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: releng/10.3/sys/netinet/tcp_timewait.c 282826 2015-05-13 00:28:36Z gnn $");
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
53#include <vm/uma.h>
54
55#include <net/route.h>
56#include <net/if.h>
57#include <net/vnet.h>
58
59#include <netinet/in.h>
60#include <netinet/in_pcb.h>
61#include <netinet/in_systm.h>
62#include <netinet/in_var.h>
63#include <netinet/ip.h>
64#include <netinet/ip_icmp.h>
65#include <netinet/ip_var.h>
66#ifdef INET6
67#include <netinet/ip6.h>
68#include <netinet6/in6_pcb.h>
69#include <netinet6/ip6_var.h>
70#include <netinet6/scope6_var.h>
71#include <netinet6/nd6.h>
72#endif
73#include <netinet/tcp.h>
74#include <netinet/tcp_fsm.h>
75#include <netinet/tcp_seq.h>
76#include <netinet/tcp_timer.h>
77#include <netinet/tcp_var.h>
78#ifdef INET6
79#include <netinet6/tcp6_var.h>
80#endif
81#include <netinet/tcpip.h>
82#ifdef TCPDEBUG
83#include <netinet/tcp_debug.h>
84#endif
85#ifdef INET6
86#include <netinet6/ip6protosw.h>
87#endif
88
89#include <machine/in_cksum.h>
90
91#include <security/mac/mac_framework.h>
92
93static VNET_DEFINE(uma_zone_t, tcptw_zone);
94#define	V_tcptw_zone		VNET(tcptw_zone)
95static int	maxtcptw;
96
97/*
98 * The timed wait queue contains references to each of the TCP sessions
99 * currently in the TIME_WAIT state.  The queue pointers, including the
100 * queue pointers in each tcptw structure, are protected using the global
101 * timewait lock, which must be held over queue iteration and modification.
102 *
103 * Rules on tcptw usage:
104 *  - a inpcb is always freed _after_ its tcptw
105 *  - a tcptw relies on its inpcb reference counting for memory stability
106 *  - a tcptw is dereferenceable only while its inpcb is locked
107 */
108static VNET_DEFINE(TAILQ_HEAD(, tcptw), twq_2msl);
109#define	V_twq_2msl		VNET(twq_2msl)
110
111/* Global timewait lock */
112static VNET_DEFINE(struct rwlock, tw_lock);
113#define	V_tw_lock		VNET(tw_lock)
114
115#define	TW_LOCK_INIT(tw, d)	rw_init_flags(&(tw), (d), 0)
116#define	TW_LOCK_DESTROY(tw)	rw_destroy(&(tw))
117#define	TW_RLOCK(tw)		rw_rlock(&(tw))
118#define	TW_WLOCK(tw)		rw_wlock(&(tw))
119#define	TW_RUNLOCK(tw)		rw_runlock(&(tw))
120#define	TW_WUNLOCK(tw)		rw_wunlock(&(tw))
121#define	TW_LOCK_ASSERT(tw)	rw_assert(&(tw), RA_LOCKED)
122#define	TW_RLOCK_ASSERT(tw)	rw_assert(&(tw), RA_RLOCKED)
123#define	TW_WLOCK_ASSERT(tw)	rw_assert(&(tw), RA_WLOCKED)
124#define	TW_UNLOCK_ASSERT(tw)	rw_assert(&(tw), RA_UNLOCKED)
125
126static void	tcp_tw_2msl_reset(struct tcptw *, int);
127static void	tcp_tw_2msl_stop(struct tcptw *, int);
128static int	tcp_twrespond(struct tcptw *, int);
129
130static int
131tcptw_auto_size(void)
132{
133	int halfrange;
134
135	/*
136	 * Max out at half the ephemeral port range so that TIME_WAIT
137	 * sockets don't tie up too many ephemeral ports.
