xref: /freebsd/sys/netinet/tcp_timewait.c (revision 66fd12cf4896eb08ad8e7a2627537f84ead84dd3)
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 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/syslog.h>
53 #include <sys/protosw.h>
54 #include <sys/random.h>
55 
56 #include <vm/uma.h>
57 
58 #include <net/route.h>
59 #include <net/if.h>
60 #include <net/if_var.h>
61 #include <net/vnet.h>
62 
63 #include <netinet/in.h>
64 #include <netinet/in_kdtrace.h>
65 #include <netinet/in_pcb.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/in_var.h>
68 #include <netinet/ip.h>
69 #include <netinet/ip_icmp.h>
70 #include <netinet/ip_var.h>
71 #ifdef INET6
72 #include <netinet/ip6.h>
73 #include <netinet6/in6_pcb.h>
74 #include <netinet6/ip6_var.h>
75 #include <netinet6/scope6_var.h>
76 #include <netinet6/nd6.h>
77 #endif
78 #include <netinet/tcp.h>
79 #include <netinet/tcp_fsm.h>
80 #include <netinet/tcp_seq.h>
81 #include <netinet/tcp_timer.h>
82 #include <netinet/tcp_var.h>
83 #include <netinet/tcpip.h>
84 
85 #include <netinet/udp.h>
86 #include <netinet/udp_var.h>
87 
88 #include <netipsec/ipsec_support.h>
89 
90 #include <machine/in_cksum.h>
91 
92 #include <security/mac/mac_framework.h>
93 
94 VNET_DEFINE_STATIC(bool, nolocaltimewait) = true;
95 #define	V_nolocaltimewait	VNET(nolocaltimewait)
96 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, nolocaltimewait,
97     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nolocaltimewait), true,
98     "Do not create TCP TIME_WAIT state for local connections");
99 
100 /*
101  * Move a TCP connection into TIME_WAIT state.
102  *    inp is locked, and is unlocked before returning.
103  *
104  * This function used to free tcpcb and allocate a compressed TCP time-wait
105  * structure tcptw.  This served well for 20 years but is no longer relevant
106  * on modern machines in the modern internet.  However, the function remains
107  * so that TCP stacks require less modification and we don't burn the bridge
108  * to go back to using compressed time-wait.
109  */
110 void
111 tcp_twstart(struct tcpcb *tp)
112 {
113 	struct inpcb *inp = tptoinpcb(tp);
114 #ifdef INET6
115 	bool isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
116 #endif
117 
118 	NET_EPOCH_ASSERT();
119 	INP_WLOCK_ASSERT(inp);
120 
121 	/* A dropped inp should never transition to TIME_WAIT state. */
122 	KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: "
123 	    "(inp->inp_flags & INP_DROPPED) != 0"));
124 
125 	tcp_state_change(tp, TCPS_TIME_WAIT);
126 	soisdisconnected(inp->inp_socket);
127 
128 	if (tp->t_flags & TF_ACKNOW)
129 		tcp_output(tp);
130 
131 	if (V_nolocaltimewait && (
132 #ifdef INET6
133 	    isipv6 ? in6_localaddr(&inp->in6p_faddr) :
134 #endif
135 #ifdef INET
136 	    in_localip(inp->inp_faddr)
137 #else
138 	    false
139 #endif
140 	    )) {
141 		if ((tp = tcp_close(tp)) != NULL)
142 			INP_WUNLOCK(inp);
143 		return;
144 	}
145 
146 	tcp_timer_activate(tp, TT_2MSL, 2 * V_tcp_msl);
147 	INP_WUNLOCK(inp);
148 }
149 
150 /*
151  * Returns true if the TIME_WAIT state was killed and we should start over,
152  * looking for a pcb in the listen state.  Otherwise returns false and frees
153  * the mbuf.
154  *
155  * For pure SYN-segments the PCB shall be read-locked and the tcpopt pointer
156  * may be NULL.  For the rest write-lock and valid tcpopt.
157  */
158 bool
159 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
160     struct mbuf *m, int tlen)
161 {
162 	struct tcpcb *tp = intotcpcb(inp);
163 	char *s;
164 	int thflags;
165 	tcp_seq seq;
166 
167 	NET_EPOCH_ASSERT();
168 	INP_LOCK_ASSERT(inp);
169 
170 	thflags = tcp_get_flags(th);
171 #ifdef INVARIANTS
172 	if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN)
173 		INP_RLOCK_ASSERT(inp);
174 	else {
175 		INP_WLOCK_ASSERT(inp);
176 		KASSERT(to != NULL,
177 		    ("%s: called without options on a non-SYN segment",
178 		    __func__));
179 	}
180 #endif
181 
182 	/*
183 	 * NOTE: for FIN_WAIT_2 (to be added later),
184 	 * must validate sequence number before accepting RST
185 	 */
186 
187 	/*
188 	 * If the segment contains RST:
189 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
190 	 *      RFC 1337.
