPSTAT_INC(tcps_pawsdrop); if (tlen) goto dropafterack; goto drop; } } /* * In the SYN-RECEIVED state, validate that the packet belongs to * this connection before trimming the data to fit the receive * window. Check the sequence number versus IRS since we know * the sequence numbers haven't wrapped. This is a partial fix * for the "LAND" DoS attack. */ if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) { rstreason = BANDLIM_RST_OPENPORT; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } todrop = tp->rcv_nxt - th->th_seq; if (todrop > 0) { if (thflags & TH_SYN) { thflags &= ~TH_SYN; th->th_seq++; if (th->th_urp > 1) th->th_urp--; else thflags &= ~TH_URG; todrop--; } /* * Following if statement from Stevens, vol. 2, p. 960. */ if (todrop > tlen || (todrop == tlen && (thflags & TH_FIN) == 0)) { /* * Any valid FIN must be to the left of the window. * At this point the FIN must be a duplicate or out * of sequence; drop it. */ thflags &= ~TH_FIN; /* * Send an ACK to resynchronize and drop any data. * But keep on processing for RST or ACK. */ tp->t_flags |= TF_ACKNOW; todrop = tlen; TCPSTAT_INC(tcps_rcvduppack); TCPSTAT_ADD(tcps_rcvdupbyte, todrop); } else { TCPSTAT_INC(tcps_rcvpartduppack); TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop); } /* * DSACK - add SACK block for dropped range */ if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) { tcp_update_sack_list(tp, th->th_seq, th->th_seq + todrop); /* * ACK now, as the next in-sequence segment * will clear the DSACK block again */ tp->t_flags |= TF_ACKNOW; } drop_hdrlen += todrop; /* drop from the top afterwards */ th->th_seq += todrop; tlen -= todrop; if (th->th_urp > todrop) th->th_urp -= todrop; else { thflags &= ~TH_URG; th->th_urp = 0; } } /* * If new data are received on a connection after the * user processes are gone, then RST the other end. */ if ((tp->t_flags & TF_CLOSED) && tlen) { if ((s = tcp_log_addrs(inc, th, NULL, NULL))) { log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data " "after socket was closed, " "sending RST and removing tcpcb\n", s, __func__, tcpstates[tp->t_state], tlen); free(s, M_TCPLOG); } tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE); /* tcp_close will kill the inp pre-log the Reset */ tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST); tp = tcp_close(tp); TCPSTAT_INC(tcps_rcvafterclose); rstreason = BANDLIM_UNLIMITED; goto dropwithreset; } /* * If segment ends after window, drop trailing data * (and PUSH and FIN); if nothing left, just ACK. */ todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd); if (todrop > 0) { TCPSTAT_INC(tcps_rcvpackafterwin); if (todrop >= tlen) { TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen); /* * If window is closed can only take segments at * window edge, and have to drop data and PUSH from * incoming segments. Continue processing, but * remember to ack. Otherwise, drop segment * and ack. */ if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) { tp->t_flags |= TF_ACKNOW; TCPSTAT_INC(tcps_rcvwinprobe); } else goto dropafterack; } else TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop); m_adj(m, -todrop); tlen -= todrop; thflags &= ~(TH_PUSH|TH_FIN); } /* * If last ACK falls within this segment's sequence numbers, * record its timestamp. * NOTE: * 1) That the test incorporates suggestions from the latest * proposal of the tcplw@cray.com list (Braden 1993/04/26). * 2) That updating only on newer timestamps interferes with * our earlier PAWS tests, so this check should be solely * predicated on the sequence space of this segment. * 3) That we modify the segment boundary check to be * Last.ACK.Sent <= SEG.SEQ + SEG.Len * instead of RFC1323's * Last.ACK.Sent < SEG.SEQ + SEG.Len, * This modified check allows us to overcome RFC1323's * limitations as described in Stevens TCP/IP Illustrated * Vol. 2 p.869. In such cases, we can still calculate the * RTT correctly when RCV.NXT == Last.ACK.Sent. */ if ((to.to_flags & TOF_TS) != 0 && SEQ_LEQ(th->th_seq, tp->last_ack_sent) && SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen + ((thflags & (TH_SYN|TH_FIN)) != 0))) { tp->ts_recent_age = tcp_ts_getticks(); tp->ts_recent = to.to_tsval; } /* * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN * flag is on (half-synchronized state), then queue data for * later processing; else drop segment and return. */ if ((thflags & TH_ACK) == 0) { if (tp->t_state == TCPS_SYN_RECEIVED || (tp->t_flags & TF_NEEDSYN)) { if (tp->t_state == TCPS_SYN_RECEIVED && IS_FASTOPEN(tp->t_flags)) { tp->snd_wnd = tiwin; cc_conn_init(tp); } goto step6; } else if (tp->t_flags & TF_ACKNOW) goto dropafterack; else goto drop; } /* * Ack processing. */ switch (tp->t_state) { /* * In SYN_RECEIVED state, the ack ACKs our SYN, so enter * ESTABLISHED state and continue processing. * The ACK was checked above. */ case TCPS_SYN_RECEIVED: TCPSTAT_INC(tcps_connects); if (tp->t_flags & TF_SONOTCONN) { /* * Usually SYN_RECEIVED had been created from a LISTEN, * and solisten_enqueue() has already marked the socket * layer as connected. If it didn't, which can happen * only with an accept_filter(9), then the tp is marked * with TF_SONOTCONN. The other reason for this mark * to be set is a simultaneous open, a SYN_RECEIVED * that had been created from SYN_SENT. */ tp->t_flags &= ~TF_SONOTCONN; soisconnected(so); } /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; } tp->snd_wnd = tiwin; /* * Make transitions: * SYN-RECEIVED -> ESTABLISHED * SYN-RECEIVED* -> FIN-WAIT-1 */ tp->t_starttime = ticks; if (IS_FASTOPEN(tp->t_flags) && tp->t_tfo_pending) { tcp_fastopen_decrement_counter(tp->t_tfo_pending); tp->t_tfo_pending = NULL; } if (tp->t_flags & TF_NEEDFIN) { tp->t_acktime = ticks; tcp_state_change(tp, TCPS_FIN_WAIT_1); tp->t_flags &= ~TF_NEEDFIN; } else { tcp_state_change(tp, TCPS_ESTABLISHED); TCP_PROBE5(accept__established, NULL, tp, m, tp, th); /* * TFO connections call cc_conn_init() during SYN * processing. Calling it again here for such * connections is not harmless as it would undo the * snd_cwnd reduction that occurs when a TFO SYN|ACK * is retransmitted. */ if (!IS_FASTOPEN(tp->t_flags)) cc_conn_init(tp); tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp)); } /* * Account for the ACK of our SYN prior to * regular ACK processing below, except for * simultaneous SYN, which is handled later. */ if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN)) incforsyn = 1; /* * If segment contains data or ACK, will call tcp_reass() * later; if not, do so now to pass queued data to user. */ if (tlen == 0 && (thflags & TH_FIN) == 0) { (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0, (struct mbuf *)0); tcp_handle_wakeup(tp); } tp->snd_wl1 = th->th_seq - 1; /* FALLTHROUGH */ /* * In ESTABLISHED state: drop duplicate ACKs; ACK out of range * ACKs. If the ack is in the range * tp->snd_una < th->th_ack <= tp->snd_max * then advance tp->snd_una to th->th_ack and drop * data from the retransmission queue. If this ACK reflects * more up to date window information we update our window information. */ case TCPS_ESTABLISHED: case TCPS_FIN_WAIT_1: case TCPS_FIN_WAIT_2: case TCPS_CLOSE_WAIT: case TCPS_CLOSING: case TCPS_LAST_ACK: if (SEQ_GT(th->th_ack, tp->snd_max)) { TCPSTAT_INC(tcps_rcvacktoomuch); goto dropafterack; } if (tcp_is_sack_recovery(tp, &to)) { sack_changed = tcp_sack_doack(tp, &to, th->th_ack); if ((sack_changed != SACK_NOCHANGE) && (tp->t_flags & TF_LRD)) { tcp_sack_lost_retransmission(tp, th); } } else /* * Reset the value so that previous (valid) value * from the last ack with SACK doesn't get used. */ tp->sackhint.sacked_bytes = 0; #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */ hhook_run_tcp_est_in(tp, th, &to); #endif if (SEQ_LEQ(th->th_ack, tp->snd_una)) { maxseg = tcp_maxseg(tp); if (tlen == 0 && (tiwin == tp->snd_wnd || (tp->t_flags & TF_SACK_PERMIT))) { /* * If this is the first time we've seen a * FIN from the remote, this is not a * duplicate and it needs to be processed * normally. This happens during a * simultaneous close. */ if ((thflags & TH_FIN) && (TCPS_HAVERCVDFIN(tp->t_state) == 0)) { tp->t_dupacks = 0; break; } TCPSTAT_INC(tcps_rcvdupack); /* * If we have outstanding data (other than * a window probe), this is a completely * duplicate ack (ie, window info didn't * change and FIN isn't set), * the ack is the biggest we've * seen and we've seen exactly our rexmt * threshold of them, assume a packet * has been dropped and retransmit it. * Kludge snd_nxt & the congestion * window so we send only this one * packet. * * We know we're losing at the current * window size so do congestion avoidance * (set ssthresh to half the current window * and pull our congestion window back to * the new ssthresh). * * Dup acks mean that packets have left the * network (they're now cached at the receiver) * so bump cwnd by the amount in the receiver * to keep a constant cwnd packets in the * network. * * When using TCP ECN, notify the peer that * we reduced the cwnd. */ /* * Following 2 kinds of acks should not affect * dupack counting: * 1) Old acks * 2) Acks with SACK but without any new SACK * information in them. These could result from * any anomaly in the network like a switch * duplicating packets or a possible DoS attack. */ if (th->th_ack != tp->snd_una || (tcp_is_sack_recovery(tp, &to) && - (sack_changed == SACK_NOCHANGE))) + (sack_changed == SACK_NOCHANGE))) { break; - else if (!tcp_timer_active(tp, TT_REXMT)) + } else if (!tcp_timer_active(tp, TT_REXMT)) { tp->t_dupacks = 0; - else if (++tp->t_dupacks > tcprexmtthresh || - IN_FASTRECOVERY(tp->t_flags)) { + } else if (++tp->t_dupacks > tcprexmtthresh || + IN_FASTRECOVERY(tp->t_flags)) { cc_ack_received(tp, th, nsegs, CC_DUPACK); if (V_tcp_do_prr && IN_FASTRECOVERY(tp->t_flags) && (tp->t_flags & TF_SACK_PERMIT)) { tcp_do_prr_ack(tp, th, &to, sack_changed, &maxseg); } else if (tcp_is_sack_recovery(tp, &to) && - IN_FASTRECOVERY(tp->t_flags)) { + IN_FASTRECOVERY(tp->t_flags)) { int awnd; /* * Compute the amount of data in flight first. * We can inject new data into the pipe iff * we have less than 1/2 the original window's * worth of data in flight. */ - if (V_tcp_do_newsack) + if (V_tcp_do_newsack) { awnd = tcp_compute_pipe(tp); - else + } else { awnd = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; - + } if (awnd < tp->snd_ssthresh) { tp->snd_cwnd += maxseg; if (tp->snd_cwnd > tp->snd_ssthresh) tp->snd_cwnd = tp->snd_ssthresh; } - } else + } else { tp->snd_cwnd += maxseg; + } (void) tcp_output(tp); goto drop; } else if (tp->t_dupacks == tcprexmtthresh || (tp->t_flags & TF_SACK_PERMIT && V_tcp_do_newsack && tp->sackhint.sacked_bytes > (tcprexmtthresh - 1) * maxseg)) { enter_recovery: /* * Above is the RFC6675 trigger condition of * more than (dupthresh-1)*maxseg sacked data. * If the count of holes in the * scoreboard is >= dupthresh, we could * also enter loss recovery, but don't * have that value readily available. */ tp->t_dupacks = tcprexmtthresh; tcp_seq onxt = tp->snd_nxt; /* * If we're doing sack, or prr, check * to see if we're already in sack * recovery. If we're not doing sack, * check to see if we're in newreno * recovery. */ if (V_tcp_do_prr || (tp->t_flags & TF_SACK_PERMIT)) { if (IN_FASTRECOVERY(tp->t_flags)) { tp->t_dupacks = 0; break; } } else { if (SEQ_LEQ(th->th_ack, tp->snd_recover)) { tp->t_dupacks = 0; break; } } /* Congestion signal before ack. */ cc_cong_signal(tp, th, CC_NDUPACK); cc_ack_received(tp, th, nsegs, CC_DUPACK); tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; if (V_tcp_do_prr) { /* * snd_ssthresh is already updated by * cc_cong_signal. */ if (tcp_is_sack_recovery(tp, &to)) { /* * Exclude Limited Transmit * segments here */ tp->sackhint.prr_delivered = maxseg; } else { tp->sackhint.prr_delivered = imin(tp->snd_max - tp->snd_una, imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); } tp->sackhint.recover_fs = max(1, tp->snd_nxt - tp->snd_una); } if (tcp_is_sack_recovery(tp, &to)) { - TCPSTAT_INC( - tcps_sack_recovery_episode); + TCPSTAT_INC(tcps_sack_recovery_episode); tp->snd_recover = tp->snd_nxt; tp->snd_cwnd = maxseg; (void) tcp_output(tp); - if (SEQ_GT(th->th_ack, tp->snd_una)) + if (SEQ_GT(th->th_ack, tp->snd_una)) { goto resume_partialack; + } goto drop; } tp->snd_nxt = th->th_ack; tp->snd_cwnd = maxseg; (void) tcp_output(tp); KASSERT(tp->snd_limited <= 2, ("%s: tp->snd_limited too big", __func__)); tp->snd_cwnd = tp->snd_ssthresh + maxseg * (tp->t_dupacks - tp->snd_limited); if (SEQ_GT(onxt, tp->snd_nxt)) tp->snd_nxt = onxt; goto drop; } else if (V_tcp_do_rfc3042) { /* * Process first and second duplicate * ACKs. Each indicates a segment * leaving the network, creating room * for more. Make sure we can send a * packet on reception of each duplicate * ACK by increasing snd_cwnd by one * segment. Restore the original * snd_cwnd after packet transmission. */ - cc_ack_received(tp, th, nsegs, - CC_DUPACK); + cc_ack_received(tp, th, nsegs, CC_DUPACK); uint32_t oldcwnd = tp->snd_cwnd; tcp_seq oldsndmax = tp->snd_max; u_int sent; int avail; KASSERT(tp->t_dupacks == 1 || tp->t_dupacks == 2, ("%s: dupacks not 1 or 2", __func__)); if (tp->t_dupacks == 1) tp->snd_limited = 0; tp->snd_cwnd = (tp->snd_nxt - tp->snd_una) + (tp->t_dupacks - tp->snd_limited) * maxseg; /* * Only call tcp_output when there * is new data available to be sent * or we need to send an ACK. */ SOCKBUF_LOCK(&so->so_snd); - avail = sbavail(&so->so_snd) - - (tp->snd_nxt - tp->snd_una); + avail = sbavail(&so->so_snd); SOCKBUF_UNLOCK(&so->so_snd); - if (avail > 0 || tp->t_flags & TF_ACKNOW) + if (tp->t_flags & TF_ACKNOW || + (avail >= + SEQ_SUB(tp->snd_nxt, tp->snd_una))) { (void) tcp_output(tp); - sent = tp->snd_max - oldsndmax; + } + sent = SEQ_SUB(tp->snd_max, oldsndmax); if (sent > maxseg) { KASSERT((tp->t_dupacks == 2 && tp->snd_limited == 0) || (sent == maxseg + 1 && tp->t_flags & TF_SENTFIN), ("%s: sent too much", __func__)); tp->snd_limited = 2; - } else if (sent > 0) + } else if (sent > 0) { ++tp->snd_limited; + } tp->snd_cwnd = oldcwnd; goto drop; } } break; } else { /* * This ack is advancing the left edge, reset the * counter. */ tp->t_dupacks = 0; /* * If this ack also has new SACK info, increment the * counter as per rfc6675. The variable * sack_changed tracks all changes to the SACK * scoreboard, including when partial ACKs without * SACK options are received, and clear the scoreboard * from the left side. Such partial ACKs should not be * counted as dupacks here. */ if (tcp_is_sack_recovery(tp, &to) && (sack_changed != SACK_NOCHANGE)) { tp->t_dupacks++; /* limit overhead by setting maxseg last */ if (!IN_FASTRECOVERY(tp->t_flags) && (tp->sackhint.sacked_bytes > ((tcprexmtthresh - 1) * (maxseg = tcp_maxseg(tp))))) { goto enter_recovery; } } } resume_partialack: KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("%s: th_ack <= snd_una", __func__)); /* * If the congestion window was inflated to account * for the other side's cached packets, retract it. */ if (IN_FASTRECOVERY(tp->t_flags)) { if (SEQ_LT(th->th_ack, tp->snd_recover)) { if (tp->t_flags & TF_SACK_PERMIT) { if (V_tcp_do_prr && to.to_flags & TOF_SACK) { tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; tcp_do_prr_ack(tp, th, &to, sack_changed, &maxseg); tp->t_flags |= TF_ACKNOW; (void) tcp_output(tp); } else { tcp_sack_partialack(tp, th, &maxseg); } } else { tcp_newreno_partial_ack(tp, th); } } else { cc_post_recovery(tp, th); } } else if (IN_CONGRECOVERY(tp->t_flags)) { if (SEQ_LT(th->th_ack, tp->snd_recover)) { if (V_tcp_do_prr) { tp->sackhint.delivered_data = BYTES_THIS_ACK(tp, th); tp->snd_fack = th->th_ack; /* * During ECN cwnd reduction * always use PRR-SSRB */ tcp_do_prr_ack(tp, th, &to, SACK_CHANGE, &maxseg); (void) tcp_output(tp); } } else { cc_post_recovery(tp, th); } } /* * If we reach this point, ACK is not a duplicate, * i.e., it ACKs something we sent. */ if (tp->t_flags & TF_NEEDSYN) { /* * T/TCP: Connection was half-synchronized, and our * SYN has been ACK'd (so connection is now fully * synchronized). Go to non-starred state, * increment snd_una for ACK of SYN, and check if * we can do window scaling. */ tp->t_flags &= ~TF_NEEDSYN; tp->snd_una++; /* Do window scaling? */ if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) == (TF_RCVD_SCALE|TF_REQ_SCALE)) { tp->rcv_scale = tp->request_r_scale; /* Send window already scaled. */ } } process_ACK: INP_WLOCK_ASSERT(inp); /* * Adjust for the SYN bit in sequence space, * but don't account for it in cwnd calculations. * This is for the SYN_RECEIVED, non-simultaneous * SYN case. SYN_SENT and simultaneous SYN are * treated elsewhere. */ if (incforsyn) tp->snd_una++; acked = BYTES_THIS_ACK(tp, th); KASSERT(acked >= 0, ("%s: acked unexepectedly negative " "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__, tp->snd_una, th->th_ack, tp, m)); TCPSTAT_ADD(tcps_rcvackpack, nsegs); TCPSTAT_ADD(tcps_rcvackbyte, acked); /* * If