138	 */
139	if (V_ipport_lastauto > V_ipport_firstauto)
140		halfrange = (V_ipport_lastauto - V_ipport_firstauto) / 2;
141	else
142		halfrange = (V_ipport_firstauto - V_ipport_lastauto) / 2;
143	/* Protect against goofy port ranges smaller than 32. */
144	return (imin(imax(halfrange, 32), maxsockets / 5));
145}
146
147static int
148sysctl_maxtcptw(SYSCTL_HANDLER_ARGS)
149{
150	int error, new;
151
152	if (maxtcptw == 0)
153		new = tcptw_auto_size();
154	else
155		new = maxtcptw;
156	error = sysctl_handle_int(oidp, &new, 0, req);
157	if (error == 0 && req->newptr)
158		if (new >= 32) {
159			maxtcptw = new;
160			uma_zone_set_max(V_tcptw_zone, maxtcptw);
161		}
162	return (error);
163}
164
165SYSCTL_PROC(_net_inet_tcp, OID_AUTO, maxtcptw, CTLTYPE_INT|CTLFLAG_RW,
166    &maxtcptw, 0, sysctl_maxtcptw, "IU",
167    "Maximum number of compressed TCP TIME_WAIT entries");
168
169VNET_DEFINE(int, nolocaltimewait) = 0;
170#define	V_nolocaltimewait	VNET(nolocaltimewait)
171SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, nolocaltimewait, CTLFLAG_RW,
172    &VNET_NAME(nolocaltimewait), 0,
173    "Do not create compressed TCP TIME_WAIT entries for local connections");
174
175void
176tcp_tw_zone_change(void)
177{
178
179	if (maxtcptw == 0)
180		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
181}
182
183void
184tcp_tw_init(void)
185{
186
187	V_tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
188	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
189	TUNABLE_INT_FETCH("net.inet.tcp.maxtcptw", &maxtcptw);
190	if (maxtcptw == 0)
191		uma_zone_set_max(V_tcptw_zone, tcptw_auto_size());
192	else
193		uma_zone_set_max(V_tcptw_zone, maxtcptw);
194	TAILQ_INIT(&V_twq_2msl);
195	TW_LOCK_INIT(V_tw_lock, "tcptw");
196}
197
198#ifdef VIMAGE
199void
200tcp_tw_destroy(void)
201{
202	struct tcptw *tw;
203
204	INP_INFO_WLOCK(&V_tcbinfo);
205	while ((tw = TAILQ_FIRST(&V_twq_2msl)) != NULL)
206		tcp_twclose(tw, 0);
207	INP_INFO_WUNLOCK(&V_tcbinfo);
208
209	TW_LOCK_DESTROY(V_tw_lock);
210	uma_zdestroy(V_tcptw_zone);
211}
212#endif
213
214/*
215 * Move a TCP connection into TIME_WAIT state.
216 *    tcbinfo is locked.
217 *    inp is locked, and is unlocked before returning.
218 */
219void
220tcp_twstart(struct tcpcb *tp)
221{
222	struct tcptw *tw;
223	struct inpcb *inp = tp->t_inpcb;
224	int acknow;
225	struct socket *so;
226#ifdef INET6
227	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
228#endif
229
230	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
231	INP_WLOCK_ASSERT(inp);
232
233	if (V_nolocaltimewait) {
234		int error = 0;
235#ifdef INET6
236		if (isipv6)
237			error = in6_localaddr(&inp->in6p_faddr);
238#endif
239#if defined(INET6) && defined(INET)
240		else
241#endif
242#ifdef INET
243			error = in_localip(inp->inp_faddr);
244#endif
245		if (error) {
246			tp = tcp_close(tp);
247			if (tp != NULL)
248				INP_WUNLOCK(inp);
249			return;
250		}
251	}
252
253
254	/*
255	 * For use only by DTrace.  We do not reference the state
256	 * after this point so modifying it in place is not a problem.
257	 */
258	tcp_state_change(tp, TCPS_TIME_WAIT);
259
260	tw = uma_zalloc(V_tcptw_zone, M_NOWAIT);
261	if (tw == NULL) {
262		/*
263		 * Reached limit on total number of TIMEWAIT connections
264		 * allowed. Remove a connection from TIMEWAIT queue in LRU
265		 * fashion to make room for this connection.
266		 *
267		 * pcbinfo lock is needed here to prevent deadlock as
268		 * two inpcb locks can be acquired simultaneously.
