xref: /freebsd/sys/netinet/tcp_timewait.c (revision 3823d5e198425b4f5e5a80267d195769d1063773)
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 static int	tcp_twrespond(struct tcptw *, int);
126 
127 /*
128  * tw_pcbref() bumps the reference count on an tw in order to maintain
129  * stability of an tw pointer despite the tw lock being released.
130  */
131 static void
132 tw_pcbref(struct tcptw *tw)
133 {
134 
135 	KASSERT(tw->tw_refcount > 0, ("%s: refcount 0", __func__));
136 	refcount_acquire(&tw->tw_refcount);
137 }
138 
139 /*
140  * Drop a refcount on an tw elevated using tw_pcbref().
141  */
142 static int
143 tw_pcbrele(struct tcptw *tw)
144 {
145 
146 	KASSERT(tw->tw_refcount > 0, ("%s: refcount 0", __func__));
147 	if (!refcount_release(&tw->tw_refcount))
148 		return (0);
149 	uma_zfree(V_tcptw_zone, tw);
150 	return (1);
151 }
152 
153 static int
154 tcptw_auto_size(void)
155 {
156 	int halfrange;
157 
158 	/*
159 	 * Max out at half the ephemeral port range so that TIME_WAIT
160 	 * sockets don't tie up too many ephemeral ports.
161 	 */
162 	if (V_ipport_lastauto > V_ipport_firstauto)
163 		halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
164 	else
165 		halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
166 	/* Protect against goofy port ranges smaller than 32. */
167 	return (imin(imax(halfrange, 32), maxsockets / 5));
168 }
169 
170 static int
171 sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
172 {
173 	int error, new;
174 
175 	if (maxtcptw == 0)
176 		new = tcptw_auto_size();
177 	else
178 		new = maxtcptw;
179 	error = sysctl_handle_int(oidp, &new, 0, req);
180 	if (error == 0 && req->newptr)
181 		if (new >= 32) {
182 			maxtcptw = new;
183 			uma_zone_set_max(V_tcptw_zone, maxtcptw);
184 		}
185 	return (error);
186 }
187 
188 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
189     &maxtcptw, 0, sysctl_maxtcptw, "IU",
190     "Maximum number of compressed TCP TIME_WAIT entries");
191 
192 VNET_DEFINE(int, nolocaltimewait) = 0;
193 #define	V_nolocaltimewait	VNET(nolocaltimewait)
194 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
195     &VNET_NAME(nolocaltimewait), 0,
196     "Do not create compressed TCP TIME_WAIT entries for local connections");
197 
198 void
199 tcp_tw_zone_change(void)
200 {
201 
202 	if (maxtcptw == 0)
203 		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
204 }
205 
206 void
207 tcp_tw_init(void)
208 {
209 
210 	V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
211 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
212 	TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
213 	if (maxtcptw == 0)
214 		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
215 	else
216 		uma_zone_set_max(V_tcptw_zone, maxtcptw);
217 	TAILQ_INIT(&V_twq_2msl);
218 	TW_LOCK_INIT(V_tw_lock, "tcptw");
219 }
220 
221 #ifdef VIMAGE
222 void
223 tcp_tw_destroy(void)
224 {
225 	struct tcptw *tw;
226 
227 	INP_INFO_WLOCK(&V_tcbinfo);
228 	while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
229 		tcp_twclose(tw, 0);
230 	INP_INFO_WUNLOCK(&V_tcbinfo);
231 
232 	TW_LOCK_DESTROY(V_tw_lock);
233 	uma_zdestroy(V_tcptw_zone);
234 }
235 #endif
236 
237 /*
238  * Move a TCP connection into TIME_WAIT state.
239  *    tcbinfo is locked.
240  *    inp is locked, and is unlocked before returning.
241  */
242 void
243 tcp_twstart(struct tcpcb *tp)
244 {
245 	struct tcptw *tw;
246 	struct inpcb *inp = tp->t_inpcb;
247 	int acknow;
248 	struct socket *so;
249 #ifdef INET6
250 	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
251 #endif
252 
253 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
254 	INP_WLOCK_ASSERT(inp);
255 
256 	if (V_nolocaltimewait) {
257 		int error = 0;
258 #ifdef INET6
259 		if (isipv6)
260 			error = in6_localaddr(&inp->in6p_faddr);
261 #endif
262 #if defined(INET6) && defined(INET)
263 		else
264 #endif
265 #ifdef INET
266 			error = in_localip(inp->inp_faddr);
267 #endif
268 		if (error) {
269 			tp = tcp_close(tp);
270 			if (tp != NULL)
271 				INP_WUNLOCK(inp);
272 			return;
273 		}
274 	}
275 
276 	tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
277 	if (tw == NULL) {
278 		/*
279 		 * Reached limit on total number of TIMEWAIT connections
280 		 * allowed. Remove a connection from TIMEWAIT queue in LRU
281 		 * fashion to make room for this connection.