we just performed our first retransmit, and the ACK * arrives within our recovery window, then it was a mistake * to do the retransmit in the first place. Recover our * original cwnd and ssthresh, and proceed to transmit where * we left off. */ if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID && tp->t_badrxtwin != 0 && to.to_flags & TOF_TS && to.to_tsecr != 0 && TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) cc_cong_signal(tp, th, CC_RTO_ERR); /* * If we have a timestamp reply, update smoothed * round trip time. If no timestamp is present but * transmit timer is running and timed sequence * number was acked, update smoothed round trip time. * Since we now have an rtt measurement, cancel the * timer backoff (cf., Phil Karn's retransmit alg.). * Recompute the initial retransmit timer. * * Some boxes send broken timestamp replies * during the SYN+ACK phase, ignore * timestamps of 0 or we could calculate a * huge RTT and blow up the retransmit timer. */ if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) { uint32_t t; t = tcp_ts_getticks() - to.to_tsecr; if (!tp->t_rttlow || tp->t_rttlow > t) tp->t_rttlow = t; tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1); } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) { if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime) tp->t_rttlow = ticks - tp->t_rtttime; tcp_xmit_timer(tp, ticks - tp->t_rtttime); } SOCKBUF_LOCK(&so->so_snd); /* * Clear t_acktime if remote side has ACKd all data in the * socket buffer and FIN (if applicable). * Otherwise, update t_acktime if we received a sufficiently * large ACK. */ if ((tp->t_state <= TCPS_CLOSE_WAIT && acked == sbavail(&so->so_snd)) || acked > sbavail(&so->so_snd)) tp->t_acktime = 0; else if (acked > 1) tp->t_acktime = ticks; /* * If all outstanding data is acked, stop retransmit * timer and remember to restart (more output or persist). * If there is more data to be acked, restart retransmit * timer, using current (possibly backed-off) value. */ if (th->th_ack == tp->snd_max) { tcp_timer_activate(tp, TT_REXMT, 0); needoutput = 1; } else if (!tcp_timer_active(tp, TT_PERSIST)) tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); /* * If no data (only SYN) was ACK'd, * skip rest of ACK processing. */ if (acked == 0) { SOCKBUF_UNLOCK(&so->so_snd); goto step6; } /* * Let the congestion control algorithm update congestion * control related information. This typically means increasing * the congestion window. */ cc_ack_received(tp, th, nsegs, CC_ACK); if (acked > sbavail(&so->so_snd)) { if (tp->snd_wnd >= sbavail(&so->so_snd)) tp->snd_wnd -= sbavail(&so->so_snd); else tp->snd_wnd = 0; mfree = sbcut_locked(&so->so_snd, (int)sbavail(&so->so_snd)); ourfinisacked = 1; } else { mfree = sbcut_locked(&so->so_snd, acked); if (tp->snd_wnd >= (uint32_t) acked) tp->snd_wnd -= acked; else tp->snd_wnd = 0; ourfinisacked = 0; } /* NB: sowwakeup_locked() does an implicit unlock. */ sowwakeup_locked(so); m_freem(mfree); /* Detect una wraparound. */ if (!IN_RECOVERY(tp->t_flags) && SEQ_GT(tp->snd_una, tp->snd_recover) && SEQ_LEQ(th->th_ack, tp->snd_recover)) tp->snd_recover = th->th_ack - 1; /* XXXLAS: Can this be moved up into cc_post_recovery? */ if (IN_RECOVERY(tp->t_flags) && SEQ_GEQ(th->th_ack, tp->snd_recover)) { EXIT_RECOVERY(tp->t_flags); } tp->snd_una = th->th_ack; if (tp->t_flags & TF_SACK_PERMIT) { if (SEQ_GT(tp->snd_una, tp->snd_recover)) tp->snd_recover = tp->snd_una; } if (SEQ_LT(tp->snd_nxt, tp->snd_una)) tp->snd_nxt = tp->snd_una; switch (tp->t_state) { /* * In FIN_WAIT_1 STATE in addition to the processing * for the ESTABLISHED state if our FIN is now acknowledged * then enter FIN_WAIT_2. */ case TCPS_FIN_WAIT_1: if (ourfinisacked) { /* * If we can't receive any more * data, then closing user can proceed. * Starting the timer is contrary to the * specification, but if we don't get a FIN * we'll hang forever. */ if (so->so_rcv.sb_state & SBS_CANTRCVMORE) { soisdisconnected(so); tcp_timer_activate(tp, TT_2MSL, (tcp_fast_finwait2_recycle ? tcp_finwait2_timeout : TP_MAXIDLE(tp))); } tcp_state_change(tp, TCPS_FIN_WAIT_2); } break; /* * In CLOSING STATE in addition to the processing for * the ESTABLISHED state if the ACK acknowledges our FIN * then enter the TIME-WAIT state, otherwise ignore * the segment. */ case TCPS_CLOSING: if (ourfinisacked) { tcp_twstart(tp); m_freem(m); return; } break; /* * In LAST_ACK, we may still be waiting for data to drain * and/or to be acked, as well as for the ack of our FIN. * If our FIN is now acknowledged, delete the TCB, * enter the closed state and return. */ case TCPS_LAST_ACK: if (ourfinisacked) { tp = tcp_close(tp); goto drop; } break; } } step6: INP_WLOCK_ASSERT(inp); /* * Update window information. * Don't look at window if no ACK: TAC's send garbage on first SYN. */ if ((thflags & TH_ACK) && (SEQ_LT(tp->snd_wl1, th->th_seq) || (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) || (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) { /* keep track of pure window updates */ if (tlen == 0 && tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd) TCPSTAT_INC(tcps_rcvwinupd); tp->snd_wnd = tiwin; tp->snd_wl1 = th->th_seq; tp->snd_wl2 = th->th_ack; if (tp->snd_wnd > tp->max_sndwnd) tp->max_sndwnd = tp->snd_wnd; needoutput = 1; } /* * Process segments with URG. */ if ((thflags & TH_URG) && th->th_urp && TCPS_HAVERCVDFIN(tp->t_state) == 0) { /* * This is a kludge, but if we receive and accept * random urgent pointers, we'll crash in * soreceive. It's hard to imagine someone * actually wanting to send this much urgent data. */ SOCKBUF_LOCK(&so->so_rcv); if (th->th_urp + sbavail(&so->so_rcv) > sb_max) { th->th_urp = 0; /* XXX */ thflags &= ~TH_URG; /* XXX */ SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */ goto dodata; /* XXX */ } /* * If this segment advances the known urgent pointer, * then mark the data stream. This should not happen * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since * a FIN has been received from the remote side. * In these states we ignore the URG. * * According to RFC961 (Assigned Protocols), * the urgent pointer points to the last octet * of urgent data. We continue, however, * to consider it to indicate the first octet * of data past the urgent section as the original * spec states (in one of two places). */ if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) { tp->rcv_up = th->th_seq + th->th_urp; so->so_oobmark = sbavail(&so->so_rcv) + (tp->rcv_up - tp->rcv_nxt) - 1; if (so->so_oobmark == 0) so->so_rcv.sb_state |= SBS_RCVATMARK; sohasoutofband(so); tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA); } SOCKBUF_UNLOCK(&so->so_rcv); /* * Remove out of band data so doesn't get presented to user. * This can happen independent of advancing the URG pointer, * but if two URG's are pending at once, some out-of-band * data may creep in... ick. */ if (th->th_urp <= (uint32_t)tlen && !(so->so_options & SO_OOBINLINE)) { /* hdr drop is delayed */ tcp_pulloutofband(so, th, m, drop_hdrlen); } } else { /* * If no out of band data is expected, * pull receive urgent pointer along * with the receive window. */ if (SEQ_GT(tp->rcv_nxt, tp->rcv_up)) tp->rcv_up = tp->rcv_nxt; } dodata: /* XXX */ INP_WLOCK_ASSERT(inp); /* * Process the segment text, merging it into the TCP sequencing queue, * and arranging for acknowledgment of receipt if necessary. * This process logically involves adjusting tp->rcv_wnd as data * is presented to the user (this happens in tcp_usrreq.c, * case PRU_RCVD). If a FIN has already been received on this * connection then we just ignore the text. */ tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) && IS_FASTOPEN(tp->t_flags)); if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) && TCPS_HAVERCVDFIN(tp->t_state) == 0) { tcp_seq save_start = th->th_seq; tcp_seq save_rnxt = tp->rcv_nxt; int save_tlen = tlen; m_adj(m, drop_hdrlen); /* delayed header drop */ /* * Insert segment which includes th into TCP reassembly queue * with control block tp. Set thflags to whether reassembly now * includes a segment with FIN. This handles the common case * inline (segment is the next to be received on an established * connection, and the queue is empty), avoiding linkage into * and removal from the queue and repetition of various * conversions. * Set DELACK for segments received in order, but ack * immediately when segments are out of order (so * fast retransmit can work). */ if (th->th_seq == tp->rcv_nxt && SEGQ_EMPTY(tp) && (TCPS_HAVEESTABLISHED(tp->t_state) || tfo_syn)) { if (DELAY_ACK(tp, tlen) || tfo_syn) tp->t_flags |= TF_DELACK; else tp->t_flags |= TF_ACKNOW; tp->rcv_nxt += tlen; if (tlen && ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) && (tp->t_fbyte_in == 0)) { tp->t_fbyte_in = ticks; if (tp->t_fbyte_in == 0) tp->t_fbyte_in = 1; if (tp->t_fbyte_out && tp->t_fbyte_in) tp->t_flags2 |= TF2_FBYTES_COMPLETE; } thflags = tcp_get_flags(th) & TH_FIN; TCPSTAT_INC(tcps_rcvpack); TCPSTAT_ADD(tcps_rcvbyte, tlen); SOCKBUF_LOCK(&so->so_rcv); if (so->so_rcv.sb_state & SBS_CANTRCVMORE) m_freem(m); else sbappendstream_locked(&so->so_rcv, m, 0); tp->t_flags |= TF_WAKESOR; } else { /* * XXX: Due to the header drop above "th" is * theoretically invalid by now. Fortunately * m_adj() doesn't actually frees any mbufs * when trimming from the head. */ tcp_seq temp = save_start; thflags = tcp_reass(tp, th, &temp, &tlen, m); tp->t_flags |= TF_ACKNOW; } if ((tp->t_flags & TF_SACK_PERMIT) && (save_tlen > 0) && TCPS_HAVEESTABLISHED(tp->t_state)) { if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) { /* * DSACK actually handled in the fastpath * above. */ tcp_update_sack_list(tp, save_start, save_start + save_tlen); } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) { if ((tp->rcv_numsacks >= 1) && (tp->sackblks[0].end == save_start)) { /* * Partial overlap, recorded at todrop * above. */ tcp_update_sack_list(tp, tp->sackblks[0].start, tp->sackblks[0].end); } else { tcp_update_dsack_list(tp, save_start, save_start + save_tlen); } } else if (tlen >= save_tlen) { /* Update of sackblks. */ tcp_update_dsack_list(tp, save_start, save_start + save_tlen); } else if (tlen > 0) { tcp_update_dsack_list(tp, save_start, save_start + tlen); } } tcp_handle_wakeup(tp); #if 0 /* * Note the amount of data that peer has sent into * our window, in order to estimate the sender's * buffer size. * XXX: Unused. */ if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt); else len = so->so_rcv.sb_hiwat; #endif } else { m_freem(m); thflags &= ~TH_FIN; } /* * If FIN is received ACK the FIN and let the user know * that the connection is closing. */ if (thflags & TH_FIN) { if (TCPS_HAVERCVDFIN(tp->t_state) == 0) { /* The socket upcall is handled by socantrcvmore. */ socantrcvmore(so); /* * If connection is half-synchronized * (ie NEEDSYN flag on) then delay ACK, * so it may be piggybacked when SYN is sent. * Otherwise, since we received a FIN then no * more input can be expected, send ACK now. */ if (tp->t_flags & TF_NEEDSYN) tp->t_flags |= TF_DELACK; else tp->t_flags |= TF_ACKNOW; tp->rcv_nxt++; } switch (tp->t_state) { /* * In SYN_RECEIVED and ESTABLISHED STATES * enter the CLOSE_WAIT state. */ case TCPS_SYN_RECEIVED: tp->t_starttime = ticks; /* FALLTHROUGH */ case TCPS_ESTABLISHED: tcp_state_change(tp, TCPS_CLOSE_WAIT); break; /* * If still in FIN_WAIT_1 STATE FIN has not been acked so * enter the CLOSING state. */ case TCPS_FIN_WAIT_1: tcp_state_change(tp, TCPS_CLOSING); break; /* * In FIN_WAIT_2 state enter the TIME_WAIT state, * starting the time-wait timer, turning off the other * standard timers. */ case TCPS_FIN_WAIT_2: tcp_twstart(tp); return; } } TCP_PROBE3(debug__input, tp, th, m); /* * Return any desired output. */ - if (needoutput || (tp->t_flags & TF_ACKNOW)) + if (needoutput || (tp->t_flags & TF_ACKNOW)) { (void) tcp_output(tp); - + } check_delack: INP_WLOCK_ASSERT(inp); if (tp->t_flags & TF_DELACK) { tp->t_flags &= ~TF_DELACK; tcp_timer_activate(tp, TT_DELACK, tcp_delacktime); } INP_WUNLOCK(inp); return; dropafterack: /* * Generate an ACK dropping incoming segment if it occupies * sequence space, where the ACK reflects our state. * * We can now skip the test for the RST flag since all * paths to this code happen after packets containing * RST have been dropped. * * In the SYN-RECEIVED state, don't send an ACK unless the * segment we received passes the SYN-RECEIVED ACK test. * If it fails send a RST. This breaks the loop in the * "LAND" DoS attack, and also prevents an ACK storm * between two listening ports that have been sent forged * SYN segments, each with the source address of the other. */ if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) && (SEQ_GT(tp->snd_una, th->th_ack) || SEQ_GT(th->th_ack, tp->snd_max)) ) { rstreason = BANDLIM_RST_OPENPORT; tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT); goto dropwithreset; } TCP_PROBE3(debug__input, tp, th, m); tp->t_flags |= TF_ACKNOW; (void) tcp_output(tp); INP_WUNLOCK(inp); m_freem(m); return; dropwithreset: if (tp != NULL) { tcp_dropwithreset(m, th, tp, tlen, rstreason); INP_WUNLOCK(inp); } else tcp_dropwithreset(m, th, NULL, tlen, rstreason); return; drop: /* * Drop space held by incoming segment and return. */ TCP_PROBE3(debug__input, tp, th, m); if (tp != NULL) { INP_WUNLOCK(inp); } m_freem(m); } /* * Issue RST and make ACK acceptable to originator of segment. * The mbuf must still include the original packet header. * tp may be NULL. */ void tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp, int tlen, int rstreason) { #ifdef INET struct ip *ip; #endif #ifdef INET6 struct ip6_hdr *ip6; #endif if (tp != NULL) { INP_LOCK_ASSERT(tptoinpcb(tp)); } /* Don't bother if destination was broadcast/multicast. */ if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST)) goto drop; #ifdef INET6 if (mtod(m, struct ip *)->ip_v == 6) { ip6 = mtod(m, struct ip6_hdr *); if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) || IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) goto drop; /* IPv6 anycast check is done at tcp6_input() */ } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip = mtod(m, struct ip *); if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) || IN_MULTICAST(ntohl(ip->ip_src.s_addr)) || ip->ip_src.s_addr == htonl(INADDR_BROADCAST) || in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) goto drop; } #endif /* Perform bandwidth limiting. */ if (badport_bandlim(rstreason) < 0) goto drop; /* tcp_respond consumes the mbuf chain. */ if (tcp_get_flags(th) & TH_ACK) { tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack, TH_RST); } else { if (tcp_get_flags(th) & TH_SYN) tlen++; if (tcp_get_flags(th) & TH_FIN) tlen++; tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK); } return; drop: m_freem(m); } /* * Parse TCP options and place in tcpopt. */ void tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags) { int opt, optlen; to->to_flags = 0; for (; cnt > 0; cnt -= optlen, cp += optlen) { opt = cp[0]; if (opt == TCPOPT_EOL) break; if (opt == TCPOPT_NOP) optlen = 1; else { if (cnt < 2) break; optlen = cp[1]; if (optlen < 2 || optlen > cnt) break; } switch (opt) { case TCPOPT_MAXSEG: if (optlen != TCPOLEN_MAXSEG) continue; if (!(flags & TO_SYN)) continue; to->to_flags |= TOF_MSS; bcopy((char *)cp + 2, (char *)&to->to_mss, sizeof(to->to_mss)); to->to_mss = ntohs(to->to_mss); break; case TCPOPT_WINDOW: if (optlen != TCPOLEN_WINDOW) continue; if (!(flags & TO_SYN)) continue; to->to_flags |= TOF_SCALE; to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT); break; case TCPOPT_TIMESTAMP: if (optlen != TCPOLEN_TIMESTAMP) continue; to->to_flags |= TOF_TS; bcopy((char *)cp + 2, (char *)&to->to_tsval, sizeof(to->to_tsval)); to->to_tsval = ntohl(to->to_tsval); bcopy((char *)cp + 6, (char *)&to->to_tsecr, sizeof(to->to_tsecr)); to->to_tsecr = ntohl(to->to_tsecr); break; case TCPOPT_SIGNATURE: /* * In order to reply to a host which has set the * TCP_SIGNATURE option in its initial SYN, we have * to record the fact that the option was observed * here for the syncache code to perform the correct * response. */ if (optlen != TCPOLEN_SIGNATURE) continue; to->to_flags |= TOF_SIGNATURE; to->to_signature = cp + 2; break; case TCPOPT_SACK_PERMITTED: if (optlen != TCPOLEN_SACK_PERMITTED) continue; if (!(flags & TO_SYN)) continue; if (!V_tcp_do_sack) continue; to->to_flags |= TOF_SACKPERM; break; case TCPOPT_SACK: if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0) continue; if (flags & TO_SYN) continue; to->to_flags |= TOF_SACK; to->to_nsacks = (optlen - 2) / TCPOLEN_SACK; to->to_sacks = cp + 2; TCPSTAT_INC(tcps_sack_rcv_blocks); break; case TCPOPT_FAST_OPEN: /* * Cookie length validation is performed by the * server side cookie checking code or the client * side cookie cache update code. */ if (!