xref: /freebsd/sys/netinet/tcp_timewait.c (revision 058ac3e8063366dafa634d9107642e12b038bf09)
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/tcp_hpts.h>
84 #include <netinet/tcpip.h>
85 
86 #include <netinet/udp.h>
87 #include <netinet/udp_var.h>
88 
89 #include <netipsec/ipsec_support.h>
90 
91 #include <machine/in_cksum.h>
92 
93 #include <security/mac/mac_framework.h>
94 
95 VNET_DEFINE_STATIC(bool, nolocaltimewait) = true;
96 #define	V_nolocaltimewait	VNET(nolocaltimewait)
97 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, nolocaltimewait,
98     CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(nolocaltimewait), true,
99     "Do not create TCP TIME_WAIT state for local connections");
100 
101 /*
102  * Move a TCP connection into TIME_WAIT state.
103  *    inp is locked, and is unlocked before returning.
104  *
105  * This function used to free tcpcb and allocate a compressed TCP time-wait
106  * structure tcptw.  This served well for 20 years but is no longer relevant
107  * on modern machines in the modern internet.  However, the function remains
108  * so that TCP stacks require less modification and we don't burn the bridge
109  * to go back to using compressed time-wait.
110  */
111 void
112 tcp_twstart(struct tcpcb *tp)
113 {
114 	struct inpcb *inp = tptoinpcb(tp);
115 #ifdef INET6
116 	bool isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
117 #endif
118 
119 	NET_EPOCH_ASSERT();
120 	INP_WLOCK_ASSERT(inp);
121 
122 	/* A dropped inp should never transition to TIME_WAIT state. */
123 	KASSERT((inp->inp_flags & INP_DROPPED) == 0, ("tcp_twstart: "
124 	    "(inp->inp_flags & INP_DROPPED) != 0"));
125 
126 	tcp_state_change(tp, TCPS_TIME_WAIT);
127 	soisdisconnected(inp->inp_socket);
128 
129 	if (tp->t_flags & TF_ACKNOW)
130 		tcp_output(tp);
131 
132 	if (V_nolocaltimewait && (
133 #ifdef INET6
134 	    isipv6 ? in6_localaddr(&inp->in6p_faddr) :
135 #endif
136 #ifdef INET
137 	    in_localip(inp->inp_faddr)
138 #else
139 	    false
140 #endif
141 	    )) {
142 		if ((tp = tcp_close(tp)) != NULL)
143 			INP_WUNLOCK(inp);
144 		return;
145 	}
146 
147 	tcp_timer_activate(tp, TT_2MSL, 2 * V_tcp_msl);
148 	INP_WUNLOCK(inp);
149 }
150 
151 /*
152  * Returns true if the TIME_WAIT state was killed and we should start over,
153  * looking for a pcb in the listen state.  Otherwise returns false and frees
154  * the mbuf.
155  *
156  * For pure SYN-segments the PCB shall be read-locked and the tcpopt pointer
157  * may be NULL.  For the rest write-lock and valid tcpopt.
158  */
159 bool
160 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
161     struct mbuf *m, int tlen)
162 {
163 	struct tcpcb *tp = intotcpcb(inp);
164 	char *s;
165 	int thflags;
166 	tcp_seq seq;
167 
168 	NET_EPOCH_ASSERT();
169 	INP_LOCK_ASSERT(inp);
170 
171 	thflags = tcp_get_flags(th);
172 #ifdef INVARIANTS
173 	if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN)
174 		INP_RLOCK_ASSERT(inp);
175 	else {
176 		INP_WLOCK_ASSERT(inp);
177 		KASSERT(to != NULL,
178 		    ("%s: called without options on a non-SYN segment",
179 		    __func__));
180 	}
181 #endif
182 
183 	/*
184 	 * NOTE: for FIN_WAIT_2 (to be added later),
185 	 * must validate sequence number before accepting RST
186 	 */
187 
188 	/*
189 	 * If the segment contains RST:
190 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
191 	 *      RFC 1337.
192 	 */
193 	if (thflags & TH_RST)
194 		goto drop;
195 
196 #if 0
197 /* PAWS not needed at the moment */
198 	/*
199 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
200 	 * and it's less than ts_recent, drop it.