269		 */
270		tw = tcp_tw_2msl_scan(1);
271		if (tw == NULL) {
272			tp = tcp_close(tp);
273			if (tp != NULL)
274				INP_WUNLOCK(inp);
275			return;
276		}
277	}
278	/*
279	 * The tcptw will hold a reference on its inpcb until tcp_twclose
280	 * is called
281	 */
282	tw->tw_inpcb = inp;
283	in_pcbref(inp);	/* Reference from tw */
284
285	/*
286	 * Recover last window size sent.
287	 */
288	if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
289		tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
290	else
291		tw->last_win = 0;
292
293	/*
294	 * Set t_recent if timestamps are used on the connection.
295	 */
296	if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
297	    (TF_REQ_TSTMP|TF_RCVD_TSTMP)) {
298		tw->t_recent = tp->ts_recent;
299		tw->ts_offset = tp->ts_offset;
300	} else {
301		tw->t_recent = 0;
302		tw->ts_offset = 0;
303	}
304
305	tw->snd_nxt = tp->snd_nxt;
306	tw->rcv_nxt = tp->rcv_nxt;
307	tw->iss     = tp->iss;
308	tw->irs     = tp->irs;
309	tw->t_starttime = tp->t_starttime;
310	tw->tw_time = 0;
311
312/* XXX
313 * If this code will
314 * be used for fin-wait-2 state also, then we may need
315 * a ts_recent from the last segment.
316 */
317	acknow = tp->t_flags & TF_ACKNOW;
318
319	/*
320	 * First, discard tcpcb state, which includes stopping its timers and
321	 * freeing it.  tcp_discardcb() used to also release the inpcb, but
322	 * that work is now done in the caller.
323	 *
324	 * Note: soisdisconnected() call used to be made in tcp_discardcb(),
325	 * and might not be needed here any longer.
326	 */
327	tcp_discardcb(tp);
328	so = inp->inp_socket;
329	soisdisconnected(so);
330	tw->tw_cred = crhold(so->so_cred);
331	SOCK_LOCK(so);
332	tw->tw_so_options = so->so_options;
333	SOCK_UNLOCK(so);
334	if (acknow)
335		tcp_twrespond(tw, TH_ACK);
336	inp->inp_ppcb = tw;
337	inp->inp_flags |= INP_TIMEWAIT;
338	tcp_tw_2msl_reset(tw, 0);
339
340	/*
341	 * If the inpcb owns the sole reference to the socket, then we can
342	 * detach and free the socket as it is not needed in time wait.
343	 */
344	if (inp->inp_flags & INP_SOCKREF) {
345		KASSERT(so->so_state & SS_PROTOREF,
346		    ("tcp_twstart: !SS_PROTOREF"));
347		inp->inp_flags &= ~INP_SOCKREF;
348		INP_WUNLOCK(inp);
349		ACCEPT_LOCK();
350		SOCK_LOCK(so);
351		so->so_state &= ~SS_PROTOREF;
352		sofree(so);
353	} else
354		INP_WUNLOCK(inp);
355}
356
357#if 0
358/*
359 * The appromixate rate of ISN increase of Microsoft TCP stacks;
360 * the actual rate is slightly higher due to the addition of
361 * random positive increments.
362 *
363 * Most other new OSes use semi-randomized ISN values, so we
364 * do not need to worry about them.
365 */
366#define	MS_ISN_BYTES_PER_SECOND		250000
367
368/*
369 * Determine if the ISN we will generate has advanced beyond the last
370 * sequence number used by the previous connection.  If so, indicate
371 * that it is safe to recycle this tw socket by returning 1.
372 */
373int
374tcp_twrecycleable(struct tcptw *tw)
375{
376	tcp_seq new_iss = tw->iss;
377	tcp_seq new_irs = tw->irs;
378
379	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
380	new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
381	new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
382
383	if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
384		return (1);
385	else
386		return (0);
387}
388#endif
389
390/*
391 * Returns 1 if the TIME_WAIT state was killed and we should start over,
392 * looking for a pcb in the listen state.  Returns 0 otherwise.
