xref: /freebsd/sys/netinet/tcp_timewait.c (revision 0572ccaa4543b0abef8ef81e384c1d04de9f3da1)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)tcp_subr.c	8.2 (Berkeley) 5/24/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 #include "opt_tcpdebug.h"
38 
39 #include <sys/param.h>
40 #include <sys/systm.h>
41 #include <sys/callout.h>
42 #include <sys/kernel.h>
43 #include <sys/sysctl.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/priv.h>
47 #include <sys/proc.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/protosw.h>
51 #include <sys/random.h>
52 #include <sys/refcount.h>
53 
54 #include <vm/uma.h>
55 
56 #include <net/route.h>
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/vnet.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_pcb.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/in_var.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_icmp.h>
67 #include <netinet/ip_var.h>
68 #ifdef INET6
69 #include <netinet/ip6.h>
70 #include <netinet6/in6_pcb.h>
71 #include <netinet6/ip6_var.h>
72 #include <netinet6/scope6_var.h>
73 #include <netinet6/nd6.h>
74 #endif
75 #include <netinet/tcp.h>
76 #include <netinet/tcp_fsm.h>
77 #include <netinet/tcp_seq.h>
78 #include <netinet/tcp_timer.h>
79 #include <netinet/tcp_var.h>
80 #ifdef INET6
81 #include <netinet6/tcp6_var.h>
82 #endif
83 #include <netinet/tcpip.h>
84 #ifdef TCPDEBUG
85 #include <netinet/tcp_debug.h>
86 #endif
87 #ifdef INET6
88 #include <netinet6/ip6protosw.h>
89 #endif
90 
91 #include <machine/in_cksum.h>
92 
93 #include <security/mac/mac_framework.h>
94 
95 static VNET_DEFINE(uma_zone_t, tcptw_zone);
96 #define	V_tcptw_zone		VNET(tcptw_zone)
97 static int	maxtcptw;
98 
99 /*
100  * The timed wait queue contains references to each of the TCP sessions
101  * currently in the TIME_WAIT state.  The queue pointers, including the
102  * queue pointers in each tcptw structure, are protected using the global
103  * timewait lock, which must be held over queue iteration and modification.
104  */
105 static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl);
106 #define	V_twq_2msl		VNET(twq_2msl)
107 
108 /* Global timewait lock */
109 static VNET_DEFINE(struct rwlock, tw_lock);
110 #define	V_tw_lock		VNET(tw_lock)
111 
112 #define	TW_LOCK_INIT(tw, d)	rw_init_flags(&(tw), (d), 0)
113 #define	TW_LOCK_DESTROY(tw)	rw_destroy(&(tw))
114 #define	TW_RLOCK(tw)		rw_rlock(&(tw))
115 #define	TW_WLOCK(tw)		rw_wlock(&(tw))
116 #define	TW_RUNLOCK(tw)		rw_runlock(&(tw))
117 #define	TW_WUNLOCK(tw)		rw_wunlock(&(tw))
118 #define	TW_LOCK_ASSERT(tw)	rw_assert(&(tw), RA_LOCKED)
119 #define	TW_RLOCK_ASSERT(tw)	rw_assert(&(tw), RA_RLOCKED)
120 #define	TW_WLOCK_ASSERT(tw)	rw_assert(&(tw), RA_WLOCKED)
121 #define	TW_UNLOCK_ASSERT(tw)	rw_assert(&(tw), RA_UNLOCKED)
122 
123 static void	tcp_tw_2msl_reset(struct tcptw *, int);
124 static void	tcp_tw_2msl_stop(struct tcptw *, int);
125 
126 /*
127  * tw_pcbref() bumps the reference count on an tw in order to maintain
128  * stability of an tw pointer despite the tw lock being released.
129  */
130 static void
131 tw_pcbref(struct tcptw *tw)
132 {
133 
134 	KASSERT(tw->tw_refcount > 0, ("%s: refcount 0", __func__));
135 	refcount_acquire(&tw->tw_refcount);
136 }
137 
138 /*
139  * Drop a refcount on an tw elevated using tw_pcbref().
140  */
141 static int
142 tw_pcbrele(struct tcptw *tw)
143 {
144 
145 	KASSERT(tw->tw_refcount > 0, ("%s: refcount 0", __func__));
146 	if (!refcount_release(&tw->tw_refcount))
147 		return (0);
148 	uma_zfree(V_tcptw_zone, tw);
149 	return (1);
150 }
151 
152 static int
153 tcptw_auto_size(void)
154 {
155 	int halfrange;
156 
157 	/*
158 	 * Max out at half the ephemeral port range so that TIME_WAIT
159 	 * sockets don't tie up too many ephemeral ports.