(flags & TO_SYN)) continue; if (!V_tcp_fastopen_client_enable && !V_tcp_fastopen_server_enable) continue; to->to_flags |= TOF_FASTOPEN; to->to_tfo_len = optlen - 2; to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL; break; default: continue; } } } /* * Pull out of band byte out of a segment so * it doesn't appear in the user's data queue. * It is still reflected in the segment length for * sequencing purposes. */ void tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off) { int cnt = off + th->th_urp - 1; while (cnt >= 0) { if (m->m_len > cnt) { char *cp = mtod(m, caddr_t) + cnt; struct tcpcb *tp = sototcpcb(so); INP_WLOCK_ASSERT(tptoinpcb(tp)); tp->t_iobc = *cp; tp->t_oobflags |= TCPOOB_HAVEDATA; bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1)); m->m_len--; if (m->m_flags & M_PKTHDR) m->m_pkthdr.len--; return; } cnt -= m->m_len; m = m->m_next; if (m == NULL) break; } panic("tcp_pulloutofband"); } /* * Collect new round-trip time estimate * and update averages and current timeout. */ void tcp_xmit_timer(struct tcpcb *tp, int rtt) { int delta; INP_WLOCK_ASSERT(tptoinpcb(tp)); TCPSTAT_INC(tcps_rttupdated); if (tp->t_rttupdated < UCHAR_MAX) tp->t_rttupdated++; #ifdef STATS stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT, imax(0, rtt * 1000 / hz)); #endif if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) { /* * srtt is stored as fixed point with 5 bits after the * binary point (i.e., scaled by 8). The following magic * is equivalent to the smoothing algorithm in rfc793 with * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed * point). Adjust rtt to origin 0. */ delta = ((rtt - 1) << TCP_DELTA_SHIFT) - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT)); if ((tp->t_srtt += delta) <= 0) tp->t_srtt = 1; /* * We accumulate a smoothed rtt variance (actually, a * smoothed mean difference), then set the retransmit * timer to smoothed rtt + 4 times the smoothed variance. * rttvar is stored as fixed point with 4 bits after the * binary point (scaled by 16). The following is * equivalent to rfc793 smoothing with an alpha of .75 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces * rfc793's wired-in beta. */ if (delta < 0) delta = -delta; delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT); if ((tp->t_rttvar += delta) <= 0) tp->t_rttvar = 1; } else { /* * No rtt measurement yet - use the unsmoothed rtt. * Set the variance to half the rtt (so our first * retransmit happens at 3*rtt). */ tp->t_srtt = rtt << TCP_RTT_SHIFT; tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1); } tp->t_rtttime = 0; tp->t_rxtshift = 0; /* * the retransmit should happen at rtt + 4 * rttvar. * Because of the way we do the smoothing, srtt and rttvar * will each average +1/2 tick of bias. When we compute * the retransmit timer, we want 1/2 tick of rounding and * 1 extra tick because of +-1/2 tick uncertainty in the * firing of the timer. The bias will give us exactly the * 1.5 tick we need. But, because the bias is * statistical, we have to test that we don't drop below * the minimum feasible timer (which is 2 ticks). */ TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp), max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX); /* * We received an ack for a packet that wasn't retransmitted; * it is probably safe to discard any error indications we've * received recently. This isn't quite right, but close enough * for now (a route might have failed after we sent a segment, * and the return path might not be symmetrical). */ tp->t_softerror = 0; } /* * Determine a reasonable value for maxseg size. * If the route is known, check route for mtu. * If none, use an mss that can be handled on the outgoing interface * without forcing IP to fragment. If no route is found, route has no mtu, * or the destination isn't local, use a default, hopefully conservative * size (usually 512 or the default IP max size, but no more than the mtu * of the interface), as we can't discover anything about intervening * gateways or networks. We also initialize the congestion/slow start * window to be a single segment if the destination isn't local. * While looking at the routing entry, we also initialize other path-dependent * parameters from pre-set or cached values in the routing entry. * * NOTE that resulting t_maxseg doesn't include space for TCP options or * IP options, e.g. IPSEC data, since length of this data may vary, and * thus it is calculated for every segment separately in tcp_output(). * * NOTE that this routine is only called when we process an incoming * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS * settings are handled in tcp_mssopt(). */ void tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer, struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap) { int mss = 0; uint32_t maxmtu = 0; struct inpcb *inp = tptoinpcb(tp); struct hc_metrics_lite metrics; #ifdef INET6 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0; size_t min_protoh = isipv6 ? sizeof (struct ip6_hdr) + sizeof (struct tcphdr) : sizeof (struct tcpiphdr); #else size_t min_protoh = sizeof(struct tcpiphdr); #endif INP_WLOCK_ASSERT(inp); if (tp->t_port) min_protoh += V_tcp_udp_tunneling_overhead; if (mtuoffer != -1) { KASSERT(offer == -1, ("%s: conflict", __func__)); offer = mtuoffer - min_protoh; } /* Initialize. */ #ifdef INET6 if (isipv6) { maxmtu = tcp_maxmtu6(&inp->inp_inc, cap); tp->t_maxseg = V_tcp_v6mssdflt; } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { maxmtu = tcp_maxmtu(&inp->inp_inc, cap); tp->t_maxseg = V_tcp_mssdflt; } #endif /* * No route to sender, stay with default mss and return. */ if (maxmtu == 0) { /* * In case we return early we need to initialize metrics * to a defined state as tcp_hc_get() would do for us * if there was no cache hit. */ if (metricptr != NULL) bzero(metricptr, sizeof(struct hc_metrics_lite)); return; } /* What have we got? */ switch (offer) { case 0: /* * Offer == 0 means that there was no MSS on the SYN * segment, in this case we use tcp_mssdflt as * already assigned to t_maxseg above. */ offer = tp->t_maxseg; break; case -1: /* * Offer == -1 means that we didn't receive SYN yet. */ /* FALLTHROUGH */ default: /* * Prevent DoS attack with too small MSS. Round up * to at least minmss. */ offer = max(offer, V_tcp_minmss); } /* * rmx information is now retrieved from tcp_hostcache. */ tcp_hc_get(&inp->inp_inc, &metrics); if (metricptr != NULL) bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite)); /* * If there's a discovered mtu in tcp hostcache, use it. * Else, use the link mtu. */ if (metrics.rmx_mtu) mss = min(metrics.rmx_mtu, maxmtu) - min_protoh; else { #ifdef INET6 if (isipv6) { mss = maxmtu - min_protoh; if (!V_path_mtu_discovery && !in6_localaddr(&inp->in6p_faddr)) mss = min(mss, V_tcp_v6mssdflt); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { mss = maxmtu - min_protoh; if (!V_path_mtu_discovery && !in_localaddr(inp->inp_faddr)) mss = min(mss, V_tcp_mssdflt); } #endif /* * XXX - The above conditional (mss = maxmtu - min_protoh) * probably violates the TCP spec. * The problem is that, since we don't know the * other end's MSS, we are supposed to use a conservative * default. But, if we do that, then MTU discovery will * never actually take place, because the conservative * default is much less than the MTUs typically seen * on the Internet today. For the moment, we'll sweep * this under the carpet. * * The conservative default might not actually be a problem * if the only case this occurs is when sending an initial * SYN with options and data to a host we've never talked * to before. Then, they will reply with an MSS value which * will get recorded and the new parameters should get * recomputed. For Further Study. */ } mss = min(mss, offer); /* * Sanity check: make sure that maxseg will be large * enough to allow some data on segments even if the * all the option space is used (40bytes). Otherwise * funny things may happen in tcp_output. * * XXXGL: shouldn't we reserve space for IP/IPv6 options? */ mss = max(mss, 64); tp->t_maxseg = mss; } void tcp_mss(struct tcpcb *tp, int offer) { int mss; uint32_t bufsize; struct inpcb *inp = tptoinpcb(tp); struct socket *so; struct hc_metrics_lite metrics; struct tcp_ifcap cap; KASSERT(tp != NULL, ("%s: tp == NULL", __func__)); bzero(&cap, sizeof(cap)); tcp_mss_update(tp, offer, -1, &metrics, &cap); mss = tp->t_maxseg; /* * If there's a pipesize, change the socket buffer to that size, * don't change if sb_hiwat is different than default (then it * has been changed on purpose with setsockopt). * Make the socket buffers an integral number of mss units; * if the mss is larger than the socket buffer, decrease the mss. */ so = inp->inp_socket; SOCKBUF_LOCK(&so->so_snd); if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe) bufsize = metrics.rmx_sendpipe; else bufsize = so->so_snd.sb_hiwat; if (bufsize < mss) mss = bufsize; else { bufsize = roundup(bufsize, mss); if (bufsize > sb_max) bufsize = sb_max; if (bufsize > so->so_snd.sb_hiwat) (void)sbreserve_locked(so, SO_SND, bufsize, NULL); } SOCKBUF_UNLOCK(&so->so_snd); /* * Sanity check: make sure that maxseg will be large * enough to allow some data on segments even if the * all the option space is used (40bytes). Otherwise * funny things may happen in tcp_output. * * XXXGL: shouldn't we reserve space for IP/IPv6 options? */ tp->t_maxseg = max(mss, 64); SOCKBUF_LOCK(&so->so_rcv); if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe) bufsize = metrics.rmx_recvpipe; else bufsize = so->so_rcv.sb_hiwat; if (bufsize > mss) { bufsize = roundup(bufsize, mss); if (bufsize > sb_max) bufsize = sb_max; if (bufsize > so->so_rcv.sb_hiwat) (void)sbreserve_locked(so, SO_RCV, bufsize, NULL); } SOCKBUF_UNLOCK(&so->so_rcv); /* Check the interface for TSO capabilities. */ if (cap.ifcap & CSUM_TSO) { tp->t_flags |= TF_TSO; tp->t_tsomax = cap.tsomax; tp->t_tsomaxsegcount = cap.tsomaxsegcount; tp->t_tsomaxsegsize = cap.tsomaxsegsize; } } /* * Determine the MSS option to send on an outgoing SYN. */ int tcp_mssopt(struct in_conninfo *inc) { int mss = 0; uint32_t thcmtu = 0; uint32_t maxmtu = 0; size_t min_protoh; KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer")); #ifdef INET6 if (inc->inc_flags & INC_ISIPV6) { mss = V_tcp_v6mssdflt; maxmtu = tcp_maxmtu6(inc, NULL); min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET { mss = V_tcp_mssdflt; maxmtu = tcp_maxmtu(inc, NULL); min_protoh = sizeof(struct tcpiphdr); } #endif #if defined(INET6) || defined(INET) thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */ #endif if (maxmtu && thcmtu) mss = min(maxmtu, thcmtu) - min_protoh; else if (maxmtu || thcmtu) mss = max(maxmtu, thcmtu) - min_protoh; return (mss); } void tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, sackstatus_t sack_changed, u_int *maxsegp) { int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0; u_int maxseg; INP_WLOCK_ASSERT(tptoinpcb(tp)); if (*maxsegp == 0) { *maxsegp = tcp_maxseg(tp); } maxseg = *maxsegp; /* * Compute the amount of data that this ACK is indicating * (del_data) and an estimate of how many bytes are in the * network. */ if (tcp_is_sack_recovery(tp, to) || (IN_CONGRECOVERY(tp->t_flags) && !IN_FASTRECOVERY(tp->t_flags))) { del_data = tp->sackhint.delivered_data; if (V_tcp_do_newsack) pipe = tcp_compute_pipe(tp); else pipe = (tp->snd_nxt - tp->snd_fack) + tp->sackhint.sack_bytes_rexmit; } else { if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg + - tp->snd_recover - tp->snd_una)) + tp->snd_recover - tp->snd_una)) { del_data = maxseg; + } pipe = imax(0, tp->snd_max - tp->snd_una - imin(INT_MAX / 65536, tp->t_dupacks) * maxseg); } tp->sackhint.prr_delivered += del_data; /* * Proportional Rate Reduction */ if (pipe >= tp->snd_ssthresh) { if (tp->sackhint.recover_fs == 0) tp->sackhint.recover_fs = imax(1, tp->snd_nxt - tp->snd_una); snd_cnt = howmany((long)tp->sackhint.prr_delivered * tp->snd_ssthresh, tp->sackhint.recover_fs) - tp->sackhint.prr_out + maxseg - 1; } else { /* * PRR 6937bis heuristic: * - A partial ack without SACK block beneath snd_recover * indicates further loss. * - An SACK scoreboard update adding a new hole indicates * further loss, so be conservative and send at most one * segment. * - Prevent ACK splitting attacks, by being conservative * when no new data is acked. */ - if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) + if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) { limit = tp->sackhint.prr_delivered - tp->sackhint.prr_out; - else + } else { limit = imax(tp->sackhint.prr_delivered - tp->sackhint.prr_out, del_data) + maxseg; + } snd_cnt = imin((tp->snd_ssthresh - pipe), limit); } snd_cnt = imax(snd_cnt, 0) / maxseg; /* * Send snd_cnt new data into the network in response to this ack. * If there is going to be a SACK retransmission, adjust snd_cwnd * accordingly. */ if (IN_FASTRECOVERY(tp->t_flags)) { if (tcp_is_sack_recovery(tp, to)) { tp->snd_cwnd = tp->snd_nxt - tp->snd_recover + tp->sackhint.sack_bytes_rexmit + (snd_cnt * maxseg); } else { tp->snd_cwnd = (tp->snd_max - tp->snd_una) + (snd_cnt * maxseg); } - } else if (IN_CONGRECOVERY(tp->t_flags)) + } else if (IN_CONGRECOVERY(tp->t_flags)) { tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg); + } tp->snd_cwnd = imax(maxseg, tp->snd_cwnd); } /* * On a partial ack arrives, force the retransmission of the * next unacknowledged segment. Do not clear tp->t_dupacks. * By setting snd_nxt to ti_ack, this forces retransmission timer to * be started again. */ void tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th) { tcp_seq onxt = tp->snd_nxt; uint32_t ocwnd = tp->snd_cwnd; u_int maxseg = tcp_maxseg(tp); INP_WLOCK_ASSERT(tptoinpcb(tp)); tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; tp->snd_nxt = th->th_ack; /* * Set snd_cwnd to one segment beyond acknowledged offset. * (tp->snd_una has not yet been updated when this function is called.) */ tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th); tp->t_flags |= TF_ACKNOW; (void) tcp_output(tp); tp->snd_cwnd = ocwnd; if (SEQ_GT(onxt, tp->snd_nxt)) tp->snd_nxt = onxt; /* * Partial window deflation. Relies on fact that tp->snd_una * not updated yet. */ if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th)) tp->snd_cwnd -= BYTES_THIS_ACK(tp, th); else tp->snd_cwnd = 0; tp->snd_cwnd += maxseg; } int tcp_compute_pipe(struct tcpcb *tp) { if (tp->t_fb->tfb_compute_pipe == NULL) { return (tp->snd_max - tp->snd_una + tp->sackhint.sack_bytes_rexmit - tp->sackhint.sacked_bytes - tp->sackhint.lost_bytes); } else { return((*tp->t_fb->tfb_compute_pipe)(tp)); } } uint32_t tcp_compute_initwnd(uint32_t maxseg) { /* * Calculate the Initial Window, also used as Restart Window * * RFC5681 Section 3.1 specifies the default conservative values. * RFC3390 specifies slightly more aggressive values. * RFC6928 increases it to ten segments. * Support for user specified value for initial flight size. */ if (V_tcp_initcwnd_segments) return min(V_tcp_initcwnd_segments * maxseg, max(2 * maxseg, V_tcp_initcwnd_segments * 1460)); else if (V_tcp_do_rfc3390) return min(4 * maxseg, max(2 * maxseg, 4380)); else { /* Per RFC5681 Section 3.1 */ if (maxseg > 2190) return (2 * maxseg); else if (maxseg > 1095) return (3 * maxseg); else return (4 * maxseg); } } diff --git a/sys/netinet/tcp_output.c b/sys/netinet/tcp_output.c index 58f63b593b2a..50dc05e9c55a 100644 --- a/sys/netinet/tcp_output.c +++ b/sys/netinet/tcp_output.c @@ -1,2164 +1,2160 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include "opt_inet.h" #include "opt_inet6.h" #include "opt_ipsec.h" #include "opt_kern_tls.h" #include #include #include #include #ifdef TCP_HHOOK #include #endif #include #ifdef KERN_TLS #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef INET6 #include #include #include #endif #include #define TCPOUTFLAGS #include #include #include #include #include #include #include #include #include #ifdef TCPPCAP #include #endif #ifdef TCP_OFFLOAD #include #endif #include #include #include #include #include #include VNET_DEFINE(int, path_mtu_discovery) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, path_mtu_discovery, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(path_mtu_discovery), 1, "Enable Path MTU Discovery"); VNET_DEFINE(int, tcp_do_tso) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, tso, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_tso), 0, "Enable TCP Segmentation Offload"); VNET_DEFINE(int, tcp_sendspace) = 1024*32; #define V_tcp_sendspace VNET(tcp_sendspace) SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sendspace), 0, "Initial send socket buffer size"); VNET_DEFINE(int, tcp_do_autosndbuf) = 1; SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_autosndbuf), 0, "Enable automatic send buffer sizing"); VNET_DEFINE(int, tcp_autosndbuf_inc) = 8*1024; SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_inc, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autosndbuf_inc), 