393 */
394int
395tcp_twcheck(struct inpcb *inp, struct tcpopt *to __unused, struct tcphdr *th,
396    struct mbuf *m, int tlen)
397{
398	struct tcptw *tw;
399	int thflags;
400	tcp_seq seq;
401
402	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
403	INP_WLOCK_ASSERT(inp);
404
405	/*
406	 * XXXRW: Time wait state for inpcb has been recycled, but inpcb is
407	 * still present.  This is undesirable, but temporarily necessary
408	 * until we work out how to handle inpcb's who's timewait state has
409	 * been removed.
410	 */
411	tw = intotw(inp);
412	if (tw == NULL)
413		goto drop;
414
415	thflags = th->th_flags;
416
417	/*
418	 * NOTE: for FIN_WAIT_2 (to be added later),
419	 * must validate sequence number before accepting RST
420	 */
421
422	/*
423	 * If the segment contains RST:
424	 *	Drop the segment - see Stevens, vol. 2, p. 964 and
425	 *      RFC 1337.
426	 */
427	if (thflags & TH_RST)
428		goto drop;
429
430#if 0
431/* PAWS not needed at the moment */
432	/*
433	 * RFC 1323 PAWS: If we have a timestamp reply on this segment
434	 * and it's less than ts_recent, drop it.
435	 */
436	if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
437	    TSTMP_LT(to.to_tsval, tp->ts_recent)) {
438		if ((thflags & TH_ACK) == 0)
439			goto drop;
440		goto ack;
441	}
442	/*
443	 * ts_recent is never updated because we never accept new segments.
444	 */
445#endif
446
447	/*
448	 * If a new connection request is received
449	 * while in TIME_WAIT, drop the old connection
450	 * and start over if the sequence numbers
451	 * are above the previous ones.
452	 */
453	if ((thflags & TH_SYN) && SEQ_GT(th->th_seq, tw->rcv_nxt)) {
454		tcp_twclose(tw, 0);
455		return (1);
456	}
457
458	/*
459	 * Drop the segment if it does not contain an ACK.
460	 */
461	if ((thflags & TH_ACK) == 0)
462		goto drop;
463
464	/*
465	 * Reset the 2MSL timer if this is a duplicate FIN.
466	 */
467	if (thflags & TH_FIN) {
468		seq = th->th_seq + tlen + (thflags & TH_SYN ? 1 : 0);
469		if (seq + 1 == tw->rcv_nxt)
470			tcp_tw_2msl_reset(tw, 1);
471	}
472
473	/*
474	 * Acknowledge the segment if it has data or is not a duplicate ACK.
475	 */
476	if (thflags != TH_ACK || tlen != 0 ||
477	    th->th_seq != tw->rcv_nxt || th->th_ack != tw->snd_nxt)
478		tcp_twrespond(tw, TH_ACK);
479drop:
480	INP_WUNLOCK(inp);
481	m_freem(m);
482	return (0);
483}
484
485void
486tcp_twclose(struct tcptw *tw, int reuse)
487{
488	struct socket *so;
489	struct inpcb *inp;
490
491	/*
492	 * At this point, we are in one of two situations:
493	 *
494	 * (1) We have no socket, just an inpcb<->twtcp pair.  We can free
495	 *     all state.
496	 *
497	 * (2) We have a socket -- if we own a reference, release it and
498	 *     notify the socket layer.
499	 */
500	inp = tw->tw_inpcb;
501	KASSERT((inp->inp_flags & INP_TIMEWAIT), ("tcp_twclose: !timewait"));
502	KASSERT(intotw(inp) == tw, ("tcp_twclose: inp_ppcb != tw"));
503	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);	/* in_pcbfree() */
504	INP_WLOCK_ASSERT(inp);
505
506	tcp_tw_2msl_stop(tw, reuse);
507	inp->inp_ppcb = NULL;
508	in_pcbdrop(inp);
509
510	so = inp->inp_socket;
511	if (so != NULL) {
512		/*
513		 * If there's a socket, handle two cases: first, we own a
514		 * strong reference, which we will now release, or we don't
515		 * in which case another reference exists (XXXRW: think
516		 * about this more), and we don't need to take action.