282 		 */
283 		tw = tcp_tw_2msl_reuse();
284 		if (tw == NULL) {
285 			tp = tcp_close(tp);
286 			if (tp != NULL)
287 				INP_WUNLOCK(inp);
288 			return;
289 		}
290 	}
291 	tw->tw_inpcb = inp;
292 	refcount_init(&tw->tw_refcount, 1);
293 
294 	/*
295 	 * Recover last window size sent.
296 	 */
297 	if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
298 		tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
299 	else
300 		tw->last_win = 0;
301 
302 	/*
303 	 * Set t_recent if timestamps are used on the connection.
304 	 */
305 	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
306 	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
307 		tw->t_recent = tp->ts_recent;
308 		tw->ts_offset = tp->ts_offset;
309 	} else {
310 		tw->t_recent = 0;
311 		tw->ts_offset = 0;
312 	}
313 
314 	tw->snd_nxt = tp->snd_nxt;
315 	tw->rcv_nxt = tp->rcv_nxt;
316 	tw->iss     = tp->iss;
317 	tw->irs     = tp->irs;
318 	tw->t_starttime = tp->t_starttime;
319 	tw->tw_time = 0;
320 
321 /* XXX
322  * If this code will
323  * be used for fin-wait-2 state also, then we may need
324  * a ts_recent from the last segment.
325  */
326 	acknow = tp->t_flags & TF_ACKNOW;
327 
328 	/*
329 	 * First, discard tcpcb state, which includes stopping its timers and
330 	 * freeing it.  tcp_discardcb() used to also release the inpcb, but
331 	 * that work is now done in the caller.
332 	 *
333 	 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
334 	 * and might not be needed here any longer.
335 	 */
336 	tcp_discardcb(tp);
337 	so = inp->inp_socket;
338 	soisdisconnected(so);
339 	tw->tw_cred = crhold(so->so_cred);
340 	SOCK_LOCK(so);
341 	tw->tw_so_options = so->so_options;
342 	SOCK_UNLOCK(so);
343 	if (acknow)
344 		tcp_twrespond(tw, TH_ACK);
345 	inp->inp_ppcb = tw;
346 	inp->inp_flags |= INP_TIMEWAIT;
347 	tcp_tw_2msl_reset(tw, 0);
348 
349 	/*
350 	 * If the inpcb owns the sole reference to the socket, then we can
351 	 * detach and free the socket as it is not needed in time wait.
352 	 */
353 	if (inp->inp_flags & INP_SOCKREF) {
354 		KASSERT(so->so_state & SS_PROTOREF,
355 		    ("tcp_twstart: !SS_PROTOREF"));
356 		inp->inp_flags &= ~INP_SOCKREF;
357 		INP_WUNLOCK(inp);
358 		ACCEPT_LOCK();
359 		SOCK_LOCK(so);
360 		so->so_state &= ~SS_PROTOREF;
361 		sofree(so);
362 	} else
363 		INP_WUNLOCK(inp);
364 }
365 
366 /*
367  * Returns 1 if the TIME_WAIT state was killed and we should start over,
368  * looking for a pcb in the listen state.  Returns 0 otherwise.
369  */
370 int
371 tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th,
372     struct mbuf *m, int tlen)
373 {
374 	struct tcptw *tw;
375 	int thflags;
376 	tcp_seq seq;
377 
378 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
379 	INP_WLOCK_ASSERT(inp);
380 
381 	/*
382 	 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
383 	 * still present.  This is undesirable, but temporarily necessary
384 	 * until we work out how to handle inpcb's who's timewait state has
385 	 * been removed.
386 	 */
387 	tw = intotw(inp);
388 	if (tw == NULL)
389 		goto drop;
390 
391 	thflags = th->th_flags;
392 
393 	/*
394 	 * NOTE: for FIN_WAIT_2 (to be added later),
395 	 * must validate sequence number before accepting RST
396 	 */
397 
398 	/*
399 	 * If the segment contains RST:
400 	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
401 	 *      RFC 1337.
402 	 */
403 	if (thflags & TH_RST)
404 		goto drop;
405 
406 #if 0
407 /* PAWS not needed at the moment */
408 	/*
409 	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
410 	 * and it's less than ts_recent, drop it.
411 	 */
412 	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
413 	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
414 		if ((thflags & TH_ACK) == 0)
415 			goto drop;
416 		goto ack;
417 	}
418 	/*
419 	 * ts_recent is never updated because we never accept new segments.