0, "Incrementor step size of automatic send buffer"); VNET_DEFINE(int, tcp_autosndbuf_max) = 2*1024*1024; SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_max, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_autosndbuf_max), 0, "Max size of automatic send buffer"); VNET_DEFINE(int, tcp_sendbuf_auto_lowat) = 0; #define V_tcp_sendbuf_auto_lowat VNET(tcp_sendbuf_auto_lowat) SYSCTL_INT(_net_inet_tcp, OID_AUTO, sendbuf_auto_lowat, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sendbuf_auto_lowat), 0, "Modify threshold for auto send buffer growth to account for SO_SNDLOWAT"); /* * Make sure that either retransmit or persist timer is set for SYN, FIN and * non-ACK. */ #define TCP_XMIT_TIMER_ASSERT(tp, len, th_flags) \ KASSERT(((len) == 0 && ((th_flags) & (TH_SYN | TH_FIN)) == 0) ||\ tcp_timer_active((tp), TT_REXMT) || \ tcp_timer_active((tp), TT_PERSIST), \ ("neither rexmt nor persist timer is set")) #ifdef TCP_HHOOK /* * Wrapper for the TCP established output helper hook. */ void hhook_run_tcp_est_out(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to, uint32_t len, int tso) { struct tcp_hhook_data hhook_data; if (V_tcp_hhh[HHOOK_TCP_EST_OUT]->hhh_nhooks > 0) { hhook_data.tp = tp; hhook_data.th = th; hhook_data.to = to; hhook_data.len = len; hhook_data.tso = tso; hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_OUT], &hhook_data, &tp->t_osd); } } #endif /* * CC wrapper hook functions */ void cc_after_idle(struct tcpcb *tp) { INP_WLOCK_ASSERT(tptoinpcb(tp)); if (CC_ALGO(tp)->after_idle != NULL) CC_ALGO(tp)->after_idle(&tp->t_ccv); } /* * Tcp output routine: figure out what should be sent and send it. */ int tcp_default_output(struct tcpcb *tp) { struct socket *so = tptosocket(tp); struct inpcb *inp = tptoinpcb(tp); int32_t len; uint32_t recwin, sendwin; uint16_t flags; int off, error = 0; /* Keep compiler happy */ u_int if_hw_tsomaxsegcount = 0; u_int if_hw_tsomaxsegsize = 0; struct mbuf *m; struct ip *ip = NULL; struct tcphdr *th; u_char opt[TCP_MAXOLEN]; unsigned ipoptlen, optlen, hdrlen, ulen; #if defined(IPSEC) || defined(IPSEC_SUPPORT) unsigned ipsec_optlen = 0; #endif int idle, sendalot, curticks; int sack_rxmit, sack_bytes_rxmt; struct sackhole *p; int tso, mtu; struct tcpopt to; struct udphdr *udp = NULL; struct tcp_log_buffer *lgb; unsigned int wanted_cookie = 0; unsigned int dont_sendalot = 0; #if 0 int maxburst = TCP_MAXBURST; #endif #ifdef INET6 struct ip6_hdr *ip6 = NULL; const bool isipv6 = (inp->inp_vflag & INP_IPV6) != 0; #endif #ifdef KERN_TLS const bool hw_tls = tp->t_nic_ktls_xmit != 0; #else const bool hw_tls = false; #endif NET_EPOCH_ASSERT(); INP_WLOCK_ASSERT(inp); #ifdef TCP_OFFLOAD if (tp->t_flags & TF_TOE) return (tcp_offload_output(tp)); #endif /* * For TFO connections in SYN_SENT or SYN_RECEIVED, * only allow the initial SYN or SYN|ACK and those sent * by the retransmit timer. */ if (IS_FASTOPEN(tp->t_flags) && ((tp->t_state == TCPS_SYN_SENT) || (tp->t_state == TCPS_SYN_RECEIVED)) && SEQ_GT(tp->snd_max, tp->snd_una) && /* initial SYN or SYN|ACK sent */ (tp->snd_nxt != tp->snd_una)) /* not a retransmit */ return (0); /* * Determine length of data that should be transmitted, * and flags that will be used. * If there is some data or critical controls (SYN, RST) * to send, then transmit; otherwise, investigate further. */ idle = (tp->t_flags & TF_LASTIDLE) || (tp->snd_max == tp->snd_una); if (idle && (((ticks - tp->t_rcvtime) >= tp->t_rxtcur) || (tp->t_sndtime && ((ticks - tp->t_sndtime) >= tp->t_rxtcur)))) cc_after_idle(tp); tp->t_flags &= ~TF_LASTIDLE; if (idle) { if (tp->t_flags & TF_MORETOCOME) { tp->t_flags |= TF_LASTIDLE; idle = 0; } } again: /* * If we've recently taken a timeout, snd_max will be greater than * snd_nxt. There may be SACK information that allows us to avoid * resending already delivered data. Adjust snd_nxt accordingly. */ if ((tp->t_flags & TF_SACK_PERMIT) && SEQ_LT(tp->snd_nxt, tp->snd_max)) tcp_sack_adjust(tp); sendalot = 0; tso = 0; mtu = 0; off = tp->snd_nxt - tp->snd_una; sendwin = min(tp->snd_wnd, tp->snd_cwnd); flags = tcp_outflags[tp->t_state]; /* * Send any SACK-generated retransmissions. If we're explicitly trying * to send out new data (when sendalot is 1), bypass this function. * If we retransmit in fast recovery mode, decrement snd_cwnd, since * we're replacing a (future) new transmission with a retransmission * now, and we previously incremented snd_cwnd in tcp_input(). */ /* * Still in sack recovery , reset rxmit flag to zero. */ sack_rxmit = 0; sack_bytes_rxmt = 0; len = 0; p = NULL; if ((tp->t_flags & TF_SACK_PERMIT) && IN_FASTRECOVERY(tp->t_flags) && (p = tcp_sack_output(tp, &sack_bytes_rxmt))) { uint32_t cwin; cwin = imax(min(tp->snd_wnd, tp->snd_cwnd) - sack_bytes_rxmt, 0); /* Do not retransmit SACK segments beyond snd_recover */ if (SEQ_GT(p->end, tp->snd_recover)) { /* * (At least) part of sack hole extends beyond * snd_recover. Check to see if we can rexmit data * for this hole. */ if (SEQ_GEQ(p->rxmit, tp->snd_recover)) { /* * Can't rexmit any more data for this hole. * That data will be rexmitted in the next * sack recovery episode, when snd_recover * moves past p->rxmit. */ p = NULL; goto after_sack_rexmit; } else { /* Can rexmit part of the current hole */ len = ((int32_t)ulmin(cwin, SEQ_SUB(tp->snd_recover, p->rxmit))); } } else { len = ((int32_t)ulmin(cwin, SEQ_SUB(p->end, p->rxmit))); } if (len > 0) { off = SEQ_SUB(p->rxmit, tp->snd_una); KASSERT(off >= 0,("%s: sack block to the left of una : %d", __func__, off)); sack_rxmit = 1; sendalot = 1; } } after_sack_rexmit: /* * Get standard flags, and add SYN or FIN if requested by 'hidden' * state flags. */ if (tp->t_flags & TF_NEEDFIN) flags |= TH_FIN; if (tp->t_flags & TF_NEEDSYN) flags |= TH_SYN; SOCKBUF_LOCK(&so->so_snd); /* * If in persist timeout with window of 0, send 1 byte. * Otherwise, if window is small but nonzero * and timer expired, we will send what we can * and go to transmit state. */ if (tp->t_flags & TF_FORCEDATA) { if (sendwin == 0) { /* * If we still have some data to send, then * clear the FIN bit. Usually this would * happen below when it realizes that we * aren't sending all the data. However, * if we have exactly 1 byte of unsent data, * then it won't clear the FIN bit below, * and if we are in persist state, we wind * up sending the packet without recording * that we sent the FIN bit. * * We can't just blindly clear the FIN bit, * because if we don't have any more data * to send then the probe will be the FIN * itself. */ if (off < sbused(&so->so_snd)) flags &= ~TH_FIN; sendwin = 1; } else { tcp_timer_activate(tp, TT_PERSIST, 0); tp->t_rxtshift = 0; } } /* * If snd_nxt == snd_max and we have transmitted a FIN, the * offset will be > 0 even if so_snd.sb_cc is 0, resulting in * a negative length. This can also occur when TCP opens up * its congestion window while receiving additional duplicate * acks after fast-retransmit because TCP will reset snd_nxt * to snd_max after the fast-retransmit. * * In the normal retransmit-FIN-only case, however, snd_nxt will * be set to snd_una, the offset will be 0, and the length may * wind up 0. * * If sack_rxmit is true we are retransmitting from the scoreboard * in which case len is already set. */ if (sack_rxmit == 0) { - if (sack_bytes_rxmt == 0) + if (sack_bytes_rxmt == 0) { len = ((int32_t)min(sbavail(&so->so_snd), sendwin) - off); - else { + } else { int32_t cwin; /* * We are inside of a SACK recovery episode and are * sending new data, having retransmitted all the * data possible in the scoreboard. */ len = ((int32_t)min(sbavail(&so->so_snd), tp->snd_wnd) - off); /* * Don't remove this (len > 0) check ! * We explicitly check for len > 0 here (although it * isn't really necessary), to work around a gcc * optimization issue - to force gcc to compute * len above. Without this check, the computation * of len is bungled by the optimizer. */ if (len > 0) { cwin = tp->snd_cwnd - imax(0, (int32_t) (tp->snd_nxt - tp->snd_recover)) - sack_bytes_rxmt; if (cwin < 0) cwin = 0; len = imin(len, cwin); } } } /* * Lop off SYN bit if it has already been sent. However, if this * is SYN-SENT state and if segment contains data and if we don't * know that foreign host supports TAO, suppress sending segment. */ if ((flags & TH_SYN) && SEQ_GT(tp->snd_nxt, tp->snd_una)) { if (tp->t_state != TCPS_SYN_RECEIVED) flags &= ~TH_SYN; /* * When sending additional segments following a TFO SYN|ACK, * do not include the SYN bit. */ if (IS_FASTOPEN(tp->t_flags) && (tp->t_state == TCPS_SYN_RECEIVED)) flags &= ~TH_SYN; off--, len++; } /* * Be careful not to send data and/or FIN on SYN segments. * This measure is needed to prevent interoperability problems * with not fully conformant TCP implementations. */ if ((flags & TH_SYN) && (tp->t_flags & TF_NOOPT)) { len = 0; flags &= ~TH_FIN; } /* * On TFO sockets, ensure no data is sent in the following cases: * * - When retransmitting SYN|ACK on a passively-created socket * * - When retransmitting SYN on an actively created socket * * - When sending a zero-length cookie (cookie request) on an * actively created socket * * - When the socket is in the CLOSED state (RST is being sent) */ if (IS_FASTOPEN(tp->t_flags) && (((flags & TH_SYN) && (tp->t_rxtshift > 0)) || ((tp->t_state == TCPS_SYN_SENT) && (tp->t_tfo_client_cookie_len == 0)) || (flags & TH_RST))) len = 0; if (len <= 0) { /* * If FIN has been sent but not acked, * but we haven't been called to retransmit, * len will be < 0. Otherwise, window shrank * after we sent into it. If window shrank to 0, * cancel pending retransmit, pull snd_nxt back * to (closed) window, and set the persist timer * if it isn't already going. If the window didn't * close completely, just wait for an ACK. * * We also do a general check here to ensure that * we will set the persist timer when we have data * to send, but a 0-byte window. This makes sure * the persist timer is set even if the packet * hits one of the "goto send" lines below. */ len = 0; if ((sendwin == 0) && (TCPS_HAVEESTABLISHED(tp->t_state)) && (off < (int) sbavail(&so->so_snd)) && !tcp_timer_active(tp, TT_PERSIST)) { tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rxtshift = 0; tp->snd_nxt = tp->snd_una; if (!tcp_timer_active(tp, TT_PERSIST)) tcp_setpersist(tp); } } /* len will be >= 0 after this point. */ KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); tcp_sndbuf_autoscale(tp, so, sendwin); /* * Decide if we can use TCP Segmentation Offloading (if supported by * hardware). * * TSO may only be used if we are in a pure bulk sending state. The * presence of TCP-MD5, SACK retransmits, SACK advertizements and * IP options prevent using TSO. With TSO the TCP header is the same * (except for the sequence number) for all generated packets. This * makes it impossible to transmit any options which vary per generated * segment or packet. * * IPv4 handling has a clear separation of ip options and ip header * flags while IPv6 combines both in in6p_outputopts. ip6_optlen() does * the right thing below to provide length of just ip options and thus * checking for ipoptlen is enough to decide if ip options are present. */ #if defined(IPSEC) || defined(IPSEC_SUPPORT) /* * Pre-calculate here as we save another lookup into the darknesses * of IPsec that way and can actually decide if TSO is ok. */ #ifdef INET6 if (isipv6 && IPSEC_ENABLED(ipv6)) ipsec_optlen = IPSEC_HDRSIZE(ipv6, inp); #ifdef INET else #endif #endif /* INET6 */ #ifdef INET if (IPSEC_ENABLED(ipv4)) ipsec_optlen = IPSEC_HDRSIZE(ipv4, inp); #endif /* INET */ #endif /* IPSEC */ #ifdef INET6 if (isipv6) ipoptlen = ip6_optlen(inp); else #endif if (inp->inp_options) ipoptlen = inp->inp_options->m_len - offsetof(struct ipoption, ipopt_list); else ipoptlen = 0; #if defined(IPSEC) || defined(IPSEC_SUPPORT) ipoptlen += ipsec_optlen; #endif if ((tp->t_flags & TF_TSO) && V_tcp_do_tso && len > tp->t_maxseg && (tp->t_port == 0) && ((tp->t_flags & TF_SIGNATURE) == 0) && tp->rcv_numsacks == 0 && sack_rxmit == 0 && ipoptlen == 0 && !(flags & TH_SYN)) tso = 1; - if (sack_rxmit) { - if (SEQ_LT(p->rxmit + len, tp->snd_una + sbused(&so->so_snd))) - flags &= ~TH_FIN; - } else { - if (SEQ_LT(tp->snd_nxt + len, tp->snd_una + - sbused(&so->so_snd))) + if (SEQ_LT((sack_rxmit ? p->rxmit : tp->snd_nxt) + len, + tp->snd_una + sbused(&so->so_snd))) { flags &= ~TH_FIN; } recwin = lmin(lmax(sbspace(&so->so_rcv), 0), (long)TCP_MAXWIN << tp->rcv_scale); /* * Sender silly window avoidance. We transmit under the following * conditions when len is non-zero: * * - We have a full segment (or more with TSO) * - This is the last buffer in a write()/send() and we are * either idle or running NODELAY * - we've timed out (e.g. persist timer) * - we have more then 1/2 the maximum send window's worth of * data (receiver may be limited the window size) * - we need to retransmit */ if (len) { if (len >= tp->t_maxseg) goto send; /* * As the TCP header options are now * considered when setting up the initial * window, we would not send the last segment * if we skip considering the option length here. * Note: this may not work when tcp headers change * very dynamically in the future. */ if ((((tp->t_flags & TF_SIGNATURE) ? PADTCPOLEN(TCPOLEN_SIGNATURE) : 0) + ((tp->t_flags & TF_RCVD_TSTMP) ? PADTCPOLEN(TCPOLEN_TIMESTAMP) : 0) + len) >= tp->t_maxseg) goto send; /* * NOTE! on localhost connections an 'ack' from the remote * end may occur synchronously with the output and cause * us to flush a buffer queued with moretocome. XXX * * note: the len + off check is almost certainly unnecessary. */ if (!(tp->t_flags & TF_MORETOCOME) && /* normal case */ (idle || (tp->t_flags & TF_NODELAY)) && (uint32_t)len + (uint32_t)off >= sbavail(&so->so_snd) && (tp->t_flags & TF_NOPUSH) == 0) { goto send; } if (tp->t_flags & TF_FORCEDATA) /* typ. timeout case */ goto send; if (len >= tp->max_sndwnd / 2 && tp->max_sndwnd > 0) goto send; if (SEQ_LT(tp->snd_nxt, tp->snd_max)) /* retransmit case */ goto send; if (sack_rxmit) goto send; } /* * Sending of standalone window updates. * * Window updates are important when we close our window due to a * full socket buffer and are opening it again after the application * reads data from it. Once the window has opened again and the * remote end starts to send again the ACK clock takes over and * provides the most current window information. * * We must avoid the silly window syndrome whereas every read * from the receive buffer, no matter how small, causes a window * update to be sent. We also should avoid sending a flurry of * window updates when the socket buffer had queued a lot of data * and the application is doing small reads. * * Prevent a flurry of pointless window updates by only sending * an update when we can increase the advertized window by more * than 1/4th of the socket buffer capacity. When the buffer is * getting full or is very small be more aggressive and send an * update whenever we can increase by two mss sized segments. * In all other situations the ACK's to new incoming data will * carry further window increases. * * Don't send an independent window update if a delayed * ACK is pending (it will get piggy-backed on it) or the * remote side already has done a half-close and won't send * more data. */ if (recwin > 0 && !(tp->t_flags & TF_NEEDSYN) && !