1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995 5 * The Regents of the University of California. All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of the University nor the names of its contributors 16 * may be used to endorse or promote products derived from this software 17 * without specific prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 29 * SUCH DAMAGE. 30 * 31 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95 32 */ 33 34 #include <sys/cdefs.h> 35 __FBSDID("$FreeBSD$"); 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_ipsec.h" 40 #include "opt_tcpdebug.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/callout.h> 45 #include <sys/kernel.h> 46 #include <sys/sysctl.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #include <sys/priv.h> 50 #include <sys/proc.h> 51 #include <sys/socket.h> 52 #include <sys/socketvar.h> 53 #include <sys/syslog.h> 54 #include <sys/protosw.h> 55 #include <sys/random.h> 56 57 #include <vm/uma.h> 58 59 #include <net/route.h> 60 #include <net/if.h> 61 #include <net/if_var.h> 62 #include <net/vnet.h> 63 64 #include <netinet/in.h> 65 #include <netinet/in_kdtrace.h> 66 #include <netinet/in_pcb.h> 67 #include <netinet/in_systm.h> 68 #include <netinet/in_var.h> 69 #include <netinet/ip.h> 70 #include <netinet/ip_icmp.h> 71 #include <netinet/ip_var.h> 72 #ifdef INET6 73 #include <netinet/ip6.h> 74 #include <netinet6/in6_pcb.h> 75 #include <netinet6/ip6_var.h> 76 #include <netinet6/scope6_var.h> 77 #include <netinet6/nd6.h> 78 #endif 79 #include <netinet/tcp.h> 80 #include <netinet/tcp_fsm.h> 81 #include <netinet/tcp_seq.h> 82 #include <netinet/tcp_timer.h> 83 #include <netinet/tcp_var.h> 84 #include <netinet/tcp_hpts.h> 85 #include <netinet/tcpip.h> 86 #ifdef TCPDEBUG 87 #include <netinet/tcp_debug.h> 88 #endif 89 90 #include <netinet/udp.h> 91 #include <netinet/udp_var.h> 92 93 #include <netipsec/ipsec_support.h> 94 95 #include <machine/in_cksum.h> 96 97 #include <security/mac/mac_framework.h> 98 99 VNET_DEFINE_STATIC(bool, nolocaltimewait) = true; 100 #define V_nolocaltimewait VNET(nolocaltimewait) 101 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, nolocaltimewait, 102 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nolocaltimewait), true, 103 "Do not create TCP TIME_WAIT state for local connections"); 104 105 /* 106 * Move a TCP connection into TIME_WAIT state. 107 * inp is locked, and is unlocked before returning. 108 * 109 * This function used to free tcpcb and allocate a compressed TCP time-wait 110 * structure tcptw. This served well for 20 years but is no longer relevant 111 * on modern machines in the modern internet. However, the function remains 112 * so that TCP stacks require less modification and we don't burn the bridge 113 * to go back to using compressed time-wait. 114 */ 115 void 116 tcp_twstart(struct tcpcb *tp) 117 { 118 struct inpcb *inp = tptoinpcb(tp); 119 #ifdef INET6 120 bool isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6; 121 #endif 122 123 NET_EPOCH_ASSERT(); 124 INP_WLOCK_ASSERT(inp); 125 126 /* A dropped inp should never transition to TIME_WAIT state. */ 127 KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: " 128 "(inp->inp_flags & INP_DROPPED) != 0")); 129 130 tcp_state_change(tp, TCPS_TIME_WAIT); 131 soisdisconnected(inp->inp_socket); 132 133 if (tp->t_flags & TF_ACKNOW) 134 tcp_output(tp); 135 136 if (V_nolocaltimewait && ( 137 #ifdef INET6 138 isipv6 ? in6_localaddr(&inp->in6p_faddr) : 139 #endif 140 #ifdef INET 141 in_localip(inp->inp_faddr) 142 #else 143 false 144 #endif 145 )) { 146 if ((tp = tcp_close(tp)) != NULL) 147 INP_WUNLOCK(inp); 148 return; 149 } 150 151 tcp_timer_activate(tp, TT_2MSL, 2 * V_tcp_msl); 152 INP_WUNLOCK(inp); 153 } 154 155 /* 156 * Returns true if the TIME_WAIT state was killed and we should start over, 157 * looking for a pcb in the listen state. Otherwise returns false and frees 158 * the mbuf. 159 * 160 * For pure SYN-segments the PCB shall be read-locked and the tcpopt pointer 161 * may be NULL. For the rest write-lock and valid tcpopt. 