160 	 */
161 	if (V_ipport_lastauto > V_ipport_firstauto)
162 		halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
163 	else
164 		halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
165 	/* Protect against goofy port ranges smaller than 32. */
166 	return (imin(imax(halfrange, 32), maxsockets / 5));
167 }
168 
169 static int
170 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
171 {
172 	int error, new;
173 
174 	if (maxtcptw == 0)
175 		new = tcptw_auto_size();
176 	else
177 		new = maxtcptw;
178 	error = sysctl_handle_int(oidp, &new, 0, req);
179 	if (error == 0 && req->newptr)
180 		if (new >= 32) {
181 			maxtcptw = new;
182 			uma_zone_set_max(V_tcptw_zone, maxtcptw);
183 		}
184 	return (error);
185 }
186 
187 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
188     &maxtcptw, 0, sysctl_maxtcptw, "IU",
189     "Maximum number of compressed TCP TIME_WAIT entries");
190 
191 VNET_DEFINE(int, nolocaltimewait) = 0;
192 #define	V_nolocaltimewait	VNET(nolocaltimewait)
193 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
194     &VNET_NAME(nolocaltimewait), 0,
195     "Do not create compressed TCP TIME_WAIT entries for local connections");
196 
197 void
198 tcp_tw_zone_change(void)
199 {
200 
201 	if (maxtcptw == 0)
202 		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
203 }
204 
205 void
206 tcp_tw_init(void)
207 {
208 
209 	V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
210 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
211 	TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
212 	if (maxtcptw == 0)
213 		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
214 	else
215 		uma_zone_set_max(V_tcptw_zone, maxtcptw);
216 	TAILQ_INIT(&V_twq_2msl);
217 	TW_LOCK_INIT(V_tw_lock, "tcptw");
218 }
219 
220 #ifdef VIMAGE
221 void
222 tcp_tw_destroy(void)
223 {
224 	struct tcptw *tw;
225 
226 	INP_INFO_WLOCK(&V_tcbinfo);
227 	while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
228 		tcp_twclose(tw, 0);
229 	INP_INFO_WUNLOCK(&V_tcbinfo);
230 
231 	TW_LOCK_DESTROY(V_tw_lock);
232 	uma_zdestroy(V_tcptw_zone);
233 }
234 #endif
235 
236 /*
237  * Move a TCP connection into TIME_WAIT state.
238  *    tcbinfo is locked.
239  *    inp is locked, and is unlocked before returning.
240  */
241 void
242 tcp_twstart(struct tcpcb *tp)
243 {
244 	struct tcptw *tw;
245 	struct inpcb *inp = tp->t_inpcb;
246 	int acknow;
247 	struct socket *so;
248 #ifdef INET6
249 	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
250 #endif
251 
252 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
253 	INP_WLOCK_ASSERT(inp);
254 
255 	if (V_nolocaltimewait) {
256 		int error = 0;
257 #ifdef INET6
258 		if (isipv6)
259 			error = in6_localaddr(&inp->in6p_faddr);
260 #endif
261 #if defined(INET6) && defined(INET)
262 		else
263 #endif
264 #ifdef INET
265 			error = in_localip(inp->inp_faddr);
266 #endif
267 		if (error) {
268 			tp = tcp_close(tp);
269 			if (tp != NULL)
270 				INP_WUNLOCK(inp);
271 			return;
272 		}
273 	}
274 
275 	tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
276 	if (tw == NULL) {
277 		tw = tcp_tw_2msl_reuse();
278 		if (tw == NULL) {
279 			tp = tcp_close(tp);
280 			if (tp != NULL)
281 				INP_WUNLOCK(inp);
282 			return;
283 		}
284 	}
285 	tw->tw_inpcb = inp;
286 	refcount_init(&tw->tw_refcount, 1);
287 
288 	/*
289 	 * Recover last window size sent.
290 	 */
291 	if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
292 		tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
293 	else
294 		tw->last_win = 0;
295 
296 	/*
297 	 * Set t_recent if timestamps are used on the connection.