517		 */
518		if (inp->inp_flags & INP_SOCKREF) {
519			inp->inp_flags &= ~INP_SOCKREF;
520			INP_WUNLOCK(inp);
521			ACCEPT_LOCK();
522			SOCK_LOCK(so);
523			KASSERT(so->so_state & SS_PROTOREF,
524			    ("tcp_twclose: INP_SOCKREF && !SS_PROTOREF"));
525			so->so_state &= ~SS_PROTOREF;
526			sofree(so);
527		} else {
528			/*
529			 * If we don't own the only reference, the socket and
530			 * inpcb need to be left around to be handled by
531			 * tcp_usr_detach() later.
532			 */
533			INP_WUNLOCK(inp);
534		}
535	} else {
536		/*
537		 * The socket has been already cleaned-up for us, only free the
538		 * inpcb.
539		 */
540		in_pcbfree(inp);
541	}
542	TCPSTAT_INC(tcps_closed);
543}
544
545static int
546tcp_twrespond(struct tcptw *tw, int flags)
547{
548	struct inpcb *inp = tw->tw_inpcb;
549#if defined(INET6) || defined(INET)
550	struct tcphdr *th = NULL;
551#endif
552	struct mbuf *m;
553#ifdef INET
554	struct ip *ip = NULL;
555#endif
556	u_int hdrlen, optlen;
557	int error = 0;			/* Keep compiler happy */
558	struct tcpopt to;
559#ifdef INET6
560	struct ip6_hdr *ip6 = NULL;
561	int isipv6 = inp->inp_inc.inc_flags & INC_ISIPV6;
562#endif
563	hdrlen = 0;                     /* Keep compiler happy */
564
565	INP_WLOCK_ASSERT(inp);
566
567	m = m_gethdr(M_NOWAIT, MT_DATA);
568	if (m == NULL)
569		return (ENOBUFS);
570	m->m_data += max_linkhdr;
571
572#ifdef MAC
573	mac_inpcb_create_mbuf(inp, m);
574#endif
575
576#ifdef INET6
577	if (isipv6) {
578		hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
579		ip6 = mtod(m, struct ip6_hdr *);
580		th = (struct tcphdr *)(ip6 + 1);
581		tcpip_fillheaders(inp, ip6, th);
582	}
583#endif
584#if defined(INET6) && defined(INET)
585	else
586#endif
587#ifdef INET
588	{
589		hdrlen = sizeof(struct tcpiphdr);
590		ip = mtod(m, struct ip *);
591		th = (struct tcphdr *)(ip + 1);
592		tcpip_fillheaders(inp, ip, th);
593	}
594#endif
595	to.to_flags = 0;
596
597	/*
598	 * Send a timestamp and echo-reply if both our side and our peer
599	 * have sent timestamps in our SYN's and this is not a RST.
600	 */
601	if (tw->t_recent && flags == TH_ACK) {
602		to.to_flags |= TOF_TS;
603		to.to_tsval = tcp_ts_getticks() + tw->ts_offset;
604		to.to_tsecr = tw->t_recent;
605	}
606	optlen = tcp_addoptions(&to, (u_char *)(th + 1));
607
608	m->m_len = hdrlen + optlen;
609	m->m_pkthdr.len = m->m_len;
610
611	KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
612
613	th->th_seq = htonl(tw->snd_nxt);
614	th->th_ack = htonl(tw->rcv_nxt);
615	th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
616	th->th_flags = flags;
617	th->th_win = htons(tw->last_win);
618
619	m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
620#ifdef INET6
621	if (isipv6) {
622		m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
623		th->th_sum = in6_cksum_pseudo(ip6,
624		    sizeof(struct tcphdr) + optlen, IPPROTO_TCP, 0);
625		ip6->ip6_hlim = in6_selecthlim(inp, NULL);
626		error = ip6_output(m, inp->in6p_outputopts, NULL,
627		    (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
628	}
629#endif
630#if defined(INET6) && defined(INET)
631	else
632#endif
633#ifdef INET
634	{
635		m->m_pkthdr.csum_flags = CSUM_TCP;