201 	 */
202 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
203 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
204 		if ((thflags & TH_ACK) == 0)
205 			goto drop;
206 		goto ack;
207 	}
208 	/*
209 	 * ts_recent is never updated because we never accept new segments.
210 	 */
211 #endif
212 
213 	/* Honor the drop_synfin sysctl variable. */
214 	if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
215 		if ((s = tcp_log_addrs(&inp->inp_inc, th, NULL, NULL))) {
216 			log(LOG_DEBUG, "%s; %s: "
217 			    "SYN|FIN segment ignored (based on "
218 			    "sysctl setting)\n", s, __func__);
219 			free(s, M_TCPLOG);
220 		}
221 		goto drop;
222 	}
223 
224 	/*
225 	 * If a new connection request is received
226 	 * while in TIME_WAIT, drop the old connection
227 	 * and start over if the sequence numbers
228 	 * are above the previous ones.
229 	 * Allow UDP port number changes in this case.
230 	 */
231 	if (((thflags & (TH_SYN | TH_ACK)) == TH_SYN) &&
232 	    SEQ_GT(th->th_seq, tp->rcv_nxt)) {
233 		/*
234 		 * In case we can't upgrade our lock just pretend we have
235 		 * lost this packet.
236 		 */
237 		if (INP_TRY_UPGRADE(inp) == 0)
238 			goto drop;
239 		if ((tp = tcp_close(tp)) != NULL)
240 			INP_WUNLOCK(inp);
241 		TCPSTAT_INC(tcps_tw_recycles);
242 		return (true);
243 	}
244 
245 	/*
246 	 * Send RST if UDP port numbers don't match
247 	 */
248 	if (tp->t_port != m->m_pkthdr.tcp_tun_port) {
249 		if (tcp_get_flags(th) & TH_ACK) {
250 			tcp_respond(tp, mtod(m, void *), th, m,
251 			    (tcp_seq)0, th->th_ack, TH_RST);
252 		} else {
253 			if (tcp_get_flags(th) & TH_SYN)
254 				tlen++;
255 			if (tcp_get_flags(th) & TH_FIN)
256 				tlen++;
257 			tcp_respond(tp, mtod(m, void *), th, m,
258 			    th->th_seq+tlen, (tcp_seq)0, TH_RST|TH_ACK);
259 		}
260 		INP_UNLOCK(inp);
261 		TCPSTAT_INC(tcps_tw_resets);
262 		return (false);
263 	}
264 
265 	/*
266 	 * Drop the segment if it does not contain an ACK.
267 	 */
268 	if ((thflags & TH_ACK) == 0)
269 		goto drop;
270 
271 	INP_WLOCK_ASSERT(inp);
272 
273 	/*
274 	 * If timestamps were negotiated during SYN/ACK and a
275 	 * segment without a timestamp is received, silently drop
276 	 * the segment, unless the missing timestamps are tolerated.
277 	 * See section 3.2 of RFC 7323.
278 	 */
279 	if (((to->to_flags & TOF_TS) == 0) && (tp->ts_recent != 0) &&
280 	    (V_tcp_tolerate_missing_ts == 0)) {
281 		goto drop;
282 	}
283 
284 	/*
285 	 * Reset the 2MSL timer if this is a duplicate FIN.
286 	 */
287 	if (thflags & TH_FIN) {
288 		seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
289 		if (seq + 1 == tp->rcv_nxt)
290 			tcp_timer_activate(tp, TT_2MSL, 2 * V_tcp_msl);
291 	}
292 
293 	/*
294 	 * Acknowledge the segment if it has data or is not a duplicate ACK.
295 	 */
296 	if (thflags != TH_ACK || tlen != 0 ||
297 	    th->th_seq != tp->rcv_nxt || th->th_ack != tp->snd_nxt) {
298 		TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
299 		tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
300 		    tp->snd_nxt, TH_ACK);
301 		INP_UNLOCK(inp);
302 		TCPSTAT_INC(tcps_tw_responds);
303 		return (false);
304 	}
305 drop:
306 	TCP_PROBE5(receive, NULL, NULL, m, NULL, th);
307 	INP_UNLOCK(inp);
308 	m_freem(m);
309 	return (false);
310 }
311