(tp->t_flags & TF_DELACK) && !TCPS_HAVERCVDFIN(tp->t_state)) { /* * "adv" is the amount we could increase the window, * taking into account that we are limited by * TCP_MAXWIN << tp->rcv_scale. */ int32_t adv; int oldwin; adv = recwin; if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt)) { oldwin = (tp->rcv_adv - tp->rcv_nxt); if (adv > oldwin) adv -= oldwin; else adv = 0; } else oldwin = 0; /* * If the new window size ends up being the same as or less * than the old size when it is scaled, then don't force * a window update. */ if (oldwin >> tp->rcv_scale >= (adv + oldwin) >> tp->rcv_scale) goto dontupdate; if (adv >= (int32_t)(2 * tp->t_maxseg) && (adv >= (int32_t)(so->so_rcv.sb_hiwat / 4) || recwin <= (so->so_rcv.sb_hiwat / 8) || so->so_rcv.sb_hiwat <= 8 * tp->t_maxseg || adv >= TCP_MAXWIN << tp->rcv_scale)) goto send; if (2 * adv >= (int32_t)so->so_rcv.sb_hiwat) goto send; } dontupdate: /* * Send if we owe the peer an ACK, RST, SYN, or urgent data. ACKNOW * is also a catch-all for the retransmit timer timeout case. */ if (tp->t_flags & TF_ACKNOW) goto send; if ((flags & TH_RST) || ((flags & TH_SYN) && (tp->t_flags & TF_NEEDSYN) == 0)) goto send; if (SEQ_GT(tp->snd_up, tp->snd_una)) goto send; /* * If our state indicates that FIN should be sent * and we have not yet done so, then we need to send. */ if (flags & TH_FIN && ((tp->t_flags & TF_SENTFIN) == 0 || tp->snd_nxt == tp->snd_una)) goto send; /* * In SACK, it is possible for tcp_output to fail to send a segment * after the retransmission timer has been turned off. Make sure * that the retransmission timer is set. */ if ((tp->t_flags & TF_SACK_PERMIT) && SEQ_GT(tp->snd_max, tp->snd_una) && !tcp_timer_active(tp, TT_REXMT) && !tcp_timer_active(tp, TT_PERSIST)) { tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); goto just_return; } /* * TCP window updates are not reliable, rather a polling protocol * using ``persist'' packets is used to insure receipt of window * updates. The three ``states'' for the output side are: * idle not doing retransmits or persists * persisting to move a small or zero window * (re)transmitting and thereby not persisting * * tcp_timer_active(tp, TT_PERSIST) * is true when we are in persist state. * (tp->t_flags & TF_FORCEDATA) * is set when we are called to send a persist packet. * tcp_timer_active(tp, TT_REXMT) * is set when we are retransmitting * The output side is idle when both timers are zero. * * If send window is too small, there is data to transmit, and no * retransmit or persist is pending, then go to persist state. * If nothing happens soon, send when timer expires: * if window is nonzero, transmit what we can, * otherwise force out a byte. */ if (sbavail(&so->so_snd) && !tcp_timer_active(tp, TT_REXMT) && !tcp_timer_active(tp, TT_PERSIST)) { tp->t_rxtshift = 0; tcp_setpersist(tp); } /* * No reason to send a segment, just return. */ just_return: SOCKBUF_UNLOCK(&so->so_snd); return (0); send: SOCKBUF_LOCK_ASSERT(&so->so_snd); if (len > 0) { if (len >= tp->t_maxseg) tp->t_flags2 |= TF2_PLPMTU_MAXSEGSNT; else tp->t_flags2 &= ~TF2_PLPMTU_MAXSEGSNT; } /* * Before ESTABLISHED, force sending of initial options * unless TCP set not to do any options. * NOTE: we assume that the IP/TCP header plus TCP options * always fit in a single mbuf, leaving room for a maximum * link header, i.e. * max_linkhdr + sizeof (struct tcpiphdr) + optlen <= MCLBYTES */ optlen = 0; #ifdef INET6 if (isipv6) hdrlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr); else #endif hdrlen = sizeof (struct tcpiphdr); if (flags & TH_SYN) { tp->snd_nxt = tp->iss; } /* * Compute options for segment. * We only have to care about SYN and established connection * segments. Options for SYN-ACK segments are handled in TCP * syncache. */ to.to_flags = 0; if ((tp->t_flags & TF_NOOPT) == 0) { /* Maximum segment size. */ if (flags & TH_SYN) { to.to_mss = tcp_mssopt(&inp->inp_inc); if (tp->t_port) to.to_mss -= V_tcp_udp_tunneling_overhead; to.to_flags |= TOF_MSS; /* * On SYN or SYN|ACK transmits on TFO connections, * only include the TFO option if it is not a * retransmit, as the presence of the TFO option may * have caused the original SYN or SYN|ACK to have * been dropped by a middlebox. */ if (IS_FASTOPEN(tp->t_flags) && (tp->t_rxtshift == 0)) { if (tp->t_state == TCPS_SYN_RECEIVED) { to.to_tfo_len = TCP_FASTOPEN_COOKIE_LEN; to.to_tfo_cookie = (u_int8_t *)&tp->t_tfo_cookie.server; to.to_flags |= TOF_FASTOPEN; wanted_cookie = 1; } else if (tp->t_state == TCPS_SYN_SENT) { to.to_tfo_len = tp->t_tfo_client_cookie_len; to.to_tfo_cookie = tp->t_tfo_cookie.client; to.to_flags |= TOF_FASTOPEN; wanted_cookie = 1; /* * If we wind up having more data to * send with the SYN than can fit in * one segment, don't send any more * until the SYN|ACK comes back from * the other end. */ dont_sendalot = 1; } } } /* Window scaling. */ if ((flags & TH_SYN) && (tp->t_flags & TF_REQ_SCALE)) { to.to_wscale = tp->request_r_scale; to.to_flags |= TOF_SCALE; } /* Timestamps. */ if ((tp->t_flags & TF_RCVD_TSTMP) || ((flags & TH_SYN) && (tp->t_flags & TF_REQ_TSTMP))) { curticks = tcp_ts_getticks(); to.to_tsval = curticks + tp->ts_offset; to.to_tsecr = tp->ts_recent; to.to_flags |= TOF_TS; if (tp->t_rxtshift == 1) tp->t_badrxtwin = curticks; } /* Set receive buffer autosizing timestamp. */ if (tp->rfbuf_ts == 0 && (so->so_rcv.sb_flags & SB_AUTOSIZE)) tp->rfbuf_ts = tcp_ts_getticks(); /* Selective ACK's. */ if (tp->t_flags & TF_SACK_PERMIT) { if (flags & TH_SYN) to.to_flags |= TOF_SACKPERM; else if (TCPS_HAVEESTABLISHED(tp->t_state) && tp->rcv_numsacks > 0) { to.to_flags |= TOF_SACK; to.to_nsacks = tp->rcv_numsacks; to.to_sacks = (u_char *)tp->sackblks; } } #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) /* TCP-MD5 (RFC2385). */ /* * Check that TCP_MD5SIG is enabled in tcpcb to * account the size needed to set this TCP option. */ if (tp->t_flags & TF_SIGNATURE) to.to_flags |= TOF_SIGNATURE; #endif /* TCP_SIGNATURE */ /* Processing the options. */ hdrlen += optlen = tcp_addoptions(&to, opt); /* * If we wanted a TFO option to be added, but it was unable * to fit, ensure no data is sent. */ if (IS_FASTOPEN(tp->t_flags) && wanted_cookie && !(to.to_flags & TOF_FASTOPEN)) len = 0; } if (tp->t_port) { if (V_tcp_udp_tunneling_port == 0) { /* The port was removed?? */ SOCKBUF_UNLOCK(&so->so_snd); return (EHOSTUNREACH); } hdrlen += sizeof(struct udphdr); } /* * Adjust data length if insertion of options will * bump the packet length beyond the t_maxseg length. * Clear the FIN bit because we cut off the tail of * the segment. */ if (len + optlen + ipoptlen > tp->t_maxseg) { flags &= ~TH_FIN; if (tso) { u_int if_hw_tsomax; u_int moff; int max_len; /* extract TSO information */ if_hw_tsomax = tp->t_tsomax; if_hw_tsomaxsegcount = tp->t_tsomaxsegcount; if_hw_tsomaxsegsize = tp->t_tsomaxsegsize; /* * Limit a TSO burst to prevent it from * overflowing or exceeding the maximum length * allowed by the network interface: */ KASSERT(ipoptlen == 0, ("%s: TSO can't do IP options", __func__)); /* * Check if we should limit by maximum payload * length: */ if (if_hw_tsomax != 0) { /* compute maximum TSO length */ max_len = (if_hw_tsomax - hdrlen - max_linkhdr); if (max_len <= 0) { len = 0; } else if (len > max_len) { sendalot = 1; len = max_len; } } /* * Prevent the last segment from being * fractional unless the send sockbuf can be * emptied: */ max_len = (tp->t_maxseg - optlen); if (((uint32_t)off + (uint32_t)len) < sbavail(&so->so_snd)) { moff = len % max_len; if (moff != 0) { len -= moff; sendalot = 1; } } /* * In case there are too many small fragments * don't use TSO: */ if (len <= max_len) { len = max_len; sendalot = 1; tso = 0; } /* * Send the FIN in a separate segment * after the bulk sending is done. * We don't trust the TSO implementations * to clear the FIN flag on all but the * last segment. */ if (tp->t_flags & TF_NEEDFIN) sendalot = 1; } else { if (optlen + ipoptlen >= tp->t_maxseg) { /* * Since we don't have enough space to put * the IP header chain and the TCP header in * one packet as required by RFC 7112, don't * send it. Also ensure that at least one * byte of the payload can be put into the * TCP segment. */ SOCKBUF_UNLOCK(&so->so_snd); error = EMSGSIZE; sack_rxmit = 0; goto out; } len = tp->t_maxseg - optlen - ipoptlen; sendalot = 1; if (dont_sendalot) sendalot = 0; } } else tso = 0; KASSERT(len + hdrlen + ipoptlen <= IP_MAXPACKET, ("%s: len > IP_MAXPACKET", __func__)); /*#ifdef DIAGNOSTIC*/ #ifdef INET6 if (max_linkhdr + hdrlen > MCLBYTES) #else if (max_linkhdr + hdrlen > MHLEN) #endif panic("tcphdr too big"); /*#endif*/ /* * This KASSERT is here to catch edge cases at a well defined place. * Before, those had triggered (random) panic conditions further down. */ KASSERT(len >= 0, ("[%s:%d]: len < 0", __func__, __LINE__)); /* * Grab a header mbuf, attaching a copy of data to * be transmitted, and initialize the header from * the template for sends on this connection. */ if (len) { struct mbuf *mb; struct sockbuf *msb; u_int moff; if ((tp->t_flags & TF_FORCEDATA) && len == 1) { TCPSTAT_INC(tcps_sndprobe); #ifdef STATS if (SEQ_LT(tp->snd_nxt, tp->snd_max)) stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RETXPB, len); else stats_voi_update_abs_u64(tp->t_stats, VOI_TCP_TXPB, len); #endif /* STATS */ } else if (SEQ_LT(tp->snd_nxt, tp->snd_max) || sack_rxmit) { tp->t_sndrexmitpack++; TCPSTAT_INC(tcps_sndrexmitpack); TCPSTAT_ADD(tcps_sndrexmitbyte, len); if (sack_rxmit) { TCPSTAT_INC(tcps_sack_rexmits); TCPSTAT_ADD(tcps_sack_rexmit_bytes, len); } #ifdef STATS stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RETXPB, len); #endif /* STATS */ } else { TCPSTAT_INC(tcps_sndpack); TCPSTAT_ADD(tcps_sndbyte, len); #ifdef STATS stats_voi_update_abs_u64(tp->t_stats, VOI_TCP_TXPB, len); #endif /* STATS */ } #ifdef INET6 if (MHLEN < hdrlen + max_linkhdr) m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); else #endif m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { SOCKBUF_UNLOCK(&so->so_snd); error = ENOBUFS; sack_rxmit = 0; goto out; } m->m_data += max_linkhdr; m->m_len = hdrlen; /* * Start the m_copy functions from the closest mbuf * to the offset in the socket buffer chain. */ mb = sbsndptr_noadv(&so->so_snd, off, &moff); if (len <= MHLEN - hdrlen - max_linkhdr && !hw_tls) { m_copydata(mb, moff, len, mtod(m, caddr_t) + hdrlen); if (SEQ_LT(tp->snd_nxt, tp->snd_max)) sbsndptr_adv(&so->so_snd, mb, len); m->m_len += len; } else { if (SEQ_LT(tp->snd_nxt, tp->snd_max)) msb = NULL; else msb = &so->so_snd; m->m_next = tcp_m_copym(mb, moff, &len, if_hw_tsomaxsegcount, if_hw_tsomaxsegsize, msb, hw_tls); if (len <= (tp->t_maxseg - optlen)) { /* * Must have ran out of mbufs for the copy * shorten it to no longer need tso. Lets * not put on sendalot since we are low on * mbufs. */ tso = 0; } if (m->m_next == NULL) { SOCKBUF_UNLOCK(&so->so_snd); (void) m_free(m); error = ENOBUFS; sack_rxmit = 0; goto out; } } /* * If we're sending everything we've got, set PUSH. * (This will keep happy those implementations which only * give data to the user when a buffer fills or * a PUSH comes in.) */ if (((uint32_t)off + (uint32_t)len == sbused(&so->so_snd)) && !(flags & TH_SYN)) flags |= TH_PUSH; SOCKBUF_UNLOCK(&so->so_snd); } else { SOCKBUF_UNLOCK(&so->so_snd); if (tp->t_flags & TF_ACKNOW) TCPSTAT_INC(tcps_sndacks); else if (flags & (TH_SYN|TH_FIN|TH_RST)) TCPSTAT_INC(tcps_sndctrl); else if (SEQ_GT(tp->snd_up, tp->snd_una)) TCPSTAT_INC(tcps_sndurg); else TCPSTAT_INC(tcps_sndwinup); m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) { error = ENOBUFS; sack_rxmit = 0; goto out; } #ifdef INET6 if (isipv6 && (MHLEN < hdrlen + max_linkhdr) && MHLEN >= hdrlen) { M_ALIGN(m, hdrlen); } else #endif m->m_data += max_linkhdr; m->m_len = hdrlen; } SOCKBUF_UNLOCK_ASSERT(&so->so_snd); m->m_pkthdr.rcvif = (struct ifnet *)0; #ifdef MAC mac_inpcb_create_mbuf(inp, m); #endif #ifdef INET6 if (isipv6) { ip6 = mtod(m, struct ip6_hdr *); if (tp->t_port) { udp = (struct udphdr *)((caddr_t)ip6 + sizeof(struct ip6_hdr)); udp->uh_sport = htons(V_tcp_udp_tunneling_port); udp->uh_dport = tp->t_port; ulen = hdrlen + len - sizeof(struct ip6_hdr); udp->uh_ulen = htons(ulen); th = (struct tcphdr *)(udp + 1); } else { th = (struct tcphdr *)(ip6 + 1); } tcpip_fillheaders(inp, tp->t_port, ip6, th); } else #endif /* INET6 */ { ip = mtod(m, struct ip *); if (tp->t_port) { udp = (struct udphdr *)((caddr_t)ip + sizeof(struct ip)); udp->uh_sport = htons(V_tcp_udp_tunneling_port); udp->uh_dport = tp->t_port; ulen = hdrlen + len - sizeof(struct ip); udp->uh_ulen = htons(ulen); th = (struct tcphdr *)(udp + 1); } else th = (struct tcphdr *)(ip + 1); tcpip_fillheaders(inp, tp->t_port, ip, th); } /* * Fill in fields, remembering maximum advertised * window for use in delaying messages about window sizes. * If resending a FIN, be sure not to use a new sequence number. */ if (flags & TH_FIN && tp->t_flags & TF_SENTFIN && tp->snd_nxt == tp->snd_max) tp->snd_nxt--; /* * If we are starting a connection, send ECN setup * SYN packet. If we are on a retransmit, we may * resend those bits a number of times as per * RFC 3168. */ if (tp->t_state == TCPS_SYN_SENT && V_tcp_do_ecn) { flags |= tcp_ecn_output_syn_sent(tp); } /* Also handle parallel SYN for ECN */ if ((TCPS_HAVERCVDSYN(tp->t_state)) && (tp->t_flags2 & (TF2_ECN_PERMIT | TF2_ACE_PERMIT))) { int ect = tcp_ecn_output_established(tp, &flags, len, sack_rxmit); if ((tp->t_state == TCPS_SYN_RECEIVED) && (tp->t_flags2 & TF2_ECN_SND_ECE)) tp->t_flags2 &= ~TF2_ECN_SND_ECE; #ifdef INET6 if (isipv6) { ip6->ip6_flow &= ~htonl(IPTOS_ECN_MASK << IPV6_FLOWLABEL_LEN); ip6->ip6_flow |= htonl(ect << IPV6_FLOWLABEL_LEN); } else #endif { ip->ip_tos &= ~IPTOS_ECN_MASK; ip->ip_tos |= ect; } } /* * If we are doing retransmissions, then snd_nxt will * not reflect the first unsent octet. For ACK only * packets, we do not want the sequence number of the * retransmitted packet, we want the sequence number * of the next unsent octet. So, if there is no data * (and no SYN or FIN), use snd_max instead of snd_nxt * when filling in ti_seq. But if we are in persist * state, snd_max might reflect one byte beyond the * right edge of the window, so use snd_nxt in that * case, since we know we aren't doing a retransmission. * (retransmit and persist are mutually exclusive...) */ if (sack_rxmit == 0) { if (len || (flags & (TH_SYN|TH_FIN)) || tcp_timer_active(tp, TT_PERSIST)) th->th_seq = htonl(tp->snd_nxt); else th->th_seq = htonl(tp->snd_max); } else { th->th_seq = htonl(p->rxmit); p->rxmit += len; /* * Lost Retransmission Detection * trigger resending of a (then * still existing) hole, when * fack acks recoverypoint. */ if ((tp->t_flags & TF_LRD) && SEQ_GEQ(p->rxmit, p->end)) p->rxmit = tp->snd_recover; tp->sackhint.sack_bytes_rexmit += len; } if (IN_RECOVERY(tp->t_flags)) { /* * Account all bytes transmitted while * IN_RECOVERY, simplifying PRR and * Lost Retransmit Detection */ tp->sackhint.prr_out += len; } th->th_ack = htonl(tp->rcv_nxt); if (optlen) { bcopy(opt, th + 1, optlen); th->th_off = (sizeof (struct tcphdr) + optlen) >> 2; } tcp_set_flags(th, flags); /* * Calculate receive window. Don't shrink window, * but avoid silly window syndrome. * If a RST segment is sent, advertise a window of zero. */ if (flags & TH_RST) { recwin = 0; } else { if (recwin < (so->so_rcv.sb_hiwat / 4) && recwin < tp->t_maxseg) recwin = 0; if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt) && recwin < (tp->rcv_adv - tp->rcv_nxt)) recwin = (tp->rcv_adv - tp->rcv_nxt); } /* * According to RFC1323 the window field in a SYN (i.e., a * or ) segment itself is never scaled. The * case is handled in syncache. */ if (flags & TH_SYN) th->th_win = htons((u_short) (min(sbspace(&so->so_rcv), TCP_MAXWIN))); else { /* Avoid shrinking window with window scaling. */ recwin = roundup2(recwin, 1 << tp->rcv_scale); th->th_win = htons((u_short)(recwin >> tp->rcv_scale)); } /* * Adjust the RXWIN0SENT flag - indicate that we have advertised * a 0 window. This may cause the remote transmitter to stall. This * flag tells soreceive() to disable delayed acknowledgements when * draining the buffer. This can occur if the receiver is attempting * to read more data than can be buffered prior to transmitting on * the connection. */ if (th->th_win == 0) { tp->t_sndzerowin++; tp->t_flags |= TF_RXWIN0SENT; } else tp->t_flags &= ~TF_RXWIN0SENT; if (SEQ_GT(tp->snd_up, tp->snd_nxt)) { th->th_urp = htons((u_short)(tp->snd_up - tp->snd_nxt)); th->th_flags |= TH_URG; } else /* * If no urgent pointer to send, then we pull * the urgent pointer to the left edge of the send window * so that it doesn't drift into the send window on sequence * number wraparound. */ tp->snd_up = tp->snd_una; /* drag it along */ /* * Put TCP length in extended header, and then * checksum extended header and data. */ m->m_pkthdr.len = hdrlen + len; /* in6_cksum() need this */ #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE) if (to.to_flags & TOF_SIGNATURE) { /* * Calculate MD5 signature and put it into the place * determined before. * NOTE: since TCP options buffer doesn't point into * mbuf's data, calculate offset and use it. */ if (!TCPMD5_ENABLED() || (error = TCPMD5_OUTPUT(m, th, (u_char *)(th + 1) + (to.to_signature - opt))) != 0) { /* * Do not send segment if the calculation of MD5 * digest has failed. */ m_freem(m); goto out; } } #endif #ifdef INET6 if (isipv6) { /* * There is no need to fill in ip6_plen right now. * It will be filled later by ip6_output. */ if (tp->t_port) { m->m_pkthdr.csum_flags = CSUM_UDP_IPV6; m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); udp->uh_sum = in6_cksum_pseudo(ip6, ulen, IPPROTO_UDP, 0); th->th_sum = htons(0); UDPSTAT_INC(udps_opackets); } else { m->m_pkthdr.csum_flags = CSUM_TCP_IPV6; m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); th->th_sum = in6_cksum_pseudo(ip6, sizeof(struct tcphdr) + optlen + len, IPPROTO_TCP, 0); } } #endif #if defined(INET6) && defined(INET) else #endif #ifdef INET { if (tp->t_port) { m->m_pkthdr.csum_flags = CSUM_UDP; m->m_pkthdr.csum_data = offsetof(struct udphdr, uh_sum); udp->uh_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(ulen + IPPROTO_UDP)); th->th_sum = htons(0); UDPSTAT_INC(udps_opackets); } else { m->m_pkthdr.csum_flags = CSUM_TCP; m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum); th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr, htons(sizeof(struct