298 	 */
299 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
300 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
301 		tw->t_recent = tp->ts_recent;
302 		tw->ts_offset = tp->ts_offset;
303 	} else {
304 		tw->t_recent = 0;
305 		tw->ts_offset = 0;
306 	}
307 
308 	tw->snd_nxt = tp->snd_nxt;
309 	tw->rcv_nxt = tp->rcv_nxt;
310 	tw->iss     = tp->iss;
311 	tw->irs     = tp->irs;
312 	tw->t_starttime = tp->t_starttime;
313 	tw->tw_time = 0;
314 
315 /* XXX
316  * If this code will
317  * be used for fin-wait-2 state also, then we may need
318  * a ts_recent from the last segment.
319  */
320 	acknow = tp->t_flags & TF_ACKNOW;
321 
322 	/*
323 	 * First, discard tcpcb state, which includes stopping its timers and
324 	 * freeing it.  tcp_discardcb() used to also release the inpcb, but
325 	 * that work is now done in the caller.
326 	 *
327 	 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
328 	 * and might not be needed here any longer.
329 	 */
330 	tcp_discardcb(tp);
331 	so = inp->inp_socket;
332 	soisdisconnected(so);
333 	tw->tw_cred = crhold(so->so_cred);
334 	SOCK_LOCK(so);
335 	tw->tw_so_options = so->so_options;
336 	SOCK_UNLOCK(so);
337 	if (acknow)
338 		tcp_twrespond(tw, TH_ACK);
339 	inp->inp_ppcb = tw;
340 	inp->inp_flags |= INP_TIMEWAIT;
341 	tcp_tw_2msl_reset(tw, 0);
342 
343 	/*
344 	 * If the inpcb owns the sole reference to the socket, then we can
345 	 * detach and free the socket as it is not needed in time wait.
346 	 */
347 	if (inp->inp_flags & INP_SOCKREF) {
348 		KASSERT(so->so_state & SS_PROTOREF,
349 		    ("tcp_twstart: !SS_PROTOREF"));
350 		inp->inp_flags &= ~INP_SOCKREF;
351 		INP_WUNLOCK(inp);
352 		ACCEPT_LOCK();
353 		SOCK_LOCK(so);
354 		so->so_state &= ~SS_PROTOREF;
355 		sofree(so);
356 	} else
357 		INP_WUNLOCK(inp);
358 }
359 
360 /*
361  * Returns 1 if the TIME_WAIT state was killed and we should start over,
362  * looking for a pcb in the listen state.  Returns 0 otherwise.
363  */
364 int
365 tcp_twcheck(struct inpcb *inp, struct tcpopt *to, struct tcphdr *th,
366     struct mbuf *m, int tlen)
367 {
368 	struct tcptw *tw;
369 	int thflags;
370 	tcp_seq seq;
371 
372 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
373 	INP_WLOCK_ASSERT(inp);
374 
375 	/*
376 	 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
377 	 * still present.  This is undesirable, but temporarily necessary
378 	 * until we work out how to handle inpcb's who's timewait state has
379 	 * been removed.
380 	 */
381 	tw = intotw(inp);
382 	if (tw == NULL)
383 		goto drop;
384 
385 	thflags = th->th_flags;
386 
387 	/*
388 	 * NOTE: for FIN_WAIT_2 (to be added later),
389 	 * must validate sequence number before accepting RST
390 	 */
391 
392 	/*
393 	 * If the segment contains RST:
394 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
395 	 *      RFC 1337.
396 	 */
397 	if (thflags & TH_RST)
398 		goto drop;
399 
400 #if 0
401 /* PAWS not needed at the moment */
402 	/*
403 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
404 	 * and it's less than ts_recent, drop it.
405 	 */
406 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
407 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
408 		if ((thflags & TH_ACK) == 0)
409 			goto drop;
410 		goto ack;
411 	}
412 	/*
413 	 * ts_recent is never updated because we never accept new segments.
414 	 */
415 #endif
416 
417 	/*
418 	 * If a new connection request is received
419 	 * while in TIME_WAIT, drop the old connection
420 	 * and start over if the sequence numbers
421 	 * are above the previous ones.
422 	 */
423 	if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
424 		tcp_twclose(tw, 0);
425 		return (1);
426 	}
427 
428 	/*
429 	 * Drop the segment if it does not contain an ACK.
430 	 */
431 	if ((thflags & TH_ACK) == 0)
432 		goto drop;
433 
434 	/*
435 	 * Reset the 2MSL timer if this is a duplicate FIN.