636		th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
637		    htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
638		ip->ip_len = htons(m->m_pkthdr.len);
639		if (V_path_mtu_discovery)
640			ip->ip_off |= htons(IP_DF);
641		error = ip_output(m, inp->inp_options, NULL,
642		    ((tw->tw_so_options & SO_DONTROUTE) ? IP_ROUTETOIF : 0),
643		    NULL, inp);
644	}
645#endif
646	if (flags & TH_ACK)
647		TCPSTAT_INC(tcps_sndacks);
648	else
649		TCPSTAT_INC(tcps_sndctrl);
650	TCPSTAT_INC(tcps_sndtotal);
651	return (error);
652}
653
654static void
655tcp_tw_2msl_reset(struct tcptw *tw, int rearm)
656{
657
658	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
659	INP_WLOCK_ASSERT(tw->tw_inpcb);
660
661	TW_WLOCK(V_tw_lock);
662	if (rearm)
663		TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
664	tw->tw_time = ticks + 2 * tcp_msl;
665	TAILQ_INSERT_TAIL(&V_twq_2msl, tw, tw_2msl);
666	TW_WUNLOCK(V_tw_lock);
667}
668
669static void
670tcp_tw_2msl_stop(struct tcptw *tw, int reuse)
671{
672	struct ucred *cred;
673	struct inpcb *inp;
674	int released;
675
676	INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
677
678	TW_WLOCK(V_tw_lock);
679	inp = tw->tw_inpcb;
680	tw->tw_inpcb = NULL;
681
682	TAILQ_REMOVE(&V_twq_2msl, tw, tw_2msl);
683	cred = tw->tw_cred;
684	tw->tw_cred = NULL;
685	TW_WUNLOCK(V_tw_lock);
686
687	if (cred != NULL)
688		crfree(cred);
689
690	released = in_pcbrele_wlocked(inp);
691	KASSERT(!released, ("%s: inp should not be released here", __func__));
692
693	if (!reuse)
694		uma_zfree(V_tcptw_zone, tw);
695}
696
697struct tcptw *
698tcp_tw_2msl_scan(int reuse)
699{
700	struct tcptw *tw;
701	struct inpcb *inp;
702
703#ifdef INVARIANTS
704	if (reuse) {
705		/*
706		 * pcbinfo lock is needed in reuse case to prevent deadlock
707		 * as two inpcb locks can be acquired simultaneously:
708		 *  - the inpcb transitioning to TIME_WAIT state in
709		 *    tcp_tw_start(),
710		 *  - the inpcb closed by tcp_twclose().
711		 */
712		INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
713	}
714#endif
715
716	for (;;) {
717		TW_RLOCK(V_tw_lock);
718		tw = TAILQ_FIRST(&V_twq_2msl);
719		if (tw == NULL || (!reuse && (tw->tw_time - ticks) > 0)) {
720			TW_RUNLOCK(V_tw_lock);
721			break;
722		}
723		KASSERT(tw->tw_inpcb != NULL, ("%s: tw->tw_inpcb == NULL",
724		    __func__));
725
726		inp = tw->tw_inpcb;
727		in_pcbref(inp);
728		TW_RUNLOCK(V_tw_lock);
729
730		if (INP_INFO_TRY_WLOCK(&V_tcbinfo)) {
731
732			INP_WLOCK(inp);
733			tw = intotw(inp);
734			if (in_pcbrele_wlocked(inp)) {
735				KASSERT(tw == NULL, ("%s: held last inp "
736				    "reference but tw not NULL", __func__));
737				INP_INFO_WUNLOCK(&V_tcbinfo);
738				continue;
739			}
740
741			if (tw == NULL) {
742				/* tcp_twclose() has already been called */
743				INP_WUNLOCK(inp);
744				INP_INFO_WUNLOCK(&V_tcbinfo);
745				continue;
746			}
747
748			tcp_twclose(tw, reuse);
749			INP_INFO_WUNLOCK(&V_tcbinfo);
750			if (reuse)
751			    return tw;
752		} else {
753			/* INP_INFO lock is busy, continue later. */
754			INP_WLOCK(inp);
755			if (!in_pcbrele_wlocked(inp))
756				INP_WUNLOCK(inp);
757			break;
758		}
759	}
760
761	return NULL;
762}
763