420 	 */
421 #endif
422 
423 	/*
424 	 * If a new connection request is received
425 	 * while in TIME_WAIT, drop the old connection
426 	 * and start over if the sequence numbers
427 	 * are above the previous ones.
428 	 */
429 	if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
430 		tcp_twclose(tw, 0);
431 		return (1);
432 	}
433 
434 	/*
435 	 * Drop the segment if it does not contain an ACK.
436 	 */
437 	if ((thflags & TH_ACK) == 0)
438 		goto drop;
439 
440 	/*
441 	 * Reset the 2MSL timer if this is a duplicate FIN.
442 	 */
443 	if (thflags & TH_FIN) {
444 		seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
445 		if (seq + 1 == tw->rcv_nxt)
446 			tcp_tw_2msl_reset(tw, 1);
447 	}
448 
449 	/*
450 	 * Acknowledge the segment if it has data or is not a duplicate ACK.
451 	 */
452 	if (thflags != TH_ACK || tlen != 0 ||
453 	    th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
454 		tcp_twrespond(tw, TH_ACK);
455 drop:
456 	INP_WUNLOCK(inp);
457 	m_freem(m);
458 	return (0);
459 }
460 
461 void
462 tcp_twclose(struct tcptw *tw, int reuse)
463 {
464 	struct socket *so;
465 	struct inpcb *inp;
466 
467 	/*
468 	 * At this point, we are in one of two situations:
469 	 *
470 	 * (1) We have no socket, just an inpcb<->twtcp pair.  We can free
471 	 *     all state.
472 	 *
473 	 * (2) We have a socket -- if we own a reference, release it and
474 	 *     notify the socket layer.
475 	 */
476 	inp = tw->tw_inpcb;
477 	KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
478 	KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
479 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);	/* in_pcbfree() */
480 	INP_WLOCK_ASSERT(inp);
481 
482 	tw->tw_inpcb = NULL;
483 	tcp_tw_2msl_stop(tw, reuse);
484 	inp->inp_ppcb = NULL;
485 	in_pcbdrop(inp);
486 
487 	so = inp->inp_socket;
488 	if (so != NULL) {
489 		/*
490 		 * If there's a socket, handle two cases: first, we own a
491 		 * strong reference, which we will now release, or we don't
492 		 * in which case another reference exists (XXXRW: think
493 		 * about this more), and we don't need to take action.
494 		 */
495 		if (inp->inp_flags & INP_SOCKREF) {
496 			inp->inp_flags &= ~INP_SOCKREF;
497 			INP_WUNLOCK(inp);
498 			ACCEPT_LOCK();
499 			SOCK_LOCK(so);
500 			KASSERT(so->so_state & SS_PROTOREF,
501 			    ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
502 			so->so_state &= ~SS_PROTOREF;
503 			sofree(so);
504 		} else {
505 			/*
506 			 * If we don't own the only reference, the socket and
507 			 * inpcb need to be left around to be handled by
508 			 * tcp_usr_detach() later.
509 			 */
510 			INP_WUNLOCK(inp);
511 		}
512 	} else
513 		in_pcbfree(inp);
514 	TCPSTAT_INC(tcps_closed);
515 }
516 
517 static int
518 tcp_twrespond(struct tcptw *tw, int flags)
519 {
520 	struct inpcb *inp = tw->tw_inpcb;
521 #if defined(INET6) || defined(INET)
522 	struct tcphdr *th = NULL;
523 #endif
524 	struct mbuf *m;
525 #ifdef INET
526 	struct ip *ip = NULL;
527 #endif
528 	u_int hdrlen, optlen;
529 	int error = 0;			/* Keep compiler happy */
530 	struct tcpopt to;
531 #ifdef INET6
532 	struct ip6_hdr *ip6 = NULL;
533 	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
534 #endif
535 	hdrlen = 0;                     /* Keep compiler happy */
536 
537 	INP_WLOCK_ASSERT(inp);
538 
539 	m = m_gethdr(M_NOWAIT, MT_DATA);
540 	if (m == NULL)
541 		return (ENOBUFS);
542 	m->m_data += max_linkhdr;
543 
544 #ifdef MAC
545 	mac_inpcb_create_mbuf(inp, m);
546 #endif
547 
548 #ifdef INET6
549 	if (isipv6) {
550 		hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
551 		ip6 = mtod(m, struct ip6_hdr *);
552 		th = (struct tcphdr *)(ip6 + 1);
553 		tcpip_fillheaders(inp, ip6, th);
554 	}
555 #endif
556 #if defined(INET6) && defined(INET)
557 	else
558 #endif
559 #ifdef INET
560 	{
561 		hdrlen = sizeof(struct tcpiphdr);
562 		ip = mtod(m, struct ip *);
563 		th = (struct tcphdr *)(ip + 1);
564 		tcpip_fillheaders(inp, ip, th);
565 	}
566 #endif
567 	to.to_flags = 0;
568 
569 	/*
570 	 * Send a timestamp and echo-reply if both our side and our peer
571 	 * have sent timestamps in our SYN's and this is not a RST.