436 	 */
437 	if (thflags & TH_FIN) {
438 		seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
439 		if (seq + 1 == tw->rcv_nxt)
440 			tcp_tw_2msl_reset(tw, 1);
441 	}
442 
443 	/*
444 	 * Acknowledge the segment if it has data or is not a duplicate ACK.
445 	 */
446 	if (thflags != TH_ACK || tlen != 0 ||
447 	    th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
448 		tcp_twrespond(tw, TH_ACK);
449 drop:
450 	INP_WUNLOCK(inp);
451 	m_freem(m);
452 	return (0);
453 }
454 
455 void
456 tcp_twclose(struct tcptw *tw, int reuse)
457 {
458 	struct socket *so;
459 	struct inpcb *inp;
460 
461 	/*
462 	 * At this point, we are in one of two situations:
463 	 *
464 	 * (1) We have no socket, just an inpcb<->twtcp pair.  We can free
465 	 *     all state.
466 	 *
467 	 * (2) We have a socket -- if we own a reference, release it and
468 	 *     notify the socket layer.
469 	 */
470 	inp = tw->tw_inpcb;
471 	KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
472 	KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
473 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);	/* in_pcbfree() */
474 	INP_WLOCK_ASSERT(inp);
475 
476 	tw->tw_inpcb = NULL;
477 	tcp_tw_2msl_stop(tw, reuse);
478 	inp->inp_ppcb = NULL;
479 	in_pcbdrop(inp);
480 
481 	so = inp->inp_socket;
482 	if (so != NULL) {
483 		/*
484 		 * If there's a socket, handle two cases: first, we own a
485 		 * strong reference, which we will now release, or we don't
486 		 * in which case another reference exists (XXXRW: think
487 		 * about this more), and we don't need to take action.
488 		 */
489 		if (inp->inp_flags & INP_SOCKREF) {
490 			inp->inp_flags &= ~INP_SOCKREF;
491 			INP_WUNLOCK(inp);
492 			ACCEPT_LOCK();
493 			SOCK_LOCK(so);
494 			KASSERT(so->so_state & SS_PROTOREF,
495 			    ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
496 			so->so_state &= ~SS_PROTOREF;
497 			sofree(so);
498 		} else {
499 			/*
500 			 * If we don't own the only reference, the socket and
501 			 * inpcb need to be left around to be handled by
502 			 * tcp_usr_detach() later.
503 			 */
504 			INP_WUNLOCK(inp);
505 		}
506 	} else
507 		in_pcbfree(inp);
508 	TCPSTAT_INC(tcps_closed);
509 }
510 
511 int
512 tcp_twrespond(struct tcptw *tw, int flags)
513 {
514 	struct inpcb *inp = tw->tw_inpcb;
515 #if defined(INET6) || defined(INET)
516 	struct tcphdr *th = NULL;
517 #endif
518 	struct mbuf *m;
519 #ifdef INET
520 	struct ip *ip = NULL;
521 #endif
522 	u_int hdrlen, optlen;
523 	int error = 0;			/* Keep compiler happy */
524 	struct tcpopt to;
525 #ifdef INET6
526 	struct ip6_hdr *ip6 = NULL;
527 	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
528 #endif
529 	hdrlen = 0;                     /* Keep compiler happy */
530 
531 	INP_WLOCK_ASSERT(inp);
532 
533 	m = m_gethdr(M_NOWAIT, MT_DATA);
534 	if (m == NULL)
535 		return (ENOBUFS);
536 	m->m_data += max_linkhdr;
537 
538 #ifdef MAC
539 	mac_inpcb_create_mbuf(inp, m);
540 #endif
541 
542 #ifdef INET6
543 	if (isipv6) {
544 		hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
545 		ip6 = mtod(m, struct ip6_hdr *);
546 		th = (struct tcphdr *)(ip6 + 1);
547 		tcpip_fillheaders(inp, ip6, th);
548 	}
549 #endif
550 #if defined(INET6) && defined(INET)
551 	else
552 #endif
553 #ifdef INET
554 	{
555 		hdrlen = sizeof(struct tcpiphdr);
556 		ip = mtod(m, struct ip *);
557 		th = (struct tcphdr *)(ip + 1);
558 		tcpip_fillheaders(inp, ip, th);
559 	}
560 #endif
561 	to.to_flags = 0;
562 
563 	/*
564 	 * Send a timestamp and echo-reply if both our side and our peer
565 	 * have sent timestamps in our SYN's and this is not a RST.