tcphdr) + IPPROTO_TCP + len + optlen)); } /* IP version must be set here for ipv4/ipv6 checking later */ KASSERT(ip->ip_v == IPVERSION, ("%s: IP version incorrect: %d", __func__, ip->ip_v)); } #endif /* * Enable TSO and specify the size of the segments. * The TCP pseudo header checksum is always provided. */ if (tso) { KASSERT(len > tp->t_maxseg - optlen, ("%s: len <= tso_segsz", __func__)); m->m_pkthdr.csum_flags |= CSUM_TSO; m->m_pkthdr.tso_segsz = tp->t_maxseg - optlen; } KASSERT(len + hdrlen == m_length(m, NULL), ("%s: mbuf chain shorter than expected: %d + %u != %u", __func__, len, hdrlen, m_length(m, NULL))); #ifdef TCP_HHOOK /* Run HHOOK_TCP_ESTABLISHED_OUT helper hooks. */ hhook_run_tcp_est_out(tp, th, &to, len, tso); #endif TCP_PROBE3(debug__output, tp, th, m); /* We're getting ready to send; log now. */ /* XXXMT: We are not honoring verbose logging. */ if (tcp_bblogging_on(tp)) lgb = tcp_log_event(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_OUT, ERRNO_UNK, len, NULL, false, NULL, NULL, 0, NULL); else lgb = NULL; /* * Fill in IP length and desired time to live and * send to IP level. There should be a better way * to handle ttl and tos; we could keep them in * the template, but need a way to checksum without them. */ /* * m->m_pkthdr.len should have been set before checksum calculation, * because in6_cksum() need it. */ #ifdef INET6 if (isipv6) { /* * we separately set hoplimit for every segment, since the * user might want to change the value via setsockopt. * Also, desired default hop limit might be changed via * Neighbor Discovery. */ ip6->ip6_hlim = in6_selecthlim(inp, NULL); /* * Set the packet size here for the benefit of DTrace probes. * ip6_output() will set it properly; it's supposed to include * the option header lengths as well. */ ip6->ip6_plen = htons(m->m_pkthdr.len - sizeof(*ip6)); if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) tp->t_flags2 |= TF2_PLPMTU_PMTUD; else tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; if (tp->t_state == TCPS_SYN_SENT) TCP_PROBE5(connect__request, NULL, tp, ip6, tp, th); TCP_PROBE5(send, NULL, tp, ip6, tp, th); #ifdef TCPPCAP /* Save packet, if requested. */ tcp_pcap_add(th, m, &(tp->t_outpkts)); #endif /* TODO: IPv6 IP6TOS_ECT bit on */ error = ip6_output(m, inp->in6p_outputopts, &inp->inp_route6, ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), NULL, NULL, inp); if (error == EMSGSIZE && inp->inp_route6.ro_nh != NULL) mtu = inp->inp_route6.ro_nh->nh_mtu; } #endif /* INET6 */ #if defined(INET) && defined(INET6) else #endif #ifdef INET { ip->ip_len = htons(m->m_pkthdr.len); #ifdef INET6 if (inp->inp_vflag & INP_IPV6PROTO) ip->ip_ttl = in6_selecthlim(inp, NULL); #endif /* INET6 */ /* * If we do path MTU discovery, then we set DF on every packet. * This might not be the best thing to do according to RFC3390 * Section 2. However the tcp hostcache migitates the problem * so it affects only the first tcp connection with a host. * * NB: Don't set DF on small MTU/MSS to have a safe fallback. */ if (V_path_mtu_discovery && tp->t_maxseg > V_tcp_minmss) { tp->t_flags2 |= TF2_PLPMTU_PMTUD; if (tp->t_port == 0 || len < V_tcp_minmss) { ip->ip_off |= htons(IP_DF); } } else { tp->t_flags2 &= ~TF2_PLPMTU_PMTUD; } if (tp->t_state == TCPS_SYN_SENT) TCP_PROBE5(connect__request, NULL, tp, ip, tp, th); TCP_PROBE5(send, NULL, tp, ip, tp, th); #ifdef TCPPCAP /* Save packet, if requested. */ tcp_pcap_add(th, m, &(tp->t_outpkts)); #endif error = ip_output(m, inp->inp_options, &inp->inp_route, ((so->so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0), 0, inp); if (error == EMSGSIZE && inp->inp_route.ro_nh != NULL) mtu = inp->inp_route.ro_nh->nh_mtu; } #endif /* INET */ if (lgb != NULL) { lgb->tlb_errno = error; lgb = NULL; } out: if (error == 0) tcp_account_for_send(tp, len, (tp->snd_nxt != tp->snd_max), 0, hw_tls); /* * In transmit state, time the transmission and arrange for * the retransmit. In persist state, just set snd_max. */ if ((tp->t_flags & TF_FORCEDATA) == 0 || !tcp_timer_active(tp, TT_PERSIST)) { tcp_seq startseq = tp->snd_nxt; /* * Advance snd_nxt over sequence space of this segment. */ if (flags & (TH_SYN|TH_FIN)) { if (flags & TH_SYN) tp->snd_nxt++; if (flags & TH_FIN) { tp->snd_nxt++; tp->t_flags |= TF_SENTFIN; } } if (sack_rxmit) goto timer; tp->snd_nxt += len; if (SEQ_GT(tp->snd_nxt, tp->snd_max)) { /* * Update "made progress" indication if we just * added new data to an empty socket buffer. */ if (tp->snd_una == tp->snd_max) tp->t_acktime = ticks; tp->snd_max = tp->snd_nxt; /* * Time this transmission if not a retransmission and * not currently timing anything. */ tp->t_sndtime = ticks; if (tp->t_rtttime == 0) { tp->t_rtttime = ticks; tp->t_rtseq = startseq; TCPSTAT_INC(tcps_segstimed); } #ifdef STATS if (!(tp->t_flags & TF_GPUTINPROG) && len) { tp->t_flags |= TF_GPUTINPROG; tp->gput_seq = startseq; tp->gput_ack = startseq + ulmin(sbavail(&so->so_snd) - off, sendwin); tp->gput_ts = tcp_ts_getticks(); } #endif /* STATS */ } /* * Set retransmit timer if not currently set, * and not doing a pure ack or a keep-alive probe. * Initial value for retransmit timer is smoothed * round-trip time + 2 * round-trip time variance. * Initialize shift counter which is used for backoff * of retransmit time. */ timer: if (!tcp_timer_active(tp, TT_REXMT) && ((sack_rxmit && tp->snd_nxt != tp->snd_max) || (tp->snd_nxt != tp->snd_una))) { if (tcp_timer_active(tp, TT_PERSIST)) { tcp_timer_activate(tp, TT_PERSIST, 0); tp->t_rxtshift = 0; } tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp)); } else if (len == 0 && sbavail(&so->so_snd) && !tcp_timer_active(tp, TT_REXMT) && !tcp_timer_active(tp, TT_PERSIST)) { /* * Avoid a situation where we do not set persist timer * after a zero window condition. For example: * 1) A -> B: packet with enough data to fill the window * 2) B -> A: ACK for #1 + new data (0 window * advertisement) * 3) A -> B: ACK for #2, 0 len packet * * In this case, A will not activate the persist timer, * because it chose to send a packet. Unless tcp_output * is called for some other reason (delayed ack timer, * another input packet from B, socket syscall), A will * not send zero window probes. * * So, if you send a 0-length packet, but there is data * in the socket buffer, and neither the rexmt or * persist timer is already set, then activate the * persist timer. */ tp->t_rxtshift = 0; tcp_setpersist(tp); } } else { /* * Persist case, update snd_max but since we are in * persist mode (no window) we do not update snd_nxt. */ int xlen = len; if (flags & TH_SYN) ++xlen; if (flags & TH_FIN) { ++xlen; tp->t_flags |= TF_SENTFIN; } if (SEQ_GT(tp->snd_nxt + xlen, tp->snd_max)) tp->snd_max = tp->snd_nxt + xlen; } if ((error == 0) && (TCPS_HAVEESTABLISHED(tp->t_state) && (tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks > 0)) { /* Clean up any DSACK's sent */ tcp_clean_dsack_blocks(tp); } if (error) { /* * We know that the packet was lost, so back out the * sequence number advance, if any. * * If the error is EPERM the packet got blocked by the * local firewall. Normally we should terminate the * connection but the blocking may have been spurious * due to a firewall reconfiguration cycle. So we treat * it like a packet loss and let the retransmit timer and * timeouts do their work over time. * XXX: It is a POLA question whether calling tcp_drop right * away would be the really correct behavior instead. */ if (((tp->t_flags & TF_FORCEDATA) == 0 || !tcp_timer_active(tp, TT_PERSIST)) && ((flags & TH_SYN) == 0) && (error != EPERM)) { if (sack_rxmit) { p->rxmit = SEQ_MIN(p->end, p->rxmit) - len; tp->sackhint.sack_bytes_rexmit -= len; KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, ("sackhint bytes rtx >= 0")); KASSERT((flags & TH_FIN) == 0, ("error while FIN with SACK rxmit")); } else { tp->snd_nxt -= len; if (flags & TH_FIN) tp->snd_nxt--; } if (IN_RECOVERY(tp->t_flags)) tp->sackhint.prr_out -= len; } SOCKBUF_UNLOCK_ASSERT(&so->so_snd); /* Check gotos. */ switch (error) { case EACCES: case EPERM: tp->t_softerror = error; return (error); case ENOBUFS: TCP_XMIT_TIMER_ASSERT(tp, len, flags); tp->snd_cwnd = tp->t_maxseg; return (0); case EMSGSIZE: /* * For some reason the interface we used initially * to send segments changed to another or lowered * its MTU. * If TSO was active we either got an interface * without TSO capabilits or TSO was turned off. * If we obtained mtu from ip_output() then update * it and try again. */ if (tso) tp->t_flags &= ~TF_TSO; if (mtu != 0) { tcp_mss_update(tp, -1, mtu, NULL, NULL); goto again; } return (error); case EHOSTDOWN: case EHOSTUNREACH: case ENETDOWN: case ENETUNREACH: if (TCPS_HAVERCVDSYN(tp->t_state)) { tp->t_softerror = error; return (0); } /* FALLTHROUGH */ default: return (error); } } TCPSTAT_INC(tcps_sndtotal); /* * Data sent (as far as we can tell). * If this advertises a larger window than any other segment, * then remember the size of the advertised window. * Any pending ACK has now been sent. */ if (SEQ_GT(tp->rcv_nxt + recwin, tp->rcv_adv)) tp->rcv_adv = tp->rcv_nxt + recwin; tp->last_ack_sent = tp->rcv_nxt; tp->t_flags &= ~(TF_ACKNOW | TF_DELACK); if (tcp_timer_active(tp, TT_DELACK)) tcp_timer_activate(tp, TT_DELACK, 0); if (sendalot) goto again; return (0); } void tcp_setpersist(struct tcpcb *tp) { int t = ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1; int tt; int maxunacktime; tp->t_flags &= ~TF_PREVVALID; if (tcp_timer_active(tp, TT_REXMT)) panic("tcp_setpersist: retransmit pending"); /* * If the state is already closed, don't bother. */ if (tp->t_state == TCPS_CLOSED) return; /* * Start/restart persistence timer. */ TCPT_RANGESET(tt, t * tcp_backoff[tp->t_rxtshift], tcp_persmin, tcp_persmax); if (TP_MAXUNACKTIME(tp) && tp->t_acktime) { maxunacktime = tp->t_acktime + TP_MAXUNACKTIME(tp) - ticks; if (maxunacktime < 1) maxunacktime = 1; if (maxunacktime < tt) tt = maxunacktime; } tcp_timer_activate(tp, TT_PERSIST, tt); if (tp->t_rxtshift < V_tcp_retries) tp->t_rxtshift++; } /* * Insert TCP options according to the supplied parameters to the place * optp in a consistent way. Can handle unaligned destinations. * * The order of the option processing is crucial for optimal packing and * alignment for the scarce option space. * * The optimal order for a SYN/SYN-ACK segment is: * MSS (4) + NOP (1) + Window scale (3) + SACK permitted (2) + * Timestamp (10) + Signature (18) = 38 bytes out of a maximum of 40. * * The SACK options should be last. SACK blocks consume 8*n+2 bytes. * So a full size SACK blocks option is 34 bytes (with 4 SACK blocks). * At minimum we need 10 bytes (to generate 1 SACK block). If both * TCP Timestamps (12 bytes) and TCP Signatures (18 bytes) are present, * we only have 10 bytes for SACK options (40 - (12 + 18)). */ int tcp_addoptions(struct tcpopt *to, u_char *optp) { u_int32_t mask, optlen = 0; for (mask = 1; mask < TOF_MAXOPT; mask <<= 1) { if ((to->to_flags & mask) != mask) continue; if (optlen == TCP_MAXOLEN) break; switch (to->to_flags & mask) { case TOF_MSS: while (optlen % 4) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_MAXSEG) continue; optlen += TCPOLEN_MAXSEG; *optp++ = TCPOPT_MAXSEG; *optp++ = TCPOLEN_MAXSEG; to->to_mss = htons(to->to_mss); bcopy((u_char *)&to->to_mss, optp, sizeof(to->to_mss)); optp += sizeof(to->to_mss); break; case TOF_SCALE: while (!optlen || optlen % 2 != 1) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_WINDOW) continue; optlen += TCPOLEN_WINDOW; *optp++ = TCPOPT_WINDOW; *optp++ = TCPOLEN_WINDOW; *optp++ = to->to_wscale; break; case TOF_SACKPERM: while (optlen % 2) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_SACK_PERMITTED) continue; optlen += TCPOLEN_SACK_PERMITTED; *optp++ = TCPOPT_SACK_PERMITTED; *optp++ = TCPOLEN_SACK_PERMITTED; break; case TOF_TS: while (!optlen || optlen % 4 != 2) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_TIMESTAMP) continue; optlen += TCPOLEN_TIMESTAMP; *optp++ = TCPOPT_TIMESTAMP; *optp++ = TCPOLEN_TIMESTAMP; to->to_tsval = htonl(to->to_tsval); to->to_tsecr = htonl(to->to_tsecr); bcopy((u_char *)&to->to_tsval, optp, sizeof(to->to_tsval)); optp += sizeof(to->to_tsval); bcopy((u_char *)&to->to_tsecr, optp, sizeof(to->to_tsecr)); optp += sizeof(to->to_tsecr); break; case TOF_SIGNATURE: { int siglen = TCPOLEN_SIGNATURE - 2; while (!optlen || optlen % 4 != 2) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_SIGNATURE) { to->to_flags &= ~TOF_SIGNATURE; continue; } optlen += TCPOLEN_SIGNATURE; *optp++ = TCPOPT_SIGNATURE; *optp++ = TCPOLEN_SIGNATURE; to->to_signature = optp; while (siglen--) *optp++ = 0; break; } case TOF_SACK: { int sackblks = 0; struct sackblk *sack = (struct sackblk *)to->to_sacks; tcp_seq sack_seq; while (!optlen || optlen % 4 != 2) { optlen += TCPOLEN_NOP; *optp++ = TCPOPT_NOP; } if (TCP_MAXOLEN - optlen < TCPOLEN_SACKHDR + TCPOLEN_SACK) continue; optlen += TCPOLEN_SACKHDR; *optp++ = TCPOPT_SACK; sackblks = min(to->to_nsacks, (TCP_MAXOLEN - optlen) / TCPOLEN_SACK); *optp++ = TCPOLEN_SACKHDR + sackblks * TCPOLEN_SACK; while (sackblks--) { sack_seq = htonl(sack->start); bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); optp += sizeof(sack_seq); sack_seq = htonl(sack->end); bcopy((u_char *)&sack_seq, optp, sizeof(sack_seq)); optp += sizeof(sack_seq); optlen += TCPOLEN_SACK; sack++; } TCPSTAT_INC(tcps_sack_send_blocks); break; } case TOF_FASTOPEN: { int total_len; /* XXX is there any point to aligning this option? */ total_len = TCPOLEN_FAST_OPEN_EMPTY + to->to_tfo_len; if (TCP_MAXOLEN - optlen < total_len) { to->to_flags &= ~TOF_FASTOPEN; continue; } *optp++ = TCPOPT_FAST_OPEN; *optp++ = total_len; if (to->to_tfo_len > 0) { bcopy(to->to_tfo_cookie, optp, to->to_tfo_len); optp += to->to_tfo_len; } optlen += total_len; break; } default: panic("%s: unknown TCP option type", __func__); break; } } /* Terminate and pad TCP options to a 4 byte boundary. */ if (optlen % 4) { optlen += TCPOLEN_EOL; *optp++ = TCPOPT_EOL; } /* * According to RFC 793 (STD0007): * "The content of the header beyond the End-of-Option option * must be header padding (i.e., zero)." * and later: "The padding is composed of zeros." */ while (optlen % 4) { optlen += TCPOLEN_PAD; *optp++ = TCPOPT_PAD; } KASSERT(optlen <= TCP_MAXOLEN, ("%s: TCP options too long", __func__)); return (optlen); } /* * This is a copy of m_copym(), taking the TSO segment size/limit * constraints into account, and advancing the sndptr as it goes. */ struct mbuf * tcp_m_copym(struct mbuf *m, int32_t off0, int32_t *plen, int32_t seglimit, int32_t segsize, struct sockbuf *sb, bool hw_tls) { #ifdef KERN_TLS struct ktls_session *tls, *ntls; struct mbuf *start __diagused; #endif struct mbuf *n, **np; struct mbuf *top; int32_t off = off0; int32_t len = *plen; int32_t fragsize; int32_t len_cp = 0; int32_t *pkthdrlen; uint32_t mlen, frags; bool copyhdr; KASSERT(off >= 0, ("tcp_m_copym, negative off %d", off)); KASSERT(len >= 0, ("tcp_m_copym, negative len %d", len)); if (off == 0 && m->m_flags & M_PKTHDR) copyhdr = true; else copyhdr = false; while (off > 0) { KASSERT(m != NULL, ("tcp_m_copym, offset > size of mbuf chain")); if (off < m->m_len) break; off -= m->m_len; if ((sb) && (m == sb->sb_sndptr)) { sb->sb_sndptroff += m->m_len; sb->sb_sndptr = m->m_next; } m = m->m_next; } np = ⊤ top = NULL; pkthdrlen = NULL; #ifdef KERN_TLS if (hw_tls && (m->m_flags & M_EXTPG)) tls = m->m_epg_tls; else tls = NULL; start = m; #endif while (len > 0) { if (m == NULL) { KASSERT(len == M_COPYALL, ("tcp_m_copym, length > size of mbuf chain")); *plen = len_cp; if (pkthdrlen != NULL) *pkthdrlen = len_cp; break; } #ifdef KERN_TLS if (hw_tls) { if (m->m_flags & M_EXTPG) ntls = m->m_epg_tls; else ntls = NULL; /* * Avoid mixing TLS records with handshake * data or TLS records from different * sessions. */ if (tls != ntls) { MPASS(m != start); *plen = len_cp; if (pkthdrlen != NULL) *pkthdrlen = len_cp; break; } } #endif mlen = min(len, m->m_len - off); if (seglimit) { /* * For M_EXTPG mbufs, add 3 segments * + 1 in case we are crossing page boundaries * + 2 in case the TLS hdr/trailer are used * It is cheaper to just add the segments * than it is to take the cache miss to look * at the mbuf ext_pgs state in detail. */ if (m->m_flags & M_EXTPG) { fragsize = min(segsize, PAGE_SIZE); frags = 3; } else { fragsize = segsize; frags = 0; } /* Break if we really can't fit anymore. */ if ((frags + 1) >= seglimit) { *plen = len_cp; if (pkthdrlen != NULL) *pkthdrlen = len_cp; break; } /* * Reduce size if you can't copy the whole * mbuf. If we can't copy the whole mbuf, also * adjust len so the loop will end after this * mbuf. */ if ((frags + howmany(mlen, fragsize)) >= seglimit) { mlen = (seglimit - frags - 1) * fragsize; len = mlen; *plen = len_cp + len; if (pkthdrlen != NULL) *pkthdrlen = *plen; } frags += howmany(mlen, fragsize); if (frags == 0) frags++; seglimit -= frags; KASSERT(seglimit > 0, ("%s: seglimit went too low", __func__)); } if (copyhdr) n = m_gethdr(M_NOWAIT, m->m_type); else n = m_get(M_NOWAIT, m->m_type); *np = n; if (n == NULL) goto nospace; if (copyhdr) { if (!m_dup_pkthdr(n, m, M_NOWAIT)) goto nospace; if (len == M_COPYALL) n->m_pkthdr.len -= off0; else n->m_pkthdr.len = len; pkthdrlen = &n->m_pkthdr.len; copyhdr = false; } n->m_len = mlen; len_cp += n->m_len; if (m->m_flags & (M_EXT | M_EXTPG)) { n->m_data = m->m_data + off; mb_dupcl(n, m); } else bcopy(mtod(m, caddr_t)+off, mtod(n, caddr_t), (u_int)n->m_len); if (sb && (sb->sb_sndptr == m) && ((n->m_len + off) >= m->m_len) && m->m_next) { sb->sb_sndptroff += m->m_len; sb->sb_sndptr = m->m_next; } off = 0; if (len != M_COPYALL) { len -= n->m_len; } m = m->m_next; np = &n->m_next; } return (top); nospace: m_freem(top); return (NULL); } void tcp_sndbuf_autoscale(struct tcpcb *tp, struct socket *so, uint32_t sendwin) { /* * Automatic sizing of send socket buffer. Often the send buffer * size is not optimally adjusted to the actual network conditions * at hand (delay bandwidth product). Setting the buffer size too * small limits throughput on links with high bandwidth and high * delay (eg. trans-continental/oceanic links). Setting the * buffer size too big consumes too much real kernel memory, * especially with many connections on busy servers. * * The criteria to step up the send buffer one notch are: * 1. receive window of remote host is larger than send buffer * (with a fudge factor of 5/4th); * 2. send buffer is filled to 7/8th with data (so we actually * have data to make use of it); * 3. send buffer fill has not hit maximal automatic size; * 4. our send window (slow start and cogestion controlled) is * larger than sent but unacknowledged data in send buffer. * * The remote host receive window scaling factor may limit the * growing of the send buffer before it reaches its allowed * maximum. * * It scales directly with slow start or congestion window * and does at most one step per received ACK. This fast * scaling has the drawback of growing the send buffer beyond * what is strictly necessary to make full use of a given * delay*bandwidth product. However testing has shown this not * to be much of an problem. At worst we are trading wasting * of available bandwidth (the non-use of it) for wasting some * socket buffer memory. * * TODO: Shrink send buffer during idle periods together * with congestion window. Requires another timer. Has to * wait for upcoming tcp timer rewrite. * * XXXGL: should there be used sbused() or sbavail()? */ if (V_tcp_do_autosndbuf && so->so_snd.sb_flags & SB_AUTOSIZE) { int lowat; lowat = V_tcp_sendbuf_auto_lowat ? so->so_snd.sb_lowat : 0; if ((tp->snd_wnd / 4 * 5) >= so->so_snd.sb_hiwat - lowat && sbused(&so->so_snd) >= (so->so_snd.sb_hiwat / 8 * 7) - lowat && sbused(&so->so_snd) < V_tcp_autosndbuf_max && sendwin >= (sbused(&so->so_snd) - (tp->snd_nxt - tp->snd_una))) { if (!sbreserve_locked(so, SO_SND, min(so->so_snd.sb_hiwat + V_tcp_autosndbuf_inc, V_tcp_autosndbuf_max), curthread)) so->so_snd.sb_flags &= ~SB_AUTOSIZE; } } } diff --git a/sys/netinet/tcp_sack.c b/sys/netinet/tcp_sack.c index f517bb9fcdb7..0c557dc4579d 100644 --- a/sys/netinet/tcp_sack.c +++ b/sys/netinet/tcp_sack.c @@ -1,1143 +1,1150 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995 * The Regents of the University of California. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /*- * * NRL grants permission for redistribution and use in source and binary * forms, with or without modification, of the software and documentation * created at NRL provided that the following conditions are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgements: * This product includes software developed by the University of * California, Berkeley and its contributors. * This product includes software developed at the Information * Technology Division, US Naval Research Laboratory. * 4. Neither the name of the NRL nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THE SOFTWARE PROVIDED BY NRL IS PROVIDED BY NRL AND CONTRIBUTORS ``AS * IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A * PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL NRL OR * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * The views and conclusions contained in the software and documentation * are those of the authors and should not be interpreted as representing * official policies, either expressed or implied, of the US Naval * Research Laboratory (NRL). */ #include #include "opt_inet.h" #include "opt_inet6.h" #include #include #include #include #include #include #include /* for proc0 declaration */ #include #include #include #include #include #include /* before tcp_seq.h, for tcp_random18() */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include VNET_DECLARE(struct uma_zone *, sack_hole_zone); #define V_sack_hole_zone VNET(sack_hole_zone) SYSCTL_NODE(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "TCP SACK"); VNET_DEFINE(int, tcp_do_sack) = 1; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_sack), 0, "Enable/Disable TCP SACK support"); VNET_DEFINE(int, tcp_do_newsack) = 1; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, revised, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_newsack), 0, "Use revised SACK loss recovery per RFC 6675"); VNET_DEFINE(int, tcp_do_lrd) = 1; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, lrd, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_do_lrd), 1, "Perform Lost Retransmission Detection"); VNET_DEFINE(int, tcp_sack_maxholes) = 128; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, maxholes, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sack_maxholes), 0, "Maximum number of TCP SACK holes allowed per connection"); VNET_DEFINE(int, tcp_sack_globalmaxholes) = 65536; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalmaxholes, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_sack_globalmaxholes), 0, "Global maximum number of TCP SACK holes"); VNET_DEFINE(int, tcp_sack_globalholes) = 0; SYSCTL_INT(_net_inet_tcp_sack, OID_AUTO, globalholes, CTLFLAG_VNET | CTLFLAG_RD, &VNET_NAME(tcp_sack_globalholes), 0, "Global number of TCP SACK holes currently allocated"); int tcp_dsack_block_exists(struct tcpcb *tp) { /* Return true if a DSACK block exists */ if (tp->rcv_numsacks == 0) return (0); if (SEQ_LEQ(tp->sackblks[0].end, tp->rcv_nxt)) return(1); return (0); } /* * This function will find overlaps with the currently stored sackblocks * and add any overlap as a dsack block upfront */ void tcp_update_dsack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end) { struct sackblk head_blk,mid_blk,saved_blks[MAX_SACK_BLKS]; int i, j, n, identical; tcp_seq start, end; INP_WLOCK_ASSERT(tptoinpcb(tp)); KASSERT(SEQ_LT(rcv_start, rcv_end), ("rcv_start < rcv_end")); if (SEQ_LT(rcv_end, tp->rcv_nxt) || ((rcv_end == tp->rcv_nxt) && (tp->rcv_numsacks > 0 ) && (tp->sackblks[0].end == tp->rcv_nxt))) { saved_blks[0].start = rcv_start; saved_blks[0].end = rcv_end; } else { saved_blks[0].start = saved_blks[0].end = 0; } head_blk.start = head_blk.end = 0; mid_blk.start = rcv_start; mid_blk.end = rcv_end; identical = 0; for (i = 0; i < tp->rcv_numsacks; i++) { start = tp->sackblks[i].start; end = tp->sackblks[i].end; if (SEQ_LT(rcv_end, start)) { /* pkt left to sack blk */ continue; } if (SEQ_GT(rcv_start, end)) { /* pkt right to sack blk */ continue; } if (SEQ_GT(tp->rcv_nxt, end)) { if ((SEQ_MAX(rcv_start, start) != SEQ_MIN(rcv_end, end)) && (SEQ_GT(head_blk.start, SEQ_MAX(rcv_start, start)) || (head_blk.start == head_blk.end))) { head_blk.start = SEQ_MAX(rcv_start, start); head_blk.end = SEQ_MIN(rcv_end, end); } continue; } if (((head_blk.start == head_blk.end) || SEQ_LT(start, head_blk.start)) && (SEQ_GT(end, rcv_start) && SEQ_LEQ(start, rcv_end))) { head_blk.start = start; head_blk.end = end; } mid_blk.start = SEQ_MIN(mid_blk.start, start); mid_blk.end = SEQ_MAX(mid_blk.end, end); if ((mid_blk.start == start) && (mid_blk.end == end)) identical = 1; } if (SEQ_LT(head_blk.start, head_blk.end)) { /* store overlapping range */ saved_blks[0].start = SEQ_MAX(rcv_start, head_blk.start); saved_blks[0].end = SEQ_MIN(rcv_end, head_blk.end); } n = 1; /* * Second, if not ACKed, store the SACK block that * overlaps with the DSACK block unless it is identical */ if ((SEQ_LT(tp->rcv_nxt, mid_blk.end) && !((mid_blk.start == saved_blks[0].start) && (mid_blk.end == saved_blks[0].end))) || identical == 1) { saved_blks[n].start = mid_blk.start; saved_blks[n++].end = mid_blk.end; } for (j = 0; (j < tp->rcv_numsacks) && (n < MAX_SACK_BLKS); j++) { if (((SEQ_LT(tp->sackblks[j].end, mid_blk.start) || SEQ_GT(tp->sackblks[j].start, mid_blk.end)) && (SEQ_GT(tp->sackblks[j].start, tp->rcv_nxt)))) saved_blks[n++] = tp->sackblks[j]; } j = 0; for (i = 0; i < n; i++) { /* we can end up with a stale initial entry */ if (SEQ_LT(saved_blks[i].start, saved_blks[i].end)) { tp->sackblks[j++] = saved_blks[i]; } } tp->rcv_numsacks = j; } /* * This function is called upon receipt of new valid data (while not in * header prediction mode), and it updates the ordered list of sacks. */ void tcp_update_sack_list(struct tcpcb *tp, tcp_seq rcv_start, tcp_seq rcv_end) { /* * First reported block MUST be the most recent one. Subsequent * blocks SHOULD be in the order in which they arrived at the * receiver. These two conditions make the implementation fully * compliant with RFC 2018. */ struct sackblk head_blk, saved_blks[MAX_SACK_BLKS]; int num_head, num_saved, i; INP_WLOCK_ASSERT(tptoinpcb(tp)); /* Check arguments. */ KASSERT(SEQ_LEQ(rcv_start, rcv_end), ("rcv_start <= rcv_end")); if ((rcv_start == rcv_end) && (tp->rcv_numsacks >= 1) && (rcv_end == tp->sackblks[0].end)) { /* retaining DSACK block below rcv_nxt (todrop) */ head_blk = tp->sackblks[0]; } else { /* SACK block for the received segment. */ head_blk.start = rcv_start; head_blk.end = rcv_end; } /* * Merge updated SACK blocks into head_blk, and save unchanged SACK * blocks into saved_blks[]. num_saved will have the number of the * saved SACK blocks. */ num_saved = 0; for (i = 0; i < tp->rcv_numsacks; i++) { tcp_seq start = tp->sackblks[i].start; tcp_seq end = tp->sackblks[i].end; if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) { /* * Discard this SACK block. */ } else if (SEQ_LEQ(head_blk.start, end) && SEQ_GEQ(head_blk.end, start)) { /* * Merge this SACK block into head_blk. This SACK * block itself will be discarded. */ /* * |-| * |---| merge * * |-| * |---| merge * * |-----| * |-| DSACK smaller * * |-| * |-----| DSACK smaller */ if (head_blk.start == end) head_blk.start = start; else if (head_blk.end == start) head_blk.end = end; else { if (SEQ_LT(head_blk.start, start)) { tcp_seq temp = start; start = head_blk.start; head_blk.start = temp; } if (SEQ_GT(head_blk.end, end)) { tcp_seq temp = end; end = head_blk.end; head_blk.end = temp; } if ((head_blk.start != start) || (head_blk.end != end)) { if ((num_saved >= 1) && SEQ_GEQ(saved_blks[num_saved-1].start, start) && SEQ_LEQ(saved_blks[num_saved-1].end, end)) num_saved--; saved_blks[num_saved].start = start; saved_blks[num_saved].end = end; num_saved++; } } } else { /* * This block supercedes the prior block */ if ((num_saved >= 1) && SEQ_GEQ(saved_blks[num_saved-1].start, start) && SEQ_LEQ(saved_blks[num_saved-1].end, end)) num_saved--; /* * Save this SACK block. */ saved_blks[num_saved].start = start; saved_blks[num_saved].end = end; num_saved++; } } /* * Update SACK list in tp->sackblks[]. */ num_head = 0; if (SEQ_LT(rcv_start, rcv_end)) { /* * The received data segment is an out-of-order segment. Put * head_blk at the top of SACK list. */ tp->sackblks[0] = head_blk; num_head = 1; /* * If the number of saved SACK blocks exceeds its limit, * discard the last SACK block. */ if (num_saved >= MAX_SACK_BLKS) num_saved--; } if ((rcv_start == rcv_end) && (rcv_start == tp->sackblks[0].end)) { num_head = 1; } if (num_saved > 0) { /* * Copy the saved SACK blocks back. */ bcopy(saved_blks, &tp->sackblks[num_head], sizeof(struct sackblk) * num_saved); } /* Save the number of SACK blocks. */ tp->rcv_numsacks = num_head + num_saved; } void tcp_clean_dsack_blocks(struct tcpcb *tp) { struct sackblk saved_blks[MAX_SACK_BLKS]; int num_saved, i; INP_WLOCK_ASSERT(tptoinpcb(tp)); /* * Clean up any DSACK blocks that * are in our queue of sack blocks. * */ num_saved = 0; for (i = 0; i < tp->rcv_numsacks; i++) { tcp_seq start = tp->sackblks[i].start; tcp_seq end = tp->sackblks[i].end; if (SEQ_GEQ(start, end) || SEQ_LEQ(start, tp->rcv_nxt)) { /* * Discard this D-SACK block. */ continue; } /* * Save this SACK block. */ saved_blks[num_saved].start = start; saved_blks[num_saved].end = end; num_saved++; } if (num_saved > 0) { /* * Copy the saved SACK blocks back. */ bcopy(saved_blks, &tp->sackblks[0], sizeof(struct sackblk) * num_saved); } tp->rcv_numsacks = num_saved; } /* * Delete all receiver-side SACK information. */ void tcp_clean_sackreport(struct tcpcb *tp) { int i; INP_WLOCK_ASSERT(tptoinpcb(tp)); tp->rcv_numsacks = 0; for (i = 0; i < MAX_SACK_BLKS; i++) tp->sackblks[i].start = tp->sackblks[i].end=0; } /* * Allocate struct sackhole. */ static struct sackhole * tcp_sackhole_alloc(struct tcpcb *tp, tcp_seq start, tcp_seq end) { struct sackhole *hole; if (tp->snd_numholes >= V_tcp_sack_maxholes || V_tcp_sack_globalholes >= V_tcp_sack_globalmaxholes) { TCPSTAT_INC(tcps_sack_sboverflow); return NULL; } hole = (struct sackhole *)uma_zalloc(V_sack_hole_zone, M_NOWAIT); if (hole == NULL) return NULL; hole->start = start; hole->end = end; hole->rxmit = start; tp->snd_numholes++; atomic_add_int(&V_tcp_sack_globalholes, 1); return hole; } /* * Free struct sackhole. */ static void tcp_sackhole_free(struct tcpcb *tp, struct sackhole *hole) { uma_zfree(V_sack_hole_zone, hole); tp->snd_numholes--; atomic_subtract_int(&V_tcp_sack_globalholes, 1); KASSERT(tp->snd_numholes >= 0, ("tp->snd_numholes >= 0")); KASSERT(V_tcp_sack_globalholes >= 0, ("tcp_sack_globalholes >= 0")); } /* * Insert new SACK hole into scoreboard. */ static struct sackhole * tcp_sackhole_insert(struct tcpcb *tp, tcp_seq start, tcp_seq end, struct sackhole *after) { struct sackhole *hole; /* Allocate a new SACK hole. */ hole = tcp_sackhole_alloc(tp, start, end); if (hole == NULL) return NULL; /* Insert the new SACK hole into scoreboard. */ if (after != NULL) TAILQ_INSERT_AFTER(&tp->snd_holes, after, hole, scblink); else TAILQ_INSERT_TAIL(&tp->snd_holes, hole, scblink); /* Update SACK hint. */ if (tp->sackhint.nexthole == NULL) tp->sackhint.nexthole = hole; return hole; } /* * Remove SACK hole from scoreboard. */ static void tcp_sackhole_remove(struct tcpcb *tp, struct sackhole *hole) { /* Update SACK hint. */ if (tp->sackhint.nexthole == hole) tp->sackhint.nexthole = TAILQ_NEXT(hole, scblink); /* Remove this SACK hole. */ TAILQ_REMOVE(&tp->snd_holes, hole, scblink); /* Free this SACK hole. */ tcp_sackhole_free(tp, hole); } /* * Process cumulative ACK and the TCP SACK option to update the scoreboard. * tp->snd_holes is an ordered list of holes (oldest to newest, in terms of * the sequence space). * Returns SACK_NEWLOSS if incoming ACK indicates ongoing loss (hole split, new hole), * SACK_CHANGE if incoming ACK has previously unknown SACK information, * SACK_NOCHANGE otherwise. */ sackstatus_t tcp_sack_doack(struct tcpcb *tp, struct tcpopt *to, tcp_seq th_ack) { struct sackhole *cur, *temp; struct sackblk sack, sack_blocks[TCP_MAX_SACK + 1], *sblkp; int i, j, num_sack_blks; sackstatus_t sack_changed; int delivered_data, left_edge_delta; tcp_seq loss_hiack = 0; int loss_thresh = 0; int loss_sblks = 0; int notlost_bytes = 0; INP_WLOCK_ASSERT(tptoinpcb(tp)); num_sack_blks = 0; sack_changed = SACK_NOCHANGE; delivered_data = 0; left_edge_delta = 0; /* * If SND.UNA will be advanced by SEG.ACK, and if SACK holes exist, * treat [SND.UNA, SEG.ACK) as if it is a SACK block. * Account changes to SND.UNA always in delivered data. */ if (SEQ_LT(tp->snd_una, th_ack) && !TAILQ_EMPTY(&tp->snd_holes)) { left_edge_delta = th_ack - tp->snd_una; sack_blocks[num_sack_blks].start = tp->snd_una; sack_blocks[num_sack_blks++].end = th_ack; /* * Pulling snd_fack forward if we got here * due to DSACK blocks */ if (SEQ_LT(tp->snd_fack, th_ack)) { delivered_data += th_ack - tp->snd_una; tp->snd_fack = th_ack; sack_changed = SACK_CHANGE; } } /* * Append received valid SACK blocks to sack_blocks[], but only if we * received new blocks from the other side. */ if (to->to_flags & TOF_SACK) { for (i = 0; i < to->to_nsacks; i++) { bcopy((to->to_sacks + i * TCPOLEN_SACK), &sack, sizeof(sack)); sack.start = ntohl(sack.start); sack.end = ntohl(sack.end); if (SEQ_GT(sack.end, sack.start) && SEQ_GT(sack.start, tp->snd_una) && SEQ_GT(sack.start, th_ack) && SEQ_LT(sack.start, tp->snd_max) && SEQ_GT(sack.end, tp->snd_una) && SEQ_LEQ(sack.end, tp->snd_max)) { sack_blocks[num_sack_blks++] = sack; } else