572 	 */
573 	if (tw->t_recent && flags == TH_ACK) {
574 		to.to_flags |= TOF_TS;
575 		to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
576 		to.to_tsecr = tw->t_recent;
577 	}
578 	optlen = tcp_addoptions(&to, (u_char *)(th + 1));
579 
580 	m->m_len = hdrlen + optlen;
581 	m->m_pkthdr.len = m->m_len;
582 
583 	KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
584 
585 	th->th_seq = htonl(tw->snd_nxt);
586 	th->th_ack = htonl(tw->rcv_nxt);
587 	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
588 	th->th_flags = flags;
589 	th->th_win = htons(tw->last_win);
590 
591 	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
592 #ifdef INET6
593 	if (isipv6) {
594 		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
595 		th->th_sum = in6_cksum_pseudo(ip6,
596 		    sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
597 		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
598 		error = ip6_output(m, inp->in6p_outputopts, NULL,
599 		    (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
600 	}
601 #endif
602 #if defined(INET6) && defined(INET)
603 	else
604 #endif
605 #ifdef INET
606 	{
607 		m->m_pkthdr.csum_flags = CSUM_TCP;
608 		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
609 		    htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
610 		ip->ip_len = htons(m->m_pkthdr.len);
611 		if (V_path_mtu_discovery)
612 			ip->ip_off |= htons(IP_DF);
613 		error = ip_output(m, inp->inp_options, NULL,
614 		    ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
615 		    NULL, inp);
616 	}
617 #endif
618 	if (flags & TH_ACK)
619 		TCPSTAT_INC(tcps_sndacks);
620 	else
621 		TCPSTAT_INC(tcps_sndctrl);
622 	TCPSTAT_INC(tcps_sndtotal);
623 	return (error);
624 }
625 
626 static void
627 tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
628 {
629 
630 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
631 	INP_WLOCK_ASSERT(tw->tw_inpcb);
632 
633 	TW_WLOCK(V_tw_lock);
634 	if (rearm)
635 		TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
636 	tw->tw_time = ticks + 2 * tcp_msl;
637 	TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
638 	TW_WUNLOCK(V_tw_lock);
639 }
640 
641 static void
642 tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
643 {
644 
645 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
646 
647 	TW_WLOCK(V_tw_lock);
648 	TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
649 	crfree(tw->tw_cred);
650 	tw->tw_cred = NULL;
651 	TW_WUNLOCK(V_tw_lock);
652 
653 	if (!reuse)
654 		tw_pcbrele(tw);
655 }
656 
657 struct tcptw *
658 tcp_tw_2msl_reuse(void)
659 {
660 	struct tcptw *tw;
661 
662 	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
663 
664 	TW_WLOCK(V_tw_lock);
665 	tw = TAILQ_FIRST(&V_twq_2msl);
666 	if (tw == NULL) {
667 		TW_WUNLOCK(V_tw_lock);
668 		return NULL;
669 	}
670 	TW_WUNLOCK(V_tw_lock);
671 
672 	INP_WLOCK(tw->tw_inpcb);
673 	tcp_twclose(tw, 1);
674 
675 	return (tw);
676 }
677 
678 void
679 tcp_tw_2msl_scan(void)
680 {
681 	struct tcptw *tw;
682 
683 	for (;;) {
684 		TW_RLOCK(V_tw_lock);
685 		tw = TAILQ_FIRST(&V_twq_2msl);
686 		if (tw == NULL || tw->tw_time - ticks > 0) {
687 			TW_RUNLOCK(V_tw_lock);
688 			break;
689 		}
690 		tw_pcbref(tw);
691 		TW_RUNLOCK(V_tw_lock);
692 
693 		/* Close timewait state */
694 		if (INP_INFO_TRY_WLOCK(&V_tcbinfo)) {
695 			if (tw_pcbrele(tw)) {
696 				INP_INFO_WUNLOCK(&V_tcbinfo);
697 				continue;
698 			}
699 
700 			KASSERT(tw->tw_inpcb != NULL,
701 			    ("%s: tw->tw_inpcb == NULL", __func__));
702 			INP_WLOCK(tw->tw_inpcb);
703 			tcp_twclose(tw, 0);
704 			INP_INFO_WUNLOCK(&V_tcbinfo);
705 		} else {
706 			/* INP_INFO lock is busy; continue later. */
707 			tw_pcbrele(tw);
708 			break;
709 		}
710 	}
711 }
712