566 	 */
567 	if (tw->t_recent && flags == TH_ACK) {
568 		to.to_flags |= TOF_TS;
569 		to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
570 		to.to_tsecr = tw->t_recent;
571 	}
572 	optlen = tcp_addoptions(&to, (u_char *)(th + 1));
573 
574 	m->m_len = hdrlen + optlen;
575 	m->m_pkthdr.len = m->m_len;
576 
577 	KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
578 
579 	th->th_seq = htonl(tw->snd_nxt);
580 	th->th_ack = htonl(tw->rcv_nxt);
581 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
582 	th->th_flags = flags;
583 	th->th_win = htons(tw->last_win);
584 
585 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
586 #ifdef INET6
587 	if (isipv6) {
588 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
589 		th->th_sum = in6_cksum_pseudo(ip6,
590 		    sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
591 		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
592 		error = ip6_output(m, inp->in6p_outputopts, NULL,
593 		    (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
594 	}
595 #endif
596 #if defined(INET6) && defined(INET)
597 	else
598 #endif
599 #ifdef INET
600 	{
601 		m->m_pkthdr.csum_flags = CSUM_TCP;
602 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
603 		    htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
604 		ip->ip_len = htons(m->m_pkthdr.len);
605 		if (V_path_mtu_discovery)
606 			ip->ip_off |= htons(IP_DF);
607 		error = ip_output(m, inp->inp_options, NULL,
608 		    ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
609 		    NULL, inp);
610 	}
611 #endif
612 	if (flags & TH_ACK)
613 		TCPSTAT_INC(tcps_sndacks);
614 	else
615 		TCPSTAT_INC(tcps_sndctrl);
616 	TCPSTAT_INC(tcps_sndtotal);
617 	return (error);
618 }
619 
620 static void
621 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
622 {
623 
624 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
625 	INP_WLOCK_ASSERT(tw->tw_inpcb);
626 
627 	TW_WLOCK(V_tw_lock);
628 	if (rearm)
629 		TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
630 	tw->tw_time = ticks + 2 * tcp_msl;
631 	TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
632 	TW_WUNLOCK(V_tw_lock);
633 }
634 
635 static void
636 tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
637 {
638 
639 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
640 
641 	TW_WLOCK(V_tw_lock);
642 	TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
643 	crfree(tw->tw_cred);
644 	tw->tw_cred = NULL;
645 	TW_WUNLOCK(V_tw_lock);
646 
647 	if (!reuse)
648 		tw_pcbrele(tw);
649 }
650 
651 struct tcptw *
652 tcp_tw_2msl_reuse(void)
653 {
654 	struct tcptw *tw;
655 
656 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
657 
658 	TW_WLOCK(V_tw_lock);
659 	tw = TAILQ_FIRST(&V_twq_2msl);
660 	if (tw == NULL) {
661 		TW_WUNLOCK(V_tw_lock);
662 		return NULL;
663 	}
664 	TW_WUNLOCK(V_tw_lock);
665 
666 	INP_WLOCK(tw->tw_inpcb);
667 	tcp_twclose(tw, 1);
668 
669 	return (tw);
670 }
671 
672 void
673 tcp_tw_2msl_scan(void)
674 {
675 	struct tcptw *tw;
676 
677 	for (;;) {
678 		TW_RLOCK(V_tw_lock);
679 		tw = TAILQ_FIRST(&V_twq_2msl);
680 		if (tw == NULL || tw->tw_time - ticks > 0) {
681 			TW_RUNLOCK(V_tw_lock);
682 			break;
683 		}
684 		tw_pcbref(tw);
685 		TW_RUNLOCK(V_tw_lock);
686 
687 		/* Close timewait state */
688 		if (INP_INFO_TRY_WLOCK(&V_tcbinfo)) {
689 			if (tw_pcbrele(tw)) {
690 				INP_INFO_WUNLOCK(&V_tcbinfo);
691 				continue;
692 			}
693 
694 			KASSERT(tw->tw_inpcb != NULL,
695 			    ("%s: tw->tw_inpcb == NULL", __func__));
696 			INP_WLOCK(tw->tw_inpcb);
697 			tcp_twclose(tw, 0);
698 			INP_INFO_WUNLOCK(&V_tcbinfo);
699 		} else {
700 			/* INP_INFO lock is busy; continue later. */
701 			tw_pcbrele(tw);
702 			break;
703 		}
704 	}
705 }
706