if (SEQ_LEQ(sack.start, th_ack) && SEQ_LEQ(sack.end, th_ack)) { /* * Its a D-SACK block. */ tcp_record_dsack(tp, sack.start, sack.end, 0); } } } /* * Return if SND.UNA is not advanced and no valid SACK block is * received. */ if (num_sack_blks == 0) return (sack_changed); /* * Sort the SACK blocks so we can update the scoreboard with just one * pass. The overhead of sorting up to 4+1 elements is less than * making up to 4+1 passes over the scoreboard. */ for (i = 0; i < num_sack_blks; i++) { for (j = i + 1; j < num_sack_blks; j++) { if (SEQ_GT(sack_blocks[i].end, sack_blocks[j].end)) { sack = sack_blocks[i]; sack_blocks[i] = sack_blocks[j]; sack_blocks[j] = sack; } } } if (TAILQ_EMPTY(&tp->snd_holes)) { /* * Empty scoreboard. Need to initialize snd_fack (it may be * uninitialized or have a bogus value). Scoreboard holes * (from the sack blocks received) are created later below * (in the logic that adds holes to the tail of the * scoreboard). */ tp->snd_fack = SEQ_MAX(tp->snd_una, th_ack); tp->sackhint.sacked_bytes = 0; /* reset */ tp->sackhint.hole_bytes = 0; } /* * In the while-loop below, incoming SACK blocks (sack_blocks[]) and * SACK holes (snd_holes) are traversed from their tails with just * one pass in order to reduce the number of compares especially when * the bandwidth-delay product is large. * * Note: Typically, in the first RTT of SACK recovery, the highest * three or four SACK blocks with the same ack number are received. * In the second RTT, if retransmitted data segments are not lost, * the highest three or four SACK blocks with ack number advancing * are received. */ sblkp = &sack_blocks[num_sack_blks - 1]; /* Last SACK block */ tp->sackhint.last_sack_ack = sblkp->end; if (SEQ_LT(tp->snd_fack, sblkp->start)) { /* * The highest SACK block is beyond fack. First, * check if there was a successful Rescue Retransmission, * and move this hole left. With normal holes, snd_fack * is always to the right of the end. */ if (((temp = TAILQ_LAST(&tp->snd_holes, sackhole_head)) != NULL) && SEQ_LEQ(tp->snd_fack,temp->end)) { tp->sackhint.hole_bytes -= temp->end - temp->start; temp->start = SEQ_MAX(tp->snd_fack, SEQ_MAX(tp->snd_una, th_ack)); temp->end = sblkp->start; temp->rxmit = temp->start; delivered_data += sblkp->end - sblkp->start; tp->sackhint.hole_bytes += temp->end - temp->start; KASSERT(tp->sackhint.hole_bytes >= 0, ("sackhint hole bytes >= 0")); tp->snd_fack = sblkp->end; sblkp--; sack_changed = SACK_NEWLOSS; } else { /* * Append a new SACK hole at the tail. If the * second or later highest SACK blocks are also * beyond the current fack, they will be inserted * by way of hole splitting in the while-loop below. */ temp = tcp_sackhole_insert(tp, tp->snd_fack,sblkp->start,NULL); if (temp != NULL) { delivered_data += sblkp->end - sblkp->start; tp->sackhint.hole_bytes += temp->end - temp->start; tp->snd_fack = sblkp->end; /* Go to the previous sack block. */ sblkp--; sack_changed = SACK_CHANGE; } else { /* * We failed to add a new hole based on the current * sack block. Skip over all the sack blocks that * fall completely to the right of snd_fack and * proceed to trim the scoreboard based on the * remaining sack blocks. This also trims the * scoreboard for th_ack (which is sack_blocks[0]). */ while (sblkp >= sack_blocks && SEQ_LT(tp->snd_fack, sblkp->start)) sblkp--; if (sblkp >= sack_blocks && SEQ_LT(tp->snd_fack, sblkp->end)) { delivered_data += sblkp->end - tp->snd_fack; tp->snd_fack = sblkp->end; /* * While the Scoreboard didn't change in * size, we only ended up here because * some SACK data had to be dismissed. */ sack_changed = SACK_NEWLOSS; } } } } else if (SEQ_LT(tp->snd_fack, sblkp->end)) { /* fack is advanced. */ delivered_data += sblkp->end - tp->snd_fack; tp->snd_fack = sblkp->end; sack_changed = SACK_CHANGE; } cur = TAILQ_LAST(&tp->snd_holes, sackhole_head); /* Last SACK hole. */ loss_hiack = tp->snd_fack; /* * Since the incoming sack blocks are sorted, we can process them * making one sweep of the scoreboard. */ while (cur != NULL) { if (!(sblkp >= sack_blocks)) { if (((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) break; loss_thresh += loss_hiack - cur->end; loss_hiack = cur->start; loss_sblks++; if (!((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) { notlost_bytes += cur->end - cur->start; } else { break; } cur = TAILQ_PREV(cur, sackhole_head, scblink); continue; } if (SEQ_GEQ(sblkp->start, cur->end)) { /* * SACKs data beyond the current hole. Go to the * previous sack block. */ sblkp--; continue; } if (SEQ_LEQ(sblkp->end, cur->start)) { /* * SACKs data before the current hole. Go to the * previous hole. */ loss_thresh += loss_hiack - cur->end; loss_hiack = cur->start; loss_sblks++; if (!((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) notlost_bytes += cur->end - cur->start; cur = TAILQ_PREV(cur, sackhole_head, scblink); continue; } tp->sackhint.sack_bytes_rexmit -= (SEQ_MIN(cur->rxmit, cur->end) - cur->start); KASSERT(tp->sackhint.sack_bytes_rexmit >= 0, ("sackhint bytes rtx >= 0")); sack_changed = SACK_CHANGE; if (SEQ_LEQ(sblkp->start, cur->start)) { /* Data acks at least the beginning of hole. */ if (SEQ_GEQ(sblkp->end, cur->end)) { /* Acks entire hole, so delete hole. */ delivered_data += (cur->end - cur->start); temp = cur; cur = TAILQ_PREV(cur, sackhole_head, scblink); tp->sackhint.hole_bytes -= temp->end - temp->start; tcp_sackhole_remove(tp, temp); /* * The sack block may ack all or part of the * next hole too, so continue onto the next * hole. */ continue; } else { /* Move start of hole forward. */ delivered_data += (sblkp->end - cur->start); tp->sackhint.hole_bytes -= sblkp->end - cur->start; cur->start = sblkp->end; cur->rxmit = SEQ_MAX(cur->rxmit, cur->start); } } else { /* Data acks at least the end of hole. */ if (SEQ_GEQ(sblkp->end, cur->end)) { /* Move end of hole backward. */ delivered_data += (cur->end - sblkp->start); tp->sackhint.hole_bytes -= cur->end - sblkp->start; cur->end = sblkp->start; cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); if ((tp->t_flags & TF_LRD) && SEQ_GEQ(cur->rxmit, cur->end)) cur->rxmit = tp->snd_recover; } else { /* * ACKs some data in middle of a hole; need * to split current hole */ temp = tcp_sackhole_insert(tp, sblkp->end, cur->end, cur); sack_changed = SACK_NEWLOSS; if (temp != NULL) { if (SEQ_GT(cur->rxmit, temp->rxmit)) { temp->rxmit = cur->rxmit; tp->sackhint.sack_bytes_rexmit += (SEQ_MIN(temp->rxmit, temp->end) - temp->start); } tp->sackhint.hole_bytes -= sblkp->end - sblkp->start; loss_thresh += loss_hiack - temp->end; loss_hiack = temp->start; loss_sblks++; if (!((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) notlost_bytes += temp->end - temp->start; cur->end = sblkp->start; cur->rxmit = SEQ_MIN(cur->rxmit, cur->end); if ((tp->t_flags & TF_LRD) && SEQ_GEQ(cur->rxmit, cur->end)) cur->rxmit = tp->snd_recover; delivered_data += (sblkp->end - sblkp->start); } } } tp->sackhint.sack_bytes_rexmit += (SEQ_MIN(cur->rxmit, cur->end) - cur->start); /* * Testing sblkp->start against cur->start tells us whether * we're done with the sack block or the sack hole. * Accordingly, we advance one or the other. */ if (SEQ_LEQ(sblkp->start, cur->start)) { loss_thresh += loss_hiack - cur->end; loss_hiack = cur->start; loss_sblks++; if (!((loss_sblks >= tcprexmtthresh) || (loss_thresh > (tcprexmtthresh-1)*tp->t_maxseg))) notlost_bytes += cur->end - cur->start; cur = TAILQ_PREV(cur, sackhole_head, scblink); } else { sblkp--; } } KASSERT(!(TAILQ_EMPTY(&tp->snd_holes) && (tp->sackhint.hole_bytes != 0)), ("SACK scoreboard empty, but accounting non-zero\n")); KASSERT(notlost_bytes <= tp->sackhint.hole_bytes, ("SACK: more bytes marked notlost than in scoreboard holes")); if (!(to->to_flags & TOF_SACK)) /* * If this ACK did not contain any * SACK blocks, any only moved the * left edge right, it is a pure * cumulative ACK. Do not count * DupAck for this. Also required * for RFC6675 rescue retransmission. */ sack_changed = SACK_NOCHANGE; tp->sackhint.delivered_data = delivered_data; tp->sackhint.sacked_bytes += delivered_data - left_edge_delta; tp->sackhint.lost_bytes = tp->sackhint.hole_bytes - notlost_bytes; KASSERT((delivered_data >= 0), ("delivered_data < 0")); KASSERT((tp->sackhint.sacked_bytes >= 0), ("sacked_bytes < 0")); return (sack_changed); } /* * Free all SACK holes to clear the scoreboard. */ void tcp_free_sackholes(struct tcpcb *tp) { struct sackhole *q; INP_WLOCK_ASSERT(tptoinpcb(tp)); while ((q = TAILQ_FIRST(&tp->snd_holes)) != NULL) tcp_sackhole_remove(tp, q); tp->sackhint.sack_bytes_rexmit = 0; tp->sackhint.delivered_data = 0; tp->sackhint.sacked_bytes = 0; tp->sackhint.hole_bytes = 0; tp->sackhint.lost_bytes = 0; KASSERT(tp->snd_numholes == 0, ("tp->snd_numholes == 0")); KASSERT(tp->sackhint.nexthole == NULL, ("tp->sackhint.nexthole == NULL")); } /* * Resend all the currently existing SACK holes of * the scoreboard. This is in line with the Errata to * RFC 2018, which allows the use of SACK data past * an RTO to good effect typically. */ void tcp_resend_sackholes(struct tcpcb *tp) { struct sackhole *p; INP_WLOCK_ASSERT(tptoinpcb(tp)); TAILQ_FOREACH(p, &tp->snd_holes, scblink) { p->rxmit = p->start; } tp->sackhint.nexthole = TAILQ_FIRST(&tp->snd_holes); tp->sackhint.sack_bytes_rexmit = 0; } /* * Partial ack handling within a sack recovery episode. Keeping this very * simple for now. When a partial ack is received, force snd_cwnd to a value * that will allow the sender to transmit no more than 2 segments. If * necessary, a better scheme can be adopted at a later point, but for now, * the goal is to prevent the sender from bursting a large amount of data in * the midst of sack recovery. */ void tcp_sack_partialack(struct tcpcb *tp, struct tcphdr *th, u_int *maxsegp) { struct sackhole *temp; int num_segs = 1; u_int maxseg; INP_WLOCK_ASSERT(tptoinpcb(tp)); if (*maxsegp == 0) { *maxsegp = tcp_maxseg(tp); } maxseg = *maxsegp; tcp_timer_activate(tp, TT_REXMT, 0); tp->t_rtttime = 0; /* Send one or 2 segments based on how much new data was acked. */ if ((BYTES_THIS_ACK(tp, th) / maxseg) >= 2) num_segs = 2; tp->snd_cwnd = (tp->sackhint.sack_bytes_rexmit + (tp->snd_nxt - tp->snd_recover) + num_segs * maxseg); if (tp->snd_cwnd > tp->snd_ssthresh) tp->snd_cwnd = tp->snd_ssthresh; tp->t_flags |= TF_ACKNOW; /* * RFC6675 rescue retransmission * Add a hole between th_ack (snd_una is not yet set) and snd_max, * if this was a pure cumulative ACK and no data was send beyond * recovery point. Since the data in the socket has not been freed * at this point, we check if the scoreboard is empty, and the ACK * delivered some new data, indicating a full ACK. Also, if the * recovery point is still at snd_max, we are probably application * limited. However, this inference might not always be true. The * rescue retransmission may rarely be slightly premature * compared to RFC6675. * The corresponding ACK+SACK will cause any further outstanding * segments to be retransmitted. This addresses a corner case, when * the trailing packets of a window are lost and no further data * is available for sending. */ if ((V_tcp_do_newsack) && SEQ_LT(th->th_ack, tp->snd_recover) && TAILQ_EMPTY(&tp->snd_holes) && (tp->sackhint.delivered_data > 0)) { /* * Exclude FIN sequence space in * the hole for the rescue retransmission, * and also don't create a hole, if only * the ACK for a FIN is outstanding. */ tcp_seq highdata = tp->snd_max; if (tp->t_flags & TF_SENTFIN) highdata--; highdata = SEQ_MIN(highdata, tp->snd_recover); - if (th->th_ack != highdata) { + if (SEQ_LT(th->th_ack, highdata)) { tp->snd_fack = th->th_ack; if ((temp = tcp_sackhole_insert(tp, SEQ_MAX(th->th_ack, - highdata - maxseg), highdata, NULL)) != NULL) - tp->sackhint.hole_bytes += temp->end - - temp->start; + highdata - maxseg), highdata, NULL)) != NULL) { + tp->sackhint.hole_bytes += + temp->end - temp->start; + } } } (void) tcp_output(tp); } /* * Returns the next hole to retransmit and the number of retransmitted bytes * from the scoreboard. We store both the next hole and the number of * retransmitted bytes as hints (and recompute these on the fly upon SACK/ACK * reception). This avoids scoreboard traversals completely. * * The loop here will traverse *at most* one link. Here's the argument. For * the loop to traverse more than 1 link before finding the next hole to * retransmit, we would need to have at least 1 node following the current * hint with (rxmit == end). But, for all holes following the current hint, * (start == rxmit), since we have not yet retransmitted from them. * Therefore, in order to traverse more 1 link in the loop below, we need to * have at least one node following the current hint with (start == rxmit == * end). But that can't happen, (start == end) means that all the data in * that hole has been sacked, in which case, the hole would have been removed * from the scoreboard. */ struct sackhole * tcp_sack_output(struct tcpcb *tp, int *sack_bytes_rexmt) { struct sackhole *hole = NULL; INP_WLOCK_ASSERT(tptoinpcb(tp)); *sack_bytes_rexmt = tp->sackhint.sack_bytes_rexmit; hole = tp->sackhint.nexthole; if (hole == NULL) return (hole); if (SEQ_GEQ(hole->rxmit, hole->end)) { for (;;) { hole = TAILQ_NEXT(hole, scblink); if (hole == NULL) return (hole); if (SEQ_LT(hole->rxmit, hole->end)) { tp->sackhint.nexthole = hole; break; } } } KASSERT(SEQ_LT(hole->start, hole->end), ("%s: hole.start >= hole.end", __func__)); if (!(V_tcp_do_newsack)) { KASSERT(SEQ_LT(hole->start, tp->snd_fack), ("%s: hole.start >= snd.fack", __func__)); KASSERT(SEQ_LT(hole->end, tp->snd_fack), ("%s: hole.end >= snd.fack", __func__)); KASSERT(SEQ_LT(hole->rxmit, tp->snd_fack), ("%s: hole.rxmit >= snd.fack", __func__)); if (SEQ_GEQ(hole->start, hole->end) || SEQ_GEQ(hole->start, tp->snd_fack) || SEQ_GEQ(hole->end, tp->snd_fack) || SEQ_GEQ(hole->rxmit, tp->snd_fack)) { log(LOG_CRIT,"tcp: invalid SACK hole (%u-%u,%u) vs fwd ack %u, ignoring.\n", hole->start, hole->end, hole->rxmit, tp->snd_fack); return (NULL); } } return (hole); } /* * After a timeout, the SACK list may be rebuilt. This SACK information * should be used to avoid retransmitting SACKed data. This function * traverses the SACK list to see if snd_nxt should be moved forward. */ void tcp_sack_adjust(struct tcpcb *tp) { struct sackhole *p, *cur = TAILQ_FIRST(&tp->snd_holes); INP_WLOCK_ASSERT(tptoinpcb(tp)); - if (cur == NULL) - return; /* No holes */ - if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack)) - return; /* We're already beyond any SACKed blocks */ + if (cur == NULL) { + /* No holes */ + return; + } + if (SEQ_GEQ(tp->snd_nxt, tp->snd_fack)) { + /* We're already beyond any SACKed blocks */ + return; + } /*- * Two cases for which we want to advance snd_nxt: * i) snd_nxt lies between end of one hole and beginning of another * ii) snd_nxt lies between end of last hole and snd_fack */ while ((p = TAILQ_NEXT(cur, scblink)) != NULL) { - if (SEQ_LT(tp->snd_nxt, cur->end)) + if (SEQ_LT(tp->snd_nxt, cur->end)) { return; - if (SEQ_GEQ(tp->snd_nxt, p->start)) + } + if (SEQ_GEQ(tp->snd_nxt, p->start)) { cur = p; - else { + } else { tp->snd_nxt = p->start; return; } } - if (SEQ_LT(tp->snd_nxt, cur->end)) + if (SEQ_LT(tp->snd_nxt, cur->end)) { return; + } tp->snd_nxt = tp->snd_fack; } /* * Lost Retransmission Detection * Check is FACK is beyond the rexmit of the leftmost hole. * If yes, we restart sending from still existing holes, * and adjust cwnd via the congestion control module. */ void tcp_sack_lost_retransmission(struct tcpcb *tp, struct tcphdr *th) { struct sackhole *temp; if (IN_RECOVERY(tp->t_flags) && SEQ_GT(tp->snd_fack, tp->snd_recover) && ((temp = TAILQ_FIRST(&tp->snd_holes)) != NULL) && SEQ_GEQ(temp->rxmit, temp->end) && SEQ_GEQ(tp->snd_fack, temp->rxmit)) { TCPSTAT_INC(tcps_sack_lostrexmt); /* * Start retransmissions from the first hole, and * subsequently all other remaining holes, including * those, which had been sent completely before. */ tp->sackhint.nexthole = temp; TAILQ_FOREACH(temp, &tp->snd_holes, scblink) { if (SEQ_GEQ(tp->snd_fack, temp->rxmit) && SEQ_GEQ(temp->rxmit, temp->end)) temp->rxmit = temp->start; } /* * Remember the old ssthresh, to deduct the beta factor used * by the CC module. Finally, set cwnd to ssthresh just * prior to invoking another cwnd reduction by the CC * module, to not shrink it excessively. */ tp->snd_cwnd = tp->snd_ssthresh; /* * Formally exit recovery, and let the CC module adjust * ssthresh as intended. */ EXIT_RECOVERY(tp->t_flags); cc_cong_signal(tp, th, CC_NDUPACK); /* * For PRR, adjust recover_fs as if this new reduction * initialized this variable. * cwnd will be adjusted by SACK or PRR processing * subsequently, only set it to a safe value here. */ tp->snd_cwnd = tcp_maxseg(tp); tp->sackhint.recover_fs = (tp->snd_max - tp->snd_una) - tp->sackhint.recover_fs; } }