162 */ 163 bool 164 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th, 165 struct mbuf *m, int tlen) 166 { 167 struct tcpcb *tp = intotcpcb(inp); 168 char *s; 169 int thflags; 170 tcp_seq seq; 171 172 NET_EPOCH_ASSERT(); 173 INP_LOCK_ASSERT(inp); 174 175 thflags = tcp_get_flags(th); 176 #ifdef INVARIANTS 177 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) 178 INP_RLOCK_ASSERT(inp); 179 else { 180 INP_WLOCK_ASSERT(inp); 181 KASSERT(to != NULL, 182 ("%s: called without options on a non-SYN segment", 183 __func__)); 184 } 185 #endif 186 187 /* 188 * NOTE: for FIN_WAIT_2 (to be added later), 189 * must validate sequence number before accepting RST 190 */ 191 192 /* 193 * If the segment contains RST: 194 * Drop the segment - see Stevens, vol. 2, p. 964 and 195 * RFC 1337. 196 */ 197 if (thflags & TH_RST) 198 goto drop; 199 200 #if 0 201 /* PAWS not needed at the moment */ 202 /* 203 * RFC 1323 PAWS: If we have a timestamp reply on this segment 204 * and it's less than ts_recent, drop it. 205 */ 206 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent && 207 TSTMP_LT(to.to_tsval, tp->ts_recent)) { 208 if ((thflags & TH_ACK) == 0) 209 goto drop; 210 goto ack; 211 } 212 /* 213 * ts_recent is never updated because we never accept new segments. 214 */ 215 #endif 216 217 /* Honor the drop_synfin sysctl variable. */ 218 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) { 219 if ((s = tcp_log_addrs(&inp->inp_inc, th, NULL, NULL))) { 220 log(LOG_DEBUG, "%s; %s: " 221 "SYN|FIN segment ignored (based on " 222 "sysctl setting)\n", s, __func__); 223 free(s, M_TCPLOG); 224 } 225 goto drop; 226 } 227 228 /* 229 * If a new connection request is received 230 * while in TIME_WAIT, drop the old connection 231 * and start over if the sequence numbers 232 * are above the previous ones. 233 * Allow UDP port number changes in this case. 234 */ 235 if (((thflags & (TH_SYN | TH_ACK)) == TH_SYN) && 236 SEQ_GT(th->th_seq, tp->rcv_nxt)) { 237 /* 238 * In case we can't upgrade our lock just pretend we have 239 * lost this packet. 240 */ 241 if (INP_TRY_UPGRADE(inp) == 0) 242 goto drop; 243 if ((tp = tcp_close(tp)) != NULL) 244 INP_WUNLOCK(inp); 245 TCPSTAT_INC(tcps_tw_recycles); 246 return (true); 247 } 248 249 /* 250 * Send RST if UDP port numbers don't match 251 */ 252 if (tp->t_port != m->m_pkthdr.tcp_tun_port) { 253 if (tcp_get_flags(th) & TH_ACK) { 254 tcp_respond(tp, mtod(m, void *), th, m, 255 (tcp_seq)0, th->th_ack, TH_RST); 256 } else { 257 if (tcp_get_flags(th) & TH_SYN) 258 tlen++; 259 if (tcp_get_flags(th) & TH_FIN) 260 tlen++; 261 tcp_respond(tp, mtod(m, void *), th, m, 262 th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK); 263 } 264 INP_UNLOCK(inp); 265 TCPSTAT_INC(tcps_tw_resets); 266 return (false); 267 } 268 269 /* 270 * Drop the segment if it does not contain an ACK. 271 */ 272 if ((thflags & TH_ACK) == 0) 273 goto drop; 274 275 INP_WLOCK_ASSERT(inp); 276 277 /* 278 * If timestamps were negotiated during SYN/ACK and a 279 * segment without a timestamp is received, silently drop 280 * the segment, unless the missing timestamps are tolerated. 281 * See section 3.2 of RFC 7323. 282 */ 283 if (((to->to_flags & TOF_TS) == 0) && (tp->ts_recent != 0) && 284 (V_tcp_tolerate_missing_ts == 0)) { 285 goto drop; 286 } 287 288 /* 289 * Reset the 2MSL timer if this is a duplicate FIN. 290 */ 291 if (thflags & TH_FIN) { 292 seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0); 293 if (seq + 1 == tp->rcv_nxt) 294 tcp_timer_activate(tp, TT_2MSL, 2 * V_tcp_msl); 295 } 296 297 /* 298 * Acknowledge the segment if it has data or is not a duplicate ACK. 299 */ 300 if (thflags != TH_ACK || tlen != 0 || 301 th->th_seq != tp->rcv_nxt || th->th_ack != tp->snd_nxt) { 302 TCP_PROBE5(receive, NULL, NULL, m, NULL, th); 303 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt, 304 tp->snd_nxt, TH_ACK); 305 INP_UNLOCK(inp); 306 TCPSTAT_INC(tcps_tw_responds); 307 return (false); 308 } 309 drop: 310 TCP_PROBE5(receive, NULL, NULL, m, NULL, th); 311 INP_UNLOCK(inp); 312 m_freem(m); 313 return (false); 314 } 315