191 	 */
192 	if (thflags & TH_RST)
193 		goto drop;
194 
195 #if 0
196 /* PAWS not needed at the moment */
197 	/*
198 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
199 	 * and it's less than ts_recent, drop it.
200 	 */
201 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
202 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
203 		if ((thflags & TH_ACK) == 0)
204 			goto drop;
205 		goto ack;
206 	}
207 	/*
208 	 * ts_recent is never updated because we never accept new segments.
209 	 */
210 #endif
211 
212 	/* Honor the drop_synfin sysctl variable. */
213 	if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
214 		if ((s = tcp_log_addrs(&inp->inp_inc, th, NULL, NULL))) {
215 			log(LOG_DEBUG, "%s; %s: "
216 			    "SYN|FIN segment ignored (based on "
217 			    "sysctl setting)\n", s, __func__);
218 			free(s, M_TCPLOG);
219 		}
220 		goto drop;
221 	}
222 
223 	/*
224 	 * If a new connection request is received
225 	 * while in TIME_WAIT, drop the old connection
226 	 * and start over if the sequence numbers
227 	 * are above the previous ones.
228 	 * Allow UDP port number changes in this case.
229 	 */
230 	if (((thflags & (TH_SYN | TH_ACK)) == TH_SYN) &&
231 	    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
232 		/*
233 		 * In case we can't upgrade our lock just pretend we have
234 		 * lost this packet.
235 		 */
236 		if (INP_TRY_UPGRADE(inp) == 0)
237 			goto drop;
238 		if ((tp = tcp_close(tp)) != NULL)
239 			INP_WUNLOCK(inp);
240 		TCPSTAT_INC(tcps_tw_recycles);
241 		return (true);
242 	}
243 
244 	/*
245 	 * Send RST if UDP port numbers don't match
246 	 */
247 	if (tp->t_port != m->m_pkthdr.tcp_tun_port) {
248 		if (tcp_get_flags(th) & TH_ACK) {
249 			tcp_respond(tp, mtod(m, void *), th, m,
250 			    (tcp_seq)0, th->th_ack, TH_RST);
251 		} else {
252 			if (tcp_get_flags(th) & TH_SYN)
253 				tlen++;
254 			if (tcp_get_flags(th) & TH_FIN)
255 				tlen++;
256 			tcp_respond(tp, mtod(m, void *), th, m,
257 			    th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK);
258 		}
259 		INP_UNLOCK(inp);
260 		TCPSTAT_INC(tcps_tw_resets);
261 		return (false);
262 	}
263 
264 	/*
265 	 * Drop the segment if it does not contain an ACK.
266 	 */
267 	if ((thflags & TH_ACK) == 0)
268 		goto drop;
269 
270 	INP_WLOCK_ASSERT(inp);
271 
272 	/*
273 	 * If timestamps were negotiated during SYN/ACK and a
274 	 * segment without a timestamp is received, silently drop
275 	 * the segment, unless the missing timestamps are tolerated.
276 	 * See section 3.2 of RFC 7323.
277 	 */
278 	if (((to->to_flags & TOF_TS) == 0) && (tp->ts_recent != 0) &&
279 	    (V_tcp_tolerate_missing_ts == 0)) {
280 		goto drop;
281 	}
282 
283 	/*
284 	 * Reset the 2MSL timer if this is a duplicate FIN.
285 	 */
286 	if (thflags & TH_FIN) {
287 		seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
288 		if (seq + 1 == tp->rcv_nxt)
289 			tcp_timer_activate(tp, TT_2MSL, 2 * V_tcp_msl);
290 	}
291 
292 	/*
293 	 * Acknowledge the segment if it has data or is not a duplicate ACK.
294 	 */
295 	if (thflags != TH_ACK || tlen != 0 ||
296 	    th->th_seq != tp->rcv_nxt || th->th_ack != tp->snd_nxt) {
297 		TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
298 		tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
299 		    tp->snd_nxt, TH_ACK);
300 		INP_UNLOCK(inp);
301 		TCPSTAT_INC(tcps_tw_responds);
302 		return (false);
303 	}
304 drop:
305 	TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
306 	INP_UNLOCK(inp);
307 	m_freem(m);
308 	return (false);
309 }
310